resolve-undo: record resolved conflicts in a new index extension section
[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 #include "resolve-undo.h"
18
19 /* Index extensions.
20 *
21 * The first letter should be 'A'..'Z' for extensions that are not
22 * necessary for a correct operation (i.e. optimization data).
23 * When new extensions are added that _needs_ to be understood in
24 * order to correctly interpret the index file, pick character that
25 * is outside the range, to cause the reader to abort.
26 */
27
28 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
29 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
30 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUN" */
31
32 struct index_state the_index;
33
34 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
35 {
36 istate->cache[nr] = ce;
37 add_name_hash(istate, ce);
38 }
39
40 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
41 {
42 struct cache_entry *old = istate->cache[nr];
43
44 remove_name_hash(old);
45 set_index_entry(istate, nr, ce);
46 istate->cache_changed = 1;
47 }
48
49 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
50 {
51 struct cache_entry *old = istate->cache[nr], *new;
52 int namelen = strlen(new_name);
53
54 new = xmalloc(cache_entry_size(namelen));
55 copy_cache_entry(new, old);
56 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
57 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
58 memcpy(new->name, new_name, namelen + 1);
59
60 cache_tree_invalidate_path(istate->cache_tree, old->name);
61 remove_index_entry_at(istate, nr);
62 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
63 }
64
65 /*
66 * This only updates the "non-critical" parts of the directory
67 * cache, ie the parts that aren't tracked by GIT, and only used
68 * to validate the cache.
69 */
70 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
71 {
72 ce->ce_ctime.sec = (unsigned int)st->st_ctime;
73 ce->ce_mtime.sec = (unsigned int)st->st_mtime;
74 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
75 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
76 ce->ce_dev = st->st_dev;
77 ce->ce_ino = st->st_ino;
78 ce->ce_uid = st->st_uid;
79 ce->ce_gid = st->st_gid;
80 ce->ce_size = st->st_size;
81
82 if (assume_unchanged)
83 ce->ce_flags |= CE_VALID;
84
85 if (S_ISREG(st->st_mode))
86 ce_mark_uptodate(ce);
87 }
88
89 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
90 {
91 int match = -1;
92 int fd = open(ce->name, O_RDONLY);
93
94 if (fd >= 0) {
95 unsigned char sha1[20];
96 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
97 match = hashcmp(sha1, ce->sha1);
98 /* index_fd() closed the file descriptor already */
99 }
100 return match;
101 }
102
103 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
104 {
105 int match = -1;
106 void *buffer;
107 unsigned long size;
108 enum object_type type;
109 struct strbuf sb = STRBUF_INIT;
110
111 if (strbuf_readlink(&sb, ce->name, expected_size))
112 return -1;
113
114 buffer = read_sha1_file(ce->sha1, &type, &size);
115 if (buffer) {
116 if (size == sb.len)
117 match = memcmp(buffer, sb.buf, size);
118 free(buffer);
119 }
120 strbuf_release(&sb);
121 return match;
122 }
123
124 static int ce_compare_gitlink(struct cache_entry *ce)
125 {
126 unsigned char sha1[20];
127
128 /*
129 * We don't actually require that the .git directory
130 * under GITLINK directory be a valid git directory. It
131 * might even be missing (in case nobody populated that
132 * sub-project).
133 *
134 * If so, we consider it always to match.
135 */
136 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
137 return 0;
138 return hashcmp(sha1, ce->sha1);
139 }
140
141 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
142 {
143 switch (st->st_mode & S_IFMT) {
144 case S_IFREG:
145 if (ce_compare_data(ce, st))
146 return DATA_CHANGED;
147 break;
148 case S_IFLNK:
149 if (ce_compare_link(ce, xsize_t(st->st_size)))
150 return DATA_CHANGED;
151 break;
152 case S_IFDIR:
153 if (S_ISGITLINK(ce->ce_mode))
154 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
155 default:
156 return TYPE_CHANGED;
157 }
158 return 0;
159 }
160
161 int is_empty_blob_sha1(const unsigned char *sha1)
162 {
163 static const unsigned char empty_blob_sha1[20] = {
164 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
165 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
166 };
167
168 return !hashcmp(sha1, empty_blob_sha1);
169 }
170
171 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
172 {
173 unsigned int changed = 0;
174
175 if (ce->ce_flags & CE_REMOVE)
176 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
177
178 switch (ce->ce_mode & S_IFMT) {
179 case S_IFREG:
180 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
181 /* We consider only the owner x bit to be relevant for
182 * "mode changes"
183 */
184 if (trust_executable_bit &&
185 (0100 & (ce->ce_mode ^ st->st_mode)))
186 changed |= MODE_CHANGED;
187 break;
188 case S_IFLNK:
189 if (!S_ISLNK(st->st_mode) &&
190 (has_symlinks || !S_ISREG(st->st_mode)))
191 changed |= TYPE_CHANGED;
192 break;
193 case S_IFGITLINK:
194 /* We ignore most of the st_xxx fields for gitlinks */
195 if (!S_ISDIR(st->st_mode))
196 changed |= TYPE_CHANGED;
197 else if (ce_compare_gitlink(ce))
198 changed |= DATA_CHANGED;
199 return changed;
200 default:
201 die("internal error: ce_mode is %o", ce->ce_mode);
202 }
203 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
204 changed |= MTIME_CHANGED;
205 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
206 changed |= CTIME_CHANGED;
207
208 #ifdef USE_NSEC
209 if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
210 changed |= MTIME_CHANGED;
211 if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
212 changed |= CTIME_CHANGED;
213 #endif
214
215 if (ce->ce_uid != (unsigned int) st->st_uid ||
216 ce->ce_gid != (unsigned int) st->st_gid)
217 changed |= OWNER_CHANGED;
218 if (ce->ce_ino != (unsigned int) st->st_ino)
219 changed |= INODE_CHANGED;
220
221 #ifdef USE_STDEV
222 /*
223 * st_dev breaks on network filesystems where different
224 * clients will have different views of what "device"
225 * the filesystem is on
226 */
227 if (ce->ce_dev != (unsigned int) st->st_dev)
228 changed |= INODE_CHANGED;
229 #endif
230
231 if (ce->ce_size != (unsigned int) st->st_size)
232 changed |= DATA_CHANGED;
233
234 /* Racily smudged entry? */
235 if (!ce->ce_size) {
236 if (!is_empty_blob_sha1(ce->sha1))
237 changed |= DATA_CHANGED;
238 }
239
240 return changed;
241 }
242
243 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
244 {
245 return (!S_ISGITLINK(ce->ce_mode) &&
246 istate->timestamp.sec &&
247 #ifdef USE_NSEC
248 /* nanosecond timestamped files can also be racy! */
249 (istate->timestamp.sec < ce->ce_mtime.sec ||
250 (istate->timestamp.sec == ce->ce_mtime.sec &&
251 istate->timestamp.nsec <= ce->ce_mtime.nsec))
252 #else
253 istate->timestamp.sec <= ce->ce_mtime.sec
254 #endif
255 );
256 }
257
258 int ie_match_stat(const struct index_state *istate,
259 struct cache_entry *ce, struct stat *st,
260 unsigned int options)
261 {
262 unsigned int changed;
263 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
264 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
265
266 /*
267 * If it's marked as always valid in the index, it's
268 * valid whatever the checked-out copy says.
