Merge branch 'pm/build-persistent-https-with-recent-go' into maint
[git/debian.git] / read-cache.c
blobdb2776605529bd43f77ca48e84b9e2ded95b493e
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "tempfile.h"
9 #include "lockfile.h"
10 #include "cache-tree.h"
11 #include "refs.h"
12 #include "dir.h"
13 #include "tree.h"
14 #include "commit.h"
15 #include "blob.h"
16 #include "resolve-undo.h"
17 #include "strbuf.h"
18 #include "varint.h"
19 #include "split-index.h"
20 #include "utf8.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)
29 /* Index extensions.
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);
64 free(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;
79 new->index = 0;
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)
103 int changed = 0;
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;
111 #ifdef USE_NSEC
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;
117 #endif
119 if (check_stat) {
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;
127 #ifdef USE_STDEV
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;
135 #endif
137 if (sd->sd_size != (unsigned int) st->st_size)
138 changed |= DATA_CHANGED;
140 return 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)
161 int match = -1;
162 int fd = open(ce->name, O_RDONLY);
164 if (fd >= 0) {
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 */
170 return match;
173 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
175 int match = -1;
176 void *buffer;
177 unsigned long size;
178 enum object_type type;
179 struct strbuf sb = STRBUF_INIT;
181 if (strbuf_readlink(&sb, ce->name, expected_size))
182 return -1;
184 buffer = read_sha1_file(ce->sha1, &type, &size);
185 if (buffer) {
186 if (size == sb.len)
187 match = memcmp(buffer, sb.buf, size);
188 free(buffer);
190 strbuf_release(&sb);
191 return match;
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
202 * sub-project).
204 * If so, we consider it always to match.
206 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
207 return 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) {
214 case S_IFREG:
215 if (ce_compare_data(ce, st))
216 return DATA_CHANGED;
217 break;
218 case S_IFLNK:
219 if (ce_compare_link(ce, xsize_t(st->st_size)))
220 return DATA_CHANGED;
221 break;
222 case S_IFDIR:
223 if (S_ISGITLINK(ce->ce_mode))
224 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
225 default:
226 return TYPE_CHANGED;
228 return 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) {
239 case S_IFREG:
240 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
241 /* We consider only the owner x bit to be relevant for
242 * "mode changes"
244 if (trust_executable_bit &&
245 (0100 & (ce->ce_mode ^ st->st_mode)))
246 changed |= MODE_CHANGED;
247 break;
248 case S_IFLNK:
249 if (!S_ISLNK(st->st_mode) &&
250 (has_symlinks || !S_ISREG(st->st_mode)))
251 changed |= TYPE_CHANGED;
252 break;
253 case S_IFGITLINK:
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;
259 return changed;
260 default:
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;
272 return changed;
275 static int is_racy_stat(const struct index_state *istate,
276 const struct stat_data *sd)
278 return (istate->timestamp.sec &&
279 #ifdef USE_NSEC
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))
284 #else
285 istate->timestamp.sec <= sd->sd_mtime.sec
286 #endif
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))
321 return 0;
322 if (!ignore_valid && (ce->ce_flags & CE_VALID))
323 return 0;
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:
339 * echo frotz >file
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;
354 else
355 changed |= ce_modified_check_fs(ce, st);
358 return changed;
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);
368 if (!changed)
369 return 0;
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))
375 return 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))
393 return changed;
395 changed_fs = ce_modified_check_fs(ce, st);
396 if (changed_fs)
397 return changed | changed_fs;
398 return 0;
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;
406 int cmp;
408 cmp = memcmp(name1, name2, len);
409 if (cmp)
410 return cmp;
411 c1 = name1[len];
412 c2 = name2[len];
413 if (!c1 && S_ISDIR(mode1))
414 c1 = '/';
415 if (!c2 && S_ISDIR(mode2))
416 c2 = '/';
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);
437 if (cmp)
438 return cmp;
439 /* Directories and files compare equal (same length, same name) */
440 if (len1 == len2)
441 return 0;
442 c1 = name1[len];
443 if (!c1 && S_ISDIR(mode1))
444 c1 = '/';
445 c2 = name2[len];
446 if (!c2 && S_ISDIR(mode2))
447 c2 = '/';
448 if (c1 == '/' && !c2)
449 return 0;
450 if (c2 == '/' && !c1)
451 return 0;
452 return c1 - c2;
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);
459 if (cmp)
460 return cmp;
461 if (len1 < len2)
462 return -1;
463 if (len1 > len2)
464 return 1;
465 return 0;
468 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
470 int cmp;
472 cmp = name_compare(name1, len1, name2, len2);
473 if (cmp)
474 return cmp;
476 if (stage1 < stage2)
477 return -1;
478 if (stage1 > stage2)
479 return 1;
480 return 0;
483 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
485 int first, last;
487 first = 0;
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));
493 if (!cmp)
494 return next;
495 if (cmp < 0) {
496 last = next;
497 continue;
499 first = next+1;
501 return -first-1;
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;
518 istate->cache_nr--;
519 if (pos >= istate->cache_nr)
520 return 0;
521 memmove(istate->cache + pos,
522 istate->cache + pos + 1,
523 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
524 return 1;
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;
535 unsigned int i, j;
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]);
542 else
543 ce_array[j++] = ce_array[i];
545 if (j == istate->cache_nr)
546 return;
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));
554 if (pos < 0)
555 pos = -pos-1;
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);
560 return 0;
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;
574 if (pos >= 0)
575 return pos;
577 /* maybe unmerged? */
578 pos = -1 - pos;
579 if (pos >= istate->cache_nr ||
580 compare_name((ce = istate->cache[pos]), path, namelen))
581 return -1;
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))
587 pos++;
588 return pos;
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)
610 int len;
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);
622 return new;
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] == '/')
651 namelen--;
653 size = cache_entry_size(namelen);
654 ce = xcalloc(1, size);
655 memcpy(ce->name, path, namelen);
656 ce->ce_namelen = namelen;
657 if (!intent_only)
658 fill_stat_cache_info(ce, st);
659 else
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);
666 else {
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.
