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