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 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
23 unsigned int options);
25 /* Mask for the name length in ce_flags in the on-disk index */
27 #define CE_NAMEMASK (0x0fff)
31 * The first letter should be 'A'..'Z' for extensions that are not
32 * necessary for a correct operation (i.e. optimization data).
33 * When new extensions are added that _needs_ to be understood in
34 * order to correctly interpret the index file, pick character that
35 * is outside the range, to cause the reader to abort.
38 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
39 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
40 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
41 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
42 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
44 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
45 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
46 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
47 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
49 struct index_state the_index;
50 static const char *alternate_index_output;
52 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
54 istate->cache[nr] = ce;
55 add_name_hash(istate, ce);
58 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
60 struct cache_entry *old = istate->cache[nr];
62 replace_index_entry_in_base(istate, old, ce);
63 remove_name_hash(istate, old);
65 set_index_entry(istate, nr, ce);
66 ce->ce_flags |= CE_UPDATE_IN_BASE;
67 istate->cache_changed |= CE_ENTRY_CHANGED;
70 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
72 struct cache_entry *old = istate->cache[nr], *new;
73 int namelen = strlen(new_name);
75 new = xmalloc(cache_entry_size(namelen));
76 copy_cache_entry(new, old);
77 new->ce_flags &= ~CE_HASHED;
78 new->ce_namelen = namelen;
80 memcpy(new->name, new_name, namelen + 1);
82 cache_tree_invalidate_path(istate, old->name);
83 untracked_cache_remove_from_index(istate, old->name);
84 remove_index_entry_at(istate, nr);
85 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
88 void fill_stat_data(struct stat_data *sd, struct stat *st)
90 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
91 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
92 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
93 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
94 sd->sd_dev = st->st_dev;
95 sd->sd_ino = st->st_ino;
96 sd->sd_uid = st->st_uid;
97 sd->sd_gid = st->st_gid;
98 sd->sd_size = st->st_size;
101 int match_stat_data(const struct stat_data *sd, struct stat *st)
105 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
106 changed |= MTIME_CHANGED;
107 if (trust_ctime && check_stat &&
108 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
109 changed |= CTIME_CHANGED;
112 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
113 changed |= MTIME_CHANGED;
114 if (trust_ctime && check_stat &&
115 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
116 changed |= CTIME_CHANGED;
120 if (sd->sd_uid != (unsigned int) st->st_uid ||
121 sd->sd_gid != (unsigned int) st->st_gid)
122 changed |= OWNER_CHANGED;
123 if (sd->sd_ino != (unsigned int) st->st_ino)
124 changed |= INODE_CHANGED;
129 * st_dev breaks on network filesystems where different
130 * clients will have different views of what "device"
131 * the filesystem is on
133 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
134 changed |= INODE_CHANGED;
137 if (sd->sd_size != (unsigned int) st->st_size)
138 changed |= DATA_CHANGED;
144 * This only updates the "non-critical" parts of the directory
145 * cache, ie the parts that aren't tracked by GIT, and only used
146 * to validate the cache.
148 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
150 fill_stat_data(&ce->ce_stat_data, st);
152 if (assume_unchanged)
153 ce->ce_flags |= CE_VALID;
155 if (S_ISREG(st->st_mode))
156 ce_mark_uptodate(ce);
159 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
162 int fd = open(ce->name, O_RDONLY);
165 unsigned char sha1[20];
166 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
167 match = hashcmp(sha1, ce->sha1);
168 /* index_fd() closed the file descriptor already */
173 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
178 enum object_type type;
179 struct strbuf sb = STRBUF_INIT;
181 if (strbuf_readlink(&sb, ce->name, expected_size))
184 buffer = read_sha1_file(ce->sha1, &type, &size);
187 match = memcmp(buffer, sb.buf, size);
194 static int ce_compare_gitlink(const struct cache_entry *ce)
196 unsigned char sha1[20];
199 * We don't actually require that the .git directory
200 * under GITLINK directory be a valid git directory. It
201 * might even be missing (in case nobody populated that
204 * If so, we consider it always to match.
206 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
208 return hashcmp(sha1, ce->sha1);
211 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
213 switch (st->st_mode & S_IFMT) {
215 if (ce_compare_data(ce, st))
219 if (ce_compare_link(ce, xsize_t(st->st_size)))
223 if (S_ISGITLINK(ce->ce_mode))
224 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
231 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
233 unsigned int changed = 0;
235 if (ce->ce_flags & CE_REMOVE)
236 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
238 switch (ce->ce_mode & S_IFMT) {
240 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
241 /* We consider only the owner x bit to be relevant for
244 if (trust_executable_bit &&
245 (0100 & (ce->ce_mode ^ st->st_mode)))
246 changed |= MODE_CHANGED;
249 if (!S_ISLNK(st->st_mode) &&
250 (has_symlinks || !S_ISREG(st->st_mode)))
251 changed |= TYPE_CHANGED;
254 /* We ignore most of the st_xxx fields for gitlinks */
255 if (!S_ISDIR(st->st_mode))
256 changed |= TYPE_CHANGED;
257 else if (ce_compare_gitlink(ce))
258 changed |= DATA_CHANGED;
261 die("internal error: ce_mode is %o", ce->ce_mode);
264 changed |= match_stat_data(&ce->ce_stat_data, st);
266 /* Racily smudged entry? */
267 if (!ce->ce_stat_data.sd_size) {
268 if (!is_empty_blob_sha1(ce->sha1))
269 changed |= DATA_CHANGED;
275 static int is_racy_stat(const struct index_state *istate,
276 const struct stat_data *sd)
278 return (istate->timestamp.sec &&
280 /* nanosecond timestamped files can also be racy! */
281 (istate->timestamp.sec < sd->sd_mtime.sec ||
282 (istate->timestamp.sec == sd->sd_mtime.sec &&
283 istate->timestamp.nsec <= sd->sd_mtime.nsec))
285 istate->timestamp.sec <= sd->sd_mtime.sec
290 static int is_racy_timestamp(const struct index_state *istate,
291 const struct cache_entry *ce)
293 return (!S_ISGITLINK(ce->ce_mode) &&
294 is_racy_stat(istate, &ce->ce_stat_data));
297 int match_stat_data_racy(const struct index_state *istate,
298 const struct stat_data *sd, struct stat *st)
300 if (is_racy_stat(istate, sd))
301 return MTIME_CHANGED;
302 return match_stat_data(sd, st);
305 int ie_match_stat(const struct index_state *istate,
306 const struct cache_entry *ce, struct stat *st,
307 unsigned int options)
309 unsigned int changed;
310 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
311 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
312 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
315 * If it's marked as always valid in the index, it's
316 * valid whatever the checked-out copy says.
