Merge branch 'bp/refresh-cache-ent-rehash-fix'
[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_entry = istate->cache[nr], *new_entry;
75 int namelen = strlen(new_name);
76
77 new_entry = xmalloc(cache_entry_size(namelen));
78 copy_cache_entry(new_entry, old_entry);
79 new_entry->ce_flags &= ~CE_HASHED;
80 new_entry->ce_namelen = namelen;
81 new_entry->index = 0;
82 memcpy(new_entry->name, new_name, namelen + 1);
83
84 cache_tree_invalidate_path(istate, old_entry->name);
85 untracked_cache_remove_from_index(istate, old_entry->name);
86 remove_index_entry_at(istate, nr);
87 add_index_entry(istate, new_entry, 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_entry;
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_entry = xcalloc(1, cache_entry_size(len));
627 memcpy(new_entry->name, alias->name, len);
628 copy_cache_entry(new_entry, ce);
629 save_or_free_index_entry(istate, ce);
630 return new_entry;
631 }
632
633 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
634 {
635 struct object_id oid;
636 if (write_object_file("", 0, blob_type, &oid))
637 die("cannot create an empty blob in the object database");
638 oidcpy(&ce->oid, &oid);
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 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1222 istate->cache_nr - pos - 1);
1223 set_index_entry(istate, pos, ce);
1224 istate->cache_changed |= CE_ENTRY_ADDED;
1225 return 0;
1226 }
1227
1228 /*
1229 * "refresh" does not calculate a new sha1 file or bring the
1230 * cache up-to-date for mode/content changes. But what it
1231 * _does_ do is to "re-match" the stat information of a file
1232 * with the cache, so that you can refresh the cache for a
1233 * file that hasn't been changed but where the stat entry is
1234 * out of date.
1235 *
1236 * For example, you'd want to do this after doing a "git-read-tree",
1237 * to link up the stat cache details with the proper files.
1238 */
1239 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1240 struct cache_entry *ce,
1241 unsigned int options, int *err,
1242 int *changed_ret)
1243 {
1244 struct stat st;
1245 struct cache_entry *updated;
1246 int changed, size;
1247 int refresh = options & CE_MATCH_REFRESH;
1248 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1249 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1250 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1251 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1252
1253 if (!refresh || ce_uptodate(ce))
1254 return ce;
1255
1256 if (!ignore_fsmonitor)
1257 refresh_fsmonitor(istate);
1258 /*
1259 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1260 * that the change to the work tree does not matter and told
1261 * us not to worry.
1262 */
1263 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1264 ce_mark_uptodate(ce);
1265 return ce;
1266 }
1267 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1268 ce_mark_uptodate(ce);
1269 return ce;
1270 }
1271 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1272 ce_mark_uptodate(ce);
1273 return ce;
1274 }
1275
1276 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1277 if (ignore_missing)
1278 return ce;
1279 if (err)
1280 *err = ENOENT;
1281 return NULL;
1282 }
1283
1284 if (lstat(ce->name, &st) < 0) {
1285 if (ignore_missing && errno == ENOENT)
1286 return ce;
1287 if (err)
1288 *err = errno;
1289 return NULL;
1290 }
1291
1292 changed = ie_match_stat(istate, ce, &st, options);
1293 if (changed_ret)
1294 *changed_ret = changed;
1295 if (!changed) {
1296 /*
1297 * The path is unchanged. If we were told to ignore
1298 * valid bit, then we did the actual stat check and
1299 * found that the entry is unmodified. If the entry
1300 * is not marked VALID, this is the place to mark it
1301 * valid again, under "assume unchanged" mode.
1302 */
1303 if (ignore_valid && assume_unchanged &&
1304 !(ce->ce_flags & CE_VALID))
1305 ; /* mark this one VALID again */
1306 else {
1307 /*
1308 * We do not mark the index itself "modified"
1309 * because CE_UPTODATE flag is in-core only;
1310 * we are not going to write this change out.
1311 */
1312 if (!S_ISGITLINK(ce->ce_mode)) {
1313 ce_mark_uptodate(ce);
1314 mark_fsmonitor_valid(ce);
1315 }
1316 return ce;
1317 }
1318 }
1319
1320 if (ie_modified(istate, ce, &st, options)) {
1321 if (err)
1322 *err = EINVAL;
1323 return NULL;
1324 }
1325
1326 size = ce_size(ce);
1327 updated = xmalloc(size);
1328 copy_cache_entry(updated, ce);
1329 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1330 fill_stat_cache_info(updated, &st);
1331 /*
1332 * If ignore_valid is not set, we should leave CE_VALID bit
1333 * alone. Otherwise, paths marked with --no-assume-unchanged
1334 * (i.e. things to be edited) will reacquire CE_VALID bit
1335 * automatically, which is not really what we want.
