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 #include "dir.h"
12 /* Index extensions.
13 *
14 * The first letter should be 'A'..'Z' for extensions that are not
15 * necessary for a correct operation (i.e. optimization data).
16 * When new extensions are added that _needs_ to be understood in
17 * order to correctly interpret the index file, pick character that
18 * is outside the range, to cause the reader to abort.
19 */
21 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
22 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
24 struct index_state the_index;
26 /*
27 * This only updates the "non-critical" parts of the directory
28 * cache, ie the parts that aren't tracked by GIT, and only used
29 * to validate the cache.
30 */
31 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
32 {
33 ce->ce_ctime = st->st_ctime;
34 ce->ce_mtime = st->st_mtime;
35 ce->ce_dev = st->st_dev;
36 ce->ce_ino = st->st_ino;
37 ce->ce_uid = st->st_uid;
38 ce->ce_gid = st->st_gid;
39 ce->ce_size = st->st_size;
41 if (assume_unchanged)
42 ce->ce_flags |= CE_VALID;
44 if (S_ISREG(st->st_mode))
45 ce_mark_uptodate(ce);
46 }
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);
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 }
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;
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 }
89 static int ce_compare_gitlink(struct cache_entry *ce)
90 {
91 unsigned char sha1[20];
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 }
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(ce->ce_mode))
119 return 0;
120 default:
121 return TYPE_CHANGED;
122 }
123 return 0;
124 }
126 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
127 {
128 unsigned int changed = 0;
130 if (ce->ce_flags & CE_REMOVE)
131 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
133 switch (ce->ce_mode & S_IFMT) {
134 case S_IFREG:
135 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
136 /* We consider only the owner x bit to be relevant for
137 * "mode changes"
138 */
139 if (trust_executable_bit &&
140 (0100 & (ce->ce_mode ^ st->st_mode)))
141 changed |= MODE_CHANGED;
142 break;
143 case S_IFLNK:
144 if (!S_ISLNK(st->st_mode) &&
145 (has_symlinks || !S_ISREG(st->st_mode)))
146 changed |= TYPE_CHANGED;
147 break;
148 case S_IFGITLINK:
149 if (!S_ISDIR(st->st_mode))
150 changed |= TYPE_CHANGED;
151 else if (ce_compare_gitlink(ce))
152 changed |= DATA_CHANGED;
153 return changed;
154 default:
155 die("internal error: ce_mode is %o", ce->ce_mode);
156 }
157 if (ce->ce_mtime != (unsigned int) st->st_mtime)
158 changed |= MTIME_CHANGED;
159 if (ce->ce_ctime != (unsigned int) st->st_ctime)
160 changed |= CTIME_CHANGED;
162 if (ce->ce_uid != (unsigned int) st->st_uid ||
163 ce->ce_gid != (unsigned int) st->st_gid)
164 changed |= OWNER_CHANGED;
165 if (ce->ce_ino != (unsigned int) st->st_ino)
166 changed |= INODE_CHANGED;
168 #ifdef USE_STDEV
169 /*
170 * st_dev breaks on network filesystems where different
171 * clients will have different views of what "device"
172 * the filesystem is on
173 */
174 if (ce->ce_dev != (unsigned int) st->st_dev)
175 changed |= INODE_CHANGED;
176 #endif
178 if (ce->ce_size != (unsigned int) st->st_size)
179 changed |= DATA_CHANGED;
181 return changed;
182 }
184 int ie_match_stat(struct index_state *istate,
185 struct cache_entry *ce, struct stat *st,
186 unsigned int options)
187 {
188 unsigned int changed;
189 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
190 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
192 /*
193 * If it's marked as always valid in the index, it's
194 * valid whatever the checked-out copy says.
195 */
196 if (!ignore_valid && (ce->ce_flags & CE_VALID))
197 return 0;
199 changed = ce_match_stat_basic(ce, st);
201 /*
202 * Within 1 second of this sequence:
203 * echo xyzzy >file && git-update-index --add file
204 * running this command:
205 * echo frotz >file
206 * would give a falsely clean cache entry. The mtime and
207 * length match the cache, and other stat fields do not change.
208 *
209 * We could detect this at update-index time (the cache entry
210 * being registered/updated records the same time as "now")
211 * and delay the return from git-update-index, but that would
212 * effectively mean we can make at most one commit per second,
213 * which is not acceptable. Instead, we check cache entries
214 * whose mtime are the same as the index file timestamp more
215 * carefully than others.
