tempfile: do not delete tempfile on failed close
[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 "config.h"
9 #include "tempfile.h"
10 #include "lockfile.h"
11 #include "cache-tree.h"
12 #include "refs.h"
13 #include "dir.h"
14 #include "tree.h"
15 #include "commit.h"
16 #include "blob.h"
17 #include "resolve-undo.h"
18 #include "strbuf.h"
19 #include "varint.h"
20 #include "split-index.h"
21 #include "utf8.h"
22
23 /* Mask for the name length in ce_flags in the on-disk index */
24
25 #define CE_NAMEMASK (0x0fff)
26
27 /* Index extensions.
28 *
29 * The first letter should be 'A'..'Z' for extensions that are not
30 * necessary for a correct operation (i.e. optimization data).
31 * When new extensions are added that _needs_ to be understood in
32 * order to correctly interpret the index file, pick character that
33 * is outside the range, to cause the reader to abort.
34 */
35
36 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
37 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
38 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
39 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
40 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
41
42 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
43 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
44 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
45 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
46
47 struct index_state the_index;
48 static const char *alternate_index_output;
49
50 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
51 {
52 istate->cache[nr] = ce;
53 add_name_hash(istate, ce);
54 }
55
56 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
57 {
58 struct cache_entry *old = istate->cache[nr];
59
60 replace_index_entry_in_base(istate, old, ce);
61 remove_name_hash(istate, old);
62 free(old);
63 set_index_entry(istate, nr, ce);
64 ce->ce_flags |= CE_UPDATE_IN_BASE;
65 istate->cache_changed |= CE_ENTRY_CHANGED;
66 }
67
68 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
69 {
70 struct cache_entry *old = istate->cache[nr], *new;
71 int namelen = strlen(new_name);
72
73 new = xmalloc(cache_entry_size(namelen));
74 copy_cache_entry(new, old);
75 new->ce_flags &= ~CE_HASHED;
76 new->ce_namelen = namelen;
77 new->index = 0;
78 memcpy(new->name, new_name, namelen + 1);
79
80 cache_tree_invalidate_path(istate, old->name);
81 untracked_cache_remove_from_index(istate, old->name);
82 remove_index_entry_at(istate, nr);
83 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
84 }
85
86 void fill_stat_data(struct stat_data *sd, struct stat *st)
87 {
88 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
89 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
90 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
91 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
92 sd->sd_dev = st->st_dev;
93 sd->sd_ino = st->st_ino;
94 sd->sd_uid = st->st_uid;
95 sd->sd_gid = st->st_gid;
96 sd->sd_size = st->st_size;
97 }
98
99 int match_stat_data(const struct stat_data *sd, struct stat *st)
100 {
101 int changed = 0;
102
103 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
104 changed |= MTIME_CHANGED;
105 if (trust_ctime && check_stat &&
106 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
107 changed |= CTIME_CHANGED;
108
109 #ifdef USE_NSEC
110 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
111 changed |= MTIME_CHANGED;
112 if (trust_ctime && check_stat &&
113 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
114 changed |= CTIME_CHANGED;
115 #endif
116
117 if (check_stat) {
118 if (sd->sd_uid != (unsigned int) st->st_uid ||
119 sd->sd_gid != (unsigned int) st->st_gid)
120 changed |= OWNER_CHANGED;
121 if (sd->sd_ino != (unsigned int) st->st_ino)
122 changed |= INODE_CHANGED;
123 }
124
125 #ifdef USE_STDEV
126 /*
127 * st_dev breaks on network filesystems where different
128 * clients will have different views of what "device"
129 * the filesystem is on
130 */
131 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
132 changed |= INODE_CHANGED;
133 #endif
134
135 if (sd->sd_size != (unsigned int) st->st_size)
136 changed |= DATA_CHANGED;
137
138 return changed;
139 }
140
141 /*
142 * This only updates the "non-critical" parts of the directory
143 * cache, ie the parts that aren't tracked by GIT, and only used
144 * to validate the cache.
145 */
146 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
147 {
148 fill_stat_data(&ce->ce_stat_data, st);
149
150 if (assume_unchanged)
151 ce->ce_flags |= CE_VALID;
152
153 if (S_ISREG(st->st_mode))
154 ce_mark_uptodate(ce);
155 }
156
157 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
158 {
159 int match = -1;
160 int fd = git_open_cloexec(ce->name, O_RDONLY);
161
162 if (fd >= 0) {
163 struct object_id oid;
164 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
165 match = oidcmp(&oid, &ce->oid);
166 /* index_fd() closed the file descriptor already */
167 }
168 return match;
169 }
170
171 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
172 {
173 int match = -1;
174 void *buffer;
175 unsigned long size;
176 enum object_type type;
177 struct strbuf sb = STRBUF_INIT;
178
179 if (strbuf_readlink(&sb, ce->name, expected_size))
180 return -1;
181
182 buffer = read_sha1_file(ce->oid.hash, &type, &size);
183 if (buffer) {
184 if (size == sb.len)
185 match = memcmp(buffer, sb.buf, size);
186 free(buffer);
187 }
188 strbuf_release(&sb);
189 return match;
190 }
191
192 static int ce_compare_gitlink(const struct cache_entry *ce)
193 {
194 unsigned char sha1[20];
195
196 /*
197 * We don't actually require that the .git directory
198 * under GITLINK directory be a valid git directory. It
199 * might even be missing (in case nobody populated that
200 * sub-project).
201 *
202 * If so, we consider it always to match.
203 */
204 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
205 return 0;
206 return hashcmp(sha1, ce->oid.hash);
207 }
208
209 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
210 {
211 switch (st->st_mode & S_IFMT) {
212 case S_IFREG:
213 if (ce_compare_data(ce, st))
214 return DATA_CHANGED;
215 break;
216 case S_IFLNK:
217 if (ce_compare_link(ce, xsize_t(st->st_size)))
218 return DATA_CHANGED;
219 break;
220 case S_IFDIR:
221 if (S_ISGITLINK(ce->ce_mode))
222 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
223 default:
224 return TYPE_CHANGED;
225 }
226 return 0;
227 }
228
229 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
230 {
231 unsigned int changed = 0;
232
233 if (ce->ce_flags & CE_REMOVE)
234 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
235
236 switch (ce->ce_mode & S_IFMT) {
237 case S_IFREG:
238 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
239 /* We consider only the owner x bit to be relevant for
240 * "mode changes"
241 */
242 if (trust_executable_bit &&
243 (0100 & (ce->ce_mode ^ st->st_mode)))
244 changed |= MODE_CHANGED;
245 break;
246 case S_IFLNK:
247 if (!S_ISLNK(st->st_mode) &&
248 (has_symlinks || !S_ISREG(st->st_mode)))
249 changed |= TYPE_CHANGED;
250 break;
251 case S_IFGITLINK:
252 /* We ignore most of the st_xxx fields for gitlinks */
253 if (!S_ISDIR(st->st_mode))
254 changed |= TYPE_CHANGED;
255 else if (ce_compare_gitlink(ce))
256 changed |= DATA_CHANGED;
257 return changed;
258 default:
259 die("internal error: ce_mode is %o", ce->ce_mode);
260 }
261
262 changed |= match_stat_data(&ce->ce_stat_data, st);
263
264 /* Racily smudged entry? */
265 if (!ce->ce_stat_data.sd_size) {
266 if (!is_empty_blob_sha1(ce->oid.hash))
267 changed |= DATA_CHANGED;
268 }
269
270 return changed;
271 }
272
273 static int is_racy_stat(const struct index_state *istate,
274 const struct stat_data *sd)
275 {
276 return (istate->timestamp.sec &&
277 #ifdef USE_NSEC
278 /* nanosecond timestamped files can also be racy! */
279 (istate->timestamp.sec < sd->sd_mtime.sec ||
280 (istate->timestamp.sec == sd->sd_mtime.sec &&
281 istate->timestamp.nsec <= sd->sd_mtime.nsec))
282 #else
283 istate->timestamp.sec <= sd->sd_mtime.sec
284 #endif
285 );
286 }
287
288 static int is_racy_timestamp(const struct index_state *istate,
289 const struct cache_entry *ce)
290 {
291 return (!S_ISGITLINK(ce->ce_mode) &&
292 is_racy_stat(istate, &ce->ce_stat_data));
293 }
294
295 int match_stat_data_racy(const struct index_state *istate,
296 const struct stat_data *sd, struct stat *st)
297 {
298 if (is_racy_stat(istate, sd))
299 return MTIME_CHANGED;
300 return match_stat_data(sd, st);
301 }
302
303 int ie_match_stat(const struct index_state *istate,
304 const struct cache_entry *ce, struct stat *st,
305 unsigned int options)
306 {
307 unsigned int changed;
308 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
309 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
310 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
311
312 /*
313 * If it's marked as always valid in the index, it's
314 * valid whatever the checked-out copy says.
315 *
316 * skip-worktree has the same effect with higher precedence
317 */
318 if (!ignore_skip_worktree && ce_skip_worktree(ce))
319 return 0;
320 if (!ignore_valid && (ce->ce_flags & CE_VALID))
321 return 0;
322
323 /*
324 * Intent-to-add entries have not been added, so the index entry
325 * by definition never matches what is in the work tree until it
326 * actually gets added.
327 */
328 if (ce_intent_to_add(ce))
329 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
330
331 changed = ce_match_stat_basic(ce, st);
332
333 /*
334 * Within 1 second of this sequence:
335 * echo xyzzy >file && git-update-index --add file
336 * running this command:
337 * echo frotz >file
338 * would give a falsely clean cache entry. The mtime and
339 * length match the cache, and other stat fields do not change.
340 *
341 * We could detect this at update-index time (the cache entry
342 * being registered/updated records the same time as "now")
343 * and delay the return from git-update-index, but that would
344 * effectively mean we can make at most one commit per second,
345 * which is not acceptable. Instead, we check cache entries
346 * whose mtime are the same as the index file timestamp more
347 * carefully than others.
