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 int fd = open(ce->name, O_RDONLY);
162 unsigned char sha1[20];
163 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
164 match = hashcmp(sha1, ce->sha1);
165 /* index_fd() closed the file descriptor already */
170 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
175 enum object_type type;
176 struct strbuf sb = STRBUF_INIT;
178 if (strbuf_readlink(&sb, ce->name, expected_size))
181 buffer = read_sha1_file(ce->sha1, &type, &size);
184 match = memcmp(buffer, sb.buf, size);
191 static int ce_compare_gitlink(const struct cache_entry *ce)
193 unsigned char sha1[20];
196 * We don't actually require that the .git directory
197 * under GITLINK directory be a valid git directory. It
198 * might even be missing (in case nobody populated that
201 * If so, we consider it always to match.
203 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
205 return hashcmp(sha1, ce->sha1);
208 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
210 switch (st->st_mode & S_IFMT) {
212 if (ce_compare_data(ce, st))
216 if (ce_compare_link(ce, xsize_t(st->st_size)))
220 if (S_ISGITLINK(ce->ce_mode))
221 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
228 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
230 unsigned int changed = 0;
232 if (ce->ce_flags & CE_REMOVE)
233 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
235 switch (ce->ce_mode & S_IFMT) {
237 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
238 /* We consider only the owner x bit to be relevant for
241 if (trust_executable_bit &&
242 (0100 & (ce->ce_mode ^ st->st_mode)))
243 changed |= MODE_CHANGED;
246 if (!S_ISLNK(st->st_mode) &&
247 (has_symlinks || !S_ISREG(st->st_mode)))
248 changed |= TYPE_CHANGED;
251 /* We ignore most of the st_xxx fields for gitlinks */
252 if (!S_ISDIR(st->st_mode))
253 changed |= TYPE_CHANGED;
254 else if (ce_compare_gitlink(ce))
255 changed |= DATA_CHANGED;
258 die("internal error: ce_mode is %o", ce->ce_mode);
261 changed |= match_stat_data(&ce->ce_stat_data, st);
263 /* Racily smudged entry? */
264 if (!ce->ce_stat_data.sd_size) {
265 if (!is_empty_blob_sha1(ce->sha1))
266 changed |= DATA_CHANGED;
272 static int is_racy_stat(const struct index_state *istate,
273 const struct stat_data *sd)
275 return (istate->timestamp.sec &&
277 /* nanosecond timestamped files can also be racy! */
278 (istate->timestamp.sec < sd->sd_mtime.sec ||
279 (istate->timestamp.sec == sd->sd_mtime.sec &&
280 istate->timestamp.nsec <= sd->sd_mtime.nsec))
282 istate->timestamp.sec <= sd->sd_mtime.sec
287 static int is_racy_timestamp(const struct index_state *istate,
288 const struct cache_entry *ce)
290 return (!S_ISGITLINK(ce->ce_mode) &&
291 is_racy_stat(istate, &ce->ce_stat_data));
294 int match_stat_data_racy(const struct index_state *istate,
295 const struct stat_data *sd, struct stat *st)
297 if (is_racy_stat(istate, sd))
298 return MTIME_CHANGED;
299 return match_stat_data(sd, st);
302 int ie_match_stat(const struct index_state *istate,
303 const struct cache_entry *ce, struct stat *st,
304 unsigned int options)
306 unsigned int changed;
307 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
308 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
309 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
312 * If it's marked as always valid in the index, it's
313 * valid whatever the checked-out copy says.
315 * skip-worktree has the same effect with higher precedence
317 if (!ignore_skip_worktree && ce_skip_worktree(ce))
319 if (!ignore_valid && (ce->ce_flags & CE_VALID))
323 * Intent-to-add entries have not been added, so the index entry
324 * by definition never matches what is in the work tree until it
325 * actually gets added.
327 if (ce_intent_to_add(ce))
328 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
330 changed = ce_match_stat_basic(ce, st);
333 * Within 1 second of this sequence:
334 * echo xyzzy >file && git-update-index --add file
335 * running this command:
337 * would give a falsely clean cache entry. The mtime and
338 * length match the cache, and other stat fields do not change.
340 * We could detect this at update-index time (the cache entry
341 * being registered/updated records the same time as "now")
342 * and delay the return from git-update-index, but that would
343 * effectively mean we can make at most one commit per second,
344 * which is not acceptable. Instead, we check cache entries
345 * whose mtime are the same as the index file timestamp more
346 * carefully than others.
348 if (!changed && is_racy_timestamp(istate, ce)) {
349 if (assume_racy_is_modified)
350 changed |= DATA_CHANGED;
352 changed |= ce_modified_check_fs(ce, st);
358 int ie_modified(const struct index_state *istate,
359 const struct cache_entry *ce,
360 struct stat *st, unsigned int options)
362 int changed, changed_fs;
364 changed = ie_match_stat(istate, ce, st, options);
368 * If the mode or type has changed, there's no point in trying
369 * to refresh the entry - it's not going to match
371 if (changed & (MODE_CHANGED | TYPE_CHANGED))
375 * Immediately after read-tree or update-index --cacheinfo,
376 * the length field is zero, as we have never even read the
377 * lstat(2) information once, and we cannot trust DATA_CHANGED
378 * returned by ie_match_stat() which in turn was returned by
379 * ce_match_stat_basic() to signal that the filesize of the
380 * blob changed. We have to actually go to the filesystem to
381 * see if the contents match, and if so, should answer "unchanged".
383 * The logic does not apply to gitlinks, as ce_match_stat_basic()
384 * already has checked the actual HEAD from the filesystem in the
385 * subproject. If ie_match_stat() already said it is different,
386 * then we know it is.
388 if ((changed & DATA_CHANGED) &&
389 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
392 changed_fs = ce_modified_check_fs(ce, st);
394 return changed | changed_fs;
398 int base_name_compare(const char *name1, int len1, int mode1,
399 const char *name2, int len2, int mode2)
401 unsigned char c1, c2;
402 int len = len1 < len2 ? len1 : len2;
405 cmp = memcmp(name1, name2, len);
410 if (!c1 && S_ISDIR(mode1))
412 if (!c2 && S_ISDIR(mode2))
414 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
418 * df_name_compare() is identical to base_name_compare(), except it
419 * compares conflicting directory/file entries as equal. Note that
420 * while a directory name compares as equal to a regular file, they
421 * then individually compare _differently_ to a filename that has
422 * a dot after the basename (because '\0' < '.' < '/').
