Sync with Git 2.15.2
[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 unsigned char sha1[20];
635 if (write_sha1_file("", 0, blob_type, sha1))
636 die("cannot create an empty blob in the object database");
637 hashcpy(ce->oid.hash, sha1);
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, mode)) {
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, unsigned mode)
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. Note that we match
834 * case-insensitively here, even if ignore_case is not set.
835 * This outlaws ".GIT" everywhere out of an abundance of caution,
836 * since there's really no good reason to allow it.
837 *
838 * Once we've seen ".git", we can also find ".gitmodules", etc (also
839 * case-insensitively).
840 */
841 case 'g':
842 case 'G':
843 if (rest[1] != 'i' && rest[1] != 'I')
844 break;
845 if (rest[2] != 't' && rest[2] != 'T')
846 break;
847 if (rest[3] == '\0' || is_dir_sep(rest[3]))
848 return 0;
849 if (S_ISLNK(mode)) {
850 rest += 3;
851 if (skip_iprefix(rest, "modules", &rest) &&
852 (*rest == '\0' || is_dir_sep(*rest)))
853 return 0;
854 }
855 break;
856 case '.':
857 if (rest[1] == '\0' || is_dir_sep(rest[1]))
858 return 0;
859 }
860 return 1;
861 }
862
863 int verify_path(const char *path, unsigned mode)
864 {
865 char c;
866
867 if (has_dos_drive_prefix(path))
868 return 0;
869
870 goto inside;
871 for (;;) {
872 if (!c)
873 return 1;
874 if (is_dir_sep(c)) {
875 inside:
876 if (protect_hfs) {
877 if (is_hfs_dotgit(path))
878 return 0;
879 if (S_ISLNK(mode)) {
880 if (is_hfs_dotgitmodules(path))
881 return 0;
882 }
883 }
884 if (protect_ntfs) {
885 if (is_ntfs_dotgit(path))
886 return 0;
887 if (S_ISLNK(mode)) {
888 if (is_ntfs_dotgitmodules(path))
889 return 0;
890 }
891 }
892
893 c = *path++;
894 if ((c == '.' && !verify_dotfile(path, mode)) ||
895 is_dir_sep(c) || c == '\0')
896 return 0;
897 }
898 c = *path++;
899 }
900 }
901
902 /*
903 * Do we have another file that has the beginning components being a
904 * proper superset of the name we're trying to add?
905 */
906 static int has_file_name(struct index_state *istate,
907 const struct cache_entry *ce, int pos, int ok_to_replace)
908 {
909 int retval = 0;
910 int len = ce_namelen(ce);
911 int stage = ce_stage(ce);
912 const char *name = ce->name;
913
914 while (pos < istate->cache_nr) {
915 struct cache_entry *p = istate->cache[pos++];
916
917 if (len >= ce_namelen(p))
918 break;
919 if (memcmp(name, p->name, len))
920 break;
921 if (ce_stage(p) != stage)
922 continue;
923 if (p->name[len] != '/')
924 continue;
925 if (p->ce_flags & CE_REMOVE)
926 continue;
927 retval = -1;
928 if (!ok_to_replace)
929 break;
930 remove_index_entry_at(istate, --pos);
931 }
932 return retval;
933 }
934
935
936 /*
937 * Like strcmp(), but also return the offset of the first change.
938 * If strings are equal, return the length.
939 */
940 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
941 {
942 size_t k;
943
944 if (!first_change)
945 return strcmp(s1, s2);
946
947 for (k = 0; s1[k] == s2[k]; k++)
948 if (s1[k] == '\0')
949 break;
950
951 *first_change = k;
952 return (unsigned char)s1[k] - (unsigned char)s2[k];
953 }
954
955 /*
956 * Do we have another file with a pathname that is a proper
957 * subset of the name we're trying to add?
958 *
959 * That is, is there another file in the index with a path
960 * that matches a sub-directory in the given entry?
961 */
962 static int has_dir_name(struct index_state *istate,
963 const struct cache_entry *ce, int pos, int ok_to_replace)
964 {
965 int retval = 0;
966 int stage = ce_stage(ce);
967 const char *name = ce->name;
968 const char *slash = name + ce_namelen(ce);
969 size_t len_eq_last;
970 int cmp_last = 0;
971
972 /*
973 * We are frequently called during an iteration on a sorted
974 * list of pathnames and while building a new index. Therefore,
975 * there is a high probability that this entry will eventually
976 * be appended to the index, rather than inserted in the middle.
977 * If we can confirm that, we can avoid binary searches on the
978 * components of the pathname.
979 *
980 * Compare the entry's full path with the last path in the index.
981 */
982 if (istate->cache_nr > 0) {
983 cmp_last = strcmp_offset(name,
984 istate->cache[istate->cache_nr - 1]->name,
985 &len_eq_last);
986 if (cmp_last > 0) {
987 if (len_eq_last == 0) {
988 /*
989 * The entry sorts AFTER the last one in the
990 * index and their paths have no common prefix,
991 * so there cannot be a F/D conflict.
992 */
993 return retval;
994 } else {
995 /*
996 * The entry sorts AFTER the last one in the
997 * index, but has a common prefix. Fall through
998 * to the loop below to disect the entry's path
999 * and see where the difference is.
1000 */
1001 }
1002 } else if (cmp_last == 0) {
1003 /*
1004 * The entry exactly matches the last one in the
1005 * index, but because of multiple stage and CE_REMOVE
1006 * items, we fall through and let the regular search
1007 * code handle it.
1008 */
1009 }
1010 }
1011
1012 for (;;) {
1013 size_t len;
1014
1015 for (;;) {
1016 if (*--slash == '/')
1017 break;
1018 if (slash <= ce->name)
1019 return retval;
1020 }
1021 len = slash - name;
1022
1023 if (cmp_last > 0) {
1024 /*
1025 * (len + 1) is a directory boundary (including
1026 * the trailing slash). And since the loop is
1027 * decrementing "slash", the first iteration is
1028 * the longest directory prefix; subsequent
1029 * iterations consider parent directories.
1030 */
1031
1032 if (len + 1 <= len_eq_last) {
1033 /*
1034 * The directory prefix (including the trailing
1035 * slash) also appears as a prefix in the last
1036 * entry, so the remainder cannot collide (because
1037 * strcmp said the whole path was greater).
1038 *
1039 * EQ: last: xxx/A
1040 * this: xxx/B
1041 *
1042 * LT: last: xxx/file_A
1043 * this: xxx/file_B
1044 */
1045 return retval;
1046 }
1047
1048 if (len > len_eq_last) {
1049 /*
1050 * This part of the directory prefix (excluding
1051 * the trailing slash) is longer than the known
1052 * equal portions, so this sub-directory cannot
1053 * collide with a file.
