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"
20 * The first letter should be 'A'..'Z' for extensions that are not
21 * necessary for a correct operation (i.e. optimization data).
22 * When new extensions are added that _needs_ to be understood in
23 * order to correctly interpret the index file, pick character that
24 * is outside the range, to cause the reader to abort.
27 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
28 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
30 struct index_state the_index;
32 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
34 istate->cache[nr] = ce;
35 add_name_hash(istate, ce);
38 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
40 struct cache_entry *old = istate->cache[nr];
42 remove_name_hash(old);
43 set_index_entry(istate, nr, ce);
44 istate->cache_changed = 1;
47 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
49 struct cache_entry *old = istate->cache[nr], *new;
50 int namelen = strlen(new_name);
52 new = xmalloc(cache_entry_size(namelen));
53 copy_cache_entry(new, old);
54 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
55 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
56 memcpy(new->name, new_name, namelen + 1);
58 cache_tree_invalidate_path(istate->cache_tree, old->name);
59 remove_index_entry_at(istate, nr);
60 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
64 * This only updates the "non-critical" parts of the directory
65 * cache, ie the parts that aren't tracked by GIT, and only used
66 * to validate the cache.
68 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
70 ce->ce_ctime = st->st_ctime;
71 ce->ce_mtime = st->st_mtime;
72 ce->ce_dev = st->st_dev;
73 ce->ce_ino = st->st_ino;
74 ce->ce_uid = st->st_uid;
75 ce->ce_gid = st->st_gid;
76 ce->ce_size = st->st_size;
79 ce->ce_flags |= CE_VALID;
81 if (S_ISREG(st->st_mode))
85 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
88 int fd = open(ce->name, O_RDONLY);
91 unsigned char sha1[20];
92 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
93 match = hashcmp(sha1, ce->sha1);
94 /* index_fd() closed the file descriptor already */
99 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
104 enum object_type type;
105 struct strbuf sb = STRBUF_INIT;
107 if (strbuf_readlink(&sb, ce->name, expected_size))
110 buffer = read_sha1_file(ce->sha1, &type, &size);
113 match = memcmp(buffer, sb.buf, size);
120 static int ce_compare_gitlink(struct cache_entry *ce)
122 unsigned char sha1[20];
125 * We don't actually require that the .git directory
126 * under GITLINK directory be a valid git directory. It
127 * might even be missing (in case nobody populated that
130 * If so, we consider it always to match.
132 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
134 return hashcmp(sha1, ce->sha1);
137 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
139 switch (st->st_mode & S_IFMT) {
141 if (ce_compare_data(ce, st))
145 if (ce_compare_link(ce, xsize_t(st->st_size)))
149 if (S_ISGITLINK(ce->ce_mode))
150 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
157 int is_empty_blob_sha1(const unsigned char *sha1)
159 static const unsigned char empty_blob_sha1[20] = {
160 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
161 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
164 return !hashcmp(sha1, empty_blob_sha1);
167 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
169 unsigned int changed = 0;
171 if (ce->ce_flags & CE_REMOVE)
172 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
174 switch (ce->ce_mode & S_IFMT) {
176 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
177 /* We consider only the owner x bit to be relevant for
180 if (trust_executable_bit &&
181 (0100 & (ce->ce_mode ^ st->st_mode)))
182 changed |= MODE_CHANGED;
185 if (!S_ISLNK(st->st_mode) &&
186 (has_symlinks || !S_ISREG(st->st_mode)))
187 changed |= TYPE_CHANGED;
190 /* We ignore most of the st_xxx fields for gitlinks */
191 if (!S_ISDIR(st->st_mode))
192 changed |= TYPE_CHANGED;
193 else if (ce_compare_gitlink(ce))
194 changed |= DATA_CHANGED;
197 die("internal error: ce_mode is %o", ce->ce_mode);
199 if (ce->ce_mtime != (unsigned int) st->st_mtime)
200 changed |= MTIME_CHANGED;
201 if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime)
202 changed |= CTIME_CHANGED;
204 if (ce->ce_uid != (unsigned int) st->st_uid ||
205 ce->ce_gid != (unsigned int) st->st_gid)
206 changed |= OWNER_CHANGED;
207 if (ce->ce_ino != (unsigned int) st->st_ino)
208 changed |= INODE_CHANGED;
212 * st_dev breaks on network filesystems where different
213 * clients will have different views of what "device"
214 * the filesystem is on
216 if (ce->ce_dev != (unsigned int) st->st_dev)
217 changed |= INODE_CHANGED;
220 if (ce->ce_size != (unsigned int) st->st_size)
221 changed |= DATA_CHANGED;
223 /* Racily smudged entry? */
225 if (!is_empty_blob_sha1(ce->sha1))
226 changed |= DATA_CHANGED;
232 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
234 return (!S_ISGITLINK(ce->ce_mode) &&
236 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
239 int ie_match_stat(const struct index_state *istate,
240 struct cache_entry *ce, struct stat *st,
241 unsigned int options)
243 unsigned int changed;
244 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
245 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
248 * If it's marked as always valid in the index, it's
249 * valid whatever the checked-out copy says.
