archive.c: make archiver static
[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 "cache-tree.h"
9 #include "refs.h"
10 #include "dir.h"
11 #include "tree.h"
12 #include "commit.h"
13 #include "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16
17 /* Index extensions.
18 *
19 * The first letter should be 'A'..'Z' for extensions that are not
20 * necessary for a correct operation (i.e. optimization data).
21 * When new extensions are added that _needs_ to be understood in
22 * order to correctly interpret the index file, pick character that
23 * is outside the range, to cause the reader to abort.
24 */
25
26 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
27 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
28
29 struct index_state the_index;
30
31 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
32 {
33 istate->cache[nr] = ce;
34 add_name_hash(istate, ce);
35 }
36
37 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
38 {
39 struct cache_entry *old = istate->cache[nr];
40
41 remove_name_hash(old);
42 set_index_entry(istate, nr, ce);
43 istate->cache_changed = 1;
44 }
45
46 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
47 {
48 struct cache_entry *old = istate->cache[nr], *new;
49 int namelen = strlen(new_name);
50
51 new = xmalloc(cache_entry_size(namelen));
52 copy_cache_entry(new, old);
53 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
54 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
55 memcpy(new->name, new_name, namelen + 1);
56
57 cache_tree_invalidate_path(istate->cache_tree, old->name);
58 remove_index_entry_at(istate, nr);
59 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
60 }
61
62 /*
63 * This only updates the "non-critical" parts of the directory
64 * cache, ie the parts that aren't tracked by GIT, and only used
65 * to validate the cache.
66 */
67 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
68 {
69 ce->ce_ctime = st->st_ctime;
70 ce->ce_mtime = st->st_mtime;
71 ce->ce_dev = st->st_dev;
72 ce->ce_ino = st->st_ino;
73 ce->ce_uid = st->st_uid;
74 ce->ce_gid = st->st_gid;
75 ce->ce_size = st->st_size;
76
77 if (assume_unchanged)
78 ce->ce_flags |= CE_VALID;
79
80 if (S_ISREG(st->st_mode))
81 ce_mark_uptodate(ce);
82 }
83
84 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
85 {
86 int match = -1;
87 int fd = open(ce->name, O_RDONLY);
88
89 if (fd >= 0) {
90 unsigned char sha1[20];
91 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
92 match = hashcmp(sha1, ce->sha1);
93 /* index_fd() closed the file descriptor already */
94 }
95 return match;
96 }
97
98 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
99 {
100 int match = -1;
101 char *target;
102 void *buffer;
103 unsigned long size;
104 enum object_type type;
105 int len;
106
107 target = xmalloc(expected_size);
108 len = readlink(ce->name, target, expected_size);
109 if (len != expected_size) {
110 free(target);
111 return -1;
112 }
113 buffer = read_sha1_file(ce->sha1, &type, &size);
114 if (!buffer) {
115 free(target);
116 return -1;
117 }
118 if (size == expected_size)
119 match = memcmp(buffer, target, size);
120 free(buffer);
121 free(target);
122 return match;
123 }
124
125 static int ce_compare_gitlink(struct cache_entry *ce)
126 {
127 unsigned char sha1[20];
128
129 /*
130 * We don't actually require that the .git directory
131 * under GITLINK directory be a valid git directory. It
132 * might even be missing (in case nobody populated that
133 * sub-project).
134 *
135 * If so, we consider it always to match.
136 */
137 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
138 return 0;
139 return hashcmp(sha1, ce->sha1);
140 }
141
142 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
143 {
144 switch (st->st_mode & S_IFMT) {
145 case S_IFREG:
146 if (ce_compare_data(ce, st))
147 return DATA_CHANGED;
148 break;
149 case S_IFLNK:
150 if (ce_compare_link(ce, xsize_t(st->st_size)))
151 return DATA_CHANGED;
152 break;
153 case S_IFDIR:
154 if (S_ISGITLINK(ce->ce_mode))
155 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
156 default:
157 return TYPE_CHANGED;
158 }
159 return 0;
160 }
161
162 static int is_empty_blob_sha1(const unsigned char *sha1)
163 {
164 static const unsigned char empty_blob_sha1[20] = {
165 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
166 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
167 };
168
169 return !hashcmp(sha1, empty_blob_sha1);
170 }
171
172 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
173 {
174 unsigned int changed = 0;
175
176 if (ce->ce_flags & CE_REMOVE)
177 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
178
179 switch (ce->ce_mode & S_IFMT) {
180 case S_IFREG:
181 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
182 /* We consider only the owner x bit to be relevant for
183 * "mode changes"
184 */
185 if (trust_executable_bit &&
186 (0100 & (ce->ce_mode ^ st->st_mode)))
187 changed |= MODE_CHANGED;
188 break;
189 case S_IFLNK:
190 if (!S_ISLNK(st->st_mode) &&
191 (has_symlinks || !S_ISREG(st->st_mode)))
192 changed |= TYPE_CHANGED;
193 break;
194 case S_IFGITLINK:
195 /* We ignore most of the st_xxx fields for gitlinks */
196 if (!S_ISDIR(st->st_mode))
197 changed |= TYPE_CHANGED;
198 else if (ce_compare_gitlink(ce))
199 changed |= DATA_CHANGED;
200 return changed;
201 default:
202 die("internal error: ce_mode is %o", ce->ce_mode);
203 }
204 if (ce->ce_mtime != (unsigned int) st->st_mtime)
205 changed |= MTIME_CHANGED;
206 if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime)
207 changed |= CTIME_CHANGED;
208
209 if (ce->ce_uid != (unsigned int) st->st_uid ||
210 ce->ce_gid != (unsigned int) st->st_gid)
211 changed |= OWNER_CHANGED;
212 if (ce->ce_ino != (unsigned int) st->st_ino)
213 changed |= INODE_CHANGED;
214
215 #ifdef USE_STDEV
216 /*
217 * st_dev breaks on network filesystems where different
218 * clients will have different views of what "device"
219 * the filesystem is on
220 */
221 if (ce->ce_dev != (unsigned int) st->st_dev)
222 changed |= INODE_CHANGED;
223 #endif
224
225 if (ce->ce_size != (unsigned int) st->st_size)
226 changed |= DATA_CHANGED;
227
228 /* Racily smudged entry? */
229 if (!ce->ce_size) {
230 if (!is_empty_blob_sha1(ce->sha1))
231 changed |= DATA_CHANGED;
232 }
233
234 return changed;
235 }
236
237 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
238 {
239 return (!S_ISGITLINK(ce->ce_mode) &&
240 istate->timestamp &&
241 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
242 }
243
244 int ie_match_stat(const struct index_state *istate,
245 struct cache_entry *ce, struct stat *st,
246 unsigned int options)
247 {
248 unsigned int changed;
249 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
250 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
251
252 /*
253 * If it's marked as always valid in the index, it's
254 * valid whatever the checked-out copy says.
