Do this by searching a hash table instead of a linked list.
Problem reported by Martin Dørum in https://mort.coffee/home/tar/
via Gavin Smith in:
https://lists.gnu.org/r/bug-tar/2022-07/msg00003.html
* src/extract.c: Include hash.h.
Improve performance a bit on non-birthtime hosts
(struct delayed_link.has_predecessor): New member.
(delayed_link_head): Remove, replacing with ...
(delayed_link_table): ... this new variable. All uses
of linked list replaced with hash table.
(dl_hash, dl_compare): New functions for hash table.
(create_placeholder_file): Initialize has_predecessor.
(apply_delayed_link): New function, with body taken from
most of the old apply_delayed_link.
(apply_delayed_links): Use it. Respect has_predecessor.
Don’t bother freeing as we are about to exit.
* src/extract.c (HAVE_BIRTHTIME, BIRTHTIME_EQ): New macros.
(struct delayed_link, create_placeholder_file, extract_link)
(apply_delayed_links): Avoid unnecessary work on platforms
like GNU/Linux that lack birthtime.
* src/buffer.c, src/delete.c: Do not include system-ioctl.h.
* src/buffer.c (guess_seekable_archive): Remove. This is now done
by get_archive_status, in a different way.
(get_archive_status): New function that gets archive_stat
unless remote, and sets seekable_archive etc.
(_open_archive): Prefer bool for boolean.
(_open_archive, new_volume): Get archive status consistently
by calling get_archive_status in both places.
* src/buffer.c (backspace_output):
* src/compare.c (verify_volume):
* src/delete.c (move_archive):
Let mtioseek worry about mtio.
* src/common.h (archive_stat): New global, replacing ar_dev and
ar_ino. All uses changed.
* src/delete.c (move_archive): Check for integer overflow.
Also report overflow if the archive position would go negative.
* src/system.c: Include system-ioctl.h, for MTIOCTOP etc.
(mtioseek): New function, which also checks for integer overflow.
(sys_save_archive_dev_ino): Remove.
(archive_stat): Now
(sys_get_archive_stat): Also initialize mtioseekable_archive.
(sys_file_is_archive): Don’t return true if the archive is /dev/null
since it’s not a problem in that case.
(sys_detect_dev_null_output): Cache dev_null_stat.
doc: omit MS-DOS mentions in doc
It’s really FAT32 we’re worried about now, not MS-DOS.
And doschk is no longer a GNU program.
Problem caught by GCC 12.
* src/tar.c (easprintf): New static function, which never returns
a null pointer. All uses of xasprintf replaced by uses of this
function.
* src/extract.c (is_directory_link): New arg ST. Caller changed.
(is_directory_link, open_output_file):
Use readlinkat, not fstatat, to determine whether a string
names a symlink. This avoids EOVERFLOW issues.
(extract_dir): Avoid duplicate calls to fstatat when
keep_directory_symlink_option && fstatat_flags == 0
and the file is a symlink to an existing file.
Problem reported in:
https://lists.gnu.org/r/bug-tar/2022-03/msg00000.html
* src/extract.c (make_directories): Retry the file creation as
long as the directory exists, regardless of whether tar itself
created the directory.
Copyright-paperwork-exempt: Yes
Problem reported by James Abbatiello in:
https://lists.gnu.org/r/bug-tar/2022-03/msg00000.html
* src/extract.c (make_directories): Do not assume that when
mkdirat fails with errno == EEXIST that there is an existing file
that can be statted. It could be a dangling symlink. Instead,
wait until the end and stat it.
* src/create.c (dump_file0): Remove an fstatat call that is
unnecessary because the file wasn’t read so we can treat the first
fstatat as atomic. Warn “file changed” when the file’s size,
mtime, user ID, group ID, or mode changes, instead of when the
file’s size or ctime changes. Also, when such a change happens,
do not change exit status if --ignore-failed-read. Finally, don’t
attempt to change atime back if it didn’t change.
* src/create.c (dump_file0): For clarity, change diagnostic
wording from "file is the archive; not dumped" to "archive cannot
contain itself; not dumped". All test cases and documentation changed.
* src/create.c (start_header): Leave the devmajor and devminor
fields empty for files that are not character and block special
devices, even when the archive format is pax, ustar or v7.
