nfsd_splice_actor() has a clause in its loop that chops up a compound page
into individual pages such that if the same page is seen twice in a row, it
is discarded the second time. This is a problem with the advent of
shmem_splice_read() as that inserts zero_pages into the pipe in lieu of
pages that aren't present in the pagecache.
Fix this by assuming that the last page is being extended only if the
currently stored length + starting offset is not currently on a page
boundary.
This can be tested by NFS-exporting a tmpfs filesystem on the test machine
and truncating it to more than a page in size (eg. truncate -s 8192) and
then reading it by NFS. The first page will be all zeros, but thereafter
garbage will be read.
Note: I wonder if we can ever get a situation now where we get a splice
that gives us contiguous parts of a page in separate actor calls. As NFSD
can only be splicing from a file (I think), there are only three sources of
the page: copy_splice_read(), shmem_splice_read() and file_splice_read().
The first allocates pages for the data it reads, so the problem cannot
occur; the second should never see a partial page; and the third waits for
each page to become available before we're allowed to read from it.
Fixes: bd194b1871 ("shmem: Implement splice-read")
Reported-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: NeilBrown <neilb@suse.de>
cc: Hugh Dickins <hughd@google.com>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
cc: linux-nfs@vger.kernel.org
cc: linux-fsdevel@vger.kernel.org
cc: linux-mm@kvack.org
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
If the client is calling TEST_STATEID, then it is because some event
occurred that requires it to check all the stateids for validity and
call FREE_STATEID on the ones that have been revoked. In this case,
either the stateid exists in the list of stateids associated with that
nfs4_client, in which case it should be tested, or it does not. There
are no additional conditions to be considered.
Reported-by: "Frank Ch. Eigler" <fche@redhat.com>
Fixes: 7df302f75e ("NFSD: TEST_STATEID should not return NFS4ERR_STALE_STATEID")
Cc: stable@vger.kernel.org # v5.7+
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
In nfsd4_encode_fattr(), TIME_CREATE was being written out after all
other times. However, they should be written out in an order that
matches the bit flags in bmval1, which in this case are
#define FATTR4_WORD1_TIME_ACCESS (1UL << 15)
#define FATTR4_WORD1_TIME_CREATE (1UL << 18)
#define FATTR4_WORD1_TIME_DELTA (1UL << 19)
#define FATTR4_WORD1_TIME_METADATA (1UL << 20)
#define FATTR4_WORD1_TIME_MODIFY (1UL << 21)
so TIME_CREATE should come second.
I noticed this on a FreeBSD NFSv4.2 client, which supports creation
times. On this client, file times were weirdly permuted. With this
patch applied on the server, times looked normal on the client.
Fixes: e377a3e698 ("nfsd: Add support for the birth time attribute")
Link: https://unix.stackexchange.com/q/749605/56202
Signed-off-by: Tavian Barnes <tavianator@tavianator.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
There are a few assignments to variable len where the value is not
being read and so the assignments are redundant and can be removed.
In one case, the variable len can be removed completely. Cleans up
4 clang scan warnings of the form:
fs/nfsd/export.c💯7: warning: Although the value stored to 'len'
is used in the enclosing expression, the value is never actually
read from 'len' [deadcode.DeadStores]
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Commit 7d81ee8722 ("svcrdma: Single-stage RDMA Read") changed the
behavior of svc_rdma_recvfrom() but neglected to update the
documenting comment.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
I find the naming of nfsd_init_net() and nfsd_startup_net() to be
confusingly similar. Rename the namespace initialization and tear-
down ops and add comments to distinguish their separate purposes.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Commit f5f9d4a314 ("nfsd: move reply cache initialization into nfsd
startup") moved the initialization of the reply cache into nfsd startup,
but didn't account for the stats counters, which can be accessed before
nfsd is ever started. The result can be a NULL pointer dereference when
someone accesses /proc/fs/nfsd/reply_cache_stats while nfsd is still
shut down.
This is a regression and a user-triggerable oops in the right situation:
- non-x86_64 arch
- /proc/fs/nfsd is mounted in the namespace
- nfsd is not started in the namespace
- unprivileged user calls "cat /proc/fs/nfsd/reply_cache_stats"
Although this is easy to trigger on some arches (like aarch64), on
x86_64, calling this_cpu_ptr(NULL) evidently returns a pointer to the
fixed_percpu_data. That struct looks just enough like a newly
initialized percpu var to allow nfsd_reply_cache_stats_show to access
it without Oopsing.
Move the initialization of the per-net+per-cpu reply-cache counters
back into nfsd_init_net, while leaving the rest of the reply cache
allocations to be done at nfsd startup time.