1336 */
1337 if (!ignore_valid && assume_unchanged &&
1338 !(ce->ce_flags & CE_VALID))
1339 updated->ce_flags &= ~CE_VALID;
1340
1341 /* istate->cache_changed is updated in the caller */
1342 return updated;
1343 }
1344
1345 static void show_file(const char * fmt, const char * name, int in_porcelain,
1346 int * first, const char *header_msg)
1347 {
1348 if (in_porcelain && *first && header_msg) {
1349 printf("%s\n", header_msg);
1350 *first = 0;
1351 }
1352 printf(fmt, name);
1353 }
1354
1355 int refresh_index(struct index_state *istate, unsigned int flags,
1356 const struct pathspec *pathspec,
1357 char *seen, const char *header_msg)
1358 {
1359 int i;
1360 int has_errors = 0;
1361 int really = (flags & REFRESH_REALLY) != 0;
1362 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1363 int quiet = (flags & REFRESH_QUIET) != 0;
1364 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1365 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1366 int first = 1;
1367 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1368 unsigned int options = (CE_MATCH_REFRESH |
1369 (really ? CE_MATCH_IGNORE_VALID : 0) |
1370 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1371 const char *modified_fmt;
1372 const char *deleted_fmt;
1373 const char *typechange_fmt;
1374 const char *added_fmt;
1375 const char *unmerged_fmt;
1376 uint64_t start = getnanotime();
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_entry;
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_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1414 if (new_entry == ce)
1415 continue;
1416 if (!new_entry) {
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_entry);
1446 }
1447 trace_performance_since(start, "refresh index");
1448 return has_errors;
1449 }
1450
1451 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1452 unsigned int options)
1453 {
1454 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1455 }
1456
1457
1458 /*****************************************************************
1459 * Index File I/O
1460 *****************************************************************/
1461
1462 #define INDEX_FORMAT_DEFAULT 3
1463
1464 static unsigned int get_index_format_default(void)
1465 {
1466 char *envversion = getenv("GIT_INDEX_VERSION");
1467 char *endp;
1468 int value;
1469 unsigned int version = INDEX_FORMAT_DEFAULT;
1470
1471 if (!envversion) {
1472 if (!git_config_get_int("index.version", &value))
1473 version = value;
1474 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1475 warning(_("index.version set, but the value is invalid.\n"
1476 "Using version %i"), INDEX_FORMAT_DEFAULT);
1477 return INDEX_FORMAT_DEFAULT;
1478 }
1479 return version;
1480 }
1481
1482 version = strtoul(envversion, &endp, 10);
1483 if (*endp ||
1484 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1485 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1486 "Using version %i"), INDEX_FORMAT_DEFAULT);
1487 version = INDEX_FORMAT_DEFAULT;
1488 }
1489 return version;
1490 }
1491
1492 /*
1493 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1494 * Again - this is just a (very strong in practice) heuristic that
1495 * the inode hasn't changed.
1496 *
1497 * We save the fields in big-endian order to allow using the
1498 * index file over NFS transparently.
1499 */
1500 struct ondisk_cache_entry {
1501 struct cache_time ctime;
1502 struct cache_time mtime;
1503 uint32_t dev;
1504 uint32_t ino;
1505 uint32_t mode;
1506 uint32_t uid;
1507 uint32_t gid;
1508 uint32_t size;
1509 unsigned char sha1[20];
1510 uint16_t flags;
1511 char name[FLEX_ARRAY]; /* more */
1512 };
1513
1514 /*
1515 * This struct is used when CE_EXTENDED bit is 1
1516 * The struct must match ondisk_cache_entry exactly from
1517 * ctime till flags
1518 */
1519 struct ondisk_cache_entry_extended {
1520 struct cache_time ctime;
1521 struct cache_time mtime;
1522 uint32_t dev;
1523 uint32_t ino;
1524 uint32_t mode;
1525 uint32_t uid;
1526 uint32_t gid;
1527 uint32_t size;
1528 unsigned char sha1[20];
1529 uint16_t flags;
1530 uint16_t flags2;
1531 char name[FLEX_ARRAY]; /* more */
1532 };
1533
1534 /* These are only used for v3 or lower */
1535 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1536 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1537 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1538 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1539 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1540 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1541 ondisk_cache_entry_size(ce_namelen(ce)))
1542
1543 /* Allow fsck to force verification of the index checksum. */
1544 int verify_index_checksum;
1545
1546 /* Allow fsck to force verification of the cache entry order. */
1547 int verify_ce_order;
1548
1549 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1550 {
1551 git_hash_ctx c;
1552 unsigned char hash[GIT_MAX_RAWSZ];
1553 int hdr_version;
1554
1555 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1556 return error("bad signature");
1557 hdr_version = ntohl(hdr->hdr_version);
1558 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1559 return error("bad index version %d", hdr_version);
1560
1561 if (!verify_index_checksum)
1562 return 0;
1563
1564 the_hash_algo->init_fn(&c);
1565 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1566 the_hash_algo->final_fn(hash, &c);
1567 if (hashcmp(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1568 return error("bad index file sha1 signature");
1569 return 0;
1570 }
1571
1572 static int read_index_extension(struct index_state *istate,
1573 const char *ext, void *data, unsigned long sz)
1574 {
1575 switch (CACHE_EXT(ext)) {
1576 case CACHE_EXT_TREE:
1577 istate->cache_tree = cache_tree_read(data, sz);
1578 break;
1579 case CACHE_EXT_RESOLVE_UNDO:
1580 istate->resolve_undo = resolve_undo_read(data, sz);
1581 break;
1582 case CACHE_EXT_LINK:
1583 if (read_link_extension(istate, data, sz))
1584 return -1;
1585 break;
1586 case CACHE_EXT_UNTRACKED:
1587 istate->untracked = read_untracked_extension(data, sz);
1588 break;
1589 case CACHE_EXT_FSMONITOR:
1590 read_fsmonitor_extension(istate, data, sz);
1591 break;
1592 default:
1593 if (*ext < 'A' || 'Z' < *ext)
1594 return error("index uses %.4s extension, which we do not understand",
1595 ext);
1596 fprintf(stderr, "ignoring %.4s extension\n", ext);
1597 break;
1598 }
1599 return 0;
1600 }
1601
1602 int hold_locked_index(struct lock_file *lk, int lock_flags)
1603 {
1604 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1605 }
1606
1607 int read_index(struct index_state *istate)
1608 {
1609 return read_index_from(istate, get_index_file(), get_git_dir());
1610 }
1611
1612 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1613 unsigned int flags,
1614 const char *name,
1615 size_t len)
1616 {
1617 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1618
1619 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1620 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1621 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1622 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1623 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1624 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1625 ce->ce_mode = get_be32(&ondisk->mode);
1626 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1627 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1628 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1629 ce->ce_flags = flags & ~CE_NAMEMASK;
1630 ce->ce_namelen = len;
1631 ce->index = 0;
1632 hashcpy(ce->oid.hash, ondisk->sha1);
1633 memcpy(ce->name, name, len);
1634 ce->name[len] = '\0';
1635 return ce;
1636 }
1637
1638 /*
1639 * Adjacent cache entries tend to share the leading paths, so it makes
1640 * sense to only store the differences in later entries. In the v4
1641 * on-disk format of the index, each on-disk cache entry stores the
1642 * number of bytes to be stripped from the end of the previous name,
1643 * and the bytes to append to the result, to come up with its name.