This avoids generating irrelevant differences which helps with
reproducible builds, and is more compatible with what Solaris 10
tar does.
* src/common.h (get_directory_entries):
Add _GL_ATTRIBUTE_MALLOC _GL_ATTRIBUTE_DEALLOC_FREE.
Problem found by gcc -Wsuggest-attribute=malloc and
current Gnulib.
* src/common.h (xheader_xattr_free,xheader_xattr_copy): Remove protos.
(xattr_map_init,xattr_map_copy)
(xattr_map_add,xattr_map_free): New protos.
* src/tar.h (xattr_map): New struct.
(tar_stat_info): Replace xattr_map_size and xattr_map with one
field: xattr_map.
* src/xattrs.c (XATTRS_PREFIX,XATTRS_PREFIX_LEN): New defines.
(xheader_xattr_init,xattr_map_init)
(xattr_map_free,xattr_map_add)
(xheader_xattr_add,xattr_map_copy): New functions.
All uses changed.
* src/create.c (start_header): Update to use struct xattr_map.
* src/extract.c: Update to use struct xattr_map.
* src/tar.c: Likewise.
* src/xheader.c (xheader_xattr_init,xheader_xattr_free)
(xheader_xattr_add,xheader_xattr_copy): Remove.
(xattr_coder,xattr_decoder): Use xattr_map_ functions.
This fixes handling of expressions like 's/s/@/2'
Fix suggested by Anthony Heading.
* src/transform.c (_single_transform_name_to_obstack): Avoid duplicating
initial prefix if replace is not needed.
The copy_string_unquote function does not handle arguments which only
consist of a single quote. A string is only quoted if two quoting
characters exist.
How to reproduce:
tar --checkpoint-action exec=\"
* src/names.c (read_name_from_file): Do not remove trailing slash
here, since name_buffer might contain an option (e.g. -C /).
(read_next_name): Remove trailing slash when we're sure we're dealing
with a file name.
See 163e96a0.
* src/Makefile.am (LDADD): Remove, folding into tar_LDADD.
* src/Makefile.am (tar_LDADD), tests/Makefile.am (LDADD):
Add the libraries gnulib-tool currently recommends: LIB_ACL,
LIB_CLOCK_GETTIME, LIB_GETRANDOM, LIB_HARD_LOCALE, LIB_HAS_ACL,
LIB_MBRTOWC, LIB_SETLOCALE_NULL. Otherwise, tar won’t link on AIX
7.1 with xlc because the -lpthread option is missing.
* src/sparse.c (sparse_extract_file): Set *SIZE to
stat.st_size so that the caller does not use *SIZE
when uninitalized. Problem found with GCC 10 and
--enable-gcc-warnings CFLAGS='-O2 -flto -fanalyzer'.
With GCC 10.2.1, ‘./configure --enable-gcc-warnings CFLAGS='-O2
-flto -fanalyzer' issued a false alarm about uninitialized
variable use. Pacify GCC by using a variant of the code.
* src/buffer.c (zip_program): Omit last placeholder entry.
(n_zip_programs): New constant.
(find_zip_program): Use it instead of placeholder.
(first_decompress_program): Set *PSTATE to maximum value
if skipping the table. This avoids confusing gcc -flto
into thinking *PSTATE is used uninitialized.
(next_decompress_program): Simplify now that *PSTATE is maximal
when skipping.
With today’s compilers ‘inline’ is typically not needed for
performance (at least the way GNU Tar uses it) and it gets in the
way of portability.
* configure.ac: Omit AC_C_INLINE; no longer needed here.
* lib/attr-xattr.in.h (setxattr, lsetxattr, fsetxattr, getxattr)
(lgetxattr, fgetxattr, listxattr, llistxattr, flistxattr):
* lib/wordsplit.c (skip_delim_internal, skip_delim)
(skip_delim_real, exptab_matches):
* src/delete.c (flush_file):
* src/extract.c (safe_dir_mode):
* src/misc.c (ptr_align):
Now just static, not static inline.
* lib/wordsplit.h (wordsplit_getwords): Remove; no longer used.
* src/common.h (name_more_files): Now COMMON_INLINE, not
extern inline - which is not portable according to C99,
the way we were using it.
