2 Copyright (C) Andrew Tridgell 1996
3 Copyright (C) Paul Mackerras 1996
4 Copyright (C) 2001, 2002 by Martin Pool <mbp@samba.org>
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 * Generate and receive file lists
24 * @sa http://lists.samba.org/pipermail/rsync/2000-June/002351.html
36 extern int do_progress;
37 extern int always_checksum;
39 extern int ignore_errors;
40 extern int numeric_ids;
43 extern int filesfrom_fd;
44 extern int one_file_system;
45 extern int copy_dirlinks;
46 extern int keep_dirlinks;
47 extern int preserve_links;
48 extern int preserve_hard_links;
49 extern int preserve_devices;
50 extern int preserve_specials;
51 extern int preserve_uid;
52 extern int preserve_gid;
53 extern int relative_paths;
54 extern int implied_dirs;
55 extern int prune_empty_dirs;
56 extern int copy_links;
57 extern int copy_unsafe_links;
58 extern int protocol_version;
59 extern int sanitize_paths;
60 extern const char *io_write_phase;
61 extern struct stats stats;
62 extern struct file_list *the_file_list;
64 extern char curr_dir[MAXPATHLEN];
66 extern struct chmod_mode_struct *chmod_modes;
68 extern struct filter_list_struct filter_list;
69 extern struct filter_list_struct server_filter_list;
73 dev_t filesystem_dev; /* used to implement -x */
74 unsigned int file_struct_len;
76 static char empty_sum[MD4_SUM_LENGTH];
77 static int flist_count_offset;
79 static void clean_flist(struct file_list *flist, int strip_root, int no_dups);
80 static void output_flist(struct file_list *flist);
86 /* Figure out how big the file_struct is without trailing padding */
87 file_struct_len = offsetof(struct file_struct, flags) + sizeof f.flags;
88 checksum_len = protocol_version < 21 ? 2 : MD4_SUM_LENGTH;
91 static int show_filelist_p(void)
93 return verbose && xfer_dirs && !am_server;
96 static void start_filelist_progress(char *kind)
98 rprintf(FINFO, "%s ... ", kind);
99 if (verbose > 1 || do_progress)
100 rprintf(FINFO, "\n");
104 static void emit_filelist_progress(int count)
106 rprintf(FINFO, " %d files...\r", count);
109 static void maybe_emit_filelist_progress(int count)
111 if (do_progress && show_filelist_p() && (count % 100) == 0)
112 emit_filelist_progress(count);
115 static void finish_filelist_progress(const struct file_list *flist)
118 /* This overwrites the progress line */
119 rprintf(FINFO, "%d file%sto consider\n",
120 flist->count, flist->count == 1 ? " " : "s ");
122 rprintf(FINFO, "done\n");
125 void show_flist_stats(void)
130 static void list_file_entry(struct file_struct *f)
132 char permbuf[PERMSTRING_SIZE];
135 /* this can happen if duplicate names were removed */
139 permstring(permbuf, f->mode);
142 if (preserve_links && S_ISLNK(f->mode)) {
143 rprintf(FINFO, "%s %11.0f %s %s -> %s\n",
145 (double)f->length, timestring(f->modtime),
146 f_name(f, NULL), f->u.link);
150 rprintf(FINFO, "%s %11.0f %s %s\n",
152 (double)f->length, timestring(f->modtime),
158 * Stat either a symlink or its referent, depending on the settings of
159 * copy_links, copy_unsafe_links, etc.
161 * @retval -1 on error
163 * @retval 0 for success
165 * @post If @p path is a symlink, then @p linkbuf (of size @c
166 * MAXPATHLEN) contains the symlink target.
168 * @post @p buffer contains information about the link or the
169 * referrent as appropriate, if they exist.
171 static int readlink_stat(const char *path, STRUCT_STAT *buffer, char *linkbuf)
175 return do_stat(path, buffer);
176 if (link_stat(path, buffer, copy_dirlinks) < 0)
178 if (S_ISLNK(buffer->st_mode)) {
179 int l = readlink((char *)path, linkbuf, MAXPATHLEN - 1);
183 if (copy_unsafe_links && unsafe_symlink(linkbuf, path)) {
185 rprintf(FINFO,"copying unsafe symlink \"%s\" -> \"%s\"\n",
188 return do_stat(path, buffer);
193 return do_stat(path, buffer);
197 int link_stat(const char *path, STRUCT_STAT *buffer, int follow_dirlinks)
201 return do_stat(path, buffer);
202 if (do_lstat(path, buffer) < 0)
204 if (follow_dirlinks && S_ISLNK(buffer->st_mode)) {
206 if (do_stat(path, &st) == 0 && S_ISDIR(st.st_mode))
211 return do_stat(path, buffer);
215 /* This function is used to check if a file should be included/excluded
216 * from the list of files based on its name and type etc. The value of
217 * filter_level is set to either SERVER_FILTERS or ALL_FILTERS. */
218 static int is_excluded(char *fname, int is_dir, int filter_level)
220 #if 0 /* This currently never happens, so avoid a useless compare. */
221 if (filter_level == NO_FILTERS)
225 /* never exclude '.', even if somebody does --exclude '*' */
226 if (fname[0] == '.' && !fname[1])
228 /* Handle the -R version of the '.' dir. */
229 if (fname[0] == '/') {
230 int len = strlen(fname);
231 if (fname[len-1] == '.' && fname[len-2] == '/')
235 if (server_filter_list.head
236 && check_filter(&server_filter_list, fname, is_dir) < 0)
238 if (filter_level != ALL_FILTERS)
241 && check_filter(&filter_list, fname, is_dir) < 0)
246 static int to_wire_mode(mode_t mode)
249 if (S_ISLNK(mode) && (_S_IFLNK != 0120000))
250 return (mode & ~(_S_IFMT)) | 0120000;
255 static mode_t from_wire_mode(int mode)
257 if ((mode & (_S_IFMT)) == 0120000 && (_S_IFLNK != 0120000))
258 return (mode & ~(_S_IFMT)) | _S_IFLNK;
262 static void send_directory(int f, struct file_list *flist,
263 char *fbuf, int len);
265 static char *flist_dir;
266 static int flist_dir_len;
270 * Make sure @p flist is big enough to hold at least @p flist->count
273 void flist_expand(struct file_list *flist)
275 struct file_struct **new_ptr;
277 if (flist->count < flist->malloced)
280 if (flist->malloced < FLIST_START)
281 flist->malloced = FLIST_START;
282 else if (flist->malloced >= FLIST_LINEAR)
283 flist->malloced += FLIST_LINEAR;
285 flist->malloced *= 2;
288 * In case count jumped or we are starting the list
289 * with a known size just set it.
