Merge branch 'bc/object-id'
[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 "tempfile.h"
10 #include "lockfile.h"
11 #include "cache-tree.h"
12 #include "refs.h"
13 #include "dir.h"
14 #include "tree.h"
15 #include "commit.h"
16 #include "blob.h"
17 #include "resolve-undo.h"
18 #include "strbuf.h"
19 #include "varint.h"
20 #include "split-index.h"
21 #include "utf8.h"
22
23 /* Mask for the name length in ce_flags in the on-disk index */
24
25 #define CE_NAMEMASK (0x0fff)
26
27 /* Index extensions.
28 *
29 * The first letter should be 'A'..'Z' for extensions that are not
30 * necessary for a correct operation (i.e. optimization data).
31 * When new extensions are added that _needs_ to be understood in
32 * order to correctly interpret the index file, pick character that
33 * is outside the range, to cause the reader to abort.
34 */
35
36 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
37 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
38 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
39 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
40 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
41
42 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
43 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
44 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
45 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
46
47 struct index_state the_index;
48 static const char *alternate_index_output;
49
50 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
51 {
52 istate->cache[nr] = ce;
53 add_name_hash(istate, ce);
54 }
55
56 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
57 {
58 struct cache_entry *old = istate->cache[nr];
59
60 replace_index_entry_in_base(istate, old, ce);
61 remove_name_hash(istate, old);
62 free(old);
63 set_index_entry(istate, nr, ce);
64 ce->ce_flags |= CE_UPDATE_IN_BASE;
65 istate->cache_changed |= CE_ENTRY_CHANGED;
66 }
67
68 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
69 {
70 struct cache_entry *old = istate->cache[nr], *new;
71 int namelen = strlen(new_name);
72
73 new = xmalloc(cache_entry_size(namelen));
74 copy_cache_entry(new, old);
75 new->ce_flags &= ~CE_HASHED;
76 new->ce_namelen = namelen;
77 new->index = 0;
78 memcpy(new->name, new_name, namelen + 1);
79
80 cache_tree_invalidate_path(istate, old->name);
81 untracked_cache_remove_from_index(istate, old->name);
82 remove_index_entry_at(istate, nr);
83 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
84 }
85
86 void fill_stat_data(struct stat_data *sd, struct stat *st)
87 {
88 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
89 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
90 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
91 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
92 sd->sd_dev = st->st_dev;
93 sd->sd_ino = st->st_ino;
94 sd->sd_uid = st->st_uid;
95 sd->sd_gid = st->st_gid;
96 sd->sd_size = st->st_size;
97 }
98
99 int match_stat_data(const struct stat_data *sd, struct stat *st)
100 {
101 int changed = 0;
102
103 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
104 changed |= MTIME_CHANGED;
105 if (trust_ctime && check_stat &&
106 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
107 changed |= CTIME_CHANGED;
108
109 #ifdef USE_NSEC
110 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
111 changed |= MTIME_CHANGED;
112 if (trust_ctime && check_stat &&
113 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
114 changed |= CTIME_CHANGED;
115 #endif
116
117 if (check_stat) {
118 if (sd->sd_uid != (unsigned int) st->st_uid ||
119 sd->sd_gid != (unsigned int) st->st_gid)
120 changed |= OWNER_CHANGED;
121 if (sd->sd_ino != (unsigned int) st->st_ino)
122 changed |= INODE_CHANGED;
123 }
124
125 #ifdef USE_STDEV
126 /*
127 * st_dev breaks on network filesystems where different
128 * clients will have different views of what "device"
129 * the filesystem is on
130 */
131 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
132 changed |= INODE_CHANGED;
133 #endif
134
135 if (sd->sd_size != (unsigned int) st->st_size)
136 changed |= DATA_CHANGED;
137
138 return changed;
139 }
140
141 /*
142 * This only updates the "non-critical" parts of the directory
143 * cache, ie the parts that aren't tracked by GIT, and only used
144 * to validate the cache.
145 */
146 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
147 {
148 fill_stat_data(&ce->ce_stat_data, st);
149
150 if (assume_unchanged)
151 ce->ce_flags |= CE_VALID;
152
153 if (S_ISREG(st->st_mode))
154 ce_mark_uptodate(ce);
155 }
156
157 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
158 {
159 int match = -1;
160 int fd = git_open_cloexec(ce->name, O_RDONLY);
161
162 if (fd >= 0) {
163 struct object_id oid;
164 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
165 match = oidcmp(&oid, &ce->oid);
166 /* index_fd() closed the file descriptor already */
167 }
168 return match;
169 }
170
171 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
172 {
173 int match = -1;
174 void *buffer;
175 unsigned long size;
176 enum object_type type;
177 struct strbuf sb = STRBUF_INIT;
178
179 if (strbuf_readlink(&sb, ce->name, expected_size))
180 return -1;
181
182 buffer = read_sha1_file(ce->oid.hash, &type, &size);
183 if (buffer) {
184 if (size == sb.len)
185 match = memcmp(buffer, sb.buf, size);
186 free(buffer);
187 }
188 strbuf_release(&sb);
189 return match;
190 }
191
192 static int ce_compare_gitlink(const struct cache_entry *ce)
193 {
194 struct object_id oid;
195
196 /*
197 * We don't actually require that the .git directory
198 * under GITLINK directory be a valid git directory. It
199 * might even be missing (in case nobody populated that
200 * sub-project).
201 *
202 * If so, we consider it always to match.
203 */
204 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
205 return 0;
206 return oidcmp(&oid, &ce->oid);
207 }
208
209 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
210 {
211 switch (st->st_mode & S_IFMT) {
212 case S_IFREG:
213 if (ce_compare_data(ce, st))
214 return DATA_CHANGED;
215 break;
216 case S_IFLNK:
217 if (ce_compare_link(ce, xsize_t(st->st_size)))
218 return DATA_CHANGED;
219 break;
220 case S_IFDIR:
221 if (S_ISGITLINK(ce->ce_mode))
222 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
223 /* else fallthrough */
224 default:
225 return TYPE_CHANGED;
226 }
227 return 0;
228 }
229
230 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
231 {
232 unsigned int changed = 0;
233
234 if (ce->ce_flags & CE_REMOVE)
235 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
236
237 switch (ce->ce_mode & S_IFMT) {
238 case S_IFREG:
239 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
240 /* We consider only the owner x bit to be relevant for
241 * "mode changes"
242 */
243 if (trust_executable_bit &&
244 (0100 & (ce->ce_mode ^ st->st_mode)))
245 changed |= MODE_CHANGED;
246 break;
247 case S_IFLNK:
248 if (!S_ISLNK(st->st_mode) &&
249 (has_symlinks || !S_ISREG(st->st_mode)))
250 changed |= TYPE_CHANGED;
251 break;
252 case S_IFGITLINK:
253 /* We ignore most of the st_xxx fields for gitlinks */
254 if (!S_ISDIR(st->st_mode))
255 changed |= TYPE_CHANGED;
256 else if (ce_compare_gitlink(ce))
257 changed |= DATA_CHANGED;
258 return changed;
259 default:
260 die("internal error: ce_mode is %o", ce->ce_mode);
261 }
262
263 changed |= match_stat_data(&ce->ce_stat_data, st);
264
265 /* Racily smudged entry? */
266 if (!ce->ce_stat_data.sd_size) {
267 if (!is_empty_blob_sha1(ce->oid.hash))
268 changed |= DATA_CHANGED;
269 }
270
271 return changed;
272 }
273
274 static int is_racy_stat(const struct index_state *istate,
275 const struct stat_data *sd)
276 {
277 return (istate->timestamp.sec &&
278 #ifdef USE_NSEC
279 /* nanosecond timestamped files can also be racy! */
280 (istate->timestamp.sec < sd->sd_mtime.sec ||
281 (istate->timestamp.sec == sd->sd_mtime.sec &&
282 istate->timestamp.nsec <= sd->sd_mtime.nsec))
283 #else
284 istate->timestamp.sec <= sd->sd_mtime.sec
285 #endif
286 );
287 }
288
289 static int is_racy_timestamp(const struct index_state *istate,
290 const struct cache_entry *ce)
291 {
292 return (!S_ISGITLINK(ce->ce_mode) &&
293 is_racy_stat(istate, &ce->ce_stat_data));
294 }
295
296 int match_stat_data_racy(const struct index_state *istate,
297 const struct stat_data *sd, struct stat *st)
298 {
299 if (is_racy_stat(istate, sd))
300 return MTIME_CHANGED;
301 return match_stat_data(sd, st);
302 }
303
304 int ie_match_stat(const struct index_state *istate,
305 const struct cache_entry *ce, struct stat *st,
306 unsigned int options)
307 {
308 unsigned int changed;
309 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
310 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
311 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
312
313 /*
314 * If it's marked as always valid in the index, it's
315 * valid whatever the checked-out copy says.
316 *
317 * skip-worktree has the same effect with higher precedence
318 */
319 if (!ignore_skip_worktree && ce_skip_worktree(ce))
320 return 0;
321 if (!ignore_valid && (ce->ce_flags & CE_VALID))
322 return 0;
323
324 /*
325 * Intent-to-add entries have not been added, so the index entry
326 * by definition never matches what is in the work tree until it
327 * actually gets added.
328 */
329 if (ce_intent_to_add(ce))
330 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
331
332 changed = ce_match_stat_basic(ce, st);
333
334 /*
335 * Within 1 second of this sequence:
336 * echo xyzzy >file && git-update-index --add file
337 * running this command:
338 * echo frotz >file
339 * would give a falsely clean cache entry. The mtime and
340 * length match the cache, and other stat fields do not change.
341 *
342 * We could detect this at update-index time (the cache entry
343 * being registered/updated records the same time as "now")
344 * and delay the return from git-update-index, but that would
345 * effectively mean we can make at most one commit per second,
346 * which is not acceptable. Instead, we check cache entries
347 * whose mtime are the same as the index file timestamp more
348 * carefully than others.
