0e0c93edc9be5a7af5de1097eafe9d860e7d183e
[git/git.git] / read-cache.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #include "cache.h"
7 #include "config.h"
8 #include "diff.h"
9 #include "diffcore.h"
10 #include "tempfile.h"
11 #include "lockfile.h"
12 #include "cache-tree.h"
13 #include "refs.h"
14 #include "dir.h"
15 #include "object-store.h"
16 #include "tree.h"
17 #include "commit.h"
18 #include "blob.h"
19 #include "resolve-undo.h"
20 #include "strbuf.h"
21 #include "varint.h"
22 #include "split-index.h"
23 #include "utf8.h"
24 #include "fsmonitor.h"
25 #include "thread-utils.h"
26 #include "progress.h"
27
28 /* Mask for the name length in ce_flags in the on-disk index */
29
30 #define CE_NAMEMASK (0x0fff)
31
32 /* Index extensions.
33 *
34 * The first letter should be 'A'..'Z' for extensions that are not
35 * necessary for a correct operation (i.e. optimization data).
36 * When new extensions are added that _needs_ to be understood in
37 * order to correctly interpret the index file, pick character that
38 * is outside the range, to cause the reader to abort.
39 */
40
41 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
42 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
43 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
44 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
45 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
46 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
47 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
48 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
49
50 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
51 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
52 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
53 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
54
55
56 /*
57 * This is an estimate of the pathname length in the index. We use
58 * this for V4 index files to guess the un-deltafied size of the index
59 * in memory because of pathname deltafication. This is not required
60 * for V2/V3 index formats because their pathnames are not compressed.
61 * If the initial amount of memory set aside is not sufficient, the
62 * mem pool will allocate extra memory.
63 */
64 #define CACHE_ENTRY_PATH_LENGTH 80
65
66 static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
67 {
68 struct cache_entry *ce;
69 ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
70 ce->mem_pool_allocated = 1;
71 return ce;
72 }
73
74 static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
75 {
76 struct cache_entry * ce;
77 ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
78 ce->mem_pool_allocated = 1;
79 return ce;
80 }
81
82 static struct mem_pool *find_mem_pool(struct index_state *istate)
83 {
84 struct mem_pool **pool_ptr;
85
86 if (istate->split_index && istate->split_index->base)
87 pool_ptr = &istate->split_index->base->ce_mem_pool;
88 else
89 pool_ptr = &istate->ce_mem_pool;
90
91 if (!*pool_ptr)
92 mem_pool_init(pool_ptr, 0);
93
94 return *pool_ptr;
95 }
96
97 static const char *alternate_index_output;
98
99 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
100 {
101 istate->cache[nr] = ce;
102 add_name_hash(istate, ce);
103 }
104
105 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
106 {
107 struct cache_entry *old = istate->cache[nr];
108
109 replace_index_entry_in_base(istate, old, ce);
110 remove_name_hash(istate, old);
111 discard_cache_entry(old);
112 ce->ce_flags &= ~CE_HASHED;
113 set_index_entry(istate, nr, ce);
114 ce->ce_flags |= CE_UPDATE_IN_BASE;
115 mark_fsmonitor_invalid(istate, ce);
116 istate->cache_changed |= CE_ENTRY_CHANGED;
117 }
118
119 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
120 {
121 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
122 int namelen = strlen(new_name);
123
124 new_entry = make_empty_cache_entry(istate, namelen);
125 copy_cache_entry(new_entry, old_entry);
126 new_entry->ce_flags &= ~CE_HASHED;
127 new_entry->ce_namelen = namelen;
128 new_entry->index = 0;
129 memcpy(new_entry->name, new_name, namelen + 1);
130
131 cache_tree_invalidate_path(istate, old_entry->name);
132 untracked_cache_remove_from_index(istate, old_entry->name);
133 remove_index_entry_at(istate, nr);
134 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
135 }
136
137 void fill_stat_data(struct stat_data *sd, struct stat *st)
138 {
139 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
140 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
141 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
142 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
143 sd->sd_dev = st->st_dev;
144 sd->sd_ino = st->st_ino;
145 sd->sd_uid = st->st_uid;
146 sd->sd_gid = st->st_gid;
147 sd->sd_size = st->st_size;
148 }
149
150 int match_stat_data(const struct stat_data *sd, struct stat *st)
151 {
152 int changed = 0;
153
154 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
155 changed |= MTIME_CHANGED;
156 if (trust_ctime && check_stat &&
157 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
158 changed |= CTIME_CHANGED;
159
160 #ifdef USE_NSEC
161 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
162 changed |= MTIME_CHANGED;
163 if (trust_ctime && check_stat &&
164 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
165 changed |= CTIME_CHANGED;
166 #endif
167
168 if (check_stat) {
169 if (sd->sd_uid != (unsigned int) st->st_uid ||
170 sd->sd_gid != (unsigned int) st->st_gid)
171 changed |= OWNER_CHANGED;
172 if (sd->sd_ino != (unsigned int) st->st_ino)
173 changed |= INODE_CHANGED;
174 }
175
176 #ifdef USE_STDEV
177 /*
178 * st_dev breaks on network filesystems where different
179 * clients will have different views of what "device"
180 * the filesystem is on
181 */
182 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
183 changed |= INODE_CHANGED;
184 #endif
185
186 if (sd->sd_size != (unsigned int) st->st_size)
187 changed |= DATA_CHANGED;
188
189 return changed;
190 }
191
192 /*
193 * This only updates the "non-critical" parts of the directory
194 * cache, ie the parts that aren't tracked by GIT, and only used
195 * to validate the cache.
196 */
197 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
198 {
199 fill_stat_data(&ce->ce_stat_data, st);
200
201 if (assume_unchanged)
202 ce->ce_flags |= CE_VALID;
203
204 if (S_ISREG(st->st_mode)) {
205 ce_mark_uptodate(ce);
206 mark_fsmonitor_valid(ce);
207 }
208 }
209
210 static int ce_compare_data(struct index_state *istate,
211 const struct cache_entry *ce,
212 struct stat *st)
213 {
214 int match = -1;
215 int fd = git_open_cloexec(ce->name, O_RDONLY);
216
217 if (fd >= 0) {
218 struct object_id oid;
219 if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
220 match = !oideq(&oid, &ce->oid);
221 /* index_fd() closed the file descriptor already */
222 }
223 return match;
224 }
225
226 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
227 {
228 int match = -1;
229 void *buffer;
230 unsigned long size;
231 enum object_type type;
232 struct strbuf sb = STRBUF_INIT;
233
234 if (strbuf_readlink(&sb, ce->name, expected_size))
235 return -1;
236
237 buffer = read_object_file(&ce->oid, &type, &size);
238 if (buffer) {
239 if (size == sb.len)
240 match = memcmp(buffer, sb.buf, size);
241 free(buffer);
242 }
243 strbuf_release(&sb);
244 return match;
245 }
246
247 static int ce_compare_gitlink(const struct cache_entry *ce)
248 {
249 struct object_id oid;
250
251 /*
252 * We don't actually require that the .git directory
253 * under GITLINK directory be a valid git directory. It
254 * might even be missing (in case nobody populated that
255 * sub-project).
256 *
257 * If so, we consider it always to match.
258 */
259 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
260 return 0;
261 return !oideq(&oid, &ce->oid);
262 }
263
264 static int ce_modified_check_fs(struct index_state *istate,
265 const struct cache_entry *ce,
266 struct stat *st)
267 {
268 switch (st->st_mode & S_IFMT) {
269 case S_IFREG:
270 if (ce_compare_data(istate, ce, st))
271 return DATA_CHANGED;
272 break;
273 case S_IFLNK:
274 if (ce_compare_link(ce, xsize_t(st->st_size)))
275 return DATA_CHANGED;
276 break;
277 case S_IFDIR:
278 if (S_ISGITLINK(ce->ce_mode))
279 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
280 /* else fallthrough */
281 default:
282 return TYPE_CHANGED;
283 }
284 return 0;
285 }
286
287 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
288 {
289 unsigned int changed = 0;
290
291 if (ce->ce_flags & CE_REMOVE)
292 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
293
294 switch (ce->ce_mode & S_IFMT) {
295 case S_IFREG:
296 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
297 /* We consider only the owner x bit to be relevant for
298 * "mode changes"
299 */
300 if (trust_executable_bit &&
301 (0100 & (ce->ce_mode ^ st->st_mode)))
302 changed |= MODE_CHANGED;
303 break;
304 case S_IFLNK:
305 if (!S_ISLNK(st->st_mode) &&
306 (has_symlinks || !S_ISREG(st->st_mode)))
307 changed |= TYPE_CHANGED;
308 break;
309 case S_IFGITLINK:
310 /* We ignore most of the st_xxx fields for gitlinks */
311 if (!S_ISDIR(st->st_mode))
312 changed |= TYPE_CHANGED;
313 else if (ce_compare_gitlink(ce))
314 changed |= DATA_CHANGED;
315 return changed;
316 default:
317 BUG("unsupported ce_mode: %o", ce->ce_mode);
318 }
319
320 changed |= match_stat_data(&ce->ce_stat_data, st);
321
322 /* Racily smudged entry? */
323 if (!ce->ce_stat_data.sd_size) {
324 if (!is_empty_blob_sha1(ce->oid.hash))
325 changed |= DATA_CHANGED;
326 }
327
328 return changed;
329 }
330
331 static int is_racy_stat(const struct index_state *istate,
332 const struct stat_data *sd)
333 {
334 return (istate->timestamp.sec &&
335 #ifdef USE_NSEC
336 /* nanosecond timestamped files can also be racy! */
337 (istate->timestamp.sec < sd->sd_mtime.sec ||
338 (istate->timestamp.sec == sd->sd_mtime.sec &&
339 istate->timestamp.nsec <= sd->sd_mtime.nsec))
340 #else
341 istate->timestamp.sec <= sd->sd_mtime.sec
342 #endif
343 );
344 }
345
346 int is_racy_timestamp(const struct index_state *istate,
347 const struct cache_entry *ce)
348 {
349 return (!S_ISGITLINK(ce->ce_mode) &&
350 is_racy_stat(istate, &ce->ce_stat_data));
351 }
352
353 int match_stat_data_racy(const struct index_state *istate,
354 const struct stat_data *sd, struct stat *st)
355 {
356 if (is_racy_stat(istate, sd))
357 return MTIME_CHANGED;
358 return match_stat_data(sd, st);
359 }
360
361 int ie_match_stat(struct index_state *istate,
362 const struct cache_entry *ce, struct stat *st,
363 unsigned int options)
364 {
365 unsigned int changed;
366 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
367 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
368 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
369 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
370
371 if (!ignore_fsmonitor)
372 refresh_fsmonitor(istate);
373 /*
374 * If it's marked as always valid in the index, it's
375 * valid whatever the checked-out copy says.
376 *
377 * skip-worktree has the same effect with higher precedence
378 */
379 if (!ignore_skip_worktree && ce_skip_worktree(ce))
380 return 0;
381 if (!ignore_valid && (ce->ce_flags & CE_VALID))
382 return 0;
383 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
384 return 0;
385
386 /*
387 * Intent-to-add entries have not been added, so the index entry
388 * by definition never matches what is in the work tree until it
389 * actually gets added.
390 */
391 if (ce_intent_to_add(ce))
392 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
393
394 changed = ce_match_stat_basic(ce, st);
395
396 /*
397 * Within 1 second of this sequence:
398 * echo xyzzy >file && git-update-index --add file
399 * running this command:
400 * echo frotz >file
401 * would give a falsely clean cache entry. The mtime and
402 * length match the cache, and other stat fields do not change.
403 *
404 * We could detect this at update-index time (the cache entry
405 * being registered/updated records the same time as "now")
406 * and delay the return from git-update-index, but that would
407 * effectively mean we can make at most one commit per second,
408 * which is not acceptable. Instead, we check cache entries
409 * whose mtime are the same as the index file timestamp more
410 * carefully than others.
411 */
412 if (!changed && is_racy_timestamp(istate, ce)) {
413 if (assume_racy_is_modified)
414 changed |= DATA_CHANGED;
415 else
416 changed |= ce_modified_check_fs(istate, ce, st);
417 }
418
419 return changed;
420 }
421
422 int ie_modified(struct index_state *istate,
423 const struct cache_entry *ce,
424 struct stat *st, unsigned int options)
425 {
426 int changed, changed_fs;
427
428 changed = ie_match_stat(istate, ce, st, options);
429 if (!changed)
430 return 0;
431 /*
432 * If the mode or type has changed, there's no point in trying
433 * to refresh the entry - it's not going to match
434 */
435 if (changed & (MODE_CHANGED | TYPE_CHANGED))
436 return changed;
437
438 /*
439 * Immediately after read-tree or update-index --cacheinfo,
440 * the length field is zero, as we have never even read the
441 * lstat(2) information once, and we cannot trust DATA_CHANGED
442 * returned by ie_match_stat() which in turn was returned by
443 * ce_match_stat_basic() to signal that the filesize of the
444 * blob changed. We have to actually go to the filesystem to
445 * see if the contents match, and if so, should answer "unchanged".
446 *
447 * The logic does not apply to gitlinks, as ce_match_stat_basic()
448 * already has checked the actual HEAD from the filesystem in the
449 * subproject. If ie_match_stat() already said it is different,
450 * then we know it is.
