ie_match_stat(): do not ignore skip-worktree bit with CE_MATCH_IGNORE_VALID
[git/git.git] / read-cache.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "cache-tree.h"
9 #include "refs.h"
10 #include "dir.h"
11 #include "tree.h"
12 #include "commit.h"
13 #include "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16 #include "blob.h"
17
18 /* Index extensions.
19 *
20 * The first letter should be 'A'..'Z' for extensions that are not
21 * necessary for a correct operation (i.e. optimization data).
22 * When new extensions are added that _needs_ to be understood in
23 * order to correctly interpret the index file, pick character that
24 * is outside the range, to cause the reader to abort.
25 */
26
27 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
28 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
29
30 struct index_state the_index;
31
32 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
33 {
34 istate->cache[nr] = ce;
35 add_name_hash(istate, ce);
36 }
37
38 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
39 {
40 struct cache_entry *old = istate->cache[nr];
41
42 remove_name_hash(old);
43 set_index_entry(istate, nr, ce);
44 istate->cache_changed = 1;
45 }
46
47 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
48 {
49 struct cache_entry *old = istate->cache[nr], *new;
50 int namelen = strlen(new_name);
51
52 new = xmalloc(cache_entry_size(namelen));
53 copy_cache_entry(new, old);
54 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
55 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
56 memcpy(new->name, new_name, namelen + 1);
57
58 cache_tree_invalidate_path(istate->cache_tree, old->name);
59 remove_index_entry_at(istate, nr);
60 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
61 }
62
63 /*
64 * This only updates the "non-critical" parts of the directory
65 * cache, ie the parts that aren't tracked by GIT, and only used
66 * to validate the cache.
67 */
68 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
69 {
70 ce->ce_ctime.sec = (unsigned int)st->st_ctime;
71 ce->ce_mtime.sec = (unsigned int)st->st_mtime;
72 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
73 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
74 ce->ce_dev = st->st_dev;
75 ce->ce_ino = st->st_ino;
76 ce->ce_uid = st->st_uid;
77 ce->ce_gid = st->st_gid;
78 ce->ce_size = st->st_size;
79
80 if (assume_unchanged)
81 ce->ce_flags |= CE_VALID;
82
83 if (S_ISREG(st->st_mode))
84 ce_mark_uptodate(ce);
85 }
86
87 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
88 {
89 int match = -1;
90 int fd = open(ce->name, O_RDONLY);
91
92 if (fd >= 0) {
93 unsigned char sha1[20];
94 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
95 match = hashcmp(sha1, ce->sha1);
96 /* index_fd() closed the file descriptor already */
97 }
98 return match;
99 }
100
101 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
102 {
103 int match = -1;
104 void *buffer;
105 unsigned long size;
106 enum object_type type;
107 struct strbuf sb = STRBUF_INIT;
108
109 if (strbuf_readlink(&sb, ce->name, expected_size))
110 return -1;
111
112 buffer = read_sha1_file(ce->sha1, &type, &size);
113 if (buffer) {
114 if (size == sb.len)
115 match = memcmp(buffer, sb.buf, size);
116 free(buffer);
117 }
118 strbuf_release(&sb);
119 return match;
120 }
121
122 static int ce_compare_gitlink(struct cache_entry *ce)
123 {
124 unsigned char sha1[20];
125
126 /*
127 * We don't actually require that the .git directory
128 * under GITLINK directory be a valid git directory. It
129 * might even be missing (in case nobody populated that
130 * sub-project).
131 *
132 * If so, we consider it always to match.
133 */
134 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
135 return 0;
136 return hashcmp(sha1, ce->sha1);
137 }
138
139 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
140 {
141 switch (st->st_mode & S_IFMT) {
142 case S_IFREG:
143 if (ce_compare_data(ce, st))
144 return DATA_CHANGED;
145 break;
146 case S_IFLNK:
147 if (ce_compare_link(ce, xsize_t(st->st_size)))
148 return DATA_CHANGED;
149 break;
150 case S_IFDIR:
151 if (S_ISGITLINK(ce->ce_mode))
152 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
153 default:
154 return TYPE_CHANGED;
155 }
156 return 0;
157 }
158
159 int is_empty_blob_sha1(const unsigned char *sha1)
160 {
161 static const unsigned char empty_blob_sha1[20] = {
162 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
163 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
164 };
165
166 return !hashcmp(sha1, empty_blob_sha1);
167 }
168
169 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
170 {
171 unsigned int changed = 0;
172
173 if (ce->ce_flags & CE_REMOVE)
174 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
175
176 switch (ce->ce_mode & S_IFMT) {
177 case S_IFREG:
178 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
179 /* We consider only the owner x bit to be relevant for
180 * "mode changes"
181 */
182 if (trust_executable_bit &&
183 (0100 & (ce->ce_mode ^ st->st_mode)))
184 changed |= MODE_CHANGED;
185 break;
186 case S_IFLNK:
187 if (!S_ISLNK(st->st_mode) &&
188 (has_symlinks || !S_ISREG(st->st_mode)))
189 changed |= TYPE_CHANGED;
190 break;
191 case S_IFGITLINK:
192 /* We ignore most of the st_xxx fields for gitlinks */
193 if (!S_ISDIR(st->st_mode))
194 changed |= TYPE_CHANGED;
195 else if (ce_compare_gitlink(ce))
196 changed |= DATA_CHANGED;
197 return changed;
198 default:
199 die("internal error: ce_mode is %o", ce->ce_mode);
200 }
201 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
202 changed |= MTIME_CHANGED;
203 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
204 changed |= CTIME_CHANGED;
205
206 #ifdef USE_NSEC
207 if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
208 changed |= MTIME_CHANGED;
209 if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
210 changed |= CTIME_CHANGED;
211 #endif
212
213 if (ce->ce_uid != (unsigned int) st->st_uid ||
214 ce->ce_gid != (unsigned int) st->st_gid)
215 changed |= OWNER_CHANGED;
216 if (ce->ce_ino != (unsigned int) st->st_ino)
217 changed |= INODE_CHANGED;
218
219 #ifdef USE_STDEV
220 /*
221 * st_dev breaks on network filesystems where different
222 * clients will have different views of what "device"
223 * the filesystem is on
224 */
225 if (ce->ce_dev != (unsigned int) st->st_dev)
226 changed |= INODE_CHANGED;
227 #endif
228
229 if (ce->ce_size != (unsigned int) st->st_size)
230 changed |= DATA_CHANGED;
231
232 /* Racily smudged entry? */
233 if (!ce->ce_size) {
234 if (!is_empty_blob_sha1(ce->sha1))
235 changed |= DATA_CHANGED;
236 }
237
238 return changed;
239 }
240
241 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
242 {
243 return (!S_ISGITLINK(ce->ce_mode) &&
244 istate->timestamp.sec &&
245 #ifdef USE_NSEC
246 /* nanosecond timestamped files can also be racy! */
247 (istate->timestamp.sec < ce->ce_mtime.sec ||
248 (istate->timestamp.sec == ce->ce_mtime.sec &&
249 istate->timestamp.nsec <= ce->ce_mtime.nsec))
250 #else
251 istate->timestamp.sec <= ce->ce_mtime.sec
252 #endif
253 );
254 }
255
256 int ie_match_stat(const struct index_state *istate,
257 struct cache_entry *ce, struct stat *st,
258 unsigned int options)
259 {
260 unsigned int changed;
261 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
262 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
263 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
264
265 /*
266 * If it's marked as always valid in the index, it's
267 * valid whatever the checked-out copy says.
