Optimize "diff --cached" performance.
[git/git.git] / read-cache.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "cache-tree.h"
9 #include "refs.h"
10
11 /* Index extensions.
12 *
13 * The first letter should be 'A'..'Z' for extensions that are not
14 * necessary for a correct operation (i.e. optimization data).
15 * When new extensions are added that _needs_ to be understood in
16 * order to correctly interpret the index file, pick character that
17 * is outside the range, to cause the reader to abort.
18 */
19
20 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
21 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
22
23 struct index_state the_index;
24
25 /*
26 * This only updates the "non-critical" parts of the directory
27 * cache, ie the parts that aren't tracked by GIT, and only used
28 * to validate the cache.
29 */
30 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
31 {
32 ce->ce_ctime.sec = htonl(st->st_ctime);
33 ce->ce_mtime.sec = htonl(st->st_mtime);
34 #ifdef USE_NSEC
35 ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
36 ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
37 #endif
38 ce->ce_dev = htonl(st->st_dev);
39 ce->ce_ino = htonl(st->st_ino);
40 ce->ce_uid = htonl(st->st_uid);
41 ce->ce_gid = htonl(st->st_gid);
42 ce->ce_size = htonl(st->st_size);
43
44 if (assume_unchanged)
45 ce->ce_flags |= htons(CE_VALID);
46 }
47
48 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
49 {
50 int match = -1;
51 int fd = open(ce->name, O_RDONLY);
52
53 if (fd >= 0) {
54 unsigned char sha1[20];
55 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
56 match = hashcmp(sha1, ce->sha1);
57 /* index_fd() closed the file descriptor already */
58 }
59 return match;
60 }
61
62 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
63 {
64 int match = -1;
65 char *target;
66 void *buffer;
67 unsigned long size;
68 enum object_type type;
69 int len;
70
71 target = xmalloc(expected_size);
72 len = readlink(ce->name, target, expected_size);
73 if (len != expected_size) {
74 free(target);
75 return -1;
76 }
77 buffer = read_sha1_file(ce->sha1, &type, &size);
78 if (!buffer) {
79 free(target);
80 return -1;
81 }
82 if (size == expected_size)
83 match = memcmp(buffer, target, size);
84 free(buffer);
85 free(target);
86 return match;
87 }
88
89 static int ce_compare_gitlink(struct cache_entry *ce)
90 {
91 unsigned char sha1[20];
92
93 /*
94 * We don't actually require that the .git directory
95 * under GITLINK directory be a valid git directory. It
96 * might even be missing (in case nobody populated that
97 * sub-project).
98 *
99 * If so, we consider it always to match.
100 */
101 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
102 return 0;
103 return hashcmp(sha1, ce->sha1);
104 }
105
106 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
107 {
108 switch (st->st_mode & S_IFMT) {
109 case S_IFREG:
110 if (ce_compare_data(ce, st))
111 return DATA_CHANGED;
112 break;
113 case S_IFLNK:
114 if (ce_compare_link(ce, xsize_t(st->st_size)))
115 return DATA_CHANGED;
116 break;
117 case S_IFDIR:
118 if (S_ISGITLINK(ntohl(ce->ce_mode)))
119 return 0;
120 default:
121 return TYPE_CHANGED;
122 }
123 return 0;
124 }
125
126 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
127 {
128 unsigned int changed = 0;
129
130 switch (ntohl(ce->ce_mode) & S_IFMT) {
131 case S_IFREG:
132 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
133 /* We consider only the owner x bit to be relevant for
134 * "mode changes"
135 */
136 if (trust_executable_bit &&
137 (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
138 changed |= MODE_CHANGED;
139 break;
140 case S_IFLNK:
141 if (!S_ISLNK(st->st_mode) &&
142 (has_symlinks || !S_ISREG(st->st_mode)))
143 changed |= TYPE_CHANGED;
144 break;
145 case S_IFGITLINK:
146 if (!S_ISDIR(st->st_mode))
147 changed |= TYPE_CHANGED;
148 else if (ce_compare_gitlink(ce))
149 changed |= DATA_CHANGED;
150 return changed;
151 default:
152 die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
153 }
154 if (ce->ce_mtime.sec != htonl(st->st_mtime))
155 changed |= MTIME_CHANGED;
156 if (ce->ce_ctime.sec != htonl(st->st_ctime))
157 changed |= CTIME_CHANGED;
158
159 #ifdef USE_NSEC
160 /*
161 * nsec seems unreliable - not all filesystems support it, so
162 * as long as it is in the inode cache you get right nsec
163 * but after it gets flushed, you get zero nsec.