451 */
452 if ((changed & DATA_CHANGED) &&
453 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
454 return changed;
455
456 changed_fs = ce_modified_check_fs(istate, ce, st);
457 if (changed_fs)
458 return changed | changed_fs;
459 return 0;
460 }
461
462 int base_name_compare(const char *name1, int len1, int mode1,
463 const char *name2, int len2, int mode2)
464 {
465 unsigned char c1, c2;
466 int len = len1 < len2 ? len1 : len2;
467 int cmp;
468
469 cmp = memcmp(name1, name2, len);
470 if (cmp)
471 return cmp;
472 c1 = name1[len];
473 c2 = name2[len];
474 if (!c1 && S_ISDIR(mode1))
475 c1 = '/';
476 if (!c2 && S_ISDIR(mode2))
477 c2 = '/';
478 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
479 }
480
481 /*
482 * df_name_compare() is identical to base_name_compare(), except it
483 * compares conflicting directory/file entries as equal. Note that
484 * while a directory name compares as equal to a regular file, they
485 * then individually compare _differently_ to a filename that has
486 * a dot after the basename (because '\0' < '.' < '/').
487 *
488 * This is used by routines that want to traverse the git namespace
489 * but then handle conflicting entries together when possible.
490 */
491 int df_name_compare(const char *name1, int len1, int mode1,
492 const char *name2, int len2, int mode2)
493 {
494 int len = len1 < len2 ? len1 : len2, cmp;
495 unsigned char c1, c2;
496
497 cmp = memcmp(name1, name2, len);
498 if (cmp)
499 return cmp;
500 /* Directories and files compare equal (same length, same name) */
501 if (len1 == len2)
502 return 0;
503 c1 = name1[len];
504 if (!c1 && S_ISDIR(mode1))
505 c1 = '/';
506 c2 = name2[len];
507 if (!c2 && S_ISDIR(mode2))
508 c2 = '/';
509 if (c1 == '/' && !c2)
510 return 0;
511 if (c2 == '/' && !c1)
512 return 0;
513 return c1 - c2;
514 }
515
516 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
517 {
518 size_t min_len = (len1 < len2) ? len1 : len2;
519 int cmp = memcmp(name1, name2, min_len);
520 if (cmp)
521 return cmp;
522 if (len1 < len2)
523 return -1;
524 if (len1 > len2)
525 return 1;
526 return 0;
527 }
528
529 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
530 {
531 int cmp;
532
533 cmp = name_compare(name1, len1, name2, len2);
534 if (cmp)
535 return cmp;
536
537 if (stage1 < stage2)
538 return -1;
539 if (stage1 > stage2)
540 return 1;
541 return 0;
542 }
543
544 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
545 {
546 int first, last;
547
548 first = 0;
549 last = istate->cache_nr;
550 while (last > first) {
551 int next = (last + first) >> 1;
552 struct cache_entry *ce = istate->cache[next];
553 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
554 if (!cmp)
555 return next;
556 if (cmp < 0) {
557 last = next;
558 continue;
559 }
560 first = next+1;
561 }
562 return -first-1;
563 }
564
565 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
566 {
567 return index_name_stage_pos(istate, name, namelen, 0);
568 }
569
570 int remove_index_entry_at(struct index_state *istate, int pos)
571 {
572 struct cache_entry *ce = istate->cache[pos];
573
574 record_resolve_undo(istate, ce);
575 remove_name_hash(istate, ce);
576 save_or_free_index_entry(istate, ce);
577 istate->cache_changed |= CE_ENTRY_REMOVED;
578 istate->cache_nr--;
579 if (pos >= istate->cache_nr)
580 return 0;
581 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
582 istate->cache_nr - pos);
583 return 1;
584 }
585
586 /*
587 * Remove all cache entries marked for removal, that is where
588 * CE_REMOVE is set in ce_flags. This is much more effective than
589 * calling remove_index_entry_at() for each entry to be removed.
590 */
591 void remove_marked_cache_entries(struct index_state *istate)
592 {
593 struct cache_entry **ce_array = istate->cache;
594 unsigned int i, j;
595
596 for (i = j = 0; i < istate->cache_nr; i++) {
597 if (ce_array[i]->ce_flags & CE_REMOVE) {
598 remove_name_hash(istate, ce_array[i]);
599 save_or_free_index_entry(istate, ce_array[i]);
600 }
601 else
602 ce_array[j++] = ce_array[i];
603 }
604 if (j == istate->cache_nr)
605 return;
606 istate->cache_changed |= CE_ENTRY_REMOVED;
607 istate->cache_nr = j;
608 }
609
610 int remove_file_from_index(struct index_state *istate, const char *path)
611 {
612 int pos = index_name_pos(istate, path, strlen(path));
613 if (pos < 0)
614 pos = -pos-1;
615 cache_tree_invalidate_path(istate, path);
616 untracked_cache_remove_from_index(istate, path);
617 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
618 remove_index_entry_at(istate, pos);
619 return 0;
620 }
621
622 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
623 {
624 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
625 }
626
627 static int index_name_pos_also_unmerged(struct index_state *istate,
628 const char *path, int namelen)
629 {
630 int pos = index_name_pos(istate, path, namelen);
631 struct cache_entry *ce;
632
633 if (pos >= 0)
634 return pos;
635
636 /* maybe unmerged? */
637 pos = -1 - pos;
638 if (pos >= istate->cache_nr ||
639 compare_name((ce = istate->cache[pos]), path, namelen))
640 return -1;
641
642 /* order of preference: stage 2, 1, 3 */
643 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
644 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
645 !compare_name(ce, path, namelen))
646 pos++;
647 return pos;
648 }
649
650 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
651 {
652 int len = ce_namelen(ce);
653 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
654 }
655
656 /*
657 * If we add a filename that aliases in the cache, we will use the
658 * name that we already have - but we don't want to update the same
659 * alias twice, because that implies that there were actually two
660 * different files with aliasing names!
661 *
662 * So we use the CE_ADDED flag to verify that the alias was an old
663 * one before we accept it as
664 */
665 static struct cache_entry *create_alias_ce(struct index_state *istate,
666 struct cache_entry *ce,
667 struct cache_entry *alias)
668 {
669 int len;
670 struct cache_entry *new_entry;
671
672 if (alias->ce_flags & CE_ADDED)
673 die(_("will not add file alias '%s' ('%s' already exists in index)"),
674 ce->name, alias->name);
675
676 /* Ok, create the new entry using the name of the existing alias */
677 len = ce_namelen(alias);
678 new_entry = make_empty_cache_entry(istate, len);
679 memcpy(new_entry->name, alias->name, len);
680 copy_cache_entry(new_entry, ce);
681 save_or_free_index_entry(istate, ce);
682 return new_entry;
683 }
684
685 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
686 {
687 struct object_id oid;
688 if (write_object_file("", 0, blob_type, &oid))
689 die(_("cannot create an empty blob in the object database"));
690 oidcpy(&ce->oid, &oid);
691 }
692
693 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
694 {
695 int namelen, was_same;
696 mode_t st_mode = st->st_mode;
697 struct cache_entry *ce, *alias = NULL;
698 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
699 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
700 int pretend = flags & ADD_CACHE_PRETEND;
701 int intent_only = flags & ADD_CACHE_INTENT;
702 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
703 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
704 int hash_flags = HASH_WRITE_OBJECT;
705
706 if (flags & ADD_CACHE_RENORMALIZE)
707 hash_flags |= HASH_RENORMALIZE;
708
709 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
710 return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
711
712 namelen = strlen(path);
713 if (S_ISDIR(st_mode)) {
714 while (namelen && path[namelen-1] == '/')
715 namelen--;
716 }
717 ce = make_empty_cache_entry(istate, namelen);
718 memcpy(ce->name, path, namelen);
719 ce->ce_namelen = namelen;
720 if (!intent_only)
721 fill_stat_cache_info(ce, st);
722 else
723 ce->ce_flags |= CE_INTENT_TO_ADD;
724
725
726 if (trust_executable_bit && has_symlinks) {
727 ce->ce_mode = create_ce_mode(st_mode);
728 } else {
729 /* If there is an existing entry, pick the mode bits and type
730 * from it, otherwise assume unexecutable regular file.
731 */
732 struct cache_entry *ent;
733 int pos = index_name_pos_also_unmerged(istate, path, namelen);
734
735 ent = (0 <= pos) ? istate->cache[pos] : NULL;
736 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
737 }
738
739 /* When core.ignorecase=true, determine if a directory of the same name but differing
740 * case already exists within the Git repository. If it does, ensure the directory
741 * case of the file being added to the repository matches (is folded into) the existing
742 * entry's directory case.
743 */
744 if (ignore_case) {
745 adjust_dirname_case(istate, ce->name);
746 }
747 if (!(flags & ADD_CACHE_RENORMALIZE)) {
748 alias = index_file_exists(istate, ce->name,
749 ce_namelen(ce), ignore_case);
750 if (alias &&
751 !ce_stage(alias) &&
752 !ie_match_stat(istate, alias, st, ce_option)) {
753 /* Nothing changed, really */
754 if (!S_ISGITLINK(alias->ce_mode))
755 ce_mark_uptodate(alias);
756 alias->ce_flags |= CE_ADDED;
757
758 discard_cache_entry(ce);
759 return 0;
760 }
761 }
762 if (!intent_only) {
763 if (index_path(istate, &ce->oid, path, st, hash_flags)) {
764 discard_cache_entry(ce);
765 return error(_("unable to index file '%s'"), path);
766 }
767 } else
768 set_object_name_for_intent_to_add_entry(ce);
769
770 if (ignore_case && alias && different_name(ce, alias))
771 ce = create_alias_ce(istate, ce, alias);
772 ce->ce_flags |= CE_ADDED;
773
774 /* It was suspected to be racily clean, but it turns out to be Ok */
775 was_same = (alias &&
776 !ce_stage(alias) &&
777 oideq(&alias->oid, &ce->oid) &&
778 ce->ce_mode == alias->ce_mode);
779
780 if (pretend)
781 discard_cache_entry(ce);
782 else if (add_index_entry(istate, ce, add_option)) {
783 discard_cache_entry(ce);
784 return error(_("unable to add '%s' to index"), path);
785 }
786 if (verbose && !was_same)
787 printf("add '%s'\n", path);
788 return 0;
789 }
790
791 int add_file_to_index(struct index_state *istate, const char *path, int flags)
792 {
793 struct stat st;
794 if (lstat(path, &st))
795 die_errno(_("unable to stat '%s'"), path);
796 return add_to_index(istate, path, &st, flags);
797 }
798
799 struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
800 {
801 return mem_pool__ce_calloc(find_mem_pool(istate), len);
802 }
803
804 struct cache_entry *make_empty_transient_cache_entry(size_t len)
805 {
806 return xcalloc(1, cache_entry_size(len));
807 }
808
809 struct cache_entry *make_cache_entry(struct index_state *istate,
810 unsigned int mode,
811 const struct object_id *oid,
812 const char *path,
813 int stage,
814 unsigned int refresh_options)
815 {
816 struct cache_entry *ce, *ret;
817 int len;
818
819 if (!verify_path(path, mode)) {
820 error(_("invalid path '%s'"), path);
821 return NULL;
822 }
823
824 len = strlen(path);
825 ce = make_empty_cache_entry(istate, len);
826
827 oidcpy(&ce->oid, oid);
828 memcpy(ce->name, path, len);
829 ce->ce_flags = create_ce_flags(stage);
830 ce->ce_namelen = len;
831 ce->ce_mode = create_ce_mode(mode);
832
833 ret = refresh_cache_entry(istate, ce, refresh_options);
834 if (ret != ce)
835 discard_cache_entry(ce);
836 return ret;
837 }
838
839 struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
840 const char *path, int stage)
841 {
842 struct cache_entry *ce;
843 int len;
844
845 if (!verify_path(path, mode)) {
846 error(_("invalid path '%s'"), path);
847 return NULL;
848 }
849
850 len = strlen(path);
851 ce = make_empty_transient_cache_entry(len);
852
853 oidcpy(&ce->oid, oid);
854 memcpy(ce->name, path, len);
855 ce->ce_flags = create_ce_flags(stage);
856 ce->ce_namelen = len;
857 ce->ce_mode = create_ce_mode(mode);
858
859 return ce;
860 }
861
862 /*
863 * Chmod an index entry with either +x or -x.
864 *
865 * Returns -1 if the chmod for the particular cache entry failed (if it's
866 * not a regular file), -2 if an invalid flip argument is passed in, 0
867 * otherwise.
868 */
869 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
870 char flip)
871 {
872 if (!S_ISREG(ce->ce_mode))
873 return -1;
874 switch (flip) {
875 case '+':
876 ce->ce_mode |= 0111;
877 break;
878 case '-':
879 ce->ce_mode &= ~0111;
880 break;
881 default:
882 return -2;
883 }
884 cache_tree_invalidate_path(istate, ce->name);
885 ce->ce_flags |= CE_UPDATE_IN_BASE;
886 mark_fsmonitor_invalid(istate, ce);
887 istate->cache_changed |= CE_ENTRY_CHANGED;
888
889 return 0;
890 }
891
892 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
893 {
894 int len = ce_namelen(a);
895 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
896 }
897
898 /*
899 * We fundamentally don't like some paths: we don't want
900 * dot or dot-dot anywhere, and for obvious reasons don't
901 * want to recurse into ".git" either.
902 *
903 * Also, we don't want double slashes or slashes at the
904 * end that can make pathnames ambiguous.
