Merge branch 'jh/add-index-entry-optim'
[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 "tempfile.h"
9 #include "lockfile.h"
10 #include "cache-tree.h"
11 #include "refs.h"
12 #include "dir.h"
13 #include "tree.h"
14 #include "commit.h"
15 #include "blob.h"
16 #include "resolve-undo.h"
17 #include "strbuf.h"
18 #include "varint.h"
19 #include "split-index.h"
20 #include "utf8.h"
21
22 /* Mask for the name length in ce_flags in the on-disk index */
23
24 #define CE_NAMEMASK (0x0fff)
25
26 /* Index extensions.
27 *
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
33 */
34
35 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
37 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
39 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
40
41 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
42 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
43 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
44 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
45
46 struct index_state the_index;
47 static const char *alternate_index_output;
48
49 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
50 {
51 istate->cache[nr] = ce;
52 add_name_hash(istate, ce);
53 }
54
55 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
56 {
57 struct cache_entry *old = istate->cache[nr];
58
59 replace_index_entry_in_base(istate, old, ce);
60 remove_name_hash(istate, old);
61 free(old);
62 set_index_entry(istate, nr, ce);
63 ce->ce_flags |= CE_UPDATE_IN_BASE;
64 istate->cache_changed |= CE_ENTRY_CHANGED;
65 }
66
67 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
68 {
69 struct cache_entry *old = istate->cache[nr], *new;
70 int namelen = strlen(new_name);
71
72 new = xmalloc(cache_entry_size(namelen));
73 copy_cache_entry(new, old);
74 new->ce_flags &= ~CE_HASHED;
75 new->ce_namelen = namelen;
76 new->index = 0;
77 memcpy(new->name, new_name, namelen + 1);
78
79 cache_tree_invalidate_path(istate, old->name);
80 untracked_cache_remove_from_index(istate, old->name);
81 remove_index_entry_at(istate, nr);
82 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
83 }
84
85 void fill_stat_data(struct stat_data *sd, struct stat *st)
86 {
87 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
88 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
89 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
90 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
91 sd->sd_dev = st->st_dev;
92 sd->sd_ino = st->st_ino;
93 sd->sd_uid = st->st_uid;
94 sd->sd_gid = st->st_gid;
95 sd->sd_size = st->st_size;
96 }
97
98 int match_stat_data(const struct stat_data *sd, struct stat *st)
99 {
100 int changed = 0;
101
102 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
103 changed |= MTIME_CHANGED;
104 if (trust_ctime && check_stat &&
105 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
106 changed |= CTIME_CHANGED;
107
108 #ifdef USE_NSEC
109 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
110 changed |= MTIME_CHANGED;
111 if (trust_ctime && check_stat &&
112 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
113 changed |= CTIME_CHANGED;
114 #endif
115
116 if (check_stat) {
117 if (sd->sd_uid != (unsigned int) st->st_uid ||
118 sd->sd_gid != (unsigned int) st->st_gid)
119 changed |= OWNER_CHANGED;
120 if (sd->sd_ino != (unsigned int) st->st_ino)
121 changed |= INODE_CHANGED;
122 }
123
124 #ifdef USE_STDEV
125 /*
126 * st_dev breaks on network filesystems where different
127 * clients will have different views of what "device"
128 * the filesystem is on
129 */
130 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
131 changed |= INODE_CHANGED;
132 #endif
133
134 if (sd->sd_size != (unsigned int) st->st_size)
135 changed |= DATA_CHANGED;
136
137 return changed;
138 }
139
140 /*
141 * This only updates the "non-critical" parts of the directory
142 * cache, ie the parts that aren't tracked by GIT, and only used
143 * to validate the cache.
144 */
145 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
146 {
147 fill_stat_data(&ce->ce_stat_data, st);
148
149 if (assume_unchanged)
150 ce->ce_flags |= CE_VALID;
151
152 if (S_ISREG(st->st_mode))
153 ce_mark_uptodate(ce);
154 }
155
156 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
157 {
158 int match = -1;
159 int fd = git_open_cloexec(ce->name, O_RDONLY);
160
161 if (fd >= 0) {
162 unsigned char sha1[20];
163 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
164 match = hashcmp(sha1, ce->oid.hash);
165 /* index_fd() closed the file descriptor already */
166 }
167 return match;
168 }
169
170 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
171 {
172 int match = -1;
173 void *buffer;
174 unsigned long size;
175 enum object_type type;
176 struct strbuf sb = STRBUF_INIT;
177
178 if (strbuf_readlink(&sb, ce->name, expected_size))
179 return -1;
180
181 buffer = read_sha1_file(ce->oid.hash, &type, &size);
182 if (buffer) {
183 if (size == sb.len)
184 match = memcmp(buffer, sb.buf, size);
185 free(buffer);
186 }
187 strbuf_release(&sb);
188 return match;
189 }
190
191 static int ce_compare_gitlink(const struct cache_entry *ce)
192 {
193 unsigned char sha1[20];
194
195 /*
196 * We don't actually require that the .git directory
197 * under GITLINK directory be a valid git directory. It
198 * might even be missing (in case nobody populated that
199 * sub-project).
200 *
201 * If so, we consider it always to match.
202 */
203 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
204 return 0;
205 return hashcmp(sha1, ce->oid.hash);
206 }
207
208 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
209 {
210 switch (st->st_mode & S_IFMT) {
211 case S_IFREG:
212 if (ce_compare_data(ce, st))
213 return DATA_CHANGED;
214 break;
215 case S_IFLNK:
216 if (ce_compare_link(ce, xsize_t(st->st_size)))
217 return DATA_CHANGED;
218 break;
219 case S_IFDIR:
220 if (S_ISGITLINK(ce->ce_mode))
221 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
222 default:
223 return TYPE_CHANGED;
224 }
225 return 0;
226 }
227
228 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
229 {
230 unsigned int changed = 0;
231
232 if (ce->ce_flags & CE_REMOVE)
233 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
234
235 switch (ce->ce_mode & S_IFMT) {
236 case S_IFREG:
237 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
238 /* We consider only the owner x bit to be relevant for
239 * "mode changes"
240 */
241 if (trust_executable_bit &&
242 (0100 & (ce->ce_mode ^ st->st_mode)))
243 changed |= MODE_CHANGED;
244 break;
245 case S_IFLNK:
246 if (!S_ISLNK(st->st_mode) &&
247 (has_symlinks || !S_ISREG(st->st_mode)))
248 changed |= TYPE_CHANGED;
249 break;
250 case S_IFGITLINK:
251 /* We ignore most of the st_xxx fields for gitlinks */
252 if (!S_ISDIR(st->st_mode))
253 changed |= TYPE_CHANGED;
254 else if (ce_compare_gitlink(ce))
255 changed |= DATA_CHANGED;
256 return changed;
257 default:
258 die("internal error: ce_mode is %o", ce->ce_mode);
259 }
260
261 changed |= match_stat_data(&ce->ce_stat_data, st);
262
263 /* Racily smudged entry? */
264 if (!ce->ce_stat_data.sd_size) {
265 if (!is_empty_blob_sha1(ce->oid.hash))
266 changed |= DATA_CHANGED;
267 }
268
269 return changed;
270 }
271
272 static int is_racy_stat(const struct index_state *istate,
273 const struct stat_data *sd)
274 {
275 return (istate->timestamp.sec &&
276 #ifdef USE_NSEC
277 /* nanosecond timestamped files can also be racy! */
278 (istate->timestamp.sec < sd->sd_mtime.sec ||
279 (istate->timestamp.sec == sd->sd_mtime.sec &&
280 istate->timestamp.nsec <= sd->sd_mtime.nsec))
281 #else
282 istate->timestamp.sec <= sd->sd_mtime.sec
283 #endif
284 );
285 }
286
287 static int is_racy_timestamp(const struct index_state *istate,
288 const struct cache_entry *ce)
289 {
290 return (!S_ISGITLINK(ce->ce_mode) &&
291 is_racy_stat(istate, &ce->ce_stat_data));
292 }
293
294 int match_stat_data_racy(const struct index_state *istate,
295 const struct stat_data *sd, struct stat *st)
296 {
297 if (is_racy_stat(istate, sd))
298 return MTIME_CHANGED;
299 return match_stat_data(sd, st);
300 }
301
302 int ie_match_stat(const struct index_state *istate,
303 const struct cache_entry *ce, struct stat *st,
304 unsigned int options)
305 {
306 unsigned int changed;
307 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
308 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
309 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
310
311 /*
312 * If it's marked as always valid in the index, it's
313 * valid whatever the checked-out copy says.
314 *
315 * skip-worktree has the same effect with higher precedence
316 */
317 if (!ignore_skip_worktree && ce_skip_worktree(ce))
318 return 0;
319 if (!ignore_valid && (ce->ce_flags & CE_VALID))
320 return 0;
321
322 /*
323 * Intent-to-add entries have not been added, so the index entry
324 * by definition never matches what is in the work tree until it
325 * actually gets added.
326 */
327 if (ce_intent_to_add(ce))
328 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
329
330 changed = ce_match_stat_basic(ce, st);
331
332 /*
333 * Within 1 second of this sequence:
334 * echo xyzzy >file && git-update-index --add file
335 * running this command:
336 * echo frotz >file
337 * would give a falsely clean cache entry. The mtime and
338 * length match the cache, and other stat fields do not change.
339 *
340 * We could detect this at update-index time (the cache entry
341 * being registered/updated records the same time as "now")
342 * and delay the return from git-update-index, but that would
343 * effectively mean we can make at most one commit per second,
344 * which is not acceptable. Instead, we check cache entries
345 * whose mtime are the same as the index file timestamp more
346 * carefully than others.
