Mention a few NEWS items that hadn't been mentioned yet.
[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 preserve_xattrs;
34extern int make_backups;
35extern int protocol_version;
36extern int remove_source_files;
37extern int stdout_format_has_i;
38extern int maybe_ATTRS_REPORT;
39extern int unsort_ndx;
40extern char *basis_dir[];
41extern struct file_list *cur_flist;
42
43#ifdef SUPPORT_HARD_LINKS
44
45/* Starting with protocol 30, we use a simple hashtable on the sending side
46 * for hashing the st_dev and st_ino info. The receiving side gets told
47 * (via flags and a "group index") which items are hard-linked together, so
48 * we can avoid the pool of dev+inode data. For incremental recursion mode,
49 * the receiver will use a ndx hash to remember old pathnames. */
50
51static struct hashtable *dev_tbl;
52
53static struct hashtable *prior_hlinks;
54
55static struct file_list *hlink_flist;
56
57void init_hard_links(void)
58{
59 if (am_sender || protocol_version < 30)
60 dev_tbl = hashtable_create(16, SIZEOF_INT64 == 8);
61 else if (inc_recurse)
62 prior_hlinks = hashtable_create(1024, 0);
63}
64
65struct ht_int64_node *idev_find(int64 dev, int64 ino)
66{
67 static struct ht_int64_node *dev_node = NULL;
68 struct hashtable *tbl;
69
70 if (!dev_node || dev_node->key != dev) {
71 /* We keep a separate hash table of inodes for every device. */
72 dev_node = hashtable_find(dev_tbl, dev, 1);
73 if (!(tbl = dev_node->data))
74 tbl = dev_node->data = hashtable_create(512, SIZEOF_INT64 == 8);
75 } else
76 tbl = dev_node->data;
77
78 return hashtable_find(tbl, ino, 1);
79}
80
81void idev_destroy(void)
82{
83 int i;
84
85 for (i = 0; i < dev_tbl->size; i++) {
86 struct ht_int32_node *node = HT_NODE(dev_tbl, dev_tbl->nodes, i);
87 if (node->data)
88 hashtable_destroy(node->data);
89 }
90
91 hashtable_destroy(dev_tbl);
92}
93
94static int hlink_compare_gnum(int *int1, int *int2)
95{
96 struct file_struct *f1 = hlink_flist->sorted[*int1];
97 struct file_struct *f2 = hlink_flist->sorted[*int2];
98 int32 gnum1 = F_HL_GNUM(f1);
99 int32 gnum2 = F_HL_GNUM(f2);
100
101 if (gnum1 != gnum2)
102 return gnum1 > gnum2 ? 1 : -1;
103
104 return *int1 > *int2 ? 1 : -1;
105}
106
107static void match_gnums(int32 *ndx_list, int ndx_count)
108{
109 int32 from, prev;
110 struct file_struct *file, *file_next;
111 struct ht_int32_node *node = NULL;
112 int32 gnum, gnum_next;
113
114 qsort(ndx_list, ndx_count, sizeof ndx_list[0],
115 (int (*)()) hlink_compare_gnum);
116
117 for (from = 0; from < ndx_count; from++) {
118 file = hlink_flist->sorted[ndx_list[from]];
119 gnum = F_HL_GNUM(file);
120 if (inc_recurse) {
121 node = hashtable_find(prior_hlinks, gnum, 1);
122 if (!node->data) {
123 if (!(node->data = new_array0(char, 5)))
124 out_of_memory("match_gnums");
125 assert(gnum >= hlink_flist->ndx_start);
126 file->flags |= FLAG_HLINK_FIRST;
127 prev = -1;
128 } else if (CVAL(node->data, 0) == 0) {
129 struct file_list *flist;
130 prev = IVAL(node->data, 1);
