Like the optimization to avoid always-failing getfilecon calls,
this change avoids always-failing queries for whether a file has
a nontrivial ACL and for whether a file has certain "capabilities".
When such a query fails for one file (indicating no support), we know it
will always fail that way for the affected device. With this change, we
have thus eliminated nearly all failing-unsupported getxattr syscalls.
* src/ls.c (has_capability) [!HAVE_CAP]: Set errno to ENOTSUP.
(errno_unsupported): Expand the list of E* errno values to match
that of lib/acl-internal.h's ACL_NOT_WELL_SUPPORTED macro.
(file_has_acl_cache, has_capability_cache): New functions.
(gobble_file): Use them in place of non-caching ones.
* NEWS (Improvements): Mention it.
Suggested by Sven Breuner in
http://thread.gmane.org/gmane.comp.gnu.coreutils.general/2187
While eliminating most getfilecon calls saved about 33%,
eliminating these other calls can save almost all of the
remaining ~67% cost, on some remote file systems.
On systems or file systems without SELinux support, all getfilecon
and lgetfilecon calls would fail due to lack of support. We can non-
invasively cache such failure (on most recently accessed device) and
avoid the vast majority of the failing underlying getxattr syscalls.
* src/ls.c (errno_unsupported): New function.
(selinux_challenged_device): New file-scoped global.
(getfilecon_cache, lgetfilecon_cache): New error-caching wrapper
functions.
(gobble_file): Use the caching wrappers, for when many *getfilecon
calls would fail with ENOTSUP or EOPNOTSUPP.
Suggested by Sven Breuner in
http://thread.gmane.org/gmane.comp.gnu.coreutils.general/2187
Improved-by: Pádraig Brady.
Allow changing the --numeric-suffixes start number
from the default of 0.
* src/split.c (next_file_name): Initialize the suffix index
and the output filename according to start value.
(main): Check that the suffix length is large enough for the
numerical suffix start value.
* doc/coreutils.texi (split invocation): Mention it.
* NEWS (New features): Mention it.
* tests/split/numeric: New file. Test --numeric-suffixes[=FROM].
* tests/Makefile.am (TESTS): Reference the new test.
* doc/coreutils.texi (rm invocation): Mention that the -f option also
silences the message for missing operands, which is useful in scripts
e.g., for "rm -f $file_list" when $file_list is empty.
* src/rm.c (usage): Likewise.
Reported by Jérémy Magrin in http://bugs.gnu.org/10819
These edge cases were missed in the previous commit 140eca15c.
* src/dd.c (main): Include the bytes slop when truncating
without further I/O. Don't invalidate the whole file cache
in the case where 0 < count < ibs.
* tests/dd/bytes: Change to using the independent truncate
command to generate the file for comparison. Remove a redundant
test case and replace with one testing the truncation only logic.
dd now accepts the count_bytes and skip_bytes input flag and the
seek_bytes output flag, to more easily allow processing portions of a
file.
* src/dd.c (scanargs): Compute skip_records and skip_bytes when
'skip_bytes' iflag is used. Compute max_records and max_bytes when
'count_bytes' iflag is used. Compute seek_records and seek_bytes
when 'seek_bytes' oflag is used.
(skip_via_lseek): Use new 'bytes' parameter and handle potential
'records' equals to zero. Update the bytes parameter when called with
'fdesc' equal to STDOUT_FILENO. Update the header comments.
(dd_copy): Skip accordingly to skip_records AND skip_bytes. Count
accordingly to max_records AND max_bytes. Seek on output accordingly
to seek_records AND seek_bytes.
* NEWS (New features): Mention it.
* doc/coreutils.texi (dd invocation): Detail new flags and behaviors.
* tests/dd/bytes: New file. Tests for these new flags.
* tests/Makefile.am (TESTS): Add it.
But only when both A and B were hard links to the same symlink.
* src/copy.c (same_file_ok): Handle another special case: the one
in which we are moving a symlink onto a hard link to itself.
In this case, we must explicitly tell the caller to unlink the
source file. Otherwise, at least the linux-3.x kernel rename
function would do nothing, as mandated by POSIX 2008.
