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