269 */
270 if (!ignore_valid && (ce->ce_flags & CE_VALID))
271 return 0;
272
273 /*
274 * Intent-to-add entries have not been added, so the index entry
275 * by definition never matches what is in the work tree until it
276 * actually gets added.
277 */
278 if (ce->ce_flags & CE_INTENT_TO_ADD)
279 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
280
281 changed = ce_match_stat_basic(ce, st);
282
283 /*
284 * Within 1 second of this sequence:
285 * echo xyzzy >file && git-update-index --add file
286 * running this command:
287 * echo frotz >file
288 * would give a falsely clean cache entry. The mtime and
289 * length match the cache, and other stat fields do not change.
290 *
291 * We could detect this at update-index time (the cache entry
292 * being registered/updated records the same time as "now")
293 * and delay the return from git-update-index, but that would
294 * effectively mean we can make at most one commit per second,
295 * which is not acceptable. Instead, we check cache entries
296 * whose mtime are the same as the index file timestamp more
297 * carefully than others.
298 */
299 if (!changed && is_racy_timestamp(istate, ce)) {
300 if (assume_racy_is_modified)
301 changed |= DATA_CHANGED;
302 else
303 changed |= ce_modified_check_fs(ce, st);
304 }
305
306 return changed;
307 }
308
309 int ie_modified(const struct index_state *istate,
310 struct cache_entry *ce, struct stat *st, unsigned int options)
311 {
312 int changed, changed_fs;
313
314 changed = ie_match_stat(istate, ce, st, options);
315 if (!changed)
316 return 0;
317 /*
318 * If the mode or type has changed, there's no point in trying
319 * to refresh the entry - it's not going to match
320 */
321 if (changed & (MODE_CHANGED | TYPE_CHANGED))
322 return changed;
323
324 /*
325 * Immediately after read-tree or update-index --cacheinfo,
326 * the length field is zero, as we have never even read the
327 * lstat(2) information once, and we cannot trust DATA_CHANGED
328 * returned by ie_match_stat() which in turn was returned by
329 * ce_match_stat_basic() to signal that the filesize of the
330 * blob changed. We have to actually go to the filesystem to
331 * see if the contents match, and if so, should answer "unchanged".
332 *
333 * The logic does not apply to gitlinks, as ce_match_stat_basic()
334 * already has checked the actual HEAD from the filesystem in the
335 * subproject. If ie_match_stat() already said it is different,
336 * then we know it is.
337 */
338 if ((changed & DATA_CHANGED) &&
339 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
340 return changed;
341
342 changed_fs = ce_modified_check_fs(ce, st);
343 if (changed_fs)
344 return changed | changed_fs;
345 return 0;
346 }
347
348 int base_name_compare(const char *name1, int len1, int mode1,
349 const char *name2, int len2, int mode2)
350 {
351 unsigned char c1, c2;
352 int len = len1 < len2 ? len1 : len2;
353 int cmp;
354
355 cmp = memcmp(name1, name2, len);
356 if (cmp)
357 return cmp;
358 c1 = name1[len];
359 c2 = name2[len];
360 if (!c1 && S_ISDIR(mode1))
361 c1 = '/';
362 if (!c2 && S_ISDIR(mode2))
363 c2 = '/';
364 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
365 }
366
367 /*
368 * df_name_compare() is identical to base_name_compare(), except it
369 * compares conflicting directory/file entries as equal. Note that
370 * while a directory name compares as equal to a regular file, they
371 * then individually compare _differently_ to a filename that has
372 * a dot after the basename (because '\0' < '.' < '/').
373 *
374 * This is used by routines that want to traverse the git namespace
375 * but then handle conflicting entries together when possible.
376 */
377 int df_name_compare(const char *name1, int len1, int mode1,
378 const char *name2, int len2, int mode2)
379 {
380 int len = len1 < len2 ? len1 : len2, cmp;
381 unsigned char c1, c2;
382
383 cmp = memcmp(name1, name2, len);
384 if (cmp)
385 return cmp;
386 /* Directories and files compare equal (same length, same name) */
387 if (len1 == len2)
388 return 0;
389 c1 = name1[len];
390 if (!c1 && S_ISDIR(mode1))
391 c1 = '/';
392 c2 = name2[len];
393 if (!c2 && S_ISDIR(mode2))
394 c2 = '/';
395 if (c1 == '/' && !c2)
396 return 0;
397 if (c2 == '/' && !c1)
398 return 0;
399 return c1 - c2;
400 }
401
402 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
403 {
404 int len1 = flags1 & CE_NAMEMASK;
405 int len2 = flags2 & CE_NAMEMASK;
406 int len = len1 < len2 ? len1 : len2;
407 int cmp;
408
409 cmp = memcmp(name1, name2, len);
410 if (cmp)
411 return cmp;
412 if (len1 < len2)
413 return -1;
414 if (len1 > len2)
415 return 1;
416
417 /* Compare stages */
418 flags1 &= CE_STAGEMASK;
419 flags2 &= CE_STAGEMASK;
420
421 if (flags1 < flags2)
422 return -1;
423 if (flags1 > flags2)
424 return 1;
425 return 0;
426 }
427
428 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
429 {
430 int first, last;
431
432 first = 0;
433 last = istate->cache_nr;
434 while (last > first) {
435 int next = (last + first) >> 1;
436 struct cache_entry *ce = istate->cache[next];
437 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
438 if (!cmp)
439 return next;
440 if (cmp < 0) {
441 last = next;
442 continue;
443 }
444 first = next+1;
445 }
446 return -first-1;
447 }
448
449 /* Remove entry, return true if there are more entries to go.. */
450 int remove_index_entry_at(struct index_state *istate, int pos)
451 {
452 struct cache_entry *ce = istate->cache[pos];
453
454 record_resolve_undo(istate, ce);
455 remove_name_hash(ce);
456 istate->cache_changed = 1;
457 istate->cache_nr--;
458 if (pos >= istate->cache_nr)
459 return 0;
460 memmove(istate->cache + pos,
461 istate->cache + pos + 1,
462 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
463 return 1;
464 }
465
466 /*
467 * Remove all cache ententries marked for removal, that is where
468 * CE_REMOVE is set in ce_flags. This is much more effective than
469 * calling remove_index_entry_at() for each entry to be removed.