318 * skip-worktree has the same effect with higher precedence
320 if (!ignore_skip_worktree && ce_skip_worktree(ce))
322 if (!ignore_valid && (ce->ce_flags & CE_VALID))
326 * Intent-to-add entries have not been added, so the index entry
327 * by definition never matches what is in the work tree until it
328 * actually gets added.
330 if (ce_intent_to_add(ce))
331 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
333 changed = ce_match_stat_basic(ce, st);
336 * Within 1 second of this sequence:
337 * echo xyzzy >file && git-update-index --add file
338 * running this command:
340 * would give a falsely clean cache entry. The mtime and
341 * length match the cache, and other stat fields do not change.
343 * We could detect this at update-index time (the cache entry
344 * being registered/updated records the same time as "now")
345 * and delay the return from git-update-index, but that would
346 * effectively mean we can make at most one commit per second,
347 * which is not acceptable. Instead, we check cache entries
348 * whose mtime are the same as the index file timestamp more
349 * carefully than others.
351 if (!changed && is_racy_timestamp(istate, ce)) {
352 if (assume_racy_is_modified)
353 changed |= DATA_CHANGED;
355 changed |= ce_modified_check_fs(ce, st);
361 int ie_modified(const struct index_state *istate,
362 const struct cache_entry *ce,
363 struct stat *st, unsigned int options)
365 int changed, changed_fs;
367 changed = ie_match_stat(istate, ce, st, options);
371 * If the mode or type has changed, there's no point in trying
372 * to refresh the entry - it's not going to match
374 if (changed & (MODE_CHANGED | TYPE_CHANGED))
378 * Immediately after read-tree or update-index --cacheinfo,
379 * the length field is zero, as we have never even read the
380 * lstat(2) information once, and we cannot trust DATA_CHANGED
381 * returned by ie_match_stat() which in turn was returned by
382 * ce_match_stat_basic() to signal that the filesize of the
383 * blob changed. We have to actually go to the filesystem to
384 * see if the contents match, and if so, should answer "unchanged".
386 * The logic does not apply to gitlinks, as ce_match_stat_basic()
387 * already has checked the actual HEAD from the filesystem in the
388 * subproject. If ie_match_stat() already said it is different,
389 * then we know it is.
391 if ((changed & DATA_CHANGED) &&
392 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
395 changed_fs = ce_modified_check_fs(ce, st);
397 return changed | changed_fs;
401 int base_name_compare(const char *name1, int len1, int mode1,
402 const char *name2, int len2, int mode2)
404 unsigned char c1, c2;
405 int len = len1 < len2 ? len1 : len2;
408 cmp = memcmp(name1, name2, len);
413 if (!c1 && S_ISDIR(mode1))
415 if (!c2 && S_ISDIR(mode2))
417 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
421 * df_name_compare() is identical to base_name_compare(), except it
422 * compares conflicting directory/file entries as equal. Note that
423 * while a directory name compares as equal to a regular file, they
424 * then individually compare _differently_ to a filename that has
425 * a dot after the basename (because '\0' < '.' < '/').
427 * This is used by routines that want to traverse the git namespace
428 * but then handle conflicting entries together when possible.
430 int df_name_compare(const char *name1, int len1, int mode1,
431 const char *name2, int len2, int mode2)
433 int len = len1 < len2 ? len1 : len2, cmp;
434 unsigned char c1, c2;
436 cmp = memcmp(name1, name2, len);
439 /* Directories and files compare equal (same length, same name) */
443 if (!c1 && S_ISDIR(mode1))
446 if (!c2 && S_ISDIR(mode2))
448 if (c1 == '/' && !c2)
450 if (c2 == '/' && !c1)
455 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
457 size_t min_len = (len1 < len2) ? len1 : len2;
458 int cmp = memcmp(name1, name2, min_len);
468 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
472 cmp = name_compare(name1, len1, name2, len2);
483 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
488 last = istate->cache_nr;
489 while (last > first) {
490 int next = (last + first) >> 1;
491 struct cache_entry *ce = istate->cache[next];
492 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
504 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
506 return index_name_stage_pos(istate, name, namelen, 0);
509 /* Remove entry, return true if there are more entries to go.. */
510 int remove_index_entry_at(struct index_state *istate, int pos)
512 struct cache_entry *ce = istate->cache[pos];
514 record_resolve_undo(istate, ce);
515 remove_name_hash(istate, ce);
516 save_or_free_index_entry(istate, ce);
517 istate->cache_changed |= CE_ENTRY_REMOVED;
519 if (pos >= istate->cache_nr)
521 memmove(istate->cache + pos,
522 istate->cache + pos + 1,
523 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
528 * Remove all cache entries marked for removal, that is where
529 * CE_REMOVE is set in ce_flags. This is much more effective than
530 * calling remove_index_entry_at() for each entry to be removed.
532 void remove_marked_cache_entries(struct index_state *istate)
534 struct cache_entry **ce_array = istate->cache;
537 for (i = j = 0; i < istate->cache_nr; i++) {
538 if (ce_array[i]->ce_flags & CE_REMOVE) {
539 remove_name_hash(istate, ce_array[i]);
540 save_or_free_index_entry(istate, ce_array[i]);
543 ce_array[j++] = ce_array[i];
545 if (j == istate->cache_nr)
547 istate->cache_changed |= CE_ENTRY_REMOVED;
548 istate->cache_nr = j;
551 int remove_file_from_index(struct index_state *istate, const char *path)
553 int pos = index_name_pos(istate, path, strlen(path));
556 cache_tree_invalidate_path(istate, path);
557 untracked_cache_remove_from_index(istate, path);
558 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
559 remove_index_entry_at(istate, pos);
563 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
565 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
568 static int index_name_pos_also_unmerged(struct index_state *istate,
569 const char *path, int namelen)
571 int pos = index_name_pos(istate, path, namelen);
572 struct cache_entry *ce;
577 /* maybe unmerged? */
579 if (pos >= istate->cache_nr ||
580 compare_name((ce = istate->cache[pos]), path, namelen))
583 /* order of preference: stage 2, 1, 3 */
584 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
585 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
586 !compare_name(ce, path, namelen))
591 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
593 int len = ce_namelen(ce);
594 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
598 * If we add a filename that aliases in the cache, we will use the
599 * name that we already have - but we don't want to update the same
600 * alias twice, because that implies that there were actually two
601 * different files with aliasing names!