255 */
256 if (!ignore_valid && (ce->ce_flags & CE_VALID))
257 return 0;
258
259 changed = ce_match_stat_basic(ce, st);
260
261 /*
262 * Within 1 second of this sequence:
263 * echo xyzzy >file && git-update-index --add file
264 * running this command:
265 * echo frotz >file
266 * would give a falsely clean cache entry. The mtime and
267 * length match the cache, and other stat fields do not change.
268 *
269 * We could detect this at update-index time (the cache entry
270 * being registered/updated records the same time as "now")
271 * and delay the return from git-update-index, but that would
272 * effectively mean we can make at most one commit per second,
273 * which is not acceptable. Instead, we check cache entries
274 * whose mtime are the same as the index file timestamp more
275 * carefully than others.
276 */
277 if (!changed && is_racy_timestamp(istate, ce)) {
278 if (assume_racy_is_modified)
279 changed |= DATA_CHANGED;
280 else
281 changed |= ce_modified_check_fs(ce, st);
282 }
283
284 return changed;
285 }
286
287 int ie_modified(const struct index_state *istate,
288 struct cache_entry *ce, struct stat *st, unsigned int options)
289 {
290 int changed, changed_fs;
291
292 changed = ie_match_stat(istate, ce, st, options);
293 if (!changed)
294 return 0;
295 /*
296 * If the mode or type has changed, there's no point in trying
297 * to refresh the entry - it's not going to match
298 */
299 if (changed & (MODE_CHANGED | TYPE_CHANGED))
300 return changed;
301
302 /*
303 * Immediately after read-tree or update-index --cacheinfo,
304 * the length field is zero, as we have never even read the
305 * lstat(2) information once, and we cannot trust DATA_CHANGED
306 * returned by ie_match_stat() which in turn was returned by
307 * ce_match_stat_basic() to signal that the filesize of the
308 * blob changed. We have to actually go to the filesystem to
309 * see if the contents match, and if so, should answer "unchanged".
310 *
311 * The logic does not apply to gitlinks, as ce_match_stat_basic()
312 * already has checked the actual HEAD from the filesystem in the
313 * subproject. If ie_match_stat() already said it is different,
314 * then we know it is.
315 */
316 if ((changed & DATA_CHANGED) &&
317 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
318 return changed;
319
320 changed_fs = ce_modified_check_fs(ce, st);
321 if (changed_fs)
322 return changed | changed_fs;
323 return 0;
324 }
325
326 int base_name_compare(const char *name1, int len1, int mode1,
327 const char *name2, int len2, int mode2)
328 {
329 unsigned char c1, c2;
330 int len = len1 < len2 ? len1 : len2;
331 int cmp;
332
333 cmp = memcmp(name1, name2, len);
334 if (cmp)
335 return cmp;
336 c1 = name1[len];
337 c2 = name2[len];
338 if (!c1 && S_ISDIR(mode1))
339 c1 = '/';
340 if (!c2 && S_ISDIR(mode2))
341 c2 = '/';
342 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
343 }
344
345 /*
346 * df_name_compare() is identical to base_name_compare(), except it
347 * compares conflicting directory/file entries as equal. Note that
348 * while a directory name compares as equal to a regular file, they
349 * then individually compare _differently_ to a filename that has
350 * a dot after the basename (because '\0' < '.' < '/').
351 *
352 * This is used by routines that want to traverse the git namespace
353 * but then handle conflicting entries together when possible.
