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
24struct index_state the_index;
25
26/*
27 * This only updates the "non-critical" parts of the directory
28 * cache, ie the parts that aren't tracked by GIT, and only used
29 * to validate the cache.
30 */
31void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
32{
33 ce->ce_ctime.sec = htonl(st->st_ctime);
34 ce->ce_mtime.sec = htonl(st->st_mtime);
35#ifdef USE_NSEC
36 ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
37 ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
38#endif
39 ce->ce_dev = htonl(st->st_dev);
40 ce->ce_ino = htonl(st->st_ino);
41 ce->ce_uid = htonl(st->st_uid);
42 ce->ce_gid = htonl(st->st_gid);
43 ce->ce_size = htonl(st->st_size);
44
45 if (assume_unchanged)
46 ce->ce_flags |= htons(CE_VALID);
47}
48
49static int ce_compare_data(struct cache_entry *ce, struct stat *st)
50{
51 int match = -1;
52 int fd = open(ce->name, O_RDONLY);
53
54 if (fd >= 0) {
55 unsigned char sha1[20];
56 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
57 match = hashcmp(sha1, ce->sha1);
58 /* index_fd() closed the file descriptor already */
59 }
60 return match;
61}
62
63static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
64{
65 int match = -1;
66 char *target;
67 void *buffer;
68 unsigned long size;
69 enum object_type type;
70 int len;
71
72 target = xmalloc(expected_size);
73 len = readlink(ce->name, target, expected_size);
74 if (len != expected_size) {
75 free(target);
76 return -1;
77 }
78 buffer = read_sha1_file(ce->sha1, &type, &size);
79 if (!buffer) {
80 free(target);
81 return -1;
82 }
83 if (size == expected_size)
84 match = memcmp(buffer, target, size);
85 free(buffer);
86 free(target);
87 return match;
88}
89
90static int ce_compare_gitlink(struct cache_entry *ce)
91{
92 unsigned char sha1[20];
93
94 /*
95 * We don't actually require that the .git directory
96 * under GITLINK directory be a valid git directory. It
97 * might even be missing (in case nobody populated that
98 * sub-project).
99 *
100 * If so, we consider it always to match.
101 */
102 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
103 return 0;
104 return hashcmp(sha1, ce->sha1);
105}
106
107static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
108{
109 switch (st->st_mode & S_IFMT) {
110 case S_IFREG:
111 if (ce_compare_data(ce, st))
112 return DATA_CHANGED;
113 break;
114 case S_IFLNK:
115 if (ce_compare_link(ce, xsize_t(st->st_size)))
116 return DATA_CHANGED;
117 break;
118 case S_IFDIR:
119 if (S_ISGITLINK(ntohl(ce->ce_mode)))
120 return 0;
121 default:
122 return TYPE_CHANGED;
123 }
124 return 0;
125}
126
127static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
128{
129 unsigned int changed = 0;
130
131 switch (ntohl(ce->ce_mode) & S_IFMT) {
132 case S_IFREG:
133 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
134 /* We consider only the owner x bit to be relevant for
135 * "mode changes"
136 */
137 if (trust_executable_bit &&
138 (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
139 changed |= MODE_CHANGED;
140 break;
141 case S_IFLNK:
142 if (!S_ISLNK(st->st_mode) &&
143 (has_symlinks || !S_ISREG(st->st_mode)))
144 changed |= TYPE_CHANGED;
145 break;
146 case S_IFGITLINK:
147 if (!S_ISDIR(st->st_mode))
148 changed |= TYPE_CHANGED;
149 else if (ce_compare_gitlink(ce))
150 changed |= DATA_CHANGED;
151 return changed;
152 case 0: /* Special case: unmerged file in index */
153 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
154 default:
155 die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
156 }
157 if (ce->ce_mtime.sec != htonl(st->st_mtime))
158 changed |= MTIME_CHANGED;
159 if (ce->ce_ctime.sec != htonl(st->st_ctime))
160 changed |= CTIME_CHANGED;
161
162#ifdef USE_NSEC
163 /*
164 * nsec seems unreliable - not all filesystems support it, so
165 * as long as it is in the inode cache you get right nsec
166 * but after it gets flushed, you get zero nsec.