603 * So we use the CE_ADDED flag to verify that the alias was an old
604 * one before we accept it as
606 static struct cache_entry *create_alias_ce(struct index_state *istate,
607 struct cache_entry *ce,
608 struct cache_entry *alias)
611 struct cache_entry *new;
613 if (alias->ce_flags & CE_ADDED)
614 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
616 /* Ok, create the new entry using the name of the existing alias */
617 len = ce_namelen(alias);
618 new = xcalloc(1, cache_entry_size(len));
619 memcpy(new->name, alias->name, len);
620 copy_cache_entry(new, ce);
621 save_or_free_index_entry(istate, ce);
625 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
627 unsigned char sha1[20];
628 if (write_sha1_file("", 0, blob_type, sha1))
629 die("cannot create an empty blob in the object database");
630 hashcpy(ce->sha1, sha1);
633 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags, int force_mode)
635 int size, namelen, was_same;
636 mode_t st_mode = st->st_mode;
637 struct cache_entry *ce, *alias;
638 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
639 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
640 int pretend = flags & ADD_CACHE_PRETEND;
641 int intent_only = flags & ADD_CACHE_INTENT;
642 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
643 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
645 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
646 return error("%s: can only add regular files, symbolic links or git-directories", path);
648 namelen = strlen(path);
649 if (S_ISDIR(st_mode)) {
650 while (namelen && path[namelen-1] == '/')
653 size = cache_entry_size(namelen);
654 ce = xcalloc(1, size);
655 memcpy(ce->name, path, namelen);
656 ce->ce_namelen = namelen;
658 fill_stat_cache_info(ce, st);
660 ce->ce_flags |= CE_INTENT_TO_ADD;
662 if (S_ISREG(st_mode) && force_mode)
663 ce->ce_mode = create_ce_mode(force_mode);
664 else if (trust_executable_bit && has_symlinks)
665 ce->ce_mode = create_ce_mode(st_mode);
667 /* If there is an existing entry, pick the mode bits and type
668 * from it, otherwise assume unexecutable regular file.
670 struct cache_entry *ent;
671 int pos = index_name_pos_also_unmerged(istate, path, namelen);
673 ent = (0 <= pos) ? istate->cache[pos] : NULL;
674 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
677 /* When core.ignorecase=true, determine if a directory of the same name but differing
678 * case already exists within the Git repository. If it does, ensure the directory
679 * case of the file being added to the repository matches (is folded into) the existing
680 * entry's directory case.
683 adjust_dirname_case(istate, ce->name);
686 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
687 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
688 /* Nothing changed, really */
689 if (!S_ISGITLINK(alias->ce_mode))
690 ce_mark_uptodate(alias);
691 alias->ce_flags |= CE_ADDED;
697 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) {
699 return error("unable to index file %s", path);
702 set_object_name_for_intent_to_add_entry(ce);
704 if (ignore_case && alias && different_name(ce, alias))
705 ce = create_alias_ce(istate, ce, alias);
706 ce->ce_flags |= CE_ADDED;
708 /* It was suspected to be racily clean, but it turns out to be Ok */
711 !hashcmp(alias->sha1, ce->sha1) &&
712 ce->ce_mode == alias->ce_mode);
716 else if (add_index_entry(istate, ce, add_option)) {
718 return error("unable to add %s to index", path);
720 if (verbose && !was_same)
721 printf("add '%s'\n", path);
725 int add_file_to_index(struct index_state *istate, const char *path,
726 int flags, int force_mode)
729 if (lstat(path, &st))
730 die_errno("unable to stat '%s'", path);
731 return add_to_index(istate, path, &st, flags, force_mode);
734 struct cache_entry *make_cache_entry(unsigned int mode,
735 const unsigned char *sha1, const char *path, int stage,
736 unsigned int refresh_options)
739 struct cache_entry *ce, *ret;
741 if (!verify_path(path)) {
742 error("Invalid path '%s'", path);
747 size = cache_entry_size(len);
748 ce = xcalloc(1, size);
750 hashcpy(ce->sha1, sha1);
751 memcpy(ce->name, path, len);
752 ce->ce_flags = create_ce_flags(stage);
753 ce->ce_namelen = len;
754 ce->ce_mode = create_ce_mode(mode);
756 ret = refresh_cache_entry(ce, refresh_options);
763 * Chmod an index entry with either +x or -x.
765 * Returns -1 if the chmod for the particular cache entry failed (if it's
766 * not a regular file), -2 if an invalid flip argument is passed in, 0
769 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
772 if (!S_ISREG(ce->ce_mode))
779 ce->ce_mode &= ~0111;
784 cache_tree_invalidate_path(istate, ce->name);
785 ce->ce_flags |= CE_UPDATE_IN_BASE;
786 istate->cache_changed |= CE_ENTRY_CHANGED;
791 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
793 int len = ce_namelen(a);
794 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
798 * We fundamentally don't like some paths: we don't want
799 * dot or dot-dot anywhere, and for obvious reasons don't
800 * want to recurse into ".git" either.
802 * Also, we don't want double slashes or slashes at the
803 * end that can make pathnames ambiguous.
805 static int verify_dotfile(const char *rest)
808 * The first character was '.', but that
809 * has already been discarded, we now test
813 /* "." is not allowed */
814 if (*rest == '\0' || is_dir_sep(*rest))
819 * ".git" followed by NUL or slash is bad. This
820 * shares the path end test with the ".." case.
824 if (rest[1] != 'i' && rest[1] != 'I')
826 if (rest[2] != 't' && rest[2] != 'T')
831 if (rest[1] == '\0' || is_dir_sep(rest[1]))
837 int verify_path(const char *path)
841 if (has_dos_drive_prefix(path))
850 if (protect_hfs && is_hfs_dotgit(path))
852 if (protect_ntfs && is_ntfs_dotgit(path))
855 if ((c == '.' && !verify_dotfile(path)) ||
856 is_dir_sep(c) || c == '\0')
864 * Do we have another file that has the beginning components being a
865 * proper superset of the name we're trying to add?
867 static int has_file_name(struct index_state *istate,
868 const struct cache_entry *ce, int pos, int ok_to_replace)
871 int len = ce_namelen(ce);
872 int stage = ce_stage(ce);
873 const char *name = ce->name;
875 while (pos < istate->cache_nr) {
876 struct cache_entry *p = istate->cache[pos++];
878 if (len >= ce_namelen(p))
880 if (memcmp(name, p->name, len))
882 if (ce_stage(p) != stage)
884 if (p->name[len] != '/')
886 if (p->ce_flags & CE_REMOVE)
891 remove_index_entry_at(istate, --pos);
897 * Do we have another file with a pathname that is a proper
898 * subset of the name we're trying to add?
