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