Kiran Rangoon [Mon, 29 Dec 2025 03:51:00 +0000 (22:51 -0500)]
tests: correct UUID timestamp test expectations
Update expected timestamps for additional test UUIDs to match actual
output. These test cases provide coverage for various timestamp ranges
including pre-epoch, historical, and far-future dates.
Signed-off-by: Kiran Rangoon <kiranrangoon0@gmail.com> Reviewed-by: Thomas Weißschuh <thomas@t-8ch.de>
Kiran Rangoon [Mon, 29 Dec 2025 03:50:59 +0000 (22:50 -0500)]
libuuid: fix timestamp overflow for pre-1970 dates
Use int64_t arithmetic to correctly handle timestamps before Unix epoch.
This fixes the overflow that was causing UUID timestamps from 1582 to
appear as far-future dates. Update test expectations accordingly.
Signed-off-by: Kiran Rangoon <kiranrangoon0@gmail.com> Reviewed-by: Thomas Weißschuh <thomas@t-8ch.de>
Masatake YAMATO [Mon, 29 Dec 2025 07:22:24 +0000 (16:22 +0900)]
lslocks: special-case PID for filtering (skip −1)
The PID column may be printed as −1 for locks that are not tied to a
process or are inaccessible to lslocks (e.g., permission constraints).
The libsmartcols filter engine does not support negative integers, so
PID = −1 cannot be compared reliably.
Attach the PID as cell userdata only when PID >= 0, and use that value
for filter evaluation. Rows with PID = −1 are thus excluded from PID
comparisons.
Masatake YAMATO [Mon, 15 Dec 2025 21:41:59 +0000 (06:41 +0900)]
lslocks: make SIZE filterable by normalizing to bytes
The SIZE column may be printed in raw bytes or human-readable
units (KB/MB, etc.), which makes number-based comparison impossible.
Normalize SIZE to a byte value and store it in the cell via
libsmartcols `scols_cell_set_userdata`. When evaluating filter
expressions, use the stored byte value instead of the printed SIZE.
This enables consistent numeric comparisons (e.g., -Q 'SIZE > 1M')
regardless of the output format.
Masatake YAMATO [Mon, 15 Dec 2025 20:15:30 +0000 (05:15 +0900)]
lslocks: (refactor) use global bytes in get_json_type_for_column()
`get_json_type_for_column()` used to take `representing_in_bytes`, but
every caller passed the global `bytes`. The parameter added no value
and split the ownership of the setting. Drop the parameter and read
the global `bytes` instead.
This aligns lslocks with lsblk and prepares for adding `-Q/--filter`.
lsblk also reads global `bytes` defined in lsblk.c
Karel Zak [Wed, 17 Dec 2025 11:06:35 +0000 (12:06 +0100)]
Merge branch 'feat/su_accept_gid_args' of https://github.com/cgoesche/util-linux-fork
* 'feat/su_accept_gid_args' of https://github.com/cgoesche/util-linux-fork:
unshare: use MAX_OF_UINT_TYPE instead of (type)-1
chfn: enable the use of the username or UID
unshare: use the new ul_get{grp,userpw}_str() routines
login: use new xgetuserpw() instead of xgetpwnam()
chsh: use new xgetuserpw() instead of xgetpwnam()
su: accept group name and GID in -g and -G options
tests: (su) test GID argument in --group option
lib: (pwdutils.c) new library routines to get a group/passwd struct by name or GID/UID
include/c.h: add MAX_OF_UINT_TYPE macro to get max num of an uint type
chmem: do not word a configuration failure as an instruction to the user
When the current machine does not support configuring/deconfiguring
memory, simply report that it cannot be done, instead of seemingly
telling the user to skip the configuring/deconfiguring.
Also, there is no need to mention `chmem` in the message: mentioning
just the recommended option (-e or -d) is enough. Also, consistently
terminate all four messages with a period.
This improves upon commit ab2e709600 ("chmem: improve messages").
Chris Webb [Mon, 15 Dec 2025 18:48:29 +0000 (18:48 +0000)]
unshare: fix user namespace bind mounts
unshare --user=<file> always fails because we no longer have CAP_SYS_ADMIN
in the parent user namespace after unsharing to create the new one. As with
unshare --mount=<file>, fork a child to make the bind mount instead.
Chris Webb [Mon, 15 Dec 2025 18:41:08 +0000 (18:41 +0000)]
unshare: remove get_mnt_ino() check in bind_ns_files_from_child()
get_mnt_ino() was originally introduced in c84f2590 where it was used
in a loop to wait for the parent process to unshare the mount namespace
before binding the namespace in its child.
The parent and child processes are now synchronised with eventfd, so
remove this vestigial check and the now-unused get_mnt_ino() function.
This allows bind_ns_files_from_child() to be used even when the mount
namespace isn't amongst the namespaces being unshared.
Chris Webb [Sun, 14 Dec 2025 23:18:33 +0000 (23:18 +0000)]
unshare: add --owner to set user namespace owner uid and gid
As well as the mappings between lower and upper ids, a user namespace is
associated with an owner user and group in its parent. These are set from
the uid and gid when the unshare() call is made, and determine which user
in the parent namespace has CAP_SYS_ADMIN in the child and can setns()
into it.
