Added a comment to a shared iterator to avoid a warning from IBM's checker.
[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 version 2 as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License along
19 * with this program; if not, write to the Free Software Foundation, Inc.,
20 * 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA.
21 */
22
23#include "rsync.h"
24
25extern int verbose;
26extern int dry_run;
27extern int do_xfers;
28extern int link_dest;
29extern int preserve_acls;
30extern int make_backups;
31extern int protocol_version;
32extern int remove_source_files;
33extern int stdout_format_has_i;
34extern int maybe_ATTRS_REPORT;
35extern char *basis_dir[];
36extern struct file_list *cur_flist;
37#ifdef ICONV_OPTION
38extern int ic_ndx;
39#endif
40
41#ifdef SUPPORT_HARD_LINKS
42
43#define HASH_LOAD_LIMIT(size) ((size)*3/4)
44
45struct ihash_table {
46 int32 size;
47 int32 entries;
48 struct idev_node *buckets;
49} *dev_tbl;
50
51static struct idev_node *ihash_node(struct ihash_table *tbl, int64 key);
52
53/* Starting with protocol 30, we use a simple hashtable on the sending side
54 * for hashing the st_dev and st_ino info. The receiving side gets told
55 * (via flags and a "group index") which items are hard-linked together, so
56 * we can avoid the pool of dev+inode data. */
57
58static struct ihash_table *ihash_create(int size)
59{
60 struct ihash_table *tbl;
61
62 /* Pick a power of 2 that can hold the requested size. */
63 if (size & (size-1) || size < 16) {
64 int req = size;
65 size = 16;
66 while (size < req)
67 size *= 2;
68 }
69
70 if (!(tbl = new(struct ihash_table))
71 || !(tbl->buckets = new_array(struct idev_node, size)))
72 out_of_memory("ihash_create");
73 memset(tbl->buckets, 0, size * sizeof tbl->buckets[0]);
74 tbl->size = size;
75 tbl->entries = 0;
76
77 return tbl;
78}
79
80static void ihash_destroy(struct ihash_table *tbl)
81{
82 free(tbl->buckets);
83 free(tbl);
84}
85
86void init_hard_links(void)
87{
88 dev_tbl = ihash_create(16);
89}
90
91static void expand_ihash(struct ihash_table *tbl)
92{
93 struct idev_node *old_buckets = tbl->buckets;
94 int size = tbl->size * 2;
95 int i;
96
97 if (!(tbl->buckets = new_array(struct idev_node, size)))
98 out_of_memory("ihash_create");
99 memset(tbl->buckets, 0, size * sizeof (struct idev_node));
100 tbl->size = size;
101 tbl->entries = 0;
102
103 for (i = size / 2; i-- > 0; ) {
104 int64 key = old_buckets[i].key;
105 if (key == 0)
106 continue;
107 ihash_node(tbl, key)->data = old_buckets[i].data;
108 }
109
110 free(old_buckets);
111}
112
113/* This returns the node for the indicated key, either newly created,
114 * or already existing. */
115static struct idev_node *ihash_node(struct ihash_table *tbl, int64 key)
116{
117 uint32 bkt;
118
119 if (tbl->entries > HASH_LOAD_LIMIT(tbl->size))
120 expand_ihash(tbl);
121
122#if SIZEOF_INT64 < 8
123 /* Based on Jenkins One-at-a-time hash. */
124 {
125 uchar *keyp = (uchar*)&key;
126 int i;
127
128 for (bkt = 0, i = 0; i < SIZEOF_INT64; i++) {
129 bkt += keyp[i];
130 bkt += (bkt << 10);
131 bkt ^= (bkt >> 6);
132 }
133 bkt += (bkt << 3);
134 bkt ^= (bkt >> 11);
135 bkt += (bkt << 15);
136 }
137#else
138#define rot(x,k) (((x)<<(k)) ^ ((x)>>(32-(k))))
139 /* Based on Jenkins hashword() from lookup3.c. */
