completion: support restore
[git.git] / read-cache.c
blob4dc6de1b55b0a6047e49f188b28c431f0167d7ef
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 "strbuf.h"
21 #include "varint.h"
22 #include "split-index.h"
23 #include "utf8.h"
24 #include "fsmonitor.h"
25 #include "thread-utils.h"
26 #include "progress.h"
28 /* Mask for the name length in ce_flags in the on-disk index */
30 #define CE_NAMEMASK (0x0fff)
32 /* Index extensions.
34 * The first letter should be 'A'..'Z' for extensions that are not
35 * necessary for a correct operation (i.e. optimization data).
36 * When new extensions are added that _needs_ to be understood in
37 * order to correctly interpret the index file, pick character that
38 * is outside the range, to cause the reader to abort.
41 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
42 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
43 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
44 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
45 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
46 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
47 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
48 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
50 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
51 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
52 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
53 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
57 * This is an estimate of the pathname length in the index. We use
58 * this for V4 index files to guess the un-deltafied size of the index
59 * in memory because of pathname deltafication. This is not required
60 * for V2/V3 index formats because their pathnames are not compressed.
61 * If the initial amount of memory set aside is not sufficient, the
62 * mem pool will allocate extra memory.
64 #define CACHE_ENTRY_PATH_LENGTH 80
66 static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
68 struct cache_entry *ce;
69 ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
70 ce->mem_pool_allocated = 1;
71 return ce;
74 static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
76 struct cache_entry * ce;
77 ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
78 ce->mem_pool_allocated = 1;
79 return ce;
82 static struct mem_pool *find_mem_pool(struct index_state *istate)
84 struct mem_pool **pool_ptr;
86 if (istate->split_index && istate->split_index->base)
87 pool_ptr = &istate->split_index->base->ce_mem_pool;
88 else
89 pool_ptr = &istate->ce_mem_pool;
91 if (!*pool_ptr)
92 mem_pool_init(pool_ptr, 0);
94 return *pool_ptr;
97 static const char *alternate_index_output;
99 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
101 istate->cache[nr] = ce;
102 add_name_hash(istate, ce);
105 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
107 struct cache_entry *old = istate->cache[nr];
109 replace_index_entry_in_base(istate, old, ce);
110 remove_name_hash(istate, old);
111 discard_cache_entry(old);
112 ce->ce_flags &= ~CE_HASHED;
113 set_index_entry(istate, nr, ce);
114 ce->ce_flags |= CE_UPDATE_IN_BASE;
115 mark_fsmonitor_invalid(istate, ce);
116 istate->cache_changed |= CE_ENTRY_CHANGED;
119 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
121 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
122 int namelen = strlen(new_name);
124 new_entry = make_empty_cache_entry(istate, namelen);
125 copy_cache_entry(new_entry, old_entry);
126 new_entry->ce_flags &= ~CE_HASHED;
127 new_entry->ce_namelen = namelen;
128 new_entry->index = 0;
129 memcpy(new_entry->name, new_name, namelen + 1);
131 cache_tree_invalidate_path(istate, old_entry->name);
132 untracked_cache_remove_from_index(istate, old_entry->name);
133 remove_index_entry_at(istate, nr);
134 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
137 void fill_stat_data(struct stat_data *sd, struct stat *st)
139 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
140 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
141 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
142 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
143 sd->sd_dev = st->st_dev;
144 sd->sd_ino = st->st_ino;
145 sd->sd_uid = st->st_uid;
146 sd->sd_gid = st->st_gid;
147 sd->sd_size = st->st_size;
150 int match_stat_data(const struct stat_data *sd, struct stat *st)
152 int changed = 0;
154 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
155 changed |= MTIME_CHANGED;
156 if (trust_ctime && check_stat &&
157 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
158 changed |= CTIME_CHANGED;
160 #ifdef USE_NSEC
161 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
162 changed |= MTIME_CHANGED;
163 if (trust_ctime && check_stat &&
164 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
165 changed |= CTIME_CHANGED;
166 #endif
168 if (check_stat) {
169 if (sd->sd_uid != (unsigned int) st->st_uid ||
170 sd->sd_gid != (unsigned int) st->st_gid)
171 changed |= OWNER_CHANGED;
172 if (sd->sd_ino != (unsigned int) st->st_ino)
173 changed |= INODE_CHANGED;
176 #ifdef USE_STDEV
178 * st_dev breaks on network filesystems where different
179 * clients will have different views of what "device"
180 * the filesystem is on
182 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
183 changed |= INODE_CHANGED;
184 #endif
186 if (sd->sd_size != (unsigned int) st->st_size)
187 changed |= DATA_CHANGED;
189 return changed;
193 * This only updates the "non-critical" parts of the directory
194 * cache, ie the parts that aren't tracked by GIT, and only used
195 * to validate the cache.
197 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
199 fill_stat_data(&ce->ce_stat_data, st);
201 if (assume_unchanged)
202 ce->ce_flags |= CE_VALID;
204 if (S_ISREG(st->st_mode)) {
205 ce_mark_uptodate(ce);
206 mark_fsmonitor_valid(ce);
210 static int ce_compare_data(struct index_state *istate,
211 const struct cache_entry *ce,
212 struct stat *st)
214 int match = -1;
215 int fd = git_open_cloexec(ce->name, O_RDONLY);
217 if (fd >= 0) {
218 struct object_id oid;
219 if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
220 match = !oideq(&oid, &ce->oid);
221 /* index_fd() closed the file descriptor already */
223 return match;
226 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
228 int match = -1;
229 void *buffer;
230 unsigned long size;
231 enum object_type type;
232 struct strbuf sb = STRBUF_INIT;
234 if (strbuf_readlink(&sb, ce->name, expected_size))
235 return -1;
237 buffer = read_object_file(&ce->oid, &type, &size);
238 if (buffer) {
239 if (size == sb.len)
240 match = memcmp(buffer, sb.buf, size);
241 free(buffer);
243 strbuf_release(&sb);
244 return match;
247 static int ce_compare_gitlink(const struct cache_entry *ce)
249 struct object_id oid;
252 * We don't actually require that the .git directory
253 * under GITLINK directory be a valid git directory. It
254 * might even be missing (in case nobody populated that
255 * sub-project).
257 * If so, we consider it always to match.
259 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
260 return 0;
261 return !oideq(&oid, &ce->oid);
264 static int ce_modified_check_fs(struct index_state *istate,
265 const struct cache_entry *ce,
266 struct stat *st)
268 switch (st->st_mode & S_IFMT) {
269 case S_IFREG:
270 if (ce_compare_data(istate, ce, st))
271 return DATA_CHANGED;
272 break;
273 case S_IFLNK:
274 if (ce_compare_link(ce, xsize_t(st->st_size)))
275 return DATA_CHANGED;
276 break;
277 case S_IFDIR:
278 if (S_ISGITLINK(ce->ce_mode))
279 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
280 /* else fallthrough */
281 default:
282 return TYPE_CHANGED;
284 return 0;
287 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
289 unsigned int changed = 0;
291 if (ce->ce_flags & CE_REMOVE)
292 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
294 switch (ce->ce_mode & S_IFMT) {
295 case S_IFREG:
296 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
297 /* We consider only the owner x bit to be relevant for
298 * "mode changes"
300 if (trust_executable_bit &&
301 (0100 & (ce->ce_mode ^ st->st_mode)))
302 changed |= MODE_CHANGED;
303 break;
304 case S_IFLNK:
305 if (!S_ISLNK(st->st_mode) &&
306 (has_symlinks || !S_ISREG(st->st_mode)))
307 changed |= TYPE_CHANGED;
308 break;
309 case S_IFGITLINK:
310 /* We ignore most of the st_xxx fields for gitlinks */
311 if (!S_ISDIR(st->st_mode))
312 changed |= TYPE_CHANGED;
313 else if (ce_compare_gitlink(ce))
314 changed |= DATA_CHANGED;
315 return changed;
316 default:
317 BUG("unsupported ce_mode: %o", ce->ce_mode);
320 changed |= match_stat_data(&ce->ce_stat_data, st);
322 /* Racily smudged entry? */
323 if (!ce->ce_stat_data.sd_size) {
324 if (!is_empty_blob_sha1(ce->oid.hash))
325 changed |= DATA_CHANGED;
328 return changed;
331 static int is_racy_stat(const struct index_state *istate,
332 const struct stat_data *sd)
334 return (istate->timestamp.sec &&
335 #ifdef USE_NSEC
336 /* nanosecond timestamped files can also be racy! */
337 (istate->timestamp.sec < sd->sd_mtime.sec ||
338 (istate->timestamp.sec == sd->sd_mtime.sec &&
339 istate->timestamp.nsec <= sd->sd_mtime.nsec))
340 #else
341 istate->timestamp.sec <= sd->sd_mtime.sec
342 #endif
346 int is_racy_timestamp(const struct index_state *istate,
347 const struct cache_entry *ce)
349 return (!S_ISGITLINK(ce->ce_mode) &&
350 is_racy_stat(istate, &ce->ce_stat_data));
353 int match_stat_data_racy(const struct index_state *istate,
354 const struct stat_data *sd, struct stat *st)
356 if (is_racy_stat(istate, sd))
357 return MTIME_CHANGED;
358 return match_stat_data(sd, st);
361 int ie_match_stat(struct index_state *istate,
362 const struct cache_entry *ce, struct stat *st,
363 unsigned int options)
365 unsigned int changed;
366 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
367 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
368 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
369 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
371 if (!ignore_fsmonitor)
372 refresh_fsmonitor(istate);
374 * If it's marked as always valid in the index, it's
375 * valid whatever the checked-out copy says.
377 * skip-worktree has the same effect with higher precedence
379 if (!ignore_skip_worktree && ce_skip_worktree(ce))
380 return 0;
381 if (!ignore_valid && (ce->ce_flags & CE_VALID))
382 return 0;
383 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
384 return 0;
387 * Intent-to-add entries have not been added, so the index entry
388 * by definition never matches what is in the work tree until it
389 * actually gets added.
391 if (ce_intent_to_add(ce))
392 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
394 changed = ce_match_stat_basic(ce, st);
397 * Within 1 second of this sequence:
398 * echo xyzzy >file && git-update-index --add file
399 * running this command:
400 * echo frotz >file
401 * would give a falsely clean cache entry. The mtime and
402 * length match the cache, and other stat fields do not change.
404 * We could detect this at update-index time (the cache entry
405 * being registered/updated records the same time as "now")
406 * and delay the return from git-update-index, but that would
407 * effectively mean we can make at most one commit per second,
408 * which is not acceptable. Instead, we check cache entries
409 * whose mtime are the same as the index file timestamp more
410 * carefully than others.
412 if (!changed && is_racy_timestamp(istate, ce)) {
413 if (assume_racy_is_modified)
414 changed |= DATA_CHANGED;
415 else
416 changed |= ce_modified_check_fs(istate, ce, st);
419 return changed;
422 int ie_modified(struct index_state *istate,
423 const struct cache_entry *ce,
424 struct stat *st, unsigned int options)
426 int changed, changed_fs;
428 changed = ie_match_stat(istate, ce, st, options);
429 if (!changed)
430 return 0;
432 * If the mode or type has changed, there's no point in trying
433 * to refresh the entry - it's not going to match
435 if (changed & (MODE_CHANGED | TYPE_CHANGED))
436 return changed;
439 * Immediately after read-tree or update-index --cacheinfo,
440 * the length field is zero, as we have never even read the
441 * lstat(2) information once, and we cannot trust DATA_CHANGED
442 * returned by ie_match_stat() which in turn was returned by
443 * ce_match_stat_basic() to signal that the filesize of the
444 * blob changed. We have to actually go to the filesystem to
445 * see if the contents match, and if so, should answer "unchanged".
447 * The logic does not apply to gitlinks, as ce_match_stat_basic()
448 * already has checked the actual HEAD from the filesystem in the
449 * subproject. If ie_match_stat() already said it is different,
450 * then we know it is.
