implement some resilience against pack corruptions
[git/trast.git] / read-cache.c
blobf83de8c4158e08bab3e0445c4d15caf2d4d105aa
1 /*
2 * GIT - The information manager from hell
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.
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.
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 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
28 istate->cache[nr] = ce;
29 add_name_hash(istate, ce);
32 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
34 struct cache_entry *old = istate->cache[nr];
36 remove_name_hash(old);
37 set_index_entry(istate, nr, ce);
38 istate->cache_changed = 1;
42 * This only updates the "non-critical" parts of the directory
43 * cache, ie the parts that aren't tracked by GIT, and only used
44 * to validate the cache.
46 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
48 ce->ce_ctime = st->st_ctime;
49 ce->ce_mtime = st->st_mtime;
50 ce->ce_dev = st->st_dev;
51 ce->ce_ino = st->st_ino;
52 ce->ce_uid = st->st_uid;
53 ce->ce_gid = st->st_gid;
54 ce->ce_size = st->st_size;
56 if (assume_unchanged)
57 ce->ce_flags |= CE_VALID;
59 if (S_ISREG(st->st_mode))
60 ce_mark_uptodate(ce);
63 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
65 int match = -1;
66 int fd = open(ce->name, O_RDONLY);
68 if (fd >= 0) {
69 unsigned char sha1[20];
70 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
71 match = hashcmp(sha1, ce->sha1);
72 /* index_fd() closed the file descriptor already */
74 return match;
77 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
79 int match = -1;
80 char *target;
81 void *buffer;
82 unsigned long size;
83 enum object_type type;
84 int len;
86 target = xmalloc(expected_size);
87 len = readlink(ce->name, target, expected_size);
88 if (len != expected_size) {
89 free(target);
90 return -1;
92 buffer = read_sha1_file(ce->sha1, &type, &size);
93 if (!buffer) {
94 free(target);
95 return -1;
97 if (size == expected_size)
98 match = memcmp(buffer, target, size);
99 free(buffer);
100 free(target);
101 return match;
104 static int ce_compare_gitlink(struct cache_entry *ce)
106 unsigned char sha1[20];
109 * We don't actually require that the .git directory
110 * under GITLINK directory be a valid git directory. It
111 * might even be missing (in case nobody populated that
112 * sub-project).
114 * If so, we consider it always to match.
116 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
117 return 0;
118 return hashcmp(sha1, ce->sha1);
121 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
123 switch (st->st_mode & S_IFMT) {
124 case S_IFREG:
125 if (ce_compare_data(ce, st))
126 return DATA_CHANGED;
127 break;
128 case S_IFLNK:
129 if (ce_compare_link(ce, xsize_t(st->st_size)))
130 return DATA_CHANGED;
131 break;
132 case S_IFDIR:
133 if (S_ISGITLINK(ce->ce_mode))
134 return 0;
135 default:
136 return TYPE_CHANGED;
138 return 0;
141 static int is_empty_blob_sha1(const unsigned char *sha1)
143 static const unsigned char empty_blob_sha1[20] = {
144 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
145 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
148 return !hashcmp(sha1, empty_blob_sha1);
151 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
153 unsigned int changed = 0;
155 if (ce->ce_flags & CE_REMOVE)
156 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
158 switch (ce->ce_mode & S_IFMT) {
159 case S_IFREG:
160 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
161 /* We consider only the owner x bit to be relevant for
162 * "mode changes"
164 if (trust_executable_bit &&
165 (0100 & (ce->ce_mode ^ st->st_mode)))
166 changed |= MODE_CHANGED;
167 break;
168 case S_IFLNK:
169 if (!S_ISLNK(st->st_mode) &&
170 (has_symlinks || !S_ISREG(st->st_mode)))
171 changed |= TYPE_CHANGED;
172 break;
173 case S_IFGITLINK:
174 if (!S_ISDIR(st->st_mode))
175 changed |= TYPE_CHANGED;
176 else if (ce_compare_gitlink(ce))
177 changed |= DATA_CHANGED;
178 return changed;
179 default:
180 die("internal error: ce_mode is %o", ce->ce_mode);
182 if (ce->ce_mtime != (unsigned int) st->st_mtime)
183 changed |= MTIME_CHANGED;
184 if (ce->ce_ctime != (unsigned int) st->st_ctime)
185 changed |= CTIME_CHANGED;
187 if (ce->ce_uid != (unsigned int) st->st_uid ||
188 ce->ce_gid != (unsigned int) st->st_gid)
189 changed |= OWNER_CHANGED;
190 if (ce->ce_ino != (unsigned int) st->st_ino)
191 changed |= INODE_CHANGED;
193 #ifdef USE_STDEV
195 * st_dev breaks on network filesystems where different
196 * clients will have different views of what "device"
197 * the filesystem is on
199 if (ce->ce_dev != (unsigned int) st->st_dev)
200 changed |= INODE_CHANGED;
201 #endif
203 if (ce->ce_size != (unsigned int) st->st_size)
204 changed |= DATA_CHANGED;
206 /* Racily smudged entry? */
207 if (!ce->ce_size) {
208 if (!is_empty_blob_sha1(ce->sha1))
209 changed |= DATA_CHANGED;
212 return changed;
215 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
217 return (!S_ISGITLINK(ce->ce_mode) &&
218 istate->timestamp &&
219 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
222 int ie_match_stat(const struct index_state *istate,
223 struct cache_entry *ce, struct stat *st,
224 unsigned int options)
226 unsigned int changed;
227 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
228 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
231 * If it's marked as always valid in the index, it's
232 * valid whatever the checked-out copy says.