268 *
269 * skip-worktree has the same effect with higher precedence
270 */
271 if (!ignore_skip_worktree && ce_skip_worktree(ce))
272 return 0;
273 if (!ignore_valid && (ce->ce_flags & CE_VALID))
274 return 0;
275
276 /*
277 * Intent-to-add entries have not been added, so the index entry
278 * by definition never matches what is in the work tree until it
279 * actually gets added.
280 */
281 if (ce->ce_flags & CE_INTENT_TO_ADD)
282 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
283
284 changed = ce_match_stat_basic(ce, st);
285
286 /*
287 * Within 1 second of this sequence:
288 * echo xyzzy >file && git-update-index --add file
289 * running this command:
290 * echo frotz >file
291 * would give a falsely clean cache entry. The mtime and
292 * length match the cache, and other stat fields do not change.
293 *
294 * We could detect this at update-index time (the cache entry
295 * being registered/updated records the same time as "now")
296 * and delay the return from git-update-index, but that would
297 * effectively mean we can make at most one commit per second,
298 * which is not acceptable. Instead, we check cache entries
299 * whose mtime are the same as the index file timestamp more
300 * carefully than others.
301 */
302 if (!changed && is_racy_timestamp(istate, ce)) {
303 if (assume_racy_is_modified)
304 changed |= DATA_CHANGED;
305 else
306 changed |= ce_modified_check_fs(ce, st);
307 }
308
309 return changed;
310 }
311
312 int ie_modified(const struct index_state *istate,
313 struct cache_entry *ce, struct stat *st, unsigned int options)
314 {
315 int changed, changed_fs;
316
317 changed = ie_match_stat(istate, ce, st, options);
318 if (!changed)
319 return 0;
320 /*
321 * If the mode or type has changed, there's no point in trying
322 * to refresh the entry - it's not going to match
323 */
324 if (changed & (MODE_CHANGED | TYPE_CHANGED))
325 return changed;
326
327 /*
328 * Immediately after read-tree or update-index --cacheinfo,
329 * the length field is zero, as we have never even read the
330 * lstat(2) information once, and we cannot trust DATA_CHANGED
331 * returned by ie_match_stat() which in turn was returned by
332 * ce_match_stat_basic() to signal that the filesize of the
333 * blob changed. We have to actually go to the filesystem to
334 * see if the contents match, and if so, should answer "unchanged".
335 *
336 * The logic does not apply to gitlinks, as ce_match_stat_basic()
337 * already has checked the actual HEAD from the filesystem in the
338 * subproject. If ie_match_stat() already said it is different,
339 * then we know it is.
340 */
341 if ((changed & DATA_CHANGED) &&
342 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
343 return changed;
344
345 changed_fs = ce_modified_check_fs(ce, st);
346 if (changed_fs)
347 return changed | changed_fs;
348 return 0;
349 }
350
351 int base_name_compare(const char *name1, int len1, int mode1,
352 const char *name2, int len2, int mode2)
353 {
354 unsigned char c1, c2;
355 int len = len1 < len2 ? len1 : len2;
356 int cmp;
357
358 cmp = memcmp(name1, name2, len);
359 if (cmp)
360 return cmp;
361 c1 = name1[len];
362 c2 = name2[len];
363 if (!c1 && S_ISDIR(mode1))
364 c1 = '/';
365 if (!c2 && S_ISDIR(mode2))
366 c2 = '/';
367 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
368 }
369
370 /*
371 * df_name_compare() is identical to base_name_compare(), except it
372 * compares conflicting directory/file entries as equal. Note that
373 * while a directory name compares as equal to a regular file, they
374 * then individually compare _differently_ to a filename that has
375 * a dot after the basename (because '\0' < '.' < '/').
376 *
377 * This is used by routines that want to traverse the git namespace
378 * but then handle conflicting entries together when possible.
379 */
380 int df_name_compare(const char *name1, int len1, int mode1,
381 const char *name2, int len2, int mode2)
382 {
383 int len = len1 < len2 ? len1 : len2, cmp;
384 unsigned char c1, c2;
385
386 cmp = memcmp(name1, name2, len);
387 if (cmp)
388 return cmp;
389 /* Directories and files compare equal (same length, same name) */
390 if (len1 == len2)
391 return 0;
392 c1 = name1[len];
393 if (!c1 && S_ISDIR(mode1))
394 c1 = '/';
395 c2 = name2[len];
396 if (!c2 && S_ISDIR(mode2))
397 c2 = '/';
398 if (c1 == '/' && !c2)
399 return 0;
400 if (c2 == '/' && !c1)
401 return 0;
402 return c1 - c2;
403 }
404
405 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
406 {
407 int len1 = flags1 & CE_NAMEMASK;
408 int len2 = flags2 & CE_NAMEMASK;
409 int len = len1 < len2 ? len1 : len2;
410 int cmp;
411
412 cmp = memcmp(name1, name2, len);
413 if (cmp)
414 return cmp;
415 if (len1 < len2)
416 return -1;
417 if (len1 > len2)
418 return 1;
419
420 /* Compare stages */
421 flags1 &= CE_STAGEMASK;
422 flags2 &= CE_STAGEMASK;
423
424 if (flags1 < flags2)
425 return -1;
426 if (flags1 > flags2)
427 return 1;
428 return 0;
429 }
430
431 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
432 {
433 int first, last;
434
435 first = 0;
436 last = istate->cache_nr;
437 while (last > first) {
438 int next = (last + first) >> 1;
439 struct cache_entry *ce = istate->cache[next];
440 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
441 if (!cmp)
442 return next;
443 if (cmp < 0) {
444 last = next;
445 continue;
446 }
447 first = next+1;
448 }
449 return -first-1;
450 }
451
452 /* Remove entry, return true if there are more entries to go.. */
453 int remove_index_entry_at(struct index_state *istate, int pos)
454 {
455 struct cache_entry *ce = istate->cache[pos];
456
457 remove_name_hash(ce);
458 istate->cache_changed = 1;
459 istate->cache_nr--;
460 if (pos >= istate->cache_nr)
461 return 0;
462 memmove(istate->cache + pos,
463 istate->cache + pos + 1,
464 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
465 return 1;
466 }
467
468 /*
469 * Remove all cache ententries marked for removal, that is where
470 * CE_REMOVE is set in ce_flags. This is much more effective than
471 * calling remove_index_entry_at() for each entry to be removed.
