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"
19 * The first letter should be 'A'..'Z' for extensions that are not
20 * necessary for a correct operation (i.e. optimization data).
21 * When new extensions are added that _needs_ to be understood in
22 * order to correctly interpret the index file, pick character that
23 * is outside the range, to cause the reader to abort.
26 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
27 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
29 struct index_state the_index;
31 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
33 istate->cache[nr] = ce;
34 add_name_hash(istate, ce);
37 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
39 struct cache_entry *old = istate->cache[nr];
41 remove_name_hash(old);
42 set_index_entry(istate, nr, ce);
43 istate->cache_changed = 1;
46 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
48 struct cache_entry *old = istate->cache[nr], *new;
49 int namelen = strlen(new_name);
51 new = xmalloc(cache_entry_size(namelen));
52 copy_cache_entry(new, old);
53 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
54 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
55 memcpy(new->name, new_name, namelen + 1);
57 cache_tree_invalidate_path(istate->cache_tree, old->name);
58 remove_index_entry_at(istate, nr);
59 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
63 * This only updates the "non-critical" parts of the directory
64 * cache, ie the parts that aren't tracked by GIT, and only used
65 * to validate the cache.
67 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
69 ce->ce_ctime = st->st_ctime;
70 ce->ce_mtime = st->st_mtime;
71 ce->ce_dev = st->st_dev;
72 ce->ce_ino = st->st_ino;
73 ce->ce_uid = st->st_uid;
74 ce->ce_gid = st->st_gid;
75 ce->ce_size = st->st_size;
78 ce->ce_flags |= CE_VALID;
80 if (S_ISREG(st->st_mode))
84 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
87 int fd = open(ce->name, O_RDONLY);
90 unsigned char sha1[20];
91 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
92 match = hashcmp(sha1, ce->sha1);
93 /* index_fd() closed the file descriptor already */
98 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
104 enum object_type type;
107 target = xmalloc(expected_size);
108 len = readlink(ce->name, target, expected_size);
109 if (len != expected_size) {
113 buffer = read_sha1_file(ce->sha1, &type, &size);
118 if (size == expected_size)
119 match = memcmp(buffer, target, size);
125 static int ce_compare_gitlink(struct cache_entry *ce)
127 unsigned char sha1[20];
130 * We don't actually require that the .git directory
131 * under GITLINK directory be a valid git directory. It
132 * might even be missing (in case nobody populated that
135 * If so, we consider it always to match.
137 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
139 return hashcmp(sha1, ce->sha1);
142 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
144 switch (st->st_mode & S_IFMT) {
146 if (ce_compare_data(ce, st))
150 if (ce_compare_link(ce, xsize_t(st->st_size)))
154 if (S_ISGITLINK(ce->ce_mode))
155 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
162 static int is_empty_blob_sha1(const unsigned char *sha1)
164 static const unsigned char empty_blob_sha1[20] = {
165 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
166 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
169 return !hashcmp(sha1, empty_blob_sha1);
172 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
174 unsigned int changed = 0;
176 if (ce->ce_flags & CE_REMOVE)
177 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
179 switch (ce->ce_mode & S_IFMT) {
181 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
182 /* We consider only the owner x bit to be relevant for
185 if (trust_executable_bit &&
186 (0100 & (ce->ce_mode ^ st->st_mode)))
187 changed |= MODE_CHANGED;
190 if (!S_ISLNK(st->st_mode) &&
191 (has_symlinks || !S_ISREG(st->st_mode)))
192 changed |= TYPE_CHANGED;
195 /* We ignore most of the st_xxx fields for gitlinks */
196 if (!S_ISDIR(st->st_mode))
197 changed |= TYPE_CHANGED;
198 else if (ce_compare_gitlink(ce))
199 changed |= DATA_CHANGED;
202 die("internal error: ce_mode is %o", ce->ce_mode);
204 if (ce->ce_mtime != (unsigned int) st->st_mtime)
205 changed |= MTIME_CHANGED;
206 if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime)
207 changed |= CTIME_CHANGED;
209 if (ce->ce_uid != (unsigned int) st->st_uid ||
210 ce->ce_gid != (unsigned int) st->st_gid)
211 changed |= OWNER_CHANGED;
212 if (ce->ce_ino != (unsigned int) st->st_ino)
213 changed |= INODE_CHANGED;
217 * st_dev breaks on network filesystems where different
218 * clients will have different views of what "device"
219 * the filesystem is on
221 if (ce->ce_dev != (unsigned int) st->st_dev)
222 changed |= INODE_CHANGED;
225 if (ce->ce_size != (unsigned int) st->st_size)
226 changed |= DATA_CHANGED;
228 /* Racily smudged entry? */
230 if (!is_empty_blob_sha1(ce->sha1))
231 changed |= DATA_CHANGED;
237 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
239 return (!S_ISGITLINK(ce->ce_mode) &&
241 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
244 int ie_match_stat(const struct index_state *istate,
245 struct cache_entry *ce, struct stat *st,
246 unsigned int options)
248 unsigned int changed;
249 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
250 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
253 * If it's marked as always valid in the index, it's
254 * valid whatever the checked-out copy says.