470 */
471 void remove_marked_cache_entries(struct index_state *istate)
472 {
473 struct cache_entry **ce_array = istate->cache;
474 unsigned int i, j;
475
476 for (i = j = 0; i < istate->cache_nr; i++) {
477 if (ce_array[i]->ce_flags & CE_REMOVE)
478 remove_name_hash(ce_array[i]);
479 else
480 ce_array[j++] = ce_array[i];
481 }
482 istate->cache_changed = 1;
483 istate->cache_nr = j;
484 }
485
486 int remove_file_from_index(struct index_state *istate, const char *path)
487 {
488 int pos = index_name_pos(istate, path, strlen(path));
489 if (pos < 0)
490 pos = -pos-1;
491 cache_tree_invalidate_path(istate->cache_tree, path);
492 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
493 remove_index_entry_at(istate, pos);
494 return 0;
495 }
496
497 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
498 {
499 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
500 }
501
502 static int index_name_pos_also_unmerged(struct index_state *istate,
503 const char *path, int namelen)
504 {
505 int pos = index_name_pos(istate, path, namelen);
506 struct cache_entry *ce;
507
508 if (pos >= 0)
509 return pos;
510
511 /* maybe unmerged? */
512 pos = -1 - pos;
513 if (pos >= istate->cache_nr ||
514 compare_name((ce = istate->cache[pos]), path, namelen))
515 return -1;
516
517 /* order of preference: stage 2, 1, 3 */
518 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
519 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
520 !compare_name(ce, path, namelen))
521 pos++;
522 return pos;
523 }
524
525 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
526 {
527 int len = ce_namelen(ce);
528 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
529 }
530
531 /*
532 * If we add a filename that aliases in the cache, we will use the
533 * name that we already have - but we don't want to update the same
534 * alias twice, because that implies that there were actually two
535 * different files with aliasing names!
536 *
537 * So we use the CE_ADDED flag to verify that the alias was an old
538 * one before we accept it as
539 */
540 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
541 {
542 int len;
543 struct cache_entry *new;
544
545 if (alias->ce_flags & CE_ADDED)
546 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
547
548 /* Ok, create the new entry using the name of the existing alias */
549 len = ce_namelen(alias);
550 new = xcalloc(1, cache_entry_size(len));
551 memcpy(new->name, alias->name, len);
552 copy_cache_entry(new, ce);
553 free(ce);
554 return new;
555 }
556
557 static void record_intent_to_add(struct cache_entry *ce)
558 {
559 unsigned char sha1[20];
560 if (write_sha1_file("", 0, blob_type, sha1))
561 die("cannot create an empty blob in the object database");
562 hashcpy(ce->sha1, sha1);
563 }
564
565 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
566 {
567 int size, namelen, was_same;
568 mode_t st_mode = st->st_mode;
569 struct cache_entry *ce, *alias;
570 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
571 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
572 int pretend = flags & ADD_CACHE_PRETEND;
573 int intent_only = flags & ADD_CACHE_INTENT;
574 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
575 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
576
577 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
578 return error("%s: can only add regular files, symbolic links or git-directories", path);
579
580 namelen = strlen(path);
581 if (S_ISDIR(st_mode)) {
582 while (namelen && path[namelen-1] == '/')
583 namelen--;
584 }
585 size = cache_entry_size(namelen);
586 ce = xcalloc(1, size);
587 memcpy(ce->name, path, namelen);
588 ce->ce_flags = namelen;
589 if (!intent_only)
590 fill_stat_cache_info(ce, st);
591 else
592 ce->ce_flags |= CE_INTENT_TO_ADD;
593
594 if (trust_executable_bit && has_symlinks)
595 ce->ce_mode = create_ce_mode(st_mode);
596 else {
597 /* If there is an existing entry, pick the mode bits and type
598 * from it, otherwise assume unexecutable regular file.
599 */
600 struct cache_entry *ent;
601 int pos = index_name_pos_also_unmerged(istate, path, namelen);
602
603 ent = (0 <= pos) ? istate->cache[pos] : NULL;
604 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
605 }
606
607 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
608 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
609 /* Nothing changed, really */
610 free(ce);
611 ce_mark_uptodate(alias);
612 alias->ce_flags |= CE_ADDED;
613 return 0;
614 }
615 if (!intent_only) {
616 if (index_path(ce->sha1, path, st, 1))
617 return error("unable to index file %s", path);
618 } else
619 record_intent_to_add(ce);
620
621 if (ignore_case && alias && different_name(ce, alias))
622 ce = create_alias_ce(ce, alias);
623 ce->ce_flags |= CE_ADDED;
624
625 /* It was suspected to be racily clean, but it turns out to be Ok */
626 was_same = (alias &&
627 !ce_stage(alias) &&
628 !hashcmp(alias->sha1, ce->sha1) &&
629 ce->ce_mode == alias->ce_mode);
630
631 if (pretend)
632 ;
633 else if (add_index_entry(istate, ce, add_option))
634 return error("unable to add %s to index",path);
635 if (verbose && !was_same)
636 printf("add '%s'\n", path);
637 return 0;
638 }
639
640 int add_file_to_index(struct index_state *istate, const char *path, int flags)
641 {
642 struct stat st;
643 if (lstat(path, &st))
644 die_errno("unable to stat '%s'", path);
645 return add_to_index(istate, path, &st, flags);
646 }
647
648 struct cache_entry *make_cache_entry(unsigned int mode,
649 const unsigned char *sha1, const char *path, int stage,
650 int refresh)
651 {
652 int size, len;
653 struct cache_entry *ce;
654
655 if (!verify_path(path)) {
656 error("Invalid path '%s'", path);
657 return NULL;
658 }
659
660 len = strlen(path);
661 size = cache_entry_size(len);
662 ce = xcalloc(1, size);
663
664 hashcpy(ce->sha1, sha1);
665 memcpy(ce->name, path, len);
666 ce->ce_flags = create_ce_flags(len, stage);
667 ce->ce_mode = create_ce_mode(mode);
668
669 if (refresh)
670 return refresh_cache_entry(ce, 0);
671
672 return ce;
673 }
674
675 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
676 {
677 int len = ce_namelen(a);
678 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
679 }
680
681 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
682 {
683 const char *match, *name;
684 int len;
685
686 if (!pathspec)
687 return 1;
688
689 len = ce_namelen(ce);
690 name = ce->name;
691 while ((match = *pathspec++) != NULL) {
692 int matchlen = strlen(match);
693 if (matchlen > len)
694 continue;
695 if (memcmp(name, match, matchlen))
696 continue;
697 if (matchlen && name[matchlen-1] == '/')
698 return 1;
699 if (name[matchlen] == '/' || !name[matchlen])
700 return 1;
701 if (!matchlen)
702 return 1;
703 }
704 return 0;
705 }
706
707 /*
708 * We fundamentally don't like some paths: we don't want
709 * dot or dot-dot anywhere, and for obvious reasons don't
710 * want to recurse into ".git" either.