349 */
350 if (!changed && is_racy_timestamp(istate, ce)) {
351 if (assume_racy_is_modified)
352 changed |= DATA_CHANGED;
353 else
354 changed |= ce_modified_check_fs(ce, st);
355 }
356
357 return changed;
358 }
359
360 int ie_modified(const struct index_state *istate,
361 const struct cache_entry *ce,
362 struct stat *st, unsigned int options)
363 {
364 int changed, changed_fs;
365
366 changed = ie_match_stat(istate, ce, st, options);
367 if (!changed)
368 return 0;
369 /*
370 * If the mode or type has changed, there's no point in trying
371 * to refresh the entry - it's not going to match
372 */
373 if (changed & (MODE_CHANGED | TYPE_CHANGED))
374 return changed;
375
376 /*
377 * Immediately after read-tree or update-index --cacheinfo,
378 * the length field is zero, as we have never even read the
379 * lstat(2) information once, and we cannot trust DATA_CHANGED
380 * returned by ie_match_stat() which in turn was returned by
381 * ce_match_stat_basic() to signal that the filesize of the
382 * blob changed. We have to actually go to the filesystem to
383 * see if the contents match, and if so, should answer "unchanged".
384 *
385 * The logic does not apply to gitlinks, as ce_match_stat_basic()
386 * already has checked the actual HEAD from the filesystem in the
387 * subproject. If ie_match_stat() already said it is different,
388 * then we know it is.
389 */
390 if ((changed & DATA_CHANGED) &&
391 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
392 return changed;
393
394 changed_fs = ce_modified_check_fs(ce, st);
395 if (changed_fs)
396 return changed | changed_fs;
397 return 0;
398 }
399
400 int base_name_compare(const char *name1, int len1, int mode1,
401 const char *name2, int len2, int mode2)
402 {
403 unsigned char c1, c2;
404 int len = len1 < len2 ? len1 : len2;
405 int cmp;
406
407 cmp = memcmp(name1, name2, len);
408 if (cmp)
409 return cmp;
410 c1 = name1[len];
411 c2 = name2[len];
412 if (!c1 && S_ISDIR(mode1))
413 c1 = '/';
414 if (!c2 && S_ISDIR(mode2))
415 c2 = '/';
416 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
417 }
418
419 /*
420 * df_name_compare() is identical to base_name_compare(), except it
421 * compares conflicting directory/file entries as equal. Note that
422 * while a directory name compares as equal to a regular file, they
423 * then individually compare _differently_ to a filename that has
424 * a dot after the basename (because '\0' < '.' < '/').
425 *
426 * This is used by routines that want to traverse the git namespace
427 * but then handle conflicting entries together when possible.
428 */
429 int df_name_compare(const char *name1, int len1, int mode1,
430 const char *name2, int len2, int mode2)
431 {
432 int len = len1 < len2 ? len1 : len2, cmp;
433 unsigned char c1, c2;
434
435 cmp = memcmp(name1, name2, len);
436 if (cmp)
437 return cmp;
438 /* Directories and files compare equal (same length, same name) */
439 if (len1 == len2)
440 return 0;
441 c1 = name1[len];
442 if (!c1 && S_ISDIR(mode1))
443 c1 = '/';
444 c2 = name2[len];
445 if (!c2 && S_ISDIR(mode2))
446 c2 = '/';
447 if (c1 == '/' && !c2)
448 return 0;
449 if (c2 == '/' && !c1)
450 return 0;
451 return c1 - c2;
452 }
453
454 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
455 {
456 size_t min_len = (len1 < len2) ? len1 : len2;
457 int cmp = memcmp(name1, name2, min_len);
458 if (cmp)
459 return cmp;
460 if (len1 < len2)
461 return -1;
462 if (len1 > len2)
463 return 1;
464 return 0;
465 }
466
467 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
468 {
469 int cmp;
470
471 cmp = name_compare(name1, len1, name2, len2);
472 if (cmp)
473 return cmp;
474
475 if (stage1 < stage2)
476 return -1;
477 if (stage1 > stage2)
478 return 1;
479 return 0;
480 }
481
482 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
483 {
484 int first, last;
485
486 first = 0;
487 last = istate->cache_nr;
488 while (last > first) {
489 int next = (last + first) >> 1;
490 struct cache_entry *ce = istate->cache[next];
491 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
492 if (!cmp)
493 return next;
494 if (cmp < 0) {
495 last = next;
496 continue;
497 }
498 first = next+1;
499 }
500 return -first-1;
501 }
502
503 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
504 {
505 return index_name_stage_pos(istate, name, namelen, 0);
506 }
507
508 int remove_index_entry_at(struct index_state *istate, int pos)
509 {
510 struct cache_entry *ce = istate->cache[pos];
511
512 record_resolve_undo(istate, ce);
513 remove_name_hash(istate, ce);
514 save_or_free_index_entry(istate, ce);
515 istate->cache_changed |= CE_ENTRY_REMOVED;
516 istate->cache_nr--;
517 if (pos >= istate->cache_nr)
518 return 0;
519 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
520 istate->cache_nr - pos);
521 return 1;
522 }
523
524 /*
525 * Remove all cache entries marked for removal, that is where
526 * CE_REMOVE is set in ce_flags. This is much more effective than
527 * calling remove_index_entry_at() for each entry to be removed.
528 */
529 void remove_marked_cache_entries(struct index_state *istate)
530 {
531 struct cache_entry **ce_array = istate->cache;
532 unsigned int i, j;
533
534 for (i = j = 0; i < istate->cache_nr; i++) {
535 if (ce_array[i]->ce_flags & CE_REMOVE) {
536 remove_name_hash(istate, ce_array[i]);
537 save_or_free_index_entry(istate, ce_array[i]);
538 }
539 else
540 ce_array[j++] = ce_array[i];
541 }
542 if (j == istate->cache_nr)
543 return;
544 istate->cache_changed |= CE_ENTRY_REMOVED;
545 istate->cache_nr = j;
546 }
547
548 int remove_file_from_index(struct index_state *istate, const char *path)
549 {
550 int pos = index_name_pos(istate, path, strlen(path));
551 if (pos < 0)
552 pos = -pos-1;
553 cache_tree_invalidate_path(istate, path);
554 untracked_cache_remove_from_index(istate, path);
555 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
556 remove_index_entry_at(istate, pos);
557 return 0;
558 }
559
560 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
561 {
562 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
563 }
564
565 static int index_name_pos_also_unmerged(struct index_state *istate,
566 const char *path, int namelen)
567 {
568 int pos = index_name_pos(istate, path, namelen);
569 struct cache_entry *ce;
570
571 if (pos >= 0)
572 return pos;
573
574 /* maybe unmerged? */
575 pos = -1 - pos;
576 if (pos >= istate->cache_nr ||
577 compare_name((ce = istate->cache[pos]), path, namelen))
578 return -1;
579
580 /* order of preference: stage 2, 1, 3 */
581 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
582 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
583 !compare_name(ce, path, namelen))
584 pos++;
585 return pos;
586 }
587
588 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
589 {
590 int len = ce_namelen(ce);
591 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
592 }
593
594 /*
595 * If we add a filename that aliases in the cache, we will use the
596 * name that we already have - but we don't want to update the same
597 * alias twice, because that implies that there were actually two
598 * different files with aliasing names!
599 *
600 * So we use the CE_ADDED flag to verify that the alias was an old
601 * one before we accept it as
602 */
603 static struct cache_entry *create_alias_ce(struct index_state *istate,
604 struct cache_entry *ce,
605 struct cache_entry *alias)
606 {
607 int len;
608 struct cache_entry *new;
609
610 if (alias->ce_flags & CE_ADDED)
611 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
612
613 /* Ok, create the new entry using the name of the existing alias */
614 len = ce_namelen(alias);
615 new = xcalloc(1, cache_entry_size(len));
616 memcpy(new->name, alias->name, len);
617 copy_cache_entry(new, ce);
618 save_or_free_index_entry(istate, ce);
619 return new;
620 }
621
622 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
623 {
624 unsigned char sha1[20];
625 if (write_sha1_file("", 0, blob_type, sha1))
626 die("cannot create an empty blob in the object database");
627 hashcpy(ce->oid.hash, sha1);
628 }
629
630 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
631 {
632 int size, namelen, was_same;
633 mode_t st_mode = st->st_mode;
634 struct cache_entry *ce, *alias;
635 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
636 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
637 int pretend = flags & ADD_CACHE_PRETEND;
638 int intent_only = flags & ADD_CACHE_INTENT;
639 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
640 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
641
642 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
643 return error("%s: can only add regular files, symbolic links or git-directories", path);
644
645 namelen = strlen(path);
646 if (S_ISDIR(st_mode)) {
647 while (namelen && path[namelen-1] == '/')
648 namelen--;
649 }
650 size = cache_entry_size(namelen);
651 ce = xcalloc(1, size);
652 memcpy(ce->name, path, namelen);
653 ce->ce_namelen = namelen;
654 if (!intent_only)
655 fill_stat_cache_info(ce, st);
656 else
657 ce->ce_flags |= CE_INTENT_TO_ADD;
658
659
660 if (trust_executable_bit && has_symlinks) {
661 ce->ce_mode = create_ce_mode(st_mode);
662 } else {
663 /* If there is an existing entry, pick the mode bits and type
664 * from it, otherwise assume unexecutable regular file.
665 */
666 struct cache_entry *ent;
667 int pos = index_name_pos_also_unmerged(istate, path, namelen);
668
669 ent = (0 <= pos) ? istate->cache[pos] : NULL;
670 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
671 }
672
673 /* When core.ignorecase=true, determine if a directory of the same name but differing
674 * case already exists within the Git repository. If it does, ensure the directory
675 * case of the file being added to the repository matches (is folded into) the existing
676 * entry's directory case.