251 if (!ignore_valid && (ce->ce_flags & CE_VALID))
255 * Intent-to-add entries have not been added, so the index entry
256 * by definition never matches what is in the work tree until it
257 * actually gets added.
259 if (ce->ce_flags & CE_INTENT_TO_ADD)
260 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
262 changed = ce_match_stat_basic(ce, st);
265 * Within 1 second of this sequence:
266 * echo xyzzy >file && git-update-index --add file
267 * running this command:
269 * would give a falsely clean cache entry. The mtime and
270 * length match the cache, and other stat fields do not change.
272 * We could detect this at update-index time (the cache entry
273 * being registered/updated records the same time as "now")
274 * and delay the return from git-update-index, but that would
275 * effectively mean we can make at most one commit per second,
276 * which is not acceptable. Instead, we check cache entries
277 * whose mtime are the same as the index file timestamp more
278 * carefully than others.
280 if (!changed && is_racy_timestamp(istate, ce)) {
281 if (assume_racy_is_modified)
282 changed |= DATA_CHANGED;
284 changed |= ce_modified_check_fs(ce, st);
290 int ie_modified(const struct index_state *istate,
291 struct cache_entry *ce, struct stat *st, unsigned int options)
293 int changed, changed_fs;
295 changed = ie_match_stat(istate, ce, st, options);
299 * If the mode or type has changed, there's no point in trying
300 * to refresh the entry - it's not going to match
302 if (changed & (MODE_CHANGED | TYPE_CHANGED))
306 * Immediately after read-tree or update-index --cacheinfo,
307 * the length field is zero, as we have never even read the
308 * lstat(2) information once, and we cannot trust DATA_CHANGED
309 * returned by ie_match_stat() which in turn was returned by
310 * ce_match_stat_basic() to signal that the filesize of the
311 * blob changed. We have to actually go to the filesystem to
312 * see if the contents match, and if so, should answer "unchanged".
314 * The logic does not apply to gitlinks, as ce_match_stat_basic()
315 * already has checked the actual HEAD from the filesystem in the
316 * subproject. If ie_match_stat() already said it is different,
317 * then we know it is.
319 if ((changed & DATA_CHANGED) &&
320 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
323 changed_fs = ce_modified_check_fs(ce, st);
325 return changed | changed_fs;
329 int base_name_compare(const char *name1, int len1, int mode1,
330 const char *name2, int len2, int mode2)
332 unsigned char c1, c2;
333 int len = len1 < len2 ? len1 : len2;
336 cmp = memcmp(name1, name2, len);
341 if (!c1 && S_ISDIR(mode1))
343 if (!c2 && S_ISDIR(mode2))
345 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
349 * df_name_compare() is identical to base_name_compare(), except it
350 * compares conflicting directory/file entries as equal. Note that
351 * while a directory name compares as equal to a regular file, they
352 * then individually compare _differently_ to a filename that has
353 * a dot after the basename (because '\0' < '.' < '/').
355 * This is used by routines that want to traverse the git namespace
356 * but then handle conflicting entries together when possible.