216 */
217 if (!changed &&
218 istate->timestamp &&
219 istate->timestamp <= ce->ce_mtime) {
220 if (assume_racy_is_modified)
221 changed |= DATA_CHANGED;
222 else
223 changed |= ce_modified_check_fs(ce, st);
224 }
226 return changed;
227 }
229 int ie_modified(struct index_state *istate,
230 struct cache_entry *ce, struct stat *st, unsigned int options)
231 {
232 int changed, changed_fs;
234 changed = ie_match_stat(istate, ce, st, options);
235 if (!changed)
236 return 0;
237 /*
238 * If the mode or type has changed, there's no point in trying
239 * to refresh the entry - it's not going to match
240 */
241 if (changed & (MODE_CHANGED | TYPE_CHANGED))
242 return changed;
244 /* Immediately after read-tree or update-index --cacheinfo,
245 * the length field is zero. For other cases the ce_size
246 * should match the SHA1 recorded in the index entry.
247 */
248 if ((changed & DATA_CHANGED) && ce->ce_size != 0)
249 return changed;
251 changed_fs = ce_modified_check_fs(ce, st);
252 if (changed_fs)
253 return changed | changed_fs;
254 return 0;
255 }
257 int base_name_compare(const char *name1, int len1, int mode1,
258 const char *name2, int len2, int mode2)
259 {
260 unsigned char c1, c2;
261 int len = len1 < len2 ? len1 : len2;
262 int cmp;
264 cmp = memcmp(name1, name2, len);
265 if (cmp)
266 return cmp;
267 c1 = name1[len];
268 c2 = name2[len];
269 if (!c1 && S_ISDIR(mode1))
270 c1 = '/';
271 if (!c2 && S_ISDIR(mode2))
272 c2 = '/';
273 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
274 }
276 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
277 {
278 int len1 = flags1 & CE_NAMEMASK;
279 int len2 = flags2 & CE_NAMEMASK;
280 int len = len1 < len2 ? len1 : len2;
281 int cmp;
283 cmp = memcmp(name1, name2, len);
284 if (cmp)
285 return cmp;
286 if (len1 < len2)
287 return -1;
288 if (len1 > len2)
289 return 1;
291 /* Compare stages */
292 flags1 &= CE_STAGEMASK;
293 flags2 &= CE_STAGEMASK;
295 if (flags1 < flags2)
296 return -1;
297 if (flags1 > flags2)
298 return 1;
299 return 0;
300 }
302 int index_name_pos(struct index_state *istate, const char *name, int namelen)
303 {
304 int first, last;
306 first = 0;
307 last = istate->cache_nr;
308 while (last > first) {
309 int next = (last + first) >> 1;
310 struct cache_entry *ce = istate->cache[next];
311 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
312 if (!cmp)
313 return next;
314 if (cmp < 0) {
315 last = next;
316 continue;
317 }
318 first = next+1;
319 }
320 return -first-1;
321 }
323 /* Remove entry, return true if there are more entries to go.. */
324 int remove_index_entry_at(struct index_state *istate, int pos)
325 {
326 istate->cache_changed = 1;
327 istate->cache_nr--;
328 if (pos >= istate->cache_nr)
329 return 0;
330 memmove(istate->cache + pos,
331 istate->cache + pos + 1,
332 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
333 return 1;
334 }
336 int remove_file_from_index(struct index_state *istate, const char *path)
337 {
338 int pos = index_name_pos(istate, path, strlen(path));
339 if (pos < 0)
340 pos = -pos-1;
341 cache_tree_invalidate_path(istate->cache_tree, path);
342 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
343 remove_index_entry_at(istate, pos);
344 return 0;
345 }
347 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
348 {
349 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
350 }
352 static int index_name_pos_also_unmerged(struct index_state *istate,
353 const char *path, int namelen)
354 {
355 int pos = index_name_pos(istate, path, namelen);
356 struct cache_entry *ce;
358 if (pos >= 0)
359 return pos;
361 /* maybe unmerged? */
362 pos = -1 - pos;
363 if (pos >= istate->cache_nr ||
364 compare_name((ce = istate->cache[pos]), path, namelen))
365 return -1;
367 /* order of preference: stage 2, 1, 3 */
368 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
369 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
370 !compare_name(ce, path, namelen))
371 pos++;
372 return pos;
373 }
375 int add_file_to_index(struct index_state *istate, const char *path, int verbose)
376 {
377 int size, namelen, pos;
378 struct stat st;
379 struct cache_entry *ce;
380 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
382 if (lstat(path, &st))
383 die("%s: unable to stat (%s)", path, strerror(errno));
385 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
386 die("%s: can only add regular files, symbolic links or git-directories", path);
388 namelen = strlen(path);
389 if (S_ISDIR(st.st_mode)) {
390 while (namelen && path[namelen-1] == '/')
391 namelen--;
392 }
393 size = cache_entry_size(namelen);
394 ce = xcalloc(1, size);
395 memcpy(ce->name, path, namelen);
396 ce->ce_flags = namelen;
397 fill_stat_cache_info(ce, &st);
399 if (trust_executable_bit && has_symlinks)
400 ce->ce_mode = create_ce_mode(st.st_mode);
401 else {
402 /* If there is an existing entry, pick the mode bits and type
403 * from it, otherwise assume unexecutable regular file.
