userdiff: add built-in pattern for golang
[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 set_index_entry(istate, nr, ce);
66 ce->ce_flags |= CE_UPDATE_IN_BASE;
67 mark_fsmonitor_invalid(istate, ce);
68 istate->cache_changed |= CE_ENTRY_CHANGED;
69 }
70
71 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
72 {
73 struct cache_entry *old = istate->cache[nr], *new;
74 int namelen = strlen(new_name);
75
76 new = xmalloc(cache_entry_size(namelen));
77 copy_cache_entry(new, old);
78 new->ce_flags &= ~CE_HASHED;
79 new->ce_namelen = namelen;
80 new->index = 0;
81 memcpy(new->name, new_name, namelen + 1);
82
83 cache_tree_invalidate_path(istate, old->name);
84 untracked_cache_remove_from_index(istate, old->name);
85 remove_index_entry_at(istate, nr);
86 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
87 }
88
89 void fill_stat_data(struct stat_data *sd, struct stat *st)
90 {
91 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
92 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
93 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
94 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
95 sd->sd_dev = st->st_dev;
96 sd->sd_ino = st->st_ino;
97 sd->sd_uid = st->st_uid;
98 sd->sd_gid = st->st_gid;
99 sd->sd_size = st->st_size;
100 }
101
102 int match_stat_data(const struct stat_data *sd, struct stat *st)
103 {
104 int changed = 0;
105
106 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
107 changed |= MTIME_CHANGED;
108 if (trust_ctime && check_stat &&
109 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
110 changed |= CTIME_CHANGED;
111
112 #ifdef USE_NSEC
113 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
114 changed |= MTIME_CHANGED;
115 if (trust_ctime && check_stat &&
116 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
117 changed |= CTIME_CHANGED;
118 #endif
119
120 if (check_stat) {
121 if (sd->sd_uid != (unsigned int) st->st_uid ||
122 sd->sd_gid != (unsigned int) st->st_gid)
123 changed |= OWNER_CHANGED;
124 if (sd->sd_ino != (unsigned int) st->st_ino)
125 changed |= INODE_CHANGED;
126 }
127
128 #ifdef USE_STDEV
129 /*
130 * st_dev breaks on network filesystems where different
131 * clients will have different views of what "device"
132 * the filesystem is on
133 */
134 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
135 changed |= INODE_CHANGED;
136 #endif
137
138 if (sd->sd_size != (unsigned int) st->st_size)
139 changed |= DATA_CHANGED;
140
141 return changed;
142 }
143
144 /*
145 * This only updates the "non-critical" parts of the directory
146 * cache, ie the parts that aren't tracked by GIT, and only used
147 * to validate the cache.
148 */
149 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
150 {
151 fill_stat_data(&ce->ce_stat_data, st);
152
153 if (assume_unchanged)
154 ce->ce_flags |= CE_VALID;
155
156 if (S_ISREG(st->st_mode)) {
157 ce_mark_uptodate(ce);
158 mark_fsmonitor_valid(ce);
159 }
160 }
161
162 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
163 {
164 int match = -1;
165 int fd = git_open_cloexec(ce->name, O_RDONLY);
166
167 if (fd >= 0) {
168 struct object_id oid;
169 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
170 match = oidcmp(&oid, &ce->oid);
171 /* index_fd() closed the file descriptor already */
172 }
173 return match;
174 }
175
176 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
177 {
178 int match = -1;
179 void *buffer;
180 unsigned long size;
181 enum object_type type;
182 struct strbuf sb = STRBUF_INIT;
183
184 if (strbuf_readlink(&sb, ce->name, expected_size))
185 return -1;
186
187 buffer = read_sha1_file(ce->oid.hash, &type, &size);
188 if (buffer) {
189 if (size == sb.len)
190 match = memcmp(buffer, sb.buf, size);
191 free(buffer);
192 }
193 strbuf_release(&sb);
194 return match;
195 }
196
197 static int ce_compare_gitlink(const struct cache_entry *ce)
198 {
199 struct object_id oid;
200
201 /*
202 * We don't actually require that the .git directory
203 * under GITLINK directory be a valid git directory. It
204 * might even be missing (in case nobody populated that
205 * sub-project).
206 *
207 * If so, we consider it always to match.
208 */
209 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
210 return 0;
211 return oidcmp(&oid, &ce->oid);
212 }
213
214 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
215 {
216 switch (st->st_mode & S_IFMT) {
217 case S_IFREG:
218 if (ce_compare_data(ce, st))
219 return DATA_CHANGED;
220 break;
221 case S_IFLNK:
222 if (ce_compare_link(ce, xsize_t(st->st_size)))
223 return DATA_CHANGED;
224 break;
225 case S_IFDIR:
226 if (S_ISGITLINK(ce->ce_mode))
227 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
228 /* else fallthrough */
229 default:
230 return TYPE_CHANGED;
231 }
232 return 0;
233 }
234
235 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
236 {
237 unsigned int changed = 0;
238
239 if (ce->ce_flags & CE_REMOVE)
240 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
241
242 switch (ce->ce_mode & S_IFMT) {
243 case S_IFREG:
244 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
245 /* We consider only the owner x bit to be relevant for
246 * "mode changes"
247 */
248 if (trust_executable_bit &&
249 (0100 & (ce->ce_mode ^ st->st_mode)))
250 changed |= MODE_CHANGED;
251 break;
252 case S_IFLNK:
253 if (!S_ISLNK(st->st_mode) &&
254 (has_symlinks || !S_ISREG(st->st_mode)))
255 changed |= TYPE_CHANGED;
256 break;
257 case S_IFGITLINK:
258 /* We ignore most of the st_xxx fields for gitlinks */
259 if (!S_ISDIR(st->st_mode))
260 changed |= TYPE_CHANGED;
261 else if (ce_compare_gitlink(ce))
262 changed |= DATA_CHANGED;
263 return changed;
264 default:
265 die("internal error: ce_mode is %o", ce->ce_mode);
266 }
267
268 changed |= match_stat_data(&ce->ce_stat_data, st);
269
270 /* Racily smudged entry? */
271 if (!ce->ce_stat_data.sd_size) {
272 if (!is_empty_blob_sha1(ce->oid.hash))
273 changed |= DATA_CHANGED;
274 }
275
276 return changed;
277 }
278
279 static int is_racy_stat(const struct index_state *istate,
280 const struct stat_data *sd)
281 {
282 return (istate->timestamp.sec &&
283 #ifdef USE_NSEC
284 /* nanosecond timestamped files can also be racy! */
285 (istate->timestamp.sec < sd->sd_mtime.sec ||
286 (istate->timestamp.sec == sd->sd_mtime.sec &&
287 istate->timestamp.nsec <= sd->sd_mtime.nsec))
288 #else
289 istate->timestamp.sec <= sd->sd_mtime.sec
290 #endif
291 );
292 }
293
294 static int is_racy_timestamp(const struct index_state *istate,
295 const struct cache_entry *ce)
296 {
297 return (!S_ISGITLINK(ce->ce_mode) &&
298 is_racy_stat(istate, &ce->ce_stat_data));
299 }
300
301 int match_stat_data_racy(const struct index_state *istate,
302 const struct stat_data *sd, struct stat *st)
303 {
304 if (is_racy_stat(istate, sd))
305 return MTIME_CHANGED;
306 return match_stat_data(sd, st);
307 }
308
309 int ie_match_stat(struct index_state *istate,
310 const struct cache_entry *ce, struct stat *st,
311 unsigned int options)
312 {
313 unsigned int changed;
314 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
315 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
316 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
317 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
318
319 if (!ignore_fsmonitor)
320 refresh_fsmonitor(istate);
321 /*
322 * If it's marked as always valid in the index, it's
323 * valid whatever the checked-out copy says.
324 *
325 * skip-worktree has the same effect with higher precedence
326 */
327 if (!ignore_skip_worktree && ce_skip_worktree(ce))
328 return 0;
329 if (!ignore_valid && (ce->ce_flags & CE_VALID))
330 return 0;
331 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
332 return 0;
333
334 /*
335 * Intent-to-add entries have not been added, so the index entry
336 * by definition never matches what is in the work tree until it
337 * actually gets added.
338 */
339 if (ce_intent_to_add(ce))
340 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
341
342 changed = ce_match_stat_basic(ce, st);
343
344 /*
345 * Within 1 second of this sequence:
346 * echo xyzzy >file && git-update-index --add file
347 * running this command:
348 * echo frotz >file
349 * would give a falsely clean cache entry. The mtime and
350 * length match the cache, and other stat fields do not change.
351 *
352 * We could detect this at update-index time (the cache entry
353 * being registered/updated records the same time as "now")
354 * and delay the return from git-update-index, but that would
355 * effectively mean we can make at most one commit per second,
356 * which is not acceptable. Instead, we check cache entries
357 * whose mtime are the same as the index file timestamp more
358 * carefully than others.
359 */
360 if (!changed && is_racy_timestamp(istate, ce)) {
361 if (assume_racy_is_modified)
362 changed |= DATA_CHANGED;
363 else
364 changed |= ce_modified_check_fs(ce, st);
365 }
366
367 return changed;
368 }
369
370 int ie_modified(struct index_state *istate,
371 const struct cache_entry *ce,
372 struct stat *st, unsigned int options)
373 {
374 int changed, changed_fs;
375
376 changed = ie_match_stat(istate, ce, st, options);
377 if (!changed)
378 return 0;
379 /*
380 * If the mode or type has changed, there's no point in trying
381 * to refresh the entry - it's not going to match
382 */
383 if (changed & (MODE_CHANGED | TYPE_CHANGED))
384 return changed;
385
386 /*
387 * Immediately after read-tree or update-index --cacheinfo,
388 * the length field is zero, as we have never even read the
389 * lstat(2) information once, and we cannot trust DATA_CHANGED
390 * returned by ie_match_stat() which in turn was returned by
391 * ce_match_stat_basic() to signal that the filesize of the
392 * blob changed. We have to actually go to the filesystem to
393 * see if the contents match, and if so, should answer "unchanged".