234 if (!ignore_valid && (ce->ce_flags & CE_VALID))
235 return 0;
237 changed = ce_match_stat_basic(ce, st);
240 * Within 1 second of this sequence:
241 * echo xyzzy >file && git-update-index --add file
242 * running this command:
243 * echo frotz >file
244 * would give a falsely clean cache entry. The mtime and
245 * length match the cache, and other stat fields do not change.
247 * We could detect this at update-index time (the cache entry
248 * being registered/updated records the same time as "now")
249 * and delay the return from git-update-index, but that would
250 * effectively mean we can make at most one commit per second,
251 * which is not acceptable. Instead, we check cache entries
252 * whose mtime are the same as the index file timestamp more
253 * carefully than others.
255 if (!changed && is_racy_timestamp(istate, ce)) {
256 if (assume_racy_is_modified)
257 changed |= DATA_CHANGED;
258 else
259 changed |= ce_modified_check_fs(ce, st);
262 return changed;
265 int ie_modified(const struct index_state *istate,
266 struct cache_entry *ce, struct stat *st, unsigned int options)
268 int changed, changed_fs;
270 changed = ie_match_stat(istate, ce, st, options);
271 if (!changed)
272 return 0;
274 * If the mode or type has changed, there's no point in trying
275 * to refresh the entry - it's not going to match
277 if (changed & (MODE_CHANGED | TYPE_CHANGED))
278 return changed;
280 /* Immediately after read-tree or update-index --cacheinfo,
281 * the length field is zero. For other cases the ce_size
282 * should match the SHA1 recorded in the index entry.
284 if ((changed & DATA_CHANGED) && ce->ce_size != 0)
285 return changed;
287 changed_fs = ce_modified_check_fs(ce, st);
288 if (changed_fs)
289 return changed | changed_fs;
290 return 0;
293 int base_name_compare(const char *name1, int len1, int mode1,
294 const char *name2, int len2, int mode2)
296 unsigned char c1, c2;
297 int len = len1 < len2 ? len1 : len2;
298 int cmp;
300 cmp = memcmp(name1, name2, len);
301 if (cmp)
302 return cmp;
303 c1 = name1[len];
304 c2 = name2[len];
305 if (!c1 && S_ISDIR(mode1))
306 c1 = '/';
307 if (!c2 && S_ISDIR(mode2))
308 c2 = '/';
309 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
313 * df_name_compare() is identical to base_name_compare(), except it
314 * compares conflicting directory/file entries as equal. Note that
315 * while a directory name compares as equal to a regular file, they
316 * then individually compare _differently_ to a filename that has
317 * a dot after the basename (because '\0' < '.' < '/').
319 * This is used by routines that want to traverse the git namespace
320 * but then handle conflicting entries together when possible.