1644 */
1645 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1646 {
1647 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1648 size_t len = decode_varint(&cp);
1649
1650 if (name->len < len)
1651 die("malformed name field in the index");
1652 strbuf_remove(name, name->len - len, len);
1653 for (ep = cp; *ep; ep++)
1654 ; /* find the end */
1655 strbuf_add(name, cp, ep - cp);
1656 return (const char *)ep + 1 - cp_;
1657 }
1658
1659 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1660 unsigned long *ent_size,
1661 struct strbuf *previous_name)
1662 {
1663 struct cache_entry *ce;
1664 size_t len;
1665 const char *name;
1666 unsigned int flags;
1667
1668 /* On-disk flags are just 16 bits */
1669 flags = get_be16(&ondisk->flags);
1670 len = flags & CE_NAMEMASK;
1671
1672 if (flags & CE_EXTENDED) {
1673 struct ondisk_cache_entry_extended *ondisk2;
1674 int extended_flags;
1675 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1676 extended_flags = get_be16(&ondisk2->flags2) << 16;
1677 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1678 if (extended_flags & ~CE_EXTENDED_FLAGS)
1679 die("Unknown index entry format %08x", extended_flags);
1680 flags |= extended_flags;
1681 name = ondisk2->name;
1682 }
1683 else
1684 name = ondisk->name;
1685
1686 if (!previous_name) {
1687 /* v3 and earlier */
1688 if (len == CE_NAMEMASK)
1689 len = strlen(name);
1690 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1691
1692 *ent_size = ondisk_ce_size(ce);
1693 } else {
1694 unsigned long consumed;
1695 consumed = expand_name_field(previous_name, name);
1696 ce = cache_entry_from_ondisk(ondisk, flags,
1697 previous_name->buf,
1698 previous_name->len);
1699
1700 *ent_size = (name - ((char *)ondisk)) + consumed;
1701 }
1702 return ce;
1703 }
1704
1705 static void check_ce_order(struct index_state *istate)
1706 {
1707 unsigned int i;
1708
1709 if (!verify_ce_order)
1710 return;
1711
1712 for (i = 1; i < istate->cache_nr; i++) {
1713 struct cache_entry *ce = istate->cache[i - 1];
1714 struct cache_entry *next_ce = istate->cache[i];
1715 int name_compare = strcmp(ce->name, next_ce->name);
1716
1717 if (0 < name_compare)
1718 die("unordered stage entries in index");
1719 if (!name_compare) {
1720 if (!ce_stage(ce))
1721 die("multiple stage entries for merged file '%s'",
1722 ce->name);
1723 if (ce_stage(ce) > ce_stage(next_ce))
1724 die("unordered stage entries for '%s'",
1725 ce->name);
1726 }
1727 }
1728 }
1729
1730 static void tweak_untracked_cache(struct index_state *istate)
1731 {
1732 switch (git_config_get_untracked_cache()) {
1733 case -1: /* keep: do nothing */
1734 break;
1735 case 0: /* false */
1736 remove_untracked_cache(istate);
1737 break;
1738 case 1: /* true */
1739 add_untracked_cache(istate);
1740 break;
1741 default: /* unknown value: do nothing */
1742 break;
1743 }
1744 }
1745
1746 static void tweak_split_index(struct index_state *istate)
1747 {
1748 switch (git_config_get_split_index()) {
1749 case -1: /* unset: do nothing */
1750 break;
1751 case 0: /* false */
1752 remove_split_index(istate);
1753 break;
1754 case 1: /* true */
1755 add_split_index(istate);
1756 break;
1757 default: /* unknown value: do nothing */
1758 break;
1759 }
1760 }
1761
1762 static void post_read_index_from(struct index_state *istate)
1763 {
1764 check_ce_order(istate);
1765 tweak_untracked_cache(istate);
1766 tweak_split_index(istate);
1767 tweak_fsmonitor(istate);
1768 }
1769
1770 /* remember to discard_cache() before reading a different cache! */
1771 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1772 {
1773 int fd, i;
1774 struct stat st;
1775 unsigned long src_offset;
1776 struct cache_header *hdr;
1777 void *mmap;
1778 size_t mmap_size;
1779 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1780
1781 if (istate->initialized)
1782 return istate->cache_nr;
1783
1784 istate->timestamp.sec = 0;
1785 istate->timestamp.nsec = 0;
1786 fd = open(path, O_RDONLY);
1787 if (fd < 0) {
1788 if (!must_exist && errno == ENOENT)
1789 return 0;
1790 die_errno("%s: index file open failed", path);
1791 }
1792
1793 if (fstat(fd, &st))
1794 die_errno("cannot stat the open index");
1795
1796 mmap_size = xsize_t(st.st_size);
1797 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
1798 die("index file smaller than expected");
1799
1800 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1801 if (mmap == MAP_FAILED)
1802 die_errno("unable to map index file");
1803 close(fd);
1804
1805 hdr = mmap;
1806 if (verify_hdr(hdr, mmap_size) < 0)
1807 goto unmap;
1808
1809 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
1810 istate->version = ntohl(hdr->hdr_version);
1811 istate->cache_nr = ntohl(hdr->hdr_entries);
1812 istate->cache_alloc = alloc_nr(istate->cache_nr);
1813 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1814 istate->initialized = 1;
1815
1816 if (istate->version == 4)
1817 previous_name = &previous_name_buf;
1818 else
1819 previous_name = NULL;
1820
1821 src_offset = sizeof(*hdr);
1822 for (i = 0; i < istate->cache_nr; i++) {
1823 struct ondisk_cache_entry *disk_ce;
1824 struct cache_entry *ce;
1825 unsigned long consumed;
1826
1827 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1828 ce = create_from_disk(disk_ce, &consumed, previous_name);
1829 set_index_entry(istate, i, ce);
1830
1831 src_offset += consumed;
1832 }
1833 strbuf_release(&previous_name_buf);
1834 istate->timestamp.sec = st.st_mtime;
1835 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1836
1837 while (src_offset <= mmap_size - the_hash_algo->rawsz - 8) {
1838 /* After an array of active_nr index entries,
1839 * there can be arbitrary number of extended
1840 * sections, each of which is prefixed with
1841 * extension name (4-byte) and section length
1842 * in 4-byte network byte order.