900 static int has_dir_name(struct index_state *istate,
901 const struct cache_entry *ce, int pos, int ok_to_replace)
904 int stage = ce_stage(ce);
905 const char *name = ce->name;
906 const char *slash = name + ce_namelen(ce);
914 if (slash <= ce->name)
919 pos = index_name_stage_pos(istate, name, len, stage);
922 * Found one, but not so fast. This could
923 * be a marker that says "I was here, but
924 * I am being removed". Such an entry is
925 * not a part of the resulting tree, and
926 * it is Ok to have a directory at the same
929 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
933 remove_index_entry_at(istate, pos);
941 * Trivial optimization: if we find an entry that
942 * already matches the sub-directory, then we know
943 * we're ok, and we can exit.
945 while (pos < istate->cache_nr) {
946 struct cache_entry *p = istate->cache[pos];
947 if ((ce_namelen(p) <= len) ||
948 (p->name[len] != '/') ||
949 memcmp(p->name, name, len))
950 break; /* not our subdirectory */
951 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
953 * p is at the same stage as our entry, and
954 * is a subdirectory of what we are looking
955 * at, so we cannot have conflicts at our
956 * level or anything shorter.
965 /* We may be in a situation where we already have path/file and path
966 * is being added, or we already have path and path/file is being
967 * added. Either one would result in a nonsense tree that has path
968 * twice when git-write-tree tries to write it out. Prevent it.
970 * If ok-to-replace is specified, we remove the conflicting entries
971 * from the cache so the caller should recompute the insert position.
972 * When this happens, we return non-zero.
974 static int check_file_directory_conflict(struct index_state *istate,
975 const struct cache_entry *ce,
976 int pos, int ok_to_replace)
981 * When ce is an "I am going away" entry, we allow it to be added
983 if (ce->ce_flags & CE_REMOVE)
987 * We check if the path is a sub-path of a subsequent pathname
988 * first, since removing those will not change the position
991 retval = has_file_name(istate, ce, pos, ok_to_replace);
994 * Then check if the path might have a clashing sub-directory
997 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1000 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1003 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1004 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1005 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1006 int new_only = option & ADD_CACHE_NEW_ONLY;
1008 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1009 cache_tree_invalidate_path(istate, ce->name);
1010 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1012 /* existing match? Just replace it. */
1015 replace_index_entry(istate, pos, ce);
1020 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1021 untracked_cache_add_to_index(istate, ce->name);
1024 * Inserting a merged entry ("stage 0") into the index
1025 * will always replace all non-merged entries..
1027 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1028 while (ce_same_name(istate->cache[pos], ce)) {
1030 if (!remove_index_entry_at(istate, pos))
1037 if (!verify_path(ce->name))
1038 return error("Invalid path '%s'", ce->name);
1040 if (!skip_df_check &&
1041 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1043 return error("'%s' appears as both a file and as a directory",
1045 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1051 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1055 if (option & ADD_CACHE_JUST_APPEND)
1056 pos = istate->cache_nr;
1059 ret = add_index_entry_with_check(istate, ce, option);
1065 /* Make sure the array is big enough .. */
1066 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1070 if (istate->cache_nr > pos + 1)
1071 memmove(istate->cache + pos + 1,
1072 istate->cache + pos,
1073 (istate->cache_nr - pos - 1) * sizeof(ce));
1074 set_index_entry(istate, pos, ce);
1075 istate->cache_changed |= CE_ENTRY_ADDED;
1080 * "refresh" does not calculate a new sha1 file or bring the
1081 * cache up-to-date for mode/content changes. But what it
1082 * _does_ do is to "re-match" the stat information of a file
1083 * with the cache, so that you can refresh the cache for a
1084 * file that hasn't been changed but where the stat entry is
1087 * For example, you'd want to do this after doing a "git-read-tree",
1088 * to link up the stat cache details with the proper files.
1090 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1091 struct cache_entry *ce,
1092 unsigned int options, int *err,
1096 struct cache_entry *updated;
1098 int refresh = options & CE_MATCH_REFRESH;
1099 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1100 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1101 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1103 if (!refresh || ce_uptodate(ce))
1107 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1108 * that the change to the work tree does not matter and told
1111 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1112 ce_mark_uptodate(ce);
1115 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1116 ce_mark_uptodate(ce);
1120 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1128 if (lstat(ce->name, &st) < 0) {
1129 if (ignore_missing && errno == ENOENT)
1136 changed = ie_match_stat(istate, ce, &st, options);
1138 *changed_ret = changed;
1141 * The path is unchanged. If we were told to ignore
1142 * valid bit, then we did the actual stat check and
1143 * found that the entry is unmodified. If the entry
1144 * is not marked VALID, this is the place to mark it
1145 * valid again, under "assume unchanged" mode.
1147 if (ignore_valid && assume_unchanged &&
1148 !(ce->ce_flags & CE_VALID))
1149 ; /* mark this one VALID again */
1152 * We do not mark the index itself "modified"
1153 * because CE_UPTODATE flag is in-core only;
1154 * we are not going to write this change out.
1156 if (!S_ISGITLINK(ce->ce_mode))
1157 ce_mark_uptodate(ce);
1162 if (ie_modified(istate, ce, &st, options)) {
1169 updated = xmalloc(size);
1170 memcpy(updated, ce, size);
1171 fill_stat_cache_info(updated, &st);
1173 * If ignore_valid is not set, we should leave CE_VALID bit
1174 * alone. Otherwise, paths marked with --no-assume-unchanged
1175 * (i.e. things to be edited) will reacquire CE_VALID bit
1176 * automatically, which is not really what we want.