354 */
355 int df_name_compare(const char *name1, int len1, int mode1,
356 const char *name2, int len2, int mode2)
357 {
358 int len = len1 < len2 ? len1 : len2, cmp;
359 unsigned char c1, c2;
360
361 cmp = memcmp(name1, name2, len);
362 if (cmp)
363 return cmp;
364 /* Directories and files compare equal (same length, same name) */
365 if (len1 == len2)
366 return 0;
367 c1 = name1[len];
368 if (!c1 && S_ISDIR(mode1))
369 c1 = '/';
370 c2 = name2[len];
371 if (!c2 && S_ISDIR(mode2))
372 c2 = '/';
373 if (c1 == '/' && !c2)
374 return 0;
375 if (c2 == '/' && !c1)
376 return 0;
377 return c1 - c2;
378 }
379
380 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
381 {
382 int len1 = flags1 & CE_NAMEMASK;
383 int len2 = flags2 & CE_NAMEMASK;
384 int len = len1 < len2 ? len1 : len2;
385 int cmp;
386
387 cmp = memcmp(name1, name2, len);
388 if (cmp)
389 return cmp;
390 if (len1 < len2)
391 return -1;
392 if (len1 > len2)
393 return 1;
394
395 /* Compare stages */
396 flags1 &= CE_STAGEMASK;
397 flags2 &= CE_STAGEMASK;
398
399 if (flags1 < flags2)
400 return -1;
401 if (flags1 > flags2)
402 return 1;
403 return 0;
404 }
405
406 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
407 {
408 int first, last;
409
410 first = 0;
411 last = istate->cache_nr;
412 while (last > first) {
413 int next = (last + first) >> 1;
414 struct cache_entry *ce = istate->cache[next];
415 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
416 if (!cmp)
417 return next;
418 if (cmp < 0) {
419 last = next;
420 continue;
421 }
422 first = next+1;
423 }
424 return -first-1;
425 }
426
427 /* Remove entry, return true if there are more entries to go.. */
428 int remove_index_entry_at(struct index_state *istate, int pos)
429 {
430 struct cache_entry *ce = istate->cache[pos];
431
432 remove_name_hash(ce);
433 istate->cache_changed = 1;
434 istate->cache_nr--;
435 if (pos >= istate->cache_nr)
436 return 0;
437 memmove(istate->cache + pos,
438 istate->cache + pos + 1,
439 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
440 return 1;
441 }
442
443 int remove_file_from_index(struct index_state *istate, const char *path)
444 {
445 int pos = index_name_pos(istate, path, strlen(path));
446 if (pos < 0)
447 pos = -pos-1;
448 cache_tree_invalidate_path(istate->cache_tree, path);
449 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
450 remove_index_entry_at(istate, pos);
451 return 0;
452 }
453
454 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
455 {
456 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
457 }
458
459 static int index_name_pos_also_unmerged(struct index_state *istate,
460 const char *path, int namelen)
461 {
462 int pos = index_name_pos(istate, path, namelen);
463 struct cache_entry *ce;
464
465 if (pos >= 0)
466 return pos;
467
468 /* maybe unmerged? */
469 pos = -1 - pos;
470 if (pos >= istate->cache_nr ||
471 compare_name((ce = istate->cache[pos]), path, namelen))
472 return -1;
473
474 /* order of preference: stage 2, 1, 3 */
475 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
476 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
477 !compare_name(ce, path, namelen))
478 pos++;
479 return pos;
480 }
481
482 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
483 {
484 int len = ce_namelen(ce);
485 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
486 }
487
488 /*
489 * If we add a filename that aliases in the cache, we will use the
490 * name that we already have - but we don't want to update the same
491 * alias twice, because that implies that there were actually two
492 * different files with aliasing names!
493 *
494 * So we use the CE_ADDED flag to verify that the alias was an old
495 * one before we accept it as
496 */
497 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
498 {
499 int len;
500 struct cache_entry *new;
501
502 if (alias->ce_flags & CE_ADDED)
503 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
504
505 /* Ok, create the new entry using the name of the existing alias */
506 len = ce_namelen(alias);
507 new = xcalloc(1, cache_entry_size(len));
508 memcpy(new->name, alias->name, len);
509 copy_cache_entry(new, ce);
510 free(ce);
511 return new;
512 }
513
514 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
515 {
516 int size, namelen, was_same;
517 mode_t st_mode = st->st_mode;
518 struct cache_entry *ce, *alias;
519 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
520 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
521 int pretend = flags & ADD_CACHE_PRETEND;
522
523 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
524 return error("%s: can only add regular files, symbolic links or git-directories", path);
525
526 namelen = strlen(path);
527 if (S_ISDIR(st_mode)) {
528 while (namelen && path[namelen-1] == '/')
529 namelen--;
530 }
531 size = cache_entry_size(namelen);
532 ce = xcalloc(1, size);
533 memcpy(ce->name, path, namelen);
534 ce->ce_flags = namelen;
535 fill_stat_cache_info(ce, st);
536
537 if (trust_executable_bit && has_symlinks)
538 ce->ce_mode = create_ce_mode(st_mode);
539 else {
540 /* If there is an existing entry, pick the mode bits and type
541 * from it, otherwise assume unexecutable regular file.
542 */
543 struct cache_entry *ent;
544 int pos = index_name_pos_also_unmerged(istate, path, namelen);
545
546 ent = (0 <= pos) ? istate->cache[pos] : NULL;
547 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
548 }
549
550 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
551 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
552 /* Nothing changed, really */
553 free(ce);
554 ce_mark_uptodate(alias);
555 alias->ce_flags |= CE_ADDED;
556 return 0;
557 }
558 if (index_path(ce->sha1, path, st, 1))
559 return error("unable to index file %s", path);
560 if (ignore_case && alias && different_name(ce, alias))
561 ce = create_alias_ce(ce, alias);
562 ce->ce_flags |= CE_ADDED;
563
564 /* It was suspected to be racily clean, but it turns out to be Ok */
565 was_same = (alias &&
566 !ce_stage(alias) &&
567 !hashcmp(alias->sha1, ce->sha1) &&
568 ce->ce_mode == alias->ce_mode);
569
570 if (pretend)
571 ;
572 else if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
573 return error("unable to add %s to index",path);
574 if (verbose && !was_same)
575 printf("add '%s'\n", path);
576 return 0;
577 }
578
579 int add_file_to_index(struct index_state *istate, const char *path, int flags)
580 {
581 struct stat st;
582 if (lstat(path, &st))
583 die("%s: unable to stat (%s)", path, strerror(errno));
584 return add_to_index(istate, path, &st, flags);
585 }
586
587 struct cache_entry *make_cache_entry(unsigned int mode,
588 const unsigned char *sha1, const char *path, int stage,
589 int refresh)
590 {
591 int size, len;
592 struct cache_entry *ce;
593
594 if (!verify_path(path))
595 return NULL;
596
597 len = strlen(path);
598 size = cache_entry_size(len);
599 ce = xcalloc(1, size);
600
601 hashcpy(ce->sha1, sha1);
602 memcpy(ce->name, path, len);
603 ce->ce_flags = create_ce_flags(len, stage);
604 ce->ce_mode = create_ce_mode(mode);
605
606 if (refresh)
607 return refresh_cache_entry(ce, 0);
608
609 return ce;
610 }
611
612 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
613 {
614 int len = ce_namelen(a);
615 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
616 }
617
618 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
619 {
620 const char *match, *name;
621 int len;
622
623 if (!pathspec)
624 return 1;
625
626 len = ce_namelen(ce);
627 name = ce->name;
628 while ((match = *pathspec++) != NULL) {
629 int matchlen = strlen(match);
630 if (matchlen > len)
631 continue;
632 if (memcmp(name, match, matchlen))
633 continue;
634 if (matchlen && name[matchlen-1] == '/')
635 return 1;
636 if (name[matchlen] == '/' || !name[matchlen])
637 return 1;
638 if (!matchlen)
639 return 1;
640 }
641 return 0;
642 }
643
644 /*
645 * We fundamentally don't like some paths: we don't want
646 * dot or dot-dot anywhere, and for obvious reasons don't
647 * want to recurse into ".git" either.