472 */
473 void remove_marked_cache_entries(struct index_state *istate)
474 {
475 struct cache_entry **ce_array = istate->cache;
476 unsigned int i, j;
477
478 for (i = j = 0; i < istate->cache_nr; i++) {
479 if (ce_array[i]->ce_flags & CE_REMOVE)
480 remove_name_hash(ce_array[i]);
481 else
482 ce_array[j++] = ce_array[i];
483 }
484 istate->cache_changed = 1;
485 istate->cache_nr = j;
486 }
487
488 int remove_file_from_index(struct index_state *istate, const char *path)
489 {
490 int pos = index_name_pos(istate, path, strlen(path));
491 if (pos < 0)
492 pos = -pos-1;
493 cache_tree_invalidate_path(istate->cache_tree, path);
494 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
495 remove_index_entry_at(istate, pos);
496 return 0;
497 }
498
499 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
500 {
501 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
502 }
503
504 static int index_name_pos_also_unmerged(struct index_state *istate,
505 const char *path, int namelen)
506 {
507 int pos = index_name_pos(istate, path, namelen);
508 struct cache_entry *ce;
509
510 if (pos >= 0)
511 return pos;
512
513 /* maybe unmerged? */
514 pos = -1 - pos;
515 if (pos >= istate->cache_nr ||
516 compare_name((ce = istate->cache[pos]), path, namelen))
517 return -1;
518
519 /* order of preference: stage 2, 1, 3 */
520 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
521 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
522 !compare_name(ce, path, namelen))
523 pos++;
524 return pos;
525 }
526
527 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
528 {
529 int len = ce_namelen(ce);
530 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
531 }
532
533 /*
534 * If we add a filename that aliases in the cache, we will use the
535 * name that we already have - but we don't want to update the same
536 * alias twice, because that implies that there were actually two
537 * different files with aliasing names!
538 *
539 * So we use the CE_ADDED flag to verify that the alias was an old
540 * one before we accept it as
541 */
542 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
543 {
544 int len;
545 struct cache_entry *new;
546
547 if (alias->ce_flags & CE_ADDED)
548 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
549
550 /* Ok, create the new entry using the name of the existing alias */
551 len = ce_namelen(alias);
552 new = xcalloc(1, cache_entry_size(len));
553 memcpy(new->name, alias->name, len);
554 copy_cache_entry(new, ce);
555 free(ce);
556 return new;
557 }
558
559 static void record_intent_to_add(struct cache_entry *ce)
560 {
561 unsigned char sha1[20];
562 if (write_sha1_file("", 0, blob_type, sha1))
563 die("cannot create an empty blob in the object database");
564 hashcpy(ce->sha1, sha1);
565 }
566
567 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
568 {
569 int size, namelen, was_same;
570 mode_t st_mode = st->st_mode;
571 struct cache_entry *ce, *alias;
572 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
573 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
574 int pretend = flags & ADD_CACHE_PRETEND;
575 int intent_only = flags & ADD_CACHE_INTENT;
576 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
577 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
578
579 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
580 return error("%s: can only add regular files, symbolic links or git-directories", path);
581
582 namelen = strlen(path);
583 if (S_ISDIR(st_mode)) {
584 while (namelen && path[namelen-1] == '/')
585 namelen--;
586 }
587 size = cache_entry_size(namelen);
588 ce = xcalloc(1, size);
589 memcpy(ce->name, path, namelen);
590 ce->ce_flags = namelen;
591 if (!intent_only)
592 fill_stat_cache_info(ce, st);
593 else
594 ce->ce_flags |= CE_INTENT_TO_ADD;
595
596 if (trust_executable_bit && has_symlinks)
597 ce->ce_mode = create_ce_mode(st_mode);
598 else {
599 /* If there is an existing entry, pick the mode bits and type
600 * from it, otherwise assume unexecutable regular file.
601 */
602 struct cache_entry *ent;
603 int pos = index_name_pos_also_unmerged(istate, path, namelen);
604
605 ent = (0 <= pos) ? istate->cache[pos] : NULL;
606 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
607 }
608
609 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
610 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
611 /* Nothing changed, really */
612 free(ce);
613 ce_mark_uptodate(alias);
614 alias->ce_flags |= CE_ADDED;
615 return 0;
616 }
617 if (!intent_only) {
618 if (index_path(ce->sha1, path, st, 1))
619 return error("unable to index file %s", path);
620 } else
621 record_intent_to_add(ce);
622
623 if (ignore_case && alias && different_name(ce, alias))
624 ce = create_alias_ce(ce, alias);
625 ce->ce_flags |= CE_ADDED;
626
627 /* It was suspected to be racily clean, but it turns out to be Ok */
628 was_same = (alias &&
629 !ce_stage(alias) &&
630 !hashcmp(alias->sha1, ce->sha1) &&
631 ce->ce_mode == alias->ce_mode);
632
633 if (pretend)
634 ;
635 else if (add_index_entry(istate, ce, add_option))
636 return error("unable to add %s to index",path);
637 if (verbose && !was_same)
638 printf("add '%s'\n", path);
639 return 0;
640 }
641
642 int add_file_to_index(struct index_state *istate, const char *path, int flags)
643 {
644 struct stat st;
645 if (lstat(path, &st))
646 die_errno("unable to stat '%s'", path);
647 return add_to_index(istate, path, &st, flags);
648 }
649
650 struct cache_entry *make_cache_entry(unsigned int mode,
651 const unsigned char *sha1, const char *path, int stage,
652 int refresh)
653 {
654 int size, len;
655 struct cache_entry *ce;
656
657 if (!verify_path(path)) {
658 error("Invalid path '%s'", path);
659 return NULL;
660 }
661
662 len = strlen(path);
663 size = cache_entry_size(len);
664 ce = xcalloc(1, size);
665
666 hashcpy(ce->sha1, sha1);
667 memcpy(ce->name, path, len);
668 ce->ce_flags = create_ce_flags(len, stage);
669 ce->ce_mode = create_ce_mode(mode);
670
671 if (refresh)
672 return refresh_cache_entry(ce, 0);
673
674 return ce;
675 }
676
677 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
678 {
679 int len = ce_namelen(a);
680 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
681 }
682
683 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
684 {
685 const char *match, *name;
686 int len;
687
688 if (!pathspec)
689 return 1;
690
691 len = ce_namelen(ce);
692 name = ce->name;
693 while ((match = *pathspec++) != NULL) {
694 int matchlen = strlen(match);
695 if (matchlen > len)
696 continue;
697 if (memcmp(name, match, matchlen))
698 continue;
699 if (matchlen && name[matchlen-1] == '/')
700 return 1;
701 if (name[matchlen] == '/' || !name[matchlen])
702 return 1;
703 if (!matchlen)
704 return 1;
705 }
706 return 0;
707 }
708
709 /*
710 * We fundamentally don't like some paths: we don't want
711 * dot or dot-dot anywhere, and for obvious reasons don't
712 * want to recurse into ".git" either.
