linux/io_uring/io_uring.c

4730 lines
122 KiB
C
Raw Normal View History

Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
// SPDX-License-Identifier: GPL-2.0
/*
* Shared application/kernel submission and completion ring pairs, for
* supporting fast/efficient IO.
*
* A note on the read/write ordering memory barriers that are matched between
* the application and kernel side.
*
* After the application reads the CQ ring tail, it must use an
* appropriate smp_rmb() to pair with the smp_wmb() the kernel uses
* before writing the tail (using smp_load_acquire to read the tail will
* do). It also needs a smp_mb() before updating CQ head (ordering the
* entry load(s) with the head store), pairing with an implicit barrier
* through a control-dependency in io_get_cqe (smp_store_release to
* store head will do). Failure to do so could lead to reading invalid
* CQ entries.
*
* Likewise, the application must use an appropriate smp_wmb() before
* writing the SQ tail (ordering SQ entry stores with the tail store),
* which pairs with smp_load_acquire in io_get_sqring (smp_store_release
* to store the tail will do). And it needs a barrier ordering the SQ
* head load before writing new SQ entries (smp_load_acquire to read
* head will do).
*
* When using the SQ poll thread (IORING_SETUP_SQPOLL), the application
* needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after*
* updating the SQ tail; a full memory barrier smp_mb() is needed
* between.
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
*
* Also see the examples in the liburing library:
*
* git://git.kernel.dk/liburing
*
* io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens
* from data shared between the kernel and application. This is done both
* for ordering purposes, but also to ensure that once a value is loaded from
* data that the application could potentially modify, it remains stable.
*
* Copyright (C) 2018-2019 Jens Axboe
* Copyright (c) 2018-2019 Christoph Hellwig
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
*/
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/errno.h>
#include <linux/syscalls.h>
#include <net/compat.h>
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#include <linux/refcount.h>
#include <linux/uio.h>
#include <linux/bits.h>
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#include <linux/sched/signal.h>
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/fdtable.h>
#include <linux/mm.h>
#include <linux/mman.h>
#include <linux/percpu.h>
#include <linux/slab.h>
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
#include <linux/bvec.h>
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#include <linux/net.h>
#include <net/sock.h>
#include <net/af_unix.h>
#include <net/scm.h>
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#include <linux/anon_inodes.h>
#include <linux/sched/mm.h>
#include <linux/uaccess.h>
#include <linux/nospec.h>
#include <linux/highmem.h>
#include <linux/fsnotify.h>
#include <linux/fadvise.h>
#include <linux/task_work.h>
#include <linux/io_uring.h>
#include <linux/audit.h>
lsm,io_uring: add LSM hooks to io_uring A full expalantion of io_uring is beyond the scope of this commit description, but in summary it is an asynchronous I/O mechanism which allows for I/O requests and the resulting data to be queued in memory mapped "rings" which are shared between the kernel and userspace. Optionally, io_uring offers the ability for applications to spawn kernel threads to dequeue I/O requests from the ring and submit the requests in the kernel, helping to minimize the syscall overhead. Rings are accessed in userspace by memory mapping a file descriptor provided by the io_uring_setup(2), and can be shared between applications as one might do with any open file descriptor. Finally, process credentials can be registered with a given ring and any process with access to that ring can submit I/O requests using any of the registered credentials. While the io_uring functionality is widely recognized as offering a vastly improved, and high performing asynchronous I/O mechanism, its ability to allow processes to submit I/O requests with credentials other than its own presents a challenge to LSMs. When a process creates a new io_uring ring the ring's credentials are inhertied from the calling process; if this ring is shared with another process operating with different credentials there is the potential to bypass the LSMs security policy. Similarly, registering credentials with a given ring allows any process with access to that ring to submit I/O requests with those credentials. In an effort to allow LSMs to apply security policy to io_uring I/O operations, this patch adds two new LSM hooks. These hooks, in conjunction with the LSM anonymous inode support previously submitted, allow an LSM to apply access control policy to the sharing of io_uring rings as well as any io_uring credential changes requested by a process. The new LSM hooks are described below: * int security_uring_override_creds(cred) Controls if the current task, executing an io_uring operation, is allowed to override it's credentials with @cred. In cases where the current task is a user application, the current credentials will be those of the user application. In cases where the current task is a kernel thread servicing io_uring requests the current credentials will be those of the io_uring ring (inherited from the process that created the ring). * int security_uring_sqpoll(void) Controls if the current task is allowed to create an io_uring polling thread (IORING_SETUP_SQPOLL). Without a SQPOLL thread in the kernel processes must submit I/O requests via io_uring_enter(2) which allows us to compare any requested credential changes against the application making the request. With a SQPOLL thread, we can no longer compare requested credential changes against the application making the request, the comparison is made against the ring's credentials. Signed-off-by: Paul Moore <paul@paul-moore.com>
2021-02-02 08:56:49 +08:00
#include <linux/security.h>
#include <asm/shmparam.h>
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#define CREATE_TRACE_POINTS
#include <trace/events/io_uring.h>
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#include <uapi/linux/io_uring.h>
#include "io-wq.h"
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#include "io_uring.h"
#include "opdef.h"
#include "refs.h"
#include "tctx.h"
#include "sqpoll.h"
#include "fdinfo.h"
#include "kbuf.h"
#include "rsrc.h"
#include "cancel.h"
#include "net.h"
io_uring: add zc notification infrastructure Add internal part of send zerocopy notifications. There are two main structures, the first one is struct io_notif, which carries inside struct ubuf_info and maps 1:1 to it. io_uring will be binding a number of zerocopy send requests to it and ask to complete (aka flush) it. When flushed and all attached requests and skbs complete, it'll generate one and only one CQE. There are intended to be passed into the network layer as struct msghdr::msg_ubuf. The second concept is notification slots. The userspace will be able to register an array of slots and subsequently addressing them by the index in the array. Slots are independent of each other. Each slot can have only one notifier at a time (called active notifier) but many notifiers during the lifetime. When active, a notifier not going to post any completion but the userspace can attach requests to it by specifying the corresponding slot while issueing send zc requests. Eventually, the userspace will want to "flush" the notifier losing any way to attach new requests to it, however it can use the next atomatically added notifier of this slot or of any other slot. When the network layer is done with all enqueued skbs attached to a notifier and doesn't need the specified in them user data, the flushed notifier will post a CQE. Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/3ecf54c31a85762bf679b0a432c9f43ecf7e61cc.1657643355.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 04:52:38 +08:00
#include "notif.h"
#include "timeout.h"
#include "poll.h"
#include "rw.h"
#include "alloc_cache.h"
#define IORING_MAX_ENTRIES 32768
#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
#define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \
IORING_REGISTER_LAST + IORING_OP_LAST)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#define SQE_COMMON_FLAGS (IOSQE_FIXED_FILE | IOSQE_IO_LINK | \
IOSQE_IO_HARDLINK | IOSQE_ASYNC)
#define SQE_VALID_FLAGS (SQE_COMMON_FLAGS | IOSQE_BUFFER_SELECT | \
IOSQE_IO_DRAIN | IOSQE_CQE_SKIP_SUCCESS)
#define IO_REQ_CLEAN_FLAGS (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | \
REQ_F_POLLED | REQ_F_INFLIGHT | REQ_F_CREDS | \
REQ_F_ASYNC_DATA)
#define IO_REQ_CLEAN_SLOW_FLAGS (REQ_F_REFCOUNT | REQ_F_LINK | REQ_F_HARDLINK |\
IO_REQ_CLEAN_FLAGS)
#define IO_TCTX_REFS_CACHE_NR (1U << 10)
#define IO_COMPL_BATCH 32
#define IO_REQ_ALLOC_BATCH 8
enum {
IO_CHECK_CQ_OVERFLOW_BIT,
IO_CHECK_CQ_DROPPED_BIT,
};
enum {
IO_EVENTFD_OP_SIGNAL_BIT,
IO_EVENTFD_OP_FREE_BIT,
};
struct io_defer_entry {
struct list_head list;
struct io_kiocb *req;
u32 seq;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
};
/* requests with any of those set should undergo io_disarm_next() */
#define IO_DISARM_MASK (REQ_F_ARM_LTIMEOUT | REQ_F_LINK_TIMEOUT | REQ_F_FAIL)
#define IO_REQ_LINK_FLAGS (REQ_F_LINK | REQ_F_HARDLINK)
static bool io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
struct task_struct *task,
bool cancel_all);
static void io_queue_sqe(struct io_kiocb *req);
struct kmem_cache *req_cachep;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
static int __read_mostly sysctl_io_uring_disabled;
static int __read_mostly sysctl_io_uring_group = -1;
#ifdef CONFIG_SYSCTL
static struct ctl_table kernel_io_uring_disabled_table[] = {
{
.procname = "io_uring_disabled",
.data = &sysctl_io_uring_disabled,
.maxlen = sizeof(sysctl_io_uring_disabled),
.mode = 0644,
.proc_handler = proc_dointvec_minmax,
.extra1 = SYSCTL_ZERO,
.extra2 = SYSCTL_TWO,
},
{
.procname = "io_uring_group",
.data = &sysctl_io_uring_group,
.maxlen = sizeof(gid_t),
.mode = 0644,
.proc_handler = proc_dointvec,
},
{},
};
#endif
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
struct sock *io_uring_get_socket(struct file *file)
{
#if defined(CONFIG_UNIX)
if (io_is_uring_fops(file)) {
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
struct io_ring_ctx *ctx = file->private_data;
return ctx->ring_sock->sk;
}
#endif
return NULL;
}
EXPORT_SYMBOL(io_uring_get_socket);
static inline void io_submit_flush_completions(struct io_ring_ctx *ctx)
{
if (!wq_list_empty(&ctx->submit_state.compl_reqs) ||
ctx->submit_state.cqes_count)
__io_submit_flush_completions(ctx);
}
static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
{
return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
}
static inline unsigned int __io_cqring_events_user(struct io_ring_ctx *ctx)
{
return READ_ONCE(ctx->rings->cq.tail) - READ_ONCE(ctx->rings->cq.head);
}
static bool io_match_linked(struct io_kiocb *head)
{
struct io_kiocb *req;
io_for_each_link(req, head) {
if (req->flags & REQ_F_INFLIGHT)
return true;
}
return false;
io_uring: fix link traversal locking WARNING: inconsistent lock state 5.16.0-rc2-syzkaller #0 Not tainted inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage. ffff888078e11418 (&ctx->timeout_lock ){?.+.}-{2:2} , at: io_timeout_fn+0x6f/0x360 fs/io_uring.c:5943 {HARDIRQ-ON-W} state was registered at: [...] spin_unlock_irq include/linux/spinlock.h:399 [inline] __io_poll_remove_one fs/io_uring.c:5669 [inline] __io_poll_remove_one fs/io_uring.c:5654 [inline] io_poll_remove_one+0x236/0x870 fs/io_uring.c:5680 io_poll_remove_all+0x1af/0x235 fs/io_uring.c:5709 io_ring_ctx_wait_and_kill+0x1cc/0x322 fs/io_uring.c:9534 io_uring_release+0x42/0x46 fs/io_uring.c:9554 __fput+0x286/0x9f0 fs/file_table.c:280 task_work_run+0xdd/0x1a0 kernel/task_work.c:164 exit_task_work include/linux/task_work.h:32 [inline] do_exit+0xc14/0x2b40 kernel/exit.c:832 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") fixed a data race but introduced a possible deadlock and inconsistentcy in irq states. E.g. io_poll_remove_all() spin_lock_irq(timeout_lock) io_poll_remove_one() spin_lock/unlock_irq(poll_lock); spin_unlock_irq(timeout_lock) Another type of problem is freeing a request while holding ->timeout_lock, which may leads to a deadlock in io_commit_cqring() -> io_flush_timeouts() and other places. Having 3 nested locks is also too ugly. Add io_match_task_safe(), which would briefly take and release timeout_lock for race prevention inside, so the actuall request cancellation / free / etc. code doesn't have it taken. Reported-by: syzbot+ff49a3059d49b0ca0eec@syzkaller.appspotmail.com Reported-by: syzbot+847f02ec20a6609a328b@syzkaller.appspotmail.com Reported-by: syzbot+3368aadcd30425ceb53b@syzkaller.appspotmail.com Reported-by: syzbot+51ce8887cdef77c9ac83@syzkaller.appspotmail.com Reported-by: syzbot+3cb756a49d2f394a9ee3@syzkaller.appspotmail.com Fixes: 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") Cc: stable@kernel.org # 5.15+ Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/397f7ebf3f4171f1abe41f708ac1ecb5766f0b68.1637937097.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-11-26 22:38:15 +08:00
}
/*
* As io_match_task() but protected against racing with linked timeouts.
* User must not hold timeout_lock.
*/
bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task,
bool cancel_all)
io_uring: fix link traversal locking WARNING: inconsistent lock state 5.16.0-rc2-syzkaller #0 Not tainted inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage. ffff888078e11418 (&ctx->timeout_lock ){?.+.}-{2:2} , at: io_timeout_fn+0x6f/0x360 fs/io_uring.c:5943 {HARDIRQ-ON-W} state was registered at: [...] spin_unlock_irq include/linux/spinlock.h:399 [inline] __io_poll_remove_one fs/io_uring.c:5669 [inline] __io_poll_remove_one fs/io_uring.c:5654 [inline] io_poll_remove_one+0x236/0x870 fs/io_uring.c:5680 io_poll_remove_all+0x1af/0x235 fs/io_uring.c:5709 io_ring_ctx_wait_and_kill+0x1cc/0x322 fs/io_uring.c:9534 io_uring_release+0x42/0x46 fs/io_uring.c:9554 __fput+0x286/0x9f0 fs/file_table.c:280 task_work_run+0xdd/0x1a0 kernel/task_work.c:164 exit_task_work include/linux/task_work.h:32 [inline] do_exit+0xc14/0x2b40 kernel/exit.c:832 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") fixed a data race but introduced a possible deadlock and inconsistentcy in irq states. E.g. io_poll_remove_all() spin_lock_irq(timeout_lock) io_poll_remove_one() spin_lock/unlock_irq(poll_lock); spin_unlock_irq(timeout_lock) Another type of problem is freeing a request while holding ->timeout_lock, which may leads to a deadlock in io_commit_cqring() -> io_flush_timeouts() and other places. Having 3 nested locks is also too ugly. Add io_match_task_safe(), which would briefly take and release timeout_lock for race prevention inside, so the actuall request cancellation / free / etc. code doesn't have it taken. Reported-by: syzbot+ff49a3059d49b0ca0eec@syzkaller.appspotmail.com Reported-by: syzbot+847f02ec20a6609a328b@syzkaller.appspotmail.com Reported-by: syzbot+3368aadcd30425ceb53b@syzkaller.appspotmail.com Reported-by: syzbot+51ce8887cdef77c9ac83@syzkaller.appspotmail.com Reported-by: syzbot+3cb756a49d2f394a9ee3@syzkaller.appspotmail.com Fixes: 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") Cc: stable@kernel.org # 5.15+ Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/397f7ebf3f4171f1abe41f708ac1ecb5766f0b68.1637937097.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-11-26 22:38:15 +08:00
{
bool matched;
io_uring: fix link traversal locking WARNING: inconsistent lock state 5.16.0-rc2-syzkaller #0 Not tainted inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage. ffff888078e11418 (&ctx->timeout_lock ){?.+.}-{2:2} , at: io_timeout_fn+0x6f/0x360 fs/io_uring.c:5943 {HARDIRQ-ON-W} state was registered at: [...] spin_unlock_irq include/linux/spinlock.h:399 [inline] __io_poll_remove_one fs/io_uring.c:5669 [inline] __io_poll_remove_one fs/io_uring.c:5654 [inline] io_poll_remove_one+0x236/0x870 fs/io_uring.c:5680 io_poll_remove_all+0x1af/0x235 fs/io_uring.c:5709 io_ring_ctx_wait_and_kill+0x1cc/0x322 fs/io_uring.c:9534 io_uring_release+0x42/0x46 fs/io_uring.c:9554 __fput+0x286/0x9f0 fs/file_table.c:280 task_work_run+0xdd/0x1a0 kernel/task_work.c:164 exit_task_work include/linux/task_work.h:32 [inline] do_exit+0xc14/0x2b40 kernel/exit.c:832 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") fixed a data race but introduced a possible deadlock and inconsistentcy in irq states. E.g. io_poll_remove_all() spin_lock_irq(timeout_lock) io_poll_remove_one() spin_lock/unlock_irq(poll_lock); spin_unlock_irq(timeout_lock) Another type of problem is freeing a request while holding ->timeout_lock, which may leads to a deadlock in io_commit_cqring() -> io_flush_timeouts() and other places. Having 3 nested locks is also too ugly. Add io_match_task_safe(), which would briefly take and release timeout_lock for race prevention inside, so the actuall request cancellation / free / etc. code doesn't have it taken. Reported-by: syzbot+ff49a3059d49b0ca0eec@syzkaller.appspotmail.com Reported-by: syzbot+847f02ec20a6609a328b@syzkaller.appspotmail.com Reported-by: syzbot+3368aadcd30425ceb53b@syzkaller.appspotmail.com Reported-by: syzbot+51ce8887cdef77c9ac83@syzkaller.appspotmail.com Reported-by: syzbot+3cb756a49d2f394a9ee3@syzkaller.appspotmail.com Fixes: 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") Cc: stable@kernel.org # 5.15+ Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/397f7ebf3f4171f1abe41f708ac1ecb5766f0b68.1637937097.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-11-26 22:38:15 +08:00
if (task && head->task != task)
return false;
if (cancel_all)
return true;
if (head->flags & REQ_F_LINK_TIMEOUT) {
struct io_ring_ctx *ctx = head->ctx;
/* protect against races with linked timeouts */
spin_lock_irq(&ctx->timeout_lock);
matched = io_match_linked(head);
spin_unlock_irq(&ctx->timeout_lock);
} else {
matched = io_match_linked(head);
}
return matched;
io_uring: fix link traversal locking WARNING: inconsistent lock state 5.16.0-rc2-syzkaller #0 Not tainted inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage. ffff888078e11418 (&ctx->timeout_lock ){?.+.}-{2:2} , at: io_timeout_fn+0x6f/0x360 fs/io_uring.c:5943 {HARDIRQ-ON-W} state was registered at: [...] spin_unlock_irq include/linux/spinlock.h:399 [inline] __io_poll_remove_one fs/io_uring.c:5669 [inline] __io_poll_remove_one fs/io_uring.c:5654 [inline] io_poll_remove_one+0x236/0x870 fs/io_uring.c:5680 io_poll_remove_all+0x1af/0x235 fs/io_uring.c:5709 io_ring_ctx_wait_and_kill+0x1cc/0x322 fs/io_uring.c:9534 io_uring_release+0x42/0x46 fs/io_uring.c:9554 __fput+0x286/0x9f0 fs/file_table.c:280 task_work_run+0xdd/0x1a0 kernel/task_work.c:164 exit_task_work include/linux/task_work.h:32 [inline] do_exit+0xc14/0x2b40 kernel/exit.c:832 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") fixed a data race but introduced a possible deadlock and inconsistentcy in irq states. E.g. io_poll_remove_all() spin_lock_irq(timeout_lock) io_poll_remove_one() spin_lock/unlock_irq(poll_lock); spin_unlock_irq(timeout_lock) Another type of problem is freeing a request while holding ->timeout_lock, which may leads to a deadlock in io_commit_cqring() -> io_flush_timeouts() and other places. Having 3 nested locks is also too ugly. Add io_match_task_safe(), which would briefly take and release timeout_lock for race prevention inside, so the actuall request cancellation / free / etc. code doesn't have it taken. Reported-by: syzbot+ff49a3059d49b0ca0eec@syzkaller.appspotmail.com Reported-by: syzbot+847f02ec20a6609a328b@syzkaller.appspotmail.com Reported-by: syzbot+3368aadcd30425ceb53b@syzkaller.appspotmail.com Reported-by: syzbot+51ce8887cdef77c9ac83@syzkaller.appspotmail.com Reported-by: syzbot+3cb756a49d2f394a9ee3@syzkaller.appspotmail.com Fixes: 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") Cc: stable@kernel.org # 5.15+ Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/397f7ebf3f4171f1abe41f708ac1ecb5766f0b68.1637937097.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-11-26 22:38:15 +08:00
}
static inline void req_fail_link_node(struct io_kiocb *req, int res)
{
req_set_fail(req);
io_req_set_res(req, res, 0);
}
static inline void io_req_add_to_cache(struct io_kiocb *req, struct io_ring_ctx *ctx)
{
wq_stack_add_head(&req->comp_list, &ctx->submit_state.free_list);
}
static __cold void io_ring_ctx_ref_free(struct percpu_ref *ref)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
complete(&ctx->ref_comp);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
static __cold void io_fallback_req_func(struct work_struct *work)
{
struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx,
fallback_work.work);
struct llist_node *node = llist_del_all(&ctx->fallback_llist);
struct io_kiocb *req, *tmp;
struct io_tw_state ts = { .locked = true, };
percpu_ref_get(&ctx->refs);
mutex_lock(&ctx->uring_lock);
llist_for_each_entry_safe(req, tmp, node, io_task_work.node)
req->io_task_work.func(req, &ts);
if (WARN_ON_ONCE(!ts.locked))
return;
io_submit_flush_completions(ctx);
mutex_unlock(&ctx->uring_lock);
percpu_ref_put(&ctx->refs);
}
static int io_alloc_hash_table(struct io_hash_table *table, unsigned bits)
{
unsigned hash_buckets = 1U << bits;
size_t hash_size = hash_buckets * sizeof(table->hbs[0]);
table->hbs = kmalloc(hash_size, GFP_KERNEL);
if (!table->hbs)
return -ENOMEM;
table->hash_bits = bits;
init_hash_table(table, hash_buckets);
return 0;
}
static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_ring_ctx *ctx;
int hash_bits;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
if (!ctx)
return NULL;
xa_init(&ctx->io_bl_xa);
/*
* Use 5 bits less than the max cq entries, that should give us around
* 32 entries per hash list if totally full and uniformly spread, but
* don't keep too many buckets to not overconsume memory.
*/
hash_bits = ilog2(p->cq_entries) - 5;
hash_bits = clamp(hash_bits, 1, 8);
if (io_alloc_hash_table(&ctx->cancel_table, hash_bits))
goto err;
if (io_alloc_hash_table(&ctx->cancel_table_locked, hash_bits))
goto err;
if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
0, GFP_KERNEL))
goto err;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ctx->flags = p->flags;
init_waitqueue_head(&ctx->sqo_sq_wait);
INIT_LIST_HEAD(&ctx->sqd_list);
INIT_LIST_HEAD(&ctx->cq_overflow_list);
INIT_LIST_HEAD(&ctx->io_buffers_cache);
INIT_HLIST_HEAD(&ctx->io_buf_list);
io_alloc_cache_init(&ctx->rsrc_node_cache, IO_NODE_ALLOC_CACHE_MAX,
sizeof(struct io_rsrc_node));
io_alloc_cache_init(&ctx->apoll_cache, IO_ALLOC_CACHE_MAX,
sizeof(struct async_poll));
io_alloc_cache_init(&ctx->netmsg_cache, IO_ALLOC_CACHE_MAX,
sizeof(struct io_async_msghdr));
init_completion(&ctx->ref_comp);
xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
mutex_init(&ctx->uring_lock);
init_waitqueue_head(&ctx->cq_wait);
init_waitqueue_head(&ctx->poll_wq);
init_waitqueue_head(&ctx->rsrc_quiesce_wq);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
spin_lock_init(&ctx->completion_lock);
spin_lock_init(&ctx->timeout_lock);
INIT_WQ_LIST(&ctx->iopoll_list);
INIT_LIST_HEAD(&ctx->io_buffers_pages);
INIT_LIST_HEAD(&ctx->io_buffers_comp);
INIT_LIST_HEAD(&ctx->defer_list);
INIT_LIST_HEAD(&ctx->timeout_list);
INIT_LIST_HEAD(&ctx->ltimeout_list);
INIT_LIST_HEAD(&ctx->rsrc_ref_list);
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
init_llist_head(&ctx->work_llist);
INIT_LIST_HEAD(&ctx->tctx_list);
ctx->submit_state.free_list.next = NULL;
INIT_WQ_LIST(&ctx->locked_free_list);
INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func);
INIT_WQ_LIST(&ctx->submit_state.compl_reqs);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return ctx;
err:
kfree(ctx->cancel_table.hbs);
kfree(ctx->cancel_table_locked.hbs);
kfree(ctx->io_bl);
xa_destroy(&ctx->io_bl_xa);
kfree(ctx);
return NULL;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
static void io_account_cq_overflow(struct io_ring_ctx *ctx)
{
struct io_rings *r = ctx->rings;
WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1);
ctx->cq_extra--;
}
static bool req_need_defer(struct io_kiocb *req, u32 seq)
{
if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
struct io_ring_ctx *ctx = req->ctx;
return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail;
}
return false;
}
static void io_clean_op(struct io_kiocb *req)
{
if (req->flags & REQ_F_BUFFER_SELECTED) {
spin_lock(&req->ctx->completion_lock);
io_put_kbuf_comp(req);
spin_unlock(&req->ctx->completion_lock);
}
if (req->flags & REQ_F_NEED_CLEANUP) {
const struct io_cold_def *def = &io_cold_defs[req->opcode];
if (def->cleanup)
def->cleanup(req);
}
if ((req->flags & REQ_F_POLLED) && req->apoll) {
kfree(req->apoll->double_poll);
kfree(req->apoll);
req->apoll = NULL;
}
if (req->flags & REQ_F_INFLIGHT) {
struct io_uring_task *tctx = req->task->io_uring;
atomic_dec(&tctx->inflight_tracked);
}
if (req->flags & REQ_F_CREDS)
put_cred(req->creds);
if (req->flags & REQ_F_ASYNC_DATA) {
kfree(req->async_data);
req->async_data = NULL;
}
req->flags &= ~IO_REQ_CLEAN_FLAGS;
}
static inline void io_req_track_inflight(struct io_kiocb *req)
{
if (!(req->flags & REQ_F_INFLIGHT)) {
req->flags |= REQ_F_INFLIGHT;
atomic_inc(&req->task->io_uring->inflight_tracked);
}
}
static struct io_kiocb *__io_prep_linked_timeout(struct io_kiocb *req)
{
if (WARN_ON_ONCE(!req->link))
return NULL;
req->flags &= ~REQ_F_ARM_LTIMEOUT;
req->flags |= REQ_F_LINK_TIMEOUT;
/* linked timeouts should have two refs once prep'ed */
io_req_set_refcount(req);
__io_req_set_refcount(req->link, 2);
return req->link;
}
static inline struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
{
if (likely(!(req->flags & REQ_F_ARM_LTIMEOUT)))
return NULL;
return __io_prep_linked_timeout(req);
}
static noinline void __io_arm_ltimeout(struct io_kiocb *req)
{
io_queue_linked_timeout(__io_prep_linked_timeout(req));
}
static inline void io_arm_ltimeout(struct io_kiocb *req)
{
if (unlikely(req->flags & REQ_F_ARM_LTIMEOUT))
__io_arm_ltimeout(req);
}
static void io_prep_async_work(struct io_kiocb *req)
{
const struct io_issue_def *def = &io_issue_defs[req->opcode];
struct io_ring_ctx *ctx = req->ctx;
if (!(req->flags & REQ_F_CREDS)) {
req->flags |= REQ_F_CREDS;
req->creds = get_current_cred();
}
req->work.list.next = NULL;
req->work.flags = 0;
req->work.cancel_seq = atomic_read(&ctx->cancel_seq);
if (req->flags & REQ_F_FORCE_ASYNC)
req->work.flags |= IO_WQ_WORK_CONCURRENT;
if (req->file && !(req->flags & REQ_F_FIXED_FILE))
req->flags |= io_file_get_flags(req->file);
if (req->file && (req->flags & REQ_F_ISREG)) {
bool should_hash = def->hash_reg_file;
/* don't serialize this request if the fs doesn't need it */
if (should_hash && (req->file->f_flags & O_DIRECT) &&
(req->file->f_mode & FMODE_DIO_PARALLEL_WRITE))
should_hash = false;
if (should_hash || (ctx->flags & IORING_SETUP_IOPOLL))
io_wq_hash_work(&req->work, file_inode(req->file));
} else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) {
if (def->unbound_nonreg_file)
req->work.flags |= IO_WQ_WORK_UNBOUND;
}
}
static void io_prep_async_link(struct io_kiocb *req)
{
struct io_kiocb *cur;
if (req->flags & REQ_F_LINK_TIMEOUT) {
struct io_ring_ctx *ctx = req->ctx;
spin_lock_irq(&ctx->timeout_lock);
io_for_each_link(cur, req)
io_prep_async_work(cur);
spin_unlock_irq(&ctx->timeout_lock);
} else {
io_for_each_link(cur, req)
io_prep_async_work(cur);
}
}
void io_queue_iowq(struct io_kiocb *req, struct io_tw_state *ts_dont_use)
{
struct io_kiocb *link = io_prep_linked_timeout(req);
struct io_uring_task *tctx = req->task->io_uring;
BUG_ON(!tctx);
BUG_ON(!tctx->io_wq);
/* init ->work of the whole link before punting */
io_prep_async_link(req);
/*
* Not expected to happen, but if we do have a bug where this _can_
* happen, catch it here and ensure the request is marked as
* canceled. That will make io-wq go through the usual work cancel
* procedure rather than attempt to run this request (or create a new
* worker for it).
*/
if (WARN_ON_ONCE(!same_thread_group(req->task, current)))
req->work.flags |= IO_WQ_WORK_CANCEL;
trace_io_uring_queue_async_work(req, io_wq_is_hashed(&req->work));
io_wq_enqueue(tctx->io_wq, &req->work);
io_uring: fix recursive completion locking on oveflow flush syszbot reports a scenario where we recurse on the completion lock when flushing an overflow: 1 lock held by syz-executor287/6816: #0: ffff888093cdb4d8 (&ctx->completion_lock){....}-{2:2}, at: io_cqring_overflow_flush+0xc6/0xab0 fs/io_uring.c:1333 stack backtrace: CPU: 1 PID: 6816 Comm: syz-executor287 Not tainted 5.8.0-syzkaller #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1f0/0x31e lib/dump_stack.c:118 print_deadlock_bug kernel/locking/lockdep.c:2391 [inline] check_deadlock kernel/locking/lockdep.c:2432 [inline] validate_chain+0x69a4/0x88a0 kernel/locking/lockdep.c:3202 __lock_acquire+0x1161/0x2ab0 kernel/locking/lockdep.c:4426 lock_acquire+0x160/0x730 kernel/locking/lockdep.c:5005 __raw_spin_lock_irq include/linux/spinlock_api_smp.h:128 [inline] _raw_spin_lock_irq+0x67/0x80 kernel/locking/spinlock.c:167 spin_lock_irq include/linux/spinlock.h:379 [inline] io_queue_linked_timeout fs/io_uring.c:5928 [inline] __io_queue_async_work fs/io_uring.c:1192 [inline] __io_queue_deferred+0x36a/0x790 fs/io_uring.c:1237 io_cqring_overflow_flush+0x774/0xab0 fs/io_uring.c:1359 io_ring_ctx_wait_and_kill+0x2a1/0x570 fs/io_uring.c:7808 io_uring_release+0x59/0x70 fs/io_uring.c:7829 __fput+0x34f/0x7b0 fs/file_table.c:281 task_work_run+0x137/0x1c0 kernel/task_work.c:135 exit_task_work include/linux/task_work.h:25 [inline] do_exit+0x5f3/0x1f20 kernel/exit.c:806 do_group_exit+0x161/0x2d0 kernel/exit.c:903 __do_sys_exit_group+0x13/0x20 kernel/exit.c:914 __se_sys_exit_group+0x10/0x10 kernel/exit.c:912 __x64_sys_exit_group+0x37/0x40 kernel/exit.c:912 do_syscall_64+0x31/0x70 arch/x86/entry/common.c:46 entry_SYSCALL_64_after_hwframe+0x44/0xa9 Fix this by passing back the link from __io_queue_async_work(), and then let the caller handle the queueing of the link. Take care to also punt the submission reference put to the caller, as we're holding the completion lock for the __io_queue_defer() case. Hence we need to mark the io_kiocb appropriately for that case. Reported-by: syzbot+996f91b6ec3812c48042@syzkaller.appspotmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-08-10 23:55:22 +08:00
if (link)
io_queue_linked_timeout(link);
}
static __cold void io_queue_deferred(struct io_ring_ctx *ctx)
{
while (!list_empty(&ctx->defer_list)) {
struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
struct io_defer_entry, list);
if (req_need_defer(de->req, de->seq))
break;
list_del_init(&de->list);
io_req_task_queue(de->req);
kfree(de);
}
}
static void io_eventfd_ops(struct rcu_head *rcu)
{
struct io_ev_fd *ev_fd = container_of(rcu, struct io_ev_fd, rcu);
int ops = atomic_xchg(&ev_fd->ops, 0);
if (ops & BIT(IO_EVENTFD_OP_SIGNAL_BIT))
eventfd_signal_mask(ev_fd->cq_ev_fd, 1, EPOLL_URING_WAKE);
/* IO_EVENTFD_OP_FREE_BIT may not be set here depending on callback
* ordering in a race but if references are 0 we know we have to free
* it regardless.
*/
if (atomic_dec_and_test(&ev_fd->refs)) {
eventfd_ctx_put(ev_fd->cq_ev_fd);
kfree(ev_fd);
}
}
static void io_eventfd_signal(struct io_ring_ctx *ctx)
{
struct io_ev_fd *ev_fd = NULL;
rcu_read_lock();
/*
* rcu_dereference ctx->io_ev_fd once and use it for both for checking
* and eventfd_signal
*/
ev_fd = rcu_dereference(ctx->io_ev_fd);
/*
* Check again if ev_fd exists incase an io_eventfd_unregister call
* completed between the NULL check of ctx->io_ev_fd at the start of
* the function and rcu_read_lock.