347 */
348 if (!changed && is_racy_timestamp(istate, ce)) {
349 if (assume_racy_is_modified)
350 changed |= DATA_CHANGED;
351 else
352 changed |= ce_modified_check_fs(ce, st);
353 }
354
355 return changed;
356 }
357
358 int ie_modified(const struct index_state *istate,
359 const struct cache_entry *ce,
360 struct stat *st, unsigned int options)
361 {
362 int changed, changed_fs;
363
364 changed = ie_match_stat(istate, ce, st, options);
365 if (!changed)
366 return 0;
367 /*
368 * If the mode or type has changed, there's no point in trying
369 * to refresh the entry - it's not going to match
370 */
371 if (changed & (MODE_CHANGED | TYPE_CHANGED))
372 return changed;
373
374 /*
375 * Immediately after read-tree or update-index --cacheinfo,
376 * the length field is zero, as we have never even read the
377 * lstat(2) information once, and we cannot trust DATA_CHANGED
378 * returned by ie_match_stat() which in turn was returned by
379 * ce_match_stat_basic() to signal that the filesize of the
380 * blob changed. We have to actually go to the filesystem to
381 * see if the contents match, and if so, should answer "unchanged".
382 *
383 * The logic does not apply to gitlinks, as ce_match_stat_basic()
384 * already has checked the actual HEAD from the filesystem in the
385 * subproject. If ie_match_stat() already said it is different,
386 * then we know it is.
387 */
388 if ((changed & DATA_CHANGED) &&
389 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
390 return changed;
391
392 changed_fs = ce_modified_check_fs(ce, st);
393 if (changed_fs)
394 return changed | changed_fs;
395 return 0;
396 }
397
398 int base_name_compare(const char *name1, int len1, int mode1,
399 const char *name2, int len2, int mode2)
400 {
401 unsigned char c1, c2;
402 int len = len1 < len2 ? len1 : len2;
403 int cmp;
404
405 cmp = memcmp(name1, name2, len);
406 if (cmp)
407 return cmp;
408 c1 = name1[len];
409 c2 = name2[len];
410 if (!c1 && S_ISDIR(mode1))
411 c1 = '/';
412 if (!c2 && S_ISDIR(mode2))
413 c2 = '/';
414 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
415 }
416
417 /*
418 * df_name_compare() is identical to base_name_compare(), except it
419 * compares conflicting directory/file entries as equal. Note that
420 * while a directory name compares as equal to a regular file, they
421 * then individually compare _differently_ to a filename that has
422 * a dot after the basename (because '\0' < '.' < '/').
423 *
424 * This is used by routines that want to traverse the git namespace
425 * but then handle conflicting entries together when possible.
426 */
427 int df_name_compare(const char *name1, int len1, int mode1,
428 const char *name2, int len2, int mode2)
429 {
430 int len = len1 < len2 ? len1 : len2, cmp;
431 unsigned char c1, c2;
432
433 cmp = memcmp(name1, name2, len);
434 if (cmp)
435 return cmp;
436 /* Directories and files compare equal (same length, same name) */
437 if (len1 == len2)
438 return 0;
439 c1 = name1[len];
440 if (!c1 && S_ISDIR(mode1))
441 c1 = '/';
442 c2 = name2[len];
443 if (!c2 && S_ISDIR(mode2))
444 c2 = '/';
445 if (c1 == '/' && !c2)
446 return 0;
447 if (c2 == '/' && !c1)
448 return 0;
449 return c1 - c2;
450 }
451
452 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
453 {
454 size_t min_len = (len1 < len2) ? len1 : len2;
455 int cmp = memcmp(name1, name2, min_len);
456 if (cmp)
457 return cmp;
458 if (len1 < len2)
459 return -1;
460 if (len1 > len2)
461 return 1;
462 return 0;
463 }
464
465 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
466 {
467 int cmp;
468
469 cmp = name_compare(name1, len1, name2, len2);
470 if (cmp)
471 return cmp;
472
473 if (stage1 < stage2)
474 return -1;
475 if (stage1 > stage2)
476 return 1;
477 return 0;
478 }
479
480 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
481 {
482 int first, last;
483
484 first = 0;
485 last = istate->cache_nr;
486 while (last > first) {
487 int next = (last + first) >> 1;
488 struct cache_entry *ce = istate->cache[next];
489 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
490 if (!cmp)
491 return next;
492 if (cmp < 0) {
493 last = next;
494 continue;
495 }
496 first = next+1;
497 }
498 return -first-1;
499 }
500
501 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
502 {
503 return index_name_stage_pos(istate, name, namelen, 0);
504 }
505
506 int remove_index_entry_at(struct index_state *istate, int pos)
507 {
508 struct cache_entry *ce = istate->cache[pos];
509
510 record_resolve_undo(istate, ce);
511 remove_name_hash(istate, ce);
512 save_or_free_index_entry(istate, ce);
513 istate->cache_changed |= CE_ENTRY_REMOVED;
514 istate->cache_nr--;
515 if (pos >= istate->cache_nr)
516 return 0;
517 memmove(istate->cache + pos,
518 istate->cache + pos + 1,
519 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
520 return 1;
521 }
522
523 /*
524 * Remove all cache entries marked for removal, that is where
525 * CE_REMOVE is set in ce_flags. This is much more effective than
526 * calling remove_index_entry_at() for each entry to be removed.
527 */
528 void remove_marked_cache_entries(struct index_state *istate)
529 {
530 struct cache_entry **ce_array = istate->cache;
531 unsigned int i, j;
532
533 for (i = j = 0; i < istate->cache_nr; i++) {
534 if (ce_array[i]->ce_flags & CE_REMOVE) {
535 remove_name_hash(istate, ce_array[i]);
536 save_or_free_index_entry(istate, ce_array[i]);
537 }
538 else
539 ce_array[j++] = ce_array[i];
540 }
541 if (j == istate->cache_nr)
542 return;
543 istate->cache_changed |= CE_ENTRY_REMOVED;
544 istate->cache_nr = j;
545 }
546
547 int remove_file_from_index(struct index_state *istate, const char *path)
548 {
549 int pos = index_name_pos(istate, path, strlen(path));
550 if (pos < 0)
551 pos = -pos-1;
552 cache_tree_invalidate_path(istate, path);
553 untracked_cache_remove_from_index(istate, path);
554 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
555 remove_index_entry_at(istate, pos);
556 return 0;
557 }
558
559 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
560 {
561 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
562 }
563
564 static int index_name_pos_also_unmerged(struct index_state *istate,
565 const char *path, int namelen)
566 {
567 int pos = index_name_pos(istate, path, namelen);
568 struct cache_entry *ce;
569
570 if (pos >= 0)
571 return pos;
572
573 /* maybe unmerged? */
574 pos = -1 - pos;
575 if (pos >= istate->cache_nr ||
576 compare_name((ce = istate->cache[pos]), path, namelen))
577 return -1;
578
579 /* order of preference: stage 2, 1, 3 */
580 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
581 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
582 !compare_name(ce, path, namelen))
583 pos++;
584 return pos;
585 }
586
587 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
588 {
589 int len = ce_namelen(ce);
590 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
591 }
592
593 /*
594 * If we add a filename that aliases in the cache, we will use the
595 * name that we already have - but we don't want to update the same
596 * alias twice, because that implies that there were actually two
597 * different files with aliasing names!
598 *
599 * So we use the CE_ADDED flag to verify that the alias was an old
600 * one before we accept it as
601 */
602 static struct cache_entry *create_alias_ce(struct index_state *istate,
603 struct cache_entry *ce,
604 struct cache_entry *alias)
605 {
606 int len;
607 struct cache_entry *new;
608
609 if (alias->ce_flags & CE_ADDED)
610 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
611
612 /* Ok, create the new entry using the name of the existing alias */
613 len = ce_namelen(alias);
614 new = xcalloc(1, cache_entry_size(len));
615 memcpy(new->name, alias->name, len);
616 copy_cache_entry(new, ce);
617 save_or_free_index_entry(istate, ce);
618 return new;
619 }
620
621 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
622 {
623 unsigned char sha1[20];
624 if (write_sha1_file("", 0, blob_type, sha1))
625 die("cannot create an empty blob in the object database");
626 hashcpy(ce->oid.hash, sha1);
627 }
628
629 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
630 {
631 int size, namelen, was_same;
632 mode_t st_mode = st->st_mode;
633 struct cache_entry *ce, *alias;
634 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
635 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
636 int pretend = flags & ADD_CACHE_PRETEND;
637 int intent_only = flags & ADD_CACHE_INTENT;
638 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
639 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
640
641 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
642 return error("%s: can only add regular files, symbolic links or git-directories", path);
643
644 namelen = strlen(path);
645 if (S_ISDIR(st_mode)) {
646 while (namelen && path[namelen-1] == '/')
647 namelen--;
648 }
649 size = cache_entry_size(namelen);
650 ce = xcalloc(1, size);
651 memcpy(ce->name, path, namelen);
652 ce->ce_namelen = namelen;
653 if (!intent_only)
654 fill_stat_cache_info(ce, st);
655 else
656 ce->ce_flags |= CE_INTENT_TO_ADD;
657
658
659 if (trust_executable_bit && has_symlinks) {
660 ce->ce_mode = create_ce_mode(st_mode);
661 } else {
662 /* If there is an existing entry, pick the mode bits and type
663 * from it, otherwise assume unexecutable regular file.
664 */
665 struct cache_entry *ent;
666 int pos = index_name_pos_also_unmerged(istate, path, namelen);
667
668 ent = (0 <= pos) ? istate->cache[pos] : NULL;
669 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
670 }
671
672 /* When core.ignorecase=true, determine if a directory of the same name but differing
673 * case already exists within the Git repository. If it does, ensure the directory
674 * case of the file being added to the repository matches (is folded into) the existing
675 * entry's directory case.
