b4653c083a6af62ca583ad2f0c9dfb35594bdbcb
[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 "blob.h"
14 #include "resolve-undo.h"
15 #include "strbuf.h"
16 #include "varint.h"
17
18 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
19 unsigned int options);
20
21 /* Mask for the name length in ce_flags in the on-disk index */
22
23 #define CE_NAMEMASK (0x0fff)
24
25 /* Index extensions.
26 *
27 * The first letter should be 'A'..'Z' for extensions that are not
28 * necessary for a correct operation (i.e. optimization data).
29 * When new extensions are added that _needs_ to be understood in
30 * order to correctly interpret the index file, pick character that
31 * is outside the range, to cause the reader to abort.
32 */
33
34 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
35 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
36 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
37
38 struct index_state the_index;
39 static const char *alternate_index_output;
40
41 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
42 {
43 istate->cache[nr] = ce;
44 add_name_hash(istate, ce);
45 }
46
47 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
48 {
49 struct cache_entry *old = istate->cache[nr];
50
51 remove_name_hash(istate, old);
52 free(old);
53 set_index_entry(istate, nr, ce);
54 istate->cache_changed |= CE_ENTRY_CHANGED;
55 }
56
57 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
58 {
59 struct cache_entry *old = istate->cache[nr], *new;
60 int namelen = strlen(new_name);
61
62 new = xmalloc(cache_entry_size(namelen));
63 copy_cache_entry(new, old);
64 new->ce_flags &= ~CE_HASHED;
65 new->ce_namelen = namelen;
66 memcpy(new->name, new_name, namelen + 1);
67
68 cache_tree_invalidate_path(istate, old->name);
69 remove_index_entry_at(istate, nr);
70 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
71 }
72
73 void fill_stat_data(struct stat_data *sd, struct stat *st)
74 {
75 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
76 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
77 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
78 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
79 sd->sd_dev = st->st_dev;
80 sd->sd_ino = st->st_ino;
81 sd->sd_uid = st->st_uid;
82 sd->sd_gid = st->st_gid;
83 sd->sd_size = st->st_size;
84 }
85
86 int match_stat_data(const struct stat_data *sd, struct stat *st)
87 {
88 int changed = 0;
89
90 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
91 changed |= MTIME_CHANGED;
92 if (trust_ctime && check_stat &&
93 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
94 changed |= CTIME_CHANGED;
95
96 #ifdef USE_NSEC
97 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
98 changed |= MTIME_CHANGED;
99 if (trust_ctime && check_stat &&
100 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
101 changed |= CTIME_CHANGED;
102 #endif
103
104 if (check_stat) {
105 if (sd->sd_uid != (unsigned int) st->st_uid ||
106 sd->sd_gid != (unsigned int) st->st_gid)
107 changed |= OWNER_CHANGED;
108 if (sd->sd_ino != (unsigned int) st->st_ino)
109 changed |= INODE_CHANGED;
110 }
111
112 #ifdef USE_STDEV
113 /*
114 * st_dev breaks on network filesystems where different
115 * clients will have different views of what "device"
116 * the filesystem is on
117 */
118 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
119 changed |= INODE_CHANGED;
120 #endif
121
122 if (sd->sd_size != (unsigned int) st->st_size)
123 changed |= DATA_CHANGED;
124
125 return changed;
126 }
127
128 /*
129 * This only updates the "non-critical" parts of the directory
130 * cache, ie the parts that aren't tracked by GIT, and only used
131 * to validate the cache.
132 */
133 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
134 {
135 fill_stat_data(&ce->ce_stat_data, st);
136
137 if (assume_unchanged)
138 ce->ce_flags |= CE_VALID;
139
140 if (S_ISREG(st->st_mode))
141 ce_mark_uptodate(ce);
142 }
143
144 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
145 {
146 int match = -1;
147 int fd = open(ce->name, O_RDONLY);
148
149 if (fd >= 0) {
150 unsigned char sha1[20];
151 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
152 match = hashcmp(sha1, ce->sha1);
153 /* index_fd() closed the file descriptor already */
154 }
155 return match;
156 }
157
158 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
159 {
160 int match = -1;
161 void *buffer;
162 unsigned long size;
163 enum object_type type;
164 struct strbuf sb = STRBUF_INIT;
165
166 if (strbuf_readlink(&sb, ce->name, expected_size))
167 return -1;
168
169 buffer = read_sha1_file(ce->sha1, &type, &size);
170 if (buffer) {
171 if (size == sb.len)
172 match = memcmp(buffer, sb.buf, size);
173 free(buffer);
174 }
175 strbuf_release(&sb);
176 return match;
177 }
178
179 static int ce_compare_gitlink(const struct cache_entry *ce)
180 {
181 unsigned char sha1[20];
182
183 /*
184 * We don't actually require that the .git directory
185 * under GITLINK directory be a valid git directory. It
186 * might even be missing (in case nobody populated that
187 * sub-project).
188 *
189 * If so, we consider it always to match.
190 */
191 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
192 return 0;
193 return hashcmp(sha1, ce->sha1);
194 }
195
196 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
197 {
198 switch (st->st_mode & S_IFMT) {
199 case S_IFREG:
200 if (ce_compare_data(ce, st))
201 return DATA_CHANGED;
202 break;
203 case S_IFLNK:
204 if (ce_compare_link(ce, xsize_t(st->st_size)))
205 return DATA_CHANGED;
206 break;
207 case S_IFDIR:
208 if (S_ISGITLINK(ce->ce_mode))
209 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
210 default:
211 return TYPE_CHANGED;
212 }
213 return 0;
214 }
215
216 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
217 {
218 unsigned int changed = 0;
219
220 if (ce->ce_flags & CE_REMOVE)
221 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
222
223 switch (ce->ce_mode & S_IFMT) {
224 case S_IFREG:
225 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
226 /* We consider only the owner x bit to be relevant for
227 * "mode changes"
228 */
229 if (trust_executable_bit &&
230 (0100 & (ce->ce_mode ^ st->st_mode)))
231 changed |= MODE_CHANGED;
232 break;
233 case S_IFLNK:
234 if (!S_ISLNK(st->st_mode) &&
235 (has_symlinks || !S_ISREG(st->st_mode)))
236 changed |= TYPE_CHANGED;
237 break;
238 case S_IFGITLINK:
239 /* We ignore most of the st_xxx fields for gitlinks */
240 if (!S_ISDIR(st->st_mode))
241 changed |= TYPE_CHANGED;
242 else if (ce_compare_gitlink(ce))
243 changed |= DATA_CHANGED;
244 return changed;
245 default:
246 die("internal error: ce_mode is %o", ce->ce_mode);
247 }
248
249 changed |= match_stat_data(&ce->ce_stat_data, st);
250
251 /* Racily smudged entry? */
252 if (!ce->ce_stat_data.sd_size) {
253 if (!is_empty_blob_sha1(ce->sha1))
254 changed |= DATA_CHANGED;
255 }
256
257 return changed;
258 }
259
260 static int is_racy_timestamp(const struct index_state *istate,
261 const struct cache_entry *ce)
262 {
263 return (!S_ISGITLINK(ce->ce_mode) &&
264 istate->timestamp.sec &&
265 #ifdef USE_NSEC
266 /* nanosecond timestamped files can also be racy! */
267 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
268 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
269 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
270 #else
271 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
272 #endif
273 );
274 }
275
276 int ie_match_stat(const struct index_state *istate,
277 const struct cache_entry *ce, struct stat *st,
278 unsigned int options)
279 {
280 unsigned int changed;
281 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
282 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
283 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
284
285 /*
286 * If it's marked as always valid in the index, it's
287 * valid whatever the checked-out copy says.
