Implemented an extended-flag-byte in the transfer protocol by moving
[rsync/rsync.git] / flist.c
1 /*
2    Copyright (C) Andrew Tridgell 1996
3    Copyright (C) Paul Mackerras 1996
4    Copyright (C) 2001, 2002 by Martin Pool <mbp@samba.org>
5
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.
10
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.
15
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.
19 */
20
21 /** @file flist.c
22  * Generate and receive file lists
23  *
24  * @todo Get rid of the string_area optimization.  Efficiently
25  * allocating blocks is the responsibility of the system's malloc
26  * library, not of rsync.
27  *
28  * @sa http://lists.samba.org/pipermail/rsync/2000-June/002351.html
29  *
30  **/
31
32 #include "rsync.h"
33
34 extern struct stats stats;
35
36 extern int verbose;
37 extern int do_progress;
38 extern int am_server;
39 extern int always_checksum;
40
41 extern int cvs_exclude;
42
43 extern int recurse;
44 extern char *files_from;
45 extern int filesfrom_fd;
46
47 extern int one_file_system;
48 extern int make_backups;
49 extern int preserve_links;
50 extern int preserve_hard_links;
51 extern int preserve_perms;
52 extern int preserve_devices;
53 extern int preserve_uid;
54 extern int preserve_gid;
55 extern int preserve_times;
56 extern int relative_paths;
57 extern int implied_dirs;
58 extern int copy_links;
59 extern int copy_unsafe_links;
60 extern int protocol_version;
61 extern int sanitize_paths;
62
63 extern int read_batch;
64 extern int write_batch;
65
66 extern struct exclude_struct **exclude_list;
67 extern struct exclude_struct **server_exclude_list;
68 extern struct exclude_struct **local_exclude_list;
69
70 int io_error;
71
72 static struct file_struct null_file;
73
74 static void clean_flist(struct file_list *flist, int strip_root, int no_dups);
75
76
77 static int show_filelist_p(void)
78 {
79         return verbose && (recurse || files_from) && !am_server;
80 }
81
82 static void start_filelist_progress(char *kind)
83 {
84         rprintf(FINFO, "%s ... ", kind);
85         if ((verbose > 1) || do_progress)
86                 rprintf(FINFO, "\n");
87         rflush(FINFO);
88 }
89
90
91 static void emit_filelist_progress(const struct file_list *flist)
92 {
93         rprintf(FINFO, " %d files...\r", flist->count);
94 }
95
96
97 static void maybe_emit_filelist_progress(const struct file_list *flist)
98 {
99         if (do_progress && show_filelist_p() && ((flist->count % 100) == 0))
100                 emit_filelist_progress(flist);
101 }
102
103
104 static void finish_filelist_progress(const struct file_list *flist)
105 {
106         if (do_progress) {
107                 /* This overwrites the progress line */
108                 rprintf(FINFO, "%d file%sto consider\n",
109                         flist->count, flist->count == 1 ? " " : "s ");
110         } else
111                 rprintf(FINFO, "done\n");
112 }
113
114 void show_flist_stats(void)
115 {
116         /* Nothing yet */
117 }
118
119
120 static struct string_area *string_area_new(int size)
121 {
122         struct string_area *a;
123
124         if (size <= 0)
125                 size = ARENA_SIZE;
126         a = new(struct string_area);
127         if (!a)
128                 out_of_memory("string_area_new");
129         a->current = a->base = new_array(char, size);
130         if (!a->current)
131                 out_of_memory("string_area_new buffer");
132         a->end = a->base + size;
133         a->next = NULL;
134
135         return a;
136 }
137
138 static void string_area_free(struct string_area *a)
139 {
140         struct string_area *next;
141
142         for (; a; a = next) {
143                 next = a->next;
144                 free(a->base);
145         }
146 }
147
148 static char *string_area_malloc(struct string_area **ap, int size)
149 {
150         char *p;
151         struct string_area *a;
152
153         /* does the request fit into the current space? */
154         a = *ap;
155         if (a->current + size >= a->end) {
156                 /* no; get space, move new string_area to front of the list */
157                 a = string_area_new(size > ARENA_SIZE ? size : ARENA_SIZE);
158                 a->next = *ap;
159                 *ap = a;
160         }
161
162         /* have space; do the "allocation." */
163         p = a->current;
164         a->current += size;
165         return p;
166 }
167
168 static char *string_area_strdup(struct string_area **ap, const char *src)
169 {
170         char *dest = string_area_malloc(ap, strlen(src) + 1);
171         return strcpy(dest, src);
172 }
173
174 static void list_file_entry(struct file_struct *f)
175 {
176         char perms[11];
177
178         if (!f->basename)
179                 /* this can happen if duplicate names were removed */
180                 return;
181
182         permstring(perms, f->mode);
183
184         if (preserve_links && S_ISLNK(f->mode)) {
185                 rprintf(FINFO, "%s %11.0f %s %s -> %s\n",
186                         perms,
187                         (double) f->length, timestring(f->modtime),
188                         f_name(f), f->link);
189         } else {
190                 rprintf(FINFO, "%s %11.0f %s %s\n",
191                         perms,
192                         (double) f->length, timestring(f->modtime),
193                         f_name(f));
194         }
195 }
196
197
198 /**
199  * Stat either a symlink or its referent, depending on the settings of
200  * copy_links, copy_unsafe_links, etc.
201  *
202  * @retval -1 on error
203  *
204  * @retval 0 for success
205  *
206  * @post If @p path is a symlink, then @p linkbuf (of size @c
207  * MAXPATHLEN) contains the symlink target.
208  *
209  * @post @p buffer contains information about the link or the
210  * referrent as appropriate, if they exist.
