Extract helper bits from c-merge-recursive work
[git/git.git] / read-cache.c
CommitLineData
8bc9a0c7
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
e83c5163 6#include "cache.h"
bad68ec9
JH
7#include "cache-tree.h"
8
9/* Index extensions.
10 *
11 * The first letter should be 'A'..'Z' for extensions that are not
12 * necessary for a correct operation (i.e. optimization data).
13 * When new extensions are added that _needs_ to be understood in
14 * order to correctly interpret the index file, pick character that
15 * is outside the range, to cause the reader to abort.
16 */
17
18#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
19#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
e83c5163 20
e83c5163 21struct cache_entry **active_cache = NULL;
29e4d363 22static time_t index_file_timestamp;
ee267527 23unsigned int active_nr = 0, active_alloc = 0, active_cache_changed = 0;
e83c5163 24
bad68ec9
JH
25struct cache_tree *active_cache_tree = NULL;
26
8fd2cb40
JS
27int cache_errno = 0;
28
29static void *cache_mmap = NULL;
30static size_t cache_mmap_size = 0;
31
415e96c8
JH
32/*
33 * This only updates the "non-critical" parts of the directory
34 * cache, ie the parts that aren't tracked by GIT, and only used
35 * to validate the cache.
36 */
37void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
38{
39 ce->ce_ctime.sec = htonl(st->st_ctime);
40 ce->ce_mtime.sec = htonl(st->st_mtime);
2cb45e95 41#ifdef USE_NSEC
415e96c8
JH
42 ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
43 ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
44#endif
45 ce->ce_dev = htonl(st->st_dev);
46 ce->ce_ino = htonl(st->st_ino);
47 ce->ce_uid = htonl(st->st_uid);
48 ce->ce_gid = htonl(st->st_gid);
49 ce->ce_size = htonl(st->st_size);
5f73076c
JH
50
51 if (assume_unchanged)
52 ce->ce_flags |= htons(CE_VALID);
415e96c8
JH
53}
54
29e4d363
JH
55static int ce_compare_data(struct cache_entry *ce, struct stat *st)
56{
57 int match = -1;
58 int fd = open(ce->name, O_RDONLY);
59
60 if (fd >= 0) {
61 unsigned char sha1[20];
62 if (!index_fd(sha1, fd, st, 0, NULL))
63 match = memcmp(sha1, ce->sha1, 20);
64 close(fd);
65 }
66 return match;
67}
68
69static int ce_compare_link(struct cache_entry *ce, unsigned long expected_size)
70{
71 int match = -1;
72 char *target;
73 void *buffer;
74 unsigned long size;
75 char type[10];
76 int len;
77
78 target = xmalloc(expected_size);
79 len = readlink(ce->name, target, expected_size);
80 if (len != expected_size) {
81 free(target);
82 return -1;
83 }
84 buffer = read_sha1_file(ce->sha1, type, &size);
85 if (!buffer) {
86 free(target);
87 return -1;
88 }
89 if (size == expected_size)
90 match = memcmp(buffer, target, size);
91 free(buffer);
92 free(target);
93 return match;
94}
95
96static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
97{
98 switch (st->st_mode & S_IFMT) {
99 case S_IFREG:
100 if (ce_compare_data(ce, st))
101 return DATA_CHANGED;
102 break;
103 case S_IFLNK:
104 if (ce_compare_link(ce, st->st_size))
105 return DATA_CHANGED;
106 break;
107 default:
108 return TYPE_CHANGED;
109 }
110 return 0;
111}
112
407c8eb0 113static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
734aab75
LT
114{
115 unsigned int changed = 0;
116
8ae0a8c5
KS
117 switch (ntohl(ce->ce_mode) & S_IFMT) {
118 case S_IFREG:
119 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
120 /* We consider only the owner x bit to be relevant for
121 * "mode changes"
122 */
123 if (trust_executable_bit &&
124 (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
ffbe1add 125 changed |= MODE_CHANGED;
8ae0a8c5
KS
126 break;
127 case S_IFLNK:
128 changed |= !S_ISLNK(st->st_mode) ? TYPE_CHANGED : 0;
129 break;
130 default:
131 die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
132 }
ccc4feb5 133 if (ce->ce_mtime.sec != htonl(st->st_mtime))
734aab75 134 changed |= MTIME_CHANGED;
ccc4feb5
LT
135 if (ce->ce_ctime.sec != htonl(st->st_ctime))
136 changed |= CTIME_CHANGED;
137
2cb45e95 138#ifdef USE_NSEC
ccc4feb5
LT
139 /*
140 * nsec seems unreliable - not all filesystems support it, so
141 * as long as it is in the inode cache you get right nsec
142 * but after it gets flushed, you get zero nsec.
