2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
10 #include "cache-tree.h"
16 #include "resolve-undo.h"
19 #include "split-index.h"
22 /* Mask for the name length in ce_flags in the on-disk index */
24 #define CE_NAMEMASK (0x0fff)
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
35 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
37 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
39 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
41 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
42 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
43 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
44 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
46 struct index_state the_index;
47 static const char *alternate_index_output;
49 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
51 istate->cache[nr] = ce;
52 add_name_hash(istate, ce);
55 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
57 struct cache_entry *old = istate->cache[nr];
59 replace_index_entry_in_base(istate, old, ce);
60 remove_name_hash(istate, old);
62 set_index_entry(istate, nr, ce);
63 ce->ce_flags |= CE_UPDATE_IN_BASE;
64 istate->cache_changed |= CE_ENTRY_CHANGED;
67 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
69 struct cache_entry *old = istate->cache[nr], *new;
70 int namelen = strlen(new_name);
72 new = xmalloc(cache_entry_size(namelen));
73 copy_cache_entry(new, old);
74 new->ce_flags &= ~CE_HASHED;
75 new->ce_namelen = namelen;
77 memcpy(new->name, new_name, namelen + 1);
79 cache_tree_invalidate_path(istate, old->name);
80 untracked_cache_remove_from_index(istate, old->name);
81 remove_index_entry_at(istate, nr);
82 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
85 void fill_stat_data(struct stat_data *sd, struct stat *st)
87 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
88 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
89 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
90 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
91 sd->sd_dev = st->st_dev;
92 sd->sd_ino = st->st_ino;
93 sd->sd_uid = st->st_uid;
94 sd->sd_gid = st->st_gid;
95 sd->sd_size = st->st_size;
98 int match_stat_data(const struct stat_data *sd, struct stat *st)
102 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
103 changed |= MTIME_CHANGED;
104 if (trust_ctime && check_stat &&
105 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
106 changed |= CTIME_CHANGED;
109 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
110 changed |= MTIME_CHANGED;
111 if (trust_ctime && check_stat &&
112 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
113 changed |= CTIME_CHANGED;
117 if (sd->sd_uid != (unsigned int) st->st_uid ||
118 sd->sd_gid != (unsigned int) st->st_gid)
119 changed |= OWNER_CHANGED;
120 if (sd->sd_ino != (unsigned int) st->st_ino)
121 changed |= INODE_CHANGED;
126 * st_dev breaks on network filesystems where different
127 * clients will have different views of what "device"
128 * the filesystem is on
130 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
131 changed |= INODE_CHANGED;
134 if (sd->sd_size != (unsigned int) st->st_size)
135 changed |= DATA_CHANGED;
141 * This only updates the "non-critical" parts of the directory
142 * cache, ie the parts that aren't tracked by GIT, and only used
143 * to validate the cache.
145 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
147 fill_stat_data(&ce->ce_stat_data, st);
149 if (assume_unchanged)
150 ce->ce_flags |= CE_VALID;
152 if (S_ISREG(st->st_mode))
153 ce_mark_uptodate(ce);
156 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
159 static int cloexec = O_CLOEXEC;
160 int fd = open(ce->name, O_RDONLY | cloexec);
162 if ((cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
163 /* Try again w/o O_CLOEXEC: the kernel might not support it */
164 cloexec &= ~O_CLOEXEC;
165 fd = open(ce->name, O_RDONLY | cloexec);
169 unsigned char sha1[20];
170 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
171 match = hashcmp(sha1, ce->oid.hash);
172 /* index_fd() closed the file descriptor already */
177 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
182 enum object_type type;
183 struct strbuf sb = STRBUF_INIT;
185 if (strbuf_readlink(&sb, ce->name, expected_size))
188 buffer = read_sha1_file(ce->oid.hash, &type, &size);
191 match = memcmp(buffer, sb.buf, size);
198 static int ce_compare_gitlink(const struct cache_entry *ce)
200 unsigned char sha1[20];
203 * We don't actually require that the .git directory
204 * under GITLINK directory be a valid git directory. It
205 * might even be missing (in case nobody populated that
208 * If so, we consider it always to match.
210 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
212 return hashcmp(sha1, ce->oid.hash);
215 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
217 switch (st->st_mode & S_IFMT) {
219 if (ce_compare_data(ce, st))
223 if (ce_compare_link(ce, xsize_t(st->st_size)))
227 if (S_ISGITLINK(ce->ce_mode))
228 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
235 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
237 unsigned int changed = 0;
239 if (ce->ce_flags & CE_REMOVE)
240 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
242 switch (ce->ce_mode & S_IFMT) {
244 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
245 /* We consider only the owner x bit to be relevant for
248 if (trust_executable_bit &&
249 (0100 & (ce->ce_mode ^ st->st_mode)))
250 changed |= MODE_CHANGED;
253 if (!S_ISLNK(st->st_mode) &&
254 (has_symlinks || !S_ISREG(st->st_mode)))
255 changed |= TYPE_CHANGED;
258 /* We ignore most of the st_xxx fields for gitlinks */
259 if (!S_ISDIR(st->st_mode))
260 changed |= TYPE_CHANGED;
261 else if (ce_compare_gitlink(ce))
262 changed |= DATA_CHANGED;
265 die("internal error: ce_mode is %o", ce->ce_mode);
268 changed |= match_stat_data(&ce->ce_stat_data, st);
270 /* Racily smudged entry? */
271 if (!ce->ce_stat_data.sd_size) {
272 if (!is_empty_blob_sha1(ce->oid.hash))
273 changed |= DATA_CHANGED;
279 static int is_racy_stat(const struct index_state *istate,
280 const struct stat_data *sd)
282 return (istate->timestamp.sec &&
284 /* nanosecond timestamped files can also be racy! */
285 (istate->timestamp.sec < sd->sd_mtime.sec ||
286 (istate->timestamp.sec == sd->sd_mtime.sec &&
287 istate->timestamp.nsec <= sd->sd_mtime.nsec))
289 istate->timestamp.sec <= sd->sd_mtime.sec
294 static int is_racy_timestamp(const struct index_state *istate,
295 const struct cache_entry *ce)
297 return (!S_ISGITLINK(ce->ce_mode) &&
298 is_racy_stat(istate, &ce->ce_stat_data));
301 int match_stat_data_racy(const struct index_state *istate,
302 const struct stat_data *sd, struct stat *st)
304 if (is_racy_stat(istate, sd))
305 return MTIME_CHANGED;
306 return match_stat_data(sd, st);
309 int ie_match_stat(const struct index_state *istate,
310 const struct cache_entry *ce, struct stat *st,
311 unsigned int options)
313 unsigned int changed;
314 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
315 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
316 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
319 * If it's marked as always valid in the index, it's
320 * valid whatever the checked-out copy says.