211  **/
212 int readlink_stat(const char *path, STRUCT_STAT *buffer, char *linkbuf)
213 {
214 #if SUPPORT_LINKS
215         if (copy_links)
216                 return do_stat(path, buffer);
217         if (do_lstat(path, buffer) == -1)
218                 return -1;
219         if (S_ISLNK(buffer->st_mode)) {
220                 int l = readlink((char *) path, linkbuf, MAXPATHLEN - 1);
221                 if (l == -1)
222                         return -1;
223                 linkbuf[l] = 0;
224                 if (copy_unsafe_links && unsafe_symlink(linkbuf, path)) {
225                         if (verbose > 1) {
226                                 rprintf(FINFO,"copying unsafe symlink \"%s\" -> \"%s\"\n",
227                                         path, linkbuf);
228                         }
229                         return do_stat(path, buffer);
230                 }
231         }
232         return 0;
233 #else
234         return do_stat(path, buffer);
235 #endif
236 }
237
238 int link_stat(const char *path, STRUCT_STAT * buffer)
239 {
240 #if SUPPORT_LINKS
241         if (copy_links)
242                 return do_stat(path, buffer);
243         return do_lstat(path, buffer);
244 #else
245         return do_stat(path, buffer);
246 #endif
247 }
248
249 /*
250  * This function is used to check if a file should be included/excluded
251  * from the list of files based on its name and type etc.  The value of
252  * exclude_level is set to either SERVER_EXCLUDES or ALL_EXCLUDES.
253  */
254 static int check_exclude_file(char *fname, int is_dir, int exclude_level)
255 {
256 #if 0 /* This currently never happens, so avoid a useless compare. */
257         if (exclude_level == NO_EXCLUDES)
258                 return 0;
259 #endif
260         if (fname) {
261                 /* never exclude '.', even if somebody does --exclude '*' */
262                 if (fname[0] == '.' && !fname[1])
263                         return 0;
264                 /* Handle the -R version of the '.' dir. */
265                 if (fname[0] == '/') {
266                         int len = strlen(fname);
267                         if (fname[len-1] == '.' && fname[len-2] == '/')
268                                 return 0;
269                 }
270         }
271         if (server_exclude_list
272          && check_exclude(server_exclude_list, fname, is_dir))
273                 return 1;
274         if (exclude_level != ALL_EXCLUDES)
275                 return 0;
276         if (exclude_list && check_exclude(exclude_list, fname, is_dir))
277                 return 1;
278         if (local_exclude_list
279          && check_exclude(local_exclude_list, fname, is_dir))
280                 return 1;
281         return 0;
282 }
283
284 /* used by the one_file_system code */
285 static dev_t filesystem_dev;
286
287 static void set_filesystem(char *fname)
288 {
289         STRUCT_STAT st;
290         if (link_stat(fname, &st) != 0)
291                 return;
292         filesystem_dev = st.st_dev;
293 }
294
295
296 static int to_wire_mode(mode_t mode)
297 {
298         if (S_ISLNK(mode) && (_S_IFLNK != 0120000))
299                 return (mode & ~(_S_IFMT)) | 0120000;
300         return (int) mode;
301 }
302
303 static mode_t from_wire_mode(int mode)
304 {
305         if ((mode & (_S_IFMT)) == 0120000 && (_S_IFLNK != 0120000))
306                 return (mode & ~(_S_IFMT)) | _S_IFLNK;
307         return (mode_t) mode;
308 }
309
310
311 static void send_directory(int f, struct file_list *flist, char *dir);
312
313 static char *flist_dir;
314
315
316 /**
317  * Make sure @p flist is big enough to hold at least @p flist->count
318  * entries.
319  **/
320 static void flist_expand(struct file_list *flist)
321 {
322         if (flist->count >= flist->malloced) {
323                 void *new_ptr;
324
325                 if (flist->malloced < 1000)
326                         flist->malloced += 1000;
327                 else
328                         flist->malloced *= 2;
329
330                 if (flist->files) {
331                         new_ptr = realloc_array(flist->files,
332                                                 struct file_struct *,
333                                                 flist->malloced);
334                 } else {
335                         new_ptr = new_array(struct file_struct *,
336                                             flist->malloced);
337                 }
338
339                 if (verbose >= 2) {
340                         rprintf(FINFO, "expand file_list to %.0f bytes, did%s move\n",
341                                 (double)sizeof(flist->files[0])
342                                 * flist->malloced,
343                                 (new_ptr == flist->files) ? " not" : "");
344                 }
345
346                 flist->files = (struct file_struct **) new_ptr;
347
348                 if (!flist->files)
349                         out_of_memory("flist_expand");
350         }
351 }
352
353
354 static void send_file_entry(struct file_struct *file, int f,
355                             unsigned short base_flags)
356 {
357         unsigned short flags;
358         static time_t last_time;
359         static mode_t last_mode;
360         static DEV64_T last_rdev;
361         static uid_t last_uid;
362         static gid_t last_gid;
363         static char lastname[MAXPATHLEN];
364         char *fname, fbuf[MAXPATHLEN];
365         int l1, l2;
366
367         if (f == -1)
368                 return;
369
370         if (!file) {
371                 write_byte(f, 0);
372                 return;
373         }
374
375         io_write_phase = "send_file_entry";
376
377         fname = f_name_to(file, fbuf, sizeof fbuf);
378
379         flags = base_flags;
380
381         if (file->mode == last_mode)
382                 flags |= SAME_MODE;
383         else
384                 last_mode = file->mode;
385         if (preserve_devices) {
386                 if (protocol_version < 28) {
387                         if (IS_DEVICE(last_mode) && file->rdev == last_rdev)
388                                 flags |= OLD_SAME_RDEV|SAME_HIGH_RDEV;
389                         else
390                                 last_rdev = file->rdev;
391                 }
392                 else if (IS_DEVICE(last_mode)) {
393                         if ((file->rdev & ~0xFF) == (last_rdev & ~0xFF))
394                                 flags |= SAME_HIGH_RDEV;
395                         last_rdev = file->rdev;
396                 }
397         }
398         if (file->uid == last_uid)
399                 flags |= SAME_UID;
400         else
401                 last_uid = file->uid;
