The first batch post 2.24 cycle
[git.git] / read-cache.c
blob133f790fa4cd52e503da5ae6be3f78389f7674b7
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #include "cache.h"
7 #include "config.h"
8 #include "diff.h"
9 #include "diffcore.h"
10 #include "tempfile.h"
11 #include "lockfile.h"
12 #include "cache-tree.h"
13 #include "refs.h"
14 #include "dir.h"
15 #include "object-store.h"
16 #include "tree.h"
17 #include "commit.h"
18 #include "blob.h"
19 #include "resolve-undo.h"
20 #include "run-command.h"
21 #include "strbuf.h"
22 #include "varint.h"
23 #include "split-index.h"
24 #include "utf8.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
27 #include "progress.h"
29 /* Mask for the name length in ce_flags in the on-disk index */
31 #define CE_NAMEMASK (0x0fff)
33 /* Index extensions.
35 * The first letter should be 'A'..'Z' for extensions that are not
36 * necessary for a correct operation (i.e. optimization data).
37 * When new extensions are added that _needs_ to be understood in
38 * order to correctly interpret the index file, pick character that
39 * is outside the range, to cause the reader to abort.
42 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
43 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
44 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
45 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
46 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
47 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
48 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
49 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
51 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
52 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
53 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
54 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
58 * This is an estimate of the pathname length in the index. We use
59 * this for V4 index files to guess the un-deltafied size of the index
60 * in memory because of pathname deltafication. This is not required
61 * for V2/V3 index formats because their pathnames are not compressed.
62 * If the initial amount of memory set aside is not sufficient, the
63 * mem pool will allocate extra memory.
65 #define CACHE_ENTRY_PATH_LENGTH 80
67 static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
69 struct cache_entry *ce;
70 ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
71 ce->mem_pool_allocated = 1;
72 return ce;
75 static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
77 struct cache_entry * ce;
78 ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
79 ce->mem_pool_allocated = 1;
80 return ce;
83 static struct mem_pool *find_mem_pool(struct index_state *istate)
85 struct mem_pool **pool_ptr;
87 if (istate->split_index && istate->split_index->base)
88 pool_ptr = &istate->split_index->base->ce_mem_pool;
89 else
90 pool_ptr = &istate->ce_mem_pool;
92 if (!*pool_ptr)
93 mem_pool_init(pool_ptr, 0);
95 return *pool_ptr;
98 static const char *alternate_index_output;
100 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
102 istate->cache[nr] = ce;
103 add_name_hash(istate, ce);
106 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
108 struct cache_entry *old = istate->cache[nr];
110 replace_index_entry_in_base(istate, old, ce);
111 remove_name_hash(istate, old);
112 discard_cache_entry(old);
113 ce->ce_flags &= ~CE_HASHED;
114 set_index_entry(istate, nr, ce);
115 ce->ce_flags |= CE_UPDATE_IN_BASE;
116 mark_fsmonitor_invalid(istate, ce);
117 istate->cache_changed |= CE_ENTRY_CHANGED;
120 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
122 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
123 int namelen = strlen(new_name);
125 new_entry = make_empty_cache_entry(istate, namelen);
126 copy_cache_entry(new_entry, old_entry);
127 new_entry->ce_flags &= ~CE_HASHED;
128 new_entry->ce_namelen = namelen;
129 new_entry->index = 0;
130 memcpy(new_entry->name, new_name, namelen + 1);
132 cache_tree_invalidate_path(istate, old_entry->name);
133 untracked_cache_remove_from_index(istate, old_entry->name);
134 remove_index_entry_at(istate, nr);
135 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
138 void fill_stat_data(struct stat_data *sd, struct stat *st)
140 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
141 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
142 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
143 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
144 sd->sd_dev = st->st_dev;
145 sd->sd_ino = st->st_ino;
146 sd->sd_uid = st->st_uid;
147 sd->sd_gid = st->st_gid;
148 sd->sd_size = st->st_size;
151 int match_stat_data(const struct stat_data *sd, struct stat *st)
153 int changed = 0;
155 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
156 changed |= MTIME_CHANGED;
157 if (trust_ctime && check_stat &&
158 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
159 changed |= CTIME_CHANGED;
161 #ifdef USE_NSEC
162 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
163 changed |= MTIME_CHANGED;
164 if (trust_ctime && check_stat &&
165 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
166 changed |= CTIME_CHANGED;
167 #endif
169 if (check_stat) {
170 if (sd->sd_uid != (unsigned int) st->st_uid ||
171 sd->sd_gid != (unsigned int) st->st_gid)
172 changed |= OWNER_CHANGED;
173 if (sd->sd_ino != (unsigned int) st->st_ino)
174 changed |= INODE_CHANGED;
177 #ifdef USE_STDEV
179 * st_dev breaks on network filesystems where different
180 * clients will have different views of what "device"
181 * the filesystem is on
183 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
184 changed |= INODE_CHANGED;
185 #endif
187 if (sd->sd_size != (unsigned int) st->st_size)
188 changed |= DATA_CHANGED;
190 return changed;
194 * This only updates the "non-critical" parts of the directory
195 * cache, ie the parts that aren't tracked by GIT, and only used
196 * to validate the cache.
198 void fill_stat_cache_info(struct index_state *istate, struct cache_entry *ce, struct stat *st)
200 fill_stat_data(&ce->ce_stat_data, st);
202 if (assume_unchanged)
203 ce->ce_flags |= CE_VALID;
205 if (S_ISREG(st->st_mode)) {
206 ce_mark_uptodate(ce);
207 mark_fsmonitor_valid(istate, ce);
211 static int ce_compare_data(struct index_state *istate,
212 const struct cache_entry *ce,
213 struct stat *st)
215 int match = -1;
216 int fd = git_open_cloexec(ce->name, O_RDONLY);
218 if (fd >= 0) {
219 struct object_id oid;
220 if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
221 match = !oideq(&oid, &ce->oid);
222 /* index_fd() closed the file descriptor already */
224 return match;
227 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
229 int match = -1;
230 void *buffer;
231 unsigned long size;
232 enum object_type type;
233 struct strbuf sb = STRBUF_INIT;
235 if (strbuf_readlink(&sb, ce->name, expected_size))
236 return -1;
238 buffer = read_object_file(&ce->oid, &type, &size);
239 if (buffer) {
240 if (size == sb.len)
241 match = memcmp(buffer, sb.buf, size);
242 free(buffer);
244 strbuf_release(&sb);
245 return match;
248 static int ce_compare_gitlink(const struct cache_entry *ce)
250 struct object_id oid;
253 * We don't actually require that the .git directory
254 * under GITLINK directory be a valid git directory. It
255 * might even be missing (in case nobody populated that
256 * sub-project).
258 * If so, we consider it always to match.
260 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
261 return 0;
262 return !oideq(&oid, &ce->oid);
265 static int ce_modified_check_fs(struct index_state *istate,
266 const struct cache_entry *ce,
267 struct stat *st)
269 switch (st->st_mode & S_IFMT) {
270 case S_IFREG:
271 if (ce_compare_data(istate, ce, st))
272 return DATA_CHANGED;
273 break;
274 case S_IFLNK:
275 if (ce_compare_link(ce, xsize_t(st->st_size)))
276 return DATA_CHANGED;
277 break;
278 case S_IFDIR:
279 if (S_ISGITLINK(ce->ce_mode))
280 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
281 /* else fallthrough */
282 default:
283 return TYPE_CHANGED;
285 return 0;
288 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
290 unsigned int changed = 0;
292 if (ce->ce_flags & CE_REMOVE)
293 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
295 switch (ce->ce_mode & S_IFMT) {
296 case S_IFREG:
297 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
298 /* We consider only the owner x bit to be relevant for
299 * "mode changes"
301 if (trust_executable_bit &&
302 (0100 & (ce->ce_mode ^ st->st_mode)))
303 changed |= MODE_CHANGED;
304 break;
305 case S_IFLNK:
306 if (!S_ISLNK(st->st_mode) &&
307 (has_symlinks || !S_ISREG(st->st_mode)))
308 changed |= TYPE_CHANGED;
309 break;
310 case S_IFGITLINK:
311 /* We ignore most of the st_xxx fields for gitlinks */
312 if (!S_ISDIR(st->st_mode))
313 changed |= TYPE_CHANGED;
314 else if (ce_compare_gitlink(ce))
315 changed |= DATA_CHANGED;
316 return changed;
317 default:
318 BUG("unsupported ce_mode: %o", ce->ce_mode);
321 changed |= match_stat_data(&ce->ce_stat_data, st);
323 /* Racily smudged entry? */
324 if (!ce->ce_stat_data.sd_size) {
325 if (!is_empty_blob_sha1(ce->oid.hash))
326 changed |= DATA_CHANGED;
329 return changed;
332 static int is_racy_stat(const struct index_state *istate,
333 const struct stat_data *sd)
335 return (istate->timestamp.sec &&
336 #ifdef USE_NSEC
337 /* nanosecond timestamped files can also be racy! */
338 (istate->timestamp.sec < sd->sd_mtime.sec ||
339 (istate->timestamp.sec == sd->sd_mtime.sec &&
340 istate->timestamp.nsec <= sd->sd_mtime.nsec))
341 #else
342 istate->timestamp.sec <= sd->sd_mtime.sec
343 #endif
347 int is_racy_timestamp(const struct index_state *istate,
348 const struct cache_entry *ce)
350 return (!S_ISGITLINK(ce->ce_mode) &&
351 is_racy_stat(istate, &ce->ce_stat_data));
354 int match_stat_data_racy(const struct index_state *istate,
355 const struct stat_data *sd, struct stat *st)
357 if (is_racy_stat(istate, sd))
358 return MTIME_CHANGED;
359 return match_stat_data(sd, st);
362 int ie_match_stat(struct index_state *istate,
363 const struct cache_entry *ce, struct stat *st,
364 unsigned int options)
366 unsigned int changed;
367 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
368 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
369 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
370 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
372 if (!ignore_fsmonitor)
373 refresh_fsmonitor(istate);
375 * If it's marked as always valid in the index, it's
376 * valid whatever the checked-out copy says.
378 * skip-worktree has the same effect with higher precedence
380 if (!ignore_skip_worktree && ce_skip_worktree(ce))
381 return 0;
382 if (!ignore_valid && (ce->ce_flags & CE_VALID))
383 return 0;
384 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
385 return 0;
388 * Intent-to-add entries have not been added, so the index entry
389 * by definition never matches what is in the work tree until it
390 * actually gets added.
392 if (ce_intent_to_add(ce))
393 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
395 changed = ce_match_stat_basic(ce, st);
398 * Within 1 second of this sequence:
399 * echo xyzzy >file && git-update-index --add file
400 * running this command:
401 * echo frotz >file
402 * would give a falsely clean cache entry. The mtime and
403 * length match the cache, and other stat fields do not change.
405 * We could detect this at update-index time (the cache entry
406 * being registered/updated records the same time as "now")
407 * and delay the return from git-update-index, but that would
408 * effectively mean we can make at most one commit per second,
409 * which is not acceptable. Instead, we check cache entries
410 * whose mtime are the same as the index file timestamp more
411 * carefully than others.
413 if (!changed && is_racy_timestamp(istate, ce)) {
414 if (assume_racy_is_modified)
415 changed |= DATA_CHANGED;
416 else
417 changed |= ce_modified_check_fs(istate, ce, st);
420 return changed;
423 int ie_modified(struct index_state *istate,
424 const struct cache_entry *ce,
425 struct stat *st, unsigned int options)
427 int changed, changed_fs;
429 changed = ie_match_stat(istate, ce, st, options);
430 if (!changed)
431 return 0;
433 * If the mode or type has changed, there's no point in trying
434 * to refresh the entry - it's not going to match
436 if (changed & (MODE_CHANGED | TYPE_CHANGED))
437 return changed;
440 * Immediately after read-tree or update-index --cacheinfo,
441 * the length field is zero, as we have never even read the
442 * lstat(2) information once, and we cannot trust DATA_CHANGED
443 * returned by ie_match_stat() which in turn was returned by
444 * ce_match_stat_basic() to signal that the filesize of the
445 * blob changed. We have to actually go to the filesystem to
446 * see if the contents match, and if so, should answer "unchanged".
448 * The logic does not apply to gitlinks, as ce_match_stat_basic()
449 * already has checked the actual HEAD from the filesystem in the
450 * subproject. If ie_match_stat() already said it is different,
451 * then we know it is.
