We don't disable inc_recurse due to -H anymore.
[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) 2004-2007 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 am_sender;
28extern int inc_recurse;
29extern int do_xfers;
30extern int link_dest;
31extern int preserve_acls;
32extern int make_backups;
33extern int protocol_version;
34extern int remove_source_files;
35extern int stdout_format_has_i;
36extern int maybe_ATTRS_REPORT;
37extern char *basis_dir[];
38extern struct file_list *cur_flist;
39#ifdef ICONV_OPTION
40extern int ic_ndx;
41#endif
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 node->data = new_array0(char, 5);
124 assert(gnum >= hlink_flist->ndx_start);
125 }
126 }
127 if (gnum >= hlink_flist->ndx_start) {
128 file->flags |= FLAG_HLINK_FIRST;
129 prev = -1;
130 } else if (CVAL(node->data, 0) == 0) {
131 struct file_list *flist;
132 prev = IVAL(node->data, 1);
133 flist = flist_for_ndx(prev);
134 assert(flist != NULL);
135 file_next = flist->files[prev - flist->ndx_start];
136 file_next->flags &= ~FLAG_HLINK_LAST;
137 } else
138 prev = -1;
139 for ( ; from < ndx_count-1; file = file_next, gnum = gnum_next, from++) { /*SHARED ITERATOR*/
140 file_next = hlink_flist->sorted[ndx_list[from+1]];
141 gnum_next = F_HL_GNUM(file_next);
142 if (gnum != gnum_next)
143 break;
144 F_HL_PREV(file) = prev;
145 /* The linked list must use raw ndx values. */
146#ifdef ICONV_OPTION
147 if (ic_ndx)
148 prev = F_NDX(file);
149 else
150#endif
151 prev = ndx_list[from] + hlink_flist->ndx_start;
152 }
153 if (prev < 0 && !inc_recurse) {
154 file->flags &= ~(FLAG_HLINKED | FLAG_HLINK_FIRST);
155 continue;
156 }
157
158 file->flags |= FLAG_HLINK_LAST;
159 F_HL_PREV(file) = prev;
160 if (inc_recurse && CVAL(node->data, 0) == 0) {
161#ifdef ICONV_OPTION
162 if (ic_ndx)
163 prev = F_NDX(file);
164 else
165#endif
166 prev = ndx_list[from] + hlink_flist->ndx_start;
167 SIVAL(node->data, 1, prev);
168 }
169 }
170}
171
172/* Analyze the hard-links in the file-list by creating a list of all the
173 * items that have hlink data, sorting them, and matching up identical
174 * values into clusters. These will be a single linked list from last
175 * to first when we're done. */
176void match_hard_links(struct file_list *flist)
177{
178 int i, ndx_count = 0;
179 int32 *ndx_list;
180
181 if (!(ndx_list = new_array(int32, flist->used)))
182 out_of_memory("match_hard_links");
183
184 for (i = 0; i < flist->used; i++) {
185 if (F_IS_HLINKED(flist->sorted[i]))
186 ndx_list[ndx_count++] = i;
187 }
188
189 hlink_flist = flist;
190
191 if (ndx_count)
192 match_gnums(ndx_list, ndx_count);
193
194 free(ndx_list);
195 if (protocol_version < 30)
196 idev_destroy();
197}
198
199static int maybe_hard_link(struct file_struct *file, int ndx,
200 const char *fname, int statret, statx *sxp,
201 const char *oldname, STRUCT_STAT *old_stp,
202 const char *realname, int itemizing, enum logcode code)
203{
204 if (statret == 0) {
205 if (sxp->st.st_dev == old_stp->st_dev
206 && sxp->st.st_ino == old_stp->st_ino) {
207 if (itemizing) {
208 itemize(fname, file, ndx, statret, sxp,
209 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS,
210 0, "");
211 }
212 if (verbose > 1 && maybe_ATTRS_REPORT)
213 rprintf(FCLIENT, "%s is uptodate\n", fname);
214 file->flags |= FLAG_HLINK_DONE;
215 return 0;
216 }
217 if (make_backups > 0) {
218 if (!make_backup(fname))
219 return -1;
220 } else if (robust_unlink(fname)) {
221 rsyserr(FERROR, errno, "unlink %s failed",
222 full_fname(fname));