402         if (file->gid == last_gid)
403                 flags |= SAME_GID;
404         else
405                 last_gid = file->gid;
406         if (file->modtime == last_time)
407                 flags |= SAME_TIME;
408         else
409                 last_time = file->modtime;
410
411         for (l1 = 0;
412              lastname[l1] && (fname[l1] == lastname[l1]) && (l1 < 255);
413              l1++) {}
414         l2 = strlen(fname) - l1;
415
416         if (l1 > 0)
417                 flags |= SAME_NAME;
418         if (l2 > 255)
419                 flags |= LONG_NAME;
420
421         /* We must make sure we don't send a zero flags byte or
422          * the other end will terminate the flist transfer. */
423         if (flags == 0 && !S_ISDIR(last_mode))
424                 flags |= FLAG_DELETE; /* NOTE: no meaning for non-dir */
425         if (protocol_version >= 28) {
426                 if ((flags & 0xFF00) || flags == 0) {
427                         flags |= EXTENDED_FLAGS;
428                         write_byte(f, flags);
429                         write_byte(f, flags >> 8);
430                 } else
431                         write_byte(f, flags);
432         } else {
433                 if (flags == 0)
434                         flags |= LONG_NAME;
435                 write_byte(f, flags);
436         }
437         if (flags & SAME_NAME)
438                 write_byte(f, l1);
439         if (flags & LONG_NAME)
440                 write_int(f, l2);
441         else
442                 write_byte(f, l2);
443         write_buf(f, fname + l1, l2);
444
445         write_longint(f, file->length);
446         if (!(flags & SAME_TIME))
447                 write_int(f, last_time);
448         if (!(flags & SAME_MODE))
449                 write_int(f, to_wire_mode(last_mode));
450         if (preserve_uid && !(flags & SAME_UID)) {
451                 add_uid(last_uid);
452                 write_int(f, last_uid);
453         }
454         if (preserve_gid && !(flags & SAME_GID)) {
455                 add_gid(last_gid);
456                 write_int(f, last_gid);
457         }
458         if (preserve_devices && IS_DEVICE(last_mode)) {
459                 /* If SAME_HIGH_RDEV is off, OLD_SAME_RDEV is also off. */
460                 if (!(flags & SAME_HIGH_RDEV))
461                         write_int(f, last_rdev);
462                 else if (protocol_version >= 28)
463                         write_byte(f, last_rdev);
464         }
465
466 #if SUPPORT_LINKS
467         if (preserve_links && S_ISLNK(last_mode)) {
468                 write_int(f, strlen(file->link));
469                 write_buf(f, file->link, strlen(file->link));
470         }
471 #endif
472
473 #if SUPPORT_HARD_LINKS
474         if (preserve_hard_links && S_ISREG(file->mode)) {
475                 if (protocol_version < 26) {
476                         /* 32-bit dev_t and ino_t */
477                         write_int(f, (int) file->dev);
478                         write_int(f, (int) file->inode);
479                 } else {
480                         /* 64-bit dev_t and ino_t */
481                         write_longint(f, file->dev);
482                         write_longint(f, file->inode);
483                 }
484         }
485 #endif
486
487         if (always_checksum) {
488                 if (protocol_version < 21)
489                         write_buf(f, file->sum, 2);
490                 else
491                         write_buf(f, file->sum, MD4_SUM_LENGTH);
492         }
493
494         strlcpy(lastname, fname, MAXPATHLEN);
495         lastname[MAXPATHLEN - 1] = 0;
496
497         io_write_phase = "unknown";
498 }
499
500
501
502 static void receive_file_entry(struct file_struct **fptr,
503                                unsigned short flags, int f)
504 {
505         static time_t last_time;
506         static mode_t last_mode;
507         static DEV64_T last_rdev;
508         static uid_t last_uid;
509         static gid_t last_gid;
510         static char lastname[MAXPATHLEN];
511         char thisname[MAXPATHLEN];
512         unsigned int l1 = 0, l2 = 0;
513         char *p;
514         struct file_struct *file;
515
516         if (flags & SAME_NAME)
517                 l1 = read_byte(f);
518
519         if (flags & LONG_NAME)
520                 l2 = read_int(f);
521         else
522                 l2 = read_byte(f);
523
524         file = new(struct file_struct);
525         if (!file)
526                 out_of_memory("receive_file_entry");
527         memset((char *) file, 0, sizeof(*file));
528         (*fptr) = file;
529
530         if (l2 >= MAXPATHLEN - l1) {
531                 rprintf(FERROR,
532                         "overflow: flags=0x%x l1=%d l2=%d lastname=%s\n",
533                         flags, l1, l2, lastname);
534                 overflow("receive_file_entry");
535         }
536
537         strlcpy(thisname, lastname, l1 + 1);
538         read_sbuf(f, &thisname[l1], l2);
539         thisname[l1 + l2] = 0;
540
541         strlcpy(lastname, thisname, MAXPATHLEN);
542         lastname[MAXPATHLEN - 1] = 0;
543
544         clean_fname(thisname);
545
546         if (sanitize_paths) {
547                 sanitize_path(thisname, NULL);
548         }
549
550         if ((p = strrchr(thisname, '/'))) {
551                 static char *lastdir;
552                 *p = 0;
553                 if (lastdir && strcmp(thisname, lastdir) == 0)
554                         file->dirname = lastdir;
555                 else {
556                         file->dirname = strdup(thisname);
557                         lastdir = file->dirname;
558                 }
559                 file->basename = strdup(p + 1);
560         } else {
561                 file->dirname = NULL;
562                 file->basename = strdup(thisname);
563         }
564
565         if (!file->basename)
566                 out_of_memory("receive_file_entry 1");
567
568         file->flags = flags;
569         file->length = read_longint(f);
570         if (!(flags & SAME_TIME))
571                 last_time = (time_t)read_int(f);
572         file->modtime = last_time;
573         if (!(flags & SAME_MODE))
574                 last_mode = from_wire_mode(read_int(f));
575         file->mode = last_mode;
576
577         if (preserve_uid) {
578                 if (!(flags & SAME_UID))
579                         last_uid = (uid_t)read_int(f);
580                 file->uid = last_uid;
581         }
582         if (preserve_gid) {
583                 if (!(flags & SAME_GID))
584                         last_gid = (gid_t)read_int(f);
585                 file->gid = last_gid;
586         }
