racy-git: an empty blob has a fixed object name
[git/dscho.git] / read-cache.c
blob6fa3c21148979791f18bc0b5c3ed76952acad9ff
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 unsigned int hash_name(const char *name, int namelen)
28 unsigned int hash = 0x123;
30 do {
31 unsigned char c = *name++;
32 hash = hash*101 + c;
33 } while (--namelen);
34 return hash;
37 static void hash_index_entry(struct index_state *istate, struct cache_entry *ce)
39 void **pos;
40 unsigned int hash;
42 if (ce->ce_flags & CE_HASHED)
43 return;
44 ce->ce_flags |= CE_HASHED;
45 ce->next = NULL;
46 hash = hash_name(ce->name, ce_namelen(ce));
47 pos = insert_hash(hash, ce, &istate->name_hash);
48 if (pos) {
49 ce->next = *pos;
50 *pos = ce;
54 static void lazy_init_name_hash(struct index_state *istate)
56 int nr;
58 if (istate->name_hash_initialized)
59 return;
60 for (nr = 0; nr < istate->cache_nr; nr++)
61 hash_index_entry(istate, istate->cache[nr]);
62 istate->name_hash_initialized = 1;
65 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
67 ce->ce_flags &= ~CE_UNHASHED;
68 istate->cache[nr] = ce;
69 if (istate->name_hash_initialized)
70 hash_index_entry(istate, ce);
73 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
75 struct cache_entry *old = istate->cache[nr];
77 remove_index_entry(old);
78 set_index_entry(istate, nr, ce);
79 istate->cache_changed = 1;
82 int index_name_exists(struct index_state *istate, const char *name, int namelen)
84 unsigned int hash = hash_name(name, namelen);
85 struct cache_entry *ce;
87 lazy_init_name_hash(istate);
88 ce = lookup_hash(hash, &istate->name_hash);
90 while (ce) {
91 if (!(ce->ce_flags & CE_UNHASHED)) {
92 if (!cache_name_compare(name, namelen, ce->name, ce->ce_flags))
93 return 1;
95 ce = ce->next;
97 return 0;
101 * This only updates the "non-critical" parts of the directory
102 * cache, ie the parts that aren't tracked by GIT, and only used
103 * to validate the cache.
105 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
107 ce->ce_ctime = st->st_ctime;
108 ce->ce_mtime = st->st_mtime;
109 ce->ce_dev = st->st_dev;
110 ce->ce_ino = st->st_ino;
111 ce->ce_uid = st->st_uid;
112 ce->ce_gid = st->st_gid;
113 ce->ce_size = st->st_size;
115 if (assume_unchanged)
116 ce->ce_flags |= CE_VALID;
118 if (S_ISREG(st->st_mode))
119 ce_mark_uptodate(ce);
122 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
124 int match = -1;
125 int fd = open(ce->name, O_RDONLY);
127 if (fd >= 0) {
128 unsigned char sha1[20];
129 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
130 match = hashcmp(sha1, ce->sha1);
131 /* index_fd() closed the file descriptor already */
133 return match;
136 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
138 int match = -1;
139 char *target;
140 void *buffer;
141 unsigned long size;
142 enum object_type type;
143 int len;
145 target = xmalloc(expected_size);
146 len = readlink(ce->name, target, expected_size);
147 if (len != expected_size) {
148 free(target);
149 return -1;
151 buffer = read_sha1_file(ce->sha1, &type, &size);
152 if (!buffer) {
153 free(target);
154 return -1;
156 if (size == expected_size)
157 match = memcmp(buffer, target, size);
158 free(buffer);
159 free(target);
160 return match;
163 static int ce_compare_gitlink(struct cache_entry *ce)
165 unsigned char sha1[20];
168 * We don't actually require that the .git directory
169 * under GITLINK directory be a valid git directory. It
170 * might even be missing (in case nobody populated that
171 * sub-project).
173 * If so, we consider it always to match.
