Racy GIT
[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
LT
6#include "cache.h"
7
e83c5163 8struct cache_entry **active_cache = NULL;
29e4d363 9static time_t index_file_timestamp;
ee267527 10unsigned int active_nr = 0, active_alloc = 0, active_cache_changed = 0;
e83c5163 11
415e96c8
JH
12/*
13 * This only updates the "non-critical" parts of the directory
14 * cache, ie the parts that aren't tracked by GIT, and only used
15 * to validate the cache.
16 */
17void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
18{
19 ce->ce_ctime.sec = htonl(st->st_ctime);
20 ce->ce_mtime.sec = htonl(st->st_mtime);
2cb45e95 21#ifdef USE_NSEC
415e96c8
JH
22 ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
23 ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
24#endif
25 ce->ce_dev = htonl(st->st_dev);
26 ce->ce_ino = htonl(st->st_ino);
27 ce->ce_uid = htonl(st->st_uid);
28 ce->ce_gid = htonl(st->st_gid);
29 ce->ce_size = htonl(st->st_size);
30}
31
29e4d363
JH
32static int ce_compare_data(struct cache_entry *ce, struct stat *st)
33{
34 int match = -1;
35 int fd = open(ce->name, O_RDONLY);
36
37 if (fd >= 0) {
38 unsigned char sha1[20];
39 if (!index_fd(sha1, fd, st, 0, NULL))
40 match = memcmp(sha1, ce->sha1, 20);
41 close(fd);
42 }
43 return match;
44}
45
46static int ce_compare_link(struct cache_entry *ce, unsigned long expected_size)
47{
48 int match = -1;
49 char *target;
50 void *buffer;
51 unsigned long size;
52 char type[10];
53 int len;
54
55 target = xmalloc(expected_size);
56 len = readlink(ce->name, target, expected_size);
57 if (len != expected_size) {
58 free(target);
59 return -1;
60 }
61 buffer = read_sha1_file(ce->sha1, type, &size);
62 if (!buffer) {
63 free(target);
64 return -1;
65 }
66 if (size == expected_size)
67 match = memcmp(buffer, target, size);
68 free(buffer);
69 free(target);
70 return match;
71}
72
73static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
74{
75 switch (st->st_mode & S_IFMT) {
76 case S_IFREG:
77 if (ce_compare_data(ce, st))
78 return DATA_CHANGED;
79 break;
80 case S_IFLNK:
81 if (ce_compare_link(ce, st->st_size))
82 return DATA_CHANGED;
83 break;
84 default:
85 return TYPE_CHANGED;
86 }
87 return 0;
88}
89
5d728c84 90int ce_match_stat(struct cache_entry *ce, struct stat *st)
734aab75
LT
91{
92 unsigned int changed = 0;
93
8ae0a8c5
KS
94 switch (ntohl(ce->ce_mode) & S_IFMT) {
95 case S_IFREG:
96 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
97 /* We consider only the owner x bit to be relevant for
98 * "mode changes"
99 */
100 if (trust_executable_bit &&
101 (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
ffbe1add 102 changed |= MODE_CHANGED;
8ae0a8c5
KS
103 break;
104 case S_IFLNK:
105 changed |= !S_ISLNK(st->st_mode) ? TYPE_CHANGED : 0;
106 break;
107 default:
108 die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
109 }
ccc4feb5 110 if (ce->ce_mtime.sec != htonl(st->st_mtime))
734aab75 111 changed |= MTIME_CHANGED;
ccc4feb5
LT
112 if (ce->ce_ctime.sec != htonl(st->st_ctime))
113 changed |= CTIME_CHANGED;
114
2cb45e95 115#ifdef USE_NSEC
ccc4feb5
LT
116 /*
117 * nsec seems unreliable - not all filesystems support it, so
118 * as long as it is in the inode cache you get right nsec
119 * but after it gets flushed, you get zero nsec.