164 */
165 if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
166 changed |= MTIME_CHANGED;
167 if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
168 changed |= CTIME_CHANGED;
169 #endif
170
171 if (ce->ce_uid != htonl(st->st_uid) ||
172 ce->ce_gid != htonl(st->st_gid))
173 changed |= OWNER_CHANGED;
174 if (ce->ce_ino != htonl(st->st_ino))
175 changed |= INODE_CHANGED;
176
177 #ifdef USE_STDEV
178 /*
179 * st_dev breaks on network filesystems where different
180 * clients will have different views of what "device"
181 * the filesystem is on
182 */
183 if (ce->ce_dev != htonl(st->st_dev))
184 changed |= INODE_CHANGED;
185 #endif
186
187 if (ce->ce_size != htonl(st->st_size))
188 changed |= DATA_CHANGED;
189
190 return changed;
191 }
192
193 int ie_match_stat(struct index_state *istate,
194 struct cache_entry *ce, struct stat *st, int options)
195 {
196 unsigned int changed;
197 int ignore_valid = options & 01;
198 int assume_racy_is_modified = options & 02;
199
200 /*
201 * If it's marked as always valid in the index, it's
202 * valid whatever the checked-out copy says.
203 */
204 if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
205 return 0;
206
207 changed = ce_match_stat_basic(ce, st);
208
209 /*
210 * Within 1 second of this sequence:
211 * echo xyzzy >file && git-update-index --add file
212 * running this command:
213 * echo frotz >file
214 * would give a falsely clean cache entry. The mtime and
215 * length match the cache, and other stat fields do not change.
216 *
217 * We could detect this at update-index time (the cache entry
218 * being registered/updated records the same time as "now")
219 * and delay the return from git-update-index, but that would
220 * effectively mean we can make at most one commit per second,
221 * which is not acceptable. Instead, we check cache entries
222 * whose mtime are the same as the index file timestamp more
223 * carefully than others.
224 */
225 if (!changed &&
226 istate->timestamp &&
227 istate->timestamp <= ntohl(ce->ce_mtime.sec)) {
228 if (assume_racy_is_modified)
229 changed |= DATA_CHANGED;
230 else
231 changed |= ce_modified_check_fs(ce, st);
232 }
233
234 return changed;
235 }
236
237 int ie_modified(struct index_state *istate,
238 struct cache_entry *ce, struct stat *st, int really)
239 {
240 int changed, changed_fs;
241 changed = ie_match_stat(istate, ce, st, really);
242 if (!changed)
243 return 0;
244 /*
245 * If the mode or type has changed, there's no point in trying
246 * to refresh the entry - it's not going to match
247 */
248 if (changed & (MODE_CHANGED | TYPE_CHANGED))
249 return changed;
250
251 /* Immediately after read-tree or update-index --cacheinfo,
252 * the length field is zero. For other cases the ce_size
253 * should match the SHA1 recorded in the index entry.
