libuuid: avoid truncate clocks.txt to improve performance
Instead of explicitly truncating clocks.txt file, pad with
whitespaces in the end of file.
This is done to improve performance of libuuid on xfs
filesystems. Instead of truncating the file, pad it with whitespaces.
This is anyways used as a failsafe method in case truncate fails.
The reason why this regression was introduced was because of: 869ae85dae64 ("xfs: flush new eof page on truncate to avoid post-eof corruption")
An attempt to move the clocks.txt to /run (tmpfs) has been attempted before
[1] and with commit ab2e7dd17 ("libuuid: move clock state file from
/var/lib to /var/run"). The latter was reverted.
Define cs_min through a define and not a const int to avoid the
following build failure with -O0 raised since version 2.39 and
https://github.com/util-linux/util-linux/commit/2fa4168c8bc9d5438bc1dfadda293c7c21b6fa59:
libuuid/src/gen_uuid.c: In function 'uuid_generate_time_generic':
libuuid/src/gen_uuid.c:536:33: error: initializer element is not constant
THREAD_LOCAL int cache_size = cs_min;
^~~~~~
For consistency, also use define for cs_max and cs_factor
Karel Zak [Thu, 2 Nov 2023 09:41:03 +0000 (10:41 +0100)]
libmount: improve mnt_table_next_child_fs()
The function utilizes the struct libmnt_itr to iterate through the mountinfo file
but neglects the direction specified by the iterator. This a bug. The application
must manage the direction, as, for instance, umount(8) requires the children of
the mountpoint in reverse order.
Fixes: https://github.com/util-linux/util-linux/issues/2552 Signed-off-by: Karel Zak <kzak@redhat.com>
Karel Zak [Wed, 1 Nov 2023 13:47:41 +0000 (14:47 +0100)]
disk-utils: add SPDX and Copyright notices
Explicitly state the license (usually GPL-2.0-or-later; our default)
and include copyright statements in all files to prevent false positive
reports from license analysis tools. Add also add SPDX-License-Identifier
tag to all files.
Yuezhang Mo [Wed, 11 Oct 2023 10:42:11 +0000 (18:42 +0800)]
libblkid: exfat: fix fail to find volume label
Commit f98b56326 set the maximum number of iterations to 10000.
If the volume label is after the 10000th entry, the volume label
will not be found. So this commit sets the maximum number of
iterations to correct value 256×1024×1024/32.
Fixes: f98b56326 ("libblkid: [exfat] Limit maximum number of iterations in find_label") Signed-off-by: Yuezhang Mo <Yuezhang.Mo@sony.com> Reviewed-by: Andy Wu <Andy.Wu@sony.com> Reviewed-by: Aoyama Wataru <wataru.aoyama@sony.com>
Thomas Weißschuh [Mon, 25 Sep 2023 22:27:22 +0000 (00:27 +0200)]
libblkid: (ntfs) validate that sector_size is a power of two
The NTFS prober reads data based off an offset of the sector size.
If the sector size is unaligned and the read data is cached then other
probers can read unaligned values.
Sector sizes for NTFS actually only make sense as power-of-two so
validate that and as a sideeffect avoid the unaligned reads.
Also add the reproducer from OSS-Fuzz that found this issue.
Some lines in /proc/cpuinfo can be large e.g. flags and can then
truncate them in displaying them
BUFSIZ can vary quite a bit e.g. glibc/linux systems its 8192
but on musl/linux and OSX its 1024, on mingW it is 256, some tests e.g.
x86_64-64cpu-linux6.2.tar.gz has added really long line for cpu flags
line which is greater than 1024 characters and hence this test fails
on musl because lscpu -s reports truncated string
Karel Zak [Wed, 23 Aug 2023 09:50:37 +0000 (11:50 +0200)]
libmount: fix statx() includes
Using sys/stat.h and linux/stat is too tricky.h together. It seems
better to rely on libc and use sys/stat.h only. Users affected
by old libc must update to use recent util-linux.
Fixes: https://github.com/util-linux/util-linux/issues/2448 Signed-off-by: Karel Zak <kzak@redhat.com>
Colin Gillespie [Wed, 9 Aug 2023 08:28:07 +0000 (18:28 +1000)]
libblkid: (bcachefs) fix not detecting large superblocks
Probing does not detect bcachefs filesystems with a superblock larger
than 4KiB. Bcachefs superblocks grow in size and can become much larger
than this.
Increase the superblock maximum size limit to 1MiB.
