/* Traverse a file hierarchy.
- Copyright (C) 2004, 2005, 2006 Free Software Foundation, Inc.
+ Copyright (C) 2004-2009 Free Software Foundation, Inc.
- This program is free software; you can redistribute it and/or modify
+ This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
- the Free Software Foundation; either version 2, or (at your option)
- any later version.
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
- along with this program; if not, write to the Free Software Foundation,
- Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
/*-
* Copyright (c) 1990, 1993, 1994
#endif
#include <fcntl.h>
#include <errno.h>
-#include "dirfd.h"
#include <stdbool.h>
#include <stdlib.h>
#include <string.h>
#if ! _LIBC
# include "fcntl--.h"
-# include "lstat.h"
-# include "openat.h"
+# include "dirent--.h"
# include "unistd--.h"
+/* FIXME - use fcntl(F_DUPFD_CLOEXEC)/openat(O_CLOEXEC) once they are
+ supported. */
+# include "cloexec.h"
+# include "openat.h"
+# include "same-inode.h"
#endif
#include <dirent.h>
# define _D_EXACT_NAMLEN(dirent) strlen ((dirent)->d_name)
#endif
+#if HAVE_STRUCT_DIRENT_D_TYPE
+/* True if the type of the directory entry D is known. */
+# define DT_IS_KNOWN(d) ((d)->d_type != DT_UNKNOWN)
+/* True if the type of the directory entry D must be T. */
+# define DT_MUST_BE(d, t) ((d)->d_type == (t))
+# define D_TYPE(d) ((d)->d_type)
+#else
+# define DT_IS_KNOWN(d) false
+# define DT_MUST_BE(d, t) false
+# define D_TYPE(d) DT_UNKNOWN
+
+# undef DT_UNKNOWN
+# define DT_UNKNOWN 0
+
+/* Any nonzero values will do here, so long as they're distinct.
+ Undef any existing macros out of the way. */
+# undef DT_BLK
+# undef DT_CHR
+# undef DT_DIR
+# undef DT_FIFO
+# undef DT_LNK
+# undef DT_REG
+# undef DT_SOCK
+# define DT_BLK 1
+# define DT_CHR 2
+# define DT_DIR 3
+# define DT_FIFO 4
+# define DT_LNK 5
+# define DT_REG 6
+# define DT_SOCK 7
+#endif
+
+#ifndef S_IFLNK
+# define S_IFLNK 0
+#endif
+#ifndef S_IFSOCK
+# define S_IFSOCK 0
+#endif
+
+enum
+{
+ NOT_AN_INODE_NUMBER = 0
+};
+
+#ifdef D_INO_IN_DIRENT
+# define D_INO(dp) (dp)->d_ino
+#else
+/* Some systems don't have inodes, so fake them to avoid lots of ifdefs. */
+# define D_INO(dp) NOT_AN_INODE_NUMBER
+#endif
+
+/* If there are more than this many entries in a directory,
+ and the conditions mentioned below are satisfied, then sort
+ the entries on inode number before any further processing. */
+#ifndef FTS_INODE_SORT_DIR_ENTRIES_THRESHOLD
+# define FTS_INODE_SORT_DIR_ENTRIES_THRESHOLD 10000
+#endif
+enum
+{
+ _FTS_INODE_SORT_DIR_ENTRIES_THRESHOLD = FTS_INODE_SORT_DIR_ENTRIES_THRESHOLD
+};
+
+enum Fts_stat
+{
+ FTS_NO_STAT_REQUIRED = 1,
+ FTS_STAT_REQUIRED = 2
+};
+
#ifdef _LIBC
# undef close
# define close __close
# define __set_errno(Val) errno = (Val)
#endif
-#ifndef __attribute__
-# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 8) || __STRICT_ANSI__
-# define __attribute__(x) /* empty */
-# endif
-#endif
-
-#ifndef ATTRIBUTE_UNUSED
-# define ATTRIBUTE_UNUSED __attribute__ ((__unused__))
-#endif
-
/* If this host provides the openat function, then we can avoid
attempting to open "." in some initialization code below. */
#ifdef HAVE_OPENAT
# define HAVE_OPENAT_SUPPORT 0
#endif
+#ifdef NDEBUG
+# define fts_assert(expr) ((void) 0)
+#else
+# define fts_assert(expr) \
+ do \
+ { \
+ if (!(expr)) \
+ abort (); \
+ } \
+ while (false)
+#endif
+
static FTSENT *fts_alloc (FTS *, const char *, size_t) internal_function;
static FTSENT *fts_build (FTS *, int) internal_function;
static void fts_lfree (FTSENT *) internal_function;
static int fts_safe_changedir (FTS *, FTSENT *, int, const char *)
internal_function;
