9b7dcf5e351b6d2c3590d4f6e70171702af48a24
[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 modtime;
359         static mode_t mode;
360         static DEV64_T last_rdev;
361         static uid_t uid;
362         static gid_t gid;
363         static DEV64_T dev;
364         static char lastname[MAXPATHLEN];
365         char *fname, fbuf[MAXPATHLEN];
366         int l1, l2;
367
368         if (f == -1)
369                 return;
370
371         if (!file) {
372                 write_byte(f, 0);
373                 return;
374         }
375
376         io_write_phase = "send_file_entry";
377
378         fname = f_name_to(file, fbuf, sizeof fbuf);
379
380         flags = base_flags;
381
382         if (file->mode == mode)
383                 flags |= SAME_MODE;
384         else
385                 mode = file->mode;
386         if (preserve_devices) {
387                 if (protocol_version < 28) {
388                         if (IS_DEVICE(mode) && file->rdev == last_rdev) {
389                                 /* Set both flags so that the test when
390                                  * writing the data is simpler. */
391                                 flags |= SAME_RDEV_pre28|SAME_HIGH_RDEV;
392                         }
393                         else
394                                 last_rdev = file->rdev;
395                 }
396                 else if (IS_DEVICE(mode)) {
397                         if ((file->rdev & ~0xFF) == last_rdev)
398                                 flags |= SAME_HIGH_RDEV;
399                         else
400                                 last_rdev = file->rdev & ~0xFF;
401                 }
402         }
403         if (file->uid == uid)
404                 flags |= SAME_UID;
405         else
406                 uid = file->uid;
407         if (file->gid == gid)
408                 flags |= SAME_GID;
409         else
410                 gid = file->gid;
411         if (file->modtime == modtime)
412                 flags |= SAME_TIME;
413         else
414                 modtime = file->modtime;
415         if (file->flags & HAS_INODE_DATA) {
416                 if (file->dev == dev) {
417                         if (protocol_version >= 28)
418                                 flags |= SAME_DEV;
419                 }
420                 else
421                         dev = file->dev;
422                 flags |= HAS_INODE_DATA;
423         }
424
425         for (l1 = 0;
426              lastname[l1] && (fname[l1] == lastname[l1]) && (l1 < 255);
427              l1++) {}
428         l2 = strlen(fname) - l1;
429
430         if (l1 > 0)
431                 flags |= SAME_NAME;
432         if (l2 > 255)
433                 flags |= LONG_NAME;
434
435         /* We must make sure we don't send a zero flags byte or
436          * the other end will terminate the flist transfer. */
437         if (flags == 0 && !S_ISDIR(mode))
438                 flags |= FLAG_DELETE; /* NOTE: no meaning for non-dir */
439         if (protocol_version >= 28) {
440                 if ((flags & 0xFF00) || flags == 0) {
441                         flags |= EXTENDED_FLAGS;
442                         write_byte(f, flags);
443                         write_byte(f, flags >> 8);
444                 } else
445                         write_byte(f, flags);
446         } else {
447                 if (flags == 0)
448                         flags |= LONG_NAME;
449                 write_byte(f, flags);
450         }
451         if (flags & SAME_NAME)
452                 write_byte(f, l1);
453         if (flags & LONG_NAME)
454                 write_int(f, l2);
455         else
456                 write_byte(f, l2);
457         write_buf(f, fname + l1, l2);
458
459         write_longint(f, file->length);
460         if (!(flags & SAME_TIME))
461                 write_int(f, modtime);
462         if (!(flags & SAME_MODE))
463                 write_int(f, to_wire_mode(mode));
464         if (preserve_uid && !(flags & SAME_UID)) {
465                 add_uid(uid);
466                 write_int(f, uid);
467         }
468         if (preserve_gid && !(flags & SAME_GID)) {
469                 add_gid(gid);
470                 write_int(f, gid);
471         }
472         if (preserve_devices && IS_DEVICE(mode)) {
473                 /* If SAME_HIGH_RDEV is off, SAME_RDEV_pre28 is also off. */
474                 if (!(flags & SAME_HIGH_RDEV))
475                         write_int(f, file->rdev);
476                 else if (protocol_version >= 28)
477                         write_byte(f, file->rdev);
478         }
479
480 #if SUPPORT_LINKS
481         if (preserve_links && S_ISLNK(mode)) {
482                 write_int(f, strlen(file->link));
483                 write_buf(f, file->link, strlen(file->link));
484         }
485 #endif
486
487 #if SUPPORT_HARD_LINKS
488         if (flags & HAS_INODE_DATA) {
489                 if (protocol_version < 26) {
490                         /* 32-bit dev_t and ino_t */
491                         write_int(f, dev);
492                         write_int(f, file->inode);
493                 } else {
494                         /* 64-bit dev_t and ino_t */
495                         if (!(flags & SAME_DEV))
496                                 write_longint(f, dev);
497                         write_longint(f, file->inode);
498                 }
499         }
500 #endif
501
502         if (always_checksum) {
503                 if (protocol_version < 21)
504                         write_buf(f, file->sum, 2);
505                 else
506                         write_buf(f, file->sum, MD4_SUM_LENGTH);
507         }
