get rid of num_packed_objects()
[git/gitweb.git] / read-cache.c
blob54573ce2ee3b2c70d5419716b20ade61683bc289
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #include "cache.h"
7 #include "cache-tree.h"
9 /* Index extensions.
11 * The first letter should be 'A'..'Z' for extensions that are not
12 * necessary for a correct operation (i.e. optimization data).
13 * When new extensions are added that _needs_ to be understood in
14 * order to correctly interpret the index file, pick character that
15 * is outside the range, to cause the reader to abort.
18 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
19 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
21 struct cache_entry **active_cache;
22 static time_t index_file_timestamp;
23 unsigned int active_nr, active_alloc, active_cache_changed;
25 struct cache_tree *active_cache_tree;
27 static void *cache_mmap;
28 static size_t cache_mmap_size;
31 * This only updates the "non-critical" parts of the directory
32 * cache, ie the parts that aren't tracked by GIT, and only used
33 * to validate the cache.
35 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
37 ce->ce_ctime.sec = htonl(st->st_ctime);
38 ce->ce_mtime.sec = htonl(st->st_mtime);
39 #ifdef USE_NSEC
40 ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
41 ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
42 #endif
43 ce->ce_dev = htonl(st->st_dev);
44 ce->ce_ino = htonl(st->st_ino);
45 ce->ce_uid = htonl(st->st_uid);
46 ce->ce_gid = htonl(st->st_gid);
47 ce->ce_size = htonl(st->st_size);
49 if (assume_unchanged)
50 ce->ce_flags |= htons(CE_VALID);
53 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
55 int match = -1;
56 int fd = open(ce->name, O_RDONLY);
58 if (fd >= 0) {
59 unsigned char sha1[20];
60 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
61 match = hashcmp(sha1, ce->sha1);
62 /* index_fd() closed the file descriptor already */
64 return match;
67 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
69 int match = -1;
70 char *target;
71 void *buffer;
72 unsigned long size;
73 enum object_type type;
74 int len;
76 target = xmalloc(expected_size);
77 len = readlink(ce->name, target, expected_size);
78 if (len != expected_size) {
79 free(target);
80 return -1;
82 buffer = read_sha1_file(ce->sha1, &type, &size);
83 if (!buffer) {
84 free(target);
85 return -1;
87 if (size == expected_size)
88 match = memcmp(buffer, target, size);
89 free(buffer);
90 free(target);
91 return match;
94 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
96 switch (st->st_mode & S_IFMT) {
97 case S_IFREG:
98 if (ce_compare_data(ce, st))
99 return DATA_CHANGED;
100 break;
101 case S_IFLNK:
102 if (ce_compare_link(ce, xsize_t(st->st_size)))
103 return DATA_CHANGED;
104 break;
105 default:
106 return TYPE_CHANGED;
108 return 0;
111 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
113 unsigned int changed = 0;
115 switch (ntohl(ce->ce_mode) & S_IFMT) {
116 case S_IFREG:
117 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
118 /* We consider only the owner x bit to be relevant for
119 * "mode changes"
121 if (trust_executable_bit &&
122 (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
123 changed |= MODE_CHANGED;
124 break;
125 case S_IFLNK:
126 if (!S_ISLNK(st->st_mode) &&
127 (has_symlinks || !S_ISREG(st->st_mode)))
128 changed |= TYPE_CHANGED;
129 break;
130 default:
131 die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
133 if (ce->ce_mtime.sec != htonl(st->st_mtime))
134 changed |= MTIME_CHANGED;
135 if (ce->ce_ctime.sec != htonl(st->st_ctime))
136 changed |= CTIME_CHANGED;
138 #ifdef USE_NSEC
140 * nsec seems unreliable - not all filesystems support it, so
141 * as long as it is in the inode cache you get right nsec
142 * but after it gets flushed, you get zero nsec.