452 if ((changed & DATA_CHANGED) &&
453 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
454 return changed;
456 changed_fs = ce_modified_check_fs(istate, ce, st);
457 if (changed_fs)
458 return changed | changed_fs;
459 return 0;
462 int base_name_compare(const char *name1, int len1, int mode1,
463 const char *name2, int len2, int mode2)
465 unsigned char c1, c2;
466 int len = len1 < len2 ? len1 : len2;
467 int cmp;
469 cmp = memcmp(name1, name2, len);
470 if (cmp)
471 return cmp;
472 c1 = name1[len];
473 c2 = name2[len];
474 if (!c1 && S_ISDIR(mode1))
475 c1 = '/';
476 if (!c2 && S_ISDIR(mode2))
477 c2 = '/';
478 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
482 * df_name_compare() is identical to base_name_compare(), except it
483 * compares conflicting directory/file entries as equal. Note that
484 * while a directory name compares as equal to a regular file, they
485 * then individually compare _differently_ to a filename that has
486 * a dot after the basename (because '\0' < '.' < '/').
488 * This is used by routines that want to traverse the git namespace
489 * but then handle conflicting entries together when possible.
491 int df_name_compare(const char *name1, int len1, int mode1,
492 const char *name2, int len2, int mode2)
494 int len = len1 < len2 ? len1 : len2, cmp;
495 unsigned char c1, c2;
497 cmp = memcmp(name1, name2, len);
498 if (cmp)
499 return cmp;
500 /* Directories and files compare equal (same length, same name) */
501 if (len1 == len2)
502 return 0;
503 c1 = name1[len];
504 if (!c1 && S_ISDIR(mode1))
505 c1 = '/';
506 c2 = name2[len];
507 if (!c2 && S_ISDIR(mode2))
508 c2 = '/';
509 if (c1 == '/' && !c2)
510 return 0;
511 if (c2 == '/' && !c1)
512 return 0;
513 return c1 - c2;
516 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
518 size_t min_len = (len1 < len2) ? len1 : len2;
519 int cmp = memcmp(name1, name2, min_len);
520 if (cmp)
521 return cmp;
522 if (len1 < len2)
523 return -1;
524 if (len1 > len2)
525 return 1;
526 return 0;
529 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
531 int cmp;
533 cmp = name_compare(name1, len1, name2, len2);
534 if (cmp)
535 return cmp;
537 if (stage1 < stage2)
538 return -1;
539 if (stage1 > stage2)
540 return 1;
541 return 0;
544 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
546 int first, last;
548 first = 0;
549 last = istate->cache_nr;
550 while (last > first) {
551 int next = (last + first) >> 1;
552 struct cache_entry *ce = istate->cache[next];
553 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
554 if (!cmp)
555 return next;
556 if (cmp < 0) {
557 last = next;
558 continue;
560 first = next+1;
562 return -first-1;
565 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
567 return index_name_stage_pos(istate, name, namelen, 0);
570 int remove_index_entry_at(struct index_state *istate, int pos)
572 struct cache_entry *ce = istate->cache[pos];
574 record_resolve_undo(istate, ce);
575 remove_name_hash(istate, ce);
576 save_or_free_index_entry(istate, ce);
577 istate->cache_changed |= CE_ENTRY_REMOVED;
578 istate->cache_nr--;
579 if (pos >= istate->cache_nr)
580 return 0;
581 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
582 istate->cache_nr - pos);
583 return 1;
587 * Remove all cache entries marked for removal, that is where
588 * CE_REMOVE is set in ce_flags. This is much more effective than
589 * calling remove_index_entry_at() for each entry to be removed.
591 void remove_marked_cache_entries(struct index_state *istate, int invalidate)
593 struct cache_entry **ce_array = istate->cache;
594 unsigned int i, j;
596 for (i = j = 0; i < istate->cache_nr; i++) {
597 if (ce_array[i]->ce_flags & CE_REMOVE) {
598 if (invalidate) {
599 cache_tree_invalidate_path(istate,
600 ce_array[i]->name);
601 untracked_cache_remove_from_index(istate,
602 ce_array[i]->name);
604 remove_name_hash(istate, ce_array[i]);
605 save_or_free_index_entry(istate, ce_array[i]);
607 else
608 ce_array[j++] = ce_array[i];
610 if (j == istate->cache_nr)
611 return;
612 istate->cache_changed |= CE_ENTRY_REMOVED;
613 istate->cache_nr = j;
616 int remove_file_from_index(struct index_state *istate, const char *path)
618 int pos = index_name_pos(istate, path, strlen(path));
619 if (pos < 0)
620 pos = -pos-1;
621 cache_tree_invalidate_path(istate, path);
622 untracked_cache_remove_from_index(istate, path);
623 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
624 remove_index_entry_at(istate, pos);
625 return 0;
628 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
630 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
633 static int index_name_pos_also_unmerged(struct index_state *istate,
634 const char *path, int namelen)
636 int pos = index_name_pos(istate, path, namelen);
637 struct cache_entry *ce;
639 if (pos >= 0)
640 return pos;
642 /* maybe unmerged? */
643 pos = -1 - pos;
644 if (pos >= istate->cache_nr ||
645 compare_name((ce = istate->cache[pos]), path, namelen))
646 return -1;
648 /* order of preference: stage 2, 1, 3 */
649 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
650 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
651 !compare_name(ce, path, namelen))
652 pos++;
653 return pos;
656 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
658 int len = ce_namelen(ce);
659 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
663 * If we add a filename that aliases in the cache, we will use the
664 * name that we already have - but we don't want to update the same
665 * alias twice, because that implies that there were actually two
666 * different files with aliasing names!
668 * So we use the CE_ADDED flag to verify that the alias was an old
669 * one before we accept it as
671 static struct cache_entry *create_alias_ce(struct index_state *istate,
672 struct cache_entry *ce,
673 struct cache_entry *alias)
675 int len;
676 struct cache_entry *new_entry;
678 if (alias->ce_flags & CE_ADDED)
679 die(_("will not add file alias '%s' ('%s' already exists in index)"),
680 ce->name, alias->name);
682 /* Ok, create the new entry using the name of the existing alias */
683 len = ce_namelen(alias);
684 new_entry = make_empty_cache_entry(istate, len);
685 memcpy(new_entry->name, alias->name, len);
686 copy_cache_entry(new_entry, ce);
687 save_or_free_index_entry(istate, ce);
688 return new_entry;
691 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
693 struct object_id oid;
694 if (write_object_file("", 0, blob_type, &oid))
695 die(_("cannot create an empty blob in the object database"));
696 oidcpy(&ce->oid, &oid);
699 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
701 int namelen, was_same;
702 mode_t st_mode = st->st_mode;
703 struct cache_entry *ce, *alias = NULL;
704 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
705 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
706 int pretend = flags & ADD_CACHE_PRETEND;
707 int intent_only = flags & ADD_CACHE_INTENT;
708 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
709 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
710 int hash_flags = HASH_WRITE_OBJECT;
712 if (flags & ADD_CACHE_RENORMALIZE)
713 hash_flags |= HASH_RENORMALIZE;
715 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
716 return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
718 namelen = strlen(path);
719 if (S_ISDIR(st_mode)) {
720 while (namelen && path[namelen-1] == '/')
721 namelen--;
723 ce = make_empty_cache_entry(istate, namelen);
724 memcpy(ce->name, path, namelen);
725 ce->ce_namelen = namelen;
726 if (!intent_only)
727 fill_stat_cache_info(ce, st);
728 else
729 ce->ce_flags |= CE_INTENT_TO_ADD;
732 if (trust_executable_bit && has_symlinks) {
733 ce->ce_mode = create_ce_mode(st_mode);
734 } else {
735 /* If there is an existing entry, pick the mode bits and type
736 * from it, otherwise assume unexecutable regular file.
738 struct cache_entry *ent;
739 int pos = index_name_pos_also_unmerged(istate, path, namelen);
741 ent = (0 <= pos) ? istate->cache[pos] : NULL;
742 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
745 /* When core.ignorecase=true, determine if a directory of the same name but differing
746 * case already exists within the Git repository. If it does, ensure the directory
747 * case of the file being added to the repository matches (is folded into) the existing
748 * entry's directory case.
750 if (ignore_case) {
751 adjust_dirname_case(istate, ce->name);
753 if (!(flags & ADD_CACHE_RENORMALIZE)) {
754 alias = index_file_exists(istate, ce->name,
755 ce_namelen(ce), ignore_case);
756 if (alias &&
757 !ce_stage(alias) &&
758 !ie_match_stat(istate, alias, st, ce_option)) {
759 /* Nothing changed, really */
760 if (!S_ISGITLINK(alias->ce_mode))
761 ce_mark_uptodate(alias);
762 alias->ce_flags |= CE_ADDED;
764 discard_cache_entry(ce);
765 return 0;
768 if (!intent_only) {
769 if (index_path(istate, &ce->oid, path, st, hash_flags)) {
770 discard_cache_entry(ce);
771 return error(_("unable to index file '%s'"), path);
773 } else
774 set_object_name_for_intent_to_add_entry(ce);
776 if (ignore_case && alias && different_name(ce, alias))
777 ce = create_alias_ce(istate, ce, alias);
778 ce->ce_flags |= CE_ADDED;
780 /* It was suspected to be racily clean, but it turns out to be Ok */
781 was_same = (alias &&
782 !ce_stage(alias) &&
783 oideq(&alias->oid, &ce->oid) &&
784 ce->ce_mode == alias->ce_mode);
786 if (pretend)
787 discard_cache_entry(ce);
788 else if (add_index_entry(istate, ce, add_option)) {
789 discard_cache_entry(ce);
790 return error(_("unable to add '%s' to index"), path);
792 if (verbose && !was_same)
793 printf("add '%s'\n", path);
794 return 0;
797 int add_file_to_index(struct index_state *istate, const char *path, int flags)
799 struct stat st;
800 if (lstat(path, &st))
801 die_errno(_("unable to stat '%s'"), path);
802 return add_to_index(istate, path, &st, flags);
805 struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
807 return mem_pool__ce_calloc(find_mem_pool(istate), len);
810 struct cache_entry *make_empty_transient_cache_entry(size_t len)
812 return xcalloc(1, cache_entry_size(len));
815 struct cache_entry *make_cache_entry(struct index_state *istate,
816 unsigned int mode,
817 const struct object_id *oid,
818 const char *path,
819 int stage,
820 unsigned int refresh_options)
822 struct cache_entry *ce, *ret;
823 int len;
825 if (!verify_path(path, mode)) {
826 error(_("invalid path '%s'"), path);
827 return NULL;
830 len = strlen(path);
831 ce = make_empty_cache_entry(istate, len);
833 oidcpy(&ce->oid, oid);
834 memcpy(ce->name, path, len);
835 ce->ce_flags = create_ce_flags(stage);
836 ce->ce_namelen = len;
837 ce->ce_mode = create_ce_mode(mode);
839 ret = refresh_cache_entry(istate, ce, refresh_options);
840 if (ret != ce)
841 discard_cache_entry(ce);
842 return ret;
845 struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
846 const char *path, int stage)
848 struct cache_entry *ce;
849 int len;
851 if (!verify_path(path, mode)) {
852 error(_("invalid path '%s'"), path);
853 return NULL;
856 len = strlen(path);
857 ce = make_empty_transient_cache_entry(len);
859 oidcpy(&ce->oid, oid);
860 memcpy(ce->name, path, len);
861 ce->ce_flags = create_ce_flags(stage);
862 ce->ce_namelen = len;
863 ce->ce_mode = create_ce_mode(mode);
865 return ce;
869 * Chmod an index entry with either +x or -x.
871 * Returns -1 if the chmod for the particular cache entry failed (if it's
872 * not a regular file), -2 if an invalid flip argument is passed in, 0
873 * otherwise.
875 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
876 char flip)
878 if (!S_ISREG(ce->ce_mode))
879 return -1;
880 switch (flip) {
881 case '+':
882 ce->ce_mode |= 0111;
883 break;
884 case '-':
885 ce->ce_mode &= ~0111;
886 break;
887 default:
888 return -2;
890 cache_tree_invalidate_path(istate, ce->name);
891 ce->ce_flags |= CE_UPDATE_IN_BASE;
892 mark_fsmonitor_invalid(istate, ce);
893 istate->cache_changed |= CE_ENTRY_CHANGED;
895 return 0;
898 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
900 int len = ce_namelen(a);
901 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
905 * We fundamentally don't like some paths: we don't want
906 * dot or dot-dot anywhere, and for obvious reasons don't
907 * want to recurse into ".git" either.
909 * Also, we don't want double slashes or slashes at the
910 * end that can make pathnames ambiguous.
912 static int verify_dotfile(const char *rest, unsigned mode)
915 * The first character was '.', but that
916 * has already been discarded, we now test
917 * the rest.
