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