Merge branch 'mh/notes-duplicate-entries'
[git/git.git] / read-cache.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #include "cache.h"
7 #include "config.h"
8 #include "diff.h"
9 #include "diffcore.h"
10 #include "tempfile.h"
11 #include "lockfile.h"
12 #include "cache-tree.h"
13 #include "refs.h"
14 #include "dir.h"
15 #include "object-store.h"
16 #include "tree.h"
17 #include "commit.h"
18 #include "blob.h"
19 #include "resolve-undo.h"
20 #include "run-command.h"
21 #include "strbuf.h"
22 #include "varint.h"
23 #include "split-index.h"
24 #include "utf8.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
27 #include "progress.h"
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 static const char *alternate_index_output;
99
100 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
101 {
102 istate->cache[nr] = ce;
103 add_name_hash(istate, ce);
104 }
105
106 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
107 {
108 struct cache_entry *old = istate->cache[nr];
109
110 replace_index_entry_in_base(istate, old, ce);
111 remove_name_hash(istate, old);
112 discard_cache_entry(old);
113 ce->ce_flags &= ~CE_HASHED;
114 set_index_entry(istate, nr, ce);
115 ce->ce_flags |= CE_UPDATE_IN_BASE;
116 mark_fsmonitor_invalid(istate, ce);
117 istate->cache_changed |= CE_ENTRY_CHANGED;
118 }
119
120 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
121 {
122 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
123 int namelen = strlen(new_name);
124
125 new_entry = make_empty_cache_entry(istate, namelen);
126 copy_cache_entry(new_entry, old_entry);
127 new_entry->ce_flags &= ~CE_HASHED;
128 new_entry->ce_namelen = namelen;
129 new_entry->index = 0;
130 memcpy(new_entry->name, new_name, namelen + 1);
131
132 cache_tree_invalidate_path(istate, old_entry->name);
133 untracked_cache_remove_from_index(istate, old_entry->name);
134 remove_index_entry_at(istate, nr);
135 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
136 }
137
138 void fill_stat_data(struct stat_data *sd, struct stat *st)
139 {
140 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
141 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
142 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
143 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
144 sd->sd_dev = st->st_dev;
145 sd->sd_ino = st->st_ino;
146 sd->sd_uid = st->st_uid;
147 sd->sd_gid = st->st_gid;
148 sd->sd_size = st->st_size;
149 }
150
151 int match_stat_data(const struct stat_data *sd, struct stat *st)
152 {
153 int changed = 0;
154
155 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
156 changed |= MTIME_CHANGED;
157 if (trust_ctime && check_stat &&
158 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
159 changed |= CTIME_CHANGED;
160
161 #ifdef USE_NSEC
162 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
163 changed |= MTIME_CHANGED;
164 if (trust_ctime && check_stat &&
165 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
166 changed |= CTIME_CHANGED;
167 #endif
168
169 if (check_stat) {
170 if (sd->sd_uid != (unsigned int) st->st_uid ||
171 sd->sd_gid != (unsigned int) st->st_gid)
172 changed |= OWNER_CHANGED;
173 if (sd->sd_ino != (unsigned int) st->st_ino)
174 changed |= INODE_CHANGED;
175 }
176
177 #ifdef USE_STDEV
178 /*
179 * st_dev breaks on network filesystems where different
180 * clients will have different views of what "device"
181 * the filesystem is on
182 */
183 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
184 changed |= INODE_CHANGED;
185 #endif
186
187 if (sd->sd_size != (unsigned int) st->st_size)
188 changed |= DATA_CHANGED;
189
190 return changed;
191 }
192
193 /*
194 * This only updates the "non-critical" parts of the directory
195 * cache, ie the parts that aren't tracked by GIT, and only used
196 * to validate the cache.
197 */
198 void fill_stat_cache_info(struct index_state *istate, struct cache_entry *ce, struct stat *st)
199 {
200 fill_stat_data(&ce->ce_stat_data, st);
201
202 if (assume_unchanged)
203 ce->ce_flags |= CE_VALID;
204
205 if (S_ISREG(st->st_mode)) {
206 ce_mark_uptodate(ce);
207 mark_fsmonitor_valid(istate, ce);
208 }
209 }
210
211 static int ce_compare_data(struct index_state *istate,
212 const struct cache_entry *ce,
213 struct stat *st)
214 {
215 int match = -1;
216 int fd = git_open_cloexec(ce->name, O_RDONLY);
217
218 if (fd >= 0) {
219 struct object_id oid;
220 if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
221 match = !oideq(&oid, &ce->oid);
222 /* index_fd() closed the file descriptor already */
223 }
224 return match;
225 }
226
227 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
228 {
229 int match = -1;
230 void *buffer;
231 unsigned long size;
232 enum object_type type;
233 struct strbuf sb = STRBUF_INIT;
234
235 if (strbuf_readlink(&sb, ce->name, expected_size))
236 return -1;
237
238 buffer = read_object_file(&ce->oid, &type, &size);
239 if (buffer) {
240 if (size == sb.len)
241 match = memcmp(buffer, sb.buf, size);
242 free(buffer);
243 }
244 strbuf_release(&sb);
245 return match;
246 }
247
248 static int ce_compare_gitlink(const struct cache_entry *ce)
249 {
250 struct object_id oid;
251
252 /*
253 * We don't actually require that the .git directory
254 * under GITLINK directory be a valid git directory. It
255 * might even be missing (in case nobody populated that
256 * sub-project).
257 *
258 * If so, we consider it always to match.
259 */
260 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
261 return 0;
262 return !oideq(&oid, &ce->oid);
263 }
264
265 static int ce_modified_check_fs(struct index_state *istate,
266 const struct cache_entry *ce,
267 struct stat *st)
268 {
269 switch (st->st_mode & S_IFMT) {
270 case S_IFREG:
271 if (ce_compare_data(istate, ce, st))
272 return DATA_CHANGED;
273 break;
274 case S_IFLNK:
275 if (ce_compare_link(ce, xsize_t(st->st_size)))
276 return DATA_CHANGED;
277 break;
278 case S_IFDIR:
279 if (S_ISGITLINK(ce->ce_mode))
280 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
281 /* else fallthrough */
282 default:
283 return TYPE_CHANGED;
284 }
285 return 0;
286 }
287
288 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
289 {
290 unsigned int changed = 0;
291
292 if (ce->ce_flags & CE_REMOVE)
293 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
294
295 switch (ce->ce_mode & S_IFMT) {
296 case S_IFREG:
297 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
298 /* We consider only the owner x bit to be relevant for
299 * "mode changes"
300 */
301 if (trust_executable_bit &&
302 (0100 & (ce->ce_mode ^ st->st_mode)))
303 changed |= MODE_CHANGED;
304 break;
305 case S_IFLNK:
306 if (!S_ISLNK(st->st_mode) &&
307 (has_symlinks || !S_ISREG(st->st_mode)))
308 changed |= TYPE_CHANGED;
309 break;
310 case S_IFGITLINK:
311 /* We ignore most of the st_xxx fields for gitlinks */
312 if (!S_ISDIR(st->st_mode))
313 changed |= TYPE_CHANGED;
314 else if (ce_compare_gitlink(ce))
315 changed |= DATA_CHANGED;
316 return changed;
317 default:
318 BUG("unsupported ce_mode: %o", ce->ce_mode);
319 }
320
321 changed |= match_stat_data(&ce->ce_stat_data, st);
322
323 /* Racily smudged entry? */
324 if (!ce->ce_stat_data.sd_size) {
325 if (!is_empty_blob_sha1(ce->oid.hash))
326 changed |= DATA_CHANGED;
327 }
328
329 return changed;
330 }
331
332 static int is_racy_stat(const struct index_state *istate,
333 const struct stat_data *sd)
334 {
335 return (istate->timestamp.sec &&
336 #ifdef USE_NSEC
337 /* nanosecond timestamped files can also be racy! */
338 (istate->timestamp.sec < sd->sd_mtime.sec ||
339 (istate->timestamp.sec == sd->sd_mtime.sec &&
340 istate->timestamp.nsec <= sd->sd_mtime.nsec))
341 #else
342 istate->timestamp.sec <= sd->sd_mtime.sec
343 #endif
344 );
345 }
346
347 int is_racy_timestamp(const struct index_state *istate,
348 const struct cache_entry *ce)
349 {
350 return (!S_ISGITLINK(ce->ce_mode) &&
351 is_racy_stat(istate, &ce->ce_stat_data));
352 }
353
354 int match_stat_data_racy(const struct index_state *istate,
355 const struct stat_data *sd, struct stat *st)
356 {
357 if (is_racy_stat(istate, sd))
358 return MTIME_CHANGED;
359 return match_stat_data(sd, st);
360 }
361
362 int ie_match_stat(struct index_state *istate,
363 const struct cache_entry *ce, struct stat *st,
364 unsigned int options)
365 {
366 unsigned int changed;
367 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
368 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
369 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
370 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
371
372 if (!ignore_fsmonitor)
373 refresh_fsmonitor(istate);
374 /*
375 * If it's marked as always valid in the index, it's
376 * valid whatever the checked-out copy says.
377 *
378 * skip-worktree has the same effect with higher precedence
379 */
380 if (!ignore_skip_worktree && ce_skip_worktree(ce))
381 return 0;
382 if (!ignore_valid && (ce->ce_flags & CE_VALID))
383 return 0;
384 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
385 return 0;
386
387 /*
388 * Intent-to-add entries have not been added, so the index entry
389 * by definition never matches what is in the work tree until it
390 * actually gets added.
391 */
392 if (ce_intent_to_add(ce))
393 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
394
395 changed = ce_match_stat_basic(ce, st);
396
397 /*
398 * Within 1 second of this sequence:
399 * echo xyzzy >file && git-update-index --add file
400 * running this command:
401 * echo frotz >file
402 * would give a falsely clean cache entry. The mtime and
403 * length match the cache, and other stat fields do not change.
404 *
405 * We could detect this at update-index time (the cache entry
406 * being registered/updated records the same time as "now")
407 * and delay the return from git-update-index, but that would
408 * effectively mean we can make at most one commit per second,
409 * which is not acceptable. Instead, we check cache entries
410 * whose mtime are the same as the index file timestamp more
411 * carefully than others.
412 */
413 if (!changed && is_racy_timestamp(istate, ce)) {
414 if (assume_racy_is_modified)
415 changed |= DATA_CHANGED;
416 else
417 changed |= ce_modified_check_fs(istate, ce, st);
418 }
419
420 return changed;
421 }
422
423 int ie_modified(struct index_state *istate,
424 const struct cache_entry *ce,
425 struct stat *st, unsigned int options)
426 {
427 int changed, changed_fs;
428
429 changed = ie_match_stat(istate, ce, st, options);
430 if (!changed)
431 return 0;
432 /*
433 * If the mode or type has changed, there's no point in trying
434 * to refresh the entry - it's not going to match
435 */
436 if (changed & (MODE_CHANGED | TYPE_CHANGED))
437 return changed;
438
439 /*
440 * Immediately after read-tree or update-index --cacheinfo,
441 * the length field is zero, as we have never even read the
442 * lstat(2) information once, and we cannot trust DATA_CHANGED
443 * returned by ie_match_stat() which in turn was returned by
444 * ce_match_stat_basic() to signal that the filesize of the
445 * blob changed. We have to actually go to the filesystem to
446 * see if the contents match, and if so, should answer "unchanged".
447 *
448 * The logic does not apply to gitlinks, as ce_match_stat_basic()
449 * already has checked the actual HEAD from the filesystem in the
450 * subproject. If ie_match_stat() already said it is different,
451 * then we know it is.
452 */
453 if ((changed & DATA_CHANGED) &&
454 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
455 return changed;
456
457 changed_fs = ce_modified_check_fs(istate, ce, st);
458 if (changed_fs)
459 return changed | changed_fs;
460 return 0;
461 }
462
463 int base_name_compare(const char *name1, int len1, int mode1,
464 const char *name2, int len2, int mode2)
465 {
466 unsigned char c1, c2;
467 int len = len1 < len2 ? len1 : len2;
468 int cmp;
469
470 cmp = memcmp(name1, name2, len);
471 if (cmp)
472 return cmp;
473 c1 = name1[len];
474 c2 = name2[len];
475 if (!c1 && S_ISDIR(mode1))
476 c1 = '/';
477 if (!c2 && S_ISDIR(mode2))
478 c2 = '/';
479 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
480 }
481
482 /*
483 * df_name_compare() is identical to base_name_compare(), except it
484 * compares conflicting directory/file entries as equal. Note that
485 * while a directory name compares as equal to a regular file, they
486 * then individually compare _differently_ to a filename that has
487 * a dot after the basename (because '\0' < '.' < '/').
488 *
489 * This is used by routines that want to traverse the git namespace
490 * but then handle conflicting entries together when possible.