120 */
94dfb7f2 121 if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
ccc4feb5 122 changed |= MTIME_CHANGED;
94dfb7f2 123 if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
734aab75 124 changed |= CTIME_CHANGED;
ccc4feb5
LT
125#endif
126
127 if (ce->ce_uid != htonl(st->st_uid) ||
128 ce->ce_gid != htonl(st->st_gid))
734aab75 129 changed |= OWNER_CHANGED;
2cb45e95 130 if (ce->ce_ino != htonl(st->st_ino))
734aab75 131 changed |= INODE_CHANGED;
2cb45e95
LT
132
133#ifdef USE_STDEV
134 /*
135 * st_dev breaks on network filesystems where different
136 * clients will have different views of what "device"
137 * the filesystem is on
138 */
139 if (ce->ce_dev != htonl(st->st_dev))
140 changed |= INODE_CHANGED;
141#endif
142
ccc4feb5 143 if (ce->ce_size != htonl(st->st_size))
734aab75 144 changed |= DATA_CHANGED;
b0391890 145
29e4d363
JH
146 /*
147 * Within 1 second of this sequence:
148 * echo xyzzy >file && git-update-index --add file
149 * running this command:
150 * echo frotz >file
151 * would give a falsely clean cache entry. The mtime and
152 * length match the cache, and other stat fields do not change.
153 *
154 * We could detect this at update-index time (the cache entry
155 * being registered/updated records the same time as "now")
156 * and delay the return from git-update-index, but that would
157 * effectively mean we can make at most one commit per second,
158 * which is not acceptable. Instead, we check cache entries
159 * whose mtime are the same as the index file timestamp more
160 * careful than others.
161 */
162 if (!changed &&
163 index_file_timestamp &&
164 index_file_timestamp <= ntohl(ce->ce_mtime.sec))
165 changed |= ce_modified_check_fs(ce, st);
b0391890 166
29e4d363 167 return changed;
b0391890
JH
168}
169
170int ce_modified(struct cache_entry *ce, struct stat *st)
171{
29e4d363 172 int changed, changed_fs;
b0391890
JH
173 changed = ce_match_stat(ce, st);
174 if (!changed)
175 return 0;
b0391890
JH
176 /*
177 * If the mode or type has changed, there's no point in trying
178 * to refresh the entry - it's not going to match
179 */
180 if (changed & (MODE_CHANGED | TYPE_CHANGED))
181 return changed;
182
183 /* Immediately after read-tree or update-index --cacheinfo,
184 * the length field is zero. For other cases the ce_size
185 * should match the SHA1 recorded in the index entry.
186 */
187 if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
188 return changed;
189
29e4d363
JH
190 changed_fs = ce_modified_check_fs(ce, st);
191 if (changed_fs)
192 return changed | changed_fs;
b0391890
JH
193 return 0;
194}
195
958ba6c9
LT
196int base_name_compare(const char *name1, int len1, int mode1,
197 const char *name2, int len2, int mode2)
198{
199 unsigned char c1, c2;
200 int len = len1 < len2 ? len1 : len2;
201 int cmp;
202
203 cmp = memcmp(name1, name2, len);
204 if (cmp)
205 return cmp;
206 c1 = name1[len];
207 c2 = name2[len];
208 if (!c1 && S_ISDIR(mode1))
209 c1 = '/';
210 if (!c2 && S_ISDIR(mode2))
211 c2 = '/';
212 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
213}
214
95fd5bf8 215int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
eb38c22f 216{
95fd5bf8
LT
217 int len1 = flags1 & CE_NAMEMASK;
218 int len2 = flags2 & CE_NAMEMASK;
eb38c22f
LT
219 int len = len1 < len2 ? len1 : len2;
220 int cmp;
221
222 cmp = memcmp(name1, name2, len);
223 if (cmp)
224 return cmp;
225 if (len1 < len2)
226 return -1;
227 if (len1 > len2)
228 return 1;
95fd5bf8
LT
229 if (flags1 < flags2)
230 return -1;
231 if (flags1 > flags2)
232 return 1;
eb38c22f
LT
233 return 0;
234}
235
236int cache_name_pos(const char *name, int namelen)
237{
238 int first, last;
239
240 first = 0;
241 last = active_nr;
242 while (last > first) {
243 int next = (last + first) >> 1;
244 struct cache_entry *ce = active_cache[next];
972d1bb0 245 int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
eb38c22f 246 if (!cmp)
76e7f4ec 247 return next;