358 int df_name_compare(const char *name1, int len1, int mode1,
359 const char *name2, int len2, int mode2)
361 int len = len1 < len2 ? len1 : len2, cmp;
362 unsigned char c1, c2;
364 cmp = memcmp(name1, name2, len);
367 /* Directories and files compare equal (same length, same name) */
371 if (!c1 && S_ISDIR(mode1))
374 if (!c2 && S_ISDIR(mode2))
376 if (c1 == '/' && !c2)
378 if (c2 == '/' && !c1)
383 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
385 int len1 = flags1 & CE_NAMEMASK;
386 int len2 = flags2 & CE_NAMEMASK;
387 int len = len1 < len2 ? len1 : len2;
390 cmp = memcmp(name1, name2, len);
399 flags1 &= CE_STAGEMASK;
400 flags2 &= CE_STAGEMASK;
409 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
414 last = istate->cache_nr;
415 while (last > first) {
416 int next = (last + first) >> 1;
417 struct cache_entry *ce = istate->cache[next];
418 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
430 /* Remove entry, return true if there are more entries to go.. */
431 int remove_index_entry_at(struct index_state *istate, int pos)
433 struct cache_entry *ce = istate->cache[pos];
435 remove_name_hash(ce);
436 istate->cache_changed = 1;
438 if (pos >= istate->cache_nr)
440 memmove(istate->cache + pos,
441 istate->cache + pos + 1,
442 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
446 int remove_file_from_index(struct index_state *istate, const char *path)
448 int pos = index_name_pos(istate, path, strlen(path));
451 cache_tree_invalidate_path(istate->cache_tree, path);
452 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
453 remove_index_entry_at(istate, pos);
457 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
459 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
462 static int index_name_pos_also_unmerged(struct index_state *istate,
463 const char *path, int namelen)
465 int pos = index_name_pos(istate, path, namelen);
466 struct cache_entry *ce;
471 /* maybe unmerged? */
473 if (pos >= istate->cache_nr ||
474 compare_name((ce = istate->cache[pos]), path, namelen))
477 /* order of preference: stage 2, 1, 3 */
478 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
479 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
480 !compare_name(ce, path, namelen))
485 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
487 int len = ce_namelen(ce);
488 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
492 * If we add a filename that aliases in the cache, we will use the
493 * name that we already have - but we don't want to update the same
494 * alias twice, because that implies that there were actually two
495 * different files with aliasing names!
497 * So we use the CE_ADDED flag to verify that the alias was an old
498 * one before we accept it as
500 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
503 struct cache_entry *new;
505 if (alias->ce_flags & CE_ADDED)
506 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
508 /* Ok, create the new entry using the name of the existing alias */
509 len = ce_namelen(alias);
510 new = xcalloc(1, cache_entry_size(len));
511 memcpy(new->name, alias->name, len);
512 copy_cache_entry(new, ce);
517 static void record_intent_to_add(struct cache_entry *ce)
519 unsigned char sha1[20];
520 if (write_sha1_file("", 0, blob_type, sha1))
521 die("cannot create an empty blob in the object database");
522 hashcpy(ce->sha1, sha1);
525 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
527 int size, namelen, was_same;
528 mode_t st_mode = st->st_mode;
529 struct cache_entry *ce, *alias;
530 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
531 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
532 int pretend = flags & ADD_CACHE_PRETEND;
533 int intent_only = flags & ADD_CACHE_INTENT;
534 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
535 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
537 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
538 return error("%s: can only add regular files, symbolic links or git-directories", path);
540 namelen = strlen(path);
541 if (S_ISDIR(st_mode)) {
542 while (namelen && path[namelen-1] == '/')
545 size = cache_entry_size(namelen);
546 ce = xcalloc(1, size);
547 memcpy(ce->name, path, namelen);
548 ce->ce_flags = namelen;
550 fill_stat_cache_info(ce, st);
552 ce->ce_flags |= CE_INTENT_TO_ADD;
554 if (trust_executable_bit && has_symlinks)
555 ce->ce_mode = create_ce_mode(st_mode);
557 /* If there is an existing entry, pick the mode bits and type
558 * from it, otherwise assume unexecutable regular file.
560 struct cache_entry *ent;
561 int pos = index_name_pos_also_unmerged(istate, path, namelen);
563 ent = (0 <= pos) ? istate->cache[pos] : NULL;
564 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
567 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
568 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
569 /* Nothing changed, really */
571 ce_mark_uptodate(alias);
572 alias->ce_flags |= CE_ADDED;
576 if (index_path(ce->sha1, path, st, 1))
577 return error("unable to index file %s", path);
579 record_intent_to_add(ce);
581 if (ignore_case && alias && different_name(ce, alias))
582 ce = create_alias_ce(ce, alias);
583 ce->ce_flags |= CE_ADDED;
585 /* It was suspected to be racily clean, but it turns out to be Ok */
588 !hashcmp(alias->sha1, ce->sha1) &&
589 ce->ce_mode == alias->ce_mode);
593 else if (add_index_entry(istate, ce, add_option))
594 return error("unable to add %s to index",path);
595 if (verbose && !was_same)
596 printf("add '%s'\n", path);
600 int add_file_to_index(struct index_state *istate, const char *path, int flags)
603 if (lstat(path, &st))
604 die("%s: unable to stat (%s)", path, strerror(errno));
605 return add_to_index(istate, path, &st, flags);
608 struct cache_entry *make_cache_entry(unsigned int mode,
609 const unsigned char *sha1, const char *path, int stage,
613 struct cache_entry *ce;
615 if (!verify_path(path)) {
616 error("Invalid path '%s'", path);
621 size = cache_entry_size(len);
622 ce = xcalloc(1, size);
624 hashcpy(ce->sha1, sha1);
625 memcpy(ce->name, path, len);
626 ce->ce_flags = create_ce_flags(len, stage);
627 ce->ce_mode = create_ce_mode(mode);
630 return refresh_cache_entry(ce, 0);
635 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
637 int len = ce_namelen(a);
638 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
641 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
643 const char *match, *name;
649 len = ce_namelen(ce);
651 while ((match = *pathspec++) != NULL) {
652 int matchlen = strlen(match);
655 if (memcmp(name, match, matchlen))
657 if (matchlen && name[matchlen-1] == '/')
659 if (name[matchlen] == '/' || !name[matchlen])
668 * We fundamentally don't like some paths: we don't want
669 * dot or dot-dot anywhere, and for obvious reasons don't
670 * want to recurse into ".git" either.