348 */
349 if (!changed && is_racy_timestamp(istate, ce)) {
350 if (assume_racy_is_modified)
351 changed |= DATA_CHANGED;
352 else
353 changed |= ce_modified_check_fs(ce, st);
354 }
355
356 return changed;
357 }
358
359 int ie_modified(const struct index_state *istate,
360 const struct cache_entry *ce,
361 struct stat *st, unsigned int options)
362 {
363 int changed, changed_fs;
364
365 changed = ie_match_stat(istate, ce, st, options);
366 if (!changed)
367 return 0;
368 /*
369 * If the mode or type has changed, there's no point in trying
370 * to refresh the entry - it's not going to match
371 */
372 if (changed & (MODE_CHANGED | TYPE_CHANGED))
373 return changed;
374
375 /*
376 * Immediately after read-tree or update-index --cacheinfo,
377 * the length field is zero, as we have never even read the
378 * lstat(2) information once, and we cannot trust DATA_CHANGED
379 * returned by ie_match_stat() which in turn was returned by
380 * ce_match_stat_basic() to signal that the filesize of the
381 * blob changed. We have to actually go to the filesystem to
382 * see if the contents match, and if so, should answer "unchanged".
383 *
384 * The logic does not apply to gitlinks, as ce_match_stat_basic()
385 * already has checked the actual HEAD from the filesystem in the
386 * subproject. If ie_match_stat() already said it is different,
387 * then we know it is.
388 */
389 if ((changed & DATA_CHANGED) &&
390 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
391 return changed;
392
393 changed_fs = ce_modified_check_fs(ce, st);
394 if (changed_fs)
395 return changed | changed_fs;
396 return 0;
397 }
398
399 int base_name_compare(const char *name1, int len1, int mode1,
400 const char *name2, int len2, int mode2)
401 {
402 unsigned char c1, c2;
403 int len = len1 < len2 ? len1 : len2;
404 int cmp;
405
406 cmp = memcmp(name1, name2, len);
407 if (cmp)
408 return cmp;
409 c1 = name1[len];
410 c2 = name2[len];
411 if (!c1 && S_ISDIR(mode1))
412 c1 = '/';
413 if (!c2 && S_ISDIR(mode2))
414 c2 = '/';
415 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
416 }
417
418 /*
419 * df_name_compare() is identical to base_name_compare(), except it
420 * compares conflicting directory/file entries as equal. Note that
421 * while a directory name compares as equal to a regular file, they
422 * then individually compare _differently_ to a filename that has
423 * a dot after the basename (because '\0' < '.' < '/').
424 *
425 * This is used by routines that want to traverse the git namespace
426 * but then handle conflicting entries together when possible.
427 */
428 int df_name_compare(const char *name1, int len1, int mode1,
429 const char *name2, int len2, int mode2)
430 {
431 int len = len1 < len2 ? len1 : len2, cmp;
432 unsigned char c1, c2;
433
434 cmp = memcmp(name1, name2, len);
435 if (cmp)
436 return cmp;
437 /* Directories and files compare equal (same length, same name) */
438 if (len1 == len2)
439 return 0;
440 c1 = name1[len];
441 if (!c1 && S_ISDIR(mode1))
442 c1 = '/';
443 c2 = name2[len];
444 if (!c2 && S_ISDIR(mode2))
445 c2 = '/';
446 if (c1 == '/' && !c2)
447 return 0;
448 if (c2 == '/' && !c1)
449 return 0;
450 return c1 - c2;
451 }
452
453 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
454 {
455 size_t min_len = (len1 < len2) ? len1 : len2;
456 int cmp = memcmp(name1, name2, min_len);
457 if (cmp)
458 return cmp;
459 if (len1 < len2)
460 return -1;
461 if (len1 > len2)
462 return 1;
463 return 0;
464 }
465
466 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
467 {
468 int cmp;
469
470 cmp = name_compare(name1, len1, name2, len2);
471 if (cmp)
472 return cmp;
473
474 if (stage1 < stage2)
475 return -1;
476 if (stage1 > stage2)
477 return 1;
478 return 0;
479 }
480
481 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
482 {
483 int first, last;
484
485 first = 0;
486 last = istate->cache_nr;
487 while (last > first) {
488 int next = (last + first) >> 1;
489 struct cache_entry *ce = istate->cache[next];
490 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
491 if (!cmp)
492 return next;
493 if (cmp < 0) {
494 last = next;
495 continue;
496 }
497 first = next+1;
498 }
499 return -first-1;
500 }
501
502 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
503 {
504 return index_name_stage_pos(istate, name, namelen, 0);
505 }
506
507 int remove_index_entry_at(struct index_state *istate, int pos)
508 {
509 struct cache_entry *ce = istate->cache[pos];
510
511 record_resolve_undo(istate, ce);
512 remove_name_hash(istate, ce);
513 save_or_free_index_entry(istate, ce);
514 istate->cache_changed |= CE_ENTRY_REMOVED;
515 istate->cache_nr--;
516 if (pos >= istate->cache_nr)
517 return 0;
518 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
519 istate->cache_nr - pos);
520 return 1;
521 }
522
523 /*
524 * Remove all cache entries marked for removal, that is where
525 * CE_REMOVE is set in ce_flags. This is much more effective than
526 * calling remove_index_entry_at() for each entry to be removed.
527 */
528 void remove_marked_cache_entries(struct index_state *istate)
529 {
530 struct cache_entry **ce_array = istate->cache;
531 unsigned int i, j;
532
533 for (i = j = 0; i < istate->cache_nr; i++) {
534 if (ce_array[i]->ce_flags & CE_REMOVE) {
535 remove_name_hash(istate, ce_array[i]);
536 save_or_free_index_entry(istate, ce_array[i]);
537 }
538 else
539 ce_array[j++] = ce_array[i];
540 }
541 if (j == istate->cache_nr)
542 return;
543 istate->cache_changed |= CE_ENTRY_REMOVED;
544 istate->cache_nr = j;
545 }
546
547 int remove_file_from_index(struct index_state *istate, const char *path)
548 {
549 int pos = index_name_pos(istate, path, strlen(path));
550 if (pos < 0)
551 pos = -pos-1;
552 cache_tree_invalidate_path(istate, path);
553 untracked_cache_remove_from_index(istate, path);
554 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
555 remove_index_entry_at(istate, pos);
556 return 0;
557 }
558
559 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
560 {
561 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
562 }
563
564 static int index_name_pos_also_unmerged(struct index_state *istate,
565 const char *path, int namelen)
566 {
567 int pos = index_name_pos(istate, path, namelen);
568 struct cache_entry *ce;
569
570 if (pos >= 0)
571 return pos;
572
573 /* maybe unmerged? */
574 pos = -1 - pos;
575 if (pos >= istate->cache_nr ||
576 compare_name((ce = istate->cache[pos]), path, namelen))
577 return -1;
578
579 /* order of preference: stage 2, 1, 3 */
580 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
581 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
582 !compare_name(ce, path, namelen))
583 pos++;
584 return pos;
585 }
586
587 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
588 {
589 int len = ce_namelen(ce);
590 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
591 }
592
593 /*
594 * If we add a filename that aliases in the cache, we will use the
595 * name that we already have - but we don't want to update the same
596 * alias twice, because that implies that there were actually two
597 * different files with aliasing names!
598 *
599 * So we use the CE_ADDED flag to verify that the alias was an old
600 * one before we accept it as
601 */
602 static struct cache_entry *create_alias_ce(struct index_state *istate,
603 struct cache_entry *ce,
604 struct cache_entry *alias)
605 {
606 int len;
607 struct cache_entry *new;
608
609 if (alias->ce_flags & CE_ADDED)
610 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
611
612 /* Ok, create the new entry using the name of the existing alias */
613 len = ce_namelen(alias);
614 new = xcalloc(1, cache_entry_size(len));
615 memcpy(new->name, alias->name, len);
616 copy_cache_entry(new, ce);
617 save_or_free_index_entry(istate, ce);
618 return new;
619 }
620
621 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
622 {
623 unsigned char sha1[20];
624 if (write_sha1_file("", 0, blob_type, sha1))
625 die("cannot create an empty blob in the object database");
626 hashcpy(ce->oid.hash, sha1);
627 }
628
629 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
630 {
631 int size, namelen, was_same;
632 mode_t st_mode = st->st_mode;
633 struct cache_entry *ce, *alias;
634 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
635 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
636 int pretend = flags & ADD_CACHE_PRETEND;
637 int intent_only = flags & ADD_CACHE_INTENT;
638 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
639 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
640
641 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
642 return error("%s: can only add regular files, symbolic links or git-directories", path);
643
644 namelen = strlen(path);
645 if (S_ISDIR(st_mode)) {
646 while (namelen && path[namelen-1] == '/')
647 namelen--;
648 }
649 size = cache_entry_size(namelen);
650 ce = xcalloc(1, size);
651 memcpy(ce->name, path, namelen);
652 ce->ce_namelen = namelen;
653 if (!intent_only)
654 fill_stat_cache_info(ce, st);
655 else
656 ce->ce_flags |= CE_INTENT_TO_ADD;
657
658
659 if (trust_executable_bit && has_symlinks) {
660 ce->ce_mode = create_ce_mode(st_mode);
661 } else {
662 /* If there is an existing entry, pick the mode bits and type
663 * from it, otherwise assume unexecutable regular file.
664 */
665 struct cache_entry *ent;
666 int pos = index_name_pos_also_unmerged(istate, path, namelen);
667
668 ent = (0 <= pos) ? istate->cache[pos] : NULL;
669 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
670 }
671
672 /* When core.ignorecase=true, determine if a directory of the same name but differing
673 * case already exists within the Git repository. If it does, ensure the directory
674 * case of the file being added to the repository matches (is folded into) the existing
675 * entry's directory case.