144 if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
145 changed |= MTIME_CHANGED;
146 if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
147 changed |= CTIME_CHANGED;
148 #endif
150 if (ce->ce_uid != htonl(st->st_uid) ||
151 ce->ce_gid != htonl(st->st_gid))
152 changed |= OWNER_CHANGED;
153 if (ce->ce_ino != htonl(st->st_ino))
154 changed |= INODE_CHANGED;
156 #ifdef USE_STDEV
158 * st_dev breaks on network filesystems where different
159 * clients will have different views of what "device"
160 * the filesystem is on
162 if (ce->ce_dev != htonl(st->st_dev))
163 changed |= INODE_CHANGED;
164 #endif
166 if (ce->ce_size != htonl(st->st_size))
167 changed |= DATA_CHANGED;
169 return changed;
172 int ce_match_stat(struct cache_entry *ce, struct stat *st, int options)
174 unsigned int changed;
175 int ignore_valid = options & 01;
176 int assume_racy_is_modified = options & 02;
179 * If it's marked as always valid in the index, it's
180 * valid whatever the checked-out copy says.
182 if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
183 return 0;
185 changed = ce_match_stat_basic(ce, st);
188 * Within 1 second of this sequence:
189 * echo xyzzy >file && git-update-index --add file
190 * running this command:
191 * echo frotz >file
192 * would give a falsely clean cache entry. The mtime and
193 * length match the cache, and other stat fields do not change.
195 * We could detect this at update-index time (the cache entry
196 * being registered/updated records the same time as "now")
197 * and delay the return from git-update-index, but that would
198 * effectively mean we can make at most one commit per second,
199 * which is not acceptable. Instead, we check cache entries
200 * whose mtime are the same as the index file timestamp more
201 * carefully than others.
203 if (!changed &&
204 index_file_timestamp &&
205 index_file_timestamp <= ntohl(ce->ce_mtime.sec)) {
206 if (assume_racy_is_modified)
207 changed |= DATA_CHANGED;
208 else
209 changed |= ce_modified_check_fs(ce, st);
212 return changed;
215 int ce_modified(struct cache_entry *ce, struct stat *st, int really)
217 int changed, changed_fs;
218 changed = ce_match_stat(ce, st, really);
219 if (!changed)
220 return 0;
222 * If the mode or type has changed, there's no point in trying
223 * to refresh the entry - it's not going to match
225 if (changed & (MODE_CHANGED | TYPE_CHANGED))
226 return changed;
228 /* Immediately after read-tree or update-index --cacheinfo,
229 * the length field is zero. For other cases the ce_size
230 * should match the SHA1 recorded in the index entry.
232 if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
233 return changed;
235 changed_fs = ce_modified_check_fs(ce, st);
236 if (changed_fs)
237 return changed | changed_fs;
238 return 0;
241 int base_name_compare(const char *name1, int len1, int mode1,
242 const char *name2, int len2, int mode2)
244 unsigned char c1, c2;
245 int len = len1 < len2 ? len1 : len2;
246 int cmp;
248 cmp = memcmp(name1, name2, len);
249 if (cmp)
250 return cmp;
251 c1 = name1[len];
252 c2 = name2[len];
253 if (!c1 && S_ISDIR(mode1))
254 c1 = '/';
255 if (!c2 && S_ISDIR(mode2))
256 c2 = '/';
257 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
260 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
262 int len1 = flags1 & CE_NAMEMASK;
263 int len2 = flags2 & CE_NAMEMASK;
264 int len = len1 < len2 ? len1 : len2;
265 int cmp;
267 cmp = memcmp(name1, name2, len);
268 if (cmp)
269 return cmp;
270 if (len1 < len2)
271 return -1;
272 if (len1 > len2)
273 return 1;
275 /* Compare stages */
276 flags1 &= CE_STAGEMASK;
277 flags2 &= CE_STAGEMASK;
279 if (flags1 < flags2)
280 return -1;
281 if (flags1 > flags2)
282 return 1;
283 return 0;
286 int cache_name_pos(const char *name, int namelen)
288 int first, last;
290 first = 0;
291 last = active_nr;
292 while (last > first) {
293 int next = (last + first) >> 1;