672 * Also, we don't want double slashes or slashes at the
673 * end that can make pathnames ambiguous.
675 static int verify_dotfile(const char *rest)
678 * The first character was '.', but that
679 * has already been discarded, we now test
683 /* "." is not allowed */
688 * ".git" followed by NUL or slash is bad. This
689 * shares the path end test with the ".." case.
699 if (rest[1] == '\0' || rest[1] == '/')
705 int verify_path(const char *path)
722 if (verify_dotfile(path))
732 * Do we have another file that has the beginning components being a
733 * proper superset of the name we're trying to add?
735 static int has_file_name(struct index_state *istate,
736 const struct cache_entry *ce, int pos, int ok_to_replace)
739 int len = ce_namelen(ce);
740 int stage = ce_stage(ce);
741 const char *name = ce->name;
743 while (pos < istate->cache_nr) {
744 struct cache_entry *p = istate->cache[pos++];
746 if (len >= ce_namelen(p))
748 if (memcmp(name, p->name, len))
750 if (ce_stage(p) != stage)
752 if (p->name[len] != '/')
754 if (p->ce_flags & CE_REMOVE)
759 remove_index_entry_at(istate, --pos);
765 * Do we have another file with a pathname that is a proper
766 * subset of the name we're trying to add?
768 static int has_dir_name(struct index_state *istate,
769 const struct cache_entry *ce, int pos, int ok_to_replace)
772 int stage = ce_stage(ce);
773 const char *name = ce->name;
774 const char *slash = name + ce_namelen(ce);
782 if (slash <= ce->name)
787 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
790 * Found one, but not so fast. This could
791 * be a marker that says "I was here, but
792 * I am being removed". Such an entry is
793 * not a part of the resulting tree, and
794 * it is Ok to have a directory at the same
797 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
801 remove_index_entry_at(istate, pos);
809 * Trivial optimization: if we find an entry that
810 * already matches the sub-directory, then we know
811 * we're ok, and we can exit.
813 while (pos < istate->cache_nr) {
814 struct cache_entry *p = istate->cache[pos];
815 if ((ce_namelen(p) <= len) ||
816 (p->name[len] != '/') ||
817 memcmp(p->name, name, len))
818 break; /* not our subdirectory */
819 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
821 * p is at the same stage as our entry, and
822 * is a subdirectory of what we are looking
823 * at, so we cannot have conflicts at our
824 * level or anything shorter.
833 /* We may be in a situation where we already have path/file and path
834 * is being added, or we already have path and path/file is being
835 * added. Either one would result in a nonsense tree that has path
836 * twice when git-write-tree tries to write it out. Prevent it.
838 * If ok-to-replace is specified, we remove the conflicting entries
839 * from the cache so the caller should recompute the insert position.
840 * When this happens, we return non-zero.
842 static int check_file_directory_conflict(struct index_state *istate,
843 const struct cache_entry *ce,
844 int pos, int ok_to_replace)
849 * When ce is an "I am going away" entry, we allow it to be added
851 if (ce->ce_flags & CE_REMOVE)
855 * We check if the path is a sub-path of a subsequent pathname
856 * first, since removing those will not change the position
859 retval = has_file_name(istate, ce, pos, ok_to_replace);
862 * Then check if the path might have a clashing sub-directory
865 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
868 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
871 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
872 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
873 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
874 int new_only = option & ADD_CACHE_NEW_ONLY;
876 cache_tree_invalidate_path(istate->cache_tree, ce->name);
877 pos = index_name_pos(istate, ce->name, ce->ce_flags);
879 /* existing match? Just replace it. */
882 replace_index_entry(istate, pos, ce);
888 * Inserting a merged entry ("stage 0") into the index
889 * will always replace all non-merged entries..
