Got rid of the sanitizing of a symlink's link string because it
[rsync/rsync.git] / flist.c
1 /*
2  * Generate and receive file lists.
3  *
4  * Copyright (C) 1996 Andrew Tridgell
5  * Copyright (C) 1996 Paul Mackerras
6  * Copyright (C) 2001, 2002 Martin Pool <mbp@samba.org>
7  * Copyright (C) 2002, 2003, 2004, 2005, 2006 Wayne Davison
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License as published by
11  * the Free Software Foundation; either version 2 of the License, or
12  * (at your option) any later version.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write to the Free Software Foundation, Inc.,
21  * 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA.
22  */
23
24 #include "rsync.h"
25
26 extern int verbose;
27 extern int list_only;
28 extern int am_root;
29 extern int am_server;
30 extern int am_daemon;
31 extern int am_sender;
32 extern int do_progress;
33 extern int always_checksum;
34 extern int module_id;
35 extern int ignore_errors;
36 extern int numeric_ids;
37 extern int recurse;
38 extern int xfer_dirs;
39 extern int filesfrom_fd;
40 extern int one_file_system;
41 extern int copy_dirlinks;
42 extern int keep_dirlinks;
43 extern int preserve_links;
44 extern int preserve_hard_links;
45 extern int preserve_devices;
46 extern int preserve_specials;
47 extern int preserve_uid;
48 extern int preserve_gid;
49 extern int relative_paths;
50 extern int implied_dirs;
51 extern int prune_empty_dirs;
52 extern int copy_links;
53 extern int copy_unsafe_links;
54 extern int protocol_version;
55 extern int sanitize_paths;
56 extern struct stats stats;
57 extern struct file_list *the_file_list;
58
59 extern char curr_dir[MAXPATHLEN];
60
61 extern struct chmod_mode_struct *chmod_modes;
62
63 extern struct filter_list_struct filter_list;
64 extern struct filter_list_struct server_filter_list;
65
66 int io_error;
67 int checksum_len;
68 dev_t filesystem_dev; /* used to implement -x */
69 unsigned int file_struct_len;
70
71 static char empty_sum[MD4_SUM_LENGTH];
72 static int flist_count_offset;
73
74 static void clean_flist(struct file_list *flist, int strip_root, int no_dups);
75 static void output_flist(struct file_list *flist);
76
77 void init_flist(void)
78 {
79         struct file_struct f;
80
81         /* Figure out how big the file_struct is without trailing padding */
82         file_struct_len = offsetof(struct file_struct, flags) + sizeof f.flags;
83         checksum_len = protocol_version < 21 ? 2 : MD4_SUM_LENGTH;
84 }
85
86 static int show_filelist_p(void)
87 {
88         return verbose && xfer_dirs && !am_server;
89 }
90
91 static void start_filelist_progress(char *kind)
92 {
93         rprintf(FINFO, "%s ... ", kind);
94         if (verbose > 1 || do_progress)
95                 rprintf(FINFO, "\n");
96         rflush(FINFO);
97 }
98
99 static void emit_filelist_progress(int count)
100 {
101         rprintf(FINFO, " %d files...\r", count);
102 }
103
104 static void maybe_emit_filelist_progress(int count)
105 {
106         if (do_progress && show_filelist_p() && (count % 100) == 0)
107                 emit_filelist_progress(count);
108 }
109
110 static void finish_filelist_progress(const struct file_list *flist)
111 {
112         if (do_progress) {
113                 /* This overwrites the progress line */
114                 rprintf(FINFO, "%d file%sto consider\n",
115                         flist->count, flist->count == 1 ? " " : "s ");
116         } else
117                 rprintf(FINFO, "done\n");
118 }
119
120 void show_flist_stats(void)
121 {
122         /* Nothing yet */
123 }
124
125 static void list_file_entry(struct file_struct *f)
126 {
127         char permbuf[PERMSTRING_SIZE];
128
129         if (!f->basename) {
130                 /* this can happen if duplicate names were removed */
131                 return;
132         }
133
134         permstring(permbuf, f->mode);
135
136 #ifdef SUPPORT_LINKS
137         if (preserve_links && S_ISLNK(f->mode)) {
138                 rprintf(FINFO, "%s %11.0f %s %s -> %s\n",
139                         permbuf,
140                         (double)f->length, timestring(f->modtime),
141                         f_name(f, NULL), f->u.link);
142         } else
143 #endif
144         {
145                 rprintf(FINFO, "%s %11.0f %s %s\n",
146                         permbuf,
147                         (double)f->length, timestring(f->modtime),
148                         f_name(f, NULL));
149         }
150 }
151
152 /**
153  * Stat either a symlink or its referent, depending on the settings of
154  * copy_links, copy_unsafe_links, etc.
155  *
156  * @retval -1 on error
157  *
158  * @retval 0 for success
159  *
160  * @post If @p path is a symlink, then @p linkbuf (of size @c
161  * MAXPATHLEN) contains the symlink target.
162  *
163  * @post @p buffer contains information about the link or the
164  * referrent as appropriate, if they exist.
165  **/
166 static int readlink_stat(const char *path, STRUCT_STAT *buffer, char *linkbuf)
167 {
168 #ifdef SUPPORT_LINKS
169         if (copy_links)
170                 return do_stat(path, buffer);
171         if (link_stat(path, buffer, copy_dirlinks) < 0)
172                 return -1;
173         if (S_ISLNK(buffer->st_mode)) {
174                 int l = readlink((char *)path, linkbuf, MAXPATHLEN - 1);
175                 if (l == -1)
176                         return -1;
177                 linkbuf[l] = 0;
178                 if (copy_unsafe_links && unsafe_symlink(linkbuf, path)) {
179                         if (verbose > 1) {
180                                 rprintf(FINFO,"copying unsafe symlink \"%s\" -> \"%s\"\n",
181                                         path, linkbuf);
182                         }
183                         return do_stat(path, buffer);
184                 }
185         }
186         return 0;
187 #else
188         return do_stat(path, buffer);
189 #endif
190 }
191
192 int link_stat(const char *path, STRUCT_STAT *buffer, int follow_dirlinks)
193 {
194 #ifdef SUPPORT_LINKS
195         if (copy_links)
196                 return do_stat(path, buffer);
197         if (do_lstat(path, buffer) < 0)
198                 return -1;
199         if (follow_dirlinks && S_ISLNK(buffer->st_mode)) {
200                 STRUCT_STAT st;
201                 if (do_stat(path, &st) == 0 && S_ISDIR(st.st_mode))
202                         *buffer = st;
203         }
204         return 0;
205 #else
206         return do_stat(path, buffer);
207 #endif
208 }
209
210 /* This function is used to check if a file should be included/excluded
211  * from the list of files based on its name and type etc.  The value of
212  * filter_level is set to either SERVER_FILTERS or ALL_FILTERS. */
213 static int is_excluded(char *fname, int is_dir, int filter_level)
214 {
215 #if 0 /* This currently never happens, so avoid a useless compare. */
216         if (filter_level == NO_FILTERS)
217                 return 0;
218 #endif
219         if (fname) {
220                 /* never exclude '.', even if somebody does --exclude '*' */
221                 if (fname[0] == '.' && !fname[1])
222                         return 0;
223                 /* Handle the -R version of the '.' dir. */
224                 if (fname[0] == '/') {
225                         int len = strlen(fname);
226                         if (fname[len-1] == '.' && fname[len-2] == '/')
227                                 return 0;
228                 }
229         }
230         if (server_filter_list.head
231             && check_filter(&server_filter_list, fname, is_dir) < 0)
232                 return 1;
233         if (filter_level != ALL_FILTERS)
234                 return 0;
235         if (filter_list.head
236             && check_filter(&filter_list, fname, is_dir) < 0)
237                 return 1;
238         return 0;
239 }
240
241 static int to_wire_mode(mode_t mode)
242 {
243 #ifdef SUPPORT_LINKS
244         if (S_ISLNK(mode) && (_S_IFLNK != 0120000))
245                 return (mode & ~(_S_IFMT)) | 0120000;
246 #endif
247         return (int)mode;
248 }
249
250 static mode_t from_wire_mode(int mode)
251 {
252         if ((mode & (_S_IFMT)) == 0120000 && (_S_IFLNK != 0120000))
253                 return (mode & ~(_S_IFMT)) | _S_IFLNK;
254         return (mode_t)mode;
255 }
256
257 static void send_directory(int f, struct file_list *flist,
258                            char *fbuf, int len);
259
260 static char *flist_dir;
261 static int flist_dir_len;
262
263
264 /**
265  * Make sure @p flist is big enough to hold at least @p flist->count
266  * entries.