256 if (!ignore_valid && (ce->ce_flags & CE_VALID))
259 changed = ce_match_stat_basic(ce, st);
262 * Within 1 second of this sequence:
263 * echo xyzzy >file && git-update-index --add file
264 * running this command:
266 * would give a falsely clean cache entry. The mtime and
267 * length match the cache, and other stat fields do not change.
269 * We could detect this at update-index time (the cache entry
270 * being registered/updated records the same time as "now")
271 * and delay the return from git-update-index, but that would
272 * effectively mean we can make at most one commit per second,
273 * which is not acceptable. Instead, we check cache entries
274 * whose mtime are the same as the index file timestamp more
275 * carefully than others.
277 if (!changed && is_racy_timestamp(istate, ce)) {
278 if (assume_racy_is_modified)
279 changed |= DATA_CHANGED;
281 changed |= ce_modified_check_fs(ce, st);
287 int ie_modified(const struct index_state *istate,
288 struct cache_entry *ce, struct stat *st, unsigned int options)
290 int changed, changed_fs;
292 changed = ie_match_stat(istate, ce, st, options);
296 * If the mode or type has changed, there's no point in trying
297 * to refresh the entry - it's not going to match
299 if (changed & (MODE_CHANGED | TYPE_CHANGED))
303 * Immediately after read-tree or update-index --cacheinfo,
304 * the length field is zero, as we have never even read the
305 * lstat(2) information once, and we cannot trust DATA_CHANGED
306 * returned by ie_match_stat() which in turn was returned by
307 * ce_match_stat_basic() to signal that the filesize of the
308 * blob changed. We have to actually go to the filesystem to
309 * see if the contents match, and if so, should answer "unchanged".
311 * The logic does not apply to gitlinks, as ce_match_stat_basic()
312 * already has checked the actual HEAD from the filesystem in the
313 * subproject. If ie_match_stat() already said it is different,
314 * then we know it is.
316 if ((changed & DATA_CHANGED) &&
317 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
320 changed_fs = ce_modified_check_fs(ce, st);
322 return changed | changed_fs;
326 int base_name_compare(const char *name1, int len1, int mode1,
327 const char *name2, int len2, int mode2)
329 unsigned char c1, c2;
330 int len = len1 < len2 ? len1 : len2;
333 cmp = memcmp(name1, name2, len);
338 if (!c1 && S_ISDIR(mode1))
340 if (!c2 && S_ISDIR(mode2))
342 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
346 * df_name_compare() is identical to base_name_compare(), except it
347 * compares conflicting directory/file entries as equal. Note that
348 * while a directory name compares as equal to a regular file, they
349 * then individually compare _differently_ to a filename that has
350 * a dot after the basename (because '\0' < '.' < '/').
352 * This is used by routines that want to traverse the git namespace
353 * but then handle conflicting entries together when possible.