676 */
677 if (ignore_case) {
678 adjust_dirname_case(istate, ce->name);
679 }
680
681 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
682 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
683 /* Nothing changed, really */
684 if (!S_ISGITLINK(alias->ce_mode))
685 ce_mark_uptodate(alias);
686 alias->ce_flags |= CE_ADDED;
687
688 free(ce);
689 return 0;
690 }
691 if (!intent_only) {
692 if (index_path(ce->oid.hash, path, st, HASH_WRITE_OBJECT)) {
693 free(ce);
694 return error("unable to index file %s", path);
695 }
696 } else
697 set_object_name_for_intent_to_add_entry(ce);
698
699 if (ignore_case && alias && different_name(ce, alias))
700 ce = create_alias_ce(istate, ce, alias);
701 ce->ce_flags |= CE_ADDED;
702
703 /* It was suspected to be racily clean, but it turns out to be Ok */
704 was_same = (alias &&
705 !ce_stage(alias) &&
706 !oidcmp(&alias->oid, &ce->oid) &&
707 ce->ce_mode == alias->ce_mode);
708
709 if (pretend)
710 free(ce);
711 else if (add_index_entry(istate, ce, add_option)) {
712 free(ce);
713 return error("unable to add %s to index", path);
714 }
715 if (verbose && !was_same)
716 printf("add '%s'\n", path);
717 return 0;
718 }
719
720 int add_file_to_index(struct index_state *istate, const char *path, int flags)
721 {
722 struct stat st;
723 if (lstat(path, &st))
724 die_errno("unable to stat '%s'", path);
725 return add_to_index(istate, path, &st, flags);
726 }
727
728 struct cache_entry *make_cache_entry(unsigned int mode,
729 const unsigned char *sha1, const char *path, int stage,
730 unsigned int refresh_options)
731 {
732 int size, len;
733 struct cache_entry *ce, *ret;
734
735 if (!verify_path(path)) {
736 error("Invalid path '%s'", path);
737 return NULL;
738 }
739
740 len = strlen(path);
741 size = cache_entry_size(len);
742 ce = xcalloc(1, size);
743
744 hashcpy(ce->oid.hash, sha1);
745 memcpy(ce->name, path, len);
746 ce->ce_flags = create_ce_flags(stage);
747 ce->ce_namelen = len;
748 ce->ce_mode = create_ce_mode(mode);
749
750 ret = refresh_cache_entry(ce, refresh_options);
751 if (ret != ce)
752 free(ce);
753 return ret;
754 }
755
756 /*
757 * Chmod an index entry with either +x or -x.
758 *
759 * Returns -1 if the chmod for the particular cache entry failed (if it's
760 * not a regular file), -2 if an invalid flip argument is passed in, 0
761 * otherwise.
762 */
763 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
764 char flip)
765 {
766 if (!S_ISREG(ce->ce_mode))
767 return -1;
768 switch (flip) {
769 case '+':
770 ce->ce_mode |= 0111;
771 break;
772 case '-':
773 ce->ce_mode &= ~0111;
774 break;
775 default:
776 return -2;
777 }
778 cache_tree_invalidate_path(istate, ce->name);
779 ce->ce_flags |= CE_UPDATE_IN_BASE;
780 istate->cache_changed |= CE_ENTRY_CHANGED;
781
782 return 0;
783 }
784
785 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
786 {
787 int len = ce_namelen(a);
788 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
789 }
790
791 /*
792 * We fundamentally don't like some paths: we don't want
793 * dot or dot-dot anywhere, and for obvious reasons don't
794 * want to recurse into ".git" either.
795 *
796 * Also, we don't want double slashes or slashes at the
797 * end that can make pathnames ambiguous.
798 */
799 static int verify_dotfile(const char *rest)
800 {
801 /*
802 * The first character was '.', but that
803 * has already been discarded, we now test
804 * the rest.
805 */
806
807 /* "." is not allowed */
808 if (*rest == '\0' || is_dir_sep(*rest))
809 return 0;
810
811 switch (*rest) {
812 /*
813 * ".git" followed by NUL or slash is bad. This
814 * shares the path end test with the ".." case.
815 */
816 case 'g':
817 case 'G':
818 if (rest[1] != 'i' && rest[1] != 'I')
819 break;
820 if (rest[2] != 't' && rest[2] != 'T')
821 break;
822 rest += 2;
823 /* fallthrough */
824 case '.':
825 if (rest[1] == '\0' || is_dir_sep(rest[1]))
826 return 0;
827 }
828 return 1;
829 }
830
831 int verify_path(const char *path)
832 {
833 char c;
834
835 if (has_dos_drive_prefix(path))
836 return 0;
837
838 goto inside;
839 for (;;) {
840 if (!c)
841 return 1;
842 if (is_dir_sep(c)) {
843 inside:
844 if (protect_hfs && is_hfs_dotgit(path))
845 return 0;
846 if (protect_ntfs && is_ntfs_dotgit(path))
847 return 0;
848 c = *path++;
849 if ((c == '.' && !verify_dotfile(path)) ||
850 is_dir_sep(c) || c == '\0')
851 return 0;
852 }
853 c = *path++;
854 }
855 }
856
857 /*
858 * Do we have another file that has the beginning components being a
859 * proper superset of the name we're trying to add?
860 */
861 static int has_file_name(struct index_state *istate,
862 const struct cache_entry *ce, int pos, int ok_to_replace)
863 {
864 int retval = 0;
865 int len = ce_namelen(ce);
866 int stage = ce_stage(ce);
867 const char *name = ce->name;
868
869 while (pos < istate->cache_nr) {
870 struct cache_entry *p = istate->cache[pos++];
871
872 if (len >= ce_namelen(p))
873 break;
874 if (memcmp(name, p->name, len))
875 break;
876 if (ce_stage(p) != stage)
877 continue;
878 if (p->name[len] != '/')
879 continue;
880 if (p->ce_flags & CE_REMOVE)
881 continue;
882 retval = -1;
883 if (!ok_to_replace)
884 break;
885 remove_index_entry_at(istate, --pos);
886 }
887 return retval;
888 }
889
890
891 /*
892 * Like strcmp(), but also return the offset of the first change.
893 * If strings are equal, return the length.
894 */
895 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
896 {
897 size_t k;
898
899 if (!first_change)
900 return strcmp(s1, s2);
901
902 for (k = 0; s1[k] == s2[k]; k++)
903 if (s1[k] == '\0')
904 break;
905
906 *first_change = k;
907 return (unsigned char)s1[k] - (unsigned char)s2[k];
908 }
909
910 /*
911 * Do we have another file with a pathname that is a proper
912 * subset of the name we're trying to add?
913 *
914 * That is, is there another file in the index with a path
915 * that matches a sub-directory in the given entry?
916 */
917 static int has_dir_name(struct index_state *istate,
918 const struct cache_entry *ce, int pos, int ok_to_replace)
919 {
920 int retval = 0;
921 int stage = ce_stage(ce);
922 const char *name = ce->name;
923 const char *slash = name + ce_namelen(ce);
924 size_t len_eq_last;
925 int cmp_last = 0;
926
927 /*
928 * We are frequently called during an iteration on a sorted
929 * list of pathnames and while building a new index. Therefore,
930 * there is a high probability that this entry will eventually
931 * be appended to the index, rather than inserted in the middle.
932 * If we can confirm that, we can avoid binary searches on the
933 * components of the pathname.
934 *
935 * Compare the entry's full path with the last path in the index.
936 */
937 if (istate->cache_nr > 0) {
938 cmp_last = strcmp_offset(name,
939 istate->cache[istate->cache_nr - 1]->name,
940 &len_eq_last);
941 if (cmp_last > 0) {
942 if (len_eq_last == 0) {
943 /*
944 * The entry sorts AFTER the last one in the
945 * index and their paths have no common prefix,
946 * so there cannot be a F/D conflict.
947 */
948 return retval;
949 } else {
950 /*
951 * The entry sorts AFTER the last one in the
952 * index, but has a common prefix. Fall through
953 * to the loop below to disect the entry's path
954 * and see where the difference is.
955 */
956 }
957 } else if (cmp_last == 0) {
958 /*
959 * The entry exactly matches the last one in the
960 * index, but because of multiple stage and CE_REMOVE
961 * items, we fall through and let the regular search
962 * code handle it.
963 */
964 }
965 }
966
967 for (;;) {
968 size_t len;
969
970 for (;;) {
971 if (*--slash == '/')
972 break;
973 if (slash <= ce->name)
974 return retval;
975 }
976 len = slash - name;
977
978 if (cmp_last > 0) {
979 /*
980 * (len + 1) is a directory boundary (including
981 * the trailing slash). And since the loop is
982 * decrementing "slash", the first iteration is
983 * the longest directory prefix; subsequent
984 * iterations consider parent directories.
985 */
986
987 if (len + 1 <= len_eq_last) {
988 /*
989 * The directory prefix (including the trailing
990 * slash) also appears as a prefix in the last
991 * entry, so the remainder cannot collide (because
992 * strcmp said the whole path was greater).
993 *
994 * EQ: last: xxx/A
995 * this: xxx/B
996 *
997 * LT: last: xxx/file_A
998 * this: xxx/file_B
999 */
1000 return retval;
1001 }
1002
1003 if (len > len_eq_last) {
1004 /*
1005 * This part of the directory prefix (excluding
1006 * the trailing slash) is longer than the known
1007 * equal portions, so this sub-directory cannot
1008 * collide with a file.
1009 *
1010 * GT: last: xxxA
1011 * this: xxxB/file
1012 */
1013 return retval;
1014 }
1015
1016 if (istate->cache_nr > 0 &&
1017 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1018 /*
1019 * The directory prefix lines up with part of
1020 * a longer file or directory name, but sorts
1021 * after it, so this sub-directory cannot
1022 * collide with a file.
