read-cache: mark new entries for split index
[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 #include "split-index.h"
18
19 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
20 unsigned int options);
21
22 /* Mask for the name length in ce_flags in the on-disk index */
23
24 #define CE_NAMEMASK (0x0fff)
25
26 /* Index extensions.
27 *
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
33 */
34
35 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
37 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
39
40 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
41 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
42 CE_ENTRY_ADDED)
43
44 struct index_state the_index;
45 static const char *alternate_index_output;
46
47 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
48 {
49 istate->cache[nr] = ce;
50 add_name_hash(istate, ce);
51 }
52
53 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
54 {
55 struct cache_entry *old = istate->cache[nr];
56
57 remove_name_hash(istate, old);
58 free(old);
59 set_index_entry(istate, nr, ce);
60 istate->cache_changed |= CE_ENTRY_CHANGED;
61 }
62
63 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
64 {
65 struct cache_entry *old = istate->cache[nr], *new;
66 int namelen = strlen(new_name);
67
68 new = xmalloc(cache_entry_size(namelen));
69 copy_cache_entry(new, old);
70 new->ce_flags &= ~CE_HASHED;
71 new->ce_namelen = namelen;
72 new->index = 0;
73 memcpy(new->name, new_name, namelen + 1);
74
75 cache_tree_invalidate_path(istate, old->name);
76 remove_index_entry_at(istate, nr);
77 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
78 }
79
80 void fill_stat_data(struct stat_data *sd, struct stat *st)
81 {
82 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
83 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
84 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
85 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
86 sd->sd_dev = st->st_dev;
87 sd->sd_ino = st->st_ino;
88 sd->sd_uid = st->st_uid;
89 sd->sd_gid = st->st_gid;
90 sd->sd_size = st->st_size;
91 }
92
93 int match_stat_data(const struct stat_data *sd, struct stat *st)
94 {
95 int changed = 0;
96
97 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
98 changed |= MTIME_CHANGED;
99 if (trust_ctime && check_stat &&
100 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
101 changed |= CTIME_CHANGED;
102
103 #ifdef USE_NSEC
104 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
105 changed |= MTIME_CHANGED;
106 if (trust_ctime && check_stat &&
107 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
108 changed |= CTIME_CHANGED;
109 #endif
110
111 if (check_stat) {
112 if (sd->sd_uid != (unsigned int) st->st_uid ||
113 sd->sd_gid != (unsigned int) st->st_gid)
114 changed |= OWNER_CHANGED;
115 if (sd->sd_ino != (unsigned int) st->st_ino)
116 changed |= INODE_CHANGED;
117 }
118
119 #ifdef USE_STDEV
120 /*
121 * st_dev breaks on network filesystems where different
122 * clients will have different views of what "device"
123 * the filesystem is on
124 */
125 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
126 changed |= INODE_CHANGED;
127 #endif
128
129 if (sd->sd_size != (unsigned int) st->st_size)
130 changed |= DATA_CHANGED;
131
132 return changed;
133 }
134
135 /*
136 * This only updates the "non-critical" parts of the directory
137 * cache, ie the parts that aren't tracked by GIT, and only used
138 * to validate the cache.
139 */
140 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
141 {
142 fill_stat_data(&ce->ce_stat_data, st);
143
144 if (assume_unchanged)
145 ce->ce_flags |= CE_VALID;
146
147 if (S_ISREG(st->st_mode))
148 ce_mark_uptodate(ce);
149 }
150
151 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
152 {
153 int match = -1;
154 int fd = open(ce->name, O_RDONLY);
155
156 if (fd >= 0) {
157 unsigned char sha1[20];
158 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
159 match = hashcmp(sha1, ce->sha1);
160 /* index_fd() closed the file descriptor already */
161 }
162 return match;
163 }
164
165 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
166 {
167 int match = -1;
168 void *buffer;
169 unsigned long size;
170 enum object_type type;
171 struct strbuf sb = STRBUF_INIT;
172
173 if (strbuf_readlink(&sb, ce->name, expected_size))
174 return -1;
175
176 buffer = read_sha1_file(ce->sha1, &type, &size);
177 if (buffer) {
178 if (size == sb.len)
179 match = memcmp(buffer, sb.buf, size);
180 free(buffer);
181 }
182 strbuf_release(&sb);
183 return match;
184 }
185
186 static int ce_compare_gitlink(const struct cache_entry *ce)
187 {
188 unsigned char sha1[20];
189
190 /*
191 * We don't actually require that the .git directory
192 * under GITLINK directory be a valid git directory. It
193 * might even be missing (in case nobody populated that
194 * sub-project).
195 *
196 * If so, we consider it always to match.
197 */
198 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
199 return 0;
200 return hashcmp(sha1, ce->sha1);
201 }
202
203 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
204 {
205 switch (st->st_mode & S_IFMT) {
206 case S_IFREG:
207 if (ce_compare_data(ce, st))
208 return DATA_CHANGED;
209 break;
210 case S_IFLNK:
211 if (ce_compare_link(ce, xsize_t(st->st_size)))
212 return DATA_CHANGED;
213 break;
214 case S_IFDIR:
215 if (S_ISGITLINK(ce->ce_mode))
216 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
217 default:
218 return TYPE_CHANGED;
219 }
220 return 0;
221 }
222
223 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
224 {
225 unsigned int changed = 0;
226
227 if (ce->ce_flags & CE_REMOVE)
228 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
229
230 switch (ce->ce_mode & S_IFMT) {
231 case S_IFREG:
232 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
233 /* We consider only the owner x bit to be relevant for
234 * "mode changes"
235 */
236 if (trust_executable_bit &&
237 (0100 & (ce->ce_mode ^ st->st_mode)))
238 changed |= MODE_CHANGED;
239 break;
240 case S_IFLNK:
241 if (!S_ISLNK(st->st_mode) &&
242 (has_symlinks || !S_ISREG(st->st_mode)))
243 changed |= TYPE_CHANGED;
244 break;
245 case S_IFGITLINK:
246 /* We ignore most of the st_xxx fields for gitlinks */
247 if (!S_ISDIR(st->st_mode))
248 changed |= TYPE_CHANGED;
249 else if (ce_compare_gitlink(ce))
250 changed |= DATA_CHANGED;
251 return changed;
252 default:
253 die("internal error: ce_mode is %o", ce->ce_mode);
254 }
255
256 changed |= match_stat_data(&ce->ce_stat_data, st);
257
258 /* Racily smudged entry? */
259 if (!ce->ce_stat_data.sd_size) {
260 if (!is_empty_blob_sha1(ce->sha1))
261 changed |= DATA_CHANGED;
262 }
263
264 return changed;
265 }
266
267 static int is_racy_timestamp(const struct index_state *istate,
268 const struct cache_entry *ce)
269 {
270 return (!S_ISGITLINK(ce->ce_mode) &&
271 istate->timestamp.sec &&
272 #ifdef USE_NSEC
273 /* nanosecond timestamped files can also be racy! */
274 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
275 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
276 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
277 #else
278 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
279 #endif
280 );
281 }
282
283 int ie_match_stat(const struct index_state *istate,
284 const struct cache_entry *ce, struct stat *st,
285 unsigned int options)
286 {
287 unsigned int changed;
288 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
289 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
290 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
291
292 /*
293 * If it's marked as always valid in the index, it's
294 * valid whatever the checked-out copy says.
295 *
296 * skip-worktree has the same effect with higher precedence
297 */
298 if (!ignore_skip_worktree && ce_skip_worktree(ce))
299 return 0;
300 if (!ignore_valid && (ce->ce_flags & CE_VALID))
301 return 0;
302
303 /*
304 * Intent-to-add entries have not been added, so the index entry
305 * by definition never matches what is in the work tree until it
306 * actually gets added.
