Tar --remove-files relied on canonicalize_file_name,
which replaces symlinks in file name components with
the directories they point to. Due to this, tar
effectively ignored existence of symbolic links and
was unable to remove a directory that contained any
(Alexander Kozlov <akozlov@nada.kth.se>, 2010-03-15).
* gnulib.modules: Remove canonicalize.
* src/misc.c (normalize_filename): Rewrite
from scratch. The function operates only on
its input string, it makes no attempt to test
components for existence or to resolve symbolic
links.
* tests/Makefile.am (TESTSUITE_AT): Add remfiles03.at.
* tests/testsuite.at: Likewise.
* tests/remfiles03.at: New test case.
* NEWS: Update.
* src/common.h (namebuf_t): New typedef.
(namebuf_create, namebuf_free)
(namebuf_name): New prototypes.
(remname): New prototype.
* src/misc.c (struct namebuf): New structure.
(namebuf_create, namebuf_free)
(namebuf_name): New functions.
* src/create.c (dup_dir0): Remove is_avoided_name
checks. This is taken care of in update_archive.
* src/incremen.c (scan_directory): Use namebuf
to produce full file names.
* src/names.c (nametail): Remove extra level of
indirection. All uses updated.
(avoided_name_table, add_avoided_name)
(is_avoided_name): Remove.
* src/update.c (update_archive): Change algorithm.
Instead of adding unmodified files to the avoided_name
table, create namelist so that it contains only
modified files.
* tests/Makefile.am: Add update01.at, update02.at
* tests/testsuite.at: Likewise.
* tests/update.at (AT_KEYWORDS): Add update00.
* src/tar.c (set_exit_status): New function.
* src/common.h (set_exit_status): New prototype.
* src/compare.c: Use set_exit_status instead of
exit_status assignments.
* src/create.c: Likewise.
* src/misc.c: Likewise.
* src/system.c (wait_for_grandchild): Use auto variable
instead of the global exit_status.
* src/incremen.c (scan_directory): Use file_removed_diag
instead of stat_diag.
Changes to src/create.c and src/incremen.c are partially
based on patch from Alexander Peslyak <solar at openwall.com>.
The new testcases require paxutils commit f653a2b or later.
* src/common.h (struct name): New member `cmdline'.
(dump_file): Change type of the 2nd argument to bool.
(file_removed_diag, dir_removed_diag): New prototypes.
(addname): New argument `cmdline'.
(name_from_list): Change return value.
* src/create.c (dump_dir0, dump_dir): top_level is bool.
(create_archive): Update calls to name_from_list.
Take advantage of the name->cmdline to set top_level argument
during incremental backups.
(dump_file0): top_level is bool.
Do not bail out if a no-top-level file disappears during incremental
backup, use file_removed_diag instead.
(dump_filed): top_level is bool.
* src/incremen.c (update_parent_directory): Silently ignore
ENOENT. It should have already been reported elsewhere.
(scan_directory): Use dir_removed_diag to report missing directories.
* src/misc.c (file_removed_diag, dir_removed_diag): New functions.
* src/names.c (name_gather): Set ->cmdname.
(addname): Likewise. All uses updated.
(name_from_list): Return struct name const *. All uses updated.
* tests/filerem01.at: New testcase.
* tests/filerem02.at: New testcase.
* tests/Makefile.am, tests/testsuite.at: Add filerem01.at, filerem02.at
* tests/grow.at, test/truncate.at: Use new syntax for genfile --run.
* NEWS: Update.
* doc/tar.texi: Minor fix.
The modified algorithm tries to avoid dumping the same
directory twice and ensures the order of the directories
in the resulting archive is the same, whatever their order
on the command line. It also fixes the operation of
--listed-incremental -C.
* gnulib.modules: Add canonicalize
* src/common.h (incremental_level): New global.
(check_exclusion_tags): first argument is const.
(get_directory_contents): Add third argument.
(zap_slashes, normalize_filename): New prototypes.
(chdir_count): New prototype.
(WARN_VERBOSE_WARNINGS): New define.
(WARN_ALL): Exclude WARN_VERBOSE_WARNINGS.
* src/compare.c (diff_dumpdir): Update the call to get_directory_contents.