267  **/
268 void flist_expand(struct file_list *flist)
269 {
270         struct file_struct **new_ptr;
271
272         if (flist->count < flist->malloced)
273                 return;
274
275         if (flist->malloced < FLIST_START)
276                 flist->malloced = FLIST_START;
277         else if (flist->malloced >= FLIST_LINEAR)
278                 flist->malloced += FLIST_LINEAR;
279         else
280                 flist->malloced *= 2;
281
282         /*
283          * In case count jumped or we are starting the list
284          * with a known size just set it.
285          */
286         if (flist->malloced < flist->count)
287                 flist->malloced = flist->count;
288
289         new_ptr = realloc_array(flist->files, struct file_struct *,
290                                 flist->malloced);
291
292         if (verbose >= 2 && flist->malloced != FLIST_START) {
293                 rprintf(FINFO, "[%s] expand file_list to %.0f bytes, did%s move\n",
294                     who_am_i(),
295                     (double)sizeof flist->files[0] * flist->malloced,
296                     (new_ptr == flist->files) ? " not" : "");
297         }
298
299         flist->files = new_ptr;
300
301         if (!flist->files)
302                 out_of_memory("flist_expand");
303 }
304
305 static void send_file_entry(struct file_struct *file, int f)
306 {
307         unsigned short flags;
308         static time_t modtime;
309         static mode_t mode;
310         static int64 dev;
311         static dev_t rdev;
312         static uint32 rdev_major;
313         static uid_t uid;
314         static gid_t gid;
315         static char lastname[MAXPATHLEN];
316         char fname[MAXPATHLEN];
317         int l1, l2;
318
319         if (f < 0)
320                 return;
321
322         if (!file) {
323                 write_byte(f, 0);
324                 modtime = 0, mode = 0;
325                 dev = 0, rdev = makedev(0, 0);
326                 rdev_major = 0;
327                 uid = 0, gid = 0;
328                 *lastname = '\0';
329                 return;
330         }
331
332         f_name(file, fname);
333
334         flags = file->flags & XMIT_TOP_DIR;
335
336         if (file->mode == mode)
337                 flags |= XMIT_SAME_MODE;
338         else
339                 mode = file->mode;
340         if ((preserve_devices && IS_DEVICE(mode))
341          || (preserve_specials && IS_SPECIAL(mode))) {
342                 if (protocol_version < 28) {
343                         if (file->u.rdev == rdev)
344                                 flags |= XMIT_SAME_RDEV_pre28;
345                         else
346                                 rdev = file->u.rdev;
347                 } else {
348                         rdev = file->u.rdev;
349                         if ((uint32)major(rdev) == rdev_major)
350                                 flags |= XMIT_SAME_RDEV_MAJOR;
351                         else
352                                 rdev_major = major(rdev);
353                         if ((uint32)minor(rdev) <= 0xFFu)
354                                 flags |= XMIT_RDEV_MINOR_IS_SMALL;
355                 }
356         } else if (protocol_version < 28)
357                 rdev = makedev(0, 0);
358         if (file->uid == uid)
359                 flags |= XMIT_SAME_UID;
360         else
361                 uid = file->uid;
362         if (file->gid == gid)
363                 flags |= XMIT_SAME_GID;
364         else
365                 gid = file->gid;
366         if (file->modtime == modtime)
367                 flags |= XMIT_SAME_TIME;
368         else
369                 modtime = file->modtime;
370
371 #ifdef SUPPORT_HARD_LINKS
372         if (file->link_u.idev) {
373                 if (file->F_DEV == dev) {
374                         if (protocol_version >= 28)
375                                 flags |= XMIT_SAME_DEV;
376                 } else
377                         dev = file->F_DEV;
378                 flags |= XMIT_HAS_IDEV_DATA;
379         }
380 #endif
381
382         for (l1 = 0;
383             lastname[l1] && (fname[l1] == lastname[l1]) && (l1 < 255);
384             l1++) {}
385         l2 = strlen(fname+l1);
386
387         if (l1 > 0)
388                 flags |= XMIT_SAME_NAME;
389         if (l2 > 255)
390                 flags |= XMIT_LONG_NAME;
391
392         /* We must make sure we don't send a zero flag byte or the
393          * other end will terminate the flist transfer.  Note that
394          * the use of XMIT_TOP_DIR on a non-dir has no meaning, so
395          * it's harmless way to add a bit to the first flag byte. */
396         if (protocol_version >= 28) {
397                 if (!flags && !S_ISDIR(mode))
398                         flags |= XMIT_TOP_DIR;
399                 if ((flags & 0xFF00) || !flags) {
400                         flags |= XMIT_EXTENDED_FLAGS;
401                         write_byte(f, flags);
402                         write_byte(f, flags >> 8);
403                 } else
404                         write_byte(f, flags);
405         } else {
406                 if (!(flags & 0xFF))
407                         flags |= S_ISDIR(mode) ? XMIT_LONG_NAME : XMIT_TOP_DIR;
408                 write_byte(f, flags);
409         }
410         if (flags & XMIT_SAME_NAME)
411                 write_byte(f, l1);
412         if (flags & XMIT_LONG_NAME)
413                 write_int(f, l2);
414         else
415                 write_byte(f, l2);
416         write_buf(f, fname + l1, l2);
417
418         write_longint(f, file->length);
419         if (!(flags & XMIT_SAME_TIME))
420                 write_int(f, modtime);
421         if (!(flags & XMIT_SAME_MODE))
422                 write_int(f, to_wire_mode(mode));
423         if (preserve_uid && !(flags & XMIT_SAME_UID)) {
424                 if (!numeric_ids)
425                         add_uid(uid);
426                 write_int(f, uid);
427         }
428         if (preserve_gid && !(flags & XMIT_SAME_GID)) {
429                 if (!numeric_ids)
430                         add_gid(gid);
431                 write_int(f, gid);
432         }
433         if ((preserve_devices && IS_DEVICE(mode))
434          || (preserve_specials && IS_SPECIAL(mode))) {
435                 if (protocol_version < 28) {
436                         if (!(flags & XMIT_SAME_RDEV_pre28))
437                                 write_int(f, (int)rdev);
438                 } else {
439                         if (!(flags & XMIT_SAME_RDEV_MAJOR))
440                                 write_int(f, major(rdev));
441                         if (flags & XMIT_RDEV_MINOR_IS_SMALL)
442                                 write_byte(f, minor(rdev));
443                         else
444                                 write_int(f, minor(rdev));
445                 }
446         }
447
448 #ifdef SUPPORT_LINKS
449         if (preserve_links && S_ISLNK(mode)) {
450                 int len = strlen(file->u.link);
451                 write_int(f, len);
452                 write_buf(f, file->u.link, len);
453         }
454 #endif
455
456 #ifdef SUPPORT_HARD_LINKS
457         if (flags & XMIT_HAS_IDEV_DATA) {
458                 if (protocol_version < 26) {
459                         /* 32-bit dev_t and ino_t */
460                         write_int(f, dev);
461                         write_int(f, file->F_INODE);
462                 } else {
463                         /* 64-bit dev_t and ino_t */
464                         if (!(flags & XMIT_SAME_DEV))
465                                 write_longint(f, dev);
466                         write_longint(f, file->F_INODE);
467                 }
468         }
469 #endif
470
471         if (always_checksum && (S_ISREG(mode) || protocol_version < 28)) {
472                 char *sum;
473                 if (S_ISREG(mode))
474                         sum = file->u.sum;
475                 else {
476                         /* Prior to 28, we sent a useless set of nulls. */
477                         sum = empty_sum;
478                 }
479                 write_buf(f, sum, checksum_len);
480         }
481
482         strlcpy(lastname, fname, MAXPATHLEN);
483 }
484
485 static struct file_struct *receive_file_entry(struct file_list *flist,
486                                               unsigned short flags, int f)
487 {
488         static time_t modtime;
489         static mode_t mode;
490         static int64 dev;
491         static dev_t rdev;
492         static uint32 rdev_major;
493         static uid_t uid;
494         static gid_t gid;
495         static char lastname[MAXPATHLEN], *lastdir;
496         static int lastdir_depth, lastdir_len = -1;
497         static unsigned int del_hier_name_len = 0;
498         static int in_del_hier = 0;
499         char thisname[MAXPATHLEN];
500         unsigned int l1 = 0, l2 = 0;
501         int alloc_len, basename_len, dirname_len, linkname_len, sum_len;
