mv: create cross-device directory entries relative to the destination descriptor by abendrothj · Pull Request #14971 · uutils/coreutils

abendrothj · 2026-09-29T23:50:48Z

Based on #15123; only the last three commits are new here.

When a directory move falls back to copying across filesystems, each destination entry was created by joining a path, and regular files went through fs::copy. A symlink that appeared in the destination during the copy was followed, so content went wherever it pointed.

Each destination directory is now held open (search-only where possible), and files, subdirectories, symlinks, special files and hard links are created relative to it without following symlinks. A file, link or special file that shows up at a name is replaced; a directory there fails ("dest appeared"). Regular files share copy_file_data with the single-file fallback, which sets ownership and mode on the open descriptor, using libc fchown so fakeroot still works; util/check-libc-interposition.sh now checks this for mv. On Linux it copies with io::copy, so a move between btrfs subvolumes still reflinks, as fs::copy did, and single-file moves now do too. On macOS, fcopyfile copies the data and then the ACLs, extended attributes and times, which fs::copy kept for files in a directory and single-file moves didn't; if the second part fails, mv warns ("setting attributes for ...") and goes on, as for xattrs elsewhere. Source directory names are read before descending, so one descriptor stays open per level. Still by path: directory ownership, symlink xattrs, the hard link source, and the reopen that sets each directory's xattrs and mode.

On Linux, a 512 MiB file moved between two btrfs subvolumes takes 1 ms against 238 ms on main, as it now reflinks; a directory already did (0.9 ms on both). Across filesystems it is about the same: 4 x 128 MiB took 238 ms against 240 ms (btrfs to tmpfs) and 97.5 against 96.8 ms on macOS (APFS to a RAM disk); 5000 small files took 236 against 248 ms on Linux and 1580 against 1530 ms on macOS, where the separate metadata call costs about 3%.

test_mv_cross_device_dir_replaces_symlink_entry_created_mid_copy plants symlinks from a thread while 200 files are copied. It can't force the timing, so a run where none lands in time passes without proving anything; on Linux it failed 10 of 10 runs without the change and passed 30 of 30 with it, in about 0.02 s. A unit test checks the same replacement deterministically, with the symlink planted beforehand. The other new tests cover a deep tree under a low descriptor limit, an owner-read-masked directory and a write/search-only destination. Run on macOS and Linux.

Closes #14991

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

codspeed · 2026-09-30T00:19:56Z

Merging this PR will regress 1 benchmark

1 warning, 1 note

⚠️ 122 benchmarks are dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

ℹ️ 1 benchmark has no flamegraph

The flamegraph wasn't generated: most of the time was spent in runtime internals, which would leave nothing useful to analyze.

See affected benchmarks

⚡ 4 improved benchmarks
❌ 1 regressed benchmark
✅ 400 untouched benchmarks
⏩ 70 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	`BASE`	`HEAD`	Efficiency
❌	CPU Simulation	`ls_recursive_wide_tree[(10000, 1000)]`	68.1 ms	75 ms	-9.21%
⚡	CPU Simulation	`uniq_case_insensitive[10000]`	2.9 ms	2.8 ms	+5.11%
⚡	CPU Simulation	`uniq_with_count[10000]`	3.6 ms	3.5 ms	+4.13%
⚡	CPU Simulation	`uniq_heavy_duplicates[10000]`	4.2 ms	4 ms	+3.49%
⚡	CPU Simulation	`uniq_check_chars[(10000, "1")]`	4.2 ms	4.1 ms	+3.47%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

_{Comparing abendrothj:mv-cross-device-dir-dirfd (7f40614) with main (156d715)}

70 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

github-actions · 2026-09-30T00:44:15Z

GNU testsuite comparison:

Congrats! The gnu test tests/mv/mv-special-1 is no longer failing!

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

abendrothj · 2026-10-01T00:53:50Z

Opened #14991.