424 * This is used by routines that want to traverse the git namespace
425 * but then handle conflicting entries together when possible.
427 int df_name_compare(const char *name1, int len1, int mode1,
428 const char *name2, int len2, int mode2)
430 int len = len1 < len2 ? len1 : len2, cmp;
431 unsigned char c1, c2;
433 cmp = memcmp(name1, name2, len);
436 /* Directories and files compare equal (same length, same name) */
440 if (!c1 && S_ISDIR(mode1))
443 if (!c2 && S_ISDIR(mode2))
445 if (c1 == '/' && !c2)
447 if (c2 == '/' && !c1)
452 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
454 size_t min_len = (len1 < len2) ? len1 : len2;
455 int cmp = memcmp(name1, name2, min_len);
465 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
469 cmp = name_compare(name1, len1, name2, len2);
480 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
485 last = istate->cache_nr;
486 while (last > first) {
487 int next = (last + first) >> 1;
488 struct cache_entry *ce = istate->cache[next];
489 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
501 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
503 return index_name_stage_pos(istate, name, namelen, 0);
506 /* Remove entry, return true if there are more entries to go.. */
507 int remove_index_entry_at(struct index_state *istate, int pos)
509 struct cache_entry *ce = istate->cache[pos];
511 record_resolve_undo(istate, ce);
512 remove_name_hash(istate, ce);
513 save_or_free_index_entry(istate, ce);
514 istate->cache_changed |= CE_ENTRY_REMOVED;
516 if (pos >= istate->cache_nr)
518 memmove(istate->cache + pos,
519 istate->cache + pos + 1,
520 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
525 * Remove all cache entries marked for removal, that is where
526 * CE_REMOVE is set in ce_flags. This is much more effective than
527 * calling remove_index_entry_at() for each entry to be removed.
529 void remove_marked_cache_entries(struct index_state *istate)
531 struct cache_entry **ce_array = istate->cache;
534 for (i = j = 0; i < istate->cache_nr; i++) {
535 if (ce_array[i]->ce_flags & CE_REMOVE) {
536 remove_name_hash(istate, ce_array[i]);
537 save_or_free_index_entry(istate, ce_array[i]);
540 ce_array[j++] = ce_array[i];
542 if (j == istate->cache_nr)
544 istate->cache_changed |= CE_ENTRY_REMOVED;
545 istate->cache_nr = j;
548 int remove_file_from_index(struct index_state *istate, const char *path)
550 int pos = index_name_pos(istate, path, strlen(path));
553 cache_tree_invalidate_path(istate, path);
554 untracked_cache_remove_from_index(istate, path);
555 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
556 remove_index_entry_at(istate, pos);
560 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
562 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
565 static int index_name_pos_also_unmerged(struct index_state *istate,
566 const char *path, int namelen)
568 int pos = index_name_pos(istate, path, namelen);
569 struct cache_entry *ce;
574 /* maybe unmerged? */
576 if (pos >= istate->cache_nr ||
577 compare_name((ce = istate->cache[pos]), path, namelen))
580 /* order of preference: stage 2, 1, 3 */
581 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
582 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
583 !compare_name(ce, path, namelen))
588 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
590 int len = ce_namelen(ce);
591 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
595 * If we add a filename that aliases in the cache, we will use the
596 * name that we already have - but we don't want to update the same
597 * alias twice, because that implies that there were actually two
598 * different files with aliasing names!
600 * So we use the CE_ADDED flag to verify that the alias was an old
601 * one before we accept it as
603 static struct cache_entry *create_alias_ce(struct index_state *istate,
604 struct cache_entry *ce,
605 struct cache_entry *alias)
608 struct cache_entry *new;
610 if (alias->ce_flags & CE_ADDED)
611 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
613 /* Ok, create the new entry using the name of the existing alias */
614 len = ce_namelen(alias);
615 new = xcalloc(1, cache_entry_size(len));
616 memcpy(new->name, alias->name, len);
617 copy_cache_entry(new, ce);
618 save_or_free_index_entry(istate, ce);
622 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
624 unsigned char sha1[20];
625 if (write_sha1_file("", 0, blob_type, sha1))
626 die("cannot create an empty blob in the object database");
627 hashcpy(ce->sha1, sha1);
630 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
632 int size, namelen, was_same;
633 mode_t st_mode = st->st_mode;
634 struct cache_entry *ce, *alias;
635 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
636 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
637 int pretend = flags & ADD_CACHE_PRETEND;
638 int intent_only = flags & ADD_CACHE_INTENT;
639 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
640 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
642 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
643 return error("%s: can only add regular files, symbolic links or git-directories", path);
645 namelen = strlen(path);
646 if (S_ISDIR(st_mode)) {
647 while (namelen && path[namelen-1] == '/')
650 size = cache_entry_size(namelen);
651 ce = xcalloc(1, size);
652 memcpy(ce->name, path, namelen);
653 ce->ce_namelen = namelen;
655 fill_stat_cache_info(ce, st);
657 ce->ce_flags |= CE_INTENT_TO_ADD;
660 if (trust_executable_bit && has_symlinks) {
661 ce->ce_mode = create_ce_mode(st_mode);
663 /* If there is an existing entry, pick the mode bits and type
664 * from it, otherwise assume unexecutable regular file.
666 struct cache_entry *ent;
667 int pos = index_name_pos_also_unmerged(istate, path, namelen);
669 ent = (0 <= pos) ? istate->cache[pos] : NULL;
670 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
673 /* When core.ignorecase=true, determine if a directory of the same name but differing
674 * case already exists within the Git repository. If it does, ensure the directory
675 * case of the file being added to the repository matches (is folded into) the existing
676 * entry's directory case.
