-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCYCegywAKCRCRxhvAZXjc
ouJ6AQDlf+7jCQlQdeKKoN9QDFfMzG1ooemat36EpRRTONaGuAD8D9A4sUsG4+5f
4IU5Lj9oY4DEmF8HenbWK2ZHsesL2Qg=
=yPaw
-----END PGP SIGNATURE-----
Merge tag 'idmapped-mounts-v5.12' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux
Pull idmapped mounts from Christian Brauner:
"This introduces idmapped mounts which has been in the making for some
time. Simply put, different mounts can expose the same file or
directory with different ownership. This initial implementation comes
with ports for fat, ext4 and with Christoph's port for xfs with more
filesystems being actively worked on by independent people and
maintainers.
Idmapping mounts handle a wide range of long standing use-cases. Here
are just a few:
- Idmapped mounts make it possible to easily share files between
multiple users or multiple machines especially in complex
scenarios. For example, idmapped mounts will be used in the
implementation of portable home directories in
systemd-homed.service(8) where they allow users to move their home
directory to an external storage device and use it on multiple
computers where they are assigned different uids and gids. This
effectively makes it possible to assign random uids and gids at
login time.
- It is possible to share files from the host with unprivileged
containers without having to change ownership permanently through
chown(2).
- It is possible to idmap a container's rootfs and without having to
mangle every file. For example, Chromebooks use it to share the
user's Download folder with their unprivileged containers in their
Linux subsystem.
- It is possible to share files between containers with
non-overlapping idmappings.
- Filesystem that lack a proper concept of ownership such as fat can
use idmapped mounts to implement discretionary access (DAC)
permission checking.
- They allow users to efficiently changing ownership on a per-mount
basis without having to (recursively) chown(2) all files. In
contrast to chown (2) changing ownership of large sets of files is
instantenous with idmapped mounts. This is especially useful when
ownership of a whole root filesystem of a virtual machine or
container is changed. With idmapped mounts a single syscall
mount_setattr syscall will be sufficient to change the ownership of
all files.
- Idmapped mounts always take the current ownership into account as
idmappings specify what a given uid or gid is supposed to be mapped
to. This contrasts with the chown(2) syscall which cannot by itself
take the current ownership of the files it changes into account. It
simply changes the ownership to the specified uid and gid. This is
especially problematic when recursively chown(2)ing a large set of
files which is commong with the aforementioned portable home
directory and container and vm scenario.
- Idmapped mounts allow to change ownership locally, restricting it
to specific mounts, and temporarily as the ownership changes only
apply as long as the mount exists.
Several userspace projects have either already put up patches and
pull-requests for this feature or will do so should you decide to pull
this:
- systemd: In a wide variety of scenarios but especially right away
in their implementation of portable home directories.
https://systemd.io/HOME_DIRECTORY/
- container runtimes: containerd, runC, LXD:To share data between
host and unprivileged containers, unprivileged and privileged
containers, etc. The pull request for idmapped mounts support in
containerd, the default Kubernetes runtime is already up for quite
a while now: https://github.com/containerd/containerd/pull/4734
- The virtio-fs developers and several users have expressed interest
in using this feature with virtual machines once virtio-fs is
ported.
- ChromeOS: Sharing host-directories with unprivileged containers.
I've tightly synced with all those projects and all of those listed
here have also expressed their need/desire for this feature on the
mailing list. For more info on how people use this there's a bunch of
talks about this too. Here's just two recent ones:
https://www.cncf.io/wp-content/uploads/2020/12/Rootless-Containers-in-Gitpod.pdfhttps://fosdem.org/2021/schedule/event/containers_idmap/
This comes with an extensive xfstests suite covering both ext4 and
xfs:
https://git.kernel.org/brauner/xfstests-dev/h/idmapped_mounts
It covers truncation, creation, opening, xattrs, vfscaps, setid
execution, setgid inheritance and more both with idmapped and
non-idmapped mounts. It already helped to discover an unrelated xfs
setgid inheritance bug which has since been fixed in mainline. It will
be sent for inclusion with the xfstests project should you decide to
merge this.
In order to support per-mount idmappings vfsmounts are marked with
user namespaces. The idmapping of the user namespace will be used to
map the ids of vfs objects when they are accessed through that mount.
By default all vfsmounts are marked with the initial user namespace.
The initial user namespace is used to indicate that a mount is not
idmapped. All operations behave as before and this is verified in the
testsuite.
Based on prior discussions we want to attach the whole user namespace
and not just a dedicated idmapping struct. This allows us to reuse all
the helpers that already exist for dealing with idmappings instead of
introducing a whole new range of helpers. In addition, if we decide in
the future that we are confident enough to enable unprivileged users
to setup idmapped mounts the permission checking can take into account
whether the caller is privileged in the user namespace the mount is
currently marked with.
The user namespace the mount will be marked with can be specified by
passing a file descriptor refering to the user namespace as an
argument to the new mount_setattr() syscall together with the new
MOUNT_ATTR_IDMAP flag. The system call follows the openat2() pattern
of extensibility.
The following conditions must be met in order to create an idmapped
mount:
- The caller must currently have the CAP_SYS_ADMIN capability in the
user namespace the underlying filesystem has been mounted in.
- The underlying filesystem must support idmapped mounts.
- The mount must not already be idmapped. This also implies that the
idmapping of a mount cannot be altered once it has been idmapped.
- The mount must be a detached/anonymous mount, i.e. it must have
been created by calling open_tree() with the OPEN_TREE_CLONE flag
and it must not already have been visible in the filesystem.
The last two points guarantee easier semantics for userspace and the
kernel and make the implementation significantly simpler.
By default vfsmounts are marked with the initial user namespace and no
behavioral or performance changes are observed.
The manpage with a detailed description can be found here:
1d7b902e28
In order to support idmapped mounts, filesystems need to be changed
and mark themselves with the FS_ALLOW_IDMAP flag in fs_flags. The
patches to convert individual filesystem are not very large or
complicated overall as can be seen from the included fat, ext4, and
xfs ports. Patches for other filesystems are actively worked on and
will be sent out separately. The xfstestsuite can be used to verify
that port has been done correctly.
The mount_setattr() syscall is motivated independent of the idmapped
mounts patches and it's been around since July 2019. One of the most
valuable features of the new mount api is the ability to perform
mounts based on file descriptors only.
Together with the lookup restrictions available in the openat2()
RESOLVE_* flag namespace which we added in v5.6 this is the first time
we are close to hardened and race-free (e.g. symlinks) mounting and
path resolution.
While userspace has started porting to the new mount api to mount
proper filesystems and create new bind-mounts it is currently not
possible to change mount options of an already existing bind mount in
the new mount api since the mount_setattr() syscall is missing.
With the addition of the mount_setattr() syscall we remove this last
restriction and userspace can now fully port to the new mount api,
covering every use-case the old mount api could. We also add the
crucial ability to recursively change mount options for a whole mount
tree, both removing and adding mount options at the same time. This
syscall has been requested multiple times by various people and
projects.
There is a simple tool available at
https://github.com/brauner/mount-idmapped
that allows to create idmapped mounts so people can play with this
patch series. I'll add support for the regular mount binary should you
decide to pull this in the following weeks:
Here's an example to a simple idmapped mount of another user's home
directory:
u1001@f2-vm:/$ sudo ./mount --idmap both:1000:1001:1 /home/ubuntu/ /mnt
u1001@f2-vm:/$ ls -al /home/ubuntu/
total 28
drwxr-xr-x 2 ubuntu ubuntu 4096 Oct 28 22:07 .
drwxr-xr-x 4 root root 4096 Oct 28 04:00 ..
-rw------- 1 ubuntu ubuntu 3154 Oct 28 22:12 .bash_history
-rw-r--r-- 1 ubuntu ubuntu 220 Feb 25 2020 .bash_logout
-rw-r--r-- 1 ubuntu ubuntu 3771 Feb 25 2020 .bashrc
-rw-r--r-- 1 ubuntu ubuntu 807 Feb 25 2020 .profile
-rw-r--r-- 1 ubuntu ubuntu 0 Oct 16 16:11 .sudo_as_admin_successful
-rw------- 1 ubuntu ubuntu 1144 Oct 28 00:43 .viminfo
u1001@f2-vm:/$ ls -al /mnt/
total 28
drwxr-xr-x 2 u1001 u1001 4096 Oct 28 22:07 .
drwxr-xr-x 29 root root 4096 Oct 28 22:01 ..