131 flist = flist_for_ndx(prev, NULL);
132 if (flist)
133 flist->files[prev - flist->ndx_start]->flags &= ~FLAG_HLINK_LAST;
134 else {
135 /* We skipped all prior files in this
136 * group, so mark this as a "first". */
137 file->flags |= FLAG_HLINK_FIRST;
138 prev = -1;
139 }
140 } else
141 prev = -1;
142 } else {
143 file->flags |= FLAG_HLINK_FIRST;
144 prev = -1;
145 }
146 for ( ; from < ndx_count-1; file = file_next, gnum = gnum_next, from++) { /*SHARED ITERATOR*/
147 file_next = hlink_flist->sorted[ndx_list[from+1]];
148 gnum_next = F_HL_GNUM(file_next);
149 if (gnum != gnum_next)
150 break;
151 F_HL_PREV(file) = prev;
152 /* The linked list uses over-the-wire ndx values. */
153 if (unsort_ndx)
154 prev = F_NDX(file);
155 else
156 prev = ndx_list[from] + hlink_flist->ndx_start;
157 }
158 if (prev < 0 && !inc_recurse) {
159 /* Disable hard-link bit and set DONE so that
160 * HLINK_BUMP()-dependent values are unaffected. */
161 file->flags &= ~(FLAG_HLINKED | FLAG_HLINK_FIRST);
162 file->flags |= FLAG_HLINK_DONE;
163 continue;
164 }
165
166 file->flags |= FLAG_HLINK_LAST;
167 F_HL_PREV(file) = prev;
168 if (inc_recurse && CVAL(node->data, 0) == 0) {
169 if (unsort_ndx)
170 prev = F_NDX(file);
171 else
172 prev = ndx_list[from] + hlink_flist->ndx_start;
173 SIVAL(node->data, 1, prev);
174 }
175 }
176}
177
178/* Analyze the hard-links in the file-list by creating a list of all the
179 * items that have hlink data, sorting them, and matching up identical
180 * values into clusters. These will be a single linked list from last
181 * to first when we're done. */
182void match_hard_links(struct file_list *flist)
183{
184 if (!list_only && flist->used) {
185 int i, ndx_count = 0;
186 int32 *ndx_list;
187
188 if (!(ndx_list = new_array(int32, flist->used)))
189 out_of_memory("match_hard_links");
190
191 for (i = 0; i < flist->used; i++) {
192 if (F_IS_HLINKED(flist->sorted[i]))
193 ndx_list[ndx_count++] = i;
194 }
195
196 hlink_flist = flist;
197
198 if (ndx_count)
199 match_gnums(ndx_list, ndx_count);
200
201 free(ndx_list);
202 }
203 if (protocol_version < 30)
204 idev_destroy();
205}
206
207static int maybe_hard_link(struct file_struct *file, int ndx,
208 const char *fname, int statret, stat_x *sxp,
209 const char *oldname, STRUCT_STAT *old_stp,
210 const char *realname, int itemizing, enum logcode code)
211{
212 if (statret == 0) {
213 if (sxp->st.st_dev == old_stp->st_dev
214 && sxp->st.st_ino == old_stp->st_ino) {
215 if (itemizing) {
216 itemize(fname, file, ndx, statret, sxp,
217 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS,
218 0, "");
219 }
220 if (verbose > 1 && maybe_ATTRS_REPORT)
221 rprintf(FCLIENT, "%s is uptodate\n", fname);
222 file->flags |= FLAG_HLINK_DONE;
223 return 0;
224 }
225 if (make_backups > 0) {
226 if (!make_backup(fname))
227 return -1;
228 } else if (robust_unlink(fname)) {
229 rsyserr(FERROR_XFER, errno, "unlink %s failed",
230 full_fname(fname));
231 return -1;
232 }
233 }
234