491 */
492 int df_name_compare(const char *name1, int len1, int mode1,
493 const char *name2, int len2, int mode2)
494 {
495 int len = len1 < len2 ? len1 : len2, cmp;
496 unsigned char c1, c2;
497
498 cmp = memcmp(name1, name2, len);
499 if (cmp)
500 return cmp;
501 /* Directories and files compare equal (same length, same name) */
502 if (len1 == len2)
503 return 0;
504 c1 = name1[len];
505 if (!c1 && S_ISDIR(mode1))
506 c1 = '/';
507 c2 = name2[len];
508 if (!c2 && S_ISDIR(mode2))
509 c2 = '/';
510 if (c1 == '/' && !c2)
511 return 0;
512 if (c2 == '/' && !c1)
513 return 0;
514 return c1 - c2;
515 }
516
517 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
518 {
519 size_t min_len = (len1 < len2) ? len1 : len2;
520 int cmp = memcmp(name1, name2, min_len);
521 if (cmp)
522 return cmp;
523 if (len1 < len2)
524 return -1;
525 if (len1 > len2)
526 return 1;
527 return 0;
528 }
529
530 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
531 {
532 int cmp;
533
534 cmp = name_compare(name1, len1, name2, len2);
535 if (cmp)
536 return cmp;
537
538 if (stage1 < stage2)
539 return -1;
540 if (stage1 > stage2)
541 return 1;
542 return 0;
543 }
544
545 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
546 {
547 int first, last;
548
549 first = 0;
550 last = istate->cache_nr;
551 while (last > first) {
552 int next = first + ((last - first) >> 1);
553 struct cache_entry *ce = istate->cache[next];
554 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
555 if (!cmp)
556 return next;
557 if (cmp < 0) {
558 last = next;
559 continue;
560 }
561 first = next+1;
562 }
563 return -first-1;
564 }
565
566 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
567 {
568 return index_name_stage_pos(istate, name, namelen, 0);
569 }
570
571 int remove_index_entry_at(struct index_state *istate, int pos)
572 {
573 struct cache_entry *ce = istate->cache[pos];
574
575 record_resolve_undo(istate, ce);
576 remove_name_hash(istate, ce);
577 save_or_free_index_entry(istate, ce);
578 istate->cache_changed |= CE_ENTRY_REMOVED;
579 istate->cache_nr--;
580 if (pos >= istate->cache_nr)
581 return 0;
582 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
583 istate->cache_nr - pos);
584 return 1;
585 }
586
587 /*
588 * Remove all cache entries marked for removal, that is where
589 * CE_REMOVE is set in ce_flags. This is much more effective than
590 * calling remove_index_entry_at() for each entry to be removed.
591 */
592 void remove_marked_cache_entries(struct index_state *istate, int invalidate)
593 {
594 struct cache_entry **ce_array = istate->cache;
595 unsigned int i, j;
596
597 for (i = j = 0; i < istate->cache_nr; i++) {
598 if (ce_array[i]->ce_flags & CE_REMOVE) {
599 if (invalidate) {
600 cache_tree_invalidate_path(istate,
601 ce_array[i]->name);
602 untracked_cache_remove_from_index(istate,
603 ce_array[i]->name);
604 }
605 remove_name_hash(istate, ce_array[i]);
606 save_or_free_index_entry(istate, ce_array[i]);
607 }
608 else
609 ce_array[j++] = ce_array[i];
610 }
611 if (j == istate->cache_nr)
612 return;
613 istate->cache_changed |= CE_ENTRY_REMOVED;
614 istate->cache_nr = j;
615 }
616
617 int remove_file_from_index(struct index_state *istate, const char *path)
618 {
619 int pos = index_name_pos(istate, path, strlen(path));
620 if (pos < 0)
621 pos = -pos-1;
622 cache_tree_invalidate_path(istate, path);
623 untracked_cache_remove_from_index(istate, path);
624 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
625 remove_index_entry_at(istate, pos);
626 return 0;
627 }
628
629 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
630 {
631 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
632 }
633
634 static int index_name_pos_also_unmerged(struct index_state *istate,
635 const char *path, int namelen)
636 {
637 int pos = index_name_pos(istate, path, namelen);
638 struct cache_entry *ce;
639
640 if (pos >= 0)
641 return pos;
642
643 /* maybe unmerged? */
644 pos = -1 - pos;
645 if (pos >= istate->cache_nr ||
646 compare_name((ce = istate->cache[pos]), path, namelen))
647 return -1;
648
649 /* order of preference: stage 2, 1, 3 */
650 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
651 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
652 !compare_name(ce, path, namelen))
653 pos++;
654 return pos;
655 }
656
657 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
658 {
659 int len = ce_namelen(ce);
660 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
661 }
662
663 /*
664 * If we add a filename that aliases in the cache, we will use the
665 * name that we already have - but we don't want to update the same
666 * alias twice, because that implies that there were actually two
667 * different files with aliasing names!
668 *
669 * So we use the CE_ADDED flag to verify that the alias was an old
670 * one before we accept it as
671 */
672 static struct cache_entry *create_alias_ce(struct index_state *istate,
673 struct cache_entry *ce,
674 struct cache_entry *alias)
675 {
676 int len;
677 struct cache_entry *new_entry;
678
679 if (alias->ce_flags & CE_ADDED)
680 die(_("will not add file alias '%s' ('%s' already exists in index)"),
681 ce->name, alias->name);
682
683 /* Ok, create the new entry using the name of the existing alias */
684 len = ce_namelen(alias);
685 new_entry = make_empty_cache_entry(istate, len);
686 memcpy(new_entry->name, alias->name, len);
687 copy_cache_entry(new_entry, ce);
688 save_or_free_index_entry(istate, ce);
689 return new_entry;
690 }
691
692 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
693 {
694 struct object_id oid;
695 if (write_object_file("", 0, blob_type, &oid))
696 die(_("cannot create an empty blob in the object database"));
697 oidcpy(&ce->oid, &oid);
698 }
699
700 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
701 {
702 int namelen, was_same;
703 mode_t st_mode = st->st_mode;
704 struct cache_entry *ce, *alias = NULL;
705 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
706 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
707 int pretend = flags & ADD_CACHE_PRETEND;
708 int intent_only = flags & ADD_CACHE_INTENT;
709 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
710 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
711 int hash_flags = HASH_WRITE_OBJECT;
712 struct object_id oid;
713
714 if (flags & ADD_CACHE_RENORMALIZE)
715 hash_flags |= HASH_RENORMALIZE;
716
717 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
718 return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
719
720 namelen = strlen(path);
721 if (S_ISDIR(st_mode)) {
722 if (resolve_gitlink_ref(path, "HEAD", &oid) < 0)
723 return error(_("'%s' does not have a commit checked out"), path);
724 while (namelen && path[namelen-1] == '/')
725 namelen--;
726 }
727 ce = make_empty_cache_entry(istate, namelen);
728 memcpy(ce->name, path, namelen);
729 ce->ce_namelen = namelen;
730 if (!intent_only)
731 fill_stat_cache_info(istate, ce, st);
732 else
733 ce->ce_flags |= CE_INTENT_TO_ADD;
734
735
736 if (trust_executable_bit && has_symlinks) {
737 ce->ce_mode = create_ce_mode(st_mode);
738 } else {
739 /* If there is an existing entry, pick the mode bits and type
740 * from it, otherwise assume unexecutable regular file.
741 */
742 struct cache_entry *ent;
743 int pos = index_name_pos_also_unmerged(istate, path, namelen);
744
745 ent = (0 <= pos) ? istate->cache[pos] : NULL;
746 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
747 }
748
749 /* When core.ignorecase=true, determine if a directory of the same name but differing
750 * case already exists within the Git repository. If it does, ensure the directory
751 * case of the file being added to the repository matches (is folded into) the existing
752 * entry's directory case.
753 */
754 if (ignore_case) {
755 adjust_dirname_case(istate, ce->name);
756 }
757 if (!(flags & ADD_CACHE_RENORMALIZE)) {
758 alias = index_file_exists(istate, ce->name,
759 ce_namelen(ce), ignore_case);
760 if (alias &&
761 !ce_stage(alias) &&
762 !ie_match_stat(istate, alias, st, ce_option)) {
763 /* Nothing changed, really */
764 if (!S_ISGITLINK(alias->ce_mode))
765 ce_mark_uptodate(alias);
766 alias->ce_flags |= CE_ADDED;
767
768 discard_cache_entry(ce);
769 return 0;
770 }
771 }
772 if (!intent_only) {
773 if (index_path(istate, &ce->oid, path, st, hash_flags)) {
774 discard_cache_entry(ce);
775 return error(_("unable to index file '%s'"), path);
776 }
777 } else
778 set_object_name_for_intent_to_add_entry(ce);
779
780 if (ignore_case && alias && different_name(ce, alias))
781 ce = create_alias_ce(istate, ce, alias);
782 ce->ce_flags |= CE_ADDED;
783
784 /* It was suspected to be racily clean, but it turns out to be Ok */
785 was_same = (alias &&
786 !ce_stage(alias) &&
787 oideq(&alias->oid, &ce->oid) &&
788 ce->ce_mode == alias->ce_mode);
789
790 if (pretend)
791 discard_cache_entry(ce);
792 else if (add_index_entry(istate, ce, add_option)) {
793 discard_cache_entry(ce);
794 return error(_("unable to add '%s' to index"), path);
795 }
796 if (verbose && !was_same)
797 printf("add '%s'\n", path);
798 return 0;
799 }
800
801 int add_file_to_index(struct index_state *istate, const char *path, int flags)
802 {
803 struct stat st;
804 if (lstat(path, &st))
805 die_errno(_("unable to stat '%s'"), path);
806 return add_to_index(istate, path, &st, flags);
807 }
808
809 struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
810 {
811 return mem_pool__ce_calloc(find_mem_pool(istate), len);
812 }
813
814 struct cache_entry *make_empty_transient_cache_entry(size_t len)
815 {
816 return xcalloc(1, cache_entry_size(len));
817 }
818
819 struct cache_entry *make_cache_entry(struct index_state *istate,
820 unsigned int mode,
821 const struct object_id *oid,
822 const char *path,
823 int stage,
824 unsigned int refresh_options)
825 {
826 struct cache_entry *ce, *ret;
827 int len;
828
829 if (!verify_path(path, mode)) {
830 error(_("invalid path '%s'"), path);
831 return NULL;
832 }
833
834 len = strlen(path);
835 ce = make_empty_cache_entry(istate, len);
836
837 oidcpy(&ce->oid, oid);
838 memcpy(ce->name, path, len);
839 ce->ce_flags = create_ce_flags(stage);
840 ce->ce_namelen = len;
841 ce->ce_mode = create_ce_mode(mode);
842
843 ret = refresh_cache_entry(istate, ce, refresh_options);
844 if (ret != ce)
845 discard_cache_entry(ce);
846 return ret;
847 }
848
849 struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
850 const char *path, int stage)
851 {
852 struct cache_entry *ce;
853 int len;
854
855 if (!verify_path(path, mode)) {
856 error(_("invalid path '%s'"), path);
857 return NULL;
858 }
859
860 len = strlen(path);
861 ce = make_empty_transient_cache_entry(len);
862
863 oidcpy(&ce->oid, oid);
864 memcpy(ce->name, path, len);
865 ce->ce_flags = create_ce_flags(stage);
866 ce->ce_namelen = len;
867 ce->ce_mode = create_ce_mode(mode);
868
869 return ce;
870 }
871
872 /*
873 * Chmod an index entry with either +x or -x.
874 *
875 * Returns -1 if the chmod for the particular cache entry failed (if it's
876 * not a regular file), -2 if an invalid flip argument is passed in, 0
877 * otherwise.
878 */
879 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
880 char flip)
881 {
882 if (!S_ISREG(ce->ce_mode))
883 return -1;
884 switch (flip) {
885 case '+':
886 ce->ce_mode |= 0111;
887 break;
888 case '-':
889 ce->ce_mode &= ~0111;
890 break;
891 default:
892 return -2;
893 }
894 cache_tree_invalidate_path(istate, ce->name);
895 ce->ce_flags |= CE_UPDATE_IN_BASE;
896 mark_fsmonitor_invalid(istate, ce);
897 istate->cache_changed |= CE_ENTRY_CHANGED;
898
899 return 0;
900 }
901
902 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
903 {
904 int len = ce_namelen(a);
905 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
906 }
907
908 /*
909 * We fundamentally don't like some paths: we don't want
910 * dot or dot-dot anywhere, and for obvious reasons don't
911 * want to recurse into ".git" either.
912 *
913 * Also, we don't want double slashes or slashes at the
914 * end that can make pathnames ambiguous.
915 */
916 static int verify_dotfile(const char *rest, unsigned mode)
917 {
918 /*
919 * The first character was '.', but that
920 * has already been discarded, we now test
921 * the rest.
922 */
923
924 /* "." is not allowed */
925 if (*rest == '\0' || is_dir_sep(*rest))
926 return 0;
927
928 switch (*rest) {
929 /*
930 * ".git" followed by NUL or slash is bad. Note that we match
931 * case-insensitively here, even if ignore_case is not set.
932 * This outlaws ".GIT" everywhere out of an abundance of caution,
933 * since there's really no good reason to allow it.
934 *
935 * Once we've seen ".git", we can also find ".gitmodules", etc (also
936 * case-insensitively).