322 int df_name_compare(const char *name1, int len1, int mode1,
323 const char *name2, int len2, int mode2)
325 int len = len1 < len2 ? len1 : len2, cmp;
326 unsigned char c1, c2;
328 cmp = memcmp(name1, name2, len);
329 if (cmp)
330 return cmp;
331 /* Directories and files compare equal (same length, same name) */
332 if (len1 == len2)
333 return 0;
334 c1 = name1[len];
335 if (!c1 && S_ISDIR(mode1))
336 c1 = '/';
337 c2 = name2[len];
338 if (!c2 && S_ISDIR(mode2))
339 c2 = '/';
340 if (c1 == '/' && !c2)
341 return 0;
342 if (c2 == '/' && !c1)
343 return 0;
344 return c1 - c2;
347 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
349 int len1 = flags1 & CE_NAMEMASK;
350 int len2 = flags2 & CE_NAMEMASK;
351 int len = len1 < len2 ? len1 : len2;
352 int cmp;
354 cmp = memcmp(name1, name2, len);
355 if (cmp)
356 return cmp;
357 if (len1 < len2)
358 return -1;
359 if (len1 > len2)
360 return 1;
362 /* Compare stages */
363 flags1 &= CE_STAGEMASK;
364 flags2 &= CE_STAGEMASK;
366 if (flags1 < flags2)
367 return -1;
368 if (flags1 > flags2)
369 return 1;
370 return 0;
373 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
375 int first, last;
377 first = 0;
378 last = istate->cache_nr;
379 while (last > first) {
380 int next = (last + first) >> 1;
381 struct cache_entry *ce = istate->cache[next];
382 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
383 if (!cmp)
384 return next;
385 if (cmp < 0) {
386 last = next;
387 continue;
389 first = next+1;
391 return -first-1;
394 /* Remove entry, return true if there are more entries to go.. */
395 int remove_index_entry_at(struct index_state *istate, int pos)
397 struct cache_entry *ce = istate->cache[pos];
399 remove_name_hash(ce);
400 istate->cache_changed = 1;
401 istate->cache_nr--;
402 if (pos >= istate->cache_nr)
403 return 0;
404 memmove(istate->cache + pos,
405 istate->cache + pos + 1,
406 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
407 return 1;
410 int remove_file_from_index(struct index_state *istate, const char *path)
412 int pos = index_name_pos(istate, path, strlen(path));
413 if (pos < 0)
414 pos = -pos-1;
415 cache_tree_invalidate_path(istate->cache_tree, path);
416 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
417 remove_index_entry_at(istate, pos);
418 return 0;
421 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
423 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
426 static int index_name_pos_also_unmerged(struct index_state *istate,
427 const char *path, int namelen)
429 int pos = index_name_pos(istate, path, namelen);
430 struct cache_entry *ce;
432 if (pos >= 0)
433 return pos;
435 /* maybe unmerged? */
436 pos = -1 - pos;
437 if (pos >= istate->cache_nr ||
438 compare_name((ce = istate->cache[pos]), path, namelen))
439 return -1;
441 /* order of preference: stage 2, 1, 3 */
442 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
443 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
444 !compare_name(ce, path, namelen))
445 pos++;
446 return pos;
449 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
451 int len = ce_namelen(ce);
452 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
456 * If we add a filename that aliases in the cache, we will use the
457 * name that we already have - but we don't want to update the same
458 * alias twice, because that implies that there were actually two
459 * different files with aliasing names!
461 * So we use the CE_ADDED flag to verify that the alias was an old
462 * one before we accept it as
464 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
466 int len;
467 struct cache_entry *new;
469 if (alias->ce_flags & CE_ADDED)
470 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
472 /* Ok, create the new entry using the name of the existing alias */
473 len = ce_namelen(alias);
474 new = xcalloc(1, cache_entry_size(len));
475 memcpy(new->name, alias->name, len);
476 copy_cache_entry(new, ce);
477 free(ce);
478 return new;
481 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
483 int size, namelen, was_same;
484 mode_t st_mode = st->st_mode;
485 struct cache_entry *ce, *alias;
486 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
487 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
488 int pretend = flags & ADD_CACHE_PRETEND;
490 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
491 return error("%s: can only add regular files, symbolic links or git-directories", path);
493 namelen = strlen(path);
494 if (S_ISDIR(st_mode)) {
495 while (namelen && path[namelen-1] == '/')
496 namelen--;
498 size = cache_entry_size(namelen);
499 ce = xcalloc(1, size);
500 memcpy(ce->name, path, namelen);
501 ce->ce_flags = namelen;
502 fill_stat_cache_info(ce, st);
504 if (trust_executable_bit && has_symlinks)
505 ce->ce_mode = create_ce_mode(st_mode);
506 else {
507 /* If there is an existing entry, pick the mode bits and type
508 * from it, otherwise assume unexecutable regular file.