648 *
649 * Also, we don't want double slashes or slashes at the
650 * end that can make pathnames ambiguous.
651 */
652 static int verify_dotfile(const char *rest)
653 {
654 /*
655 * The first character was '.', but that
656 * has already been discarded, we now test
657 * the rest.
658 */
659 switch (*rest) {
660 /* "." is not allowed */
661 case '\0': case '/':
662 return 0;
663
664 /*
665 * ".git" followed by NUL or slash is bad. This
666 * shares the path end test with the ".." case.
667 */
668 case 'g':
669 if (rest[1] != 'i')
670 break;
671 if (rest[2] != 't')
672 break;
673 rest += 2;
674 /* fallthrough */
675 case '.':
676 if (rest[1] == '\0' || rest[1] == '/')
677 return 0;
678 }
679 return 1;
680 }
681
682 int verify_path(const char *path)
683 {
684 char c;
685
686 goto inside;
687 for (;;) {
688 if (!c)
689 return 1;
690 if (c == '/') {
691 inside:
692 c = *path++;
693 switch (c) {
694 default:
695 continue;
696 case '/': case '\0':
697 break;
698 case '.':
699 if (verify_dotfile(path))
700 continue;
701 }
702 return 0;
703 }
704 c = *path++;
705 }
706 }
707
708 /*
709 * Do we have another file that has the beginning components being a
710 * proper superset of the name we're trying to add?
711 */
712 static int has_file_name(struct index_state *istate,
713 const struct cache_entry *ce, int pos, int ok_to_replace)
714 {
715 int retval = 0;
716 int len = ce_namelen(ce);
717 int stage = ce_stage(ce);
718 const char *name = ce->name;
719
720 while (pos < istate->cache_nr) {
721 struct cache_entry *p = istate->cache[pos++];
722
723 if (len >= ce_namelen(p))
724 break;
725 if (memcmp(name, p->name, len))
726 break;
727 if (ce_stage(p) != stage)
728 continue;
729 if (p->name[len] != '/')
730 continue;
731 if (p->ce_flags & CE_REMOVE)
732 continue;
733 retval = -1;
734 if (!ok_to_replace)
735 break;
736 remove_index_entry_at(istate, --pos);
737 }
738 return retval;
739 }
740
741 /*
742 * Do we have another file with a pathname that is a proper
743 * subset of the name we're trying to add?
744 */
745 static int has_dir_name(struct index_state *istate,
746 const struct cache_entry *ce, int pos, int ok_to_replace)
747 {
748 int retval = 0;
749 int stage = ce_stage(ce);
750 const char *name = ce->name;
751 const char *slash = name + ce_namelen(ce);
752
753 for (;;) {
754 int len;
755
756 for (;;) {
757 if (*--slash == '/')
758 break;
759 if (slash <= ce->name)
760 return retval;
761 }
762 len = slash - name;
763
764 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
765 if (pos >= 0) {
766 /*
767 * Found one, but not so fast. This could
768 * be a marker that says "I was here, but
769 * I am being removed". Such an entry is
770 * not a part of the resulting tree, and
771 * it is Ok to have a directory at the same
772 * path.
773 */
774 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
775 retval = -1;
776 if (!ok_to_replace)
777 break;
778 remove_index_entry_at(istate, pos);
779 continue;
780 }
781 }
782 else
783 pos = -pos-1;
784
785 /*
786 * Trivial optimization: if we find an entry that
787 * already matches the sub-directory, then we know
788 * we're ok, and we can exit.
789 */
790 while (pos < istate->cache_nr) {
791 struct cache_entry *p = istate->cache[pos];
792 if ((ce_namelen(p) <= len) ||
793 (p->name[len] != '/') ||
794 memcmp(p->name, name, len))
795 break; /* not our subdirectory */
796 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
797 /*
798 * p is at the same stage as our entry, and
799 * is a subdirectory of what we are looking
800 * at, so we cannot have conflicts at our
801 * level or anything shorter.
802 */
803 return retval;
804 pos++;
805 }
806 }
807 return retval;
808 }
809
810 /* We may be in a situation where we already have path/file and path
811 * is being added, or we already have path and path/file is being
812 * added. Either one would result in a nonsense tree that has path
813 * twice when git-write-tree tries to write it out. Prevent it.
814 *
815 * If ok-to-replace is specified, we remove the conflicting entries
816 * from the cache so the caller should recompute the insert position.
817 * When this happens, we return non-zero.
