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)
105 enum object_type type;
108 target = xmalloc(expected_size);
109 len = readlink(ce->name, target, expected_size);
110 if (len != expected_size) {
114 buffer = read_sha1_file(ce->sha1, &type, &size);
119 if (size == expected_size)
120 match = memcmp(buffer, target, size);
126 static int ce_compare_gitlink(struct cache_entry *ce)
128 unsigned char sha1[20];
131 * We don't actually require that the .git directory
132 * under GITLINK directory be a valid git directory. It
133 * might even be missing (in case nobody populated that
136 * If so, we consider it always to match.
138 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
140 return hashcmp(sha1, ce->sha1);
143 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
145 switch (st->st_mode & S_IFMT) {
147 if (ce_compare_data(ce, st))
151 if (ce_compare_link(ce, xsize_t(st->st_size)))
155 if (S_ISGITLINK(ce->ce_mode))
156 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
163 int is_empty_blob_sha1(const unsigned char *sha1)
165 static const unsigned char empty_blob_sha1[20] = {
166 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
167 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
170 return !hashcmp(sha1, empty_blob_sha1);
173 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
175 unsigned int changed = 0;
177 if (ce->ce_flags & CE_REMOVE)
178 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
180 switch (ce->ce_mode & S_IFMT) {
182 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
183 /* We consider only the owner x bit to be relevant for
186 if (trust_executable_bit &&
187 (0100 & (ce->ce_mode ^ st->st_mode)))
188 changed |= MODE_CHANGED;
191 if (!S_ISLNK(st->st_mode) &&
192 (has_symlinks || !S_ISREG(st->st_mode)))
193 changed |= TYPE_CHANGED;
196 /* We ignore most of the st_xxx fields for gitlinks */
197 if (!S_ISDIR(st->st_mode))
198 changed |= TYPE_CHANGED;
199 else if (ce_compare_gitlink(ce))
200 changed |= DATA_CHANGED;
203 die("internal error: ce_mode is %o", ce->ce_mode);
205 if (ce->ce_mtime != (unsigned int) st->st_mtime)
206 changed |= MTIME_CHANGED;
207 if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime)
208 changed |= CTIME_CHANGED;
210 if (ce->ce_uid != (unsigned int) st->st_uid ||
211 ce->ce_gid != (unsigned int) st->st_gid)
212 changed |= OWNER_CHANGED;
213 if (ce->ce_ino != (unsigned int) st->st_ino)
214 changed |= INODE_CHANGED;
218 * st_dev breaks on network filesystems where different
219 * clients will have different views of what "device"
220 * the filesystem is on
222 if (ce->ce_dev != (unsigned int) st->st_dev)
223 changed |= INODE_CHANGED;
226 if (ce->ce_size != (unsigned int) st->st_size)
227 changed |= DATA_CHANGED;
229 /* Racily smudged entry? */
231 if (!is_empty_blob_sha1(ce->sha1))
232 changed |= DATA_CHANGED;
238 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
240 return (!S_ISGITLINK(ce->ce_mode) &&
242 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
245 int ie_match_stat(const struct index_state *istate,
246 struct cache_entry *ce, struct stat *st,
247 unsigned int options)
249 unsigned int changed;
250 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
251 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
254 * If it's marked as always valid in the index, it's
255 * valid whatever the checked-out copy says.
257 if (!ignore_valid && (ce->ce_flags & CE_VALID))
260 changed = ce_match_stat_basic(ce, st);
263 * Within 1 second of this sequence:
264 * echo xyzzy >file && git-update-index --add file
265 * running this command:
267 * would give a falsely clean cache entry. The mtime and
268 * length match the cache, and other stat fields do not change.
270 * We could detect this at update-index time (the cache entry
271 * being registered/updated records the same time as "now")
272 * and delay the return from git-update-index, but that would
273 * effectively mean we can make at most one commit per second,
274 * which is not acceptable. Instead, we check cache entries
275 * whose mtime are the same as the index file timestamp more
276 * carefully than others.