291 if (flist->malloced < flist->count)
292 flist->malloced = flist->count;
294 new_ptr = realloc_array(flist->files, struct file_struct *,
297 if (verbose >= 2 && flist->malloced != FLIST_START) {
298 rprintf(FINFO, "[%s] expand file_list to %.0f bytes, did%s move\n",
300 (double)sizeof flist->files[0] * flist->malloced,
301 (new_ptr == flist->files) ? " not" : "");
304 flist->files = new_ptr;
307 out_of_memory("flist_expand");
310 static void send_file_entry(struct file_struct *file, int f)
312 unsigned short flags;
313 static time_t modtime;
317 static uint32 rdev_major;
320 static char lastname[MAXPATHLEN];
321 char fname[MAXPATHLEN];
329 modtime = 0, mode = 0;
330 dev = 0, rdev = makedev(0, 0);
337 io_write_phase = "send_file_entry";
341 flags = file->flags & XMIT_TOP_DIR;
343 if (file->mode == mode)
344 flags |= XMIT_SAME_MODE;
347 if ((preserve_devices && IS_DEVICE(mode))
348 || (preserve_specials && IS_SPECIAL(mode))) {
349 if (protocol_version < 28) {
350 if (file->u.rdev == rdev)
351 flags |= XMIT_SAME_RDEV_pre28;
356 if ((uint32)major(rdev) == rdev_major)
357 flags |= XMIT_SAME_RDEV_MAJOR;
359 rdev_major = major(rdev);
360 if ((uint32)minor(rdev) <= 0xFFu)
361 flags |= XMIT_RDEV_MINOR_IS_SMALL;
363 } else if (protocol_version < 28)
364 rdev = makedev(0, 0);
365 if (file->uid == uid)
366 flags |= XMIT_SAME_UID;
369 if (file->gid == gid)
370 flags |= XMIT_SAME_GID;
373 if (file->modtime == modtime)
374 flags |= XMIT_SAME_TIME;
376 modtime = file->modtime;
378 #ifdef SUPPORT_HARD_LINKS
379 if (file->link_u.idev) {
380 if (file->F_DEV == dev) {
381 if (protocol_version >= 28)
382 flags |= XMIT_SAME_DEV;
385 flags |= XMIT_HAS_IDEV_DATA;
390 lastname[l1] && (fname[l1] == lastname[l1]) && (l1 < 255);
392 l2 = strlen(fname+l1);
395 flags |= XMIT_SAME_NAME;
397 flags |= XMIT_LONG_NAME;
399 /* We must make sure we don't send a zero flag byte or the
400 * other end will terminate the flist transfer. Note that
401 * the use of XMIT_TOP_DIR on a non-dir has no meaning, so
402 * it's harmless way to add a bit to the first flag byte. */
403 if (protocol_version >= 28) {
404 if (!flags && !S_ISDIR(mode))
405 flags |= XMIT_TOP_DIR;
406 if ((flags & 0xFF00) || !flags) {
407 flags |= XMIT_EXTENDED_FLAGS;
408 write_byte(f, flags);
409 write_byte(f, flags >> 8);
411 write_byte(f, flags);
414 flags |= S_ISDIR(mode) ? XMIT_LONG_NAME : XMIT_TOP_DIR;
415 write_byte(f, flags);
417 if (flags & XMIT_SAME_NAME)
419 if (flags & XMIT_LONG_NAME)
423 write_buf(f, fname + l1, l2);
425 write_longint(f, file->length);
426 if (!(flags & XMIT_SAME_TIME))
427 write_int(f, modtime);
428 if (!(flags & XMIT_SAME_MODE))
429 write_int(f, to_wire_mode(mode));
430 if (preserve_uid && !(flags & XMIT_SAME_UID)) {
435 if (preserve_gid && !(flags & XMIT_SAME_GID)) {
440 if ((preserve_devices && IS_DEVICE(mode))
441 || (preserve_specials && IS_SPECIAL(mode))) {
442 if (protocol_version < 28) {
443 if (!(flags & XMIT_SAME_RDEV_pre28))
444 write_int(f, (int)rdev);
446 if (!(flags & XMIT_SAME_RDEV_MAJOR))
447 write_int(f, major(rdev));
448 if (flags & XMIT_RDEV_MINOR_IS_SMALL)
449 write_byte(f, minor(rdev));
451 write_int(f, minor(rdev));
456 if (preserve_links && S_ISLNK(mode)) {
457 int len = strlen(file->u.link);
459 write_buf(f, file->u.link, len);
463 #ifdef SUPPORT_HARD_LINKS
464 if (flags & XMIT_HAS_IDEV_DATA) {
465 if (protocol_version < 26) {
466 /* 32-bit dev_t and ino_t */
468 write_int(f, file->F_INODE);
470 /* 64-bit dev_t and ino_t */
471 if (!(flags & XMIT_SAME_DEV))
472 write_longint(f, dev);
473 write_longint(f, file->F_INODE);
478 if (always_checksum && (S_ISREG(mode) || protocol_version < 28)) {
483 /* Prior to 28, we sent a useless set of nulls. */
486 write_buf(f, sum, checksum_len);
489 strlcpy(lastname, fname, MAXPATHLEN);
491 io_write_phase = "unknown";
494 static struct file_struct *receive_file_entry(struct file_list *flist,
495 unsigned short flags, int f)
497 static time_t modtime;
501 static uint32 rdev_major;
504 static char lastname[MAXPATHLEN], *lastdir;
505 static int lastdir_depth, lastdir_len = -1;
506 static unsigned int del_hier_name_len = 0;
507 static int in_del_hier = 0;
508 char thisname[MAXPATHLEN];
509 unsigned int l1 = 0, l2 = 0;
510 int alloc_len, basename_len, dirname_len, linkname_len, sum_len;
512 char *basename, *dirname, *bp;
513 struct file_struct *file;
516 modtime = 0, mode = 0;
517 dev = 0, rdev = makedev(0, 0);
526 if (flags & XMIT_SAME_NAME)
529 if (flags & XMIT_LONG_NAME)
534 if (l2 >= MAXPATHLEN - l1) {
536 "overflow: flags=0x%x l1=%d l2=%d lastname=%s\n",