175 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
176 return 0;
177 return hashcmp(sha1, ce->sha1);
180 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
182 switch (st->st_mode & S_IFMT) {
183 case S_IFREG:
184 if (ce_compare_data(ce, st))
185 return DATA_CHANGED;
186 break;
187 case S_IFLNK:
188 if (ce_compare_link(ce, xsize_t(st->st_size)))
189 return DATA_CHANGED;
190 break;
191 case S_IFDIR:
192 if (S_ISGITLINK(ce->ce_mode))
193 return 0;
194 default:
195 return TYPE_CHANGED;
197 return 0;
200 static int is_empty_blob_sha1(const unsigned char *sha1)
202 static const unsigned char empty_blob_sha1[20] = {
203 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
204 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
207 return !hashcmp(sha1, empty_blob_sha1);
210 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
212 unsigned int changed = 0;
214 if (ce->ce_flags & CE_REMOVE)
215 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
217 switch (ce->ce_mode & S_IFMT) {
218 case S_IFREG:
219 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
220 /* We consider only the owner x bit to be relevant for
221 * "mode changes"
223 if (trust_executable_bit &&
224 (0100 & (ce->ce_mode ^ st->st_mode)))
225 changed |= MODE_CHANGED;
226 break;
227 case S_IFLNK:
228 if (!S_ISLNK(st->st_mode) &&
229 (has_symlinks || !S_ISREG(st->st_mode)))
230 changed |= TYPE_CHANGED;
231 break;
232 case S_IFGITLINK:
233 if (!S_ISDIR(st->st_mode))
234 changed |= TYPE_CHANGED;
235 else if (ce_compare_gitlink(ce))
236 changed |= DATA_CHANGED;
237 return changed;
238 default:
239 die("internal error: ce_mode is %o", ce->ce_mode);
241 if (ce->ce_mtime != (unsigned int) st->st_mtime)
242 changed |= MTIME_CHANGED;
243 if (ce->ce_ctime != (unsigned int) st->st_ctime)
244 changed |= CTIME_CHANGED;
246 if (ce->ce_uid != (unsigned int) st->st_uid ||
247 ce->ce_gid != (unsigned int) st->st_gid)
248 changed |= OWNER_CHANGED;
249 if (ce->ce_ino != (unsigned int) st->st_ino)
250 changed |= INODE_CHANGED;
252 #ifdef USE_STDEV
254 * st_dev breaks on network filesystems where different
255 * clients will have different views of what "device"
256 * the filesystem is on
258 if (ce->ce_dev != (unsigned int) st->st_dev)
259 changed |= INODE_CHANGED;
260 #endif
262 if (ce->ce_size != (unsigned int) st->st_size)
263 changed |= DATA_CHANGED;
265 /* Racily smudged entry? */
266 if (!ce->ce_size) {
267 if (!is_empty_blob_sha1(ce->sha1))
268 changed |= DATA_CHANGED;
271 return changed;
274 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
276 return (istate->timestamp &&
277 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
280 int ie_match_stat(const struct index_state *istate,
281 struct cache_entry *ce, struct stat *st,
282 unsigned int options)
284 unsigned int changed;
285 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
286 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
289 * If it's marked as always valid in the index, it's
290 * valid whatever the checked-out copy says.
292 if (!ignore_valid && (ce->ce_flags & CE_VALID))
293 return 0;
295 changed = ce_match_stat_basic(ce, st);
298 * Within 1 second of this sequence:
299 * echo xyzzy >file && git-update-index --add file
300 * running this command:
301 * echo frotz >file
302 * would give a falsely clean cache entry. The mtime and
303 * length match the cache, and other stat fields do not change.
305 * We could detect this at update-index time (the cache entry
306 * being registered/updated records the same time as "now")
307 * and delay the return from git-update-index, but that would
308 * effectively mean we can make at most one commit per second,
309 * which is not acceptable. Instead, we check cache entries
310 * whose mtime are the same as the index file timestamp more
311 * carefully than others.
313 if (!changed && is_racy_timestamp(istate, ce)) {
314 if (assume_racy_is_modified)
315 changed |= DATA_CHANGED;
316 else
317 changed |= ce_modified_check_fs(ce, st);
320 return changed;
323 int ie_modified(const struct index_state *istate,
324 struct cache_entry *ce, struct stat *st, unsigned int options)
326 int changed, changed_fs;
328 changed = ie_match_stat(istate, ce, st, options);
329 if (!changed)
330 return 0;
332 * If the mode or type has changed, there's no point in trying
333 * to refresh the entry - it's not going to match
335 if (changed & (MODE_CHANGED | TYPE_CHANGED))
336 return changed;
338 /* Immediately after read-tree or update-index --cacheinfo,
339 * the length field is zero. For other cases the ce_size
340 * should match the SHA1 recorded in the index entry.
