5f0f5a6fbc017ebc17c7343dd1a087f4b60fbc8b
[rsync/rsync.git] / hlink.c
1 /*
2  * Routines to support hard-linking.
3  *
4  * Copyright (C) 1996 Andrew Tridgell
5  * Copyright (C) 1996 Paul Mackerras
6  * Copyright (C) 2002 Martin Pool <mbp@samba.org>
7  * Copyright (C) 2004-2008 Wayne Davison
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License as published by
11  * the Free Software Foundation; either version 3 of the License, or
12  * (at your option) any later version.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, visit the http://fsf.org website.
21  */
22
23 #include "rsync.h"
24 #include "inums.h"
25
26 extern int dry_run;
27 extern int list_only;
28 extern int am_sender;
29 extern int inc_recurse;
30 extern int do_xfers;
31 extern int link_dest;
32 extern int preserve_acls;
33 extern int make_backups;
34 extern int protocol_version;
35 extern int remove_source_files;
36 extern int stdout_format_has_i;
37 extern int maybe_ATTRS_REPORT;
38 extern int unsort_ndx;
39 extern char *basis_dir[MAX_BASIS_DIRS+1];
40 extern struct file_list *cur_flist;
41
42 #ifdef SUPPORT_HARD_LINKS
43
44 /* Starting with protocol 30, we use a simple hashtable on the sending side
45  * for hashing the st_dev and st_ino info.  The receiving side gets told
46  * (via flags and a "group index") which items are hard-linked together, so
47  * we can avoid the pool of dev+inode data.  For incremental recursion mode,
48  * the receiver will use a ndx hash to remember old pathnames. */
49
50 static struct hashtable *dev_tbl;
51
52 static struct hashtable *prior_hlinks;
53
54 static struct file_list *hlink_flist;
55
56 void init_hard_links(void)
57 {
58         if (am_sender || protocol_version < 30)
59                 dev_tbl = hashtable_create(16, SIZEOF_INT64 == 8);
60         else if (inc_recurse)
61                 prior_hlinks = hashtable_create(1024, 0);
62 }
63
64 struct ht_int64_node *idev_find(int64 dev, int64 ino)
65 {
66         static struct ht_int64_node *dev_node = NULL;
67         struct hashtable *tbl;
68
69         if (!dev_node || dev_node->key != dev) {
70                 /* We keep a separate hash table of inodes for every device. */
71                 dev_node = hashtable_find(dev_tbl, dev, 1);
72                 if (!(tbl = dev_node->data)) {
73                         tbl = dev_node->data = hashtable_create(512, SIZEOF_INT64 == 8);
74                         if (DEBUG_GTE(HLINK, 3)) {
75                                 rprintf(FINFO,
76                                     "[%s] created hashtable for dev %s\n",
77                                     who_am_i(), big_num(dev));
78                         }
79                 }
80         } else
81                 tbl = dev_node->data;
82
83         return hashtable_find(tbl, ino, 1);
84 }
85
86 void idev_destroy(void)
87 {
88         int i;
89
90         for (i = 0; i < dev_tbl->size; i++) {
91                 struct ht_int32_node *node = HT_NODE(dev_tbl, dev_tbl->nodes, i);
92                 if (node->data)
93                         hashtable_destroy(node->data);
94         }
95
96         hashtable_destroy(dev_tbl);
97 }
98
99 static int hlink_compare_gnum(int *int1, int *int2)
100 {
101         struct file_struct *f1 = hlink_flist->sorted[*int1];
102         struct file_struct *f2 = hlink_flist->sorted[*int2];
103         int32 gnum1 = F_HL_GNUM(f1);
104         int32 gnum2 = F_HL_GNUM(f2);
105
106         if (gnum1 != gnum2)
107                 return gnum1 > gnum2 ? 1 : -1;
108
109         return *int1 > *int2 ? 1 : -1;
110 }
111
112 static void match_gnums(int32 *ndx_list, int ndx_count)
113 {
114         int32 from, prev;
115         struct file_struct *file, *file_next;
116         struct ht_int32_node *node = NULL;
117         int32 gnum, gnum_next;
118
119         qsort(ndx_list, ndx_count, sizeof ndx_list[0],
120              (int (*)()) hlink_compare_gnum);
121
122         for (from = 0; from < ndx_count; from++) {
123                 file = hlink_flist->sorted[ndx_list[from]];
124                 gnum = F_HL_GNUM(file);
125                 if (inc_recurse) {
126                         node = hashtable_find(prior_hlinks, gnum, 1);
127                         if (!node->data) {
128                                 if (!(node->data = new_array0(char, 5)))
129                                         out_of_memory("match_gnums");
130                                 assert(gnum >= hlink_flist->ndx_start);
131                                 file->flags |= FLAG_HLINK_FIRST;
132                                 prev = -1;
133                         } else if (CVAL(node->data, 0) == 0) {
134                                 struct file_list *flist;
135                                 prev = IVAL(node->data, 1);
