Start the post 2.27 cycle
[git.git] / read-cache.c
blobaa427c5c170f02c0c2300f4de5e16a03c83dd975
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 = 0;
964 if (has_dos_drive_prefix(path))
965 return 0;
967 if (!is_valid_path(path))
968 return 0;
970 goto inside;
971 for (;;) {
972 if (!c)
973 return 1;
974 if (is_dir_sep(c)) {
975 inside:
976 if (protect_hfs) {
978 if (is_hfs_dotgit(path))
979 return 0;
980 if (S_ISLNK(mode)) {
981 if (is_hfs_dotgitmodules(path))
982 return 0;
985 if (protect_ntfs) {
986 #ifdef GIT_WINDOWS_NATIVE
987 if (c == '\\')
988 return 0;
989 #endif
990 if (is_ntfs_dotgit(path))
991 return 0;
992 if (S_ISLNK(mode)) {
993 if (is_ntfs_dotgitmodules(path))
994 return 0;
998 c = *path++;
999 if ((c == '.' && !verify_dotfile(path, mode)) ||
1000 is_dir_sep(c) || c == '\0')
1001 return 0;
1002 } else if (c == '\\' && protect_ntfs) {
1003 if (is_ntfs_dotgit(path))
1004 return 0;
1005 if (S_ISLNK(mode)) {
1006 if (is_ntfs_dotgitmodules(path))
1007 return 0;
1011 c = *path++;
1016 * Do we have another file that has the beginning components being a
1017 * proper superset of the name we're trying to add?
1019 static int has_file_name(struct index_state *istate,
1020 const struct cache_entry *ce, int pos, int ok_to_replace)
1022 int retval = 0;
1023 int len = ce_namelen(ce);
1024 int stage = ce_stage(ce);
1025 const char *name = ce->name;
1027 while (pos < istate->cache_nr) {
1028 struct cache_entry *p = istate->cache[pos++];
1030 if (len >= ce_namelen(p))
1031 break;
1032 if (memcmp(name, p->name, len))
1033 break;
1034 if (ce_stage(p) != stage)
1035 continue;
1036 if (p->name[len] != '/')
1037 continue;
1038 if (p->ce_flags & CE_REMOVE)
1039 continue;
1040 retval = -1;
1041 if (!ok_to_replace)
1042 break;
1043 remove_index_entry_at(istate, --pos);
1045 return retval;
1050 * Like strcmp(), but also return the offset of the first change.
1051 * If strings are equal, return the length.
1053 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1055 size_t k;
1057 if (!first_change)
1058 return strcmp(s1, s2);
1060 for (k = 0; s1[k] == s2[k]; k++)
1061 if (s1[k] == '\0')
1062 break;
1064 *first_change = k;
1065 return (unsigned char)s1[k] - (unsigned char)s2[k];
1069 * Do we have another file with a pathname that is a proper
1070 * subset of the name we're trying to add?
1072 * That is, is there another file in the index with a path
1073 * that matches a sub-directory in the given entry?
1075 static int has_dir_name(struct index_state *istate,
1076 const struct cache_entry *ce, int pos, int ok_to_replace)
1078 int retval = 0;
1079 int stage = ce_stage(ce);
1080 const char *name = ce->name;
1081 const char *slash = name + ce_namelen(ce);
1082 size_t len_eq_last;
1083 int cmp_last = 0;
1086 * We are frequently called during an iteration on a sorted
1087 * list of pathnames and while building a new index. Therefore,
1088 * there is a high probability that this entry will eventually
1089 * be appended to the index, rather than inserted in the middle.
1090 * If we can confirm that, we can avoid binary searches on the
1091 * components of the pathname.
1093 * Compare the entry's full path with the last path in the index.
1095 if (istate->cache_nr > 0) {
1096 cmp_last = strcmp_offset(name,
1097 istate->cache[istate->cache_nr - 1]->name,
1098 &len_eq_last);
1099 if (cmp_last > 0) {
1100 if (len_eq_last == 0) {
1102 * The entry sorts AFTER the last one in the
1103 * index and their paths have no common prefix,
1104 * so there cannot be a F/D conflict.
1106 return retval;
1107 } else {
1109 * The entry sorts AFTER the last one in the
1110 * index, but has a common prefix. Fall through
1111 * to the loop below to disect the entry's path
1112 * and see where the difference is.
1115 } else if (cmp_last == 0) {
1117 * The entry exactly matches the last one in the
1118 * index, but because of multiple stage and CE_REMOVE
1119 * items, we fall through and let the regular search
1120 * code handle it.
1125 for (;;) {
1126 size_t len;
1128 for (;;) {
1129 if (*--slash == '/')
1130 break;
1131 if (slash <= ce->name)
1132 return retval;
1134 len = slash - name;
1136 if (cmp_last > 0) {
1138 * (len + 1) is a directory boundary (including
1139 * the trailing slash). And since the loop is
1140 * decrementing "slash", the first iteration is
1141 * the longest directory prefix; subsequent
1142 * iterations consider parent directories.
1145 if (len + 1 <= len_eq_last) {
1147 * The directory prefix (including the trailing
1148 * slash) also appears as a prefix in the last
1149 * entry, so the remainder cannot collide (because
1150 * strcmp said the whole path was greater).
1152 * EQ: last: xxx/A
1153 * this: xxx/B
1155 * LT: last: xxx/file_A
1156 * this: xxx/file_B
1158 return retval;
1161 if (len > len_eq_last) {
1163 * This part of the directory prefix (excluding
1164 * the trailing slash) is longer than the known
1165 * equal portions, so this sub-directory cannot
1166 * collide with a file.
1168 * GT: last: xxxA
1169 * this: xxxB/file
1171 return retval;
1174 if (istate->cache_nr > 0 &&
1175 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1177 * The directory prefix lines up with part of
1178 * a longer file or directory name, but sorts
1179 * after it, so this sub-directory cannot
1180 * collide with a file.
1182 * last: xxx/yy-file (because '-' sorts before '/')
1183 * this: xxx/yy/abc
1185 return retval;
1189 * This is a possible collision. Fall through and
1190 * let the regular search code handle it.
1192 * last: xxx
1193 * this: xxx/file
1197 pos = index_name_stage_pos(istate, name, len, stage);
1198 if (pos >= 0) {
1200 * Found one, but not so fast. This could
1201 * be a marker that says "I was here, but
1202 * I am being removed". Such an entry is
1203 * not a part of the resulting tree, and
1204 * it is Ok to have a directory at the same
1205 * path.
1207 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1208 retval = -1;
1209 if (!ok_to_replace)
1210 break;
1211 remove_index_entry_at(istate, pos);
1212 continue;
1215 else
1216 pos = -pos-1;
1219 * Trivial optimization: if we find an entry that
1220 * already matches the sub-directory, then we know
1221 * we're ok, and we can exit.
1223 while (pos < istate->cache_nr) {
1224 struct cache_entry *p = istate->cache[pos];
1225 if ((ce_namelen(p) <= len) ||
1226 (p->name[len] != '/') ||
1227 memcmp(p->name, name, len))
1228 break; /* not our subdirectory */
1229 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1231 * p is at the same stage as our entry, and
1232 * is a subdirectory of what we are looking
1233 * at, so we cannot have conflicts at our
1234 * level or anything shorter.
1236 return retval;
1237 pos++;
1240 return retval;
1243 /* We may be in a situation where we already have path/file and path
1244 * is being added, or we already have path and path/file is being
1245 * added. Either one would result in a nonsense tree that has path
1246 * twice when git-write-tree tries to write it out. Prevent it.
1248 * If ok-to-replace is specified, we remove the conflicting entries
1249 * from the cache so the caller should recompute the insert position.
1250 * When this happens, we return non-zero.
1252 static int check_file_directory_conflict(struct index_state *istate,
1253 const struct cache_entry *ce,
1254 int pos, int ok_to_replace)
1256 int retval;
1259 * When ce is an "I am going away" entry, we allow it to be added
1261 if (ce->ce_flags & CE_REMOVE)
1262 return 0;
1265 * We check if the path is a sub-path of a subsequent pathname
1266 * first, since removing those will not change the position
1267 * in the array.
1269 retval = has_file_name(istate, ce, pos, ok_to_replace);
1272 * Then check if the path might have a clashing sub-directory
1273 * before it.
1275 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1278 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1280 int pos;
1281 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1282 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1283 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1284 int new_only = option & ADD_CACHE_NEW_ONLY;
1286 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1287 cache_tree_invalidate_path(istate, ce->name);
1290 * If this entry's path sorts after the last entry in the index,
1291 * we can avoid searching for it.
1293 if (istate->cache_nr > 0 &&
1294 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1295 pos = index_pos_to_insert_pos(istate->cache_nr);
1296 else
1297 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1299 /* existing match? Just replace it. */
1300 if (pos >= 0) {
1301 if (!new_only)
1302 replace_index_entry(istate, pos, ce);
1303 return 0;
1305 pos = -pos-1;
1307 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1308 untracked_cache_add_to_index(istate, ce->name);
1311 * Inserting a merged entry ("stage 0") into the index
1312 * will always replace all non-merged entries..