510 struct cache_entry *ent;
511 int pos = index_name_pos_also_unmerged(istate, path, namelen);
513 ent = (0 <= pos) ? istate->cache[pos] : NULL;
514 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
517 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
518 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
519 /* Nothing changed, really */
520 free(ce);
521 ce_mark_uptodate(alias);
522 alias->ce_flags |= CE_ADDED;
523 return 0;
525 if (index_path(ce->sha1, path, st, 1))
526 return error("unable to index file %s", path);
527 if (ignore_case && alias && different_name(ce, alias))
528 ce = create_alias_ce(ce, alias);
529 ce->ce_flags |= CE_ADDED;
531 /* It was suspected to be recily clean, but it turns out to be Ok */
532 was_same = (alias &&
533 !ce_stage(alias) &&
534 !hashcmp(alias->sha1, ce->sha1) &&
535 ce->ce_mode == alias->ce_mode);
537 if (pretend)
539 else if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
540 return error("unable to add %s to index",path);
541 if (verbose && !was_same)
542 printf("add '%s'\n", path);
543 return 0;
546 int add_file_to_index(struct index_state *istate, const char *path, int flags)
548 struct stat st;
549 if (lstat(path, &st))
550 die("%s: unable to stat (%s)", path, strerror(errno));
551 return add_to_index(istate, path, &st, flags);
554 struct cache_entry *make_cache_entry(unsigned int mode,
555 const unsigned char *sha1, const char *path, int stage,
556 int refresh)
558 int size, len;
559 struct cache_entry *ce;
561 if (!verify_path(path))
562 return NULL;
564 len = strlen(path);
565 size = cache_entry_size(len);
566 ce = xcalloc(1, size);
568 hashcpy(ce->sha1, sha1);
569 memcpy(ce->name, path, len);
570 ce->ce_flags = create_ce_flags(len, stage);
571 ce->ce_mode = create_ce_mode(mode);
573 if (refresh)
574 return refresh_cache_entry(ce, 0);
576 return ce;
579 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
581 int len = ce_namelen(a);
582 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
585 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
587 const char *match, *name;
588 int len;
590 if (!pathspec)
591 return 1;
593 len = ce_namelen(ce);
594 name = ce->name;
595 while ((match = *pathspec++) != NULL) {
596 int matchlen = strlen(match);
597 if (matchlen > len)
598 continue;
599 if (memcmp(name, match, matchlen))
600 continue;
601 if (matchlen && name[matchlen-1] == '/')
602 return 1;
603 if (name[matchlen] == '/' || !name[matchlen])
604 return 1;
605 if (!matchlen)
606 return 1;
608 return 0;
612 * We fundamentally don't like some paths: we don't want
613 * dot or dot-dot anywhere, and for obvious reasons don't
614 * want to recurse into ".git" either.
616 * Also, we don't want double slashes or slashes at the
617 * end that can make pathnames ambiguous.
619 static int verify_dotfile(const char *rest)
622 * The first character was '.', but that
623 * has already been discarded, we now test
624 * the rest.
626 switch (*rest) {
627 /* "." is not allowed */
628 case '\0': case '/':
629 return 0;
632 * ".git" followed by NUL or slash is bad. This
633 * shares the path end test with the ".." case.
635 case 'g':
636 if (rest[1] != 'i')
637 break;
638 if (rest[2] != 't')
639 break;
640 rest += 2;
641 /* fallthrough */
642 case '.':
643 if (rest[1] == '\0' || rest[1] == '/')
644 return 0;
646 return 1;
649 int verify_path(const char *path)
651 char c;
653 goto inside;
654 for (;;) {
655 if (!c)
656 return 1;
657 if (c == '/') {
658 inside:
659 c = *path++;
660 switch (c) {
661 default:
662 continue;
663 case '/': case '\0':
664 break;
665 case '.':
666 if (verify_dotfile(path))
667 continue;
669 return 0;
671 c = *path++;
676 * Do we have another file that has the beginning components being a
677 * proper superset of the name we're trying to add?
679 static int has_file_name(struct index_state *istate,
680 const struct cache_entry *ce, int pos, int ok_to_replace)
682 int retval = 0;
683 int len = ce_namelen(ce);
684 int stage = ce_stage(ce);
685 const char *name = ce->name;
687 while (pos < istate->cache_nr) {
688 struct cache_entry *p = istate->cache[pos++];
690 if (len >= ce_namelen(p))
691 break;
692 if (memcmp(name, p->name, len))
693 break;
694 if (ce_stage(p) != stage)
695 continue;
696 if (p->name[len] != '/')
697 continue;
698 if (p->ce_flags & CE_REMOVE)
699 continue;
700 retval = -1;
701 if (!ok_to_replace)
702 break;
703 remove_index_entry_at(istate, --pos);
705 return retval;
709 * Do we have another file with a pathname that is a proper
710 * subset of the name we're trying to add?
712 static int has_dir_name(struct index_state *istate,
713 const struct cache_entry *ce, int pos, int ok_to_replace)
715 int retval = 0;
716 int stage = ce_stage(ce);
717 const char *name = ce->name;
718 const char *slash = name + ce_namelen(ce);
720 for (;;) {
721 int len;
723 for (;;) {
724 if (*--slash == '/')
725 break;
726 if (slash <= ce->name)
727 return retval;
729 len = slash - name;
731 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
732 if (pos >= 0) {
734 * Found one, but not so fast. This could
735 * be a marker that says "I was here, but
736 * I am being removed". Such an entry is
737 * not a part of the resulting tree, and
738 * it is Ok to have a directory at the same
739 * path.