920 /* "." is not allowed */
921 if (*rest == '\0' || is_dir_sep(*rest))
922 return 0;
924 switch (*rest) {
926 * ".git" followed by NUL or slash is bad. Note that we match
927 * case-insensitively here, even if ignore_case is not set.
928 * This outlaws ".GIT" everywhere out of an abundance of caution,
929 * since there's really no good reason to allow it.
931 * Once we've seen ".git", we can also find ".gitmodules", etc (also
932 * case-insensitively).
934 case 'g':
935 case 'G':
936 if (rest[1] != 'i' && rest[1] != 'I')
937 break;
938 if (rest[2] != 't' && rest[2] != 'T')
939 break;
940 if (rest[3] == '\0' || is_dir_sep(rest[3]))
941 return 0;
942 if (S_ISLNK(mode)) {
943 rest += 3;
944 if (skip_iprefix(rest, "modules", &rest) &&
945 (*rest == '\0' || is_dir_sep(*rest)))
946 return 0;
948 break;
949 case '.':
950 if (rest[1] == '\0' || is_dir_sep(rest[1]))
951 return 0;
953 return 1;
956 int verify_path(const char *path, unsigned mode)
958 char c;
960 if (has_dos_drive_prefix(path))
961 return 0;
963 goto inside;
964 for (;;) {
965 if (!c)
966 return 1;
967 if (is_dir_sep(c)) {
968 inside:
969 if (protect_hfs) {
970 if (is_hfs_dotgit(path))
971 return 0;
972 if (S_ISLNK(mode)) {
973 if (is_hfs_dotgitmodules(path))
974 return 0;
977 if (protect_ntfs) {
978 if (is_ntfs_dotgit(path))
979 return 0;
980 if (S_ISLNK(mode)) {
981 if (is_ntfs_dotgitmodules(path))
982 return 0;
986 c = *path++;
987 if ((c == '.' && !verify_dotfile(path, mode)) ||
988 is_dir_sep(c) || c == '\0')
989 return 0;
991 c = *path++;
996 * Do we have another file that has the beginning components being a
997 * proper superset of the name we're trying to add?
999 static int has_file_name(struct index_state *istate,
1000 const struct cache_entry *ce, int pos, int ok_to_replace)
1002 int retval = 0;
1003 int len = ce_namelen(ce);
1004 int stage = ce_stage(ce);
1005 const char *name = ce->name;
1007 while (pos < istate->cache_nr) {
1008 struct cache_entry *p = istate->cache[pos++];
1010 if (len >= ce_namelen(p))
1011 break;
1012 if (memcmp(name, p->name, len))
1013 break;
1014 if (ce_stage(p) != stage)
1015 continue;
1016 if (p->name[len] != '/')
1017 continue;
1018 if (p->ce_flags & CE_REMOVE)
1019 continue;
1020 retval = -1;
1021 if (!ok_to_replace)
1022 break;
1023 remove_index_entry_at(istate, --pos);
1025 return retval;
1030 * Like strcmp(), but also return the offset of the first change.
1031 * If strings are equal, return the length.
1033 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1035 size_t k;
1037 if (!first_change)
1038 return strcmp(s1, s2);
1040 for (k = 0; s1[k] == s2[k]; k++)
1041 if (s1[k] == '\0')
1042 break;
1044 *first_change = k;
1045 return (unsigned char)s1[k] - (unsigned char)s2[k];
1049 * Do we have another file with a pathname that is a proper
1050 * subset of the name we're trying to add?
1052 * That is, is there another file in the index with a path
1053 * that matches a sub-directory in the given entry?
1055 static int has_dir_name(struct index_state *istate,
1056 const struct cache_entry *ce, int pos, int ok_to_replace)
1058 int retval = 0;
1059 int stage = ce_stage(ce);
1060 const char *name = ce->name;
1061 const char *slash = name + ce_namelen(ce);
1062 size_t len_eq_last;
1063 int cmp_last = 0;
1066 * We are frequently called during an iteration on a sorted
1067 * list of pathnames and while building a new index. Therefore,
1068 * there is a high probability that this entry will eventually
1069 * be appended to the index, rather than inserted in the middle.
1070 * If we can confirm that, we can avoid binary searches on the
1071 * components of the pathname.
1073 * Compare the entry's full path with the last path in the index.
1075 if (istate->cache_nr > 0) {
1076 cmp_last = strcmp_offset(name,
1077 istate->cache[istate->cache_nr - 1]->name,
1078 &len_eq_last);
1079 if (cmp_last > 0) {
1080 if (len_eq_last == 0) {
1082 * The entry sorts AFTER the last one in the
1083 * index and their paths have no common prefix,
1084 * so there cannot be a F/D conflict.
1086 return retval;
1087 } else {
1089 * The entry sorts AFTER the last one in the
1090 * index, but has a common prefix. Fall through
1091 * to the loop below to disect the entry's path
1092 * and see where the difference is.
1095 } else if (cmp_last == 0) {
1097 * The entry exactly matches the last one in the
1098 * index, but because of multiple stage and CE_REMOVE
1099 * items, we fall through and let the regular search
1100 * code handle it.
1105 for (;;) {
1106 size_t len;
1108 for (;;) {
1109 if (*--slash == '/')
1110 break;
1111 if (slash <= ce->name)
1112 return retval;
1114 len = slash - name;
1116 if (cmp_last > 0) {
1118 * (len + 1) is a directory boundary (including
1119 * the trailing slash). And since the loop is
1120 * decrementing "slash", the first iteration is
1121 * the longest directory prefix; subsequent
1122 * iterations consider parent directories.
1125 if (len + 1 <= len_eq_last) {
1127 * The directory prefix (including the trailing
1128 * slash) also appears as a prefix in the last
1129 * entry, so the remainder cannot collide (because
1130 * strcmp said the whole path was greater).
1132 * EQ: last: xxx/A
1133 * this: xxx/B
1135 * LT: last: xxx/file_A
1136 * this: xxx/file_B
1138 return retval;
1141 if (len > len_eq_last) {
1143 * This part of the directory prefix (excluding
1144 * the trailing slash) is longer than the known
1145 * equal portions, so this sub-directory cannot
1146 * collide with a file.
1148 * GT: last: xxxA
1149 * this: xxxB/file
1151 return retval;
1154 if (istate->cache_nr > 0 &&
1155 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1157 * The directory prefix lines up with part of
1158 * a longer file or directory name, but sorts
1159 * after it, so this sub-directory cannot
1160 * collide with a file.
1162 * last: xxx/yy-file (because '-' sorts before '/')
1163 * this: xxx/yy/abc
1165 return retval;
1169 * This is a possible collision. Fall through and
1170 * let the regular search code handle it.
1172 * last: xxx
1173 * this: xxx/file
1177 pos = index_name_stage_pos(istate, name, len, stage);
1178 if (pos >= 0) {
1180 * Found one, but not so fast. This could
1181 * be a marker that says "I was here, but
1182 * I am being removed". Such an entry is
1183 * not a part of the resulting tree, and
1184 * it is Ok to have a directory at the same
1185 * path.
1187 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1188 retval = -1;
1189 if (!ok_to_replace)
1190 break;
1191 remove_index_entry_at(istate, pos);
1192 continue;
1195 else
1196 pos = -pos-1;
1199 * Trivial optimization: if we find an entry that
1200 * already matches the sub-directory, then we know
1201 * we're ok, and we can exit.
1203 while (pos < istate->cache_nr) {
1204 struct cache_entry *p = istate->cache[pos];
1205 if ((ce_namelen(p) <= len) ||
1206 (p->name[len] != '/') ||
1207 memcmp(p->name, name, len))
1208 break; /* not our subdirectory */
1209 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1211 * p is at the same stage as our entry, and
1212 * is a subdirectory of what we are looking
1213 * at, so we cannot have conflicts at our
1214 * level or anything shorter.
1216 return retval;
1217 pos++;
1220 return retval;
1223 /* We may be in a situation where we already have path/file and path
1224 * is being added, or we already have path and path/file is being
1225 * added. Either one would result in a nonsense tree that has path
1226 * twice when git-write-tree tries to write it out. Prevent it.
1228 * If ok-to-replace is specified, we remove the conflicting entries
1229 * from the cache so the caller should recompute the insert position.
1230 * When this happens, we return non-zero.
1232 static int check_file_directory_conflict(struct index_state *istate,
1233 const struct cache_entry *ce,
1234 int pos, int ok_to_replace)
1236 int retval;
1239 * When ce is an "I am going away" entry, we allow it to be added
1241 if (ce->ce_flags & CE_REMOVE)
1242 return 0;
1245 * We check if the path is a sub-path of a subsequent pathname
1246 * first, since removing those will not change the position
1247 * in the array.
1249 retval = has_file_name(istate, ce, pos, ok_to_replace);
1252 * Then check if the path might have a clashing sub-directory
1253 * before it.
1255 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1258 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1260 int pos;
1261 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1262 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1263 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1264 int new_only = option & ADD_CACHE_NEW_ONLY;
1266 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1267 cache_tree_invalidate_path(istate, ce->name);
1270 * If this entry's path sorts after the last entry in the index,
1271 * we can avoid searching for it.
1273 if (istate->cache_nr > 0 &&
1274 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1275 pos = -istate->cache_nr - 1;
1276 else
1277 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1279 /* existing match? Just replace it. */
1280 if (pos >= 0) {
1281 if (!new_only)
1282 replace_index_entry(istate, pos, ce);
1283 return 0;
1285 pos = -pos-1;
1287 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1288 untracked_cache_add_to_index(istate, ce->name);
1291 * Inserting a merged entry ("stage 0") into the index
1292 * will always replace all non-merged entries..
1294 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1295 while (ce_same_name(istate->cache[pos], ce)) {
1296 ok_to_add = 1;
1297 if (!remove_index_entry_at(istate, pos))
1298 break;
1302 if (!ok_to_add)
1303 return -1;
1304 if (!verify_path(ce->name, ce->ce_mode))
1305 return error(_("invalid path '%s'"), ce->name);
1307 if (!skip_df_check &&
1308 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1309 if (!ok_to_replace)
1310 return error(_("'%s' appears as both a file and as a directory"),
1311 ce->name);
1312 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1313 pos = -pos-1;
1315 return pos + 1;
1318 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1320 int pos;
1322 if (option & ADD_CACHE_JUST_APPEND)
1323 pos = istate->cache_nr;
1324 else {
1325 int ret;
1326 ret = add_index_entry_with_check(istate, ce, option);
1327 if (ret <= 0)
1328 return ret;
1329 pos = ret - 1;
1332 /* Make sure the array is big enough .. */
1333 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1335 /* Add it in.. */
1336 istate->cache_nr++;
1337 if (istate->cache_nr > pos + 1)
1338 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1339 istate->cache_nr - pos - 1);
1340 set_index_entry(istate, pos, ce);
1341 istate->cache_changed |= CE_ENTRY_ADDED;
1342 return 0;
1346 * "refresh" does not calculate a new sha1 file or bring the
1347 * cache up-to-date for mode/content changes. But what it
1348 * _does_ do is to "re-match" the stat information of a file
1349 * with the cache, so that you can refresh the cache for a
1350 * file that hasn't been changed but where the stat entry is
1351 * out of date.
1353 * For example, you'd want to do this after doing a "git-read-tree",
1354 * to link up the stat cache details with the proper files.
1356 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1357 struct cache_entry *ce,
1358 unsigned int options, int *err,
1359 int *changed_ret)
1361 struct stat st;
1362 struct cache_entry *updated;
1363 int changed;
1364 int refresh = options & CE_MATCH_REFRESH;
1365 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1366 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1367 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1368 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1370 if (!refresh || ce_uptodate(ce))
1371 return ce;
1373 if (!ignore_fsmonitor)
1374 refresh_fsmonitor(istate);
1376 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1377 * that the change to the work tree does not matter and told
1378 * us not to worry.
1380 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1381 ce_mark_uptodate(ce);
1382 return ce;
1384 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1385 ce_mark_uptodate(ce);
1386 return ce;
1388 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1389 ce_mark_uptodate(ce);
1390 return ce;
1393 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1394 if (ignore_missing)
1395 return ce;
1396 if (err)
1397 *err = ENOENT;
1398 return NULL;
1401 if (lstat(ce->name, &st) < 0) {
1402 if (ignore_missing && errno == ENOENT)
1403 return ce;
1404 if (err)
1405 *err = errno;
1406 return NULL;
1409 changed = ie_match_stat(istate, ce, &st, options);
1410 if (changed_ret)
1411 *changed_ret = changed;
1412 if (!changed) {
1414 * The path is unchanged. If we were told to ignore
1415 * valid bit, then we did the actual stat check and
1416 * found that the entry is unmodified. If the entry
1417 * is not marked VALID, this is the place to mark it
1418 * valid again, under "assume unchanged" mode.