Kudos to Eirik who did most of the legwork to track this down.
Cc: stable@vger.kernel.org # v6.3+
Fixes: f5f9d4a314 ("nfsd: move reply cache initialization into nfsd startup")
Reported-and-tested-by: Eirik Fuller <efuller@redhat.com>
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=2215429
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Remove a couple of dprintk call sites that are of little value.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The preceding block comment before svc_register_xprt_class() is
not related to that function.
While we're here, add proper documenting comments for these two
publicly-visible functions.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This event brackets the svcrdma_post_* trace points. If this trace
event is enabled but does not appear as expected, that indicates a
chunk_ctxt leak.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Try to catch incorrect calling contexts mechanically rather than by
code review.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Clean up: de-duplicate some common code.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Micro-optimization: Call ktime_get() only when ->xpo_recvfrom() has
given us a full RPC message to process. rq_stime isn't used
otherwise, so this avoids pointless work.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Now that we have bulk page allocation and release APIs, it's more
efficient to use those than it is for nfsd threads to wait for send
completions. Previous patches have eliminated the calls to
wait_for_completion() and complete(), in order to avoid scheduler
overhead.
Now release pages-under-I/O in the send completion handler using
the efficient bulk release API.
I've measured a 7% reduction in cumulative CPU utilization in
svc_rdma_sendto(), svc_rdma_wc_send(), and svc_xprt_release(). In
particular, using release_pages() instead of complete() cuts the
time per svc_rdma_wc_send() call by two-thirds. This helps improve
scalability because svc_rdma_wc_send() is single-threaded per
connection.
Reviewed-by: Tom Talpey <tom@talpey.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
I noticed that svc_rqst_release_pages() was still unnecessarily
releasing a page when svc_rdma_recvfrom() returns zero.
Fixes: a53d5cb064 ("svcrdma: Avoid releasing a page in svc_xprt_release()")
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Get rid of the completion wait in svc_rdma_sendto(), and release
pages in the send completion handler again. A subsequent patch will
handle releasing those pages more efficiently.
Reverted by hand: patch -R would not apply cleanly.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Pre-requisite for releasing pages in the send completion handler.
Reverted by hand: patch -R would not apply cleanly.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Pre-requisite for releasing pages in the send completion handler.
Reverted by hand: patch -R would not apply cleanly.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Modified nfsd4_encode_open to encode the op_recall flag properly
for OPEN result with write delegation granted.
Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Cc: stable@vger.kernel.org
At LFS 2023, it was suggested we should publicly document the name and
email of reviewers who new contributors can trust. This also gives them
some recognition for their work as reviewers.
Acked-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Ensure that Bruce's old e-mail addresses map to his current one so
he doesn't miss out on all the fun.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The physical device's favored NUMA node ID is available when
allocating a rw_ctxt. Use that value instead of relying on the
assumption that the memory allocation happens to be running on a
node close to the device.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The physical device's favored NUMA node ID is available when
allocating a send_ctxt. Use that value instead of relying on the
assumption that the memory allocation happens to be running on a
node close to the device.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The physical device's favored NUMA node ID is available when
allocating a recv_ctxt. Use that value instead of relying on the
assumption that the memory allocation happens to be running on a
node close to the device.
This clean up eliminates the hack of destroying recv_ctxts that
were not created by the receive CQ thread -- recv_ctxts are now
always allocated on a "good" node.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The physical device's NUMA node ID is available when allocating an
svc_xprt for an incoming connection. Use that value to ensure the
svc_xprt structure is allocated on the NUMA node closest to the
device.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The below-mentioned patch was intended to simplify refcounting on the
svc_serv used by locked. The goal was to only ever have a single
reference from the single thread. To that end we dropped a call to
lockd_start_svc() (except when creating thread) which would take a
reference, and dropped the svc_put(serv) that would drop that reference.
Unfortunately we didn't also remove the svc_get() from
lockd_create_svc() in the case where the svc_serv already existed.
So after the patch:
- on the first call the svc_serv was allocated and the one reference
was given to the thread, so there are no extra references
- on subsequent calls svc_get() was called so there is now an extra
reference.
This is clearly not consistent.
The inconsistency is also clear in the current code in lockd_get()
takes *two* references, one on nlmsvc_serv and one by incrementing
nlmsvc_users. This clearly does not match lockd_put().
So: drop that svc_get() from lockd_get() (which used to be in
lockd_create_svc().
Reported-by: Ido Schimmel <idosch@idosch.org>
Closes: https://lore.kernel.org/linux-nfs/ZHsI%2FH16VX9kJQX1@shredder/T/#u
Fixes: b73a297204 ("lockd: move lockd_start_svc() call into lockd_create_svc()")
Signed-off-by: NeilBrown <neilb@suse.de>
Tested-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
nfsd calls fh_getattr to get the latest inode attrs for pre/post-op
info. In the event that fh_getattr fails, it resorts to scraping cached
values out of the inode directly.