891 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
892 while (ce_same_name(istate->cache[pos], ce)) {
894 if (!remove_index_entry_at(istate, pos))
901 if (!verify_path(ce->name))
902 return error("Invalid path '%s'", ce->name);
904 if (!skip_df_check &&
905 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
907 return error("'%s' appears as both a file and as a directory",
909 pos = index_name_pos(istate, ce->name, ce->ce_flags);
915 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
919 if (option & ADD_CACHE_JUST_APPEND)
920 pos = istate->cache_nr;
923 ret = add_index_entry_with_check(istate, ce, option);
929 /* Make sure the array is big enough .. */
930 if (istate->cache_nr == istate->cache_alloc) {
931 istate->cache_alloc = alloc_nr(istate->cache_alloc);
932 istate->cache = xrealloc(istate->cache,
933 istate->cache_alloc * sizeof(struct cache_entry *));
938 if (istate->cache_nr > pos + 1)
939 memmove(istate->cache + pos + 1,
941 (istate->cache_nr - pos - 1) * sizeof(ce));
942 set_index_entry(istate, pos, ce);
943 istate->cache_changed = 1;
948 * "refresh" does not calculate a new sha1 file or bring the
949 * cache up-to-date for mode/content changes. But what it
950 * _does_ do is to "re-match" the stat information of a file
951 * with the cache, so that you can refresh the cache for a
952 * file that hasn't been changed but where the stat entry is
955 * For example, you'd want to do this after doing a "git-read-tree",
956 * to link up the stat cache details with the proper files.
958 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
959 struct cache_entry *ce,
960 unsigned int options, int *err)
963 struct cache_entry *updated;
965 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
971 * CE_VALID means the user promised us that the change to
972 * the work tree does not matter and told us not to worry.
974 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
975 ce_mark_uptodate(ce);
979 if (lstat(ce->name, &st) < 0) {
985 changed = ie_match_stat(istate, ce, &st, options);
988 * The path is unchanged. If we were told to ignore
989 * valid bit, then we did the actual stat check and
990 * found that the entry is unmodified. If the entry
991 * is not marked VALID, this is the place to mark it
992 * valid again, under "assume unchanged" mode.
994 if (ignore_valid && assume_unchanged &&
995 !(ce->ce_flags & CE_VALID))
996 ; /* mark this one VALID again */
999 * We do not mark the index itself "modified"
1000 * because CE_UPTODATE flag is in-core only;
1001 * we are not going to write this change out.
1003 ce_mark_uptodate(ce);
1008 if (ie_modified(istate, ce, &st, options)) {
1015 updated = xmalloc(size);
1016 memcpy(updated, ce, size);
1017 fill_stat_cache_info(updated, &st);
1019 * If ignore_valid is not set, we should leave CE_VALID bit
1020 * alone. Otherwise, paths marked with --no-assume-unchanged
1021 * (i.e. things to be edited) will reacquire CE_VALID bit
1022 * automatically, which is not really what we want.
1024 if (!ignore_valid && assume_unchanged &&
1025 !(ce->ce_flags & CE_VALID))
1026 updated->ce_flags &= ~CE_VALID;
1031 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
1035 int really = (flags & REFRESH_REALLY) != 0;
1036 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1037 int quiet = (flags & REFRESH_QUIET) != 0;
1038 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1039 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1040 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1041 const char *needs_update_message;
1043 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
1044 ? "locally modified" : "needs update");
1045 for (i = 0; i < istate->cache_nr; i++) {
1046 struct cache_entry *ce, *new;
1047 int cache_errno = 0;
1049 ce = istate->cache[i];
1050 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1054 while ((i < istate->cache_nr) &&
1055 ! strcmp(istate->cache[i]->name, ce->name))
1060 printf("%s: needs merge\n", ce->name);
1065 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1068 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1072 if (not_new && cache_errno == ENOENT)
1074 if (really && cache_errno == EINVAL) {
1075 /* If we are doing --really-refresh that
1076 * means the index is not valid anymore.