404 */
405 struct cache_entry *ent;
406 int pos = index_name_pos_also_unmerged(istate, path, namelen);
408 ent = (0 <= pos) ? istate->cache[pos] : NULL;
409 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
410 }
412 pos = index_name_pos(istate, ce->name, namelen);
413 if (0 <= pos &&
414 !ce_stage(istate->cache[pos]) &&
415 !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) {
416 /* Nothing changed, really */
417 free(ce);
418 ce_mark_uptodate(istate->cache[pos]);
419 return 0;
420 }
422 if (index_path(ce->sha1, path, &st, 1))
423 die("unable to index file %s", path);
424 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
425 die("unable to add %s to index",path);
426 if (verbose)
427 printf("add '%s'\n", path);
428 return 0;
429 }
431 struct cache_entry *make_cache_entry(unsigned int mode,
432 const unsigned char *sha1, const char *path, int stage,
433 int refresh)
434 {
435 int size, len;
436 struct cache_entry *ce;
438 if (!verify_path(path))
439 return NULL;
441 len = strlen(path);
442 size = cache_entry_size(len);
443 ce = xcalloc(1, size);
445 hashcpy(ce->sha1, sha1);
446 memcpy(ce->name, path, len);
447 ce->ce_flags = create_ce_flags(len, stage);
448 ce->ce_mode = create_ce_mode(mode);
450 if (refresh)
451 return refresh_cache_entry(ce, 0);
453 return ce;
454 }
456 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
457 {
458 int len = ce_namelen(a);
459 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
460 }
462 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
463 {
464 const char *match, *name;
465 int len;
467 if (!pathspec)
468 return 1;
470 len = ce_namelen(ce);
471 name = ce->name;
472 while ((match = *pathspec++) != NULL) {
473 int matchlen = strlen(match);
474 if (matchlen > len)
475 continue;
476 if (memcmp(name, match, matchlen))
477 continue;
478 if (matchlen && name[matchlen-1] == '/')
479 return 1;
480 if (name[matchlen] == '/' || !name[matchlen])
481 return 1;
482 if (!matchlen)
483 return 1;
484 }
485 return 0;
486 }
488 /*
489 * We fundamentally don't like some paths: we don't want
490 * dot or dot-dot anywhere, and for obvious reasons don't
491 * want to recurse into ".git" either.
492 *
493 * Also, we don't want double slashes or slashes at the
494 * end that can make pathnames ambiguous.
495 */
496 static int verify_dotfile(const char *rest)
497 {
498 /*
499 * The first character was '.', but that
500 * has already been discarded, we now test
501 * the rest.
502 */
503 switch (*rest) {
504 /* "." is not allowed */
505 case '\0': case '/':
506 return 0;
508 /*
509 * ".git" followed by NUL or slash is bad. This
510 * shares the path end test with the ".." case.
511 */
512 case 'g':
513 if (rest[1] != 'i')
514 break;
515 if (rest[2] != 't')
516 break;
517 rest += 2;
518 /* fallthrough */
519 case '.':
520 if (rest[1] == '\0' || rest[1] == '/')
521 return 0;
522 }
523 return 1;
524 }
526 int verify_path(const char *path)
527 {
528 char c;
530 goto inside;
531 for (;;) {
532 if (!c)
533 return 1;
534 if (c == '/') {
535 inside:
536 c = *path++;
537 switch (c) {
538 default:
539 continue;
540 case '/': case '\0':
541 break;
542 case '.':
543 if (verify_dotfile(path))
544 continue;
545 }
546 return 0;
547 }
548 c = *path++;
549 }
550 }
552 /*
553 * Do we have another file that has the beginning components being a
554 * proper superset of the name we're trying to add?