288 *
289 * skip-worktree has the same effect with higher precedence
290 */
291 if (!ignore_skip_worktree && ce_skip_worktree(ce))
292 return 0;
293 if (!ignore_valid && (ce->ce_flags & CE_VALID))
294 return 0;
295
296 /*
297 * Intent-to-add entries have not been added, so the index entry
298 * by definition never matches what is in the work tree until it
299 * actually gets added.
300 */
301 if (ce->ce_flags & CE_INTENT_TO_ADD)
302 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
303
304 changed = ce_match_stat_basic(ce, st);
305
306 /*
307 * Within 1 second of this sequence:
308 * echo xyzzy >file && git-update-index --add file
309 * running this command:
310 * echo frotz >file
311 * would give a falsely clean cache entry. The mtime and
312 * length match the cache, and other stat fields do not change.
313 *
314 * We could detect this at update-index time (the cache entry
315 * being registered/updated records the same time as "now")
316 * and delay the return from git-update-index, but that would
317 * effectively mean we can make at most one commit per second,
318 * which is not acceptable. Instead, we check cache entries
319 * whose mtime are the same as the index file timestamp more
320 * carefully than others.
321 */
322 if (!changed && is_racy_timestamp(istate, ce)) {
323 if (assume_racy_is_modified)
324 changed |= DATA_CHANGED;
325 else
326 changed |= ce_modified_check_fs(ce, st);
327 }
328
329 return changed;
330 }
331
332 int ie_modified(const struct index_state *istate,
333 const struct cache_entry *ce,
334 struct stat *st, unsigned int options)
335 {
336 int changed, changed_fs;
337
338 changed = ie_match_stat(istate, ce, st, options);
339 if (!changed)
340 return 0;
341 /*
342 * If the mode or type has changed, there's no point in trying
343 * to refresh the entry - it's not going to match
344 */
345 if (changed & (MODE_CHANGED | TYPE_CHANGED))
346 return changed;
347
348 /*
349 * Immediately after read-tree or update-index --cacheinfo,
350 * the length field is zero, as we have never even read the
351 * lstat(2) information once, and we cannot trust DATA_CHANGED
352 * returned by ie_match_stat() which in turn was returned by
353 * ce_match_stat_basic() to signal that the filesize of the
354 * blob changed. We have to actually go to the filesystem to
355 * see if the contents match, and if so, should answer "unchanged".
356 *
357 * The logic does not apply to gitlinks, as ce_match_stat_basic()
358 * already has checked the actual HEAD from the filesystem in the
359 * subproject. If ie_match_stat() already said it is different,
360 * then we know it is.
361 */
362 if ((changed & DATA_CHANGED) &&
363 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
364 return changed;
365
366 changed_fs = ce_modified_check_fs(ce, st);
367 if (changed_fs)
368 return changed | changed_fs;
369 return 0;
370 }
371
372 int base_name_compare(const char *name1, int len1, int mode1,
373 const char *name2, int len2, int mode2)
374 {
375 unsigned char c1, c2;
376 int len = len1 < len2 ? len1 : len2;
377 int cmp;
378
379 cmp = memcmp(name1, name2, len);
380 if (cmp)
381 return cmp;
382 c1 = name1[len];
383 c2 = name2[len];
384 if (!c1 && S_ISDIR(mode1))
385 c1 = '/';
386 if (!c2 && S_ISDIR(mode2))
387 c2 = '/';
388 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
389 }
390
391 /*
392 * df_name_compare() is identical to base_name_compare(), except it
393 * compares conflicting directory/file entries as equal. Note that
394 * while a directory name compares as equal to a regular file, they
395 * then individually compare _differently_ to a filename that has
396 * a dot after the basename (because '\0' < '.' < '/').
397 *
398 * This is used by routines that want to traverse the git namespace
399 * but then handle conflicting entries together when possible.
400 */
401 int df_name_compare(const char *name1, int len1, int mode1,
402 const char *name2, int len2, int mode2)
403 {
404 int len = len1 < len2 ? len1 : len2, cmp;
405 unsigned char c1, c2;
406
407 cmp = memcmp(name1, name2, len);
408 if (cmp)
409 return cmp;
410 /* Directories and files compare equal (same length, same name) */
411 if (len1 == len2)
412 return 0;
413 c1 = name1[len];
414 if (!c1 && S_ISDIR(mode1))
415 c1 = '/';
416 c2 = name2[len];
417 if (!c2 && S_ISDIR(mode2))
418 c2 = '/';
419 if (c1 == '/' && !c2)
420 return 0;
421 if (c2 == '/' && !c1)
422 return 0;
423 return c1 - c2;
424 }
425
426 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
427 {
428 int len = len1 < len2 ? len1 : len2;
429 int cmp;
430
431 cmp = memcmp(name1, name2, len);
432 if (cmp)
433 return cmp;
434 if (len1 < len2)
435 return -1;
436 if (len1 > len2)
437 return 1;
438
439 if (stage1 < stage2)
440 return -1;
441 if (stage1 > stage2)
442 return 1;
443 return 0;
444 }
445
446 int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
447 {
448 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
449 }
450
451 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
452 {
453 int first, last;
454
455 first = 0;
456 last = istate->cache_nr;
457 while (last > first) {
458 int next = (last + first) >> 1;
459 struct cache_entry *ce = istate->cache[next];
460 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
461 if (!cmp)
462 return next;
463 if (cmp < 0) {
464 last = next;
465 continue;
466 }
467 first = next+1;
468 }
469 return -first-1;
470 }
471
472 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
473 {
474 return index_name_stage_pos(istate, name, namelen, 0);
475 }
476
477 /* Remove entry, return true if there are more entries to go.. */
478 int remove_index_entry_at(struct index_state *istate, int pos)
479 {
480 struct cache_entry *ce = istate->cache[pos];
481
482 record_resolve_undo(istate, ce);
483 remove_name_hash(istate, ce);
484 free(ce);
485 istate->cache_changed |= CE_ENTRY_REMOVED;
486 istate->cache_nr--;
487 if (pos >= istate->cache_nr)
488 return 0;
489 memmove(istate->cache + pos,
490 istate->cache + pos + 1,
491 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
492 return 1;
493 }
494
495 /*
496 * Remove all cache entries marked for removal, that is where
497 * CE_REMOVE is set in ce_flags. This is much more effective than
498 * calling remove_index_entry_at() for each entry to be removed.
499 */
500 void remove_marked_cache_entries(struct index_state *istate)
501 {
502 struct cache_entry **ce_array = istate->cache;
503 unsigned int i, j;
504
505 for (i = j = 0; i < istate->cache_nr; i++) {
506 if (ce_array[i]->ce_flags & CE_REMOVE) {
507 remove_name_hash(istate, ce_array[i]);
508 free(ce_array[i]);
509 }
510 else
511 ce_array[j++] = ce_array[i];
512 }
513 if (j == istate->cache_nr)
514 return;
515 istate->cache_changed |= CE_ENTRY_REMOVED;
516 istate->cache_nr = j;
517 }
518
519 int remove_file_from_index(struct index_state *istate, const char *path)
520 {
521 int pos = index_name_pos(istate, path, strlen(path));
522 if (pos < 0)
523 pos = -pos-1;
524 cache_tree_invalidate_path(istate, path);
525 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
526 remove_index_entry_at(istate, pos);
527 return 0;
528 }
529
530 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
531 {
532 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
533 }
534
535 static int index_name_pos_also_unmerged(struct index_state *istate,
536 const char *path, int namelen)
537 {
538 int pos = index_name_pos(istate, path, namelen);
539 struct cache_entry *ce;
540
541 if (pos >= 0)
542 return pos;
543
544 /* maybe unmerged? */
545 pos = -1 - pos;
546 if (pos >= istate->cache_nr ||
547 compare_name((ce = istate->cache[pos]), path, namelen))
548 return -1;
549
550 /* order of preference: stage 2, 1, 3 */
551 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
552 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
553 !compare_name(ce, path, namelen))
554 pos++;
555 return pos;
556 }
557
558 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
559 {
560 int len = ce_namelen(ce);
561 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
562 }
563
564 /*
565 * If we add a filename that aliases in the cache, we will use the
566 * name that we already have - but we don't want to update the same
567 * alias twice, because that implies that there were actually two
568 * different files with aliasing names!