537 flags, l1, l2, lastname);
538 overflow_exit("receive_file_entry");
541 strlcpy(thisname, lastname, l1 + 1);
542 read_sbuf(f, &thisname[l1], l2);
543 thisname[l1 + l2] = 0;
545 strlcpy(lastname, thisname, MAXPATHLEN);
547 clean_fname(thisname, 0);
550 sanitize_path(thisname, thisname, "", 0);
552 if ((basename = strrchr(thisname, '/')) != NULL) {
553 dirname_len = ++basename - thisname; /* counts future '\0' */
554 if (lastdir_len == dirname_len - 1
555 && strncmp(thisname, lastdir, lastdir_len) == 0) {
557 dirname_len = 0; /* indicates no copy is needed */
565 basename_len = strlen(basename) + 1; /* count the '\0' */
567 file_length = read_longint(f);
568 if (!(flags & XMIT_SAME_TIME))
569 modtime = (time_t)read_int(f);
570 if (!(flags & XMIT_SAME_MODE))
571 mode = from_wire_mode(read_int(f));
573 if (chmod_modes && !S_ISLNK(mode))
574 mode = tweak_mode(mode, chmod_modes);
576 if (preserve_uid && !(flags & XMIT_SAME_UID))
577 uid = (uid_t)read_int(f);
578 if (preserve_gid && !(flags & XMIT_SAME_GID))
579 gid = (gid_t)read_int(f);
581 if ((preserve_devices && IS_DEVICE(mode))
582 || (preserve_specials && IS_SPECIAL(mode))) {
583 if (protocol_version < 28) {
584 if (!(flags & XMIT_SAME_RDEV_pre28))
585 rdev = (dev_t)read_int(f);
588 if (!(flags & XMIT_SAME_RDEV_MAJOR))
589 rdev_major = read_int(f);
590 if (flags & XMIT_RDEV_MINOR_IS_SMALL)
591 rdev_minor = read_byte(f);
593 rdev_minor = read_int(f);
594 rdev = makedev(rdev_major, rdev_minor);
596 } else if (protocol_version < 28)
597 rdev = makedev(0, 0);
600 if (preserve_links && S_ISLNK(mode)) {
601 linkname_len = read_int(f) + 1; /* count the '\0' */
602 if (linkname_len <= 0 || linkname_len > MAXPATHLEN) {
603 rprintf(FERROR, "overflow: linkname_len=%d\n",
605 overflow_exit("receive_file_entry");
612 sum_len = always_checksum && S_ISREG(mode) ? MD4_SUM_LENGTH : 0;
614 alloc_len = file_struct_len + dirname_len + basename_len
615 + linkname_len + sum_len;
616 bp = pool_alloc(flist->file_pool, alloc_len, "receive_file_entry");
618 file = (struct file_struct *)bp;
619 memset(bp, 0, file_struct_len);
620 bp += file_struct_len;
622 file->modtime = modtime;
623 file->length = file_length;
629 file->dirname = lastdir = bp;
630 lastdir_len = dirname_len - 1;
631 memcpy(bp, dirname, dirname_len - 1);
634 lastdir_depth = count_dir_elements(lastdir);
635 file->dir.depth = lastdir_depth + 1;
636 } else if (dirname) {
637 file->dirname = dirname; /* we're reusing lastname */
638 file->dir.depth = lastdir_depth + 1;
643 if (basename_len == 1+1 && *basename == '.') /* +1 for '\0' */
645 if (flags & XMIT_TOP_DIR) {
646 in_del_hier = recurse;
647 del_hier_name_len = file->dir.depth == 0 ? 0 : l1 + l2;
648 if (relative_paths && del_hier_name_len > 2
649 && lastname[del_hier_name_len-1] == '.'
650 && lastname[del_hier_name_len-2] == '/')
651 del_hier_name_len -= 2;
652 file->flags |= FLAG_TOP_DIR | FLAG_DEL_HERE;
653 } else if (in_del_hier) {
654 if (!relative_paths || !del_hier_name_len
655 || (l1 >= del_hier_name_len
656 && lastname[del_hier_name_len] == '/'))
657 file->flags |= FLAG_DEL_HERE;
664 memcpy(bp, basename, basename_len);
667 if ((preserve_devices && IS_DEVICE(mode))
668 || (preserve_specials && IS_SPECIAL(mode)))
674 read_sbuf(f, bp, linkname_len - 1);
676 sanitize_path(bp, bp, "", lastdir_depth);
681 #ifdef SUPPORT_HARD_LINKS
682 if (preserve_hard_links && protocol_version < 28 && S_ISREG(mode))
683 flags |= XMIT_HAS_IDEV_DATA;
684 if (flags & XMIT_HAS_IDEV_DATA) {
686 if (protocol_version < 26) {
690 if (!(flags & XMIT_SAME_DEV))
691 dev = read_longint(f);
692 inode = read_longint(f);
694 if (flist->hlink_pool) {
695 file->link_u.idev = pool_talloc(flist->hlink_pool,
696 struct idev, 1, "inode_table");
697 file->F_INODE = inode;
703 if (always_checksum && (sum_len || protocol_version < 28)) {
706 file->u.sum = sum = bp;
709 /* Prior to 28, we get a useless set of nulls. */
712 read_buf(f, sum, checksum_len);
719 * Create a file_struct for a named file by reading its stat()
720 * information and performing extensive checks against global
723 * @return the new file, or NULL if there was an error or this file
724 * should be excluded.
726 * @todo There is a small optimization opportunity here to avoid
727 * stat()ing the file in some circumstances, which has a certain cost.
728 * We are called immediately after doing readdir(), and so we may
729 * already know the d_type of the file. We could for example avoid
730 * statting directories if we're not recursing, but this is not a very
731 * important case. Some systems may not have d_type.