711 *
712 * Also, we don't want double slashes or slashes at the
713 * end that can make pathnames ambiguous.
714 */
715 static int verify_dotfile(const char *rest)
716 {
717 /*
718 * The first character was '.', but that
719 * has already been discarded, we now test
720 * the rest.
721 */
722 switch (*rest) {
723 /* "." is not allowed */
724 case '\0': case '/':
725 return 0;
726
727 /*
728 * ".git" followed by NUL or slash is bad. This
729 * shares the path end test with the ".." case.
730 */
731 case 'g':
732 if (rest[1] != 'i')
733 break;
734 if (rest[2] != 't')
735 break;
736 rest += 2;
737 /* fallthrough */
738 case '.':
739 if (rest[1] == '\0' || rest[1] == '/')
740 return 0;
741 }
742 return 1;
743 }
744
745 int verify_path(const char *path)
746 {
747 char c;
748
749 goto inside;
750 for (;;) {
751 if (!c)
752 return 1;
753 if (c == '/') {
754 inside:
755 c = *path++;
756 switch (c) {
757 default:
758 continue;
759 case '/': case '\0':
760 break;
761 case '.':
762 if (verify_dotfile(path))
763 continue;
764 }
765 return 0;
766 }
767 c = *path++;
768 }
769 }
770
771 /*
772 * Do we have another file that has the beginning components being a
773 * proper superset of the name we're trying to add?
774 */
775 static int has_file_name(struct index_state *istate,
776 const struct cache_entry *ce, int pos, int ok_to_replace)
777 {
778 int retval = 0;
779 int len = ce_namelen(ce);
780 int stage = ce_stage(ce);
781 const char *name = ce->name;
782
783 while (pos < istate->cache_nr) {
784 struct cache_entry *p = istate->cache[pos++];
785
786 if (len >= ce_namelen(p))
787 break;
788 if (memcmp(name, p->name, len))
789 break;
790 if (ce_stage(p) != stage)
791 continue;
792 if (p->name[len] != '/')
793 continue;
794 if (p->ce_flags & CE_REMOVE)
795 continue;
796 retval = -1;
797 if (!ok_to_replace)
798 break;
799 remove_index_entry_at(istate, --pos);
800 }
801 return retval;
802 }
803
804 /*
805 * Do we have another file with a pathname that is a proper
806 * subset of the name we're trying to add?
807 */
808 static int has_dir_name(struct index_state *istate,
809 const struct cache_entry *ce, int pos, int ok_to_replace)
810 {
811 int retval = 0;
812 int stage = ce_stage(ce);
813 const char *name = ce->name;
814 const char *slash = name + ce_namelen(ce);
815
816 for (;;) {
817 int len;
818
819 for (;;) {
820 if (*--slash == '/')
821 break;
822 if (slash <= ce->name)
823 return retval;
824 }
825 len = slash - name;
826
827 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
828 if (pos >= 0) {
829 /*
830 * Found one, but not so fast. This could
831 * be a marker that says "I was here, but
832 * I am being removed". Such an entry is
833 * not a part of the resulting tree, and
834 * it is Ok to have a directory at the same
835 * path.
836 */
837 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
838 retval = -1;
839 if (!ok_to_replace)
840 break;
841 remove_index_entry_at(istate, pos);
842 continue;
843 }
844 }
845 else
846 pos = -pos-1;
847
848 /*
849 * Trivial optimization: if we find an entry that
850 * already matches the sub-directory, then we know
851 * we're ok, and we can exit.
852 */
853 while (pos < istate->cache_nr) {
854 struct cache_entry *p = istate->cache[pos];
855 if ((ce_namelen(p) <= len) ||
856 (p->name[len] != '/') ||
857 memcmp(p->name, name, len))
858 break; /* not our subdirectory */
859 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
860 /*
861 * p is at the same stage as our entry, and
862 * is a subdirectory of what we are looking
863 * at, so we cannot have conflicts at our
864 * level or anything shorter.
865 */
866 return retval;
867 pos++;
868 }
869 }
870 return retval;
871 }
872
873 /* We may be in a situation where we already have path/file and path
874 * is being added, or we already have path and path/file is being
875 * added. Either one would result in a nonsense tree that has path
876 * twice when git-write-tree tries to write it out. Prevent it.
877 *
878 * If ok-to-replace is specified, we remove the conflicting entries
879 * from the cache so the caller should recompute the insert position.
880 * When this happens, we return non-zero.
881 */
882 static int check_file_directory_conflict(struct index_state *istate,
883 const struct cache_entry *ce,
884 int pos, int ok_to_replace)
885 {
886 int retval;
887
888 /*
889 * When ce is an "I am going away" entry, we allow it to be added
890 */
891 if (ce->ce_flags & CE_REMOVE)
892 return 0;
893
894 /*
895 * We check if the path is a sub-path of a subsequent pathname
896 * first, since removing those will not change the position
897 * in the array.
898 */
899 retval = has_file_name(istate, ce, pos, ok_to_replace);
900
901 /*
902 * Then check if the path might have a clashing sub-directory
903 * before it.
904 */
905 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
906 }
907
908 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
909 {
910 int pos;
911 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
912 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
913 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
914 int new_only = option & ADD_CACHE_NEW_ONLY;
915
916 cache_tree_invalidate_path(istate->cache_tree, ce->name);
917 pos = index_name_pos(istate, ce->name, ce->ce_flags);
918
919 /* existing match? Just replace it. */
920 if (pos >= 0) {
921 if (!new_only)
922 replace_index_entry(istate, pos, ce);
923 return 0;
924 }
925 pos = -pos-1;
926
927 /*
928 * Inserting a merged entry ("stage 0") into the index
929 * will always replace all non-merged entries..
