2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
8 #include "cache-tree.h"
14 #include "resolve-undo.h"
16 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
20 * The first letter should be 'A'..'Z' for extensions that are not
21 * necessary for a correct operation (i.e. optimization data).
22 * When new extensions are added that _needs_ to be understood in
23 * order to correctly interpret the index file, pick character that
24 * is outside the range, to cause the reader to abort.
27 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
28 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
29 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
31 struct index_state the_index;
33 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
35 istate->cache[nr] = ce;
36 add_name_hash(istate, ce);
39 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
41 struct cache_entry *old = istate->cache[nr];
43 remove_name_hash(old);
44 set_index_entry(istate, nr, ce);
45 istate->cache_changed = 1;
48 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
50 struct cache_entry *old = istate->cache[nr], *new;
51 int namelen = strlen(new_name);
53 new = xmalloc(cache_entry_size(namelen));
54 copy_cache_entry(new, old);
55 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
56 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
57 memcpy(new->name, new_name, namelen + 1);
59 cache_tree_invalidate_path(istate->cache_tree, old->name);
60 remove_index_entry_at(istate, nr);
61 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
65 * This only updates the "non-critical" parts of the directory
66 * cache, ie the parts that aren't tracked by GIT, and only used
67 * to validate the cache.
69 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
71 ce->ce_ctime.sec = (unsigned int)st->st_ctime;
72 ce->ce_mtime.sec = (unsigned int)st->st_mtime;
73 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
74 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
75 ce->ce_dev = st->st_dev;
76 ce->ce_ino = st->st_ino;
77 ce->ce_uid = st->st_uid;
78 ce->ce_gid = st->st_gid;
79 ce->ce_size = st->st_size;
82 ce->ce_flags |= CE_VALID;
84 if (S_ISREG(st->st_mode))
88 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
91 int fd = open(ce->name, O_RDONLY);
94 unsigned char sha1[20];
95 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
96 match = hashcmp(sha1, ce->sha1);
97 /* index_fd() closed the file descriptor already */
102 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
107 enum object_type type;
108 struct strbuf sb = STRBUF_INIT;
110 if (strbuf_readlink(&sb, ce->name, expected_size))
113 buffer = read_sha1_file(ce->sha1, &type, &size);
116 match = memcmp(buffer, sb.buf, size);
123 static int ce_compare_gitlink(struct cache_entry *ce)
125 unsigned char sha1[20];
128 * We don't actually require that the .git directory
129 * under GITLINK directory be a valid git directory. It
130 * might even be missing (in case nobody populated that
133 * If so, we consider it always to match.
135 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
137 return hashcmp(sha1, ce->sha1);
140 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
142 switch (st->st_mode & S_IFMT) {
144 if (ce_compare_data(ce, st))
148 if (ce_compare_link(ce, xsize_t(st->st_size)))
152 if (S_ISGITLINK(ce->ce_mode))
153 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
160 static int is_empty_blob_sha1(const unsigned char *sha1)
162 static const unsigned char empty_blob_sha1[20] = {
163 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
164 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
167 return !hashcmp(sha1, empty_blob_sha1);
170 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
172 unsigned int changed = 0;
174 if (ce->ce_flags & CE_REMOVE)
175 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
177 switch (ce->ce_mode & S_IFMT) {
179 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
180 /* We consider only the owner x bit to be relevant for
183 if (trust_executable_bit &&
184 (0100 & (ce->ce_mode ^ st->st_mode)))
185 changed |= MODE_CHANGED;
188 if (!S_ISLNK(st->st_mode) &&
189 (has_symlinks || !S_ISREG(st->st_mode)))
190 changed |= TYPE_CHANGED;
193 /* We ignore most of the st_xxx fields for gitlinks */
194 if (!S_ISDIR(st->st_mode))
195 changed |= TYPE_CHANGED;
196 else if (ce_compare_gitlink(ce))
197 changed |= DATA_CHANGED;
200 die("internal error: ce_mode is %o", ce->ce_mode);
202 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
203 changed |= MTIME_CHANGED;
204 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
205 changed |= CTIME_CHANGED;
208 if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
209 changed |= MTIME_CHANGED;
210 if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
211 changed |= CTIME_CHANGED;
214 if (ce->ce_uid != (unsigned int) st->st_uid ||
215 ce->ce_gid != (unsigned int) st->st_gid)
216 changed |= OWNER_CHANGED;
217 if (ce->ce_ino != (unsigned int) st->st_ino)
218 changed |= INODE_CHANGED;
222 * st_dev breaks on network filesystems where different
223 * clients will have different views of what "device"
224 * the filesystem is on
226 if (ce->ce_dev != (unsigned int) st->st_dev)
227 changed |= INODE_CHANGED;
230 if (ce->ce_size != (unsigned int) st->st_size)
231 changed |= DATA_CHANGED;
233 /* Racily smudged entry? */
235 if (!is_empty_blob_sha1(ce->sha1))
236 changed |= DATA_CHANGED;
242 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
244 return (!S_ISGITLINK(ce->ce_mode) &&
245 istate->timestamp.sec &&
247 /* nanosecond timestamped files can also be racy! */
248 (istate->timestamp.sec < ce->ce_mtime.sec ||
249 (istate->timestamp.sec == ce->ce_mtime.sec &&
250 istate->timestamp.nsec <= ce->ce_mtime.nsec))
252 istate->timestamp.sec <= ce->ce_mtime.sec
257 int ie_match_stat(const struct index_state *istate,
258 struct cache_entry *ce, struct stat *st,
259 unsigned int options)
261 unsigned int changed;
262 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
263 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
264 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
267 * If it's marked as always valid in the index, it's
268 * valid whatever the checked-out copy says.