235 if (hard_link_one(file, fname, oldname, 0)) {
236 if (itemizing) {
237 itemize(fname, file, ndx, statret, sxp,
238 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, 0,
239 realname);
240 }
241 if (code != FNONE && verbose)
242 rprintf(code, "%s => %s\n", fname, realname);
243 return 0;
244 }
245 return -1;
246}
247
248/* Figure out if a prior entry is still there or if we just have a
249 * cached name for it. */
250static char *check_prior(struct file_struct *file, int gnum,
251 int *prev_ndx_p, struct file_list **flist_p)
252{
253 struct file_struct *fp;
254 struct ht_int32_node *node;
255 int prev_ndx = F_HL_PREV(file);
256
257 while (1) {
258 struct file_list *flist;
259 if (prev_ndx < 0
260 || (flist = flist_for_ndx(prev_ndx, NULL)) == NULL)
261 break;
262 fp = flist->files[prev_ndx - flist->ndx_start];
263 if (!(fp->flags & FLAG_SKIP_HLINK)) {
264 *prev_ndx_p = prev_ndx;
265 *flist_p = flist;
266 return NULL;
267 }
268 F_HL_PREV(file) = prev_ndx = F_HL_PREV(fp);
269 }
270
271 if (inc_recurse
272 && (node = hashtable_find(prior_hlinks, gnum, 0)) != NULL) {
273 assert(node->data != NULL);
274 if (CVAL(node->data, 0) != 0) {
275 *prev_ndx_p = -1;
276 *flist_p = NULL;
277 return node->data;
278 }
279 /* The prior file must have been skipped. */
280 F_HL_PREV(file) = -1;
281 }
282
283 *prev_ndx_p = -1;
284 *flist_p = NULL;
285 return NULL;
286}
287
288/* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not. Returns:
289 * 0 = process the file, 1 = skip the file, -1 = error occurred. */
290int hard_link_check(struct file_struct *file, int ndx, const char *fname,
291 int statret, stat_x *sxp, int itemizing,
292 enum logcode code)
293{
294 STRUCT_STAT prev_st;
295 char namebuf[MAXPATHLEN], altbuf[MAXPATHLEN];
296 char *realname, *prev_name;
297 struct file_list *flist;
298 int gnum = inc_recurse ? F_HL_GNUM(file) : -1;
299 int prev_ndx;
300
301 prev_name = realname = check_prior(file, gnum, &prev_ndx, &flist);
302
303 if (!prev_name) {
304 struct file_struct *prev_file;
305
306 if (!flist) {
307 /* The previous file was skipped, so this one is
308 * treated as if it were the first in its group. */
309 return 0;
310 }
311
312 prev_file = flist->files[prev_ndx - flist->ndx_start];
313
314 /* Is the previous link not complete yet? */
315 if (!(prev_file->flags & FLAG_HLINK_DONE)) {
316 /* Is the previous link being transferred? */
317 if (prev_file->flags & FLAG_FILE_SENT) {
318 /* Add ourselves to the list of files that will
319 * be updated when the transfer completes, and
320 * mark ourself as waiting for the transfer. */
321 F_HL_PREV(file) = F_HL_PREV(prev_file);
322 F_HL_PREV(prev_file) = ndx;
323 file->flags |= FLAG_FILE_SENT;
324 cur_flist->in_progress++;
325 return 1;
326 }
327 return 0;
328 }
329
330 /* There is a finished file to link with! */
331 if (!(prev_file->flags & FLAG_HLINK_FIRST)) {
332 /* The previous previous is FIRST when prev is not. */
333 prev_name = realname = check_prior(prev_file, gnum, &prev_ndx, &flist);
