Fix bugs preventing adding updated cache entries to the name hash
[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 #include "fsmonitor.h"
23
24 /* Mask for the name length in ce_flags in the on-disk index */
25
26 #define CE_NAMEMASK (0x0fff)
27
28 /* Index extensions.
29 *
30 * The first letter should be 'A'..'Z' for extensions that are not
31 * necessary for a correct operation (i.e. optimization data).
32 * When new extensions are added that _needs_ to be understood in
33 * order to correctly interpret the index file, pick character that
34 * is outside the range, to cause the reader to abort.
35 */
36
37 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
38 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
39 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
40 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
41 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
42 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
43
44 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
45 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
46 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
47 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
48
49 struct index_state the_index;
50 static const char *alternate_index_output;
51
52 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
53 {
54 istate->cache[nr] = ce;
55 add_name_hash(istate, ce);
56 }
57
58 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
59 {
60 struct cache_entry *old = istate->cache[nr];
61
62 replace_index_entry_in_base(istate, old, ce);
63 remove_name_hash(istate, old);
64 free(old);
65 ce->ce_flags &= ~CE_HASHED;
66 set_index_entry(istate, nr, ce);
67 ce->ce_flags |= CE_UPDATE_IN_BASE;
68 mark_fsmonitor_invalid(istate, ce);
69 istate->cache_changed |= CE_ENTRY_CHANGED;
70 }
71
72 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
73 {
74 struct cache_entry *old = istate->cache[nr], *new;
75 int namelen = strlen(new_name);
76
77 new = xmalloc(cache_entry_size(namelen));
78 copy_cache_entry(new, old);
79 new->ce_flags &= ~CE_HASHED;
80 new->ce_namelen = namelen;
81 new->index = 0;
82 memcpy(new->name, new_name, namelen + 1);
83
84 cache_tree_invalidate_path(istate, old->name);
85 untracked_cache_remove_from_index(istate, old->name);
86 remove_index_entry_at(istate, nr);
87 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
88 }
89
90 void fill_stat_data(struct stat_data *sd, struct stat *st)
91 {
92 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
93 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
94 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
95 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
96 sd->sd_dev = st->st_dev;
97 sd->sd_ino = st->st_ino;
98 sd->sd_uid = st->st_uid;
99 sd->sd_gid = st->st_gid;
100 sd->sd_size = st->st_size;
101 }
102
103 int match_stat_data(const struct stat_data *sd, struct stat *st)
104 {
105 int changed = 0;
106
107 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
108 changed |= MTIME_CHANGED;
109 if (trust_ctime && check_stat &&
110 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
111 changed |= CTIME_CHANGED;
112
113 #ifdef USE_NSEC
114 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
115 changed |= MTIME_CHANGED;
116 if (trust_ctime && check_stat &&
117 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
118 changed |= CTIME_CHANGED;
119 #endif
120
121 if (check_stat) {
122 if (sd->sd_uid != (unsigned int) st->st_uid ||
123 sd->sd_gid != (unsigned int) st->st_gid)
124 changed |= OWNER_CHANGED;
125 if (sd->sd_ino != (unsigned int) st->st_ino)
126 changed |= INODE_CHANGED;
127 }
128
129 #ifdef USE_STDEV
130 /*
131 * st_dev breaks on network filesystems where different
132 * clients will have different views of what "device"
133 * the filesystem is on
134 */
135 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
136 changed |= INODE_CHANGED;
137 #endif
138
139 if (sd->sd_size != (unsigned int) st->st_size)
140 changed |= DATA_CHANGED;
141
142 return changed;
143 }
144
145 /*
146 * This only updates the "non-critical" parts of the directory
147 * cache, ie the parts that aren't tracked by GIT, and only used
148 * to validate the cache.
149 */
150 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
151 {
152 fill_stat_data(&ce->ce_stat_data, st);
153
154 if (assume_unchanged)
155 ce->ce_flags |= CE_VALID;
156
157 if (S_ISREG(st->st_mode)) {
158 ce_mark_uptodate(ce);
159 mark_fsmonitor_valid(ce);
160 }
161 }
162
163 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
164 {
165 int match = -1;
166 int fd = git_open_cloexec(ce->name, O_RDONLY);
167
168 if (fd >= 0) {
169 struct object_id oid;
170 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
171 match = oidcmp(&oid, &ce->oid);
172 /* index_fd() closed the file descriptor already */
173 }
174 return match;
175 }
176
177 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
178 {
179 int match = -1;
180 void *buffer;
181 unsigned long size;
182 enum object_type type;
183 struct strbuf sb = STRBUF_INIT;
184
185 if (strbuf_readlink(&sb, ce->name, expected_size))
186 return -1;
187
188 buffer = read_sha1_file(ce->oid.hash, &type, &size);
189 if (buffer) {
190 if (size == sb.len)
191 match = memcmp(buffer, sb.buf, size);
192 free(buffer);
193 }
194 strbuf_release(&sb);
195 return match;
196 }
197
198 static int ce_compare_gitlink(const struct cache_entry *ce)
199 {
200 struct object_id oid;
201
202 /*
203 * We don't actually require that the .git directory
204 * under GITLINK directory be a valid git directory. It
205 * might even be missing (in case nobody populated that
206 * sub-project).
207 *
208 * If so, we consider it always to match.
209 */
210 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
211 return 0;
212 return oidcmp(&oid, &ce->oid);
213 }
214
215 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
216 {
217 switch (st->st_mode & S_IFMT) {
218 case S_IFREG:
219 if (ce_compare_data(ce, st))
220 return DATA_CHANGED;
221 break;
222 case S_IFLNK:
223 if (ce_compare_link(ce, xsize_t(st->st_size)))
224 return DATA_CHANGED;
225 break;
226 case S_IFDIR:
227 if (S_ISGITLINK(ce->ce_mode))
228 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
229 /* else fallthrough */
230 default:
231 return TYPE_CHANGED;
232 }
233 return 0;
234 }
235
236 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
237 {
238 unsigned int changed = 0;
239
240 if (ce->ce_flags & CE_REMOVE)
241 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
242
243 switch (ce->ce_mode & S_IFMT) {
244 case S_IFREG:
245 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
246 /* We consider only the owner x bit to be relevant for
247 * "mode changes"
248 */
249 if (trust_executable_bit &&
250 (0100 & (ce->ce_mode ^ st->st_mode)))
251 changed |= MODE_CHANGED;
252 break;
253 case S_IFLNK:
254 if (!S_ISLNK(st->st_mode) &&
255 (has_symlinks || !S_ISREG(st->st_mode)))
256 changed |= TYPE_CHANGED;
257 break;
258 case S_IFGITLINK:
259 /* We ignore most of the st_xxx fields for gitlinks */
260 if (!S_ISDIR(st->st_mode))
261 changed |= TYPE_CHANGED;
262 else if (ce_compare_gitlink(ce))
263 changed |= DATA_CHANGED;
264 return changed;
265 default:
266 die("internal error: ce_mode is %o", ce->ce_mode);
267 }
268
269 changed |= match_stat_data(&ce->ce_stat_data, st);
270
271 /* Racily smudged entry? */
272 if (!ce->ce_stat_data.sd_size) {
273 if (!is_empty_blob_sha1(ce->oid.hash))
274 changed |= DATA_CHANGED;
275 }
276
277 return changed;
278 }
279
280 static int is_racy_stat(const struct index_state *istate,
281 const struct stat_data *sd)
282 {
283 return (istate->timestamp.sec &&
284 #ifdef USE_NSEC
285 /* nanosecond timestamped files can also be racy! */
286 (istate->timestamp.sec < sd->sd_mtime.sec ||
287 (istate->timestamp.sec == sd->sd_mtime.sec &&
288 istate->timestamp.nsec <= sd->sd_mtime.nsec))
289 #else
290 istate->timestamp.sec <= sd->sd_mtime.sec
291 #endif
292 );
293 }
294
295 static int is_racy_timestamp(const struct index_state *istate,
296 const struct cache_entry *ce)
297 {
298 return (!S_ISGITLINK(ce->ce_mode) &&
299 is_racy_stat(istate, &ce->ce_stat_data));
300 }
301
302 int match_stat_data_racy(const struct index_state *istate,
303 const struct stat_data *sd, struct stat *st)
304 {
305 if (is_racy_stat(istate, sd))
306 return MTIME_CHANGED;
307 return match_stat_data(sd, st);
308 }
309
310 int ie_match_stat(struct index_state *istate,
311 const struct cache_entry *ce, struct stat *st,
312 unsigned int options)
313 {
314 unsigned int changed;
315 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
316 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
317 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
318 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
319
320 if (!ignore_fsmonitor)
321 refresh_fsmonitor(istate);
322 /*
323 * If it's marked as always valid in the index, it's
324 * valid whatever the checked-out copy says.
325 *
326 * skip-worktree has the same effect with higher precedence
327 */
328 if (!ignore_skip_worktree && ce_skip_worktree(ce))
329 return 0;
330 if (!ignore_valid && (ce->ce_flags & CE_VALID))
331 return 0;
332 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
333 return 0;
334
335 /*
336 * Intent-to-add entries have not been added, so the index entry
337 * by definition never matches what is in the work tree until it
338 * actually gets added.
339 */
340 if (ce_intent_to_add(ce))
341 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
342
343 changed = ce_match_stat_basic(ce, st);
344
345 /*
346 * Within 1 second of this sequence:
347 * echo xyzzy >file && git-update-index --add file
348 * running this command:
349 * echo frotz >file
350 * would give a falsely clean cache entry. The mtime and
351 * length match the cache, and other stat fields do not change.
352 *
353 * We could detect this at update-index time (the cache entry
354 * being registered/updated records the same time as "now")
355 * and delay the return from git-update-index, but that would
356 * effectively mean we can make at most one commit per second,
357 * which is not acceptable. Instead, we check cache entries
358 * whose mtime are the same as the index file timestamp more
359 * carefully than others.
