From f3337786e55909538aacfd7c29b1cf58ff444fbf Mon Sep 17 00:00:00 2001 From: Kamil Dudka Date: Mon, 12 Feb 2018 12:45:36 +0100 Subject: [PATCH 1/4] import gnulib's FTS module from upstream commit 281b825e --- gl/lib/fts.c | 424 +++++++++++++++++++++++++++++----------------------------- gl/lib/fts_.h | 10 +- 2 files changed, 221 insertions(+), 213 deletions(-) diff --git a/gl/lib/fts.c b/gl/lib/fts.c index c91d7a1..bfa73e3 100644 --- a/gl/lib/fts.c +++ b/gl/lib/fts.c @@ -1,6 +1,6 @@ /* Traverse a file hierarchy. - Copyright (C) 2004-2015 Free Software Foundation, Inc. + Copyright (C) 2004-2018 Free Software Foundation, Inc. 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 @@ -13,7 +13,7 @@ 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, see . */ + along with this program. If not, see . */ /*- * Copyright (c) 1990, 1993, 1994 @@ -46,9 +46,9 @@ #include -#if defined(LIBC_SCCS) && !defined(lint) +#if defined LIBC_SCCS && !defined GCC_LINT && !defined lint static char sccsid[] = "@(#)fts.c 8.6 (Berkeley) 8/14/94"; -#endif /* LIBC_SCCS and not lint */ +#endif #include "fts_.h" @@ -71,11 +71,7 @@ static char sccsid[] = "@(#)fts.c 8.6 (Berkeley) 8/14/94"; #if ! _LIBC # include "fcntl--.h" -# include "dirent--.h" -# include "unistd--.h" -/* FIXME - use fcntl(F_DUPFD_CLOEXEC)/openat(O_CLOEXEC) once they are - supported. */ -# include "cloexec.h" +# include "flexmember.h" # include "openat.h" # include "same-inode.h" #endif @@ -202,6 +198,14 @@ enum Fts_stat while (false) #endif +#ifndef FALLTHROUGH +# if __GNUC__ < 7 +# define FALLTHROUGH ((void) 0) +# else +# define FALLTHROUGH __attribute__ ((__fallthrough__)) +# endif +#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; @@ -296,14 +300,13 @@ static DIR * internal_function opendirat (int fd, char const *dir, int extra_flags, int *pdir_fd) { - int new_fd = openat (fd, dir, - (O_RDONLY | O_DIRECTORY | O_NOCTTY | O_NONBLOCK - | extra_flags)); + int open_flags = (O_RDONLY | O_CLOEXEC | O_DIRECTORY | O_NOCTTY + | O_NONBLOCK | extra_flags); + int new_fd = openat (fd, dir, open_flags); DIR *dirp; if (new_fd < 0) return NULL; - set_cloexec_flag (new_fd, true); dirp = fdopendir (new_fd); if (dirp) *pdir_fd = new_fd; @@ -366,15 +369,13 @@ static int internal_function diropen (FTS const *sp, char const *dir) { - int open_flags = (O_SEARCH | O_DIRECTORY | O_NOCTTY | O_NONBLOCK + int open_flags = (O_SEARCH | O_CLOEXEC | O_DIRECTORY | O_NOCTTY | O_NONBLOCK | (ISSET (FTS_PHYSICAL) ? O_NOFOLLOW : 0) | (ISSET (FTS_NOATIME) ? O_NOATIME : 0)); 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; } @@ -470,6 +471,7 @@ fts_open (char * const *argv, if ((parent = fts_alloc(sp, "", 0)) == NULL) goto mem2; parent->fts_level = FTS_ROOTPARENTLEVEL; + parent->fts_n_dirs_remaining = -1; } /* The classic fts implementation would call fts_stat with @@ -656,39 +658,139 @@ fts_close (FTS *sp) return (0); } +/* Minimum link count of a traditional Unix directory. When leaf + optimization is OK and MIN_DIR_NLINK <= st_nlink, then st_nlink is + an upper bound on the number of subdirectories (counting "." and + ".."). */ +enum { MIN_DIR_NLINK = 2 }; + +/* Whether leaf optimization is OK for a directory. */ +enum leaf_optimization + { + /* st_nlink is not reliable for this directory's subdirectories. */ + NO_LEAF_OPTIMIZATION, + + /* Leaf optimization is OK, but is not useful for avoiding stat calls. */ + OK_LEAF_OPTIMIZATION, + + /* Leaf optimization is not only OK: it is useful for avoiding + stat calls, because dirent.d_type does not work. */ + NOSTAT_LEAF_OPTIMIZATION + }; + #if defined __linux__ \ && HAVE_SYS_VFS_H && HAVE_FSTATFS && HAVE_STRUCT_STATFS_F_TYPE # include /* Linux-specific constants from coreutils' src/fs.h */ -# define S_MAGIC_TMPFS 0x1021994 +# define S_MAGIC_AFS 0x5346414F # define S_MAGIC_NFS 0x6969 +# define S_MAGIC_PROC 0x9FA0 # define S_MAGIC_REISERFS 0x52654973 +# define S_MAGIC_TMPFS 0x1021994 # define S_MAGIC_XFS 0x58465342 -# 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. */ +# ifdef HAVE___FSWORD_T +typedef __fsword_t fsword; +# else +typedef long int fsword; +# endif + +/* Map a stat.st_dev number to a file system type number f_ftype. */ +struct dev_type +{ + dev_t st_dev; + fsword f_type; +}; + +/* 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 { DEV_TYPE_HT_INITIAL_SIZE = 13 }; + +static size_t +dev_type_hash (void const *x, size_t table_size) +{ + struct dev_type const *ax = x; + uintmax_t dev = ax->st_dev; + return dev % table_size; +} + static bool -dirent_inode_sort_may_be_useful (int dir_fd) +dev_type_compare (void const *x, void const *y) +{ + struct dev_type const *ax = x; + struct dev_type const *ay = y; + return ax->st_dev == ay->st_dev; +} + +/* Return the file system type of P, or 0 if not known. + Try to cache known values. */ + +static fsword +filesystem_type (FTSENT const *p) +{ + FTS *sp = p->fts_fts; + Hash_table *h = sp->fts_leaf_optimization_works_ht; + struct dev_type *ent; + struct statfs fs_buf; + + /* 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 0; + + if (! h) + h = sp->fts_leaf_optimization_works_ht + = hash_initialize (DEV_TYPE_HT_INITIAL_SIZE, NULL, dev_type_hash, + dev_type_compare, free); + if (h) + { + struct dev_type tmp; + tmp.st_dev = p->fts_statp->st_dev; + ent = hash_lookup (h, &tmp); + if (ent) + return ent->f_type; + } + + /* Look-up failed. Query directly and cache the result. */ + if (fstatfs (p->fts_fts->fts_cwd_fd, &fs_buf) != 0) + return 0; + + if (h) + { + struct dev_type *t2 = malloc (sizeof *t2); + if (t2) + { + t2->st_dev = p->fts_statp->st_dev; + t2->f_type = fs_buf.f_type; + + ent = hash_insert (h, t2); + if (ent) + fts_assert (ent == t2); + else + free (t2); + } + } + + return fs_buf.f_type; +} + +/* Return false if it is easy to determine the file system type of the + directory P, 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 (FTSENT const *p) { /* 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) + switch (filesystem_type (p)) { case S_MAGIC_TMPFS: case S_MAGIC_NFS: @@ -701,133 +803,58 @@ dirent_inode_sort_may_be_useful (int dir_fd) } } -/* 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) +/* Given an FTS entry P for a directory D, + return true if it is both useful and valid to apply leaf optimization. + The optimization is useful only for file systems that lack usable + dirent.d_type info. The optimization is valid if an st_nlink value + of at least MIN_DIR_NLINK is an upper bound on the number of + subdirectories of D, counting "." and ".." as subdirectories. */ +static enum leaf_optimization +leaf_optimization (FTSENT const *p) { - 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) + switch (filesystem_type (p)) { - case S_MAGIC_NFS: - /* NFS provides usable dirent.d_type but not necessarily for all entries - of large directories. See . */ - return true; - - /* List here the file system types that lack usable dirent.d_type + /* List here the file system types that may lack usable dirent.d_type info, yet for which the optimization does apply. */ case S_MAGIC_REISERFS: - case S_MAGIC_XFS: - return true; - + case S_MAGIC_XFS: /* XFS lacked it until 2013-08-22 commit. */ + return NOSTAT_LEAF_OPTIMIZATION; + + case 0: + /* Leaf optimization is unsafe if the file system type is unknown. */ + FALLTHROUGH; + case S_MAGIC_AFS: + /* Although AFS mount points are not counted in st_nlink, they + act like directories. See . */ + FALLTHROUGH; + case S_MAGIC_NFS: + /* NFS provides usable dirent.d_type but not necessarily for all entries + of large directories, so as per + NFS should return true. However st_nlink values are not accurate on + all implementations as per . */ + FALLTHROUGH; 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 */ + /* Per /proc + may have bogus stat.st_nlink values. */ + return NO_LEAF_OPTIMIZATION; + default: - return false; + return OK_LEAF_OPTIMIZATION; } } #else static bool -dirent_inode_sort_may_be_useful (int dir_fd _GL_UNUSED) { return true; } -static bool -leaf_optimization_applies (int dir_fd _GL_UNUSED) { return false; } -#endif - -/* link-count-optimization entry: - map a 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) +dirent_inode_sort_may_be_useful (FTSENT const *p _GL_UNUSED) { - struct LCO_ent const *ax = x; - struct LCO_ent const *ay = y; - return ax->st_dev == ay->st_dev; + return true; } - -/* 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) +static enum leaf_optimization +leaf_optimization (FTSENT const *p _GL_UNUSED) { - 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; + return NO_LEAF_OPTIMIZATION; } +#endif /* * Special case of "/" at the end of the file name so that slashes aren't @@ -1014,13 +1041,11 @@ check_for_dir: 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 + if (parent->fts_n_dirs_remaining == 0 && ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL) - && link_count_optimize_ok (parent)) + && (leaf_optimization (parent) + == NOSTAT_LEAF_OPTIMIZATION)) { /* nothing more needed */ } @@ -1029,7 +1054,8 @@ check_for_dir: 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) + && 0 < parent->fts_n_dirs_remaining + && parent->fts_n_dirs_remaining != (nlink_t) -1) parent->fts_n_dirs_remaining--; } } @@ -1298,8 +1324,6 @@ fts_build (register FTS *sp, int type) bool descend; bool doadjust; ptrdiff_t level; - nlink_t nlinks; - bool nostat; size_t len, maxlen, new_len; char *cp; int dir_fd; @@ -1369,24 +1393,6 @@ fts_build (register FTS *sp, int type) sorting, yet not so large that we risk exhausting memory. */ max_entries = sp->fts_compar ? SIZE_MAX : FTS_MAX_READDIR_ENTRIES; - /* - * Nlinks is the number of possible entries of type directory in the - * directory if we're cheating on stat calls, 0 if we're not doing - * any stat calls at all, (nlink_t) -1 if we're statting everything. - */ - if (type == BNAMES) { - nlinks = 0; - /* Be quiet about nostat, GCC. */ - nostat = false; - } else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) { - nlinks = (cur->fts_statp->st_nlink - - (ISSET(FTS_SEEDOT) ? 0 : 2)); - nostat = true; - } else { - nlinks = -1; - nostat = false; - } - /* * If we're going to need to stat anything or we want to descend * and stay in the directory, chdir. If this fails we keep going, @@ -1408,15 +1414,22 @@ fts_build (register FTS *sp, int type) the required dirp and dir_fd. */ descend = true; } - else if (nlinks || type == BREAD) { + else + { + /* Try to descend unless it is a names-only fts_children, + or the directory is known to lack subdirectories. */ + descend = (type != BNAMES + && ! (ISSET (FTS_NOSTAT) && ISSET (FTS_PHYSICAL) + && ! ISSET (FTS_SEEDOT) + && cur->fts_statp->st_nlink == MIN_DIR_NLINK + && (leaf_optimization (cur) + != NO_LEAF_OPTIMIZATION))); + if (descend || type == BREAD) + { if (ISSET(FTS_CWDFD)) - { - dir_fd = dup (dir_fd); - if (0 <= dir_fd) - set_cloexec_flag (dir_fd, true); - } + dir_fd = fcntl (dir_fd, F_DUPFD_CLOEXEC, STDERR_FILENO + 1); if (dir_fd < 0 || fts_safe_changedir(sp, cur, dir_fd, NULL)) { - if (nlinks && type == BREAD) + if (descend && type == BREAD) cur->fts_errno = errno; cur->fts_flags |= FTS_DONTCHDIR; descend = false; @@ -1426,8 +1439,8 @@ fts_build (register FTS *sp, int type) cur->fts_dirp = NULL; } else descend = true; - } else - descend = false; + } + } /* * Figure out the max file name length that can be stored in the @@ -1458,11 +1471,19 @@ fts_build (register FTS *sp, int type) tail = NULL; nitems = 0; while (cur->fts_dirp) { - bool is_dir; size_t d_namelen; + __set_errno (0); struct dirent *dp = readdir(cur->fts_dirp); - if (dp == NULL) + if (dp == NULL) { + if (errno) { + cur->fts_errno = errno; + /* If we've not read any items yet, treat + the error as if we can't access the dir. */ + cur->fts_info = (continue_readdir || nitems) + ? FTS_ERR : FTS_DNR; + } break; + } if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name)) continue; @@ -1550,19 +1571,10 @@ mem1: saved_errno = errno; 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 && is_dir) - nlinks -= nostat; - /* We walk in directory order so "ls -f" doesn't get upset. */ p->fts_link = NULL; if (head == NULL) @@ -1621,7 +1633,8 @@ mem1: saved_errno = errno; /* If didn't find anything, return NULL. */ if (!nitems) { - if (type == BREAD) + if (type == BREAD + && cur->fts_info != FTS_DNR && cur->fts_info != FTS_ERR) cur->fts_info = FTS_DP; fts_lfree(head); return (NULL); @@ -1633,8 +1646,7 @@ mem1: saved_errno = errno; 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)) { + && dirent_inode_sort_may_be_useful (cur)) { sp->fts_compar = fts_compare_ino; head = fts_sort (sp, head, nitems); sp->fts_compar = NULL; @@ -1757,7 +1769,7 @@ fd_ring_check (FTS const *sp) I_ring fd_w = sp->fts_fd_ring; int cwd_fd = sp->fts_cwd_fd; - cwd_fd = dup (cwd_fd); + cwd_fd = fcntl (cwd_fd, F_DUPFD_CLOEXEC, STDERR_FILENO + 1); char *dot = getcwdat (cwd_fd, NULL, 0); error (0, 0, "===== check ===== cwd: %s", dot); free (dot); @@ -1766,7 +1778,8 @@ fd_ring_check (FTS const *sp) int fd = i_ring_pop (&fd_w); if (0 <= fd) { - int parent_fd = openat (cwd_fd, "..", O_SEARCH | O_NOATIME); + int open_flags = O_SEARCH | O_CLOEXEC | O_NOATIME; + int parent_fd = openat (cwd_fd, "..", open_flags); if (parent_fd < 0) { // Warn? @@ -1795,7 +1808,6 @@ internal_function fts_stat(FTS *sp, register FTSENT *p, bool follow) { struct stat *sbp = p->fts_statp; - int saved_errno; if (p->fts_level == FTS_ROOTLEVEL && ISSET(FTS_COMFOLLOW)) follow = true; @@ -1807,13 +1819,12 @@ fts_stat(FTS *sp, register FTSENT *p, bool follow) */ if (ISSET(FTS_LOGICAL) || follow) { if (stat(p->fts_accpath, sbp)) { - saved_errno = errno; if (errno == ENOENT && lstat(p->fts_accpath, sbp) == 0) { __set_errno (0); return (FTS_SLNONE); } - p->fts_errno = saved_errno; + p->fts_errno = errno; goto err; } } else if (fstatat(sp->fts_cwd_fd, p->fts_accpath, sbp, @@ -1824,8 +1835,11 @@ err: memset(sbp, 0, sizeof(struct stat)); } if (S_ISDIR(sbp->st_mode)) { - p->fts_n_dirs_remaining = (sbp->st_nlink - - (ISSET(FTS_SEEDOT) ? 0 : 2)); + p->fts_n_dirs_remaining + = ((sbp->st_nlink < MIN_DIR_NLINK + || p->fts_level <= FTS_ROOTLEVEL) + ? -1 + : sbp->st_nlink - (ISSET (FTS_SEEDOT) ? 