-#if _LGPL_PACKAGE
+#if GNULIB_FTS
+# include "fts-cycle.c"
+#else
static bool enter_dir (FTS *fts, FTSENT *ent) { return true; }
static void leave_dir (FTS *fts, FTSENT *ent) {}
static bool setup_dir (FTS *fts) { return true; }
static void free_dir (FTS *fts) {}
-#else
-# include "fcntl--.h"
-# include "fts-cycle.c"
#endif
#ifndef MAX
# define SIZE_MAX ((size_t) -1)
#endif
-#ifndef O_DIRECTORY
-# define O_DIRECTORY 0
-#endif
-
#define ISDOT(a) (a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
#define STREQ(a, b) (strcmp ((a), (b)) == 0)
#define ISSET(opt) (sp->fts_options & (opt))
#define SET(opt) (sp->fts_options |= (opt))
-#define RESTORE_INITIAL_CWD(sp) FCHDIR (sp, (ISSET (FTS_CWDFD) \
- ? AT_FDCWD \
- : sp->fts_rfd))
+/* FIXME: make this a function */
+#define RESTORE_INITIAL_CWD(sp) \
+ (fd_ring_clear (&((sp)->fts_fd_ring)), \
+ FCHDIR ((sp), (ISSET (FTS_CWDFD) ? AT_FDCWD : (sp)->fts_rfd)))
-#define FCHDIR(sp, fd) (!ISSET(FTS_NOCHDIR) \
- && (ISSET(FTS_CWDFD) \
- ? (cwd_advance_fd (sp, fd), 0) \
- : fchdir(fd)))
+/* FIXME: FTS_NOCHDIR is now misnamed.
+ Call it FTS_USE_FULL_RELATIVE_FILE_NAMES instead. */
+#define FCHDIR(sp, fd) \
+ (!ISSET(FTS_NOCHDIR) && (ISSET(FTS_CWDFD) \
+ ? (cwd_advance_fd ((sp), (fd), true), 0) \
+ : fchdir (fd)))
/* fts_build flags */
+/* FIXME: make this an enum */
#define BCHILD 1 /* fts_children */
#define BNAMES 2 /* fts_children, names only */
#define BREAD 3 /* fts_read */
# include <inttypes.h>
# include <stdint.h>
# include <stdio.h>
+# include "getcwdat.h"
bool fts_debug = false;
-# define Dprintf(x) do { if (fts_debug) printf x; } while (0)
+# define Dprintf(x) do { if (fts_debug) printf x; } while (false)
#else
# define Dprintf(x)
+# define fd_ring_check(x)
+# define fd_ring_print(a, b, c)
#endif
#define LEAVE_DIR(Fts, Ent, Tag) \
{ \
Dprintf ((" %s-leaving: %s\n", Tag, (Ent)->fts_path)); \
leave_dir (Fts, Ent); \
+ fd_ring_check (Fts); \
} \
while (false)
+static void
+fd_ring_clear (I_ring *fd_ring)
+{
+ while ( ! i_ring_empty (fd_ring))
+ {
+ int fd = i_ring_pop (fd_ring);
+ if (0 <= fd)
+ close (fd);
+ }
+}
+
+/* Overload the fts_statp->st_size member (otherwise unused, when
+ fts_info is FTS_NSOK) to indicate whether fts_read should stat
+ this entry or not. */
+static void
+fts_set_stat_required (FTSENT *p, bool required)
+{
+ fts_assert (p->fts_info == FTS_NSOK);
+ p->fts_statp->st_size = (required
+ ? FTS_STAT_REQUIRED
+ : FTS_NO_STAT_REQUIRED);
+}
+
/* file-descriptor-relative opendir. */
/* FIXME: if others need this function, move it into lib/openat.c */
static inline DIR *
internal_function
opendirat (int fd, char const *dir)
{
- int new_fd = openat (fd, dir, O_RDONLY);
+ int new_fd = openat (fd, dir,
+ O_RDONLY | O_DIRECTORY | O_NOCTTY | O_NONBLOCK);
DIR *dirp;
if (new_fd < 0)
return NULL;
+ set_cloexec_flag (new_fd, true);
dirp = fdopendir (new_fd);
if (dirp == NULL)
{
return dirp;
}
-/* Virtual fchdir. Advance SP's working directory
- file descriptor, SP->fts_cwd_fd, to FD, and close
- the previous one, ignoring any error. */
+/* Virtual fchdir. Advance SP's working directory file descriptor,
+ SP->fts_cwd_fd, to FD, and push the previous value onto the fd_ring.
+ CHDIR_DOWN_ONE is true if FD corresponds to an entry in the directory
+ open on sp->fts_cwd_fd; i.e., to move the working directory one level
+ down. */
static void
internal_function
-cwd_advance_fd (FTS *sp, int fd)
+cwd_advance_fd (FTS *sp, int fd, bool chdir_down_one)
{
int old = sp->fts_cwd_fd;
- if (old == fd && old != AT_FDCWD)
- abort ();
+ fts_assert (old != fd || old == AT_FDCWD);
+
+ if (chdir_down_one)
+ {
+ /* Push "old" onto the ring.