*/
if (unlikely(!ev_fd))
goto out;
if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
goto out;
if (ev_fd->eventfd_async && !io_wq_current_is_worker())
goto out;
if (likely(eventfd_signal_allowed())) {
eventfd_signal_mask(ev_fd->cq_ev_fd, 1, EPOLL_URING_WAKE);
} else {
atomic_inc(&ev_fd->refs);
if (!atomic_fetch_or(BIT(IO_EVENTFD_OP_SIGNAL_BIT), &ev_fd->ops))
call_rcu_hurry(&ev_fd->rcu, io_eventfd_ops);
else
atomic_dec(&ev_fd->refs);
}
out:
rcu_read_unlock();
}
static void io_eventfd_flush_signal(struct io_ring_ctx *ctx)
{
bool skip;
spin_lock(&ctx->completion_lock);
/*
* Eventfd should only get triggered when at least one event has been
* posted. Some applications rely on the eventfd notification count
* only changing IFF a new CQE has been added to the CQ ring. There's
* no depedency on 1:1 relationship between how many times this
* function is called (and hence the eventfd count) and number of CQEs
* posted to the CQ ring.
*/
skip = ctx->cached_cq_tail == ctx->evfd_last_cq_tail;
ctx->evfd_last_cq_tail = ctx->cached_cq_tail;
spin_unlock(&ctx->completion_lock);
if (skip)
return;
io_eventfd_signal(ctx);
}
void __io_commit_cqring_flush(struct io_ring_ctx *ctx)
{
if (ctx->poll_activated)
io_poll_wq_wake(ctx);
if (ctx->off_timeout_used)
io_flush_timeouts(ctx);
if (ctx->drain_active) {
spin_lock(&ctx->completion_lock);
io_queue_deferred(ctx);
spin_unlock(&ctx->completion_lock);
}
if (ctx->has_evfd)
io_eventfd_flush_signal(ctx);
}
static inline void __io_cq_lock(struct io_ring_ctx *ctx)
{
if (!ctx->lockless_cq)
spin_lock(&ctx->completion_lock);
}
static inline void io_cq_lock(struct io_ring_ctx *ctx)
__acquires(ctx->completion_lock)
{
spin_lock(&ctx->completion_lock);
}
static inline void __io_cq_unlock_post(struct io_ring_ctx *ctx)
{
io_commit_cqring(ctx);
if (!ctx->task_complete) {
if (!ctx->lockless_cq)
spin_unlock(&ctx->completion_lock);
/* IOPOLL rings only need to wake up if it's also SQPOLL */
if (!ctx->syscall_iopoll)
io_cqring_wake(ctx);
}
io_commit_cqring_flush(ctx);
}
static void io_cq_unlock_post(struct io_ring_ctx *ctx)
__releases(ctx->completion_lock)
{
io_commit_cqring(ctx);
spin_unlock(&ctx->completion_lock);
io_cqring_wake(ctx);
io_commit_cqring_flush(ctx);
}
/* Returns true if there are no backlogged entries after the flush */
static void io_cqring_overflow_kill(struct io_ring_ctx *ctx)
{
struct io_overflow_cqe *ocqe;
LIST_HEAD(list);
spin_lock(&ctx->completion_lock);
list_splice_init(&ctx->cq_overflow_list, &list);
clear_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq);
spin_unlock(&ctx->completion_lock);
while (!list_empty(&list)) {
ocqe = list_first_entry(&list, struct io_overflow_cqe, list);
list_del(&ocqe->list);
kfree(ocqe);
}
}
static void __io_cqring_overflow_flush(struct io_ring_ctx *ctx)
{
size_t cqe_size = sizeof(struct io_uring_cqe);
if (__io_cqring_events(ctx) == ctx->cq_entries)
return;
if (ctx->flags & IORING_SETUP_CQE32)
cqe_size <<= 1;
io_cq_lock(ctx);
while (!list_empty(&ctx->cq_overflow_list)) {
struct io_uring_cqe *cqe;
struct io_overflow_cqe *ocqe;
if (!io_get_cqe_overflow(ctx, &cqe, true))
break;
ocqe = list_first_entry(&ctx->cq_overflow_list,
struct io_overflow_cqe, list);
memcpy(cqe, &ocqe->cqe, cqe_size);
list_del(&ocqe->list);
kfree(ocqe);
}
if (list_empty(&ctx->cq_overflow_list)) {
clear_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq);
atomic_andnot(IORING_SQ_CQ_OVERFLOW, &ctx->rings->sq_flags);
}
io_cq_unlock_post(ctx);
}
static void io_cqring_do_overflow_flush(struct io_ring_ctx *ctx)
{
/* iopoll syncs against uring_lock, not completion_lock */
if (ctx->flags & IORING_SETUP_IOPOLL)
mutex_lock(&ctx->uring_lock);
__io_cqring_overflow_flush(ctx);
if (ctx->flags & IORING_SETUP_IOPOLL)
mutex_unlock(&ctx->uring_lock);
}
static void io_cqring_overflow_flush(struct io_ring_ctx *ctx)
{
if (test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq))
io_cqring_do_overflow_flush(ctx);
}
/* can be called by any task */
static void io_put_task_remote(struct task_struct *task)
{
struct io_uring_task *tctx = task->io_uring;
percpu_counter_sub(&tctx->inflight, 1);
if (unlikely(atomic_read(&tctx->in_cancel)))
wake_up(&tctx->wait);
put_task_struct(task);
}
/* used by a task to put its own references */
static void io_put_task_local(struct task_struct *task)
{
task->io_uring->cached_refs++;
}
/* must to be called somewhat shortly after putting a request */
static inline void io_put_task(struct task_struct *task)
{
if (likely(task == current))
io_put_task_local(task);
else
io_put_task_remote(task);
}
void io_task_refs_refill(struct io_uring_task *tctx)
{
unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR;
percpu_counter_add(&tctx->inflight, refill);
refcount_add(refill, &current->usage);
tctx->cached_refs += refill;
}
static __cold void io_uring_drop_tctx_refs(struct task_struct *task)
{
struct io_uring_task *tctx = task->io_uring;
unsigned int refs = tctx->cached_refs;
if (refs) {
tctx->cached_refs = 0;
percpu_counter_sub(&tctx->inflight, refs);
put_task_struct_many(task, refs);
}
}
static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
s32 res, u32 cflags, u64 extra1, u64 extra2)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_overflow_cqe *ocqe;
size_t ocq_size = sizeof(struct io_overflow_cqe);
bool is_cqe32 = (ctx->flags & IORING_SETUP_CQE32);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
lockdep_assert_held(&ctx->completion_lock);
if (is_cqe32)
ocq_size += sizeof(struct io_uring_cqe);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ocqe = kmalloc(ocq_size, GFP_ATOMIC | __GFP_ACCOUNT);
trace_io_uring_cqe_overflow(ctx, user_data, res, cflags, ocqe);
if (!ocqe) {
/*
* If we're in ring overflow flush mode, or in task cancel mode,
* or cannot allocate an overflow entry, then we need to drop it
* on the floor.
*/
io_account_cq_overflow(ctx);
set_bit(IO_CHECK_CQ_DROPPED_BIT, &ctx->check_cq);
return false;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
if (list_empty(&ctx->cq_overflow_list)) {
set_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq);
atomic_or(IORING_SQ_CQ_OVERFLOW, &ctx->rings->sq_flags);
}
ocqe->cqe.user_data = user_data;
ocqe->cqe.res = res;
ocqe->cqe.flags = cflags;
if (is_cqe32) {
ocqe->cqe.big_cqe[0] = extra1;
ocqe->cqe.big_cqe[1] = extra2;
}
list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
return true;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
void io_req_cqe_overflow(struct io_kiocb *req)
{
io_cqring_event_overflow(req->ctx, req->cqe.user_data,
req->cqe.res, req->cqe.flags,
req->big_cqe.extra1, req->big_cqe.extra2);
memset(&req->big_cqe, 0, sizeof(req->big_cqe));
}
/*
* writes to the cq entry need to come after reading head; the
* control dependency is enough as we're using WRITE_ONCE to
* fill the cq entry
*/
bool io_cqe_cache_refill(struct io_ring_ctx *ctx, bool overflow)
{
struct io_rings *rings = ctx->rings;
unsigned int off = ctx->cached_cq_tail & (ctx->cq_entries - 1);
unsigned int free, queued, len;
/*
* Posting into the CQ when there are pending overflowed CQEs may break
* ordering guarantees, which will affect links, F_MORE users and more.
* Force overflow the completion.
*/
if (!overflow && (ctx->check_cq & BIT(IO_CHECK_CQ_OVERFLOW_BIT)))
return false;
/* userspace may cheat modifying the tail, be safe and do min */
queued = min(__io_cqring_events(ctx), ctx->cq_entries);
free = ctx->cq_entries - queued;
/* we need a contiguous range, limit based on the current array offset */
len = min(free, ctx->cq_entries - off);
if (!len)
return false;
if (ctx->flags & IORING_SETUP_CQE32) {
off <<= 1;
len <<= 1;
}
ctx->cqe_cached = &rings->cqes[off];
ctx->cqe_sentinel = ctx->cqe_cached + len;
return true;
}
static bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data, s32 res,
u32 cflags)
{
struct io_uring_cqe *cqe;
ctx->cq_extra++;
/*
* If we can't get a cq entry, userspace overflowed the
* submission (by quite a lot). Increment the overflow count in
* the ring.
*/
if (likely(io_get_cqe(ctx, &cqe))) {
trace_io_uring_complete(ctx, NULL, user_data, res, cflags, 0, 0);
WRITE_ONCE(cqe->user_data, user_data);
WRITE_ONCE(cqe->res, res);
WRITE_ONCE(cqe->flags, cflags);
if (ctx->flags & IORING_SETUP_CQE32) {
WRITE_ONCE(cqe->big_cqe[0], 0);
WRITE_ONCE(cqe->big_cqe[1], 0);
}
return true;
}
return false;
}
static void __io_flush_post_cqes(struct io_ring_ctx *ctx)
__must_hold(&ctx->uring_lock)
{
struct io_submit_state *state = &ctx->submit_state;
unsigned int i;
lockdep_assert_held(&ctx->uring_lock);
for (i = 0; i < state->cqes_count; i++) {
struct io_uring_cqe *cqe = &ctx->completion_cqes[i];
if (!io_fill_cqe_aux(ctx, cqe->user_data, cqe->res, cqe->flags)) {
if (ctx->lockless_cq) {
spin_lock(&ctx->completion_lock);
io_cqring_event_overflow(ctx, cqe->user_data,
cqe->res, cqe->flags, 0, 0);
spin_unlock(&ctx->completion_lock);
} else {
io_cqring_event_overflow(ctx, cqe->user_data,
cqe->res, cqe->flags, 0, 0);
}
}
}
state->cqes_count = 0;
}
static bool __io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags,
bool allow_overflow)
{
bool filled;
io_cq_lock(ctx);
filled = io_fill_cqe_aux(ctx, user_data, res, cflags);
if (!filled && allow_overflow)
filled = io_cqring_event_overflow(ctx, user_data, res, cflags, 0, 0);
io_cq_unlock_post(ctx);
return filled;
}
bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags)
{
return __io_post_aux_cqe(ctx, user_data, res, cflags, true);
}
/*
* A helper for multishot requests posting additional CQEs.
* Should only be used from a task_work including IO_URING_F_MULTISHOT.
*/
bool io_fill_cqe_req_aux(struct io_kiocb *req, bool defer, s32 res, u32 cflags)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_ring_ctx *ctx = req->ctx;
u64 user_data = req->cqe.user_data;
struct io_uring_cqe *cqe;
if (!defer)
return __io_post_aux_cqe(ctx, user_data, res, cflags, false);
lockdep_assert_held(&ctx->uring_lock);
if (ctx->submit_state.cqes_count == ARRAY_SIZE(ctx->completion_cqes)) {
__io_cq_lock(ctx);
__io_flush_post_cqes(ctx);
/* no need to flush - flush is deferred */
__io_cq_unlock_post(ctx);
}
/* For defered completions this is not as strict as it is otherwise,
* however it's main job is to prevent unbounded posted completions,
* and in that it works just as well.
*/
if (test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq))
return false;
cqe = &ctx->completion_cqes[ctx->submit_state.cqes_count++];
cqe->user_data = user_data;
cqe->res = res;
cqe->flags = cflags;
return true;
}
static void __io_req_complete_post(struct io_kiocb *req, unsigned issue_flags)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_ring_ctx *ctx = req->ctx;
struct io_rsrc_node *rsrc_node = NULL;
io_cq_lock(ctx);
if (!(req->flags & REQ_F_CQE_SKIP)) {
if (!io_fill_cqe_req(ctx, req))
io_req_cqe_overflow(req);
}
/*
* If we're the last reference to this request, add to our locked
* free_list cache.
*/
if (req_ref_put_and_test(req)) {
if (req->flags & IO_REQ_LINK_FLAGS) {
if (req->flags & IO_DISARM_MASK)
io_disarm_next(req);
if (req->link) {
io_req_task_queue(req->link);
req->link = NULL;
}
}
io_put_kbuf_comp(req);
if (unlikely(req->flags & IO_REQ_CLEAN_FLAGS))
io_clean_op(req);
io_put_file(req);
rsrc_node = req->rsrc_node;
/*
* Selected buffer deallocation in io_clean_op() assumes that
* we don't hold ->completion_lock. Clean them here to avoid
* deadlocks.
*/
io_put_task_remote(req->task);
wq_list_add_head(&req->comp_list, &ctx->locked_free_list);
ctx->locked_free_nr++;
}
io_cq_unlock_post(ctx);
if (rsrc_node) {
io_ring_submit_lock(ctx, issue_flags);
io_put_rsrc_node(ctx, rsrc_node);
io_ring_submit_unlock(ctx, issue_flags);
}
}
void io_req_complete_post(struct io_kiocb *req, unsigned issue_flags)
{
if (req->ctx->task_complete && req->ctx->submitter_task != current) {
req->io_task_work.func = io_req_task_complete;
io_req_task_work_add(req);
} else if (!(issue_flags & IO_URING_F_UNLOCKED) ||
!(req->ctx->flags & IORING_SETUP_IOPOLL)) {
__io_req_complete_post(req, issue_flags);
} else {
struct io_ring_ctx *ctx = req->ctx;
mutex_lock(&ctx->uring_lock);
__io_req_complete_post(req, issue_flags & ~IO_URING_F_UNLOCKED);
mutex_unlock(&ctx->uring_lock);
}
}
void io_req_defer_failed(struct io_kiocb *req, s32 res)
__must_hold(&ctx->uring_lock)
{
const struct io_cold_def *def = &io_cold_defs[req->opcode];
lockdep_assert_held(&req->ctx->uring_lock);
req_set_fail(req);
io_req_set_res(req, res, io_put_kbuf(req, IO_URING_F_UNLOCKED));
if (def->fail)
def->fail(req);
io_req_complete_defer(req);
}
/*
* Don't initialise the fields below on every allocation, but do that in
* advance and keep them valid across allocations.
*/
static void io_preinit_req(struct io_kiocb *req, struct io_ring_ctx *ctx)
{
req->ctx = ctx;
req->link = NULL;
req->async_data = NULL;
/* not necessary, but safer to zero */
memset(&req->cqe, 0, sizeof(req->cqe));
memset(&req->big_cqe, 0, sizeof(req->big_cqe));
}
static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx,
struct io_submit_state *state)
{
spin_lock(&ctx->completion_lock);
wq_list_splice(&ctx->locked_free_list, &state->free_list);
ctx->locked_free_nr = 0;
spin_unlock(&ctx->completion_lock);
}
io_uring: remove submission references Requests are by default given with two references, submission and completion. Completion references are straightforward, they represent request ownership and are put when a request is completed or so. Submission references are a bit more trickier. They're needed when io_issue_sqe() followed deep into the submission stack (e.g. in fs, block, drivers, etc.), request may have given away for concurrent execution or already completed, and the code unwinding back to io_issue_sqe() may be accessing some pieces of our requests, e.g. file or iov. Now, we prevent such async/in-depth completions by pushing requests through task_work. Punting to io-wq is also done through task_works, apart from a couple of cases with a pretty well known context. So, there're two cases: 1) io_issue_sqe() from the task context and protected by ->uring_lock. Either requests return back to io_uring or handed to task_work, which won't be executed because we're currently controlling that task. So, we can be sure that requests are staying alive all the time and we don't need submission references to pin them. 2) io_issue_sqe() from io-wq, which doesn't hold the mutex. The role of submission reference is played by io-wq reference, which is put by io_wq_submit_work(). Hence, it should be fine. Considering that, we can carefully kill the submission reference. Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/6b68f1c763229a590f2a27148aee77767a8d7750.1628705069.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-08-12 02:28:29 +08:00
/*
* A request might get retired back into the request caches even before opcode
* handlers and io_issue_sqe() are done with it, e.g. inline completion path.
* Because of that, io_alloc_req() should be called only under ->uring_lock
* and with extra caution to not get a request that is still worked on.
*/
__cold bool __io_alloc_req_refill(struct io_ring_ctx *ctx)
io_uring: remove submission references Requests are by default given with two references, submission and completion. Completion references are straightforward, they represent request ownership and are put when a request is completed or so. Submission references are a bit more trickier. They're needed when io_issue_sqe() followed deep into the submission stack (e.g. in fs, block, drivers, etc.), request may have given away for concurrent execution or already completed, and the code unwinding back to io_issue_sqe() may be accessing some pieces of our requests, e.g. file or iov. Now, we prevent such async/in-depth completions by pushing requests through task_work. Punting to io-wq is also done through task_works, apart from a couple of cases with a pretty well known context. So, there're two cases: 1) io_issue_sqe() from the task context and protected by ->uring_lock. Either requests return back to io_uring or handed to task_work, which won't be executed because we're currently controlling that task. So, we can be sure that requests are staying alive all the time and we don't need submission references to pin them. 2) io_issue_sqe() from io-wq, which doesn't hold the mutex. The role of submission reference is played by io-wq reference, which is put by io_wq_submit_work(). Hence, it should be fine. Considering that, we can carefully kill the submission reference. Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/6b68f1c763229a590f2a27148aee77767a8d7750.1628705069.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-08-12 02:28:29 +08:00
__must_hold(&ctx->uring_lock)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
void *reqs[IO_REQ_ALLOC_BATCH];
int ret, i;
/*
* If we have more than a batch's worth of requests in our IRQ side
* locked cache, grab the lock and move them over to our submission
* side cache.
*/
if (data_race(ctx->locked_free_nr) > IO_COMPL_BATCH) {
io_flush_cached_locked_reqs(ctx, &ctx->submit_state);
if (!io_req_cache_empty(ctx))
return true;
}
ret = kmem_cache_alloc_bulk(req_cachep, gfp, ARRAY_SIZE(reqs), reqs);
/*
* Bulk alloc is all-or-nothing. If we fail to get a batch,
* retry single alloc to be on the safe side.
*/
if (unlikely(ret <= 0)) {
reqs[0] = kmem_cache_alloc(req_cachep, gfp);
if (!reqs[0])
return false;
ret = 1;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
percpu_ref_get_many(&ctx->refs, ret);
for (i = 0; i < ret; i++) {
struct io_kiocb *req = reqs[i];
io_preinit_req(req, ctx);
io_req_add_to_cache(req, ctx);
}
return true;
}
__cold void io_free_req(struct io_kiocb *req)
{
/* refs were already put, restore them for io_req_task_complete() */
req->flags &= ~REQ_F_REFCOUNT;
/* we only want to free it, don't post CQEs */
req->flags |= REQ_F_CQE_SKIP;
req->io_task_work.func = io_req_task_complete;
io_req_task_work_add(req);
}
static void __io_req_find_next_prep(struct io_kiocb *req)
{
struct io_ring_ctx *ctx = req->ctx;
spin_lock(&ctx->completion_lock);
io_disarm_next(req);
spin_unlock(&ctx->completion_lock);
}
static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
{
struct io_kiocb *nxt;
/*
* If LINK is set, we have dependent requests in this chain. If we
* didn't fail this request, queue the first one up, moving any other
* dependencies to the next request. In case of failure, fail the rest
* of the chain.
*/
if (unlikely(req->flags & IO_DISARM_MASK))
__io_req_find_next_prep(req);
nxt = req->link;
req->link = NULL;
return nxt;
}
static void ctx_flush_and_put(struct io_ring_ctx *ctx, struct io_tw_state *ts)
2021-03-01 06:04:53 +08:00
{
if (!ctx)
return;
if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
atomic_andnot(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
if (ts->locked) {
io_submit_flush_completions(ctx);
2021-03-01 06:04:53 +08:00
mutex_unlock(&ctx->uring_lock);
ts->locked = false;
2021-03-01 06:04:53 +08:00
}
percpu_ref_put(&ctx->refs);
}
static unsigned int handle_tw_list(struct llist_node *node,
struct io_ring_ctx **ctx,
struct io_tw_state *ts,
struct llist_node *last)
{
unsigned int count = 0;
while (node && node != last) {
struct llist_node *next = node->next;
struct io_kiocb *req = container_of(node, struct io_kiocb,
io_task_work.node);
prefetch(container_of(next, struct io_kiocb, io_task_work.node));
if (req->ctx != *ctx) {
ctx_flush_and_put(*ctx, ts);
*ctx = req->ctx;
/* if not contended, grab and improve batching */
ts->locked = mutex_trylock(&(*ctx)->uring_lock);
percpu_ref_get(&(*ctx)->refs);
}
INDIRECT_CALL_2(req->io_task_work.func,
io_poll_task_func, io_req_rw_complete,
req, ts);
node = next;
count++;
if (unlikely(need_resched())) {
ctx_flush_and_put(*ctx, ts);
*ctx = NULL;
cond_resched();
}
}
return count;
}
/**
* io_llist_xchg - swap all entries in a lock-less list
* @head: the head of lock-less list to delete all entries
* @new: new entry as the head of the list
*
* If list is empty, return NULL, otherwise, return the pointer to the first entry.
* The order of entries returned is from the newest to the oldest added one.
*/
static inline struct llist_node *io_llist_xchg(struct llist_head *head,
struct llist_node *new)
{
return xchg(&head->first, new);
}
/**
* io_llist_cmpxchg - possibly swap all entries in a lock-less list
* @head: the head of lock-less list to delete all entries
* @old: expected old value of the first entry of the list
* @new: new entry as the head of the list
*
* perform a cmpxchg on the first entry of the list.
*/
static inline struct llist_node *io_llist_cmpxchg(struct llist_head *head,
struct llist_node *old,
struct llist_node *new)
{
return cmpxchg(&head->first, old, new);
}
static __cold void io_fallback_tw(struct io_uring_task *tctx, bool sync)
{
struct llist_node *node = llist_del_all(&tctx->task_list);
struct io_ring_ctx *last_ctx = NULL;
struct io_kiocb *req;
while (node) {
req = container_of(node, struct io_kiocb, io_task_work.node);
node = node->next;
if (sync && last_ctx != req->ctx) {
if (last_ctx) {
flush_delayed_work(&last_ctx->fallback_work);
percpu_ref_put(&last_ctx->refs);
}
last_ctx = req->ctx;
percpu_ref_get(&last_ctx->refs);
}
if (llist_add(&req->io_task_work.node,
&req->ctx->fallback_llist))
schedule_delayed_work(&req->ctx->fallback_work, 1);
}
if (last_ctx) {
flush_delayed_work(&last_ctx->fallback_work);
percpu_ref_put(&last_ctx->refs);
}
}
void tctx_task_work(struct callback_head *cb)
io_uring: use task_work for links if possible Currently links are always done in an async fashion, unless we catch them inline after we successfully complete a request without having to resort to blocking. This isn't necessarily the most efficient approach, it'd be more ideal if we could just use the task_work handling for this. Outside of saving an async jump, we can also do less prep work for these kinds of requests. Running dependent links from the task_work handler yields some nice performance benefits. As an example, examples/link-cp from the liburing repository uses read+write links to implement a copy operation. Without this patch, the a cache fold 4G file read from a VM runs in about 3 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.986s user 0m0.051s sys 0m2.843s and a subsequent cache hot run looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.898s user 0m0.069s sys 0m0.797s With this patch in place, the cold case takes about 2.4 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.400s user 0m0.020s sys 0m2.366s and the cache hot case looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.676s user 0m0.010s sys 0m0.665s As expected, the (mostly) cache hot case yields the biggest improvement, running about 25% faster with this change, while the cache cold case yields about a 20% increase in performance. Outside of the performance increase, we're using less CPU as well, as we're not using the async offload threads at all for this anymore. Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-06-26 05:39:59 +08:00
{
struct io_tw_state ts = {};
struct io_ring_ctx *ctx = NULL;
struct io_uring_task *tctx = container_of(cb, struct io_uring_task,
task_work);
struct llist_node fake = {};
struct llist_node *node;
unsigned int loops = 0;
unsigned int count = 0;
if (unlikely(current->flags & PF_EXITING)) {
io_fallback_tw(tctx, true);
return;
}
do {
loops++;
node = io_llist_xchg(&tctx->task_list, &fake);
count += handle_tw_list(node, &ctx, &ts, &fake);
/* skip expensive cmpxchg if there are items in the list */
if (READ_ONCE(tctx->task_list.first) != &fake)
continue;
if (ts.locked && !wq_list_empty(&ctx->submit_state.compl_reqs)) {
io_submit_flush_completions(ctx);
if (READ_ONCE(tctx->task_list.first) != &fake)
continue;
}
node = io_llist_cmpxchg(&tctx->task_list, &fake, NULL);
} while (node != &fake);
ctx_flush_and_put(ctx, &ts);
/* relaxed read is enough as only the task itself sets ->in_cancel */
if (unlikely(atomic_read(&tctx->in_cancel)))
io_uring_drop_tctx_refs(current);
trace_io_uring_task_work_run(tctx, count, loops);
}
static inline void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
{
struct io_ring_ctx *ctx = req->ctx;
unsigned nr_wait, nr_tw, nr_tw_prev;
struct llist_node *first;
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK))
flags &= ~IOU_F_TWQ_LAZY_WAKE;
first = READ_ONCE(ctx->work_llist.first);
do {
nr_tw_prev = 0;
if (first) {
struct io_kiocb *first_req = container_of(first,
struct io_kiocb,
io_task_work.node);
/*
* Might be executed at any moment, rely on
* SLAB_TYPESAFE_BY_RCU to keep it alive.
*/
nr_tw_prev = READ_ONCE(first_req->nr_tw);
}
nr_tw = nr_tw_prev + 1;
/* Large enough to fail the nr_wait comparison below */
if (!(flags & IOU_F_TWQ_LAZY_WAKE))
nr_tw = INT_MAX;
req->nr_tw = nr_tw;
req->io_task_work.node.next = first;
} while (!try_cmpxchg(&ctx->work_llist.first, &first,
&req->io_task_work.node));
if (!first) {
if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
atomic_or(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
if (ctx->has_evfd)
io_eventfd_signal(ctx);
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
}
nr_wait = atomic_read(&ctx->cq_wait_nr);
/* no one is waiting */
if (!nr_wait)
return;
/* either not enough or the previous add has already woken it up */
if (nr_wait > nr_tw || nr_tw_prev >= nr_wait)
return;
/* pairs with set_current_state() in io_cqring_wait() */
smp_mb__after_atomic();
wake_up_state(ctx->submitter_task, TASK_INTERRUPTIBLE);
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
}
static void io_req_normal_work_add(struct io_kiocb *req)
{
struct io_uring_task *tctx = req->task->io_uring;
struct io_ring_ctx *ctx = req->ctx;
/* task_work already pending, we're done */
if (!llist_add(&req->io_task_work.node, &tctx->task_list))
return;
if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
atomic_or(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
if (likely(!task_work_add(req->task, &tctx->task_work, ctx->notify_method)))
return;
io_fallback_tw(tctx, false);
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
}
void __io_req_task_work_add(struct io_kiocb *req, unsigned flags)
{
if (req->ctx->flags & IORING_SETUP_DEFER_TASKRUN) {
rcu_read_lock();
io_req_local_work_add(req, flags);
rcu_read_unlock();
} else {
io_req_normal_work_add(req);
}
}
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
static void __cold io_move_task_work_from_local(struct io_ring_ctx *ctx)
{
struct llist_node *node;
node = llist_del_all(&ctx->work_llist);
while (node) {
struct io_kiocb *req = container_of(node, struct io_kiocb,
io_task_work.node);
node = node->next;
io_req_normal_work_add(req);
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
}
}
static int __io_run_local_work(struct io_ring_ctx *ctx, struct io_tw_state *ts)
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
{
struct llist_node *node;
unsigned int loops = 0;
int ret = 0;
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
if (WARN_ON_ONCE(ctx->submitter_task != current))
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
return -EEXIST;
if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
atomic_andnot(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
again:
io_uring: maintain ordering for DEFER_TASKRUN tw list We use lockless lists for the local and deferred task_work, which means that when we queue up events for processing, we ultimately process them in reverse order to how they were received. This usually doesn't matter, but for some cases, it does seem to make a big difference. Do the right thing and reverse the list before processing it, so that we know it's processed in the same order in which it was received. This makes a rather big difference for some medium load network tests, where consistency of performance was a bit all over the place. Here's a case that has 4 connections each doing two sends and receives: io_uring port=10002: rps:161.13k Bps: 1.45M idle=256ms io_uring port=10002: rps:107.27k Bps: 0.97M idle=413ms io_uring port=10002: rps:136.98k Bps: 1.23M idle=321ms io_uring port=10002: rps:155.58k Bps: 1.40M idle=268ms and after the change: io_uring port=10002: rps:205.48k Bps: 1.85M idle=140ms user=40ms io_uring port=10002: rps:203.57k Bps: 1.83M idle=139ms user=20ms io_uring port=10002: rps:218.79k Bps: 1.97M idle=106ms user=30ms io_uring port=10002: rps:217.88k Bps: 1.96M idle=110ms user=20ms io_uring port=10002: rps:222.31k Bps: 2.00M idle=101ms user=0ms io_uring port=10002: rps:218.74k Bps: 1.97M idle=102ms user=20ms io_uring port=10002: rps:208.43k Bps: 1.88M idle=125ms user=40ms using more of the time to actually process work rather than sitting idle. No effects have been observed at the peak end of the spectrum, where performance is still the same even with deep batch depths (and hence more items to sort). Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-05-19 23:51:31 +08:00
/*
* llists are in reverse order, flip it back the right way before
* running the pending items.