937 */
938 case 'g':
939 case 'G':
940 if (rest[1] != 'i' && rest[1] != 'I')
941 break;
942 if (rest[2] != 't' && rest[2] != 'T')
943 break;
944 if (rest[3] == '\0' || is_dir_sep(rest[3]))
945 return 0;
946 if (S_ISLNK(mode)) {
947 rest += 3;
948 if (skip_iprefix(rest, "modules", &rest) &&
949 (*rest == '\0' || is_dir_sep(*rest)))
950 return 0;
951 }
952 break;
953 case '.':
954 if (rest[1] == '\0' || is_dir_sep(rest[1]))
955 return 0;
956 }
957 return 1;
958 }
959
960 int verify_path(const char *path, unsigned mode)
961 {
962 char c;
963
964 if (has_dos_drive_prefix(path))
965 return 0;
966
967 goto inside;
968 for (;;) {
969 if (!c)
970 return 1;
971 if (is_dir_sep(c)) {
972 inside:
973 if (protect_hfs) {
974 if (is_hfs_dotgit(path))
975 return 0;
976 if (S_ISLNK(mode)) {
977 if (is_hfs_dotgitmodules(path))
978 return 0;
979 }
980 }
981 if (protect_ntfs) {
982 if (is_ntfs_dotgit(path))
983 return 0;
984 if (S_ISLNK(mode)) {
985 if (is_ntfs_dotgitmodules(path))
986 return 0;
987 }
988 }
989
990 c = *path++;
991 if ((c == '.' && !verify_dotfile(path, mode)) ||
992 is_dir_sep(c) || c == '\0')
993 return 0;
994 }
995 c = *path++;
996 }
997 }
998
999 /*
1000 * Do we have another file that has the beginning components being a
1001 * proper superset of the name we're trying to add?
1002 */
1003 static int has_file_name(struct index_state *istate,
1004 const struct cache_entry *ce, int pos, int ok_to_replace)
1005 {
1006 int retval = 0;
1007 int len = ce_namelen(ce);
1008 int stage = ce_stage(ce);
1009 const char *name = ce->name;
1010
1011 while (pos < istate->cache_nr) {
1012 struct cache_entry *p = istate->cache[pos++];
1013
1014 if (len >= ce_namelen(p))
1015 break;
1016 if (memcmp(name, p->name, len))
1017 break;
1018 if (ce_stage(p) != stage)
1019 continue;
1020 if (p->name[len] != '/')
1021 continue;
1022 if (p->ce_flags & CE_REMOVE)
1023 continue;
1024 retval = -1;
1025 if (!ok_to_replace)
1026 break;
1027 remove_index_entry_at(istate, --pos);
1028 }
1029 return retval;
1030 }
1031
1032
1033 /*
1034 * Like strcmp(), but also return the offset of the first change.
1035 * If strings are equal, return the length.
1036 */
1037 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1038 {
1039 size_t k;
1040
1041 if (!first_change)
1042 return strcmp(s1, s2);
1043
1044 for (k = 0; s1[k] == s2[k]; k++)
1045 if (s1[k] == '\0')
1046 break;
1047
1048 *first_change = k;
1049 return (unsigned char)s1[k] - (unsigned char)s2[k];
1050 }
1051
1052 /*
1053 * Do we have another file with a pathname that is a proper
1054 * subset of the name we're trying to add?
1055 *
1056 * That is, is there another file in the index with a path
1057 * that matches a sub-directory in the given entry?
1058 */
1059 static int has_dir_name(struct index_state *istate,
1060 const struct cache_entry *ce, int pos, int ok_to_replace)
1061 {
1062 int retval = 0;
1063 int stage = ce_stage(ce);
1064 const char *name = ce->name;
1065 const char *slash = name + ce_namelen(ce);
1066 size_t len_eq_last;
1067 int cmp_last = 0;
1068
1069 /*
1070 * We are frequently called during an iteration on a sorted
1071 * list of pathnames and while building a new index. Therefore,
1072 * there is a high probability that this entry will eventually
1073 * be appended to the index, rather than inserted in the middle.
1074 * If we can confirm that, we can avoid binary searches on the
1075 * components of the pathname.
1076 *
1077 * Compare the entry's full path with the last path in the index.
1078 */
1079 if (istate->cache_nr > 0) {
1080 cmp_last = strcmp_offset(name,
1081 istate->cache[istate->cache_nr - 1]->name,
1082 &len_eq_last);
1083 if (cmp_last > 0) {
1084 if (len_eq_last == 0) {
1085 /*
1086 * The entry sorts AFTER the last one in the
1087 * index and their paths have no common prefix,
1088 * so there cannot be a F/D conflict.
1089 */
1090 return retval;
1091 } else {
1092 /*
1093 * The entry sorts AFTER the last one in the
1094 * index, but has a common prefix. Fall through
1095 * to the loop below to disect the entry's path
1096 * and see where the difference is.
1097 */
1098 }
1099 } else if (cmp_last == 0) {
1100 /*
1101 * The entry exactly matches the last one in the
1102 * index, but because of multiple stage and CE_REMOVE
1103 * items, we fall through and let the regular search
1104 * code handle it.
1105 */
1106 }
1107 }
1108
1109 for (;;) {
1110 size_t len;
1111
1112 for (;;) {
1113 if (*--slash == '/')
1114 break;
1115 if (slash <= ce->name)
1116 return retval;
1117 }
1118 len = slash - name;
1119
1120 if (cmp_last > 0) {
1121 /*
1122 * (len + 1) is a directory boundary (including
1123 * the trailing slash). And since the loop is
1124 * decrementing "slash", the first iteration is
1125 * the longest directory prefix; subsequent
1126 * iterations consider parent directories.
1127 */
1128
1129 if (len + 1 <= len_eq_last) {
1130 /*
1131 * The directory prefix (including the trailing
1132 * slash) also appears as a prefix in the last
1133 * entry, so the remainder cannot collide (because
1134 * strcmp said the whole path was greater).
1135 *
1136 * EQ: last: xxx/A
1137 * this: xxx/B
1138 *
1139 * LT: last: xxx/file_A
1140 * this: xxx/file_B
1141 */
1142 return retval;
1143 }
1144
1145 if (len > len_eq_last) {
1146 /*
1147 * This part of the directory prefix (excluding
1148 * the trailing slash) is longer than the known
1149 * equal portions, so this sub-directory cannot
1150 * collide with a file.
1151 *
1152 * GT: last: xxxA
1153 * this: xxxB/file
1154 */
1155 return retval;
1156 }
1157
1158 if (istate->cache_nr > 0 &&
1159 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1160 /*
1161 * The directory prefix lines up with part of
1162 * a longer file or directory name, but sorts
1163 * after it, so this sub-directory cannot
1164 * collide with a file.
1165 *
1166 * last: xxx/yy-file (because '-' sorts before '/')
1167 * this: xxx/yy/abc
1168 */
1169 return retval;
1170 }
1171
1172 /*
1173 * This is a possible collision. Fall through and
1174 * let the regular search code handle it.
1175 *
1176 * last: xxx
1177 * this: xxx/file
1178 */
1179 }
1180
1181 pos = index_name_stage_pos(istate, name, len, stage);
1182 if (pos >= 0) {
1183 /*
1184 * Found one, but not so fast. This could
1185 * be a marker that says "I was here, but
1186 * I am being removed". Such an entry is
1187 * not a part of the resulting tree, and
1188 * it is Ok to have a directory at the same
1189 * path.
1190 */
1191 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1192 retval = -1;
1193 if (!ok_to_replace)
1194 break;
1195 remove_index_entry_at(istate, pos);
1196 continue;
1197 }
1198 }
1199 else
1200 pos = -pos-1;
1201
1202 /*
1203 * Trivial optimization: if we find an entry that
1204 * already matches the sub-directory, then we know
1205 * we're ok, and we can exit.
1206 */
1207 while (pos < istate->cache_nr) {
1208 struct cache_entry *p = istate->cache[pos];
1209 if ((ce_namelen(p) <= len) ||
1210 (p->name[len] != '/') ||
1211 memcmp(p->name, name, len))
1212 break; /* not our subdirectory */
1213 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1214 /*
1215 * p is at the same stage as our entry, and
1216 * is a subdirectory of what we are looking
1217 * at, so we cannot have conflicts at our
1218 * level or anything shorter.
1219 */
1220 return retval;
1221 pos++;
1222 }
1223 }
1224 return retval;
1225 }
1226
1227 /* We may be in a situation where we already have path/file and path
1228 * is being added, or we already have path and path/file is being
1229 * added. Either one would result in a nonsense tree that has path
1230 * twice when git-write-tree tries to write it out. Prevent it.
1231 *
1232 * If ok-to-replace is specified, we remove the conflicting entries
1233 * from the cache so the caller should recompute the insert position.
1234 * When this happens, we return non-zero.
1235 */
1236 static int check_file_directory_conflict(struct index_state *istate,
1237 const struct cache_entry *ce,
1238 int pos, int ok_to_replace)
1239 {
1240 int retval;
1241
1242 /*
1243 * When ce is an "I am going away" entry, we allow it to be added
1244 */
1245 if (ce->ce_flags & CE_REMOVE)
1246 return 0;
1247
1248 /*
1249 * We check if the path is a sub-path of a subsequent pathname
1250 * first, since removing those will not change the position
1251 * in the array.
1252 */
1253 retval = has_file_name(istate, ce, pos, ok_to_replace);
1254
1255 /*
1256 * Then check if the path might have a clashing sub-directory
1257 * before it.
1258 */
1259 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1260 }
1261
1262 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1263 {
1264 int pos;
1265 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1266 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1267 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1268 int new_only = option & ADD_CACHE_NEW_ONLY;
1269
1270 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1271 cache_tree_invalidate_path(istate, ce->name);
1272
1273 /*
1274 * If this entry's path sorts after the last entry in the index,
1275 * we can avoid searching for it.
1276 */
1277 if (istate->cache_nr > 0 &&
1278 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1279 pos = -istate->cache_nr - 1;
1280 else
1281 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1282
1283 /* existing match? Just replace it. */
1284 if (pos >= 0) {
1285 if (!new_only)
1286 replace_index_entry(istate, pos, ce);
1287 return 0;
1288 }
1289 pos = -pos-1;
1290
1291 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1292 untracked_cache_add_to_index(istate, ce->name);
1293
1294 /*
1295 * Inserting a merged entry ("stage 0") into the index
1296 * will always replace all non-merged entries..
1297 */
1298 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1299 while (ce_same_name(istate->cache[pos], ce)) {
1300 ok_to_add = 1;
1301 if (!remove_index_entry_at(istate, pos))
1302 break;
1303 }
1304 }
1305
1306 if (!ok_to_add)
1307 return -1;
1308 if (!verify_path(ce->name, ce->ce_mode))
1309 return error(_("invalid path '%s'"), ce->name);
1310
1311 if (!skip_df_check &&
1312 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1313 if (!ok_to_replace)
1314 return error(_("'%s' appears as both a file and as a directory"),
1315 ce->name);
1316 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1317 pos = -pos-1;
1318 }
1319 return pos + 1;
1320 }
1321
1322 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1323 {
1324 int pos;
1325
1326 if (option & ADD_CACHE_JUST_APPEND)
1327 pos = istate->cache_nr;
1328 else {
1329 int ret;
1330 ret = add_index_entry_with_check(istate, ce, option);
1331 if (ret <= 0)
1332 return ret;
1333 pos = ret - 1;
1334 }
1335
1336 /* Make sure the array is big enough .. */
1337 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1338
1339 /* Add it in.. */
1340 istate->cache_nr++;
1341 if (istate->cache_nr > pos + 1)
1342 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1343 istate->cache_nr - pos - 1);
1344 set_index_entry(istate, pos, ce);
1345 istate->cache_changed |= CE_ENTRY_ADDED;
1346 return 0;
1347 }
1348
1349 /*
1350 * "refresh" does not calculate a new sha1 file or bring the
1351 * cache up-to-date for mode/content changes. But what it
1352 * _does_ do is to "re-match" the stat information of a file
1353 * with the cache, so that you can refresh the cache for a
1354 * file that hasn't been changed but where the stat entry is
1355 * out of date.
1356 *
1357 * For example, you'd want to do this after doing a "git-read-tree",
1358 * to link up the stat cache details with the proper files.
1359 */
1360 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1361 struct cache_entry *ce,
1362 unsigned int options, int *err,
1363 int *changed_ret)
1364 {
1365 struct stat st;
1366 struct cache_entry *updated;
1367 int changed;
1368 int refresh = options & CE_MATCH_REFRESH;
1369 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1370 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1371 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1372 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1373
1374 if (!refresh || ce_uptodate(ce))
1375 return ce;
1376
1377 if (!ignore_fsmonitor)
1378 refresh_fsmonitor(istate);
1379 /*
1380 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1381 * that the change to the work tree does not matter and told
1382 * us not to worry.
1383 */
1384 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1385 ce_mark_uptodate(ce);
1386 return ce;
1387 }
1388 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1389 ce_mark_uptodate(ce);
1390 return ce;
1391 }
1392 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1393 ce_mark_uptodate(ce);
1394 return ce;
1395 }
1396
1397 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1398 if (ignore_missing)
1399 return ce;
1400 if (err)
1401 *err = ENOENT;
1402 return NULL;
1403 }
1404
1405 if (lstat(ce->name, &st) < 0) {
1406 if (ignore_missing && errno == ENOENT)
1407 return ce;
1408 if (err)
1409 *err = errno;
1410 return NULL;
1411 }
1412
1413 changed = ie_match_stat(istate, ce, &st, options);
1414 if (changed_ret)
1415 *changed_ret = changed;
1416 if (!changed) {
1417 /*
1418 * The path is unchanged. If we were told to ignore
1419 * valid bit, then we did the actual stat check and
1420 * found that the entry is unmodified. If the entry
1421 * is not marked VALID, this is the place to mark it
1422 * valid again, under "assume unchanged" mode.
1423 */
1424 if (ignore_valid && assume_unchanged &&
1425 !(ce->ce_flags & CE_VALID))
1426 ; /* mark this one VALID again */
1427 else {
1428 /*
1429 * We do not mark the index itself "modified"
1430 * because CE_UPTODATE flag is in-core only;
1431 * we are not going to write this change out.
