2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
8 #include "cache-tree.h"
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;
31 unsigned char c = *name++;
37 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
40 unsigned int hash = hash_name(ce->name, ce_namelen(ce));
42 istate->cache[nr] = ce;
43 pos = insert_hash(hash, ce, &istate->name_hash);
51 * We don't actually *remove* it, we can just mark it invalid so that
52 * we won't find it in lookups.
54 * Not only would we have to search the lists (simple enough), but
55 * we'd also have to rehash other hash buckets in case this makes the
56 * hash bucket empty (common). So it's much better to just mark
59 static void remove_hash_entry(struct index_state *istate, struct cache_entry *ce)
61 ce->ce_flags |= CE_UNHASHED;
64 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
66 struct cache_entry *old = istate->cache[nr];
69 remove_hash_entry(istate, old);
70 set_index_entry(istate, nr, ce);
72 istate->cache_changed = 1;
75 int index_name_exists(struct index_state *istate, const char *name, int namelen)
77 unsigned int hash = hash_name(name, namelen);
78 struct cache_entry *ce = lookup_hash(hash, &istate->name_hash);
81 if (!(ce->ce_flags & CE_UNHASHED)) {
82 if (!cache_name_compare(name, namelen, ce->name, ce->ce_flags))
91 * This only updates the "non-critical" parts of the directory
92 * cache, ie the parts that aren't tracked by GIT, and only used
93 * to validate the cache.
95 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
97 ce->ce_ctime = st->st_ctime;
98 ce->ce_mtime = st->st_mtime;
99 ce->ce_dev = st->st_dev;
100 ce->ce_ino = st->st_ino;
101 ce->ce_uid = st->st_uid;
102 ce->ce_gid = st->st_gid;
103 ce->ce_size = st->st_size;
105 if (assume_unchanged)
106 ce->ce_flags |= CE_VALID;
108 if (S_ISREG(st->st_mode))
109 ce_mark_uptodate(ce);
112 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
115 int fd = open(ce->name, O_RDONLY);
118 unsigned char sha1[20];
119 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
120 match = hashcmp(sha1, ce->sha1);
121 /* index_fd() closed the file descriptor already */
126 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
132 enum object_type type;
135 target = xmalloc(expected_size);
136 len = readlink(ce->name, target, expected_size);
137 if (len != expected_size) {
141 buffer = read_sha1_file(ce->sha1, &type, &size);
146 if (size == expected_size)
147 match = memcmp(buffer, target, size);
153 static int ce_compare_gitlink(struct cache_entry *ce)
155 unsigned char sha1[20];
158 * We don't actually require that the .git directory
159 * under GITLINK directory be a valid git directory. It
160 * might even be missing (in case nobody populated that
163 * If so, we consider it always to match.
165 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
167 return hashcmp(sha1, ce->sha1);
170 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
172 switch (st->st_mode & S_IFMT) {
174 if (ce_compare_data(ce, st))
178 if (ce_compare_link(ce, xsize_t(st->st_size)))
182 if (S_ISGITLINK(ce->ce_mode))
190 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
192 unsigned int changed = 0;
194 if (ce->ce_flags & CE_REMOVE)
195 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
197 switch (ce->ce_mode & S_IFMT) {
199 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
200 /* We consider only the owner x bit to be relevant for
203 if (trust_executable_bit &&
204 (0100 & (ce->ce_mode ^ st->st_mode)))
205 changed |= MODE_CHANGED;
208 if (!S_ISLNK(st->st_mode) &&
209 (has_symlinks || !S_ISREG(st->st_mode)))
210 changed |= TYPE_CHANGED;
213 if (!S_ISDIR(st->st_mode))
214 changed |= TYPE_CHANGED;
215 else if (ce_compare_gitlink(ce))
216 changed |= DATA_CHANGED;
219 die("internal error: ce_mode is %o", ce->ce_mode);
221 if (ce->ce_mtime != (unsigned int) st->st_mtime)
222 changed |= MTIME_CHANGED;
223 if (ce->ce_ctime != (unsigned int) st->st_ctime)
224 changed |= CTIME_CHANGED;
226 if (ce->ce_uid != (unsigned int) st->st_uid ||
227 ce->ce_gid != (unsigned int) st->st_gid)
228 changed |= OWNER_CHANGED;
229 if (ce->ce_ino != (unsigned int) st->st_ino)
230 changed |= INODE_CHANGED;
234 * st_dev breaks on network filesystems where different
235 * clients will have different views of what "device"
236 * the filesystem is on
238 if (ce->ce_dev != (unsigned int) st->st_dev)
239 changed |= INODE_CHANGED;
242 if (ce->ce_size != (unsigned int) st->st_size)
243 changed |= DATA_CHANGED;
248 static int is_racy_timestamp(struct index_state *istate, struct cache_entry *ce)
250 return (istate->timestamp &&
251 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
254 int ie_match_stat(struct index_state *istate,
255 struct cache_entry *ce, struct stat *st,
256 unsigned int options)
258 unsigned int changed;
259 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
260 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
263 * If it's marked as always valid in the index, it's
264 * valid whatever the checked-out copy says.