307 */
308 if (ce->ce_flags & CE_INTENT_TO_ADD)
309 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
310
311 changed = ce_match_stat_basic(ce, st);
312
313 /*
314 * Within 1 second of this sequence:
315 * echo xyzzy >file && git-update-index --add file
316 * running this command:
317 * echo frotz >file
318 * would give a falsely clean cache entry. The mtime and
319 * length match the cache, and other stat fields do not change.
320 *
321 * We could detect this at update-index time (the cache entry
322 * being registered/updated records the same time as "now")
323 * and delay the return from git-update-index, but that would
324 * effectively mean we can make at most one commit per second,
325 * which is not acceptable. Instead, we check cache entries
326 * whose mtime are the same as the index file timestamp more
327 * carefully than others.
328 */
329 if (!changed && is_racy_timestamp(istate, ce)) {
330 if (assume_racy_is_modified)
331 changed |= DATA_CHANGED;
332 else
333 changed |= ce_modified_check_fs(ce, st);
334 }
335
336 return changed;
337 }
338
339 int ie_modified(const struct index_state *istate,
340 const struct cache_entry *ce,
341 struct stat *st, unsigned int options)
342 {
343 int changed, changed_fs;
344
345 changed = ie_match_stat(istate, ce, st, options);
346 if (!changed)
347 return 0;
348 /*
349 * If the mode or type has changed, there's no point in trying
350 * to refresh the entry - it's not going to match
351 */
352 if (changed & (MODE_CHANGED | TYPE_CHANGED))
353 return changed;
354
355 /*
356 * Immediately after read-tree or update-index --cacheinfo,
357 * the length field is zero, as we have never even read the
358 * lstat(2) information once, and we cannot trust DATA_CHANGED
359 * returned by ie_match_stat() which in turn was returned by
360 * ce_match_stat_basic() to signal that the filesize of the
361 * blob changed. We have to actually go to the filesystem to
362 * see if the contents match, and if so, should answer "unchanged".
363 *
364 * The logic does not apply to gitlinks, as ce_match_stat_basic()
365 * already has checked the actual HEAD from the filesystem in the
366 * subproject. If ie_match_stat() already said it is different,
367 * then we know it is.
368 */
369 if ((changed & DATA_CHANGED) &&
370 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
371 return changed;
372
373 changed_fs = ce_modified_check_fs(ce, st);
374 if (changed_fs)
375 return changed | changed_fs;
376 return 0;
377 }
378
379 int base_name_compare(const char *name1, int len1, int mode1,
380 const char *name2, int len2, int mode2)
381 {
382 unsigned char c1, c2;
383 int len = len1 < len2 ? len1 : len2;
384 int cmp;
385
386 cmp = memcmp(name1, name2, len);
387 if (cmp)
388 return cmp;
389 c1 = name1[len];
390 c2 = name2[len];
391 if (!c1 && S_ISDIR(mode1))
392 c1 = '/';
393 if (!c2 && S_ISDIR(mode2))
394 c2 = '/';
395 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
396 }
397
398 /*
399 * df_name_compare() is identical to base_name_compare(), except it
400 * compares conflicting directory/file entries as equal. Note that
401 * while a directory name compares as equal to a regular file, they
402 * then individually compare _differently_ to a filename that has
403 * a dot after the basename (because '\0' < '.' < '/').
404 *
405 * This is used by routines that want to traverse the git namespace
406 * but then handle conflicting entries together when possible.
407 */
408 int df_name_compare(const char *name1, int len1, int mode1,
409 const char *name2, int len2, int mode2)
410 {
411 int len = len1 < len2 ? len1 : len2, cmp;
412 unsigned char c1, c2;
413
414 cmp = memcmp(name1, name2, len);
415 if (cmp)
416 return cmp;
417 /* Directories and files compare equal (same length, same name) */
418 if (len1 == len2)
419 return 0;
420 c1 = name1[len];
421 if (!c1 && S_ISDIR(mode1))
422 c1 = '/';
423 c2 = name2[len];
424 if (!c2 && S_ISDIR(mode2))
425 c2 = '/';
426 if (c1 == '/' && !c2)
427 return 0;
428 if (c2 == '/' && !c1)
429 return 0;
430 return c1 - c2;
431 }
432
433 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
434 {
435 int len = len1 < len2 ? len1 : len2;
436 int cmp;
437
438 cmp = memcmp(name1, name2, len);
439 if (cmp)
440 return cmp;
441 if (len1 < len2)
442 return -1;
443 if (len1 > len2)
444 return 1;
445
446 if (stage1 < stage2)
447 return -1;
448 if (stage1 > stage2)
449 return 1;
450 return 0;
451 }
452
453 int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
454 {
455 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
456 }
457
458 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
459 {
460 int first, last;
461
462 first = 0;
463 last = istate->cache_nr;
464 while (last > first) {
465 int next = (last + first) >> 1;
466 struct cache_entry *ce = istate->cache[next];
467 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
468 if (!cmp)
469 return next;
470 if (cmp < 0) {
471 last = next;
472 continue;
473 }
474 first = next+1;
475 }
476 return -first-1;
477 }
478
479 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
480 {
481 return index_name_stage_pos(istate, name, namelen, 0);
482 }
483
484 /* Remove entry, return true if there are more entries to go.. */
485 int remove_index_entry_at(struct index_state *istate, int pos)
486 {
487 struct cache_entry *ce = istate->cache[pos];
488
489 record_resolve_undo(istate, ce);
490 remove_name_hash(istate, ce);
491 free(ce);
492 istate->cache_changed |= CE_ENTRY_REMOVED;
493 istate->cache_nr--;
494 if (pos >= istate->cache_nr)
495 return 0;
496 memmove(istate->cache + pos,
497 istate->cache + pos + 1,
498 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
499 return 1;
500 }
501
502 /*
503 * Remove all cache entries marked for removal, that is where
504 * CE_REMOVE is set in ce_flags. This is much more effective than
505 * calling remove_index_entry_at() for each entry to be removed.
506 */
507 void remove_marked_cache_entries(struct index_state *istate)
508 {
509 struct cache_entry **ce_array = istate->cache;
510 unsigned int i, j;
511
512 for (i = j = 0; i < istate->cache_nr; i++) {
513 if (ce_array[i]->ce_flags & CE_REMOVE) {
514 remove_name_hash(istate, ce_array[i]);
515 free(ce_array[i]);
516 }
517 else
518 ce_array[j++] = ce_array[i];
519 }
520 if (j == istate->cache_nr)
521 return;
522 istate->cache_changed |= CE_ENTRY_REMOVED;
523 istate->cache_nr = j;
524 }
525
526 int remove_file_from_index(struct index_state *istate, const char *path)
527 {
528 int pos = index_name_pos(istate, path, strlen(path));
529 if (pos < 0)
530 pos = -pos-1;
531 cache_tree_invalidate_path(istate, path);
532 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
533 remove_index_entry_at(istate, pos);
534 return 0;
535 }
536
537 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
538 {
539 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
540 }
541
542 static int index_name_pos_also_unmerged(struct index_state *istate,
543 const char *path, int namelen)
544 {
545 int pos = index_name_pos(istate, path, namelen);
546 struct cache_entry *ce;
547
548 if (pos >= 0)
549 return pos;
550
551 /* maybe unmerged? */
552 pos = -1 - pos;
553 if (pos >= istate->cache_nr ||
554 compare_name((ce = istate->cache[pos]), path, namelen))
555 return -1;
556
557 /* order of preference: stage 2, 1, 3 */
558 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
559 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
560 !compare_name(ce, path, namelen))
561 pos++;
562 return pos;
563 }
564
565 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
566 {
567 int len = ce_namelen(ce);
568 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
569 }
570
571 /*
572 * If we add a filename that aliases in the cache, we will use the
573 * name that we already have - but we don't want to update the same
574 * alias twice, because that implies that there were actually two
575 * different files with aliasing names!