1078 ce->ce_flags &= ~CE_VALID;
1079 istate->cache_changed = 1;
1083 printf("%s: %s\n", ce->name, needs_update_message);
1088 replace_index_entry(istate, i, new);
1093 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1095 return refresh_cache_ent(&the_index, ce, really, NULL);
1098 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1101 unsigned char sha1[20];
1103 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1104 return error("bad signature");
1105 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1106 return error("bad index version");
1108 git_SHA1_Update(&c, hdr, size - 20);
1109 git_SHA1_Final(sha1, &c);
1110 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1111 return error("bad index file sha1 signature");
1115 static int read_index_extension(struct index_state *istate,
1116 const char *ext, void *data, unsigned long sz)
1118 switch (CACHE_EXT(ext)) {
1119 case CACHE_EXT_TREE:
1120 istate->cache_tree = cache_tree_read(data, sz);
1123 if (*ext < 'A' || 'Z' < *ext)
1124 return error("index uses %.4s extension, which we do not understand",
1126 fprintf(stderr, "ignoring %.4s extension\n", ext);
1132 int read_index(struct index_state *istate)
1134 return read_index_from(istate, get_index_file());
1137 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1142 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1143 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1144 ce->ce_dev = ntohl(ondisk->dev);
1145 ce->ce_ino = ntohl(ondisk->ino);
1146 ce->ce_mode = ntohl(ondisk->mode);
1147 ce->ce_uid = ntohl(ondisk->uid);
1148 ce->ce_gid = ntohl(ondisk->gid);
1149 ce->ce_size = ntohl(ondisk->size);
1150 /* On-disk flags are just 16 bits */
1151 ce->ce_flags = ntohs(ondisk->flags);
1153 hashcpy(ce->sha1, ondisk->sha1);
1155 len = ce->ce_flags & CE_NAMEMASK;
1157 if (ce->ce_flags & CE_EXTENDED) {
1158 struct ondisk_cache_entry_extended *ondisk2;
1160 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1161 extended_flags = ntohs(ondisk2->flags2) << 16;
1162 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1163 if (extended_flags & ~CE_EXTENDED_FLAGS)
1164 die("Unknown index entry format %08x", extended_flags);
1165 ce->ce_flags |= extended_flags;
1166 name = ondisk2->name;
1169 name = ondisk->name;
1171 if (len == CE_NAMEMASK)
1174 * NEEDSWORK: If the original index is crafted, this copy could
1177 memcpy(ce->name, name, len + 1);
1180 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1184 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1187 * Alignment can cause differences. This should be "alignof", but
1188 * since that's a gcc'ism, just use the size of a pointer.
1190 per_entry += sizeof(void *);
1191 return ondisk_size + entries*per_entry;
1194 /* remember to discard_cache() before reading a different cache! */
1195 int read_index_from(struct index_state *istate, const char *path)
1199 unsigned long src_offset, dst_offset;
1200 struct cache_header *hdr;
1205 if (istate->initialized)
1206 return istate->cache_nr;
1209 istate->timestamp = 0;
1210 fd = open(path, O_RDONLY);
1212 if (errno == ENOENT)
1214 die("index file open failed (%s)", strerror(errno));
1218 die("cannot stat the open index (%s)", strerror(errno));
1221 mmap_size = xsize_t(st.st_size);
1222 if (mmap_size < sizeof(struct cache_header) + 20)
1223 die("index file smaller than expected");
1225 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1227 if (mmap == MAP_FAILED)
1228 die("unable to map index file");
1231 if (verify_hdr(hdr, mmap_size) < 0)
1234 istate->cache_nr = ntohl(hdr->hdr_entries);
1235 istate->cache_alloc = alloc_nr(istate->cache_nr);
1236 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1239 * The disk format is actually larger than the in-memory format,
1240 * due to space for nsec etc, so even though the in-memory one
1241 * has room for a few more flags, we can allocate using the same
1244 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1245 istate->initialized = 1;
1247 src_offset = sizeof(*hdr);
1249 for (i = 0; i < istate->cache_nr; i++) {
1250 struct ondisk_cache_entry *disk_ce;
1251 struct cache_entry *ce;
1253 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1254 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1255 convert_from_disk(disk_ce, ce);
1256 set_index_entry(istate, i, ce);
1258 src_offset += ondisk_ce_size(ce);
1259 dst_offset += ce_size(ce);
1261 istate->timestamp = st.st_mtime;
1262 while (src_offset <= mmap_size - 20 - 8) {
1263 /* After an array of active_nr index entries,
1264 * there can be arbitrary number of extended
1265 * sections, each of which is prefixed with
1266 * extension name (4-byte) and section length
1267 * in 4-byte network byte order.