* tests/mv/symlink-onto-hardlink-to-self: New test.
* tests/Makefile.am (TESTS): Add it.
* NEWS (Bug fixes): Mention it.
Reported by Bernhard Voelker in http://bugs.gnu.org/10686
* src/pr.c (usage): Fix the -S description to indicate
that the argument is optional for the short option.
* doc/coreutils.texi (pr invocation): Likewise.
* src/od.c (usage): Fix the -S description to indicate
that the argument is required for the short option.
Clarify -w takes an argument and that it is optional.
Normally, mv detects a few subtle cases in which proceeding with a
same-file rename would, with very high probability, cause data loss.
Here, we have found a corner case in which one of these same-inode
tests makes mv refuse to perform a useful operation. Permit that
corner case.
* src/copy.c (same_file_ok): Detect/exempt this case.
* tests/mv/symlink-onto-hardlink: New test.
* tests/Makefile.am (TESTS): Add it.
* NEWS (Bug fixes): Mention it.
Initially reported by: Matt McCutchen in http://bugs.gnu.org/6960.
Raised again by Anders Kaseorg due to http://bugs.debian.org/654596.
Improved-by: Paul Eggert.
* src/realpath.c (path_common_prefix): Be consistent and
always include a leading '/' in the count returned.
(relpath): Account for the change in path_common_prefix()
and avoid outputting extra '/' chars in relative paths that
span the root dir.
* tests/misc/realpath: Add the two reported cases.
Reported by Mike Frysinger
* src/stat.c (usage): Indicate this is a transfer size
suggestion, rather than some persistent block size.
* doc/coreutils.texi (stat invocation): Likewise.
The previous commit introduced a couple of spacing issues,
luckily one of which caused a test to fail.
* src/stat.c (default_format): Add a space so times are aligned.
* src/tr.c (string2_extentd): Remove an extraneous space.
Add a rule to ding any source file that has a continued string
with a word in the first column of the following line.
Those tend to trigger malfunction in tools that try to map an
arbitrary line number to an enclosing function name. Of course,
very many strings do precisely this, *when they are part of the
usage function*. So we exempt the body of each usage function.
* src/dircolors.c (main): Separate a long, continued string
into two separately-quoted parts.
* src/od.c (decode_one_format): Likewise.
(decode_one_format, main): Move a space from end of
preceding line to the beginning of the continued line.
* src/tr.c (unquote, string2_extend, validate): Likewise.
* src/seq.c (main): Split in two and use string concatenation.
* src/stat.c (default_format): Use a mix of techniques.
* cfg.mk (sc_prohibit_continued_string_alpha_in_column_1): New rule.
Exempt three files in src: system.h, od.c, printf.c.
All affected lines end with \ or \n\, so run this command
until it produces no new changes (4 times):
git grep -E -l '`[^ ]+'\''.*\\' src \
|xargs perl -pi -e 's/`([^ ]+'\''.*\\)/'\''$1/'
* src/chroot.c (usage): Change ``...'' to '...', and describe the
default more accurately (also adding quotes): s,/bin/sh,'/bin/sh -i',
* src/join.c (usage): Change ` ...'' ' to "...''".
* src/fmt.c (isopen): Change `' to '` in list of bytes, so that
a subsequent change can safely perform the `...' to '...' conversion.
* src/truncate.c (main): Tweak quoting in comments to use '...',
not `...`, for consistency with the rest of comments in coreutils.
Automatically adjust both the source (now in only one place)
and all tests that expect the resulting output via this:
git grep -l 'Try.*--help' src/system.h tests \
| xargs perl -pi -e 's/Try \\?`(\S+ --help)/Try '\''$1/'
* src/split.c (lines_chunk_split): Fix logic bug that led to
unwarranted failure of "split -n l/2 /dev/zero" on NetBSD 5.1.
The same would happen when splitting a growing file, where
open/lseek-end gives one size, but by the time we read, there
is more data available.
(bytes_chunk_extract): Likewise.