1178 if (!ignore_valid && assume_unchanged &&
1179 !(ce->ce_flags & CE_VALID))
1180 updated->ce_flags &= ~CE_VALID;
1182 /* istate->cache_changed is updated in the caller */
1186 static void show_file(const char * fmt, const char * name, int in_porcelain,
1187 int * first, const char *header_msg)
1189 if (in_porcelain && *first && header_msg) {
1190 printf("%s\n", header_msg);
1196 int refresh_index(struct index_state *istate, unsigned int flags,
1197 const struct pathspec *pathspec,
1198 char *seen, const char *header_msg)
1202 int really = (flags & REFRESH_REALLY) != 0;
1203 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1204 int quiet = (flags & REFRESH_QUIET) != 0;
1205 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1206 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1208 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1209 unsigned int options = (CE_MATCH_REFRESH |
1210 (really ? CE_MATCH_IGNORE_VALID : 0) |
1211 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1212 const char *modified_fmt;
1213 const char *deleted_fmt;
1214 const char *typechange_fmt;
1215 const char *added_fmt;
1216 const char *unmerged_fmt;
1218 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1219 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1220 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1221 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1222 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1223 for (i = 0; i < istate->cache_nr; i++) {
1224 struct cache_entry *ce, *new;
1225 int cache_errno = 0;
1229 ce = istate->cache[i];
1230 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1233 if (pathspec && !ce_path_match(ce, pathspec, seen))
1237 while ((i < istate->cache_nr) &&
1238 ! strcmp(istate->cache[i]->name, ce->name))
1244 show_file(unmerged_fmt, ce->name, in_porcelain,
1245 &first, header_msg);
1253 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1259 if (really && cache_errno == EINVAL) {
1260 /* If we are doing --really-refresh that
1261 * means the index is not valid anymore.
1263 ce->ce_flags &= ~CE_VALID;
1264 ce->ce_flags |= CE_UPDATE_IN_BASE;
1265 istate->cache_changed |= CE_ENTRY_CHANGED;
1270 if (cache_errno == ENOENT)
1272 else if (ce_intent_to_add(ce))
1273 fmt = added_fmt; /* must be before other checks */
1274 else if (changed & TYPE_CHANGED)
1275 fmt = typechange_fmt;
1279 ce->name, in_porcelain, &first, header_msg);
1284 replace_index_entry(istate, i, new);
1289 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1290 unsigned int options)
1292 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1296 /*****************************************************************
1298 *****************************************************************/
1300 #define INDEX_FORMAT_DEFAULT 3
1302 static unsigned int get_index_format_default(void)
1304 char *envversion = getenv("GIT_INDEX_VERSION");
1307 unsigned int version = INDEX_FORMAT_DEFAULT;
1310 if (!git_config_get_int("index.version", &value))
1312 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1313 warning(_("index.version set, but the value is invalid.\n"
1314 "Using version %i"), INDEX_FORMAT_DEFAULT);
1315 return INDEX_FORMAT_DEFAULT;
1320 version = strtoul(envversion, &endp, 10);
1322 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1323 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1324 "Using version %i"), INDEX_FORMAT_DEFAULT);
1325 version = INDEX_FORMAT_DEFAULT;
1331 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1332 * Again - this is just a (very strong in practice) heuristic that
1333 * the inode hasn't changed.
1335 * We save the fields in big-endian order to allow using the
1336 * index file over NFS transparently.
1338 struct ondisk_cache_entry {
1339 struct cache_time ctime;
1340 struct cache_time mtime;
1347 unsigned char sha1[20];
1349 char name[FLEX_ARRAY]; /* more */
1353 * This struct is used when CE_EXTENDED bit is 1
1354 * The struct must match ondisk_cache_entry exactly from
1357 struct ondisk_cache_entry_extended {
1358 struct cache_time ctime;
1359 struct cache_time mtime;
1366 unsigned char sha1[20];
1369 char name[FLEX_ARRAY]; /* more */
1372 /* These are only used for v3 or lower */
1373 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1374 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1375 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1376 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1377 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1378 ondisk_cache_entry_size(ce_namelen(ce)))
1380 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1383 unsigned char sha1[20];
1386 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1387 return error("bad signature");
1388 hdr_version = ntohl(hdr->hdr_version);
1389 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1390 return error("bad index version %d", hdr_version);
1392 git_SHA1_Update(&c, hdr, size - 20);
1393 git_SHA1_Final(sha1, &c);
1394 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1395 return error("bad index file sha1 signature");
1399 static int read_index_extension(struct index_state *istate,
1400 const char *ext, void *data, unsigned long sz)
1402 switch (CACHE_EXT(ext)) {
1403 case CACHE_EXT_TREE:
1404 istate->cache_tree = cache_tree_read(data, sz);
1406 case CACHE_EXT_RESOLVE_UNDO:
1407 istate->resolve_undo = resolve_undo_read(data, sz);
1409 case CACHE_EXT_LINK:
1410 if (read_link_extension(istate, data, sz))
1413 case CACHE_EXT_UNTRACKED:
1414 istate->untracked = read_untracked_extension(data, sz);
1417 if (*ext < 'A' || 'Z' < *ext)
1418 return error("index uses %.4s extension, which we do not understand",
1420 fprintf(stderr, "ignoring %.4s extension\n", ext);
1426 int hold_locked_index(struct lock_file *lk, int die_on_error)
1428 return hold_lock_file_for_update(lk, get_index_file(),
1434 int read_index(struct index_state *istate)
1436 return read_index_from(istate, get_index_file());
1439 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1444 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1446 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1447 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1448 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1449 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1450 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1451 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1452 ce->ce_mode = get_be32(&ondisk->mode);
1453 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1454 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1455 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1456 ce->ce_flags = flags & ~CE_NAMEMASK;
1457 ce->ce_namelen = len;
1459 hashcpy(ce->sha1, ondisk->sha1);
1460 memcpy(ce->name, name, len);
1461 ce->name[len] = '\0';
1466 * Adjacent cache entries tend to share the leading paths, so it makes
1467 * sense to only store the differences in later entries. In the v4
1468 * on-disk format of the index, each on-disk cache entry stores the
1469 * number of bytes to be stripped from the end of the previous name,
1470 * and the bytes to append to the result, to come up with its name.