342 if ((changed & DATA_CHANGED) && ce->ce_size != 0)
343 return changed;
345 changed_fs = ce_modified_check_fs(ce, st);
346 if (changed_fs)
347 return changed | changed_fs;
348 return 0;
351 int base_name_compare(const char *name1, int len1, int mode1,
352 const char *name2, int len2, int mode2)
354 unsigned char c1, c2;
355 int len = len1 < len2 ? len1 : len2;
356 int cmp;
358 cmp = memcmp(name1, name2, len);
359 if (cmp)
360 return cmp;
361 c1 = name1[len];
362 c2 = name2[len];
363 if (!c1 && S_ISDIR(mode1))
364 c1 = '/';
365 if (!c2 && S_ISDIR(mode2))
366 c2 = '/';
367 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
371 * df_name_compare() is identical to base_name_compare(), except it
372 * compares conflicting directory/file entries as equal. Note that
373 * while a directory name compares as equal to a regular file, they
374 * then individually compare _differently_ to a filename that has
375 * a dot after the basename (because '\0' < '.' < '/').
377 * This is used by routines that want to traverse the git namespace
378 * but then handle conflicting entries together when possible.
380 int df_name_compare(const char *name1, int len1, int mode1,
381 const char *name2, int len2, int mode2)
383 int len = len1 < len2 ? len1 : len2, cmp;
384 unsigned char c1, c2;
386 cmp = memcmp(name1, name2, len);
387 if (cmp)
388 return cmp;
389 /* Directories and files compare equal (same length, same name) */
390 if (len1 == len2)
391 return 0;
392 c1 = name1[len];
393 if (!c1 && S_ISDIR(mode1))
394 c1 = '/';
395 c2 = name2[len];
396 if (!c2 && S_ISDIR(mode2))
397 c2 = '/';
398 if (c1 == '/' && !c2)
399 return 0;
400 if (c2 == '/' && !c1)
401 return 0;
402 return c1 - c2;
405 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
407 int len1 = flags1 & CE_NAMEMASK;
408 int len2 = flags2 & CE_NAMEMASK;
409 int len = len1 < len2 ? len1 : len2;
410 int cmp;
412 cmp = memcmp(name1, name2, len);
413 if (cmp)
414 return cmp;
415 if (len1 < len2)
416 return -1;
417 if (len1 > len2)
418 return 1;
420 /* Compare stages */
421 flags1 &= CE_STAGEMASK;
422 flags2 &= CE_STAGEMASK;
424 if (flags1 < flags2)
425 return -1;
426 if (flags1 > flags2)
427 return 1;
428 return 0;
431 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
433 int first, last;
435 first = 0;
436 last = istate->cache_nr;
437 while (last > first) {
438 int next = (last + first) >> 1;
439 struct cache_entry *ce = istate->cache[next];
440 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
441 if (!cmp)
442 return next;
443 if (cmp < 0) {
444 last = next;
445 continue;
447 first = next+1;
449 return -first-1;
452 /* Remove entry, return true if there are more entries to go.. */
453 int remove_index_entry_at(struct index_state *istate, int pos)
455 struct cache_entry *ce = istate->cache[pos];
457 remove_index_entry(ce);
458 istate->cache_changed = 1;
459 istate->cache_nr--;
460 if (pos >= istate->cache_nr)
461 return 0;
462 memmove(istate->cache + pos,
463 istate->cache + pos + 1,
464 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
465 return 1;
468 int remove_file_from_index(struct index_state *istate, const char *path)
470 int pos = index_name_pos(istate, path, strlen(path));
471 if (pos < 0)
472 pos = -pos-1;
473 cache_tree_invalidate_path(istate->cache_tree, path);
474 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
475 remove_index_entry_at(istate, pos);
476 return 0;
479 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
481 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
484 static int index_name_pos_also_unmerged(struct index_state *istate,
485 const char *path, int namelen)
487 int pos = index_name_pos(istate, path, namelen);
488 struct cache_entry *ce;
490 if (pos >= 0)
491 return pos;
493 /* maybe unmerged? */
494 pos = -1 - pos;
495 if (pos >= istate->cache_nr ||
496 compare_name((ce = istate->cache[pos]), path, namelen))
497 return -1;
499 /* order of preference: stage 2, 1, 3 */
500 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
501 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
502 !compare_name(ce, path, namelen))
503 pos++;
504 return pos;
507 int add_file_to_index(struct index_state *istate, const char *path, int verbose)
509 int size, namelen, pos;
510 struct stat st;
511 struct cache_entry *ce;
512 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
514 if (lstat(path, &st))
515 die("%s: unable to stat (%s)", path, strerror(errno));
517 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
518 die("%s: can only add regular files, symbolic links or git-directories", path);
520 namelen = strlen(path);
521 if (S_ISDIR(st.st_mode)) {
522 while (namelen && path[namelen-1] == '/')
523 namelen--;
525 size = cache_entry_size(namelen);
526 ce = xcalloc(1, size);
527 memcpy(ce->name, path, namelen);
528 ce->ce_flags = namelen;
529 fill_stat_cache_info(ce, &st);
531 if (trust_executable_bit && has_symlinks)
532 ce->ce_mode = create_ce_mode(st.st_mode);
533 else {
534 /* If there is an existing entry, pick the mode bits and type
535 * from it, otherwise assume unexecutable regular file.