676 */
677 if (ignore_case) {
678 adjust_dirname_case(istate, ce->name);
679 }
680
681 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
682 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
683 /* Nothing changed, really */
684 if (!S_ISGITLINK(alias->ce_mode))
685 ce_mark_uptodate(alias);
686 alias->ce_flags |= CE_ADDED;
687
688 free(ce);
689 return 0;
690 }
691 if (!intent_only) {
692 if (index_path(&ce->oid, path, st, HASH_WRITE_OBJECT)) {
693 free(ce);
694 return error("unable to index file %s", path);
695 }
696 } else
697 set_object_name_for_intent_to_add_entry(ce);
698
699 if (ignore_case && alias && different_name(ce, alias))
700 ce = create_alias_ce(istate, ce, alias);
701 ce->ce_flags |= CE_ADDED;
702
703 /* It was suspected to be racily clean, but it turns out to be Ok */
704 was_same = (alias &&
705 !ce_stage(alias) &&
706 !oidcmp(&alias->oid, &ce->oid) &&
707 ce->ce_mode == alias->ce_mode);
708
709 if (pretend)
710 free(ce);
711 else if (add_index_entry(istate, ce, add_option)) {
712 free(ce);
713 return error("unable to add %s to index", path);
714 }
715 if (verbose && !was_same)
716 printf("add '%s'\n", path);
717 return 0;
718 }
719
720 int add_file_to_index(struct index_state *istate, const char *path, int flags)
721 {
722 struct stat st;
723 if (lstat(path, &st))
724 die_errno("unable to stat '%s'", path);
725 return add_to_index(istate, path, &st, flags);
726 }
727
728 struct cache_entry *make_cache_entry(unsigned int mode,
729 const unsigned char *sha1, const char *path, int stage,
730 unsigned int refresh_options)
731 {
732 int size, len;
733 struct cache_entry *ce, *ret;
734
735 if (!verify_path(path)) {
736 error("Invalid path '%s'", path);
737 return NULL;
738 }
739
740 len = strlen(path);
741 size = cache_entry_size(len);
742 ce = xcalloc(1, size);
743
744 hashcpy(ce->oid.hash, sha1);
745 memcpy(ce->name, path, len);
746 ce->ce_flags = create_ce_flags(stage);
747 ce->ce_namelen = len;
748 ce->ce_mode = create_ce_mode(mode);
749
750 ret = refresh_cache_entry(ce, refresh_options);
751 if (ret != ce)
752 free(ce);
753 return ret;
754 }
755
756 /*
757 * Chmod an index entry with either +x or -x.
758 *
759 * Returns -1 if the chmod for the particular cache entry failed (if it's
760 * not a regular file), -2 if an invalid flip argument is passed in, 0
761 * otherwise.
762 */
763 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
764 char flip)
765 {
766 if (!S_ISREG(ce->ce_mode))
767 return -1;
768 switch (flip) {
769 case '+':
770 ce->ce_mode |= 0111;
771 break;
772 case '-':
773 ce->ce_mode &= ~0111;
774 break;
775 default:
776 return -2;
777 }
778 cache_tree_invalidate_path(istate, ce->name);
779 ce->ce_flags |= CE_UPDATE_IN_BASE;
780 istate->cache_changed |= CE_ENTRY_CHANGED;
781
782 return 0;
783 }
784
785 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
786 {
787 int len = ce_namelen(a);
788 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
789 }
790
791 /*
792 * We fundamentally don't like some paths: we don't want
793 * dot or dot-dot anywhere, and for obvious reasons don't
794 * want to recurse into ".git" either.
795 *
796 * Also, we don't want double slashes or slashes at the
797 * end that can make pathnames ambiguous.
798 */
799 static int verify_dotfile(const char *rest)
800 {
801 /*
802 * The first character was '.', but that
803 * has already been discarded, we now test
804 * the rest.
805 */
806
807 /* "." is not allowed */
808 if (*rest == '\0' || is_dir_sep(*rest))
809 return 0;
810
811 switch (*rest) {
812 /*
813 * ".git" followed by NUL or slash is bad. This
814 * shares the path end test with the ".." case.
815 */
816 case 'g':
817 case 'G':
818 if (rest[1] != 'i' && rest[1] != 'I')
819 break;
820 if (rest[2] != 't' && rest[2] != 'T')
821 break;
822 rest += 2;
823 /* fallthrough */
824 case '.':
825 if (rest[1] == '\0' || is_dir_sep(rest[1]))
826 return 0;
827 }
828 return 1;
829 }
830
831 int verify_path(const char *path)
832 {
833 char c;
834
835 if (has_dos_drive_prefix(path))
836 return 0;
837
838 goto inside;
839 for (;;) {
840 if (!c)
841 return 1;
842 if (is_dir_sep(c)) {
843 inside:
844 if (protect_hfs && is_hfs_dotgit(path))
845 return 0;
846 if (protect_ntfs && is_ntfs_dotgit(path))
847 return 0;
848 c = *path++;
849 if ((c == '.' && !verify_dotfile(path)) ||
850 is_dir_sep(c) || c == '\0')
851 return 0;
852 }
853 c = *path++;
854 }
855 }
856
857 /*
858 * Do we have another file that has the beginning components being a
859 * proper superset of the name we're trying to add?
860 */
861 static int has_file_name(struct index_state *istate,
862 const struct cache_entry *ce, int pos, int ok_to_replace)
863 {
864 int retval = 0;
865 int len = ce_namelen(ce);
866 int stage = ce_stage(ce);
867 const char *name = ce->name;
868
869 while (pos < istate->cache_nr) {
870 struct cache_entry *p = istate->cache[pos++];
871
872 if (len >= ce_namelen(p))
873 break;
874 if (memcmp(name, p->name, len))
875 break;
876 if (ce_stage(p) != stage)
877 continue;
878 if (p->name[len] != '/')
879 continue;
880 if (p->ce_flags & CE_REMOVE)
881 continue;
882 retval = -1;
883 if (!ok_to_replace)
884 break;
885 remove_index_entry_at(istate, --pos);
886 }
887 return retval;
888 }
889
890
891 /*
892 * Like strcmp(), but also return the offset of the first change.
893 * If strings are equal, return the length.
894 */
895 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
896 {
897 size_t k;
898
899 if (!first_change)
900 return strcmp(s1, s2);
901
902 for (k = 0; s1[k] == s2[k]; k++)
903 if (s1[k] == '\0')
904 break;
905
906 *first_change = k;
907 return (unsigned char)s1[k] - (unsigned char)s2[k];
908 }
909
910 /*
911 * Do we have another file with a pathname that is a proper
912 * subset of the name we're trying to add?
913 *
914 * That is, is there another file in the index with a path
915 * that matches a sub-directory in the given entry?
916 */
917 static int has_dir_name(struct index_state *istate,
918 const struct cache_entry *ce, int pos, int ok_to_replace)
919 {
920 int retval = 0;
921 int stage = ce_stage(ce);
922 const char *name = ce->name;
923 const char *slash = name + ce_namelen(ce);
924 size_t len_eq_last;
925 int cmp_last = 0;
926
927 /*
928 * We are frequently called during an iteration on a sorted
929 * list of pathnames and while building a new index. Therefore,
930 * there is a high probability that this entry will eventually
931 * be appended to the index, rather than inserted in the middle.
932 * If we can confirm that, we can avoid binary searches on the
933 * components of the pathname.
934 *
935 * Compare the entry's full path with the last path in the index.
936 */
937 if (istate->cache_nr > 0) {
938 cmp_last = strcmp_offset(name,
939 istate->cache[istate->cache_nr - 1]->name,
940 &len_eq_last);
941 if (cmp_last > 0) {
942 if (len_eq_last == 0) {
943 /*
944 * The entry sorts AFTER the last one in the
945 * index and their paths have no common prefix,
946 * so there cannot be a F/D conflict.
947 */
948 return retval;
949 } else {
950 /*
951 * The entry sorts AFTER the last one in the
952 * index, but has a common prefix. Fall through
953 * to the loop below to disect the entry's path
954 * and see where the difference is.
955 */
956 }
957 } else if (cmp_last == 0) {
958 /*
959 * The entry exactly matches the last one in the
960 * index, but because of multiple stage and CE_REMOVE
961 * items, we fall through and let the regular search
962 * code handle it.
963 */
964 }
965 }
966
967 for (;;) {
968 size_t len;
969
970 for (;;) {
971 if (*--slash == '/')
972 break;
973 if (slash <= ce->name)
974 return retval;
975 }
976 len = slash - name;
977
978 if (cmp_last > 0) {
979 /*
980 * (len + 1) is a directory boundary (including
981 * the trailing slash). And since the loop is
982 * decrementing "slash", the first iteration is
983 * the longest directory prefix; subsequent
984 * iterations consider parent directories.
985 */
986
987 if (len + 1 <= len_eq_last) {
988 /*
989 * The directory prefix (including the trailing
990 * slash) also appears as a prefix in the last
991 * entry, so the remainder cannot collide (because
992 * strcmp said the whole path was greater).
993 *
994 * EQ: last: xxx/A
995 * this: xxx/B
996 *
997 * LT: last: xxx/file_A
998 * this: xxx/file_B
999 */
1000 return retval;
1001 }
1002
1003 if (len > len_eq_last) {
1004 /*
1005 * This part of the directory prefix (excluding
1006 * the trailing slash) is longer than the known
1007 * equal portions, so this sub-directory cannot
1008 * collide with a file.
1009 *
1010 * GT: last: xxxA
1011 * this: xxxB/file
1012 */
1013 return retval;
1014 }
1015
1016 if (istate->cache_nr > 0 &&
1017 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1018 /*
1019 * The directory prefix lines up with part of
1020 * a longer file or directory name, but sorts
1021 * after it, so this sub-directory cannot
1022 * collide with a file.
1023 *
1024 * last: xxx/yy-file (because '-' sorts before '/')
1025 * this: xxx/yy/abc
1026 */
1027 return retval;
1028 }
1029
1030 /*
1031 * This is a possible collision. Fall through and
1032 * let the regular search code handle it.