1023 *
1024 * last: xxx/yy-file (because '-' sorts before '/')
1025 * this: xxx/yy/abc
1026 */
1027 return retval;
1028 }
1029
1030 /*
1031 * This is a possible collision. Fall through and
1032 * let the regular search code handle it.
1033 *
1034 * last: xxx
1035 * this: xxx/file
1036 */
1037 }
1038
1039 pos = index_name_stage_pos(istate, name, len, stage);
1040 if (pos >= 0) {
1041 /*
1042 * Found one, but not so fast. This could
1043 * be a marker that says "I was here, but
1044 * I am being removed". Such an entry is
1045 * not a part of the resulting tree, and
1046 * it is Ok to have a directory at the same
1047 * path.
1048 */
1049 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1050 retval = -1;
1051 if (!ok_to_replace)
1052 break;
1053 remove_index_entry_at(istate, pos);
1054 continue;
1055 }
1056 }
1057 else
1058 pos = -pos-1;
1059
1060 /*
1061 * Trivial optimization: if we find an entry that
1062 * already matches the sub-directory, then we know
1063 * we're ok, and we can exit.
1064 */
1065 while (pos < istate->cache_nr) {
1066 struct cache_entry *p = istate->cache[pos];
1067 if ((ce_namelen(p) <= len) ||
1068 (p->name[len] != '/') ||
1069 memcmp(p->name, name, len))
1070 break; /* not our subdirectory */
1071 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1072 /*
1073 * p is at the same stage as our entry, and
1074 * is a subdirectory of what we are looking
1075 * at, so we cannot have conflicts at our
1076 * level or anything shorter.
1077 */
1078 return retval;
1079 pos++;
1080 }
1081 }
1082 return retval;
1083 }
1084
1085 /* We may be in a situation where we already have path/file and path
1086 * is being added, or we already have path and path/file is being
1087 * added. Either one would result in a nonsense tree that has path
1088 * twice when git-write-tree tries to write it out. Prevent it.
1089 *
1090 * If ok-to-replace is specified, we remove the conflicting entries
1091 * from the cache so the caller should recompute the insert position.
1092 * When this happens, we return non-zero.
1093 */
1094 static int check_file_directory_conflict(struct index_state *istate,
1095 const struct cache_entry *ce,
1096 int pos, int ok_to_replace)
1097 {
1098 int retval;
1099
1100 /*
1101 * When ce is an "I am going away" entry, we allow it to be added
1102 */
1103 if (ce->ce_flags & CE_REMOVE)
1104 return 0;
1105
1106 /*
1107 * We check if the path is a sub-path of a subsequent pathname
1108 * first, since removing those will not change the position
1109 * in the array.
1110 */
1111 retval = has_file_name(istate, ce, pos, ok_to_replace);
1112
1113 /*
1114 * Then check if the path might have a clashing sub-directory
1115 * before it.
1116 */
1117 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1118 }
1119
1120 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1121 {
1122 int pos;
1123 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1124 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1125 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1126 int new_only = option & ADD_CACHE_NEW_ONLY;
1127
1128 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1129 cache_tree_invalidate_path(istate, ce->name);
1130
1131 /*
1132 * If this entry's path sorts after the last entry in the index,
1133 * we can avoid searching for it.
1134 */
1135 if (istate->cache_nr > 0 &&
1136 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1137 pos = -istate->cache_nr - 1;
1138 else
1139 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1140
1141 /* existing match? Just replace it. */
1142 if (pos >= 0) {
1143 if (!new_only)
1144 replace_index_entry(istate, pos, ce);
1145 return 0;
1146 }
1147 pos = -pos-1;
1148
1149 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1150 untracked_cache_add_to_index(istate, ce->name);
1151
1152 /*
1153 * Inserting a merged entry ("stage 0") into the index
1154 * will always replace all non-merged entries..
1155 */
1156 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1157 while (ce_same_name(istate->cache[pos], ce)) {
1158 ok_to_add = 1;
1159 if (!remove_index_entry_at(istate, pos))
1160 break;
1161 }
1162 }
1163
1164 if (!ok_to_add)
1165 return -1;
1166 if (!verify_path(ce->name))
1167 return error("Invalid path '%s'", ce->name);
1168
1169 if (!skip_df_check &&
1170 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1171 if (!ok_to_replace)
1172 return error("'%s' appears as both a file and as a directory",
1173 ce->name);
1174 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1175 pos = -pos-1;
1176 }
1177 return pos + 1;
1178 }
1179
1180 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1181 {
1182 int pos;
1183
1184 if (option & ADD_CACHE_JUST_APPEND)
1185 pos = istate->cache_nr;
1186 else {
1187 int ret;
1188 ret = add_index_entry_with_check(istate, ce, option);
1189 if (ret <= 0)
1190 return ret;
1191 pos = ret - 1;
1192 }
1193
1194 /* Make sure the array is big enough .. */
1195 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1196
1197 /* Add it in.. */
1198 istate->cache_nr++;
1199 if (istate->cache_nr > pos + 1)
1200 memmove(istate->cache + pos + 1,
1201 istate->cache + pos,
1202 (istate->cache_nr - pos - 1) * sizeof(ce));
1203 set_index_entry(istate, pos, ce);
1204 istate->cache_changed |= CE_ENTRY_ADDED;
1205 return 0;
1206 }
1207
1208 /*
1209 * "refresh" does not calculate a new sha1 file or bring the
1210 * cache up-to-date for mode/content changes. But what it
1211 * _does_ do is to "re-match" the stat information of a file
1212 * with the cache, so that you can refresh the cache for a
1213 * file that hasn't been changed but where the stat entry is
1214 * out of date.
1215 *
1216 * For example, you'd want to do this after doing a "git-read-tree",
1217 * to link up the stat cache details with the proper files.
1218 */
1219 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1220 struct cache_entry *ce,
1221 unsigned int options, int *err,
1222 int *changed_ret)
1223 {
1224 struct stat st;
1225 struct cache_entry *updated;
1226 int changed, size;
1227 int refresh = options & CE_MATCH_REFRESH;
1228 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1229 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1230 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1231
1232 if (!refresh || ce_uptodate(ce))
1233 return ce;
1234
1235 /*
1236 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1237 * that the change to the work tree does not matter and told
1238 * us not to worry.
1239 */
1240 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1241 ce_mark_uptodate(ce);
1242 return ce;
1243 }
1244 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1245 ce_mark_uptodate(ce);
1246 return ce;
1247 }
1248
1249 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1250 if (ignore_missing)
1251 return ce;
1252 if (err)
1253 *err = ENOENT;
1254 return NULL;
1255 }
1256
1257 if (lstat(ce->name, &st) < 0) {
1258 if (ignore_missing && errno == ENOENT)
1259 return ce;
1260 if (err)
1261 *err = errno;
1262 return NULL;
1263 }
1264
1265 changed = ie_match_stat(istate, ce, &st, options);
1266 if (changed_ret)
1267 *changed_ret = changed;
1268 if (!changed) {
1269 /*
1270 * The path is unchanged. If we were told to ignore
1271 * valid bit, then we did the actual stat check and
1272 * found that the entry is unmodified. If the entry
1273 * is not marked VALID, this is the place to mark it
1274 * valid again, under "assume unchanged" mode.
1275 */
1276 if (ignore_valid && assume_unchanged &&
1277 !(ce->ce_flags & CE_VALID))
1278 ; /* mark this one VALID again */
1279 else {
1280 /*
1281 * We do not mark the index itself "modified"
1282 * because CE_UPTODATE flag is in-core only;
1283 * we are not going to write this change out.
1284 */
1285 if (!S_ISGITLINK(ce->ce_mode))
1286 ce_mark_uptodate(ce);
1287 return ce;
1288 }
1289 }
1290
1291 if (ie_modified(istate, ce, &st, options)) {
1292 if (err)
1293 *err = EINVAL;
1294 return NULL;
1295 }
1296
1297 size = ce_size(ce);
1298 updated = xmalloc(size);
1299 memcpy(updated, ce, size);
1300 fill_stat_cache_info(updated, &st);
1301 /*
1302 * If ignore_valid is not set, we should leave CE_VALID bit
1303 * alone. Otherwise, paths marked with --no-assume-unchanged
1304 * (i.e. things to be edited) will reacquire CE_VALID bit
1305 * automatically, which is not really what we want.
1306 */
1307 if (!ignore_valid && assume_unchanged &&
1308 !(ce->ce_flags & CE_VALID))
1309 updated->ce_flags &= ~CE_VALID;
1310
1311 /* istate->cache_changed is updated in the caller */
1312 return updated;
1313 }
1314
1315 static void show_file(const char * fmt, const char * name, int in_porcelain,
1316 int * first, const char *header_msg)
1317 {
1318 if (in_porcelain && *first && header_msg) {
1319 printf("%s\n", header_msg);
1320 *first = 0;
1321 }
1322 printf(fmt, name);
1323 }
1324
1325 int refresh_index(struct index_state *istate, unsigned int flags,
1326 const struct pathspec *pathspec,
1327 char *seen, const char *header_msg)
1328 {
1329 int i;
1330 int has_errors = 0;
1331 int really = (flags & REFRESH_REALLY) != 0;
1332 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1333 int quiet = (flags & REFRESH_QUIET) != 0;
1334 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1335 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1336 int first = 1;
1337 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1338 unsigned int options = (CE_MATCH_REFRESH |
1339 (really ? CE_MATCH_IGNORE_VALID : 0) |
1340 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1341 const char *modified_fmt;
1342 const char *deleted_fmt;
1343 const char *typechange_fmt;
1344 const char *added_fmt;
1345 const char *unmerged_fmt;
1346
1347 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1348 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1349 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1350 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1351 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1352 for (i = 0; i < istate->cache_nr; i++) {
1353 struct cache_entry *ce, *new;
1354 int cache_errno = 0;
1355 int changed = 0;
1356 int filtered = 0;
1357
1358 ce = istate->cache[i];
1359 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1360 continue;
1361
1362 if (pathspec && !ce_path_match(ce, pathspec, seen))
1363 filtered = 1;
1364
1365 if (ce_stage(ce)) {
1366 while ((i < istate->cache_nr) &&
1367 ! strcmp(istate->cache[i]->name, ce->name))
1368 i++;
1369 i--;
1370 if (allow_unmerged)
1371 continue;
1372 if (!filtered)
1373 show_file(unmerged_fmt, ce->name, in_porcelain,
1374 &first, header_msg);
1375 has_errors = 1;
1376 continue;
1377 }
1378
1379 if (filtered)
1380 continue;
1381
1382 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1383 if (new == ce)
1384 continue;
1385 if (!new) {
1386 const char *fmt;
1387
1388 if (really && cache_errno == EINVAL) {
1389 /* If we are doing --really-refresh that
1390 * means the index is not valid anymore.
