Merge branch 'ag/rebase-p'
[git.git] / read-cache.c
blob372588260ea00f48ea4d031095901a575ac0e620
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 "config.h"
9 #include "tempfile.h"
10 #include "lockfile.h"
11 #include "cache-tree.h"
12 #include "refs.h"
13 #include "dir.h"
14 #include "tree.h"
15 #include "commit.h"
16 #include "blob.h"
17 #include "resolve-undo.h"
18 #include "strbuf.h"
19 #include "varint.h"
20 #include "split-index.h"
21 #include "utf8.h"
22 #include "fsmonitor.h"
24 /* Mask for the name length in ce_flags in the on-disk index */
26 #define CE_NAMEMASK (0x0fff)
28 /* Index extensions.
30 * The first letter should be 'A'..'Z' for extensions that are not
31 * necessary for a correct operation (i.e. optimization data).
32 * When new extensions are added that _needs_ to be understood in
33 * order to correctly interpret the index file, pick character that
34 * is outside the range, to cause the reader to abort.
37 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
38 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
39 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
40 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
41 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
42 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
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 | FSMONITOR_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 ce->ce_flags &= ~CE_HASHED;
66 set_index_entry(istate, nr, ce);
67 ce->ce_flags |= CE_UPDATE_IN_BASE;
68 mark_fsmonitor_invalid(istate, ce);
69 istate->cache_changed |= CE_ENTRY_CHANGED;
72 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
74 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
75 int namelen = strlen(new_name);
77 new_entry = xmalloc(cache_entry_size(namelen));
78 copy_cache_entry(new_entry, old_entry);
79 new_entry->ce_flags &= ~CE_HASHED;
80 new_entry->ce_namelen = namelen;
81 new_entry->index = 0;
82 memcpy(new_entry->name, new_name, namelen + 1);
84 cache_tree_invalidate_path(istate, old_entry->name);
85 untracked_cache_remove_from_index(istate, old_entry->name);
86 remove_index_entry_at(istate, nr);
87 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
90 void fill_stat_data(struct stat_data *sd, struct stat *st)
92 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
93 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
94 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
95 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
96 sd->sd_dev = st->st_dev;
97 sd->sd_ino = st->st_ino;
98 sd->sd_uid = st->st_uid;
99 sd->sd_gid = st->st_gid;
100 sd->sd_size = st->st_size;
103 int match_stat_data(const struct stat_data *sd, struct stat *st)
105 int changed = 0;
107 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
108 changed |= MTIME_CHANGED;
109 if (trust_ctime && check_stat &&
110 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
111 changed |= CTIME_CHANGED;
113 #ifdef USE_NSEC
114 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
115 changed |= MTIME_CHANGED;
116 if (trust_ctime && check_stat &&
117 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
118 changed |= CTIME_CHANGED;
119 #endif
121 if (check_stat) {
122 if (sd->sd_uid != (unsigned int) st->st_uid ||
123 sd->sd_gid != (unsigned int) st->st_gid)
124 changed |= OWNER_CHANGED;
125 if (sd->sd_ino != (unsigned int) st->st_ino)
126 changed |= INODE_CHANGED;
129 #ifdef USE_STDEV
131 * st_dev breaks on network filesystems where different
132 * clients will have different views of what "device"
133 * the filesystem is on
135 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
136 changed |= INODE_CHANGED;
137 #endif
139 if (sd->sd_size != (unsigned int) st->st_size)
140 changed |= DATA_CHANGED;
142 return changed;
146 * This only updates the "non-critical" parts of the directory
147 * cache, ie the parts that aren't tracked by GIT, and only used
148 * to validate the cache.
150 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
152 fill_stat_data(&ce->ce_stat_data, st);
154 if (assume_unchanged)
155 ce->ce_flags |= CE_VALID;
157 if (S_ISREG(st->st_mode)) {
158 ce_mark_uptodate(ce);
159 mark_fsmonitor_valid(ce);
163 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
165 int match = -1;
166 int fd = git_open_cloexec(ce->name, O_RDONLY);
168 if (fd >= 0) {
169 struct object_id oid;
170 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
171 match = oidcmp(&oid, &ce->oid);
172 /* index_fd() closed the file descriptor already */
174 return match;
177 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
179 int match = -1;
180 void *buffer;
181 unsigned long size;
182 enum object_type type;
183 struct strbuf sb = STRBUF_INIT;
185 if (strbuf_readlink(&sb, ce->name, expected_size))
186 return -1;
188 buffer = read_object_file(&ce->oid, &type, &size);
189 if (buffer) {
190 if (size == sb.len)
191 match = memcmp(buffer, sb.buf, size);
192 free(buffer);
194 strbuf_release(&sb);
195 return match;
198 static int ce_compare_gitlink(const struct cache_entry *ce)
200 struct object_id oid;
203 * We don't actually require that the .git directory
204 * under GITLINK directory be a valid git directory. It
205 * might even be missing (in case nobody populated that
206 * sub-project).
208 * If so, we consider it always to match.
210 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
211 return 0;
212 return oidcmp(&oid, &ce->oid);
215 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
217 switch (st->st_mode & S_IFMT) {
218 case S_IFREG:
219 if (ce_compare_data(ce, st))
220 return DATA_CHANGED;
221 break;
222 case S_IFLNK:
223 if (ce_compare_link(ce, xsize_t(st->st_size)))
224 return DATA_CHANGED;
225 break;
226 case S_IFDIR:
227 if (S_ISGITLINK(ce->ce_mode))
228 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
229 /* else fallthrough */
230 default:
231 return TYPE_CHANGED;
233 return 0;
236 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
238 unsigned int changed = 0;
240 if (ce->ce_flags & CE_REMOVE)
241 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
243 switch (ce->ce_mode & S_IFMT) {
244 case S_IFREG:
245 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
246 /* We consider only the owner x bit to be relevant for
247 * "mode changes"
249 if (trust_executable_bit &&
250 (0100 & (ce->ce_mode ^ st->st_mode)))
251 changed |= MODE_CHANGED;
252 break;
253 case S_IFLNK:
254 if (!S_ISLNK(st->st_mode) &&
255 (has_symlinks || !S_ISREG(st->st_mode)))
256 changed |= TYPE_CHANGED;
257 break;
258 case S_IFGITLINK:
259 /* We ignore most of the st_xxx fields for gitlinks */
260 if (!S_ISDIR(st->st_mode))
261 changed |= TYPE_CHANGED;
262 else if (ce_compare_gitlink(ce))
263 changed |= DATA_CHANGED;
264 return changed;
265 default:
266 die("internal error: ce_mode is %o", ce->ce_mode);
269 changed |= match_stat_data(&ce->ce_stat_data, st);
271 /* Racily smudged entry? */
272 if (!ce->ce_stat_data.sd_size) {
273 if (!is_empty_blob_sha1(ce->oid.hash))
274 changed |= DATA_CHANGED;
277 return changed;
280 static int is_racy_stat(const struct index_state *istate,
281 const struct stat_data *sd)
283 return (istate->timestamp.sec &&
284 #ifdef USE_NSEC
285 /* nanosecond timestamped files can also be racy! */
286 (istate->timestamp.sec < sd->sd_mtime.sec ||
287 (istate->timestamp.sec == sd->sd_mtime.sec &&
288 istate->timestamp.nsec <= sd->sd_mtime.nsec))
289 #else
290 istate->timestamp.sec <= sd->sd_mtime.sec
291 #endif
295 static int is_racy_timestamp(const struct index_state *istate,
296 const struct cache_entry *ce)
298 return (!S_ISGITLINK(ce->ce_mode) &&
299 is_racy_stat(istate, &ce->ce_stat_data));
302 int match_stat_data_racy(const struct index_state *istate,
303 const struct stat_data *sd, struct stat *st)
305 if (is_racy_stat(istate, sd))
306 return MTIME_CHANGED;
307 return match_stat_data(sd, st);
310 int ie_match_stat(struct index_state *istate,
311 const struct cache_entry *ce, struct stat *st,
312 unsigned int options)
314 unsigned int changed;
315 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
316 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
317 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
318 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
320 if (!ignore_fsmonitor)
321 refresh_fsmonitor(istate);
323 * If it's marked as always valid in the index, it's
324 * valid whatever the checked-out copy says.
326 * skip-worktree has the same effect with higher precedence
328 if (!ignore_skip_worktree && ce_skip_worktree(ce))
329 return 0;
330 if (!ignore_valid && (ce->ce_flags & CE_VALID))
331 return 0;
332 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
333 return 0;
336 * Intent-to-add entries have not been added, so the index entry
337 * by definition never matches what is in the work tree until it
338 * actually gets added.
340 if (ce_intent_to_add(ce))
341 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
343 changed = ce_match_stat_basic(ce, st);
346 * Within 1 second of this sequence:
347 * echo xyzzy >file && git-update-index --add file
348 * running this command:
349 * echo frotz >file
350 * would give a falsely clean cache entry. The mtime and
351 * length match the cache, and other stat fields do not change.
353 * We could detect this at update-index time (the cache entry
354 * being registered/updated records the same time as "now")
355 * and delay the return from git-update-index, but that would
356 * effectively mean we can make at most one commit per second,
357 * which is not acceptable. Instead, we check cache entries
358 * whose mtime are the same as the index file timestamp more
359 * carefully than others.