508
509         strlcpy(lastname, fname, MAXPATHLEN);
510         lastname[MAXPATHLEN - 1] = 0;
511
512         io_write_phase = "unknown";
513 }
514
515
516
517 static void receive_file_entry(struct file_struct **fptr,
518                                unsigned short flags, int f)
519 {
520         static time_t modtime;
521         static mode_t mode;
522         static DEV64_T rdev;
523         static uid_t uid;
524         static gid_t gid;
525         static DEV64_T dev;
526         static char lastname[MAXPATHLEN];
527         char thisname[MAXPATHLEN];
528         unsigned int l1 = 0, l2 = 0;
529         char *p;
530         struct file_struct *file;
531
532         if (flags & SAME_NAME)
533                 l1 = read_byte(f);
534
535         if (flags & LONG_NAME)
536                 l2 = read_int(f);
537         else
538                 l2 = read_byte(f);
539
540         file = new(struct file_struct);
541         if (!file)
542                 out_of_memory("receive_file_entry");
543         memset((char *) file, 0, sizeof(*file));
544         (*fptr) = file;
545
546         if (l2 >= MAXPATHLEN - l1) {
547                 rprintf(FERROR,
548                         "overflow: flags=0x%x l1=%d l2=%d lastname=%s\n",
549                         flags, l1, l2, lastname);
550                 overflow("receive_file_entry");
551         }
552
553         strlcpy(thisname, lastname, l1 + 1);
554         read_sbuf(f, &thisname[l1], l2);
555         thisname[l1 + l2] = 0;
556
557         strlcpy(lastname, thisname, MAXPATHLEN);
558         lastname[MAXPATHLEN - 1] = 0;
559
560         clean_fname(thisname);
561
562         if (sanitize_paths) {
563                 sanitize_path(thisname, NULL);
564         }
565
566         if ((p = strrchr(thisname, '/'))) {
567                 static char *lastdir;
568                 *p = 0;
569                 if (lastdir && strcmp(thisname, lastdir) == 0)
570                         file->dirname = lastdir;
571                 else {
572                         file->dirname = strdup(thisname);
573                         lastdir = file->dirname;
574                 }
575                 file->basename = strdup(p + 1);
576         } else {
577                 file->dirname = NULL;
578                 file->basename = strdup(thisname);
579         }
580
581         if (!file->basename)
582                 out_of_memory("receive_file_entry 1");
583
584         file->flags = flags;
585         file->length = read_longint(f);
586         if (!(flags & SAME_TIME))
587                 modtime = (time_t)read_int(f);
588         file->modtime = modtime;
589         if (!(flags & SAME_MODE))
590                 mode = from_wire_mode(read_int(f));
591         file->mode = mode;
592
593         if (preserve_uid) {
594                 if (!(flags & SAME_UID))
595                         uid = (uid_t)read_int(f);
596                 file->uid = uid;
597         }
598         if (preserve_gid) {
599                 if (!(flags & SAME_GID))
600                         gid = (gid_t)read_int(f);
601                 file->gid = gid;
602         }
603         if (preserve_devices) {
604                 if (protocol_version < 28) {
605                         if (IS_DEVICE(mode)) {
606                                 if (!(flags & SAME_RDEV_pre28))
607                                         rdev = (DEV64_T)read_int(f);
608                                 file->rdev = rdev;
609                         } else
610                                 rdev = 0;
611                 } else if (IS_DEVICE(mode)) {
612                         if (!(flags & SAME_HIGH_RDEV)) {
613                                 file->rdev = (DEV64_T)read_int(f);
614                                 rdev = file->rdev & ~0xFF;
615                         } else
616                                 file->rdev = (DEV64_T)(rdev | read_byte(f));
617                 }
618         }
619
620         if (preserve_links && S_ISLNK(mode)) {
621                 int l = read_int(f);
622                 if (l < 0) {
623                         rprintf(FERROR, "overflow: l=%d\n", l);
624                         overflow("receive_file_entry");
625                 }
626                 file->link = new_array(char, l + 1);
627                 if (!file->link)
628                         out_of_memory("receive_file_entry 2");
629                 read_sbuf(f, file->link, l);
630                 if (sanitize_paths)
631                         sanitize_path(file->link, file->dirname);
632         }
633 #if SUPPORT_HARD_LINKS
634         if (preserve_hard_links && protocol_version < 28
635             && S_ISREG(mode))
636                 file->flags |= HAS_INODE_DATA;
637         if (file->flags & HAS_INODE_DATA) {
638                 if (protocol_version < 26) {
639                         dev = read_int(f);
640                         file->inode = read_int(f);
641                 } else {
642                         if (!(flags & SAME_DEV))
643                                 dev = read_longint(f);
644                         file->inode = read_longint(f);
645                 }
646                 file->dev = dev;
647         }
648 #endif
649
650         if (always_checksum) {
651                 file->sum = new_array(char, MD4_SUM_LENGTH);
652                 if (!file->sum)
653                         out_of_memory("md4 sum");
654                 if (protocol_version < 21)
655                         read_buf(f, file->sum, 2);
656                 else
657                         read_buf(f, file->sum, MD4_SUM_LENGTH);
658         }
659
660         if (!preserve_perms) {