453 if ((changed & DATA_CHANGED) &&
454 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
455 return changed;
457 changed_fs = ce_modified_check_fs(istate, ce, st);
458 if (changed_fs)
459 return changed | changed_fs;
460 return 0;
463 int base_name_compare(const char *name1, int len1, int mode1,
464 const char *name2, int len2, int mode2)
466 unsigned char c1, c2;
467 int len = len1 < len2 ? len1 : len2;
468 int cmp;
470 cmp = memcmp(name1, name2, len);
471 if (cmp)
472 return cmp;
473 c1 = name1[len];
474 c2 = name2[len];
475 if (!c1 && S_ISDIR(mode1))
476 c1 = '/';
477 if (!c2 && S_ISDIR(mode2))
478 c2 = '/';
479 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
483 * df_name_compare() is identical to base_name_compare(), except it
484 * compares conflicting directory/file entries as equal. Note that
485 * while a directory name compares as equal to a regular file, they
486 * then individually compare _differently_ to a filename that has
487 * a dot after the basename (because '\0' < '.' < '/').
489 * This is used by routines that want to traverse the git namespace
490 * but then handle conflicting entries together when possible.
492 int df_name_compare(const char *name1, int len1, int mode1,
493 const char *name2, int len2, int mode2)
495 int len = len1 < len2 ? len1 : len2, cmp;
496 unsigned char c1, c2;
498 cmp = memcmp(name1, name2, len);
499 if (cmp)
500 return cmp;
501 /* Directories and files compare equal (same length, same name) */
502 if (len1 == len2)
503 return 0;
504 c1 = name1[len];
505 if (!c1 && S_ISDIR(mode1))
506 c1 = '/';
507 c2 = name2[len];
508 if (!c2 && S_ISDIR(mode2))
509 c2 = '/';
510 if (c1 == '/' && !c2)
511 return 0;
512 if (c2 == '/' && !c1)
513 return 0;
514 return c1 - c2;
517 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
519 size_t min_len = (len1 < len2) ? len1 : len2;
520 int cmp = memcmp(name1, name2, min_len);
521 if (cmp)
522 return cmp;
523 if (len1 < len2)
524 return -1;
525 if (len1 > len2)
526 return 1;
527 return 0;
530 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
532 int cmp;
534 cmp = name_compare(name1, len1, name2, len2);
535 if (cmp)
536 return cmp;
538 if (stage1 < stage2)
539 return -1;
540 if (stage1 > stage2)
541 return 1;
542 return 0;
545 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
547 int first, last;
549 first = 0;
550 last = istate->cache_nr;
551 while (last > first) {
552 int next = first + ((last - first) >> 1);
553 struct cache_entry *ce = istate->cache[next];
554 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
555 if (!cmp)
556 return next;
557 if (cmp < 0) {
558 last = next;
559 continue;
561 first = next+1;
563 return -first-1;
566 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
568 return index_name_stage_pos(istate, name, namelen, 0);
571 int remove_index_entry_at(struct index_state *istate, int pos)
573 struct cache_entry *ce = istate->cache[pos];
575 record_resolve_undo(istate, ce);
576 remove_name_hash(istate, ce);
577 save_or_free_index_entry(istate, ce);
578 istate->cache_changed |= CE_ENTRY_REMOVED;
579 istate->cache_nr--;
580 if (pos >= istate->cache_nr)
581 return 0;
582 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
583 istate->cache_nr - pos);
584 return 1;
588 * Remove all cache entries marked for removal, that is where
589 * CE_REMOVE is set in ce_flags. This is much more effective than
590 * calling remove_index_entry_at() for each entry to be removed.
592 void remove_marked_cache_entries(struct index_state *istate, int invalidate)
594 struct cache_entry **ce_array = istate->cache;
595 unsigned int i, j;
597 for (i = j = 0; i < istate->cache_nr; i++) {
598 if (ce_array[i]->ce_flags & CE_REMOVE) {
599 if (invalidate) {
600 cache_tree_invalidate_path(istate,
601 ce_array[i]->name);
602 untracked_cache_remove_from_index(istate,
603 ce_array[i]->name);
605 remove_name_hash(istate, ce_array[i]);
606 save_or_free_index_entry(istate, ce_array[i]);
608 else
609 ce_array[j++] = ce_array[i];
611 if (j == istate->cache_nr)
612 return;
613 istate->cache_changed |= CE_ENTRY_REMOVED;
614 istate->cache_nr = j;
617 int remove_file_from_index(struct index_state *istate, const char *path)
619 int pos = index_name_pos(istate, path, strlen(path));
620 if (pos < 0)
621 pos = -pos-1;
622 cache_tree_invalidate_path(istate, path);
623 untracked_cache_remove_from_index(istate, path);
624 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
625 remove_index_entry_at(istate, pos);
626 return 0;
629 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
631 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
634 static int index_name_pos_also_unmerged(struct index_state *istate,
635 const char *path, int namelen)
637 int pos = index_name_pos(istate, path, namelen);
638 struct cache_entry *ce;
640 if (pos >= 0)
641 return pos;
643 /* maybe unmerged? */
644 pos = -1 - pos;
645 if (pos >= istate->cache_nr ||
646 compare_name((ce = istate->cache[pos]), path, namelen))
647 return -1;
649 /* order of preference: stage 2, 1, 3 */
650 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
651 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
652 !compare_name(ce, path, namelen))
653 pos++;
654 return pos;
657 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
659 int len = ce_namelen(ce);
660 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
664 * If we add a filename that aliases in the cache, we will use the
665 * name that we already have - but we don't want to update the same
666 * alias twice, because that implies that there were actually two
667 * different files with aliasing names!
669 * So we use the CE_ADDED flag to verify that the alias was an old
670 * one before we accept it as
672 static struct cache_entry *create_alias_ce(struct index_state *istate,
673 struct cache_entry *ce,
674 struct cache_entry *alias)
676 int len;
677 struct cache_entry *new_entry;
679 if (alias->ce_flags & CE_ADDED)
680 die(_("will not add file alias '%s' ('%s' already exists in index)"),
681 ce->name, alias->name);
683 /* Ok, create the new entry using the name of the existing alias */
684 len = ce_namelen(alias);
685 new_entry = make_empty_cache_entry(istate, len);
686 memcpy(new_entry->name, alias->name, len);
687 copy_cache_entry(new_entry, ce);
688 save_or_free_index_entry(istate, ce);
689 return new_entry;
692 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
694 struct object_id oid;
695 if (write_object_file("", 0, blob_type, &oid))
696 die(_("cannot create an empty blob in the object database"));
697 oidcpy(&ce->oid, &oid);
700 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
702 int namelen, was_same;
703 mode_t st_mode = st->st_mode;
704 struct cache_entry *ce, *alias = NULL;
705 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
706 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
707 int pretend = flags & ADD_CACHE_PRETEND;
708 int intent_only = flags & ADD_CACHE_INTENT;
709 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
710 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
711 int hash_flags = HASH_WRITE_OBJECT;
712 struct object_id oid;
714 if (flags & ADD_CACHE_RENORMALIZE)
715 hash_flags |= HASH_RENORMALIZE;
717 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
718 return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
720 namelen = strlen(path);
721 if (S_ISDIR(st_mode)) {
722 if (resolve_gitlink_ref(path, "HEAD", &oid) < 0)
723 return error(_("'%s' does not have a commit checked out"), path);
724 while (namelen && path[namelen-1] == '/')
725 namelen--;
727 ce = make_empty_cache_entry(istate, namelen);
728 memcpy(ce->name, path, namelen);
729 ce->ce_namelen = namelen;
730 if (!intent_only)
731 fill_stat_cache_info(istate, ce, st);
732 else
733 ce->ce_flags |= CE_INTENT_TO_ADD;
736 if (trust_executable_bit && has_symlinks) {
737 ce->ce_mode = create_ce_mode(st_mode);
738 } else {
739 /* If there is an existing entry, pick the mode bits and type
740 * from it, otherwise assume unexecutable regular file.
742 struct cache_entry *ent;
743 int pos = index_name_pos_also_unmerged(istate, path, namelen);
745 ent = (0 <= pos) ? istate->cache[pos] : NULL;
746 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
749 /* When core.ignorecase=true, determine if a directory of the same name but differing
750 * case already exists within the Git repository. If it does, ensure the directory
751 * case of the file being added to the repository matches (is folded into) the existing
752 * entry's directory case.
754 if (ignore_case) {
755 adjust_dirname_case(istate, ce->name);
757 if (!(flags & ADD_CACHE_RENORMALIZE)) {
758 alias = index_file_exists(istate, ce->name,
759 ce_namelen(ce), ignore_case);
760 if (alias &&
761 !ce_stage(alias) &&
762 !ie_match_stat(istate, alias, st, ce_option)) {
763 /* Nothing changed, really */
764 if (!S_ISGITLINK(alias->ce_mode))
765 ce_mark_uptodate(alias);
766 alias->ce_flags |= CE_ADDED;
768 discard_cache_entry(ce);
769 return 0;
772 if (!intent_only) {
773 if (index_path(istate, &ce->oid, path, st, hash_flags)) {
774 discard_cache_entry(ce);
775 return error(_("unable to index file '%s'"), path);
777 } else
778 set_object_name_for_intent_to_add_entry(ce);
780 if (ignore_case && alias && different_name(ce, alias))
781 ce = create_alias_ce(istate, ce, alias);
782 ce->ce_flags |= CE_ADDED;
784 /* It was suspected to be racily clean, but it turns out to be Ok */
785 was_same = (alias &&
786 !ce_stage(alias) &&
787 oideq(&alias->oid, &ce->oid) &&
788 ce->ce_mode == alias->ce_mode);
790 if (pretend)
791 discard_cache_entry(ce);
792 else if (add_index_entry(istate, ce, add_option)) {
793 discard_cache_entry(ce);
794 return error(_("unable to add '%s' to index"), path);
796 if (verbose && !was_same)
797 printf("add '%s'\n", path);
798 return 0;
801 int add_file_to_index(struct index_state *istate, const char *path, int flags)
803 struct stat st;
804 if (lstat(path, &st))
805 die_errno(_("unable to stat '%s'"), path);
806 return add_to_index(istate, path, &st, flags);
809 struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
811 return mem_pool__ce_calloc(find_mem_pool(istate), len);
814 struct cache_entry *make_empty_transient_cache_entry(size_t len)
816 return xcalloc(1, cache_entry_size(len));
819 struct cache_entry *make_cache_entry(struct index_state *istate,
820 unsigned int mode,
821 const struct object_id *oid,
822 const char *path,
823 int stage,
824 unsigned int refresh_options)
826 struct cache_entry *ce, *ret;
827 int len;
829 if (!verify_path(path, mode)) {
830 error(_("invalid path '%s'"), path);
831 return NULL;
834 len = strlen(path);
835 ce = make_empty_cache_entry(istate, len);
837 oidcpy(&ce->oid, oid);
838 memcpy(ce->name, path, len);
839 ce->ce_flags = create_ce_flags(stage);
840 ce->ce_namelen = len;
841 ce->ce_mode = create_ce_mode(mode);
843 ret = refresh_cache_entry(istate, ce, refresh_options);
844 if (ret != ce)
845 discard_cache_entry(ce);
846 return ret;
849 struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
850 const char *path, int stage)
852 struct cache_entry *ce;
853 int len;
855 if (!verify_path(path, mode)) {
856 error(_("invalid path '%s'"), path);
857 return NULL;
860 len = strlen(path);
861 ce = make_empty_transient_cache_entry(len);
863 oidcpy(&ce->oid, oid);
864 memcpy(ce->name, path, len);
865 ce->ce_flags = create_ce_flags(stage);
866 ce->ce_namelen = len;
867 ce->ce_mode = create_ce_mode(mode);
869 return ce;
873 * Chmod an index entry with either +x or -x.
875 * Returns -1 if the chmod for the particular cache entry failed (if it's
876 * not a regular file), -2 if an invalid flip argument is passed in, 0
877 * otherwise.
879 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
880 char flip)
882 if (!S_ISREG(ce->ce_mode))
883 return -1;
884 switch (flip) {
885 case '+':
886 ce->ce_mode |= 0111;
887 break;
888 case '-':
889 ce->ce_mode &= ~0111;
890 break;
891 default:
892 return -2;
894 cache_tree_invalidate_path(istate, ce->name);
895 ce->ce_flags |= CE_UPDATE_IN_BASE;
896 mark_fsmonitor_invalid(istate, ce);
897 istate->cache_changed |= CE_ENTRY_CHANGED;
899 return 0;
902 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
904 int len = ce_namelen(a);
905 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
909 * We fundamentally don't like some paths: we don't want
910 * dot or dot-dot anywhere, and for obvious reasons don't
911 * want to recurse into ".git" either.