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

sylvestre · 2026-10-05T21:25:46Z

this is quite a large patch, can we make it smaller?

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

abendrothj · 2026-10-06T07:29:44Z

Smaller now: it reuses DirFd, replace_entry_at, link_at and copy_special_file from the PRs below it instead of its own per-entry helpers. Its own diff went from +588/-66 to 5 files, +490/-79, about 200 of them tests. Part of that is a late fix in 7e5ee68: ownership is set with libc fchown on the open descriptor, so fakeroot can intercept it, with a regression test whose prerequisites are in DEVELOPMENT.md. It sits on #15123, so the diff shown here includes the stack below it until that lands.

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

abendrothj · 2026-10-07T00:55:11Z

drafting this until #15120, #15122 and #13334 land; on its own it's +490/-79, about 200 of them tests.

abendrothj · 2026-10-22T08:33:42Z

Trimmed a bit more: the fakeroot test moved into util/check-libc-interposition.sh, and the clippy allow is gone. Its own diff is now +409/-96, 133 of it tests.

sylvestre · 2026-11-05T08:24:29Z

sorry but it needs to be rebased

sylvestre · 2026-11-05T08:25:03Z

+fn dest_appeared(path: &Path, e: io::Error) -> io::Error {
+    if e.kind() == io::ErrorKind::AlreadyExists {
+        io::Error::new(
+            io::ErrorKind::AlreadyExists,


two full impl DestDir blocks per platform is a lot of duplication, could the path fallback live in uucore instead?

There's one DestDir struct now, with two small create_dest_dir versions. The path versions are only for platforms where safe_traversal isn't built at all (Redox, Windows, WASI), so uucore has nowhere to put them; elsewhere DirAnchor already falls back to the path where it can't hold the directory open. Building a path-only DirAnchor on Redox would let mv drop them; I can do that in a follow-up if you'd like.

For ln -f/-sf, cp -lf/-sf and mv's cross-device symlink. Without O_PATH or O_SEARCH (OpenBSD, Hurd) it falls back to the directory's path.

ln and mv's cross-device move use it next. The no-unique-temp-name error is now translated.

On Apple targets mknodat is looked up at run time, as it exists only from macOS 13; before that the node is created by the directory's path.

Sockets and device nodes are recreated like fifos, with their mode, and the destination is replaced only once the new node exists. Fixes uutils#13145

Sockets and device nodes in the directory no longer fail the move, and fifos keep their mode.

Entries are created through the open destination directory, so a symlink that appears in it during the copy is replaced, not written through. Closes uutils#14991

Copilot AI balanced review requested due to automatic review settings September 29, 2026 23:50

Copilot AI reviewed Sep 29, 2026

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from d0cf9fe to f3fc847 Compare September 29, 2026 23:54

Copilot AI balanced review requested due to automatic review settings September 29, 2026 23:54

Copilot AI reviewed Sep 29, 2026

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from f3fc847 to e3fda2a Compare September 29, 2026 23:55

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from e3fda2a to 8d392d1 Compare September 30, 2026 02:35

Copilot AI balanced review requested due to automatic review settings September 30, 2026 02:35

Copilot AI reviewed Sep 30, 2026

Copilot AI balanced review requested due to automatic review settings October 1, 2026 01:14

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from 8d392d1 to 5e3cb57 Compare October 1, 2026 01:14

Copilot AI reviewed Oct 1, 2026

Cassian433 mentioned this pull request Oct 4, 2026

mv: keep copying xattrs after one fails in a cross-device move #15049

Merged

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from 5e3cb57 to 42731bd Compare October 5, 2026 06:13

Copilot AI balanced review requested due to automatic review settings October 5, 2026 06:13

Copilot AI reviewed Oct 5, 2026

This was referenced Oct 5, 2026

mv: no warning when the destination file system doesn't support xattrs (GNU cp-mv-enotsup-xattr, 9.12) #15079