1420 if (ignore_valid && assume_unchanged &&
1421 !(ce->ce_flags & CE_VALID))
1422 ; /* mark this one VALID again */
1423 else {
1425 * We do not mark the index itself "modified"
1426 * because CE_UPTODATE flag is in-core only;
1427 * we are not going to write this change out.
1429 if (!S_ISGITLINK(ce->ce_mode)) {
1430 ce_mark_uptodate(ce);
1431 mark_fsmonitor_valid(ce);
1433 return ce;
1437 if (ie_modified(istate, ce, &st, options)) {
1438 if (err)
1439 *err = EINVAL;
1440 return NULL;
1443 updated = make_empty_cache_entry(istate, ce_namelen(ce));
1444 copy_cache_entry(updated, ce);
1445 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1446 fill_stat_cache_info(updated, &st);
1448 * If ignore_valid is not set, we should leave CE_VALID bit
1449 * alone. Otherwise, paths marked with --no-assume-unchanged
1450 * (i.e. things to be edited) will reacquire CE_VALID bit
1451 * automatically, which is not really what we want.
1453 if (!ignore_valid && assume_unchanged &&
1454 !(ce->ce_flags & CE_VALID))
1455 updated->ce_flags &= ~CE_VALID;
1457 /* istate->cache_changed is updated in the caller */
1458 return updated;
1461 static void show_file(const char * fmt, const char * name, int in_porcelain,
1462 int * first, const char *header_msg)
1464 if (in_porcelain && *first && header_msg) {
1465 printf("%s\n", header_msg);
1466 *first = 0;
1468 printf(fmt, name);
1471 int refresh_index(struct index_state *istate, unsigned int flags,
1472 const struct pathspec *pathspec,
1473 char *seen, const char *header_msg)
1475 int i;
1476 int has_errors = 0;
1477 int really = (flags & REFRESH_REALLY) != 0;
1478 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1479 int quiet = (flags & REFRESH_QUIET) != 0;
1480 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1481 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1482 int first = 1;
1483 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1484 unsigned int options = (CE_MATCH_REFRESH |
1485 (really ? CE_MATCH_IGNORE_VALID : 0) |
1486 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1487 const char *modified_fmt;
1488 const char *deleted_fmt;
1489 const char *typechange_fmt;
1490 const char *added_fmt;
1491 const char *unmerged_fmt;
1492 struct progress *progress = NULL;
1494 if (flags & REFRESH_PROGRESS && isatty(2))
1495 progress = start_delayed_progress(_("Refresh index"),
1496 istate->cache_nr);
1498 trace_performance_enter();
1499 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1500 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1501 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1502 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1503 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1505 * Use the multi-threaded preload_index() to refresh most of the
1506 * cache entries quickly then in the single threaded loop below,
1507 * we only have to do the special cases that are left.
1509 preload_index(istate, pathspec, 0);
1510 for (i = 0; i < istate->cache_nr; i++) {
1511 struct cache_entry *ce, *new_entry;
1512 int cache_errno = 0;
1513 int changed = 0;
1514 int filtered = 0;
1516 ce = istate->cache[i];
1517 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1518 continue;
1520 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1521 filtered = 1;
1523 if (ce_stage(ce)) {
1524 while ((i < istate->cache_nr) &&
1525 ! strcmp(istate->cache[i]->name, ce->name))
1526 i++;
1527 i--;
1528 if (allow_unmerged)
1529 continue;
1530 if (!filtered)
1531 show_file(unmerged_fmt, ce->name, in_porcelain,
1532 &first, header_msg);
1533 has_errors = 1;
1534 continue;
1537 if (filtered)
1538 continue;
1540 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1541 if (new_entry == ce)
1542 continue;
1543 if (progress)
1544 display_progress(progress, i);
1545 if (!new_entry) {
1546 const char *fmt;
1548 if (really && cache_errno == EINVAL) {
1549 /* If we are doing --really-refresh that
1550 * means the index is not valid anymore.
1552 ce->ce_flags &= ~CE_VALID;
1553 ce->ce_flags |= CE_UPDATE_IN_BASE;
1554 mark_fsmonitor_invalid(istate, ce);
1555 istate->cache_changed |= CE_ENTRY_CHANGED;
1557 if (quiet)
1558 continue;
1560 if (cache_errno == ENOENT)
1561 fmt = deleted_fmt;
1562 else if (ce_intent_to_add(ce))
1563 fmt = added_fmt; /* must be before other checks */
1564 else if (changed & TYPE_CHANGED)
1565 fmt = typechange_fmt;
1566 else
1567 fmt = modified_fmt;
1568 show_file(fmt,
1569 ce->name, in_porcelain, &first, header_msg);
1570 has_errors = 1;
1571 continue;
1574 replace_index_entry(istate, i, new_entry);
1576 if (progress) {
1577 display_progress(progress, istate->cache_nr);
1578 stop_progress(&progress);
1580 trace_performance_leave("refresh index");
1581 return has_errors;
1584 struct cache_entry *refresh_cache_entry(struct index_state *istate,
1585 struct cache_entry *ce,
1586 unsigned int options)
1588 return refresh_cache_ent(istate, ce, options, NULL, NULL);
1592 /*****************************************************************
1593 * Index File I/O
1594 *****************************************************************/
1596 #define INDEX_FORMAT_DEFAULT 3
1598 static unsigned int get_index_format_default(void)
1600 char *envversion = getenv("GIT_INDEX_VERSION");
1601 char *endp;
1602 int value;
1603 unsigned int version = INDEX_FORMAT_DEFAULT;
1605 if (!envversion) {
1606 if (!git_config_get_int("index.version", &value))
1607 version = value;
1608 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1609 warning(_("index.version set, but the value is invalid.\n"
1610 "Using version %i"), INDEX_FORMAT_DEFAULT);
1611 return INDEX_FORMAT_DEFAULT;
1613 return version;
1616 version = strtoul(envversion, &endp, 10);
1617 if (*endp ||
1618 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1619 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1620 "Using version %i"), INDEX_FORMAT_DEFAULT);
1621 version = INDEX_FORMAT_DEFAULT;
1623 return version;
1627 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1628 * Again - this is just a (very strong in practice) heuristic that
1629 * the inode hasn't changed.
1631 * We save the fields in big-endian order to allow using the
1632 * index file over NFS transparently.
1634 struct ondisk_cache_entry {
1635 struct cache_time ctime;
1636 struct cache_time mtime;
1637 uint32_t dev;
1638 uint32_t ino;
1639 uint32_t mode;
1640 uint32_t uid;
1641 uint32_t gid;
1642 uint32_t size;
1643 unsigned char sha1[20];
1644 uint16_t flags;
1645 char name[FLEX_ARRAY]; /* more */
1649 * This struct is used when CE_EXTENDED bit is 1
1650 * The struct must match ondisk_cache_entry exactly from
1651 * ctime till flags
1653 struct ondisk_cache_entry_extended {
1654 struct cache_time ctime;
1655 struct cache_time mtime;
1656 uint32_t dev;
1657 uint32_t ino;
1658 uint32_t mode;
1659 uint32_t uid;
1660 uint32_t gid;
1661 uint32_t size;
1662 unsigned char sha1[20];
1663 uint16_t flags;
1664 uint16_t flags2;
1665 char name[FLEX_ARRAY]; /* more */
1668 /* These are only used for v3 or lower */
1669 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1670 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1671 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1672 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1673 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1674 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1675 ondisk_cache_entry_size(ce_namelen(ce)))
1677 /* Allow fsck to force verification of the index checksum. */
1678 int verify_index_checksum;
1680 /* Allow fsck to force verification of the cache entry order. */
1681 int verify_ce_order;
1683 static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1685 git_hash_ctx c;
1686 unsigned char hash[GIT_MAX_RAWSZ];
1687 int hdr_version;
1689 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1690 return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1691 hdr_version = ntohl(hdr->hdr_version);
1692 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1693 return error(_("bad index version %d"), hdr_version);
1695 if (!verify_index_checksum)
1696 return 0;
1698 the_hash_algo->init_fn(&c);
1699 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1700 the_hash_algo->final_fn(hash, &c);
1701 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1702 return error(_("bad index file sha1 signature"));
1703 return 0;
1706 static int read_index_extension(struct index_state *istate,
1707 const char *ext, const char *data, unsigned long sz)
1709 switch (CACHE_EXT(ext)) {
1710 case CACHE_EXT_TREE:
1711 istate->cache_tree = cache_tree_read(data, sz);
1712 break;
1713 case CACHE_EXT_RESOLVE_UNDO:
1714 istate->resolve_undo = resolve_undo_read(data, sz);
1715 break;
1716 case CACHE_EXT_LINK:
1717 if (read_link_extension(istate, data, sz))
1718 return -1;
1719 break;
1720 case CACHE_EXT_UNTRACKED:
1721 istate->untracked = read_untracked_extension(data, sz);
1722 break;
1723 case CACHE_EXT_FSMONITOR:
1724 read_fsmonitor_extension(istate, data, sz);
1725 break;
1726 case CACHE_EXT_ENDOFINDEXENTRIES:
1727 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1728 /* already handled in do_read_index() */
1729 break;
1730 default:
1731 if (*ext < 'A' || 'Z' < *ext)
1732 return error(_("index uses %.4s extension, which we do not understand"),
1733 ext);
1734 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1735 break;
1737 return 0;
1740 static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1741 unsigned int version,
1742 struct ondisk_cache_entry *ondisk,
1743 unsigned long *ent_size,
1744 const struct cache_entry *previous_ce)
1746 struct cache_entry *ce;
1747 size_t len;
1748 const char *name;
1749 unsigned int flags;
1750 size_t copy_len = 0;
1752 * Adjacent cache entries tend to share the leading paths, so it makes
1753 * sense to only store the differences in later entries. In the v4
1754 * on-disk format of the index, each on-disk cache entry stores the
1755 * number of bytes to be stripped from the end of the previous name,
1756 * and the bytes to append to the result, to come up with its name.
1758 int expand_name_field = version == 4;
1760 /* On-disk flags are just 16 bits */
1761 flags = get_be16(&ondisk->flags);
1762 len = flags & CE_NAMEMASK;
1764 if (flags & CE_EXTENDED) {
1765 struct ondisk_cache_entry_extended *ondisk2;
1766 int extended_flags;
1767 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1768 extended_flags = get_be16(&ondisk2->flags2) << 16;
1769 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1770 if (extended_flags & ~CE_EXTENDED_FLAGS)
1771 die(_("unknown index entry format 0x%08x"), extended_flags);
1772 flags |= extended_flags;
1773 name = ondisk2->name;
1775 else
1776 name = ondisk->name;
1778 if (expand_name_field) {
1779 const unsigned char *cp = (const unsigned char *)name;
1780 size_t strip_len, previous_len;
1782 /* If we're at the begining of a block, ignore the previous name */
1783 strip_len = decode_varint(&cp);
1784 if (previous_ce) {
1785 previous_len = previous_ce->ce_namelen;
1786 if (previous_len < strip_len)
1787 die(_("malformed name field in the index, near path '%s'"),
1788 previous_ce->name);
1789 copy_len = previous_len - strip_len;
1791 name = (const char *)cp;
1794 if (len == CE_NAMEMASK) {
1795 len = strlen(name);
1796 if (expand_name_field)
1797 len += copy_len;
1800 ce = mem_pool__ce_alloc(ce_mem_pool, len);
1802 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1803 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1804 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1805 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1806 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1807 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1808 ce->ce_mode = get_be32(&ondisk->mode);
1809 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1810 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1811 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1812 ce->ce_flags = flags & ~CE_NAMEMASK;
1813 ce->ce_namelen = len;
1814 ce->index = 0;
1815 hashcpy(ce->oid.hash, ondisk->sha1);
1817 if (expand_name_field) {
1818 if (copy_len)
1819 memcpy(ce->name, previous_ce->name, copy_len);
1820 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1821 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1822 } else {
1823 memcpy(ce->name, name, len + 1);
1824 *ent_size = ondisk_ce_size(ce);
1826 return ce;
1829 static void check_ce_order(struct index_state *istate)
1831 unsigned int i;
1833 if (!verify_ce_order)
1834 return;
1836 for (i = 1; i < istate->cache_nr; i++) {
1837 struct cache_entry *ce = istate->cache[i - 1];
1838 struct cache_entry *next_ce = istate->cache[i];
1839 int name_compare = strcmp(ce->name, next_ce->name);
1841 if (0 < name_compare)
1842 die(_("unordered stage entries in index"));
1843 if (!name_compare) {
1844 if (!ce_stage(ce))
1845 die(_("multiple stage entries for merged file '%s'"),
1846 ce->name);
1847 if (ce_stage(ce) > ce_stage(next_ce))
1848 die(_("unordered stage entries for '%s'"),
1849 ce->name);
1854 static void tweak_untracked_cache(struct index_state *istate)
1856 switch (git_config_get_untracked_cache()) {
1857 case -1: /* keep: do nothing */
1858 break;
1859 case 0: /* false */
1860 remove_untracked_cache(istate);
1861 break;
1862 case 1: /* true */
1863 add_untracked_cache(istate);
1864 break;
1865 default: /* unknown value: do nothing */
1866 break;
1870 static void tweak_split_index(struct index_state *istate)
1872 switch (git_config_get_split_index()) {
1873 case -1: /* unset: do nothing */
1874 break;
1875 case 0: /* false */
1876 remove_split_index(istate);
1877 break;
1878 case 1: /* true */
1879 add_split_index(istate);
1880 break;
1881 default: /* unknown value: do nothing */
1882 break;
1886 static void post_read_index_from(struct index_state *istate)
1888 check_ce_order(istate);
1889 tweak_untracked_cache(istate);
1890 tweak_split_index(istate);
1891 tweak_fsmonitor(istate);
1894 static size_t estimate_cache_size_from_compressed(unsigned int entries)
1896 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1899 static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1901 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1904 * Account for potential alignment differences.