355 int df_name_compare(const char *name1, int len1, int mode1,
356 const char *name2, int len2, int mode2)
358 int len = len1 < len2 ? len1 : len2, cmp;
359 unsigned char c1, c2;
361 cmp = memcmp(name1, name2, len);
364 /* Directories and files compare equal (same length, same name) */
368 if (!c1 && S_ISDIR(mode1))
371 if (!c2 && S_ISDIR(mode2))
373 if (c1 == '/' && !c2)
375 if (c2 == '/' && !c1)
380 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
382 int len1 = flags1 & CE_NAMEMASK;
383 int len2 = flags2 & CE_NAMEMASK;
384 int len = len1 < len2 ? len1 : len2;
387 cmp = memcmp(name1, name2, len);
396 flags1 &= CE_STAGEMASK;
397 flags2 &= CE_STAGEMASK;
406 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
411 last = istate->cache_nr;
412 while (last > first) {
413 int next = (last + first) >> 1;
414 struct cache_entry *ce = istate->cache[next];
415 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
427 /* Remove entry, return true if there are more entries to go.. */
428 int remove_index_entry_at(struct index_state *istate, int pos)
430 struct cache_entry *ce = istate->cache[pos];
432 remove_name_hash(ce);
433 istate->cache_changed = 1;
435 if (pos >= istate->cache_nr)
437 memmove(istate->cache + pos,
438 istate->cache + pos + 1,
439 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
443 int remove_file_from_index(struct index_state *istate, const char *path)
445 int pos = index_name_pos(istate, path, strlen(path));
448 cache_tree_invalidate_path(istate->cache_tree, path);
449 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
450 remove_index_entry_at(istate, pos);
454 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
456 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
459 static int index_name_pos_also_unmerged(struct index_state *istate,
460 const char *path, int namelen)
462 int pos = index_name_pos(istate, path, namelen);
463 struct cache_entry *ce;
468 /* maybe unmerged? */
470 if (pos >= istate->cache_nr ||
471 compare_name((ce = istate->cache[pos]), path, namelen))
474 /* order of preference: stage 2, 1, 3 */
475 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
476 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
477 !compare_name(ce, path, namelen))
482 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
484 int len = ce_namelen(ce);
485 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
489 * If we add a filename that aliases in the cache, we will use the
490 * name that we already have - but we don't want to update the same
491 * alias twice, because that implies that there were actually two
492 * different files with aliasing names!
494 * So we use the CE_ADDED flag to verify that the alias was an old
495 * one before we accept it as
497 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
500 struct cache_entry *new;
502 if (alias->ce_flags & CE_ADDED)
503 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
505 /* Ok, create the new entry using the name of the existing alias */
506 len = ce_namelen(alias);
507 new = xcalloc(1, cache_entry_size(len));
508 memcpy(new->name, alias->name, len);
509 copy_cache_entry(new, ce);
514 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
516 int size, namelen, was_same;
517 mode_t st_mode = st->st_mode;
518 struct cache_entry *ce, *alias;
519 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
520 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
521 int pretend = flags & ADD_CACHE_PRETEND;
523 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
524 return error("%s: can only add regular files, symbolic links or git-directories", path);
526 namelen = strlen(path);
527 if (S_ISDIR(st_mode)) {
528 while (namelen && path[namelen-1] == '/')
531 size = cache_entry_size(namelen);
532 ce = xcalloc(1, size);
533 memcpy(ce->name, path, namelen);
534 ce->ce_flags = namelen;
535 fill_stat_cache_info(ce, st);
537 if (trust_executable_bit && has_symlinks)
538 ce->ce_mode = create_ce_mode(st_mode);
540 /* If there is an existing entry, pick the mode bits and type
541 * from it, otherwise assume unexecutable regular file.