266 if (!ignore_valid && (ce->ce_flags & CE_VALID))
269 changed = ce_match_stat_basic(ce, st);
272 * Within 1 second of this sequence:
273 * echo xyzzy >file && git-update-index --add file
274 * running this command:
276 * would give a falsely clean cache entry. The mtime and
277 * length match the cache, and other stat fields do not change.
279 * We could detect this at update-index time (the cache entry
280 * being registered/updated records the same time as "now")
281 * and delay the return from git-update-index, but that would
282 * effectively mean we can make at most one commit per second,
283 * which is not acceptable. Instead, we check cache entries
284 * whose mtime are the same as the index file timestamp more
285 * carefully than others.
287 if (!changed && is_racy_timestamp(istate, ce)) {
288 if (assume_racy_is_modified)
289 changed |= DATA_CHANGED;
291 changed |= ce_modified_check_fs(ce, st);
297 int ie_modified(struct index_state *istate,
298 struct cache_entry *ce, struct stat *st, unsigned int options)
300 int changed, changed_fs;
302 changed = ie_match_stat(istate, ce, st, options);
306 * If the mode or type has changed, there's no point in trying
307 * to refresh the entry - it's not going to match
309 if (changed & (MODE_CHANGED | TYPE_CHANGED))
312 /* Immediately after read-tree or update-index --cacheinfo,
313 * the length field is zero. For other cases the ce_size
314 * should match the SHA1 recorded in the index entry.
316 if ((changed & DATA_CHANGED) && ce->ce_size != 0)
319 changed_fs = ce_modified_check_fs(ce, st);
321 return changed | changed_fs;
325 int base_name_compare(const char *name1, int len1, int mode1,
326 const char *name2, int len2, int mode2)
328 unsigned char c1, c2;
329 int len = len1 < len2 ? len1 : len2;
332 cmp = memcmp(name1, name2, len);
337 if (!c1 && S_ISDIR(mode1))
339 if (!c2 && S_ISDIR(mode2))
341 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
344 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
346 int len1 = flags1 & CE_NAMEMASK;
347 int len2 = flags2 & CE_NAMEMASK;
348 int len = len1 < len2 ? len1 : len2;
351 cmp = memcmp(name1, name2, len);
360 flags1 &= CE_STAGEMASK;
361 flags2 &= CE_STAGEMASK;
370 int index_name_pos(struct index_state *istate, const char *name, int namelen)
375 last = istate->cache_nr;
376 while (last > first) {
377 int next = (last + first) >> 1;
378 struct cache_entry *ce = istate->cache[next];
379 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
391 /* Remove entry, return true if there are more entries to go.. */
392 int remove_index_entry_at(struct index_state *istate, int pos)
394 struct cache_entry *ce = istate->cache[pos];
396 remove_hash_entry(istate, ce);
397 istate->cache_changed = 1;
399 if (pos >= istate->cache_nr)
401 memmove(istate->cache + pos,
402 istate->cache + pos + 1,
403 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
407 int remove_file_from_index(struct index_state *istate, const char *path)
409 int pos = index_name_pos(istate, path, strlen(path));
412 cache_tree_invalidate_path(istate->cache_tree, path);
413 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
414 remove_index_entry_at(istate, pos);
418 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
420 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
423 static int index_name_pos_also_unmerged(struct index_state *istate,
424 const char *path, int namelen)
426 int pos = index_name_pos(istate, path, namelen);
427 struct cache_entry *ce;
432 /* maybe unmerged? */