This can be helpful in porting to compilers like Oracle Developer
Studio that support some but not all GCC attributes.
* lib/wordsplit.c (FALLTHROUGH): Remove; now done by attribute.h.
* lib/wordsplit.h (__WORDSPLIT_ATTRIBUTE_FORMAT): Remove;
all uses replaced by ATTRIBUTE_FORMAT.
* lib/wordsplit.h, src/buffer.c, src/common.h, src/compare.c:
* src/sparse.c, src/system.c, src/xheader.c:
Prefer ATTRIBUTE_FORMAT, MAYBE_UNUSED, _Noreturn, etc. to
__attribute__.
Fedora 33 uses GCC 10.2.1, which is a bit pickier.
* configure.ac: Do not use -Wsystem-headers, as this
runs afoul of netdb.h on Fedora 33.
* gnulib.modules: Add ‘attribute’.
* lib/wordsplit.c (wsnode_new): Return the newly allocated
pointer instead of a boolean, to pacify GCC 10.2.1 which otherwise
complains about use of possibly-null pointers. All uses changed.
* src/buffer.c (try_new_volume): Don’t assume find_next_block succeeds.
(_write_volume_label): Pacify GCC 10.2.1 with an ‘assume’, since
LABEL must be nonnull here.
* src/common.h (FALLTHROUGH): Remove; now in attribute.h.
Include attribute.h, for ATTRIBUTE_NONNULL.
* src/misc.c (assign_string_or_null): New function,
taking over the old role of assign_string.
(assign_string): Assume VALUE is non-null.
(assign_null): New function, taking over the old
role of assign_string when its VALUE was nonnull.
All callers of assign_string changed to use these functions.
(assign_string_n): Clear *STRING if VALUE is null,
to fix a potential double-free.
If the archive being extracted contains multiple copies
of the same symlink, extract only the first of them and
skip the rest. The use case is described in
https://lists.gnu.org/archive/html/bug-tar/2021-01/msg00026.html
The bug was introduced by 2d3396c3ea.
* src/extract.c (create_placeholder_file): If placeholder already
exists and is registered in the delayed_link list, return immediately.
Bug was introduced by commit 34d15af1.
* src/extract.c (prepare_to_extract): When extracting over pipe,
process only regular files.
* tests/extrac24.at: New test case.
* tests/Makefile.am: Add new test case.
* tests/testsuite.at: Likewise.
Bug reported in https://savannah.gnu.org/bugs/?59897
* src/list.c (read_header): Don't return directly from the loop.
Instead set the status and break. Return the status. Free
next_long_name and next_long_link before returning.
* lib/stdopen.c (stdopen): Fix improper condition.
Avoid leaking extra file descriptor.
* src/tar.c (main): Set name of the stdout for diagnostics.
Bail out if stdopen fails.
Using such options as -f, -z, etc. is senseless in the file list file
and bypasses the option consistency checks in decode_options. Therefore,
only options related to file selection (a.k.a position-sensitive options)
are allowed in files.
* doc/tar.texi: Document changes.
* src/common.h (tar_args): Move from tar.c
(TAR_ARGS_INITIALIZER): New macro.
* src/names.c: Declare option group identifiers as an enum.
(names_parse_opt): Special handling for ARGP_KEY_ERROR.
(names_argp): Remove static qualifier.
(names_argp_children): Remove.
* src/tar.c: Declare option group identifiers as an enum.
(parse_opt): Special handling for ARGP_KEY_INIT.
(argp_children): New static variable.
(args): Remove static variable.
(more_options): Allow only options from names_argp.
(parse_default_options): Take a pointer to struct tar_args as argument.
Replace the loc member during the call to argp_parse and restore it
afterwards.
(decode_options): Use automatic variable for args.
* src/extract.c (prepare_to_extract): Return true to proceed with
the extraction, and false to skip the current member. If extracting
over a pipe, skip unlinking logic.
(extract_archive): Update accordingly.
* src/common.h (name_count): Remove extern.
(files_count): New enum.
(filename_args): New extern.
* src/names.c (name_count): Remove.
(files_count): New variable.
(name_add_name,name_add_file): Update filename_args.
* src/create.c (create_archive): Set trivial_link_count depending on
the filename_args.