361 if (!changed && is_racy_timestamp(istate, ce)) {
362 if (assume_racy_is_modified)
363 changed |= DATA_CHANGED;
364 else
365 changed |= ce_modified_check_fs(ce, st);
368 return changed;
371 int ie_modified(struct index_state *istate,
372 const struct cache_entry *ce,
373 struct stat *st, unsigned int options)
375 int changed, changed_fs;
377 changed = ie_match_stat(istate, ce, st, options);
378 if (!changed)
379 return 0;
381 * If the mode or type has changed, there's no point in trying
382 * to refresh the entry - it's not going to match
384 if (changed & (MODE_CHANGED | TYPE_CHANGED))
385 return changed;
388 * Immediately after read-tree or update-index --cacheinfo,
389 * the length field is zero, as we have never even read the
390 * lstat(2) information once, and we cannot trust DATA_CHANGED
391 * returned by ie_match_stat() which in turn was returned by
392 * ce_match_stat_basic() to signal that the filesize of the
393 * blob changed. We have to actually go to the filesystem to
394 * see if the contents match, and if so, should answer "unchanged".
396 * The logic does not apply to gitlinks, as ce_match_stat_basic()
397 * already has checked the actual HEAD from the filesystem in the
398 * subproject. If ie_match_stat() already said it is different,
399 * then we know it is.
401 if ((changed & DATA_CHANGED) &&
402 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
403 return changed;
405 changed_fs = ce_modified_check_fs(ce, st);
406 if (changed_fs)
407 return changed | changed_fs;
408 return 0;
411 int base_name_compare(const char *name1, int len1, int mode1,
412 const char *name2, int len2, int mode2)
414 unsigned char c1, c2;
415 int len = len1 < len2 ? len1 : len2;
416 int cmp;
418 cmp = memcmp(name1, name2, len);
419 if (cmp)
420 return cmp;
421 c1 = name1[len];
422 c2 = name2[len];
423 if (!c1 && S_ISDIR(mode1))
424 c1 = '/';
425 if (!c2 && S_ISDIR(mode2))
426 c2 = '/';
427 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
431 * df_name_compare() is identical to base_name_compare(), except it
432 * compares conflicting directory/file entries as equal. Note that
433 * while a directory name compares as equal to a regular file, they
434 * then individually compare _differently_ to a filename that has
435 * a dot after the basename (because '\0' < '.' < '/').
437 * This is used by routines that want to traverse the git namespace
438 * but then handle conflicting entries together when possible.
440 int df_name_compare(const char *name1, int len1, int mode1,
441 const char *name2, int len2, int mode2)
443 int len = len1 < len2 ? len1 : len2, cmp;
444 unsigned char c1, c2;
446 cmp = memcmp(name1, name2, len);
447 if (cmp)
448 return cmp;
449 /* Directories and files compare equal (same length, same name) */
450 if (len1 == len2)
451 return 0;
452 c1 = name1[len];
453 if (!c1 && S_ISDIR(mode1))
454 c1 = '/';
455 c2 = name2[len];
456 if (!c2 && S_ISDIR(mode2))
457 c2 = '/';
458 if (c1 == '/' && !c2)
459 return 0;
460 if (c2 == '/' && !c1)
461 return 0;
462 return c1 - c2;
465 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
467 size_t min_len = (len1 < len2) ? len1 : len2;
468 int cmp = memcmp(name1, name2, min_len);
469 if (cmp)
470 return cmp;
471 if (len1 < len2)
472 return -1;
473 if (len1 > len2)
474 return 1;
475 return 0;
478 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
480 int cmp;
482 cmp = name_compare(name1, len1, name2, len2);
483 if (cmp)
484 return cmp;
486 if (stage1 < stage2)
487 return -1;
488 if (stage1 > stage2)
489 return 1;
490 return 0;
493 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
495 int first, last;
497 first = 0;
498 last = istate->cache_nr;
499 while (last > first) {
500 int next = (last + first) >> 1;
501 struct cache_entry *ce = istate->cache[next];
502 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
503 if (!cmp)
504 return next;
505 if (cmp < 0) {
506 last = next;
507 continue;
509 first = next+1;
511 return -first-1;
514 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
516 return index_name_stage_pos(istate, name, namelen, 0);
519 int remove_index_entry_at(struct index_state *istate, int pos)
521 struct cache_entry *ce = istate->cache[pos];
523 record_resolve_undo(istate, ce);
524 remove_name_hash(istate, ce);
525 save_or_free_index_entry(istate, ce);
526 istate->cache_changed |= CE_ENTRY_REMOVED;
527 istate->cache_nr--;
528 if (pos >= istate->cache_nr)
529 return 0;
530 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
531 istate->cache_nr - pos);
532 return 1;
536 * Remove all cache entries marked for removal, that is where
537 * CE_REMOVE is set in ce_flags. This is much more effective than
538 * calling remove_index_entry_at() for each entry to be removed.
540 void remove_marked_cache_entries(struct index_state *istate)
542 struct cache_entry **ce_array = istate->cache;
543 unsigned int i, j;
545 for (i = j = 0; i < istate->cache_nr; i++) {
546 if (ce_array[i]->ce_flags & CE_REMOVE) {
547 remove_name_hash(istate, ce_array[i]);
548 save_or_free_index_entry(istate, ce_array[i]);
550 else
551 ce_array[j++] = ce_array[i];
553 if (j == istate->cache_nr)
554 return;
555 istate->cache_changed |= CE_ENTRY_REMOVED;
556 istate->cache_nr = j;
559 int remove_file_from_index(struct index_state *istate, const char *path)
561 int pos = index_name_pos(istate, path, strlen(path));
562 if (pos < 0)
563 pos = -pos-1;
564 cache_tree_invalidate_path(istate, path);
565 untracked_cache_remove_from_index(istate, path);
566 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
567 remove_index_entry_at(istate, pos);
568 return 0;
571 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
573 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
576 static int index_name_pos_also_unmerged(struct index_state *istate,
577 const char *path, int namelen)
579 int pos = index_name_pos(istate, path, namelen);
580 struct cache_entry *ce;
582 if (pos >= 0)
583 return pos;
585 /* maybe unmerged? */
586 pos = -1 - pos;
587 if (pos >= istate->cache_nr ||
588 compare_name((ce = istate->cache[pos]), path, namelen))
589 return -1;
591 /* order of preference: stage 2, 1, 3 */
592 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
593 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
594 !compare_name(ce, path, namelen))
595 pos++;
596 return pos;
599 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
601 int len = ce_namelen(ce);
602 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
606 * If we add a filename that aliases in the cache, we will use the
607 * name that we already have - but we don't want to update the same
608 * alias twice, because that implies that there were actually two
609 * different files with aliasing names!
611 * So we use the CE_ADDED flag to verify that the alias was an old
612 * one before we accept it as
614 static struct cache_entry *create_alias_ce(struct index_state *istate,
615 struct cache_entry *ce,
616 struct cache_entry *alias)
618 int len;
619 struct cache_entry *new_entry;
621 if (alias->ce_flags & CE_ADDED)
622 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
624 /* Ok, create the new entry using the name of the existing alias */
625 len = ce_namelen(alias);
626 new_entry = xcalloc(1, cache_entry_size(len));
627 memcpy(new_entry->name, alias->name, len);
628 copy_cache_entry(new_entry, ce);
629 save_or_free_index_entry(istate, ce);
630 return new_entry;
633 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
635 struct object_id oid;
636 if (write_object_file("", 0, blob_type, &oid))
637 die("cannot create an empty blob in the object database");
638 oidcpy(&ce->oid, &oid);
641 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
643 int size, namelen, was_same;
644 mode_t st_mode = st->st_mode;
645 struct cache_entry *ce, *alias = NULL;
646 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
647 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
648 int pretend = flags & ADD_CACHE_PRETEND;
649 int intent_only = flags & ADD_CACHE_INTENT;
650 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
651 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
652 int newflags = HASH_WRITE_OBJECT;
654 if (flags & HASH_RENORMALIZE)
655 newflags |= HASH_RENORMALIZE;
657 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
658 return error("%s: can only add regular files, symbolic links or git-directories", path);
660 namelen = strlen(path);
661 if (S_ISDIR(st_mode)) {
662 while (namelen && path[namelen-1] == '/')
663 namelen--;
665 size = cache_entry_size(namelen);
666 ce = xcalloc(1, size);
667 memcpy(ce->name, path, namelen);
668 ce->ce_namelen = namelen;
669 if (!intent_only)
670 fill_stat_cache_info(ce, st);
671 else
672 ce->ce_flags |= CE_INTENT_TO_ADD;
675 if (trust_executable_bit && has_symlinks) {
676 ce->ce_mode = create_ce_mode(st_mode);
677 } else {
678 /* If there is an existing entry, pick the mode bits and type
679 * from it, otherwise assume unexecutable regular file.
681 struct cache_entry *ent;
682 int pos = index_name_pos_also_unmerged(istate, path, namelen);
684 ent = (0 <= pos) ? istate->cache[pos] : NULL;
685 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
688 /* When core.ignorecase=true, determine if a directory of the same name but differing
689 * case already exists within the Git repository. If it does, ensure the directory
690 * case of the file being added to the repository matches (is folded into) the existing
691 * entry's directory case.