818 */
819 static int check_file_directory_conflict(struct index_state *istate,
820 const struct cache_entry *ce,
821 int pos, int ok_to_replace)
822 {
823 int retval;
824
825 /*
826 * When ce is an "I am going away" entry, we allow it to be added
827 */
828 if (ce->ce_flags & CE_REMOVE)
829 return 0;
830
831 /*
832 * We check if the path is a sub-path of a subsequent pathname
833 * first, since removing those will not change the position
834 * in the array.
835 */
836 retval = has_file_name(istate, ce, pos, ok_to_replace);
837
838 /*
839 * Then check if the path might have a clashing sub-directory
840 * before it.
841 */
842 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
843 }
844
845 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
846 {
847 int pos;
848 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
849 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
850 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
851
852 cache_tree_invalidate_path(istate->cache_tree, ce->name);
853 pos = index_name_pos(istate, ce->name, ce->ce_flags);
854
855 /* existing match? Just replace it. */
856 if (pos >= 0) {
857 replace_index_entry(istate, pos, ce);
858 return 0;
859 }
860 pos = -pos-1;
861
862 /*
863 * Inserting a merged entry ("stage 0") into the index
864 * will always replace all non-merged entries..
865 */
866 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
867 while (ce_same_name(istate->cache[pos], ce)) {
868 ok_to_add = 1;
869 if (!remove_index_entry_at(istate, pos))
870 break;
871 }
872 }
873
874 if (!ok_to_add)
875 return -1;
876 if (!verify_path(ce->name))
877 return -1;
878
879 if (!skip_df_check &&
880 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
881 if (!ok_to_replace)
882 return error("'%s' appears as both a file and as a directory",
883 ce->name);
884 pos = index_name_pos(istate, ce->name, ce->ce_flags);
885 pos = -pos-1;
886 }
887 return pos + 1;
888 }
889
890 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
891 {
892 int pos;
893
894 if (option & ADD_CACHE_JUST_APPEND)
895 pos = istate->cache_nr;
896 else {
897 int ret;
898 ret = add_index_entry_with_check(istate, ce, option);
899 if (ret <= 0)
900 return ret;
901 pos = ret - 1;
902 }
903
904 /* Make sure the array is big enough .. */
905 if (istate->cache_nr == istate->cache_alloc) {
906 istate->cache_alloc = alloc_nr(istate->cache_alloc);
907 istate->cache = xrealloc(istate->cache,
908 istate->cache_alloc * sizeof(struct cache_entry *));
909 }
910
911 /* Add it in.. */
912 istate->cache_nr++;
913 if (istate->cache_nr > pos + 1)
914 memmove(istate->cache + pos + 1,
915 istate->cache + pos,
916 (istate->cache_nr - pos - 1) * sizeof(ce));
917 set_index_entry(istate, pos, ce);
918 istate->cache_changed = 1;
919 return 0;
920 }
921
922 /*
923 * "refresh" does not calculate a new sha1 file or bring the
924 * cache up-to-date for mode/content changes. But what it
925 * _does_ do is to "re-match" the stat information of a file
926 * with the cache, so that you can refresh the cache for a
927 * file that hasn't been changed but where the stat entry is
928 * out of date.
929 *
930 * For example, you'd want to do this after doing a "git-read-tree",
931 * to link up the stat cache details with the proper files.
932 */
933 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
934 struct cache_entry *ce,
935 unsigned int options, int *err)
936 {
937 struct stat st;
938 struct cache_entry *updated;
939 int changed, size;
940 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
941
942 if (ce_uptodate(ce))
943 return ce;
944
945 /*
946 * CE_VALID means the user promised us that the change to
947 * the work tree does not matter and told us not to worry.
948 */
949 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
950 ce_mark_uptodate(ce);
951 return ce;
952 }
953
954 if (lstat(ce->name, &st) < 0) {
955 if (err)
956 *err = errno;
957 return NULL;
958 }
959
960 changed = ie_match_stat(istate, ce, &st, options);
961 if (!changed) {
962 /*
963 * The path is unchanged. If we were told to ignore
964 * valid bit, then we did the actual stat check and
965 * found that the entry is unmodified. If the entry
966 * is not marked VALID, this is the place to mark it
967 * valid again, under "assume unchanged" mode.
968 */
969 if (ignore_valid && assume_unchanged &&
970 !(ce->ce_flags & CE_VALID))
971 ; /* mark this one VALID again */
972 else {
973 /*
974 * We do not mark the index itself "modified"
975 * because CE_UPTODATE flag is in-core only;
976 * we are not going to write this change out.
977 */
978 ce_mark_uptodate(ce);
979 return ce;
980 }
981 }
982
983 if (ie_modified(istate, ce, &st, options)) {
984 if (err)
985 *err = EINVAL;
986 return NULL;
987 }
988
989 size = ce_size(ce);
990 updated = xmalloc(size);
991 memcpy(updated, ce, size);
992 fill_stat_cache_info(updated, &st);
993 /*
994 * If ignore_valid is not set, we should leave CE_VALID bit
995 * alone. Otherwise, paths marked with --no-assume-unchanged
996 * (i.e. things to be edited) will reacquire CE_VALID bit
997 * automatically, which is not really what we want.