587         if (preserve_devices) {
588                 if (protocol_version < 28) {
589                         if (IS_DEVICE(last_mode)) {
590                                 if (!(flags & OLD_SAME_RDEV))
591                                         last_rdev = (DEV64_T)read_int(f);
592                                 file->rdev = last_rdev;
593                         } else
594                                 last_rdev = 0;
595                 } else if (IS_DEVICE(last_mode)) {
596                         if (!(flags & SAME_HIGH_RDEV))
597                                 last_rdev = (DEV64_T)read_int(f);
598                         else {
599                                 last_rdev = (DEV64_T)((last_rdev & ~0xFF)
600                                                         | read_byte(f));
601                         }
602                         file->rdev = last_rdev;
603                 }
604         }
605
606         if (preserve_links && S_ISLNK(last_mode)) {
607                 int l = read_int(f);
608                 if (l < 0) {
609                         rprintf(FERROR, "overflow: l=%d\n", l);
610                         overflow("receive_file_entry");
611                 }
612                 file->link = new_array(char, l + 1);
613                 if (!file->link)
614                         out_of_memory("receive_file_entry 2");
615                 read_sbuf(f, file->link, l);
616                 if (sanitize_paths)
617                         sanitize_path(file->link, file->dirname);
618         }
619 #if SUPPORT_HARD_LINKS
620         if (preserve_hard_links && S_ISREG(file->mode)) {
621                 if (protocol_version < 26) {
622                         file->dev = read_int(f);
623                         file->inode = read_int(f);
624                 } else {
625                         file->dev = read_longint(f);
626                         file->inode = read_longint(f);
627                 }
628         }
629 #endif
630
631         if (always_checksum) {
632                 file->sum = new_array(char, MD4_SUM_LENGTH);
633                 if (!file->sum)
634                         out_of_memory("md4 sum");
635                 if (protocol_version < 21)
636                         read_buf(f, file->sum, 2);
637                 else
638                         read_buf(f, file->sum, MD4_SUM_LENGTH);
639         }
640
641         if (!preserve_perms) {
642                 extern int orig_umask;
643                 /* set an appropriate set of permissions based on original
644                    permissions and umask. This emulates what GNU cp does */
645                 file->mode &= ~orig_umask;
646         }
647 }
648
649
650 /* determine if a file in a different filesstem should be skipped
651    when one_file_system is set. We bascally only want to include
652    the mount points - but they can be hard to find! */
653 static int skip_filesystem(char *fname, STRUCT_STAT * st)
654 {
655         STRUCT_STAT st2;
656         char *p = strrchr(fname, '/');
657
658         /* skip all but directories */
659         if (!S_ISDIR(st->st_mode))
660                 return 1;
661
662         /* if its not a subdirectory then allow */
663         if (!p)
664                 return 0;
665
666         *p = 0;
667         if (link_stat(fname, &st2)) {
668                 *p = '/';
669                 return 0;
670         }
671         *p = '/';
672
673         return (st2.st_dev != filesystem_dev);
674 }
675
676 #define STRDUP(ap, p)   (ap ? string_area_strdup(ap, p) : strdup(p))
677 /* IRIX cc cares that the operands to the ternary have the same type. */
678 #define MALLOC(ap, i)   (ap ? (void*) string_area_malloc(ap, i) : malloc(i))
679
680 /**
681  * Create a file_struct for a named file by reading its stat()
682  * information and performing extensive checks against global
683  * options.
684  *
685  * @return the new file, or NULL if there was an error or this file
686  * should be excluded.
687  *
688  * @todo There is a small optimization opportunity here to avoid
689  * stat()ing the file in some circumstances, which has a certain cost.
690  * We are called immediately after doing readdir(), and so we may
691  * already know the d_type of the file.  We could for example avoid
692  * statting directories if we're not recursing, but this is not a very
693  * important case.  Some systems may not have d_type.
694  **/
695 struct file_struct *make_file(char *fname, struct string_area **ap,
696                               int exclude_level)
697 {
698         struct file_struct *file;
699         STRUCT_STAT st;
700         char sum[SUM_LENGTH];
701         char *p;
702         char cleaned_name[MAXPATHLEN];
703         char linkbuf[MAXPATHLEN];
704         extern int module_id;
705
706         strlcpy(cleaned_name, fname, MAXPATHLEN);
707         cleaned_name[MAXPATHLEN - 1] = 0;
708         clean_fname(cleaned_name);
709         if (sanitize_paths)
710                 sanitize_path(cleaned_name, NULL);
711         fname = cleaned_name;
712
713         memset(sum, 0, SUM_LENGTH);
714
715         if (readlink_stat(fname, &st, linkbuf) != 0) {
716                 int save_errno = errno;
717                 if (errno == ENOENT && exclude_level != NO_EXCLUDES) {
718                         /* either symlink pointing nowhere or file that
719                          * was removed during rsync run; see if excluded
720                          * before reporting an error */
721                         if (check_exclude_file(fname, 0, exclude_level)) {
722                                 /* file is excluded anyway, ignore silently */
723                                 return NULL;
724                         }
725                 }
726                 io_error |= IOERR_GENERAL;
727                 rprintf(FERROR, "readlink %s failed: %s\n",
728                         full_fname(fname), strerror(save_errno));
729                 return NULL;
730         }
731
732         /* backup.c calls us with exclude_level set to NO_EXCLUDES. */
733         if (exclude_level == NO_EXCLUDES)
734                 goto skip_excludes;
735
736         if (S_ISDIR(st.st_mode) && !recurse && !files_from) {
737                 rprintf(FINFO, "skipping directory %s\n", fname);
738                 return NULL;
739         }
740
741         if (one_file_system && st.st_dev != filesystem_dev) {
742                 if (skip_filesystem(fname, &st))
743                         return NULL;
744         }
745
746         if (check_exclude_file(fname, S_ISDIR(st.st_mode) != 0, exclude_level))
747                 return NULL;
748
749         if (lp_ignore_nonreadable(module_id) && access(fname, R_OK) != 0)