* src/create.c (check_exclusion_tags): First argument is const.
Use ISSLASH and DIRECTORY_SEPARATOR instead of referring to '/'.
* src/incremen.c (struct directory): New member `caname'.
(hash_directory_name): Rename to hash_directory_canonical_name. Operate
on the canonical name.
(compare_directory_names): Rename to compare_directory_canonical_names.
Operate on the canonical name.
(make_directory): Take two arguments.
(free_directory): Free caname.
(attach_directory): Create caname.
(find_directory): Use caname for lookups.
(PD_VERBOSE): Remove.
(PD_FORCE_INIT): New define.
(procdir): First argument is const.
Reinitialize directory if PD_FORCE_INIT bit is set.
Do not use PD_VERBOSE or verbose_option for issuing warnings.
Rely on WARNOPT instead.
Always set *entry.
(scan_directory): Take three arguments. The third one is a boolean
which is true if the directory is explicitly mentioned on the command
line.
(get_directory_contents): Remove. Use scan_directory instead.
All callers updated.
(read_directory_file): Truncate the file if --level=0 is given.
* src/misc.c: Include canonicalize.h
(zap_slashes, normalize_filename): New functions.
(chdir_count): New function.
* src/names.c (add_hierarchy_to_namelist): Take three arguments, as
get_directory_contents and scan_directory.
(collect_and_sort_names): Allow at most one -C, before file name
arguments.
Read directory file after eventual changing to another directory.
Avoid adding the same directory under different pathnames to
the list.
* src/tar.c: New option --level.
* tests/incr03.at, tests/incr04.at, tests/listed01.at,
tests/listed02.at, tests/rename01.at, tests/rename02.at,
tests/rename03.at: Update for new tar behavior.
* tests/multiv01.at: Do not use --listed-incremental.
* NEWS, THANKS: Update
* src/extract.c (extract_dir): reset status to 0 if the
directory already exists.
* src/misc.c (maybe_backup_file): Assign before_backup_name
and clear after_backup_name before checking if we really need
to backup the file.
* tests/backup01.at: New testcase.
* tests/extrac08.at: New testcase.
* tests/Makefile.am, tests/testsuite.at: Add extrac08.at and
backup01.at
* src/common.h (add_exclude_tag): Add decl; C99 requires this
and Forte warns about it.
* src/incremen.c: Include <mkdtemp.h> for mkdtemp prototype,
for same reason.
* src/misc.c (get_max_open_files): Rewrite to avoid code that
Forte C complains about as being unreachable.
* src/xheader.c (mtime_code): Rewrite to avoid Forte error
reported by Trond Hasle Amundsen.
* src/incremen.c (compare_dirnames): Rewrite to avoid casts.
* src/utf8.c (string_ascii_p): Likewise.
* src/xheader.c (mtime_coder, volume_size_coder, volume_offset_coder):
Likewise.
SCALAR(0x8283218)
(get_max_open_files,closeopen): New functions
(chdir_do): Do not use save_cwd if it was already used more than
max_open_files-4 times to avoid running off the file
descriptors.
* configure.ac: Prefer AC_CHECK_HEADERS_ONCE to AC_CHECK_HEADERS.
Check for stropts.h and sys/filio.h too, for _FIOSATIME.
* doc/tar.texi: Change "modification time" to "data modification
time", "change time" to "status change time", and "filesystem" to
"file system", so that we use terminology consistent with POSIX.
Use American spacing rather than French for sentence ends.
"non-dependable" -> "undependable".
(Option Summary, Attributes): Explain better the pitfalls of the
--atime-preserve option, and suggest read-only mounts,loopback
mounts, and noatime mounts for older systems.
* doc/value.texi (op-atime-preserve-system): Renamed from
op-atime-preserver-system to fix a misspelling.
* src/common.h (enum atime_preserve): Use lower case for enum values.
* src/compare.c: Don't include utimens.h; no longer needed.
(diff_file): Use set_file_atime rather than utimens; avoid closing
diff_handle until after this, so that we can set the file time stamp
via the file descriptor rather than via its name.
* src/create.c: Don't include utimens.h; no longer needed.
(dump_regular_finish): Remove. All callers now do its work inline.
(dump_dir): New arg FD. All callers changed.