905 */
906 static int verify_dotfile(const char *rest, unsigned mode)
907 {
908 /*
909 * The first character was '.', but that
910 * has already been discarded, we now test
911 * the rest.
912 */
913
914 /* "." is not allowed */
915 if (*rest == '\0' || is_dir_sep(*rest))
916 return 0;
917
918 switch (*rest) {
919 /*
920 * ".git" followed by NUL or slash is bad. Note that we match
921 * case-insensitively here, even if ignore_case is not set.
922 * This outlaws ".GIT" everywhere out of an abundance of caution,
923 * since there's really no good reason to allow it.
924 *
925 * Once we've seen ".git", we can also find ".gitmodules", etc (also
926 * case-insensitively).
927 */
928 case 'g':
929 case 'G':
930 if (rest[1] != 'i' && rest[1] != 'I')
931 break;
932 if (rest[2] != 't' && rest[2] != 'T')
933 break;
934 if (rest[3] == '\0' || is_dir_sep(rest[3]))
935 return 0;
936 if (S_ISLNK(mode)) {
937 rest += 3;
938 if (skip_iprefix(rest, "modules", &rest) &&
939 (*rest == '\0' || is_dir_sep(*rest)))
940 return 0;
941 }
942 break;
943 case '.':
944 if (rest[1] == '\0' || is_dir_sep(rest[1]))
945 return 0;
946 }
947 return 1;
948 }
949
950 int verify_path(const char *path, unsigned mode)
951 {
952 char c;
953
954 if (has_dos_drive_prefix(path))
955 return 0;
956
957 goto inside;
958 for (;;) {
959 if (!c)
960 return 1;
961 if (is_dir_sep(c)) {
962 inside:
963 if (protect_hfs) {
964 if (is_hfs_dotgit(path))
965 return 0;
966 if (S_ISLNK(mode)) {
967 if (is_hfs_dotgitmodules(path))
968 return 0;
969 }
970 }
971 if (protect_ntfs) {
972 if (is_ntfs_dotgit(path))
973 return 0;
974 if (S_ISLNK(mode)) {
975 if (is_ntfs_dotgitmodules(path))
976 return 0;
977 }
978 }
979
980 c = *path++;
981 if ((c == '.' && !verify_dotfile(path, mode)) ||
982 is_dir_sep(c) || c == '\0')
983 return 0;
984 }
985 c = *path++;
986 }
987 }
988
989 /*
990 * Do we have another file that has the beginning components being a
991 * proper superset of the name we're trying to add?
992 */
993 static int has_file_name(struct index_state *istate,
994 const struct cache_entry *ce, int pos, int ok_to_replace)
995 {
996 int retval = 0;
997 int len = ce_namelen(ce);
998 int stage = ce_stage(ce);
999 const char *name = ce->name;
1000
1001 while (pos < istate->cache_nr) {
1002 struct cache_entry *p = istate->cache[pos++];
1003
1004 if (len >= ce_namelen(p))
1005 break;
1006 if (memcmp(name, p->name, len))
1007 break;
1008 if (ce_stage(p) != stage)
1009 continue;
1010 if (p->name[len] != '/')
1011 continue;
1012 if (p->ce_flags & CE_REMOVE)
1013 continue;
1014 retval = -1;
1015 if (!ok_to_replace)
1016 break;
1017 remove_index_entry_at(istate, --pos);
1018 }
1019 return retval;
1020 }
1021
1022
1023 /*
1024 * Like strcmp(), but also return the offset of the first change.
1025 * If strings are equal, return the length.
1026 */
1027 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1028 {
1029 size_t k;
1030
1031 if (!first_change)
1032 return strcmp(s1, s2);
1033
1034 for (k = 0; s1[k] == s2[k]; k++)
1035 if (s1[k] == '\0')
1036 break;
1037
1038 *first_change = k;
1039 return (unsigned char)s1[k] - (unsigned char)s2[k];
1040 }
1041
1042 /*
1043 * Do we have another file with a pathname that is a proper
1044 * subset of the name we're trying to add?
1045 *
1046 * That is, is there another file in the index with a path
1047 * that matches a sub-directory in the given entry?
1048 */
1049 static int has_dir_name(struct index_state *istate,
1050 const struct cache_entry *ce, int pos, int ok_to_replace)
1051 {
1052 int retval = 0;
1053 int stage = ce_stage(ce);
1054 const char *name = ce->name;
1055 const char *slash = name + ce_namelen(ce);
1056 size_t len_eq_last;
1057 int cmp_last = 0;
1058
1059 /*
1060 * We are frequently called during an iteration on a sorted
1061 * list of pathnames and while building a new index. Therefore,
1062 * there is a high probability that this entry will eventually
1063 * be appended to the index, rather than inserted in the middle.
1064 * If we can confirm that, we can avoid binary searches on the
1065 * components of the pathname.
1066 *
1067 * Compare the entry's full path with the last path in the index.
1068 */
1069 if (istate->cache_nr > 0) {
1070 cmp_last = strcmp_offset(name,
1071 istate->cache[istate->cache_nr - 1]->name,
1072 &len_eq_last);
1073 if (cmp_last > 0) {
1074 if (len_eq_last == 0) {
1075 /*
1076 * The entry sorts AFTER the last one in the
1077 * index and their paths have no common prefix,
1078 * so there cannot be a F/D conflict.
1079 */
1080 return retval;
1081 } else {
1082 /*
1083 * The entry sorts AFTER the last one in the
1084 * index, but has a common prefix. Fall through
1085 * to the loop below to disect the entry's path
1086 * and see where the difference is.
1087 */
1088 }
1089 } else if (cmp_last == 0) {
1090 /*
1091 * The entry exactly matches the last one in the
1092 * index, but because of multiple stage and CE_REMOVE
1093 * items, we fall through and let the regular search
1094 * code handle it.
1095 */
1096 }
1097 }
1098
1099 for (;;) {
1100 size_t len;
1101
1102 for (;;) {
1103 if (*--slash == '/')
1104 break;
1105 if (slash <= ce->name)
1106 return retval;
1107 }
1108 len = slash - name;
1109
1110 if (cmp_last > 0) {
1111 /*
1112 * (len + 1) is a directory boundary (including
1113 * the trailing slash). And since the loop is
1114 * decrementing "slash", the first iteration is
1115 * the longest directory prefix; subsequent
1116 * iterations consider parent directories.
1117 */
1118
1119 if (len + 1 <= len_eq_last) {
1120 /*
1121 * The directory prefix (including the trailing
1122 * slash) also appears as a prefix in the last
1123 * entry, so the remainder cannot collide (because
1124 * strcmp said the whole path was greater).
1125 *
1126 * EQ: last: xxx/A
1127 * this: xxx/B
1128 *
1129 * LT: last: xxx/file_A
1130 * this: xxx/file_B
1131 */
1132 return retval;
1133 }
1134
1135 if (len > len_eq_last) {
1136 /*
1137 * This part of the directory prefix (excluding
1138 * the trailing slash) is longer than the known
1139 * equal portions, so this sub-directory cannot
1140 * collide with a file.
1141 *
1142 * GT: last: xxxA
1143 * this: xxxB/file
1144 */
1145 return retval;
1146 }
1147
1148 if (istate->cache_nr > 0 &&
1149 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1150 /*
1151 * The directory prefix lines up with part of
1152 * a longer file or directory name, but sorts
1153 * after it, so this sub-directory cannot
1154 * collide with a file.
1155 *
1156 * last: xxx/yy-file (because '-' sorts before '/')
1157 * this: xxx/yy/abc
1158 */
1159 return retval;
1160 }
1161
1162 /*
1163 * This is a possible collision. Fall through and
1164 * let the regular search code handle it.
1165 *
1166 * last: xxx
1167 * this: xxx/file
1168 */
1169 }
1170
1171 pos = index_name_stage_pos(istate, name, len, stage);
1172 if (pos >= 0) {
1173 /*
1174 * Found one, but not so fast. This could
1175 * be a marker that says "I was here, but
1176 * I am being removed". Such an entry is
1177 * not a part of the resulting tree, and
1178 * it is Ok to have a directory at the same
1179 * path.
1180 */
1181 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1182 retval = -1;
1183 if (!ok_to_replace)
1184 break;
1185 remove_index_entry_at(istate, pos);
1186 continue;
1187 }
1188 }
1189 else
1190 pos = -pos-1;
1191
1192 /*
1193 * Trivial optimization: if we find an entry that
1194 * already matches the sub-directory, then we know
1195 * we're ok, and we can exit.
1196 */
1197 while (pos < istate->cache_nr) {
1198 struct cache_entry *p = istate->cache[pos];
1199 if ((ce_namelen(p) <= len) ||
1200 (p->name[len] != '/') ||
1201 memcmp(p->name, name, len))
1202 break; /* not our subdirectory */
1203 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1204 /*
1205 * p is at the same stage as our entry, and
1206 * is a subdirectory of what we are looking
1207 * at, so we cannot have conflicts at our
1208 * level or anything shorter.
1209 */
1210 return retval;
1211 pos++;
1212 }
1213 }
1214 return retval;
1215 }
1216
1217 /* We may be in a situation where we already have path/file and path
1218 * is being added, or we already have path and path/file is being
1219 * added. Either one would result in a nonsense tree that has path
1220 * twice when git-write-tree tries to write it out. Prevent it.
1221 *
1222 * If ok-to-replace is specified, we remove the conflicting entries
1223 * from the cache so the caller should recompute the insert position.
1224 * When this happens, we return non-zero.
1225 */
1226 static int check_file_directory_conflict(struct index_state *istate,
1227 const struct cache_entry *ce,
1228 int pos, int ok_to_replace)
1229 {
1230 int retval;
1231
1232 /*
1233 * When ce is an "I am going away" entry, we allow it to be added
1234 */
1235 if (ce->ce_flags & CE_REMOVE)
1236 return 0;
1237
1238 /*
1239 * We check if the path is a sub-path of a subsequent pathname
1240 * first, since removing those will not change the position
1241 * in the array.
1242 */
1243 retval = has_file_name(istate, ce, pos, ok_to_replace);
1244
1245 /*
1246 * Then check if the path might have a clashing sub-directory
1247 * before it.
1248 */
1249 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1250 }
1251
1252 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1253 {
1254 int pos;
1255 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1256 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1257 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1258 int new_only = option & ADD_CACHE_NEW_ONLY;
1259
1260 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1261 cache_tree_invalidate_path(istate, ce->name);
1262
1263 /*
1264 * If this entry's path sorts after the last entry in the index,
1265 * we can avoid searching for it.
1266 */
1267 if (istate->cache_nr > 0 &&
1268 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1269 pos = -istate->cache_nr - 1;
1270 else
1271 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1272
1273 /* existing match? Just replace it. */
1274 if (pos >= 0) {
1275 if (!new_only)
1276 replace_index_entry(istate, pos, ce);
1277 return 0;
1278 }
1279 pos = -pos-1;
1280
1281 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1282 untracked_cache_add_to_index(istate, ce->name);
1283
1284 /*
1285 * Inserting a merged entry ("stage 0") into the index
1286 * will always replace all non-merged entries..
1287 */
1288 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1289 while (ce_same_name(istate->cache[pos], ce)) {
1290 ok_to_add = 1;
1291 if (!remove_index_entry_at(istate, pos))
1292 break;
1293 }
1294 }
1295
1296 if (!ok_to_add)
1297 return -1;
1298 if (!verify_path(ce->name, ce->ce_mode))
1299 return error(_("invalid path '%s'"), ce->name);
1300
1301 if (!skip_df_check &&
1302 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1303 if (!ok_to_replace)
1304 return error(_("'%s' appears as both a file and as a directory"),
1305 ce->name);
1306 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1307 pos = -pos-1;
1308 }
1309 return pos + 1;
1310 }
1311
1312 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1313 {
1314 int pos;
1315
1316 if (option & ADD_CACHE_JUST_APPEND)
1317 pos = istate->cache_nr;
1318 else {
1319 int ret;
1320 ret = add_index_entry_with_check(istate, ce, option);
1321 if (ret <= 0)
1322 return ret;
1323 pos = ret - 1;
1324 }
1325
1326 /* Make sure the array is big enough .. */
1327 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1328
1329 /* Add it in.. */
1330 istate->cache_nr++;
1331 if (istate->cache_nr > pos + 1)
1332 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1333 istate->cache_nr - pos - 1);
1334 set_index_entry(istate, pos, ce);
1335 istate->cache_changed |= CE_ENTRY_ADDED;
1336 return 0;
1337 }
1338
1339 /*
1340 * "refresh" does not calculate a new sha1 file or bring the
1341 * cache up-to-date for mode/content changes. But what it
1342 * _does_ do is to "re-match" the stat information of a file
1343 * with the cache, so that you can refresh the cache for a
1344 * file that hasn't been changed but where the stat entry is
1345 * out of date.
1346 *
1347 * For example, you'd want to do this after doing a "git-read-tree",
1348 * to link up the stat cache details with the proper files.
1349 */
1350 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1351 struct cache_entry *ce,
1352 unsigned int options, int *err,
1353 int *changed_ret)
1354 {
1355 struct stat st;
1356 struct cache_entry *updated;
1357 int changed;
1358 int refresh = options & CE_MATCH_REFRESH;
1359 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1360 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1361 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1362 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1363
1364 if (!refresh || ce_uptodate(ce))
1365 return ce;
1366
1367 if (!ignore_fsmonitor)
1368 refresh_fsmonitor(istate);
1369 /*
1370 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1371 * that the change to the work tree does not matter and told
1372 * us not to worry.