278 if (!changed && is_racy_timestamp(istate, ce)) {
279 if (assume_racy_is_modified)
280 changed |= DATA_CHANGED;
282 changed |= ce_modified_check_fs(ce, st);
288 int ie_modified(const struct index_state *istate,
289 struct cache_entry *ce, struct stat *st, unsigned int options)
291 int changed, changed_fs;
293 changed = ie_match_stat(istate, ce, st, options);
297 * If the mode or type has changed, there's no point in trying
298 * to refresh the entry - it's not going to match
300 if (changed & (MODE_CHANGED | TYPE_CHANGED))
304 * Immediately after read-tree or update-index --cacheinfo,
305 * the length field is zero, as we have never even read the
306 * lstat(2) information once, and we cannot trust DATA_CHANGED
307 * returned by ie_match_stat() which in turn was returned by
308 * ce_match_stat_basic() to signal that the filesize of the
309 * blob changed. We have to actually go to the filesystem to
310 * see if the contents match, and if so, should answer "unchanged".
312 * The logic does not apply to gitlinks, as ce_match_stat_basic()
313 * already has checked the actual HEAD from the filesystem in the
314 * subproject. If ie_match_stat() already said it is different,
315 * then we know it is.
317 if ((changed & DATA_CHANGED) &&
318 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
321 changed_fs = ce_modified_check_fs(ce, st);
323 return changed | changed_fs;
327 int base_name_compare(const char *name1, int len1, int mode1,
328 const char *name2, int len2, int mode2)
330 unsigned char c1, c2;
331 int len = len1 < len2 ? len1 : len2;
334 cmp = memcmp(name1, name2, len);
339 if (!c1 && S_ISDIR(mode1))
341 if (!c2 && S_ISDIR(mode2))
343 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
347 * df_name_compare() is identical to base_name_compare(), except it
348 * compares conflicting directory/file entries as equal. Note that
349 * while a directory name compares as equal to a regular file, they
350 * then individually compare _differently_ to a filename that has
351 * a dot after the basename (because '\0' < '.' < '/').
353 * This is used by routines that want to traverse the git namespace
354 * but then handle conflicting entries together when possible.
356 int df_name_compare(const char *name1, int len1, int mode1,
357 const char *name2, int len2, int mode2)
359 int len = len1 < len2 ? len1 : len2, cmp;
360 unsigned char c1, c2;
362 cmp = memcmp(name1, name2, len);
365 /* Directories and files compare equal (same length, same name) */
369 if (!c1 && S_ISDIR(mode1))
372 if (!c2 && S_ISDIR(mode2))
374 if (c1 == '/' && !c2)
376 if (c2 == '/' && !c1)
381 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
383 int len1 = flags1 & CE_NAMEMASK;
384 int len2 = flags2 & CE_NAMEMASK;
385 int len = len1 < len2 ? len1 : len2;
388 cmp = memcmp(name1, name2, len);
397 flags1 &= CE_STAGEMASK;
398 flags2 &= CE_STAGEMASK;
407 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
412 last = istate->cache_nr;
413 while (last > first) {
414 int next = (last + first) >> 1;
415 struct cache_entry *ce = istate->cache[next];
416 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
428 /* Remove entry, return true if there are more entries to go.. */
429 int remove_index_entry_at(struct index_state *istate, int pos)
431 struct cache_entry *ce = istate->cache[pos];
433 remove_name_hash(ce);
434 istate->cache_changed = 1;
436 if (pos >= istate->cache_nr)
438 memmove(istate->cache + pos,
439 istate->cache + pos + 1,
440 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
444 int remove_file_from_index(struct index_state *istate, const char *path)
446 int pos = index_name_pos(istate, path, strlen(path));
449 cache_tree_invalidate_path(istate->cache_tree, path);
450 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
451 remove_index_entry_at(istate, pos);
455 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
457 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
460 static int index_name_pos_also_unmerged(struct index_state *istate,
461 const char *path, int namelen)
463 int pos = index_name_pos(istate, path, namelen);
464 struct cache_entry *ce;
469 /* maybe unmerged? */
471 if (pos >= istate->cache_nr ||
472 compare_name((ce = istate->cache[pos]), path, namelen))
475 /* order of preference: stage 2, 1, 3 */
476 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
477 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
478 !compare_name(ce, path, namelen))
483 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
485 int len = ce_namelen(ce);
486 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
490 * If we add a filename that aliases in the cache, we will use the
491 * name that we already have - but we don't want to update the same
492 * alias twice, because that implies that there were actually two
493 * different files with aliasing names!