1314 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1315 while (ce_same_name(istate->cache[pos], ce)) {
1316 ok_to_add = 1;
1317 if (!remove_index_entry_at(istate, pos))
1318 break;
1322 if (!ok_to_add)
1323 return -1;
1324 if (!verify_path(ce->name, ce->ce_mode))
1325 return error(_("invalid path '%s'"), ce->name);
1327 if (!skip_df_check &&
1328 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1329 if (!ok_to_replace)
1330 return error(_("'%s' appears as both a file and as a directory"),
1331 ce->name);
1332 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1333 pos = -pos-1;
1335 return pos + 1;
1338 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1340 int pos;
1342 if (option & ADD_CACHE_JUST_APPEND)
1343 pos = istate->cache_nr;
1344 else {
1345 int ret;
1346 ret = add_index_entry_with_check(istate, ce, option);
1347 if (ret <= 0)
1348 return ret;
1349 pos = ret - 1;
1352 /* Make sure the array is big enough .. */
1353 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1355 /* Add it in.. */
1356 istate->cache_nr++;
1357 if (istate->cache_nr > pos + 1)
1358 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1359 istate->cache_nr - pos - 1);
1360 set_index_entry(istate, pos, ce);
1361 istate->cache_changed |= CE_ENTRY_ADDED;
1362 return 0;
1366 * "refresh" does not calculate a new sha1 file or bring the
1367 * cache up-to-date for mode/content changes. But what it
1368 * _does_ do is to "re-match" the stat information of a file
1369 * with the cache, so that you can refresh the cache for a
1370 * file that hasn't been changed but where the stat entry is
1371 * out of date.
1373 * For example, you'd want to do this after doing a "git-read-tree",
1374 * to link up the stat cache details with the proper files.
1376 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1377 struct cache_entry *ce,
1378 unsigned int options, int *err,
1379 int *changed_ret)
1381 struct stat st;
1382 struct cache_entry *updated;
1383 int changed;
1384 int refresh = options & CE_MATCH_REFRESH;
1385 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1386 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1387 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1388 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1390 if (!refresh || ce_uptodate(ce))
1391 return ce;
1393 if (!ignore_fsmonitor)
1394 refresh_fsmonitor(istate);
1396 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1397 * that the change to the work tree does not matter and told
1398 * us not to worry.
1400 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1401 ce_mark_uptodate(ce);
1402 return ce;
1404 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1405 ce_mark_uptodate(ce);
1406 return ce;
1408 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1409 ce_mark_uptodate(ce);
1410 return ce;
1413 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1414 if (ignore_missing)
1415 return ce;
1416 if (err)
1417 *err = ENOENT;
1418 return NULL;
1421 if (lstat(ce->name, &st) < 0) {
1422 if (ignore_missing && errno == ENOENT)
1423 return ce;
1424 if (err)
1425 *err = errno;
1426 return NULL;
1429 changed = ie_match_stat(istate, ce, &st, options);
1430 if (changed_ret)
1431 *changed_ret = changed;
1432 if (!changed) {
1434 * The path is unchanged. If we were told to ignore
1435 * valid bit, then we did the actual stat check and
1436 * found that the entry is unmodified. If the entry
1437 * is not marked VALID, this is the place to mark it
1438 * valid again, under "assume unchanged" mode.
1440 if (ignore_valid && assume_unchanged &&
1441 !(ce->ce_flags & CE_VALID))
1442 ; /* mark this one VALID again */
1443 else {
1445 * We do not mark the index itself "modified"
1446 * because CE_UPTODATE flag is in-core only;
1447 * we are not going to write this change out.
1449 if (!S_ISGITLINK(ce->ce_mode)) {
1450 ce_mark_uptodate(ce);
1451 mark_fsmonitor_valid(istate, ce);
1453 return ce;
1457 if (ie_modified(istate, ce, &st, options)) {
1458 if (err)
1459 *err = EINVAL;
1460 return NULL;
1463 updated = make_empty_cache_entry(istate, ce_namelen(ce));
1464 copy_cache_entry(updated, ce);
1465 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1466 fill_stat_cache_info(istate, updated, &st);
1468 * If ignore_valid is not set, we should leave CE_VALID bit
1469 * alone. Otherwise, paths marked with --no-assume-unchanged
1470 * (i.e. things to be edited) will reacquire CE_VALID bit
1471 * automatically, which is not really what we want.
1473 if (!ignore_valid && assume_unchanged &&
1474 !(ce->ce_flags & CE_VALID))
1475 updated->ce_flags &= ~CE_VALID;
1477 /* istate->cache_changed is updated in the caller */
1478 return updated;
1481 static void show_file(const char * fmt, const char * name, int in_porcelain,
1482 int * first, const char *header_msg)
1484 if (in_porcelain && *first && header_msg) {
1485 printf("%s\n", header_msg);
1486 *first = 0;
1488 printf(fmt, name);
1491 int repo_refresh_and_write_index(struct repository *repo,
1492 unsigned int refresh_flags,
1493 unsigned int write_flags,
1494 int gentle,
1495 const struct pathspec *pathspec,
1496 char *seen, const char *header_msg)
1498 struct lock_file lock_file = LOCK_INIT;
1499 int fd, ret = 0;
1501 fd = repo_hold_locked_index(repo, &lock_file, 0);
1502 if (!gentle && fd < 0)
1503 return -1;
1504 if (refresh_index(repo->index, refresh_flags, pathspec, seen, header_msg))
1505 ret = 1;
1506 if (0 <= fd && write_locked_index(repo->index, &lock_file, COMMIT_LOCK | write_flags))
1507 ret = -1;
1508 return ret;
1512 int refresh_index(struct index_state *istate, unsigned int flags,
1513 const struct pathspec *pathspec,
1514 char *seen, const char *header_msg)
1516 int i;
1517 int has_errors = 0;
1518 int really = (flags & REFRESH_REALLY) != 0;
1519 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1520 int quiet = (flags & REFRESH_QUIET) != 0;
1521 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1522 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1523 int first = 1;
1524 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1525 unsigned int options = (CE_MATCH_REFRESH |
1526 (really ? CE_MATCH_IGNORE_VALID : 0) |
1527 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1528 const char *modified_fmt;
1529 const char *deleted_fmt;
1530 const char *typechange_fmt;
1531 const char *added_fmt;
1532 const char *unmerged_fmt;
1533 struct progress *progress = NULL;
1535 if (flags & REFRESH_PROGRESS && isatty(2))
1536 progress = start_delayed_progress(_("Refresh index"),
1537 istate->cache_nr);
1539 trace_performance_enter();
1540 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1541 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1542 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1543 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1544 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1546 * Use the multi-threaded preload_index() to refresh most of the
1547 * cache entries quickly then in the single threaded loop below,
1548 * we only have to do the special cases that are left.
1550 preload_index(istate, pathspec, 0);
1551 for (i = 0; i < istate->cache_nr; i++) {
1552 struct cache_entry *ce, *new_entry;
1553 int cache_errno = 0;
1554 int changed = 0;
1555 int filtered = 0;
1557 ce = istate->cache[i];
1558 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1559 continue;
1561 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1562 filtered = 1;
1564 if (ce_stage(ce)) {
1565 while ((i < istate->cache_nr) &&
1566 ! strcmp(istate->cache[i]->name, ce->name))
1567 i++;
1568 i--;
1569 if (allow_unmerged)
1570 continue;
1571 if (!filtered)
1572 show_file(unmerged_fmt, ce->name, in_porcelain,
1573 &first, header_msg);
1574 has_errors = 1;
1575 continue;
1578 if (filtered)
1579 continue;
1581 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1582 if (new_entry == ce)
1583 continue;
1584 if (progress)
1585 display_progress(progress, i);
1586 if (!new_entry) {
1587 const char *fmt;
1589 if (really && cache_errno == EINVAL) {
1590 /* If we are doing --really-refresh that
1591 * means the index is not valid anymore.
1593 ce->ce_flags &= ~CE_VALID;
1594 ce->ce_flags |= CE_UPDATE_IN_BASE;
1595 mark_fsmonitor_invalid(istate, ce);
1596 istate->cache_changed |= CE_ENTRY_CHANGED;
1598 if (quiet)
1599 continue;
1601 if (cache_errno == ENOENT)
1602 fmt = deleted_fmt;
1603 else if (ce_intent_to_add(ce))
1604 fmt = added_fmt; /* must be before other checks */
1605 else if (changed & TYPE_CHANGED)
1606 fmt = typechange_fmt;
1607 else
1608 fmt = modified_fmt;
1609 show_file(fmt,
1610 ce->name, in_porcelain, &first, header_msg);
1611 has_errors = 1;
1612 continue;
1615 replace_index_entry(istate, i, new_entry);
1617 if (progress) {
1618 display_progress(progress, istate->cache_nr);
1619 stop_progress(&progress);
1621 trace_performance_leave("refresh index");
1622 return has_errors;
1625 struct cache_entry *refresh_cache_entry(struct index_state *istate,
1626 struct cache_entry *ce,
1627 unsigned int options)
1629 return refresh_cache_ent(istate, ce, options, NULL, NULL);
1633 /*****************************************************************
1634 * Index File I/O
1635 *****************************************************************/
1637 #define INDEX_FORMAT_DEFAULT 3
1639 static unsigned int get_index_format_default(struct repository *r)
1641 char *envversion = getenv("GIT_INDEX_VERSION");
1642 char *endp;
1643 unsigned int version = INDEX_FORMAT_DEFAULT;
1645 if (!envversion) {
1646 prepare_repo_settings(r);
1648 if (r->settings.index_version >= 0)
1649 version = r->settings.index_version;
1650 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1651 warning(_("index.version set, but the value is invalid.\n"
1652 "Using version %i"), INDEX_FORMAT_DEFAULT);
1653 return INDEX_FORMAT_DEFAULT;
1655 return version;
1658 version = strtoul(envversion, &endp, 10);
1659 if (*endp ||
1660 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1661 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1662 "Using version %i"), INDEX_FORMAT_DEFAULT);
1663 version = INDEX_FORMAT_DEFAULT;
1665 return version;
1669 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1670 * Again - this is just a (very strong in practice) heuristic that
1671 * the inode hasn't changed.
