A couple more support/rrsync tweaks:
[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 node->data = new_array0(char, 5);
123 assert(gnum >= hlink_flist->ndx_start);
124 file->flags |= FLAG_HLINK_FIRST;
125 prev = -1;
126 } else if (CVAL(node->data, 0) == 0) {
127 struct file_list *flist;
128 prev = IVAL(node->data, 1);
129 flist = flist_for_ndx(prev);
130 if (flist)
131 flist->files[prev - flist->ndx_start]->flags &= ~FLAG_HLINK_LAST;
132 else {
133 /* We skipped all prior files in this
134 * group, so mark this as a "first". */
135 file->flags |= FLAG_HLINK_FIRST;
136 prev = -1;
137 }
138 } else
139 prev = -1;
140 } else {
141 file->flags |= FLAG_HLINK_FIRST;
142 prev = -1;
143 }
144 for ( ; from < ndx_count-1; file = file_next, gnum = gnum_next, from++) { /*SHARED ITERATOR*/
145 file_next = hlink_flist->sorted[ndx_list[from+1]];
146 gnum_next = F_HL_GNUM(file_next);
147 if (gnum != gnum_next)
148 break;
149 F_HL_PREV(file) = prev;
150 /* The linked list uses over-the-wire ndx values. */
151 if (unsort_ndx)
152 prev = F_NDX(file);
153 else
154 prev = ndx_list[from] + hlink_flist->ndx_start;
155 }
156 if (prev < 0 && !inc_recurse) {
157 /* Disable hard-link bit and set DONE so that
158 * HLINK_BUMP()-dependent values are unaffected. */
159 file->flags &= ~(FLAG_HLINKED | FLAG_HLINK_FIRST);
160 file->flags |= FLAG_HLINK_DONE;
161 continue;
162 }
163
164 file->flags |= FLAG_HLINK_LAST;
165 F_HL_PREV(file) = prev;
166 if (inc_recurse && CVAL(node->data, 0) == 0) {
167 if (unsort_ndx)
168 prev = F_NDX(file);
169 else
170 prev = ndx_list[from] + hlink_flist->ndx_start;
171 SIVAL(node->data, 1, prev);
172 }
173 }
174}
175
176/* Analyze the hard-links in the file-list by creating a list of all the
177 * items that have hlink data, sorting them, and matching up identical
178 * values into clusters. These will be a single linked list from last
179 * to first when we're done. */
180void match_hard_links(struct file_list *flist)
181{
182 if (!list_only) {
183 int i, ndx_count = 0;
184 int32 *ndx_list;
185
186 if (!(ndx_list = new_array(int32, flist->used)))
187 out_of_memory("match_hard_links");
188
189 for (i = 0; i < flist->used; i++) {
190 if (F_IS_HLINKED(flist->sorted[i]))
191 ndx_list[ndx_count++] = i;
192 }
193
194 hlink_flist = flist;
195
196 if (ndx_count)
197 match_gnums(ndx_list, ndx_count);
198
199 free(ndx_list);
200 }
201 if (protocol_version < 30)
202 idev_destroy();
203}
204
205static int maybe_hard_link(struct file_struct *file, int ndx,
206 const char *fname, int statret, stat_x *sxp,
207 const char *oldname, STRUCT_STAT *old_stp,
208 const char *realname, int itemizing, enum logcode code)
209{
210 if (statret == 0) {
211 if (sxp->st.st_dev == old_stp->st_dev
212 && sxp->st.st_ino == old_stp->st_ino) {
213 if (itemizing) {
214 itemize(fname, file, ndx, statret, sxp,
215 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS,
216 0, "");
217 }
218 if (verbose > 1 && maybe_ATTRS_REPORT)
219 rprintf(FCLIENT, "%s is uptodate\n", fname);
220 file->flags |= FLAG_HLINK_DONE;
221 return 0;
222 }
223 if (make_backups > 0) {
224 if (!make_backup(fname))
225 return -1;
226 } else if (robust_unlink(fname)) {
227 rsyserr(FERROR_XFER, errno, "unlink %s failed",
228 full_fname(fname));
229 return -1;
230 }
231 }
232
233 if (hard_link_one(file, fname, oldname, 0)) {
234 if (itemizing) {
235 itemize(fname, file, ndx, statret, sxp,
236 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, 0,
237 realname);
238 }
239 if (code != FNONE && verbose)
240 rprintf(code, "%s => %s\n", fname, realname);
241 return 0;
242 }
243 return -1;
244}
245
246/* Figure out if a prior entry is still there or if we just have a
247 * cached name for it. */
248static char *check_prior(struct file_struct *file, int gnum,
249 int *prev_ndx_p, struct file_list **flist_p)
250{
251 struct file_struct *fp;
252 struct file_list *flist;
253 struct ht_int32_node *node;