913 * Also, we don't want double slashes or slashes at the
914 * end that can make pathnames ambiguous.
916 static int verify_dotfile(const char *rest, unsigned mode)
919 * The first character was '.', but that
920 * has already been discarded, we now test
921 * the rest.
924 /* "." is not allowed */
925 if (*rest == '\0' || is_dir_sep(*rest))
926 return 0;
928 switch (*rest) {
930 * ".git" followed by NUL or slash is bad. Note that we match
931 * case-insensitively here, even if ignore_case is not set.
932 * This outlaws ".GIT" everywhere out of an abundance of caution,
933 * since there's really no good reason to allow it.
935 * Once we've seen ".git", we can also find ".gitmodules", etc (also
936 * case-insensitively).
938 case 'g':
939 case 'G':
940 if (rest[1] != 'i' && rest[1] != 'I')
941 break;
942 if (rest[2] != 't' && rest[2] != 'T')
943 break;
944 if (rest[3] == '\0' || is_dir_sep(rest[3]))
945 return 0;
946 if (S_ISLNK(mode)) {
947 rest += 3;
948 if (skip_iprefix(rest, "modules", &rest) &&
949 (*rest == '\0' || is_dir_sep(*rest)))
950 return 0;
952 break;
953 case '.':
954 if (rest[1] == '\0' || is_dir_sep(rest[1]))
955 return 0;
957 return 1;
960 int verify_path(const char *path, unsigned mode)
962 char c;
964 if (has_dos_drive_prefix(path))
965 return 0;
967 goto inside;
968 for (;;) {
969 if (!c)
970 return 1;
971 if (is_dir_sep(c)) {
972 inside:
973 if (protect_hfs) {
974 if (is_hfs_dotgit(path))
975 return 0;
976 if (S_ISLNK(mode)) {
977 if (is_hfs_dotgitmodules(path))
978 return 0;
981 if (protect_ntfs) {
982 if (is_ntfs_dotgit(path))
983 return 0;
984 if (S_ISLNK(mode)) {
985 if (is_ntfs_dotgitmodules(path))
986 return 0;
990 c = *path++;
991 if ((c == '.' && !verify_dotfile(path, mode)) ||
992 is_dir_sep(c) || c == '\0')
993 return 0;
995 c = *path++;
1000 * Do we have another file that has the beginning components being a
1001 * proper superset of the name we're trying to add?
1003 static int has_file_name(struct index_state *istate,
1004 const struct cache_entry *ce, int pos, int ok_to_replace)
1006 int retval = 0;
1007 int len = ce_namelen(ce);
1008 int stage = ce_stage(ce);
1009 const char *name = ce->name;
1011 while (pos < istate->cache_nr) {
1012 struct cache_entry *p = istate->cache[pos++];
1014 if (len >= ce_namelen(p))
1015 break;
1016 if (memcmp(name, p->name, len))
1017 break;
1018 if (ce_stage(p) != stage)
1019 continue;
1020 if (p->name[len] != '/')
1021 continue;
1022 if (p->ce_flags & CE_REMOVE)
1023 continue;
1024 retval = -1;
1025 if (!ok_to_replace)
1026 break;
1027 remove_index_entry_at(istate, --pos);
1029 return retval;
1034 * Like strcmp(), but also return the offset of the first change.
1035 * If strings are equal, return the length.
1037 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1039 size_t k;
1041 if (!first_change)
1042 return strcmp(s1, s2);
1044 for (k = 0; s1[k] == s2[k]; k++)
1045 if (s1[k] == '\0')
1046 break;
1048 *first_change = k;
1049 return (unsigned char)s1[k] - (unsigned char)s2[k];
1053 * Do we have another file with a pathname that is a proper
1054 * subset of the name we're trying to add?
1056 * That is, is there another file in the index with a path
1057 * that matches a sub-directory in the given entry?
1059 static int has_dir_name(struct index_state *istate,
1060 const struct cache_entry *ce, int pos, int ok_to_replace)
1062 int retval = 0;
1063 int stage = ce_stage(ce);
1064 const char *name = ce->name;
1065 const char *slash = name + ce_namelen(ce);
1066 size_t len_eq_last;
1067 int cmp_last = 0;
1070 * We are frequently called during an iteration on a sorted
1071 * list of pathnames and while building a new index. Therefore,
1072 * there is a high probability that this entry will eventually
1073 * be appended to the index, rather than inserted in the middle.
1074 * If we can confirm that, we can avoid binary searches on the
1075 * components of the pathname.
1077 * Compare the entry's full path with the last path in the index.
1079 if (istate->cache_nr > 0) {
1080 cmp_last = strcmp_offset(name,
1081 istate->cache[istate->cache_nr - 1]->name,
1082 &len_eq_last);
1083 if (cmp_last > 0) {
1084 if (len_eq_last == 0) {
1086 * The entry sorts AFTER the last one in the
1087 * index and their paths have no common prefix,
1088 * so there cannot be a F/D conflict.
1090 return retval;
1091 } else {
1093 * The entry sorts AFTER the last one in the
1094 * index, but has a common prefix. Fall through
1095 * to the loop below to disect the entry's path
1096 * and see where the difference is.
1099 } else if (cmp_last == 0) {
1101 * The entry exactly matches the last one in the
1102 * index, but because of multiple stage and CE_REMOVE
1103 * items, we fall through and let the regular search
1104 * code handle it.
1109 for (;;) {
1110 size_t len;
1112 for (;;) {
1113 if (*--slash == '/')
1114 break;
1115 if (slash <= ce->name)
1116 return retval;
1118 len = slash - name;
1120 if (cmp_last > 0) {
1122 * (len + 1) is a directory boundary (including
1123 * the trailing slash). And since the loop is
1124 * decrementing "slash", the first iteration is
1125 * the longest directory prefix; subsequent
1126 * iterations consider parent directories.
1129 if (len + 1 <= len_eq_last) {
1131 * The directory prefix (including the trailing
1132 * slash) also appears as a prefix in the last
1133 * entry, so the remainder cannot collide (because
1134 * strcmp said the whole path was greater).
1136 * EQ: last: xxx/A
1137 * this: xxx/B
1139 * LT: last: xxx/file_A
1140 * this: xxx/file_B
1142 return retval;
1145 if (len > len_eq_last) {
1147 * This part of the directory prefix (excluding
1148 * the trailing slash) is longer than the known
1149 * equal portions, so this sub-directory cannot
1150 * collide with a file.
1152 * GT: last: xxxA
1153 * this: xxxB/file
1155 return retval;
1158 if (istate->cache_nr > 0 &&
1159 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1161 * The directory prefix lines up with part of
1162 * a longer file or directory name, but sorts
1163 * after it, so this sub-directory cannot
1164 * collide with a file.
1166 * last: xxx/yy-file (because '-' sorts before '/')
1167 * this: xxx/yy/abc
1169 return retval;
1173 * This is a possible collision. Fall through and
1174 * let the regular search code handle it.
1176 * last: xxx
1177 * this: xxx/file
1181 pos = index_name_stage_pos(istate, name, len, stage);
1182 if (pos >= 0) {
1184 * Found one, but not so fast. This could
1185 * be a marker that says "I was here, but
1186 * I am being removed". Such an entry is
1187 * not a part of the resulting tree, and
1188 * it is Ok to have a directory at the same
1189 * path.
1191 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1192 retval = -1;
1193 if (!ok_to_replace)
1194 break;
1195 remove_index_entry_at(istate, pos);
1196 continue;
1199 else
1200 pos = -pos-1;
1203 * Trivial optimization: if we find an entry that
1204 * already matches the sub-directory, then we know
1205 * we're ok, and we can exit.
1207 while (pos < istate->cache_nr) {
1208 struct cache_entry *p = istate->cache[pos];
1209 if ((ce_namelen(p) <= len) ||
1210 (p->name[len] != '/') ||
1211 memcmp(p->name, name, len))
1212 break; /* not our subdirectory */
1213 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1215 * p is at the same stage as our entry, and
1216 * is a subdirectory of what we are looking
1217 * at, so we cannot have conflicts at our
1218 * level or anything shorter.
1220 return retval;
1221 pos++;
1224 return retval;
1227 /* We may be in a situation where we already have path/file and path
1228 * is being added, or we already have path and path/file is being
1229 * added. Either one would result in a nonsense tree that has path
1230 * twice when git-write-tree tries to write it out. Prevent it.
1232 * If ok-to-replace is specified, we remove the conflicting entries
1233 * from the cache so the caller should recompute the insert position.
1234 * When this happens, we return non-zero.
1236 static int check_file_directory_conflict(struct index_state *istate,
1237 const struct cache_entry *ce,
1238 int pos, int ok_to_replace)
1240 int retval;
1243 * When ce is an "I am going away" entry, we allow it to be added
1245 if (ce->ce_flags & CE_REMOVE)
1246 return 0;
1249 * We check if the path is a sub-path of a subsequent pathname
1250 * first, since removing those will not change the position
1251 * in the array.
1253 retval = has_file_name(istate, ce, pos, ok_to_replace);
1256 * Then check if the path might have a clashing sub-directory
1257 * before it.
1259 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1262 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1264 int pos;
1265 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1266 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1267 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1268 int new_only = option & ADD_CACHE_NEW_ONLY;
1270 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1271 cache_tree_invalidate_path(istate, ce->name);
1274 * If this entry's path sorts after the last entry in the index,
1275 * we can avoid searching for it.
1277 if (istate->cache_nr > 0 &&
1278 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1279 pos = index_pos_to_insert_pos(istate->cache_nr);
1280 else
1281 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1283 /* existing match? Just replace it. */
1284 if (pos >= 0) {
1285 if (!new_only)
1286 replace_index_entry(istate, pos, ce);
1287 return 0;
1289 pos = -pos-1;
1291 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1292 untracked_cache_add_to_index(istate, ce->name);
1295 * Inserting a merged entry ("stage 0") into the index
1296 * will always replace all non-merged entries..
1298 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1299 while (ce_same_name(istate->cache[pos], ce)) {
1300 ok_to_add = 1;
1301 if (!remove_index_entry_at(istate, pos))
1302 break;
1306 if (!ok_to_add)
1307 return -1;
1308 if (!verify_path(ce->name, ce->ce_mode))
1309 return error(_("invalid path '%s'"), ce->name);
1311 if (!skip_df_check &&
1312 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1313 if (!ok_to_replace)
1314 return error(_("'%s' appears as both a file and as a directory"),
1315 ce->name);
1316 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1317 pos = -pos-1;
1319 return pos + 1;
1322 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1324 int pos;
1326 if (option & ADD_CACHE_JUST_APPEND)
1327 pos = istate->cache_nr;
1328 else {
1329 int ret;
1330 ret = add_index_entry_with_check(istate, ce, option);
1331 if (ret <= 0)
1332 return ret;
1333 pos = ret - 1;
1336 /* Make sure the array is big enough .. */
1337 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1339 /* Add it in.. */
1340 istate->cache_nr++;
1341 if (istate->cache_nr > pos + 1)
1342 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1343 istate->cache_nr - pos - 1);
1344 set_index_entry(istate, pos, ce);
1345 istate->cache_changed |= CE_ENTRY_ADDED;
1346 return 0;
1350 * "refresh" does not calculate a new sha1 file or bring the
1351 * cache up-to-date for mode/content changes. But what it
1352 * _does_ do is to "re-match" the stat information of a file
1353 * with the cache, so that you can refresh the cache for a
1354 * file that hasn't been changed but where the stat entry is
1355 * out of date.
1357 * For example, you'd want to do this after doing a "git-read-tree",
1358 * to link up the stat cache details with the proper files.
1360 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1361 struct cache_entry *ce,
1362 unsigned int options, int *err,
1363 int *changed_ret)
1365 struct stat st;
1366 struct cache_entry *updated;
1367 int changed;
1368 int refresh = options & CE_MATCH_REFRESH;
1369 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1370 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1371 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1372 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1374 if (!refresh || ce_uptodate(ce))
1375 return ce;
1377 if (!ignore_fsmonitor)
1378 refresh_fsmonitor(istate);
1380 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1381 * that the change to the work tree does not matter and told
1382 * us not to worry.
1384 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1385 ce_mark_uptodate(ce);
1386 return ce;
1388 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1389 ce_mark_uptodate(ce);
1390 return ce;
1392 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1393 ce_mark_uptodate(ce);
1394 return ce;
1397 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1398 if (ignore_missing)
1399 return ce;
1400 if (err)
1401 *err = ENOENT;
1402 return NULL;
1405 if (lstat(ce->name, &st) < 0) {
1406 if (ignore_missing && errno == ENOENT)
1407 return ce;
1408 if (err)
1409 *err = errno;
1410 return NULL;
1413 changed = ie_match_stat(istate, ce, &st, options);
1414 if (changed_ret)
1415 *changed_ret = changed;
1416 if (!changed) {
1418 * The path is unchanged. If we were told to ignore
1419 * valid bit, then we did the actual stat check and
1420 * found that the entry is unmodified. If the entry
1421 * is not marked VALID, this is the place to mark it
1422 * valid again, under "assume unchanged" mode.