1673 * We save the fields in big-endian order to allow using the
1674 * index file over NFS transparently.
1676 struct ondisk_cache_entry {
1677 struct cache_time ctime;
1678 struct cache_time mtime;
1679 uint32_t dev;
1680 uint32_t ino;
1681 uint32_t mode;
1682 uint32_t uid;
1683 uint32_t gid;
1684 uint32_t size;
1686 * unsigned char hash[hashsz];
1687 * uint16_t flags;
1688 * if (flags & CE_EXTENDED)
1689 * uint16_t flags2;
1691 unsigned char data[GIT_MAX_RAWSZ + 2 * sizeof(uint16_t)];
1692 char name[FLEX_ARRAY];
1695 /* These are only used for v3 or lower */
1696 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1697 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1698 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1699 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1700 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1701 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1702 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1704 /* Allow fsck to force verification of the index checksum. */
1705 int verify_index_checksum;
1707 /* Allow fsck to force verification of the cache entry order. */
1708 int verify_ce_order;
1710 static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1712 git_hash_ctx c;
1713 unsigned char hash[GIT_MAX_RAWSZ];
1714 int hdr_version;
1716 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1717 return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1718 hdr_version = ntohl(hdr->hdr_version);
1719 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1720 return error(_("bad index version %d"), hdr_version);
1722 if (!verify_index_checksum)
1723 return 0;
1725 the_hash_algo->init_fn(&c);
1726 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1727 the_hash_algo->final_fn(hash, &c);
1728 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1729 return error(_("bad index file sha1 signature"));
1730 return 0;
1733 static int read_index_extension(struct index_state *istate,
1734 const char *ext, const char *data, unsigned long sz)
1736 switch (CACHE_EXT(ext)) {
1737 case CACHE_EXT_TREE:
1738 istate->cache_tree = cache_tree_read(data, sz);
1739 break;
1740 case CACHE_EXT_RESOLVE_UNDO:
1741 istate->resolve_undo = resolve_undo_read(data, sz);
1742 break;
1743 case CACHE_EXT_LINK:
1744 if (read_link_extension(istate, data, sz))
1745 return -1;
1746 break;
1747 case CACHE_EXT_UNTRACKED:
1748 istate->untracked = read_untracked_extension(data, sz);
1749 break;
1750 case CACHE_EXT_FSMONITOR:
1751 read_fsmonitor_extension(istate, data, sz);
1752 break;
1753 case CACHE_EXT_ENDOFINDEXENTRIES:
1754 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1755 /* already handled in do_read_index() */
1756 break;
1757 default:
1758 if (*ext < 'A' || 'Z' < *ext)
1759 return error(_("index uses %.4s extension, which we do not understand"),
1760 ext);
1761 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1762 break;
1764 return 0;
1767 static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1768 unsigned int version,
1769 struct ondisk_cache_entry *ondisk,
1770 unsigned long *ent_size,
1771 const struct cache_entry *previous_ce)
1773 struct cache_entry *ce;
1774 size_t len;
1775 const char *name;
1776 const unsigned hashsz = the_hash_algo->rawsz;
1777 const uint16_t *flagsp = (const uint16_t *)(ondisk->data + hashsz);
1778 unsigned int flags;
1779 size_t copy_len = 0;
1781 * Adjacent cache entries tend to share the leading paths, so it makes
1782 * sense to only store the differences in later entries. In the v4
1783 * on-disk format of the index, each on-disk cache entry stores the
1784 * number of bytes to be stripped from the end of the previous name,
1785 * and the bytes to append to the result, to come up with its name.
1787 int expand_name_field = version == 4;
1789 /* On-disk flags are just 16 bits */
1790 flags = get_be16(flagsp);
1791 len = flags & CE_NAMEMASK;
1793 if (flags & CE_EXTENDED) {
1794 int extended_flags;
1795 extended_flags = get_be16(flagsp + 1) << 16;
1796 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1797 if (extended_flags & ~CE_EXTENDED_FLAGS)
1798 die(_("unknown index entry format 0x%08x"), extended_flags);
1799 flags |= extended_flags;
1800 name = (const char *)(flagsp + 2);
1802 else
1803 name = (const char *)(flagsp + 1);
1805 if (expand_name_field) {
1806 const unsigned char *cp = (const unsigned char *)name;
1807 size_t strip_len, previous_len;
1809 /* If we're at the beginning of a block, ignore the previous name */
1810 strip_len = decode_varint(&cp);
1811 if (previous_ce) {
1812 previous_len = previous_ce->ce_namelen;
1813 if (previous_len < strip_len)
1814 die(_("malformed name field in the index, near path '%s'"),
1815 previous_ce->name);
1816 copy_len = previous_len - strip_len;
1818 name = (const char *)cp;
1821 if (len == CE_NAMEMASK) {
1822 len = strlen(name);
1823 if (expand_name_field)
1824 len += copy_len;
1827 ce = mem_pool__ce_alloc(ce_mem_pool, len);
1829 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1830 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1831 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1832 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1833 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1834 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1835 ce->ce_mode = get_be32(&ondisk->mode);
1836 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1837 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1838 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1839 ce->ce_flags = flags & ~CE_NAMEMASK;
1840 ce->ce_namelen = len;
1841 ce->index = 0;
1842 hashcpy(ce->oid.hash, ondisk->data);
1843 memcpy(ce->name, name, len);
1844 ce->name[len] = '\0';
1846 if (expand_name_field) {
1847 if (copy_len)
1848 memcpy(ce->name, previous_ce->name, copy_len);
1849 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1850 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1851 } else {
1852 memcpy(ce->name, name, len + 1);
1853 *ent_size = ondisk_ce_size(ce);
1855 return ce;
1858 static void check_ce_order(struct index_state *istate)
1860 unsigned int i;
1862 if (!verify_ce_order)
1863 return;
1865 for (i = 1; i < istate->cache_nr; i++) {
1866 struct cache_entry *ce = istate->cache[i - 1];
1867 struct cache_entry *next_ce = istate->cache[i];
1868 int name_compare = strcmp(ce->name, next_ce->name);
1870 if (0 < name_compare)
1871 die(_("unordered stage entries in index"));
1872 if (!name_compare) {
1873 if (!ce_stage(ce))
1874 die(_("multiple stage entries for merged file '%s'"),
1875 ce->name);
1876 if (ce_stage(ce) > ce_stage(next_ce))
1877 die(_("unordered stage entries for '%s'"),
1878 ce->name);
1883 static void tweak_untracked_cache(struct index_state *istate)
1885 struct repository *r = the_repository;
1887 prepare_repo_settings(r);
1889 if (r->settings.core_untracked_cache == UNTRACKED_CACHE_REMOVE) {
1890 remove_untracked_cache(istate);
1891 return;
1894 if (r->settings.core_untracked_cache == UNTRACKED_CACHE_WRITE)
1895 add_untracked_cache(istate);
1898 static void tweak_split_index(struct index_state *istate)
1900 switch (git_config_get_split_index()) {
1901 case -1: /* unset: do nothing */
1902 break;
1903 case 0: /* false */
1904 remove_split_index(istate);
1905 break;
1906 case 1: /* true */
1907 add_split_index(istate);
1908 break;
1909 default: /* unknown value: do nothing */
1910 break;
1914 static void post_read_index_from(struct index_state *istate)
1916 check_ce_order(istate);
1917 tweak_untracked_cache(istate);
1918 tweak_split_index(istate);
1919 tweak_fsmonitor(istate);
1922 static size_t estimate_cache_size_from_compressed(unsigned int entries)
1924 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1927 static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1929 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1932 * Account for potential alignment differences.
1934 per_entry += align_padding_size(per_entry, 0);
1935 return ondisk_size + entries * per_entry;
1938 struct index_entry_offset
1940 /* starting byte offset into index file, count of index entries in this block */
1941 int offset, nr;
1944 struct index_entry_offset_table
1946 int nr;
1947 struct index_entry_offset entries[FLEX_ARRAY];
1950 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1951 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1953 static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1954 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1956 struct load_index_extensions
1958 pthread_t pthread;
1959 struct index_state *istate;
1960 const char *mmap;
1961 size_t mmap_size;
1962 unsigned long src_offset;
1965 static void *load_index_extensions(void *_data)
1967 struct load_index_extensions *p = _data;
1968 unsigned long src_offset = p->src_offset;
1970 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1971 /* After an array of active_nr index entries,
1972 * there can be arbitrary number of extended
1973 * sections, each of which is prefixed with
1974 * extension name (4-byte) and section length
1975 * in 4-byte network byte order.
1977 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1978 if (read_index_extension(p->istate,
1979 p->mmap + src_offset,
1980 p->mmap + src_offset + 8,
1981 extsize) < 0) {
1982 munmap((void *)p->mmap, p->mmap_size);
1983 die(_("index file corrupt"));
1985 src_offset += 8;
1986 src_offset += extsize;
1989 return NULL;
1993 * A helper function that will load the specified range of cache entries
1994 * from the memory mapped file and add them to the given index.