254 int prev_ndx = F_HL_PREV(file);
255
256 while (1) {
257 if (prev_ndx < 0) {
258 *prev_ndx_p = prev_ndx;
259 *flist_p = NULL;
260 return NULL;
261 }
262 if ((flist = flist_for_ndx(prev_ndx)) == NULL)
263 break;
264 fp = flist->files[prev_ndx - flist->ndx_start];
265 if (!(fp->flags & FLAG_SKIP_HLINK)) {
266 *prev_ndx_p = prev_ndx;
267 *flist_p = flist;
268 return NULL;
269 }
270 F_HL_PREV(file) = prev_ndx = F_HL_PREV(fp);
271 }
272
273 node = hashtable_find(prior_hlinks, gnum, 0);
274 assert(node != NULL && node->data);
275
276 if (CVAL(node->data, 0) == 0) {
277 /* The prior file must have been skipped. */
278 F_HL_PREV(file) = prev_ndx = -1;
279 *prev_ndx_p = prev_ndx;
280 *flist_p = NULL;
281 return NULL;
282 }
283
284 *prev_ndx_p = prev_ndx;
285 *flist_p = flist;
286 return node->data;
287}
288
289/* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not. Returns:
290 * 0 = process the file, 1 = skip the file, -1 = error occurred. */
291int hard_link_check(struct file_struct *file, int ndx, const char *fname,
292 int statret, stat_x *sxp, int itemizing,
293 enum logcode code)
294{
295 STRUCT_STAT prev_st;
296 char namebuf[MAXPATHLEN], altbuf[MAXPATHLEN];
297 char *realname, *prev_name;
298 struct file_list *flist;
299 int gnum = inc_recurse ? F_HL_GNUM(file) : -1;
300 int prev_ndx;
301
302 prev_name = realname = check_prior(file, gnum, &prev_ndx, &flist);
303
304 if (!prev_name) {
305 struct file_struct *prev_file;
306
307 if (!flist) {
308 /* The previous file was skipped, so this one is
309 * treated as if it were the first in its group. */
310 return 0;
311 }
312
313 prev_file = flist->files[prev_ndx - flist->ndx_start];
314
315 /* Is the previous link not complete yet? */
316 if (!(prev_file->flags & FLAG_HLINK_DONE)) {
317 /* Is the previous link being transferred? */
318 if (prev_file->flags & FLAG_FILE_SENT) {
319 /* Add ourselves to the list of files that will
320 * be updated when the transfer completes, and
321 * mark ourself as waiting for the transfer. */
322 F_HL_PREV(file) = F_HL_PREV(prev_file);
323 F_HL_PREV(prev_file) = ndx;
324 file->flags |= FLAG_FILE_SENT;
325 cur_flist->in_progress++;
326 return 1;
327 }
328 return 0;
329 }
330
331 /* There is a finished file to link with! */
332 if (!(prev_file->flags & FLAG_HLINK_FIRST)) {
333 /* The previous previous is FIRST when prev is not. */
334 prev_name = realname = check_prior(prev_file, gnum, &prev_ndx, &flist);
335 assert(prev_name != NULL || flist != NULL);
336 /* Update our previous pointer to point to the FIRST. */
337 F_HL_PREV(file) = prev_ndx;
338 }
339
340 if (!prev_name) {
341 int alt_dest;
342
343 prev_file = flist->files[prev_ndx - flist->ndx_start];
344 /* F_HL_PREV() is alt_dest value when DONE && FIRST. */
345 alt_dest = F_HL_PREV(prev_file);
346
347 if (alt_dest >= 0 && dry_run) {
348 pathjoin(namebuf, MAXPATHLEN, basis_dir[alt_dest],
349 f_name(prev_file, NULL));
350 prev_name = namebuf;
351 realname = f_name(prev_file, altbuf);
352 } else {
353 prev_name = f_name(prev_file, namebuf);
354 realname = prev_name;
355 }
356 }
357 }
358
359 if (link_stat(prev_name, &prev_st, 0) < 0) {
360 rsyserr(FERROR_XFER, errno, "stat %s failed",
361 full_fname(prev_name));
362 return -1;
363 }
364
365 if (statret < 0 && basis_dir[0] != NULL) {
366 /* If we match an alt-dest item, we don't output this as a change. */
367 char cmpbuf[MAXPATHLEN];
368 stat_x alt_sx;
369 int j = 0;
370#ifdef SUPPORT_ACLS
371 alt_sx.acc_acl = alt_sx.def_acl = NULL;
372#endif
373 do {
374 pathjoin(cmpbuf, MAXPATHLEN, basis_dir[j], fname);
375 if (link_stat(cmpbuf, &alt_sx.st, 0) < 0)
376 continue;
377 if (link_dest) {
378 if (prev_st.st_dev != alt_sx.st.st_dev
379 || prev_st.st_ino != alt_sx.st.st_ino)
380 continue;
381 statret = 1;
382 if (stdout_format_has_i == 0
383 || (verbose < 2 && stdout_format_has_i < 2)) {
384 itemizing = 0;
385 code = FNONE;
386 if (verbose > 1 && maybe_ATTRS_REPORT)