360 */
361 if (!changed && is_racy_timestamp(istate, ce)) {
362 if (assume_racy_is_modified)
363 changed |= DATA_CHANGED;
364 else
365 changed |= ce_modified_check_fs(ce, st);
366 }
367
368 return changed;
369 }
370
371 int ie_modified(struct index_state *istate,
372 const struct cache_entry *ce,
373 struct stat *st, unsigned int options)
374 {
375 int changed, changed_fs;
376
377 changed = ie_match_stat(istate, ce, st, options);
378 if (!changed)
379 return 0;
380 /*
381 * If the mode or type has changed, there's no point in trying
382 * to refresh the entry - it's not going to match
383 */
384 if (changed & (MODE_CHANGED | TYPE_CHANGED))
385 return changed;
386
387 /*
388 * Immediately after read-tree or update-index --cacheinfo,
389 * the length field is zero, as we have never even read the
390 * lstat(2) information once, and we cannot trust DATA_CHANGED
391 * returned by ie_match_stat() which in turn was returned by
392 * ce_match_stat_basic() to signal that the filesize of the
393 * blob changed. We have to actually go to the filesystem to
394 * see if the contents match, and if so, should answer "unchanged".
395 *
396 * The logic does not apply to gitlinks, as ce_match_stat_basic()
397 * already has checked the actual HEAD from the filesystem in the
398 * subproject. If ie_match_stat() already said it is different,
399 * then we know it is.
400 */
401 if ((changed & DATA_CHANGED) &&
402 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
403 return changed;
404
405 changed_fs = ce_modified_check_fs(ce, st);
406 if (changed_fs)
407 return changed | changed_fs;
408 return 0;
409 }
410
411 int base_name_compare(const char *name1, int len1, int mode1,
412 const char *name2, int len2, int mode2)
413 {
414 unsigned char c1, c2;
415 int len = len1 < len2 ? len1 : len2;
416 int cmp;
417
418 cmp = memcmp(name1, name2, len);
419 if (cmp)
420 return cmp;
421 c1 = name1[len];
422 c2 = name2[len];
423 if (!c1 && S_ISDIR(mode1))
424 c1 = '/';
425 if (!c2 && S_ISDIR(mode2))
426 c2 = '/';
427 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
428 }
429
430 /*
431 * df_name_compare() is identical to base_name_compare(), except it
432 * compares conflicting directory/file entries as equal. Note that
433 * while a directory name compares as equal to a regular file, they
434 * then individually compare _differently_ to a filename that has
435 * a dot after the basename (because '\0' < '.' < '/').
436 *
437 * This is used by routines that want to traverse the git namespace
438 * but then handle conflicting entries together when possible.
439 */
440 int df_name_compare(const char *name1, int len1, int mode1,
441 const char *name2, int len2, int mode2)
442 {
443 int len = len1 < len2 ? len1 : len2, cmp;
444 unsigned char c1, c2;
445
446 cmp = memcmp(name1, name2, len);
447 if (cmp)
448 return cmp;
449 /* Directories and files compare equal (same length, same name) */
450 if (len1 == len2)
451 return 0;
452 c1 = name1[len];
453 if (!c1 && S_ISDIR(mode1))
454 c1 = '/';
455 c2 = name2[len];
456 if (!c2 && S_ISDIR(mode2))
457 c2 = '/';
458 if (c1 == '/' && !c2)
459 return 0;
460 if (c2 == '/' && !c1)
461 return 0;
462 return c1 - c2;
463 }
464
465 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
466 {
467 size_t min_len = (len1 < len2) ? len1 : len2;
468 int cmp = memcmp(name1, name2, min_len);
469 if (cmp)
470 return cmp;
471 if (len1 < len2)
472 return -1;
473 if (len1 > len2)
474 return 1;
475 return 0;
476 }
477
478 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
479 {
480 int cmp;
481
482 cmp = name_compare(name1, len1, name2, len2);
483 if (cmp)
484 return cmp;
485
486 if (stage1 < stage2)
487 return -1;
488 if (stage1 > stage2)
489 return 1;
490 return 0;
491 }
492
493 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
494 {
495 int first, last;
496
497 first = 0;
498 last = istate->cache_nr;
499 while (last > first) {
500 int next = (last + first) >> 1;
501 struct cache_entry *ce = istate->cache[next];
502 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
503 if (!cmp)
504 return next;
505 if (cmp < 0) {
506 last = next;
507 continue;
508 }
509 first = next+1;
510 }
511 return -first-1;
512 }
513
514 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
515 {
516 return index_name_stage_pos(istate, name, namelen, 0);
517 }
518
519 int remove_index_entry_at(struct index_state *istate, int pos)
520 {
521 struct cache_entry *ce = istate->cache[pos];
522
523 record_resolve_undo(istate, ce);
524 remove_name_hash(istate, ce);
525 save_or_free_index_entry(istate, ce);
526 istate->cache_changed |= CE_ENTRY_REMOVED;
527 istate->cache_nr--;
528 if (pos >= istate->cache_nr)
529 return 0;
530 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
531 istate->cache_nr - pos);
532 return 1;
533 }
534
535 /*
536 * Remove all cache entries marked for removal, that is where
537 * CE_REMOVE is set in ce_flags. This is much more effective than
538 * calling remove_index_entry_at() for each entry to be removed.
539 */
540 void remove_marked_cache_entries(struct index_state *istate)
541 {
542 struct cache_entry **ce_array = istate->cache;
543 unsigned int i, j;
544
545 for (i = j = 0; i < istate->cache_nr; i++) {
546 if (ce_array[i]->ce_flags & CE_REMOVE) {
547 remove_name_hash(istate, ce_array[i]);
548 save_or_free_index_entry(istate, ce_array[i]);
549 }
550 else
551 ce_array[j++] = ce_array[i];
552 }
553 if (j == istate->cache_nr)
554 return;
555 istate->cache_changed |= CE_ENTRY_REMOVED;
556 istate->cache_nr = j;
557 }
558
559 int remove_file_from_index(struct index_state *istate, const char *path)
560 {
561 int pos = index_name_pos(istate, path, strlen(path));
562 if (pos < 0)
563 pos = -pos-1;
564 cache_tree_invalidate_path(istate, path);
565 untracked_cache_remove_from_index(istate, path);
566 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
567 remove_index_entry_at(istate, pos);
568 return 0;
569 }
570
571 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
572 {
573 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
574 }
575
576 static int index_name_pos_also_unmerged(struct index_state *istate,
577 const char *path, int namelen)
578 {
579 int pos = index_name_pos(istate, path, namelen);
580 struct cache_entry *ce;
581
582 if (pos >= 0)
583 return pos;
584
585 /* maybe unmerged? */
586 pos = -1 - pos;
587 if (pos >= istate->cache_nr ||
588 compare_name((ce = istate->cache[pos]), path, namelen))
589 return -1;
590
591 /* order of preference: stage 2, 1, 3 */
592 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
593 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
594 !compare_name(ce, path, namelen))
595 pos++;
596 return pos;
597 }
598
599 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
600 {
601 int len = ce_namelen(ce);
602 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
603 }
604
605 /*
606 * If we add a filename that aliases in the cache, we will use the
607 * name that we already have - but we don't want to update the same
608 * alias twice, because that implies that there were actually two
609 * different files with aliasing names!
610 *
611 * So we use the CE_ADDED flag to verify that the alias was an old
612 * one before we accept it as
613 */
614 static struct cache_entry *create_alias_ce(struct index_state *istate,
615 struct cache_entry *ce,
616 struct cache_entry *alias)
617 {
618 int len;
619 struct cache_entry *new;
620
621 if (alias->ce_flags & CE_ADDED)
622 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
623
624 /* Ok, create the new entry using the name of the existing alias */
625 len = ce_namelen(alias);
626 new = xcalloc(1, cache_entry_size(len));
627 memcpy(new->name, alias->name, len);
628 copy_cache_entry(new, ce);
629 save_or_free_index_entry(istate, ce);
630 return new;
631 }
632
633 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
634 {
635 unsigned char sha1[20];
636 if (write_sha1_file("", 0, blob_type, sha1))
637 die("cannot create an empty blob in the object database");
638 hashcpy(ce->oid.hash, sha1);
639 }
640
641 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
642 {
643 int size, namelen, was_same;
644 mode_t st_mode = st->st_mode;
645 struct cache_entry *ce, *alias = NULL;
646 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
647 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
648 int pretend = flags & ADD_CACHE_PRETEND;
649 int intent_only = flags & ADD_CACHE_INTENT;
650 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
651 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
652 int newflags = HASH_WRITE_OBJECT;
653
654 if (flags & HASH_RENORMALIZE)
655 newflags |= HASH_RENORMALIZE;
656
657 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
658 return error("%s: can only add regular files, symbolic links or git-directories", path);
659
660 namelen = strlen(path);
661 if (S_ISDIR(st_mode)) {
662 while (namelen && path[namelen-1] == '/')
663 namelen--;
664 }
665 size = cache_entry_size(namelen);
666 ce = xcalloc(1, size);
667 memcpy(ce->name, path, namelen);
668 ce->ce_namelen = namelen;
669 if (!intent_only)
670 fill_stat_cache_info(ce, st);
671 else
672 ce->ce_flags |= CE_INTENT_TO_ADD;
673
674
675 if (trust_executable_bit && has_symlinks) {
676 ce->ce_mode = create_ce_mode(st_mode);
677 } else {
678 /* If there is an existing entry, pick the mode bits and type
679 * from it, otherwise assume unexecutable regular file.
680 */
681 struct cache_entry *ent;
682 int pos = index_name_pos_also_unmerged(istate, path, namelen);
683
684 ent = (0 <= pos) ? istate->cache[pos] : NULL;
685 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
686 }
687
688 /* When core.ignorecase=true, determine if a directory of the same name but differing
689 * case already exists within the Git repository. If it does, ensure the directory
690 * case of the file being added to the repository matches (is folded into) the existing
691 * entry's directory case.