661                 extern int orig_umask;
662                 /* set an appropriate set of permissions based on original
663                    permissions and umask. This emulates what GNU cp does */
664                 file->mode &= ~orig_umask;
665         }
666 }
667
668
669 /* determine if a file in a different filesstem should be skipped
670    when one_file_system is set. We bascally only want to include
671    the mount points - but they can be hard to find! */
672 static int skip_filesystem(char *fname, STRUCT_STAT * st)
673 {
674         STRUCT_STAT st2;
675         char *p = strrchr(fname, '/');
676
677         /* skip all but directories */
678         if (!S_ISDIR(st->st_mode))
679                 return 1;
680
681         /* if its not a subdirectory then allow */
682         if (!p)
683                 return 0;
684
685         *p = 0;
686         if (link_stat(fname, &st2)) {
687                 *p = '/';
688                 return 0;
689         }
690         *p = '/';
691
692         return (st2.st_dev != filesystem_dev);
693 }
694
695 #define STRDUP(ap, p)   (ap ? string_area_strdup(ap, p) : strdup(p))
696 /* IRIX cc cares that the operands to the ternary have the same type. */
697 #define MALLOC(ap, i)   (ap ? (void*) string_area_malloc(ap, i) : malloc(i))
698
699 /**
700  * Create a file_struct for a named file by reading its stat()
701  * information and performing extensive checks against global
702  * options.
703  *
704  * @return the new file, or NULL if there was an error or this file
705  * should be excluded.
706  *
707  * @todo There is a small optimization opportunity here to avoid
708  * stat()ing the file in some circumstances, which has a certain cost.
709  * We are called immediately after doing readdir(), and so we may
710  * already know the d_type of the file.  We could for example avoid
711  * statting directories if we're not recursing, but this is not a very
712  * important case.  Some systems may not have d_type.
713  **/
714 struct file_struct *make_file(char *fname, struct string_area **ap,
715                               int exclude_level)
716 {
717         struct file_struct *file;
718         STRUCT_STAT st;
719         char sum[SUM_LENGTH];
720         char *p;
721         char cleaned_name[MAXPATHLEN];
722         char linkbuf[MAXPATHLEN];
723         extern int module_id;
724
725         strlcpy(cleaned_name, fname, MAXPATHLEN);
726         cleaned_name[MAXPATHLEN - 1] = 0;
727         clean_fname(cleaned_name);
728         if (sanitize_paths)
729                 sanitize_path(cleaned_name, NULL);
730         fname = cleaned_name;
731
732         memset(sum, 0, SUM_LENGTH);
733
734         if (readlink_stat(fname, &st, linkbuf) != 0) {
735                 int save_errno = errno;
736                 if (errno == ENOENT && exclude_level != NO_EXCLUDES) {
737                         /* either symlink pointing nowhere or file that
738                          * was removed during rsync run; see if excluded
739                          * before reporting an error */
740                         if (check_exclude_file(fname, 0, exclude_level)) {
741                                 /* file is excluded anyway, ignore silently */
742                                 return NULL;
743                         }
744                 }
745                 io_error |= IOERR_GENERAL;
746                 rprintf(FERROR, "readlink %s failed: %s\n",
747                         full_fname(fname), strerror(save_errno));
748                 return NULL;
749         }
750
751         /* backup.c calls us with exclude_level set to NO_EXCLUDES. */
752         if (exclude_level == NO_EXCLUDES)
753                 goto skip_excludes;
754
755         if (S_ISDIR(st.st_mode) && !recurse && !files_from) {
756                 rprintf(FINFO, "skipping directory %s\n", fname);
757                 return NULL;
758         }
759
760         if (one_file_system && st.st_dev != filesystem_dev) {
761                 if (skip_filesystem(fname, &st))
762                         return NULL;
763         }
764
765         if (check_exclude_file(fname, S_ISDIR(st.st_mode) != 0, exclude_level))
766                 return NULL;
767
768         if (lp_ignore_nonreadable(module_id) && access(fname, R_OK) != 0)
769                 return NULL;
770
771       skip_excludes:
772
773         if (verbose > 2)
774                 rprintf(FINFO, "make_file(%s,*,%d)\n", fname, exclude_level);
775
776         file = new(struct file_struct);
777         if (!file)
778                 out_of_memory("make_file");
779         memset((char *) file, 0, sizeof(*file));
780
781         if ((p = strrchr(fname, '/'))) {
782                 static char *lastdir;
783                 *p = 0;
784                 if (lastdir && strcmp(fname, lastdir) == 0)
785                         file->dirname = lastdir;
786                 else {
787                         file->dirname = strdup(fname);
788                         lastdir = file->dirname;
789                 }
790                 file->basename = STRDUP(ap, p + 1);
791                 *p = '/';
792         } else {
793                 file->dirname = NULL;
794                 file->basename = STRDUP(ap, fname);
795         }
796
797         file->modtime = st.st_mtime;
798         file->length = st.st_size;
799         file->mode = st.st_mode;
800         file->uid = st.st_uid;
801         file->gid = st.st_gid;
802         if (preserve_hard_links) {
803                 if (protocol_version < 28? S_ISREG(st.st_mode)
804                     : !S_ISDIR(st.st_mode) && st.st_nlink > 1) {
805                         file->dev = st.st_dev;