682 if (ignore_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;
693 free(ce);
694 return 0;
696 if (!intent_only) {
697 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) {
698 free(ce);
699 return error("unable to index file %s", path);
701 } else
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 */
709 was_same = (alias &&
710 !ce_stage(alias) &&
711 !hashcmp(alias->sha1, ce->sha1) &&
712 ce->ce_mode == alias->ce_mode);
714 if (pretend)
715 free(ce);
716 else if (add_index_entry(istate, ce, add_option)) {
717 free(ce);
718 return error("unable to add %s to index", path);
720 if (verbose && !was_same)
721 printf("add '%s'\n", path);
722 return 0;
725 int add_file_to_index(struct index_state *istate, const char *path,
726 int flags, int force_mode)
728 struct stat st;
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)
738 int size, len;
739 struct cache_entry *ce, *ret;
741 if (!verify_path(path)) {
742 error("Invalid path '%s'", path);
743 return NULL;
746 len = strlen(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);
757 if (ret != ce)
758 free(ce);
759 return ret;
762 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
764 int len = ce_namelen(a);
765 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
769 * We fundamentally don't like some paths: we don't want
770 * dot or dot-dot anywhere, and for obvious reasons don't
771 * want to recurse into ".git" either.
773 * Also, we don't want double slashes or slashes at the
774 * end that can make pathnames ambiguous.
776 static int verify_dotfile(const char *rest)
779 * The first character was '.', but that
780 * has already been discarded, we now test
781 * the rest.
784 /* "." is not allowed */
785 if (*rest == '\0' || is_dir_sep(*rest))
786 return 0;
788 switch (*rest) {
790 * ".git" followed by NUL or slash is bad. This
791 * shares the path end test with the ".." case.
793 case 'g':
794 case 'G':
795 if (rest[1] != 'i' && rest[1] != 'I')
796 break;
797 if (rest[2] != 't' && rest[2] != 'T')
798 break;
799 rest += 2;
800 /* fallthrough */
801 case '.':
802 if (rest[1] == '\0' || is_dir_sep(rest[1]))
803 return 0;
805 return 1;
808 int verify_path(const char *path)
810 char c;
812 if (has_dos_drive_prefix(path))
813 return 0;
815 goto inside;
816 for (;;) {
817 if (!c)
818 return 1;
819 if (is_dir_sep(c)) {
820 inside:
821 if (protect_hfs && is_hfs_dotgit(path))
822 return 0;
823 if (protect_ntfs && is_ntfs_dotgit(path))
824 return 0;
825 c = *path++;
826 if ((c == '.' && !verify_dotfile(path)) ||
827 is_dir_sep(c) || c == '\0')
828 return 0;
830 c = *path++;
835 * Do we have another file that has the beginning components being a
836 * proper superset of the name we're trying to add?
838 static int has_file_name(struct index_state *istate,
839 const struct cache_entry *ce, int pos, int ok_to_replace)
841 int retval = 0;
842 int len = ce_namelen(ce);
843 int stage = ce_stage(ce);
844 const char *name = ce->name;
846 while (pos < istate->cache_nr) {
847 struct cache_entry *p = istate->cache[pos++];
849 if (len >= ce_namelen(p))
850 break;
851 if (memcmp(name, p->name, len))
852 break;
853 if (ce_stage(p) != stage)
854 continue;
855 if (p->name[len] != '/')
856 continue;
857 if (p->ce_flags & CE_REMOVE)
858 continue;
859 retval = -1;
860 if (!ok_to_replace)
861 break;
862 remove_index_entry_at(istate, --pos);
864 return retval;
868 * Do we have another file with a pathname that is a proper
869 * subset of the name we're trying to add?
871 static int has_dir_name(struct index_state *istate,
872 const struct cache_entry *ce, int pos, int ok_to_replace)
874 int retval = 0;
875 int stage = ce_stage(ce);
876 const char *name = ce->name;
877 const char *slash = name + ce_namelen(ce);
879 for (;;) {
880 int len;
882 for (;;) {
883 if (*--slash == '/')
884 break;
885 if (slash <= ce->name)
886 return retval;
888 len = slash - name;
890 pos = index_name_stage_pos(istate, name, len, stage);
891 if (pos >= 0) {
893 * Found one, but not so fast. This could
894 * be a marker that says "I was here, but
895 * I am being removed". Such an entry is
896 * not a part of the resulting tree, and
897 * it is Ok to have a directory at the same
898 * path.
900 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
901 retval = -1;
902 if (!ok_to_replace)
903 break;
904 remove_index_entry_at(istate, pos);
905 continue;
908 else
909 pos = -pos-1;
912 * Trivial optimization: if we find an entry that
913 * already matches the sub-directory, then we know
914 * we're ok, and we can exit.
916 while (pos < istate->cache_nr) {
917 struct cache_entry *p = istate->cache[pos];
918 if ((ce_namelen(p) <= len) ||
919 (p->name[len] != '/') ||
920 memcmp(p->name, name, len))
921 break; /* not our subdirectory */
922 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
924 * p is at the same stage as our entry, and
925 * is a subdirectory of what we are looking
926 * at, so we cannot have conflicts at our
927 * level or anything shorter.
929 return retval;
930 pos++;
933 return retval;
936 /* We may be in a situation where we already have path/file and path
937 * is being added, or we already have path and path/file is being
938 * added. Either one would result in a nonsense tree that has path
939 * twice when git-write-tree tries to write it out. Prevent it.
941 * If ok-to-replace is specified, we remove the conflicting entries
942 * from the cache so the caller should recompute the insert position.
943 * When this happens, we return non-zero.
945 static int check_file_directory_conflict(struct index_state *istate,
946 const struct cache_entry *ce,
947 int pos, int ok_to_replace)
949 int retval;
952 * When ce is an "I am going away" entry, we allow it to be added
954 if (ce->ce_flags & CE_REMOVE)
955 return 0;
958 * We check if the path is a sub-path of a subsequent pathname
959 * first, since removing those will not change the position
960 * in the array.
962 retval = has_file_name(istate, ce, pos, ok_to_replace);
965 * Then check if the path might have a clashing sub-directory
966 * before it.
968 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
971 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
973 int pos;
974 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
975 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
976 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
977 int new_only = option & ADD_CACHE_NEW_ONLY;
979 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
980 cache_tree_invalidate_path(istate, ce->name);
981 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
983 /* existing match? Just replace it. */
984 if (pos >= 0) {
985 if (!new_only)
986 replace_index_entry(istate, pos, ce);
987 return 0;
989 pos = -pos-1;
991 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
992 untracked_cache_add_to_index(istate, ce->name);
995 * Inserting a merged entry ("stage 0") into the index
996 * will always replace all non-merged entries..