143 */
94dfb7f2 144 if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
ccc4feb5 145 changed |= MTIME_CHANGED;
94dfb7f2 146 if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
734aab75 147 changed |= CTIME_CHANGED;
ccc4feb5
LT
148#endif
149
150 if (ce->ce_uid != htonl(st->st_uid) ||
151 ce->ce_gid != htonl(st->st_gid))
734aab75 152 changed |= OWNER_CHANGED;
2cb45e95 153 if (ce->ce_ino != htonl(st->st_ino))
734aab75 154 changed |= INODE_CHANGED;
2cb45e95
LT
155
156#ifdef USE_STDEV
157 /*
158 * st_dev breaks on network filesystems where different
159 * clients will have different views of what "device"
160 * the filesystem is on
161 */
162 if (ce->ce_dev != htonl(st->st_dev))
163 changed |= INODE_CHANGED;
164#endif
165
ccc4feb5 166 if (ce->ce_size != htonl(st->st_size))
734aab75 167 changed |= DATA_CHANGED;
b0391890 168
407c8eb0
JH
169 return changed;
170}
171
5f73076c 172int ce_match_stat(struct cache_entry *ce, struct stat *st, int ignore_valid)
407c8eb0 173{
5f73076c
JH
174 unsigned int changed;
175
176 /*
177 * If it's marked as always valid in the index, it's
178 * valid whatever the checked-out copy says.
179 */
180 if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
181 return 0;
182
183 changed = ce_match_stat_basic(ce, st);
407c8eb0 184
29e4d363
JH
185 /*
186 * Within 1 second of this sequence:
187 * echo xyzzy >file && git-update-index --add file
188 * running this command:
189 * echo frotz >file
190 * would give a falsely clean cache entry. The mtime and
191 * length match the cache, and other stat fields do not change.
192 *
193 * We could detect this at update-index time (the cache entry
194 * being registered/updated records the same time as "now")
195 * and delay the return from git-update-index, but that would
196 * effectively mean we can make at most one commit per second,
197 * which is not acceptable. Instead, we check cache entries
198 * whose mtime are the same as the index file timestamp more
5f73076c 199 * carefully than others.
29e4d363
JH
200 */
201 if (!changed &&
202 index_file_timestamp &&
203 index_file_timestamp <= ntohl(ce->ce_mtime.sec))
204 changed |= ce_modified_check_fs(ce, st);
b0391890 205
29e4d363 206 return changed;
b0391890
JH
207}
208
5f73076c 209int ce_modified(struct cache_entry *ce, struct stat *st, int really)
b0391890 210{
29e4d363 211 int changed, changed_fs;
5f73076c 212 changed = ce_match_stat(ce, st, really);
b0391890
JH
213 if (!changed)
214 return 0;
b0391890
JH
215 /*
216 * If the mode or type has changed, there's no point in trying
217 * to refresh the entry - it's not going to match
218 */
219 if (changed & (MODE_CHANGED | TYPE_CHANGED))
220 return changed;
221
222 /* Immediately after read-tree or update-index --cacheinfo,
223 * the length field is zero. For other cases the ce_size
224 * should match the SHA1 recorded in the index entry.
