| Age | Commit message (Collapse) | Author |
|---|
|
* src/remove.c (remove_entry) ["can unlink directories"]: Fix a
mistakenly reversed condition.
* NEWS: Mention this bug fix.
Reported by Pieter Bowman.
|
|
* src/remove.c (fd_to_subdirp): Remove unused parameter.
(remove_cwd_entries, remove_dir): Update callers.
|
|
an extremely long name in the current directory.
* src/remove.c (full_filename_): Rename a local variable.
Use two separate "truncated" variables, in case the second
call to right_justify does not indicate any truncation -- which
would happen only if the single component FILENAME were longer
than 511 bytes and DIR_NAME were empty.
|
|
When operating on a relative name longer than 511 bytes,
and (when either processing a directory that is neither writable
nor readable (but still searchable) or when determining whether
to prompt), and encountering an ENOMEM error while forming the
file name, rm would operate on a truncated-to-511-byte name
starting with "[...]" rather than the intended one.
* NEWS: Describe the bugs.
* src/remove.c: Correct two misuses of full_filename:
(full_filename0, xfull_filename): New functions.
(full_filename_): Rewrite to use full_filename0.
(AD_pop_and_chdir): Use xfull_filename, not full_filename.
(write_protected_non_symlink): Likewise.
|
|
* src/remove.c (full_filename_): Save and restore errno.
Spotted by Bruno Haible.
|
|
A cross-partition move of a file in a sticky tmpdir and owned by
another user would evoke an invalid diagnostic after copying it:
mv: cannot remove `x': Operation not permitted
Either of the following (mv.c, remove.c) changes would fix the bug by
itself. I think it's slightly better to use both; the added cost is
minimal: mv: an extra lstat-per-mv-cmdline-arg-that-goes-cross-partition,
rm: an extra lstat-per-unlink-that-fails-w/EPERM.
* src/remove.c (remove_entry): Also lstat the file upon EPERM.
* src/mv.c (rm_option_init): Initialize root_dev_ino just as is done
in rm, so that a cross-partition invoked remove.c:rm call works the
same way as one invoked from the command-line use of "rm". That
setting of root_dev_ino makes rm() do the equivalent of an additional
lstat for each argument, which in turn gives rm enough information to
issue the right diagnostic.
* tests/mv/sticky-to-xpart (version): New file. Test for the above.
* tests/mv/Makefile.am (TESTS): Add sticky-to-xpart.
Arrange for "make check-root" to run the new root-only test.
* tests/Makefile.am (tb): New target, to run the new root-only test.
(all_t): Add tb.
* src/c99-to-c89.diff: Adjust offsets.
|
|
On some systems (those with openat et al), when rm would fail to
remove a symlink, it would fail with the misleading diagnostic,
"Too many levels of symbolic links".
* NEWS: Mention the bug fix.
* src/remove.c (is_nondir_lstat): New function.
(remove_entry): Use it to catch failed-to-remove symlink (and any
other non-dir) here so that we don't fall through and try to treat
it as directory, which -- with a symlink -- would provoke the bogus
ELOOP failure.
* tests/rm/fail-eacces: Add a test for the above.
* src/c99-to-c89.diff: Adjust offsets.
|
|
Don't perform a "."-relative lstat, when the file in question
may well not be in ".". Although this is a bug, a few attempts
to exercise it on a linux-2.6.22 system failed. You probably need
a pre-openat system to trigger the failure. The consequence of this
bug would be a lower-quality diagnostic upon failed dir removal.
* src/remove.c (is_dir_lstat): Add a parameter, fd_cwd.
Use it instead of hard-coding AT_FDCWD.
(remove_entry): Call is_dir_lstat with fd_cwd.
Signed-off-by: Jim Meyering <jim@meyering.net>
|
|
* src/remove.c (push_dir): Don't copy trailing slashes onto the stack.
Reported by François Pinard.
* tests/rm/v-slash: New file. Test for the above change.