394 *
395 * The logic does not apply to gitlinks, as ce_match_stat_basic()
396 * already has checked the actual HEAD from the filesystem in the
397 * subproject. If ie_match_stat() already said it is different,
398 * then we know it is.
399 */
400 if ((changed & DATA_CHANGED) &&
401 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
402 return changed;
403
404 changed_fs = ce_modified_check_fs(ce, st);
405 if (changed_fs)
406 return changed | changed_fs;
407 return 0;
408 }
409
410 int base_name_compare(const char *name1, int len1, int mode1,
411 const char *name2, int len2, int mode2)
412 {
413 unsigned char c1, c2;
414 int len = len1 < len2 ? len1 : len2;
415 int cmp;
416
417 cmp = memcmp(name1, name2, len);
418 if (cmp)
419 return cmp;
420 c1 = name1[len];
421 c2 = name2[len];
422 if (!c1 && S_ISDIR(mode1))
423 c1 = '/';
424 if (!c2 && S_ISDIR(mode2))
425 c2 = '/';
426 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
427 }
428
429 /*
430 * df_name_compare() is identical to base_name_compare(), except it
431 * compares conflicting directory/file entries as equal. Note that
432 * while a directory name compares as equal to a regular file, they
433 * then individually compare _differently_ to a filename that has
434 * a dot after the basename (because '\0' < '.' < '/').
435 *
436 * This is used by routines that want to traverse the git namespace
437 * but then handle conflicting entries together when possible.
438 */
439 int df_name_compare(const char *name1, int len1, int mode1,
440 const char *name2, int len2, int mode2)
441 {
442 int len = len1 < len2 ? len1 : len2, cmp;
443 unsigned char c1, c2;
444
445 cmp = memcmp(name1, name2, len);
446 if (cmp)
447 return cmp;
448 /* Directories and files compare equal (same length, same name) */
449 if (len1 == len2)
450 return 0;
451 c1 = name1[len];
452 if (!c1 && S_ISDIR(mode1))
453 c1 = '/';
454 c2 = name2[len];
455 if (!c2 && S_ISDIR(mode2))
456 c2 = '/';
457 if (c1 == '/' && !c2)
458 return 0;
459 if (c2 == '/' && !c1)
460 return 0;
461 return c1 - c2;
462 }
463
464 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
465 {
466 size_t min_len = (len1 < len2) ? len1 : len2;
467 int cmp = memcmp(name1, name2, min_len);
468 if (cmp)
469 return cmp;
470 if (len1 < len2)
471 return -1;
472 if (len1 > len2)
473 return 1;
474 return 0;
475 }
476
477 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
478 {
479 int cmp;
480
481 cmp = name_compare(name1, len1, name2, len2);
482 if (cmp)
483 return cmp;
484
485 if (stage1 < stage2)
486 return -1;
487 if (stage1 > stage2)
488 return 1;
489 return 0;
490 }
491
492 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
493 {
494 int first, last;
495
496 first = 0;
497 last = istate->cache_nr;
498 while (last > first) {
499 int next = (last + first) >> 1;
500 struct cache_entry *ce = istate->cache[next];
501 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
502 if (!cmp)
503 return next;
504 if (cmp < 0) {
505 last = next;
506 continue;
507 }
508 first = next+1;
509 }
510 return -first-1;
511 }
512
513 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
514 {
515 return index_name_stage_pos(istate, name, namelen, 0);
516 }
517
518 int remove_index_entry_at(struct index_state *istate, int pos)
519 {
520 struct cache_entry *ce = istate->cache[pos];
521
522 record_resolve_undo(istate, ce);
523 remove_name_hash(istate, ce);
524 save_or_free_index_entry(istate, ce);
525 istate->cache_changed |= CE_ENTRY_REMOVED;
526 istate->cache_nr--;
527 if (pos >= istate->cache_nr)
528 return 0;
529 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
530 istate->cache_nr - pos);
531 return 1;
532 }
533
534 /*
535 * Remove all cache entries marked for removal, that is where
536 * CE_REMOVE is set in ce_flags. This is much more effective than
537 * calling remove_index_entry_at() for each entry to be removed.
538 */
539 void remove_marked_cache_entries(struct index_state *istate)
540 {
541 struct cache_entry **ce_array = istate->cache;
542 unsigned int i, j;
543
544 for (i = j = 0; i < istate->cache_nr; i++) {
545 if (ce_array[i]->ce_flags & CE_REMOVE) {
546 remove_name_hash(istate, ce_array[i]);
547 save_or_free_index_entry(istate, ce_array[i]);
548 }
549 else
550 ce_array[j++] = ce_array[i];
551 }
552 if (j == istate->cache_nr)
553 return;
554 istate->cache_changed |= CE_ENTRY_REMOVED;
555 istate->cache_nr = j;
556 }
557
558 int remove_file_from_index(struct index_state *istate, const char *path)
559 {
560 int pos = index_name_pos(istate, path, strlen(path));
561 if (pos < 0)
562 pos = -pos-1;
563 cache_tree_invalidate_path(istate, path);
564 untracked_cache_remove_from_index(istate, path);
565 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
566 remove_index_entry_at(istate, pos);
567 return 0;
568 }
569
570 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
571 {
572 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
573 }
574
575 static int index_name_pos_also_unmerged(struct index_state *istate,
576 const char *path, int namelen)
577 {
578 int pos = index_name_pos(istate, path, namelen);
579 struct cache_entry *ce;
580
581 if (pos >= 0)
582 return pos;
583
584 /* maybe unmerged? */
585 pos = -1 - pos;
586 if (pos >= istate->cache_nr ||
587 compare_name((ce = istate->cache[pos]), path, namelen))
588 return -1;
589
590 /* order of preference: stage 2, 1, 3 */
591 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
592 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
593 !compare_name(ce, path, namelen))
594 pos++;
595 return pos;
596 }
597
598 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
599 {
600 int len = ce_namelen(ce);
601 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
602 }
603
604 /*
605 * If we add a filename that aliases in the cache, we will use the
606 * name that we already have - but we don't want to update the same
607 * alias twice, because that implies that there were actually two
608 * different files with aliasing names!
609 *
610 * So we use the CE_ADDED flag to verify that the alias was an old
611 * one before we accept it as
612 */
613 static struct cache_entry *create_alias_ce(struct index_state *istate,
614 struct cache_entry *ce,
615 struct cache_entry *alias)
616 {
617 int len;
618 struct cache_entry *new;
619
620 if (alias->ce_flags & CE_ADDED)
621 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
622
623 /* Ok, create the new entry using the name of the existing alias */
624 len = ce_namelen(alias);
625 new = xcalloc(1, cache_entry_size(len));
626 memcpy(new->name, alias->name, len);
627 copy_cache_entry(new, ce);
628 save_or_free_index_entry(istate, ce);
629 return new;
630 }
631
632 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
633 {
634 struct object_id oid;
635 if (write_object_file("", 0, blob_type, &oid))
636 die("cannot create an empty blob in the object database");
637 oidcpy(&ce->oid, &oid);
638 }
639
640 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
641 {
642 int size, namelen, was_same;
643 mode_t st_mode = st->st_mode;
644 struct cache_entry *ce, *alias = NULL;
645 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
646 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
647 int pretend = flags & ADD_CACHE_PRETEND;
648 int intent_only = flags & ADD_CACHE_INTENT;
649 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
650 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
651 int newflags = HASH_WRITE_OBJECT;
652
653 if (flags & HASH_RENORMALIZE)
654 newflags |= HASH_RENORMALIZE;
655
656 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
657 return error("%s: can only add regular files, symbolic links or git-directories", path);
658
659 namelen = strlen(path);
660 if (S_ISDIR(st_mode)) {
661 while (namelen && path[namelen-1] == '/')
662 namelen--;
663 }
664 size = cache_entry_size(namelen);
665 ce = xcalloc(1, size);
666 memcpy(ce->name, path, namelen);
667 ce->ce_namelen = namelen;
668 if (!intent_only)
669 fill_stat_cache_info(ce, st);
670 else
671 ce->ce_flags |= CE_INTENT_TO_ADD;
672
673
674 if (trust_executable_bit && has_symlinks) {
675 ce->ce_mode = create_ce_mode(st_mode);
676 } else {
677 /* If there is an existing entry, pick the mode bits and type
678 * from it, otherwise assume unexecutable regular file.
679 */
680 struct cache_entry *ent;
681 int pos = index_name_pos_also_unmerged(istate, path, namelen);
682
683 ent = (0 <= pos) ? istate->cache[pos] : NULL;
684 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
685 }
686
687 /* When core.ignorecase=true, determine if a directory of the same name but differing
688 * case already exists within the Git repository. If it does, ensure the directory
689 * case of the file being added to the repository matches (is folded into) the existing
690 * entry's directory case.