1472 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1474 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1475 size_t len = decode_varint(&cp);
1477 if (name->len < len)
1478 die("malformed name field in the index");
1479 strbuf_remove(name, name->len - len, len);
1480 for (ep = cp; *ep; ep++)
1481 ; /* find the end */
1482 strbuf_add(name, cp, ep - cp);
1483 return (const char *)ep + 1 - cp_;
1486 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1487 unsigned long *ent_size,
1488 struct strbuf *previous_name)
1490 struct cache_entry *ce;
1495 /* On-disk flags are just 16 bits */
1496 flags = get_be16(&ondisk->flags);
1497 len = flags & CE_NAMEMASK;
1499 if (flags & CE_EXTENDED) {
1500 struct ondisk_cache_entry_extended *ondisk2;
1502 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1503 extended_flags = get_be16(&ondisk2->flags2) << 16;
1504 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1505 if (extended_flags & ~CE_EXTENDED_FLAGS)
1506 die("Unknown index entry format %08x", extended_flags);
1507 flags |= extended_flags;
1508 name = ondisk2->name;
1511 name = ondisk->name;
1513 if (!previous_name) {
1514 /* v3 and earlier */
1515 if (len == CE_NAMEMASK)
1517 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1519 *ent_size = ondisk_ce_size(ce);
1521 unsigned long consumed;
1522 consumed = expand_name_field(previous_name, name);
1523 ce = cache_entry_from_ondisk(ondisk, flags,
1525 previous_name->len);
1527 *ent_size = (name - ((char *)ondisk)) + consumed;
1532 static void check_ce_order(struct index_state *istate)
1536 for (i = 1; i < istate->cache_nr; i++) {
1537 struct cache_entry *ce = istate->cache[i - 1];
1538 struct cache_entry *next_ce = istate->cache[i];
1539 int name_compare = strcmp(ce->name, next_ce->name);
1541 if (0 < name_compare)
1542 die("unordered stage entries in index");
1543 if (!name_compare) {
1545 die("multiple stage entries for merged file '%s'",
1547 if (ce_stage(ce) > ce_stage(next_ce))
1548 die("unordered stage entries for '%s'",
1554 static void tweak_untracked_cache(struct index_state *istate)
1556 switch (git_config_get_untracked_cache()) {
1557 case -1: /* keep: do nothing */
1560 remove_untracked_cache(istate);
1563 add_untracked_cache(istate);
1565 default: /* unknown value: do nothing */
1570 static void post_read_index_from(struct index_state *istate)
1572 check_ce_order(istate);
1573 tweak_untracked_cache(istate);
1576 /* remember to discard_cache() before reading a different cache! */
1577 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1581 unsigned long src_offset;
1582 struct cache_header *hdr;
1585 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1587 if (istate->initialized)
1588 return istate->cache_nr;
1590 istate->timestamp.sec = 0;
1591 istate->timestamp.nsec = 0;
1592 fd = open(path, O_RDONLY);
1594 if (!must_exist && errno == ENOENT)
1596 die_errno("%s: index file open failed", path);
1600 die_errno("cannot stat the open index");
1602 mmap_size = xsize_t(st.st_size);
1603 if (mmap_size < sizeof(struct cache_header) + 20)
1604 die("index file smaller than expected");
1606 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1607 if (mmap == MAP_FAILED)
1608 die_errno("unable to map index file");
1612 if (verify_hdr(hdr, mmap_size) < 0)
1615 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1616 istate->version = ntohl(hdr->hdr_version);
1617 istate->cache_nr = ntohl(hdr->hdr_entries);
1618 istate->cache_alloc = alloc_nr(istate->cache_nr);
1619 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1620 istate->initialized = 1;
1622 if (istate->version == 4)
1623 previous_name = &previous_name_buf;
1625 previous_name = NULL;
1627 src_offset = sizeof(*hdr);
1628 for (i = 0; i < istate->cache_nr; i++) {
1629 struct ondisk_cache_entry *disk_ce;
1630 struct cache_entry *ce;
1631 unsigned long consumed;
1633 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1634 ce = create_from_disk(disk_ce, &consumed, previous_name);
1635 set_index_entry(istate, i, ce);
1637 src_offset += consumed;
1639 strbuf_release(&previous_name_buf);
1640 istate->timestamp.sec = st.st_mtime;
1641 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1643 while (src_offset <= mmap_size - 20 - 8) {
1644 /* After an array of active_nr index entries,
1645 * there can be arbitrary number of extended
1646 * sections, each of which is prefixed with
1647 * extension name (4-byte) and section length
1648 * in 4-byte network byte order.
1651 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1652 extsize = ntohl(extsize);
1653 if (read_index_extension(istate,
1654 (const char *) mmap + src_offset,
1655 (char *) mmap + src_offset + 8,
1659 src_offset += extsize;
1661 munmap(mmap, mmap_size);
1662 return istate->cache_nr;
1665 munmap(mmap, mmap_size);
1666 die("index file corrupt");
1669 int read_index_from(struct index_state *istate, const char *path)
1671 struct split_index *split_index;
1674 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1675 if (istate->initialized)
1676 return istate->cache_nr;
1678 ret = do_read_index(istate, path, 0);
1680 split_index = istate->split_index;
1681 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1682 post_read_index_from(istate);
1686 if (split_index->base)
1687 discard_index(split_index->base);
1689 split_index->base = xcalloc(1, sizeof(*split_index->base));
1690 ret = do_read_index(split_index->base,
1691 git_path("sharedindex.%s",
1692 sha1_to_hex(split_index->base_sha1)), 1);
1693 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1694 die("broken index, expect %s in %s, got %s",
1695 sha1_to_hex(split_index->base_sha1),
1696 git_path("sharedindex.%s",
1697 sha1_to_hex(split_index->base_sha1)),
1698 sha1_to_hex(split_index->base->sha1));
1699 merge_base_index(istate);
1700 post_read_index_from(istate);
1704 int is_index_unborn(struct index_state *istate)
1706 return (!istate->cache_nr && !istate->timestamp.sec);
1709 int discard_index(struct index_state *istate)
1713 for (i = 0; i < istate->cache_nr; i++) {
1714 if (istate->cache[i]->index &&
1715 istate->split_index &&
1716 istate->split_index->base &&
1717 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1718 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1720 free(istate->cache[i]);
1722 resolve_undo_clear_index(istate);
1723 istate->cache_nr = 0;
1724 istate->cache_changed = 0;
1725 istate->timestamp.sec = 0;
1726 istate->timestamp.nsec = 0;
1727 free_name_hash(istate);
1728 cache_tree_free(&(istate->cache_tree));
1729 istate->initialized = 0;
1730 free(istate->cache);
1731 istate->cache = NULL;
1732 istate->cache_alloc = 0;
1733 discard_split_index(istate);
1734 free_untracked_cache(istate->untracked);
1735 istate->untracked = NULL;
1739 int unmerged_index(const struct index_state *istate)
1742 for (i = 0; i < istate->cache_nr; i++) {
1743 if (ce_stage(istate->cache[i]))
1749 #define WRITE_BUFFER_SIZE 8192
1750 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1751 static unsigned long write_buffer_len;
1753 static int ce_write_flush(git_SHA_CTX *context, int fd)
1755 unsigned int buffered = write_buffer_len;
1757 git_SHA1_Update(context, write_buffer, buffered);
1758 if (write_in_full(fd, write_buffer, buffered) != buffered)
1760 write_buffer_len = 0;
1765 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1768 unsigned int buffered = write_buffer_len;
1769 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1772 memcpy(write_buffer + buffered, data, partial);
1773 buffered += partial;
1774 if (buffered == WRITE_BUFFER_SIZE) {
1775 write_buffer_len = buffered;
1776 if (ce_write_flush(context, fd))
1780 write_buffer_len = buffered;
1782 data = (char *) data + partial;
1787 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1788 unsigned int ext, unsigned int sz)
1792 return ((ce_write(context, fd, &ext, 4) < 0) ||
1793 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1796 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1798 unsigned int left = write_buffer_len;
1801 write_buffer_len = 0;
1802 git_SHA1_Update(context, write_buffer, left);
1805 /* Flush first if not enough space for SHA1 signature */
1806 if (left + 20 > WRITE_BUFFER_SIZE) {
1807 if (write_in_full(fd, write_buffer, left) != left)
1812 /* Append the SHA1 signature at the end */
1813 git_SHA1_Final(write_buffer + left, context);
1814 hashcpy(sha1, write_buffer + left);
1816 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1819 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1822 * The only thing we care about in this function is to smudge the
1823 * falsely clean entry due to touch-update-touch race, so we leave
1824 * everything else as they are. We are called for entries whose
1825 * ce_stat_data.sd_mtime match the index file mtime.