136                                 flist = flist_for_ndx(prev, NULL);
137                                 if (flist)
138                                         flist->files[prev - flist->ndx_start]->flags &= ~FLAG_HLINK_LAST;
139                                 else {
140                                         /* We skipped all prior files in this
141                                          * group, so mark this as a "first". */
142                                         file->flags |= FLAG_HLINK_FIRST;
143                                         prev = -1;
144                                 }
145                         } else
146                                 prev = -1;
147                 } else {
148                         file->flags |= FLAG_HLINK_FIRST;
149                         prev = -1;
150                 }
151                 for ( ; from < ndx_count-1; file = file_next, gnum = gnum_next, from++) { /*SHARED ITERATOR*/
152                         file_next = hlink_flist->sorted[ndx_list[from+1]];
153                         gnum_next = F_HL_GNUM(file_next);
154                         if (gnum != gnum_next)
155                                 break;
156                         F_HL_PREV(file) = prev;
157                         /* The linked list uses over-the-wire ndx values. */
158                         if (unsort_ndx)
159                                 prev = F_NDX(file);
160                         else
161                                 prev = ndx_list[from] + hlink_flist->ndx_start;
162                 }
163                 if (prev < 0 && !inc_recurse) {
164                         /* Disable hard-link bit and set DONE so that
165                          * HLINK_BUMP()-dependent values are unaffected. */
166                         file->flags &= ~(FLAG_HLINKED | FLAG_HLINK_FIRST);
167                         file->flags |= FLAG_HLINK_DONE;
168                         continue;
169                 }
170
171                 file->flags |= FLAG_HLINK_LAST;
172                 F_HL_PREV(file) = prev;
173                 if (inc_recurse && CVAL(node->data, 0) == 0) {
174                         if (unsort_ndx)
175                                 prev = F_NDX(file);
176                         else
177                                 prev = ndx_list[from] + hlink_flist->ndx_start;
178                         SIVAL(node->data, 1, prev);
179                 }
180         }
181 }
182
183 /* Analyze the hard-links in the file-list by creating a list of all the
184  * items that have hlink data, sorting them, and matching up identical
185  * values into clusters.  These will be a single linked list from last
186  * to first when we're done. */
187 void match_hard_links(struct file_list *flist)
188 {
189         if (!list_only && flist->used) {
190                 int i, ndx_count = 0;
191                 int32 *ndx_list;
192
193                 if (!(ndx_list = new_array(int32, flist->used)))
194                         out_of_memory("match_hard_links");
195
196                 for (i = 0; i < flist->used; i++) {
197                         if (F_IS_HLINKED(flist->sorted[i]))
198                                 ndx_list[ndx_count++] = i;
199                 }
200
201                 hlink_flist = flist;
202
203                 if (ndx_count)
204                         match_gnums(ndx_list, ndx_count);
205
206                 free(ndx_list);
207         }
208         if (protocol_version < 30)
209                 idev_destroy();
210 }
211
212 static int maybe_hard_link(struct file_struct *file, int ndx,
213                            const char *fname, int statret, stat_x *sxp,
214                            const char *oldname, STRUCT_STAT *old_stp,
215                            const char *realname, int itemizing, enum logcode code)
216 {
217         if (statret == 0) {
218                 if (sxp->st.st_dev == old_stp->st_dev
219                  && sxp->st.st_ino == old_stp->st_ino) {
220                         if (itemizing) {
221                                 itemize(fname, file, ndx, statret, sxp,
222                                         ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS,
223                                         0, "");
224                         }
225                         if (INFO_GTE(NAME, 2) && maybe_ATTRS_REPORT)
226                                 rprintf(FCLIENT, "%s is uptodate\n", fname);
227                         file->flags |= FLAG_HLINK_DONE;
228                         return 0;
229                 }
230                 if (make_backups > 0) {
231                         if (!make_backup(fname))
232                                 return -1;
233                 } else if (robust_unlink(fname)) {
234                         rsyserr(FERROR_XFER, errno, "unlink %s failed",