254 */
255 if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
256 return changed;
257
258 changed_fs = ce_modified_check_fs(ce, st);
259 if (changed_fs)
260 return changed | changed_fs;
261 return 0;
262 }
263
264 int base_name_compare(const char *name1, int len1, int mode1,
265 const char *name2, int len2, int mode2)
266 {
267 unsigned char c1, c2;
268 int len = len1 < len2 ? len1 : len2;
269 int cmp;
270
271 cmp = memcmp(name1, name2, len);
272 if (cmp)
273 return cmp;
274 c1 = name1[len];
275 c2 = name2[len];
276 if (!c1 && S_ISDIR(mode1))
277 c1 = '/';
278 if (!c2 && S_ISDIR(mode2))
279 c2 = '/';
280 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
281 }
282
283 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
284 {
285 int len1 = flags1 & CE_NAMEMASK;
286 int len2 = flags2 & CE_NAMEMASK;
287 int len = len1 < len2 ? len1 : len2;
288 int cmp;
289
290 cmp = memcmp(name1, name2, len);
291 if (cmp)
292 return cmp;
293 if (len1 < len2)
294 return -1;
295 if (len1 > len2)
296 return 1;
297
298 /* Compare stages */
299 flags1 &= CE_STAGEMASK;
300 flags2 &= CE_STAGEMASK;
301
302 if (flags1 < flags2)
303 return -1;
304 if (flags1 > flags2)
305 return 1;
306 return 0;
307 }
308
309 int index_name_pos(struct index_state *istate, const char *name, int namelen)
310 {
311 int first, last;
312
313 first = 0;
314 last = istate->cache_nr;
315 while (last > first) {
316 int next = (last + first) >> 1;
317 struct cache_entry *ce = istate->cache[next];
318 int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
319 if (!cmp)
320 return next;
321 if (cmp < 0) {
322 last = next;
323 continue;
324 }
325 first = next+1;
326 }
327 return -first-1;
328 }
329
330 /* Remove entry, return true if there are more entries to go.. */
331 int remove_index_entry_at(struct index_state *istate, int pos)
332 {
333 istate->cache_changed = 1;
334 istate->cache_nr--;
335 if (pos >= istate->cache_nr)
336 return 0;
337 memmove(istate->cache + pos,
338 istate->cache + pos + 1,
339 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
340 return 1;
341 }
342
343 int remove_file_from_index(struct index_state *istate, const char *path)
344 {
345 int pos = index_name_pos(istate, path, strlen(path));
346 if (pos < 0)
347 pos = -pos-1;
348 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
349 remove_index_entry_at(istate, pos);
350 return 0;
351 }
352
353 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
354 {
355 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
356 }
357
358 static int index_name_pos_also_unmerged(struct index_state *istate,
359 const char *path, int namelen)
360 {
361 int pos = index_name_pos(istate, path, namelen);
362 struct cache_entry *ce;
363
364 if (pos >= 0)
365 return pos;
366
367 /* maybe unmerged? */
368 pos = -1 - pos;
369 if (pos >= istate->cache_nr ||
370 compare_name((ce = istate->cache[pos]), path, namelen))
371 return -1;
372
373 /* order of preference: stage 2, 1, 3 */
374 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
375 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
376 !compare_name(ce, path, namelen))
377 pos++;
378 return pos;
379 }
380
381 int add_file_to_index(struct index_state *istate, const char *path, int verbose)
382 {
383 int size, namelen, pos;
384 struct stat st;
385 struct cache_entry *ce;
386
387 if (lstat(path, &st))
388 die("%s: unable to stat (%s)", path, strerror(errno));
389
390 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
391 die("%s: can only add regular files, symbolic links or git-directories", path);
392
393 namelen = strlen(path);
394 if (S_ISDIR(st.st_mode)) {
395 while (namelen && path[namelen-1] == '/')
396 namelen--;
397 }
398 size = cache_entry_size(namelen);
399 ce = xcalloc(1, size);
400 memcpy(ce->name, path, namelen);
401 ce->ce_flags = htons(namelen);
402 fill_stat_cache_info(ce, &st);
403
404 if (trust_executable_bit && has_symlinks)
405 ce->ce_mode = create_ce_mode(st.st_mode);
406 else {
407 /* If there is an existing entry, pick the mode bits and type
408 * from it, otherwise assume unexecutable regular file.
409 */
410 struct cache_entry *ent;
411 int pos = index_name_pos_also_unmerged(istate, path, namelen);
412
413 ent = (0 <= pos) ? istate->cache[pos] : NULL;
414 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
415 }
416
417 pos = index_name_pos(istate, ce->name, namelen);
418 if (0 <= pos &&
419 !ce_stage(istate->cache[pos]) &&
420 !ie_modified(istate, istate->cache[pos], &st, 1)) {
421 /* Nothing changed, really */
422 free(ce);
423 return 0;
424 }
425
426 if (index_path(ce->sha1, path, &st, 1))
427 die("unable to index file %s", path);
428 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
429 die("unable to add %s to index",path);
430 if (verbose)
431 printf("add '%s'\n", path);
432 cache_tree_invalidate_path(istate->cache_tree, path);
433 return 0;
434 }
435
436 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
437 {
438 int len = ce_namelen(a);
439 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
440 }
441
442 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
443 {
444 const char *match, *name;
445 int len;
446
447 if (!pathspec)
448 return 1;
449
450 len = ce_namelen(ce);
451 name = ce->name;
452 while ((match = *pathspec++) != NULL) {
453 int matchlen = strlen(match);
454 if (matchlen > len)
455 continue;
456 if (memcmp(name, match, matchlen))
457 continue;
458 if (matchlen && name[matchlen-1] == '/')
459 return 1;
460 if (name[matchlen] == '/' || !name[matchlen])
461 return 1;
462 if (!matchlen)
463 return 1;
464 }
465 return 0;
466 }
467
468 /*
469 * We fundamentally don't like some paths: we don't want
470 * dot or dot-dot anywhere, and for obvious reasons don't
471 * want to recurse into ".git" either.