1827 * Note that this actually does not do much for gitlinks, for
1828 * which ce_match_stat_basic() always goes to the actual
1829 * contents. The caller checks with is_racy_timestamp() which
1830 * always says "no" for gitlinks, so we are not called for them ;-)
1834 if (lstat(ce->name, &st) < 0)
1836 if (ce_match_stat_basic(ce, &st))
1838 if (ce_modified_check_fs(ce, &st)) {
1839 /* This is "racily clean"; smudge it. Note that this
1840 * is a tricky code. At first glance, it may appear
1841 * that it can break with this sequence:
1843 * $ echo xyzzy >frotz
1844 * $ git-update-index --add frotz
1847 * $ echo filfre >nitfol
1848 * $ git-update-index --add nitfol
1850 * but it does not. When the second update-index runs,
1851 * it notices that the entry "frotz" has the same timestamp
1852 * as index, and if we were to smudge it by resetting its
1853 * size to zero here, then the object name recorded
1854 * in index is the 6-byte file but the cached stat information
1855 * becomes zero --- which would then match what we would
1856 * obtain from the filesystem next time we stat("frotz").
1858 * However, the second update-index, before calling
1859 * this function, notices that the cached size is 6
1860 * bytes and what is on the filesystem is an empty
1861 * file, and never calls us, so the cached size information
1862 * for "frotz" stays 6 which does not match the filesystem.
1864 ce->ce_stat_data.sd_size = 0;
1868 /* Copy miscellaneous fields but not the name */
1869 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1870 struct cache_entry *ce)
1874 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1875 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1876 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1877 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1878 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1879 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1880 ondisk->mode = htonl(ce->ce_mode);
1881 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1882 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1883 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1884 hashcpy(ondisk->sha1, ce->sha1);
1886 flags = ce->ce_flags & ~CE_NAMEMASK;
1887 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1888 ondisk->flags = htons(flags);
1889 if (ce->ce_flags & CE_EXTENDED) {
1890 struct ondisk_cache_entry_extended *ondisk2;
1891 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1892 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1893 return ondisk2->name;
1896 return ondisk->name;
1900 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1901 struct strbuf *previous_name)
1904 struct ondisk_cache_entry *ondisk;
1905 int saved_namelen = saved_namelen; /* compiler workaround */
1909 if (ce->ce_flags & CE_STRIP_NAME) {
1910 saved_namelen = ce_namelen(ce);
1914 if (!previous_name) {
1915 size = ondisk_ce_size(ce);
1916 ondisk = xcalloc(1, size);
1917 name = copy_cache_entry_to_ondisk(ondisk, ce);
1918 memcpy(name, ce->name, ce_namelen(ce));
1920 int common, to_remove, prefix_size;
1921 unsigned char to_remove_vi[16];
1923 (ce->name[common] &&
1924 common < previous_name->len &&
1925 ce->name[common] == previous_name->buf[common]);
1927 ; /* still matching */
1928 to_remove = previous_name->len - common;
1929 prefix_size = encode_varint(to_remove, to_remove_vi);
1931 if (ce->ce_flags & CE_EXTENDED)
1932 size = offsetof(struct ondisk_cache_entry_extended, name);
1934 size = offsetof(struct ondisk_cache_entry, name);
1935 size += prefix_size + (ce_namelen(ce) - common + 1);
1937 ondisk = xcalloc(1, size);
1938 name = copy_cache_entry_to_ondisk(ondisk, ce);
1939 memcpy(name, to_remove_vi, prefix_size);
1940 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1942 strbuf_splice(previous_name, common, to_remove,
1943 ce->name + common, ce_namelen(ce) - common);
1945 if (ce->ce_flags & CE_STRIP_NAME) {
1946 ce->ce_namelen = saved_namelen;
1947 ce->ce_flags &= ~CE_STRIP_NAME;
1950 result = ce_write(c, fd, ondisk, size);
1956 * This function verifies if index_state has the correct sha1 of the
1957 * index file. Don't die if we have any other failure, just return 0.