Add an --owner=<uid>:<gid> option which allows a privileged user to
create a user namespace on behalf of another user, mapping parent ids
and/or bind-mounting the namespace with privileges that the new owner
would not have.
Simplify the control flow around map_ids_from_child() vs mapping them
inline to avoid too many special cases. We reset mapuser and mapgroup to
-1 to signal that the mapping has been delegated to the child helper.
For completeness, we maintain the semantics of --map-root-user and
--map-current-user, binding the invoking user to root or itself in the
new namespace. However, when --owner is used, these must be handled by
a forked child as with --map-users and --map-groups.
Karel Zak [Thu, 11 Dec 2025 12:39:40 +0000 (13:39 +0100)]
fallocate: refactor --report-holes and --dig-holes output
This commit consolidates duplicate printf() calls and calculations by
introducing two new helper functions:
- update_holestat(): Tracks hole statistics and prints verbose output
for both HOLETYPE_FILE and HOLETYPE_DATA. Statistics are always
collected regardless of report mode.
- summary_holestat(): Prints summary output for file holes, data holes,
and "converted to sparse holes" message. Uses consistent formatting
with hole counts and percentages.
Changes:
- Made 'report' a global variable (like 'verbose')
- Renamed struct holes_report to struct holestat for consistency
- Changed struct fields to uintmax_t to eliminate casts in printf
- Removed 'report' parameter from dig_holes() function
- Unified dig and report modes to use single dig_holes() call
- All output strings wrapped in _() for translation
- Consistent printf() usage instead of mixed fprintf()
- Percentage format: "(%ju bytes, %.2f%% of the file)"
The refactoring eliminates ~40 lines of duplicate code while maintaining
all functionality and improving code maintainability.
Karel Zak [Wed, 10 Dec 2025 11:01:06 +0000 (12:01 +0100)]
Merge branch 'fix/consolidate_validate_pfd_ino' of https://github.com/cgoesche/util-linux-fork
* 'fix/consolidate_validate_pfd_ino' of https://github.com/cgoesche/util-linux-fork:
lib: (pidutils.c) use uint64_t instead of ino_t for seamless cross-compatibility
kill: use uint64_t as type for kill_control->pidfd_ino
lib: (pidfd-utils.c) remove extraneous _GNU_SOURCE feature test macro
lib: (pidfd-utils) new helper function to retrieve pidfd inode number
lib: (pidfd-utils.c) add a helper routine to check the pidfd fs type
kill: replace USE_KILL_WITH_PIDFD_INO ifdef with USE_PIDFD_INO_SUPPORT
include: (pidfd-utils.h) conditionally define pidfd inode support
libmount: add pidfs magic number for fstype check
include: (statfs_magic.h) add pidfs magic number
kill: use ul_get_valid_pidfd_or_err() to validate user provided pidfd inodes
include: add helper routines for opening and validating pidfds
Karel Zak [Tue, 9 Dec 2025 12:12:41 +0000 (13:12 +0100)]
libblkid: Keep NTFS name unmodified and mount driver independent
We need stable filesystem names (types) even though there are multiple
filesystem drivers. libmount now provides a way to map the stable
fs-types to various mount-types.
Karel Zak [Tue, 9 Dec 2025 12:02:11 +0000 (13:02 +0100)]
libmount: add option to override fs-type with mount-type
This patch introduces an internal libmount method to convert fs-type
(as provided by libblkid or udevd) to mount-type to specify a different
mount driver.
Currently, the mapping from fs-type to mount-type is hardcoded in
libmount as a temporary solution. The final implementation should
provide configuration files (e.g., /etc/mount/fs.d/<name>) for these
mappings.
The current default mapping is implemented only for NTFS. It can be
modified during compilation with:
./configure --with-ntfs-mounttype=TYPE
or
meson setup build -D ntfs-mounttype=TYPE
The default is "ntfs3".
Addresses: https://github.com/util-linux/util-linux/pull/3618
Addresses: https://github.com/systemd/systemd/pull/39982 Signed-off-by: Karel Zak <kzak@redhat.com>
Karel Zak [Tue, 25 Nov 2025 11:56:07 +0000 (12:56 +0100)]
libmount: read from udev, add --disable-libmount-udev-support
The library traditionally uses libblkid to obtain device properties
(such as FS-type if not specified). This can be a relatively costly
operation to scan the device and requires read access to the device.
All relevant libblkid information is usually cached by the udev DB.
This commit adds the possibility to reuse the information from udev,
with a fallback to libblkid if udev is not available.
Karel Zak [Mon, 8 Dec 2025 08:59:41 +0000 (09:59 +0100)]
Merge branch 'fix/login_shell_arg_null_str' of https://github.com/cgoesche/util-linux-fork
* 'fix/login_shell_arg_null_str' of https://github.com/cgoesche/util-linux-fork:
login: duplicate --shell argument to avoid nulling through explicit_bzero()
include: add helper routines for opening and validating pidfds
With the new ul_get_valid_pidfd_or_err() routine util-linux
tools can now simply validate pidfd inode numbers before
opening a file descriptor for a given PID.