930 */
931 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
932 while (ce_same_name(istate->cache[pos], ce)) {
933 ok_to_add = 1;
934 if (!remove_index_entry_at(istate, pos))
935 break;
936 }
937 }
938
939 if (!ok_to_add)
940 return -1;
941 if (!verify_path(ce->name))
942 return error("Invalid path '%s'", ce->name);
943
944 if (!skip_df_check &&
945 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
946 if (!ok_to_replace)
947 return error("'%s' appears as both a file and as a directory",
948 ce->name);
949 pos = index_name_pos(istate, ce->name, ce->ce_flags);
950 pos = -pos-1;
951 }
952 return pos + 1;
953 }
954
955 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
956 {
957 int pos;
958
959 if (option & ADD_CACHE_JUST_APPEND)
960 pos = istate->cache_nr;
961 else {
962 int ret;
963 ret = add_index_entry_with_check(istate, ce, option);
964 if (ret <= 0)
965 return ret;
966 pos = ret - 1;
967 }
968
969 /* Make sure the array is big enough .. */
970 if (istate->cache_nr == istate->cache_alloc) {
971 istate->cache_alloc = alloc_nr(istate->cache_alloc);
972 istate->cache = xrealloc(istate->cache,
973 istate->cache_alloc * sizeof(struct cache_entry *));
974 }
975
976 /* Add it in.. */
977 istate->cache_nr++;
978 if (istate->cache_nr > pos + 1)
979 memmove(istate->cache + pos + 1,
980 istate->cache + pos,
981 (istate->cache_nr - pos - 1) * sizeof(ce));
982 set_index_entry(istate, pos, ce);
983 istate->cache_changed = 1;
984 return 0;
985 }
986
987 /*
988 * "refresh" does not calculate a new sha1 file or bring the
989 * cache up-to-date for mode/content changes. But what it
990 * _does_ do is to "re-match" the stat information of a file
991 * with the cache, so that you can refresh the cache for a
992 * file that hasn't been changed but where the stat entry is
993 * out of date.
994 *
995 * For example, you'd want to do this after doing a "git-read-tree",
996 * to link up the stat cache details with the proper files.
997 */
998 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
999 struct cache_entry *ce,
1000 unsigned int options, int *err)
1001 {
1002 struct stat st;
1003 struct cache_entry *updated;
1004 int changed, size;
1005 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1006
1007 if (ce_uptodate(ce))
1008 return ce;
1009
1010 /*
1011 * CE_VALID means the user promised us that the change to
1012 * the work tree does not matter and told us not to worry.
1013 */
1014 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1015 ce_mark_uptodate(ce);
1016 return ce;
1017 }
1018
1019 if (lstat(ce->name, &st) < 0) {
1020 if (err)
1021 *err = errno;
1022 return NULL;
1023 }
1024
1025 changed = ie_match_stat(istate, ce, &st, options);
1026 if (!changed) {
1027 /*
1028 * The path is unchanged. If we were told to ignore
1029 * valid bit, then we did the actual stat check and
1030 * found that the entry is unmodified. If the entry
1031 * is not marked VALID, this is the place to mark it
1032 * valid again, under "assume unchanged" mode.
1033 */
1034 if (ignore_valid && assume_unchanged &&
1035 !(ce->ce_flags & CE_VALID))
1036 ; /* mark this one VALID again */
1037 else {
1038 /*
1039 * We do not mark the index itself "modified"
1040 * because CE_UPTODATE flag is in-core only;
1041 * we are not going to write this change out.
1042 */
1043 ce_mark_uptodate(ce);
1044 return ce;
1045 }
1046 }
1047
1048 if (ie_modified(istate, ce, &st, options)) {
1049 if (err)
1050 *err = EINVAL;
1051 return NULL;
1052 }
1053
1054 size = ce_size(ce);
1055 updated = xmalloc(size);
1056 memcpy(updated, ce, size);
1057 fill_stat_cache_info(updated, &st);
1058 /*
1059 * If ignore_valid is not set, we should leave CE_VALID bit
1060 * alone. Otherwise, paths marked with --no-assume-unchanged
1061 * (i.e. things to be edited) will reacquire CE_VALID bit
1062 * automatically, which is not really what we want.
1063 */
1064 if (!ignore_valid && assume_unchanged &&
1065 !(ce->ce_flags & CE_VALID))
1066 updated->ce_flags &= ~CE_VALID;
1067
1068 return updated;
1069 }
1070
1071 static void show_file(const char * fmt, const char * name, int in_porcelain,
1072 int * first, char *header_msg)
1073 {
1074 if (in_porcelain && *first && header_msg) {
1075 printf("%s\n", header_msg);
1076 *first=0;
1077 }
1078 printf(fmt, name);
1079 }
1080
1081 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1082 char *seen, char *header_msg)
1083 {
1084 int i;
1085 int has_errors = 0;
1086 int really = (flags & REFRESH_REALLY) != 0;
1087 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1088 int quiet = (flags & REFRESH_QUIET) != 0;
1089 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1090 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1091 int first = 1;
1092 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1093 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1094 const char *needs_update_fmt;
1095 const char *needs_merge_fmt;
1096
1097 needs_update_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1098 needs_merge_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1099 for (i = 0; i < istate->cache_nr; i++) {
1100 struct cache_entry *ce, *new;
1101 int cache_errno = 0;
1102
1103 ce = istate->cache[i];
1104 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1105 continue;
1106
1107 if (ce_stage(ce)) {
1108 while ((i < istate->cache_nr) &&
1109 ! strcmp(istate->cache[i]->name, ce->name))
1110 i++;
1111 i--;
1112 if (allow_unmerged)
1113 continue;
1114 show_file(needs_merge_fmt, ce->name, in_porcelain, &first, header_msg);
1115 has_errors = 1;
1116 continue;
1117 }
1118
1119 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1120 continue;
1121
1122 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1123 if (new == ce)
1124 continue;
1125 if (!new) {
1126 if (not_new && cache_errno == ENOENT)
1127 continue;
1128 if (really && cache_errno == EINVAL) {
1129 /* If we are doing --really-refresh that
1130 * means the index is not valid anymore.