270 * skip-worktree has the same effect with higher precedence
272 if (!ignore_skip_worktree && ce_skip_worktree(ce))
274 if (!ignore_valid && (ce->ce_flags & CE_VALID))
278 * Intent-to-add entries have not been added, so the index entry
279 * by definition never matches what is in the work tree until it
280 * actually gets added.
282 if (ce->ce_flags & CE_INTENT_TO_ADD)
283 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
285 changed = ce_match_stat_basic(ce, st);
288 * Within 1 second of this sequence:
289 * echo xyzzy >file && git-update-index --add file
290 * running this command:
292 * would give a falsely clean cache entry. The mtime and
293 * length match the cache, and other stat fields do not change.
295 * We could detect this at update-index time (the cache entry
296 * being registered/updated records the same time as "now")
297 * and delay the return from git-update-index, but that would
298 * effectively mean we can make at most one commit per second,
299 * which is not acceptable. Instead, we check cache entries
300 * whose mtime are the same as the index file timestamp more
301 * carefully than others.
303 if (!changed && is_racy_timestamp(istate, ce)) {
304 if (assume_racy_is_modified)
305 changed |= DATA_CHANGED;
307 changed |= ce_modified_check_fs(ce, st);
313 int ie_modified(const struct index_state *istate,
314 struct cache_entry *ce, struct stat *st, unsigned int options)
316 int changed, changed_fs;
318 changed = ie_match_stat(istate, ce, st, options);
322 * If the mode or type has changed, there's no point in trying
323 * to refresh the entry - it's not going to match
325 if (changed & (MODE_CHANGED | TYPE_CHANGED))
329 * Immediately after read-tree or update-index --cacheinfo,
330 * the length field is zero, as we have never even read the
331 * lstat(2) information once, and we cannot trust DATA_CHANGED
332 * returned by ie_match_stat() which in turn was returned by
333 * ce_match_stat_basic() to signal that the filesize of the
334 * blob changed. We have to actually go to the filesystem to
335 * see if the contents match, and if so, should answer "unchanged".
337 * The logic does not apply to gitlinks, as ce_match_stat_basic()
338 * already has checked the actual HEAD from the filesystem in the
339 * subproject. If ie_match_stat() already said it is different,
340 * then we know it is.
342 if ((changed & DATA_CHANGED) &&
343 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
346 changed_fs = ce_modified_check_fs(ce, st);
348 return changed | changed_fs;
352 int base_name_compare(const char *name1, int len1, int mode1,
353 const char *name2, int len2, int mode2)
355 unsigned char c1, c2;
356 int len = len1 < len2 ? len1 : len2;
359 cmp = memcmp(name1, name2, len);
364 if (!c1 && S_ISDIR(mode1))
366 if (!c2 && S_ISDIR(mode2))
368 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
372 * df_name_compare() is identical to base_name_compare(), except it
373 * compares conflicting directory/file entries as equal. Note that
374 * while a directory name compares as equal to a regular file, they
375 * then individually compare _differently_ to a filename that has
376 * a dot after the basename (because '\0' < '.' < '/').
378 * This is used by routines that want to traverse the git namespace
379 * but then handle conflicting entries together when possible.
381 int df_name_compare(const char *name1, int len1, int mode1,
382 const char *name2, int len2, int mode2)
384 int len = len1 < len2 ? len1 : len2, cmp;
385 unsigned char c1, c2;
387 cmp = memcmp(name1, name2, len);
390 /* Directories and files compare equal (same length, same name) */
394 if (!c1 && S_ISDIR(mode1))
397 if (!c2 && S_ISDIR(mode2))
399 if (c1 == '/' && !c2)
401 if (c2 == '/' && !c1)
406 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
408 int len1 = flags1 & CE_NAMEMASK;
409 int len2 = flags2 & CE_NAMEMASK;
410 int len = len1 < len2 ? len1 : len2;
413 cmp = memcmp(name1, name2, len);
422 flags1 &= CE_STAGEMASK;
423 flags2 &= CE_STAGEMASK;
432 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
437 last = istate->cache_nr;
438 while (last > first) {
439 int next = (last + first) >> 1;
440 struct cache_entry *ce = istate->cache[next];
441 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
453 /* Remove entry, return true if there are more entries to go.. */
454 int remove_index_entry_at(struct index_state *istate, int pos)
456 struct cache_entry *ce = istate->cache[pos];
458 record_resolve_undo(istate, ce);
459 remove_name_hash(ce);
460 istate->cache_changed = 1;
462 if (pos >= istate->cache_nr)
464 memmove(istate->cache + pos,
465 istate->cache + pos + 1,
466 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
471 * Remove all cache ententries marked for removal, that is where
472 * CE_REMOVE is set in ce_flags. This is much more effective than
473 * calling remove_index_entry_at() for each entry to be removed.