1432 */
1433 if (!S_ISGITLINK(ce->ce_mode)) {
1434 ce_mark_uptodate(ce);
1435 mark_fsmonitor_valid(istate, ce);
1436 }
1437 return ce;
1438 }
1439 }
1440
1441 if (ie_modified(istate, ce, &st, options)) {
1442 if (err)
1443 *err = EINVAL;
1444 return NULL;
1445 }
1446
1447 updated = make_empty_cache_entry(istate, ce_namelen(ce));
1448 copy_cache_entry(updated, ce);
1449 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1450 fill_stat_cache_info(istate, updated, &st);
1451 /*
1452 * If ignore_valid is not set, we should leave CE_VALID bit
1453 * alone. Otherwise, paths marked with --no-assume-unchanged
1454 * (i.e. things to be edited) will reacquire CE_VALID bit
1455 * automatically, which is not really what we want.
1456 */
1457 if (!ignore_valid && assume_unchanged &&
1458 !(ce->ce_flags & CE_VALID))
1459 updated->ce_flags &= ~CE_VALID;
1460
1461 /* istate->cache_changed is updated in the caller */
1462 return updated;
1463 }
1464
1465 static void show_file(const char * fmt, const char * name, int in_porcelain,
1466 int * first, const char *header_msg)
1467 {
1468 if (in_porcelain && *first && header_msg) {
1469 printf("%s\n", header_msg);
1470 *first = 0;
1471 }
1472 printf(fmt, name);
1473 }
1474
1475 int refresh_index(struct index_state *istate, unsigned int flags,
1476 const struct pathspec *pathspec,
1477 char *seen, const char *header_msg)
1478 {
1479 int i;
1480 int has_errors = 0;
1481 int really = (flags & REFRESH_REALLY) != 0;
1482 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1483 int quiet = (flags & REFRESH_QUIET) != 0;
1484 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1485 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1486 int first = 1;
1487 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1488 unsigned int options = (CE_MATCH_REFRESH |
1489 (really ? CE_MATCH_IGNORE_VALID : 0) |
1490 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1491 const char *modified_fmt;
1492 const char *deleted_fmt;
1493 const char *typechange_fmt;
1494 const char *added_fmt;
1495 const char *unmerged_fmt;
1496 struct progress *progress = NULL;
1497
1498 if (flags & REFRESH_PROGRESS && isatty(2))
1499 progress = start_delayed_progress(_("Refresh index"),
1500 istate->cache_nr);
1501
1502 trace_performance_enter();
1503 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1504 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1505 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1506 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1507 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1508 /*
1509 * Use the multi-threaded preload_index() to refresh most of the
1510 * cache entries quickly then in the single threaded loop below,
1511 * we only have to do the special cases that are left.
1512 */
1513 preload_index(istate, pathspec, 0);
1514 for (i = 0; i < istate->cache_nr; i++) {
1515 struct cache_entry *ce, *new_entry;
1516 int cache_errno = 0;
1517 int changed = 0;
1518 int filtered = 0;
1519
1520 ce = istate->cache[i];
1521 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1522 continue;
1523
1524 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1525 filtered = 1;
1526
1527 if (ce_stage(ce)) {
1528 while ((i < istate->cache_nr) &&
1529 ! strcmp(istate->cache[i]->name, ce->name))
1530 i++;
1531 i--;
1532 if (allow_unmerged)
1533 continue;
1534 if (!filtered)
1535 show_file(unmerged_fmt, ce->name, in_porcelain,
1536 &first, header_msg);
1537 has_errors = 1;
1538 continue;
1539 }
1540
1541 if (filtered)
1542 continue;
1543
1544 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1545 if (new_entry == ce)
1546 continue;
1547 if (progress)
1548 display_progress(progress, i);
1549 if (!new_entry) {
1550 const char *fmt;
1551
1552 if (really && cache_errno == EINVAL) {
1553 /* If we are doing --really-refresh that
1554 * means the index is not valid anymore.
1555 */
1556 ce->ce_flags &= ~CE_VALID;
1557 ce->ce_flags |= CE_UPDATE_IN_BASE;
1558 mark_fsmonitor_invalid(istate, ce);
1559 istate->cache_changed |= CE_ENTRY_CHANGED;
1560 }
1561 if (quiet)
1562 continue;
1563
1564 if (cache_errno == ENOENT)
1565 fmt = deleted_fmt;
1566 else if (ce_intent_to_add(ce))
1567 fmt = added_fmt; /* must be before other checks */
1568 else if (changed & TYPE_CHANGED)
1569 fmt = typechange_fmt;
1570 else
1571 fmt = modified_fmt;
1572 show_file(fmt,
1573 ce->name, in_porcelain, &first, header_msg);
1574 has_errors = 1;
1575 continue;
1576 }
1577
1578 replace_index_entry(istate, i, new_entry);
1579 }
1580 if (progress) {
1581 display_progress(progress, istate->cache_nr);
1582 stop_progress(&progress);
1583 }
1584 trace_performance_leave("refresh index");
1585 return has_errors;
1586 }
1587
1588 struct cache_entry *refresh_cache_entry(struct index_state *istate,
1589 struct cache_entry *ce,
1590 unsigned int options)
1591 {
1592 return refresh_cache_ent(istate, ce, options, NULL, NULL);
1593 }
1594
1595
1596 /*****************************************************************
1597 * Index File I/O
1598 *****************************************************************/
1599
1600 #define INDEX_FORMAT_DEFAULT 3
1601
1602 static unsigned int get_index_format_default(struct repository *r)
1603 {
1604 char *envversion = getenv("GIT_INDEX_VERSION");
1605 char *endp;
1606 unsigned int version = INDEX_FORMAT_DEFAULT;
1607
1608 if (!envversion) {
1609 prepare_repo_settings(r);
1610
1611 if (r->settings.index_version >= 0)
1612 version = r->settings.index_version;
1613 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1614 warning(_("index.version set, but the value is invalid.\n"
1615 "Using version %i"), INDEX_FORMAT_DEFAULT);
1616 return INDEX_FORMAT_DEFAULT;
1617 }
1618 return version;
1619 }
1620
1621 version = strtoul(envversion, &endp, 10);
1622 if (*endp ||
1623 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1624 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1625 "Using version %i"), INDEX_FORMAT_DEFAULT);
1626 version = INDEX_FORMAT_DEFAULT;
1627 }
1628 return version;
1629 }
1630
1631 /*
1632 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1633 * Again - this is just a (very strong in practice) heuristic that
1634 * the inode hasn't changed.
1635 *
1636 * We save the fields in big-endian order to allow using the
1637 * index file over NFS transparently.
1638 */
1639 struct ondisk_cache_entry {
1640 struct cache_time ctime;
1641 struct cache_time mtime;
1642 uint32_t dev;
1643 uint32_t ino;
1644 uint32_t mode;
1645 uint32_t uid;
1646 uint32_t gid;
1647 uint32_t size;
1648 /*
1649 * unsigned char hash[hashsz];
1650 * uint16_t flags;
1651 * if (flags & CE_EXTENDED)
1652 * uint16_t flags2;
1653 */
1654 unsigned char data[GIT_MAX_RAWSZ + 2 * sizeof(uint16_t)];
1655 char name[FLEX_ARRAY];
1656 };
1657
1658 /* These are only used for v3 or lower */
1659 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1660 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1661 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1662 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1663 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1664 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1665 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1666
1667 /* Allow fsck to force verification of the index checksum. */
1668 int verify_index_checksum;
1669
1670 /* Allow fsck to force verification of the cache entry order. */
1671 int verify_ce_order;
1672
1673 static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1674 {
1675 git_hash_ctx c;
1676 unsigned char hash[GIT_MAX_RAWSZ];
1677 int hdr_version;
1678
1679 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1680 return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1681 hdr_version = ntohl(hdr->hdr_version);
1682 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1683 return error(_("bad index version %d"), hdr_version);
1684
1685 if (!verify_index_checksum)
1686 return 0;
1687
1688 the_hash_algo->init_fn(&c);
1689 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1690 the_hash_algo->final_fn(hash, &c);
1691 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1692 return error(_("bad index file sha1 signature"));
1693 return 0;
1694 }
1695
1696 static int read_index_extension(struct index_state *istate,
1697 const char *ext, const char *data, unsigned long sz)
1698 {
1699 switch (CACHE_EXT(ext)) {
1700 case CACHE_EXT_TREE:
1701 istate->cache_tree = cache_tree_read(data, sz);
1702 break;
1703 case CACHE_EXT_RESOLVE_UNDO:
1704 istate->resolve_undo = resolve_undo_read(data, sz);
1705 break;
1706 case CACHE_EXT_LINK:
1707 if (read_link_extension(istate, data, sz))
1708 return -1;
1709 break;
1710 case CACHE_EXT_UNTRACKED:
1711 istate->untracked = read_untracked_extension(data, sz);
1712 break;
1713 case CACHE_EXT_FSMONITOR:
1714 read_fsmonitor_extension(istate, data, sz);
1715 break;
1716 case CACHE_EXT_ENDOFINDEXENTRIES:
1717 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1718 /* already handled in do_read_index() */
1719 break;
1720 default:
1721 if (*ext < 'A' || 'Z' < *ext)
1722 return error(_("index uses %.4s extension, which we do not understand"),
1723 ext);
1724 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1725 break;
1726 }
1727 return 0;
1728 }
1729
1730 static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1731 unsigned int version,
1732 struct ondisk_cache_entry *ondisk,
1733 unsigned long *ent_size,
1734 const struct cache_entry *previous_ce)
1735 {
1736 struct cache_entry *ce;
1737 size_t len;
1738 const char *name;
1739 const unsigned hashsz = the_hash_algo->rawsz;
1740 const uint16_t *flagsp = (const uint16_t *)(ondisk->data + hashsz);
1741 unsigned int flags;
1742 size_t copy_len = 0;
1743 /*
1744 * Adjacent cache entries tend to share the leading paths, so it makes
1745 * sense to only store the differences in later entries. In the v4
1746 * on-disk format of the index, each on-disk cache entry stores the
1747 * number of bytes to be stripped from the end of the previous name,
1748 * and the bytes to append to the result, to come up with its name.
1749 */
1750 int expand_name_field = version == 4;
1751
1752 /* On-disk flags are just 16 bits */
1753 flags = get_be16(flagsp);
1754 len = flags & CE_NAMEMASK;
1755
1756 if (flags & CE_EXTENDED) {
1757 int extended_flags;
1758 extended_flags = get_be16(flagsp + 1) << 16;
1759 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1760 if (extended_flags & ~CE_EXTENDED_FLAGS)
1761 die(_("unknown index entry format 0x%08x"), extended_flags);
1762 flags |= extended_flags;
1763 name = (const char *)(flagsp + 2);
1764 }
1765 else
1766 name = (const char *)(flagsp + 1);
1767
1768 if (expand_name_field) {
1769 const unsigned char *cp = (const unsigned char *)name;
1770 size_t strip_len, previous_len;
1771
1772 /* If we're at the begining of a block, ignore the previous name */
1773 strip_len = decode_varint(&cp);
1774 if (previous_ce) {
1775 previous_len = previous_ce->ce_namelen;
1776 if (previous_len < strip_len)
1777 die(_("malformed name field in the index, near path '%s'"),
1778 previous_ce->name);
1779 copy_len = previous_len - strip_len;
1780 }
1781 name = (const char *)cp;
1782 }
1783
1784 if (len == CE_NAMEMASK) {
1785 len = strlen(name);
1786 if (expand_name_field)
1787 len += copy_len;
1788 }
1789
1790 ce = mem_pool__ce_alloc(ce_mem_pool, len);
1791
1792 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1793 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1794 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1795 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1796 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1797 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1798 ce->ce_mode = get_be32(&ondisk->mode);
1799 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1800 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1801 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1802 ce->ce_flags = flags & ~CE_NAMEMASK;
1803 ce->ce_namelen = len;
1804 ce->index = 0;
1805 hashcpy(ce->oid.hash, ondisk->data);
1806 memcpy(ce->name, name, len);
1807 ce->name[len] = '\0';
1808
1809 if (expand_name_field) {
1810 if (copy_len)
1811 memcpy(ce->name, previous_ce->name, copy_len);
1812 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1813 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1814 } else {
1815 memcpy(ce->name, name, len + 1);
1816 *ent_size = ondisk_ce_size(ce);
1817 }
1818 return ce;
1819 }
1820
1821 static void check_ce_order(struct index_state *istate)
1822 {
1823 unsigned int i;
1824
1825 if (!verify_ce_order)
1826 return;
1827
1828 for (i = 1; i < istate->cache_nr; i++) {
1829 struct cache_entry *ce = istate->cache[i - 1];
1830 struct cache_entry *next_ce = istate->cache[i];
1831 int name_compare = strcmp(ce->name, next_ce->name);
1832
1833 if (0 < name_compare)
1834 die(_("unordered stage entries in index"));
1835 if (!name_compare) {
1836 if (!ce_stage(ce))
1837 die(_("multiple stage entries for merged file '%s'"),
1838 ce->name);
1839 if (ce_stage(ce) > ce_stage(next_ce))
1840 die(_("unordered stage entries for '%s'"),
1841 ce->name);
1842 }
1843 }
1844 }
1845
1846 static void tweak_untracked_cache(struct index_state *istate)
1847 {
1848 struct repository *r = the_repository;
1849
1850 prepare_repo_settings(r);
1851
1852 if (r->settings.core_untracked_cache == UNTRACKED_CACHE_REMOVE) {
1853 remove_untracked_cache(istate);
1854 return;
1855 }
1856
1857 if (r->settings.core_untracked_cache == UNTRACKED_CACHE_WRITE)
1858 add_untracked_cache(istate);
1859 }
1860
1861 static void tweak_split_index(struct index_state *istate)
1862 {
1863 switch (git_config_get_split_index()) {
1864 case -1: /* unset: do nothing */
1865 break;
1866 case 0: /* false */
1867 remove_split_index(istate);
1868 break;
1869 case 1: /* true */
1870 add_split_index(istate);
1871 break;
1872 default: /* unknown value: do nothing */
1873 break;
1874 }
1875 }
1876
1877 static void post_read_index_from(struct index_state *istate)
1878 {
1879 check_ce_order(istate);
1880 tweak_untracked_cache(istate);
1881 tweak_split_index(istate);
1882 tweak_fsmonitor(istate);
1883 }
1884
1885 static size_t estimate_cache_size_from_compressed(unsigned int entries)
1886 {
1887 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1888 }
1889
1890 static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1891 {
1892 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1893
1894 /*
1895 * Account for potential alignment differences.