576 *
577 * So we use the CE_ADDED flag to verify that the alias was an old
578 * one before we accept it as
579 */
580 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
581 {
582 int len;
583 struct cache_entry *new;
584
585 if (alias->ce_flags & CE_ADDED)
586 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
587
588 /* Ok, create the new entry using the name of the existing alias */
589 len = ce_namelen(alias);
590 new = xcalloc(1, cache_entry_size(len));
591 memcpy(new->name, alias->name, len);
592 copy_cache_entry(new, ce);
593 free(ce);
594 return new;
595 }
596
597 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
598 {
599 unsigned char sha1[20];
600 if (write_sha1_file("", 0, blob_type, sha1))
601 die("cannot create an empty blob in the object database");
602 hashcpy(ce->sha1, sha1);
603 }
604
605 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
606 {
607 int size, namelen, was_same;
608 mode_t st_mode = st->st_mode;
609 struct cache_entry *ce, *alias;
610 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
611 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
612 int pretend = flags & ADD_CACHE_PRETEND;
613 int intent_only = flags & ADD_CACHE_INTENT;
614 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
615 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
616
617 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
618 return error("%s: can only add regular files, symbolic links or git-directories", path);
619
620 namelen = strlen(path);
621 if (S_ISDIR(st_mode)) {
622 while (namelen && path[namelen-1] == '/')
623 namelen--;
624 }
625 size = cache_entry_size(namelen);
626 ce = xcalloc(1, size);
627 memcpy(ce->name, path, namelen);
628 ce->ce_namelen = namelen;
629 if (!intent_only)
630 fill_stat_cache_info(ce, st);
631 else
632 ce->ce_flags |= CE_INTENT_TO_ADD;
633
634 if (trust_executable_bit && has_symlinks)
635 ce->ce_mode = create_ce_mode(st_mode);
636 else {
637 /* If there is an existing entry, pick the mode bits and type
638 * from it, otherwise assume unexecutable regular file.
639 */
640 struct cache_entry *ent;
641 int pos = index_name_pos_also_unmerged(istate, path, namelen);
642
643 ent = (0 <= pos) ? istate->cache[pos] : NULL;
644 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
645 }
646
647 /* When core.ignorecase=true, determine if a directory of the same name but differing
648 * case already exists within the Git repository. If it does, ensure the directory
649 * case of the file being added to the repository matches (is folded into) the existing
650 * entry's directory case.
651 */
652 if (ignore_case) {
653 const char *startPtr = ce->name;
654 const char *ptr = startPtr;
655 while (*ptr) {
656 while (*ptr && *ptr != '/')
657 ++ptr;
658 if (*ptr == '/') {
659 struct cache_entry *foundce;
660 ++ptr;
661 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
662 if (foundce) {
663 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
664 startPtr = ptr;
665 }
666 }
667 }
668 }
669
670 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
671 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
672 /* Nothing changed, really */
673 free(ce);
674 if (!S_ISGITLINK(alias->ce_mode))
675 ce_mark_uptodate(alias);
676 alias->ce_flags |= CE_ADDED;
677 return 0;
678 }
679 if (!intent_only) {
680 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
681 return error("unable to index file %s", path);
682 } else
683 set_object_name_for_intent_to_add_entry(ce);
684
685 if (ignore_case && alias && different_name(ce, alias))
686 ce = create_alias_ce(ce, alias);
687 ce->ce_flags |= CE_ADDED;
688
689 /* It was suspected to be racily clean, but it turns out to be Ok */
690 was_same = (alias &&
691 !ce_stage(alias) &&
692 !hashcmp(alias->sha1, ce->sha1) &&
693 ce->ce_mode == alias->ce_mode);
694
695 if (pretend)
696 ;
697 else if (add_index_entry(istate, ce, add_option))
698 return error("unable to add %s to index",path);
699 if (verbose && !was_same)
700 printf("add '%s'\n", path);
701 return 0;
702 }
703
704 int add_file_to_index(struct index_state *istate, const char *path, int flags)
705 {
706 struct stat st;
707 if (lstat(path, &st))
708 die_errno("unable to stat '%s'", path);
709 return add_to_index(istate, path, &st, flags);
710 }
711
712 struct cache_entry *make_cache_entry(unsigned int mode,
713 const unsigned char *sha1, const char *path, int stage,
714 unsigned int refresh_options)
715 {
716 int size, len;
717 struct cache_entry *ce;
718
719 if (!verify_path(path)) {
720 error("Invalid path '%s'", path);
721 return NULL;
722 }
723
724 len = strlen(path);
725 size = cache_entry_size(len);
726 ce = xcalloc(1, size);
727
728 hashcpy(ce->sha1, sha1);
729 memcpy(ce->name, path, len);
730 ce->ce_flags = create_ce_flags(stage);
731 ce->ce_namelen = len;
732 ce->ce_mode = create_ce_mode(mode);
733
734 return refresh_cache_entry(ce, refresh_options);
735 }
736
737 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
738 {
739 int len = ce_namelen(a);
740 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
741 }
742
743 /*
744 * We fundamentally don't like some paths: we don't want
745 * dot or dot-dot anywhere, and for obvious reasons don't
746 * want to recurse into ".git" either.
747 *
748 * Also, we don't want double slashes or slashes at the
749 * end that can make pathnames ambiguous.
750 */
751 static int verify_dotfile(const char *rest)
752 {
753 /*
754 * The first character was '.', but that
755 * has already been discarded, we now test
756 * the rest.
757 */
758
759 /* "." is not allowed */
760 if (*rest == '\0' || is_dir_sep(*rest))
761 return 0;
762
763 switch (*rest) {
764 /*
765 * ".git" followed by NUL or slash is bad. This
766 * shares the path end test with the ".." case.
767 */
768 case 'g':
769 if (rest[1] != 'i')
770 break;
771 if (rest[2] != 't')
772 break;
773 rest += 2;
774 /* fallthrough */
775 case '.':
776 if (rest[1] == '\0' || is_dir_sep(rest[1]))
777 return 0;
778 }
779 return 1;
780 }
781
782 int verify_path(const char *path)
783 {
784 char c;
785
786 if (has_dos_drive_prefix(path))
787 return 0;
788
789 goto inside;
790 for (;;) {
791 if (!c)
792 return 1;
793 if (is_dir_sep(c)) {
794 inside:
795 c = *path++;
796 if ((c == '.' && !verify_dotfile(path)) ||
797 is_dir_sep(c) || c == '\0')
798 return 0;
799 }
800 c = *path++;
801 }
802 }
803
804 /*
805 * Do we have another file that has the beginning components being a
806 * proper superset of the name we're trying to add?
807 */
808 static int has_file_name(struct index_state *istate,
809 const struct cache_entry *ce, int pos, int ok_to_replace)
810 {
811 int retval = 0;
812 int len = ce_namelen(ce);
813 int stage = ce_stage(ce);
814 const char *name = ce->name;
815
816 while (pos < istate->cache_nr) {
817 struct cache_entry *p = istate->cache[pos++];
818
819 if (len >= ce_namelen(p))
820 break;
821 if (memcmp(name, p->name, len))
822 break;
823 if (ce_stage(p) != stage)
824 continue;
825 if (p->name[len] != '/')
826 continue;
827 if (p->ce_flags & CE_REMOVE)
828 continue;
829 retval = -1;
830 if (!ok_to_replace)
831 break;
832 remove_index_entry_at(istate, --pos);
833 }
834 return retval;
835 }
836
837 /*
838 * Do we have another file with a pathname that is a proper
839 * subset of the name we're trying to add?