-rw------- 1 u1001 u1001 3154 Oct 28 22:12 .bash_history
-rw-r--r-- 1 u1001 u1001 220 Feb 25 2020 .bash_logout
-rw-r--r-- 1 u1001 u1001 3771 Feb 25 2020 .bashrc
-rw-r--r-- 1 u1001 u1001 807 Feb 25 2020 .profile
-rw-r--r-- 1 u1001 u1001 0 Oct 16 16:11 .sudo_as_admin_successful
-rw------- 1 u1001 u1001 1144 Oct 28 00:43 .viminfo
u1001@f2-vm:/$ touch /mnt/my-file
u1001@f2-vm:/$ setfacl -m u:1001:rwx /mnt/my-file
u1001@f2-vm:/$ sudo setcap -n 1001 cap_net_raw+ep /mnt/my-file
u1001@f2-vm:/$ ls -al /mnt/my-file
-rw-rwxr--+ 1 u1001 u1001 0 Oct 28 22:14 /mnt/my-file
u1001@f2-vm:/$ ls -al /home/ubuntu/my-file
-rw-rwxr--+ 1 ubuntu ubuntu 0 Oct 28 22:14 /home/ubuntu/my-file
u1001@f2-vm:/$ getfacl /mnt/my-file
getfacl: Removing leading '/' from absolute path names
# file: mnt/my-file
# owner: u1001
# group: u1001
user::rw-
user:u1001:rwx
group::rw-
mask::rwx
other::r--
u1001@f2-vm:/$ getfacl /home/ubuntu/my-file
getfacl: Removing leading '/' from absolute path names
# file: home/ubuntu/my-file
# owner: ubuntu
# group: ubuntu
user::rw-
user:ubuntu:rwx
group::rw-
mask::rwx
other::r--"
* tag 'idmapped-mounts-v5.12' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux: (41 commits)
xfs: remove the possibly unused mp variable in xfs_file_compat_ioctl
xfs: support idmapped mounts
ext4: support idmapped mounts
fat: handle idmapped mounts
tests: add mount_setattr() selftests
fs: introduce MOUNT_ATTR_IDMAP
fs: add mount_setattr()
fs: add attr_flags_to_mnt_flags helper
fs: split out functions to hold writers
namespace: only take read lock in do_reconfigure_mnt()
mount: make {lock,unlock}_mount_hash() static
namespace: take lock_mount_hash() directly when changing flags
nfs: do not export idmapped mounts
overlayfs: do not mount on top of idmapped mounts
ecryptfs: do not mount on top of idmapped mounts
ima: handle idmapped mounts
apparmor: handle idmapped mounts
fs: make helpers idmap mount aware
exec: handle idmapped mounts
would_dump: handle idmapped mounts
...
For core file operations such as changing directories or chrooting,
determining file access, changing mode or ownership the vfs will verify
that the caller is privileged over the inode. Extend the various helpers
to handle idmapped mounts. If the inode is accessed through an idmapped
mount map it into the mount's user namespace. Afterwards the permissions
checks are identical to non-idmapped mounts. When changing file
ownership we need to map the uid and gid from the mount's user
namespace. If the initial user namespace is passed nothing changes so
non-idmapped mounts will see identical behavior as before.
Link: https://lore.kernel.org/r/20210121131959.646623-17-christian.brauner@ubuntu.com
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Howells <dhowells@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Reviewed-by: James Morris <jamorris@linux.microsoft.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
When truncating files the vfs will verify that the caller is privileged
over the inode. Extend it to handle idmapped mounts. If the inode is
accessed through an idmapped mount it is mapped according to the mount's
user namespace. Afterwards the permissions checks are identical to
non-idmapped mounts. If the initial user namespace is passed nothing
changes so non-idmapped mounts will see identical behavior as before.
Link: https://lore.kernel.org/r/20210121131959.646623-16-christian.brauner@ubuntu.com
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Howells <dhowells@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
When file attributes are changed most filesystems rely on the
setattr_prepare(), setattr_copy(), and notify_change() helpers for
initialization and permission checking. Let them handle idmapped mounts.
If the inode is accessed through an idmapped mount map it into the
mount's user namespace. Afterwards the checks are identical to
non-idmapped mounts. If the initial user namespace is passed nothing
changes so non-idmapped mounts will see identical behavior as before.
Helpers that perform checks on the ia_uid and ia_gid fields in struct
iattr assume that ia_uid and ia_gid are intended values and have already
been mapped correctly at the userspace-kernelspace boundary as we
already do today. If the initial user namespace is passed nothing
changes so non-idmapped mounts will see identical behavior as before.
Link: https://lore.kernel.org/r/20210121131959.646623-8-christian.brauner@ubuntu.com
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Howells <dhowells@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
The two helpers inode_permission() and generic_permission() are used by
the vfs to perform basic permission checking by verifying that the
caller is privileged over an inode. In order to handle idmapped mounts
we extend the two helpers with an additional user namespace argument.
On idmapped mounts the two helpers will make sure to map the inode
according to the mount's user namespace and then peform identical
permission checks to inode_permission() and generic_permission(). If the
initial user namespace is passed nothing changes so non-idmapped mounts
will see identical behavior as before.
Link: https://lore.kernel.org/r/20210121131959.646623-6-christian.brauner@ubuntu.com
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Howells <dhowells@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: James Morris <jamorris@linux.microsoft.com>
Acked-by: Serge Hallyn <serge@hallyn.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Add two simple helpers to check permissions on a file and path
respectively and convert over some callers. It simplifies quite a few
codepaths and also reduces the churn in later patches quite a bit.
Christoph also correctly points out that this makes codepaths (e.g.
ioctls) way easier to follow that would otherwise have to do more
complex argument passing than necessary.
Link: https://lore.kernel.org/r/20210121131959.646623-4-christian.brauner@ubuntu.com
Cc: David Howells <dhowells@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Suggested-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: James Morris <jamorris@linux.microsoft.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Now that we support non-blocking path resolution internally, expose it
via openat2() in the struct open_how ->resolve flags. This allows
applications using openat2() to limit path resolution to the extent that
it is already cached.
If the lookup cannot be satisfied in a non-blocking manner, openat2(2)
will return -1/-EAGAIN.
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Pull execve updates from Eric Biederman:
"This set of changes ultimately fixes the interaction of posix file
lock and exec. Fundamentally most of the change is just moving where
unshare_files is called during exec, and tweaking the users of
files_struct so that the count of files_struct is not unnecessarily
played with.
Along the way fcheck and related helpers were renamed to more
accurately reflect what they do.
There were also many other small changes that fell out, as this is the
first time in a long time much of this code has been touched.
Benchmarks haven't turned up any practical issues but Al Viro has
observed a possibility for a lot of pounding on task_lock. So I have
some changes in progress to convert put_files_struct to always rcu
free files_struct. That wasn't ready for the merge window so that will
have to wait until next time"
* 'exec-for-v5.11' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace: (27 commits)
exec: Move io_uring_task_cancel after the point of no return
coredump: Document coredump code exclusively used by cell spufs
file: Remove get_files_struct
file: Rename __close_fd_get_file close_fd_get_file
file: Replace ksys_close with close_fd
file: Rename __close_fd to close_fd and remove the files parameter
file: Merge __alloc_fd into alloc_fd
file: In f_dupfd read RLIMIT_NOFILE once.
file: Merge __fd_install into fd_install
proc/fd: In fdinfo seq_show don't use get_files_struct
bpf/task_iter: In task_file_seq_get_next use task_lookup_next_fd_rcu
proc/fd: In proc_readfd_common use task_lookup_next_fd_rcu
file: Implement task_lookup_next_fd_rcu
kcmp: In get_file_raw_ptr use task_lookup_fd_rcu
proc/fd: In tid_fd_mode use task_lookup_fd_rcu
file: Implement task_lookup_fd_rcu
file: Rename fcheck lookup_fd_rcu
file: Replace fcheck_files with files_lookup_fd_rcu
file: Factor files_lookup_fd_locked out of fcheck_files
file: Rename __fcheck_files to files_lookup_fd_raw
...