741 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
742 retval = -1;
743 if (!ok_to_replace)
744 break;
745 remove_index_entry_at(istate, pos);
746 continue;
749 else
750 pos = -pos-1;
753 * Trivial optimization: if we find an entry that
754 * already matches the sub-directory, then we know
755 * we're ok, and we can exit.
757 while (pos < istate->cache_nr) {
758 struct cache_entry *p = istate->cache[pos];
759 if ((ce_namelen(p) <= len) ||
760 (p->name[len] != '/') ||
761 memcmp(p->name, name, len))
762 break; /* not our subdirectory */
763 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
765 * p is at the same stage as our entry, and
766 * is a subdirectory of what we are looking
767 * at, so we cannot have conflicts at our
768 * level or anything shorter.
770 return retval;
771 pos++;
774 return retval;
777 /* We may be in a situation where we already have path/file and path
778 * is being added, or we already have path and path/file is being
779 * added. Either one would result in a nonsense tree that has path
780 * twice when git-write-tree tries to write it out. Prevent it.
782 * If ok-to-replace is specified, we remove the conflicting entries
783 * from the cache so the caller should recompute the insert position.
784 * When this happens, we return non-zero.
786 static int check_file_directory_conflict(struct index_state *istate,
787 const struct cache_entry *ce,
788 int pos, int ok_to_replace)
790 int retval;
793 * When ce is an "I am going away" entry, we allow it to be added
795 if (ce->ce_flags & CE_REMOVE)
796 return 0;
799 * We check if the path is a sub-path of a subsequent pathname
800 * first, since removing those will not change the position
801 * in the array.
803 retval = has_file_name(istate, ce, pos, ok_to_replace);
806 * Then check if the path might have a clashing sub-directory
807 * before it.
809 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
812 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
814 int pos;
815 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
816 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
817 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
819 cache_tree_invalidate_path(istate->cache_tree, ce->name);
820 pos = index_name_pos(istate, ce->name, ce->ce_flags);
822 /* existing match? Just replace it. */
823 if (pos >= 0) {
824 replace_index_entry(istate, pos, ce);
825 return 0;
827 pos = -pos-1;
830 * Inserting a merged entry ("stage 0") into the index
831 * will always replace all non-merged entries..
833 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
834 while (ce_same_name(istate->cache[pos], ce)) {
835 ok_to_add = 1;
836 if (!remove_index_entry_at(istate, pos))
837 break;
841 if (!ok_to_add)
842 return -1;
843 if (!verify_path(ce->name))
844 return -1;
846 if (!skip_df_check &&
847 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
848 if (!ok_to_replace)
849 return error("'%s' appears as both a file and as a directory",
850 ce->name);
851 pos = index_name_pos(istate, ce->name, ce->ce_flags);
852 pos = -pos-1;
854 return pos + 1;
857 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
859 int pos;
861 if (option & ADD_CACHE_JUST_APPEND)
862 pos = istate->cache_nr;
863 else {
864 int ret;
865 ret = add_index_entry_with_check(istate, ce, option);
866 if (ret <= 0)
867 return ret;
868 pos = ret - 1;
871 /* Make sure the array is big enough .. */
872 if (istate->cache_nr == istate->cache_alloc) {
873 istate->cache_alloc = alloc_nr(istate->cache_alloc);
874 istate->cache = xrealloc(istate->cache,
875 istate->cache_alloc * sizeof(struct cache_entry *));
878 /* Add it in.. */
879 istate->cache_nr++;
880 if (istate->cache_nr > pos + 1)
881 memmove(istate->cache + pos + 1,
882 istate->cache + pos,
883 (istate->cache_nr - pos - 1) * sizeof(ce));
884 set_index_entry(istate, pos, ce);
885 istate->cache_changed = 1;
886 return 0;
890 * "refresh" does not calculate a new sha1 file or bring the
891 * cache up-to-date for mode/content changes. But what it
892 * _does_ do is to "re-match" the stat information of a file
893 * with the cache, so that you can refresh the cache for a
894 * file that hasn't been changed but where the stat entry is
895 * out of date.
897 * For example, you'd want to do this after doing a "git-read-tree",
898 * to link up the stat cache details with the proper files.
900 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
901 struct cache_entry *ce,
902 unsigned int options, int *err)
904 struct stat st;
905 struct cache_entry *updated;
906 int changed, size;
907 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
909 if (ce_uptodate(ce))
910 return ce;
913 * CE_VALID means the user promised us that the change to
914 * the work tree does not matter and told us not to worry.
916 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
917 ce_mark_uptodate(ce);
918 return ce;
921 if (lstat(ce->name, &st) < 0) {
922 if (err)
923 *err = errno;
924 return NULL;
927 changed = ie_match_stat(istate, ce, &st, options);
928 if (!changed) {
930 * The path is unchanged. If we were told to ignore
931 * valid bit, then we did the actual stat check and
932 * found that the entry is unmodified. If the entry
933 * is not marked VALID, this is the place to mark it
934 * valid again, under "assume unchanged" mode.