472 *
473 * Also, we don't want double slashes or slashes at the
474 * end that can make pathnames ambiguous.
475 */
476 static int verify_dotfile(const char *rest)
477 {
478 /*
479 * The first character was '.', but that
480 * has already been discarded, we now test
481 * the rest.
482 */
483 switch (*rest) {
484 /* "." is not allowed */
485 case '\0': case '/':
486 return 0;
487
488 /*
489 * ".git" followed by NUL or slash is bad. This
490 * shares the path end test with the ".." case.
491 */
492 case 'g':
493 if (rest[1] != 'i')
494 break;
495 if (rest[2] != 't')
496 break;
497 rest += 2;
498 /* fallthrough */
499 case '.':
500 if (rest[1] == '\0' || rest[1] == '/')
501 return 0;
502 }
503 return 1;
504 }
505
506 int verify_path(const char *path)
507 {
508 char c;
509
510 goto inside;
511 for (;;) {
512 if (!c)
513 return 1;
514 if (c == '/') {
515 inside:
516 c = *path++;
517 switch (c) {
518 default:
519 continue;
520 case '/': case '\0':
521 break;
522 case '.':
523 if (verify_dotfile(path))
524 continue;
525 }
526 return 0;
527 }
528 c = *path++;
529 }
530 }
531
532 /*
533 * Do we have another file that has the beginning components being a
534 * proper superset of the name we're trying to add?
535 */
536 static int has_file_name(struct index_state *istate,
537 const struct cache_entry *ce, int pos, int ok_to_replace)
538 {
539 int retval = 0;
540 int len = ce_namelen(ce);
541 int stage = ce_stage(ce);
542 const char *name = ce->name;
543
544 while (pos < istate->cache_nr) {
545 struct cache_entry *p = istate->cache[pos++];
546
547 if (len >= ce_namelen(p))
548 break;
549 if (memcmp(name, p->name, len))
550 break;
551 if (ce_stage(p) != stage)
552 continue;
553 if (p->name[len] != '/')
554 continue;
555 if (!ce_stage(p) && !p->ce_mode)
556 continue;
557 retval = -1;
558 if (!ok_to_replace)
559 break;
560 remove_index_entry_at(istate, --pos);
561 }
562 return retval;
563 }
564
565 /*
566 * Do we have another file with a pathname that is a proper
567 * subset of the name we're trying to add?
568 */
569 static int has_dir_name(struct index_state *istate,
570 const struct cache_entry *ce, int pos, int ok_to_replace)
571 {
572 int retval = 0;
573 int stage = ce_stage(ce);
574 const char *name = ce->name;
575 const char *slash = name + ce_namelen(ce);
576
577 for (;;) {
578 int len;
579
580 for (;;) {
581 if (*--slash == '/')
582 break;
583 if (slash <= ce->name)
584 return retval;
585 }
586 len = slash - name;
587
588 pos = index_name_pos(istate, name, ntohs(create_ce_flags(len, stage)));
589 if (pos >= 0) {
590 /*
591 * Found one, but not so fast. This could
592 * be a marker that says "I was here, but
593 * I am being removed". Such an entry is
594 * not a part of the resulting tree, and
595 * it is Ok to have a directory at the same
596 * path.