The function __close_fd was added to support binder[1]. Now that
binder has been fixed to no longer need __close_fd[2] all calls
to __close_fd pass current->files.
Therefore transform the files parameter into a local variable
initialized to current->files, and rename __close_fd to close_fd to
reflect this change, and keep it in sync with the similar changes to
__alloc_fd, and __fd_install.
This removes the need for callers to care about the extra care that
needs to be take if anything except current->files is passed, by
limiting the callers to only operation on current->files.
[1] 483ce1d4b8 ("take descriptor-related part of close() to file.c")
[2] 44d8047f1d ("binder: use standard functions to allocate fds")
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
v1: https://lkml.kernel.org/r/20200817220425.9389-17-ebiederm@xmission.com
Link: https://lkml.kernel.org/r/20201120231441.29911-21-ebiederm@xmission.com
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
This was an oversight in the original implementation, as it makes no
sense to specify both scoping flags to the same openat2(2) invocation
(before this patch, the result of such an invocation was equivalent to
RESOLVE_IN_ROOT being ignored).
This is a userspace-visible ABI change, but the only user of openat2(2)
at the moment is LXC which doesn't specify both flags and so no
userspace programs will break as a result.
Fixes: fddb5d430a ("open: introduce openat2(2) syscall")
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Cc: <stable@vger.kernel.org> # v5.6+
Link: https://lore.kernel.org/r/20201027235044.5240-2-cyphar@cyphar.com
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
The execve(2)/uselib(2) syscalls have always rejected non-regular files.
Recently, it was noticed that a deadlock was introduced when trying to
execute pipes, as the S_ISREG() test was happening too late. This was
fixed in commit 73601ea5b7 ("fs/open.c: allow opening only regular files
during execve()"), but it was added after inode_permission() had already
run, which meant LSMs could see bogus attempts to execute non-regular
files.
Move the test into the other inode type checks (which already look for
other pathological conditions[1]). Since there is no need to use
FMODE_EXEC while we still have access to "acc_mode", also switch the test
to MAY_EXEC.
Also include a comment with the redundant S_ISREG() checks at the end of
execve(2)/uselib(2) to note that they are present to avoid any mistakes.
My notes on the call path, and related arguments, checks, etc:
do_open_execat()
struct open_flags open_exec_flags = {
.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
.acc_mode = MAY_EXEC,
...
do_filp_open(dfd, filename, open_flags)
path_openat(nameidata, open_flags, flags)
file = alloc_empty_file(open_flags, current_cred());
do_open(nameidata, file, open_flags)
may_open(path, acc_mode, open_flag)
/* new location of MAY_EXEC vs S_ISREG() test */
inode_permission(inode, MAY_OPEN | acc_mode)
security_inode_permission(inode, acc_mode)
vfs_open(path, file)
do_dentry_open(file, path->dentry->d_inode, open)
/* old location of FMODE_EXEC vs S_ISREG() test */
security_file_open(f)
open()
[1] https://lore.kernel.org/lkml/202006041910.9EF0C602@keescook/
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Aleksa Sarai <cyphar@cyphar.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christian Brauner <christian.brauner@ubuntu.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Eric Biggers <ebiggers3@gmail.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Link: http://lkml.kernel.org/r/20200605160013.3954297-3-keescook@chromium.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull init and set_fs() cleanups from Al Viro:
"Christoph's 'getting rid of ksys_...() uses under KERNEL_DS' series"
* 'hch.init_path' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (50 commits)
init: add an init_dup helper
init: add an init_utimes helper
init: add an init_stat helper
init: add an init_mknod helper
init: add an init_mkdir helper
init: add an init_symlink helper
init: add an init_link helper
init: add an init_eaccess helper
init: add an init_chmod helper
init: add an init_chown helper
init: add an init_chroot helper
init: add an init_chdir helper
init: add an init_rmdir helper
init: add an init_unlink helper
init: add an init_umount helper
init: add an init_mount helper
init: mark create_dev as __init
init: mark console_on_rootfs as __init
init: initialize ramdisk_execute_command at compile time
devtmpfs: refactor devtmpfsd()
...
Add a simple helper to check if a file exists based on kernel space file
name and switch the early init code over to it. Note that this
theoretically changes behavior as it always is based on the effective
permissions. But during early init that doesn't make a difference.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Add a simple helper to chroot with a kernel space file name and switch
the early init code over to it. Remove the now unused ksys_chroot.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Add a simple helper to chdir with a kernel space file name and switch
the early init code over to it. Remove the now unused ksys_chdir.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Add a helper for struct file based chmode operations. To be used by
the initramfs code soon.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a helper for struct file based chown operations. To be used by
the initramfs code soon.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
One of the use-cases of close_range() is to drop file descriptors just before
execve(). This would usually be expressed in the sequence:
unshare(CLONE_FILES);
close_range(3, ~0U);
as pointed out by Linus it might be desirable to have this be a part of
close_range() itself under a new flag CLOSE_RANGE_UNSHARE.
This expands {dup,unshare)_fd() to take a max_fds argument that indicates the
maximum number of file descriptors to copy from the old struct files. When the
user requests that all file descriptors are supposed to be closed via
close_range(min, max) then we can cap via unshare_fd(min) and hence don't need
to do any of the heavy fput() work for everything above min.
The patch makes it so that if CLOSE_RANGE_UNSHARE is requested and we do in
fact currently share our file descriptor table we create a new private copy.
We then close all fds in the requested range and finally after we're done we
install the new fd table.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
This adds the close_range() syscall. It allows to efficiently close a range
of file descriptors up to all file descriptors of a calling task.
I was contacted by FreeBSD as they wanted to have the same close_range()
syscall as we proposed here. We've coordinated this and in the meantime, Kyle
was fast enough to merge close_range() into FreeBSD already in April:
https://reviews.freebsd.org/D21627https://svnweb.freebsd.org/base?view=revision&revision=359836
and the current plan is to backport close_range() to FreeBSD 12.2 (cf. [2])
once its merged in Linux too. Python is in the process of switching to
close_range() on FreeBSD and they are waiting on us to merge this to switch on
Linux as well: https://bugs.python.org/issue38061
The syscall came up in a recent discussion around the new mount API and
making new file descriptor types cloexec by default. During this
discussion, Al suggested the close_range() syscall (cf. [1]). Note, a
syscall in this manner has been requested by various people over time.
First, it helps to close all file descriptors of an exec()ing task. This
can be done safely via (quoting Al's example from [1] verbatim):
/* that exec is sensitive */
unshare(CLONE_FILES);
/* we don't want anything past stderr here */
close_range(3, ~0U);
execve(....);
The code snippet above is one way of working around the problem that file
descriptors are not cloexec by default. This is aggravated by the fact that
we can't just switch them over without massively regressing userspace. For
a whole class of programs having an in-kernel method of closing all file
descriptors is very helpful (e.g. demons, service managers, programming
language standard libraries, container managers etc.).
(Please note, unshare(CLONE_FILES) should only be needed if the calling
task is multi-threaded and shares the file descriptor table with another
thread in which case two threads could race with one thread allocating file
descriptors and the other one closing them via close_range(). For the
general case close_range() before the execve() is sufficient.)
Second, it allows userspace to avoid implementing closing all file
descriptors by parsing through /proc/<pid>/fd/* and calling close() on each
file descriptor. From looking at various large(ish) userspace code bases
this or similar patterns are very common in:
- service managers (cf. [4])
- libcs (cf. [6])
- container runtimes (cf. [5])
- programming language runtimes/standard libraries
- Python (cf. [2])
- Rust (cf. [7], [8])
As Dmitry pointed out there's even a long-standing glibc bug about missing
kernel support for this task (cf. [3]).
In addition, the syscall will also work for tasks that do not have procfs
mounted and on kernels that do not have procfs support compiled in. In such
situations the only way to make sure that all file descriptors are closed
is to call close() on each file descriptor up to UINT_MAX or RLIMIT_NOFILE,
OPEN_MAX trickery (cf. comment [8] on Rust).
