* tests/ls/stat-free-color.sh: Add newfstatat to the list
of syscalls to trace. Also add all "stat" syscalls to the
list of syscalls that we verify that strace supports.
Also only create a single dangling symlink to check, since
we already only check for a single "stat" call.
Fixes http://bugs.gnu.org/16075 seen on AArch64
* tests/misc/shuf-reservoir.sh: Restrict the valgrind
"exit on leak" behavior to developer environments where
specific "lint" code is enabled to avoid inconsequential leaks.
Previously, the test triggered another error diagnostic:
shuf: invalid input range ‘-e’
and therefore eclipsed the expected one:
shuf: cannot combine -e and -i options
While at it, reindent a line with more than 80 characters, present
since the previous commit to silence sc_long_lines.
* tests/misc/shuf.sh: Pass a valid range to the -i option.
Original problem reported by Philipp Thomas in
<http://bugs.gnu.org/16061>.
* NEWS: shuf --repeat, not shuf --repetitions.
* doc/coreutils.texi (shuf invocation):
* src/shuf.c (usage, long_opts, main):
* tests/misc/shuf.sh:
Likewise. Also, the default head-count is infinity.
Since v8.21-172-g33660b4, df not only treats symbolic link arguments
differently, as stated there, but now generally processes special file
arguments in a non-canonicalized form correctly:
$ cd /dev && df-old sdb
Filesystem 1K-blocks Used Available Use% Mounted on
devtmpfs 1014572 48 1014524 1% /dev
$ cd /dev && df-new sdb
Filesystem 1K-blocks Used Available Use% Mounted on
/dev/sdb 10190136 6039532 3609932 63% /home
Document df's new behavior.
* doc/coreutils.texi (df invocation): In the paragraph describing
df's behavior regarding special file arguments, relax the condition
for such special files from "... is an absolute name of ..." to
"... resolves to ...".
* NEWS (Bug fixes): Mention the new behavior also here.
This could trigger on SELinux systems where we build --qithout-selinux
or where the SELinux development libraries are not installed.
* init.cfg (require_selinux_enforcing_): Call require_selinux_()
to determine if the current build supports SELinux. This avoids
a false failure in tests/mkdir/selinux.sh where only mkdir would
determine that SELinux was disabled and thus ignore invalid contexts.
(require_selinux_): Refactor a little to distinguish whether it's
the build or the (file) system that doesn't support SELinux.
* src/selinux.h (ignorable_ctx_err): A new function used
to determine if a warning should be given after a call
to defaultcon() or restorecon().
* src/cp.c (main): Fix the setfscreatecon() call to use
the argument passed by the user.
* src/mkdir.c (make_ancestor): Show all but "ignoreable" errors
from defaultcon() and restorecon().
* tests/misc/selinux.sh: Add a test run as root in selinux enforcing
mode, to ensure cp --context=invalid is honored and fails immediately.
libcrypto is generally available and has well optimized
crypto hash routines particular to various systems.
For example, testing sha1sum with openssl-1.0.0j
on an i3-2310M, gives a performance boost of about 40%:
$ time sha1sum.old --tag ~/test.iso
SHA1 (/home/padraig/test.iso) = 3c27f7ed01965fd2b89e22128fd62dc51a3bef30
real 0m4.692s
user 0m4.499s
sys 0m0.162s
$ time sha1sum.new --tag ~/test.iso
SHA1 (/home/padraig/test.iso) = 3c27f7ed01965fd2b89e22128fd62dc51a3bef30
real 0m2.685s
user 0m2.512s
sys 0m0.170s
* configure.ac: By default, enable use of libcrypto if available.
* src/local.mk: Link with libcrypto.
* NEWS: Mention the md5sum and sha*sum improvements.
* src/df.c (get_disk): Use the same heuristic used in
get_point() to select the shortest file system mount point,
in an attempt to show the base mounted file system.
* NEWS: Mention the bug fix.
This is so the matching for the device is done on the canonical name
of the disk node, rather than on the path of the symlink.
In any case the user will generally want to use the symlink target.
* src/df.c (get_disk): Canonicalize the passed file,
before matching against the list of mounted file system devices.
Note we pass the original symlink name to the "file" output field,
as the symlink target is usually available through the "source" field.
* tests/df/df-symlink.sh: Test the dereferencing operation.
* tests/local.mk: Mention the new test.
* NEWS: Mention the fix.
Reported by Ondrej Oprala
Note tests/init.sh and bootstrap are still in sync with gnulib.