* NEWS (Bug fixes): Mention this.
* tests/split/l-chunk: The latter case was not exercised.
Add code to do that.
Bug introduced with the chunk-selecting feature in v8.7-25-gbe10739.
Co-authored-by: Jim Meyering <meyering@redhat.com>
On systems with recent kernel/tools, a symlink from /etc/mtab to
/proc/mounts, and a by-UUID mount (i.e., soon, nearly everyone),
you will see something like the following when running "df -hT":
(this has been truncated to fit in a width-limited ChangeLog file)
Filesystem Type Siz...
rootfs rootfs 11G
udev devtmpfs 3.8G
tmpfs tmpfs 774M
/dev/disk/by-uuid/828fc648-9f30-43d8-a0b1-f7096a2edb66 ext4 11G
tmpfs tmpfs 1.6G
/dev/sda2 ext3 494M
/dev/sda5 ext4 12G
/dev/sda6 ext4 9.9G
Contrast that with what we're used to seeing (modulo the
two entries mounted on "/", which is a separate problem):
Filesystem Type Size Used Avail Use% Mounted on
rootfs rootfs 11G 1.9G 8.0G 19% /
udev devtmpfs 3.8G 0 3.8G 0% /dev
tmpfs tmpfs 774M 376K 774M 1% /run
/dev/sda3 ext4 11G 1.9G 8.0G 19% /
tmpfs tmpfs 1.6G 8.0K 1.6G 1% /run/shm
/dev/sda2 ext3 494M 78M 392M 17% /boot
/dev/sda5 ext4 12G 7.6G 3.7G 68% /usr
/dev/sda6 ext4 9.9G 6.6G 2.8G 71% /var
When that long /dev/disk/by-uuid/... name is merely a symlink
to a much shorter (and often more useful) device name like
"/dev/sda3", and when it's part of a listing of all file systems,
I would much prefer to see only the latter. Similarly, when using
an encrypted root file system, you would see a name like
/dev/mapper/luks-828fc648-9f30-43d8-a0b1-f7196a2edb66 pointing
to say, /dev/dm-0, I prefer the shorter name.
I.e., if I explicitly run
"df -hT /dev/disk/by-uuid/828fc648-9f30-43d8-a0b1-f7096a2edb66",
then, it's fine -- and expected -- to print to the long name.
It was explicitly given. However, with no non-option argument,
df should print the shorter name. Note that performing this
translation at a lower level (via a change to gnulib's mountlist.c)
would make it impossible to distinguish those two cases.
* src/df.c: Include "canonicalize.h".
(get_dev): Add a parameter, telling when we're in process-all-
mount-points mode; update all callers. When true, resolve
UUID-suffixed symlinks.
* NEWS (Changes in behavior): Mention it.
Reported by Dan Jacobson in http://bugs.gnu.org/10363
This program is compatible with other realpath(1)
implementations, and also incorporates relpath like support,
through the --relative options. The relpath support
was suggested by Peng Yu, who also provided an initial
implemenation of that functionality.
* AUTHORS: Add my name.
* NEWS: Mention the new command.
* README: Likewise.
* doc/coreutils.texi (realpath invocation): Add realpath info.
* man/Makefile.am (realpath.1): Add dependency.
* man/realpath.x: New template.
* man/.gitignore: Ignore generated man page.
* po/POTFILES.in: Add src/realpath.c.
* src/.gitignore: Exclude realpath.
* src/Makefile.am (EXTRA_PROGRAMS): Add realpath.
* src/realpath.c: New file.
* scripts/git-hooks/commit-msg: Add realpath to the list of prefixes.
* tests/Makefile.am (TESTS): Add misc/realpath.
* tests/misc/realpath: New file.
Without this change, gcc's -Werror=format would complain that
the '%lx' format requires 'long unsigned int', not 'int'.
* src/tail.c (fremote): Use a temporary variable.
* src/tail.c (start_lines): Do not form potentially-invalid address.
Use safe_read's return value as a pointer offset only after
ensuring that it is not SAFE_READ_ERROR (size_t)(-1).