167 */
168 if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
169 changed |= MTIME_CHANGED;
170 if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
171 changed |= CTIME_CHANGED;
172#endif
173
174 if (ce->ce_uid != htonl(st->st_uid) ||
175 ce->ce_gid != htonl(st->st_gid))
176 changed |= OWNER_CHANGED;
177 if (ce->ce_ino != htonl(st->st_ino))
178 changed |= INODE_CHANGED;
179
180#ifdef USE_STDEV
181 /*
182 * st_dev breaks on network filesystems where different
183 * clients will have different views of what "device"
184 * the filesystem is on
185 */
186 if (ce->ce_dev != htonl(st->st_dev))
187 changed |= INODE_CHANGED;
188#endif
189
190 if (ce->ce_size != htonl(st->st_size))
191 changed |= DATA_CHANGED;
192
193 return changed;
194}
195
196int ie_match_stat(struct index_state *istate,
197 struct cache_entry *ce, struct stat *st, int options)
198{
199 unsigned int changed;
200 int ignore_valid = options & 01;
201 int assume_racy_is_modified = options & 02;
202
203 /*
204 * If it's marked as always valid in the index, it's
205 * valid whatever the checked-out copy says.
206 */
207 if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
208 return 0;
209
210 changed = ce_match_stat_basic(ce, st);
211
212 /*
213 * Within 1 second of this sequence:
214 * echo xyzzy >file && git-update-index --add file
215 * running this command:
216 * echo frotz >file
217 * would give a falsely clean cache entry. The mtime and
218 * length match the cache, and other stat fields do not change.
219 *
220 * We could detect this at update-index time (the cache entry
221 * being registered/updated records the same time as "now")
222 * and delay the return from git-update-index, but that would
223 * effectively mean we can make at most one commit per second,
224 * which is not acceptable. Instead, we check cache entries
225 * whose mtime are the same as the index file timestamp more
226 * carefully than others.
227 */
228 if (!changed &&
229 istate->timestamp &&
230 istate->timestamp <= ntohl(ce->ce_mtime.sec)) {
231 if (assume_racy_is_modified)
232 changed |= DATA_CHANGED;
233 else
234 changed |= ce_modified_check_fs(ce, st);
235 }
236
237 return changed;
238}
239
240int ie_modified(struct index_state *istate,
241 struct cache_entry *ce, struct stat *st, int really)
242{
243 int changed, changed_fs;
244 changed = ie_match_stat(istate, ce, st, really);
245 if (!changed)
246 return 0;
247 /*
248 * If the mode or type has changed, there's no point in trying
249 * to refresh the entry - it's not going to match
250 */
251 if (changed & (MODE_CHANGED | TYPE_CHANGED))
252 return changed;
253
254 /* Immediately after read-tree or update-index --cacheinfo,
255 * the length field is zero. For other cases the ce_size
256 * should match the SHA1 recorded in the index entry.
257 */
258 if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
259 return changed;
260
261 changed_fs = ce_modified_check_fs(ce, st);
262 if (changed_fs)
263 return changed | changed_fs;
264 return 0;
265}
266
267int base_name_compare(const char *name1, int len1, int mode1,
268 const char *name2, int len2, int mode2)
269{
270 unsigned char c1, c2;
271 int len = len1 < len2 ? len1 : len2;
272 int cmp;
273
274 cmp = memcmp(name1, name2, len);
275 if (cmp)
276 return cmp;
277 c1 = name1[len];
278 c2 = name2[len];
279 if (!c1 && S_ISDIR(mode1))
280 c1 = '/';
281 if (!c2 && S_ISDIR(mode2))
282 c2 = '/';
283 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
284}
285
286int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
287{
288 int len1 = flags1 & CE_NAMEMASK;
289 int len2 = flags2 & CE_NAMEMASK;
290 int len = len1 < len2 ? len1 : len2;
291 int cmp;
292
293 cmp = memcmp(name1, name2, len);
294 if (cmp)
295 return cmp;
296 if (len1 < len2)
297 return -1;
298 if (len1 > len2)
299 return 1;
300
301 /* Compare stages */
302 flags1 &= CE_STAGEMASK;
303 flags2 &= CE_STAGEMASK;
304
305 if (flags1 < flags2)
306 return -1;
307 if (flags1 > flags2)
308 return 1;
309 return 0;
310}
311
312int index_name_pos(struct index_state *istate, const char *name, int namelen)
313{
314 int first, last;
315
316 first = 0;
317 last = istate->cache_nr;
318 while (last > first) {
319 int next = (last + first) >> 1;
320 struct cache_entry *ce = istate->cache[next];
321 int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
322 if (!cmp)
323 return next;
324 if (cmp < 0) {
325 last = next;
326 continue;
327 }
328 first = next+1;
329 }
330 return -first-1;
331}
332
333/* Remove entry, return true if there are more entries to go.. */