597 */
598 if (stage || istate->cache[pos]->ce_mode) {
599 retval = -1;
600 if (!ok_to_replace)
601 break;
602 remove_index_entry_at(istate, pos);
603 continue;
604 }
605 }
606 else
607 pos = -pos-1;
608
609 /*
610 * Trivial optimization: if we find an entry that
611 * already matches the sub-directory, then we know
612 * we're ok, and we can exit.
613 */
614 while (pos < istate->cache_nr) {
615 struct cache_entry *p = istate->cache[pos];
616 if ((ce_namelen(p) <= len) ||
617 (p->name[len] != '/') ||
618 memcmp(p->name, name, len))
619 break; /* not our subdirectory */
620 if (ce_stage(p) == stage && (stage || p->ce_mode))
621 /* p is at the same stage as our entry, and
622 * is a subdirectory of what we are looking
623 * at, so we cannot have conflicts at our
624 * level or anything shorter.
625 */
626 return retval;
627 pos++;
628 }
629 }
630 return retval;
631 }
632
633 /* We may be in a situation where we already have path/file and path
634 * is being added, or we already have path and path/file is being
635 * added. Either one would result in a nonsense tree that has path
636 * twice when git-write-tree tries to write it out. Prevent it.
637 *
638 * If ok-to-replace is specified, we remove the conflicting entries
639 * from the cache so the caller should recompute the insert position.
640 * When this happens, we return non-zero.
641 */
642 static int check_file_directory_conflict(struct index_state *istate,
643 const struct cache_entry *ce,
644 int pos, int ok_to_replace)
645 {
646 int retval;
647
648 /*
649 * When ce is an "I am going away" entry, we allow it to be added
650 */
651 if (!ce_stage(ce) && !ce->ce_mode)
652 return 0;
653
654 /*
655 * We check if the path is a sub-path of a subsequent pathname
656 * first, since removing those will not change the position
657 * in the array.
658 */
659 retval = has_file_name(istate, ce, pos, ok_to_replace);
660
661 /*
662 * Then check if the path might have a clashing sub-directory
663 * before it.
664 */
665 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
666 }
667
668 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
669 {
670 int pos;
671 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
672 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
673 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
674
675 pos = index_name_pos(istate, ce->name, ntohs(ce->ce_flags));
676
677 /* existing match? Just replace it. */
678 if (pos >= 0) {
679 istate->cache_changed = 1;
680 istate->cache[pos] = ce;
681 return 0;
682 }
683 pos = -pos-1;
684
685 /*
686 * Inserting a merged entry ("stage 0") into the index
687 * will always replace all non-merged entries..
688 */
689 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
690 while (ce_same_name(istate->cache[pos], ce)) {
691 ok_to_add = 1;
692 if (!remove_index_entry_at(istate, pos))
693 break;
694 }
695 }
696
697 if (!ok_to_add)
698 return -1;
699 if (!verify_path(ce->name))
700 return -1;
701
702 if (!skip_df_check &&
703 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
704 if (!ok_to_replace)
705 return error("'%s' appears as both a file and as a directory",
706 ce->name);
707 pos = index_name_pos(istate, ce->name, ntohs(ce->ce_flags));
708 pos = -pos-1;
709 }
710 return pos + 1;
711 }
712
713 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
714 {
715 int pos;
716
717 if (option & ADD_CACHE_JUST_APPEND)
718 pos = istate->cache_nr;
719 else {
720 int ret;
721 ret = add_index_entry_with_check(istate, ce, option);
722 if (ret <= 0)
723 return ret;
724 pos = ret - 1;
725 }
726
727 /* Make sure the array is big enough .. */
728 if (istate->cache_nr == istate->cache_alloc) {
729 istate->cache_alloc = alloc_nr(istate->cache_alloc);
730 istate->cache = xrealloc(istate->cache,
731 istate->cache_alloc * sizeof(struct cache_entry *));
732 }
733
734 /* Add it in.. */
735 istate->cache_nr++;
736 if (istate->cache_nr > pos + 1)
737 memmove(istate->cache + pos + 1,
738 istate->cache + pos,
739 (istate->cache_nr - pos - 1) * sizeof(ce));
740 istate->cache[pos] = ce;
741 istate->cache_changed = 1;
742 return 0;
743 }
744
745 /*
746 * "refresh" does not calculate a new sha1 file or bring the
747 * cache up-to-date for mode/content changes. But what it
748 * _does_ do is to "re-match" the stat information of a file
749 * with the cache, so that you can refresh the cache for a
750 * file that hasn't been changed but where the stat entry is
751 * out of date.