713 *
714 * Also, we don't want double slashes or slashes at the
715 * end that can make pathnames ambiguous.
716 */
717 static int verify_dotfile(const char *rest)
718 {
719 /*
720 * The first character was '.', but that
721 * has already been discarded, we now test
722 * the rest.
723 */
724 switch (*rest) {
725 /* "." is not allowed */
726 case '\0': case '/':
727 return 0;
728
729 /*
730 * ".git" followed by NUL or slash is bad. This
731 * shares the path end test with the ".." case.
732 */
733 case 'g':
734 if (rest[1] != 'i')
735 break;
736 if (rest[2] != 't')
737 break;
738 rest += 2;
739 /* fallthrough */
740 case '.':
741 if (rest[1] == '\0' || rest[1] == '/')
742 return 0;
743 }
744 return 1;
745 }
746
747 int verify_path(const char *path)
748 {
749 char c;
750
751 goto inside;
752 for (;;) {
753 if (!c)
754 return 1;
755 if (c == '/') {
756 inside:
757 c = *path++;
758 switch (c) {
759 default:
760 continue;
761 case '/': case '\0':
762 break;
763 case '.':
764 if (verify_dotfile(path))
765 continue;
766 }
767 return 0;
768 }
769 c = *path++;
770 }
771 }
772
773 /*
774 * Do we have another file that has the beginning components being a
775 * proper superset of the name we're trying to add?
776 */
777 static int has_file_name(struct index_state *istate,
778 const struct cache_entry *ce, int pos, int ok_to_replace)
779 {
780 int retval = 0;
781 int len = ce_namelen(ce);
782 int stage = ce_stage(ce);
783 const char *name = ce->name;
784
785 while (pos < istate->cache_nr) {
786 struct cache_entry *p = istate->cache[pos++];
787
788 if (len >= ce_namelen(p))
789 break;
790 if (memcmp(name, p->name, len))
791 break;
792 if (ce_stage(p) != stage)
793 continue;
794 if (p->name[len] != '/')
795 continue;
796 if (p->ce_flags & CE_REMOVE)
797 continue;
798 retval = -1;
799 if (!ok_to_replace)
800 break;
801 remove_index_entry_at(istate, --pos);
802 }
803 return retval;
804 }
805
806 /*
807 * Do we have another file with a pathname that is a proper
808 * subset of the name we're trying to add?
809 */
810 static int has_dir_name(struct index_state *istate,
811 const struct cache_entry *ce, int pos, int ok_to_replace)
812 {
813 int retval = 0;
814 int stage = ce_stage(ce);
815 const char *name = ce->name;
816 const char *slash = name + ce_namelen(ce);
817
818 for (;;) {
819 int len;
820
821 for (;;) {
822 if (*--slash == '/')
823 break;
824 if (slash <= ce->name)
825 return retval;
826 }
827 len = slash - name;
828
829 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
830 if (pos >= 0) {
831 /*
832 * Found one, but not so fast. This could
833 * be a marker that says "I was here, but
834 * I am being removed". Such an entry is
835 * not a part of the resulting tree, and
836 * it is Ok to have a directory at the same
837 * path.
838 */
839 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
840 retval = -1;
841 if (!ok_to_replace)
842 break;
843 remove_index_entry_at(istate, pos);
844 continue;
845 }
846 }
847 else
848 pos = -pos-1;
849
850 /*
851 * Trivial optimization: if we find an entry that
852 * already matches the sub-directory, then we know
853 * we're ok, and we can exit.
854 */
855 while (pos < istate->cache_nr) {
856 struct cache_entry *p = istate->cache[pos];
857 if ((ce_namelen(p) <= len) ||
858 (p->name[len] != '/') ||
859 memcmp(p->name, name, len))
860 break; /* not our subdirectory */
861 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
862 /*
863 * p is at the same stage as our entry, and
864 * is a subdirectory of what we are looking
865 * at, so we cannot have conflicts at our
866 * level or anything shorter.
867 */
868 return retval;
869 pos++;
870 }
871 }
872 return retval;
873 }
874
875 /* We may be in a situation where we already have path/file and path
876 * is being added, or we already have path and path/file is being
877 * added. Either one would result in a nonsense tree that has path
878 * twice when git-write-tree tries to write it out. Prevent it.
879 *
880 * If ok-to-replace is specified, we remove the conflicting entries
881 * from the cache so the caller should recompute the insert position.
882 * When this happens, we return non-zero.
883 */
884 static int check_file_directory_conflict(struct index_state *istate,
885 const struct cache_entry *ce,
886 int pos, int ok_to_replace)
887 {
888 int retval;
889
890 /*
891 * When ce is an "I am going away" entry, we allow it to be added
892 */
893 if (ce->ce_flags & CE_REMOVE)
894 return 0;
895
896 /*
897 * We check if the path is a sub-path of a subsequent pathname
898 * first, since removing those will not change the position
899 * in the array.
900 */
901 retval = has_file_name(istate, ce, pos, ok_to_replace);
902
903 /*
904 * Then check if the path might have a clashing sub-directory
905 * before it.