555 */
556 static int has_file_name(struct index_state *istate,
557 const struct cache_entry *ce, int pos, int ok_to_replace)
558 {
559 int retval = 0;
560 int len = ce_namelen(ce);
561 int stage = ce_stage(ce);
562 const char *name = ce->name;
564 while (pos < istate->cache_nr) {
565 struct cache_entry *p = istate->cache[pos++];
567 if (len >= ce_namelen(p))
568 break;
569 if (memcmp(name, p->name, len))
570 break;
571 if (ce_stage(p) != stage)
572 continue;
573 if (p->name[len] != '/')
574 continue;
575 if (p->ce_flags & CE_REMOVE)
576 continue;
577 retval = -1;
578 if (!ok_to_replace)
579 break;
580 remove_index_entry_at(istate, --pos);
581 }
582 return retval;
583 }
585 /*
586 * Do we have another file with a pathname that is a proper
587 * subset of the name we're trying to add?
588 */
589 static int has_dir_name(struct index_state *istate,
590 const struct cache_entry *ce, int pos, int ok_to_replace)
591 {
592 int retval = 0;
593 int stage = ce_stage(ce);
594 const char *name = ce->name;
595 const char *slash = name + ce_namelen(ce);
597 for (;;) {
598 int len;
600 for (;;) {
601 if (*--slash == '/')
602 break;
603 if (slash <= ce->name)
604 return retval;
605 }
606 len = slash - name;
608 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
609 if (pos >= 0) {
610 /*
611 * Found one, but not so fast. This could
612 * be a marker that says "I was here, but
613 * I am being removed". Such an entry is
614 * not a part of the resulting tree, and
615 * it is Ok to have a directory at the same
616 * path.
617 */
618 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
619 retval = -1;
620 if (!ok_to_replace)
621 break;
622 remove_index_entry_at(istate, pos);
623 continue;
624 }
625 }
626 else
627 pos = -pos-1;
629 /*
630 * Trivial optimization: if we find an entry that
631 * already matches the sub-directory, then we know
632 * we're ok, and we can exit.
633 */
634 while (pos < istate->cache_nr) {
635 struct cache_entry *p = istate->cache[pos];
636 if ((ce_namelen(p) <= len) ||
637 (p->name[len] != '/') ||
638 memcmp(p->name, name, len))
639 break; /* not our subdirectory */
640 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
641 /*
642 * p is at the same stage as our entry, and
643 * is a subdirectory of what we are looking
644 * at, so we cannot have conflicts at our
645 * level or anything shorter.
646 */
647 return retval;
648 pos++;
649 }
650 }
651 return retval;
652 }
654 /* We may be in a situation where we already have path/file and path
655 * is being added, or we already have path and path/file is being
656 * added. Either one would result in a nonsense tree that has path
657 * twice when git-write-tree tries to write it out. Prevent it.
658 *
659 * If ok-to-replace is specified, we remove the conflicting entries
660 * from the cache so the caller should recompute the insert position.
661 * When this happens, we return non-zero.
662 */
663 static int check_file_directory_conflict(struct index_state *istate,
664 const struct cache_entry *ce,
665 int pos, int ok_to_replace)
666 {
667 int retval;
669 /*
670 * When ce is an "I am going away" entry, we allow it to be added
671 */
672 if (ce->ce_flags & CE_REMOVE)
673 return 0;
675 /*
676 * We check if the path is a sub-path of a subsequent pathname
677 * first, since removing those will not change the position
678 * in the array.
679 */
680 retval = has_file_name(istate, ce, pos, ok_to_replace);
682 /*
683 * Then check if the path might have a clashing sub-directory
684 * before it.
685 */
686 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
687 }
689 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
690 {
691 int pos;
692 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
693 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
694 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
696 cache_tree_invalidate_path(istate->cache_tree, ce->name);
697 pos = index_name_pos(istate, ce->name, ce->ce_flags);
699 /* existing match? Just replace it. */
700 if (pos >= 0) {
701 istate->cache_changed = 1;
702 istate->cache[pos] = ce;
703 return 0;
704 }
705 pos = -pos-1;
707 /*
708 * Inserting a merged entry ("stage 0") into the index
709 * will always replace all non-merged entries..