502         OFF_T file_length;
503         char *basename, *dirname, *bp;
504         struct file_struct *file;
505
506         if (!flist) {
507                 modtime = 0, mode = 0;
508                 dev = 0, rdev = makedev(0, 0);
509                 rdev_major = 0;
510                 uid = 0, gid = 0;
511                 *lastname = '\0';
512                 lastdir_len = -1;
513                 in_del_hier = 0;
514                 return NULL;
515         }
516
517         if (flags & XMIT_SAME_NAME)
518                 l1 = read_byte(f);
519
520         if (flags & XMIT_LONG_NAME)
521                 l2 = read_int(f);
522         else
523                 l2 = read_byte(f);
524
525         if (l2 >= MAXPATHLEN - l1) {
526                 rprintf(FERROR,
527                         "overflow: flags=0x%x l1=%d l2=%d lastname=%s\n",
528                         flags, l1, l2, lastname);
529                 overflow_exit("receive_file_entry");
530         }
531
532         strlcpy(thisname, lastname, l1 + 1);
533         read_sbuf(f, &thisname[l1], l2);
534         thisname[l1 + l2] = 0;
535
536         strlcpy(lastname, thisname, MAXPATHLEN);
537
538         clean_fname(thisname, 0);
539
540         if (sanitize_paths)
541                 sanitize_path(thisname, thisname, "", 0);
542
543         if ((basename = strrchr(thisname, '/')) != NULL) {
544                 dirname_len = ++basename - thisname; /* counts future '\0' */
545                 if (lastdir_len == dirname_len - 1
546                     && strncmp(thisname, lastdir, lastdir_len) == 0) {
547                         dirname = lastdir;
548                         dirname_len = 0; /* indicates no copy is needed */
549                 } else
550                         dirname = thisname;
551         } else {
552                 basename = thisname;
553                 dirname = NULL;
554                 dirname_len = 0;
555         }
556         basename_len = strlen(basename) + 1; /* count the '\0' */
557
558         file_length = read_longint(f);
559         if (!(flags & XMIT_SAME_TIME))
560                 modtime = (time_t)read_int(f);
561         if (!(flags & XMIT_SAME_MODE))
562                 mode = from_wire_mode(read_int(f));
563
564         if (chmod_modes && !S_ISLNK(mode))
565                 mode = tweak_mode(mode, chmod_modes);
566
567         if (preserve_uid && !(flags & XMIT_SAME_UID))
568                 uid = (uid_t)read_int(f);
569         if (preserve_gid && !(flags & XMIT_SAME_GID))
570                 gid = (gid_t)read_int(f);
571
572         if ((preserve_devices && IS_DEVICE(mode))
573          || (preserve_specials && IS_SPECIAL(mode))) {
574                 if (protocol_version < 28) {
575                         if (!(flags & XMIT_SAME_RDEV_pre28))
576                                 rdev = (dev_t)read_int(f);
577                 } else {
578                         uint32 rdev_minor;
579                         if (!(flags & XMIT_SAME_RDEV_MAJOR))
580                                 rdev_major = read_int(f);
581                         if (flags & XMIT_RDEV_MINOR_IS_SMALL)
582                                 rdev_minor = read_byte(f);
583                         else
584                                 rdev_minor = read_int(f);
585                         rdev = makedev(rdev_major, rdev_minor);
586                 }
587         } else if (protocol_version < 28)
588                 rdev = makedev(0, 0);
589
590 #ifdef SUPPORT_LINKS
591         if (preserve_links && S_ISLNK(mode)) {
592                 linkname_len = read_int(f) + 1; /* count the '\0' */
593                 if (linkname_len <= 0 || linkname_len > MAXPATHLEN) {
594                         rprintf(FERROR, "overflow: linkname_len=%d\n",
595                                 linkname_len - 1);
596                         overflow_exit("receive_file_entry");
597                 }
598         }
599         else
600 #endif
601                 linkname_len = 0;
602
603         sum_len = always_checksum && S_ISREG(mode) ? MD4_SUM_LENGTH : 0;
604
605         alloc_len = file_struct_len + dirname_len + basename_len
606                   + linkname_len + sum_len;
607         bp = pool_alloc(flist->file_pool, alloc_len, "receive_file_entry");
608
609         file = (struct file_struct *)bp;
610         memset(bp, 0, file_struct_len);
611         bp += file_struct_len;
612
613         file->modtime = modtime;
614         file->length = file_length;
615         file->mode = mode;
616         file->uid = uid;
617         file->gid = gid;
618
619         if (dirname_len) {
620                 file->dirname = lastdir = bp;
621                 lastdir_len = dirname_len - 1;
622                 memcpy(bp, dirname, dirname_len - 1);
623                 bp += dirname_len;
624                 bp[-1] = '\0';
625                 lastdir_depth = count_dir_elements(lastdir);
626                 file->dir.depth = lastdir_depth + 1;
627         } else if (dirname) {
628                 file->dirname = dirname; /* we're reusing lastname */
629                 file->dir.depth = lastdir_depth + 1;
630         } else
631                 file->dir.depth = 1;
632
633         if (S_ISDIR(mode)) {
634                 if (basename_len == 1+1 && *basename == '.') /* +1 for '\0' */
635                         file->dir.depth--;
636                 if (flags & XMIT_TOP_DIR) {
637                         in_del_hier = recurse;
638                         del_hier_name_len = file->dir.depth == 0 ? 0 : l1 + l2;
639                         if (relative_paths && del_hier_name_len > 2
640                             && lastname[del_hier_name_len-1] == '.'
641                             && lastname[del_hier_name_len-2] == '/')
642                                 del_hier_name_len -= 2;
643                         file->flags |= FLAG_TOP_DIR | FLAG_DEL_HERE;
644                 } else if (in_del_hier) {
645                         if (!relative_paths || !del_hier_name_len
646                          || (l1 >= del_hier_name_len
647                           && lastname[del_hier_name_len] == '/'))
648                                 file->flags |= FLAG_DEL_HERE;
649                         else
650                                 in_del_hier = 0;
651                 }
652         }
653
654         file->basename = bp;
655         memcpy(bp, basename, basename_len);
656         bp += basename_len;
657
658         if ((preserve_devices && IS_DEVICE(mode))
659          || (preserve_specials && IS_SPECIAL(mode)))
660                 file->u.rdev = rdev;
661
662 #ifdef SUPPORT_LINKS
663         if (linkname_len) {
664                 file->u.link = bp;
665                 read_sbuf(f, bp, linkname_len - 1);
666                 bp += linkname_len;
667         }
668 #endif
669
670 #ifdef SUPPORT_HARD_LINKS
671         if (preserve_hard_links && protocol_version < 28 && S_ISREG(mode))
672                 flags |= XMIT_HAS_IDEV_DATA;
673         if (flags & XMIT_HAS_IDEV_DATA) {
674                 int64 inode;
675                 if (protocol_version < 26) {
676                         dev = read_int(f);
677                         inode = read_int(f);
678                 } else {
679                         if (!(flags & XMIT_SAME_DEV))
680                                 dev = read_longint(f);
681                         inode = read_longint(f);
682                 }
683                 if (flist->hlink_pool) {
684                         file->link_u.idev = pool_talloc(flist->hlink_pool,
685                             struct idev, 1, "inode_table");
686                         file->F_INODE = inode;
687                         file->F_DEV = dev;
688                 }
689         }
690 #endif
691
692         if (always_checksum && (sum_len || protocol_version < 28)) {
693                 char *sum;
694                 if (sum_len) {
695                         file->u.sum = sum = bp;
696                         /*bp += sum_len;*/
697                 } else {
698                         /* Prior to 28, we get a useless set of nulls. */
699                         sum = empty_sum;
700                 }
701                 read_buf(f, sum, checksum_len);
702         }
703
704         return file;
705 }
706
707 /**
708  * Create a file_struct for a named file by reading its stat()
709  * information and performing extensive checks against global
710  * options.
711  *
712  * @return the new file, or NULL if there was an error or this file
713  * should be excluded.
714  *
715  * @todo There is a small optimization opportunity here to avoid
716  * stat()ing the file in some circumstances, which has a certain cost.
717  * We are called immediately after doing readdir(), and so we may
718  * already know the d_type of the file.  We could for example avoid
719  * statting directories if we're not recursing, but this is not a very
720  * important case.  Some systems may not have d_type.