936 if (ignore_valid && assume_unchanged &&
937 !(ce->ce_flags & CE_VALID))
938 ; /* mark this one VALID again */
939 else {
941 * We do not mark the index itself "modified"
942 * because CE_UPTODATE flag is in-core only;
943 * we are not going to write this change out.
945 ce_mark_uptodate(ce);
946 return ce;
950 if (ie_modified(istate, ce, &st, options)) {
951 if (err)
952 *err = EINVAL;
953 return NULL;
956 size = ce_size(ce);
957 updated = xmalloc(size);
958 memcpy(updated, ce, size);
959 fill_stat_cache_info(updated, &st);
961 * If ignore_valid is not set, we should leave CE_VALID bit
962 * alone. Otherwise, paths marked with --no-assume-unchanged
963 * (i.e. things to be edited) will reacquire CE_VALID bit
964 * automatically, which is not really what we want.
966 if (!ignore_valid && assume_unchanged &&
967 !(ce->ce_flags & CE_VALID))
968 updated->ce_flags &= ~CE_VALID;
970 return updated;
973 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
975 int i;
976 int has_errors = 0;
977 int really = (flags & REFRESH_REALLY) != 0;
978 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
979 int quiet = (flags & REFRESH_QUIET) != 0;
980 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
981 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
982 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
984 for (i = 0; i < istate->cache_nr; i++) {
985 struct cache_entry *ce, *new;
986 int cache_errno = 0;
988 ce = istate->cache[i];
989 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
990 continue;
992 if (ce_stage(ce)) {
993 while ((i < istate->cache_nr) &&
994 ! strcmp(istate->cache[i]->name, ce->name))
995 i++;
996 i--;
997 if (allow_unmerged)
998 continue;
999 printf("%s: needs merge\n", ce->name);
1000 has_errors = 1;
1001 continue;
1004 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1005 continue;
1007 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1008 if (new == ce)
1009 continue;
1010 if (!new) {
1011 if (not_new && cache_errno == ENOENT)
1012 continue;
1013 if (really && cache_errno == EINVAL) {
1014 /* If we are doing --really-refresh that
1015 * means the index is not valid anymore.
1017 ce->ce_flags &= ~CE_VALID;
1018 istate->cache_changed = 1;
1020 if (quiet)
1021 continue;
1022 printf("%s: needs update\n", ce->name);
1023 has_errors = 1;
1024 continue;
1027 replace_index_entry(istate, i, new);
1029 return has_errors;
1032 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1034 return refresh_cache_ent(&the_index, ce, really, NULL);
1037 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1039 SHA_CTX c;
1040 unsigned char sha1[20];
1042 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1043 return error("bad signature");
1044 if (hdr->hdr_version != htonl(2))
1045 return error("bad index version");
1046 SHA1_Init(&c);
1047 SHA1_Update(&c, hdr, size - 20);
1048 SHA1_Final(sha1, &c);
1049 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1050 return error("bad index file sha1 signature");
1051 return 0;
1054 static int read_index_extension(struct index_state *istate,
1055 const char *ext, void *data, unsigned long sz)
1057 switch (CACHE_EXT(ext)) {
1058 case CACHE_EXT_TREE:
1059 istate->cache_tree = cache_tree_read(data, sz);
1060 break;
1061 default:
1062 if (*ext < 'A' || 'Z' < *ext)
1063 return error("index uses %.4s extension, which we do not understand",
1064 ext);
1065 fprintf(stderr, "ignoring %.4s extension\n", ext);
1066 break;
1068 return 0;
1071 int read_index(struct index_state *istate)
1073 return read_index_from(istate, get_index_file());
1076 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1078 size_t len;
1080 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1081 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1082 ce->ce_dev = ntohl(ondisk->dev);
1083 ce->ce_ino = ntohl(ondisk->ino);
1084 ce->ce_mode = ntohl(ondisk->mode);
1085 ce->ce_uid = ntohl(ondisk->uid);
1086 ce->ce_gid = ntohl(ondisk->gid);
1087 ce->ce_size = ntohl(ondisk->size);
1088 /* On-disk flags are just 16 bits */
1089 ce->ce_flags = ntohs(ondisk->flags);
1090 hashcpy(ce->sha1, ondisk->sha1);
1092 len = ce->ce_flags & CE_NAMEMASK;
1093 if (len == CE_NAMEMASK)
1094 len = strlen(ondisk->name);
1096 * NEEDSWORK: If the original index is crafted, this copy could
1097 * go unchecked.