752 *
753 * For example, you'd want to do this after doing a "git-read-tree",
754 * to link up the stat cache details with the proper files.
755 */
756 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
757 struct cache_entry *ce, int really, int *err)
758 {
759 struct stat st;
760 struct cache_entry *updated;
761 int changed, size;
762
763 if (lstat(ce->name, &st) < 0) {
764 if (err)
765 *err = errno;
766 return NULL;
767 }
768
769 changed = ie_match_stat(istate, ce, &st, really);
770 if (!changed) {
771 if (really && assume_unchanged &&
772 !(ce->ce_flags & htons(CE_VALID)))
773 ; /* mark this one VALID again */
774 else
775 return ce;
776 }
777
778 if (ie_modified(istate, ce, &st, really)) {
779 if (err)
780 *err = EINVAL;
781 return NULL;
782 }
783
784 size = ce_size(ce);
785 updated = xmalloc(size);
786 memcpy(updated, ce, size);
787 fill_stat_cache_info(updated, &st);
788
789 /* In this case, if really is not set, we should leave
790 * CE_VALID bit alone. Otherwise, paths marked with
791 * --no-assume-unchanged (i.e. things to be edited) will
792 * reacquire CE_VALID bit automatically, which is not
793 * really what we want.
794 */
795 if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
796 updated->ce_flags &= ~htons(CE_VALID);
797
798 return updated;
799 }
800
801 int refresh_index(struct index_state *istate, unsigned int flags)
802 {
803 int i;
804 int has_errors = 0;
805 int really = (flags & REFRESH_REALLY) != 0;
806 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
807 int quiet = (flags & REFRESH_QUIET) != 0;
808 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
809
810 for (i = 0; i < istate->cache_nr; i++) {
811 struct cache_entry *ce, *new;
812 int cache_errno = 0;
813
814 ce = istate->cache[i];
815 if (ce_stage(ce)) {
816 while ((i < istate->cache_nr) &&
817 ! strcmp(istate->cache[i]->name, ce->name))
818 i++;
819 i--;
820 if (allow_unmerged)
821 continue;
822 printf("%s: needs merge\n", ce->name);
823 has_errors = 1;
824 continue;
825 }
826
827 new = refresh_cache_ent(istate, ce, really, &cache_errno);
828 if (new == ce)
829 continue;
830 if (!new) {
831 if (not_new && cache_errno == ENOENT)
832 continue;
833 if (really && cache_errno == EINVAL) {
834 /* If we are doing --really-refresh that
835 * means the index is not valid anymore.
836 */
837 ce->ce_flags &= ~htons(CE_VALID);
838 istate->cache_changed = 1;
839 }
840 if (quiet)
841 continue;
842 printf("%s: needs update\n", ce->name);
843 has_errors = 1;
844 continue;
845 }
846 istate->cache_changed = 1;
847 /* You can NOT just free istate->cache[i] here, since it
848 * might not be necessarily malloc()ed but can also come
849 * from mmap(). */
850 istate->cache[i] = new;
851 }
852 return has_errors;
853 }
854
855 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
856 {
857 return refresh_cache_ent(&the_index, ce, really, NULL);
858 }
859
860 static int verify_hdr(struct cache_header *hdr, unsigned long size)
861 {
862 SHA_CTX c;
863 unsigned char sha1[20];
864
865 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
866 return error("bad signature");
867 if (hdr->hdr_version != htonl(2))
868 return error("bad index version");
869 SHA1_Init(&c);
870 SHA1_Update(&c, hdr, size - 20);
871 SHA1_Final(sha1, &c);
872 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
873 return error("bad index file sha1 signature");
874 return 0;
875 }
876
877 static int read_index_extension(struct index_state *istate,
878 const char *ext, void *data, unsigned long sz)
879 {
880 switch (CACHE_EXT(ext)) {
881 case CACHE_EXT_TREE:
882 istate->cache_tree = cache_tree_read(data, sz);
883 break;
884 default:
885 if (*ext < 'A' || 'Z' < *ext)
886 return error("index uses %.4s extension, which we do not understand",
887 ext);
888 fprintf(stderr, "ignoring %.4s extension\n", ext);