1033 *
1034 * last: xxx
1035 * this: xxx/file
1036 */
1037 }
1038
1039 pos = index_name_stage_pos(istate, name, len, stage);
1040 if (pos >= 0) {
1041 /*
1042 * Found one, but not so fast. This could
1043 * be a marker that says "I was here, but
1044 * I am being removed". Such an entry is
1045 * not a part of the resulting tree, and
1046 * it is Ok to have a directory at the same
1047 * path.
1048 */
1049 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1050 retval = -1;
1051 if (!ok_to_replace)
1052 break;
1053 remove_index_entry_at(istate, pos);
1054 continue;
1055 }
1056 }
1057 else
1058 pos = -pos-1;
1059
1060 /*
1061 * Trivial optimization: if we find an entry that
1062 * already matches the sub-directory, then we know
1063 * we're ok, and we can exit.
1064 */
1065 while (pos < istate->cache_nr) {
1066 struct cache_entry *p = istate->cache[pos];
1067 if ((ce_namelen(p) <= len) ||
1068 (p->name[len] != '/') ||
1069 memcmp(p->name, name, len))
1070 break; /* not our subdirectory */
1071 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1072 /*
1073 * p is at the same stage as our entry, and
1074 * is a subdirectory of what we are looking
1075 * at, so we cannot have conflicts at our
1076 * level or anything shorter.
1077 */
1078 return retval;
1079 pos++;
1080 }
1081 }
1082 return retval;
1083 }
1084
1085 /* We may be in a situation where we already have path/file and path
1086 * is being added, or we already have path and path/file is being
1087 * added. Either one would result in a nonsense tree that has path
1088 * twice when git-write-tree tries to write it out. Prevent it.
1089 *
1090 * If ok-to-replace is specified, we remove the conflicting entries
1091 * from the cache so the caller should recompute the insert position.
1092 * When this happens, we return non-zero.
1093 */
1094 static int check_file_directory_conflict(struct index_state *istate,
1095 const struct cache_entry *ce,
1096 int pos, int ok_to_replace)
1097 {
1098 int retval;
1099
1100 /*
1101 * When ce is an "I am going away" entry, we allow it to be added
1102 */
1103 if (ce->ce_flags & CE_REMOVE)
1104 return 0;
1105
1106 /*
1107 * We check if the path is a sub-path of a subsequent pathname
1108 * first, since removing those will not change the position
1109 * in the array.
1110 */
1111 retval = has_file_name(istate, ce, pos, ok_to_replace);
1112
1113 /*
1114 * Then check if the path might have a clashing sub-directory
1115 * before it.
1116 */
1117 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1118 }
1119
1120 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1121 {
1122 int pos;
1123 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1124 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1125 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1126 int new_only = option & ADD_CACHE_NEW_ONLY;
1127
1128 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1129 cache_tree_invalidate_path(istate, ce->name);
1130
1131 /*
1132 * If this entry's path sorts after the last entry in the index,
1133 * we can avoid searching for it.
1134 */
1135 if (istate->cache_nr > 0 &&
1136 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1137 pos = -istate->cache_nr - 1;
1138 else
1139 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1140
1141 /* existing match? Just replace it. */
1142 if (pos >= 0) {
1143 if (!new_only)
1144 replace_index_entry(istate, pos, ce);
1145 return 0;
1146 }
1147 pos = -pos-1;
1148
1149 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1150 untracked_cache_add_to_index(istate, ce->name);
1151
1152 /*
1153 * Inserting a merged entry ("stage 0") into the index
1154 * will always replace all non-merged entries..
1155 */
1156 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1157 while (ce_same_name(istate->cache[pos], ce)) {
1158 ok_to_add = 1;
1159 if (!remove_index_entry_at(istate, pos))
1160 break;
1161 }
1162 }
1163
1164 if (!ok_to_add)
1165 return -1;
1166 if (!verify_path(ce->name))
1167 return error("Invalid path '%s'", ce->name);
1168
1169 if (!skip_df_check &&
1170 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1171 if (!ok_to_replace)
1172 return error("'%s' appears as both a file and as a directory",
1173 ce->name);
1174 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1175 pos = -pos-1;
1176 }
1177 return pos + 1;
1178 }
1179
1180 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1181 {
1182 int pos;
1183
1184 if (option & ADD_CACHE_JUST_APPEND)
1185 pos = istate->cache_nr;
1186 else {
1187 int ret;
1188 ret = add_index_entry_with_check(istate, ce, option);
1189 if (ret <= 0)
1190 return ret;
1191 pos = ret - 1;
1192 }
1193
1194 /* Make sure the array is big enough .. */
1195 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1196
1197 /* Add it in.. */
1198 istate->cache_nr++;
1199 if (istate->cache_nr > pos + 1)
1200 memmove(istate->cache + pos + 1,
1201 istate->cache + pos,
1202 (istate->cache_nr - pos - 1) * sizeof(ce));
1203 set_index_entry(istate, pos, ce);
1204 istate->cache_changed |= CE_ENTRY_ADDED;
1205 return 0;
1206 }
1207
1208 /*
1209 * "refresh" does not calculate a new sha1 file or bring the
1210 * cache up-to-date for mode/content changes. But what it
1211 * _does_ do is to "re-match" the stat information of a file
1212 * with the cache, so that you can refresh the cache for a
1213 * file that hasn't been changed but where the stat entry is
1214 * out of date.
1215 *
1216 * For example, you'd want to do this after doing a "git-read-tree",
1217 * to link up the stat cache details with the proper files.
1218 */
1219 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1220 struct cache_entry *ce,
1221 unsigned int options, int *err,
1222 int *changed_ret)
1223 {
1224 struct stat st;
1225 struct cache_entry *updated;
1226 int changed, size;
1227 int refresh = options & CE_MATCH_REFRESH;
1228 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1229 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1230 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1231
1232 if (!refresh || ce_uptodate(ce))
1233 return ce;
1234
1235 /*
1236 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1237 * that the change to the work tree does not matter and told
1238 * us not to worry.
1239 */
1240 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1241 ce_mark_uptodate(ce);
1242 return ce;
1243 }
1244 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1245 ce_mark_uptodate(ce);
1246 return ce;
1247 }
1248
1249 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1250 if (ignore_missing)
1251 return ce;
1252 if (err)
1253 *err = ENOENT;
1254 return NULL;
1255 }
1256
1257 if (lstat(ce->name, &st) < 0) {
1258 if (ignore_missing && errno == ENOENT)
1259 return ce;
1260 if (err)
1261 *err = errno;
1262 return NULL;
1263 }
1264
1265 changed = ie_match_stat(istate, ce, &st, options);
1266 if (changed_ret)
1267 *changed_ret = changed;
1268 if (!changed) {
1269 /*
1270 * The path is unchanged. If we were told to ignore
1271 * valid bit, then we did the actual stat check and
1272 * found that the entry is unmodified. If the entry
1273 * is not marked VALID, this is the place to mark it
1274 * valid again, under "assume unchanged" mode.
1275 */
1276 if (ignore_valid && assume_unchanged &&
1277 !(ce->ce_flags & CE_VALID))
1278 ; /* mark this one VALID again */
1279 else {
1280 /*
1281 * We do not mark the index itself "modified"
1282 * because CE_UPTODATE flag is in-core only;
1283 * we are not going to write this change out.
1284 */
1285 if (!S_ISGITLINK(ce->ce_mode))
1286 ce_mark_uptodate(ce);
1287 return ce;
1288 }
1289 }
1290
1291 if (ie_modified(istate, ce, &st, options)) {
1292 if (err)
1293 *err = EINVAL;
1294 return NULL;
1295 }
1296
1297 size = ce_size(ce);
1298 updated = xmalloc(size);
1299 memcpy(updated, ce, size);
1300 fill_stat_cache_info(updated, &st);
1301 /*
1302 * If ignore_valid is not set, we should leave CE_VALID bit
1303 * alone. Otherwise, paths marked with --no-assume-unchanged
1304 * (i.e. things to be edited) will reacquire CE_VALID bit
1305 * automatically, which is not really what we want.
1306 */
1307 if (!ignore_valid && assume_unchanged &&
1308 !(ce->ce_flags & CE_VALID))
1309 updated->ce_flags &= ~CE_VALID;
1310
1311 /* istate->cache_changed is updated in the caller */
1312 return updated;
1313 }
1314
1315 static void show_file(const char * fmt, const char * name, int in_porcelain,
1316 int * first, const char *header_msg)
1317 {
1318 if (in_porcelain && *first && header_msg) {
1319 printf("%s\n", header_msg);
1320 *first = 0;
1321 }
1322 printf(fmt, name);
1323 }
1324
1325 int refresh_index(struct index_state *istate, unsigned int flags,
1326 const struct pathspec *pathspec,
1327 char *seen, const char *header_msg)
1328 {
1329 int i;
1330 int has_errors = 0;
1331 int really = (flags & REFRESH_REALLY) != 0;
1332 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1333 int quiet = (flags & REFRESH_QUIET) != 0;
1334 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1335 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1336 int first = 1;
1337 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1338 unsigned int options = (CE_MATCH_REFRESH |
1339 (really ? CE_MATCH_IGNORE_VALID : 0) |
1340 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1341 const char *modified_fmt;
1342 const char *deleted_fmt;
1343 const char *typechange_fmt;
1344 const char *added_fmt;
1345 const char *unmerged_fmt;
1346
1347 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1348 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1349 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1350 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1351 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1352 for (i = 0; i < istate->cache_nr; i++) {
1353 struct cache_entry *ce, *new;
1354 int cache_errno = 0;
1355 int changed = 0;
1356 int filtered = 0;
1357
1358 ce = istate->cache[i];
1359 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1360 continue;
1361
1362 if (pathspec && !ce_path_match(ce, pathspec, seen))
1363 filtered = 1;
1364
1365 if (ce_stage(ce)) {
1366 while ((i < istate->cache_nr) &&
1367 ! strcmp(istate->cache[i]->name, ce->name))
1368 i++;
1369 i--;
1370 if (allow_unmerged)
1371 continue;
1372 if (!filtered)
1373 show_file(unmerged_fmt, ce->name, in_porcelain,
1374 &first, header_msg);
1375 has_errors = 1;
1376 continue;
1377 }
1378
1379 if (filtered)
1380 continue;
1381
1382 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1383 if (new == ce)
1384 continue;
1385 if (!new) {
1386 const char *fmt;
1387
1388 if (really && cache_errno == EINVAL) {
1389 /* If we are doing --really-refresh that
1390 * means the index is not valid anymore.