537 struct cache_entry *ent;
538 int pos = index_name_pos_also_unmerged(istate, path, namelen);
540 ent = (0 <= pos) ? istate->cache[pos] : NULL;
541 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
544 pos = index_name_pos(istate, ce->name, namelen);
545 if (0 <= pos &&
546 !ce_stage(istate->cache[pos]) &&
547 !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) {
548 /* Nothing changed, really */
549 free(ce);
550 ce_mark_uptodate(istate->cache[pos]);
551 return 0;
554 if (index_path(ce->sha1, path, &st, 1))
555 die("unable to index file %s", path);
556 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
557 die("unable to add %s to index",path);
558 if (verbose)
559 printf("add '%s'\n", path);
560 return 0;
563 struct cache_entry *make_cache_entry(unsigned int mode,
564 const unsigned char *sha1, const char *path, int stage,
565 int refresh)
567 int size, len;
568 struct cache_entry *ce;
570 if (!verify_path(path))
571 return NULL;
573 len = strlen(path);
574 size = cache_entry_size(len);
575 ce = xcalloc(1, size);
577 hashcpy(ce->sha1, sha1);
578 memcpy(ce->name, path, len);
579 ce->ce_flags = create_ce_flags(len, stage);
580 ce->ce_mode = create_ce_mode(mode);
582 if (refresh)
583 return refresh_cache_entry(ce, 0);
585 return ce;
588 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
590 int len = ce_namelen(a);
591 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
594 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
596 const char *match, *name;
597 int len;
599 if (!pathspec)
600 return 1;
602 len = ce_namelen(ce);
603 name = ce->name;
604 while ((match = *pathspec++) != NULL) {
605 int matchlen = strlen(match);
606 if (matchlen > len)
607 continue;
608 if (memcmp(name, match, matchlen))
609 continue;
610 if (matchlen && name[matchlen-1] == '/')
611 return 1;
612 if (name[matchlen] == '/' || !name[matchlen])
613 return 1;
614 if (!matchlen)
615 return 1;
617 return 0;
621 * We fundamentally don't like some paths: we don't want
622 * dot or dot-dot anywhere, and for obvious reasons don't
623 * want to recurse into ".git" either.
625 * Also, we don't want double slashes or slashes at the
626 * end that can make pathnames ambiguous.
628 static int verify_dotfile(const char *rest)
631 * The first character was '.', but that
632 * has already been discarded, we now test
633 * the rest.
635 switch (*rest) {
636 /* "." is not allowed */
637 case '\0': case '/':
638 return 0;
641 * ".git" followed by NUL or slash is bad. This
642 * shares the path end test with the ".." case.
644 case 'g':
645 if (rest[1] != 'i')
646 break;
647 if (rest[2] != 't')
648 break;
649 rest += 2;
650 /* fallthrough */
651 case '.':
652 if (rest[1] == '\0' || rest[1] == '/')
653 return 0;
655 return 1;
658 int verify_path(const char *path)
660 char c;
662 goto inside;
663 for (;;) {
664 if (!c)
665 return 1;
666 if (c == '/') {
667 inside:
668 c = *path++;
669 switch (c) {
670 default:
671 continue;
672 case '/': case '\0':
673 break;
674 case '.':
675 if (verify_dotfile(path))
676 continue;
678 return 0;
680 c = *path++;
685 * Do we have another file that has the beginning components being a
686 * proper superset of the name we're trying to add?
688 static int has_file_name(struct index_state *istate,
689 const struct cache_entry *ce, int pos, int ok_to_replace)
691 int retval = 0;
692 int len = ce_namelen(ce);
693 int stage = ce_stage(ce);
694 const char *name = ce->name;
696 while (pos < istate->cache_nr) {
697 struct cache_entry *p = istate->cache[pos++];
699 if (len >= ce_namelen(p))
700 break;
701 if (memcmp(name, p->name, len))
702 break;
703 if (ce_stage(p) != stage)
704 continue;
705 if (p->name[len] != '/')
706 continue;
707 if (p->ce_flags & CE_REMOVE)
708 continue;
709 retval = -1;
710 if (!ok_to_replace)
711 break;
712 remove_index_entry_at(istate, --pos);
714 return retval;
718 * Do we have another file with a pathname that is a proper
719 * subset of the name we're trying to add?