140 {
141 uint32 a, b, c;
142
143 /* Set up the internal state */
144 a = b = c = 0xdeadbeef + (8 << 2);
145
146 b += (uint32)(key >> 32);
147 a += (uint32)key;
148 c ^= b; c -= rot(b, 14);
149 a ^= c; a -= rot(c, 11);
150 b ^= a; b -= rot(a, 25);
151 c ^= b; c -= rot(b, 16);
152 a ^= c; a -= rot(c, 4);
153 b ^= a; b -= rot(a, 14);
154 c ^= b; c -= rot(b, 24);
155 bkt = c;
156 }
157#endif
158
159 /* If it already exists, return it. */
160 while (1) {
161 bkt &= tbl->size - 1;
162 if (tbl->buckets[bkt].key == key)
163 return &tbl->buckets[bkt];
164 if (tbl->buckets[bkt].key == 0)
165 break;
166 bkt++;
167 }
168
169 /* Otherwise, take over this empty spot and then return it. */
170 tbl->buckets[bkt].key = key;
171 tbl->entries++;
172 return &tbl->buckets[bkt];
173}
174
175struct idev_node *idev_node(int64 dev, int64 ino)
176{
177 static struct idev_node *dev_node = NULL;
178 struct ihash_table *tbl;
179
180 if (!dev_node || dev_node->key != dev) {
181 /* We keep a separate hash table of inodes for every device. */
182 dev_node = ihash_node(dev_tbl, dev);
183 if (!(tbl = dev_node->data))
184 tbl = dev_node->data = ihash_create(512);
185 } else
186 tbl = dev_node->data;
187
188 return ihash_node(tbl, ino);
189}
190
191void idev_destroy(void)
192{
193 int i;
194
195 for (i = 0; i < dev_tbl->size; i++) {
196 if (dev_tbl->buckets[i].data)
197 ihash_destroy(dev_tbl->buckets[i].data);
198 }
199
200 ihash_destroy(dev_tbl);
201}
202
203static int hlink_compare_gnum(int *int1, int *int2)
204{
205 struct file_struct *f1 = cur_flist->sorted[*int1];
206 struct file_struct *f2 = cur_flist->sorted[*int2];
207 int32 gnum1 = F_HL_GNUM(f1);
208 int32 gnum2 = F_HL_GNUM(f2);
209
210 if (gnum1 != gnum2)
211 return gnum1 > gnum2 ? 1 : -1;
212
213 return *int1 > *int2 ? 1 : -1;
214}
215
216static void match_gnums(int32 *ndx_list, int ndx_count)
217{
218 int32 from, prev;
219 struct file_struct *file, *file_next;
220 int32 gnum, gnum_next;
221
222 qsort(ndx_list, ndx_count, sizeof ndx_list[0],
223 (int (*)()) hlink_compare_gnum);
224
225 for (from = 0; from < ndx_count; from++) {
226 for (file = cur_flist->sorted[ndx_list[from]], gnum = F_HL_GNUM(file), prev = -1;
227 from < ndx_count-1;
228 file = file_next, gnum = gnum_next, from++) /*SHARED ITERATOR*/
229 {
230 file_next = cur_flist->sorted[ndx_list[from+1]];
231 gnum_next = F_HL_GNUM(file_next);
232 if (gnum != gnum_next)
233 break;
234 if (prev < 0)
235 file->flags |= FLAG_HLINK_FIRST;
236 F_HL_PREV(file) = prev;
237 /* The linked list must use raw ndx values. */
238#ifdef ICONV_OPTION
239 if (ic_ndx)
240 prev = F_NDX(file);
241 else
242#endif
243 prev = ndx_list[from];
244 }
245 if (prev < 0)
246 file->flags &= ~FLAG_HLINKED;
247 else {
248 file->flags |= FLAG_HLINK_LAST;
249 F_HL_PREV(file) = prev;
250 }
251 }
252}
253
254/* Analyze the hard-links in the file-list by creating a list of all the
255 * items that have hlink data, sorting them, and matching up identical
256 * values into clusters. These will be a single linked list from last
257 * to first when we're done. */
258void match_hard_links(void)
259{
260 int i, ndx_count = 0;
261 int32 *ndx_list;
262
263 if (!(ndx_list = new_array(int32, cur_flist->count)))
264 out_of_memory("match_hard_links");
265
266 for (i = 0; i < cur_flist->count; i++) {