1843 */
1844 uint32_t extsize;
1845 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1846 extsize = ntohl(extsize);
1847 if (read_index_extension(istate,
1848 (const char *) mmap + src_offset,
1849 (char *) mmap + src_offset + 8,
1850 extsize) < 0)
1851 goto unmap;
1852 src_offset += 8;
1853 src_offset += extsize;
1854 }
1855 munmap(mmap, mmap_size);
1856 return istate->cache_nr;
1857
1858 unmap:
1859 munmap(mmap, mmap_size);
1860 die("index file corrupt");
1861 }
1862
1863 /*
1864 * Signal that the shared index is used by updating its mtime.
1865 *
1866 * This way, shared index can be removed if they have not been used
1867 * for some time.
1868 */
1869 static void freshen_shared_index(const char *shared_index, int warn)
1870 {
1871 if (!check_and_freshen_file(shared_index, 1) && warn)
1872 warning("could not freshen shared index '%s'", shared_index);
1873 }
1874
1875 int read_index_from(struct index_state *istate, const char *path,
1876 const char *gitdir)
1877 {
1878 uint64_t start = getnanotime();
1879 struct split_index *split_index;
1880 int ret;
1881 char *base_sha1_hex;
1882 char *base_path;
1883
1884 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1885 if (istate->initialized)
1886 return istate->cache_nr;
1887
1888 ret = do_read_index(istate, path, 0);
1889 trace_performance_since(start, "read cache %s", path);
1890
1891 split_index = istate->split_index;
1892 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1893 post_read_index_from(istate);
1894 return ret;
1895 }
1896
1897 if (split_index->base)
1898 discard_index(split_index->base);
1899 else
1900 split_index->base = xcalloc(1, sizeof(*split_index->base));
1901
1902 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1903 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_sha1_hex);
1904 ret = do_read_index(split_index->base, base_path, 1);
1905 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1906 die("broken index, expect %s in %s, got %s",
1907 base_sha1_hex, base_path,
1908 sha1_to_hex(split_index->base->sha1));
1909
1910 freshen_shared_index(base_path, 0);
1911 merge_base_index(istate);
1912 post_read_index_from(istate);
1913 trace_performance_since(start, "read cache %s", base_path);
1914 free(base_path);
1915 return ret;
1916 }
1917
1918 int is_index_unborn(struct index_state *istate)
1919 {
1920 return (!istate->cache_nr && !istate->timestamp.sec);
1921 }
1922
1923 int discard_index(struct index_state *istate)
1924 {
1925 int i;
1926
1927 for (i = 0; i < istate->cache_nr; i++) {
1928 if (istate->cache[i]->index &&
1929 istate->split_index &&
1930 istate->split_index->base &&
1931 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1932 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1933 continue;
1934 free(istate->cache[i]);
1935 }
1936 resolve_undo_clear_index(istate);
1937 istate->cache_nr = 0;
1938 istate->cache_changed = 0;
1939 istate->timestamp.sec = 0;
1940 istate->timestamp.nsec = 0;
1941 free_name_hash(istate);
1942 cache_tree_free(&(istate->cache_tree));
1943 istate->initialized = 0;
1944 FREE_AND_NULL(istate->cache);
1945 istate->cache_alloc = 0;
1946 discard_split_index(istate);
1947 free_untracked_cache(istate->untracked);
1948 istate->untracked = NULL;
1949 return 0;
1950 }
1951
1952 int unmerged_index(const struct index_state *istate)
1953 {
1954 int i;
1955 for (i = 0; i < istate->cache_nr; i++) {
1956 if (ce_stage(istate->cache[i]))
1957 return 1;
1958 }
1959 return 0;
1960 }
1961
1962 #define WRITE_BUFFER_SIZE 8192
1963 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1964 static unsigned long write_buffer_len;
1965
1966 static int ce_write_flush(git_hash_ctx *context, int fd)
1967 {
1968 unsigned int buffered = write_buffer_len;
1969 if (buffered) {
1970 the_hash_algo->update_fn(context, write_buffer, buffered);
1971 if (write_in_full(fd, write_buffer, buffered) < 0)
1972 return -1;
1973 write_buffer_len = 0;
1974 }
1975 return 0;
1976 }
1977
1978 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
1979 {
1980 while (len) {
1981 unsigned int buffered = write_buffer_len;
1982 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1983 if (partial > len)
1984 partial = len;
1985 memcpy(write_buffer + buffered, data, partial);
1986 buffered += partial;
1987 if (buffered == WRITE_BUFFER_SIZE) {
1988 write_buffer_len = buffered;
1989 if (ce_write_flush(context, fd))
1990 return -1;
1991 buffered = 0;
1992 }
1993 write_buffer_len = buffered;
1994 len -= partial;
1995 data = (char *) data + partial;
1996 }
1997 return 0;
1998 }
1999
2000 static int write_index_ext_header(git_hash_ctx *context, int fd,
2001 unsigned int ext, unsigned int sz)
2002 {
2003 ext = htonl(ext);
2004 sz = htonl(sz);
2005 return ((ce_write(context, fd, &ext, 4) < 0) ||
2006 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2007 }
2008
2009 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2010 {
2011 unsigned int left = write_buffer_len;
2012
2013 if (left) {
2014 write_buffer_len = 0;
2015 the_hash_algo->update_fn(context, write_buffer, left);
2016 }
2017
2018 /* Flush first if not enough space for hash signature */
2019 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2020 if (write_in_full(fd, write_buffer, left) < 0)
2021 return -1;
2022 left = 0;
2023 }
2024
2025 /* Append the hash signature at the end */
2026 the_hash_algo->final_fn(write_buffer + left, context);
2027 hashcpy(hash, write_buffer + left);
2028 left += the_hash_algo->rawsz;
2029 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2030 }
2031
2032 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2033 {
2034 /*
2035 * The only thing we care about in this function is to smudge the
2036 * falsely clean entry due to touch-update-touch race, so we leave
2037 * everything else as they are. We are called for entries whose
2038 * ce_stat_data.sd_mtime match the index file mtime.