806                         file->inode = st.st_ino;
807                         file->flags |= HAS_INODE_DATA;
808                 }
809         }
810 #ifdef HAVE_STRUCT_STAT_ST_RDEV
811         file->rdev = st.st_rdev;
812 #endif
813
814 #if SUPPORT_LINKS
815         if (S_ISLNK(st.st_mode))
816                 file->link = STRDUP(ap, linkbuf);
817 #endif
818
819         if (always_checksum) {
820                 file->sum = (char *) MALLOC(ap, MD4_SUM_LENGTH);
821                 if (!file->sum)
822                         out_of_memory("md4 sum");
823                 /* drat. we have to provide a null checksum for non-regular
824                    files in order to be compatible with earlier versions
825                    of rsync */
826                 if (S_ISREG(st.st_mode)) {
827                         file_checksum(fname, file->sum, st.st_size);
828                 } else {
829                         memset(file->sum, 0, MD4_SUM_LENGTH);
830                 }
831         }
832
833         if (flist_dir) {
834                 static char *lastdir;
835                 if (lastdir && strcmp(lastdir, flist_dir) == 0)
836                         file->basedir = lastdir;
837                 else {
838                         file->basedir = strdup(flist_dir);
839                         lastdir = file->basedir;
840                 }
841         } else
842                 file->basedir = NULL;
843
844         if (!S_ISDIR(st.st_mode))
845                 stats.total_size += st.st_size;
846
847         return file;
848 }
849
850
851 void send_file_name(int f, struct file_list *flist, char *fname,
852                     int recursive, unsigned short base_flags)
853 {
854         struct file_struct *file;
855         char fbuf[MAXPATHLEN];
856         extern int delete_excluded;
857
858         /* f is set to -1 when calculating deletion file list */
859         file = make_file(fname, &flist->string_area,
860                          f == -1 && delete_excluded? SERVER_EXCLUDES
861                                                    : ALL_EXCLUDES);
862
863         if (!file)
864                 return;
865
866         maybe_emit_filelist_progress(flist);
867
868         flist_expand(flist);
869
870         if (write_batch)
871                 file->flags |= FLAG_DELETE;
872
873         if (file->basename[0]) {
874                 flist->files[flist->count++] = file;
875                 send_file_entry(file, f, base_flags);
876         }
877
878         if (S_ISDIR(file->mode) && recursive) {
879                 struct exclude_struct **last_exclude_list =
880                     local_exclude_list;
881                 send_directory(f, flist, f_name_to(file, fbuf, sizeof fbuf));
882                 local_exclude_list = last_exclude_list;
883                 return;
884         }
885 }
886
887
888 static void send_directory(int f, struct file_list *flist, char *dir)
889 {
890         DIR *d;
891         struct dirent *di;
892         char fname[MAXPATHLEN];
893         int l;
894         char *p;
895
896         d = opendir(dir);
897         if (!d) {
898                 io_error |= IOERR_GENERAL;
899                 rprintf(FERROR, "opendir %s failed: %s\n",
900                         full_fname(dir), strerror(errno));
901                 return;
902         }
903
904         strlcpy(fname, dir, MAXPATHLEN);
905         l = strlen(fname);
906         if (fname[l - 1] != '/') {
907                 if (l == MAXPATHLEN - 1) {
908                         io_error |= IOERR_GENERAL;
909                         rprintf(FERROR, "skipping long-named directory: %s\n",
910                                 full_fname(fname));
911                         closedir(d);
912                         return;
913                 }
914                 strlcat(fname, "/", MAXPATHLEN);
915                 l++;
916         }
917         p = fname + strlen(fname);
918
919         local_exclude_list = NULL;
920
921         if (cvs_exclude) {
922                 if (strlen(fname) + strlen(".cvsignore") <= MAXPATHLEN - 1) {
923                         strcpy(p, ".cvsignore");
924                         add_exclude_file(&exclude_list,fname,MISSING_OK,ADD_EXCLUDE);
925                 } else {
926                         io_error |= IOERR_GENERAL;
927                         rprintf(FINFO,
928                                 "cannot cvs-exclude in long-named directory %s\n",
929                                 full_fname(fname));
930                 }
931         }
932
933         for (errno = 0, di = readdir(d); di; errno = 0, di = readdir(d)) {
934                 char *dname = d_name(di);
935                 if (dname[0] == '.' && (dname[1] == '\0'
936                     || (dname[1] == '.' && dname[2] == '\0')))
937                         continue;
938                 strlcpy(p, dname, MAXPATHLEN - l);
939                 send_file_name(f, flist, fname, recurse, 0);
940         }
941         if (errno) {
942                 io_error |= IOERR_GENERAL;
943                 rprintf(FERROR, "readdir(%s): (%d) %s\n",
944                     dir, errno, strerror(errno));
945         }
946
947         if (local_exclude_list)
948                 free_exclude_list(&local_exclude_list); /* Zeros pointer too */
949
950         closedir(d);
951 }
952
953
954 /**
955  * The delete_files() function in receiver.c sets f to -1 so that we just
956  * construct the file list in memory without sending it over the wire.  It
957  * also has the side-effect of ignoring user-excludes if delete_excluded
958  * is set (so that the delete list includes user-excluded files).