334 assert(prev_name != NULL || flist != NULL);
335 /* Update our previous pointer to point to the FIRST. */
336 F_HL_PREV(file) = prev_ndx;
337 }
338
339 if (!prev_name) {
340 int alt_dest;
341
342 prev_file = flist->files[prev_ndx - flist->ndx_start];
343 /* F_HL_PREV() is alt_dest value when DONE && FIRST. */
344 alt_dest = F_HL_PREV(prev_file);
345
346 if (alt_dest >= 0 && dry_run) {
347 pathjoin(namebuf, MAXPATHLEN, basis_dir[alt_dest],
348 f_name(prev_file, NULL));
349 prev_name = namebuf;
350 realname = f_name(prev_file, altbuf);
351 } else {
352 prev_name = f_name(prev_file, namebuf);
353 realname = prev_name;
354 }
355 }
356 }
357
358 if (link_stat(prev_name, &prev_st, 0) < 0) {
359 rsyserr(FERROR_XFER, errno, "stat %s failed",
360 full_fname(prev_name));
361 return -1;
362 }
363
364 if (statret < 0 && basis_dir[0] != NULL) {
365 /* If we match an alt-dest item, we don't output this as a change. */
366 char cmpbuf[MAXPATHLEN];
367 stat_x alt_sx;
368 int j = 0;
369#ifdef SUPPORT_ACLS
370 alt_sx.acc_acl = alt_sx.def_acl = NULL;
371#endif
372#ifdef SUPPORT_XATTRS
373 alt_sx.xattr = NULL;
374#endif
375 do {
376 pathjoin(cmpbuf, MAXPATHLEN, basis_dir[j], fname);
377 if (link_stat(cmpbuf, &alt_sx.st, 0) < 0)
378 continue;
379 if (link_dest) {
380 if (prev_st.st_dev != alt_sx.st.st_dev
381 || prev_st.st_ino != alt_sx.st.st_ino)
382 continue;
383 statret = 1;
384 if (stdout_format_has_i == 0
385 || (verbose < 2 && stdout_format_has_i < 2)) {
386 itemizing = 0;
387 code = FNONE;
388 if (verbose > 1 && maybe_ATTRS_REPORT)
389 rprintf(FCLIENT, "%s is uptodate\n", fname);
390 }
391 break;
392 }
393 if (!unchanged_file(cmpbuf, file, &alt_sx.st))
394 continue;
395 statret = 1;
396 if (unchanged_attrs(cmpbuf, file, &alt_sx))
397 break;
398 } while (basis_dir[++j] != NULL);
399 if (statret == 1) {
400 sxp->st = alt_sx.st;
401#ifdef SUPPORT_ACLS
402 if (preserve_acls && !S_ISLNK(file->mode)) {
403 free_acl(sxp);
404 if (!ACL_READY(alt_sx))
405 get_acl(cmpbuf, sxp);
406 else {
407 sxp->acc_acl = alt_sx.acc_acl;
408 sxp->def_acl = alt_sx.def_acl;
409 alt_sx.acc_acl = alt_sx.def_acl = NULL;
410 }
411 }
412#endif
413#ifdef SUPPORT_XATTRS
414 if (preserve_xattrs) {
415 free_xattr(sxp);
416 if (!XATTR_READY(alt_sx))
417 get_xattr(cmpbuf, sxp);
418 else {
419 sxp->xattr = alt_sx.xattr;
420 alt_sx.xattr = NULL;
421 }
422 }
423#endif
424 } else {
425#ifdef SUPPORT_ACLS
426 if (preserve_acls)
427 free_acl(&alt_sx);
428#endif
429#ifdef SUPPORT_XATTRS
430 if (preserve_xattrs)
431 free_xattr(&alt_sx);
432#endif
433 }
434 }
435
436 if (maybe_hard_link(file, ndx, fname, statret, sxp, prev_name, &prev_st,
437 realname, itemizing, code) < 0)
438 return -1;
439
440 if (remove_source_files == 1 && do_xfers)
441 send_msg_int(MSG_SUCCESS, ndx);
442
443 return 1;
444}
445
446int hard_link_one(struct file_struct *file, const char *fname,
447 const char *oldname, int terse)
448{
449 if (do_link(oldname, fname) < 0) {
450 enum logcode code;