2039 *
2040 * Note that this actually does not do much for gitlinks, for
2041 * which ce_match_stat_basic() always goes to the actual
2042 * contents. The caller checks with is_racy_timestamp() which
2043 * always says "no" for gitlinks, so we are not called for them ;-)
2044 */
2045 struct stat st;
2046
2047 if (lstat(ce->name, &st) < 0)
2048 return;
2049 if (ce_match_stat_basic(ce, &st))
2050 return;
2051 if (ce_modified_check_fs(ce, &st)) {
2052 /* This is "racily clean"; smudge it. Note that this
2053 * is a tricky code. At first glance, it may appear
2054 * that it can break with this sequence:
2055 *
2056 * $ echo xyzzy >frotz
2057 * $ git-update-index --add frotz
2058 * $ : >frotz
2059 * $ sleep 3
2060 * $ echo filfre >nitfol
2061 * $ git-update-index --add nitfol
2062 *
2063 * but it does not. When the second update-index runs,
2064 * it notices that the entry "frotz" has the same timestamp
2065 * as index, and if we were to smudge it by resetting its
2066 * size to zero here, then the object name recorded
2067 * in index is the 6-byte file but the cached stat information
2068 * becomes zero --- which would then match what we would
2069 * obtain from the filesystem next time we stat("frotz").
2070 *
2071 * However, the second update-index, before calling
2072 * this function, notices that the cached size is 6
2073 * bytes and what is on the filesystem is an empty
2074 * file, and never calls us, so the cached size information
2075 * for "frotz" stays 6 which does not match the filesystem.
2076 */
2077 ce->ce_stat_data.sd_size = 0;
2078 }
2079 }
2080
2081 /* Copy miscellaneous fields but not the name */
2082 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2083 struct cache_entry *ce)
2084 {
2085 short flags;
2086
2087 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2088 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2089 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2090 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2091 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2092 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2093 ondisk->mode = htonl(ce->ce_mode);
2094 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2095 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2096 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2097 hashcpy(ondisk->sha1, ce->oid.hash);
2098
2099 flags = ce->ce_flags & ~CE_NAMEMASK;
2100 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2101 ondisk->flags = htons(flags);
2102 if (ce->ce_flags & CE_EXTENDED) {
2103 struct ondisk_cache_entry_extended *ondisk2;
2104 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2105 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2106 }
2107 }
2108
2109 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2110 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2111 {
2112 int size;
2113 int saved_namelen = saved_namelen; /* compiler workaround */
2114 int result;
2115 static unsigned char padding[8] = { 0x00 };
2116
2117 if (ce->ce_flags & CE_STRIP_NAME) {
2118 saved_namelen = ce_namelen(ce);
2119 ce->ce_namelen = 0;
2120 }
2121
2122 if (ce->ce_flags & CE_EXTENDED)
2123 size = offsetof(struct ondisk_cache_entry_extended, name);
2124 else
2125 size = offsetof(struct ondisk_cache_entry, name);
2126
2127 if (!previous_name) {
2128 int len = ce_namelen(ce);
2129 copy_cache_entry_to_ondisk(ondisk, ce);
2130 result = ce_write(c, fd, ondisk, size);
2131 if (!result)
2132 result = ce_write(c, fd, ce->name, len);
2133 if (!result)
2134 result = ce_write(c, fd, padding, align_padding_size(size, len));
2135 } else {
2136 int common, to_remove, prefix_size;
2137 unsigned char to_remove_vi[16];
2138 for (common = 0;
2139 (ce->name[common] &&
2140 common < previous_name->len &&
2141 ce->name[common] == previous_name->buf[common]);
2142 common++)
2143 ; /* still matching */
2144 to_remove = previous_name->len - common;
2145 prefix_size = encode_varint(to_remove, to_remove_vi);
2146
2147 copy_cache_entry_to_ondisk(ondisk, ce);
2148 result = ce_write(c, fd, ondisk, size);
2149 if (!result)
2150 result = ce_write(c, fd, to_remove_vi, prefix_size);
2151 if (!result)
2152 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2153 if (!result)
2154 result = ce_write(c, fd, padding, 1);
2155
2156 strbuf_splice(previous_name, common, to_remove,
2157 ce->name + common, ce_namelen(ce) - common);
2158 }
2159 if (ce->ce_flags & CE_STRIP_NAME) {
2160 ce->ce_namelen = saved_namelen;
2161 ce->ce_flags &= ~CE_STRIP_NAME;
2162 }
2163
2164 return result;
2165 }
2166
2167 /*
2168 * This function verifies if index_state has the correct sha1 of the
2169 * index file. Don't die if we have any other failure, just return 0.