906 */
907 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
908 }
909
910 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
911 {
912 int pos;
913 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
914 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
915 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
916 int new_only = option & ADD_CACHE_NEW_ONLY;
917
918 cache_tree_invalidate_path(istate->cache_tree, ce->name);
919 pos = index_name_pos(istate, ce->name, ce->ce_flags);
920
921 /* existing match? Just replace it. */
922 if (pos >= 0) {
923 if (!new_only)
924 replace_index_entry(istate, pos, ce);
925 return 0;
926 }
927 pos = -pos-1;
928
929 /*
930 * Inserting a merged entry ("stage 0") into the index
931 * will always replace all non-merged entries..
932 */
933 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
934 while (ce_same_name(istate->cache[pos], ce)) {
935 ok_to_add = 1;
936 if (!remove_index_entry_at(istate, pos))
937 break;
938 }
939 }
940
941 if (!ok_to_add)
942 return -1;
943 if (!verify_path(ce->name))
944 return error("Invalid path '%s'", ce->name);
945
946 if (!skip_df_check &&
947 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
948 if (!ok_to_replace)
949 return error("'%s' appears as both a file and as a directory",
950 ce->name);
951 pos = index_name_pos(istate, ce->name, ce->ce_flags);
952 pos = -pos-1;
953 }
954 return pos + 1;
955 }
956
957 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
958 {
959 int pos;
960
961 if (option & ADD_CACHE_JUST_APPEND)
962 pos = istate->cache_nr;
963 else {
964 int ret;
965 ret = add_index_entry_with_check(istate, ce, option);
966 if (ret <= 0)
967 return ret;
968 pos = ret - 1;
969 }
970
971 /* Make sure the array is big enough .. */
972 if (istate->cache_nr == istate->cache_alloc) {
973 istate->cache_alloc = alloc_nr(istate->cache_alloc);
974 istate->cache = xrealloc(istate->cache,
975 istate->cache_alloc * sizeof(struct cache_entry *));
976 }
977
978 /* Add it in.. */
979 istate->cache_nr++;
980 if (istate->cache_nr > pos + 1)
981 memmove(istate->cache + pos + 1,
982 istate->cache + pos,
983 (istate->cache_nr - pos - 1) * sizeof(ce));
984 set_index_entry(istate, pos, ce);
985 istate->cache_changed = 1;
986 return 0;
987 }
988
989 /*
990 * "refresh" does not calculate a new sha1 file or bring the
991 * cache up-to-date for mode/content changes. But what it
992 * _does_ do is to "re-match" the stat information of a file
993 * with the cache, so that you can refresh the cache for a
994 * file that hasn't been changed but where the stat entry is
995 * out of date.
996 *
997 * For example, you'd want to do this after doing a "git-read-tree",
998 * to link up the stat cache details with the proper files.
999 */
1000 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1001 struct cache_entry *ce,
1002 unsigned int options, int *err)
1003 {
1004 struct stat st;
1005 struct cache_entry *updated;
1006 int changed, size;
1007 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1008 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1009
1010 if (ce_uptodate(ce))
1011 return ce;
1012
1013 /*
1014 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1015 * that the change to the work tree does not matter and told
1016 * us not to worry.
1017 */
1018 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1019 ce_mark_uptodate(ce);
1020 return ce;
1021 }
1022 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1023 ce_mark_uptodate(ce);
1024 return ce;
1025 }
1026
1027 if (lstat(ce->name, &st) < 0) {
1028 if (err)
1029 *err = errno;
1030 return NULL;
1031 }
1032
1033 changed = ie_match_stat(istate, ce, &st, options);
1034 if (!changed) {
1035 /*
1036 * The path is unchanged. If we were told to ignore
1037 * valid bit, then we did the actual stat check and
1038 * found that the entry is unmodified. If the entry
1039 * is not marked VALID, this is the place to mark it
1040 * valid again, under "assume unchanged" mode.
1041 */
1042 if (ignore_valid && assume_unchanged &&
1043 !(ce->ce_flags & CE_VALID))
1044 ; /* mark this one VALID again */
1045 else {
1046 /*
1047 * We do not mark the index itself "modified"
1048 * because CE_UPTODATE flag is in-core only;
1049 * we are not going to write this change out.
1050 */
1051 ce_mark_uptodate(ce);
1052 return ce;
1053 }
1054 }
1055
1056 if (ie_modified(istate, ce, &st, options)) {
1057 if (err)
1058 *err = EINVAL;
1059 return NULL;
1060 }
1061
1062 size = ce_size(ce);
1063 updated = xmalloc(size);
1064 memcpy(updated, ce, size);
1065 fill_stat_cache_info(updated, &st);
1066 /*
1067 * If ignore_valid is not set, we should leave CE_VALID bit
1068 * alone. Otherwise, paths marked with --no-assume-unchanged
1069 * (i.e. things to be edited) will reacquire CE_VALID bit
1070 * automatically, which is not really what we want.
1071 */
1072 if (!ignore_valid && assume_unchanged &&
1073 !(ce->ce_flags & CE_VALID))
1074 updated->ce_flags &= ~CE_VALID;
1075
1076 return updated;
1077 }
1078
1079 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
1080 {
1081 int i;
1082 int has_errors = 0;
1083 int really = (flags & REFRESH_REALLY) != 0;
1084 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1085 int quiet = (flags & REFRESH_QUIET) != 0;
1086 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1087 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1088 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1089 const char *needs_update_message;
1090
1091 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
1092 ? "locally modified" : "needs update");
1093 for (i = 0; i < istate->cache_nr; i++) {
1094 struct cache_entry *ce, *new;
1095 int cache_errno = 0;
1096
1097 ce = istate->cache[i];
1098 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1099 continue;
1100
1101 if (ce_stage(ce)) {
1102 while ((i < istate->cache_nr) &&
1103 ! strcmp(istate->cache[i]->name, ce->name))
1104 i++;
1105 i--;
1106 if (allow_unmerged)
1107 continue;
1108 printf("%s: needs merge\n", ce->name);
1109 has_errors = 1;
1110 continue;
1111 }
1112
1113 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1114 continue;
1115
1116 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1117 if (new == ce)
1118 continue;
1119 if (!new) {
1120 if (not_new && cache_errno == ENOENT)
1121 continue;
1122 if (really && cache_errno == EINVAL) {
1123 /* If we are doing --really-refresh that
1124 * means the index is not valid anymore.