225 */
226 if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
227 return changed;
228
29e4d363
JH
229 changed_fs = ce_modified_check_fs(ce, st);
230 if (changed_fs)
231 return changed | changed_fs;
b0391890
JH
232 return 0;
233}
234
958ba6c9
LT
235int base_name_compare(const char *name1, int len1, int mode1,
236 const char *name2, int len2, int mode2)
237{
238 unsigned char c1, c2;
239 int len = len1 < len2 ? len1 : len2;
240 int cmp;
241
242 cmp = memcmp(name1, name2, len);
243 if (cmp)
244 return cmp;
245 c1 = name1[len];
246 c2 = name2[len];
247 if (!c1 && S_ISDIR(mode1))
248 c1 = '/';
249 if (!c2 && S_ISDIR(mode2))
250 c2 = '/';
251 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
252}
253
95fd5bf8 254int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
eb38c22f 255{
95fd5bf8
LT
256 int len1 = flags1 & CE_NAMEMASK;
257 int len2 = flags2 & CE_NAMEMASK;
eb38c22f
LT
258 int len = len1 < len2 ? len1 : len2;
259 int cmp;
260
261 cmp = memcmp(name1, name2, len);
262 if (cmp)
263 return cmp;
264 if (len1 < len2)
265 return -1;
266 if (len1 > len2)
267 return 1;
5f73076c 268
7b80be15
JH
269 /* Compare stages */
270 flags1 &= CE_STAGEMASK;
271 flags2 &= CE_STAGEMASK;
5f73076c 272
95fd5bf8
LT
273 if (flags1 < flags2)
274 return -1;
275 if (flags1 > flags2)
276 return 1;
eb38c22f
LT
277 return 0;
278}
279
280int cache_name_pos(const char *name, int namelen)
281{
282 int first, last;
283
284 first = 0;
285 last = active_nr;
286 while (last > first) {
287 int next = (last + first) >> 1;
288 struct cache_entry *ce = active_cache[next];
972d1bb0 289 int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
eb38c22f 290 if (!cmp)
76e7f4ec 291 return next;
eb38c22f
LT
292 if (cmp < 0) {
293 last = next;
294 continue;
295 }
296 first = next+1;
297 }
76e7f4ec 298 return -first-1;
eb38c22f
LT
299}
300
7b937ca3 301/* Remove entry, return true if there are more entries to go.. */
dbbce55b 302int remove_cache_entry_at(int pos)
7b937ca3 303{
ee267527 304 active_cache_changed = 1;
7b937ca3
LT
305 active_nr--;
306 if (pos >= active_nr)
307 return 0;
308 memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
309 return 1;
310}
311
6b5ee137 312int remove_file_from_cache(const char *path)
197ee8c9
LT
313{
314 int pos = cache_name_pos(path, strlen(path));
c4e3cca1
JH
315 if (pos < 0)
316 pos = -pos-1;
317 while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
dbbce55b 318 remove_cache_entry_at(pos);
197ee8c9
LT
319 return 0;
320}
321
dbbce55b 322int ce_same_name(struct cache_entry *a, struct cache_entry *b)
7b937ca3
LT
323{
324 int len = ce_namelen(a);
325 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
326}
327
c0fd1f51
LT
328int ce_path_match(const struct cache_entry *ce, const char **pathspec)
329{
330 const char *match, *name;
331 int len;
332
333 if (!pathspec)
334 return 1;
335
336 len = ce_namelen(ce);
337 name = ce->name;
338 while ((match = *pathspec++) != NULL) {
339 int matchlen = strlen(match);
340 if (matchlen > len)
341 continue;
342 if (memcmp(name, match, matchlen))
343 continue;
344 if (matchlen && name[matchlen-1] == '/')
345 return 1;
346 if (name[matchlen] == '/' || !name[matchlen])
347 return 1;
f332726e
LT
348 if (!matchlen)
349 return 1;
c0fd1f51
LT
350 }
351 return 0;
352}
353
8dcf39c4
LT
354/*
355 * We fundamentally don't like some paths: we don't want
356 * dot or dot-dot anywhere, and for obvious reasons don't
357 * want to recurse into ".git" either.
358 *
359 * Also, we don't want double slashes or slashes at the
360 * end that can make pathnames ambiguous.
361 */
362static int verify_dotfile(const char *rest)
363{
364 /*
365 * The first character was '.', but that
366 * has already been discarded, we now test
367 * the rest.