1131 */
1132 ce->ce_flags &= ~CE_VALID;
1133 istate->cache_changed = 1;
1134 }
1135 if (quiet)
1136 continue;
1137 show_file(needs_update_fmt, ce->name, in_porcelain, &first, header_msg);
1138 has_errors = 1;
1139 continue;
1140 }
1141
1142 replace_index_entry(istate, i, new);
1143 }
1144 return has_errors;
1145 }
1146
1147 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1148 {
1149 return refresh_cache_ent(&the_index, ce, really, NULL);
1150 }
1151
1152 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1153 {
1154 git_SHA_CTX c;
1155 unsigned char sha1[20];
1156
1157 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1158 return error("bad signature");
1159 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1160 return error("bad index version");
1161 git_SHA1_Init(&c);
1162 git_SHA1_Update(&c, hdr, size - 20);
1163 git_SHA1_Final(sha1, &c);
1164 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1165 return error("bad index file sha1 signature");
1166 return 0;
1167 }
1168
1169 static int read_index_extension(struct index_state *istate,
1170 const char *ext, void *data, unsigned long sz)
1171 {
1172 switch (CACHE_EXT(ext)) {
1173 case CACHE_EXT_TREE:
1174 istate->cache_tree = cache_tree_read(data, sz);
1175 break;
1176 case CACHE_EXT_RESOLVE_UNDO:
1177 istate->resolve_undo = resolve_undo_read(data, sz);
1178 break;
1179 default:
1180 if (*ext < 'A' || 'Z' < *ext)
1181 return error("index uses %.4s extension, which we do not understand",
1182 ext);
1183 fprintf(stderr, "ignoring %.4s extension\n", ext);
1184 break;
1185 }
1186 return 0;
1187 }
1188
1189 int read_index(struct index_state *istate)
1190 {
1191 return read_index_from(istate, get_index_file());
1192 }
1193
1194 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1195 {
1196 size_t len;
1197 const char *name;
1198
1199 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1200 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1201 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1202 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1203 ce->ce_dev = ntohl(ondisk->dev);
1204 ce->ce_ino = ntohl(ondisk->ino);
1205 ce->ce_mode = ntohl(ondisk->mode);
1206 ce->ce_uid = ntohl(ondisk->uid);
1207 ce->ce_gid = ntohl(ondisk->gid);
1208 ce->ce_size = ntohl(ondisk->size);
1209 /* On-disk flags are just 16 bits */
1210 ce->ce_flags = ntohs(ondisk->flags);
1211
1212 hashcpy(ce->sha1, ondisk->sha1);
1213
1214 len = ce->ce_flags & CE_NAMEMASK;
1215
1216 if (ce->ce_flags & CE_EXTENDED) {
1217 struct ondisk_cache_entry_extended *ondisk2;
1218 int extended_flags;
1219 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1220 extended_flags = ntohs(ondisk2->flags2) << 16;
1221 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1222 if (extended_flags & ~CE_EXTENDED_FLAGS)
1223 die("Unknown index entry format %08x", extended_flags);
1224 ce->ce_flags |= extended_flags;
1225 name = ondisk2->name;
1226 }
1227 else
1228 name = ondisk->name;
1229
1230 if (len == CE_NAMEMASK)
1231 len = strlen(name);
1232 /*
1233 * NEEDSWORK: If the original index is crafted, this copy could
1234 * go unchecked.
1235 */
1236 memcpy(ce->name, name, len + 1);
1237 }
1238
1239 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1240 {
1241 long per_entry;
1242
1243 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1244
1245 /*
1246 * Alignment can cause differences. This should be "alignof", but
1247 * since that's a gcc'ism, just use the size of a pointer.
1248 */
1249 per_entry += sizeof(void *);
1250 return ondisk_size + entries*per_entry;
1251 }
1252
1253 /* remember to discard_cache() before reading a different cache! */
1254 int read_index_from(struct index_state *istate, const char *path)
1255 {
1256 int fd, i;
1257 struct stat st;
1258 unsigned long src_offset, dst_offset;
1259 struct cache_header *hdr;
1260 void *mmap;
1261 size_t mmap_size;
1262
1263 errno = EBUSY;
1264 if (istate->initialized)
1265 return istate->cache_nr;
1266
1267 errno = ENOENT;
1268 istate->timestamp.sec = 0;
1269 istate->timestamp.nsec = 0;
1270 fd = open(path, O_RDONLY);
1271 if (fd < 0) {
1272 if (errno == ENOENT)
1273 return 0;
1274 die_errno("index file open failed");
1275 }
1276
1277 if (fstat(fd, &st))
1278 die_errno("cannot stat the open index");
1279
1280 errno = EINVAL;
1281 mmap_size = xsize_t(st.st_size);
1282 if (mmap_size < sizeof(struct cache_header) + 20)
1283 die("index file smaller than expected");
1284
1285 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1286 close(fd);
1287 if (mmap == MAP_FAILED)
1288 die_errno("unable to map index file");
1289
1290 hdr = mmap;
1291 if (verify_hdr(hdr, mmap_size) < 0)
1292 goto unmap;
1293
1294 istate->cache_nr = ntohl(hdr->hdr_entries);
1295 istate->cache_alloc = alloc_nr(istate->cache_nr);
1296 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1297
1298 /*
1299 * The disk format is actually larger than the in-memory format,
1300 * due to space for nsec etc, so even though the in-memory one
1301 * has room for a few more flags, we can allocate using the same
1302 * index size
1303 */
1304 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1305 istate->initialized = 1;
1306
1307 src_offset = sizeof(*hdr);
1308 dst_offset = 0;
1309 for (i = 0; i < istate->cache_nr; i++) {
1310 struct ondisk_cache_entry *disk_ce;
1311 struct cache_entry *ce;
1312
1313 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1314 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1315 convert_from_disk(disk_ce, ce);
1316 set_index_entry(istate, i, ce);
1317
1318 src_offset += ondisk_ce_size(ce);
1319 dst_offset += ce_size(ce);
1320 }
1321 istate->timestamp.sec = st.st_mtime;
1322 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1323
1324 while (src_offset <= mmap_size - 20 - 8) {
1325 /* After an array of active_nr index entries,
1326 * there can be arbitrary number of extended
1327 * sections, each of which is prefixed with
1328 * extension name (4-byte) and section length
1329 * in 4-byte network byte order.