840 */
841 static int has_dir_name(struct index_state *istate,
842 const struct cache_entry *ce, int pos, int ok_to_replace)
843 {
844 int retval = 0;
845 int stage = ce_stage(ce);
846 const char *name = ce->name;
847 const char *slash = name + ce_namelen(ce);
848
849 for (;;) {
850 int len;
851
852 for (;;) {
853 if (*--slash == '/')
854 break;
855 if (slash <= ce->name)
856 return retval;
857 }
858 len = slash - name;
859
860 pos = index_name_stage_pos(istate, name, len, stage);
861 if (pos >= 0) {
862 /*
863 * Found one, but not so fast. This could
864 * be a marker that says "I was here, but
865 * I am being removed". Such an entry is
866 * not a part of the resulting tree, and
867 * it is Ok to have a directory at the same
868 * path.
869 */
870 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
871 retval = -1;
872 if (!ok_to_replace)
873 break;
874 remove_index_entry_at(istate, pos);
875 continue;
876 }
877 }
878 else
879 pos = -pos-1;
880
881 /*
882 * Trivial optimization: if we find an entry that
883 * already matches the sub-directory, then we know
884 * we're ok, and we can exit.
885 */
886 while (pos < istate->cache_nr) {
887 struct cache_entry *p = istate->cache[pos];
888 if ((ce_namelen(p) <= len) ||
889 (p->name[len] != '/') ||
890 memcmp(p->name, name, len))
891 break; /* not our subdirectory */
892 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
893 /*
894 * p is at the same stage as our entry, and
895 * is a subdirectory of what we are looking
896 * at, so we cannot have conflicts at our
897 * level or anything shorter.
898 */
899 return retval;
900 pos++;
901 }
902 }
903 return retval;
904 }
905
906 /* We may be in a situation where we already have path/file and path
907 * is being added, or we already have path and path/file is being
908 * added. Either one would result in a nonsense tree that has path
909 * twice when git-write-tree tries to write it out. Prevent it.
910 *
911 * If ok-to-replace is specified, we remove the conflicting entries
912 * from the cache so the caller should recompute the insert position.
913 * When this happens, we return non-zero.
914 */
915 static int check_file_directory_conflict(struct index_state *istate,
916 const struct cache_entry *ce,
917 int pos, int ok_to_replace)
918 {
919 int retval;
920
921 /*
922 * When ce is an "I am going away" entry, we allow it to be added
923 */
924 if (ce->ce_flags & CE_REMOVE)
925 return 0;
926
927 /*
928 * We check if the path is a sub-path of a subsequent pathname
929 * first, since removing those will not change the position
930 * in the array.
931 */
932 retval = has_file_name(istate, ce, pos, ok_to_replace);
933
934 /*
935 * Then check if the path might have a clashing sub-directory
936 * before it.
937 */
938 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
939 }
940
941 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
942 {
943 int pos;
944 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
945 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
946 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
947 int new_only = option & ADD_CACHE_NEW_ONLY;
948
949 cache_tree_invalidate_path(istate, ce->name);
950 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
951
952 /* existing match? Just replace it. */
953 if (pos >= 0) {
954 if (!new_only)
955 replace_index_entry(istate, pos, ce);
956 return 0;
957 }
958 pos = -pos-1;
959
960 /*
961 * Inserting a merged entry ("stage 0") into the index
962 * will always replace all non-merged entries..
963 */
964 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
965 while (ce_same_name(istate->cache[pos], ce)) {
966 ok_to_add = 1;
967 if (!remove_index_entry_at(istate, pos))
968 break;
969 }
970 }
971
972 if (!ok_to_add)
973 return -1;
974 if (!verify_path(ce->name))
975 return error("Invalid path '%s'", ce->name);
976
977 if (!skip_df_check &&
978 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
979 if (!ok_to_replace)
980 return error("'%s' appears as both a file and as a directory",
981 ce->name);
982 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
983 pos = -pos-1;
984 }
985 return pos + 1;
986 }
987
988 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
989 {
990 int pos;
991
992 if (option & ADD_CACHE_JUST_APPEND)
993 pos = istate->cache_nr;
994 else {
995 int ret;
996 ret = add_index_entry_with_check(istate, ce, option);
997 if (ret <= 0)
998 return ret;
999 pos = ret - 1;
1000 }
1001
1002 /* Make sure the array is big enough .. */
1003 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1004
1005 /* Add it in.. */
1006 istate->cache_nr++;
1007 if (istate->cache_nr > pos + 1)
1008 memmove(istate->cache + pos + 1,
1009 istate->cache + pos,
1010 (istate->cache_nr - pos - 1) * sizeof(ce));
1011 set_index_entry(istate, pos, ce);
1012 istate->cache_changed |= CE_ENTRY_ADDED;
1013 return 0;
1014 }
1015
1016 /*
1017 * "refresh" does not calculate a new sha1 file or bring the
1018 * cache up-to-date for mode/content changes. But what it
1019 * _does_ do is to "re-match" the stat information of a file
1020 * with the cache, so that you can refresh the cache for a
1021 * file that hasn't been changed but where the stat entry is
1022 * out of date.
1023 *
1024 * For example, you'd want to do this after doing a "git-read-tree",
1025 * to link up the stat cache details with the proper files.
1026 */
1027 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1028 struct cache_entry *ce,
1029 unsigned int options, int *err,
1030 int *changed_ret)
1031 {
1032 struct stat st;
1033 struct cache_entry *updated;
1034 int changed, size;
1035 int refresh = options & CE_MATCH_REFRESH;
1036 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1037 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1038 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1039
1040 if (!refresh || ce_uptodate(ce))
1041 return ce;
1042
1043 /*
1044 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1045 * that the change to the work tree does not matter and told
1046 * us not to worry.
1047 */
1048 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1049 ce_mark_uptodate(ce);
1050 return ce;
1051 }
1052 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1053 ce_mark_uptodate(ce);
1054 return ce;
1055 }
1056
1057 if (lstat(ce->name, &st) < 0) {
1058 if (ignore_missing && errno == ENOENT)
1059 return ce;
1060 if (err)
1061 *err = errno;
1062 return NULL;
1063 }
1064
1065 changed = ie_match_stat(istate, ce, &st, options);
1066 if (changed_ret)
1067 *changed_ret = changed;
1068 if (!changed) {
1069 /*
1070 * The path is unchanged. If we were told to ignore
1071 * valid bit, then we did the actual stat check and
1072 * found that the entry is unmodified. If the entry
1073 * is not marked VALID, this is the place to mark it
1074 * valid again, under "assume unchanged" mode.
1075 */
1076 if (ignore_valid && assume_unchanged &&
1077 !(ce->ce_flags & CE_VALID))
1078 ; /* mark this one VALID again */
1079 else {
1080 /*
1081 * We do not mark the index itself "modified"
1082 * because CE_UPTODATE flag is in-core only;
1083 * we are not going to write this change out.
1084 */
1085 if (!S_ISGITLINK(ce->ce_mode))
1086 ce_mark_uptodate(ce);
1087 return ce;
1088 }
1089 }
1090
1091 if (ie_modified(istate, ce, &st, options)) {
1092 if (err)
1093 *err = EINVAL;
1094 return NULL;
1095 }
1096
1097 size = ce_size(ce);
1098 updated = xmalloc(size);
1099 memcpy(updated, ce, size);
1100 fill_stat_cache_info(updated, &st);
1101 /*
1102 * If ignore_valid is not set, we should leave CE_VALID bit
1103 * alone. Otherwise, paths marked with --no-assume-unchanged
1104 * (i.e. things to be edited) will reacquire CE_VALID bit
1105 * automatically, which is not really what we want.