1996 static unsigned long load_cache_entry_block(struct index_state *istate,
1997 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1998 unsigned long start_offset, const struct cache_entry *previous_ce)
2000 int i;
2001 unsigned long src_offset = start_offset;
2003 for (i = offset; i < offset + nr; i++) {
2004 struct ondisk_cache_entry *disk_ce;
2005 struct cache_entry *ce;
2006 unsigned long consumed;
2008 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
2009 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
2010 set_index_entry(istate, i, ce);
2012 src_offset += consumed;
2013 previous_ce = ce;
2015 return src_offset - start_offset;
2018 static unsigned long load_all_cache_entries(struct index_state *istate,
2019 const char *mmap, size_t mmap_size, unsigned long src_offset)
2021 unsigned long consumed;
2023 if (istate->version == 4) {
2024 mem_pool_init(&istate->ce_mem_pool,
2025 estimate_cache_size_from_compressed(istate->cache_nr));
2026 } else {
2027 mem_pool_init(&istate->ce_mem_pool,
2028 estimate_cache_size(mmap_size, istate->cache_nr));
2031 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2032 0, istate->cache_nr, mmap, src_offset, NULL);
2033 return consumed;
2037 * Mostly randomly chosen maximum thread counts: we
2038 * cap the parallelism to online_cpus() threads, and we want
2039 * to have at least 10000 cache entries per thread for it to
2040 * be worth starting a thread.
2043 #define THREAD_COST (10000)
2045 struct load_cache_entries_thread_data
2047 pthread_t pthread;
2048 struct index_state *istate;
2049 struct mem_pool *ce_mem_pool;
2050 int offset;
2051 const char *mmap;
2052 struct index_entry_offset_table *ieot;
2053 int ieot_start; /* starting index into the ieot array */
2054 int ieot_blocks; /* count of ieot entries to process */
2055 unsigned long consumed; /* return # of bytes in index file processed */
2059 * A thread proc to run the load_cache_entries() computation
2060 * across multiple background threads.
2062 static void *load_cache_entries_thread(void *_data)
2064 struct load_cache_entries_thread_data *p = _data;
2065 int i;
2067 /* iterate across all ieot blocks assigned to this thread */
2068 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2069 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2070 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2071 p->offset += p->ieot->entries[i].nr;
2073 return NULL;
2076 static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2077 int nr_threads, struct index_entry_offset_table *ieot)
2079 int i, offset, ieot_blocks, ieot_start, err;
2080 struct load_cache_entries_thread_data *data;
2081 unsigned long consumed = 0;
2083 /* a little sanity checking */
2084 if (istate->name_hash_initialized)
2085 BUG("the name hash isn't thread safe");
2087 mem_pool_init(&istate->ce_mem_pool, 0);
2089 /* ensure we have no more threads than we have blocks to process */
2090 if (nr_threads > ieot->nr)
2091 nr_threads = ieot->nr;
2092 data = xcalloc(nr_threads, sizeof(*data));
2094 offset = ieot_start = 0;
2095 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2096 for (i = 0; i < nr_threads; i++) {
2097 struct load_cache_entries_thread_data *p = &data[i];
2098 int nr, j;
2100 if (ieot_start + ieot_blocks > ieot->nr)
2101 ieot_blocks = ieot->nr - ieot_start;
2103 p->istate = istate;
2104 p->offset = offset;
2105 p->mmap = mmap;
2106 p->ieot = ieot;
2107 p->ieot_start = ieot_start;
2108 p->ieot_blocks = ieot_blocks;
2110 /* create a mem_pool for each thread */
2111 nr = 0;
2112 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2113 nr += p->ieot->entries[j].nr;
2114 if (istate->version == 4) {
2115 mem_pool_init(&p->ce_mem_pool,
2116 estimate_cache_size_from_compressed(nr));
2117 } else {
2118 mem_pool_init(&p->ce_mem_pool,
2119 estimate_cache_size(mmap_size, nr));
2122 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2123 if (err)
2124 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2126 /* increment by the number of cache entries in the ieot block being processed */
2127 for (j = 0; j < ieot_blocks; j++)
2128 offset += ieot->entries[ieot_start + j].nr;
2129 ieot_start += ieot_blocks;
2132 for (i = 0; i < nr_threads; i++) {
2133 struct load_cache_entries_thread_data *p = &data[i];
2135 err = pthread_join(p->pthread, NULL);
2136 if (err)
2137 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2138 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2139 consumed += p->consumed;
2142 free(data);
2144 return consumed;
2147 /* remember to discard_cache() before reading a different cache! */
2148 int do_read_index(struct index_state *istate, const char *path, int must_exist)
2150 int fd;
2151 struct stat st;
2152 unsigned long src_offset;
2153 const struct cache_header *hdr;
2154 const char *mmap;
2155 size_t mmap_size;
2156 struct load_index_extensions p;
2157 size_t extension_offset = 0;
2158 int nr_threads, cpus;
2159 struct index_entry_offset_table *ieot = NULL;
2161 if (istate->initialized)
2162 return istate->cache_nr;
2164 istate->timestamp.sec = 0;
2165 istate->timestamp.nsec = 0;
2166 fd = open(path, O_RDONLY);
2167 if (fd < 0) {
2168 if (!must_exist && errno == ENOENT)
2169 return 0;
2170 die_errno(_("%s: index file open failed"), path);
2173 if (fstat(fd, &st))
2174 die_errno(_("%s: cannot stat the open index"), path);
2176 mmap_size = xsize_t(st.st_size);
2177 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2178 die(_("%s: index file smaller than expected"), path);
2180 mmap = xmmap_gently(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2181 if (mmap == MAP_FAILED)
2182 die_errno(_("%s: unable to map index file"), path);
2183 close(fd);
2185 hdr = (const struct cache_header *)mmap;
2186 if (verify_hdr(hdr, mmap_size) < 0)
2187 goto unmap;
2189 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2190 istate->version = ntohl(hdr->hdr_version);
2191 istate->cache_nr = ntohl(hdr->hdr_entries);
2192 istate->cache_alloc = alloc_nr(istate->cache_nr);
2193 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2194 istate->initialized = 1;
2196 p.istate = istate;
2197 p.mmap = mmap;
2198 p.mmap_size = mmap_size;
2200 src_offset = sizeof(*hdr);
2202 if (git_config_get_index_threads(&nr_threads))
2203 nr_threads = 1;
2205 /* TODO: does creating more threads than cores help? */
2206 if (!nr_threads) {
2207 nr_threads = istate->cache_nr / THREAD_COST;
2208 cpus = online_cpus();
2209 if (nr_threads > cpus)
2210 nr_threads = cpus;
2213 if (!HAVE_THREADS)
2214 nr_threads = 1;
2216 if (nr_threads > 1) {
2217 extension_offset = read_eoie_extension(mmap, mmap_size);
2218 if (extension_offset) {
2219 int err;
2221 p.src_offset = extension_offset;
2222 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2223 if (err)
2224 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2226 nr_threads--;
2231 * Locate and read the index entry offset table so that we can use it
2232 * to multi-thread the reading of the cache entries.
2234 if (extension_offset && nr_threads > 1)
2235 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2237 if (ieot) {
2238 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, nr_threads, ieot);
2239 free(ieot);
2240 } else {
2241 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2244 istate->timestamp.sec = st.st_mtime;
2245 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2247 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2248 if (extension_offset) {
2249 int ret = pthread_join(p.pthread, NULL);
2250 if (ret)
2251 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2252 } else {
2253 p.src_offset = src_offset;
2254 load_index_extensions(&p);
2256 munmap((void *)mmap, mmap_size);
2259 * TODO trace2: replace "the_repository" with the actual repo instance
2260 * that is associated with the given "istate".
2262 trace2_data_intmax("index", the_repository, "read/version",
2263 istate->version);
2264 trace2_data_intmax("index", the_repository, "read/cache_nr",
2265 istate->cache_nr);
2267 return istate->cache_nr;
2269 unmap:
2270 munmap((void *)mmap, mmap_size);
2271 die(_("index file corrupt"));
2275 * Signal that the shared index is used by updating its mtime.
2277 * This way, shared index can be removed if they have not been used
2278 * for some time.
2280 static void freshen_shared_index(const char *shared_index, int warn)
2282 if (!check_and_freshen_file(shared_index, 1) && warn)
2283 warning(_("could not freshen shared index '%s'"), shared_index);
2286 int read_index_from(struct index_state *istate, const char *path,
2287 const char *gitdir)
2289 struct split_index *split_index;
2290 int ret;
2291 char *base_oid_hex;
2292 char *base_path;
2294 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2295 if (istate->initialized)
2296 return istate->cache_nr;
2299 * TODO trace2: replace "the_repository" with the actual repo instance
2300 * that is associated with the given "istate".