721  **/
722 struct file_struct *make_file(char *fname, struct file_list *flist,
723                               STRUCT_STAT *stp, unsigned short flags,
724                               int filter_level)
725 {
726         static char *lastdir;
727         static int lastdir_len = -1;
728         struct file_struct *file;
729         STRUCT_STAT st;
730         char sum[SUM_LENGTH];
731         char thisname[MAXPATHLEN];
732         char linkname[MAXPATHLEN];
733         int alloc_len, basename_len, dirname_len, linkname_len, sum_len;
734         char *basename, *dirname, *bp;
735
736         if (!flist || !flist->count)    /* Ignore lastdir when invalid. */
737                 lastdir_len = -1;
738
739         if (strlcpy(thisname, fname, sizeof thisname)
740             >= sizeof thisname - flist_dir_len) {
741                 rprintf(FINFO, "skipping overly long name: %s\n", fname);
742                 return NULL;
743         }
744         clean_fname(thisname, 0);
745         if (sanitize_paths)
746                 sanitize_path(thisname, thisname, "", 0);
747
748         memset(sum, 0, SUM_LENGTH);
749
750         if (stp && S_ISDIR(stp->st_mode))
751                 st = *stp; /* Needed for "symlink/." with --relative. */
752         else if (readlink_stat(thisname, &st, linkname) != 0) {
753                 int save_errno = errno;
754                 /* See if file is excluded before reporting an error. */
755                 if (filter_level != NO_FILTERS
756                     && is_excluded(thisname, 0, filter_level))
757                         return NULL;
758                 if (save_errno == ENOENT) {
759 #ifdef SUPPORT_LINKS
760                         /* Avoid "vanished" error if symlink points nowhere. */
761                         if (copy_links && do_lstat(thisname, &st) == 0
762                             && S_ISLNK(st.st_mode)) {
763                                 io_error |= IOERR_GENERAL;
764                                 rprintf(FERROR, "symlink has no referent: %s\n",
765                                         full_fname(thisname));
766                         } else
767 #endif
768                         {
769                                 enum logcode c = am_daemon && protocol_version < 28
770                                     ? FERROR : FINFO;
771                                 io_error |= IOERR_VANISHED;
772                                 rprintf(c, "file has vanished: %s\n",
773                                         full_fname(thisname));
774                         }
775                 } else {
776                         io_error |= IOERR_GENERAL;
777                         rsyserr(FERROR, save_errno, "readlink %s failed",
778                                 full_fname(thisname));
779                 }
780                 return NULL;
781         }
782
783         /* backup.c calls us with filter_level set to NO_FILTERS. */
784         if (filter_level == NO_FILTERS)
785                 goto skip_filters;
786
787         if (S_ISDIR(st.st_mode) && !xfer_dirs) {
788                 rprintf(FINFO, "skipping directory %s\n", thisname);
789                 return NULL;
790         }
791
792         /* We only care about directories because we need to avoid recursing
793          * into a mount-point directory, not to avoid copying a symlinked
794          * file if -L (or similar) was specified. */
795         if (one_file_system && st.st_dev != filesystem_dev
796          && S_ISDIR(st.st_mode)) {
797                 if (one_file_system > 1) {
798                         if (verbose > 2) {
799                                 rprintf(FINFO, "skipping mount-point dir %s\n",
800                                         thisname);
801                         }
802                         return NULL;
803                 }
804                 flags |= FLAG_MOUNT_POINT;
805         }
806
807         if (is_excluded(thisname, S_ISDIR(st.st_mode) != 0, filter_level))
808                 return NULL;
809
810         if (lp_ignore_nonreadable(module_id)) {
811 #ifdef SUPPORT_LINKS
812                 if (!S_ISLNK(st.st_mode))
813 #endif
814                         if (access(thisname, R_OK) != 0)
815                                 return NULL;
816         }
817
818   skip_filters:
819
820         if (verbose > 2) {
821                 rprintf(FINFO, "[%s] make_file(%s,*,%d)\n",
822                         who_am_i(), thisname, filter_level);
823         }
824
825         if ((basename = strrchr(thisname, '/')) != NULL) {
826                 dirname_len = ++basename - thisname; /* counts future '\0' */
827                 if (lastdir_len == dirname_len - 1
828                     && strncmp(thisname, lastdir, lastdir_len) == 0) {
829                         dirname = lastdir;
830                         dirname_len = 0; /* indicates no copy is needed */
831                 } else
832                         dirname = thisname;
833         } else {
834                 basename = thisname;
835                 dirname = NULL;
836                 dirname_len = 0;
837         }
838         basename_len = strlen(basename) + 1; /* count the '\0' */
839
840 #ifdef SUPPORT_LINKS
841         linkname_len = S_ISLNK(st.st_mode) ? strlen(linkname) + 1 : 0;
842 #else
843         linkname_len = 0;
844 #endif
845
846         sum_len = always_checksum && am_sender && S_ISREG(st.st_mode)
847                 ? MD4_SUM_LENGTH : 0;
848
849         alloc_len = file_struct_len + dirname_len + basename_len
850             + linkname_len + sum_len;
851         if (flist)
852                 bp = pool_alloc(flist->file_pool, alloc_len, "make_file");
853         else {
854                 if (!(bp = new_array(char, alloc_len)))
855                         out_of_memory("make_file");
856         }
857
858         file = (struct file_struct *)bp;
859         memset(bp, 0, file_struct_len);
860         bp += file_struct_len;
861
862         file->flags = flags;
863         file->modtime = st.st_mtime;
864         file->length = st.st_size;
865         file->mode = st.st_mode;
866         file->uid = st.st_uid;
867         file->gid = st.st_gid;
868
869 #ifdef SUPPORT_HARD_LINKS
870         if (flist && flist->hlink_pool) {
871                 if (protocol_version < 28) {
872                         if (S_ISREG(st.st_mode))
873                                 file->link_u.idev = pool_talloc(
874                                     flist->hlink_pool, struct idev, 1,
875                                     "inode_table");
876                 } else {
877                         if (!S_ISDIR(st.st_mode) && st.st_nlink > 1)
878                                 file->link_u.idev = pool_talloc(
879                                     flist->hlink_pool, struct idev, 1,
880                                     "inode_table");
881                 }
882         }
883         if (file->link_u.idev) {
884                 file->F_DEV = st.st_dev;
885                 file->F_INODE = st.st_ino;
886         }
887 #endif
888
889         if (dirname_len) {
890                 file->dirname = lastdir = bp;
891                 lastdir_len = dirname_len - 1;
892                 memcpy(bp, dirname, dirname_len - 1);
893                 bp += dirname_len;
894                 bp[-1] = '\0';
895         } else if (dirname)
896                 file->dirname = dirname;
897
898         file->basename = bp;
899         memcpy(bp, basename, basename_len);
900         bp += basename_len;
901
902 #ifdef HAVE_STRUCT_STAT_ST_RDEV
903         if ((preserve_devices && IS_DEVICE(st.st_mode))
904          || (preserve_specials && IS_SPECIAL(st.st_mode)))
905                 file->u.rdev = st.st_rdev;
906 #endif
907
908 #ifdef SUPPORT_LINKS
909         if (linkname_len) {
910                 file->u.link = bp;
911                 memcpy(bp, linkname, linkname_len);
912                 bp += linkname_len;
913         }
914 #endif
915
916         if (sum_len) {
917                 file->u.sum = bp;
918                 file_checksum(thisname, bp, st.st_size);
919                 /*bp += sum_len;*/
920         }
921
922         file->dir.root = flist_dir;
923
924         /* This code is only used by the receiver when it is building
925          * a list of files for a delete pass. */
926         if (keep_dirlinks && linkname_len && flist) {
927                 STRUCT_STAT st2;
928                 int save_mode = file->mode;
929                 file->mode = S_IFDIR; /* Find a directory with our name. */
930                 if (flist_find(the_file_list, file) >= 0
931                     && do_stat(thisname, &st2) == 0 && S_ISDIR(st2.st_mode)) {
932                         file->modtime = st2.st_mtime;
933                         file->length = st2.st_size;
934                         file->mode = st2.st_mode;
935                         file->uid = st2.st_uid;
936                         file->gid = st2.st_gid;
937                         file->u.link = NULL;
938                 } else
939                         file->mode = save_mode;
940         }
941
942         if (S_ISREG(st.st_mode) || S_ISLNK(st.st_mode))
943                 stats.total_size += st.st_size;
944
945         return file;
946 }
947
948 static struct file_struct *send_file_name(int f, struct file_list *flist,
949                                           char *fname, STRUCT_STAT *stp,
950                                           unsigned short flags)
951 {
952         struct file_struct *file;
953
954         file = make_file(fname, flist, stp, flags,