495 * So we use the CE_ADDED flag to verify that the alias was an old
496 * one before we accept it as
498 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
501 struct cache_entry *new;
503 if (alias->ce_flags & CE_ADDED)
504 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
506 /* Ok, create the new entry using the name of the existing alias */
507 len = ce_namelen(alias);
508 new = xcalloc(1, cache_entry_size(len));
509 memcpy(new->name, alias->name, len);
510 copy_cache_entry(new, ce);
515 static void record_intent_to_add(struct cache_entry *ce)
517 unsigned char sha1[20];
518 if (write_sha1_file("", 0, blob_type, sha1))
519 die("cannot create an empty blob in the object database");
520 hashcpy(ce->sha1, sha1);
523 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
525 int size, namelen, was_same;
526 mode_t st_mode = st->st_mode;
527 struct cache_entry *ce, *alias;
528 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
529 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
530 int pretend = flags & ADD_CACHE_PRETEND;
531 int intent_only = flags & ADD_CACHE_INTENT;
532 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
533 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
535 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
536 return error("%s: can only add regular files, symbolic links or git-directories", path);
538 namelen = strlen(path);
539 if (S_ISDIR(st_mode)) {
540 while (namelen && path[namelen-1] == '/')
543 size = cache_entry_size(namelen);
544 ce = xcalloc(1, size);
545 memcpy(ce->name, path, namelen);
546 ce->ce_flags = namelen;
548 fill_stat_cache_info(ce, st);
550 if (trust_executable_bit && has_symlinks)
551 ce->ce_mode = create_ce_mode(st_mode);
553 /* If there is an existing entry, pick the mode bits and type
554 * from it, otherwise assume unexecutable regular file.
556 struct cache_entry *ent;
557 int pos = index_name_pos_also_unmerged(istate, path, namelen);
559 ent = (0 <= pos) ? istate->cache[pos] : NULL;
560 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
563 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
564 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
565 /* Nothing changed, really */
567 ce_mark_uptodate(alias);
568 alias->ce_flags |= CE_ADDED;
572 if (index_path(ce->sha1, path, st, 1))
573 return error("unable to index file %s", path);
575 record_intent_to_add(ce);
577 if (ignore_case && alias && different_name(ce, alias))
578 ce = create_alias_ce(ce, alias);
579 ce->ce_flags |= CE_ADDED;
581 /* It was suspected to be racily clean, but it turns out to be Ok */
584 !hashcmp(alias->sha1, ce->sha1) &&
585 ce->ce_mode == alias->ce_mode);
589 else if (add_index_entry(istate, ce, add_option))
590 return error("unable to add %s to index",path);
591 if (verbose && !was_same)
592 printf("add '%s'\n", path);
596 int add_file_to_index(struct index_state *istate, const char *path, int flags)
599 if (lstat(path, &st))
600 die("%s: unable to stat (%s)", path, strerror(errno));
601 return add_to_index(istate, path, &st, flags);
604 struct cache_entry *make_cache_entry(unsigned int mode,
605 const unsigned char *sha1, const char *path, int stage,
609 struct cache_entry *ce;
611 if (!verify_path(path)) {
612 error("Invalid path '%s'", path);
617 size = cache_entry_size(len);
618 ce = xcalloc(1, size);
620 hashcpy(ce->sha1, sha1);
621 memcpy(ce->name, path, len);
622 ce->ce_flags = create_ce_flags(len, stage);
623 ce->ce_mode = create_ce_mode(mode);
626 return refresh_cache_entry(ce, 0);
631 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
633 int len = ce_namelen(a);
634 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
637 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
639 const char *match, *name;
645 len = ce_namelen(ce);
647 while ((match = *pathspec++) != NULL) {
648 int matchlen = strlen(match);
651 if (memcmp(name, match, matchlen))
653 if (matchlen && name[matchlen-1] == '/')
655 if (name[matchlen] == '/' || !name[matchlen])
664 * We fundamentally don't like some paths: we don't want
665 * dot or dot-dot anywhere, and for obvious reasons don't
666 * want to recurse into ".git" either.
668 * Also, we don't want double slashes or slashes at the
669 * end that can make pathnames ambiguous.
671 static int verify_dotfile(const char *rest)
674 * The first character was '.', but that
675 * has already been discarded, we now test
679 /* "." is not allowed */
684 * ".git" followed by NUL or slash is bad. This
685 * shares the path end test with the ".." case.