1896 */
1897 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1898 return ondisk_size + entries * per_entry;
1899 }
1900
1901 struct index_entry_offset
1902 {
1903 /* starting byte offset into index file, count of index entries in this block */
1904 int offset, nr;
1905 };
1906
1907 struct index_entry_offset_table
1908 {
1909 int nr;
1910 struct index_entry_offset entries[FLEX_ARRAY];
1911 };
1912
1913 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1914 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1915
1916 static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1917 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1918
1919 struct load_index_extensions
1920 {
1921 pthread_t pthread;
1922 struct index_state *istate;
1923 const char *mmap;
1924 size_t mmap_size;
1925 unsigned long src_offset;
1926 };
1927
1928 static void *load_index_extensions(void *_data)
1929 {
1930 struct load_index_extensions *p = _data;
1931 unsigned long src_offset = p->src_offset;
1932
1933 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1934 /* After an array of active_nr index entries,
1935 * there can be arbitrary number of extended
1936 * sections, each of which is prefixed with
1937 * extension name (4-byte) and section length
1938 * in 4-byte network byte order.
1939 */
1940 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1941 if (read_index_extension(p->istate,
1942 p->mmap + src_offset,
1943 p->mmap + src_offset + 8,
1944 extsize) < 0) {
1945 munmap((void *)p->mmap, p->mmap_size);
1946 die(_("index file corrupt"));
1947 }
1948 src_offset += 8;
1949 src_offset += extsize;
1950 }
1951
1952 return NULL;
1953 }
1954
1955 /*
1956 * A helper function that will load the specified range of cache entries
1957 * from the memory mapped file and add them to the given index.
1958 */
1959 static unsigned long load_cache_entry_block(struct index_state *istate,
1960 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1961 unsigned long start_offset, const struct cache_entry *previous_ce)
1962 {
1963 int i;
1964 unsigned long src_offset = start_offset;
1965
1966 for (i = offset; i < offset + nr; i++) {
1967 struct ondisk_cache_entry *disk_ce;
1968 struct cache_entry *ce;
1969 unsigned long consumed;
1970
1971 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1972 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1973 set_index_entry(istate, i, ce);
1974
1975 src_offset += consumed;
1976 previous_ce = ce;
1977 }
1978 return src_offset - start_offset;
1979 }
1980
1981 static unsigned long load_all_cache_entries(struct index_state *istate,
1982 const char *mmap, size_t mmap_size, unsigned long src_offset)
1983 {
1984 unsigned long consumed;
1985
1986 if (istate->version == 4) {
1987 mem_pool_init(&istate->ce_mem_pool,
1988 estimate_cache_size_from_compressed(istate->cache_nr));
1989 } else {
1990 mem_pool_init(&istate->ce_mem_pool,
1991 estimate_cache_size(mmap_size, istate->cache_nr));
1992 }
1993
1994 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
1995 0, istate->cache_nr, mmap, src_offset, NULL);
1996 return consumed;
1997 }
1998
1999 /*
2000 * Mostly randomly chosen maximum thread counts: we
2001 * cap the parallelism to online_cpus() threads, and we want
2002 * to have at least 10000 cache entries per thread for it to
2003 * be worth starting a thread.
2004 */
2005
2006 #define THREAD_COST (10000)
2007
2008 struct load_cache_entries_thread_data
2009 {
2010 pthread_t pthread;
2011 struct index_state *istate;
2012 struct mem_pool *ce_mem_pool;
2013 int offset;
2014 const char *mmap;
2015 struct index_entry_offset_table *ieot;
2016 int ieot_start; /* starting index into the ieot array */
2017 int ieot_blocks; /* count of ieot entries to process */
2018 unsigned long consumed; /* return # of bytes in index file processed */
2019 };
2020
2021 /*
2022 * A thread proc to run the load_cache_entries() computation
2023 * across multiple background threads.
2024 */
2025 static void *load_cache_entries_thread(void *_data)
2026 {
2027 struct load_cache_entries_thread_data *p = _data;
2028 int i;
2029
2030 /* iterate across all ieot blocks assigned to this thread */
2031 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2032 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2033 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2034 p->offset += p->ieot->entries[i].nr;
2035 }
2036 return NULL;
2037 }
2038
2039 static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2040 int nr_threads, struct index_entry_offset_table *ieot)
2041 {
2042 int i, offset, ieot_blocks, ieot_start, err;
2043 struct load_cache_entries_thread_data *data;
2044 unsigned long consumed = 0;
2045
2046 /* a little sanity checking */
2047 if (istate->name_hash_initialized)
2048 BUG("the name hash isn't thread safe");
2049
2050 mem_pool_init(&istate->ce_mem_pool, 0);
2051
2052 /* ensure we have no more threads than we have blocks to process */
2053 if (nr_threads > ieot->nr)
2054 nr_threads = ieot->nr;
2055 data = xcalloc(nr_threads, sizeof(*data));
2056
2057 offset = ieot_start = 0;
2058 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2059 for (i = 0; i < nr_threads; i++) {
2060 struct load_cache_entries_thread_data *p = &data[i];
2061 int nr, j;
2062
2063 if (ieot_start + ieot_blocks > ieot->nr)
2064 ieot_blocks = ieot->nr - ieot_start;
2065
2066 p->istate = istate;
2067 p->offset = offset;
2068 p->mmap = mmap;
2069 p->ieot = ieot;
2070 p->ieot_start = ieot_start;
2071 p->ieot_blocks = ieot_blocks;
2072
2073 /* create a mem_pool for each thread */
2074 nr = 0;
2075 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2076 nr += p->ieot->entries[j].nr;
2077 if (istate->version == 4) {
2078 mem_pool_init(&p->ce_mem_pool,
2079 estimate_cache_size_from_compressed(nr));
2080 } else {
2081 mem_pool_init(&p->ce_mem_pool,
2082 estimate_cache_size(mmap_size, nr));
2083 }
2084
2085 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2086 if (err)
2087 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2088
2089 /* increment by the number of cache entries in the ieot block being processed */
2090 for (j = 0; j < ieot_blocks; j++)
2091 offset += ieot->entries[ieot_start + j].nr;
2092 ieot_start += ieot_blocks;
2093 }
2094
2095 for (i = 0; i < nr_threads; i++) {
2096 struct load_cache_entries_thread_data *p = &data[i];
2097
2098 err = pthread_join(p->pthread, NULL);
2099 if (err)
2100 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2101 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2102 consumed += p->consumed;
2103 }
2104
2105 free(data);
2106
2107 return consumed;
2108 }
2109
2110 /* remember to discard_cache() before reading a different cache! */
2111 int do_read_index(struct index_state *istate, const char *path, int must_exist)
2112 {
2113 int fd;
2114 struct stat st;
2115 unsigned long src_offset;
2116 const struct cache_header *hdr;
2117 const char *mmap;
2118 size_t mmap_size;
2119 struct load_index_extensions p;
2120 size_t extension_offset = 0;
2121 int nr_threads, cpus;
2122 struct index_entry_offset_table *ieot = NULL;
2123
2124 if (istate->initialized)
2125 return istate->cache_nr;
2126
2127 istate->timestamp.sec = 0;
2128 istate->timestamp.nsec = 0;
2129 fd = open(path, O_RDONLY);
2130 if (fd < 0) {
2131 if (!must_exist && errno == ENOENT)
2132 return 0;
2133 die_errno(_("%s: index file open failed"), path);
2134 }
2135
2136 if (fstat(fd, &st))
2137 die_errno(_("%s: cannot stat the open index"), path);
2138
2139 mmap_size = xsize_t(st.st_size);
2140 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2141 die(_("%s: index file smaller than expected"), path);
2142
2143 mmap = xmmap_gently(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2144 if (mmap == MAP_FAILED)
2145 die_errno(_("%s: unable to map index file"), path);
2146 close(fd);
2147
2148 hdr = (const struct cache_header *)mmap;
2149 if (verify_hdr(hdr, mmap_size) < 0)
2150 goto unmap;
2151
2152 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2153 istate->version = ntohl(hdr->hdr_version);
2154 istate->cache_nr = ntohl(hdr->hdr_entries);
2155 istate->cache_alloc = alloc_nr(istate->cache_nr);
2156 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2157 istate->initialized = 1;
2158
2159 p.istate = istate;
2160 p.mmap = mmap;
2161 p.mmap_size = mmap_size;
2162
2163 src_offset = sizeof(*hdr);
2164
2165 if (git_config_get_index_threads(&nr_threads))
2166 nr_threads = 1;
2167
2168 /* TODO: does creating more threads than cores help? */
2169 if (!nr_threads) {
2170 nr_threads = istate->cache_nr / THREAD_COST;
2171 cpus = online_cpus();
2172 if (nr_threads > cpus)
2173 nr_threads = cpus;
2174 }
2175
2176 if (!HAVE_THREADS)
2177 nr_threads = 1;
2178
2179 if (nr_threads > 1) {
2180 extension_offset = read_eoie_extension(mmap, mmap_size);
2181 if (extension_offset) {
2182 int err;
2183
2184 p.src_offset = extension_offset;
2185 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2186 if (err)
2187 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2188
2189 nr_threads--;
2190 }
2191 }
2192
2193 /*
2194 * Locate and read the index entry offset table so that we can use it
2195 * to multi-thread the reading of the cache entries.
2196 */
2197 if (extension_offset && nr_threads > 1)
2198 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2199
2200 if (ieot) {
2201 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, nr_threads, ieot);
2202 free(ieot);
2203 } else {
2204 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2205 }
2206
2207 istate->timestamp.sec = st.st_mtime;
2208 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2209
2210 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2211 if (extension_offset) {
2212 int ret = pthread_join(p.pthread, NULL);
2213 if (ret)
2214 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2215 } else {
2216 p.src_offset = src_offset;
2217 load_index_extensions(&p);
2218 }
2219 munmap((void *)mmap, mmap_size);
2220
2221 /*
2222 * TODO trace2: replace "the_repository" with the actual repo instance
2223 * that is associated with the given "istate".
2224 */
2225 trace2_data_intmax("index", the_repository, "read/version",
2226 istate->version);
2227 trace2_data_intmax("index", the_repository, "read/cache_nr",
2228 istate->cache_nr);
2229
2230 return istate->cache_nr;
2231
2232 unmap:
2233 munmap((void *)mmap, mmap_size);
2234 die(_("index file corrupt"));
2235 }
2236
2237 /*
2238 * Signal that the shared index is used by updating its mtime.
2239 *
2240 * This way, shared index can be removed if they have not been used
2241 * for some time.
2242 */
2243 static void freshen_shared_index(const char *shared_index, int warn)
2244 {
2245 if (!check_and_freshen_file(shared_index, 1) && warn)
2246 warning(_("could not freshen shared index '%s'"), shared_index);
2247 }
2248
2249 int read_index_from(struct index_state *istate, const char *path,
2250 const char *gitdir)
2251 {
2252 struct split_index *split_index;
2253 int ret;
2254 char *base_oid_hex;
2255 char *base_path;
2256
2257 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2258 if (istate->initialized)
2259 return istate->cache_nr;
2260
2261 /*
2262 * TODO trace2: replace "the_repository" with the actual repo instance
2263 * that is associated with the given "istate".
2264 */
2265 trace2_region_enter_printf("index", "do_read_index", the_repository,
2266 "%s", path);
2267 trace_performance_enter();
2268 ret = do_read_index(istate, path, 0);
2269 trace_performance_leave("read cache %s", path);
2270 trace2_region_leave_printf("index", "do_read_index", the_repository,
2271 "%s", path);
2272
2273 split_index = istate->split_index;
2274 if (!split_index || is_null_oid(&split_index->base_oid)) {
2275 post_read_index_from(istate);
2276 return ret;
2277 }
2278
2279 trace_performance_enter();
2280 if (split_index->base)
2281 discard_index(split_index->base);
2282 else
2283 split_index->base = xcalloc(1, sizeof(*split_index->base));
2284
2285 base_oid_hex = oid_to_hex(&split_index->base_oid);
2286 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2287 trace2_region_enter_printf("index", "shared/do_read_index",
2288 the_repository, "%s", base_path);
2289 ret = do_read_index(split_index->base, base_path, 1);
2290 trace2_region_leave_printf("index", "shared/do_read_index",
2291 the_repository, "%s", base_path);
2292 if (!oideq(&split_index->base_oid, &split_index->base->oid))
2293 die(_("broken index, expect %s in %s, got %s"),
2294 base_oid_hex, base_path,
2295 oid_to_hex(&split_index->base->oid));
2296
2297 freshen_shared_index(base_path, 0);
2298 merge_base_index(istate);
2299 post_read_index_from(istate);
2300 trace_performance_leave("read cache %s", base_path);
2301 free(base_path);
2302 return ret;
2303 }
2304
2305 int is_index_unborn(struct index_state *istate)
2306 {
2307 return (!istate->cache_nr && !istate->timestamp.sec);
2308 }
2309
2310 int discard_index(struct index_state *istate)
2311 {
2312 /*
2313 * Cache entries in istate->cache[] should have been allocated
2314 * from the memory pool associated with this index, or from an
2315 * associated split_index. There is no need to free individual
2316 * cache entries. validate_cache_entries can detect when this
2317 * assertion does not hold.