955                          f == -2 ? SERVER_FILTERS : ALL_FILTERS);
956         if (!file)
957                 return NULL;
958
959         if (chmod_modes && !S_ISLNK(file->mode))
960                 file->mode = tweak_mode(file->mode, chmod_modes);
961
962         maybe_emit_filelist_progress(flist->count + flist_count_offset);
963
964         flist_expand(flist);
965
966         if (file->basename[0]) {
967                 flist->files[flist->count++] = file;
968                 send_file_entry(file, f);
969         }
970         return file;
971 }
972
973 static void send_if_directory(int f, struct file_list *flist,
974                               struct file_struct *file,
975                               char *fbuf, unsigned int ol)
976 {
977         char is_dot_dir = fbuf[ol-1] == '.' && (ol == 1 || fbuf[ol-2] == '/');
978
979         if (S_ISDIR(file->mode)
980             && !(file->flags & FLAG_MOUNT_POINT) && f_name(file, fbuf)) {
981                 void *save_filters;
982                 unsigned int len = strlen(fbuf);
983                 if (len > 1 && fbuf[len-1] == '/')
984                         fbuf[--len] = '\0';
985                 if (len >= MAXPATHLEN - 1) {
986                         io_error |= IOERR_GENERAL;
987                         rprintf(FERROR, "skipping long-named directory: %s\n",
988                                 full_fname(fbuf));
989                         return;
990                 }
991                 save_filters = push_local_filters(fbuf, len);
992                 send_directory(f, flist, fbuf, len);
993                 pop_local_filters(save_filters);
994                 fbuf[ol] = '\0';
995                 if (is_dot_dir)
996                         fbuf[ol-1] = '.';
997         }
998 }
999
1000 /* This function is normally called by the sender, but the receiving side also
1001  * calls it from get_dirlist() with f set to -1 so that we just construct the
1002  * file list in memory without sending it over the wire.  Also, get_dirlist()
1003  * might call this with f set to -2, which also indicates that local filter
1004  * rules should be ignored. */
1005 static void send_directory(int f, struct file_list *flist,
1006                            char *fbuf, int len)
1007 {
1008         struct dirent *di;
1009         unsigned remainder;
1010         char *p;
1011         DIR *d;
1012         int start = flist->count;
1013
1014         if (!(d = opendir(fbuf))) {
1015                 io_error |= IOERR_GENERAL;
1016                 rsyserr(FERROR, errno, "opendir %s failed", full_fname(fbuf));
1017                 return;
1018         }
1019
1020         p = fbuf + len;
1021         if (len != 1 || *fbuf != '/')
1022                 *p++ = '/';
1023         *p = '\0';
1024         remainder = MAXPATHLEN - (p - fbuf);
1025
1026         for (errno = 0, di = readdir(d); di; errno = 0, di = readdir(d)) {
1027                 char *dname = d_name(di);
1028                 if (dname[0] == '.' && (dname[1] == '\0'
1029                     || (dname[1] == '.' && dname[2] == '\0')))
1030                         continue;
1031                 if (strlcpy(p, dname, remainder) >= remainder) {
1032                         io_error |= IOERR_GENERAL;
1033                         rprintf(FINFO,
1034                                 "cannot send long-named file %s\n",
1035                                 full_fname(fbuf));
1036                         continue;
1037                 }
1038
1039                 send_file_name(f, flist, fbuf, NULL, 0);
1040         }
1041
1042         fbuf[len] = '\0';
1043
1044         if (errno) {
1045                 io_error |= IOERR_GENERAL;
1046                 rsyserr(FERROR, errno, "readdir(%s)", full_fname(fbuf));
1047         }
1048
1049         closedir(d);
1050
1051         if (recurse) {
1052                 int i, end = flist->count - 1;
1053                 for (i = start; i <= end; i++)
1054                         send_if_directory(f, flist, flist->files[i], fbuf, len);
1055         }
1056 }
1057
1058 struct file_list *send_file_list(int f, int argc, char *argv[])
1059 {
1060         int len;
1061         STRUCT_STAT st;
1062         char *p, *dir, olddir[sizeof curr_dir];
1063         char lastpath[MAXPATHLEN] = "";
1064         struct file_list *flist;
1065         struct timeval start_tv, end_tv;
1066         int64 start_write;
1067         int use_ff_fd = 0;
1068
1069         if (show_filelist_p())
1070                 start_filelist_progress("building file list");
1071
1072         start_write = stats.total_written;
1073         gettimeofday(&start_tv, NULL);
1074
1075         flist = flist_new(WITH_HLINK, "send_file_list");
1076
1077         io_start_buffering_out();
1078         if (filesfrom_fd >= 0) {
1079                 if (argv[0] && !push_dir(argv[0])) {
1080                         rsyserr(FERROR, errno, "push_dir %s failed",
1081                                 full_fname(argv[0]));
1082                         exit_cleanup(RERR_FILESELECT);
1083                 }
1084                 use_ff_fd = 1;
1085         }
1086
1087         while (1) {
1088                 char fbuf[MAXPATHLEN];
1089                 char *fn;
1090                 int is_dot_dir;
1091
1092                 if (use_ff_fd) {
1093                         if (read_filesfrom_line(filesfrom_fd, fbuf) == 0)
1094                                 break;
1095                         sanitize_path(fbuf, fbuf, "", 0);
1096                 } else {
1097                         if (argc-- == 0)
1098                                 break;
1099                         strlcpy(fbuf, *argv++, MAXPATHLEN);
1100                         if (sanitize_paths)
1101                                 sanitize_path(fbuf, fbuf, "", 0);
1102                 }
1103
1104                 len = strlen(fbuf);
1105                 if (relative_paths) {
1106                         /* We clean up fbuf below. */
1107                         is_dot_dir = 0;
1108                 } else if (!len || fbuf[len - 1] == '/') {
1109                         if (len == 2 && fbuf[0] == '.') {
1110                                 /* Turn "./" into just "." rather than "./." */
1111                                 fbuf[1] = '\0';
1112                         } else {
1113                                 if (len + 1 >= MAXPATHLEN)
1114                                         overflow_exit("send_file_list");
1115                                 fbuf[len++] = '.';
1116                                 fbuf[len] = '\0';
1117                         }
1118                         is_dot_dir = 1;
1119                 } else if (len > 1 && fbuf[len-1] == '.' && fbuf[len-2] == '.'
1120                     && (len == 2 || fbuf[len-3] == '/')) {
1121                         if (len + 2 >= MAXPATHLEN)
1122                                 overflow_exit("send_file_list");
1123                         fbuf[len++] = '/';
1124                         fbuf[len++] = '.';
1125                         fbuf[len] = '\0';
1126                         is_dot_dir = 1;
1127                 } else {
1128                         is_dot_dir = fbuf[len-1] == '.'
1129                                    && (len == 1 || fbuf[len-2] == '/');
1130                 }
1131
1132                 if (link_stat(fbuf, &st, copy_dirlinks) != 0) {
1133                         io_error |= IOERR_GENERAL;
1134                         rsyserr(FERROR, errno, "link_stat %s failed",
1135                                 full_fname(fbuf));
1136                         continue;
1137                 }
1138
1139                 if (S_ISDIR(st.st_mode) && !xfer_dirs) {
1140                         rprintf(FINFO, "skipping directory %s\n", fbuf);
1141                         continue;
1142                 }
1143
1144                 dir = NULL;
1145                 olddir[0] = '\0';
1146
1147                 if (!relative_paths) {
1148                         p = strrchr(fbuf, '/');
1149                         if (p) {
1150                                 *p = '\0';
1151                                 if (p == fbuf)
1152                                         dir = "/";
1153                                 else
1154                                         dir = fbuf;
1155                                 len -= p - fbuf + 1;
1156                                 fn = p + 1;
1157                         } else
1158                                 fn = fbuf;
1159                 } else {
1160                         if ((p = strstr(fbuf, "/./")) != NULL) {
1161                                 *p = '\0';
1162                                 if (p == fbuf)
1163                                         dir = "/";
1164                                 else
1165                                         dir = fbuf;
1166                                 len -= p - fbuf + 3;
1167                                 fn = p + 3;
1168                         } else
1169                                 fn = fbuf;
1170                         /* Get rid of trailing "/" and "/.". */
1171                         while (len) {
1172                                 if (fn[len - 1] == '/') {
1173                                         is_dot_dir = 1;
1174                                         if (!--len && !dir) {
1175                                                 len++;
1176                                                 break;
1177                                         }
1178                                 }
1179                                 else if (len >= 2 && fn[len - 1] == '.'