679 adjust_dirname_case(istate, ce->name);
682 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
683 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
684 /* Nothing changed, really */
685 if (!S_ISGITLINK(alias->ce_mode))
686 ce_mark_uptodate(alias);
687 alias->ce_flags |= CE_ADDED;
693 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) {
695 return error("unable to index file %s", path);
698 set_object_name_for_intent_to_add_entry(ce);
700 if (ignore_case && alias && different_name(ce, alias))
701 ce = create_alias_ce(istate, ce, alias);
702 ce->ce_flags |= CE_ADDED;
704 /* It was suspected to be racily clean, but it turns out to be Ok */
707 !hashcmp(alias->sha1, ce->sha1) &&
708 ce->ce_mode == alias->ce_mode);
712 else if (add_index_entry(istate, ce, add_option)) {
714 return error("unable to add %s to index", path);
716 if (verbose && !was_same)
717 printf("add '%s'\n", path);
721 int add_file_to_index(struct index_state *istate, const char *path, int flags)
724 if (lstat(path, &st))
725 die_errno("unable to stat '%s'", path);
726 return add_to_index(istate, path, &st, flags);
729 struct cache_entry *make_cache_entry(unsigned int mode,
730 const unsigned char *sha1, const char *path, int stage,
731 unsigned int refresh_options)
734 struct cache_entry *ce, *ret;
736 if (!verify_path(path)) {
737 error("Invalid path '%s'", path);
742 size = cache_entry_size(len);
743 ce = xcalloc(1, size);
745 hashcpy(ce->sha1, sha1);
746 memcpy(ce->name, path, len);
747 ce->ce_flags = create_ce_flags(stage);
748 ce->ce_namelen = len;
749 ce->ce_mode = create_ce_mode(mode);
751 ret = refresh_cache_entry(ce, refresh_options);
758 * Chmod an index entry with either +x or -x.
760 * Returns -1 if the chmod for the particular cache entry failed (if it's
761 * not a regular file), -2 if an invalid flip argument is passed in, 0
764 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
767 if (!S_ISREG(ce->ce_mode))
774 ce->ce_mode &= ~0111;
779 cache_tree_invalidate_path(istate, ce->name);
780 ce->ce_flags |= CE_UPDATE_IN_BASE;
781 istate->cache_changed |= CE_ENTRY_CHANGED;
786 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
788 int len = ce_namelen(a);
789 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
793 * We fundamentally don't like some paths: we don't want
794 * dot or dot-dot anywhere, and for obvious reasons don't
795 * want to recurse into ".git" either.
797 * Also, we don't want double slashes or slashes at the
798 * end that can make pathnames ambiguous.
800 static int verify_dotfile(const char *rest)
803 * The first character was '.', but that
804 * has already been discarded, we now test
808 /* "." is not allowed */
809 if (*rest == '\0' || is_dir_sep(*rest))
814 * ".git" followed by NUL or slash is bad. This
815 * shares the path end test with the ".." case.
819 if (rest[1] != 'i' && rest[1] != 'I')
821 if (rest[2] != 't' && rest[2] != 'T')
826 if (rest[1] == '\0' || is_dir_sep(rest[1]))
832 int verify_path(const char *path)
836 if (has_dos_drive_prefix(path))
845 if (protect_hfs && is_hfs_dotgit(path))
847 if (protect_ntfs && is_ntfs_dotgit(path))
850 if ((c == '.' && !verify_dotfile(path)) ||
851 is_dir_sep(c) || c == '\0')
859 * Do we have another file that has the beginning components being a
860 * proper superset of the name we're trying to add?
862 static int has_file_name(struct index_state *istate,
863 const struct cache_entry *ce, int pos, int ok_to_replace)
866 int len = ce_namelen(ce);
867 int stage = ce_stage(ce);
868 const char *name = ce->name;
870 while (pos < istate->cache_nr) {
871 struct cache_entry *p = istate->cache[pos++];
873 if (len >= ce_namelen(p))
875 if (memcmp(name, p->name, len))
877 if (ce_stage(p) != stage)
879 if (p->name[len] != '/')
881 if (p->ce_flags & CE_REMOVE)
886 remove_index_entry_at(istate, --pos);
892 * Do we have another file with a pathname that is a proper
893 * subset of the name we're trying to add?
895 static int has_dir_name(struct index_state *istate,
896 const struct cache_entry *ce, int pos, int ok_to_replace)
899 int stage = ce_stage(ce);
900 const char *name = ce->name;
901 const char *slash = name + ce_namelen(ce);
909 if (slash <= ce->name)
914 pos = index_name_stage_pos(istate, name, len, stage);
917 * Found one, but not so fast. This could
918 * be a marker that says "I was here, but
919 * I am being removed". Such an entry is
920 * not a part of the resulting tree, and
921 * it is Ok to have a directory at the same
924 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
928 remove_index_entry_at(istate, pos);
936 * Trivial optimization: if we find an entry that
937 * already matches the sub-directory, then we know
938 * we're ok, and we can exit.
940 while (pos < istate->cache_nr) {
941 struct cache_entry *p = istate->cache[pos];
942 if ((ce_namelen(p) <= len) ||
943 (p->name[len] != '/') ||
944 memcmp(p->name, name, len))
945 break; /* not our subdirectory */
946 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
948 * p is at the same stage as our entry, and
949 * is a subdirectory of what we are looking
950 * at, so we cannot have conflicts at our
951 * level or anything shorter.
960 /* We may be in a situation where we already have path/file and path
961 * is being added, or we already have path and path/file is being
962 * added. Either one would result in a nonsense tree that has path
963 * twice when git-write-tree tries to write it out. Prevent it.
965 * If ok-to-replace is specified, we remove the conflicting entries
966 * from the cache so the caller should recompute the insert position.
967 * When this happens, we return non-zero.
969 static int check_file_directory_conflict(struct index_state *istate,
970 const struct cache_entry *ce,
971 int pos, int ok_to_replace)
976 * When ce is an "I am going away" entry, we allow it to be added
978 if (ce->ce_flags & CE_REMOVE)
982 * We check if the path is a sub-path of a subsequent pathname
983 * first, since removing those will not change the position
986 retval = has_file_name(istate, ce, pos, ok_to_replace);
989 * Then check if the path might have a clashing sub-directory
992 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
995 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
998 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
999 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1000 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1001 int new_only = option & ADD_CACHE_NEW_ONLY;
1003 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1004 cache_tree_invalidate_path(istate, ce->name);
1005 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1007 /* existing match? Just replace it. */
1010 replace_index_entry(istate, pos, ce);
1015 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1016 untracked_cache_add_to_index(istate, ce->name);
1019 * Inserting a merged entry ("stage 0") into the index
1020 * will always replace all non-merged entries..