677 */
678 if (ignore_case) {
679 adjust_dirname_case(istate, ce->name);
680 }
681
682 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
683 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
684 /* Nothing changed, really */
685 if (!S_ISGITLINK(alias->ce_mode))
686 ce_mark_uptodate(alias);
687 alias->ce_flags |= CE_ADDED;
688
689 free(ce);
690 return 0;
691 }
692 if (!intent_only) {
693 if (index_path(&ce->oid, path, st, HASH_WRITE_OBJECT)) {
694 free(ce);
695 return error("unable to index file %s", path);
696 }
697 } else
698 set_object_name_for_intent_to_add_entry(ce);
699
700 if (ignore_case && alias && different_name(ce, alias))
701 ce = create_alias_ce(istate, ce, alias);
702 ce->ce_flags |= CE_ADDED;
703
704 /* It was suspected to be racily clean, but it turns out to be Ok */
705 was_same = (alias &&
706 !ce_stage(alias) &&
707 !oidcmp(&alias->oid, &ce->oid) &&
708 ce->ce_mode == alias->ce_mode);
709
710 if (pretend)
711 free(ce);
712 else if (add_index_entry(istate, ce, add_option)) {
713 free(ce);
714 return error("unable to add %s to index", path);
715 }
716 if (verbose && !was_same)
717 printf("add '%s'\n", path);
718 return 0;
719 }
720
721 int add_file_to_index(struct index_state *istate, const char *path, int flags)
722 {
723 struct stat st;
724 if (lstat(path, &st))
725 die_errno("unable to stat '%s'", path);
726 return add_to_index(istate, path, &st, flags);
727 }
728
729 struct cache_entry *make_cache_entry(unsigned int mode,
730 const unsigned char *sha1, const char *path, int stage,
731 unsigned int refresh_options)
732 {
733 int size, len;
734 struct cache_entry *ce, *ret;
735
736 if (!verify_path(path)) {
737 error("Invalid path '%s'", path);
738 return NULL;
739 }
740
741 len = strlen(path);
742 size = cache_entry_size(len);
743 ce = xcalloc(1, size);
744
745 hashcpy(ce->oid.hash, sha1);
746 memcpy(ce->name, path, len);
747 ce->ce_flags = create_ce_flags(stage);
748 ce->ce_namelen = len;
749 ce->ce_mode = create_ce_mode(mode);
750
751 ret = refresh_cache_entry(ce, refresh_options);
752 if (ret != ce)
753 free(ce);
754 return ret;
755 }
756
757 /*
758 * Chmod an index entry with either +x or -x.
759 *
760 * Returns -1 if the chmod for the particular cache entry failed (if it's
761 * not a regular file), -2 if an invalid flip argument is passed in, 0
762 * otherwise.
763 */
764 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
765 char flip)
766 {
767 if (!S_ISREG(ce->ce_mode))
768 return -1;
769 switch (flip) {
770 case '+':
771 ce->ce_mode |= 0111;
772 break;
773 case '-':
774 ce->ce_mode &= ~0111;
775 break;
776 default:
777 return -2;
778 }
779 cache_tree_invalidate_path(istate, ce->name);
780 ce->ce_flags |= CE_UPDATE_IN_BASE;
781 istate->cache_changed |= CE_ENTRY_CHANGED;
782
783 return 0;
784 }
785
786 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
787 {
788 int len = ce_namelen(a);
789 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
790 }
791
792 /*
793 * We fundamentally don't like some paths: we don't want
794 * dot or dot-dot anywhere, and for obvious reasons don't
795 * want to recurse into ".git" either.
796 *
797 * Also, we don't want double slashes or slashes at the
798 * end that can make pathnames ambiguous.
799 */
800 static int verify_dotfile(const char *rest)
801 {
802 /*
803 * The first character was '.', but that
804 * has already been discarded, we now test
805 * the rest.
806 */
807
808 /* "." is not allowed */
809 if (*rest == '\0' || is_dir_sep(*rest))
810 return 0;
811
812 switch (*rest) {
813 /*
814 * ".git" followed by NUL or slash is bad. This
815 * shares the path end test with the ".." case.
816 */
817 case 'g':
818 case 'G':
819 if (rest[1] != 'i' && rest[1] != 'I')
820 break;
821 if (rest[2] != 't' && rest[2] != 'T')
822 break;
823 rest += 2;
824 /* fallthrough */
825 case '.':
826 if (rest[1] == '\0' || is_dir_sep(rest[1]))
827 return 0;
828 }
829 return 1;
830 }
831
832 int verify_path(const char *path)
833 {
834 char c;
835
836 if (has_dos_drive_prefix(path))
837 return 0;
838
839 goto inside;
840 for (;;) {
841 if (!c)
842 return 1;
843 if (is_dir_sep(c)) {
844 inside:
845 if (protect_hfs && is_hfs_dotgit(path))
846 return 0;
847 if (protect_ntfs && is_ntfs_dotgit(path))
848 return 0;
849 c = *path++;
850 if ((c == '.' && !verify_dotfile(path)) ||
851 is_dir_sep(c) || c == '\0')
852 return 0;
853 }
854 c = *path++;
855 }
856 }
857
858 /*
859 * Do we have another file that has the beginning components being a
860 * proper superset of the name we're trying to add?
861 */
862 static int has_file_name(struct index_state *istate,
863 const struct cache_entry *ce, int pos, int ok_to_replace)
864 {
865 int retval = 0;
866 int len = ce_namelen(ce);
867 int stage = ce_stage(ce);
868 const char *name = ce->name;
869
870 while (pos < istate->cache_nr) {
871 struct cache_entry *p = istate->cache[pos++];
872
873 if (len >= ce_namelen(p))
874 break;
875 if (memcmp(name, p->name, len))
876 break;
877 if (ce_stage(p) != stage)
878 continue;
879 if (p->name[len] != '/')
880 continue;
881 if (p->ce_flags & CE_REMOVE)
882 continue;
883 retval = -1;
884 if (!ok_to_replace)
885 break;
886 remove_index_entry_at(istate, --pos);
887 }
888 return retval;
889 }
890
891
892 /*
893 * Like strcmp(), but also return the offset of the first change.
894 * If strings are equal, return the length.
895 */
896 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
897 {
898 size_t k;
899
900 if (!first_change)
901 return strcmp(s1, s2);
902
903 for (k = 0; s1[k] == s2[k]; k++)
904 if (s1[k] == '\0')
905 break;
906
907 *first_change = k;
908 return (unsigned char)s1[k] - (unsigned char)s2[k];
909 }
910
911 /*
912 * Do we have another file with a pathname that is a proper
913 * subset of the name we're trying to add?
914 *
915 * That is, is there another file in the index with a path
916 * that matches a sub-directory in the given entry?
917 */
918 static int has_dir_name(struct index_state *istate,
919 const struct cache_entry *ce, int pos, int ok_to_replace)
920 {
921 int retval = 0;
922 int stage = ce_stage(ce);
923 const char *name = ce->name;
924 const char *slash = name + ce_namelen(ce);
925 size_t len_eq_last;
926 int cmp_last = 0;
927
928 /*
929 * We are frequently called during an iteration on a sorted
930 * list of pathnames and while building a new index. Therefore,
931 * there is a high probability that this entry will eventually
932 * be appended to the index, rather than inserted in the middle.
933 * If we can confirm that, we can avoid binary searches on the
934 * components of the pathname.
935 *
936 * Compare the entry's full path with the last path in the index.
937 */
938 if (istate->cache_nr > 0) {
939 cmp_last = strcmp_offset(name,
940 istate->cache[istate->cache_nr - 1]->name,
941 &len_eq_last);
942 if (cmp_last > 0) {
943 if (len_eq_last == 0) {
944 /*
945 * The entry sorts AFTER the last one in the
946 * index and their paths have no common prefix,
947 * so there cannot be a F/D conflict.
948 */
949 return retval;
950 } else {
951 /*
952 * The entry sorts AFTER the last one in the
953 * index, but has a common prefix. Fall through
954 * to the loop below to disect the entry's path
955 * and see where the difference is.
956 */
957 }
958 } else if (cmp_last == 0) {
959 /*
960 * The entry exactly matches the last one in the
961 * index, but because of multiple stage and CE_REMOVE
962 * items, we fall through and let the regular search
963 * code handle it.
964 */
965 }
966 }
967
968 for (;;) {
969 size_t len;
970
971 for (;;) {
972 if (*--slash == '/')
973 break;
974 if (slash <= ce->name)
975 return retval;
976 }
977 len = slash - name;
978
979 if (cmp_last > 0) {
980 /*
981 * (len + 1) is a directory boundary (including
982 * the trailing slash). And since the loop is
983 * decrementing "slash", the first iteration is
984 * the longest directory prefix; subsequent
985 * iterations consider parent directories.
986 */
987
988 if (len + 1 <= len_eq_last) {
989 /*
990 * The directory prefix (including the trailing
991 * slash) also appears as a prefix in the last
992 * entry, so the remainder cannot collide (because
993 * strcmp said the whole path was greater).
994 *
995 * EQ: last: xxx/A
996 * this: xxx/B
997 *
998 * LT: last: xxx/file_A
999 * this: xxx/file_B
1000 */
1001 return retval;
1002 }
1003
1004 if (len > len_eq_last) {
1005 /*
1006 * This part of the directory prefix (excluding
1007 * the trailing slash) is longer than the known
1008 * equal portions, so this sub-directory cannot
1009 * collide with a file.
1010 *
1011 * GT: last: xxxA
1012 * this: xxxB/file
1013 */
1014 return retval;
1015 }
1016
1017 if (istate->cache_nr > 0 &&
1018 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1019 /*
1020 * The directory prefix lines up with part of
1021 * a longer file or directory name, but sorts
1022 * after it, so this sub-directory cannot
1023 * collide with a file.
1024 *
1025 * last: xxx/yy-file (because '-' sorts before '/')
1026 * this: xxx/yy/abc
1027 */
1028 return retval;
1029 }
1030
1031 /*
1032 * This is a possible collision. Fall through and
1033 * let the regular search code handle it.