733 struct file_struct *make_file(char *fname, struct file_list *flist,
734 STRUCT_STAT *stp, unsigned short flags,
737 static char *lastdir;
738 static int lastdir_len = -1;
739 struct file_struct *file;
741 char sum[SUM_LENGTH];
742 char thisname[MAXPATHLEN];
743 char linkname[MAXPATHLEN];
744 int alloc_len, basename_len, dirname_len, linkname_len, sum_len;
745 char *basename, *dirname, *bp;
747 if (!flist || !flist->count) /* Ignore lastdir when invalid. */
750 if (strlcpy(thisname, fname, sizeof thisname)
751 >= sizeof thisname - flist_dir_len) {
752 rprintf(FINFO, "skipping overly long name: %s\n", fname);
755 clean_fname(thisname, 0);
757 sanitize_path(thisname, thisname, "", 0);
759 memset(sum, 0, SUM_LENGTH);
761 if (stp && S_ISDIR(stp->st_mode))
762 st = *stp; /* Needed for "symlink/." with --relative. */
763 else if (readlink_stat(thisname, &st, linkname) != 0) {
764 int save_errno = errno;
765 /* See if file is excluded before reporting an error. */
766 if (filter_level != NO_FILTERS
767 && is_excluded(thisname, 0, filter_level))
769 if (save_errno == ENOENT) {
771 /* Avoid "vanished" error if symlink points nowhere. */
772 if (copy_links && do_lstat(thisname, &st) == 0
773 && S_ISLNK(st.st_mode)) {
774 io_error |= IOERR_GENERAL;
775 rprintf(FERROR, "symlink has no referent: %s\n",
776 full_fname(thisname));
780 enum logcode c = am_daemon && protocol_version < 28
782 io_error |= IOERR_VANISHED;
783 rprintf(c, "file has vanished: %s\n",
784 full_fname(thisname));
787 io_error |= IOERR_GENERAL;
788 rsyserr(FERROR, save_errno, "readlink %s failed",
789 full_fname(thisname));
794 /* backup.c calls us with filter_level set to NO_FILTERS. */
795 if (filter_level == NO_FILTERS)
798 if (S_ISDIR(st.st_mode) && !xfer_dirs) {
799 rprintf(FINFO, "skipping directory %s\n", thisname);
803 /* We only care about directories because we need to avoid recursing
804 * into a mount-point directory, not to avoid copying a symlinked
805 * file if -L (or similar) was specified. */
806 if (one_file_system && st.st_dev != filesystem_dev
807 && S_ISDIR(st.st_mode)) {
808 if (one_file_system > 1) {
810 rprintf(FINFO, "skipping mount-point dir %s\n",
815 flags |= FLAG_MOUNT_POINT;
818 if (is_excluded(thisname, S_ISDIR(st.st_mode) != 0, filter_level))
821 if (lp_ignore_nonreadable(module_id)) {
823 if (!S_ISLNK(st.st_mode))
825 if (access(thisname, R_OK) != 0)
832 rprintf(FINFO, "[%s] make_file(%s,*,%d)\n",
833 who_am_i(), thisname, filter_level);
836 if ((basename = strrchr(thisname, '/')) != NULL) {
837 dirname_len = ++basename - thisname; /* counts future '\0' */
838 if (lastdir_len == dirname_len - 1
839 && strncmp(thisname, lastdir, lastdir_len) == 0) {
841 dirname_len = 0; /* indicates no copy is needed */
849 basename_len = strlen(basename) + 1; /* count the '\0' */
852 linkname_len = S_ISLNK(st.st_mode) ? strlen(linkname) + 1 : 0;
857 sum_len = always_checksum && am_sender && S_ISREG(st.st_mode)
858 ? MD4_SUM_LENGTH : 0;
860 alloc_len = file_struct_len + dirname_len + basename_len
861 + linkname_len + sum_len;
863 bp = pool_alloc(flist->file_pool, alloc_len, "make_file");
865 if (!(bp = new_array(char, alloc_len)))
866 out_of_memory("make_file");
869 file = (struct file_struct *)bp;
870 memset(bp, 0, file_struct_len);
871 bp += file_struct_len;
874 file->modtime = st.st_mtime;
875 file->length = st.st_size;
876 file->mode = st.st_mode;
877 file->uid = st.st_uid;
878 file->gid = st.st_gid;
880 #ifdef SUPPORT_HARD_LINKS
881 if (flist && flist->hlink_pool) {
882 if (protocol_version < 28) {
883 if (S_ISREG(st.st_mode))
884 file->link_u.idev = pool_talloc(
885 flist->hlink_pool, struct idev, 1,
888 if (!S_ISDIR(st.st_mode) && st.st_nlink > 1)
889 file->link_u.idev = pool_talloc(
890 flist->hlink_pool, struct idev, 1,
894 if (file->link_u.idev) {
895 file->F_DEV = st.st_dev;
896 file->F_INODE = st.st_ino;
901 file->dirname = lastdir = bp;
902 lastdir_len = dirname_len - 1;
903 memcpy(bp, dirname, dirname_len - 1);
907 file->dirname = dirname;
910 memcpy(bp, basename, basename_len);
913 #ifdef HAVE_STRUCT_STAT_ST_RDEV
914 if ((preserve_devices && IS_DEVICE(st.st_mode))
915 || (preserve_specials && IS_SPECIAL(st.st_mode)))
916 file->u.rdev = st.st_rdev;
922 memcpy(bp, linkname, linkname_len);
929 file_checksum(thisname, bp, st.st_size);
933 file->dir.root = flist_dir;
935 /* This code is only used by the receiver when it is building
936 * a list of files for a delete pass. */
937 if (keep_dirlinks && linkname_len && flist) {
939 int save_mode = file->mode;
940 file->mode = S_IFDIR; /* Find a directory with our name. */
941 if (flist_find(the_file_list, file) >= 0
942 && do_stat(thisname, &st2) == 0 && S_ISDIR(st2.st_mode)) {