334int remove_index_entry_at(struct index_state *istate, int pos)
335{
336 istate->cache_changed = 1;
337 istate->cache_nr--;
338 if (pos >= istate->cache_nr)
339 return 0;
340 memmove(istate->cache + pos,
341 istate->cache + pos + 1,
342 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
343 return 1;
344}
345
346int remove_file_from_index(struct index_state *istate, const char *path)
347{
348 int pos = index_name_pos(istate, path, strlen(path));
349 if (pos < 0)
350 pos = -pos-1;
351 cache_tree_invalidate_path(istate->cache_tree, path);
352 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
353 remove_index_entry_at(istate, pos);
354 return 0;
355}
356
357static int compare_name(struct cache_entry *ce, const char *path, int namelen)
358{
359 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
360}
361
362static int index_name_pos_also_unmerged(struct index_state *istate,
363 const char *path, int namelen)
364{
365 int pos = index_name_pos(istate, path, namelen);
366 struct cache_entry *ce;
367
368 if (pos >= 0)
369 return pos;
370
371 /* maybe unmerged? */
372 pos = -1 - pos;
373 if (pos >= istate->cache_nr ||
374 compare_name((ce = istate->cache[pos]), path, namelen))
375 return -1;
376
377 /* order of preference: stage 2, 1, 3 */
378 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
379 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
380 !compare_name(ce, path, namelen))
381 pos++;
382 return pos;
383}
384
385int add_file_to_index(struct index_state *istate, const char *path, int verbose)
386{
387 int size, namelen, pos;
388 struct stat st;
389 struct cache_entry *ce;
390
391 if (lstat(path, &st))
392 die("%s: unable to stat (%s)", path, strerror(errno));
393
394 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
395 die("%s: can only add regular files, symbolic links or git-directories", path);
396
397 namelen = strlen(path);
398 if (S_ISDIR(st.st_mode)) {
399 while (namelen && path[namelen-1] == '/')
400 namelen--;
401 }
402 size = cache_entry_size(namelen);
403 ce = xcalloc(1, size);
404 memcpy(ce->name, path, namelen);
405 ce->ce_flags = htons(namelen);
406 fill_stat_cache_info(ce, &st);
407
408 if (trust_executable_bit && has_symlinks)
409 ce->ce_mode = create_ce_mode(st.st_mode);
410 else {
411 /* If there is an existing entry, pick the mode bits and type
412 * from it, otherwise assume unexecutable regular file.
413 */
414 struct cache_entry *ent;
415 int pos = index_name_pos_also_unmerged(istate, path, namelen);
416
417 ent = (0 <= pos) ? istate->cache[pos] : NULL;
418 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
419 }
420
421 pos = index_name_pos(istate, ce->name, namelen);
422 if (0 <= pos &&
423 !ce_stage(istate->cache[pos]) &&
424 !ie_modified(istate, istate->cache[pos], &st, 1)) {
425 /* Nothing changed, really */
426 free(ce);
427 return 0;
428 }
429
430 if (index_path(ce->sha1, path, &st, 1))
431 die("unable to index file %s", path);
432 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
433 die("unable to add %s to index",path);
434 if (verbose)
435 printf("add '%s'\n", path);
436 return 0;
437}
438
439struct cache_entry *make_cache_entry(unsigned int mode,
440 const unsigned char *sha1, const char *path, int stage,
441 int refresh)
442{
443 int size, len;
444 struct cache_entry *ce;
445
446 if (!verify_path(path))
447 return NULL;
448
449 len = strlen(path);
450 size = cache_entry_size(len);
451 ce = xcalloc(1, size);
452
453 hashcpy(ce->sha1, sha1);
454 memcpy(ce->name, path, len);
455 ce->ce_flags = create_ce_flags(len, stage);
456 ce->ce_mode = create_ce_mode(mode);
457
458 if (refresh)
459 return refresh_cache_entry(ce, 0);
460
461 return ce;
462}
463
464int ce_same_name(struct cache_entry *a, struct cache_entry *b)
465{
466 int len = ce_namelen(a);
467 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
468}
469
470int ce_path_match(const struct cache_entry *ce, const char **pathspec)
471{
472 const char *match, *name;
473 int len;
474
475 if (!pathspec)
476 return 1;
477
478 len = ce_namelen(ce);
479 name = ce->name;
480 while ((match = *pathspec++) != NULL) {
481 int matchlen = strlen(match);
482 if (matchlen > len)
483 continue;
484 if (memcmp(name, match, matchlen))
485 continue;
486 if (matchlen && name[matchlen-1] == '/')
487 return 1;
488 if (name[matchlen] == '/' || !name[matchlen])
489 return 1;
490 if (!matchlen)
491 return 1;
492 }
493 return 0;
494}
495
496/*
497 * We fundamentally don't like some paths: we don't want
498 * dot or dot-dot anywhere, and for obvious reasons don't
499 * want to recurse into ".git" either.