889 break;
890 }
891 return 0;
892 }
893
894 int read_index(struct index_state *istate)
895 {
896 return read_index_from(istate, get_index_file());
897 }
898
899 /* remember to discard_cache() before reading a different cache! */
900 int read_index_from(struct index_state *istate, const char *path)
901 {
902 int fd, i;
903 struct stat st;
904 unsigned long offset;
905 struct cache_header *hdr;
906
907 errno = EBUSY;
908 if (istate->mmap)
909 return istate->cache_nr;
910
911 errno = ENOENT;
912 istate->timestamp = 0;
913 fd = open(path, O_RDONLY);
914 if (fd < 0) {
915 if (errno == ENOENT)
916 return 0;
917 die("index file open failed (%s)", strerror(errno));
918 }
919
920 if (fstat(fd, &st))
921 die("cannot stat the open index (%s)", strerror(errno));
922
923 errno = EINVAL;
924 istate->mmap_size = xsize_t(st.st_size);
925 if (istate->mmap_size < sizeof(struct cache_header) + 20)
926 die("index file smaller than expected");
927
928 istate->mmap = xmmap(NULL, istate->mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
929 close(fd);
930
931 hdr = istate->mmap;
932 if (verify_hdr(hdr, istate->mmap_size) < 0)
933 goto unmap;
934
935 istate->cache_nr = ntohl(hdr->hdr_entries);
936 istate->cache_alloc = alloc_nr(istate->cache_nr);
937 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
938
939 offset = sizeof(*hdr);
940 for (i = 0; i < istate->cache_nr; i++) {
941 struct cache_entry *ce;
942
943 ce = (struct cache_entry *)((char *)(istate->mmap) + offset);
944 offset = offset + ce_size(ce);
945 istate->cache[i] = ce;
946 }
947 istate->timestamp = st.st_mtime;
948 while (offset <= istate->mmap_size - 20 - 8) {
949 /* After an array of active_nr index entries,
950 * there can be arbitrary number of extended
951 * sections, each of which is prefixed with
952 * extension name (4-byte) and section length
953 * in 4-byte network byte order.
954 */
955 unsigned long extsize;
956 memcpy(&extsize, (char *)(istate->mmap) + offset + 4, 4);
957 extsize = ntohl(extsize);
958 if (read_index_extension(istate,
959 ((const char *) (istate->mmap)) + offset,
960 (char *) (istate->mmap) + offset + 8,
961 extsize) < 0)
962 goto unmap;
963 offset += 8;
964 offset += extsize;
965 }
966 return istate->cache_nr;
967
968 unmap:
969 munmap(istate->mmap, istate->mmap_size);
970 errno = EINVAL;
971 die("index file corrupt");
972 }
973
974 int discard_index(struct index_state *istate)
975 {
976 int ret;
977
978 istate->cache_nr = 0;
979 istate->cache_changed = 0;
980 istate->timestamp = 0;
981 cache_tree_free(&(istate->cache_tree));
982 if (istate->mmap == NULL)
983 return 0;
984 ret = munmap(istate->mmap, istate->mmap_size);
985 istate->mmap = NULL;
986 istate->mmap_size = 0;
987
988 /* no need to throw away allocated active_cache */
989 return ret;
990 }
991
992 #define WRITE_BUFFER_SIZE 8192
993 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
994 static unsigned long write_buffer_len;
995
996 static int ce_write_flush(SHA_CTX *context, int fd)
997 {
998 unsigned int buffered = write_buffer_len;
999 if (buffered) {
1000 SHA1_Update(context, write_buffer, buffered);
1001 if (write_in_full(fd, write_buffer, buffered) != buffered)
1002 return -1;
1003 write_buffer_len = 0;
1004 }
1005 return 0;
1006 }
1007
1008 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1009 {
1010 while (len) {
1011 unsigned int buffered = write_buffer_len;
1012 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1013 if (partial > len)
1014 partial = len;
1015 memcpy(write_buffer + buffered, data, partial);
1016 buffered += partial;
1017 if (buffered == WRITE_BUFFER_SIZE) {
1018 write_buffer_len = buffered;
1019 if (ce_write_flush(context, fd))
1020 return -1;
1021 buffered = 0;
1022 }
1023 write_buffer_len = buffered;
1024 len -= partial;
1025 data = (char *) data + partial;
1026 }