* tests/rm/Makefile.am (TESTS): Add v-slash.
|
|
|
|
|
|
with 'rm -r dir' (without -f), if we are root, or if we are removing
a directory tree that is full of symbolic links.
* bootstrap.conf (gnulib_modules): Add write-any-file.
* src/copy.c: Include write-any-file.h.
(UNWRITABLE): Remove macro, replacing with....
(writable_destination): New function, which uses can_write_any_file
to avoid the need for euidaccess when we are privileged.
(overwrite_prompt, abandon_move): Use it.
* src/remove.c: Include write-any-file.h.
(D_TYPE): New macro.
(DT_UNKNOWN, DT_DIR, DT_LNK) [!HAVE_STRUCT_DIRENT_D_TYPE]: New macros.
(write_protected_non_symlink): Don't bother to stat if we can write
any file.
(prompt): New arg PDIRENT_TYPE. All callers changed.
Use readdir dirent type to avoid the need for 'lstat' on each directory
entry in cases like 'rm -r dir', if we are root, or if the tree is
full of symbolic links.
(DT_IS_KNOWN, DT_MUST_BE): Remove.
(remove_entry): New arg DIRENT_TYPE_ARG. All callers changed.
|
|
* src/remove.c (write_protected_non_symlink): Return int, not bool,
so that we can indicate failure too (as a postive error number).
(prompt): If write_protected_non_symlink fails, report that error
number and fail rather than charging ahead and removing the dubious
entry. Redo the logic of printing a diagnostic so that we need to
invoke quote (full_filename (...)) only once. More details at:
<http://thread.gmane.org/gmane.comp.gnu.coreutils.bugs/9952/focus=9996>
|
|
* NEWS: Mention this.
* src/remove.h (enum rm_interactive): New ternary type.
(struct rm_options) [interactive]: Use it, here -- rather than bool.
* src/remove.c (prompt): Reflect type change.
* src/mv.c (rm_option_init): Initialize to RMI_NEVER now.
* src/rm.c (main): Add a FIXME comment for '-d' option.
Adapt to type change of rm_options.interactive.
* tests/rm/i-never: New file. Test for the above fix.
* tests/rm/Makefile.am (TESTS): Add i-never.
|
|
Root-only test failure reported by Alex van Hout and Jon Grosshart in
<http://thread.gmane.org/gmane.comp.gnu.coreutils.bugs/9415/focus=9415>.
|
|
* src/remove.c (AD_pop_and_chdir): Change **DIRP parameter to *DIRP,
now that this function never modifies the pointer. Adjust comments
and code accordingly.
(remove_dir): Set "dirp" to NULL right after AD_pop_and_chdir call,
now that AD_pop_and_chdir no longer does that.
|
|
directory before opening ".", to avoid trouble with file system
implementations that cache readdir results at opendir-time.
* src/remove.c (AD_pop_and_chdir): Add a file descriptor parameter.
Don't update **DIRP. Don't call fdopendir here.
(remove_dir): Call fdopendir here instead.
Report and patch from Mikulas Patocka:
<http://lists.gnu.org/archive/html/bug-coreutils/2006-12/msg00170.html>
|
|
* src/remove.c (remove_entry): Handle EACCES for a non-directory, too.
Karl Berry reported that a cross-partition "mv /etc/issue ~"
failed with the um,... suboptimal diagnostic,
"mv: cannot remove `/etc/issue': Not a directory".
* tests/rm/Makefile.am (TESTS): Add fail-eacces.
* tests/rm/fail-eacces: New file.
* NEWS: Mention that both mv and rm are affected.
|
|
Update callers.
|
|
while trying to retain the readability of C99 as much as possible.
* src/remove.c (close_preserve_errno): Remove.
(fd_to_subdirp): Rewrite to avoid the need for decl after statement.
Signed-off-by: Paul Eggert <eggert@cs.ucla.edu>
|
|
Use an else clause in place of a "continue" statement.
Signed-off-by: Paul Eggert <eggert@cs.ucla.edu>
|
|
after statement.
Signed-off-by: Paul Eggert <eggert@cs.ucla.edu>
|
|
a pointer argument. All uses changed.