Signed-off-by: Christian Goeschel Ndjomouo <cgoesc2@wgu.edu>
Karel Zak [Thu, 4 Dec 2025 13:26:40 +0000 (14:26 +0100)]
Merge branch 'feat/login_shell_option' of https://github.com/cgoesche/util-linux-fork
* 'feat/login_shell_option' of https://github.com/cgoesche/util-linux-fork:
login: document -s and --shell on the man page
login: define shell to log in to with -s or --shell
Karel Zak [Wed, 3 Dec 2025 10:40:57 +0000 (11:40 +0100)]
Merge branch 'fix/issue_3865' of https://github.com/lord2y/util-linux
* 'fix/issue_3865' of https://github.com/lord2y/util-linux:
login-utils, sys-utils: use _PATH_BSHELL consistently
*: use ul_default_shell() for interactive shell spawning
lib: introduce ul_default_shell() for consistent shell resolution
Karel Zak [Wed, 3 Dec 2025 10:14:13 +0000 (11:14 +0100)]
Merge branch 'feat/check_bash_comp_integrity' of https://github.com/cgoesche/util-linux-fork
* 'feat/check_bash_comp_integrity' of https://github.com/cgoesche/util-linux-fork:
bash-completion: (lslogins) fix typo in long option
tools: (checkcompletion.sh) test the integrity of long options completion
tools: new helper for extraction of program long options from source files
Martin Minkus [Wed, 3 Dec 2025 03:51:05 +0000 (19:51 -0800)]
libfdisk: modernize ZFS GPT type description
The GPT type GUID 6A898CC3-1DD2-11B2-99A6-080020736631 is currently
described as "Solaris /usr & Apple ZFS". This reflects early Solaris
and Apple experiments with ZFS, but today the same GUID is widely used
by OpenZFS implementations on Linux, illumos, FreeBSD, and other
platforms to mark ZFS pool member partitions.
Apple's ZFS work was discontinued long ago, while OpenZFS has become
the actively maintained and de facto standard implementation. Update
the human-readable description to the simpler and more accurate:
"ZFS pool member"
A short comment is added to note the GUID’s historical Solaris /usr
origin and its brief use by Apple. Only the description string is
changed; the GUID itself and its semantics remain unchanged.
unshare: use the new ul_get{grp,userpw}_str() routines
This change refactors get_group() and get_user(), so that it
uses the new routines ul_getgrp_str() and ul_getuserpw_str(),
to simplify the code and remove the overkill mem allocations.
Signed-off-by: Christian Goeschel Ndjomouo <cgoesc2@wgu.edu>
Alessandro Ratti [Sun, 23 Nov 2025 15:26:37 +0000 (16:26 +0100)]
login-utils, sys-utils: use _PATH_BSHELL consistently
Remove local DEFAULT_SHELL definitions and hardcoded "/bin/sh" strings
in favor of the standard _PATH_BSHELL macro from <paths.h>. This
provides consistency across the codebase while following libc
conventions.
These tools already perform their own passwd lookups and only need a
fallback value, so they don't require the full ul_default_shell()
resolution logic.
Affected tools:
- su(1): already checks pw_shell validity
- sulogin(8): emergency login with explicit shell handling
- setpriv(1): already has passwd entry for environment setup
Alessandro Ratti [Sun, 23 Nov 2025 15:24:52 +0000 (16:24 +0100)]
*: use ul_default_shell() for interactive shell spawning
Update tools that spawn interactive shells to use ul_default_shell()
for consistent shell resolution. This ensures these tools respect both
$SHELL and the user's configured shell from the passwd database before
falling back to _PATH_BSHELL.
Affected tools:
- script(1): fixes history truncation when invoked without $SHELL
- scriptlive(1): consistent with script(1) behavior
- flock(1): for -c command execution
- more(1): for shell escape feature
- exec_shell (used by unshare(1) and nsenter(1))
This change addresses user reports of data loss due to tools defaulting
to /bin/sh instead of the user's configured shell, particularly affecting
command history with different HISTSIZE configurations.
Alessandro Ratti [Sun, 23 Nov 2025 15:20:57 +0000 (16:20 +0100)]
lib: introduce ul_default_shell() for consistent shell resolution
Add a new ul_default_shell() function to provide consistent shell
resolution across util-linux tools. The function follows a priority
order: $SHELL environment variable, user's shell from passwd database,
and finally _PATH_BSHELL as fallback.
The function supports flags to control its behavior:
- UL_SHELL_NOENV: skip $SHELL environment variable check
- UL_SHELL_NOPWD: skip passwd database lookup
This addresses the issue where tools like script(1) would default to
/bin/sh without respecting the user's configured shell, potentially
causing data loss.
Addresses: https://github.com/util-linux/util-linux/issues/3865 Suggested-by: Karel Zak <kzak@redhat.com> Suggested-by: Thomas Weißschuh <thomas@t-8ch.de> Signed-off-by: Alessandro Ratti <alessandro@0x65c.net>
projects / thirdparty / util-linux.git / log