+ If the displaced file descriptor is non-negative, close it. */
+ int prev_fd_in_slot = i_ring_push (&sp->fts_fd_ring, old);
+ fd_ring_print (sp, stderr, "post-push");
+ if (0 <= prev_fd_in_slot)
+ close (prev_fd_in_slot); /* ignore any close failure */
+ }
+ else if ( ! ISSET (FTS_NOCHDIR))
+ {
+ if (0 <= old)
+ close (old); /* ignore any close failure */
+ }
+
sp->fts_cwd_fd = fd;
- if (0 <= old)
- close (old); /* ignore any close failure */
}
/* Open the directory DIR if possible, and return a file
int open_flags = (O_RDONLY | O_DIRECTORY | O_NOCTTY | O_NONBLOCK
| (ISSET (FTS_PHYSICAL) ? O_NOFOLLOW : 0));
- return (ISSET (FTS_CWDFD)
- ? openat (sp->fts_cwd_fd, dir, open_flags)
- : open (dir, open_flags));
+ int fd = (ISSET (FTS_CWDFD)
+ ? openat (sp->fts_cwd_fd, dir, open_flags)
+ : open (dir, open_flags));
+ if (0 <= fd)
+ set_cloexec_flag (fd, true);
+ return fd;
}
FTS *
FTSENT *parent = NULL;
FTSENT *tmp = NULL; /* pacify gcc */
size_t len;
+ bool defer_stat;
/* Options check. */
if (options & ~FTS_OPTIONMASK) {
openat via /proc, this technique can still fail, but
only in extreme conditions, e.g., when the working
directory cannot be saved (i.e. save_cwd fails) --
- and that happens only on Linux only when "." is unreadable
+ and that happens on Linux only when "." is unreadable
and the CWD would be longer than PATH_MAX.
FIXME: once Linux kernel openat support is well established,
replace the above open call and this entire if/else block
parent->fts_level = FTS_ROOTPARENTLEVEL;
}
+ /* The classic fts implementation would call fts_stat with
+ a new entry for each iteration of the loop below.
+ If the comparison function is not specified or if the
+ FTS_DEFER_STAT option is in effect, don't stat any entry
+ in this loop. This is an attempt to minimize the interval
+ between the initial stat/lstat/fstatat and the point at which
+ a directory argument is first opened. This matters for any
+ directory command line argument that resides on a file system
+ without genuine i-nodes. If you specify FTS_DEFER_STAT along
+ with a comparison function, that function must not access any
+ data via the fts_statp pointer. */
+ defer_stat = (compar == NULL || ISSET(FTS_DEFER_STAT));
+
/* Allocate/initialize root(s). */
for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
/* Don't allow zero-length file names. */
p->fts_level = FTS_ROOTLEVEL;
p->fts_parent = parent;
p->fts_accpath = p->fts_name;
- p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW) != 0);
-
- /* Command-line "." and ".." are real directories. */
- if (p->fts_info == FTS_DOT)
- p->fts_info = FTS_D;
+ /* Even when defer_stat is true, be sure to stat the first
+ command line argument, since fts_read (at least with
+ FTS_XDEV) requires that. */
+ if (defer_stat && root != NULL) {
+ p->fts_info = FTS_NSOK;
+ fts_set_stat_required(p, true);
+ } else {
+ p->fts_info = fts_stat(sp, p, false);
+ }
/*
* If comparison routine supplied, traverse in sorted
sp->fts_cur->fts_link = root;
sp->fts_cur->fts_info = FTS_INIT;
if (! setup_dir (sp))
- goto mem3;
+ goto mem3;
/*
* If using chdir(2), grab a file descriptor pointing to dot to ensure
&& (sp->fts_rfd = diropen (sp, ".")) < 0)
SET(FTS_NOCHDIR);
+ i_ring_init (&sp->fts_fd_ring, -1);
return (sp);
mem3: fts_lfree(root);
p->fts_namelen = len;
}
p->fts_accpath = p->fts_path = sp->fts_path;
- sp->fts_dev = p->fts_statp->st_dev;
}
int
/* Free up child linked list, sort array, file name buffer. */
if (sp->fts_child)
fts_lfree(sp->fts_child);
- if (sp->fts_array)
- free(sp->fts_array);
+ free(sp->fts_array);
free(sp->fts_path);
if (ISSET(FTS_CWDFD))
{
if (0 <= sp->fts_cwd_fd)
- close (sp->fts_cwd_fd);
+ if (close (sp->fts_cwd_fd))
+ saved_errno = errno;
}
else if (!ISSET(FTS_NOCHDIR))
{
/* Return to original directory, save errno if necessary. */
if (fchdir(sp->fts_rfd))
saved_errno = errno;
- close(sp->fts_rfd);
+