* gnulib: Sync two configure check fixes.
- Avoid generating core dumps from regex configure check
- Fix compile error in getcwd configure check
Note tests/init.sh and bootstrap are still in sync with gnulib.
* gnulib: Sync many fixes/changes, including the base64
encoding speedup mentioned in commit v8.21-145-g9120845
Also included is support for enabling use of libcrypto hash routines
which are generally well optimized for particular systems.
Add a new rule to ensure the use of quote() instead of '%s' or `%s'
in format strings of diagnostics messages.
* cfg.mk (sc_prohibit_quotes_notation): Add rule.
* TODO: Remove the entry regarding the '%s' notation.
* src/mkfifo.c (main): Remove the offending and in this case even
duplicate quoting in the format string of the error diagnostic.
* src/mknod.c (main): Likewise.
* src/df.c (decode_output_arg): Change two invocations of error()
according to the above new rule.
* src/numfmt.c: Fix numerous wrong quote notations to fit the above
new rule, mostly in internal debugging diagnostic messages.
* doc/coreutils.texi (ls invocation): Remove the note about
OSX terminals not aligning properly as this is no longer the case.
Tested by: Philipp Thomas
* src/longlong.h: Sync with the latest longlong.h from libgmp to:
- avoid arm asm when being compiled for the thumb instruction [sub]set
- avoid old powerpc assembly that is incompatible with newer GCC
- add arm64 optimized count_trailing_zeros()
- add sparc64 optimized add_ssaaaa() and umul_ppmm()
* src/extent-scan.c (extent_need_sync): Remove the FIXME comment about
removing the work around. As discussed in [1], the needed FIEMAP fix
never made it into the Linux kernel.
* src/realpath.c (longopts): Remove the FIXME comment about deprecating
the --strip option as it's a valid alias for --no-symlinks option.
Also discussed in [1].
[1] http://lists.gnu.org/archive/html/coreutils/2013-11/msg00103.html
This option has been undocumented for 12 years [1], and warned
about for a year [2].
[1] commit FILEUTILS-4_1_4-23-gd177203
[2] commit v8.17-43-g453ce92
* src/df.c (MEGABYTES_OPTION): Remove.
(long_options): Remove "megabytes" element.
(main): In the option parsing loop, remove the MEGABYTES_OPTION case.
* NEWS: Mention the change.
* src/selinux.c: Don't include the system "fts.h" as
that disallows _FILE_OFFSET_BITS=64 which gnulib auto enables
to support large files on 32 bit systems. Instead include
our "xfts.h" which includes the less limited gnulib replacement,
and also a checked version of xfts_open().
(restorecon): Use the checked xfts_open() rather than the standard
fts_open().
Prompted by the continuous integration build failure at:
http://hydra.nixos.org/build/6934169
Handle both newer selinux libraries with mode_to_security_class(),
and systems without selinux at all. We could easily adjust
gnulib to provide the necessary stubs for use by this module,
but it's more efficient to just stub out the module completely,
when not using selinux.
* src/selinux.h: Define stubs for the two module functions,
when SELinux is not available.
* src/selinux.c: Exclude all logic in preference for the stubs
when selinux isn't used. Also when newer selinux libs are used,
don't use our conflicting static version of mode_to_security_class().
m4/jm-macros.m4: Check for the system mode_to_security_class().
* src/selinux.c: This module introduced in commit v8.21-159-gd8e27ab
doesn't need to include <selinux/flask.h>. That header file
isn't catered for by gnulib, but is not needed as we're not
explicitly referencing any class IDs.
Prompted by the continuous integration build failure at:
http://hydra.nixos.org/build/6920020
cp, mv, install, mkdir, mkfifo, mknod are adjusted so that:
-Z no longer accepts an argument.
-Z or --context without an argument do not warn without SELinux.
--context with an argument will warn without SELinux.
* src/local.mk: Reference the new selinux module where required.
* src/system.h: Make the argument to --context optional.
* src/mkdir.c: Likewise. Also handle the SMACK case for --context.
Note we currently silently ignore -Z with SMACK.
* src/mkfifo.c: Likewise.
* src/mknod.c: Likewise.
* src/install.c: Likewise. Note install(1) by default already
set the context for target files to their system default,
albeit with an older method. Use the -Z option to select between
the old and new context restoration behavior, and document
the differences and details for how context restoration
is done in new and old methods, with a view disabling the
old method entirely in future.
* src/cp.c: Make the argument to --context optional.