959  **/
960 struct file_list *send_file_list(int f, int argc, char *argv[])
961 {
962         int l;
963         STRUCT_STAT st;
964         char *p, *dir, *olddir;
965         char lastpath[MAXPATHLEN] = "";
966         struct file_list *flist;
967         int64 start_write;
968         int use_ff_fd = 0;
969
970         if (show_filelist_p() && f != -1)
971                 start_filelist_progress("building file list");
972
973         start_write = stats.total_written;
974
975         flist = flist_new();
976
977         if (f != -1) {
978                 io_start_buffering_out(f);
979                 if (filesfrom_fd >= 0) {
980                         if (argv[0] && !push_dir(argv[0], 0)) {
981                                 rprintf(FERROR, "push_dir %s failed: %s\n",
982                                         full_fname(argv[0]), strerror(errno));
983                                 exit_cleanup(RERR_FILESELECT);
984                         }
985                         use_ff_fd = 1;
986                 }
987         }
988
989         while (1) {
990                 char fname2[MAXPATHLEN];
991                 char *fname = fname2;
992
993                 if (use_ff_fd) {
994                         if (read_filesfrom_line(filesfrom_fd, fname) == 0)
995                                 break;
996                         sanitize_path(fname, NULL);
997                 } else {
998                         if (argc-- == 0)
999                                 break;
1000                         strlcpy(fname, *argv++, MAXPATHLEN);
1001                         if (sanitize_paths)
1002                                 sanitize_path(fname, NULL);
1003                 }
1004
1005                 l = strlen(fname);
1006                 if (fname[l - 1] == '/') {
1007                         if (l == 2 && fname[0] == '.') {
1008                                 /* Turn "./" into just "." rather than "./." */
1009                                 fname[1] = '\0';
1010                         } else {
1011                                 strlcat(fname, ".", MAXPATHLEN);
1012                         }
1013                 }
1014
1015                 if (link_stat(fname, &st) != 0) {
1016                         if (f != -1) {
1017                                 io_error |= IOERR_GENERAL;
1018                                 rprintf(FERROR, "link_stat %s failed: %s\n",
1019                                         full_fname(fname), strerror(errno));
1020                         }
1021                         continue;
1022                 }
1023
1024                 if (S_ISDIR(st.st_mode) && !recurse && !files_from) {
1025                         rprintf(FINFO, "skipping directory %s\n", fname);
1026                         continue;
1027                 }
1028
1029                 dir = NULL;
1030                 olddir = NULL;
1031
1032                 if (!relative_paths) {
1033                         p = strrchr(fname, '/');
1034                         if (p) {
1035                                 *p = 0;
1036                                 if (p == fname)
1037                                         dir = "/";
1038                                 else
1039                                         dir = fname;
1040                                 fname = p + 1;
1041                         }
1042                 } else if (f != -1 && implied_dirs && (p=strrchr(fname,'/')) && p != fname) {
1043                         /* this ensures we send the intermediate directories,
1044                            thus getting their permissions right */
1045                         char *lp = lastpath, *fn = fname, *slash = fname;
1046                         *p = 0;
1047                         /* Skip any initial directories in our path that we
1048                          * have in common with lastpath. */
1049                         while (*fn && *lp == *fn) {
1050                                 if (*fn == '/')
1051                                         slash = fn;
1052                                 lp++, fn++;
1053                         }
1054                         *p = '/';
1055                         if (fn != p || (*lp && *lp != '/')) {
1056                                 int copy_links_saved = copy_links;
1057                                 int recurse_saved = recurse;
1058                                 copy_links = copy_unsafe_links;
1059                                 /* set recurse to 1 to prevent make_file
1060                                  * from ignoring directory, but still
1061                                  * turn off the recursive parameter to
1062                                  * send_file_name */
1063                                 recurse = 1;
1064                                 while ((slash = strchr(slash+1, '/')) != 0) {
1065                                         *slash = 0;
1066                                         send_file_name(f, flist, fname, 0, 0);
1067                                         *slash = '/';
1068                                 }
1069                                 copy_links = copy_links_saved;
1070                                 recurse = recurse_saved;
1071                                 *p = 0;
1072                                 strlcpy(lastpath, fname, sizeof lastpath);