1125 */
1126 ce->ce_flags &= ~CE_VALID;
1127 istate->cache_changed = 1;
1128 }
1129 if (quiet)
1130 continue;
1131 printf("%s: %s\n", ce->name, needs_update_message);
1132 has_errors = 1;
1133 continue;
1134 }
1135
1136 replace_index_entry(istate, i, new);
1137 }
1138 return has_errors;
1139 }
1140
1141 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1142 {
1143 return refresh_cache_ent(&the_index, ce, really, NULL);
1144 }
1145
1146 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1147 {
1148 git_SHA_CTX c;
1149 unsigned char sha1[20];
1150
1151 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1152 return error("bad signature");
1153 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1154 return error("bad index version");
1155 git_SHA1_Init(&c);
1156 git_SHA1_Update(&c, hdr, size - 20);
1157 git_SHA1_Final(sha1, &c);
1158 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1159 return error("bad index file sha1 signature");
1160 return 0;
1161 }
1162
1163 static int read_index_extension(struct index_state *istate,
1164 const char *ext, void *data, unsigned long sz)
1165 {
1166 switch (CACHE_EXT(ext)) {
1167 case CACHE_EXT_TREE:
1168 istate->cache_tree = cache_tree_read(data, sz);
1169 break;
1170 default:
1171 if (*ext < 'A' || 'Z' < *ext)
1172 return error("index uses %.4s extension, which we do not understand",
1173 ext);
1174 fprintf(stderr, "ignoring %.4s extension\n", ext);
1175 break;
1176 }
1177 return 0;
1178 }
1179
1180 int read_index(struct index_state *istate)
1181 {
1182 return read_index_from(istate, get_index_file());
1183 }
1184
1185 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1186 {
1187 size_t len;
1188 const char *name;
1189
1190 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1191 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1192 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1193 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1194 ce->ce_dev = ntohl(ondisk->dev);
1195 ce->ce_ino = ntohl(ondisk->ino);
1196 ce->ce_mode = ntohl(ondisk->mode);
1197 ce->ce_uid = ntohl(ondisk->uid);
1198 ce->ce_gid = ntohl(ondisk->gid);
1199 ce->ce_size = ntohl(ondisk->size);
1200 /* On-disk flags are just 16 bits */
1201 ce->ce_flags = ntohs(ondisk->flags);
1202
1203 hashcpy(ce->sha1, ondisk->sha1);
1204
1205 len = ce->ce_flags & CE_NAMEMASK;
1206
1207 if (ce->ce_flags & CE_EXTENDED) {
1208 struct ondisk_cache_entry_extended *ondisk2;
1209 int extended_flags;
1210 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1211 extended_flags = ntohs(ondisk2->flags2) << 16;
1212 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1213 if (extended_flags & ~CE_EXTENDED_FLAGS)
1214 die("Unknown index entry format %08x", extended_flags);
1215 ce->ce_flags |= extended_flags;
1216 name = ondisk2->name;
1217 }
1218 else
1219 name = ondisk->name;
1220
1221 if (len == CE_NAMEMASK)
1222 len = strlen(name);
1223 /*
1224 * NEEDSWORK: If the original index is crafted, this copy could
1225 * go unchecked.
1226 */
1227 memcpy(ce->name, name, len + 1);
1228 }
1229
1230 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1231 {
1232 long per_entry;
1233
1234 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1235
1236 /*
1237 * Alignment can cause differences. This should be "alignof", but
1238 * since that's a gcc'ism, just use the size of a pointer.
1239 */
1240 per_entry += sizeof(void *);
1241 return ondisk_size + entries*per_entry;
1242 }
1243
1244 /* remember to discard_cache() before reading a different cache! */
1245 int read_index_from(struct index_state *istate, const char *path)
1246 {
1247 int fd, i;
1248 struct stat st;
1249 unsigned long src_offset, dst_offset;
1250 struct cache_header *hdr;
1251 void *mmap;
1252 size_t mmap_size;
1253
1254 errno = EBUSY;
1255 if (istate->initialized)
1256 return istate->cache_nr;
1257
1258 errno = ENOENT;
1259 istate->timestamp.sec = 0;
1260 istate->timestamp.nsec = 0;
1261 fd = open(path, O_RDONLY);
1262 if (fd < 0) {
1263 if (errno == ENOENT)
1264 return 0;
1265 die_errno("index file open failed");
1266 }
1267
1268 if (fstat(fd, &st))
1269 die_errno("cannot stat the open index");
1270
1271 errno = EINVAL;
1272 mmap_size = xsize_t(st.st_size);
1273 if (mmap_size < sizeof(struct cache_header) + 20)
1274 die("index file smaller than expected");
1275
1276 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1277 close(fd);
1278 if (mmap == MAP_FAILED)
1279 die_errno("unable to map index file");
1280
1281 hdr = mmap;
1282 if (verify_hdr(hdr, mmap_size) < 0)
1283 goto unmap;
1284
1285 istate->cache_nr = ntohl(hdr->hdr_entries);
1286 istate->cache_alloc = alloc_nr(istate->cache_nr);
1287 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1288
1289 /*
1290 * The disk format is actually larger than the in-memory format,
1291 * due to space for nsec etc, so even though the in-memory one
1292 * has room for a few more flags, we can allocate using the same
1293 * index size
1294 */
1295 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1296 istate->initialized = 1;
1297
1298 src_offset = sizeof(*hdr);
1299 dst_offset = 0;
1300 for (i = 0; i < istate->cache_nr; i++) {
1301 struct ondisk_cache_entry *disk_ce;
1302 struct cache_entry *ce;
1303
1304 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1305 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1306 convert_from_disk(disk_ce, ce);
1307 set_index_entry(istate, i, ce);
1308
1309 src_offset += ondisk_ce_size(ce);
1310 dst_offset += ce_size(ce);
1311 }
1312 istate->timestamp.sec = st.st_mtime;
1313 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1314
1315 while (src_offset <= mmap_size - 20 - 8) {
1316 /* After an array of active_nr index entries,
1317 * there can be arbitrary number of extended
1318 * sections, each of which is prefixed with
1319 * extension name (4-byte) and section length
1320 * in 4-byte network byte order.