692 */
693 if (ignore_case) {
694 adjust_dirname_case(istate, ce->name);
695 }
696 if (!(flags & HASH_RENORMALIZE)) {
697 alias = index_file_exists(istate, ce->name,
698 ce_namelen(ce), ignore_case);
699 if (alias &&
700 !ce_stage(alias) &&
701 !ie_match_stat(istate, alias, st, ce_option)) {
702 /* Nothing changed, really */
703 if (!S_ISGITLINK(alias->ce_mode))
704 ce_mark_uptodate(alias);
705 alias->ce_flags |= CE_ADDED;
706
707 free(ce);
708 return 0;
709 }
710 }
711 if (!intent_only) {
712 if (index_path(&ce->oid, path, st, newflags)) {
713 free(ce);
714 return error("unable to index file %s", path);
715 }
716 } else
717 set_object_name_for_intent_to_add_entry(ce);
718
719 if (ignore_case && alias && different_name(ce, alias))
720 ce = create_alias_ce(istate, ce, alias);
721 ce->ce_flags |= CE_ADDED;
722
723 /* It was suspected to be racily clean, but it turns out to be Ok */
724 was_same = (alias &&
725 !ce_stage(alias) &&
726 !oidcmp(&alias->oid, &ce->oid) &&
727 ce->ce_mode == alias->ce_mode);
728
729 if (pretend)
730 free(ce);
731 else if (add_index_entry(istate, ce, add_option)) {
732 free(ce);
733 return error("unable to add %s to index", path);
734 }
735 if (verbose && !was_same)
736 printf("add '%s'\n", path);
737 return 0;
738 }
739
740 int add_file_to_index(struct index_state *istate, const char *path, int flags)
741 {
742 struct stat st;
743 if (lstat(path, &st))
744 die_errno("unable to stat '%s'", path);
745 return add_to_index(istate, path, &st, flags);
746 }
747
748 struct cache_entry *make_cache_entry(unsigned int mode,
749 const unsigned char *sha1, const char *path, int stage,
750 unsigned int refresh_options)
751 {
752 int size, len;
753 struct cache_entry *ce, *ret;
754
755 if (!verify_path(path)) {
756 error("Invalid path '%s'", path);
757 return NULL;
758 }
759
760 len = strlen(path);
761 size = cache_entry_size(len);
762 ce = xcalloc(1, size);
763
764 hashcpy(ce->oid.hash, sha1);
765 memcpy(ce->name, path, len);
766 ce->ce_flags = create_ce_flags(stage);
767 ce->ce_namelen = len;
768 ce->ce_mode = create_ce_mode(mode);
769
770 ret = refresh_cache_entry(ce, refresh_options);
771 if (ret != ce)
772 free(ce);
773 return ret;
774 }
775
776 /*
777 * Chmod an index entry with either +x or -x.
778 *
779 * Returns -1 if the chmod for the particular cache entry failed (if it's
780 * not a regular file), -2 if an invalid flip argument is passed in, 0
781 * otherwise.
782 */
783 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
784 char flip)
785 {
786 if (!S_ISREG(ce->ce_mode))
787 return -1;
788 switch (flip) {
789 case '+':
790 ce->ce_mode |= 0111;
791 break;
792 case '-':
793 ce->ce_mode &= ~0111;
794 break;
795 default:
796 return -2;
797 }
798 cache_tree_invalidate_path(istate, ce->name);
799 ce->ce_flags |= CE_UPDATE_IN_BASE;
800 mark_fsmonitor_invalid(istate, ce);
801 istate->cache_changed |= CE_ENTRY_CHANGED;
802
803 return 0;
804 }
805
806 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
807 {
808 int len = ce_namelen(a);
809 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
810 }
811
812 /*
813 * We fundamentally don't like some paths: we don't want
814 * dot or dot-dot anywhere, and for obvious reasons don't
815 * want to recurse into ".git" either.
816 *
817 * Also, we don't want double slashes or slashes at the
818 * end that can make pathnames ambiguous.
819 */
820 static int verify_dotfile(const char *rest)
821 {
822 /*
823 * The first character was '.', but that
824 * has already been discarded, we now test
825 * the rest.
826 */
827
828 /* "." is not allowed */
829 if (*rest == '\0' || is_dir_sep(*rest))
830 return 0;
831
832 switch (*rest) {
833 /*
834 * ".git" followed by NUL or slash is bad. This
835 * shares the path end test with the ".." case.
836 */
837 case 'g':
838 case 'G':
839 if (rest[1] != 'i' && rest[1] != 'I')
840 break;
841 if (rest[2] != 't' && rest[2] != 'T')
842 break;
843 rest += 2;
844 /* fallthrough */
845 case '.':
846 if (rest[1] == '\0' || is_dir_sep(rest[1]))
847 return 0;
848 }
849 return 1;
850 }
851
852 int verify_path(const char *path)
853 {
854 char c;
855
856 if (has_dos_drive_prefix(path))
857 return 0;
858
859 goto inside;
860 for (;;) {
861 if (!c)
862 return 1;
863 if (is_dir_sep(c)) {
864 inside:
865 if (protect_hfs && is_hfs_dotgit(path))
866 return 0;
867 if (protect_ntfs && is_ntfs_dotgit(path))
868 return 0;
869 c = *path++;
870 if ((c == '.' && !verify_dotfile(path)) ||
871 is_dir_sep(c) || c == '\0')
872 return 0;
873 }
874 c = *path++;
875 }
876 }
877
878 /*
879 * Do we have another file that has the beginning components being a
880 * proper superset of the name we're trying to add?
881 */
882 static int has_file_name(struct index_state *istate,
883 const struct cache_entry *ce, int pos, int ok_to_replace)
884 {
885 int retval = 0;
886 int len = ce_namelen(ce);
887 int stage = ce_stage(ce);
888 const char *name = ce->name;
889
890 while (pos < istate->cache_nr) {
891 struct cache_entry *p = istate->cache[pos++];
892
893 if (len >= ce_namelen(p))
894 break;
895 if (memcmp(name, p->name, len))
896 break;
897 if (ce_stage(p) != stage)
898 continue;
899 if (p->name[len] != '/')
900 continue;
901 if (p->ce_flags & CE_REMOVE)
902 continue;
903 retval = -1;
904 if (!ok_to_replace)
905 break;
906 remove_index_entry_at(istate, --pos);
907 }
908 return retval;
909 }
910
911
912 /*
913 * Like strcmp(), but also return the offset of the first change.
914 * If strings are equal, return the length.
915 */
916 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
917 {
918 size_t k;
919
920 if (!first_change)
921 return strcmp(s1, s2);
922
923 for (k = 0; s1[k] == s2[k]; k++)
924 if (s1[k] == '\0')
925 break;
926
927 *first_change = k;
928 return (unsigned char)s1[k] - (unsigned char)s2[k];
929 }
930
931 /*
932 * Do we have another file with a pathname that is a proper
933 * subset of the name we're trying to add?
934 *
935 * That is, is there another file in the index with a path
936 * that matches a sub-directory in the given entry?
937 */
938 static int has_dir_name(struct index_state *istate,
939 const struct cache_entry *ce, int pos, int ok_to_replace)
940 {
941 int retval = 0;
942 int stage = ce_stage(ce);
943 const char *name = ce->name;
944 const char *slash = name + ce_namelen(ce);
945 size_t len_eq_last;
946 int cmp_last = 0;
947
948 /*
949 * We are frequently called during an iteration on a sorted
950 * list of pathnames and while building a new index. Therefore,
951 * there is a high probability that this entry will eventually
952 * be appended to the index, rather than inserted in the middle.
953 * If we can confirm that, we can avoid binary searches on the
954 * components of the pathname.
955 *
956 * Compare the entry's full path with the last path in the index.
957 */
958 if (istate->cache_nr > 0) {
959 cmp_last = strcmp_offset(name,
960 istate->cache[istate->cache_nr - 1]->name,
961 &len_eq_last);
962 if (cmp_last > 0) {
963 if (len_eq_last == 0) {
964 /*
965 * The entry sorts AFTER the last one in the
966 * index and their paths have no common prefix,
967 * so there cannot be a F/D conflict.
968 */
969 return retval;
970 } else {
971 /*
972 * The entry sorts AFTER the last one in the
973 * index, but has a common prefix. Fall through
974 * to the loop below to disect the entry's path
975 * and see where the difference is.
976 */
977 }
978 } else if (cmp_last == 0) {
979 /*
980 * The entry exactly matches the last one in the
981 * index, but because of multiple stage and CE_REMOVE
982 * items, we fall through and let the regular search
983 * code handle it.
984 */
985 }
986 }
987
988 for (;;) {
989 size_t len;
990
991 for (;;) {
992 if (*--slash == '/')
993 break;
994 if (slash <= ce->name)
995 return retval;
996 }
997 len = slash - name;
998
999 if (cmp_last > 0) {
1000 /*
1001 * (len + 1) is a directory boundary (including
1002 * the trailing slash). And since the loop is
1003 * decrementing "slash", the first iteration is
1004 * the longest directory prefix; subsequent
1005 * iterations consider parent directories.
1006 */
1007
1008 if (len + 1 <= len_eq_last) {
1009 /*
1010 * The directory prefix (including the trailing
1011 * slash) also appears as a prefix in the last
1012 * entry, so the remainder cannot collide (because
1013 * strcmp said the whole path was greater).
1014 *
1015 * EQ: last: xxx/A
1016 * this: xxx/B
1017 *
1018 * LT: last: xxx/file_A
1019 * this: xxx/file_B
1020 */
1021 return retval;
1022 }
1023
1024 if (len > len_eq_last) {
1025 /*
1026 * This part of the directory prefix (excluding
1027 * the trailing slash) is longer than the known
1028 * equal portions, so this sub-directory cannot
1029 * collide with a file.
1030 *
1031 * GT: last: xxxA
1032 * this: xxxB/file
1033 */
1034 return retval;
1035 }
1036
1037 if (istate->cache_nr > 0 &&
1038 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1039 /*
1040 * The directory prefix lines up with part of
1041 * a longer file or directory name, but sorts
1042 * after it, so this sub-directory cannot
1043 * collide with a file.
1044 *
1045 * last: xxx/yy-file (because '-' sorts before '/')
1046 * this: xxx/yy/abc
1047 */
1048 return retval;
1049 }
1050
1051 /*
1052 * This is a possible collision. Fall through and
1053 * let the regular search code handle it.