267 if (F_IS_HLINKED(cur_flist->sorted[i]))
268 ndx_list[ndx_count++] = i;
269 }
270
271 if (ndx_count)
272 match_gnums(ndx_list, ndx_count);
273
274 free(ndx_list);
275 if (protocol_version < 30)
276 idev_destroy();
277}
278
279static int maybe_hard_link(struct file_struct *file, int ndx,
280 const char *fname, int statret, statx *sxp,
281 const char *oldname, STRUCT_STAT *old_stp,
282 const char *realname, int itemizing, enum logcode code)
283{
284 if (statret == 0) {
285 if (sxp->st.st_dev == old_stp->st_dev
286 && sxp->st.st_ino == old_stp->st_ino) {
287 if (itemizing) {
288 itemize(fname, file, ndx, statret, sxp,
289 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS,
290 0, "");
291 }
292 if (verbose > 1 && maybe_ATTRS_REPORT)
293 rprintf(FCLIENT, "%s is uptodate\n", fname);
294 file->flags |= FLAG_HLINK_DONE;
295 return 0;
296 }
297 if (make_backups > 0) {
298 if (!make_backup(fname))
299 return -1;
300 } else if (robust_unlink(fname)) {
301 rsyserr(FERROR, errno, "unlink %s failed",
302 full_fname(fname));
303 return -1;
304 }
305 }
306
307 if (hard_link_one(file, fname, oldname, 0)) {
308 if (itemizing) {
309 itemize(fname, file, ndx, statret, sxp,
310 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, 0,
311 realname);
312 }
313 if (code != FNONE && verbose)
314 rprintf(code, "%s => %s\n", fname, realname);
315 return 0;
316 }
317 return -1;
318}
319
320/* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not. Returns:
321 * 0 = process the file, 1 = skip the file, -1 = error occurred. */
322int hard_link_check(struct file_struct *file, int ndx, const char *fname,
323 int statret, statx *sxp, int itemizing,
324 enum logcode code)
325{
326 STRUCT_STAT prev_st;
327 char prev_name[MAXPATHLEN], altbuf[MAXPATHLEN], *realname;
328 int alt_dest, prev_ndx = F_HL_PREV(file);
329 struct file_struct *prev_file = cur_flist->files[prev_ndx];
330
331 /* Is the previous link is not complete yet? */
332 if (!(prev_file->flags & FLAG_HLINK_DONE)) {
333 /* Is the previous link being transferred? */
334 if (prev_file->flags & FLAG_FILE_SENT) {
335 /* Add ourselves to the list of files that will be
336 * updated when the transfer completes, and mark
337 * ourself as waiting for the transfer. */
338 F_HL_PREV(file) = F_HL_PREV(prev_file);
339 F_HL_PREV(prev_file) = ndx;
340 file->flags |= FLAG_FILE_SENT;
341 return 1;
342 }
343 return 0;
344 }
345
346 /* There is a finished file to link with! */
347 if (!(prev_file->flags & FLAG_HLINK_FIRST)) {
348 /* The previous previous will be marked with FIRST. */
349 prev_ndx = F_HL_PREV(prev_file);
350 prev_file = cur_flist->files[prev_ndx];
351 /* Update our previous pointer to point to the first. */
352 F_HL_PREV(file) = prev_ndx;
353 }
354 alt_dest = F_HL_PREV(prev_file); /* alternate value when DONE && FIRST */
355 if (alt_dest >= 0 && dry_run) {
356 pathjoin(prev_name, MAXPATHLEN, basis_dir[alt_dest],
357 f_name(prev_file, NULL));
358 f_name(prev_file, altbuf);
359 realname = altbuf;
360 } else {
361 f_name(prev_file, prev_name);
362 realname = prev_name;
363 }
364
365 if (link_stat(prev_name, &prev_st, 0) < 0) {
366 rsyserr(FERROR, errno, "stat %s failed",
367 full_fname(prev_name));
368 return -1;
369 }
370
371 if (statret < 0 && basis_dir[0] != NULL) {