235                                 full_fname(fname));
236                         return -1;
237                 }
238         }
239
240         if (hard_link_one(file, fname, oldname, 0)) {
241                 if (itemizing) {
242                         itemize(fname, file, ndx, statret, sxp,
243                                 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, 0,
244                                 realname);
245                 }
246                 if (code != FNONE && INFO_GTE(NAME, 1))
247                         rprintf(code, "%s => %s\n", fname, realname);
248                 return 0;
249         }
250         return -1;
251 }
252
253 /* Figure out if a prior entry is still there or if we just have a
254  * cached name for it. */
255 static char *check_prior(struct file_struct *file, int gnum,
256                          int *prev_ndx_p, struct file_list **flist_p)
257 {
258         struct file_struct *fp;
259         struct ht_int32_node *node;
260         int prev_ndx = F_HL_PREV(file);
261
262         while (1) {
263                 struct file_list *flist;
264                 if (prev_ndx < 0
265                  || (flist = flist_for_ndx(prev_ndx, NULL)) == NULL)
266                         break;
267                 fp = flist->files[prev_ndx - flist->ndx_start];
268                 if (!(fp->flags & FLAG_SKIP_HLINK)) {
269                         *prev_ndx_p = prev_ndx;
270                         *flist_p = flist;
271                         return NULL;
272                 }
273                 F_HL_PREV(file) = prev_ndx = F_HL_PREV(fp);
274         }
275
276         if (inc_recurse
277          && (node = hashtable_find(prior_hlinks, gnum, 0)) != NULL) {
278                 assert(node->data != NULL);
279                 if (CVAL(node->data, 0) != 0) {
280                         *prev_ndx_p = -1;
281                         *flist_p = NULL;
282                         return node->data;
283                 }
284                 /* The prior file must have been skipped. */
285                 F_HL_PREV(file) = -1;
286         }
287
288         *prev_ndx_p = -1;
289         *flist_p = NULL;
290         return NULL;
291 }
292
293 /* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not.  Returns:
294  * 0 = process the file, 1 = skip the file, -1 = error occurred. */
295 int hard_link_check(struct file_struct *file, int ndx, const char *fname,
296                     int statret, stat_x *sxp, int itemizing,
297                     enum logcode code)
298 {
299         STRUCT_STAT prev_st;
300         char namebuf[MAXPATHLEN], altbuf[MAXPATHLEN];
301         char *realname, *prev_name;
302         struct file_list *flist;
303         int gnum = inc_recurse ? F_HL_GNUM(file) : -1;
304         int prev_ndx;
305
306         prev_name = realname = check_prior(file, gnum, &prev_ndx, &flist);
307
308         if (!prev_name) {
309                 struct file_struct *prev_file;
310
311                 if (!flist) {
312                         /* The previous file was skipped, so this one is
313                          * treated as if it were the first in its group. */
314                         if (DEBUG_GTE(HLINK, 2)) {
315                                 rprintf(FINFO, "hlink for %d (%s,%d): virtual first\n",
316                                         ndx, f_name(file, NULL), gnum);
317                         }
318                         return 0;
319                 }
320
321                 prev_file = flist->files[prev_ndx - flist->ndx_start];
322
323                 /* Is the previous link not complete yet? */
324                 if (!(prev_file->flags & FLAG_HLINK_DONE)) {
325                         /* Is the previous link being transferred? */
326                         if (prev_file->flags & FLAG_FILE_SENT) {
327                                 /* Add ourselves to the list of files that will
328                                  * be updated when the transfer completes, and
329                                  * mark ourself as waiting for the transfer. */
330                                 F_HL_PREV(file) = F_HL_PREV(prev_file);
331                                 F_HL_PREV(prev_file) = ndx;
332                                 file->flags |= FLAG_FILE_SENT;
333                                 cur_flist->in_progress++;
334                                 if (DEBUG_GTE(HLINK, 2)) {
335                                         rprintf(FINFO, "hlink for %d (%s,%d): waiting for %d\n",
336                                                 ndx, f_name(file, NULL), gnum, F_HL_PREV(file));
337                                 }
338                                 return 1;
339                         }