500 *
501 * Also, we don't want double slashes or slashes at the
502 * end that can make pathnames ambiguous.
503 */
504static int verify_dotfile(const char *rest)
505{
506 /*
507 * The first character was '.', but that
508 * has already been discarded, we now test
509 * the rest.
510 */
511 switch (*rest) {
512 /* "." is not allowed */
513 case '\0': case '/':
514 return 0;
515
516 /*
517 * ".git" followed by NUL or slash is bad. This
518 * shares the path end test with the ".." case.
519 */
520 case 'g':
521 if (rest[1] != 'i')
522 break;
523 if (rest[2] != 't')
524 break;
525 rest += 2;
526 /* fallthrough */
527 case '.':
528 if (rest[1] == '\0' || rest[1] == '/')
529 return 0;
530 }
531 return 1;
532}
533
534int verify_path(const char *path)
535{
536 char c;
537
538 goto inside;
539 for (;;) {
540 if (!c)
541 return 1;
542 if (c == '/') {
543inside:
544 c = *path++;
545 switch (c) {
546 default:
547 continue;
548 case '/': case '\0':
549 break;
550 case '.':
551 if (verify_dotfile(path))
552 continue;
553 }
554 return 0;
555 }
556 c = *path++;
557 }
558}
559
560/*
561 * Do we have another file that has the beginning components being a
562 * proper superset of the name we're trying to add?
563 */
564static int has_file_name(struct index_state *istate,
565 const struct cache_entry *ce, int pos, int ok_to_replace)
566{
567 int retval = 0;
568 int len = ce_namelen(ce);
569 int stage = ce_stage(ce);
570 const char *name = ce->name;
571
572 while (pos < istate->cache_nr) {
573 struct cache_entry *p = istate->cache[pos++];
574
575 if (len >= ce_namelen(p))
576 break;
577 if (memcmp(name, p->name, len))
578 break;
579 if (ce_stage(p) != stage)
580 continue;
581 if (p->name[len] != '/')
582 continue;
583 if (!ce_stage(p) && !p->ce_mode)
584 continue;
585 retval = -1;
586 if (!ok_to_replace)
587 break;
588 remove_index_entry_at(istate, --pos);
589 }
590 return retval;
591}
592
593/*
594 * Do we have another file with a pathname that is a proper
595 * subset of the name we're trying to add?
596 */
597static int has_dir_name(struct index_state *istate,
598 const struct cache_entry *ce, int pos, int ok_to_replace)
599{
600 int retval = 0;
601 int stage = ce_stage(ce);
602 const char *name = ce->name;
603 const char *slash = name + ce_namelen(ce);
604
605 for (;;) {
606 int len;
607
608 for (;;) {
609 if (*--slash == '/')
610 break;
611 if (slash <= ce->name)
612 return retval;
613 }
614 len = slash - name;
615
616 pos = index_name_pos(istate, name, ntohs(create_ce_flags(len, stage)));
617 if (pos >= 0) {
618 /*
619 * Found one, but not so fast. This could
620 * be a marker that says "I was here, but
621 * I am being removed". Such an entry is
622 * not a part of the resulting tree, and
623 * it is Ok to have a directory at the same
624 * path.
