utf8: fix overflow when returning string width
[alt-git.git] / read-cache.c
blob29144cf879e795c3f5415cd5dd50f53a5dfb0cac
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #include "cache.h"
7 #include "config.h"
8 #include "diff.h"
9 #include "diffcore.h"
10 #include "tempfile.h"
11 #include "lockfile.h"
12 #include "cache-tree.h"
13 #include "refs.h"
14 #include "dir.h"
15 #include "object-store.h"
16 #include "tree.h"
17 #include "commit.h"
18 #include "blob.h"
19 #include "resolve-undo.h"
20 #include "run-command.h"
21 #include "strbuf.h"
22 #include "varint.h"
23 #include "split-index.h"
24 #include "utf8.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
27 #include "progress.h"
29 /* Mask for the name length in ce_flags in the on-disk index */
31 #define CE_NAMEMASK (0x0fff)
33 /* Index extensions.
35 * The first letter should be 'A'..'Z' for extensions that are not
36 * necessary for a correct operation (i.e. optimization data).
37 * When new extensions are added that _needs_ to be understood in
38 * order to correctly interpret the index file, pick character that
39 * is outside the range, to cause the reader to abort.
42 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
43 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
44 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
45 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
46 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
47 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
48 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
49 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
51 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
52 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
53 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
54 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
58 * This is an estimate of the pathname length in the index. We use
59 * this for V4 index files to guess the un-deltafied size of the index
60 * in memory because of pathname deltafication. This is not required
61 * for V2/V3 index formats because their pathnames are not compressed.
62 * If the initial amount of memory set aside is not sufficient, the
63 * mem pool will allocate extra memory.
65 #define CACHE_ENTRY_PATH_LENGTH 80
67 static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
69 struct cache_entry *ce;
70 ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
71 ce->mem_pool_allocated = 1;
72 return ce;
75 static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
77 struct cache_entry * ce;
78 ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
79 ce->mem_pool_allocated = 1;
80 return ce;
83 static struct mem_pool *find_mem_pool(struct index_state *istate)
85 struct mem_pool **pool_ptr;
87 if (istate->split_index && istate->split_index->base)
88 pool_ptr = &istate->split_index->base->ce_mem_pool;
89 else
90 pool_ptr = &istate->ce_mem_pool;
92 if (!*pool_ptr) {
93 *pool_ptr = xmalloc(sizeof(**pool_ptr));
94 mem_pool_init(*pool_ptr, 0);
97 return *pool_ptr;
100 static const char *alternate_index_output;
102 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
104 istate->cache[nr] = ce;
105 add_name_hash(istate, ce);
108 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
110 struct cache_entry *old = istate->cache[nr];
112 replace_index_entry_in_base(istate, old, ce);
113 remove_name_hash(istate, old);
114 discard_cache_entry(old);
115 ce->ce_flags &= ~CE_HASHED;
116 set_index_entry(istate, nr, ce);
117 ce->ce_flags |= CE_UPDATE_IN_BASE;
118 mark_fsmonitor_invalid(istate, ce);
119 istate->cache_changed |= CE_ENTRY_CHANGED;
122 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
124 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
125 int namelen = strlen(new_name);
127 new_entry = make_empty_cache_entry(istate, namelen);
128 copy_cache_entry(new_entry, old_entry);
129 new_entry->ce_flags &= ~CE_HASHED;
130 new_entry->ce_namelen = namelen;
131 new_entry->index = 0;
132 memcpy(new_entry->name, new_name, namelen + 1);
134 cache_tree_invalidate_path(istate, old_entry->name);
135 untracked_cache_remove_from_index(istate, old_entry->name);
136 remove_index_entry_at(istate, nr);
137 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
140 void fill_stat_data(struct stat_data *sd, struct stat *st)
142 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
143 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
144 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
145 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
146 sd->sd_dev = st->st_dev;
147 sd->sd_ino = st->st_ino;
148 sd->sd_uid = st->st_uid;
149 sd->sd_gid = st->st_gid;
150 sd->sd_size = st->st_size;
153 int match_stat_data(const struct stat_data *sd, struct stat *st)
155 int changed = 0;
157 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
158 changed |= MTIME_CHANGED;
159 if (trust_ctime && check_stat &&
160 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
161 changed |= CTIME_CHANGED;
163 #ifdef USE_NSEC
164 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
165 changed |= MTIME_CHANGED;
166 if (trust_ctime && check_stat &&
167 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
168 changed |= CTIME_CHANGED;
169 #endif
171 if (check_stat) {
172 if (sd->sd_uid != (unsigned int) st->st_uid ||
173 sd->sd_gid != (unsigned int) st->st_gid)
174 changed |= OWNER_CHANGED;
175 if (sd->sd_ino != (unsigned int) st->st_ino)
176 changed |= INODE_CHANGED;
179 #ifdef USE_STDEV
181 * st_dev breaks on network filesystems where different
182 * clients will have different views of what "device"
183 * the filesystem is on
185 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
186 changed |= INODE_CHANGED;
187 #endif
189 if (sd->sd_size != (unsigned int) st->st_size)
190 changed |= DATA_CHANGED;
192 return changed;
196 * This only updates the "non-critical" parts of the directory
197 * cache, ie the parts that aren't tracked by GIT, and only used
198 * to validate the cache.
200 void fill_stat_cache_info(struct index_state *istate, struct cache_entry *ce, struct stat *st)
202 fill_stat_data(&ce->ce_stat_data, st);
204 if (assume_unchanged)
205 ce->ce_flags |= CE_VALID;
207 if (S_ISREG(st->st_mode)) {
208 ce_mark_uptodate(ce);
209 mark_fsmonitor_valid(istate, ce);
213 static int ce_compare_data(struct index_state *istate,
214 const struct cache_entry *ce,
215 struct stat *st)
217 int match = -1;
218 int fd = git_open_cloexec(ce->name, O_RDONLY);
220 if (fd >= 0) {
221 struct object_id oid;
222 if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
223 match = !oideq(&oid, &ce->oid);
224 /* index_fd() closed the file descriptor already */
226 return match;
229 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
231 int match = -1;
232 void *buffer;
233 unsigned long size;
234 enum object_type type;
235 struct strbuf sb = STRBUF_INIT;
237 if (strbuf_readlink(&sb, ce->name, expected_size))
238 return -1;
240 buffer = read_object_file(&ce->oid, &type, &size);
241 if (buffer) {
242 if (size == sb.len)
243 match = memcmp(buffer, sb.buf, size);
244 free(buffer);
246 strbuf_release(&sb);
247 return match;
250 static int ce_compare_gitlink(const struct cache_entry *ce)
252 struct object_id oid;
255 * We don't actually require that the .git directory
256 * under GITLINK directory be a valid git directory. It
257 * might even be missing (in case nobody populated that
258 * sub-project).
260 * If so, we consider it always to match.
262 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
263 return 0;
264 return !oideq(&oid, &ce->oid);
267 static int ce_modified_check_fs(struct index_state *istate,
268 const struct cache_entry *ce,
269 struct stat *st)
271 switch (st->st_mode & S_IFMT) {
272 case S_IFREG:
273 if (ce_compare_data(istate, ce, st))
274 return DATA_CHANGED;
275 break;
276 case S_IFLNK:
277 if (ce_compare_link(ce, xsize_t(st->st_size)))
278 return DATA_CHANGED;
279 break;
280 case S_IFDIR:
281 if (S_ISGITLINK(ce->ce_mode))
282 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
283 /* else fallthrough */
284 default:
285 return TYPE_CHANGED;
287 return 0;
290 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
292 unsigned int changed = 0;
294 if (ce->ce_flags & CE_REMOVE)
295 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
297 switch (ce->ce_mode & S_IFMT) {
298 case S_IFREG:
299 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
300 /* We consider only the owner x bit to be relevant for
301 * "mode changes"
303 if (trust_executable_bit &&
304 (0100 & (ce->ce_mode ^ st->st_mode)))
305 changed |= MODE_CHANGED;
306 break;
307 case S_IFLNK:
308 if (!S_ISLNK(st->st_mode) &&
309 (has_symlinks || !S_ISREG(st->st_mode)))
310 changed |= TYPE_CHANGED;
311 break;
312 case S_IFGITLINK:
313 /* We ignore most of the st_xxx fields for gitlinks */
314 if (!S_ISDIR(st->st_mode))
315 changed |= TYPE_CHANGED;
316 else if (ce_compare_gitlink(ce))
317 changed |= DATA_CHANGED;
318 return changed;
319 default:
320 BUG("unsupported ce_mode: %o", ce->ce_mode);
323 changed |= match_stat_data(&ce->ce_stat_data, st);
325 /* Racily smudged entry? */
326 if (!ce->ce_stat_data.sd_size) {
327 if (!is_empty_blob_sha1(ce->oid.hash))
328 changed |= DATA_CHANGED;
331 return changed;
334 static int is_racy_stat(const struct index_state *istate,
335 const struct stat_data *sd)
337 return (istate->timestamp.sec &&
338 #ifdef USE_NSEC
339 /* nanosecond timestamped files can also be racy! */
340 (istate->timestamp.sec < sd->sd_mtime.sec ||
341 (istate->timestamp.sec == sd->sd_mtime.sec &&
342 istate->timestamp.nsec <= sd->sd_mtime.nsec))
343 #else
344 istate->timestamp.sec <= sd->sd_mtime.sec
345 #endif
349 int is_racy_timestamp(const struct index_state *istate,
350 const struct cache_entry *ce)
352 return (!S_ISGITLINK(ce->ce_mode) &&
353 is_racy_stat(istate, &ce->ce_stat_data));
356 int match_stat_data_racy(const struct index_state *istate,
357 const struct stat_data *sd, struct stat *st)
359 if (is_racy_stat(istate, sd))
360 return MTIME_CHANGED;
361 return match_stat_data(sd, st);
364 int ie_match_stat(struct index_state *istate,
365 const struct cache_entry *ce, struct stat *st,
366 unsigned int options)
368 unsigned int changed;
369 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
370 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
371 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
372 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
374 if (!ignore_fsmonitor)
375 refresh_fsmonitor(istate);
377 * If it's marked as always valid in the index, it's
378 * valid whatever the checked-out copy says.
380 * skip-worktree has the same effect with higher precedence
382 if (!ignore_skip_worktree && ce_skip_worktree(ce))
383 return 0;
384 if (!ignore_valid && (ce->ce_flags & CE_VALID))
385 return 0;
386 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
387 return 0;
390 * Intent-to-add entries have not been added, so the index entry
391 * by definition never matches what is in the work tree until it
392 * actually gets added.
394 if (ce_intent_to_add(ce))
395 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
397 changed = ce_match_stat_basic(ce, st);
400 * Within 1 second of this sequence:
401 * echo xyzzy >file && git-update-index --add file
402 * running this command:
403 * echo frotz >file
404 * would give a falsely clean cache entry. The mtime and
405 * length match the cache, and other stat fields do not change.
407 * We could detect this at update-index time (the cache entry
408 * being registered/updated records the same time as "now")
409 * and delay the return from git-update-index, but that would
410 * effectively mean we can make at most one commit per second,
411 * which is not acceptable. Instead, we check cache entries
412 * whose mtime are the same as the index file timestamp more
413 * carefully than others.
415 if (!changed && is_racy_timestamp(istate, ce)) {
416 if (assume_racy_is_modified)
417 changed |= DATA_CHANGED;
418 else
419 changed |= ce_modified_check_fs(istate, ce, st);
422 return changed;
425 int ie_modified(struct index_state *istate,
426 const struct cache_entry *ce,
427 struct stat *st, unsigned int options)
429 int changed, changed_fs;
431 changed = ie_match_stat(istate, ce, st, options);
432 if (!changed)
433 return 0;
435 * If the mode or type has changed, there's no point in trying
436 * to refresh the entry - it's not going to match
438 if (changed & (MODE_CHANGED | TYPE_CHANGED))
439 return changed;
442 * Immediately after read-tree or update-index --cacheinfo,
443 * the length field is zero, as we have never even read the
444 * lstat(2) information once, and we cannot trust DATA_CHANGED
445 * returned by ie_match_stat() which in turn was returned by
446 * ce_match_stat_basic() to signal that the filesize of the
447 * blob changed. We have to actually go to the filesystem to
448 * see if the contents match, and if so, should answer "unchanged".
450 * The logic does not apply to gitlinks, as ce_match_stat_basic()
451 * already has checked the actual HEAD from the filesystem in the
452 * subproject. If ie_match_stat() already said it is different,
453 * then we know it is.