1099 memcpy(ce->name, ondisk->name, len + 1);
1102 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1104 long per_entry;
1106 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1109 * Alignment can cause differences. This should be "alignof", but
1110 * since that's a gcc'ism, just use the size of a pointer.
1112 per_entry += sizeof(void *);
1113 return ondisk_size + entries*per_entry;
1116 /* remember to discard_cache() before reading a different cache! */
1117 int read_index_from(struct index_state *istate, const char *path)
1119 int fd, i;
1120 struct stat st;
1121 unsigned long src_offset, dst_offset;
1122 struct cache_header *hdr;
1123 void *mmap;
1124 size_t mmap_size;
1126 errno = EBUSY;
1127 if (istate->alloc)
1128 return istate->cache_nr;
1130 errno = ENOENT;
1131 istate->timestamp = 0;
1132 fd = open(path, O_RDONLY);
1133 if (fd < 0) {
1134 if (errno == ENOENT)
1135 return 0;
1136 die("index file open failed (%s)", strerror(errno));
1139 if (fstat(fd, &st))
1140 die("cannot stat the open index (%s)", strerror(errno));
1142 errno = EINVAL;
1143 mmap_size = xsize_t(st.st_size);
1144 if (mmap_size < sizeof(struct cache_header) + 20)
1145 die("index file smaller than expected");
1147 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1148 close(fd);
1149 if (mmap == MAP_FAILED)
1150 die("unable to map index file");
1152 hdr = mmap;
1153 if (verify_hdr(hdr, mmap_size) < 0)
1154 goto unmap;
1156 istate->cache_nr = ntohl(hdr->hdr_entries);
1157 istate->cache_alloc = alloc_nr(istate->cache_nr);
1158 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1161 * The disk format is actually larger than the in-memory format,
1162 * due to space for nsec etc, so even though the in-memory one
1163 * has room for a few more flags, we can allocate using the same
1164 * index size
1166 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1168 src_offset = sizeof(*hdr);
1169 dst_offset = 0;
1170 for (i = 0; i < istate->cache_nr; i++) {
1171 struct ondisk_cache_entry *disk_ce;
1172 struct cache_entry *ce;
1174 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1175 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1176 convert_from_disk(disk_ce, ce);
1177 set_index_entry(istate, i, ce);
1179 src_offset += ondisk_ce_size(ce);
1180 dst_offset += ce_size(ce);
1182 istate->timestamp = st.st_mtime;
1183 while (src_offset <= mmap_size - 20 - 8) {
1184 /* After an array of active_nr index entries,
1185 * there can be arbitrary number of extended
1186 * sections, each of which is prefixed with
1187 * extension name (4-byte) and section length
1188 * in 4-byte network byte order.
1190 unsigned long extsize;
1191 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1192 extsize = ntohl(extsize);
1193 if (read_index_extension(istate,
1194 (const char *) mmap + src_offset,
1195 (char *) mmap + src_offset + 8,
1196 extsize) < 0)
1197 goto unmap;
1198 src_offset += 8;
1199 src_offset += extsize;
1201 munmap(mmap, mmap_size);
1202 return istate->cache_nr;
1204 unmap:
1205 munmap(mmap, mmap_size);
1206 errno = EINVAL;
1207 die("index file corrupt");
1210 int discard_index(struct index_state *istate)
1212 istate->cache_nr = 0;
1213 istate->cache_changed = 0;
1214 istate->timestamp = 0;
1215 free_hash(&istate->name_hash);
1216 cache_tree_free(&(istate->cache_tree));
1217 free(istate->alloc);
1218 istate->alloc = NULL;
1220 /* no need to throw away allocated active_cache */
1221 return 0;
1224 int unmerged_index(const struct index_state *istate)
1226 int i;
1227 for (i = 0; i < istate->cache_nr; i++) {
1228 if (ce_stage(istate->cache[i]))
1229 return 1;
1231 return 0;
1234 #define WRITE_BUFFER_SIZE 8192
1235 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1236 static unsigned long write_buffer_len;
1238 static int ce_write_flush(SHA_CTX *context, int fd)
1240 unsigned int buffered = write_buffer_len;
1241 if (buffered) {
1242 SHA1_Update(context, write_buffer, buffered);
1243 if (write_in_full(fd, write_buffer, buffered) != buffered)
1244 return -1;
1245 write_buffer_len = 0;
1247 return 0;
1250 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1252 while (len) {
1253 unsigned int buffered = write_buffer_len;
1254 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1255 if (partial > len)
1256 partial = len;
1257 memcpy(write_buffer + buffered, data, partial);
1258 buffered += partial;
1259 if (buffered == WRITE_BUFFER_SIZE) {
1260 write_buffer_len = buffered;
1261 if (ce_write_flush(context, fd))
1262 return -1;
1263 buffered = 0;
1265 write_buffer_len = buffered;
1266 len -= partial;
1267 data = (char *) data + partial;
1269 return 0;
1272 static int write_index_ext_header(SHA_CTX *context, int fd,
1273 unsigned int ext, unsigned int sz)
1275 ext = htonl(ext);
1276 sz = htonl(sz);
1277 return ((ce_write(context, fd, &ext, 4) < 0) ||
1278 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1281 static int ce_flush(SHA_CTX *context, int fd)
1283 unsigned int left = write_buffer_len;
1285 if (left) {
1286 write_buffer_len = 0;
1287 SHA1_Update(context, write_buffer, left);
1290 /* Flush first if not enough space for SHA1 signature */
1291 if (left + 20 > WRITE_BUFFER_SIZE) {
1292 if (write_in_full(fd, write_buffer, left) != left)
1293 return -1;
1294 left = 0;
1297 /* Append the SHA1 signature at the end */
1298 SHA1_Final(write_buffer + left, context);
1299 left += 20;
1300 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1303 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1306 * The only thing we care about in this function is to smudge the
1307 * falsely clean entry due to touch-update-touch race, so we leave
1308 * everything else as they are. We are called for entries whose
1309 * ce_mtime match the index file mtime.