2170 */
2171 static int verify_index_from(const struct index_state *istate, const char *path)
2172 {
2173 int fd;
2174 ssize_t n;
2175 struct stat st;
2176 unsigned char hash[GIT_MAX_RAWSZ];
2177
2178 if (!istate->initialized)
2179 return 0;
2180
2181 fd = open(path, O_RDONLY);
2182 if (fd < 0)
2183 return 0;
2184
2185 if (fstat(fd, &st))
2186 goto out;
2187
2188 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2189 goto out;
2190
2191 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2192 if (n != the_hash_algo->rawsz)
2193 goto out;
2194
2195 if (hashcmp(istate->sha1, hash))
2196 goto out;
2197
2198 close(fd);
2199 return 1;
2200
2201 out:
2202 close(fd);
2203 return 0;
2204 }
2205
2206 static int verify_index(const struct index_state *istate)
2207 {
2208 return verify_index_from(istate, get_index_file());
2209 }
2210
2211 static int has_racy_timestamp(struct index_state *istate)
2212 {
2213 int entries = istate->cache_nr;
2214 int i;
2215
2216 for (i = 0; i < entries; i++) {
2217 struct cache_entry *ce = istate->cache[i];
2218 if (is_racy_timestamp(istate, ce))
2219 return 1;
2220 }
2221 return 0;
2222 }
2223
2224 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2225 {
2226 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2227 verify_index(istate))
2228 write_locked_index(istate, lockfile, COMMIT_LOCK);
2229 else
2230 rollback_lock_file(lockfile);
2231 }
2232
2233 /*
2234 * On success, `tempfile` is closed. If it is the temporary file
2235 * of a `struct lock_file`, we will therefore effectively perform
2236 * a 'close_lock_file_gently()`. Since that is an implementation
2237 * detail of lockfiles, callers of `do_write_index()` should not
2238 * rely on it.
2239 */
2240 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2241 int strip_extensions)
2242 {
2243 uint64_t start = getnanotime();
2244 int newfd = tempfile->fd;
2245 git_hash_ctx c;
2246 struct cache_header hdr;
2247 int i, err = 0, removed, extended, hdr_version;
2248 struct cache_entry **cache = istate->cache;
2249 int entries = istate->cache_nr;
2250 struct stat st;
2251 struct ondisk_cache_entry_extended ondisk;
2252 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2253 int drop_cache_tree = istate->drop_cache_tree;
2254
2255 for (i = removed = extended = 0; i < entries; i++) {
2256 if (cache[i]->ce_flags & CE_REMOVE)
2257 removed++;
2258
2259 /* reduce extended entries if possible */
2260 cache[i]->ce_flags &= ~CE_EXTENDED;
2261 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2262 extended++;
2263 cache[i]->ce_flags |= CE_EXTENDED;
2264 }
2265 }
2266
2267 if (!istate->version) {
2268 istate->version = get_index_format_default();
2269 if (getenv("GIT_TEST_SPLIT_INDEX"))
2270 init_split_index(istate);
2271 }
2272
2273 /* demote version 3 to version 2 when the latter suffices */
2274 if (istate->version == 3 || istate->version == 2)
2275 istate->version = extended ? 3 : 2;
2276
2277 hdr_version = istate->version;
2278
2279 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2280 hdr.hdr_version = htonl(hdr_version);
2281 hdr.hdr_entries = htonl(entries - removed);
2282
2283 the_hash_algo->init_fn(&c);
2284 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2285 return -1;
2286
2287 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2288
2289 for (i = 0; i < entries; i++) {
2290 struct cache_entry *ce = cache[i];
2291 if (ce->ce_flags & CE_REMOVE)
2292 continue;
2293 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2294 ce_smudge_racily_clean_entry(ce);
2295 if (is_null_oid(&ce->oid)) {
2296 static const char msg[] = "cache entry has null sha1: %s";
2297 static int allow = -1;
2298
2299 if (allow < 0)
2300 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2301 if (allow)
2302 warning(msg, ce->name);
2303 else
2304 err = error(msg, ce->name);
2305
2306 drop_cache_tree = 1;
2307 }
2308 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2309 err = -1;
2310
2311 if (err)
2312 break;
2313 }
2314 strbuf_release(&previous_name_buf);
2315
2316 if (err)
2317 return err;
2318
2319 /* Write extension data here */
2320 if (!strip_extensions && istate->split_index) {
2321 struct strbuf sb = STRBUF_INIT;
2322
2323 err = write_link_extension(&sb, istate) < 0 ||
2324 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2325 sb.len) < 0 ||
2326 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2327 strbuf_release(&sb);
2328 if (err)
2329 return -1;
2330 }
2331 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2332 struct strbuf sb = STRBUF_INIT;
2333
2334 cache_tree_write(&sb, istate->cache_tree);
2335 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2336 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2337 strbuf_release(&sb);
2338 if (err)
2339 return -1;
2340 }
2341 if (!strip_extensions && istate->resolve_undo) {
2342 struct strbuf sb = STRBUF_INIT;
2343
2344 resolve_undo_write(&sb, istate->resolve_undo);
2345 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2346 sb.len) < 0
2347 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2348 strbuf_release(&sb);
2349 if (err)
2350 return -1;
2351 }
2352 if (!strip_extensions && istate->untracked) {
2353 struct strbuf sb = STRBUF_INIT;
2354
2355 write_untracked_extension(&sb, istate->untracked);
2356 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2357 sb.len) < 0 ||