322 * skip-worktree has the same effect with higher precedence
324 if (!ignore_skip_worktree && ce_skip_worktree(ce))
326 if (!ignore_valid && (ce->ce_flags & CE_VALID))
330 * Intent-to-add entries have not been added, so the index entry
331 * by definition never matches what is in the work tree until it
332 * actually gets added.
334 if (ce_intent_to_add(ce))
335 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
337 changed = ce_match_stat_basic(ce, st);
340 * Within 1 second of this sequence:
341 * echo xyzzy >file && git-update-index --add file
342 * running this command:
344 * would give a falsely clean cache entry. The mtime and
345 * length match the cache, and other stat fields do not change.
347 * We could detect this at update-index time (the cache entry
348 * being registered/updated records the same time as "now")
349 * and delay the return from git-update-index, but that would
350 * effectively mean we can make at most one commit per second,
351 * which is not acceptable. Instead, we check cache entries
352 * whose mtime are the same as the index file timestamp more
353 * carefully than others.
355 if (!changed && is_racy_timestamp(istate, ce)) {
356 if (assume_racy_is_modified)
357 changed |= DATA_CHANGED;
359 changed |= ce_modified_check_fs(ce, st);
365 int ie_modified(const struct index_state *istate,
366 const struct cache_entry *ce,
367 struct stat *st, unsigned int options)
369 int changed, changed_fs;
371 changed = ie_match_stat(istate, ce, st, options);
375 * If the mode or type has changed, there's no point in trying
376 * to refresh the entry - it's not going to match
378 if (changed & (MODE_CHANGED | TYPE_CHANGED))
382 * Immediately after read-tree or update-index --cacheinfo,
383 * the length field is zero, as we have never even read the
384 * lstat(2) information once, and we cannot trust DATA_CHANGED
385 * returned by ie_match_stat() which in turn was returned by
386 * ce_match_stat_basic() to signal that the filesize of the
387 * blob changed. We have to actually go to the filesystem to
388 * see if the contents match, and if so, should answer "unchanged".
390 * The logic does not apply to gitlinks, as ce_match_stat_basic()
391 * already has checked the actual HEAD from the filesystem in the
392 * subproject. If ie_match_stat() already said it is different,
393 * then we know it is.
395 if ((changed & DATA_CHANGED) &&
396 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
399 changed_fs = ce_modified_check_fs(ce, st);
401 return changed | changed_fs;
405 int base_name_compare(const char *name1, int len1, int mode1,
406 const char *name2, int len2, int mode2)
408 unsigned char c1, c2;
409 int len = len1 < len2 ? len1 : len2;
412 cmp = memcmp(name1, name2, len);
417 if (!c1 && S_ISDIR(mode1))
419 if (!c2 && S_ISDIR(mode2))
421 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
425 * df_name_compare() is identical to base_name_compare(), except it
426 * compares conflicting directory/file entries as equal. Note that
427 * while a directory name compares as equal to a regular file, they
428 * then individually compare _differently_ to a filename that has
429 * a dot after the basename (because '\0' < '.' < '/').
431 * This is used by routines that want to traverse the git namespace
432 * but then handle conflicting entries together when possible.
434 int df_name_compare(const char *name1, int len1, int mode1,
435 const char *name2, int len2, int mode2)
437 int len = len1 < len2 ? len1 : len2, cmp;
438 unsigned char c1, c2;
440 cmp = memcmp(name1, name2, len);
443 /* Directories and files compare equal (same length, same name) */
447 if (!c1 && S_ISDIR(mode1))
450 if (!c2 && S_ISDIR(mode2))
452 if (c1 == '/' && !c2)
454 if (c2 == '/' && !c1)
459 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
461 size_t min_len = (len1 < len2) ? len1 : len2;
462 int cmp = memcmp(name1, name2, min_len);
472 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
476 cmp = name_compare(name1, len1, name2, len2);
487 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
492 last = istate->cache_nr;
493 while (last > first) {
494 int next = (last + first) >> 1;
495 struct cache_entry *ce = istate->cache[next];
496 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
508 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
510 return index_name_stage_pos(istate, name, namelen, 0);
513 /* Remove entry, return true if there are more entries to go.. */
514 int remove_index_entry_at(struct index_state *istate, int pos)
516 struct cache_entry *ce = istate->cache[pos];
518 record_resolve_undo(istate, ce);
519 remove_name_hash(istate, ce);
520 save_or_free_index_entry(istate, ce);
521 istate->cache_changed |= CE_ENTRY_REMOVED;
523 if (pos >= istate->cache_nr)
525 memmove(istate->cache + pos,
526 istate->cache + pos + 1,
527 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
532 * Remove all cache entries marked for removal, that is where
533 * CE_REMOVE is set in ce_flags. This is much more effective than
534 * calling remove_index_entry_at() for each entry to be removed.
536 void remove_marked_cache_entries(struct index_state *istate)
538 struct cache_entry **ce_array = istate->cache;
541 for (i = j = 0; i < istate->cache_nr; i++) {
542 if (ce_array[i]->ce_flags & CE_REMOVE) {
543 remove_name_hash(istate, ce_array[i]);
544 save_or_free_index_entry(istate, ce_array[i]);
547 ce_array[j++] = ce_array[i];
549 if (j == istate->cache_nr)
551 istate->cache_changed |= CE_ENTRY_REMOVED;
552 istate->cache_nr = j;
555 int remove_file_from_index(struct index_state *istate, const char *path)
557 int pos = index_name_pos(istate, path, strlen(path));
560 cache_tree_invalidate_path(istate, path);
561 untracked_cache_remove_from_index(istate, path);
562 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
563 remove_index_entry_at(istate, pos);
567 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
569 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
572 static int index_name_pos_also_unmerged(struct index_state *istate,
573 const char *path, int namelen)
575 int pos = index_name_pos(istate, path, namelen);
576 struct cache_entry *ce;
581 /* maybe unmerged? */
583 if (pos >= istate->cache_nr ||
584 compare_name((ce = istate->cache[pos]), path, namelen))
587 /* order of preference: stage 2, 1, 3 */
588 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
589 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
590 !compare_name(ce, path, namelen))
595 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
597 int len = ce_namelen(ce);
598 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
602 * If we add a filename that aliases in the cache, we will use the
603 * name that we already have - but we don't want to update the same
604 * alias twice, because that implies that there were actually two
605 * different files with aliasing names!