Signed-off-by: Paul Eggert <eggert@cs.ucla.edu>
|
|
Update the caller in remove_dir.
Signed-off-by: Paul Eggert <eggert@cs.ucla.edu>
|
|
Replace each of two uses with literal AT_FDCWD.
Signed-off-by: Paul Eggert <eggert@cs.ucla.edu>
|
|
Suggested by Steve McIntyre in <http://bugs.debian.org/392925>.
* src/remove.h (struct rm_options) [one_file_system]: New member.
* src/rm.c (rm_option_init): Initialize it.
(usage): Document the option.
* src/mv.c (rm_option_init): Likewise.
* src/remove.c (remove_dir): With --one-file-system and --recursive,
for each directory command line argument, do not affect a file system
different from that of the starting directory. And give a diagnostic.
* src/rm.c (ONE_FILE_SYSTEM): New enum.
(main): Handle new option.
* tests/rm/one-file-system: Test the above.
* tests/rm/Makefile.am (TESTS): Add one-file-system.
* tests/Makefile.am (check-root): Add the rm/one-file-system
test to the list.
(EXTRA_DIST): Add other-fs-tmpdir.
* tests/mv/setup: Removed. Renamed to...
* tests/other-fs-tmpdir: ...this new file.
* tests/mv/Makefile.am (EXTRA_DIST): Remove setup.
* tests/mv/acl: Reflect renaming: use ../other-fs-tmpdir.
* tests/mv/backup-is-src: Likewise.
* tests/mv/hard-link-1: Likewise.
* tests/mv/leak-fd: Likewise.
* tests/mv/mv-special-1: Likewise.
* tests/mv/part-fail: Likewise.
* tests/mv/part-hardlink: Likewise.
* tests/mv/part-rename: Likewise.
* tests/mv/part-symlink: Likewise.
* tests/mv/partition-perm: Likewise.
* tests/mv/to-symlink: Likewise.
* tests/mv/into-self-2: Likewise.
[doc/ChangeLog]
* coreutils.texi (rm invocation): Describe --one-file-system.
|
|
HAVE_WORKING_O_NOATIME and HAVE_WORKING_O_NOFOLLOW, too.
(usage): Output info about noatime and nofollow only if
they are known to work.
* src/remove.c (AD_push): Inspect HAVE_WORKING_O_NOFOLLOW rather
than O_NOFOLLOW, when testing whether it's possible to avoid a
race condition reliably.
|
|
Paul Eggert pointed out that the specified file may exist,
in spite of such an errno value.
* tests/rm/Makefile.am (TESTS): Remove ignore-name-too-long.
* tests/rm/ignore-name-too-long: Remove file.
|
|
of a missing file under some very unusual conditions (with errno
being any of ENOENT, ENOTDIR, ENAMETOOLONG).
|
|
* src/remove.c (ignorable_missing): New function.
Use it everywhere, rather than open-coding the test.
Andreas Schwab reported the ENOTDIR problem.
(ignorable_missing): Similarly, don't fail for ENAMETOOLONG.
* NEWS: Mention the bug fix.
* tests/rm/ignorable: New file. Test for the ENOTDIR case.
* tests/rm/ignore-name-too-long: New file. Test for ENAMETOOLONG.
* tests/rm/Makefile.am (TESTS): Add the new file names.
|
|
Since any system may be affected by the Darwin readdir bug,
perform the extra rewinddir unconditionally. The performance
impact of rewinding a directory is negligible.
* src/remove.c (NEED_REWIND): Define to use
CONSECUTIVE_READDIR_UNLINK_THRESHOLD unconditionally.
[m4/ChangeLog]
* readdir.m4: Remove file once again.
* jm-macros.m4: Remove reference to gl_FUNC_READDIR.
|
|
and NFS, whereby rm would not remove all files in a directory.
* src/remove.c (CONSECUTIVE_READDIR_UNLINK_THRESHOLD): Reduce to 10.