455 if ((changed & DATA_CHANGED) &&
456 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
457 return changed;
459 changed_fs = ce_modified_check_fs(istate, ce, st);
460 if (changed_fs)
461 return changed | changed_fs;
462 return 0;
465 int base_name_compare(const char *name1, int len1, int mode1,
466 const char *name2, int len2, int mode2)
468 unsigned char c1, c2;
469 int len = len1 < len2 ? len1 : len2;
470 int cmp;
472 cmp = memcmp(name1, name2, len);
473 if (cmp)
474 return cmp;
475 c1 = name1[len];
476 c2 = name2[len];
477 if (!c1 && S_ISDIR(mode1))
478 c1 = '/';
479 if (!c2 && S_ISDIR(mode2))
480 c2 = '/';
481 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
485 * df_name_compare() is identical to base_name_compare(), except it
486 * compares conflicting directory/file entries as equal. Note that
487 * while a directory name compares as equal to a regular file, they
488 * then individually compare _differently_ to a filename that has
489 * a dot after the basename (because '\0' < '.' < '/').
491 * This is used by routines that want to traverse the git namespace
492 * but then handle conflicting entries together when possible.
494 int df_name_compare(const char *name1, int len1, int mode1,
495 const char *name2, int len2, int mode2)
497 int len = len1 < len2 ? len1 : len2, cmp;
498 unsigned char c1, c2;
500 cmp = memcmp(name1, name2, len);
501 if (cmp)
502 return cmp;
503 /* Directories and files compare equal (same length, same name) */
504 if (len1 == len2)
505 return 0;
506 c1 = name1[len];
507 if (!c1 && S_ISDIR(mode1))
508 c1 = '/';
509 c2 = name2[len];
510 if (!c2 && S_ISDIR(mode2))
511 c2 = '/';
512 if (c1 == '/' && !c2)
513 return 0;
514 if (c2 == '/' && !c1)
515 return 0;
516 return c1 - c2;
519 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
521 size_t min_len = (len1 < len2) ? len1 : len2;
522 int cmp = memcmp(name1, name2, min_len);
523 if (cmp)
524 return cmp;
525 if (len1 < len2)
526 return -1;
527 if (len1 > len2)
528 return 1;
529 return 0;
532 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
534 int cmp;
536 cmp = name_compare(name1, len1, name2, len2);
537 if (cmp)
538 return cmp;
540 if (stage1 < stage2)
541 return -1;
542 if (stage1 > stage2)
543 return 1;
544 return 0;
547 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
549 int first, last;
551 first = 0;
552 last = istate->cache_nr;
553 while (last > first) {
554 int next = first + ((last - first) >> 1);
555 struct cache_entry *ce = istate->cache[next];
556 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
557 if (!cmp)
558 return next;
559 if (cmp < 0) {
560 last = next;
561 continue;
563 first = next+1;
565 return -first-1;
568 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
570 return index_name_stage_pos(istate, name, namelen, 0);
573 int remove_index_entry_at(struct index_state *istate, int pos)
575 struct cache_entry *ce = istate->cache[pos];
577 record_resolve_undo(istate, ce);
578 remove_name_hash(istate, ce);
579 save_or_free_index_entry(istate, ce);
580 istate->cache_changed |= CE_ENTRY_REMOVED;
581 istate->cache_nr--;
582 if (pos >= istate->cache_nr)
583 return 0;
584 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
585 istate->cache_nr - pos);
586 return 1;
590 * Remove all cache entries marked for removal, that is where
591 * CE_REMOVE is set in ce_flags. This is much more effective than
592 * calling remove_index_entry_at() for each entry to be removed.
594 void remove_marked_cache_entries(struct index_state *istate, int invalidate)
596 struct cache_entry **ce_array = istate->cache;
597 unsigned int i, j;
599 for (i = j = 0; i < istate->cache_nr; i++) {
600 if (ce_array[i]->ce_flags & CE_REMOVE) {
601 if (invalidate) {
602 cache_tree_invalidate_path(istate,
603 ce_array[i]->name);
604 untracked_cache_remove_from_index(istate,
605 ce_array[i]->name);
607 remove_name_hash(istate, ce_array[i]);
608 save_or_free_index_entry(istate, ce_array[i]);
610 else
611 ce_array[j++] = ce_array[i];
613 if (j == istate->cache_nr)
614 return;
615 istate->cache_changed |= CE_ENTRY_REMOVED;
616 istate->cache_nr = j;
619 int remove_file_from_index(struct index_state *istate, const char *path)
621 int pos = index_name_pos(istate, path, strlen(path));
622 if (pos < 0)
623 pos = -pos-1;
624 cache_tree_invalidate_path(istate, path);
625 untracked_cache_remove_from_index(istate, path);
626 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
627 remove_index_entry_at(istate, pos);
628 return 0;
631 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
633 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
636 static int index_name_pos_also_unmerged(struct index_state *istate,
637 const char *path, int namelen)
639 int pos = index_name_pos(istate, path, namelen);
640 struct cache_entry *ce;
642 if (pos >= 0)
643 return pos;
645 /* maybe unmerged? */
646 pos = -1 - pos;
647 if (pos >= istate->cache_nr ||
648 compare_name((ce = istate->cache[pos]), path, namelen))
649 return -1;
651 /* order of preference: stage 2, 1, 3 */
652 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
653 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
654 !compare_name(ce, path, namelen))
655 pos++;
656 return pos;
659 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
661 int len = ce_namelen(ce);
662 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
666 * If we add a filename that aliases in the cache, we will use the
667 * name that we already have - but we don't want to update the same
668 * alias twice, because that implies that there were actually two
669 * different files with aliasing names!
671 * So we use the CE_ADDED flag to verify that the alias was an old
672 * one before we accept it as
674 static struct cache_entry *create_alias_ce(struct index_state *istate,
675 struct cache_entry *ce,
676 struct cache_entry *alias)
678 int len;
679 struct cache_entry *new_entry;
681 if (alias->ce_flags & CE_ADDED)
682 die(_("will not add file alias '%s' ('%s' already exists in index)"),
683 ce->name, alias->name);
685 /* Ok, create the new entry using the name of the existing alias */
686 len = ce_namelen(alias);
687 new_entry = make_empty_cache_entry(istate, len);
688 memcpy(new_entry->name, alias->name, len);
689 copy_cache_entry(new_entry, ce);
690 save_or_free_index_entry(istate, ce);
691 return new_entry;
694 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
696 struct object_id oid;
697 if (write_object_file("", 0, blob_type, &oid))
698 die(_("cannot create an empty blob in the object database"));
699 oidcpy(&ce->oid, &oid);
702 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
704 int namelen, was_same;
705 mode_t st_mode = st->st_mode;
706 struct cache_entry *ce, *alias = NULL;
707 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
708 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
709 int pretend = flags & ADD_CACHE_PRETEND;
710 int intent_only = flags & ADD_CACHE_INTENT;
711 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
712 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
713 int hash_flags = HASH_WRITE_OBJECT;
714 struct object_id oid;
716 if (flags & ADD_CACHE_RENORMALIZE)
717 hash_flags |= HASH_RENORMALIZE;
719 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
720 return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
722 namelen = strlen(path);
723 if (S_ISDIR(st_mode)) {
724 if (resolve_gitlink_ref(path, "HEAD", &oid) < 0)
725 return error(_("'%s' does not have a commit checked out"), path);
726 while (namelen && path[namelen-1] == '/')
727 namelen--;
729 ce = make_empty_cache_entry(istate, namelen);
730 memcpy(ce->name, path, namelen);
731 ce->ce_namelen = namelen;
732 if (!intent_only)
733 fill_stat_cache_info(istate, ce, st);
734 else
735 ce->ce_flags |= CE_INTENT_TO_ADD;
738 if (trust_executable_bit && has_symlinks) {
739 ce->ce_mode = create_ce_mode(st_mode);
740 } else {
741 /* If there is an existing entry, pick the mode bits and type
742 * from it, otherwise assume unexecutable regular file.
744 struct cache_entry *ent;
745 int pos = index_name_pos_also_unmerged(istate, path, namelen);
747 ent = (0 <= pos) ? istate->cache[pos] : NULL;
748 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
751 /* When core.ignorecase=true, determine if a directory of the same name but differing
752 * case already exists within the Git repository. If it does, ensure the directory
753 * case of the file being added to the repository matches (is folded into) the existing
754 * entry's directory case.
756 if (ignore_case) {
757 adjust_dirname_case(istate, ce->name);
759 if (!(flags & ADD_CACHE_RENORMALIZE)) {
760 alias = index_file_exists(istate, ce->name,
761 ce_namelen(ce), ignore_case);
762 if (alias &&
763 !ce_stage(alias) &&
764 !ie_match_stat(istate, alias, st, ce_option)) {
765 /* Nothing changed, really */
766 if (!S_ISGITLINK(alias->ce_mode))
767 ce_mark_uptodate(alias);
768 alias->ce_flags |= CE_ADDED;
770 discard_cache_entry(ce);
771 return 0;
774 if (!intent_only) {
775 if (index_path(istate, &ce->oid, path, st, hash_flags)) {
776 discard_cache_entry(ce);
777 return error(_("unable to index file '%s'"), path);
779 } else
780 set_object_name_for_intent_to_add_entry(ce);
782 if (ignore_case && alias && different_name(ce, alias))
783 ce = create_alias_ce(istate, ce, alias);
784 ce->ce_flags |= CE_ADDED;
786 /* It was suspected to be racily clean, but it turns out to be Ok */
787 was_same = (alias &&
788 !ce_stage(alias) &&
789 oideq(&alias->oid, &ce->oid) &&
790 ce->ce_mode == alias->ce_mode);
792 if (pretend)
793 discard_cache_entry(ce);
794 else if (add_index_entry(istate, ce, add_option)) {
795 discard_cache_entry(ce);
796 return error(_("unable to add '%s' to index"), path);
798 if (verbose && !was_same)
799 printf("add '%s'\n", path);
800 return 0;
803 int add_file_to_index(struct index_state *istate, const char *path, int flags)
805 struct stat st;
806 if (lstat(path, &st))
807 die_errno(_("unable to stat '%s'"), path);
808 return add_to_index(istate, path, &st, flags);
811 struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
813 return mem_pool__ce_calloc(find_mem_pool(istate), len);
816 struct cache_entry *make_empty_transient_cache_entry(size_t len)
818 return xcalloc(1, cache_entry_size(len));
821 struct cache_entry *make_cache_entry(struct index_state *istate,
822 unsigned int mode,
823 const struct object_id *oid,
824 const char *path,
825 int stage,
826 unsigned int refresh_options)
828 struct cache_entry *ce, *ret;
829 int len;
831 if (!verify_path(path, mode)) {
832 error(_("invalid path '%s'"), path);
833 return NULL;
836 len = strlen(path);
837 ce = make_empty_cache_entry(istate, len);
839 oidcpy(&ce->oid, oid);
840 memcpy(ce->name, path, len);
841 ce->ce_flags = create_ce_flags(stage);
842 ce->ce_namelen = len;
843 ce->ce_mode = create_ce_mode(mode);
845 ret = refresh_cache_entry(istate, ce, refresh_options);
846 if (ret != ce)
847 discard_cache_entry(ce);
848 return ret;
851 struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
852 const char *path, int stage)
854 struct cache_entry *ce;
855 int len;
857 if (!verify_path(path, mode)) {
858 error(_("invalid path '%s'"), path);
859 return NULL;
862 len = strlen(path);
863 ce = make_empty_transient_cache_entry(len);
865 oidcpy(&ce->oid, oid);
866 memcpy(ce->name, path, len);
867 ce->ce_flags = create_ce_flags(stage);
868 ce->ce_namelen = len;
869 ce->ce_mode = create_ce_mode(mode);
871 return ce;
875 * Chmod an index entry with either +x or -x.
877 * Returns -1 if the chmod for the particular cache entry failed (if it's
878 * not a regular file), -2 if an invalid flip argument is passed in, 0
879 * otherwise.
881 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
882 char flip)
884 if (!S_ISREG(ce->ce_mode))
885 return -1;
886 switch (flip) {
887 case '+':
888 ce->ce_mode |= 0111;
889 break;
890 case '-':
891 ce->ce_mode &= ~0111;
892 break;
893 default:
894 return -2;
896 cache_tree_invalidate_path(istate, ce->name);
897 ce->ce_flags |= CE_UPDATE_IN_BASE;
898 mark_fsmonitor_invalid(istate, ce);
899 istate->cache_changed |= CE_ENTRY_CHANGED;
901 return 0;
904 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
906 int len = ce_namelen(a);
907 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
911 * We fundamentally don't like some paths: we don't want
912 * dot or dot-dot anywhere, and for obvious reasons don't
913 * want to recurse into ".git" either.