721 static int has_dir_name(struct index_state *istate,
722 const struct cache_entry *ce, int pos, int ok_to_replace)
724 int retval = 0;
725 int stage = ce_stage(ce);
726 const char *name = ce->name;
727 const char *slash = name + ce_namelen(ce);
729 for (;;) {
730 int len;
732 for (;;) {
733 if (*--slash == '/')
734 break;
735 if (slash <= ce->name)
736 return retval;
738 len = slash - name;
740 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
741 if (pos >= 0) {
743 * Found one, but not so fast. This could
744 * be a marker that says "I was here, but
745 * I am being removed". Such an entry is
746 * not a part of the resulting tree, and
747 * it is Ok to have a directory at the same
748 * path.
750 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
751 retval = -1;
752 if (!ok_to_replace)
753 break;
754 remove_index_entry_at(istate, pos);
755 continue;
758 else
759 pos = -pos-1;
762 * Trivial optimization: if we find an entry that
763 * already matches the sub-directory, then we know
764 * we're ok, and we can exit.
766 while (pos < istate->cache_nr) {
767 struct cache_entry *p = istate->cache[pos];
768 if ((ce_namelen(p) <= len) ||
769 (p->name[len] != '/') ||
770 memcmp(p->name, name, len))
771 break; /* not our subdirectory */
772 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
774 * p is at the same stage as our entry, and
775 * is a subdirectory of what we are looking
776 * at, so we cannot have conflicts at our
777 * level or anything shorter.
779 return retval;
780 pos++;
783 return retval;
786 /* We may be in a situation where we already have path/file and path
787 * is being added, or we already have path and path/file is being
788 * added. Either one would result in a nonsense tree that has path
789 * twice when git-write-tree tries to write it out. Prevent it.
791 * If ok-to-replace is specified, we remove the conflicting entries
792 * from the cache so the caller should recompute the insert position.
793 * When this happens, we return non-zero.
795 static int check_file_directory_conflict(struct index_state *istate,
796 const struct cache_entry *ce,
797 int pos, int ok_to_replace)
799 int retval;
802 * When ce is an "I am going away" entry, we allow it to be added
804 if (ce->ce_flags & CE_REMOVE)
805 return 0;
808 * We check if the path is a sub-path of a subsequent pathname
809 * first, since removing those will not change the position
810 * in the array.
812 retval = has_file_name(istate, ce, pos, ok_to_replace);
815 * Then check if the path might have a clashing sub-directory
816 * before it.
818 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
821 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
823 int pos;
824 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
825 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
826 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
828 cache_tree_invalidate_path(istate->cache_tree, ce->name);
829 pos = index_name_pos(istate, ce->name, ce->ce_flags);
831 /* existing match? Just replace it. */
832 if (pos >= 0) {
833 replace_index_entry(istate, pos, ce);
834 return 0;
836 pos = -pos-1;
839 * Inserting a merged entry ("stage 0") into the index
840 * will always replace all non-merged entries..
842 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
843 while (ce_same_name(istate->cache[pos], ce)) {
844 ok_to_add = 1;
845 if (!remove_index_entry_at(istate, pos))
846 break;
850 if (!ok_to_add)
851 return -1;
852 if (!verify_path(ce->name))
853 return -1;
855 if (!skip_df_check &&
856 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
857 if (!ok_to_replace)
858 return error("'%s' appears as both a file and as a directory",
859 ce->name);
860 pos = index_name_pos(istate, ce->name, ce->ce_flags);
861 pos = -pos-1;
863 return pos + 1;
866 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
868 int pos;
870 if (option & ADD_CACHE_JUST_APPEND)
871 pos = istate->cache_nr;
872 else {
873 int ret;
874 ret = add_index_entry_with_check(istate, ce, option);
875 if (ret <= 0)
876 return ret;
877 pos = ret - 1;
880 /* Make sure the array is big enough .. */
881 if (istate->cache_nr == istate->cache_alloc) {
882 istate->cache_alloc = alloc_nr(istate->cache_alloc);
883 istate->cache = xrealloc(istate->cache,
884 istate->cache_alloc * sizeof(struct cache_entry *));
887 /* Add it in.. */
888 istate->cache_nr++;
889 if (istate->cache_nr > pos + 1)
890 memmove(istate->cache + pos + 1,
891 istate->cache + pos,
892 (istate->cache_nr - pos - 1) * sizeof(ce));
893 set_index_entry(istate, pos, ce);
894 istate->cache_changed = 1;
895 return 0;
899 * "refresh" does not calculate a new sha1 file or bring the
900 * cache up-to-date for mode/content changes. But what it
901 * _does_ do is to "re-match" the stat information of a file
902 * with the cache, so that you can refresh the cache for a
903 * file that hasn't been changed but where the stat entry is
904 * out of date.