625 */
626 if (stage || istate->cache[pos]->ce_mode) {
627 retval = -1;
628 if (!ok_to_replace)
629 break;
630 remove_index_entry_at(istate, pos);
631 continue;
632 }
633 }
634 else
635 pos = -pos-1;
636
637 /*
638 * Trivial optimization: if we find an entry that
639 * already matches the sub-directory, then we know
640 * we're ok, and we can exit.
641 */
642 while (pos < istate->cache_nr) {
643 struct cache_entry *p = istate->cache[pos];
644 if ((ce_namelen(p) <= len) ||
645 (p->name[len] != '/') ||
646 memcmp(p->name, name, len))
647 break; /* not our subdirectory */
648 if (ce_stage(p) == stage && (stage || p->ce_mode))
649 /* p is at the same stage as our entry, and
650 * is a subdirectory of what we are looking
651 * at, so we cannot have conflicts at our
652 * level or anything shorter.
653 */
654 return retval;
655 pos++;
656 }
657 }
658 return retval;
659}
660
661/* We may be in a situation where we already have path/file and path
662 * is being added, or we already have path and path/file is being
663 * added. Either one would result in a nonsense tree that has path
664 * twice when git-write-tree tries to write it out. Prevent it.
665 *
666 * If ok-to-replace is specified, we remove the conflicting entries
667 * from the cache so the caller should recompute the insert position.
668 * When this happens, we return non-zero.
669 */
670static int check_file_directory_conflict(struct index_state *istate,
671 const struct cache_entry *ce,
672 int pos, int ok_to_replace)
673{
674 int retval;
675
676 /*
677 * When ce is an "I am going away" entry, we allow it to be added
678 */
679 if (!ce_stage(ce) && !ce->ce_mode)
680 return 0;
681
682 /*
683 * We check if the path is a sub-path of a subsequent pathname
684 * first, since removing those will not change the position
685 * in the array.
686 */
687 retval = has_file_name(istate, ce, pos, ok_to_replace);
688
689 /*
690 * Then check if the path might have a clashing sub-directory
691 * before it.
692 */
693 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
694}
695
696static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
697{
698 int pos;
699 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
700 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
701 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
702
703 cache_tree_invalidate_path(istate->cache_tree, ce->name);
704 pos = index_name_pos(istate, ce->name, ntohs(ce->ce_flags));
705
706 /* existing match? Just replace it. */
707 if (pos >= 0) {
708 istate->cache_changed = 1;
709 istate->cache[pos] = ce;
710 return 0;
711 }
712 pos = -pos-1;
713
714 /*
715 * Inserting a merged entry ("stage 0") into the index
716 * will always replace all non-merged entries..
717 */
718 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
719 while (ce_same_name(istate->cache[pos], ce)) {
720 ok_to_add = 1;
721 if (!remove_index_entry_at(istate, pos))
722 break;
723 }
724 }
725
726 if (!ok_to_add)
727 return -1;
728 if (!verify_path(ce->name))
729 return -1;
730
731 if (!skip_df_check &&
732 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
733 if (!ok_to_replace)
734 return error("'%s' appears as both a file and as a directory",
735 ce->name);
736 pos = index_name_pos(istate, ce->name, ntohs(ce->ce_flags));
737 pos = -pos-1;
738 }
739 return pos + 1;
740}
741
742int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
743{
744 int pos;
745
746 if (option & ADD_CACHE_JUST_APPEND)
747 pos = istate->cache_nr;
748 else {
749 int ret;
750 ret = add_index_entry_with_check(istate, ce, option);
751 if (ret <= 0)
752 return ret;
753 pos = ret - 1;
754 }
755
756 /* Make sure the array is big enough .. */
757 if (istate->cache_nr == istate->cache_alloc) {
758 istate->cache_alloc = alloc_nr(istate->cache_alloc);
759 istate->cache = xrealloc(istate->cache,
760 istate->cache_alloc * sizeof(struct cache_entry *));
761 }
762
763 /* Add it in.. */
764 istate->cache_nr++;
765 if (istate->cache_nr > pos + 1)
766 memmove(istate->cache + pos + 1,
767 istate->cache + pos,
768 (istate->cache_nr - pos - 1) * sizeof(ce));
769 istate->cache[pos] = ce;
770 istate->cache_changed = 1;
771 return 0;
772}
773
774/*
775 * "refresh" does not calculate a new sha1 file or bring the
776 * cache up-to-date for mode/content changes. But what it
777 * _does_ do is to "re-match" the stat information of a file
778 * with the cache, so that you can refresh the cache for a
779 * file that hasn't been changed but where the stat entry is
780 * out of date.