eb38c22f
LT
248 if (cmp < 0) {
249 last = next;
250 continue;
251 }
252 first = next+1;
253 }
76e7f4ec 254 return -first-1;
eb38c22f
LT
255}
256
7b937ca3 257/* Remove entry, return true if there are more entries to go.. */
dbbce55b 258int remove_cache_entry_at(int pos)
7b937ca3 259{
ee267527 260 active_cache_changed = 1;
7b937ca3
LT
261 active_nr--;
262 if (pos >= active_nr)
263 return 0;
264 memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
265 return 1;
266}
267
6b5ee137 268int remove_file_from_cache(const char *path)
197ee8c9
LT
269{
270 int pos = cache_name_pos(path, strlen(path));
c4e3cca1
JH
271 if (pos < 0)
272 pos = -pos-1;
273 while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
dbbce55b 274 remove_cache_entry_at(pos);
197ee8c9
LT
275 return 0;
276}
277
dbbce55b 278int ce_same_name(struct cache_entry *a, struct cache_entry *b)
7b937ca3
LT
279{
280 int len = ce_namelen(a);
281 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
282}
283
c0fd1f51
LT
284int ce_path_match(const struct cache_entry *ce, const char **pathspec)
285{
286 const char *match, *name;
287 int len;
288
289 if (!pathspec)
290 return 1;
291
292 len = ce_namelen(ce);
293 name = ce->name;
294 while ((match = *pathspec++) != NULL) {
295 int matchlen = strlen(match);
296 if (matchlen > len)
297 continue;
298 if (memcmp(name, match, matchlen))
299 continue;
300 if (matchlen && name[matchlen-1] == '/')
301 return 1;
302 if (name[matchlen] == '/' || !name[matchlen])
303 return 1;
f332726e
LT
304 if (!matchlen)
305 return 1;
c0fd1f51
LT
306 }
307 return 0;
308}
309
12676608
LT
310/*
311 * Do we have another file that has the beginning components being a
312 * proper superset of the name we're trying to add?
0f1e4f04 313 */
12676608 314static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 315{
12676608
LT
316 int retval = 0;
317 int len = ce_namelen(ce);
b155725d 318 int stage = ce_stage(ce);
12676608 319 const char *name = ce->name;
0f1e4f04 320
12676608
LT
321 while (pos < active_nr) {
322 struct cache_entry *p = active_cache[pos++];
0f1e4f04 323
12676608 324 if (len >= ce_namelen(p))
0f1e4f04 325 break;
12676608
LT
326 if (memcmp(name, p->name, len))
327 break;
b155725d
JH
328 if (ce_stage(p) != stage)
329 continue;
12676608
LT
330 if (p->name[len] != '/')
331 continue;
12676608
LT
332 retval = -1;
333 if (!ok_to_replace)
334 break;
335 remove_cache_entry_at(--pos);
0f1e4f04 336 }
12676608
LT
337 return retval;
338}
0f1e4f04 339
12676608
LT
340/*
341 * Do we have another file with a pathname that is a proper
342 * subset of the name we're trying to add?
343 */
344static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace)
345{
346 int retval = 0;
b155725d 347 int stage = ce_stage(ce);
12676608
LT
348 const char *name = ce->name;
349 const char *slash = name + ce_namelen(ce);
0f1e4f04 350
12676608
LT
351 for (;;) {
352 int len;
0f1e4f04 353
12676608
LT
354 for (;;) {
355 if (*--slash == '/')
356 break;
357 if (slash <= ce->name)
358 return retval;
359 }
360 len = slash - name;
0f1e4f04 361
b155725d 362 pos = cache_name_pos(name, ntohs(create_ce_flags(len, stage)));
12676608
LT
363 if (pos >= 0) {
364 retval = -1;
365 if (ok_to_replace)
366 break;
dbbce55b 367 remove_cache_entry_at(pos);
12676608
LT
368 continue;
369 }
370
371 /*
372 * Trivial optimization: if we find an entry that
373 * already matches the sub-directory, then we know
b155725d 374 * we're ok, and we can exit.
12676608
LT
375 */
376 pos = -pos-1;
b155725d 377 while (pos < active_nr) {
12676608 378 struct cache_entry *p = active_cache[pos];
b155725d
JH
379 if ((ce_namelen(p) <= len) ||
380 (p->name[len] != '/') ||
381 memcmp(p->name, name, len))
382 break; /* not our subdirectory */
383 if (ce_stage(p) == stage)
384 /* p is at the same stage as our entry, and
385 * is a subdirectory of what we are looking
386 * at, so we cannot have conflicts at our
387 * level or anything shorter.