294 struct cache_entry *ce = active_cache[next];
295 int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
296 if (!cmp)
297 return next;
298 if (cmp < 0) {
299 last = next;
300 continue;
302 first = next+1;
304 return -first-1;
307 /* Remove entry, return true if there are more entries to go.. */
308 int remove_cache_entry_at(int pos)
310 active_cache_changed = 1;
311 active_nr--;
312 if (pos >= active_nr)
313 return 0;
314 memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
315 return 1;
318 int remove_file_from_cache(const char *path)
320 int pos = cache_name_pos(path, strlen(path));
321 if (pos < 0)
322 pos = -pos-1;
323 while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
324 remove_cache_entry_at(pos);
325 return 0;
328 int add_file_to_cache(const char *path, int verbose)
330 int size, namelen;
331 struct stat st;
332 struct cache_entry *ce;
334 if (lstat(path, &st))
335 die("%s: unable to stat (%s)", path, strerror(errno));
337 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode))
338 die("%s: can only add regular files or symbolic links", path);
340 namelen = strlen(path);
341 size = cache_entry_size(namelen);
342 ce = xcalloc(1, size);
343 memcpy(ce->name, path, namelen);
344 ce->ce_flags = htons(namelen);
345 fill_stat_cache_info(ce, &st);
347 if (trust_executable_bit && has_symlinks)
348 ce->ce_mode = create_ce_mode(st.st_mode);
349 else {
350 /* If there is an existing entry, pick the mode bits and type
351 * from it, otherwise assume unexecutable regular file.
353 struct cache_entry *ent;
354 int pos = cache_name_pos(path, namelen);
356 ent = (0 <= pos) ? active_cache[pos] : NULL;
357 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
360 if (index_path(ce->sha1, path, &st, 1))
361 die("unable to index file %s", path);
362 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
363 die("unable to add %s to index",path);
364 if (verbose)
365 printf("add '%s'\n", path);
366 cache_tree_invalidate_path(active_cache_tree, path);
367 return 0;
370 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
372 int len = ce_namelen(a);
373 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
376 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
378 const char *match, *name;
379 int len;
381 if (!pathspec)
382 return 1;
384 len = ce_namelen(ce);
385 name = ce->name;
386 while ((match = *pathspec++) != NULL) {
387 int matchlen = strlen(match);
388 if (matchlen > len)
389 continue;
390 if (memcmp(name, match, matchlen))
391 continue;
392 if (matchlen && name[matchlen-1] == '/')
393 return 1;
394 if (name[matchlen] == '/' || !name[matchlen])
395 return 1;
396 if (!matchlen)
397 return 1;
399 return 0;
403 * We fundamentally don't like some paths: we don't want
404 * dot or dot-dot anywhere, and for obvious reasons don't
405 * want to recurse into ".git" either.
407 * Also, we don't want double slashes or slashes at the
408 * end that can make pathnames ambiguous.
410 static int verify_dotfile(const char *rest)
413 * The first character was '.', but that
414 * has already been discarded, we now test
415 * the rest.
417 switch (*rest) {
418 /* "." is not allowed */
419 case '\0': case '/':
420 return 0;
423 * ".git" followed by NUL or slash is bad. This
424 * shares the path end test with the ".." case.
426 case 'g':
427 if (rest[1] != 'i')
428 break;
429 if (rest[2] != 't')
430 break;
431 rest += 2;
432 /* fallthrough */
433 case '.':
434 if (rest[1] == '\0' || rest[1] == '/')
435 return 0;
437 return 1;
440 int verify_path(const char *path)
442 char c;
444 goto inside;
445 for (;;) {
446 if (!c)
447 return 1;
448 if (c == '/') {
449 inside:
450 c = *path++;
451 switch (c) {
452 default:
453 continue;
454 case '/': case '\0':
455 break;
456 case '.':
457 if (verify_dotfile(path))
458 continue;
460 return 0;
462 c = *path++;
467 * Do we have another file that has the beginning components being a
468 * proper superset of the name we're trying to add?