607 * So we use the CE_ADDED flag to verify that the alias was an old
608 * one before we accept it as
610 static struct cache_entry *create_alias_ce(struct index_state *istate,
611 struct cache_entry *ce,
612 struct cache_entry *alias)
615 struct cache_entry *new;
617 if (alias->ce_flags & CE_ADDED)
618 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
620 /* Ok, create the new entry using the name of the existing alias */
621 len = ce_namelen(alias);
622 new = xcalloc(1, cache_entry_size(len));
623 memcpy(new->name, alias->name, len);
624 copy_cache_entry(new, ce);
625 save_or_free_index_entry(istate, ce);
629 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
631 unsigned char sha1[20];
632 if (write_sha1_file("", 0, blob_type, sha1))
633 die("cannot create an empty blob in the object database");
634 hashcpy(ce->oid.hash, sha1);
637 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
639 int size, namelen, was_same;
640 mode_t st_mode = st->st_mode;
641 struct cache_entry *ce, *alias;
642 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
643 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
644 int pretend = flags & ADD_CACHE_PRETEND;
645 int intent_only = flags & ADD_CACHE_INTENT;
646 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
647 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
649 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
650 return error("%s: can only add regular files, symbolic links or git-directories", path);
652 namelen = strlen(path);
653 if (S_ISDIR(st_mode)) {
654 while (namelen && path[namelen-1] == '/')
657 size = cache_entry_size(namelen);
658 ce = xcalloc(1, size);
659 memcpy(ce->name, path, namelen);
660 ce->ce_namelen = namelen;
662 fill_stat_cache_info(ce, st);
664 ce->ce_flags |= CE_INTENT_TO_ADD;
667 if (trust_executable_bit && has_symlinks) {
668 ce->ce_mode = create_ce_mode(st_mode);
670 /* If there is an existing entry, pick the mode bits and type
671 * from it, otherwise assume unexecutable regular file.
673 struct cache_entry *ent;
674 int pos = index_name_pos_also_unmerged(istate, path, namelen);
676 ent = (0 <= pos) ? istate->cache[pos] : NULL;
677 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
680 /* When core.ignorecase=true, determine if a directory of the same name but differing
681 * case already exists within the Git repository. If it does, ensure the directory
682 * case of the file being added to the repository matches (is folded into) the existing
683 * entry's directory case.
686 adjust_dirname_case(istate, ce->name);
689 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
690 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
691 /* Nothing changed, really */
692 if (!S_ISGITLINK(alias->ce_mode))
693 ce_mark_uptodate(alias);
694 alias->ce_flags |= CE_ADDED;
700 if (index_path(ce->oid.hash, path, st, HASH_WRITE_OBJECT)) {
702 return error("unable to index file %s", path);
705 set_object_name_for_intent_to_add_entry(ce);
707 if (ignore_case && alias && different_name(ce, alias))
708 ce = create_alias_ce(istate, ce, alias);
709 ce->ce_flags |= CE_ADDED;
711 /* It was suspected to be racily clean, but it turns out to be Ok */
714 !oidcmp(&alias->oid, &ce->oid) &&
715 ce->ce_mode == alias->ce_mode);
719 else if (add_index_entry(istate, ce, add_option)) {
721 return error("unable to add %s to index", path);
723 if (verbose && !was_same)
724 printf("add '%s'\n", path);
728 int add_file_to_index(struct index_state *istate, const char *path, int flags)
731 if (lstat(path, &st))
732 die_errno("unable to stat '%s'", path);
733 return add_to_index(istate, path, &st, flags);
736 struct cache_entry *make_cache_entry(unsigned int mode,
737 const unsigned char *sha1, const char *path, int stage,
738 unsigned int refresh_options)
741 struct cache_entry *ce, *ret;
743 if (!verify_path(path)) {
744 error("Invalid path '%s'", path);
749 size = cache_entry_size(len);
750 ce = xcalloc(1, size);
752 hashcpy(ce->oid.hash, sha1);
753 memcpy(ce->name, path, len);
754 ce->ce_flags = create_ce_flags(stage);
755 ce->ce_namelen = len;
756 ce->ce_mode = create_ce_mode(mode);
758 ret = refresh_cache_entry(ce, refresh_options);
765 * Chmod an index entry with either +x or -x.
767 * Returns -1 if the chmod for the particular cache entry failed (if it's
768 * not a regular file), -2 if an invalid flip argument is passed in, 0
771 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
774 if (!S_ISREG(ce->ce_mode))
781 ce->ce_mode &= ~0111;
786 cache_tree_invalidate_path(istate, ce->name);
787 ce->ce_flags |= CE_UPDATE_IN_BASE;
788 istate->cache_changed |= CE_ENTRY_CHANGED;
793 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
795 int len = ce_namelen(a);
796 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
800 * We fundamentally don't like some paths: we don't want
801 * dot or dot-dot anywhere, and for obvious reasons don't
802 * want to recurse into ".git" either.
804 * Also, we don't want double slashes or slashes at the
805 * end that can make pathnames ambiguous.
807 static int verify_dotfile(const char *rest)
810 * The first character was '.', but that
811 * has already been discarded, we now test
815 /* "." is not allowed */
816 if (*rest == '\0' || is_dir_sep(*rest))
821 * ".git" followed by NUL or slash is bad. This
822 * shares the path end test with the ".." case.
826 if (rest[1] != 'i' && rest[1] != 'I')
828 if (rest[2] != 't' && rest[2] != 'T')
833 if (rest[1] == '\0' || is_dir_sep(rest[1]))
839 int verify_path(const char *path)
843 if (has_dos_drive_prefix(path))
852 if (protect_hfs && is_hfs_dotgit(path))
854 if (protect_ntfs && is_ntfs_dotgit(path))
857 if ((c == '.' && !verify_dotfile(path)) ||
858 is_dir_sep(c) || c == '\0')
866 * Do we have another file that has the beginning components being a
867 * proper superset of the name we're trying to add?
869 static int has_file_name(struct index_state *istate,
870 const struct cache_entry *ce, int pos, int ok_to_replace)
873 int len = ce_namelen(ce);
874 int stage = ce_stage(ce);
875 const char *name = ce->name;
877 while (pos < istate->cache_nr) {
878 struct cache_entry *p = istate->cache[pos++];
880 if (len >= ce_namelen(p))
882 if (memcmp(name, p->name, len))
884 if (ce_stage(p) != stage)
886 if (p->name[len] != '/')
888 if (p->ce_flags & CE_REMOVE)
893 remove_index_entry_at(istate, --pos);
899 * Do we have another file with a pathname that is a proper
900 * subset of the name we're trying to add?