1269 unsigned long extsize;
1270 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1271 extsize = ntohl(extsize);
1272 if (read_index_extension(istate,
1273 (const char *) mmap + src_offset,
1274 (char *) mmap + src_offset + 8,
1278 src_offset += extsize;
1280 munmap(mmap, mmap_size);
1281 return istate->cache_nr;
1284 munmap(mmap, mmap_size);
1286 die("index file corrupt");
1289 int is_index_unborn(struct index_state *istate)
1291 return (!istate->cache_nr && !istate->alloc && !istate->timestamp);
1294 int discard_index(struct index_state *istate)
1296 istate->cache_nr = 0;
1297 istate->cache_changed = 0;
1298 istate->timestamp = 0;
1299 istate->name_hash_initialized = 0;
1300 free_hash(&istate->name_hash);
1301 cache_tree_free(&(istate->cache_tree));
1302 free(istate->alloc);
1303 istate->alloc = NULL;
1304 istate->initialized = 0;
1306 /* no need to throw away allocated active_cache */
1310 int unmerged_index(const struct index_state *istate)
1313 for (i = 0; i < istate->cache_nr; i++) {
1314 if (ce_stage(istate->cache[i]))
1320 #define WRITE_BUFFER_SIZE 8192
1321 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1322 static unsigned long write_buffer_len;
1324 static int ce_write_flush(git_SHA_CTX *context, int fd)
1326 unsigned int buffered = write_buffer_len;
1328 git_SHA1_Update(context, write_buffer, buffered);
1329 if (write_in_full(fd, write_buffer, buffered) != buffered)
1331 write_buffer_len = 0;
1336 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1339 unsigned int buffered = write_buffer_len;
1340 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1343 memcpy(write_buffer + buffered, data, partial);
1344 buffered += partial;
1345 if (buffered == WRITE_BUFFER_SIZE) {
1346 write_buffer_len = buffered;
1347 if (ce_write_flush(context, fd))
1351 write_buffer_len = buffered;
1353 data = (char *) data + partial;
1358 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1359 unsigned int ext, unsigned int sz)
1363 return ((ce_write(context, fd, &ext, 4) < 0) ||
1364 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1367 static int ce_flush(git_SHA_CTX *context, int fd)
1369 unsigned int left = write_buffer_len;
1372 write_buffer_len = 0;
1373 git_SHA1_Update(context, write_buffer, left);
1376 /* Flush first if not enough space for SHA1 signature */
1377 if (left + 20 > WRITE_BUFFER_SIZE) {
1378 if (write_in_full(fd, write_buffer, left) != left)
1383 /* Append the SHA1 signature at the end */
1384 git_SHA1_Final(write_buffer + left, context);
1386 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1389 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1392 * The only thing we care about in this function is to smudge the
1393 * falsely clean entry due to touch-update-touch race, so we leave
1394 * everything else as they are. We are called for entries whose
1395 * ce_mtime match the index file mtime.
1397 * Note that this actually does not do much for gitlinks, for
1398 * which ce_match_stat_basic() always goes to the actual
1399 * contents. The caller checks with is_racy_timestamp() which
1400 * always says "no" for gitlinks, so we are not called for them ;-)
1404 if (lstat(ce->name, &st) < 0)
1406 if (ce_match_stat_basic(ce, &st))
1408 if (ce_modified_check_fs(ce, &st)) {
1409 /* This is "racily clean"; smudge it. Note that this
1410 * is a tricky code. At first glance, it may appear
1411 * that it can break with this sequence:
1413 * $ echo xyzzy >frotz
1414 * $ git-update-index --add frotz
1417 * $ echo filfre >nitfol
1418 * $ git-update-index --add nitfol
1420 * but it does not. When the second update-index runs,
1421 * it notices that the entry "frotz" has the same timestamp
1422 * as index, and if we were to smudge it by resetting its
1423 * size to zero here, then the object name recorded
1424 * in index is the 6-byte file but the cached stat information
1425 * becomes zero --- which would then match what we would
1426 * obtain from the filesystem next time we stat("frotz").
1428 * However, the second update-index, before calling
1429 * this function, notices that the cached size is 6
1430 * bytes and what is on the filesystem is an empty
1431 * file, and never calls us, so the cached size information
1432 * for "frotz" stays 6 which does not match the filesystem.