1330 */
1331 unsigned long extsize;
1332 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1333 extsize = ntohl(extsize);
1334 if (read_index_extension(istate,
1335 (const char *) mmap + src_offset,
1336 (char *) mmap + src_offset + 8,
1337 extsize) < 0)
1338 goto unmap;
1339 src_offset += 8;
1340 src_offset += extsize;
1341 }
1342 munmap(mmap, mmap_size);
1343 return istate->cache_nr;
1344
1345 unmap:
1346 munmap(mmap, mmap_size);
1347 errno = EINVAL;
1348 die("index file corrupt");
1349 }
1350
1351 int is_index_unborn(struct index_state *istate)
1352 {
1353 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
1354 }
1355
1356 int discard_index(struct index_state *istate)
1357 {
1358 resolve_undo_clear_index(istate);
1359 istate->cache_nr = 0;
1360 istate->cache_changed = 0;
1361 istate->timestamp.sec = 0;
1362 istate->timestamp.nsec = 0;
1363 istate->name_hash_initialized = 0;
1364 free_hash(&istate->name_hash);
1365 cache_tree_free(&(istate->cache_tree));
1366 free(istate->alloc);
1367 istate->alloc = NULL;
1368 istate->initialized = 0;
1369
1370 /* no need to throw away allocated active_cache */
1371 return 0;
1372 }
1373
1374 int unmerged_index(const struct index_state *istate)
1375 {
1376 int i;
1377 for (i = 0; i < istate->cache_nr; i++) {
1378 if (ce_stage(istate->cache[i]))
1379 return 1;
1380 }
1381 return 0;
1382 }
1383
1384 #define WRITE_BUFFER_SIZE 8192
1385 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1386 static unsigned long write_buffer_len;
1387
1388 static int ce_write_flush(git_SHA_CTX *context, int fd)
1389 {
1390 unsigned int buffered = write_buffer_len;
1391 if (buffered) {
1392 git_SHA1_Update(context, write_buffer, buffered);
1393 if (write_in_full(fd, write_buffer, buffered) != buffered)
1394 return -1;
1395 write_buffer_len = 0;
1396 }
1397 return 0;
1398 }
1399
1400 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1401 {
1402 while (len) {
1403 unsigned int buffered = write_buffer_len;
1404 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1405 if (partial > len)
1406 partial = len;
1407 memcpy(write_buffer + buffered, data, partial);
1408 buffered += partial;
1409 if (buffered == WRITE_BUFFER_SIZE) {
1410 write_buffer_len = buffered;
1411 if (ce_write_flush(context, fd))
1412 return -1;
1413 buffered = 0;
1414 }
1415 write_buffer_len = buffered;
1416 len -= partial;
1417 data = (char *) data + partial;
1418 }
1419 return 0;
1420 }
1421
1422 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1423 unsigned int ext, unsigned int sz)
1424 {
1425 ext = htonl(ext);
1426 sz = htonl(sz);
1427 return ((ce_write(context, fd, &ext, 4) < 0) ||
1428 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1429 }
1430
1431 static int ce_flush(git_SHA_CTX *context, int fd)
1432 {
1433 unsigned int left = write_buffer_len;
1434
1435 if (left) {
1436 write_buffer_len = 0;
1437 git_SHA1_Update(context, write_buffer, left);
1438 }
1439
1440 /* Flush first if not enough space for SHA1 signature */
1441 if (left + 20 > WRITE_BUFFER_SIZE) {
1442 if (write_in_full(fd, write_buffer, left) != left)
1443 return -1;
1444 left = 0;
1445 }
1446
1447 /* Append the SHA1 signature at the end */
1448 git_SHA1_Final(write_buffer + left, context);
1449 left += 20;
1450 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1451 }
1452
1453 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1454 {
1455 /*
1456 * The only thing we care about in this function is to smudge the
1457 * falsely clean entry due to touch-update-touch race, so we leave
1458 * everything else as they are. We are called for entries whose
1459 * ce_mtime match the index file mtime.
1460 *
1461 * Note that this actually does not do much for gitlinks, for
1462 * which ce_match_stat_basic() always goes to the actual
1463 * contents. The caller checks with is_racy_timestamp() which
1464 * always says "no" for gitlinks, so we are not called for them ;-)
1465 */
1466 struct stat st;
1467
1468 if (lstat(ce->name, &st) < 0)
1469 return;
1470 if (ce_match_stat_basic(ce, &st))
1471 return;
1472 if (ce_modified_check_fs(ce, &st)) {
1473 /* This is "racily clean"; smudge it. Note that this
1474 * is a tricky code. At first glance, it may appear
1475 * that it can break with this sequence:
1476 *
1477 * $ echo xyzzy >frotz
1478 * $ git-update-index --add frotz
1479 * $ : >frotz
1480 * $ sleep 3
1481 * $ echo filfre >nitfol
1482 * $ git-update-index --add nitfol
1483 *
1484 * but it does not. When the second update-index runs,
1485 * it notices that the entry "frotz" has the same timestamp
1486 * as index, and if we were to smudge it by resetting its
1487 * size to zero here, then the object name recorded
1488 * in index is the 6-byte file but the cached stat information
1489 * becomes zero --- which would then match what we would
1490 * obtain from the filesystem next time we stat("frotz").
1491 *
1492 * However, the second update-index, before calling
1493 * this function, notices that the cached size is 6
1494 * bytes and what is on the filesystem is an empty
1495 * file, and never calls us, so the cached size information
1496 * for "frotz" stays 6 which does not match the filesystem.