569 *
570 * So we use the CE_ADDED flag to verify that the alias was an old
571 * one before we accept it as
572 */
573 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
574 {
575 int len;
576 struct cache_entry *new;
577
578 if (alias->ce_flags & CE_ADDED)
579 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
580
581 /* Ok, create the new entry using the name of the existing alias */
582 len = ce_namelen(alias);
583 new = xcalloc(1, cache_entry_size(len));
584 memcpy(new->name, alias->name, len);
585 copy_cache_entry(new, ce);
586 free(ce);
587 return new;
588 }
589
590 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
591 {
592 unsigned char sha1[20];
593 if (write_sha1_file("", 0, blob_type, sha1))
594 die("cannot create an empty blob in the object database");
595 hashcpy(ce->sha1, sha1);
596 }
597
598 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
599 {
600 int size, namelen, was_same;
601 mode_t st_mode = st->st_mode;
602 struct cache_entry *ce, *alias;
603 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
604 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
605 int pretend = flags & ADD_CACHE_PRETEND;
606 int intent_only = flags & ADD_CACHE_INTENT;
607 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
608 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
609
610 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
611 return error("%s: can only add regular files, symbolic links or git-directories", path);
612
613 namelen = strlen(path);
614 if (S_ISDIR(st_mode)) {
615 while (namelen && path[namelen-1] == '/')
616 namelen--;
617 }
618 size = cache_entry_size(namelen);
619 ce = xcalloc(1, size);
620 memcpy(ce->name, path, namelen);
621 ce->ce_namelen = namelen;
622 if (!intent_only)
623 fill_stat_cache_info(ce, st);
624 else
625 ce->ce_flags |= CE_INTENT_TO_ADD;
626
627 if (trust_executable_bit && has_symlinks)
628 ce->ce_mode = create_ce_mode(st_mode);
629 else {
630 /* If there is an existing entry, pick the mode bits and type
631 * from it, otherwise assume unexecutable regular file.
632 */
633 struct cache_entry *ent;
634 int pos = index_name_pos_also_unmerged(istate, path, namelen);
635
636 ent = (0 <= pos) ? istate->cache[pos] : NULL;
637 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
638 }
639
640 /* When core.ignorecase=true, determine if a directory of the same name but differing
641 * case already exists within the Git repository. If it does, ensure the directory
642 * case of the file being added to the repository matches (is folded into) the existing
643 * entry's directory case.
644 */
645 if (ignore_case) {
646 const char *startPtr = ce->name;
647 const char *ptr = startPtr;
648 while (*ptr) {
649 while (*ptr && *ptr != '/')
650 ++ptr;
651 if (*ptr == '/') {
652 struct cache_entry *foundce;
653 ++ptr;
654 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
655 if (foundce) {
656 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
657 startPtr = ptr;
658 }
659 }
660 }
661 }
662
663 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
664 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
665 /* Nothing changed, really */
666 free(ce);
667 if (!S_ISGITLINK(alias->ce_mode))
668 ce_mark_uptodate(alias);
669 alias->ce_flags |= CE_ADDED;
670 return 0;
671 }
672 if (!intent_only) {
673 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
674 return error("unable to index file %s", path);
675 } else
676 set_object_name_for_intent_to_add_entry(ce);
677
678 if (ignore_case && alias && different_name(ce, alias))
679 ce = create_alias_ce(ce, alias);
680 ce->ce_flags |= CE_ADDED;
681
682 /* It was suspected to be racily clean, but it turns out to be Ok */
683 was_same = (alias &&
684 !ce_stage(alias) &&
685 !hashcmp(alias->sha1, ce->sha1) &&
686 ce->ce_mode == alias->ce_mode);
687
688 if (pretend)
689 ;
690 else if (add_index_entry(istate, ce, add_option))
691 return error("unable to add %s to index",path);
692 if (verbose && !was_same)
693 printf("add '%s'\n", path);
694 return 0;
695 }
696
697 int add_file_to_index(struct index_state *istate, const char *path, int flags)
698 {
699 struct stat st;
700 if (lstat(path, &st))
701 die_errno("unable to stat '%s'", path);
702 return add_to_index(istate, path, &st, flags);
703 }
704
705 struct cache_entry *make_cache_entry(unsigned int mode,
706 const unsigned char *sha1, const char *path, int stage,
707 unsigned int refresh_options)
708 {
709 int size, len;
710 struct cache_entry *ce;
711
712 if (!verify_path(path)) {
713 error("Invalid path '%s'", path);
714 return NULL;
715 }
716
717 len = strlen(path);
718 size = cache_entry_size(len);
719 ce = xcalloc(1, size);
720
721 hashcpy(ce->sha1, sha1);
722 memcpy(ce->name, path, len);
723 ce->ce_flags = create_ce_flags(stage);
724 ce->ce_namelen = len;
725 ce->ce_mode = create_ce_mode(mode);
726
727 return refresh_cache_entry(ce, refresh_options);
728 }
729
730 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
731 {
732 int len = ce_namelen(a);
733 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
734 }
735
736 /*
737 * We fundamentally don't like some paths: we don't want
738 * dot or dot-dot anywhere, and for obvious reasons don't
739 * want to recurse into ".git" either.
740 *
741 * Also, we don't want double slashes or slashes at the
742 * end that can make pathnames ambiguous.
743 */
744 static int verify_dotfile(const char *rest)
745 {
746 /*
747 * The first character was '.', but that
748 * has already been discarded, we now test
749 * the rest.
750 */
751
752 /* "." is not allowed */
753 if (*rest == '\0' || is_dir_sep(*rest))
754 return 0;
755
756 switch (*rest) {
757 /*
758 * ".git" followed by NUL or slash is bad. This
759 * shares the path end test with the ".." case.
760 */
761 case 'g':
762 if (rest[1] != 'i')
763 break;
764 if (rest[2] != 't')
765 break;
766 rest += 2;
767 /* fallthrough */
768 case '.':
769 if (rest[1] == '\0' || is_dir_sep(rest[1]))
770 return 0;
771 }
772 return 1;
773 }
774
775 int verify_path(const char *path)
776 {
777 char c;
778
779 if (has_dos_drive_prefix(path))
780 return 0;
781
782 goto inside;
783 for (;;) {
784 if (!c)
785 return 1;
786 if (is_dir_sep(c)) {
787 inside:
788 c = *path++;
789 if ((c == '.' && !verify_dotfile(path)) ||
790 is_dir_sep(c) || c == '\0')
791 return 0;
792 }
793 c = *path++;
794 }
795 }
796
797 /*
798 * Do we have another file that has the beginning components being a
799 * proper superset of the name we're trying to add?