434 if (pos >= istate->cache_nr ||
435 compare_name((ce = istate->cache[pos]), path, namelen))
438 /* order of preference: stage 2, 1, 3 */
439 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
440 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
441 !compare_name(ce, path, namelen))
446 int add_file_to_index(struct index_state *istate, const char *path, int verbose)
448 int size, namelen, pos;
450 struct cache_entry *ce;
451 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
453 if (lstat(path, &st))
454 die("%s: unable to stat (%s)", path, strerror(errno));
456 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
457 die("%s: can only add regular files, symbolic links or git-directories", path);
459 namelen = strlen(path);
460 if (S_ISDIR(st.st_mode)) {
461 while (namelen && path[namelen-1] == '/')
464 size = cache_entry_size(namelen);
465 ce = xcalloc(1, size);
466 memcpy(ce->name, path, namelen);
467 ce->ce_flags = namelen;
468 fill_stat_cache_info(ce, &st);
470 if (trust_executable_bit && has_symlinks)
471 ce->ce_mode = create_ce_mode(st.st_mode);
473 /* If there is an existing entry, pick the mode bits and type
474 * from it, otherwise assume unexecutable regular file.
476 struct cache_entry *ent;
477 int pos = index_name_pos_also_unmerged(istate, path, namelen);
479 ent = (0 <= pos) ? istate->cache[pos] : NULL;
480 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
483 pos = index_name_pos(istate, ce->name, namelen);
485 !ce_stage(istate->cache[pos]) &&
486 !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) {
487 /* Nothing changed, really */
489 ce_mark_uptodate(istate->cache[pos]);
493 if (index_path(ce->sha1, path, &st, 1))
494 die("unable to index file %s", path);
495 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
496 die("unable to add %s to index",path);
498 printf("add '%s'\n", path);
502 struct cache_entry *make_cache_entry(unsigned int mode,
503 const unsigned char *sha1, const char *path, int stage,
507 struct cache_entry *ce;
509 if (!verify_path(path))
513 size = cache_entry_size(len);
514 ce = xcalloc(1, size);
516 hashcpy(ce->sha1, sha1);
517 memcpy(ce->name, path, len);
518 ce->ce_flags = create_ce_flags(len, stage);
519 ce->ce_mode = create_ce_mode(mode);
522 return refresh_cache_entry(ce, 0);
527 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
529 int len = ce_namelen(a);
530 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
533 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
535 const char *match, *name;
541 len = ce_namelen(ce);
543 while ((match = *pathspec++) != NULL) {
544 int matchlen = strlen(match);
547 if (memcmp(name, match, matchlen))
549 if (matchlen && name[matchlen-1] == '/')
551 if (name[matchlen] == '/' || !name[matchlen])
560 * We fundamentally don't like some paths: we don't want
561 * dot or dot-dot anywhere, and for obvious reasons don't
562 * want to recurse into ".git" either.
564 * Also, we don't want double slashes or slashes at the
565 * end that can make pathnames ambiguous.
567 static int verify_dotfile(const char *rest)
570 * The first character was '.', but that
571 * has already been discarded, we now test
575 /* "." is not allowed */
580 * ".git" followed by NUL or slash is bad. This
581 * shares the path end test with the ".." case.
591 if (rest[1] == '\0' || rest[1] == '/')
597 int verify_path(const char *path)
614 if (verify_dotfile(path))
624 * Do we have another file that has the beginning components being a
625 * proper superset of the name we're trying to add?