1022 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1023 while (ce_same_name(istate->cache[pos], ce)) {
1025 if (!remove_index_entry_at(istate, pos))
1032 if (!verify_path(ce->name))
1033 return error("Invalid path '%s'", ce->name);
1035 if (!skip_df_check &&
1036 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1038 return error("'%s' appears as both a file and as a directory",
1040 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1046 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1050 if (option & ADD_CACHE_JUST_APPEND)
1051 pos = istate->cache_nr;
1054 ret = add_index_entry_with_check(istate, ce, option);
1060 /* Make sure the array is big enough .. */
1061 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1065 if (istate->cache_nr > pos + 1)
1066 memmove(istate->cache + pos + 1,
1067 istate->cache + pos,
1068 (istate->cache_nr - pos - 1) * sizeof(ce));
1069 set_index_entry(istate, pos, ce);
1070 istate->cache_changed |= CE_ENTRY_ADDED;
1075 * "refresh" does not calculate a new sha1 file or bring the
1076 * cache up-to-date for mode/content changes. But what it
1077 * _does_ do is to "re-match" the stat information of a file
1078 * with the cache, so that you can refresh the cache for a
1079 * file that hasn't been changed but where the stat entry is
1082 * For example, you'd want to do this after doing a "git-read-tree",
1083 * to link up the stat cache details with the proper files.
1085 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1086 struct cache_entry *ce,
1087 unsigned int options, int *err,
1091 struct cache_entry *updated;
1093 int refresh = options & CE_MATCH_REFRESH;
1094 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1095 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1096 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1098 if (!refresh || ce_uptodate(ce))
1102 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1103 * that the change to the work tree does not matter and told
1106 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1107 ce_mark_uptodate(ce);
1110 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1111 ce_mark_uptodate(ce);
1115 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1123 if (lstat(ce->name, &st) < 0) {
1124 if (ignore_missing && errno == ENOENT)
1131 changed = ie_match_stat(istate, ce, &st, options);
1133 *changed_ret = changed;
1136 * The path is unchanged. If we were told to ignore
1137 * valid bit, then we did the actual stat check and
1138 * found that the entry is unmodified. If the entry
1139 * is not marked VALID, this is the place to mark it
1140 * valid again, under "assume unchanged" mode.
1142 if (ignore_valid && assume_unchanged &&
1143 !(ce->ce_flags & CE_VALID))
1144 ; /* mark this one VALID again */
1147 * We do not mark the index itself "modified"
1148 * because CE_UPTODATE flag is in-core only;
1149 * we are not going to write this change out.
1151 if (!S_ISGITLINK(ce->ce_mode))
1152 ce_mark_uptodate(ce);
1157 if (ie_modified(istate, ce, &st, options)) {
1164 updated = xmalloc(size);
1165 memcpy(updated, ce, size);
1166 fill_stat_cache_info(updated, &st);
1168 * If ignore_valid is not set, we should leave CE_VALID bit
1169 * alone. Otherwise, paths marked with --no-assume-unchanged
1170 * (i.e. things to be edited) will reacquire CE_VALID bit
1171 * automatically, which is not really what we want.
1173 if (!ignore_valid && assume_unchanged &&
1174 !(ce->ce_flags & CE_VALID))
1175 updated->ce_flags &= ~CE_VALID;
1177 /* istate->cache_changed is updated in the caller */
1181 static void show_file(const char * fmt, const char * name, int in_porcelain,
1182 int * first, const char *header_msg)
1184 if (in_porcelain && *first && header_msg) {
1185 printf("%s\n", header_msg);
1191 int refresh_index(struct index_state *istate, unsigned int flags,
1192 const struct pathspec *pathspec,
1193 char *seen, const char *header_msg)
1197 int really = (flags & REFRESH_REALLY) != 0;
1198 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1199 int quiet = (flags & REFRESH_QUIET) != 0;
1200 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1201 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1203 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1204 unsigned int options = (CE_MATCH_REFRESH |
1205 (really ? CE_MATCH_IGNORE_VALID : 0) |
1206 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1207 const char *modified_fmt;
1208 const char *deleted_fmt;
1209 const char *typechange_fmt;
1210 const char *added_fmt;
1211 const char *unmerged_fmt;
1213 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1214 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1215 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1216 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1217 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1218 for (i = 0; i < istate->cache_nr; i++) {
1219 struct cache_entry *ce, *new;
1220 int cache_errno = 0;
1224 ce = istate->cache[i];
1225 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1228 if (pathspec && !ce_path_match(ce, pathspec, seen))
1232 while ((i < istate->cache_nr) &&
1233 ! strcmp(istate->cache[i]->name, ce->name))
1239 show_file(unmerged_fmt, ce->name, in_porcelain,
1240 &first, header_msg);
1248 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1254 if (really && cache_errno == EINVAL) {
1255 /* If we are doing --really-refresh that
1256 * means the index is not valid anymore.
1258 ce->ce_flags &= ~CE_VALID;
1259 ce->ce_flags |= CE_UPDATE_IN_BASE;
1260 istate->cache_changed |= CE_ENTRY_CHANGED;
1265 if (cache_errno == ENOENT)
1267 else if (ce_intent_to_add(ce))
1268 fmt = added_fmt; /* must be before other checks */
1269 else if (changed & TYPE_CHANGED)
1270 fmt = typechange_fmt;
1274 ce->name, in_porcelain, &first, header_msg);
1279 replace_index_entry(istate, i, new);
1284 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1285 unsigned int options)
1287 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1291 /*****************************************************************
1293 *****************************************************************/
1295 #define INDEX_FORMAT_DEFAULT 3
1297 static unsigned int get_index_format_default(void)
1299 char *envversion = getenv("GIT_INDEX_VERSION");
1302 unsigned int version = INDEX_FORMAT_DEFAULT;
1305 if (!git_config_get_int("index.version", &value))
1307 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1308 warning(_("index.version set, but the value is invalid.\n"
1309 "Using version %i"), INDEX_FORMAT_DEFAULT);
1310 return INDEX_FORMAT_DEFAULT;
1315 version = strtoul(envversion, &endp, 10);
1317 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1318 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1319 "Using version %i"), INDEX_FORMAT_DEFAULT);
1320 version = INDEX_FORMAT_DEFAULT;
1326 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1327 * Again - this is just a (very strong in practice) heuristic that
1328 * the inode hasn't changed.