2318 */
2319 validate_cache_entries(istate);
2320
2321 resolve_undo_clear_index(istate);
2322 istate->cache_nr = 0;
2323 istate->cache_changed = 0;
2324 istate->timestamp.sec = 0;
2325 istate->timestamp.nsec = 0;
2326 free_name_hash(istate);
2327 cache_tree_free(&(istate->cache_tree));
2328 istate->initialized = 0;
2329 istate->fsmonitor_has_run_once = 0;
2330 FREE_AND_NULL(istate->cache);
2331 istate->cache_alloc = 0;
2332 discard_split_index(istate);
2333 free_untracked_cache(istate->untracked);
2334 istate->untracked = NULL;
2335
2336 if (istate->ce_mem_pool) {
2337 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2338 istate->ce_mem_pool = NULL;
2339 }
2340
2341 return 0;
2342 }
2343
2344 /*
2345 * Validate the cache entries of this index.
2346 * All cache entries associated with this index
2347 * should have been allocated by the memory pool
2348 * associated with this index, or by a referenced
2349 * split index.
2350 */
2351 void validate_cache_entries(const struct index_state *istate)
2352 {
2353 int i;
2354
2355 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2356 return;
2357
2358 for (i = 0; i < istate->cache_nr; i++) {
2359 if (!istate) {
2360 BUG("cache entry is not allocated from expected memory pool");
2361 } else if (!istate->ce_mem_pool ||
2362 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2363 if (!istate->split_index ||
2364 !istate->split_index->base ||
2365 !istate->split_index->base->ce_mem_pool ||
2366 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2367 BUG("cache entry is not allocated from expected memory pool");
2368 }
2369 }
2370 }
2371
2372 if (istate->split_index)
2373 validate_cache_entries(istate->split_index->base);
2374 }
2375
2376 int unmerged_index(const struct index_state *istate)
2377 {
2378 int i;
2379 for (i = 0; i < istate->cache_nr; i++) {
2380 if (ce_stage(istate->cache[i]))
2381 return 1;
2382 }
2383 return 0;
2384 }
2385
2386 int repo_index_has_changes(struct repository *repo,
2387 struct tree *tree,
2388 struct strbuf *sb)
2389 {
2390 struct index_state *istate = repo->index;
2391 struct object_id cmp;
2392 int i;
2393
2394 if (tree)
2395 cmp = tree->object.oid;
2396 if (tree || !get_oid_tree("HEAD", &cmp)) {
2397 struct diff_options opt;
2398
2399 repo_diff_setup(repo, &opt);
2400 opt.flags.exit_with_status = 1;
2401 if (!sb)
2402 opt.flags.quick = 1;
2403 do_diff_cache(&cmp, &opt);
2404 diffcore_std(&opt);
2405 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2406 if (i)
2407 strbuf_addch(sb, ' ');
2408 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2409 }
2410 diff_flush(&opt);
2411 return opt.flags.has_changes != 0;
2412 } else {
2413 for (i = 0; sb && i < istate->cache_nr; i++) {
2414 if (i)
2415 strbuf_addch(sb, ' ');
2416 strbuf_addstr(sb, istate->cache[i]->name);
2417 }
2418 return !!istate->cache_nr;
2419 }
2420 }
2421
2422 #define WRITE_BUFFER_SIZE 8192
2423 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2424 static unsigned long write_buffer_len;
2425
2426 static int ce_write_flush(git_hash_ctx *context, int fd)
2427 {
2428 unsigned int buffered = write_buffer_len;
2429 if (buffered) {
2430 the_hash_algo->update_fn(context, write_buffer, buffered);
2431 if (write_in_full(fd, write_buffer, buffered) < 0)
2432 return -1;
2433 write_buffer_len = 0;
2434 }
2435 return 0;
2436 }
2437
2438 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2439 {
2440 while (len) {
2441 unsigned int buffered = write_buffer_len;
2442 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2443 if (partial > len)
2444 partial = len;
2445 memcpy(write_buffer + buffered, data, partial);
2446 buffered += partial;
2447 if (buffered == WRITE_BUFFER_SIZE) {
2448 write_buffer_len = buffered;
2449 if (ce_write_flush(context, fd))
2450 return -1;
2451 buffered = 0;
2452 }
2453 write_buffer_len = buffered;
2454 len -= partial;
2455 data = (char *) data + partial;
2456 }
2457 return 0;
2458 }
2459
2460 static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2461 int fd, unsigned int ext, unsigned int sz)
2462 {
2463 ext = htonl(ext);
2464 sz = htonl(sz);
2465 if (eoie_context) {
2466 the_hash_algo->update_fn(eoie_context, &ext, 4);
2467 the_hash_algo->update_fn(eoie_context, &sz, 4);
2468 }
2469 return ((ce_write(context, fd, &ext, 4) < 0) ||
2470 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2471 }
2472
2473 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2474 {
2475 unsigned int left = write_buffer_len;
2476
2477 if (left) {
2478 write_buffer_len = 0;
2479 the_hash_algo->update_fn(context, write_buffer, left);
2480 }
2481
2482 /* Flush first if not enough space for hash signature */
2483 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2484 if (write_in_full(fd, write_buffer, left) < 0)
2485 return -1;
2486 left = 0;
2487 }
2488
2489 /* Append the hash signature at the end */
2490 the_hash_algo->final_fn(write_buffer + left, context);
2491 hashcpy(hash, write_buffer + left);
2492 left += the_hash_algo->rawsz;
2493 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2494 }
2495
2496 static void ce_smudge_racily_clean_entry(struct index_state *istate,
2497 struct cache_entry *ce)
2498 {
2499 /*
2500 * The only thing we care about in this function is to smudge the
2501 * falsely clean entry due to touch-update-touch race, so we leave
2502 * everything else as they are. We are called for entries whose
2503 * ce_stat_data.sd_mtime match the index file mtime.
2504 *
2505 * Note that this actually does not do much for gitlinks, for
2506 * which ce_match_stat_basic() always goes to the actual
2507 * contents. The caller checks with is_racy_timestamp() which
2508 * always says "no" for gitlinks, so we are not called for them ;-)
2509 */
2510 struct stat st;
2511
2512 if (lstat(ce->name, &st) < 0)
2513 return;
2514 if (ce_match_stat_basic(ce, &st))
2515 return;
2516 if (ce_modified_check_fs(istate, ce, &st)) {
2517 /* This is "racily clean"; smudge it. Note that this
2518 * is a tricky code. At first glance, it may appear
2519 * that it can break with this sequence:
2520 *
2521 * $ echo xyzzy >frotz
2522 * $ git-update-index --add frotz
2523 * $ : >frotz
2524 * $ sleep 3
2525 * $ echo filfre >nitfol
2526 * $ git-update-index --add nitfol
2527 *
2528 * but it does not. When the second update-index runs,
2529 * it notices that the entry "frotz" has the same timestamp
2530 * as index, and if we were to smudge it by resetting its
2531 * size to zero here, then the object name recorded
2532 * in index is the 6-byte file but the cached stat information
2533 * becomes zero --- which would then match what we would
2534 * obtain from the filesystem next time we stat("frotz").
2535 *
2536 * However, the second update-index, before calling
2537 * this function, notices that the cached size is 6
2538 * bytes and what is on the filesystem is an empty
2539 * file, and never calls us, so the cached size information
2540 * for "frotz" stays 6 which does not match the filesystem.
2541 */
2542 ce->ce_stat_data.sd_size = 0;
2543 }
2544 }
2545
2546 /* Copy miscellaneous fields but not the name */
2547 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2548 struct cache_entry *ce)
2549 {
2550 short flags;
2551 const unsigned hashsz = the_hash_algo->rawsz;
2552 uint16_t *flagsp = (uint16_t *)(ondisk->data + hashsz);
2553
2554 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2555 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2556 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2557 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2558 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2559 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2560 ondisk->mode = htonl(ce->ce_mode);
2561 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2562 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2563 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2564 hashcpy(ondisk->data, ce->oid.hash);
2565
2566 flags = ce->ce_flags & ~CE_NAMEMASK;
2567 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2568 flagsp[0] = htons(flags);
2569 if (ce->ce_flags & CE_EXTENDED) {
2570 flagsp[1] = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2571 }
2572 }
2573
2574 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2575 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2576 {
2577 int size;
2578 int result;
2579 unsigned int saved_namelen;
2580 int stripped_name = 0;
2581 static unsigned char padding[8] = { 0x00 };
2582
2583 if (ce->ce_flags & CE_STRIP_NAME) {
2584 saved_namelen = ce_namelen(ce);
2585 ce->ce_namelen = 0;
2586 stripped_name = 1;
2587 }
2588
2589 size = offsetof(struct ondisk_cache_entry,data) + ondisk_data_size(ce->ce_flags, 0);
2590
2591 if (!previous_name) {
2592 int len = ce_namelen(ce);
2593 copy_cache_entry_to_ondisk(ondisk, ce);
2594 result = ce_write(c, fd, ondisk, size);
2595 if (!result)
2596 result = ce_write(c, fd, ce->name, len);
2597 if (!result)
2598 result = ce_write(c, fd, padding, align_padding_size(size, len));
2599 } else {
2600 int common, to_remove, prefix_size;
2601 unsigned char to_remove_vi[16];
2602 for (common = 0;
2603 (ce->name[common] &&
2604 common < previous_name->len &&
2605 ce->name[common] == previous_name->buf[common]);
2606 common++)
2607 ; /* still matching */
2608 to_remove = previous_name->len - common;
2609 prefix_size = encode_varint(to_remove, to_remove_vi);
2610
2611 copy_cache_entry_to_ondisk(ondisk, ce);
2612 result = ce_write(c, fd, ondisk, size);
2613 if (!result)
2614 result = ce_write(c, fd, to_remove_vi, prefix_size);
2615 if (!result)
2616 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2617 if (!result)
2618 result = ce_write(c, fd, padding, 1);
2619
2620 strbuf_splice(previous_name, common, to_remove,
2621 ce->name + common, ce_namelen(ce) - common);
2622 }
2623 if (stripped_name) {
2624 ce->ce_namelen = saved_namelen;
2625 ce->ce_flags &= ~CE_STRIP_NAME;
2626 }
2627
2628 return result;
2629 }
2630
2631 /*
2632 * This function verifies if index_state has the correct sha1 of the
2633 * index file. Don't die if we have any other failure, just return 0.
2634 */
2635 static int verify_index_from(const struct index_state *istate, const char *path)
2636 {
2637 int fd;
2638 ssize_t n;
2639 struct stat st;
2640 unsigned char hash[GIT_MAX_RAWSZ];
2641
2642 if (!istate->initialized)
2643 return 0;
2644
2645 fd = open(path, O_RDONLY);
2646 if (fd < 0)
2647 return 0;
2648
2649 if (fstat(fd, &st))
2650 goto out;
2651
2652 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2653 goto out;
2654
2655 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2656 if (n != the_hash_algo->rawsz)
2657 goto out;
2658
2659 if (!hasheq(istate->oid.hash, hash))
2660 goto out;
2661
2662 close(fd);
2663 return 1;
2664
2665 out:
2666 close(fd);
2667 return 0;
2668 }
2669
2670 static int repo_verify_index(struct repository *repo)
2671 {
2672 return verify_index_from(repo->index, repo->index_file);
2673 }
2674
2675 static int has_racy_timestamp(struct index_state *istate)
2676 {
2677 int entries = istate->cache_nr;
2678 int i;
2679
2680 for (i = 0; i < entries; i++) {
2681 struct cache_entry *ce = istate->cache[i];
2682 if (is_racy_timestamp(istate, ce))
2683 return 1;
2684 }
2685 return 0;
2686 }
2687
2688 void repo_update_index_if_able(struct repository *repo,
2689 struct lock_file *lockfile)
2690 {
2691 if ((repo->index->cache_changed ||
2692 has_racy_timestamp(repo->index)) &&
2693 repo_verify_index(repo))
2694 write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2695 else
2696 rollback_lock_file(lockfile);
2697 }
2698
2699 static int record_eoie(void)
2700 {
2701 int val;
2702
2703 if (!git_config_get_bool("index.recordendofindexentries", &val))
2704 return val;
2705
2706 /*
2707 * As a convenience, the end of index entries extension
2708 * used for threading is written by default if the user
2709 * explicitly requested threaded index reads.
2710 */
2711 return !git_config_get_index_threads(&val) && val != 1;
2712 }
2713
2714 static int record_ieot(void)
2715 {
2716 int val;
2717
2718 if (!git_config_get_bool("index.recordoffsettable", &val))
2719 return val;
2720
2721 /*
2722 * As a convenience, the offset table used for threading is
2723 * written by default if the user explicitly requested
2724 * threaded index reads.