543 struct cache_entry *ent;
544 int pos = index_name_pos_also_unmerged(istate, path, namelen);
546 ent = (0 <= pos) ? istate->cache[pos] : NULL;
547 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
550 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
551 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
552 /* Nothing changed, really */
554 ce_mark_uptodate(alias);
555 alias->ce_flags |= CE_ADDED;
558 if (index_path(ce->sha1, path, st, 1))
559 return error("unable to index file %s", path);
560 if (ignore_case && alias && different_name(ce, alias))
561 ce = create_alias_ce(ce, alias);
562 ce->ce_flags |= CE_ADDED;
564 /* It was suspected to be racily clean, but it turns out to be Ok */
567 !hashcmp(alias->sha1, ce->sha1) &&
568 ce->ce_mode == alias->ce_mode);
572 else if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
573 return error("unable to add %s to index",path);
574 if (verbose && !was_same)
575 printf("add '%s'\n", path);
579 int add_file_to_index(struct index_state *istate, const char *path, int flags)
582 if (lstat(path, &st))
583 die("%s: unable to stat (%s)", path, strerror(errno));
584 return add_to_index(istate, path, &st, flags);
587 struct cache_entry *make_cache_entry(unsigned int mode,
588 const unsigned char *sha1, const char *path, int stage,
592 struct cache_entry *ce;
594 if (!verify_path(path))
598 size = cache_entry_size(len);
599 ce = xcalloc(1, size);
601 hashcpy(ce->sha1, sha1);
602 memcpy(ce->name, path, len);
603 ce->ce_flags = create_ce_flags(len, stage);
604 ce->ce_mode = create_ce_mode(mode);
607 return refresh_cache_entry(ce, 0);
612 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
614 int len = ce_namelen(a);
615 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
618 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
620 const char *match, *name;
626 len = ce_namelen(ce);
628 while ((match = *pathspec++) != NULL) {
629 int matchlen = strlen(match);
632 if (memcmp(name, match, matchlen))
634 if (matchlen && name[matchlen-1] == '/')
636 if (name[matchlen] == '/' || !name[matchlen])
645 * We fundamentally don't like some paths: we don't want
646 * dot or dot-dot anywhere, and for obvious reasons don't
647 * want to recurse into ".git" either.
649 * Also, we don't want double slashes or slashes at the
650 * end that can make pathnames ambiguous.
652 static int verify_dotfile(const char *rest)
655 * The first character was '.', but that
656 * has already been discarded, we now test
660 /* "." is not allowed */
665 * ".git" followed by NUL or slash is bad. This
666 * shares the path end test with the ".." case.
676 if (rest[1] == '\0' || rest[1] == '/')
682 int verify_path(const char *path)
699 if (verify_dotfile(path))
709 * Do we have another file that has the beginning components being a
710 * proper superset of the name we're trying to add?
712 static int has_file_name(struct index_state *istate,
713 const struct cache_entry *ce, int pos, int ok_to_replace)
716 int len = ce_namelen(ce);
717 int stage = ce_stage(ce);
718 const char *name = ce->name;
720 while (pos < istate->cache_nr) {
721 struct cache_entry *p = istate->cache[pos++];
723 if (len >= ce_namelen(p))
725 if (memcmp(name, p->name, len))
727 if (ce_stage(p) != stage)
729 if (p->name[len] != '/')
731 if (p->ce_flags & CE_REMOVE)
736 remove_index_entry_at(istate, --pos);
742 * Do we have another file with a pathname that is a proper
743 * subset of the name we're trying to add?
745 static int has_dir_name(struct index_state *istate,
746 const struct cache_entry *ce, int pos, int ok_to_replace)
749 int stage = ce_stage(ce);
750 const char *name = ce->name;
751 const char *slash = name + ce_namelen(ce);
759 if (slash <= ce->name)
764 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
767 * Found one, but not so fast. This could
768 * be a marker that says "I was here, but
769 * I am being removed". Such an entry is
770 * not a part of the resulting tree, and
771 * it is Ok to have a directory at the same
774 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
778 remove_index_entry_at(istate, pos);
786 * Trivial optimization: if we find an entry that
787 * already matches the sub-directory, then we know
788 * we're ok, and we can exit.
790 while (pos < istate->cache_nr) {
791 struct cache_entry *p = istate->cache[pos];
792 if ((ce_namelen(p) <= len) ||
793 (p->name[len] != '/') ||
794 memcmp(p->name, name, len))
795 break; /* not our subdirectory */
796 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
798 * p is at the same stage as our entry, and
799 * is a subdirectory of what we are looking
800 * at, so we cannot have conflicts at our
801 * level or anything shorter.
810 /* We may be in a situation where we already have path/file and path
811 * is being added, or we already have path and path/file is being
812 * added. Either one would result in a nonsense tree that has path
813 * twice when git-write-tree tries to write it out. Prevent it.
815 * If ok-to-replace is specified, we remove the conflicting entries
816 * from the cache so the caller should recompute the insert position.
817 * When this happens, we return non-zero.