*/
node = llist_reverse_order(io_llist_xchg(&ctx->work_llist, NULL));
while (node) {
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
struct llist_node *next = node->next;
struct io_kiocb *req = container_of(node, struct io_kiocb,
io_task_work.node);
prefetch(container_of(next, struct io_kiocb, io_task_work.node));
INDIRECT_CALL_2(req->io_task_work.func,
io_poll_task_func, io_req_rw_complete,
req, ts);
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
ret++;
node = next;
}
loops++;
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
if (!llist_empty(&ctx->work_llist))
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
goto again;
if (ts->locked) {
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
io_submit_flush_completions(ctx);
if (!llist_empty(&ctx->work_llist))
goto again;
}
trace_io_uring_local_work_run(ctx, ret, loops);
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
return ret;
}
static inline int io_run_local_work_locked(struct io_ring_ctx *ctx)
{
struct io_tw_state ts = { .locked = true, };
int ret;
if (llist_empty(&ctx->work_llist))
return 0;
ret = __io_run_local_work(ctx, &ts);
/* shouldn't happen! */
if (WARN_ON_ONCE(!ts.locked))
mutex_lock(&ctx->uring_lock);
return ret;
}
static int io_run_local_work(struct io_ring_ctx *ctx)
{
struct io_tw_state ts = {};
int ret;
ts.locked = mutex_trylock(&ctx->uring_lock);
ret = __io_run_local_work(ctx, &ts);
if (ts.locked)
mutex_unlock(&ctx->uring_lock);
return ret;
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
}
static void io_req_task_cancel(struct io_kiocb *req, struct io_tw_state *ts)
io_uring: use task_work for links if possible Currently links are always done in an async fashion, unless we catch them inline after we successfully complete a request without having to resort to blocking. This isn't necessarily the most efficient approach, it'd be more ideal if we could just use the task_work handling for this. Outside of saving an async jump, we can also do less prep work for these kinds of requests. Running dependent links from the task_work handler yields some nice performance benefits. As an example, examples/link-cp from the liburing repository uses read+write links to implement a copy operation. Without this patch, the a cache fold 4G file read from a VM runs in about 3 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.986s user 0m0.051s sys 0m2.843s and a subsequent cache hot run looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.898s user 0m0.069s sys 0m0.797s With this patch in place, the cold case takes about 2.4 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.400s user 0m0.020s sys 0m2.366s and the cache hot case looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.676s user 0m0.010s sys 0m0.665s As expected, the (mostly) cache hot case yields the biggest improvement, running about 25% faster with this change, while the cache cold case yields about a 20% increase in performance. Outside of the performance increase, we're using less CPU as well, as we're not using the async offload threads at all for this anymore. Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-06-26 05:39:59 +08:00
{
io_tw_lock(req->ctx, ts);
io_req_defer_failed(req, req->cqe.res);
io_uring: use task_work for links if possible Currently links are always done in an async fashion, unless we catch them inline after we successfully complete a request without having to resort to blocking. This isn't necessarily the most efficient approach, it'd be more ideal if we could just use the task_work handling for this. Outside of saving an async jump, we can also do less prep work for these kinds of requests. Running dependent links from the task_work handler yields some nice performance benefits. As an example, examples/link-cp from the liburing repository uses read+write links to implement a copy operation. Without this patch, the a cache fold 4G file read from a VM runs in about 3 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.986s user 0m0.051s sys 0m2.843s and a subsequent cache hot run looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.898s user 0m0.069s sys 0m0.797s With this patch in place, the cold case takes about 2.4 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.400s user 0m0.020s sys 0m2.366s and the cache hot case looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.676s user 0m0.010s sys 0m0.665s As expected, the (mostly) cache hot case yields the biggest improvement, running about 25% faster with this change, while the cache cold case yields about a 20% increase in performance. Outside of the performance increase, we're using less CPU as well, as we're not using the async offload threads at all for this anymore. Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-06-26 05:39:59 +08:00
}
void io_req_task_submit(struct io_kiocb *req, struct io_tw_state *ts)
io_uring: use task_work for links if possible Currently links are always done in an async fashion, unless we catch them inline after we successfully complete a request without having to resort to blocking. This isn't necessarily the most efficient approach, it'd be more ideal if we could just use the task_work handling for this. Outside of saving an async jump, we can also do less prep work for these kinds of requests. Running dependent links from the task_work handler yields some nice performance benefits. As an example, examples/link-cp from the liburing repository uses read+write links to implement a copy operation. Without this patch, the a cache fold 4G file read from a VM runs in about 3 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.986s user 0m0.051s sys 0m2.843s and a subsequent cache hot run looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.898s user 0m0.069s sys 0m0.797s With this patch in place, the cold case takes about 2.4 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.400s user 0m0.020s sys 0m2.366s and the cache hot case looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.676s user 0m0.010s sys 0m0.665s As expected, the (mostly) cache hot case yields the biggest improvement, running about 25% faster with this change, while the cache cold case yields about a 20% increase in performance. Outside of the performance increase, we're using less CPU as well, as we're not using the async offload threads at all for this anymore. Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-06-26 05:39:59 +08:00
{
io_tw_lock(req->ctx, ts);
/* req->task == current here, checking PF_EXITING is safe */
if (unlikely(req->task->flags & PF_EXITING))
io_req_defer_failed(req, -EFAULT);
else if (req->flags & REQ_F_FORCE_ASYNC)
io_queue_iowq(req, ts);
else
io_queue_sqe(req);
io_uring: use task_work for links if possible Currently links are always done in an async fashion, unless we catch them inline after we successfully complete a request without having to resort to blocking. This isn't necessarily the most efficient approach, it'd be more ideal if we could just use the task_work handling for this. Outside of saving an async jump, we can also do less prep work for these kinds of requests. Running dependent links from the task_work handler yields some nice performance benefits. As an example, examples/link-cp from the liburing repository uses read+write links to implement a copy operation. Without this patch, the a cache fold 4G file read from a VM runs in about 3 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.986s user 0m0.051s sys 0m2.843s and a subsequent cache hot run looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.898s user 0m0.069s sys 0m0.797s With this patch in place, the cold case takes about 2.4 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.400s user 0m0.020s sys 0m2.366s and the cache hot case looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.676s user 0m0.010s sys 0m0.665s As expected, the (mostly) cache hot case yields the biggest improvement, running about 25% faster with this change, while the cache cold case yields about a 20% increase in performance. Outside of the performance increase, we're using less CPU as well, as we're not using the async offload threads at all for this anymore. Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-06-26 05:39:59 +08:00
}
void io_req_task_queue_fail(struct io_kiocb *req, int ret)
io_uring: use task_work for links if possible Currently links are always done in an async fashion, unless we catch them inline after we successfully complete a request without having to resort to blocking. This isn't necessarily the most efficient approach, it'd be more ideal if we could just use the task_work handling for this. Outside of saving an async jump, we can also do less prep work for these kinds of requests. Running dependent links from the task_work handler yields some nice performance benefits. As an example, examples/link-cp from the liburing repository uses read+write links to implement a copy operation. Without this patch, the a cache fold 4G file read from a VM runs in about 3 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.986s user 0m0.051s sys 0m2.843s and a subsequent cache hot run looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.898s user 0m0.069s sys 0m0.797s With this patch in place, the cold case takes about 2.4 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.400s user 0m0.020s sys 0m2.366s and the cache hot case looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.676s user 0m0.010s sys 0m0.665s As expected, the (mostly) cache hot case yields the biggest improvement, running about 25% faster with this change, while the cache cold case yields about a 20% increase in performance. Outside of the performance increase, we're using less CPU as well, as we're not using the async offload threads at all for this anymore. Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-06-26 05:39:59 +08:00
{
io_req_set_res(req, ret, 0);
req->io_task_work.func = io_req_task_cancel;
io_req_task_work_add(req);
io_uring: use task_work for links if possible Currently links are always done in an async fashion, unless we catch them inline after we successfully complete a request without having to resort to blocking. This isn't necessarily the most efficient approach, it'd be more ideal if we could just use the task_work handling for this. Outside of saving an async jump, we can also do less prep work for these kinds of requests. Running dependent links from the task_work handler yields some nice performance benefits. As an example, examples/link-cp from the liburing repository uses read+write links to implement a copy operation. Without this patch, the a cache fold 4G file read from a VM runs in about 3 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.986s user 0m0.051s sys 0m2.843s and a subsequent cache hot run looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.898s user 0m0.069s sys 0m0.797s With this patch in place, the cold case takes about 2.4 seconds: $ time examples/link-cp /data/file /dev/null real 0m2.400s user 0m0.020s sys 0m2.366s and the cache hot case looks like this: $ time examples/link-cp /data/file /dev/null real 0m0.676s user 0m0.010s sys 0m0.665s As expected, the (mostly) cache hot case yields the biggest improvement, running about 25% faster with this change, while the cache cold case yields about a 20% increase in performance. Outside of the performance increase, we're using less CPU as well, as we're not using the async offload threads at all for this anymore. Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-06-26 05:39:59 +08:00
}
void io_req_task_queue(struct io_kiocb *req)
{
req->io_task_work.func = io_req_task_submit;
io_req_task_work_add(req);
}
void io_queue_next(struct io_kiocb *req)
{
struct io_kiocb *nxt = io_req_find_next(req);
if (nxt)
io_req_task_queue(nxt);
}
static void io_free_batch_list(struct io_ring_ctx *ctx,
struct io_wq_work_node *node)
__must_hold(&ctx->uring_lock)
{
do {
struct io_kiocb *req = container_of(node, struct io_kiocb,
comp_list);
if (unlikely(req->flags & IO_REQ_CLEAN_SLOW_FLAGS)) {
if (req->flags & REQ_F_REFCOUNT) {
node = req->comp_list.next;
if (!req_ref_put_and_test(req))
continue;
}
if ((req->flags & REQ_F_POLLED) && req->apoll) {
struct async_poll *apoll = req->apoll;
if (apoll->double_poll)
kfree(apoll->double_poll);
if (!io_alloc_cache_put(&ctx->apoll_cache, &apoll->cache))
kfree(apoll);
req->flags &= ~REQ_F_POLLED;
}
if (req->flags & IO_REQ_LINK_FLAGS)
io_queue_next(req);
if (unlikely(req->flags & IO_REQ_CLEAN_FLAGS))
io_clean_op(req);
}
io_put_file(req);
io_req_put_rsrc_locked(req, ctx);
io_put_task(req->task);
node = req->comp_list.next;
io_req_add_to_cache(req, ctx);
} while (node);
}
void __io_submit_flush_completions(struct io_ring_ctx *ctx)
__must_hold(&ctx->uring_lock)
{
struct io_submit_state *state = &ctx->submit_state;
struct io_wq_work_node *node;
__io_cq_lock(ctx);
/* must come first to preserve CQE ordering in failure cases */
if (state->cqes_count)
__io_flush_post_cqes(ctx);
__wq_list_for_each(node, &state->compl_reqs) {
struct io_kiocb *req = container_of(node, struct io_kiocb,
comp_list);
if (!(req->flags & REQ_F_CQE_SKIP) &&
unlikely(!io_fill_cqe_req(ctx, req))) {
if (ctx->lockless_cq) {
spin_lock(&ctx->completion_lock);
io_req_cqe_overflow(req);
spin_unlock(&ctx->completion_lock);
} else {
io_req_cqe_overflow(req);
}
}
}
__io_cq_unlock_post(ctx);
if (!wq_list_empty(&ctx->submit_state.compl_reqs)) {
io_free_batch_list(ctx, state->compl_reqs.first);
INIT_WQ_LIST(&state->compl_reqs);
}
}
static unsigned io_cqring_events(struct io_ring_ctx *ctx)
{
/* See comment at the top of this file */
smp_rmb();
return __io_cqring_events(ctx);
}
/*
* We can't just wait for polled events to come to us, we have to actively
* find and complete them.
*/
static __cold void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
{
if (!(ctx->flags & IORING_SETUP_IOPOLL))
return;
mutex_lock(&ctx->uring_lock);
while (!wq_list_empty(&ctx->iopoll_list)) {
/* let it sleep and repeat later if can't complete a request */
if (io_do_iopoll(ctx, true) == 0)
break;
/*
* Ensure we allow local-to-the-cpu processing to take place,
* in this case we need to ensure that we reap all events.
* Also let task_work, etc. to progress by releasing the mutex
*/
if (need_resched()) {
mutex_unlock(&ctx->uring_lock);
cond_resched();
mutex_lock(&ctx->uring_lock);
}
}
mutex_unlock(&ctx->uring_lock);
}
static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
{
unsigned int nr_events = 0;
unsigned long check_cq;
if (!io_allowed_run_tw(ctx))
return -EEXIST;
check_cq = READ_ONCE(ctx->check_cq);
if (unlikely(check_cq)) {
if (check_cq & BIT(IO_CHECK_CQ_OVERFLOW_BIT))
__io_cqring_overflow_flush(ctx);
/*
* Similarly do not spin if we have not informed the user of any
* dropped CQE.
*/
if (check_cq & BIT(IO_CHECK_CQ_DROPPED_BIT))
return -EBADR;
}
/*
* Don't enter poll loop if we already have events pending.
* If we do, we can potentially be spinning for commands that
* already triggered a CQE (eg in error).
*/
if (io_cqring_events(ctx))
return 0;
do {
int ret = 0;
/*
* If a submit got punted to a workqueue, we can have the
* application entering polling for a command before it gets
* issued. That app will hold the uring_lock for the duration
* of the poll right here, so we need to take a breather every
* now and then to ensure that the issue has a chance to add
* the poll to the issued list. Otherwise we can spin here
* forever, while the workqueue is stuck trying to acquire the
* very same mutex.
*/
if (wq_list_empty(&ctx->iopoll_list) ||
io_task_work_pending(ctx)) {
u32 tail = ctx->cached_cq_tail;
(void) io_run_local_work_locked(ctx);
if (task_work_pending(current) ||
wq_list_empty(&ctx->iopoll_list)) {
mutex_unlock(&ctx->uring_lock);
io_run_task_work();
mutex_lock(&ctx->uring_lock);
}
/* some requests don't go through iopoll_list */
if (tail != ctx->cached_cq_tail ||
wq_list_empty(&ctx->iopoll_list))
break;
}
ret = io_do_iopoll(ctx, !min);
if (unlikely(ret < 0))
return ret;
if (task_sigpending(current))
return -EINTR;
if (need_resched())
break;
nr_events += ret;
} while (nr_events < min);
return 0;
}
void io_req_task_complete(struct io_kiocb *req, struct io_tw_state *ts)
{
if (ts->locked)
io_req_complete_defer(req);
else
io_req_complete_post(req, IO_URING_F_UNLOCKED);
}
/*
* After the iocb has been issued, it's safe to be found on the poll list.
* Adding the kiocb to the list AFTER submission ensures that we don't
* find it from a io_do_iopoll() thread before the issuer is done
* accessing the kiocb cookie.
*/
static void io_iopoll_req_issued(struct io_kiocb *req, unsigned int issue_flags)
{
struct io_ring_ctx *ctx = req->ctx;
const bool needs_lock = issue_flags & IO_URING_F_UNLOCKED;
/* workqueue context doesn't hold uring_lock, grab it now */
if (unlikely(needs_lock))
mutex_lock(&ctx->uring_lock);
/*
* Track whether we have multiple files in our lists. This will impact
* how we do polling eventually, not spinning if we're on potentially
* different devices.
*/
if (wq_list_empty(&ctx->iopoll_list)) {
ctx->poll_multi_queue = false;
} else if (!ctx->poll_multi_queue) {
struct io_kiocb *list_req;
list_req = container_of(ctx->iopoll_list.first, struct io_kiocb,
comp_list);
if (list_req->file != req->file)
ctx->poll_multi_queue = true;
}
/*
* For fast devices, IO may have already completed. If it has, add
* it to the front so we find it first.
*/
if (READ_ONCE(req->iopoll_completed))
wq_list_add_head(&req->comp_list, &ctx->iopoll_list);
else
wq_list_add_tail(&req->comp_list, &ctx->iopoll_list);
io_uring: fix poll_list race for SETUP_IOPOLL|SETUP_SQPOLL After making ext4 support iopoll method: let ext4_file_operations's iopoll method be iomap_dio_iopoll(), we found fio can easily hang in fio_ioring_getevents() with below fio job: rm -f testfile; sync; sudo fio -name=fiotest -filename=testfile -iodepth=128 -thread -rw=write -ioengine=io_uring -hipri=1 -sqthread_poll=1 -direct=1 -bs=4k -size=10G -numjobs=8 -runtime=2000 -group_reporting with IORING_SETUP_SQPOLL and IORING_SETUP_IOPOLL enabled. There are two issues that results in this hang, one reason is that when IORING_SETUP_SQPOLL and IORING_SETUP_IOPOLL are enabled, fio does not use io_uring_enter to get completed events, it relies on kernel io_sq_thread to poll for completed events. Another reason is that there is a race: when io_submit_sqes() in io_sq_thread() submits a batch of sqes, variable 'inflight' will record the number of submitted reqs, then io_sq_thread will poll for reqs which have been added to poll_list. But note, if some previous reqs have been punted to io worker, these reqs will won't be in poll_list timely. io_sq_thread() will only poll for a part of previous submitted reqs, and then find poll_list is empty, reset variable 'inflight' to be zero. If app just waits these deferred reqs and does not wake up io_sq_thread again, then hang happens. For app that entirely relies on io_sq_thread to poll completed requests, let io_iopoll_req_issued() wake up io_sq_thread properly when adding new element to poll_list, and when io_sq_thread prepares to sleep, check whether poll_list is empty again, if not empty, continue to poll. Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-02-25 22:12:08 +08:00
if (unlikely(needs_lock)) {
/*
* If IORING_SETUP_SQPOLL is enabled, sqes are either handle
* in sq thread task context or in io worker task context. If
* current task context is sq thread, we don't need to check
* whether should wake up sq thread.
*/
if ((ctx->flags & IORING_SETUP_SQPOLL) &&
wq_has_sleeper(&ctx->sq_data->wait))
wake_up(&ctx->sq_data->wait);
mutex_unlock(&ctx->uring_lock);
}
}
unsigned int io_file_get_flags(struct file *file)
{
unsigned int res = 0;
if (S_ISREG(file_inode(file)->i_mode))
res |= REQ_F_ISREG;
if ((file->f_flags & O_NONBLOCK) || (file->f_mode & FMODE_NOWAIT))
res |= REQ_F_SUPPORT_NOWAIT;
return res;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
bool io_alloc_async_data(struct io_kiocb *req)
{
WARN_ON_ONCE(!io_cold_defs[req->opcode].async_size);
req->async_data = kmalloc(io_cold_defs[req->opcode].async_size, GFP_KERNEL);
if (req->async_data) {
req->flags |= REQ_F_ASYNC_DATA;
return false;
}
return true;
}
int io_req_prep_async(struct io_kiocb *req)
{
const struct io_cold_def *cdef = &io_cold_defs[req->opcode];
const struct io_issue_def *def = &io_issue_defs[req->opcode];
/* assign early for deferred execution for non-fixed file */
2023-02-28 12:54:59 +08:00
if (def->needs_file && !(req->flags & REQ_F_FIXED_FILE) && !req->file)
req->file = io_file_get_normal(req, req->cqe.fd);
if (!cdef->prep_async)
return 0;
if (WARN_ON_ONCE(req_has_async_data(req)))
return -EFAULT;
if (!def->manual_alloc) {
if (io_alloc_async_data(req))
return -EAGAIN;
}
return cdef->prep_async(req);
}
static u32 io_get_sequence(struct io_kiocb *req)
{
u32 seq = req->ctx->cached_sq_head;
struct io_kiocb *cur;
/* need original cached_sq_head, but it was increased for each req */
io_for_each_link(cur, req)
seq--;
return seq;
}
static __cold void io_drain_req(struct io_kiocb *req)
__must_hold(&ctx->uring_lock)
{
struct io_ring_ctx *ctx = req->ctx;
struct io_defer_entry *de;
int ret;
u32 seq = io_get_sequence(req);
/* Still need defer if there is pending req in defer list. */
spin_lock(&ctx->completion_lock);
if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) {
spin_unlock(&ctx->completion_lock);
queue:
ctx->drain_active = false;
io_req_task_queue(req);
return;
}
spin_unlock(&ctx->completion_lock);
io_prep_async_link(req);
de = kmalloc(sizeof(*de), GFP_KERNEL);
if (!de) {
ret = -ENOMEM;
io_req_defer_failed(req, ret);
return;
}
spin_lock(&ctx->completion_lock);
if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
spin_unlock(&ctx->completion_lock);
kfree(de);
goto queue;
}
trace_io_uring_defer(req);
de->req = req;
de->seq = seq;
list_add_tail(&de->list, &ctx->defer_list);
spin_unlock(&ctx->completion_lock);
}
static bool io_assign_file(struct io_kiocb *req, const struct io_issue_def *def,
unsigned int issue_flags)
{
if (req->file || !def->needs_file)
return true;
if (req->flags & REQ_F_FIXED_FILE)
req->file = io_file_get_fixed(req, req->cqe.fd, issue_flags);
else
req->file = io_file_get_normal(req, req->cqe.fd);
return !!req->file;
}
static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
const struct io_issue_def *def = &io_issue_defs[req->opcode];
const struct cred *creds = NULL;
int ret;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (unlikely(!io_assign_file(req, def, issue_flags)))
return -EBADF;
if (unlikely((req->flags & REQ_F_CREDS) && req->creds != current_cred()))
creds = override_creds(req->creds);
if (!def->audit_skip)
audit_uring_entry(req->opcode);
ret = def->issue(req, issue_flags);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (!def->audit_skip)
audit_uring_exit(!ret, ret);
if (creds)
revert_creds(creds);
if (ret == IOU_OK) {
if (issue_flags & IO_URING_F_COMPLETE_DEFER)
io_req_complete_defer(req);
else
io_req_complete_post(req, issue_flags);
return 0;
}
if (ret != IOU_ISSUE_SKIP_COMPLETE)
return ret;
/* If the op doesn't have a file, we're not polling for it */
if ((req->ctx->flags & IORING_SETUP_IOPOLL) && def->iopoll_queue)
io_iopoll_req_issued(req, issue_flags);
return 0;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
int io_poll_issue(struct io_kiocb *req, struct io_tw_state *ts)
{
io_tw_lock(req->ctx, ts);
return io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_MULTISHOT|
IO_URING_F_COMPLETE_DEFER);
}
struct io_wq_work *io_wq_free_work(struct io_wq_work *work)
{
struct io_kiocb *req = container_of(work, struct io_kiocb, work);
struct io_kiocb *nxt = NULL;
if (req_ref_put_and_test(req)) {
if (req->flags & IO_REQ_LINK_FLAGS)
nxt = io_req_find_next(req);
io_free_req(req);
}
return nxt ? &nxt->work : NULL;
}
void io_wq_submit_work(struct io_wq_work *work)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_kiocb *req = container_of(work, struct io_kiocb, work);
const struct io_issue_def *def = &io_issue_defs[req->opcode];
unsigned int issue_flags = IO_URING_F_UNLOCKED | IO_URING_F_IOWQ;
bool needs_poll = false;
int ret = 0, err = -ECANCELED;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
/* one will be dropped by ->io_wq_free_work() after returning to io-wq */
if (!(req->flags & REQ_F_REFCOUNT))
__io_req_set_refcount(req, 2);
else
req_ref_get(req);
io_uring: remove submission references Requests are by default given with two references, submission and completion. Completion references are straightforward, they represent request ownership and are put when a request is completed or so. Submission references are a bit more trickier. They're needed when io_issue_sqe() followed deep into the submission stack (e.g. in fs, block, drivers, etc.), request may have given away for concurrent execution or already completed, and the code unwinding back to io_issue_sqe() may be accessing some pieces of our requests, e.g. file or iov. Now, we prevent such async/in-depth completions by pushing requests through task_work. Punting to io-wq is also done through task_works, apart from a couple of cases with a pretty well known context. So, there're two cases: 1) io_issue_sqe() from the task context and protected by ->uring_lock. Either requests return back to io_uring or handed to task_work, which won't be executed because we're currently controlling that task. So, we can be sure that requests are staying alive all the time and we don't need submission references to pin them. 2) io_issue_sqe() from io-wq, which doesn't hold the mutex. The role of submission reference is played by io-wq reference, which is put by io_wq_submit_work(). Hence, it should be fine. Considering that, we can carefully kill the submission reference. Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/6b68f1c763229a590f2a27148aee77767a8d7750.1628705069.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-08-12 02:28:29 +08:00
io_arm_ltimeout(req);
io_uring: fix io_try_cancel_userdata race for iowq WARNING: CPU: 1 PID: 5870 at fs/io_uring.c:5975 io_try_cancel_userdata+0x30f/0x540 fs/io_uring.c:5975 CPU: 0 PID: 5870 Comm: iou-wrk-5860 Not tainted 5.14.0-rc6-next-20210820-syzkaller #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:io_try_cancel_userdata+0x30f/0x540 fs/io_uring.c:5975 Call Trace: io_async_cancel fs/io_uring.c:6014 [inline] io_issue_sqe+0x22d5/0x65a0 fs/io_uring.c:6407 io_wq_submit_work+0x1dc/0x300 fs/io_uring.c:6511 io_worker_handle_work+0xa45/0x1840 fs/io-wq.c:533 io_wqe_worker+0x2cc/0xbb0 fs/io-wq.c:582 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:295 io_try_cancel_userdata() can be called from io_async_cancel() executing in the io-wq context, so the warning fires, which is there to alert anyone accessing task->io_uring->io_wq in a racy way. However, io_wq_put_and_exit() always first waits for all threads to complete, so the only detail left is to zero tctx->io_wq after the context is removed. note: one little assumption is that when IO_WQ_WORK_CANCEL, the executor won't touch ->io_wq, because io_wq_destroy() might cancel left pending requests in such a way. Cc: stable@vger.kernel.org Reported-by: syzbot+b0c9d1588ae92866515f@syzkaller.appspotmail.com Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/dfdd37a80cfa9ffd3e59538929c99cdd55d8699e.1629721757.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-08-23 20:30:44 +08:00
/* either cancelled or io-wq is dying, so don't touch tctx->iowq */
if (work->flags & IO_WQ_WORK_CANCEL) {
fail:
io_req_task_queue_fail(req, err);
return;
}
if (!io_assign_file(req, def, issue_flags)) {
err = -EBADF;
work->flags |= IO_WQ_WORK_CANCEL;
goto fail;
}
if (req->flags & REQ_F_FORCE_ASYNC) {
bool opcode_poll = def->pollin || def->pollout;
if (opcode_poll && file_can_poll(req->file)) {
needs_poll = true;
issue_flags |= IO_URING_F_NONBLOCK;
}
}
do {
ret = io_issue_sqe(req, issue_flags);
if (ret != -EAGAIN)
break;
/*
* If REQ_F_NOWAIT is set, then don't wait or retry with
* poll. -EAGAIN is final for that case.
*/
if (req->flags & REQ_F_NOWAIT)
break;
/*
* We can get EAGAIN for iopolled IO even though we're
* forcing a sync submission from here, since we can't
* wait for request slots on the block side.
*/
if (!needs_poll) {
if (!(req->ctx->flags & IORING_SETUP_IOPOLL))
break;
if (io_wq_worker_stopped())
break;
cond_resched();
continue;
io_uring: implement async hybrid mode for pollable requests The current logic of requests with IOSQE_ASYNC is first queueing it to io-worker, then execute it in a synchronous way. For unbound works like pollable requests(e.g. read/write a socketfd), the io-worker may stuck there waiting for events for a long time. And thus other works wait in the list for a long time too. Let's introduce a new way for unbound works (currently pollable requests), with this a request will first be queued to io-worker, then executed in a nonblock try rather than a synchronous way. Failure of that leads it to arm poll stuff and then the worker can begin to handle other works. The detail process of this kind of requests is: step1: original context: queue it to io-worker step2: io-worker context: nonblock try(the old logic is a synchronous try here) | |--fail--> arm poll | |--(fail/ready)-->synchronous issue | |--(succeed)-->worker finish it's job, tw take over the req This works much better than the old IOSQE_ASYNC logic in cases where unbound max_worker is relatively small. In this case, number of io-worker eazily increments to max_worker, new worker cannot be created and running workers stuck there handling old works in IOSQE_ASYNC mode. In my 64-core machine, set unbound max_worker to 20, run echo-server, turns out: (arguments: register_file, connetion number is 1000, message size is 12 Byte) original IOSQE_ASYNC: 76664.151 tps after this patch: 166934.985 tps Suggested-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Hao Xu <haoxu@linux.alibaba.com> Link: https://lore.kernel.org/r/20211018133445.103438-1-haoxu@linux.alibaba.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-10-18 21:34:45 +08:00
}
if (io_arm_poll_handler(req, issue_flags) == IO_APOLL_OK)
return;
/* aborted or ready, in either case retry blocking */
needs_poll = false;
issue_flags &= ~IO_URING_F_NONBLOCK;
} while (1);
/* avoid locking problems by failing it from a clean context */
if (ret < 0)
io_req_task_queue_fail(req, ret);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
inline struct file *io_file_get_fixed(struct io_kiocb *req, int fd,
unsigned int issue_flags)
{
struct io_ring_ctx *ctx = req->ctx;
struct io_fixed_file *slot;
struct file *file = NULL;
io_ring_submit_lock(ctx, issue_flags);
if (unlikely((unsigned int)fd >= ctx->nr_user_files))
goto out;
fd = array_index_nospec(fd, ctx->nr_user_files);
slot = io_fixed_file_slot(&ctx->file_table, fd);
file = io_slot_file(slot);
req->flags |= io_slot_flags(slot);
io_req_set_rsrc_node(req, ctx, 0);
out:
io_ring_submit_unlock(ctx, issue_flags);
return file;
}
struct file *io_file_get_normal(struct io_kiocb *req, int fd)
{
struct file *file = fget(fd);
trace_io_uring_file_get(req, fd);
/* we don't allow fixed io_uring files */
if (file && io_is_uring_fops(file))
io_req_track_inflight(req);
return file;
}
static void io_queue_async(struct io_kiocb *req, int ret)
__must_hold(&req->ctx->uring_lock)
{
struct io_kiocb *linked_timeout;
if (ret != -EAGAIN || (req->flags & REQ_F_NOWAIT)) {
io_req_defer_failed(req, ret);
return;
}
linked_timeout = io_prep_linked_timeout(req);
switch (io_arm_poll_handler(req, 0)) {
case IO_APOLL_READY:
io_kbuf_recycle(req, 0);
io_req_task_queue(req);
break;
case IO_APOLL_ABORTED:
io_kbuf_recycle(req, 0);
io_queue_iowq(req, NULL);
break;
case IO_APOLL_OK:
break;
}
if (linked_timeout)
io_queue_linked_timeout(linked_timeout);
}
static inline void io_queue_sqe(struct io_kiocb *req)
__must_hold(&req->ctx->uring_lock)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
int ret;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
/*
* We async punt it if the file wasn't marked NOWAIT, or if the file
* doesn't support non-blocking read/write attempts
*/
if (likely(!ret))
io_arm_ltimeout(req);
else
io_queue_async(req, ret);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
static void io_queue_sqe_fallback(struct io_kiocb *req)
__must_hold(&req->ctx->uring_lock)
{
if (unlikely(req->flags & REQ_F_FAIL)) {
/*
* We don't submit, fail them all, for that replace hardlinks
* with normal links. Extra REQ_F_LINK is tolerated.
*/
req->flags &= ~REQ_F_HARDLINK;
req->flags |= REQ_F_LINK;
io_req_defer_failed(req, req->cqe.res);
} else {
int ret = io_req_prep_async(req);
if (unlikely(ret)) {
io_req_defer_failed(req, ret);
return;
}
if (unlikely(req->ctx->drain_active))
io_drain_req(req);
else
io_queue_iowq(req, NULL);
}
}
/*
* Check SQE restrictions (opcode and flags).
*
* Returns 'true' if SQE is allowed, 'false' otherwise.
*/
static inline bool io_check_restriction(struct io_ring_ctx *ctx,
struct io_kiocb *req,
unsigned int sqe_flags)
{
if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
return false;
if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
ctx->restrictions.sqe_flags_required)
return false;
if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
ctx->restrictions.sqe_flags_required))
return false;
return true;
}
static void io_init_req_drain(struct io_kiocb *req)
{
struct io_ring_ctx *ctx = req->ctx;
struct io_kiocb *head = ctx->submit_state.link.head;
ctx->drain_active = true;
if (head) {
/*
* If we need to drain a request in the middle of a link, drain
* the head request and the next request/link after the current
* link. Considering sequential execution of links,
* REQ_F_IO_DRAIN will be maintained for every request of our
* link.
*/
head->flags |= REQ_F_IO_DRAIN | REQ_F_FORCE_ASYNC;
ctx->drain_next = true;
}
}
static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
const struct io_uring_sqe *sqe)
__must_hold(&ctx->uring_lock)
{
const struct io_issue_def *def;
unsigned int sqe_flags;
int personality;
u8 opcode;
/* req is partially pre-initialised, see io_preinit_req() */
req->opcode = opcode = READ_ONCE(sqe->opcode);
/* same numerical values with corresponding REQ_F_*, safe to copy */
req->flags = sqe_flags = READ_ONCE(sqe->flags);
req->cqe.user_data = READ_ONCE(sqe->user_data);
req->file = NULL;
req->rsrc_node = NULL;
req->task = current;
if (unlikely(opcode >= IORING_OP_LAST)) {
req->opcode = 0;
return -EINVAL;
}
def = &io_issue_defs[opcode];
if (unlikely(sqe_flags & ~SQE_COMMON_FLAGS)) {
/* enforce forwards compatibility on users */
if (sqe_flags & ~SQE_VALID_FLAGS)
return -EINVAL;
if (sqe_flags & IOSQE_BUFFER_SELECT) {
if (!def->buffer_select)
return -EOPNOTSUPP;
req->buf_index = READ_ONCE(sqe->buf_group);
}
if (sqe_flags & IOSQE_CQE_SKIP_SUCCESS)
ctx->drain_disabled = true;
if (sqe_flags & IOSQE_IO_DRAIN) {
if (ctx->drain_disabled)
return -EOPNOTSUPP;
io_init_req_drain(req);
}
}
if (unlikely(ctx->restricted || ctx->drain_active || ctx->drain_next)) {
if (ctx->restricted && !io_check_restriction(ctx, req, sqe_flags))
return -EACCES;
/* knock it to the slow queue path, will be drained there */
if (ctx->drain_active)
req->flags |= REQ_F_FORCE_ASYNC;
/* if there is no link, we're at "next" request and need to drain */
if (unlikely(ctx->drain_next) && !ctx->submit_state.link.head) {
ctx->drain_next = false;
ctx->drain_active = true;
req->flags |= REQ_F_IO_DRAIN | REQ_F_FORCE_ASYNC;
}
}
if (!def->ioprio && sqe->ioprio)
return -EINVAL;
if (!def->iopoll && (ctx->flags & IORING_SETUP_IOPOLL))
return -EINVAL;
if (def->needs_file) {
struct io_submit_state *state = &ctx->submit_state;
req->cqe.fd = READ_ONCE(sqe->fd);
/*
* Plug now if we have more than 2 IO left after this, and the
* target is potentially a read/write to block based storage.