1054 *
1055 * GT: last: xxxA
1056 * this: xxxB/file
1057 */
1058 return retval;
1059 }
1060
1061 if (istate->cache_nr > 0 &&
1062 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1063 /*
1064 * The directory prefix lines up with part of
1065 * a longer file or directory name, but sorts
1066 * after it, so this sub-directory cannot
1067 * collide with a file.
1068 *
1069 * last: xxx/yy-file (because '-' sorts before '/')
1070 * this: xxx/yy/abc
1071 */
1072 return retval;
1073 }
1074
1075 /*
1076 * This is a possible collision. Fall through and
1077 * let the regular search code handle it.
1078 *
1079 * last: xxx
1080 * this: xxx/file
1081 */
1082 }
1083
1084 pos = index_name_stage_pos(istate, name, len, stage);
1085 if (pos >= 0) {
1086 /*
1087 * Found one, but not so fast. This could
1088 * be a marker that says "I was here, but
1089 * I am being removed". Such an entry is
1090 * not a part of the resulting tree, and
1091 * it is Ok to have a directory at the same
1092 * path.
1093 */
1094 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1095 retval = -1;
1096 if (!ok_to_replace)
1097 break;
1098 remove_index_entry_at(istate, pos);
1099 continue;
1100 }
1101 }
1102 else
1103 pos = -pos-1;
1104
1105 /*
1106 * Trivial optimization: if we find an entry that
1107 * already matches the sub-directory, then we know
1108 * we're ok, and we can exit.
1109 */
1110 while (pos < istate->cache_nr) {
1111 struct cache_entry *p = istate->cache[pos];
1112 if ((ce_namelen(p) <= len) ||
1113 (p->name[len] != '/') ||
1114 memcmp(p->name, name, len))
1115 break; /* not our subdirectory */
1116 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1117 /*
1118 * p is at the same stage as our entry, and
1119 * is a subdirectory of what we are looking
1120 * at, so we cannot have conflicts at our
1121 * level or anything shorter.
1122 */
1123 return retval;
1124 pos++;
1125 }
1126 }
1127 return retval;
1128 }
1129
1130 /* We may be in a situation where we already have path/file and path
1131 * is being added, or we already have path and path/file is being
1132 * added. Either one would result in a nonsense tree that has path
1133 * twice when git-write-tree tries to write it out. Prevent it.
1134 *
1135 * If ok-to-replace is specified, we remove the conflicting entries
1136 * from the cache so the caller should recompute the insert position.
1137 * When this happens, we return non-zero.
1138 */
1139 static int check_file_directory_conflict(struct index_state *istate,
1140 const struct cache_entry *ce,
1141 int pos, int ok_to_replace)
1142 {
1143 int retval;
1144
1145 /*
1146 * When ce is an "I am going away" entry, we allow it to be added
1147 */
1148 if (ce->ce_flags & CE_REMOVE)
1149 return 0;
1150
1151 /*
1152 * We check if the path is a sub-path of a subsequent pathname
1153 * first, since removing those will not change the position
1154 * in the array.
1155 */
1156 retval = has_file_name(istate, ce, pos, ok_to_replace);
1157
1158 /*
1159 * Then check if the path might have a clashing sub-directory
1160 * before it.
1161 */
1162 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1163 }
1164
1165 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1166 {
1167 int pos;
1168 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1169 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1170 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1171 int new_only = option & ADD_CACHE_NEW_ONLY;
1172
1173 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1174 cache_tree_invalidate_path(istate, ce->name);
1175
1176 /*
1177 * If this entry's path sorts after the last entry in the index,
1178 * we can avoid searching for it.
1179 */
1180 if (istate->cache_nr > 0 &&
1181 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1182 pos = -istate->cache_nr - 1;
1183 else
1184 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1185
1186 /* existing match? Just replace it. */
1187 if (pos >= 0) {
1188 if (!new_only)
1189 replace_index_entry(istate, pos, ce);
1190 return 0;
1191 }
1192 pos = -pos-1;
1193
1194 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1195 untracked_cache_add_to_index(istate, ce->name);
1196
1197 /*
1198 * Inserting a merged entry ("stage 0") into the index
1199 * will always replace all non-merged entries..
1200 */
1201 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1202 while (ce_same_name(istate->cache[pos], ce)) {
1203 ok_to_add = 1;
1204 if (!remove_index_entry_at(istate, pos))
1205 break;
1206 }
1207 }
1208
1209 if (!ok_to_add)
1210 return -1;
1211 if (!verify_path(ce->name, ce->ce_mode))
1212 return error("Invalid path '%s'", ce->name);
1213
1214 if (!skip_df_check &&
1215 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1216 if (!ok_to_replace)
1217 return error("'%s' appears as both a file and as a directory",
1218 ce->name);
1219 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1220 pos = -pos-1;
1221 }
1222 return pos + 1;
1223 }
1224
1225 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1226 {
1227 int pos;
1228
1229 if (option & ADD_CACHE_JUST_APPEND)
1230 pos = istate->cache_nr;
1231 else {
1232 int ret;
1233 ret = add_index_entry_with_check(istate, ce, option);
1234 if (ret <= 0)
1235 return ret;
1236 pos = ret - 1;
1237 }
1238
1239 /* Make sure the array is big enough .. */
1240 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1241
1242 /* Add it in.. */
1243 istate->cache_nr++;
1244 if (istate->cache_nr > pos + 1)
1245 memmove(istate->cache + pos + 1,
1246 istate->cache + pos,
1247 (istate->cache_nr - pos - 1) * sizeof(ce));
1248 set_index_entry(istate, pos, ce);
1249 istate->cache_changed |= CE_ENTRY_ADDED;
1250 return 0;
1251 }
1252
1253 /*
1254 * "refresh" does not calculate a new sha1 file or bring the
1255 * cache up-to-date for mode/content changes. But what it
1256 * _does_ do is to "re-match" the stat information of a file
1257 * with the cache, so that you can refresh the cache for a
1258 * file that hasn't been changed but where the stat entry is
1259 * out of date.
1260 *
1261 * For example, you'd want to do this after doing a "git-read-tree",
1262 * to link up the stat cache details with the proper files.
1263 */
1264 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1265 struct cache_entry *ce,
1266 unsigned int options, int *err,
1267 int *changed_ret)
1268 {
1269 struct stat st;
1270 struct cache_entry *updated;
1271 int changed, size;
1272 int refresh = options & CE_MATCH_REFRESH;
1273 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1274 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1275 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1276 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1277
1278 if (!refresh || ce_uptodate(ce))
1279 return ce;
1280
1281 if (!ignore_fsmonitor)
1282 refresh_fsmonitor(istate);
1283 /*
1284 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1285 * that the change to the work tree does not matter and told
1286 * us not to worry.