1073                                 *p = '/';
1074                         }
1075                 }
1076
1077                 if (!*fname)
1078                         fname = ".";
1079
1080                 if (dir && *dir) {
1081                         olddir = push_dir(dir, 1);
1082
1083                         if (!olddir) {
1084                                 io_error |= IOERR_GENERAL;
1085                                 rprintf(FERROR, "push_dir %s failed: %s\n",
1086                                         full_fname(dir), strerror(errno));
1087                                 continue;
1088                         }
1089
1090                         flist_dir = dir;
1091                 }
1092
1093                 if (one_file_system)
1094                         set_filesystem(fname);
1095
1096                 send_file_name(f, flist, fname, recurse, FLAG_DELETE);
1097
1098                 if (olddir != NULL) {
1099                         flist_dir = NULL;
1100                         if (pop_dir(olddir) != 0) {
1101                                 rprintf(FERROR, "pop_dir %s failed: %s\n",
1102                                         full_fname(dir), strerror(errno));
1103                                 exit_cleanup(RERR_FILESELECT);
1104                         }
1105                 }
1106         }
1107
1108         if (f != -1)
1109                 send_file_entry(NULL, f, 0);
1110
1111         if (show_filelist_p() && f != -1)
1112                 finish_filelist_progress(flist);
1113
1114         clean_flist(flist, 0, 0);
1115
1116         /* now send the uid/gid list. This was introduced in protocol
1117            version 15 */
1118         if (f != -1)
1119                 send_uid_list(f);
1120
1121         /* send the io_error flag */
1122         if (f != -1) {
1123                 extern int module_id;
1124                 write_int(f, lp_ignore_errors(module_id) ? 0 : io_error);
1125         }
1126
1127         if (f != -1) {
1128                 io_end_buffering();
1129                 stats.flist_size = stats.total_written - start_write;
1130                 stats.num_files = flist->count;
1131                 if (write_batch)
1132                         write_batch_flist_info(flist->count, flist->files);
1133         }
1134
1135         if (verbose > 2)
1136                 rprintf(FINFO, "send_file_list done\n");
1137
1138         return flist;
1139 }
1140
1141
1142 struct file_list *recv_file_list(int f)
1143 {
1144         struct file_list *flist;
1145         unsigned short flags;
1146         int64 start_read;
1147         extern int list_only;
1148
1149         if (show_filelist_p())
1150                 start_filelist_progress("receiving file list");
1151
1152         start_read = stats.total_read;
1153
1154         flist = new(struct file_list);
1155         if (!flist)
1156                 goto oom;
1157
1158         flist->count = 0;
1159         flist->malloced = 1000;
1160         flist->files = new_array(struct file_struct *, flist->malloced);
1161         if (!flist->files)
1162                 goto oom;
1163
1164
1165         while ((flags = read_byte(f)) != 0) {
1166                 int i = flist->count;
1167
1168                 flist_expand(flist);
1169
1170                 if (protocol_version >= 28 && (flags & EXTENDED_FLAGS))
1171                         flags |= read_byte(f) << 8;
1172                 receive_file_entry(&flist->files[i], flags, f);
1173
1174                 if (S_ISREG(flist->files[i]->mode))
1175                         stats.total_size += flist->files[i]->length;
1176
1177                 flist->count++;
1178
1179                 maybe_emit_filelist_progress(flist);
1180
1181                 if (verbose > 2) {
1182                         rprintf(FINFO, "recv_file_name(%s)\n",
1183                                 f_name(flist->files[i]));
1184                 }
1185         }
1186
1187
1188         if (verbose > 2)
1189                 rprintf(FINFO, "received %d names\n", flist->count);
1190
1191         clean_flist(flist, relative_paths, 1);
1192
1193         if (show_filelist_p())
1194                 finish_filelist_progress(flist);
1195
1196         /* now recv the uid/gid list. This was introduced in protocol version 15 */
1197         if (f != -1)
1198                 recv_uid_list(f, flist);
1199
1200         /* recv the io_error flag */
1201         if (f != -1 && !read_batch) {   /* dw-added readbatch */
1202                 extern int module_id;
1203                 extern int ignore_errors;
1204                 if (lp_ignore_errors(module_id) || ignore_errors)
1205                         read_int(f);
1206                 else
1207                         io_error |= read_int(f);
1208         }
1209
1210         if (list_only) {
1211                 int i;
1212                 for (i = 0; i < flist->count; i++)
1213                         list_file_entry(flist->files[i]);
1214         }
1215
1216
1217         if (verbose > 2)
1218                 rprintf(FINFO, "recv_file_list done\n");
1219
1220         stats.flist_size = stats.total_read - start_read;
1221         stats.num_files = flist->count;
1222
1223         return flist;
1224
1225       oom:
1226         out_of_memory("recv_file_list");