1034 *
1035 * last: xxx
1036 * this: xxx/file
1037 */
1038 }
1039
1040 pos = index_name_stage_pos(istate, name, len, stage);
1041 if (pos >= 0) {
1042 /*
1043 * Found one, but not so fast. This could
1044 * be a marker that says "I was here, but
1045 * I am being removed". Such an entry is
1046 * not a part of the resulting tree, and
1047 * it is Ok to have a directory at the same
1048 * path.
1049 */
1050 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1051 retval = -1;
1052 if (!ok_to_replace)
1053 break;
1054 remove_index_entry_at(istate, pos);
1055 continue;
1056 }
1057 }
1058 else
1059 pos = -pos-1;
1060
1061 /*
1062 * Trivial optimization: if we find an entry that
1063 * already matches the sub-directory, then we know
1064 * we're ok, and we can exit.
1065 */
1066 while (pos < istate->cache_nr) {
1067 struct cache_entry *p = istate->cache[pos];
1068 if ((ce_namelen(p) <= len) ||
1069 (p->name[len] != '/') ||
1070 memcmp(p->name, name, len))
1071 break; /* not our subdirectory */
1072 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1073 /*
1074 * p is at the same stage as our entry, and
1075 * is a subdirectory of what we are looking
1076 * at, so we cannot have conflicts at our
1077 * level or anything shorter.
1078 */
1079 return retval;
1080 pos++;
1081 }
1082 }
1083 return retval;
1084 }
1085
1086 /* We may be in a situation where we already have path/file and path
1087 * is being added, or we already have path and path/file is being
1088 * added. Either one would result in a nonsense tree that has path
1089 * twice when git-write-tree tries to write it out. Prevent it.
1090 *
1091 * If ok-to-replace is specified, we remove the conflicting entries
1092 * from the cache so the caller should recompute the insert position.
1093 * When this happens, we return non-zero.
1094 */
1095 static int check_file_directory_conflict(struct index_state *istate,
1096 const struct cache_entry *ce,
1097 int pos, int ok_to_replace)
1098 {
1099 int retval;
1100
1101 /*
1102 * When ce is an "I am going away" entry, we allow it to be added
1103 */
1104 if (ce->ce_flags & CE_REMOVE)
1105 return 0;
1106
1107 /*
1108 * We check if the path is a sub-path of a subsequent pathname
1109 * first, since removing those will not change the position
1110 * in the array.
1111 */
1112 retval = has_file_name(istate, ce, pos, ok_to_replace);
1113
1114 /*
1115 * Then check if the path might have a clashing sub-directory
1116 * before it.
1117 */
1118 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1119 }
1120
1121 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1122 {
1123 int pos;
1124 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1125 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1126 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1127 int new_only = option & ADD_CACHE_NEW_ONLY;
1128
1129 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1130 cache_tree_invalidate_path(istate, ce->name);
1131
1132 /*
1133 * If this entry's path sorts after the last entry in the index,
1134 * we can avoid searching for it.
1135 */
1136 if (istate->cache_nr > 0 &&
1137 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1138 pos = -istate->cache_nr - 1;
1139 else
1140 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1141
1142 /* existing match? Just replace it. */
1143 if (pos >= 0) {
1144 if (!new_only)
1145 replace_index_entry(istate, pos, ce);
1146 return 0;
1147 }
1148 pos = -pos-1;
1149
1150 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1151 untracked_cache_add_to_index(istate, ce->name);
1152
1153 /*
1154 * Inserting a merged entry ("stage 0") into the index
1155 * will always replace all non-merged entries..
1156 */
1157 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1158 while (ce_same_name(istate->cache[pos], ce)) {
1159 ok_to_add = 1;
1160 if (!remove_index_entry_at(istate, pos))
1161 break;
1162 }
1163 }
1164
1165 if (!ok_to_add)
1166 return -1;
1167 if (!verify_path(ce->name))
1168 return error("Invalid path '%s'", ce->name);
1169
1170 if (!skip_df_check &&
1171 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1172 if (!ok_to_replace)
1173 return error("'%s' appears as both a file and as a directory",
1174 ce->name);
1175 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1176 pos = -pos-1;
1177 }
1178 return pos + 1;
1179 }
1180
1181 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1182 {
1183 int pos;
1184
1185 if (option & ADD_CACHE_JUST_APPEND)
1186 pos = istate->cache_nr;
1187 else {
1188 int ret;
1189 ret = add_index_entry_with_check(istate, ce, option);
1190 if (ret <= 0)
1191 return ret;
1192 pos = ret - 1;
1193 }
1194
1195 /* Make sure the array is big enough .. */
1196 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1197
1198 /* Add it in.. */
1199 istate->cache_nr++;
1200 if (istate->cache_nr > pos + 1)
1201 memmove(istate->cache + pos + 1,
1202 istate->cache + pos,
1203 (istate->cache_nr - pos - 1) * sizeof(ce));
1204 set_index_entry(istate, pos, ce);
1205 istate->cache_changed |= CE_ENTRY_ADDED;
1206 return 0;
1207 }
1208
1209 /*
1210 * "refresh" does not calculate a new sha1 file or bring the
1211 * cache up-to-date for mode/content changes. But what it
1212 * _does_ do is to "re-match" the stat information of a file
1213 * with the cache, so that you can refresh the cache for a
1214 * file that hasn't been changed but where the stat entry is
1215 * out of date.
1216 *
1217 * For example, you'd want to do this after doing a "git-read-tree",
1218 * to link up the stat cache details with the proper files.
1219 */
1220 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1221 struct cache_entry *ce,
1222 unsigned int options, int *err,
1223 int *changed_ret)
1224 {
1225 struct stat st;
1226 struct cache_entry *updated;
1227 int changed, size;
1228 int refresh = options & CE_MATCH_REFRESH;
1229 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1230 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1231 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1232
1233 if (!refresh || ce_uptodate(ce))
1234 return ce;
1235
1236 /*
1237 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1238 * that the change to the work tree does not matter and told
1239 * us not to worry.
1240 */
1241 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1242 ce_mark_uptodate(ce);
1243 return ce;
1244 }
1245 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1246 ce_mark_uptodate(ce);
1247 return ce;
1248 }
1249
1250 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1251 if (ignore_missing)
1252 return ce;
1253 if (err)
1254 *err = ENOENT;
1255 return NULL;
1256 }
1257
1258 if (lstat(ce->name, &st) < 0) {
1259 if (ignore_missing && errno == ENOENT)
1260 return ce;
1261 if (err)
1262 *err = errno;
1263 return NULL;
1264 }
1265
1266 changed = ie_match_stat(istate, ce, &st, options);
1267 if (changed_ret)
1268 *changed_ret = changed;
1269 if (!changed) {
1270 /*
1271 * The path is unchanged. If we were told to ignore
1272 * valid bit, then we did the actual stat check and
1273 * found that the entry is unmodified. If the entry
1274 * is not marked VALID, this is the place to mark it
1275 * valid again, under "assume unchanged" mode.
1276 */
1277 if (ignore_valid && assume_unchanged &&
1278 !(ce->ce_flags & CE_VALID))
1279 ; /* mark this one VALID again */
1280 else {
1281 /*
1282 * We do not mark the index itself "modified"
1283 * because CE_UPTODATE flag is in-core only;
1284 * we are not going to write this change out.
1285 */
1286 if (!S_ISGITLINK(ce->ce_mode))
1287 ce_mark_uptodate(ce);
1288 return ce;
1289 }
1290 }
1291
1292 if (ie_modified(istate, ce, &st, options)) {
1293 if (err)
1294 *err = EINVAL;
1295 return NULL;
1296 }
1297
1298 size = ce_size(ce);
1299 updated = xmalloc(size);
1300 memcpy(updated, ce, size);
1301 fill_stat_cache_info(updated, &st);
1302 /*
1303 * If ignore_valid is not set, we should leave CE_VALID bit
1304 * alone. Otherwise, paths marked with --no-assume-unchanged
1305 * (i.e. things to be edited) will reacquire CE_VALID bit
1306 * automatically, which is not really what we want.
1307 */
1308 if (!ignore_valid && assume_unchanged &&
1309 !(ce->ce_flags & CE_VALID))
1310 updated->ce_flags &= ~CE_VALID;
1311
1312 /* istate->cache_changed is updated in the caller */
1313 return updated;
1314 }
1315
1316 static void show_file(const char * fmt, const char * name, int in_porcelain,
1317 int * first, const char *header_msg)
1318 {
1319 if (in_porcelain && *first && header_msg) {
1320 printf("%s\n", header_msg);
1321 *first = 0;
1322 }
1323 printf(fmt, name);
1324 }
1325
1326 int refresh_index(struct index_state *istate, unsigned int flags,
1327 const struct pathspec *pathspec,
1328 char *seen, const char *header_msg)
1329 {
1330 int i;
1331 int has_errors = 0;
1332 int really = (flags & REFRESH_REALLY) != 0;
1333 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1334 int quiet = (flags & REFRESH_QUIET) != 0;
1335 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1336 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1337 int first = 1;
1338 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1339 unsigned int options = (CE_MATCH_REFRESH |
1340 (really ? CE_MATCH_IGNORE_VALID : 0) |
1341 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1342 const char *modified_fmt;
1343 const char *deleted_fmt;
1344 const char *typechange_fmt;
1345 const char *added_fmt;
1346 const char *unmerged_fmt;
1347
1348 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1349 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1350 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1351 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1352 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1353 for (i = 0; i < istate->cache_nr; i++) {
1354 struct cache_entry *ce, *new;
1355 int cache_errno = 0;
1356 int changed = 0;
1357 int filtered = 0;
1358
1359 ce = istate->cache[i];
1360 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1361 continue;
1362
1363 if (pathspec && !ce_path_match(ce, pathspec, seen))
1364 filtered = 1;
1365
1366 if (ce_stage(ce)) {
1367 while ((i < istate->cache_nr) &&
1368 ! strcmp(istate->cache[i]->name, ce->name))
1369 i++;
1370 i--;
1371 if (allow_unmerged)
1372 continue;
1373 if (!filtered)
1374 show_file(unmerged_fmt, ce->name, in_porcelain,
1375 &first, header_msg);
1376 has_errors = 1;
1377 continue;
1378 }
1379
1380 if (filtered)
1381 continue;
1382
1383 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1384 if (new == ce)
1385 continue;
1386 if (!new) {
1387 const char *fmt;
1388
1389 if (really && cache_errno == EINVAL) {
1390 /* If we are doing --really-refresh that
1391 * means the index is not valid anymore.