1906 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1907 return ondisk_size + entries * per_entry;
1910 struct index_entry_offset
1912 /* starting byte offset into index file, count of index entries in this block */
1913 int offset, nr;
1916 struct index_entry_offset_table
1918 int nr;
1919 struct index_entry_offset entries[FLEX_ARRAY];
1922 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1923 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1925 static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1926 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1928 struct load_index_extensions
1930 pthread_t pthread;
1931 struct index_state *istate;
1932 const char *mmap;
1933 size_t mmap_size;
1934 unsigned long src_offset;
1937 static void *load_index_extensions(void *_data)
1939 struct load_index_extensions *p = _data;
1940 unsigned long src_offset = p->src_offset;
1942 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1943 /* After an array of active_nr index entries,
1944 * there can be arbitrary number of extended
1945 * sections, each of which is prefixed with
1946 * extension name (4-byte) and section length
1947 * in 4-byte network byte order.
1949 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1950 if (read_index_extension(p->istate,
1951 p->mmap + src_offset,
1952 p->mmap + src_offset + 8,
1953 extsize) < 0) {
1954 munmap((void *)p->mmap, p->mmap_size);
1955 die(_("index file corrupt"));
1957 src_offset += 8;
1958 src_offset += extsize;
1961 return NULL;
1965 * A helper function that will load the specified range of cache entries
1966 * from the memory mapped file and add them to the given index.
1968 static unsigned long load_cache_entry_block(struct index_state *istate,
1969 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1970 unsigned long start_offset, const struct cache_entry *previous_ce)
1972 int i;
1973 unsigned long src_offset = start_offset;
1975 for (i = offset; i < offset + nr; i++) {
1976 struct ondisk_cache_entry *disk_ce;
1977 struct cache_entry *ce;
1978 unsigned long consumed;
1980 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1981 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1982 set_index_entry(istate, i, ce);
1984 src_offset += consumed;
1985 previous_ce = ce;
1987 return src_offset - start_offset;
1990 static unsigned long load_all_cache_entries(struct index_state *istate,
1991 const char *mmap, size_t mmap_size, unsigned long src_offset)
1993 unsigned long consumed;
1995 if (istate->version == 4) {
1996 mem_pool_init(&istate->ce_mem_pool,
1997 estimate_cache_size_from_compressed(istate->cache_nr));
1998 } else {
1999 mem_pool_init(&istate->ce_mem_pool,
2000 estimate_cache_size(mmap_size, istate->cache_nr));
2003 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2004 0, istate->cache_nr, mmap, src_offset, NULL);
2005 return consumed;
2009 * Mostly randomly chosen maximum thread counts: we
2010 * cap the parallelism to online_cpus() threads, and we want
2011 * to have at least 10000 cache entries per thread for it to
2012 * be worth starting a thread.
2015 #define THREAD_COST (10000)
2017 struct load_cache_entries_thread_data
2019 pthread_t pthread;
2020 struct index_state *istate;
2021 struct mem_pool *ce_mem_pool;
2022 int offset;
2023 const char *mmap;
2024 struct index_entry_offset_table *ieot;
2025 int ieot_start; /* starting index into the ieot array */
2026 int ieot_blocks; /* count of ieot entries to process */
2027 unsigned long consumed; /* return # of bytes in index file processed */
2031 * A thread proc to run the load_cache_entries() computation
2032 * across multiple background threads.
2034 static void *load_cache_entries_thread(void *_data)
2036 struct load_cache_entries_thread_data *p = _data;
2037 int i;
2039 /* iterate across all ieot blocks assigned to this thread */
2040 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2041 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2042 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2043 p->offset += p->ieot->entries[i].nr;
2045 return NULL;
2048 static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2049 unsigned long src_offset, int nr_threads, struct index_entry_offset_table *ieot)
2051 int i, offset, ieot_blocks, ieot_start, err;
2052 struct load_cache_entries_thread_data *data;
2053 unsigned long consumed = 0;
2055 /* a little sanity checking */
2056 if (istate->name_hash_initialized)
2057 BUG("the name hash isn't thread safe");
2059 mem_pool_init(&istate->ce_mem_pool, 0);
2061 /* ensure we have no more threads than we have blocks to process */
2062 if (nr_threads > ieot->nr)
2063 nr_threads = ieot->nr;
2064 data = xcalloc(nr_threads, sizeof(*data));
2066 offset = ieot_start = 0;
2067 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2068 for (i = 0; i < nr_threads; i++) {
2069 struct load_cache_entries_thread_data *p = &data[i];
2070 int nr, j;
2072 if (ieot_start + ieot_blocks > ieot->nr)
2073 ieot_blocks = ieot->nr - ieot_start;
2075 p->istate = istate;
2076 p->offset = offset;
2077 p->mmap = mmap;
2078 p->ieot = ieot;
2079 p->ieot_start = ieot_start;
2080 p->ieot_blocks = ieot_blocks;
2082 /* create a mem_pool for each thread */
2083 nr = 0;
2084 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2085 nr += p->ieot->entries[j].nr;
2086 if (istate->version == 4) {
2087 mem_pool_init(&p->ce_mem_pool,
2088 estimate_cache_size_from_compressed(nr));
2089 } else {
2090 mem_pool_init(&p->ce_mem_pool,
2091 estimate_cache_size(mmap_size, nr));
2094 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2095 if (err)
2096 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2098 /* increment by the number of cache entries in the ieot block being processed */
2099 for (j = 0; j < ieot_blocks; j++)
2100 offset += ieot->entries[ieot_start + j].nr;
2101 ieot_start += ieot_blocks;
2104 for (i = 0; i < nr_threads; i++) {
2105 struct load_cache_entries_thread_data *p = &data[i];
2107 err = pthread_join(p->pthread, NULL);
2108 if (err)
2109 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2110 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2111 consumed += p->consumed;
2114 free(data);
2116 return consumed;
2119 /* remember to discard_cache() before reading a different cache! */
2120 int do_read_index(struct index_state *istate, const char *path, int must_exist)
2122 int fd;
2123 struct stat st;
2124 unsigned long src_offset;
2125 const struct cache_header *hdr;
2126 const char *mmap;
2127 size_t mmap_size;
2128 struct load_index_extensions p;
2129 size_t extension_offset = 0;
2130 int nr_threads, cpus;
2131 struct index_entry_offset_table *ieot = NULL;
2133 if (istate->initialized)
2134 return istate->cache_nr;
2136 istate->timestamp.sec = 0;
2137 istate->timestamp.nsec = 0;
2138 fd = open(path, O_RDONLY);
2139 if (fd < 0) {
2140 if (!must_exist && errno == ENOENT)
2141 return 0;
2142 die_errno(_("%s: index file open failed"), path);
2145 if (fstat(fd, &st))
2146 die_errno(_("%s: cannot stat the open index"), path);
2148 mmap_size = xsize_t(st.st_size);
2149 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2150 die(_("%s: index file smaller than expected"), path);
2152 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2153 if (mmap == MAP_FAILED)
2154 die_errno(_("%s: unable to map index file"), path);
2155 close(fd);
2157 hdr = (const struct cache_header *)mmap;
2158 if (verify_hdr(hdr, mmap_size) < 0)
2159 goto unmap;
2161 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2162 istate->version = ntohl(hdr->hdr_version);
2163 istate->cache_nr = ntohl(hdr->hdr_entries);
2164 istate->cache_alloc = alloc_nr(istate->cache_nr);
2165 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2166 istate->initialized = 1;
2168 p.istate = istate;
2169 p.mmap = mmap;
2170 p.mmap_size = mmap_size;
2172 src_offset = sizeof(*hdr);
2174 if (git_config_get_index_threads(&nr_threads))
2175 nr_threads = 1;
2177 /* TODO: does creating more threads than cores help? */
2178 if (!nr_threads) {
2179 nr_threads = istate->cache_nr / THREAD_COST;
2180 cpus = online_cpus();
2181 if (nr_threads > cpus)
2182 nr_threads = cpus;
2185 if (!HAVE_THREADS)
2186 nr_threads = 1;
2188 if (nr_threads > 1) {
2189 extension_offset = read_eoie_extension(mmap, mmap_size);
2190 if (extension_offset) {
2191 int err;
2193 p.src_offset = extension_offset;
2194 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2195 if (err)
2196 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2198 nr_threads--;
2203 * Locate and read the index entry offset table so that we can use it
2204 * to multi-thread the reading of the cache entries.
2206 if (extension_offset && nr_threads > 1)
2207 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2209 if (ieot) {
2210 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, src_offset, nr_threads, ieot);
2211 free(ieot);
2212 } else {
2213 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2216 istate->timestamp.sec = st.st_mtime;
2217 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2219 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2220 if (extension_offset) {
2221 int ret = pthread_join(p.pthread, NULL);
2222 if (ret)
2223 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2224 } else {
2225 p.src_offset = src_offset;
2226 load_index_extensions(&p);
2228 munmap((void *)mmap, mmap_size);
2231 * TODO trace2: replace "the_repository" with the actual repo instance
2232 * that is associated with the given "istate".
2234 trace2_data_intmax("index", the_repository, "read/version",
2235 istate->version);
2236 trace2_data_intmax("index", the_repository, "read/cache_nr",
2237 istate->cache_nr);
2239 return istate->cache_nr;
2241 unmap:
2242 munmap((void *)mmap, mmap_size);
2243 die(_("index file corrupt"));
2247 * Signal that the shared index is used by updating its mtime.
2249 * This way, shared index can be removed if they have not been used
2250 * for some time.
2252 static void freshen_shared_index(const char *shared_index, int warn)
2254 if (!check_and_freshen_file(shared_index, 1) && warn)
2255 warning(_("could not freshen shared index '%s'"), shared_index);
2258 int read_index_from(struct index_state *istate, const char *path,
2259 const char *gitdir)
2261 struct split_index *split_index;
2262 int ret;
2263 char *base_oid_hex;
2264 char *base_path;
2266 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2267 if (istate->initialized)
2268 return istate->cache_nr;
2271 * TODO trace2: replace "the_repository" with the actual repo instance
2272 * that is associated with the given "istate".