+ /* If close fails, record errno only if saved_errno is zero,
+ so that we report the probably-more-meaningful fchdir errno. */
+ if (close (sp->fts_rfd))
+ if (saved_errno == 0)
+ saved_errno = errno;
}
+ fd_ring_clear (&sp->fts_fd_ring);
+
+#if GNULIB_FTS
+ if (sp->fts_leaf_optimization_works_ht)
+ hash_free (sp->fts_leaf_optimization_works_ht);
+#endif
+
free_dir (sp);
/* Free up the stream pointer. */
return (0);
}
+#if defined __linux__ \
+ && HAVE_SYS_VFS_H && HAVE_FSTATFS && HAVE_STRUCT_STATFS_F_TYPE
+
+#include <sys/vfs.h>
+
+/* Linux-specific constants from coreutils' src/fs.h */
+# define S_MAGIC_TMPFS 0x1021994
+# define S_MAGIC_NFS 0x6969
+# define S_MAGIC_REISERFS 0x52654973
+# define S_MAGIC_PROC 0x9FA0
+
+/* Return false if it is easy to determine the file system type of
+ the directory on which DIR_FD is open, and sorting dirents on
+ inode numbers is known not to improve traversal performance with
+ that type of file system. Otherwise, return true. */
+static bool
+dirent_inode_sort_may_be_useful (int dir_fd)
+{
+ /* Skip the sort only if we can determine efficiently
+ that skipping it is the right thing to do.
+ The cost of performing an unnecessary sort is negligible,
+ while the cost of *not* performing it can be O(N^2) with
+ a very large constant. */
+ struct statfs fs_buf;
+
+ /* If fstatfs fails, assume sorting would be useful. */
+ if (fstatfs (dir_fd, &fs_buf) != 0)
+ return true;
+
+ /* FIXME: what about when f_type is not an integral type?
+ deal with that if/when it's encountered. */
+ switch (fs_buf.f_type)
+ {
+ case S_MAGIC_TMPFS:
+ case S_MAGIC_NFS:
+ /* On a file system of any of these types, sorting
+ is unnecessary, and hence wasteful. */
+ return false;
+
+ default:
+ return true;
+ }
+}
+
+/* Given a file descriptor DIR_FD open on a directory D,
+ return true if it is valid to apply the leaf-optimization
+ technique of counting directories in D via stat.st_nlink. */
+static bool
+leaf_optimization_applies (int dir_fd)
+{
+ struct statfs fs_buf;
+
+ /* If fstatfs fails, assume we can't use the optimization. */
+ if (fstatfs (dir_fd, &fs_buf) != 0)
+ return false;
+
+ /* FIXME: do we need to detect AFS mount points? I doubt it,
+ unless fstatfs can report S_MAGIC_REISERFS for such a directory. */
+
+ switch (fs_buf.f_type)
+ {
+ /* List here the file system types that lack useable dirent.d_type
+ info, yet for which the optimization does apply. */
+ case S_MAGIC_REISERFS:
+ return true;
+
+ case S_MAGIC_PROC:
+ /* Explicitly listing this or any other file system type for which
+ the optimization is not applicable is not necessary, but we leave
+ it here to document the risk. Per http://bugs.debian.org/143111,
+ /proc may have bogus stat.st_nlink values. */
+ /* fall through */
+ default:
+ return false;
+ }
+}
+
+#else
+static bool
+dirent_inode_sort_may_be_useful (int dir_fd _UNUSED_PARAMETER_) { return true; }
+static bool
+leaf_optimization_applies (int dir_fd _UNUSED_PARAMETER_) { return false; }
+#endif
+
+#if GNULIB_FTS
+/* link-count-optimization entry:
+ map an stat.st_dev number to a boolean: leaf_optimization_works */
+struct LCO_ent
+{
+ dev_t st_dev;
+ bool opt_ok;
+};
+
+/* Use a tiny initial size. If a traversal encounters more than
+ a few devices, the cost of growing/rehashing this table will be
+ rendered negligible by the number of inodes processed. */
+enum { LCO_HT_INITIAL_SIZE = 13 };
+
+static size_t
+LCO_hash (void const *x, size_t table_size)
+{
+ struct LCO_ent const *ax = x;
+ return (uintmax_t) ax->st_dev % table_size;
+}
+
+static bool
+LCO_compare (void const *x, void const *y)
+{
+ struct LCO_ent const *ax = x;
+ struct LCO_ent const *ay = y;
+ return ax->st_dev == ay->st_dev;
+}
+
+/* Ask the same question as leaf_optimization_applies, but query
+ the cache first (FTS.fts_leaf_optimization_works_ht), and if necessary,
+ update that cache. */
+static bool
+link_count_optimize_ok (FTSENT const *p)