2302 trace2_region_enter_printf("index", "do_read_index", the_repository,
2303 "%s", path);
2304 trace_performance_enter();
2305 ret = do_read_index(istate, path, 0);
2306 trace_performance_leave("read cache %s", path);
2307 trace2_region_leave_printf("index", "do_read_index", the_repository,
2308 "%s", path);
2310 split_index = istate->split_index;
2311 if (!split_index || is_null_oid(&split_index->base_oid)) {
2312 post_read_index_from(istate);
2313 return ret;
2316 trace_performance_enter();
2317 if (split_index->base)
2318 discard_index(split_index->base);
2319 else
2320 split_index->base = xcalloc(1, sizeof(*split_index->base));
2322 base_oid_hex = oid_to_hex(&split_index->base_oid);
2323 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2324 trace2_region_enter_printf("index", "shared/do_read_index",
2325 the_repository, "%s", base_path);
2326 ret = do_read_index(split_index->base, base_path, 1);
2327 trace2_region_leave_printf("index", "shared/do_read_index",
2328 the_repository, "%s", base_path);
2329 if (!oideq(&split_index->base_oid, &split_index->base->oid))
2330 die(_("broken index, expect %s in %s, got %s"),
2331 base_oid_hex, base_path,
2332 oid_to_hex(&split_index->base->oid));
2334 freshen_shared_index(base_path, 0);
2335 merge_base_index(istate);
2336 post_read_index_from(istate);
2337 trace_performance_leave("read cache %s", base_path);
2338 free(base_path);
2339 return ret;
2342 int is_index_unborn(struct index_state *istate)
2344 return (!istate->cache_nr && !istate->timestamp.sec);
2347 int discard_index(struct index_state *istate)
2350 * Cache entries in istate->cache[] should have been allocated
2351 * from the memory pool associated with this index, or from an
2352 * associated split_index. There is no need to free individual
2353 * cache entries. validate_cache_entries can detect when this
2354 * assertion does not hold.
2356 validate_cache_entries(istate);
2358 resolve_undo_clear_index(istate);
2359 istate->cache_nr = 0;
2360 istate->cache_changed = 0;
2361 istate->timestamp.sec = 0;
2362 istate->timestamp.nsec = 0;
2363 free_name_hash(istate);
2364 cache_tree_free(&(istate->cache_tree));
2365 istate->initialized = 0;
2366 istate->fsmonitor_has_run_once = 0;
2367 FREE_AND_NULL(istate->cache);
2368 istate->cache_alloc = 0;
2369 discard_split_index(istate);
2370 free_untracked_cache(istate->untracked);
2371 istate->untracked = NULL;
2373 if (istate->ce_mem_pool) {
2374 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2375 istate->ce_mem_pool = NULL;
2378 return 0;
2382 * Validate the cache entries of this index.
2383 * All cache entries associated with this index
2384 * should have been allocated by the memory pool
2385 * associated with this index, or by a referenced
2386 * split index.
2388 void validate_cache_entries(const struct index_state *istate)
2390 int i;
2392 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2393 return;
2395 for (i = 0; i < istate->cache_nr; i++) {
2396 if (!istate) {
2397 BUG("cache entry is not allocated from expected memory pool");
2398 } else if (!istate->ce_mem_pool ||
2399 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2400 if (!istate->split_index ||
2401 !istate->split_index->base ||
2402 !istate->split_index->base->ce_mem_pool ||
2403 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2404 BUG("cache entry is not allocated from expected memory pool");
2409 if (istate->split_index)
2410 validate_cache_entries(istate->split_index->base);
2413 int unmerged_index(const struct index_state *istate)
2415 int i;
2416 for (i = 0; i < istate->cache_nr; i++) {
2417 if (ce_stage(istate->cache[i]))
2418 return 1;
2420 return 0;
2423 int repo_index_has_changes(struct repository *repo,
2424 struct tree *tree,
2425 struct strbuf *sb)
2427 struct index_state *istate = repo->index;
2428 struct object_id cmp;
2429 int i;
2431 if (tree)
2432 cmp = tree->object.oid;
2433 if (tree || !get_oid_tree("HEAD", &cmp)) {
2434 struct diff_options opt;
2436 repo_diff_setup(repo, &opt);
2437 opt.flags.exit_with_status = 1;
2438 if (!sb)
2439 opt.flags.quick = 1;
2440 do_diff_cache(&cmp, &opt);
2441 diffcore_std(&opt);
2442 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2443 if (i)
2444 strbuf_addch(sb, ' ');
2445 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2447 diff_flush(&opt);
2448 return opt.flags.has_changes != 0;
2449 } else {
2450 for (i = 0; sb && i < istate->cache_nr; i++) {
2451 if (i)
2452 strbuf_addch(sb, ' ');
2453 strbuf_addstr(sb, istate->cache[i]->name);
2455 return !!istate->cache_nr;
2459 #define WRITE_BUFFER_SIZE 8192
2460 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2461 static unsigned long write_buffer_len;
2463 static int ce_write_flush(git_hash_ctx *context, int fd)
2465 unsigned int buffered = write_buffer_len;
2466 if (buffered) {
2467 the_hash_algo->update_fn(context, write_buffer, buffered);
2468 if (write_in_full(fd, write_buffer, buffered) < 0)
2469 return -1;
2470 write_buffer_len = 0;
2472 return 0;
2475 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2477 while (len) {
2478 unsigned int buffered = write_buffer_len;
2479 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2480 if (partial > len)
2481 partial = len;
2482 memcpy(write_buffer + buffered, data, partial);
2483 buffered += partial;
2484 if (buffered == WRITE_BUFFER_SIZE) {
2485 write_buffer_len = buffered;
2486 if (ce_write_flush(context, fd))
2487 return -1;
2488 buffered = 0;
2490 write_buffer_len = buffered;
2491 len -= partial;
2492 data = (char *) data + partial;
2494 return 0;
2497 static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2498 int fd, unsigned int ext, unsigned int sz)
2500 ext = htonl(ext);
2501 sz = htonl(sz);
2502 if (eoie_context) {
2503 the_hash_algo->update_fn(eoie_context, &ext, 4);
2504 the_hash_algo->update_fn(eoie_context, &sz, 4);
2506 return ((ce_write(context, fd, &ext, 4) < 0) ||
2507 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2510 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2512 unsigned int left = write_buffer_len;
2514 if (left) {
2515 write_buffer_len = 0;
2516 the_hash_algo->update_fn(context, write_buffer, left);
2519 /* Flush first if not enough space for hash signature */
2520 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2521 if (write_in_full(fd, write_buffer, left) < 0)
2522 return -1;
2523 left = 0;
2526 /* Append the hash signature at the end */
2527 the_hash_algo->final_fn(write_buffer + left, context);
2528 hashcpy(hash, write_buffer + left);
2529 left += the_hash_algo->rawsz;
2530 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2533 static void ce_smudge_racily_clean_entry(struct index_state *istate,
2534 struct cache_entry *ce)
2537 * The only thing we care about in this function is to smudge the
2538 * falsely clean entry due to touch-update-touch race, so we leave
2539 * everything else as they are. We are called for entries whose
2540 * ce_stat_data.sd_mtime match the index file mtime.
2542 * Note that this actually does not do much for gitlinks, for
2543 * which ce_match_stat_basic() always goes to the actual
2544 * contents. The caller checks with is_racy_timestamp() which
2545 * always says "no" for gitlinks, so we are not called for them ;-)
2547 struct stat st;
2549 if (lstat(ce->name, &st) < 0)
2550 return;
2551 if (ce_match_stat_basic(ce, &st))
2552 return;
2553 if (ce_modified_check_fs(istate, ce, &st)) {
2554 /* This is "racily clean"; smudge it. Note that this
2555 * is a tricky code. At first glance, it may appear
2556 * that it can break with this sequence:
2558 * $ echo xyzzy >frotz
2559 * $ git-update-index --add frotz
2560 * $ : >frotz
2561 * $ sleep 3
2562 * $ echo filfre >nitfol
2563 * $ git-update-index --add nitfol
2565 * but it does not. When the second update-index runs,
2566 * it notices that the entry "frotz" has the same timestamp
2567 * as index, and if we were to smudge it by resetting its
2568 * size to zero here, then the object name recorded
2569 * in index is the 6-byte file but the cached stat information
2570 * becomes zero --- which would then match what we would
2571 * obtain from the filesystem next time we stat("frotz").
2573 * However, the second update-index, before calling
2574 * this function, notices that the cached size is 6
2575 * bytes and what is on the filesystem is an empty
2576 * file, and never calls us, so the cached size information
2577 * for "frotz" stays 6 which does not match the filesystem.