2274 trace2_region_enter_printf("index", "do_read_index", the_repository,
2275 "%s", path);
2276 trace_performance_enter();
2277 ret = do_read_index(istate, path, 0);
2278 trace_performance_leave("read cache %s", path);
2279 trace2_region_leave_printf("index", "do_read_index", the_repository,
2280 "%s", path);
2282 split_index = istate->split_index;
2283 if (!split_index || is_null_oid(&split_index->base_oid)) {
2284 post_read_index_from(istate);
2285 return ret;
2288 trace_performance_enter();
2289 if (split_index->base)
2290 discard_index(split_index->base);
2291 else
2292 split_index->base = xcalloc(1, sizeof(*split_index->base));
2294 base_oid_hex = oid_to_hex(&split_index->base_oid);
2295 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2296 trace2_region_enter_printf("index", "shared/do_read_index",
2297 the_repository, "%s", base_path);
2298 ret = do_read_index(split_index->base, base_path, 1);
2299 trace2_region_leave_printf("index", "shared/do_read_index",
2300 the_repository, "%s", base_path);
2301 if (!oideq(&split_index->base_oid, &split_index->base->oid))
2302 die(_("broken index, expect %s in %s, got %s"),
2303 base_oid_hex, base_path,
2304 oid_to_hex(&split_index->base->oid));
2306 freshen_shared_index(base_path, 0);
2307 merge_base_index(istate);
2308 post_read_index_from(istate);
2309 trace_performance_leave("read cache %s", base_path);
2310 free(base_path);
2311 return ret;
2314 int is_index_unborn(struct index_state *istate)
2316 return (!istate->cache_nr && !istate->timestamp.sec);
2319 int discard_index(struct index_state *istate)
2322 * Cache entries in istate->cache[] should have been allocated
2323 * from the memory pool associated with this index, or from an
2324 * associated split_index. There is no need to free individual
2325 * cache entries. validate_cache_entries can detect when this
2326 * assertion does not hold.
2328 validate_cache_entries(istate);
2330 resolve_undo_clear_index(istate);
2331 istate->cache_nr = 0;
2332 istate->cache_changed = 0;
2333 istate->timestamp.sec = 0;
2334 istate->timestamp.nsec = 0;
2335 free_name_hash(istate);
2336 cache_tree_free(&(istate->cache_tree));
2337 istate->initialized = 0;
2338 FREE_AND_NULL(istate->cache);
2339 istate->cache_alloc = 0;
2340 discard_split_index(istate);
2341 free_untracked_cache(istate->untracked);
2342 istate->untracked = NULL;
2344 if (istate->ce_mem_pool) {
2345 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2346 istate->ce_mem_pool = NULL;
2349 return 0;
2353 * Validate the cache entries of this index.
2354 * All cache entries associated with this index
2355 * should have been allocated by the memory pool
2356 * associated with this index, or by a referenced
2357 * split index.
2359 void validate_cache_entries(const struct index_state *istate)
2361 int i;
2363 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2364 return;
2366 for (i = 0; i < istate->cache_nr; i++) {
2367 if (!istate) {
2368 BUG("cache entry is not allocated from expected memory pool");
2369 } else if (!istate->ce_mem_pool ||
2370 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2371 if (!istate->split_index ||
2372 !istate->split_index->base ||
2373 !istate->split_index->base->ce_mem_pool ||
2374 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2375 BUG("cache entry is not allocated from expected memory pool");
2380 if (istate->split_index)
2381 validate_cache_entries(istate->split_index->base);
2384 int unmerged_index(const struct index_state *istate)
2386 int i;
2387 for (i = 0; i < istate->cache_nr; i++) {
2388 if (ce_stage(istate->cache[i]))
2389 return 1;
2391 return 0;
2394 int repo_index_has_changes(struct repository *repo,
2395 struct tree *tree,
2396 struct strbuf *sb)
2398 struct index_state *istate = repo->index;
2399 struct object_id cmp;
2400 int i;
2402 if (tree)
2403 cmp = tree->object.oid;
2404 if (tree || !get_oid_tree("HEAD", &cmp)) {
2405 struct diff_options opt;
2407 repo_diff_setup(repo, &opt);
2408 opt.flags.exit_with_status = 1;
2409 if (!sb)
2410 opt.flags.quick = 1;
2411 do_diff_cache(&cmp, &opt);
2412 diffcore_std(&opt);
2413 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2414 if (i)
2415 strbuf_addch(sb, ' ');
2416 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2418 diff_flush(&opt);
2419 return opt.flags.has_changes != 0;
2420 } else {
2421 for (i = 0; sb && i < istate->cache_nr; i++) {
2422 if (i)
2423 strbuf_addch(sb, ' ');
2424 strbuf_addstr(sb, istate->cache[i]->name);
2426 return !!istate->cache_nr;
2430 #define WRITE_BUFFER_SIZE 8192
2431 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2432 static unsigned long write_buffer_len;
2434 static int ce_write_flush(git_hash_ctx *context, int fd)
2436 unsigned int buffered = write_buffer_len;
2437 if (buffered) {
2438 the_hash_algo->update_fn(context, write_buffer, buffered);
2439 if (write_in_full(fd, write_buffer, buffered) < 0)
2440 return -1;
2441 write_buffer_len = 0;
2443 return 0;
2446 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2448 while (len) {
2449 unsigned int buffered = write_buffer_len;
2450 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2451 if (partial > len)
2452 partial = len;
2453 memcpy(write_buffer + buffered, data, partial);
2454 buffered += partial;
2455 if (buffered == WRITE_BUFFER_SIZE) {
2456 write_buffer_len = buffered;
2457 if (ce_write_flush(context, fd))
2458 return -1;
2459 buffered = 0;
2461 write_buffer_len = buffered;
2462 len -= partial;
2463 data = (char *) data + partial;
2465 return 0;
2468 static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2469 int fd, unsigned int ext, unsigned int sz)
2471 ext = htonl(ext);
2472 sz = htonl(sz);
2473 if (eoie_context) {
2474 the_hash_algo->update_fn(eoie_context, &ext, 4);
2475 the_hash_algo->update_fn(eoie_context, &sz, 4);
2477 return ((ce_write(context, fd, &ext, 4) < 0) ||
2478 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2481 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2483 unsigned int left = write_buffer_len;
2485 if (left) {
2486 write_buffer_len = 0;
2487 the_hash_algo->update_fn(context, write_buffer, left);
2490 /* Flush first if not enough space for hash signature */
2491 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2492 if (write_in_full(fd, write_buffer, left) < 0)
2493 return -1;
2494 left = 0;
2497 /* Append the hash signature at the end */
2498 the_hash_algo->final_fn(write_buffer + left, context);
2499 hashcpy(hash, write_buffer + left);
2500 left += the_hash_algo->rawsz;
2501 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2504 static void ce_smudge_racily_clean_entry(struct index_state *istate,
2505 struct cache_entry *ce)
2508 * The only thing we care about in this function is to smudge the
2509 * falsely clean entry due to touch-update-touch race, so we leave
2510 * everything else as they are. We are called for entries whose
2511 * ce_stat_data.sd_mtime match the index file mtime.
2513 * Note that this actually does not do much for gitlinks, for
2514 * which ce_match_stat_basic() always goes to the actual
2515 * contents. The caller checks with is_racy_timestamp() which
2516 * always says "no" for gitlinks, so we are not called for them ;-)
2518 struct stat st;
2520 if (lstat(ce->name, &st) < 0)
2521 return;
2522 if (ce_match_stat_basic(ce, &st))
2523 return;
2524 if (ce_modified_check_fs(istate, ce, &st)) {
2525 /* This is "racily clean"; smudge it. Note that this
2526 * is a tricky code. At first glance, it may appear
2527 * that it can break with this sequence:
2529 * $ echo xyzzy >frotz
2530 * $ git-update-index --add frotz
2531 * $ : >frotz
2532 * $ sleep 3
2533 * $ echo filfre >nitfol
2534 * $ git-update-index --add nitfol
2536 * but it does not. When the second update-index runs,
2537 * it notices that the entry "frotz" has the same timestamp
2538 * as index, and if we were to smudge it by resetting its
2539 * size to zero here, then the object name recorded
2540 * in index is the 6-byte file but the cached stat information
2541 * becomes zero --- which would then match what we would
2542 * obtain from the filesystem next time we stat("frotz").
2544 * However, the second update-index, before calling
2545 * this function, notices that the cached size is 6
2546 * bytes and what is on the filesystem is an empty
2547 * file, and never calls us, so the cached size information
2548 * for "frotz" stays 6 which does not match the filesystem.
2550 ce->ce_stat_data.sd_size = 0;
2554 /* Copy miscellaneous fields but not the name */
2555 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2556 struct cache_entry *ce)
2558 short flags;
2560 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2561 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2562 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2563 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2564 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2565 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2566 ondisk->mode = htonl(ce->ce_mode);
2567 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2568 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2569 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2570 hashcpy(ondisk->sha1, ce->oid.hash);
2572 flags = ce->ce_flags & ~CE_NAMEMASK;
2573 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2574 ondisk->flags = htons(flags);
2575 if (ce->ce_flags & CE_EXTENDED) {
2576 struct ondisk_cache_entry_extended *ondisk2;
2577 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2578 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2582 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2583 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2585 int size;
2586 int result;
2587 unsigned int saved_namelen;
2588 int stripped_name = 0;
2589 static unsigned char padding[8] = { 0x00 };
2591 if (ce->ce_flags & CE_STRIP_NAME) {
2592 saved_namelen = ce_namelen(ce);
2593 ce->ce_namelen = 0;
2594 stripped_name = 1;
2597 if (ce->ce_flags & CE_EXTENDED)
2598 size = offsetof(struct ondisk_cache_entry_extended, name);
2599 else
2600 size = offsetof(struct ondisk_cache_entry, name);
2602 if (!previous_name) {
2603 int len = ce_namelen(ce);
2604 copy_cache_entry_to_ondisk(ondisk, ce);
2605 result = ce_write(c, fd, ondisk, size);
2606 if (!result)
2607 result = ce_write(c, fd, ce->name, len);
2608 if (!result)
2609 result = ce_write(c, fd, padding, align_padding_size(size, len));
2610 } else {
2611 int common, to_remove, prefix_size;
2612 unsigned char to_remove_vi[16];
2613 for (common = 0;
2614 (ce->name[common] &&
2615 common < previous_name->len &&
2616 ce->name[common] == previous_name->buf[common]);
2617 common++)
2618 ; /* still matching */
2619 to_remove = previous_name->len - common;
2620 prefix_size = encode_varint(to_remove, to_remove_vi);
2622 copy_cache_entry_to_ondisk(ondisk, ce);
2623 result = ce_write(c, fd, ondisk, size);
2624 if (!result)
2625 result = ce_write(c, fd, to_remove_vi, prefix_size);
2626 if (!result)
2627 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2628 if (!result)
2629 result = ce_write(c, fd, padding, 1);
2631 strbuf_splice(previous_name, common, to_remove,
2632 ce->name + common, ce_namelen(ce) - common);
2634 if (stripped_name) {
2635 ce->ce_namelen = saved_namelen;
2636 ce->ce_flags &= ~CE_STRIP_NAME;
2639 return result;
2643 * This function verifies if index_state has the correct sha1 of the
2644 * index file. Don't die if we have any other failure, just return 0.
2646 static int verify_index_from(const struct index_state *istate, const char *path)
2648 int fd;
2649 ssize_t n;
2650 struct stat st;
2651 unsigned char hash[GIT_MAX_RAWSZ];
2653 if (!istate->initialized)
2654 return 0;
2656 fd = open(path, O_RDONLY);
2657 if (fd < 0)
2658 return 0;
2660 if (fstat(fd, &st))
2661 goto out;
2663 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2664 goto out;
2666 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2667 if (n != the_hash_algo->rawsz)
2668 goto out;
2670 if (!hasheq(istate->oid.hash, hash))
2671 goto out;
2673 close(fd);
2674 return 1;
2676 out:
2677 close(fd);
2678 return 0;
2681 static int repo_verify_index(struct repository *repo)
2683 return verify_index_from(repo->index, repo->index_file);
2686 static int has_racy_timestamp(struct index_state *istate)
2688 int entries = istate->cache_nr;
2689 int i;
2691 for (i = 0; i < entries; i++) {
2692 struct cache_entry *ce = istate->cache[i];
2693 if (is_racy_timestamp(istate, ce))
2694 return 1;
2696 return 0;
2699 void repo_update_index_if_able(struct repository *repo,
2700 struct lock_file *lockfile)
2702 if ((repo->index->cache_changed ||
2703 has_racy_timestamp(repo->index)) &&
2704 repo_verify_index(repo))
2705 write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2706 else
2707 rollback_lock_file(lockfile);
2710 static int record_eoie(void)
2712 int val;
2714 if (!git_config_get_bool("index.recordendofindexentries", &val))
2715 return val;
2718 * As a convenience, the end of index entries extension
2719 * used for threading is written by default if the user
2720 * explicitly requested threaded index reads.