The performance is striking. For good measure, comparing the following
simple close_all_fds() userspace implementation that is essentially just
glibc's version in [6]:
static int close_all_fds(void)
{
int dir_fd;
DIR *dir;
struct dirent *direntp;
dir = opendir("/proc/self/fd");
if (!dir)
return -1;
dir_fd = dirfd(dir);
while ((direntp = readdir(dir))) {
int fd;
if (strcmp(direntp->d_name, ".") == 0)
continue;
if (strcmp(direntp->d_name, "..") == 0)
continue;
fd = atoi(direntp->d_name);
if (fd == dir_fd || fd == 0 || fd == 1 || fd == 2)
continue;
close(fd);
}
closedir(dir);
return 0;
}
to close_range() yields:
1. closing 4 open files:
- close_all_fds(): ~280 us
- close_range(): ~24 us
2. closing 1000 open files:
- close_all_fds(): ~5000 us
- close_range(): ~800 us
close_range() is designed to allow for some flexibility. Specifically, it
does not simply always close all open file descriptors of a task. Instead,
callers can specify an upper bound.
This is e.g. useful for scenarios where specific file descriptors are
created with well-known numbers that are supposed to be excluded from
getting closed.
For extra paranoia close_range() comes with a flags argument. This can e.g.
be used to implement extension. Once can imagine userspace wanting to stop
at the first error instead of ignoring errors under certain circumstances.
There might be other valid ideas in the future. In any case, a flag
argument doesn't hurt and keeps us on the safe side.
From an implementation side this is kept rather dumb. It saw some input
from David and Jann but all nonsense is obviously my own!
- Errors to close file descriptors are currently ignored. (Could be changed
by setting a flag in the future if needed.)
- __close_range() is a rather simplistic wrapper around __close_fd().
My reasoning behind this is based on the nature of how __close_fd() needs
to release an fd. But maybe I misunderstood specifics:
We take the files_lock and rcu-dereference the fdtable of the calling
task, we find the entry in the fdtable, get the file and need to release
files_lock before calling filp_close().
In the meantime the fdtable might have been altered so we can't just
retake the spinlock and keep the old rcu-reference of the fdtable
around. Instead we need to grab a fresh reference to the fdtable.
If my reasoning is correct then there's really no point in fancyfying
__close_range(): We just need to rcu-dereference the fdtable of the
calling task once to cap the max_fd value correctly and then go on
calling __close_fd() in a loop.
/* References */
[1]: https://lore.kernel.org/lkml/20190516165021.GD17978@ZenIV.linux.org.uk/
[2]: 9e4f2f3a6b/Modules/_posixsubprocess.c (L220)
[3]: https://sourceware.org/bugzilla/show_bug.cgi?id=10353#c7
[4]: 5238e95759/src/basic/fd-util.c (L217)
[5]: ddf4b77e11/src/lxc/start.c (L236)
[6]: https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/grantpt.c;h=2030e07fa6e652aac32c775b8c6e005844c3c4eb;hb=HEAD#l17
Note that this is an internal implementation that is not exported.
Currently, libc seems to not provide an exported version of this
because of missing kernel support to do this.
Note, in a recent patch series Florian made grantpt() a nop thereby
removing the code referenced here.
[7]: https://github.com/rust-lang/rust/issues/12148
[8]: 5f47c0613e/src/libstd/sys/unix/process2.rs (L303-L308)
Rust's solution is slightly different but is equally unperformant.
Rust calls getdtablesize() which is a glibc library function that
simply returns the current RLIMIT_NOFILE or OPEN_MAX values. Rust then
goes on to call close() on each fd. That's obviously overkill for most
tasks. Rarely, tasks - especially non-demons - hit RLIMIT_NOFILE or
OPEN_MAX.
Let's be nice and assume an unprivileged user with RLIMIT_NOFILE set
to 1024. Even in this case, there's a very high chance that in the
common case Rust is calling the close() syscall 1021 times pointlessly
if the task just has 0, 1, and 2 open.
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Kyle Evans <self@kyle-evans.net>
Cc: Jann Horn <jannh@google.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Dmitry V. Levin <ldv@altlinux.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Florian Weimer <fweimer@redhat.com>
Cc: linux-api@vger.kernel.org
Merge updates from Andrew Morton:
"A few little subsystems and a start of a lot of MM patches.
Subsystems affected by this patch series: squashfs, ocfs2, parisc,
vfs. With mm subsystems: slab-generic, slub, debug, pagecache, gup,
swap, memcg, pagemap, memory-failure, vmalloc, kasan"
* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (128 commits)
kasan: move kasan_report() into report.c
mm/mm_init.c: report kasan-tag information stored in page->flags
ubsan: entirely disable alignment checks under UBSAN_TRAP
kasan: fix clang compilation warning due to stack protector
x86/mm: remove vmalloc faulting
mm: remove vmalloc_sync_(un)mappings()
x86/mm/32: implement arch_sync_kernel_mappings()
x86/mm/64: implement arch_sync_kernel_mappings()
mm/ioremap: track which page-table levels were modified
mm/vmalloc: track which page-table levels were modified
mm: add functions to track page directory modifications
s390: use __vmalloc_node in stack_alloc
powerpc: use __vmalloc_node in alloc_vm_stack
arm64: use __vmalloc_node in arch_alloc_vmap_stack
mm: remove vmalloc_user_node_flags
mm: switch the test_vmalloc module to use __vmalloc_node
mm: remove __vmalloc_node_flags_caller
mm: remove both instances of __vmalloc_node_flags
mm: remove the prot argument to __vmalloc_node
mm: remove the pgprot argument to __vmalloc
...
Patch series "vfs: have syncfs() return error when there are writeback
errors", v6.
Currently, syncfs does not return errors when one of the inodes fails to
be written back. It will return errors based on the legacy AS_EIO and
AS_ENOSPC flags when syncing out the block device fails, but that's not
particularly helpful for filesystems that aren't backed by a blockdev.
It's also possible for a stray sync to lose those errors.
The basic idea in this set is to track writeback errors at the
superblock level, so that we can quickly and easily check whether
something bad happened without having to fsync each file individually.
syncfs is then changed to reliably report writeback errors after they
occur, much in the same fashion as fsync does now.
This patch (of 2):
Usually we suggest that applications call fsync when they want to ensure
that all data written to the file has made it to the backing store, but
that can be inefficient when there are a lot of open files.
Calling syncfs on the filesystem can be more efficient in some
situations, but the error reporting doesn't currently work the way most
people expect. If a single inode on a filesystem reports a writeback
error, syncfs won't necessarily return an error. syncfs only returns an
error if __sync_blockdev fails, and on some filesystems that's a no-op.
It would be better if syncfs reported an error if there were any
writeback failures. Then applications could call syncfs to see if there
are any errors on any open files, and could then call fsync on all of
the other descriptors to figure out which one failed.
This patch adds a new errseq_t to struct super_block, and has
mapping_set_error also record writeback errors there.
To report those errors, we also need to keep an errseq_t in struct file
to act as a cursor. This patch adds a dedicated field for that purpose,
which slots nicely into 4 bytes of padding at the end of struct file on
x86_64.
An earlier version of this patch used an O_PATH file descriptor to cue
the kernel that the open file should track the superblock error and not
the inode's writeback error.
I think that API is just too weird though. This is simpler and should
make syncfs error reporting "just work" even if someone is multiplexing
fsync and syncfs on the same fds.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Andres Freund <andres@anarazel.de>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: David Howells <dhowells@redhat.com>
Link: http://lkml.kernel.org/r/20200428135155.19223-1-jlayton@kernel.org
Link: http://lkml.kernel.org/r/20200428135155.19223-2-jlayton@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
POSIX defines faccessat() as having a fourth "flags" argument, while the
linux syscall doesn't have it. Glibc tries to emulate AT_EACCESS and
AT_SYMLINK_NOFOLLOW, but AT_EACCESS emulation is broken.
Add a new faccessat(2) syscall with the added flags argument and implement
both flags.