2579 ce->ce_stat_data.sd_size = 0;
2583 /* Copy miscellaneous fields but not the name */
2584 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2585 struct cache_entry *ce)
2587 short flags;
2588 const unsigned hashsz = the_hash_algo->rawsz;
2589 uint16_t *flagsp = (uint16_t *)(ondisk->data + hashsz);
2591 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2592 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2593 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2594 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2595 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2596 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2597 ondisk->mode = htonl(ce->ce_mode);
2598 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2599 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2600 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2601 hashcpy(ondisk->data, ce->oid.hash);
2603 flags = ce->ce_flags & ~CE_NAMEMASK;
2604 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2605 flagsp[0] = htons(flags);
2606 if (ce->ce_flags & CE_EXTENDED) {
2607 flagsp[1] = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2611 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2612 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2614 int size;
2615 int result;
2616 unsigned int saved_namelen;
2617 int stripped_name = 0;
2618 static unsigned char padding[8] = { 0x00 };
2620 if (ce->ce_flags & CE_STRIP_NAME) {
2621 saved_namelen = ce_namelen(ce);
2622 ce->ce_namelen = 0;
2623 stripped_name = 1;
2626 size = offsetof(struct ondisk_cache_entry,data) + ondisk_data_size(ce->ce_flags, 0);
2628 if (!previous_name) {
2629 int len = ce_namelen(ce);
2630 copy_cache_entry_to_ondisk(ondisk, ce);
2631 result = ce_write(c, fd, ondisk, size);
2632 if (!result)
2633 result = ce_write(c, fd, ce->name, len);
2634 if (!result)
2635 result = ce_write(c, fd, padding, align_padding_size(size, len));
2636 } else {
2637 int common, to_remove, prefix_size;
2638 unsigned char to_remove_vi[16];
2639 for (common = 0;
2640 (ce->name[common] &&
2641 common < previous_name->len &&
2642 ce->name[common] == previous_name->buf[common]);
2643 common++)
2644 ; /* still matching */
2645 to_remove = previous_name->len - common;
2646 prefix_size = encode_varint(to_remove, to_remove_vi);
2648 copy_cache_entry_to_ondisk(ondisk, ce);
2649 result = ce_write(c, fd, ondisk, size);
2650 if (!result)
2651 result = ce_write(c, fd, to_remove_vi, prefix_size);
2652 if (!result)
2653 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2654 if (!result)
2655 result = ce_write(c, fd, padding, 1);
2657 strbuf_splice(previous_name, common, to_remove,
2658 ce->name + common, ce_namelen(ce) - common);
2660 if (stripped_name) {
2661 ce->ce_namelen = saved_namelen;
2662 ce->ce_flags &= ~CE_STRIP_NAME;
2665 return result;
2669 * This function verifies if index_state has the correct sha1 of the
2670 * index file. Don't die if we have any other failure, just return 0.
2672 static int verify_index_from(const struct index_state *istate, const char *path)
2674 int fd;
2675 ssize_t n;
2676 struct stat st;
2677 unsigned char hash[GIT_MAX_RAWSZ];
2679 if (!istate->initialized)
2680 return 0;
2682 fd = open(path, O_RDONLY);
2683 if (fd < 0)
2684 return 0;
2686 if (fstat(fd, &st))
2687 goto out;
2689 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2690 goto out;
2692 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2693 if (n != the_hash_algo->rawsz)
2694 goto out;
2696 if (!hasheq(istate->oid.hash, hash))
2697 goto out;
2699 close(fd);
2700 return 1;
2702 out:
2703 close(fd);
2704 return 0;
2707 static int repo_verify_index(struct repository *repo)
2709 return verify_index_from(repo->index, repo->index_file);
2712 static int has_racy_timestamp(struct index_state *istate)
2714 int entries = istate->cache_nr;
2715 int i;
2717 for (i = 0; i < entries; i++) {
2718 struct cache_entry *ce = istate->cache[i];
2719 if (is_racy_timestamp(istate, ce))
2720 return 1;
2722 return 0;
2725 void repo_update_index_if_able(struct repository *repo,
2726 struct lock_file *lockfile)
2728 if ((repo->index->cache_changed ||
2729 has_racy_timestamp(repo->index)) &&
2730 repo_verify_index(repo))
2731 write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2732 else
2733 rollback_lock_file(lockfile);
2736 static int record_eoie(void)
2738 int val;
2740 if (!git_config_get_bool("index.recordendofindexentries", &val))
2741 return val;
2744 * As a convenience, the end of index entries extension
2745 * used for threading is written by default if the user
2746 * explicitly requested threaded index reads.
2748 return !git_config_get_index_threads(&val) && val != 1;
2751 static int record_ieot(void)
2753 int val;
2755 if (!git_config_get_bool("index.recordoffsettable", &val))
2756 return val;
2759 * As a convenience, the offset table used for threading is
2760 * written by default if the user explicitly requested
2761 * threaded index reads.
2763 return !git_config_get_index_threads(&val) && val != 1;
2767 * On success, `tempfile` is closed. If it is the temporary file
2768 * of a `struct lock_file`, we will therefore effectively perform
2769 * a 'close_lock_file_gently()`. Since that is an implementation
2770 * detail of lockfiles, callers of `do_write_index()` should not
2771 * rely on it.
2773 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2774 int strip_extensions)
2776 uint64_t start = getnanotime();
2777 int newfd = tempfile->fd;
2778 git_hash_ctx c, eoie_c;
2779 struct cache_header hdr;
2780 int i, err = 0, removed, extended, hdr_version;
2781 struct cache_entry **cache = istate->cache;
2782 int entries = istate->cache_nr;
2783 struct stat st;
2784 struct ondisk_cache_entry ondisk;
2785 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2786 int drop_cache_tree = istate->drop_cache_tree;
2787 off_t offset;
2788 int ieot_entries = 1;
2789 struct index_entry_offset_table *ieot = NULL;
2790 int nr, nr_threads;
2792 for (i = removed = extended = 0; i < entries; i++) {
2793 if (cache[i]->ce_flags & CE_REMOVE)
2794 removed++;
2796 /* reduce extended entries if possible */
2797 cache[i]->ce_flags &= ~CE_EXTENDED;
2798 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2799 extended++;
2800 cache[i]->ce_flags |= CE_EXTENDED;
2804 if (!istate->version) {
2805 istate->version = get_index_format_default(the_repository);
2806 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2807 init_split_index(istate);
2810 /* demote version 3 to version 2 when the latter suffices */
2811 if (istate->version == 3 || istate->version == 2)
2812 istate->version = extended ? 3 : 2;
2814 hdr_version = istate->version;
2816 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2817 hdr.hdr_version = htonl(hdr_version);
2818 hdr.hdr_entries = htonl(entries - removed);
2820 the_hash_algo->init_fn(&c);
2821 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2822 return -1;
2824 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))
2825 nr_threads = 1;
2827 if (nr_threads != 1 && record_ieot()) {
2828 int ieot_blocks, cpus;
2831 * ensure default number of ieot blocks maps evenly to the
2832 * default number of threads that will process them leaving
2833 * room for the thread to load the index extensions.
2835 if (!nr_threads) {
2836 ieot_blocks = istate->cache_nr / THREAD_COST;
2837 cpus = online_cpus();
2838 if (ieot_blocks > cpus - 1)
2839 ieot_blocks = cpus - 1;
2840 } else {
2841 ieot_blocks = nr_threads;
2842 if (ieot_blocks > istate->cache_nr)
2843 ieot_blocks = istate->cache_nr;
2847 * no reason to write out the IEOT extension if we don't
2848 * have enough blocks to utilize multi-threading
2850 if (ieot_blocks > 1) {
2851 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2852 + (ieot_blocks * sizeof(struct index_entry_offset)));
2853 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2857 offset = lseek(newfd, 0, SEEK_CUR);
2858 if (offset < 0) {
2859 free(ieot);
2860 return -1;
2862 offset += write_buffer_len;
2863 nr = 0;
2864 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2866 for (i = 0; i < entries; i++) {
2867 struct cache_entry *ce = cache[i];
2868 if (ce->ce_flags & CE_REMOVE)
2869 continue;
2870 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2871 ce_smudge_racily_clean_entry(istate, ce);
2872 if (is_null_oid(&ce->oid)) {
2873 static const char msg[] = "cache entry has null sha1: %s";
2874 static int allow = -1;
2876 if (allow < 0)
2877 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2878 if (allow)
2879 warning(msg, ce->name);
2880 else
2881 err = error(msg, ce->name);
2883 drop_cache_tree = 1;
2885 if (ieot && i && (i % ieot_entries == 0)) {
2886 ieot->entries[ieot->nr].nr = nr;
2887 ieot->entries[ieot->nr].offset = offset;
2888 ieot->nr++;
2890 * If we have a V4 index, set the first byte to an invalid
2891 * character to ensure there is nothing common with the previous
2892 * entry
2894 if (previous_name)
2895 previous_name->buf[0] = 0;
2896 nr = 0;
2897 offset = lseek(newfd, 0, SEEK_CUR);
2898 if (offset < 0) {
2899 free(ieot);
2900 return -1;
2902 offset += write_buffer_len;
2904 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2905 err = -1;
2907 if (err)
2908 break;
2909 nr++;
2911 if (ieot && nr) {
2912 ieot->entries[ieot->nr].nr = nr;
2913 ieot->entries[ieot->nr].offset = offset;
2914 ieot->nr++;
2916 strbuf_release(&previous_name_buf);
2918 if (err) {
2919 free(ieot);
2920 return err;
2923 /* Write extension data here */
2924 offset = lseek(newfd, 0, SEEK_CUR);
2925 if (offset < 0) {
2926 free(ieot);
2927 return -1;
2929 offset += write_buffer_len;
2930 the_hash_algo->init_fn(&eoie_c);
2933 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2934 * can minimize the number of extensions we have to scan through to
2935 * find it during load. Write it out regardless of the
2936 * strip_extensions parameter as we need it when loading the shared
2937 * index.