693 if (ignore_case) {
694 adjust_dirname_case(istate, ce->name);
696 if (!(flags & HASH_RENORMALIZE)) {
697 alias = index_file_exists(istate, ce->name,
698 ce_namelen(ce), ignore_case);
699 if (alias &&
700 !ce_stage(alias) &&
701 !ie_match_stat(istate, alias, st, ce_option)) {
702 /* Nothing changed, really */
703 if (!S_ISGITLINK(alias->ce_mode))
704 ce_mark_uptodate(alias);
705 alias->ce_flags |= CE_ADDED;
707 free(ce);
708 return 0;
711 if (!intent_only) {
712 if (index_path(&ce->oid, path, st, newflags)) {
713 free(ce);
714 return error("unable to index file %s", path);
716 } else
717 set_object_name_for_intent_to_add_entry(ce);
719 if (ignore_case && alias && different_name(ce, alias))
720 ce = create_alias_ce(istate, ce, alias);
721 ce->ce_flags |= CE_ADDED;
723 /* It was suspected to be racily clean, but it turns out to be Ok */
724 was_same = (alias &&
725 !ce_stage(alias) &&
726 !oidcmp(&alias->oid, &ce->oid) &&
727 ce->ce_mode == alias->ce_mode);
729 if (pretend)
730 free(ce);
731 else if (add_index_entry(istate, ce, add_option)) {
732 free(ce);
733 return error("unable to add %s to index", path);
735 if (verbose && !was_same)
736 printf("add '%s'\n", path);
737 return 0;
740 int add_file_to_index(struct index_state *istate, const char *path, int flags)
742 struct stat st;
743 if (lstat(path, &st))
744 die_errno("unable to stat '%s'", path);
745 return add_to_index(istate, path, &st, flags);
748 struct cache_entry *make_cache_entry(unsigned int mode,
749 const unsigned char *sha1, const char *path, int stage,
750 unsigned int refresh_options)
752 int size, len;
753 struct cache_entry *ce, *ret;
755 if (!verify_path(path, mode)) {
756 error("Invalid path '%s'", path);
757 return NULL;
760 len = strlen(path);
761 size = cache_entry_size(len);
762 ce = xcalloc(1, size);
764 hashcpy(ce->oid.hash, sha1);
765 memcpy(ce->name, path, len);
766 ce->ce_flags = create_ce_flags(stage);
767 ce->ce_namelen = len;
768 ce->ce_mode = create_ce_mode(mode);
770 ret = refresh_cache_entry(ce, refresh_options);
771 if (ret != ce)
772 free(ce);
773 return ret;
777 * Chmod an index entry with either +x or -x.
779 * Returns -1 if the chmod for the particular cache entry failed (if it's
780 * not a regular file), -2 if an invalid flip argument is passed in, 0
781 * otherwise.
783 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
784 char flip)
786 if (!S_ISREG(ce->ce_mode))
787 return -1;
788 switch (flip) {
789 case '+':
790 ce->ce_mode |= 0111;
791 break;
792 case '-':
793 ce->ce_mode &= ~0111;
794 break;
795 default:
796 return -2;
798 cache_tree_invalidate_path(istate, ce->name);
799 ce->ce_flags |= CE_UPDATE_IN_BASE;
800 mark_fsmonitor_invalid(istate, ce);
801 istate->cache_changed |= CE_ENTRY_CHANGED;
803 return 0;
806 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
808 int len = ce_namelen(a);
809 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
813 * We fundamentally don't like some paths: we don't want
814 * dot or dot-dot anywhere, and for obvious reasons don't
815 * want to recurse into ".git" either.
817 * Also, we don't want double slashes or slashes at the
818 * end that can make pathnames ambiguous.
820 static int verify_dotfile(const char *rest, unsigned mode)
823 * The first character was '.', but that
824 * has already been discarded, we now test
825 * the rest.
828 /* "." is not allowed */
829 if (*rest == '\0' || is_dir_sep(*rest))
830 return 0;
832 switch (*rest) {
834 * ".git" followed by NUL or slash is bad. Note that we match
835 * case-insensitively here, even if ignore_case is not set.
836 * This outlaws ".GIT" everywhere out of an abundance of caution,
837 * since there's really no good reason to allow it.
839 * Once we've seen ".git", we can also find ".gitmodules", etc (also
840 * case-insensitively).
842 case 'g':
843 case 'G':
844 if (rest[1] != 'i' && rest[1] != 'I')
845 break;
846 if (rest[2] != 't' && rest[2] != 'T')
847 break;
848 if (rest[3] == '\0' || is_dir_sep(rest[3]))
849 return 0;
850 if (S_ISLNK(mode)) {
851 rest += 3;
852 if (skip_iprefix(rest, "modules", &rest) &&
853 (*rest == '\0' || is_dir_sep(*rest)))
854 return 0;
856 break;
857 case '.':
858 if (rest[1] == '\0' || is_dir_sep(rest[1]))
859 return 0;
861 return 1;
864 int verify_path(const char *path, unsigned mode)
866 char c;
868 if (has_dos_drive_prefix(path))
869 return 0;
871 goto inside;
872 for (;;) {
873 if (!c)
874 return 1;
875 if (is_dir_sep(c)) {
876 inside:
877 if (protect_hfs) {
878 if (is_hfs_dotgit(path))
879 return 0;
880 if (S_ISLNK(mode)) {
881 if (is_hfs_dotgitmodules(path))
882 return 0;
885 if (protect_ntfs) {
886 if (is_ntfs_dotgit(path))
887 return 0;
888 if (S_ISLNK(mode)) {
889 if (is_ntfs_dotgitmodules(path))
890 return 0;
894 c = *path++;
895 if ((c == '.' && !verify_dotfile(path, mode)) ||
896 is_dir_sep(c) || c == '\0')
897 return 0;
899 c = *path++;
904 * Do we have another file that has the beginning components being a
905 * proper superset of the name we're trying to add?
907 static int has_file_name(struct index_state *istate,
908 const struct cache_entry *ce, int pos, int ok_to_replace)
910 int retval = 0;
911 int len = ce_namelen(ce);
912 int stage = ce_stage(ce);
913 const char *name = ce->name;
915 while (pos < istate->cache_nr) {
916 struct cache_entry *p = istate->cache[pos++];
918 if (len >= ce_namelen(p))
919 break;
920 if (memcmp(name, p->name, len))
921 break;
922 if (ce_stage(p) != stage)
923 continue;
924 if (p->name[len] != '/')
925 continue;
926 if (p->ce_flags & CE_REMOVE)
927 continue;
928 retval = -1;
929 if (!ok_to_replace)
930 break;
931 remove_index_entry_at(istate, --pos);
933 return retval;
938 * Like strcmp(), but also return the offset of the first change.
939 * If strings are equal, return the length.
941 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
943 size_t k;
945 if (!first_change)
946 return strcmp(s1, s2);
948 for (k = 0; s1[k] == s2[k]; k++)
949 if (s1[k] == '\0')
950 break;
952 *first_change = k;
953 return (unsigned char)s1[k] - (unsigned char)s2[k];
957 * Do we have another file with a pathname that is a proper
958 * subset of the name we're trying to add?
960 * That is, is there another file in the index with a path
961 * that matches a sub-directory in the given entry?
963 static int has_dir_name(struct index_state *istate,
964 const struct cache_entry *ce, int pos, int ok_to_replace)
966 int retval = 0;
967 int stage = ce_stage(ce);
968 const char *name = ce->name;
969 const char *slash = name + ce_namelen(ce);
970 size_t len_eq_last;
971 int cmp_last = 0;
974 * We are frequently called during an iteration on a sorted
975 * list of pathnames and while building a new index. Therefore,
976 * there is a high probability that this entry will eventually
977 * be appended to the index, rather than inserted in the middle.
978 * If we can confirm that, we can avoid binary searches on the
979 * components of the pathname.
981 * Compare the entry's full path with the last path in the index.
983 if (istate->cache_nr > 0) {
984 cmp_last = strcmp_offset(name,
985 istate->cache[istate->cache_nr - 1]->name,
986 &len_eq_last);
987 if (cmp_last > 0) {
988 if (len_eq_last == 0) {
990 * The entry sorts AFTER the last one in the
991 * index and their paths have no common prefix,
992 * so there cannot be a F/D conflict.
994 return retval;
995 } else {
997 * The entry sorts AFTER the last one in the
998 * index, but has a common prefix. Fall through
999 * to the loop below to disect the entry's path
1000 * and see where the difference is.
1003 } else if (cmp_last == 0) {
1005 * The entry exactly matches the last one in the
1006 * index, but because of multiple stage and CE_REMOVE
1007 * items, we fall through and let the regular search
1008 * code handle it.
1013 for (;;) {
1014 size_t len;
1016 for (;;) {
1017 if (*--slash == '/')
1018 break;
1019 if (slash <= ce->name)
1020 return retval;
1022 len = slash - name;
1024 if (cmp_last > 0) {
1026 * (len + 1) is a directory boundary (including
1027 * the trailing slash). And since the loop is
1028 * decrementing "slash", the first iteration is
1029 * the longest directory prefix; subsequent
1030 * iterations consider parent directories.
1033 if (len + 1 <= len_eq_last) {
1035 * The directory prefix (including the trailing
1036 * slash) also appears as a prefix in the last
1037 * entry, so the remainder cannot collide (because
1038 * strcmp said the whole path was greater).
1040 * EQ: last: xxx/A
1041 * this: xxx/B
1043 * LT: last: xxx/file_A
1044 * this: xxx/file_B
1046 return retval;
1049 if (len > len_eq_last) {
1051 * This part of the directory prefix (excluding
1052 * the trailing slash) is longer than the known
1053 * equal portions, so this sub-directory cannot
1054 * collide with a file.