475 void remove_marked_cache_entries(struct index_state *istate)
477 struct cache_entry **ce_array = istate->cache;
480 for (i = j = 0; i < istate->cache_nr; i++) {
481 if (ce_array[i]->ce_flags & CE_REMOVE)
482 remove_name_hash(ce_array[i]);
484 ce_array[j++] = ce_array[i];
486 istate->cache_changed = 1;
487 istate->cache_nr = j;
490 int remove_file_from_index(struct index_state *istate, const char *path)
492 int pos = index_name_pos(istate, path, strlen(path));
495 cache_tree_invalidate_path(istate->cache_tree, path);
496 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
497 remove_index_entry_at(istate, pos);
501 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
503 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
506 static int index_name_pos_also_unmerged(struct index_state *istate,
507 const char *path, int namelen)
509 int pos = index_name_pos(istate, path, namelen);
510 struct cache_entry *ce;
515 /* maybe unmerged? */
517 if (pos >= istate->cache_nr ||
518 compare_name((ce = istate->cache[pos]), path, namelen))
521 /* order of preference: stage 2, 1, 3 */
522 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
523 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
524 !compare_name(ce, path, namelen))
529 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
531 int len = ce_namelen(ce);
532 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
536 * If we add a filename that aliases in the cache, we will use the
537 * name that we already have - but we don't want to update the same
538 * alias twice, because that implies that there were actually two
539 * different files with aliasing names!
541 * So we use the CE_ADDED flag to verify that the alias was an old
542 * one before we accept it as
544 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
547 struct cache_entry *new;
549 if (alias->ce_flags & CE_ADDED)
550 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
552 /* Ok, create the new entry using the name of the existing alias */
553 len = ce_namelen(alias);
554 new = xcalloc(1, cache_entry_size(len));
555 memcpy(new->name, alias->name, len);
556 copy_cache_entry(new, ce);
561 static void record_intent_to_add(struct cache_entry *ce)
563 unsigned char sha1[20];
564 if (write_sha1_file("", 0, blob_type, sha1))
565 die("cannot create an empty blob in the object database");
566 hashcpy(ce->sha1, sha1);
569 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
571 int size, namelen, was_same;
572 mode_t st_mode = st->st_mode;
573 struct cache_entry *ce, *alias;
574 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
575 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
576 int pretend = flags & ADD_CACHE_PRETEND;
577 int intent_only = flags & ADD_CACHE_INTENT;
578 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
579 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
581 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
582 return error("%s: can only add regular files, symbolic links or git-directories", path);
584 namelen = strlen(path);
585 if (S_ISDIR(st_mode)) {
586 while (namelen && path[namelen-1] == '/')
589 size = cache_entry_size(namelen);
590 ce = xcalloc(1, size);
591 memcpy(ce->name, path, namelen);
592 ce->ce_flags = namelen;
594 fill_stat_cache_info(ce, st);
596 ce->ce_flags |= CE_INTENT_TO_ADD;
598 if (trust_executable_bit && has_symlinks)
599 ce->ce_mode = create_ce_mode(st_mode);
601 /* If there is an existing entry, pick the mode bits and type
602 * from it, otherwise assume unexecutable regular file.
604 struct cache_entry *ent;
605 int pos = index_name_pos_also_unmerged(istate, path, namelen);
607 ent = (0 <= pos) ? istate->cache[pos] : NULL;
608 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
611 /* When core.ignorecase=true, determine if a directory of the same name but differing
612 * case already exists within the Git repository. If it does, ensure the directory
613 * case of the file being added to the repository matches (is folded into) the existing
614 * entry's directory case.