1287 */
1288 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1289 ce_mark_uptodate(ce);
1290 return ce;
1291 }
1292 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1293 ce_mark_uptodate(ce);
1294 return ce;
1295 }
1296 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1297 ce_mark_uptodate(ce);
1298 return ce;
1299 }
1300
1301 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1302 if (ignore_missing)
1303 return ce;
1304 if (err)
1305 *err = ENOENT;
1306 return NULL;
1307 }
1308
1309 if (lstat(ce->name, &st) < 0) {
1310 if (ignore_missing && errno == ENOENT)
1311 return ce;
1312 if (err)
1313 *err = errno;
1314 return NULL;
1315 }
1316
1317 changed = ie_match_stat(istate, ce, &st, options);
1318 if (changed_ret)
1319 *changed_ret = changed;
1320 if (!changed) {
1321 /*
1322 * The path is unchanged. If we were told to ignore
1323 * valid bit, then we did the actual stat check and
1324 * found that the entry is unmodified. If the entry
1325 * is not marked VALID, this is the place to mark it
1326 * valid again, under "assume unchanged" mode.
1327 */
1328 if (ignore_valid && assume_unchanged &&
1329 !(ce->ce_flags & CE_VALID))
1330 ; /* mark this one VALID again */
1331 else {
1332 /*
1333 * We do not mark the index itself "modified"
1334 * because CE_UPTODATE flag is in-core only;
1335 * we are not going to write this change out.
1336 */
1337 if (!S_ISGITLINK(ce->ce_mode)) {
1338 ce_mark_uptodate(ce);
1339 mark_fsmonitor_valid(ce);
1340 }
1341 return ce;
1342 }
1343 }
1344
1345 if (ie_modified(istate, ce, &st, options)) {
1346 if (err)
1347 *err = EINVAL;
1348 return NULL;
1349 }
1350
1351 size = ce_size(ce);
1352 updated = xmalloc(size);
1353 memcpy(updated, ce, size);
1354 fill_stat_cache_info(updated, &st);
1355 /*
1356 * If ignore_valid is not set, we should leave CE_VALID bit
1357 * alone. Otherwise, paths marked with --no-assume-unchanged
1358 * (i.e. things to be edited) will reacquire CE_VALID bit
1359 * automatically, which is not really what we want.
1360 */
1361 if (!ignore_valid && assume_unchanged &&
1362 !(ce->ce_flags & CE_VALID))
1363 updated->ce_flags &= ~CE_VALID;
1364
1365 /* istate->cache_changed is updated in the caller */
1366 return updated;
1367 }
1368
1369 static void show_file(const char * fmt, const char * name, int in_porcelain,
1370 int * first, const char *header_msg)
1371 {
1372 if (in_porcelain && *first && header_msg) {
1373 printf("%s\n", header_msg);
1374 *first = 0;
1375 }
1376 printf(fmt, name);
1377 }
1378
1379 int refresh_index(struct index_state *istate, unsigned int flags,
1380 const struct pathspec *pathspec,
1381 char *seen, const char *header_msg)
1382 {
1383 int i;
1384 int has_errors = 0;
1385 int really = (flags & REFRESH_REALLY) != 0;
1386 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1387 int quiet = (flags & REFRESH_QUIET) != 0;
1388 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1389 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1390 int first = 1;
1391 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1392 unsigned int options = (CE_MATCH_REFRESH |
1393 (really ? CE_MATCH_IGNORE_VALID : 0) |
1394 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1395 const char *modified_fmt;
1396 const char *deleted_fmt;
1397 const char *typechange_fmt;
1398 const char *added_fmt;
1399 const char *unmerged_fmt;
1400
1401 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1402 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1403 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1404 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1405 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1406 for (i = 0; i < istate->cache_nr; i++) {
1407 struct cache_entry *ce, *new;
1408 int cache_errno = 0;
1409 int changed = 0;
1410 int filtered = 0;
1411
1412 ce = istate->cache[i];
1413 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1414 continue;
1415
1416 if (pathspec && !ce_path_match(ce, pathspec, seen))
1417 filtered = 1;
1418
1419 if (ce_stage(ce)) {
1420 while ((i < istate->cache_nr) &&
1421 ! strcmp(istate->cache[i]->name, ce->name))
1422 i++;
1423 i--;
1424 if (allow_unmerged)
1425 continue;
1426 if (!filtered)
1427 show_file(unmerged_fmt, ce->name, in_porcelain,
1428 &first, header_msg);
1429 has_errors = 1;
1430 continue;
1431 }
1432
1433 if (filtered)
1434 continue;
1435
1436 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1437 if (new == ce)
1438 continue;
1439 if (!new) {
1440 const char *fmt;
1441
1442 if (really && cache_errno == EINVAL) {
1443 /* If we are doing --really-refresh that
1444 * means the index is not valid anymore.
1445 */
1446 ce->ce_flags &= ~CE_VALID;
1447 ce->ce_flags |= CE_UPDATE_IN_BASE;
1448 mark_fsmonitor_invalid(istate, ce);
1449 istate->cache_changed |= CE_ENTRY_CHANGED;
1450 }
1451 if (quiet)
1452 continue;
1453
1454 if (cache_errno == ENOENT)
1455 fmt = deleted_fmt;
1456 else if (ce_intent_to_add(ce))
1457 fmt = added_fmt; /* must be before other checks */
1458 else if (changed & TYPE_CHANGED)
1459 fmt = typechange_fmt;
1460 else
1461 fmt = modified_fmt;
1462 show_file(fmt,
1463 ce->name, in_porcelain, &first, header_msg);
1464 has_errors = 1;
1465 continue;
1466 }
1467
1468 replace_index_entry(istate, i, new);
1469 }
1470 return has_errors;
1471 }
1472
1473 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1474 unsigned int options)
1475 {
1476 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1477 }
1478
1479
1480 /*****************************************************************
1481 * Index File I/O
1482 *****************************************************************/
1483
1484 #define INDEX_FORMAT_DEFAULT 3
1485
1486 static unsigned int get_index_format_default(void)
1487 {
1488 char *envversion = getenv("GIT_INDEX_VERSION");
1489 char *endp;
1490 int value;
1491 unsigned int version = INDEX_FORMAT_DEFAULT;
1492
1493 if (!envversion) {
1494 if (!git_config_get_int("index.version", &value))
1495 version = value;
1496 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1497 warning(_("index.version set, but the value is invalid.\n"
1498 "Using version %i"), INDEX_FORMAT_DEFAULT);
1499 return INDEX_FORMAT_DEFAULT;
1500 }
1501 return version;
1502 }
1503
1504 version = strtoul(envversion, &endp, 10);
1505 if (*endp ||
1506 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1507 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1508 "Using version %i"), INDEX_FORMAT_DEFAULT);
1509 version = INDEX_FORMAT_DEFAULT;
1510 }
1511 return version;
1512 }
1513
1514 /*
1515 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1516 * Again - this is just a (very strong in practice) heuristic that
1517 * the inode hasn't changed.