1497 */
1498 ce->ce_size = 0;
1499 }
1500 }
1501
1502 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1503 {
1504 int size = ondisk_ce_size(ce);
1505 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1506 char *name;
1507
1508 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1509 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1510 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1511 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1512 ondisk->dev = htonl(ce->ce_dev);
1513 ondisk->ino = htonl(ce->ce_ino);
1514 ondisk->mode = htonl(ce->ce_mode);
1515 ondisk->uid = htonl(ce->ce_uid);
1516 ondisk->gid = htonl(ce->ce_gid);
1517 ondisk->size = htonl(ce->ce_size);
1518 hashcpy(ondisk->sha1, ce->sha1);
1519 ondisk->flags = htons(ce->ce_flags);
1520 if (ce->ce_flags & CE_EXTENDED) {
1521 struct ondisk_cache_entry_extended *ondisk2;
1522 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1523 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1524 name = ondisk2->name;
1525 }
1526 else
1527 name = ondisk->name;
1528 memcpy(name, ce->name, ce_namelen(ce));
1529
1530 return ce_write(c, fd, ondisk, size);
1531 }
1532
1533 int write_index(struct index_state *istate, int newfd)
1534 {
1535 git_SHA_CTX c;
1536 struct cache_header hdr;
1537 int i, err, removed, extended;
1538 struct cache_entry **cache = istate->cache;
1539 int entries = istate->cache_nr;
1540 struct stat st;
1541
1542 for (i = removed = extended = 0; i < entries; i++) {
1543 if (cache[i]->ce_flags & CE_REMOVE)
1544 removed++;
1545
1546 /* reduce extended entries if possible */
1547 cache[i]->ce_flags &= ~CE_EXTENDED;
1548 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1549 extended++;
1550 cache[i]->ce_flags |= CE_EXTENDED;
1551 }
1552 }
1553
1554 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1555 /* for extended format, increase version so older git won't try to read it */
1556 hdr.hdr_version = htonl(extended ? 3 : 2);
1557 hdr.hdr_entries = htonl(entries - removed);
1558
1559 git_SHA1_Init(&c);
1560 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1561 return -1;
1562
1563 for (i = 0; i < entries; i++) {
1564 struct cache_entry *ce = cache[i];
1565 if (ce->ce_flags & CE_REMOVE)
1566 continue;
1567 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1568 ce_smudge_racily_clean_entry(ce);
1569 if (ce_write_entry(&c, newfd, ce) < 0)
1570 return -1;
1571 }
1572
1573 /* Write extension data here */
1574 if (istate->cache_tree) {
1575 struct strbuf sb = STRBUF_INIT;
1576
1577 cache_tree_write(&sb, istate->cache_tree);
1578 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1579 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1580 strbuf_release(&sb);
1581 if (err)
1582 return -1;
1583 }
1584 if (istate->resolve_undo) {
1585 struct strbuf sb = STRBUF_INIT;
1586
1587 resolve_undo_write(&sb, istate->resolve_undo);
1588 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1589 sb.len) < 0
1590 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1591 strbuf_release(&sb);
1592 if (err)
1593 return -1;
1594 }
1595
1596 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1597 return -1;
1598 istate->timestamp.sec = (unsigned int)st.st_mtime;
1599 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1600 return 0;
1601 }
1602
1603 /*
1604 * Read the index file that is potentially unmerged into given
1605 * index_state, dropping any unmerged entries. Returns true if
1606 * the index is unmerged. Callers who want to refuse to work
1607 * from an unmerged state can call this and check its return value,
1608 * instead of calling read_cache().
1609 */
1610 int read_index_unmerged(struct index_state *istate)
1611 {
1612 int i;
1613 int unmerged = 0;
1614
1615 read_index(istate);
1616 for (i = 0; i < istate->cache_nr; i++) {
1617 struct cache_entry *ce = istate->cache[i];
1618 struct cache_entry *new_ce;
1619 int size, len;
1620
1621 if (!ce_stage(ce))
1622 continue;
1623 unmerged = 1;
1624 len = strlen(ce->name);
1625 size = cache_entry_size(len);
1626 new_ce = xcalloc(1, size);
1627 hashcpy(new_ce->sha1, ce->sha1);
1628 memcpy(new_ce->name, ce->name, len);
1629 new_ce->ce_flags = create_ce_flags(len, 0);
1630 new_ce->ce_mode = ce->ce_mode;
1631 if (add_index_entry(istate, new_ce, 0))
1632 return error("%s: cannot drop to stage #0",
1633 ce->name);
1634 i = index_name_pos(istate, new_ce->name, len);
1635 }
1636 return unmerged;
1637 }
1638
1639 struct update_callback_data
1640 {
1641 int flags;
1642 int add_errors;
1643 };
1644
1645 static void update_callback(struct diff_queue_struct *q,
1646 struct diff_options *opt, void *cbdata)
1647 {
1648 int i;
1649 struct update_callback_data *data = cbdata;
1650
1651 for (i = 0; i < q->nr; i++) {
1652 struct diff_filepair *p = q->queue[i];
1653 const char *path = p->one->path;
1654 switch (p->status) {
1655 default:
1656 die("unexpected diff status %c", p->status);
1657 case DIFF_STATUS_UNMERGED:
1658 /*
1659 * ADD_CACHE_IGNORE_REMOVAL is unset if "git
1660 * add -u" is calling us, In such a case, a
1661 * missing work tree file needs to be removed
1662 * if there is an unmerged entry at stage #2,
1663 * but such a diff record is followed by
1664 * another with DIFF_STATUS_DELETED (and if
1665 * there is no stage #2, we won't see DELETED
1666 * nor MODIFIED). We can simply continue
1667 * either way.
1668 */
1669 if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL))
1670 continue;
1671 /*
1672 * Otherwise, it is "git add path" is asking
1673 * to explicitly add it; we fall through. A
1674 * missing work tree file is an error and is
1675 * caught by add_file_to_index() in such a
1676 * case.
1677 */
1678 case DIFF_STATUS_MODIFIED:
1679 case DIFF_STATUS_TYPE_CHANGED:
1680 if (add_file_to_index(&the_index, path, data->flags)) {
1681 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1682 die("updating files failed");
1683 data->add_errors++;
1684 }
1685 break;
1686 case DIFF_STATUS_DELETED:
1687 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1688 break;
1689 if (!(data->flags & ADD_CACHE_PRETEND))
1690 remove_file_from_index(&the_index, path);
1691 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1692 printf("remove '%s'\n", path);
1693 break;
1694 }
1695 }
1696 }
1697
1698 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1699 {
1700 struct update_callback_data data;
1701 struct rev_info rev;
1702 init_revisions(&rev, prefix);
1703 setup_revisions(0, NULL, &rev, NULL);
1704 rev.prune_data = pathspec;
1705 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1706 rev.diffopt.format_callback = update_callback;
1707 data.flags = flags;
1708 data.add_errors = 0;
1709 rev.diffopt.format_callback_data = &data;
1710 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1711 return !!data.add_errors;
1712 }
1713
1714 /*
1715 * Returns 1 if the path is an "other" path with respect to
1716 * the index; that is, the path is not mentioned in the index at all,
1717 * either as a file, a directory with some files in the index,
1718 * or as an unmerged entry.
1719 *
1720 * We helpfully remove a trailing "/" from directories so that
1721 * the output of read_directory can be used as-is.
1722 */
1723 int index_name_is_other(const struct index_state *istate, const char *name,
1724 int namelen)
1725 {
1726 int pos;
1727 if (namelen && name[namelen - 1] == '/')
1728 namelen--;
1729 pos = index_name_pos(istate, name, namelen);
1730 if (0 <= pos)
1731 return 0; /* exact match */
1732 pos = -pos - 1;
1733 if (pos < istate->cache_nr) {
1734 struct cache_entry *ce = istate->cache[pos];
1735 if (ce_namelen(ce) == namelen &&
1736 !memcmp(ce->name, name, namelen))
1737 return 0; /* Yup, this one exists unmerged */
1738 }
1739 return 1;
1740 }