Note -Z implies --no-preserve=context. I.E. -Z overrides
that aspect of -a no matter what order specified.
(struct cp_options): Document the context handling options.
(main): Check/adjust option combinations after all
options are processed, to both simplify processing
and to make handling independent of order of options
on the command line. Also improve the diagnostics
from a failed call to setfscreatecon().
(set_process_security_ctx): A new function,
refactored to set the default context from the source file,
or with the type adjusted as per the system default for
the destination path.
(set_file_security_ctx): A new function refactored to
set the security context of an existing file, either based on
the process context or the default system context for a path.
(copy_internal): Use the refactored functions to simplify
error handling and consistently fail or warn as needed.
(copy_reg): Likewise.
(copy_internal): With --preserve=context, also copy
context from non regular files. Note for directories this may
impact the copying of subsequent files to that directory?
(copy_attr): If we're handling SELinux explicitly,
then exclude to avoid the redudant copy with --preserve=context,
and the problematic copy with -Z. Note SELinux attribute exclusion
also now honors cp -a --no-preserve=context. Note there was a
very small window over 10 years ago, where attr_copy_file was
available, while attr_copy_check_permissions was not, so we
don't bother adding an explicit m4 check for the latter function.
* src/mv.c: Support --context, but don't allow specifying an argument.
* src/chcon.c: Adjust a comment to be specific to SELinux.
* src/runcon.c: Likewise.
* src/copy.c: Honor the context settings to "restorecon" as appropriate.
* src/copy.h: Add a new setting to select "restorecon" functionality.
* tests/mkdir/selinux.sh: s/-Z/--context=/
* tests/cp/cp-a-selinux.sh: Augment this test with cases
testing basic -Z functionality, and also test the various
invalid option combinations and option precedence.
* tests/mkdir/restorecon.sh: Add a new test for the
more involved mkdir -Z handling, since the directory changing
and non existent directories need to be specially handled.
Also check the similar but simpler handling of -Z by mk{nod,fifo}.
* tests/local.mk: Reference the new test.
* doc/coreutils.texi (cp invocation): Update as per interface changes.
(mv invocation): Likewise.
(install invocation): Likewise.
(mkfifo invocation): Likewise.
(mknod invocation): Likewise.
(mkdir invocation): Likewise.
* NEWS: Mention the new feature and change in behavior.
* src/selinux.c: A new module implementing "restorecon" functionality.
There are two main functions to adjust the type of the referenced
file system item. defaultcon() will setup the process context so
that new items will have the required context without races. This is
the preferred method. For existing files, the equivalent restorecon()
is available which has two modes. With the "local" parameter set to
false, restorecon() will adjust the type according to the system
configuration for that file, and set to true will update the context
as per the context for the current process (disregarding type).
* src/selinux.h: Likewise.
* po/POTFILES.in: Reference the new module.
A sync operation is very often expensive. For illustration
I timed the following python script which indicated that
each ext4 dir sync was taking about 2ms and 12ms, on an
SSD and traditional disk respectively.
import os
d=os.open(".", os.O_DIRECTORY|os.O_RDONLY)
for i in range(1000):
os.fdatasync(d)
So syncing for each character for each file can result
in significant delays. Often this overhead is redundant,
as only the data is sensitive and not the file name.
Even if the names are sensitive, your file system may
employ synchronous metadata updates, which also makes
explicit syncing redundant.
* tests/misc/shred-remove.sh: Ensure all the new parameters
actually unlink the file.
* doc/coreutils.texi (shred invocation): Describe the new
parameters to the --remove option.
* src/shred.c (Usage): Likewise.
(main): Parse the new options.
(wipename): Inspect the new enum to see which of
the now optional tasks to perform.
* NEWS: Mention the new feature.
* THANKS.in: Add reporter Joseph D. Wagner
Previous behavior failed to read contents of a (re)appearing file,
when symlinked by tail's watched file. Also we now diagnose other
edge cases when running in inotify mode, where an initially
missing or regular file changes to a symlink.
* src/tail.c (main): If any arg is a symlink, use polling mode.
(recheck): Diagnose the edge case where a symlink appears during
inotify processing.
* tests/tail-2/symlink.sh: Test the fix. Mention the edge cases.
* tests/local.mk: Reference the new test.
* NEWS: Mention the fix.
Reported by: Ondrej Oprala
* src/df.c (usage): Document the new 'file' --output field.
(get_dev): Add a new parameter to pass the specified
argument from the command line through. Use '-' if a
command line parameter is not being used.