691 */
692 if (ignore_case) {
693 adjust_dirname_case(istate, ce->name);
694 }
695 if (!(flags & HASH_RENORMALIZE)) {
696 alias = index_file_exists(istate, ce->name,
697 ce_namelen(ce), ignore_case);
698 if (alias &&
699 !ce_stage(alias) &&
700 !ie_match_stat(istate, alias, st, ce_option)) {
701 /* Nothing changed, really */
702 if (!S_ISGITLINK(alias->ce_mode))
703 ce_mark_uptodate(alias);
704 alias->ce_flags |= CE_ADDED;
705
706 free(ce);
707 return 0;
708 }
709 }
710 if (!intent_only) {
711 if (index_path(&ce->oid, path, st, newflags)) {
712 free(ce);
713 return error("unable to index file %s", path);
714 }
715 } else
716 set_object_name_for_intent_to_add_entry(ce);
717
718 if (ignore_case && alias && different_name(ce, alias))
719 ce = create_alias_ce(istate, ce, alias);
720 ce->ce_flags |= CE_ADDED;
721
722 /* It was suspected to be racily clean, but it turns out to be Ok */
723 was_same = (alias &&
724 !ce_stage(alias) &&
725 !oidcmp(&alias->oid, &ce->oid) &&
726 ce->ce_mode == alias->ce_mode);
727
728 if (pretend)
729 free(ce);
730 else if (add_index_entry(istate, ce, add_option)) {
731 free(ce);
732 return error("unable to add %s to index", path);
733 }
734 if (verbose && !was_same)
735 printf("add '%s'\n", path);
736 return 0;
737 }
738
739 int add_file_to_index(struct index_state *istate, const char *path, int flags)
740 {
741 struct stat st;
742 if (lstat(path, &st))
743 die_errno("unable to stat '%s'", path);
744 return add_to_index(istate, path, &st, flags);
745 }
746
747 struct cache_entry *make_cache_entry(unsigned int mode,
748 const unsigned char *sha1, const char *path, int stage,
749 unsigned int refresh_options)
750 {
751 int size, len;
752 struct cache_entry *ce, *ret;
753
754 if (!verify_path(path)) {
755 error("Invalid path '%s'", path);
756 return NULL;
757 }
758
759 len = strlen(path);
760 size = cache_entry_size(len);
761 ce = xcalloc(1, size);
762
763 hashcpy(ce->oid.hash, sha1);
764 memcpy(ce->name, path, len);
765 ce->ce_flags = create_ce_flags(stage);
766 ce->ce_namelen = len;
767 ce->ce_mode = create_ce_mode(mode);
768
769 ret = refresh_cache_entry(ce, refresh_options);
770 if (ret != ce)
771 free(ce);
772 return ret;
773 }
774
775 /*
776 * Chmod an index entry with either +x or -x.
777 *
778 * Returns -1 if the chmod for the particular cache entry failed (if it's
779 * not a regular file), -2 if an invalid flip argument is passed in, 0
780 * otherwise.
781 */
782 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
783 char flip)
784 {
785 if (!S_ISREG(ce->ce_mode))
786 return -1;
787 switch (flip) {
788 case '+':
789 ce->ce_mode |= 0111;
790 break;
791 case '-':
792 ce->ce_mode &= ~0111;
793 break;
794 default:
795 return -2;
796 }
797 cache_tree_invalidate_path(istate, ce->name);
798 ce->ce_flags |= CE_UPDATE_IN_BASE;
799 mark_fsmonitor_invalid(istate, ce);
800 istate->cache_changed |= CE_ENTRY_CHANGED;
801
802 return 0;
803 }
804
805 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
806 {
807 int len = ce_namelen(a);
808 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
809 }
810
811 /*
812 * We fundamentally don't like some paths: we don't want
813 * dot or dot-dot anywhere, and for obvious reasons don't
814 * want to recurse into ".git" either.
815 *
816 * Also, we don't want double slashes or slashes at the
817 * end that can make pathnames ambiguous.
818 */
819 static int verify_dotfile(const char *rest)
820 {
821 /*
822 * The first character was '.', but that
823 * has already been discarded, we now test
824 * the rest.
825 */
826
827 /* "." is not allowed */
828 if (*rest == '\0' || is_dir_sep(*rest))
829 return 0;
830
831 switch (*rest) {
832 /*
833 * ".git" followed by NUL or slash is bad. This
834 * shares the path end test with the ".." case.
835 */
836 case 'g':
837 case 'G':
838 if (rest[1] != 'i' && rest[1] != 'I')
839 break;
840 if (rest[2] != 't' && rest[2] != 'T')
841 break;
842 rest += 2;
843 /* fallthrough */
844 case '.':
845 if (rest[1] == '\0' || is_dir_sep(rest[1]))
846 return 0;
847 }
848 return 1;
849 }
850
851 int verify_path(const char *path)
852 {
853 char c;
854
855 if (has_dos_drive_prefix(path))
856 return 0;
857
858 goto inside;
859 for (;;) {
860 if (!c)
861 return 1;
862 if (is_dir_sep(c)) {
863 inside:
864 if (protect_hfs && is_hfs_dotgit(path))
865 return 0;
866 if (protect_ntfs && is_ntfs_dotgit(path))
867 return 0;
868 c = *path++;
869 if ((c == '.' && !verify_dotfile(path)) ||
870 is_dir_sep(c) || c == '\0')
871 return 0;
872 }
873 c = *path++;
874 }
875 }
876
877 /*
878 * Do we have another file that has the beginning components being a
879 * proper superset of the name we're trying to add?
880 */
881 static int has_file_name(struct index_state *istate,
882 const struct cache_entry *ce, int pos, int ok_to_replace)
883 {
884 int retval = 0;
885 int len = ce_namelen(ce);
886 int stage = ce_stage(ce);
887 const char *name = ce->name;
888
889 while (pos < istate->cache_nr) {
890 struct cache_entry *p = istate->cache[pos++];
891
892 if (len >= ce_namelen(p))
893 break;
894 if (memcmp(name, p->name, len))
895 break;
896 if (ce_stage(p) != stage)
897 continue;
898 if (p->name[len] != '/')
899 continue;
900 if (p->ce_flags & CE_REMOVE)
901 continue;
902 retval = -1;
903 if (!ok_to_replace)
904 break;
905 remove_index_entry_at(istate, --pos);
906 }
907 return retval;
908 }
909
910
911 /*
912 * Like strcmp(), but also return the offset of the first change.
913 * If strings are equal, return the length.
914 */
915 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
916 {
917 size_t k;
918
919 if (!first_change)
920 return strcmp(s1, s2);
921
922 for (k = 0; s1[k] == s2[k]; k++)
923 if (s1[k] == '\0')
924 break;
925
926 *first_change = k;
927 return (unsigned char)s1[k] - (unsigned char)s2[k];
928 }
929
930 /*
931 * Do we have another file with a pathname that is a proper
932 * subset of the name we're trying to add?
933 *
934 * That is, is there another file in the index with a path
935 * that matches a sub-directory in the given entry?
936 */
937 static int has_dir_name(struct index_state *istate,
938 const struct cache_entry *ce, int pos, int ok_to_replace)
939 {
940 int retval = 0;
941 int stage = ce_stage(ce);
942 const char *name = ce->name;
943 const char *slash = name + ce_namelen(ce);
944 size_t len_eq_last;
945 int cmp_last = 0;
946
947 /*
948 * We are frequently called during an iteration on a sorted
949 * list of pathnames and while building a new index. Therefore,
950 * there is a high probability that this entry will eventually
951 * be appended to the index, rather than inserted in the middle.
952 * If we can confirm that, we can avoid binary searches on the
953 * components of the pathname.
954 *
955 * Compare the entry's full path with the last path in the index.
956 */
957 if (istate->cache_nr > 0) {
958 cmp_last = strcmp_offset(name,
959 istate->cache[istate->cache_nr - 1]->name,
960 &len_eq_last);
961 if (cmp_last > 0) {
962 if (len_eq_last == 0) {
963 /*
964 * The entry sorts AFTER the last one in the
965 * index and their paths have no common prefix,
966 * so there cannot be a F/D conflict.
967 */
968 return retval;
969 } else {
970 /*
971 * The entry sorts AFTER the last one in the
972 * index, but has a common prefix. Fall through
973 * to the loop below to disect the entry's path
974 * and see where the difference is.
975 */
976 }
977 } else if (cmp_last == 0) {
978 /*
979 * The entry exactly matches the last one in the
980 * index, but because of multiple stage and CE_REMOVE
981 * items, we fall through and let the regular search
982 * code handle it.
983 */
984 }
985 }
986
987 for (;;) {
988 size_t len;
989
990 for (;;) {
991 if (*--slash == '/')
992 break;
993 if (slash <= ce->name)
994 return retval;
995 }
996 len = slash - name;
997
998 if (cmp_last > 0) {
999 /*
1000 * (len + 1) is a directory boundary (including
1001 * the trailing slash). And since the loop is
1002 * decrementing "slash", the first iteration is
1003 * the longest directory prefix; subsequent
1004 * iterations consider parent directories.
1005 */
1006
1007 if (len + 1 <= len_eq_last) {
1008 /*
1009 * The directory prefix (including the trailing
1010 * slash) also appears as a prefix in the last
1011 * entry, so the remainder cannot collide (because
1012 * strcmp said the whole path was greater).
1013 *
1014 * EQ: last: xxx/A
1015 * this: xxx/B
1016 *
1017 * LT: last: xxx/file_A
1018 * this: xxx/file_B
1019 */
1020 return retval;
1021 }
1022
1023 if (len > len_eq_last) {
1024 /*
1025 * This part of the directory prefix (excluding
1026 * the trailing slash) is longer than the known
1027 * equal portions, so this sub-directory cannot
1028 * collide with a file.
1029 *
1030 * GT: last: xxxA
1031 * this: xxxB/file
1032 */
1033 return retval;
1034 }
1035
1036 if (istate->cache_nr > 0 &&
1037 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1038 /*
1039 * The directory prefix lines up with part of
1040 * a longer file or directory name, but sorts
1041 * after it, so this sub-directory cannot
1042 * collide with a file.
1043 *
1044 * last: xxx/yy-file (because '-' sorts before '/')
1045 * this: xxx/yy/abc
1046 */
1047 return retval;
1048 }
1049
1050 /*
1051 * This is a possible collision. Fall through and
1052 * let the regular search code handle it.
1053 *
1054 * last: xxx
1055 * this: xxx/file
1056 */
1057 }
1058
1059 pos = index_name_stage_pos(istate, name, len, stage);
1060 if (pos >= 0) {
1061 /*
1062 * Found one, but not so fast. This could
1063 * be a marker that says "I was here, but
1064 * I am being removed". Such an entry is
1065 * not a part of the resulting tree, and
1066 * it is Ok to have a directory at the same
1067 * path.