800 */
801 static int has_file_name(struct index_state *istate,
802 const struct cache_entry *ce, int pos, int ok_to_replace)
803 {
804 int retval = 0;
805 int len = ce_namelen(ce);
806 int stage = ce_stage(ce);
807 const char *name = ce->name;
808
809 while (pos < istate->cache_nr) {
810 struct cache_entry *p = istate->cache[pos++];
811
812 if (len >= ce_namelen(p))
813 break;
814 if (memcmp(name, p->name, len))
815 break;
816 if (ce_stage(p) != stage)
817 continue;
818 if (p->name[len] != '/')
819 continue;
820 if (p->ce_flags & CE_REMOVE)
821 continue;
822 retval = -1;
823 if (!ok_to_replace)
824 break;
825 remove_index_entry_at(istate, --pos);
826 }
827 return retval;
828 }
829
830 /*
831 * Do we have another file with a pathname that is a proper
832 * subset of the name we're trying to add?
833 */
834 static int has_dir_name(struct index_state *istate,
835 const struct cache_entry *ce, int pos, int ok_to_replace)
836 {
837 int retval = 0;
838 int stage = ce_stage(ce);
839 const char *name = ce->name;
840 const char *slash = name + ce_namelen(ce);
841
842 for (;;) {
843 int len;
844
845 for (;;) {
846 if (*--slash == '/')
847 break;
848 if (slash <= ce->name)
849 return retval;
850 }
851 len = slash - name;
852
853 pos = index_name_stage_pos(istate, name, len, stage);
854 if (pos >= 0) {
855 /*
856 * Found one, but not so fast. This could
857 * be a marker that says "I was here, but
858 * I am being removed". Such an entry is
859 * not a part of the resulting tree, and
860 * it is Ok to have a directory at the same
861 * path.
862 */
863 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
864 retval = -1;
865 if (!ok_to_replace)
866 break;
867 remove_index_entry_at(istate, pos);
868 continue;
869 }
870 }
871 else
872 pos = -pos-1;
873
874 /*
875 * Trivial optimization: if we find an entry that
876 * already matches the sub-directory, then we know
877 * we're ok, and we can exit.
878 */
879 while (pos < istate->cache_nr) {
880 struct cache_entry *p = istate->cache[pos];
881 if ((ce_namelen(p) <= len) ||
882 (p->name[len] != '/') ||
883 memcmp(p->name, name, len))
884 break; /* not our subdirectory */
885 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
886 /*
887 * p is at the same stage as our entry, and
888 * is a subdirectory of what we are looking
889 * at, so we cannot have conflicts at our
890 * level or anything shorter.
891 */
892 return retval;
893 pos++;
894 }
895 }
896 return retval;
897 }
898
899 /* We may be in a situation where we already have path/file and path
900 * is being added, or we already have path and path/file is being
901 * added. Either one would result in a nonsense tree that has path
902 * twice when git-write-tree tries to write it out. Prevent it.
903 *
904 * If ok-to-replace is specified, we remove the conflicting entries
905 * from the cache so the caller should recompute the insert position.
906 * When this happens, we return non-zero.
907 */
908 static int check_file_directory_conflict(struct index_state *istate,
909 const struct cache_entry *ce,
910 int pos, int ok_to_replace)
911 {
912 int retval;
913
914 /*
915 * When ce is an "I am going away" entry, we allow it to be added
916 */
917 if (ce->ce_flags & CE_REMOVE)
918 return 0;
919
920 /*
921 * We check if the path is a sub-path of a subsequent pathname
922 * first, since removing those will not change the position
923 * in the array.
924 */
925 retval = has_file_name(istate, ce, pos, ok_to_replace);
926
927 /*
928 * Then check if the path might have a clashing sub-directory
929 * before it.
930 */
931 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
932 }
933
934 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
935 {
936 int pos;
937 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
938 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
939 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
940 int new_only = option & ADD_CACHE_NEW_ONLY;
941
942 cache_tree_invalidate_path(istate, ce->name);
943 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
944
945 /* existing match? Just replace it. */
946 if (pos >= 0) {
947 if (!new_only)
948 replace_index_entry(istate, pos, ce);
949 return 0;
950 }
951 pos = -pos-1;
952
953 /*
954 * Inserting a merged entry ("stage 0") into the index
955 * will always replace all non-merged entries..
956 */
957 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
958 while (ce_same_name(istate->cache[pos], ce)) {
959 ok_to_add = 1;
960 if (!remove_index_entry_at(istate, pos))
961 break;
962 }
963 }
964
965 if (!ok_to_add)
966 return -1;
967 if (!verify_path(ce->name))
968 return error("Invalid path '%s'", ce->name);
969
970 if (!skip_df_check &&
971 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
972 if (!ok_to_replace)
973 return error("'%s' appears as both a file and as a directory",
974 ce->name);
975 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
976 pos = -pos-1;
977 }
978 return pos + 1;
979 }
980
981 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
982 {
983 int pos;
984
985 if (option & ADD_CACHE_JUST_APPEND)
986 pos = istate->cache_nr;
987 else {
988 int ret;
989 ret = add_index_entry_with_check(istate, ce, option);
990 if (ret <= 0)
991 return ret;
992 pos = ret - 1;
993 }
994
995 /* Make sure the array is big enough .. */
996 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
997
998 /* Add it in.. */
999 istate->cache_nr++;
1000 if (istate->cache_nr > pos + 1)
1001 memmove(istate->cache + pos + 1,
1002 istate->cache + pos,
1003 (istate->cache_nr - pos - 1) * sizeof(ce));
1004 set_index_entry(istate, pos, ce);
1005 istate->cache_changed |= CE_ENTRY_ADDED;
1006 return 0;
1007 }
1008
1009 /*
1010 * "refresh" does not calculate a new sha1 file or bring the
1011 * cache up-to-date for mode/content changes. But what it
1012 * _does_ do is to "re-match" the stat information of a file
1013 * with the cache, so that you can refresh the cache for a
1014 * file that hasn't been changed but where the stat entry is
1015 * out of date.
1016 *
1017 * For example, you'd want to do this after doing a "git-read-tree",
1018 * to link up the stat cache details with the proper files.
1019 */
1020 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1021 struct cache_entry *ce,
1022 unsigned int options, int *err,
1023 int *changed_ret)
1024 {
1025 struct stat st;
1026 struct cache_entry *updated;
1027 int changed, size;
1028 int refresh = options & CE_MATCH_REFRESH;
1029 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1030 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1031 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1032
1033 if (!refresh || ce_uptodate(ce))
1034 return ce;
1035
1036 /*
1037 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1038 * that the change to the work tree does not matter and told
1039 * us not to worry.
1040 */
1041 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1042 ce_mark_uptodate(ce);
1043 return ce;
1044 }
1045 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1046 ce_mark_uptodate(ce);
1047 return ce;
1048 }
1049
1050 if (lstat(ce->name, &st) < 0) {
1051 if (ignore_missing && errno == ENOENT)
1052 return ce;
1053 if (err)
1054 *err = errno;
1055 return NULL;
1056 }
1057
1058 changed = ie_match_stat(istate, ce, &st, options);
1059 if (changed_ret)
1060 *changed_ret = changed;
1061 if (!changed) {
1062 /*
1063 * The path is unchanged. If we were told to ignore
1064 * valid bit, then we did the actual stat check and
1065 * found that the entry is unmodified. If the entry
1066 * is not marked VALID, this is the place to mark it
1067 * valid again, under "assume unchanged" mode.
1068 */
1069 if (ignore_valid && assume_unchanged &&
1070 !(ce->ce_flags & CE_VALID))
1071 ; /* mark this one VALID again */
1072 else {
1073 /*
1074 * We do not mark the index itself "modified"
1075 * because CE_UPTODATE flag is in-core only;
1076 * we are not going to write this change out.