340                         if (DEBUG_GTE(HLINK, 2)) {
341                                 rprintf(FINFO, "hlink for %d (%s,%d): looking for a leader\n",
342                                         ndx, f_name(file, NULL), gnum);
343                         }
344                         return 0;
345                 }
346
347                 /* There is a finished file to link with! */
348                 if (!(prev_file->flags & FLAG_HLINK_FIRST)) {
349                         /* The previous previous is FIRST when prev is not. */
350                         prev_name = realname = check_prior(prev_file, gnum, &prev_ndx, &flist);
351                         /* Update our previous pointer to point to the FIRST. */
352                         F_HL_PREV(file) = prev_ndx;
353                 }
354
355                 if (!prev_name) {
356                         int alt_dest;
357
358                         assert(flist != NULL);
359                         prev_file = flist->files[prev_ndx - flist->ndx_start];
360                         /* F_HL_PREV() is alt_dest value when DONE && FIRST. */
361                         alt_dest = F_HL_PREV(prev_file);
362                         if (DEBUG_GTE(HLINK, 2)) {
363                                 rprintf(FINFO, "hlink for %d (%s,%d): found flist match (alt %d)\n",
364                                         ndx, f_name(file, NULL), gnum, alt_dest);
365                         }
366
367                         if (alt_dest >= 0 && dry_run) {
368                                 pathjoin(namebuf, MAXPATHLEN, basis_dir[alt_dest],
369                                          f_name(prev_file, NULL));
370                                 prev_name = namebuf;
371                                 realname = f_name(prev_file, altbuf);
372                         } else {
373                                 prev_name = f_name(prev_file, namebuf);
374                                 realname = prev_name;
375                         }
376                 }
377         }
378
379         if (DEBUG_GTE(HLINK, 2)) {
380                 rprintf(FINFO, "hlink for %d (%s,%d): leader is %d (%s)\n",
381                         ndx, f_name(file, NULL), gnum, prev_ndx, prev_name);
382         }
383
384         if (link_stat(prev_name, &prev_st, 0) < 0) {
385                 rsyserr(FERROR_XFER, errno, "stat %s failed",
386                         full_fname(prev_name));
387                 return -1;
388         }
389
390         if (statret < 0 && basis_dir[0] != NULL) {
391                 /* If we match an alt-dest item, we don't output this as a change. */
392                 char cmpbuf[MAXPATHLEN];
393                 stat_x alt_sx;
394                 int j = 0;
395 #ifdef SUPPORT_ACLS
396                 alt_sx.acc_acl = alt_sx.def_acl = NULL;
397 #endif
398                 do {
399                         pathjoin(cmpbuf, MAXPATHLEN, basis_dir[j], fname);
400                         if (link_stat(cmpbuf, &alt_sx.st, 0) < 0)
401                                 continue;
402                         if (link_dest) {
403                                 if (prev_st.st_dev != alt_sx.st.st_dev
404                                  || prev_st.st_ino != alt_sx.st.st_ino)
405                                         continue;
406                                 statret = 1;
407                                 if (stdout_format_has_i == 0
408                                  || (!INFO_GTE(NAME, 2) && stdout_format_has_i < 2)) {
409                                         itemizing = 0;
410                                         code = FNONE;
411                                         if (INFO_GTE(NAME, 2) && maybe_ATTRS_REPORT)
412                                                 rprintf(FCLIENT, "%s is uptodate\n", fname);
413                                 }
414                                 break;
415                         }
416                         if (!unchanged_file(cmpbuf, file, &alt_sx.st))
417                                 continue;
418                         statret = 1;
419                         if (unchanged_attrs(cmpbuf, file, &alt_sx))
420                                 break;
421                 } while (basis_dir[++j] != NULL);
422                 if (statret == 1) {
423                         sxp->st = alt_sx.st;
424 #ifdef SUPPORT_ACLS
425                         if (preserve_acls && !S_ISLNK(file->mode)) {
426                                 if (!ACL_READY(*sxp))
427                                         get_acl(cmpbuf, sxp);
428                                 else {
429                                         sxp->acc_acl = alt_sx.acc_acl;
430                                         sxp->def_acl = alt_sx.def_acl;
431                                 }