Closed

mv: warn when the destination doesn't support xattrs #15101

Merged

Copilot AI balanced review requested due to automatic review settings October 6, 2026 07:15

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from 42731bd to 7e5ee68 Compare October 6, 2026 07:15

Copilot AI reviewed Oct 6, 2026

Copilot AI balanced review requested due to automatic review settings October 6, 2026 07:58

Copilot AI reviewed Oct 6, 2026

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from b125862 to b3b9dde Compare October 7, 2026 00:54

Copilot AI balanced review requested due to automatic review settings October 7, 2026 00:54

Copilot AI reviewed Oct 7, 2026

abendrothj marked this pull request as draft October 7, 2026 00:55

abendrothj force-pushed the mv-cross-device-dir-dirfd branch 2 times, most recently from 52022fc to 6903ee6 Compare October 12, 2026 10:05

kitten-lily mentioned this pull request Oct 15, 2026

Evaluate uutils and other changes in Ubuntu 26.10 starlit-os/krytis#1158

Open

abendrothj force-pushed the mv-cross-device-dir-dirfd branch 3 times, most recently from d767c93 to 4d87582 Compare October 22, 2026 08:33

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from 4d87582 to e4aae21 Compare October 22, 2026 08:50

sylvestre reviewed Nov 5, 2026

View reviewed changes

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from e4aae21 to 0b6fdcb Compare November 7, 2026 06:33

abendrothj added 4 commits November 10, 2026 18:18

uucore: let replace_link work in unreadable directories

f5642c0

For ln -f/-sf, cp -lf/-sf and mv's cross-device symlink. Without O_PATH or O_SEARCH (OpenBSD, Hurd) it falls back to the directory's path.

uucore: move replace_link's atomic replace into DirAnchor

e5f5eae

ln and mv's cross-device move use it next. The no-unique-temp-name error is now translated.

uucore: add DirFd::mknod_at

7a3c044

On Apple targets mknodat is looked up at run time, as it exists only from macOS 13; before that the node is created by the directory's path.

uucore: give DirAnchor mkdir, chmod, open, unlink and rename

901707f

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from 0b6fdcb to f0b4cef Compare November 11, 2026 04:22

abendrothj added 5 commits November 10, 2026 20:53

mv: recreate special files when moving across devices

2cfb03e

Sockets and device nodes are recreated like fifos, with their mode, and the destination is replaced only once the new node exists. Fixes uutils#13145

mv: recreate special files in directories moved across devices

7abea17

Sockets and device nodes in the directory no longer fail the move, and fifos keep their mode.

uucore: let DirAnchor open subdirectories and create files in them

37bb1d6

uucore: add copy_xattrs_fd_ignore_unsupported

03ef6e7

mv: create cross-device directory entries relative to the destination

7f40614

Entries are created through the open destination directory, so a symlink that appears in it during the copy is replaced, not written through. Closes uutils#14991

abendrothj force-pushed the mv-cross-device-dir-dirfd branch from f0b4cef to 7f40614 Compare November 11, 2026 04:54

Conversation

abendrothj commented Sep 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

codspeed Bot commented Sep 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will regress 1 benchmark

Performance Changes

Footnotes

Uh oh!

github-actions Bot commented Sep 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

abendrothj commented Oct 1, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

sylvestre commented Oct 5, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

abendrothj commented Oct 6, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

abendrothj commented Oct 7, 2026

Uh oh!

abendrothj commented Oct 22, 2026

Uh oh!

sylvestre commented Nov 5, 2026

Uh oh!

Uh oh!

sylvestre Nov 5, 2026

Choose a reason for hiding this comment

Uh oh!

abendrothj Nov 7, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

abendrothj commented Sep 29, 2026 •

edited

Loading

codspeed Bot commented Sep 30, 2026 •

edited

Loading

github-actions Bot commented Sep 30, 2026 •

edited

Loading