1392 */
1393 ce->ce_flags &= ~CE_VALID;
1394 ce->ce_flags |= CE_UPDATE_IN_BASE;
1395 istate->cache_changed |= CE_ENTRY_CHANGED;
1396 }
1397 if (quiet)
1398 continue;
1399
1400 if (cache_errno == ENOENT)
1401 fmt = deleted_fmt;
1402 else if (ce_intent_to_add(ce))
1403 fmt = added_fmt; /* must be before other checks */
1404 else if (changed & TYPE_CHANGED)
1405 fmt = typechange_fmt;
1406 else
1407 fmt = modified_fmt;
1408 show_file(fmt,
1409 ce->name, in_porcelain, &first, header_msg);
1410 has_errors = 1;
1411 continue;
1412 }
1413
1414 replace_index_entry(istate, i, new);
1415 }
1416 return has_errors;
1417 }
1418
1419 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1420 unsigned int options)
1421 {
1422 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1423 }
1424
1425
1426 /*****************************************************************
1427 * Index File I/O
1428 *****************************************************************/
1429
1430 #define INDEX_FORMAT_DEFAULT 3
1431
1432 static unsigned int get_index_format_default(void)
1433 {
1434 char *envversion = getenv("GIT_INDEX_VERSION");
1435 char *endp;
1436 int value;
1437 unsigned int version = INDEX_FORMAT_DEFAULT;
1438
1439 if (!envversion) {
1440 if (!git_config_get_int("index.version", &value))
1441 version = value;
1442 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1443 warning(_("index.version set, but the value is invalid.\n"
1444 "Using version %i"), INDEX_FORMAT_DEFAULT);
1445 return INDEX_FORMAT_DEFAULT;
1446 }
1447 return version;
1448 }
1449
1450 version = strtoul(envversion, &endp, 10);
1451 if (*endp ||
1452 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1453 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1454 "Using version %i"), INDEX_FORMAT_DEFAULT);
1455 version = INDEX_FORMAT_DEFAULT;
1456 }
1457 return version;
1458 }
1459
1460 /*
1461 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1462 * Again - this is just a (very strong in practice) heuristic that
1463 * the inode hasn't changed.
1464 *
1465 * We save the fields in big-endian order to allow using the
1466 * index file over NFS transparently.
1467 */
1468 struct ondisk_cache_entry {
1469 struct cache_time ctime;
1470 struct cache_time mtime;
1471 uint32_t dev;
1472 uint32_t ino;
1473 uint32_t mode;
1474 uint32_t uid;
1475 uint32_t gid;
1476 uint32_t size;
1477 unsigned char sha1[20];
1478 uint16_t flags;
1479 char name[FLEX_ARRAY]; /* more */
1480 };
1481
1482 /*
1483 * This struct is used when CE_EXTENDED bit is 1
1484 * The struct must match ondisk_cache_entry exactly from
1485 * ctime till flags
1486 */
1487 struct ondisk_cache_entry_extended {
1488 struct cache_time ctime;
1489 struct cache_time mtime;
1490 uint32_t dev;
1491 uint32_t ino;
1492 uint32_t mode;
1493 uint32_t uid;
1494 uint32_t gid;
1495 uint32_t size;
1496 unsigned char sha1[20];
1497 uint16_t flags;
1498 uint16_t flags2;
1499 char name[FLEX_ARRAY]; /* more */
1500 };
1501
1502 /* These are only used for v3 or lower */
1503 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1504 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1505 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1506 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1507 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1508 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1509 ondisk_cache_entry_size(ce_namelen(ce)))
1510
1511 /* Allow fsck to force verification of the index checksum. */
1512 int verify_index_checksum;
1513
1514 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1515 {
1516 git_SHA_CTX c;
1517 unsigned char sha1[20];
1518 int hdr_version;
1519
1520 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1521 return error("bad signature");
1522 hdr_version = ntohl(hdr->hdr_version);
1523 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1524 return error("bad index version %d", hdr_version);
1525
1526 if (!verify_index_checksum)
1527 return 0;
1528
1529 git_SHA1_Init(&c);
1530 git_SHA1_Update(&c, hdr, size - 20);
1531 git_SHA1_Final(sha1, &c);
1532 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1533 return error("bad index file sha1 signature");
1534 return 0;
1535 }
1536
1537 static int read_index_extension(struct index_state *istate,
1538 const char *ext, void *data, unsigned long sz)
1539 {
1540 switch (CACHE_EXT(ext)) {
1541 case CACHE_EXT_TREE:
1542 istate->cache_tree = cache_tree_read(data, sz);
1543 break;
1544 case CACHE_EXT_RESOLVE_UNDO:
1545 istate->resolve_undo = resolve_undo_read(data, sz);
1546 break;
1547 case CACHE_EXT_LINK:
1548 if (read_link_extension(istate, data, sz))
1549 return -1;
1550 break;
1551 case CACHE_EXT_UNTRACKED:
1552 istate->untracked = read_untracked_extension(data, sz);
1553 break;
1554 default:
1555 if (*ext < 'A' || 'Z' < *ext)
1556 return error("index uses %.4s extension, which we do not understand",
1557 ext);
1558 fprintf(stderr, "ignoring %.4s extension\n", ext);
1559 break;
1560 }
1561 return 0;
1562 }
1563
1564 int hold_locked_index(struct lock_file *lk, int lock_flags)
1565 {
1566 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1567 }
1568
1569 int read_index(struct index_state *istate)
1570 {
1571 return read_index_from(istate, get_index_file());
1572 }
1573
1574 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1575 unsigned int flags,
1576 const char *name,
1577 size_t len)
1578 {
1579 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1580
1581 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1582 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1583 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1584 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1585 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1586 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1587 ce->ce_mode = get_be32(&ondisk->mode);
1588 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1589 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1590 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1591 ce->ce_flags = flags & ~CE_NAMEMASK;
1592 ce->ce_namelen = len;
1593 ce->index = 0;
1594 hashcpy(ce->oid.hash, ondisk->sha1);
1595 memcpy(ce->name, name, len);
1596 ce->name[len] = '\0';
1597 return ce;
1598 }
1599
1600 /*
1601 * Adjacent cache entries tend to share the leading paths, so it makes
1602 * sense to only store the differences in later entries. In the v4
1603 * on-disk format of the index, each on-disk cache entry stores the
1604 * number of bytes to be stripped from the end of the previous name,
1605 * and the bytes to append to the result, to come up with its name.