617 const char *startPtr = ce->name;
618 const char *ptr = startPtr;
620 while (*ptr && *ptr != '/')
623 struct cache_entry *foundce;
625 foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case);
627 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
634 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
635 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
636 /* Nothing changed, really */
638 if (!S_ISGITLINK(alias->ce_mode))
639 ce_mark_uptodate(alias);
640 alias->ce_flags |= CE_ADDED;
644 if (index_path(ce->sha1, path, st, 1))
645 return error("unable to index file %s", path);
647 record_intent_to_add(ce);
649 if (ignore_case && alias && different_name(ce, alias))
650 ce = create_alias_ce(ce, alias);
651 ce->ce_flags |= CE_ADDED;
653 /* It was suspected to be racily clean, but it turns out to be Ok */
656 !hashcmp(alias->sha1, ce->sha1) &&
657 ce->ce_mode == alias->ce_mode);
661 else if (add_index_entry(istate, ce, add_option))
662 return error("unable to add %s to index",path);
663 if (verbose && !was_same)
664 printf("add '%s'\n", path);
668 int add_file_to_index(struct index_state *istate, const char *path, int flags)
671 if (lstat(path, &st))
672 die_errno("unable to stat '%s'", path);
673 return add_to_index(istate, path, &st, flags);
676 struct cache_entry *make_cache_entry(unsigned int mode,
677 const unsigned char *sha1, const char *path, int stage,
681 struct cache_entry *ce;
683 if (!verify_path(path)) {
684 error("Invalid path '%s'", path);
689 size = cache_entry_size(len);
690 ce = xcalloc(1, size);
692 hashcpy(ce->sha1, sha1);
693 memcpy(ce->name, path, len);
694 ce->ce_flags = create_ce_flags(len, stage);
695 ce->ce_mode = create_ce_mode(mode);
698 return refresh_cache_entry(ce, 0);
703 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
705 int len = ce_namelen(a);
706 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
709 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
711 const char *match, *name;
717 len = ce_namelen(ce);
719 while ((match = *pathspec++) != NULL) {
720 int matchlen = strlen(match);
723 if (memcmp(name, match, matchlen))
725 if (matchlen && name[matchlen-1] == '/')
727 if (name[matchlen] == '/' || !name[matchlen])
736 * We fundamentally don't like some paths: we don't want
737 * dot or dot-dot anywhere, and for obvious reasons don't
738 * want to recurse into ".git" either.
740 * Also, we don't want double slashes or slashes at the
741 * end that can make pathnames ambiguous.
743 static int verify_dotfile(const char *rest)
746 * The first character was '.', but that
747 * has already been discarded, we now test
751 /* "." is not allowed */
756 * ".git" followed by NUL or slash is bad. This
757 * shares the path end test with the ".." case.
767 if (rest[1] == '\0' || rest[1] == '/')
773 int verify_path(const char *path)
777 if (has_dos_drive_prefix(path))
787 if ((c == '.' && !verify_dotfile(path)) ||
788 is_dir_sep(c) || c == '\0')
796 * Do we have another file that has the beginning components being a
797 * proper superset of the name we're trying to add?
799 static int has_file_name(struct index_state *istate,
800 const struct cache_entry *ce, int pos, int ok_to_replace)
803 int len = ce_namelen(ce);
804 int stage = ce_stage(ce);
805 const char *name = ce->name;
807 while (pos < istate->cache_nr) {
808 struct cache_entry *p = istate->cache[pos++];
810 if (len >= ce_namelen(p))
812 if (memcmp(name, p->name, len))
814 if (ce_stage(p) != stage)
816 if (p->name[len] != '/')
818 if (p->ce_flags & CE_REMOVE)
823 remove_index_entry_at(istate, --pos);
829 * Do we have another file with a pathname that is a proper
830 * subset of the name we're trying to add?
832 static int has_dir_name(struct index_state *istate,
833 const struct cache_entry *ce, int pos, int ok_to_replace)
836 int stage = ce_stage(ce);
837 const char *name = ce->name;
838 const char *slash = name + ce_namelen(ce);
846 if (slash <= ce->name)
851 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
854 * Found one, but not so fast. This could
855 * be a marker that says "I was here, but
856 * I am being removed". Such an entry is
857 * not a part of the resulting tree, and
858 * it is Ok to have a directory at the same
861 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
865 remove_index_entry_at(istate, pos);
873 * Trivial optimization: if we find an entry that
874 * already matches the sub-directory, then we know
875 * we're ok, and we can exit.
877 while (pos < istate->cache_nr) {
878 struct cache_entry *p = istate->cache[pos];
879 if ((ce_namelen(p) <= len) ||
880 (p->name[len] != '/') ||
881 memcmp(p->name, name, len))
882 break; /* not our subdirectory */
883 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
885 * p is at the same stage as our entry, and
886 * is a subdirectory of what we are looking
887 * at, so we cannot have conflicts at our
888 * level or anything shorter.
897 /* We may be in a situation where we already have path/file and path
898 * is being added, or we already have path and path/file is being
899 * added. Either one would result in a nonsense tree that has path
900 * twice when git-write-tree tries to write it out. Prevent it.
902 * If ok-to-replace is specified, we remove the conflicting entries
903 * from the cache so the caller should recompute the insert position.
904 * When this happens, we return non-zero.
