A few more improvements and a mention of the latest fix.
[rsync/rsync.git] / hlink.c
... / ...
CommitLineData
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
25extern int verbose;
26extern int dry_run;
27extern int list_only;
28extern int am_sender;
29extern int inc_recurse;
30extern int do_xfers;
31extern int link_dest;
32extern int preserve_acls;
33extern int make_backups;
34extern int protocol_version;
35extern int remove_source_files;
36extern int stdout_format_has_i;
37extern int maybe_ATTRS_REPORT;
38extern int unsort_ndx;
39extern char *basis_dir[];
40extern 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
50static struct hashtable *dev_tbl;
51
52static struct hashtable *prior_hlinks;
53
54static struct file_list *hlink_flist;
55
56void 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
64struct 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 } else
75 tbl = dev_node->data;
76
77 return hashtable_find(tbl, ino, 1);
78}
79
80void idev_destroy(void)
81{
82 int i;
83
84 for (i = 0; i < dev_tbl->size; i++) {
85 struct ht_int32_node *node = HT_NODE(dev_tbl, dev_tbl->nodes, i);
86 if (node->data)
87 hashtable_destroy(node->data);
88 }
89
90 hashtable_destroy(dev_tbl);
91}
92
93static int hlink_compare_gnum(int *int1, int *int2)
94{
95 struct file_struct *f1 = hlink_flist->sorted[*int1];
96 struct file_struct *f2 = hlink_flist->sorted[*int2];
97 int32 gnum1 = F_HL_GNUM(f1);
98 int32 gnum2 = F_HL_GNUM(f2);
99
100 if (gnum1 != gnum2)
101 return gnum1 > gnum2 ? 1 : -1;
102
103 return *int1 > *int2 ? 1 : -1;
104}
105
106static void match_gnums(int32 *ndx_list, int ndx_count)
107{
108 int32 from, prev;
109 struct file_struct *file, *file_next;
110 struct ht_int32_node *node = NULL;
111 int32 gnum, gnum_next;
112
113 qsort(ndx_list, ndx_count, sizeof ndx_list[0],
114 (int (*)()) hlink_compare_gnum);
115
116 for (from = 0; from < ndx_count; from++) {
117 file = hlink_flist->sorted[ndx_list[from]];
118 gnum = F_HL_GNUM(file);
119 if (inc_recurse) {
120 node = hashtable_find(prior_hlinks, gnum, 1);
121 if (!node->data) {
122 if (!(node->data = new_array0(char, 5)))
123 out_of_memory("match_gnums");
124 assert(gnum >= hlink_flist->ndx_start);
125 file->flags |= FLAG_HLINK_FIRST;
126 prev = -1;
127 } else if (CVAL(node->data, 0) == 0) {
128 struct file_list *flist;
129 prev = IVAL(node->data, 1);
130 flist = flist_for_ndx(prev, NULL);
131 if (flist)
132 flist->files[prev - flist->ndx_start]->flags &= ~FLAG_HLINK_LAST;
133 else {
134 /* We skipped all prior files in this
135 * group, so mark this as a "first". */
136 file->flags |= FLAG_HLINK_FIRST;
137 prev = -1;
138 }
139 } else
140 prev = -1;
141 } else {
142 file->flags |= FLAG_HLINK_FIRST;
143 prev = -1;
144 }
145 for ( ; from < ndx_count-1; file = file_next, gnum = gnum_next, from++) { /*SHARED ITERATOR*/
146 file_next = hlink_flist->sorted[ndx_list[from+1]];
147 gnum_next = F_HL_GNUM(file_next);
148 if (gnum != gnum_next)
149 break;
150 F_HL_PREV(file) = prev;
151 /* The linked list uses over-the-wire ndx values. */
152 if (unsort_ndx)
153 prev = F_NDX(file);
154 else
155 prev = ndx_list[from] + hlink_flist->ndx_start;
156 }
157 if (prev < 0 && !inc_recurse) {
158 /* Disable hard-link bit and set DONE so that
159 * HLINK_BUMP()-dependent values are unaffected. */
160 file->flags &= ~(FLAG_HLINKED | FLAG_HLINK_FIRST);
161 file->flags |= FLAG_HLINK_DONE;
162 continue;
163 }
164