1959 static int verify_index_from(const struct index_state *istate, const char *path)
1964 unsigned char sha1[20];
1966 if (!istate->initialized)
1969 fd = open(path, O_RDONLY);
1976 if (st.st_size < sizeof(struct cache_header) + 20)
1979 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1983 if (hashcmp(istate->sha1, sha1))
1994 static int verify_index(const struct index_state *istate)
1996 return verify_index_from(istate, get_index_file());
1999 static int has_racy_timestamp(struct index_state *istate)
2001 int entries = istate->cache_nr;
2004 for (i = 0; i < entries; i++) {
2005 struct cache_entry *ce = istate->cache[i];
2006 if (is_racy_timestamp(istate, ce))
2013 * Opportunistically update the index but do not complain if we can't
2015 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2017 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2018 verify_index(istate) &&
2019 write_locked_index(istate, lockfile, COMMIT_LOCK))
2020 rollback_lock_file(lockfile);
2023 static int do_write_index(struct index_state *istate, int newfd,
2024 int strip_extensions)
2027 struct cache_header hdr;
2028 int i, err, removed, extended, hdr_version;
2029 struct cache_entry **cache = istate->cache;
2030 int entries = istate->cache_nr;
2032 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2034 for (i = removed = extended = 0; i < entries; i++) {
2035 if (cache[i]->ce_flags & CE_REMOVE)
2038 /* reduce extended entries if possible */
2039 cache[i]->ce_flags &= ~CE_EXTENDED;
2040 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2042 cache[i]->ce_flags |= CE_EXTENDED;
2046 if (!istate->version) {
2047 istate->version = get_index_format_default();
2048 if (getenv("GIT_TEST_SPLIT_INDEX"))
2049 init_split_index(istate);
2052 /* demote version 3 to version 2 when the latter suffices */
2053 if (istate->version == 3 || istate->version == 2)
2054 istate->version = extended ? 3 : 2;
2056 hdr_version = istate->version;
2058 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2059 hdr.hdr_version = htonl(hdr_version);
2060 hdr.hdr_entries = htonl(entries - removed);
2063 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2066 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2067 for (i = 0; i < entries; i++) {
2068 struct cache_entry *ce = cache[i];
2069 if (ce->ce_flags & CE_REMOVE)
2071 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2072 ce_smudge_racily_clean_entry(ce);
2073 if (is_null_sha1(ce->sha1)) {
2074 static const char msg[] = "cache entry has null sha1: %s";
2075 static int allow = -1;
2078 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2080 warning(msg, ce->name);
2082 return error(msg, ce->name);
2084 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2087 strbuf_release(&previous_name_buf);
2089 /* Write extension data here */
2090 if (!strip_extensions && istate->split_index) {
2091 struct strbuf sb = STRBUF_INIT;
2093 err = write_link_extension(&sb, istate) < 0 ||
2094 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2096 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2097 strbuf_release(&sb);
2101 if (!strip_extensions && istate->cache_tree) {
2102 struct strbuf sb = STRBUF_INIT;
2104 cache_tree_write(&sb, istate->cache_tree);
2105 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2106 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2107 strbuf_release(&sb);
2111 if (!strip_extensions && istate->resolve_undo) {
2112 struct strbuf sb = STRBUF_INIT;
2114 resolve_undo_write(&sb, istate->resolve_undo);
2115 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2117 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2118 strbuf_release(&sb);
2122 if (!strip_extensions && istate->untracked) {
2123 struct strbuf sb = STRBUF_INIT;
2125 write_untracked_extension(&sb, istate->untracked);
2126 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2128 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2129 strbuf_release(&sb);
2134 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2136 istate->timestamp.sec = (unsigned int)st.st_mtime;
2137 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2141 void set_alternate_index_output(const char *name)
2143 alternate_index_output = name;
2146 static int commit_locked_index(struct lock_file *lk)
2148 if (alternate_index_output)
2149 return commit_lock_file_to(lk, alternate_index_output);
2151 return commit_lock_file(lk);
2154 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2157 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2160 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2161 (COMMIT_LOCK | CLOSE_LOCK));
2162 if (flags & COMMIT_LOCK)
2163 return commit_locked_index(lock);
2164 else if (flags & CLOSE_LOCK)
2165 return close_lock_file(lock);
2170 static int write_split_index(struct index_state *istate,
2171 struct lock_file *lock,
2175 prepare_to_write_split_index(istate);
2176 ret = do_write_locked_index(istate, lock, flags);
2177 finish_writing_split_index(istate);
2181 static struct tempfile temporary_sharedindex;
2183 static int write_shared_index(struct index_state *istate,
2184 struct lock_file *lock, unsigned flags)
2186 struct split_index *si = istate->split_index;
2189 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2191 hashclr(si->base_sha1);
2192 return do_write_locked_index(istate, lock, flags);
2194 move_cache_to_base_index(istate);
2195 ret = do_write_index(si->base, fd, 1);
2197 delete_tempfile(&temporary_sharedindex);
2200 ret = rename_tempfile(&temporary_sharedindex,
2201 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2203 hashcpy(si->base_sha1, si->base->sha1);
2207 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2210 struct split_index *si = istate->split_index;
2212 if (!si || alternate_index_output ||
2213 (istate->cache_changed & ~EXTMASK)) {
2215 hashclr(si->base_sha1);
2216 return do_write_locked_index(istate, lock, flags);
2219 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2220 int v = si->base_sha1[0];
2222 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2224 if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2225 int ret = write_shared_index(istate, lock, flags);
2230 return write_split_index(istate, lock, flags);
2234 * Read the index file that is potentially unmerged into given
2235 * index_state, dropping any unmerged entries. Returns true if
2236 * the index is unmerged. Callers who want to refuse to work
2237 * from an unmerged state can call this and check its return value,
2238 * instead of calling read_cache().
2240 int read_index_unmerged(struct index_state *istate)
2246 for (i = 0; i < istate->cache_nr; i++) {
2247 struct cache_entry *ce = istate->cache[i];
2248 struct cache_entry *new_ce;
2254 len = ce_namelen(ce);
2255 size = cache_entry_size(len);
2256 new_ce = xcalloc(1, size);
2257 memcpy(new_ce->name, ce->name, len);
2258 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2259 new_ce->ce_namelen = len;
2260 new_ce->ce_mode = ce->ce_mode;
2261 if (add_index_entry(istate, new_ce, 0))
2262 return error("%s: cannot drop to stage #0",
2269 * Returns 1 if the path is an "other" path with respect to
2270 * the index; that is, the path is not mentioned in the index at all,
2271 * either as a file, a directory with some files in the index,
2272 * or as an unmerged entry.
2274 * We helpfully remove a trailing "/" from directories so that
2275 * the output of read_directory can be used as-is.
2277 int index_name_is_other(const struct index_state *istate, const char *name,
2281 if (namelen && name[namelen - 1] == '/')
2283 pos = index_name_pos(istate, name, namelen);
2285 return 0; /* exact match */
2287 if (pos < istate->cache_nr) {
2288 struct cache_entry *ce = istate->cache[pos];
2289 if (ce_namelen(ce) == namelen &&
2290 !memcmp(ce->name, name, namelen))
2291 return 0; /* Yup, this one exists unmerged */
2296 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2300 enum object_type type;
2304 pos = index_name_pos(istate, path, len);
2307 * We might be in the middle of a merge, in which
2308 * case we would read stage #2 (ours).
2312 (pos < 0 && i < istate->cache_nr &&
2313 !strcmp(istate->cache[i]->name, path));
2315 if (ce_stage(istate->cache[i]) == 2)
2320 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2321 if (!data || type != OBJ_BLOB) {
2330 void stat_validity_clear(struct stat_validity *sv)
2336 int stat_validity_check(struct stat_validity *sv, const char *path)
2340 if (stat(path, &st) < 0)
2341 return sv->sd == NULL;
2344 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2347 void stat_validity_update(struct stat_validity *sv, int fd)
2351 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2352 stat_validity_clear(sv);
2355 sv->sd = xcalloc(1, sizeof(struct stat_data));
2356 fill_stat_data(sv->sd, &st);