1054 *
1055 * last: xxx
1056 * this: xxx/file
1057 */
1058 }
1059
1060 pos = index_name_stage_pos(istate, name, len, stage);
1061 if (pos >= 0) {
1062 /*
1063 * Found one, but not so fast. This could
1064 * be a marker that says "I was here, but
1065 * I am being removed". Such an entry is
1066 * not a part of the resulting tree, and
1067 * it is Ok to have a directory at the same
1068 * path.
1069 */
1070 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1071 retval = -1;
1072 if (!ok_to_replace)
1073 break;
1074 remove_index_entry_at(istate, pos);
1075 continue;
1076 }
1077 }
1078 else
1079 pos = -pos-1;
1080
1081 /*
1082 * Trivial optimization: if we find an entry that
1083 * already matches the sub-directory, then we know
1084 * we're ok, and we can exit.
1085 */
1086 while (pos < istate->cache_nr) {
1087 struct cache_entry *p = istate->cache[pos];
1088 if ((ce_namelen(p) <= len) ||
1089 (p->name[len] != '/') ||
1090 memcmp(p->name, name, len))
1091 break; /* not our subdirectory */
1092 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1093 /*
1094 * p is at the same stage as our entry, and
1095 * is a subdirectory of what we are looking
1096 * at, so we cannot have conflicts at our
1097 * level or anything shorter.
1098 */
1099 return retval;
1100 pos++;
1101 }
1102 }
1103 return retval;
1104 }
1105
1106 /* We may be in a situation where we already have path/file and path
1107 * is being added, or we already have path and path/file is being
1108 * added. Either one would result in a nonsense tree that has path
1109 * twice when git-write-tree tries to write it out. Prevent it.
1110 *
1111 * If ok-to-replace is specified, we remove the conflicting entries
1112 * from the cache so the caller should recompute the insert position.
1113 * When this happens, we return non-zero.
1114 */
1115 static int check_file_directory_conflict(struct index_state *istate,
1116 const struct cache_entry *ce,
1117 int pos, int ok_to_replace)
1118 {
1119 int retval;
1120
1121 /*
1122 * When ce is an "I am going away" entry, we allow it to be added
1123 */
1124 if (ce->ce_flags & CE_REMOVE)
1125 return 0;
1126
1127 /*
1128 * We check if the path is a sub-path of a subsequent pathname
1129 * first, since removing those will not change the position
1130 * in the array.
1131 */
1132 retval = has_file_name(istate, ce, pos, ok_to_replace);
1133
1134 /*
1135 * Then check if the path might have a clashing sub-directory
1136 * before it.
1137 */
1138 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1139 }
1140
1141 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1142 {
1143 int pos;
1144 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1145 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1146 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1147 int new_only = option & ADD_CACHE_NEW_ONLY;
1148
1149 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1150 cache_tree_invalidate_path(istate, ce->name);
1151
1152 /*
1153 * If this entry's path sorts after the last entry in the index,
1154 * we can avoid searching for it.
1155 */
1156 if (istate->cache_nr > 0 &&
1157 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1158 pos = -istate->cache_nr - 1;
1159 else
1160 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1161
1162 /* existing match? Just replace it. */
1163 if (pos >= 0) {
1164 if (!new_only)
1165 replace_index_entry(istate, pos, ce);
1166 return 0;
1167 }
1168 pos = -pos-1;
1169
1170 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1171 untracked_cache_add_to_index(istate, ce->name);
1172
1173 /*
1174 * Inserting a merged entry ("stage 0") into the index
1175 * will always replace all non-merged entries..
1176 */
1177 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1178 while (ce_same_name(istate->cache[pos], ce)) {
1179 ok_to_add = 1;
1180 if (!remove_index_entry_at(istate, pos))
1181 break;
1182 }
1183 }
1184
1185 if (!ok_to_add)
1186 return -1;
1187 if (!verify_path(ce->name))
1188 return error("Invalid path '%s'", ce->name);
1189
1190 if (!skip_df_check &&
1191 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1192 if (!ok_to_replace)
1193 return error("'%s' appears as both a file and as a directory",
1194 ce->name);
1195 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1196 pos = -pos-1;
1197 }
1198 return pos + 1;
1199 }
1200
1201 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1202 {
1203 int pos;
1204
1205 if (option & ADD_CACHE_JUST_APPEND)
1206 pos = istate->cache_nr;
1207 else {
1208 int ret;
1209 ret = add_index_entry_with_check(istate, ce, option);
1210 if (ret <= 0)
1211 return ret;
1212 pos = ret - 1;
1213 }
1214
1215 /* Make sure the array is big enough .. */
1216 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1217
1218 /* Add it in.. */
1219 istate->cache_nr++;
1220 if (istate->cache_nr > pos + 1)
1221 memmove(istate->cache + pos + 1,
1222 istate->cache + pos,
1223 (istate->cache_nr - pos - 1) * sizeof(ce));
1224 set_index_entry(istate, pos, ce);
1225 istate->cache_changed |= CE_ENTRY_ADDED;
1226 return 0;
1227 }
1228
1229 /*
1230 * "refresh" does not calculate a new sha1 file or bring the
1231 * cache up-to-date for mode/content changes. But what it
1232 * _does_ do is to "re-match" the stat information of a file
1233 * with the cache, so that you can refresh the cache for a
1234 * file that hasn't been changed but where the stat entry is
1235 * out of date.
1236 *
1237 * For example, you'd want to do this after doing a "git-read-tree",
1238 * to link up the stat cache details with the proper files.
1239 */
1240 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1241 struct cache_entry *ce,
1242 unsigned int options, int *err,
1243 int *changed_ret)
1244 {
1245 struct stat st;
1246 struct cache_entry *updated;
1247 int changed, size;
1248 int refresh = options & CE_MATCH_REFRESH;
1249 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1250 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1251 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1252 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1253
1254 if (!refresh || ce_uptodate(ce))
1255 return ce;
1256
1257 if (!ignore_fsmonitor)
1258 refresh_fsmonitor(istate);
1259 /*
1260 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1261 * that the change to the work tree does not matter and told
1262 * us not to worry.
1263 */
1264 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1265 ce_mark_uptodate(ce);
1266 return ce;
1267 }
1268 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1269 ce_mark_uptodate(ce);
1270 return ce;
1271 }
1272 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1273 ce_mark_uptodate(ce);
1274 return ce;
1275 }
1276
1277 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1278 if (ignore_missing)
1279 return ce;
1280 if (err)
1281 *err = ENOENT;
1282 return NULL;
1283 }
1284
1285 if (lstat(ce->name, &st) < 0) {
1286 if (ignore_missing && errno == ENOENT)
1287 return ce;
1288 if (err)
1289 *err = errno;
1290 return NULL;
1291 }
1292
1293 changed = ie_match_stat(istate, ce, &st, options);
1294 if (changed_ret)
1295 *changed_ret = changed;
1296 if (!changed) {
1297 /*
1298 * The path is unchanged. If we were told to ignore
1299 * valid bit, then we did the actual stat check and
1300 * found that the entry is unmodified. If the entry
1301 * is not marked VALID, this is the place to mark it
1302 * valid again, under "assume unchanged" mode.
1303 */
1304 if (ignore_valid && assume_unchanged &&
1305 !(ce->ce_flags & CE_VALID))
1306 ; /* mark this one VALID again */
1307 else {
1308 /*
1309 * We do not mark the index itself "modified"
1310 * because CE_UPTODATE flag is in-core only;
1311 * we are not going to write this change out.
1312 */
1313 if (!S_ISGITLINK(ce->ce_mode)) {
1314 ce_mark_uptodate(ce);
1315 mark_fsmonitor_valid(ce);
1316 }
1317 return ce;
1318 }
1319 }
1320
1321 if (ie_modified(istate, ce, &st, options)) {
1322 if (err)
1323 *err = EINVAL;
1324 return NULL;
1325 }
1326
1327 size = ce_size(ce);
1328 updated = xmalloc(size);
1329 copy_cache_entry(updated, ce);
1330 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1331 fill_stat_cache_info(updated, &st);
1332 /*
1333 * If ignore_valid is not set, we should leave CE_VALID bit
1334 * alone. Otherwise, paths marked with --no-assume-unchanged
1335 * (i.e. things to be edited) will reacquire CE_VALID bit
1336 * automatically, which is not really what we want.
1337 */
1338 if (!ignore_valid && assume_unchanged &&
1339 !(ce->ce_flags & CE_VALID))
1340 updated->ce_flags &= ~CE_VALID;
1341
1342 /* istate->cache_changed is updated in the caller */
1343 return updated;
1344 }
1345
1346 static void show_file(const char * fmt, const char * name, int in_porcelain,
1347 int * first, const char *header_msg)
1348 {
1349 if (in_porcelain && *first && header_msg) {
1350 printf("%s\n", header_msg);
1351 *first = 0;
1352 }
1353 printf(fmt, name);
1354 }
1355
1356 int refresh_index(struct index_state *istate, unsigned int flags,
1357 const struct pathspec *pathspec,
1358 char *seen, const char *header_msg)
1359 {
1360 int i;
1361 int has_errors = 0;
1362 int really = (flags & REFRESH_REALLY) != 0;
1363 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1364 int quiet = (flags & REFRESH_QUIET) != 0;
1365 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1366 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1367 int first = 1;
1368 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1369 unsigned int options = (CE_MATCH_REFRESH |
1370 (really ? CE_MATCH_IGNORE_VALID : 0) |
1371 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1372 const char *modified_fmt;
1373 const char *deleted_fmt;
1374 const char *typechange_fmt;
1375 const char *added_fmt;
1376 const char *unmerged_fmt;
1377
1378 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1379 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1380 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1381 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1382 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1383 for (i = 0; i < istate->cache_nr; i++) {
1384 struct cache_entry *ce, *new;
1385 int cache_errno = 0;
1386 int changed = 0;
1387 int filtered = 0;
1388
1389 ce = istate->cache[i];
1390 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1391 continue;
1392
1393 if (pathspec && !ce_path_match(ce, pathspec, seen))
1394 filtered = 1;
1395
1396 if (ce_stage(ce)) {
1397 while ((i < istate->cache_nr) &&
1398 ! strcmp(istate->cache[i]->name, ce->name))
1399 i++;
1400 i--;
1401 if (allow_unmerged)
1402 continue;
1403 if (!filtered)
1404 show_file(unmerged_fmt, ce->name, in_porcelain,
1405 &first, header_msg);
1406 has_errors = 1;
1407 continue;
1408 }
1409
1410 if (filtered)
1411 continue;
1412
1413 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1414 if (new == ce)
1415 continue;
1416 if (!new) {
1417 const char *fmt;
1418
1419 if (really && cache_errno == EINVAL) {
1420 /* If we are doing --really-refresh that
1421 * means the index is not valid anymore.