1106 */
1107 if (!ignore_valid && assume_unchanged &&
1108 !(ce->ce_flags & CE_VALID))
1109 updated->ce_flags &= ~CE_VALID;
1110
1111 /* istate->cache_changed is updated in the caller */
1112 return updated;
1113 }
1114
1115 static void show_file(const char * fmt, const char * name, int in_porcelain,
1116 int * first, const char *header_msg)
1117 {
1118 if (in_porcelain && *first && header_msg) {
1119 printf("%s\n", header_msg);
1120 *first = 0;
1121 }
1122 printf(fmt, name);
1123 }
1124
1125 int refresh_index(struct index_state *istate, unsigned int flags,
1126 const struct pathspec *pathspec,
1127 char *seen, const char *header_msg)
1128 {
1129 int i;
1130 int has_errors = 0;
1131 int really = (flags & REFRESH_REALLY) != 0;
1132 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1133 int quiet = (flags & REFRESH_QUIET) != 0;
1134 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1135 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1136 int first = 1;
1137 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1138 unsigned int options = (CE_MATCH_REFRESH |
1139 (really ? CE_MATCH_IGNORE_VALID : 0) |
1140 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1141 const char *modified_fmt;
1142 const char *deleted_fmt;
1143 const char *typechange_fmt;
1144 const char *added_fmt;
1145 const char *unmerged_fmt;
1146
1147 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1148 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1149 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1150 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1151 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1152 for (i = 0; i < istate->cache_nr; i++) {
1153 struct cache_entry *ce, *new;
1154 int cache_errno = 0;
1155 int changed = 0;
1156 int filtered = 0;
1157
1158 ce = istate->cache[i];
1159 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1160 continue;
1161
1162 if (pathspec && !ce_path_match(ce, pathspec, seen))
1163 filtered = 1;
1164
1165 if (ce_stage(ce)) {
1166 while ((i < istate->cache_nr) &&
1167 ! strcmp(istate->cache[i]->name, ce->name))
1168 i++;
1169 i--;
1170 if (allow_unmerged)
1171 continue;
1172 if (!filtered)
1173 show_file(unmerged_fmt, ce->name, in_porcelain,
1174 &first, header_msg);
1175 has_errors = 1;
1176 continue;
1177 }
1178
1179 if (filtered)
1180 continue;
1181
1182 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1183 if (new == ce)
1184 continue;
1185 if (!new) {
1186 const char *fmt;
1187
1188 if (really && cache_errno == EINVAL) {
1189 /* If we are doing --really-refresh that
1190 * means the index is not valid anymore.
1191 */
1192 ce->ce_flags &= ~CE_VALID;
1193 istate->cache_changed |= CE_ENTRY_CHANGED;
1194 }
1195 if (quiet)
1196 continue;
1197
1198 if (cache_errno == ENOENT)
1199 fmt = deleted_fmt;
1200 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1201 fmt = added_fmt; /* must be before other checks */
1202 else if (changed & TYPE_CHANGED)
1203 fmt = typechange_fmt;
1204 else
1205 fmt = modified_fmt;
1206 show_file(fmt,
1207 ce->name, in_porcelain, &first, header_msg);
1208 has_errors = 1;
1209 continue;
1210 }
1211
1212 replace_index_entry(istate, i, new);
1213 }
1214 return has_errors;
1215 }
1216
1217 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1218 unsigned int options)
1219 {
1220 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1221 }
1222
1223
1224 /*****************************************************************
1225 * Index File I/O
1226 *****************************************************************/
1227
1228 #define INDEX_FORMAT_DEFAULT 3
1229
1230 static int index_format_config(const char *var, const char *value, void *cb)
1231 {
1232 unsigned int *version = cb;
1233 if (!strcmp(var, "index.version")) {
1234 *version = git_config_int(var, value);
1235 return 0;
1236 }
1237 return 1;
1238 }
1239
1240 static unsigned int get_index_format_default(void)
1241 {
1242 char *envversion = getenv("GIT_INDEX_VERSION");
1243 char *endp;
1244 unsigned int version = INDEX_FORMAT_DEFAULT;
1245
1246 if (!envversion) {
1247 git_config(index_format_config, &version);
1248 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1249 warning(_("index.version set, but the value is invalid.\n"
1250 "Using version %i"), INDEX_FORMAT_DEFAULT);
1251 return INDEX_FORMAT_DEFAULT;
1252 }
1253 return version;
1254 }
1255
1256 version = strtoul(envversion, &endp, 10);
1257 if (*endp ||
1258 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1259 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1260 "Using version %i"), INDEX_FORMAT_DEFAULT);
1261 version = INDEX_FORMAT_DEFAULT;
1262 }
1263 return version;
1264 }
1265
1266 /*
1267 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1268 * Again - this is just a (very strong in practice) heuristic that
1269 * the inode hasn't changed.
1270 *
1271 * We save the fields in big-endian order to allow using the
1272 * index file over NFS transparently.
1273 */
1274 struct ondisk_cache_entry {
1275 struct cache_time ctime;
1276 struct cache_time mtime;
1277 uint32_t dev;
1278 uint32_t ino;
1279 uint32_t mode;
1280 uint32_t uid;
1281 uint32_t gid;
1282 uint32_t size;
1283 unsigned char sha1[20];
1284 uint16_t flags;
1285 char name[FLEX_ARRAY]; /* more */
1286 };
1287
1288 /*
1289 * This struct is used when CE_EXTENDED bit is 1
1290 * The struct must match ondisk_cache_entry exactly from
1291 * ctime till flags
1292 */
1293 struct ondisk_cache_entry_extended {
1294 struct cache_time ctime;
1295 struct cache_time mtime;
1296 uint32_t dev;
1297 uint32_t ino;
1298 uint32_t mode;
1299 uint32_t uid;
1300 uint32_t gid;
1301 uint32_t size;
1302 unsigned char sha1[20];
1303 uint16_t flags;
1304 uint16_t flags2;
1305 char name[FLEX_ARRAY]; /* more */
1306 };
1307
1308 /* These are only used for v3 or lower */
1309 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1310 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1311 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1312 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1313 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1314 ondisk_cache_entry_size(ce_namelen(ce)))
1315
1316 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1317 {
1318 git_SHA_CTX c;
1319 unsigned char sha1[20];
1320 int hdr_version;
1321
1322 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1323 return error("bad signature");
1324 hdr_version = ntohl(hdr->hdr_version);
1325 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1326 return error("bad index version %d", hdr_version);
1327 git_SHA1_Init(&c);
1328 git_SHA1_Update(&c, hdr, size - 20);
1329 git_SHA1_Final(sha1, &c);
1330 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1331 return error("bad index file sha1 signature");
1332 return 0;
1333 }
1334
1335 static int read_index_extension(struct index_state *istate,
1336 const char *ext, void *data, unsigned long sz)
1337 {
1338 switch (CACHE_EXT(ext)) {
1339 case CACHE_EXT_TREE:
1340 istate->cache_tree = cache_tree_read(data, sz);
1341 break;
1342 case CACHE_EXT_RESOLVE_UNDO:
1343 istate->resolve_undo = resolve_undo_read(data, sz);
1344 break;
1345 case CACHE_EXT_LINK:
1346 if (read_link_extension(istate, data, sz))
1347 return -1;
1348 break;
1349 default:
1350 if (*ext < 'A' || 'Z' < *ext)
1351 return error("index uses %.4s extension, which we do not understand",
1352 ext);
1353 fprintf(stderr, "ignoring %.4s extension\n", ext);
1354 break;
1355 }
1356 return 0;
1357 }
1358
1359 int read_index(struct index_state *istate)
1360 {
1361 return read_index_from(istate, get_index_file());
1362 }
1363
1364 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1365 unsigned int flags,
1366 const char *name,
1367 size_t len)
1368 {
1369 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1370
1371 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1372 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1373 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1374 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1375 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1376 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1377 ce->ce_mode = get_be32(&ondisk->mode);
1378 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1379 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1380 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1381 ce->ce_flags = flags & ~CE_NAMEMASK;
1382 ce->ce_namelen = len;
1383 ce->index = 0;
1384 hashcpy(ce->sha1, ondisk->sha1);
1385 memcpy(ce->name, name, len);
1386 ce->name[len] = '\0';
1387 return ce;
1388 }
1389
1390 /*
1391 * Adjacent cache entries tend to share the leading paths, so it makes
1392 * sense to only store the differences in later entries. In the v4
1393 * on-disk format of the index, each on-disk cache entry stores the
1394 * number of bytes to be stripped from the end of the previous name,
1395 * and the bytes to append to the result, to come up with its name.