902 static int has_dir_name(struct index_state *istate,
903 const struct cache_entry *ce, int pos, int ok_to_replace)
906 int stage = ce_stage(ce);
907 const char *name = ce->name;
908 const char *slash = name + ce_namelen(ce);
916 if (slash <= ce->name)
921 pos = index_name_stage_pos(istate, name, len, stage);
924 * Found one, but not so fast. This could
925 * be a marker that says "I was here, but
926 * I am being removed". Such an entry is
927 * not a part of the resulting tree, and
928 * it is Ok to have a directory at the same
931 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
935 remove_index_entry_at(istate, pos);
943 * Trivial optimization: if we find an entry that
944 * already matches the sub-directory, then we know
945 * we're ok, and we can exit.
947 while (pos < istate->cache_nr) {
948 struct cache_entry *p = istate->cache[pos];
949 if ((ce_namelen(p) <= len) ||
950 (p->name[len] != '/') ||
951 memcmp(p->name, name, len))
952 break; /* not our subdirectory */
953 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
955 * p is at the same stage as our entry, and
956 * is a subdirectory of what we are looking
957 * at, so we cannot have conflicts at our
958 * level or anything shorter.
967 /* We may be in a situation where we already have path/file and path
968 * is being added, or we already have path and path/file is being
969 * added. Either one would result in a nonsense tree that has path
970 * twice when git-write-tree tries to write it out. Prevent it.
972 * If ok-to-replace is specified, we remove the conflicting entries
973 * from the cache so the caller should recompute the insert position.
974 * When this happens, we return non-zero.
976 static int check_file_directory_conflict(struct index_state *istate,
977 const struct cache_entry *ce,
978 int pos, int ok_to_replace)
983 * When ce is an "I am going away" entry, we allow it to be added
985 if (ce->ce_flags & CE_REMOVE)
989 * We check if the path is a sub-path of a subsequent pathname
990 * first, since removing those will not change the position
993 retval = has_file_name(istate, ce, pos, ok_to_replace);
996 * Then check if the path might have a clashing sub-directory
999 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1002 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1005 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1006 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1007 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1008 int new_only = option & ADD_CACHE_NEW_ONLY;
1010 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1011 cache_tree_invalidate_path(istate, ce->name);
1012 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1014 /* existing match? Just replace it. */
1017 replace_index_entry(istate, pos, ce);
1022 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1023 untracked_cache_add_to_index(istate, ce->name);
1026 * Inserting a merged entry ("stage 0") into the index
1027 * will always replace all non-merged entries..
1029 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1030 while (ce_same_name(istate->cache[pos], ce)) {
1032 if (!remove_index_entry_at(istate, pos))
1039 if (!verify_path(ce->name))
1040 return error("Invalid path '%s'", ce->name);
1042 if (!skip_df_check &&
1043 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1045 return error("'%s' appears as both a file and as a directory",
1047 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1053 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1057 if (option & ADD_CACHE_JUST_APPEND)
1058 pos = istate->cache_nr;
1061 ret = add_index_entry_with_check(istate, ce, option);
1067 /* Make sure the array is big enough .. */
1068 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1072 if (istate->cache_nr > pos + 1)
1073 memmove(istate->cache + pos + 1,
1074 istate->cache + pos,
1075 (istate->cache_nr - pos - 1) * sizeof(ce));
1076 set_index_entry(istate, pos, ce);
1077 istate->cache_changed |= CE_ENTRY_ADDED;
1082 * "refresh" does not calculate a new sha1 file or bring the
1083 * cache up-to-date for mode/content changes. But what it
1084 * _does_ do is to "re-match" the stat information of a file
1085 * with the cache, so that you can refresh the cache for a
1086 * file that hasn't been changed but where the stat entry is
1089 * For example, you'd want to do this after doing a "git-read-tree",
1090 * to link up the stat cache details with the proper files.
1092 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1093 struct cache_entry *ce,
1094 unsigned int options, int *err,
1098 struct cache_entry *updated;
1100 int refresh = options & CE_MATCH_REFRESH;
1101 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1102 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1103 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1105 if (!refresh || ce_uptodate(ce))
1109 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1110 * that the change to the work tree does not matter and told
1113 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1114 ce_mark_uptodate(ce);
1117 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1118 ce_mark_uptodate(ce);
1122 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1130 if (lstat(ce->name, &st) < 0) {
1131 if (ignore_missing && errno == ENOENT)
1138 changed = ie_match_stat(istate, ce, &st, options);
1140 *changed_ret = changed;
1143 * The path is unchanged. If we were told to ignore
1144 * valid bit, then we did the actual stat check and
1145 * found that the entry is unmodified. If the entry
1146 * is not marked VALID, this is the place to mark it
1147 * valid again, under "assume unchanged" mode.
1149 if (ignore_valid && assume_unchanged &&
1150 !(ce->ce_flags & CE_VALID))
1151 ; /* mark this one VALID again */
1154 * We do not mark the index itself "modified"
1155 * because CE_UPTODATE flag is in-core only;
1156 * we are not going to write this change out.
1158 if (!S_ISGITLINK(ce->ce_mode))
1159 ce_mark_uptodate(ce);
1164 if (ie_modified(istate, ce, &st, options)) {
1171 updated = xmalloc(size);
1172 memcpy(updated, ce, size);
1173 fill_stat_cache_info(updated, &st);
1175 * If ignore_valid is not set, we should leave CE_VALID bit
1176 * alone. Otherwise, paths marked with --no-assume-unchanged
1177 * (i.e. things to be edited) will reacquire CE_VALID bit
1178 * automatically, which is not really what we want.