998 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
999 while (ce_same_name(istate->cache[pos], ce)) {
1000 ok_to_add = 1;
1001 if (!remove_index_entry_at(istate, pos))
1002 break;
1006 if (!ok_to_add)
1007 return -1;
1008 if (!verify_path(ce->name))
1009 return error("Invalid path '%s'", ce->name);
1011 if (!skip_df_check &&
1012 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1013 if (!ok_to_replace)
1014 return error("'%s' appears as both a file and as a directory",
1015 ce->name);
1016 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1017 pos = -pos-1;
1019 return pos + 1;
1022 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1024 int pos;
1026 if (option & ADD_CACHE_JUST_APPEND)
1027 pos = istate->cache_nr;
1028 else {
1029 int ret;
1030 ret = add_index_entry_with_check(istate, ce, option);
1031 if (ret <= 0)
1032 return ret;
1033 pos = ret - 1;
1036 /* Make sure the array is big enough .. */
1037 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1039 /* Add it in.. */
1040 istate->cache_nr++;
1041 if (istate->cache_nr > pos + 1)
1042 memmove(istate->cache + pos + 1,
1043 istate->cache + pos,
1044 (istate->cache_nr - pos - 1) * sizeof(ce));
1045 set_index_entry(istate, pos, ce);
1046 istate->cache_changed |= CE_ENTRY_ADDED;
1047 return 0;
1051 * "refresh" does not calculate a new sha1 file or bring the
1052 * cache up-to-date for mode/content changes. But what it
1053 * _does_ do is to "re-match" the stat information of a file
1054 * with the cache, so that you can refresh the cache for a
1055 * file that hasn't been changed but where the stat entry is
1056 * out of date.
1058 * For example, you'd want to do this after doing a "git-read-tree",
1059 * to link up the stat cache details with the proper files.
1061 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1062 struct cache_entry *ce,
1063 unsigned int options, int *err,
1064 int *changed_ret)
1066 struct stat st;
1067 struct cache_entry *updated;
1068 int changed, size;
1069 int refresh = options & CE_MATCH_REFRESH;
1070 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1071 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1072 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1074 if (!refresh || ce_uptodate(ce))
1075 return ce;
1078 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1079 * that the change to the work tree does not matter and told
1080 * us not to worry.
1082 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1083 ce_mark_uptodate(ce);
1084 return ce;
1086 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1087 ce_mark_uptodate(ce);
1088 return ce;
1091 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1092 if (ignore_missing)
1093 return ce;
1094 if (err)
1095 *err = ENOENT;
1096 return NULL;
1099 if (lstat(ce->name, &st) < 0) {
1100 if (ignore_missing && errno == ENOENT)
1101 return ce;
1102 if (err)
1103 *err = errno;
1104 return NULL;
1107 changed = ie_match_stat(istate, ce, &st, options);
1108 if (changed_ret)
1109 *changed_ret = changed;
1110 if (!changed) {
1112 * The path is unchanged. If we were told to ignore
1113 * valid bit, then we did the actual stat check and
1114 * found that the entry is unmodified. If the entry
1115 * is not marked VALID, this is the place to mark it
1116 * valid again, under "assume unchanged" mode.
1118 if (ignore_valid && assume_unchanged &&
1119 !(ce->ce_flags & CE_VALID))
1120 ; /* mark this one VALID again */
1121 else {
1123 * We do not mark the index itself "modified"
1124 * because CE_UPTODATE flag is in-core only;
1125 * we are not going to write this change out.
1127 if (!S_ISGITLINK(ce->ce_mode))
1128 ce_mark_uptodate(ce);
1129 return ce;
1133 if (ie_modified(istate, ce, &st, options)) {
1134 if (err)
1135 *err = EINVAL;
1136 return NULL;
1139 size = ce_size(ce);
1140 updated = xmalloc(size);
1141 memcpy(updated, ce, size);
1142 fill_stat_cache_info(updated, &st);
1144 * If ignore_valid is not set, we should leave CE_VALID bit
1145 * alone. Otherwise, paths marked with --no-assume-unchanged
1146 * (i.e. things to be edited) will reacquire CE_VALID bit
1147 * automatically, which is not really what we want.
1149 if (!ignore_valid && assume_unchanged &&
1150 !(ce->ce_flags & CE_VALID))
1151 updated->ce_flags &= ~CE_VALID;
1153 /* istate->cache_changed is updated in the caller */
1154 return updated;
1157 static void show_file(const char * fmt, const char * name, int in_porcelain,
1158 int * first, const char *header_msg)
1160 if (in_porcelain && *first && header_msg) {
1161 printf("%s\n", header_msg);
1162 *first = 0;
1164 printf(fmt, name);
1167 int refresh_index(struct index_state *istate, unsigned int flags,
1168 const struct pathspec *pathspec,
1169 char *seen, const char *header_msg)
1171 int i;
1172 int has_errors = 0;
1173 int really = (flags & REFRESH_REALLY) != 0;
1174 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1175 int quiet = (flags & REFRESH_QUIET) != 0;
1176 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1177 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1178 int first = 1;
1179 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1180 unsigned int options = (CE_MATCH_REFRESH |
1181 (really ? CE_MATCH_IGNORE_VALID : 0) |
1182 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1183 const char *modified_fmt;
1184 const char *deleted_fmt;
1185 const char *typechange_fmt;
1186 const char *added_fmt;
1187 const char *unmerged_fmt;
1189 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1190 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1191 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1192 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1193 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1194 for (i = 0; i < istate->cache_nr; i++) {
1195 struct cache_entry *ce, *new;
1196 int cache_errno = 0;
1197 int changed = 0;
1198 int filtered = 0;
1200 ce = istate->cache[i];
1201 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1202 continue;
1204 if (pathspec && !ce_path_match(ce, pathspec, seen))
1205 filtered = 1;
1207 if (ce_stage(ce)) {
1208 while ((i < istate->cache_nr) &&
1209 ! strcmp(istate->cache[i]->name, ce->name))
1210 i++;
1211 i--;
1212 if (allow_unmerged)
1213 continue;
1214 if (!filtered)
1215 show_file(unmerged_fmt, ce->name, in_porcelain,
1216 &first, header_msg);
1217 has_errors = 1;
1218 continue;
1221 if (filtered)
1222 continue;
1224 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1225 if (new == ce)
1226 continue;
1227 if (!new) {
1228 const char *fmt;
1230 if (really && cache_errno == EINVAL) {
1231 /* If we are doing --really-refresh that
1232 * means the index is not valid anymore.