943 file->modtime = st2.st_mtime;
944 file->length = st2.st_size;
945 file->mode = st2.st_mode;
946 file->uid = st2.st_uid;
947 file->gid = st2.st_gid;
950 file->mode = save_mode;
953 if (S_ISREG(st.st_mode) || S_ISLNK(st.st_mode))
954 stats.total_size += st.st_size;
959 static struct file_struct *send_file_name(int f, struct file_list *flist,
960 char *fname, STRUCT_STAT *stp,
961 unsigned short flags)
963 struct file_struct *file;
965 file = make_file(fname, flist, stp, flags,
966 f == -2 ? SERVER_FILTERS : ALL_FILTERS);
970 if (chmod_modes && !S_ISLNK(file->mode))
971 file->mode = tweak_mode(file->mode, chmod_modes);
973 maybe_emit_filelist_progress(flist->count + flist_count_offset);
977 if (file->basename[0]) {
978 flist->files[flist->count++] = file;
979 send_file_entry(file, f);
984 static void send_if_directory(int f, struct file_list *flist,
985 struct file_struct *file,
986 char *fbuf, unsigned int ol)
988 char is_dot_dir = fbuf[ol-1] == '.' && (ol == 1 || fbuf[ol-2] == '/');
990 if (S_ISDIR(file->mode)
991 && !(file->flags & FLAG_MOUNT_POINT) && f_name(file, fbuf)) {
993 unsigned int len = strlen(fbuf);
994 if (len > 1 && fbuf[len-1] == '/')
996 if (len >= MAXPATHLEN - 1) {
997 io_error |= IOERR_GENERAL;
998 rprintf(FERROR, "skipping long-named directory: %s\n",
1002 save_filters = push_local_filters(fbuf, len);
1003 send_directory(f, flist, fbuf, len);
1004 pop_local_filters(save_filters);
1011 /* This function is normally called by the sender, but the receiving side also
1012 * calls it from get_dirlist() with f set to -1 so that we just construct the
1013 * file list in memory without sending it over the wire. Also, get_dirlist()
1014 * might call this with f set to -2, which also indicates that local filter
1015 * rules should be ignored. */
1016 static void send_directory(int f, struct file_list *flist,
1017 char *fbuf, int len)
1023 int start = flist->count;
1025 if (!(d = opendir(fbuf))) {
1026 io_error |= IOERR_GENERAL;
1027 rsyserr(FERROR, errno, "opendir %s failed", full_fname(fbuf));
1032 if (len != 1 || *fbuf != '/')
1035 remainder = MAXPATHLEN - (p - fbuf);
1037 for (errno = 0, di = readdir(d); di; errno = 0, di = readdir(d)) {
1038 char *dname = d_name(di);
1039 if (dname[0] == '.' && (dname[1] == '\0'
1040 || (dname[1] == '.' && dname[2] == '\0')))
1042 if (strlcpy(p, dname, remainder) >= remainder) {
1043 io_error |= IOERR_GENERAL;
1045 "cannot send long-named file %s\n",
1050 send_file_name(f, flist, fbuf, NULL, 0);
1056 io_error |= IOERR_GENERAL;
1057 rsyserr(FERROR, errno, "readdir(%s)", full_fname(fbuf));
1063 int i, end = flist->count - 1;
1064 for (i = start; i <= end; i++)
1065 send_if_directory(f, flist, flist->files[i], fbuf, len);
1069 struct file_list *send_file_list(int f, int argc, char *argv[])
1073 char *p, *dir, olddir[sizeof curr_dir];
1074 char lastpath[MAXPATHLEN] = "";
1075 struct file_list *flist;
1076 struct timeval start_tv, end_tv;
1080 if (show_filelist_p())
1081 start_filelist_progress("building file list");
1083 start_write = stats.total_written;
1084 gettimeofday(&start_tv, NULL);
1086 flist = flist_new(WITH_HLINK, "send_file_list");
1088 io_start_buffering_out();
1089 if (filesfrom_fd >= 0) {
1090 if (argv[0] && !push_dir(argv[0])) {
1091 rsyserr(FERROR, errno, "push_dir %s failed",
1092 full_fname(argv[0]));
1093 exit_cleanup(RERR_FILESELECT);
1099 char fbuf[MAXPATHLEN];
1104 if (read_filesfrom_line(filesfrom_fd, fbuf) == 0)
1106 sanitize_path(fbuf, fbuf, "", 0);
1110 strlcpy(fbuf, *argv++, MAXPATHLEN);
1112 sanitize_path(fbuf, fbuf, "", 0);
1116 if (relative_paths) {
1117 /* We clean up fbuf below. */
1119 } else if (!len || fbuf[len - 1] == '/') {
1120 if (len == 2 && fbuf[0] == '.') {
1121 /* Turn "./" into just "." rather than "./." */
1124 if (len + 1 >= MAXPATHLEN)
1125 overflow_exit("send_file_list");
1130 } else if (len > 1 && fbuf[len-1] == '.' && fbuf[len-2] == '.'
1131 && (len == 2 || fbuf[len-3] == '/')) {
1132 if (len + 2 >= MAXPATHLEN)
1133 overflow_exit("send_file_list");
1139 is_dot_dir = fbuf[len-1] == '.'
1140 && (len == 1 || fbuf[len-2] == '/');
1143 if (link_stat(fbuf, &st, copy_dirlinks) != 0) {
1144 io_error |= IOERR_GENERAL;
1145 rsyserr(FERROR, errno, "link_stat %s failed",
1150 if (S_ISDIR(st.st_mode) && !xfer_dirs) {
1151 rprintf(FINFO, "skipping directory %s\n", fbuf);
1158 if (!relative_paths) {
1159 p = strrchr(fbuf, '/');
1166 len -= p - fbuf + 1;
1171 if ((p = strstr(fbuf, "/./")) != NULL) {
1177 len -= p - fbuf + 3;
1181 /* Get rid of trailing "/" and "/.". */
1183 if (fn[len - 1] == '/') {
1185 if (!--len && !dir) {
1190 else if (len >= 2 && fn[len - 1] == '.'