Validate the superblock isn't larger than the maximum size defined in
the superblocks layout section.
Filipe Manana [Thu, 17 Aug 2023 09:20:13 +0000 (10:20 +0100)]
libmount: Fix regression when mounting with atime
A regression was introduced in v2.39 that causes mounting with the atime
option to fail:
$ mkfs.ext4 -F /dev/sdi
$ mount -o atime /dev/sdi /mnt/sdi
mount: /mnt/sdi: not mount point or bad option.
dmesg(1) may have more information after failed mount system call.
The failure comes from the mount_setattr(2) call returning -EINVAL. This
is because we pass an invalid value for the attr_clr argument. From a
strace capture we have:
We can't pass MOUNT_ATTR_NOATIME to mount_setattr(2) through the attr_clr
argument because all atime options are exclusive, so in order to set atime
one has to pass MOUNT_ATTR__ATIME to attr_clr and leave attr_set as
MOUNT_ATTR_RELATIME (which is defined as a value of 0).
This can be read from the man page for mount_setattr(2) and also from the
kernel source:
$ cat fs/namespace.c
static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
struct mount_kattr *kattr, unsigned int flags)
{
(...)
/*
* Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
* users wanting to transition to a different atime setting cannot
* simply specify the atime setting in @attr_set, but must also
* specify MOUNT_ATTR__ATIME in the @attr_clr field.
* So ensure that MOUNT_ATTR__ATIME can't be partially set in
* @attr_clr and that @attr_set can't have any atime bits set if
* MOUNT_ATTR__ATIME isn't set in @attr_clr.
*/
if (attr->attr_clr & MOUNT_ATTR__ATIME) {
if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
return -EINVAL;
/*
* Clear all previous time settings as they are mutually
* exclusive.
*/
kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
switch (attr->attr_set & MOUNT_ATTR__ATIME) {
case MOUNT_ATTR_RELATIME:
kattr->attr_set |= MNT_RELATIME;
break;
case MOUNT_ATTR_NOATIME:
kattr->attr_set |= MNT_NOATIME;
break;
case MOUNT_ATTR_STRICTATIME:
break;
default:
return -EINVAL;
}
(...)
So fix this by setting attr_clr MOUNT_ATTR__ATIME if we want to clear any
atime related option.
Signed-off-by: Filipe Manana <fdmanana@kernel.org>
Karel Zak [Wed, 9 Aug 2023 09:26:28 +0000 (11:26 +0200)]
zramctl: add hint about supported algorithms
It seems the current list of the algorithms is confusing for
end-users, because it's inaccurate in many cases. Let's explain why
the list cannot be "perfect".
Addresses: https://bugzilla.redhat.com/show_bug.cgi?id=2203324 Signed-off-by: Karel Zak <kzak@redhat.com>
Markus Mayer [Tue, 1 Aug 2023 20:08:24 +0000 (13:08 -0700)]
include: define pidfd syscalls if needed
If the kernel headers are too old to provide the pidfd syscall numbers,
let's define them ourselves. This can be helpful while cross-compiling.
The runtime environment may provide a kernel that is new enough to
handle the calls, even if the toolchain doesn't.
Karel Zak [Wed, 2 Aug 2023 10:57:37 +0000 (12:57 +0200)]
column: fix -l
The original implementation is complicated and broken.
It's possible to calculate the rest of the string (for the last
column) from the current position rather than calculate it
continuously. Use the last wcstok() result also means that it will
work as expected independently on "greedy" mode (skips repeating
separators.
# printf 'a b c d\n1 2 3 4\n' | ./column -t -o '|' -l3
a|b|c d
1|2|3 4
Karel Zak [Thu, 3 Aug 2023 11:06:30 +0000 (13:06 +0200)]
libmount: improve EPERM interpretation
In some cases mount(2)/open_tree(2) returns EPERM for root user. In
this case libmount reports it as "mount point is not a directory".
It does not makes sense for bind mount where target could be a
regular file.
This patch is not ideal, the error handler is generic, but semantic
for new mount API and mount(2) is different. For example now it checks
for regular file, but the new API supports bind over symlinks, so
proper fix will require lstat() and S_ISLNK(), etc. We need to move
error messages to hook_mount.c and mount_mount_legacy.c to make it
more specific.
Fixes: https://github.com/util-linux/util-linux/issues/2413 Signed-off-by: Karel Zak <kzak@redhat.com>
Karel Zak [Thu, 3 Aug 2023 10:39:19 +0000 (12:39 +0200)]
libmount: ifdef statx() call
In this case the statx() is use to get mount ID. It's optional and not
required. Let's #ifdef the statx() call and also check for stx_mnt_id
struct member.
