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