0 : MIN_DIR_NLINK)); if (ISDOT(p->fts_name)) { /* Command-line "." and ".." are real directories. */ return (p->fts_level == FTS_ROOTLEVEL ? FTS_D : FTS_DOT); @@ -1914,17 +1928,7 @@ fts_alloc (FTS *sp, const char *name, register size_t namelen) * The file name is a variable length array. Allocate the FTSENT * structure and the file name in one chunk. */ - len = offsetof(FTSENT, fts_name) + namelen + 1; - /* Align the allocation size so that it works for FTSENT, - so that trailing padding may be referenced by direct access - to the flexible array members, without triggering undefined behavior - by accessing bytes beyond the heap allocation. This implicit access - was seen for example with ISDOT() and GCC 5.1.1 at -O2. - Do not use alignof (FTSENT) here, since C11 prohibits - taking the alignment of a structure containing a flexible - array member. */ - len += alignof (max_align_t) - 1; - len &= ~ (alignof (max_align_t) - 1); + len = FLEXSIZEOF(FTSENT, fts_name, namelen + 1); if ((p = malloc(len)) == NULL) return (NULL); diff --git a/gl/lib/fts_.h b/gl/lib/fts_.h index b9a3f12..70cc9e3 100644 --- a/gl/lib/fts_.h +++ b/gl/lib/fts_.h @@ -1,6 +1,6 @@ /* Traverse a file hierarchy. - Copyright (C) 2004-2015 Free Software Foundation, Inc. + Copyright (C) 2004-2018 Free Software Foundation, Inc. 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 @@ -13,7 +13,7 @@ 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, see . */ + along with this program. If not, see . */ /* * Copyright (c) 1989, 1993 @@ -220,7 +220,11 @@ typedef struct _ftsent { ptrdiff_t fts_level; /* depth (-1 to N) */ size_t fts_namelen; /* strlen(fts_name) */ - nlink_t fts_n_dirs_remaining; /* count down from st_nlink */ + + /* If not (nlink_t) -1, an upper bound on the number of + remaining subdirectories of interest. If this becomes + zero, some work can be avoided. */ + nlink_t fts_n_dirs_remaining; # define FTS_D 1 /* preorder directory */ # define FTS_DC 2 /* directory that causes cycles */ -- 2.13.6 From ea88dd373c60feab541fe037369805f326dc3494 Mon Sep 17 00:00:00 2001 From: rpm-build Date: Mon, 12 Feb 2018 18:58:30 +0100 Subject: [PATCH 2/4] fts: remove dependency on gnulib's fleximember.h ... by reverting upstream commit edb9d82948cb23f67a19e1b435047a0570225df3 --- gl/lib/fts.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-) diff --git a/gl/lib/fts.c b/gl/lib/fts.c index bfa73e3..c37ebe2 100644 --- a/gl/lib/fts.c +++ b/gl/lib/fts.c @@ -71,7 +71,6 @@ static char sccsid[] = "@(#)fts.c 8.6 (Berkeley) 8/14/94"; #if ! _LIBC # include "fcntl--.h" -# include "flexmember.h" # include "openat.h" # include "same-inode.h" #endif @@ -1928,7 +1927,17 @@ fts_alloc (FTS *sp, const char *name, register size_t namelen) * The file name is a variable length array. Allocate the FTSENT * structure and the file name in one chunk. */ - len = FLEXSIZEOF(FTSENT, fts_name, namelen + 1); + len = offsetof(FTSENT, fts_name) + namelen + 1; + /* Align the allocation size so that it works for FTSENT, + so that trailing padding may be referenced by direct access + to the flexible array members, without triggering undefined behavior + by accessing bytes beyond the heap allocation. This implicit access + was seen for example with ISDOT() and GCC 5.1.1 at -O2. + Do not use alignof (FTSENT) here, since C11 prohibits + taking the alignment of a structure containing a flexible + array member. */ + len += alignof (max_align_t) - 1; + len &= ~ (alignof (max_align_t) - 1); if ((p = malloc(len)) == NULL) return (NULL); -- 2.13.6 From 9c1720c99bbf8998dfdaa5976bca8bdc6d93f8e7 Mon Sep 17 00:00:00 2001 From: Paul Eggert Date: Thu, 5 Apr 2018 08:48:01 -0700 Subject: [PATCH 3/4] fts: treat CIFS like NFS Problem reported by Kamil Dudka in: https://lists.gnu.org/r/bug-gnulib/2018-04/msg00015.html * lib/fts.c (S_MAGIC_CIFS): New macro. (dirent_inode_sort_may_be_useful, leaf_optimization): Treat CIFS like NFS. Upstream-commit: 2e53df541a30d438859087ed4b5a396e04697b9b Signed-off-by: Kamil Dudka --- gl/lib/fts.