1424 if (ignore_valid && assume_unchanged &&
1425 !(ce->ce_flags & CE_VALID))
1426 ; /* mark this one VALID again */
1427 else {
1429 * We do not mark the index itself "modified"
1430 * because CE_UPTODATE flag is in-core only;
1431 * we are not going to write this change out.
1433 if (!S_ISGITLINK(ce->ce_mode)) {
1434 ce_mark_uptodate(ce);
1435 mark_fsmonitor_valid(istate, ce);
1437 return ce;
1441 if (ie_modified(istate, ce, &st, options)) {
1442 if (err)
1443 *err = EINVAL;
1444 return NULL;
1447 updated = make_empty_cache_entry(istate, ce_namelen(ce));
1448 copy_cache_entry(updated, ce);
1449 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1450 fill_stat_cache_info(istate, updated, &st);
1452 * If ignore_valid is not set, we should leave CE_VALID bit
1453 * alone. Otherwise, paths marked with --no-assume-unchanged
1454 * (i.e. things to be edited) will reacquire CE_VALID bit
1455 * automatically, which is not really what we want.
1457 if (!ignore_valid && assume_unchanged &&
1458 !(ce->ce_flags & CE_VALID))
1459 updated->ce_flags &= ~CE_VALID;
1461 /* istate->cache_changed is updated in the caller */
1462 return updated;
1465 static void show_file(const char * fmt, const char * name, int in_porcelain,
1466 int * first, const char *header_msg)
1468 if (in_porcelain && *first && header_msg) {
1469 printf("%s\n", header_msg);
1470 *first = 0;
1472 printf(fmt, name);
1475 int repo_refresh_and_write_index(struct repository *repo,
1476 unsigned int refresh_flags,
1477 unsigned int write_flags,
1478 int gentle,
1479 const struct pathspec *pathspec,
1480 char *seen, const char *header_msg)
1482 struct lock_file lock_file = LOCK_INIT;
1483 int fd, ret = 0;
1485 fd = repo_hold_locked_index(repo, &lock_file, 0);
1486 if (!gentle && fd < 0)
1487 return -1;
1488 if (refresh_index(repo->index, refresh_flags, pathspec, seen, header_msg))
1489 ret = 1;
1490 if (0 <= fd && write_locked_index(repo->index, &lock_file, COMMIT_LOCK | write_flags))
1491 ret = -1;
1492 return ret;
1496 int refresh_index(struct index_state *istate, unsigned int flags,
1497 const struct pathspec *pathspec,
1498 char *seen, const char *header_msg)
1500 int i;
1501 int has_errors = 0;
1502 int really = (flags & REFRESH_REALLY) != 0;
1503 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1504 int quiet = (flags & REFRESH_QUIET) != 0;
1505 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1506 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1507 int first = 1;
1508 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1509 unsigned int options = (CE_MATCH_REFRESH |
1510 (really ? CE_MATCH_IGNORE_VALID : 0) |
1511 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1512 const char *modified_fmt;
1513 const char *deleted_fmt;
1514 const char *typechange_fmt;
1515 const char *added_fmt;
1516 const char *unmerged_fmt;
1517 struct progress *progress = NULL;
1519 if (flags & REFRESH_PROGRESS && isatty(2))
1520 progress = start_delayed_progress(_("Refresh index"),
1521 istate->cache_nr);
1523 trace_performance_enter();
1524 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1525 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1526 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1527 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1528 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1530 * Use the multi-threaded preload_index() to refresh most of the
1531 * cache entries quickly then in the single threaded loop below,
1532 * we only have to do the special cases that are left.
1534 preload_index(istate, pathspec, 0);
1535 for (i = 0; i < istate->cache_nr; i++) {
1536 struct cache_entry *ce, *new_entry;
1537 int cache_errno = 0;
1538 int changed = 0;
1539 int filtered = 0;
1541 ce = istate->cache[i];
1542 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1543 continue;
1545 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1546 filtered = 1;
1548 if (ce_stage(ce)) {
1549 while ((i < istate->cache_nr) &&
1550 ! strcmp(istate->cache[i]->name, ce->name))
1551 i++;
1552 i--;
1553 if (allow_unmerged)
1554 continue;
1555 if (!filtered)
1556 show_file(unmerged_fmt, ce->name, in_porcelain,
1557 &first, header_msg);
1558 has_errors = 1;
1559 continue;
1562 if (filtered)
1563 continue;
1565 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1566 if (new_entry == ce)
1567 continue;
1568 if (progress)
1569 display_progress(progress, i);
1570 if (!new_entry) {
1571 const char *fmt;
1573 if (really && cache_errno == EINVAL) {
1574 /* If we are doing --really-refresh that
1575 * means the index is not valid anymore.
1577 ce->ce_flags &= ~CE_VALID;
1578 ce->ce_flags |= CE_UPDATE_IN_BASE;
1579 mark_fsmonitor_invalid(istate, ce);
1580 istate->cache_changed |= CE_ENTRY_CHANGED;
1582 if (quiet)
1583 continue;
1585 if (cache_errno == ENOENT)
1586 fmt = deleted_fmt;
1587 else if (ce_intent_to_add(ce))
1588 fmt = added_fmt; /* must be before other checks */
1589 else if (changed & TYPE_CHANGED)
1590 fmt = typechange_fmt;
1591 else
1592 fmt = modified_fmt;
1593 show_file(fmt,
1594 ce->name, in_porcelain, &first, header_msg);
1595 has_errors = 1;
1596 continue;
1599 replace_index_entry(istate, i, new_entry);
1601 if (progress) {
1602 display_progress(progress, istate->cache_nr);
1603 stop_progress(&progress);
1605 trace_performance_leave("refresh index");
1606 return has_errors;
1609 struct cache_entry *refresh_cache_entry(struct index_state *istate,
1610 struct cache_entry *ce,
1611 unsigned int options)
1613 return refresh_cache_ent(istate, ce, options, NULL, NULL);
1617 /*****************************************************************
1618 * Index File I/O
1619 *****************************************************************/
1621 #define INDEX_FORMAT_DEFAULT 3
1623 static unsigned int get_index_format_default(struct repository *r)
1625 char *envversion = getenv("GIT_INDEX_VERSION");
1626 char *endp;
1627 unsigned int version = INDEX_FORMAT_DEFAULT;
1629 if (!envversion) {
1630 prepare_repo_settings(r);
1632 if (r->settings.index_version >= 0)
1633 version = r->settings.index_version;
1634 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1635 warning(_("index.version set, but the value is invalid.\n"
1636 "Using version %i"), INDEX_FORMAT_DEFAULT);
1637 return INDEX_FORMAT_DEFAULT;
1639 return version;
1642 version = strtoul(envversion, &endp, 10);
1643 if (*endp ||
1644 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1645 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1646 "Using version %i"), INDEX_FORMAT_DEFAULT);
1647 version = INDEX_FORMAT_DEFAULT;
1649 return version;
1653 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1654 * Again - this is just a (very strong in practice) heuristic that
1655 * the inode hasn't changed.
1657 * We save the fields in big-endian order to allow using the
1658 * index file over NFS transparently.
1660 struct ondisk_cache_entry {
1661 struct cache_time ctime;
1662 struct cache_time mtime;
1663 uint32_t dev;
1664 uint32_t ino;
1665 uint32_t mode;
1666 uint32_t uid;
1667 uint32_t gid;
1668 uint32_t size;
1670 * unsigned char hash[hashsz];
1671 * uint16_t flags;
1672 * if (flags & CE_EXTENDED)
1673 * uint16_t flags2;
1675 unsigned char data[GIT_MAX_RAWSZ + 2 * sizeof(uint16_t)];
1676 char name[FLEX_ARRAY];
1679 /* These are only used for v3 or lower */
1680 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1681 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1682 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1683 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1684 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1685 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1686 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1688 /* Allow fsck to force verification of the index checksum. */
1689 int verify_index_checksum;
1691 /* Allow fsck to force verification of the cache entry order. */
1692 int verify_ce_order;
1694 static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1696 git_hash_ctx c;
1697 unsigned char hash[GIT_MAX_RAWSZ];
1698 int hdr_version;
1700 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1701 return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1702 hdr_version = ntohl(hdr->hdr_version);
1703 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1704 return error(_("bad index version %d"), hdr_version);
1706 if (!verify_index_checksum)
1707 return 0;
1709 the_hash_algo->init_fn(&c);
1710 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1711 the_hash_algo->final_fn(hash, &c);
1712 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1713 return error(_("bad index file sha1 signature"));
1714 return 0;
1717 static int read_index_extension(struct index_state *istate,
1718 const char *ext, const char *data, unsigned long sz)
1720 switch (CACHE_EXT(ext)) {
1721 case CACHE_EXT_TREE:
1722 istate->cache_tree = cache_tree_read(data, sz);
1723 break;
1724 case CACHE_EXT_RESOLVE_UNDO:
1725 istate->resolve_undo = resolve_undo_read(data, sz);
1726 break;
1727 case CACHE_EXT_LINK:
1728 if (read_link_extension(istate, data, sz))
1729 return -1;
1730 break;
1731 case CACHE_EXT_UNTRACKED:
1732 istate->untracked = read_untracked_extension(data, sz);
1733 break;
1734 case CACHE_EXT_FSMONITOR:
1735 read_fsmonitor_extension(istate, data, sz);
1736 break;
1737 case CACHE_EXT_ENDOFINDEXENTRIES:
1738 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1739 /* already handled in do_read_index() */
1740 break;
1741 default:
1742 if (*ext < 'A' || 'Z' < *ext)
1743 return error(_("index uses %.4s extension, which we do not understand"),
1744 ext);
1745 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1746 break;
1748 return 0;
1751 static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1752 unsigned int version,
1753 struct ondisk_cache_entry *ondisk,
1754 unsigned long *ent_size,
1755 const struct cache_entry *previous_ce)
1757 struct cache_entry *ce;
1758 size_t len;
1759 const char *name;
1760 const unsigned hashsz = the_hash_algo->rawsz;
1761 const uint16_t *flagsp = (const uint16_t *)(ondisk->data + hashsz);
1762 unsigned int flags;
1763 size_t copy_len = 0;
1765 * Adjacent cache entries tend to share the leading paths, so it makes
1766 * sense to only store the differences in later entries. In the v4
1767 * on-disk format of the index, each on-disk cache entry stores the
1768 * number of bytes to be stripped from the end of the previous name,
1769 * and the bytes to append to the result, to come up with its name.