1180                                                   && fn[len - 2] == '/') {
1181                                         is_dot_dir = 1;
1182                                         if (!(len -= 2) && !dir) {
1183                                                 len++;
1184                                                 break;
1185                                         }
1186                                 } else
1187                                         break;
1188                         }
1189                         fn[len] = '\0';
1190                         /* Reject a ".." dir in the active part of the path. */
1191                         for (p = fn; (p = strstr(p, "..")) != NULL; p += 2) {
1192                                 if ((p[2] == '/' || p[2] == '\0')
1193                                  && (p == fn || p[-1] == '/')) {
1194                                         rprintf(FERROR,
1195                                             "found \"..\" dir in relative path: %s\n",
1196                                             fbuf);
1197                                         exit_cleanup(RERR_SYNTAX);
1198                                 }
1199                         }
1200                 }
1201
1202                 if (!*fn) {
1203                         len = 1;
1204                         fn = ".";
1205                 }
1206
1207                 if (dir && *dir) {
1208                         static char *lastdir;
1209                         static int lastdir_len;
1210
1211                         strlcpy(olddir, curr_dir, sizeof olddir);
1212
1213                         if (!push_dir(dir)) {
1214                                 io_error |= IOERR_GENERAL;
1215                                 rsyserr(FERROR, errno, "push_dir %s failed",
1216                                         full_fname(dir));
1217                                 continue;
1218                         }
1219
1220                         if (lastdir && strcmp(lastdir, dir) == 0) {
1221                                 flist_dir = lastdir;
1222                                 flist_dir_len = lastdir_len;
1223                         } else {
1224                                 flist_dir = lastdir = strdup(dir);
1225                                 flist_dir_len = lastdir_len = strlen(dir);
1226                         }
1227                 }
1228
1229                 if (fn != fbuf)
1230                         memmove(fbuf, fn, len + 1);
1231
1232                 if (implied_dirs && (p=strrchr(fbuf,'/')) && p != fbuf) {
1233                         /* Send the implied directories at the start of the
1234                          * source spec, so we get their permissions right. */
1235                         char *lp = lastpath, *slash = fbuf;
1236                         *p = '\0';
1237                         /* Skip any initial directories in our path that we
1238                          * have in common with lastpath. */
1239                         for (fn = fbuf; *fn && *lp == *fn; lp++, fn++) {
1240                                 if (*fn == '/')
1241                                         slash = fn;
1242                         }
1243                         *p = '/';
1244                         if (fn != p || (*lp && *lp != '/')) {
1245                                 int save_copy_links = copy_links;
1246                                 int save_xfer_dirs = xfer_dirs;
1247                                 copy_links |= copy_unsafe_links;
1248                                 xfer_dirs = 1;
1249                                 while ((slash = strchr(slash+1, '/')) != 0) {
1250                                         *slash = '\0';
1251                                         send_file_name(f, flist, fbuf, NULL, 0);
1252                                         *slash = '/';
1253                                 }
1254                                 copy_links = save_copy_links;
1255                                 xfer_dirs = save_xfer_dirs;
1256                                 *p = '\0';
1257                                 strlcpy(lastpath, fbuf, sizeof lastpath);
1258                                 *p = '/';
1259                         }
1260                 }
1261
1262                 if (one_file_system)
1263                         filesystem_dev = st.st_dev;
1264
1265                 if (recurse || (xfer_dirs && is_dot_dir)) {
1266                         struct file_struct *file;
1267                         file = send_file_name(f, flist, fbuf, &st, FLAG_TOP_DIR);
1268                         if (file)
1269                                 send_if_directory(f, flist, file, fbuf, len);
1270                 } else
1271                         send_file_name(f, flist, fbuf, &st, 0);
1272
1273                 if (olddir[0]) {
1274                         flist_dir = NULL;
1275                         flist_dir_len = 0;
1276                         if (!pop_dir(olddir)) {
1277                                 rsyserr(FERROR, errno, "pop_dir %s failed",
1278                                         full_fname(olddir));
1279                                 exit_cleanup(RERR_FILESELECT);
1280                         }
1281                 }
1282         }
1283
1284         gettimeofday(&end_tv, NULL);
1285         stats.flist_buildtime = (int64)(end_tv.tv_sec - start_tv.tv_sec) * 1000
1286                               + (end_tv.tv_usec - start_tv.tv_usec) / 1000;
1287         if (stats.flist_buildtime == 0)
1288                 stats.flist_buildtime = 1;
1289         start_tv = end_tv;
1290
1291         send_file_entry(NULL, f);
1292
1293         if (show_filelist_p())
1294                 finish_filelist_progress(flist);
1295
1296         gettimeofday(&end_tv, NULL);
1297         stats.flist_xfertime = (int64)(end_tv.tv_sec - start_tv.tv_sec) * 1000
1298                              + (end_tv.tv_usec - start_tv.tv_usec) / 1000;
1299
1300         if (flist->hlink_pool) {
1301                 pool_destroy(flist->hlink_pool);
1302                 flist->hlink_pool = NULL;
1303         }
1304
1305         /* Sort the list without removing any duplicates.  This allows the
1306          * receiving side to ask for any name they like, which gives us the
1307          * flexibility to change the way we unduplicate names in the future
1308          * without causing a compatibility problem with older versions. */
1309         clean_flist(flist, 0, 0);
1310
1311         send_uid_list(f);
1312
1313         /* send the io_error flag */
1314         write_int(f, lp_ignore_errors(module_id) ? 0 : io_error);
1315
1316         io_end_buffering();
1317         stats.flist_size = stats.total_written - start_write;
1318         stats.num_files = flist->count;
1319
1320         if (verbose > 3)
1321                 output_flist(flist);
1322
1323         if (verbose > 2)
1324                 rprintf(FINFO, "send_file_list done\n");
1325
1326         return flist;
1327 }
1328
1329 struct file_list *recv_file_list(int f)
1330 {
1331         struct file_list *flist;
1332         unsigned short flags;
1333         int64 start_read;
1334
1335         if (show_filelist_p())
1336                 start_filelist_progress("receiving file list");
1337
1338         start_read = stats.total_read;
1339
1340         flist = flist_new(WITH_HLINK, "recv_file_list");
1341
1342         flist->count = 0;
1343         flist->malloced = 1000;
1344         flist->files = new_array(struct file_struct *, flist->malloced);
1345         if (!flist->files)
1346                 goto oom;
1347
1348         while ((flags = read_byte(f)) != 0) {
1349                 struct file_struct *file;
1350
1351                 flist_expand(flist);
1352
1353                 if (protocol_version >= 28 && (flags & XMIT_EXTENDED_FLAGS))
1354                         flags |= read_byte(f) << 8;
1355                 file = receive_file_entry(flist, flags, f);
1356
1357                 if (S_ISREG(file->mode) || S_ISLNK(file->mode))
1358                         stats.total_size += file->length;
1359
1360                 flist->files[flist->count++] = file;
1361
1362                 maybe_emit_filelist_progress(flist->count);
1363
1364                 if (verbose > 2) {
1365                         rprintf(FINFO, "recv_file_name(%s)\n",
1366                                 f_name(file, NULL));
1367                 }
1368         }
1369         receive_file_entry(NULL, 0, 0); /* Signal that we're done. */
1370
1371         if (verbose > 2)
1372                 rprintf(FINFO, "received %d names\n", flist->count);
1373
1374         if (show_filelist_p())
1375                 finish_filelist_progress(flist);
1376
1377         clean_flist(flist, relative_paths, 1);
1378
1379         if (f >= 0) {
1380                 recv_uid_list(f, flist);
1381
1382                 /* Recv the io_error flag */
1383                 if (lp_ignore_errors(module_id) || ignore_errors)
1384                         read_int(f);
1385                 else
1386                         io_error |= read_int(f);
1387         }
1388
1389         if (verbose > 3)
1390                 output_flist(flist);
1391
1392         if (list_only) {
1393                 int i;
1394                 for (i = 0; i < flist->count; i++)
1395                         list_file_entry(flist->files[i]);
1396         }
1397
1398         if (verbose > 2)
1399                 rprintf(FINFO, "recv_file_list done\n");
1400
1401         stats.flist_size = stats.total_read - start_read;
1402         stats.num_files = flist->count;
1403
1404         return flist;
1405
1406   oom:
1407         out_of_memory("recv_file_list");
1408         return NULL;            /* not reached */
1409 }
1410
1411 static int file_compare(struct file_struct **file1, struct file_struct **file2)
1412 {
1413         return f_name_cmp(*file1, *file2);
1414 }
1415
1416 /* Search for an identically-named item in the file list.  Note that the
1417  * items must agree in their directory-ness, or no match is returned. */
1418 int flist_find(struct file_list *flist, struct file_struct *f)
1419 {
1420         int low = flist->low, high = flist->high;
1421         int diff, mid, mid_up;
1422
1423         while (low <= high) {
1424                 mid = (low + high) / 2;
1425                 if (flist->files[mid]->basename)
1426                         mid_up = mid;
1427                 else {
1428                         /* Scan for the next non-empty entry using the cached
1429                          * distance values.  If the value isn't fully up-to-
1430                          * date, update it. */
1431                         mid_up = mid + flist->files[mid]->dir.depth;
1432                         if (!flist->files[mid_up]->basename) {
1433                                 do {
1434                                     mid_up += flist->files[mid_up]->dir.depth;
1435                                 } while (!flist->files[mid_up]->basename);
1436                                 flist->files[mid]->dir.depth = mid_up - mid;
1437                         }
1438                         if (mid_up > high) {
1439                                 /* If there's nothing left above us, set high to
1440                                  * a non-empty entry below us and continue. */
1441                                 high = mid - flist->files[mid]->length;
1442                                 if (!flist->files[high]->basename) {
1443                                         do {
1444                                             high -= flist->files[high]->length;
1445                                         } while (!flist->files[high]->basename);
1446                                         flist->files[mid]->length = mid - high;
1447                                 }
1448                                 continue;
1449                         }
1450                 }
1451                 diff = f_name_cmp(flist->files[mid_up], f);
1452                 if (diff == 0) {
1453                         if (protocol_version < 29
1454                             && S_ISDIR(flist->files[mid_up]->mode)
1455                             != S_ISDIR(f->mode))
1456                                 return -1;
1457                         return mid_up;
1458                 }
1459                 if (diff < 0)
1460                         low = mid_up + 1;
1461                 else
1462                         high = mid - 1;
1463         }
1464         return -1;
1465 }
1466
1467 /*
1468  * Free up any resources a file_struct has allocated
1469  * and clear the file.