627 static int has_file_name(struct index_state *istate,
628 const struct cache_entry *ce, int pos, int ok_to_replace)
631 int len = ce_namelen(ce);
632 int stage = ce_stage(ce);
633 const char *name = ce->name;
635 while (pos < istate->cache_nr) {
636 struct cache_entry *p = istate->cache[pos++];
638 if (len >= ce_namelen(p))
640 if (memcmp(name, p->name, len))
642 if (ce_stage(p) != stage)
644 if (p->name[len] != '/')
646 if (p->ce_flags & CE_REMOVE)
651 remove_index_entry_at(istate, --pos);
657 * Do we have another file with a pathname that is a proper
658 * subset of the name we're trying to add?
660 static int has_dir_name(struct index_state *istate,
661 const struct cache_entry *ce, int pos, int ok_to_replace)
664 int stage = ce_stage(ce);
665 const char *name = ce->name;
666 const char *slash = name + ce_namelen(ce);
674 if (slash <= ce->name)
679 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
682 * Found one, but not so fast. This could
683 * be a marker that says "I was here, but
684 * I am being removed". Such an entry is
685 * not a part of the resulting tree, and
686 * it is Ok to have a directory at the same
689 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
693 remove_index_entry_at(istate, pos);
701 * Trivial optimization: if we find an entry that
702 * already matches the sub-directory, then we know
703 * we're ok, and we can exit.
705 while (pos < istate->cache_nr) {
706 struct cache_entry *p = istate->cache[pos];
707 if ((ce_namelen(p) <= len) ||
708 (p->name[len] != '/') ||
709 memcmp(p->name, name, len))
710 break; /* not our subdirectory */
711 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
713 * p is at the same stage as our entry, and
714 * is a subdirectory of what we are looking
715 * at, so we cannot have conflicts at our
716 * level or anything shorter.
725 /* We may be in a situation where we already have path/file and path
726 * is being added, or we already have path and path/file is being
727 * added. Either one would result in a nonsense tree that has path
728 * twice when git-write-tree tries to write it out. Prevent it.
730 * If ok-to-replace is specified, we remove the conflicting entries
731 * from the cache so the caller should recompute the insert position.
732 * When this happens, we return non-zero.
734 static int check_file_directory_conflict(struct index_state *istate,
735 const struct cache_entry *ce,
736 int pos, int ok_to_replace)
741 * When ce is an "I am going away" entry, we allow it to be added
743 if (ce->ce_flags & CE_REMOVE)
747 * We check if the path is a sub-path of a subsequent pathname
748 * first, since removing those will not change the position
751 retval = has_file_name(istate, ce, pos, ok_to_replace);
754 * Then check if the path might have a clashing sub-directory
757 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
760 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
763 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
764 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
765 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
767 cache_tree_invalidate_path(istate->cache_tree, ce->name);
768 pos = index_name_pos(istate, ce->name, ce->ce_flags);
770 /* existing match? Just replace it. */
772 replace_index_entry(istate, pos, ce);
778 * Inserting a merged entry ("stage 0") into the index
779 * will always replace all non-merged entries..
781 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
782 while (ce_same_name(istate->cache[pos], ce)) {
784 if (!remove_index_entry_at(istate, pos))
791 if (!verify_path(ce->name))
794 if (!skip_df_check &&
795 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
797 return error("'%s' appears as both a file and as a directory",
799 pos = index_name_pos(istate, ce->name, ce->ce_flags);
805 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
809 if (option & ADD_CACHE_JUST_APPEND)
810 pos = istate->cache_nr;
813 ret = add_index_entry_with_check(istate, ce, option);
819 /* Make sure the array is big enough .. */
820 if (istate->cache_nr == istate->cache_alloc) {
821 istate->cache_alloc = alloc_nr(istate->cache_alloc);
822 istate->cache = xrealloc(istate->cache,
823 istate->cache_alloc * sizeof(struct cache_entry *));
828 if (istate->cache_nr > pos + 1)
829 memmove(istate->cache + pos + 1,
831 (istate->cache_nr - pos - 1) * sizeof(ce));
832 set_index_entry(istate, pos, ce);
833 istate->cache_changed = 1;
838 * "refresh" does not calculate a new sha1 file or bring the
839 * cache up-to-date for mode/content changes. But what it
840 * _does_ do is to "re-match" the stat information of a file
841 * with the cache, so that you can refresh the cache for a
842 * file that hasn't been changed but where the stat entry is
845 * For example, you'd want to do this after doing a "git-read-tree",
846 * to link up the stat cache details with the proper files.