368 */
369 switch (*rest) {
370 /* "." is not allowed */
371 case '\0': case '/':
372 return 0;
373
374 /*
375 * ".git" followed by NUL or slash is bad. This
376 * shares the path end test with the ".." case.
377 */
378 case 'g':
379 if (rest[1] != 'i')
380 break;
381 if (rest[2] != 't')
382 break;
383 rest += 2;
384 /* fallthrough */
385 case '.':
386 if (rest[1] == '\0' || rest[1] == '/')
387 return 0;
388 }
389 return 1;
390}
391
392int verify_path(const char *path)
393{
394 char c;
395
396 goto inside;
397 for (;;) {
398 if (!c)
399 return 1;
400 if (c == '/') {
401inside:
402 c = *path++;
403 switch (c) {
404 default:
405 continue;
406 case '/': case '\0':
407 break;
408 case '.':
409 if (verify_dotfile(path))
410 continue;
411 }
412 return 0;
413 }
414 c = *path++;
415 }
416}
417
12676608
LT
418/*
419 * Do we have another file that has the beginning components being a
420 * proper superset of the name we're trying to add?
0f1e4f04 421 */
12676608 422static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 423{
12676608
LT
424 int retval = 0;
425 int len = ce_namelen(ce);
b155725d 426 int stage = ce_stage(ce);
12676608 427 const char *name = ce->name;
0f1e4f04 428
12676608
LT
429 while (pos < active_nr) {
430 struct cache_entry *p = active_cache[pos++];
0f1e4f04 431
12676608 432 if (len >= ce_namelen(p))
0f1e4f04 433 break;
12676608
LT
434 if (memcmp(name, p->name, len))
435 break;
b155725d
JH
436 if (ce_stage(p) != stage)
437 continue;
12676608
LT
438 if (p->name[len] != '/')
439 continue;
12676608
LT
440 retval = -1;
441 if (!ok_to_replace)
442 break;
443 remove_cache_entry_at(--pos);
0f1e4f04 444 }
12676608
LT
445 return retval;
446}
0f1e4f04 447
12676608
LT
448/*
449 * Do we have another file with a pathname that is a proper
450 * subset of the name we're trying to add?
451 */
452static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace)
453{
454 int retval = 0;
b155725d 455 int stage = ce_stage(ce);
12676608
LT
456 const char *name = ce->name;
457 const char *slash = name + ce_namelen(ce);
0f1e4f04 458
12676608
LT
459 for (;;) {
460 int len;
0f1e4f04 461
12676608
LT
462 for (;;) {
463 if (*--slash == '/')
464 break;
465 if (slash <= ce->name)
466 return retval;
467 }
468 len = slash - name;
0f1e4f04 469
b155725d 470 pos = cache_name_pos(name, ntohs(create_ce_flags(len, stage)));
12676608
LT
471 if (pos >= 0) {
472 retval = -1;
473 if (ok_to_replace)
474 break;
dbbce55b 475 remove_cache_entry_at(pos);
12676608
LT
476 continue;
477 }
478
479 /*
480 * Trivial optimization: if we find an entry that
481 * already matches the sub-directory, then we know
b155725d 482 * we're ok, and we can exit.
12676608
LT
483 */
484 pos = -pos-1;
b155725d 485 while (pos < active_nr) {
12676608 486 struct cache_entry *p = active_cache[pos];
b155725d
JH
487 if ((ce_namelen(p) <= len) ||
488 (p->name[len] != '/') ||
489 memcmp(p->name, name, len))
490 break; /* not our subdirectory */
491 if (ce_stage(p) == stage)
492 /* p is at the same stage as our entry, and
493 * is a subdirectory of what we are looking
494 * at, so we cannot have conflicts at our
495 * level or anything shorter.
496 */
497 return retval;
498 pos++;
192268c1 499 }
0f1e4f04 500 }
12676608
LT
501 return retval;
502}
503
504/* We may be in a situation where we already have path/file and path
505 * is being added, or we already have path and path/file is being
506 * added. Either one would result in a nonsense tree that has path
507 * twice when git-write-tree tries to write it out. Prevent it.