1771 int expand_name_field = version == 4;
1773 /* On-disk flags are just 16 bits */
1774 flags = get_be16(flagsp);
1775 len = flags & CE_NAMEMASK;
1777 if (flags & CE_EXTENDED) {
1778 int extended_flags;
1779 extended_flags = get_be16(flagsp + 1) << 16;
1780 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1781 if (extended_flags & ~CE_EXTENDED_FLAGS)
1782 die(_("unknown index entry format 0x%08x"), extended_flags);
1783 flags |= extended_flags;
1784 name = (const char *)(flagsp + 2);
1786 else
1787 name = (const char *)(flagsp + 1);
1789 if (expand_name_field) {
1790 const unsigned char *cp = (const unsigned char *)name;
1791 size_t strip_len, previous_len;
1793 /* If we're at the begining of a block, ignore the previous name */
1794 strip_len = decode_varint(&cp);
1795 if (previous_ce) {
1796 previous_len = previous_ce->ce_namelen;
1797 if (previous_len < strip_len)
1798 die(_("malformed name field in the index, near path '%s'"),
1799 previous_ce->name);
1800 copy_len = previous_len - strip_len;
1802 name = (const char *)cp;
1805 if (len == CE_NAMEMASK) {
1806 len = strlen(name);
1807 if (expand_name_field)
1808 len += copy_len;
1811 ce = mem_pool__ce_alloc(ce_mem_pool, len);
1813 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1814 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1815 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1816 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1817 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1818 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1819 ce->ce_mode = get_be32(&ondisk->mode);
1820 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1821 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1822 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1823 ce->ce_flags = flags & ~CE_NAMEMASK;
1824 ce->ce_namelen = len;
1825 ce->index = 0;
1826 hashcpy(ce->oid.hash, ondisk->data);
1827 memcpy(ce->name, name, len);
1828 ce->name[len] = '\0';
1830 if (expand_name_field) {
1831 if (copy_len)
1832 memcpy(ce->name, previous_ce->name, copy_len);
1833 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1834 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1835 } else {
1836 memcpy(ce->name, name, len + 1);
1837 *ent_size = ondisk_ce_size(ce);
1839 return ce;
1842 static void check_ce_order(struct index_state *istate)
1844 unsigned int i;
1846 if (!verify_ce_order)
1847 return;
1849 for (i = 1; i < istate->cache_nr; i++) {
1850 struct cache_entry *ce = istate->cache[i - 1];
1851 struct cache_entry *next_ce = istate->cache[i];
1852 int name_compare = strcmp(ce->name, next_ce->name);
1854 if (0 < name_compare)
1855 die(_("unordered stage entries in index"));
1856 if (!name_compare) {
1857 if (!ce_stage(ce))
1858 die(_("multiple stage entries for merged file '%s'"),
1859 ce->name);
1860 if (ce_stage(ce) > ce_stage(next_ce))
1861 die(_("unordered stage entries for '%s'"),
1862 ce->name);
1867 static void tweak_untracked_cache(struct index_state *istate)
1869 struct repository *r = the_repository;
1871 prepare_repo_settings(r);
1873 if (r->settings.core_untracked_cache == UNTRACKED_CACHE_REMOVE) {
1874 remove_untracked_cache(istate);
1875 return;
1878 if (r->settings.core_untracked_cache == UNTRACKED_CACHE_WRITE)
1879 add_untracked_cache(istate);
1882 static void tweak_split_index(struct index_state *istate)
1884 switch (git_config_get_split_index()) {
1885 case -1: /* unset: do nothing */
1886 break;
1887 case 0: /* false */
1888 remove_split_index(istate);
1889 break;
1890 case 1: /* true */
1891 add_split_index(istate);
1892 break;
1893 default: /* unknown value: do nothing */
1894 break;
1898 static void post_read_index_from(struct index_state *istate)
1900 check_ce_order(istate);
1901 tweak_untracked_cache(istate);
1902 tweak_split_index(istate);
1903 tweak_fsmonitor(istate);
1906 static size_t estimate_cache_size_from_compressed(unsigned int entries)
1908 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1911 static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1913 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1916 * Account for potential alignment differences.
1918 per_entry += align_padding_size(per_entry, 0);
1919 return ondisk_size + entries * per_entry;
1922 struct index_entry_offset
1924 /* starting byte offset into index file, count of index entries in this block */
1925 int offset, nr;
1928 struct index_entry_offset_table
1930 int nr;
1931 struct index_entry_offset entries[FLEX_ARRAY];
1934 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1935 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1937 static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1938 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1940 struct load_index_extensions
1942 pthread_t pthread;
1943 struct index_state *istate;
1944 const char *mmap;
1945 size_t mmap_size;
1946 unsigned long src_offset;
1949 static void *load_index_extensions(void *_data)
1951 struct load_index_extensions *p = _data;
1952 unsigned long src_offset = p->src_offset;
1954 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1955 /* After an array of active_nr index entries,
1956 * there can be arbitrary number of extended
1957 * sections, each of which is prefixed with
1958 * extension name (4-byte) and section length
1959 * in 4-byte network byte order.
1961 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1962 if (read_index_extension(p->istate,
1963 p->mmap + src_offset,
1964 p->mmap + src_offset + 8,
1965 extsize) < 0) {
1966 munmap((void *)p->mmap, p->mmap_size);
1967 die(_("index file corrupt"));
1969 src_offset += 8;
1970 src_offset += extsize;
1973 return NULL;
1977 * A helper function that will load the specified range of cache entries
1978 * from the memory mapped file and add them to the given index.
1980 static unsigned long load_cache_entry_block(struct index_state *istate,
1981 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1982 unsigned long start_offset, const struct cache_entry *previous_ce)
1984 int i;
1985 unsigned long src_offset = start_offset;
1987 for (i = offset; i < offset + nr; i++) {
1988 struct ondisk_cache_entry *disk_ce;
1989 struct cache_entry *ce;
1990 unsigned long consumed;
1992 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1993 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1994 set_index_entry(istate, i, ce);
1996 src_offset += consumed;
1997 previous_ce = ce;
1999 return src_offset - start_offset;
2002 static unsigned long load_all_cache_entries(struct index_state *istate,
2003 const char *mmap, size_t mmap_size, unsigned long src_offset)
2005 unsigned long consumed;
2007 if (istate->version == 4) {
2008 mem_pool_init(&istate->ce_mem_pool,
2009 estimate_cache_size_from_compressed(istate->cache_nr));
2010 } else {
2011 mem_pool_init(&istate->ce_mem_pool,
2012 estimate_cache_size(mmap_size, istate->cache_nr));
2015 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2016 0, istate->cache_nr, mmap, src_offset, NULL);
2017 return consumed;
2021 * Mostly randomly chosen maximum thread counts: we
2022 * cap the parallelism to online_cpus() threads, and we want
2023 * to have at least 10000 cache entries per thread for it to
2024 * be worth starting a thread.
2027 #define THREAD_COST (10000)
2029 struct load_cache_entries_thread_data
2031 pthread_t pthread;
2032 struct index_state *istate;
2033 struct mem_pool *ce_mem_pool;
2034 int offset;
2035 const char *mmap;
2036 struct index_entry_offset_table *ieot;
2037 int ieot_start; /* starting index into the ieot array */
2038 int ieot_blocks; /* count of ieot entries to process */
2039 unsigned long consumed; /* return # of bytes in index file processed */
2043 * A thread proc to run the load_cache_entries() computation
2044 * across multiple background threads.
2046 static void *load_cache_entries_thread(void *_data)
2048 struct load_cache_entries_thread_data *p = _data;
2049 int i;
2051 /* iterate across all ieot blocks assigned to this thread */
2052 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2053 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2054 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2055 p->offset += p->ieot->entries[i].nr;
2057 return NULL;
2060 static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2061 int nr_threads, struct index_entry_offset_table *ieot)
2063 int i, offset, ieot_blocks, ieot_start, err;
2064 struct load_cache_entries_thread_data *data;
2065 unsigned long consumed = 0;
2067 /* a little sanity checking */
2068 if (istate->name_hash_initialized)
2069 BUG("the name hash isn't thread safe");
2071 mem_pool_init(&istate->ce_mem_pool, 0);
2073 /* ensure we have no more threads than we have blocks to process */
2074 if (nr_threads > ieot->nr)
2075 nr_threads = ieot->nr;
2076 data = xcalloc(nr_threads, sizeof(*data));
2078 offset = ieot_start = 0;
2079 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2080 for (i = 0; i < nr_threads; i++) {
2081 struct load_cache_entries_thread_data *p = &data[i];
2082 int nr, j;
2084 if (ieot_start + ieot_blocks > ieot->nr)
2085 ieot_blocks = ieot->nr - ieot_start;
2087 p->istate = istate;
2088 p->offset = offset;
2089 p->mmap = mmap;
2090 p->ieot = ieot;
2091 p->ieot_start = ieot_start;
2092 p->ieot_blocks = ieot_blocks;
2094 /* create a mem_pool for each thread */
2095 nr = 0;
2096 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2097 nr += p->ieot->entries[j].nr;
2098 if (istate->version == 4) {
2099 mem_pool_init(&p->ce_mem_pool,
2100 estimate_cache_size_from_compressed(nr));
2101 } else {
2102 mem_pool_init(&p->ce_mem_pool,
2103 estimate_cache_size(mmap_size, nr));
2106 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2107 if (err)
2108 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2110 /* increment by the number of cache entries in the ieot block being processed */
2111 for (j = 0; j < ieot_blocks; j++)
2112 offset += ieot->entries[ieot_start + j].nr;
2113 ieot_start += ieot_blocks;
2116 for (i = 0; i < nr_threads; i++) {
2117 struct load_cache_entries_thread_data *p = &data[i];
2119 err = pthread_join(p->pthread, NULL);
2120 if (err)
2121 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2122 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2123 consumed += p->consumed;
2126 free(data);
2128 return consumed;
2131 /* remember to discard_cache() before reading a different cache! */
2132 int do_read_index(struct index_state *istate, const char *path, int must_exist)
2134 int fd;
2135 struct stat st;
2136 unsigned long src_offset;
2137 const struct cache_header *hdr;
2138 const char *mmap;
2139 size_t mmap_size;
2140 struct load_index_extensions p;
2141 size_t extension_offset = 0;
2142 int nr_threads, cpus;
2143 struct index_entry_offset_table *ieot = NULL;
2145 if (istate->initialized)
2146 return istate->cache_nr;
2148 istate->timestamp.sec = 0;
2149 istate->timestamp.nsec = 0;
2150 fd = open(path, O_RDONLY);
2151 if (fd < 0) {
2152 if (!must_exist && errno == ENOENT)
2153 return 0;
2154 die_errno(_("%s: index file open failed"), path);
2157 if (fstat(fd, &st))
2158 die_errno(_("%s: cannot stat the open index"), path);
2160 mmap_size = xsize_t(st.st_size);
2161 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2162 die(_("%s: index file smaller than expected"), path);
2164 mmap = xmmap_gently(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2165 if (mmap == MAP_FAILED)
2166 die_errno(_("%s: unable to map index file"), path);
2167 close(fd);
2169 hdr = (const struct cache_header *)mmap;
2170 if (verify_hdr(hdr, mmap_size) < 0)
2171 goto unmap;
2173 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2174 istate->version = ntohl(hdr->hdr_version);
2175 istate->cache_nr = ntohl(hdr->hdr_entries);
2176 istate->cache_alloc = alloc_nr(istate->cache_nr);
2177 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2178 istate->initialized = 1;
2180 p.istate = istate;
2181 p.mmap = mmap;
2182 p.mmap_size = mmap_size;
2184 src_offset = sizeof(*hdr);
2186 if (git_config_get_index_threads(&nr_threads))
2187 nr_threads = 1;
2189 /* TODO: does creating more threads than cores help? */
2190 if (!nr_threads) {
2191 nr_threads = istate->cache_nr / THREAD_COST;
2192 cpus = online_cpus();
2193 if (nr_threads > cpus)
2194 nr_threads = cpus;
2197 if (!HAVE_THREADS)
2198 nr_threads = 1;
2200 if (nr_threads > 1) {
2201 extension_offset = read_eoie_extension(mmap, mmap_size);
2202 if (extension_offset) {
2203 int err;
2205 p.src_offset = extension_offset;
2206 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2207 if (err)
2208 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2210 nr_threads--;
2215 * Locate and read the index entry offset table so that we can use it
2216 * to multi-thread the reading of the cache entries.
2218 if (extension_offset && nr_threads > 1)
2219 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2221 if (ieot) {
2222 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, nr_threads, ieot);
2223 free(ieot);
2224 } else {
2225 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2228 istate->timestamp.sec = st.st_mtime;
2229 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2231 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2232 if (extension_offset) {
2233 int ret = pthread_join(p.pthread, NULL);
2234 if (ret)
2235 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2236 } else {
2237 p.src_offset = src_offset;
2238 load_index_extensions(&p);
2240 munmap((void *)mmap, mmap_size);
2243 * TODO trace2: replace "the_repository" with the actual repo instance
2244 * that is associated with the given "istate".
2246 trace2_data_intmax("index", the_repository, "read/version",
2247 istate->version);
2248 trace2_data_intmax("index", the_repository, "read/cache_nr",
2249 istate->cache_nr);
2251 return istate->cache_nr;
2253 unmap:
2254 munmap((void *)mmap, mmap_size);
2255 die(_("index file corrupt"));
2259 * Signal that the shared index is used by updating its mtime.
2261 * This way, shared index can be removed if they have not been used
2262 * for some time.
2264 static void freshen_shared_index(const char *shared_index, int warn)
2266 if (!check_and_freshen_file(shared_index, 1) && warn)
2267 warning(_("could not freshen shared index '%s'"), shared_index);
2270 int read_index_from(struct index_state *istate, const char *path,
2271 const char *gitdir)
2273 struct split_index *split_index;
2274 int ret;
2275 char *base_oid_hex;
2276 char *base_path;
2278 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2279 if (istate->initialized)
2280 return istate->cache_nr;
2283 * TODO trace2: replace "the_repository" with the actual repo instance
2284 * that is associated with the given "istate".