1077 */
1078 if (!S_ISGITLINK(ce->ce_mode))
1079 ce_mark_uptodate(ce);
1080 return ce;
1081 }
1082 }
1083
1084 if (ie_modified(istate, ce, &st, options)) {
1085 if (err)
1086 *err = EINVAL;
1087 return NULL;
1088 }
1089
1090 size = ce_size(ce);
1091 updated = xmalloc(size);
1092 memcpy(updated, ce, size);
1093 fill_stat_cache_info(updated, &st);
1094 /*
1095 * If ignore_valid is not set, we should leave CE_VALID bit
1096 * alone. Otherwise, paths marked with --no-assume-unchanged
1097 * (i.e. things to be edited) will reacquire CE_VALID bit
1098 * automatically, which is not really what we want.
1099 */
1100 if (!ignore_valid && assume_unchanged &&
1101 !(ce->ce_flags & CE_VALID))
1102 updated->ce_flags &= ~CE_VALID;
1103
1104 /* istate->cache_changed is updated in the caller */
1105 return updated;
1106 }
1107
1108 static void show_file(const char * fmt, const char * name, int in_porcelain,
1109 int * first, const char *header_msg)
1110 {
1111 if (in_porcelain && *first && header_msg) {
1112 printf("%s\n", header_msg);
1113 *first = 0;
1114 }
1115 printf(fmt, name);
1116 }
1117
1118 int refresh_index(struct index_state *istate, unsigned int flags,
1119 const struct pathspec *pathspec,
1120 char *seen, const char *header_msg)
1121 {
1122 int i;
1123 int has_errors = 0;
1124 int really = (flags & REFRESH_REALLY) != 0;
1125 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1126 int quiet = (flags & REFRESH_QUIET) != 0;
1127 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1128 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1129 int first = 1;
1130 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1131 unsigned int options = (CE_MATCH_REFRESH |
1132 (really ? CE_MATCH_IGNORE_VALID : 0) |
1133 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1134 const char *modified_fmt;
1135 const char *deleted_fmt;
1136 const char *typechange_fmt;
1137 const char *added_fmt;
1138 const char *unmerged_fmt;
1139
1140 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1141 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1142 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1143 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1144 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1145 for (i = 0; i < istate->cache_nr; i++) {
1146 struct cache_entry *ce, *new;
1147 int cache_errno = 0;
1148 int changed = 0;
1149 int filtered = 0;
1150
1151 ce = istate->cache[i];
1152 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1153 continue;
1154
1155 if (pathspec && !ce_path_match(ce, pathspec, seen))
1156 filtered = 1;
1157
1158 if (ce_stage(ce)) {
1159 while ((i < istate->cache_nr) &&
1160 ! strcmp(istate->cache[i]->name, ce->name))
1161 i++;
1162 i--;
1163 if (allow_unmerged)
1164 continue;
1165 if (!filtered)
1166 show_file(unmerged_fmt, ce->name, in_porcelain,
1167 &first, header_msg);
1168 has_errors = 1;
1169 continue;
1170 }
1171
1172 if (filtered)
1173 continue;
1174
1175 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1176 if (new == ce)
1177 continue;
1178 if (!new) {
1179 const char *fmt;
1180
1181 if (really && cache_errno == EINVAL) {
1182 /* If we are doing --really-refresh that
1183 * means the index is not valid anymore.
1184 */
1185 ce->ce_flags &= ~CE_VALID;
1186 istate->cache_changed |= CE_ENTRY_CHANGED;
1187 }
1188 if (quiet)
1189 continue;
1190
1191 if (cache_errno == ENOENT)
1192 fmt = deleted_fmt;
1193 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1194 fmt = added_fmt; /* must be before other checks */
1195 else if (changed & TYPE_CHANGED)
1196 fmt = typechange_fmt;
1197 else
1198 fmt = modified_fmt;
1199 show_file(fmt,
1200 ce->name, in_porcelain, &first, header_msg);
1201 has_errors = 1;
1202 continue;
1203 }
1204
1205 replace_index_entry(istate, i, new);
1206 }
1207 return has_errors;
1208 }
1209
1210 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1211 unsigned int options)
1212 {
1213 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1214 }
1215
1216
1217 /*****************************************************************
1218 * Index File I/O
1219 *****************************************************************/
1220
1221 #define INDEX_FORMAT_DEFAULT 3
1222
1223 static int index_format_config(const char *var, const char *value, void *cb)
1224 {
1225 unsigned int *version = cb;
1226 if (!strcmp(var, "index.version")) {
1227 *version = git_config_int(var, value);
1228 return 0;
1229 }
1230 return 1;
1231 }
1232
1233 static unsigned int get_index_format_default(void)
1234 {
1235 char *envversion = getenv("GIT_INDEX_VERSION");
1236 char *endp;
1237 unsigned int version = INDEX_FORMAT_DEFAULT;
1238
1239 if (!envversion) {
1240 git_config(index_format_config, &version);
1241 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1242 warning(_("index.version set, but the value is invalid.\n"
1243 "Using version %i"), INDEX_FORMAT_DEFAULT);
1244 return INDEX_FORMAT_DEFAULT;
1245 }
1246 return version;
1247 }
1248
1249 version = strtoul(envversion, &endp, 10);
1250 if (*endp ||
1251 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1252 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1253 "Using version %i"), INDEX_FORMAT_DEFAULT);
1254 version = INDEX_FORMAT_DEFAULT;
1255 }
1256 return version;
1257 }
1258
1259 /*
1260 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1261 * Again - this is just a (very strong in practice) heuristic that
1262 * the inode hasn't changed.
1263 *
1264 * We save the fields in big-endian order to allow using the
1265 * index file over NFS transparently.
1266 */
1267 struct ondisk_cache_entry {
1268 struct cache_time ctime;
1269 struct cache_time mtime;
1270 uint32_t dev;
1271 uint32_t ino;
1272 uint32_t mode;
1273 uint32_t uid;
1274 uint32_t gid;
1275 uint32_t size;
1276 unsigned char sha1[20];
1277 uint16_t flags;
1278 char name[FLEX_ARRAY]; /* more */
1279 };
1280
1281 /*
1282 * This struct is used when CE_EXTENDED bit is 1
1283 * The struct must match ondisk_cache_entry exactly from
1284 * ctime till flags
1285 */
1286 struct ondisk_cache_entry_extended {
1287 struct cache_time ctime;
1288 struct cache_time mtime;
1289 uint32_t dev;
1290 uint32_t ino;
1291 uint32_t mode;
1292 uint32_t uid;
1293 uint32_t gid;
1294 uint32_t size;
1295 unsigned char sha1[20];
1296 uint16_t flags;
1297 uint16_t flags2;
1298 char name[FLEX_ARRAY]; /* more */
1299 };
1300
1301 /* These are only used for v3 or lower */
1302 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1303 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1304 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1305 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1306 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1307 ondisk_cache_entry_size(ce_namelen(ce)))
1308
1309 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1310 {
1311 git_SHA_CTX c;
1312 unsigned char sha1[20];
1313 int hdr_version;
1314
1315 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1316 return error("bad signature");
1317 hdr_version = ntohl(hdr->hdr_version);
1318 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1319 return error("bad index version %d", hdr_version);
1320 git_SHA1_Init(&c);
1321 git_SHA1_Update(&c, hdr, size - 20);
1322 git_SHA1_Final(sha1, &c);
1323 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1324 return error("bad index file sha1 signature");
1325 return 0;
1326 }
1327
1328 static int read_index_extension(struct index_state *istate,
1329 const char *ext, void *data, unsigned long sz)
1330 {
1331 switch (CACHE_EXT(ext)) {
1332 case CACHE_EXT_TREE:
1333 istate->cache_tree = cache_tree_read(data, sz);
1334 break;
1335 case CACHE_EXT_RESOLVE_UNDO:
1336 istate->resolve_undo = resolve_undo_read(data, sz);
1337 break;
1338 default:
1339 if (*ext < 'A' || 'Z' < *ext)
1340 return error("index uses %.4s extension, which we do not understand",
1341 ext);
1342 fprintf(stderr, "ignoring %.4s extension\n", ext);
1343 break;
1344 }
1345 return 0;
1346 }
1347
1348 int read_index(struct index_state *istate)
1349 {
1350 return read_index_from(istate, get_index_file());
1351 }
1352
1353 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1354 unsigned int flags,
1355 const char *name,
1356 size_t len)
1357 {
1358 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1359
1360 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1361 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1362 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1363 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1364 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1365 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1366 ce->ce_mode = get_be32(&ondisk->mode);
1367 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1368 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1369 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1370 ce->ce_flags = flags & ~CE_NAMEMASK;
1371 ce->ce_namelen = len;
1372 hashcpy(ce->sha1, ondisk->sha1);
1373 memcpy(ce->name, name, len);
1374 ce->name[len] = '\0';
1375 return ce;
1376 }
1377
1378 /*
1379 * Adjacent cache entries tend to share the leading paths, so it makes
1380 * sense to only store the differences in later entries. In the v4
1381 * on-disk format of the index, each on-disk cache entry stores the
1382 * number of bytes to be stripped from the end of the previous name,
1383 * and the bytes to append to the result, to come up with its name.