508 *
509 * If ok-to-replace is specified, we remove the conflicting entries
510 * from the cache so the caller should recompute the insert position.
511 * When this happens, we return non-zero.
512 */
513static int check_file_directory_conflict(const struct cache_entry *ce, int pos, int ok_to_replace)
514{
515 /*
516 * We check if the path is a sub-path of a subsequent pathname
517 * first, since removing those will not change the position
518 * in the array
519 */
520 int retval = has_file_name(ce, pos, ok_to_replace);
521 /*
522 * Then check if the path might have a clashing sub-directory
523 * before it.
524 */
525 return retval + has_dir_name(ce, pos, ok_to_replace);
0f1e4f04
JH
526}
527
192268c1 528int add_cache_entry(struct cache_entry *ce, int option)
197ee8c9
LT
529{
530 int pos;
192268c1
JH
531 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
532 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 533 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
5f73076c 534
972d1bb0 535 pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
197ee8c9 536
3e09cdfd 537 /* existing match? Just replace it. */
76e7f4ec 538 if (pos >= 0) {
ee267527 539 active_cache_changed = 1;
76e7f4ec 540 active_cache[pos] = ce;
197ee8c9
LT
541 return 0;
542 }
76e7f4ec 543 pos = -pos-1;
197ee8c9 544
7b937ca3
LT
545 /*
546 * Inserting a merged entry ("stage 0") into the index
547 * will always replace all non-merged entries..
548 */
549 if (pos < active_nr && ce_stage(ce) == 0) {
dbbce55b 550 while (ce_same_name(active_cache[pos], ce)) {
7b937ca3 551 ok_to_add = 1;
dbbce55b 552 if (!remove_cache_entry_at(pos))
7b937ca3
LT
553 break;
554 }
555 }
556
121481ab
LT
557 if (!ok_to_add)
558 return -1;
8dcf39c4
LT
559 if (!verify_path(ce->name))
560 return -1;
121481ab 561
3e09cdfd
JH
562 if (!skip_df_check &&
563 check_file_directory_conflict(ce, pos, ok_to_replace)) {
192268c1
JH
564 if (!ok_to_replace)
565 return -1;
972d1bb0 566 pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
192268c1
JH
567 pos = -pos-1;
568 }
0f1e4f04 569
197ee8c9
LT
570 /* Make sure the array is big enough .. */
571 if (active_nr == active_alloc) {
572 active_alloc = alloc_nr(active_alloc);
812666c8 573 active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
197ee8c9
LT
574 }
575
576 /* Add it in.. */
577 active_nr++;
578 if (active_nr > pos)
579 memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
580 active_cache[pos] = ce;
ee267527 581 active_cache_changed = 1;
197ee8c9
LT
582 return 0;
583}
584
405e5b2f
LT
585/*
586 * "refresh" does not calculate a new sha1 file or bring the
587 * cache up-to-date for mode/content changes. But what it
588 * _does_ do is to "re-match" the stat information of a file
589 * with the cache, so that you can refresh the cache for a
590 * file that hasn't been changed but where the stat entry is
591 * out of date.
592 *
593 * For example, you'd want to do this after doing a "git-read-tree",
594 * to link up the stat cache details with the proper files.
595 */
8fd2cb40 596struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
405e5b2f
LT
597{
598 struct stat st;
599 struct cache_entry *updated;
600 int changed, size;
601
8fd2cb40
JS
602 if (lstat(ce->name, &st) < 0) {
603 cache_errno = errno;
604 return NULL;
605 }
405e5b2f
LT
606
607 changed = ce_match_stat(ce, &st, really);
608 if (!changed) {
609 if (really && assume_unchanged &&
610 !(ce->ce_flags & htons(CE_VALID)))
611 ; /* mark this one VALID again */
612 else
8fd2cb40 613 return ce;
405e5b2f
LT
614 }
615
8fd2cb40
JS
616 if (ce_modified(ce, &st, really)) {
617 cache_errno = EINVAL;
618 return NULL;
619 }
405e5b2f
LT
620
621 size = ce_size(ce);
622 updated = xmalloc(size);
623 memcpy(updated, ce, size);
624 fill_stat_cache_info(updated, &st);
625
626 /* In this case, if really is not set, we should leave
627 * CE_VALID bit alone. Otherwise, paths marked with
628 * --no-assume-unchanged (i.e. things to be edited) will
629 * reacquire CE_VALID bit automatically, which is not
630 * really what we want.