* doc/coreutils.texi (df invocation): Describe the new 'file' field.
* tests/df/df-output.sh: Adjust all fields test, and
add a specific test for --output=file.
* NEWS: Mention the new feature.
* src/timeout.c (cleanup): When calling settimeout() from
this signal handler, ensure we don't call out to error()
or gettext(), which are not async-signal-safe.
Also reset the errno which may be cleared by settimeout().
* tests/local.mk (factor-tests:) Add -f to the mv command that
replaces any existing generated tests. This is required to avoid
prompts when root initially generates the tests, and they subsequently
need to be regenerated by a non root user.
* src/sort.c: (async_safe_die): A new limited version of error(),
that outputs fixed strings and unconverted errnos to stderr.
This is safe to call in the limited context of a signal handler,
or in this particular case, between the fork() and exec() of
a multithreaded process.
(move_fd_or_die): Use the async_safe_die() rather than error().
(maybe_create_temp): Likewise.
(open_temp): Likewise.
Fixes http://bugs.gnu.org/15970
* src/ls.c (usage): Mention -k only changes the display
for disk usage (directory total, and ls -s), and imply
that it can be overridden (by --block-size, and -h).
* doc/coreutils.texi (block size): Mention that ls -k
handling is different to other utilities.
Addresses http://bugs.gnu.org/14525
Recent commit 2da7009d changed the error diagnostic of rm(1) trying
to remove "." or "..". Enhance the corresponding test.
* tests/rm/r-4.sh: Ensure rm(1) outputs the expected error diagnostic.
With newer perl, "make syntax-check" issues many warnings like:
-i used with no filenames on the command line, reading from STDIN.
* cfg.mk (sc_check-AUTHORS): Remove the -i flag in the perl
invocation as it is reading from a pipe.
The error diagnostic
"rm: cannot remove directory: '.'"
does not give the user a hint for the reason.
Issue a clearer error message.
* src/remove.c (rm_fts): Enhance the error diagnostic in the above
case to emphasize that skipping is done deliberately.
In the corresponding comment, mention that POSIX mandates this
behavior. Likewise in the subsequent comment for skipping "/".
* doc/coreutils.texi (rm invocation): In the paragraph describing
the above behavior, mention that POSIX mandates it.
Fix a recent regression introduced in commit v8.21-127-g5ee7d8f
Also related to this is the recent query about root run `rm -I`
ignoring the mode bits of a file: https://bugzilla.redhat.com/1013171
* tests/rm/interactive-once.sh: Avoid the messages and
corresponding file presence checks with write protected files
when running as root.
This adds support for using a constant or "stick" parity bit.
* src/stty.c (usage): Mention the new flag.
* tests/misc/stty.sh: Adjust for the new flag.
* NEWS: Mention the improvement.
* docs/coreutils.texi (stty invocation): Mention the new flag.
Since the I/O overhead is significant to the relatively
simple processing done by this utility, use fputs() rather
than fputc() to output '\n'.
Time to process a 100MiB file was measured to
decrease from 0.417s to 0.383s, i.e. an 8% improvement.
Related to these changes, is a processing improvement in
gnulib, which increases throughput by 60% when processing
full buffers, which improves processing of a 100MiB file
with standard wrapped output, down to 0.256s.
http://git.sv.gnu.org/gitweb/?p=gnulib.git;a=commit;h=43fd1e7b
Also increase the encoding buffer size from 3 to 30KiB.
This was seen to give a further 8% improvement, taking
processing time down to 0.235s in the wrapped output case.
The decoding size buffer is not adjusted,
due to the noted caveat with --ignore-garbage.
* src/base64.c (BLOCKSIZE): Split into ENC_ and DEC_ variants,
with the former increased from 3KiB to 30KiB.
(wrap_write): Use the simpler fputc() rather than fputs()
to output the '\n' character. Also check against EOF
rather than < 0 for errors.
(do_encode): Likewise.
* NEWS: Mention the large increase in performance, which
with the I/O improvements in coreutils and the processing
improvement in gnulib, amount to about a 60% throughput increase.
* src/shred.c (dopass): Exit early to avoid redundant heap
allocation, and more importantly avoiding a file sync
when we're writting no data, as this can have side effects.
Also with --verbose, this avoids printing of "pass status"
which could be confusing as to whether data was actually written.
* tests/misc/shred-passes.sh: Ensure the status for data
passes are not written when not doing any data writes.