1321 */
1322 unsigned long extsize;
1323 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1324 extsize = ntohl(extsize);
1325 if (read_index_extension(istate,
1326 (const char *) mmap + src_offset,
1327 (char *) mmap + src_offset + 8,
1328 extsize) < 0)
1329 goto unmap;
1330 src_offset += 8;
1331 src_offset += extsize;
1332 }
1333 munmap(mmap, mmap_size);
1334 return istate->cache_nr;
1335
1336 unmap:
1337 munmap(mmap, mmap_size);
1338 errno = EINVAL;
1339 die("index file corrupt");
1340 }
1341
1342 int is_index_unborn(struct index_state *istate)
1343 {
1344 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
1345 }
1346
1347 int discard_index(struct index_state *istate)
1348 {
1349 istate->cache_nr = 0;
1350 istate->cache_changed = 0;
1351 istate->timestamp.sec = 0;
1352 istate->timestamp.nsec = 0;
1353 istate->name_hash_initialized = 0;
1354 free_hash(&istate->name_hash);
1355 cache_tree_free(&(istate->cache_tree));
1356 free(istate->alloc);
1357 istate->alloc = NULL;
1358 istate->initialized = 0;
1359
1360 /* no need to throw away allocated active_cache */
1361 return 0;
1362 }
1363
1364 int unmerged_index(const struct index_state *istate)
1365 {
1366 int i;
1367 for (i = 0; i < istate->cache_nr; i++) {
1368 if (ce_stage(istate->cache[i]))
1369 return 1;
1370 }
1371 return 0;
1372 }
1373
1374 #define WRITE_BUFFER_SIZE 8192
1375 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1376 static unsigned long write_buffer_len;
1377
1378 static int ce_write_flush(git_SHA_CTX *context, int fd)
1379 {
1380 unsigned int buffered = write_buffer_len;
1381 if (buffered) {
1382 git_SHA1_Update(context, write_buffer, buffered);
1383 if (write_in_full(fd, write_buffer, buffered) != buffered)
1384 return -1;
1385 write_buffer_len = 0;
1386 }
1387 return 0;
1388 }
1389
1390 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1391 {
1392 while (len) {
1393 unsigned int buffered = write_buffer_len;
1394 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1395 if (partial > len)
1396 partial = len;
1397 memcpy(write_buffer + buffered, data, partial);
1398 buffered += partial;
1399 if (buffered == WRITE_BUFFER_SIZE) {
1400 write_buffer_len = buffered;
1401 if (ce_write_flush(context, fd))
1402 return -1;
1403 buffered = 0;
1404 }
1405 write_buffer_len = buffered;
1406 len -= partial;
1407 data = (char *) data + partial;
1408 }
1409 return 0;
1410 }
1411
1412 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1413 unsigned int ext, unsigned int sz)
1414 {
1415 ext = htonl(ext);
1416 sz = htonl(sz);
1417 return ((ce_write(context, fd, &ext, 4) < 0) ||
1418 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1419 }
1420
1421 static int ce_flush(git_SHA_CTX *context, int fd)
1422 {
1423 unsigned int left = write_buffer_len;
1424
1425 if (left) {
1426 write_buffer_len = 0;
1427 git_SHA1_Update(context, write_buffer, left);
1428 }
1429
1430 /* Flush first if not enough space for SHA1 signature */
1431 if (left + 20 > WRITE_BUFFER_SIZE) {
1432 if (write_in_full(fd, write_buffer, left) != left)
1433 return -1;
1434 left = 0;
1435 }
1436
1437 /* Append the SHA1 signature at the end */
1438 git_SHA1_Final(write_buffer + left, context);
1439 left += 20;
1440 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1441 }
1442
1443 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1444 {
1445 /*
1446 * The only thing we care about in this function is to smudge the
1447 * falsely clean entry due to touch-update-touch race, so we leave
1448 * everything else as they are. We are called for entries whose
1449 * ce_mtime match the index file mtime.
1450 *
1451 * Note that this actually does not do much for gitlinks, for
1452 * which ce_match_stat_basic() always goes to the actual
1453 * contents. The caller checks with is_racy_timestamp() which
1454 * always says "no" for gitlinks, so we are not called for them ;-)
1455 */
1456 struct stat st;
1457
1458 if (lstat(ce->name, &st) < 0)
1459 return;
1460 if (ce_match_stat_basic(ce, &st))
1461 return;
1462 if (ce_modified_check_fs(ce, &st)) {
1463 /* This is "racily clean"; smudge it. Note that this
1464 * is a tricky code. At first glance, it may appear
1465 * that it can break with this sequence:
1466 *
1467 * $ echo xyzzy >frotz
1468 * $ git-update-index --add frotz
1469 * $ : >frotz
1470 * $ sleep 3
1471 * $ echo filfre >nitfol
1472 * $ git-update-index --add nitfol
1473 *
1474 * but it does not. When the second update-index runs,
1475 * it notices that the entry "frotz" has the same timestamp
1476 * as index, and if we were to smudge it by resetting its
1477 * size to zero here, then the object name recorded
1478 * in index is the 6-byte file but the cached stat information
1479 * becomes zero --- which would then match what we would
1480 * obtain from the filesystem next time we stat("frotz").
1481 *
1482 * However, the second update-index, before calling
1483 * this function, notices that the cached size is 6
1484 * bytes and what is on the filesystem is an empty
1485 * file, and never calls us, so the cached size information
1486 * for "frotz" stays 6 which does not match the filesystem.