1391 */
1392 ce->ce_flags &= ~CE_VALID;
1393 ce->ce_flags |= CE_UPDATE_IN_BASE;
1394 istate->cache_changed |= CE_ENTRY_CHANGED;
1395 }
1396 if (quiet)
1397 continue;
1398
1399 if (cache_errno == ENOENT)
1400 fmt = deleted_fmt;
1401 else if (ce_intent_to_add(ce))
1402 fmt = added_fmt; /* must be before other checks */
1403 else if (changed & TYPE_CHANGED)
1404 fmt = typechange_fmt;
1405 else
1406 fmt = modified_fmt;
1407 show_file(fmt,
1408 ce->name, in_porcelain, &first, header_msg);
1409 has_errors = 1;
1410 continue;
1411 }
1412
1413 replace_index_entry(istate, i, new);
1414 }
1415 return has_errors;
1416 }
1417
1418 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1419 unsigned int options)
1420 {
1421 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1422 }
1423
1424
1425 /*****************************************************************
1426 * Index File I/O
1427 *****************************************************************/
1428
1429 #define INDEX_FORMAT_DEFAULT 3
1430
1431 static unsigned int get_index_format_default(void)
1432 {
1433 char *envversion = getenv("GIT_INDEX_VERSION");
1434 char *endp;
1435 int value;
1436 unsigned int version = INDEX_FORMAT_DEFAULT;
1437
1438 if (!envversion) {
1439 if (!git_config_get_int("index.version", &value))
1440 version = value;
1441 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1442 warning(_("index.version set, but the value is invalid.\n"
1443 "Using version %i"), INDEX_FORMAT_DEFAULT);
1444 return INDEX_FORMAT_DEFAULT;
1445 }
1446 return version;
1447 }
1448
1449 version = strtoul(envversion, &endp, 10);
1450 if (*endp ||
1451 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1452 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1453 "Using version %i"), INDEX_FORMAT_DEFAULT);
1454 version = INDEX_FORMAT_DEFAULT;
1455 }
1456 return version;
1457 }
1458
1459 /*
1460 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1461 * Again - this is just a (very strong in practice) heuristic that
1462 * the inode hasn't changed.
1463 *
1464 * We save the fields in big-endian order to allow using the
1465 * index file over NFS transparently.
1466 */
1467 struct ondisk_cache_entry {
1468 struct cache_time ctime;
1469 struct cache_time mtime;
1470 uint32_t dev;
1471 uint32_t ino;
1472 uint32_t mode;
1473 uint32_t uid;
1474 uint32_t gid;
1475 uint32_t size;
1476 unsigned char sha1[20];
1477 uint16_t flags;
1478 char name[FLEX_ARRAY]; /* more */
1479 };
1480
1481 /*
1482 * This struct is used when CE_EXTENDED bit is 1
1483 * The struct must match ondisk_cache_entry exactly from
1484 * ctime till flags
1485 */
1486 struct ondisk_cache_entry_extended {
1487 struct cache_time ctime;
1488 struct cache_time mtime;
1489 uint32_t dev;
1490 uint32_t ino;
1491 uint32_t mode;
1492 uint32_t uid;
1493 uint32_t gid;
1494 uint32_t size;
1495 unsigned char sha1[20];
1496 uint16_t flags;
1497 uint16_t flags2;
1498 char name[FLEX_ARRAY]; /* more */
1499 };
1500
1501 /* These are only used for v3 or lower */
1502 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1503 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1504 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1505 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1506 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1507 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1508 ondisk_cache_entry_size(ce_namelen(ce)))
1509
1510 /* Allow fsck to force verification of the index checksum. */
1511 int verify_index_checksum;
1512
1513 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1514 {
1515 git_SHA_CTX c;
1516 unsigned char sha1[20];
1517 int hdr_version;
1518
1519 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1520 return error("bad signature");
1521 hdr_version = ntohl(hdr->hdr_version);
1522 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1523 return error("bad index version %d", hdr_version);
1524
1525 if (!verify_index_checksum)
1526 return 0;
1527
1528 git_SHA1_Init(&c);
1529 git_SHA1_Update(&c, hdr, size - 20);
1530 git_SHA1_Final(sha1, &c);
1531 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1532 return error("bad index file sha1 signature");
1533 return 0;
1534 }
1535
1536 static int read_index_extension(struct index_state *istate,
1537 const char *ext, void *data, unsigned long sz)
1538 {
1539 switch (CACHE_EXT(ext)) {
1540 case CACHE_EXT_TREE:
1541 istate->cache_tree = cache_tree_read(data, sz);
1542 break;
1543 case CACHE_EXT_RESOLVE_UNDO:
1544 istate->resolve_undo = resolve_undo_read(data, sz);
1545 break;
1546 case CACHE_EXT_LINK:
1547 if (read_link_extension(istate, data, sz))
1548 return -1;
1549 break;
1550 case CACHE_EXT_UNTRACKED:
1551 istate->untracked = read_untracked_extension(data, sz);
1552 break;
1553 default:
1554 if (*ext < 'A' || 'Z' < *ext)
1555 return error("index uses %.4s extension, which we do not understand",
1556 ext);
1557 fprintf(stderr, "ignoring %.4s extension\n", ext);
1558 break;
1559 }
1560 return 0;
1561 }
1562
1563 int hold_locked_index(struct lock_file *lk, int lock_flags)
1564 {
1565 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1566 }
1567
1568 int read_index(struct index_state *istate)
1569 {
1570 return read_index_from(istate, get_index_file());
1571 }
1572
1573 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1574 unsigned int flags,
1575 const char *name,
1576 size_t len)
1577 {
1578 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1579
1580 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1581 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1582 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1583 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1584 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1585 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1586 ce->ce_mode = get_be32(&ondisk->mode);
1587 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1588 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1589 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1590 ce->ce_flags = flags & ~CE_NAMEMASK;
1591 ce->ce_namelen = len;
1592 ce->index = 0;
1593 hashcpy(ce->oid.hash, ondisk->sha1);
1594 memcpy(ce->name, name, len);
1595 ce->name[len] = '\0';
1596 return ce;
1597 }
1598
1599 /*
1600 * Adjacent cache entries tend to share the leading paths, so it makes
1601 * sense to only store the differences in later entries. In the v4
1602 * on-disk format of the index, each on-disk cache entry stores the
1603 * number of bytes to be stripped from the end of the previous name,
1604 * and the bytes to append to the result, to come up with its name.