* src/shred.c (dopass): In the periodic pattern case increase the
I/O block size from 12KiB to 60KiB (also a multiple of 3 and 4096).
* NEWS: Adjust accordingly.
Since direct I/O is now enabled with commit v8.21-139-gebaf961
we must handle the case where we write an odd size at the
end of a file (with --exact), or we specify an odd --size that
is larger than 64KiB, or in the very unlikely case of a device
with an odd size. This issue was present since direct I/O
support was first added in v5.3.0, but latent since v6.0.
Theoretically this could have also been an issue after that on
systems which didn't have alignment constraints, but did have
size constraints for direct I/O.
* src/shred.c (dopass): On the first pass for a file, always
retry a write that fails with EINVAL, so we handle direct I/O
failure at either the start or end of the file. Adjust the comment
as the original case is out of date and implicitly handled
by this more general fix.
* tests/misc/shred-exact.sh: Add a test case.
* NEWS: Add a "bug fix" entry for shred since there are
two related issues now fixed.
* src/copy.c (create_hard_link): Add a bool 'dereference' parameter,
and pass AT_SYMLINK_FOLLOW as 'flags' to linkat() when dereference
is true.
(should_dereference): Add new 'bool' function to determine if a
file should be dereferenced or not.
(copy_internal): Use the above new should_dereference() and remember
its return value in a new local bool 'dereference' variable. Use that
in all three calls to create_hard_link().
* src/cp.c (main): after parsing the options, if x.dereference is
still DEFEF_UNDEFINED and the x.recursive is true, then only set
x.dereference to DEREF_NEVER iff --link was not specified.
* doc/coreutils.texi (cp invocation): Mention that cp(1) does not
follow symbolic links in the source when --link is specified.
Likewise in the description of the -R option when used together with
that option.
* tests/cp/same-file.sh: Adapt the expected results for the -fl,
the -bl and the -bfl tests.
* tests/cp/link-deref.sh: Add a new test.
* tests/local.mk (all_tests): Reference the above new test.
* NEWS (Changes in behavior): Mention the change.
This fixes http://bugs.gnu.org/15173
Co-authored-by: Bernhard Voelker <mail@bernhard-voelker.de>
Commit v5.92-1057-g43d487b introduced a regression
in coreutils 6.0 where it removed the page alignment
of the buffer to write, thus disabling direct I/O.
We want to use direct I/O when possible to avoid
impacting the page cache at least, as we know we don't
want to cache the data we're writing.
* src/shred.c (dopass): Allocate the buffer on the heap,
while using a more general calculation to allow to have
the output size independent from the fillpattern() size
constraint of a multiple of 3. Also we dispense with the
union as it's no longer needed given we're aligning on
a page boundary and thus don't need to explicitly handle
uint32_t alignment.
* src/md5sum.c (usage): s/three/four/ in the message pertaining
to the --check related options. Also clarify that --strict
is just significant for the formatting of the checksum lines.
Also since we're changing both strings, move the --strict description
in with the description of the other options and order alphabetically.
* THANKS.in: Added reporter: Daniel Mach
* src/md5sum.c (main): Add a comment as to why we continue
to escape names that do not have '\n' but do have '\\' chars.
(print_filename): Use the predetermined boolean to decide
whether to escape or not, so that in the common case we
can output the file name directly, rather than inspecting each char.
* tests/misc/md5sum.pl: Add case to show '\\' chars cause escaping.
* tests/misc/sha1sum.pl: Likewise.
Although the above man pages depend on src/md5sum.c as a shared
source, the build of the man pages directly requires their own
executables to exist.
* man/local.mk (man/sha1sum.1): Change the dependency from
'src/md5sum' to 'src/sha1sum'.
(man/sha224sum.1): s/md5sum/sha224sum/
(man/sha256sum.1): s/md5sum/sha256sum/
(man/sha384sum.1): s/md5sum/sha384sum/
(man/sha512sum.1): s/md5sum/sha512sum/
Reported by Pádraig Brady in
http://lists.gnu.org/archive/html/coreutils/2013-11/msg00006.html
* README-prereq: Update as per the latest required versions
in bootstrap.conf. Also add a missing cd command.
Reported by Aaron Davies
Fixes http://bugs.gnu.org/15612
* doc/coreutils.texi (cp invocation): Give more detail about what's
happening in the example, explicitly calling out the --no-dereference
option required to make the -H and -L options significant.
Also mention the option order significance of the -H and -L options.
Fixes http://bugs.gnu.org/15579