1068 */
1069 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1070 retval = -1;
1071 if (!ok_to_replace)
1072 break;
1073 remove_index_entry_at(istate, pos);
1074 continue;
1075 }
1076 }
1077 else
1078 pos = -pos-1;
1079
1080 /*
1081 * Trivial optimization: if we find an entry that
1082 * already matches the sub-directory, then we know
1083 * we're ok, and we can exit.
1084 */
1085 while (pos < istate->cache_nr) {
1086 struct cache_entry *p = istate->cache[pos];
1087 if ((ce_namelen(p) <= len) ||
1088 (p->name[len] != '/') ||
1089 memcmp(p->name, name, len))
1090 break; /* not our subdirectory */
1091 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1092 /*
1093 * p is at the same stage as our entry, and
1094 * is a subdirectory of what we are looking
1095 * at, so we cannot have conflicts at our
1096 * level or anything shorter.
1097 */
1098 return retval;
1099 pos++;
1100 }
1101 }
1102 return retval;
1103 }
1104
1105 /* We may be in a situation where we already have path/file and path
1106 * is being added, or we already have path and path/file is being
1107 * added. Either one would result in a nonsense tree that has path
1108 * twice when git-write-tree tries to write it out. Prevent it.
1109 *
1110 * If ok-to-replace is specified, we remove the conflicting entries
1111 * from the cache so the caller should recompute the insert position.
1112 * When this happens, we return non-zero.
1113 */
1114 static int check_file_directory_conflict(struct index_state *istate,
1115 const struct cache_entry *ce,
1116 int pos, int ok_to_replace)
1117 {
1118 int retval;
1119
1120 /*
1121 * When ce is an "I am going away" entry, we allow it to be added
1122 */
1123 if (ce->ce_flags & CE_REMOVE)
1124 return 0;
1125
1126 /*
1127 * We check if the path is a sub-path of a subsequent pathname
1128 * first, since removing those will not change the position
1129 * in the array.
1130 */
1131 retval = has_file_name(istate, ce, pos, ok_to_replace);
1132
1133 /*
1134 * Then check if the path might have a clashing sub-directory
1135 * before it.
1136 */
1137 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1138 }
1139
1140 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1141 {
1142 int pos;
1143 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1144 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1145 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1146 int new_only = option & ADD_CACHE_NEW_ONLY;
1147
1148 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1149 cache_tree_invalidate_path(istate, ce->name);
1150
1151 /*
1152 * If this entry's path sorts after the last entry in the index,
1153 * we can avoid searching for it.
1154 */
1155 if (istate->cache_nr > 0 &&
1156 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1157 pos = -istate->cache_nr - 1;
1158 else
1159 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1160
1161 /* existing match? Just replace it. */
1162 if (pos >= 0) {
1163 if (!new_only)
1164 replace_index_entry(istate, pos, ce);
1165 return 0;
1166 }
1167 pos = -pos-1;
1168
1169 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1170 untracked_cache_add_to_index(istate, ce->name);
1171
1172 /*
1173 * Inserting a merged entry ("stage 0") into the index
1174 * will always replace all non-merged entries..
1175 */
1176 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1177 while (ce_same_name(istate->cache[pos], ce)) {
1178 ok_to_add = 1;
1179 if (!remove_index_entry_at(istate, pos))
1180 break;
1181 }
1182 }
1183
1184 if (!ok_to_add)
1185 return -1;
1186 if (!verify_path(ce->name))
1187 return error("Invalid path '%s'", ce->name);
1188
1189 if (!skip_df_check &&
1190 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1191 if (!ok_to_replace)
1192 return error("'%s' appears as both a file and as a directory",
1193 ce->name);
1194 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1195 pos = -pos-1;
1196 }
1197 return pos + 1;
1198 }
1199
1200 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1201 {
1202 int pos;
1203
1204 if (option & ADD_CACHE_JUST_APPEND)
1205 pos = istate->cache_nr;
1206 else {
1207 int ret;
1208 ret = add_index_entry_with_check(istate, ce, option);
1209 if (ret <= 0)
1210 return ret;
1211 pos = ret - 1;
1212 }
1213
1214 /* Make sure the array is big enough .. */
1215 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1216
1217 /* Add it in.. */
1218 istate->cache_nr++;
1219 if (istate->cache_nr > pos + 1)
1220 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1221 istate->cache_nr - pos - 1);
1222 set_index_entry(istate, pos, ce);
1223 istate->cache_changed |= CE_ENTRY_ADDED;
1224 return 0;
1225 }
1226
1227 /*
1228 * "refresh" does not calculate a new sha1 file or bring the
1229 * cache up-to-date for mode/content changes. But what it
1230 * _does_ do is to "re-match" the stat information of a file
1231 * with the cache, so that you can refresh the cache for a
1232 * file that hasn't been changed but where the stat entry is
1233 * out of date.
1234 *
1235 * For example, you'd want to do this after doing a "git-read-tree",
1236 * to link up the stat cache details with the proper files.
1237 */
1238 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1239 struct cache_entry *ce,
1240 unsigned int options, int *err,
1241 int *changed_ret)
1242 {
1243 struct stat st;
1244 struct cache_entry *updated;
1245 int changed, size;
1246 int refresh = options & CE_MATCH_REFRESH;
1247 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1248 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1249 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1250 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1251
1252 if (!refresh || ce_uptodate(ce))
1253 return ce;
1254
1255 if (!ignore_fsmonitor)
1256 refresh_fsmonitor(istate);
1257 /*
1258 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1259 * that the change to the work tree does not matter and told
1260 * us not to worry.
1261 */
1262 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1263 ce_mark_uptodate(ce);
1264 return ce;
1265 }
1266 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1267 ce_mark_uptodate(ce);
1268 return ce;
1269 }
1270 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1271 ce_mark_uptodate(ce);
1272 return ce;
1273 }
1274
1275 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1276 if (ignore_missing)
1277 return ce;
1278 if (err)
1279 *err = ENOENT;
1280 return NULL;
1281 }
1282
1283 if (lstat(ce->name, &st) < 0) {
1284 if (ignore_missing && errno == ENOENT)
1285 return ce;
1286 if (err)
1287 *err = errno;
1288 return NULL;
1289 }
1290
1291 changed = ie_match_stat(istate, ce, &st, options);
1292 if (changed_ret)
1293 *changed_ret = changed;
1294 if (!changed) {
1295 /*
1296 * The path is unchanged. If we were told to ignore
1297 * valid bit, then we did the actual stat check and
1298 * found that the entry is unmodified. If the entry
1299 * is not marked VALID, this is the place to mark it
1300 * valid again, under "assume unchanged" mode.
1301 */
1302 if (ignore_valid && assume_unchanged &&
1303 !(ce->ce_flags & CE_VALID))
1304 ; /* mark this one VALID again */
1305 else {
1306 /*
1307 * We do not mark the index itself "modified"
1308 * because CE_UPTODATE flag is in-core only;
1309 * we are not going to write this change out.
1310 */
1311 if (!S_ISGITLINK(ce->ce_mode)) {
1312 ce_mark_uptodate(ce);
1313 mark_fsmonitor_valid(ce);
1314 }
1315 return ce;
1316 }
1317 }
1318
1319 if (ie_modified(istate, ce, &st, options)) {
1320 if (err)
1321 *err = EINVAL;
1322 return NULL;
1323 }
1324
1325 size = ce_size(ce);
1326 updated = xmalloc(size);
1327 memcpy(updated, ce, size);
1328 fill_stat_cache_info(updated, &st);
1329 /*
1330 * If ignore_valid is not set, we should leave CE_VALID bit
1331 * alone. Otherwise, paths marked with --no-assume-unchanged
1332 * (i.e. things to be edited) will reacquire CE_VALID bit
1333 * automatically, which is not really what we want.
1334 */
1335 if (!ignore_valid && assume_unchanged &&
1336 !(ce->ce_flags & CE_VALID))
1337 updated->ce_flags &= ~CE_VALID;
1338
1339 /* istate->cache_changed is updated in the caller */
1340 return updated;
1341 }
1342
1343 static void show_file(const char * fmt, const char * name, int in_porcelain,
1344 int * first, const char *header_msg)
1345 {
1346 if (in_porcelain && *first && header_msg) {
1347 printf("%s\n", header_msg);
1348 *first = 0;
1349 }
1350 printf(fmt, name);
1351 }
1352
1353 int refresh_index(struct index_state *istate, unsigned int flags,
1354 const struct pathspec *pathspec,
1355 char *seen, const char *header_msg)
1356 {
1357 int i;
1358 int has_errors = 0;
1359 int really = (flags & REFRESH_REALLY) != 0;
1360 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1361 int quiet = (flags & REFRESH_QUIET) != 0;
1362 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1363 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1364 int first = 1;
1365 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1366 unsigned int options = (CE_MATCH_REFRESH |
1367 (really ? CE_MATCH_IGNORE_VALID : 0) |
1368 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1369 const char *modified_fmt;
1370 const char *deleted_fmt;
1371 const char *typechange_fmt;
1372 const char *added_fmt;
1373 const char *unmerged_fmt;
1374 uint64_t start = getnanotime();
1375
1376 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1377 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1378 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1379 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1380 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1381 for (i = 0; i < istate->cache_nr; i++) {
1382 struct cache_entry *ce, *new;
1383 int cache_errno = 0;
1384 int changed = 0;
1385 int filtered = 0;
1386
1387 ce = istate->cache[i];
1388 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1389 continue;
1390
1391 if (pathspec && !ce_path_match(ce, pathspec, seen))
1392 filtered = 1;
1393
1394 if (ce_stage(ce)) {
1395 while ((i < istate->cache_nr) &&
1396 ! strcmp(istate->cache[i]->name, ce->name))
1397 i++;
1398 i--;
1399 if (allow_unmerged)
1400 continue;
1401 if (!filtered)
1402 show_file(unmerged_fmt, ce->name, in_porcelain,
1403 &first, header_msg);
1404 has_errors = 1;
1405 continue;
1406 }
1407
1408 if (filtered)
1409 continue;
1410
1411 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1412 if (new == ce)
1413 continue;
1414 if (!new) {
1415 const char *fmt;
1416
1417 if (really && cache_errno == EINVAL) {
1418 /* If we are doing --really-refresh that
1419 * means the index is not valid anymore.