1311 struct stat st;
1313 if (lstat(ce->name, &st) < 0)
1314 return;
1315 if (ce_match_stat_basic(ce, &st))
1316 return;
1317 if (ce_modified_check_fs(ce, &st)) {
1318 /* This is "racily clean"; smudge it. Note that this
1319 * is a tricky code. At first glance, it may appear
1320 * that it can break with this sequence:
1322 * $ echo xyzzy >frotz
1323 * $ git-update-index --add frotz
1324 * $ : >frotz
1325 * $ sleep 3
1326 * $ echo filfre >nitfol
1327 * $ git-update-index --add nitfol
1329 * but it does not. When the second update-index runs,
1330 * it notices that the entry "frotz" has the same timestamp
1331 * as index, and if we were to smudge it by resetting its
1332 * size to zero here, then the object name recorded
1333 * in index is the 6-byte file but the cached stat information
1334 * becomes zero --- which would then match what we would
1335 * obtain from the filesystem next time we stat("frotz").
1337 * However, the second update-index, before calling
1338 * this function, notices that the cached size is 6
1339 * bytes and what is on the filesystem is an empty
1340 * file, and never calls us, so the cached size information
1341 * for "frotz" stays 6 which does not match the filesystem.
1343 ce->ce_size = 0;
1347 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1349 int size = ondisk_ce_size(ce);
1350 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1352 ondisk->ctime.sec = htonl(ce->ce_ctime);
1353 ondisk->ctime.nsec = 0;
1354 ondisk->mtime.sec = htonl(ce->ce_mtime);
1355 ondisk->mtime.nsec = 0;
1356 ondisk->dev = htonl(ce->ce_dev);
1357 ondisk->ino = htonl(ce->ce_ino);
1358 ondisk->mode = htonl(ce->ce_mode);
1359 ondisk->uid = htonl(ce->ce_uid);
1360 ondisk->gid = htonl(ce->ce_gid);
1361 ondisk->size = htonl(ce->ce_size);
1362 hashcpy(ondisk->sha1, ce->sha1);
1363 ondisk->flags = htons(ce->ce_flags);
1364 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1366 return ce_write(c, fd, ondisk, size);
1369 int write_index(const struct index_state *istate, int newfd)
1371 SHA_CTX c;
1372 struct cache_header hdr;
1373 int i, err, removed;
1374 struct cache_entry **cache = istate->cache;
1375 int entries = istate->cache_nr;
1377 for (i = removed = 0; i < entries; i++)
1378 if (cache[i]->ce_flags & CE_REMOVE)
1379 removed++;
1381 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1382 hdr.hdr_version = htonl(2);
1383 hdr.hdr_entries = htonl(entries - removed);
1385 SHA1_Init(&c);
1386 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1387 return -1;
1389 for (i = 0; i < entries; i++) {
1390 struct cache_entry *ce = cache[i];
1391 if (ce->ce_flags & CE_REMOVE)
1392 continue;
1393 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1394 ce_smudge_racily_clean_entry(ce);
1395 if (ce_write_entry(&c, newfd, ce) < 0)
1396 return -1;
1399 /* Write extension data here */
1400 if (istate->cache_tree) {
1401 struct strbuf sb;
1403 strbuf_init(&sb, 0);
1404 cache_tree_write(&sb, istate->cache_tree);
1405 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1406 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1407 strbuf_release(&sb);
1408 if (err)
1409 return -1;
1411 return ce_flush(&c, newfd);