819 static int check_file_directory_conflict(struct index_state *istate,
820 const struct cache_entry *ce,
821 int pos, int ok_to_replace)
826 * When ce is an "I am going away" entry, we allow it to be added
828 if (ce->ce_flags & CE_REMOVE)
832 * We check if the path is a sub-path of a subsequent pathname
833 * first, since removing those will not change the position
836 retval = has_file_name(istate, ce, pos, ok_to_replace);
839 * Then check if the path might have a clashing sub-directory
842 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
845 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
848 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
849 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
850 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
852 cache_tree_invalidate_path(istate->cache_tree, ce->name);
853 pos = index_name_pos(istate, ce->name, ce->ce_flags);
855 /* existing match? Just replace it. */
857 replace_index_entry(istate, pos, ce);
863 * Inserting a merged entry ("stage 0") into the index
864 * will always replace all non-merged entries..
866 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
867 while (ce_same_name(istate->cache[pos], ce)) {
869 if (!remove_index_entry_at(istate, pos))
876 if (!verify_path(ce->name))
879 if (!skip_df_check &&
880 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
882 return error("'%s' appears as both a file and as a directory",
884 pos = index_name_pos(istate, ce->name, ce->ce_flags);
890 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
894 if (option & ADD_CACHE_JUST_APPEND)
895 pos = istate->cache_nr;
898 ret = add_index_entry_with_check(istate, ce, option);
904 /* Make sure the array is big enough .. */
905 if (istate->cache_nr == istate->cache_alloc) {
906 istate->cache_alloc = alloc_nr(istate->cache_alloc);
907 istate->cache = xrealloc(istate->cache,
908 istate->cache_alloc * sizeof(struct cache_entry *));
913 if (istate->cache_nr > pos + 1)
914 memmove(istate->cache + pos + 1,
916 (istate->cache_nr - pos - 1) * sizeof(ce));
917 set_index_entry(istate, pos, ce);
918 istate->cache_changed = 1;
923 * "refresh" does not calculate a new sha1 file or bring the
924 * cache up-to-date for mode/content changes. But what it
925 * _does_ do is to "re-match" the stat information of a file
926 * with the cache, so that you can refresh the cache for a
927 * file that hasn't been changed but where the stat entry is
930 * For example, you'd want to do this after doing a "git-read-tree",
931 * to link up the stat cache details with the proper files.
933 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
934 struct cache_entry *ce,
935 unsigned int options, int *err)
938 struct cache_entry *updated;
940 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
946 * CE_VALID means the user promised us that the change to
947 * the work tree does not matter and told us not to worry.
949 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
950 ce_mark_uptodate(ce);
954 if (lstat(ce->name, &st) < 0) {
960 changed = ie_match_stat(istate, ce, &st, options);
963 * The path is unchanged. If we were told to ignore
964 * valid bit, then we did the actual stat check and
965 * found that the entry is unmodified. If the entry
966 * is not marked VALID, this is the place to mark it
967 * valid again, under "assume unchanged" mode.
969 if (ignore_valid && assume_unchanged &&
970 !(ce->ce_flags & CE_VALID))
971 ; /* mark this one VALID again */
974 * We do not mark the index itself "modified"
975 * because CE_UPTODATE flag is in-core only;
976 * we are not going to write this change out.
978 ce_mark_uptodate(ce);
983 if (ie_modified(istate, ce, &st, options)) {
990 updated = xmalloc(size);
991 memcpy(updated, ce, size);
992 fill_stat_cache_info(updated, &st);
994 * If ignore_valid is not set, we should leave CE_VALID bit
995 * alone. Otherwise, paths marked with --no-assume-unchanged
996 * (i.e. things to be edited) will reacquire CE_VALID bit
997 * automatically, which is not really what we want.
999 if (!ignore_valid && assume_unchanged &&
1000 !(ce->ce_flags & CE_VALID))
1001 updated->ce_flags &= ~CE_VALID;
1006 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
1010 int really = (flags & REFRESH_REALLY) != 0;
1011 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1012 int quiet = (flags & REFRESH_QUIET) != 0;
1013 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1014 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1015 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1016 const char *needs_update_message;
1018 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
1019 ? "locally modified" : "needs update");
1020 for (i = 0; i < istate->cache_nr; i++) {
1021 struct cache_entry *ce, *new;
1022 int cache_errno = 0;
1024 ce = istate->cache[i];
1025 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1029 while ((i < istate->cache_nr) &&
1030 ! strcmp(istate->cache[i]->name, ce->name))
1035 printf("%s: needs merge\n", ce->name);
1040 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1043 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1047 if (not_new && cache_errno == ENOENT)
1049 if (really && cache_errno == EINVAL) {
1050 /* If we are doing --really-refresh that
1051 * means the index is not valid anymore.