2939 if (ieot) {
2940 struct strbuf sb = STRBUF_INIT;
2942 write_ieot_extension(&sb, ieot);
2943 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2944 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2945 strbuf_release(&sb);
2946 free(ieot);
2947 if (err)
2948 return -1;
2951 if (!strip_extensions && istate->split_index &&
2952 !is_null_oid(&istate->split_index->base_oid)) {
2953 struct strbuf sb = STRBUF_INIT;
2955 err = write_link_extension(&sb, istate) < 0 ||
2956 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2957 sb.len) < 0 ||
2958 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2959 strbuf_release(&sb);
2960 if (err)
2961 return -1;
2963 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2964 struct strbuf sb = STRBUF_INIT;
2966 cache_tree_write(&sb, istate->cache_tree);
2967 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2968 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2969 strbuf_release(&sb);
2970 if (err)
2971 return -1;
2973 if (!strip_extensions && istate->resolve_undo) {
2974 struct strbuf sb = STRBUF_INIT;
2976 resolve_undo_write(&sb, istate->resolve_undo);
2977 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2978 sb.len) < 0
2979 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2980 strbuf_release(&sb);
2981 if (err)
2982 return -1;
2984 if (!strip_extensions && istate->untracked) {
2985 struct strbuf sb = STRBUF_INIT;
2987 write_untracked_extension(&sb, istate->untracked);
2988 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2989 sb.len) < 0 ||
2990 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2991 strbuf_release(&sb);
2992 if (err)
2993 return -1;
2995 if (!strip_extensions && istate->fsmonitor_last_update) {
2996 struct strbuf sb = STRBUF_INIT;
2998 write_fsmonitor_extension(&sb, istate);
2999 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
3000 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
3001 strbuf_release(&sb);
3002 if (err)
3003 return -1;
3007 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3008 * so that it can be found and processed before all the index entries are
3009 * read. Write it out regardless of the strip_extensions parameter as we need it
3010 * when loading the shared index.
3012 if (offset && record_eoie()) {
3013 struct strbuf sb = STRBUF_INIT;
3015 write_eoie_extension(&sb, &eoie_c, offset);
3016 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
3017 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
3018 strbuf_release(&sb);
3019 if (err)
3020 return -1;
3023 if (ce_flush(&c, newfd, istate->oid.hash))
3024 return -1;
3025 if (close_tempfile_gently(tempfile)) {
3026 error(_("could not close '%s'"), tempfile->filename.buf);
3027 return -1;
3029 if (stat(tempfile->filename.buf, &st))
3030 return -1;
3031 istate->timestamp.sec = (unsigned int)st.st_mtime;
3032 istate->timestamp.nsec = ST_MTIME_NSEC(st);
3033 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
3036 * TODO trace2: replace "the_repository" with the actual repo instance
3037 * that is associated with the given "istate".
3039 trace2_data_intmax("index", the_repository, "write/version",
3040 istate->version);
3041 trace2_data_intmax("index", the_repository, "write/cache_nr",
3042 istate->cache_nr);
3044 return 0;
3047 void set_alternate_index_output(const char *name)
3049 alternate_index_output = name;
3052 static int commit_locked_index(struct lock_file *lk)
3054 if (alternate_index_output)
3055 return commit_lock_file_to(lk, alternate_index_output);
3056 else
3057 return commit_lock_file(lk);
3060 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
3061 unsigned flags)
3063 int ret;
3066 * TODO trace2: replace "the_repository" with the actual repo instance
3067 * that is associated with the given "istate".
3069 trace2_region_enter_printf("index", "do_write_index", the_repository,
3070 "%s", lock->tempfile->filename.buf);
3071 ret = do_write_index(istate, lock->tempfile, 0);
3072 trace2_region_leave_printf("index", "do_write_index", the_repository,
3073 "%s", lock->tempfile->filename.buf);
3075 if (ret)
3076 return ret;
3077 if (flags & COMMIT_LOCK)
3078 ret = commit_locked_index(lock);
3079 else
3080 ret = close_lock_file_gently(lock);
3082 run_hook_le(NULL, "post-index-change",
3083 istate->updated_workdir ? "1" : "0",
3084 istate->updated_skipworktree ? "1" : "0", NULL);
3085 istate->updated_workdir = 0;
3086 istate->updated_skipworktree = 0;
3088 return ret;
3091 static int write_split_index(struct index_state *istate,
3092 struct lock_file *lock,
3093 unsigned flags)
3095 int ret;
3096 prepare_to_write_split_index(istate);
3097 ret = do_write_locked_index(istate, lock, flags);
3098 finish_writing_split_index(istate);
3099 return ret;
3102 static const char *shared_index_expire = "2.weeks.ago";
3104 static unsigned long get_shared_index_expire_date(void)
3106 static unsigned long shared_index_expire_date;
3107 static int shared_index_expire_date_prepared;
3109 if (!shared_index_expire_date_prepared) {
3110 git_config_get_expiry("splitindex.sharedindexexpire",
3111 &shared_index_expire);
3112 shared_index_expire_date = approxidate(shared_index_expire);
3113 shared_index_expire_date_prepared = 1;
3116 return shared_index_expire_date;
3119 static int should_delete_shared_index(const char *shared_index_path)
3121 struct stat st;
3122 unsigned long expiration;
3124 /* Check timestamp */
3125 expiration = get_shared_index_expire_date();
3126 if (!expiration)
3127 return 0;
3128 if (stat(shared_index_path, &st))
3129 return error_errno(_("could not stat '%s'"), shared_index_path);
3130 if (st.st_mtime > expiration)
3131 return 0;
3133 return 1;
3136 static int clean_shared_index_files(const char *current_hex)
3138 struct dirent *de;
3139 DIR *dir = opendir(get_git_dir());
3141 if (!dir)
3142 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3144 while ((de = readdir(dir)) != NULL) {
3145 const char *sha1_hex;
3146 const char *shared_index_path;
3147 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3148 continue;
3149 if (!strcmp(sha1_hex, current_hex))
3150 continue;
3151 shared_index_path = git_path("%s", de->d_name);
3152 if (should_delete_shared_index(shared_index_path) > 0 &&
3153 unlink(shared_index_path))
3154 warning_errno(_("unable to unlink: %s"), shared_index_path);
3156 closedir(dir);
3158 return 0;
3161 static int write_shared_index(struct index_state *istate,
3162 struct tempfile **temp)
3164 struct split_index *si = istate->split_index;
3165 int ret;
3167 move_cache_to_base_index(istate);
3169 trace2_region_enter_printf("index", "shared/do_write_index",
3170 the_repository, "%s", (*temp)->filename.buf);
3171 ret = do_write_index(si->base, *temp, 1);
3172 trace2_region_leave_printf("index", "shared/do_write_index",
3173 the_repository, "%s", (*temp)->filename.buf);
3175 if (ret)
3176 return ret;
3177 ret = adjust_shared_perm(get_tempfile_path(*temp));
3178 if (ret) {
3179 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3180 return ret;
3182 ret = rename_tempfile(temp,
3183 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3184 if (!ret) {
3185 oidcpy(&si->base_oid, &si->base->oid);
3186 clean_shared_index_files(oid_to_hex(&si->base->oid));
3189 return ret;
3192 static const int default_max_percent_split_change = 20;
3194 static int too_many_not_shared_entries(struct index_state *istate)
3196 int i, not_shared = 0;
3197 int max_split = git_config_get_max_percent_split_change();
3199 switch (max_split) {
3200 case -1:
3201 /* not or badly configured: use the default value */
3202 max_split = default_max_percent_split_change;
3203 break;
3204 case 0:
3205 return 1; /* 0% means always write a new shared index */
3206 case 100:
3207 return 0; /* 100% means never write a new shared index */
3208 default:
3209 break; /* just use the configured value */
3212 /* Count not shared entries */
3213 for (i = 0; i < istate->cache_nr; i++) {
3214 struct cache_entry *ce = istate->cache[i];
3215 if (!ce->index)
3216 not_shared++;
3219 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3222 int write_locked_index(struct index_state *istate, struct lock_file *lock,
3223 unsigned flags)
3225 int new_shared_index, ret;
3226 struct split_index *si = istate->split_index;
3228 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3229 cache_tree_verify(the_repository, istate);
3231 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3232 if (flags & COMMIT_LOCK)
3233 rollback_lock_file(lock);
3234 return 0;
3237 if (istate->fsmonitor_last_update)
3238 fill_fsmonitor_bitmap(istate);
3240 if (!si || alternate_index_output ||
3241 (istate->cache_changed & ~EXTMASK)) {
3242 if (si)
3243 oidclr(&si->base_oid);
3244 ret = do_write_locked_index(istate, lock, flags);
3245 goto out;
3248 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3249 int v = si->base_oid.hash[0];
3250 if ((v & 15) < 6)
3251 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3253 if (too_many_not_shared_entries(istate))
3254 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3256 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3258 if (new_shared_index) {
3259 struct tempfile *temp;
3260 int saved_errno;
3262 /* Same initial permissions as the main .git/index file */
3263 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3264 if (!temp) {
3265 oidclr(&si->base_oid);
3266 ret = do_write_locked_index(istate, lock, flags);
3267 goto out;
3269 ret = write_shared_index(istate, &temp);
3271 saved_errno = errno;
3272 if (is_tempfile_active(temp))
3273 delete_tempfile(&temp);
3274 errno = saved_errno;
3276 if (ret)
3277 goto out;
3280 ret = write_split_index(istate, lock, flags);
3282 /* Freshen the shared index only if the split-index was written */
3283 if (!ret && !new_shared_index && !is_null_oid(&si->base_oid)) {
3284 const char *shared_index = git_path("sharedindex.%s",
3285 oid_to_hex(&si->base_oid));
3286 freshen_shared_index(shared_index, 1);
3289 out:
3290 if (flags & COMMIT_LOCK)
3291 rollback_lock_file(lock);
3292 return ret;
3296 * Read the index file that is potentially unmerged into given
3297 * index_state, dropping any unmerged entries to stage #0 (potentially
3298 * resulting in a path appearing as both a file and a directory in the
3299 * index; the caller is responsible to clear out the extra entries
3300 * before writing the index to a tree). Returns true if the index is
3301 * unmerged. Callers who want to refuse to work from an unmerged
3302 * state can call this and check its return value, instead of calling
3303 * read_cache().