848 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
849 struct cache_entry *ce,
850 unsigned int options, int *err)
853 struct cache_entry *updated;
855 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
860 if (lstat(ce->name, &st) < 0) {
866 changed = ie_match_stat(istate, ce, &st, options);
869 * The path is unchanged. If we were told to ignore
870 * valid bit, then we did the actual stat check and
871 * found that the entry is unmodified. If the entry
872 * is not marked VALID, this is the place to mark it
873 * valid again, under "assume unchanged" mode.
875 if (ignore_valid && assume_unchanged &&
876 !(ce->ce_flags & CE_VALID))
877 ; /* mark this one VALID again */
880 * We do not mark the index itself "modified"
881 * because CE_UPTODATE flag is in-core only;
882 * we are not going to write this change out.
884 ce_mark_uptodate(ce);
889 if (ie_modified(istate, ce, &st, options)) {
896 updated = xmalloc(size);
897 memcpy(updated, ce, size);
898 fill_stat_cache_info(updated, &st);
900 * If ignore_valid is not set, we should leave CE_VALID bit
901 * alone. Otherwise, paths marked with --no-assume-unchanged
902 * (i.e. things to be edited) will reacquire CE_VALID bit
903 * automatically, which is not really what we want.
905 if (!ignore_valid && assume_unchanged &&
906 !(ce->ce_flags & CE_VALID))
907 updated->ce_flags &= ~CE_VALID;
912 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
916 int really = (flags & REFRESH_REALLY) != 0;
917 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
918 int quiet = (flags & REFRESH_QUIET) != 0;
919 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
920 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
922 for (i = 0; i < istate->cache_nr; i++) {
923 struct cache_entry *ce, *new;
926 ce = istate->cache[i];
928 while ((i < istate->cache_nr) &&
929 ! strcmp(istate->cache[i]->name, ce->name))
934 printf("%s: needs merge\n", ce->name);
939 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
942 new = refresh_cache_ent(istate, ce, options, &cache_errno);
946 if (not_new && cache_errno == ENOENT)
948 if (really && cache_errno == EINVAL) {
949 /* If we are doing --really-refresh that
950 * means the index is not valid anymore.
952 ce->ce_flags &= ~CE_VALID;
953 istate->cache_changed = 1;
957 printf("%s: needs update\n", ce->name);
962 replace_index_entry(istate, i, new);
967 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
969 return refresh_cache_ent(&the_index, ce, really, NULL);
972 static int verify_hdr(struct cache_header *hdr, unsigned long size)
975 unsigned char sha1[20];
977 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
978 return error("bad signature");
979 if (hdr->hdr_version != htonl(2))
980 return error("bad index version");
982 SHA1_Update(&c, hdr, size - 20);
983 SHA1_Final(sha1, &c);
984 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
985 return error("bad index file sha1 signature");
989 static int read_index_extension(struct index_state *istate,
990 const char *ext, void *data, unsigned long sz)
992 switch (CACHE_EXT(ext)) {
994 istate->cache_tree = cache_tree_read(data, sz);
997 if (*ext < 'A' || 'Z' < *ext)
998 return error("index uses %.4s extension, which we do not understand",
1000 fprintf(stderr, "ignoring %.4s extension\n", ext);
1006 int read_index(struct index_state *istate)
1008 return read_index_from(istate, get_index_file());
1011 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1015 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1016 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1017 ce->ce_dev = ntohl(ondisk->dev);
1018 ce->ce_ino = ntohl(ondisk->ino);
1019 ce->ce_mode = ntohl(ondisk->mode);
1020 ce->ce_uid = ntohl(ondisk->uid);
1021 ce->ce_gid = ntohl(ondisk->gid);
1022 ce->ce_size = ntohl(ondisk->size);
1023 /* On-disk flags are just 16 bits */
1024 ce->ce_flags = ntohs(ondisk->flags);
1025 hashcpy(ce->sha1, ondisk->sha1);
1027 len = ce->ce_flags & CE_NAMEMASK;
1028 if (len == CE_NAMEMASK)
1029 len = strlen(ondisk->name);
1031 * NEEDSWORK: If the original index is crafted, this copy could
1034 memcpy(ce->name, ondisk->name, len + 1);
1037 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1041 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1044 * Alignment can cause differences. This should be "alignof", but
1045 * since that's a gcc'ism, just use the size of a pointer.