1180 if (!ignore_valid && assume_unchanged &&
1181 !(ce->ce_flags & CE_VALID))
1182 updated->ce_flags &= ~CE_VALID;
1184 /* istate->cache_changed is updated in the caller */
1188 static void show_file(const char * fmt, const char * name, int in_porcelain,
1189 int * first, const char *header_msg)
1191 if (in_porcelain && *first && header_msg) {
1192 printf("%s\n", header_msg);
1198 int refresh_index(struct index_state *istate, unsigned int flags,
1199 const struct pathspec *pathspec,
1200 char *seen, const char *header_msg)
1204 int really = (flags & REFRESH_REALLY) != 0;
1205 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1206 int quiet = (flags & REFRESH_QUIET) != 0;
1207 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1208 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1210 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1211 unsigned int options = (CE_MATCH_REFRESH |
1212 (really ? CE_MATCH_IGNORE_VALID : 0) |
1213 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1214 const char *modified_fmt;
1215 const char *deleted_fmt;
1216 const char *typechange_fmt;
1217 const char *added_fmt;
1218 const char *unmerged_fmt;
1220 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1221 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1222 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1223 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1224 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1225 for (i = 0; i < istate->cache_nr; i++) {
1226 struct cache_entry *ce, *new;
1227 int cache_errno = 0;
1231 ce = istate->cache[i];
1232 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1235 if (pathspec && !ce_path_match(ce, pathspec, seen))
1239 while ((i < istate->cache_nr) &&
1240 ! strcmp(istate->cache[i]->name, ce->name))
1246 show_file(unmerged_fmt, ce->name, in_porcelain,
1247 &first, header_msg);
1255 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1261 if (really && cache_errno == EINVAL) {
1262 /* If we are doing --really-refresh that
1263 * means the index is not valid anymore.
1265 ce->ce_flags &= ~CE_VALID;
1266 ce->ce_flags |= CE_UPDATE_IN_BASE;
1267 istate->cache_changed |= CE_ENTRY_CHANGED;
1272 if (cache_errno == ENOENT)
1274 else if (ce_intent_to_add(ce))
1275 fmt = added_fmt; /* must be before other checks */
1276 else if (changed & TYPE_CHANGED)
1277 fmt = typechange_fmt;
1281 ce->name, in_porcelain, &first, header_msg);
1286 replace_index_entry(istate, i, new);
1291 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1292 unsigned int options)
1294 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1298 /*****************************************************************
1300 *****************************************************************/
1302 #define INDEX_FORMAT_DEFAULT 3
1304 static unsigned int get_index_format_default(void)
1306 char *envversion = getenv("GIT_INDEX_VERSION");
1309 unsigned int version = INDEX_FORMAT_DEFAULT;
1312 if (!git_config_get_int("index.version", &value))
1314 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1315 warning(_("index.version set, but the value is invalid.\n"
1316 "Using version %i"), INDEX_FORMAT_DEFAULT);
1317 return INDEX_FORMAT_DEFAULT;
1322 version = strtoul(envversion, &endp, 10);
1324 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1325 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1326 "Using version %i"), INDEX_FORMAT_DEFAULT);
1327 version = INDEX_FORMAT_DEFAULT;
1333 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1334 * Again - this is just a (very strong in practice) heuristic that
1335 * the inode hasn't changed.
1337 * We save the fields in big-endian order to allow using the
1338 * index file over NFS transparently.
1340 struct ondisk_cache_entry {
1341 struct cache_time ctime;
1342 struct cache_time mtime;
1349 unsigned char sha1[20];
1351 char name[FLEX_ARRAY]; /* more */
1355 * This struct is used when CE_EXTENDED bit is 1
1356 * The struct must match ondisk_cache_entry exactly from
1359 struct ondisk_cache_entry_extended {
1360 struct cache_time ctime;
1361 struct cache_time mtime;
1368 unsigned char sha1[20];
1371 char name[FLEX_ARRAY]; /* more */
1374 /* These are only used for v3 or lower */
1375 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1376 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1377 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1378 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1379 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1380 ondisk_cache_entry_size(ce_namelen(ce)))
1382 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1385 unsigned char sha1[20];
1388 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1389 return error("bad signature");
1390 hdr_version = ntohl(hdr->hdr_version);
1391 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1392 return error("bad index version %d", hdr_version);
1394 git_SHA1_Update(&c, hdr, size - 20);
1395 git_SHA1_Final(sha1, &c);
1396 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1397 return error("bad index file sha1 signature");
1401 static int read_index_extension(struct index_state *istate,
1402 const char *ext, void *data, unsigned long sz)
1404 switch (CACHE_EXT(ext)) {
1405 case CACHE_EXT_TREE:
1406 istate->cache_tree = cache_tree_read(data, sz);
1408 case CACHE_EXT_RESOLVE_UNDO:
1409 istate->resolve_undo = resolve_undo_read(data, sz);
1411 case CACHE_EXT_LINK:
1412 if (read_link_extension(istate, data, sz))
1415 case CACHE_EXT_UNTRACKED:
1416 istate->untracked = read_untracked_extension(data, sz);
1419 if (*ext < 'A' || 'Z' < *ext)
1420 return error("index uses %.4s extension, which we do not understand",
1422 fprintf(stderr, "ignoring %.4s extension\n", ext);
1428 int hold_locked_index(struct lock_file *lk, int lock_flags)
1430 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1433 int read_index(struct index_state *istate)
1435 return read_index_from(istate, get_index_file());
1438 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1443 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1445 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1446 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1447 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1448 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1449 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1450 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1451 ce->ce_mode = get_be32(&ondisk->mode);
1452 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1453 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1454 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1455 ce->ce_flags = flags & ~CE_NAMEMASK;
1456 ce->ce_namelen = len;
1458 hashcpy(ce->oid.hash, ondisk->sha1);
1459 memcpy(ce->name, name, len);
1460 ce->name[len] = '\0';
1465 * Adjacent cache entries tend to share the leading paths, so it makes
1466 * sense to only store the differences in later entries. In the v4
1467 * on-disk format of the index, each on-disk cache entry stores the
1468 * number of bytes to be stripped from the end of the previous name,
1469 * and the bytes to append to the result, to come up with its name.