1227         return NULL;            /* not reached */
1228 }
1229
1230
1231 int file_compare(struct file_struct **file1, struct file_struct **file2)
1232 {
1233         struct file_struct *f1 = *file1;
1234         struct file_struct *f2 = *file2;
1235
1236         if (!f1->basename && !f2->basename)
1237                 return 0;
1238         if (!f1->basename)
1239                 return -1;
1240         if (!f2->basename)
1241                 return 1;
1242         if (f1->dirname == f2->dirname)
1243                 return u_strcmp(f1->basename, f2->basename);
1244         return f_name_cmp(f1, f2);
1245 }
1246
1247
1248 int flist_find(struct file_list *flist, struct file_struct *f)
1249 {
1250         int low = 0, high = flist->count - 1;
1251
1252         while (high >= 0 && !flist->files[high]->basename) high--;
1253
1254         if (high < 0)
1255                 return -1;
1256
1257         while (low != high) {
1258                 int mid = (low + high) / 2;
1259                 int ret = file_compare(&flist->files[flist_up(flist, mid)],&f);
1260                 if (ret == 0)
1261                         return flist_up(flist, mid);
1262                 if (ret > 0)
1263                         high = mid;
1264                 else
1265                         low = mid + 1;
1266         }
1267
1268         if (file_compare(&flist->files[flist_up(flist, low)], &f) == 0)
1269                 return flist_up(flist, low);
1270         return -1;
1271 }
1272
1273
1274 /*
1275  * free up one file
1276  */
1277 void free_file(struct file_struct *file)
1278 {
1279         if (!file)
1280                 return;
1281         if (file->basename)
1282                 free(file->basename);
1283         if (file->link)
1284                 free(file->link);
1285         if (file->sum)
1286                 free(file->sum);
1287         *file = null_file;
1288 }
1289
1290
1291 /*
1292  * allocate a new file list
1293  */
1294 struct file_list *flist_new(void)
1295 {
1296         struct file_list *flist;
1297
1298         flist = new(struct file_list);
1299         if (!flist)
1300                 out_of_memory("send_file_list");
1301
1302         flist->count = 0;
1303         flist->malloced = 0;
1304         flist->files = NULL;
1305
1306 #if ARENA_SIZE > 0
1307         flist->string_area = string_area_new(0);
1308 #else
1309         flist->string_area = NULL;
1310 #endif
1311         return flist;
1312 }
1313
1314 /*
1315  * free up all elements in a flist
1316  */
1317 void flist_free(struct file_list *flist)
1318 {
1319         int i;
1320         for (i = 1; i < flist->count; i++) {
1321                 if (!flist->string_area)
1322                         free_file(flist->files[i]);
1323                 free(flist->files[i]);
1324         }
1325         /* FIXME: I don't think we generally need to blank the flist
1326          * since it's about to be freed.  This will just cause more
1327          * memory traffic.  If you want a freed-memory debugger, you
1328          * know where to get it. */
1329         memset((char *) flist->files, 0,
1330                sizeof(flist->files[0]) * flist->count);
1331         free(flist->files);
1332         if (flist->string_area)
1333                 string_area_free(flist->string_area);
1334         memset((char *) flist, 0, sizeof(*flist));
1335         free(flist);
1336 }
1337
1338
1339 /*
1340  * This routine ensures we don't have any duplicate names in our file list.
1341  * duplicate names can cause corruption because of the pipelining
1342  */
1343 static void clean_flist(struct file_list *flist, int strip_root, int no_dups)
1344 {
1345         int i, prev_i = 0;
1346
1347         if (!flist || flist->count == 0)
1348                 return;
1349
1350         qsort(flist->files, flist->count,
1351               sizeof(flist->files[0]), (int (*)()) file_compare);
1352
1353         for (i = no_dups? 0 : flist->count; i < flist->count; i++) {
1354                 if (flist->files[i]->basename) {
1355                         prev_i = i;
1356                         break;
1357                 }
1358         }
1359         while (++i < flist->count) {
1360                 if (!flist->files[i]->basename)
1361                         continue;
1362                 if (f_name_cmp(flist->files[i], flist->files[prev_i]) == 0) {
1363                         if (verbose > 1 && !am_server) {
1364                                 rprintf(FINFO,
1365                                         "removing duplicate name %s from file list %d\n",
1366                                         f_name(flist->files[i]), i);
1367                         }
1368                         /* Make sure that if we unduplicate '.', that we don't
1369                          * lose track of a user-specified starting point (or
1370                          * else deletions will mysteriously fail with -R). */
1371                         if (flist->files[i]->flags & FLAG_DELETE)
1372                                 flist->files[prev_i]->flags |= FLAG_DELETE;
1373                         /* it's not great that the flist knows the semantics of
1374                          * the file memory usage, but i'd rather not add a flag
1375                          * byte to that struct.