470 static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replace)
472 int retval = 0;
473 int len = ce_namelen(ce);
474 int stage = ce_stage(ce);
475 const char *name = ce->name;
477 while (pos < active_nr) {
478 struct cache_entry *p = active_cache[pos++];
480 if (len >= ce_namelen(p))
481 break;
482 if (memcmp(name, p->name, len))
483 break;
484 if (ce_stage(p) != stage)
485 continue;
486 if (p->name[len] != '/')
487 continue;
488 if (!ce_stage(p) && !p->ce_mode)
489 continue;
490 retval = -1;
491 if (!ok_to_replace)
492 break;
493 remove_cache_entry_at(--pos);
495 return retval;
499 * Do we have another file with a pathname that is a proper
500 * subset of the name we're trying to add?
502 static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace)
504 int retval = 0;
505 int stage = ce_stage(ce);
506 const char *name = ce->name;
507 const char *slash = name + ce_namelen(ce);
509 for (;;) {
510 int len;
512 for (;;) {
513 if (*--slash == '/')
514 break;
515 if (slash <= ce->name)
516 return retval;
518 len = slash - name;
520 pos = cache_name_pos(name, ntohs(create_ce_flags(len, stage)));
521 if (pos >= 0) {
523 * Found one, but not so fast. This could
524 * be a marker that says "I was here, but
525 * I am being removed". Such an entry is
526 * not a part of the resulting tree, and
527 * it is Ok to have a directory at the same
528 * path.
530 if (stage || active_cache[pos]->ce_mode) {
531 retval = -1;
532 if (!ok_to_replace)
533 break;
534 remove_cache_entry_at(pos);
535 continue;
538 else
539 pos = -pos-1;
542 * Trivial optimization: if we find an entry that
543 * already matches the sub-directory, then we know
544 * we're ok, and we can exit.
546 while (pos < active_nr) {
547 struct cache_entry *p = active_cache[pos];
548 if ((ce_namelen(p) <= len) ||
549 (p->name[len] != '/') ||
550 memcmp(p->name, name, len))
551 break; /* not our subdirectory */
552 if (ce_stage(p) == stage && (stage || p->ce_mode))
553 /* p is at the same stage as our entry, and
554 * is a subdirectory of what we are looking
555 * at, so we cannot have conflicts at our
556 * level or anything shorter.
558 return retval;
559 pos++;
562 return retval;
565 /* We may be in a situation where we already have path/file and path
566 * is being added, or we already have path and path/file is being
567 * added. Either one would result in a nonsense tree that has path
568 * twice when git-write-tree tries to write it out. Prevent it.
570 * If ok-to-replace is specified, we remove the conflicting entries
571 * from the cache so the caller should recompute the insert position.
572 * When this happens, we return non-zero.
574 static int check_file_directory_conflict(const struct cache_entry *ce, int pos, int ok_to_replace)
576 int retval;
579 * When ce is an "I am going away" entry, we allow it to be added
581 if (!ce_stage(ce) && !ce->ce_mode)
582 return 0;
585 * We check if the path is a sub-path of a subsequent pathname
586 * first, since removing those will not change the position
587 * in the array.
589 retval = has_file_name(ce, pos, ok_to_replace);
592 * Then check if the path might have a clashing sub-directory
593 * before it.
595 return retval + has_dir_name(ce, pos, ok_to_replace);
598 int add_cache_entry(struct cache_entry *ce, int option)
600 int pos;
601 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
602 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
603 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
605 pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
607 /* existing match? Just replace it. */
608 if (pos >= 0) {
609 active_cache_changed = 1;
610 active_cache[pos] = ce;
611 return 0;
613 pos = -pos-1;
616 * Inserting a merged entry ("stage 0") into the index
617 * will always replace all non-merged entries..