695 if (rest[1] == '\0' || rest[1] == '/')
701 int verify_path(const char *path)
718 if (verify_dotfile(path))
728 * Do we have another file that has the beginning components being a
729 * proper superset of the name we're trying to add?
731 static int has_file_name(struct index_state *istate,
732 const struct cache_entry *ce, int pos, int ok_to_replace)
735 int len = ce_namelen(ce);
736 int stage = ce_stage(ce);
737 const char *name = ce->name;
739 while (pos < istate->cache_nr) {
740 struct cache_entry *p = istate->cache[pos++];
742 if (len >= ce_namelen(p))
744 if (memcmp(name, p->name, len))
746 if (ce_stage(p) != stage)
748 if (p->name[len] != '/')
750 if (p->ce_flags & CE_REMOVE)
755 remove_index_entry_at(istate, --pos);
761 * Do we have another file with a pathname that is a proper
762 * subset of the name we're trying to add?
764 static int has_dir_name(struct index_state *istate,
765 const struct cache_entry *ce, int pos, int ok_to_replace)
768 int stage = ce_stage(ce);
769 const char *name = ce->name;
770 const char *slash = name + ce_namelen(ce);
778 if (slash <= ce->name)
783 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
786 * Found one, but not so fast. This could
787 * be a marker that says "I was here, but
788 * I am being removed". Such an entry is
789 * not a part of the resulting tree, and
790 * it is Ok to have a directory at the same
793 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
797 remove_index_entry_at(istate, pos);
805 * Trivial optimization: if we find an entry that
806 * already matches the sub-directory, then we know
807 * we're ok, and we can exit.
809 while (pos < istate->cache_nr) {
810 struct cache_entry *p = istate->cache[pos];
811 if ((ce_namelen(p) <= len) ||
812 (p->name[len] != '/') ||
813 memcmp(p->name, name, len))
814 break; /* not our subdirectory */
815 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
817 * p is at the same stage as our entry, and
818 * is a subdirectory of what we are looking
819 * at, so we cannot have conflicts at our
820 * level or anything shorter.
829 /* We may be in a situation where we already have path/file and path
830 * is being added, or we already have path and path/file is being
831 * added. Either one would result in a nonsense tree that has path
832 * twice when git-write-tree tries to write it out. Prevent it.
834 * If ok-to-replace is specified, we remove the conflicting entries
835 * from the cache so the caller should recompute the insert position.
836 * When this happens, we return non-zero.
838 static int check_file_directory_conflict(struct index_state *istate,
839 const struct cache_entry *ce,
840 int pos, int ok_to_replace)
845 * When ce is an "I am going away" entry, we allow it to be added
847 if (ce->ce_flags & CE_REMOVE)
851 * We check if the path is a sub-path of a subsequent pathname
852 * first, since removing those will not change the position
855 retval = has_file_name(istate, ce, pos, ok_to_replace);
858 * Then check if the path might have a clashing sub-directory
861 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
864 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
867 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
868 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
869 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
870 int new_only = option & ADD_CACHE_NEW_ONLY;
872 cache_tree_invalidate_path(istate->cache_tree, ce->name);
873 pos = index_name_pos(istate, ce->name, ce->ce_flags);
875 /* existing match? Just replace it. */
878 replace_index_entry(istate, pos, ce);
884 * Inserting a merged entry ("stage 0") into the index
885 * will always replace all non-merged entries..
887 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
888 while (ce_same_name(istate->cache[pos], ce)) {
890 if (!remove_index_entry_at(istate, pos))
897 if (!verify_path(ce->name))
898 return error("Invalid path '%s'", ce->name);
900 if (!skip_df_check &&
901 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
903 return error("'%s' appears as both a file and as a directory",
905 pos = index_name_pos(istate, ce->name, ce->ce_flags);
911 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
915 if (option & ADD_CACHE_JUST_APPEND)
916 pos = istate->cache_nr;
919 ret = add_index_entry_with_check(istate, ce, option);
925 /* Make sure the array is big enough .. */
926 if (istate->cache_nr == istate->cache_alloc) {
927 istate->cache_alloc = alloc_nr(istate->cache_alloc);
928 istate->cache = xrealloc(istate->cache,
929 istate->cache_alloc * sizeof(struct cache_entry *));
934 if (istate->cache_nr > pos + 1)
935 memmove(istate->cache + pos + 1,
937 (istate->cache_nr - pos - 1) * sizeof(ce));
938 set_index_entry(istate, pos, ce);
939 istate->cache_changed = 1;
944 * "refresh" does not calculate a new sha1 file or bring the
945 * cache up-to-date for mode/content changes. But what it
946 * _does_ do is to "re-match" the stat information of a file
947 * with the cache, so that you can refresh the cache for a
948 * file that hasn't been changed but where the stat entry is
951 * For example, you'd want to do this after doing a "git-read-tree",
952 * to link up the stat cache details with the proper files.