1330 * We save the fields in big-endian order to allow using the
1331 * index file over NFS transparently.
1333 struct ondisk_cache_entry {
1334 struct cache_time ctime;
1335 struct cache_time mtime;
1342 unsigned char sha1[20];
1344 char name[FLEX_ARRAY]; /* more */
1348 * This struct is used when CE_EXTENDED bit is 1
1349 * The struct must match ondisk_cache_entry exactly from
1352 struct ondisk_cache_entry_extended {
1353 struct cache_time ctime;
1354 struct cache_time mtime;
1361 unsigned char sha1[20];
1364 char name[FLEX_ARRAY]; /* more */
1367 /* These are only used for v3 or lower */
1368 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1369 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1370 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1371 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1372 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1373 ondisk_cache_entry_size(ce_namelen(ce)))
1375 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1378 unsigned char sha1[20];
1381 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1382 return error("bad signature");
1383 hdr_version = ntohl(hdr->hdr_version);
1384 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1385 return error("bad index version %d", hdr_version);
1387 git_SHA1_Update(&c, hdr, size - 20);
1388 git_SHA1_Final(sha1, &c);
1389 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1390 return error("bad index file sha1 signature");
1394 static int read_index_extension(struct index_state *istate,
1395 const char *ext, void *data, unsigned long sz)
1397 switch (CACHE_EXT(ext)) {
1398 case CACHE_EXT_TREE:
1399 istate->cache_tree = cache_tree_read(data, sz);
1401 case CACHE_EXT_RESOLVE_UNDO:
1402 istate->resolve_undo = resolve_undo_read(data, sz);
1404 case CACHE_EXT_LINK:
1405 if (read_link_extension(istate, data, sz))
1408 case CACHE_EXT_UNTRACKED:
1409 istate->untracked = read_untracked_extension(data, sz);
1412 if (*ext < 'A' || 'Z' < *ext)
1413 return error("index uses %.4s extension, which we do not understand",
1415 fprintf(stderr, "ignoring %.4s extension\n", ext);
1421 int hold_locked_index(struct lock_file *lk, int die_on_error)
1423 return hold_lock_file_for_update(lk, get_index_file(),
1429 int read_index(struct index_state *istate)
1431 return read_index_from(istate, get_index_file());
1434 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1439 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1441 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1442 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1443 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1444 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1445 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1446 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1447 ce->ce_mode = get_be32(&ondisk->mode);
1448 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1449 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1450 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1451 ce->ce_flags = flags & ~CE_NAMEMASK;
1452 ce->ce_namelen = len;
1454 hashcpy(ce->sha1, ondisk->sha1);
1455 memcpy(ce->name, name, len);
1456 ce->name[len] = '\0';
1461 * Adjacent cache entries tend to share the leading paths, so it makes
1462 * sense to only store the differences in later entries. In the v4
1463 * on-disk format of the index, each on-disk cache entry stores the
1464 * number of bytes to be stripped from the end of the previous name,
1465 * and the bytes to append to the result, to come up with its name.
1467 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1469 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1470 size_t len = decode_varint(&cp);
1472 if (name->len < len)
1473 die("malformed name field in the index");
1474 strbuf_remove(name, name->len - len, len);
1475 for (ep = cp; *ep; ep++)
1476 ; /* find the end */
1477 strbuf_add(name, cp, ep - cp);
1478 return (const char *)ep + 1 - cp_;
1481 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1482 unsigned long *ent_size,
1483 struct strbuf *previous_name)
1485 struct cache_entry *ce;
1490 /* On-disk flags are just 16 bits */
1491 flags = get_be16(&ondisk->flags);
1492 len = flags & CE_NAMEMASK;
1494 if (flags & CE_EXTENDED) {
1495 struct ondisk_cache_entry_extended *ondisk2;
1497 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1498 extended_flags = get_be16(&ondisk2->flags2) << 16;
1499 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1500 if (extended_flags & ~CE_EXTENDED_FLAGS)
1501 die("Unknown index entry format %08x", extended_flags);
1502 flags |= extended_flags;
1503 name = ondisk2->name;
1506 name = ondisk->name;
1508 if (!previous_name) {
1509 /* v3 and earlier */
1510 if (len == CE_NAMEMASK)
1512 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1514 *ent_size = ondisk_ce_size(ce);
1516 unsigned long consumed;
1517 consumed = expand_name_field(previous_name, name);
1518 ce = cache_entry_from_ondisk(ondisk, flags,
1520 previous_name->len);
1522 *ent_size = (name - ((char *)ondisk)) + consumed;
1527 static void check_ce_order(struct index_state *istate)
1531 for (i = 1; i < istate->cache_nr; i++) {
1532 struct cache_entry *ce = istate->cache[i - 1];
1533 struct cache_entry *next_ce = istate->cache[i];
1534 int name_compare = strcmp(ce->name, next_ce->name);
1536 if (0 < name_compare)
1537 die("unordered stage entries in index");
1538 if (!name_compare) {
1540 die("multiple stage entries for merged file '%s'",
1542 if (ce_stage(ce) > ce_stage(next_ce))
1543 die("unordered stage entries for '%s'",
1549 static void tweak_untracked_cache(struct index_state *istate)
1551 switch (git_config_get_untracked_cache()) {
1552 case -1: /* keep: do nothing */
1555 remove_untracked_cache(istate);
1558 add_untracked_cache(istate);
1560 default: /* unknown value: do nothing */
1565 static void post_read_index_from(struct index_state *istate)
1567 check_ce_order(istate);
1568 tweak_untracked_cache(istate);
1571 /* remember to discard_cache() before reading a different cache! */
1572 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1576 unsigned long src_offset;
1577 struct cache_header *hdr;