(NEED_REWIND): New macro, so that we incur the cost of the work-around
rewinddir only on afflicted systems.
* NEWS: Clarify and correct.
* tests/rm/readdir-bug: New file. Test for the above fix.
* tests/rm/Makefile.am (TESTS): Add it.
Prompted by testing and analysis from Bruno Haible:
http://lists.gnu.org/archive/html/bug-coreutils/2006-09/msg00326.html
|
|
it from removing a directory containing 188 or more entries.
* src/remove.c (CONSECUTIVE_READDIR_UNLINK_THRESHOLD): Decrease by
20, go work around the buggy readdir on Darwin 8.6.1 with NFS.
Reported by Matthew Woehlke.
|
|
it's own line.
(rm_1): Move declaration of "st" and new cache_stat_init call
"down" to nearer where they're used.
|
|
(cache_fstatat, cache_stat_init): New functions.
(cache_statted, cache_stat_ok): New functions.
(write_protected_non_symlink): Remove struct stat ** buf_p arg,
which is no longer needed with the new functions. All callers
changed.
(prompt, is_dir_lstat, remove_entry, remove_dir):
New struct stat * arg. All callers changed.
(write_protected_non_symlink, prompt, is_dir_lstat, remove_entry):
(remove_cwd_entries, remove_dir, rm_1):
Use and maintain the file status cache.
(prompt, remove_entry): Omit the first "directory" in the diagnostic
"Cannot remove directory `foo': is a directory". This causes "rm"
to pass a test case that it would otherwise fail now that it
"knows" more about its argument. I think the diagnostic is better
without the first "directory" anyway.
(prompt): Remove the no-longer-needed IS_DIR arg; all callers changed.
(rm_1): Reject attempts to remove /, ./, or ../.
|
|
(dot_or_dotdot): ...new static inline function.
* src/remove.c (rm_1): Reflect this renaming.
* src/ls.c (basename_is_dot_or_dotdot): Likewise.
|
|
(AD_mark_helper): Free malloc'd filename if hash_insert says
that string is already in the hash table.
|
|
being recorded.
* src/remove.c (remove_cwd_entries): Don't call cycle_check here.
(AD_push): Call it from here instead.
|
|
* src/remove.c (AD_stack_clear): New function.
(rm_1): Use it.
(AD_pop_and_chdir): Free *prev_dir just before longjmp.
|
|
|
|
fd_to_subdirp failure, not just when errno == EACCES.
* src/remove.c (remove_dir): Use unlinkat-with-AT_REMOVEDIR, not
rmdir, here, even though rmdir may happen to be adequate.
|
|
* src/remove.c (remove_cwd_entries): If we can't open a directory,
and the failure is not being ignored, try to remove the directory
with rmdir (aka unlinkat-with-AT_REMOVEDIR), in case it's empty.
Problem report and test case from Paul Eggert in
<http://article.gmane.org/gmane.comp.gnu.core-utils.bugs/7425>.
* tests/rm/empty-inacc: New test, for the above.
|
|
|
|
|
|
CYCLE_CHECK_REFLECT_CHDIR_UP, rather than open-coding it.
|
|
could cause the removal of that directory to fail with an erroneous
diagnostic about a directory cycle. Reported by Vineet Chadha.
(AD_pop_and_chdir): If the directory we're about to
leave (and try to rmdir) is the one whose dev_ino is being used to
detect a cycle, reset cycle_check_state.dev_ino to that of the parent.
|
|
|
|
(close_preserve_errno): New function.
(fd_to_subdirp): Don't print a diagnostic in this function.
Do it from the callers instead, unless rmdir succeeds.
(remove_cwd_entries, remove_dir): Adjust callers.
|
|
First off, the attribute should have been `volatile' (not static)
to avoid longjmp-related risk of clobber. Secondly, now there is
no longer any risk of a local variable being clobbered, so there's
no need for any attribute at all.
|
|
(AD_pop_and_chdir): Use gotos to avoid some duplication.
(AD_push): Rewrite an assertion so that the entire computation
goes away when assertions are turned off.
|