1234 ce->ce_flags &= ~CE_VALID;
1235 ce->ce_flags |= CE_UPDATE_IN_BASE;
1236 istate->cache_changed |= CE_ENTRY_CHANGED;
1238 if (quiet)
1239 continue;
1241 if (cache_errno == ENOENT)
1242 fmt = deleted_fmt;
1243 else if (ce_intent_to_add(ce))
1244 fmt = added_fmt; /* must be before other checks */
1245 else if (changed & TYPE_CHANGED)
1246 fmt = typechange_fmt;
1247 else
1248 fmt = modified_fmt;
1249 show_file(fmt,
1250 ce->name, in_porcelain, &first, header_msg);
1251 has_errors = 1;
1252 continue;
1255 replace_index_entry(istate, i, new);
1257 return has_errors;
1260 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1261 unsigned int options)
1263 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1267 /*****************************************************************
1268 * Index File I/O
1269 *****************************************************************/
1271 #define INDEX_FORMAT_DEFAULT 3
1273 static unsigned int get_index_format_default(void)
1275 char *envversion = getenv("GIT_INDEX_VERSION");
1276 char *endp;
1277 int value;
1278 unsigned int version = INDEX_FORMAT_DEFAULT;
1280 if (!envversion) {
1281 if (!git_config_get_int("index.version", &value))
1282 version = value;
1283 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1284 warning(_("index.version set, but the value is invalid.\n"
1285 "Using version %i"), INDEX_FORMAT_DEFAULT);
1286 return INDEX_FORMAT_DEFAULT;
1288 return version;
1291 version = strtoul(envversion, &endp, 10);
1292 if (*endp ||
1293 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1294 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1295 "Using version %i"), INDEX_FORMAT_DEFAULT);
1296 version = INDEX_FORMAT_DEFAULT;
1298 return version;
1302 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1303 * Again - this is just a (very strong in practice) heuristic that
1304 * the inode hasn't changed.
1306 * We save the fields in big-endian order to allow using the
1307 * index file over NFS transparently.
1309 struct ondisk_cache_entry {
1310 struct cache_time ctime;
1311 struct cache_time mtime;
1312 uint32_t dev;
1313 uint32_t ino;
1314 uint32_t mode;
1315 uint32_t uid;
1316 uint32_t gid;
1317 uint32_t size;
1318 unsigned char sha1[20];
1319 uint16_t flags;
1320 char name[FLEX_ARRAY]; /* more */
1324 * This struct is used when CE_EXTENDED bit is 1
1325 * The struct must match ondisk_cache_entry exactly from
1326 * ctime till flags
1328 struct ondisk_cache_entry_extended {
1329 struct cache_time ctime;
1330 struct cache_time mtime;
1331 uint32_t dev;
1332 uint32_t ino;
1333 uint32_t mode;
1334 uint32_t uid;
1335 uint32_t gid;
1336 uint32_t size;
1337 unsigned char sha1[20];
1338 uint16_t flags;
1339 uint16_t flags2;
1340 char name[FLEX_ARRAY]; /* more */
1343 /* These are only used for v3 or lower */
1344 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1345 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1346 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1347 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1348 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1349 ondisk_cache_entry_size(ce_namelen(ce)))
1351 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1353 git_SHA_CTX c;
1354 unsigned char sha1[20];
1355 int hdr_version;
1357 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1358 return error("bad signature");
1359 hdr_version = ntohl(hdr->hdr_version);
1360 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1361 return error("bad index version %d", hdr_version);
1362 git_SHA1_Init(&c);
1363 git_SHA1_Update(&c, hdr, size - 20);
1364 git_SHA1_Final(sha1, &c);
1365 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1366 return error("bad index file sha1 signature");
1367 return 0;
1370 static int read_index_extension(struct index_state *istate,
1371 const char *ext, void *data, unsigned long sz)
1373 switch (CACHE_EXT(ext)) {
1374 case CACHE_EXT_TREE:
1375 istate->cache_tree = cache_tree_read(data, sz);
1376 break;
1377 case CACHE_EXT_RESOLVE_UNDO:
1378 istate->resolve_undo = resolve_undo_read(data, sz);
1379 break;
1380 case CACHE_EXT_LINK:
1381 if (read_link_extension(istate, data, sz))
1382 return -1;
1383 break;
1384 case CACHE_EXT_UNTRACKED:
1385 istate->untracked = read_untracked_extension(data, sz);
1386 break;
1387 default:
1388 if (*ext < 'A' || 'Z' < *ext)
1389 return error("index uses %.4s extension, which we do not understand",
1390 ext);
1391 fprintf(stderr, "ignoring %.4s extension\n", ext);
1392 break;
1394 return 0;
1397 int hold_locked_index(struct lock_file *lk, int die_on_error)
1399 return hold_lock_file_for_update(lk, get_index_file(),
1400 die_on_error
1401 ? LOCK_DIE_ON_ERROR
1402 : 0);
1405 int read_index(struct index_state *istate)
1407 return read_index_from(istate, get_index_file());
1410 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1411 unsigned int flags,
1412 const char *name,
1413 size_t len)
1415 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1417 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1418 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1419 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1420 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1421 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1422 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1423 ce->ce_mode = get_be32(&ondisk->mode);
1424 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1425 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1426 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1427 ce->ce_flags = flags & ~CE_NAMEMASK;
1428 ce->ce_namelen = len;
1429 ce->index = 0;
1430 hashcpy(ce->sha1, ondisk->sha1);
1431 memcpy(ce->name, name, len);
1432 ce->name[len] = '\0';
1433 return ce;
1437 * Adjacent cache entries tend to share the leading paths, so it makes
1438 * sense to only store the differences in later entries. In the v4
1439 * on-disk format of the index, each on-disk cache entry stores the
1440 * number of bytes to be stripped from the end of the previous name,
1441 * and the bytes to append to the result, to come up with its name.