906 * For example, you'd want to do this after doing a "git-read-tree",
907 * to link up the stat cache details with the proper files.
909 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
910 struct cache_entry *ce,
911 unsigned int options, int *err)
913 struct stat st;
914 struct cache_entry *updated;
915 int changed, size;
916 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
918 if (ce_uptodate(ce))
919 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 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
983 for (i = 0; i < istate->cache_nr; i++) {
984 struct cache_entry *ce, *new;
985 int cache_errno = 0;
987 ce = istate->cache[i];
988 if (ce_stage(ce)) {
989 while ((i < istate->cache_nr) &&
990 ! strcmp(istate->cache[i]->name, ce->name))
991 i++;
992 i--;
993 if (allow_unmerged)
994 continue;
995 printf("%s: needs merge\n", ce->name);
996 has_errors = 1;
997 continue;
1000 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1001 continue;
1003 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1004 if (new == ce)
1005 continue;
1006 if (!new) {
1007 if (not_new && cache_errno == ENOENT)
1008 continue;
1009 if (really && cache_errno == EINVAL) {
1010 /* If we are doing --really-refresh that
1011 * means the index is not valid anymore.
1013 ce->ce_flags &= ~CE_VALID;
1014 istate->cache_changed = 1;
1016 if (quiet)
1017 continue;
1018 printf("%s: needs update\n", ce->name);
1019 has_errors = 1;
1020 continue;
1023 replace_index_entry(istate, i, new);
1025 return has_errors;
1028 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1030 return refresh_cache_ent(&the_index, ce, really, NULL);
1033 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1035 SHA_CTX c;
1036 unsigned char sha1[20];
1038 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1039 return error("bad signature");
1040 if (hdr->hdr_version != htonl(2))
1041 return error("bad index version");
1042 SHA1_Init(&c);
1043 SHA1_Update(&c, hdr, size - 20);
1044 SHA1_Final(sha1, &c);
1045 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1046 return error("bad index file sha1 signature");
1047 return 0;
1050 static int read_index_extension(struct index_state *istate,
1051 const char *ext, void *data, unsigned long sz)
1053 switch (CACHE_EXT(ext)) {
1054 case CACHE_EXT_TREE:
1055 istate->cache_tree = cache_tree_read(data, sz);
1056 break;
1057 default:
1058 if (*ext < 'A' || 'Z' < *ext)
1059 return error("index uses %.4s extension, which we do not understand",
1060 ext);
1061 fprintf(stderr, "ignoring %.4s extension\n", ext);
1062 break;
1064 return 0;
1067 int read_index(struct index_state *istate)
1069 return read_index_from(istate, get_index_file());
1072 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1074 size_t len;
1076 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1077 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1078 ce->ce_dev = ntohl(ondisk->dev);
1079 ce->ce_ino = ntohl(ondisk->ino);
1080 ce->ce_mode = ntohl(ondisk->mode);
1081 ce->ce_uid = ntohl(ondisk->uid);
1082 ce->ce_gid = ntohl(ondisk->gid);
1083 ce->ce_size = ntohl(ondisk->size);
1084 /* On-disk flags are just 16 bits */
1085 ce->ce_flags = ntohs(ondisk->flags);
1086 hashcpy(ce->sha1, ondisk->sha1);
1088 len = ce->ce_flags & CE_NAMEMASK;
1089 if (len == CE_NAMEMASK)
1090 len = strlen(ondisk->name);
1092 * NEEDSWORK: If the original index is crafted, this copy could
1093 * go unchecked.
1095 memcpy(ce->name, ondisk->name, len + 1);
1098 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1100 long per_entry;
1102 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1105 * Alignment can cause differences. This should be "alignof", but
1106 * since that's a gcc'ism, just use the size of a pointer.