915 * Also, we don't want double slashes or slashes at the
916 * end that can make pathnames ambiguous.
918 static int verify_dotfile(const char *rest, unsigned mode)
921 * The first character was '.', but that
922 * has already been discarded, we now test
923 * the rest.
926 /* "." is not allowed */
927 if (*rest == '\0' || is_dir_sep(*rest))
928 return 0;
930 switch (*rest) {
932 * ".git" followed by NUL or slash is bad. Note that we match
933 * case-insensitively here, even if ignore_case is not set.
934 * This outlaws ".GIT" everywhere out of an abundance of caution,
935 * since there's really no good reason to allow it.
937 * Once we've seen ".git", we can also find ".gitmodules", etc (also
938 * case-insensitively).
940 case 'g':
941 case 'G':
942 if (rest[1] != 'i' && rest[1] != 'I')
943 break;
944 if (rest[2] != 't' && rest[2] != 'T')
945 break;
946 if (rest[3] == '\0' || is_dir_sep(rest[3]))
947 return 0;
948 if (S_ISLNK(mode)) {
949 rest += 3;
950 if (skip_iprefix(rest, "modules", &rest) &&
951 (*rest == '\0' || is_dir_sep(*rest)))
952 return 0;
954 break;
955 case '.':
956 if (rest[1] == '\0' || is_dir_sep(rest[1]))
957 return 0;
959 return 1;
962 int verify_path(const char *path, unsigned mode)
964 char c = 0;
966 if (has_dos_drive_prefix(path))
967 return 0;
969 if (!is_valid_path(path))
970 return 0;
972 goto inside;
973 for (;;) {
974 if (!c)
975 return 1;
976 if (is_dir_sep(c)) {
977 inside:
978 if (protect_hfs) {
980 if (is_hfs_dotgit(path))
981 return 0;
982 if (S_ISLNK(mode)) {
983 if (is_hfs_dotgitmodules(path))
984 return 0;
987 if (protect_ntfs) {
988 #ifdef GIT_WINDOWS_NATIVE
989 if (c == '\\')
990 return 0;
991 #endif
992 if (is_ntfs_dotgit(path))
993 return 0;
994 if (S_ISLNK(mode)) {
995 if (is_ntfs_dotgitmodules(path))
996 return 0;
1000 c = *path++;
1001 if ((c == '.' && !verify_dotfile(path, mode)) ||
1002 is_dir_sep(c) || c == '\0')
1003 return 0;
1004 } else if (c == '\\' && protect_ntfs) {
1005 if (is_ntfs_dotgit(path))
1006 return 0;
1007 if (S_ISLNK(mode)) {
1008 if (is_ntfs_dotgitmodules(path))
1009 return 0;
1013 c = *path++;
1018 * Do we have another file that has the beginning components being a
1019 * proper superset of the name we're trying to add?
1021 static int has_file_name(struct index_state *istate,
1022 const struct cache_entry *ce, int pos, int ok_to_replace)
1024 int retval = 0;
1025 int len = ce_namelen(ce);
1026 int stage = ce_stage(ce);
1027 const char *name = ce->name;
1029 while (pos < istate->cache_nr) {
1030 struct cache_entry *p = istate->cache[pos++];
1032 if (len >= ce_namelen(p))
1033 break;
1034 if (memcmp(name, p->name, len))
1035 break;
1036 if (ce_stage(p) != stage)
1037 continue;
1038 if (p->name[len] != '/')
1039 continue;
1040 if (p->ce_flags & CE_REMOVE)
1041 continue;
1042 retval = -1;
1043 if (!ok_to_replace)
1044 break;
1045 remove_index_entry_at(istate, --pos);
1047 return retval;
1052 * Like strcmp(), but also return the offset of the first change.
1053 * If strings are equal, return the length.
1055 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1057 size_t k;
1059 if (!first_change)
1060 return strcmp(s1, s2);
1062 for (k = 0; s1[k] == s2[k]; k++)
1063 if (s1[k] == '\0')
1064 break;
1066 *first_change = k;
1067 return (unsigned char)s1[k] - (unsigned char)s2[k];
1071 * Do we have another file with a pathname that is a proper
1072 * subset of the name we're trying to add?
1074 * That is, is there another file in the index with a path
1075 * that matches a sub-directory in the given entry?
1077 static int has_dir_name(struct index_state *istate,
1078 const struct cache_entry *ce, int pos, int ok_to_replace)
1080 int retval = 0;
1081 int stage = ce_stage(ce);
1082 const char *name = ce->name;
1083 const char *slash = name + ce_namelen(ce);
1084 size_t len_eq_last;
1085 int cmp_last = 0;
1088 * We are frequently called during an iteration on a sorted
1089 * list of pathnames and while building a new index. Therefore,
1090 * there is a high probability that this entry will eventually
1091 * be appended to the index, rather than inserted in the middle.
1092 * If we can confirm that, we can avoid binary searches on the
1093 * components of the pathname.
1095 * Compare the entry's full path with the last path in the index.
1097 if (istate->cache_nr > 0) {
1098 cmp_last = strcmp_offset(name,
1099 istate->cache[istate->cache_nr - 1]->name,
1100 &len_eq_last);
1101 if (cmp_last > 0) {
1102 if (len_eq_last == 0) {
1104 * The entry sorts AFTER the last one in the
1105 * index and their paths have no common prefix,
1106 * so there cannot be a F/D conflict.
1108 return retval;
1109 } else {
1111 * The entry sorts AFTER the last one in the
1112 * index, but has a common prefix. Fall through
1113 * to the loop below to disect the entry's path
1114 * and see where the difference is.
1117 } else if (cmp_last == 0) {
1119 * The entry exactly matches the last one in the
1120 * index, but because of multiple stage and CE_REMOVE
1121 * items, we fall through and let the regular search
1122 * code handle it.
1127 for (;;) {
1128 size_t len;
1130 for (;;) {
1131 if (*--slash == '/')
1132 break;
1133 if (slash <= ce->name)
1134 return retval;
1136 len = slash - name;
1138 if (cmp_last > 0) {
1140 * (len + 1) is a directory boundary (including
1141 * the trailing slash). And since the loop is
1142 * decrementing "slash", the first iteration is
1143 * the longest directory prefix; subsequent
1144 * iterations consider parent directories.
1147 if (len + 1 <= len_eq_last) {
1149 * The directory prefix (including the trailing
1150 * slash) also appears as a prefix in the last
1151 * entry, so the remainder cannot collide (because
1152 * strcmp said the whole path was greater).
1154 * EQ: last: xxx/A
1155 * this: xxx/B
1157 * LT: last: xxx/file_A
1158 * this: xxx/file_B
1160 return retval;
1163 if (len > len_eq_last) {
1165 * This part of the directory prefix (excluding
1166 * the trailing slash) is longer than the known
1167 * equal portions, so this sub-directory cannot
1168 * collide with a file.
1170 * GT: last: xxxA
1171 * this: xxxB/file
1173 return retval;
1177 * This is a possible collision. Fall through and
1178 * let the regular search code handle it.
1180 * last: xxx
1181 * this: xxx/file
1185 pos = index_name_stage_pos(istate, name, len, stage);
1186 if (pos >= 0) {
1188 * Found one, but not so fast. This could
1189 * be a marker that says "I was here, but
1190 * I am being removed". Such an entry is
1191 * not a part of the resulting tree, and
1192 * it is Ok to have a directory at the same
1193 * path.
1195 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1196 retval = -1;
1197 if (!ok_to_replace)
1198 break;
1199 remove_index_entry_at(istate, pos);
1200 continue;
1203 else
1204 pos = -pos-1;
1207 * Trivial optimization: if we find an entry that
1208 * already matches the sub-directory, then we know
1209 * we're ok, and we can exit.
1211 while (pos < istate->cache_nr) {
1212 struct cache_entry *p = istate->cache[pos];
1213 if ((ce_namelen(p) <= len) ||
1214 (p->name[len] != '/') ||
1215 memcmp(p->name, name, len))
1216 break; /* not our subdirectory */
1217 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1219 * p is at the same stage as our entry, and
1220 * is a subdirectory of what we are looking
1221 * at, so we cannot have conflicts at our
1222 * level or anything shorter.
1224 return retval;
1225 pos++;
1228 return retval;
1231 /* We may be in a situation where we already have path/file and path
1232 * is being added, or we already have path and path/file is being
1233 * added. Either one would result in a nonsense tree that has path
1234 * twice when git-write-tree tries to write it out. Prevent it.
1236 * If ok-to-replace is specified, we remove the conflicting entries
1237 * from the cache so the caller should recompute the insert position.
1238 * When this happens, we return non-zero.
1240 static int check_file_directory_conflict(struct index_state *istate,
1241 const struct cache_entry *ce,
1242 int pos, int ok_to_replace)
1244 int retval;
1247 * When ce is an "I am going away" entry, we allow it to be added
1249 if (ce->ce_flags & CE_REMOVE)
1250 return 0;
1253 * We check if the path is a sub-path of a subsequent pathname
1254 * first, since removing those will not change the position
1255 * in the array.
1257 retval = has_file_name(istate, ce, pos, ok_to_replace);
1260 * Then check if the path might have a clashing sub-directory
1261 * before it.
1263 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1266 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1268 int pos;
1269 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1270 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1271 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1272 int new_only = option & ADD_CACHE_NEW_ONLY;
1274 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1275 cache_tree_invalidate_path(istate, ce->name);
1278 * If this entry's path sorts after the last entry in the index,
1279 * we can avoid searching for it.
1281 if (istate->cache_nr > 0 &&
1282 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1283 pos = index_pos_to_insert_pos(istate->cache_nr);
1284 else
1285 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1287 /* existing match? Just replace it. */
1288 if (pos >= 0) {
1289 if (!new_only)
1290 replace_index_entry(istate, pos, ce);
1291 return 0;
1293 pos = -pos-1;
1295 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1296 untracked_cache_add_to_index(istate, ce->name);
1299 * Inserting a merged entry ("stage 0") into the index
1300 * will always replace all non-merged entries..
1302 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1303 while (ce_same_name(istate->cache[pos], ce)) {
1304 ok_to_add = 1;
1305 if (!remove_index_entry_at(istate, pos))
1306 break;
1310 if (!ok_to_add)
1311 return -1;
1312 if (!verify_path(ce->name, ce->ce_mode))
1313 return error(_("invalid path '%s'"), ce->name);
1315 if (!skip_df_check &&
1316 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1317 if (!ok_to_replace)
1318 return error(_("'%s' appears as both a file and as a directory"),
1319 ce->name);
1320 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1321 pos = -pos-1;
1323 return pos + 1;
1326 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1328 int pos;
1330 if (option & ADD_CACHE_JUST_APPEND)
1331 pos = istate->cache_nr;
1332 else {
1333 int ret;
1334 ret = add_index_entry_with_check(istate, ce, option);
1335 if (ret <= 0)
1336 return ret;
1337 pos = ret - 1;
1340 /* Make sure the array is big enough .. */
1341 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1343 /* Add it in.. */
1344 istate->cache_nr++;
1345 if (istate->cache_nr > pos + 1)
1346 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1347 istate->cache_nr - pos - 1);
1348 set_index_entry(istate, pos, ce);
1349 istate->cache_changed |= CE_ENTRY_ADDED;
1350 return 0;
1354 * "refresh" does not calculate a new sha1 file or bring the
1355 * cache up-to-date for mode/content changes. But what it
1356 * _does_ do is to "re-match" the stat information of a file
1357 * with the cache, so that you can refresh the cache for a
1358 * file that hasn't been changed but where the stat entry is
1359 * out of date.
1361 * For example, you'd want to do this after doing a "git-read-tree",
1362 * to link up the stat cache details with the proper files.
1364 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1365 struct cache_entry *ce,
1366 unsigned int options, int *err,
1367 int *changed_ret)
1369 struct stat st;
1370 struct cache_entry *updated;
1371 int changed;
1372 int refresh = options & CE_MATCH_REFRESH;
1373 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1374 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1375 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1376 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1378 if (!refresh || ce_uptodate(ce))
1379 return ce;
1381 if (!ignore_fsmonitor)
1382 refresh_fsmonitor(istate);
1384 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1385 * that the change to the work tree does not matter and told
1386 * us not to worry.