906 static int check_file_directory_conflict(struct index_state *istate,
907 const struct cache_entry *ce,
908 int pos, int ok_to_replace)
913 * When ce is an "I am going away" entry, we allow it to be added
915 if (ce->ce_flags & CE_REMOVE)
919 * We check if the path is a sub-path of a subsequent pathname
920 * first, since removing those will not change the position
923 retval = has_file_name(istate, ce, pos, ok_to_replace);
926 * Then check if the path might have a clashing sub-directory
929 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
932 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
935 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
936 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
937 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
938 int new_only = option & ADD_CACHE_NEW_ONLY;
940 cache_tree_invalidate_path(istate->cache_tree, ce->name);
941 pos = index_name_pos(istate, ce->name, ce->ce_flags);
943 /* existing match? Just replace it. */
946 replace_index_entry(istate, pos, ce);
952 * Inserting a merged entry ("stage 0") into the index
953 * will always replace all non-merged entries..
955 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
956 while (ce_same_name(istate->cache[pos], ce)) {
958 if (!remove_index_entry_at(istate, pos))
965 if (!verify_path(ce->name))
966 return error("Invalid path '%s'", ce->name);
968 if (!skip_df_check &&
969 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
971 return error("'%s' appears as both a file and as a directory",
973 pos = index_name_pos(istate, ce->name, ce->ce_flags);
979 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
983 if (option & ADD_CACHE_JUST_APPEND)
984 pos = istate->cache_nr;
987 ret = add_index_entry_with_check(istate, ce, option);
993 /* Make sure the array is big enough .. */
994 if (istate->cache_nr == istate->cache_alloc) {
995 istate->cache_alloc = alloc_nr(istate->cache_alloc);
996 istate->cache = xrealloc(istate->cache,
997 istate->cache_alloc * sizeof(struct cache_entry *));
1002 if (istate->cache_nr > pos + 1)
1003 memmove(istate->cache + pos + 1,
1004 istate->cache + pos,
1005 (istate->cache_nr - pos - 1) * sizeof(ce));
1006 set_index_entry(istate, pos, ce);
1007 istate->cache_changed = 1;
1012 * "refresh" does not calculate a new sha1 file or bring the
1013 * cache up-to-date for mode/content changes. But what it
1014 * _does_ do is to "re-match" the stat information of a file
1015 * with the cache, so that you can refresh the cache for a
1016 * file that hasn't been changed but where the stat entry is
1019 * For example, you'd want to do this after doing a "git-read-tree",
1020 * to link up the stat cache details with the proper files.
1022 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1023 struct cache_entry *ce,
1024 unsigned int options, int *err)
1027 struct cache_entry *updated;
1029 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1030 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1032 if (ce_uptodate(ce))
1036 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1037 * that the change to the work tree does not matter and told
1040 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1041 ce_mark_uptodate(ce);
1044 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1045 ce_mark_uptodate(ce);
1049 if (lstat(ce->name, &st) < 0) {
1055 changed = ie_match_stat(istate, ce, &st, options);
1058 * The path is unchanged. If we were told to ignore
1059 * valid bit, then we did the actual stat check and
1060 * found that the entry is unmodified. If the entry
1061 * is not marked VALID, this is the place to mark it
1062 * valid again, under "assume unchanged" mode.
1064 if (ignore_valid && assume_unchanged &&
1065 !(ce->ce_flags & CE_VALID))
1066 ; /* mark this one VALID again */
1069 * We do not mark the index itself "modified"
1070 * because CE_UPTODATE flag is in-core only;
1071 * we are not going to write this change out.
1073 if (!S_ISGITLINK(ce->ce_mode))
1074 ce_mark_uptodate(ce);
1079 if (ie_modified(istate, ce, &st, options)) {
1086 updated = xmalloc(size);
1087 memcpy(updated, ce, size);
1088 fill_stat_cache_info(updated, &st);
1090 * If ignore_valid is not set, we should leave CE_VALID bit
1091 * alone. Otherwise, paths marked with --no-assume-unchanged
1092 * (i.e. things to be edited) will reacquire CE_VALID bit
1093 * automatically, which is not really what we want.
1095 if (!ignore_valid && assume_unchanged &&
1096 !(ce->ce_flags & CE_VALID))
1097 updated->ce_flags &= ~CE_VALID;
1102 static void show_file(const char * fmt, const char * name, int in_porcelain,
1103 int * first, char *header_msg)
1105 if (in_porcelain && *first && header_msg) {
1106 printf("%s\n", header_msg);
1112 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1113 char *seen, char *header_msg)
1117 int really = (flags & REFRESH_REALLY) != 0;
1118 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1119 int quiet = (flags & REFRESH_QUIET) != 0;
1120 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1121 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1123 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1124 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1125 const char *needs_update_fmt;
1126 const char *needs_merge_fmt;
1128 needs_update_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1129 needs_merge_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1130 for (i = 0; i < istate->cache_nr; i++) {
1131 struct cache_entry *ce, *new;
1132 int cache_errno = 0;
1134 ce = istate->cache[i];
1135 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1139 while ((i < istate->cache_nr) &&
1140 ! strcmp(istate->cache[i]->name, ce->name))
1145 show_file(needs_merge_fmt, ce->name, in_porcelain, &first, header_msg);
1150 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1153 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1157 if (not_new && cache_errno == ENOENT)
1159 if (really && cache_errno == EINVAL) {
1160 /* If we are doing --really-refresh that
1161 * means the index is not valid anymore.