Use fdsavedir rather than savedir.
(unknown_file_error): Arg is a const pointer now.
(dump_file0): 2nd arg is a const pointer now.
Treat directories more like files, with respect to --atime-preserve.
For example, also warn if a directory changes while we are dumping it.
Prefer file descriptors to file names when retrieving/setting file
attributes; this saves path-resolution time and allows us to avoid
changing mtime/ctime on Solaris when restoring atime as root.
Use O_DIRECTORY when opening directories, to avoid some race conditions.
Do not reset atime if mtime has changed. Report an error if
we cannot reset atime.
First cut at adding support for --atime-preserve=system.
* doc/tar.texi (Option Summary): First cut at documenting it.
All other uses of --atime-preserve changed to --atime-preserve=replace.
* doc/value.texi (op-atime-preserve-replace, op-atime-preserver-system):
New.
(op-atime-preserve): Mention METHOD.
* src/common.h (atime_preserve): New enum.
(atime_preserve_option): Now of the enum type rather than bool.
All uses changed.
* src/compare.c (diff_file): Read with O_NOATIME if asked for.
* src/create.c (dump_file0): Read regular and CTG files with O_NOATIME
if asked for.
* src/tar.c (usage): Mention new usage.
(parse_opt): Parse new usage.
by struct stat; keep them to full nanosecond resolution.
This affects behavior only on older hosts or file systems
that have lower-resolution time stamps.
* src/common.h (OLDER_STAT_TIME): Parenthesize arg.
(OLDER_TAR_STAT_TIME): New macro.
(code_timespec): New function.
(BILLION, LOG10_BILLION, TIMESPEC_STRSIZE_BOUND): New constants.
* src/compare.c (diff_file): Use full time stamp resolution.
* src/create.c (start_header, dump_file0): Likewise.
(start_header, dump_file0): Adjust to new structure layout.
(dump_regular_finish): Simplify by using timespec_cmp.
* src/extract.c (struct delayed_set_stat): Don't store stat info
that we don't need, to save space. All uses changed.
(struct delayed_set_stat, struct delayed_link, file_newer_p):
(create_placeholder_file, extract_link, apply_delayed_links):
Use full time stamp resolution.
(check_time): Use code_timespec rather than rolling our own code.
(set_stat, delay_set_stat): Arg now points to tar_stat_info to
avoid losing time information. All callers changed.
* src/list.c (read_and, decode_header, print_heaeder):
Use full time stamp resolution.
* src/misc.c (code_timespec): New function.
* src/tar.h (struct tar_stat_info): Record atime, mtime, ctime
separately, for benefit of hosts with lower resolution.
* src/update.c (update_archive): Use full time stamp resolution.
* src/xheader.c (code_time): Use new code_timespec function
to simplify code.
(atime_coder, atime_decoder, ctime_coder, ctime_decoder):
(mtime_coder, mtime_decoder): Use full time stamp resolution.
Report time stamps to full resolution in environment.
Report memory allocation failures rather than ignoring them.
* src/system.c (time_to_env): New function.
(oct_to_env, str_to_env, chr_to_env): Report memory allocation failures.
(stat_to_env): Report full resolution in time stamps.
Invalid header fields go undiagnosed.
Some valid time strings are ignored.
* NEWS: Better support for full-resolution time stamps.
The -v option now prints time stamps only to 1-minute resolution.
* gnulib.modules: Add utimens.
* lib/.cvsignore: Add imaxtostr.c, inttostr.c, inttostr.h,
offtostr.c, umaxtostr.c, utimens.c, utimens.h. Remove paxconvert.c.
* lib/Makefile.tmpl (libtar_a_SOURCES): Remove paxconvert.c.
* lib/paxconvert.c: Remove; superseded by umaxtostr.c.
* po/POTFILES.in: Remove lib/paxconvert.c. Add lib/xalloc-die.c,
lib/obstack.c.
* src/buffer.c (set_start_time, compute_duration, start_time):
Use gettime rather than rolling our own code.
* src/common.h (OLDGNU_NAME_FIELD_SIZE, MAXOCTAL11, MAXOCTAL7): Remove.
(newer_ctime_option): Remove.
(timespec_lt): New function.