1580 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1582 if (istate->initialized)
1583 return istate->cache_nr;
1585 istate->timestamp.sec = 0;
1586 istate->timestamp.nsec = 0;
1587 fd = open(path, O_RDONLY);
1589 if (!must_exist && errno == ENOENT)
1591 die_errno("%s: index file open failed", path);
1595 die_errno("cannot stat the open index");
1597 mmap_size = xsize_t(st.st_size);
1598 if (mmap_size < sizeof(struct cache_header) + 20)
1599 die("index file smaller than expected");
1601 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1602 if (mmap == MAP_FAILED)
1603 die_errno("unable to map index file");
1607 if (verify_hdr(hdr, mmap_size) < 0)
1610 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1611 istate->version = ntohl(hdr->hdr_version);
1612 istate->cache_nr = ntohl(hdr->hdr_entries);
1613 istate->cache_alloc = alloc_nr(istate->cache_nr);
1614 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1615 istate->initialized = 1;
1617 if (istate->version == 4)
1618 previous_name = &previous_name_buf;
1620 previous_name = NULL;
1622 src_offset = sizeof(*hdr);
1623 for (i = 0; i < istate->cache_nr; i++) {
1624 struct ondisk_cache_entry *disk_ce;
1625 struct cache_entry *ce;
1626 unsigned long consumed;
1628 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1629 ce = create_from_disk(disk_ce, &consumed, previous_name);
1630 set_index_entry(istate, i, ce);
1632 src_offset += consumed;
1634 strbuf_release(&previous_name_buf);
1635 istate->timestamp.sec = st.st_mtime;
1636 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1638 while (src_offset <= mmap_size - 20 - 8) {
1639 /* After an array of active_nr index entries,
1640 * there can be arbitrary number of extended
1641 * sections, each of which is prefixed with
1642 * extension name (4-byte) and section length
1643 * in 4-byte network byte order.
1646 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1647 extsize = ntohl(extsize);
1648 if (read_index_extension(istate,
1649 (const char *) mmap + src_offset,
1650 (char *) mmap + src_offset + 8,
1654 src_offset += extsize;
1656 munmap(mmap, mmap_size);
1657 return istate->cache_nr;
1660 munmap(mmap, mmap_size);
1661 die("index file corrupt");
1664 int read_index_from(struct index_state *istate, const char *path)
1666 struct split_index *split_index;
1669 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1670 if (istate->initialized)
1671 return istate->cache_nr;
1673 ret = do_read_index(istate, path, 0);
1675 split_index = istate->split_index;
1676 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1677 post_read_index_from(istate);
1681 if (split_index->base)
1682 discard_index(split_index->base);
1684 split_index->base = xcalloc(1, sizeof(*split_index->base));
1685 ret = do_read_index(split_index->base,
1686 git_path("sharedindex.%s",
1687 sha1_to_hex(split_index->base_sha1)), 1);
1688 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1689 die("broken index, expect %s in %s, got %s",
1690 sha1_to_hex(split_index->base_sha1),
1691 git_path("sharedindex.%s",
1692 sha1_to_hex(split_index->base_sha1)),
1693 sha1_to_hex(split_index->base->sha1));
1694 merge_base_index(istate);
1695 post_read_index_from(istate);
1699 int is_index_unborn(struct index_state *istate)
1701 return (!istate->cache_nr && !istate->timestamp.sec);
1704 int discard_index(struct index_state *istate)
1708 for (i = 0; i < istate->cache_nr; i++) {
1709 if (istate->cache[i]->index &&
1710 istate->split_index &&
1711 istate->split_index->base &&
1712 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1713 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1715 free(istate->cache[i]);
1717 resolve_undo_clear_index(istate);
1718 istate->cache_nr = 0;
1719 istate->cache_changed = 0;
1720 istate->timestamp.sec = 0;
1721 istate->timestamp.nsec = 0;
1722 free_name_hash(istate);
1723 cache_tree_free(&(istate->cache_tree));
1724 istate->initialized = 0;
1725 free(istate->cache);
1726 istate->cache = NULL;
1727 istate->cache_alloc = 0;
1728 discard_split_index(istate);
1729 free_untracked_cache(istate->untracked);
1730 istate->untracked = NULL;
1734 int unmerged_index(const struct index_state *istate)
1737 for (i = 0; i < istate->cache_nr; i++) {
1738 if (ce_stage(istate->cache[i]))
1744 #define WRITE_BUFFER_SIZE 8192
1745 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1746 static unsigned long write_buffer_len;
1748 static int ce_write_flush(git_SHA_CTX *context, int fd)
1750 unsigned int buffered = write_buffer_len;
1752 git_SHA1_Update(context, write_buffer, buffered);
1753 if (write_in_full(fd, write_buffer, buffered) != buffered)
1755 write_buffer_len = 0;
1760 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1763 unsigned int buffered = write_buffer_len;
1764 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1767 memcpy(write_buffer + buffered, data, partial);
1768 buffered += partial;
1769 if (buffered == WRITE_BUFFER_SIZE) {
1770 write_buffer_len = buffered;
1771 if (ce_write_flush(context, fd))
1775 write_buffer_len = buffered;
1777 data = (char *) data + partial;
1782 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1783 unsigned int ext, unsigned int sz)
1787 return ((ce_write(context, fd, &ext, 4) < 0) ||
1788 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1791 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1793 unsigned int left = write_buffer_len;
1796 write_buffer_len = 0;
1797 git_SHA1_Update(context, write_buffer, left);
1800 /* Flush first if not enough space for SHA1 signature */
1801 if (left + 20 > WRITE_BUFFER_SIZE) {
1802 if (write_in_full(fd, write_buffer, left) != left)
1807 /* Append the SHA1 signature at the end */
1808 git_SHA1_Final(write_buffer + left, context);
1809 hashcpy(sha1, write_buffer + left);
1811 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1814 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1817 * The only thing we care about in this function is to smudge the
1818 * falsely clean entry due to touch-update-touch race, so we leave
1819 * everything else as they are. We are called for entries whose
1820 * ce_stat_data.sd_mtime match the index file mtime.