954 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
955 struct cache_entry *ce,
956 unsigned int options, int *err)
959 struct cache_entry *updated;
961 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
967 * CE_VALID means the user promised us that the change to
968 * the work tree does not matter and told us not to worry.
970 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
971 ce_mark_uptodate(ce);
975 if (lstat(ce->name, &st) < 0) {
981 changed = ie_match_stat(istate, ce, &st, options);
984 * The path is unchanged. If we were told to ignore
985 * valid bit, then we did the actual stat check and
986 * found that the entry is unmodified. If the entry
987 * is not marked VALID, this is the place to mark it
988 * valid again, under "assume unchanged" mode.
990 if (ignore_valid && assume_unchanged &&
991 !(ce->ce_flags & CE_VALID))
992 ; /* mark this one VALID again */
995 * We do not mark the index itself "modified"
996 * because CE_UPTODATE flag is in-core only;
997 * we are not going to write this change out.
999 ce_mark_uptodate(ce);
1004 if (ie_modified(istate, ce, &st, options)) {
1011 updated = xmalloc(size);
1012 memcpy(updated, ce, size);
1013 fill_stat_cache_info(updated, &st);
1015 * If ignore_valid is not set, we should leave CE_VALID bit
1016 * alone. Otherwise, paths marked with --no-assume-unchanged
1017 * (i.e. things to be edited) will reacquire CE_VALID bit
1018 * automatically, which is not really what we want.
1020 if (!ignore_valid && assume_unchanged &&
1021 !(ce->ce_flags & CE_VALID))
1022 updated->ce_flags &= ~CE_VALID;
1027 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
1031 int really = (flags & REFRESH_REALLY) != 0;
1032 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1033 int quiet = (flags & REFRESH_QUIET) != 0;
1034 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1035 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1036 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1037 const char *needs_update_message;
1039 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
1040 ? "locally modified" : "needs update");
1041 for (i = 0; i < istate->cache_nr; i++) {
1042 struct cache_entry *ce, *new;
1043 int cache_errno = 0;
1045 ce = istate->cache[i];
1046 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1050 while ((i < istate->cache_nr) &&
1051 ! strcmp(istate->cache[i]->name, ce->name))
1056 printf("%s: needs merge\n", ce->name);
1061 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1064 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1068 if (not_new && cache_errno == ENOENT)
1070 if (really && cache_errno == EINVAL) {
1071 /* If we are doing --really-refresh that
1072 * means the index is not valid anymore.