710 */
711 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
712 while (ce_same_name(istate->cache[pos], ce)) {
713 ok_to_add = 1;
714 if (!remove_index_entry_at(istate, pos))
715 break;
716 }
717 }
719 if (!ok_to_add)
720 return -1;
721 if (!verify_path(ce->name))
722 return -1;
724 if (!skip_df_check &&
725 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
726 if (!ok_to_replace)
727 return error("'%s' appears as both a file and as a directory",
728 ce->name);
729 pos = index_name_pos(istate, ce->name, ce->ce_flags);
730 pos = -pos-1;
731 }
732 return pos + 1;
733 }
735 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
736 {
737 int pos;
739 if (option & ADD_CACHE_JUST_APPEND)
740 pos = istate->cache_nr;
741 else {
742 int ret;
743 ret = add_index_entry_with_check(istate, ce, option);
744 if (ret <= 0)
745 return ret;
746 pos = ret - 1;
747 }
749 /* Make sure the array is big enough .. */
750 if (istate->cache_nr == istate->cache_alloc) {
751 istate->cache_alloc = alloc_nr(istate->cache_alloc);
752 istate->cache = xrealloc(istate->cache,
753 istate->cache_alloc * sizeof(struct cache_entry *));
754 }
756 /* Add it in.. */
757 istate->cache_nr++;
758 if (istate->cache_nr > pos + 1)
759 memmove(istate->cache + pos + 1,
760 istate->cache + pos,
761 (istate->cache_nr - pos - 1) * sizeof(ce));
762 istate->cache[pos] = ce;
763 istate->cache_changed = 1;
764 return 0;
765 }
767 /*
768 * "refresh" does not calculate a new sha1 file or bring the
769 * cache up-to-date for mode/content changes. But what it
770 * _does_ do is to "re-match" the stat information of a file
771 * with the cache, so that you can refresh the cache for a
772 * file that hasn't been changed but where the stat entry is
773 * out of date.
774 *
775 * For example, you'd want to do this after doing a "git-read-tree",
776 * to link up the stat cache details with the proper files.
777 */
778 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
779 struct cache_entry *ce,
780 unsigned int options, int *err)
781 {
782 struct stat st;
783 struct cache_entry *updated;
784 int changed, size;
785 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
787 if (ce_uptodate(ce))
788 return ce;
790 if (lstat(ce->name, &st) < 0) {
791 if (err)
792 *err = errno;
793 return NULL;
794 }
796 changed = ie_match_stat(istate, ce, &st, options);
797 if (!changed) {
798 /*
799 * The path is unchanged. If we were told to ignore
800 * valid bit, then we did the actual stat check and
801 * found that the entry is unmodified. If the entry
802 * is not marked VALID, this is the place to mark it
803 * valid again, under "assume unchanged" mode.
804 */
805 if (ignore_valid && assume_unchanged &&
806 !(ce->ce_flags & CE_VALID))
807 ; /* mark this one VALID again */
808 else {
809 /*
810 * We do not mark the index itself "modified"
811 * because CE_UPTODATE flag is in-core only;
812 * we are not going to write this change out.
813 */
814 ce_mark_uptodate(ce);
815 return ce;
816 }
817 }
819 if (ie_modified(istate, ce, &st, options)) {
820 if (err)
821 *err = EINVAL;
822 return NULL;
823 }
825 size = ce_size(ce);
826 updated = xmalloc(size);
827 memcpy(updated, ce, size);
828 fill_stat_cache_info(updated, &st);
829 /*
830 * If ignore_valid is not set, we should leave CE_VALID bit
831 * alone. Otherwise, paths marked with --no-assume-unchanged
832 * (i.e. things to be edited) will reacquire CE_VALID bit
833 * automatically, which is not really what we want.