631 */
632 if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
633 updated->ce_flags &= ~htons(CE_VALID);
634
635 return updated;
636}
637
638int refresh_cache(unsigned int flags)
639{
640 int i;
641 int has_errors = 0;
642 int really = (flags & REFRESH_REALLY) != 0;
643 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
644 int quiet = (flags & REFRESH_QUIET) != 0;
645 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
646
647 for (i = 0; i < active_nr; i++) {
648 struct cache_entry *ce, *new;
649 ce = active_cache[i];
650 if (ce_stage(ce)) {
651 while ((i < active_nr) &&
652 ! strcmp(active_cache[i]->name, ce->name))
653 i++;
654 i--;
655 if (allow_unmerged)
656 continue;
657 printf("%s: needs merge\n", ce->name);
658 has_errors = 1;
659 continue;
660 }
661
8fd2cb40
JS
662 new = refresh_cache_entry(ce, really);
663 if (new == ce)
405e5b2f 664 continue;
8fd2cb40
JS
665 if (!new) {
666 if (not_new && cache_errno == ENOENT)
405e5b2f 667 continue;
8fd2cb40 668 if (really && cache_errno == EINVAL) {
405e5b2f
LT
669 /* If we are doing --really-refresh that
670 * means the index is not valid anymore.
671 */
672 ce->ce_flags &= ~htons(CE_VALID);
673 active_cache_changed = 1;
674 }
675 if (quiet)
676 continue;
677 printf("%s: needs update\n", ce->name);
678 has_errors = 1;
679 continue;
680 }
681 active_cache_changed = 1;
682 /* You can NOT just free active_cache[i] here, since it
683 * might not be necessarily malloc()ed but can also come
684 * from mmap(). */
685 active_cache[i] = new;
686 }
687 return has_errors;
688}
689
bad68ec9 690static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163
LT
691{
692 SHA_CTX c;
bad68ec9 693 unsigned char sha1[20];
e83c5163 694
ccc4feb5 695 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 696 return error("bad signature");
ca9be054
LT
697 if (hdr->hdr_version != htonl(2))
698 return error("bad index version");
e83c5163 699 SHA1_Init(&c);
ca9be054 700 SHA1_Update(&c, hdr, size - 20);
e83c5163 701 SHA1_Final(sha1, &c);
1d7f171c 702 if (memcmp(sha1, (char *) hdr + size - 20, 20))
ca9be054 703 return error("bad index file sha1 signature");
e83c5163
LT
704 return 0;
705}
706
bad68ec9
JH
707static int read_index_extension(const char *ext, void *data, unsigned long sz)
708{
709 switch (CACHE_EXT(ext)) {
710 case CACHE_EXT_TREE:
711 active_cache_tree = cache_tree_read(data, sz);
712 break;
713 default:
714 if (*ext < 'A' || 'Z' < *ext)
715 return error("index uses %.4s extension, which we do not understand",
716 ext);
717 fprintf(stderr, "ignoring %.4s extension\n", ext);
718 break;
719 }
720 return 0;
721}
722
723int read_cache(void)
8fd2cb40
JS
724{
725 return read_cache_from(get_index_file());
726}
727
728/* remember to discard_cache() before reading a different cache! */
729int read_cache_from(const char *path)
e83c5163
LT
730{
731 int fd, i;
732 struct stat st;
8fd2cb40 733 unsigned long offset;
e83c5163
LT
734 struct cache_header *hdr;
735
736 errno = EBUSY;
8fd2cb40 737 if (cache_mmap)
5d1a5c02
LT
738 return active_nr;
739
e83c5163 740 errno = ENOENT;
29e4d363 741 index_file_timestamp = 0;
8fd2cb40 742 fd = open(path, O_RDONLY);
5d1a5c02
LT
743 if (fd < 0) {
744 if (errno == ENOENT)
745 return 0;
746 die("index file open failed (%s)", strerror(errno));
747 }
e83c5163 748
8fd2cb40 749 cache_mmap = MAP_FAILED;