619 if (pos < active_nr && ce_stage(ce) == 0) {
620 while (ce_same_name(active_cache[pos], ce)) {
621 ok_to_add = 1;
622 if (!remove_cache_entry_at(pos))
623 break;
627 if (!ok_to_add)
628 return -1;
629 if (!verify_path(ce->name))
630 return -1;
632 if (!skip_df_check &&
633 check_file_directory_conflict(ce, pos, ok_to_replace)) {
634 if (!ok_to_replace)
635 return error("'%s' appears as both a file and as a directory", ce->name);
636 pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
637 pos = -pos-1;
640 /* Make sure the array is big enough .. */
641 if (active_nr == active_alloc) {
642 active_alloc = alloc_nr(active_alloc);
643 active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
646 /* Add it in.. */
647 active_nr++;
648 if (active_nr > pos)
649 memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
650 active_cache[pos] = ce;
651 active_cache_changed = 1;
652 return 0;
656 * "refresh" does not calculate a new sha1 file or bring the
657 * cache up-to-date for mode/content changes. But what it
658 * _does_ do is to "re-match" the stat information of a file
659 * with the cache, so that you can refresh the cache for a
660 * file that hasn't been changed but where the stat entry is
661 * out of date.
663 * For example, you'd want to do this after doing a "git-read-tree",
664 * to link up the stat cache details with the proper files.
666 static struct cache_entry *refresh_cache_ent(struct cache_entry *ce, int really, int *err)
668 struct stat st;
669 struct cache_entry *updated;
670 int changed, size;
672 if (lstat(ce->name, &st) < 0) {
673 if (err)
674 *err = errno;
675 return NULL;
678 changed = ce_match_stat(ce, &st, really);
679 if (!changed) {
680 if (really && assume_unchanged &&
681 !(ce->ce_flags & htons(CE_VALID)))
682 ; /* mark this one VALID again */
683 else
684 return ce;
687 if (ce_modified(ce, &st, really)) {
688 if (err)
689 *err = EINVAL;
690 return NULL;
693 size = ce_size(ce);
694 updated = xmalloc(size);
695 memcpy(updated, ce, size);
696 fill_stat_cache_info(updated, &st);
698 /* In this case, if really is not set, we should leave
699 * CE_VALID bit alone. Otherwise, paths marked with
700 * --no-assume-unchanged (i.e. things to be edited) will
701 * reacquire CE_VALID bit automatically, which is not
702 * really what we want.
704 if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
705 updated->ce_flags &= ~htons(CE_VALID);
707 return updated;
710 int refresh_cache(unsigned int flags)
712 int i;
713 int has_errors = 0;
714 int really = (flags & REFRESH_REALLY) != 0;
715 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
716 int quiet = (flags & REFRESH_QUIET) != 0;
717 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
719 for (i = 0; i < active_nr; i++) {
720 struct cache_entry *ce, *new;
721 int cache_errno = 0;
723 ce = active_cache[i];
724 if (ce_stage(ce)) {
725 while ((i < active_nr) &&
726 ! strcmp(active_cache[i]->name, ce->name))
727 i++;
728 i--;
729 if (allow_unmerged)
730 continue;
731 printf("%s: needs merge\n", ce->name);
732 has_errors = 1;
733 continue;
736 new = refresh_cache_ent(ce, really, &cache_errno);
737 if (new == ce)
738 continue;
739 if (!new) {
740 if (not_new && cache_errno == ENOENT)
741 continue;
742 if (really && cache_errno == EINVAL) {
743 /* If we are doing --really-refresh that
744 * means the index is not valid anymore.