2358 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2359 strbuf_release(&sb);
2360 if (err)
2361 return -1;
2362 }
2363 if (!strip_extensions && istate->fsmonitor_last_update) {
2364 struct strbuf sb = STRBUF_INIT;
2365
2366 write_fsmonitor_extension(&sb, istate);
2367 err = write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2368 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2369 strbuf_release(&sb);
2370 if (err)
2371 return -1;
2372 }
2373
2374 if (ce_flush(&c, newfd, istate->sha1))
2375 return -1;
2376 if (close_tempfile_gently(tempfile)) {
2377 error(_("could not close '%s'"), tempfile->filename.buf);
2378 return -1;
2379 }
2380 if (stat(tempfile->filename.buf, &st))
2381 return -1;
2382 istate->timestamp.sec = (unsigned int)st.st_mtime;
2383 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2384 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
2385 return 0;
2386 }
2387
2388 void set_alternate_index_output(const char *name)
2389 {
2390 alternate_index_output = name;
2391 }
2392
2393 static int commit_locked_index(struct lock_file *lk)
2394 {
2395 if (alternate_index_output)
2396 return commit_lock_file_to(lk, alternate_index_output);
2397 else
2398 return commit_lock_file(lk);
2399 }
2400
2401 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2402 unsigned flags)
2403 {
2404 int ret = do_write_index(istate, lock->tempfile, 0);
2405 if (ret)
2406 return ret;
2407 if (flags & COMMIT_LOCK)
2408 return commit_locked_index(lock);
2409 return close_lock_file_gently(lock);
2410 }
2411
2412 static int write_split_index(struct index_state *istate,
2413 struct lock_file *lock,
2414 unsigned flags)
2415 {
2416 int ret;
2417 prepare_to_write_split_index(istate);
2418 ret = do_write_locked_index(istate, lock, flags);
2419 finish_writing_split_index(istate);
2420 return ret;
2421 }
2422
2423 static const char *shared_index_expire = "2.weeks.ago";
2424
2425 static unsigned long get_shared_index_expire_date(void)
2426 {
2427 static unsigned long shared_index_expire_date;
2428 static int shared_index_expire_date_prepared;
2429
2430 if (!shared_index_expire_date_prepared) {
2431 git_config_get_expiry("splitindex.sharedindexexpire",
2432 &shared_index_expire);
2433 shared_index_expire_date = approxidate(shared_index_expire);
2434 shared_index_expire_date_prepared = 1;
2435 }
2436
2437 return shared_index_expire_date;
2438 }
2439
2440 static int should_delete_shared_index(const char *shared_index_path)
2441 {
2442 struct stat st;
2443 unsigned long expiration;
2444
2445 /* Check timestamp */
2446 expiration = get_shared_index_expire_date();
2447 if (!expiration)
2448 return 0;
2449 if (stat(shared_index_path, &st))
2450 return error_errno(_("could not stat '%s'"), shared_index_path);
2451 if (st.st_mtime > expiration)
2452 return 0;
2453
2454 return 1;
2455 }
2456
2457 static int clean_shared_index_files(const char *current_hex)
2458 {
2459 struct dirent *de;
2460 DIR *dir = opendir(get_git_dir());
2461
2462 if (!dir)
2463 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2464
2465 while ((de = readdir(dir)) != NULL) {
2466 const char *sha1_hex;
2467 const char *shared_index_path;
2468 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2469 continue;
2470 if (!strcmp(sha1_hex, current_hex))
2471 continue;
2472 shared_index_path = git_path("%s", de->d_name);
2473 if (should_delete_shared_index(shared_index_path) > 0 &&
2474 unlink(shared_index_path))
2475 warning_errno(_("unable to unlink: %s"), shared_index_path);
2476 }
2477 closedir(dir);
2478
2479 return 0;
2480 }
2481
2482 static int write_shared_index(struct index_state *istate,
2483 struct tempfile **temp)
2484 {
2485 struct split_index *si = istate->split_index;
2486 int ret;
2487
2488 move_cache_to_base_index(istate);
2489 ret = do_write_index(si->base, *temp, 1);
2490 if (ret)
2491 return ret;
2492 ret = adjust_shared_perm(get_tempfile_path(*temp));
2493 if (ret) {
2494 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
2495 return ret;
2496 }
2497 ret = rename_tempfile(temp,
2498 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2499 if (!ret) {
2500 hashcpy(si->base_sha1, si->base->sha1);
2501 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2502 }
2503
2504 return ret;
2505 }
2506
2507 static const int default_max_percent_split_change = 20;
2508
2509 static int too_many_not_shared_entries(struct index_state *istate)
2510 {
2511 int i, not_shared = 0;
2512 int max_split = git_config_get_max_percent_split_change();
2513
2514 switch (max_split) {
2515 case -1:
2516 /* not or badly configured: use the default value */
2517 max_split = default_max_percent_split_change;
2518 break;
2519 case 0:
2520 return 1; /* 0% means always write a new shared index */
2521 case 100:
2522 return 0; /* 100% means never write a new shared index */
2523 default:
2524 break; /* just use the configured value */
2525 }
2526
2527 /* Count not shared entries */
2528 for (i = 0; i < istate->cache_nr; i++) {
2529 struct cache_entry *ce = istate->cache[i];
2530 if (!ce->index)
2531 not_shared++;
2532 }
2533
2534 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2535 }
2536
2537 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2538 unsigned flags)
2539 {
2540 int new_shared_index, ret;
2541 struct split_index *si = istate->split_index;
2542
2543 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