Since these attributes are optional, we can just skip providing them
altogether when this happens.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Neil Brown <neilb@suse.de>
Now that the preparation of an rq_vec has been removed from the
generic read path, nfsd_splice_read() no longer needs to reset
rq_next_page.
nfsd4_encode_read() calls nfsd_splice_read() directly. As far as I
can ascertain, resetting rq_next_page for NFSv4 splice reads is
unnecessary because rq_next_page is already set correctly.
Moreover, resetting it might even be incorrect if previous
operations in the COMPOUND have already consumed at least a page of
the send buffer. I would expect that the result would be encoding
the READ payload over previously-encoded results.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Accrue the following benefits:
a) Deduplicate this common bit of code.
b) Don't prepare rq_vec for NFSv2 and NFSv3 spliced reads, which
don't use rq_vec. This is already the case for
nfsd4_encode_read().
c) Eventually, converting NFSD's read path to use a bvec iterator
will be simpler.
In the next patch, nfsd_iter_read() will replace nfsd_readv() for
all NFS versions.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
A GETATTR with a large result can advance xdr->page_ptr without
updating rq_next_page. If a splice READ follows that GETATTR in the
COMPOUND, nfsd_splice_actor can start splicing at the wrong page.
I've also seen READLINK and READDIR leave rq_next_page in an
unmodified state.
There are potentially a myriad of combinations like this, so play it
safe: move the rq_next_page update to nfsd4_encode_operation.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Commit 15b23ef5d3 ("nfsd4: fix corruption of NFSv4 read data")
encountered exactly the same issue: after a splice read, a
filesystem-owned page is left in rq_pages[]; the symptoms are the
same as described there.
If the computed number of pages in nfsd4_encode_splice_read() is not
exactly the same as the actual number of pages that were consumed by
nfsd_splice_actor() (say, because of a bug) then hilarity ensues.
Instead of recomputing the page offset based on the size of the
payload, use rq_next_page, which is already properly updated by
nfsd_splice_actor(), to cause svc_rqst_release_pages() to operate
correctly in every instance.
This is a defensive change since we believe that after commit
27c934dd88 ("nfsd: don't replace page in rq_pages if it's a
continuation of last page") has been applied, there are no known
opportunities for nfsd_splice_actor() to screw up. So I'm not
marking it for stable backport.
Reported-by: Andy Zlotek <andy.zlotek@oracle.com>
Suggested-by: Calum Mackay <calum.mackay@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
svc_init_buffer() is careful to allocate the initial set of server
thread buffer pages from memory on the local NUMA node.
svc_alloc_arg() should also be that careful.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Add trace log eye-catchers that record the arguments used to
configure NFSD. This helps when troubleshooting the NFSD
administrative interfaces.
These tracepoints can capture NFSD start-up and shutdown times and
parameters, changes in lease time and thread count, and a request
to end the namespace's NFSv4 grace period, in addition to the set
of NFS versions that are enabled.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
For easier readability, follow the common convention:
if (error)
handle_error;
continue_normally;
No behavior change is expected.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Capture a timestamp and pointer address during the creation and
destruction of struct svc_sock to record its lifetime. This helps
to diagnose transport reference counting issues.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The -ENOMEM arm could fire repeatedly if the system runs low on
memory, so remove it.
Don't bother to trace -EAGAIN error events, since those fire after
a listener is created (with no work done) and once again after an
accept has been handled successfully (again, with no work done).
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When enabled, this dprintk() fires for every incoming RPC, which is
an enormous amount of log traffic. These days, after the first few
hundred log messages, the system journald is just going to mute it,
along with all other NFSD debug output.
Let's rely on trace points for this high-traffic information
instead.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The correct function name is svc_tcp_listen_data_ready().
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
After the listener svc_sock is freed, and before invoking svc_tcp_accept()
for the established child sock, there is a window that the newsock
retaining a freed listener svc_sock in sk_user_data which cloning from
parent. In the race window, if data is received on the newsock, we will
observe use-after-free report in svc_tcp_listen_data_ready().