746 ce->ce_flags &= ~htons(CE_VALID);
747 active_cache_changed = 1;
749 if (quiet)
750 continue;
751 printf("%s: needs update\n", ce->name);
752 has_errors = 1;
753 continue;
755 active_cache_changed = 1;
756 /* You can NOT just free active_cache[i] here, since it
757 * might not be necessarily malloc()ed but can also come
758 * from mmap(). */
759 active_cache[i] = new;
761 return has_errors;
764 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
766 return refresh_cache_ent(ce, really, NULL);
769 static int verify_hdr(struct cache_header *hdr, unsigned long size)
771 SHA_CTX c;
772 unsigned char sha1[20];
774 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
775 return error("bad signature");
776 if (hdr->hdr_version != htonl(2))
777 return error("bad index version");
778 SHA1_Init(&c);
779 SHA1_Update(&c, hdr, size - 20);
780 SHA1_Final(sha1, &c);
781 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
782 return error("bad index file sha1 signature");
783 return 0;
786 static int read_index_extension(const char *ext, void *data, unsigned long sz)
788 switch (CACHE_EXT(ext)) {
789 case CACHE_EXT_TREE:
790 active_cache_tree = cache_tree_read(data, sz);
791 break;
792 default:
793 if (*ext < 'A' || 'Z' < *ext)
794 return error("index uses %.4s extension, which we do not understand",
795 ext);
796 fprintf(stderr, "ignoring %.4s extension\n", ext);
797 break;
799 return 0;
802 int read_cache(void)
804 return read_cache_from(get_index_file());
807 /* remember to discard_cache() before reading a different cache! */
808 int read_cache_from(const char *path)
810 int fd, i;
811 struct stat st;
812 unsigned long offset;
813 struct cache_header *hdr;
815 errno = EBUSY;
816 if (cache_mmap)
817 return active_nr;
819 errno = ENOENT;
820 index_file_timestamp = 0;
821 fd = open(path, O_RDONLY);
822 if (fd < 0) {
823 if (errno == ENOENT)
824 return 0;
825 die("index file open failed (%s)", strerror(errno));
828 if (!fstat(fd, &st)) {
829 cache_mmap_size = xsize_t(st.st_size);
830 errno = EINVAL;
831 if (cache_mmap_size >= sizeof(struct cache_header) + 20)
832 cache_mmap = xmmap(NULL, cache_mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
833 else
834 die("index file smaller than expected");
835 } else
836 die("cannot stat the open index (%s)", strerror(errno));
837 close(fd);
839 hdr = cache_mmap;
840 if (verify_hdr(hdr, cache_mmap_size) < 0)
841 goto unmap;
843 active_nr = ntohl(hdr->hdr_entries);
844 active_alloc = alloc_nr(active_nr);
845 active_cache = xcalloc(active_alloc, sizeof(struct cache_entry *));
847 offset = sizeof(*hdr);
848 for (i = 0; i < active_nr; i++) {
849 struct cache_entry *ce = (struct cache_entry *) ((char *) cache_mmap + offset);
850 offset = offset + ce_size(ce);
851 active_cache[i] = ce;
853 index_file_timestamp = st.st_mtime;
854 while (offset <= cache_mmap_size - 20 - 8) {
855 /* After an array of active_nr index entries,
856 * there can be arbitrary number of extended
857 * sections, each of which is prefixed with
858 * extension name (4-byte) and section length
859 * in 4-byte network byte order.
861 unsigned long extsize;
862 memcpy(&extsize, (char *) cache_mmap + offset + 4, 4);
863 extsize = ntohl(extsize);
864 if (read_index_extension(((const char *) cache_mmap) + offset,
865 (char *) cache_mmap + offset + 8,
866 extsize) < 0)
867 goto unmap;
868 offset += 8;
869 offset += extsize;
871 return active_nr;
873 unmap:
874 munmap(cache_mmap, cache_mmap_size);
875 errno = EINVAL;
876 die("index file corrupt");
879 int discard_cache(void)
881 int ret;
883 active_nr = active_cache_changed = 0;
884 index_file_timestamp = 0;
885 cache_tree_free(&active_cache_tree);
886 if (cache_mmap == NULL)
887 return 0;
888 ret = munmap(cache_mmap, cache_mmap_size);
889 cache_mmap = NULL;
890 cache_mmap_size = 0;
892 /* no need to throw away allocated active_cache */
893 return ret;
896 #define WRITE_BUFFER_SIZE 8192
897 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
898 static unsigned long write_buffer_len;
900 static int ce_write_flush(SHA_CTX *context, int fd)
902 unsigned int buffered = write_buffer_len;
903 if (buffered) {
904 SHA1_Update(context, write_buffer, buffered);