1388 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1389 ce_mark_uptodate(ce);
1390 return ce;
1392 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1393 ce_mark_uptodate(ce);
1394 return ce;
1396 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1397 ce_mark_uptodate(ce);
1398 return ce;
1401 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1402 if (ignore_missing)
1403 return ce;
1404 if (err)
1405 *err = ENOENT;
1406 return NULL;
1409 if (lstat(ce->name, &st) < 0) {
1410 if (ignore_missing && errno == ENOENT)
1411 return ce;
1412 if (err)
1413 *err = errno;
1414 return NULL;
1417 changed = ie_match_stat(istate, ce, &st, options);
1418 if (changed_ret)
1419 *changed_ret = changed;
1420 if (!changed) {
1422 * The path is unchanged. If we were told to ignore
1423 * valid bit, then we did the actual stat check and
1424 * found that the entry is unmodified. If the entry
1425 * is not marked VALID, this is the place to mark it
1426 * valid again, under "assume unchanged" mode.
1428 if (ignore_valid && assume_unchanged &&
1429 !(ce->ce_flags & CE_VALID))
1430 ; /* mark this one VALID again */
1431 else {
1433 * We do not mark the index itself "modified"
1434 * because CE_UPTODATE flag is in-core only;
1435 * we are not going to write this change out.
1437 if (!S_ISGITLINK(ce->ce_mode)) {
1438 ce_mark_uptodate(ce);
1439 mark_fsmonitor_valid(istate, ce);
1441 return ce;
1445 if (ie_modified(istate, ce, &st, options)) {
1446 if (err)
1447 *err = EINVAL;
1448 return NULL;
1451 updated = make_empty_cache_entry(istate, ce_namelen(ce));
1452 copy_cache_entry(updated, ce);
1453 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1454 fill_stat_cache_info(istate, updated, &st);
1456 * If ignore_valid is not set, we should leave CE_VALID bit
1457 * alone. Otherwise, paths marked with --no-assume-unchanged
1458 * (i.e. things to be edited) will reacquire CE_VALID bit
1459 * automatically, which is not really what we want.
1461 if (!ignore_valid && assume_unchanged &&
1462 !(ce->ce_flags & CE_VALID))
1463 updated->ce_flags &= ~CE_VALID;
1465 /* istate->cache_changed is updated in the caller */
1466 return updated;
1469 static void show_file(const char * fmt, const char * name, int in_porcelain,
1470 int * first, const char *header_msg)
1472 if (in_porcelain && *first && header_msg) {
1473 printf("%s\n", header_msg);
1474 *first = 0;
1476 printf(fmt, name);
1479 int repo_refresh_and_write_index(struct repository *repo,
1480 unsigned int refresh_flags,
1481 unsigned int write_flags,
1482 int gentle,
1483 const struct pathspec *pathspec,
1484 char *seen, const char *header_msg)
1486 struct lock_file lock_file = LOCK_INIT;
1487 int fd, ret = 0;
1489 fd = repo_hold_locked_index(repo, &lock_file, 0);
1490 if (!gentle && fd < 0)
1491 return -1;
1492 if (refresh_index(repo->index, refresh_flags, pathspec, seen, header_msg))
1493 ret = 1;
1494 if (0 <= fd && write_locked_index(repo->index, &lock_file, COMMIT_LOCK | write_flags))
1495 ret = -1;
1496 return ret;
1500 int refresh_index(struct index_state *istate, unsigned int flags,
1501 const struct pathspec *pathspec,
1502 char *seen, const char *header_msg)
1504 int i;
1505 int has_errors = 0;
1506 int really = (flags & REFRESH_REALLY) != 0;
1507 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1508 int quiet = (flags & REFRESH_QUIET) != 0;
1509 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1510 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1511 int first = 1;
1512 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1513 unsigned int options = (CE_MATCH_REFRESH |
1514 (really ? CE_MATCH_IGNORE_VALID : 0) |
1515 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1516 const char *modified_fmt;
1517 const char *deleted_fmt;
1518 const char *typechange_fmt;
1519 const char *added_fmt;
1520 const char *unmerged_fmt;
1521 struct progress *progress = NULL;
1523 if (flags & REFRESH_PROGRESS && isatty(2))
1524 progress = start_delayed_progress(_("Refresh index"),
1525 istate->cache_nr);
1527 trace_performance_enter();
1528 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1529 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1530 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1531 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1532 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1534 * Use the multi-threaded preload_index() to refresh most of the
1535 * cache entries quickly then in the single threaded loop below,
1536 * we only have to do the special cases that are left.
1538 preload_index(istate, pathspec, 0);
1539 for (i = 0; i < istate->cache_nr; i++) {
1540 struct cache_entry *ce, *new_entry;
1541 int cache_errno = 0;
1542 int changed = 0;
1543 int filtered = 0;
1545 ce = istate->cache[i];
1546 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1547 continue;
1549 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1550 filtered = 1;
1552 if (ce_stage(ce)) {
1553 while ((i < istate->cache_nr) &&
1554 ! strcmp(istate->cache[i]->name, ce->name))
1555 i++;
1556 i--;
1557 if (allow_unmerged)
1558 continue;
1559 if (!filtered)
1560 show_file(unmerged_fmt, ce->name, in_porcelain,
1561 &first, header_msg);
1562 has_errors = 1;
1563 continue;
1566 if (filtered)
1567 continue;
1569 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1570 if (new_entry == ce)
1571 continue;
1572 if (progress)
1573 display_progress(progress, i);
1574 if (!new_entry) {
1575 const char *fmt;
1577 if (really && cache_errno == EINVAL) {
1578 /* If we are doing --really-refresh that
1579 * means the index is not valid anymore.
1581 ce->ce_flags &= ~CE_VALID;
1582 ce->ce_flags |= CE_UPDATE_IN_BASE;
1583 mark_fsmonitor_invalid(istate, ce);
1584 istate->cache_changed |= CE_ENTRY_CHANGED;
1586 if (quiet)
1587 continue;
1589 if (cache_errno == ENOENT)
1590 fmt = deleted_fmt;
1591 else if (ce_intent_to_add(ce))
1592 fmt = added_fmt; /* must be before other checks */
1593 else if (changed & TYPE_CHANGED)
1594 fmt = typechange_fmt;
1595 else
1596 fmt = modified_fmt;
1597 show_file(fmt,
1598 ce->name, in_porcelain, &first, header_msg);
1599 has_errors = 1;
1600 continue;
1603 replace_index_entry(istate, i, new_entry);
1605 if (progress) {
1606 display_progress(progress, istate->cache_nr);
1607 stop_progress(&progress);
1609 trace_performance_leave("refresh index");
1610 return has_errors;
1613 struct cache_entry *refresh_cache_entry(struct index_state *istate,
1614 struct cache_entry *ce,
1615 unsigned int options)
1617 return refresh_cache_ent(istate, ce, options, NULL, NULL);
1621 /*****************************************************************
1622 * Index File I/O
1623 *****************************************************************/
1625 #define INDEX_FORMAT_DEFAULT 3
1627 static unsigned int get_index_format_default(struct repository *r)
1629 char *envversion = getenv("GIT_INDEX_VERSION");
1630 char *endp;
1631 unsigned int version = INDEX_FORMAT_DEFAULT;
1633 if (!envversion) {
1634 prepare_repo_settings(r);
1636 if (r->settings.index_version >= 0)
1637 version = r->settings.index_version;
1638 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1639 warning(_("index.version set, but the value is invalid.\n"
1640 "Using version %i"), INDEX_FORMAT_DEFAULT);
1641 return INDEX_FORMAT_DEFAULT;
1643 return version;
1646 version = strtoul(envversion, &endp, 10);
1647 if (*endp ||
1648 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1649 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1650 "Using version %i"), INDEX_FORMAT_DEFAULT);
1651 version = INDEX_FORMAT_DEFAULT;
1653 return version;
1657 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1658 * Again - this is just a (very strong in practice) heuristic that
1659 * the inode hasn't changed.
1661 * We save the fields in big-endian order to allow using the
1662 * index file over NFS transparently.
1664 struct ondisk_cache_entry {
1665 struct cache_time ctime;
1666 struct cache_time mtime;
1667 uint32_t dev;
1668 uint32_t ino;
1669 uint32_t mode;
1670 uint32_t uid;
1671 uint32_t gid;
1672 uint32_t size;
1674 * unsigned char hash[hashsz];
1675 * uint16_t flags;
1676 * if (flags & CE_EXTENDED)
1677 * uint16_t flags2;
1679 unsigned char data[GIT_MAX_RAWSZ + 2 * sizeof(uint16_t)];
1680 char name[FLEX_ARRAY];
1683 /* These are only used for v3 or lower */
1684 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1685 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1686 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1687 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1688 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1689 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1690 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1692 /* Allow fsck to force verification of the index checksum. */
1693 int verify_index_checksum;
1695 /* Allow fsck to force verification of the cache entry order. */
1696 int verify_ce_order;
1698 static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1700 git_hash_ctx c;
1701 unsigned char hash[GIT_MAX_RAWSZ];
1702 int hdr_version;
1704 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1705 return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1706 hdr_version = ntohl(hdr->hdr_version);
1707 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1708 return error(_("bad index version %d"), hdr_version);
1710 if (!verify_index_checksum)
1711 return 0;
1713 the_hash_algo->init_fn(&c);
1714 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1715 the_hash_algo->final_fn(hash, &c);
1716 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1717 return error(_("bad index file sha1 signature"));
1718 return 0;
1721 static int read_index_extension(struct index_state *istate,
1722 const char *ext, const char *data, unsigned long sz)
1724 switch (CACHE_EXT(ext)) {
1725 case CACHE_EXT_TREE:
1726 istate->cache_tree = cache_tree_read(data, sz);
1727 break;
1728 case CACHE_EXT_RESOLVE_UNDO:
1729 istate->resolve_undo = resolve_undo_read(data, sz);
1730 break;
1731 case CACHE_EXT_LINK:
1732 if (read_link_extension(istate, data, sz))
1733 return -1;
1734 break;
1735 case CACHE_EXT_UNTRACKED:
1736 istate->untracked = read_untracked_extension(data, sz);
1737 break;
1738 case CACHE_EXT_FSMONITOR:
1739 read_fsmonitor_extension(istate, data, sz);
1740 break;
1741 case CACHE_EXT_ENDOFINDEXENTRIES:
1742 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1743 /* already handled in do_read_index() */
1744 break;
1745 default:
1746 if (*ext < 'A' || 'Z' < *ext)
1747 return error(_("index uses %.4s extension, which we do not understand"),
1748 ext);
1749 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1750 break;
1752 return 0;
1755 static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1756 unsigned int version,
1757 struct ondisk_cache_entry *ondisk,
1758 unsigned long *ent_size,
1759 const struct cache_entry *previous_ce)
1761 struct cache_entry *ce;
1762 size_t len;
1763 const char *name;
1764 const unsigned hashsz = the_hash_algo->rawsz;
1765 const uint16_t *flagsp = (const uint16_t *)(ondisk->data + hashsz);
1766 unsigned int flags;
1767 size_t copy_len = 0;
1769 * Adjacent cache entries tend to share the leading paths, so it makes
1770 * sense to only store the differences in later entries. In the v4
1771 * on-disk format of the index, each on-disk cache entry stores the
1772 * number of bytes to be stripped from the end of the previous name,
1773 * and the bytes to append to the result, to come up with its name.