1420 */
1421 ce->ce_flags &= ~CE_VALID;
1422 ce->ce_flags |= CE_UPDATE_IN_BASE;
1423 mark_fsmonitor_invalid(istate, ce);
1424 istate->cache_changed |= CE_ENTRY_CHANGED;
1425 }
1426 if (quiet)
1427 continue;
1428
1429 if (cache_errno == ENOENT)
1430 fmt = deleted_fmt;
1431 else if (ce_intent_to_add(ce))
1432 fmt = added_fmt; /* must be before other checks */
1433 else if (changed & TYPE_CHANGED)
1434 fmt = typechange_fmt;
1435 else
1436 fmt = modified_fmt;
1437 show_file(fmt,
1438 ce->name, in_porcelain, &first, header_msg);
1439 has_errors = 1;
1440 continue;
1441 }
1442
1443 replace_index_entry(istate, i, new);
1444 }
1445 trace_performance_since(start, "refresh index");
1446 return has_errors;
1447 }
1448
1449 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1450 unsigned int options)
1451 {
1452 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1453 }
1454
1455
1456 /*****************************************************************
1457 * Index File I/O
1458 *****************************************************************/
1459
1460 #define INDEX_FORMAT_DEFAULT 3
1461
1462 static unsigned int get_index_format_default(void)
1463 {
1464 char *envversion = getenv("GIT_INDEX_VERSION");
1465 char *endp;
1466 int value;
1467 unsigned int version = INDEX_FORMAT_DEFAULT;
1468
1469 if (!envversion) {
1470 if (!git_config_get_int("index.version", &value))
1471 version = value;
1472 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1473 warning(_("index.version set, but the value is invalid.\n"
1474 "Using version %i"), INDEX_FORMAT_DEFAULT);
1475 return INDEX_FORMAT_DEFAULT;
1476 }
1477 return version;
1478 }
1479
1480 version = strtoul(envversion, &endp, 10);
1481 if (*endp ||
1482 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1483 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1484 "Using version %i"), INDEX_FORMAT_DEFAULT);
1485 version = INDEX_FORMAT_DEFAULT;
1486 }
1487 return version;
1488 }
1489
1490 /*
1491 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1492 * Again - this is just a (very strong in practice) heuristic that
1493 * the inode hasn't changed.
1494 *
1495 * We save the fields in big-endian order to allow using the
1496 * index file over NFS transparently.
1497 */
1498 struct ondisk_cache_entry {
1499 struct cache_time ctime;
1500 struct cache_time mtime;
1501 uint32_t dev;
1502 uint32_t ino;
1503 uint32_t mode;
1504 uint32_t uid;
1505 uint32_t gid;
1506 uint32_t size;
1507 unsigned char sha1[20];
1508 uint16_t flags;
1509 char name[FLEX_ARRAY]; /* more */
1510 };
1511
1512 /*
1513 * This struct is used when CE_EXTENDED bit is 1
1514 * The struct must match ondisk_cache_entry exactly from
1515 * ctime till flags
1516 */
1517 struct ondisk_cache_entry_extended {
1518 struct cache_time ctime;
1519 struct cache_time mtime;
1520 uint32_t dev;
1521 uint32_t ino;
1522 uint32_t mode;
1523 uint32_t uid;
1524 uint32_t gid;
1525 uint32_t size;
1526 unsigned char sha1[20];
1527 uint16_t flags;
1528 uint16_t flags2;
1529 char name[FLEX_ARRAY]; /* more */
1530 };
1531
1532 /* These are only used for v3 or lower */
1533 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1534 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1535 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1536 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1537 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1538 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1539 ondisk_cache_entry_size(ce_namelen(ce)))
1540
1541 /* Allow fsck to force verification of the index checksum. */
1542 int verify_index_checksum;
1543
1544 /* Allow fsck to force verification of the cache entry order. */
1545 int verify_ce_order;
1546
1547 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1548 {
1549 git_hash_ctx c;
1550 unsigned char hash[GIT_MAX_RAWSZ];
1551 int hdr_version;
1552
1553 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1554 return error("bad signature");
1555 hdr_version = ntohl(hdr->hdr_version);
1556 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1557 return error("bad index version %d", hdr_version);
1558
1559 if (!verify_index_checksum)
1560 return 0;
1561
1562 the_hash_algo->init_fn(&c);
1563 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1564 the_hash_algo->final_fn(hash, &c);
1565 if (hashcmp(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1566 return error("bad index file sha1 signature");
1567 return 0;
1568 }
1569
1570 static int read_index_extension(struct index_state *istate,
1571 const char *ext, void *data, unsigned long sz)
1572 {
1573 switch (CACHE_EXT(ext)) {
1574 case CACHE_EXT_TREE:
1575 istate->cache_tree = cache_tree_read(data, sz);
1576 break;
1577 case CACHE_EXT_RESOLVE_UNDO:
1578 istate->resolve_undo = resolve_undo_read(data, sz);
1579 break;
1580 case CACHE_EXT_LINK:
1581 if (read_link_extension(istate, data, sz))
1582 return -1;
1583 break;
1584 case CACHE_EXT_UNTRACKED:
1585 istate->untracked = read_untracked_extension(data, sz);
1586 break;
1587 case CACHE_EXT_FSMONITOR:
1588 read_fsmonitor_extension(istate, data, sz);
1589 break;
1590 default:
1591 if (*ext < 'A' || 'Z' < *ext)
1592 return error("index uses %.4s extension, which we do not understand",
1593 ext);
1594 fprintf(stderr, "ignoring %.4s extension\n", ext);
1595 break;
1596 }
1597 return 0;
1598 }
1599
1600 int hold_locked_index(struct lock_file *lk, int lock_flags)
1601 {
1602 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1603 }
1604
1605 int read_index(struct index_state *istate)
1606 {
1607 return read_index_from(istate, get_index_file(), get_git_dir());
1608 }
1609
1610 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1611 unsigned int flags,
1612 const char *name,
1613 size_t len)
1614 {
1615 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1616
1617 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1618 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1619 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1620 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1621 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1622 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1623 ce->ce_mode = get_be32(&ondisk->mode);
1624 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1625 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1626 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1627 ce->ce_flags = flags & ~CE_NAMEMASK;
1628 ce->ce_namelen = len;
1629 ce->index = 0;
1630 hashcpy(ce->oid.hash, ondisk->sha1);
1631 memcpy(ce->name, name, len);
1632 ce->name[len] = '\0';
1633 return ce;
1634 }
1635
1636 /*
1637 * Adjacent cache entries tend to share the leading paths, so it makes
1638 * sense to only store the differences in later entries. In the v4
1639 * on-disk format of the index, each on-disk cache entry stores the
1640 * number of bytes to be stripped from the end of the previous name,
1641 * and the bytes to append to the result, to come up with its name.
1642 */
1643 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1644 {
1645 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1646 size_t len = decode_varint(&cp);
1647
1648 if (name->len < len)
1649 die("malformed name field in the index");
1650 strbuf_remove(name, name->len - len, len);
1651 for (ep = cp; *ep; ep++)
1652 ; /* find the end */
1653 strbuf_add(name, cp, ep - cp);
1654 return (const char *)ep + 1 - cp_;
1655 }
1656
1657 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1658 unsigned long *ent_size,
1659 struct strbuf *previous_name)
1660 {
1661 struct cache_entry *ce;
1662 size_t len;
1663 const char *name;
1664 unsigned int flags;
1665
1666 /* On-disk flags are just 16 bits */
1667 flags = get_be16(&ondisk->flags);
1668 len = flags & CE_NAMEMASK;
1669
1670 if (flags & CE_EXTENDED) {
1671 struct ondisk_cache_entry_extended *ondisk2;
1672 int extended_flags;
1673 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1674 extended_flags = get_be16(&ondisk2->flags2) << 16;
1675 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1676 if (extended_flags & ~CE_EXTENDED_FLAGS)
1677 die("Unknown index entry format %08x", extended_flags);
1678 flags |= extended_flags;
1679 name = ondisk2->name;
1680 }
1681 else
1682 name = ondisk->name;
1683
1684 if (!previous_name) {
1685 /* v3 and earlier */
1686 if (len == CE_NAMEMASK)
1687 len = strlen(name);
1688 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1689
1690 *ent_size = ondisk_ce_size(ce);
1691 } else {
1692 unsigned long consumed;
1693 consumed = expand_name_field(previous_name, name);
1694 ce = cache_entry_from_ondisk(ondisk, flags,
1695 previous_name->buf,
1696 previous_name->len);
1697
1698 *ent_size = (name - ((char *)ondisk)) + consumed;
1699 }
1700 return ce;
1701 }
1702
1703 static void check_ce_order(struct index_state *istate)
1704 {
1705 unsigned int i;
1706
1707 if (!verify_ce_order)
1708 return;
1709
1710 for (i = 1; i < istate->cache_nr; i++) {
1711 struct cache_entry *ce = istate->cache[i - 1];
1712 struct cache_entry *next_ce = istate->cache[i];
1713 int name_compare = strcmp(ce->name, next_ce->name);
1714
1715 if (0 < name_compare)
1716 die("unordered stage entries in index");
1717 if (!name_compare) {
1718 if (!ce_stage(ce))
1719 die("multiple stage entries for merged file '%s'",
1720 ce->name);
1721 if (ce_stage(ce) > ce_stage(next_ce))
1722 die("unordered stage entries for '%s'",
1723 ce->name);
1724 }
1725 }
1726 }
1727
1728 static void tweak_untracked_cache(struct index_state *istate)
1729 {
1730 switch (git_config_get_untracked_cache()) {
1731 case -1: /* keep: do nothing */
1732 break;
1733 case 0: /* false */