1438 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1440 int size = ondisk_ce_size(ce);
1441 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1444 ondisk->ctime.sec = htonl(ce->ce_ctime);
1445 ondisk->ctime.nsec = 0;
1446 ondisk->mtime.sec = htonl(ce->ce_mtime);
1447 ondisk->mtime.nsec = 0;
1448 ondisk->dev = htonl(ce->ce_dev);
1449 ondisk->ino = htonl(ce->ce_ino);
1450 ondisk->mode = htonl(ce->ce_mode);
1451 ondisk->uid = htonl(ce->ce_uid);
1452 ondisk->gid = htonl(ce->ce_gid);
1453 ondisk->size = htonl(ce->ce_size);
1454 hashcpy(ondisk->sha1, ce->sha1);
1455 ondisk->flags = htons(ce->ce_flags);
1456 if (ce->ce_flags & CE_EXTENDED) {
1457 struct ondisk_cache_entry_extended *ondisk2;
1458 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1459 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1460 name = ondisk2->name;
1463 name = ondisk->name;
1464 memcpy(name, ce->name, ce_namelen(ce));
1466 return ce_write(c, fd, ondisk, size);
1469 int write_index(const struct index_state *istate, int newfd)
1472 struct cache_header hdr;
1473 int i, err, removed, extended;
1474 struct cache_entry **cache = istate->cache;
1475 int entries = istate->cache_nr;
1477 for (i = removed = extended = 0; i < entries; i++) {
1478 if (cache[i]->ce_flags & CE_REMOVE)
1481 /* reduce extended entries if possible */
1482 cache[i]->ce_flags &= ~CE_EXTENDED;
1483 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1485 cache[i]->ce_flags |= CE_EXTENDED;
1489 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1490 /* for extended format, increase version so older git won't try to read it */
1491 hdr.hdr_version = htonl(extended ? 3 : 2);
1492 hdr.hdr_entries = htonl(entries - removed);
1495 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1498 for (i = 0; i < entries; i++) {
1499 struct cache_entry *ce = cache[i];
1500 if (ce->ce_flags & CE_REMOVE)
1502 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1503 ce_smudge_racily_clean_entry(ce);
1504 if (ce_write_entry(&c, newfd, ce) < 0)
1508 /* Write extension data here */
1509 if (istate->cache_tree) {
1510 struct strbuf sb = STRBUF_INIT;
1512 cache_tree_write(&sb, istate->cache_tree);
1513 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1514 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1515 strbuf_release(&sb);
1519 return ce_flush(&c, newfd);
1523 * Read the index file that is potentially unmerged into given
1524 * index_state, dropping any unmerged entries. Returns true if
1525 * the index is unmerged. Callers who want to refuse to work
1526 * from an unmerged state can call this and check its return value,
1527 * instead of calling read_cache().
1529 int read_index_unmerged(struct index_state *istate)
1535 for (i = 0; i < istate->cache_nr; i++) {
1536 struct cache_entry *ce = istate->cache[i];
1537 struct cache_entry *new_ce;
1543 len = strlen(ce->name);
1544 size = cache_entry_size(len);
1545 new_ce = xcalloc(1, size);
1546 hashcpy(new_ce->sha1, ce->sha1);
1547 memcpy(new_ce->name, ce->name, len);
1548 new_ce->ce_flags = create_ce_flags(len, 0);
1549 new_ce->ce_mode = ce->ce_mode;
1550 if (add_index_entry(istate, new_ce, 0))
1551 return error("%s: cannot drop to stage #0",
1553 i = index_name_pos(istate, new_ce->name, len);
1558 struct update_callback_data
1564 static void update_callback(struct diff_queue_struct *q,
1565 struct diff_options *opt, void *cbdata)
1568 struct update_callback_data *data = cbdata;
1570 for (i = 0; i < q->nr; i++) {
1571 struct diff_filepair *p = q->queue[i];
1572 const char *path = p->one->path;
1573 switch (p->status) {
1575 die("unexpected diff status %c", p->status);
1576 case DIFF_STATUS_UNMERGED:
1577 case DIFF_STATUS_MODIFIED:
1578 case DIFF_STATUS_TYPE_CHANGED:
1579 if (add_file_to_index(&the_index, path, data->flags)) {
1580 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1581 die("updating files failed");
1585 case DIFF_STATUS_DELETED:
1586 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1588 if (!(data->flags & ADD_CACHE_PRETEND))
1589 remove_file_from_index(&the_index, path);
1590 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1591 printf("remove '%s'\n", path);
1597 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1599 struct update_callback_data data;
1600 struct rev_info rev;
1601 init_revisions(&rev, prefix);
1602 setup_revisions(0, NULL, &rev, NULL);
1603 rev.prune_data = pathspec;
1604 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1605 rev.diffopt.format_callback = update_callback;
1607 data.add_errors = 0;
1608 rev.diffopt.format_callback_data = &data;
1609 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1610 return !!data.add_errors;
1614 * Returns 1 if the path is an "other" path with respect to
1615 * the index; that is, the path is not mentioned in the index at all,
1616 * either as a file, a directory with some files in the index,
1617 * or as an unmerged entry.
1619 * We helpfully remove a trailing "/" from directories so that
1620 * the output of read_directory can be used as-is.
1622 int index_name_is_other(const struct index_state *istate, const char *name,
1626 if (namelen && name[namelen - 1] == '/')
1628 pos = index_name_pos(istate, name, namelen);
1630 return 0; /* exact match */
1632 if (pos < istate->cache_nr) {
1633 struct cache_entry *ce = istate->cache[pos];
1634 if (ce_namelen(ce) == namelen &&
1635 !memcmp(ce->name, name, namelen))
1636 return 0; /* Yup, this one exists unmerged */