1053 ce->ce_flags &= ~CE_VALID;
1054 istate->cache_changed = 1;
1058 printf("%s: %s\n", ce->name, needs_update_message);
1063 replace_index_entry(istate, i, new);
1068 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1070 return refresh_cache_ent(&the_index, ce, really, NULL);
1073 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1076 unsigned char sha1[20];
1078 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1079 return error("bad signature");
1080 if (hdr->hdr_version != htonl(2))
1081 return error("bad index version");
1083 SHA1_Update(&c, hdr, size - 20);
1084 SHA1_Final(sha1, &c);
1085 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1086 return error("bad index file sha1 signature");
1090 static int read_index_extension(struct index_state *istate,
1091 const char *ext, void *data, unsigned long sz)
1093 switch (CACHE_EXT(ext)) {
1094 case CACHE_EXT_TREE:
1095 istate->cache_tree = cache_tree_read(data, sz);
1098 if (*ext < 'A' || 'Z' < *ext)
1099 return error("index uses %.4s extension, which we do not understand",
1101 fprintf(stderr, "ignoring %.4s extension\n", ext);
1107 int read_index(struct index_state *istate)
1109 return read_index_from(istate, get_index_file());
1112 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1116 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1117 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1118 ce->ce_dev = ntohl(ondisk->dev);
1119 ce->ce_ino = ntohl(ondisk->ino);
1120 ce->ce_mode = ntohl(ondisk->mode);
1121 ce->ce_uid = ntohl(ondisk->uid);
1122 ce->ce_gid = ntohl(ondisk->gid);
1123 ce->ce_size = ntohl(ondisk->size);
1124 /* On-disk flags are just 16 bits */
1125 ce->ce_flags = ntohs(ondisk->flags);
1127 /* For future extension: we do not understand this entry yet */
1128 if (ce->ce_flags & CE_EXTENDED)
1129 die("Unknown index entry format");
1130 hashcpy(ce->sha1, ondisk->sha1);
1132 len = ce->ce_flags & CE_NAMEMASK;
1133 if (len == CE_NAMEMASK)
1134 len = strlen(ondisk->name);
1136 * NEEDSWORK: If the original index is crafted, this copy could
1139 memcpy(ce->name, ondisk->name, len + 1);
1142 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1146 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1149 * Alignment can cause differences. This should be "alignof", but
1150 * since that's a gcc'ism, just use the size of a pointer.
1152 per_entry += sizeof(void *);
1153 return ondisk_size + entries*per_entry;
1156 /* remember to discard_cache() before reading a different cache! */
1157 int read_index_from(struct index_state *istate, const char *path)
1161 unsigned long src_offset, dst_offset;
1162 struct cache_header *hdr;
1167 if (istate->initialized)
1168 return istate->cache_nr;
1171 istate->timestamp = 0;
1172 fd = open(path, O_RDONLY);
1174 if (errno == ENOENT)
1176 die("index file open failed (%s)", strerror(errno));
1180 die("cannot stat the open index (%s)", strerror(errno));
1183 mmap_size = xsize_t(st.st_size);
1184 if (mmap_size < sizeof(struct cache_header) + 20)
1185 die("index file smaller than expected");
1187 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1189 if (mmap == MAP_FAILED)
1190 die("unable to map index file");
1193 if (verify_hdr(hdr, mmap_size) < 0)
1196 istate->cache_nr = ntohl(hdr->hdr_entries);
1197 istate->cache_alloc = alloc_nr(istate->cache_nr);
1198 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1201 * The disk format is actually larger than the in-memory format,
1202 * due to space for nsec etc, so even though the in-memory one
1203 * has room for a few more flags, we can allocate using the same
1206 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1207 istate->initialized = 1;
1209 src_offset = sizeof(*hdr);
1211 for (i = 0; i < istate->cache_nr; i++) {
1212 struct ondisk_cache_entry *disk_ce;
1213 struct cache_entry *ce;
1215 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1216 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1217 convert_from_disk(disk_ce, ce);
1218 set_index_entry(istate, i, ce);
1220 src_offset += ondisk_ce_size(ce);
1221 dst_offset += ce_size(ce);
1223 istate->timestamp = st.st_mtime;
1224 while (src_offset <= mmap_size - 20 - 8) {
1225 /* After an array of active_nr index entries,
1226 * there can be arbitrary number of extended
1227 * sections, each of which is prefixed with
1228 * extension name (4-byte) and section length
1229 * in 4-byte network byte order.