834 */
835 if (!ignore_valid && assume_unchanged &&
836 !(ce->ce_flags & CE_VALID))
837 updated->ce_flags &= ~CE_VALID;
839 return updated;
840 }
842 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
843 {
844 int i;
845 int has_errors = 0;
846 int really = (flags & REFRESH_REALLY) != 0;
847 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
848 int quiet = (flags & REFRESH_QUIET) != 0;
849 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
850 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
852 for (i = 0; i < istate->cache_nr; i++) {
853 struct cache_entry *ce, *new;
854 int cache_errno = 0;
856 ce = istate->cache[i];
857 if (ce_stage(ce)) {
858 while ((i < istate->cache_nr) &&
859 ! strcmp(istate->cache[i]->name, ce->name))
860 i++;
861 i--;
862 if (allow_unmerged)
863 continue;
864 printf("%s: needs merge\n", ce->name);
865 has_errors = 1;
866 continue;
867 }
869 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
870 continue;
872 new = refresh_cache_ent(istate, ce, options, &cache_errno);
873 if (new == ce)
874 continue;
875 if (!new) {
876 if (not_new && cache_errno == ENOENT)
877 continue;
878 if (really && cache_errno == EINVAL) {
879 /* If we are doing --really-refresh that
880 * means the index is not valid anymore.
881 */
882 ce->ce_flags &= ~CE_VALID;
883 istate->cache_changed = 1;
884 }
885 if (quiet)
886 continue;
887 printf("%s: needs update\n", ce->name);
888 has_errors = 1;
889 continue;
890 }
891 istate->cache_changed = 1;
892 /* You can NOT just free istate->cache[i] here, since it
893 * might not be necessarily malloc()ed but can also come
894 * from mmap(). */
895 istate->cache[i] = new;
896 }
897 return has_errors;
898 }
900 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
901 {
902 return refresh_cache_ent(&the_index, ce, really, NULL);
903 }
905 static int verify_hdr(struct cache_header *hdr, unsigned long size)
906 {
907 SHA_CTX c;
908 unsigned char sha1[20];
910 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
911 return error("bad signature");
912 if (hdr->hdr_version != htonl(2))
913 return error("bad index version");
914 SHA1_Init(&c);
915 SHA1_Update(&c, hdr, size - 20);
916 SHA1_Final(sha1, &c);
917 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
918 return error("bad index file sha1 signature");
919 return 0;
920 }
922 static int read_index_extension(struct index_state *istate,
923 const char *ext, void *data, unsigned long sz)
924 {
925 switch (CACHE_EXT(ext)) {
926 case CACHE_EXT_TREE:
927 istate->cache_tree = cache_tree_read(data, sz);
928 break;
929 default:
930 if (*ext < 'A' || 'Z' < *ext)
931 return error("index uses %.4s extension, which we do not understand",
932 ext);
933 fprintf(stderr, "ignoring %.4s extension\n", ext);
934 break;
935 }
936 return 0;
937 }
939 int read_index(struct index_state *istate)
940 {
941 return read_index_from(istate, get_index_file());
942 }
944 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
945 {
946 size_t len;
948 ce->ce_ctime = ntohl(ondisk->ctime.sec);
949 ce->ce_mtime = ntohl(ondisk->mtime.sec);
950 ce->ce_dev = ntohl(ondisk->dev);
951 ce->ce_ino = ntohl(ondisk->ino);
952 ce->ce_mode = ntohl(ondisk->mode);
953 ce->ce_uid = ntohl(ondisk->uid);
954 ce->ce_gid = ntohl(ondisk->gid);
955 ce->ce_size = ntohl(ondisk->size);
956 /* On-disk flags are just 16 bits */
957 ce->ce_flags = ntohs(ondisk->flags);
958 hashcpy(ce->sha1, ondisk->sha1);
960 len = ce->ce_flags & CE_NAMEMASK;
961 if (len == CE_NAMEMASK)
962 len = strlen(ondisk->name);
963 /*
964 * NEEDSWORK: If the original index is crafted, this copy could
965 * go unchecked.