1518 *
1519 * We save the fields in big-endian order to allow using the
1520 * index file over NFS transparently.
1521 */
1522 struct ondisk_cache_entry {
1523 struct cache_time ctime;
1524 struct cache_time mtime;
1525 uint32_t dev;
1526 uint32_t ino;
1527 uint32_t mode;
1528 uint32_t uid;
1529 uint32_t gid;
1530 uint32_t size;
1531 unsigned char sha1[20];
1532 uint16_t flags;
1533 char name[FLEX_ARRAY]; /* more */
1534 };
1535
1536 /*
1537 * This struct is used when CE_EXTENDED bit is 1
1538 * The struct must match ondisk_cache_entry exactly from
1539 * ctime till flags
1540 */
1541 struct ondisk_cache_entry_extended {
1542 struct cache_time ctime;
1543 struct cache_time mtime;
1544 uint32_t dev;
1545 uint32_t ino;
1546 uint32_t mode;
1547 uint32_t uid;
1548 uint32_t gid;
1549 uint32_t size;
1550 unsigned char sha1[20];
1551 uint16_t flags;
1552 uint16_t flags2;
1553 char name[FLEX_ARRAY]; /* more */
1554 };
1555
1556 /* These are only used for v3 or lower */
1557 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1558 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1559 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1560 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1561 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1562 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1563 ondisk_cache_entry_size(ce_namelen(ce)))
1564
1565 /* Allow fsck to force verification of the index checksum. */
1566 int verify_index_checksum;
1567
1568 /* Allow fsck to force verification of the cache entry order. */
1569 int verify_ce_order;
1570
1571 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1572 {
1573 git_SHA_CTX c;
1574 unsigned char sha1[20];
1575 int hdr_version;
1576
1577 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1578 return error("bad signature");
1579 hdr_version = ntohl(hdr->hdr_version);
1580 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1581 return error("bad index version %d", hdr_version);
1582
1583 if (!verify_index_checksum)
1584 return 0;
1585
1586 git_SHA1_Init(&c);
1587 git_SHA1_Update(&c, hdr, size - 20);
1588 git_SHA1_Final(sha1, &c);
1589 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1590 return error("bad index file sha1 signature");
1591 return 0;
1592 }
1593
1594 static int read_index_extension(struct index_state *istate,
1595 const char *ext, void *data, unsigned long sz)
1596 {
1597 switch (CACHE_EXT(ext)) {
1598 case CACHE_EXT_TREE:
1599 istate->cache_tree = cache_tree_read(data, sz);
1600 break;
1601 case CACHE_EXT_RESOLVE_UNDO:
1602 istate->resolve_undo = resolve_undo_read(data, sz);
1603 break;
1604 case CACHE_EXT_LINK:
1605 if (read_link_extension(istate, data, sz))
1606 return -1;
1607 break;
1608 case CACHE_EXT_UNTRACKED:
1609 istate->untracked = read_untracked_extension(data, sz);
1610 break;
1611 case CACHE_EXT_FSMONITOR:
1612 read_fsmonitor_extension(istate, data, sz);
1613 break;
1614 default:
1615 if (*ext < 'A' || 'Z' < *ext)
1616 return error("index uses %.4s extension, which we do not understand",
1617 ext);
1618 fprintf(stderr, "ignoring %.4s extension\n", ext);
1619 break;
1620 }
1621 return 0;
1622 }
1623
1624 int hold_locked_index(struct lock_file *lk, int lock_flags)
1625 {
1626 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1627 }
1628
1629 int read_index(struct index_state *istate)
1630 {
1631 return read_index_from(istate, get_index_file(), get_git_dir());
1632 }
1633
1634 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1635 unsigned int flags,
1636 const char *name,
1637 size_t len)
1638 {
1639 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1640
1641 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1642 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1643 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1644 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1645 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1646 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1647 ce->ce_mode = get_be32(&ondisk->mode);
1648 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1649 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1650 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1651 ce->ce_flags = flags & ~CE_NAMEMASK;
1652 ce->ce_namelen = len;
1653 ce->index = 0;
1654 hashcpy(ce->oid.hash, ondisk->sha1);
1655 memcpy(ce->name, name, len);
1656 ce->name[len] = '\0';
1657 return ce;
1658 }
1659
1660 /*
1661 * Adjacent cache entries tend to share the leading paths, so it makes
1662 * sense to only store the differences in later entries. In the v4
1663 * on-disk format of the index, each on-disk cache entry stores the
1664 * number of bytes to be stripped from the end of the previous name,
1665 * and the bytes to append to the result, to come up with its name.