1422 */
1423 ce->ce_flags &= ~CE_VALID;
1424 ce->ce_flags |= CE_UPDATE_IN_BASE;
1425 mark_fsmonitor_invalid(istate, ce);
1426 istate->cache_changed |= CE_ENTRY_CHANGED;
1427 }
1428 if (quiet)
1429 continue;
1430
1431 if (cache_errno == ENOENT)
1432 fmt = deleted_fmt;
1433 else if (ce_intent_to_add(ce))
1434 fmt = added_fmt; /* must be before other checks */
1435 else if (changed & TYPE_CHANGED)
1436 fmt = typechange_fmt;
1437 else
1438 fmt = modified_fmt;
1439 show_file(fmt,
1440 ce->name, in_porcelain, &first, header_msg);
1441 has_errors = 1;
1442 continue;
1443 }
1444
1445 replace_index_entry(istate, i, new);
1446 }
1447 return has_errors;
1448 }
1449
1450 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1451 unsigned int options)
1452 {
1453 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1454 }
1455
1456
1457 /*****************************************************************
1458 * Index File I/O
1459 *****************************************************************/
1460
1461 #define INDEX_FORMAT_DEFAULT 3
1462
1463 static unsigned int get_index_format_default(void)
1464 {
1465 char *envversion = getenv("GIT_INDEX_VERSION");
1466 char *endp;
1467 int value;
1468 unsigned int version = INDEX_FORMAT_DEFAULT;
1469
1470 if (!envversion) {
1471 if (!git_config_get_int("index.version", &value))
1472 version = value;
1473 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1474 warning(_("index.version set, but the value is invalid.\n"
1475 "Using version %i"), INDEX_FORMAT_DEFAULT);
1476 return INDEX_FORMAT_DEFAULT;
1477 }
1478 return version;
1479 }
1480
1481 version = strtoul(envversion, &endp, 10);
1482 if (*endp ||
1483 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1484 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1485 "Using version %i"), INDEX_FORMAT_DEFAULT);
1486 version = INDEX_FORMAT_DEFAULT;
1487 }
1488 return version;
1489 }
1490
1491 /*
1492 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1493 * Again - this is just a (very strong in practice) heuristic that
1494 * the inode hasn't changed.
1495 *
1496 * We save the fields in big-endian order to allow using the
1497 * index file over NFS transparently.
1498 */
1499 struct ondisk_cache_entry {
1500 struct cache_time ctime;
1501 struct cache_time mtime;
1502 uint32_t dev;
1503 uint32_t ino;
1504 uint32_t mode;
1505 uint32_t uid;
1506 uint32_t gid;
1507 uint32_t size;
1508 unsigned char sha1[20];
1509 uint16_t flags;
1510 char name[FLEX_ARRAY]; /* more */
1511 };
1512
1513 /*
1514 * This struct is used when CE_EXTENDED bit is 1
1515 * The struct must match ondisk_cache_entry exactly from
1516 * ctime till flags
1517 */
1518 struct ondisk_cache_entry_extended {
1519 struct cache_time ctime;
1520 struct cache_time mtime;
1521 uint32_t dev;
1522 uint32_t ino;
1523 uint32_t mode;
1524 uint32_t uid;
1525 uint32_t gid;
1526 uint32_t size;
1527 unsigned char sha1[20];
1528 uint16_t flags;
1529 uint16_t flags2;
1530 char name[FLEX_ARRAY]; /* more */
1531 };
1532
1533 /* These are only used for v3 or lower */
1534 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1535 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1536 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1537 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1538 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1539 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1540 ondisk_cache_entry_size(ce_namelen(ce)))
1541
1542 /* Allow fsck to force verification of the index checksum. */
1543 int verify_index_checksum;
1544
1545 /* Allow fsck to force verification of the cache entry order. */
1546 int verify_ce_order;
1547
1548 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1549 {
1550 git_SHA_CTX c;
1551 unsigned char sha1[20];
1552 int hdr_version;
1553
1554 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1555 return error("bad signature");
1556 hdr_version = ntohl(hdr->hdr_version);
1557 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1558 return error("bad index version %d", hdr_version);
1559
1560 if (!verify_index_checksum)
1561 return 0;
1562
1563 git_SHA1_Init(&c);
1564 git_SHA1_Update(&c, hdr, size - 20);
1565 git_SHA1_Final(sha1, &c);
1566 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1567 return error("bad index file sha1 signature");
1568 return 0;
1569 }
1570
1571 static int read_index_extension(struct index_state *istate,
1572 const char *ext, void *data, unsigned long sz)
1573 {
1574 switch (CACHE_EXT(ext)) {
1575 case CACHE_EXT_TREE:
1576 istate->cache_tree = cache_tree_read(data, sz);
1577 break;
1578 case CACHE_EXT_RESOLVE_UNDO:
1579 istate->resolve_undo = resolve_undo_read(data, sz);
1580 break;
1581 case CACHE_EXT_LINK:
1582 if (read_link_extension(istate, data, sz))
1583 return -1;
1584 break;
1585 case CACHE_EXT_UNTRACKED:
1586 istate->untracked = read_untracked_extension(data, sz);
1587 break;
1588 case CACHE_EXT_FSMONITOR:
1589 read_fsmonitor_extension(istate, data, sz);
1590 break;
1591 default:
1592 if (*ext < 'A' || 'Z' < *ext)
1593 return error("index uses %.4s extension, which we do not understand",
1594 ext);
1595 fprintf(stderr, "ignoring %.4s extension\n", ext);
1596 break;
1597 }
1598 return 0;
1599 }
1600
1601 int hold_locked_index(struct lock_file *lk, int lock_flags)
1602 {
1603 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1604 }
1605
1606 int read_index(struct index_state *istate)
1607 {
1608 return read_index_from(istate, get_index_file());
1609 }
1610
1611 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1612 unsigned int flags,
1613 const char *name,
1614 size_t len)
1615 {
1616 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1617
1618 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1619 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1620 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1621 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1622 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1623 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1624 ce->ce_mode = get_be32(&ondisk->mode);
1625 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1626 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1627 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1628 ce->ce_flags = flags & ~CE_NAMEMASK;
1629 ce->ce_namelen = len;
1630 ce->index = 0;
1631 hashcpy(ce->oid.hash, ondisk->sha1);
1632 memcpy(ce->name, name, len);
1633 ce->name[len] = '\0';
1634 return ce;
1635 }
1636
1637 /*
1638 * Adjacent cache entries tend to share the leading paths, so it makes
1639 * sense to only store the differences in later entries. In the v4
1640 * on-disk format of the index, each on-disk cache entry stores the
1641 * number of bytes to be stripped from the end of the previous name,
1642 * and the bytes to append to the result, to come up with its name.