1163 ce->ce_flags &= ~CE_VALID;
1164 istate->cache_changed = 1;
1168 show_file(needs_update_fmt, ce->name, in_porcelain, &first, header_msg);
1173 replace_index_entry(istate, i, new);
1178 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1180 return refresh_cache_ent(&the_index, ce, really, NULL);
1183 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1186 unsigned char sha1[20];
1188 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1189 return error("bad signature");
1190 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1191 return error("bad index version");
1193 git_SHA1_Update(&c, hdr, size - 20);
1194 git_SHA1_Final(sha1, &c);
1195 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1196 return error("bad index file sha1 signature");
1200 static int read_index_extension(struct index_state *istate,
1201 const char *ext, void *data, unsigned long sz)
1203 switch (CACHE_EXT(ext)) {
1204 case CACHE_EXT_TREE:
1205 istate->cache_tree = cache_tree_read(data, sz);
1207 case CACHE_EXT_RESOLVE_UNDO:
1208 istate->resolve_undo = resolve_undo_read(data, sz);
1211 if (*ext < 'A' || 'Z' < *ext)
1212 return error("index uses %.4s extension, which we do not understand",
1214 fprintf(stderr, "ignoring %.4s extension\n", ext);
1220 int read_index(struct index_state *istate)
1222 return read_index_from(istate, get_index_file());
1225 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1230 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1231 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1232 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1233 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1234 ce->ce_dev = ntohl(ondisk->dev);
1235 ce->ce_ino = ntohl(ondisk->ino);
1236 ce->ce_mode = ntohl(ondisk->mode);
1237 ce->ce_uid = ntohl(ondisk->uid);
1238 ce->ce_gid = ntohl(ondisk->gid);
1239 ce->ce_size = ntohl(ondisk->size);
1240 /* On-disk flags are just 16 bits */
1241 ce->ce_flags = ntohs(ondisk->flags);
1243 hashcpy(ce->sha1, ondisk->sha1);
1245 len = ce->ce_flags & CE_NAMEMASK;
1247 if (ce->ce_flags & CE_EXTENDED) {
1248 struct ondisk_cache_entry_extended *ondisk2;
1250 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1251 extended_flags = ntohs(ondisk2->flags2) << 16;
1252 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1253 if (extended_flags & ~CE_EXTENDED_FLAGS)
1254 die("Unknown index entry format %08x", extended_flags);
1255 ce->ce_flags |= extended_flags;
1256 name = ondisk2->name;
1259 name = ondisk->name;
1261 if (len == CE_NAMEMASK)
1264 * NEEDSWORK: If the original index is crafted, this copy could
1267 memcpy(ce->name, name, len + 1);
1270 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1274 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1277 * Alignment can cause differences. This should be "alignof", but
1278 * since that's a gcc'ism, just use the size of a pointer.
1280 per_entry += sizeof(void *);
1281 return ondisk_size + entries*per_entry;
1284 /* remember to discard_cache() before reading a different cache! */
1285 int read_index_from(struct index_state *istate, const char *path)
1289 unsigned long src_offset, dst_offset;
1290 struct cache_header *hdr;
1295 if (istate->initialized)
1296 return istate->cache_nr;
1299 istate->timestamp.sec = 0;
1300 istate->timestamp.nsec = 0;
1301 fd = open(path, O_RDONLY);
1303 if (errno == ENOENT)
1305 die_errno("index file open failed");
1309 die_errno("cannot stat the open index");
1312 mmap_size = xsize_t(st.st_size);
1313 if (mmap_size < sizeof(struct cache_header) + 20)
1314 die("index file smaller than expected");
1316 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1318 if (mmap == MAP_FAILED)
1319 die_errno("unable to map index file");
1322 if (verify_hdr(hdr, mmap_size) < 0)
1325 istate->cache_nr = ntohl(hdr->hdr_entries);
1326 istate->cache_alloc = alloc_nr(istate->cache_nr);
1327 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1330 * The disk format is actually larger than the in-memory format,
1331 * due to space for nsec etc, so even though the in-memory one
1332 * has room for a few more flags, we can allocate using the same
1335 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1336 istate->initialized = 1;
1338 src_offset = sizeof(*hdr);
1340 for (i = 0; i < istate->cache_nr; i++) {
1341 struct ondisk_cache_entry *disk_ce;
1342 struct cache_entry *ce;
1344 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1345 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1346 convert_from_disk(disk_ce, ce);
1347 set_index_entry(istate, i, ce);
1349 src_offset += ondisk_ce_size(ce);
1350 dst_offset += ce_size(ce);
1352 istate->timestamp.sec = st.st_mtime;
1353 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1355 while (src_offset <= mmap_size - 20 - 8) {
1356 /* After an array of active_nr index entries,
1357 * there can be arbitrary number of extended
1358 * sections, each of which is prefixed with
1359 * extension name (4-byte) and section length
1360 * in 4-byte network byte order.