223 return -1;
224 }
225 }
226
227 if (hard_link_one(file, fname, oldname, 0)) {
228 if (itemizing) {
229 itemize(fname, file, ndx, statret, sxp,
230 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, 0,
231 realname);
232 }
233 if (code != FNONE && verbose)
234 rprintf(code, "%s => %s\n", fname, realname);
235 return 0;
236 }
237 return -1;
238}
239
240/* Figure out if a prior entry is still there or if we just have a
241 * cached name for it. Never called with a FLAG_HLINK_FIRST entry. */
242static char *check_prior(int prev_ndx, int gnum)
243{
244 struct file_list *flist = flist_for_ndx(prev_ndx);
245 struct ht_int32_node *node;
246
247 if (flist) {
248 hlink_flist = flist;
249 return NULL;
250 }
251
252 node = hashtable_find(prior_hlinks, gnum, 0);
253 assert(node != NULL && node->data);
254 assert(CVAL(node->data, 0) != 0);
255 return node->data;
256}
257
258/* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not. Returns:
259 * 0 = process the file, 1 = skip the file, -1 = error occurred. */
260int hard_link_check(struct file_struct *file, int ndx, const char *fname,
261 int statret, statx *sxp, int itemizing,
262 enum logcode code)
263{
264 STRUCT_STAT prev_st;
265 char namebuf[MAXPATHLEN], altbuf[MAXPATHLEN];
266 char *realname, *prev_name;
267 int gnum = F_HL_GNUM(file);
268 int prev_ndx = F_HL_PREV(file);
269
270 prev_name = realname = check_prior(prev_ndx, gnum);
271
272 if (!prev_name) {
273 struct file_struct *prev_file = hlink_flist->files[prev_ndx - hlink_flist->ndx_start];
274
275 /* Is the previous link is not complete yet? */
276 if (!(prev_file->flags & FLAG_HLINK_DONE)) {
277 /* Is the previous link being transferred? */
278 if (prev_file->flags & FLAG_FILE_SENT) {
279 /* Add ourselves to the list of files that will be
280 * updated when the transfer completes, and mark
281 * ourself as waiting for the transfer. */
282 F_HL_PREV(file) = F_HL_PREV(prev_file);
283 F_HL_PREV(prev_file) = ndx;
284 file->flags |= FLAG_FILE_SENT;
285 cur_flist->in_progress++;
286 return 1;
287 }
288 return 0;
289 }
290
291 /* There is a finished file to link with! */
292 if (!(prev_file->flags & FLAG_HLINK_FIRST)) {
293 /* The previous previous will be marked with FIRST. */
294 prev_ndx = F_HL_PREV(prev_file);
295 prev_name = realname = check_prior(prev_ndx, gnum);
296 /* Update our previous pointer to point to the first. */
297 F_HL_PREV(file) = prev_ndx;
298 }
299 }
300 if (!prev_name) {
301 struct file_struct *prev_file = hlink_flist->files[prev_ndx - hlink_flist->ndx_start];
302 int alt_dest = F_HL_PREV(prev_file); /* alternate value when DONE && FIRST */
303
304 if (alt_dest >= 0 && dry_run) {
305 pathjoin(namebuf, MAXPATHLEN, basis_dir[alt_dest],
306 f_name(prev_file, NULL));
307 prev_name = namebuf;
308 realname = f_name(prev_file, altbuf);
309 } else {
310 prev_name = f_name(prev_file, namebuf);
311 realname = prev_name;
312 }
313 }
314
315 if (link_stat(prev_name, &prev_st, 0) < 0) {
316 rsyserr(FERROR, errno, "stat %s failed",
317 full_fname(prev_name));
318 return -1;
319 }
320
321 if (statret < 0 && basis_dir[0] != NULL) {
322 /* If we match an alt-dest item, we don't output this as a change. */
323 char cmpbuf[MAXPATHLEN];
324 statx alt_sx;
325 int j = 0;
326#ifdef SUPPORT_ACLS
327 alt_sx.acc_acl = alt_sx.def_acl = NULL;
328#endif
329 do {
330 pathjoin(cmpbuf, MAXPATHLEN, basis_dir[j], fname);
331 if (link_stat(cmpbuf, &alt_sx.st, 0) < 0)
332 continue;
333 if (link_dest) {
334 if (prev_st.st_dev != alt_sx.st.st_dev