Spotted by coverity.
Also, move declaration of "p" to be closer to first use.
* src/chgrp.c (usage): Group associated options together,
to aid users. Also minimize the differences between
individual messages across these four commands, to
aid translators.
* src/chmod.c: Likewise.
* src/chown.c: Likewise.
* src/chcon.c (usage): Likewise. Document the
--dereference option.
Suggested by Paul Eggert and Jari Aalto
Before, we would use inotify in that case, which would work as long
as updates were taking place locally, but not at all when remote.
Move hard-coded list of known remote FS types into a more
maintainable table in stat.c, alongside the list of FS
names and magic numbers. Generate a new is_local_fs_type function.
* src/Makefile.am (fs-is-local.h): New rule, generated file.
* src/extract-magic: Revamp to parse local/remote keyword after
each magic number in src/stat.c's case statements.
Accept new --local option.
* src/.gitignore: Ignore the generated fs-is-local.h.
* src/tail.c [HAVE_INOTIFY]: Include fs-is-local.h.
(fremote) [HAVE_INOTIFY]: Use the new function in place of
the switch stmt with hard-coded list of FS types.
Emit a warning when processing a file on a file system of unknown type.
* NEWS (Changes in behavior): Mention it.
Suggested by Sven Breuner.
* src/stat.c (human_fstype): Add a case: fhgfs, 0x19830326.
* src/tail.c (fremote): Add S_MAGIC_FHGFS.
* NEWS (Bug fixes): Update the entry for GPFS to mention FhGFS, too.
Reported by Sven Breuner.
This fixes Bug#10293, which I guess was introduced in commit
95c948b06a dated 2003-10-02.
* NEWS: Document fix.
* src/du.c (process_file): Don't count files in different file
systems if -x is given.
* tests/du/one-file-system: Test for this bug.
* src/ls.c (decode_switches): Replace our use of XARGMATCH
with open-coded version so that we can give a better diagnostic.
* tests/ls/time-style-diag: New file.
* tests/Makefile.am (TESTS): Add it.
Reported by Dan Jacobson in http://bugs.gnu.org/10253
with suggestions from Eric Blake and Paul Eggert.
* src/shred.c: Remove obsolete TODO comment. The first two and the
last item were done, adding --recursive (-r) is neither necessary
nor appropriate, and I don't want to add --interactive. I don't
see a need for the others. Prompted by comments from Amr Ali.
Starting with commit adc30a83, when using --color, ls inhibited
interrupts to avoid corrupting the state of an output terminal.
However, for very large directories, that inhibition rendered ls
uninterruptible for too long, including a potentially long period
even before any output is generated.
* src/ls.c: Two phases of processing are time-consuming enough that
they can provoke this: the readdir loop and the printing loop. The
printing was supposed to be covered by a call to process_signals in
(print_name_with_quoting): ... but that call was mistakenly guarded
by a condition that might be false for many or even all files being
processed. Call process_signals unconditionally.
(print_dir): Also call process_signals in the readdir loop.
* NEWS (Bug fixes): Mention it.
Reported by Arkadiusz Miśkiewicz in http://bugs.gnu.org/10243
Co-authored-by: Eric Blake <eblake@redhat.com>
* src/Makefile.am (fs_normalize_perl_subst, fs-magic, fs-kernel-magic):
Undo previous patch; it missed a \#.
(fs_normalize_perl_subst): Use \043 rather than \#.
\043 is portable to all ASCIIish platforms, whereas \# is portable
only to platforms that are compatible with GNU make (and are
incompatible with POSIX make). Porting this to EBCDIC is left as
an exercise for the reader....
* src/Makefile.am (fs_normalize_perl_subst): Don't make unportable
assumption about \# in the right hand side of a macro definition.
This works with GNU make, but not with POSIX make.
Problem reported by Basavaraj B (Bug#10220).
(fs-magic, fs-kernel-magic): Do the #-substitution here instead.
* src/test.c (unary_operator): gcc reported that initializations
in two case statements were skipped. Enclose in braces.
* src/od.c (decode_one_format): Likewise.
