On 04/12/2024 11:55, Pádraig Brady wrote:
On 02/12/2024 16:19, Göran Uddeborg wrote:When using "tail --follow=name", but without "--retry", on a file supporting inotify, the command doesn't finish if the file is moved to a new name.Repeat this way on a local filesystem: echo apa > apa tail --follow=name apa & mv apa bepa "tail" will print an error message saying "No such file or directory", but it will continue running. I would expect a following message "no files remaining" and "tail" to exit. If the file system doesn't support inotify, if "apa" is on NFS for example, "tail" exits as expected. If the file is removed rather than moved ("rm apa") "tail" also exits as expected. This is tested on Fedora 41 using coreutils-9.5-11.fc41.x86_64 kernel-6.11.6-300.fc41.x86_64Yes that is a bug. The info docs state this: "In that [renamed file] case, use ‘--follow=name’ to track the named file, perhaps by reopening it periodically to see if it has been removed and recreated by some other program. Note that the inotify-based implementation handles this case without the need for any periodic reopening." But that description only alludes to the implementation, and not any functional difference I would say. I.e. inotify should not imply --retry. Interestingly if you `rm depa` before recreating "apa" in the example above, then tail does exit. Also tail will exit immediately if "apa" is not present at startup. Also tail will exit if you move "apa" to a different directory. All of those existing behaviors gives us latitude to adjust this behavior I think, so that tail does in fact exit for the case above with inotify. BTW one can test the non-inotify behavior on any file system with the undocumented ---disable-inotify option.
I'll push the attached later to fix this. Marking this as done. thanks, Pádraig
From fd01fc8075a0df4e9036fdb962b589e60c134e26 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A1draig=20Brady?= <p...@draigbrady.com> Date: Wed, 4 Dec 2024 19:40:55 +0000 Subject: [PATCH] tail: ensure --follow=name unfollows renamed files Require --retry to continue to track files upon rename. We already unfollowed a file if it was renamed to another file system (unlinked), so this makes the behavior consistent if renaming to a file in the same file system. I.e. --follow=name without --retry, means unfollow if the name is unlinked or moved, so this change ensures that behavior for all rename cases. Related commits: v8.0-121-g3b997a9bc, v8.23-161-gd313a0b24 * src/tail.c (tail_forever_notify): Remove watch for a renamed file if --retry is not specified. * tests/tail/F-vs-rename.sh: Related test cleanup. * tests/tail/follow-name.sh: Add a test case. * NEWS: Mention the bug fix. Fixes https://bugs.gnu.org/74653 --- NEWS | 4 ++++ src/tail.c | 5 +++-- tests/tail/F-vs-rename.sh | 10 +++++----- tests/tail/follow-name.sh | 13 ++++++++++--- 4 files changed, 22 insertions(+), 10 deletions(-) diff --git a/NEWS b/NEWS index 5bd9199e4..c1e604ffa 100644 --- a/NEWS +++ b/NEWS @@ -28,6 +28,10 @@ GNU coreutils NEWS -*- outline -*- 'shuf' generates more-random output when the output is small. [bug introduced in coreutils-8.6] + `tail --follow=name` no longer waits indefinitely for watched + file names that are moved elsewhere within the same file system. + [bug introduced in coreutils-8.24] + 'tail -c 4096 /dev/zero' no longer loops forever. [This bug was present in "the beginning".] diff --git a/src/tail.c b/src/tail.c index 33acf9ab3..8cefb9c07 100644 --- a/src/tail.c +++ b/src/tail.c @@ -1812,10 +1812,11 @@ tail_forever_inotify (int wd, struct File_spec *f, size_t n_files, if (ev->mask & (IN_ATTRIB | IN_DELETE | IN_DELETE_SELF | IN_MOVE_SELF)) { /* Note for IN_MOVE_SELF (the file we're watching has - been clobbered via a rename) we leave the watch + been clobbered via a rename) without --retry we leave the watch in place since it may still be part of the set of watched names. */ - if (ev->mask & IN_DELETE_SELF) + if (ev->mask & IN_DELETE_SELF + || (!reopen_inaccessible_files && (ev->mask & IN_MOVE_SELF))) { inotify_rm_watch (wd, fspec->wd); hash_remove (wd_to_name, fspec); diff --git a/tests/tail/F-vs-rename.sh b/tests/tail/F-vs-rename.sh index df2422985..2b66033b8 100755 --- a/tests/tail/F-vs-rename.sh +++ b/tests/tail/F-vs-rename.sh @@ -63,26 +63,26 @@ for mode in '' '---disable-inotify'; do { cat out; fail=1; } # Wait up to 12.7s for "x" to be displayed: file='b' data='x' retry_delay_ check_tail_output .1 7 || - { echo "$0: b: unexpected delay?"; cat out; fail=1; } + { echo "$0: b: unexpected delay 1?"; cat out; fail=1; } echo x2 > a # Wait up to 12.7s for this to appear in the output: # "tail: '...' has appeared; following new file" tail_re='has appeared' retry_delay_ check_tail_output .1 7 || - { echo "$0: a: unexpected delay?"; cat out; fail=1; } + { echo "$0: a: unexpected delay 2?"; cat out; fail=1; } # Wait up to 12.7s for "x2" to be displayed: file='a' data='x2' retry_delay_ check_tail_output .1 7 || - { echo "$0: a: unexpected delay 2?"; cat out; fail=1; } + { echo "$0: a: unexpected delay 3?"; cat out; fail=1; } echo y >> b # Wait up to 12.7s for "y" to appear in the output: file='b' data='y' retry_delay_ check_tail_output .1 7 || - { echo "$0: b: unexpected delay 2?"; cat out; fail=1; } + { echo "$0: b: unexpected delay 4?"; cat out; fail=1; } echo z >> a # Wait up to 12.7s for "z" to appear in the output: file='a' data='z' retry_delay_ check_tail_output .1 7 || - { echo "$0: a: unexpected delay 3?"; cat out; fail=1; } + { echo "$0: a: unexpected delay 5?"; cat out; fail=1; } cleanup_ done diff --git a/tests/tail/follow-name.sh b/tests/tail/follow-name.sh index 64864edfb..f92839c9e 100755 --- a/tests/tail/follow-name.sh +++ b/tests/tail/follow-name.sh @@ -23,13 +23,20 @@ cat <<\EOF > exp || framework_failure_ tail: cannot open 'no-such' for reading: No such file or directory tail: no files remaining EOF - returns_ 1 timeout 10 tail --follow=name no-such > out 2> err || fail=1 - # Remove an inconsequential inotify warning so # we can compare against the above error sed '/inotify cannot be used/d' err > k && mv k err - compare exp err || fail=1 +# Between coreutils 8.34 and 9.5 inclusive, tail would have +# waited indefinitely when a file was moved to the same file system +cleanup_() { kill $pid 2>/dev/null && wait $pid; } +touch file || framework_failure_ +timeout 10 tail --follow=name file & pid=$! +sleep .1 # Usually in inotify loop here +mv file file.unfollow || framework_failure_ +wait $pid +test $? = 1 || fail=1 + Exit $fail -- 2.47.0
