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