2722 return !git_config_get_index_threads(&val) && val != 1;
2725 static int record_ieot(void)
2727 int val;
2729 if (!git_config_get_bool("index.recordoffsettable", &val))
2730 return val;
2733 * As a convenience, the offset table used for threading is
2734 * written by default if the user explicitly requested
2735 * threaded index reads.
2737 return !git_config_get_index_threads(&val) && val != 1;
2741 * On success, `tempfile` is closed. If it is the temporary file
2742 * of a `struct lock_file`, we will therefore effectively perform
2743 * a 'close_lock_file_gently()`. Since that is an implementation
2744 * detail of lockfiles, callers of `do_write_index()` should not
2745 * rely on it.
2747 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2748 int strip_extensions)
2750 uint64_t start = getnanotime();
2751 int newfd = tempfile->fd;
2752 git_hash_ctx c, eoie_c;
2753 struct cache_header hdr;
2754 int i, err = 0, removed, extended, hdr_version;
2755 struct cache_entry **cache = istate->cache;
2756 int entries = istate->cache_nr;
2757 struct stat st;
2758 struct ondisk_cache_entry_extended ondisk;
2759 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2760 int drop_cache_tree = istate->drop_cache_tree;
2761 off_t offset;
2762 int ieot_entries = 1;
2763 struct index_entry_offset_table *ieot = NULL;
2764 int nr, nr_threads;
2766 for (i = removed = extended = 0; i < entries; i++) {
2767 if (cache[i]->ce_flags & CE_REMOVE)
2768 removed++;
2770 /* reduce extended entries if possible */
2771 cache[i]->ce_flags &= ~CE_EXTENDED;
2772 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2773 extended++;
2774 cache[i]->ce_flags |= CE_EXTENDED;
2778 if (!istate->version) {
2779 istate->version = get_index_format_default();
2780 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2781 init_split_index(istate);
2784 /* demote version 3 to version 2 when the latter suffices */
2785 if (istate->version == 3 || istate->version == 2)
2786 istate->version = extended ? 3 : 2;
2788 hdr_version = istate->version;
2790 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2791 hdr.hdr_version = htonl(hdr_version);
2792 hdr.hdr_entries = htonl(entries - removed);
2794 the_hash_algo->init_fn(&c);
2795 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2796 return -1;
2798 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))
2799 nr_threads = 1;
2801 if (nr_threads != 1 && record_ieot()) {
2802 int ieot_blocks, cpus;
2805 * ensure default number of ieot blocks maps evenly to the
2806 * default number of threads that will process them leaving
2807 * room for the thread to load the index extensions.
2809 if (!nr_threads) {
2810 ieot_blocks = istate->cache_nr / THREAD_COST;
2811 cpus = online_cpus();
2812 if (ieot_blocks > cpus - 1)
2813 ieot_blocks = cpus - 1;
2814 } else {
2815 ieot_blocks = nr_threads;
2816 if (ieot_blocks > istate->cache_nr)
2817 ieot_blocks = istate->cache_nr;
2821 * no reason to write out the IEOT extension if we don't
2822 * have enough blocks to utilize multi-threading
2824 if (ieot_blocks > 1) {
2825 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2826 + (ieot_blocks * sizeof(struct index_entry_offset)));
2827 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2831 offset = lseek(newfd, 0, SEEK_CUR);
2832 if (offset < 0) {
2833 free(ieot);
2834 return -1;
2836 offset += write_buffer_len;
2837 nr = 0;
2838 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2840 for (i = 0; i < entries; i++) {
2841 struct cache_entry *ce = cache[i];
2842 if (ce->ce_flags & CE_REMOVE)
2843 continue;
2844 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2845 ce_smudge_racily_clean_entry(istate, ce);
2846 if (is_null_oid(&ce->oid)) {
2847 static const char msg[] = "cache entry has null sha1: %s";
2848 static int allow = -1;
2850 if (allow < 0)
2851 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2852 if (allow)
2853 warning(msg, ce->name);
2854 else
2855 err = error(msg, ce->name);
2857 drop_cache_tree = 1;
2859 if (ieot && i && (i % ieot_entries == 0)) {
2860 ieot->entries[ieot->nr].nr = nr;
2861 ieot->entries[ieot->nr].offset = offset;
2862 ieot->nr++;
2864 * If we have a V4 index, set the first byte to an invalid
2865 * character to ensure there is nothing common with the previous
2866 * entry
2868 if (previous_name)
2869 previous_name->buf[0] = 0;
2870 nr = 0;
2871 offset = lseek(newfd, 0, SEEK_CUR);
2872 if (offset < 0) {
2873 free(ieot);
2874 return -1;
2876 offset += write_buffer_len;
2878 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2879 err = -1;
2881 if (err)
2882 break;
2883 nr++;
2885 if (ieot && nr) {
2886 ieot->entries[ieot->nr].nr = nr;
2887 ieot->entries[ieot->nr].offset = offset;
2888 ieot->nr++;
2890 strbuf_release(&previous_name_buf);
2892 if (err) {
2893 free(ieot);
2894 return err;
2897 /* Write extension data here */
2898 offset = lseek(newfd, 0, SEEK_CUR);
2899 if (offset < 0) {
2900 free(ieot);
2901 return -1;
2903 offset += write_buffer_len;
2904 the_hash_algo->init_fn(&eoie_c);
2907 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2908 * can minimize the number of extensions we have to scan through to
2909 * find it during load. Write it out regardless of the
2910 * strip_extensions parameter as we need it when loading the shared
2911 * index.
2913 if (ieot) {
2914 struct strbuf sb = STRBUF_INIT;
2916 write_ieot_extension(&sb, ieot);
2917 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2918 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2919 strbuf_release(&sb);
2920 free(ieot);
2921 if (err)
2922 return -1;
2925 if (!strip_extensions && istate->split_index &&
2926 !is_null_oid(&istate->split_index->base_oid)) {
2927 struct strbuf sb = STRBUF_INIT;
2929 err = write_link_extension(&sb, istate) < 0 ||
2930 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2931 sb.len) < 0 ||
2932 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2933 strbuf_release(&sb);
2934 if (err)
2935 return -1;
2937 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2938 struct strbuf sb = STRBUF_INIT;
2940 cache_tree_write(&sb, istate->cache_tree);
2941 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2942 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2943 strbuf_release(&sb);
2944 if (err)
2945 return -1;
2947 if (!strip_extensions && istate->resolve_undo) {
2948 struct strbuf sb = STRBUF_INIT;
2950 resolve_undo_write(&sb, istate->resolve_undo);
2951 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2952 sb.len) < 0
2953 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2954 strbuf_release(&sb);
2955 if (err)
2956 return -1;
2958 if (!strip_extensions && istate->untracked) {
2959 struct strbuf sb = STRBUF_INIT;
2961 write_untracked_extension(&sb, istate->untracked);
2962 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2963 sb.len) < 0 ||
2964 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2965 strbuf_release(&sb);
2966 if (err)
2967 return -1;
2969 if (!strip_extensions && istate->fsmonitor_last_update) {
2970 struct strbuf sb = STRBUF_INIT;
2972 write_fsmonitor_extension(&sb, istate);
2973 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2974 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2975 strbuf_release(&sb);
2976 if (err)
2977 return -1;
2981 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2982 * so that it can be found and processed before all the index entries are
2983 * read. Write it out regardless of the strip_extensions parameter as we need it
2984 * when loading the shared index.
2986 if (offset && record_eoie()) {
2987 struct strbuf sb = STRBUF_INIT;
2989 write_eoie_extension(&sb, &eoie_c, offset);
2990 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
2991 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2992 strbuf_release(&sb);
2993 if (err)
2994 return -1;
2997 if (ce_flush(&c, newfd, istate->oid.hash))
2998 return -1;
2999 if (close_tempfile_gently(tempfile)) {
3000 error(_("could not close '%s'"), tempfile->filename.buf);
3001 return -1;
3003 if (stat(tempfile->filename.buf, &st))
3004 return -1;
3005 istate->timestamp.sec = (unsigned int)st.st_mtime;
3006 istate->timestamp.nsec = ST_MTIME_NSEC(st);
3007 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
3010 * TODO trace2: replace "the_repository" with the actual repo instance
3011 * that is associated with the given "istate".
3013 trace2_data_intmax("index", the_repository, "write/version",
3014 istate->version);
3015 trace2_data_intmax("index", the_repository, "write/cache_nr",
3016 istate->cache_nr);
3018 return 0;
3021 void set_alternate_index_output(const char *name)
3023 alternate_index_output = name;
3026 static int commit_locked_index(struct lock_file *lk)
3028 if (alternate_index_output)
3029 return commit_lock_file_to(lk, alternate_index_output);
3030 else
3031 return commit_lock_file(lk);
3034 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
3035 unsigned flags)
3037 int ret;
3040 * TODO trace2: replace "the_repository" with the actual repo instance
3041 * that is associated with the given "istate".
3043 trace2_region_enter_printf("index", "do_write_index", the_repository,
3044 "%s", lock->tempfile->filename.buf);
3045 ret = do_write_index(istate, lock->tempfile, 0);
3046 trace2_region_leave_printf("index", "do_write_index", the_repository,
3047 "%s", lock->tempfile->filename.buf);
3049 if (ret)
3050 return ret;
3051 if (flags & COMMIT_LOCK)
3052 return commit_locked_index(lock);
3053 return close_lock_file_gently(lock);
3056 static int write_split_index(struct index_state *istate,
3057 struct lock_file *lock,
3058 unsigned flags)
3060 int ret;
3061 prepare_to_write_split_index(istate);
3062 ret = do_write_locked_index(istate, lock, flags);
3063 finish_writing_split_index(istate);
3064 return ret;
3067 static const char *shared_index_expire = "2.weeks.ago";
3069 static unsigned long get_shared_index_expire_date(void)
3071 static unsigned long shared_index_expire_date;
3072 static int shared_index_expire_date_prepared;
3074 if (!shared_index_expire_date_prepared) {
3075 git_config_get_expiry("splitindex.sharedindexexpire",
3076 &shared_index_expire);
3077 shared_index_expire_date = approxidate(shared_index_expire);
3078 shared_index_expire_date_prepared = 1;
3081 return shared_index_expire_date;
3084 static int should_delete_shared_index(const char *shared_index_path)
3086 struct stat st;
3087 unsigned long expiration;
3089 /* Check timestamp */
3090 expiration = get_shared_index_expire_date();
3091 if (!expiration)
3092 return 0;
3093 if (stat(shared_index_path, &st))
3094 return error_errno(_("could not stat '%s'"), shared_index_path);
3095 if (st.st_mtime > expiration)
3096 return 0;
3098 return 1;
3101 static int clean_shared_index_files(const char *current_hex)
3103 struct dirent *de;
3104 DIR *dir = opendir(get_git_dir());
3106 if (!dir)
3107 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3109 while ((de = readdir(dir)) != NULL) {
3110 const char *sha1_hex;
3111 const char *shared_index_path;
3112 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3113 continue;
3114 if (!strcmp(sha1_hex, current_hex))
3115 continue;
3116 shared_index_path = git_path("%s", de->d_name);
3117 if (should_delete_shared_index(shared_index_path) > 0 &&
3118 unlink(shared_index_path))
3119 warning_errno(_("unable to unlink: %s"), shared_index_path);
3121 closedir(dir);
3123 return 0;
3126 static int write_shared_index(struct index_state *istate,
3127 struct tempfile **temp)
3129 struct split_index *si = istate->split_index;
3130 int ret;
3132 move_cache_to_base_index(istate);
3134 trace2_region_enter_printf("index", "shared/do_write_index",
3135 the_repository, "%s", (*temp)->filename.buf);
3136 ret = do_write_index(si->base, *temp, 1);
3137 trace2_region_enter_printf("index", "shared/do_write_index",
3138 the_repository, "%s", (*temp)->filename.buf);
3140 if (ret)
3141 return ret;
3142 ret = adjust_shared_perm(get_tempfile_path(*temp));
3143 if (ret) {
3144 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3145 return ret;
3147 ret = rename_tempfile(temp,
3148 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3149 if (!ret) {
3150 oidcpy(&si->base_oid, &si->base->oid);
3151 clean_shared_index_files(oid_to_hex(&si->base->oid));
3154 return ret;
3157 static const int default_max_percent_split_change = 20;
3159 static int too_many_not_shared_entries(struct index_state *istate)
3161 int i, not_shared = 0;
3162 int max_split = git_config_get_max_percent_split_change();
3164 switch (max_split) {
3165 case -1:
3166 /* not or badly configured: use the default value */
3167 max_split = default_max_percent_split_change;
3168 break;
3169 case 0:
3170 return 1; /* 0% means always write a new shared index */
3171 case 100:
3172 return 0; /* 100% means never write a new shared index */
3173 default:
3174 break; /* just use the configured value */
3177 /* Count not shared entries */
3178 for (i = 0; i < istate->cache_nr; i++) {
3179 struct cache_entry *ce = istate->cache[i];
3180 if (!ce->index)
3181 not_shared++;
3184 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3187 int write_locked_index(struct index_state *istate, struct lock_file *lock,
3188 unsigned flags)
3190 int new_shared_index, ret;
3191 struct split_index *si = istate->split_index;
3193 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3194 cache_tree_verify(the_repository, istate);
3196 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3197 if (flags & COMMIT_LOCK)
3198 rollback_lock_file(lock);
3199 return 0;
3202 if (istate->fsmonitor_last_update)
3203 fill_fsmonitor_bitmap(istate);
3205 if (!si || alternate_index_output ||
3206 (istate->cache_changed & ~EXTMASK)) {
3207 if (si)
3208 oidclr(&si->base_oid);
3209 ret = do_write_locked_index(istate, lock, flags);
3210 goto out;
3213 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3214 int v = si->base_oid.hash[0];
3215 if ((v & 15) < 6)
3216 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3218 if (too_many_not_shared_entries(istate))
3219 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3221 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3223 if (new_shared_index) {
3224 struct tempfile *temp;
3225 int saved_errno;
3227 /* Same initial permissions as the main .git/index file */
3228 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3229 if (!temp) {
3230 oidclr(&si->base_oid);
3231 ret = do_write_locked_index(istate, lock, flags);
3232 goto out;
3234 ret = write_shared_index(istate, &temp);
3236 saved_errno = errno;
3237 if (is_tempfile_active(temp))
3238 delete_tempfile(&temp);
3239 errno = saved_errno;
3241 if (ret)
3242 goto out;
3245 ret = write_split_index(istate, lock, flags);
3247 /* Freshen the shared index only if the split-index was written */
3248 if (!ret && !new_shared_index && !is_null_oid(&si->base_oid)) {
3249 const char *shared_index = git_path("sharedindex.%s",
3250 oid_to_hex(&si->base_oid));
3251 freshen_shared_index(shared_index, 1);
3254 out:
3255 if (flags & COMMIT_LOCK)
3256 rollback_lock_file(lock);
3257 return ret;
3261 * Read the index file that is potentially unmerged into given
3262 * index_state, dropping any unmerged entries to stage #0 (potentially
3263 * resulting in a path appearing as both a file and a directory in the
3264 * index; the caller is responsible to clear out the extra entries
3265 * before writing the index to a tree). Returns true if the index is
3266 * unmerged. Callers who want to refuse to work from an unmerged
3267 * state can call this and check its return value, instead of calling
3268 * read_cache().