998 */
999 if (!ignore_valid && assume_unchanged &&
1000 !(ce->ce_flags & CE_VALID))
1001 updated->ce_flags &= ~CE_VALID;
1002
1003 return updated;
1004 }
1005
1006 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
1007 {
1008 int i;
1009 int has_errors = 0;
1010 int really = (flags & REFRESH_REALLY) != 0;
1011 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1012 int quiet = (flags & REFRESH_QUIET) != 0;
1013 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1014 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1015 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1016 const char *needs_update_message;
1017
1018 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
1019 ? "locally modified" : "needs update");
1020 for (i = 0; i < istate->cache_nr; i++) {
1021 struct cache_entry *ce, *new;
1022 int cache_errno = 0;
1023
1024 ce = istate->cache[i];
1025 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1026 continue;
1027
1028 if (ce_stage(ce)) {
1029 while ((i < istate->cache_nr) &&
1030 ! strcmp(istate->cache[i]->name, ce->name))
1031 i++;
1032 i--;
1033 if (allow_unmerged)
1034 continue;
1035 printf("%s: needs merge\n", ce->name);
1036 has_errors = 1;
1037 continue;
1038 }
1039
1040 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1041 continue;
1042
1043 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1044 if (new == ce)
1045 continue;
1046 if (!new) {
1047 if (not_new && cache_errno == ENOENT)
1048 continue;
1049 if (really && cache_errno == EINVAL) {
1050 /* If we are doing --really-refresh that
1051 * means the index is not valid anymore.
1052 */
1053 ce->ce_flags &= ~CE_VALID;
1054 istate->cache_changed = 1;
1055 }
1056 if (quiet)
1057 continue;
1058 printf("%s: %s\n", ce->name, needs_update_message);
1059 has_errors = 1;
1060 continue;
1061 }
1062
1063 replace_index_entry(istate, i, new);
1064 }
1065 return has_errors;
1066 }
1067
1068 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1069 {
1070 return refresh_cache_ent(&the_index, ce, really, NULL);
1071 }
1072
1073 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1074 {
1075 SHA_CTX c;
1076 unsigned char sha1[20];
1077
1078 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1079 return error("bad signature");
1080 if (hdr->hdr_version != htonl(2))
1081 return error("bad index version");
1082 SHA1_Init(&c);
1083 SHA1_Update(&c, hdr, size - 20);
1084 SHA1_Final(sha1, &c);
1085 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1086 return error("bad index file sha1 signature");
1087 return 0;
1088 }
1089
1090 static int read_index_extension(struct index_state *istate,
1091 const char *ext, void *data, unsigned long sz)
1092 {
1093 switch (CACHE_EXT(ext)) {
1094 case CACHE_EXT_TREE:
1095 istate->cache_tree = cache_tree_read(data, sz);
1096 break;
1097 default:
1098 if (*ext < 'A' || 'Z' < *ext)
1099 return error("index uses %.4s extension, which we do not understand",
1100 ext);
1101 fprintf(stderr, "ignoring %.4s extension\n", ext);
1102 break;
1103 }
1104 return 0;
1105 }
1106
1107 int read_index(struct index_state *istate)
1108 {
1109 return read_index_from(istate, get_index_file());
1110 }
1111
1112 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1113 {
1114 size_t len;
1115
1116 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1117 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1118 ce->ce_dev = ntohl(ondisk->dev);
1119 ce->ce_ino = ntohl(ondisk->ino);
1120 ce->ce_mode = ntohl(ondisk->mode);
1121 ce->ce_uid = ntohl(ondisk->uid);
1122 ce->ce_gid = ntohl(ondisk->gid);
1123 ce->ce_size = ntohl(ondisk->size);
1124 /* On-disk flags are just 16 bits */
1125 ce->ce_flags = ntohs(ondisk->flags);
1126
1127 /* For future extension: we do not understand this entry yet */
1128 if (ce->ce_flags & CE_EXTENDED)
1129 die("Unknown index entry format");
1130 hashcpy(ce->sha1, ondisk->sha1);
1131
1132 len = ce->ce_flags & CE_NAMEMASK;
1133 if (len == CE_NAMEMASK)
1134 len = strlen(ondisk->name);
1135 /*
1136 * NEEDSWORK: If the original index is crafted, this copy could
1137 * go unchecked.
1138 */
1139 memcpy(ce->name, ondisk->name, len + 1);
1140 }
1141
1142 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1143 {
1144 long per_entry;
1145
1146 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1147
1148 /*
1149 * Alignment can cause differences. This should be "alignof", but
1150 * since that's a gcc'ism, just use the size of a pointer.
1151 */
1152 per_entry += sizeof(void *);
1153 return ondisk_size + entries*per_entry;
1154 }
1155
1156 /* remember to discard_cache() before reading a different cache! */
1157 int read_index_from(struct index_state *istate, const char *path)
1158 {
1159 int fd, i;
1160 struct stat st;
1161 unsigned long src_offset, dst_offset;
1162 struct cache_header *hdr;
1163 void *mmap;
1164 size_t mmap_size;
1165
1166 errno = EBUSY;
1167 if (istate->initialized)
1168 return istate->cache_nr;
1169
1170 errno = ENOENT;
1171 istate->timestamp = 0;
1172 fd = open(path, O_RDONLY);
1173 if (fd < 0) {
1174 if (errno == ENOENT)
1175 return 0;
1176 die("index file open failed (%s)", strerror(errno));
1177 }
1178
1179 if (fstat(fd, &st))
1180 die("cannot stat the open index (%s)", strerror(errno));
1181
1182 errno = EINVAL;
1183 mmap_size = xsize_t(st.st_size);
1184 if (mmap_size < sizeof(struct cache_header) + 20)
1185 die("index file smaller than expected");
1186
1187 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1188 close(fd);
1189 if (mmap == MAP_FAILED)
1190 die("unable to map index file");
1191
1192 hdr = mmap;
1193 if (verify_hdr(hdr, mmap_size) < 0)
1194 goto unmap;
1195
1196 istate->cache_nr = ntohl(hdr->hdr_entries);
1197 istate->cache_alloc = alloc_nr(istate->cache_nr);
1198 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1199
1200 /*
1201 * The disk format is actually larger than the in-memory format,
1202 * due to space for nsec etc, so even though the in-memory one
1203 * has room for a few more flags, we can allocate using the same
1204 * index size
1205 */
1206 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1207 istate->initialized = 1;
1208
1209 src_offset = sizeof(*hdr);
1210 dst_offset = 0;
1211 for (i = 0; i < istate->cache_nr; i++) {
1212 struct ondisk_cache_entry *disk_ce;
1213 struct cache_entry *ce;
1214
1215 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1216 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1217 convert_from_disk(disk_ce, ce);
1218 set_index_entry(istate, i, ce);
1219
1220 src_offset += ondisk_ce_size(ce);
1221 dst_offset += ce_size(ce);
1222 }
1223 istate->timestamp = st.st_mtime;
1224 while (src_offset <= mmap_size - 20 - 8) {
1225 /* After an array of active_nr index entries,
1226 * there can be arbitrary number of extended
1227 * sections, each of which is prefixed with
1228 * extension name (4-byte) and section length
1229 * in 4-byte network byte order.