1643 */
1644 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1645 {
1646 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1647 size_t len = decode_varint(&cp);
1648
1649 if (name->len < len)
1650 die("malformed name field in the index");
1651 strbuf_remove(name, name->len - len, len);
1652 for (ep = cp; *ep; ep++)
1653 ; /* find the end */
1654 strbuf_add(name, cp, ep - cp);
1655 return (const char *)ep + 1 - cp_;
1656 }
1657
1658 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1659 unsigned long *ent_size,
1660 struct strbuf *previous_name)
1661 {
1662 struct cache_entry *ce;
1663 size_t len;
1664 const char *name;
1665 unsigned int flags;
1666
1667 /* On-disk flags are just 16 bits */
1668 flags = get_be16(&ondisk->flags);
1669 len = flags & CE_NAMEMASK;
1670
1671 if (flags & CE_EXTENDED) {
1672 struct ondisk_cache_entry_extended *ondisk2;
1673 int extended_flags;
1674 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1675 extended_flags = get_be16(&ondisk2->flags2) << 16;
1676 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1677 if (extended_flags & ~CE_EXTENDED_FLAGS)
1678 die("Unknown index entry format %08x", extended_flags);
1679 flags |= extended_flags;
1680 name = ondisk2->name;
1681 }
1682 else
1683 name = ondisk->name;
1684
1685 if (!previous_name) {
1686 /* v3 and earlier */
1687 if (len == CE_NAMEMASK)
1688 len = strlen(name);
1689 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1690
1691 *ent_size = ondisk_ce_size(ce);
1692 } else {
1693 unsigned long consumed;
1694 consumed = expand_name_field(previous_name, name);
1695 ce = cache_entry_from_ondisk(ondisk, flags,
1696 previous_name->buf,
1697 previous_name->len);
1698
1699 *ent_size = (name - ((char *)ondisk)) + consumed;
1700 }
1701 return ce;
1702 }
1703
1704 static void check_ce_order(struct index_state *istate)
1705 {
1706 unsigned int i;
1707
1708 if (!verify_ce_order)
1709 return;
1710
1711 for (i = 1; i < istate->cache_nr; i++) {
1712 struct cache_entry *ce = istate->cache[i - 1];
1713 struct cache_entry *next_ce = istate->cache[i];
1714 int name_compare = strcmp(ce->name, next_ce->name);
1715
1716 if (0 < name_compare)
1717 die("unordered stage entries in index");
1718 if (!name_compare) {
1719 if (!ce_stage(ce))
1720 die("multiple stage entries for merged file '%s'",
1721 ce->name);
1722 if (ce_stage(ce) > ce_stage(next_ce))
1723 die("unordered stage entries for '%s'",
1724 ce->name);
1725 }
1726 }
1727 }
1728
1729 static void tweak_untracked_cache(struct index_state *istate)
1730 {
1731 switch (git_config_get_untracked_cache()) {
1732 case -1: /* keep: do nothing */
1733 break;
1734 case 0: /* false */
1735 remove_untracked_cache(istate);
1736 break;
1737 case 1: /* true */
1738 add_untracked_cache(istate);
1739 break;
1740 default: /* unknown value: do nothing */
1741 break;
1742 }
1743 }
1744
1745 static void tweak_split_index(struct index_state *istate)
1746 {
1747 switch (git_config_get_split_index()) {
1748 case -1: /* unset: do nothing */
1749 break;
1750 case 0: /* false */
1751 remove_split_index(istate);
1752 break;
1753 case 1: /* true */
1754 add_split_index(istate);
1755 break;
1756 default: /* unknown value: do nothing */
1757 break;
1758 }
1759 }
1760
1761 static void post_read_index_from(struct index_state *istate)
1762 {
1763 check_ce_order(istate);
1764 tweak_untracked_cache(istate);
1765 tweak_split_index(istate);
1766 tweak_fsmonitor(istate);
1767 }
1768
1769 /* remember to discard_cache() before reading a different cache! */
1770 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1771 {
1772 int fd, i;
1773 struct stat st;
1774 unsigned long src_offset;
1775 struct cache_header *hdr;
1776 void *mmap;
1777 size_t mmap_size;
1778 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1779
1780 if (istate->initialized)
1781 return istate->cache_nr;
1782
1783 istate->timestamp.sec = 0;
1784 istate->timestamp.nsec = 0;
1785 fd = open(path, O_RDONLY);
1786 if (fd < 0) {
1787 if (!must_exist && errno == ENOENT)
1788 return 0;
1789 die_errno("%s: index file open failed", path);
1790 }
1791
1792 if (fstat(fd, &st))
1793 die_errno("cannot stat the open index");
1794
1795 mmap_size = xsize_t(st.st_size);
1796 if (mmap_size < sizeof(struct cache_header) + 20)
1797 die("index file smaller than expected");
1798
1799 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1800 if (mmap == MAP_FAILED)
1801 die_errno("unable to map index file");
1802 close(fd);
1803
1804 hdr = mmap;
1805 if (verify_hdr(hdr, mmap_size) < 0)
1806 goto unmap;
1807
1808 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1809 istate->version = ntohl(hdr->hdr_version);
1810 istate->cache_nr = ntohl(hdr->hdr_entries);
1811 istate->cache_alloc = alloc_nr(istate->cache_nr);
1812 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1813 istate->initialized = 1;
1814
1815 if (istate->version == 4)
1816 previous_name = &previous_name_buf;
1817 else
1818 previous_name = NULL;
1819
1820 src_offset = sizeof(*hdr);
1821 for (i = 0; i < istate->cache_nr; i++) {
1822 struct ondisk_cache_entry *disk_ce;
1823 struct cache_entry *ce;
1824 unsigned long consumed;
1825
1826 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1827 ce = create_from_disk(disk_ce, &consumed, previous_name);
1828 set_index_entry(istate, i, ce);
1829
1830 src_offset += consumed;
1831 }
1832 strbuf_release(&previous_name_buf);
1833 istate->timestamp.sec = st.st_mtime;
1834 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1835
1836 while (src_offset <= mmap_size - 20 - 8) {
1837 /* After an array of active_nr index entries,
1838 * there can be arbitrary number of extended
1839 * sections, each of which is prefixed with
1840 * extension name (4-byte) and section length
1841 * in 4-byte network byte order.
1842 */
1843 uint32_t extsize;
1844 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1845 extsize = ntohl(extsize);
1846 if (read_index_extension(istate,
1847 (const char *) mmap + src_offset,
1848 (char *) mmap + src_offset + 8,
1849 extsize) < 0)
1850 goto unmap;
1851 src_offset += 8;
1852 src_offset += extsize;
1853 }
1854 munmap(mmap, mmap_size);
1855 return istate->cache_nr;
1856
1857 unmap:
1858 munmap(mmap, mmap_size);
1859 die("index file corrupt");
1860 }
1861
1862 /*
1863 * Signal that the shared index is used by updating its mtime.
1864 *
1865 * This way, shared index can be removed if they have not been used
1866 * for some time.
1867 */
1868 static void freshen_shared_index(char *base_sha1_hex, int warn)
1869 {
1870 char *shared_index = git_pathdup("sharedindex.%s", base_sha1_hex);
1871 if (!check_and_freshen_file(shared_index, 1) && warn)
1872 warning("could not freshen shared index '%s'", shared_index);
1873 free(shared_index);
1874 }
1875
1876 int read_index_from(struct index_state *istate, const char *path)
1877 {
1878 struct split_index *split_index;
1879 int ret;
1880 char *base_sha1_hex;
1881 const char *base_path;
1882
1883 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1884 if (istate->initialized)
1885 return istate->cache_nr;
1886
1887 ret = do_read_index(istate, path, 0);
1888
1889 split_index = istate->split_index;
1890 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1891 post_read_index_from(istate);
1892 return ret;
1893 }
1894
1895 if (split_index->base)
1896 discard_index(split_index->base);
1897 else
1898 split_index->base = xcalloc(1, sizeof(*split_index->base));
1899
1900 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1901 base_path = git_path("sharedindex.%s", base_sha1_hex);
1902 ret = do_read_index(split_index->base, base_path, 1);
1903 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1904 die("broken index, expect %s in %s, got %s",
1905 base_sha1_hex, base_path,
1906 sha1_to_hex(split_index->base->sha1));
1907
1908 freshen_shared_index(base_sha1_hex, 0);
1909 merge_base_index(istate);
1910 post_read_index_from(istate);
1911 return ret;
1912 }
1913
1914 int is_index_unborn(struct index_state *istate)
1915 {
1916 return (!istate->cache_nr && !istate->timestamp.sec);
1917 }
1918
1919 int discard_index(struct index_state *istate)
1920 {
1921 int i;
1922
1923 for (i = 0; i < istate->cache_nr; i++) {
1924 if (istate->cache[i]->index &&
1925 istate->split_index &&
1926 istate->split_index->base &&
1927 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1928 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1929 continue;
1930 free(istate->cache[i]);
1931 }
1932 resolve_undo_clear_index(istate);
1933 istate->cache_nr = 0;
1934 istate->cache_changed = 0;
1935 istate->timestamp.sec = 0;
1936 istate->timestamp.nsec = 0;
1937 free_name_hash(istate);
1938 cache_tree_free(&(istate->cache_tree));
1939 istate->initialized = 0;
1940 FREE_AND_NULL(istate->cache);
1941 istate->cache_alloc = 0;
1942 discard_split_index(istate);
1943 free_untracked_cache(istate->untracked);
1944 istate->untracked = NULL;
1945 return 0;
1946 }
1947
1948 int unmerged_index(const struct index_state *istate)
1949 {
1950 int i;
1951 for (i = 0; i < istate->cache_nr; i++) {
1952 if (ce_stage(istate->cache[i]))
1953 return 1;
1954 }
1955 return 0;
1956 }
1957
1958 #define WRITE_BUFFER_SIZE 8192
1959 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1960 static unsigned long write_buffer_len;
1961
1962 static int ce_write_flush(git_SHA_CTX *context, int fd)
1963 {
1964 unsigned int buffered = write_buffer_len;
1965 if (buffered) {
1966 git_SHA1_Update(context, write_buffer, buffered);
1967 if (write_in_full(fd, write_buffer, buffered) < 0)
1968 return -1;
1969 write_buffer_len = 0;
1970 }
1971 return 0;
1972 }
1973
1974 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1975 {
1976 while (len) {
1977 unsigned int buffered = write_buffer_len;
1978 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1979 if (partial > len)
1980 partial = len;
1981 memcpy(write_buffer + buffered, data, partial);
1982 buffered += partial;
1983 if (buffered == WRITE_BUFFER_SIZE) {
1984 write_buffer_len = buffered;
1985 if (ce_write_flush(context, fd))
1986 return -1;
1987 buffered = 0;
1988 }
1989 write_buffer_len = buffered;
1990 len -= partial;
1991 data = (char *) data + partial;
1992 }
1993 return 0;
1994 }
1995
1996 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1997 unsigned int ext, unsigned int sz)
1998 {
1999 ext = htonl(ext);
2000 sz = htonl(sz);
2001 return ((ce_write(context, fd, &ext, 4) < 0) ||
2002 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2003 }
2004
2005 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
2006 {
2007 unsigned int left = write_buffer_len;
2008
2009 if (left) {
2010 write_buffer_len = 0;
2011 git_SHA1_Update(context, write_buffer, left);
2012 }
2013
2014 /* Flush first if not enough space for SHA1 signature */
2015 if (left + 20 > WRITE_BUFFER_SIZE) {
2016 if (write_in_full(fd, write_buffer, left) < 0)
2017 return -1;
2018 left = 0;
2019 }
2020
2021 /* Append the SHA1 signature at the end */
2022 git_SHA1_Final(write_buffer + left, context);
2023 hashcpy(sha1, write_buffer + left);
2024 left += 20;
2025 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2026 }
2027
2028 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2029 {
2030 /*
2031 * The only thing we care about in this function is to smudge the
2032 * falsely clean entry due to touch-update-touch race, so we leave
2033 * everything else as they are. We are called for entries whose
2034 * ce_stat_data.sd_mtime match the index file mtime.
2035 *
2036 * Note that this actually does not do much for gitlinks, for
2037 * which ce_match_stat_basic() always goes to the actual
2038 * contents. The caller checks with is_racy_timestamp() which
2039 * always says "no" for gitlinks, so we are not called for them ;-)
2040 */
2041 struct stat st;
2042
2043 if (lstat(ce->name, &st) < 0)
2044 return;
2045 if (ce_match_stat_basic(ce, &st))
2046 return;
2047 if (ce_modified_check_fs(ce, &st)) {
2048 /* This is "racily clean"; smudge it. Note that this
2049 * is a tricky code. At first glance, it may appear
2050 * that it can break with this sequence:
2051 *
2052 * $ echo xyzzy >frotz
2053 * $ git-update-index --add frotz
2054 * $ : >frotz
2055 * $ sleep 3
2056 * $ echo filfre >nitfol
2057 * $ git-update-index --add nitfol
2058 *
2059 * but it does not. When the second update-index runs,
2060 * it notices that the entry "frotz" has the same timestamp
2061 * as index, and if we were to smudge it by resetting its
2062 * size to zero here, then the object name recorded
2063 * in index is the 6-byte file but the cached stat information
2064 * becomes zero --- which would then match what we would
2065 * obtain from the filesystem next time we stat("frotz").