*/
if (state->need_plug && def->plug) {
state->plug_started = true;
state->need_plug = false;
blk_start_plug_nr_ios(&state->plug, state->submit_nr);
}
}
personality = READ_ONCE(sqe->personality);
if (personality) {
selinux/stable-5.16 PR 20211101 -----BEGIN PGP SIGNATURE----- iQJIBAABCAAyFiEES0KozwfymdVUl37v6iDy2pc3iXMFAmGANbAUHHBhdWxAcGF1 bC1tb29yZS5jb20ACgkQ6iDy2pc3iXNaMBAAg+9gZr0F7xiafu8JFZqZfx/AQdJ2 G2cn3le+/tXGZmF8m/+82lOaR6LeQLatgSDJNSkXWkKr0nRwseQJDbtRfvYJdn0t Ax05/Fmz6OGxQ2wgRYgaFiSrKpE5p3NhDtiLFVdkCJaQNe/8DZOc7NhBl6EjZf3x ubhl2hUiJ4AmiXGwcYhr4uKgP4nhW8OM1/OkskVi+bBMmLA8KTY9kslmIDP5E3BW 29W4qhqeLNQupY5dGMEMVcyxY9ZUWpO39q4uOaQVZrUGE7xABkj/jhnxT5gFTSlI pu8VhsYXm9KuRVveIsv0L5SZfadwoM9YAl7ki1wD3W5rHqOAte3rBTm6VmNlQwfU MqxP65Jiyxudxet5Be3/dCRH/+MDQuwBxivgmZXbeVxor2SeznVb0GDaEUC5FSHu CJIgWtQzsPJMxgAEGXN4F3QGP0htTTJni56GUPOsrf4TIBW02TT+oLTLFRIokQQL INNOfwVSRXElnCsvxsHR4oB+JZ9pJyBaAmeupcQ6jmcKiWlbLj4s+W0U0pM5h91v hmMpz7KMxrX6gVL4gB2Jj4aN3r5YRbq26NBu6D+wdwwBTeTTocaHSpAqkv4buClf uNk3cG8Hkp8TTg9cM8jYgpxMyzKH/AI/Uw3VhEa1xCiq2Ck3DgfnZvnvcRRaZevU FPgmwgqePJXGi60= =sb8J -----END PGP SIGNATURE----- Merge tag 'selinux-pr-20211101' of git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/selinux Pull selinux updates from Paul Moore: - Add LSM/SELinux/Smack controls and auditing for io-uring. As usual, the individual commit descriptions have more detail, but we were basically missing two things which we're adding here: + establishment of a proper audit context so that auditing of io-uring ops works similarly to how it does for syscalls (with some io-uring additions because io-uring ops are *not* syscalls) + additional LSM hooks to enable access control points for some of the more unusual io-uring features, e.g. credential overrides. The additional audit callouts and LSM hooks were done in conjunction with the io-uring folks, based on conversations and RFC patches earlier in the year. - Fixup the binder credential handling so that the proper credentials are used in the LSM hooks; the commit description and the code comment which is removed in these patches are helpful to understand the background and why this is the proper fix. - Enable SELinux genfscon policy support for securityfs, allowing improved SELinux filesystem labeling for other subsystems which make use of securityfs, e.g. IMA. * tag 'selinux-pr-20211101' of git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/selinux: security: Return xattr name from security_dentry_init_security() selinux: fix a sock regression in selinux_ip_postroute_compat() binder: use cred instead of task for getsecid binder: use cred instead of task for selinux checks binder: use euid from cred instead of using task LSM: Avoid warnings about potentially unused hook variables selinux: fix all of the W=1 build warnings selinux: make better use of the nf_hook_state passed to the NF hooks selinux: fix race condition when computing ocontext SIDs selinux: remove unneeded ipv6 hook wrappers selinux: remove the SELinux lockdown implementation selinux: enable genfscon labeling for securityfs Smack: Brutalist io_uring support selinux: add support for the io_uring access controls lsm,io_uring: add LSM hooks to io_uring io_uring: convert io_uring to the secure anon inode interface fs: add anon_inode_getfile_secure() similar to anon_inode_getfd_secure() audit: add filtering for io_uring records audit,io_uring,io-wq: add some basic audit support to io_uring audit: prepare audit_context for use in calling contexts beyond syscalls
2021-11-02 12:06:18 +08:00
int ret;
req->creds = xa_load(&ctx->personalities, personality);
if (!req->creds)
return -EINVAL;
get_cred(req->creds);
lsm,io_uring: add LSM hooks to io_uring A full expalantion of io_uring is beyond the scope of this commit description, but in summary it is an asynchronous I/O mechanism which allows for I/O requests and the resulting data to be queued in memory mapped "rings" which are shared between the kernel and userspace. Optionally, io_uring offers the ability for applications to spawn kernel threads to dequeue I/O requests from the ring and submit the requests in the kernel, helping to minimize the syscall overhead. Rings are accessed in userspace by memory mapping a file descriptor provided by the io_uring_setup(2), and can be shared between applications as one might do with any open file descriptor. Finally, process credentials can be registered with a given ring and any process with access to that ring can submit I/O requests using any of the registered credentials. While the io_uring functionality is widely recognized as offering a vastly improved, and high performing asynchronous I/O mechanism, its ability to allow processes to submit I/O requests with credentials other than its own presents a challenge to LSMs. When a process creates a new io_uring ring the ring's credentials are inhertied from the calling process; if this ring is shared with another process operating with different credentials there is the potential to bypass the LSMs security policy. Similarly, registering credentials with a given ring allows any process with access to that ring to submit I/O requests with those credentials. In an effort to allow LSMs to apply security policy to io_uring I/O operations, this patch adds two new LSM hooks. These hooks, in conjunction with the LSM anonymous inode support previously submitted, allow an LSM to apply access control policy to the sharing of io_uring rings as well as any io_uring credential changes requested by a process. The new LSM hooks are described below: * int security_uring_override_creds(cred) Controls if the current task, executing an io_uring operation, is allowed to override it's credentials with @cred. In cases where the current task is a user application, the current credentials will be those of the user application. In cases where the current task is a kernel thread servicing io_uring requests the current credentials will be those of the io_uring ring (inherited from the process that created the ring). * int security_uring_sqpoll(void) Controls if the current task is allowed to create an io_uring polling thread (IORING_SETUP_SQPOLL). Without a SQPOLL thread in the kernel processes must submit I/O requests via io_uring_enter(2) which allows us to compare any requested credential changes against the application making the request. With a SQPOLL thread, we can no longer compare requested credential changes against the application making the request, the comparison is made against the ring's credentials. Signed-off-by: Paul Moore <paul@paul-moore.com>
2021-02-02 08:56:49 +08:00
ret = security_uring_override_creds(req->creds);
if (ret) {
put_cred(req->creds);
return ret;
}
req->flags |= REQ_F_CREDS;
}
return def->prep(req, sqe);
}
static __cold int io_submit_fail_init(const struct io_uring_sqe *sqe,
struct io_kiocb *req, int ret)
{
struct io_ring_ctx *ctx = req->ctx;
struct io_submit_link *link = &ctx->submit_state.link;
struct io_kiocb *head = link->head;
trace_io_uring_req_failed(sqe, req, ret);
/*
* Avoid breaking links in the middle as it renders links with SQPOLL
* unusable. Instead of failing eagerly, continue assembling the link if
* applicable and mark the head with REQ_F_FAIL. The link flushing code
* should find the flag and handle the rest.
*/
req_fail_link_node(req, ret);
if (head && !(head->flags & REQ_F_FAIL))
req_fail_link_node(head, -ECANCELED);
if (!(req->flags & IO_REQ_LINK_FLAGS)) {
if (head) {
link->last->link = req;
link->head = NULL;
req = head;
}
io_queue_sqe_fallback(req);
return ret;
}
if (head)
link->last->link = req;
else
link->head = req;
link->last = req;
return 0;
}
static inline int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
const struct io_uring_sqe *sqe)
__must_hold(&ctx->uring_lock)
{
struct io_submit_link *link = &ctx->submit_state.link;
int ret;
ret = io_init_req(ctx, req, sqe);
if (unlikely(ret))
return io_submit_fail_init(sqe, req, ret);
trace_io_uring_submit_req(req);
/*
* If we already have a head request, queue this one for async
* submittal once the head completes. If we don't have a head but
* IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
* submitted sync once the chain is complete. If none of those
* conditions are true (normal request), then just queue it.
*/
if (unlikely(link->head)) {
ret = io_req_prep_async(req);
if (unlikely(ret))
return io_submit_fail_init(sqe, req, ret);
trace_io_uring_link(req, link->head);
link->last->link = req;
link->last = req;
if (req->flags & IO_REQ_LINK_FLAGS)
return 0;
/* last request of the link, flush it */
req = link->head;
link->head = NULL;
if (req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL))
goto fallback;
} else if (unlikely(req->flags & (IO_REQ_LINK_FLAGS |
REQ_F_FORCE_ASYNC | REQ_F_FAIL))) {
if (req->flags & IO_REQ_LINK_FLAGS) {
link->head = req;
link->last = req;
} else {
fallback:
io_queue_sqe_fallback(req);
}
return 0;
}
io_queue_sqe(req);
return 0;
}
/*
* Batched submission is done, ensure local IO is flushed out.
*/
static void io_submit_state_end(struct io_ring_ctx *ctx)
{
struct io_submit_state *state = &ctx->submit_state;
if (unlikely(state->link.head))
io_queue_sqe_fallback(state->link.head);
/* flush only after queuing links as they can generate completions */
io_submit_flush_completions(ctx);
if (state->plug_started)
blk_finish_plug(&state->plug);
}
/*
* Start submission side cache.
*/
static void io_submit_state_start(struct io_submit_state *state,
unsigned int max_ios)
{
state->plug_started = false;
state->need_plug = max_ios > 2;
state->submit_nr = max_ios;
/* set only head, no need to init link_last in advance */
state->link.head = NULL;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
static void io_commit_sqring(struct io_ring_ctx *ctx)
{
struct io_rings *rings = ctx->rings;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
/*
* Ensure any loads from the SQEs are done at this point,
* since once we write the new head, the application could
* write new data to them.
*/
smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
/*
* Fetch an sqe, if one is available. Note this returns a pointer to memory
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
* that is mapped by userspace. This means that care needs to be taken to
* ensure that reads are stable, as we cannot rely on userspace always
* being a good citizen. If members of the sqe are validated and then later
* used, it's important that those reads are done through READ_ONCE() to
* prevent a re-load down the line.
*/
static bool io_get_sqe(struct io_ring_ctx *ctx, const struct io_uring_sqe **sqe)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
unsigned mask = ctx->sq_entries - 1;
unsigned head = ctx->cached_sq_head++ & mask;
if (!(ctx->flags & IORING_SETUP_NO_SQARRAY)) {
head = READ_ONCE(ctx->sq_array[head]);
if (unlikely(head >= ctx->sq_entries)) {
/* drop invalid entries */
spin_lock(&ctx->completion_lock);
ctx->cq_extra--;
spin_unlock(&ctx->completion_lock);
WRITE_ONCE(ctx->rings->sq_dropped,
READ_ONCE(ctx->rings->sq_dropped) + 1);
return false;
}
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
/*
* The cached sq head (or cq tail) serves two purposes:
*
* 1) allows us to batch the cost of updating the user visible
* head updates.
* 2) allows the kernel side to track the head on its own, even
* though the application is the one updating it.
*/
/* double index for 128-byte SQEs, twice as long */
if (ctx->flags & IORING_SETUP_SQE128)
head <<= 1;
*sqe = &ctx->sq_sqes[head];
return true;
}
int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
__must_hold(&ctx->uring_lock)
{
unsigned int entries = io_sqring_entries(ctx);
unsigned int left;
int ret;
if (unlikely(!entries))
return 0;
/* make sure SQ entry isn't read before tail */
ret = left = min(nr, entries);
io_get_task_refs(left);
io_submit_state_start(&ctx->submit_state, left);
do {
const struct io_uring_sqe *sqe;
struct io_kiocb *req;
if (unlikely(!io_alloc_req(ctx, &req)))
break;
if (unlikely(!io_get_sqe(ctx, &sqe))) {
io_req_add_to_cache(req, ctx);
break;
}
/*
* Continue submitting even for sqe failure if the
* ring was setup with IORING_SETUP_SUBMIT_ALL
*/
if (unlikely(io_submit_sqe(ctx, req, sqe)) &&
!(ctx->flags & IORING_SETUP_SUBMIT_ALL)) {
left--;
break;
}
} while (--left);
if (unlikely(left)) {
ret -= left;
/* try again if it submitted nothing and can't allocate a req */
if (!ret && io_req_cache_empty(ctx))
ret = -EAGAIN;
current->io_uring->cached_refs += left;
}
io_submit_state_end(ctx);
/* Commit SQ ring head once we've consumed and submitted all SQEs */
io_commit_sqring(ctx);
return ret;
}
struct io_wait_queue {
struct wait_queue_entry wq;
struct io_ring_ctx *ctx;
unsigned cq_tail;
unsigned nr_timeouts;
ktime_t timeout;
};
static inline bool io_has_work(struct io_ring_ctx *ctx)
{
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
return test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq) ||
!llist_empty(&ctx->work_llist);
}
static inline bool io_should_wake(struct io_wait_queue *iowq)
{
struct io_ring_ctx *ctx = iowq->ctx;
int dist = READ_ONCE(ctx->rings->cq.tail) - (int) iowq->cq_tail;
/*
* Wake up if we have enough events, or if a timeout occurred since we
* started waiting. For timeouts, we always want to return to userspace,
* regardless of event count.
*/
return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
}
static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
int wake_flags, void *key)
{
struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, wq);
/*
* Cannot safely flush overflowed CQEs from here, ensure we wake up
* the task, and the next invocation will do it.
*/
if (io_should_wake(iowq) || io_has_work(iowq->ctx))
return autoremove_wake_function(curr, mode, wake_flags, key);
return -1;
}
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
int io_run_task_work_sig(struct io_ring_ctx *ctx)
{
if (!llist_empty(&ctx->work_llist)) {
__set_current_state(TASK_RUNNING);
if (io_run_local_work(ctx) > 0)
return 0;
}
if (io_run_task_work() > 0)
return 0;
if (task_sigpending(current))
return -EINTR;
return 0;
}
static bool current_pending_io(void)
{
struct io_uring_task *tctx = current->io_uring;
if (!tctx)
return false;
return percpu_counter_read_positive(&tctx->inflight);
}
/* when returns >0, the caller should retry */
static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
struct io_wait_queue *iowq)
{
int io_wait, ret;
2023-07-08 00:20:07 +08:00
if (unlikely(READ_ONCE(ctx->check_cq)))
return 1;
if (unlikely(!llist_empty(&ctx->work_llist)))
return 1;
if (unlikely(test_thread_flag(TIF_NOTIFY_SIGNAL)))
return 1;
if (unlikely(task_sigpending(current)))
return -EINTR;
if (unlikely(io_should_wake(iowq)))
return 0;
2023-07-08 00:20:07 +08:00
/*
* Mark us as being in io_wait if we have pending requests, so cpufreq
* can take into account that the task is waiting for IO - turns out
* to be important for low QD IO.
2023-07-08 00:20:07 +08:00
*/
io_wait = current->in_iowait;
if (current_pending_io())
current->in_iowait = 1;
2023-07-08 00:20:07 +08:00
ret = 0;
if (iowq->timeout == KTIME_MAX)
schedule();
else if (!schedule_hrtimeout(&iowq->timeout, HRTIMER_MODE_ABS))
2023-07-08 00:20:07 +08:00
ret = -ETIME;
current->in_iowait = io_wait;
2023-07-08 00:20:07 +08:00
return ret;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
/*
* Wait until events become available, if we don't already have some. The
* application must reap them itself, as they reside on the shared cq ring.
*/
static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
const sigset_t __user *sig, size_t sigsz,
struct __kernel_timespec __user *uts)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_wait_queue iowq;
struct io_rings *rings = ctx->rings;
int ret;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (!io_allowed_run_tw(ctx))
return -EEXIST;
if (!llist_empty(&ctx->work_llist))
io_run_local_work(ctx);
io_run_task_work();
io_cqring_overflow_flush(ctx);
/* if user messes with these they will just get an early return */
if (__io_cqring_events_user(ctx) >= min_events)
return 0;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (sig) {
#ifdef CONFIG_COMPAT
if (in_compat_syscall())
ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
sigsz);
else
#endif
ret = set_user_sigmask(sig, sigsz);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (ret)
return ret;
}
init_waitqueue_func_entry(&iowq.wq, io_wake_function);
iowq.wq.private = current;
INIT_LIST_HEAD(&iowq.wq.entry);
iowq.ctx = ctx;
iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events;
iowq.timeout = KTIME_MAX;
if (uts) {
struct timespec64 ts;
if (get_timespec64(&ts, uts))
return -EFAULT;
iowq.timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns());
}
trace_io_uring_cqring_wait(ctx, min_events);
do {
unsigned long check_cq;
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) {
int nr_wait = (int) iowq.cq_tail - READ_ONCE(ctx->rings->cq.tail);
atomic_set(&ctx->cq_wait_nr, nr_wait);
set_current_state(TASK_INTERRUPTIBLE);
} else {
prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq,
TASK_INTERRUPTIBLE);
}
ret = io_cqring_wait_schedule(ctx, &iowq);
__set_current_state(TASK_RUNNING);
atomic_set(&ctx->cq_wait_nr, 0);
/*
* Run task_work after scheduling and before io_should_wake().
* If we got woken because of task_work being processed, run it
* now rather than let the caller do another wait loop.
*/
io_run_task_work();
if (!llist_empty(&ctx->work_llist))
io_run_local_work(ctx);
/*
* Non-local task_work will be run on exit to userspace, but
* if we're using DEFER_TASKRUN, then we could have waited
* with a timeout for a number of requests. If the timeout
* hits, we could have some requests ready to process. Ensure
* this break is _after_ we have run task_work, to avoid
* deferring running potentially pending requests until the
* next time we wait for events.
*/
if (ret < 0)
break;
check_cq = READ_ONCE(ctx->check_cq);
if (unlikely(check_cq)) {
/* let the caller flush overflows, retry */
if (check_cq & BIT(IO_CHECK_CQ_OVERFLOW_BIT))
io_cqring_do_overflow_flush(ctx);
if (check_cq & BIT(IO_CHECK_CQ_DROPPED_BIT)) {
ret = -EBADR;
break;
}
}
if (io_should_wake(&iowq)) {
ret = 0;
break;
}
cond_resched();
} while (1);
if (!(ctx->flags & IORING_SETUP_DEFER_TASKRUN))
finish_wait(&ctx->cq_wait, &iowq.wq);
restore_saved_sigmask_unless(ret == -EINTR);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
void io_mem_free(void *ptr)
{
if (!ptr)
return;
folio_put(virt_to_folio(ptr));
}
static void io_pages_free(struct page ***pages, int npages)
{
struct page **page_array;
int i;
if (!pages)
return;
io_uring: fix crash with IORING_SETUP_NO_MMAP and invalid SQ ring address If we specify a valid CQ ring address but an invalid SQ ring address, we'll correctly spot this and free the allocated pages and clear them to NULL. However, we don't clear the ring page count, and hence will attempt to free the pages again. We've already cleared the address of the page array when freeing them, but we don't check for that. This causes the following crash: Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000 Oops [#1] Modules linked in: CPU: 0 PID: 20 Comm: kworker/u2:1 Not tainted 6.6.0-rc5-dirty #56 Hardware name: ucbbar,riscvemu-bare (DT) Workqueue: events_unbound io_ring_exit_work epc : io_pages_free+0x2a/0x58 ra : io_rings_free+0x3a/0x50 epc : ffffffff808811a2 ra : ffffffff80881406 sp : ffff8f80000c3cd0 status: 0000000200000121 badaddr: 0000000000000000 cause: 000000000000000d [<ffffffff808811a2>] io_pages_free+0x2a/0x58 [<ffffffff80881406>] io_rings_free+0x3a/0x50 [<ffffffff80882176>] io_ring_exit_work+0x37e/0x424 [<ffffffff80027234>] process_one_work+0x10c/0x1f4 [<ffffffff8002756e>] worker_thread+0x252/0x31c [<ffffffff8002f5e4>] kthread+0xc4/0xe0 [<ffffffff8000332a>] ret_from_fork+0xa/0x1c Check for a NULL array in io_pages_free(), but also clear the page counts when we free them to be on the safer side. Reported-by: rtm@csail.mit.edu Fixes: 03d89a2de25b ("io_uring: support for user allocated memory for rings/sqes") Cc: stable@vger.kernel.org Reviewed-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-18 22:09:27 +08:00
page_array = *pages;
io_uring: fix crash with IORING_SETUP_NO_MMAP and invalid SQ ring address If we specify a valid CQ ring address but an invalid SQ ring address, we'll correctly spot this and free the allocated pages and clear them to NULL. However, we don't clear the ring page count, and hence will attempt to free the pages again. We've already cleared the address of the page array when freeing them, but we don't check for that. This causes the following crash: Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000 Oops [#1] Modules linked in: CPU: 0 PID: 20 Comm: kworker/u2:1 Not tainted 6.6.0-rc5-dirty #56 Hardware name: ucbbar,riscvemu-bare (DT) Workqueue: events_unbound io_ring_exit_work epc : io_pages_free+0x2a/0x58 ra : io_rings_free+0x3a/0x50 epc : ffffffff808811a2 ra : ffffffff80881406 sp : ffff8f80000c3cd0 status: 0000000200000121 badaddr: 0000000000000000 cause: 000000000000000d [<ffffffff808811a2>] io_pages_free+0x2a/0x58 [<ffffffff80881406>] io_rings_free+0x3a/0x50 [<ffffffff80882176>] io_ring_exit_work+0x37e/0x424 [<ffffffff80027234>] process_one_work+0x10c/0x1f4 [<ffffffff8002756e>] worker_thread+0x252/0x31c [<ffffffff8002f5e4>] kthread+0xc4/0xe0 [<ffffffff8000332a>] ret_from_fork+0xa/0x1c Check for a NULL array in io_pages_free(), but also clear the page counts when we free them to be on the safer side. Reported-by: rtm@csail.mit.edu Fixes: 03d89a2de25b ("io_uring: support for user allocated memory for rings/sqes") Cc: stable@vger.kernel.org Reviewed-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-18 22:09:27 +08:00
if (!page_array)
return;
for (i = 0; i < npages; i++)
unpin_user_page(page_array[i]);
kvfree(page_array);
*pages = NULL;
}
static void *__io_uaddr_map(struct page ***pages, unsigned short *npages,
unsigned long uaddr, size_t size)
{
struct page **page_array;
unsigned int nr_pages;
void *page_addr;
int ret, i;
*npages = 0;
if (uaddr & (PAGE_SIZE - 1) || !size)
return ERR_PTR(-EINVAL);
nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
if (nr_pages > USHRT_MAX)
return ERR_PTR(-EINVAL);
page_array = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
if (!page_array)
return ERR_PTR(-ENOMEM);
ret = pin_user_pages_fast(uaddr, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
page_array);
if (ret != nr_pages) {
err:
io_pages_free(&page_array, ret > 0 ? ret : 0);
return ret < 0 ? ERR_PTR(ret) : ERR_PTR(-EFAULT);
}
page_addr = page_address(page_array[0]);
for (i = 0; i < nr_pages; i++) {
ret = -EINVAL;
/*
* Can't support mapping user allocated ring memory on 32-bit
* archs where it could potentially reside in highmem. Just
* fail those with -EINVAL, just like we did on kernels that
* didn't support this feature.
*/
if (PageHighMem(page_array[i]))
goto err;
/*
* No support for discontig pages for now, should either be a
* single normal page, or a huge page. Later on we can add
* support for remapping discontig pages, for now we will
* just fail them with EINVAL.
*/
if (page_address(page_array[i]) != page_addr)
goto err;
page_addr += PAGE_SIZE;
}
*pages = page_array;
*npages = nr_pages;
return page_to_virt(page_array[0]);
}
static void *io_rings_map(struct io_ring_ctx *ctx, unsigned long uaddr,
size_t size)
{
return __io_uaddr_map(&ctx->ring_pages, &ctx->n_ring_pages, uaddr,
size);
}
static void *io_sqes_map(struct io_ring_ctx *ctx, unsigned long uaddr,
size_t size)
{
return __io_uaddr_map(&ctx->sqe_pages, &ctx->n_sqe_pages, uaddr,
size);
}
static void io_rings_free(struct io_ring_ctx *ctx)
{
if (!(ctx->flags & IORING_SETUP_NO_MMAP)) {
io_mem_free(ctx->rings);
io_mem_free(ctx->sq_sqes);
ctx->rings = NULL;
ctx->sq_sqes = NULL;
} else {
io_pages_free(&ctx->ring_pages, ctx->n_ring_pages);
io_uring: fix crash with IORING_SETUP_NO_MMAP and invalid SQ ring address If we specify a valid CQ ring address but an invalid SQ ring address, we'll correctly spot this and free the allocated pages and clear them to NULL. However, we don't clear the ring page count, and hence will attempt to free the pages again. We've already cleared the address of the page array when freeing them, but we don't check for that. This causes the following crash: Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000 Oops [#1] Modules linked in: CPU: 0 PID: 20 Comm: kworker/u2:1 Not tainted 6.6.0-rc5-dirty #56 Hardware name: ucbbar,riscvemu-bare (DT) Workqueue: events_unbound io_ring_exit_work epc : io_pages_free+0x2a/0x58 ra : io_rings_free+0x3a/0x50 epc : ffffffff808811a2 ra : ffffffff80881406 sp : ffff8f80000c3cd0 status: 0000000200000121 badaddr: 0000000000000000 cause: 000000000000000d [<ffffffff808811a2>] io_pages_free+0x2a/0x58 [<ffffffff80881406>] io_rings_free+0x3a/0x50 [<ffffffff80882176>] io_ring_exit_work+0x37e/0x424 [<ffffffff80027234>] process_one_work+0x10c/0x1f4 [<ffffffff8002756e>] worker_thread+0x252/0x31c [<ffffffff8002f5e4>] kthread+0xc4/0xe0 [<ffffffff8000332a>] ret_from_fork+0xa/0x1c Check for a NULL array in io_pages_free(), but also clear the page counts when we free them to be on the safer side. Reported-by: rtm@csail.mit.edu Fixes: 03d89a2de25b ("io_uring: support for user allocated memory for rings/sqes") Cc: stable@vger.kernel.org Reviewed-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-18 22:09:27 +08:00
ctx->n_ring_pages = 0;
io_pages_free(&ctx->sqe_pages, ctx->n_sqe_pages);
io_uring: fix crash with IORING_SETUP_NO_MMAP and invalid SQ ring address If we specify a valid CQ ring address but an invalid SQ ring address, we'll correctly spot this and free the allocated pages and clear them to NULL. However, we don't clear the ring page count, and hence will attempt to free the pages again. We've already cleared the address of the page array when freeing them, but we don't check for that. This causes the following crash: Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000 Oops [#1] Modules linked in: CPU: 0 PID: 20 Comm: kworker/u2:1 Not tainted 6.6.0-rc5-dirty #56 Hardware name: ucbbar,riscvemu-bare (DT) Workqueue: events_unbound io_ring_exit_work epc : io_pages_free+0x2a/0x58 ra : io_rings_free+0x3a/0x50 epc : ffffffff808811a2 ra : ffffffff80881406 sp : ffff8f80000c3cd0 status: 0000000200000121 badaddr: 0000000000000000 cause: 000000000000000d [<ffffffff808811a2>] io_pages_free+0x2a/0x58 [<ffffffff80881406>] io_rings_free+0x3a/0x50 [<ffffffff80882176>] io_ring_exit_work+0x37e/0x424 [<ffffffff80027234>] process_one_work+0x10c/0x1f4 [<ffffffff8002756e>] worker_thread+0x252/0x31c [<ffffffff8002f5e4>] kthread+0xc4/0xe0 [<ffffffff8000332a>] ret_from_fork+0xa/0x1c Check for a NULL array in io_pages_free(), but also clear the page counts when we free them to be on the safer side. Reported-by: rtm@csail.mit.edu Fixes: 03d89a2de25b ("io_uring: support for user allocated memory for rings/sqes") Cc: stable@vger.kernel.org Reviewed-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-18 22:09:27 +08:00
ctx->n_sqe_pages = 0;
}
}
void *io_mem_alloc(size_t size)
{
gfp_t gfp = GFP_KERNEL_ACCOUNT | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP;
void *ret;
ret = (void *) __get_free_pages(gfp, get_order(size));
if (ret)
return ret;
return ERR_PTR(-ENOMEM);
}
static unsigned long rings_size(struct io_ring_ctx *ctx, unsigned int sq_entries,
unsigned int cq_entries, size_t *sq_offset)
{
struct io_rings *rings;
size_t off, sq_array_size;
off = struct_size(rings, cqes, cq_entries);
if (off == SIZE_MAX)
return SIZE_MAX;
if (ctx->flags & IORING_SETUP_CQE32) {
if (check_shl_overflow(off, 1, &off))
return SIZE_MAX;
}
#ifdef CONFIG_SMP
off = ALIGN(off, SMP_CACHE_BYTES);
if (off == 0)
return SIZE_MAX;
#endif
if (ctx->flags & IORING_SETUP_NO_SQARRAY) {
if (sq_offset)
*sq_offset = SIZE_MAX;
return off;
}
if (sq_offset)
*sq_offset = off;
sq_array_size = array_size(sizeof(u32), sq_entries);
if (sq_array_size == SIZE_MAX)
return SIZE_MAX;
if (check_add_overflow(off, sq_array_size, &off))
return SIZE_MAX;
io_uring: don't hold uring_lock when calling io_run_task_work* Abaci reported the below issue: [ 141.400455] hrtimer: interrupt took 205853 ns [ 189.869316] process 'usr/local/ilogtail/ilogtail_0.16.26' started with executable stack [ 250.188042] [ 250.188327] ============================================ [ 250.189015] WARNING: possible recursive locking detected [ 250.189732] 5.11.0-rc4 #1 Not tainted [ 250.190267] -------------------------------------------- [ 250.190917] a.out/7363 is trying to acquire lock: [ 250.191506] ffff888114dbcbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __io_req_task_submit+0x29/0xa0 [ 250.192599] [ 250.192599] but task is already holding lock: [ 250.193309] ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.194426] [ 250.194426] other info that might help us debug this: [ 250.195238] Possible unsafe locking scenario: [ 250.195238] [ 250.196019] CPU0 [ 250.196411] ---- [ 250.196803] lock(&ctx->uring_lock); [ 250.197420] lock(&ctx->uring_lock); [ 250.197966] [ 250.197966] *** DEADLOCK *** [ 250.197966] [ 250.198837] May be due to missing lock nesting notation [ 250.198837] [ 250.199780] 1 lock held by a.out/7363: [ 250.200373] #0: ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.201645] [ 250.201645] stack backtrace: [ 250.202298] CPU: 0 PID: 7363 Comm: a.out Not tainted 5.11.0-rc4 #1 [ 250.203144] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 250.203887] Call Trace: [ 250.204302] dump_stack+0xac/0xe3 [ 250.204804] __lock_acquire+0xab6/0x13a0 [ 250.205392] lock_acquire+0x2c3/0x390 [ 250.205928] ? __io_req_task_submit+0x29/0xa0 [ 250.206541] __mutex_lock+0xae/0x9f0 [ 250.207071] ? __io_req_task_submit+0x29/0xa0 [ 250.207745] ? 0xffffffffa0006083 [ 250.208248] ? __io_req_task_submit+0x29/0xa0 [ 250.208845] ? __io_req_task_submit+0x29/0xa0 [ 250.209452] ? __io_req_task_submit+0x5/0xa0 [ 250.210083] __io_req_task_submit+0x29/0xa0 [ 250.210687] io_async_task_func+0x23d/0x4c0 [ 250.211278] task_work_run+0x89/0xd0 [ 250.211884] io_run_task_work_sig+0x50/0xc0 [ 250.212464] io_sqe_files_unregister+0xb2/0x1f0 [ 250.213109] __io_uring_register+0x115a/0x1750 [ 250.213718] ? __x64_sys_io_uring_register+0xad/0x210 [ 250.214395] ? __fget_files+0x15a/0x260 [ 250.214956] __x64_sys_io_uring_register+0xbe/0x210 [ 250.215620] ? trace_hardirqs_on+0x46/0x110 [ 250.216205] do_syscall_64+0x2d/0x40 [ 250.216731] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 250.217455] RIP: 0033:0x7f0fa17e5239 [ 250.218034] Code: 01 00 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01 48 [ 250.220343] RSP: 002b:00007f0fa1eeac48 EFLAGS: 00000246 ORIG_RAX: 00000000000001ab [ 250.221360] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0fa17e5239 [ 250.222272] RDX: 0000000000000000 RSI: 0000000000000003 RDI: 0000000000000008 [ 250.223185] RBP: 00007f0fa1eeae20 R08: 0000000000000000 R09: 0000000000000000 [ 250.224091] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 [ 250.224999] R13: 0000000000021000 R14: 0000000000000000 R15: 00007f0fa1eeb700 This is caused by calling io_run_task_work_sig() to do work under uring_lock while the caller io_sqe_files_unregister() already held uring_lock. To fix this issue, briefly drop uring_lock when calling io_run_task_work_sig(), and there are two things to concern: - hold uring_lock in io_ring_ctx_free() around io_sqe_files_unregister() this is for consistency of lock/unlock. - add new fixed rsrc ref node before dropping uring_lock it's not safe to do io_uring_enter-->percpu_ref_get() with a dying one. - check if rsrc_data->refs is dying to avoid parallel io_sqe_files_unregister Reported-by: Abaci <abaci@linux.alibaba.com> Fixes: 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs") Suggested-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Hao Xu <haoxu@linux.alibaba.com> [axboe: fixes from Pavel folded in] Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-02-19 17:19:36 +08:00
return off;