1363 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1364 extsize = ntohl(extsize);
1365 if (read_index_extension(istate,
1366 (const char *) mmap + src_offset,
1367 (char *) mmap + src_offset + 8,
1371 src_offset += extsize;
1373 munmap(mmap, mmap_size);
1374 return istate->cache_nr;
1377 munmap(mmap, mmap_size);
1379 die("index file corrupt");
1382 int is_index_unborn(struct index_state *istate)
1384 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
1387 int discard_index(struct index_state *istate)
1389 resolve_undo_clear_index(istate);
1390 istate->cache_nr = 0;
1391 istate->cache_changed = 0;
1392 istate->timestamp.sec = 0;
1393 istate->timestamp.nsec = 0;
1394 istate->name_hash_initialized = 0;
1395 free_hash(&istate->name_hash);
1396 cache_tree_free(&(istate->cache_tree));
1397 free(istate->alloc);
1398 istate->alloc = NULL;
1399 istate->initialized = 0;
1401 /* no need to throw away allocated active_cache */
1405 int unmerged_index(const struct index_state *istate)
1408 for (i = 0; i < istate->cache_nr; i++) {
1409 if (ce_stage(istate->cache[i]))
1415 #define WRITE_BUFFER_SIZE 8192
1416 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1417 static unsigned long write_buffer_len;
1419 static int ce_write_flush(git_SHA_CTX *context, int fd)
1421 unsigned int buffered = write_buffer_len;
1423 git_SHA1_Update(context, write_buffer, buffered);
1424 if (write_in_full(fd, write_buffer, buffered) != buffered)
1426 write_buffer_len = 0;
1431 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1434 unsigned int buffered = write_buffer_len;
1435 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1438 memcpy(write_buffer + buffered, data, partial);
1439 buffered += partial;
1440 if (buffered == WRITE_BUFFER_SIZE) {
1441 write_buffer_len = buffered;
1442 if (ce_write_flush(context, fd))
1446 write_buffer_len = buffered;
1448 data = (char *) data + partial;
1453 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1454 unsigned int ext, unsigned int sz)
1458 return ((ce_write(context, fd, &ext, 4) < 0) ||
1459 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1462 static int ce_flush(git_SHA_CTX *context, int fd)
1464 unsigned int left = write_buffer_len;
1467 write_buffer_len = 0;
1468 git_SHA1_Update(context, write_buffer, left);
1471 /* Flush first if not enough space for SHA1 signature */
1472 if (left + 20 > WRITE_BUFFER_SIZE) {
1473 if (write_in_full(fd, write_buffer, left) != left)
1478 /* Append the SHA1 signature at the end */
1479 git_SHA1_Final(write_buffer + left, context);
1481 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1484 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1487 * The only thing we care about in this function is to smudge the
1488 * falsely clean entry due to touch-update-touch race, so we leave
1489 * everything else as they are. We are called for entries whose
1490 * ce_mtime match the index file mtime.
1492 * Note that this actually does not do much for gitlinks, for
1493 * which ce_match_stat_basic() always goes to the actual
1494 * contents. The caller checks with is_racy_timestamp() which
1495 * always says "no" for gitlinks, so we are not called for them ;-)
1499 if (lstat(ce->name, &st) < 0)
1501 if (ce_match_stat_basic(ce, &st))
1503 if (ce_modified_check_fs(ce, &st)) {
1504 /* This is "racily clean"; smudge it. Note that this
1505 * is a tricky code. At first glance, it may appear
1506 * that it can break with this sequence:
1508 * $ echo xyzzy >frotz
1509 * $ git-update-index --add frotz
1512 * $ echo filfre >nitfol
1513 * $ git-update-index --add nitfol
1515 * but it does not. When the second update-index runs,
1516 * it notices that the entry "frotz" has the same timestamp
1517 * as index, and if we were to smudge it by resetting its
1518 * size to zero here, then the object name recorded
1519 * in index is the 6-byte file but the cached stat information
1520 * becomes zero --- which would then match what we would
1521 * obtain from the filesystem next time we stat("frotz").
1523 * However, the second update-index, before calling
1524 * this function, notices that the cached size is 6
1525 * bytes and what is on the filesystem is an empty
1526 * file, and never calls us, so the cached size information
1527 * for "frotz" stays 6 which does not match the filesystem.