1775 int expand_name_field = version == 4;
1777 /* On-disk flags are just 16 bits */
1778 flags = get_be16(flagsp);
1779 len = flags & CE_NAMEMASK;
1781 if (flags & CE_EXTENDED) {
1782 int extended_flags;
1783 extended_flags = get_be16(flagsp + 1) << 16;
1784 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1785 if (extended_flags & ~CE_EXTENDED_FLAGS)
1786 die(_("unknown index entry format 0x%08x"), extended_flags);
1787 flags |= extended_flags;
1788 name = (const char *)(flagsp + 2);
1790 else
1791 name = (const char *)(flagsp + 1);
1793 if (expand_name_field) {
1794 const unsigned char *cp = (const unsigned char *)name;
1795 size_t strip_len, previous_len;
1797 /* If we're at the beginning of a block, ignore the previous name */
1798 strip_len = decode_varint(&cp);
1799 if (previous_ce) {
1800 previous_len = previous_ce->ce_namelen;
1801 if (previous_len < strip_len)
1802 die(_("malformed name field in the index, near path '%s'"),
1803 previous_ce->name);
1804 copy_len = previous_len - strip_len;
1806 name = (const char *)cp;
1809 if (len == CE_NAMEMASK) {
1810 len = strlen(name);
1811 if (expand_name_field)
1812 len += copy_len;
1815 ce = mem_pool__ce_alloc(ce_mem_pool, len);
1817 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1818 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1819 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1820 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1821 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1822 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1823 ce->ce_mode = get_be32(&ondisk->mode);
1824 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1825 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1826 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1827 ce->ce_flags = flags & ~CE_NAMEMASK;
1828 ce->ce_namelen = len;
1829 ce->index = 0;
1830 hashcpy(ce->oid.hash, ondisk->data);
1831 memcpy(ce->name, name, len);
1832 ce->name[len] = '\0';
1834 if (expand_name_field) {
1835 if (copy_len)
1836 memcpy(ce->name, previous_ce->name, copy_len);
1837 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1838 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1839 } else {
1840 memcpy(ce->name, name, len + 1);
1841 *ent_size = ondisk_ce_size(ce);
1843 return ce;
1846 static void check_ce_order(struct index_state *istate)
1848 unsigned int i;
1850 if (!verify_ce_order)
1851 return;
1853 for (i = 1; i < istate->cache_nr; i++) {
1854 struct cache_entry *ce = istate->cache[i - 1];
1855 struct cache_entry *next_ce = istate->cache[i];
1856 int name_compare = strcmp(ce->name, next_ce->name);
1858 if (0 < name_compare)
1859 die(_("unordered stage entries in index"));
1860 if (!name_compare) {
1861 if (!ce_stage(ce))
1862 die(_("multiple stage entries for merged file '%s'"),
1863 ce->name);
1864 if (ce_stage(ce) > ce_stage(next_ce))
1865 die(_("unordered stage entries for '%s'"),
1866 ce->name);
1871 static void tweak_untracked_cache(struct index_state *istate)
1873 struct repository *r = the_repository;
1875 prepare_repo_settings(r);
1877 if (r->settings.core_untracked_cache == UNTRACKED_CACHE_REMOVE) {
1878 remove_untracked_cache(istate);
1879 return;
1882 if (r->settings.core_untracked_cache == UNTRACKED_CACHE_WRITE)
1883 add_untracked_cache(istate);
1886 static void tweak_split_index(struct index_state *istate)
1888 switch (git_config_get_split_index()) {
1889 case -1: /* unset: do nothing */
1890 break;
1891 case 0: /* false */
1892 remove_split_index(istate);
1893 break;
1894 case 1: /* true */
1895 add_split_index(istate);
1896 break;
1897 default: /* unknown value: do nothing */
1898 break;
1902 static void post_read_index_from(struct index_state *istate)
1904 check_ce_order(istate);
1905 tweak_untracked_cache(istate);
1906 tweak_split_index(istate);
1907 tweak_fsmonitor(istate);
1910 static size_t estimate_cache_size_from_compressed(unsigned int entries)
1912 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1915 static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1917 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1920 * Account for potential alignment differences.
1922 per_entry += align_padding_size(per_entry, 0);
1923 return ondisk_size + entries * per_entry;
1926 struct index_entry_offset
1928 /* starting byte offset into index file, count of index entries in this block */
1929 int offset, nr;
1932 struct index_entry_offset_table
1934 int nr;
1935 struct index_entry_offset entries[FLEX_ARRAY];
1938 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1939 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1941 static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1942 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1944 struct load_index_extensions
1946 pthread_t pthread;
1947 struct index_state *istate;
1948 const char *mmap;
1949 size_t mmap_size;
1950 unsigned long src_offset;
1953 static void *load_index_extensions(void *_data)
1955 struct load_index_extensions *p = _data;
1956 unsigned long src_offset = p->src_offset;
1958 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1959 /* After an array of active_nr index entries,
1960 * there can be arbitrary number of extended
1961 * sections, each of which is prefixed with
1962 * extension name (4-byte) and section length
1963 * in 4-byte network byte order.
1965 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1966 if (read_index_extension(p->istate,
1967 p->mmap + src_offset,
1968 p->mmap + src_offset + 8,
1969 extsize) < 0) {
1970 munmap((void *)p->mmap, p->mmap_size);
1971 die(_("index file corrupt"));
1973 src_offset += 8;
1974 src_offset += extsize;
1977 return NULL;
1981 * A helper function that will load the specified range of cache entries
1982 * from the memory mapped file and add them to the given index.
1984 static unsigned long load_cache_entry_block(struct index_state *istate,
1985 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1986 unsigned long start_offset, const struct cache_entry *previous_ce)
1988 int i;
1989 unsigned long src_offset = start_offset;
1991 for (i = offset; i < offset + nr; i++) {
1992 struct ondisk_cache_entry *disk_ce;
1993 struct cache_entry *ce;
1994 unsigned long consumed;
1996 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1997 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1998 set_index_entry(istate, i, ce);
2000 src_offset += consumed;
2001 previous_ce = ce;
2003 return src_offset - start_offset;
2006 static unsigned long load_all_cache_entries(struct index_state *istate,
2007 const char *mmap, size_t mmap_size, unsigned long src_offset)
2009 unsigned long consumed;
2011 istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2012 if (istate->version == 4) {
2013 mem_pool_init(istate->ce_mem_pool,
2014 estimate_cache_size_from_compressed(istate->cache_nr));
2015 } else {
2016 mem_pool_init(istate->ce_mem_pool,
2017 estimate_cache_size(mmap_size, istate->cache_nr));
2020 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2021 0, istate->cache_nr, mmap, src_offset, NULL);
2022 return consumed;
2026 * Mostly randomly chosen maximum thread counts: we
2027 * cap the parallelism to online_cpus() threads, and we want
2028 * to have at least 10000 cache entries per thread for it to
2029 * be worth starting a thread.
2032 #define THREAD_COST (10000)
2034 struct load_cache_entries_thread_data
2036 pthread_t pthread;
2037 struct index_state *istate;
2038 struct mem_pool *ce_mem_pool;
2039 int offset;
2040 const char *mmap;
2041 struct index_entry_offset_table *ieot;
2042 int ieot_start; /* starting index into the ieot array */
2043 int ieot_blocks; /* count of ieot entries to process */
2044 unsigned long consumed; /* return # of bytes in index file processed */
2048 * A thread proc to run the load_cache_entries() computation
2049 * across multiple background threads.
2051 static void *load_cache_entries_thread(void *_data)
2053 struct load_cache_entries_thread_data *p = _data;
2054 int i;
2056 /* iterate across all ieot blocks assigned to this thread */
2057 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2058 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2059 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2060 p->offset += p->ieot->entries[i].nr;
2062 return NULL;
2065 static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2066 int nr_threads, struct index_entry_offset_table *ieot)
2068 int i, offset, ieot_blocks, ieot_start, err;
2069 struct load_cache_entries_thread_data *data;
2070 unsigned long consumed = 0;
2072 /* a little sanity checking */
2073 if (istate->name_hash_initialized)
2074 BUG("the name hash isn't thread safe");
2076 istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2077 mem_pool_init(istate->ce_mem_pool, 0);
2079 /* ensure we have no more threads than we have blocks to process */
2080 if (nr_threads > ieot->nr)
2081 nr_threads = ieot->nr;
2082 data = xcalloc(nr_threads, sizeof(*data));
2084 offset = ieot_start = 0;
2085 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2086 for (i = 0; i < nr_threads; i++) {
2087 struct load_cache_entries_thread_data *p = &data[i];
2088 int nr, j;
2090 if (ieot_start + ieot_blocks > ieot->nr)
2091 ieot_blocks = ieot->nr - ieot_start;
2093 p->istate = istate;
2094 p->offset = offset;
2095 p->mmap = mmap;
2096 p->ieot = ieot;
2097 p->ieot_start = ieot_start;
2098 p->ieot_blocks = ieot_blocks;
2100 /* create a mem_pool for each thread */
2101 nr = 0;
2102 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2103 nr += p->ieot->entries[j].nr;
2104 p->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2105 if (istate->version == 4) {
2106 mem_pool_init(p->ce_mem_pool,
2107 estimate_cache_size_from_compressed(nr));
2108 } else {
2109 mem_pool_init(p->ce_mem_pool,
2110 estimate_cache_size(mmap_size, nr));
2113 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2114 if (err)
2115 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2117 /* increment by the number of cache entries in the ieot block being processed */
2118 for (j = 0; j < ieot_blocks; j++)
2119 offset += ieot->entries[ieot_start + j].nr;
2120 ieot_start += ieot_blocks;
2123 for (i = 0; i < nr_threads; i++) {
2124 struct load_cache_entries_thread_data *p = &data[i];
2126 err = pthread_join(p->pthread, NULL);
2127 if (err)
2128 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2129 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2130 consumed += p->consumed;
2133 free(data);
2135 return consumed;
2138 /* remember to discard_cache() before reading a different cache! */
2139 int do_read_index(struct index_state *istate, const char *path, int must_exist)
2141 int fd;
2142 struct stat st;
2143 unsigned long src_offset;
2144 const struct cache_header *hdr;
2145 const char *mmap;
2146 size_t mmap_size;
2147 struct load_index_extensions p;
2148 size_t extension_offset = 0;
2149 int nr_threads, cpus;
2150 struct index_entry_offset_table *ieot = NULL;
2152 if (istate->initialized)
2153 return istate->cache_nr;
2155 istate->timestamp.sec = 0;
2156 istate->timestamp.nsec = 0;
2157 fd = open(path, O_RDONLY);
2158 if (fd < 0) {
2159 if (!must_exist && errno == ENOENT)
2160 return 0;
2161 die_errno(_("%s: index file open failed"), path);
2164 if (fstat(fd, &st))
2165 die_errno(_("%s: cannot stat the open index"), path);
2167 mmap_size = xsize_t(st.st_size);
2168 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2169 die(_("%s: index file smaller than expected"), path);
2171 mmap = xmmap_gently(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2172 if (mmap == MAP_FAILED)
2173 die_errno(_("%s: unable to map index file"), path);
2174 close(fd);
2176 hdr = (const struct cache_header *)mmap;
2177 if (verify_hdr(hdr, mmap_size) < 0)
2178 goto unmap;
2180 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2181 istate->version = ntohl(hdr->hdr_version);
2182 istate->cache_nr = ntohl(hdr->hdr_entries);
2183 istate->cache_alloc = alloc_nr(istate->cache_nr);
2184 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2185 istate->initialized = 1;
2187 p.istate = istate;
2188 p.mmap = mmap;
2189 p.mmap_size = mmap_size;
2191 src_offset = sizeof(*hdr);
2193 if (git_config_get_index_threads(&nr_threads))
2194 nr_threads = 1;
2196 /* TODO: does creating more threads than cores help? */
2197 if (!nr_threads) {
2198 nr_threads = istate->cache_nr / THREAD_COST;
2199 cpus = online_cpus();
2200 if (nr_threads > cpus)
2201 nr_threads = cpus;
2204 if (!HAVE_THREADS)
2205 nr_threads = 1;
2207 if (nr_threads > 1) {
2208 extension_offset = read_eoie_extension(mmap, mmap_size);
2209 if (extension_offset) {
2210 int err;
2212 p.src_offset = extension_offset;
2213 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2214 if (err)
2215 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2217 nr_threads--;
2222 * Locate and read the index entry offset table so that we can use it
2223 * to multi-thread the reading of the cache entries.
2225 if (extension_offset && nr_threads > 1)
2226 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2228 if (ieot) {
2229 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, nr_threads, ieot);
2230 free(ieot);
2231 } else {
2232 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2235 istate->timestamp.sec = st.st_mtime;
2236 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2238 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2239 if (extension_offset) {
2240 int ret = pthread_join(p.pthread, NULL);
2241 if (ret)
2242 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2243 } else {
2244 p.src_offset = src_offset;
2245 load_index_extensions(&p);
2247 munmap((void *)mmap, mmap_size);
2250 * TODO trace2: replace "the_repository" with the actual repo instance
2251 * that is associated with the given "istate".
2253 trace2_data_intmax("index", the_repository, "read/version",
2254 istate->version);
2255 trace2_data_intmax("index", the_repository, "read/cache_nr",
2256 istate->cache_nr);
2258 return istate->cache_nr;
2260 unmap:
2261 munmap((void *)mmap, mmap_size);
2262 die(_("index file corrupt"));
2266 * Signal that the shared index is used by updating its mtime.
2268 * This way, shared index can be removed if they have not been used
2269 * for some time.
2271 static void freshen_shared_index(const char *shared_index, int warn)
2273 if (!check_and_freshen_file(shared_index, 1) && warn)
2274 warning(_("could not freshen shared index '%s'"), shared_index);
2277 int read_index_from(struct index_state *istate, const char *path,
2278 const char *gitdir)
2280 struct split_index *split_index;
2281 int ret;
2282 char *base_oid_hex;
2283 char *base_path;
2285 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2286 if (istate->initialized)
2287 return istate->cache_nr;
2290 * TODO trace2: replace "the_repository" with the actual repo instance
2291 * that is associated with the given "istate".