1230 */
1231 unsigned long extsize;
1232 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1233 extsize = ntohl(extsize);
1234 if (read_index_extension(istate,
1235 (const char *) mmap + src_offset,
1236 (char *) mmap + src_offset + 8,
1237 extsize) < 0)
1238 goto unmap;
1239 src_offset += 8;
1240 src_offset += extsize;
1241 }
1242 munmap(mmap, mmap_size);
1243 return istate->cache_nr;
1244
1245 unmap:
1246 munmap(mmap, mmap_size);
1247 errno = EINVAL;
1248 die("index file corrupt");
1249 }
1250
1251 int discard_index(struct index_state *istate)
1252 {
1253 istate->cache_nr = 0;
1254 istate->cache_changed = 0;
1255 istate->timestamp = 0;
1256 istate->name_hash_initialized = 0;
1257 free_hash(&istate->name_hash);
1258 cache_tree_free(&(istate->cache_tree));
1259 free(istate->alloc);
1260 istate->alloc = NULL;
1261 istate->initialized = 0;
1262
1263 /* no need to throw away allocated active_cache */
1264 return 0;
1265 }
1266
1267 int unmerged_index(const struct index_state *istate)
1268 {
1269 int i;
1270 for (i = 0; i < istate->cache_nr; i++) {
1271 if (ce_stage(istate->cache[i]))
1272 return 1;
1273 }
1274 return 0;
1275 }
1276
1277 #define WRITE_BUFFER_SIZE 8192
1278 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1279 static unsigned long write_buffer_len;
1280
1281 static int ce_write_flush(SHA_CTX *context, int fd)
1282 {
1283 unsigned int buffered = write_buffer_len;
1284 if (buffered) {
1285 SHA1_Update(context, write_buffer, buffered);
1286 if (write_in_full(fd, write_buffer, buffered) != buffered)
1287 return -1;
1288 write_buffer_len = 0;
1289 }
1290 return 0;
1291 }
1292
1293 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1294 {
1295 while (len) {
1296 unsigned int buffered = write_buffer_len;
1297 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1298 if (partial > len)
1299 partial = len;
1300 memcpy(write_buffer + buffered, data, partial);
1301 buffered += partial;
1302 if (buffered == WRITE_BUFFER_SIZE) {
1303 write_buffer_len = buffered;
1304 if (ce_write_flush(context, fd))
1305 return -1;
1306 buffered = 0;
1307 }
1308 write_buffer_len = buffered;
1309 len -= partial;
1310 data = (char *) data + partial;
1311 }
1312 return 0;
1313 }
1314
1315 static int write_index_ext_header(SHA_CTX *context, int fd,
1316 unsigned int ext, unsigned int sz)
1317 {
1318 ext = htonl(ext);
1319 sz = htonl(sz);
1320 return ((ce_write(context, fd, &ext, 4) < 0) ||
1321 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1322 }
1323
1324 static int ce_flush(SHA_CTX *context, int fd)
1325 {
1326 unsigned int left = write_buffer_len;
1327
1328 if (left) {
1329 write_buffer_len = 0;
1330 SHA1_Update(context, write_buffer, left);
1331 }
1332
1333 /* Flush first if not enough space for SHA1 signature */
1334 if (left + 20 > WRITE_BUFFER_SIZE) {
1335 if (write_in_full(fd, write_buffer, left) != left)
1336 return -1;
1337 left = 0;
1338 }
1339
1340 /* Append the SHA1 signature at the end */
1341 SHA1_Final(write_buffer + left, context);
1342 left += 20;
1343 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1344 }
1345
1346 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1347 {
1348 /*
1349 * The only thing we care about in this function is to smudge the
1350 * falsely clean entry due to touch-update-touch race, so we leave
1351 * everything else as they are. We are called for entries whose
1352 * ce_mtime match the index file mtime.
1353 *
1354 * Note that this actually does not do much for gitlinks, for
1355 * which ce_match_stat_basic() always goes to the actual
1356 * contents. The caller checks with is_racy_timestamp() which
1357 * always says "no" for gitlinks, so we are not called for them ;-)
1358 */
1359 struct stat st;
1360
1361 if (lstat(ce->name, &st) < 0)
1362 return;
1363 if (ce_match_stat_basic(ce, &st))
1364 return;
1365 if (ce_modified_check_fs(ce, &st)) {
1366 /* This is "racily clean"; smudge it. Note that this
1367 * is a tricky code. At first glance, it may appear
1368 * that it can break with this sequence:
1369 *
1370 * $ echo xyzzy >frotz
1371 * $ git-update-index --add frotz
1372 * $ : >frotz
1373 * $ sleep 3
1374 * $ echo filfre >nitfol
1375 * $ git-update-index --add nitfol
1376 *
1377 * but it does not. When the second update-index runs,
1378 * it notices that the entry "frotz" has the same timestamp
1379 * as index, and if we were to smudge it by resetting its
1380 * size to zero here, then the object name recorded
1381 * in index is the 6-byte file but the cached stat information
1382 * becomes zero --- which would then match what we would
1383 * obtain from the filesystem next time we stat("frotz").