781 *
782 * For example, you'd want to do this after doing a "git-read-tree",
783 * to link up the stat cache details with the proper files.
784 */
785static struct cache_entry *refresh_cache_ent(struct index_state *istate,
786 struct cache_entry *ce, int really, int *err)
787{
788 struct stat st;
789 struct cache_entry *updated;
790 int changed, size;
791
792 if (lstat(ce->name, &st) < 0) {
793 if (err)
794 *err = errno;
795 return NULL;
796 }
797
798 changed = ie_match_stat(istate, ce, &st, really);
799 if (!changed) {
800 if (really && assume_unchanged &&
801 !(ce->ce_flags & htons(CE_VALID)))
802 ; /* mark this one VALID again */
803 else
804 return ce;
805 }
806
807 if (ie_modified(istate, ce, &st, really)) {
808 if (err)
809 *err = EINVAL;
810 return NULL;
811 }
812
813 size = ce_size(ce);
814 updated = xmalloc(size);
815 memcpy(updated, ce, size);
816 fill_stat_cache_info(updated, &st);
817
818 /* In this case, if really is not set, we should leave
819 * CE_VALID bit alone. Otherwise, paths marked with
820 * --no-assume-unchanged (i.e. things to be edited) will
821 * reacquire CE_VALID bit automatically, which is not
822 * really what we want.
823 */
824 if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
825 updated->ce_flags &= ~htons(CE_VALID);
826
827 return updated;
828}
829
830int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
831{
832 int i;
833 int has_errors = 0;
834 int really = (flags & REFRESH_REALLY) != 0;
835 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
836 int quiet = (flags & REFRESH_QUIET) != 0;
837 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
838
839 for (i = 0; i < istate->cache_nr; i++) {
840 struct cache_entry *ce, *new;
841 int cache_errno = 0;
842
843 ce = istate->cache[i];
844 if (ce_stage(ce)) {
845 while ((i < istate->cache_nr) &&
846 ! strcmp(istate->cache[i]->name, ce->name))
847 i++;
848 i--;
849 if (allow_unmerged)
850 continue;
851 printf("%s: needs merge\n", ce->name);
852 has_errors = 1;
853 continue;
854 }
855
856 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
857 continue;
858
859 new = refresh_cache_ent(istate, ce, really, &cache_errno);
860 if (new == ce)
861 continue;
862 if (!new) {
863 if (not_new && cache_errno == ENOENT)
864 continue;
865 if (really && cache_errno == EINVAL) {
866 /* If we are doing --really-refresh that
867 * means the index is not valid anymore.
868 */
869 ce->ce_flags &= ~htons(CE_VALID);
870 istate->cache_changed = 1;
871 }
872 if (quiet)
873 continue;
874 printf("%s: needs update\n", ce->name);
875 has_errors = 1;
876 continue;
877 }
878 istate->cache_changed = 1;
879 /* You can NOT just free istate->cache[i] here, since it
880 * might not be necessarily malloc()ed but can also come
881 * from mmap(). */
882 istate->cache[i] = new;
883 }
884 return has_errors;
885}
886
887struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
888{
889 return refresh_cache_ent(&the_index, ce, really, NULL);
890}
891
892static int verify_hdr(struct cache_header *hdr, unsigned long size)
893{
894 SHA_CTX c;
895 unsigned char sha1[20];
896
897 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
898 return error("bad signature");
899 if (hdr->hdr_version != htonl(2))
900 return error("bad index version");
901 SHA1_Init(&c);
902 SHA1_Update(&c, hdr, size - 20);
903 SHA1_Final(sha1, &c);
904 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
905 return error("bad index file sha1 signature");
906 return 0;
907}
908
909static int read_index_extension(struct index_state *istate,
910 const char *ext, void *data, unsigned long sz)
911{
912 switch (CACHE_EXT(ext)) {
913 case CACHE_EXT_TREE:
914 istate->cache_tree = cache_tree_read(data, sz);
915 break;
916 default:
917 if (*ext < 'A' || 'Z' < *ext)
918 return error("index uses %.4s extension, which we do not understand",
919 ext);