1191 && fn[len - 2] == '/') {
1193 if (!(len -= 2) && !dir) {
1201 /* Reject a ".." dir in the active part of the path. */
1202 for (p = fn; (p = strstr(p, "..")) != NULL; p += 2) {
1203 if ((p[2] == '/' || p[2] == '\0')
1204 && (p == fn || p[-1] == '/')) {
1206 "found \"..\" dir in relative path: %s\n",
1208 exit_cleanup(RERR_SYNTAX);
1219 static char *lastdir;
1220 static int lastdir_len;
1222 strlcpy(olddir, curr_dir, sizeof olddir);
1224 if (!push_dir(dir)) {
1225 io_error |= IOERR_GENERAL;
1226 rsyserr(FERROR, errno, "push_dir %s failed",
1231 if (lastdir && strcmp(lastdir, dir) == 0) {
1232 flist_dir = lastdir;
1233 flist_dir_len = lastdir_len;
1235 flist_dir = lastdir = strdup(dir);
1236 flist_dir_len = lastdir_len = strlen(dir);
1241 memmove(fbuf, fn, len + 1);
1243 if (implied_dirs && (p=strrchr(fbuf,'/')) && p != fbuf) {
1244 /* Send the implied directories at the start of the
1245 * source spec, so we get their permissions right. */
1246 char *lp = lastpath, *slash = fbuf;
1248 /* Skip any initial directories in our path that we
1249 * have in common with lastpath. */
1250 for (fn = fbuf; *fn && *lp == *fn; lp++, fn++) {
1255 if (fn != p || (*lp && *lp != '/')) {
1256 int save_copy_links = copy_links;
1257 int save_xfer_dirs = xfer_dirs;
1258 copy_links |= copy_unsafe_links;
1260 while ((slash = strchr(slash+1, '/')) != 0) {
1262 send_file_name(f, flist, fbuf, NULL, 0);
1265 copy_links = save_copy_links;
1266 xfer_dirs = save_xfer_dirs;
1268 strlcpy(lastpath, fbuf, sizeof lastpath);
1273 if (one_file_system)
1274 filesystem_dev = st.st_dev;
1276 if (recurse || (xfer_dirs && is_dot_dir)) {
1277 struct file_struct *file;
1278 file = send_file_name(f, flist, fbuf, &st, FLAG_TOP_DIR);
1280 send_if_directory(f, flist, file, fbuf, len);
1282 send_file_name(f, flist, fbuf, &st, 0);
1287 if (!pop_dir(olddir)) {
1288 rsyserr(FERROR, errno, "pop_dir %s failed",
1289 full_fname(olddir));
1290 exit_cleanup(RERR_FILESELECT);
1295 gettimeofday(&end_tv, NULL);
1296 stats.flist_buildtime = (int64)(end_tv.tv_sec - start_tv.tv_sec) * 1000
1297 + (end_tv.tv_usec - start_tv.tv_usec) / 1000;
1298 if (stats.flist_buildtime == 0)
1299 stats.flist_buildtime = 1;
1302 send_file_entry(NULL, f);
1304 if (show_filelist_p())
1305 finish_filelist_progress(flist);
1307 gettimeofday(&end_tv, NULL);
1308 stats.flist_xfertime = (int64)(end_tv.tv_sec - start_tv.tv_sec) * 1000
1309 + (end_tv.tv_usec - start_tv.tv_usec) / 1000;
1311 if (flist->hlink_pool) {
1312 pool_destroy(flist->hlink_pool);
1313 flist->hlink_pool = NULL;
1316 /* Sort the list without removing any duplicates. This allows the
1317 * receiving side to ask for any name they like, which gives us the
1318 * flexibility to change the way we unduplicate names in the future
1319 * without causing a compatibility problem with older versions. */
1320 clean_flist(flist, 0, 0);
1324 /* send the io_error flag */
1325 write_int(f, lp_ignore_errors(module_id) ? 0 : io_error);
1328 stats.flist_size = stats.total_written - start_write;
1329 stats.num_files = flist->count;
1332 output_flist(flist);
1335 rprintf(FINFO, "send_file_list done\n");
1340 struct file_list *recv_file_list(int f)
1342 struct file_list *flist;
1343 unsigned short flags;
1346 if (show_filelist_p())
1347 start_filelist_progress("receiving file list");
1349 start_read = stats.total_read;
1351 flist = flist_new(WITH_HLINK, "recv_file_list");
1354 flist->malloced = 1000;
1355 flist->files = new_array(struct file_struct *, flist->malloced);
1359 while ((flags = read_byte(f)) != 0) {
1360 struct file_struct *file;
1362 flist_expand(flist);
1364 if (protocol_version >= 28 && (flags & XMIT_EXTENDED_FLAGS))
1365 flags |= read_byte(f) << 8;
1366 file = receive_file_entry(flist, flags, f);
1368 if (S_ISREG(file->mode) || S_ISLNK(file->mode))
1369 stats.total_size += file->length;
1371 flist->files[flist->count++] = file;
1373 maybe_emit_filelist_progress(flist->count);
1376 rprintf(FINFO, "recv_file_name(%s)\n",
1377 f_name(file, NULL));
1380 receive_file_entry(NULL, 0, 0); /* Signal that we're done. */
1383 rprintf(FINFO, "received %d names\n", flist->count);
1385 if (show_filelist_p())
1386 finish_filelist_progress(flist);
1388 clean_flist(flist, relative_paths, 1);
1391 recv_uid_list(f, flist);
1393 /* Recv the io_error flag */
1394 if (lp_ignore_errors(module_id) || ignore_errors)
1397 io_error |= read_int(f);
1401 output_flist(flist);
1405 for (i = 0; i < flist->count; i++)
1406 list_file_entry(flist->files[i]);
1410 rprintf(FINFO, "recv_file_list done\n");
1412 stats.flist_size = stats.total_read - start_read;
1413 stats.num_files = flist->count;
1418 out_of_memory("recv_file_list");
1419 return NULL; /* not reached */
1422 static int file_compare(struct file_struct **file1, struct file_struct **file2)
1424 return f_name_cmp(*file1, *file2);
1427 /* Search for an identically-named item in the file list. Note that the
1428 * items must agree in their directory-ness, or no match is returned. */
1429 int flist_find(struct file_list *flist, struct file_struct *f)
1431 int low = flist->low, high = flist->high;
1432 int diff, mid, mid_up;
1434 while (low <= high) {
1435 mid = (low + high) / 2;
1436 if (flist->files[mid]->basename)
1439 /* Scan for the next non-empty entry using the cached
1440 * distance values. If the value isn't fully up-to-
1441 * date, update it. */
1442 mid_up = mid + flist->files[mid]->dir.depth;
1443 if (!flist->files[mid_up]->basename) {
1445 mid_up += flist->files[mid_up]->dir.depth;
1446 } while (!flist->files[mid_up]->basename);
1447 flist->files[mid]->dir.depth = mid_up - mid;
1449 if (mid_up > high) {
1450 /* If there's nothing left above us, set high to
1451 * a non-empty entry below us and continue. */
1452 high = mid - flist->files[mid]->length;
1453 if (!flist->files[high]->basename) {
1455 high -= flist->files[high]->length;
1456 } while (!flist->files[high]->basename);
1457 flist->files[mid]->length = mid - high;
1462 diff = f_name_cmp(flist->files[mid_up], f);
1464 if (protocol_version < 29
1465 && S_ISDIR(flist->files[mid_up]->mode)
1466 != S_ISDIR(f->mode))
1479 * Free up any resources a file_struct has allocated
1480 * and clear the file.