165 file->flags |= FLAG_HLINK_LAST;
166 F_HL_PREV(file) = prev;
167 if (inc_recurse && CVAL(node->data, 0) == 0) {
168 if (unsort_ndx)
169 prev = F_NDX(file);
170 else
171 prev = ndx_list[from] + hlink_flist->ndx_start;
172 SIVAL(node->data, 1, prev);
173 }
174 }
175}
176
177/* Analyze the hard-links in the file-list by creating a list of all the
178 * items that have hlink data, sorting them, and matching up identical
179 * values into clusters. These will be a single linked list from last
180 * to first when we're done. */
181void match_hard_links(struct file_list *flist)
182{
183 if (!list_only && flist->used) {
184 int i, ndx_count = 0;
185 int32 *ndx_list;
186
187 if (!(ndx_list = new_array(int32, flist->used)))
188 out_of_memory("match_hard_links");
189
190 for (i = 0; i < flist->used; i++) {
191 if (F_IS_HLINKED(flist->sorted[i]))
192 ndx_list[ndx_count++] = i;
193 }
194
195 hlink_flist = flist;
196
197 if (ndx_count)
198 match_gnums(ndx_list, ndx_count);
199
200 free(ndx_list);
201 }
202 if (protocol_version < 30)
203 idev_destroy();
204}
205
206static int maybe_hard_link(struct file_struct *file, int ndx,
207 const char *fname, int statret, stat_x *sxp,
208 const char *oldname, STRUCT_STAT *old_stp,
209 const char *realname, int itemizing, enum logcode code)
210{
211 if (statret == 0) {
212 if (sxp->st.st_dev == old_stp->st_dev
213 && sxp->st.st_ino == old_stp->st_ino) {
214 if (itemizing) {
215 itemize(fname, file, ndx, statret, sxp,
216 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS,
217 0, "");
218 }
219 if (verbose > 1 && maybe_ATTRS_REPORT)
220 rprintf(FCLIENT, "%s is uptodate\n", fname);
221 file->flags |= FLAG_HLINK_DONE;
222 return 0;
223 }
224 if (make_backups > 0) {
225 if (!make_backup(fname))
226 return -1;
227 } else if (robust_unlink(fname)) {
228 rsyserr(FERROR_XFER, errno, "unlink %s failed",
229 full_fname(fname));
230 return -1;
231 }
232 }
233
234 if (hard_link_one(file, fname, oldname, 0)) {
235 if (itemizing) {
236 itemize(fname, file, ndx, statret, sxp,
237 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, 0,
238 realname);
239 }
240 if (code != FNONE && verbose)
241 rprintf(code, "%s => %s\n", fname, realname);
242 return 0;
243 }
244 return -1;
245}
246
247/* Figure out if a prior entry is still there or if we just have a
248 * cached name for it. */
249static char *check_prior(struct file_struct *file, int gnum,
250 int *prev_ndx_p, struct file_list **flist_p)
251{
252 struct file_struct *fp;
253 struct ht_int32_node *node;
254 int prev_ndx = F_HL_PREV(file);
255
256 while (1) {
257 struct file_list *flist;
258 if (prev_ndx < 0
259 || (flist = flist_for_ndx(prev_ndx, NULL)) == NULL)
260 break;
261 fp = flist->files[prev_ndx - flist->ndx_start];
262 if (!(fp->flags & FLAG_SKIP_HLINK)) {
263 *prev_ndx_p = prev_ndx;
264 *flist_p = flist;
265 return NULL;
266 }
267 F_HL_PREV(file) = prev_ndx = F_HL_PREV(fp);
268 }
269
270 if (inc_recurse
271 && (node = hashtable_find(prior_hlinks, gnum, 0)) != NULL) {
272 assert(node->data != NULL);
273 if (CVAL(node->data, 0) != 0) {
274 *prev_ndx_p = -1;
275 *flist_p = NULL;
276 return node->data;
277 }
278 /* The prior file must have been skipped. */
279 F_HL_PREV(file) = -1;
280 }
281
282 *prev_ndx_p = -1;
283 *flist_p = NULL;
284 return NULL;
285}
286
287/* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not. Returns:
288 * 0 = process the file, 1 = skip the file, -1 = error occurred. */