io_uring: don't hold uring_lock when calling io_run_task_work* Abaci reported the below issue: [ 141.400455] hrtimer: interrupt took 205853 ns [ 189.869316] process 'usr/local/ilogtail/ilogtail_0.16.26' started with executable stack [ 250.188042] [ 250.188327] ============================================ [ 250.189015] WARNING: possible recursive locking detected [ 250.189732] 5.11.0-rc4 #1 Not tainted [ 250.190267] -------------------------------------------- [ 250.190917] a.out/7363 is trying to acquire lock: [ 250.191506] ffff888114dbcbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __io_req_task_submit+0x29/0xa0 [ 250.192599] [ 250.192599] but task is already holding lock: [ 250.193309] ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.194426] [ 250.194426] other info that might help us debug this: [ 250.195238] Possible unsafe locking scenario: [ 250.195238] [ 250.196019] CPU0 [ 250.196411] ---- [ 250.196803] lock(&ctx->uring_lock); [ 250.197420] lock(&ctx->uring_lock); [ 250.197966] [ 250.197966] *** DEADLOCK *** [ 250.197966] [ 250.198837] May be due to missing lock nesting notation [ 250.198837] [ 250.199780] 1 lock held by a.out/7363: [ 250.200373] #0: ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.201645] [ 250.201645] stack backtrace: [ 250.202298] CPU: 0 PID: 7363 Comm: a.out Not tainted 5.11.0-rc4 #1 [ 250.203144] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 250.203887] Call Trace: [ 250.204302] dump_stack+0xac/0xe3 [ 250.204804] __lock_acquire+0xab6/0x13a0 [ 250.205392] lock_acquire+0x2c3/0x390 [ 250.205928] ? __io_req_task_submit+0x29/0xa0 [ 250.206541] __mutex_lock+0xae/0x9f0 [ 250.207071] ? __io_req_task_submit+0x29/0xa0 [ 250.207745] ? 0xffffffffa0006083 [ 250.208248] ? __io_req_task_submit+0x29/0xa0 [ 250.208845] ? __io_req_task_submit+0x29/0xa0 [ 250.209452] ? __io_req_task_submit+0x5/0xa0 [ 250.210083] __io_req_task_submit+0x29/0xa0 [ 250.210687] io_async_task_func+0x23d/0x4c0 [ 250.211278] task_work_run+0x89/0xd0 [ 250.211884] io_run_task_work_sig+0x50/0xc0 [ 250.212464] io_sqe_files_unregister+0xb2/0x1f0 [ 250.213109] __io_uring_register+0x115a/0x1750 [ 250.213718] ? __x64_sys_io_uring_register+0xad/0x210 [ 250.214395] ? __fget_files+0x15a/0x260 [ 250.214956] __x64_sys_io_uring_register+0xbe/0x210 [ 250.215620] ? trace_hardirqs_on+0x46/0x110 [ 250.216205] do_syscall_64+0x2d/0x40 [ 250.216731] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 250.217455] RIP: 0033:0x7f0fa17e5239 [ 250.218034] Code: 01 00 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01 48 [ 250.220343] RSP: 002b:00007f0fa1eeac48 EFLAGS: 00000246 ORIG_RAX: 00000000000001ab [ 250.221360] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0fa17e5239 [ 250.222272] RDX: 0000000000000000 RSI: 0000000000000003 RDI: 0000000000000008 [ 250.223185] RBP: 00007f0fa1eeae20 R08: 0000000000000000 R09: 0000000000000000 [ 250.224091] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 [ 250.224999] R13: 0000000000021000 R14: 0000000000000000 R15: 00007f0fa1eeb700 This is caused by calling io_run_task_work_sig() to do work under uring_lock while the caller io_sqe_files_unregister() already held uring_lock. To fix this issue, briefly drop uring_lock when calling io_run_task_work_sig(), and there are two things to concern: - hold uring_lock in io_ring_ctx_free() around io_sqe_files_unregister() this is for consistency of lock/unlock. - add new fixed rsrc ref node before dropping uring_lock it's not safe to do io_uring_enter-->percpu_ref_get() with a dying one. - check if rsrc_data->refs is dying to avoid parallel io_sqe_files_unregister Reported-by: Abaci <abaci@linux.alibaba.com> Fixes: 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs") Suggested-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Hao Xu <haoxu@linux.alibaba.com> [axboe: fixes from Pavel folded in] Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-02-19 17:19:36 +08:00
}
static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg,
unsigned int eventfd_async)
io_uring: don't hold uring_lock when calling io_run_task_work* Abaci reported the below issue: [ 141.400455] hrtimer: interrupt took 205853 ns [ 189.869316] process 'usr/local/ilogtail/ilogtail_0.16.26' started with executable stack [ 250.188042] [ 250.188327] ============================================ [ 250.189015] WARNING: possible recursive locking detected [ 250.189732] 5.11.0-rc4 #1 Not tainted [ 250.190267] -------------------------------------------- [ 250.190917] a.out/7363 is trying to acquire lock: [ 250.191506] ffff888114dbcbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __io_req_task_submit+0x29/0xa0 [ 250.192599] [ 250.192599] but task is already holding lock: [ 250.193309] ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.194426] [ 250.194426] other info that might help us debug this: [ 250.195238] Possible unsafe locking scenario: [ 250.195238] [ 250.196019] CPU0 [ 250.196411] ---- [ 250.196803] lock(&ctx->uring_lock); [ 250.197420] lock(&ctx->uring_lock); [ 250.197966] [ 250.197966] *** DEADLOCK *** [ 250.197966] [ 250.198837] May be due to missing lock nesting notation [ 250.198837] [ 250.199780] 1 lock held by a.out/7363: [ 250.200373] #0: ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.201645] [ 250.201645] stack backtrace: [ 250.202298] CPU: 0 PID: 7363 Comm: a.out Not tainted 5.11.0-rc4 #1 [ 250.203144] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 250.203887] Call Trace: [ 250.204302] dump_stack+0xac/0xe3 [ 250.204804] __lock_acquire+0xab6/0x13a0 [ 250.205392] lock_acquire+0x2c3/0x390 [ 250.205928] ? __io_req_task_submit+0x29/0xa0 [ 250.206541] __mutex_lock+0xae/0x9f0 [ 250.207071] ? __io_req_task_submit+0x29/0xa0 [ 250.207745] ? 0xffffffffa0006083 [ 250.208248] ? __io_req_task_submit+0x29/0xa0 [ 250.208845] ? __io_req_task_submit+0x29/0xa0 [ 250.209452] ? __io_req_task_submit+0x5/0xa0 [ 250.210083] __io_req_task_submit+0x29/0xa0 [ 250.210687] io_async_task_func+0x23d/0x4c0 [ 250.211278] task_work_run+0x89/0xd0 [ 250.211884] io_run_task_work_sig+0x50/0xc0 [ 250.212464] io_sqe_files_unregister+0xb2/0x1f0 [ 250.213109] __io_uring_register+0x115a/0x1750 [ 250.213718] ? __x64_sys_io_uring_register+0xad/0x210 [ 250.214395] ? __fget_files+0x15a/0x260 [ 250.214956] __x64_sys_io_uring_register+0xbe/0x210 [ 250.215620] ? trace_hardirqs_on+0x46/0x110 [ 250.216205] do_syscall_64+0x2d/0x40 [ 250.216731] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 250.217455] RIP: 0033:0x7f0fa17e5239 [ 250.218034] Code: 01 00 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01 48 [ 250.220343] RSP: 002b:00007f0fa1eeac48 EFLAGS: 00000246 ORIG_RAX: 00000000000001ab [ 250.221360] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0fa17e5239 [ 250.222272] RDX: 0000000000000000 RSI: 0000000000000003 RDI: 0000000000000008 [ 250.223185] RBP: 00007f0fa1eeae20 R08: 0000000000000000 R09: 0000000000000000 [ 250.224091] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 [ 250.224999] R13: 0000000000021000 R14: 0000000000000000 R15: 00007f0fa1eeb700 This is caused by calling io_run_task_work_sig() to do work under uring_lock while the caller io_sqe_files_unregister() already held uring_lock. To fix this issue, briefly drop uring_lock when calling io_run_task_work_sig(), and there are two things to concern: - hold uring_lock in io_ring_ctx_free() around io_sqe_files_unregister() this is for consistency of lock/unlock. - add new fixed rsrc ref node before dropping uring_lock it's not safe to do io_uring_enter-->percpu_ref_get() with a dying one. - check if rsrc_data->refs is dying to avoid parallel io_sqe_files_unregister Reported-by: Abaci <abaci@linux.alibaba.com> Fixes: 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs") Suggested-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Hao Xu <haoxu@linux.alibaba.com> [axboe: fixes from Pavel folded in] Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-02-19 17:19:36 +08:00
{
struct io_ev_fd *ev_fd;
__s32 __user *fds = arg;
int fd;
ev_fd = rcu_dereference_protected(ctx->io_ev_fd,
lockdep_is_held(&ctx->uring_lock));
if (ev_fd)
return -EBUSY;
io_uring: don't hold uring_lock when calling io_run_task_work* Abaci reported the below issue: [ 141.400455] hrtimer: interrupt took 205853 ns [ 189.869316] process 'usr/local/ilogtail/ilogtail_0.16.26' started with executable stack [ 250.188042] [ 250.188327] ============================================ [ 250.189015] WARNING: possible recursive locking detected [ 250.189732] 5.11.0-rc4 #1 Not tainted [ 250.190267] -------------------------------------------- [ 250.190917] a.out/7363 is trying to acquire lock: [ 250.191506] ffff888114dbcbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __io_req_task_submit+0x29/0xa0 [ 250.192599] [ 250.192599] but task is already holding lock: [ 250.193309] ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.194426] [ 250.194426] other info that might help us debug this: [ 250.195238] Possible unsafe locking scenario: [ 250.195238] [ 250.196019] CPU0 [ 250.196411] ---- [ 250.196803] lock(&ctx->uring_lock); [ 250.197420] lock(&ctx->uring_lock); [ 250.197966] [ 250.197966] *** DEADLOCK *** [ 250.197966] [ 250.198837] May be due to missing lock nesting notation [ 250.198837] [ 250.199780] 1 lock held by a.out/7363: [ 250.200373] #0: ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.201645] [ 250.201645] stack backtrace: [ 250.202298] CPU: 0 PID: 7363 Comm: a.out Not tainted 5.11.0-rc4 #1 [ 250.203144] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 250.203887] Call Trace: [ 250.204302] dump_stack+0xac/0xe3 [ 250.204804] __lock_acquire+0xab6/0x13a0 [ 250.205392] lock_acquire+0x2c3/0x390 [ 250.205928] ? __io_req_task_submit+0x29/0xa0 [ 250.206541] __mutex_lock+0xae/0x9f0 [ 250.207071] ? __io_req_task_submit+0x29/0xa0 [ 250.207745] ? 0xffffffffa0006083 [ 250.208248] ? __io_req_task_submit+0x29/0xa0 [ 250.208845] ? __io_req_task_submit+0x29/0xa0 [ 250.209452] ? __io_req_task_submit+0x5/0xa0 [ 250.210083] __io_req_task_submit+0x29/0xa0 [ 250.210687] io_async_task_func+0x23d/0x4c0 [ 250.211278] task_work_run+0x89/0xd0 [ 250.211884] io_run_task_work_sig+0x50/0xc0 [ 250.212464] io_sqe_files_unregister+0xb2/0x1f0 [ 250.213109] __io_uring_register+0x115a/0x1750 [ 250.213718] ? __x64_sys_io_uring_register+0xad/0x210 [ 250.214395] ? __fget_files+0x15a/0x260 [ 250.214956] __x64_sys_io_uring_register+0xbe/0x210 [ 250.215620] ? trace_hardirqs_on+0x46/0x110 [ 250.216205] do_syscall_64+0x2d/0x40 [ 250.216731] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 250.217455] RIP: 0033:0x7f0fa17e5239 [ 250.218034] Code: 01 00 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01 48 [ 250.220343] RSP: 002b:00007f0fa1eeac48 EFLAGS: 00000246 ORIG_RAX: 00000000000001ab [ 250.221360] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0fa17e5239 [ 250.222272] RDX: 0000000000000000 RSI: 0000000000000003 RDI: 0000000000000008 [ 250.223185] RBP: 00007f0fa1eeae20 R08: 0000000000000000 R09: 0000000000000000 [ 250.224091] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 [ 250.224999] R13: 0000000000021000 R14: 0000000000000000 R15: 00007f0fa1eeb700 This is caused by calling io_run_task_work_sig() to do work under uring_lock while the caller io_sqe_files_unregister() already held uring_lock. To fix this issue, briefly drop uring_lock when calling io_run_task_work_sig(), and there are two things to concern: - hold uring_lock in io_ring_ctx_free() around io_sqe_files_unregister() this is for consistency of lock/unlock. - add new fixed rsrc ref node before dropping uring_lock it's not safe to do io_uring_enter-->percpu_ref_get() with a dying one. - check if rsrc_data->refs is dying to avoid parallel io_sqe_files_unregister Reported-by: Abaci <abaci@linux.alibaba.com> Fixes: 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs") Suggested-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Hao Xu <haoxu@linux.alibaba.com> [axboe: fixes from Pavel folded in] Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-02-19 17:19:36 +08:00
if (copy_from_user(&fd, fds, sizeof(*fds)))
return -EFAULT;
ev_fd = kmalloc(sizeof(*ev_fd), GFP_KERNEL);
if (!ev_fd)
return -ENOMEM;
ev_fd->cq_ev_fd = eventfd_ctx_fdget(fd);
if (IS_ERR(ev_fd->cq_ev_fd)) {
int ret = PTR_ERR(ev_fd->cq_ev_fd);
kfree(ev_fd);
return ret;
}
spin_lock(&ctx->completion_lock);
ctx->evfd_last_cq_tail = ctx->cached_cq_tail;
spin_unlock(&ctx->completion_lock);
ev_fd->eventfd_async = eventfd_async;
ctx->has_evfd = true;
rcu_assign_pointer(ctx->io_ev_fd, ev_fd);
atomic_set(&ev_fd->refs, 1);
atomic_set(&ev_fd->ops, 0);
return 0;
}
static int io_eventfd_unregister(struct io_ring_ctx *ctx)
{
struct io_ev_fd *ev_fd;
ev_fd = rcu_dereference_protected(ctx->io_ev_fd,
lockdep_is_held(&ctx->uring_lock));
if (ev_fd) {
ctx->has_evfd = false;
rcu_assign_pointer(ctx->io_ev_fd, NULL);
if (!atomic_fetch_or(BIT(IO_EVENTFD_OP_FREE_BIT), &ev_fd->ops))
call_rcu(&ev_fd->rcu, io_eventfd_ops);
return 0;
}
return -ENXIO;
}
static void io_req_caches_free(struct io_ring_ctx *ctx)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_kiocb *req;
int nr = 0;
mutex_lock(&ctx->uring_lock);
io_flush_cached_locked_reqs(ctx, &ctx->submit_state);
while (!io_req_cache_empty(ctx)) {
req = io_extract_req(ctx);
kmem_cache_free(req_cachep, req);
nr++;
}
if (nr)
percpu_ref_put_many(&ctx->refs, nr);
mutex_unlock(&ctx->uring_lock);
}
static void io_rsrc_node_cache_free(struct io_cache_entry *entry)
{
kfree(container_of(entry, struct io_rsrc_node, cache));
}
static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
io_sq_thread_finish(ctx);
/* __io_rsrc_put_work() may need uring_lock to progress, wait w/o it */
if (WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)))
return;
io_uring: don't hold uring_lock when calling io_run_task_work* Abaci reported the below issue: [ 141.400455] hrtimer: interrupt took 205853 ns [ 189.869316] process 'usr/local/ilogtail/ilogtail_0.16.26' started with executable stack [ 250.188042] [ 250.188327] ============================================ [ 250.189015] WARNING: possible recursive locking detected [ 250.189732] 5.11.0-rc4 #1 Not tainted [ 250.190267] -------------------------------------------- [ 250.190917] a.out/7363 is trying to acquire lock: [ 250.191506] ffff888114dbcbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __io_req_task_submit+0x29/0xa0 [ 250.192599] [ 250.192599] but task is already holding lock: [ 250.193309] ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.194426] [ 250.194426] other info that might help us debug this: [ 250.195238] Possible unsafe locking scenario: [ 250.195238] [ 250.196019] CPU0 [ 250.196411] ---- [ 250.196803] lock(&ctx->uring_lock); [ 250.197420] lock(&ctx->uring_lock); [ 250.197966] [ 250.197966] *** DEADLOCK *** [ 250.197966] [ 250.198837] May be due to missing lock nesting notation [ 250.198837] [ 250.199780] 1 lock held by a.out/7363: [ 250.200373] #0: ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210 [ 250.201645] [ 250.201645] stack backtrace: [ 250.202298] CPU: 0 PID: 7363 Comm: a.out Not tainted 5.11.0-rc4 #1 [ 250.203144] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 250.203887] Call Trace: [ 250.204302] dump_stack+0xac/0xe3 [ 250.204804] __lock_acquire+0xab6/0x13a0 [ 250.205392] lock_acquire+0x2c3/0x390 [ 250.205928] ? __io_req_task_submit+0x29/0xa0 [ 250.206541] __mutex_lock+0xae/0x9f0 [ 250.207071] ? __io_req_task_submit+0x29/0xa0 [ 250.207745] ? 0xffffffffa0006083 [ 250.208248] ? __io_req_task_submit+0x29/0xa0 [ 250.208845] ? __io_req_task_submit+0x29/0xa0 [ 250.209452] ? __io_req_task_submit+0x5/0xa0 [ 250.210083] __io_req_task_submit+0x29/0xa0 [ 250.210687] io_async_task_func+0x23d/0x4c0 [ 250.211278] task_work_run+0x89/0xd0 [ 250.211884] io_run_task_work_sig+0x50/0xc0 [ 250.212464] io_sqe_files_unregister+0xb2/0x1f0 [ 250.213109] __io_uring_register+0x115a/0x1750 [ 250.213718] ? __x64_sys_io_uring_register+0xad/0x210 [ 250.214395] ? __fget_files+0x15a/0x260 [ 250.214956] __x64_sys_io_uring_register+0xbe/0x210 [ 250.215620] ? trace_hardirqs_on+0x46/0x110 [ 250.216205] do_syscall_64+0x2d/0x40 [ 250.216731] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 250.217455] RIP: 0033:0x7f0fa17e5239 [ 250.218034] Code: 01 00 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01 48 [ 250.220343] RSP: 002b:00007f0fa1eeac48 EFLAGS: 00000246 ORIG_RAX: 00000000000001ab [ 250.221360] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0fa17e5239 [ 250.222272] RDX: 0000000000000000 RSI: 0000000000000003 RDI: 0000000000000008 [ 250.223185] RBP: 00007f0fa1eeae20 R08: 0000000000000000 R09: 0000000000000000 [ 250.224091] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 [ 250.224999] R13: 0000000000021000 R14: 0000000000000000 R15: 00007f0fa1eeb700 This is caused by calling io_run_task_work_sig() to do work under uring_lock while the caller io_sqe_files_unregister() already held uring_lock. To fix this issue, briefly drop uring_lock when calling io_run_task_work_sig(), and there are two things to concern: - hold uring_lock in io_ring_ctx_free() around io_sqe_files_unregister() this is for consistency of lock/unlock. - add new fixed rsrc ref node before dropping uring_lock it's not safe to do io_uring_enter-->percpu_ref_get() with a dying one. - check if rsrc_data->refs is dying to avoid parallel io_sqe_files_unregister Reported-by: Abaci <abaci@linux.alibaba.com> Fixes: 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs") Suggested-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Hao Xu <haoxu@linux.alibaba.com> [axboe: fixes from Pavel folded in] Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-02-19 17:19:36 +08:00
mutex_lock(&ctx->uring_lock);
if (ctx->buf_data)
__io_sqe_buffers_unregister(ctx);
if (ctx->file_data)
__io_sqe_files_unregister(ctx);
io_cqring_overflow_kill(ctx);
io_eventfd_unregister(ctx);
io_alloc_cache_free(&ctx->apoll_cache, io_apoll_cache_free);
io_alloc_cache_free(&ctx->netmsg_cache, io_netmsg_cache_free);
io_destroy_buffers(ctx);
mutex_unlock(&ctx->uring_lock);
if (ctx->sq_creds)
put_cred(ctx->sq_creds);
if (ctx->submitter_task)
put_task_struct(ctx->submitter_task);
/* there are no registered resources left, nobody uses it */
if (ctx->rsrc_node)
io_rsrc_node_destroy(ctx, ctx->rsrc_node);
WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list));
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#if defined(CONFIG_UNIX)
if (ctx->ring_sock) {
ctx->ring_sock->file = NULL; /* so that iput() is called */
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
sock_release(ctx->ring_sock);
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#endif
WARN_ON_ONCE(!list_empty(&ctx->ltimeout_list));
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
io_alloc_cache_free(&ctx->rsrc_node_cache, io_rsrc_node_cache_free);
if (ctx->mm_account) {
mmdrop(ctx->mm_account);
ctx->mm_account = NULL;
}
io_rings_free(ctx);
io_kbuf_mmap_list_free(ctx);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
percpu_ref_exit(&ctx->refs);
free_uid(ctx->user);
io_req_caches_free(ctx);
if (ctx->hash_map)
io_wq_put_hash(ctx->hash_map);
kfree(ctx->cancel_table.hbs);
kfree(ctx->cancel_table_locked.hbs);
kfree(ctx->io_bl);
xa_destroy(&ctx->io_bl_xa);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
kfree(ctx);
}
static __cold void io_activate_pollwq_cb(struct callback_head *cb)
{
struct io_ring_ctx *ctx = container_of(cb, struct io_ring_ctx,
poll_wq_task_work);
mutex_lock(&ctx->uring_lock);
ctx->poll_activated = true;
mutex_unlock(&ctx->uring_lock);
/*
* Wake ups for some events between start of polling and activation
* might've been lost due to loose synchronisation.
*/
wake_up_all(&ctx->poll_wq);
percpu_ref_put(&ctx->refs);
}
static __cold void io_activate_pollwq(struct io_ring_ctx *ctx)
{
spin_lock(&ctx->completion_lock);
/* already activated or in progress */
if (ctx->poll_activated || ctx->poll_wq_task_work.func)
goto out;
if (WARN_ON_ONCE(!ctx->task_complete))
goto out;
if (!ctx->submitter_task)
goto out;
/*
* with ->submitter_task only the submitter task completes requests, we
* only need to sync with it, which is done by injecting a tw
*/
init_task_work(&ctx->poll_wq_task_work, io_activate_pollwq_cb);
percpu_ref_get(&ctx->refs);
if (task_work_add(ctx->submitter_task, &ctx->poll_wq_task_work, TWA_SIGNAL))
percpu_ref_put(&ctx->refs);
out:
spin_unlock(&ctx->completion_lock);
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
static __poll_t io_uring_poll(struct file *file, poll_table *wait)
{
struct io_ring_ctx *ctx = file->private_data;
__poll_t mask = 0;
if (unlikely(!ctx->poll_activated))
io_activate_pollwq(ctx);
poll_wait(file, &ctx->poll_wq, wait);
/*
* synchronizes with barrier from wq_has_sleeper call in
* io_commit_cqring
*/
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
smp_rmb();
if (!io_sqring_full(ctx))
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
mask |= EPOLLOUT | EPOLLWRNORM;
io_uring: fix possible deadlock in io_uring_poll Abaci reported follow issue: [ 30.615891] ====================================================== [ 30.616648] WARNING: possible circular locking dependency detected [ 30.617423] 5.11.0-rc3-next-20210115 #1 Not tainted [ 30.618035] ------------------------------------------------------ [ 30.618914] a.out/1128 is trying to acquire lock: [ 30.619520] ffff88810b063868 (&ep->mtx){+.+.}-{3:3}, at: __ep_eventpoll_poll+0x9f/0x220 [ 30.620505] [ 30.620505] but task is already holding lock: [ 30.621218] ffff88810e952be8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_enter+0x3f0/0x5b0 [ 30.622349] [ 30.622349] which lock already depends on the new lock. [ 30.622349] [ 30.623289] [ 30.623289] the existing dependency chain (in reverse order) is: [ 30.624243] [ 30.624243] -> #1 (&ctx->uring_lock){+.+.}-{3:3}: [ 30.625263] lock_acquire+0x2c7/0x390 [ 30.625868] __mutex_lock+0xae/0x9f0 [ 30.626451] io_cqring_overflow_flush.part.95+0x6d/0x70 [ 30.627278] io_uring_poll+0xcb/0xd0 [ 30.627890] ep_item_poll.isra.14+0x4e/0x90 [ 30.628531] do_epoll_ctl+0xb7e/0x1120 [ 30.629122] __x64_sys_epoll_ctl+0x70/0xb0 [ 30.629770] do_syscall_64+0x2d/0x40 [ 30.630332] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 30.631187] [ 30.631187] -> #0 (&ep->mtx){+.+.}-{3:3}: [ 30.631985] check_prevs_add+0x226/0xb00 [ 30.632584] __lock_acquire+0x1237/0x13a0 [ 30.633207] lock_acquire+0x2c7/0x390 [ 30.633740] __mutex_lock+0xae/0x9f0 [ 30.634258] __ep_eventpoll_poll+0x9f/0x220 [ 30.634879] __io_arm_poll_handler+0xbf/0x220 [ 30.635462] io_issue_sqe+0xa6b/0x13e0 [ 30.635982] __io_queue_sqe+0x10b/0x550 [ 30.636648] io_queue_sqe+0x235/0x470 [ 30.637281] io_submit_sqes+0xcce/0xf10 [ 30.637839] __x64_sys_io_uring_enter+0x3fb/0x5b0 [ 30.638465] do_syscall_64+0x2d/0x40 [ 30.638999] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 30.639643] [ 30.639643] other info that might help us debug this: [ 30.639643] [ 30.640618] Possible unsafe locking scenario: [ 30.640618] [ 30.641402] CPU0 CPU1 [ 30.641938] ---- ---- [ 30.642664] lock(&ctx->uring_lock); [ 30.643425] lock(&ep->mtx); [ 30.644498] lock(&ctx->uring_lock); [ 30.645668] lock(&ep->mtx); [ 30.646321] [ 30.646321] *** DEADLOCK *** [ 30.646321] [ 30.647642] 1 lock held by a.out/1128: [ 30.648424] #0: ffff88810e952be8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_enter+0x3f0/0x5b0 [ 30.649954] [ 30.649954] stack backtrace: [ 30.650592] CPU: 1 PID: 1128 Comm: a.out Not tainted 5.11.0-rc3-next-20210115 #1 [ 30.651554] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 30.652290] Call Trace: [ 30.652688] dump_stack+0xac/0xe3 [ 30.653164] check_noncircular+0x11e/0x130 [ 30.653747] ? check_prevs_add+0x226/0xb00 [ 30.654303] check_prevs_add+0x226/0xb00 [ 30.654845] ? add_lock_to_list.constprop.49+0xac/0x1d0 [ 30.655564] __lock_acquire+0x1237/0x13a0 [ 30.656262] lock_acquire+0x2c7/0x390 [ 30.656788] ? __ep_eventpoll_poll+0x9f/0x220 [ 30.657379] ? __io_queue_proc.isra.88+0x180/0x180 [ 30.658014] __mutex_lock+0xae/0x9f0 [ 30.658524] ? __ep_eventpoll_poll+0x9f/0x220 [ 30.659112] ? mark_held_locks+0x5a/0x80 [ 30.659648] ? __ep_eventpoll_poll+0x9f/0x220 [ 30.660229] ? _raw_spin_unlock_irqrestore+0x2d/0x40 [ 30.660885] ? trace_hardirqs_on+0x46/0x110 [ 30.661471] ? __io_queue_proc.isra.88+0x180/0x180 [ 30.662102] ? __ep_eventpoll_poll+0x9f/0x220 [ 30.662696] __ep_eventpoll_poll+0x9f/0x220 [ 30.663273] ? __ep_eventpoll_poll+0x220/0x220 [ 30.663875] __io_arm_poll_handler+0xbf/0x220 [ 30.664463] io_issue_sqe+0xa6b/0x13e0 [ 30.664984] ? __lock_acquire+0x782/0x13a0 [ 30.665544] ? __io_queue_proc.isra.88+0x180/0x180 [ 30.666170] ? __io_queue_sqe+0x10b/0x550 [ 30.666725] __io_queue_sqe+0x10b/0x550 [ 30.667252] ? __fget_files+0x131/0x260 [ 30.667791] ? io_req_prep+0xd8/0x1090 [ 30.668316] ? io_queue_sqe+0x235/0x470 [ 30.668868] io_queue_sqe+0x235/0x470 [ 30.669398] io_submit_sqes+0xcce/0xf10 [ 30.669931] ? xa_load+0xe4/0x1c0 [ 30.670425] __x64_sys_io_uring_enter+0x3fb/0x5b0 [ 30.671051] ? lockdep_hardirqs_on_prepare+0xde/0x180 [ 30.671719] ? syscall_enter_from_user_mode+0x2b/0x80 [ 30.672380] do_syscall_64+0x2d/0x40 [ 30.672901] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 30.673503] RIP: 0033:0x7fd89c813239 [ 30.673962] Code: 01 00 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01 48 [ 30.675920] RSP: 002b:00007ffc65a7c628 EFLAGS: 00000217 ORIG_RAX: 00000000000001aa [ 30.676791] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fd89c813239 [ 30.677594] RDX: 0000000000000000 RSI: 0000000000000014 RDI: 0000000000000003 [ 30.678678] RBP: 00007ffc65a7c720 R08: 0000000000000000 R09: 0000000003000000 [ 30.679492] R10: 0000000000000000 R11: 0000000000000217 R12: 0000000000400ff0 [ 30.680282] R13: 00007ffc65a7c840 R14: 0000000000000000 R15: 0000000000000000 This might happen if we do epoll_wait on a uring fd while reading/writing the former epoll fd in a sqe in the former uring instance. So let's don't flush cqring overflow list, just do a simple check. Reported-by: Abaci <abaci@linux.alibaba.com> Fixes: 6c503150ae33 ("io_uring: patch up IOPOLL overflow_flush sync") Signed-off-by: Hao Xu <haoxu@linux.alibaba.com> Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-02-05 16:34:21 +08:00
/*
* Don't flush cqring overflow list here, just do a simple check.
* Otherwise there could possible be ABBA deadlock:
* CPU0 CPU1
* ---- ----
* lock(&ctx->uring_lock);
* lock(&ep->mtx);
* lock(&ctx->uring_lock);
* lock(&ep->mtx);
*
* Users may get EPOLLIN meanwhile seeing nothing in cqring, this
* pushes them to do the flush.
io_uring: fix possible deadlock in io_uring_poll Abaci reported follow issue: [ 30.615891] ====================================================== [ 30.616648] WARNING: possible circular locking dependency detected [ 30.617423] 5.11.0-rc3-next-20210115 #1 Not tainted [ 30.618035] ------------------------------------------------------ [ 30.618914] a.out/1128 is trying to acquire lock: [ 30.619520] ffff88810b063868 (&ep->mtx){+.+.}-{3:3}, at: __ep_eventpoll_poll+0x9f/0x220 [ 30.620505] [ 30.620505] but task is already holding lock: [ 30.621218] ffff88810e952be8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_enter+0x3f0/0x5b0 [ 30.622349] [ 30.622349] which lock already depends on the new lock. [ 30.622349] [ 30.623289] [ 30.623289] the existing dependency chain (in reverse order) is: [ 30.624243] [ 30.624243] -> #1 (&ctx->uring_lock){+.+.}-{3:3}: [ 30.625263] lock_acquire+0x2c7/0x390 [ 30.625868] __mutex_lock+0xae/0x9f0 [ 30.626451] io_cqring_overflow_flush.part.95+0x6d/0x70 [ 30.627278] io_uring_poll+0xcb/0xd0 [ 30.627890] ep_item_poll.isra.14+0x4e/0x90 [ 30.628531] do_epoll_ctl+0xb7e/0x1120 [ 30.629122] __x64_sys_epoll_ctl+0x70/0xb0 [ 30.629770] do_syscall_64+0x2d/0x40 [ 30.630332] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 30.631187] [ 30.631187] -> #0 (&ep->mtx){+.+.}-{3:3}: [ 30.631985] check_prevs_add+0x226/0xb00 [ 30.632584] __lock_acquire+0x1237/0x13a0 [ 30.633207] lock_acquire+0x2c7/0x390 [ 30.633740] __mutex_lock+0xae/0x9f0 [ 30.634258] __ep_eventpoll_poll+0x9f/0x220 [ 30.634879] __io_arm_poll_handler+0xbf/0x220 [ 30.635462] io_issue_sqe+0xa6b/0x13e0 [ 30.635982] __io_queue_sqe+0x10b/0x550 [ 30.636648] io_queue_sqe+0x235/0x470 [ 30.637281] io_submit_sqes+0xcce/0xf10 [ 30.637839] __x64_sys_io_uring_enter+0x3fb/0x5b0 [ 30.638465] do_syscall_64+0x2d/0x40 [ 30.638999] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 30.639643] [ 30.639643] other info that might help us debug this: [ 30.639643] [ 30.640618] Possible unsafe locking scenario: [ 30.640618] [ 30.641402] CPU0 CPU1 [ 30.641938] ---- ---- [ 30.642664] lock(&ctx->uring_lock); [ 30.643425] lock(&ep->mtx); [ 30.644498] lock(&ctx->uring_lock); [ 30.645668] lock(&ep->mtx); [ 30.646321] [ 30.646321] *** DEADLOCK *** [ 30.646321] [ 30.647642] 1 lock held by a.out/1128: [ 30.648424] #0: ffff88810e952be8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_enter+0x3f0/0x5b0 [ 30.649954] [ 30.649954] stack backtrace: [ 30.650592] CPU: 1 PID: 1128 Comm: a.out Not tainted 5.11.0-rc3-next-20210115 #1 [ 30.651554] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 30.652290] Call Trace: [ 30.652688] dump_stack+0xac/0xe3 [ 30.653164] check_noncircular+0x11e/0x130 [ 30.653747] ? check_prevs_add+0x226/0xb00 [ 30.654303] check_prevs_add+0x226/0xb00 [ 30.654845] ? add_lock_to_list.constprop.49+0xac/0x1d0 [ 30.655564] __lock_acquire+0x1237/0x13a0 [ 30.656262] lock_acquire+0x2c7/0x390 [ 30.656788] ? __ep_eventpoll_poll+0x9f/0x220 [ 30.657379] ? __io_queue_proc.isra.88+0x180/0x180 [ 30.658014] __mutex_lock+0xae/0x9f0 [ 30.658524] ? __ep_eventpoll_poll+0x9f/0x220 [ 30.659112] ? mark_held_locks+0x5a/0x80 [ 30.659648] ? __ep_eventpoll_poll+0x9f/0x220 [ 30.660229] ? _raw_spin_unlock_irqrestore+0x2d/0x40 [ 30.660885] ? trace_hardirqs_on+0x46/0x110 [ 30.661471] ? __io_queue_proc.isra.88+0x180/0x180 [ 30.662102] ? __ep_eventpoll_poll+0x9f/0x220 [ 30.662696] __ep_eventpoll_poll+0x9f/0x220 [ 30.663273] ? __ep_eventpoll_poll+0x220/0x220 [ 30.663875] __io_arm_poll_handler+0xbf/0x220 [ 30.664463] io_issue_sqe+0xa6b/0x13e0 [ 30.664984] ? __lock_acquire+0x782/0x13a0 [ 30.665544] ? __io_queue_proc.isra.88+0x180/0x180 [ 30.666170] ? __io_queue_sqe+0x10b/0x550 [ 30.666725] __io_queue_sqe+0x10b/0x550 [ 30.667252] ? __fget_files+0x131/0x260 [ 30.667791] ? io_req_prep+0xd8/0x1090 [ 30.668316] ? io_queue_sqe+0x235/0x470 [ 30.668868] io_queue_sqe+0x235/0x470 [ 30.669398] io_submit_sqes+0xcce/0xf10 [ 30.669931] ? xa_load+0xe4/0x1c0 [ 30.670425] __x64_sys_io_uring_enter+0x3fb/0x5b0 [ 30.671051] ? lockdep_hardirqs_on_prepare+0xde/0x180 [ 30.671719] ? syscall_enter_from_user_mode+0x2b/0x80 [ 30.672380] do_syscall_64+0x2d/0x40 [ 30.672901] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 30.673503] RIP: 0033:0x7fd89c813239 [ 30.673962] Code: 01 00 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01 48 [ 30.675920] RSP: 002b:00007ffc65a7c628 EFLAGS: 00000217 ORIG_RAX: 00000000000001aa [ 30.676791] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fd89c813239 [ 30.677594] RDX: 0000000000000000 RSI: 0000000000000014 RDI: 0000000000000003 [ 30.678678] RBP: 00007ffc65a7c720 R08: 0000000000000000 R09: 0000000003000000 [ 30.679492] R10: 0000000000000000 R11: 0000000000000217 R12: 0000000000400ff0 [ 30.680282] R13: 00007ffc65a7c840 R14: 0000000000000000 R15: 0000000000000000 This might happen if we do epoll_wait on a uring fd while reading/writing the former epoll fd in a sqe in the former uring instance. So let's don't flush cqring overflow list, just do a simple check. Reported-by: Abaci <abaci@linux.alibaba.com> Fixes: 6c503150ae33 ("io_uring: patch up IOPOLL overflow_flush sync") Signed-off-by: Hao Xu <haoxu@linux.alibaba.com> Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-02-05 16:34:21 +08:00
*/
if (__io_cqring_events_user(ctx) || io_has_work(ctx))
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
mask |= EPOLLIN | EPOLLRDNORM;
return mask;
}
static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
{
const struct cred *creds;
creds = xa_erase(&ctx->personalities, id);
if (creds) {
put_cred(creds);
return 0;
}
return -EINVAL;
}
struct io_tctx_exit {
struct callback_head task_work;
struct completion completion;
struct io_ring_ctx *ctx;
};
static __cold void io_tctx_exit_cb(struct callback_head *cb)
{
struct io_uring_task *tctx = current->io_uring;
struct io_tctx_exit *work;
work = container_of(cb, struct io_tctx_exit, task_work);
/*
* When @in_cancel, we're in cancellation and it's racy to remove the
* node. It'll be removed by the end of cancellation, just ignore it.
io_uring: Fix a null-ptr-deref in io_tctx_exit_cb() Syzkaller reports a NULL deref bug as follows: BUG: KASAN: null-ptr-deref in io_tctx_exit_cb+0x53/0xd3 Read of size 4 at addr 0000000000000138 by task file1/1955 CPU: 1 PID: 1955 Comm: file1 Not tainted 6.1.0-rc7-00103-gef4d3ea40565 #75 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-2.el7 04/01/2014 Call Trace: <TASK> dump_stack_lvl+0xcd/0x134 ? io_tctx_exit_cb+0x53/0xd3 kasan_report+0xbb/0x1f0 ? io_tctx_exit_cb+0x53/0xd3 kasan_check_range+0x140/0x190 io_tctx_exit_cb+0x53/0xd3 task_work_run+0x164/0x250 ? task_work_cancel+0x30/0x30 get_signal+0x1c3/0x2440 ? lock_downgrade+0x6e0/0x6e0 ? lock_downgrade+0x6e0/0x6e0 ? exit_signals+0x8b0/0x8b0 ? do_raw_read_unlock+0x3b/0x70 ? do_raw_spin_unlock+0x50/0x230 arch_do_signal_or_restart+0x82/0x2470 ? kmem_cache_free+0x260/0x4b0 ? putname+0xfe/0x140 ? get_sigframe_size+0x10/0x10 ? do_execveat_common.isra.0+0x226/0x710 ? lockdep_hardirqs_on+0x79/0x100 ? putname+0xfe/0x140 ? do_execveat_common.isra.0+0x238/0x710 exit_to_user_mode_prepare+0x15f/0x250 syscall_exit_to_user_mode+0x19/0x50 do_syscall_64+0x42/0xb0 entry_SYSCALL_64_after_hwframe+0x63/0xcd RIP: 0023:0x0 Code: Unable to access opcode bytes at 0xffffffffffffffd6. RSP: 002b:00000000fffb7790 EFLAGS: 00000200 ORIG_RAX: 000000000000000b RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000 RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000 RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000 </TASK> Kernel panic - not syncing: panic_on_warn set ... This happens because the adding of task_work from io_ring_exit_work() isn't synchronized with canceling all work items from eg exec. The execution of the two are ordered in that they are both run by the task itself, but if io_tctx_exit_cb() is queued while we're canceling all work items off exec AND gets executed when the task exits to userspace rather than in the main loop in io_uring_cancel_generic(), then we can find current->io_uring == NULL and hit the above crash. It's safe to add this NULL check here, because the execution of the two paths are done by the task itself. Cc: stable@vger.kernel.org Fixes: d56d938b4bef ("io_uring: do ctx initiated file note removal") Reported-by: syzkaller <syzkaller@googlegroups.com> Signed-off-by: Harshit Mogalapalli <harshit.m.mogalapalli@oracle.com> Link: https://lore.kernel.org/r/20221206093833.3812138-1-harshit.m.mogalapalli@oracle.com [axboe: add code comment and also put an explanation in the commit msg] Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-12-06 17:38:32 +08:00
* tctx can be NULL if the queueing of this task_work raced with
* work cancelation off the exec path.