920 fprintf(stderr, "ignoring %.4s extension\n", ext);
921 break;
922 }
923 return 0;
924}
925
926int read_index(struct index_state *istate)
927{
928 return read_index_from(istate, get_index_file());
929}
930
931/* remember to discard_cache() before reading a different cache! */
932int read_index_from(struct index_state *istate, const char *path)
933{
934 int fd, i;
935 struct stat st;
936 unsigned long offset;
937 struct cache_header *hdr;
938
939 errno = EBUSY;
940 if (istate->mmap)
941 return istate->cache_nr;
942
943 errno = ENOENT;
944 istate->timestamp = 0;
945 fd = open(path, O_RDONLY);
946 if (fd < 0) {
947 if (errno == ENOENT)
948 return 0;
949 die("index file open failed (%s)", strerror(errno));
950 }
951
952 if (fstat(fd, &st))
953 die("cannot stat the open index (%s)", strerror(errno));
954
955 errno = EINVAL;
956 istate->mmap_size = xsize_t(st.st_size);
957 if (istate->mmap_size < sizeof(struct cache_header) + 20)
958 die("index file smaller than expected");
959
960 istate->mmap = xmmap(NULL, istate->mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
961 close(fd);
962
963 hdr = istate->mmap;
964 if (verify_hdr(hdr, istate->mmap_size) < 0)
965 goto unmap;
966
967 istate->cache_nr = ntohl(hdr->hdr_entries);
968 istate->cache_alloc = alloc_nr(istate->cache_nr);
969 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
970
971 offset = sizeof(*hdr);
972 for (i = 0; i < istate->cache_nr; i++) {
973 struct cache_entry *ce;
974
975 ce = (struct cache_entry *)((char *)(istate->mmap) + offset);
976 offset = offset + ce_size(ce);
977 istate->cache[i] = ce;
978 }
979 istate->timestamp = st.st_mtime;
980 while (offset <= istate->mmap_size - 20 - 8) {
981 /* After an array of active_nr index entries,
982 * there can be arbitrary number of extended
983 * sections, each of which is prefixed with
984 * extension name (4-byte) and section length
985 * in 4-byte network byte order.
986 */
987 unsigned long extsize;
988 memcpy(&extsize, (char *)(istate->mmap) + offset + 4, 4);
989 extsize = ntohl(extsize);
990 if (read_index_extension(istate,
991 ((const char *) (istate->mmap)) + offset,
992 (char *) (istate->mmap) + offset + 8,
993 extsize) < 0)
994 goto unmap;
995 offset += 8;
996 offset += extsize;
997 }
998 return istate->cache_nr;
999
1000unmap:
1001 munmap(istate->mmap, istate->mmap_size);
1002 errno = EINVAL;
1003 die("index file corrupt");
1004}
1005
1006int discard_index(struct index_state *istate)
1007{
1008 int ret;
1009
1010 istate->cache_nr = 0;
1011 istate->cache_changed = 0;
1012 istate->timestamp = 0;
1013 cache_tree_free(&(istate->cache_tree));
1014 if (istate->mmap == NULL)
1015 return 0;
1016 ret = munmap(istate->mmap, istate->mmap_size);
1017 istate->mmap = NULL;
1018 istate->mmap_size = 0;
1019
1020 /* no need to throw away allocated active_cache */
1021 return ret;
1022}
1023
1024#define WRITE_BUFFER_SIZE 8192
1025static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1026static unsigned long write_buffer_len;
1027
1028static int ce_write_flush(SHA_CTX *context, int fd)
1029{
1030 unsigned int buffered = write_buffer_len;
1031 if (buffered) {
1032 SHA1_Update(context, write_buffer, buffered);
1033 if (write_in_full(fd, write_buffer, buffered) != buffered)
1034 return -1;
1035 write_buffer_len = 0;
1036 }
1037 return 0;
1038}
1039
1040static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1041{
1042 while (len) {
1043 unsigned int buffered = write_buffer_len;
1044 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1045 if (partial > len)
1046 partial = len;
1047 memcpy(write_buffer + buffered, data, partial);
1048 buffered += partial;
1049 if (buffered == WRITE_BUFFER_SIZE) {
1050 write_buffer_len = buffered;
1051 if (ce_write_flush(context, fd))
1052 return -1;
1053 buffered = 0;
1054 }
1055 write_buffer_len = buffered;
1056 len -= partial;