1231 unsigned long extsize;
1232 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1233 extsize = ntohl(extsize);
1234 if (read_index_extension(istate,
1235 (const char *) mmap + src_offset,
1236 (char *) mmap + src_offset + 8,
1240 src_offset += extsize;
1242 munmap(mmap, mmap_size);
1243 return istate->cache_nr;
1246 munmap(mmap, mmap_size);
1248 die("index file corrupt");
1251 int discard_index(struct index_state *istate)
1253 istate->cache_nr = 0;
1254 istate->cache_changed = 0;
1255 istate->timestamp = 0;
1256 free_hash(&istate->name_hash);
1257 cache_tree_free(&(istate->cache_tree));
1258 free(istate->alloc);
1259 istate->alloc = NULL;
1260 istate->initialized = 0;
1262 /* no need to throw away allocated active_cache */
1266 int unmerged_index(const struct index_state *istate)
1269 for (i = 0; i < istate->cache_nr; i++) {
1270 if (ce_stage(istate->cache[i]))
1276 #define WRITE_BUFFER_SIZE 8192
1277 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1278 static unsigned long write_buffer_len;
1280 static int ce_write_flush(SHA_CTX *context, int fd)
1282 unsigned int buffered = write_buffer_len;
1284 SHA1_Update(context, write_buffer, buffered);
1285 if (write_in_full(fd, write_buffer, buffered) != buffered)
1287 write_buffer_len = 0;
1292 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1295 unsigned int buffered = write_buffer_len;
1296 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1299 memcpy(write_buffer + buffered, data, partial);
1300 buffered += partial;
1301 if (buffered == WRITE_BUFFER_SIZE) {
1302 write_buffer_len = buffered;
1303 if (ce_write_flush(context, fd))
1307 write_buffer_len = buffered;
1309 data = (char *) data + partial;
1314 static int write_index_ext_header(SHA_CTX *context, int fd,
1315 unsigned int ext, unsigned int sz)
1319 return ((ce_write(context, fd, &ext, 4) < 0) ||
1320 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1323 static int ce_flush(SHA_CTX *context, int fd)
1325 unsigned int left = write_buffer_len;
1328 write_buffer_len = 0;
1329 SHA1_Update(context, write_buffer, left);
1332 /* Flush first if not enough space for SHA1 signature */
1333 if (left + 20 > WRITE_BUFFER_SIZE) {
1334 if (write_in_full(fd, write_buffer, left) != left)
1339 /* Append the SHA1 signature at the end */
1340 SHA1_Final(write_buffer + left, context);
1342 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1345 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1348 * The only thing we care about in this function is to smudge the
1349 * falsely clean entry due to touch-update-touch race, so we leave
1350 * everything else as they are. We are called for entries whose
1351 * ce_mtime match the index file mtime.
1353 * Note that this actually does not do much for gitlinks, for
1354 * which ce_match_stat_basic() always goes to the actual
1355 * contents. The caller checks with is_racy_timestamp() which
1356 * always says "no" for gitlinks, so we are not called for them ;-)
1360 if (lstat(ce->name, &st) < 0)
1362 if (ce_match_stat_basic(ce, &st))
1364 if (ce_modified_check_fs(ce, &st)) {
1365 /* This is "racily clean"; smudge it. Note that this
1366 * is a tricky code. At first glance, it may appear
1367 * that it can break with this sequence:
1369 * $ echo xyzzy >frotz
1370 * $ git-update-index --add frotz
1373 * $ echo filfre >nitfol
1374 * $ git-update-index --add nitfol
1376 * but it does not. When the second update-index runs,
1377 * it notices that the entry "frotz" has the same timestamp
1378 * as index, and if we were to smudge it by resetting its
1379 * size to zero here, then the object name recorded
1380 * in index is the 6-byte file but the cached stat information
1381 * becomes zero --- which would then match what we would
1382 * obtain from the filesystem next time we stat("frotz").