e83c5163 750 if (!fstat(fd, &st)) {
8fd2cb40 751 cache_mmap_size = st.st_size;
e83c5163 752 errno = EINVAL;
8fd2cb40
JS
753 if (cache_mmap_size >= sizeof(struct cache_header) + 20)
754 cache_mmap = mmap(NULL, cache_mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
e83c5163
LT
755 }
756 close(fd);
8fd2cb40 757 if (cache_mmap == MAP_FAILED)
5d1a5c02 758 die("index file mmap failed (%s)", strerror(errno));
e83c5163 759
8fd2cb40
JS
760 hdr = cache_mmap;
761 if (verify_hdr(hdr, cache_mmap_size) < 0)
e83c5163
LT
762 goto unmap;
763
ccc4feb5 764 active_nr = ntohl(hdr->hdr_entries);
e83c5163 765 active_alloc = alloc_nr(active_nr);
28cc4ab4 766 active_cache = xcalloc(active_alloc, sizeof(struct cache_entry *));
e83c5163
LT
767
768 offset = sizeof(*hdr);
ccc4feb5 769 for (i = 0; i < active_nr; i++) {
8fd2cb40 770 struct cache_entry *ce = (struct cache_entry *) ((char *) cache_mmap + offset);
e83c5163
LT
771 offset = offset + ce_size(ce);
772 active_cache[i] = ce;
773 }
29e4d363 774 index_file_timestamp = st.st_mtime;
8fd2cb40 775 while (offset <= cache_mmap_size - 20 - 8) {
bad68ec9
JH
776 /* After an array of active_nr index entries,
777 * there can be arbitrary number of extended
778 * sections, each of which is prefixed with
779 * extension name (4-byte) and section length
780 * in 4-byte network byte order.
781 */
782 unsigned long extsize;
8fd2cb40 783 memcpy(&extsize, (char *) cache_mmap + offset + 4, 4);
bad68ec9 784 extsize = ntohl(extsize);
8fd2cb40
JS
785 if (read_index_extension(((const char *) cache_mmap) + offset,
786 (char *) cache_mmap + offset + 8,
1d7f171c 787 extsize) < 0)
bad68ec9
JH
788 goto unmap;
789 offset += 8;
790 offset += extsize;
791 }
e83c5163
LT
792 return active_nr;
793
794unmap:
8fd2cb40 795 munmap(cache_mmap, cache_mmap_size);
e83c5163 796 errno = EINVAL;
5d1a5c02 797 die("index file corrupt");
e83c5163
LT
798}
799
4990aadc 800#define WRITE_BUFFER_SIZE 8192
bf0f910d 801static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
802static unsigned long write_buffer_len;
803
ca9be054 804static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
805{
806 while (len) {
807 unsigned int buffered = write_buffer_len;
808 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
809 if (partial > len)
810 partial = len;
811 memcpy(write_buffer + buffered, data, partial);
812 buffered += partial;
813 if (buffered == WRITE_BUFFER_SIZE) {
ca9be054 814 SHA1_Update(context, write_buffer, WRITE_BUFFER_SIZE);
4990aadc
LT
815 if (write(fd, write_buffer, WRITE_BUFFER_SIZE) != WRITE_BUFFER_SIZE)
816 return -1;
817 buffered = 0;
818 }
819 write_buffer_len = buffered;
820 len -= partial;
1d7f171c 821 data = (char *) data + partial;
4990aadc
LT
822 }
823 return 0;
824}
825
bad68ec9 826static int write_index_ext_header(SHA_CTX *context, int fd,
ac58c7b1 827 unsigned int ext, unsigned int sz)
bad68ec9
JH
828{
829 ext = htonl(ext);
830 sz = htonl(sz);
831 if ((ce_write(context, fd, &ext, 4) < 0) ||
832 (ce_write(context, fd, &sz, 4) < 0))
833 return -1;
834 return 0;
835}
836
837static int ce_flush(SHA_CTX *context, int fd)
4990aadc
LT
838{
839 unsigned int left = write_buffer_len;
ca9be054 840
4990aadc
LT
841 if (left) {
842 write_buffer_len = 0;
ca9be054 843 SHA1_Update(context, write_buffer, left);
4990aadc 844 }
ca9be054 845
2c865d9a
QH
846 /* Flush first if not enough space for SHA1 signature */