1373 */
1374 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1375 ce_mark_uptodate(ce);
1376 return ce;
1377 }
1378 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1379 ce_mark_uptodate(ce);
1380 return ce;
1381 }
1382 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1383 ce_mark_uptodate(ce);
1384 return ce;
1385 }
1386
1387 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1388 if (ignore_missing)
1389 return ce;
1390 if (err)
1391 *err = ENOENT;
1392 return NULL;
1393 }
1394
1395 if (lstat(ce->name, &st) < 0) {
1396 if (ignore_missing && errno == ENOENT)
1397 return ce;
1398 if (err)
1399 *err = errno;
1400 return NULL;
1401 }
1402
1403 changed = ie_match_stat(istate, ce, &st, options);
1404 if (changed_ret)
1405 *changed_ret = changed;
1406 if (!changed) {
1407 /*
1408 * The path is unchanged. If we were told to ignore
1409 * valid bit, then we did the actual stat check and
1410 * found that the entry is unmodified. If the entry
1411 * is not marked VALID, this is the place to mark it
1412 * valid again, under "assume unchanged" mode.
1413 */
1414 if (ignore_valid && assume_unchanged &&
1415 !(ce->ce_flags & CE_VALID))
1416 ; /* mark this one VALID again */
1417 else {
1418 /*
1419 * We do not mark the index itself "modified"
1420 * because CE_UPTODATE flag is in-core only;
1421 * we are not going to write this change out.
1422 */
1423 if (!S_ISGITLINK(ce->ce_mode)) {
1424 ce_mark_uptodate(ce);
1425 mark_fsmonitor_valid(ce);
1426 }
1427 return ce;
1428 }
1429 }
1430
1431 if (ie_modified(istate, ce, &st, options)) {
1432 if (err)
1433 *err = EINVAL;
1434 return NULL;
1435 }
1436
1437 updated = make_empty_cache_entry(istate, ce_namelen(ce));
1438 copy_cache_entry(updated, ce);
1439 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1440 fill_stat_cache_info(updated, &st);
1441 /*
1442 * If ignore_valid is not set, we should leave CE_VALID bit
1443 * alone. Otherwise, paths marked with --no-assume-unchanged
1444 * (i.e. things to be edited) will reacquire CE_VALID bit
1445 * automatically, which is not really what we want.
1446 */
1447 if (!ignore_valid && assume_unchanged &&
1448 !(ce->ce_flags & CE_VALID))
1449 updated->ce_flags &= ~CE_VALID;
1450
1451 /* istate->cache_changed is updated in the caller */
1452 return updated;
1453 }
1454
1455 static void show_file(const char * fmt, const char * name, int in_porcelain,
1456 int * first, const char *header_msg)
1457 {
1458 if (in_porcelain && *first && header_msg) {
1459 printf("%s\n", header_msg);
1460 *first = 0;
1461 }
1462 printf(fmt, name);
1463 }
1464
1465 int refresh_index(struct index_state *istate, unsigned int flags,
1466 const struct pathspec *pathspec,
1467 char *seen, const char *header_msg)
1468 {
1469 int i;
1470 int has_errors = 0;
1471 int really = (flags & REFRESH_REALLY) != 0;
1472 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1473 int quiet = (flags & REFRESH_QUIET) != 0;
1474 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1475 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1476 int first = 1;
1477 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1478 unsigned int options = (CE_MATCH_REFRESH |
1479 (really ? CE_MATCH_IGNORE_VALID : 0) |
1480 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1481 const char *modified_fmt;
1482 const char *deleted_fmt;
1483 const char *typechange_fmt;
1484 const char *added_fmt;
1485 const char *unmerged_fmt;
1486 struct progress *progress = NULL;
1487
1488 if (flags & REFRESH_PROGRESS && isatty(2))
1489 progress = start_delayed_progress(_("Refresh index"),
1490 istate->cache_nr);
1491
1492 trace_performance_enter();
1493 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1494 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1495 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1496 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1497 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1498 /*
1499 * Use the multi-threaded preload_index() to refresh most of the
1500 * cache entries quickly then in the single threaded loop below,
1501 * we only have to do the special cases that are left.
1502 */
1503 preload_index(istate, pathspec, 0);
1504 for (i = 0; i < istate->cache_nr; i++) {
1505 struct cache_entry *ce, *new_entry;
1506 int cache_errno = 0;
1507 int changed = 0;
1508 int filtered = 0;
1509
1510 ce = istate->cache[i];
1511 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1512 continue;
1513
1514 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1515 filtered = 1;
1516
1517 if (ce_stage(ce)) {
1518 while ((i < istate->cache_nr) &&
1519 ! strcmp(istate->cache[i]->name, ce->name))
1520 i++;
1521 i--;
1522 if (allow_unmerged)
1523 continue;
1524 if (!filtered)
1525 show_file(unmerged_fmt, ce->name, in_porcelain,
1526 &first, header_msg);
1527 has_errors = 1;
1528 continue;
1529 }
1530
1531 if (filtered)
1532 continue;
1533
1534 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1535 if (new_entry == ce)
1536 continue;
1537 if (progress)
1538 display_progress(progress, i);
1539 if (!new_entry) {
1540 const char *fmt;
1541
1542 if (really && cache_errno == EINVAL) {
1543 /* If we are doing --really-refresh that
1544 * means the index is not valid anymore.
1545 */
1546 ce->ce_flags &= ~CE_VALID;
1547 ce->ce_flags |= CE_UPDATE_IN_BASE;
1548 mark_fsmonitor_invalid(istate, ce);
1549 istate->cache_changed |= CE_ENTRY_CHANGED;
1550 }
1551 if (quiet)
1552 continue;
1553
1554 if (cache_errno == ENOENT)
1555 fmt = deleted_fmt;
1556 else if (ce_intent_to_add(ce))
1557 fmt = added_fmt; /* must be before other checks */
1558 else if (changed & TYPE_CHANGED)
1559 fmt = typechange_fmt;
1560 else
1561 fmt = modified_fmt;
1562 show_file(fmt,
1563 ce->name, in_porcelain, &first, header_msg);
1564 has_errors = 1;
1565 continue;
1566 }
1567
1568 replace_index_entry(istate, i, new_entry);
1569 }
1570 if (progress) {
1571 display_progress(progress, istate->cache_nr);
1572 stop_progress(&progress);
1573 }
1574 trace_performance_leave("refresh index");
1575 return has_errors;
1576 }
1577
1578 struct cache_entry *refresh_cache_entry(struct index_state *istate,
1579 struct cache_entry *ce,
1580 unsigned int options)
1581 {
1582 return refresh_cache_ent(istate, ce, options, NULL, NULL);
1583 }
1584
1585
1586 /*****************************************************************
1587 * Index File I/O
1588 *****************************************************************/
1589
1590 #define INDEX_FORMAT_DEFAULT 3
1591
1592 static unsigned int get_index_format_default(void)
1593 {
1594 char *envversion = getenv("GIT_INDEX_VERSION");
1595 char *endp;
1596 int value;
1597 unsigned int version = INDEX_FORMAT_DEFAULT;
1598
1599 if (!envversion) {
1600 if (!git_config_get_int("index.version", &value))
1601 version = value;
1602 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1603 warning(_("index.version set, but the value is invalid.\n"
1604 "Using version %i"), INDEX_FORMAT_DEFAULT);
1605 return INDEX_FORMAT_DEFAULT;
1606 }
1607 return version;
1608 }
1609
1610 version = strtoul(envversion, &endp, 10);
1611 if (*endp ||
1612 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1613 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1614 "Using version %i"), INDEX_FORMAT_DEFAULT);
1615 version = INDEX_FORMAT_DEFAULT;
1616 }
1617 return version;
1618 }
1619
1620 /*
1621 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1622 * Again - this is just a (very strong in practice) heuristic that
1623 * the inode hasn't changed.
1624 *
1625 * We save the fields in big-endian order to allow using the
1626 * index file over NFS transparently.
1627 */
1628 struct ondisk_cache_entry {
1629 struct cache_time ctime;
1630 struct cache_time mtime;
1631 uint32_t dev;
1632 uint32_t ino;
1633 uint32_t mode;
1634 uint32_t uid;
1635 uint32_t gid;
1636 uint32_t size;
1637 unsigned char sha1[20];
1638 uint16_t flags;
1639 char name[FLEX_ARRAY]; /* more */
1640 };
1641
1642 /*
1643 * This struct is used when CE_EXTENDED bit is 1
1644 * The struct must match ondisk_cache_entry exactly from
1645 * ctime till flags
1646 */
1647 struct ondisk_cache_entry_extended {
1648 struct cache_time ctime;
1649 struct cache_time mtime;
1650 uint32_t dev;
1651 uint32_t ino;
1652 uint32_t mode;
1653 uint32_t uid;
1654 uint32_t gid;
1655 uint32_t size;
1656 unsigned char sha1[20];
1657 uint16_t flags;
1658 uint16_t flags2;
1659 char name[FLEX_ARRAY]; /* more */
1660 };
1661
1662 /* These are only used for v3 or lower */
1663 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1664 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1665 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1666 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1667 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1668 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1669 ondisk_cache_entry_size(ce_namelen(ce)))
1670
1671 /* Allow fsck to force verification of the index checksum. */
1672 int verify_index_checksum;
1673
1674 /* Allow fsck to force verification of the cache entry order. */
1675 int verify_ce_order;
1676
1677 static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1678 {
1679 git_hash_ctx c;
1680 unsigned char hash[GIT_MAX_RAWSZ];
1681 int hdr_version;
1682
1683 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1684 return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1685 hdr_version = ntohl(hdr->hdr_version);
1686 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1687 return error(_("bad index version %d"), hdr_version);
1688
1689 if (!verify_index_checksum)
1690 return 0;
1691
1692 the_hash_algo->init_fn(&c);
1693 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1694 the_hash_algo->final_fn(hash, &c);
1695 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1696 return error(_("bad index file sha1 signature"));
1697 return 0;
1698 }
1699
1700 static int read_index_extension(struct index_state *istate,
1701 const char *ext, const char *data, unsigned long sz)
1702 {
1703 switch (CACHE_EXT(ext)) {
1704 case CACHE_EXT_TREE:
1705 istate->cache_tree = cache_tree_read(data, sz);
1706 break;
1707 case CACHE_EXT_RESOLVE_UNDO:
1708 istate->resolve_undo = resolve_undo_read(data, sz);
1709 break;
1710 case CACHE_EXT_LINK:
1711 if (read_link_extension(istate, data, sz))
1712 return -1;
1713 break;
1714 case CACHE_EXT_UNTRACKED:
1715 istate->untracked = read_untracked_extension(data, sz);
1716 break;
1717 case CACHE_EXT_FSMONITOR:
1718 read_fsmonitor_extension(istate, data, sz);
1719 break;
1720 case CACHE_EXT_ENDOFINDEXENTRIES:
1721 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1722 /* already handled in do_read_index() */
1723 break;
1724 default:
1725 if (*ext < 'A' || 'Z' < *ext)
1726 return error(_("index uses %.4s extension, which we do not understand"),
1727 ext);
1728 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1729 break;
1730 }
1731 return 0;
1732 }
1733
1734 static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1735 unsigned int version,
1736 struct ondisk_cache_entry *ondisk,
1737 unsigned long *ent_size,
1738 const struct cache_entry *previous_ce)
1739 {
1740 struct cache_entry *ce;
1741 size_t len;
1742 const char *name;
1743 unsigned int flags;
1744 size_t copy_len = 0;
1745 /*
1746 * Adjacent cache entries tend to share the leading paths, so it makes
1747 * sense to only store the differences in later entries. In the v4
1748 * on-disk format of the index, each on-disk cache entry stores the
1749 * number of bytes to be stripped from the end of the previous name,
1750 * and the bytes to append to the result, to come up with its name.