2293 trace2_region_enter_printf("index", "do_read_index", the_repository,
2294 "%s", path);
2295 trace_performance_enter();
2296 ret = do_read_index(istate, path, 0);
2297 trace_performance_leave("read cache %s", path);
2298 trace2_region_leave_printf("index", "do_read_index", the_repository,
2299 "%s", path);
2301 split_index = istate->split_index;
2302 if (!split_index || is_null_oid(&split_index->base_oid)) {
2303 post_read_index_from(istate);
2304 return ret;
2307 trace_performance_enter();
2308 if (split_index->base)
2309 discard_index(split_index->base);
2310 else
2311 split_index->base = xcalloc(1, sizeof(*split_index->base));
2313 base_oid_hex = oid_to_hex(&split_index->base_oid);
2314 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2315 trace2_region_enter_printf("index", "shared/do_read_index",
2316 the_repository, "%s", base_path);
2317 ret = do_read_index(split_index->base, base_path, 1);
2318 trace2_region_leave_printf("index", "shared/do_read_index",
2319 the_repository, "%s", base_path);
2320 if (!oideq(&split_index->base_oid, &split_index->base->oid))
2321 die(_("broken index, expect %s in %s, got %s"),
2322 base_oid_hex, base_path,
2323 oid_to_hex(&split_index->base->oid));
2325 freshen_shared_index(base_path, 0);
2326 merge_base_index(istate);
2327 post_read_index_from(istate);
2328 trace_performance_leave("read cache %s", base_path);
2329 free(base_path);
2330 return ret;
2333 int is_index_unborn(struct index_state *istate)
2335 return (!istate->cache_nr && !istate->timestamp.sec);
2338 int discard_index(struct index_state *istate)
2341 * Cache entries in istate->cache[] should have been allocated
2342 * from the memory pool associated with this index, or from an
2343 * associated split_index. There is no need to free individual
2344 * cache entries. validate_cache_entries can detect when this
2345 * assertion does not hold.
2347 validate_cache_entries(istate);
2349 resolve_undo_clear_index(istate);
2350 istate->cache_nr = 0;
2351 istate->cache_changed = 0;
2352 istate->timestamp.sec = 0;
2353 istate->timestamp.nsec = 0;
2354 free_name_hash(istate);
2355 cache_tree_free(&(istate->cache_tree));
2356 istate->initialized = 0;
2357 istate->fsmonitor_has_run_once = 0;
2358 FREE_AND_NULL(istate->cache);
2359 istate->cache_alloc = 0;
2360 discard_split_index(istate);
2361 free_untracked_cache(istate->untracked);
2362 istate->untracked = NULL;
2364 if (istate->ce_mem_pool) {
2365 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2366 FREE_AND_NULL(istate->ce_mem_pool);
2369 return 0;
2373 * Validate the cache entries of this index.
2374 * All cache entries associated with this index
2375 * should have been allocated by the memory pool
2376 * associated with this index, or by a referenced
2377 * split index.
2379 void validate_cache_entries(const struct index_state *istate)
2381 int i;
2383 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2384 return;
2386 for (i = 0; i < istate->cache_nr; i++) {
2387 if (!istate) {
2388 BUG("cache entry is not allocated from expected memory pool");
2389 } else if (!istate->ce_mem_pool ||
2390 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2391 if (!istate->split_index ||
2392 !istate->split_index->base ||
2393 !istate->split_index->base->ce_mem_pool ||
2394 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2395 BUG("cache entry is not allocated from expected memory pool");
2400 if (istate->split_index)
2401 validate_cache_entries(istate->split_index->base);
2404 int unmerged_index(const struct index_state *istate)
2406 int i;
2407 for (i = 0; i < istate->cache_nr; i++) {
2408 if (ce_stage(istate->cache[i]))
2409 return 1;
2411 return 0;
2414 int repo_index_has_changes(struct repository *repo,
2415 struct tree *tree,
2416 struct strbuf *sb)
2418 struct index_state *istate = repo->index;
2419 struct object_id cmp;
2420 int i;
2422 if (tree)
2423 cmp = tree->object.oid;
2424 if (tree || !get_oid_tree("HEAD", &cmp)) {
2425 struct diff_options opt;
2427 repo_diff_setup(repo, &opt);
2428 opt.flags.exit_with_status = 1;
2429 if (!sb)
2430 opt.flags.quick = 1;
2431 do_diff_cache(&cmp, &opt);
2432 diffcore_std(&opt);
2433 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2434 if (i)
2435 strbuf_addch(sb, ' ');
2436 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2438 diff_flush(&opt);
2439 return opt.flags.has_changes != 0;
2440 } else {
2441 for (i = 0; sb && i < istate->cache_nr; i++) {
2442 if (i)
2443 strbuf_addch(sb, ' ');
2444 strbuf_addstr(sb, istate->cache[i]->name);
2446 return !!istate->cache_nr;
2450 #define WRITE_BUFFER_SIZE 8192
2451 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2452 static unsigned long write_buffer_len;
2454 static int ce_write_flush(git_hash_ctx *context, int fd)
2456 unsigned int buffered = write_buffer_len;
2457 if (buffered) {
2458 the_hash_algo->update_fn(context, write_buffer, buffered);
2459 if (write_in_full(fd, write_buffer, buffered) < 0)
2460 return -1;
2461 write_buffer_len = 0;
2463 return 0;
2466 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2468 while (len) {
2469 unsigned int buffered = write_buffer_len;
2470 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2471 if (partial > len)
2472 partial = len;
2473 memcpy(write_buffer + buffered, data, partial);
2474 buffered += partial;
2475 if (buffered == WRITE_BUFFER_SIZE) {
2476 write_buffer_len = buffered;
2477 if (ce_write_flush(context, fd))
2478 return -1;
2479 buffered = 0;
2481 write_buffer_len = buffered;
2482 len -= partial;
2483 data = (char *) data + partial;
2485 return 0;
2488 static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2489 int fd, unsigned int ext, unsigned int sz)
2491 ext = htonl(ext);
2492 sz = htonl(sz);
2493 if (eoie_context) {
2494 the_hash_algo->update_fn(eoie_context, &ext, 4);
2495 the_hash_algo->update_fn(eoie_context, &sz, 4);
2497 return ((ce_write(context, fd, &ext, 4) < 0) ||
2498 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2501 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2503 unsigned int left = write_buffer_len;
2505 if (left) {
2506 write_buffer_len = 0;
2507 the_hash_algo->update_fn(context, write_buffer, left);
2510 /* Flush first if not enough space for hash signature */
2511 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2512 if (write_in_full(fd, write_buffer, left) < 0)
2513 return -1;
2514 left = 0;
2517 /* Append the hash signature at the end */
2518 the_hash_algo->final_fn(write_buffer + left, context);
2519 hashcpy(hash, write_buffer + left);
2520 left += the_hash_algo->rawsz;
2521 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2524 static void ce_smudge_racily_clean_entry(struct index_state *istate,
2525 struct cache_entry *ce)
2528 * The only thing we care about in this function is to smudge the
2529 * falsely clean entry due to touch-update-touch race, so we leave
2530 * everything else as they are. We are called for entries whose
2531 * ce_stat_data.sd_mtime match the index file mtime.
2533 * Note that this actually does not do much for gitlinks, for
2534 * which ce_match_stat_basic() always goes to the actual
2535 * contents. The caller checks with is_racy_timestamp() which
2536 * always says "no" for gitlinks, so we are not called for them ;-)
2538 struct stat st;
2540 if (lstat(ce->name, &st) < 0)
2541 return;
2542 if (ce_match_stat_basic(ce, &st))
2543 return;
2544 if (ce_modified_check_fs(istate, ce, &st)) {
2545 /* This is "racily clean"; smudge it. Note that this
2546 * is a tricky code. At first glance, it may appear
2547 * that it can break with this sequence:
2549 * $ echo xyzzy >frotz
2550 * $ git-update-index --add frotz
2551 * $ : >frotz
2552 * $ sleep 3
2553 * $ echo filfre >nitfol
2554 * $ git-update-index --add nitfol
2556 * but it does not. When the second update-index runs,
2557 * it notices that the entry "frotz" has the same timestamp
2558 * as index, and if we were to smudge it by resetting its
2559 * size to zero here, then the object name recorded
2560 * in index is the 6-byte file but the cached stat information
2561 * becomes zero --- which would then match what we would
2562 * obtain from the filesystem next time we stat("frotz").
2564 * However, the second update-index, before calling
2565 * this function, notices that the cached size is 6
2566 * bytes and what is on the filesystem is an empty
2567 * file, and never calls us, so the cached size information
2568 * for "frotz" stays 6 which does not match the filesystem.
2570 ce->ce_stat_data.sd_size = 0;
2574 /* Copy miscellaneous fields but not the name */
2575 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2576 struct cache_entry *ce)
2578 short flags;
2579 const unsigned hashsz = the_hash_algo->rawsz;
2580 uint16_t *flagsp = (uint16_t *)(ondisk->data + hashsz);
2582 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2583 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2584 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2585 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2586 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2587 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2588 ondisk->mode = htonl(ce->ce_mode);
2589 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2590 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2591 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2592 hashcpy(ondisk->data, ce->oid.hash);
2594 flags = ce->ce_flags & ~CE_NAMEMASK;
2595 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2596 flagsp[0] = htons(flags);
2597 if (ce->ce_flags & CE_EXTENDED) {
2598 flagsp[1] = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2602 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2603 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2605 int size;
2606 int result;
2607 unsigned int saved_namelen;
2608 int stripped_name = 0;
2609 static unsigned char padding[8] = { 0x00 };
2611 if (ce->ce_flags & CE_STRIP_NAME) {
2612 saved_namelen = ce_namelen(ce);
2613 ce->ce_namelen = 0;
2614 stripped_name = 1;
2617 size = offsetof(struct ondisk_cache_entry,data) + ondisk_data_size(ce->ce_flags, 0);
2619 if (!previous_name) {
2620 int len = ce_namelen(ce);
2621 copy_cache_entry_to_ondisk(ondisk, ce);
2622 result = ce_write(c, fd, ondisk, size);
2623 if (!result)
2624 result = ce_write(c, fd, ce->name, len);
2625 if (!result)
2626 result = ce_write(c, fd, padding, align_padding_size(size, len));
2627 } else {
2628 int common, to_remove, prefix_size;
2629 unsigned char to_remove_vi[16];
2630 for (common = 0;
2631 (ce->name[common] &&
2632 common < previous_name->len &&
2633 ce->name[common] == previous_name->buf[common]);
2634 common++)
2635 ; /* still matching */
2636 to_remove = previous_name->len - common;
2637 prefix_size = encode_varint(to_remove, to_remove_vi);
2639 copy_cache_entry_to_ondisk(ondisk, ce);
2640 result = ce_write(c, fd, ondisk, size);
2641 if (!result)
2642 result = ce_write(c, fd, to_remove_vi, prefix_size);
2643 if (!result)
2644 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2645 if (!result)
2646 result = ce_write(c, fd, padding, 1);
2648 strbuf_splice(previous_name, common, to_remove,
2649 ce->name + common, ce_namelen(ce) - common);
2651 if (stripped_name) {
2652 ce->ce_namelen = saved_namelen;
2653 ce->ce_flags &= ~CE_STRIP_NAME;
2656 return result;
2660 * This function verifies if index_state has the correct sha1 of the
2661 * index file. Don't die if we have any other failure, just return 0.
2663 static int verify_index_from(const struct index_state *istate, const char *path)
2665 int fd;
2666 ssize_t n;
2667 struct stat st;
2668 unsigned char hash[GIT_MAX_RAWSZ];
2670 if (!istate->initialized)
2671 return 0;
2673 fd = open(path, O_RDONLY);
2674 if (fd < 0)
2675 return 0;
2677 if (fstat(fd, &st))
2678 goto out;
2680 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2681 goto out;
2683 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2684 if (n != the_hash_algo->rawsz)
2685 goto out;
2687 if (!hasheq(istate->oid.hash, hash))
2688 goto out;
2690 close(fd);
2691 return 1;
2693 out:
2694 close(fd);
2695 return 0;
2698 static int repo_verify_index(struct repository *repo)
2700 return verify_index_from(repo->index, repo->index_file);
2703 static int has_racy_timestamp(struct index_state *istate)
2705 int entries = istate->cache_nr;
2706 int i;
2708 for (i = 0; i < entries; i++) {
2709 struct cache_entry *ce = istate->cache[i];
2710 if (is_racy_timestamp(istate, ce))
2711 return 1;
2713 return 0;
2716 void repo_update_index_if_able(struct repository *repo,
2717 struct lock_file *lockfile)
2719 if ((repo->index->cache_changed ||
2720 has_racy_timestamp(repo->index)) &&
2721 repo_verify_index(repo))
2722 write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2723 else
2724 rollback_lock_file(lockfile);
2727 static int record_eoie(void)
2729 int val;
2731 if (!git_config_get_bool("index.recordendofindexentries", &val))
2732 return val;
2735 * As a convenience, the end of index entries extension
2736 * used for threading is written by default if the user
2737 * explicitly requested threaded index reads.