1056 * GT: last: xxxA
1057 * this: xxxB/file
1059 return retval;
1062 if (istate->cache_nr > 0 &&
1063 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1065 * The directory prefix lines up with part of
1066 * a longer file or directory name, but sorts
1067 * after it, so this sub-directory cannot
1068 * collide with a file.
1070 * last: xxx/yy-file (because '-' sorts before '/')
1071 * this: xxx/yy/abc
1073 return retval;
1077 * This is a possible collision. Fall through and
1078 * let the regular search code handle it.
1080 * last: xxx
1081 * this: xxx/file
1085 pos = index_name_stage_pos(istate, name, len, stage);
1086 if (pos >= 0) {
1088 * Found one, but not so fast. This could
1089 * be a marker that says "I was here, but
1090 * I am being removed". Such an entry is
1091 * not a part of the resulting tree, and
1092 * it is Ok to have a directory at the same
1093 * path.
1095 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1096 retval = -1;
1097 if (!ok_to_replace)
1098 break;
1099 remove_index_entry_at(istate, pos);
1100 continue;
1103 else
1104 pos = -pos-1;
1107 * Trivial optimization: if we find an entry that
1108 * already matches the sub-directory, then we know
1109 * we're ok, and we can exit.
1111 while (pos < istate->cache_nr) {
1112 struct cache_entry *p = istate->cache[pos];
1113 if ((ce_namelen(p) <= len) ||
1114 (p->name[len] != '/') ||
1115 memcmp(p->name, name, len))
1116 break; /* not our subdirectory */
1117 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1119 * p is at the same stage as our entry, and
1120 * is a subdirectory of what we are looking
1121 * at, so we cannot have conflicts at our
1122 * level or anything shorter.
1124 return retval;
1125 pos++;
1128 return retval;
1131 /* We may be in a situation where we already have path/file and path
1132 * is being added, or we already have path and path/file is being
1133 * added. Either one would result in a nonsense tree that has path
1134 * twice when git-write-tree tries to write it out. Prevent it.
1136 * If ok-to-replace is specified, we remove the conflicting entries
1137 * from the cache so the caller should recompute the insert position.
1138 * When this happens, we return non-zero.
1140 static int check_file_directory_conflict(struct index_state *istate,
1141 const struct cache_entry *ce,
1142 int pos, int ok_to_replace)
1144 int retval;
1147 * When ce is an "I am going away" entry, we allow it to be added
1149 if (ce->ce_flags & CE_REMOVE)
1150 return 0;
1153 * We check if the path is a sub-path of a subsequent pathname
1154 * first, since removing those will not change the position
1155 * in the array.
1157 retval = has_file_name(istate, ce, pos, ok_to_replace);
1160 * Then check if the path might have a clashing sub-directory
1161 * before it.
1163 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1166 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1168 int pos;
1169 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1170 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1171 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1172 int new_only = option & ADD_CACHE_NEW_ONLY;
1174 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1175 cache_tree_invalidate_path(istate, ce->name);
1178 * If this entry's path sorts after the last entry in the index,
1179 * we can avoid searching for it.
1181 if (istate->cache_nr > 0 &&
1182 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1183 pos = -istate->cache_nr - 1;
1184 else
1185 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1187 /* existing match? Just replace it. */
1188 if (pos >= 0) {
1189 if (!new_only)
1190 replace_index_entry(istate, pos, ce);
1191 return 0;
1193 pos = -pos-1;
1195 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1196 untracked_cache_add_to_index(istate, ce->name);
1199 * Inserting a merged entry ("stage 0") into the index
1200 * will always replace all non-merged entries..
1202 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1203 while (ce_same_name(istate->cache[pos], ce)) {
1204 ok_to_add = 1;
1205 if (!remove_index_entry_at(istate, pos))
1206 break;
1210 if (!ok_to_add)
1211 return -1;
1212 if (!verify_path(ce->name, ce->ce_mode))
1213 return error("Invalid path '%s'", ce->name);
1215 if (!skip_df_check &&
1216 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1217 if (!ok_to_replace)
1218 return error("'%s' appears as both a file and as a directory",
1219 ce->name);
1220 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1221 pos = -pos-1;
1223 return pos + 1;
1226 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1228 int pos;
1230 if (option & ADD_CACHE_JUST_APPEND)
1231 pos = istate->cache_nr;
1232 else {
1233 int ret;
1234 ret = add_index_entry_with_check(istate, ce, option);
1235 if (ret <= 0)
1236 return ret;
1237 pos = ret - 1;
1240 /* Make sure the array is big enough .. */
1241 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1243 /* Add it in.. */
1244 istate->cache_nr++;
1245 if (istate->cache_nr > pos + 1)
1246 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1247 istate->cache_nr - pos - 1);
1248 set_index_entry(istate, pos, ce);
1249 istate->cache_changed |= CE_ENTRY_ADDED;
1250 return 0;
1254 * "refresh" does not calculate a new sha1 file or bring the
1255 * cache up-to-date for mode/content changes. But what it
1256 * _does_ do is to "re-match" the stat information of a file
1257 * with the cache, so that you can refresh the cache for a
1258 * file that hasn't been changed but where the stat entry is
1259 * out of date.
1261 * For example, you'd want to do this after doing a "git-read-tree",
1262 * to link up the stat cache details with the proper files.
1264 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1265 struct cache_entry *ce,
1266 unsigned int options, int *err,
1267 int *changed_ret)
1269 struct stat st;
1270 struct cache_entry *updated;
1271 int changed, size;
1272 int refresh = options & CE_MATCH_REFRESH;
1273 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1274 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1275 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1276 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1278 if (!refresh || ce_uptodate(ce))
1279 return ce;
1281 if (!ignore_fsmonitor)
1282 refresh_fsmonitor(istate);
1284 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1285 * that the change to the work tree does not matter and told
1286 * us not to worry.
1288 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1289 ce_mark_uptodate(ce);
1290 return ce;
1292 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1293 ce_mark_uptodate(ce);
1294 return ce;
1296 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1297 ce_mark_uptodate(ce);
1298 return ce;
1301 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1302 if (ignore_missing)
1303 return ce;
1304 if (err)
1305 *err = ENOENT;
1306 return NULL;
1309 if (lstat(ce->name, &st) < 0) {
1310 if (ignore_missing && errno == ENOENT)
1311 return ce;
1312 if (err)
1313 *err = errno;
1314 return NULL;
1317 changed = ie_match_stat(istate, ce, &st, options);
1318 if (changed_ret)
1319 *changed_ret = changed;
1320 if (!changed) {
1322 * The path is unchanged. If we were told to ignore
1323 * valid bit, then we did the actual stat check and
1324 * found that the entry is unmodified. If the entry
1325 * is not marked VALID, this is the place to mark it
1326 * valid again, under "assume unchanged" mode.
1328 if (ignore_valid && assume_unchanged &&
1329 !(ce->ce_flags & CE_VALID))
1330 ; /* mark this one VALID again */
1331 else {
1333 * We do not mark the index itself "modified"
1334 * because CE_UPTODATE flag is in-core only;
1335 * we are not going to write this change out.
1337 if (!S_ISGITLINK(ce->ce_mode)) {
1338 ce_mark_uptodate(ce);
1339 mark_fsmonitor_valid(ce);
1341 return ce;
1345 if (ie_modified(istate, ce, &st, options)) {
1346 if (err)
1347 *err = EINVAL;
1348 return NULL;
1351 size = ce_size(ce);
1352 updated = xmalloc(size);
1353 copy_cache_entry(updated, ce);
1354 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1355 fill_stat_cache_info(updated, &st);
1357 * If ignore_valid is not set, we should leave CE_VALID bit
1358 * alone. Otherwise, paths marked with --no-assume-unchanged
1359 * (i.e. things to be edited) will reacquire CE_VALID bit
1360 * automatically, which is not really what we want.
1362 if (!ignore_valid && assume_unchanged &&
1363 !(ce->ce_flags & CE_VALID))
1364 updated->ce_flags &= ~CE_VALID;
1366 /* istate->cache_changed is updated in the caller */
1367 return updated;
1370 static void show_file(const char * fmt, const char * name, int in_porcelain,
1371 int * first, const char *header_msg)
1373 if (in_porcelain && *first && header_msg) {
1374 printf("%s\n", header_msg);
1375 *first = 0;
1377 printf(fmt, name);
1380 int refresh_index(struct index_state *istate, unsigned int flags,
1381 const struct pathspec *pathspec,
1382 char *seen, const char *header_msg)
1384 int i;
1385 int has_errors = 0;
1386 int really = (flags & REFRESH_REALLY) != 0;
1387 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1388 int quiet = (flags & REFRESH_QUIET) != 0;
1389 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1390 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1391 int first = 1;
1392 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1393 unsigned int options = (CE_MATCH_REFRESH |
1394 (really ? CE_MATCH_IGNORE_VALID : 0) |
1395 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1396 const char *modified_fmt;
1397 const char *deleted_fmt;
1398 const char *typechange_fmt;
1399 const char *added_fmt;
1400 const char *unmerged_fmt;
1401 uint64_t start = getnanotime();
1403 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1404 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1405 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1406 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1407 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1408 for (i = 0; i < istate->cache_nr; i++) {
1409 struct cache_entry *ce, *new_entry;
1410 int cache_errno = 0;
1411 int changed = 0;
1412 int filtered = 0;
1414 ce = istate->cache[i];
1415 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1416 continue;
1418 if (pathspec && !ce_path_match(ce, pathspec, seen))
1419 filtered = 1;
1421 if (ce_stage(ce)) {
1422 while ((i < istate->cache_nr) &&
1423 ! strcmp(istate->cache[i]->name, ce->name))
1424 i++;
1425 i--;
1426 if (allow_unmerged)
1427 continue;
1428 if (!filtered)
1429 show_file(unmerged_fmt, ce->name, in_porcelain,
1430 &first, header_msg);
1431 has_errors = 1;
1432 continue;
1435 if (filtered)
1436 continue;
1438 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1439 if (new_entry == ce)
1440 continue;
1441 if (!new_entry) {
1442 const char *fmt;
1444 if (really && cache_errno == EINVAL) {
1445 /* If we are doing --really-refresh that
1446 * means the index is not valid anymore.