1533 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1535 int size = ondisk_ce_size(ce);
1536 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1540 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1541 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1542 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1543 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1544 ondisk->dev = htonl(ce->ce_dev);
1545 ondisk->ino = htonl(ce->ce_ino);
1546 ondisk->mode = htonl(ce->ce_mode);
1547 ondisk->uid = htonl(ce->ce_uid);
1548 ondisk->gid = htonl(ce->ce_gid);
1549 ondisk->size = htonl(ce->ce_size);
1550 hashcpy(ondisk->sha1, ce->sha1);
1551 ondisk->flags = htons(ce->ce_flags);
1552 if (ce->ce_flags & CE_EXTENDED) {
1553 struct ondisk_cache_entry_extended *ondisk2;
1554 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1555 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1556 name = ondisk2->name;
1559 name = ondisk->name;
1560 memcpy(name, ce->name, ce_namelen(ce));
1562 result = ce_write(c, fd, ondisk, size);
1567 static int has_racy_timestamp(struct index_state *istate)
1569 int entries = istate->cache_nr;
1572 for (i = 0; i < entries; i++) {
1573 struct cache_entry *ce = istate->cache[i];
1574 if (is_racy_timestamp(istate, ce))
1581 * Opportunisticly update the index but do not complain if we can't
1583 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1585 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1586 !write_index(istate, lockfile->fd))
1587 commit_locked_index(lockfile);
1589 rollback_lock_file(lockfile);
1592 int write_index(struct index_state *istate, int newfd)
1595 struct cache_header hdr;
1596 int i, err, removed, extended;
1597 struct cache_entry **cache = istate->cache;
1598 int entries = istate->cache_nr;
1601 for (i = removed = extended = 0; i < entries; i++) {
1602 if (cache[i]->ce_flags & CE_REMOVE)
1605 /* reduce extended entries if possible */
1606 cache[i]->ce_flags &= ~CE_EXTENDED;
1607 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1609 cache[i]->ce_flags |= CE_EXTENDED;
1613 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1614 /* for extended format, increase version so older git won't try to read it */
1615 hdr.hdr_version = htonl(extended ? 3 : 2);
1616 hdr.hdr_entries = htonl(entries - removed);
1619 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1622 for (i = 0; i < entries; i++) {
1623 struct cache_entry *ce = cache[i];
1624 if (ce->ce_flags & CE_REMOVE)
1626 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1627 ce_smudge_racily_clean_entry(ce);
1628 if (ce_write_entry(&c, newfd, ce) < 0)
1632 /* Write extension data here */
1633 if (istate->cache_tree) {
1634 struct strbuf sb = STRBUF_INIT;
1636 cache_tree_write(&sb, istate->cache_tree);
1637 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1638 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1639 strbuf_release(&sb);
1643 if (istate->resolve_undo) {
1644 struct strbuf sb = STRBUF_INIT;
1646 resolve_undo_write(&sb, istate->resolve_undo);
1647 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1649 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1650 strbuf_release(&sb);
1655 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1657 istate->timestamp.sec = (unsigned int)st.st_mtime;
1658 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1663 * Read the index file that is potentially unmerged into given
1664 * index_state, dropping any unmerged entries. Returns true if
1665 * the index is unmerged. Callers who want to refuse to work
1666 * from an unmerged state can call this and check its return value,
1667 * instead of calling read_cache().
1669 int read_index_unmerged(struct index_state *istate)
1675 for (i = 0; i < istate->cache_nr; i++) {
1676 struct cache_entry *ce = istate->cache[i];
1677 struct cache_entry *new_ce;
1683 len = strlen(ce->name);
1684 size = cache_entry_size(len);
1685 new_ce = xcalloc(1, size);
1686 memcpy(new_ce->name, ce->name, len);
1687 new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;
1688 new_ce->ce_mode = ce->ce_mode;
1689 if (add_index_entry(istate, new_ce, 0))
1690 return error("%s: cannot drop to stage #0",
1692 i = index_name_pos(istate, new_ce->name, len);
1698 * Returns 1 if the path is an "other" path with respect to
1699 * the index; that is, the path is not mentioned in the index at all,
1700 * either as a file, a directory with some files in the index,
1701 * or as an unmerged entry.
1703 * We helpfully remove a trailing "/" from directories so that
1704 * the output of read_directory can be used as-is.
1706 int index_name_is_other(const struct index_state *istate, const char *name,
1710 if (namelen && name[namelen - 1] == '/')
1712 pos = index_name_pos(istate, name, namelen);
1714 return 0; /* exact match */
1716 if (pos < istate->cache_nr) {
1717 struct cache_entry *ce = istate->cache[pos];
1718 if (ce_namelen(ce) == namelen &&
1719 !memcmp(ce->name, name, namelen))
1720 return 0; /* Yup, this one exists unmerged */