1470  */
1471 void clear_file(struct file_struct *file, struct file_list *flist)
1472 {
1473         if (flist->hlink_pool && file->link_u.idev)
1474                 pool_free(flist->hlink_pool, 0, file->link_u.idev);
1475         memset(file, 0, file_struct_len);
1476         /* In an empty entry, dir.depth is an offset to the next non-empty
1477          * entry.  Likewise for length in the opposite direction.  We assume
1478          * that we're alone for now since flist_find() will adjust the counts
1479          * it runs into that aren't up-to-date. */
1480         file->length = file->dir.depth = 1;
1481 }
1482
1483 /*
1484  * allocate a new file list
1485  */
1486 struct file_list *flist_new(int with_hlink, char *msg)
1487 {
1488         struct file_list *flist;
1489
1490         flist = new(struct file_list);
1491         if (!flist)
1492                 out_of_memory(msg);
1493
1494         memset(flist, 0, sizeof (struct file_list));
1495
1496         if (!(flist->file_pool = pool_create(FILE_EXTENT, 0,
1497             out_of_memory, POOL_INTERN)))
1498                 out_of_memory(msg);
1499
1500 #ifdef SUPPORT_HARD_LINKS
1501         if (with_hlink && preserve_hard_links) {
1502                 if (!(flist->hlink_pool = pool_create(HLINK_EXTENT,
1503                     sizeof (struct idev), out_of_memory, POOL_INTERN)))
1504                         out_of_memory(msg);
1505         }
1506 #endif
1507
1508         return flist;
1509 }
1510
1511 /*
1512  * free up all elements in a flist
1513  */
1514 void flist_free(struct file_list *flist)
1515 {
1516         pool_destroy(flist->file_pool);
1517         pool_destroy(flist->hlink_pool);
1518         free(flist->files);
1519         free(flist);
1520 }
1521
1522 /*
1523  * This routine ensures we don't have any duplicate names in our file list.
1524  * duplicate names can cause corruption because of the pipelining
1525  */
1526 static void clean_flist(struct file_list *flist, int strip_root, int no_dups)
1527 {
1528         char fbuf[MAXPATHLEN];
1529         int i, prev_i = 0;
1530
1531         if (!flist)
1532                 return;
1533         if (flist->count == 0) {
1534                 flist->high = -1;
1535                 return;
1536         }
1537
1538         qsort(flist->files, flist->count,
1539             sizeof flist->files[0], (int (*)())file_compare);
1540
1541         for (i = no_dups? 0 : flist->count; i < flist->count; i++) {
1542                 if (flist->files[i]->basename) {
1543                         prev_i = i;
1544                         break;
1545                 }
1546         }
1547         flist->low = prev_i;
1548         while (++i < flist->count) {
1549                 int j;
1550                 struct file_struct *file = flist->files[i];
1551
1552                 if (!file->basename)
1553                         continue;
1554                 if (f_name_cmp(file, flist->files[prev_i]) == 0)
1555                         j = prev_i;
1556                 else if (protocol_version >= 29 && S_ISDIR(file->mode)) {
1557                         int save_mode = file->mode;
1558                         /* Make sure that this directory doesn't duplicate a
1559                          * non-directory earlier in the list. */
1560                         flist->high = prev_i;
1561                         file->mode = S_IFREG;
1562                         j = flist_find(flist, file);
1563                         file->mode = save_mode;
1564                 } else
1565                         j = -1;
1566                 if (j >= 0) {
1567                         struct file_struct *fp = flist->files[j];
1568                         int keep, drop;
1569                         /* If one is a dir and the other is not, we want to
1570                          * keep the dir because it might have contents in the
1571                          * list. */
1572                         if (S_ISDIR(file->mode) != S_ISDIR(fp->mode)) {
1573                                 if (S_ISDIR(file->mode))
1574                                         keep = i, drop = j;
1575                                 else
1576                                         keep = j, drop = i;
1577                         } else
1578                                 keep = j, drop = i;
1579                         if (verbose > 1 && !am_server) {
1580                                 rprintf(FINFO,
1581                                         "removing duplicate name %s from file list (%d)\n",
1582                                         f_name(file, fbuf), drop);
1583                         }
1584                         /* Make sure we don't lose track of a user-specified
1585                          * top directory. */
1586                         flist->files[keep]->flags |= flist->files[drop]->flags
1587                                                    & (FLAG_TOP_DIR|FLAG_DEL_HERE);
1588
1589                         clear_file(flist->files[drop], flist);
1590
1591                         if (keep == i) {
1592                                 if (flist->low == drop) {
1593                                         for (j = drop + 1;
1594                                              j < i && !flist->files[j]->basename;
1595                                              j++) {}
1596                                         flist->low = j;
1597                                 }
1598                                 prev_i = i;
1599                         }
1600                 } else
1601                         prev_i = i;
1602         }
1603         flist->high = no_dups ? prev_i : flist->count - 1;
1604
1605         if (strip_root) {
1606                 /* We need to strip off the leading slashes for relative
1607                  * paths, but this must be done _after_ the sorting phase. */
1608                 for (i = flist->low; i <= flist->high; i++) {
1609                         struct file_struct *file = flist->files[i];
1610
1611                         if (!file->dirname)
1612                                 continue;
1613                         while (*file->dirname == '/')
1614                                 file->dirname++;
1615                         if (!*file->dirname)
1616                                 file->dirname = NULL;
1617                 }
1618         }
1619
1620         if (prune_empty_dirs && no_dups) {
1621                 int j, prev_depth = 0;
1622
1623                 prev_i = 0; /* It's OK that this isn't really true. */
1624
1625                 for (i = flist->low; i <= flist->high; i++) {
1626                         struct file_struct *fp, *file = flist->files[i];
1627
1628                         /* This temporarily abuses the dir.depth value for a
1629                          * directory that is in a chain that might get pruned.
1630                          * We restore the old value if it gets a reprieve. */
1631                         if (S_ISDIR(file->mode) && file->dir.depth) {
1632                                 /* Dump empty dirs when coming back down. */
1633                                 for (j = prev_depth; j >= file->dir.depth; j--) {
1634                                         fp = flist->files[prev_i];
1635                                         if (fp->dir.depth >= 0)
1636                                                 break;
1637                                         prev_i = -fp->dir.depth-1;
1638                                         clear_file(fp, flist);
1639                                 }
1640                                 prev_depth = file->dir.depth;
1641                                 if (is_excluded(f_name(file, fbuf), 1,
1642                                                        ALL_FILTERS)) {
1643                                         /* Keep dirs through this dir. */
1644                                         for (j = prev_depth-1; ; j--) {
1645                                                 fp = flist->files[prev_i];
1646                                                 if (fp->dir.depth >= 0)
1647                                                         break;
1648                                                 prev_i = -fp->dir.depth-1;
1649                                                 fp->dir.depth = j;
1650                                         }
1651                                 } else
1652                                         file->dir.depth = -prev_i-1;
1653                                 prev_i = i;
1654                         } else {
1655                                 /* Keep dirs through this non-dir. */
1656                                 for (j = prev_depth; ; j--) {
1657                                         fp = flist->files[prev_i];
1658                                         if (fp->dir.depth >= 0)
1659                                                 break;
1660                                         prev_i = -fp->dir.depth-1;
1661                                         fp->dir.depth = j;
1662                                 }
1663                         }
1664                 }
1665                 /* Dump empty all remaining empty dirs. */
1666                 while (1) {
1667                         struct file_struct *fp = flist->files[prev_i];
1668                         if (fp->dir.depth >= 0)
1669                                 break;
1670                         prev_i = -fp->dir.depth-1;
1671                         clear_file(fp, flist);
1672                 }
1673
1674                 for (i = flist->low; i <= flist->high; i++) {
1675                         if (flist->files[i]->basename)
1676                                 break;
1677                 }
1678                 flist->low = i;
1679                 for (i = flist->high; i >= flist->low; i--) {
1680                         if (flist->files[i]->basename)
1681                                 break;
1682                 }
1683                 flist->high = i;
1684         }
1685 }
1686
1687 static void output_flist(struct file_list *flist)
1688 {
1689         char uidbuf[16], gidbuf[16], depthbuf[16];
1690         struct file_struct *file;
1691         const char *who = who_am_i();
1692         int i;
1693
1694         for (i = 0; i < flist->count; i++) {
1695                 file = flist->files[i];
1696                 if ((am_root || am_sender) && preserve_uid)
1697                         sprintf(uidbuf, " uid=%ld", (long)file->uid);
1698                 else
1699                         *uidbuf = '\0';
1700                 if (preserve_gid && file->gid != GID_NONE)
1701                         sprintf(gidbuf, " gid=%ld", (long)file->gid);
1702                 else
1703                         *gidbuf = '\0';
1704                 if (!am_sender)
1705                         sprintf(depthbuf, "%d", file->dir.depth);
1706                 rprintf(FINFO, "[%s] i=%d %s %s%s%s%s mode=0%o len=%.0f%s%s flags=%x\n",
1707                         who, i, am_sender ? NS(file->dir.root) : depthbuf,
1708                         file->dirname ? file->dirname : "",
1709                         file->dirname ? "/" : "", NS(file->basename),
1710                         S_ISDIR(file->mode) ? "/" : "", (int)file->mode,
1711                         (double)file->length, uidbuf, gidbuf, file->flags);
1712         }
1713 }
1714
1715 enum fnc_state { s_DIR, s_SLASH, s_BASE, s_TRAILING };
1716 enum fnc_type { t_PATH, t_ITEM };
1717
1718 /* Compare the names of two file_struct entities, similar to how strcmp()
1719  * would do if it were operating on the joined strings.