+{
+ FTS *sp = p->fts_fts;
+ Hash_table *h = sp->fts_leaf_optimization_works_ht;
+ struct LCO_ent tmp;
+ struct LCO_ent *ent;
+ bool opt_ok;
+ struct LCO_ent *t2;
+
+ /* If we're not in CWDFD mode, don't bother with this optimization,
+ since the caller is not serious about performance. */
+ if (!ISSET(FTS_CWDFD))
+ return false;
+
+ /* map st_dev to the boolean, leaf_optimization_works */
+ if (h == NULL)
+ {
+ h = sp->fts_leaf_optimization_works_ht
+ = hash_initialize (LCO_HT_INITIAL_SIZE, NULL, LCO_hash,
+ LCO_compare, free);
+ if (h == NULL)
+ return false;
+ }
+ tmp.st_dev = p->fts_statp->st_dev;
+ ent = hash_lookup (h, &tmp);
+ if (ent)
+ return ent->opt_ok;
+
+ /* Look-up failed. Query directly and cache the result. */
+ t2 = malloc (sizeof *t2);
+ if (t2 == NULL)
+ return false;
+
+ /* Is it ok to perform the optimization in the dir, FTS_CWD_FD? */
+ opt_ok = leaf_optimization_applies (sp->fts_cwd_fd);
+ t2->opt_ok = opt_ok;
+ t2->st_dev = p->fts_statp->st_dev;
+
+ ent = hash_insert (h, t2);
+ if (ent == NULL)
+ {
+ /* insertion failed */
+ free (t2);
+ return false;
+ }
+ fts_assert (ent == t2);
+
+ return opt_ok;
+}
+#endif
+
/*
* Special case of "/" at the end of the file name so that slashes aren't
* appended which would cause file names to be written as "....//foo".
/* If fts_build's call to fts_safe_changedir failed
because it was not able to fchdir into a
subdirectory, tell the caller. */
- if (p->fts_errno)
+ if (p->fts_errno && p->fts_info != FTS_DNR)
p->fts_info = FTS_ERR;
LEAVE_DIR (sp, p, "2");
return (p);
/* Move to the next node on this level. */
next: tmp = p;
if ((p = p->fts_link) != NULL) {
+ sp->fts_cur = p;
free(tmp);
/*
if (p->fts_level == FTS_ROOTLEVEL) {
if (RESTORE_INITIAL_CWD(sp)) {
SET(FTS_STOP);
- sp->fts_cur = p;
return (NULL);
}
+ free_dir(sp);
fts_load(sp, p);
+ setup_dir(sp);
goto check_for_dir;
}
memmove(t, p->fts_name, p->fts_namelen + 1);
check_for_dir:
sp->fts_cur = p;
+ if (p->fts_info == FTS_NSOK)
+ {
+ if (p->fts_statp->st_size == FTS_STAT_REQUIRED)
+ {
+ FTSENT *parent = p->fts_parent;
+ if (FTS_ROOTLEVEL < p->fts_level
+ /* ->fts_n_dirs_remaining is not valid
+ for command-line-specified names. */
+ && parent->fts_n_dirs_remaining == 0
+ && ISSET(FTS_NOSTAT)
+ && ISSET(FTS_PHYSICAL)
+ && link_count_optimize_ok (parent))
+ {
+ /* nothing more needed */
+ }
+ else
+ {
+ p->fts_info = fts_stat(sp, p, false);
+ if (S_ISDIR(p->fts_statp->st_mode)
+ && p->fts_level != FTS_ROOTLEVEL
+ && parent->fts_n_dirs_remaining)
+ parent->fts_n_dirs_remaining--;
+ }
+ }
+ else
+ fts_assert (p->fts_statp->st_size == FTS_NO_STAT_REQUIRED);
+ }
+
if (p->fts_info == FTS_D)
{
- Dprintf ((" %s-entering: %s\n", sp, p->fts_path));
+ /* Now that P->fts_statp is guaranteed to be valid,
+ if this is a command-line directory, record its
+ device number, to be used for FTS_XDEV. */
+ if (p->fts_level == FTS_ROOTLEVEL)
+ sp->fts_dev = p->fts_statp->st_dev;
+ Dprintf ((" entering: %s\n", p->fts_path));
if (! enter_dir (sp, p))
{
__set_errno (ENOMEM);
/* Move up to the parent node. */
p = tmp->fts_parent;
+ sp->fts_cur = p;
free(tmp);
if (p->fts_level == FTS_ROOTPARENTLEVEL) {
return (sp->fts_cur = NULL);
}
+ fts_assert (p->fts_info != FTS_NSOK);
+
/* NUL terminate the file name. */
sp->fts_path[p->fts_pathlen] = '\0';
p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
if (p->fts_errno == 0)
LEAVE_DIR (sp, p, "3");
- sp->fts_cur = p;
return ISSET(FTS_STOP) ? NULL : p;
}
*/
/* ARGSUSED */
int
-fts_set(FTS *sp ATTRIBUTE_UNUSED, FTSENT *p, int instr)
+fts_set(FTS *sp _UNUSED_PARAMETER_, FTSENT *p, int instr)
{
if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
instr != FTS_NOINSTR && instr != FTS_SKIP) {
sp->fts_child = fts_build(sp, instr);
if (ISSET(FTS_CWDFD))
{
- cwd_advance_fd (sp, fd);
+ cwd_advance_fd (sp, fd, true);
}
else
{
return (sp->fts_child);
}
+/* A comparison function to sort on increasing inode number.