The value of AT_EACCESS is defined in glibc headers to be the same as
AT_REMOVEDIR. Use this value for the kernel interface as well, together
with the explanatory comment.
Also add AT_EMPTY_PATH support, which is not documented by POSIX, but can
be useful and is trivial to implement.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Split out a helper that overrides the credentials in preparation for
actually doing the access check.
This prepares for the next patch that optionally disables the creds
override.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Pull vfs pathwalk sanitizing from Al Viro:
"Massive pathwalk rewrite and cleanups.
Several iterations have been posted; hopefully this thing is getting
readable and understandable now. Pretty much all parts of pathname
resolutions are affected...
The branch is identical to what has sat in -next, except for commit
message in "lift all calls of step_into() out of follow_dotdot/
follow_dotdot_rcu", crediting Qian Cai for reporting the bug; only
commit message changed there."
* 'work.dotdot1' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (69 commits)
lookup_open(): don't bother with fallbacks to lookup+create
atomic_open(): no need to pass struct open_flags anymore
open_last_lookups(): move complete_walk() into do_open()
open_last_lookups(): lift O_EXCL|O_CREAT handling into do_open()
open_last_lookups(): don't abuse complete_walk() when all we want is unlazy
open_last_lookups(): consolidate fsnotify_create() calls
take post-lookup part of do_last() out of loop
link_path_walk(): sample parent's i_uid and i_mode for the last component
__nd_alloc_stack(): make it return bool
reserve_stack(): switch to __nd_alloc_stack()
pick_link(): take reserving space on stack into a new helper
pick_link(): more straightforward handling of allocation failures
fold path_to_nameidata() into its only remaining caller
pick_link(): pass it struct path already with normal refcounting rules
fs/namei.c: kill follow_mount()
non-RCU analogue of the previous commit
helper for mount rootwards traversal
follow_dotdot(): be lazy about changing nd->path
follow_dotdot_rcu(): be lazy about changing nd->path
follow_dotdot{,_rcu}(): massage loops
...
several iterations of ->atomic_open() calling conventions ago, we
used to need fput() if ->atomic_open() failed at some point after
successful finish_open(). Now (since 2016) it's not needed -
struct file carries enough state to make fput() work regardless
of the point in struct file lifecycle and discarding it on
failure exits in open() got unified. Unfortunately, I'd missed
the fact that we had an instance of ->atomic_open() (cifs one)
that used to need that fput(), as well as the stale comment in
finish_open() demanding such late failure handling. Trivially
fixed...
Fixes: fe9ec8291f "do_last(): take fput() on error after opening to out:"
Cc: stable@kernel.org # v4.7+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
O_CREAT | O_EXCL means "-EEXIST if we run into a trailing symlink".
As it is, we might or might not have LOOKUP_FOLLOW in op->intent
in that case - that depends upon having O_NOFOLLOW in open flags.
It doesn't matter, since we won't be checking it in that case -
do_last() bails out earlier.
However, making sure it's not set (i.e. acting as if we had an explicit
O_NOFOLLOW) makes the behaviour more explicit and allows to reorder the
check for O_CREAT | O_EXCL in do_last() with the call of step_into()
immediately following it.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
/* Background. */
For a very long time, extending openat(2) with new features has been
incredibly frustrating. This stems from the fact that openat(2) is
possibly the most famous counter-example to the mantra "don't silently
accept garbage from userspace" -- it doesn't check whether unknown flags
are present[1].
This means that (generally) the addition of new flags to openat(2) has
been fraught with backwards-compatibility issues (O_TMPFILE has to be
defined as __O_TMPFILE|O_DIRECTORY|[O_RDWR or O_WRONLY] to ensure old
kernels gave errors, since it's insecure to silently ignore the
flag[2]). All new security-related flags therefore have a tough road to
being added to openat(2).
Userspace also has a hard time figuring out whether a particular flag is
supported on a particular kernel. While it is now possible with
contemporary kernels (thanks to [3]), older kernels will expose unknown
flag bits through fcntl(F_GETFL). Giving a clear -EINVAL during
openat(2) time matches modern syscall designs and is far more
fool-proof.
In addition, the newly-added path resolution restriction LOOKUP flags
(which we would like to expose to user-space) don't feel related to the
pre-existing O_* flag set -- they affect all components of path lookup.
We'd therefore like to add a new flag argument.
Adding a new syscall allows us to finally fix the flag-ignoring problem,
and we can make it extensible enough so that we will hopefully never
need an openat3(2).
/* Syscall Prototype. */
/*
* open_how is an extensible structure (similar in interface to
* clone3(2) or sched_setattr(2)). The size parameter must be set to
* sizeof(struct open_how), to allow for future extensions. All future
* extensions will be appended to open_how, with their zero value
* acting as a no-op default.
*/
struct open_how { /* ... */ };
int openat2(int dfd, const char *pathname,
struct open_how *how, size_t size);
/* Description. */
The initial version of 'struct open_how' contains the following fields:
flags
Used to specify openat(2)-style flags. However, any unknown flag
bits or otherwise incorrect flag combinations (like O_PATH|O_RDWR)
will result in -EINVAL. In addition, this field is 64-bits wide to
allow for more O_ flags than currently permitted with openat(2).
mode
The file mode for O_CREAT or O_TMPFILE.
Must be set to zero if flags does not contain O_CREAT or O_TMPFILE.
resolve
Restrict path resolution (in contrast to O_* flags they affect all
path components). The current set of flags are as follows (at the
moment, all of the RESOLVE_ flags are implemented as just passing
the corresponding LOOKUP_ flag).
RESOLVE_NO_XDEV => LOOKUP_NO_XDEV
RESOLVE_NO_SYMLINKS => LOOKUP_NO_SYMLINKS
RESOLVE_NO_MAGICLINKS => LOOKUP_NO_MAGICLINKS
RESOLVE_BENEATH => LOOKUP_BENEATH
RESOLVE_IN_ROOT => LOOKUP_IN_ROOT
open_how does not contain an embedded size field, because it is of
little benefit (userspace can figure out the kernel open_how size at
runtime fairly easily without it). It also only contains u64s (even
though ->mode arguably should be a u16) to avoid having padding fields
which are never used in the future.
Note that as a result of the new how->flags handling, O_PATH|O_TMPFILE
is no longer permitted for openat(2). As far as I can tell, this has
always been a bug and appears to not be used by userspace (and I've not
seen any problems on my machines by disallowing it). If it turns out
this breaks something, we can special-case it and only permit it for
openat(2) but not openat2(2).
After input from Florian Weimer, the new open_how and flag definitions
are inside a separate header from uapi/linux/fcntl.h, to avoid problems
that glibc has with importing that header.
/* Testing. */
In a follow-up patch there are over 200 selftests which ensure that this
syscall has the correct semantics and will correctly handle several
attack scenarios.
In addition, I've written a userspace library[4] which provides
convenient wrappers around openat2(RESOLVE_IN_ROOT) (this is necessary
because no other syscalls support RESOLVE_IN_ROOT, and thus lots of care
must be taken when using RESOLVE_IN_ROOT'd file descriptors with other
syscalls). During the development of this patch, I've run numerous
verification tests using libpathrs (showing that the API is reasonably
usable by userspace).
/* Future Work. */
Additional RESOLVE_ flags have been suggested during the review period.
These can be easily implemented separately (such as blocking auto-mount
during resolution).
Furthermore, there are some other proposed changes to the openat(2)
interface (the most obvious example is magic-link hardening[5]) which
would be a good opportunity to add a way for userspace to restrict how
O_PATH file descriptors can be re-opened.
Another possible avenue of future work would be some kind of
CHECK_FIELDS[6] flag which causes the kernel to indicate to userspace
which openat2(2) flags and fields are supported by the current kernel
(to avoid userspace having to go through several guesses to figure it
out).
[1]: https://lwn.net/Articles/588444/
[2]: https://lore.kernel.org/lkml/CA+55aFyyxJL1LyXZeBsf2ypriraj5ut1XkNDsunRBqgVjZU_6Q@mail.gmail.com
[3]: commit 629e014bb8 ("fs: completely ignore unknown open flags")
[4]: https://sourceware.org/bugzilla/show_bug.cgi?id=17523
[5]: https://lore.kernel.org/lkml/20190930183316.10190-2-cyphar@cyphar.com/
[6]: https://youtu.be/ggD-eb3yPVs
Suggested-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
This reverts commit 0be0ee7181.