387 rprintf(FCLIENT, "%s is uptodate\n", fname);
388 }
389 break;
390 }
391 if (!unchanged_file(cmpbuf, file, &alt_sx.st))
392 continue;
393 statret = 1;
394 if (unchanged_attrs(cmpbuf, file, &alt_sx))
395 break;
396 } while (basis_dir[++j] != NULL);
397 if (statret == 1) {
398 sxp->st = alt_sx.st;
399#ifdef SUPPORT_ACLS
400 if (preserve_acls && !S_ISLNK(file->mode)) {
401 if (!ACL_READY(*sxp))
402 get_acl(cmpbuf, sxp);
403 else {
404 sxp->acc_acl = alt_sx.acc_acl;
405 sxp->def_acl = alt_sx.def_acl;
406 }
407 }
408#endif
409 }
410#ifdef SUPPORT_ACLS
411 else if (preserve_acls)
412 free_acl(&alt_sx);
413#endif
414 }
415
416 if (maybe_hard_link(file, ndx, fname, statret, sxp, prev_name, &prev_st,
417 realname, itemizing, code) < 0)
418 return -1;
419
420 if (remove_source_files == 1 && do_xfers)
421 send_msg_int(MSG_SUCCESS, ndx);
422
423 return 1;
424}
425
426int hard_link_one(struct file_struct *file, const char *fname,
427 const char *oldname, int terse)
428{
429 if (do_link(oldname, fname) < 0) {
430 enum logcode code;
431 if (terse) {
432 if (!verbose)
433 return 0;
434 code = FINFO;
435 } else
436 code = FERROR_XFER;
437 rsyserr(code, errno, "link %s => %s failed",
438 full_fname(fname), oldname);
439 return 0;
440 }
441
442 file->flags |= FLAG_HLINK_DONE;
443
444 return 1;
445}
446
447void finish_hard_link(struct file_struct *file, const char *fname, int fin_ndx,
448 STRUCT_STAT *stp, int itemizing, enum logcode code,
449 int alt_dest)
450{
451 stat_x prev_sx;
452 STRUCT_STAT st;
453 char alt_name[MAXPATHLEN], *prev_name;
454 const char *our_name;
455 struct file_list *flist;
456 int prev_statret, ndx, prev_ndx = F_HL_PREV(file);
457
458 if (stp == NULL && prev_ndx >= 0) {
459 if (link_stat(fname, &st, 0) < 0) {
460 rsyserr(FERROR_XFER, errno, "stat %s failed",
461 full_fname(fname));
462 return;
463 }
464 stp = &st;
465 }
466
467 /* FIRST combined with DONE means we were the first to get done. */
468 file->flags |= FLAG_HLINK_FIRST | FLAG_HLINK_DONE;
469 F_HL_PREV(file) = alt_dest;
470 if (alt_dest >= 0 && dry_run) {
471 pathjoin(alt_name, MAXPATHLEN, basis_dir[alt_dest],
472 f_name(file, NULL));
473 our_name = alt_name;
474 } else
475 our_name = fname;
476
477#ifdef SUPPORT_ACLS
478 prev_sx.acc_acl = prev_sx.def_acl = NULL;
479#endif
480
481 while ((ndx = prev_ndx) >= 0) {
482 int val;
483 flist = flist_for_ndx(ndx);
484 assert(flist != NULL);
485 file = flist->files[ndx - flist->ndx_start];
486 file->flags = (file->flags & ~FLAG_HLINK_FIRST) | FLAG_HLINK_DONE;
487 prev_ndx = F_HL_PREV(file);
488 F_HL_PREV(file) = fin_ndx;
489 prev_name = f_name(file, NULL);
490 prev_statret = link_stat(prev_name, &prev_sx.st, 0);
491 val = maybe_hard_link(file, ndx, prev_name, prev_statret, &prev_sx,
492 our_name, stp, fname, itemizing, code);
493 flist->in_progress--;
494#ifdef SUPPORT_ACLS
495 if (preserve_acls)
496 free_acl(&prev_sx);
497#endif
498 if (val < 0)
499 continue;
500 if (remove_source_files == 1 && do_xfers)
501 send_msg_int(MSG_SUCCESS, ndx);
502 }
503
504 if (inc_recurse) {
505 int gnum = F_HL_GNUM(file);
506 struct ht_int32_node *node = hashtable_find(prior_hlinks, gnum, 0);
507 assert(node != NULL && node->data != NULL);
508 assert(CVAL(node->data, 0) == 0);
509 free(node->data);
510 if (!(node->data = strdup(our_name)))
511 out_of_memory("finish_hard_link");
512 }
513}
514
515int skip_hard_link(struct file_struct *file, struct file_list **flist_p)
516{
517 struct file_list *flist;
518 int prev_ndx;
519
520 file->flags |= FLAG_SKIP_HLINK;
521 if (!(file->flags & FLAG_HLINK_LAST))
522 return -1;
523
524 check_prior(file, F_HL_GNUM(file), &prev_ndx, &flist);
525 if (prev_ndx >= 0) {
526 file = flist->files[prev_ndx - flist->ndx_start];
527 if (file->flags & (FLAG_HLINK_DONE|FLAG_FILE_SENT))
528 return -1;
529 file->flags |= FLAG_HLINK_LAST;
530 *flist_p = flist;
531 }
532
533 return prev_ndx;
534}
535#endif