432                         }
433 #endif
434                 }
435 #ifdef SUPPORT_ACLS
436                 else if (preserve_acls)
437                         free_acl(&alt_sx);
438 #endif
439         }
440
441         if (maybe_hard_link(file, ndx, fname, statret, sxp, prev_name, &prev_st,
442                             realname, itemizing, code) < 0)
443                 return -1;
444
445         if (remove_source_files == 1 && do_xfers)
446                 send_msg_int(MSG_SUCCESS, ndx);
447
448         return 1;
449 }
450
451 int hard_link_one(struct file_struct *file, const char *fname,
452                   const char *oldname, int terse)
453 {
454         if (do_link(oldname, fname) < 0) {
455                 enum logcode code;
456                 if (terse) {
457                         if (!INFO_GTE(NAME, 1))
458                                 return 0;
459                         code = FINFO;
460                 } else
461                         code = FERROR_XFER;
462                 rsyserr(code, errno, "link %s => %s failed",
463                         full_fname(fname), oldname);
464                 return 0;
465         }
466
467         file->flags |= FLAG_HLINK_DONE;
468
469         return 1;
470 }
471
472 void finish_hard_link(struct file_struct *file, const char *fname, int fin_ndx,
473                       STRUCT_STAT *stp, int itemizing, enum logcode code,
474                       int alt_dest)
475 {
476         stat_x prev_sx;
477         STRUCT_STAT st;
478         char prev_name[MAXPATHLEN], alt_name[MAXPATHLEN];
479         const char *our_name;
480         struct file_list *flist;
481         int prev_statret, ndx, prev_ndx = F_HL_PREV(file);
482
483         if (stp == NULL && prev_ndx >= 0) {
484                 if (link_stat(fname, &st, 0) < 0) {
485                         rsyserr(FERROR_XFER, errno, "stat %s failed",
486                                 full_fname(fname));
487                         return;
488                 }
489                 stp = &st;
490         }
491
492         /* FIRST combined with DONE means we were the first to get done. */
493         file->flags |= FLAG_HLINK_FIRST | FLAG_HLINK_DONE;
494         F_HL_PREV(file) = alt_dest;
495         if (alt_dest >= 0 && dry_run) {
496                 pathjoin(alt_name, MAXPATHLEN, basis_dir[alt_dest],
497                          f_name(file, NULL));
498                 our_name = alt_name;
499         } else
500                 our_name = fname;
501
502 #ifdef SUPPORT_ACLS
503         prev_sx.acc_acl = prev_sx.def_acl = NULL;
504 #endif
505
506         while ((ndx = prev_ndx) >= 0) {
507                 int val;
508                 flist = flist_for_ndx(ndx, "finish_hard_link");
509                 file = flist->files[ndx - flist->ndx_start];
510                 file->flags = (file->flags & ~FLAG_HLINK_FIRST) | FLAG_HLINK_DONE;
511                 prev_ndx = F_HL_PREV(file);
512                 F_HL_PREV(file) = fin_ndx;
513                 prev_statret = link_stat(f_name(file, prev_name), &prev_sx.st, 0);
514                 val = maybe_hard_link(file, ndx, prev_name, prev_statret, &prev_sx,
515                                       our_name, stp, fname, itemizing, code);
516                 flist->in_progress--;
517 #ifdef SUPPORT_ACLS
518                 if (preserve_acls)
519                         free_acl(&prev_sx);
520 #endif
521                 if (val < 0)
522                         continue;
523                 if (remove_source_files == 1 && do_xfers)
524                         send_msg_int(MSG_SUCCESS, ndx);
525         }
526
527         if (inc_recurse) {
528                 int gnum = F_HL_GNUM(file);
529                 struct ht_int32_node *node = hashtable_find(prior_hlinks, gnum, 0);
530                 assert(node != NULL && node->data != NULL);
531                 assert(CVAL(node->data, 0) == 0);
532                 free(node->data);
533                 if (!(node->data = strdup(our_name)))
534                         out_of_memory("finish_hard_link");
535         }
536 }
537
538 int skip_hard_link(struct file_struct *file, struct file_list **flist_p)
539 {
540         struct file_list *flist;
541         int prev_ndx;
542
543         file->flags |= FLAG_SKIP_HLINK;
544         if (!(file->flags & FLAG_HLINK_LAST))
545                 return -1;
546
547         check_prior(file, F_HL_GNUM(file), &prev_ndx, &flist);
548         if (prev_ndx >= 0) {
549                 file = flist->files[prev_ndx - flist->ndx_start];
550                 if (file->flags & (FLAG_HLINK_DONE|FLAG_FILE_SENT))
551                         return -1;
552                 file->flags |= FLAG_HLINK_LAST;
553                 *flist_p = flist;
554         }
555
556         return prev_ndx;
557 }
558 #endif