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/gl/lib/fts.c b/gl/lib/fts.c index c37ebe2..508ceac 100644 --- a/gl/lib/fts.c +++ b/gl/lib/fts.c @@ -684,6 +684,7 @@ enum leaf_optimization /* Linux-specific constants from coreutils' src/fs.h */ # define S_MAGIC_AFS 0x5346414F +# define S_MAGIC_CIFS 0xFF534D42 # define S_MAGIC_NFS 0x6969 # define S_MAGIC_PROC 0x9FA0 # define S_MAGIC_REISERFS 0x52654973 @@ -791,8 +792,9 @@ dirent_inode_sort_may_be_useful (FTSENT const *p) switch (filesystem_type (p)) { - case S_MAGIC_TMPFS: + case S_MAGIC_CIFS: case S_MAGIC_NFS: + case S_MAGIC_TMPFS: /* On a file system of any of these types, sorting is unnecessary, and hence wasteful. */ return false; @@ -826,6 +828,10 @@ leaf_optimization (FTSENT const *p) /* Although AFS mount points are not counted in st_nlink, they act like directories. See . */ FALLTHROUGH; + case S_MAGIC_CIFS: + /* Leaf optimization causes 'find' to abort. See + . */ + FALLTHROUGH; case S_MAGIC_NFS: /* NFS provides usable dirent.d_type but not necessarily for all entries of large directories, so as per -- 2.14.3 From ff64329a046e76ba553c15373ed61bbed814d286 Mon Sep 17 00:00:00 2001 From: Paul Eggert Date: Wed, 11 Apr 2018 12:50:35 -0700 Subject: [PATCH 4/4] fts: fix bug in find across filesystems This fixes a bug I introduced last summer. Problem reported by Kamil Dudka in: https://lists.gnu.org/r/bug-gnulib/2018-04/msg00033.html * lib/fts.c (filesystem_type, dirent_inode_sort_may_be_useful) (leaf_optimization): New arg for file descriptor. All callers changed. (fts_build): Check for whether inodes should be sorted before closing the directory. Upstream-commit: 81b8c0d3be98f5a77403599de3d06329b3e7673e Signed-off-by: Kamil Dudka --- gl/lib/fts.c | 55 +++++++++++++++++++++++++++++++------------------------ 1 file changed, 31 insertions(+), 24 deletions(-) diff --git a/gl/lib/fts.c b/gl/lib/fts.c index 508ceac..175f12a 100644 --- a/gl/lib/fts.c +++ b/gl/lib/fts.c @@ -725,11 +725,12 @@ dev_type_compare (void const *x, void const *y) return ax->st_dev == ay->st_dev; } -/* Return the file system type of P, or 0 if not known. +/* Return the file system type of P with file descriptor FD, or 0 if not known. + If FD is negative, P's file descriptor is unavailable. Try to cache known values. */ static fsword -filesystem_type (FTSENT const *p) +filesystem_type (FTSENT const *p, int fd) { FTS *sp = p->fts_fts; Hash_table *h = sp->fts_leaf_optimization_works_ht; @@ -755,7 +756,7 @@ filesystem_type (FTSENT const *p) } /* Look-up failed. Query directly and cache the result. */ - if (fstatfs (p->fts_fts->fts_cwd_fd, &fs_buf) != 0) + if (fd < 0 || fstatfs (fd, &fs_buf) != 0) return 0; if (h) @@ -777,12 +778,12 @@ filesystem_type (FTSENT const *p) return fs_buf.f_type; } -/* Return false if it is easy to determine the file system type of the - directory P, and sorting dirents on inode numbers is known not to - improve traversal performance with that type of file system. - Otherwise, return true. */ +/* Return true if sorting dirents on inode numbers is known to improve + traversal