1666 */
1667 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1668 {
1669 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1670 size_t len = decode_varint(&cp);
1671
1672 if (name->len < len)
1673 die("malformed name field in the index");
1674 strbuf_remove(name, name->len - len, len);
1675 for (ep = cp; *ep; ep++)
1676 ; /* find the end */
1677 strbuf_add(name, cp, ep - cp);
1678 return (const char *)ep + 1 - cp_;
1679 }
1680
1681 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1682 unsigned long *ent_size,
1683 struct strbuf *previous_name)
1684 {
1685 struct cache_entry *ce;
1686 size_t len;
1687 const char *name;
1688 unsigned int flags;
1689
1690 /* On-disk flags are just 16 bits */
1691 flags = get_be16(&ondisk->flags);
1692 len = flags & CE_NAMEMASK;
1693
1694 if (flags & CE_EXTENDED) {
1695 struct ondisk_cache_entry_extended *ondisk2;
1696 int extended_flags;
1697 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1698 extended_flags = get_be16(&ondisk2->flags2) << 16;
1699 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1700 if (extended_flags & ~CE_EXTENDED_FLAGS)
1701 die("Unknown index entry format %08x", extended_flags);
1702 flags |= extended_flags;
1703 name = ondisk2->name;
1704 }
1705 else
1706 name = ondisk->name;
1707
1708 if (!previous_name) {
1709 /* v3 and earlier */
1710 if (len == CE_NAMEMASK)
1711 len = strlen(name);
1712 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1713
1714 *ent_size = ondisk_ce_size(ce);
1715 } else {
1716 unsigned long consumed;
1717 consumed = expand_name_field(previous_name, name);
1718 ce = cache_entry_from_ondisk(ondisk, flags,
1719 previous_name->buf,
1720 previous_name->len);
1721
1722 *ent_size = (name - ((char *)ondisk)) + consumed;
1723 }
1724 return ce;
1725 }
1726
1727 static void check_ce_order(struct index_state *istate)
1728 {
1729 unsigned int i;
1730
1731 if (!verify_ce_order)
1732 return;
1733
1734 for (i = 1; i < istate->cache_nr; i++) {
1735 struct cache_entry *ce = istate->cache[i - 1];
1736 struct cache_entry *next_ce = istate->cache[i];
1737 int name_compare = strcmp(ce->name, next_ce->name);
1738
1739 if (0 < name_compare)
1740 die("unordered stage entries in index");
1741 if (!name_compare) {
1742 if (!ce_stage(ce))
1743 die("multiple stage entries for merged file '%s'",
1744 ce->name);
1745 if (ce_stage(ce) > ce_stage(next_ce))
1746 die("unordered stage entries for '%s'",
1747 ce->name);
1748 }
1749 }
1750 }
1751
1752 static void tweak_untracked_cache(struct index_state *istate)
1753 {
1754 switch (git_config_get_untracked_cache()) {
1755 case -1: /* keep: do nothing */
1756 break;
1757 case 0: /* false */
1758 remove_untracked_cache(istate);
1759 break;
1760 case 1: /* true */
1761 add_untracked_cache(istate);
1762 break;
1763 default: /* unknown value: do nothing */
1764 break;
1765 }
1766 }
1767
1768 static void tweak_split_index(struct index_state *istate)
1769 {
1770 switch (git_config_get_split_index()) {
1771 case -1: /* unset: do nothing */
1772 break;
1773 case 0: /* false */
1774 remove_split_index(istate);
1775 break;
1776 case 1: /* true */
1777 add_split_index(istate);
1778 break;
1779 default: /* unknown value: do nothing */
1780 break;
1781 }
1782 }
1783
1784 static void post_read_index_from(struct index_state *istate)
1785 {
1786 check_ce_order(istate);
1787 tweak_untracked_cache(istate);
1788 tweak_split_index(istate);
1789 tweak_fsmonitor(istate);
1790 }
1791
1792 /* remember to discard_cache() before reading a different cache! */
1793 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1794 {
1795 int fd, i;
1796 struct stat st;
1797 unsigned long src_offset;
1798 struct cache_header *hdr;
1799 void *mmap;
1800 size_t mmap_size;
1801 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1802
1803 if (istate->initialized)
1804 return istate->cache_nr;
1805
1806 istate->timestamp.sec = 0;
1807 istate->timestamp.nsec = 0;
1808 fd = open(path, O_RDONLY);
1809 if (fd < 0) {
1810 if (!must_exist && errno == ENOENT)
1811 return 0;
1812 die_errno("%s: index file open failed", path);
1813 }
1814
1815 if (fstat(fd, &st))
1816 die_errno("cannot stat the open index");
1817
1818 mmap_size = xsize_t(st.st_size);
1819 if (mmap_size < sizeof(struct cache_header) + 20)
1820 die("index file smaller than expected");
1821
1822 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1823 if (mmap == MAP_FAILED)
1824 die_errno("unable to map index file");
1825 close(fd);
1826
1827 hdr = mmap;
1828 if (verify_hdr(hdr, mmap_size) < 0)
1829 goto unmap;
1830
1831 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1832 istate->version = ntohl(hdr->hdr_version);
1833 istate->cache_nr = ntohl(hdr->hdr_entries);
1834 istate->cache_alloc = alloc_nr(istate->cache_nr);
1835 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1836 istate->initialized = 1;
1837
1838 if (istate->version == 4)
1839 previous_name = &previous_name_buf;
1840 else
1841 previous_name = NULL;
1842
1843 src_offset = sizeof(*hdr);
1844 for (i = 0; i < istate->cache_nr; i++) {
1845 struct ondisk_cache_entry *disk_ce;
1846 struct cache_entry *ce;
1847 unsigned long consumed;
1848
1849 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1850 ce = create_from_disk(disk_ce, &consumed, previous_name);
1851 set_index_entry(istate, i, ce);
1852
1853 src_offset += consumed;
1854 }
1855 strbuf_release(&previous_name_buf);
1856 istate->timestamp.sec = st.st_mtime;
1857 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1858
1859 while (src_offset <= mmap_size - 20 - 8) {
1860 /* After an array of active_nr index entries,
1861 * there can be arbitrary number of extended
1862 * sections, each of which is prefixed with
1863 * extension name (4-byte) and section length
1864 * in 4-byte network byte order.
1865 */
1866 uint32_t extsize;
1867 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1868 extsize = ntohl(extsize);
1869 if (read_index_extension(istate,
1870 (const char *) mmap + src_offset,
1871 (char *) mmap + src_offset + 8,
1872 extsize) < 0)
1873 goto unmap;
1874 src_offset += 8;
1875 src_offset += extsize;
1876 }
1877 munmap(mmap, mmap_size);
1878 return istate->cache_nr;
1879
1880 unmap:
1881 munmap(mmap, mmap_size);
1882 die("index file corrupt");
1883 }
1884
1885 /*
1886 * Signal that the shared index is used by updating its mtime.
1887 *
1888 * This way, shared index can be removed if they have not been used
1889 * for some time.