1822 * Note that this actually does not do much for gitlinks, for
1823 * which ce_match_stat_basic() always goes to the actual
1824 * contents. The caller checks with is_racy_timestamp() which
1825 * always says "no" for gitlinks, so we are not called for them ;-)
1829 if (lstat(ce->name, &st) < 0)
1831 if (ce_match_stat_basic(ce, &st))
1833 if (ce_modified_check_fs(ce, &st)) {
1834 /* This is "racily clean"; smudge it. Note that this
1835 * is a tricky code. At first glance, it may appear
1836 * that it can break with this sequence:
1838 * $ echo xyzzy >frotz
1839 * $ git-update-index --add frotz
1842 * $ echo filfre >nitfol
1843 * $ git-update-index --add nitfol
1845 * but it does not. When the second update-index runs,
1846 * it notices that the entry "frotz" has the same timestamp
1847 * as index, and if we were to smudge it by resetting its
1848 * size to zero here, then the object name recorded
1849 * in index is the 6-byte file but the cached stat information
1850 * becomes zero --- which would then match what we would
1851 * obtain from the filesystem next time we stat("frotz").
1853 * However, the second update-index, before calling
1854 * this function, notices that the cached size is 6
1855 * bytes and what is on the filesystem is an empty
1856 * file, and never calls us, so the cached size information
1857 * for "frotz" stays 6 which does not match the filesystem.
1859 ce->ce_stat_data.sd_size = 0;
1863 /* Copy miscellaneous fields but not the name */
1864 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1865 struct cache_entry *ce)
1869 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1870 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1871 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1872 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1873 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1874 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1875 ondisk->mode = htonl(ce->ce_mode);
1876 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1877 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1878 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1879 hashcpy(ondisk->sha1, ce->sha1);
1881 flags = ce->ce_flags & ~CE_NAMEMASK;
1882 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1883 ondisk->flags = htons(flags);
1884 if (ce->ce_flags & CE_EXTENDED) {
1885 struct ondisk_cache_entry_extended *ondisk2;
1886 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1887 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1888 return ondisk2->name;
1891 return ondisk->name;
1895 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1896 struct strbuf *previous_name)
1899 struct ondisk_cache_entry *ondisk;
1900 int saved_namelen = saved_namelen; /* compiler workaround */
1904 if (ce->ce_flags & CE_STRIP_NAME) {
1905 saved_namelen = ce_namelen(ce);
1909 if (!previous_name) {
1910 size = ondisk_ce_size(ce);
1911 ondisk = xcalloc(1, size);
1912 name = copy_cache_entry_to_ondisk(ondisk, ce);
1913 memcpy(name, ce->name, ce_namelen(ce));
1915 int common, to_remove, prefix_size;
1916 unsigned char to_remove_vi[16];
1918 (ce->name[common] &&
1919 common < previous_name->len &&
1920 ce->name[common] == previous_name->buf[common]);
1922 ; /* still matching */
1923 to_remove = previous_name->len - common;
1924 prefix_size = encode_varint(to_remove, to_remove_vi);
1926 if (ce->ce_flags & CE_EXTENDED)
1927 size = offsetof(struct ondisk_cache_entry_extended, name);
1929 size = offsetof(struct ondisk_cache_entry, name);
1930 size += prefix_size + (ce_namelen(ce) - common + 1);
1932 ondisk = xcalloc(1, size);
1933 name = copy_cache_entry_to_ondisk(ondisk, ce);
1934 memcpy(name, to_remove_vi, prefix_size);
1935 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1937 strbuf_splice(previous_name, common, to_remove,
1938 ce->name + common, ce_namelen(ce) - common);
1940 if (ce->ce_flags & CE_STRIP_NAME) {
1941 ce->ce_namelen = saved_namelen;
1942 ce->ce_flags &= ~CE_STRIP_NAME;
1945 result = ce_write(c, fd, ondisk, size);
1951 * This function verifies if index_state has the correct sha1 of the
1952 * index file. Don't die if we have any other failure, just return 0.
1954 static int verify_index_from(const struct index_state *istate, const char *path)
1959 unsigned char sha1[20];
1961 if (!istate->initialized)
1964 fd = open(path, O_RDONLY);
1971 if (st.st_size < sizeof(struct cache_header) + 20)
1974 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1978 if (hashcmp(istate->sha1, sha1))
1989 static int verify_index(const struct index_state *istate)
1991 return verify_index_from(istate, get_index_file());
1994 static int has_racy_timestamp(struct index_state *istate)
1996 int entries = istate->cache_nr;
1999 for (i = 0; i < entries; i++) {
2000 struct cache_entry *ce = istate->cache[i];
2001 if (is_racy_timestamp(istate, ce))
2008 * Opportunistically update the index but do not complain if we can't
2010 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2012 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2013 verify_index(istate) &&
2014 write_locked_index(istate, lockfile, COMMIT_LOCK))
2015 rollback_lock_file(lockfile);
2018 static int do_write_index(struct index_state *istate, int newfd,
2019 int strip_extensions)
2022 struct cache_header hdr;
2023 int i, err, removed, extended, hdr_version;
2024 struct cache_entry **cache = istate->cache;
2025 int entries = istate->cache_nr;
2027 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2029 for (i = removed = extended = 0; i < entries; i++) {
2030 if (cache[i]->ce_flags & CE_REMOVE)
2033 /* reduce extended entries if possible */
2034 cache[i]->ce_flags &= ~CE_EXTENDED;
2035 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2037 cache[i]->ce_flags |= CE_EXTENDED;
2041 if (!istate->version) {
2042 istate->version = get_index_format_default();
2043 if (getenv("GIT_TEST_SPLIT_INDEX"))
2044 init_split_index(istate);
2047 /* demote version 3 to version 2 when the latter suffices */
2048 if (istate->version == 3 || istate->version == 2)
2049 istate->version = extended ? 3 : 2;
2051 hdr_version = istate->version;
2053 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2054 hdr.hdr_version = htonl(hdr_version);
2055 hdr.hdr_entries = htonl(entries - removed);
2058 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2061 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2062 for (i = 0; i < entries; i++) {
2063 struct cache_entry *ce = cache[i];