750                 return NULL;
751
752       skip_excludes:
753
754         if (verbose > 2)
755                 rprintf(FINFO, "make_file(%s,*,%d)\n", fname, exclude_level);
756
757         file = new(struct file_struct);
758         if (!file)
759                 out_of_memory("make_file");
760         memset((char *) file, 0, sizeof(*file));
761
762         if ((p = strrchr(fname, '/'))) {
763                 static char *lastdir;
764                 *p = 0;
765                 if (lastdir && strcmp(fname, lastdir) == 0)
766                         file->dirname = lastdir;
767                 else {
768                         file->dirname = strdup(fname);
769                         lastdir = file->dirname;
770                 }
771                 file->basename = STRDUP(ap, p + 1);
772                 *p = '/';
773         } else {
774                 file->dirname = NULL;
775                 file->basename = STRDUP(ap, fname);
776         }
777
778         file->modtime = st.st_mtime;
779         file->length = st.st_size;
780         file->mode = st.st_mode;
781         file->uid = st.st_uid;
782         file->gid = st.st_gid;
783         file->dev = st.st_dev;
784         file->inode = st.st_ino;
785 #ifdef HAVE_STRUCT_STAT_ST_RDEV
786         file->rdev = st.st_rdev;
787 #endif
788
789 #if SUPPORT_LINKS
790         if (S_ISLNK(st.st_mode))
791                 file->link = STRDUP(ap, linkbuf);
792 #endif
793
794         if (always_checksum) {
795                 file->sum = (char *) MALLOC(ap, MD4_SUM_LENGTH);
796                 if (!file->sum)
797                         out_of_memory("md4 sum");
798                 /* drat. we have to provide a null checksum for non-regular
799                    files in order to be compatible with earlier versions
800                    of rsync */
801                 if (S_ISREG(st.st_mode)) {
802                         file_checksum(fname, file->sum, st.st_size);
803                 } else {
804                         memset(file->sum, 0, MD4_SUM_LENGTH);
805                 }
806         }
807
808         if (flist_dir) {
809                 static char *lastdir;
810                 if (lastdir && strcmp(lastdir, flist_dir) == 0)
811                         file->basedir = lastdir;
812                 else {
813                         file->basedir = strdup(flist_dir);
814                         lastdir = file->basedir;
815                 }
816         } else
817                 file->basedir = NULL;
818
819         if (!S_ISDIR(st.st_mode))
820                 stats.total_size += st.st_size;
821
822         return file;
823 }
824
825
826 void send_file_name(int f, struct file_list *flist, char *fname,
827                     int recursive, unsigned short base_flags)
828 {
829         struct file_struct *file;
830         char fbuf[MAXPATHLEN];
831         extern int delete_excluded;
832
833         /* f is set to -1 when calculating deletion file list */
834         file = make_file(fname, &flist->string_area,
835                          f == -1 && delete_excluded? SERVER_EXCLUDES
836                                                    : ALL_EXCLUDES);
837
838         if (!file)
839                 return;
840
841         maybe_emit_filelist_progress(flist);
842
843         flist_expand(flist);
844
845         if (write_batch)
846                 file->flags |= FLAG_DELETE;
847
848         if (file->basename[0]) {
849                 flist->files[flist->count++] = file;
850                 send_file_entry(file, f, base_flags);
851         }
852
853         if (S_ISDIR(file->mode) && recursive) {
854                 struct exclude_struct **last_exclude_list =
855                     local_exclude_list;
856                 send_directory(f, flist, f_name_to(file, fbuf, sizeof fbuf));
857                 local_exclude_list = last_exclude_list;
858                 return;
859         }
860 }
861
862
863 static void send_directory(int f, struct file_list *flist, char *dir)
864 {
865         DIR *d;
866         struct dirent *di;
867         char fname[MAXPATHLEN];
868         int l;
869         char *p;
870
871         d = opendir(dir);
872         if (!d) {
873                 io_error |= IOERR_GENERAL;
874                 rprintf(FERROR, "opendir %s failed: %s\n",
875                         full_fname(dir), strerror(errno));
876                 return;
877         }
878
879         strlcpy(fname, dir, MAXPATHLEN);
880         l = strlen(fname);
881         if (fname[l - 1] != '/') {
882                 if (l == MAXPATHLEN - 1) {
883                         io_error |= IOERR_GENERAL;
884                         rprintf(FERROR, "skipping long-named directory: %s\n",
885                                 full_fname(fname));
886                         closedir(d);
887                         return;
888                 }
889                 strlcat(fname, "/", MAXPATHLEN);
890                 l++;
891         }
892         p = fname + strlen(fname);
893
894         local_exclude_list = NULL;
895
896         if (cvs_exclude) {
897                 if (strlen(fname) + strlen(".cvsignore") <= MAXPATHLEN - 1) {
898                         strcpy(p, ".cvsignore");
899                         add_exclude_file(&exclude_list,fname,MISSING_OK,ADD_EXCLUDE);
900                 } else {
901                         io_error |= IOERR_GENERAL;
902                         rprintf(FINFO,
903                                 "cannot cvs-exclude in long-named directory %s\n",
904                                 full_fname(fname));
905                 }
906         }
907
908         for (errno = 0, di = readdir(d); di; errno = 0, di = readdir(d)) {
909                 char *dname = d_name(di);
910                 if (dname[0] == '.' && (dname[1] == '\0'
911                     || (dname[1] == '.' && dname[2] == '\0')))
912                         continue;
913                 strlcpy(p, dname, MAXPATHLEN - l);
914                 send_file_name(f, flist, fname, recurse, 0);
915         }
916         if (errno) {
917                 io_error |= IOERR_GENERAL;
918                 rprintf(FERROR, "readdir(%s): (%d) %s\n",
919                     dir, errno, strerror(errno));
920         }
921
922         if (local_exclude_list)
923                 free_exclude_list(&local_exclude_list); /* Zeros pointer too */
924
925         closedir(d);
926 }
927
928
929 /**
930  * The delete_files() function in receiver.c sets f to -1 so that we just
931  * construct the file list in memory without sending it over the wire.  It
932  * also has the side-effect of ignoring user-excludes if delete_excluded
933  * is set (so that the delete list includes user-excluded files).