1448 ce->ce_flags &= ~CE_VALID;
1449 ce->ce_flags |= CE_UPDATE_IN_BASE;
1450 mark_fsmonitor_invalid(istate, ce);
1451 istate->cache_changed |= CE_ENTRY_CHANGED;
1453 if (quiet)
1454 continue;
1456 if (cache_errno == ENOENT)
1457 fmt = deleted_fmt;
1458 else if (ce_intent_to_add(ce))
1459 fmt = added_fmt; /* must be before other checks */
1460 else if (changed & TYPE_CHANGED)
1461 fmt = typechange_fmt;
1462 else
1463 fmt = modified_fmt;
1464 show_file(fmt,
1465 ce->name, in_porcelain, &first, header_msg);
1466 has_errors = 1;
1467 continue;
1470 replace_index_entry(istate, i, new_entry);
1472 trace_performance_since(start, "refresh index");
1473 return has_errors;
1476 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1477 unsigned int options)
1479 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1483 /*****************************************************************
1484 * Index File I/O
1485 *****************************************************************/
1487 #define INDEX_FORMAT_DEFAULT 3
1489 static unsigned int get_index_format_default(void)
1491 char *envversion = getenv("GIT_INDEX_VERSION");
1492 char *endp;
1493 int value;
1494 unsigned int version = INDEX_FORMAT_DEFAULT;
1496 if (!envversion) {
1497 if (!git_config_get_int("index.version", &value))
1498 version = value;
1499 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1500 warning(_("index.version set, but the value is invalid.\n"
1501 "Using version %i"), INDEX_FORMAT_DEFAULT);
1502 return INDEX_FORMAT_DEFAULT;
1504 return version;
1507 version = strtoul(envversion, &endp, 10);
1508 if (*endp ||
1509 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1510 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1511 "Using version %i"), INDEX_FORMAT_DEFAULT);
1512 version = INDEX_FORMAT_DEFAULT;
1514 return version;
1518 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1519 * Again - this is just a (very strong in practice) heuristic that
1520 * the inode hasn't changed.
1522 * We save the fields in big-endian order to allow using the
1523 * index file over NFS transparently.
1525 struct ondisk_cache_entry {
1526 struct cache_time ctime;
1527 struct cache_time mtime;
1528 uint32_t dev;
1529 uint32_t ino;
1530 uint32_t mode;
1531 uint32_t uid;
1532 uint32_t gid;
1533 uint32_t size;
1534 unsigned char sha1[20];
1535 uint16_t flags;
1536 char name[FLEX_ARRAY]; /* more */
1540 * This struct is used when CE_EXTENDED bit is 1
1541 * The struct must match ondisk_cache_entry exactly from
1542 * ctime till flags
1544 struct ondisk_cache_entry_extended {
1545 struct cache_time ctime;
1546 struct cache_time mtime;
1547 uint32_t dev;
1548 uint32_t ino;
1549 uint32_t mode;
1550 uint32_t uid;
1551 uint32_t gid;
1552 uint32_t size;
1553 unsigned char sha1[20];
1554 uint16_t flags;
1555 uint16_t flags2;
1556 char name[FLEX_ARRAY]; /* more */
1559 /* These are only used for v3 or lower */
1560 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1561 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1562 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1563 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1564 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1565 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1566 ondisk_cache_entry_size(ce_namelen(ce)))
1568 /* Allow fsck to force verification of the index checksum. */
1569 int verify_index_checksum;
1571 /* Allow fsck to force verification of the cache entry order. */
1572 int verify_ce_order;
1574 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1576 git_hash_ctx c;
1577 unsigned char hash[GIT_MAX_RAWSZ];
1578 int hdr_version;
1580 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1581 return error("bad signature");
1582 hdr_version = ntohl(hdr->hdr_version);
1583 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1584 return error("bad index version %d", hdr_version);
1586 if (!verify_index_checksum)
1587 return 0;
1589 the_hash_algo->init_fn(&c);
1590 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1591 the_hash_algo->final_fn(hash, &c);
1592 if (hashcmp(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1593 return error("bad index file sha1 signature");
1594 return 0;
1597 static int read_index_extension(struct index_state *istate,
1598 const char *ext, void *data, unsigned long sz)
1600 switch (CACHE_EXT(ext)) {
1601 case CACHE_EXT_TREE:
1602 istate->cache_tree = cache_tree_read(data, sz);
1603 break;
1604 case CACHE_EXT_RESOLVE_UNDO:
1605 istate->resolve_undo = resolve_undo_read(data, sz);
1606 break;
1607 case CACHE_EXT_LINK:
1608 if (read_link_extension(istate, data, sz))
1609 return -1;
1610 break;
1611 case CACHE_EXT_UNTRACKED:
1612 istate->untracked = read_untracked_extension(data, sz);
1613 break;
1614 case CACHE_EXT_FSMONITOR:
1615 read_fsmonitor_extension(istate, data, sz);
1616 break;
1617 default:
1618 if (*ext < 'A' || 'Z' < *ext)
1619 return error("index uses %.4s extension, which we do not understand",
1620 ext);
1621 fprintf(stderr, "ignoring %.4s extension\n", ext);
1622 break;
1624 return 0;
1627 int hold_locked_index(struct lock_file *lk, int lock_flags)
1629 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1632 int read_index(struct index_state *istate)
1634 return read_index_from(istate, get_index_file(), get_git_dir());
1637 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1638 unsigned int flags,
1639 const char *name,
1640 size_t len)
1642 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1644 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1645 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1646 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1647 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1648 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1649 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1650 ce->ce_mode = get_be32(&ondisk->mode);
1651 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1652 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1653 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1654 ce->ce_flags = flags & ~CE_NAMEMASK;
1655 ce->ce_namelen = len;
1656 ce->index = 0;
1657 hashcpy(ce->oid.hash, ondisk->sha1);
1658 memcpy(ce->name, name, len);
1659 ce->name[len] = '\0';
1660 return ce;
1664 * Adjacent cache entries tend to share the leading paths, so it makes
1665 * sense to only store the differences in later entries. In the v4
1666 * on-disk format of the index, each on-disk cache entry stores the
1667 * number of bytes to be stripped from the end of the previous name,
1668 * and the bytes to append to the result, to come up with its name.