1443 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1445 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1446 size_t len = decode_varint(&cp);
1448 if (name->len < len)
1449 die("malformed name field in the index");
1450 strbuf_remove(name, name->len - len, len);
1451 for (ep = cp; *ep; ep++)
1452 ; /* find the end */
1453 strbuf_add(name, cp, ep - cp);
1454 return (const char *)ep + 1 - cp_;
1457 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1458 unsigned long *ent_size,
1459 struct strbuf *previous_name)
1461 struct cache_entry *ce;
1462 size_t len;
1463 const char *name;
1464 unsigned int flags;
1466 /* On-disk flags are just 16 bits */
1467 flags = get_be16(&ondisk->flags);
1468 len = flags & CE_NAMEMASK;
1470 if (flags & CE_EXTENDED) {
1471 struct ondisk_cache_entry_extended *ondisk2;
1472 int extended_flags;
1473 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1474 extended_flags = get_be16(&ondisk2->flags2) << 16;
1475 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1476 if (extended_flags & ~CE_EXTENDED_FLAGS)
1477 die("Unknown index entry format %08x", extended_flags);
1478 flags |= extended_flags;
1479 name = ondisk2->name;
1481 else
1482 name = ondisk->name;
1484 if (!previous_name) {
1485 /* v3 and earlier */
1486 if (len == CE_NAMEMASK)
1487 len = strlen(name);
1488 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1490 *ent_size = ondisk_ce_size(ce);
1491 } else {
1492 unsigned long consumed;
1493 consumed = expand_name_field(previous_name, name);
1494 ce = cache_entry_from_ondisk(ondisk, flags,
1495 previous_name->buf,
1496 previous_name->len);
1498 *ent_size = (name - ((char *)ondisk)) + consumed;
1500 return ce;
1503 static void check_ce_order(struct index_state *istate)
1505 unsigned int i;
1507 for (i = 1; i < istate->cache_nr; i++) {
1508 struct cache_entry *ce = istate->cache[i - 1];
1509 struct cache_entry *next_ce = istate->cache[i];
1510 int name_compare = strcmp(ce->name, next_ce->name);
1512 if (0 < name_compare)
1513 die("unordered stage entries in index");
1514 if (!name_compare) {
1515 if (!ce_stage(ce))
1516 die("multiple stage entries for merged file '%s'",
1517 ce->name);
1518 if (ce_stage(ce) > ce_stage(next_ce))
1519 die("unordered stage entries for '%s'",
1520 ce->name);
1525 static void tweak_untracked_cache(struct index_state *istate)
1527 switch (git_config_get_untracked_cache()) {
1528 case -1: /* keep: do nothing */
1529 break;
1530 case 0: /* false */
1531 remove_untracked_cache(istate);
1532 break;
1533 case 1: /* true */
1534 add_untracked_cache(istate);
1535 break;
1536 default: /* unknown value: do nothing */
1537 break;
1541 static void post_read_index_from(struct index_state *istate)
1543 check_ce_order(istate);
1544 tweak_untracked_cache(istate);
1547 /* remember to discard_cache() before reading a different cache! */
1548 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1550 int fd, i;
1551 struct stat st;
1552 unsigned long src_offset;
1553 struct cache_header *hdr;
1554 void *mmap;
1555 size_t mmap_size;
1556 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1558 if (istate->initialized)
1559 return istate->cache_nr;
1561 istate->timestamp.sec = 0;
1562 istate->timestamp.nsec = 0;
1563 fd = open(path, O_RDONLY);
1564 if (fd < 0) {
1565 if (!must_exist && errno == ENOENT)
1566 return 0;
1567 die_errno("%s: index file open failed", path);
1570 if (fstat(fd, &st))
1571 die_errno("cannot stat the open index");
1573 mmap_size = xsize_t(st.st_size);
1574 if (mmap_size < sizeof(struct cache_header) + 20)
1575 die("index file smaller than expected");
1577 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1578 if (mmap == MAP_FAILED)
1579 die_errno("unable to map index file");
1580 close(fd);
1582 hdr = mmap;
1583 if (verify_hdr(hdr, mmap_size) < 0)
1584 goto unmap;
1586 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1587 istate->version = ntohl(hdr->hdr_version);
1588 istate->cache_nr = ntohl(hdr->hdr_entries);
1589 istate->cache_alloc = alloc_nr(istate->cache_nr);
1590 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1591 istate->initialized = 1;
1593 if (istate->version == 4)
1594 previous_name = &previous_name_buf;
1595 else
1596 previous_name = NULL;
1598 src_offset = sizeof(*hdr);
1599 for (i = 0; i < istate->cache_nr; i++) {
1600 struct ondisk_cache_entry *disk_ce;
1601 struct cache_entry *ce;
1602 unsigned long consumed;
1604 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1605 ce = create_from_disk(disk_ce, &consumed, previous_name);
1606 set_index_entry(istate, i, ce);
1608 src_offset += consumed;
1610 strbuf_release(&previous_name_buf);
1611 istate->timestamp.sec = st.st_mtime;
1612 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1614 while (src_offset <= mmap_size - 20 - 8) {
1615 /* After an array of active_nr index entries,
1616 * there can be arbitrary number of extended
1617 * sections, each of which is prefixed with
1618 * extension name (4-byte) and section length
1619 * in 4-byte network byte order.