+ For some file system types, sorting either way makes a huge
+ performance difference for a directory with very many entries,
+ but sorting on increasing values is slightly better than sorting
+ on decreasing values. The difference is in the 5% range. */
+static int
+fts_compare_ino (struct _ftsent const **a, struct _ftsent const **b)
+{
+ return (a[0]->fts_statp->st_ino < b[0]->fts_statp->st_ino ? -1
+ : b[0]->fts_statp->st_ino < a[0]->fts_statp->st_ino ? 1 : 0);
+}
+
+/* Map the dirent.d_type value, DTYPE, to the corresponding stat.st_mode
+ S_IF* bit and set ST.st_mode, thus clearing all other bits in that field. */
+static void
+set_stat_type (struct stat *st, unsigned int dtype)
+{
+ mode_t type;
+ switch (dtype)
+ {
+ case DT_BLK:
+ type = S_IFBLK;
+ break;
+ case DT_CHR:
+ type = S_IFCHR;
+ break;
+ case DT_DIR:
+ type = S_IFDIR;
+ break;
+ case DT_FIFO:
+ type = S_IFIFO;
+ break;
+ case DT_LNK:
+ type = S_IFLNK;
+ break;
+ case DT_REG:
+ type = S_IFREG;
+ break;
+ case DT_SOCK:
+ type = S_IFSOCK;
+ break;
+ default:
+ type = 0;
+ }
+ st->st_mode = type;
+}
+
/*
* This is the tricky part -- do not casually change *anything* in here. The
* idea is to build the linked list of entries that are used by fts_children
}
return (NULL);
}
+ /* Rather than calling fts_stat for each and every entry encountered
+ in the readdir loop (below), stat each directory only right after
+ opening it. */
+ if (cur->fts_info == FTS_NSOK)
+ cur->fts_info = fts_stat(sp, cur, false);
/*
* Nlinks is the number of possible entries of type directory in the
if (nlinks || type == BREAD) {
int dir_fd = dirfd(dirp);
if (ISSET(FTS_CWDFD) && 0 <= dir_fd)
- dir_fd = dup (dir_fd);
+ {
+ dir_fd = dup (dir_fd);
+ set_cloexec_flag (dir_fd, true);
+ }
if (dir_fd < 0 || fts_safe_changedir(sp, cur, dir_fd, NULL)) {
if (nlinks && type == BREAD)
cur->fts_errno = errno;
/* Read the directory, attaching each entry to the `link' pointer. */
doadjust = false;
for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
+ bool is_dir;
+
if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
continue;
* structures already allocated.
*/
mem1: saved_errno = errno;
- if (p)
- free(p);
+ free(p);
fts_lfree(head);
closedir(dirp);
cur->fts_info = FTS_ERR;
new_len = len + _D_EXACT_NAMLEN (dp);
if (new_len < len) {
/*
- * In the unlikely even that we would end up
+ * In the unlikely event that we would end up
* with a file name longer than SIZE_MAX, free up
* the current structure and the structures already
* allocated, then error out with ENAMETOOLONG.