388 */
389 return retval;
390 pos++;
192268c1 391 }
0f1e4f04 392 }
12676608
LT
393 return retval;
394}
395
396/* We may be in a situation where we already have path/file and path
397 * is being added, or we already have path and path/file is being
398 * added. Either one would result in a nonsense tree that has path
399 * twice when git-write-tree tries to write it out. Prevent it.
400 *
401 * If ok-to-replace is specified, we remove the conflicting entries
402 * from the cache so the caller should recompute the insert position.
403 * When this happens, we return non-zero.
404 */
405static int check_file_directory_conflict(const struct cache_entry *ce, int pos, int ok_to_replace)
406{
407 /*
408 * We check if the path is a sub-path of a subsequent pathname
409 * first, since removing those will not change the position
410 * in the array
411 */
412 int retval = has_file_name(ce, pos, ok_to_replace);
413 /*
414 * Then check if the path might have a clashing sub-directory
415 * before it.
416 */
417 return retval + has_dir_name(ce, pos, ok_to_replace);
0f1e4f04
JH
418}
419
192268c1 420int add_cache_entry(struct cache_entry *ce, int option)
197ee8c9
LT
421{
422 int pos;
192268c1
JH
423 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
424 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 425 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
972d1bb0 426 pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
197ee8c9 427
3e09cdfd 428 /* existing match? Just replace it. */
76e7f4ec 429 if (pos >= 0) {
ee267527 430 active_cache_changed = 1;
76e7f4ec 431 active_cache[pos] = ce;
197ee8c9
LT
432 return 0;
433 }
76e7f4ec 434 pos = -pos-1;
197ee8c9 435
7b937ca3
LT
436 /*
437 * Inserting a merged entry ("stage 0") into the index
438 * will always replace all non-merged entries..
439 */
440 if (pos < active_nr && ce_stage(ce) == 0) {
dbbce55b 441 while (ce_same_name(active_cache[pos], ce)) {
7b937ca3 442 ok_to_add = 1;
dbbce55b 443 if (!remove_cache_entry_at(pos))
7b937ca3
LT
444 break;
445 }
446 }
447
121481ab
LT
448 if (!ok_to_add)
449 return -1;
450
3e09cdfd
JH
451 if (!skip_df_check &&
452 check_file_directory_conflict(ce, pos, ok_to_replace)) {
192268c1
JH
453 if (!ok_to_replace)
454 return -1;
972d1bb0 455 pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
192268c1
JH
456 pos = -pos-1;
457 }
0f1e4f04 458
197ee8c9
LT
459 /* Make sure the array is big enough .. */
460 if (active_nr == active_alloc) {
461 active_alloc = alloc_nr(active_alloc);
812666c8 462 active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
197ee8c9
LT
463 }
464
465 /* Add it in.. */
466 active_nr++;
467 if (active_nr > pos)
468 memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
469 active_cache[pos] = ce;
ee267527 470 active_cache_changed = 1;
197ee8c9
LT
471 return 0;
472}
473
e83c5163
LT
474static int verify_hdr(struct cache_header *hdr, unsigned long size)
475{
476 SHA_CTX c;
477 unsigned char sha1[20];
478
ccc4feb5 479 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 480 return error("bad signature");
ca9be054
LT
481 if (hdr->hdr_version != htonl(2))
482 return error("bad index version");
e83c5163 483 SHA1_Init(&c);
ca9be054 484 SHA1_Update(&c, hdr, size - 20);
e83c5163 485 SHA1_Final(sha1, &c);
ca9be054
LT
486 if (memcmp(sha1, (void *)hdr + size - 20, 20))
487 return error("bad index file sha1 signature");
e83c5163
LT
488 return 0;
489}
490
491int read_cache(void)
492{
493 int fd, i;
494 struct stat st;
495 unsigned long size, offset;
496 void *map;
497 struct cache_header *hdr;
498
499 errno = EBUSY;
500 if (active_cache)
5d1a5c02
LT
501 return active_nr;
502
e83c5163 503 errno = ENOENT;
