(O_NOFOLLOW, STREQ): Define.
(diropen_fd): Remove function. Merge it into sole caller...
(diropen): ...here. Use O_NOFOLLOW when appropriate.
(fts_safe_changedir): Call fstat for dev/inode check, only if the
previous open/openat call may have opened the wrong directory.
it's easier for legacy applications designed for the version
of fts in glibc or BSD to convert to this more robust version.
Add a new mode, FTS_CWDFD, by which to enable the improved
(openat- based -- aka no-chdir) semantics.
* fts_.h (FTS_CWDFD): Define. Callers must use this fts_open
option to enable the more robust behavior.
(FTS_OPTIONMASK): Widen accordingly.
* fts.c: Restore removed code, reverting the default behavior.
Assume that filemode.h includes sys/types.h and sys/stat.h.
(HAVE_ST_DM_MODE): New macro, moved here from ls.c.
(ftypelet): Reorder to put common cases first, for efficiency.
Add 'P', 'w'. Remove 'M', since it's now the caller's responsibility
to do 'M'.
(strmode): Renamed from mode_string, and now stores 12 bytes instead
of 10, for compatibility with FreeBSD. All callers changed.
(filemodestring): Now stores 12 bytes instead of 10, and sets file types
that can't be deduced solely from st_mode. First arg is now a const
pointer.
fsusage.h now does that. Include fsusage.h first, to test interface.
Prefer statvfs if it works, since it's blessed by POSIX. Attempt
at most one method (the old code could have generated decls that
didn't conform to C89, not that this was ever exercised).
Preserve // when it is special.
Preserve relative files that look like drive letters.
(base_len): Preserve // when it is special.
(last_component): New method, similar to old base_name semantics.
[FILE_SYSTEM_ACCEPTS_DRIVE_LETTER_PREFIX]: Don't treat 1: as a drive prefix.
(IS_ABSOLUTE_FILE_NAME): Treat all drive letters as absolute on
platforms like cygwin with FILE_SYSTEM_DRIVE_PREFIX_IS_ABSOLUTE.
(last_component): New method.
don't match the stat.st_ino values for the same names.
(__getcwd): When no d_ino value matches the target inode
number, iterate through all entries again, using lstat instead.
Reported by Kenshi Muto in http://bugs.debian.org/355810.