1487 */
1488 ce->ce_size = 0;
1489 }
1490 }
1491
1492 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1493 {
1494 int size = ondisk_ce_size(ce);
1495 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1496 char *name;
1497
1498 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1499 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1500 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1501 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1502 ondisk->dev = htonl(ce->ce_dev);
1503 ondisk->ino = htonl(ce->ce_ino);
1504 ondisk->mode = htonl(ce->ce_mode);
1505 ondisk->uid = htonl(ce->ce_uid);
1506 ondisk->gid = htonl(ce->ce_gid);
1507 ondisk->size = htonl(ce->ce_size);
1508 hashcpy(ondisk->sha1, ce->sha1);
1509 ondisk->flags = htons(ce->ce_flags);
1510 if (ce->ce_flags & CE_EXTENDED) {
1511 struct ondisk_cache_entry_extended *ondisk2;
1512 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1513 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1514 name = ondisk2->name;
1515 }
1516 else
1517 name = ondisk->name;
1518 memcpy(name, ce->name, ce_namelen(ce));
1519
1520 return ce_write(c, fd, ondisk, size);
1521 }
1522
1523 int write_index(struct index_state *istate, int newfd)
1524 {
1525 git_SHA_CTX c;
1526 struct cache_header hdr;
1527 int i, err, removed, extended;
1528 struct cache_entry **cache = istate->cache;
1529 int entries = istate->cache_nr;
1530 struct stat st;
1531
1532 for (i = removed = extended = 0; i < entries; i++) {
1533 if (cache[i]->ce_flags & CE_REMOVE)
1534 removed++;
1535
1536 /* reduce extended entries if possible */
1537 cache[i]->ce_flags &= ~CE_EXTENDED;
1538 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1539 extended++;
1540 cache[i]->ce_flags |= CE_EXTENDED;
1541 }
1542 }
1543
1544 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1545 /* for extended format, increase version so older git won't try to read it */
1546 hdr.hdr_version = htonl(extended ? 3 : 2);
1547 hdr.hdr_entries = htonl(entries - removed);
1548
1549 git_SHA1_Init(&c);
1550 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1551 return -1;
1552
1553 for (i = 0; i < entries; i++) {
1554 struct cache_entry *ce = cache[i];
1555 if (ce->ce_flags & CE_REMOVE)
1556 continue;
1557 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1558 ce_smudge_racily_clean_entry(ce);
1559 if (ce_write_entry(&c, newfd, ce) < 0)
1560 return -1;
1561 }
1562
1563 /* Write extension data here */
1564 if (istate->cache_tree) {
1565 struct strbuf sb = STRBUF_INIT;
1566
1567 cache_tree_write(&sb, istate->cache_tree);
1568 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1569 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1570 strbuf_release(&sb);
1571 if (err)
1572 return -1;
1573 }
1574
1575 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1576 return -1;
1577 istate->timestamp.sec = (unsigned int)st.st_mtime;
1578 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1579 return 0;
1580 }
1581
1582 /*
1583 * Read the index file that is potentially unmerged into given
1584 * index_state, dropping any unmerged entries. Returns true if
1585 * the index is unmerged. Callers who want to refuse to work
1586 * from an unmerged state can call this and check its return value,
1587 * instead of calling read_cache().
1588 */
1589 int read_index_unmerged(struct index_state *istate)
1590 {
1591 int i;
1592 int unmerged = 0;
1593
1594 read_index(istate);
1595 for (i = 0; i < istate->cache_nr; i++) {
1596 struct cache_entry *ce = istate->cache[i];
1597 struct cache_entry *new_ce;
1598 int size, len;
1599
1600 if (!ce_stage(ce))
1601 continue;
1602 unmerged = 1;
1603 len = strlen(ce->name);
1604 size = cache_entry_size(len);
1605 new_ce = xcalloc(1, size);
1606 hashcpy(new_ce->sha1, ce->sha1);
1607 memcpy(new_ce->name, ce->name, len);
1608 new_ce->ce_flags = create_ce_flags(len, 0);
1609 new_ce->ce_mode = ce->ce_mode;
1610 if (add_index_entry(istate, new_ce, 0))
1611 return error("%s: cannot drop to stage #0",
1612 ce->name);
1613 i = index_name_pos(istate, new_ce->name, len);
1614 }
1615 return unmerged;
1616 }
1617
1618 struct update_callback_data
1619 {
1620 int flags;
1621 int add_errors;
1622 };
1623
1624 static void update_callback(struct diff_queue_struct *q,
1625 struct diff_options *opt, void *cbdata)
1626 {
1627 int i;
1628 struct update_callback_data *data = cbdata;
1629
1630 for (i = 0; i < q->nr; i++) {
1631 struct diff_filepair *p = q->queue[i];
1632 const char *path = p->one->path;
1633 switch (p->status) {
1634 default:
1635 die("unexpected diff status %c", p->status);
1636 case DIFF_STATUS_UNMERGED:
1637 /*
1638 * ADD_CACHE_IGNORE_REMOVAL is unset if "git
1639 * add -u" is calling us, In such a case, a
1640 * missing work tree file needs to be removed
1641 * if there is an unmerged entry at stage #2,
1642 * but such a diff record is followed by
1643 * another with DIFF_STATUS_DELETED (and if
1644 * there is no stage #2, we won't see DELETED
1645 * nor MODIFIED). We can simply continue
1646 * either way.
1647 */
1648 if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL))
1649 continue;
1650 /*
1651 * Otherwise, it is "git add path" is asking
1652 * to explicitly add it; we fall through. A
1653 * missing work tree file is an error and is
1654 * caught by add_file_to_index() in such a
1655 * case.
1656 */
1657 case DIFF_STATUS_MODIFIED:
1658 case DIFF_STATUS_TYPE_CHANGED:
1659 if (add_file_to_index(&the_index, path, data->flags)) {
1660 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1661 die("updating files failed");
1662 data->add_errors++;
1663 }
1664 break;
1665 case DIFF_STATUS_DELETED:
1666 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1667 break;
1668 if (!(data->flags & ADD_CACHE_PRETEND))
1669 remove_file_from_index(&the_index, path);
1670 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1671 printf("remove '%s'\n", path);
1672 break;
1673 }
1674 }
1675 }
1676
1677 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1678 {
1679 struct update_callback_data data;
1680 struct rev_info rev;
1681 init_revisions(&rev, prefix);
1682 setup_revisions(0, NULL, &rev, NULL);
1683 rev.prune_data = pathspec;
1684 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1685 rev.diffopt.format_callback = update_callback;
1686 data.flags = flags;
1687 data.add_errors = 0;
1688 rev.diffopt.format_callback_data = &data;
1689 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1690 return !!data.add_errors;
1691 }
1692
1693 /*
1694 * Returns 1 if the path is an "other" path with respect to
1695 * the index; that is, the path is not mentioned in the index at all,
1696 * either as a file, a directory with some files in the index,
1697 * or as an unmerged entry.
1698 *
1699 * We helpfully remove a trailing "/" from directories so that
1700 * the output of read_directory can be used as-is.
1701 */
1702 int index_name_is_other(const struct index_state *istate, const char *name,
1703 int namelen)
1704 {
1705 int pos;
1706 if (namelen && name[namelen - 1] == '/')
1707 namelen--;
1708 pos = index_name_pos(istate, name, namelen);
1709 if (0 <= pos)
1710 return 0; /* exact match */
1711 pos = -pos - 1;
1712 if (pos < istate->cache_nr) {
1713 struct cache_entry *ce = istate->cache[pos];
1714 if (ce_namelen(ce) == namelen &&
1715 !memcmp(ce->name, name, namelen))
1716 return 0; /* Yup, this one exists unmerged */
1717 }
1718 return 1;
1719 }