*/
if (tctx && !atomic_read(&tctx->in_cancel))
io_uring_del_tctx_node((unsigned long)work->ctx);
complete(&work->completion);
}
static __cold bool io_cancel_ctx_cb(struct io_wq_work *work, void *data)
{
struct io_kiocb *req = container_of(work, struct io_kiocb, work);
return req->ctx == data;
}
static __cold void io_ring_exit_work(struct work_struct *work)
{
struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work);
unsigned long timeout = jiffies + HZ * 60 * 5;
unsigned long interval = HZ / 20;
struct io_tctx_exit exit;
struct io_tctx_node *node;
int ret;
/*
* If we're doing polled IO and end up having requests being
* submitted async (out-of-line), then completions can come in while
* we're waiting for refs to drop. We need to reap these manually,
* as nobody else will be looking for them.
*/
do {
if (test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq)) {
mutex_lock(&ctx->uring_lock);
io_cqring_overflow_kill(ctx);
mutex_unlock(&ctx->uring_lock);
}
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN)
io_move_task_work_from_local(ctx);
while (io_uring_try_cancel_requests(ctx, NULL, true))
cond_resched();
if (ctx->sq_data) {
struct io_sq_data *sqd = ctx->sq_data;
struct task_struct *tsk;
io_sq_thread_park(sqd);
tsk = sqd->thread;
if (tsk && tsk->io_uring && tsk->io_uring->io_wq)
io_wq_cancel_cb(tsk->io_uring->io_wq,
io_cancel_ctx_cb, ctx, true);
io_sq_thread_unpark(sqd);
}
io_req_caches_free(ctx);
if (WARN_ON_ONCE(time_after(jiffies, timeout))) {
/* there is little hope left, don't run it too often */
interval = HZ * 60;
}
/*
* This is really an uninterruptible wait, as it has to be
* complete. But it's also run from a kworker, which doesn't
* take signals, so it's fine to make it interruptible. This
* avoids scenarios where we knowingly can wait much longer
* on completions, for example if someone does a SIGSTOP on
* a task that needs to finish task_work to make this loop
* complete. That's a synthetic situation that should not
* cause a stuck task backtrace, and hence a potential panic
* on stuck tasks if that is enabled.
*/
} while (!wait_for_completion_interruptible_timeout(&ctx->ref_comp, interval));
init_completion(&exit.completion);
init_task_work(&exit.task_work, io_tctx_exit_cb);
exit.ctx = ctx;
mutex_lock(&ctx->uring_lock);
while (!list_empty(&ctx->tctx_list)) {
WARN_ON_ONCE(time_after(jiffies, timeout));
node = list_first_entry(&ctx->tctx_list, struct io_tctx_node,
ctx_node);
/* don't spin on a single task if cancellation failed */
list_rotate_left(&ctx->tctx_list);
ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL);
if (WARN_ON_ONCE(ret))
continue;
mutex_unlock(&ctx->uring_lock);
/*
* See comment above for
* wait_for_completion_interruptible_timeout() on why this
* wait is marked as interruptible.
*/
wait_for_completion_interruptible(&exit.completion);
mutex_lock(&ctx->uring_lock);
}
mutex_unlock(&ctx->uring_lock);
spin_lock(&ctx->completion_lock);
spin_unlock(&ctx->completion_lock);
/* pairs with RCU read section in io_req_local_work_add() */
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN)
synchronize_rcu();
io_ring_ctx_free(ctx);
}
static __cold void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
unsigned long index;
struct creds *creds;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
mutex_lock(&ctx->uring_lock);
percpu_ref_kill(&ctx->refs);
xa_for_each(&ctx->personalities, index, creds)
io_unregister_personality(ctx, index);
if (ctx->rings)
io_poll_remove_all(ctx, NULL, true);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
mutex_unlock(&ctx->uring_lock);
/*
* If we failed setting up the ctx, we might not have any rings
* and therefore did not submit any requests
*/
if (ctx->rings)
io_kill_timeouts(ctx, NULL, true);
flush_delayed_work(&ctx->fallback_work);
INIT_WORK(&ctx->exit_work, io_ring_exit_work);
/*
* Use system_unbound_wq to avoid spawning tons of event kworkers
* if we're exiting a ton of rings at the same time. It just adds
* noise and overhead, there's no discernable change in runtime
* over using system_wq.
*/
queue_work(system_unbound_wq, &ctx->exit_work);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
static int io_uring_release(struct inode *inode, struct file *file)
{
struct io_ring_ctx *ctx = file->private_data;
file->private_data = NULL;
io_ring_ctx_wait_and_kill(ctx);
return 0;
}
struct io_task_cancel {
struct task_struct *task;
bool all;
};
static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
{
struct io_kiocb *req = container_of(work, struct io_kiocb, work);
struct io_task_cancel *cancel = data;
io_uring: fix link traversal locking WARNING: inconsistent lock state 5.16.0-rc2-syzkaller #0 Not tainted inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage. ffff888078e11418 (&ctx->timeout_lock ){?.+.}-{2:2} , at: io_timeout_fn+0x6f/0x360 fs/io_uring.c:5943 {HARDIRQ-ON-W} state was registered at: [...] spin_unlock_irq include/linux/spinlock.h:399 [inline] __io_poll_remove_one fs/io_uring.c:5669 [inline] __io_poll_remove_one fs/io_uring.c:5654 [inline] io_poll_remove_one+0x236/0x870 fs/io_uring.c:5680 io_poll_remove_all+0x1af/0x235 fs/io_uring.c:5709 io_ring_ctx_wait_and_kill+0x1cc/0x322 fs/io_uring.c:9534 io_uring_release+0x42/0x46 fs/io_uring.c:9554 __fput+0x286/0x9f0 fs/file_table.c:280 task_work_run+0xdd/0x1a0 kernel/task_work.c:164 exit_task_work include/linux/task_work.h:32 [inline] do_exit+0xc14/0x2b40 kernel/exit.c:832 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") fixed a data race but introduced a possible deadlock and inconsistentcy in irq states. E.g. io_poll_remove_all() spin_lock_irq(timeout_lock) io_poll_remove_one() spin_lock/unlock_irq(poll_lock); spin_unlock_irq(timeout_lock) Another type of problem is freeing a request while holding ->timeout_lock, which may leads to a deadlock in io_commit_cqring() -> io_flush_timeouts() and other places. Having 3 nested locks is also too ugly. Add io_match_task_safe(), which would briefly take and release timeout_lock for race prevention inside, so the actuall request cancellation / free / etc. code doesn't have it taken. Reported-by: syzbot+ff49a3059d49b0ca0eec@syzkaller.appspotmail.com Reported-by: syzbot+847f02ec20a6609a328b@syzkaller.appspotmail.com Reported-by: syzbot+3368aadcd30425ceb53b@syzkaller.appspotmail.com Reported-by: syzbot+51ce8887cdef77c9ac83@syzkaller.appspotmail.com Reported-by: syzbot+3cb756a49d2f394a9ee3@syzkaller.appspotmail.com Fixes: 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") Cc: stable@kernel.org # 5.15+ Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/397f7ebf3f4171f1abe41f708ac1ecb5766f0b68.1637937097.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-11-26 22:38:15 +08:00
return io_match_task_safe(req, cancel->task, cancel->all);
}
static __cold bool io_cancel_defer_files(struct io_ring_ctx *ctx,
struct task_struct *task,
bool cancel_all)
{
struct io_defer_entry *de;
LIST_HEAD(list);
spin_lock(&ctx->completion_lock);
list_for_each_entry_reverse(de, &ctx->defer_list, list) {
io_uring: fix link traversal locking WARNING: inconsistent lock state 5.16.0-rc2-syzkaller #0 Not tainted inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage. ffff888078e11418 (&ctx->timeout_lock ){?.+.}-{2:2} , at: io_timeout_fn+0x6f/0x360 fs/io_uring.c:5943 {HARDIRQ-ON-W} state was registered at: [...] spin_unlock_irq include/linux/spinlock.h:399 [inline] __io_poll_remove_one fs/io_uring.c:5669 [inline] __io_poll_remove_one fs/io_uring.c:5654 [inline] io_poll_remove_one+0x236/0x870 fs/io_uring.c:5680 io_poll_remove_all+0x1af/0x235 fs/io_uring.c:5709 io_ring_ctx_wait_and_kill+0x1cc/0x322 fs/io_uring.c:9534 io_uring_release+0x42/0x46 fs/io_uring.c:9554 __fput+0x286/0x9f0 fs/file_table.c:280 task_work_run+0xdd/0x1a0 kernel/task_work.c:164 exit_task_work include/linux/task_work.h:32 [inline] do_exit+0xc14/0x2b40 kernel/exit.c:832 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") fixed a data race but introduced a possible deadlock and inconsistentcy in irq states. E.g. io_poll_remove_all() spin_lock_irq(timeout_lock) io_poll_remove_one() spin_lock/unlock_irq(poll_lock); spin_unlock_irq(timeout_lock) Another type of problem is freeing a request while holding ->timeout_lock, which may leads to a deadlock in io_commit_cqring() -> io_flush_timeouts() and other places. Having 3 nested locks is also too ugly. Add io_match_task_safe(), which would briefly take and release timeout_lock for race prevention inside, so the actuall request cancellation / free / etc. code doesn't have it taken. Reported-by: syzbot+ff49a3059d49b0ca0eec@syzkaller.appspotmail.com Reported-by: syzbot+847f02ec20a6609a328b@syzkaller.appspotmail.com Reported-by: syzbot+3368aadcd30425ceb53b@syzkaller.appspotmail.com Reported-by: syzbot+51ce8887cdef77c9ac83@syzkaller.appspotmail.com Reported-by: syzbot+3cb756a49d2f394a9ee3@syzkaller.appspotmail.com Fixes: 674ee8e1b4a41 ("io_uring: correct link-list traversal locking") Cc: stable@kernel.org # 5.15+ Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/397f7ebf3f4171f1abe41f708ac1ecb5766f0b68.1637937097.git.asml.silence@gmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-11-26 22:38:15 +08:00
if (io_match_task_safe(de->req, task, cancel_all)) {
list_cut_position(&list, &ctx->defer_list, &de->list);
break;
}
}
spin_unlock(&ctx->completion_lock);
if (list_empty(&list))
return false;
while (!list_empty(&list)) {
de = list_first_entry(&list, struct io_defer_entry, list);
list_del_init(&de->list);
io_req_task_queue_fail(de->req, -ECANCELED);
kfree(de);
}
return true;
}
static __cold bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx)
{
struct io_tctx_node *node;
enum io_wq_cancel cret;
bool ret = false;
mutex_lock(&ctx->uring_lock);
list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
struct io_uring_task *tctx = node->task->io_uring;
/*
* io_wq will stay alive while we hold uring_lock, because it's
* killed after ctx nodes, which requires to take the lock.
*/
if (!tctx || !tctx->io_wq)
continue;
cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true);
ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
}
mutex_unlock(&ctx->uring_lock);
return ret;
}
static __cold bool io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
struct task_struct *task,
bool cancel_all)
{
struct io_task_cancel cancel = { .task = task, .all = cancel_all, };
struct io_uring_task *tctx = task ? task->io_uring : NULL;
enum io_wq_cancel cret;
bool ret = false;
/* set it so io_req_local_work_add() would wake us up */
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) {
atomic_set(&ctx->cq_wait_nr, 1);
smp_mb();
}
/* failed during ring init, it couldn't have issued any requests */
if (!ctx->rings)
return false;
if (!task) {
ret |= io_uring_try_cancel_iowq(ctx);
} else if (tctx && tctx->io_wq) {
/*
* Cancels requests of all rings, not only @ctx, but
* it's fine as the task is in exit/exec.
*/
cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb,
&cancel, true);
ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
}
/* SQPOLL thread does its own polling */
if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) ||
(ctx->sq_data && ctx->sq_data->thread == current)) {
while (!wq_list_empty(&ctx->iopoll_list)) {
io_iopoll_try_reap_events(ctx);
ret = true;
io_uring: add a conditional reschedule to the IOPOLL cancelation loop If the kernel is configured with CONFIG_PREEMPT_NONE, we could be sitting in a tight loop reaping events but not giving them a chance to finish. This results in a trace ala: rcu: INFO: rcu_sched self-detected stall on CPU rcu: 2-...!: (5249 ticks this GP) idle=935c/1/0x4000000000000000 softirq=4265/4274 fqs=1 (t=5251 jiffies g=465 q=4135 ncpus=4) rcu: rcu_sched kthread starved for 5249 jiffies! g465 f0x0 RCU_GP_WAIT_FQS(5) ->state=0x0 ->cpu=0 rcu: Unless rcu_sched kthread gets sufficient CPU time, OOM is now expected behavior. rcu: RCU grace-period kthread stack dump: task:rcu_sched state:R running task stack:0 pid:12 ppid:2 flags:0x00000008 Call trace: __switch_to+0xb0/0xc8 __schedule+0x43c/0x520 schedule+0x4c/0x98 schedule_timeout+0xbc/0xdc rcu_gp_fqs_loop+0x308/0x344 rcu_gp_kthread+0xd8/0xf0 kthread+0xb8/0xc8 ret_from_fork+0x10/0x20 rcu: Stack dump where RCU GP kthread last ran: Task dump for CPU 0: task:kworker/u8:10 state:R running task stack:0 pid:89 ppid:2 flags:0x0000000a Workqueue: events_unbound io_ring_exit_work Call trace: __switch_to+0xb0/0xc8 0xffff0000c8fefd28 CPU: 2 PID: 95 Comm: kworker/u8:13 Not tainted 6.2.0-rc5-00042-g40316e337c80-dirty #2759 Hardware name: linux,dummy-virt (DT) Workqueue: events_unbound io_ring_exit_work pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--) pc : io_do_iopoll+0x344/0x360 lr : io_do_iopoll+0xb8/0x360 sp : ffff800009bebc60 x29: ffff800009bebc60 x28: 0000000000000000 x27: 0000000000000000 x26: ffff0000c0f67d48 x25: ffff0000c0f67840 x24: ffff800008950024 x23: 0000000000000001 x22: 0000000000000000 x21: ffff0000c27d3200 x20: ffff0000c0f67840 x19: ffff0000c0f67800 x18: 0000000000000000 x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000 x14: 0000000000000001 x13: 0000000000000001 x12: 0000000000000000 x11: 0000000000000179 x10: 0000000000000870 x9 : ffff800009bebd60 x8 : ffff0000c27d3ad0 x7 : fefefefefefefeff x6 : 0000646e756f626e x5 : ffff0000c0f67840 x4 : 0000000000000000 x3 : ffff0000c2398000 x2 : 0000000000000000 x1 : 0000000000000000 x0 : 0000000000000000 Call trace: io_do_iopoll+0x344/0x360 io_uring_try_cancel_requests+0x21c/0x334 io_ring_exit_work+0x90/0x40c process_one_work+0x1a4/0x254 worker_thread+0x1ec/0x258 kthread+0xb8/0xc8 ret_from_fork+0x10/0x20 Add a cond_resched() in the cancelation IOPOLL loop to fix this. Cc: stable@vger.kernel.org # 5.10+ Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-28 00:28:13 +08:00
cond_resched();
}
}
if ((ctx->flags & IORING_SETUP_DEFER_TASKRUN) &&
io_allowed_defer_tw_run(ctx))
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
ret |= io_run_local_work(ctx) > 0;
ret |= io_cancel_defer_files(ctx, task, cancel_all);
mutex_lock(&ctx->uring_lock);
ret |= io_poll_remove_all(ctx, task, cancel_all);
mutex_unlock(&ctx->uring_lock);
ret |= io_kill_timeouts(ctx, task, cancel_all);
if (task)
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
ret |= io_run_task_work() > 0;
return ret;
}
static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked)
{
if (tracked)
return atomic_read(&tctx->inflight_tracked);
return percpu_counter_sum(&tctx->inflight);
}
/*
* Find any io_uring ctx that this task has registered or done IO on, and cancel
* requests. @sqd should be not-null IFF it's an SQPOLL thread cancellation.
*/
__cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd)
{
struct io_uring_task *tctx = current->io_uring;
struct io_ring_ctx *ctx;
struct io_tctx_node *node;
unsigned long index;
s64 inflight;
DEFINE_WAIT(wait);
WARN_ON_ONCE(sqd && sqd->thread != current);
if (!current->io_uring)
return;
if (tctx->io_wq)
io_wq_exit_start(tctx->io_wq);
atomic_inc(&tctx->in_cancel);
do {
bool loop = false;
io_uring_drop_tctx_refs(current);
/* read completions before cancelations */
inflight = tctx_inflight(tctx, !cancel_all);
if (!inflight)
break;
if (!sqd) {
xa_for_each(&tctx->xa, index, node) {
/* sqpoll task will cancel all its requests */
if (node->ctx->sq_data)
continue;
loop |= io_uring_try_cancel_requests(node->ctx,
current, cancel_all);
}
} else {
list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
loop |= io_uring_try_cancel_requests(ctx,
current,
cancel_all);
}
if (loop) {
cond_resched();
continue;
}
prepare_to_wait(&tctx->wait, &wait, TASK_INTERRUPTIBLE);
io_run_task_work();
io_uring_drop_tctx_refs(current);
xa_for_each(&tctx->xa, index, node) {
if (!llist_empty(&node->ctx->work_llist)) {
WARN_ON_ONCE(node->ctx->submitter_task &&
node->ctx->submitter_task != current);
goto end_wait;
}
}
/*
* If we've seen completions, retry without waiting. This
* avoids a race where a completion comes in before we did
* prepare_to_wait().
*/
if (inflight == tctx_inflight(tctx, !cancel_all))
schedule();
end_wait:
finish_wait(&tctx->wait, &wait);
} while (1);
io_uring_clean_tctx(tctx);
if (cancel_all) {
/*
* We shouldn't run task_works after cancel, so just leave
* ->in_cancel set for normal exit.
*/
atomic_dec(&tctx->in_cancel);
/* for exec all current's requests should be gone, kill tctx */
__io_uring_free(current);
}
}
void __io_uring_cancel(bool cancel_all)
{
io_uring_cancel_generic(cancel_all, NULL);
}
static void *io_uring_validate_mmap_request(struct file *file,
loff_t pgoff, size_t sz)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_ring_ctx *ctx = file->private_data;
loff_t offset = pgoff << PAGE_SHIFT;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
struct page *page;
void *ptr;
switch (offset & IORING_OFF_MMAP_MASK) {
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
case IORING_OFF_SQ_RING:
case IORING_OFF_CQ_RING:
/* Don't allow mmap if the ring was setup without it */
if (ctx->flags & IORING_SETUP_NO_MMAP)
return ERR_PTR(-EINVAL);
ptr = ctx->rings;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
break;
case IORING_OFF_SQES:
/* Don't allow mmap if the ring was setup without it */
if (ctx->flags & IORING_SETUP_NO_MMAP)
return ERR_PTR(-EINVAL);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ptr = ctx->sq_sqes;
break;
case IORING_OFF_PBUF_RING: {
unsigned int bgid;
bgid = (offset & ~IORING_OFF_MMAP_MASK) >> IORING_OFF_PBUF_SHIFT;
io_uring: free io_buffer_list entries via RCU commit 5cf4f52e6d8aa2d3b7728f568abbf9d42a3af252 upstream. mmap_lock nests under uring_lock out of necessity, as we may be doing user copies with uring_lock held. However, for mmap of provided buffer rings, we attempt to grab uring_lock with mmap_lock already held from do_mmap(). This makes lockdep, rightfully, complain: WARNING: possible circular locking dependency detected 6.7.0-rc1-00009-gff3337ebaf94-dirty #4438 Not tainted ------------------------------------------------------ buf-ring.t/442 is trying to acquire lock: ffff00020e1480a8 (&ctx->uring_lock){+.+.}-{3:3}, at: io_uring_validate_mmap_request.isra.0+0x4c/0x140 but task is already holding lock: ffff0000dc226190 (&mm->mmap_lock){++++}-{3:3}, at: vm_mmap_pgoff+0x124/0x264 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&mm->mmap_lock){++++}-{3:3}: __might_fault+0x90/0xbc io_register_pbuf_ring+0x94/0x488 __arm64_sys_io_uring_register+0x8dc/0x1318 invoke_syscall+0x5c/0x17c el0_svc_common.constprop.0+0x108/0x130 do_el0_svc+0x2c/0x38 el0_svc+0x4c/0x94 el0t_64_sync_handler+0x118/0x124 el0t_64_sync+0x168/0x16c -> #0 (&ctx->uring_lock){+.+.}-{3:3}: __lock_acquire+0x19a0/0x2d14 lock_acquire+0x2e0/0x44c __mutex_lock+0x118/0x564 mutex_lock_nested+0x20/0x28 io_uring_validate_mmap_request.isra.0+0x4c/0x140 io_uring_mmu_get_unmapped_area+0x3c/0x98 get_unmapped_area+0xa4/0x158 do_mmap+0xec/0x5b4 vm_mmap_pgoff+0x158/0x264 ksys_mmap_pgoff+0x1d4/0x254 __arm64_sys_mmap+0x80/0x9c invoke_syscall+0x5c/0x17c el0_svc_common.constprop.0+0x108/0x130 do_el0_svc+0x2c/0x38 el0_svc+0x4c/0x94 el0t_64_sync_handler+0x118/0x124 el0t_64_sync+0x168/0x16c From that mmap(2) path, we really just need to ensure that the buffer list doesn't go away from underneath us. For the lower indexed entries, they never go away until the ring is freed and we can always sanely reference those as long as the caller has a file reference. For the higher indexed ones in our xarray, we just need to ensure that the buffer list remains valid while we return the address of it. Free the higher indexed io_buffer_list entries via RCU. With that we can avoid needing ->uring_lock inside mmap(2), and simply hold the RCU read lock around the buffer list lookup and address check. To ensure that the arrayed lookup either returns a valid fully formulated entry via RCU lookup, add an 'is_ready' flag that we access with store and release memory ordering. This isn't needed for the xarray lookups, but doesn't hurt either. Since this isn't a fast path, retain it across both types. Similarly, for the allocated array inside the ctx, ensure we use the proper load/acquire as setup could in theory be running in parallel with mmap. While in there, add a few lockdep checks for documentation purposes. Cc: stable@vger.kernel.org Fixes: c56e022c0a27 ("io_uring: add support for user mapped provided buffer ring") Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2023-11-28 08:54:40 +08:00
rcu_read_lock();
ptr = io_pbuf_get_address(ctx, bgid);
io_uring: free io_buffer_list entries via RCU commit 5cf4f52e6d8aa2d3b7728f568abbf9d42a3af252 upstream. mmap_lock nests under uring_lock out of necessity, as we may be doing user copies with uring_lock held. However, for mmap of provided buffer rings, we attempt to grab uring_lock with mmap_lock already held from do_mmap(). This makes lockdep, rightfully, complain: WARNING: possible circular locking dependency detected 6.7.0-rc1-00009-gff3337ebaf94-dirty #4438 Not tainted ------------------------------------------------------ buf-ring.t/442 is trying to acquire lock: ffff00020e1480a8 (&ctx->uring_lock){+.+.}-{3:3}, at: io_uring_validate_mmap_request.isra.0+0x4c/0x140 but task is already holding lock: ffff0000dc226190 (&mm->mmap_lock){++++}-{3:3}, at: vm_mmap_pgoff+0x124/0x264 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&mm->mmap_lock){++++}-{3:3}: __might_fault+0x90/0xbc io_register_pbuf_ring+0x94/0x488 __arm64_sys_io_uring_register+0x8dc/0x1318 invoke_syscall+0x5c/0x17c el0_svc_common.constprop.0+0x108/0x130 do_el0_svc+0x2c/0x38 el0_svc+0x4c/0x94 el0t_64_sync_handler+0x118/0x124 el0t_64_sync+0x168/0x16c -> #0 (&ctx->uring_lock){+.+.}-{3:3}: __lock_acquire+0x19a0/0x2d14 lock_acquire+0x2e0/0x44c __mutex_lock+0x118/0x564 mutex_lock_nested+0x20/0x28 io_uring_validate_mmap_request.isra.0+0x4c/0x140 io_uring_mmu_get_unmapped_area+0x3c/0x98 get_unmapped_area+0xa4/0x158 do_mmap+0xec/0x5b4 vm_mmap_pgoff+0x158/0x264 ksys_mmap_pgoff+0x1d4/0x254 __arm64_sys_mmap+0x80/0x9c invoke_syscall+0x5c/0x17c el0_svc_common.constprop.0+0x108/0x130 do_el0_svc+0x2c/0x38 el0_svc+0x4c/0x94 el0t_64_sync_handler+0x118/0x124 el0t_64_sync+0x168/0x16c From that mmap(2) path, we really just need to ensure that the buffer list doesn't go away from underneath us. For the lower indexed entries, they never go away until the ring is freed and we can always sanely reference those as long as the caller has a file reference. For the higher indexed ones in our xarray, we just need to ensure that the buffer list remains valid while we return the address of it. Free the higher indexed io_buffer_list entries via RCU. With that we can avoid needing ->uring_lock inside mmap(2), and simply hold the RCU read lock around the buffer list lookup and address check. To ensure that the arrayed lookup either returns a valid fully formulated entry via RCU lookup, add an 'is_ready' flag that we access with store and release memory ordering. This isn't needed for the xarray lookups, but doesn't hurt either. Since this isn't a fast path, retain it across both types. Similarly, for the allocated array inside the ctx, ensure we use the proper load/acquire as setup could in theory be running in parallel with mmap. While in there, add a few lockdep checks for documentation purposes. Cc: stable@vger.kernel.org Fixes: c56e022c0a27 ("io_uring: add support for user mapped provided buffer ring") Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2023-11-28 08:54:40 +08:00
rcu_read_unlock();
if (!ptr)
return ERR_PTR(-EINVAL);
break;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
default:
return ERR_PTR(-EINVAL);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
page = virt_to_head_page(ptr);
if (sz > page_size(page))
return ERR_PTR(-EINVAL);
return ptr;
}
#ifdef CONFIG_MMU
static __cold int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
{
size_t sz = vma->vm_end - vma->vm_start;
unsigned long pfn;
void *ptr;
ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
if (IS_ERR(ptr))
return PTR_ERR(ptr);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
}
static unsigned long io_uring_mmu_get_unmapped_area(struct file *filp,
unsigned long addr, unsigned long len,
unsigned long pgoff, unsigned long flags)
{
void *ptr;
/*
* Do not allow to map to user-provided address to avoid breaking the
* aliasing rules. Userspace is not able to guess the offset address of
* kernel kmalloc()ed memory area.
*/
if (addr)
return -EINVAL;
ptr = io_uring_validate_mmap_request(filp, pgoff, len);
if (IS_ERR(ptr))
return -ENOMEM;
/*
* Some architectures have strong cache aliasing requirements.
* For such architectures we need a coherent mapping which aliases
* kernel memory *and* userspace memory. To achieve that:
* - use a NULL file pointer to reference physical memory, and
* - use the kernel virtual address of the shared io_uring context
* (instead of the userspace-provided address, which has to be 0UL
* anyway).
* - use the same pgoff which the get_unmapped_area() uses to
* calculate the page colouring.
* For architectures without such aliasing requirements, the
* architecture will return any suitable mapping because addr is 0.
*/
filp = NULL;
flags |= MAP_SHARED;
pgoff = 0; /* has been translated to ptr above */
#ifdef SHM_COLOUR
addr = (uintptr_t) ptr;
pgoff = addr >> PAGE_SHIFT;
#else
addr = 0UL;
#endif
return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
}
#else /* !CONFIG_MMU */
static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
{
mm/nommu: factor out check for NOMMU shared mappings into is_nommu_shared_mapping() Patch series "mm/nommu: don't use VM_MAYSHARE for MAP_PRIVATE mappings". Trying to reduce the confusion around VM_SHARED and VM_MAYSHARE first requires !CONFIG_MMU to stop using VM_MAYSHARE for MAP_PRIVATE mappings. CONFIG_MMU only sets VM_MAYSHARE for MAP_SHARED mappings. This paves the way for further VM_MAYSHARE and VM_SHARED cleanups: for example, renaming VM_MAYSHARED to VM_MAP_SHARED to make it cleaner what is actually means. Let's first get the weird case out of the way and not use VM_MAYSHARE in MAP_PRIVATE mappings, using a new VM_MAYOVERLAY flag instead. This patch (of 3): We want to stop using VM_MAYSHARE in private mappings to pave the way for clarifying the semantics of VM_MAYSHARE vs. VM_SHARED and reduce the confusion. While CONFIG_MMU uses VM_MAYSHARE to represent MAP_SHARED, !CONFIG_MMU also sets VM_MAYSHARE for selected R/O private file mappings that are an effective overlay of a file mapping. Let's factor out all relevant VM_MAYSHARE checks in !CONFIG_MMU code into is_nommu_shared_mapping() first. Note that whenever VM_SHARED is set, VM_MAYSHARE must be set as well (unless there is a serious BUG). So there is not need to test for VM_SHARED manually. No functional change intended. Link: https://lkml.kernel.org/r/20230102160856.500584-1-david@redhat.com Link: https://lkml.kernel.org/r/20230102160856.500584-2-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Nicolas Pitre <nico@fluxnic.net> Cc: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-03 00:08:54 +08:00
return is_nommu_shared_mapping(vma->vm_flags) ? 0 : -EINVAL;
}
static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
{
return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
}
static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
unsigned long addr, unsigned long len,
unsigned long pgoff, unsigned long flags)
{
void *ptr;
ptr = io_uring_validate_mmap_request(file, pgoff, len);
if (IS_ERR(ptr))
return PTR_ERR(ptr);
return (unsigned long) ptr;
}
#endif /* !CONFIG_MMU */
static int io_validate_ext_arg(unsigned flags, const void __user *argp, size_t argsz)
{
if (flags & IORING_ENTER_EXT_ARG) {
struct io_uring_getevents_arg arg;
if (argsz != sizeof(arg))
return -EINVAL;
if (copy_from_user(&arg, argp, sizeof(arg)))
return -EFAULT;
}
return 0;
}
static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
struct __kernel_timespec __user **ts,
const sigset_t __user **sig)
{
struct io_uring_getevents_arg arg;
/*
* If EXT_ARG isn't set, then we have no timespec and the argp pointer
* is just a pointer to the sigset_t.