1890 */
1891 static void freshen_shared_index(const char *shared_index, int warn)
1892 {
1893 if (!check_and_freshen_file(shared_index, 1) && warn)
1894 warning("could not freshen shared index '%s'", shared_index);
1895 }
1896
1897 int read_index_from(struct index_state *istate, const char *path,
1898 const char *gitdir)
1899 {
1900 struct split_index *split_index;
1901 int ret;
1902 char *base_sha1_hex;
1903 char *base_path;
1904
1905 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1906 if (istate->initialized)
1907 return istate->cache_nr;
1908
1909 ret = do_read_index(istate, path, 0);
1910
1911 split_index = istate->split_index;
1912 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1913 post_read_index_from(istate);
1914 return ret;
1915 }
1916
1917 if (split_index->base)
1918 discard_index(split_index->base);
1919 else
1920 split_index->base = xcalloc(1, sizeof(*split_index->base));
1921
1922 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1923 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_sha1_hex);
1924 ret = do_read_index(split_index->base, base_path, 1);
1925 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1926 die("broken index, expect %s in %s, got %s",
1927 base_sha1_hex, base_path,
1928 sha1_to_hex(split_index->base->sha1));
1929
1930 freshen_shared_index(base_path, 0);
1931 merge_base_index(istate);
1932 post_read_index_from(istate);
1933 free(base_path);
1934 return ret;
1935 }
1936
1937 int is_index_unborn(struct index_state *istate)
1938 {
1939 return (!istate->cache_nr && !istate->timestamp.sec);
1940 }
1941
1942 int discard_index(struct index_state *istate)
1943 {
1944 int i;
1945
1946 for (i = 0; i < istate->cache_nr; i++) {
1947 if (istate->cache[i]->index &&
1948 istate->split_index &&
1949 istate->split_index->base &&
1950 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1951 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1952 continue;
1953 free(istate->cache[i]);
1954 }
1955 resolve_undo_clear_index(istate);
1956 istate->cache_nr = 0;
1957 istate->cache_changed = 0;
1958 istate->timestamp.sec = 0;
1959 istate->timestamp.nsec = 0;
1960 free_name_hash(istate);
1961 cache_tree_free(&(istate->cache_tree));
1962 istate->initialized = 0;
1963 FREE_AND_NULL(istate->cache);
1964 istate->cache_alloc = 0;
1965 discard_split_index(istate);
1966 free_untracked_cache(istate->untracked);
1967 istate->untracked = NULL;
1968 return 0;
1969 }
1970
1971 int unmerged_index(const struct index_state *istate)
1972 {
1973 int i;
1974 for (i = 0; i < istate->cache_nr; i++) {
1975 if (ce_stage(istate->cache[i]))
1976 return 1;
1977 }
1978 return 0;
1979 }
1980
1981 #define WRITE_BUFFER_SIZE 8192
1982 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1983 static unsigned long write_buffer_len;
1984
1985 static int ce_write_flush(git_SHA_CTX *context, int fd)
1986 {
1987 unsigned int buffered = write_buffer_len;
1988 if (buffered) {
1989 git_SHA1_Update(context, write_buffer, buffered);
1990 if (write_in_full(fd, write_buffer, buffered) < 0)
1991 return -1;
1992 write_buffer_len = 0;
1993 }
1994 return 0;
1995 }
1996
1997 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1998 {
1999 while (len) {
2000 unsigned int buffered = write_buffer_len;
2001 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2002 if (partial > len)
2003 partial = len;
2004 memcpy(write_buffer + buffered, data, partial);
2005 buffered += partial;
2006 if (buffered == WRITE_BUFFER_SIZE) {
2007 write_buffer_len = buffered;
2008 if (ce_write_flush(context, fd))
2009 return -1;
2010 buffered = 0;
2011 }
2012 write_buffer_len = buffered;
2013 len -= partial;
2014 data = (char *) data + partial;
2015 }
2016 return 0;
2017 }
2018
2019 static int write_index_ext_header(git_SHA_CTX *context, int fd,
2020 unsigned int ext, unsigned int sz)
2021 {
2022 ext = htonl(ext);
2023 sz = htonl(sz);
2024 return ((ce_write(context, fd, &ext, 4) < 0) ||
2025 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2026 }
2027
2028 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
2029 {
2030 unsigned int left = write_buffer_len;
2031
2032 if (left) {
2033 write_buffer_len = 0;
2034 git_SHA1_Update(context, write_buffer, left);
2035 }
2036
2037 /* Flush first if not enough space for SHA1 signature */
2038 if (left + 20 > WRITE_BUFFER_SIZE) {
2039 if (write_in_full(fd, write_buffer, left) < 0)
2040 return -1;
2041 left = 0;
2042 }
2043
2044 /* Append the SHA1 signature at the end */
2045 git_SHA1_Final(write_buffer + left, context);
2046 hashcpy(sha1, write_buffer + left);
2047 left += 20;
2048 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2049 }
2050
2051 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2052 {
2053 /*
2054 * The only thing we care about in this function is to smudge the
2055 * falsely clean entry due to touch-update-touch race, so we leave
2056 * everything else as they are. We are called for entries whose
2057 * ce_stat_data.sd_mtime match the index file mtime.
2058 *
2059 * Note that this actually does not do much for gitlinks, for
2060 * which ce_match_stat_basic() always goes to the actual
2061 * contents. The caller checks with is_racy_timestamp() which
2062 * always says "no" for gitlinks, so we are not called for them ;-)
2063 */
2064 struct stat st;
2065
2066 if (lstat(ce->name, &st) < 0)
2067 return;
2068 if (ce_match_stat_basic(ce, &st))
2069 return;
2070 if (ce_modified_check_fs(ce, &st)) {
2071 /* This is "racily clean"; smudge it. Note that this
2072 * is a tricky code. At first glance, it may appear
2073 * that it can break with this sequence:
2074 *
2075 * $ echo xyzzy >frotz
2076 * $ git-update-index --add frotz
2077 * $ : >frotz
2078 * $ sleep 3
2079 * $ echo filfre >nitfol
2080 * $ git-update-index --add nitfol
2081 *
2082 * but it does not. When the second update-index runs,
2083 * it notices that the entry "frotz" has the same timestamp
2084 * as index, and if we were to smudge it by resetting its
2085 * size to zero here, then the object name recorded
2086 * in index is the 6-byte file but the cached stat information
2087 * becomes zero --- which would then match what we would
2088 * obtain from the filesystem next time we stat("frotz").
2089 *
2090 * However, the second update-index, before calling
2091 * this function, notices that the cached size is 6
2092 * bytes and what is on the filesystem is an empty
2093 * file, and never calls us, so the cached size information
2094 * for "frotz" stays 6 which does not match the filesystem.