1074 ce->ce_flags &= ~CE_VALID;
1075 istate->cache_changed = 1;
1079 printf("%s: %s\n", ce->name, needs_update_message);
1084 replace_index_entry(istate, i, new);
1089 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1091 return refresh_cache_ent(&the_index, ce, really, NULL);
1094 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1097 unsigned char sha1[20];
1099 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1100 return error("bad signature");
1101 if (hdr->hdr_version != htonl(2))
1102 return error("bad index version");
1104 git_SHA1_Update(&c, hdr, size - 20);
1105 git_SHA1_Final(sha1, &c);
1106 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1107 return error("bad index file sha1 signature");
1111 static int read_index_extension(struct index_state *istate,
1112 const char *ext, void *data, unsigned long sz)
1114 switch (CACHE_EXT(ext)) {
1115 case CACHE_EXT_TREE:
1116 istate->cache_tree = cache_tree_read(data, sz);
1119 if (*ext < 'A' || 'Z' < *ext)
1120 return error("index uses %.4s extension, which we do not understand",
1122 fprintf(stderr, "ignoring %.4s extension\n", ext);
1128 int read_index(struct index_state *istate)
1130 return read_index_from(istate, get_index_file());
1133 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1137 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1138 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1139 ce->ce_dev = ntohl(ondisk->dev);
1140 ce->ce_ino = ntohl(ondisk->ino);
1141 ce->ce_mode = ntohl(ondisk->mode);
1142 ce->ce_uid = ntohl(ondisk->uid);
1143 ce->ce_gid = ntohl(ondisk->gid);
1144 ce->ce_size = ntohl(ondisk->size);
1145 /* On-disk flags are just 16 bits */
1146 ce->ce_flags = ntohs(ondisk->flags);
1148 /* For future extension: we do not understand this entry yet */
1149 if (ce->ce_flags & CE_EXTENDED)
1150 die("Unknown index entry format");
1151 hashcpy(ce->sha1, ondisk->sha1);
1153 len = ce->ce_flags & CE_NAMEMASK;
1154 if (len == CE_NAMEMASK)
1155 len = strlen(ondisk->name);
1157 * NEEDSWORK: If the original index is crafted, this copy could
1160 memcpy(ce->name, ondisk->name, len + 1);
1163 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1167 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1170 * Alignment can cause differences. This should be "alignof", but
1171 * since that's a gcc'ism, just use the size of a pointer.
1173 per_entry += sizeof(void *);
1174 return ondisk_size + entries*per_entry;
1177 /* remember to discard_cache() before reading a different cache! */
1178 int read_index_from(struct index_state *istate, const char *path)
1182 unsigned long src_offset, dst_offset;
1183 struct cache_header *hdr;
1188 if (istate->initialized)
1189 return istate->cache_nr;
1192 istate->timestamp = 0;
1193 fd = open(path, O_RDONLY);
1195 if (errno == ENOENT)
1197 die("index file open failed (%s)", strerror(errno));
1201 die("cannot stat the open index (%s)", strerror(errno));
1204 mmap_size = xsize_t(st.st_size);
1205 if (mmap_size < sizeof(struct cache_header) + 20)
1206 die("index file smaller than expected");
1208 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1210 if (mmap == MAP_FAILED)
1211 die("unable to map index file");
1214 if (verify_hdr(hdr, mmap_size) < 0)
1217 istate->cache_nr = ntohl(hdr->hdr_entries);
1218 istate->cache_alloc = alloc_nr(istate->cache_nr);
1219 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1222 * The disk format is actually larger than the in-memory format,
1223 * due to space for nsec etc, so even though the in-memory one
1224 * has room for a few more flags, we can allocate using the same
1227 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1228 istate->initialized = 1;
1230 src_offset = sizeof(*hdr);
1232 for (i = 0; i < istate->cache_nr; i++) {
1233 struct ondisk_cache_entry *disk_ce;
1234 struct cache_entry *ce;
1236 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1237 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1238 convert_from_disk(disk_ce, ce);
1239 set_index_entry(istate, i, ce);
1241 src_offset += ondisk_ce_size(ce);
1242 dst_offset += ce_size(ce);
1244 istate->timestamp = st.st_mtime;
1245 while (src_offset <= mmap_size - 20 - 8) {
1246 /* After an array of active_nr index entries,
1247 * there can be arbitrary number of extended
1248 * sections, each of which is prefixed with
1249 * extension name (4-byte) and section length
1250 * in 4-byte network byte order.