2725 */
2726 return !git_config_get_index_threads(&val) && val != 1;
2727 }
2728
2729 /*
2730 * On success, `tempfile` is closed. If it is the temporary file
2731 * of a `struct lock_file`, we will therefore effectively perform
2732 * a 'close_lock_file_gently()`. Since that is an implementation
2733 * detail of lockfiles, callers of `do_write_index()` should not
2734 * rely on it.
2735 */
2736 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2737 int strip_extensions)
2738 {
2739 uint64_t start = getnanotime();
2740 int newfd = tempfile->fd;
2741 git_hash_ctx c, eoie_c;
2742 struct cache_header hdr;
2743 int i, err = 0, removed, extended, hdr_version;
2744 struct cache_entry **cache = istate->cache;
2745 int entries = istate->cache_nr;
2746 struct stat st;
2747 struct ondisk_cache_entry ondisk;
2748 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2749 int drop_cache_tree = istate->drop_cache_tree;
2750 off_t offset;
2751 int ieot_entries = 1;
2752 struct index_entry_offset_table *ieot = NULL;
2753 int nr, nr_threads;
2754
2755 for (i = removed = extended = 0; i < entries; i++) {
2756 if (cache[i]->ce_flags & CE_REMOVE)
2757 removed++;
2758
2759 /* reduce extended entries if possible */
2760 cache[i]->ce_flags &= ~CE_EXTENDED;
2761 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2762 extended++;
2763 cache[i]->ce_flags |= CE_EXTENDED;
2764 }
2765 }
2766
2767 if (!istate->version) {
2768 istate->version = get_index_format_default(the_repository);
2769 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2770 init_split_index(istate);
2771 }
2772
2773 /* demote version 3 to version 2 when the latter suffices */
2774 if (istate->version == 3 || istate->version == 2)
2775 istate->version = extended ? 3 : 2;
2776
2777 hdr_version = istate->version;
2778
2779 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2780 hdr.hdr_version = htonl(hdr_version);
2781 hdr.hdr_entries = htonl(entries - removed);
2782
2783 the_hash_algo->init_fn(&c);
2784 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2785 return -1;
2786
2787 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))
2788 nr_threads = 1;
2789
2790 if (nr_threads != 1 && record_ieot()) {
2791 int ieot_blocks, cpus;
2792
2793 /*
2794 * ensure default number of ieot blocks maps evenly to the
2795 * default number of threads that will process them leaving
2796 * room for the thread to load the index extensions.
2797 */
2798 if (!nr_threads) {
2799 ieot_blocks = istate->cache_nr / THREAD_COST;
2800 cpus = online_cpus();
2801 if (ieot_blocks > cpus - 1)
2802 ieot_blocks = cpus - 1;
2803 } else {
2804 ieot_blocks = nr_threads;
2805 if (ieot_blocks > istate->cache_nr)
2806 ieot_blocks = istate->cache_nr;
2807 }
2808
2809 /*
2810 * no reason to write out the IEOT extension if we don't
2811 * have enough blocks to utilize multi-threading
2812 */
2813 if (ieot_blocks > 1) {
2814 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2815 + (ieot_blocks * sizeof(struct index_entry_offset)));
2816 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2817 }
2818 }
2819
2820 offset = lseek(newfd, 0, SEEK_CUR);
2821 if (offset < 0) {
2822 free(ieot);
2823 return -1;
2824 }
2825 offset += write_buffer_len;
2826 nr = 0;
2827 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2828
2829 for (i = 0; i < entries; i++) {
2830 struct cache_entry *ce = cache[i];
2831 if (ce->ce_flags & CE_REMOVE)
2832 continue;
2833 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2834 ce_smudge_racily_clean_entry(istate, ce);
2835 if (is_null_oid(&ce->oid)) {
2836 static const char msg[] = "cache entry has null sha1: %s";
2837 static int allow = -1;
2838
2839 if (allow < 0)
2840 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2841 if (allow)
2842 warning(msg, ce->name);
2843 else
2844 err = error(msg, ce->name);
2845
2846 drop_cache_tree = 1;
2847 }
2848 if (ieot && i && (i % ieot_entries == 0)) {
2849 ieot->entries[ieot->nr].nr = nr;
2850 ieot->entries[ieot->nr].offset = offset;
2851 ieot->nr++;
2852 /*
2853 * If we have a V4 index, set the first byte to an invalid
2854 * character to ensure there is nothing common with the previous
2855 * entry
2856 */
2857 if (previous_name)
2858 previous_name->buf[0] = 0;
2859 nr = 0;
2860 offset = lseek(newfd, 0, SEEK_CUR);
2861 if (offset < 0) {
2862 free(ieot);
2863 return -1;
2864 }
2865 offset += write_buffer_len;
2866 }
2867 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2868 err = -1;
2869
2870 if (err)
2871 break;
2872 nr++;
2873 }
2874 if (ieot && nr) {
2875 ieot->entries[ieot->nr].nr = nr;
2876 ieot->entries[ieot->nr].offset = offset;
2877 ieot->nr++;
2878 }
2879 strbuf_release(&previous_name_buf);
2880
2881 if (err) {
2882 free(ieot);
2883 return err;
2884 }
2885
2886 /* Write extension data here */
2887 offset = lseek(newfd, 0, SEEK_CUR);
2888 if (offset < 0) {
2889 free(ieot);
2890 return -1;
2891 }
2892 offset += write_buffer_len;
2893 the_hash_algo->init_fn(&eoie_c);
2894
2895 /*
2896 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2897 * can minimize the number of extensions we have to scan through to
2898 * find it during load. Write it out regardless of the
2899 * strip_extensions parameter as we need it when loading the shared
2900 * index.
2901 */
2902 if (ieot) {
2903 struct strbuf sb = STRBUF_INIT;
2904
2905 write_ieot_extension(&sb, ieot);
2906 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2907 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2908 strbuf_release(&sb);
2909 free(ieot);
2910 if (err)
2911 return -1;
2912 }
2913
2914 if (!strip_extensions && istate->split_index &&
2915 !is_null_oid(&istate->split_index->base_oid)) {
2916 struct strbuf sb = STRBUF_INIT;
2917
2918 err = write_link_extension(&sb, istate) < 0 ||
2919 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2920 sb.len) < 0 ||
2921 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2922 strbuf_release(&sb);
2923 if (err)
2924 return -1;
2925 }
2926 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2927 struct strbuf sb = STRBUF_INIT;
2928
2929 cache_tree_write(&sb, istate->cache_tree);
2930 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2931 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2932 strbuf_release(&sb);
2933 if (err)
2934 return -1;
2935 }
2936 if (!strip_extensions && istate->resolve_undo) {
2937 struct strbuf sb = STRBUF_INIT;
2938
2939 resolve_undo_write(&sb, istate->resolve_undo);
2940 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2941 sb.len) < 0
2942 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2943 strbuf_release(&sb);
2944 if (err)
2945 return -1;
2946 }
2947 if (!strip_extensions && istate->untracked) {
2948 struct strbuf sb = STRBUF_INIT;
2949
2950 write_untracked_extension(&sb, istate->untracked);
2951 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2952 sb.len) < 0 ||
2953 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2954 strbuf_release(&sb);
2955 if (err)
2956 return -1;
2957 }
2958 if (!strip_extensions && istate->fsmonitor_last_update) {
2959 struct strbuf sb = STRBUF_INIT;
2960
2961 write_fsmonitor_extension(&sb, istate);
2962 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2963 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2964 strbuf_release(&sb);
2965 if (err)
2966 return -1;
2967 }
2968
2969 /*
2970 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2971 * so that it can be found and processed before all the index entries are
2972 * read. Write it out regardless of the strip_extensions parameter as we need it
2973 * when loading the shared index.
2974 */
2975 if (offset && record_eoie()) {
2976 struct strbuf sb = STRBUF_INIT;
2977
2978 write_eoie_extension(&sb, &eoie_c, offset);
2979 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
2980 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2981 strbuf_release(&sb);
2982 if (err)
2983 return -1;
2984 }
2985
2986 if (ce_flush(&c, newfd, istate->oid.hash))
2987 return -1;
2988 if (close_tempfile_gently(tempfile)) {
2989 error(_("could not close '%s'"), tempfile->filename.buf);
2990 return -1;
2991 }
2992 if (stat(tempfile->filename.buf, &st))
2993 return -1;
2994 istate->timestamp.sec = (unsigned int)st.st_mtime;
2995 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2996 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
2997
2998 /*
2999 * TODO trace2: replace "the_repository" with the actual repo instance
3000 * that is associated with the given "istate".
3001 */
3002 trace2_data_intmax("index", the_repository, "write/version",
3003 istate->version);
3004 trace2_data_intmax("index", the_repository, "write/cache_nr",
3005 istate->cache_nr);
3006
3007 return 0;
3008 }
3009
3010 void set_alternate_index_output(const char *name)
3011 {
3012 alternate_index_output = name;
3013 }
3014
3015 static int commit_locked_index(struct lock_file *lk)
3016 {
3017 if (alternate_index_output)
3018 return commit_lock_file_to(lk, alternate_index_output);
3019 else
3020 return commit_lock_file(lk);
3021 }
3022
3023 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
3024 unsigned flags)
3025 {
3026 int ret;
3027
3028 /*
3029 * TODO trace2: replace "the_repository" with the actual repo instance
3030 * that is associated with the given "istate".
3031 */
3032 trace2_region_enter_printf("index", "do_write_index", the_repository,
3033 "%s", lock->tempfile->filename.buf);
3034 ret = do_write_index(istate, lock->tempfile, 0);
3035 trace2_region_leave_printf("index", "do_write_index", the_repository,
3036 "%s", lock->tempfile->filename.buf);
3037
3038 if (ret)
3039 return ret;
3040 if (flags & COMMIT_LOCK)
3041 ret = commit_locked_index(lock);
3042 else
3043 ret = close_lock_file_gently(lock);
3044
3045 run_hook_le(NULL, "post-index-change",
3046 istate->updated_workdir ? "1" : "0",
3047 istate->updated_skipworktree ? "1" : "0", NULL);
3048 istate->updated_workdir = 0;
3049 istate->updated_skipworktree = 0;
3050
3051 return ret;
3052 }
3053
3054 static int write_split_index(struct index_state *istate,
3055 struct lock_file *lock,
3056 unsigned flags)
3057 {
3058 int ret;
3059 prepare_to_write_split_index(istate);
3060 ret = do_write_locked_index(istate, lock, flags);
3061 finish_writing_split_index(istate);
3062 return ret;
3063 }
3064
3065 static const char *shared_index_expire = "2.weeks.ago";
3066
3067 static unsigned long get_shared_index_expire_date(void)
3068 {
3069 static unsigned long shared_index_expire_date;
3070 static int shared_index_expire_date_prepared;
3071
3072 if (!shared_index_expire_date_prepared) {
3073 git_config_get_expiry("splitindex.sharedindexexpire",
3074 &shared_index_expire);
3075 shared_index_expire_date = approxidate(shared_index_expire);
3076 shared_index_expire_date_prepared = 1;
3077 }
3078
3079 return shared_index_expire_date;
3080 }
3081
3082 static int should_delete_shared_index(const char *shared_index_path)
3083 {
3084 struct stat st;
3085 unsigned long expiration;
3086
3087 /* Check timestamp */
3088 expiration = get_shared_index_expire_date();
3089 if (!expiration)
3090 return 0;
3091 if (stat(shared_index_path, &st))
3092 return error_errno(_("could not stat '%s'"), shared_index_path);
3093 if (st.st_mtime > expiration)
3094 return 0;
3095
3096 return 1;
3097 }
3098
3099 static int clean_shared_index_files(const char *current_hex)
3100 {
3101 struct dirent *de;
3102 DIR *dir = opendir(get_git_dir());
3103
3104 if (!dir)
3105 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3106
3107 while ((de = readdir(dir)) != NULL) {
3108 const char *sha1_hex;
3109 const char *shared_index_path;
3110 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3111 continue;
3112 if (!strcmp(sha1_hex, current_hex))
3113 continue;
3114 shared_index_path = git_path("%s", de->d_name);
3115 if (should_delete_shared_index(shared_index_path) > 0 &&
3116 unlink(shared_index_path))
3117 warning_errno(_("unable to unlink: %s"), shared_index_path);
3118 }
3119 closedir(dir);
3120
3121 return 0;
3122 }
3123
3124 static int write_shared_index(struct index_state *istate,
3125 struct tempfile **temp)
3126 {
3127 struct split_index *si = istate->split_index;
3128 int ret;
3129
3130 move_cache_to_base_index(istate);
3131
3132 trace2_region_enter_printf("index", "shared/do_write_index",
3133 the_repository, "%s", (*temp)->filename.buf);
3134 ret = do_write_index(si->base, *temp, 1);
3135 trace2_region_leave_printf("index", "shared/do_write_index",
3136 the_repository, "%s", (*temp)->filename.buf);
3137
3138 if (ret)
3139 return ret;
3140 ret = adjust_shared_perm(get_tempfile_path(*temp));
3141 if (ret) {
3142 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3143 return ret;
3144 }
3145 ret = rename_tempfile(temp,
3146 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3147 if (!ret) {
3148 oidcpy(&si->base_oid, &si->base->oid);
3149 clean_shared_index_files(oid_to_hex(&si->base->oid));
3150 }
3151
3152 return ret;
3153 }
3154
3155 static const int default_max_percent_split_change = 20;
3156
3157 static int too_many_not_shared_entries(struct index_state *istate)
3158 {
3159 int i, not_shared = 0;
3160 int max_split = git_config_get_max_percent_split_change();
3161
3162 switch (max_split) {
3163 case -1:
3164 /* not or badly configured: use the default value */
3165 max_split = default_max_percent_split_change;
3166 break;
3167 case 0:
3168 return 1; /* 0% means always write a new shared index */
3169 case 100:
3170 return 0; /* 100% means never write a new shared index */
3171 default:
3172 break; /* just use the configured value */
3173 }
3174
3175 /* Count not shared entries */
3176 for (i = 0; i < istate->cache_nr; i++) {
3177 struct cache_entry *ce = istate->cache[i];
3178 if (!ce->index)
3179 not_shared++;
3180 }
3181
3182 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3183 }
3184
3185 int write_locked_index(struct index_state *istate, struct lock_file *lock,
3186 unsigned flags)
3187 {
3188 int new_shared_index, ret;
3189 struct split_index *si = istate->split_index;
3190
3191 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3192 cache_tree_verify(the_repository, istate);
3193
3194 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3195 if (flags & COMMIT_LOCK)
3196 rollback_lock_file(lock);
3197 return 0;
3198 }
3199
3200 if (istate->fsmonitor_last_update)
3201 fill_fsmonitor_bitmap(istate);
3202
3203 if (!si || alternate_index_output ||
3204 (istate->cache_changed & ~EXTMASK)) {
3205 if (si)
3206 oidclr(&si->base_oid);
3207 ret = do_write_locked_index(istate, lock, flags);
3208 goto out;
3209 }
3210
3211 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3212 int v = si->base_oid.hash[0];
3213 if ((v & 15) < 6)
3214 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3215 }
3216 if (too_many_not_shared_entries(istate))
3217 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3218
3219 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3220
3221 if (new_shared_index) {
3222 struct tempfile *temp;
3223 int saved_errno;
3224
3225 /* Same initial permissions as the main .git/index file */
3226 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3227 if (!temp) {
3228 oidclr(&si->base_oid);
3229 ret = do_write_locked_index(istate, lock, flags);
3230 goto out;
3231 }
3232 ret = write_shared_index(istate, &temp);
3233
3234 saved_errno = errno;
3235 if (is_tempfile_active(temp))
3236 delete_tempfile(&temp);
3237 errno = saved_errno;
3238
3239 if (ret)
3240 goto out;
3241 }
3242
3243 ret = write_split_index(istate, lock, flags);
3244
3245 /* Freshen the shared index only if the split-index was written */
3246 if (!ret && !new_shared_index && !is_null_oid(&si->base_oid)) {
3247 const char *shared_index = git_path("sharedindex.%s",
3248 oid_to_hex(&si->base_oid));
3249 freshen_shared_index(shared_index, 1);
3250 }
3251
3252 out:
3253 if (flags & COMMIT_LOCK)
3254 rollback_lock_file(lock);
3255 return ret;
3256 }
3257
3258 /*
3259 * Read the index file that is potentially unmerged into given
3260 * index_state, dropping any unmerged entries to stage #0 (potentially
3261 * resulting in a path appearing as both a file and a directory in the
3262 * index; the caller is responsible to clear out the extra entries
3263 * before writing the index to a tree). Returns true if the index is
3264 * unmerged. Callers who want to refuse to work from an unmerged
3265 * state can call this and check its return value, instead of calling
3266 * read_cache().