1391 */
1392 ce->ce_flags &= ~CE_VALID;
1393 ce->ce_flags |= CE_UPDATE_IN_BASE;
1394 istate->cache_changed |= CE_ENTRY_CHANGED;
1395 }
1396 if (quiet)
1397 continue;
1398
1399 if (cache_errno == ENOENT)
1400 fmt = deleted_fmt;
1401 else if (ce_intent_to_add(ce))
1402 fmt = added_fmt; /* must be before other checks */
1403 else if (changed & TYPE_CHANGED)
1404 fmt = typechange_fmt;
1405 else
1406 fmt = modified_fmt;
1407 show_file(fmt,
1408 ce->name, in_porcelain, &first, header_msg);
1409 has_errors = 1;
1410 continue;
1411 }
1412
1413 replace_index_entry(istate, i, new);
1414 }
1415 return has_errors;
1416 }
1417
1418 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1419 unsigned int options)
1420 {
1421 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1422 }
1423
1424
1425 /*****************************************************************
1426 * Index File I/O
1427 *****************************************************************/
1428
1429 #define INDEX_FORMAT_DEFAULT 3
1430
1431 static unsigned int get_index_format_default(void)
1432 {
1433 char *envversion = getenv("GIT_INDEX_VERSION");
1434 char *endp;
1435 int value;
1436 unsigned int version = INDEX_FORMAT_DEFAULT;
1437
1438 if (!envversion) {
1439 if (!git_config_get_int("index.version", &value))
1440 version = value;
1441 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1442 warning(_("index.version set, but the value is invalid.\n"
1443 "Using version %i"), INDEX_FORMAT_DEFAULT);
1444 return INDEX_FORMAT_DEFAULT;
1445 }
1446 return version;
1447 }
1448
1449 version = strtoul(envversion, &endp, 10);
1450 if (*endp ||
1451 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1452 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1453 "Using version %i"), INDEX_FORMAT_DEFAULT);
1454 version = INDEX_FORMAT_DEFAULT;
1455 }
1456 return version;
1457 }
1458
1459 /*
1460 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1461 * Again - this is just a (very strong in practice) heuristic that
1462 * the inode hasn't changed.
1463 *
1464 * We save the fields in big-endian order to allow using the
1465 * index file over NFS transparently.
1466 */
1467 struct ondisk_cache_entry {
1468 struct cache_time ctime;
1469 struct cache_time mtime;
1470 uint32_t dev;
1471 uint32_t ino;
1472 uint32_t mode;
1473 uint32_t uid;
1474 uint32_t gid;
1475 uint32_t size;
1476 unsigned char sha1[20];
1477 uint16_t flags;
1478 char name[FLEX_ARRAY]; /* more */
1479 };
1480
1481 /*
1482 * This struct is used when CE_EXTENDED bit is 1
1483 * The struct must match ondisk_cache_entry exactly from
1484 * ctime till flags
1485 */
1486 struct ondisk_cache_entry_extended {
1487 struct cache_time ctime;
1488 struct cache_time mtime;
1489 uint32_t dev;
1490 uint32_t ino;
1491 uint32_t mode;
1492 uint32_t uid;
1493 uint32_t gid;
1494 uint32_t size;
1495 unsigned char sha1[20];
1496 uint16_t flags;
1497 uint16_t flags2;
1498 char name[FLEX_ARRAY]; /* more */
1499 };
1500
1501 /* These are only used for v3 or lower */
1502 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1503 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1504 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1505 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1506 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1507 ondisk_cache_entry_size(ce_namelen(ce)))
1508
1509 /* Allow fsck to force verification of the index checksum. */
1510 int verify_index_checksum;
1511
1512 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1513 {
1514 git_SHA_CTX c;
1515 unsigned char sha1[20];
1516 int hdr_version;
1517
1518 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1519 return error("bad signature");
1520 hdr_version = ntohl(hdr->hdr_version);
1521 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1522 return error("bad index version %d", hdr_version);
1523
1524 if (!verify_index_checksum)
1525 return 0;
1526
1527 git_SHA1_Init(&c);
1528 git_SHA1_Update(&c, hdr, size - 20);
1529 git_SHA1_Final(sha1, &c);
1530 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1531 return error("bad index file sha1 signature");
1532 return 0;
1533 }
1534
1535 static int read_index_extension(struct index_state *istate,
1536 const char *ext, void *data, unsigned long sz)
1537 {
1538 switch (CACHE_EXT(ext)) {
1539 case CACHE_EXT_TREE:
1540 istate->cache_tree = cache_tree_read(data, sz);
1541 break;
1542 case CACHE_EXT_RESOLVE_UNDO:
1543 istate->resolve_undo = resolve_undo_read(data, sz);
1544 break;
1545 case CACHE_EXT_LINK:
1546 if (read_link_extension(istate, data, sz))
1547 return -1;
1548 break;
1549 case CACHE_EXT_UNTRACKED:
1550 istate->untracked = read_untracked_extension(data, sz);
1551 break;
1552 default:
1553 if (*ext < 'A' || 'Z' < *ext)
1554 return error("index uses %.4s extension, which we do not understand",
1555 ext);
1556 fprintf(stderr, "ignoring %.4s extension\n", ext);
1557 break;
1558 }
1559 return 0;
1560 }
1561
1562 int hold_locked_index(struct lock_file *lk, int lock_flags)
1563 {
1564 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1565 }
1566
1567 int read_index(struct index_state *istate)
1568 {
1569 return read_index_from(istate, get_index_file());
1570 }
1571
1572 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1573 unsigned int flags,
1574 const char *name,
1575 size_t len)
1576 {
1577 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1578
1579 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1580 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1581 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1582 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1583 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1584 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1585 ce->ce_mode = get_be32(&ondisk->mode);
1586 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1587 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1588 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1589 ce->ce_flags = flags & ~CE_NAMEMASK;
1590 ce->ce_namelen = len;
1591 ce->index = 0;
1592 hashcpy(ce->oid.hash, ondisk->sha1);
1593 memcpy(ce->name, name, len);
1594 ce->name[len] = '\0';
1595 return ce;
1596 }
1597
1598 /*
1599 * Adjacent cache entries tend to share the leading paths, so it makes
1600 * sense to only store the differences in later entries. In the v4
1601 * on-disk format of the index, each on-disk cache entry stores the
1602 * number of bytes to be stripped from the end of the previous name,
1603 * and the bytes to append to the result, to come up with its name.