1670 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1672 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1673 size_t len = decode_varint(&cp);
1675 if (name->len < len)
1676 die("malformed name field in the index");
1677 strbuf_remove(name, name->len - len, len);
1678 for (ep = cp; *ep; ep++)
1679 ; /* find the end */
1680 strbuf_add(name, cp, ep - cp);
1681 return (const char *)ep + 1 - cp_;
1684 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1685 unsigned long *ent_size,
1686 struct strbuf *previous_name)
1688 struct cache_entry *ce;
1689 size_t len;
1690 const char *name;
1691 unsigned int flags;
1693 /* On-disk flags are just 16 bits */
1694 flags = get_be16(&ondisk->flags);
1695 len = flags & CE_NAMEMASK;
1697 if (flags & CE_EXTENDED) {
1698 struct ondisk_cache_entry_extended *ondisk2;
1699 int extended_flags;
1700 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1701 extended_flags = get_be16(&ondisk2->flags2) << 16;
1702 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1703 if (extended_flags & ~CE_EXTENDED_FLAGS)
1704 die("Unknown index entry format %08x", extended_flags);
1705 flags |= extended_flags;
1706 name = ondisk2->name;
1708 else
1709 name = ondisk->name;
1711 if (!previous_name) {
1712 /* v3 and earlier */
1713 if (len == CE_NAMEMASK)
1714 len = strlen(name);
1715 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1717 *ent_size = ondisk_ce_size(ce);
1718 } else {
1719 unsigned long consumed;
1720 consumed = expand_name_field(previous_name, name);
1721 ce = cache_entry_from_ondisk(ondisk, flags,
1722 previous_name->buf,
1723 previous_name->len);
1725 *ent_size = (name - ((char *)ondisk)) + consumed;
1727 return ce;
1730 static void check_ce_order(struct index_state *istate)
1732 unsigned int i;
1734 if (!verify_ce_order)
1735 return;
1737 for (i = 1; i < istate->cache_nr; i++) {
1738 struct cache_entry *ce = istate->cache[i - 1];
1739 struct cache_entry *next_ce = istate->cache[i];
1740 int name_compare = strcmp(ce->name, next_ce->name);
1742 if (0 < name_compare)
1743 die("unordered stage entries in index");
1744 if (!name_compare) {
1745 if (!ce_stage(ce))
1746 die("multiple stage entries for merged file '%s'",
1747 ce->name);
1748 if (ce_stage(ce) > ce_stage(next_ce))
1749 die("unordered stage entries for '%s'",
1750 ce->name);
1755 static void tweak_untracked_cache(struct index_state *istate)
1757 switch (git_config_get_untracked_cache()) {
1758 case -1: /* keep: do nothing */
1759 break;
1760 case 0: /* false */
1761 remove_untracked_cache(istate);
1762 break;
1763 case 1: /* true */
1764 add_untracked_cache(istate);
1765 break;
1766 default: /* unknown value: do nothing */
1767 break;
1771 static void tweak_split_index(struct index_state *istate)
1773 switch (git_config_get_split_index()) {
1774 case -1: /* unset: do nothing */
1775 break;
1776 case 0: /* false */
1777 remove_split_index(istate);
1778 break;
1779 case 1: /* true */
1780 add_split_index(istate);
1781 break;
1782 default: /* unknown value: do nothing */
1783 break;
1787 static void post_read_index_from(struct index_state *istate)
1789 check_ce_order(istate);
1790 tweak_untracked_cache(istate);
1791 tweak_split_index(istate);
1792 tweak_fsmonitor(istate);
1795 /* remember to discard_cache() before reading a different cache! */
1796 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1798 int fd, i;
1799 struct stat st;
1800 unsigned long src_offset;
1801 struct cache_header *hdr;
1802 void *mmap;
1803 size_t mmap_size;
1804 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1806 if (istate->initialized)
1807 return istate->cache_nr;
1809 istate->timestamp.sec = 0;
1810 istate->timestamp.nsec = 0;
1811 fd = open(path, O_RDONLY);
1812 if (fd < 0) {
1813 if (!must_exist && errno == ENOENT)
1814 return 0;
1815 die_errno("%s: index file open failed", path);
1818 if (fstat(fd, &st))
1819 die_errno("cannot stat the open index");
1821 mmap_size = xsize_t(st.st_size);
1822 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
1823 die("index file smaller than expected");
1825 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1826 if (mmap == MAP_FAILED)
1827 die_errno("unable to map index file");
1828 close(fd);
1830 hdr = mmap;
1831 if (verify_hdr(hdr, mmap_size) < 0)
1832 goto unmap;
1834 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
1835 istate->version = ntohl(hdr->hdr_version);
1836 istate->cache_nr = ntohl(hdr->hdr_entries);
1837 istate->cache_alloc = alloc_nr(istate->cache_nr);
1838 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1839 istate->initialized = 1;
1841 if (istate->version == 4)
1842 previous_name = &previous_name_buf;
1843 else
1844 previous_name = NULL;
1846 src_offset = sizeof(*hdr);
1847 for (i = 0; i < istate->cache_nr; i++) {
1848 struct ondisk_cache_entry *disk_ce;
1849 struct cache_entry *ce;
1850 unsigned long consumed;
1852 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1853 ce = create_from_disk(disk_ce, &consumed, previous_name);
1854 set_index_entry(istate, i, ce);
1856 src_offset += consumed;
1858 strbuf_release(&previous_name_buf);
1859 istate->timestamp.sec = st.st_mtime;
1860 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1862 while (src_offset <= mmap_size - the_hash_algo->rawsz - 8) {
1863 /* After an array of active_nr index entries,
1864 * there can be arbitrary number of extended
1865 * sections, each of which is prefixed with
1866 * extension name (4-byte) and section length
1867 * in 4-byte network byte order.
1869 uint32_t extsize;
1870 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1871 extsize = ntohl(extsize);
1872 if (read_index_extension(istate,
1873 (const char *) mmap + src_offset,
1874 (char *) mmap + src_offset + 8,
1875 extsize) < 0)
1876 goto unmap;
1877 src_offset += 8;
1878 src_offset += extsize;
1880 munmap(mmap, mmap_size);
1881 return istate->cache_nr;
1883 unmap:
1884 munmap(mmap, mmap_size);
1885 die("index file corrupt");
1889 * Signal that the shared index is used by updating its mtime.
1891 * This way, shared index can be removed if they have not been used
1892 * for some time.
1894 static void freshen_shared_index(const char *shared_index, int warn)
1896 if (!check_and_freshen_file(shared_index, 1) && warn)
1897 warning("could not freshen shared index '%s'", shared_index);
1900 int read_index_from(struct index_state *istate, const char *path,
1901 const char *gitdir)
1903 uint64_t start = getnanotime();
1904 struct split_index *split_index;
1905 int ret;
1906 char *base_oid_hex;
1907 char *base_path;
1909 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1910 if (istate->initialized)
1911 return istate->cache_nr;
1913 ret = do_read_index(istate, path, 0);
1914 trace_performance_since(start, "read cache %s", path);
1916 split_index = istate->split_index;
1917 if (!split_index || is_null_oid(&split_index->base_oid)) {
1918 post_read_index_from(istate);
1919 return ret;
1922 if (split_index->base)
1923 discard_index(split_index->base);
1924 else
1925 split_index->base = xcalloc(1, sizeof(*split_index->base));
1927 base_oid_hex = oid_to_hex(&split_index->base_oid);
1928 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
1929 ret = do_read_index(split_index->base, base_path, 1);
1930 if (oidcmp(&split_index->base_oid, &split_index->base->oid))
1931 die("broken index, expect %s in %s, got %s",
1932 base_oid_hex, base_path,
1933 oid_to_hex(&split_index->base->oid));
1935 freshen_shared_index(base_path, 0);
1936 merge_base_index(istate);
1937 post_read_index_from(istate);
1938 trace_performance_since(start, "read cache %s", base_path);
1939 free(base_path);
1940 return ret;
1943 int is_index_unborn(struct index_state *istate)
1945 return (!istate->cache_nr && !istate->timestamp.sec);
1948 int discard_index(struct index_state *istate)
1950 int i;
1952 for (i = 0; i < istate->cache_nr; i++) {
1953 if (istate->cache[i]->index &&
1954 istate->split_index &&
1955 istate->split_index->base &&
1956 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1957 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1958 continue;
1959 free(istate->cache[i]);
1961 resolve_undo_clear_index(istate);
1962 istate->cache_nr = 0;
1963 istate->cache_changed = 0;
1964 istate->timestamp.sec = 0;
1965 istate->timestamp.nsec = 0;
1966 free_name_hash(istate);
1967 cache_tree_free(&(istate->cache_tree));
1968 istate->initialized = 0;
1969 FREE_AND_NULL(istate->cache);
1970 istate->cache_alloc = 0;
1971 discard_split_index(istate);
1972 free_untracked_cache(istate->untracked);
1973 istate->untracked = NULL;
1974 return 0;
1977 int unmerged_index(const struct index_state *istate)
1979 int i;
1980 for (i = 0; i < istate->cache_nr; i++) {
1981 if (ce_stage(istate->cache[i]))
1982 return 1;
1984 return 0;
1987 #define WRITE_BUFFER_SIZE 8192
1988 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1989 static unsigned long write_buffer_len;
1991 static int ce_write_flush(git_hash_ctx *context, int fd)
1993 unsigned int buffered = write_buffer_len;
1994 if (buffered) {
1995 the_hash_algo->update_fn(context, write_buffer, buffered);
1996 if (write_in_full(fd, write_buffer, buffered) < 0)
1997 return -1;
1998 write_buffer_len = 0;
2000 return 0;
2003 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2005 while (len) {
2006 unsigned int buffered = write_buffer_len;
2007 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2008 if (partial > len)
2009 partial = len;
2010 memcpy(write_buffer + buffered, data, partial);
2011 buffered += partial;
2012 if (buffered == WRITE_BUFFER_SIZE) {
2013 write_buffer_len = buffered;
2014 if (ce_write_flush(context, fd))
2015 return -1;
2016 buffered = 0;
2018 write_buffer_len = buffered;
2019 len -= partial;
2020 data = (char *) data + partial;
2022 return 0;
2025 static int write_index_ext_header(git_hash_ctx *context, int fd,
2026 unsigned int ext, unsigned int sz)
2028 ext = htonl(ext);
2029 sz = htonl(sz);
2030 return ((ce_write(context, fd, &ext, 4) < 0) ||
2031 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2034 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2036 unsigned int left = write_buffer_len;
2038 if (left) {
2039 write_buffer_len = 0;
2040 the_hash_algo->update_fn(context, write_buffer, left);
2043 /* Flush first if not enough space for hash signature */
2044 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2045 if (write_in_full(fd, write_buffer, left) < 0)
2046 return -1;
2047 left = 0;
2050 /* Append the hash signature at the end */
2051 the_hash_algo->final_fn(write_buffer + left, context);
2052 hashcpy(hash, write_buffer + left);
2053 left += the_hash_algo->rawsz;
2054 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2057 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2060 * The only thing we care about in this function is to smudge the
2061 * falsely clean entry due to touch-update-touch race, so we leave
2062 * everything else as they are. We are called for entries whose
2063 * ce_stat_data.sd_mtime match the index file mtime.