1605 */
1606 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1607 {
1608 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1609 size_t len = decode_varint(&cp);
1610
1611 if (name->len < len)
1612 die("malformed name field in the index");
1613 strbuf_remove(name, name->len - len, len);
1614 for (ep = cp; *ep; ep++)
1615 ; /* find the end */
1616 strbuf_add(name, cp, ep - cp);
1617 return (const char *)ep + 1 - cp_;
1618 }
1619
1620 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1621 unsigned long *ent_size,
1622 struct strbuf *previous_name)
1623 {
1624 struct cache_entry *ce;
1625 size_t len;
1626 const char *name;
1627 unsigned int flags;
1628
1629 /* On-disk flags are just 16 bits */
1630 flags = get_be16(&ondisk->flags);
1631 len = flags & CE_NAMEMASK;
1632
1633 if (flags & CE_EXTENDED) {
1634 struct ondisk_cache_entry_extended *ondisk2;
1635 int extended_flags;
1636 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1637 extended_flags = get_be16(&ondisk2->flags2) << 16;
1638 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1639 if (extended_flags & ~CE_EXTENDED_FLAGS)
1640 die("Unknown index entry format %08x", extended_flags);
1641 flags |= extended_flags;
1642 name = ondisk2->name;
1643 }
1644 else
1645 name = ondisk->name;
1646
1647 if (!previous_name) {
1648 /* v3 and earlier */
1649 if (len == CE_NAMEMASK)
1650 len = strlen(name);
1651 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1652
1653 *ent_size = ondisk_ce_size(ce);
1654 } else {
1655 unsigned long consumed;
1656 consumed = expand_name_field(previous_name, name);
1657 ce = cache_entry_from_ondisk(ondisk, flags,
1658 previous_name->buf,
1659 previous_name->len);
1660
1661 *ent_size = (name - ((char *)ondisk)) + consumed;
1662 }
1663 return ce;
1664 }
1665
1666 static void check_ce_order(struct index_state *istate)
1667 {
1668 unsigned int i;
1669
1670 for (i = 1; i < istate->cache_nr; i++) {
1671 struct cache_entry *ce = istate->cache[i - 1];
1672 struct cache_entry *next_ce = istate->cache[i];
1673 int name_compare = strcmp(ce->name, next_ce->name);
1674
1675 if (0 < name_compare)
1676 die("unordered stage entries in index");
1677 if (!name_compare) {
1678 if (!ce_stage(ce))
1679 die("multiple stage entries for merged file '%s'",
1680 ce->name);
1681 if (ce_stage(ce) > ce_stage(next_ce))
1682 die("unordered stage entries for '%s'",
1683 ce->name);
1684 }
1685 }
1686 }
1687
1688 static void tweak_untracked_cache(struct index_state *istate)
1689 {
1690 switch (git_config_get_untracked_cache()) {
1691 case -1: /* keep: do nothing */
1692 break;
1693 case 0: /* false */
1694 remove_untracked_cache(istate);
1695 break;
1696 case 1: /* true */
1697 add_untracked_cache(istate);
1698 break;
1699 default: /* unknown value: do nothing */
1700 break;
1701 }
1702 }
1703
1704 static void tweak_split_index(struct index_state *istate)
1705 {
1706 switch (git_config_get_split_index()) {
1707 case -1: /* unset: do nothing */
1708 break;
1709 case 0: /* false */
1710 remove_split_index(istate);
1711 break;
1712 case 1: /* true */
1713 add_split_index(istate);
1714 break;
1715 default: /* unknown value: do nothing */
1716 break;
1717 }
1718 }
1719
1720 static void post_read_index_from(struct index_state *istate)
1721 {
1722 check_ce_order(istate);
1723 tweak_untracked_cache(istate);
1724 tweak_split_index(istate);
1725 }
1726
1727 /* remember to discard_cache() before reading a different cache! */
1728 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1729 {
1730 int fd, i;
1731 struct stat st;
1732 unsigned long src_offset;
1733 struct cache_header *hdr;
1734 void *mmap;
1735 size_t mmap_size;
1736 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1737
1738 if (istate->initialized)
1739 return istate->cache_nr;
1740
1741 istate->timestamp.sec = 0;
1742 istate->timestamp.nsec = 0;
1743 fd = open(path, O_RDONLY);
1744 if (fd < 0) {
1745 if (!must_exist && errno == ENOENT)
1746 return 0;
1747 die_errno("%s: index file open failed", path);
1748 }
1749
1750 if (fstat(fd, &st))
1751 die_errno("cannot stat the open index");
1752
1753 mmap_size = xsize_t(st.st_size);
1754 if (mmap_size < sizeof(struct cache_header) + 20)
1755 die("index file smaller than expected");
1756
1757 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1758 if (mmap == MAP_FAILED)
1759 die_errno("unable to map index file");
1760 close(fd);
1761
1762 hdr = mmap;
1763 if (verify_hdr(hdr, mmap_size) < 0)
1764 goto unmap;
1765
1766 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1767 istate->version = ntohl(hdr->hdr_version);
1768 istate->cache_nr = ntohl(hdr->hdr_entries);
1769 istate->cache_alloc = alloc_nr(istate->cache_nr);
1770 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1771 istate->initialized = 1;
1772
1773 if (istate->version == 4)
1774 previous_name = &previous_name_buf;
1775 else
1776 previous_name = NULL;
1777
1778 src_offset = sizeof(*hdr);
1779 for (i = 0; i < istate->cache_nr; i++) {
1780 struct ondisk_cache_entry *disk_ce;
1781 struct cache_entry *ce;
1782 unsigned long consumed;
1783
1784 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1785 ce = create_from_disk(disk_ce, &consumed, previous_name);
1786 set_index_entry(istate, i, ce);
1787
1788 src_offset += consumed;
1789 }
1790 strbuf_release(&previous_name_buf);
1791 istate->timestamp.sec = st.st_mtime;
1792 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1793
1794 while (src_offset <= mmap_size - 20 - 8) {
1795 /* After an array of active_nr index entries,
1796 * there can be arbitrary number of extended
1797 * sections, each of which is prefixed with
1798 * extension name (4-byte) and section length
1799 * in 4-byte network byte order.
1800 */
1801 uint32_t extsize;
1802 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1803 extsize = ntohl(extsize);
1804 if (read_index_extension(istate,
1805 (const char *) mmap + src_offset,
1806 (char *) mmap + src_offset + 8,
1807 extsize) < 0)
1808 goto unmap;
1809 src_offset += 8;
1810 src_offset += extsize;
1811 }
1812 munmap(mmap, mmap_size);
1813 return istate->cache_nr;
1814
1815 unmap:
1816 munmap(mmap, mmap_size);
1817 die("index file corrupt");
1818 }
1819
1820 /*
1821 * Signal that the shared index is used by updating its mtime.
1822 *
1823 * This way, shared index can be removed if they have not been used
1824 * for some time.
1825 */
1826 static void freshen_shared_index(char *base_sha1_hex, int warn)
1827 {
1828 char *shared_index = git_pathdup("sharedindex.%s", base_sha1_hex);
1829 if (!check_and_freshen_file(shared_index, 1) && warn)
1830 warning("could not freshen shared index '%s'", shared_index);
1831 free(shared_index);
1832 }
1833
1834 int read_index_from(struct index_state *istate, const char *path)
1835 {
1836 struct split_index *split_index;
1837 int ret;
1838 char *base_sha1_hex;
1839 const char *base_path;
1840
1841 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1842 if (istate->initialized)
1843 return istate->cache_nr;
1844
1845 ret = do_read_index(istate, path, 0);
1846
1847 split_index = istate->split_index;
1848 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1849 post_read_index_from(istate);
1850 return ret;
1851 }
1852
1853 if (split_index->base)
1854 discard_index(split_index->base);
1855 else
1856 split_index->base = xcalloc(1, sizeof(*split_index->base));
1857
1858 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1859 base_path = git_path("sharedindex.%s", base_sha1_hex);
1860 ret = do_read_index(split_index->base, base_path, 1);
1861 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1862 die("broken index, expect %s in %s, got %s",
1863 base_sha1_hex, base_path,
1864 sha1_to_hex(split_index->base->sha1));
1865
1866 freshen_shared_index(base_sha1_hex, 0);
1867 merge_base_index(istate);
1868 post_read_index_from(istate);
1869 return ret;
1870 }
1871
1872 int is_index_unborn(struct index_state *istate)
1873 {
1874 return (!istate->cache_nr && !istate->timestamp.sec);
1875 }
1876
1877 int discard_index(struct index_state *istate)
1878 {
1879 int i;
1880
1881 for (i = 0; i < istate->cache_nr; i++) {
1882 if (istate->cache[i]->index &&
1883 istate->split_index &&
1884 istate->split_index->base &&
1885 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1886 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1887 continue;
1888 free(istate->cache[i]);
1889 }
1890 resolve_undo_clear_index(istate);
1891 istate->cache_nr = 0;
1892 istate->cache_changed = 0;
1893 istate->timestamp.sec = 0;
1894 istate->timestamp.nsec = 0;
1895 free_name_hash(istate);
1896 cache_tree_free(&(istate->cache_tree));
1897 istate->initialized = 0;
1898 FREE_AND_NULL(istate->cache);
1899 istate->cache_alloc = 0;
1900 discard_split_index(istate);
1901 free_untracked_cache(istate->untracked);
1902 istate->untracked = NULL;
1903 return 0;
1904 }
1905
1906 int unmerged_index(const struct index_state *istate)
1907 {
1908 int i;
1909 for (i = 0; i < istate->cache_nr; i++) {
1910 if (ce_stage(istate->cache[i]))
1911 return 1;
1912 }
1913 return 0;
1914 }
1915
1916 #define WRITE_BUFFER_SIZE 8192
1917 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1918 static unsigned long write_buffer_len;
1919
1920 static int ce_write_flush(git_SHA_CTX *context, int fd)
1921 {
1922 unsigned int buffered = write_buffer_len;
1923 if (buffered) {
1924 git_SHA1_Update(context, write_buffer, buffered);
1925 if (write_in_full(fd, write_buffer, buffered) != buffered)
1926 return -1;
1927 write_buffer_len = 0;
1928 }
1929 return 0;
1930 }
1931
1932 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1933 {
1934 while (len) {
1935 unsigned int buffered = write_buffer_len;
1936 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1937 if (partial > len)
1938 partial = len;
1939 memcpy(write_buffer + buffered, data, partial);
1940 buffered += partial;
1941 if (buffered == WRITE_BUFFER_SIZE) {
1942 write_buffer_len = buffered;
1943 if (ce_write_flush(context, fd))
1944 return -1;
1945 buffered = 0;
1946 }
1947 write_buffer_len = buffered;
1948 len -= partial;
1949 data = (char *) data + partial;
1950 }
1951 return 0;
1952 }
1953
1954 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1955 unsigned int ext, unsigned int sz)
1956 {
1957 ext = htonl(ext);
1958 sz = htonl(sz);
1959 return ((ce_write(context, fd, &ext, 4) < 0) ||
1960 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1961 }
1962
1963 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1964 {
1965 unsigned int left = write_buffer_len;
1966
1967 if (left) {
1968 write_buffer_len = 0;
1969 git_SHA1_Update(context, write_buffer, left);
1970 }
1971
1972 /* Flush first if not enough space for SHA1 signature */
1973 if (left + 20 > WRITE_BUFFER_SIZE) {
1974 if (write_in_full(fd, write_buffer, left) != left)
1975 return -1;
1976 left = 0;
1977 }
1978
1979 /* Append the SHA1 signature at the end */
1980 git_SHA1_Final(write_buffer + left, context);
1981 hashcpy(sha1, write_buffer + left);
1982 left += 20;
1983 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1984 }
1985
1986 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1987 {
1988 /*
1989 * The only thing we care about in this function is to smudge the
1990 * falsely clean entry due to touch-update-touch race, so we leave
1991 * everything else as they are. We are called for entries whose
1992 * ce_stat_data.sd_mtime match the index file mtime.
1993 *
1994 * Note that this actually does not do much for gitlinks, for
1995 * which ce_match_stat_basic() always goes to the actual
1996 * contents. The caller checks with is_racy_timestamp() which
1997 * always says "no" for gitlinks, so we are not called for them ;-)
1998 */
1999 struct stat st;
2000
2001 if (lstat(ce->name, &st) < 0)
2002 return;
2003 if (ce_match_stat_basic(ce, &st))
2004 return;
2005 if (ce_modified_check_fs(ce, &st)) {
2006 /* This is "racily clean"; smudge it. Note that this
2007 * is a tricky code. At first glance, it may appear
2008 * that it can break with this sequence:
2009 *
2010 * $ echo xyzzy >frotz
2011 * $ git-update-index --add frotz
2012 * $ : >frotz
2013 * $ sleep 3
2014 * $ echo filfre >nitfol
2015 * $ git-update-index --add nitfol
2016 *
2017 * but it does not. When the second update-index runs,
2018 * it notices that the entry "frotz" has the same timestamp
2019 * as index, and if we were to smudge it by resetting its
2020 * size to zero here, then the object name recorded
2021 * in index is the 6-byte file but the cached stat information
2022 * becomes zero --- which would then match what we would
2023 * obtain from the filesystem next time we stat("frotz").