I was hoping it would be benign to switch over entirely to FMODE_STREAM,
and we'd have just a couple of small fixups we'd need, but it looks like
we're not quite there yet.
While it worked fine on both my desktop and laptop, they are fairly
similar in other respects, and run mostly the same loads. Kenneth
Crudup reports that it seems to break both his vmware installation and
the KDE upower service. In both cases apparently leading to timeouts
due to waitinmg for the f_pos lock.
There are a number of character devices in particular that definitely
want stream-like behavior, but that currently don't get marked as
streams, and as a result get the exclusion between concurrent
read()/write() on the same file descriptor. Which doesn't work well for
them.
The most obvious example if this is /dev/console and /dev/tty, which use
console_fops and tty_fops respectively (and ptmx_fops for the pty master
side). It may be that it's just this that causes problems, but we
clearly weren't ready yet.
Because there's a number of other likely common cases that don't have
llseek implementations and would seem to act as stream devices:
/dev/fuse (fuse_dev_operations)
/dev/mcelog (mce_chrdev_ops)
/dev/mei0 (mei_fops)
/dev/net/tun (tun_fops)
/dev/nvme0 (nvme_dev_fops)
/dev/tpm0 (tpm_fops)
/proc/self/ns/mnt (ns_file_operations)
/dev/snd/pcm* (snd_pcm_f_ops[])
and while some of these could be trivially automatically detected by the
vfs layer when the character device is opened by just noticing that they
have no read or write operations either, it often isn't that obvious.
Some character devices most definitely do use the file position, even if
they don't allow seeking: the firmware update code, for example, uses
simple_read_from_buffer() that does use f_pos, but doesn't allow seeking
back and forth.
We'll revisit this when there's a better way to detect the problem and
fix it (possibly with a coccinelle script to do more of the FMODE_STREAM
annotations).
Reported-by: Kenneth R. Crudup <kenny@panix.com>
Cc: Kirill Smelkov <kirr@nexedi.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fdget_pos() is used by file operations that will read and update f_pos:
things like "read()", "write()" and "lseek()" (but not, for example,
"pread()/pwrite" that get their file positions elsewhere).
However, it had two separate escape clauses for this, because not
everybody wants or needs serialization of the file position.
The first and most obvious case is the "file descriptor doesn't have a
position at all", ie a stream-like file. Except we didn't actually use
FMODE_STREAM, but instead used FMODE_ATOMIC_POS. The reason for that
was that FMODE_STREAM didn't exist back in the days, but also that we
didn't want to mark all the special cases, so we only marked the ones
that _required_ position atomicity according to POSIX - regular files
and directories.
The case one was intentionally lazy, but now that we _do_ have
FMODE_STREAM we could and should just use it. With the change to use
FMODE_STREAM, there are no remaining uses for FMODE_ATOMIC_POS, and all
the code to set it is deleted.
Any cases where we don't want the serialization because the driver (or
subsystem) doesn't use the file position should just be updated to do
"stream_open()". We've done that for all the obvious and common
situations, we may need a few more. Quoting Kirill Smelkov in the
original FMODE_STREAM thread (see link below for full email):
"And I appreciate if people could help at least somehow with "getting
rid of mixed case entirely" (i.e. always lock f_pos_lock on
!FMODE_STREAM), because this transition starts to diverge from my
particular use-case too far. To me it makes sense to do that
transition as follows:
- convert nonseekable_open -> stream_open via stream_open.cocci;
- audit other nonseekable_open calls and convert left users that
truly don't depend on position to stream_open;
- extend stream_open.cocci to analyze alloc_file_pseudo as well (this
will cover pipes and sockets), or maybe convert pipes and sockets
to FMODE_STREAM manually;
- extend stream_open.cocci to analyze file_operations that use
no_llseek or noop_llseek, but do not use nonseekable_open or
alloc_file_pseudo. This might find files that have stream semantic
but are opened differently;
- extend stream_open.cocci to analyze file_operations whose
.read/.write do not use ppos at all (independently of how file was
opened);
- ...
- after that remove FMODE_ATOMIC_POS and always take f_pos_lock if
!FMODE_STREAM;
- gather bug reports for deadlocked read/write and convert missed
cases to FMODE_STREAM, probably extending stream_open.cocci along
the road to catch similar cases
i.e. always take f_pos_lock unless a file is explicitly marked as
being stream, and try to find and cover all files that are streams"
We have not done the "extend stream_open.cocci to analyze
alloc_file_pseudo" as well, but the previous commit did manually handle
the case of pipes and sockets.
The other case where we can avoid locking f_pos is the "this file
descriptor only has a single user and it is us, and thus there is no
need to lock it".
The second test was correct, although a bit subtle and worth just
re-iterating here. There are two kinds of other sources of references
to the same file descriptor: file descriptors that have been explicitly
shared across fork() or with dup(), and file tables having elevated
reference counts due to threading (or explicit file sharing with
clone()).
The first case would have incremented the file count explicitly, and in
the second case the previous __fdget() would have incremented it for us
and set the FDPUT_FPUT flag.
But in both cases the file count would be greater than one, so the
"file_count(file) > 1" test catches both situations. Also note that if
file_count is 1, that also means that no other thread can have access to
the file table, so there also cannot be races with concurrent calls to
dup()/fork()/clone() that would increment the file count any other way.
Link: https://lore.kernel.org/linux-fsdevel/20190413184404.GA13490@deco.navytux.spb.ru
Cc: Kirill Smelkov <kirr@nexedi.com>
Cc: Eic Dumazet <edumazet@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: Marco Elver <elver@google.com>
Cc: Andrea Parri <parri.andrea@gmail.com>
Cc: Paul McKenney <paulmck@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In previous patch, an application could put part of its text section in
THP via madvise(). These THPs will be protected from writes when the
application is still running (TXTBSY). However, after the application
exits, the file is available for writes.
This patch avoids writes to file THP by dropping page cache for the file
when the file is open for write. A new counter nr_thps is added to struct
address_space. In do_dentry_open(), if the file is open for write and
nr_thps is non-zero, we drop page cache for the whole file.
Link: http://lkml.kernel.org/r/20190801184244.3169074-8-songliubraving@fb.com
Signed-off-by: Song Liu <songliubraving@fb.com>
Reported-by: kbuild test robot <lkp@intel.com>
Acked-by: Rik van Riel <riel@surriel.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It turns out that 'access()' (and 'faccessat()') can cause a lot of RCU
work because it installs a temporary credential that gets allocated and
freed for each system call.
The allocation and freeing overhead is mostly benign, but because
credentials can be accessed under the RCU read lock, the freeing
involves a RCU grace period.
Which is not a huge deal normally, but if you have a lot of access()
calls, this causes a fair amount of seconday damage: instead of having a
nice alloc/free patterns that hits in hot per-CPU slab caches, you have
all those delayed free's, and on big machines with hundreds of cores,
the RCU overhead can end up being enormous.
But it turns out that all of this is entirely unnecessary. Exactly
because access() only installs the credential as the thread-local
subjective credential, the temporary cred pointer doesn't actually need
to be RCU free'd at all. Once we're done using it, we can just free it
synchronously and avoid all the RCU overhead.
So add a 'non_rcu' flag to 'struct cred', which can be set by users that
know they only use it in non-RCU context (there are other potential
users for this). We can make it a union with the rcu freeing list head
that we need for the RCU case, so this doesn't need any extra storage.
Note that this also makes 'get_current_cred()' clear the new non_rcu
flag, in case we have filesystems that take a long-term reference to the
cred and then expect the RCU delayed freeing afterwards. It's not
entirely clear that this is required, but it makes for clear semantics:
the subjective cred remains non-RCU as long as you only access it
synchronously using the thread-local accessors, but you _can_ use it as
a generic cred if you want to.