2739 return !git_config_get_index_threads(&val) && val != 1;
2742 static int record_ieot(void)
2744 int val;
2746 if (!git_config_get_bool("index.recordoffsettable", &val))
2747 return val;
2750 * As a convenience, the offset table used for threading is
2751 * written by default if the user explicitly requested
2752 * threaded index reads.
2754 return !git_config_get_index_threads(&val) && val != 1;
2758 * On success, `tempfile` is closed. If it is the temporary file
2759 * of a `struct lock_file`, we will therefore effectively perform
2760 * a 'close_lock_file_gently()`. Since that is an implementation
2761 * detail of lockfiles, callers of `do_write_index()` should not
2762 * rely on it.
2764 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2765 int strip_extensions)
2767 uint64_t start = getnanotime();
2768 int newfd = tempfile->fd;
2769 git_hash_ctx c, eoie_c;
2770 struct cache_header hdr;
2771 int i, err = 0, removed, extended, hdr_version;
2772 struct cache_entry **cache = istate->cache;
2773 int entries = istate->cache_nr;
2774 struct stat st;
2775 struct ondisk_cache_entry ondisk;
2776 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2777 int drop_cache_tree = istate->drop_cache_tree;
2778 off_t offset;
2779 int ieot_entries = 1;
2780 struct index_entry_offset_table *ieot = NULL;
2781 int nr, nr_threads;
2783 for (i = removed = extended = 0; i < entries; i++) {
2784 if (cache[i]->ce_flags & CE_REMOVE)
2785 removed++;
2787 /* reduce extended entries if possible */
2788 cache[i]->ce_flags &= ~CE_EXTENDED;
2789 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2790 extended++;
2791 cache[i]->ce_flags |= CE_EXTENDED;
2795 if (!istate->version) {
2796 istate->version = get_index_format_default(the_repository);
2797 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2798 init_split_index(istate);
2801 /* demote version 3 to version 2 when the latter suffices */
2802 if (istate->version == 3 || istate->version == 2)
2803 istate->version = extended ? 3 : 2;
2805 hdr_version = istate->version;
2807 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2808 hdr.hdr_version = htonl(hdr_version);
2809 hdr.hdr_entries = htonl(entries - removed);
2811 the_hash_algo->init_fn(&c);
2812 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2813 return -1;
2815 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))
2816 nr_threads = 1;
2818 if (nr_threads != 1 && record_ieot()) {
2819 int ieot_blocks, cpus;
2822 * ensure default number of ieot blocks maps evenly to the
2823 * default number of threads that will process them leaving
2824 * room for the thread to load the index extensions.
2826 if (!nr_threads) {
2827 ieot_blocks = istate->cache_nr / THREAD_COST;
2828 cpus = online_cpus();
2829 if (ieot_blocks > cpus - 1)
2830 ieot_blocks = cpus - 1;
2831 } else {
2832 ieot_blocks = nr_threads;
2833 if (ieot_blocks > istate->cache_nr)
2834 ieot_blocks = istate->cache_nr;
2838 * no reason to write out the IEOT extension if we don't
2839 * have enough blocks to utilize multi-threading
2841 if (ieot_blocks > 1) {
2842 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2843 + (ieot_blocks * sizeof(struct index_entry_offset)));
2844 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2848 offset = lseek(newfd, 0, SEEK_CUR);
2849 if (offset < 0) {
2850 free(ieot);
2851 return -1;
2853 offset += write_buffer_len;
2854 nr = 0;
2855 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2857 for (i = 0; i < entries; i++) {
2858 struct cache_entry *ce = cache[i];
2859 if (ce->ce_flags & CE_REMOVE)
2860 continue;
2861 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2862 ce_smudge_racily_clean_entry(istate, ce);
2863 if (is_null_oid(&ce->oid)) {
2864 static const char msg[] = "cache entry has null sha1: %s";
2865 static int allow = -1;
2867 if (allow < 0)
2868 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2869 if (allow)
2870 warning(msg, ce->name);
2871 else
2872 err = error(msg, ce->name);
2874 drop_cache_tree = 1;
2876 if (ieot && i && (i % ieot_entries == 0)) {
2877 ieot->entries[ieot->nr].nr = nr;
2878 ieot->entries[ieot->nr].offset = offset;
2879 ieot->nr++;
2881 * If we have a V4 index, set the first byte to an invalid
2882 * character to ensure there is nothing common with the previous
2883 * entry
2885 if (previous_name)
2886 previous_name->buf[0] = 0;
2887 nr = 0;
2888 offset = lseek(newfd, 0, SEEK_CUR);
2889 if (offset < 0) {
2890 free(ieot);
2891 return -1;
2893 offset += write_buffer_len;
2895 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2896 err = -1;
2898 if (err)
2899 break;
2900 nr++;
2902 if (ieot && nr) {
2903 ieot->entries[ieot->nr].nr = nr;
2904 ieot->entries[ieot->nr].offset = offset;
2905 ieot->nr++;
2907 strbuf_release(&previous_name_buf);
2909 if (err) {
2910 free(ieot);
2911 return err;
2914 /* Write extension data here */
2915 offset = lseek(newfd, 0, SEEK_CUR);
2916 if (offset < 0) {
2917 free(ieot);
2918 return -1;
2920 offset += write_buffer_len;
2921 the_hash_algo->init_fn(&eoie_c);
2924 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2925 * can minimize the number of extensions we have to scan through to
2926 * find it during load. Write it out regardless of the
2927 * strip_extensions parameter as we need it when loading the shared
2928 * index.
2930 if (ieot) {
2931 struct strbuf sb = STRBUF_INIT;
2933 write_ieot_extension(&sb, ieot);
2934 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2935 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2936 strbuf_release(&sb);
2937 free(ieot);
2938 if (err)
2939 return -1;
2942 if (!strip_extensions && istate->split_index &&
2943 !is_null_oid(&istate->split_index->base_oid)) {
2944 struct strbuf sb = STRBUF_INIT;
2946 err = write_link_extension(&sb, istate) < 0 ||
2947 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2948 sb.len) < 0 ||
2949 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2950 strbuf_release(&sb);
2951 if (err)
2952 return -1;
2954 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2955 struct strbuf sb = STRBUF_INIT;
2957 cache_tree_write(&sb, istate->cache_tree);
2958 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2959 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2960 strbuf_release(&sb);
2961 if (err)
2962 return -1;
2964 if (!strip_extensions && istate->resolve_undo) {
2965 struct strbuf sb = STRBUF_INIT;
2967 resolve_undo_write(&sb, istate->resolve_undo);
2968 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2969 sb.len) < 0
2970 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2971 strbuf_release(&sb);
2972 if (err)
2973 return -1;
2975 if (!strip_extensions && istate->untracked) {
2976 struct strbuf sb = STRBUF_INIT;
2978 write_untracked_extension(&sb, istate->untracked);
2979 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2980 sb.len) < 0 ||
2981 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2982 strbuf_release(&sb);
2983 if (err)
2984 return -1;
2986 if (!strip_extensions && istate->fsmonitor_last_update) {
2987 struct strbuf sb = STRBUF_INIT;
2989 write_fsmonitor_extension(&sb, istate);
2990 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2991 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2992 strbuf_release(&sb);
2993 if (err)
2994 return -1;
2998 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2999 * so that it can be found and processed before all the index entries are
3000 * read. Write it out regardless of the strip_extensions parameter as we need it
3001 * when loading the shared index.
3003 if (offset && record_eoie()) {
3004 struct strbuf sb = STRBUF_INIT;
3006 write_eoie_extension(&sb, &eoie_c, offset);
3007 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
3008 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
3009 strbuf_release(&sb);
3010 if (err)
3011 return -1;
3014 if (ce_flush(&c, newfd, istate->oid.hash))
3015 return -1;
3016 if (close_tempfile_gently(tempfile)) {
3017 error(_("could not close '%s'"), get_tempfile_path(tempfile));
3018 return -1;
3020 if (stat(get_tempfile_path(tempfile), &st))
3021 return -1;
3022 istate->timestamp.sec = (unsigned int)st.st_mtime;
3023 istate->timestamp.nsec = ST_MTIME_NSEC(st);
3024 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
3027 * TODO trace2: replace "the_repository" with the actual repo instance
3028 * that is associated with the given "istate".
3030 trace2_data_intmax("index", the_repository, "write/version",
3031 istate->version);
3032 trace2_data_intmax("index", the_repository, "write/cache_nr",
3033 istate->cache_nr);
3035 return 0;
3038 void set_alternate_index_output(const char *name)
3040 alternate_index_output = name;
3043 static int commit_locked_index(struct lock_file *lk)
3045 if (alternate_index_output)
3046 return commit_lock_file_to(lk, alternate_index_output);
3047 else
3048 return commit_lock_file(lk);
3051 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
3052 unsigned flags)
3054 int ret;
3057 * TODO trace2: replace "the_repository" with the actual repo instance
3058 * that is associated with the given "istate".
3060 trace2_region_enter_printf("index", "do_write_index", the_repository,
3061 "%s", get_lock_file_path(lock));
3062 ret = do_write_index(istate, lock->tempfile, 0);
3063 trace2_region_leave_printf("index", "do_write_index", the_repository,
3064 "%s", get_lock_file_path(lock));
3066 if (ret)
3067 return ret;
3068 if (flags & COMMIT_LOCK)
3069 ret = commit_locked_index(lock);
3070 else
3071 ret = close_lock_file_gently(lock);
3073 run_hook_le(NULL, "post-index-change",
3074 istate->updated_workdir ? "1" : "0",
3075 istate->updated_skipworktree ? "1" : "0", NULL);
3076 istate->updated_workdir = 0;
3077 istate->updated_skipworktree = 0;
3079 return ret;
3082 static int write_split_index(struct index_state *istate,
3083 struct lock_file *lock,
3084 unsigned flags)
3086 int ret;
3087 prepare_to_write_split_index(istate);
3088 ret = do_write_locked_index(istate, lock, flags);
3089 finish_writing_split_index(istate);
3090 return ret;
3093 static const char *shared_index_expire = "2.weeks.ago";
3095 static unsigned long get_shared_index_expire_date(void)
3097 static unsigned long shared_index_expire_date;
3098 static int shared_index_expire_date_prepared;
3100 if (!shared_index_expire_date_prepared) {
3101 git_config_get_expiry("splitindex.sharedindexexpire",
3102 &shared_index_expire);
3103 shared_index_expire_date = approxidate(shared_index_expire);
3104 shared_index_expire_date_prepared = 1;
3107 return shared_index_expire_date;
3110 static int should_delete_shared_index(const char *shared_index_path)
3112 struct stat st;
3113 unsigned long expiration;
3115 /* Check timestamp */
3116 expiration = get_shared_index_expire_date();
3117 if (!expiration)
3118 return 0;
3119 if (stat(shared_index_path, &st))
3120 return error_errno(_("could not stat '%s'"), shared_index_path);
3121 if (st.st_mtime > expiration)
3122 return 0;
3124 return 1;
3127 static int clean_shared_index_files(const char *current_hex)
3129 struct dirent *de;
3130 DIR *dir = opendir(get_git_dir());
3132 if (!dir)
3133 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3135 while ((de = readdir(dir)) != NULL) {
3136 const char *sha1_hex;
3137 const char *shared_index_path;
3138 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3139 continue;
3140 if (!strcmp(sha1_hex, current_hex))
3141 continue;
3142 shared_index_path = git_path("%s", de->d_name);
3143 if (should_delete_shared_index(shared_index_path) > 0 &&
3144 unlink(shared_index_path))
3145 warning_errno(_("unable to unlink: %s"), shared_index_path);
3147 closedir(dir);
3149 return 0;
3152 static int write_shared_index(struct index_state *istate,
3153 struct tempfile **temp)
3155 struct split_index *si = istate->split_index;
3156 int ret;
3158 move_cache_to_base_index(istate);
3160 trace2_region_enter_printf("index", "shared/do_write_index",
3161 the_repository, "%s", get_tempfile_path(*temp));
3162 ret = do_write_index(si->base, *temp, 1);
3163 trace2_region_leave_printf("index", "shared/do_write_index",
3164 the_repository, "%s", get_tempfile_path(*temp));
3166 if (ret)
3167 return ret;
3168 ret = adjust_shared_perm(get_tempfile_path(*temp));
3169 if (ret) {
3170 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3171 return ret;
3173 ret = rename_tempfile(temp,
3174 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3175 if (!ret) {
3176 oidcpy(&si->base_oid, &si->base->oid);
3177 clean_shared_index_files(oid_to_hex(&si->base->oid));
3180 return ret;
3183 static const int default_max_percent_split_change = 20;
3185 static int too_many_not_shared_entries(struct index_state *istate)
3187 int i, not_shared = 0;
3188 int max_split = git_config_get_max_percent_split_change();
3190 switch (max_split) {
3191 case -1:
3192 /* not or badly configured: use the default value */
3193 max_split = default_max_percent_split_change;
3194 break;
3195 case 0:
3196 return 1; /* 0% means always write a new shared index */
3197 case 100:
3198 return 0; /* 100% means never write a new shared index */
3199 default:
3200 break; /* just use the configured value */
3203 /* Count not shared entries */
3204 for (i = 0; i < istate->cache_nr; i++) {
3205 struct cache_entry *ce = istate->cache[i];
3206 if (!ce->index)
3207 not_shared++;
3210 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3213 int write_locked_index(struct index_state *istate, struct lock_file *lock,
3214 unsigned flags)
3216 int new_shared_index, ret;
3217 struct split_index *si = istate->split_index;
3219 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3220 cache_tree_verify(the_repository, istate);
3222 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3223 if (flags & COMMIT_LOCK)
3224 rollback_lock_file(lock);
3225 return 0;
3228 if (istate->fsmonitor_last_update)
3229 fill_fsmonitor_bitmap(istate);
3231 if (!si || alternate_index_output ||
3232 (istate->cache_changed & ~EXTMASK)) {
3233 if (si)
3234 oidclr(&si->base_oid);
3235 ret = do_write_locked_index(istate, lock, flags);
3236 goto out;
3239 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3240 int v = si->base_oid.hash[0];
3241 if ((v & 15) < 6)
3242 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3244 if (too_many_not_shared_entries(istate))
3245 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3247 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3249 if (new_shared_index) {
3250 struct tempfile *temp;
3251 int saved_errno;
3253 /* Same initial permissions as the main .git/index file */
3254 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3255 if (!temp) {
3256 oidclr(&si->base_oid);
3257 ret = do_write_locked_index(istate, lock, flags);
3258 goto out;
3260 ret = write_shared_index(istate, &temp);
3262 saved_errno = errno;
3263 if (is_tempfile_active(temp))
3264 delete_tempfile(&temp);
3265 errno = saved_errno;
3267 if (ret)
3268 goto out;
3271 ret = write_split_index(istate, lock, flags);
3273 /* Freshen the shared index only if the split-index was written */
3274 if (!ret && !new_shared_index && !is_null_oid(&si->base_oid)) {
3275 const char *shared_index = git_path("sharedindex.%s",
3276 oid_to_hex(&si->base_oid));
3277 freshen_shared_index(shared_index, 1);
3280 out:
3281 if (flags & COMMIT_LOCK)
3282 rollback_lock_file(lock);
3283 return ret;
3287 * Read the index file that is potentially unmerged into given
3288 * index_state, dropping any unmerged entries to stage #0 (potentially
3289 * resulting in a path appearing as both a file and a directory in the
3290 * index; the caller is responsible to clear out the extra entries
3291 * before writing the index to a tree). Returns true if the index is
3292 * unmerged. Callers who want to refuse to work from an unmerged
3293 * state can call this and check its return value, instead of calling
3294 * read_cache().