2066 *
2067 * However, the second update-index, before calling
2068 * this function, notices that the cached size is 6
2069 * bytes and what is on the filesystem is an empty
2070 * file, and never calls us, so the cached size information
2071 * for "frotz" stays 6 which does not match the filesystem.
2072 */
2073 ce->ce_stat_data.sd_size = 0;
2074 }
2075 }
2076
2077 /* Copy miscellaneous fields but not the name */
2078 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2079 struct cache_entry *ce)
2080 {
2081 short flags;
2082
2083 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2084 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2085 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2086 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2087 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2088 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2089 ondisk->mode = htonl(ce->ce_mode);
2090 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2091 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2092 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2093 hashcpy(ondisk->sha1, ce->oid.hash);
2094
2095 flags = ce->ce_flags & ~CE_NAMEMASK;
2096 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2097 ondisk->flags = htons(flags);
2098 if (ce->ce_flags & CE_EXTENDED) {
2099 struct ondisk_cache_entry_extended *ondisk2;
2100 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2101 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2102 }
2103 }
2104
2105 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
2106 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2107 {
2108 int size;
2109 int saved_namelen = saved_namelen; /* compiler workaround */
2110 int result;
2111 static unsigned char padding[8] = { 0x00 };
2112
2113 if (ce->ce_flags & CE_STRIP_NAME) {
2114 saved_namelen = ce_namelen(ce);
2115 ce->ce_namelen = 0;
2116 }
2117
2118 if (ce->ce_flags & CE_EXTENDED)
2119 size = offsetof(struct ondisk_cache_entry_extended, name);
2120 else
2121 size = offsetof(struct ondisk_cache_entry, name);
2122
2123 if (!previous_name) {
2124 int len = ce_namelen(ce);
2125 copy_cache_entry_to_ondisk(ondisk, ce);
2126 result = ce_write(c, fd, ondisk, size);
2127 if (!result)
2128 result = ce_write(c, fd, ce->name, len);
2129 if (!result)
2130 result = ce_write(c, fd, padding, align_padding_size(size, len));
2131 } else {
2132 int common, to_remove, prefix_size;
2133 unsigned char to_remove_vi[16];
2134 for (common = 0;
2135 (ce->name[common] &&
2136 common < previous_name->len &&
2137 ce->name[common] == previous_name->buf[common]);
2138 common++)
2139 ; /* still matching */
2140 to_remove = previous_name->len - common;
2141 prefix_size = encode_varint(to_remove, to_remove_vi);
2142
2143 copy_cache_entry_to_ondisk(ondisk, ce);
2144 result = ce_write(c, fd, ondisk, size);
2145 if (!result)
2146 result = ce_write(c, fd, to_remove_vi, prefix_size);
2147 if (!result)
2148 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2149 if (!result)
2150 result = ce_write(c, fd, padding, 1);
2151
2152 strbuf_splice(previous_name, common, to_remove,
2153 ce->name + common, ce_namelen(ce) - common);
2154 }
2155 if (ce->ce_flags & CE_STRIP_NAME) {
2156 ce->ce_namelen = saved_namelen;
2157 ce->ce_flags &= ~CE_STRIP_NAME;
2158 }
2159
2160 return result;
2161 }
2162
2163 /*
2164 * This function verifies if index_state has the correct sha1 of the
2165 * index file. Don't die if we have any other failure, just return 0.
2166 */
2167 static int verify_index_from(const struct index_state *istate, const char *path)
2168 {
2169 int fd;
2170 ssize_t n;
2171 struct stat st;
2172 unsigned char sha1[20];
2173
2174 if (!istate->initialized)
2175 return 0;
2176
2177 fd = open(path, O_RDONLY);
2178 if (fd < 0)
2179 return 0;
2180
2181 if (fstat(fd, &st))
2182 goto out;
2183
2184 if (st.st_size < sizeof(struct cache_header) + 20)
2185 goto out;
2186
2187 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2188 if (n != 20)
2189 goto out;
2190
2191 if (hashcmp(istate->sha1, sha1))
2192 goto out;
2193
2194 close(fd);
2195 return 1;
2196
2197 out:
2198 close(fd);
2199 return 0;
2200 }
2201
2202 static int verify_index(const struct index_state *istate)
2203 {
2204 return verify_index_from(istate, get_index_file());
2205 }
2206
2207 static int has_racy_timestamp(struct index_state *istate)
2208 {
2209 int entries = istate->cache_nr;
2210 int i;
2211
2212 for (i = 0; i < entries; i++) {
2213 struct cache_entry *ce = istate->cache[i];
2214 if (is_racy_timestamp(istate, ce))
2215 return 1;
2216 }
2217 return 0;
2218 }
2219
2220 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2221 {
2222 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2223 verify_index(istate))
2224 write_locked_index(istate, lockfile, COMMIT_LOCK);
2225 else
2226 rollback_lock_file(lockfile);
2227 }
2228
2229 /*
2230 * On success, `tempfile` is closed. If it is the temporary file
2231 * of a `struct lock_file`, we will therefore effectively perform
2232 * a 'close_lock_file_gently()`. Since that is an implementation
2233 * detail of lockfiles, callers of `do_write_index()` should not
2234 * rely on it.
2235 */
2236 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2237 int strip_extensions)
2238 {
2239 int newfd = tempfile->fd;
2240 git_SHA_CTX c;
2241 struct cache_header hdr;
2242 int i, err = 0, removed, extended, hdr_version;
2243 struct cache_entry **cache = istate->cache;
2244 int entries = istate->cache_nr;
2245 struct stat st;
2246 struct ondisk_cache_entry_extended ondisk;
2247 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2248 int drop_cache_tree = 0;
2249
2250 for (i = removed = extended = 0; i < entries; i++) {
2251 if (cache[i]->ce_flags & CE_REMOVE)
2252 removed++;
2253
2254 /* reduce extended entries if possible */
2255 cache[i]->ce_flags &= ~CE_EXTENDED;
2256 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2257 extended++;
2258 cache[i]->ce_flags |= CE_EXTENDED;
2259 }
2260 }
2261
2262 if (!istate->version) {
2263 istate->version = get_index_format_default();
2264 if (getenv("GIT_TEST_SPLIT_INDEX"))
2265 init_split_index(istate);
2266 }
2267
2268 /* demote version 3 to version 2 when the latter suffices */
2269 if (istate->version == 3 || istate->version == 2)
2270 istate->version = extended ? 3 : 2;
2271
2272 hdr_version = istate->version;
2273
2274 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2275 hdr.hdr_version = htonl(hdr_version);
2276 hdr.hdr_entries = htonl(entries - removed);
2277
2278 git_SHA1_Init(&c);
2279 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2280 return -1;
2281
2282 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2283
2284 for (i = 0; i < entries; i++) {
2285 struct cache_entry *ce = cache[i];
2286 if (ce->ce_flags & CE_REMOVE)
2287 continue;
2288 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2289 ce_smudge_racily_clean_entry(ce);
2290 if (is_null_oid(&ce->oid)) {
2291 static const char msg[] = "cache entry has null sha1: %s";
2292 static int allow = -1;
2293
2294 if (allow < 0)
2295 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2296 if (allow)
2297 warning(msg, ce->name);
2298 else
2299 err = error(msg, ce->name);
2300
2301 drop_cache_tree = 1;
2302 }
2303 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2304 err = -1;
2305
2306 if (err)
2307 break;
2308 }
2309 strbuf_release(&previous_name_buf);
2310
2311 if (err)
2312 return err;
2313
2314 /* Write extension data here */
2315 if (!strip_extensions && istate->split_index) {
2316 struct strbuf sb = STRBUF_INIT;
2317
2318 err = write_link_extension(&sb, istate) < 0 ||
2319 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2320 sb.len) < 0 ||
2321 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2322 strbuf_release(&sb);
2323 if (err)
2324 return -1;
2325 }
2326 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2327 struct strbuf sb = STRBUF_INIT;
2328
2329 cache_tree_write(&sb, istate->cache_tree);
2330 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2331 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2332 strbuf_release(&sb);
2333 if (err)
2334 return -1;
2335 }
2336 if (!strip_extensions && istate->resolve_undo) {
2337 struct strbuf sb = STRBUF_INIT;
2338
2339 resolve_undo_write(&sb, istate->resolve_undo);
2340 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2341 sb.len) < 0
2342 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2343 strbuf_release(&sb);
2344 if (err)
2345 return -1;
2346 }
2347 if (!strip_extensions && istate->untracked) {
2348 struct strbuf sb = STRBUF_INIT;
2349
2350 write_untracked_extension(&sb, istate->untracked);
2351 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2352 sb.len) < 0 ||
2353 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2354 strbuf_release(&sb);
2355 if (err)
2356 return -1;
2357 }
2358 if (!strip_extensions && istate->fsmonitor_last_update) {
2359 struct strbuf sb = STRBUF_INIT;
2360
2361 write_fsmonitor_extension(&sb, istate);
2362 err = write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2363 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2364 strbuf_release(&sb);
2365 if (err)
2366 return -1;
2367 }
2368
2369 if (ce_flush(&c, newfd, istate->sha1))
2370 return -1;
2371 if (close_tempfile_gently(tempfile)) {
2372 error(_("could not close '%s'"), tempfile->filename.buf);
2373 return -1;
2374 }
2375 if (stat(tempfile->filename.buf, &st))
2376 return -1;
2377 istate->timestamp.sec = (unsigned int)st.st_mtime;
2378 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2379 return 0;