289int hard_link_check(struct file_struct *file, int ndx, const char *fname,
290 int statret, stat_x *sxp, int itemizing,
291 enum logcode code)
292{
293 STRUCT_STAT prev_st;
294 char namebuf[MAXPATHLEN], altbuf[MAXPATHLEN];
295 char *realname, *prev_name;
296 struct file_list *flist;
297 int gnum = inc_recurse ? F_HL_GNUM(file) : -1;
298 int prev_ndx;
299
300 prev_name = realname = check_prior(file, gnum, &prev_ndx, &flist);
301
302 if (!prev_name) {
303 struct file_struct *prev_file;
304
305 if (!flist) {
306 /* The previous file was skipped, so this one is
307 * treated as if it were the first in its group. */
308 return 0;
309 }
310
311 prev_file = flist->files[prev_ndx - flist->ndx_start];
312
313 /* Is the previous link not complete yet? */
314 if (!(prev_file->flags & FLAG_HLINK_DONE)) {
315 /* Is the previous link being transferred? */
316 if (prev_file->flags & FLAG_FILE_SENT) {
317 /* Add ourselves to the list of files that will
318 * be updated when the transfer completes, and
319 * mark ourself as waiting for the transfer. */
320 F_HL_PREV(file) = F_HL_PREV(prev_file);
321 F_HL_PREV(prev_file) = ndx;
322 file->flags |= FLAG_FILE_SENT;
323 cur_flist->in_progress++;
324 return 1;
325 }
326 return 0;
327 }
328
329 /* There is a finished file to link with! */
330 if (!(prev_file->flags & FLAG_HLINK_FIRST)) {
331 /* The previous previous is FIRST when prev is not. */
332 prev_name = realname = check_prior(prev_file, gnum, &prev_ndx, &flist);
333 assert(prev_name != NULL || flist != NULL);
334 /* Update our previous pointer to point to the FIRST. */
335 F_HL_PREV(file) = prev_ndx;
336 }
337
338 if (!prev_name) {
339 int alt_dest;
340
341 prev_file = flist->files[prev_ndx - flist->ndx_start];
342 /* F_HL_PREV() is alt_dest value when DONE && FIRST. */
343 alt_dest = F_HL_PREV(prev_file);
344
345 if (alt_dest >= 0 && dry_run) {
346 pathjoin(namebuf, MAXPATHLEN, basis_dir[alt_dest],
347 f_name(prev_file, NULL));
348 prev_name = namebuf;
349 realname = f_name(prev_file, altbuf);
350 } else {
351 prev_name = f_name(prev_file, namebuf);
352 realname = prev_name;
353 }
354 }
355 }
356
357 if (link_stat(prev_name, &prev_st, 0) < 0) {
358 rsyserr(FERROR_XFER, errno, "stat %s failed",
359 full_fname(prev_name));
360 return -1;
361 }
362
363 if (statret < 0 && basis_dir[0] != NULL) {
364 /* If we match an alt-dest item, we don't output this as a change. */
365 char cmpbuf[MAXPATHLEN];
366 stat_x alt_sx;
367 int j = 0;
368#ifdef SUPPORT_ACLS
369 alt_sx.acc_acl = alt_sx.def_acl = NULL;
370#endif
371 do {
372 pathjoin(cmpbuf, MAXPATHLEN, basis_dir[j], fname);
373 if (link_stat(cmpbuf, &alt_sx.st, 0) < 0)
374 continue;
375 if (link_dest) {
376 if (prev_st.st_dev != alt_sx.st.st_dev
377 || prev_st.st_ino != alt_sx.st.st_ino)
378 continue;
379 statret = 1;
380 if (stdout_format_has_i == 0
381 || (verbose < 2 && stdout_format_has_i < 2)) {
382 itemizing = 0;
383 code = FNONE;
384 if (verbose > 1 && maybe_ATTRS_REPORT)
385 rprintf(FCLIENT, "%s is uptodate\n", fname);
386 }
387 break;
388 }
389 if (!unchanged_file(cmpbuf, file, &alt_sx.st))
390 continue;
391 statret = 1;
392 if (unchanged_attrs(cmpbuf, file, &alt_sx))
393 break;
394 } while (basis_dir[++j] != NULL);
395 if (statret == 1) {
396 sxp->st = alt_sx.st;
397#ifdef SUPPORT_ACLS
398 if (preserve_acls && !S_ISLNK(file->mode)) {
399 if (!ACL_READY(*sxp))
400 get_acl(cmpbuf, sxp);
401 else {