1604 */
1605 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1606 {
1607 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1608 size_t len = decode_varint(&cp);
1609
1610 if (name->len < len)
1611 die("malformed name field in the index");
1612 strbuf_remove(name, name->len - len, len);
1613 for (ep = cp; *ep; ep++)
1614 ; /* find the end */
1615 strbuf_add(name, cp, ep - cp);
1616 return (const char *)ep + 1 - cp_;
1617 }
1618
1619 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1620 unsigned long *ent_size,
1621 struct strbuf *previous_name)
1622 {
1623 struct cache_entry *ce;
1624 size_t len;
1625 const char *name;
1626 unsigned int flags;
1627
1628 /* On-disk flags are just 16 bits */
1629 flags = get_be16(&ondisk->flags);
1630 len = flags & CE_NAMEMASK;
1631
1632 if (flags & CE_EXTENDED) {
1633 struct ondisk_cache_entry_extended *ondisk2;
1634 int extended_flags;
1635 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1636 extended_flags = get_be16(&ondisk2->flags2) << 16;
1637 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1638 if (extended_flags & ~CE_EXTENDED_FLAGS)
1639 die("Unknown index entry format %08x", extended_flags);
1640 flags |= extended_flags;
1641 name = ondisk2->name;
1642 }
1643 else
1644 name = ondisk->name;
1645
1646 if (!previous_name) {
1647 /* v3 and earlier */
1648 if (len == CE_NAMEMASK)
1649 len = strlen(name);
1650 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1651
1652 *ent_size = ondisk_ce_size(ce);
1653 } else {
1654 unsigned long consumed;
1655 consumed = expand_name_field(previous_name, name);
1656 ce = cache_entry_from_ondisk(ondisk, flags,
1657 previous_name->buf,
1658 previous_name->len);
1659
1660 *ent_size = (name - ((char *)ondisk)) + consumed;
1661 }
1662 return ce;
1663 }
1664
1665 static void check_ce_order(struct index_state *istate)
1666 {
1667 unsigned int i;
1668
1669 for (i = 1; i < istate->cache_nr; i++) {
1670 struct cache_entry *ce = istate->cache[i - 1];
1671 struct cache_entry *next_ce = istate->cache[i];
1672 int name_compare = strcmp(ce->name, next_ce->name);
1673
1674 if (0 < name_compare)
1675 die("unordered stage entries in index");
1676 if (!name_compare) {
1677 if (!ce_stage(ce))
1678 die("multiple stage entries for merged file '%s'",
1679 ce->name);
1680 if (ce_stage(ce) > ce_stage(next_ce))
1681 die("unordered stage entries for '%s'",
1682 ce->name);
1683 }
1684 }
1685 }
1686
1687 static void tweak_untracked_cache(struct index_state *istate)
1688 {
1689 switch (git_config_get_untracked_cache()) {
1690 case -1: /* keep: do nothing */
1691 break;
1692 case 0: /* false */
1693 remove_untracked_cache(istate);
1694 break;
1695 case 1: /* true */
1696 add_untracked_cache(istate);
1697 break;
1698 default: /* unknown value: do nothing */
1699 break;
1700 }
1701 }
1702
1703 static void tweak_split_index(struct index_state *istate)
1704 {
1705 switch (git_config_get_split_index()) {
1706 case -1: /* unset: do nothing */
1707 break;
1708 case 0: /* false */
1709 remove_split_index(istate);
1710 break;
1711 case 1: /* true */
1712 add_split_index(istate);
1713 break;
1714 default: /* unknown value: do nothing */
1715 break;
1716 }
1717 }
1718
1719 static void post_read_index_from(struct index_state *istate)
1720 {
1721 check_ce_order(istate);
1722 tweak_untracked_cache(istate);
1723 tweak_split_index(istate);
1724 }
1725
1726 /* remember to discard_cache() before reading a different cache! */
1727 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1728 {
1729 int fd, i;
1730 struct stat st;
1731 unsigned long src_offset;
1732 struct cache_header *hdr;
1733 void *mmap;
1734 size_t mmap_size;
1735 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1736
1737 if (istate->initialized)
1738 return istate->cache_nr;
1739
1740 istate->timestamp.sec = 0;
1741 istate->timestamp.nsec = 0;
1742 fd = open(path, O_RDONLY);
1743 if (fd < 0) {
1744 if (!must_exist && errno == ENOENT)
1745 return 0;
1746 die_errno("%s: index file open failed", path);
1747 }
1748
1749 if (fstat(fd, &st))
1750 die_errno("cannot stat the open index");
1751
1752 mmap_size = xsize_t(st.st_size);
1753 if (mmap_size < sizeof(struct cache_header) + 20)
1754 die("index file smaller than expected");
1755
1756 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1757 if (mmap == MAP_FAILED)
1758 die_errno("unable to map index file");
1759 close(fd);
1760
1761 hdr = mmap;
1762 if (verify_hdr(hdr, mmap_size) < 0)
1763 goto unmap;
1764
1765 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1766 istate->version = ntohl(hdr->hdr_version);
1767 istate->cache_nr = ntohl(hdr->hdr_entries);
1768 istate->cache_alloc = alloc_nr(istate->cache_nr);
1769 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1770 istate->initialized = 1;
1771
1772 if (istate->version == 4)
1773 previous_name = &previous_name_buf;
1774 else
1775 previous_name = NULL;
1776
1777 src_offset = sizeof(*hdr);
1778 for (i = 0; i < istate->cache_nr; i++) {
1779 struct ondisk_cache_entry *disk_ce;
1780 struct cache_entry *ce;
1781 unsigned long consumed;
1782
1783 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1784 ce = create_from_disk(disk_ce, &consumed, previous_name);
1785 set_index_entry(istate, i, ce);
1786
1787 src_offset += consumed;
1788 }
1789 strbuf_release(&previous_name_buf);
1790 istate->timestamp.sec = st.st_mtime;
1791 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1792
1793 while (src_offset <= mmap_size - 20 - 8) {
1794 /* After an array of active_nr index entries,
1795 * there can be arbitrary number of extended
1796 * sections, each of which is prefixed with
1797 * extension name (4-byte) and section length
1798 * in 4-byte network byte order.
1799 */
1800 uint32_t extsize;
1801 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1802 extsize = ntohl(extsize);
1803 if (read_index_extension(istate,
1804 (const char *) mmap + src_offset,
1805 (char *) mmap + src_offset + 8,
1806 extsize) < 0)
1807 goto unmap;
1808 src_offset += 8;
1809 src_offset += extsize;
1810 }
1811 munmap(mmap, mmap_size);
1812 return istate->cache_nr;
1813
1814 unmap:
1815 munmap(mmap, mmap_size);
1816 die("index file corrupt");
1817 }
1818
1819 /*
1820 * Signal that the shared index is used by updating its mtime.
1821 *
1822 * This way, shared index can be removed if they have not been used
1823 * for some time.
1824 */
1825 static void freshen_shared_index(char *base_sha1_hex, int warn)
1826 {
1827 const char *shared_index = git_path("sharedindex.%s", base_sha1_hex);
1828 if (!check_and_freshen_file(shared_index, 1) && warn)
1829 warning("could not freshen shared index '%s'", shared_index);
1830 }
1831
1832 int read_index_from(struct index_state *istate, const char *path)
1833 {
1834 struct split_index *split_index;
1835 int ret;
1836 char *base_sha1_hex;
1837 const char *base_path;
1838
1839 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1840 if (istate->initialized)
1841 return istate->cache_nr;
1842
1843 ret = do_read_index(istate, path, 0);
1844
1845 split_index = istate->split_index;
1846 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1847 post_read_index_from(istate);
1848 return ret;
1849 }
1850
1851 if (split_index->base)
1852 discard_index(split_index->base);
1853 else
1854 split_index->base = xcalloc(1, sizeof(*split_index->base));
1855
1856 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1857 base_path = git_path("sharedindex.%s", base_sha1_hex);
1858 ret = do_read_index(split_index->base, base_path, 1);
1859 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1860 die("broken index, expect %s in %s, got %s",
1861 base_sha1_hex, base_path,
1862 sha1_to_hex(split_index->base->sha1));
1863
1864 freshen_shared_index(base_sha1_hex, 0);
1865 merge_base_index(istate);
1866 post_read_index_from(istate);
1867 return ret;
1868 }
1869
1870 int is_index_unborn(struct index_state *istate)
1871 {
1872 return (!istate->cache_nr && !istate->timestamp.sec);
1873 }
1874
1875 int discard_index(struct index_state *istate)
1876 {
1877 int i;
1878
1879 for (i = 0; i < istate->cache_nr; i++) {
1880 if (istate->cache[i]->index &&
1881 istate->split_index &&
1882 istate->split_index->base &&
1883 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1884 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1885 continue;
1886 free(istate->cache[i]);
1887 }
1888 resolve_undo_clear_index(istate);
1889 istate->cache_nr = 0;
1890 istate->cache_changed = 0;
1891 istate->timestamp.sec = 0;
1892 istate->timestamp.nsec = 0;
1893 free_name_hash(istate);
1894 cache_tree_free(&(istate->cache_tree));
1895 istate->initialized = 0;
1896 free(istate->cache);
1897 istate->cache = NULL;
1898 istate->cache_alloc = 0;
1899 discard_split_index(istate);
1900 free_untracked_cache(istate->untracked);
1901 istate->untracked = NULL;
1902 return 0;
1903 }
1904
1905 int unmerged_index(const struct index_state *istate)
1906 {
1907 int i;
1908 for (i = 0; i < istate->cache_nr; i++) {
1909 if (ce_stage(istate->cache[i]))
1910 return 1;
1911 }
1912 return 0;
1913 }
1914
1915 #define WRITE_BUFFER_SIZE 8192
1916 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1917 static unsigned long write_buffer_len;
1918
1919 static int ce_write_flush(git_SHA_CTX *context, int fd)
1920 {
1921 unsigned int buffered = write_buffer_len;
1922 if (buffered) {
1923 git_SHA1_Update(context, write_buffer, buffered);
1924 if (write_in_full(fd, write_buffer, buffered) != buffered)
1925 return -1;
1926 write_buffer_len = 0;
1927 }
1928 return 0;
1929 }
1930
1931 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1932 {
1933 while (len) {
1934 unsigned int buffered = write_buffer_len;
1935 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1936 if (partial > len)
1937 partial = len;
1938 memcpy(write_buffer + buffered, data, partial);
1939 buffered += partial;
1940 if (buffered == WRITE_BUFFER_SIZE) {
1941 write_buffer_len = buffered;
1942 if (ce_write_flush(context, fd))
1943 return -1;
1944 buffered = 0;
1945 }
1946 write_buffer_len = buffered;
1947 len -= partial;
1948 data = (char *) data + partial;
1949 }
1950 return 0;
1951 }
1952
1953 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1954 unsigned int ext, unsigned int sz)
1955 {
1956 ext = htonl(ext);
1957 sz = htonl(sz);
1958 return ((ce_write(context, fd, &ext, 4) < 0) ||
1959 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1960 }
1961
1962 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1963 {
1964 unsigned int left = write_buffer_len;
1965
1966 if (left) {
1967 write_buffer_len = 0;
1968 git_SHA1_Update(context, write_buffer, left);
1969 }
1970
1971 /* Flush first if not enough space for SHA1 signature */
1972 if (left + 20 > WRITE_BUFFER_SIZE) {
1973 if (write_in_full(fd, write_buffer, left) != left)
1974 return -1;
1975 left = 0;
1976 }
1977
1978 /* Append the SHA1 signature at the end */
1979 git_SHA1_Final(write_buffer + left, context);
1980 hashcpy(sha1, write_buffer + left);
1981 left += 20;
1982 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1983 }
1984
1985 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1986 {
1987 /*
1988 * The only thing we care about in this function is to smudge the
1989 * falsely clean entry due to touch-update-touch race, so we leave
1990 * everything else as they are. We are called for entries whose
1991 * ce_stat_data.sd_mtime match the index file mtime.