2380 }
2381
2382 void set_alternate_index_output(const char *name)
2383 {
2384 alternate_index_output = name;
2385 }
2386
2387 static int commit_locked_index(struct lock_file *lk)
2388 {
2389 if (alternate_index_output)
2390 return commit_lock_file_to(lk, alternate_index_output);
2391 else
2392 return commit_lock_file(lk);
2393 }
2394
2395 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2396 unsigned flags)
2397 {
2398 int ret = do_write_index(istate, lock->tempfile, 0);
2399 if (ret)
2400 return ret;
2401 if (flags & COMMIT_LOCK)
2402 return commit_locked_index(lock);
2403 return close_lock_file_gently(lock);
2404 }
2405
2406 static int write_split_index(struct index_state *istate,
2407 struct lock_file *lock,
2408 unsigned flags)
2409 {
2410 int ret;
2411 prepare_to_write_split_index(istate);
2412 ret = do_write_locked_index(istate, lock, flags);
2413 finish_writing_split_index(istate);
2414 return ret;
2415 }
2416
2417 static const char *shared_index_expire = "2.weeks.ago";
2418
2419 static unsigned long get_shared_index_expire_date(void)
2420 {
2421 static unsigned long shared_index_expire_date;
2422 static int shared_index_expire_date_prepared;
2423
2424 if (!shared_index_expire_date_prepared) {
2425 git_config_get_expiry("splitindex.sharedindexexpire",
2426 &shared_index_expire);
2427 shared_index_expire_date = approxidate(shared_index_expire);
2428 shared_index_expire_date_prepared = 1;
2429 }
2430
2431 return shared_index_expire_date;
2432 }
2433
2434 static int should_delete_shared_index(const char *shared_index_path)
2435 {
2436 struct stat st;
2437 unsigned long expiration;
2438
2439 /* Check timestamp */
2440 expiration = get_shared_index_expire_date();
2441 if (!expiration)
2442 return 0;
2443 if (stat(shared_index_path, &st))
2444 return error_errno(_("could not stat '%s'"), shared_index_path);
2445 if (st.st_mtime > expiration)
2446 return 0;
2447
2448 return 1;
2449 }
2450
2451 static int clean_shared_index_files(const char *current_hex)
2452 {
2453 struct dirent *de;
2454 DIR *dir = opendir(get_git_dir());
2455
2456 if (!dir)
2457 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2458
2459 while ((de = readdir(dir)) != NULL) {
2460 const char *sha1_hex;
2461 const char *shared_index_path;
2462 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2463 continue;
2464 if (!strcmp(sha1_hex, current_hex))
2465 continue;
2466 shared_index_path = git_path("%s", de->d_name);
2467 if (should_delete_shared_index(shared_index_path) > 0 &&
2468 unlink(shared_index_path))
2469 warning_errno(_("unable to unlink: %s"), shared_index_path);
2470 }
2471 closedir(dir);
2472
2473 return 0;
2474 }
2475
2476 static int write_shared_index(struct index_state *istate,
2477 struct lock_file *lock, unsigned flags)
2478 {
2479 struct tempfile *temp;
2480 struct split_index *si = istate->split_index;
2481 int ret;
2482
2483 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2484 if (!temp) {
2485 hashclr(si->base_sha1);
2486 return do_write_locked_index(istate, lock, flags);
2487 }
2488 move_cache_to_base_index(istate);
2489 ret = do_write_index(si->base, temp, 1);
2490 if (ret) {
2491 delete_tempfile(&temp);
2492 return ret;
2493 }
2494 ret = adjust_shared_perm(get_tempfile_path(temp));
2495 if (ret) {
2496 int save_errno = errno;
2497 error("cannot fix permission bits on %s", get_tempfile_path(temp));
2498 delete_tempfile(&temp);
2499 errno = save_errno;
2500 return ret;
2501 }
2502 ret = rename_tempfile(&temp,
2503 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2504 if (!ret) {
2505 hashcpy(si->base_sha1, si->base->sha1);
2506 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2507 }
2508
2509 return ret;
2510 }
2511
2512 static const int default_max_percent_split_change = 20;
2513
2514 static int too_many_not_shared_entries(struct index_state *istate)
2515 {
2516 int i, not_shared = 0;
2517 int max_split = git_config_get_max_percent_split_change();
2518
2519 switch (max_split) {
2520 case -1:
2521 /* not or badly configured: use the default value */
2522 max_split = default_max_percent_split_change;
2523 break;
2524 case 0:
2525 return 1; /* 0% means always write a new shared index */
2526 case 100:
2527 return 0; /* 100% means never write a new shared index */
2528 default:
2529 break; /* just use the configured value */
2530 }
2531
2532 /* Count not shared entries */
2533 for (i = 0; i < istate->cache_nr; i++) {
2534 struct cache_entry *ce = istate->cache[i];
2535 if (!ce->index)
2536 not_shared++;
2537 }
2538
2539 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2540 }
2541
2542 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2543 unsigned flags)
2544 {
2545 int new_shared_index, ret;
2546 struct split_index *si = istate->split_index;
2547
2548 if (istate->fsmonitor_last_update)
2549 fill_fsmonitor_bitmap(istate);
2550
2551 if (!si || alternate_index_output ||
2552 (istate->cache_changed & ~EXTMASK)) {
2553 if (si)
2554 hashclr(si->base_sha1);
2555 ret = do_write_locked_index(istate, lock, flags);
2556 goto out;
2557 }
2558
2559 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2560 int v = si->base_sha1[0];
2561 if ((v & 15) < 6)
2562 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2563 }
2564 if (too_many_not_shared_entries(istate))
2565 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2566
2567 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2568
2569 if (new_shared_index) {
2570 ret = write_shared_index(istate, lock, flags);
2571 if (ret)
2572 goto out;
2573 }
2574
2575 ret = write_split_index(istate, lock, flags);
2576
2577 /* Freshen the shared index only if the split-index was written */
2578 if (!ret && !new_shared_index)
2579 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2580
2581 out:
2582 if (flags & COMMIT_LOCK)
2583 rollback_lock_file(lock);
2584 return ret;
2585 }
2586
2587 /*
2588 * Read the index file that is potentially unmerged into given
2589 * index_state, dropping any unmerged entries. Returns true if
2590 * the index is unmerged. Callers who want to refuse to work
2591 * from an unmerged state can call this and check its return value,
2592 * instead of calling read_cache().
2593 */
2594 int read_index_unmerged(struct index_state *istate)
2595 {
2596 int i;
2597 int unmerged = 0;
2598
2599 read_index(istate);
2600 for (i = 0; i < istate->cache_nr; i++) {
2601 struct cache_entry *ce = istate->cache[i];
2602 struct cache_entry *new_ce;
2603 int size, len;
2604
2605 if (!ce_stage(ce))
2606 continue;
2607 unmerged = 1;
2608 len = ce_namelen(ce);
2609 size = cache_entry_size(len);
2610 new_ce = xcalloc(1, size);
2611 memcpy(new_ce->name, ce->name, len);
2612 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2613 new_ce->ce_namelen = len;
2614 new_ce->ce_mode = ce->ce_mode;
2615 if (add_index_entry(istate, new_ce, 0))
2616 return error("%s: cannot drop to stage #0",
2617 new_ce->name);
2618 }
2619 return unmerged;
2620 }
2621
2622 /*
2623 * Returns 1 if the path is an "other" path with respect to
2624 * the index; that is, the path is not mentioned in the index at all,
2625 * either as a file, a directory with some files in the index,
2626 * or as an unmerged entry.
2627 *
2628 * We helpfully remove a trailing "/" from directories so that
2629 * the output of read_directory can be used as-is.
2630 */
2631 int index_name_is_other(const struct index_state *istate, const char *name,
2632 int namelen)
2633 {
2634 int pos;
2635 if (namelen && name[namelen - 1] == '/')
2636 namelen--;
2637 pos = index_name_pos(istate, name, namelen);
2638 if (0 <= pos)
2639 return 0; /* exact match */
2640 pos = -pos - 1;
2641 if (pos < istate->cache_nr) {
2642 struct cache_entry *ce = istate->cache[pos];
2643 if (ce_namelen(ce) == namelen &&
2644 !memcmp(ce->name, name, namelen))
2645 return 0; /* Yup, this one exists unmerged */
2646 }
2647 return 1;
2648 }
2649
2650 void *read_blob_data_from_index(const struct index_state *istate,
2651 const char *path, unsigned long *size)
2652 {
2653 int pos, len;
2654 unsigned long sz;
2655 enum object_type type;
2656 void *data;
2657
2658 len = strlen(path);
2659 pos = index_name_pos(istate, path, len);
2660 if (pos < 0) {
2661 /*
2662 * We might be in the middle of a merge, in which
2663 * case we would read stage #2 (ours).
2664 */
2665 int i;
2666 for (i = -pos - 1;
2667 (pos < 0 && i < istate->cache_nr &&
2668 !strcmp(istate->cache[i]->name, path));
2669 i++)
2670 if (ce_stage(istate->cache[i]) == 2)
2671 pos = i;
2672 }
2673 if (pos < 0)
2674 return NULL;
2675 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2676 if (!data || type != OBJ_BLOB) {
2677 free(data);
2678 return NULL;
2679 }
2680 if (size)
2681 *size = sz;
2682 return data;
2683 }
2684
2685 void stat_validity_clear(struct stat_validity *sv)
2686 {
2687 FREE_AND_NULL(sv->sd);
2688 }
2689
2690 int stat_validity_check(struct stat_validity *sv, const char *path)
2691 {
2692 struct stat st;
2693
2694 if (stat(path, &st) < 0)
2695 return sv->sd == NULL;
2696 if (!sv->sd)
2697 return 0;
2698 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2699 }
2700
2701 void stat_validity_update(struct stat_validity *sv, int fd)
2702 {
2703 struct stat st;
2704
2705 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2706 stat_validity_clear(sv);
2707 else {
2708 if (!sv->sd)
2709 sv->sd = xcalloc(1, sizeof(struct stat_data));
2710 fill_stat_data(sv->sd, &st);
2711 }
2712 }
2713
2714 void move_index_extensions(struct index_state *dst, struct index_state *src)
2715 {
2716 dst->untracked = src->untracked;
2717 src->untracked = NULL;
2718 }