1471 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1473 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1474 size_t len = decode_varint(&cp);
1476 if (name->len < len)
1477 die("malformed name field in the index");
1478 strbuf_remove(name, name->len - len, len);
1479 for (ep = cp; *ep; ep++)
1480 ; /* find the end */
1481 strbuf_add(name, cp, ep - cp);
1482 return (const char *)ep + 1 - cp_;
1485 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1486 unsigned long *ent_size,
1487 struct strbuf *previous_name)
1489 struct cache_entry *ce;
1494 /* On-disk flags are just 16 bits */
1495 flags = get_be16(&ondisk->flags);
1496 len = flags & CE_NAMEMASK;
1498 if (flags & CE_EXTENDED) {
1499 struct ondisk_cache_entry_extended *ondisk2;
1501 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1502 extended_flags = get_be16(&ondisk2->flags2) << 16;
1503 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1504 if (extended_flags & ~CE_EXTENDED_FLAGS)
1505 die("Unknown index entry format %08x", extended_flags);
1506 flags |= extended_flags;
1507 name = ondisk2->name;
1510 name = ondisk->name;
1512 if (!previous_name) {
1513 /* v3 and earlier */
1514 if (len == CE_NAMEMASK)
1516 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1518 *ent_size = ondisk_ce_size(ce);
1520 unsigned long consumed;
1521 consumed = expand_name_field(previous_name, name);
1522 ce = cache_entry_from_ondisk(ondisk, flags,
1524 previous_name->len);
1526 *ent_size = (name - ((char *)ondisk)) + consumed;
1531 static void check_ce_order(struct index_state *istate)
1535 for (i = 1; i < istate->cache_nr; i++) {
1536 struct cache_entry *ce = istate->cache[i - 1];
1537 struct cache_entry *next_ce = istate->cache[i];
1538 int name_compare = strcmp(ce->name, next_ce->name);
1540 if (0 < name_compare)
1541 die("unordered stage entries in index");
1542 if (!name_compare) {
1544 die("multiple stage entries for merged file '%s'",
1546 if (ce_stage(ce) > ce_stage(next_ce))
1547 die("unordered stage entries for '%s'",
1553 static void tweak_untracked_cache(struct index_state *istate)
1555 switch (git_config_get_untracked_cache()) {
1556 case -1: /* keep: do nothing */
1559 remove_untracked_cache(istate);
1562 add_untracked_cache(istate);
1564 default: /* unknown value: do nothing */
1569 static void post_read_index_from(struct index_state *istate)
1571 check_ce_order(istate);
1572 tweak_untracked_cache(istate);
1575 /* remember to discard_cache() before reading a different cache! */
1576 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1580 unsigned long src_offset;
1581 struct cache_header *hdr;
1584 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1586 if (istate->initialized)
1587 return istate->cache_nr;
1589 istate->timestamp.sec = 0;
1590 istate->timestamp.nsec = 0;
1591 fd = open(path, O_RDONLY);
1593 if (!must_exist && errno == ENOENT)
1595 die_errno("%s: index file open failed", path);
1599 die_errno("cannot stat the open index");
1601 mmap_size = xsize_t(st.st_size);
1602 if (mmap_size < sizeof(struct cache_header) + 20)
1603 die("index file smaller than expected");
1605 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1606 if (mmap == MAP_FAILED)
1607 die_errno("unable to map index file");
1611 if (verify_hdr(hdr, mmap_size) < 0)
1614 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1615 istate->version = ntohl(hdr->hdr_version);
1616 istate->cache_nr = ntohl(hdr->hdr_entries);
1617 istate->cache_alloc = alloc_nr(istate->cache_nr);
1618 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1619 istate->initialized = 1;
1621 if (istate->version == 4)
1622 previous_name = &previous_name_buf;
1624 previous_name = NULL;
1626 src_offset = sizeof(*hdr);
1627 for (i = 0; i < istate->cache_nr; i++) {
1628 struct ondisk_cache_entry *disk_ce;
1629 struct cache_entry *ce;
1630 unsigned long consumed;
1632 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1633 ce = create_from_disk(disk_ce, &consumed, previous_name);
1634 set_index_entry(istate, i, ce);
1636 src_offset += consumed;
1638 strbuf_release(&previous_name_buf);
1639 istate->timestamp.sec = st.st_mtime;
1640 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1642 while (src_offset <= mmap_size - 20 - 8) {
1643 /* After an array of active_nr index entries,
1644 * there can be arbitrary number of extended
1645 * sections, each of which is prefixed with
1646 * extension name (4-byte) and section length
1647 * in 4-byte network byte order.
1650 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1651 extsize = ntohl(extsize);
1652 if (read_index_extension(istate,
1653 (const char *) mmap + src_offset,
1654 (char *) mmap + src_offset + 8,
1658 src_offset += extsize;
1660 munmap(mmap, mmap_size);
1661 return istate->cache_nr;
1664 munmap(mmap, mmap_size);
1665 die("index file corrupt");
1668 int read_index_from(struct index_state *istate, const char *path)
1670 struct split_index *split_index;
1673 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1674 if (istate->initialized)
1675 return istate->cache_nr;
1677 ret = do_read_index(istate, path, 0);
1679 split_index = istate->split_index;
1680 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1681 post_read_index_from(istate);
1685 if (split_index->base)
1686 discard_index(split_index->base);
1688 split_index->base = xcalloc(1, sizeof(*split_index->base));
1689 ret = do_read_index(split_index->base,
1690 git_path("sharedindex.%s",
1691 sha1_to_hex(split_index->base_sha1)), 1);
1692 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1693 die("broken index, expect %s in %s, got %s",
1694 sha1_to_hex(split_index->base_sha1),
1695 git_path("sharedindex.%s",
1696 sha1_to_hex(split_index->base_sha1)),
1697 sha1_to_hex(split_index->base->sha1));
1698 merge_base_index(istate);
1699 post_read_index_from(istate);
1703 int is_index_unborn(struct index_state *istate)
1705 return (!istate->cache_nr && !istate->timestamp.sec);
1708 int discard_index(struct index_state *istate)
1712 for (i = 0; i < istate->cache_nr; i++) {
1713 if (istate->cache[i]->index &&
1714 istate->split_index &&
1715 istate->split_index->base &&
1716 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1717 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1719 free(istate->cache[i]);
1721 resolve_undo_clear_index(istate);
1722 istate->cache_nr = 0;
1723 istate->cache_changed = 0;
1724 istate->timestamp.sec = 0;
1725 istate->timestamp.nsec = 0;
1726 free_name_hash(istate);
1727 cache_tree_free(&(istate->cache_tree));
1728 istate->initialized = 0;
1729 free(istate->cache);
1730 istate->cache = NULL;
1731 istate->cache_alloc = 0;
1732 discard_split_index(istate);
1733 free_untracked_cache(istate->untracked);
1734 istate->untracked = NULL;
1738 int unmerged_index(const struct index_state *istate)
1741 for (i = 0; i < istate->cache_nr; i++) {
1742 if (ce_stage(istate->cache[i]))
1748 #define WRITE_BUFFER_SIZE 8192
1749 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1750 static unsigned long write_buffer_len;
1752 static int ce_write_flush(git_SHA_CTX *context, int fd)
1754 unsigned int buffered = write_buffer_len;
1756 git_SHA1_Update(context, write_buffer, buffered);
1757 if (write_in_full(fd, write_buffer, buffered) != buffered)
1759 write_buffer_len = 0;
1764 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1767 unsigned int buffered = write_buffer_len;
1768 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1771 memcpy(write_buffer + buffered, data, partial);
1772 buffered += partial;
1773 if (buffered == WRITE_BUFFER_SIZE) {
1774 write_buffer_len = buffered;
1775 if (ce_write_flush(context, fd))
1779 write_buffer_len = buffered;
1781 data = (char *) data + partial;
1786 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1787 unsigned int ext, unsigned int sz)
1791 return ((ce_write(context, fd, &ext, 4) < 0) ||
1792 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1795 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1797 unsigned int left = write_buffer_len;
1800 write_buffer_len = 0;
1801 git_SHA1_Update(context, write_buffer, left);
1804 /* Flush first if not enough space for SHA1 signature */
1805 if (left + 20 > WRITE_BUFFER_SIZE) {
1806 if (write_in_full(fd, write_buffer, left) != left)
1811 /* Append the SHA1 signature at the end */
1812 git_SHA1_Final(write_buffer + left, context);
1813 hashcpy(sha1, write_buffer + left);
1815 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1818 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1821 * The only thing we care about in this function is to smudge the
1822 * falsely clean entry due to touch-update-touch race, so we leave
1823 * everything else as they are. We are called for entries whose
1824 * ce_stat_data.sd_mtime match the index file mtime.