1396 */
1397 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1398 {
1399 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1400 size_t len = decode_varint(&cp);
1401
1402 if (name->len < len)
1403 die("malformed name field in the index");
1404 strbuf_remove(name, name->len - len, len);
1405 for (ep = cp; *ep; ep++)
1406 ; /* find the end */
1407 strbuf_add(name, cp, ep - cp);
1408 return (const char *)ep + 1 - cp_;
1409 }
1410
1411 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1412 unsigned long *ent_size,
1413 struct strbuf *previous_name)
1414 {
1415 struct cache_entry *ce;
1416 size_t len;
1417 const char *name;
1418 unsigned int flags;
1419
1420 /* On-disk flags are just 16 bits */
1421 flags = get_be16(&ondisk->flags);
1422 len = flags & CE_NAMEMASK;
1423
1424 if (flags & CE_EXTENDED) {
1425 struct ondisk_cache_entry_extended *ondisk2;
1426 int extended_flags;
1427 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1428 extended_flags = get_be16(&ondisk2->flags2) << 16;
1429 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1430 if (extended_flags & ~CE_EXTENDED_FLAGS)
1431 die("Unknown index entry format %08x", extended_flags);
1432 flags |= extended_flags;
1433 name = ondisk2->name;
1434 }
1435 else
1436 name = ondisk->name;
1437
1438 if (!previous_name) {
1439 /* v3 and earlier */
1440 if (len == CE_NAMEMASK)
1441 len = strlen(name);
1442 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1443
1444 *ent_size = ondisk_ce_size(ce);
1445 } else {
1446 unsigned long consumed;
1447 consumed = expand_name_field(previous_name, name);
1448 ce = cache_entry_from_ondisk(ondisk, flags,
1449 previous_name->buf,
1450 previous_name->len);
1451
1452 *ent_size = (name - ((char *)ondisk)) + consumed;
1453 }
1454 return ce;
1455 }
1456
1457 /* remember to discard_cache() before reading a different cache! */
1458 static int do_read_index(struct index_state *istate, const char *path,
1459 int must_exist)
1460 {
1461 int fd, i;
1462 struct stat st;
1463 unsigned long src_offset;
1464 struct cache_header *hdr;
1465 void *mmap;
1466 size_t mmap_size;
1467 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1468
1469 if (istate->initialized)
1470 return istate->cache_nr;
1471
1472 istate->timestamp.sec = 0;
1473 istate->timestamp.nsec = 0;
1474 fd = open(path, O_RDONLY);
1475 if (fd < 0) {
1476 if (!must_exist && errno == ENOENT)
1477 return 0;
1478 die_errno("%s: index file open failed", path);
1479 }
1480
1481 if (fstat(fd, &st))
1482 die_errno("cannot stat the open index");
1483
1484 mmap_size = xsize_t(st.st_size);
1485 if (mmap_size < sizeof(struct cache_header) + 20)
1486 die("index file smaller than expected");
1487
1488 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1489 if (mmap == MAP_FAILED)
1490 die_errno("unable to map index file");
1491 close(fd);
1492
1493 hdr = mmap;
1494 if (verify_hdr(hdr, mmap_size) < 0)
1495 goto unmap;
1496
1497 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1498 istate->version = ntohl(hdr->hdr_version);
1499 istate->cache_nr = ntohl(hdr->hdr_entries);
1500 istate->cache_alloc = alloc_nr(istate->cache_nr);
1501 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1502 istate->initialized = 1;
1503
1504 if (istate->version == 4)
1505 previous_name = &previous_name_buf;
1506 else
1507 previous_name = NULL;
1508
1509 src_offset = sizeof(*hdr);
1510 for (i = 0; i < istate->cache_nr; i++) {
1511 struct ondisk_cache_entry *disk_ce;
1512 struct cache_entry *ce;
1513 unsigned long consumed;
1514
1515 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1516 ce = create_from_disk(disk_ce, &consumed, previous_name);
1517 set_index_entry(istate, i, ce);
1518
1519 src_offset += consumed;
1520 }
1521 strbuf_release(&previous_name_buf);
1522 istate->timestamp.sec = st.st_mtime;
1523 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1524
1525 while (src_offset <= mmap_size - 20 - 8) {
1526 /* After an array of active_nr index entries,
1527 * there can be arbitrary number of extended
1528 * sections, each of which is prefixed with
1529 * extension name (4-byte) and section length
1530 * in 4-byte network byte order.
1531 */
1532 uint32_t extsize;
1533 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1534 extsize = ntohl(extsize);
1535 if (read_index_extension(istate,
1536 (const char *) mmap + src_offset,
1537 (char *) mmap + src_offset + 8,
1538 extsize) < 0)
1539 goto unmap;
1540 src_offset += 8;
1541 src_offset += extsize;
1542 }
1543 munmap(mmap, mmap_size);
1544 return istate->cache_nr;
1545
1546 unmap:
1547 munmap(mmap, mmap_size);
1548 die("index file corrupt");
1549 }
1550
1551 int read_index_from(struct index_state *istate, const char *path)
1552 {
1553 struct split_index *split_index;
1554 int ret;
1555
1556 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1557 if (istate->initialized)
1558 return istate->cache_nr;
1559
1560 ret = do_read_index(istate, path, 0);
1561 split_index = istate->split_index;
1562 if (!split_index)
1563 return ret;
1564
1565 if (is_null_sha1(split_index->base_sha1))
1566 return ret;
1567 if (istate->cache_nr)
1568 die("index in split-index mode must contain no entries");
1569
1570 if (split_index->base)
1571 discard_index(split_index->base);
1572 else
1573 split_index->base = xcalloc(1, sizeof(*split_index->base));
1574 ret = do_read_index(split_index->base,
1575 git_path("sharedindex.%s",
1576 sha1_to_hex(split_index->base_sha1)), 1);
1577 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1578 die("broken index, expect %s in %s, got %s",
1579 sha1_to_hex(split_index->base_sha1),
1580 git_path("sharedindex.%s",
1581 sha1_to_hex(split_index->base_sha1)),
1582 sha1_to_hex(split_index->base->sha1));
1583 merge_base_index(istate);
1584 return ret;
1585 }
1586
1587 int is_index_unborn(struct index_state *istate)
1588 {
1589 return (!istate->cache_nr && !istate->timestamp.sec);
1590 }
1591
1592 int discard_index(struct index_state *istate)
1593 {
1594 int i;
1595
1596 for (i = 0; i < istate->cache_nr; i++) {
1597 if (istate->cache[i]->index &&
1598 istate->split_index &&
1599 istate->split_index->base &&
1600 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1601 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1602 continue;
1603 free(istate->cache[i]);
1604 }
1605 resolve_undo_clear_index(istate);
1606 istate->cache_nr = 0;
1607 istate->cache_changed = 0;
1608 istate->timestamp.sec = 0;
1609 istate->timestamp.nsec = 0;
1610 free_name_hash(istate);
1611 cache_tree_free(&(istate->cache_tree));
1612 istate->initialized = 0;
1613 free(istate->cache);
1614 istate->cache = NULL;
1615 istate->cache_alloc = 0;
1616 discard_split_index(istate);
1617 return 0;
1618 }
1619
1620 int unmerged_index(const struct index_state *istate)
1621 {
1622 int i;
1623 for (i = 0; i < istate->cache_nr; i++) {
1624 if (ce_stage(istate->cache[i]))
1625 return 1;
1626 }
1627 return 0;
1628 }
1629
1630 #define WRITE_BUFFER_SIZE 8192
1631 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1632 static unsigned long write_buffer_len;
1633
1634 static int ce_write_flush(git_SHA_CTX *context, int fd)
1635 {
1636 unsigned int buffered = write_buffer_len;
1637 if (buffered) {
1638 git_SHA1_Update(context, write_buffer, buffered);
1639 if (write_in_full(fd, write_buffer, buffered) != buffered)
1640 return -1;
1641 write_buffer_len = 0;
1642 }
1643 return 0;
1644 }
1645
1646 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1647 {
1648 while (len) {
1649 unsigned int buffered = write_buffer_len;
1650 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1651 if (partial > len)
1652 partial = len;
1653 memcpy(write_buffer + buffered, data, partial);
1654 buffered += partial;
1655 if (buffered == WRITE_BUFFER_SIZE) {
1656 write_buffer_len = buffered;
1657 if (ce_write_flush(context, fd))
1658 return -1;
1659 buffered = 0;
1660 }
1661 write_buffer_len = buffered;
1662 len -= partial;
1663 data = (char *) data + partial;
1664 }
1665 return 0;
1666 }
1667
1668 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1669 unsigned int ext, unsigned int sz)
1670 {
1671 ext = htonl(ext);
1672 sz = htonl(sz);
1673 return ((ce_write(context, fd, &ext, 4) < 0) ||
1674 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1675 }
1676
1677 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1678 {
1679 unsigned int left = write_buffer_len;
1680
1681 if (left) {
1682 write_buffer_len = 0;
1683 git_SHA1_Update(context, write_buffer, left);
1684 }
1685
1686 /* Flush first if not enough space for SHA1 signature */
1687 if (left + 20 > WRITE_BUFFER_SIZE) {
1688 if (write_in_full(fd, write_buffer, left) != left)
1689 return -1;
1690 left = 0;
1691 }
1692
1693 /* Append the SHA1 signature at the end */
1694 git_SHA1_Final(write_buffer + left, context);
1695 hashcpy(sha1, write_buffer + left);
1696 left += 20;
1697 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1698 }
1699
1700 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1701 {
1702 /*
1703 * The only thing we care about in this function is to smudge the
1704 * falsely clean entry due to touch-update-touch race, so we leave
1705 * everything else as they are. We are called for entries whose
1706 * ce_stat_data.sd_mtime match the index file mtime.