1621 uint32_t extsize;
1622 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1623 extsize = ntohl(extsize);
1624 if (read_index_extension(istate,
1625 (const char *) mmap + src_offset,
1626 (char *) mmap + src_offset + 8,
1627 extsize) < 0)
1628 goto unmap;
1629 src_offset += 8;
1630 src_offset += extsize;
1632 munmap(mmap, mmap_size);
1633 return istate->cache_nr;
1635 unmap:
1636 munmap(mmap, mmap_size);
1637 die("index file corrupt");
1640 int read_index_from(struct index_state *istate, const char *path)
1642 struct split_index *split_index;
1643 int ret;
1645 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1646 if (istate->initialized)
1647 return istate->cache_nr;
1649 ret = do_read_index(istate, path, 0);
1651 split_index = istate->split_index;
1652 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1653 post_read_index_from(istate);
1654 return ret;
1657 if (split_index->base)
1658 discard_index(split_index->base);
1659 else
1660 split_index->base = xcalloc(1, sizeof(*split_index->base));
1661 ret = do_read_index(split_index->base,
1662 git_path("sharedindex.%s",
1663 sha1_to_hex(split_index->base_sha1)), 1);
1664 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1665 die("broken index, expect %s in %s, got %s",
1666 sha1_to_hex(split_index->base_sha1),
1667 git_path("sharedindex.%s",
1668 sha1_to_hex(split_index->base_sha1)),
1669 sha1_to_hex(split_index->base->sha1));
1670 merge_base_index(istate);
1671 post_read_index_from(istate);
1672 return ret;
1675 int is_index_unborn(struct index_state *istate)
1677 return (!istate->cache_nr && !istate->timestamp.sec);
1680 int discard_index(struct index_state *istate)
1682 int i;
1684 for (i = 0; i < istate->cache_nr; i++) {
1685 if (istate->cache[i]->index &&
1686 istate->split_index &&
1687 istate->split_index->base &&
1688 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1689 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1690 continue;
1691 free(istate->cache[i]);
1693 resolve_undo_clear_index(istate);
1694 istate->cache_nr = 0;
1695 istate->cache_changed = 0;
1696 istate->timestamp.sec = 0;
1697 istate->timestamp.nsec = 0;
1698 free_name_hash(istate);
1699 cache_tree_free(&(istate->cache_tree));
1700 istate->initialized = 0;
1701 free(istate->cache);
1702 istate->cache = NULL;
1703 istate->cache_alloc = 0;
1704 discard_split_index(istate);
1705 free_untracked_cache(istate->untracked);
1706 istate->untracked = NULL;
1707 return 0;
1710 int unmerged_index(const struct index_state *istate)
1712 int i;
1713 for (i = 0; i < istate->cache_nr; i++) {
1714 if (ce_stage(istate->cache[i]))
1715 return 1;
1717 return 0;
1720 #define WRITE_BUFFER_SIZE 8192
1721 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1722 static unsigned long write_buffer_len;
1724 static int ce_write_flush(git_SHA_CTX *context, int fd)
1726 unsigned int buffered = write_buffer_len;
1727 if (buffered) {
1728 git_SHA1_Update(context, write_buffer, buffered);
1729 if (write_in_full(fd, write_buffer, buffered) != buffered)
1730 return -1;
1731 write_buffer_len = 0;
1733 return 0;
1736 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1738 while (len) {
1739 unsigned int buffered = write_buffer_len;
1740 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1741 if (partial > len)
1742 partial = len;
1743 memcpy(write_buffer + buffered, data, partial);
1744 buffered += partial;
1745 if (buffered == WRITE_BUFFER_SIZE) {
1746 write_buffer_len = buffered;
1747 if (ce_write_flush(context, fd))
1748 return -1;
1749 buffered = 0;
1751 write_buffer_len = buffered;
1752 len -= partial;
1753 data = (char *) data + partial;
1755 return 0;
1758 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1759 unsigned int ext, unsigned int sz)
1761 ext = htonl(ext);
1762 sz = htonl(sz);
1763 return ((ce_write(context, fd, &ext, 4) < 0) ||
1764 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1767 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1769 unsigned int left = write_buffer_len;
1771 if (left) {
1772 write_buffer_len = 0;
1773 git_SHA1_Update(context, write_buffer, left);
1776 /* Flush first if not enough space for SHA1 signature */
1777 if (left + 20 > WRITE_BUFFER_SIZE) {
1778 if (write_in_full(fd, write_buffer, left) != left)
1779 return -1;
1780 left = 0;
1783 /* Append the SHA1 signature at the end */
1784 git_SHA1_Final(write_buffer + left, context);
1785 hashcpy(sha1, write_buffer + left);
1786 left += 20;
1787 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1790 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1793 * The only thing we care about in this function is to smudge the
1794 * falsely clean entry due to touch-update-touch race, so we leave
1795 * everything else as they are. We are called for entries whose
1796 * ce_stat_data.sd_mtime match the index file mtime.
1798 * Note that this actually does not do much for gitlinks, for
1799 * which ce_match_stat_basic() always goes to the actual
1800 * contents. The caller checks with is_racy_timestamp() which
1801 * always says "no" for gitlinks, so we are not called for them ;-)
1803 struct stat st;
1805 if (lstat(ce->name, &st) < 0)
1806 return;
1807 if (ce_match_stat_basic(ce, &st))
1808 return;
1809 if (ce_modified_check_fs(ce, &st)) {
1810 /* This is "racily clean"; smudge it. Note that this
1811 * is a tricky code. At first glance, it may appear
1812 * that it can break with this sequence:
1814 * $ echo xyzzy >frotz
1815 * $ git-update-index --add frotz
1816 * $ : >frotz
1817 * $ sleep 3
1818 * $ echo filfre >nitfol
1819 * $ git-update-index --add nitfol
1821 * but it does not. When the second update-index runs,
1822 * it notices that the entry "frotz" has the same timestamp
1823 * as index, and if we were to smudge it by resetting its
1824 * size to zero here, then the object name recorded
1825 * in index is the 6-byte file but the cached stat information
1826 * becomes zero --- which would then match what we would
1827 * obtain from the filesystem next time we stat("frotz").
1829 * However, the second update-index, before calling
1830 * this function, notices that the cached size is 6
1831 * bytes and what is on the filesystem is an empty
1832 * file, and never calls us, so the cached size information
1833 * for "frotz" stays 6 which does not match the filesystem.