905 if (write_in_full(fd, write_buffer, buffered) != buffered)
906 return -1;
907 write_buffer_len = 0;
909 return 0;
912 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
914 while (len) {
915 unsigned int buffered = write_buffer_len;
916 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
917 if (partial > len)
918 partial = len;
919 memcpy(write_buffer + buffered, data, partial);
920 buffered += partial;
921 if (buffered == WRITE_BUFFER_SIZE) {
922 write_buffer_len = buffered;
923 if (ce_write_flush(context, fd))
924 return -1;
925 buffered = 0;
927 write_buffer_len = buffered;
928 len -= partial;
929 data = (char *) data + partial;
931 return 0;
934 static int write_index_ext_header(SHA_CTX *context, int fd,
935 unsigned int ext, unsigned int sz)
937 ext = htonl(ext);
938 sz = htonl(sz);
939 return ((ce_write(context, fd, &ext, 4) < 0) ||
940 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
943 static int ce_flush(SHA_CTX *context, int fd)
945 unsigned int left = write_buffer_len;
947 if (left) {
948 write_buffer_len = 0;
949 SHA1_Update(context, write_buffer, left);
952 /* Flush first if not enough space for SHA1 signature */
953 if (left + 20 > WRITE_BUFFER_SIZE) {
954 if (write_in_full(fd, write_buffer, left) != left)
955 return -1;
956 left = 0;
959 /* Append the SHA1 signature at the end */
960 SHA1_Final(write_buffer + left, context);
961 left += 20;
962 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
965 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
968 * The only thing we care about in this function is to smudge the
969 * falsely clean entry due to touch-update-touch race, so we leave
970 * everything else as they are. We are called for entries whose
971 * ce_mtime match the index file mtime.
973 struct stat st;
975 if (lstat(ce->name, &st) < 0)
976 return;
977 if (ce_match_stat_basic(ce, &st))
978 return;
979 if (ce_modified_check_fs(ce, &st)) {
980 /* This is "racily clean"; smudge it. Note that this
981 * is a tricky code. At first glance, it may appear
982 * that it can break with this sequence:
984 * $ echo xyzzy >frotz
985 * $ git-update-index --add frotz
986 * $ : >frotz
987 * $ sleep 3
988 * $ echo filfre >nitfol
989 * $ git-update-index --add nitfol
991 * but it does not. When the second update-index runs,
992 * it notices that the entry "frotz" has the same timestamp
993 * as index, and if we were to smudge it by resetting its
994 * size to zero here, then the object name recorded
995 * in index is the 6-byte file but the cached stat information
996 * becomes zero --- which would then match what we would
997 * obtain from the filesystem next time we stat("frotz").
999 * However, the second update-index, before calling
1000 * this function, notices that the cached size is 6
1001 * bytes and what is on the filesystem is an empty
1002 * file, and never calls us, so the cached size information
1003 * for "frotz" stays 6 which does not match the filesystem.
1005 ce->ce_size = htonl(0);
1009 int write_cache(int newfd, struct cache_entry **cache, int entries)
1011 SHA_CTX c;
1012 struct cache_header hdr;
1013 int i, removed;
1015 for (i = removed = 0; i < entries; i++)
1016 if (!cache[i]->ce_mode)
1017 removed++;
1019 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1020 hdr.hdr_version = htonl(2);
1021 hdr.hdr_entries = htonl(entries - removed);
1023 SHA1_Init(&c);
1024 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1025 return -1;
1027 for (i = 0; i < entries; i++) {
1028 struct cache_entry *ce = cache[i];
1029 if (!ce->ce_mode)
1030 continue;
1031 if (index_file_timestamp &&
1032 index_file_timestamp <= ntohl(ce->ce_mtime.sec))
1033 ce_smudge_racily_clean_entry(ce);
1034 if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
1035 return -1;
1038 /* Write extension data here */
1039 if (active_cache_tree) {
1040 unsigned long sz;
1041 void *data = cache_tree_write(active_cache_tree, &sz);
1042 if (data &&
1043 !write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sz) &&
1044 !ce_write(&c, newfd, data, sz))
1045 free(data);
1046 else {
1047 free(data);
1048 return -1;
1051 return ce_flush(&c, newfd);