1606 */
1607 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1608 {
1609 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1610 size_t len = decode_varint(&cp);
1611
1612 if (name->len < len)
1613 die("malformed name field in the index");
1614 strbuf_remove(name, name->len - len, len);
1615 for (ep = cp; *ep; ep++)
1616 ; /* find the end */
1617 strbuf_add(name, cp, ep - cp);
1618 return (const char *)ep + 1 - cp_;
1619 }
1620
1621 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1622 unsigned long *ent_size,
1623 struct strbuf *previous_name)
1624 {
1625 struct cache_entry *ce;
1626 size_t len;
1627 const char *name;
1628 unsigned int flags;
1629
1630 /* On-disk flags are just 16 bits */
1631 flags = get_be16(&ondisk->flags);
1632 len = flags & CE_NAMEMASK;
1633
1634 if (flags & CE_EXTENDED) {
1635 struct ondisk_cache_entry_extended *ondisk2;
1636 int extended_flags;
1637 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1638 extended_flags = get_be16(&ondisk2->flags2) << 16;
1639 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1640 if (extended_flags & ~CE_EXTENDED_FLAGS)
1641 die("Unknown index entry format %08x", extended_flags);
1642 flags |= extended_flags;
1643 name = ondisk2->name;
1644 }
1645 else
1646 name = ondisk->name;
1647
1648 if (!previous_name) {
1649 /* v3 and earlier */
1650 if (len == CE_NAMEMASK)
1651 len = strlen(name);
1652 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1653
1654 *ent_size = ondisk_ce_size(ce);
1655 } else {
1656 unsigned long consumed;
1657 consumed = expand_name_field(previous_name, name);
1658 ce = cache_entry_from_ondisk(ondisk, flags,
1659 previous_name->buf,
1660 previous_name->len);
1661
1662 *ent_size = (name - ((char *)ondisk)) + consumed;
1663 }
1664 return ce;
1665 }
1666
1667 static void check_ce_order(struct index_state *istate)
1668 {
1669 unsigned int i;
1670
1671 for (i = 1; i < istate->cache_nr; i++) {
1672 struct cache_entry *ce = istate->cache[i - 1];
1673 struct cache_entry *next_ce = istate->cache[i];
1674 int name_compare = strcmp(ce->name, next_ce->name);
1675
1676 if (0 < name_compare)
1677 die("unordered stage entries in index");
1678 if (!name_compare) {
1679 if (!ce_stage(ce))
1680 die("multiple stage entries for merged file '%s'",
1681 ce->name);
1682 if (ce_stage(ce) > ce_stage(next_ce))
1683 die("unordered stage entries for '%s'",
1684 ce->name);
1685 }
1686 }
1687 }
1688
1689 static void tweak_untracked_cache(struct index_state *istate)
1690 {
1691 switch (git_config_get_untracked_cache()) {
1692 case -1: /* keep: do nothing */
1693 break;
1694 case 0: /* false */
1695 remove_untracked_cache(istate);
1696 break;
1697 case 1: /* true */
1698 add_untracked_cache(istate);
1699 break;
1700 default: /* unknown value: do nothing */
1701 break;
1702 }
1703 }
1704
1705 static void tweak_split_index(struct index_state *istate)
1706 {
1707 switch (git_config_get_split_index()) {
1708 case -1: /* unset: do nothing */
1709 break;
1710 case 0: /* false */
1711 remove_split_index(istate);
1712 break;
1713 case 1: /* true */
1714 add_split_index(istate);
1715 break;
1716 default: /* unknown value: do nothing */
1717 break;
1718 }
1719 }
1720
1721 static void post_read_index_from(struct index_state *istate)
1722 {
1723 check_ce_order(istate);
1724 tweak_untracked_cache(istate);
1725 tweak_split_index(istate);
1726 }
1727
1728 /* remember to discard_cache() before reading a different cache! */
1729 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1730 {
1731 int fd, i;
1732 struct stat st;
1733 unsigned long src_offset;
1734 struct cache_header *hdr;
1735 void *mmap;
1736 size_t mmap_size;
1737 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1738
1739 if (istate->initialized)
1740 return istate->cache_nr;
1741
1742 istate->timestamp.sec = 0;
1743 istate->timestamp.nsec = 0;
1744 fd = open(path, O_RDONLY);
1745 if (fd < 0) {
1746 if (!must_exist && errno == ENOENT)
1747 return 0;
1748 die_errno("%s: index file open failed", path);
1749 }
1750
1751 if (fstat(fd, &st))
1752 die_errno("cannot stat the open index");
1753
1754 mmap_size = xsize_t(st.st_size);
1755 if (mmap_size < sizeof(struct cache_header) + 20)
1756 die("index file smaller than expected");
1757
1758 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1759 if (mmap == MAP_FAILED)
1760 die_errno("unable to map index file");
1761 close(fd);
1762
1763 hdr = mmap;
1764 if (verify_hdr(hdr, mmap_size) < 0)
1765 goto unmap;
1766
1767 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1768 istate->version = ntohl(hdr->hdr_version);
1769 istate->cache_nr = ntohl(hdr->hdr_entries);
1770 istate->cache_alloc = alloc_nr(istate->cache_nr);
1771 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1772 istate->initialized = 1;
1773
1774 if (istate->version == 4)
1775 previous_name = &previous_name_buf;
1776 else
1777 previous_name = NULL;
1778
1779 src_offset = sizeof(*hdr);
1780 for (i = 0; i < istate->cache_nr; i++) {
1781 struct ondisk_cache_entry *disk_ce;
1782 struct cache_entry *ce;
1783 unsigned long consumed;
1784
1785 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1786 ce = create_from_disk(disk_ce, &consumed, previous_name);
1787 set_index_entry(istate, i, ce);
1788
1789 src_offset += consumed;
1790 }
1791 strbuf_release(&previous_name_buf);
1792 istate->timestamp.sec = st.st_mtime;
1793 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1794
1795 while (src_offset <= mmap_size - 20 - 8) {
1796 /* After an array of active_nr index entries,
1797 * there can be arbitrary number of extended
1798 * sections, each of which is prefixed with
1799 * extension name (4-byte) and section length
1800 * in 4-byte network byte order.
1801 */
1802 uint32_t extsize;
1803 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1804 extsize = ntohl(extsize);
1805 if (read_index_extension(istate,
1806 (const char *) mmap + src_offset,
1807 (char *) mmap + src_offset + 8,
1808 extsize) < 0)
1809 goto unmap;
1810 src_offset += 8;
1811 src_offset += extsize;
1812 }
1813 munmap(mmap, mmap_size);
1814 return istate->cache_nr;
1815
1816 unmap:
1817 munmap(mmap, mmap_size);
1818 die("index file corrupt");
1819 }
1820
1821 /*
1822 * Signal that the shared index is used by updating its mtime.
1823 *
1824 * This way, shared index can be removed if they have not been used
1825 * for some time.
1826 */
1827 static void freshen_shared_index(char *base_sha1_hex, int warn)
1828 {
1829 char *shared_index = git_pathdup("sharedindex.%s", base_sha1_hex);
1830 if (!check_and_freshen_file(shared_index, 1) && warn)
1831 warning("could not freshen shared index '%s'", shared_index);
1832 free(shared_index);
1833 }
1834
1835 int read_index_from(struct index_state *istate, const char *path)
1836 {
1837 struct split_index *split_index;
1838 int ret;
1839 char *base_sha1_hex;
1840 const char *base_path;
1841
1842 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1843 if (istate->initialized)
1844 return istate->cache_nr;
1845
1846 ret = do_read_index(istate, path, 0);
1847
1848 split_index = istate->split_index;
1849 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1850 post_read_index_from(istate);
1851 return ret;
1852 }
1853
1854 if (split_index->base)
1855 discard_index(split_index->base);
1856 else
1857 split_index->base = xcalloc(1, sizeof(*split_index->base));
1858
1859 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1860 base_path = git_path("sharedindex.%s", base_sha1_hex);
1861 ret = do_read_index(split_index->base, base_path, 1);
1862 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1863 die("broken index, expect %s in %s, got %s",
1864 base_sha1_hex, base_path,
1865 sha1_to_hex(split_index->base->sha1));
1866
1867 freshen_shared_index(base_sha1_hex, 0);
1868 merge_base_index(istate);
1869 post_read_index_from(istate);
1870 return ret;
1871 }
1872
1873 int is_index_unborn(struct index_state *istate)
1874 {
1875 return (!istate->cache_nr && !istate->timestamp.sec);
1876 }
1877
1878 int discard_index(struct index_state *istate)
1879 {
1880 int i;
1881
1882 for (i = 0; i < istate->cache_nr; i++) {
1883 if (istate->cache[i]->index &&
1884 istate->split_index &&
1885 istate->split_index->base &&
1886 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1887 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1888 continue;
1889 free(istate->cache[i]);
1890 }
1891 resolve_undo_clear_index(istate);
1892 istate->cache_nr = 0;
1893 istate->cache_changed = 0;
1894 istate->timestamp.sec = 0;
1895 istate->timestamp.nsec = 0;
1896 free_name_hash(istate);
1897 cache_tree_free(&(istate->cache_tree));
1898 istate->initialized = 0;
1899 FREE_AND_NULL(istate->cache);
1900 istate->cache_alloc = 0;
1901 discard_split_index(istate);
1902 free_untracked_cache(istate->untracked);
1903 istate->untracked = NULL;
1904 return 0;
1905 }
1906
1907 int unmerged_index(const struct index_state *istate)
1908 {
1909 int i;
1910 for (i = 0; i < istate->cache_nr; i++) {
1911 if (ce_stage(istate->cache[i]))
1912 return 1;
1913 }
1914 return 0;
1915 }
1916
1917 #define WRITE_BUFFER_SIZE 8192
1918 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1919 static unsigned long write_buffer_len;
1920
1921 static int ce_write_flush(git_SHA_CTX *context, int fd)
1922 {
1923 unsigned int buffered = write_buffer_len;
1924 if (buffered) {
1925 git_SHA1_Update(context, write_buffer, buffered);
1926 if (write_in_full(fd, write_buffer, buffered) < 0)
1927 return -1;
1928 write_buffer_len = 0;
1929 }
1930 return 0;
1931 }
1932
1933 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1934 {
1935 while (len) {
1936 unsigned int buffered = write_buffer_len;
1937 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1938 if (partial > len)
1939 partial = len;
1940 memcpy(write_buffer + buffered, data, partial);
1941 buffered += partial;
1942 if (buffered == WRITE_BUFFER_SIZE) {
1943 write_buffer_len = buffered;
1944 if (ce_write_flush(context, fd))
1945 return -1;
1946 buffered = 0;
1947 }
1948 write_buffer_len = buffered;
1949 len -= partial;
1950 data = (char *) data + partial;
1951 }
1952 return 0;
1953 }
1954
1955 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1956 unsigned int ext, unsigned int sz)
1957 {
1958 ext = htonl(ext);
1959 sz = htonl(sz);
1960 return ((ce_write(context, fd, &ext, 4) < 0) ||
1961 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1962 }
1963
1964 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1965 {
1966 unsigned int left = write_buffer_len;
1967
1968 if (left) {
1969 write_buffer_len = 0;
1970 git_SHA1_Update(context, write_buffer, left);
1971 }
1972
1973 /* Flush first if not enough space for SHA1 signature */
1974 if (left + 20 > WRITE_BUFFER_SIZE) {
1975 if (write_in_full(fd, write_buffer, left) < 0)
1976 return -1;
1977 left = 0;
1978 }
1979
1980 /* Append the SHA1 signature at the end */
1981 git_SHA1_Final(write_buffer + left, context);
1982 hashcpy(sha1, write_buffer + left);
1983 left += 20;
1984 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
1985 }
1986
1987 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1988 {
1989 /*
1990 * The only thing we care about in this function is to smudge the
1991 * falsely clean entry due to touch-update-touch race, so we leave
1992 * everything else as they are. We are called for entries whose
1993 * ce_stat_data.sd_mtime match the index file mtime.
1994 *
1995 * Note that this actually does not do much for gitlinks, for
1996 * which ce_match_stat_basic() always goes to the actual
1997 * contents. The caller checks with is_racy_timestamp() which
1998 * always says "no" for gitlinks, so we are not called for them ;-)
1999 */
2000 struct stat st;
2001
2002 if (lstat(ce->name, &st) < 0)
2003 return;
2004 if (ce_match_stat_basic(ce, &st))
2005 return;
2006 if (ce_modified_check_fs(ce, &st)) {
2007 /* This is "racily clean"; smudge it. Note that this
2008 * is a tricky code. At first glance, it may appear
2009 * that it can break with this sequence:
2010 *
2011 * $ echo xyzzy >frotz
2012 * $ git-update-index --add frotz
2013 * $ : >frotz
2014 * $ sleep 3
2015 * $ echo filfre >nitfol
2016 * $ git-update-index --add nitfol
2017 *
2018 * but it does not. When the second update-index runs,
2019 * it notices that the entry "frotz" has the same timestamp
2020 * as index, and if we were to smudge it by resetting its
2021 * size to zero here, then the object name recorded
2022 * in index is the 6-byte file but the cached stat information
2023 * becomes zero --- which would then match what we would
2024 * obtain from the filesystem next time we stat("frotz").