1835 ce->ce_stat_data.sd_size = 0;
1839 /* Copy miscellaneous fields but not the name */
1840 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1841 struct cache_entry *ce)
1843 short flags;
1845 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1846 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1847 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1848 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1849 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1850 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1851 ondisk->mode = htonl(ce->ce_mode);
1852 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1853 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1854 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1855 hashcpy(ondisk->sha1, ce->sha1);
1857 flags = ce->ce_flags & ~CE_NAMEMASK;
1858 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1859 ondisk->flags = htons(flags);
1860 if (ce->ce_flags & CE_EXTENDED) {
1861 struct ondisk_cache_entry_extended *ondisk2;
1862 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1863 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1864 return ondisk2->name;
1866 else {
1867 return ondisk->name;
1871 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1872 struct strbuf *previous_name)
1874 int size;
1875 struct ondisk_cache_entry *ondisk;
1876 int saved_namelen = saved_namelen; /* compiler workaround */
1877 char *name;
1878 int result;
1880 if (ce->ce_flags & CE_STRIP_NAME) {
1881 saved_namelen = ce_namelen(ce);
1882 ce->ce_namelen = 0;
1885 if (!previous_name) {
1886 size = ondisk_ce_size(ce);
1887 ondisk = xcalloc(1, size);
1888 name = copy_cache_entry_to_ondisk(ondisk, ce);
1889 memcpy(name, ce->name, ce_namelen(ce));
1890 } else {
1891 int common, to_remove, prefix_size;
1892 unsigned char to_remove_vi[16];
1893 for (common = 0;
1894 (ce->name[common] &&
1895 common < previous_name->len &&
1896 ce->name[common] == previous_name->buf[common]);
1897 common++)
1898 ; /* still matching */
1899 to_remove = previous_name->len - common;
1900 prefix_size = encode_varint(to_remove, to_remove_vi);
1902 if (ce->ce_flags & CE_EXTENDED)
1903 size = offsetof(struct ondisk_cache_entry_extended, name);
1904 else
1905 size = offsetof(struct ondisk_cache_entry, name);
1906 size += prefix_size + (ce_namelen(ce) - common + 1);
1908 ondisk = xcalloc(1, size);
1909 name = copy_cache_entry_to_ondisk(ondisk, ce);
1910 memcpy(name, to_remove_vi, prefix_size);
1911 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1913 strbuf_splice(previous_name, common, to_remove,
1914 ce->name + common, ce_namelen(ce) - common);
1916 if (ce->ce_flags & CE_STRIP_NAME) {
1917 ce->ce_namelen = saved_namelen;
1918 ce->ce_flags &= ~CE_STRIP_NAME;
1921 result = ce_write(c, fd, ondisk, size);
1922 free(ondisk);
1923 return result;
1927 * This function verifies if index_state has the correct sha1 of the
1928 * index file. Don't die if we have any other failure, just return 0.
1930 static int verify_index_from(const struct index_state *istate, const char *path)
1932 int fd;
1933 ssize_t n;
1934 struct stat st;
1935 unsigned char sha1[20];
1937 if (!istate->initialized)
1938 return 0;
1940 fd = open(path, O_RDONLY);
1941 if (fd < 0)
1942 return 0;
1944 if (fstat(fd, &st))
1945 goto out;
1947 if (st.st_size < sizeof(struct cache_header) + 20)
1948 goto out;
1950 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1951 if (n != 20)
1952 goto out;
1954 if (hashcmp(istate->sha1, sha1))
1955 goto out;
1957 close(fd);
1958 return 1;
1960 out:
1961 close(fd);
1962 return 0;
1965 static int verify_index(const struct index_state *istate)
1967 return verify_index_from(istate, get_index_file());
1970 static int has_racy_timestamp(struct index_state *istate)
1972 int entries = istate->cache_nr;
1973 int i;
1975 for (i = 0; i < entries; i++) {
1976 struct cache_entry *ce = istate->cache[i];
1977 if (is_racy_timestamp(istate, ce))
1978 return 1;
1980 return 0;
1984 * Opportunistically update the index but do not complain if we can't
1986 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1988 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1989 verify_index(istate) &&
1990 write_locked_index(istate, lockfile, COMMIT_LOCK))
1991 rollback_lock_file(lockfile);
1994 static int do_write_index(struct index_state *istate, int newfd,
1995 int strip_extensions)
1997 git_SHA_CTX c;
1998 struct cache_header hdr;
1999 int i, err, removed, extended, hdr_version;
2000 struct cache_entry **cache = istate->cache;
2001 int entries = istate->cache_nr;
2002 struct stat st;
2003 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2005 for (i = removed = extended = 0; i < entries; i++) {
2006 if (cache[i]->ce_flags & CE_REMOVE)
2007 removed++;
2009 /* reduce extended entries if possible */
2010 cache[i]->ce_flags &= ~CE_EXTENDED;
2011 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2012 extended++;
2013 cache[i]->ce_flags |= CE_EXTENDED;
2017 if (!istate->version) {
2018 istate->version = get_index_format_default();
2019 if (getenv("GIT_TEST_SPLIT_INDEX"))
2020 init_split_index(istate);
2023 /* demote version 3 to version 2 when the latter suffices */
2024 if (istate->version == 3 || istate->version == 2)
2025 istate->version = extended ? 3 : 2;
2027 hdr_version = istate->version;
2029 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2030 hdr.hdr_version = htonl(hdr_version);
2031 hdr.hdr_entries = htonl(entries - removed);
2033 git_SHA1_Init(&c);
2034 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2035 return -1;
2037 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2038 for (i = 0; i < entries; i++) {
2039 struct cache_entry *ce = cache[i];
2040 if (ce->ce_flags & CE_REMOVE)
2041 continue;
2042 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2043 ce_smudge_racily_clean_entry(ce);
2044 if (is_null_sha1(ce->sha1)) {
2045 static const char msg[] = "cache entry has null sha1: %s";
2046 static int allow = -1;
2048 if (allow < 0)
2049 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2050 if (allow)
2051 warning(msg, ce->name);
2052 else
2053 return error(msg, ce->name);
2055 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2056 return -1;
2058 strbuf_release(&previous_name_buf);
2060 /* Write extension data here */
2061 if (!strip_extensions && istate->split_index) {
2062 struct strbuf sb = STRBUF_INIT;
2064 err = write_link_extension(&sb, istate) < 0 ||
2065 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2066 sb.len) < 0 ||
2067 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2068 strbuf_release(&sb);
2069 if (err)
2070 return -1;
2072 if (!strip_extensions && istate->cache_tree) {
2073 struct strbuf sb = STRBUF_INIT;
2075 cache_tree_write(&sb, istate->cache_tree);