*/
if (!(flags & IORING_ENTER_EXT_ARG)) {
*sig = (const sigset_t __user *) argp;
*ts = NULL;
return 0;
}
/*
* EXT_ARG is set - ensure we agree on the size of it and copy in our
* timespec and sigset_t pointers if good.
*/
if (*argsz != sizeof(arg))
return -EINVAL;
if (copy_from_user(&arg, argp, sizeof(arg)))
return -EFAULT;
if (arg.pad)
return -EINVAL;
*sig = u64_to_user_ptr(arg.sigmask);
*argsz = arg.sigmask_sz;
*ts = u64_to_user_ptr(arg.ts);
return 0;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
u32, min_complete, u32, flags, const void __user *, argp,
size_t, argsz)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_ring_ctx *ctx;
struct file *file;
long ret;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
io_uring: add support for registering ring file descriptors Lots of workloads use multiple threads, in which case the file table is shared between them. This makes getting and putting the ring file descriptor for each io_uring_enter(2) system call more expensive, as it involves an atomic get and put for each call. Similarly to how we allow registering normal file descriptors to avoid this overhead, add support for an io_uring_register(2) API that allows to register the ring fds themselves: 1) IORING_REGISTER_RING_FDS - takes an array of io_uring_rsrc_update structs, and registers them with the task. 2) IORING_UNREGISTER_RING_FDS - takes an array of io_uring_src_update structs, and unregisters them. When a ring fd is registered, it is internally represented by an offset. This offset is returned to the application, and the application then uses this offset and sets IORING_ENTER_REGISTERED_RING for the io_uring_enter(2) system call. This works just like using a registered file descriptor, rather than a real one, in an SQE, where IOSQE_FIXED_FILE gets set to tell io_uring that we're using an internal offset/descriptor rather than a real file descriptor. In initial testing, this provides a nice bump in performance for threaded applications in real world cases where the batch count (eg number of requests submitted per io_uring_enter(2) invocation) is low. In a microbenchmark, submitting NOP requests, we see the following increases in performance: Requests per syscall Baseline Registered Increase ---------------------------------------------------------------- 1 ~7030K ~8080K +15% 2 ~13120K ~14800K +13% 4 ~22740K ~25300K +11% Co-developed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-03-04 23:22:22 +08:00
IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG |
IORING_ENTER_REGISTERED_RING)))
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return -EINVAL;
io_uring: add support for registering ring file descriptors Lots of workloads use multiple threads, in which case the file table is shared between them. This makes getting and putting the ring file descriptor for each io_uring_enter(2) system call more expensive, as it involves an atomic get and put for each call. Similarly to how we allow registering normal file descriptors to avoid this overhead, add support for an io_uring_register(2) API that allows to register the ring fds themselves: 1) IORING_REGISTER_RING_FDS - takes an array of io_uring_rsrc_update structs, and registers them with the task. 2) IORING_UNREGISTER_RING_FDS - takes an array of io_uring_src_update structs, and unregisters them. When a ring fd is registered, it is internally represented by an offset. This offset is returned to the application, and the application then uses this offset and sets IORING_ENTER_REGISTERED_RING for the io_uring_enter(2) system call. This works just like using a registered file descriptor, rather than a real one, in an SQE, where IOSQE_FIXED_FILE gets set to tell io_uring that we're using an internal offset/descriptor rather than a real file descriptor. In initial testing, this provides a nice bump in performance for threaded applications in real world cases where the batch count (eg number of requests submitted per io_uring_enter(2) invocation) is low. In a microbenchmark, submitting NOP requests, we see the following increases in performance: Requests per syscall Baseline Registered Increase ---------------------------------------------------------------- 1 ~7030K ~8080K +15% 2 ~13120K ~14800K +13% 4 ~22740K ~25300K +11% Co-developed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-03-04 23:22:22 +08:00
/*
* Ring fd has been registered via IORING_REGISTER_RING_FDS, we
* need only dereference our task private array to find it.
*/
if (flags & IORING_ENTER_REGISTERED_RING) {
struct io_uring_task *tctx = current->io_uring;
if (unlikely(!tctx || fd >= IO_RINGFD_REG_MAX))
io_uring: add support for registering ring file descriptors Lots of workloads use multiple threads, in which case the file table is shared between them. This makes getting and putting the ring file descriptor for each io_uring_enter(2) system call more expensive, as it involves an atomic get and put for each call. Similarly to how we allow registering normal file descriptors to avoid this overhead, add support for an io_uring_register(2) API that allows to register the ring fds themselves: 1) IORING_REGISTER_RING_FDS - takes an array of io_uring_rsrc_update structs, and registers them with the task. 2) IORING_UNREGISTER_RING_FDS - takes an array of io_uring_src_update structs, and unregisters them. When a ring fd is registered, it is internally represented by an offset. This offset is returned to the application, and the application then uses this offset and sets IORING_ENTER_REGISTERED_RING for the io_uring_enter(2) system call. This works just like using a registered file descriptor, rather than a real one, in an SQE, where IOSQE_FIXED_FILE gets set to tell io_uring that we're using an internal offset/descriptor rather than a real file descriptor. In initial testing, this provides a nice bump in performance for threaded applications in real world cases where the batch count (eg number of requests submitted per io_uring_enter(2) invocation) is low. In a microbenchmark, submitting NOP requests, we see the following increases in performance: Requests per syscall Baseline Registered Increase ---------------------------------------------------------------- 1 ~7030K ~8080K +15% 2 ~13120K ~14800K +13% 4 ~22740K ~25300K +11% Co-developed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-03-04 23:22:22 +08:00
return -EINVAL;
fd = array_index_nospec(fd, IO_RINGFD_REG_MAX);
file = tctx->registered_rings[fd];
if (unlikely(!file))
return -EBADF;
io_uring: add support for registering ring file descriptors Lots of workloads use multiple threads, in which case the file table is shared between them. This makes getting and putting the ring file descriptor for each io_uring_enter(2) system call more expensive, as it involves an atomic get and put for each call. Similarly to how we allow registering normal file descriptors to avoid this overhead, add support for an io_uring_register(2) API that allows to register the ring fds themselves: 1) IORING_REGISTER_RING_FDS - takes an array of io_uring_rsrc_update structs, and registers them with the task. 2) IORING_UNREGISTER_RING_FDS - takes an array of io_uring_src_update structs, and unregisters them. When a ring fd is registered, it is internally represented by an offset. This offset is returned to the application, and the application then uses this offset and sets IORING_ENTER_REGISTERED_RING for the io_uring_enter(2) system call. This works just like using a registered file descriptor, rather than a real one, in an SQE, where IOSQE_FIXED_FILE gets set to tell io_uring that we're using an internal offset/descriptor rather than a real file descriptor. In initial testing, this provides a nice bump in performance for threaded applications in real world cases where the batch count (eg number of requests submitted per io_uring_enter(2) invocation) is low. In a microbenchmark, submitting NOP requests, we see the following increases in performance: Requests per syscall Baseline Registered Increase ---------------------------------------------------------------- 1 ~7030K ~8080K +15% 2 ~13120K ~14800K +13% 4 ~22740K ~25300K +11% Co-developed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-03-04 23:22:22 +08:00
} else {
file = fget(fd);
if (unlikely(!file))
return -EBADF;
ret = -EOPNOTSUPP;
if (unlikely(!io_is_uring_fops(file)))
goto out;
io_uring: add support for registering ring file descriptors Lots of workloads use multiple threads, in which case the file table is shared between them. This makes getting and putting the ring file descriptor for each io_uring_enter(2) system call more expensive, as it involves an atomic get and put for each call. Similarly to how we allow registering normal file descriptors to avoid this overhead, add support for an io_uring_register(2) API that allows to register the ring fds themselves: 1) IORING_REGISTER_RING_FDS - takes an array of io_uring_rsrc_update structs, and registers them with the task. 2) IORING_UNREGISTER_RING_FDS - takes an array of io_uring_src_update structs, and unregisters them. When a ring fd is registered, it is internally represented by an offset. This offset is returned to the application, and the application then uses this offset and sets IORING_ENTER_REGISTERED_RING for the io_uring_enter(2) system call. This works just like using a registered file descriptor, rather than a real one, in an SQE, where IOSQE_FIXED_FILE gets set to tell io_uring that we're using an internal offset/descriptor rather than a real file descriptor. In initial testing, this provides a nice bump in performance for threaded applications in real world cases where the batch count (eg number of requests submitted per io_uring_enter(2) invocation) is low. In a microbenchmark, submitting NOP requests, we see the following increases in performance: Requests per syscall Baseline Registered Increase ---------------------------------------------------------------- 1 ~7030K ~8080K +15% 2 ~13120K ~14800K +13% 4 ~22740K ~25300K +11% Co-developed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-03-04 23:22:22 +08:00
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ctx = file->private_data;
ret = -EBADFD;
if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED))
goto out;
/*
* For SQ polling, the thread will do all submissions and completions.
* Just return the requested submit count, and wake the thread if
* we were asked to.
*/
ret = 0;
if (ctx->flags & IORING_SETUP_SQPOLL) {
io_cqring_overflow_flush(ctx);
if (unlikely(ctx->sq_data->thread == NULL)) {
ret = -EOWNERDEAD;
goto out;
}
if (flags & IORING_ENTER_SQ_WAKEUP)
wake_up(&ctx->sq_data->wait);
if (flags & IORING_ENTER_SQ_WAIT)
io_sqpoll_wait_sq(ctx);
ret = to_submit;
} else if (to_submit) {
ret = io_uring_add_tctx_node(ctx);
if (unlikely(ret))
goto out;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
mutex_lock(&ctx->uring_lock);
ret = io_submit_sqes(ctx, to_submit);
if (ret != to_submit) {
mutex_unlock(&ctx->uring_lock);
goto out;
}
if (flags & IORING_ENTER_GETEVENTS) {
if (ctx->syscall_iopoll)
goto iopoll_locked;
/*
* Ignore errors, we'll soon call io_cqring_wait() and
* it should handle ownership problems if any.
*/
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN)
(void)io_run_local_work_locked(ctx);
}
mutex_unlock(&ctx->uring_lock);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (flags & IORING_ENTER_GETEVENTS) {
int ret2;
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
if (ctx->syscall_iopoll) {
/*
* We disallow the app entering submit/complete with
* polling, but we still need to lock the ring to
* prevent racing with polled issue that got punted to
* a workqueue.
*/
mutex_lock(&ctx->uring_lock);
iopoll_locked:
ret2 = io_validate_ext_arg(flags, argp, argsz);
if (likely(!ret2)) {
min_complete = min(min_complete,
ctx->cq_entries);
ret2 = io_iopoll_check(ctx, min_complete);
}
mutex_unlock(&ctx->uring_lock);
} else {
const sigset_t __user *sig;
struct __kernel_timespec __user *ts;
ret2 = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
if (likely(!ret2)) {
min_complete = min(min_complete,
ctx->cq_entries);
ret2 = io_cqring_wait(ctx, min_complete, sig,
argsz, ts);
}
}
if (!ret) {
ret = ret2;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
/*
* EBADR indicates that one or more CQE were dropped.
* Once the user has been informed we can clear the bit
* as they are obviously ok with those drops.
*/
if (unlikely(ret2 == -EBADR))
clear_bit(IO_CHECK_CQ_DROPPED_BIT,
&ctx->check_cq);
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
out:
if (!(flags & IORING_ENTER_REGISTERED_RING))
fput(file);
return ret;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
static const struct file_operations io_uring_fops = {
.release = io_uring_release,
.mmap = io_uring_mmap,
#ifndef CONFIG_MMU
.get_unmapped_area = io_uring_nommu_get_unmapped_area,
.mmap_capabilities = io_uring_nommu_mmap_capabilities,
#else
.get_unmapped_area = io_uring_mmu_get_unmapped_area,
#endif
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
.poll = io_uring_poll,
#ifdef CONFIG_PROC_FS
.show_fdinfo = io_uring_show_fdinfo,
#endif
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
};
bool io_is_uring_fops(struct file *file)
{
return file->f_op == &io_uring_fops;
}
static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx,
struct io_uring_params *p)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_rings *rings;
size_t size, sq_array_offset;
void *ptr;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
/* make sure these are sane, as we already accounted them */
ctx->sq_entries = p->sq_entries;
ctx->cq_entries = p->cq_entries;
size = rings_size(ctx, p->sq_entries, p->cq_entries, &sq_array_offset);
if (size == SIZE_MAX)
return -EOVERFLOW;
if (!(ctx->flags & IORING_SETUP_NO_MMAP))
rings = io_mem_alloc(size);
else
rings = io_rings_map(ctx, p->cq_off.user_addr, size);
if (IS_ERR(rings))
return PTR_ERR(rings);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ctx->rings = rings;
if (!(ctx->flags & IORING_SETUP_NO_SQARRAY))
ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
rings->sq_ring_mask = p->sq_entries - 1;
rings->cq_ring_mask = p->cq_entries - 1;
rings->sq_ring_entries = p->sq_entries;
rings->cq_ring_entries = p->cq_entries;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (p->flags & IORING_SETUP_SQE128)
size = array_size(2 * sizeof(struct io_uring_sqe), p->sq_entries);
else
size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
if (size == SIZE_MAX) {
io_rings_free(ctx);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return -EOVERFLOW;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (!(ctx->flags & IORING_SETUP_NO_MMAP))
ptr = io_mem_alloc(size);
else
ptr = io_sqes_map(ctx, p->sq_off.user_addr, size);
if (IS_ERR(ptr)) {
io_rings_free(ctx);
return PTR_ERR(ptr);
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ctx->sq_sqes = ptr;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return 0;
}
static int io_uring_install_fd(struct file *file)
{
int fd;
fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
if (fd < 0)
return fd;
fd_install(fd, file);
return fd;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
/*
* Allocate an anonymous fd, this is what constitutes the application
* visible backing of an io_uring instance. The application mmaps this
* fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
* we have to tie this fd to a socket for file garbage collection purposes.
*/
static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct file *file;
#if defined(CONFIG_UNIX)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
int ret;
ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
&ctx->ring_sock);
if (ret)
return ERR_PTR(ret);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#endif
file = anon_inode_getfile_secure("[io_uring]", &io_uring_fops, ctx,
O_RDWR | O_CLOEXEC, NULL);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#if defined(CONFIG_UNIX)
if (IS_ERR(file)) {
sock_release(ctx->ring_sock);
ctx->ring_sock = NULL;
} else {
ctx->ring_sock->file = file;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
#endif
return file;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,
struct io_uring_params __user *params)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
{
struct io_ring_ctx *ctx;
struct io_uring_task *tctx;
struct file *file;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
int ret;
if (!entries)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return -EINVAL;
if (entries > IORING_MAX_ENTRIES) {
if (!(p->flags & IORING_SETUP_CLAMP))
return -EINVAL;
entries = IORING_MAX_ENTRIES;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if ((p->flags & IORING_SETUP_REGISTERED_FD_ONLY)
&& !(p->flags & IORING_SETUP_NO_MMAP))
return -EINVAL;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
/*
* Use twice as many entries for the CQ ring. It's possible for the
* application to drive a higher depth than the size of the SQ ring,
* since the sqes are only used at submission time. This allows for
* some flexibility in overcommitting a bit. If the application has
* set IORING_SETUP_CQSIZE, it will have passed in the desired number
* of CQ ring entries manually.
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
*/
p->sq_entries = roundup_pow_of_two(entries);
if (p->flags & IORING_SETUP_CQSIZE) {
/*
* If IORING_SETUP_CQSIZE is set, we do the same roundup
* to a power-of-two, if it isn't already. We do NOT impose
* any cq vs sq ring sizing.
*/
io_uring: fix shift-out-of-bounds when round up cq size Abaci Fuzz reported a shift-out-of-bounds BUG in io_uring_create(): [ 59.598207] UBSAN: shift-out-of-bounds in ./include/linux/log2.h:57:13 [ 59.599665] shift exponent 64 is too large for 64-bit type 'long unsigned int' [ 59.601230] CPU: 0 PID: 963 Comm: a.out Not tainted 5.10.0-rc4+ #3 [ 59.602502] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 59.603673] Call Trace: [ 59.604286] dump_stack+0x107/0x163 [ 59.605237] ubsan_epilogue+0xb/0x5a [ 59.606094] __ubsan_handle_shift_out_of_bounds.cold+0xb2/0x20e [ 59.607335] ? lock_downgrade+0x6c0/0x6c0 [ 59.608182] ? rcu_read_lock_sched_held+0xaf/0xe0 [ 59.609166] io_uring_create.cold+0x99/0x149 [ 59.610114] io_uring_setup+0xd6/0x140 [ 59.610975] ? io_uring_create+0x2510/0x2510 [ 59.611945] ? lockdep_hardirqs_on_prepare+0x286/0x400 [ 59.613007] ? syscall_enter_from_user_mode+0x27/0x80 [ 59.614038] ? trace_hardirqs_on+0x5b/0x180 [ 59.615056] do_syscall_64+0x2d/0x40 [ 59.615940] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 59.617007] RIP: 0033:0x7f2bb8a0b239 This is caused by roundup_pow_of_two() if the input entries larger enough, e.g. 2^32-1. For sq_entries, it will check first and we allow at most IORING_MAX_ENTRIES, so it is okay. But for cq_entries, we do round up first, that may overflow and truncate it to 0, which is not the expected behavior. So check the cq size first and then do round up. Fixes: 88ec3211e463 ("io_uring: round-up cq size before comparing with rounded sq size") Reported-by: Abaci Fuzz <abaci@linux.alibaba.com> Signed-off-by: Joseph Qi <joseph.qi@linux.alibaba.com> Reviewed-by: Stefano Garzarella <sgarzare@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-11-24 15:03:03 +08:00
if (!p->cq_entries)
return -EINVAL;
if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
if (!(p->flags & IORING_SETUP_CLAMP))
return -EINVAL;
p->cq_entries = IORING_MAX_CQ_ENTRIES;
}
io_uring: fix shift-out-of-bounds when round up cq size Abaci Fuzz reported a shift-out-of-bounds BUG in io_uring_create(): [ 59.598207] UBSAN: shift-out-of-bounds in ./include/linux/log2.h:57:13 [ 59.599665] shift exponent 64 is too large for 64-bit type 'long unsigned int' [ 59.601230] CPU: 0 PID: 963 Comm: a.out Not tainted 5.10.0-rc4+ #3 [ 59.602502] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [ 59.603673] Call Trace: [ 59.604286] dump_stack+0x107/0x163 [ 59.605237] ubsan_epilogue+0xb/0x5a [ 59.606094] __ubsan_handle_shift_out_of_bounds.cold+0xb2/0x20e [ 59.607335] ? lock_downgrade+0x6c0/0x6c0 [ 59.608182] ? rcu_read_lock_sched_held+0xaf/0xe0 [ 59.609166] io_uring_create.cold+0x99/0x149 [ 59.610114] io_uring_setup+0xd6/0x140 [ 59.610975] ? io_uring_create+0x2510/0x2510 [ 59.611945] ? lockdep_hardirqs_on_prepare+0x286/0x400 [ 59.613007] ? syscall_enter_from_user_mode+0x27/0x80 [ 59.614038] ? trace_hardirqs_on+0x5b/0x180 [ 59.615056] do_syscall_64+0x2d/0x40 [ 59.615940] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 59.617007] RIP: 0033:0x7f2bb8a0b239 This is caused by roundup_pow_of_two() if the input entries larger enough, e.g. 2^32-1. For sq_entries, it will check first and we allow at most IORING_MAX_ENTRIES, so it is okay. But for cq_entries, we do round up first, that may overflow and truncate it to 0, which is not the expected behavior. So check the cq size first and then do round up. Fixes: 88ec3211e463 ("io_uring: round-up cq size before comparing with rounded sq size") Reported-by: Abaci Fuzz <abaci@linux.alibaba.com> Signed-off-by: Joseph Qi <joseph.qi@linux.alibaba.com> Reviewed-by: Stefano Garzarella <sgarzare@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-11-24 15:03:03 +08:00
p->cq_entries = roundup_pow_of_two(p->cq_entries);
if (p->cq_entries < p->sq_entries)
return -EINVAL;
} else {
p->cq_entries = 2 * p->sq_entries;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ctx = io_ring_ctx_alloc(p);
if (!ctx)
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return -ENOMEM;
if ((ctx->flags & IORING_SETUP_DEFER_TASKRUN) &&
!(ctx->flags & IORING_SETUP_IOPOLL) &&
!(ctx->flags & IORING_SETUP_SQPOLL))
ctx->task_complete = true;
if (ctx->task_complete || (ctx->flags & IORING_SETUP_IOPOLL))
ctx->lockless_cq = true;
/*
* lazy poll_wq activation relies on ->task_complete for synchronisation
* purposes, see io_activate_pollwq()
*/
if (!ctx->task_complete)
ctx->poll_activated = true;
/*
* When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
* space applications don't need to do io completion events
* polling again, they can rely on io_sq_thread to do polling
* work, which can reduce cpu usage and uring_lock contention.
*/
if (ctx->flags & IORING_SETUP_IOPOLL &&
!(ctx->flags & IORING_SETUP_SQPOLL))
ctx->syscall_iopoll = 1;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ctx->compat = in_compat_syscall();
if (!ns_capable_noaudit(&init_user_ns, CAP_IPC_LOCK))
ctx->user = get_uid(current_user());
/*
* For SQPOLL, we just need a wakeup, always. For !SQPOLL, if
* COOP_TASKRUN is set, then IPIs are never needed by the app.
*/
ret = -EINVAL;
if (ctx->flags & IORING_SETUP_SQPOLL) {
/* IPI related flags don't make sense with SQPOLL */
if (ctx->flags & (IORING_SETUP_COOP_TASKRUN |
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
IORING_SETUP_TASKRUN_FLAG |
IORING_SETUP_DEFER_TASKRUN))
goto err;
ctx->notify_method = TWA_SIGNAL_NO_IPI;
} else if (ctx->flags & IORING_SETUP_COOP_TASKRUN) {
ctx->notify_method = TWA_SIGNAL_NO_IPI;
} else {
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
if (ctx->flags & IORING_SETUP_TASKRUN_FLAG &&
!(ctx->flags & IORING_SETUP_DEFER_TASKRUN))
goto err;
ctx->notify_method = TWA_SIGNAL;
}
io_uring: add IORING_SETUP_DEFER_TASKRUN Allow deferring async tasks until the user calls io_uring_enter(2) with the IORING_ENTER_GETEVENTS flag. Enable this mode with a flag at io_uring_setup time. This functionality requires that the later io_uring_enter will be called from the same submission task, and therefore restrict this flag to work only when IORING_SETUP_SINGLE_ISSUER is also set. Being able to hand pick when tasks are run prevents the problem where there is current work to be done, however task work runs anyway. For example, a common workload would obtain a batch of CQEs, and process each one. Interrupting this to additional taskwork would add latency but not gain anything. If instead task work is deferred to just before more CQEs are obtained then no additional latency is added. The way this is implemented is by trying to keep task work local to a io_ring_ctx, rather than to the submission task. This is required, as the application will want to wake up only a single io_ring_ctx at a time to process work, and so the lists of work have to be kept separate. This has some other benefits like not having to check the task continually in handle_tw_list (and potentially unlocking/locking those), and reducing locks in the submit & process completions path. There are networking cases where using this option can reduce request latency by 50%. For example a contrived example using [1] where the client sends 2k data and receives the same data back while doing some system calls (to trigger task work) shows this reduction. The reason ends up being that if sending responses is delayed by processing task work, then the client side sits idle. Whereas reordering the sends first means that the client runs it's workload in parallel with the local task work. [1]: Using https://github.com/DylanZA/netbench/tree/defer_run Client: ./netbench --client_only 1 --control_port 10000 --host <host> --tx "epoll --threads 16 --per_thread 1 --size 2048 --resp 2048 --workload 1000" Server: ./netbench --server_only 1 --control_port 10000 --rx "io_uring --defer_taskrun 0 --workload 100" --rx "io_uring --defer_taskrun 1 --workload 100" Signed-off-by: Dylan Yudaken <dylany@fb.com> Link: https://lore.kernel.org/r/20220830125013.570060-5-dylany@fb.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-08-30 20:50:10 +08:00
/*
* For DEFER_TASKRUN we require the completion task to be the same as the
* submission task. This implies that there is only one submitter, so enforce
* that.
*/
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN &&
!(ctx->flags & IORING_SETUP_SINGLE_ISSUER)) {
goto err;
}
/*
* This is just grabbed for accounting purposes. When a process exits,
* the mm is exited and dropped before the files, hence we need to hang
* on to this mm purely for the purposes of being able to unaccount
* memory (locked/pinned vm). It's not used for anything else.
*/
mmgrab(current->mm);
ctx->mm_account = current->mm;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
ret = io_allocate_scq_urings(ctx, p);
if (ret)
goto err;
ret = io_sq_offload_create(ctx, p);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
if (ret)
goto err;
ret = io_rsrc_init(ctx);
if (ret)
goto err;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
p->sq_off.head = offsetof(struct io_rings, sq.head);
p->sq_off.tail = offsetof(struct io_rings, sq.tail);
p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
p->sq_off.flags = offsetof(struct io_rings, sq_flags);
p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
if (!(ctx->flags & IORING_SETUP_NO_SQARRAY))
p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
p->sq_off.resv1 = 0;
if (!(ctx->flags & IORING_SETUP_NO_MMAP))
p->sq_off.user_addr = 0;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
p->cq_off.head = offsetof(struct io_rings, cq.head);
p->cq_off.tail = offsetof(struct io_rings, cq.tail);
p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
p->cq_off.cqes = offsetof(struct io_rings, cqes);
p->cq_off.flags = offsetof(struct io_rings, cq_flags);
p->cq_off.resv1 = 0;
if (!(ctx->flags & IORING_SETUP_NO_MMAP))
p->cq_off.user_addr = 0;
p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS |
IORING_FEAT_RSRC_TAGS | IORING_FEAT_CQE_SKIP |
IORING_FEAT_LINKED_FILE | IORING_FEAT_REG_REG_RING;
if (copy_to_user(params, p, sizeof(*p))) {
ret = -EFAULT;
goto err;
}
io_uring: don't touch 'ctx' after installing file descriptor As soon as we install the file descriptor, we have to assume that it can get arbitrarily closed. We currently account memory (and note that we did) after installing the ring fd, which means that it could be a potential use-after-free condition if the fd is closed right after being installed, but before we fiddle with the ctx. In fact, syzbot reported this exact scenario: BUG: KASAN: use-after-free in io_account_mem fs/io_uring.c:7397 [inline] BUG: KASAN: use-after-free in io_uring_create fs/io_uring.c:8369 [inline] BUG: KASAN: use-after-free in io_uring_setup+0x2797/0x2910 fs/io_uring.c:8400 Read of size 1 at addr ffff888087a41044 by task syz-executor.5/18145 CPU: 0 PID: 18145 Comm: syz-executor.5 Not tainted 5.8.0-rc7-next-20200729-syzkaller #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x18f/0x20d lib/dump_stack.c:118 print_address_description.constprop.0.cold+0xae/0x497 mm/kasan/report.c:383 __kasan_report mm/kasan/report.c:513 [inline] kasan_report.cold+0x1f/0x37 mm/kasan/report.c:530 io_account_mem fs/io_uring.c:7397 [inline] io_uring_create fs/io_uring.c:8369 [inline] io_uring_setup+0x2797/0x2910 fs/io_uring.c:8400 do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x45c429 Code: 8d b6 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 5b b6 fb ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007f8f121d0c78 EFLAGS: 00000246 ORIG_RAX: 00000000000001a9 RAX: ffffffffffffffda RBX: 0000000000008540 RCX: 000000000045c429 RDX: 0000000000000000 RSI: 0000000020000040 RDI: 0000000000000196 RBP: 000000000078bf38 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 000000000078bf0c R13: 00007fff86698cff R14: 00007f8f121d19c0 R15: 000000000078bf0c Move the accounting of the ring used locked memory before we get and install the ring file descriptor. Cc: stable@vger.kernel.org Reported-by: syzbot+9d46305e76057f30c74e@syzkaller.appspotmail.com Fixes: 309758254ea6 ("io_uring: report pinned memory usage") Reviewed-by: Stefano Garzarella <sgarzare@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-07-31 03:43:53 +08:00
if (ctx->flags & IORING_SETUP_SINGLE_ISSUER
&& !(ctx->flags & IORING_SETUP_R_DISABLED))
WRITE_ONCE(ctx->submitter_task, get_task_struct(current));
file = io_uring_get_file(ctx);
if (IS_ERR(file)) {
ret = PTR_ERR(file);
goto err;
}
ret = __io_uring_add_tctx_node(ctx);
if (ret)
goto err_fput;
tctx = current->io_uring;
/*
* Install ring fd as the very last thing, so we don't risk someone
* having closed it before we finish setup
*/
if (p->flags & IORING_SETUP_REGISTERED_FD_ONLY)
ret = io_ring_add_registered_file(tctx, file, 0, IO_RINGFD_REG_MAX);
else
ret = io_uring_install_fd(file);
if (ret < 0)
goto err_fput;
trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return ret;
err:
io_ring_ctx_wait_and_kill(ctx);
return ret;
err_fput:
fput(file);
return ret;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
/*
* Sets up an aio uring context, and returns the fd. Applications asks for a
* ring size, we return the actual sq/cq ring sizes (among other things) in the
* params structure passed in.
*/
static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
{
struct io_uring_params p;
int i;
if (copy_from_user(&p, params, sizeof(p)))
return -EFAULT;
for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
if (p.resv[i])
return -EINVAL;
}
if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
IORING_SETUP_R_DISABLED | IORING_SETUP_SUBMIT_ALL |
IORING_SETUP_COOP_TASKRUN | IORING_SETUP_TASKRUN_FLAG |
IORING_SETUP_SQE128 | IORING_SETUP_CQE32 |
IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN |
IORING_SETUP_NO_MMAP | IORING_SETUP_REGISTERED_FD_ONLY |
IORING_SETUP_NO_SQARRAY))
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return -EINVAL;
return io_uring_create(entries, &p, params);
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
}
static inline bool io_uring_allowed(void)
{
int disabled = READ_ONCE(sysctl_io_uring_disabled);
kgid_t io_uring_group;
if (disabled == 2)
return false;
if (disabled == 0 || capable(CAP_SYS_ADMIN))
return true;
io_uring_group = make_kgid(&init_user_ns, sysctl_io_uring_group);
if (!gid_valid(io_uring_group))
return false;
return in_group_p(io_uring_group);
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
SYSCALL_DEFINE2(io_uring_setup, u32, entries,
struct io_uring_params __user *, params)
{
if (!io_uring_allowed())
return -EPERM;
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return io_uring_setup(entries, params);
}
static __cold int io_probe(struct io_ring_ctx *ctx, void __user *arg,
unsigned nr_args)
{
struct io_uring_probe *p;
size_t size;
int i, ret;
size = struct_size(p, ops, nr_args);
if (size == SIZE_MAX)
return -EOVERFLOW;
p = kzalloc(size, GFP_KERNEL);
if (!p)
return -ENOMEM;
ret = -EFAULT;
if (copy_from_user(p, arg, size))
goto out;
ret = -EINVAL;
if (memchr_inv(p, 0, size))
goto out;
p->last_op = IORING_OP_LAST - 1;
if (nr_args > IORING_OP_LAST)
nr_args = IORING_OP_LAST;
for (i = 0; i < nr_args; i++) {
p->ops[i].op = i;
if (!io_issue_defs[i].not_supported)
p->ops[i].flags = IO_URING_OP_SUPPORTED;
}
p->ops_len = i;
ret = 0;
if (copy_to_user(arg, p, size))
ret = -EFAULT;
out:
kfree(p);
return ret;
}
static int io_register_personality(struct io_ring_ctx *ctx)
{
const struct cred *creds;
u32 id;
int ret;
creds = get_current_cred();
ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds,
XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL);
if (ret < 0) {
put_cred(creds);
return ret;
}
return id;
}
static __cold int io_register_restrictions(struct io_ring_ctx *ctx,
void __user *arg, unsigned int nr_args)
{
struct io_uring_restriction *res;
size_t size;
int i, ret;
/* Restrictions allowed only if rings started disabled */
if (!(ctx->flags & IORING_SETUP_R_DISABLED))
return -EBADFD;
/* We allow only a single restrictions registration */
if (ctx->restrictions.registered)
return -EBUSY;
if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
return -EINVAL;
size = array_size(nr_args, sizeof(*res));
if (size == SIZE_MAX)
return -EOVERFLOW;
res = memdup_user(arg, size);
if (IS_ERR(res))
return PTR_ERR(res);
ret = 0;
for (i = 0; i < nr_args; i++) {
switch (res[i].opcode) {
case IORING_RESTRICTION_REGISTER_OP:
if (res[i].register_op >= IORING_REGISTER_LAST) {
ret = -EINVAL;
goto out;
}
__set_bit(res[i].register_op,
ctx->restrictions.register_op);
break;
case IORING_RESTRICTION_SQE_OP:
if (res[i].sqe_op >= IORING_OP_LAST) {
ret = -EINVAL;
goto out;
}
__set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
break;
case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
break;
case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
break;
default:
ret = -EINVAL;
goto out;
}
}
out:
/* Reset all restrictions if an error happened */
if (ret != 0)
memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
else
ctx->restrictions.registered = true;
kfree(res);
return ret;
}
static int io_register_enable_rings(struct io_ring_ctx *ctx)
{
if (!(ctx->flags & IORING_SETUP_R_DISABLED))
return -EBADFD;
if (ctx->flags & IORING_SETUP_SINGLE_ISSUER && !ctx->submitter_task) {
WRITE_ONCE(ctx->submitter_task, get_task_struct(current));
/*
* Lazy activation attempts would fail if it was polled before
* submitter_task is set.