if (dp->d_type == DT_WHT)
p->fts_flags |= FTS_ISW;
#endif
+ /* Store dirent.d_ino, in case we need to sort
+ entries before processing them. */
+ p->fts_statp->st_ino = D_INO (dp);
/* Build a file name for fts_stat to stat. */
if (ISSET(FTS_NOCHDIR)) {
memmove(cp, p->fts_name, p->fts_namelen + 1);
} else
p->fts_accpath = p->fts_name;
- /* Stat it. */
- p->fts_info = fts_stat(sp, p, false);
+
+ if (sp->fts_compar == NULL || ISSET(FTS_DEFER_STAT)) {
+ /* Record what fts_read will have to do with this
+ entry. In many cases, it will simply fts_stat it,
+ but we can take advantage of any d_type information
+ to optimize away the unnecessary stat calls. I.e.,
+ if FTS_NOSTAT is in effect and we're not following
+ symlinks (FTS_PHYSICAL) and d_type indicates this
+ is *not* a directory, then we won't have to stat it
+ at all. If it *is* a directory, then (currently)
+ we stat it regardless, in order to get device and
+ inode numbers. Some day we might optimize that
+ away, too, for directories where d_ino is known to
+ be valid. */
+ bool skip_stat = (ISSET(FTS_PHYSICAL)
+ && ISSET(FTS_NOSTAT)
+ && DT_IS_KNOWN(dp)
+ && ! DT_MUST_BE(dp, DT_DIR));
+ p->fts_info = FTS_NSOK;
+ /* Propagate dirent.d_type information back
+ to caller, when possible. */
+ set_stat_type (p->fts_statp, D_TYPE (dp));
+ fts_set_stat_required(p, !skip_stat);
+ is_dir = (ISSET(FTS_PHYSICAL)
+ && DT_MUST_BE(dp, DT_DIR));
+ } else {
+ p->fts_info = fts_stat(sp, p, false);
+ is_dir = (p->fts_info == FTS_D
+ || p->fts_info == FTS_DC
+ || p->fts_info == FTS_DOT);
+ }
/* Decrement link count if applicable. */
- if (nlinks > 0 && (p->fts_info == FTS_D ||
- p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
+ if (nlinks > 0 && is_dir)
nlinks -= nostat;
/* We walk in directory order so "ls -f" doesn't get upset. */
return (NULL);
}
+ /* If there are many entries, no sorting function has been specified,
+ and this file system is of a type that may be slow with a large
+ number of entries, then sort the directory entries on increasing
+ inode numbers. */
+ if (nitems > _FTS_INODE_SORT_DIR_ENTRIES_THRESHOLD
+ && !sp->fts_compar
+ && ISSET (FTS_CWDFD)
+ && dirent_inode_sort_may_be_useful (sp->fts_cwd_fd)) {
+ sp->fts_compar = fts_compare_ino;
+ head = fts_sort (sp, head, nitems);
+ sp->fts_compar = NULL;
+ }
+
/* Sort the entries. */
if (sp->fts_compar && nitems > 1)
head = fts_sort(sp, head, nitems);
}
}
}
+
+static bool
+same_fd (int fd1, int fd2)
+{
+ struct stat sb1, sb2;
+ return (fstat (fd1, &sb1) == 0
+ && fstat (fd2, &sb2) == 0
+ && SAME_INODE (sb1, sb2));
+}
+
+static void
+fd_ring_print (FTS const *sp, FILE *stream, char const *msg)
+{
+ I_ring const *fd_ring = &sp->fts_fd_ring;
+ unsigned int i = fd_ring->fts_front;
+ char *cwd = getcwdat (sp->fts_cwd_fd, NULL, 0);
+ fprintf (stream, "=== %s ========== %s\n", msg, cwd);
+ free (cwd);
+ if (i_ring_empty (fd_ring))
+ return;
+
+ while (true)
+ {
+ int fd = fd_ring->fts_fd_ring[i];
+ if (fd < 0)
+ fprintf (stream, "%d: %d:\n", i, fd);
+ else
+ {
+ char *wd = getcwdat (fd, NULL, 0);
+ fprintf (stream, "%d: %d: %s\n", i, fd, wd);
+ free (wd);
+ }
+ if (i == fd_ring->fts_back)
+ break;
+ i = (i + I_RING_SIZE - 1) % I_RING_SIZE;
+ }
+}
+
+/* Ensure that each file descriptor on the fd_ring matches a
+ parent, grandparent, etc. of the current working directory. */
+static void
+fd_ring_check (FTS const *sp)
+{
+ if (!fts_debug)
+ return;
+
+ /* Make a writable copy. */
+ I_ring fd_w = sp->fts_fd_ring;
+
+ int cwd_fd = sp->fts_cwd_fd;
+ cwd_fd = dup (cwd_fd);
+ char *dot = getcwdat (cwd_fd, NULL, 0);
+ error (0, 0, "===== check ===== cwd: %s", dot);
+ free (dot);
+ while ( ! i_ring_empty (&fd_w))
+ {
+ int fd = i_ring_pop (&fd_w);
+ if (0 <= fd)
+ {
+ int parent_fd = openat (cwd_fd, "..", O_RDONLY);
+ if (parent_fd < 0)
+ {
+ // Warn?