1734 remove_untracked_cache(istate);
1735 break;
1736 case 1: /* true */
1737 add_untracked_cache(istate);
1738 break;
1739 default: /* unknown value: do nothing */
1740 break;
1741 }
1742 }
1743
1744 static void tweak_split_index(struct index_state *istate)
1745 {
1746 switch (git_config_get_split_index()) {
1747 case -1: /* unset: do nothing */
1748 break;
1749 case 0: /* false */
1750 remove_split_index(istate);
1751 break;
1752 case 1: /* true */
1753 add_split_index(istate);
1754 break;
1755 default: /* unknown value: do nothing */
1756 break;
1757 }
1758 }
1759
1760 static void post_read_index_from(struct index_state *istate)
1761 {
1762 check_ce_order(istate);
1763 tweak_untracked_cache(istate);
1764 tweak_split_index(istate);
1765 tweak_fsmonitor(istate);
1766 }
1767
1768 /* remember to discard_cache() before reading a different cache! */
1769 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1770 {
1771 int fd, i;
1772 struct stat st;
1773 unsigned long src_offset;
1774 struct cache_header *hdr;
1775 void *mmap;
1776 size_t mmap_size;
1777 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1778
1779 if (istate->initialized)
1780 return istate->cache_nr;
1781
1782 istate->timestamp.sec = 0;
1783 istate->timestamp.nsec = 0;
1784 fd = open(path, O_RDONLY);
1785 if (fd < 0) {
1786 if (!must_exist && errno == ENOENT)
1787 return 0;
1788 die_errno("%s: index file open failed", path);
1789 }
1790
1791 if (fstat(fd, &st))
1792 die_errno("cannot stat the open index");
1793
1794 mmap_size = xsize_t(st.st_size);
1795 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
1796 die("index file smaller than expected");
1797
1798 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1799 if (mmap == MAP_FAILED)
1800 die_errno("unable to map index file");
1801 close(fd);
1802
1803 hdr = mmap;
1804 if (verify_hdr(hdr, mmap_size) < 0)
1805 goto unmap;
1806
1807 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
1808 istate->version = ntohl(hdr->hdr_version);
1809 istate->cache_nr = ntohl(hdr->hdr_entries);
1810 istate->cache_alloc = alloc_nr(istate->cache_nr);
1811 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1812 istate->initialized = 1;
1813
1814 if (istate->version == 4)
1815 previous_name = &previous_name_buf;
1816 else
1817 previous_name = NULL;
1818
1819 src_offset = sizeof(*hdr);
1820 for (i = 0; i < istate->cache_nr; i++) {
1821 struct ondisk_cache_entry *disk_ce;
1822 struct cache_entry *ce;
1823 unsigned long consumed;
1824
1825 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1826 ce = create_from_disk(disk_ce, &consumed, previous_name);
1827 set_index_entry(istate, i, ce);
1828
1829 src_offset += consumed;
1830 }
1831 strbuf_release(&previous_name_buf);
1832 istate->timestamp.sec = st.st_mtime;
1833 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1834
1835 while (src_offset <= mmap_size - the_hash_algo->rawsz - 8) {
1836 /* After an array of active_nr index entries,
1837 * there can be arbitrary number of extended
1838 * sections, each of which is prefixed with
1839 * extension name (4-byte) and section length
1840 * in 4-byte network byte order.
1841 */
1842 uint32_t extsize;
1843 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1844 extsize = ntohl(extsize);
1845 if (read_index_extension(istate,
1846 (const char *) mmap + src_offset,
1847 (char *) mmap + src_offset + 8,
1848 extsize) < 0)
1849 goto unmap;
1850 src_offset += 8;
1851 src_offset += extsize;
1852 }
1853 munmap(mmap, mmap_size);
1854 return istate->cache_nr;
1855
1856 unmap:
1857 munmap(mmap, mmap_size);
1858 die("index file corrupt");
1859 }
1860
1861 /*
1862 * Signal that the shared index is used by updating its mtime.
1863 *
1864 * This way, shared index can be removed if they have not been used
1865 * for some time.
1866 */
1867 static void freshen_shared_index(const char *shared_index, int warn)
1868 {
1869 if (!check_and_freshen_file(shared_index, 1) && warn)
1870 warning("could not freshen shared index '%s'", shared_index);
1871 }
1872
1873 int read_index_from(struct index_state *istate, const char *path,
1874 const char *gitdir)
1875 {
1876 uint64_t start = getnanotime();
1877 struct split_index *split_index;
1878 int ret;
1879 char *base_sha1_hex;
1880 char *base_path;
1881
1882 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1883 if (istate->initialized)
1884 return istate->cache_nr;
1885
1886 ret = do_read_index(istate, path, 0);
1887 trace_performance_since(start, "read cache %s", path);
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 = xstrfmt("%s/sharedindex.%s", gitdir, 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_path, 0);
1909 merge_base_index(istate);
1910 post_read_index_from(istate);
1911 trace_performance_since(start, "read cache %s", base_path);
1912 free(base_path);
1913 return ret;
1914 }
1915
1916 int is_index_unborn(struct index_state *istate)
1917 {
1918 return (!istate->cache_nr && !istate->timestamp.sec);
1919 }
1920
1921 int discard_index(struct index_state *istate)
1922 {
1923 int i;
1924
1925 for (i = 0; i < istate->cache_nr; i++) {
1926 if (istate->cache[i]->index &&
1927 istate->split_index &&
1928 istate->split_index->base &&
1929 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1930 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1931 continue;
1932 free(istate->cache[i]);
1933 }
1934 resolve_undo_clear_index(istate);
1935 istate->cache_nr = 0;
1936 istate->cache_changed = 0;
1937 istate->timestamp.sec = 0;
1938 istate->timestamp.nsec = 0;
1939 free_name_hash(istate);
1940 cache_tree_free(&(istate->cache_tree));
1941 istate->initialized = 0;
1942 FREE_AND_NULL(istate->cache);
1943 istate->cache_alloc = 0;
1944 discard_split_index(istate);
1945 free_untracked_cache(istate->untracked);
1946 istate->untracked = NULL;
1947 return 0;
1948 }
1949
1950 int unmerged_index(const struct index_state *istate)
1951 {
1952 int i;
1953 for (i = 0; i < istate->cache_nr; i++) {
1954 if (ce_stage(istate->cache[i]))
1955 return 1;
1956 }
1957 return 0;
1958 }
1959
1960 #define WRITE_BUFFER_SIZE 8192
1961 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1962 static unsigned long write_buffer_len;
1963
1964 static int ce_write_flush(git_hash_ctx *context, int fd)
1965 {
1966 unsigned int buffered = write_buffer_len;
1967 if (buffered) {
1968 the_hash_algo->update_fn(context, write_buffer, buffered);
1969 if (write_in_full(fd, write_buffer, buffered) < 0)
1970 return -1;
1971 write_buffer_len = 0;
1972 }
1973 return 0;
1974 }
1975
1976 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
1977 {
1978 while (len) {
1979 unsigned int buffered = write_buffer_len;
1980 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1981 if (partial > len)
1982 partial = len;
1983 memcpy(write_buffer + buffered, data, partial);
1984 buffered += partial;
1985 if (buffered == WRITE_BUFFER_SIZE) {
1986 write_buffer_len = buffered;
1987 if (ce_write_flush(context, fd))
1988 return -1;
1989 buffered = 0;
1990 }
1991 write_buffer_len = buffered;
1992 len -= partial;
1993 data = (char *) data + partial;
1994 }
1995 return 0;
1996 }
1997
1998 static int write_index_ext_header(git_hash_ctx *context, int fd,
1999 unsigned int ext, unsigned int sz)
2000 {
2001 ext = htonl(ext);
2002 sz = htonl(sz);
2003 return ((ce_write(context, fd, &ext, 4) < 0) ||
2004 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2005 }
2006
2007 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2008 {
2009 unsigned int left = write_buffer_len;
2010
2011 if (left) {
2012 write_buffer_len = 0;
2013 the_hash_algo->update_fn(context, write_buffer, left);
2014 }
2015
2016 /* Flush first if not enough space for hash signature */
2017 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2018 if (write_in_full(fd, write_buffer, left) < 0)
2019 return -1;
2020 left = 0;
2021 }
2022
2023 /* Append the hash signature at the end */
2024 the_hash_algo->final_fn(write_buffer + left, context);
2025 hashcpy(hash, write_buffer + left);
2026 left += the_hash_algo->rawsz;
2027 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2028 }
2029
2030 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2031 {
2032 /*
2033 * The only thing we care about in this function is to smudge the
2034 * falsely clean entry due to touch-update-touch race, so we leave
2035 * everything else as they are. We are called for entries whose
2036 * ce_stat_data.sd_mtime match the index file mtime.
2037 *
2038 * Note that this actually does not do much for gitlinks, for
2039 * which ce_match_stat_basic() always goes to the actual
2040 * contents. The caller checks with is_racy_timestamp() which
2041 * always says "no" for gitlinks, so we are not called for them ;-)
2042 */
2043 struct stat st;
2044
2045 if (lstat(ce->name, &st) < 0)
2046 return;
2047 if (ce_match_stat_basic(ce, &st))
2048 return;
2049 if (ce_modified_check_fs(ce, &st)) {
2050 /* This is "racily clean"; smudge it. Note that this
2051 * is a tricky code. At first glance, it may appear
2052 * that it can break with this sequence:
2053 *
2054 * $ echo xyzzy >frotz
2055 * $ git-update-index --add frotz
2056 * $ : >frotz
2057 * $ sleep 3
2058 * $ echo filfre >nitfol
2059 * $ git-update-index --add nitfol
2060 *
2061 * but it does not. When the second update-index runs,
2062 * it notices that the entry "frotz" has the same timestamp
2063 * as index, and if we were to smudge it by resetting its
2064 * size to zero here, then the object name recorded
2065 * in index is the 6-byte file but the cached stat information
2066 * becomes zero --- which would then match what we would
2067 * obtain from the filesystem next time we stat("frotz").