2095 */
2096 ce->ce_stat_data.sd_size = 0;
2097 }
2098 }
2099
2100 /* Copy miscellaneous fields but not the name */
2101 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2102 struct cache_entry *ce)
2103 {
2104 short flags;
2105
2106 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2107 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2108 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2109 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2110 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2111 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2112 ondisk->mode = htonl(ce->ce_mode);
2113 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2114 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2115 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2116 hashcpy(ondisk->sha1, ce->oid.hash);
2117
2118 flags = ce->ce_flags & ~CE_NAMEMASK;
2119 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2120 ondisk->flags = htons(flags);
2121 if (ce->ce_flags & CE_EXTENDED) {
2122 struct ondisk_cache_entry_extended *ondisk2;
2123 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2124 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2125 }
2126 }
2127
2128 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
2129 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2130 {
2131 int size;
2132 int saved_namelen = saved_namelen; /* compiler workaround */
2133 int result;
2134 static unsigned char padding[8] = { 0x00 };
2135
2136 if (ce->ce_flags & CE_STRIP_NAME) {
2137 saved_namelen = ce_namelen(ce);
2138 ce->ce_namelen = 0;
2139 }
2140
2141 if (ce->ce_flags & CE_EXTENDED)
2142 size = offsetof(struct ondisk_cache_entry_extended, name);
2143 else
2144 size = offsetof(struct ondisk_cache_entry, name);
2145
2146 if (!previous_name) {
2147 int len = ce_namelen(ce);
2148 copy_cache_entry_to_ondisk(ondisk, ce);
2149 result = ce_write(c, fd, ondisk, size);
2150 if (!result)
2151 result = ce_write(c, fd, ce->name, len);
2152 if (!result)
2153 result = ce_write(c, fd, padding, align_padding_size(size, len));
2154 } else {
2155 int common, to_remove, prefix_size;
2156 unsigned char to_remove_vi[16];
2157 for (common = 0;
2158 (ce->name[common] &&
2159 common < previous_name->len &&
2160 ce->name[common] == previous_name->buf[common]);
2161 common++)
2162 ; /* still matching */
2163 to_remove = previous_name->len - common;
2164 prefix_size = encode_varint(to_remove, to_remove_vi);
2165
2166 copy_cache_entry_to_ondisk(ondisk, ce);
2167 result = ce_write(c, fd, ondisk, size);
2168 if (!result)
2169 result = ce_write(c, fd, to_remove_vi, prefix_size);
2170 if (!result)
2171 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2172 if (!result)
2173 result = ce_write(c, fd, padding, 1);
2174
2175 strbuf_splice(previous_name, common, to_remove,
2176 ce->name + common, ce_namelen(ce) - common);
2177 }
2178 if (ce->ce_flags & CE_STRIP_NAME) {
2179 ce->ce_namelen = saved_namelen;
2180 ce->ce_flags &= ~CE_STRIP_NAME;
2181 }
2182
2183 return result;
2184 }
2185
2186 /*
2187 * This function verifies if index_state has the correct sha1 of the
2188 * index file. Don't die if we have any other failure, just return 0.
2189 */
2190 static int verify_index_from(const struct index_state *istate, const char *path)
2191 {
2192 int fd;
2193 ssize_t n;
2194 struct stat st;
2195 unsigned char sha1[20];
2196
2197 if (!istate->initialized)
2198 return 0;
2199
2200 fd = open(path, O_RDONLY);
2201 if (fd < 0)
2202 return 0;
2203
2204 if (fstat(fd, &st))
2205 goto out;
2206
2207 if (st.st_size < sizeof(struct cache_header) + 20)
2208 goto out;
2209
2210 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2211 if (n != 20)
2212 goto out;
2213
2214 if (hashcmp(istate->sha1, sha1))
2215 goto out;
2216
2217 close(fd);
2218 return 1;
2219
2220 out:
2221 close(fd);
2222 return 0;
2223 }
2224
2225 static int verify_index(const struct index_state *istate)
2226 {
2227 return verify_index_from(istate, get_index_file());
2228 }
2229
2230 static int has_racy_timestamp(struct index_state *istate)
2231 {
2232 int entries = istate->cache_nr;
2233 int i;
2234
2235 for (i = 0; i < entries; i++) {
2236 struct cache_entry *ce = istate->cache[i];
2237 if (is_racy_timestamp(istate, ce))
2238 return 1;
2239 }
2240 return 0;
2241 }
2242
2243 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2244 {
2245 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2246 verify_index(istate))
2247 write_locked_index(istate, lockfile, COMMIT_LOCK);
2248 else
2249 rollback_lock_file(lockfile);
2250 }
2251
2252 /*
2253 * On success, `tempfile` is closed. If it is the temporary file
2254 * of a `struct lock_file`, we will therefore effectively perform
2255 * a 'close_lock_file_gently()`. Since that is an implementation
2256 * detail of lockfiles, callers of `do_write_index()` should not
2257 * rely on it.
2258 */
2259 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2260 int strip_extensions)
2261 {
2262 int newfd = tempfile->fd;
2263 git_SHA_CTX c;
2264 struct cache_header hdr;
2265 int i, err = 0, removed, extended, hdr_version;
2266 struct cache_entry **cache = istate->cache;
2267 int entries = istate->cache_nr;
2268 struct stat st;
2269 struct ondisk_cache_entry_extended ondisk;
2270 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2271 int drop_cache_tree = istate->drop_cache_tree;
2272
2273 for (i = removed = extended = 0; i < entries; i++) {
2274 if (cache[i]->ce_flags & CE_REMOVE)
2275 removed++;
2276
2277 /* reduce extended entries if possible */
2278 cache[i]->ce_flags &= ~CE_EXTENDED;
2279 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2280 extended++;
2281 cache[i]->ce_flags |= CE_EXTENDED;
2282 }
2283 }
2284
2285 if (!istate->version) {
2286 istate->version = get_index_format_default();
2287 if (getenv("GIT_TEST_SPLIT_INDEX"))
2288 init_split_index(istate);
2289 }
2290
2291 /* demote version 3 to version 2 when the latter suffices */
2292 if (istate->version == 3 || istate->version == 2)
2293 istate->version = extended ? 3 : 2;
2294
2295 hdr_version = istate->version;
2296
2297 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2298 hdr.hdr_version = htonl(hdr_version);
2299 hdr.hdr_entries = htonl(entries - removed);
2300
2301 git_SHA1_Init(&c);
2302 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2303 return -1;
2304
2305 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2306
2307 for (i = 0; i < entries; i++) {
2308 struct cache_entry *ce = cache[i];
2309 if (ce->ce_flags & CE_REMOVE)
2310 continue;
2311 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2312 ce_smudge_racily_clean_entry(ce);
2313 if (is_null_oid(&ce->oid)) {
2314 static const char msg[] = "cache entry has null sha1: %s";
2315 static int allow = -1;
2316
2317 if (allow < 0)
2318 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2319 if (allow)
2320 warning(msg, ce->name);