934  **/
935 struct file_list *send_file_list(int f, int argc, char *argv[])
936 {
937         int l;
938         STRUCT_STAT st;
939         char *p, *dir, *olddir;
940         char lastpath[MAXPATHLEN] = "";
941         struct file_list *flist;
942         int64 start_write;
943         int use_ff_fd = 0;
944
945         if (show_filelist_p() && f != -1)
946                 start_filelist_progress("building file list");
947
948         start_write = stats.total_written;
949
950         flist = flist_new();
951
952         if (f != -1) {
953                 io_start_buffering_out(f);
954                 if (filesfrom_fd >= 0) {
955                         if (argv[0] && !push_dir(argv[0], 0)) {
956                                 rprintf(FERROR, "push_dir %s failed: %s\n",
957                                         full_fname(argv[0]), strerror(errno));
958                                 exit_cleanup(RERR_FILESELECT);
959                         }
960                         use_ff_fd = 1;
961                 }
962         }
963
964         while (1) {
965                 char fname2[MAXPATHLEN];
966                 char *fname = fname2;
967
968                 if (use_ff_fd) {
969                         if (read_filesfrom_line(filesfrom_fd, fname) == 0)
970                                 break;
971                         sanitize_path(fname, NULL);
972                 } else {
973                         if (argc-- == 0)
974                                 break;
975                         strlcpy(fname, *argv++, MAXPATHLEN);
976                         if (sanitize_paths)
977                                 sanitize_path(fname, NULL);
978                 }
979
980                 l = strlen(fname);
981                 if (fname[l - 1] == '/') {
982                         if (l == 2 && fname[0] == '.') {
983                                 /* Turn "./" into just "." rather than "./." */
984                                 fname[1] = '\0';
985                         } else {
986                                 strlcat(fname, ".", MAXPATHLEN);
987                         }
988                 }
989
990                 if (link_stat(fname, &st) != 0) {
991                         if (f != -1) {
992                                 io_error |= IOERR_GENERAL;
993                                 rprintf(FERROR, "link_stat %s failed: %s\n",
994                                         full_fname(fname), strerror(errno));
995                         }
996                         continue;
997                 }
998
999                 if (S_ISDIR(st.st_mode) && !recurse && !files_from) {
1000                         rprintf(FINFO, "skipping directory %s\n", fname);
1001                         continue;
1002                 }
1003
1004                 dir = NULL;
1005                 olddir = NULL;
1006
1007                 if (!relative_paths) {
1008                         p = strrchr(fname, '/');
1009                         if (p) {
1010                                 *p = 0;
1011                                 if (p == fname)
1012                                         dir = "/";
1013                                 else
1014                                         dir = fname;
1015                                 fname = p + 1;
1016                         }
1017                 } else if (f != -1 && implied_dirs && (p=strrchr(fname,'/')) && p != fname) {
1018                         /* this ensures we send the intermediate directories,
1019                            thus getting their permissions right */
1020                         char *lp = lastpath, *fn = fname, *slash = fname;
1021                         *p = 0;
1022                         /* Skip any initial directories in our path that we
1023                          * have in common with lastpath. */
1024                         while (*fn && *lp == *fn) {
1025                                 if (*fn == '/')
1026                                         slash = fn;
1027                                 lp++, fn++;
1028                         }
1029                         *p = '/';
1030                         if (fn != p || (*lp && *lp != '/')) {
1031                                 int copy_links_saved = copy_links;
1032                                 int recurse_saved = recurse;
1033                                 copy_links = copy_unsafe_links;
1034                                 /* set recurse to 1 to prevent make_file
1035                                  * from ignoring directory, but still
1036                                  * turn off the recursive parameter to
1037                                  * send_file_name */
1038                                 recurse = 1;
1039                                 while ((slash = strchr(slash+1, '/')) != 0) {
1040                                         *slash = 0;
1041                                         send_file_name(f, flist, fname, 0, 0);
1042                                         *slash = '/';
1043                                 }
1044                                 copy_links = copy_links_saved;
1045                                 recurse = recurse_saved;
1046                                 *p = 0;
1047                                 strlcpy(lastpath, fname, sizeof lastpath);
1048                                 *p = '/';
1049                         }
1050                 }
1051
1052                 if (!*fname)
1053                         fname = ".";
1054
1055                 if (dir && *dir) {
1056                         olddir = push_dir(dir, 1);
1057
1058                         if (!olddir) {
1059                                 io_error |= IOERR_GENERAL;
1060                                 rprintf(FERROR, "push_dir %s failed: %s\n",
1061                                         full_fname(dir), strerror(errno));
1062                                 continue;
1063                         }
1064
1065                         flist_dir = dir;
1066                 }
1067
1068                 if (one_file_system)
1069                         set_filesystem(fname);
1070
1071                 send_file_name(f, flist, fname, recurse, FLAG_DELETE);
1072
1073                 if (olddir != NULL) {
1074                         flist_dir = NULL;
1075                         if (pop_dir(olddir) != 0) {
1076                                 rprintf(FERROR, "pop_dir %s failed: %s\n",
1077                                         full_fname(dir), strerror(errno));
1078                                 exit_cleanup(RERR_FILESELECT);
1079                         }
1080                 }
1081         }
1082
1083         if (f != -1)
1084                 send_file_entry(NULL, f, 0);
1085
1086         if (show_filelist_p() && f != -1)
1087                 finish_filelist_progress(flist);
1088
1089         clean_flist(flist, 0, 0);
1090
1091         /* now send the uid/gid list. This was introduced in protocol
1092            version 15 */
1093         if (f != -1)