2065 * Note that this actually does not do much for gitlinks, for
2066 * which ce_match_stat_basic() always goes to the actual
2067 * contents. The caller checks with is_racy_timestamp() which
2068 * always says "no" for gitlinks, so we are not called for them ;-)
2070 struct stat st;
2072 if (lstat(ce->name, &st) < 0)
2073 return;
2074 if (ce_match_stat_basic(ce, &st))
2075 return;
2076 if (ce_modified_check_fs(ce, &st)) {
2077 /* This is "racily clean"; smudge it. Note that this
2078 * is a tricky code. At first glance, it may appear
2079 * that it can break with this sequence:
2081 * $ echo xyzzy >frotz
2082 * $ git-update-index --add frotz
2083 * $ : >frotz
2084 * $ sleep 3
2085 * $ echo filfre >nitfol
2086 * $ git-update-index --add nitfol
2088 * but it does not. When the second update-index runs,
2089 * it notices that the entry "frotz" has the same timestamp
2090 * as index, and if we were to smudge it by resetting its
2091 * size to zero here, then the object name recorded
2092 * in index is the 6-byte file but the cached stat information
2093 * becomes zero --- which would then match what we would
2094 * obtain from the filesystem next time we stat("frotz").
2096 * However, the second update-index, before calling
2097 * this function, notices that the cached size is 6
2098 * bytes and what is on the filesystem is an empty
2099 * file, and never calls us, so the cached size information
2100 * for "frotz" stays 6 which does not match the filesystem.
2102 ce->ce_stat_data.sd_size = 0;
2106 /* Copy miscellaneous fields but not the name */
2107 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2108 struct cache_entry *ce)
2110 short flags;
2112 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2113 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2114 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2115 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2116 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2117 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2118 ondisk->mode = htonl(ce->ce_mode);
2119 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2120 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2121 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2122 hashcpy(ondisk->sha1, ce->oid.hash);
2124 flags = ce->ce_flags & ~CE_NAMEMASK;
2125 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2126 ondisk->flags = htons(flags);
2127 if (ce->ce_flags & CE_EXTENDED) {
2128 struct ondisk_cache_entry_extended *ondisk2;
2129 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2130 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2134 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2135 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2137 int size;
2138 int result;
2139 unsigned int saved_namelen;
2140 int stripped_name = 0;
2141 static unsigned char padding[8] = { 0x00 };
2143 if (ce->ce_flags & CE_STRIP_NAME) {
2144 saved_namelen = ce_namelen(ce);
2145 ce->ce_namelen = 0;
2146 stripped_name = 1;
2149 if (ce->ce_flags & CE_EXTENDED)
2150 size = offsetof(struct ondisk_cache_entry_extended, name);
2151 else
2152 size = offsetof(struct ondisk_cache_entry, name);
2154 if (!previous_name) {
2155 int len = ce_namelen(ce);
2156 copy_cache_entry_to_ondisk(ondisk, ce);
2157 result = ce_write(c, fd, ondisk, size);
2158 if (!result)
2159 result = ce_write(c, fd, ce->name, len);
2160 if (!result)
2161 result = ce_write(c, fd, padding, align_padding_size(size, len));
2162 } else {
2163 int common, to_remove, prefix_size;
2164 unsigned char to_remove_vi[16];
2165 for (common = 0;
2166 (ce->name[common] &&
2167 common < previous_name->len &&
2168 ce->name[common] == previous_name->buf[common]);
2169 common++)
2170 ; /* still matching */
2171 to_remove = previous_name->len - common;
2172 prefix_size = encode_varint(to_remove, to_remove_vi);
2174 copy_cache_entry_to_ondisk(ondisk, ce);
2175 result = ce_write(c, fd, ondisk, size);
2176 if (!result)
2177 result = ce_write(c, fd, to_remove_vi, prefix_size);
2178 if (!result)
2179 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2180 if (!result)
2181 result = ce_write(c, fd, padding, 1);
2183 strbuf_splice(previous_name, common, to_remove,
2184 ce->name + common, ce_namelen(ce) - common);
2186 if (stripped_name) {
2187 ce->ce_namelen = saved_namelen;
2188 ce->ce_flags &= ~CE_STRIP_NAME;
2191 return result;
2195 * This function verifies if index_state has the correct sha1 of the
2196 * index file. Don't die if we have any other failure, just return 0.
2198 static int verify_index_from(const struct index_state *istate, const char *path)
2200 int fd;
2201 ssize_t n;
2202 struct stat st;
2203 unsigned char hash[GIT_MAX_RAWSZ];
2205 if (!istate->initialized)
2206 return 0;
2208 fd = open(path, O_RDONLY);
2209 if (fd < 0)
2210 return 0;
2212 if (fstat(fd, &st))
2213 goto out;
2215 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2216 goto out;
2218 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2219 if (n != the_hash_algo->rawsz)
2220 goto out;
2222 if (hashcmp(istate->oid.hash, hash))
2223 goto out;
2225 close(fd);
2226 return 1;
2228 out:
2229 close(fd);
2230 return 0;
2233 static int verify_index(const struct index_state *istate)
2235 return verify_index_from(istate, get_index_file());
2238 static int has_racy_timestamp(struct index_state *istate)
2240 int entries = istate->cache_nr;
2241 int i;
2243 for (i = 0; i < entries; i++) {
2244 struct cache_entry *ce = istate->cache[i];
2245 if (is_racy_timestamp(istate, ce))
2246 return 1;
2248 return 0;
2251 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2253 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2254 verify_index(istate))
2255 write_locked_index(istate, lockfile, COMMIT_LOCK);
2256 else
2257 rollback_lock_file(lockfile);
2261 * On success, `tempfile` is closed. If it is the temporary file
2262 * of a `struct lock_file`, we will therefore effectively perform
2263 * a 'close_lock_file_gently()`. Since that is an implementation
2264 * detail of lockfiles, callers of `do_write_index()` should not
2265 * rely on it.
2267 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2268 int strip_extensions)
2270 uint64_t start = getnanotime();
2271 int newfd = tempfile->fd;
2272 git_hash_ctx c;
2273 struct cache_header hdr;
2274 int i, err = 0, removed, extended, hdr_version;
2275 struct cache_entry **cache = istate->cache;
2276 int entries = istate->cache_nr;
2277 struct stat st;
2278 struct ondisk_cache_entry_extended ondisk;
2279 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2280 int drop_cache_tree = istate->drop_cache_tree;
2282 for (i = removed = extended = 0; i < entries; i++) {
2283 if (cache[i]->ce_flags & CE_REMOVE)
2284 removed++;
2286 /* reduce extended entries if possible */
2287 cache[i]->ce_flags &= ~CE_EXTENDED;
2288 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2289 extended++;
2290 cache[i]->ce_flags |= CE_EXTENDED;
2294 if (!istate->version) {
2295 istate->version = get_index_format_default();
2296 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2297 init_split_index(istate);
2300 /* demote version 3 to version 2 when the latter suffices */
2301 if (istate->version == 3 || istate->version == 2)
2302 istate->version = extended ? 3 : 2;
2304 hdr_version = istate->version;
2306 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2307 hdr.hdr_version = htonl(hdr_version);
2308 hdr.hdr_entries = htonl(entries - removed);
2310 the_hash_algo->init_fn(&c);
2311 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2312 return -1;
2314 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2316 for (i = 0; i < entries; i++) {
2317 struct cache_entry *ce = cache[i];
2318 if (ce->ce_flags & CE_REMOVE)
2319 continue;
2320 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2321 ce_smudge_racily_clean_entry(ce);
2322 if (is_null_oid(&ce->oid)) {
2323 static const char msg[] = "cache entry has null sha1: %s";
2324 static int allow = -1;
2326 if (allow < 0)
2327 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2328 if (allow)
2329 warning(msg, ce->name);
2330 else
2331 err = error(msg, ce->name);
2333 drop_cache_tree = 1;
2335 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2336 err = -1;
2338 if (err)
2339 break;
2341 strbuf_release(&previous_name_buf);
2343 if (err)
2344 return err;
2346 /* Write extension data here */
2347 if (!strip_extensions && istate->split_index) {
2348 struct strbuf sb = STRBUF_INIT;
2350 err = write_link_extension(&sb, istate) < 0 ||
2351 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2352 sb.len) < 0 ||
2353 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2354 strbuf_release(&sb);
2355 if (err)
2356 return -1;
2358 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2359 struct strbuf sb = STRBUF_INIT;
2361 cache_tree_write(&sb, istate->cache_tree);
2362 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2363 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2364 strbuf_release(&sb);
2365 if (err)
2366 return -1;
2368 if (!strip_extensions && istate->resolve_undo) {
2369 struct strbuf sb = STRBUF_INIT;
2371 resolve_undo_write(&sb, istate->resolve_undo);
2372 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2373 sb.len) < 0
2374 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2375 strbuf_release(&sb);
2376 if (err)
2377 return -1;
2379 if (!strip_extensions && istate->untracked) {
2380 struct strbuf sb = STRBUF_INIT;
2382 write_untracked_extension(&sb, istate->untracked);
2383 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2384 sb.len) < 0 ||
2385 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2386 strbuf_release(&sb);
2387 if (err)
2388 return -1;
2390 if (!strip_extensions && istate->fsmonitor_last_update) {
2391 struct strbuf sb = STRBUF_INIT;