1707 *
1708 * Note that this actually does not do much for gitlinks, for
1709 * which ce_match_stat_basic() always goes to the actual
1710 * contents. The caller checks with is_racy_timestamp() which
1711 * always says "no" for gitlinks, so we are not called for them ;-)
1712 */
1713 struct stat st;
1714
1715 if (lstat(ce->name, &st) < 0)
1716 return;
1717 if (ce_match_stat_basic(ce, &st))
1718 return;
1719 if (ce_modified_check_fs(ce, &st)) {
1720 /* This is "racily clean"; smudge it. Note that this
1721 * is a tricky code. At first glance, it may appear
1722 * that it can break with this sequence:
1723 *
1724 * $ echo xyzzy >frotz
1725 * $ git-update-index --add frotz
1726 * $ : >frotz
1727 * $ sleep 3
1728 * $ echo filfre >nitfol
1729 * $ git-update-index --add nitfol
1730 *
1731 * but it does not. When the second update-index runs,
1732 * it notices that the entry "frotz" has the same timestamp
1733 * as index, and if we were to smudge it by resetting its
1734 * size to zero here, then the object name recorded
1735 * in index is the 6-byte file but the cached stat information
1736 * becomes zero --- which would then match what we would
1737 * obtain from the filesystem next time we stat("frotz").
1738 *
1739 * However, the second update-index, before calling
1740 * this function, notices that the cached size is 6
1741 * bytes and what is on the filesystem is an empty
1742 * file, and never calls us, so the cached size information
1743 * for "frotz" stays 6 which does not match the filesystem.
1744 */
1745 ce->ce_stat_data.sd_size = 0;
1746 }
1747 }
1748
1749 /* Copy miscellaneous fields but not the name */
1750 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1751 struct cache_entry *ce)
1752 {
1753 short flags;
1754
1755 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1756 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1757 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1758 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1759 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1760 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1761 ondisk->mode = htonl(ce->ce_mode);
1762 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1763 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1764 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1765 hashcpy(ondisk->sha1, ce->sha1);
1766
1767 flags = ce->ce_flags & ~CE_NAMEMASK;
1768 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1769 ondisk->flags = htons(flags);
1770 if (ce->ce_flags & CE_EXTENDED) {
1771 struct ondisk_cache_entry_extended *ondisk2;
1772 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1773 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1774 return ondisk2->name;
1775 }
1776 else {
1777 return ondisk->name;
1778 }
1779 }
1780
1781 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1782 struct strbuf *previous_name)
1783 {
1784 int size;
1785 struct ondisk_cache_entry *ondisk;
1786 char *name;
1787 int result;
1788
1789 if (!previous_name) {
1790 size = ondisk_ce_size(ce);
1791 ondisk = xcalloc(1, size);
1792 name = copy_cache_entry_to_ondisk(ondisk, ce);
1793 memcpy(name, ce->name, ce_namelen(ce));
1794 } else {
1795 int common, to_remove, prefix_size;
1796 unsigned char to_remove_vi[16];
1797 for (common = 0;
1798 (ce->name[common] &&
1799 common < previous_name->len &&
1800 ce->name[common] == previous_name->buf[common]);
1801 common++)
1802 ; /* still matching */
1803 to_remove = previous_name->len - common;
1804 prefix_size = encode_varint(to_remove, to_remove_vi);
1805
1806 if (ce->ce_flags & CE_EXTENDED)
1807 size = offsetof(struct ondisk_cache_entry_extended, name);
1808 else
1809 size = offsetof(struct ondisk_cache_entry, name);
1810 size += prefix_size + (ce_namelen(ce) - common + 1);
1811
1812 ondisk = xcalloc(1, size);
1813 name = copy_cache_entry_to_ondisk(ondisk, ce);
1814 memcpy(name, to_remove_vi, prefix_size);
1815 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1816
1817 strbuf_splice(previous_name, common, to_remove,
1818 ce->name + common, ce_namelen(ce) - common);
1819 }
1820
1821 result = ce_write(c, fd, ondisk, size);
1822 free(ondisk);
1823 return result;
1824 }
1825
1826 static int has_racy_timestamp(struct index_state *istate)
1827 {
1828 int entries = istate->cache_nr;
1829 int i;
1830
1831 for (i = 0; i < entries; i++) {
1832 struct cache_entry *ce = istate->cache[i];
1833 if (is_racy_timestamp(istate, ce))
1834 return 1;
1835 }
1836 return 0;
1837 }
1838
1839 /*
1840 * Opportunistically update the index but do not complain if we can't
1841 */
1842 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1843 {
1844 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1845 write_locked_index(istate, lockfile, COMMIT_LOCK))
1846 rollback_lock_file(lockfile);
1847 }
1848
1849 static int do_write_index(struct index_state *istate, int newfd)
1850 {
1851 git_SHA_CTX c;
1852 struct cache_header hdr;
1853 int i, err, removed, extended, hdr_version;
1854 struct cache_entry **cache = istate->cache;
1855 int entries = istate->cache_nr;
1856 struct stat st;
1857 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1858
1859 for (i = removed = extended = 0; i < entries; i++) {
1860 if (cache[i]->ce_flags & CE_REMOVE)
1861 removed++;
1862
1863 /* reduce extended entries if possible */
1864 cache[i]->ce_flags &= ~CE_EXTENDED;
1865 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1866 extended++;
1867 cache[i]->ce_flags |= CE_EXTENDED;
1868 }
1869 }
1870
1871 if (!istate->version)
1872 istate->version = get_index_format_default();
1873
1874 /* demote version 3 to version 2 when the latter suffices */
1875 if (istate->version == 3 || istate->version == 2)
1876 istate->version = extended ? 3 : 2;
1877
1878 hdr_version = istate->version;
1879
1880 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1881 hdr.hdr_version = htonl(hdr_version);
1882 hdr.hdr_entries = htonl(entries - removed);
1883
1884 git_SHA1_Init(&c);
1885 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1886 return -1;
1887
1888 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1889 for (i = 0; i < entries; i++) {
1890 struct cache_entry *ce = cache[i];
1891 if (ce->ce_flags & CE_REMOVE)
1892 continue;
1893 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1894 ce_smudge_racily_clean_entry(ce);
1895 if (is_null_sha1(ce->sha1)) {
1896 static const char msg[] = "cache entry has null sha1: %s";
1897 static int allow = -1;
1898
1899 if (allow < 0)
1900 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1901 if (allow)