2068 *
2069 * However, the second update-index, before calling
2070 * this function, notices that the cached size is 6
2071 * bytes and what is on the filesystem is an empty
2072 * file, and never calls us, so the cached size information
2073 * for "frotz" stays 6 which does not match the filesystem.
2074 */
2075 ce->ce_stat_data.sd_size = 0;
2076 }
2077 }
2078
2079 /* Copy miscellaneous fields but not the name */
2080 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2081 struct cache_entry *ce)
2082 {
2083 short flags;
2084
2085 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2086 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2087 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2088 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2089 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2090 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2091 ondisk->mode = htonl(ce->ce_mode);
2092 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2093 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2094 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2095 hashcpy(ondisk->sha1, ce->oid.hash);
2096
2097 flags = ce->ce_flags & ~CE_NAMEMASK;
2098 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2099 ondisk->flags = htons(flags);
2100 if (ce->ce_flags & CE_EXTENDED) {
2101 struct ondisk_cache_entry_extended *ondisk2;
2102 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2103 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2104 }
2105 }
2106
2107 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2108 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2109 {
2110 int size;
2111 int saved_namelen = saved_namelen; /* compiler workaround */
2112 int result;
2113 static unsigned char padding[8] = { 0x00 };
2114
2115 if (ce->ce_flags & CE_STRIP_NAME) {
2116 saved_namelen = ce_namelen(ce);
2117 ce->ce_namelen = 0;
2118 }
2119
2120 if (ce->ce_flags & CE_EXTENDED)
2121 size = offsetof(struct ondisk_cache_entry_extended, name);
2122 else
2123 size = offsetof(struct ondisk_cache_entry, name);
2124
2125 if (!previous_name) {
2126 int len = ce_namelen(ce);
2127 copy_cache_entry_to_ondisk(ondisk, ce);
2128 result = ce_write(c, fd, ondisk, size);
2129 if (!result)
2130 result = ce_write(c, fd, ce->name, len);
2131 if (!result)
2132 result = ce_write(c, fd, padding, align_padding_size(size, len));
2133 } else {
2134 int common, to_remove, prefix_size;
2135 unsigned char to_remove_vi[16];
2136 for (common = 0;
2137 (ce->name[common] &&
2138 common < previous_name->len &&
2139 ce->name[common] == previous_name->buf[common]);
2140 common++)
2141 ; /* still matching */
2142 to_remove = previous_name->len - common;
2143 prefix_size = encode_varint(to_remove, to_remove_vi);
2144
2145 copy_cache_entry_to_ondisk(ondisk, ce);
2146 result = ce_write(c, fd, ondisk, size);
2147 if (!result)
2148 result = ce_write(c, fd, to_remove_vi, prefix_size);
2149 if (!result)
2150 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2151 if (!result)
2152 result = ce_write(c, fd, padding, 1);
2153
2154 strbuf_splice(previous_name, common, to_remove,
2155 ce->name + common, ce_namelen(ce) - common);
2156 }
2157 if (ce->ce_flags & CE_STRIP_NAME) {
2158 ce->ce_namelen = saved_namelen;
2159 ce->ce_flags &= ~CE_STRIP_NAME;
2160 }
2161
2162 return result;
2163 }
2164
2165 /*
2166 * This function verifies if index_state has the correct sha1 of the
2167 * index file. Don't die if we have any other failure, just return 0.
2168 */
2169 static int verify_index_from(const struct index_state *istate, const char *path)
2170 {
2171 int fd;
2172 ssize_t n;
2173 struct stat st;
2174 unsigned char hash[GIT_MAX_RAWSZ];
2175
2176 if (!istate->initialized)
2177 return 0;
2178
2179 fd = open(path, O_RDONLY);
2180 if (fd < 0)
2181 return 0;
2182
2183 if (fstat(fd, &st))
2184 goto out;
2185
2186 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2187 goto out;
2188
2189 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2190 if (n != the_hash_algo->rawsz)
2191 goto out;
2192
2193 if (hashcmp(istate->sha1, hash))
2194 goto out;
2195
2196 close(fd);
2197 return 1;
2198
2199 out:
2200 close(fd);
2201 return 0;
2202 }
2203
2204 static int verify_index(const struct index_state *istate)
2205 {
2206 return verify_index_from(istate, get_index_file());
2207 }
2208
2209 static int has_racy_timestamp(struct index_state *istate)
2210 {
2211 int entries = istate->cache_nr;
2212 int i;
2213
2214 for (i = 0; i < entries; i++) {
2215 struct cache_entry *ce = istate->cache[i];
2216 if (is_racy_timestamp(istate, ce))
2217 return 1;
2218 }
2219 return 0;
2220 }
2221
2222 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2223 {
2224 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2225 verify_index(istate))
2226 write_locked_index(istate, lockfile, COMMIT_LOCK);
2227 else
2228 rollback_lock_file(lockfile);
2229 }
2230
2231 /*
2232 * On success, `tempfile` is closed. If it is the temporary file
2233 * of a `struct lock_file`, we will therefore effectively perform
2234 * a 'close_lock_file_gently()`. Since that is an implementation
2235 * detail of lockfiles, callers of `do_write_index()` should not
2236 * rely on it.
2237 */
2238 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2239 int strip_extensions)
2240 {
2241 uint64_t start = getnanotime();
2242 int newfd = tempfile->fd;
2243 git_hash_ctx c;
2244 struct cache_header hdr;
2245 int i, err = 0, removed, extended, hdr_version;
2246 struct cache_entry **cache = istate->cache;
2247 int entries = istate->cache_nr;
2248 struct stat st;
2249 struct ondisk_cache_entry_extended ondisk;
2250 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2251 int drop_cache_tree = istate->drop_cache_tree;
2252
2253 for (i = removed = extended = 0; i < entries; i++) {
2254 if (cache[i]->ce_flags & CE_REMOVE)
2255 removed++;
2256
2257 /* reduce extended entries if possible */
2258 cache[i]->ce_flags &= ~CE_EXTENDED;
2259 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2260 extended++;
2261 cache[i]->ce_flags |= CE_EXTENDED;
2262 }
2263 }
2264
2265 if (!istate->version) {
2266 istate->version = get_index_format_default();
2267 if (getenv("GIT_TEST_SPLIT_INDEX"))
2268 init_split_index(istate);
2269 }
2270
2271 /* demote version 3 to version 2 when the latter suffices */
2272 if (istate->version == 3 || istate->version == 2)
2273 istate->version = extended ? 3 : 2;
2274
2275 hdr_version = istate->version;
2276
2277 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2278 hdr.hdr_version = htonl(hdr_version);
2279 hdr.hdr_entries = htonl(entries - removed);
2280
2281 the_hash_algo->init_fn(&c);
2282 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2283 return -1;
2284
2285 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2286
2287 for (i = 0; i < entries; i++) {
2288 struct cache_entry *ce = cache[i];
2289 if (ce->ce_flags & CE_REMOVE)
2290 continue;
2291 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2292 ce_smudge_racily_clean_entry(ce);
2293 if (is_null_oid(&ce->oid)) {
2294 static const char msg[] = "cache entry has null sha1: %s";
2295 static int allow = -1;
2296
2297 if (allow < 0)
2298 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2299 if (allow)
2300 warning(msg, ce->name);
2301 else
2302 err = error(msg, ce->name);
2303
2304 drop_cache_tree = 1;
2305 }
2306 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2307 err = -1;
2308
2309 if (err)
2310 break;
2311 }
2312 strbuf_release(&previous_name_buf);
2313
2314 if (err)
2315 return err;
2316
2317 /* Write extension data here */
2318 if (!strip_extensions && istate->split_index) {
2319 struct strbuf sb = STRBUF_INIT;
2320
2321 err = write_link_extension(&sb, istate) < 0 ||
2322 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2323 sb.len) < 0 ||
2324 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2325 strbuf_release(&sb);
2326 if (err)
2327 return -1;
2328 }
2329 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2330 struct strbuf sb = STRBUF_INIT;
2331
2332 cache_tree_write(&sb, istate->cache_tree);
2333 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2334 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2335 strbuf_release(&sb);
2336 if (err)
2337 return -1;
2338 }
2339 if (!strip_extensions && istate->resolve_undo) {
2340 struct strbuf sb = STRBUF_INIT;
2341
2342 resolve_undo_write(&sb, istate->resolve_undo);
2343 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2344 sb.len) < 0
2345 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2346 strbuf_release(&sb);
2347 if (err)
2348 return -1;
2349 }
2350 if (!strip_extensions && istate->untracked) {
2351 struct strbuf sb = STRBUF_INIT;
2352
2353 write_untracked_extension(&sb, istate->untracked);
2354 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2355 sb.len) < 0 ||
2356 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2357 strbuf_release(&sb);
2358 if (err)
2359 return -1;
2360 }
2361 if (!strip_extensions && istate->fsmonitor_last_update) {
2362 struct strbuf sb = STRBUF_INIT;
2363
2364 write_fsmonitor_extension(&sb, istate);
2365 err = write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2366 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2367 strbuf_release(&sb);
2368 if (err)
2369 return -1;
2370 }
2371
2372 if (ce_flush(&c, newfd, istate->sha1))
2373 return -1;
2374 if (close_tempfile_gently(tempfile)) {
2375 error(_("could not close '%s'"), tempfile->filename.buf);
2376 return -1;
2377 }
2378 if (stat(tempfile->filename.buf, &st))
2379 return -1;
2380 istate->timestamp.sec = (unsigned int)st.st_mtime;
2381 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2382 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