1027 return 0;
1028 }
1029
1030 static int write_index_ext_header(SHA_CTX *context, int fd,
1031 unsigned int ext, unsigned int sz)
1032 {
1033 ext = htonl(ext);
1034 sz = htonl(sz);
1035 return ((ce_write(context, fd, &ext, 4) < 0) ||
1036 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1037 }
1038
1039 static int ce_flush(SHA_CTX *context, int fd)
1040 {
1041 unsigned int left = write_buffer_len;
1042
1043 if (left) {
1044 write_buffer_len = 0;
1045 SHA1_Update(context, write_buffer, left);
1046 }
1047
1048 /* Flush first if not enough space for SHA1 signature */
1049 if (left + 20 > WRITE_BUFFER_SIZE) {
1050 if (write_in_full(fd, write_buffer, left) != left)
1051 return -1;
1052 left = 0;
1053 }
1054
1055 /* Append the SHA1 signature at the end */
1056 SHA1_Final(write_buffer + left, context);
1057 left += 20;
1058 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1059 }
1060
1061 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1062 {
1063 /*
1064 * The only thing we care about in this function is to smudge the
1065 * falsely clean entry due to touch-update-touch race, so we leave
1066 * everything else as they are. We are called for entries whose
1067 * ce_mtime match the index file mtime.
1068 */
1069 struct stat st;
1070
1071 if (lstat(ce->name, &st) < 0)
1072 return;
1073 if (ce_match_stat_basic(ce, &st))
1074 return;
1075 if (ce_modified_check_fs(ce, &st)) {
1076 /* This is "racily clean"; smudge it. Note that this
1077 * is a tricky code. At first glance, it may appear
1078 * that it can break with this sequence:
1079 *
1080 * $ echo xyzzy >frotz
1081 * $ git-update-index --add frotz
1082 * $ : >frotz
1083 * $ sleep 3
1084 * $ echo filfre >nitfol
1085 * $ git-update-index --add nitfol
1086 *
1087 * but it does not. When the second update-index runs,
1088 * it notices that the entry "frotz" has the same timestamp
1089 * as index, and if we were to smudge it by resetting its
1090 * size to zero here, then the object name recorded
1091 * in index is the 6-byte file but the cached stat information
1092 * becomes zero --- which would then match what we would
1093 * obtain from the filesystem next time we stat("frotz").
1094 *
1095 * However, the second update-index, before calling
1096 * this function, notices that the cached size is 6
1097 * bytes and what is on the filesystem is an empty
1098 * file, and never calls us, so the cached size information
1099 * for "frotz" stays 6 which does not match the filesystem.
1100 */
1101 ce->ce_size = htonl(0);
1102 }
1103 }
1104
1105 int write_index(struct index_state *istate, int newfd)
1106 {
1107 SHA_CTX c;
1108 struct cache_header hdr;
1109 int i, removed;
1110 struct cache_entry **cache = istate->cache;
1111 int entries = istate->cache_nr;
1112
1113 for (i = removed = 0; i < entries; i++)
1114 if (!cache[i]->ce_mode)
1115 removed++;
1116
1117 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1118 hdr.hdr_version = htonl(2);
1119 hdr.hdr_entries = htonl(entries - removed);
1120
1121 SHA1_Init(&c);
1122 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1123 return -1;
1124
1125 for (i = 0; i < entries; i++) {
1126 struct cache_entry *ce = cache[i];
1127 if (!ce->ce_mode)
1128 continue;
1129 if (istate->timestamp &&
1130 istate->timestamp <= ntohl(ce->ce_mtime.sec))
1131 ce_smudge_racily_clean_entry(ce);
1132 if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
1133 return -1;
1134 }
1135
1136 /* Write extension data here */
1137 if (istate->cache_tree) {
1138 unsigned long sz;
1139 void *data = cache_tree_write(istate->cache_tree, &sz);
1140 if (data &&
1141 !write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sz) &&
1142 !ce_write(&c, newfd, data, sz))
1143 free(data);
1144 else {
1145 free(data);
1146 return -1;
1147 }
1148 }
1149 return ce_flush(&c, newfd);
1150 }