1384 * However, the second update-index, before calling
1385 * this function, notices that the cached size is 6
1386 * bytes and what is on the filesystem is an empty
1387 * file, and never calls us, so the cached size information
1388 * for "frotz" stays 6 which does not match the filesystem.
1394 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1396 int size = ondisk_ce_size(ce);
1397 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1399 ondisk->ctime.sec = htonl(ce->ce_ctime);
1400 ondisk->ctime.nsec = 0;
1401 ondisk->mtime.sec = htonl(ce->ce_mtime);
1402 ondisk->mtime.nsec = 0;
1403 ondisk->dev = htonl(ce->ce_dev);
1404 ondisk->ino = htonl(ce->ce_ino);
1405 ondisk->mode = htonl(ce->ce_mode);
1406 ondisk->uid = htonl(ce->ce_uid);
1407 ondisk->gid = htonl(ce->ce_gid);
1408 ondisk->size = htonl(ce->ce_size);
1409 hashcpy(ondisk->sha1, ce->sha1);
1410 ondisk->flags = htons(ce->ce_flags);
1411 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1413 return ce_write(c, fd, ondisk, size);
1416 int write_index(const struct index_state *istate, int newfd)
1419 struct cache_header hdr;
1420 int i, err, removed;
1421 struct cache_entry **cache = istate->cache;
1422 int entries = istate->cache_nr;
1424 for (i = removed = 0; i < entries; i++)
1425 if (cache[i]->ce_flags & CE_REMOVE)
1428 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1429 hdr.hdr_version = htonl(2);
1430 hdr.hdr_entries = htonl(entries - removed);
1433 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1436 for (i = 0; i < entries; i++) {
1437 struct cache_entry *ce = cache[i];
1438 if (ce->ce_flags & CE_REMOVE)
1440 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1441 ce_smudge_racily_clean_entry(ce);
1442 if (ce_write_entry(&c, newfd, ce) < 0)
1446 /* Write extension data here */
1447 if (istate->cache_tree) {
1450 strbuf_init(&sb, 0);
1451 cache_tree_write(&sb, istate->cache_tree);
1452 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1453 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1454 strbuf_release(&sb);
1458 return ce_flush(&c, newfd);
1462 * Read the index file that is potentially unmerged into given
1463 * index_state, dropping any unmerged entries. Returns true is
1464 * the index is unmerged. Callers who want to refuse to work
1465 * from an unmerged state can call this and check its return value,
1466 * instead of calling read_cache().
1468 int read_index_unmerged(struct index_state *istate)
1471 struct cache_entry **dst;
1472 struct cache_entry *last = NULL;
1475 dst = istate->cache;
1476 for (i = 0; i < istate->cache_nr; i++) {
1477 struct cache_entry *ce = istate->cache[i];
1479 remove_name_hash(ce);
1480 if (last && !strcmp(ce->name, last->name))
1482 cache_tree_invalidate_path(istate->cache_tree, ce->name);
1488 istate->cache_nr = dst - istate->cache;
1492 struct update_callback_data
1498 static void update_callback(struct diff_queue_struct *q,
1499 struct diff_options *opt, void *cbdata)
1502 struct update_callback_data *data = cbdata;
1504 for (i = 0; i < q->nr; i++) {
1505 struct diff_filepair *p = q->queue[i];
1506 const char *path = p->one->path;
1507 switch (p->status) {
1509 die("unexpected diff status %c", p->status);
1510 case DIFF_STATUS_UNMERGED:
1511 case DIFF_STATUS_MODIFIED:
1512 case DIFF_STATUS_TYPE_CHANGED:
1513 if (add_file_to_index(&the_index, path, data->flags)) {
1514 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1515 die("updating files failed");
1519 case DIFF_STATUS_DELETED:
1520 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1522 if (!(data->flags & ADD_CACHE_PRETEND))
1523 remove_file_from_index(&the_index, path);
1524 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1525 printf("remove '%s'\n", path);
1531 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1533 struct update_callback_data data;
1534 struct rev_info rev;
1535 init_revisions(&rev, prefix);
1536 setup_revisions(0, NULL, &rev, NULL);
1537 rev.prune_data = pathspec;
1538 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1539 rev.diffopt.format_callback = update_callback;
1541 data.add_errors = 0;
1542 rev.diffopt.format_callback_data = &data;
1543 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1544 return !!data.add_errors;