2076 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2077 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2078 strbuf_release(&sb);
2079 if (err)
2080 return -1;
2082 if (!strip_extensions && istate->resolve_undo) {
2083 struct strbuf sb = STRBUF_INIT;
2085 resolve_undo_write(&sb, istate->resolve_undo);
2086 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2087 sb.len) < 0
2088 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2089 strbuf_release(&sb);
2090 if (err)
2091 return -1;
2093 if (!strip_extensions && istate->untracked) {
2094 struct strbuf sb = STRBUF_INIT;
2096 write_untracked_extension(&sb, istate->untracked);
2097 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2098 sb.len) < 0 ||
2099 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2100 strbuf_release(&sb);
2101 if (err)
2102 return -1;
2105 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2106 return -1;
2107 istate->timestamp.sec = (unsigned int)st.st_mtime;
2108 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2109 return 0;
2112 void set_alternate_index_output(const char *name)
2114 alternate_index_output = name;
2117 static int commit_locked_index(struct lock_file *lk)
2119 if (alternate_index_output)
2120 return commit_lock_file_to(lk, alternate_index_output);
2121 else
2122 return commit_lock_file(lk);
2125 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2126 unsigned flags)
2128 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2129 if (ret)
2130 return ret;
2131 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2132 (COMMIT_LOCK | CLOSE_LOCK));
2133 if (flags & COMMIT_LOCK)
2134 return commit_locked_index(lock);
2135 else if (flags & CLOSE_LOCK)
2136 return close_lock_file(lock);
2137 else
2138 return ret;
2141 static int write_split_index(struct index_state *istate,
2142 struct lock_file *lock,
2143 unsigned flags)
2145 int ret;
2146 prepare_to_write_split_index(istate);
2147 ret = do_write_locked_index(istate, lock, flags);
2148 finish_writing_split_index(istate);
2149 return ret;
2152 static struct tempfile temporary_sharedindex;
2154 static int write_shared_index(struct index_state *istate,
2155 struct lock_file *lock, unsigned flags)
2157 struct split_index *si = istate->split_index;
2158 int fd, ret;
2160 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2161 if (fd < 0) {
2162 hashclr(si->base_sha1);
2163 return do_write_locked_index(istate, lock, flags);
2165 move_cache_to_base_index(istate);
2166 ret = do_write_index(si->base, fd, 1);
2167 if (ret) {
2168 delete_tempfile(&temporary_sharedindex);
2169 return ret;
2171 ret = rename_tempfile(&temporary_sharedindex,
2172 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2173 if (!ret)
2174 hashcpy(si->base_sha1, si->base->sha1);
2175 return ret;
2178 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2179 unsigned flags)
2181 struct split_index *si = istate->split_index;
2183 if (!si || alternate_index_output ||
2184 (istate->cache_changed & ~EXTMASK)) {
2185 if (si)
2186 hashclr(si->base_sha1);
2187 return do_write_locked_index(istate, lock, flags);
2190 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2191 int v = si->base_sha1[0];
2192 if ((v & 15) < 6)
2193 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2195 if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2196 int ret = write_shared_index(istate, lock, flags);
2197 if (ret)
2198 return ret;
2201 return write_split_index(istate, lock, flags);
2205 * Read the index file that is potentially unmerged into given
2206 * index_state, dropping any unmerged entries. Returns true if
2207 * the index is unmerged. Callers who want to refuse to work
2208 * from an unmerged state can call this and check its return value,
2209 * instead of calling read_cache().
2211 int read_index_unmerged(struct index_state *istate)
2213 int i;
2214 int unmerged = 0;
2216 read_index(istate);
2217 for (i = 0; i < istate->cache_nr; i++) {
2218 struct cache_entry *ce = istate->cache[i];
2219 struct cache_entry *new_ce;
2220 int size, len;
2222 if (!ce_stage(ce))
2223 continue;
2224 unmerged = 1;
2225 len = ce_namelen(ce);
2226 size = cache_entry_size(len);
2227 new_ce = xcalloc(1, size);
2228 memcpy(new_ce->name, ce->name, len);
2229 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2230 new_ce->ce_namelen = len;
2231 new_ce->ce_mode = ce->ce_mode;
2232 if (add_index_entry(istate, new_ce, 0))
2233 return error("%s: cannot drop to stage #0",
2234 new_ce->name);
2236 return unmerged;
2240 * Returns 1 if the path is an "other" path with respect to
2241 * the index; that is, the path is not mentioned in the index at all,
2242 * either as a file, a directory with some files in the index,
2243 * or as an unmerged entry.
2245 * We helpfully remove a trailing "/" from directories so that
2246 * the output of read_directory can be used as-is.
2248 int index_name_is_other(const struct index_state *istate, const char *name,
2249 int namelen)
2251 int pos;
2252 if (namelen && name[namelen - 1] == '/')
2253 namelen--;
2254 pos = index_name_pos(istate, name, namelen);
2255 if (0 <= pos)
2256 return 0; /* exact match */
2257 pos = -pos - 1;
2258 if (pos < istate->cache_nr) {
2259 struct cache_entry *ce = istate->cache[pos];
2260 if (ce_namelen(ce) == namelen &&
2261 !memcmp(ce->name, name, namelen))
2262 return 0; /* Yup, this one exists unmerged */
2264 return 1;
2267 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2269 int pos, len;
2270 unsigned long sz;
2271 enum object_type type;
2272 void *data;
2274 len = strlen(path);
2275 pos = index_name_pos(istate, path, len);
2276 if (pos < 0) {
2278 * We might be in the middle of a merge, in which
2279 * case we would read stage #2 (ours).
2281 int i;
2282 for (i = -pos - 1;
2283 (pos < 0 && i < istate->cache_nr &&
2284 !strcmp(istate->cache[i]->name, path));
2285 i++)
2286 if (ce_stage(istate->cache[i]) == 2)
2287 pos = i;
2289 if (pos < 0)
2290 return NULL;
2291 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2292 if (!data || type != OBJ_BLOB) {
2293 free(data);
2294 return NULL;
2296 if (size)
2297 *size = sz;
2298 return data;
2301 void stat_validity_clear(struct stat_validity *sv)
2303 free(sv->sd);
2304 sv->sd = NULL;
2307 int stat_validity_check(struct stat_validity *sv, const char *path)
2309 struct stat st;
2311 if (stat(path, &st) < 0)
2312 return sv->sd == NULL;
2313 if (!sv->sd)
2314 return 0;
2315 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2318 void stat_validity_update(struct stat_validity *sv, int fd)
2320 struct stat st;
2322 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2323 stat_validity_clear(sv);
2324 else {
2325 if (!sv->sd)
2326 sv->sd = xcalloc(1, sizeof(struct stat_data));
2327 fill_stat_data(sv->sd, &st);