1992 *
1993 * Note that this actually does not do much for gitlinks, for
1994 * which ce_match_stat_basic() always goes to the actual
1995 * contents. The caller checks with is_racy_timestamp() which
1996 * always says "no" for gitlinks, so we are not called for them ;-)
1997 */
1998 struct stat st;
1999
2000 if (lstat(ce->name, &st) < 0)
2001 return;
2002 if (ce_match_stat_basic(ce, &st))
2003 return;
2004 if (ce_modified_check_fs(ce, &st)) {
2005 /* This is "racily clean"; smudge it. Note that this
2006 * is a tricky code. At first glance, it may appear
2007 * that it can break with this sequence:
2008 *
2009 * $ echo xyzzy >frotz
2010 * $ git-update-index --add frotz
2011 * $ : >frotz
2012 * $ sleep 3
2013 * $ echo filfre >nitfol
2014 * $ git-update-index --add nitfol
2015 *
2016 * but it does not. When the second update-index runs,
2017 * it notices that the entry "frotz" has the same timestamp
2018 * as index, and if we were to smudge it by resetting its
2019 * size to zero here, then the object name recorded
2020 * in index is the 6-byte file but the cached stat information
2021 * becomes zero --- which would then match what we would
2022 * obtain from the filesystem next time we stat("frotz").
2023 *
2024 * However, the second update-index, before calling
2025 * this function, notices that the cached size is 6
2026 * bytes and what is on the filesystem is an empty
2027 * file, and never calls us, so the cached size information
2028 * for "frotz" stays 6 which does not match the filesystem.
2029 */
2030 ce->ce_stat_data.sd_size = 0;
2031 }
2032 }
2033
2034 /* Copy miscellaneous fields but not the name */
2035 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2036 struct cache_entry *ce)
2037 {
2038 short flags;
2039
2040 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2041 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2042 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2043 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2044 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2045 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2046 ondisk->mode = htonl(ce->ce_mode);
2047 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2048 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2049 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2050 hashcpy(ondisk->sha1, ce->oid.hash);
2051
2052 flags = ce->ce_flags & ~CE_NAMEMASK;
2053 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2054 ondisk->flags = htons(flags);
2055 if (ce->ce_flags & CE_EXTENDED) {
2056 struct ondisk_cache_entry_extended *ondisk2;
2057 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2058 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2059 return ondisk2->name;
2060 }
2061 else {
2062 return ondisk->name;
2063 }
2064 }
2065
2066 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
2067 struct strbuf *previous_name)
2068 {
2069 int size;
2070 struct ondisk_cache_entry *ondisk;
2071 int saved_namelen = saved_namelen; /* compiler workaround */
2072 char *name;
2073 int result;
2074
2075 if (ce->ce_flags & CE_STRIP_NAME) {
2076 saved_namelen = ce_namelen(ce);
2077 ce->ce_namelen = 0;
2078 }
2079
2080 if (!previous_name) {
2081 size = ondisk_ce_size(ce);
2082 ondisk = xcalloc(1, size);
2083 name = copy_cache_entry_to_ondisk(ondisk, ce);
2084 memcpy(name, ce->name, ce_namelen(ce));
2085 } else {
2086 int common, to_remove, prefix_size;
2087 unsigned char to_remove_vi[16];
2088 for (common = 0;
2089 (ce->name[common] &&
2090 common < previous_name->len &&
2091 ce->name[common] == previous_name->buf[common]);
2092 common++)
2093 ; /* still matching */
2094 to_remove = previous_name->len - common;
2095 prefix_size = encode_varint(to_remove, to_remove_vi);
2096
2097 if (ce->ce_flags & CE_EXTENDED)
2098 size = offsetof(struct ondisk_cache_entry_extended, name);
2099 else
2100 size = offsetof(struct ondisk_cache_entry, name);
2101 size += prefix_size + (ce_namelen(ce) - common + 1);
2102
2103 ondisk = xcalloc(1, size);
2104 name = copy_cache_entry_to_ondisk(ondisk, ce);
2105 memcpy(name, to_remove_vi, prefix_size);
2106 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
2107
2108 strbuf_splice(previous_name, common, to_remove,
2109 ce->name + common, ce_namelen(ce) - common);
2110 }
2111 if (ce->ce_flags & CE_STRIP_NAME) {
2112 ce->ce_namelen = saved_namelen;
2113 ce->ce_flags &= ~CE_STRIP_NAME;
2114 }
2115
2116 result = ce_write(c, fd, ondisk, size);
2117 free(ondisk);
2118 return result;
2119 }
2120
2121 /*
2122 * This function verifies if index_state has the correct sha1 of the
2123 * index file. Don't die if we have any other failure, just return 0.
2124 */
2125 static int verify_index_from(const struct index_state *istate, const char *path)
2126 {
2127 int fd;
2128 ssize_t n;
2129 struct stat st;
2130 unsigned char sha1[20];
2131
2132 if (!istate->initialized)
2133 return 0;
2134
2135 fd = open(path, O_RDONLY);
2136 if (fd < 0)
2137 return 0;
2138
2139 if (fstat(fd, &st))
2140 goto out;
2141
2142 if (st.st_size < sizeof(struct cache_header) + 20)
2143 goto out;
2144
2145 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2146 if (n != 20)
2147 goto out;
2148
2149 if (hashcmp(istate->sha1, sha1))
2150 goto out;
2151
2152 close(fd);
2153 return 1;
2154
2155 out:
2156 close(fd);
2157 return 0;
2158 }
2159
2160 static int verify_index(const struct index_state *istate)
2161 {
2162 return verify_index_from(istate, get_index_file());
2163 }
2164
2165 static int has_racy_timestamp(struct index_state *istate)
2166 {
2167 int entries = istate->cache_nr;
2168 int i;
2169
2170 for (i = 0; i < entries; i++) {
2171 struct cache_entry *ce = istate->cache[i];
2172 if (is_racy_timestamp(istate, ce))
2173 return 1;
2174 }
2175 return 0;
2176 }
2177
2178 /*
2179 * Opportunistically update the index but do not complain if we can't
2180 */
2181 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2182 {
2183 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2184 verify_index(istate) &&
2185 write_locked_index(istate, lockfile, COMMIT_LOCK))
2186 rollback_lock_file(lockfile);
2187 }
2188
2189 static int do_write_index(struct index_state *istate, int newfd,
2190 int strip_extensions)
2191 {
2192 git_SHA_CTX c;
2193 struct cache_header hdr;
2194 int i, err, removed, extended, hdr_version;
2195 struct cache_entry **cache = istate->cache;
2196 int entries = istate->cache_nr;
2197 struct stat st;
2198 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2199
2200 for (i = removed = extended = 0; i < entries; i++) {
2201 if (cache[i]->ce_flags & CE_REMOVE)
2202 removed++;
2203
2204 /* reduce extended entries if possible */
2205 cache[i]->ce_flags &= ~CE_EXTENDED;
2206 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2207 extended++;
2208 cache[i]->ce_flags |= CE_EXTENDED;
2209 }
2210 }
2211
2212 if (!istate->version) {
2213 istate->version = get_index_format_default();
2214 if (getenv("GIT_TEST_SPLIT_INDEX"))
2215 init_split_index(istate);
2216 }
2217
2218 /* demote version 3 to version 2 when the latter suffices */
2219 if (istate->version == 3 || istate->version == 2)
2220 istate->version = extended ? 3 : 2;
2221
2222 hdr_version = istate->version;
2223
2224 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2225 hdr.hdr_version = htonl(hdr_version);
2226 hdr.hdr_entries = htonl(entries - removed);
2227
2228 git_SHA1_Init(&c);
2229 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2230 return -1;
2231
2232 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2233 for (i = 0; i < entries; i++) {
2234 struct cache_entry *ce = cache[i];
2235 if (ce->ce_flags & CE_REMOVE)
2236 continue;
2237 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2238 ce_smudge_racily_clean_entry(ce);
2239 if (is_null_oid(&ce->oid)) {
2240 static const char msg[] = "cache entry has null sha1: %s";
2241 static int allow = -1;
2242
2243 if (allow < 0)
2244 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2245 if (allow)
2246 warning(msg, ce->name);
2247 else
2248 return error(msg, ce->name);
2249 }
2250 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2251 return -1;
2252 }
2253 strbuf_release(&previous_name_buf);
2254
2255 /* Write extension data here */
2256 if (!strip_extensions && istate->split_index) {
2257 struct strbuf sb = STRBUF_INIT;
2258
2259 err = write_link_extension(&sb, istate) < 0 ||
2260 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2261 sb.len) < 0 ||
2262 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2263 strbuf_release(&sb);
2264 if (err)
2265 return -1;
2266 }
2267 if (!strip_extensions && istate->cache_tree) {
2268 struct strbuf sb = STRBUF_INIT;
2269
2270 cache_tree_write(&sb, istate->cache_tree);
2271 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2272 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2273 strbuf_release(&sb);
2274 if (err)
2275 return -1;
2276 }
2277 if (!strip_extensions && istate->resolve_undo) {
2278 struct strbuf sb = STRBUF_INIT;
2279
2280 resolve_undo_write(&sb, istate->resolve_undo);
2281 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2282 sb.len) < 0
2283 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2284 strbuf_release(&sb);
2285 if (err)
2286 return -1;
2287 }
2288 if (!strip_extensions && istate->untracked) {
2289 struct strbuf sb = STRBUF_INIT;
2290
2291 write_untracked_extension(&sb, istate->untracked);
2292 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2293 sb.len) < 0 ||