3270 int repo_read_index_unmerged(struct repository *repo)
3272 struct index_state *istate;
3273 int i;
3274 int unmerged = 0;
3276 repo_read_index(repo);
3277 istate = repo->index;
3278 for (i = 0; i < istate->cache_nr; i++) {
3279 struct cache_entry *ce = istate->cache[i];
3280 struct cache_entry *new_ce;
3281 int len;
3283 if (!ce_stage(ce))
3284 continue;
3285 unmerged = 1;
3286 len = ce_namelen(ce);
3287 new_ce = make_empty_cache_entry(istate, len);
3288 memcpy(new_ce->name, ce->name, len);
3289 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3290 new_ce->ce_namelen = len;
3291 new_ce->ce_mode = ce->ce_mode;
3292 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3293 return error(_("%s: cannot drop to stage #0"),
3294 new_ce->name);
3296 return unmerged;
3300 * Returns 1 if the path is an "other" path with respect to
3301 * the index; that is, the path is not mentioned in the index at all,
3302 * either as a file, a directory with some files in the index,
3303 * or as an unmerged entry.
3305 * We helpfully remove a trailing "/" from directories so that
3306 * the output of read_directory can be used as-is.
3308 int index_name_is_other(const struct index_state *istate, const char *name,
3309 int namelen)
3311 int pos;
3312 if (namelen && name[namelen - 1] == '/')
3313 namelen--;
3314 pos = index_name_pos(istate, name, namelen);
3315 if (0 <= pos)
3316 return 0; /* exact match */
3317 pos = -pos - 1;
3318 if (pos < istate->cache_nr) {
3319 struct cache_entry *ce = istate->cache[pos];
3320 if (ce_namelen(ce) == namelen &&
3321 !memcmp(ce->name, name, namelen))
3322 return 0; /* Yup, this one exists unmerged */
3324 return 1;
3327 void *read_blob_data_from_index(const struct index_state *istate,
3328 const char *path, unsigned long *size)
3330 int pos, len;
3331 unsigned long sz;
3332 enum object_type type;
3333 void *data;
3335 len = strlen(path);
3336 pos = index_name_pos(istate, path, len);
3337 if (pos < 0) {
3339 * We might be in the middle of a merge, in which
3340 * case we would read stage #2 (ours).
3342 int i;
3343 for (i = -pos - 1;
3344 (pos < 0 && i < istate->cache_nr &&
3345 !strcmp(istate->cache[i]->name, path));
3346 i++)
3347 if (ce_stage(istate->cache[i]) == 2)
3348 pos = i;
3350 if (pos < 0)
3351 return NULL;
3352 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3353 if (!data || type != OBJ_BLOB) {
3354 free(data);
3355 return NULL;
3357 if (size)
3358 *size = sz;
3359 return data;
3362 void stat_validity_clear(struct stat_validity *sv)
3364 FREE_AND_NULL(sv->sd);
3367 int stat_validity_check(struct stat_validity *sv, const char *path)
3369 struct stat st;
3371 if (stat(path, &st) < 0)
3372 return sv->sd == NULL;
3373 if (!sv->sd)
3374 return 0;
3375 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3378 void stat_validity_update(struct stat_validity *sv, int fd)
3380 struct stat st;
3382 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3383 stat_validity_clear(sv);
3384 else {
3385 if (!sv->sd)
3386 sv->sd = xcalloc(1, sizeof(struct stat_data));
3387 fill_stat_data(sv->sd, &st);
3391 void move_index_extensions(struct index_state *dst, struct index_state *src)
3393 dst->untracked = src->untracked;
3394 src->untracked = NULL;
3395 dst->cache_tree = src->cache_tree;
3396 src->cache_tree = NULL;
3399 struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3400 struct index_state *istate)
3402 unsigned int size = ce_size(ce);
3403 int mem_pool_allocated;
3404 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3405 mem_pool_allocated = new_entry->mem_pool_allocated;
3407 memcpy(new_entry, ce, size);
3408 new_entry->mem_pool_allocated = mem_pool_allocated;
3409 return new_entry;
3412 void discard_cache_entry(struct cache_entry *ce)
3414 if (ce && should_validate_cache_entries())
3415 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3417 if (ce && ce->mem_pool_allocated)
3418 return;
3420 free(ce);
3423 int should_validate_cache_entries(void)
3425 static int validate_index_cache_entries = -1;
3427 if (validate_index_cache_entries < 0) {
3428 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3429 validate_index_cache_entries = 1;
3430 else
3431 validate_index_cache_entries = 0;
3434 return validate_index_cache_entries;
3437 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3438 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3440 static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3443 * The end of index entries (EOIE) extension is guaranteed to be last
3444 * so that it can be found by scanning backwards from the EOF.
3446 * "EOIE"
3447 * <4-byte length>
3448 * <4-byte offset>
3449 * <20-byte hash>
3451 const char *index, *eoie;
3452 uint32_t extsize;
3453 size_t offset, src_offset;
3454 unsigned char hash[GIT_MAX_RAWSZ];
3455 git_hash_ctx c;
3457 /* ensure we have an index big enough to contain an EOIE extension */
3458 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3459 return 0;
3461 /* validate the extension signature */
3462 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3463 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3464 return 0;
3465 index += sizeof(uint32_t);
3467 /* validate the extension size */
3468 extsize = get_be32(index);
3469 if (extsize != EOIE_SIZE)
3470 return 0;
3471 index += sizeof(uint32_t);
3474 * Validate the offset we're going to look for the first extension
3475 * signature is after the index header and before the eoie extension.
3477 offset = get_be32(index);
3478 if (mmap + offset < mmap + sizeof(struct cache_header))
3479 return 0;
3480 if (mmap + offset >= eoie)
3481 return 0;
3482 index += sizeof(uint32_t);
3485 * The hash is computed over extension types and their sizes (but not
3486 * their contents). E.g. if we have "TREE" extension that is N-bytes
3487 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3488 * then the hash would be:
3490 * SHA-1("TREE" + <binary representation of N> +
3491 * "REUC" + <binary representation of M>)
3493 src_offset = offset;
3494 the_hash_algo->init_fn(&c);
3495 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3496 /* After an array of active_nr index entries,
3497 * there can be arbitrary number of extended
3498 * sections, each of which is prefixed with
3499 * extension name (4-byte) and section length
3500 * in 4-byte network byte order.
3502 uint32_t extsize;
3503 memcpy(&extsize, mmap + src_offset + 4, 4);
3504 extsize = ntohl(extsize);
3506 /* verify the extension size isn't so large it will wrap around */
3507 if (src_offset + 8 + extsize < src_offset)
3508 return 0;
3510 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3512 src_offset += 8;
3513 src_offset += extsize;
3515 the_hash_algo->final_fn(hash, &c);
3516 if (!hasheq(hash, (const unsigned char *)index))
3517 return 0;
3519 /* Validate that the extension offsets returned us back to the eoie extension. */
3520 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3521 return 0;
3523 return offset;
3526 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3528 uint32_t buffer;
3529 unsigned char hash[GIT_MAX_RAWSZ];
3531 /* offset */
3532 put_be32(&buffer, offset);
3533 strbuf_add(sb, &buffer, sizeof(uint32_t));
3535 /* hash */
3536 the_hash_algo->final_fn(hash, eoie_context);
3537 strbuf_add(sb, hash, the_hash_algo->rawsz);
3540 #define IEOT_VERSION (1)
3542 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3544 const char *index = NULL;
3545 uint32_t extsize, ext_version;
3546 struct index_entry_offset_table *ieot;
3547 int i, nr;
3549 /* find the IEOT extension */
3550 if (!offset)
3551 return NULL;
3552 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3553 extsize = get_be32(mmap + offset + 4);
3554 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3555 index = mmap + offset + 4 + 4;
3556 break;
3558 offset += 8;
3559 offset += extsize;
3561 if (!index)
3562 return NULL;
3564 /* validate the version is IEOT_VERSION */
3565 ext_version = get_be32(index);
3566 if (ext_version != IEOT_VERSION) {
3567 error("invalid IEOT version %d", ext_version);
3568 return NULL;
3570 index += sizeof(uint32_t);
3572 /* extension size - version bytes / bytes per entry */
3573 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3574 if (!nr) {
3575 error("invalid number of IEOT entries %d", nr);
3576 return NULL;
3578 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3579 + (nr * sizeof(struct index_entry_offset)));
3580 ieot->nr = nr;
3581 for (i = 0; i < nr; i++) {
3582 ieot->entries[i].offset = get_be32(index);
3583 index += sizeof(uint32_t);
3584 ieot->entries[i].nr = get_be32(index);
3585 index += sizeof(uint32_t);
3588 return ieot;
3591 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3593 uint32_t buffer;
3594 int i;
3596 /* version */
3597 put_be32(&buffer, IEOT_VERSION);
3598 strbuf_add(sb, &buffer, sizeof(uint32_t));
3600 /* ieot */
3601 for (i = 0; i < ieot->nr; i++) {
3603 /* offset */
3604 put_be32(&buffer, ieot->entries[i].offset);
3605 strbuf_add(sb, &buffer, sizeof(uint32_t));
3607 /* count */
3608 put_be32(&buffer, ieot->entries[i].nr);
3609 strbuf_add(sb, &buffer, sizeof(uint32_t));