2544 if (flags & COMMIT_LOCK)
2545 rollback_lock_file(lock);
2546 return 0;
2547 }
2548
2549 if (istate->fsmonitor_last_update)
2550 fill_fsmonitor_bitmap(istate);
2551
2552 if (!si || alternate_index_output ||
2553 (istate->cache_changed & ~EXTMASK)) {
2554 if (si)
2555 hashclr(si->base_sha1);
2556 ret = do_write_locked_index(istate, lock, flags);
2557 goto out;
2558 }
2559
2560 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2561 int v = si->base_sha1[0];
2562 if ((v & 15) < 6)
2563 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2564 }
2565 if (too_many_not_shared_entries(istate))
2566 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2567
2568 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2569
2570 if (new_shared_index) {
2571 struct tempfile *temp;
2572 int saved_errno;
2573
2574 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2575 if (!temp) {
2576 hashclr(si->base_sha1);
2577 ret = do_write_locked_index(istate, lock, flags);
2578 goto out;
2579 }
2580 ret = write_shared_index(istate, &temp);
2581
2582 saved_errno = errno;
2583 if (is_tempfile_active(temp))
2584 delete_tempfile(&temp);
2585 errno = saved_errno;
2586
2587 if (ret)
2588 goto out;
2589 }
2590
2591 ret = write_split_index(istate, lock, flags);
2592
2593 /* Freshen the shared index only if the split-index was written */
2594 if (!ret && !new_shared_index) {
2595 const char *shared_index = git_path("sharedindex.%s",
2596 sha1_to_hex(si->base_sha1));
2597 freshen_shared_index(shared_index, 1);
2598 }
2599
2600 out:
2601 if (flags & COMMIT_LOCK)
2602 rollback_lock_file(lock);
2603 return ret;
2604 }
2605
2606 /*
2607 * Read the index file that is potentially unmerged into given
2608 * index_state, dropping any unmerged entries. Returns true if
2609 * the index is unmerged. Callers who want to refuse to work
2610 * from an unmerged state can call this and check its return value,
2611 * instead of calling read_cache().
2612 */
2613 int read_index_unmerged(struct index_state *istate)
2614 {
2615 int i;
2616 int unmerged = 0;
2617
2618 read_index(istate);
2619 for (i = 0; i < istate->cache_nr; i++) {
2620 struct cache_entry *ce = istate->cache[i];
2621 struct cache_entry *new_ce;
2622 int size, len;
2623
2624 if (!ce_stage(ce))
2625 continue;
2626 unmerged = 1;
2627 len = ce_namelen(ce);
2628 size = cache_entry_size(len);
2629 new_ce = xcalloc(1, size);
2630 memcpy(new_ce->name, ce->name, len);
2631 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2632 new_ce->ce_namelen = len;
2633 new_ce->ce_mode = ce->ce_mode;
2634 if (add_index_entry(istate, new_ce, 0))
2635 return error("%s: cannot drop to stage #0",
2636 new_ce->name);
2637 }
2638 return unmerged;
2639 }
2640
2641 /*
2642 * Returns 1 if the path is an "other" path with respect to
2643 * the index; that is, the path is not mentioned in the index at all,
2644 * either as a file, a directory with some files in the index,
2645 * or as an unmerged entry.
2646 *
2647 * We helpfully remove a trailing "/" from directories so that
2648 * the output of read_directory can be used as-is.
2649 */
2650 int index_name_is_other(const struct index_state *istate, const char *name,
2651 int namelen)
2652 {
2653 int pos;
2654 if (namelen && name[namelen - 1] == '/')
2655 namelen--;
2656 pos = index_name_pos(istate, name, namelen);
2657 if (0 <= pos)
2658 return 0; /* exact match */
2659 pos = -pos - 1;
2660 if (pos < istate->cache_nr) {
2661 struct cache_entry *ce = istate->cache[pos];
2662 if (ce_namelen(ce) == namelen &&
2663 !memcmp(ce->name, name, namelen))
2664 return 0; /* Yup, this one exists unmerged */
2665 }
2666 return 1;
2667 }
2668
2669 void *read_blob_data_from_index(const struct index_state *istate,
2670 const char *path, unsigned long *size)
2671 {
2672 int pos, len;
2673 unsigned long sz;
2674 enum object_type type;
2675 void *data;
2676
2677 len = strlen(path);
2678 pos = index_name_pos(istate, path, len);
2679 if (pos < 0) {
2680 /*
2681 * We might be in the middle of a merge, in which
2682 * case we would read stage #2 (ours).
2683 */
2684 int i;
2685 for (i = -pos - 1;
2686 (pos < 0 && i < istate->cache_nr &&
2687 !strcmp(istate->cache[i]->name, path));
2688 i++)
2689 if (ce_stage(istate->cache[i]) == 2)
2690 pos = i;
2691 }
2692 if (pos < 0)
2693 return NULL;
2694 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2695 if (!data || type != OBJ_BLOB) {
2696 free(data);
2697 return NULL;
2698 }
2699 if (size)
2700 *size = sz;
2701 return data;
2702 }
2703
2704 void stat_validity_clear(struct stat_validity *sv)
2705 {
2706 FREE_AND_NULL(sv->sd);
2707 }
2708
2709 int stat_validity_check(struct stat_validity *sv, const char *path)
2710 {
2711 struct stat st;
2712
2713 if (stat(path, &st) < 0)
2714 return sv->sd == NULL;
2715 if (!sv->sd)
2716 return 0;
2717 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2718 }
2719
2720 void stat_validity_update(struct stat_validity *sv, int fd)
2721 {
2722 struct stat st;
2723
2724 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2725 stat_validity_clear(sv);
2726 else {
2727 if (!sv->sd)
2728 sv->sd = xcalloc(1, sizeof(struct stat_data));
2729 fill_stat_data(sv->sd, &st);
2730 }
2731 }
2732
2733 void move_index_extensions(struct index_state *dst, struct index_state *src)
2734 {
2735 dst->untracked = src->untracked;
2736 src->untracked = NULL;
2737 }