1720  *
1721  * Some differences beginning with protocol_version 29: (1) directory names
1722  * are compared with an assumed trailing slash so that they compare in a
1723  * way that would cause them to sort immediately prior to any content they
1724  * may have; (2) a directory of any name compares after a non-directory of
1725  * any name at the same depth; (3) a directory with name "." compares prior
1726  * to anything else.  These changes mean that a directory and a non-dir
1727  * with the same name will not compare as equal (protocol_version >= 29).
1728  *
1729  * The dirname component can be an empty string, but the basename component
1730  * cannot (and never is in the current codebase).  The basename component
1731  * may be NULL (for a removed item), in which case it is considered to be
1732  * after any existing item. */
1733 int f_name_cmp(struct file_struct *f1, struct file_struct *f2)
1734 {
1735         int dif;
1736         const uchar *c1, *c2;
1737         enum fnc_state state1, state2;
1738         enum fnc_type type1, type2;
1739         enum fnc_type t_path = protocol_version >= 29 ? t_PATH : t_ITEM;
1740
1741         if (!f1 || !f1->basename) {
1742                 if (!f2 || !f2->basename)
1743                         return 0;
1744                 return -1;
1745         }
1746         if (!f2 || !f2->basename)
1747                 return 1;
1748
1749         c1 = (uchar*)f1->dirname;
1750         c2 = (uchar*)f2->dirname;
1751         if (c1 == c2)
1752                 c1 = c2 = NULL;
1753         if (!c1) {
1754                 type1 = S_ISDIR(f1->mode) ? t_path : t_ITEM;
1755                 c1 = (uchar*)f1->basename;
1756                 if (type1 == t_PATH && *c1 == '.' && !c1[1]) {
1757                         type1 = t_ITEM;
1758                         state1 = s_TRAILING;
1759                         c1 = (uchar*)"";
1760                 } else
1761                         state1 = s_BASE;
1762         } else if (!*c1) {
1763                 type1 = t_path;
1764                 state1 = s_SLASH;
1765                 c1 = (uchar*)"/";
1766         } else {
1767                 type1 = t_path;
1768                 state1 = s_DIR;
1769         }
1770         if (!c2) {
1771                 type2 = S_ISDIR(f2->mode) ? t_path : t_ITEM;
1772                 c2 = (uchar*)f2->basename;
1773                 if (type2 == t_PATH && *c2 == '.' && !c2[1]) {
1774                         type2 = t_ITEM;
1775                         state2 = s_TRAILING;
1776                         c2 = (uchar*)"";
1777                 } else
1778                         state2 = s_BASE;
1779         } else if (!*c2) {
1780                 type2 = t_path;
1781                 state2 = s_SLASH;
1782                 c2 = (uchar*)"/";
1783         } else {
1784                 type2 = t_path;
1785                 state2 = s_DIR;
1786         }
1787
1788         if (type1 != type2)
1789                 return type1 == t_PATH ? 1 : -1;
1790
1791         while (1) {
1792                 if ((dif = (int)*c1++ - (int)*c2++) != 0)
1793                         break;
1794                 if (!*c1) {
1795                         switch (state1) {
1796                         case s_DIR:
1797                                 state1 = s_SLASH;
1798                                 c1 = (uchar*)"/";
1799                                 break;
1800                         case s_SLASH:
1801                                 type1 = S_ISDIR(f1->mode) ? t_path : t_ITEM;
1802                                 c1 = (uchar*)f1->basename;
1803                                 if (type1 == t_PATH && *c1 == '.' && !c1[1]) {
1804                                         type1 = t_ITEM;
1805                                         state1 = s_TRAILING;
1806                                         c1 = (uchar*)"";
1807                                 } else
1808                                         state1 = s_BASE;
1809                                 break;
1810                         case s_BASE:
1811                                 state1 = s_TRAILING;
1812                                 if (type1 == t_PATH) {
1813                                         c1 = (uchar*)"/";
1814                                         break;
1815                                 }
1816                                 /* FALL THROUGH */
1817                         case s_TRAILING:
1818                                 type1 = t_ITEM;
1819                                 break;
1820                         }
1821                         if (*c2 && type1 != type2)
1822                                 return type1 == t_PATH ? 1 : -1;
1823                 }
1824                 if (!*c2) {
1825                         switch (state2) {
1826                         case s_DIR:
1827                                 state2 = s_SLASH;
1828                                 c2 = (uchar*)"/";
1829                                 break;
1830                         case s_SLASH:
1831                                 type2 = S_ISDIR(f2->mode) ? t_path : t_ITEM;
1832                                 c2 = (uchar*)f2->basename;
1833                                 if (type2 == t_PATH && *c2 == '.' && !c2[1]) {
1834                                         type2 = t_ITEM;
1835                                         state2 = s_TRAILING;
1836                                         c2 = (uchar*)"";
1837                                 } else
1838                                         state2 = s_BASE;
1839                                 break;
1840                         case s_BASE:
1841                                 state2 = s_TRAILING;
1842                                 if (type2 == t_PATH) {
1843                                         c2 = (uchar*)"/";
1844                                         break;
1845                                 }
1846                                 /* FALL THROUGH */
1847                         case s_TRAILING:
1848                                 if (!*c1)
1849                                         return 0;
1850                                 type2 = t_ITEM;
1851                                 break;
1852                         }
1853                         if (type1 != type2)
1854                                 return type1 == t_PATH ? 1 : -1;
1855                 }
1856         }
1857
1858         return dif;
1859 }
1860
1861 /* Return a copy of the full filename of a flist entry, using the indicated
1862  * buffer or one of 5 static buffers if fbuf is NULL.  No size-checking is
1863  * done because we checked the size when creating the file_struct entry.
1864  */
1865 char *f_name(struct file_struct *f, char *fbuf)
1866 {
1867         if (!f || !f->basename)
1868                 return NULL;
1869
1870         if (!fbuf) {
1871                 static char names[5][MAXPATHLEN];
1872                 static unsigned int n;
1873
1874                 n = (n + 1) % (sizeof names / sizeof names[0]);
1875
1876                 fbuf = names[n];
1877         }
1878
1879         if (f->dirname) {
1880                 int len = strlen(f->dirname);
1881                 memcpy(fbuf, f->dirname, len);
1882                 fbuf[len] = '/';
1883                 strcpy(fbuf + len + 1, f->basename);
1884         } else
1885                 strcpy(fbuf, f->basename);
1886
1887         return fbuf;
1888 }
1889
1890 /* Do a non-recursive scan of the named directory, possibly ignoring all
1891  * exclude rules except for the daemon's.  If "dlen" is >=0, it is the length
1892  * of the dirname string, and also indicates that "dirname" is a MAXPATHLEN
1893  * buffer (the functions we call will append names onto the end, but the old
1894  * dir value will be restored on exit). */
1895 struct file_list *get_dirlist(char *dirname, int dlen,
1896                               int ignore_filter_rules)
1897 {
1898         struct file_list *dirlist;
1899         char dirbuf[MAXPATHLEN];
1900         int save_recurse = recurse;
1901         int save_xfer_dirs = xfer_dirs;
1902
1903         if (dlen < 0) {
1904                 dlen = strlcpy(dirbuf, dirname, MAXPATHLEN);
1905                 if (dlen >= MAXPATHLEN)
1906                         return NULL;
1907                 dirname = dirbuf;
1908         }
1909
1910         dirlist = flist_new(WITHOUT_HLINK, "get_dirlist");
1911
1912         recurse = 0;
1913         xfer_dirs = 1;
1914         send_directory(ignore_filter_rules ? -2 : -1, dirlist, dirname, dlen);
1915         xfer_dirs = save_xfer_dirs;
1916         recurse = save_recurse;
1917         if (do_progress)
1918                 flist_count_offset += dirlist->count;
1919
1920         clean_flist(dirlist, 0, 0);
1921
1922         if (verbose > 3)
1923                 output_flist(dirlist);
1924
1925         return dirlist;
1926 }