Fixes: https://github.com/util-linux/util-linux/issues/2415 Signed-off-by: Karel Zak <kzak@redhat.com>
Markus Mayer [Tue, 1 Aug 2023 19:59:27 +0000 (12:59 -0700)]
libmount: check for struct statx
Let's ensure struct statx exists before we try to use it. Checking for
the existence of linux/stat.h is not sufficient. This is because
[uapi/]linux/stat.h has existed since Linux 3.7, however struct statx
was only introduced with Linux 4.11.
The problem arises if one happens ot be using kernel headers from within
the aforementioned range, such as Linux 4.9.
CC libmount/src/la-utils.lo
In file included from libmount/src/utils.c:31:
./include/fileutils.h:100:33: warning: declaration of 'struct statx' will not
be visible outside of this function [-Wvisibility]
unsigned int mask, struct statx *stx)
^
libmount/src/utils.c:117:16: error: variable has incomplete type 'struct statx'
struct statx stx = { 0 };
^
libmount/src/utils.c:117:10: note: forward declaration of 'struct statx'
struct statx stx = { 0 };
^
libmount/src/utils.c:125:5: error: use of undeclared identifier 'STATX_TYPE'
STATX_TYPE
^
libmount/src/utils.c:126:8: error: use of undeclared identifier 'STATX_MODE'
| STATX_MODE
^
libmount/src/utils.c:127:8: error: use of undeclared identifier 'STATX_INO'
| STATX_INO,
^
1 warning and 4 errors generated.
make[4]: *** [Makefile:11269: libmount/src/la-utils.lo] Error 1
Checking for the presence of struct statx explicitly avoids this
problem.
Markus Mayer [Tue, 1 Aug 2023 19:59:26 +0000 (12:59 -0700)]
libmount: check for linux/mount.h
Check that linux/mount.h is present before using features that rely on
it.
It is necessary to check for the presence of linux/mount.h explicitly.
Simply relying on the presence of several syscalls (e.g. SYS_move_mount,
SYS_open_tree) does not work reliably, at least not in a cross-compiling
environment.
The syscall definitions are provided by the C library. Meanwhile,
linux/mount.h is provided by the kernel. This opens the possibility for
discrepancies.
A problem arises if the C library (e.g. musl-1.2.3) defines all "mount
fd API" syscalls (and HAVE_MOUNTFD_API becomes true), but the kernel
headers are old enough to not provide linux/mount.h. The resulting error
looks as follows. This example is using an LLVM-13 cross-compiler from
x86_64 to aarch64 with musl-1.2.3 as the C library.
CC libmount/src/la-hooks.lo
In file included from libmount/src/hooks.c:30:
./include/mount-api-utils.h:11:10: fatal error: 'linux/mount.h' file not found
^~~~~~~~~~~~~~~
1 error generated.
make[4]: *** [Makefile:11185: libmount/src/la-hooks.lo] Error 1
To prevent this condition, we add a check to configure that will test
for the presence of linux/mount.h in addition to testing for the mount
fd API. Only if both conditions are met can we actually use the mount
fd API.
If the systemcall is not available ENOSYS is returned.
Under glibc the statx implementation also has its own fallback logic.
As AT_STATX_DONT_SYNC can't be implemented correctly in that fallback
logic the wrapper will return EINVAL in case the emulation is needed and
AT_STATX_DONT_SYNC is set.
So also use our own fallback in that case.
Fixes: #2409 Signed-off-by: Thomas Weißschuh <thomas@t-8ch.de>
Karel Zak [Tue, 1 Aug 2023 12:49:48 +0000 (14:49 +0200)]
libmount: cleanup --fake mode
It was originally designed to play nasty games with /etc/mtab (mount
when /etc is read-only and update later when it's write-able).
The --fake is completely useless with the new API due to complexity
where we cannot skip any step, because the next stuff depends on it.
So, it makes more sense skip all functionality where libmount does
anything significant.
This commit add --fake check to hooks logic to skip all hooks as the
hooks are place where libmount implements mount related invasive
operations (create mountpoint, namespaces, create superblock, move,
mount, etc.).
Frankly, --fake without mtab is useless.