1108 per_entry += sizeof(void *);
1109 return ondisk_size + entries*per_entry;
1112 /* remember to discard_cache() before reading a different cache! */
1113 int read_index_from(struct index_state *istate, const char *path)
1115 int fd, i;
1116 struct stat st;
1117 unsigned long src_offset, dst_offset;
1118 struct cache_header *hdr;
1119 void *mmap;
1120 size_t mmap_size;
1122 errno = EBUSY;
1123 if (istate->alloc)
1124 return istate->cache_nr;
1126 errno = ENOENT;
1127 istate->timestamp = 0;
1128 fd = open(path, O_RDONLY);
1129 if (fd < 0) {
1130 if (errno == ENOENT)
1131 return 0;
1132 die("index file open failed (%s)", strerror(errno));
1135 if (fstat(fd, &st))
1136 die("cannot stat the open index (%s)", strerror(errno));
1138 errno = EINVAL;
1139 mmap_size = xsize_t(st.st_size);
1140 if (mmap_size < sizeof(struct cache_header) + 20)
1141 die("index file smaller than expected");
1143 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1144 close(fd);
1145 if (mmap == MAP_FAILED)
1146 die("unable to map index file");
1148 hdr = mmap;
1149 if (verify_hdr(hdr, mmap_size) < 0)
1150 goto unmap;
1152 istate->cache_nr = ntohl(hdr->hdr_entries);
1153 istate->cache_alloc = alloc_nr(istate->cache_nr);
1154 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1157 * The disk format is actually larger than the in-memory format,
1158 * due to space for nsec etc, so even though the in-memory one
1159 * has room for a few more flags, we can allocate using the same
1160 * index size
1162 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1164 src_offset = sizeof(*hdr);
1165 dst_offset = 0;
1166 for (i = 0; i < istate->cache_nr; i++) {
1167 struct ondisk_cache_entry *disk_ce;
1168 struct cache_entry *ce;
1170 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1171 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1172 convert_from_disk(disk_ce, ce);
1173 set_index_entry(istate, i, ce);
1175 src_offset += ondisk_ce_size(ce);
1176 dst_offset += ce_size(ce);
1178 istate->timestamp = st.st_mtime;
1179 while (src_offset <= mmap_size - 20 - 8) {
1180 /* After an array of active_nr index entries,
1181 * there can be arbitrary number of extended
1182 * sections, each of which is prefixed with
1183 * extension name (4-byte) and section length
1184 * in 4-byte network byte order.
1186 unsigned long extsize;
1187 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1188 extsize = ntohl(extsize);
1189 if (read_index_extension(istate,
1190 (const char *) mmap + src_offset,
1191 (char *) mmap + src_offset + 8,
1192 extsize) < 0)
1193 goto unmap;
1194 src_offset += 8;
1195 src_offset += extsize;
1197 munmap(mmap, mmap_size);
1198 return istate->cache_nr;
1200 unmap:
1201 munmap(mmap, mmap_size);
1202 errno = EINVAL;
1203 die("index file corrupt");
1206 int discard_index(struct index_state *istate)
1208 istate->cache_nr = 0;
1209 istate->cache_changed = 0;
1210 istate->timestamp = 0;
1211 free_hash(&istate->name_hash);
1212 cache_tree_free(&(istate->cache_tree));
1213 free(istate->alloc);
1214 istate->alloc = NULL;
1216 /* no need to throw away allocated active_cache */
1217 return 0;
1220 int unmerged_index(const struct index_state *istate)
1222 int i;
1223 for (i = 0; i < istate->cache_nr; i++) {
1224 if (ce_stage(istate->cache[i]))
1225 return 1;
1227 return 0;
1230 #define WRITE_BUFFER_SIZE 8192
1231 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1232 static unsigned long write_buffer_len;
1234 static int ce_write_flush(SHA_CTX *context, int fd)
1236 unsigned int buffered = write_buffer_len;
1237 if (buffered) {
1238 SHA1_Update(context, write_buffer, buffered);
1239 if (write_in_full(fd, write_buffer, buffered) != buffered)
1240 return -1;
1241 write_buffer_len = 0;
1243 return 0;
1246 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1248 while (len) {
1249 unsigned int buffered = write_buffer_len;
1250 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1251 if (partial > len)
1252 partial = len;
1253 memcpy(write_buffer + buffered, data, partial);
1254 buffered += partial;
1255 if (buffered == WRITE_BUFFER_SIZE) {
1256 write_buffer_len = buffered;
1257 if (ce_write_flush(context, fd))
1258 return -1;
1259 buffered = 0;
1261 write_buffer_len = buffered;
1262 len -= partial;
1263 data = (char *) data + partial;
1265 return 0;
1268 static int write_index_ext_header(SHA_CTX *context, int fd,
1269 unsigned int ext, unsigned int sz)
1271 ext = htonl(ext);
1272 sz = htonl(sz);
1273 return ((ce_write(context, fd, &ext, 4) < 0) ||
1274 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1277 static int ce_flush(SHA_CTX *context, int fd)
1279 unsigned int left = write_buffer_len;
1281 if (left) {
1282 write_buffer_len = 0;
1283 SHA1_Update(context, write_buffer, left);
1286 /* Flush first if not enough space for SHA1 signature */
1287 if (left + 20 > WRITE_BUFFER_SIZE) {
1288 if (write_in_full(fd, write_buffer, left) != left)
1289 return -1;
1290 left = 0;
1293 /* Append the SHA1 signature at the end */
1294 SHA1_Final(write_buffer + left, context);
1295 left += 20;
1296 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1299 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1302 * The only thing we care about in this function is to smudge the
1303 * falsely clean entry due to touch-update-touch race, so we leave
1304 * everything else as they are. We are called for entries whose
1305 * ce_mtime match the index file mtime.