2064 if (ce->ce_flags & CE_REMOVE)
2066 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2067 ce_smudge_racily_clean_entry(ce);
2068 if (is_null_sha1(ce->sha1)) {
2069 static const char msg[] = "cache entry has null sha1: %s";
2070 static int allow = -1;
2073 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2075 warning(msg, ce->name);
2077 return error(msg, ce->name);
2079 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2082 strbuf_release(&previous_name_buf);
2084 /* Write extension data here */
2085 if (!strip_extensions && istate->split_index) {
2086 struct strbuf sb = STRBUF_INIT;
2088 err = write_link_extension(&sb, istate) < 0 ||
2089 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2091 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2092 strbuf_release(&sb);
2096 if (!strip_extensions && istate->cache_tree) {
2097 struct strbuf sb = STRBUF_INIT;
2099 cache_tree_write(&sb, istate->cache_tree);
2100 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2101 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2102 strbuf_release(&sb);
2106 if (!strip_extensions && istate->resolve_undo) {
2107 struct strbuf sb = STRBUF_INIT;
2109 resolve_undo_write(&sb, istate->resolve_undo);
2110 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2112 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2113 strbuf_release(&sb);
2117 if (!strip_extensions && istate->untracked) {
2118 struct strbuf sb = STRBUF_INIT;
2120 write_untracked_extension(&sb, istate->untracked);
2121 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2123 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2124 strbuf_release(&sb);
2129 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2131 istate->timestamp.sec = (unsigned int)st.st_mtime;
2132 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2136 void set_alternate_index_output(const char *name)
2138 alternate_index_output = name;
2141 static int commit_locked_index(struct lock_file *lk)
2143 if (alternate_index_output)
2144 return commit_lock_file_to(lk, alternate_index_output);
2146 return commit_lock_file(lk);
2149 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2152 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2155 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2156 (COMMIT_LOCK | CLOSE_LOCK));
2157 if (flags & COMMIT_LOCK)
2158 return commit_locked_index(lock);
2159 else if (flags & CLOSE_LOCK)
2160 return close_lock_file(lock);
2165 static int write_split_index(struct index_state *istate,
2166 struct lock_file *lock,
2170 prepare_to_write_split_index(istate);
2171 ret = do_write_locked_index(istate, lock, flags);
2172 finish_writing_split_index(istate);
2176 static struct tempfile temporary_sharedindex;
2178 static int write_shared_index(struct index_state *istate,
2179 struct lock_file *lock, unsigned flags)
2181 struct split_index *si = istate->split_index;
2184 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2186 hashclr(si->base_sha1);
2187 return do_write_locked_index(istate, lock, flags);
2189 move_cache_to_base_index(istate);
2190 ret = do_write_index(si->base, fd, 1);
2192 delete_tempfile(&temporary_sharedindex);
2195 ret = rename_tempfile(&temporary_sharedindex,
2196 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2198 hashcpy(si->base_sha1, si->base->sha1);
2202 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2205 struct split_index *si = istate->split_index;
2207 if (!si || alternate_index_output ||
2208 (istate->cache_changed & ~EXTMASK)) {
2210 hashclr(si->base_sha1);
2211 return do_write_locked_index(istate, lock, flags);
2214 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2215 int v = si->base_sha1[0];
2217 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2219 if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2220 int ret = write_shared_index(istate, lock, flags);
2225 return write_split_index(istate, lock, flags);
2229 * Read the index file that is potentially unmerged into given
2230 * index_state, dropping any unmerged entries. Returns true if
2231 * the index is unmerged. Callers who want to refuse to work
2232 * from an unmerged state can call this and check its return value,
2233 * instead of calling read_cache().
2235 int read_index_unmerged(struct index_state *istate)
2241 for (i = 0; i < istate->cache_nr; i++) {
2242 struct cache_entry *ce = istate->cache[i];
2243 struct cache_entry *new_ce;
2249 len = ce_namelen(ce);
2250 size = cache_entry_size(len);
2251 new_ce = xcalloc(1, size);
2252 memcpy(new_ce->name, ce->name, len);
2253 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2254 new_ce->ce_namelen = len;
2255 new_ce->ce_mode = ce->ce_mode;
2256 if (add_index_entry(istate, new_ce, 0))
2257 return error("%s: cannot drop to stage #0",
2264 * Returns 1 if the path is an "other" path with respect to
2265 * the index; that is, the path is not mentioned in the index at all,
2266 * either as a file, a directory with some files in the index,
2267 * or as an unmerged entry.
2269 * We helpfully remove a trailing "/" from directories so that
2270 * the output of read_directory can be used as-is.
2272 int index_name_is_other(const struct index_state *istate, const char *name,
2276 if (namelen && name[namelen - 1] == '/')
2278 pos = index_name_pos(istate, name, namelen);
2280 return 0; /* exact match */
2282 if (pos < istate->cache_nr) {
2283 struct cache_entry *ce = istate->cache[pos];
2284 if (ce_namelen(ce) == namelen &&
2285 !memcmp(ce->name, name, namelen))
2286 return 0; /* Yup, this one exists unmerged */
2291 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2295 enum object_type type;
2299 pos = index_name_pos(istate, path, len);
2302 * We might be in the middle of a merge, in which
2303 * case we would read stage #2 (ours).
2307 (pos < 0 && i < istate->cache_nr &&
2308 !strcmp(istate->cache[i]->name, path));
2310 if (ce_stage(istate->cache[i]) == 2)
2315 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2316 if (!data || type != OBJ_BLOB) {
2325 void stat_validity_clear(struct stat_validity *sv)
2331 int stat_validity_check(struct stat_validity *sv, const char *path)
2335 if (stat(path, &st) < 0)
2336 return sv->sd == NULL;
2339 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2342 void stat_validity_update(struct stat_validity *sv, int fd)
2346 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2347 stat_validity_clear(sv);
2350 sv->sd = xcalloc(1, sizeof(struct stat_data));
2351 fill_stat_data(sv->sd, &st);