Reproduce by two tasks:
1. while :; do rpc.nfsd 0 ; rpc.nfsd; done
2. while :; do echo "" | ncat -4 127.0.0.1 2049 ; done
KASAN report:
==================================================================
BUG: KASAN: slab-use-after-free in svc_tcp_listen_data_ready+0x1cf/0x1f0 [sunrpc]
Read of size 8 at addr ffff888139d96228 by task nc/102553
CPU: 7 PID: 102553 Comm: nc Not tainted 6.3.0+ #18
Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 11/12/2020
Call Trace:
<IRQ>
dump_stack_lvl+0x33/0x50
print_address_description.constprop.0+0x27/0x310
print_report+0x3e/0x70
kasan_report+0xae/0xe0
svc_tcp_listen_data_ready+0x1cf/0x1f0 [sunrpc]
tcp_data_queue+0x9f4/0x20e0
tcp_rcv_established+0x666/0x1f60
tcp_v4_do_rcv+0x51c/0x850
tcp_v4_rcv+0x23fc/0x2e80
ip_protocol_deliver_rcu+0x62/0x300
ip_local_deliver_finish+0x267/0x350
ip_local_deliver+0x18b/0x2d0
ip_rcv+0x2fb/0x370
__netif_receive_skb_one_core+0x166/0x1b0
process_backlog+0x24c/0x5e0
__napi_poll+0xa2/0x500
net_rx_action+0x854/0xc90
__do_softirq+0x1bb/0x5de
do_softirq+0xcb/0x100
</IRQ>
<TASK>
...
</TASK>
Allocated by task 102371:
kasan_save_stack+0x1e/0x40
kasan_set_track+0x21/0x30
__kasan_kmalloc+0x7b/0x90
svc_setup_socket+0x52/0x4f0 [sunrpc]
svc_addsock+0x20d/0x400 [sunrpc]
__write_ports_addfd+0x209/0x390 [nfsd]
write_ports+0x239/0x2c0 [nfsd]
nfsctl_transaction_write+0xac/0x110 [nfsd]
vfs_write+0x1c3/0xae0
ksys_write+0xed/0x1c0
do_syscall_64+0x38/0x90
entry_SYSCALL_64_after_hwframe+0x72/0xdc
Freed by task 102551:
kasan_save_stack+0x1e/0x40
kasan_set_track+0x21/0x30
kasan_save_free_info+0x2a/0x50
__kasan_slab_free+0x106/0x190
__kmem_cache_free+0x133/0x270
svc_xprt_free+0x1e2/0x350 [sunrpc]
svc_xprt_destroy_all+0x25a/0x440 [sunrpc]
nfsd_put+0x125/0x240 [nfsd]
nfsd_svc+0x2cb/0x3c0 [nfsd]
write_threads+0x1ac/0x2a0 [nfsd]
nfsctl_transaction_write+0xac/0x110 [nfsd]
vfs_write+0x1c3/0xae0
ksys_write+0xed/0x1c0
do_syscall_64+0x38/0x90
entry_SYSCALL_64_after_hwframe+0x72/0xdc
Fix the UAF by simply doing nothing in svc_tcp_listen_data_ready()
if state != TCP_LISTEN, that will avoid dereferencing svsk for all
child socket.
Link: https://lore.kernel.org/lkml/20230507091131.23540-1-dinghui@sangfor.com.cn/
Fixes: fa9251afc3 ("SUNRPC: Call the default socket callbacks instead of open coding")
Signed-off-by: Ding Hui <dinghui@sangfor.com.cn>
Cc: <stable@vger.kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
We've aligned setgid behavior over multiple kernel releases. The details
can be found in commit cf619f8919 ("Merge tag 'fs.ovl.setgid.v6.2' of
git://git.kernel.org/pub/scm/linux/kernel/git/vfs/idmapping") and
commit 426b4ca2d6 ("Merge tag 'fs.setgid.v6.0' of
git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux").
Consistent setgid stripping behavior is now encapsulated in the
setattr_should_drop_sgid() helper which is used by all filesystems that
strip setgid bits outside of vfs proper. Usually ATTR_KILL_SGID is
raised in e.g., chown_common() and is subject to the
setattr_should_drop_sgid() check to determine whether the setgid bit can
be retained. Since nfsd is raising ATTR_KILL_SGID unconditionally it
will cause notify_change() to strip it even if the caller had the
necessary privileges to retain it. Ensure that nfsd only raises
ATR_KILL_SGID if the caller lacks the necessary privileges to retain the
setgid bit.
Without this patch the setgid stripping tests in LTP will fail:
> As you can see, the problem is S_ISGID (0002000) was dropped on a
> non-group-executable file while chown was invoked by super-user, while
[...]
> fchown02.c:66: TFAIL: testfile2: wrong mode permissions 0100700, expected 0102700
[...]
> chown02.c:57: TFAIL: testfile2: wrong mode permissions 0100700, expected 0102700
With this patch all tests pass.
Reported-by: Sherry Yang <sherry.yang@oracle.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