2286 trace2_region_enter_printf("index", "do_read_index", the_repository,
2287 "%s", path);
2288 trace_performance_enter();
2289 ret = do_read_index(istate, path, 0);
2290 trace_performance_leave("read cache %s", path);
2291 trace2_region_leave_printf("index", "do_read_index", the_repository,
2292 "%s", path);
2294 split_index = istate->split_index;
2295 if (!split_index || is_null_oid(&split_index->base_oid)) {
2296 post_read_index_from(istate);
2297 return ret;
2300 trace_performance_enter();
2301 if (split_index->base)
2302 discard_index(split_index->base);
2303 else
2304 split_index->base = xcalloc(1, sizeof(*split_index->base));
2306 base_oid_hex = oid_to_hex(&split_index->base_oid);
2307 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2308 trace2_region_enter_printf("index", "shared/do_read_index",
2309 the_repository, "%s", base_path);
2310 ret = do_read_index(split_index->base, base_path, 1);
2311 trace2_region_leave_printf("index", "shared/do_read_index",
2312 the_repository, "%s", base_path);
2313 if (!oideq(&split_index->base_oid, &split_index->base->oid))
2314 die(_("broken index, expect %s in %s, got %s"),
2315 base_oid_hex, base_path,
2316 oid_to_hex(&split_index->base->oid));
2318 freshen_shared_index(base_path, 0);
2319 merge_base_index(istate);
2320 post_read_index_from(istate);
2321 trace_performance_leave("read cache %s", base_path);
2322 free(base_path);
2323 return ret;
2326 int is_index_unborn(struct index_state *istate)
2328 return (!istate->cache_nr && !istate->timestamp.sec);
2331 int discard_index(struct index_state *istate)
2334 * Cache entries in istate->cache[] should have been allocated
2335 * from the memory pool associated with this index, or from an
2336 * associated split_index. There is no need to free individual
2337 * cache entries. validate_cache_entries can detect when this
2338 * assertion does not hold.
2340 validate_cache_entries(istate);
2342 resolve_undo_clear_index(istate);
2343 istate->cache_nr = 0;
2344 istate->cache_changed = 0;
2345 istate->timestamp.sec = 0;
2346 istate->timestamp.nsec = 0;
2347 free_name_hash(istate);
2348 cache_tree_free(&(istate->cache_tree));
2349 istate->initialized = 0;
2350 istate->fsmonitor_has_run_once = 0;
2351 FREE_AND_NULL(istate->cache);
2352 istate->cache_alloc = 0;
2353 discard_split_index(istate);
2354 free_untracked_cache(istate->untracked);
2355 istate->untracked = NULL;
2357 if (istate->ce_mem_pool) {
2358 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2359 istate->ce_mem_pool = NULL;
2362 return 0;
2366 * Validate the cache entries of this index.
2367 * All cache entries associated with this index
2368 * should have been allocated by the memory pool
2369 * associated with this index, or by a referenced
2370 * split index.
2372 void validate_cache_entries(const struct index_state *istate)
2374 int i;
2376 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2377 return;
2379 for (i = 0; i < istate->cache_nr; i++) {
2380 if (!istate) {
2381 BUG("cache entry is not allocated from expected memory pool");
2382 } else if (!istate->ce_mem_pool ||
2383 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2384 if (!istate->split_index ||
2385 !istate->split_index->base ||
2386 !istate->split_index->base->ce_mem_pool ||
2387 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2388 BUG("cache entry is not allocated from expected memory pool");
2393 if (istate->split_index)
2394 validate_cache_entries(istate->split_index->base);
2397 int unmerged_index(const struct index_state *istate)
2399 int i;
2400 for (i = 0; i < istate->cache_nr; i++) {
2401 if (ce_stage(istate->cache[i]))
2402 return 1;
2404 return 0;
2407 int repo_index_has_changes(struct repository *repo,
2408 struct tree *tree,
2409 struct strbuf *sb)
2411 struct index_state *istate = repo->index;
2412 struct object_id cmp;
2413 int i;
2415 if (tree)
2416 cmp = tree->object.oid;
2417 if (tree || !get_oid_tree("HEAD", &cmp)) {
2418 struct diff_options opt;
2420 repo_diff_setup(repo, &opt);
2421 opt.flags.exit_with_status = 1;
2422 if (!sb)
2423 opt.flags.quick = 1;
2424 do_diff_cache(&cmp, &opt);
2425 diffcore_std(&opt);
2426 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2427 if (i)
2428 strbuf_addch(sb, ' ');
2429 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2431 diff_flush(&opt);
2432 return opt.flags.has_changes != 0;
2433 } else {
2434 for (i = 0; sb && i < istate->cache_nr; i++) {
2435 if (i)
2436 strbuf_addch(sb, ' ');
2437 strbuf_addstr(sb, istate->cache[i]->name);
2439 return !!istate->cache_nr;
2443 #define WRITE_BUFFER_SIZE 8192
2444 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2445 static unsigned long write_buffer_len;
2447 static int ce_write_flush(git_hash_ctx *context, int fd)
2449 unsigned int buffered = write_buffer_len;
2450 if (buffered) {
2451 the_hash_algo->update_fn(context, write_buffer, buffered);
2452 if (write_in_full(fd, write_buffer, buffered) < 0)
2453 return -1;
2454 write_buffer_len = 0;
2456 return 0;
2459 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2461 while (len) {
2462 unsigned int buffered = write_buffer_len;
2463 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2464 if (partial > len)
2465 partial = len;
2466 memcpy(write_buffer + buffered, data, partial);
2467 buffered += partial;
2468 if (buffered == WRITE_BUFFER_SIZE) {
2469 write_buffer_len = buffered;
2470 if (ce_write_flush(context, fd))
2471 return -1;
2472 buffered = 0;
2474 write_buffer_len = buffered;
2475 len -= partial;
2476 data = (char *) data + partial;
2478 return 0;
2481 static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2482 int fd, unsigned int ext, unsigned int sz)
2484 ext = htonl(ext);
2485 sz = htonl(sz);
2486 if (eoie_context) {
2487 the_hash_algo->update_fn(eoie_context, &ext, 4);
2488 the_hash_algo->update_fn(eoie_context, &sz, 4);
2490 return ((ce_write(context, fd, &ext, 4) < 0) ||
2491 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2494 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2496 unsigned int left = write_buffer_len;
2498 if (left) {
2499 write_buffer_len = 0;
2500 the_hash_algo->update_fn(context, write_buffer, left);
2503 /* Flush first if not enough space for hash signature */
2504 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2505 if (write_in_full(fd, write_buffer, left) < 0)
2506 return -1;
2507 left = 0;
2510 /* Append the hash signature at the end */
2511 the_hash_algo->final_fn(write_buffer + left, context);
2512 hashcpy(hash, write_buffer + left);
2513 left += the_hash_algo->rawsz;
2514 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2517 static void ce_smudge_racily_clean_entry(struct index_state *istate,
2518 struct cache_entry *ce)
2521 * The only thing we care about in this function is to smudge the
2522 * falsely clean entry due to touch-update-touch race, so we leave
2523 * everything else as they are. We are called for entries whose
2524 * ce_stat_data.sd_mtime match the index file mtime.
2526 * Note that this actually does not do much for gitlinks, for
2527 * which ce_match_stat_basic() always goes to the actual
2528 * contents. The caller checks with is_racy_timestamp() which
2529 * always says "no" for gitlinks, so we are not called for them ;-)
2531 struct stat st;
2533 if (lstat(ce->name, &st) < 0)
2534 return;
2535 if (ce_match_stat_basic(ce, &st))
2536 return;
2537 if (ce_modified_check_fs(istate, ce, &st)) {
2538 /* This is "racily clean"; smudge it. Note that this
2539 * is a tricky code. At first glance, it may appear
2540 * that it can break with this sequence:
2542 * $ echo xyzzy >frotz
2543 * $ git-update-index --add frotz
2544 * $ : >frotz
2545 * $ sleep 3
2546 * $ echo filfre >nitfol
2547 * $ git-update-index --add nitfol
2549 * but it does not. When the second update-index runs,
2550 * it notices that the entry "frotz" has the same timestamp
2551 * as index, and if we were to smudge it by resetting its
2552 * size to zero here, then the object name recorded
2553 * in index is the 6-byte file but the cached stat information
2554 * becomes zero --- which would then match what we would
2555 * obtain from the filesystem next time we stat("frotz").
2557 * However, the second update-index, before calling
2558 * this function, notices that the cached size is 6
2559 * bytes and what is on the filesystem is an empty
2560 * file, and never calls us, so the cached size information
2561 * for "frotz" stays 6 which does not match the filesystem.
2563 ce->ce_stat_data.sd_size = 0;
2567 /* Copy miscellaneous fields but not the name */
2568 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2569 struct cache_entry *ce)
2571 short flags;
2572 const unsigned hashsz = the_hash_algo->rawsz;
2573 uint16_t *flagsp = (uint16_t *)(ondisk->data + hashsz);
2575 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2576 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2577 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2578 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2579 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2580 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2581 ondisk->mode = htonl(ce->ce_mode);
2582 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2583 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2584 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2585 hashcpy(ondisk->data, ce->oid.hash);
2587 flags = ce->ce_flags & ~CE_NAMEMASK;
2588 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2589 flagsp[0] = htons(flags);
2590 if (ce->ce_flags & CE_EXTENDED) {
2591 flagsp[1] = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2595 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2596 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2598 int size;
2599 int result;
2600 unsigned int saved_namelen;
2601 int stripped_name = 0;
2602 static unsigned char padding[8] = { 0x00 };
2604 if (ce->ce_flags & CE_STRIP_NAME) {
2605 saved_namelen = ce_namelen(ce);
2606 ce->ce_namelen = 0;
2607 stripped_name = 1;
2610 size = offsetof(struct ondisk_cache_entry,data) + ondisk_data_size(ce->ce_flags, 0);
2612 if (!previous_name) {
2613 int len = ce_namelen(ce);
2614 copy_cache_entry_to_ondisk(ondisk, ce);
2615 result = ce_write(c, fd, ondisk, size);
2616 if (!result)
2617 result = ce_write(c, fd, ce->name, len);
2618 if (!result)
2619 result = ce_write(c, fd, padding, align_padding_size(size, len));
2620 } else {
2621 int common, to_remove, prefix_size;
2622 unsigned char to_remove_vi[16];
2623 for (common = 0;
2624 (ce->name[common] &&
2625 common < previous_name->len &&
2626 ce->name[common] == previous_name->buf[common]);
2627 common++)
2628 ; /* still matching */
2629 to_remove = previous_name->len - common;
2630 prefix_size = encode_varint(to_remove, to_remove_vi);
2632 copy_cache_entry_to_ondisk(ondisk, ce);
2633 result = ce_write(c, fd, ondisk, size);
2634 if (!result)
2635 result = ce_write(c, fd, to_remove_vi, prefix_size);
2636 if (!result)
2637 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2638 if (!result)
2639 result = ce_write(c, fd, padding, 1);
2641 strbuf_splice(previous_name, common, to_remove,
2642 ce->name + common, ce_namelen(ce) - common);
2644 if (stripped_name) {
2645 ce->ce_namelen = saved_namelen;
2646 ce->ce_flags &= ~CE_STRIP_NAME;
2649 return result;
2653 * This function verifies if index_state has the correct sha1 of the
2654 * index file. Don't die if we have any other failure, just return 0.
2656 static int verify_index_from(const struct index_state *istate, const char *path)
2658 int fd;
2659 ssize_t n;
2660 struct stat st;
2661 unsigned char hash[GIT_MAX_RAWSZ];
2663 if (!istate->initialized)
2664 return 0;
2666 fd = open(path, O_RDONLY);
2667 if (fd < 0)
2668 return 0;
2670 if (fstat(fd, &st))
2671 goto out;
2673 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2674 goto out;
2676 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2677 if (n != the_hash_algo->rawsz)
2678 goto out;
2680 if (!hasheq(istate->oid.hash, hash))
2681 goto out;
2683 close(fd);
2684 return 1;
2686 out:
2687 close(fd);
2688 return 0;
2691 static int repo_verify_index(struct repository *repo)
2693 return verify_index_from(repo->index, repo->index_file);
2696 static int has_racy_timestamp(struct index_state *istate)
2698 int entries = istate->cache_nr;
2699 int i;
2701 for (i = 0; i < entries; i++) {
2702 struct cache_entry *ce = istate->cache[i];
2703 if (is_racy_timestamp(istate, ce))
2704 return 1;
2706 return 0;
2709 void repo_update_index_if_able(struct repository *repo,
2710 struct lock_file *lockfile)
2712 if ((repo->index->cache_changed ||
2713 has_racy_timestamp(repo->index)) &&
2714 repo_verify_index(repo))
2715 write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2716 else
2717 rollback_lock_file(lockfile);
2720 static int record_eoie(void)
2722 int val;
2724 if (!git_config_get_bool("index.recordendofindexentries", &val))
2725 return val;
2728 * As a convenience, the end of index entries extension
2729 * used for threading is written by default if the user
2730 * explicitly requested threaded index reads.