1057 data = (char *) data + partial;
1058 }
1059 return 0;
1060}
1061
1062static int write_index_ext_header(SHA_CTX *context, int fd,
1063 unsigned int ext, unsigned int sz)
1064{
1065 ext = htonl(ext);
1066 sz = htonl(sz);
1067 return ((ce_write(context, fd, &ext, 4) < 0) ||
1068 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1069}
1070
1071static int ce_flush(SHA_CTX *context, int fd)
1072{
1073 unsigned int left = write_buffer_len;
1074
1075 if (left) {
1076 write_buffer_len = 0;
1077 SHA1_Update(context, write_buffer, left);
1078 }
1079
1080 /* Flush first if not enough space for SHA1 signature */
1081 if (left + 20 > WRITE_BUFFER_SIZE) {
1082 if (write_in_full(fd, write_buffer, left) != left)
1083 return -1;
1084 left = 0;
1085 }
1086
1087 /* Append the SHA1 signature at the end */
1088 SHA1_Final(write_buffer + left, context);
1089 left += 20;
1090 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1091}
1092
1093static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1094{
1095 /*
1096 * The only thing we care about in this function is to smudge the
1097 * falsely clean entry due to touch-update-touch race, so we leave
1098 * everything else as they are. We are called for entries whose
1099 * ce_mtime match the index file mtime.
1100 */
1101 struct stat st;
1102
1103 if (lstat(ce->name, &st) < 0)
1104 return;
1105 if (ce_match_stat_basic(ce, &st))
1106 return;
1107 if (ce_modified_check_fs(ce, &st)) {
1108 /* This is "racily clean"; smudge it. Note that this
1109 * is a tricky code. At first glance, it may appear
1110 * that it can break with this sequence:
1111 *
1112 * $ echo xyzzy >frotz
1113 * $ git-update-index --add frotz
1114 * $ : >frotz
1115 * $ sleep 3
1116 * $ echo filfre >nitfol
1117 * $ git-update-index --add nitfol
1118 *
1119 * but it does not. When the second update-index runs,
1120 * it notices that the entry "frotz" has the same timestamp
1121 * as index, and if we were to smudge it by resetting its
1122 * size to zero here, then the object name recorded
1123 * in index is the 6-byte file but the cached stat information
1124 * becomes zero --- which would then match what we would
1125 * obtain from the filesystem next time we stat("frotz").
1126 *
1127 * However, the second update-index, before calling
1128 * this function, notices that the cached size is 6
1129 * bytes and what is on the filesystem is an empty
1130 * file, and never calls us, so the cached size information
1131 * for "frotz" stays 6 which does not match the filesystem.
1132 */
1133 ce->ce_size = htonl(0);
1134 }
1135}
1136
1137int write_index(struct index_state *istate, int newfd)
1138{
1139 SHA_CTX c;
1140 struct cache_header hdr;
1141 int i, err, removed;
1142 struct cache_entry **cache = istate->cache;
1143 int entries = istate->cache_nr;
1144
1145 for (i = removed = 0; i < entries; i++)
1146 if (!cache[i]->ce_mode)
1147 removed++;
1148
1149 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1150 hdr.hdr_version = htonl(2);
1151 hdr.hdr_entries = htonl(entries - removed);
1152
1153 SHA1_Init(&c);
1154 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1155 return -1;
1156
1157 for (i = 0; i < entries; i++) {
1158 struct cache_entry *ce = cache[i];
1159 if (!ce->ce_mode)
1160 continue;
1161 if (istate->timestamp &&
1162 istate->timestamp <= ntohl(ce->ce_mtime.sec))
1163 ce_smudge_racily_clean_entry(ce);
1164 if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
1165 return -1;
1166 }
1167
1168 /* Write extension data here */
1169 if (istate->cache_tree) {
1170 struct strbuf sb;
1171
1172 strbuf_init(&sb, 0);
1173 cache_tree_write(&sb, istate->cache_tree);
1174 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1175 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1176 strbuf_release(&sb);
1177 if (err)
1178 return -1;
1179 }
1180 return ce_flush(&c, newfd);
1181}