1902 warning(msg, ce->name);
1903 else
1904 return error(msg, ce->name);
1905 }
1906 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1907 return -1;
1908 }
1909 strbuf_release(&previous_name_buf);
1910
1911 /* Write extension data here */
1912 if (istate->split_index) {
1913 struct strbuf sb = STRBUF_INIT;
1914
1915 err = write_link_extension(&sb, istate) < 0 ||
1916 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
1917 sb.len) < 0 ||
1918 ce_write(&c, newfd, sb.buf, sb.len) < 0;
1919 strbuf_release(&sb);
1920 if (err)
1921 return -1;
1922 }
1923 if (istate->cache_tree) {
1924 struct strbuf sb = STRBUF_INIT;
1925
1926 cache_tree_write(&sb, istate->cache_tree);
1927 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1928 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1929 strbuf_release(&sb);
1930 if (err)
1931 return -1;
1932 }
1933 if (istate->resolve_undo) {
1934 struct strbuf sb = STRBUF_INIT;
1935
1936 resolve_undo_write(&sb, istate->resolve_undo);
1937 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1938 sb.len) < 0
1939 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1940 strbuf_release(&sb);
1941 if (err)
1942 return -1;
1943 }
1944
1945 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
1946 return -1;
1947 istate->timestamp.sec = (unsigned int)st.st_mtime;
1948 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1949 return 0;
1950 }
1951
1952 void set_alternate_index_output(const char *name)
1953 {
1954 alternate_index_output = name;
1955 }
1956
1957 static int commit_locked_index(struct lock_file *lk)
1958 {
1959 if (alternate_index_output) {
1960 if (lk->fd >= 0 && close_lock_file(lk))
1961 return -1;
1962 if (rename(lk->filename, alternate_index_output))
1963 return -1;
1964 lk->filename[0] = 0;
1965 return 0;
1966 } else {
1967 return commit_lock_file(lk);
1968 }
1969 }
1970
1971 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
1972 unsigned flags)
1973 {
1974 int ret = do_write_index(istate, lock->fd);
1975 if (ret)
1976 return ret;
1977 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
1978 (COMMIT_LOCK | CLOSE_LOCK));
1979 if (flags & COMMIT_LOCK)
1980 return commit_locked_index(lock);
1981 else if (flags & CLOSE_LOCK)
1982 return close_lock_file(lock);
1983 else
1984 return ret;
1985 }
1986
1987 static int write_split_index(struct index_state *istate,
1988 struct lock_file *lock,
1989 unsigned flags)
1990 {
1991 int ret;
1992 prepare_to_write_split_index(istate);
1993 ret = do_write_locked_index(istate, lock, flags);
1994 finish_writing_split_index(istate);
1995 return ret;
1996 }
1997
1998 int write_locked_index(struct index_state *istate, struct lock_file *lock,
1999 unsigned flags)
2000 {
2001 struct split_index *si = istate->split_index;
2002
2003 if (!si || (istate->cache_changed & ~EXTMASK)) {
2004 if (si)
2005 hashclr(si->base_sha1);
2006 return do_write_locked_index(istate, lock, flags);
2007 }
2008
2009 return write_split_index(istate, lock, flags);
2010 }
2011
2012 /*
2013 * Read the index file that is potentially unmerged into given
2014 * index_state, dropping any unmerged entries. Returns true if
2015 * the index is unmerged. Callers who want to refuse to work
2016 * from an unmerged state can call this and check its return value,
2017 * instead of calling read_cache().
2018 */
2019 int read_index_unmerged(struct index_state *istate)
2020 {
2021 int i;
2022 int unmerged = 0;
2023
2024 read_index(istate);
2025 for (i = 0; i < istate->cache_nr; i++) {
2026 struct cache_entry *ce = istate->cache[i];
2027 struct cache_entry *new_ce;
2028 int size, len;
2029
2030 if (!ce_stage(ce))
2031 continue;
2032 unmerged = 1;
2033 len = ce_namelen(ce);
2034 size = cache_entry_size(len);
2035 new_ce = xcalloc(1, size);
2036 memcpy(new_ce->name, ce->name, len);
2037 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2038 new_ce->ce_namelen = len;
2039 new_ce->ce_mode = ce->ce_mode;
2040 if (add_index_entry(istate, new_ce, 0))
2041 return error("%s: cannot drop to stage #0",
2042 new_ce->name);
2043 i = index_name_pos(istate, new_ce->name, len);
2044 }
2045 return unmerged;
2046 }
2047
2048 /*
2049 * Returns 1 if the path is an "other" path with respect to
2050 * the index; that is, the path is not mentioned in the index at all,
2051 * either as a file, a directory with some files in the index,
2052 * or as an unmerged entry.
2053 *
2054 * We helpfully remove a trailing "/" from directories so that
2055 * the output of read_directory can be used as-is.
2056 */
2057 int index_name_is_other(const struct index_state *istate, const char *name,
2058 int namelen)
2059 {
2060 int pos;
2061 if (namelen && name[namelen - 1] == '/')
2062 namelen--;
2063 pos = index_name_pos(istate, name, namelen);
2064 if (0 <= pos)
2065 return 0; /* exact match */
2066 pos = -pos - 1;
2067 if (pos < istate->cache_nr) {
2068 struct cache_entry *ce = istate->cache[pos];
2069 if (ce_namelen(ce) == namelen &&
2070 !memcmp(ce->name, name, namelen))
2071 return 0; /* Yup, this one exists unmerged */
2072 }
2073 return 1;
2074 }
2075
2076 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2077 {
2078 int pos, len;
2079 unsigned long sz;
2080 enum object_type type;
2081 void *data;
2082
2083 len = strlen(path);
2084 pos = index_name_pos(istate, path, len);
2085 if (pos < 0) {
2086 /*
2087 * We might be in the middle of a merge, in which
2088 * case we would read stage #2 (ours).
2089 */
2090 int i;
2091 for (i = -pos - 1;
2092 (pos < 0 && i < istate->cache_nr &&
2093 !strcmp(istate->cache[i]->name, path));
2094 i++)
2095 if (ce_stage(istate->cache[i]) == 2)
2096 pos = i;
2097 }
2098 if (pos < 0)
2099 return NULL;
2100 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2101 if (!data || type != OBJ_BLOB) {
2102 free(data);
2103 return NULL;
2104 }
2105 if (size)
2106 *size = sz;
2107 return data;
2108 }
2109
2110 void stat_validity_clear(struct stat_validity *sv)
2111 {
2112 free(sv->sd);
2113 sv->sd = NULL;
2114 }
2115
2116 int stat_validity_check(struct stat_validity *sv, const char *path)
2117 {
2118 struct stat st;
2119
2120 if (stat(path, &st) < 0)
2121 return sv->sd == NULL;
2122 if (!sv->sd)
2123 return 0;
2124 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2125 }
2126
2127 void stat_validity_update(struct stat_validity *sv, int fd)
2128 {
2129 struct stat st;
2130
2131 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2132 stat_validity_clear(sv);
2133 else {
2134 if (!sv->sd)
2135 sv->sd = xcalloc(1, sizeof(struct stat_data));
2136 fill_stat_data(sv->sd, &st);
2137 }
2138 }