2732 return !git_config_get_index_threads(&val) && val != 1;
2735 static int record_ieot(void)
2737 int val;
2739 if (!git_config_get_bool("index.recordoffsettable", &val))
2740 return val;
2743 * As a convenience, the offset table used for threading is
2744 * written by default if the user explicitly requested
2745 * threaded index reads.
2747 return !git_config_get_index_threads(&val) && val != 1;
2751 * On success, `tempfile` is closed. If it is the temporary file
2752 * of a `struct lock_file`, we will therefore effectively perform
2753 * a 'close_lock_file_gently()`. Since that is an implementation
2754 * detail of lockfiles, callers of `do_write_index()` should not
2755 * rely on it.
2757 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2758 int strip_extensions)
2760 uint64_t start = getnanotime();
2761 int newfd = tempfile->fd;
2762 git_hash_ctx c, eoie_c;
2763 struct cache_header hdr;
2764 int i, err = 0, removed, extended, hdr_version;
2765 struct cache_entry **cache = istate->cache;
2766 int entries = istate->cache_nr;
2767 struct stat st;
2768 struct ondisk_cache_entry ondisk;
2769 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2770 int drop_cache_tree = istate->drop_cache_tree;
2771 off_t offset;
2772 int ieot_entries = 1;
2773 struct index_entry_offset_table *ieot = NULL;
2774 int nr, nr_threads;
2776 for (i = removed = extended = 0; i < entries; i++) {
2777 if (cache[i]->ce_flags & CE_REMOVE)
2778 removed++;
2780 /* reduce extended entries if possible */
2781 cache[i]->ce_flags &= ~CE_EXTENDED;
2782 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2783 extended++;
2784 cache[i]->ce_flags |= CE_EXTENDED;
2788 if (!istate->version) {
2789 istate->version = get_index_format_default(the_repository);
2790 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2791 init_split_index(istate);
2794 /* demote version 3 to version 2 when the latter suffices */
2795 if (istate->version == 3 || istate->version == 2)
2796 istate->version = extended ? 3 : 2;
2798 hdr_version = istate->version;
2800 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2801 hdr.hdr_version = htonl(hdr_version);
2802 hdr.hdr_entries = htonl(entries - removed);
2804 the_hash_algo->init_fn(&c);
2805 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2806 return -1;
2808 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))
2809 nr_threads = 1;
2811 if (nr_threads != 1 && record_ieot()) {
2812 int ieot_blocks, cpus;
2815 * ensure default number of ieot blocks maps evenly to the
2816 * default number of threads that will process them leaving
2817 * room for the thread to load the index extensions.
2819 if (!nr_threads) {
2820 ieot_blocks = istate->cache_nr / THREAD_COST;
2821 cpus = online_cpus();
2822 if (ieot_blocks > cpus - 1)
2823 ieot_blocks = cpus - 1;
2824 } else {
2825 ieot_blocks = nr_threads;
2826 if (ieot_blocks > istate->cache_nr)
2827 ieot_blocks = istate->cache_nr;
2831 * no reason to write out the IEOT extension if we don't
2832 * have enough blocks to utilize multi-threading
2834 if (ieot_blocks > 1) {
2835 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2836 + (ieot_blocks * sizeof(struct index_entry_offset)));
2837 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2841 offset = lseek(newfd, 0, SEEK_CUR);
2842 if (offset < 0) {
2843 free(ieot);
2844 return -1;
2846 offset += write_buffer_len;
2847 nr = 0;
2848 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2850 for (i = 0; i < entries; i++) {
2851 struct cache_entry *ce = cache[i];
2852 if (ce->ce_flags & CE_REMOVE)
2853 continue;
2854 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2855 ce_smudge_racily_clean_entry(istate, ce);
2856 if (is_null_oid(&ce->oid)) {
2857 static const char msg[] = "cache entry has null sha1: %s";
2858 static int allow = -1;
2860 if (allow < 0)
2861 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2862 if (allow)
2863 warning(msg, ce->name);
2864 else
2865 err = error(msg, ce->name);
2867 drop_cache_tree = 1;
2869 if (ieot && i && (i % ieot_entries == 0)) {
2870 ieot->entries[ieot->nr].nr = nr;
2871 ieot->entries[ieot->nr].offset = offset;
2872 ieot->nr++;
2874 * If we have a V4 index, set the first byte to an invalid
2875 * character to ensure there is nothing common with the previous
2876 * entry
2878 if (previous_name)
2879 previous_name->buf[0] = 0;
2880 nr = 0;
2881 offset = lseek(newfd, 0, SEEK_CUR);
2882 if (offset < 0) {
2883 free(ieot);
2884 return -1;
2886 offset += write_buffer_len;
2888 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2889 err = -1;
2891 if (err)
2892 break;
2893 nr++;
2895 if (ieot && nr) {
2896 ieot->entries[ieot->nr].nr = nr;
2897 ieot->entries[ieot->nr].offset = offset;
2898 ieot->nr++;
2900 strbuf_release(&previous_name_buf);
2902 if (err) {
2903 free(ieot);
2904 return err;
2907 /* Write extension data here */
2908 offset = lseek(newfd, 0, SEEK_CUR);
2909 if (offset < 0) {
2910 free(ieot);
2911 return -1;
2913 offset += write_buffer_len;
2914 the_hash_algo->init_fn(&eoie_c);
2917 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2918 * can minimize the number of extensions we have to scan through to
2919 * find it during load. Write it out regardless of the
2920 * strip_extensions parameter as we need it when loading the shared
2921 * index.
2923 if (ieot) {
2924 struct strbuf sb = STRBUF_INIT;
2926 write_ieot_extension(&sb, ieot);
2927 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2928 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2929 strbuf_release(&sb);
2930 free(ieot);
2931 if (err)
2932 return -1;
2935 if (!strip_extensions && istate->split_index &&
2936 !is_null_oid(&istate->split_index->base_oid)) {
2937 struct strbuf sb = STRBUF_INIT;
2939 err = write_link_extension(&sb, istate) < 0 ||
2940 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2941 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 && !drop_cache_tree && istate->cache_tree) {
2948 struct strbuf sb = STRBUF_INIT;
2950 cache_tree_write(&sb, istate->cache_tree);
2951 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2952 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2953 strbuf_release(&sb);
2954 if (err)
2955 return -1;
2957 if (!strip_extensions && istate->resolve_undo) {
2958 struct strbuf sb = STRBUF_INIT;
2960 resolve_undo_write(&sb, istate->resolve_undo);
2961 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2962 sb.len) < 0
2963 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2964 strbuf_release(&sb);
2965 if (err)
2966 return -1;
2968 if (!strip_extensions && istate->untracked) {
2969 struct strbuf sb = STRBUF_INIT;
2971 write_untracked_extension(&sb, istate->untracked);
2972 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2973 sb.len) < 0 ||
2974 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2975 strbuf_release(&sb);
2976 if (err)
2977 return -1;
2979 if (!strip_extensions && istate->fsmonitor_last_update) {
2980 struct strbuf sb = STRBUF_INIT;
2982 write_fsmonitor_extension(&sb, istate);
2983 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2984 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2985 strbuf_release(&sb);
2986 if (err)
2987 return -1;
2991 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2992 * so that it can be found and processed before all the index entries are
2993 * read. Write it out regardless of the strip_extensions parameter as we need it
2994 * when loading the shared index.
2996 if (offset && record_eoie()) {
2997 struct strbuf sb = STRBUF_INIT;
2999 write_eoie_extension(&sb, &eoie_c, offset);
3000 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
3001 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
3002 strbuf_release(&sb);
3003 if (err)
3004 return -1;
3007 if (ce_flush(&c, newfd, istate->oid.hash))
3008 return -1;
3009 if (close_tempfile_gently(tempfile)) {
3010 error(_("could not close '%s'"), tempfile->filename.buf);
3011 return -1;
3013 if (stat(tempfile->filename.buf, &st))
3014 return -1;
3015 istate->timestamp.sec = (unsigned int)st.st_mtime;
3016 istate->timestamp.nsec = ST_MTIME_NSEC(st);
3017 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
3020 * TODO trace2: replace "the_repository" with the actual repo instance
3021 * that is associated with the given "istate".
3023 trace2_data_intmax("index", the_repository, "write/version",
3024 istate->version);
3025 trace2_data_intmax("index", the_repository, "write/cache_nr",
3026 istate->cache_nr);
3028 return 0;
3031 void set_alternate_index_output(const char *name)
3033 alternate_index_output = name;
3036 static int commit_locked_index(struct lock_file *lk)
3038 if (alternate_index_output)
3039 return commit_lock_file_to(lk, alternate_index_output);
3040 else
3041 return commit_lock_file(lk);
3044 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
3045 unsigned flags)
3047 int ret;
3050 * TODO trace2: replace "the_repository" with the actual repo instance
3051 * that is associated with the given "istate".
3053 trace2_region_enter_printf("index", "do_write_index", the_repository,
3054 "%s", lock->tempfile->filename.buf);
3055 ret = do_write_index(istate, lock->tempfile, 0);
3056 trace2_region_leave_printf("index", "do_write_index", the_repository,
3057 "%s", lock->tempfile->filename.buf);
3059 if (ret)
3060 return ret;
3061 if (flags & COMMIT_LOCK)
3062 ret = commit_locked_index(lock);
3063 else
3064 ret = close_lock_file_gently(lock);
3066 run_hook_le(NULL, "post-index-change",
3067 istate->updated_workdir ? "1" : "0",
3068 istate->updated_skipworktree ? "1" : "0", NULL);
3069 istate->updated_workdir = 0;
3070 istate->updated_skipworktree = 0;
3072 return ret;
3075 static int write_split_index(struct index_state *istate,
3076 struct lock_file *lock,
3077 unsigned flags)
3079 int ret;
3080 prepare_to_write_split_index(istate);
3081 ret = do_write_locked_index(istate, lock, flags);
3082 finish_writing_split_index(istate);
3083 return ret;
3086 static const char *shared_index_expire = "2.weeks.ago";
3088 static unsigned long get_shared_index_expire_date(void)
3090 static unsigned long shared_index_expire_date;
3091 static int shared_index_expire_date_prepared;
3093 if (!shared_index_expire_date_prepared) {
3094 git_config_get_expiry("splitindex.sharedindexexpire",
3095 &shared_index_expire);
3096 shared_index_expire_date = approxidate(shared_index_expire);
3097 shared_index_expire_date_prepared = 1;
3100 return shared_index_expire_date;
3103 static int should_delete_shared_index(const char *shared_index_path)
3105 struct stat st;
3106 unsigned long expiration;
3108 /* Check timestamp */
3109 expiration = get_shared_index_expire_date();
3110 if (!expiration)
3111 return 0;
3112 if (stat(shared_index_path, &st))
3113 return error_errno(_("could not stat '%s'"), shared_index_path);
3114 if (st.st_mtime > expiration)
3115 return 0;
3117 return 1;
3120 static int clean_shared_index_files(const char *current_hex)
3122 struct dirent *de;
3123 DIR *dir = opendir(get_git_dir());
3125 if (!dir)
3126 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3128 while ((de = readdir(dir)) != NULL) {
3129 const char *sha1_hex;
3130 const char *shared_index_path;
3131 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3132 continue;
3133 if (!strcmp(sha1_hex, current_hex))
3134 continue;
3135 shared_index_path = git_path("%s", de->d_name);
3136 if (should_delete_shared_index(shared_index_path) > 0 &&
3137 unlink(shared_index_path))
3138 warning_errno(_("unable to unlink: %s"), shared_index_path);
3140 closedir(dir);
3142 return 0;
3145 static int write_shared_index(struct index_state *istate,
3146 struct tempfile **temp)
3148 struct split_index *si = istate->split_index;
3149 int ret;
3151 move_cache_to_base_index(istate);
3153 trace2_region_enter_printf("index", "shared/do_write_index",
3154 the_repository, "%s", (*temp)->filename.buf);
3155 ret = do_write_index(si->base, *temp, 1);
3156 trace2_region_leave_printf("index", "shared/do_write_index",
3157 the_repository, "%s", (*temp)->filename.buf);
3159 if (ret)
3160 return ret;
3161 ret = adjust_shared_perm(get_tempfile_path(*temp));
3162 if (ret) {
3163 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3164 return ret;
3166 ret = rename_tempfile(temp,
3167 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3168 if (!ret) {
3169 oidcpy(&si->base_oid, &si->base->oid);
3170 clean_shared_index_files(oid_to_hex(&si->base->oid));
3173 return ret;
3176 static const int default_max_percent_split_change = 20;
3178 static int too_many_not_shared_entries(struct index_state *istate)
3180 int i, not_shared = 0;
3181 int max_split = git_config_get_max_percent_split_change();
3183 switch (max_split) {
3184 case -1:
3185 /* not or badly configured: use the default value */
3186 max_split = default_max_percent_split_change;
3187 break;
3188 case 0:
3189 return 1; /* 0% means always write a new shared index */
3190 case 100:
3191 return 0; /* 100% means never write a new shared index */
3192 default:
3193 break; /* just use the configured value */
3196 /* Count not shared entries */
3197 for (i = 0; i < istate->cache_nr; i++) {
3198 struct cache_entry *ce = istate->cache[i];
3199 if (!ce->index)
3200 not_shared++;
3203 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3206 int write_locked_index(struct index_state *istate, struct lock_file *lock,
3207 unsigned flags)
3209 int new_shared_index, ret;
3210 struct split_index *si = istate->split_index;
3212 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3213 cache_tree_verify(the_repository, istate);
3215 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3216 if (flags & COMMIT_LOCK)
3217 rollback_lock_file(lock);
3218 return 0;
3221 if (istate->fsmonitor_last_update)
3222 fill_fsmonitor_bitmap(istate);
3224 if (!si || alternate_index_output ||
3225 (istate->cache_changed & ~EXTMASK)) {
3226 if (si)
3227 oidclr(&si->base_oid);
3228 ret = do_write_locked_index(istate, lock, flags);
3229 goto out;
3232 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3233 int v = si->base_oid.hash[0];
3234 if ((v & 15) < 6)
3235 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3237 if (too_many_not_shared_entries(istate))
3238 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3240 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3242 if (new_shared_index) {
3243 struct tempfile *temp;
3244 int saved_errno;
3246 /* Same initial permissions as the main .git/index file */
3247 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3248 if (!temp) {
3249 oidclr(&si->base_oid);
3250 ret = do_write_locked_index(istate, lock, flags);
3251 goto out;
3253 ret = write_shared_index(istate, &temp);
3255 saved_errno = errno;
3256 if (is_tempfile_active(temp))
3257 delete_tempfile(&temp);
3258 errno = saved_errno;
3260 if (ret)
3261 goto out;
3264 ret = write_split_index(istate, lock, flags);
3266 /* Freshen the shared index only if the split-index was written */
3267 if (!ret && !new_shared_index && !is_null_oid(&si->base_oid)) {
3268 const char *shared_index = git_path("sharedindex.%s",
3269 oid_to_hex(&si->base_oid));
3270 freshen_shared_index(shared_index, 1);
3273 out:
3274 if (flags & COMMIT_LOCK)
3275 rollback_lock_file(lock);
3276 return ret;
3280 * Read the index file that is potentially unmerged into given
3281 * index_state, dropping any unmerged entries to stage #0 (potentially
3282 * resulting in a path appearing as both a file and a directory in the
3283 * index; the caller is responsible to clear out the extra entries
3284 * before writing the index to a tree). Returns true if the index is
3285 * unmerged. Callers who want to refuse to work from an unmerged
3286 * state can call this and check its return value, instead of calling
3287 * read_cache().