966 */
967 memcpy(ce->name, ondisk->name, len + 1);
968 }
970 /* remember to discard_cache() before reading a different cache! */
971 int read_index_from(struct index_state *istate, const char *path)
972 {
973 int fd, i;
974 struct stat st;
975 unsigned long src_offset, dst_offset;
976 struct cache_header *hdr;
977 void *mmap;
978 size_t mmap_size;
980 errno = EBUSY;
981 if (istate->alloc)
982 return istate->cache_nr;
984 errno = ENOENT;
985 istate->timestamp = 0;
986 fd = open(path, O_RDONLY);
987 if (fd < 0) {
988 if (errno == ENOENT)
989 return 0;
990 die("index file open failed (%s)", strerror(errno));
991 }
993 if (fstat(fd, &st))
994 die("cannot stat the open index (%s)", strerror(errno));
996 errno = EINVAL;
997 mmap_size = xsize_t(st.st_size);
998 if (mmap_size < sizeof(struct cache_header) + 20)
999 die("index file smaller than expected");
1001 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1002 close(fd);
1003 if (mmap == MAP_FAILED)
1004 die("unable to map index file");
1006 hdr = mmap;
1007 if (verify_hdr(hdr, mmap_size) < 0)
1008 goto unmap;
1010 istate->cache_nr = ntohl(hdr->hdr_entries);
1011 istate->cache_alloc = alloc_nr(istate->cache_nr);
1012 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1014 /*
1015 * The disk format is actually larger than the in-memory format,
1016 * due to space for nsec etc, so even though the in-memory one
1017 * has room for a few more flags, we can allocate using the same
1018 * index size
1019 */
1020 istate->alloc = xmalloc(mmap_size);
1022 src_offset = sizeof(*hdr);
1023 dst_offset = 0;
1024 for (i = 0; i < istate->cache_nr; i++) {
1025 struct ondisk_cache_entry *disk_ce;
1026 struct cache_entry *ce;
1028 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1029 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1030 convert_from_disk(disk_ce, ce);
1031 istate->cache[i] = ce;
1033 src_offset += ondisk_ce_size(ce);
1034 dst_offset += ce_size(ce);
1035 }
1036 istate->timestamp = st.st_mtime;
1037 while (src_offset <= mmap_size - 20 - 8) {
1038 /* After an array of active_nr index entries,
1039 * there can be arbitrary number of extended
1040 * sections, each of which is prefixed with
1041 * extension name (4-byte) and section length
1042 * in 4-byte network byte order.
1043 */
1044 unsigned long extsize;
1045 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1046 extsize = ntohl(extsize);
1047 if (read_index_extension(istate,
1048 (const char *) mmap + src_offset,
1049 (char *) mmap + src_offset + 8,
1050 extsize) < 0)
1051 goto unmap;
1052 src_offset += 8;
1053 src_offset += extsize;
1054 }
1055 munmap(mmap, mmap_size);
1056 return istate->cache_nr;
1058 unmap:
1059 munmap(mmap, mmap_size);
1060 errno = EINVAL;
1061 die("index file corrupt");
1062 }
1064 int discard_index(struct index_state *istate)
1065 {
1066 istate->cache_nr = 0;
1067 istate->cache_changed = 0;
1068 istate->timestamp = 0;
1069 cache_tree_free(&(istate->cache_tree));
1070 free(istate->alloc);
1071 istate->alloc = NULL;
1073 /* no need to throw away allocated active_cache */
1074 return 0;
1075 }
1077 #define WRITE_BUFFER_SIZE 8192
1078 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1079 static unsigned long write_buffer_len;
1081 static int ce_write_flush(SHA_CTX *context, int fd)
1082 {
1083 unsigned int buffered = write_buffer_len;
1084 if (buffered) {
1085 SHA1_Update(context, write_buffer, buffered);
1086 if (write_in_full(fd, write_buffer, buffered) != buffered)
1087 return -1;
1088 write_buffer_len = 0;
1089 }
1090 return 0;
1091 }
1093 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1094 {
1095 while (len) {
1096 unsigned int buffered = write_buffer_len;
1097 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1098 if (partial > len)
1099 partial = len;
1100 memcpy(write_buffer + buffered, data, partial);
1101 buffered += partial;
1102 if (buffered == WRITE_BUFFER_SIZE) {
1103 write_buffer_len = buffered;
1104 if (ce_write_flush(context, fd))
1105 return -1;
1106 buffered = 0;
1107 }
1108 write_buffer_len = buffered;
1109 len -= partial;
1110 data = (char *) data + partial;
1111 }
1112 return 0;
1113 }
1115 static int write_index_ext_header(SHA_CTX *context, int fd,
1116 unsigned int ext, unsigned int sz)
1117 {
1118 ext = htonl(ext);
1119 sz = htonl(sz);
1120 return ((ce_write(context, fd, &ext, 4) < 0) ||
1121 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1122 }
1124 static int ce_flush(SHA_CTX *context, int fd)
1125 {
1126 unsigned int left = write_buffer_len;
1128 if (left) {
1129 write_buffer_len = 0;
1130 SHA1_Update(context, write_buffer, left);
1131 }
1133 /* Flush first if not enough space for SHA1 signature */
1134 if (left + 20 > WRITE_BUFFER_SIZE) {
1135 if (write_in_full(fd, write_buffer, left) != left)
1136 return -1;
1137 left = 0;
1138 }
1140 /* Append the SHA1 signature at the end */
1141 SHA1_Final(write_buffer + left, context);
1142 left += 20;
1143 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1144 }
1146 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1147 {
1148 /*
1149 * The only thing we care about in this function is to smudge the
1150 * falsely clean entry due to touch-update-touch race, so we leave
1151 * everything else as they are. We are called for entries whose
1152 * ce_mtime match the index file mtime.