1047 per_entry += sizeof(void *);
1048 return ondisk_size + entries*per_entry;
1051 /* remember to discard_cache() before reading a different cache! */
1052 int read_index_from(struct index_state *istate, const char *path)
1056 unsigned long src_offset, dst_offset;
1057 struct cache_header *hdr;
1063 return istate->cache_nr;
1066 istate->timestamp = 0;
1067 fd = open(path, O_RDONLY);
1069 if (errno == ENOENT)
1071 die("index file open failed (%s)", strerror(errno));
1075 die("cannot stat the open index (%s)", strerror(errno));
1078 mmap_size = xsize_t(st.st_size);
1079 if (mmap_size < sizeof(struct cache_header) + 20)
1080 die("index file smaller than expected");
1082 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1084 if (mmap == MAP_FAILED)
1085 die("unable to map index file");
1088 if (verify_hdr(hdr, mmap_size) < 0)
1091 istate->cache_nr = ntohl(hdr->hdr_entries);
1092 istate->cache_alloc = alloc_nr(istate->cache_nr);
1093 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1096 * The disk format is actually larger than the in-memory format,
1097 * due to space for nsec etc, so even though the in-memory one
1098 * has room for a few more flags, we can allocate using the same
1101 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1103 src_offset = sizeof(*hdr);
1105 for (i = 0; i < istate->cache_nr; i++) {
1106 struct ondisk_cache_entry *disk_ce;
1107 struct cache_entry *ce;
1109 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1110 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1111 convert_from_disk(disk_ce, ce);
1112 set_index_entry(istate, i, ce);
1114 src_offset += ondisk_ce_size(ce);
1115 dst_offset += ce_size(ce);
1117 istate->timestamp = st.st_mtime;
1118 while (src_offset <= mmap_size - 20 - 8) {
1119 /* After an array of active_nr index entries,
1120 * there can be arbitrary number of extended
1121 * sections, each of which is prefixed with
1122 * extension name (4-byte) and section length
1123 * in 4-byte network byte order.
1125 unsigned long extsize;
1126 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1127 extsize = ntohl(extsize);
1128 if (read_index_extension(istate,
1129 (const char *) mmap + src_offset,
1130 (char *) mmap + src_offset + 8,
1134 src_offset += extsize;
1136 munmap(mmap, mmap_size);
1137 return istate->cache_nr;
1140 munmap(mmap, mmap_size);
1142 die("index file corrupt");
1145 int discard_index(struct index_state *istate)
1147 istate->cache_nr = 0;
1148 istate->cache_changed = 0;
1149 istate->timestamp = 0;
1150 free_hash(&istate->name_hash);
1151 cache_tree_free(&(istate->cache_tree));
1152 free(istate->alloc);
1153 istate->alloc = NULL;
1155 /* no need to throw away allocated active_cache */
1159 #define WRITE_BUFFER_SIZE 8192
1160 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1161 static unsigned long write_buffer_len;
1163 static int ce_write_flush(SHA_CTX *context, int fd)
1165 unsigned int buffered = write_buffer_len;
1167 SHA1_Update(context, write_buffer, buffered);
1168 if (write_in_full(fd, write_buffer, buffered) != buffered)
1170 write_buffer_len = 0;
1175 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1178 unsigned int buffered = write_buffer_len;
1179 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1182 memcpy(write_buffer + buffered, data, partial);
1183 buffered += partial;
1184 if (buffered == WRITE_BUFFER_SIZE) {
1185 write_buffer_len = buffered;
1186 if (ce_write_flush(context, fd))
1190 write_buffer_len = buffered;
1192 data = (char *) data + partial;
1197 static int write_index_ext_header(SHA_CTX *context, int fd,
1198 unsigned int ext, unsigned int sz)
1202 return ((ce_write(context, fd, &ext, 4) < 0) ||
1203 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1206 static int ce_flush(SHA_CTX *context, int fd)
1208 unsigned int left = write_buffer_len;
1211 write_buffer_len = 0;
1212 SHA1_Update(context, write_buffer, left);
1215 /* Flush first if not enough space for SHA1 signature */
1216 if (left + 20 > WRITE_BUFFER_SIZE) {
1217 if (write_in_full(fd, write_buffer, left) != left)
1222 /* Append the SHA1 signature at the end */
1223 SHA1_Final(write_buffer + left, context);
1225 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1228 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1231 * The only thing we care about in this function is to smudge the
1232 * falsely clean entry due to touch-update-touch race, so we leave
1233 * everything else as they are. We are called for entries whose
1234 * ce_mtime match the index file mtime.