performance for the directory P with descriptor DIR_FD. + Return false otherwise. When in doubt, return true. + DIR_FD is negative if unavailable. */ static bool -dirent_inode_sort_may_be_useful (FTSENT const *p) +dirent_inode_sort_may_be_useful (FTSENT const *p, int dir_fd) { /* Skip the sort only if we can determine efficiently that skipping it is the right thing to do. @@ -790,7 +791,7 @@ dirent_inode_sort_may_be_useful (FTSENT const *p) while the cost of *not* performing it can be O(N^2) with a very large constant. */ - switch (filesystem_type (p)) + switch (filesystem_type (p, dir_fd)) { case S_MAGIC_CIFS: case S_MAGIC_NFS: @@ -804,16 +805,17 @@ dirent_inode_sort_may_be_useful (FTSENT const *p) } } -/* Given an FTS entry P for a directory D, +/* Given an FTS entry P for a directory with descriptor DIR_FD, return true if it is both useful and valid to apply leaf optimization. The optimization is useful only for file systems that lack usable dirent.d_type info. The optimization is valid if an st_nlink value of at least MIN_DIR_NLINK is an upper bound on the number of - subdirectories of D, counting "." and ".." as subdirectories. */ + subdirectories of D, counting "." and ".." as subdirectories. + DIR_FD is negative if unavailable. */ static enum leaf_optimization -leaf_optimization (FTSENT const *p) +leaf_optimization (FTSENT const *p, int dir_fd) { - switch (filesystem_type (p)) + switch (filesystem_type (p, dir_fd)) { /* List here the file system types that may lack usable dirent.d_type info, yet for which the optimization does apply. */ @@ -850,12 +852,13 @@ leaf_optimization (FTSENT const *p) #else static bool -dirent_inode_sort_may_be_useful (FTSENT const *p _GL_UNUSED) +dirent_inode_sort_may_be_useful (FTSENT const *p _GL_UNUSED, + int dir_fd _GL_UNUSED) { return true; } static enum leaf_optimization -leaf_optimization (FTSENT const *p _GL_UNUSED) +leaf_optimization (FTSENT const *p _GL_UNUSED, int dir_fd _GL_UNUSED) { return NO_LEAF_OPTIMIZATION; } @@ -1049,7 +1052,7 @@ check_for_dir: if (parent->fts_n_dirs_remaining == 0 && ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL) - && (leaf_optimization (parent) + && (leaf_optimization (parent, sp->fts_cwd_fd) == NOSTAT_LEAF_OPTIMIZATION)) { /* nothing more needed */ @@ -1334,6 +1337,7 @@ fts_build (register FTS *sp, int type) int dir_fd; FTSENT *cur = sp->fts_cur; bool continue_readdir = !!cur->fts_dirp; + bool sort_by_inode = false; size_t max_entries; /* When cur->fts_dirp is non-NULL, that means we should @@ -1427,7 +1431,7 @@ fts_build (register FTS *sp, int type) && ! (ISSET (FTS_NOSTAT) && ISSET (FTS_PHYSICAL) && ! ISSET (FTS_SEEDOT) && cur->fts_statp->st_nlink == MIN_DIR_NLINK - && (leaf_optimization (cur) + && (leaf_optimization (cur, dir_fd) != NO_LEAF_OPTIMIZATION))); if (descend || type == BREAD) { @@ -1588,6 +1592,15 @@ mem1: saved_errno = errno; tail->fts_link = p; tail = p; } + + /* 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, arrange to sort the + directory entries on increasing inode numbers. */ + if (nitems == _FTS_INODE_SORT_DIR_ENTRIES_THRESHOLD + && !sp->fts_compar) + sort_by_inode = dirent_inode_sort_may_be_useful (cur, dir_fd); + ++nitems; if (max_entries <= nitems) { /* When there are too many dir entries, leave @@ -1645,13 +1658,7 @@ mem1: saved_errno = errno; 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 - && dirent_inode_sort_may_be_useful (cur)) { + if (sort_by_inode) { sp->fts_compar = fts_compare_ino; head = fts_sort (sp, head, nitems); sp->fts_compar = NULL; -- 2.14.3