1153 */
1154 struct stat st;
1156 if (lstat(ce->name, &st) < 0)
1157 return;
1158 if (ce_match_stat_basic(ce, &st))
1159 return;
1160 if (ce_modified_check_fs(ce, &st)) {
1161 /* This is "racily clean"; smudge it. Note that this
1162 * is a tricky code. At first glance, it may appear
1163 * that it can break with this sequence:
1164 *
1165 * $ echo xyzzy >frotz
1166 * $ git-update-index --add frotz
1167 * $ : >frotz
1168 * $ sleep 3
1169 * $ echo filfre >nitfol
1170 * $ git-update-index --add nitfol
1171 *
1172 * but it does not. When the second update-index runs,
1173 * it notices that the entry "frotz" has the same timestamp
1174 * as index, and if we were to smudge it by resetting its
1175 * size to zero here, then the object name recorded
1176 * in index is the 6-byte file but the cached stat information
1177 * becomes zero --- which would then match what we would
1178 * obtain from the filesystem next time we stat("frotz").
1179 *
1180 * However, the second update-index, before calling
1181 * this function, notices that the cached size is 6
1182 * bytes and what is on the filesystem is an empty
1183 * file, and never calls us, so the cached size information
1184 * for "frotz" stays 6 which does not match the filesystem.
1185 */
1186 ce->ce_size = 0;
1187 }
1188 }
1190 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1191 {
1192 int size = ondisk_ce_size(ce);
1193 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1195 ondisk->ctime.sec = htonl(ce->ce_ctime);
1196 ondisk->ctime.nsec = 0;
1197 ondisk->mtime.sec = htonl(ce->ce_mtime);
1198 ondisk->mtime.nsec = 0;
1199 ondisk->dev = htonl(ce->ce_dev);
1200 ondisk->ino = htonl(ce->ce_ino);
1201 ondisk->mode = htonl(ce->ce_mode);
1202 ondisk->uid = htonl(ce->ce_uid);
1203 ondisk->gid = htonl(ce->ce_gid);
1204 ondisk->size = htonl(ce->ce_size);
1205 hashcpy(ondisk->sha1, ce->sha1);
1206 ondisk->flags = htons(ce->ce_flags);
1207 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1209 return ce_write(c, fd, ondisk, size);
1210 }
1212 int write_index(struct index_state *istate, int newfd)
1213 {
1214 SHA_CTX c;
1215 struct cache_header hdr;
1216 int i, err, removed;
1217 struct cache_entry **cache = istate->cache;
1218 int entries = istate->cache_nr;
1220 for (i = removed = 0; i < entries; i++)
1221 if (cache[i]->ce_flags & CE_REMOVE)
1222 removed++;
1224 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1225 hdr.hdr_version = htonl(2);
1226 hdr.hdr_entries = htonl(entries - removed);
1228 SHA1_Init(&c);
1229 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1230 return -1;
1232 for (i = 0; i < entries; i++) {
1233 struct cache_entry *ce = cache[i];
1234 if (ce->ce_flags & CE_REMOVE)
1235 continue;
1236 if (istate->timestamp &&
1237 istate->timestamp <= ce->ce_mtime)
1238 ce_smudge_racily_clean_entry(ce);
1239 if (ce_write_entry(&c, newfd, ce) < 0)
1240 return -1;
1241 }
1243 /* Write extension data here */
1244 if (istate->cache_tree) {
1245 struct strbuf sb;
1247 strbuf_init(&sb, 0);
1248 cache_tree_write(&sb, istate->cache_tree);
1249 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1250 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1251 strbuf_release(&sb);
1252 if (err)
1253 return -1;
1254 }
1255 return ce_flush(&c, newfd);
1256 }