1094                 send_uid_list(f);
1095
1096         /* send the io_error flag */
1097         if (f != -1) {
1098                 extern int module_id;
1099                 write_int(f, lp_ignore_errors(module_id) ? 0 : io_error);
1100         }
1101
1102         if (f != -1) {
1103                 io_end_buffering();
1104                 stats.flist_size = stats.total_written - start_write;
1105                 stats.num_files = flist->count;
1106                 if (write_batch)
1107                         write_batch_flist_info(flist->count, flist->files);
1108         }
1109
1110         if (verbose > 2)
1111                 rprintf(FINFO, "send_file_list done\n");
1112
1113         return flist;
1114 }
1115
1116
1117 struct file_list *recv_file_list(int f)
1118 {
1119         struct file_list *flist;
1120         unsigned short flags;
1121         int64 start_read;
1122         extern int list_only;
1123
1124         if (show_filelist_p())
1125                 start_filelist_progress("receiving file list");
1126
1127         start_read = stats.total_read;
1128
1129         flist = new(struct file_list);
1130         if (!flist)
1131                 goto oom;
1132
1133         flist->count = 0;
1134         flist->malloced = 1000;
1135         flist->files = new_array(struct file_struct *, flist->malloced);
1136         if (!flist->files)
1137                 goto oom;
1138
1139
1140         while ((flags = read_byte(f)) != 0) {
1141                 int i = flist->count;
1142
1143                 flist_expand(flist);
1144
1145                 if (protocol_version >= 28 && (flags & EXTENDED_FLAGS))
1146                         flags |= read_byte(f) << 8;
1147                 receive_file_entry(&flist->files[i], flags, f);
1148
1149                 if (S_ISREG(flist->files[i]->mode))
1150                         stats.total_size += flist->files[i]->length;
1151
1152                 flist->count++;
1153
1154                 maybe_emit_filelist_progress(flist);
1155
1156                 if (verbose > 2) {
1157                         rprintf(FINFO, "recv_file_name(%s)\n",
1158                                 f_name(flist->files[i]));
1159                 }
1160         }
1161
1162
1163         if (verbose > 2)
1164                 rprintf(FINFO, "received %d names\n", flist->count);
1165
1166         clean_flist(flist, relative_paths, 1);
1167
1168         if (show_filelist_p())
1169                 finish_filelist_progress(flist);
1170
1171         /* now recv the uid/gid list. This was introduced in protocol version 15 */
1172         if (f != -1)
1173                 recv_uid_list(f, flist);
1174
1175         /* recv the io_error flag */
1176         if (f != -1 && !read_batch) {   /* dw-added readbatch */
1177                 extern int module_id;
1178                 extern int ignore_errors;
1179                 if (lp_ignore_errors(module_id) || ignore_errors)
1180                         read_int(f);
1181                 else
1182                         io_error |= read_int(f);
1183         }
1184
1185         if (list_only) {
1186                 int i;
1187                 for (i = 0; i < flist->count; i++)
1188                         list_file_entry(flist->files[i]);
1189         }
1190
1191
1192         if (verbose > 2)
1193                 rprintf(FINFO, "recv_file_list done\n");
1194
1195         stats.flist_size = stats.total_read - start_read;
1196         stats.num_files = flist->count;
1197
1198         return flist;
1199
1200       oom:
1201         out_of_memory("recv_file_list");
1202         return NULL;            /* not reached */
1203 }
1204
1205
1206 int file_compare(struct file_struct **file1, struct file_struct **file2)
1207 {
1208         struct file_struct *f1 = *file1;
1209         struct file_struct *f2 = *file2;
1210
1211         if (!f1->basename && !f2->basename)
1212                 return 0;
1213         if (!f1->basename)
1214                 return -1;
1215         if (!f2->basename)
1216                 return 1;
1217         if (f1->dirname == f2->dirname)
1218                 return u_strcmp(f1->basename, f2->basename);
1219         return f_name_cmp(f1, f2);
1220 }
1221
1222
1223 int flist_find(struct file_list *flist, struct file_struct *f)
1224 {
1225         int low = 0, high = flist->count - 1;
1226
1227         while (high >= 0 && !flist->files[high]->basename) high--;
1228
1229         if (high < 0)
1230                 return -1;
1231
1232         while (low != high) {
1233                 int mid = (low + high) / 2;
1234                 int ret = file_compare(&flist->files[flist_up(flist, mid)],&f);
1235                 if (ret == 0)
1236                         return flist_up(flist, mid);
1237                 if (ret > 0)
1238                         high = mid;
1239                 else
1240                         low = mid + 1;
1241         }
1242
1243         if (file_compare(&flist->files[flist_up(flist, low)], &f) == 0)
1244                 return flist_up(flist, low);
1245         return -1;
1246 }
1247
1248
1249 /*
1250  * free up one file
1251  */
1252 void free_file(struct file_struct *file)
1253 {
1254         if (!file)
1255                 return;
1256         if (file->basename)
1257                 free(file->basename);
1258         if (file->link)
1259                 free(file->link);
1260         if (file->sum)
1261                 free(file->sum);
1262         *file = null_file;
1263 }
1264
1265
1266 /*
1267  * allocate a new file list
1268  */
1269 struct file_list *flist_new(void)
1270 {
1271         struct file_list *flist;
1272
1273         flist = new(struct file_list);
1274         if (!flist)
1275                 out_of_memory("send_file_list");
1276
1277         flist->count = 0;
1278         flist->malloced = 0;
1279         flist->files = NULL;
1280
1281 #if ARENA_SIZE > 0
1282         flist->string_area = string_area_new(0);
1283 #else
1284         flist->string_area = NULL;
1285 #endif
1286         return flist;
1287 }
1288
1289 /*
1290  * free up all elements in a flist
1291  */
1292 void flist_free(struct file_list *flist)
1293 {
1294         int i;
1295         for (i = 1; i < flist->count; i++) {
1296                 if (!flist->string_area)
1297                         free_file(flist->files[i]);
1298                 free(flist->files[i]);
1299         }
1300         /* FIXME: I don't think we generally need to blank the flist
1301          * since it's about to be freed.  This will just cause more
1302          * memory traffic.  If you want a freed-memory debugger, you
1303          * know where to get it. */
1304         memset((char *) flist->files, 0,
1305                sizeof(flist->files[0]) * flist->count);
1306         free(flist->files);
1307         if (flist->string_area)
1308                 string_area_free(flist->string_area);
1309         memset((char *) flist, 0, sizeof(*flist));
1310         free(flist);
1311 }
1312
1313
1314 /*
1315  * This routine ensures we don't have any duplicate names in our file list.