2294 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2295 strbuf_release(&sb);
2296 if (err)
2297 return -1;
2298 }
2299
2300 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2301 return -1;
2302 istate->timestamp.sec = (unsigned int)st.st_mtime;
2303 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2304 return 0;
2305 }
2306
2307 void set_alternate_index_output(const char *name)
2308 {
2309 alternate_index_output = name;
2310 }
2311
2312 static int commit_locked_index(struct lock_file *lk)
2313 {
2314 if (alternate_index_output)
2315 return commit_lock_file_to(lk, alternate_index_output);
2316 else
2317 return commit_lock_file(lk);
2318 }
2319
2320 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2321 unsigned flags)
2322 {
2323 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2324 if (ret)
2325 return ret;
2326 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2327 (COMMIT_LOCK | CLOSE_LOCK));
2328 if (flags & COMMIT_LOCK)
2329 return commit_locked_index(lock);
2330 else if (flags & CLOSE_LOCK)
2331 return close_lock_file(lock);
2332 else
2333 return ret;
2334 }
2335
2336 static int write_split_index(struct index_state *istate,
2337 struct lock_file *lock,
2338 unsigned flags)
2339 {
2340 int ret;
2341 prepare_to_write_split_index(istate);
2342 ret = do_write_locked_index(istate, lock, flags);
2343 finish_writing_split_index(istate);
2344 return ret;
2345 }
2346
2347 static const char *shared_index_expire = "2.weeks.ago";
2348
2349 static unsigned long get_shared_index_expire_date(void)
2350 {
2351 static unsigned long shared_index_expire_date;
2352 static int shared_index_expire_date_prepared;
2353
2354 if (!shared_index_expire_date_prepared) {
2355 git_config_get_expiry("splitindex.sharedindexexpire",
2356 &shared_index_expire);
2357 shared_index_expire_date = approxidate(shared_index_expire);
2358 shared_index_expire_date_prepared = 1;
2359 }
2360
2361 return shared_index_expire_date;
2362 }
2363
2364 static int should_delete_shared_index(const char *shared_index_path)
2365 {
2366 struct stat st;
2367 unsigned long expiration;
2368
2369 /* Check timestamp */
2370 expiration = get_shared_index_expire_date();
2371 if (!expiration)
2372 return 0;
2373 if (stat(shared_index_path, &st))
2374 return error_errno(_("could not stat '%s"), shared_index_path);
2375 if (st.st_mtime > expiration)
2376 return 0;
2377
2378 return 1;
2379 }
2380
2381 static int clean_shared_index_files(const char *current_hex)
2382 {
2383 struct dirent *de;
2384 DIR *dir = opendir(get_git_dir());
2385
2386 if (!dir)
2387 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2388
2389 while ((de = readdir(dir)) != NULL) {
2390 const char *sha1_hex;
2391 const char *shared_index_path;
2392 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2393 continue;
2394 if (!strcmp(sha1_hex, current_hex))
2395 continue;
2396 shared_index_path = git_path("%s", de->d_name);
2397 if (should_delete_shared_index(shared_index_path) > 0 &&
2398 unlink(shared_index_path))
2399 warning_errno(_("unable to unlink: %s"), shared_index_path);
2400 }
2401 closedir(dir);
2402
2403 return 0;
2404 }
2405
2406 static struct tempfile temporary_sharedindex;
2407
2408 static int write_shared_index(struct index_state *istate,
2409 struct lock_file *lock, unsigned flags)
2410 {
2411 struct split_index *si = istate->split_index;
2412 int fd, ret;
2413
2414 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2415 if (fd < 0) {
2416 hashclr(si->base_sha1);
2417 return do_write_locked_index(istate, lock, flags);
2418 }
2419 move_cache_to_base_index(istate);
2420 ret = do_write_index(si->base, fd, 1);
2421 if (ret) {
2422 delete_tempfile(&temporary_sharedindex);
2423 return ret;
2424 }
2425 ret = rename_tempfile(&temporary_sharedindex,
2426 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2427 if (!ret) {
2428 hashcpy(si->base_sha1, si->base->sha1);
2429 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2430 }
2431
2432 return ret;
2433 }
2434
2435 static const int default_max_percent_split_change = 20;
2436
2437 static int too_many_not_shared_entries(struct index_state *istate)
2438 {
2439 int i, not_shared = 0;
2440 int max_split = git_config_get_max_percent_split_change();
2441
2442 switch (max_split) {
2443 case -1:
2444 /* not or badly configured: use the default value */
2445 max_split = default_max_percent_split_change;
2446 break;
2447 case 0:
2448 return 1; /* 0% means always write a new shared index */
2449 case 100:
2450 return 0; /* 100% means never write a new shared index */
2451 default:
2452 break; /* just use the configured value */
2453 }
2454
2455 /* Count not shared entries */
2456 for (i = 0; i < istate->cache_nr; i++) {
2457 struct cache_entry *ce = istate->cache[i];
2458 if (!ce->index)
2459 not_shared++;
2460 }
2461
2462 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2463 }
2464
2465 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2466 unsigned flags)
2467 {
2468 int new_shared_index, ret;
2469 struct split_index *si = istate->split_index;
2470
2471 if (!si || alternate_index_output ||
2472 (istate->cache_changed & ~EXTMASK)) {
2473 if (si)
2474 hashclr(si->base_sha1);
2475 return do_write_locked_index(istate, lock, flags);
2476 }
2477
2478 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2479 int v = si->base_sha1[0];
2480 if ((v & 15) < 6)
2481 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2482 }
2483 if (too_many_not_shared_entries(istate))
2484 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2485
2486 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2487
2488 if (new_shared_index) {
2489 ret = write_shared_index(istate, lock, flags);
2490 if (ret)
2491 return ret;
2492 }
2493
2494 ret = write_split_index(istate, lock, flags);
2495
2496 /* Freshen the shared index only if the split-index was written */
2497 if (!ret && !new_shared_index)
2498 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2499
2500 return ret;
2501 }
2502
2503 /*
2504 * Read the index file that is potentially unmerged into given
2505 * index_state, dropping any unmerged entries. Returns true if
2506 * the index is unmerged. Callers who want to refuse to work
2507 * from an unmerged state can call this and check its return value,
2508 * instead of calling read_cache().
2509 */
2510 int read_index_unmerged(struct index_state *istate)
2511 {
2512 int i;
2513 int unmerged = 0;
2514
2515 read_index(istate);
2516 for (i = 0; i < istate->cache_nr; i++) {
2517 struct cache_entry *ce = istate->cache[i];
2518 struct cache_entry *new_ce;
2519 int size, len;
2520
2521 if (!ce_stage(ce))
2522 continue;
2523 unmerged = 1;
2524 len = ce_namelen(ce);
2525 size = cache_entry_size(len);
2526 new_ce = xcalloc(1, size);
2527 memcpy(new_ce->name, ce->name, len);
2528 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2529 new_ce->ce_namelen = len;
2530 new_ce->ce_mode = ce->ce_mode;
2531 if (add_index_entry(istate, new_ce, 0))
2532 return error("%s: cannot drop to stage #0",
2533 new_ce->name);
2534 }
2535 return unmerged;
2536 }
2537
2538 /*
2539 * Returns 1 if the path is an "other" path with respect to
2540 * the index; that is, the path is not mentioned in the index at all,
2541 * either as a file, a directory with some files in the index,
2542 * or as an unmerged entry.
2543 *
2544 * We helpfully remove a trailing "/" from directories so that
2545 * the output of read_directory can be used as-is.
2546 */
2547 int index_name_is_other(const struct index_state *istate, const char *name,
2548 int namelen)
2549 {
2550 int pos;
2551 if (namelen && name[namelen - 1] == '/')
2552 namelen--;
2553 pos = index_name_pos(istate, name, namelen);
2554 if (0 <= pos)
2555 return 0; /* exact match */
2556 pos = -pos - 1;
2557 if (pos < istate->cache_nr) {
2558 struct cache_entry *ce = istate->cache[pos];
2559 if (ce_namelen(ce) == namelen &&
2560 !memcmp(ce->name, name, namelen))
2561 return 0; /* Yup, this one exists unmerged */
2562 }
2563 return 1;
2564 }
2565
2566 void *read_blob_data_from_index(const struct index_state *istate,
2567 const char *path, unsigned long *size)
2568 {
2569 int pos, len;
2570 unsigned long sz;
2571 enum object_type type;
2572 void *data;
2573
2574 len = strlen(path);
2575 pos = index_name_pos(istate, path, len);
2576 if (pos < 0) {
2577 /*
2578 * We might be in the middle of a merge, in which
2579 * case we would read stage #2 (ours).
2580 */
2581 int i;
2582 for (i = -pos - 1;
2583 (pos < 0 && i < istate->cache_nr &&
2584 !strcmp(istate->cache[i]->name, path));
2585 i++)
2586 if (ce_stage(istate->cache[i]) == 2)
2587 pos = i;
2588 }
2589 if (pos < 0)
2590 return NULL;
2591 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2592 if (!data || type != OBJ_BLOB) {
2593 free(data);
2594 return NULL;
2595 }
2596 if (size)
2597 *size = sz;
2598 return data;
2599 }
2600
2601 void stat_validity_clear(struct stat_validity *sv)
2602 {
2603 free(sv->sd);
2604 sv->sd = NULL;
2605 }
2606
2607 int stat_validity_check(struct stat_validity *sv, const char *path)
2608 {
2609 struct stat st;
2610
2611 if (stat(path, &st) < 0)
2612 return sv->sd == NULL;
2613 if (!sv->sd)
2614 return 0;
2615 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2616 }
2617
2618 void stat_validity_update(struct stat_validity *sv, int fd)
2619 {
2620 struct stat st;
2621
2622 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2623 stat_validity_clear(sv);
2624 else {
2625 if (!sv->sd)
2626 sv->sd = xcalloc(1, sizeof(struct stat_data));
2627 fill_stat_data(sv->sd, &st);
2628 }
2629 }