1252 unsigned long extsize;
1253 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1254 extsize = ntohl(extsize);
1255 if (read_index_extension(istate,
1256 (const char *) mmap + src_offset,
1257 (char *) mmap + src_offset + 8,
1261 src_offset += extsize;
1263 munmap(mmap, mmap_size);
1264 return istate->cache_nr;
1267 munmap(mmap, mmap_size);
1269 die("index file corrupt");
1272 int is_index_unborn(struct index_state *istate)
1274 return (!istate->cache_nr && !istate->alloc && !istate->timestamp);
1277 int discard_index(struct index_state *istate)
1279 istate->cache_nr = 0;
1280 istate->cache_changed = 0;
1281 istate->timestamp = 0;
1282 istate->name_hash_initialized = 0;
1283 free_hash(&istate->name_hash);
1284 cache_tree_free(&(istate->cache_tree));
1285 free(istate->alloc);
1286 istate->alloc = NULL;
1287 istate->initialized = 0;
1289 /* no need to throw away allocated active_cache */
1293 int unmerged_index(const struct index_state *istate)
1296 for (i = 0; i < istate->cache_nr; i++) {
1297 if (ce_stage(istate->cache[i]))
1303 #define WRITE_BUFFER_SIZE 8192
1304 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1305 static unsigned long write_buffer_len;
1307 static int ce_write_flush(git_SHA_CTX *context, int fd)
1309 unsigned int buffered = write_buffer_len;
1311 git_SHA1_Update(context, write_buffer, buffered);
1312 if (write_in_full(fd, write_buffer, buffered) != buffered)
1314 write_buffer_len = 0;
1319 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1322 unsigned int buffered = write_buffer_len;
1323 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1326 memcpy(write_buffer + buffered, data, partial);
1327 buffered += partial;
1328 if (buffered == WRITE_BUFFER_SIZE) {
1329 write_buffer_len = buffered;
1330 if (ce_write_flush(context, fd))
1334 write_buffer_len = buffered;
1336 data = (char *) data + partial;
1341 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1342 unsigned int ext, unsigned int sz)
1346 return ((ce_write(context, fd, &ext, 4) < 0) ||
1347 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1350 static int ce_flush(git_SHA_CTX *context, int fd)
1352 unsigned int left = write_buffer_len;
1355 write_buffer_len = 0;
1356 git_SHA1_Update(context, write_buffer, left);
1359 /* Flush first if not enough space for SHA1 signature */
1360 if (left + 20 > WRITE_BUFFER_SIZE) {
1361 if (write_in_full(fd, write_buffer, left) != left)
1366 /* Append the SHA1 signature at the end */
1367 git_SHA1_Final(write_buffer + left, context);
1369 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1372 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1375 * The only thing we care about in this function is to smudge the
1376 * falsely clean entry due to touch-update-touch race, so we leave
1377 * everything else as they are. We are called for entries whose
1378 * ce_mtime match the index file mtime.
1380 * Note that this actually does not do much for gitlinks, for
1381 * which ce_match_stat_basic() always goes to the actual
1382 * contents. The caller checks with is_racy_timestamp() which
1383 * always says "no" for gitlinks, so we are not called for them ;-)
1387 if (lstat(ce->name, &st) < 0)
1389 if (ce_match_stat_basic(ce, &st))
1391 if (ce_modified_check_fs(ce, &st)) {
1392 /* This is "racily clean"; smudge it. Note that this
1393 * is a tricky code. At first glance, it may appear
1394 * that it can break with this sequence:
1396 * $ echo xyzzy >frotz
1397 * $ git-update-index --add frotz
1400 * $ echo filfre >nitfol
1401 * $ git-update-index --add nitfol
1403 * but it does not. When the second update-index runs,
1404 * it notices that the entry "frotz" has the same timestamp
1405 * as index, and if we were to smudge it by resetting its
1406 * size to zero here, then the object name recorded
1407 * in index is the 6-byte file but the cached stat information
1408 * becomes zero --- which would then match what we would
1409 * obtain from the filesystem next time we stat("frotz").
1411 * However, the second update-index, before calling
1412 * this function, notices that the cached size is 6
1413 * bytes and what is on the filesystem is an empty
1414 * file, and never calls us, so the cached size information
1415 * for "frotz" stays 6 which does not match the filesystem.