1751 */
1752 int expand_name_field = version == 4;
1753
1754 /* On-disk flags are just 16 bits */
1755 flags = get_be16(&ondisk->flags);
1756 len = flags & CE_NAMEMASK;
1757
1758 if (flags & CE_EXTENDED) {
1759 struct ondisk_cache_entry_extended *ondisk2;
1760 int extended_flags;
1761 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1762 extended_flags = get_be16(&ondisk2->flags2) << 16;
1763 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1764 if (extended_flags & ~CE_EXTENDED_FLAGS)
1765 die(_("unknown index entry format 0x%08x"), extended_flags);
1766 flags |= extended_flags;
1767 name = ondisk2->name;
1768 }
1769 else
1770 name = ondisk->name;
1771
1772 if (expand_name_field) {
1773 const unsigned char *cp = (const unsigned char *)name;
1774 size_t strip_len, previous_len;
1775
1776 /* If we're at the begining of a block, ignore the previous name */
1777 strip_len = decode_varint(&cp);
1778 if (previous_ce) {
1779 previous_len = previous_ce->ce_namelen;
1780 if (previous_len < strip_len)
1781 die(_("malformed name field in the index, near path '%s'"),
1782 previous_ce->name);
1783 copy_len = previous_len - strip_len;
1784 }
1785 name = (const char *)cp;
1786 }
1787
1788 if (len == CE_NAMEMASK) {
1789 len = strlen(name);
1790 if (expand_name_field)
1791 len += copy_len;
1792 }
1793
1794 ce = mem_pool__ce_alloc(ce_mem_pool, len);
1795
1796 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1797 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1798 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1799 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1800 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1801 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1802 ce->ce_mode = get_be32(&ondisk->mode);
1803 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1804 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1805 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1806 ce->ce_flags = flags & ~CE_NAMEMASK;
1807 ce->ce_namelen = len;
1808 ce->index = 0;
1809 hashcpy(ce->oid.hash, ondisk->sha1);
1810
1811 if (expand_name_field) {
1812 if (copy_len)
1813 memcpy(ce->name, previous_ce->name, copy_len);
1814 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1815 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1816 } else {
1817 memcpy(ce->name, name, len + 1);
1818 *ent_size = ondisk_ce_size(ce);
1819 }
1820 return ce;
1821 }
1822
1823 static void check_ce_order(struct index_state *istate)
1824 {
1825 unsigned int i;
1826
1827 if (!verify_ce_order)
1828 return;
1829
1830 for (i = 1; i < istate->cache_nr; i++) {
1831 struct cache_entry *ce = istate->cache[i - 1];
1832 struct cache_entry *next_ce = istate->cache[i];
1833 int name_compare = strcmp(ce->name, next_ce->name);
1834
1835 if (0 < name_compare)
1836 die(_("unordered stage entries in index"));
1837 if (!name_compare) {
1838 if (!ce_stage(ce))
1839 die(_("multiple stage entries for merged file '%s'"),
1840 ce->name);
1841 if (ce_stage(ce) > ce_stage(next_ce))
1842 die(_("unordered stage entries for '%s'"),
1843 ce->name);
1844 }
1845 }
1846 }
1847
1848 static void tweak_untracked_cache(struct index_state *istate)
1849 {
1850 switch (git_config_get_untracked_cache()) {
1851 case -1: /* keep: do nothing */
1852 break;
1853 case 0: /* false */
1854 remove_untracked_cache(istate);
1855 break;
1856 case 1: /* true */
1857 add_untracked_cache(istate);
1858 break;
1859 default: /* unknown value: do nothing */
1860 break;
1861 }
1862 }
1863
1864 static void tweak_split_index(struct index_state *istate)
1865 {
1866 switch (git_config_get_split_index()) {
1867 case -1: /* unset: do nothing */
1868 break;
1869 case 0: /* false */
1870 remove_split_index(istate);
1871 break;
1872 case 1: /* true */
1873 add_split_index(istate);
1874 break;
1875 default: /* unknown value: do nothing */
1876 break;
1877 }
1878 }
1879
1880 static void post_read_index_from(struct index_state *istate)
1881 {
1882 check_ce_order(istate);
1883 tweak_untracked_cache(istate);
1884 tweak_split_index(istate);
1885 tweak_fsmonitor(istate);
1886 }
1887
1888 static size_t estimate_cache_size_from_compressed(unsigned int entries)
1889 {
1890 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1891 }
1892
1893 static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1894 {
1895 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1896
1897 /*
1898 * Account for potential alignment differences.
1899 */
1900 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1901 return ondisk_size + entries * per_entry;
1902 }
1903
1904 struct index_entry_offset
1905 {
1906 /* starting byte offset into index file, count of index entries in this block */
1907 int offset, nr;
1908 };
1909
1910 struct index_entry_offset_table
1911 {
1912 int nr;
1913 struct index_entry_offset entries[FLEX_ARRAY];
1914 };
1915
1916 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1917 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1918
1919 static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1920 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1921
1922 struct load_index_extensions
1923 {
1924 pthread_t pthread;
1925 struct index_state *istate;
1926 const char *mmap;
1927 size_t mmap_size;
1928 unsigned long src_offset;
1929 };
1930
1931 static void *load_index_extensions(void *_data)
1932 {
1933 struct load_index_extensions *p = _data;
1934 unsigned long src_offset = p->src_offset;
1935
1936 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1937 /* After an array of active_nr index entries,
1938 * there can be arbitrary number of extended
1939 * sections, each of which is prefixed with
1940 * extension name (4-byte) and section length
1941 * in 4-byte network byte order.
1942 */
1943 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1944 if (read_index_extension(p->istate,
1945 p->mmap + src_offset,
1946 p->mmap + src_offset + 8,
1947 extsize) < 0) {
1948 munmap((void *)p->mmap, p->mmap_size);
1949 die(_("index file corrupt"));
1950 }
1951 src_offset += 8;
1952 src_offset += extsize;
1953 }
1954
1955 return NULL;
1956 }
1957
1958 /*
1959 * A helper function that will load the specified range of cache entries
1960 * from the memory mapped file and add them to the given index.
1961 */
1962 static unsigned long load_cache_entry_block(struct index_state *istate,
1963 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1964 unsigned long start_offset, const struct cache_entry *previous_ce)
1965 {
1966 int i;
1967 unsigned long src_offset = start_offset;
1968
1969 for (i = offset; i < offset + nr; i++) {
1970 struct ondisk_cache_entry *disk_ce;
1971 struct cache_entry *ce;
1972 unsigned long consumed;
1973
1974 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1975 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1976 set_index_entry(istate, i, ce);
1977
1978 src_offset += consumed;
1979 previous_ce = ce;
1980 }
1981 return src_offset - start_offset;
1982 }
1983
1984 static unsigned long load_all_cache_entries(struct index_state *istate,
1985 const char *mmap, size_t mmap_size, unsigned long src_offset)
1986 {
1987 unsigned long consumed;
1988
1989 if (istate->version == 4) {
1990 mem_pool_init(&istate->ce_mem_pool,
1991 estimate_cache_size_from_compressed(istate->cache_nr));
1992 } else {
1993 mem_pool_init(&istate->ce_mem_pool,
1994 estimate_cache_size(mmap_size, istate->cache_nr));
1995 }
1996
1997 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
1998 0, istate->cache_nr, mmap, src_offset, NULL);
1999 return consumed;
2000 }
2001
2002 /*
2003 * Mostly randomly chosen maximum thread counts: we
2004 * cap the parallelism to online_cpus() threads, and we want
2005 * to have at least 10000 cache entries per thread for it to
2006 * be worth starting a thread.
2007 */
2008
2009 #define THREAD_COST (10000)
2010
2011 struct load_cache_entries_thread_data
2012 {
2013 pthread_t pthread;
2014 struct index_state *istate;
2015 struct mem_pool *ce_mem_pool;
2016 int offset;
2017 const char *mmap;
2018 struct index_entry_offset_table *ieot;
2019 int ieot_start; /* starting index into the ieot array */
2020 int ieot_blocks; /* count of ieot entries to process */
2021 unsigned long consumed; /* return # of bytes in index file processed */
2022 };
2023
2024 /*
2025 * A thread proc to run the load_cache_entries() computation
2026 * across multiple background threads.
2027 */
2028 static void *load_cache_entries_thread(void *_data)
2029 {
2030 struct load_cache_entries_thread_data *p = _data;
2031 int i;
2032
2033 /* iterate across all ieot blocks assigned to this thread */
2034 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2035 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2036 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2037 p->offset += p->ieot->entries[i].nr;
2038 }
2039 return NULL;
2040 }
2041
2042 static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2043 unsigned long src_offset, int nr_threads, struct index_entry_offset_table *ieot)
2044 {
2045 int i, offset, ieot_blocks, ieot_start, err;
2046 struct load_cache_entries_thread_data *data;
2047 unsigned long consumed = 0;
2048
2049 /* a little sanity checking */
2050 if (istate->name_hash_initialized)
2051 BUG("the name hash isn't thread safe");
2052
2053 mem_pool_init(&istate->ce_mem_pool, 0);
2054
2055 /* ensure we have no more threads than we have blocks to process */
2056 if (nr_threads > ieot->nr)
2057 nr_threads = ieot->nr;
2058 data = xcalloc(nr_threads, sizeof(*data));
2059
2060 offset = ieot_start = 0;
2061 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2062 for (i = 0; i < nr_threads; i++) {
2063 struct load_cache_entries_thread_data *p = &data[i];
2064 int nr, j;
2065
2066 if (ieot_start + ieot_blocks > ieot->nr)
2067 ieot_blocks = ieot->nr - ieot_start;
2068
2069 p->istate = istate;
2070 p->offset = offset;
2071 p->mmap = mmap;
2072 p->ieot = ieot;
2073 p->ieot_start = ieot_start;
2074 p->ieot_blocks = ieot_blocks;
2075
2076 /* create a mem_pool for each thread */
2077 nr = 0;
2078 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2079 nr += p->ieot->entries[j].nr;
2080 if (istate->version == 4) {
2081 mem_pool_init(&p->ce_mem_pool,
2082 estimate_cache_size_from_compressed(nr));
2083 } else {
2084 mem_pool_init(&p->ce_mem_pool,
2085 estimate_cache_size(mmap_size, nr));
2086 }
2087
2088 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2089 if (err)
2090 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2091
2092 /* increment by the number of cache entries in the ieot block being processed */
2093 for (j = 0; j < ieot_blocks; j++)
2094 offset += ieot->entries[ieot_start + j].nr;
2095 ieot_start += ieot_blocks;
2096 }
2097
2098 for (i = 0; i < nr_threads; i++) {
2099 struct load_cache_entries_thread_data *p = &data[i];
2100
2101 err = pthread_join(p->pthread, NULL);
2102 if (err)
2103 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2104 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2105 consumed += p->consumed;
2106 }
2107
2108 free(data);
2109
2110 return consumed;
2111 }
2112
2113 /* remember to discard_cache() before reading a different cache! */
2114 int do_read_index(struct index_state *istate, const char *path, int must_exist)
2115 {
2116 int fd;
2117 struct stat st;
2118 unsigned long src_offset;
2119 const struct cache_header *hdr;
2120 const char *mmap;
2121 size_t mmap_size;
2122 struct load_index_extensions p;
2123 size_t extension_offset = 0;
2124 int nr_threads, cpus;
2125 struct index_entry_offset_table *ieot = NULL;
2126
2127 if (istate->initialized)
2128 return istate->cache_nr;
2129
2130 istate->timestamp.sec = 0;
2131 istate->timestamp.nsec = 0;
2132 fd = open(path, O_RDONLY);
2133 if (fd < 0) {
2134 if (!must_exist && errno == ENOENT)
2135 return 0;
2136 die_errno(_("%s: index file open failed"), path);
2137 }
2138
2139 if (fstat(fd, &st))
2140 die_errno(_("%s: cannot stat the open index"), path);
2141
2142 mmap_size = xsize_t(st.st_size);
2143 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2144 die(_("%s: index file smaller than expected"), path);
2145
2146 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2147 if (mmap == MAP_FAILED)
2148 die_errno(_("%s: unable to map index file"), path);
2149 close(fd);
2150
2151 hdr = (const struct cache_header *)mmap;
2152 if (verify_hdr(hdr, mmap_size) < 0)
2153 goto unmap;
2154
2155 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2156 istate->version = ntohl(hdr->hdr_version);
2157 istate->cache_nr = ntohl(hdr->hdr_entries);
2158 istate->cache_alloc = alloc_nr(istate->cache_nr);
2159 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2160 istate->initialized = 1;
2161
2162 p.istate = istate;
2163 p.mmap = mmap;
2164 p.mmap_size = mmap_size;
2165
2166 src_offset = sizeof(*hdr);
2167
2168 if (git_config_get_index_threads(&nr_threads))
2169 nr_threads = 1;
2170
2171 /* TODO: does creating more threads than cores help? */
2172 if (!nr_threads) {
2173 nr_threads = istate->cache_nr / THREAD_COST;
2174 cpus = online_cpus();
2175 if (nr_threads > cpus)
2176 nr_threads = cpus;
2177 }
2178
2179 if (!HAVE_THREADS)
2180 nr_threads = 1;
2181
2182 if (nr_threads > 1) {
2183 extension_offset = read_eoie_extension(mmap, mmap_size);
2184 if (extension_offset) {
2185 int err;
2186
2187 p.src_offset = extension_offset;
2188 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2189 if (err)
2190 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2191
2192 nr_threads--;
2193 }
2194 }
2195
2196 /*
2197 * Locate and read the index entry offset table so that we can use it
2198 * to multi-thread the reading of the cache entries.
2199 */
2200 if (extension_offset && nr_threads > 1)
2201 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2202
2203 if (ieot) {
2204 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, src_offset, nr_threads, ieot);
2205 free(ieot);
2206 } else {
2207 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2208 }
2209
2210 istate->timestamp.sec = st.st_mtime;
2211 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2212
2213 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2214 if (extension_offset) {
2215 int ret = pthread_join(p.pthread, NULL);
2216 if (ret)
2217 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2218 } else {
2219 p.src_offset = src_offset;
2220 load_index_extensions(&p);
2221 }
2222 munmap((void *)mmap, mmap_size);
2223 return istate->cache_nr;
2224
2225 unmap:
2226 munmap((void *)mmap, mmap_size);
2227 die(_("index file corrupt"));
2228 }
2229
2230 /*
2231 * Signal that the shared index is used by updating its mtime.
2232 *
2233 * This way, shared index can be removed if they have not been used
2234 * for some time.