1307 struct stat st;
1309 if (lstat(ce->name, &st) < 0)
1310 return;
1311 if (ce_match_stat_basic(ce, &st))
1312 return;
1313 if (ce_modified_check_fs(ce, &st)) {
1314 /* This is "racily clean"; smudge it. Note that this
1315 * is a tricky code. At first glance, it may appear
1316 * that it can break with this sequence:
1318 * $ echo xyzzy >frotz
1319 * $ git-update-index --add frotz
1320 * $ : >frotz
1321 * $ sleep 3
1322 * $ echo filfre >nitfol
1323 * $ git-update-index --add nitfol
1325 * but it does not. When the second update-index runs,
1326 * it notices that the entry "frotz" has the same timestamp
1327 * as index, and if we were to smudge it by resetting its
1328 * size to zero here, then the object name recorded
1329 * in index is the 6-byte file but the cached stat information
1330 * becomes zero --- which would then match what we would
1331 * obtain from the filesystem next time we stat("frotz").
1333 * However, the second update-index, before calling
1334 * this function, notices that the cached size is 6
1335 * bytes and what is on the filesystem is an empty
1336 * file, and never calls us, so the cached size information
1337 * for "frotz" stays 6 which does not match the filesystem.
1339 ce->ce_size = 0;
1343 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1345 int size = ondisk_ce_size(ce);
1346 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1348 ondisk->ctime.sec = htonl(ce->ce_ctime);
1349 ondisk->ctime.nsec = 0;
1350 ondisk->mtime.sec = htonl(ce->ce_mtime);
1351 ondisk->mtime.nsec = 0;
1352 ondisk->dev = htonl(ce->ce_dev);
1353 ondisk->ino = htonl(ce->ce_ino);
1354 ondisk->mode = htonl(ce->ce_mode);
1355 ondisk->uid = htonl(ce->ce_uid);
1356 ondisk->gid = htonl(ce->ce_gid);
1357 ondisk->size = htonl(ce->ce_size);
1358 hashcpy(ondisk->sha1, ce->sha1);
1359 ondisk->flags = htons(ce->ce_flags);
1360 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1362 return ce_write(c, fd, ondisk, size);
1365 int write_index(const struct index_state *istate, int newfd)
1367 SHA_CTX c;
1368 struct cache_header hdr;
1369 int i, err, removed;
1370 struct cache_entry **cache = istate->cache;
1371 int entries = istate->cache_nr;
1373 for (i = removed = 0; i < entries; i++)
1374 if (cache[i]->ce_flags & CE_REMOVE)
1375 removed++;
1377 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1378 hdr.hdr_version = htonl(2);
1379 hdr.hdr_entries = htonl(entries - removed);
1381 SHA1_Init(&c);
1382 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1383 return -1;
1385 for (i = 0; i < entries; i++) {
1386 struct cache_entry *ce = cache[i];
1387 if (ce->ce_flags & CE_REMOVE)
1388 continue;
1389 if (is_racy_timestamp(istate, ce))
1390 ce_smudge_racily_clean_entry(ce);
1391 if (ce_write_entry(&c, newfd, ce) < 0)
1392 return -1;
1395 /* Write extension data here */
1396 if (istate->cache_tree) {
1397 struct strbuf sb;
1399 strbuf_init(&sb, 0);
1400 cache_tree_write(&sb, istate->cache_tree);
1401 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1402 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1403 strbuf_release(&sb);
1404 if (err)
1405 return -1;
1407 return ce_flush(&c, newfd);