3305 int repo_read_index_unmerged(struct repository *repo)
3307 struct index_state *istate;
3308 int i;
3309 int unmerged = 0;
3311 repo_read_index(repo);
3312 istate = repo->index;
3313 for (i = 0; i < istate->cache_nr; i++) {
3314 struct cache_entry *ce = istate->cache[i];
3315 struct cache_entry *new_ce;
3316 int len;
3318 if (!ce_stage(ce))
3319 continue;
3320 unmerged = 1;
3321 len = ce_namelen(ce);
3322 new_ce = make_empty_cache_entry(istate, len);
3323 memcpy(new_ce->name, ce->name, len);
3324 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3325 new_ce->ce_namelen = len;
3326 new_ce->ce_mode = ce->ce_mode;
3327 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3328 return error(_("%s: cannot drop to stage #0"),
3329 new_ce->name);
3331 return unmerged;
3335 * Returns 1 if the path is an "other" path with respect to
3336 * the index; that is, the path is not mentioned in the index at all,
3337 * either as a file, a directory with some files in the index,
3338 * or as an unmerged entry.
3340 * We helpfully remove a trailing "/" from directories so that
3341 * the output of read_directory can be used as-is.
3343 int index_name_is_other(const struct index_state *istate, const char *name,
3344 int namelen)
3346 int pos;
3347 if (namelen && name[namelen - 1] == '/')
3348 namelen--;
3349 pos = index_name_pos(istate, name, namelen);
3350 if (0 <= pos)
3351 return 0; /* exact match */
3352 pos = -pos - 1;
3353 if (pos < istate->cache_nr) {
3354 struct cache_entry *ce = istate->cache[pos];
3355 if (ce_namelen(ce) == namelen &&
3356 !memcmp(ce->name, name, namelen))
3357 return 0; /* Yup, this one exists unmerged */
3359 return 1;
3362 void *read_blob_data_from_index(const struct index_state *istate,
3363 const char *path, unsigned long *size)
3365 int pos, len;
3366 unsigned long sz;
3367 enum object_type type;
3368 void *data;
3370 len = strlen(path);
3371 pos = index_name_pos(istate, path, len);
3372 if (pos < 0) {
3374 * We might be in the middle of a merge, in which
3375 * case we would read stage #2 (ours).
3377 int i;
3378 for (i = -pos - 1;
3379 (pos < 0 && i < istate->cache_nr &&
3380 !strcmp(istate->cache[i]->name, path));
3381 i++)
3382 if (ce_stage(istate->cache[i]) == 2)
3383 pos = i;
3385 if (pos < 0)
3386 return NULL;
3387 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3388 if (!data || type != OBJ_BLOB) {
3389 free(data);
3390 return NULL;
3392 if (size)
3393 *size = sz;
3394 return data;
3397 void stat_validity_clear(struct stat_validity *sv)
3399 FREE_AND_NULL(sv->sd);
3402 int stat_validity_check(struct stat_validity *sv, const char *path)
3404 struct stat st;
3406 if (stat(path, &st) < 0)
3407 return sv->sd == NULL;
3408 if (!sv->sd)
3409 return 0;
3410 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3413 void stat_validity_update(struct stat_validity *sv, int fd)
3415 struct stat st;
3417 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3418 stat_validity_clear(sv);
3419 else {
3420 if (!sv->sd)
3421 sv->sd = xcalloc(1, sizeof(struct stat_data));
3422 fill_stat_data(sv->sd, &st);
3426 void move_index_extensions(struct index_state *dst, struct index_state *src)
3428 dst->untracked = src->untracked;
3429 src->untracked = NULL;
3430 dst->cache_tree = src->cache_tree;
3431 src->cache_tree = NULL;
3434 struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3435 struct index_state *istate)
3437 unsigned int size = ce_size(ce);
3438 int mem_pool_allocated;
3439 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3440 mem_pool_allocated = new_entry->mem_pool_allocated;
3442 memcpy(new_entry, ce, size);
3443 new_entry->mem_pool_allocated = mem_pool_allocated;
3444 return new_entry;
3447 void discard_cache_entry(struct cache_entry *ce)
3449 if (ce && should_validate_cache_entries())
3450 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3452 if (ce && ce->mem_pool_allocated)
3453 return;
3455 free(ce);
3458 int should_validate_cache_entries(void)
3460 static int validate_index_cache_entries = -1;
3462 if (validate_index_cache_entries < 0) {
3463 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3464 validate_index_cache_entries = 1;
3465 else
3466 validate_index_cache_entries = 0;
3469 return validate_index_cache_entries;
3472 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3473 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3475 static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3478 * The end of index entries (EOIE) extension is guaranteed to be last
3479 * so that it can be found by scanning backwards from the EOF.
3481 * "EOIE"
3482 * <4-byte length>
3483 * <4-byte offset>
3484 * <20-byte hash>
3486 const char *index, *eoie;
3487 uint32_t extsize;
3488 size_t offset, src_offset;
3489 unsigned char hash[GIT_MAX_RAWSZ];
3490 git_hash_ctx c;
3492 /* ensure we have an index big enough to contain an EOIE extension */
3493 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3494 return 0;
3496 /* validate the extension signature */
3497 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3498 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3499 return 0;
3500 index += sizeof(uint32_t);
3502 /* validate the extension size */
3503 extsize = get_be32(index);
3504 if (extsize != EOIE_SIZE)
3505 return 0;
3506 index += sizeof(uint32_t);
3509 * Validate the offset we're going to look for the first extension
3510 * signature is after the index header and before the eoie extension.
3512 offset = get_be32(index);
3513 if (mmap + offset < mmap + sizeof(struct cache_header))
3514 return 0;
3515 if (mmap + offset >= eoie)
3516 return 0;
3517 index += sizeof(uint32_t);
3520 * The hash is computed over extension types and their sizes (but not
3521 * their contents). E.g. if we have "TREE" extension that is N-bytes
3522 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3523 * then the hash would be:
3525 * SHA-1("TREE" + <binary representation of N> +
3526 * "REUC" + <binary representation of M>)
3528 src_offset = offset;
3529 the_hash_algo->init_fn(&c);
3530 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3531 /* After an array of active_nr index entries,
3532 * there can be arbitrary number of extended
3533 * sections, each of which is prefixed with
3534 * extension name (4-byte) and section length
3535 * in 4-byte network byte order.
3537 uint32_t extsize;
3538 memcpy(&extsize, mmap + src_offset + 4, 4);
3539 extsize = ntohl(extsize);
3541 /* verify the extension size isn't so large it will wrap around */
3542 if (src_offset + 8 + extsize < src_offset)
3543 return 0;
3545 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3547 src_offset += 8;
3548 src_offset += extsize;
3550 the_hash_algo->final_fn(hash, &c);
3551 if (!hasheq(hash, (const unsigned char *)index))
3552 return 0;
3554 /* Validate that the extension offsets returned us back to the eoie extension. */
3555 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3556 return 0;
3558 return offset;
3561 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3563 uint32_t buffer;
3564 unsigned char hash[GIT_MAX_RAWSZ];
3566 /* offset */
3567 put_be32(&buffer, offset);
3568 strbuf_add(sb, &buffer, sizeof(uint32_t));
3570 /* hash */
3571 the_hash_algo->final_fn(hash, eoie_context);
3572 strbuf_add(sb, hash, the_hash_algo->rawsz);
3575 #define IEOT_VERSION (1)
3577 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3579 const char *index = NULL;
3580 uint32_t extsize, ext_version;
3581 struct index_entry_offset_table *ieot;
3582 int i, nr;
3584 /* find the IEOT extension */
3585 if (!offset)
3586 return NULL;
3587 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3588 extsize = get_be32(mmap + offset + 4);
3589 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3590 index = mmap + offset + 4 + 4;
3591 break;
3593 offset += 8;
3594 offset += extsize;
3596 if (!index)
3597 return NULL;
3599 /* validate the version is IEOT_VERSION */
3600 ext_version = get_be32(index);
3601 if (ext_version != IEOT_VERSION) {
3602 error("invalid IEOT version %d", ext_version);
3603 return NULL;
3605 index += sizeof(uint32_t);
3607 /* extension size - version bytes / bytes per entry */
3608 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3609 if (!nr) {
3610 error("invalid number of IEOT entries %d", nr);
3611 return NULL;
3613 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3614 + (nr * sizeof(struct index_entry_offset)));
3615 ieot->nr = nr;
3616 for (i = 0; i < nr; i++) {
3617 ieot->entries[i].offset = get_be32(index);
3618 index += sizeof(uint32_t);
3619 ieot->entries[i].nr = get_be32(index);
3620 index += sizeof(uint32_t);
3623 return ieot;
3626 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3628 uint32_t buffer;
3629 int i;
3631 /* version */
3632 put_be32(&buffer, IEOT_VERSION);
3633 strbuf_add(sb, &buffer, sizeof(uint32_t));
3635 /* ieot */
3636 for (i = 0; i < ieot->nr; i++) {
3638 /* offset */
3639 put_be32(&buffer, ieot->entries[i].offset);
3640 strbuf_add(sb, &buffer, sizeof(uint32_t));
3642 /* count */
3643 put_be32(&buffer, ieot->entries[i].nr);
3644 strbuf_add(sb, &buffer, sizeof(uint32_t));