2235 */
2236 static void freshen_shared_index(const char *shared_index, int warn)
2237 {
2238 if (!check_and_freshen_file(shared_index, 1) && warn)
2239 warning(_("could not freshen shared index '%s'"), shared_index);
2240 }
2241
2242 int read_index_from(struct index_state *istate, const char *path,
2243 const char *gitdir)
2244 {
2245 struct split_index *split_index;
2246 int ret;
2247 char *base_oid_hex;
2248 char *base_path;
2249
2250 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2251 if (istate->initialized)
2252 return istate->cache_nr;
2253
2254 trace_performance_enter();
2255 ret = do_read_index(istate, path, 0);
2256 trace_performance_leave("read cache %s", path);
2257
2258 split_index = istate->split_index;
2259 if (!split_index || is_null_oid(&split_index->base_oid)) {
2260 post_read_index_from(istate);
2261 return ret;
2262 }
2263
2264 trace_performance_enter();
2265 if (split_index->base)
2266 discard_index(split_index->base);
2267 else
2268 split_index->base = xcalloc(1, sizeof(*split_index->base));
2269
2270 base_oid_hex = oid_to_hex(&split_index->base_oid);
2271 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2272 ret = do_read_index(split_index->base, base_path, 1);
2273 if (!oideq(&split_index->base_oid, &split_index->base->oid))
2274 die(_("broken index, expect %s in %s, got %s"),
2275 base_oid_hex, base_path,
2276 oid_to_hex(&split_index->base->oid));
2277
2278 freshen_shared_index(base_path, 0);
2279 merge_base_index(istate);
2280 post_read_index_from(istate);
2281 trace_performance_leave("read cache %s", base_path);
2282 free(base_path);
2283 return ret;
2284 }
2285
2286 int is_index_unborn(struct index_state *istate)
2287 {
2288 return (!istate->cache_nr && !istate->timestamp.sec);
2289 }
2290
2291 int discard_index(struct index_state *istate)
2292 {
2293 /*
2294 * Cache entries in istate->cache[] should have been allocated
2295 * from the memory pool associated with this index, or from an
2296 * associated split_index. There is no need to free individual
2297 * cache entries. validate_cache_entries can detect when this
2298 * assertion does not hold.
2299 */
2300 validate_cache_entries(istate);
2301
2302 resolve_undo_clear_index(istate);
2303 istate->cache_nr = 0;
2304 istate->cache_changed = 0;
2305 istate->timestamp.sec = 0;
2306 istate->timestamp.nsec = 0;
2307 free_name_hash(istate);
2308 cache_tree_free(&(istate->cache_tree));
2309 istate->initialized = 0;
2310 FREE_AND_NULL(istate->cache);
2311 istate->cache_alloc = 0;
2312 discard_split_index(istate);
2313 free_untracked_cache(istate->untracked);
2314 istate->untracked = NULL;
2315
2316 if (istate->ce_mem_pool) {
2317 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2318 istate->ce_mem_pool = NULL;
2319 }
2320
2321 return 0;
2322 }
2323
2324 /*
2325 * Validate the cache entries of this index.
2326 * All cache entries associated with this index
2327 * should have been allocated by the memory pool
2328 * associated with this index, or by a referenced
2329 * split index.
2330 */
2331 void validate_cache_entries(const struct index_state *istate)
2332 {
2333 int i;
2334
2335 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2336 return;
2337
2338 for (i = 0; i < istate->cache_nr; i++) {
2339 if (!istate) {
2340 BUG("cache entry is not allocated from expected memory pool");
2341 } else if (!istate->ce_mem_pool ||
2342 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2343 if (!istate->split_index ||
2344 !istate->split_index->base ||
2345 !istate->split_index->base->ce_mem_pool ||
2346 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2347 BUG("cache entry is not allocated from expected memory pool");
2348 }
2349 }
2350 }
2351
2352 if (istate->split_index)
2353 validate_cache_entries(istate->split_index->base);
2354 }
2355
2356 int unmerged_index(const struct index_state *istate)
2357 {
2358 int i;
2359 for (i = 0; i < istate->cache_nr; i++) {
2360 if (ce_stage(istate->cache[i]))
2361 return 1;
2362 }
2363 return 0;
2364 }
2365
2366 int repo_index_has_changes(struct repository *repo,
2367 struct tree *tree,
2368 struct strbuf *sb)
2369 {
2370 struct index_state *istate = repo->index;
2371 struct object_id cmp;
2372 int i;
2373
2374 if (tree)
2375 cmp = tree->object.oid;
2376 if (tree || !get_oid_tree("HEAD", &cmp)) {
2377 struct diff_options opt;
2378
2379 repo_diff_setup(repo, &opt);
2380 opt.flags.exit_with_status = 1;
2381 if (!sb)
2382 opt.flags.quick = 1;
2383 do_diff_cache(&cmp, &opt);
2384 diffcore_std(&opt);
2385 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2386 if (i)
2387 strbuf_addch(sb, ' ');
2388 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2389 }
2390 diff_flush(&opt);
2391 return opt.flags.has_changes != 0;
2392 } else {
2393 for (i = 0; sb && i < istate->cache_nr; i++) {
2394 if (i)
2395 strbuf_addch(sb, ' ');
2396 strbuf_addstr(sb, istate->cache[i]->name);
2397 }
2398 return !!istate->cache_nr;
2399 }
2400 }
2401
2402 #define WRITE_BUFFER_SIZE 8192
2403 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2404 static unsigned long write_buffer_len;
2405
2406 static int ce_write_flush(git_hash_ctx *context, int fd)
2407 {
2408 unsigned int buffered = write_buffer_len;
2409 if (buffered) {
2410 the_hash_algo->update_fn(context, write_buffer, buffered);
2411 if (write_in_full(fd, write_buffer, buffered) < 0)
2412 return -1;
2413 write_buffer_len = 0;
2414 }
2415 return 0;
2416 }
2417
2418 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2419 {
2420 while (len) {
2421 unsigned int buffered = write_buffer_len;
2422 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2423 if (partial > len)
2424 partial = len;
2425 memcpy(write_buffer + buffered, data, partial);
2426 buffered += partial;
2427 if (buffered == WRITE_BUFFER_SIZE) {
2428 write_buffer_len = buffered;
2429 if (ce_write_flush(context, fd))
2430 return -1;
2431 buffered = 0;
2432 }
2433 write_buffer_len = buffered;
2434 len -= partial;
2435 data = (char *) data + partial;
2436 }
2437 return 0;
2438 }
2439
2440 static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2441 int fd, unsigned int ext, unsigned int sz)
2442 {
2443 ext = htonl(ext);
2444 sz = htonl(sz);
2445 if (eoie_context) {
2446 the_hash_algo->update_fn(eoie_context, &ext, 4);
2447 the_hash_algo->update_fn(eoie_context, &sz, 4);
2448 }
2449 return ((ce_write(context, fd, &ext, 4) < 0) ||
2450 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2451 }
2452
2453 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2454 {
2455 unsigned int left = write_buffer_len;
2456
2457 if (left) {
2458 write_buffer_len = 0;
2459 the_hash_algo->update_fn(context, write_buffer, left);
2460 }
2461
2462 /* Flush first if not enough space for hash signature */
2463 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2464 if (write_in_full(fd, write_buffer, left) < 0)
2465 return -1;
2466 left = 0;
2467 }
2468
2469 /* Append the hash signature at the end */
2470 the_hash_algo->final_fn(write_buffer + left, context);
2471 hashcpy(hash, write_buffer + left);
2472 left += the_hash_algo->rawsz;
2473 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2474 }
2475
2476 static void ce_smudge_racily_clean_entry(struct index_state *istate,
2477 struct cache_entry *ce)
2478 {
2479 /*
2480 * The only thing we care about in this function is to smudge the
2481 * falsely clean entry due to touch-update-touch race, so we leave
2482 * everything else as they are. We are called for entries whose
2483 * ce_stat_data.sd_mtime match the index file mtime.
2484 *
2485 * Note that this actually does not do much for gitlinks, for
2486 * which ce_match_stat_basic() always goes to the actual
2487 * contents. The caller checks with is_racy_timestamp() which
2488 * always says "no" for gitlinks, so we are not called for them ;-)
2489 */
2490 struct stat st;
2491
2492 if (lstat(ce->name, &st) < 0)
2493 return;
2494 if (ce_match_stat_basic(ce, &st))
2495 return;
2496 if (ce_modified_check_fs(istate, ce, &st)) {
2497 /* This is "racily clean"; smudge it. Note that this
2498 * is a tricky code. At first glance, it may appear
2499 * that it can break with this sequence:
2500 *
2501 * $ echo xyzzy >frotz
2502 * $ git-update-index --add frotz
2503 * $ : >frotz
2504 * $ sleep 3
2505 * $ echo filfre >nitfol
2506 * $ git-update-index --add nitfol
2507 *
2508 * but it does not. When the second update-index runs,
2509 * it notices that the entry "frotz" has the same timestamp
2510 * as index, and if we were to smudge it by resetting its
2511 * size to zero here, then the object name recorded
2512 * in index is the 6-byte file but the cached stat information
2513 * becomes zero --- which would then match what we would
2514 * obtain from the filesystem next time we stat("frotz").
2515 *
2516 * However, the second update-index, before calling
2517 * this function, notices that the cached size is 6
2518 * bytes and what is on the filesystem is an empty
2519 * file, and never calls us, so the cached size information
2520 * for "frotz" stays 6 which does not match the filesystem.
2521 */
2522 ce->ce_stat_data.sd_size = 0;
2523 }
2524 }
2525
2526 /* Copy miscellaneous fields but not the name */
2527 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2528 struct cache_entry *ce)
2529 {
2530 short flags;
2531
2532 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2533 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2534 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2535 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2536 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2537 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2538 ondisk->mode = htonl(ce->ce_mode);
2539 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2540 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2541 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2542 hashcpy(ondisk->sha1, ce->oid.hash);
2543
2544 flags = ce->ce_flags & ~CE_NAMEMASK;
2545 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2546 ondisk->flags = htons(flags);
2547 if (ce->ce_flags & CE_EXTENDED) {
2548 struct ondisk_cache_entry_extended *ondisk2;
2549 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2550 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2551 }
2552 }
2553
2554 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2555 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2556 {
2557 int size;
2558 int result;
2559 unsigned int saved_namelen;
2560 int stripped_name = 0;
2561 static unsigned char padding[8] = { 0x00 };
2562
2563 if (ce->ce_flags & CE_STRIP_NAME) {
2564 saved_namelen = ce_namelen(ce);
2565 ce->ce_namelen = 0;
2566 stripped_name = 1;
2567 }
2568
2569 if (ce->ce_flags & CE_EXTENDED)
2570 size = offsetof(struct ondisk_cache_entry_extended, name);
2571 else
2572 size = offsetof(struct ondisk_cache_entry, name);
2573
2574 if (!previous_name) {
2575 int len = ce_namelen(ce);
2576 copy_cache_entry_to_ondisk(ondisk, ce);
2577 result = ce_write(c, fd, ondisk, size);
2578 if (!result)
2579 result = ce_write(c, fd, ce->name, len);
2580 if (!result)
2581 result = ce_write(c, fd, padding, align_padding_size(size, len));
2582 } else {
2583 int common, to_remove, prefix_size;
2584 unsigned char to_remove_vi[16];
2585 for (common = 0;
2586 (ce->name[common] &&
2587 common < previous_name->len &&
2588 ce->name[common] == previous_name->buf[common]);
2589 common++)
2590 ; /* still matching */
2591 to_remove = previous_name->len - common;
2592 prefix_size = encode_varint(to_remove, to_remove_vi);
2593
2594 copy_cache_entry_to_ondisk(ondisk, ce);
2595 result = ce_write(c, fd, ondisk, size);
2596 if (!result)
2597 result = ce_write(c, fd, to_remove_vi, prefix_size);
2598 if (!result)
2599 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2600 if (!result)
2601 result = ce_write(c, fd, padding, 1);
2602
2603 strbuf_splice(previous_name, common, to_remove,
2604 ce->name + common, ce_namelen(ce) - common);
2605 }
2606 if (stripped_name) {
2607 ce->ce_namelen = saved_namelen;
2608 ce->ce_flags &= ~CE_STRIP_NAME;
2609 }
2610
2611 return result;
2612 }
2613
2614 /*
2615 * This function verifies if index_state has the correct sha1 of the
2616 * index file. Don't die if we have any other failure, just return 0.
2617 */
2618 static int verify_index_from(const struct index_state *istate, const char *path)
2619 {
2620 int fd;
2621 ssize_t n;
2622 struct stat st;
2623 unsigned char hash[GIT_MAX_RAWSZ];
2624
2625 if (!istate->initialized)
2626 return 0;
2627
2628 fd = open(path, O_RDONLY);
2629 if (fd < 0)
2630 return 0;
2631
2632 if (fstat(fd, &st))
2633 goto out;
2634
2635 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2636 goto out;
2637
2638 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2639 if (n != the_hash_algo->rawsz)
2640 goto out;
2641
2642 if (!hasheq(istate->oid.hash, hash))
2643 goto out;
2644
2645 close(fd);
2646 return 1;
2647
2648 out:
2649 close(fd);
2650 return 0;
2651 }
2652
2653 static int repo_verify_index(struct repository *repo)
2654 {
2655 return verify_index_from(repo->index, repo->index_file);
2656 }
2657
2658 static int has_racy_timestamp(struct index_state *istate)
2659 {
2660 int entries = istate->cache_nr;
2661 int i;
2662
2663 for (i = 0; i < entries; i++) {
2664 struct cache_entry *ce = istate->cache[i];
2665 if (is_racy_timestamp(istate, ce))
2666 return 1;
2667 }
2668 return 0;
2669 }
2670
2671 void repo_update_index_if_able(struct repository *repo,
2672 struct lock_file *lockfile)
2673 {
2674 if ((repo->index->cache_changed ||
2675 has_racy_timestamp(repo->index)) &&
2676 repo_verify_index(repo))
2677 write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2678 else
2679 rollback_lock_file(lockfile);
2680 }
2681
2682 static int record_eoie(void)
2683 {
2684 int val;
2685
2686 if (!git_config_get_bool("index.recordendofindexentries", &val))
2687 return val;
2688
2689 /*
2690 * As a convenience, the end of index entries extension
2691 * used for threading is written by default if the user
2692 * explicitly requested threaded index reads.