1238 if (lstat(ce->name, &st) < 0)
1240 if (ce_match_stat_basic(ce, &st))
1242 if (ce_modified_check_fs(ce, &st)) {
1243 /* This is "racily clean"; smudge it. Note that this
1244 * is a tricky code. At first glance, it may appear
1245 * that it can break with this sequence:
1247 * $ echo xyzzy >frotz
1248 * $ git-update-index --add frotz
1251 * $ echo filfre >nitfol
1252 * $ git-update-index --add nitfol
1254 * but it does not. When the second update-index runs,
1255 * it notices that the entry "frotz" has the same timestamp
1256 * as index, and if we were to smudge it by resetting its
1257 * size to zero here, then the object name recorded
1258 * in index is the 6-byte file but the cached stat information
1259 * becomes zero --- which would then match what we would
1260 * obtain from the filesystem next time we stat("frotz").
1262 * However, the second update-index, before calling
1263 * this function, notices that the cached size is 6
1264 * bytes and what is on the filesystem is an empty
1265 * file, and never calls us, so the cached size information
1266 * for "frotz" stays 6 which does not match the filesystem.
1272 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1274 int size = ondisk_ce_size(ce);
1275 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1277 ondisk->ctime.sec = htonl(ce->ce_ctime);
1278 ondisk->ctime.nsec = 0;
1279 ondisk->mtime.sec = htonl(ce->ce_mtime);
1280 ondisk->mtime.nsec = 0;
1281 ondisk->dev = htonl(ce->ce_dev);
1282 ondisk->ino = htonl(ce->ce_ino);
1283 ondisk->mode = htonl(ce->ce_mode);
1284 ondisk->uid = htonl(ce->ce_uid);
1285 ondisk->gid = htonl(ce->ce_gid);
1286 ondisk->size = htonl(ce->ce_size);
1287 hashcpy(ondisk->sha1, ce->sha1);
1288 ondisk->flags = htons(ce->ce_flags);
1289 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1291 return ce_write(c, fd, ondisk, size);
1294 int write_index(struct index_state *istate, int newfd)
1297 struct cache_header hdr;
1298 int i, err, removed;
1299 struct cache_entry **cache = istate->cache;
1300 int entries = istate->cache_nr;
1302 for (i = removed = 0; i < entries; i++)
1303 if (cache[i]->ce_flags & CE_REMOVE)
1306 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1307 hdr.hdr_version = htonl(2);
1308 hdr.hdr_entries = htonl(entries - removed);
1311 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1314 for (i = 0; i < entries; i++) {
1315 struct cache_entry *ce = cache[i];
1316 if (ce->ce_flags & CE_REMOVE)
1318 if (is_racy_timestamp(istate, ce))
1319 ce_smudge_racily_clean_entry(ce);
1320 if (ce_write_entry(&c, newfd, ce) < 0)
1324 /* Write extension data here */
1325 if (istate->cache_tree) {
1328 strbuf_init(&sb, 0);
1329 cache_tree_write(&sb, istate->cache_tree);
1330 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1331 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1332 strbuf_release(&sb);
1336 return ce_flush(&c, newfd);