2393 write_fsmonitor_extension(&sb, istate);
2394 err = write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2395 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2396 strbuf_release(&sb);
2397 if (err)
2398 return -1;
2401 if (ce_flush(&c, newfd, istate->oid.hash))
2402 return -1;
2403 if (close_tempfile_gently(tempfile)) {
2404 error(_("could not close '%s'"), tempfile->filename.buf);
2405 return -1;
2407 if (stat(tempfile->filename.buf, &st))
2408 return -1;
2409 istate->timestamp.sec = (unsigned int)st.st_mtime;
2410 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2411 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
2412 return 0;
2415 void set_alternate_index_output(const char *name)
2417 alternate_index_output = name;
2420 static int commit_locked_index(struct lock_file *lk)
2422 if (alternate_index_output)
2423 return commit_lock_file_to(lk, alternate_index_output);
2424 else
2425 return commit_lock_file(lk);
2428 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2429 unsigned flags)
2431 int ret = do_write_index(istate, lock->tempfile, 0);
2432 if (ret)
2433 return ret;
2434 if (flags & COMMIT_LOCK)
2435 return commit_locked_index(lock);
2436 return close_lock_file_gently(lock);
2439 static int write_split_index(struct index_state *istate,
2440 struct lock_file *lock,
2441 unsigned flags)
2443 int ret;
2444 prepare_to_write_split_index(istate);
2445 ret = do_write_locked_index(istate, lock, flags);
2446 finish_writing_split_index(istate);
2447 return ret;
2450 static const char *shared_index_expire = "2.weeks.ago";
2452 static unsigned long get_shared_index_expire_date(void)
2454 static unsigned long shared_index_expire_date;
2455 static int shared_index_expire_date_prepared;
2457 if (!shared_index_expire_date_prepared) {
2458 git_config_get_expiry("splitindex.sharedindexexpire",
2459 &shared_index_expire);
2460 shared_index_expire_date = approxidate(shared_index_expire);
2461 shared_index_expire_date_prepared = 1;
2464 return shared_index_expire_date;
2467 static int should_delete_shared_index(const char *shared_index_path)
2469 struct stat st;
2470 unsigned long expiration;
2472 /* Check timestamp */
2473 expiration = get_shared_index_expire_date();
2474 if (!expiration)
2475 return 0;
2476 if (stat(shared_index_path, &st))
2477 return error_errno(_("could not stat '%s'"), shared_index_path);
2478 if (st.st_mtime > expiration)
2479 return 0;
2481 return 1;
2484 static int clean_shared_index_files(const char *current_hex)
2486 struct dirent *de;
2487 DIR *dir = opendir(get_git_dir());
2489 if (!dir)
2490 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2492 while ((de = readdir(dir)) != NULL) {
2493 const char *sha1_hex;
2494 const char *shared_index_path;
2495 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2496 continue;
2497 if (!strcmp(sha1_hex, current_hex))
2498 continue;
2499 shared_index_path = git_path("%s", de->d_name);
2500 if (should_delete_shared_index(shared_index_path) > 0 &&
2501 unlink(shared_index_path))
2502 warning_errno(_("unable to unlink: %s"), shared_index_path);
2504 closedir(dir);
2506 return 0;
2509 static int write_shared_index(struct index_state *istate,
2510 struct tempfile **temp)
2512 struct split_index *si = istate->split_index;
2513 int ret;
2515 move_cache_to_base_index(istate);
2516 ret = do_write_index(si->base, *temp, 1);
2517 if (ret)
2518 return ret;
2519 ret = adjust_shared_perm(get_tempfile_path(*temp));
2520 if (ret) {
2521 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
2522 return ret;
2524 ret = rename_tempfile(temp,
2525 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
2526 if (!ret) {
2527 oidcpy(&si->base_oid, &si->base->oid);
2528 clean_shared_index_files(oid_to_hex(&si->base->oid));
2531 return ret;
2534 static const int default_max_percent_split_change = 20;
2536 static int too_many_not_shared_entries(struct index_state *istate)
2538 int i, not_shared = 0;
2539 int max_split = git_config_get_max_percent_split_change();
2541 switch (max_split) {
2542 case -1:
2543 /* not or badly configured: use the default value */
2544 max_split = default_max_percent_split_change;
2545 break;
2546 case 0:
2547 return 1; /* 0% means always write a new shared index */
2548 case 100:
2549 return 0; /* 100% means never write a new shared index */
2550 default:
2551 break; /* just use the configured value */
2554 /* Count not shared entries */
2555 for (i = 0; i < istate->cache_nr; i++) {
2556 struct cache_entry *ce = istate->cache[i];
2557 if (!ce->index)
2558 not_shared++;
2561 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2564 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2565 unsigned flags)
2567 int new_shared_index, ret;
2568 struct split_index *si = istate->split_index;
2570 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
2571 if (flags & COMMIT_LOCK)
2572 rollback_lock_file(lock);
2573 return 0;
2576 if (istate->fsmonitor_last_update)
2577 fill_fsmonitor_bitmap(istate);
2579 if (!si || alternate_index_output ||
2580 (istate->cache_changed & ~EXTMASK)) {
2581 if (si)
2582 oidclr(&si->base_oid);
2583 ret = do_write_locked_index(istate, lock, flags);
2584 goto out;
2587 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2588 int v = si->base_oid.hash[0];
2589 if ((v & 15) < 6)
2590 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2592 if (too_many_not_shared_entries(istate))
2593 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2595 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2597 if (new_shared_index) {
2598 struct tempfile *temp;
2599 int saved_errno;
2601 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2602 if (!temp) {
2603 oidclr(&si->base_oid);
2604 ret = do_write_locked_index(istate, lock, flags);
2605 goto out;
2607 ret = write_shared_index(istate, &temp);
2609 saved_errno = errno;
2610 if (is_tempfile_active(temp))
2611 delete_tempfile(&temp);
2612 errno = saved_errno;
2614 if (ret)
2615 goto out;
2618 ret = write_split_index(istate, lock, flags);
2620 /* Freshen the shared index only if the split-index was written */
2621 if (!ret && !new_shared_index) {
2622 const char *shared_index = git_path("sharedindex.%s",
2623 oid_to_hex(&si->base_oid));
2624 freshen_shared_index(shared_index, 1);
2627 out:
2628 if (flags & COMMIT_LOCK)
2629 rollback_lock_file(lock);
2630 return ret;
2634 * Read the index file that is potentially unmerged into given
2635 * index_state, dropping any unmerged entries. Returns true if
2636 * the index is unmerged. Callers who want to refuse to work
2637 * from an unmerged state can call this and check its return value,
2638 * instead of calling read_cache().
2640 int read_index_unmerged(struct index_state *istate)
2642 int i;
2643 int unmerged = 0;
2645 read_index(istate);
2646 for (i = 0; i < istate->cache_nr; i++) {
2647 struct cache_entry *ce = istate->cache[i];
2648 struct cache_entry *new_ce;
2649 int size, len;
2651 if (!ce_stage(ce))
2652 continue;
2653 unmerged = 1;
2654 len = ce_namelen(ce);
2655 size = cache_entry_size(len);
2656 new_ce = xcalloc(1, size);
2657 memcpy(new_ce->name, ce->name, len);
2658 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2659 new_ce->ce_namelen = len;
2660 new_ce->ce_mode = ce->ce_mode;
2661 if (add_index_entry(istate, new_ce, 0))
2662 return error("%s: cannot drop to stage #0",
2663 new_ce->name);
2665 return unmerged;
2669 * Returns 1 if the path is an "other" path with respect to
2670 * the index; that is, the path is not mentioned in the index at all,
2671 * either as a file, a directory with some files in the index,
2672 * or as an unmerged entry.
2674 * We helpfully remove a trailing "/" from directories so that
2675 * the output of read_directory can be used as-is.
2677 int index_name_is_other(const struct index_state *istate, const char *name,
2678 int namelen)
2680 int pos;
2681 if (namelen && name[namelen - 1] == '/')
2682 namelen--;
2683 pos = index_name_pos(istate, name, namelen);
2684 if (0 <= pos)
2685 return 0; /* exact match */
2686 pos = -pos - 1;
2687 if (pos < istate->cache_nr) {
2688 struct cache_entry *ce = istate->cache[pos];
2689 if (ce_namelen(ce) == namelen &&
2690 !memcmp(ce->name, name, namelen))
2691 return 0; /* Yup, this one exists unmerged */
2693 return 1;
2696 void *read_blob_data_from_index(const struct index_state *istate,
2697 const char *path, unsigned long *size)
2699 int pos, len;
2700 unsigned long sz;
2701 enum object_type type;
2702 void *data;
2704 len = strlen(path);
2705 pos = index_name_pos(istate, path, len);
2706 if (pos < 0) {
2708 * We might be in the middle of a merge, in which
2709 * case we would read stage #2 (ours).
2711 int i;
2712 for (i = -pos - 1;
2713 (pos < 0 && i < istate->cache_nr &&
2714 !strcmp(istate->cache[i]->name, path));
2715 i++)
2716 if (ce_stage(istate->cache[i]) == 2)
2717 pos = i;
2719 if (pos < 0)
2720 return NULL;
2721 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
2722 if (!data || type != OBJ_BLOB) {
2723 free(data);
2724 return NULL;
2726 if (size)
2727 *size = sz;
2728 return data;
2731 void stat_validity_clear(struct stat_validity *sv)
2733 FREE_AND_NULL(sv->sd);
2736 int stat_validity_check(struct stat_validity *sv, const char *path)
2738 struct stat st;
2740 if (stat(path, &st) < 0)
2741 return sv->sd == NULL;
2742 if (!sv->sd)
2743 return 0;
2744 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2747 void stat_validity_update(struct stat_validity *sv, int fd)
2749 struct stat st;
2751 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2752 stat_validity_clear(sv);
2753 else {
2754 if (!sv->sd)
2755 sv->sd = xcalloc(1, sizeof(struct stat_data));
2756 fill_stat_data(sv->sd, &st);
2760 void move_index_extensions(struct index_state *dst, struct index_state *src)
2762 dst->untracked = src->untracked;
2763 src->untracked = NULL;