1384 */
1385 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1386 {
1387 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1388 size_t len = decode_varint(&cp);
1389
1390 if (name->len < len)
1391 die("malformed name field in the index");
1392 strbuf_remove(name, name->len - len, len);
1393 for (ep = cp; *ep; ep++)
1394 ; /* find the end */
1395 strbuf_add(name, cp, ep - cp);
1396 return (const char *)ep + 1 - cp_;
1397 }
1398
1399 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1400 unsigned long *ent_size,
1401 struct strbuf *previous_name)
1402 {
1403 struct cache_entry *ce;
1404 size_t len;
1405 const char *name;
1406 unsigned int flags;
1407
1408 /* On-disk flags are just 16 bits */
1409 flags = get_be16(&ondisk->flags);
1410 len = flags & CE_NAMEMASK;
1411
1412 if (flags & CE_EXTENDED) {
1413 struct ondisk_cache_entry_extended *ondisk2;
1414 int extended_flags;
1415 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1416 extended_flags = get_be16(&ondisk2->flags2) << 16;
1417 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1418 if (extended_flags & ~CE_EXTENDED_FLAGS)
1419 die("Unknown index entry format %08x", extended_flags);
1420 flags |= extended_flags;
1421 name = ondisk2->name;
1422 }
1423 else
1424 name = ondisk->name;
1425
1426 if (!previous_name) {
1427 /* v3 and earlier */
1428 if (len == CE_NAMEMASK)
1429 len = strlen(name);
1430 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1431
1432 *ent_size = ondisk_ce_size(ce);
1433 } else {
1434 unsigned long consumed;
1435 consumed = expand_name_field(previous_name, name);
1436 ce = cache_entry_from_ondisk(ondisk, flags,
1437 previous_name->buf,
1438 previous_name->len);
1439
1440 *ent_size = (name - ((char *)ondisk)) + consumed;
1441 }
1442 return ce;
1443 }
1444
1445 /* remember to discard_cache() before reading a different cache! */
1446 int read_index_from(struct index_state *istate, const char *path)
1447 {
1448 int fd, i;
1449 struct stat st;
1450 unsigned long src_offset;
1451 struct cache_header *hdr;
1452 void *mmap;
1453 size_t mmap_size;
1454 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1455
1456 if (istate->initialized)
1457 return istate->cache_nr;
1458
1459 istate->timestamp.sec = 0;
1460 istate->timestamp.nsec = 0;
1461 fd = open(path, O_RDONLY);
1462 if (fd < 0) {
1463 if (errno == ENOENT)
1464 return 0;
1465 die_errno("index file open failed");
1466 }
1467
1468 if (fstat(fd, &st))
1469 die_errno("cannot stat the open index");
1470
1471 mmap_size = xsize_t(st.st_size);
1472 if (mmap_size < sizeof(struct cache_header) + 20)
1473 die("index file smaller than expected");
1474
1475 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1476 if (mmap == MAP_FAILED)
1477 die_errno("unable to map index file");
1478 close(fd);
1479
1480 hdr = mmap;
1481 if (verify_hdr(hdr, mmap_size) < 0)
1482 goto unmap;
1483
1484 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1485 istate->version = ntohl(hdr->hdr_version);
1486 istate->cache_nr = ntohl(hdr->hdr_entries);
1487 istate->cache_alloc = alloc_nr(istate->cache_nr);
1488 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1489 istate->initialized = 1;
1490
1491 if (istate->version == 4)
1492 previous_name = &previous_name_buf;
1493 else
1494 previous_name = NULL;
1495
1496 src_offset = sizeof(*hdr);
1497 for (i = 0; i < istate->cache_nr; i++) {
1498 struct ondisk_cache_entry *disk_ce;
1499 struct cache_entry *ce;
1500 unsigned long consumed;
1501
1502 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1503 ce = create_from_disk(disk_ce, &consumed, previous_name);
1504 set_index_entry(istate, i, ce);
1505
1506 src_offset += consumed;
1507 }
1508 strbuf_release(&previous_name_buf);
1509 istate->timestamp.sec = st.st_mtime;
1510 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1511
1512 while (src_offset <= mmap_size - 20 - 8) {
1513 /* After an array of active_nr index entries,
1514 * there can be arbitrary number of extended
1515 * sections, each of which is prefixed with
1516 * extension name (4-byte) and section length
1517 * in 4-byte network byte order.