*/
if (wq_has_sleeper(&ctx->poll_wq))
io_activate_pollwq(ctx);
}
if (ctx->restrictions.registered)
ctx->restricted = 1;
ctx->flags &= ~IORING_SETUP_R_DISABLED;
if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait))
wake_up(&ctx->sq_data->wait);
return 0;
}
static __cold int __io_register_iowq_aff(struct io_ring_ctx *ctx,
cpumask_var_t new_mask)
{
int ret;
if (!(ctx->flags & IORING_SETUP_SQPOLL)) {
ret = io_wq_cpu_affinity(current->io_uring, new_mask);
} else {
mutex_unlock(&ctx->uring_lock);
ret = io_sqpoll_wq_cpu_affinity(ctx, new_mask);
mutex_lock(&ctx->uring_lock);
}
return ret;
}
static __cold int io_register_iowq_aff(struct io_ring_ctx *ctx,
void __user *arg, unsigned len)
{
cpumask_var_t new_mask;
int ret;
if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
return -ENOMEM;
cpumask_clear(new_mask);
if (len > cpumask_size())
len = cpumask_size();
if (in_compat_syscall()) {
ret = compat_get_bitmap(cpumask_bits(new_mask),
(const compat_ulong_t __user *)arg,
len * 8 /* CHAR_BIT */);
} else {
ret = copy_from_user(new_mask, arg, len);
}
if (ret) {
free_cpumask_var(new_mask);
return -EFAULT;
}
ret = __io_register_iowq_aff(ctx, new_mask);
free_cpumask_var(new_mask);
return ret;
}
static __cold int io_unregister_iowq_aff(struct io_ring_ctx *ctx)
{
return __io_register_iowq_aff(ctx, NULL);
}
static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx,
void __user *arg)
__must_hold(&ctx->uring_lock)
{
struct io_tctx_node *node;
struct io_uring_task *tctx = NULL;
struct io_sq_data *sqd = NULL;
__u32 new_count[2];
int i, ret;
if (copy_from_user(new_count, arg, sizeof(new_count)))
return -EFAULT;
for (i = 0; i < ARRAY_SIZE(new_count); i++)
if (new_count[i] > INT_MAX)
return -EINVAL;
if (ctx->flags & IORING_SETUP_SQPOLL) {
sqd = ctx->sq_data;
if (sqd) {
io_uring: drop ctx->uring_lock before acquiring sqd->lock The SQPOLL thread dictates the lock order, and we hold the ctx->uring_lock for all the registration opcodes. We also hold a ref to the ctx, and we do drop the lock for other reasons to quiesce, so it's fine to drop the ctx lock temporarily to grab the sqd->lock. This fixes the following lockdep splat: ====================================================== WARNING: possible circular locking dependency detected 5.14.0-syzkaller #0 Not tainted ------------------------------------------------------ syz-executor.5/25433 is trying to acquire lock: ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: io_register_iowq_max_workers fs/io_uring.c:10551 [inline] ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: __io_uring_register fs/io_uring.c:10757 [inline] ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: __do_sys_io_uring_register+0x10aa/0x2e70 fs/io_uring.c:10792 but task is already holding lock: ffff8880885b40a8 (&ctx->uring_lock){+.+.}-{3:3}, at: __do_sys_io_uring_register+0x2e1/0x2e70 fs/io_uring.c:10791 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&ctx->uring_lock){+.+.}-{3:3}: __mutex_lock_common kernel/locking/mutex.c:596 [inline] __mutex_lock+0x131/0x12f0 kernel/locking/mutex.c:729 __io_sq_thread fs/io_uring.c:7291 [inline] io_sq_thread+0x65a/0x1370 fs/io_uring.c:7368 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:295 -> #0 (&sqd->lock){+.+.}-{3:3}: check_prev_add kernel/locking/lockdep.c:3051 [inline] check_prevs_add kernel/locking/lockdep.c:3174 [inline] validate_chain kernel/locking/lockdep.c:3789 [inline] __lock_acquire+0x2a07/0x54a0 kernel/locking/lockdep.c:5015 lock_acquire kernel/locking/lockdep.c:5625 [inline] lock_acquire+0x1ab/0x510 kernel/locking/lockdep.c:5590 __mutex_lock_common kernel/locking/mutex.c:596 [inline] __mutex_lock+0x131/0x12f0 kernel/locking/mutex.c:729 io_register_iowq_max_workers fs/io_uring.c:10551 [inline] __io_uring_register fs/io_uring.c:10757 [inline] __do_sys_io_uring_register+0x10aa/0x2e70 fs/io_uring.c:10792 do_syscall_x64 arch/x86/entry/common.c:50 [inline] do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80 entry_SYSCALL_64_after_hwframe+0x44/0xae other info that might help us debug this: Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&ctx->uring_lock); lock(&sqd->lock); lock(&ctx->uring_lock); lock(&sqd->lock); *** DEADLOCK *** Fixes: 2e480058ddc2 ("io-wq: provide a way to limit max number of workers") Reported-by: syzbot+97fa56483f69d677969f@syzkaller.appspotmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-09-09 09:07:26 +08:00
/*
* Observe the correct sqd->lock -> ctx->uring_lock
* ordering. Fine to drop uring_lock here, we hold
* a ref to the ctx.
*/
refcount_inc(&sqd->refs);
io_uring: drop ctx->uring_lock before acquiring sqd->lock The SQPOLL thread dictates the lock order, and we hold the ctx->uring_lock for all the registration opcodes. We also hold a ref to the ctx, and we do drop the lock for other reasons to quiesce, so it's fine to drop the ctx lock temporarily to grab the sqd->lock. This fixes the following lockdep splat: ====================================================== WARNING: possible circular locking dependency detected 5.14.0-syzkaller #0 Not tainted ------------------------------------------------------ syz-executor.5/25433 is trying to acquire lock: ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: io_register_iowq_max_workers fs/io_uring.c:10551 [inline] ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: __io_uring_register fs/io_uring.c:10757 [inline] ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: __do_sys_io_uring_register+0x10aa/0x2e70 fs/io_uring.c:10792 but task is already holding lock: ffff8880885b40a8 (&ctx->uring_lock){+.+.}-{3:3}, at: __do_sys_io_uring_register+0x2e1/0x2e70 fs/io_uring.c:10791 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&ctx->uring_lock){+.+.}-{3:3}: __mutex_lock_common kernel/locking/mutex.c:596 [inline] __mutex_lock+0x131/0x12f0 kernel/locking/mutex.c:729 __io_sq_thread fs/io_uring.c:7291 [inline] io_sq_thread+0x65a/0x1370 fs/io_uring.c:7368 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:295 -> #0 (&sqd->lock){+.+.}-{3:3}: check_prev_add kernel/locking/lockdep.c:3051 [inline] check_prevs_add kernel/locking/lockdep.c:3174 [inline] validate_chain kernel/locking/lockdep.c:3789 [inline] __lock_acquire+0x2a07/0x54a0 kernel/locking/lockdep.c:5015 lock_acquire kernel/locking/lockdep.c:5625 [inline] lock_acquire+0x1ab/0x510 kernel/locking/lockdep.c:5590 __mutex_lock_common kernel/locking/mutex.c:596 [inline] __mutex_lock+0x131/0x12f0 kernel/locking/mutex.c:729 io_register_iowq_max_workers fs/io_uring.c:10551 [inline] __io_uring_register fs/io_uring.c:10757 [inline] __do_sys_io_uring_register+0x10aa/0x2e70 fs/io_uring.c:10792 do_syscall_x64 arch/x86/entry/common.c:50 [inline] do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80 entry_SYSCALL_64_after_hwframe+0x44/0xae other info that might help us debug this: Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&ctx->uring_lock); lock(&sqd->lock); lock(&ctx->uring_lock); lock(&sqd->lock); *** DEADLOCK *** Fixes: 2e480058ddc2 ("io-wq: provide a way to limit max number of workers") Reported-by: syzbot+97fa56483f69d677969f@syzkaller.appspotmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-09-09 09:07:26 +08:00
mutex_unlock(&ctx->uring_lock);
mutex_lock(&sqd->lock);
io_uring: drop ctx->uring_lock before acquiring sqd->lock The SQPOLL thread dictates the lock order, and we hold the ctx->uring_lock for all the registration opcodes. We also hold a ref to the ctx, and we do drop the lock for other reasons to quiesce, so it's fine to drop the ctx lock temporarily to grab the sqd->lock. This fixes the following lockdep splat: ====================================================== WARNING: possible circular locking dependency detected 5.14.0-syzkaller #0 Not tainted ------------------------------------------------------ syz-executor.5/25433 is trying to acquire lock: ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: io_register_iowq_max_workers fs/io_uring.c:10551 [inline] ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: __io_uring_register fs/io_uring.c:10757 [inline] ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: __do_sys_io_uring_register+0x10aa/0x2e70 fs/io_uring.c:10792 but task is already holding lock: ffff8880885b40a8 (&ctx->uring_lock){+.+.}-{3:3}, at: __do_sys_io_uring_register+0x2e1/0x2e70 fs/io_uring.c:10791 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&ctx->uring_lock){+.+.}-{3:3}: __mutex_lock_common kernel/locking/mutex.c:596 [inline] __mutex_lock+0x131/0x12f0 kernel/locking/mutex.c:729 __io_sq_thread fs/io_uring.c:7291 [inline] io_sq_thread+0x65a/0x1370 fs/io_uring.c:7368 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:295 -> #0 (&sqd->lock){+.+.}-{3:3}: check_prev_add kernel/locking/lockdep.c:3051 [inline] check_prevs_add kernel/locking/lockdep.c:3174 [inline] validate_chain kernel/locking/lockdep.c:3789 [inline] __lock_acquire+0x2a07/0x54a0 kernel/locking/lockdep.c:5015 lock_acquire kernel/locking/lockdep.c:5625 [inline] lock_acquire+0x1ab/0x510 kernel/locking/lockdep.c:5590 __mutex_lock_common kernel/locking/mutex.c:596 [inline] __mutex_lock+0x131/0x12f0 kernel/locking/mutex.c:729 io_register_iowq_max_workers fs/io_uring.c:10551 [inline] __io_uring_register fs/io_uring.c:10757 [inline] __do_sys_io_uring_register+0x10aa/0x2e70 fs/io_uring.c:10792 do_syscall_x64 arch/x86/entry/common.c:50 [inline] do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80 entry_SYSCALL_64_after_hwframe+0x44/0xae other info that might help us debug this: Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&ctx->uring_lock); lock(&sqd->lock); lock(&ctx->uring_lock); lock(&sqd->lock); *** DEADLOCK *** Fixes: 2e480058ddc2 ("io-wq: provide a way to limit max number of workers") Reported-by: syzbot+97fa56483f69d677969f@syzkaller.appspotmail.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
2021-09-09 09:07:26 +08:00
mutex_lock(&ctx->uring_lock);
if (sqd->thread)
tctx = sqd->thread->io_uring;
}
} else {
tctx = current->io_uring;
}
BUILD_BUG_ON(sizeof(new_count) != sizeof(ctx->iowq_limits));
for (i = 0; i < ARRAY_SIZE(new_count); i++)
if (new_count[i])
ctx->iowq_limits[i] = new_count[i];
ctx->iowq_limits_set = true;
if (tctx && tctx->io_wq) {
ret = io_wq_max_workers(tctx->io_wq, new_count);
if (ret)
goto err;
} else {
memset(new_count, 0, sizeof(new_count));
}
if (sqd) {
mutex_unlock(&sqd->lock);
io_put_sq_data(sqd);
}
if (copy_to_user(arg, new_count, sizeof(new_count)))
return -EFAULT;
/* that's it for SQPOLL, only the SQPOLL task creates requests */
if (sqd)
return 0;
/* now propagate the restriction to all registered users */
list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
struct io_uring_task *tctx = node->task->io_uring;
if (WARN_ON_ONCE(!tctx->io_wq))
continue;
for (i = 0; i < ARRAY_SIZE(new_count); i++)
new_count[i] = ctx->iowq_limits[i];
/* ignore errors, it always returns zero anyway */
(void)io_wq_max_workers(tctx->io_wq, new_count);
}
return 0;
err:
if (sqd) {
mutex_unlock(&sqd->lock);
io_put_sq_data(sqd);
}
return ret;
}
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
void __user *arg, unsigned nr_args)
__releases(ctx->uring_lock)
__acquires(ctx->uring_lock)
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
{
int ret;
/*
* We don't quiesce the refs for register anymore and so it can't be
* dying as we're holding a file ref here.
*/
if (WARN_ON_ONCE(percpu_ref_is_dying(&ctx->refs)))
return -ENXIO;
if (ctx->submitter_task && ctx->submitter_task != current)
return -EEXIST;
if (ctx->restricted) {
opcode = array_index_nospec(opcode, IORING_REGISTER_LAST);
if (!test_bit(opcode, ctx->restrictions.register_op))
return -EACCES;
}
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
switch (opcode) {
case IORING_REGISTER_BUFFERS:
ret = -EFAULT;
if (!arg)
break;
ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL);
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
break;
case IORING_UNREGISTER_BUFFERS:
ret = -EINVAL;
if (arg || nr_args)
break;
ret = io_sqe_buffers_unregister(ctx);
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
break;
case IORING_REGISTER_FILES:
ret = -EFAULT;
if (!arg)
break;
ret = io_sqe_files_register(ctx, arg, nr_args, NULL);
break;
case IORING_UNREGISTER_FILES:
ret = -EINVAL;
if (arg || nr_args)
break;
ret = io_sqe_files_unregister(ctx);
break;
case IORING_REGISTER_FILES_UPDATE:
ret = io_register_files_update(ctx, arg, nr_args);
break;
case IORING_REGISTER_EVENTFD:
ret = -EINVAL;
if (nr_args != 1)
break;
ret = io_eventfd_register(ctx, arg, 0);
break;
case IORING_REGISTER_EVENTFD_ASYNC:
ret = -EINVAL;
if (nr_args != 1)
break;
ret = io_eventfd_register(ctx, arg, 1);
break;
case IORING_UNREGISTER_EVENTFD:
ret = -EINVAL;
if (arg || nr_args)
break;
ret = io_eventfd_unregister(ctx);
break;
case IORING_REGISTER_PROBE:
ret = -EINVAL;
if (!arg || nr_args > 256)
break;
ret = io_probe(ctx, arg, nr_args);
break;
case IORING_REGISTER_PERSONALITY:
ret = -EINVAL;
if (arg || nr_args)
break;
ret = io_register_personality(ctx);
break;
case IORING_UNREGISTER_PERSONALITY:
ret = -EINVAL;
if (arg)
break;
ret = io_unregister_personality(ctx, nr_args);
break;
case IORING_REGISTER_ENABLE_RINGS:
ret = -EINVAL;
if (arg || nr_args)
break;
ret = io_register_enable_rings(ctx);
break;
case IORING_REGISTER_RESTRICTIONS:
ret = io_register_restrictions(ctx, arg, nr_args);
break;
case IORING_REGISTER_FILES2:
ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE);
break;
case IORING_REGISTER_FILES_UPDATE2:
ret = io_register_rsrc_update(ctx, arg, nr_args,
IORING_RSRC_FILE);
break;
case IORING_REGISTER_BUFFERS2:
ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER);
break;
case IORING_REGISTER_BUFFERS_UPDATE:
ret = io_register_rsrc_update(ctx, arg, nr_args,
IORING_RSRC_BUFFER);
break;
case IORING_REGISTER_IOWQ_AFF:
ret = -EINVAL;
if (!arg || !nr_args)
break;
ret = io_register_iowq_aff(ctx, arg, nr_args);
break;
case IORING_UNREGISTER_IOWQ_AFF:
ret = -EINVAL;
if (arg || nr_args)
break;
ret = io_unregister_iowq_aff(ctx);
break;
case IORING_REGISTER_IOWQ_MAX_WORKERS:
ret = -EINVAL;
if (!arg || nr_args != 2)
break;
ret = io_register_iowq_max_workers(ctx, arg);
break;
io_uring: add support for registering ring file descriptors Lots of workloads use multiple threads, in which case the file table is shared between them. This makes getting and putting the ring file descriptor for each io_uring_enter(2) system call more expensive, as it involves an atomic get and put for each call. Similarly to how we allow registering normal file descriptors to avoid this overhead, add support for an io_uring_register(2) API that allows to register the ring fds themselves: 1) IORING_REGISTER_RING_FDS - takes an array of io_uring_rsrc_update structs, and registers them with the task. 2) IORING_UNREGISTER_RING_FDS - takes an array of io_uring_src_update structs, and unregisters them. When a ring fd is registered, it is internally represented by an offset. This offset is returned to the application, and the application then uses this offset and sets IORING_ENTER_REGISTERED_RING for the io_uring_enter(2) system call. This works just like using a registered file descriptor, rather than a real one, in an SQE, where IOSQE_FIXED_FILE gets set to tell io_uring that we're using an internal offset/descriptor rather than a real file descriptor. In initial testing, this provides a nice bump in performance for threaded applications in real world cases where the batch count (eg number of requests submitted per io_uring_enter(2) invocation) is low. In a microbenchmark, submitting NOP requests, we see the following increases in performance: Requests per syscall Baseline Registered Increase ---------------------------------------------------------------- 1 ~7030K ~8080K +15% 2 ~13120K ~14800K +13% 4 ~22740K ~25300K +11% Co-developed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-03-04 23:22:22 +08:00
case IORING_REGISTER_RING_FDS:
ret = io_ringfd_register(ctx, arg, nr_args);
break;
case IORING_UNREGISTER_RING_FDS:
ret = io_ringfd_unregister(ctx, arg, nr_args);
break;
io_uring: add support for ring mapped supplied buffers Provided buffers allow an application to supply io_uring with buffers that can then be grabbed for a read/receive request, when the data source is ready to deliver data. The existing scheme relies on using IORING_OP_PROVIDE_BUFFERS to do that, but it can be difficult to use in real world applications. It's pretty efficient if the application is able to supply back batches of provided buffers when they have been consumed and the application is ready to recycle them, but if fragmentation occurs in the buffer space, it can become difficult to supply enough buffers at the time. This hurts efficiency. Add a register op, IORING_REGISTER_PBUF_RING, which allows an application to setup a shared queue for each buffer group of provided buffers. The application can then supply buffers simply by adding them to this ring, and the kernel can consume then just as easily. The ring shares the head with the application, the tail remains private in the kernel. Provided buffers setup with IORING_REGISTER_PBUF_RING cannot use IORING_OP_{PROVIDE,REMOVE}_BUFFERS for adding or removing entries to the ring, they must use the mapped ring. Mapped provided buffer rings can co-exist with normal provided buffers, just not within the same group ID. To gauge overhead of the existing scheme and evaluate the mapped ring approach, a simple NOP benchmark was written. It uses a ring of 128 entries, and submits/completes 32 at the time. 'Replenish' is how many buffers are provided back at the time after they have been consumed: Test Replenish NOPs/sec ================================================================ No provided buffers NA ~30M Provided buffers 32 ~16M Provided buffers 1 ~10M Ring buffers 32 ~27M Ring buffers 1 ~27M The ring mapped buffers perform almost as well as not using provided buffers at all, and they don't care if you provided 1 or more back at the same time. This means application can just replenish as they go, rather than need to batch and compact, further reducing overhead in the application. The NOP benchmark above doesn't need to do any compaction, so that overhead isn't even reflected in the above test. Co-developed-by: Dylan Yudaken <dylany@fb.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-05-01 04:38:53 +08:00
case IORING_REGISTER_PBUF_RING:
ret = -EINVAL;
if (!arg || nr_args != 1)
break;
ret = io_register_pbuf_ring(ctx, arg);
break;
case IORING_UNREGISTER_PBUF_RING:
ret = -EINVAL;
if (!arg || nr_args != 1)
break;
ret = io_unregister_pbuf_ring(ctx, arg);
break;
2022-06-19 00:00:50 +08:00
case IORING_REGISTER_SYNC_CANCEL:
ret = -EINVAL;
if (!arg || nr_args != 1)
break;
ret = io_sync_cancel(ctx, arg);
break;
case IORING_REGISTER_FILE_ALLOC_RANGE:
ret = -EINVAL;
if (!arg || nr_args)
break;
ret = io_register_file_alloc_range(ctx, arg);
break;
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
default:
ret = -EINVAL;
break;
}
return ret;
}
SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
void __user *, arg, unsigned int, nr_args)
{
struct io_ring_ctx *ctx;
long ret = -EBADF;
struct file *file;
bool use_registered_ring;
use_registered_ring = !!(opcode & IORING_REGISTER_USE_REGISTERED_RING);
opcode &= ~IORING_REGISTER_USE_REGISTERED_RING;
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
if (opcode >= IORING_REGISTER_LAST)
return -EINVAL;
if (use_registered_ring) {
/*
* Ring fd has been registered via IORING_REGISTER_RING_FDS, we
* need only dereference our task private array to find it.
*/
struct io_uring_task *tctx = current->io_uring;
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
if (unlikely(!tctx || fd >= IO_RINGFD_REG_MAX))
return -EINVAL;
fd = array_index_nospec(fd, IO_RINGFD_REG_MAX);
file = tctx->registered_rings[fd];
if (unlikely(!file))
return -EBADF;
} else {
file = fget(fd);
if (unlikely(!file))
return -EBADF;
ret = -EOPNOTSUPP;
if (!io_is_uring_fops(file))
goto out_fput;
}
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
ctx = file->private_data;
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
mutex_lock(&ctx->uring_lock);
ret = __io_uring_register(ctx, opcode, arg, nr_args);
mutex_unlock(&ctx->uring_lock);
trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, ret);
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
out_fput:
if (!use_registered_ring)
fput(file);
io_uring: add support for pre-mapped user IO buffers If we have fixed user buffers, we can map them into the kernel when we setup the io_uring. That avoids the need to do get_user_pages() for each and every IO. To utilize this feature, the application must call io_uring_register() after having setup an io_uring instance, passing in IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to an iovec array, and the nr_args should contain how many iovecs the application wishes to map. If successful, these buffers are now mapped into the kernel, eligible for IO. To use these fixed buffers, the application must use the IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len must point to somewhere inside the indexed buffer. The application may register buffers throughout the lifetime of the io_uring instance. It can call io_uring_register() with IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of buffers, and then register a new set. The application need not unregister buffers explicitly before shutting down the io_uring instance. It's perfectly valid to setup a larger buffer, and then sometimes only use parts of it for an IO. As long as the range is within the originally mapped region, it will work just fine. For now, buffers must not be file backed. If file backed buffers are passed in, the registration will fail with -1/EOPNOTSUPP. This restriction may be relaxed in the future. RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat arbitrary 1G per buffer size is also imposed. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-10 00:16:05 +08:00
return ret;
}
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
static int __init io_uring_init(void)
{
#define __BUILD_BUG_VERIFY_OFFSET_SIZE(stype, eoffset, esize, ename) do { \
BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
BUILD_BUG_ON(sizeof_field(stype, ename) != esize); \
} while (0)
#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
__BUILD_BUG_VERIFY_OFFSET_SIZE(struct io_uring_sqe, eoffset, sizeof(etype), ename)
#define BUILD_BUG_SQE_ELEM_SIZE(eoffset, esize, ename) \
__BUILD_BUG_VERIFY_OFFSET_SIZE(struct io_uring_sqe, eoffset, esize, ename)
BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
BUILD_BUG_SQE_ELEM(0, __u8, opcode);
BUILD_BUG_SQE_ELEM(1, __u8, flags);
BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
BUILD_BUG_SQE_ELEM(4, __s32, fd);
BUILD_BUG_SQE_ELEM(8, __u64, off);
BUILD_BUG_SQE_ELEM(8, __u64, addr2);
BUILD_BUG_SQE_ELEM(8, __u32, cmd_op);
BUILD_BUG_SQE_ELEM(12, __u32, __pad1);
BUILD_BUG_SQE_ELEM(16, __u64, addr);
BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
BUILD_BUG_SQE_ELEM(24, __u32, len);
BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
BUILD_BUG_SQE_ELEM(28, __u32, rename_flags);
BUILD_BUG_SQE_ELEM(28, __u32, unlink_flags);
BUILD_BUG_SQE_ELEM(28, __u32, hardlink_flags);
BUILD_BUG_SQE_ELEM(28, __u32, xattr_flags);
BUILD_BUG_SQE_ELEM(28, __u32, msg_ring_flags);
BUILD_BUG_SQE_ELEM(32, __u64, user_data);
BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
BUILD_BUG_SQE_ELEM(40, __u16, buf_group);
BUILD_BUG_SQE_ELEM(42, __u16, personality);
BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
BUILD_BUG_SQE_ELEM(44, __u32, file_index);
BUILD_BUG_SQE_ELEM(44, __u16, addr_len);
BUILD_BUG_SQE_ELEM(46, __u16, __pad3[0]);
BUILD_BUG_SQE_ELEM(48, __u64, addr3);
BUILD_BUG_SQE_ELEM_SIZE(48, 0, cmd);
BUILD_BUG_SQE_ELEM(56, __u64, __pad2);
BUILD_BUG_ON(sizeof(struct io_uring_files_update) !=
sizeof(struct io_uring_rsrc_update));
BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) >
sizeof(struct io_uring_rsrc_update2));
/* ->buf_index is u16 */
io_uring: add support for ring mapped supplied buffers Provided buffers allow an application to supply io_uring with buffers that can then be grabbed for a read/receive request, when the data source is ready to deliver data. The existing scheme relies on using IORING_OP_PROVIDE_BUFFERS to do that, but it can be difficult to use in real world applications. It's pretty efficient if the application is able to supply back batches of provided buffers when they have been consumed and the application is ready to recycle them, but if fragmentation occurs in the buffer space, it can become difficult to supply enough buffers at the time. This hurts efficiency. Add a register op, IORING_REGISTER_PBUF_RING, which allows an application to setup a shared queue for each buffer group of provided buffers. The application can then supply buffers simply by adding them to this ring, and the kernel can consume then just as easily. The ring shares the head with the application, the tail remains private in the kernel. Provided buffers setup with IORING_REGISTER_PBUF_RING cannot use IORING_OP_{PROVIDE,REMOVE}_BUFFERS for adding or removing entries to the ring, they must use the mapped ring. Mapped provided buffer rings can co-exist with normal provided buffers, just not within the same group ID. To gauge overhead of the existing scheme and evaluate the mapped ring approach, a simple NOP benchmark was written. It uses a ring of 128 entries, and submits/completes 32 at the time. 'Replenish' is how many buffers are provided back at the time after they have been consumed: Test Replenish NOPs/sec ================================================================ No provided buffers NA ~30M Provided buffers 32 ~16M Provided buffers 1 ~10M Ring buffers 32 ~27M Ring buffers 1 ~27M The ring mapped buffers perform almost as well as not using provided buffers at all, and they don't care if you provided 1 or more back at the same time. This means application can just replenish as they go, rather than need to batch and compact, further reducing overhead in the application. The NOP benchmark above doesn't need to do any compaction, so that overhead isn't even reflected in the above test. Co-developed-by: Dylan Yudaken <dylany@fb.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-05-01 04:38:53 +08:00
BUILD_BUG_ON(offsetof(struct io_uring_buf_ring, bufs) != 0);
BUILD_BUG_ON(offsetof(struct io_uring_buf, resv) !=
offsetof(struct io_uring_buf_ring, tail));
/* should fit into one byte */
BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8));
BUILD_BUG_ON(SQE_COMMON_FLAGS >= (1 << 8));
BUILD_BUG_ON((SQE_VALID_FLAGS | SQE_COMMON_FLAGS) != SQE_VALID_FLAGS);
BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int));
BUILD_BUG_ON(sizeof(atomic_t) != sizeof(u32));
io_uring_optable_init();
io_uring: annotate the struct io_kiocb slab for appropriate user copy When compiling the kernel with clang and having HARDENED_USERCOPY enabled, the liburing openat2.t test case fails during request setup: usercopy: Kernel memory overwrite attempt detected to SLUB object 'io_kiocb' (offset 24, size 24)! ------------[ cut here ]------------ kernel BUG at mm/usercopy.c:102! invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC CPU: 3 PID: 413 Comm: openat2.t Tainted: G N 6.4.3-g6995e2de6891-dirty #19 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.1-0-g3208b098f51a-prebuilt.qemu.org 04/01/2014 RIP: 0010:usercopy_abort+0x84/0x90 Code: ce 49 89 ce 48 c7 c3 68 48 98 82 48 0f 44 de 48 c7 c7 56 c6 94 82 4c 89 de 48 89 c1 41 52 41 56 53 e8 e0 51 c5 00 48 83 c4 18 <0f> 0b 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 55 41 57 41 56 RSP: 0018:ffffc900016b3da0 EFLAGS: 00010296 RAX: 0000000000000062 RBX: ffffffff82984868 RCX: 4e9b661ac6275b00 RDX: ffff8881b90ec580 RSI: ffffffff82949a64 RDI: 00000000ffffffff RBP: 0000000000000018 R08: 0000000000000000 R09: 0000000000000000 R10: ffffc900016b3c88 R11: ffffc900016b3c30 R12: 00007ffe549659e0 R13: ffff888119014000 R14: 0000000000000018 R15: 0000000000000018 FS: 00007f862e3ca680(0000) GS:ffff8881b90c0000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00005571483542a8 CR3: 0000000118c11000 CR4: 00000000003506e0 Call Trace: <TASK> ? __die_body+0x63/0xb0 ? die+0x9d/0xc0 ? do_trap+0xa7/0x180 ? usercopy_abort+0x84/0x90 ? do_error_trap+0xc6/0x110 ? usercopy_abort+0x84/0x90 ? handle_invalid_op+0x2c/0x40 ? usercopy_abort+0x84/0x90 ? exc_invalid_op+0x2f/0x40 ? asm_exc_invalid_op+0x16/0x20 ? usercopy_abort+0x84/0x90 __check_heap_object+0xe2/0x110 __check_object_size+0x142/0x3d0 io_openat2_prep+0x68/0x140 io_submit_sqes+0x28a/0x680 __se_sys_io_uring_enter+0x120/0x580 do_syscall_64+0x3d/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 RIP: 0033:0x55714834de26 Code: ca 01 0f b6 82 d0 00 00 00 8b ba cc 00 00 00 45 31 c0 31 d2 41 b9 08 00 00 00 83 e0 01 c1 e0 04 41 09 c2 b8 aa 01 00 00 0f 05 <c3> 66 0f 1f 84 00 00 00 00 00 89 30 eb 89 0f 1f 40 00 8b 00 a8 06 RSP: 002b:00007ffe549659c8 EFLAGS: 00000246 ORIG_RAX: 00000000000001aa RAX: ffffffffffffffda RBX: 00007ffe54965a50 RCX: 000055714834de26 RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000003 RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000008 R10: 0000000000000000 R11: 0000000000000246 R12: 000055714834f057 R13: 00007ffe54965a50 R14: 0000000000000001 R15: 0000557148351dd8 </TASK> Modules linked in: ---[ end trace 0000000000000000 ]--- when it tries to copy struct open_how from userspace into the per-command space in the io_kiocb. There's nothing wrong with the copy, but we're missing the appropriate annotations for allowing user copies to/from the io_kiocb slab. Allow copies in the per-command area, which is from the 'file' pointer to when 'opcode' starts. We do have existing user copies there, but they are not all annotated like the one that openat2_prep() uses, copy_struct_from_user(). But in practice opcodes should be allowed to copy data into their per-command area in the io_kiocb. Reported-by: Breno Leitao <leitao@debian.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-08-03 04:38:01 +08:00
/*
* Allow user copy in the per-command field, which starts after the
* file in io_kiocb and until the opcode field. The openat2 handling
* requires copying in user memory into the io_kiocb object in that
* range, and HARDENED_USERCOPY will complain if we haven't
* correctly annotated this range.
*/
req_cachep = kmem_cache_create_usercopy("io_kiocb",
sizeof(struct io_kiocb), 0,
SLAB_HWCACHE_ALIGN | SLAB_PANIC |
SLAB_ACCOUNT | SLAB_TYPESAFE_BY_RCU,
offsetof(struct io_kiocb, cmd.data),
sizeof_field(struct io_kiocb, cmd.data), NULL);
#ifdef CONFIG_SYSCTL
register_sysctl_init("kernel", kernel_io_uring_disabled_table);
#endif
Add io_uring IO interface The submission queue (SQ) and completion queue (CQ) rings are shared between the application and the kernel. This eliminates the need to copy data back and forth to submit and complete IO. IO submissions use the io_uring_sqe data structure, and completions are generated in the form of io_uring_cqe data structures. The SQ ring is an index into the io_uring_sqe array, which makes it possible to submit a batch of IOs without them being contiguous in the ring. The CQ ring is always contiguous, as completion events are inherently unordered, and hence any io_uring_cqe entry can point back to an arbitrary submission. Two new system calls are added for this: io_uring_setup(entries, params) Sets up an io_uring instance for doing async IO. On success, returns a file descriptor that the application can mmap to gain access to the SQ ring, CQ ring, and io_uring_sqes. io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) Initiates IO against the rings mapped to this fd, or waits for them to complete, or both. The behavior is controlled by the parameters passed in. If 'to_submit' is non-zero, then we'll try and submit new IO. If IORING_ENTER_GETEVENTS is set, the kernel will wait for 'min_complete' events, if they aren't already available. It's valid to set IORING_ENTER_GETEVENTS and 'min_complete' == 0 at the same time, this allows the kernel to return already completed events without waiting for them. This is useful only for polling, as for IRQ driven IO, the application can just check the CQ ring without entering the kernel. With this setup, it's possible to do async IO with a single system call. Future developments will enable polled IO with this interface, and polled submission as well. The latter will enable an application to do IO without doing ANY system calls at all. For IRQ driven IO, an application only needs to enter the kernel for completions if it wants to wait for them to occur. Each io_uring is backed by a workqueue, to support buffered async IO as well. We will only punt to an async context if the command would need to wait for IO on the device side. Any data that can be accessed directly in the page cache is done inline. This avoids the slowness issue of usual threadpools, since cached data is accessed as quickly as a sync interface. Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-01-08 01:46:33 +08:00
return 0;
};
__initcall(io_uring_init);