1376                          * XXX can i use a bit in the flags field? */
1377                         if (flist->string_area)
1378                                 flist->files[i][0] = null_file;
1379                         else
1380                                 free_file(flist->files[i]);
1381                 }
1382                 else
1383                         prev_i = i;
1384         }
1385
1386         if (strip_root) {
1387                 /* we need to strip off the root directory in the case
1388                    of relative paths, but this must be done _after_
1389                    the sorting phase */
1390                 for (i = 0; i < flist->count; i++) {
1391                         if (flist->files[i]->dirname &&
1392                             flist->files[i]->dirname[0] == '/') {
1393                                 memmove(&flist->files[i]->dirname[0],
1394                                         &flist->files[i]->dirname[1],
1395                                         strlen(flist->files[i]->dirname));
1396                         }
1397
1398                         if (flist->files[i]->dirname &&
1399                             !flist->files[i]->dirname[0]) {
1400                                 flist->files[i]->dirname = NULL;
1401                         }
1402                 }
1403         }
1404
1405         if (verbose <= 3)
1406                 return;
1407
1408         for (i = 0; i < flist->count; i++) {
1409                 rprintf(FINFO, "[%ld] i=%d %s %s mode=0%o len=%.0f\n",
1410                         (long) getpid(), i,
1411                         NS(flist->files[i]->dirname),
1412                         NS(flist->files[i]->basename),
1413                         (int) flist->files[i]->mode,
1414                         (double) flist->files[i]->length);
1415         }
1416 }
1417
1418
1419 enum fnc_state { fnc_DIR, fnc_SLASH, fnc_BASE };
1420
1421 /* Compare the names of two file_struct entities, just like strcmp()
1422  * would do if it were operating on the joined strings.  We assume
1423  * that there are no 0-length strings.
1424  */
1425 int f_name_cmp(struct file_struct *f1, struct file_struct *f2)
1426 {
1427         int dif;
1428         const uchar *c1, *c2;
1429         enum fnc_state state1, state2;
1430
1431         if (!f1 || !f1->basename) {
1432                 if (!f2 || !f2->basename)
1433                         return 0;
1434                 return -1;
1435         }
1436         if (!f2 || !f2->basename)
1437                 return 1;
1438
1439         if (!(c1 = (uchar*)f1->dirname)) {
1440                 state1 = fnc_BASE;
1441                 c1 = (uchar*)f1->basename;
1442         }
1443         else
1444                 state1 = fnc_DIR;
1445         if (!(c2 = (uchar*)f2->dirname)) {
1446                 state2 = fnc_BASE;
1447                 c2 = (uchar*)f2->basename;
1448         }
1449         else
1450                 state2 = fnc_DIR;
1451
1452         while (1) {
1453                 if ((dif = (int)*c1 - (int)*c2) != 0)
1454                         break;
1455                 if (!*++c1) {
1456                         switch (state1) {
1457                         case fnc_DIR:
1458                                 state1 = fnc_SLASH;
1459                                 c1 = (uchar*)"/";
1460                                 break;
1461                         case fnc_SLASH:
1462                                 state1 = fnc_BASE;
1463                                 c1 = (uchar*)f1->basename;
1464                                 break;
1465                         case fnc_BASE:
1466                                 break;
1467                         }
1468                 }
1469                 if (!*++c2) {
1470                         switch (state2) {
1471                         case fnc_DIR:
1472                                 state2 = fnc_SLASH;
1473                                 c2 = (uchar*)"/";
1474                                 break;
1475                         case fnc_SLASH:
1476                                 state2 = fnc_BASE;
1477                                 c2 = (uchar*)f2->basename;
1478                                 break;
1479                         case fnc_BASE:
1480                                 if (!*c1)
1481                                         return 0;
1482                                 break;
1483                         }
1484                 }
1485         }
1486
1487         return dif;
1488 }
1489
1490
1491 /* Return a copy of the full filename of a flist entry, using the indicated
1492  * buffer.
1493  */
1494 char *f_name_to(struct file_struct *f, char *fbuf, int bsize)
1495 {
1496         if (!f || !f->basename)
1497                 return NULL;
1498
1499         if (f->dirname) {
1500                 int off = strlcpy(fbuf, f->dirname, bsize);
1501                 off += strlcpy(fbuf + off, "/", bsize - off);
1502                 strlcpy(fbuf + off, f->basename, bsize - off);
1503         } else
1504                 strlcpy(fbuf, f->basename, bsize);
1505         return fbuf;
1506 }
1507
1508
1509 /* Like f_name_to(), but we rotate through 5 static buffers of our own.
1510  */
1511 char *f_name(struct file_struct *f)
1512 {
1513         static char names[5][MAXPATHLEN];
1514         static unsigned int n;
1515
1516         n = (n + 1) % (sizeof names / sizeof names[0]);
1517
1518         return f_name_to(f, names[n], sizeof names[0]);
1519 }