1518 */
1519 uint32_t extsize;
1520 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1521 extsize = ntohl(extsize);
1522 if (read_index_extension(istate,
1523 (const char *) mmap + src_offset,
1524 (char *) mmap + src_offset + 8,
1525 extsize) < 0)
1526 goto unmap;
1527 src_offset += 8;
1528 src_offset += extsize;
1529 }
1530 munmap(mmap, mmap_size);
1531 return istate->cache_nr;
1532
1533 unmap:
1534 munmap(mmap, mmap_size);
1535 die("index file corrupt");
1536 }
1537
1538 int is_index_unborn(struct index_state *istate)
1539 {
1540 return (!istate->cache_nr && !istate->timestamp.sec);
1541 }
1542
1543 int discard_index(struct index_state *istate)
1544 {
1545 int i;
1546
1547 for (i = 0; i < istate->cache_nr; i++)
1548 free(istate->cache[i]);
1549 resolve_undo_clear_index(istate);
1550 istate->cache_nr = 0;
1551 istate->cache_changed = 0;
1552 istate->timestamp.sec = 0;
1553 istate->timestamp.nsec = 0;
1554 free_name_hash(istate);
1555 cache_tree_free(&(istate->cache_tree));
1556 istate->initialized = 0;
1557 free(istate->cache);
1558 istate->cache = NULL;
1559 istate->cache_alloc = 0;
1560 return 0;
1561 }
1562
1563 int unmerged_index(const struct index_state *istate)
1564 {
1565 int i;
1566 for (i = 0; i < istate->cache_nr; i++) {
1567 if (ce_stage(istate->cache[i]))
1568 return 1;
1569 }
1570 return 0;
1571 }
1572
1573 #define WRITE_BUFFER_SIZE 8192
1574 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1575 static unsigned long write_buffer_len;
1576
1577 static int ce_write_flush(git_SHA_CTX *context, int fd)
1578 {
1579 unsigned int buffered = write_buffer_len;
1580 if (buffered) {
1581 git_SHA1_Update(context, write_buffer, buffered);
1582 if (write_in_full(fd, write_buffer, buffered) != buffered)
1583 return -1;
1584 write_buffer_len = 0;
1585 }
1586 return 0;
1587 }
1588
1589 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1590 {
1591 while (len) {
1592 unsigned int buffered = write_buffer_len;
1593 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1594 if (partial > len)
1595 partial = len;
1596 memcpy(write_buffer + buffered, data, partial);
1597 buffered += partial;
1598 if (buffered == WRITE_BUFFER_SIZE) {
1599 write_buffer_len = buffered;
1600 if (ce_write_flush(context, fd))
1601 return -1;
1602 buffered = 0;
1603 }
1604 write_buffer_len = buffered;
1605 len -= partial;
1606 data = (char *) data + partial;
1607 }
1608 return 0;
1609 }
1610
1611 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1612 unsigned int ext, unsigned int sz)
1613 {
1614 ext = htonl(ext);
1615 sz = htonl(sz);
1616 return ((ce_write(context, fd, &ext, 4) < 0) ||
1617 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1618 }
1619
1620 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1621 {
1622 unsigned int left = write_buffer_len;
1623
1624 if (left) {
1625 write_buffer_len = 0;
1626 git_SHA1_Update(context, write_buffer, left);
1627 }
1628
1629 /* Flush first if not enough space for SHA1 signature */
1630 if (left + 20 > WRITE_BUFFER_SIZE) {
1631 if (write_in_full(fd, write_buffer, left) != left)
1632 return -1;
1633 left = 0;
1634 }
1635
1636 /* Append the SHA1 signature at the end */
1637 git_SHA1_Final(write_buffer + left, context);
1638 hashcpy(sha1, write_buffer + left);
1639 left += 20;
1640 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1641 }
1642
1643 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1644 {
1645 /*
1646 * The only thing we care about in this function is to smudge the
1647 * falsely clean entry due to touch-update-touch race, so we leave
1648 * everything else as they are. We are called for entries whose
1649 * ce_stat_data.sd_mtime match the index file mtime.
1650 *
1651 * Note that this actually does not do much for gitlinks, for
1652 * which ce_match_stat_basic() always goes to the actual
1653 * contents. The caller checks with is_racy_timestamp() which
1654 * always says "no" for gitlinks, so we are not called for them ;-)
1655 */
1656 struct stat st;
1657
1658 if (lstat(ce->name, &st) < 0)
1659 return;
1660 if (ce_match_stat_basic(ce, &st))
1661 return;
1662 if (ce_modified_check_fs(ce, &st)) {
1663 /* This is "racily clean"; smudge it. Note that this
1664 * is a tricky code. At first glance, it may appear
1665 * that it can break with this sequence:
1666 *
1667 * $ echo xyzzy >frotz
1668 * $ git-update-index --add frotz
1669 * $ : >frotz
1670 * $ sleep 3
1671 * $ echo filfre >nitfol
1672 * $ git-update-index --add nitfol
1673 *
1674 * but it does not. When the second update-index runs,
1675 * it notices that the entry "frotz" has the same timestamp
1676 * as index, and if we were to smudge it by resetting its
1677 * size to zero here, then the object name recorded
1678 * in index is the 6-byte file but the cached stat information
1679 * becomes zero --- which would then match what we would
1680 * obtain from the filesystem next time we stat("frotz").
1681 *
1682 * However, the second update-index, before calling
1683 * this function, notices that the cached size is 6
1684 * bytes and what is on the filesystem is an empty
1685 * file, and never calls us, so the cached size information
1686 * for "frotz" stays 6 which does not match the filesystem.
1687 */
1688 ce->ce_stat_data.sd_size = 0;
1689 }
1690 }
1691
1692 /* Copy miscellaneous fields but not the name */
1693 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1694 struct cache_entry *ce)
1695 {
1696 short flags;
1697
1698 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1699 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1700 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1701 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1702 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1703 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1704 ondisk->mode = htonl(ce->ce_mode);
1705 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1706 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1707 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1708 hashcpy(ondisk->sha1, ce->sha1);
1709
1710 flags = ce->ce_flags & ~CE_NAMEMASK;
1711 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1712 ondisk->flags = htons(flags);
1713 if (ce->ce_flags & CE_EXTENDED) {
1714 struct ondisk_cache_entry_extended *ondisk2;
1715 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1716 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1717 return ondisk2->name;
1718 }
1719 else {
1720 return ondisk->name;
1721 }
1722 }
1723
1724 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1725 struct strbuf *previous_name)
1726 {
1727 int size;
1728 struct ondisk_cache_entry *ondisk;
1729 char *name;
1730 int result;
1731
1732 if (!previous_name) {
1733 size = ondisk_ce_size(ce);
1734 ondisk = xcalloc(1, size);
1735 name = copy_cache_entry_to_ondisk(ondisk, ce);
1736 memcpy(name, ce->name, ce_namelen(ce));
1737 } else {
1738 int common, to_remove, prefix_size;
1739 unsigned char to_remove_vi[16];
1740 for (common = 0;
1741 (ce->name[common] &&
1742 common < previous_name->len &&
1743 ce->name[common] == previous_name->buf[common]);
1744 common++)
1745 ; /* still matching */
1746 to_remove = previous_name->len - common;
1747 prefix_size = encode_varint(to_remove, to_remove_vi);
1748
1749 if (ce->ce_flags & CE_EXTENDED)
1750 size = offsetof(struct ondisk_cache_entry_extended, name);
1751 else
1752 size = offsetof(struct ondisk_cache_entry, name);
1753 size += prefix_size + (ce_namelen(ce) - common + 1);
1754
1755 ondisk = xcalloc(1, size);
1756 name = copy_cache_entry_to_ondisk(ondisk, ce);
1757 memcpy(name, to_remove_vi, prefix_size);
1758 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1759
1760 strbuf_splice(previous_name, common, to_remove,
1761 ce->name + common, ce_namelen(ce) - common);
1762 }
1763
1764 result = ce_write(c, fd, ondisk, size);
1765 free(ondisk);
1766 return result;
1767 }
1768
1769 static int has_racy_timestamp(struct index_state *istate)
1770 {
1771 int entries = istate->cache_nr;