3289 int repo_read_index_unmerged(struct repository *repo)
3291 struct index_state *istate;
3292 int i;
3293 int unmerged = 0;
3295 repo_read_index(repo);
3296 istate = repo->index;
3297 for (i = 0; i < istate->cache_nr; i++) {
3298 struct cache_entry *ce = istate->cache[i];
3299 struct cache_entry *new_ce;
3300 int len;
3302 if (!ce_stage(ce))
3303 continue;
3304 unmerged = 1;
3305 len = ce_namelen(ce);
3306 new_ce = make_empty_cache_entry(istate, len);
3307 memcpy(new_ce->name, ce->name, len);
3308 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3309 new_ce->ce_namelen = len;
3310 new_ce->ce_mode = ce->ce_mode;
3311 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3312 return error(_("%s: cannot drop to stage #0"),
3313 new_ce->name);
3315 return unmerged;
3319 * Returns 1 if the path is an "other" path with respect to
3320 * the index; that is, the path is not mentioned in the index at all,
3321 * either as a file, a directory with some files in the index,
3322 * or as an unmerged entry.
3324 * We helpfully remove a trailing "/" from directories so that
3325 * the output of read_directory can be used as-is.
3327 int index_name_is_other(const struct index_state *istate, const char *name,
3328 int namelen)
3330 int pos;
3331 if (namelen && name[namelen - 1] == '/')
3332 namelen--;
3333 pos = index_name_pos(istate, name, namelen);
3334 if (0 <= pos)
3335 return 0; /* exact match */
3336 pos = -pos - 1;
3337 if (pos < istate->cache_nr) {
3338 struct cache_entry *ce = istate->cache[pos];
3339 if (ce_namelen(ce) == namelen &&
3340 !memcmp(ce->name, name, namelen))
3341 return 0; /* Yup, this one exists unmerged */
3343 return 1;
3346 void *read_blob_data_from_index(const struct index_state *istate,
3347 const char *path, unsigned long *size)
3349 int pos, len;
3350 unsigned long sz;
3351 enum object_type type;
3352 void *data;
3354 len = strlen(path);
3355 pos = index_name_pos(istate, path, len);
3356 if (pos < 0) {
3358 * We might be in the middle of a merge, in which
3359 * case we would read stage #2 (ours).
3361 int i;
3362 for (i = -pos - 1;
3363 (pos < 0 && i < istate->cache_nr &&
3364 !strcmp(istate->cache[i]->name, path));
3365 i++)
3366 if (ce_stage(istate->cache[i]) == 2)
3367 pos = i;
3369 if (pos < 0)
3370 return NULL;
3371 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3372 if (!data || type != OBJ_BLOB) {
3373 free(data);
3374 return NULL;
3376 if (size)
3377 *size = sz;
3378 return data;
3381 void stat_validity_clear(struct stat_validity *sv)
3383 FREE_AND_NULL(sv->sd);
3386 int stat_validity_check(struct stat_validity *sv, const char *path)
3388 struct stat st;
3390 if (stat(path, &st) < 0)
3391 return sv->sd == NULL;
3392 if (!sv->sd)
3393 return 0;
3394 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3397 void stat_validity_update(struct stat_validity *sv, int fd)
3399 struct stat st;
3401 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3402 stat_validity_clear(sv);
3403 else {
3404 if (!sv->sd)
3405 sv->sd = xcalloc(1, sizeof(struct stat_data));
3406 fill_stat_data(sv->sd, &st);
3410 void move_index_extensions(struct index_state *dst, struct index_state *src)
3412 dst->untracked = src->untracked;
3413 src->untracked = NULL;
3414 dst->cache_tree = src->cache_tree;
3415 src->cache_tree = NULL;
3418 struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3419 struct index_state *istate)
3421 unsigned int size = ce_size(ce);
3422 int mem_pool_allocated;
3423 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3424 mem_pool_allocated = new_entry->mem_pool_allocated;
3426 memcpy(new_entry, ce, size);
3427 new_entry->mem_pool_allocated = mem_pool_allocated;
3428 return new_entry;
3431 void discard_cache_entry(struct cache_entry *ce)
3433 if (ce && should_validate_cache_entries())
3434 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3436 if (ce && ce->mem_pool_allocated)
3437 return;
3439 free(ce);
3442 int should_validate_cache_entries(void)
3444 static int validate_index_cache_entries = -1;
3446 if (validate_index_cache_entries < 0) {
3447 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3448 validate_index_cache_entries = 1;
3449 else
3450 validate_index_cache_entries = 0;
3453 return validate_index_cache_entries;
3456 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3457 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3459 static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3462 * The end of index entries (EOIE) extension is guaranteed to be last
3463 * so that it can be found by scanning backwards from the EOF.
3465 * "EOIE"
3466 * <4-byte length>
3467 * <4-byte offset>
3468 * <20-byte hash>
3470 const char *index, *eoie;
3471 uint32_t extsize;
3472 size_t offset, src_offset;
3473 unsigned char hash[GIT_MAX_RAWSZ];
3474 git_hash_ctx c;
3476 /* ensure we have an index big enough to contain an EOIE extension */
3477 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3478 return 0;
3480 /* validate the extension signature */
3481 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3482 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3483 return 0;
3484 index += sizeof(uint32_t);
3486 /* validate the extension size */
3487 extsize = get_be32(index);
3488 if (extsize != EOIE_SIZE)
3489 return 0;
3490 index += sizeof(uint32_t);
3493 * Validate the offset we're going to look for the first extension
3494 * signature is after the index header and before the eoie extension.
3496 offset = get_be32(index);
3497 if (mmap + offset < mmap + sizeof(struct cache_header))
3498 return 0;
3499 if (mmap + offset >= eoie)
3500 return 0;
3501 index += sizeof(uint32_t);
3504 * The hash is computed over extension types and their sizes (but not
3505 * their contents). E.g. if we have "TREE" extension that is N-bytes
3506 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3507 * then the hash would be:
3509 * SHA-1("TREE" + <binary representation of N> +
3510 * "REUC" + <binary representation of M>)
3512 src_offset = offset;
3513 the_hash_algo->init_fn(&c);
3514 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3515 /* After an array of active_nr index entries,
3516 * there can be arbitrary number of extended
3517 * sections, each of which is prefixed with
3518 * extension name (4-byte) and section length
3519 * in 4-byte network byte order.
3521 uint32_t extsize;
3522 memcpy(&extsize, mmap + src_offset + 4, 4);
3523 extsize = ntohl(extsize);
3525 /* verify the extension size isn't so large it will wrap around */
3526 if (src_offset + 8 + extsize < src_offset)
3527 return 0;
3529 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3531 src_offset += 8;
3532 src_offset += extsize;
3534 the_hash_algo->final_fn(hash, &c);
3535 if (!hasheq(hash, (const unsigned char *)index))
3536 return 0;
3538 /* Validate that the extension offsets returned us back to the eoie extension. */
3539 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3540 return 0;
3542 return offset;
3545 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3547 uint32_t buffer;
3548 unsigned char hash[GIT_MAX_RAWSZ];
3550 /* offset */
3551 put_be32(&buffer, offset);
3552 strbuf_add(sb, &buffer, sizeof(uint32_t));
3554 /* hash */
3555 the_hash_algo->final_fn(hash, eoie_context);
3556 strbuf_add(sb, hash, the_hash_algo->rawsz);
3559 #define IEOT_VERSION (1)
3561 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3563 const char *index = NULL;
3564 uint32_t extsize, ext_version;
3565 struct index_entry_offset_table *ieot;
3566 int i, nr;
3568 /* find the IEOT extension */
3569 if (!offset)
3570 return NULL;
3571 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3572 extsize = get_be32(mmap + offset + 4);
3573 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3574 index = mmap + offset + 4 + 4;
3575 break;
3577 offset += 8;
3578 offset += extsize;
3580 if (!index)
3581 return NULL;
3583 /* validate the version is IEOT_VERSION */
3584 ext_version = get_be32(index);
3585 if (ext_version != IEOT_VERSION) {
3586 error("invalid IEOT version %d", ext_version);
3587 return NULL;
3589 index += sizeof(uint32_t);
3591 /* extension size - version bytes / bytes per entry */
3592 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3593 if (!nr) {
3594 error("invalid number of IEOT entries %d", nr);
3595 return NULL;
3597 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3598 + (nr * sizeof(struct index_entry_offset)));
3599 ieot->nr = nr;
3600 for (i = 0; i < nr; i++) {
3601 ieot->entries[i].offset = get_be32(index);
3602 index += sizeof(uint32_t);
3603 ieot->entries[i].nr = get_be32(index);
3604 index += sizeof(uint32_t);
3607 return ieot;
3610 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3612 uint32_t buffer;
3613 int i;
3615 /* version */
3616 put_be32(&buffer, IEOT_VERSION);
3617 strbuf_add(sb, &buffer, sizeof(uint32_t));
3619 /* ieot */
3620 for (i = 0; i < ieot->nr; i++) {
3622 /* offset */
3623 put_be32(&buffer, ieot->entries[i].offset);
3624 strbuf_add(sb, &buffer, sizeof(uint32_t));
3626 /* count */
3627 put_be32(&buffer, ieot->entries[i].nr);
3628 strbuf_add(sb, &buffer, sizeof(uint32_t));