2025 *
2026 * However, the second update-index, before calling
2027 * this function, notices that the cached size is 6
2028 * bytes and what is on the filesystem is an empty
2029 * file, and never calls us, so the cached size information
2030 * for "frotz" stays 6 which does not match the filesystem.
2031 */
2032 ce->ce_stat_data.sd_size = 0;
2033 }
2034 }
2035
2036 /* Copy miscellaneous fields but not the name */
2037 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2038 struct cache_entry *ce)
2039 {
2040 short flags;
2041
2042 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2043 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2044 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2045 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2046 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2047 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2048 ondisk->mode = htonl(ce->ce_mode);
2049 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2050 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2051 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2052 hashcpy(ondisk->sha1, ce->oid.hash);
2053
2054 flags = ce->ce_flags & ~CE_NAMEMASK;
2055 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2056 ondisk->flags = htons(flags);
2057 if (ce->ce_flags & CE_EXTENDED) {
2058 struct ondisk_cache_entry_extended *ondisk2;
2059 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2060 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2061 }
2062 }
2063
2064 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
2065 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2066 {
2067 int size;
2068 int saved_namelen = saved_namelen; /* compiler workaround */
2069 int result;
2070 static unsigned char padding[8] = { 0x00 };
2071
2072 if (ce->ce_flags & CE_STRIP_NAME) {
2073 saved_namelen = ce_namelen(ce);
2074 ce->ce_namelen = 0;
2075 }
2076
2077 if (ce->ce_flags & CE_EXTENDED)
2078 size = offsetof(struct ondisk_cache_entry_extended, name);
2079 else
2080 size = offsetof(struct ondisk_cache_entry, name);
2081
2082 if (!previous_name) {
2083 int len = ce_namelen(ce);
2084 copy_cache_entry_to_ondisk(ondisk, ce);
2085 result = ce_write(c, fd, ondisk, size);
2086 if (!result)
2087 result = ce_write(c, fd, ce->name, len);
2088 if (!result)
2089 result = ce_write(c, fd, padding, align_padding_size(size, len));
2090 } else {
2091 int common, to_remove, prefix_size;
2092 unsigned char to_remove_vi[16];
2093 for (common = 0;
2094 (ce->name[common] &&
2095 common < previous_name->len &&
2096 ce->name[common] == previous_name->buf[common]);
2097 common++)
2098 ; /* still matching */
2099 to_remove = previous_name->len - common;
2100 prefix_size = encode_varint(to_remove, to_remove_vi);
2101
2102 copy_cache_entry_to_ondisk(ondisk, ce);
2103 result = ce_write(c, fd, ondisk, size);
2104 if (!result)
2105 result = ce_write(c, fd, to_remove_vi, prefix_size);
2106 if (!result)
2107 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2108 if (!result)
2109 result = ce_write(c, fd, padding, 1);
2110
2111 strbuf_splice(previous_name, common, to_remove,
2112 ce->name + common, ce_namelen(ce) - common);
2113 }
2114 if (ce->ce_flags & CE_STRIP_NAME) {
2115 ce->ce_namelen = saved_namelen;
2116 ce->ce_flags &= ~CE_STRIP_NAME;
2117 }
2118
2119 return result;
2120 }
2121
2122 /*
2123 * This function verifies if index_state has the correct sha1 of the
2124 * index file. Don't die if we have any other failure, just return 0.
2125 */
2126 static int verify_index_from(const struct index_state *istate, const char *path)
2127 {
2128 int fd;
2129 ssize_t n;
2130 struct stat st;
2131 unsigned char sha1[20];
2132
2133 if (!istate->initialized)
2134 return 0;
2135
2136 fd = open(path, O_RDONLY);
2137 if (fd < 0)
2138 return 0;
2139
2140 if (fstat(fd, &st))
2141 goto out;
2142
2143 if (st.st_size < sizeof(struct cache_header) + 20)
2144 goto out;
2145
2146 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2147 if (n != 20)
2148 goto out;
2149
2150 if (hashcmp(istate->sha1, sha1))
2151 goto out;
2152
2153 close(fd);
2154 return 1;
2155
2156 out:
2157 close(fd);
2158 return 0;
2159 }
2160
2161 static int verify_index(const struct index_state *istate)
2162 {
2163 return verify_index_from(istate, get_index_file());
2164 }
2165
2166 static int has_racy_timestamp(struct index_state *istate)
2167 {
2168 int entries = istate->cache_nr;
2169 int i;
2170
2171 for (i = 0; i < entries; i++) {
2172 struct cache_entry *ce = istate->cache[i];
2173 if (is_racy_timestamp(istate, ce))
2174 return 1;
2175 }
2176 return 0;
2177 }
2178
2179 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2180 {
2181 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2182 verify_index(istate))
2183 write_locked_index(istate, lockfile, COMMIT_LOCK);
2184 else
2185 rollback_lock_file(lockfile);
2186 }
2187
2188 /*
2189 * On success, `tempfile` is closed. If it is the temporary file
2190 * of a `struct lock_file`, we will therefore effectively perform
2191 * a 'close_lock_file_gently()`. Since that is an implementation
2192 * detail of lockfiles, callers of `do_write_index()` should not
2193 * rely on it.
2194 */
2195 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2196 int strip_extensions)
2197 {
2198 int newfd = tempfile->fd;
2199 git_SHA_CTX c;
2200 struct cache_header hdr;
2201 int i, err = 0, removed, extended, hdr_version;
2202 struct cache_entry **cache = istate->cache;
2203 int entries = istate->cache_nr;
2204 struct stat st;
2205 struct ondisk_cache_entry_extended ondisk;
2206 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2207 int drop_cache_tree = 0;
2208
2209 for (i = removed = extended = 0; i < entries; i++) {
2210 if (cache[i]->ce_flags & CE_REMOVE)
2211 removed++;
2212
2213 /* reduce extended entries if possible */
2214 cache[i]->ce_flags &= ~CE_EXTENDED;
2215 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2216 extended++;
2217 cache[i]->ce_flags |= CE_EXTENDED;
2218 }
2219 }
2220
2221 if (!istate->version) {
2222 istate->version = get_index_format_default();
2223 if (getenv("GIT_TEST_SPLIT_INDEX"))
2224 init_split_index(istate);
2225 }
2226
2227 /* demote version 3 to version 2 when the latter suffices */
2228 if (istate->version == 3 || istate->version == 2)
2229 istate->version = extended ? 3 : 2;
2230
2231 hdr_version = istate->version;
2232
2233 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2234 hdr.hdr_version = htonl(hdr_version);
2235 hdr.hdr_entries = htonl(entries - removed);
2236
2237 git_SHA1_Init(&c);
2238 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2239 return -1;
2240
2241 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2242
2243 for (i = 0; i < entries; i++) {
2244 struct cache_entry *ce = cache[i];
2245 if (ce->ce_flags & CE_REMOVE)
2246 continue;
2247 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2248 ce_smudge_racily_clean_entry(ce);
2249 if (is_null_oid(&ce->oid)) {
2250 static const char msg[] = "cache entry has null sha1: %s";
2251 static int allow = -1;
2252
2253 if (allow < 0)
2254 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2255 if (allow)
2256 warning(msg, ce->name);
2257 else
2258 err = error(msg, ce->name);
2259
2260 drop_cache_tree = 1;
2261 }
2262 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2263 err = -1;
2264
2265 if (err)
2266 break;
2267 }
2268 strbuf_release(&previous_name_buf);
2269
2270 if (err)
2271 return err;
2272
2273 /* Write extension data here */
2274 if (!strip_extensions && istate->split_index) {
2275 struct strbuf sb = STRBUF_INIT;
2276
2277 err = write_link_extension(&sb, istate) < 0 ||
2278 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2279 sb.len) < 0 ||
2280 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2281 strbuf_release(&sb);
2282 if (err)
2283 return -1;
2284 }
2285 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2286 struct strbuf sb = STRBUF_INIT;
2287
2288 cache_tree_write(&sb, istate->cache_tree);
2289 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2290 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2291 strbuf_release(&sb);
2292 if (err)
2293 return -1;
2294 }
2295 if (!strip_extensions && istate->resolve_undo) {
2296 struct strbuf sb = STRBUF_INIT;
2297
2298 resolve_undo_write(&sb, istate->resolve_undo);
2299 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2300 sb.len) < 0
2301 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2302 strbuf_release(&sb);
2303 if (err)
2304 return -1;
2305 }
2306 if (!strip_extensions && istate->untracked) {
2307 struct strbuf sb = STRBUF_INIT;
2308
2309 write_untracked_extension(&sb, istate->untracked);
2310 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2311 sb.len) < 0 ||
2312 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2313 strbuf_release(&sb);
2314 if (err)
2315 return -1;
2316 }
2317
2318 if (ce_flush(&c, newfd, istate->sha1))
2319 return -1;
2320 if (close_tempfile_gently(tempfile)) {
2321 error(_("could not close '%s'"), tempfile->filename.buf);
2322 return -1;
2323 }
2324 if (stat(tempfile->filename.buf, &st))
2325 return -1;
2326 istate->timestamp.sec = (unsigned int)st.st_mtime;
2327 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2328 return 0;
2329 }
2330
2331 void set_alternate_index_output(const char *name)
2332 {
2333 alternate_index_output = name;
2334 }
2335