3296 int repo_read_index_unmerged(struct repository *repo)
3298 struct index_state *istate;
3299 int i;
3300 int unmerged = 0;
3302 repo_read_index(repo);
3303 istate = repo->index;
3304 for (i = 0; i < istate->cache_nr; i++) {
3305 struct cache_entry *ce = istate->cache[i];
3306 struct cache_entry *new_ce;
3307 int len;
3309 if (!ce_stage(ce))
3310 continue;
3311 unmerged = 1;
3312 len = ce_namelen(ce);
3313 new_ce = make_empty_cache_entry(istate, len);
3314 memcpy(new_ce->name, ce->name, len);
3315 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3316 new_ce->ce_namelen = len;
3317 new_ce->ce_mode = ce->ce_mode;
3318 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3319 return error(_("%s: cannot drop to stage #0"),
3320 new_ce->name);
3322 return unmerged;
3326 * Returns 1 if the path is an "other" path with respect to
3327 * the index; that is, the path is not mentioned in the index at all,
3328 * either as a file, a directory with some files in the index,
3329 * or as an unmerged entry.
3331 * We helpfully remove a trailing "/" from directories so that
3332 * the output of read_directory can be used as-is.
3334 int index_name_is_other(const struct index_state *istate, const char *name,
3335 int namelen)
3337 int pos;
3338 if (namelen && name[namelen - 1] == '/')
3339 namelen--;
3340 pos = index_name_pos(istate, name, namelen);
3341 if (0 <= pos)
3342 return 0; /* exact match */
3343 pos = -pos - 1;
3344 if (pos < istate->cache_nr) {
3345 struct cache_entry *ce = istate->cache[pos];
3346 if (ce_namelen(ce) == namelen &&
3347 !memcmp(ce->name, name, namelen))
3348 return 0; /* Yup, this one exists unmerged */
3350 return 1;
3353 void *read_blob_data_from_index(const struct index_state *istate,
3354 const char *path, unsigned long *size)
3356 int pos, len;
3357 unsigned long sz;
3358 enum object_type type;
3359 void *data;
3361 len = strlen(path);
3362 pos = index_name_pos(istate, path, len);
3363 if (pos < 0) {
3365 * We might be in the middle of a merge, in which
3366 * case we would read stage #2 (ours).
3368 int i;
3369 for (i = -pos - 1;
3370 (pos < 0 && i < istate->cache_nr &&
3371 !strcmp(istate->cache[i]->name, path));
3372 i++)
3373 if (ce_stage(istate->cache[i]) == 2)
3374 pos = i;
3376 if (pos < 0)
3377 return NULL;
3378 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3379 if (!data || type != OBJ_BLOB) {
3380 free(data);
3381 return NULL;
3383 if (size)
3384 *size = sz;
3385 return data;
3388 void stat_validity_clear(struct stat_validity *sv)
3390 FREE_AND_NULL(sv->sd);
3393 int stat_validity_check(struct stat_validity *sv, const char *path)
3395 struct stat st;
3397 if (stat(path, &st) < 0)
3398 return sv->sd == NULL;
3399 if (!sv->sd)
3400 return 0;
3401 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3404 void stat_validity_update(struct stat_validity *sv, int fd)
3406 struct stat st;
3408 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3409 stat_validity_clear(sv);
3410 else {
3411 if (!sv->sd)
3412 sv->sd = xcalloc(1, sizeof(struct stat_data));
3413 fill_stat_data(sv->sd, &st);
3417 void move_index_extensions(struct index_state *dst, struct index_state *src)
3419 dst->untracked = src->untracked;
3420 src->untracked = NULL;
3421 dst->cache_tree = src->cache_tree;
3422 src->cache_tree = NULL;
3425 struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3426 struct index_state *istate)
3428 unsigned int size = ce_size(ce);
3429 int mem_pool_allocated;
3430 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3431 mem_pool_allocated = new_entry->mem_pool_allocated;
3433 memcpy(new_entry, ce, size);
3434 new_entry->mem_pool_allocated = mem_pool_allocated;
3435 return new_entry;
3438 void discard_cache_entry(struct cache_entry *ce)
3440 if (ce && should_validate_cache_entries())
3441 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3443 if (ce && ce->mem_pool_allocated)
3444 return;
3446 free(ce);
3449 int should_validate_cache_entries(void)
3451 static int validate_index_cache_entries = -1;
3453 if (validate_index_cache_entries < 0) {
3454 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3455 validate_index_cache_entries = 1;
3456 else
3457 validate_index_cache_entries = 0;
3460 return validate_index_cache_entries;
3463 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3464 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3466 static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3469 * The end of index entries (EOIE) extension is guaranteed to be last
3470 * so that it can be found by scanning backwards from the EOF.
3472 * "EOIE"
3473 * <4-byte length>
3474 * <4-byte offset>
3475 * <20-byte hash>
3477 const char *index, *eoie;
3478 uint32_t extsize;
3479 size_t offset, src_offset;
3480 unsigned char hash[GIT_MAX_RAWSZ];
3481 git_hash_ctx c;
3483 /* ensure we have an index big enough to contain an EOIE extension */
3484 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3485 return 0;
3487 /* validate the extension signature */
3488 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3489 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3490 return 0;
3491 index += sizeof(uint32_t);
3493 /* validate the extension size */
3494 extsize = get_be32(index);
3495 if (extsize != EOIE_SIZE)
3496 return 0;
3497 index += sizeof(uint32_t);
3500 * Validate the offset we're going to look for the first extension
3501 * signature is after the index header and before the eoie extension.
3503 offset = get_be32(index);
3504 if (mmap + offset < mmap + sizeof(struct cache_header))
3505 return 0;
3506 if (mmap + offset >= eoie)
3507 return 0;
3508 index += sizeof(uint32_t);
3511 * The hash is computed over extension types and their sizes (but not
3512 * their contents). E.g. if we have "TREE" extension that is N-bytes
3513 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3514 * then the hash would be:
3516 * SHA-1("TREE" + <binary representation of N> +
3517 * "REUC" + <binary representation of M>)
3519 src_offset = offset;
3520 the_hash_algo->init_fn(&c);
3521 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3522 /* After an array of active_nr index entries,
3523 * there can be arbitrary number of extended
3524 * sections, each of which is prefixed with
3525 * extension name (4-byte) and section length
3526 * in 4-byte network byte order.
3528 uint32_t extsize;
3529 memcpy(&extsize, mmap + src_offset + 4, 4);
3530 extsize = ntohl(extsize);
3532 /* verify the extension size isn't so large it will wrap around */
3533 if (src_offset + 8 + extsize < src_offset)
3534 return 0;
3536 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3538 src_offset += 8;
3539 src_offset += extsize;
3541 the_hash_algo->final_fn(hash, &c);
3542 if (!hasheq(hash, (const unsigned char *)index))
3543 return 0;
3545 /* Validate that the extension offsets returned us back to the eoie extension. */
3546 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3547 return 0;
3549 return offset;
3552 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3554 uint32_t buffer;
3555 unsigned char hash[GIT_MAX_RAWSZ];
3557 /* offset */
3558 put_be32(&buffer, offset);
3559 strbuf_add(sb, &buffer, sizeof(uint32_t));
3561 /* hash */
3562 the_hash_algo->final_fn(hash, eoie_context);
3563 strbuf_add(sb, hash, the_hash_algo->rawsz);
3566 #define IEOT_VERSION (1)
3568 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3570 const char *index = NULL;
3571 uint32_t extsize, ext_version;
3572 struct index_entry_offset_table *ieot;
3573 int i, nr;
3575 /* find the IEOT extension */
3576 if (!offset)
3577 return NULL;
3578 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3579 extsize = get_be32(mmap + offset + 4);
3580 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3581 index = mmap + offset + 4 + 4;
3582 break;
3584 offset += 8;
3585 offset += extsize;
3587 if (!index)
3588 return NULL;
3590 /* validate the version is IEOT_VERSION */
3591 ext_version = get_be32(index);
3592 if (ext_version != IEOT_VERSION) {
3593 error("invalid IEOT version %d", ext_version);
3594 return NULL;
3596 index += sizeof(uint32_t);
3598 /* extension size - version bytes / bytes per entry */
3599 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3600 if (!nr) {
3601 error("invalid number of IEOT entries %d", nr);
3602 return NULL;
3604 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3605 + (nr * sizeof(struct index_entry_offset)));
3606 ieot->nr = nr;
3607 for (i = 0; i < nr; i++) {
3608 ieot->entries[i].offset = get_be32(index);
3609 index += sizeof(uint32_t);
3610 ieot->entries[i].nr = get_be32(index);
3611 index += sizeof(uint32_t);
3614 return ieot;
3617 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3619 uint32_t buffer;
3620 int i;
3622 /* version */
3623 put_be32(&buffer, IEOT_VERSION);
3624 strbuf_add(sb, &buffer, sizeof(uint32_t));
3626 /* ieot */
3627 for (i = 0; i < ieot->nr; i++) {
3629 /* offset */
3630 put_be32(&buffer, ieot->entries[i].offset);
3631 strbuf_add(sb, &buffer, sizeof(uint32_t));
3633 /* count */
3634 put_be32(&buffer, ieot->entries[i].nr);
3635 strbuf_add(sb, &buffer, sizeof(uint32_t));