It is possible that we should just remove the whole RCU markings for
->cred entirely. Only ->real_cred is really supposed to be accessed
through RCU, and the long-term cred copies that nfs uses might want to
explicitly re-enable RCU freeing if required, rather than have
get_current_cred() do it implicitly.
But this is a "minimal semantic changes" change for the immediate
problem.
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paul E. McKenney <paulmck@linux.ibm.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Jan Glauber <jglauber@marvell.com>
Cc: Jiri Kosina <jikos@kernel.org>
Cc: Jayachandran Chandrasekharan Nair <jnair@marvell.com>
Cc: Greg KH <greg@kroah.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add SPDX license identifiers to all files which:
- Have no license information of any form
- Have EXPORT_.*_SYMBOL_GPL inside which was used in the
initial scan/conversion to ignore the file
These files fall under the project license, GPL v2 only. The resulting SPDX
license identifier is:
GPL-2.0-only
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This amends commit 10dce8af34 ("fs: stream_open - opener for
stream-like files so that read and write can run simultaneously without
deadlock") in how position is passed into .read()/.write() handler for
stream-like files:
Rasmus noticed that we currently pass 0 as position and ignore any position
change if that is done by a file implementation. This papers over bugs if ppos
is used in files that declare themselves as being stream-like as such bugs will
go unnoticed. Even if a file implementation is correctly converted into using
stream_open, its read/write later could be changed to use ppos and even though
that won't be working correctly, that bug might go unnoticed without someone
doing wrong behaviour analysis. It is thus better to pass ppos=NULL into
read/write for stream-like files as that don't give any chance for ppos usage
bugs because it will oops if ppos is ever used inside .read() or .write().
Note 1: rw_verify_area, new_sync_{read,write} needs to be updated
because they are called by vfs_read/vfs_write & friends before
file_operations .read/.write .
Note 2: if file backend uses new-style .read_iter/.write_iter, position
is still passed into there as non-pointer kiocb.ki_pos . Currently
stream_open.cocci (semantic patch added by 10dce8af34) ignores files
whose file_operations has *_iter methods.
Suggested-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Kirill Smelkov <kirr@nexedi.com>
Commit 9c225f2655 ("vfs: atomic f_pos accesses as per POSIX") added
locking for file.f_pos access and in particular made concurrent read and
write not possible - now both those functions take f_pos lock for the
whole run, and so if e.g. a read is blocked waiting for data, write will
deadlock waiting for that read to complete.
This caused regression for stream-like files where previously read and
write could run simultaneously, but after that patch could not do so
anymore. See e.g. commit 581d21a2d0 ("xenbus: fix deadlock on writes
to /proc/xen/xenbus") which fixes such regression for particular case of
/proc/xen/xenbus.
The patch that added f_pos lock in 2014 did so to guarantee POSIX thread
safety for read/write/lseek and added the locking to file descriptors of
all regular files. In 2014 that thread-safety problem was not new as it
was already discussed earlier in 2006.
However even though 2006'th version of Linus's patch was adding f_pos
locking "only for files that are marked seekable with FMODE_LSEEK (thus
avoiding the stream-like objects like pipes and sockets)", the 2014
version - the one that actually made it into the tree as 9c225f2655 -
is doing so irregardless of whether a file is seekable or not.
See
https://lore.kernel.org/lkml/53022DB1.4070805@gmail.com/https://lwn.net/Articles/180387https://lwn.net/Articles/180396
for historic context.
The reason that it did so is, probably, that there are many files that
are marked non-seekable, but e.g. their read implementation actually
depends on knowing current position to correctly handle the read. Some
examples:
kernel/power/user.c snapshot_read
fs/debugfs/file.c u32_array_read
fs/fuse/control.c fuse_conn_waiting_read + ...
drivers/hwmon/asus_atk0110.c atk_debugfs_ggrp_read
arch/s390/hypfs/inode.c hypfs_read_iter
...
Despite that, many nonseekable_open users implement read and write with
pure stream semantics - they don't depend on passed ppos at all. And for
those cases where read could wait for something inside, it creates a
situation similar to xenbus - the write could be never made to go until
read is done, and read is waiting for some, potentially external, event,
for potentially unbounded time -> deadlock.
Besides xenbus, there are 14 such places in the kernel that I've found
with semantic patch (see below):
drivers/xen/evtchn.c:667:8-24: ERROR: evtchn_fops: .read() can deadlock .write()
drivers/isdn/capi/capi.c:963:8-24: ERROR: capi_fops: .read() can deadlock .write()
drivers/input/evdev.c:527:1-17: ERROR: evdev_fops: .read() can deadlock .write()
drivers/char/pcmcia/cm4000_cs.c:1685:7-23: ERROR: cm4000_fops: .read() can deadlock .write()
net/rfkill/core.c:1146:8-24: ERROR: rfkill_fops: .read() can deadlock .write()
drivers/s390/char/fs3270.c:488:1-17: ERROR: fs3270_fops: .read() can deadlock .write()
drivers/usb/misc/ldusb.c:310:1-17: ERROR: ld_usb_fops: .read() can deadlock .write()
drivers/hid/uhid.c:635:1-17: ERROR: uhid_fops: .read() can deadlock .write()
net/batman-adv/icmp_socket.c:80:1-17: ERROR: batadv_fops: .read() can deadlock .write()
drivers/media/rc/lirc_dev.c:198:1-17: ERROR: lirc_fops: .read() can deadlock .write()
drivers/leds/uleds.c:77:1-17: ERROR: uleds_fops: .read() can deadlock .write()
drivers/input/misc/uinput.c:400:1-17: ERROR: uinput_fops: .read() can deadlock .write()
drivers/infiniband/core/user_mad.c:985:7-23: ERROR: umad_fops: .read() can deadlock .write()
drivers/gnss/core.c:45:1-17: ERROR: gnss_fops: .read() can deadlock .write()
In addition to the cases above another regression caused by f_pos
locking is that now FUSE filesystems that implement open with
FOPEN_NONSEEKABLE flag, can no longer implement bidirectional
stream-like files - for the same reason as above e.g. read can deadlock
write locking on file.f_pos in the kernel.
FUSE's FOPEN_NONSEEKABLE was added in 2008 in a7c1b990f7 ("fuse:
implement nonseekable open") to support OSSPD. OSSPD implements /dev/dsp
in userspace with FOPEN_NONSEEKABLE flag, with corresponding read and
write routines not depending on current position at all, and with both
read and write being potentially blocking operations:
See
https://github.com/libfuse/osspdhttps://lwn.net/Articles/308445https://github.com/libfuse/osspd/blob/14a9cff0/osspd.c#L1406https://github.com/libfuse/osspd/blob/14a9cff0/osspd.c#L1438-L1477https://github.com/libfuse/osspd/blob/14a9cff0/osspd.c#L1479-L1510
Corresponding libfuse example/test also describes FOPEN_NONSEEKABLE as
"somewhat pipe-like files ..." with read handler not using offset.
However that test implements only read without write and cannot exercise
the deadlock scenario:
https://github.com/libfuse/libfuse/blob/fuse-3.4.2-3-ga1bff7d/example/poll.c#L124-L131https://github.com/libfuse/libfuse/blob/fuse-3.4.2-3-ga1bff7d/example/poll.c#L146-L163https://github.com/libfuse/libfuse/blob/fuse-3.4.2-3-ga1bff7d/example/poll.c#L209-L216
I've actually hit the read vs write deadlock for real while implementing
my FUSE filesystem where there is /head/watch file, for which open
creates separate bidirectional socket-like stream in between filesystem
and its user with both read and write being later performed
simultaneously. And there it is semantically not easy to split the
stream into two separate read-only and write-only channels:
https://lab.nexedi.com/kirr/wendelin.core/blob/f13aa600/wcfs/wcfs.go#L88-169
Let's fix this regression. The plan is:
1. We can't change nonseekable_open to include &~FMODE_ATOMIC_POS -
doing so would break many in-kernel nonseekable_open users which
actually use ppos in read/write handlers.
2. Add stream_open() to kernel to open stream-like non-seekable file
descriptors. Read and write on such file descriptors would never use
nor change ppos. And with that property on stream-like files read and
write will be running without taking f_pos lock - i.e. read and write
could be running simultaneously.