3267 */
3268 int repo_read_index_unmerged(struct repository *repo)
3269 {
3270 struct index_state *istate;
3271 int i;
3272 int unmerged = 0;
3273
3274 repo_read_index(repo);
3275 istate = repo->index;
3276 for (i = 0; i < istate->cache_nr; i++) {
3277 struct cache_entry *ce = istate->cache[i];
3278 struct cache_entry *new_ce;
3279 int len;
3280
3281 if (!ce_stage(ce))
3282 continue;
3283 unmerged = 1;
3284 len = ce_namelen(ce);
3285 new_ce = make_empty_cache_entry(istate, len);
3286 memcpy(new_ce->name, ce->name, len);
3287 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3288 new_ce->ce_namelen = len;
3289 new_ce->ce_mode = ce->ce_mode;
3290 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3291 return error(_("%s: cannot drop to stage #0"),
3292 new_ce->name);
3293 }
3294 return unmerged;
3295 }
3296
3297 /*
3298 * Returns 1 if the path is an "other" path with respect to
3299 * the index; that is, the path is not mentioned in the index at all,
3300 * either as a file, a directory with some files in the index,
3301 * or as an unmerged entry.
3302 *
3303 * We helpfully remove a trailing "/" from directories so that
3304 * the output of read_directory can be used as-is.
3305 */
3306 int index_name_is_other(const struct index_state *istate, const char *name,
3307 int namelen)
3308 {
3309 int pos;
3310 if (namelen && name[namelen - 1] == '/')
3311 namelen--;
3312 pos = index_name_pos(istate, name, namelen);
3313 if (0 <= pos)
3314 return 0; /* exact match */
3315 pos = -pos - 1;
3316 if (pos < istate->cache_nr) {
3317 struct cache_entry *ce = istate->cache[pos];
3318 if (ce_namelen(ce) == namelen &&
3319 !memcmp(ce->name, name, namelen))
3320 return 0; /* Yup, this one exists unmerged */
3321 }
3322 return 1;
3323 }
3324
3325 void *read_blob_data_from_index(const struct index_state *istate,
3326 const char *path, unsigned long *size)
3327 {
3328 int pos, len;
3329 unsigned long sz;
3330 enum object_type type;
3331 void *data;
3332
3333 len = strlen(path);
3334 pos = index_name_pos(istate, path, len);
3335 if (pos < 0) {
3336 /*
3337 * We might be in the middle of a merge, in which
3338 * case we would read stage #2 (ours).
3339 */
3340 int i;
3341 for (i = -pos - 1;
3342 (pos < 0 && i < istate->cache_nr &&
3343 !strcmp(istate->cache[i]->name, path));
3344 i++)
3345 if (ce_stage(istate->cache[i]) == 2)
3346 pos = i;
3347 }
3348 if (pos < 0)
3349 return NULL;
3350 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3351 if (!data || type != OBJ_BLOB) {
3352 free(data);
3353 return NULL;
3354 }
3355 if (size)
3356 *size = sz;
3357 return data;
3358 }
3359
3360 void stat_validity_clear(struct stat_validity *sv)
3361 {
3362 FREE_AND_NULL(sv->sd);
3363 }
3364
3365 int stat_validity_check(struct stat_validity *sv, const char *path)
3366 {
3367 struct stat st;
3368
3369 if (stat(path, &st) < 0)
3370 return sv->sd == NULL;
3371 if (!sv->sd)
3372 return 0;
3373 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3374 }
3375
3376 void stat_validity_update(struct stat_validity *sv, int fd)
3377 {
3378 struct stat st;
3379
3380 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3381 stat_validity_clear(sv);
3382 else {
3383 if (!sv->sd)
3384 sv->sd = xcalloc(1, sizeof(struct stat_data));
3385 fill_stat_data(sv->sd, &st);
3386 }
3387 }
3388
3389 void move_index_extensions(struct index_state *dst, struct index_state *src)
3390 {
3391 dst->untracked = src->untracked;
3392 src->untracked = NULL;
3393 dst->cache_tree = src->cache_tree;
3394 src->cache_tree = NULL;
3395 }
3396
3397 struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3398 struct index_state *istate)
3399 {
3400 unsigned int size = ce_size(ce);
3401 int mem_pool_allocated;
3402 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3403 mem_pool_allocated = new_entry->mem_pool_allocated;
3404
3405 memcpy(new_entry, ce, size);
3406 new_entry->mem_pool_allocated = mem_pool_allocated;
3407 return new_entry;
3408 }
3409
3410 void discard_cache_entry(struct cache_entry *ce)
3411 {
3412 if (ce && should_validate_cache_entries())
3413 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3414
3415 if (ce && ce->mem_pool_allocated)
3416 return;
3417
3418 free(ce);
3419 }
3420
3421 int should_validate_cache_entries(void)
3422 {
3423 static int validate_index_cache_entries = -1;
3424
3425 if (validate_index_cache_entries < 0) {
3426 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3427 validate_index_cache_entries = 1;
3428 else
3429 validate_index_cache_entries = 0;
3430 }
3431
3432 return validate_index_cache_entries;
3433 }
3434
3435 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3436 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3437
3438 static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3439 {
3440 /*
3441 * The end of index entries (EOIE) extension is guaranteed to be last
3442 * so that it can be found by scanning backwards from the EOF.
3443 *
3444 * "EOIE"
3445 * <4-byte length>
3446 * <4-byte offset>
3447 * <20-byte hash>
3448 */
3449 const char *index, *eoie;
3450 uint32_t extsize;
3451 size_t offset, src_offset;
3452 unsigned char hash[GIT_MAX_RAWSZ];
3453 git_hash_ctx c;
3454
3455 /* ensure we have an index big enough to contain an EOIE extension */
3456 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3457 return 0;
3458
3459 /* validate the extension signature */
3460 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3461 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3462 return 0;
3463 index += sizeof(uint32_t);
3464
3465 /* validate the extension size */
3466 extsize = get_be32(index);
3467 if (extsize != EOIE_SIZE)
3468 return 0;
3469 index += sizeof(uint32_t);
3470
3471 /*
3472 * Validate the offset we're going to look for the first extension
3473 * signature is after the index header and before the eoie extension.
3474 */
3475 offset = get_be32(index);
3476 if (mmap + offset < mmap + sizeof(struct cache_header))
3477 return 0;
3478 if (mmap + offset >= eoie)
3479 return 0;
3480 index += sizeof(uint32_t);
3481
3482 /*
3483 * The hash is computed over extension types and their sizes (but not
3484 * their contents). E.g. if we have "TREE" extension that is N-bytes
3485 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3486 * then the hash would be:
3487 *
3488 * SHA-1("TREE" + <binary representation of N> +
3489 * "REUC" + <binary representation of M>)
3490 */
3491 src_offset = offset;
3492 the_hash_algo->init_fn(&c);
3493 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3494 /* After an array of active_nr index entries,
3495 * there can be arbitrary number of extended
3496 * sections, each of which is prefixed with
3497 * extension name (4-byte) and section length
3498 * in 4-byte network byte order.
3499 */
3500 uint32_t extsize;
3501 memcpy(&extsize, mmap + src_offset + 4, 4);
3502 extsize = ntohl(extsize);
3503
3504 /* verify the extension size isn't so large it will wrap around */
3505 if (src_offset + 8 + extsize < src_offset)
3506 return 0;
3507
3508 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3509
3510 src_offset += 8;
3511 src_offset += extsize;
3512 }
3513 the_hash_algo->final_fn(hash, &c);
3514 if (!hasheq(hash, (const unsigned char *)index))
3515 return 0;
3516
3517 /* Validate that the extension offsets returned us back to the eoie extension. */
3518 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3519 return 0;
3520
3521 return offset;
3522 }
3523
3524 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3525 {
3526 uint32_t buffer;
3527 unsigned char hash[GIT_MAX_RAWSZ];
3528
3529 /* offset */
3530 put_be32(&buffer, offset);
3531 strbuf_add(sb, &buffer, sizeof(uint32_t));
3532
3533 /* hash */
3534 the_hash_algo->final_fn(hash, eoie_context);
3535 strbuf_add(sb, hash, the_hash_algo->rawsz);
3536 }
3537
3538 #define IEOT_VERSION (1)
3539
3540 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3541 {
3542 const char *index = NULL;
3543 uint32_t extsize, ext_version;
3544 struct index_entry_offset_table *ieot;
3545 int i, nr;
3546
3547 /* find the IEOT extension */
3548 if (!offset)
3549 return NULL;
3550 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3551 extsize = get_be32(mmap + offset + 4);
3552 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3553 index = mmap + offset + 4 + 4;
3554 break;
3555 }
3556 offset += 8;
3557 offset += extsize;
3558 }
3559 if (!index)
3560 return NULL;
3561
3562 /* validate the version is IEOT_VERSION */
3563 ext_version = get_be32(index);
3564 if (ext_version != IEOT_VERSION) {
3565 error("invalid IEOT version %d", ext_version);
3566 return NULL;
3567 }
3568 index += sizeof(uint32_t);
3569
3570 /* extension size - version bytes / bytes per entry */
3571 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3572 if (!nr) {
3573 error("invalid number of IEOT entries %d", nr);
3574 return NULL;
3575 }
3576 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3577 + (nr * sizeof(struct index_entry_offset)));
3578 ieot->nr = nr;
3579 for (i = 0; i < nr; i++) {
3580 ieot->entries[i].offset = get_be32(index);
3581 index += sizeof(uint32_t);
3582 ieot->entries[i].nr = get_be32(index);
3583 index += sizeof(uint32_t);
3584 }
3585
3586 return ieot;
3587 }
3588
3589 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3590 {
3591 uint32_t buffer;
3592 int i;
3593
3594 /* version */
3595 put_be32(&buffer, IEOT_VERSION);
3596 strbuf_add(sb, &buffer, sizeof(uint32_t));
3597
3598 /* ieot */
3599 for (i = 0; i < ieot->nr; i++) {
3600
3601 /* offset */
3602 put_be32(&buffer, ieot->entries[i].offset);
3603 strbuf_add(sb, &buffer, sizeof(uint32_t));
3604
3605 /* count */
3606 put_be32(&buffer, ieot->entries[i].nr);
3607 strbuf_add(sb, &buffer, sizeof(uint32_t));
3608 }
3609 }