1772 int i;
1773
1774 for (i = 0; i < entries; i++) {
1775 struct cache_entry *ce = istate->cache[i];
1776 if (is_racy_timestamp(istate, ce))
1777 return 1;
1778 }
1779 return 0;
1780 }
1781
1782 /*
1783 * Opportunistically update the index but do not complain if we can't
1784 */
1785 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1786 {
1787 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1788 write_locked_index(istate, lockfile, COMMIT_LOCK))
1789 rollback_lock_file(lockfile);
1790 }
1791
1792 static int do_write_index(struct index_state *istate, int newfd)
1793 {
1794 git_SHA_CTX c;
1795 struct cache_header hdr;
1796 int i, err, removed, extended, hdr_version;
1797 struct cache_entry **cache = istate->cache;
1798 int entries = istate->cache_nr;
1799 struct stat st;
1800 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1801
1802 for (i = removed = extended = 0; i < entries; i++) {
1803 if (cache[i]->ce_flags & CE_REMOVE)
1804 removed++;
1805
1806 /* reduce extended entries if possible */
1807 cache[i]->ce_flags &= ~CE_EXTENDED;
1808 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1809 extended++;
1810 cache[i]->ce_flags |= CE_EXTENDED;
1811 }
1812 }
1813
1814 if (!istate->version)
1815 istate->version = get_index_format_default();
1816
1817 /* demote version 3 to version 2 when the latter suffices */
1818 if (istate->version == 3 || istate->version == 2)
1819 istate->version = extended ? 3 : 2;
1820
1821 hdr_version = istate->version;
1822
1823 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1824 hdr.hdr_version = htonl(hdr_version);
1825 hdr.hdr_entries = htonl(entries - removed);
1826
1827 git_SHA1_Init(&c);
1828 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1829 return -1;
1830
1831 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1832 for (i = 0; i < entries; i++) {
1833 struct cache_entry *ce = cache[i];
1834 if (ce->ce_flags & CE_REMOVE)
1835 continue;
1836 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1837 ce_smudge_racily_clean_entry(ce);
1838 if (is_null_sha1(ce->sha1)) {
1839 static const char msg[] = "cache entry has null sha1: %s";
1840 static int allow = -1;
1841
1842 if (allow < 0)
1843 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1844 if (allow)
1845 warning(msg, ce->name);
1846 else
1847 return error(msg, ce->name);
1848 }
1849 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1850 return -1;
1851 }
1852 strbuf_release(&previous_name_buf);
1853
1854 /* Write extension data here */
1855 if (istate->cache_tree) {
1856 struct strbuf sb = STRBUF_INIT;
1857
1858 cache_tree_write(&sb, istate->cache_tree);
1859 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1860 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1861 strbuf_release(&sb);
1862 if (err)
1863 return -1;
1864 }
1865 if (istate->resolve_undo) {
1866 struct strbuf sb = STRBUF_INIT;
1867
1868 resolve_undo_write(&sb, istate->resolve_undo);
1869 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1870 sb.len) < 0
1871 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1872 strbuf_release(&sb);
1873 if (err)
1874 return -1;
1875 }
1876
1877 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
1878 return -1;
1879 istate->timestamp.sec = (unsigned int)st.st_mtime;
1880 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1881 return 0;
1882 }
1883
1884 void set_alternate_index_output(const char *name)
1885 {
1886 alternate_index_output = name;
1887 }
1888
1889 static int commit_locked_index(struct lock_file *lk)
1890 {
1891 if (alternate_index_output) {
1892 if (lk->fd >= 0 && close_lock_file(lk))
1893 return -1;
1894 if (rename(lk->filename, alternate_index_output))
1895 return -1;
1896 lk->filename[0] = 0;
1897 return 0;
1898 } else {
1899 return commit_lock_file(lk);
1900 }
1901 }
1902
1903 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
1904 unsigned flags)
1905 {
1906 int ret = do_write_index(istate, lock->fd);
1907 if (ret)
1908 return ret;
1909 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
1910 (COMMIT_LOCK | CLOSE_LOCK));
1911 if (flags & COMMIT_LOCK)
1912 return commit_locked_index(lock);
1913 else if (flags & CLOSE_LOCK)
1914 return close_lock_file(lock);
1915 else
1916 return ret;
1917 }
1918
1919 int write_locked_index(struct index_state *istate, struct lock_file *lock,
1920 unsigned flags)
1921 {
1922 return do_write_locked_index(istate, lock, flags);
1923 }
1924
1925 /*
1926 * Read the index file that is potentially unmerged into given
1927 * index_state, dropping any unmerged entries. Returns true if
1928 * the index is unmerged. Callers who want to refuse to work
1929 * from an unmerged state can call this and check its return value,
1930 * instead of calling read_cache().
1931 */
1932 int read_index_unmerged(struct index_state *istate)
1933 {
1934 int i;
1935 int unmerged = 0;
1936
1937 read_index(istate);
1938 for (i = 0; i < istate->cache_nr; i++) {
1939 struct cache_entry *ce = istate->cache[i];
1940 struct cache_entry *new_ce;
1941 int size, len;
1942
1943 if (!ce_stage(ce))
1944 continue;
1945 unmerged = 1;
1946 len = ce_namelen(ce);
1947 size = cache_entry_size(len);
1948 new_ce = xcalloc(1, size);
1949 memcpy(new_ce->name, ce->name, len);
1950 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1951 new_ce->ce_namelen = len;
1952 new_ce->ce_mode = ce->ce_mode;
1953 if (add_index_entry(istate, new_ce, 0))
1954 return error("%s: cannot drop to stage #0",
1955 new_ce->name);
1956 i = index_name_pos(istate, new_ce->name, len);
1957 }
1958 return unmerged;
1959 }
1960
1961 /*
1962 * Returns 1 if the path is an "other" path with respect to
1963 * the index; that is, the path is not mentioned in the index at all,
1964 * either as a file, a directory with some files in the index,
1965 * or as an unmerged entry.
1966 *
1967 * We helpfully remove a trailing "/" from directories so that
1968 * the output of read_directory can be used as-is.
1969 */
1970 int index_name_is_other(const struct index_state *istate, const char *name,
1971 int namelen)
1972 {
1973 int pos;
1974 if (namelen && name[namelen - 1] == '/')
1975 namelen--;
1976 pos = index_name_pos(istate, name, namelen);
1977 if (0 <= pos)
1978 return 0; /* exact match */
1979 pos = -pos - 1;
1980 if (pos < istate->cache_nr) {
1981 struct cache_entry *ce = istate->cache[pos];
1982 if (ce_namelen(ce) == namelen &&
1983 !memcmp(ce->name, name, namelen))
1984 return 0; /* Yup, this one exists unmerged */
1985 }
1986 return 1;
1987 }
1988
1989 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
1990 {
1991 int pos, len;
1992 unsigned long sz;
1993 enum object_type type;
1994 void *data;
1995
1996 len = strlen(path);
1997 pos = index_name_pos(istate, path, len);
1998 if (pos < 0) {
1999 /*
2000 * We might be in the middle of a merge, in which
2001 * case we would read stage #2 (ours).
2002 */
2003 int i;
2004 for (i = -pos - 1;
2005 (pos < 0 && i < istate->cache_nr &&
2006 !strcmp(istate->cache[i]->name, path));
2007 i++)
2008 if (ce_stage(istate->cache[i]) == 2)
2009 pos = i;
2010 }
2011 if (pos < 0)
2012 return NULL;
2013 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2014 if (!data || type != OBJ_BLOB) {
2015 free(data);
2016 return NULL;
2017 }
2018 if (size)
2019 *size = sz;
2020 return data;
2021 }
2022
2023 void stat_validity_clear(struct stat_validity *sv)
2024 {
2025 free(sv->sd);
2026 sv->sd = NULL;
2027 }
2028
2029 int stat_validity_check(struct stat_validity *sv, const char *path)
2030 {
2031 struct stat st;
2032
2033 if (stat(path, &st) < 0)
2034 return sv->sd == NULL;
2035 if (!sv->sd)
2036 return 0;
2037 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2038 }
2039
2040 void stat_validity_update(struct stat_validity *sv, int fd)
2041 {
2042 struct stat st;
2043
2044 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2045 stat_validity_clear(sv);
2046 else {
2047 if (!sv->sd)
2048 sv->sd = xcalloc(1, sizeof(struct stat_data));
2049 fill_stat_data(sv->sd, &st);
2050 }
2051 }