+ break;
+ }
+ if (!same_fd (fd, parent_fd))
+ {
+ char *cwd = getcwdat (fd, NULL, 0);
+ error (0, errno, "ring : %s", cwd);
+ char *c2 = getcwdat (parent_fd, NULL, 0);
+ error (0, errno, "parent: %s", c2);
+ free (cwd);
+ free (c2);
+ fts_assert (0);
+ }
+ close (cwd_fd);
+ cwd_fd = parent_fd;
+ }
+ }
+ close (cwd_fd);
+}
#endif
static unsigned short int
struct stat *sbp = p->fts_statp;
int saved_errno;
+ if (p->fts_level == FTS_ROOTLEVEL && ISSET(FTS_COMFOLLOW))
+ follow = true;
+
#if defined FTS_WHITEOUT && 0
/* check for whiteout */
if (p->fts_flags & FTS_ISW) {
}
if (S_ISDIR(sbp->st_mode)) {
- if (ISDOT(p->fts_name))
- return (FTS_DOT);
+ p->fts_n_dirs_remaining = (sbp->st_nlink
+ - (ISSET(FTS_SEEDOT) ? 0 : 2));
+ if (ISDOT(p->fts_name)) {
+ /* Command-line "." and ".." are real directories. */
+ return (p->fts_level == FTS_ROOTLEVEL ? FTS_D : FTS_DOT);
+ }
-#if _LGPL_PACKAGE
+#if !GNULIB_FTS
{
/*
* Cycle detection is done by brute force when the directory
* 40 so don't realloc one entry at a time.
*/
if (nitems > sp->fts_nitems) {
- struct _ftsent **a;
+ FTSENT **a;
sp->fts_nitems = nitems + 40;
if (SIZE_MAX / sizeof *a < sp->fts_nitems
* See if fts_pathlen would overflow.
*/
if (new_len < sp->fts_pathlen) {
- if (sp->fts_path) {
- free(sp->fts_path);
- sp->fts_path = NULL;
- }
+ free(sp->fts_path);
sp->fts_path = NULL;
__set_errno (ENAMETOOLONG);
return false;
fts_safe_changedir (FTS *sp, FTSENT *p, int fd, char const *dir)
{
int ret;
+ bool is_dotdot = dir && STREQ (dir, "..");
+ int newfd;
+
+ /* This clause handles the unusual case in which FTS_NOCHDIR
+ is specified, along with FTS_CWDFD. In that case, there is
+ no need to change even the virtual cwd file descriptor.
+ However, if FD is non-negative, we do close it here. */
+ if (ISSET (FTS_NOCHDIR))
+ {
+ if (ISSET (FTS_CWDFD) && 0 <= fd)
+ close (fd);
+ return 0;
+ }
- int newfd = fd;
- if (ISSET(FTS_NOCHDIR)) {
- if (ISSET(FTS_CWDFD) && 0 <= fd)
- close (fd);
- return (0);
- }
- if (fd < 0 && (newfd = diropen (sp, dir)) < 0)
- return (-1);
+ if (fd < 0 && is_dotdot && ISSET (FTS_CWDFD))
+ {
+ /* When possible, skip the diropen and subsequent fstat+dev/ino
+ comparison. I.e., when changing to parent directory
+ (chdir ("..")), use a file descriptor from the ring and
+ save the overhead of diropen+fstat, as well as avoiding
+ failure when we lack "x" access to the virtual cwd. */
+ if ( ! i_ring_empty (&sp->fts_fd_ring))
+ {
+ int parent_fd;
+ fd_ring_print (sp, stderr, "pre-pop");
+ parent_fd = i_ring_pop (&sp->fts_fd_ring);
+ is_dotdot = true;
+ if (0 <= parent_fd)
+ {
+ fd = parent_fd;
+ dir = NULL;
+ }
+ }
+ }
- /* The following dev/inode check is necessary if we're doing
- a `logical' traversal (through symlinks, a la chown -L),
- if the system lacks O_NOFOLLOW support, or if we're changing
- to "..". In the latter case, O_NOFOLLOW can't help. In
- general (when the target is not ".."), diropen's use of
- O_NOFOLLOW ensures we don't mistakenly follow a symlink,
- so we can avoid the expense of this fstat. */
- if (ISSET(FTS_LOGICAL) || O_NOFOLLOW == 0
+ newfd = fd;
+ if (fd < 0 && (newfd = diropen (sp, dir)) < 0)
+ return -1;
+
+ /* The following dev/inode check is necessary if we're doing a
+ `logical' traversal (through symlinks, a la chown -L), if the
+ system lacks O_NOFOLLOW support, or if we're changing to ".."
+ (but not via a popped file descriptor). When changing to the
+ name "..", O_NOFOLLOW can't help. In general, when the target is
+ not "..", diropen's use of O_NOFOLLOW ensures we don't mistakenly
+ follow a symlink, so we can avoid the expense of this fstat. */
+ if (ISSET(FTS_LOGICAL) || ! HAVE_WORKING_O_NOFOLLOW
|| (dir && STREQ (dir, "..")))
{
struct stat sb;
if (ISSET(FTS_CWDFD))
{
- cwd_advance_fd (sp, newfd);
+ cwd_advance_fd (sp, newfd, ! is_dotdot);
return 0;
}
(void)close(newfd);
__set_errno (oerrno);
}
- return (ret);
+ return ret;
}
projects / gnulib.git / blobdiff