372 /* If we match an alt-dest item, we don't output this as a change. */
373 char cmpbuf[MAXPATHLEN];
374 statx alt_sx;
375 int j = 0;
376#ifdef SUPPORT_ACLS
377 alt_sx.acc_acl = alt_sx.def_acl = NULL;
378#endif
379 do {
380 pathjoin(cmpbuf, MAXPATHLEN, basis_dir[j], fname);
381 if (link_stat(cmpbuf, &alt_sx.st, 0) < 0)
382 continue;
383 if (link_dest) {
384 if (prev_st.st_dev != alt_sx.st.st_dev
385 || prev_st.st_ino != alt_sx.st.st_ino)
386 continue;
387 statret = 1;
388 if (verbose < 2 || !stdout_format_has_i) {
389 itemizing = 0;
390 code = FNONE;
391 if (verbose > 1 && maybe_ATTRS_REPORT)
392 rprintf(FCLIENT, "%s is uptodate\n", fname);
393 }
394 break;
395 }
396 if (!unchanged_file(cmpbuf, file, &alt_sx.st))
397 continue;
398 statret = 1;
399 if (unchanged_attrs(cmpbuf, file, &alt_sx))
400 break;
401 } while (basis_dir[++j] != NULL);
402 if (statret == 1) {
403 sxp->st = alt_sx.st;
404#ifdef SUPPORT_ACLS
405 if (preserve_acls && !S_ISLNK(file->mode)) {
406 if (!ACL_READY(*sxp))
407 get_acl(cmpbuf, sxp);
408 else {
409 sxp->acc_acl = alt_sx.acc_acl;
410 sxp->def_acl = alt_sx.def_acl;
411 }
412 }
413#endif
414 }
415#ifdef SUPPORT_ACLS
416 else if (preserve_acls)
417 free_acl(&alt_sx);
418#endif
419 }
420
421 if (maybe_hard_link(file, ndx, fname, statret, sxp, prev_name, &prev_st,
422 realname, itemizing, code) < 0)
423 return -1;
424
425 if (remove_source_files == 1 && do_xfers)
426 send_msg_int(MSG_SUCCESS, ndx);
427
428 return 1;
429}
430
431int hard_link_one(struct file_struct *file, const char *fname,
432 const char *oldname, int terse)
433{
434 if (do_link(oldname, fname) < 0) {
435 enum logcode code;
436 if (terse) {
437 if (!verbose)
438 return -1;
439 code = FINFO;
440 } else
441 code = FERROR;
442 rsyserr(code, errno, "link %s => %s failed",
443 full_fname(fname), oldname);
444 return 0;
445 }
446
447 file->flags |= FLAG_HLINK_DONE;
448
449 return 1;
450}
451
452void finish_hard_link(struct file_struct *file, const char *fname,
453 STRUCT_STAT *stp, int itemizing, enum logcode code,
454 int alt_dest)
455{
456 statx prev_sx;
457 STRUCT_STAT st;
458 char alt_name[MAXPATHLEN], *prev_name;
459 const char *our_name;
460 int prev_statret, ndx, prev_ndx = F_HL_PREV(file);
461
462 if (stp == NULL && prev_ndx >= 0) {
463 if (link_stat(fname, &st, 0) < 0) {
464 rsyserr(FERROR, errno, "stat %s failed",
465 full_fname(fname));
466 return;
467 }
468 stp = &st;
469 }
470
471 /* FIRST combined with DONE means we were the first to get done. */
472 file->flags |= FLAG_HLINK_FIRST | FLAG_HLINK_DONE;
473 F_HL_PREV(file) = alt_dest;
474 if (alt_dest >= 0 && dry_run) {
475 pathjoin(alt_name, MAXPATHLEN, basis_dir[alt_dest],
476 f_name(file, NULL));
477 our_name = alt_name;
478 } else
479 our_name = fname;
480
481#ifdef SUPPORT_ACLS
482 prev_sx.acc_acl = prev_sx.def_acl = NULL;
483#endif
484
485 while ((ndx = prev_ndx) >= 0) {
486 int val;
487 file = cur_flist->files[ndx];
488 file->flags = (file->flags & ~FLAG_HLINK_FIRST) | FLAG_HLINK_DONE;
489 prev_ndx = F_HL_PREV(file);
490 prev_name = f_name(file, NULL);
491 prev_statret = link_stat(prev_name, &prev_sx.st, 0);
492 val = maybe_hard_link(file, ndx, prev_name, prev_statret, &prev_sx,
493 our_name, stp, fname, itemizing, code);
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#endif