402 sxp->acc_acl = alt_sx.acc_acl;
403 sxp->def_acl = alt_sx.def_acl;
404 }
405 }
406#endif
407 }
408#ifdef SUPPORT_ACLS
409 else if (preserve_acls)
410 free_acl(&alt_sx);
411#endif
412 }
413
414 if (maybe_hard_link(file, ndx, fname, statret, sxp, prev_name, &prev_st,
415 realname, itemizing, code) < 0)
416 return -1;
417
418 if (remove_source_files == 1 && do_xfers)
419 send_msg_int(MSG_SUCCESS, ndx);
420
421 return 1;
422}
423
424int hard_link_one(struct file_struct *file, const char *fname,
425 const char *oldname, int terse)
426{
427 if (do_link(oldname, fname) < 0) {
428 enum logcode code;
429 if (terse) {
430 if (!verbose)
431 return 0;
432 code = FINFO;
433 } else
434 code = FERROR_XFER;
435 rsyserr(code, errno, "link %s => %s failed",
436 full_fname(fname), oldname);
437 return 0;
438 }
439
440 file->flags |= FLAG_HLINK_DONE;
441
442 return 1;
443}
444
445void finish_hard_link(struct file_struct *file, const char *fname, int fin_ndx,
446 STRUCT_STAT *stp, int itemizing, enum logcode code,
447 int alt_dest)
448{
449 stat_x prev_sx;
450 STRUCT_STAT st;
451 char prev_name[MAXPATHLEN], alt_name[MAXPATHLEN];
452 const char *our_name;
453 struct file_list *flist;
454 int prev_statret, ndx, prev_ndx = F_HL_PREV(file);
455
456 if (stp == NULL && prev_ndx >= 0) {
457 if (link_stat(fname, &st, 0) < 0) {
458 rsyserr(FERROR_XFER, errno, "stat %s failed",
459 full_fname(fname));
460 return;
461 }
462 stp = &st;
463 }
464
465 /* FIRST combined with DONE means we were the first to get done. */
466 file->flags |= FLAG_HLINK_FIRST | FLAG_HLINK_DONE;
467 F_HL_PREV(file) = alt_dest;
468 if (alt_dest >= 0 && dry_run) {
469 pathjoin(alt_name, MAXPATHLEN, basis_dir[alt_dest],
470 f_name(file, NULL));
471 our_name = alt_name;
472 } else
473 our_name = fname;
474
475#ifdef SUPPORT_ACLS
476 prev_sx.acc_acl = prev_sx.def_acl = NULL;
477#endif
478
479 while ((ndx = prev_ndx) >= 0) {
480 int val;
481 flist = flist_for_ndx(ndx, "finish_hard_link");
482 file = flist->files[ndx - flist->ndx_start];
483 file->flags = (file->flags & ~FLAG_HLINK_FIRST) | FLAG_HLINK_DONE;
484 prev_ndx = F_HL_PREV(file);
485 F_HL_PREV(file) = fin_ndx;
486 prev_statret = link_stat(f_name(file, prev_name), &prev_sx.st, 0);
487 val = maybe_hard_link(file, ndx, prev_name, prev_statret, &prev_sx,
488 our_name, stp, fname, itemizing, code);
489 flist->in_progress--;
490#ifdef SUPPORT_ACLS
491 if (preserve_acls)
492 free_acl(&prev_sx);
493#endif
494 if (val < 0)
495 continue;
496 if (remove_source_files == 1 && do_xfers)
497 send_msg_int(MSG_SUCCESS, ndx);
498 }
499
500 if (inc_recurse) {
501 int gnum = F_HL_GNUM(file);
502 struct ht_int32_node *node = hashtable_find(prior_hlinks, gnum, 0);
503 assert(node != NULL && node->data != NULL);
504 assert(CVAL(node->data, 0) == 0);
505 free(node->data);
506 if (!(node->data = strdup(our_name)))
507 out_of_memory("finish_hard_link");
508 }
509}
510
511int skip_hard_link(struct file_struct *file, struct file_list **flist_p)
512{
513 struct file_list *flist;
514 int prev_ndx;
515
516 file->flags |= FLAG_SKIP_HLINK;
517 if (!(file->flags & FLAG_HLINK_LAST))
518 return -1;
519
520 check_prior(file, F_HL_GNUM(file), &prev_ndx, &flist);
521 if (prev_ndx >= 0) {
522 file = flist->files[prev_ndx - flist->ndx_start];
523 if (file->flags & (FLAG_HLINK_DONE|FLAG_FILE_SENT))
524 return -1;
525 file->flags |= FLAG_HLINK_LAST;
526 *flist_p = flist;
527 }
528
529 return prev_ndx;
530}
531#endif