1482 void clear_file(struct file_struct *file, struct file_list *flist)
1484 if (flist->hlink_pool && file->link_u.idev)
1485 pool_free(flist->hlink_pool, 0, file->link_u.idev);
1486 memset(file, 0, file_struct_len);
1487 /* In an empty entry, dir.depth is an offset to the next non-empty
1488 * entry. Likewise for length in the opposite direction. We assume
1489 * that we're alone for now since flist_find() will adjust the counts
1490 * it runs into that aren't up-to-date. */
1491 file->length = file->dir.depth = 1;
1495 * allocate a new file list
1497 struct file_list *flist_new(int with_hlink, char *msg)
1499 struct file_list *flist;
1501 flist = new(struct file_list);
1505 memset(flist, 0, sizeof (struct file_list));
1507 if (!(flist->file_pool = pool_create(FILE_EXTENT, 0,
1508 out_of_memory, POOL_INTERN)))
1511 #ifdef SUPPORT_HARD_LINKS
1512 if (with_hlink && preserve_hard_links) {
1513 if (!(flist->hlink_pool = pool_create(HLINK_EXTENT,
1514 sizeof (struct idev), out_of_memory, POOL_INTERN)))
1523 * free up all elements in a flist
1525 void flist_free(struct file_list *flist)
1527 pool_destroy(flist->file_pool);
1528 pool_destroy(flist->hlink_pool);
1534 * This routine ensures we don't have any duplicate names in our file list.
1535 * duplicate names can cause corruption because of the pipelining
1537 static void clean_flist(struct file_list *flist, int strip_root, int no_dups)
1539 char fbuf[MAXPATHLEN];
1544 if (flist->count == 0) {
1549 qsort(flist->files, flist->count,
1550 sizeof flist->files[0], (int (*)())file_compare);
1552 for (i = no_dups? 0 : flist->count; i < flist->count; i++) {
1553 if (flist->files[i]->basename) {
1558 flist->low = prev_i;
1559 while (++i < flist->count) {
1561 struct file_struct *file = flist->files[i];
1563 if (!file->basename)
1565 if (f_name_cmp(file, flist->files[prev_i]) == 0)
1567 else if (protocol_version >= 29 && S_ISDIR(file->mode)) {
1568 int save_mode = file->mode;
1569 /* Make sure that this directory doesn't duplicate a
1570 * non-directory earlier in the list. */
1571 flist->high = prev_i;
1572 file->mode = S_IFREG;
1573 j = flist_find(flist, file);
1574 file->mode = save_mode;
1578 struct file_struct *fp = flist->files[j];
1580 /* If one is a dir and the other is not, we want to
1581 * keep the dir because it might have contents in the
1583 if (S_ISDIR(file->mode) != S_ISDIR(fp->mode)) {
1584 if (S_ISDIR(file->mode))
1590 if (verbose > 1 && !am_server) {
1592 "removing duplicate name %s from file list (%d)\n",
1593 f_name(file, fbuf), drop);
1595 /* Make sure we don't lose track of a user-specified
1597 flist->files[keep]->flags |= flist->files[drop]->flags
1598 & (FLAG_TOP_DIR|FLAG_DEL_HERE);
1600 clear_file(flist->files[drop], flist);
1603 if (flist->low == drop) {
1605 j < i && !flist->files[j]->basename;
1614 flist->high = no_dups ? prev_i : flist->count - 1;
1617 /* We need to strip off the leading slashes for relative
1618 * paths, but this must be done _after_ the sorting phase. */
1619 for (i = flist->low; i <= flist->high; i++) {
1620 struct file_struct *file = flist->files[i];
1624 while (*file->dirname == '/')
1626 if (!*file->dirname)
1627 file->dirname = NULL;
1631 if (prune_empty_dirs && no_dups) {
1632 int j, prev_depth = 0;
1634 prev_i = 0; /* It's OK that this isn't really true. */
1636 for (i = flist->low; i <= flist->high; i++) {
1637 struct file_struct *fp, *file = flist->files[i];
1639 /* This temporarily abuses the dir.depth value for a
1640 * directory that is in a chain that might get pruned.