2383 return 0;
2384 }
2385
2386 void set_alternate_index_output(const char *name)
2387 {
2388 alternate_index_output = name;
2389 }
2390
2391 static int commit_locked_index(struct lock_file *lk)
2392 {
2393 if (alternate_index_output)
2394 return commit_lock_file_to(lk, alternate_index_output);
2395 else
2396 return commit_lock_file(lk);
2397 }
2398
2399 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2400 unsigned flags)
2401 {
2402 int ret = do_write_index(istate, lock->tempfile, 0);
2403 if (ret)
2404 return ret;
2405 if (flags & COMMIT_LOCK)
2406 return commit_locked_index(lock);
2407 return close_lock_file_gently(lock);
2408 }
2409
2410 static int write_split_index(struct index_state *istate,
2411 struct lock_file *lock,
2412 unsigned flags)
2413 {
2414 int ret;
2415 prepare_to_write_split_index(istate);
2416 ret = do_write_locked_index(istate, lock, flags);
2417 finish_writing_split_index(istate);
2418 return ret;
2419 }
2420
2421 static const char *shared_index_expire = "2.weeks.ago";
2422
2423 static unsigned long get_shared_index_expire_date(void)
2424 {
2425 static unsigned long shared_index_expire_date;
2426 static int shared_index_expire_date_prepared;
2427
2428 if (!shared_index_expire_date_prepared) {
2429 git_config_get_expiry("splitindex.sharedindexexpire",
2430 &shared_index_expire);
2431 shared_index_expire_date = approxidate(shared_index_expire);
2432 shared_index_expire_date_prepared = 1;
2433 }
2434
2435 return shared_index_expire_date;
2436 }
2437
2438 static int should_delete_shared_index(const char *shared_index_path)
2439 {
2440 struct stat st;
2441 unsigned long expiration;
2442
2443 /* Check timestamp */
2444 expiration = get_shared_index_expire_date();
2445 if (!expiration)
2446 return 0;
2447 if (stat(shared_index_path, &st))
2448 return error_errno(_("could not stat '%s'"), shared_index_path);
2449 if (st.st_mtime > expiration)
2450 return 0;
2451
2452 return 1;
2453 }
2454
2455 static int clean_shared_index_files(const char *current_hex)
2456 {
2457 struct dirent *de;
2458 DIR *dir = opendir(get_git_dir());
2459
2460 if (!dir)
2461 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2462
2463 while ((de = readdir(dir)) != NULL) {
2464 const char *sha1_hex;
2465 const char *shared_index_path;
2466 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2467 continue;
2468 if (!strcmp(sha1_hex, current_hex))
2469 continue;
2470 shared_index_path = git_path("%s", de->d_name);
2471 if (should_delete_shared_index(shared_index_path) > 0 &&
2472 unlink(shared_index_path))
2473 warning_errno(_("unable to unlink: %s"), shared_index_path);
2474 }
2475 closedir(dir);
2476
2477 return 0;
2478 }
2479
2480 static int write_shared_index(struct index_state *istate,
2481 struct tempfile **temp)
2482 {
2483 struct split_index *si = istate->split_index;
2484 int ret;
2485
2486 move_cache_to_base_index(istate);
2487 ret = do_write_index(si->base, *temp, 1);
2488 if (ret)
2489 return ret;
2490 ret = adjust_shared_perm(get_tempfile_path(*temp));
2491 if (ret) {
2492 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
2493 return ret;
2494 }
2495 ret = rename_tempfile(temp,
2496 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2497 if (!ret) {
2498 hashcpy(si->base_sha1, si->base->sha1);
2499 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2500 }
2501
2502 return ret;
2503 }
2504
2505 static const int default_max_percent_split_change = 20;
2506
2507 static int too_many_not_shared_entries(struct index_state *istate)
2508 {
2509 int i, not_shared = 0;
2510 int max_split = git_config_get_max_percent_split_change();
2511
2512 switch (max_split) {
2513 case -1:
2514 /* not or badly configured: use the default value */
2515 max_split = default_max_percent_split_change;
2516 break;
2517 case 0:
2518 return 1; /* 0% means always write a new shared index */
2519 case 100:
2520 return 0; /* 100% means never write a new shared index */
2521 default:
2522 break; /* just use the configured value */
2523 }
2524
2525 /* Count not shared entries */
2526 for (i = 0; i < istate->cache_nr; i++) {
2527 struct cache_entry *ce = istate->cache[i];
2528 if (!ce->index)
2529 not_shared++;
2530 }
2531
2532 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2533 }
2534
2535 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2536 unsigned flags)
2537 {
2538 int new_shared_index, ret;
2539 struct split_index *si = istate->split_index;
2540
2541 if (istate->fsmonitor_last_update)
2542 fill_fsmonitor_bitmap(istate);
2543
2544 if (!si || alternate_index_output ||
2545 (istate->cache_changed & ~EXTMASK)) {
2546 if (si)
2547 hashclr(si->base_sha1);
2548 ret = do_write_locked_index(istate, lock, flags);
2549 goto out;
2550 }
2551
2552 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2553 int v = si->base_sha1[0];
2554 if ((v & 15) < 6)
2555 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2556 }
2557 if (too_many_not_shared_entries(istate))
2558 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2559
2560 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2561
2562 if (new_shared_index) {
2563 struct tempfile *temp;
2564 int saved_errno;
2565
2566 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2567 if (!temp) {
2568 hashclr(si->base_sha1);
2569 ret = do_write_locked_index(istate, lock, flags);
2570 goto out;
2571 }
2572 ret = write_shared_index(istate, &temp);
2573
2574 saved_errno = errno;
2575 if (is_tempfile_active(temp))
2576 delete_tempfile(&temp);
2577 errno = saved_errno;
2578
2579 if (ret)
2580 goto out;
2581 }
2582
2583 ret = write_split_index(istate, lock, flags);
2584
2585 /* Freshen the shared index only if the split-index was written */
2586 if (!ret && !new_shared_index) {
2587 const char *shared_index = git_path("sharedindex.%s",
2588 sha1_to_hex(si->base_sha1));
2589 freshen_shared_index(shared_index, 1);
2590 }
2591
2592 out:
2593 if (flags & COMMIT_LOCK)
2594 rollback_lock_file(lock);
2595 return ret;
2596 }
2597
2598 /*
2599 * Read the index file that is potentially unmerged into given
2600 * index_state, dropping any unmerged entries. Returns true if
2601 * the index is unmerged. Callers who want to refuse to work
2602 * from an unmerged state can call this and check its return value,
2603 * instead of calling read_cache().
2604 */
2605 int read_index_unmerged(struct index_state *istate)
2606 {
2607 int i;
2608 int unmerged = 0;
2609
2610 read_index(istate);
2611 for (i = 0; i < istate->cache_nr; i++) {
2612 struct cache_entry *ce = istate->cache[i];
2613 struct cache_entry *new_ce;
2614 int size, len;
2615
2616 if (!ce_stage(ce))
2617 continue;
2618 unmerged = 1;
2619 len = ce_namelen(ce);
2620 size = cache_entry_size(len);
2621 new_ce = xcalloc(1, size);
2622 memcpy(new_ce->name, ce->name, len);
2623 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2624 new_ce->ce_namelen = len;
2625 new_ce->ce_mode = ce->ce_mode;
2626 if (add_index_entry(istate, new_ce, 0))
2627 return error("%s: cannot drop to stage #0",
2628 new_ce->name);
2629 }
2630 return unmerged;
2631 }
2632
2633 /*
2634 * Returns 1 if the path is an "other" path with respect to
2635 * the index; that is, the path is not mentioned in the index at all,
2636 * either as a file, a directory with some files in the index,
2637 * or as an unmerged entry.
2638 *
2639 * We helpfully remove a trailing "/" from directories so that
2640 * the output of read_directory can be used as-is.
2641 */
2642 int index_name_is_other(const struct index_state *istate, const char *name,
2643 int namelen)
2644 {
2645 int pos;
2646 if (namelen && name[namelen - 1] == '/')
2647 namelen--;
2648 pos = index_name_pos(istate, name, namelen);
2649 if (0 <= pos)
2650 return 0; /* exact match */
2651 pos = -pos - 1;
2652 if (pos < istate->cache_nr) {
2653 struct cache_entry *ce = istate->cache[pos];
2654 if (ce_namelen(ce) == namelen &&
2655 !memcmp(ce->name, name, namelen))
2656 return 0; /* Yup, this one exists unmerged */
2657 }
2658 return 1;
2659 }
2660
2661 void *read_blob_data_from_index(const struct index_state *istate,
2662 const char *path, unsigned long *size)
2663 {
2664 int pos, len;
2665 unsigned long sz;
2666 enum object_type type;
2667 void *data;
2668
2669 len = strlen(path);
2670 pos = index_name_pos(istate, path, len);
2671 if (pos < 0) {
2672 /*
2673 * We might be in the middle of a merge, in which
2674 * case we would read stage #2 (ours).
2675 */
2676 int i;
2677 for (i = -pos - 1;
2678 (pos < 0 && i < istate->cache_nr &&
2679 !strcmp(istate->cache[i]->name, path));
2680 i++)
2681 if (ce_stage(istate->cache[i]) == 2)
2682 pos = i;
2683 }
2684 if (pos < 0)
2685 return NULL;
2686 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2687 if (!data || type != OBJ_BLOB) {
2688 free(data);
2689 return NULL;
2690 }
2691 if (size)
2692 *size = sz;
2693 return data;
2694 }
2695
2696 void stat_validity_clear(struct stat_validity *sv)
2697 {
2698 FREE_AND_NULL(sv->sd);
2699 }
2700
2701 int stat_validity_check(struct stat_validity *sv, const char *path)
2702 {
2703 struct stat st;
2704
2705 if (stat(path, &st) < 0)
2706 return sv->sd == NULL;
2707 if (!sv->sd)
2708 return 0;
2709 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2710 }
2711
2712 void stat_validity_update(struct stat_validity *sv, int fd)
2713 {
2714 struct stat st;
2715
2716 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2717 stat_validity_clear(sv);
2718 else {
2719 if (!sv->sd)
2720 sv->sd = xcalloc(1, sizeof(struct stat_data));
2721 fill_stat_data(sv->sd, &st);
2722 }
2723 }
2724
2725 void move_index_extensions(struct index_state *dst, struct index_state *src)
2726 {
2727 dst->untracked = src->untracked;
2728 src->untracked = NULL;
2729 }