2024 *
2025 * However, the second update-index, before calling
2026 * this function, notices that the cached size is 6
2027 * bytes and what is on the filesystem is an empty
2028 * file, and never calls us, so the cached size information
2029 * for "frotz" stays 6 which does not match the filesystem.
2030 */
2031 ce->ce_stat_data.sd_size = 0;
2032 }
2033 }
2034
2035 /* Copy miscellaneous fields but not the name */
2036 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2037 struct cache_entry *ce)
2038 {
2039 short flags;
2040
2041 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2042 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2043 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2044 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2045 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2046 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2047 ondisk->mode = htonl(ce->ce_mode);
2048 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2049 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2050 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2051 hashcpy(ondisk->sha1, ce->oid.hash);
2052
2053 flags = ce->ce_flags & ~CE_NAMEMASK;
2054 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2055 ondisk->flags = htons(flags);
2056 if (ce->ce_flags & CE_EXTENDED) {
2057 struct ondisk_cache_entry_extended *ondisk2;
2058 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2059 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2060 }
2061 }
2062
2063 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
2064 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2065 {
2066 int size;
2067 int saved_namelen = saved_namelen; /* compiler workaround */
2068 int result;
2069 static unsigned char padding[8] = { 0x00 };
2070
2071 if (ce->ce_flags & CE_STRIP_NAME) {
2072 saved_namelen = ce_namelen(ce);
2073 ce->ce_namelen = 0;
2074 }
2075
2076 if (ce->ce_flags & CE_EXTENDED)
2077 size = offsetof(struct ondisk_cache_entry_extended, name);
2078 else
2079 size = offsetof(struct ondisk_cache_entry, name);
2080
2081 if (!previous_name) {
2082 int len = ce_namelen(ce);
2083 copy_cache_entry_to_ondisk(ondisk, ce);
2084 result = ce_write(c, fd, ondisk, size);
2085 if (!result)
2086 result = ce_write(c, fd, ce->name, len);
2087 if (!result)
2088 result = ce_write(c, fd, padding, align_padding_size(size, len));
2089 } else {
2090 int common, to_remove, prefix_size;
2091 unsigned char to_remove_vi[16];
2092 for (common = 0;
2093 (ce->name[common] &&
2094 common < previous_name->len &&
2095 ce->name[common] == previous_name->buf[common]);
2096 common++)
2097 ; /* still matching */
2098 to_remove = previous_name->len - common;
2099 prefix_size = encode_varint(to_remove, to_remove_vi);
2100
2101 copy_cache_entry_to_ondisk(ondisk, ce);
2102 result = ce_write(c, fd, ondisk, size);
2103 if (!result)
2104 result = ce_write(c, fd, to_remove_vi, prefix_size);
2105 if (!result)
2106 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common + 1);
2107
2108 strbuf_splice(previous_name, common, to_remove,
2109 ce->name + common, ce_namelen(ce) - common);
2110 }
2111 if (ce->ce_flags & CE_STRIP_NAME) {
2112 ce->ce_namelen = saved_namelen;
2113 ce->ce_flags &= ~CE_STRIP_NAME;
2114 }
2115
2116 return result;
2117 }
2118
2119 /*
2120 * This function verifies if index_state has the correct sha1 of the
2121 * index file. Don't die if we have any other failure, just return 0.
2122 */
2123 static int verify_index_from(const struct index_state *istate, const char *path)
2124 {
2125 int fd;
2126 ssize_t n;
2127 struct stat st;
2128 unsigned char sha1[20];
2129
2130 if (!istate->initialized)
2131 return 0;
2132
2133 fd = open(path, O_RDONLY);
2134 if (fd < 0)
2135 return 0;
2136
2137 if (fstat(fd, &st))
2138 goto out;
2139
2140 if (st.st_size < sizeof(struct cache_header) + 20)
2141 goto out;
2142
2143 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2144 if (n != 20)
2145 goto out;
2146
2147 if (hashcmp(istate->sha1, sha1))
2148 goto out;
2149
2150 close(fd);
2151 return 1;
2152
2153 out:
2154 close(fd);
2155 return 0;
2156 }
2157
2158 static int verify_index(const struct index_state *istate)
2159 {
2160 return verify_index_from(istate, get_index_file());
2161 }
2162
2163 static int has_racy_timestamp(struct index_state *istate)
2164 {
2165 int entries = istate->cache_nr;
2166 int i;
2167
2168 for (i = 0; i < entries; i++) {
2169 struct cache_entry *ce = istate->cache[i];
2170 if (is_racy_timestamp(istate, ce))
2171 return 1;
2172 }
2173 return 0;
2174 }
2175
2176 /*
2177 * Opportunistically update the index but do not complain if we can't
2178 */
2179 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2180 {
2181 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2182 verify_index(istate) &&
2183 write_locked_index(istate, lockfile, COMMIT_LOCK))
2184 rollback_lock_file(lockfile);
2185 }
2186
2187 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2188 int strip_extensions)
2189 {
2190 int newfd = tempfile->fd;
2191 git_SHA_CTX c;
2192 struct cache_header hdr;
2193 int i, err = 0, removed, extended, hdr_version;
2194 struct cache_entry **cache = istate->cache;
2195 int entries = istate->cache_nr;
2196 struct stat st;
2197 struct ondisk_cache_entry_extended ondisk;
2198 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2199 int drop_cache_tree = 0;
2200
2201 for (i = removed = extended = 0; i < entries; i++) {
2202 if (cache[i]->ce_flags & CE_REMOVE)
2203 removed++;
2204
2205 /* reduce extended entries if possible */
2206 cache[i]->ce_flags &= ~CE_EXTENDED;
2207 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2208 extended++;
2209 cache[i]->ce_flags |= CE_EXTENDED;
2210 }
2211 }
2212
2213 if (!istate->version) {
2214 istate->version = get_index_format_default();
2215 if (getenv("GIT_TEST_SPLIT_INDEX"))
2216 init_split_index(istate);
2217 }
2218
2219 /* demote version 3 to version 2 when the latter suffices */
2220 if (istate->version == 3 || istate->version == 2)
2221 istate->version = extended ? 3 : 2;
2222
2223 hdr_version = istate->version;
2224
2225 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2226 hdr.hdr_version = htonl(hdr_version);
2227 hdr.hdr_entries = htonl(entries - removed);
2228
2229 git_SHA1_Init(&c);
2230 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2231 return -1;
2232
2233 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2234
2235 for (i = 0; i < entries; i++) {
2236 struct cache_entry *ce = cache[i];
2237 if (ce->ce_flags & CE_REMOVE)
2238 continue;
2239 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2240 ce_smudge_racily_clean_entry(ce);
2241 if (is_null_oid(&ce->oid)) {
2242 static const char msg[] = "cache entry has null sha1: %s";
2243 static int allow = -1;
2244
2245 if (allow < 0)
2246 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2247 if (allow)
2248 warning(msg, ce->name);
2249 else
2250 err = error(msg, ce->name);
2251
2252 drop_cache_tree = 1;
2253 }
2254 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2255 err = -1;
2256
2257 if (err)
2258 break;
2259 }
2260 strbuf_release(&previous_name_buf);
2261
2262 if (err)
2263 return err;
2264
2265 /* Write extension data here */
2266 if (!strip_extensions && istate->split_index) {
2267 struct strbuf sb = STRBUF_INIT;
2268
2269 err = write_link_extension(&sb, istate) < 0 ||
2270 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2271 sb.len) < 0 ||
2272 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2273 strbuf_release(&sb);
2274 if (err)
2275 return -1;
2276 }
2277 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2278 struct strbuf sb = STRBUF_INIT;
2279
2280 cache_tree_write(&sb, istate->cache_tree);
2281 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2282 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2283 strbuf_release(&sb);
2284 if (err)
2285 return -1;
2286 }
2287 if (!strip_extensions && istate->resolve_undo) {
2288 struct strbuf sb = STRBUF_INIT;
2289
2290 resolve_undo_write(&sb, istate->resolve_undo);
2291 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2292 sb.len) < 0
2293 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2294 strbuf_release(&sb);
2295 if (err)
2296 return -1;
2297 }
2298 if (!strip_extensions && istate->untracked) {
2299 struct strbuf sb = STRBUF_INIT;
2300
2301 write_untracked_extension(&sb, istate->untracked);
2302 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2303 sb.len) < 0 ||
2304 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2305 strbuf_release(&sb);
2306 if (err)
2307 return -1;
2308 }
2309
2310 if (ce_flush(&c, newfd, istate->sha1))
2311 return -1;
2312 if (close_tempfile_gently(tempfile)) {
2313 error(_("could not close '%s'"), tempfile->filename.buf);
2314 delete_tempfile(tempfile);
2315 return -1;
2316 }
2317 if (stat(tempfile->filename.buf, &st))
2318 return -1;
2319 istate->timestamp.sec = (unsigned int)st.st_mtime;
2320 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2321 return 0;
2322 }
2323
2324 void set_alternate_index_output(const char *name)
2325 {
2326 alternate_index_output = name;
2327 }
2328
2329 static int commit_locked_index(struct lock_file *lk)
2330 {
2331 if (alternate_index_output)
2332 return commit_lock_file_to(lk, alternate_index_output);
2333 else
2334 return commit_lock_file(lk);
2335 }
2336