451 if (terse) {
452 if (!verbose)
453 return 0;
454 code = FINFO;
455 } else
456 code = FERROR_XFER;
457 rsyserr(code, errno, "link %s => %s failed",
458 full_fname(fname), oldname);
459 return 0;
460 }
461
462 file->flags |= FLAG_HLINK_DONE;
463
464 return 1;
465}
466
467void finish_hard_link(struct file_struct *file, const char *fname, int fin_ndx,
468 STRUCT_STAT *stp, int itemizing, enum logcode code,
469 int alt_dest)
470{
471 stat_x prev_sx;
472 STRUCT_STAT st;
473 char prev_name[MAXPATHLEN], alt_name[MAXPATHLEN];
474 const char *our_name;
475 struct file_list *flist;
476 int prev_statret, ndx, prev_ndx = F_HL_PREV(file);
477
478 if (stp == NULL && prev_ndx >= 0) {
479 if (link_stat(fname, &st, 0) < 0) {
480 rsyserr(FERROR_XFER, errno, "stat %s failed",
481 full_fname(fname));
482 return;
483 }
484 stp = &st;
485 }
486
487 /* FIRST combined with DONE means we were the first to get done. */
488 file->flags |= FLAG_HLINK_FIRST | FLAG_HLINK_DONE;
489 F_HL_PREV(file) = alt_dest;
490 if (alt_dest >= 0 && dry_run) {
491 pathjoin(alt_name, MAXPATHLEN, basis_dir[alt_dest],
492 f_name(file, NULL));
493 our_name = alt_name;
494 } else
495 our_name = fname;
496
497#ifdef SUPPORT_ACLS
498 prev_sx.acc_acl = prev_sx.def_acl = NULL;
499#endif
500#ifdef SUPPORT_XATTRS
501 prev_sx.xattr = NULL;
502#endif
503
504 while ((ndx = prev_ndx) >= 0) {
505 int val;
506 flist = flist_for_ndx(ndx, "finish_hard_link");
507 file = flist->files[ndx - flist->ndx_start];
508 file->flags = (file->flags & ~FLAG_HLINK_FIRST) | FLAG_HLINK_DONE;
509 prev_ndx = F_HL_PREV(file);
510 F_HL_PREV(file) = fin_ndx;
511 prev_statret = link_stat(f_name(file, prev_name), &prev_sx.st, 0);
512 val = maybe_hard_link(file, ndx, prev_name, prev_statret, &prev_sx,
513 our_name, stp, fname, itemizing, code);
514 flist->in_progress--;
515#ifdef SUPPORT_ACLS
516 if (preserve_acls)
517 free_acl(&prev_sx);
518#endif
519#ifdef SUPPORT_XATTRS
520 if (preserve_xattrs)
521 free_xattr(&prev_sx);
522#endif
523 if (val < 0)
524 continue;
525 if (remove_source_files == 1 && do_xfers)
526 send_msg_int(MSG_SUCCESS, ndx);
527 }
528
529 if (inc_recurse) {
530 int gnum = F_HL_GNUM(file);
531 struct ht_int32_node *node = hashtable_find(prior_hlinks, gnum, 0);
532 assert(node != NULL && node->data != NULL);
533 assert(CVAL(node->data, 0) == 0);
534 free(node->data);
535 if (!(node->data = strdup(our_name)))
536 out_of_memory("finish_hard_link");
537 }
538}
539
540int skip_hard_link(struct file_struct *file, struct file_list **flist_p)
541{
542 struct file_list *flist;
543 int prev_ndx;
544
545 file->flags |= FLAG_SKIP_HLINK;
546 if (!(file->flags & FLAG_HLINK_LAST))
547 return -1;
548
549 check_prior(file, F_HL_GNUM(file), &prev_ndx, &flist);
550 if (prev_ndx >= 0) {
551 file = flist->files[prev_ndx - flist->ndx_start];
552 if (file->flags & (FLAG_HLINK_DONE|FLAG_FILE_SENT))
553 return -1;
554 file->flags |= FLAG_HLINK_LAST;
555 *flist_p = flist;
556 }
557
558 return prev_ndx;
559}
560#endif