1421 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1423 int size = ondisk_ce_size(ce);
1424 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1426 ondisk->ctime.sec = htonl(ce->ce_ctime);
1427 ondisk->ctime.nsec = 0;
1428 ondisk->mtime.sec = htonl(ce->ce_mtime);
1429 ondisk->mtime.nsec = 0;
1430 ondisk->dev = htonl(ce->ce_dev);
1431 ondisk->ino = htonl(ce->ce_ino);
1432 ondisk->mode = htonl(ce->ce_mode);
1433 ondisk->uid = htonl(ce->ce_uid);
1434 ondisk->gid = htonl(ce->ce_gid);
1435 ondisk->size = htonl(ce->ce_size);
1436 hashcpy(ondisk->sha1, ce->sha1);
1437 ondisk->flags = htons(ce->ce_flags);
1438 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1440 return ce_write(c, fd, ondisk, size);
1443 int write_index(const struct index_state *istate, int newfd)
1446 struct cache_header hdr;
1447 int i, err, removed;
1448 struct cache_entry **cache = istate->cache;
1449 int entries = istate->cache_nr;
1451 for (i = removed = 0; i < entries; i++)
1452 if (cache[i]->ce_flags & CE_REMOVE)
1455 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1456 hdr.hdr_version = htonl(2);
1457 hdr.hdr_entries = htonl(entries - removed);
1460 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1463 for (i = 0; i < entries; i++) {
1464 struct cache_entry *ce = cache[i];
1465 if (ce->ce_flags & CE_REMOVE)
1467 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1468 ce_smudge_racily_clean_entry(ce);
1469 if (ce_write_entry(&c, newfd, ce) < 0)
1473 /* Write extension data here */
1474 if (istate->cache_tree) {
1475 struct strbuf sb = STRBUF_INIT;
1477 cache_tree_write(&sb, istate->cache_tree);
1478 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1479 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1480 strbuf_release(&sb);
1484 return ce_flush(&c, newfd);
1488 * Read the index file that is potentially unmerged into given
1489 * index_state, dropping any unmerged entries. Returns true is
1490 * the index is unmerged. Callers who want to refuse to work
1491 * from an unmerged state can call this and check its return value,
1492 * instead of calling read_cache().
1494 int read_index_unmerged(struct index_state *istate)
1500 for (i = 0; i < istate->cache_nr; i++) {
1501 struct cache_entry *ce = istate->cache[i];
1502 struct cache_entry *new_ce;
1508 len = strlen(ce->name);
1509 size = cache_entry_size(len);
1510 new_ce = xcalloc(1, size);
1511 hashcpy(new_ce->sha1, ce->sha1);
1512 memcpy(new_ce->name, ce->name, len);
1513 new_ce->ce_flags = create_ce_flags(len, 0);
1514 new_ce->ce_mode = ce->ce_mode;
1515 if (add_index_entry(istate, new_ce, 0))
1516 return error("%s: cannot drop to stage #0",
1518 i = index_name_pos(istate, new_ce->name, len);
1523 struct update_callback_data
1529 static void update_callback(struct diff_queue_struct *q,
1530 struct diff_options *opt, void *cbdata)
1533 struct update_callback_data *data = cbdata;
1535 for (i = 0; i < q->nr; i++) {
1536 struct diff_filepair *p = q->queue[i];
1537 const char *path = p->one->path;
1538 switch (p->status) {
1540 die("unexpected diff status %c", p->status);
1541 case DIFF_STATUS_UNMERGED:
1542 case DIFF_STATUS_MODIFIED:
1543 case DIFF_STATUS_TYPE_CHANGED:
1544 if (add_file_to_index(&the_index, path, data->flags)) {
1545 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1546 die("updating files failed");
1550 case DIFF_STATUS_DELETED:
1551 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1553 if (!(data->flags & ADD_CACHE_PRETEND))
1554 remove_file_from_index(&the_index, path);
1555 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1556 printf("remove '%s'\n", path);
1562 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1564 struct update_callback_data data;
1565 struct rev_info rev;
1566 init_revisions(&rev, prefix);
1567 setup_revisions(0, NULL, &rev, NULL);
1568 rev.prune_data = pathspec;
1569 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1570 rev.diffopt.format_callback = update_callback;
1572 data.add_errors = 0;
1573 rev.diffopt.format_callback_data = &data;
1574 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1575 return !!data.add_errors;
1579 * Returns 1 if the path is an "other" path with respect to
1580 * the index; that is, the path is not mentioned in the index at all,
1581 * either as a file, a directory with some files in the index,
1582 * or as an unmerged entry.
1584 * We helpfully remove a trailing "/" from directories so that
1585 * the output of read_directory can be used as-is.
1587 int index_name_is_other(const struct index_state *istate, const char *name,
1591 if (namelen && name[namelen - 1] == '/')
1593 pos = index_name_pos(istate, name, namelen);
1595 return 0; /* exact match */
1597 if (pos < istate->cache_nr) {
1598 struct cache_entry *ce = istate->cache[pos];
1599 if (ce_namelen(ce) == namelen &&
1600 !memcmp(ce->name, name, namelen))
1601 return 0; /* Yup, this one exists unmerged */