(OLDER_STAT_TIME): Use it.
(string_to_chars): First arg is char const *, not char *.
(tartime): Time arg is now struct timespec. New bool arg.
All callers changed.
(code_ns_fraction): New decl.
(sys_stat_nanoseconds): Remove decl.
(get_stat_atime, get_stat_ctime, get_stat_mtime): New functions.
(set_stat_atime, set_stat_ctime, set_stat_mtime): New functions.
* src/compare.c: Include utimens.h rather than rolling our own.
(diff_dir, diff_file, diff_link, diff_symlink, diff_special):
Prototype.
(diff_dumpdir, diff_multivol): Prototype.
(diff_file): Support higher-resolution time stamps.
* src/create.c: Include utimens.h rather than rolling our own.
(MAX_OCTAL_VAL): New macro.
(tar_copy_str, string_to_chars): Don't bother to zero-fill;
the destination is already zeroed.
(string_to_chars): First arg is char const *.
(start_private_header): Use MINOR_TO_CHARS, not MAJOR_TO_CHARS,
for minor device number.
(write_header_name, dump_hard_link, dump_file0):
Simplify test for old GNU format.
(start_header): Put in placeholders for uid, etc., even when
using extended headers, for benefit of older "tar" implementations.
Don't assume uintmax_t is wider than 32 bits.
Output extended header for mtime if needed.
(dump_regular_finish, dump_file0):
Support extended time stamp resolution.
* src/extract.c: Include utimens.h rather than rolling our own.
(check_time): Support extended time stamp resolution.
* src/list.c: Include <inttostr.h>.
(tartime): Use umaxtostr rather than stringify_uintmax_t_backwards.
* src/xheader.c: Include <inttostr.h>.
Do not include <xstrtol.h>.
(strtoimax) [!HAVE_DECL_STRTOIMAX && !defined strtoimax]: New decl.
(strtoumax) [!HAVE_DECL_STRTOUMAX && !defined strtoumax]: New decl.
(BILLION, LOG10_BILLION): New constants.
(to_decimal): Remove; superseded by inttostr. All callers changed
to use umaxtostr.
(xheader_format_name): Don't assume pids and uintmax_t values
fit in 63 bytes (!) when printed.
(decode_record): Don't bother to check for ERANGE; an out of range
value must be treater than len_max anyway.
If the length is out of range, output it in the diagnostic.
(format_uintmax): Remove; all callers changed to use umaxtostr.
(xheader_print): Don't assume sizes can be printed in 99 bytes (!).
(out_of_range_header): New function.
(decode_time): Use it.
(code_time): Accept struct timespec, not time_t and unsigned long.
All callers changed. Size sbuf properly, and remove unnecessary check.
Don't assume time stamps can fit in 199 bytes.
Handle negative time stamps. Handle fractional time stamps
more consistently. Don't output unnecessary trailing zeros.
(decode_time): Yield struct timespec, not time_t and unsigned long.
All callers changed.
Handle negative time stamps. Truncate towards minus infinity
consistently. Improve overflow checks, and output a better
diagnostic on overflow.
(code_num): Don't assume uintmax_t can be printed in 99 bytes (!).
(decode_num): New function, for better diagnostics.
(atime_coder, atime_decoder, gid_decoder, ctime_coder):
(ctime_decoder, mtime_coder, mtime_decoder, size_decoder):
(uid_decoder, sparse_size_decoder, sparse_numblocks_decoder):
(sparse_offset_decoder, sparse_numbytes_decoder):
Use decode_num, etc., instead of xstrtoumax, etc.
* src/xheader.c (sparse_numblocks_decoder): Remove unchecked use
of `calloc'. Use xcalloc instead.
(decode_time, gid_decoder, size_decoder, uid_decoder):
(sparse_size_decoder, sparse_offset_decoder, sparse_numblocks_decoder):
Ensure that the result of calling xstrtoumax is no larger than
the maximum value for the target type. Upon any failure, exit with
a diagnostic.
(sparse_numblocks_decoder): Avoid buffer overrun/heap corruption:
use x2nrealloc, rather than `n *= 2' and xrealloc(p, n,....
(decode_time): Rewrite to accept time strings like
1119018481.000000000. Before, such strings were always ignored.