Fixes: https://github.com/util-linux/util-linux/issues/2395 Signed-off-by: Karel Zak <kzak@redhat.com>
Debarshi Ray [Thu, 29 Jun 2023 11:04:41 +0000 (13:04 +0200)]
libmount: update documentation for MNT_ERR_APPLYFLAGS
The implementation using the new FD based mount kernel API (ie.,
fsconfig/fsopen) uses MNT_ERR_APPLYFLAGS for failed mount_setattr(2)
calls, which involves more mount attributes (eg., MOUNT_ATTR_RDONLY,
MOUNT_ATTR_NOSUID, etc.) in addition to the MS_PROPAGATION flags (eg.,
MS_SHARED, MS_UNBINDABLE, etc.).
Note that mount_setattr(2) is part of the new FD based mount kernel API,
and is not used by the classic mount(2) based version.
Karel Zak [Thu, 20 Jul 2023 10:34:14 +0000 (12:34 +0200)]
libmount: use some MS_* flags as superblock flags
The old mount(2) API usually utilizes MS_* flags to set up the VFS
node. However, there are some exceptions like "sync" (MS_SYNCHRONOUS),
where the flag is used (by kernel) for the superblock instead. The new
API addresses this issue, ensuring that these options are used for
fsconfig().
This commit introduces MNT_SUPERBLOCK to identify these options in the
libmount options Linux map, and it enforces the new mount code to
utilize these options for fsconfig(FSCONFIG_SET_FLAG).
Reported-by: Abbink Esger <esger.abbink.ext@siemens.com> Signed-off-by: Karel Zak <kzak@redhat.com>
(cherry picked from commit 54e4a6b145fd6ef943d93e16de748283e687855d) Fixes: #2439
Karel Zak [Mon, 5 Jun 2023 10:59:41 +0000 (12:59 +0200)]
libmount: use mount(2) for remount on Linux < 5.14
It seems mount_setattr() is supported on Linux < 5.14, but it's without
MOUNT_ATTR_NOSYMFOLLOW. That's problem for remount where we reset all
VFS flags.
The most simple (but not elegant) is to check for kernel version and
fallback to mount(2) on remount.
Addresses: https://github.com/util-linux/util-linux/issues/2283 Signed-off-by: Karel Zak <kzak@redhat.com>
Debarshi Ray [Thu, 13 Jul 2023 09:07:27 +0000 (11:07 +0200)]
libmount: handle failure to apply flags as part of a mount operation
If a mount operation with extra flags runs into an EPERM when applying
the flags, then mnt_context_get_mount_excode() returns 'Unknown error
5005' and MNT_EX_FAIL. Here's an example:
Create a mount point on the host with 'nodev,nosuid,noexec':
$ dd if=/dev/zero of=/var/tmp/loopfile bs=40960 count=1024
$ sudo losetup --find /var/tmp/loopfile
$ sudo mkfs.ext4 /dev/loop0
$ sudo mkdir /mnt/a
$ sudo mount -o nosuid,nodev,noexec /dev/loop0 /mnt/a
Check the mount options to be sure:
$ findmnt --output OPTIONS,PROPAGATION /mnt/a
OPTIONS PROPAGATION
rw,nosuid,nodev,noexec,relatime,seclabel shared
Enter a mount and user namespace:
$ podman run \
--interactive \
--privileged \
--rm \
--tty \
--volume /:/run/host:rslave \
registry.fedoraproject.org/fedora:38 \
/bin/bash
Try to bind mount the mount point from the host inside the namespace
with some extra flags:
# mkdir ~/b
# mount --bind -o ro /run/host/mnt/a ~/b
mount: /root/b: filesystem was mounted, but any subsequent operation
failed: Unknown error 5005.
# echo $?
32
It will be better to show something more human-readable than 'Unknown
error 5005'.
Secondly, an exit code of 32 means 'mount failure', which isn't quite
correct here. The mount operation is split into two mount(2) calls,
where the first one uses MS_BIND to create the bind mount, and the
second uses MS_REMOUNT | MS_BIND | MS_RDONLY to apply the 'ro' flag.
Here, the first mount(2) does succeed:
# findmnt --output OPTIONS,PROPAGATION ~/b
OPTIONS PROPAGATION
rw,nosuid,nodev,noexec,relatime,seclabel private,slave
It's only the application of the 'ro' flag with the second mount(2) that
fails with an EPERM. Hence, an exit code of 1 that means 'incorrect
invocation or permissions' seems more appropriate.
Signed-off-by: Debarshi Ray <rishi@fedoraproject.org>
projects / thirdparty / util-linux.git / log
