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