2337 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2338 unsigned flags)
2339 {
2340 int ret = do_write_index(istate, &lock->tempfile, 0);
2341 if (ret)
2342 return ret;
2343 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2344 (COMMIT_LOCK | CLOSE_LOCK));
2345 if (flags & COMMIT_LOCK)
2346 return commit_locked_index(lock);
2347 else if (flags & CLOSE_LOCK)
2348 return close_lock_file(lock);
2349 else
2350 return ret;
2351 }
2352
2353 static int write_split_index(struct index_state *istate,
2354 struct lock_file *lock,
2355 unsigned flags)
2356 {
2357 int ret;
2358 prepare_to_write_split_index(istate);
2359 ret = do_write_locked_index(istate, lock, flags);
2360 finish_writing_split_index(istate);
2361 return ret;
2362 }
2363
2364 static const char *shared_index_expire = "2.weeks.ago";
2365
2366 static unsigned long get_shared_index_expire_date(void)
2367 {
2368 static unsigned long shared_index_expire_date;
2369 static int shared_index_expire_date_prepared;
2370
2371 if (!shared_index_expire_date_prepared) {
2372 git_config_get_expiry("splitindex.sharedindexexpire",
2373 &shared_index_expire);
2374 shared_index_expire_date = approxidate(shared_index_expire);
2375 shared_index_expire_date_prepared = 1;
2376 }
2377
2378 return shared_index_expire_date;
2379 }
2380
2381 static int should_delete_shared_index(const char *shared_index_path)
2382 {
2383 struct stat st;
2384 unsigned long expiration;
2385
2386 /* Check timestamp */
2387 expiration = get_shared_index_expire_date();
2388 if (!expiration)
2389 return 0;
2390 if (stat(shared_index_path, &st))
2391 return error_errno(_("could not stat '%s'"), shared_index_path);
2392 if (st.st_mtime > expiration)
2393 return 0;
2394
2395 return 1;
2396 }
2397
2398 static int clean_shared_index_files(const char *current_hex)
2399 {
2400 struct dirent *de;
2401 DIR *dir = opendir(get_git_dir());
2402
2403 if (!dir)
2404 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2405
2406 while ((de = readdir(dir)) != NULL) {
2407 const char *sha1_hex;
2408 const char *shared_index_path;
2409 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2410 continue;
2411 if (!strcmp(sha1_hex, current_hex))
2412 continue;
2413 shared_index_path = git_path("%s", de->d_name);
2414 if (should_delete_shared_index(shared_index_path) > 0 &&
2415 unlink(shared_index_path))
2416 warning_errno(_("unable to unlink: %s"), shared_index_path);
2417 }
2418 closedir(dir);
2419
2420 return 0;
2421 }
2422
2423 static struct tempfile temporary_sharedindex;
2424
2425 static int write_shared_index(struct index_state *istate,
2426 struct lock_file *lock, unsigned flags)
2427 {
2428 struct split_index *si = istate->split_index;
2429 int fd, ret;
2430
2431 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2432 if (fd < 0) {
2433 hashclr(si->base_sha1);
2434 return do_write_locked_index(istate, lock, flags);
2435 }
2436 move_cache_to_base_index(istate);
2437 ret = do_write_index(si->base, &temporary_sharedindex, 1);
2438 if (ret) {
2439 delete_tempfile(&temporary_sharedindex);
2440 return ret;
2441 }
2442 ret = adjust_shared_perm(get_tempfile_path(&temporary_sharedindex));
2443 if (ret) {
2444 int save_errno = errno;
2445 error("cannot fix permission bits on %s", get_tempfile_path(&temporary_sharedindex));
2446 delete_tempfile(&temporary_sharedindex);
2447 errno = save_errno;
2448 return ret;
2449 }
2450 ret = rename_tempfile(&temporary_sharedindex,
2451 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2452 if (!ret) {
2453 hashcpy(si->base_sha1, si->base->sha1);
2454 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2455 }
2456
2457 return ret;
2458 }
2459
2460 static const int default_max_percent_split_change = 20;
2461
2462 static int too_many_not_shared_entries(struct index_state *istate)
2463 {
2464 int i, not_shared = 0;
2465 int max_split = git_config_get_max_percent_split_change();
2466
2467 switch (max_split) {
2468 case -1:
2469 /* not or badly configured: use the default value */
2470 max_split = default_max_percent_split_change;
2471 break;
2472 case 0:
2473 return 1; /* 0% means always write a new shared index */
2474 case 100:
2475 return 0; /* 100% means never write a new shared index */
2476 default:
2477 break; /* just use the configured value */
2478 }
2479
2480 /* Count not shared entries */
2481 for (i = 0; i < istate->cache_nr; i++) {
2482 struct cache_entry *ce = istate->cache[i];
2483 if (!ce->index)
2484 not_shared++;
2485 }
2486
2487 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2488 }
2489
2490 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2491 unsigned flags)
2492 {
2493 int new_shared_index, ret;
2494 struct split_index *si = istate->split_index;
2495
2496 if (!si || alternate_index_output ||
2497 (istate->cache_changed & ~EXTMASK)) {
2498 if (si)
2499 hashclr(si->base_sha1);
2500 return do_write_locked_index(istate, lock, flags);
2501 }
2502
2503 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2504 int v = si->base_sha1[0];
2505 if ((v & 15) < 6)
2506 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2507 }
2508 if (too_many_not_shared_entries(istate))
2509 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2510
2511 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2512
2513 if (new_shared_index) {
2514 ret = write_shared_index(istate, lock, flags);
2515 if (ret)
2516 return ret;
2517 }
2518
2519 ret = write_split_index(istate, lock, flags);
2520
2521 /* Freshen the shared index only if the split-index was written */
2522 if (!ret && !new_shared_index)
2523 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2524
2525 return ret;
2526 }
2527
2528 /*
2529 * Read the index file that is potentially unmerged into given
2530 * index_state, dropping any unmerged entries. Returns true if
2531 * the index is unmerged. Callers who want to refuse to work
2532 * from an unmerged state can call this and check its return value,
2533 * instead of calling read_cache().
2534 */
2535 int read_index_unmerged(struct index_state *istate)
2536 {
2537 int i;
2538 int unmerged = 0;
2539
2540 read_index(istate);
2541 for (i = 0; i < istate->cache_nr; i++) {
2542 struct cache_entry *ce = istate->cache[i];
2543 struct cache_entry *new_ce;
2544 int size, len;
2545
2546 if (!ce_stage(ce))
2547 continue;
2548 unmerged = 1;
2549 len = ce_namelen(ce);
2550 size = cache_entry_size(len);
2551 new_ce = xcalloc(1, size);
2552 memcpy(new_ce->name, ce->name, len);
2553 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2554 new_ce->ce_namelen = len;
2555 new_ce->ce_mode = ce->ce_mode;
2556 if (add_index_entry(istate, new_ce, 0))
2557 return error("%s: cannot drop to stage #0",
2558 new_ce->name);
2559 }
2560 return unmerged;
2561 }
2562
2563 /*
2564 * Returns 1 if the path is an "other" path with respect to
2565 * the index; that is, the path is not mentioned in the index at all,
2566 * either as a file, a directory with some files in the index,
2567 * or as an unmerged entry.
2568 *
2569 * We helpfully remove a trailing "/" from directories so that
2570 * the output of read_directory can be used as-is.
2571 */
2572 int index_name_is_other(const struct index_state *istate, const char *name,
2573 int namelen)
2574 {
2575 int pos;
2576 if (namelen && name[namelen - 1] == '/')
2577 namelen--;
2578 pos = index_name_pos(istate, name, namelen);
2579 if (0 <= pos)
2580 return 0; /* exact match */
2581 pos = -pos - 1;
2582 if (pos < istate->cache_nr) {
2583 struct cache_entry *ce = istate->cache[pos];
2584 if (ce_namelen(ce) == namelen &&
2585 !memcmp(ce->name, name, namelen))
2586 return 0; /* Yup, this one exists unmerged */
2587 }
2588 return 1;
2589 }
2590
2591 void *read_blob_data_from_index(const struct index_state *istate,
2592 const char *path, unsigned long *size)
2593 {
2594 int pos, len;
2595 unsigned long sz;
2596 enum object_type type;
2597 void *data;
2598
2599 len = strlen(path);
2600 pos = index_name_pos(istate, path, len);
2601 if (pos < 0) {
2602 /*
2603 * We might be in the middle of a merge, in which
2604 * case we would read stage #2 (ours).
2605 */
2606 int i;
2607 for (i = -pos - 1;
2608 (pos < 0 && i < istate->cache_nr &&
2609 !strcmp(istate->cache[i]->name, path));
2610 i++)
2611 if (ce_stage(istate->cache[i]) == 2)
2612 pos = i;
2613 }
2614 if (pos < 0)
2615 return NULL;
2616 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2617 if (!data || type != OBJ_BLOB) {
2618 free(data);
2619 return NULL;
2620 }
2621 if (size)
2622 *size = sz;
2623 return data;
2624 }
2625
2626 void stat_validity_clear(struct stat_validity *sv)
2627 {
2628 FREE_AND_NULL(sv->sd);
2629 }
2630
2631 int stat_validity_check(struct stat_validity *sv, const char *path)
2632 {
2633 struct stat st;
2634
2635 if (stat(path, &st) < 0)
2636 return sv->sd == NULL;
2637 if (!sv->sd)
2638 return 0;
2639 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2640 }
2641
2642 void stat_validity_update(struct stat_validity *sv, int fd)
2643 {
2644 struct stat st;
2645
2646 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2647 stat_validity_clear(sv);
2648 else {
2649 if (!sv->sd)
2650 sv->sd = xcalloc(1, sizeof(struct stat_data));
2651 fill_stat_data(sv->sd, &st);
2652 }
2653 }
2654
2655 void move_index_extensions(struct index_state *dst, struct index_state *src)
2656 {
2657 dst->untracked = src->untracked;
2658 src->untracked = NULL;
2659 }