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