3. With semantic patch search and convert to stream_open all in-kernel
nonseekable_open users for which read and write actually do not
depend on ppos and where there is no other methods in file_operations
which assume @offset access.
4. Add FOPEN_STREAM to fs/fuse/ and open in-kernel file-descriptors via
steam_open if that bit is present in filesystem open reply.
It was tempting to change fs/fuse/ open handler to use stream_open
instead of nonseekable_open on just FOPEN_NONSEEKABLE flags, but
grepping through Debian codesearch shows users of FOPEN_NONSEEKABLE,
and in particular GVFS which actually uses offset in its read and
write handlers
https://codesearch.debian.net/search?q=-%3Enonseekable+%3Dhttps://gitlab.gnome.org/GNOME/gvfs/blob/1.40.0-6-gcbc54396/client/gvfsfusedaemon.c#L1080https://gitlab.gnome.org/GNOME/gvfs/blob/1.40.0-6-gcbc54396/client/gvfsfusedaemon.c#L1247-1346https://gitlab.gnome.org/GNOME/gvfs/blob/1.40.0-6-gcbc54396/client/gvfsfusedaemon.c#L1399-1481
so if we would do such a change it will break a real user.
5. Add stream_open and FOPEN_STREAM handling to stable kernels starting
from v3.14+ (the kernel where 9c225f2655 first appeared).
This will allow to patch OSSPD and other FUSE filesystems that
provide stream-like files to return FOPEN_STREAM | FOPEN_NONSEEKABLE
in their open handler and this way avoid the deadlock on all kernel
versions. This should work because fs/fuse/ ignores unknown open
flags returned from a filesystem and so passing FOPEN_STREAM to a
kernel that is not aware of this flag cannot hurt. In turn the kernel
that is not aware of FOPEN_STREAM will be < v3.14 where just
FOPEN_NONSEEKABLE is sufficient to implement streams without read vs
write deadlock.
This patch adds stream_open, converts /proc/xen/xenbus to it and adds
semantic patch to automatically locate in-kernel places that are either
required to be converted due to read vs write deadlock, or that are just
safe to be converted because read and write do not use ppos and there
are no other funky methods in file_operations.
Regarding semantic patch I've verified each generated change manually -
that it is correct to convert - and each other nonseekable_open instance
left - that it is either not correct to convert there, or that it is not
converted due to current stream_open.cocci limitations.
The script also does not convert files that should be valid to convert,
but that currently have .llseek = noop_llseek or generic_file_llseek for
unknown reason despite file being opened with nonseekable_open (e.g.
drivers/input/mousedev.c)
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Yongzhi Pan <panyongzhi@gmail.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: Tejun Heo <tj@kernel.org>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Julia Lawall <Julia.Lawall@lip6.fr>
Cc: Nikolaus Rath <Nikolaus@rath.org>
Cc: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Kirill Smelkov <kirr@nexedi.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
syzbot is hitting lockdep warning [1] due to trying to open a fifo
during an execve() operation. But we don't need to open non regular
files during an execve() operation, for all files which we will need are
the executable file itself and the interpreter programs like /bin/sh and
ld-linux.so.2 .
Since the manpage for execve(2) says that execve() returns EACCES when
the file or a script interpreter is not a regular file, and the manpage
for uselib(2) says that uselib() can return EACCES, and we use
FMODE_EXEC when opening for execve()/uselib(), we can bail out if a non
regular file is requested with FMODE_EXEC set.
Since this deadlock followed by khungtaskd warnings is trivially
reproducible by a local unprivileged user, and syzbot's frequent crash
due to this deadlock defers finding other bugs, let's workaround this
deadlock until we get a chance to find a better solution.
[1] https://syzkaller.appspot.com/bug?id=b5095bfec44ec84213bac54742a82483aad578ce
Link: http://lkml.kernel.org/r/1552044017-7890-1-git-send-email-penguin-kernel@I-love.SAKURA.ne.jp
Reported-by: syzbot <syzbot+e93a80c1bb7c5c56e522461c149f8bf55eab1b2b@syzkaller.appspotmail.com>
Fixes: 8924feff66 ("splice: lift pipe_lock out of splice_to_pipe()")
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Eric Biggers <ebiggers3@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: <stable@vger.kernel.org> [4.9+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This contains two new features:
1) Stack file operations: this allows removal of several hacks from the
VFS, proper interaction of read-only open files with copy-up,
possibility to implement fs modifying ioctls properly, and others.
2) Metadata only copy-up: when file is on lower layer and only metadata is
modified (except size) then only copy up the metadata and continue to
use the data from the lower file.
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQSQHSd0lITzzeNWNm3h3BK/laaZPAUCW3srhAAKCRDh3BK/laaZ
PC6tAQCP+KklcN+TvNp502f+O/kATahSpgnun4NY1/p4I8JV+AEAzdlkTN3+MiAO
fn9brN6mBK7h59DO3hqedPLJy2vrgwg=
=QDXH
-----END PGP SIGNATURE-----
Merge tag 'ovl-update-4.19' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/vfs
Pull overlayfs updates from Miklos Szeredi:
"This contains two new features:
- Stack file operations: this allows removal of several hacks from
the VFS, proper interaction of read-only open files with copy-up,
possibility to implement fs modifying ioctls properly, and others.
- Metadata only copy-up: when file is on lower layer and only
metadata is modified (except size) then only copy up the metadata
and continue to use the data from the lower file"
* tag 'ovl-update-4.19' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/vfs: (66 commits)
ovl: Enable metadata only feature
ovl: Do not do metacopy only for ioctl modifying file attr
ovl: Do not do metadata only copy-up for truncate operation
ovl: add helper to force data copy-up
ovl: Check redirect on index as well
ovl: Set redirect on upper inode when it is linked
ovl: Set redirect on metacopy files upon rename
ovl: Do not set dentry type ORIGIN for broken hardlinks
ovl: Add an inode flag OVL_CONST_INO
ovl: Treat metacopy dentries as type OVL_PATH_MERGE
ovl: Check redirects for metacopy files
ovl: Move some dir related ovl_lookup_single() code in else block
ovl: Do not expose metacopy only dentry from d_real()
ovl: Open file with data except for the case of fsync
ovl: Add helper ovl_inode_realdata()
ovl: Store lower data inode in ovl_inode
ovl: Fix ovl_getattr() to get number of blocks from lower
ovl: Add helper ovl_dentry_lowerdata() to get lower data dentry
ovl: Copy up meta inode data from lowest data inode
ovl: Modify ovl_lookup() and friends to lookup metacopy dentry
...
This reverts commit 4d0c5ba2ff.
We now get write access on both overlay and underlying layers so this patch
is no longer needed for correct operation.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
This reverts commit 7c6893e3c9.
Overlayfs no longer relies on the vfs for checking writability of files.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Let overlayfs do its thing when opening a file.
This enables stacking and fixes the corner case when a file is opened for
read, modified through a writable open, and data is read from the read-only
file. After this patch the read-only open will not return stale data even
in this case.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Stacking file operations in overlay will store an extra open file for each
overlay file opened.
The overhead is just that of "struct file" which is about 256bytes, because
overlay already pins an extra dentry and inode when the file is open, which
add up to a much larger overhead.
For fear of breaking working setups, don't start accounting the extra file.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
open a file by given inode, faking ->f_path. Use with shitloads
of caution - at the very least you'd damn better make sure that
some dentry alias of that inode is pinned down by the path in
question. Again, this is no general-purpose interface and I hope
it will eventually go away. Right now overlayfs wants something
like that, but nothing else should.
Any out-of-tree code with bright idea of using this one *will*
eventually get hurt, with zero notice and great delight on my part.
I refuse to use EXPORT_SYMBOL_GPL(), especially in situations when
it's really EXPORT_SYMBOL_DONT_USE_IT(), but don't take that export
as "you are welcome to use it".
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
FMODE_OPENED can be used to distingusish "successful open" from the
"called finish_no_open(), do it yourself" cases. Since finish_no_open()
has been adjusted, no changes in the instances were actually needed.
The caller has been adjusted.
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