1641 * We restore the old value if it gets a reprieve. */
1642 if (S_ISDIR(file->mode) && file->dir.depth) {
1643 /* Dump empty dirs when coming back down. */
1644 for (j = prev_depth; j >= file->dir.depth; j--) {
1645 fp = flist->files[prev_i];
1646 if (fp->dir.depth >= 0)
1648 prev_i = -fp->dir.depth-1;
1649 clear_file(fp, flist);
1651 prev_depth = file->dir.depth;
1652 if (is_excluded(f_name(file, fbuf), 1,
1654 /* Keep dirs through this dir. */
1655 for (j = prev_depth-1; ; j--) {
1656 fp = flist->files[prev_i];
1657 if (fp->dir.depth >= 0)
1659 prev_i = -fp->dir.depth-1;
1663 file->dir.depth = -prev_i-1;
1666 /* Keep dirs through this non-dir. */
1667 for (j = prev_depth; ; j--) {
1668 fp = flist->files[prev_i];
1669 if (fp->dir.depth >= 0)
1671 prev_i = -fp->dir.depth-1;
1676 /* Dump empty all remaining empty dirs. */
1678 struct file_struct *fp = flist->files[prev_i];
1679 if (fp->dir.depth >= 0)
1681 prev_i = -fp->dir.depth-1;
1682 clear_file(fp, flist);
1685 for (i = flist->low; i <= flist->high; i++) {
1686 if (flist->files[i]->basename)
1690 for (i = flist->high; i >= flist->low; i--) {
1691 if (flist->files[i]->basename)
1698 static void output_flist(struct file_list *flist)
1700 char uidbuf[16], gidbuf[16], depthbuf[16];
1701 struct file_struct *file;
1702 const char *who = who_am_i();
1705 for (i = 0; i < flist->count; i++) {
1706 file = flist->files[i];
1707 if ((am_root || am_sender) && preserve_uid)
1708 sprintf(uidbuf, " uid=%ld", (long)file->uid);
1711 if (preserve_gid && file->gid != GID_NONE)
1712 sprintf(gidbuf, " gid=%ld", (long)file->gid);
1716 sprintf(depthbuf, "%d", file->dir.depth);
1717 rprintf(FINFO, "[%s] i=%d %s %s%s%s%s mode=0%o len=%.0f%s%s flags=%x\n",
1718 who, i, am_sender ? NS(file->dir.root) : depthbuf,
1719 file->dirname ? file->dirname : "",
1720 file->dirname ? "/" : "", NS(file->basename),
1721 S_ISDIR(file->mode) ? "/" : "", (int)file->mode,
1722 (double)file->length, uidbuf, gidbuf, file->flags);
1726 enum fnc_state { s_DIR, s_SLASH, s_BASE, s_TRAILING };
1727 enum fnc_type { t_PATH, t_ITEM };
1729 /* Compare the names of two file_struct entities, similar to how strcmp()
1730 * would do if it were operating on the joined strings.
1732 * Some differences beginning with protocol_version 29: (1) directory names
1733 * are compared with an assumed trailing slash so that they compare in a
1734 * way that would cause them to sort immediately prior to any content they
1735 * may have; (2) a directory of any name compares after a non-directory of
1736 * any name at the same depth; (3) a directory with name "." compares prior
1737 * to anything else. These changes mean that a directory and a non-dir
1738 * with the same name will not compare as equal (protocol_version >= 29).
1740 * The dirname component can be an empty string, but the basename component
1741 * cannot (and never is in the current codebase). The basename component
1742 * may be NULL (for a removed item), in which case it is considered to be
1743 * after any existing item. */
1744 int f_name_cmp(struct file_struct *f1, struct file_struct *f2)
1747 const uchar *c1, *c2;
1748 enum fnc_state state1, state2;
1749 enum fnc_type type1, type2;
1750 enum fnc_type t_path = protocol_version >= 29 ? t_PATH : t_ITEM;
1752 if (!f1 || !f1->basename) {
1753 if (!f2 || !f2->basename)
1757 if (!f2 || !f2->basename)
1760 c1 = (uchar*)f1->dirname;
1761 c2 = (uchar*)f2->dirname;
1765 type1 = S_ISDIR(f1->mode) ? t_path : t_ITEM;
1766 c1 = (uchar*)f1->basename;
1767 if (type1 == t_PATH && *c1 == '.' && !c1[1]) {
1769 state1 = s_TRAILING;
1782 type2 = S_ISDIR(f2->mode) ? t_path : t_ITEM;
1783 c2 = (uchar*)f2->basename;
1784 if (type2 == t_PATH && *c2 == '.' && !c2[1]) {
1786 state2 = s_TRAILING;
1800 return type1 == t_PATH ? 1 : -1;
1803 if ((dif = (int)*c1++ - (int)*c2++) != 0)
1812 type1 = S_ISDIR(f1->mode) ? t_path : t_ITEM;
1813 c1 = (uchar*)f1->basename;
1814 if (type1 == t_PATH && *c1 == '.' && !c1[1]) {
1816 state1 = s_TRAILING;
1822 state1 = s_TRAILING;
1823 if (type1 == t_PATH) {
1832 if (*c2 && type1 != type2)
1833 return type1 == t_PATH ? 1 : -1;
1842 type2 = S_ISDIR(f2->mode) ? t_path : t_ITEM;
1843 c2 = (uchar*)f2->basename;
1844 if (type2 == t_PATH && *c2 == '.' && !c2[1]) {
1846 state2 = s_TRAILING;
1852 state2 = s_TRAILING;
1853 if (type2 == t_PATH) {
1865 return type1 == t_PATH ? 1 : -1;
1872 /* Return a copy of the full filename of a flist entry, using the indicated
1873 * buffer or one of 5 static buffers if fbuf is NULL. No size-checking is
1874 * done because we checked the size when creating the file_struct entry.
1876 char *f_name(struct file_struct *f, char *fbuf)
1878 if (!f || !f->basename)
1882 static char names[5][MAXPATHLEN];
1883 static unsigned int n;
1885 n = (n + 1) % (sizeof names / sizeof names[0]);
1891 int len = strlen(f->dirname);
1892 memcpy(fbuf, f->dirname, len);
1894 strcpy(fbuf + len + 1, f->basename);
1896 strcpy(fbuf, f->basename);
1901 /* Do a non-recursive scan of the named directory, possibly ignoring all
1902 * exclude rules except for the daemon's. If "dlen" is >=0, it is the length
1903 * of the dirname string, and also indicates that "dirname" is a MAXPATHLEN
1904 * buffer (the functions we call will append names onto the end, but the old
1905 * dir value will be restored on exit). */
1906 struct file_list *get_dirlist(char *dirname, int dlen,
1907 int ignore_filter_rules)
1909 struct file_list *dirlist;
1910 char dirbuf[MAXPATHLEN];
1911 int save_recurse = recurse;
1912 int save_xfer_dirs = xfer_dirs;
1915 dlen = strlcpy(dirbuf, dirname, MAXPATHLEN);
1916 if (dlen >= MAXPATHLEN)
1921 dirlist = flist_new(WITHOUT_HLINK, "get_dirlist");
1925 send_directory(ignore_filter_rules ? -2 : -1, dirlist, dirname, dlen);
1926 xfer_dirs = save_xfer_dirs;
1927 recurse = save_recurse;
1929 flist_count_offset += dirlist->count;
1931 clean_flist(dirlist, 0, 0);
1934 output_flist(dirlist);