2693 */
2694 return !git_config_get_index_threads(&val) && val != 1;
2695 }
2696
2697 static int record_ieot(void)
2698 {
2699 int val;
2700
2701 if (!git_config_get_bool("index.recordoffsettable", &val))
2702 return val;
2703
2704 /*
2705 * As a convenience, the offset table used for threading is
2706 * written by default if the user explicitly requested
2707 * threaded index reads.
2708 */
2709 return !git_config_get_index_threads(&val) && val != 1;
2710 }
2711
2712 /*
2713 * On success, `tempfile` is closed. If it is the temporary file
2714 * of a `struct lock_file`, we will therefore effectively perform
2715 * a 'close_lock_file_gently()`. Since that is an implementation
2716 * detail of lockfiles, callers of `do_write_index()` should not
2717 * rely on it.
2718 */
2719 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2720 int strip_extensions)
2721 {
2722 uint64_t start = getnanotime();
2723 int newfd = tempfile->fd;
2724 git_hash_ctx c, eoie_c;
2725 struct cache_header hdr;
2726 int i, err = 0, removed, extended, hdr_version;
2727 struct cache_entry **cache = istate->cache;
2728 int entries = istate->cache_nr;
2729 struct stat st;
2730 struct ondisk_cache_entry_extended ondisk;
2731 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2732 int drop_cache_tree = istate->drop_cache_tree;
2733 off_t offset;
2734 int ieot_entries = 1;
2735 struct index_entry_offset_table *ieot = NULL;
2736 int nr, nr_threads;
2737
2738 for (i = removed = extended = 0; i < entries; i++) {
2739 if (cache[i]->ce_flags & CE_REMOVE)
2740 removed++;
2741
2742 /* reduce extended entries if possible */
2743 cache[i]->ce_flags &= ~CE_EXTENDED;
2744 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2745 extended++;
2746 cache[i]->ce_flags |= CE_EXTENDED;
2747 }
2748 }
2749
2750 if (!istate->version) {
2751 istate->version = get_index_format_default();
2752 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2753 init_split_index(istate);
2754 }
2755
2756 /* demote version 3 to version 2 when the latter suffices */
2757 if (istate->version == 3 || istate->version == 2)
2758 istate->version = extended ? 3 : 2;
2759
2760 hdr_version = istate->version;
2761
2762 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2763 hdr.hdr_version = htonl(hdr_version);
2764 hdr.hdr_entries = htonl(entries - removed);
2765
2766 the_hash_algo->init_fn(&c);
2767 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2768 return -1;
2769
2770 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))
2771 nr_threads = 1;
2772
2773 if (nr_threads != 1 && record_ieot()) {
2774 int ieot_blocks, cpus;
2775
2776 /*
2777 * ensure default number of ieot blocks maps evenly to the
2778 * default number of threads that will process them leaving
2779 * room for the thread to load the index extensions.
2780 */
2781 if (!nr_threads) {
2782 ieot_blocks = istate->cache_nr / THREAD_COST;
2783 cpus = online_cpus();
2784 if (ieot_blocks > cpus - 1)
2785 ieot_blocks = cpus - 1;
2786 } else {
2787 ieot_blocks = nr_threads;
2788 if (ieot_blocks > istate->cache_nr)
2789 ieot_blocks = istate->cache_nr;
2790 }
2791
2792 /*
2793 * no reason to write out the IEOT extension if we don't
2794 * have enough blocks to utilize multi-threading
2795 */
2796 if (ieot_blocks > 1) {
2797 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2798 + (ieot_blocks * sizeof(struct index_entry_offset)));
2799 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2800 }
2801 }
2802
2803 offset = lseek(newfd, 0, SEEK_CUR);
2804 if (offset < 0) {
2805 free(ieot);
2806 return -1;
2807 }
2808 offset += write_buffer_len;
2809 nr = 0;
2810 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2811
2812 for (i = 0; i < entries; i++) {
2813 struct cache_entry *ce = cache[i];
2814 if (ce->ce_flags & CE_REMOVE)
2815 continue;
2816 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2817 ce_smudge_racily_clean_entry(istate, ce);
2818 if (is_null_oid(&ce->oid)) {
2819 static const char msg[] = "cache entry has null sha1: %s";
2820 static int allow = -1;
2821
2822 if (allow < 0)
2823 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2824 if (allow)
2825 warning(msg, ce->name);
2826 else
2827 err = error(msg, ce->name);
2828
2829 drop_cache_tree = 1;
2830 }
2831 if (ieot && i && (i % ieot_entries == 0)) {
2832 ieot->entries[ieot->nr].nr = nr;
2833 ieot->entries[ieot->nr].offset = offset;
2834 ieot->nr++;
2835 /*
2836 * If we have a V4 index, set the first byte to an invalid
2837 * character to ensure there is nothing common with the previous
2838 * entry
2839 */
2840 if (previous_name)
2841 previous_name->buf[0] = 0;
2842 nr = 0;
2843 offset = lseek(newfd, 0, SEEK_CUR);
2844 if (offset < 0) {
2845 free(ieot);
2846 return -1;
2847 }
2848 offset += write_buffer_len;
2849 }
2850 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2851 err = -1;
2852
2853 if (err)
2854 break;
2855 nr++;
2856 }
2857 if (ieot && nr) {
2858 ieot->entries[ieot->nr].nr = nr;
2859 ieot->entries[ieot->nr].offset = offset;
2860 ieot->nr++;
2861 }
2862 strbuf_release(&previous_name_buf);
2863
2864 if (err) {
2865 free(ieot);
2866 return err;
2867 }
2868
2869 /* Write extension data here */
2870 offset = lseek(newfd, 0, SEEK_CUR);
2871 if (offset < 0) {
2872 free(ieot);
2873 return -1;
2874 }
2875 offset += write_buffer_len;
2876 the_hash_algo->init_fn(&eoie_c);
2877
2878 /*
2879 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2880 * can minimize the number of extensions we have to scan through to
2881 * find it during load. Write it out regardless of the
2882 * strip_extensions parameter as we need it when loading the shared
2883 * index.
2884 */
2885 if (ieot) {
2886 struct strbuf sb = STRBUF_INIT;
2887
2888 write_ieot_extension(&sb, ieot);
2889 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2890 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2891 strbuf_release(&sb);
2892 free(ieot);
2893 if (err)
2894 return -1;
2895 }
2896
2897 if (!strip_extensions && istate->split_index) {
2898 struct strbuf sb = STRBUF_INIT;
2899
2900 err = write_link_extension(&sb, istate) < 0 ||
2901 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2902 sb.len) < 0 ||
2903 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2904 strbuf_release(&sb);
2905 if (err)
2906 return -1;
2907 }
2908 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2909 struct strbuf sb = STRBUF_INIT;
2910
2911 cache_tree_write(&sb, istate->cache_tree);
2912 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2913 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2914 strbuf_release(&sb);
2915 if (err)
2916 return -1;
2917 }
2918 if (!strip_extensions && istate->resolve_undo) {
2919 struct strbuf sb = STRBUF_INIT;
2920
2921 resolve_undo_write(&sb, istate->resolve_undo);
2922 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2923 sb.len) < 0
2924 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2925 strbuf_release(&sb);
2926 if (err)
2927 return -1;
2928 }
2929 if (!strip_extensions && istate->untracked) {
2930 struct strbuf sb = STRBUF_INIT;
2931
2932 write_untracked_extension(&sb, istate->untracked);
2933 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2934 sb.len) < 0 ||
2935 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2936 strbuf_release(&sb);
2937 if (err)
2938 return -1;
2939 }
2940 if (!strip_extensions && istate->fsmonitor_last_update) {
2941 struct strbuf sb = STRBUF_INIT;
2942
2943 write_fsmonitor_extension(&sb, istate);
2944 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2945 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2946 strbuf_release(&sb);
2947 if (err)
2948 return -1;
2949 }
2950
2951 /*
2952 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2953 * so that it can be found and processed before all the index entries are
2954 * read. Write it out regardless of the strip_extensions parameter as we need it
2955 * when loading the shared index.
2956 */
2957 if (offset && record_eoie()) {
2958 struct strbuf sb = STRBUF_INIT;
2959
2960 write_eoie_extension(&sb, &eoie_c, offset);
2961 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
2962 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2963 strbuf_release(&sb);
2964 if (err)
2965 return -1;
2966 }
2967
2968 if (ce_flush(&c, newfd, istate->oid.hash))
2969 return -1;
2970 if (close_tempfile_gently(tempfile)) {
2971 error(_("could not close '%s'"), tempfile->filename.buf);
2972 return -1;
2973 }
2974 if (stat(tempfile->filename.buf, &st))
2975 return -1;
2976 istate->timestamp.sec = (unsigned int)st.st_mtime;
2977 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2978 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
2979 return 0;
2980 }
2981
2982 void set_alternate_index_output(const char *name)
2983 {
2984 alternate_index_output = name;
2985 }
2986
2987 static int commit_locked_index(struct lock_file *lk)
2988 {
2989 if (alternate_index_output)
2990 return commit_lock_file_to(lk, alternate_index_output);
2991 else
2992 return commit_lock_file(lk);
2993 }
2994
2995 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2996 unsigned flags)
2997 {
2998 int ret = do_write_index(istate, lock->tempfile, 0);
2999 if (ret)
3000 return ret;
3001 if (flags & COMMIT_LOCK)
3002 return commit_locked_index(lock);
3003 return close_lock_file_gently(lock);
3004 }
3005
3006 static int write_split_index(struct index_state *istate,
3007 struct lock_file *lock,
3008 unsigned flags)
3009 {
3010 int ret;
3011 prepare_to_write_split_index(istate);
3012 ret = do_write_locked_index(istate, lock, flags);
3013 finish_writing_split_index(istate);
3014 return ret;
3015 }
3016
3017 static const char *shared_index_expire = "2.weeks.ago";
3018
3019 static unsigned long get_shared_index_expire_date(void)
3020 {
3021 static unsigned long shared_index_expire_date;
3022 static int shared_index_expire_date_prepared;
3023
3024 if (!shared_index_expire_date_prepared) {
3025 git_config_get_expiry("splitindex.sharedindexexpire",
3026 &shared_index_expire);
3027 shared_index_expire_date = approxidate(shared_index_expire);
3028 shared_index_expire_date_prepared = 1;
3029 }
3030
3031 return shared_index_expire_date;
3032 }
3033
3034 static int should_delete_shared_index(const char *shared_index_path)
3035 {
3036 struct stat st;
3037 unsigned long expiration;
3038
3039 /* Check timestamp */
3040 expiration = get_shared_index_expire_date();
3041 if (!expiration)
3042 return 0;
3043 if (stat(shared_index_path, &st))
3044 return error_errno(_("could not stat '%s'"), shared_index_path);
3045 if (st.st_mtime > expiration)
3046 return 0;
3047
3048 return 1;
3049 }
3050
3051 static int clean_shared_index_files(const char *current_hex)
3052 {
3053 struct dirent *de;
3054 DIR *dir = opendir(get_git_dir());
3055
3056 if (!dir)
3057 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3058
3059 while ((de = readdir(dir)) != NULL) {
3060 const char *sha1_hex;
3061 const char *shared_index_path;
3062 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3063 continue;
3064 if (!strcmp(sha1_hex, current_hex))
3065 continue;
3066 shared_index_path = git_path("%s", de->d_name);
3067 if (should_delete_shared_index(shared_index_path) > 0 &&
3068 unlink(shared_index_path))
3069 warning_errno(_("unable to unlink: %s"), shared_index_path);
3070 }
3071 closedir(dir);
3072
3073 return 0;
3074 }
3075
3076 static int write_shared_index(struct index_state *istate,
3077 struct tempfile **temp)
3078 {
3079 struct split_index *si = istate->split_index;
3080 int ret;
3081
3082 move_cache_to_base_index(istate);
3083 ret = do_write_index(si->base, *temp, 1);
3084 if (ret)
3085 return ret;
3086 ret = adjust_shared_perm(get_tempfile_path(*temp));
3087 if (ret) {
3088 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3089 return ret;
3090 }
3091 ret = rename_tempfile(temp,
3092 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3093 if (!ret) {
3094 oidcpy(&si->base_oid, &si->base->oid);
3095 clean_shared_index_files(oid_to_hex(&si->base->oid));
3096 }
3097
3098 return ret;
3099 }
3100
3101 static const int default_max_percent_split_change = 20;
3102
3103 static int too_many_not_shared_entries(struct index_state *istate)
3104 {
3105 int i, not_shared = 0;
3106 int max_split = git_config_get_max_percent_split_change();
3107
3108 switch (max_split) {
3109 case -1:
3110 /* not or badly configured: use the default value */
3111 max_split = default_max_percent_split_change;
3112 break;
3113 case 0:
3114 return 1; /* 0% means always write a new shared index */
3115 case 100:
3116 return 0; /* 100% means never write a new shared index */
3117 default:
3118 break; /* just use the configured value */
3119 }
3120
3121 /* Count not shared entries */
3122 for (i = 0; i < istate->cache_nr; i++) {
3123 struct cache_entry *ce = istate->cache[i];
3124 if (!ce->index)
3125 not_shared++;
3126 }
3127
3128 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3129 }
3130
3131 int write_locked_index(struct index_state *istate, struct lock_file *lock,
3132 unsigned flags)
3133 {
3134 int new_shared_index, ret;
3135 struct split_index *si = istate->split_index;
3136
3137 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3138 cache_tree_verify(the_repository, istate);
3139
3140 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3141 if (flags & COMMIT_LOCK)
3142 rollback_lock_file(lock);
3143 return 0;
3144 }
3145
3146 if (istate->fsmonitor_last_update)
3147 fill_fsmonitor_bitmap(istate);
3148
3149 if (!si || alternate_index_output ||
3150 (istate->cache_changed & ~EXTMASK)) {
3151 if (si)
3152 oidclr(&si->base_oid);
3153 ret = do_write_locked_index(istate, lock, flags);
3154 goto out;
3155 }
3156
3157 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3158 int v = si->base_oid.hash[0];
3159 if ((v & 15) < 6)
3160 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3161 }
3162 if (too_many_not_shared_entries(istate))
3163 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3164
3165 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3166
3167 if (new_shared_index) {
3168 struct tempfile *temp;
3169 int saved_errno;
3170
3171 /* Same initial permissions as the main .git/index file */
3172 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3173 if (!temp) {
3174 oidclr(&si->base_oid);
3175 ret = do_write_locked_index(istate, lock, flags);
3176 goto out;
3177 }
3178 ret = write_shared_index(istate, &temp);
3179
3180 saved_errno = errno;
3181 if (is_tempfile_active(temp))
3182 delete_tempfile(&temp);
3183 errno = saved_errno;
3184
3185 if (ret)
3186 goto out;
3187 }
3188
3189 ret = write_split_index(istate, lock, flags);
3190
3191 /* Freshen the shared index only if the split-index was written */
3192 if (!ret && !new_shared_index) {
3193 const char *shared_index = git_path("sharedindex.%s",
3194 oid_to_hex(&si->base_oid));
3195 freshen_shared_index(shared_index, 1);
3196 }
3197
3198 out:
3199 if (flags & COMMIT_LOCK)
3200 rollback_lock_file(lock);
3201 return ret;
3202 }
3203
3204 /*
3205 * Read the index file that is potentially unmerged into given
3206 * index_state, dropping any unmerged entries to stage #0 (potentially
3207 * resulting in a path appearing as both a file and a directory in the
3208 * index; the caller is responsible to clear out the extra entries
3209 * before writing the index to a tree). Returns true if the index is
3210 * unmerged. Callers who want to refuse to work from an unmerged
3211 * state can call this and check its return value, instead of calling
3212 * read_cache().