29e4d363 504 index_file_timestamp = 0;
bb233d69 505 fd = open(get_index_file(), O_RDONLY);
5d1a5c02
LT
506 if (fd < 0) {
507 if (errno == ENOENT)
508 return 0;
509 die("index file open failed (%s)", strerror(errno));
510 }
e83c5163 511
19b2860c 512 size = 0; // avoid gcc warning
e35f9824 513 map = MAP_FAILED;
e83c5163 514 if (!fstat(fd, &st)) {
e83c5163
LT
515 size = st.st_size;
516 errno = EINVAL;
ca9be054 517 if (size >= sizeof(struct cache_header) + 20)
520fc241 518 map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
e83c5163
LT
519 }
520 close(fd);
e35f9824 521 if (map == MAP_FAILED)
5d1a5c02 522 die("index file mmap failed (%s)", strerror(errno));
e83c5163
LT
523
524 hdr = map;
525 if (verify_hdr(hdr, size) < 0)
526 goto unmap;
527
ccc4feb5 528 active_nr = ntohl(hdr->hdr_entries);
e83c5163
LT
529 active_alloc = alloc_nr(active_nr);
530 active_cache = calloc(active_alloc, sizeof(struct cache_entry *));
531
532 offset = sizeof(*hdr);
ccc4feb5 533 for (i = 0; i < active_nr; i++) {
e83c5163
LT
534 struct cache_entry *ce = map + offset;
535 offset = offset + ce_size(ce);
536 active_cache[i] = ce;
537 }
29e4d363 538 index_file_timestamp = st.st_mtime;
e83c5163
LT
539 return active_nr;
540
541unmap:
542 munmap(map, size);
543 errno = EINVAL;
5d1a5c02 544 die("index file corrupt");
e83c5163
LT
545}
546
4990aadc 547#define WRITE_BUFFER_SIZE 8192
bf0f910d 548static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
549static unsigned long write_buffer_len;
550
ca9be054 551static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
552{
553 while (len) {
554 unsigned int buffered = write_buffer_len;
555 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
556 if (partial > len)
557 partial = len;
558 memcpy(write_buffer + buffered, data, partial);
559 buffered += partial;
560 if (buffered == WRITE_BUFFER_SIZE) {
ca9be054 561 SHA1_Update(context, write_buffer, WRITE_BUFFER_SIZE);
4990aadc
LT
562 if (write(fd, write_buffer, WRITE_BUFFER_SIZE) != WRITE_BUFFER_SIZE)
563 return -1;
564 buffered = 0;
565 }
566 write_buffer_len = buffered;
567 len -= partial;
568 data += partial;
569 }
570 return 0;
571}
572
ca9be054 573static int ce_flush(SHA_CTX *context, int fd)
4990aadc
LT
574{
575 unsigned int left = write_buffer_len;
ca9be054 576
4990aadc
LT
577 if (left) {
578 write_buffer_len = 0;
ca9be054 579 SHA1_Update(context, write_buffer, left);
4990aadc 580 }
ca9be054 581
2c865d9a
QH
582 /* Flush first if not enough space for SHA1 signature */
583 if (left + 20 > WRITE_BUFFER_SIZE) {
584 if (write(fd, write_buffer, left) != left)
585 return -1;
586 left = 0;
587 }
588
ca9be054
LT
589 /* Append the SHA1 signature at the end */
590 SHA1_Final(write_buffer + left, context);
591 left += 20;
592 if (write(fd, write_buffer, left) != left)
593 return -1;
4990aadc
LT
594 return 0;
595}
596
197ee8c9
LT
597int write_cache(int newfd, struct cache_entry **cache, int entries)
598{
599 SHA_CTX c;
600 struct cache_header hdr;
025a0709
JH
601 int i, removed;
602
603 for (i = removed = 0; i < entries; i++)
604 if (!cache[i]->ce_mode)
605 removed++;
197ee8c9 606
ccc4feb5 607 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
ca9be054 608 hdr.hdr_version = htonl(2);
025a0709 609 hdr.hdr_entries = htonl(entries - removed);
197ee8c9
LT
610
611 SHA1_Init(&c);
ca9be054 612 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
613 return -1;
614
615 for (i = 0; i < entries; i++) {
616 struct cache_entry *ce = cache[i];
aa16021e
LT
617 if (!ce->ce_mode)
618 continue;
ca9be054 619 if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
197ee8c9
LT
620 return -1;
621 }
ca9be054 622 return ce_flush(&c, newfd);
197ee8c9 623}