847 if (left + 20 > WRITE_BUFFER_SIZE) {
848 if (write(fd, write_buffer, left) != left)
849 return -1;
850 left = 0;
851 }
852
ca9be054 853 /* Append the SHA1 signature at the end */
bad68ec9 854 SHA1_Final(write_buffer + left, context);
ca9be054
LT
855 left += 20;
856 if (write(fd, write_buffer, left) != left)
857 return -1;
4990aadc
LT
858 return 0;
859}
860
407c8eb0
JH
861static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
862{
863 /*
864 * The only thing we care about in this function is to smudge the
865 * falsely clean entry due to touch-update-touch race, so we leave
866 * everything else as they are. We are called for entries whose
867 * ce_mtime match the index file mtime.
868 */
869 struct stat st;
870
871 if (lstat(ce->name, &st) < 0)
872 return;
873 if (ce_match_stat_basic(ce, &st))
874 return;
875 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
876 /* This is "racily clean"; smudge it. Note that this
877 * is a tricky code. At first glance, it may appear
878 * that it can break with this sequence:
879 *
880 * $ echo xyzzy >frotz
881 * $ git-update-index --add frotz
882 * $ : >frotz
883 * $ sleep 3
884 * $ echo filfre >nitfol
885 * $ git-update-index --add nitfol
886 *
887 * but it does not. Whe the second update-index runs,
888 * it notices that the entry "frotz" has the same timestamp
889 * as index, and if we were to smudge it by resetting its
890 * size to zero here, then the object name recorded
891 * in index is the 6-byte file but the cached stat information
892 * becomes zero --- which would then match what we would
893 * obtain from the filesystem next time we stat("frotz").
894 *
895 * However, the second update-index, before calling
896 * this function, notices that the cached size is 6
897 * bytes and what is on the filesystem is an empty
898 * file, and never calls us, so the cached size information
899 * for "frotz" stays 6 which does not match the filesystem.
900 */
407c8eb0
JH
901 ce->ce_size = htonl(0);
902 }
903}
904
bad68ec9 905int write_cache(int newfd, struct cache_entry **cache, int entries)
197ee8c9
LT
906{
907 SHA_CTX c;
908 struct cache_header hdr;
025a0709
JH
909 int i, removed;
910
911 for (i = removed = 0; i < entries; i++)
912 if (!cache[i]->ce_mode)
913 removed++;
197ee8c9 914
ccc4feb5 915 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
ca9be054 916 hdr.hdr_version = htonl(2);
025a0709 917 hdr.hdr_entries = htonl(entries - removed);
197ee8c9
LT
918
919 SHA1_Init(&c);
ca9be054 920 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
921 return -1;
922
923 for (i = 0; i < entries; i++) {
924 struct cache_entry *ce = cache[i];
aa16021e
LT
925 if (!ce->ce_mode)
926 continue;
407c8eb0
JH
927 if (index_file_timestamp &&
928 index_file_timestamp <= ntohl(ce->ce_mtime.sec))
929 ce_smudge_racily_clean_entry(ce);
ca9be054 930 if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
197ee8c9
LT
931 return -1;
932 }
1af1c2b6 933
bad68ec9
JH
934 /* Write extension data here */
935 if (active_cache_tree) {
936 unsigned long sz;
937 void *data = cache_tree_write(active_cache_tree, &sz);
938 if (data &&
939 !write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sz) &&
940 !ce_write(&c, newfd, data, sz))
941 ;
942 else {
943 free(data);
944 return -1;
945 }
946 }
947 return ce_flush(&c, newfd);
197ee8c9 948}