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