2336 static int commit_locked_index(struct lock_file *lk)
2337 {
2338 if (alternate_index_output)
2339 return commit_lock_file_to(lk, alternate_index_output);
2340 else
2341 return commit_lock_file(lk);
2342 }
2343
2344 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2345 unsigned flags)
2346 {
2347 int ret = do_write_index(istate, lock->tempfile, 0);
2348 if (ret)
2349 return ret;
2350 if (flags & COMMIT_LOCK)
2351 return commit_locked_index(lock);
2352 return close_lock_file_gently(lock);
2353 }
2354
2355 static int write_split_index(struct index_state *istate,
2356 struct lock_file *lock,
2357 unsigned flags)
2358 {
2359 int ret;
2360 prepare_to_write_split_index(istate);
2361 ret = do_write_locked_index(istate, lock, flags);
2362 finish_writing_split_index(istate);
2363 return ret;
2364 }
2365
2366 static const char *shared_index_expire = "2.weeks.ago";
2367
2368 static unsigned long get_shared_index_expire_date(void)
2369 {
2370 static unsigned long shared_index_expire_date;
2371 static int shared_index_expire_date_prepared;
2372
2373 if (!shared_index_expire_date_prepared) {
2374 git_config_get_expiry("splitindex.sharedindexexpire",
2375 &shared_index_expire);
2376 shared_index_expire_date = approxidate(shared_index_expire);
2377 shared_index_expire_date_prepared = 1;
2378 }
2379
2380 return shared_index_expire_date;
2381 }
2382
2383 static int should_delete_shared_index(const char *shared_index_path)
2384 {
2385 struct stat st;
2386 unsigned long expiration;
2387
2388 /* Check timestamp */
2389 expiration = get_shared_index_expire_date();
2390 if (!expiration)
2391 return 0;
2392 if (stat(shared_index_path, &st))
2393 return error_errno(_("could not stat '%s'"), shared_index_path);
2394 if (st.st_mtime > expiration)
2395 return 0;
2396
2397 return 1;
2398 }
2399
2400 static int clean_shared_index_files(const char *current_hex)
2401 {
2402 struct dirent *de;
2403 DIR *dir = opendir(get_git_dir());
2404
2405 if (!dir)
2406 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2407
2408 while ((de = readdir(dir)) != NULL) {
2409 const char *sha1_hex;
2410 const char *shared_index_path;
2411 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2412 continue;
2413 if (!strcmp(sha1_hex, current_hex))
2414 continue;
2415 shared_index_path = git_path("%s", de->d_name);
2416 if (should_delete_shared_index(shared_index_path) > 0 &&
2417 unlink(shared_index_path))
2418 warning_errno(_("unable to unlink: %s"), shared_index_path);
2419 }
2420 closedir(dir);
2421
2422 return 0;
2423 }
2424
2425 static int write_shared_index(struct index_state *istate,
2426 struct lock_file *lock, unsigned flags)
2427 {
2428 struct tempfile *temp;
2429 struct split_index *si = istate->split_index;
2430 int ret;
2431
2432 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2433 if (!temp) {
2434 hashclr(si->base_sha1);
2435 return do_write_locked_index(istate, lock, flags);
2436 }
2437 move_cache_to_base_index(istate);
2438 ret = do_write_index(si->base, temp, 1);
2439 if (ret) {
2440 delete_tempfile(&temp);
2441 return ret;
2442 }
2443 ret = adjust_shared_perm(get_tempfile_path(temp));
2444 if (ret) {
2445 int save_errno = errno;
2446 error("cannot fix permission bits on %s", get_tempfile_path(temp));
2447 delete_tempfile(&temp);
2448 errno = save_errno;
2449 return ret;
2450 }
2451 ret = rename_tempfile(&temp,
2452 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2453 if (!ret) {
2454 hashcpy(si->base_sha1, si->base->sha1);
2455 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2456 }
2457
2458 return ret;
2459 }
2460
2461 static const int default_max_percent_split_change = 20;
2462
2463 static int too_many_not_shared_entries(struct index_state *istate)
2464 {
2465 int i, not_shared = 0;
2466 int max_split = git_config_get_max_percent_split_change();
2467
2468 switch (max_split) {
2469 case -1:
2470 /* not or badly configured: use the default value */
2471 max_split = default_max_percent_split_change;
2472 break;
2473 case 0:
2474 return 1; /* 0% means always write a new shared index */
2475 case 100:
2476 return 0; /* 100% means never write a new shared index */
2477 default:
2478 break; /* just use the configured value */
2479 }
2480
2481 /* Count not shared entries */
2482 for (i = 0; i < istate->cache_nr; i++) {
2483 struct cache_entry *ce = istate->cache[i];
2484 if (!ce->index)
2485 not_shared++;
2486 }
2487
2488 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2489 }
2490
2491 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2492 unsigned flags)
2493 {
2494 int new_shared_index, ret;
2495 struct split_index *si = istate->split_index;
2496
2497 if (!si || alternate_index_output ||
2498 (istate->cache_changed & ~EXTMASK)) {
2499 if (si)
2500 hashclr(si->base_sha1);
2501 ret = do_write_locked_index(istate, lock, flags);
2502 goto out;
2503 }
2504
2505 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2506 int v = si->base_sha1[0];
2507 if ((v & 15) < 6)
2508 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2509 }
2510 if (too_many_not_shared_entries(istate))
2511 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2512
2513 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2514
2515 if (new_shared_index) {
2516 ret = write_shared_index(istate, lock, flags);
2517 if (ret)
2518 goto out;
2519 }
2520
2521 ret = write_split_index(istate, lock, flags);
2522
2523 /* Freshen the shared index only if the split-index was written */
2524 if (!ret && !new_shared_index)
2525 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2526
2527 out:
2528 if (flags & COMMIT_LOCK)
2529 rollback_lock_file(lock);
2530 return ret;
2531 }
2532
2533 /*
2534 * Read the index file that is potentially unmerged into given
2535 * index_state, dropping any unmerged entries. Returns true if
2536 * the index is unmerged. Callers who want to refuse to work
2537 * from an unmerged state can call this and check its return value,
2538 * instead of calling read_cache().
2539 */
2540 int read_index_unmerged(struct index_state *istate)
2541 {
2542 int i;
2543 int unmerged = 0;
2544
2545 read_index(istate);
2546 for (i = 0; i < istate->cache_nr; i++) {
2547 struct cache_entry *ce = istate->cache[i];
2548 struct cache_entry *new_ce;
2549 int size, len;
2550
2551 if (!ce_stage(ce))
2552 continue;
2553 unmerged = 1;
2554 len = ce_namelen(ce);
2555 size = cache_entry_size(len);
2556 new_ce = xcalloc(1, size);
2557 memcpy(new_ce->name, ce->name, len);
2558 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2559 new_ce->ce_namelen = len;
2560 new_ce->ce_mode = ce->ce_mode;
2561 if (add_index_entry(istate, new_ce, 0))
2562 return error("%s: cannot drop to stage #0",
2563 new_ce->name);
2564 }
2565 return unmerged;
2566 }
2567
2568 /*
2569 * Returns 1 if the path is an "other" path with respect to
2570 * the index; that is, the path is not mentioned in the index at all,
2571 * either as a file, a directory with some files in the index,
2572 * or as an unmerged entry.
2573 *
2574 * We helpfully remove a trailing "/" from directories so that
2575 * the output of read_directory can be used as-is.
2576 */
2577 int index_name_is_other(const struct index_state *istate, const char *name,
2578 int namelen)
2579 {
2580 int pos;
2581 if (namelen && name[namelen - 1] == '/')
2582 namelen--;
2583 pos = index_name_pos(istate, name, namelen);
2584 if (0 <= pos)
2585 return 0; /* exact match */
2586 pos = -pos - 1;
2587 if (pos < istate->cache_nr) {
2588 struct cache_entry *ce = istate->cache[pos];
2589 if (ce_namelen(ce) == namelen &&
2590 !memcmp(ce->name, name, namelen))
2591 return 0; /* Yup, this one exists unmerged */
2592 }
2593 return 1;
2594 }
2595
2596 void *read_blob_data_from_index(const struct index_state *istate,
2597 const char *path, unsigned long *size)
2598 {
2599 int pos, len;
2600 unsigned long sz;
2601 enum object_type type;
2602 void *data;
2603
2604 len = strlen(path);
2605 pos = index_name_pos(istate, path, len);
2606 if (pos < 0) {
2607 /*
2608 * We might be in the middle of a merge, in which
2609 * case we would read stage #2 (ours).
2610 */
2611 int i;
2612 for (i = -pos - 1;
2613 (pos < 0 && i < istate->cache_nr &&
2614 !strcmp(istate->cache[i]->name, path));
2615 i++)
2616 if (ce_stage(istate->cache[i]) == 2)
2617 pos = i;
2618 }
2619 if (pos < 0)
2620 return NULL;
2621 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2622 if (!data || type != OBJ_BLOB) {
2623 free(data);
2624 return NULL;
2625 }
2626 if (size)
2627 *size = sz;
2628 return data;
2629 }
2630
2631 void stat_validity_clear(struct stat_validity *sv)
2632 {
2633 FREE_AND_NULL(sv->sd);
2634 }
2635
2636 int stat_validity_check(struct stat_validity *sv, const char *path)
2637 {
2638 struct stat st;
2639
2640 if (stat(path, &st) < 0)
2641 return sv->sd == NULL;
2642 if (!sv->sd)
2643 return 0;
2644 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2645 }
2646
2647 void stat_validity_update(struct stat_validity *sv, int fd)
2648 {
2649 struct stat st;
2650
2651 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2652 stat_validity_clear(sv);
2653 else {
2654 if (!sv->sd)
2655 sv->sd = xcalloc(1, sizeof(struct stat_data));
2656 fill_stat_data(sv->sd, &st);
2657 }
2658 }
2659
2660 void move_index_extensions(struct index_state *dst, struct index_state *src)
2661 {
2662 dst->untracked = src->untracked;
2663 src->untracked = NULL;
2664 }