1826 * Note that this actually does not do much for gitlinks, for
1827 * which ce_match_stat_basic() always goes to the actual
1828 * contents. The caller checks with is_racy_timestamp() which
1829 * always says "no" for gitlinks, so we are not called for them ;-)
1833 if (lstat(ce->name, &st) < 0)
1835 if (ce_match_stat_basic(ce, &st))
1837 if (ce_modified_check_fs(ce, &st)) {
1838 /* This is "racily clean"; smudge it. Note that this
1839 * is a tricky code. At first glance, it may appear
1840 * that it can break with this sequence:
1842 * $ echo xyzzy >frotz
1843 * $ git-update-index --add frotz
1846 * $ echo filfre >nitfol
1847 * $ git-update-index --add nitfol
1849 * but it does not. When the second update-index runs,
1850 * it notices that the entry "frotz" has the same timestamp
1851 * as index, and if we were to smudge it by resetting its
1852 * size to zero here, then the object name recorded
1853 * in index is the 6-byte file but the cached stat information
1854 * becomes zero --- which would then match what we would
1855 * obtain from the filesystem next time we stat("frotz").
1857 * However, the second update-index, before calling
1858 * this function, notices that the cached size is 6
1859 * bytes and what is on the filesystem is an empty
1860 * file, and never calls us, so the cached size information
1861 * for "frotz" stays 6 which does not match the filesystem.
1863 ce->ce_stat_data.sd_size = 0;
1867 /* Copy miscellaneous fields but not the name */
1868 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1869 struct cache_entry *ce)
1873 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1874 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1875 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1876 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1877 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1878 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1879 ondisk->mode = htonl(ce->ce_mode);
1880 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1881 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1882 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1883 hashcpy(ondisk->sha1, ce->oid.hash);
1885 flags = ce->ce_flags & ~CE_NAMEMASK;
1886 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1887 ondisk->flags = htons(flags);
1888 if (ce->ce_flags & CE_EXTENDED) {
1889 struct ondisk_cache_entry_extended *ondisk2;
1890 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1891 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1892 return ondisk2->name;
1895 return ondisk->name;
1899 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1900 struct strbuf *previous_name)
1903 struct ondisk_cache_entry *ondisk;
1904 int saved_namelen = saved_namelen; /* compiler workaround */
1908 if (ce->ce_flags & CE_STRIP_NAME) {
1909 saved_namelen = ce_namelen(ce);
1913 if (!previous_name) {
1914 size = ondisk_ce_size(ce);
1915 ondisk = xcalloc(1, size);
1916 name = copy_cache_entry_to_ondisk(ondisk, ce);
1917 memcpy(name, ce->name, ce_namelen(ce));
1919 int common, to_remove, prefix_size;
1920 unsigned char to_remove_vi[16];
1922 (ce->name[common] &&
1923 common < previous_name->len &&
1924 ce->name[common] == previous_name->buf[common]);
1926 ; /* still matching */
1927 to_remove = previous_name->len - common;
1928 prefix_size = encode_varint(to_remove, to_remove_vi);
1930 if (ce->ce_flags & CE_EXTENDED)
1931 size = offsetof(struct ondisk_cache_entry_extended, name);
1933 size = offsetof(struct ondisk_cache_entry, name);
1934 size += prefix_size + (ce_namelen(ce) - common + 1);
1936 ondisk = xcalloc(1, size);
1937 name = copy_cache_entry_to_ondisk(ondisk, ce);
1938 memcpy(name, to_remove_vi, prefix_size);
1939 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1941 strbuf_splice(previous_name, common, to_remove,
1942 ce->name + common, ce_namelen(ce) - common);
1944 if (ce->ce_flags & CE_STRIP_NAME) {
1945 ce->ce_namelen = saved_namelen;
1946 ce->ce_flags &= ~CE_STRIP_NAME;
1949 result = ce_write(c, fd, ondisk, size);
1955 * This function verifies if index_state has the correct sha1 of the
1956 * index file. Don't die if we have any other failure, just return 0.