1316  * duplicate names can cause corruption because of the pipelining
1317  */
1318 static void clean_flist(struct file_list *flist, int strip_root, int no_dups)
1319 {
1320         int i, prev_i = 0;
1321
1322         if (!flist || flist->count == 0)
1323                 return;
1324
1325         qsort(flist->files, flist->count,
1326               sizeof(flist->files[0]), (int (*)()) file_compare);
1327
1328         for (i = no_dups? 0 : flist->count; i < flist->count; i++) {
1329                 if (flist->files[i]->basename) {
1330                         prev_i = i;
1331                         break;
1332                 }
1333         }
1334         while (++i < flist->count) {
1335                 if (!flist->files[i]->basename)
1336                         continue;
1337                 if (f_name_cmp(flist->files[i], flist->files[prev_i]) == 0) {
1338                         if (verbose > 1 && !am_server) {
1339                                 rprintf(FINFO,
1340                                         "removing duplicate name %s from file list %d\n",
1341                                         f_name(flist->files[i]), i);
1342                         }
1343                         /* Make sure that if we unduplicate '.', that we don't
1344                          * lose track of a user-specified starting point (or
1345                          * else deletions will mysteriously fail with -R). */
1346                         if (flist->files[i]->flags & FLAG_DELETE)
1347                                 flist->files[prev_i]->flags |= FLAG_DELETE;
1348                         /* it's not great that the flist knows the semantics of
1349                          * the file memory usage, but i'd rather not add a flag
1350                          * byte to that struct.
1351                          * XXX can i use a bit in the flags field? */
1352                         if (flist->string_area)
1353                                 flist->files[i][0] = null_file;
1354                         else
1355                                 free_file(flist->files[i]);
1356                 }
1357                 else
1358                         prev_i = i;
1359         }
1360
1361         if (strip_root) {
1362                 /* we need to strip off the root directory in the case
1363                    of relative paths, but this must be done _after_
1364                    the sorting phase */
1365                 for (i = 0; i < flist->count; i++) {
1366                         if (flist->files[i]->dirname &&
1367                             flist->files[i]->dirname[0] == '/') {
1368                                 memmove(&flist->files[i]->dirname[0],
1369                                         &flist->files[i]->dirname[1],
1370                                         strlen(flist->files[i]->dirname));
1371                         }
1372
1373                         if (flist->files[i]->dirname &&
1374                             !flist->files[i]->dirname[0]) {
1375                                 flist->files[i]->dirname = NULL;
1376                         }
1377                 }
1378         }
1379
1380         if (verbose <= 3)
1381                 return;
1382
1383         for (i = 0; i < flist->count; i++) {
1384                 rprintf(FINFO, "[%ld] i=%d %s %s mode=0%o len=%.0f\n",
1385                         (long) getpid(), i,
1386                         NS(flist->files[i]->dirname),
1387                         NS(flist->files[i]->basename),
1388                         (int) flist->files[i]->mode,
1389                         (double) flist->files[i]->length);
1390         }
1391 }
1392
1393
1394 enum fnc_state { fnc_DIR, fnc_SLASH, fnc_BASE };
1395
1396 /* Compare the names of two file_struct entities, just like strcmp()
1397  * would do if it were operating on the joined strings.  We assume
1398  * that there are no 0-length strings.
1399  */
1400 int f_name_cmp(struct file_struct *f1, struct file_struct *f2)
1401 {
1402         int dif;
1403         const uchar *c1, *c2;
1404         enum fnc_state state1, state2;
1405
1406         if (!f1 || !f1->basename) {
1407                 if (!f2 || !f2->basename)
1408                         return 0;
1409                 return -1;
1410         }
1411         if (!f2 || !f2->basename)
1412                 return 1;
1413
1414         if (!(c1 = (uchar*)f1->dirname)) {
1415                 state1 = fnc_BASE;
1416                 c1 = (uchar*)f1->basename;
1417         }
1418         else
1419                 state1 = fnc_DIR;
1420         if (!(c2 = (uchar*)f2->dirname)) {
1421                 state2 = fnc_BASE;
1422                 c2 = (uchar*)f2->basename;
1423         }
1424         else
1425                 state2 = fnc_DIR;
1426
1427         while (1) {
1428                 if ((dif = (int)*c1 - (int)*c2) != 0)
1429                         break;
1430                 if (!*++c1) {
1431                         switch (state1) {
1432                         case fnc_DIR:
1433                                 state1 = fnc_SLASH;
1434                                 c1 = (uchar*)"/";
1435                                 break;
1436                         case fnc_SLASH:
1437                                 state1 = fnc_BASE;
1438                                 c1 = (uchar*)f1->basename;
1439                                 break;
1440                         case fnc_BASE:
1441                                 break;
1442                         }
1443                 }
1444                 if (!*++c2) {
1445                         switch (state2) {
1446                         case fnc_DIR:
1447                                 state2 = fnc_SLASH;
1448                                 c2 = (uchar*)"/";
1449                                 break;
1450                         case fnc_SLASH:
1451                                 state2 = fnc_BASE;
1452                                 c2 = (uchar*)f2->basename;
1453                                 break;
1454                         case fnc_BASE:
1455                                 if (!*c1)
1456                                         return 0;
1457                                 break;
1458                         }
1459                 }
1460         }
1461
1462         return dif;
1463 }
1464
1465
1466 /* Return a copy of the full filename of a flist entry, using the indicated
1467  * buffer.
1468  */
1469 char *f_name_to(struct file_struct *f, char *fbuf, int bsize)
1470 {
1471         if (!f || !f->basename)
1472                 return NULL;
1473
1474         if (f->dirname) {
1475                 int off = strlcpy(fbuf, f->dirname, bsize);
1476                 off += strlcpy(fbuf + off, "/", bsize - off);
1477                 strlcpy(fbuf + off, f->basename, bsize - off);
1478         } else
1479                 strlcpy(fbuf, f->basename, bsize);
1480         return fbuf;
1481 }
1482
1483
1484 /* Like f_name_to(), but we rotate through 5 static buffers of our own.
1485  */
1486 char *f_name(struct file_struct *f)
1487 {
1488         static char names[5][MAXPATHLEN];
1489         static unsigned int n;
1490
1491         n = (n + 1) % (sizeof names / sizeof names[0]);
1492
1493         return f_name_to(f, names[n], sizeof names[0]);
1494 }