335 || prev_st.st_ino != alt_sx.st.st_ino)
336 continue;
337 statret = 1;
338 if (verbose < 2 || !stdout_format_has_i) {
339 itemizing = 0;
340 code = FNONE;
341 if (verbose > 1 && maybe_ATTRS_REPORT)
342 rprintf(FCLIENT, "%s is uptodate\n", fname);
343 }
344 break;
345 }
346 if (!unchanged_file(cmpbuf, file, &alt_sx.st))
347 continue;
348 statret = 1;
349 if (unchanged_attrs(cmpbuf, file, &alt_sx))
350 break;
351 } while (basis_dir[++j] != NULL);
352 if (statret == 1) {
353 sxp->st = alt_sx.st;
354#ifdef SUPPORT_ACLS
355 if (preserve_acls && !S_ISLNK(file->mode)) {
356 if (!ACL_READY(*sxp))
357 get_acl(cmpbuf, sxp);
358 else {
359 sxp->acc_acl = alt_sx.acc_acl;
360 sxp->def_acl = alt_sx.def_acl;
361 }
362 }
363#endif
364 }
365#ifdef SUPPORT_ACLS
366 else if (preserve_acls)
367 free_acl(&alt_sx);
368#endif
369 }
370
371 if (maybe_hard_link(file, ndx, fname, statret, sxp, prev_name, &prev_st,
372 realname, itemizing, code) < 0)
373 return -1;
374
375 if (remove_source_files == 1 && do_xfers)
376 send_msg_int(MSG_SUCCESS, ndx);
377
378 return 1;
379}
380
381int hard_link_one(struct file_struct *file, const char *fname,
382 const char *oldname, int terse)
383{
384 if (do_link(oldname, fname) < 0) {
385 enum logcode code;
386 if (terse) {
387 if (!verbose)
388 return -1;
389 code = FINFO;
390 } else
391 code = FERROR;
392 rsyserr(code, errno, "link %s => %s failed",
393 full_fname(fname), oldname);
394 return 0;
395 }
396
397 file->flags |= FLAG_HLINK_DONE;
398
399 return 1;
400}
401
402void finish_hard_link(struct file_struct *file, const char *fname,
403 STRUCT_STAT *stp, int itemizing, enum logcode code,
404 int alt_dest)
405{
406 statx prev_sx;
407 STRUCT_STAT st;
408 char alt_name[MAXPATHLEN], *prev_name;
409 const char *our_name;
410 struct file_list *flist;
411 int prev_statret, ndx, prev_ndx = F_HL_PREV(file);
412
413 if (stp == NULL && prev_ndx >= 0) {
414 if (link_stat(fname, &st, 0) < 0) {
415 rsyserr(FERROR, errno, "stat %s failed",
416 full_fname(fname));
417 return;
418 }
419 stp = &st;
420 }
421
422 /* FIRST combined with DONE means we were the first to get done. */
423 file->flags |= FLAG_HLINK_FIRST | FLAG_HLINK_DONE;
424 F_HL_PREV(file) = alt_dest;
425 if (alt_dest >= 0 && dry_run) {
426 pathjoin(alt_name, MAXPATHLEN, basis_dir[alt_dest],
427 f_name(file, NULL));
428 our_name = alt_name;
429 } else
430 our_name = fname;
431
432#ifdef SUPPORT_ACLS
433 prev_sx.acc_acl = prev_sx.def_acl = NULL;
434#endif
435
436 while ((ndx = prev_ndx) >= 0) {
437 int val;
438 flist = flist_for_ndx(ndx);
439 assert(flist != NULL);
440 file = flist->files[ndx - flist->ndx_start];
441 file->flags = (file->flags & ~FLAG_HLINK_FIRST) | FLAG_HLINK_DONE;
442 prev_ndx = F_HL_PREV(file);
443 prev_name = f_name(file, NULL);
444 prev_statret = link_stat(prev_name, &prev_sx.st, 0);
445 val = maybe_hard_link(file, ndx, prev_name, prev_statret, &prev_sx,
446 our_name, stp, fname, itemizing, code);
447 flist->in_progress--;
448#ifdef SUPPORT_ACLS
449 if (preserve_acls)
450 free_acl(&prev_sx);
451#endif
452 if (val < 0)
453 continue;
454 if (remove_source_files == 1 && do_xfers)
455 send_msg_int(MSG_SUCCESS, ndx);
456 }
457
458 if (inc_recurse) {
459 int gnum = F_HL_GNUM(file);
460 struct ht_int32_node *node = hashtable_find(prior_hlinks, gnum, 0);
461 assert(node != NULL && node->data != NULL);
462 assert(CVAL(node->data, 0) == 0);
463 free(node->data);
464 if (!(node->data = strdup(our_name)))
465 out_of_memory("finish_hard_link");
466 }
467}
468#endif