3213 */
3214 int repo_read_index_unmerged(struct repository *repo)
3215 {
3216 struct index_state *istate;
3217 int i;
3218 int unmerged = 0;
3219
3220 repo_read_index(repo);
3221 istate = repo->index;
3222 for (i = 0; i < istate->cache_nr; i++) {
3223 struct cache_entry *ce = istate->cache[i];
3224 struct cache_entry *new_ce;
3225 int len;
3226
3227 if (!ce_stage(ce))
3228 continue;
3229 unmerged = 1;
3230 len = ce_namelen(ce);
3231 new_ce = make_empty_cache_entry(istate, len);
3232 memcpy(new_ce->name, ce->name, len);
3233 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3234 new_ce->ce_namelen = len;
3235 new_ce->ce_mode = ce->ce_mode;
3236 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3237 return error(_("%s: cannot drop to stage #0"),
3238 new_ce->name);
3239 }
3240 return unmerged;
3241 }
3242
3243 /*
3244 * Returns 1 if the path is an "other" path with respect to
3245 * the index; that is, the path is not mentioned in the index at all,
3246 * either as a file, a directory with some files in the index,
3247 * or as an unmerged entry.
3248 *
3249 * We helpfully remove a trailing "/" from directories so that
3250 * the output of read_directory can be used as-is.
3251 */
3252 int index_name_is_other(const struct index_state *istate, const char *name,
3253 int namelen)
3254 {
3255 int pos;
3256 if (namelen && name[namelen - 1] == '/')
3257 namelen--;
3258 pos = index_name_pos(istate, name, namelen);
3259 if (0 <= pos)
3260 return 0; /* exact match */
3261 pos = -pos - 1;
3262 if (pos < istate->cache_nr) {
3263 struct cache_entry *ce = istate->cache[pos];
3264 if (ce_namelen(ce) == namelen &&
3265 !memcmp(ce->name, name, namelen))
3266 return 0; /* Yup, this one exists unmerged */
3267 }
3268 return 1;
3269 }
3270
3271 void *read_blob_data_from_index(const struct index_state *istate,
3272 const char *path, unsigned long *size)
3273 {
3274 int pos, len;
3275 unsigned long sz;
3276 enum object_type type;
3277 void *data;
3278
3279 len = strlen(path);
3280 pos = index_name_pos(istate, path, len);
3281 if (pos < 0) {
3282 /*
3283 * We might be in the middle of a merge, in which
3284 * case we would read stage #2 (ours).
3285 */
3286 int i;
3287 for (i = -pos - 1;
3288 (pos < 0 && i < istate->cache_nr &&
3289 !strcmp(istate->cache[i]->name, path));
3290 i++)
3291 if (ce_stage(istate->cache[i]) == 2)
3292 pos = i;
3293 }
3294 if (pos < 0)
3295 return NULL;
3296 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3297 if (!data || type != OBJ_BLOB) {
3298 free(data);
3299 return NULL;
3300 }
3301 if (size)
3302 *size = sz;
3303 return data;
3304 }
3305
3306 void stat_validity_clear(struct stat_validity *sv)
3307 {
3308 FREE_AND_NULL(sv->sd);
3309 }
3310
3311 int stat_validity_check(struct stat_validity *sv, const char *path)
3312 {
3313 struct stat st;
3314
3315 if (stat(path, &st) < 0)
3316 return sv->sd == NULL;
3317 if (!sv->sd)
3318 return 0;
3319 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3320 }
3321
3322 void stat_validity_update(struct stat_validity *sv, int fd)
3323 {
3324 struct stat st;
3325
3326 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3327 stat_validity_clear(sv);
3328 else {
3329 if (!sv->sd)
3330 sv->sd = xcalloc(1, sizeof(struct stat_data));
3331 fill_stat_data(sv->sd, &st);
3332 }
3333 }
3334
3335 void move_index_extensions(struct index_state *dst, struct index_state *src)
3336 {
3337 dst->untracked = src->untracked;
3338 src->untracked = NULL;
3339 dst->cache_tree = src->cache_tree;
3340 src->cache_tree = NULL;
3341 }
3342
3343 struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3344 struct index_state *istate)
3345 {
3346 unsigned int size = ce_size(ce);
3347 int mem_pool_allocated;
3348 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3349 mem_pool_allocated = new_entry->mem_pool_allocated;
3350
3351 memcpy(new_entry, ce, size);
3352 new_entry->mem_pool_allocated = mem_pool_allocated;
3353 return new_entry;
3354 }
3355
3356 void discard_cache_entry(struct cache_entry *ce)
3357 {
3358 if (ce && should_validate_cache_entries())
3359 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3360
3361 if (ce && ce->mem_pool_allocated)
3362 return;
3363
3364 free(ce);
3365 }
3366
3367 int should_validate_cache_entries(void)
3368 {
3369 static int validate_index_cache_entries = -1;
3370
3371 if (validate_index_cache_entries < 0) {
3372 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3373 validate_index_cache_entries = 1;
3374 else
3375 validate_index_cache_entries = 0;
3376 }
3377
3378 return validate_index_cache_entries;
3379 }
3380
3381 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3382 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3383
3384 static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3385 {
3386 /*
3387 * The end of index entries (EOIE) extension is guaranteed to be last
3388 * so that it can be found by scanning backwards from the EOF.
3389 *
3390 * "EOIE"
3391 * <4-byte length>
3392 * <4-byte offset>
3393 * <20-byte hash>
3394 */
3395 const char *index, *eoie;
3396 uint32_t extsize;
3397 size_t offset, src_offset;
3398 unsigned char hash[GIT_MAX_RAWSZ];
3399 git_hash_ctx c;
3400
3401 /* ensure we have an index big enough to contain an EOIE extension */
3402 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3403 return 0;
3404
3405 /* validate the extension signature */
3406 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3407 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3408 return 0;
3409 index += sizeof(uint32_t);
3410
3411 /* validate the extension size */
3412 extsize = get_be32(index);
3413 if (extsize != EOIE_SIZE)
3414 return 0;
3415 index += sizeof(uint32_t);
3416
3417 /*
3418 * Validate the offset we're going to look for the first extension
3419 * signature is after the index header and before the eoie extension.
3420 */
3421 offset = get_be32(index);
3422 if (mmap + offset < mmap + sizeof(struct cache_header))
3423 return 0;
3424 if (mmap + offset >= eoie)
3425 return 0;
3426 index += sizeof(uint32_t);
3427
3428 /*
3429 * The hash is computed over extension types and their sizes (but not
3430 * their contents). E.g. if we have "TREE" extension that is N-bytes
3431 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3432 * then the hash would be:
3433 *
3434 * SHA-1("TREE" + <binary representation of N> +
3435 * "REUC" + <binary representation of M>)
3436 */
3437 src_offset = offset;
3438 the_hash_algo->init_fn(&c);
3439 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3440 /* After an array of active_nr index entries,
3441 * there can be arbitrary number of extended
3442 * sections, each of which is prefixed with
3443 * extension name (4-byte) and section length
3444 * in 4-byte network byte order.
3445 */
3446 uint32_t extsize;
3447 memcpy(&extsize, mmap + src_offset + 4, 4);
3448 extsize = ntohl(extsize);
3449
3450 /* verify the extension size isn't so large it will wrap around */
3451 if (src_offset + 8 + extsize < src_offset)
3452 return 0;
3453
3454 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3455
3456 src_offset += 8;
3457 src_offset += extsize;
3458 }
3459 the_hash_algo->final_fn(hash, &c);
3460 if (!hasheq(hash, (const unsigned char *)index))
3461 return 0;
3462
3463 /* Validate that the extension offsets returned us back to the eoie extension. */
3464 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3465 return 0;
3466
3467 return offset;
3468 }
3469
3470 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3471 {
3472 uint32_t buffer;
3473 unsigned char hash[GIT_MAX_RAWSZ];
3474
3475 /* offset */
3476 put_be32(&buffer, offset);
3477 strbuf_add(sb, &buffer, sizeof(uint32_t));
3478
3479 /* hash */
3480 the_hash_algo->final_fn(hash, eoie_context);
3481 strbuf_add(sb, hash, the_hash_algo->rawsz);
3482 }
3483
3484 #define IEOT_VERSION (1)
3485
3486 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3487 {
3488 const char *index = NULL;
3489 uint32_t extsize, ext_version;
3490 struct index_entry_offset_table *ieot;
3491 int i, nr;
3492
3493 /* find the IEOT extension */
3494 if (!offset)
3495 return NULL;
3496 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3497 extsize = get_be32(mmap + offset + 4);
3498 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3499 index = mmap + offset + 4 + 4;
3500 break;
3501 }
3502 offset += 8;
3503 offset += extsize;
3504 }
3505 if (!index)
3506 return NULL;
3507
3508 /* validate the version is IEOT_VERSION */
3509 ext_version = get_be32(index);
3510 if (ext_version != IEOT_VERSION) {
3511 error("invalid IEOT version %d", ext_version);
3512 return NULL;
3513 }
3514 index += sizeof(uint32_t);
3515
3516 /* extension size - version bytes / bytes per entry */
3517 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3518 if (!nr) {
3519 error("invalid number of IEOT entries %d", nr);
3520 return NULL;
3521 }
3522 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3523 + (nr * sizeof(struct index_entry_offset)));
3524 ieot->nr = nr;
3525 for (i = 0; i < nr; i++) {
3526 ieot->entries[i].offset = get_be32(index);
3527 index += sizeof(uint32_t);
3528 ieot->entries[i].nr = get_be32(index);
3529 index += sizeof(uint32_t);
3530 }
3531
3532 return ieot;
3533 }
3534
3535 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3536 {
3537 uint32_t buffer;
3538 int i;
3539
3540 /* version */
3541 put_be32(&buffer, IEOT_VERSION);
3542 strbuf_add(sb, &buffer, sizeof(uint32_t));
3543
3544 /* ieot */
3545 for (i = 0; i < ieot->nr; i++) {
3546
3547 /* offset */
3548 put_be32(&buffer, ieot->entries[i].offset);
3549 strbuf_add(sb, &buffer, sizeof(uint32_t));
3550
3551 /* count */
3552 put_be32(&buffer, ieot->entries[i].nr);
3553 strbuf_add(sb, &buffer, sizeof(uint32_t));
3554 }
3555 }