1958 static int verify_index_from(const struct index_state *istate, const char *path)
1963 unsigned char sha1[20];
1965 if (!istate->initialized)
1968 fd = open(path, O_RDONLY);
1975 if (st.st_size < sizeof(struct cache_header) + 20)
1978 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1982 if (hashcmp(istate->sha1, sha1))
1993 static int verify_index(const struct index_state *istate)
1995 return verify_index_from(istate, get_index_file());
1998 static int has_racy_timestamp(struct index_state *istate)
2000 int entries = istate->cache_nr;
2003 for (i = 0; i < entries; i++) {
2004 struct cache_entry *ce = istate->cache[i];
2005 if (is_racy_timestamp(istate, ce))
2012 * Opportunistically update the index but do not complain if we can't
2014 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2016 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2017 verify_index(istate) &&
2018 write_locked_index(istate, lockfile, COMMIT_LOCK))
2019 rollback_lock_file(lockfile);
2022 static int do_write_index(struct index_state *istate, int newfd,
2023 int strip_extensions)
2026 struct cache_header hdr;
2027 int i, err, removed, extended, hdr_version;
2028 struct cache_entry **cache = istate->cache;
2029 int entries = istate->cache_nr;
2031 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2033 for (i = removed = extended = 0; i < entries; i++) {
2034 if (cache[i]->ce_flags & CE_REMOVE)
2037 /* reduce extended entries if possible */
2038 cache[i]->ce_flags &= ~CE_EXTENDED;
2039 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2041 cache[i]->ce_flags |= CE_EXTENDED;
2045 if (!istate->version) {
2046 istate->version = get_index_format_default();
2047 if (getenv("GIT_TEST_SPLIT_INDEX"))
2048 init_split_index(istate);
2051 /* demote version 3 to version 2 when the latter suffices */
2052 if (istate->version == 3 || istate->version == 2)
2053 istate->version = extended ? 3 : 2;
2055 hdr_version = istate->version;
2057 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2058 hdr.hdr_version = htonl(hdr_version);
2059 hdr.hdr_entries = htonl(entries - removed);
2062 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2065 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2066 for (i = 0; i < entries; i++) {
2067 struct cache_entry *ce = cache[i];
2068 if (ce->ce_flags & CE_REMOVE)
2070 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2071 ce_smudge_racily_clean_entry(ce);
2072 if (is_null_oid(&ce->oid)) {
2073 static const char msg[] = "cache entry has null sha1: %s";
2074 static int allow = -1;
2077 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2079 warning(msg, ce->name);
2081 return error(msg, ce->name);
2083 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2086 strbuf_release(&previous_name_buf);
2088 /* Write extension data here */
2089 if (!strip_extensions && istate->split_index) {
2090 struct strbuf sb = STRBUF_INIT;
2092 err = write_link_extension(&sb, istate) < 0 ||
2093 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2095 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2096 strbuf_release(&sb);
2100 if (!strip_extensions && istate->cache_tree) {
2101 struct strbuf sb = STRBUF_INIT;
2103 cache_tree_write(&sb, istate->cache_tree);
2104 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2105 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2106 strbuf_release(&sb);
2110 if (!strip_extensions && istate->resolve_undo) {
2111 struct strbuf sb = STRBUF_INIT;
2113 resolve_undo_write(&sb, istate->resolve_undo);
2114 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2116 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2117 strbuf_release(&sb);
2121 if (!strip_extensions && istate->untracked) {
2122 struct strbuf sb = STRBUF_INIT;
2124 write_untracked_extension(&sb, istate->untracked);
2125 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2127 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2128 strbuf_release(&sb);
2133 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2135 istate->timestamp.sec = (unsigned int)st.st_mtime;
2136 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2140 void set_alternate_index_output(const char *name)
2142 alternate_index_output = name;
2145 static int commit_locked_index(struct lock_file *lk)
2147 if (alternate_index_output)
2148 return commit_lock_file_to(lk, alternate_index_output);
2150 return commit_lock_file(lk);
2153 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2156 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2159 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2160 (COMMIT_LOCK | CLOSE_LOCK));
2161 if (flags & COMMIT_LOCK)
2162 return commit_locked_index(lock);
2163 else if (flags & CLOSE_LOCK)
2164 return close_lock_file(lock);
2169 static int write_split_index(struct index_state *istate,
2170 struct lock_file *lock,
2174 prepare_to_write_split_index(istate);
2175 ret = do_write_locked_index(istate, lock, flags);
2176 finish_writing_split_index(istate);
2180 static struct tempfile temporary_sharedindex;
2182 static int write_shared_index(struct index_state *istate,
2183 struct lock_file *lock, unsigned flags)
2185 struct split_index *si = istate->split_index;
2188 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2190 hashclr(si->base_sha1);
2191 return do_write_locked_index(istate, lock, flags);
2193 move_cache_to_base_index(istate);
2194 ret = do_write_index(si->base, fd, 1);
2196 delete_tempfile(&temporary_sharedindex);
2199 ret = rename_tempfile(&temporary_sharedindex,
2200 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2202 hashcpy(si->base_sha1, si->base->sha1);
2206 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2209 struct split_index *si = istate->split_index;
2211 if (!si || alternate_index_output ||
2212 (istate->cache_changed & ~EXTMASK)) {
2214 hashclr(si->base_sha1);
2215 return do_write_locked_index(istate, lock, flags);
2218 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2219 int v = si->base_sha1[0];
2221 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2223 if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2224 int ret = write_shared_index(istate, lock, flags);
2229 return write_split_index(istate, lock, flags);
2233 * Read the index file that is potentially unmerged into given
2234 * index_state, dropping any unmerged entries. Returns true if
2235 * the index is unmerged. Callers who want to refuse to work
2236 * from an unmerged state can call this and check its return value,
2237 * instead of calling read_cache().
2239 int read_index_unmerged(struct index_state *istate)
2245 for (i = 0; i < istate->cache_nr; i++) {
2246 struct cache_entry *ce = istate->cache[i];
2247 struct cache_entry *new_ce;
2253 len = ce_namelen(ce);
2254 size = cache_entry_size(len);
2255 new_ce = xcalloc(1, size);
2256 memcpy(new_ce->name, ce->name, len);
2257 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2258 new_ce->ce_namelen = len;
2259 new_ce->ce_mode = ce->ce_mode;
2260 if (add_index_entry(istate, new_ce, 0))
2261 return error("%s: cannot drop to stage #0",
2268 * Returns 1 if the path is an "other" path with respect to
2269 * the index; that is, the path is not mentioned in the index at all,
2270 * either as a file, a directory with some files in the index,
2271 * or as an unmerged entry.
2273 * We helpfully remove a trailing "/" from directories so that
2274 * the output of read_directory can be used as-is.
2276 int index_name_is_other(const struct index_state *istate, const char *name,
2280 if (namelen && name[namelen - 1] == '/')
2282 pos = index_name_pos(istate, name, namelen);
2284 return 0; /* exact match */
2286 if (pos < istate->cache_nr) {
2287 struct cache_entry *ce = istate->cache[pos];
2288 if (ce_namelen(ce) == namelen &&
2289 !memcmp(ce->name, name, namelen))
2290 return 0; /* Yup, this one exists unmerged */
2295 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2299 enum object_type type;
2303 pos = index_name_pos(istate, path, len);
2306 * We might be in the middle of a merge, in which
2307 * case we would read stage #2 (ours).
2311 (pos < 0 && i < istate->cache_nr &&
2312 !strcmp(istate->cache[i]->name, path));
2314 if (ce_stage(istate->cache[i]) == 2)
2319 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2320 if (!data || type != OBJ_BLOB) {
2329 void stat_validity_clear(struct stat_validity *sv)
2335 int stat_validity_check(struct stat_validity *sv, const char *path)
2339 if (stat(path, &st) < 0)
2340 return sv->sd == NULL;
2343 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2346 void stat_validity_update(struct stat_validity *sv, int fd)
2350 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2351 stat_validity_clear(sv);
2354 sv->sd = xcalloc(1, sizeof(struct stat_data));
2355 fill_stat_data(sv->sd, &st);