1384 *
1385 * However, the second update-index, before calling
1386 * this function, notices that the cached size is 6
1387 * bytes and what is on the filesystem is an empty
1388 * file, and never calls us, so the cached size information
1389 * for "frotz" stays 6 which does not match the filesystem.
1390 */
1391 ce->ce_size = 0;
1392 }
1393 }
1394
1395 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1396 {
1397 int size = ondisk_ce_size(ce);
1398 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1399
1400 ondisk->ctime.sec = htonl(ce->ce_ctime);
1401 ondisk->ctime.nsec = 0;
1402 ondisk->mtime.sec = htonl(ce->ce_mtime);
1403 ondisk->mtime.nsec = 0;
1404 ondisk->dev = htonl(ce->ce_dev);
1405 ondisk->ino = htonl(ce->ce_ino);
1406 ondisk->mode = htonl(ce->ce_mode);
1407 ondisk->uid = htonl(ce->ce_uid);
1408 ondisk->gid = htonl(ce->ce_gid);
1409 ondisk->size = htonl(ce->ce_size);
1410 hashcpy(ondisk->sha1, ce->sha1);
1411 ondisk->flags = htons(ce->ce_flags);
1412 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1413
1414 return ce_write(c, fd, ondisk, size);
1415 }
1416
1417 int write_index(const struct index_state *istate, int newfd)
1418 {
1419 SHA_CTX c;
1420 struct cache_header hdr;
1421 int i, err, removed;
1422 struct cache_entry **cache = istate->cache;
1423 int entries = istate->cache_nr;
1424
1425 for (i = removed = 0; i < entries; i++)
1426 if (cache[i]->ce_flags & CE_REMOVE)
1427 removed++;
1428
1429 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1430 hdr.hdr_version = htonl(2);
1431 hdr.hdr_entries = htonl(entries - removed);
1432
1433 SHA1_Init(&c);
1434 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1435 return -1;
1436
1437 for (i = 0; i < entries; i++) {
1438 struct cache_entry *ce = cache[i];
1439 if (ce->ce_flags & CE_REMOVE)
1440 continue;
1441 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1442 ce_smudge_racily_clean_entry(ce);
1443 if (ce_write_entry(&c, newfd, ce) < 0)
1444 return -1;
1445 }
1446
1447 /* Write extension data here */
1448 if (istate->cache_tree) {
1449 struct strbuf sb;
1450
1451 strbuf_init(&sb, 0);
1452 cache_tree_write(&sb, istate->cache_tree);
1453 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1454 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1455 strbuf_release(&sb);
1456 if (err)
1457 return -1;
1458 }
1459 return ce_flush(&c, newfd);
1460 }
1461
1462 /*
1463 * Read the index file that is potentially unmerged into given
1464 * index_state, dropping any unmerged entries. Returns true is
1465 * the index is unmerged. Callers who want to refuse to work
1466 * from an unmerged state can call this and check its return value,
1467 * instead of calling read_cache().
1468 */
1469 int read_index_unmerged(struct index_state *istate)
1470 {
1471 int i;
1472 struct cache_entry **dst;
1473 struct cache_entry *last = NULL;
1474
1475 read_index(istate);
1476 dst = istate->cache;
1477 for (i = 0; i < istate->cache_nr; i++) {
1478 struct cache_entry *ce = istate->cache[i];
1479 if (ce_stage(ce)) {
1480 remove_name_hash(ce);
1481 if (last && !strcmp(ce->name, last->name))
1482 continue;
1483 cache_tree_invalidate_path(istate->cache_tree, ce->name);
1484 last = ce;
1485 continue;
1486 }
1487 *dst++ = ce;
1488 }
1489 istate->cache_nr = dst - istate->cache;
1490 return !!last;
1491 }
1492
1493 struct update_callback_data
1494 {
1495 int flags;
1496 int add_errors;
1497 };
1498
1499 static void update_callback(struct diff_queue_struct *q,
1500 struct diff_options *opt, void *cbdata)
1501 {
1502 int i;
1503 struct update_callback_data *data = cbdata;
1504
1505 for (i = 0; i < q->nr; i++) {
1506 struct diff_filepair *p = q->queue[i];
1507 const char *path = p->one->path;
1508 switch (p->status) {
1509 default:
1510 die("unexpected diff status %c", p->status);
1511 case DIFF_STATUS_UNMERGED:
1512 case DIFF_STATUS_MODIFIED:
1513 case DIFF_STATUS_TYPE_CHANGED:
1514 if (add_file_to_index(&the_index, path, data->flags)) {
1515 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1516 die("updating files failed");
1517 data->add_errors++;
1518 }
1519 break;
1520 case DIFF_STATUS_DELETED:
1521 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1522 break;
1523 if (!(data->flags & ADD_CACHE_PRETEND))
1524 remove_file_from_index(&the_index, path);
1525 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1526 printf("remove '%s'\n", path);
1527 break;
1528 }
1529 }
1530 }
1531
1532 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1533 {
1534 struct update_callback_data data;
1535 struct rev_info rev;
1536 init_revisions(&rev, prefix);
1537 setup_revisions(0, NULL, &rev, NULL);
1538 rev.prune_data = pathspec;
1539 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1540 rev.diffopt.format_callback = update_callback;
1541 data.flags = flags;
1542 data.add_errors = 0;
1543 rev.diffopt.format_callback_data = &data;
1544 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1545 return !!data.add_errors;
1546 }
1547