2321 else
2322 err = error(msg, ce->name);
2323
2324 drop_cache_tree = 1;
2325 }
2326 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2327 err = -1;
2328
2329 if (err)
2330 break;
2331 }
2332 strbuf_release(&previous_name_buf);
2333
2334 if (err)
2335 return err;
2336
2337 /* Write extension data here */
2338 if (!strip_extensions && istate->split_index) {
2339 struct strbuf sb = STRBUF_INIT;
2340
2341 err = write_link_extension(&sb, istate) < 0 ||
2342 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2343 sb.len) < 0 ||
2344 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2345 strbuf_release(&sb);
2346 if (err)
2347 return -1;
2348 }
2349 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2350 struct strbuf sb = STRBUF_INIT;
2351
2352 cache_tree_write(&sb, istate->cache_tree);
2353 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2354 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2355 strbuf_release(&sb);
2356 if (err)
2357 return -1;
2358 }
2359 if (!strip_extensions && istate->resolve_undo) {
2360 struct strbuf sb = STRBUF_INIT;
2361
2362 resolve_undo_write(&sb, istate->resolve_undo);
2363 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2364 sb.len) < 0
2365 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2366 strbuf_release(&sb);
2367 if (err)
2368 return -1;
2369 }
2370 if (!strip_extensions && istate->untracked) {
2371 struct strbuf sb = STRBUF_INIT;
2372
2373 write_untracked_extension(&sb, istate->untracked);
2374 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2375 sb.len) < 0 ||
2376 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2377 strbuf_release(&sb);
2378 if (err)
2379 return -1;
2380 }
2381 if (!strip_extensions && istate->fsmonitor_last_update) {
2382 struct strbuf sb = STRBUF_INIT;
2383
2384 write_fsmonitor_extension(&sb, istate);
2385 err = write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2386 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2387 strbuf_release(&sb);
2388 if (err)
2389 return -1;
2390 }
2391
2392 if (ce_flush(&c, newfd, istate->sha1))
2393 return -1;
2394 if (close_tempfile_gently(tempfile)) {
2395 error(_("could not close '%s'"), tempfile->filename.buf);
2396 return -1;
2397 }
2398 if (stat(tempfile->filename.buf, &st))
2399 return -1;
2400 istate->timestamp.sec = (unsigned int)st.st_mtime;
2401 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2402 return 0;
2403 }
2404
2405 void set_alternate_index_output(const char *name)
2406 {
2407 alternate_index_output = name;
2408 }
2409
2410 static int commit_locked_index(struct lock_file *lk)
2411 {
2412 if (alternate_index_output)
2413 return commit_lock_file_to(lk, alternate_index_output);
2414 else
2415 return commit_lock_file(lk);
2416 }
2417
2418 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2419 unsigned flags)
2420 {
2421 int ret = do_write_index(istate, lock->tempfile, 0);
2422 if (ret)
2423 return ret;
2424 if (flags & COMMIT_LOCK)
2425 return commit_locked_index(lock);
2426 return close_lock_file_gently(lock);
2427 }
2428
2429 static int write_split_index(struct index_state *istate,
2430 struct lock_file *lock,
2431 unsigned flags)
2432 {
2433 int ret;
2434 prepare_to_write_split_index(istate);
2435 ret = do_write_locked_index(istate, lock, flags);
2436 finish_writing_split_index(istate);
2437 return ret;
2438 }
2439
2440 static const char *shared_index_expire = "2.weeks.ago";
2441
2442 static unsigned long get_shared_index_expire_date(void)
2443 {
2444 static unsigned long shared_index_expire_date;
2445 static int shared_index_expire_date_prepared;
2446
2447 if (!shared_index_expire_date_prepared) {
2448 git_config_get_expiry("splitindex.sharedindexexpire",
2449 &shared_index_expire);
2450 shared_index_expire_date = approxidate(shared_index_expire);
2451 shared_index_expire_date_prepared = 1;
2452 }
2453
2454 return shared_index_expire_date;
2455 }
2456
2457 static int should_delete_shared_index(const char *shared_index_path)
2458 {
2459 struct stat st;
2460 unsigned long expiration;
2461
2462 /* Check timestamp */
2463 expiration = get_shared_index_expire_date();
2464 if (!expiration)
2465 return 0;
2466 if (stat(shared_index_path, &st))
2467 return error_errno(_("could not stat '%s'"), shared_index_path);
2468 if (st.st_mtime > expiration)
2469 return 0;
2470
2471 return 1;
2472 }
2473
2474 static int clean_shared_index_files(const char *current_hex)
2475 {
2476 struct dirent *de;
2477 DIR *dir = opendir(get_git_dir());
2478
2479 if (!dir)
2480 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2481
2482 while ((de = readdir(dir)) != NULL) {
2483 const char *sha1_hex;
2484 const char *shared_index_path;
2485 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2486 continue;
2487 if (!strcmp(sha1_hex, current_hex))
2488 continue;
2489 shared_index_path = git_path("%s", de->d_name);
2490 if (should_delete_shared_index(shared_index_path) > 0 &&
2491 unlink(shared_index_path))
2492 warning_errno(_("unable to unlink: %s"), shared_index_path);
2493 }
2494 closedir(dir);
2495
2496 return 0;
2497 }
2498
2499 static int write_shared_index(struct index_state *istate,
2500 struct tempfile **temp)
2501 {
2502 struct split_index *si = istate->split_index;
2503 int ret;
2504
2505 move_cache_to_base_index(istate);
2506 ret = do_write_index(si->base, *temp, 1);
2507 if (ret)
2508 return ret;
2509 ret = adjust_shared_perm(get_tempfile_path(*temp));
2510 if (ret) {
2511 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
2512 return ret;
2513 }
2514 ret = rename_tempfile(temp,
2515 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2516 if (!ret) {
2517 hashcpy(si->base_sha1, si->base->sha1);
2518 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2519 }
2520
2521 return ret;
2522 }
2523
2524 static const int default_max_percent_split_change = 20;
2525
2526 static int too_many_not_shared_entries(struct index_state *istate)
2527 {
2528 int i, not_shared = 0;
2529 int max_split = git_config_get_max_percent_split_change();
2530
2531 switch (max_split) {
2532 case -1:
2533 /* not or badly configured: use the default value */
2534 max_split = default_max_percent_split_change;
2535 break;
2536 case 0:
2537 return 1; /* 0% means always write a new shared index */
2538 case 100:
2539 return 0; /* 100% means never write a new shared index */
2540 default:
2541 break; /* just use the configured value */
2542 }
2543
2544 /* Count not shared entries */
2545 for (i = 0; i < istate->cache_nr; i++) {
2546 struct cache_entry *ce = istate->cache[i];
2547 if (!ce->index)
2548 not_shared++;
2549 }
2550
2551 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2552 }
2553
2554 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2555 unsigned flags)
2556 {
2557 int new_shared_index, ret;
2558 struct split_index *si = istate->split_index;
2559
2560 if (istate->fsmonitor_last_update)
2561 fill_fsmonitor_bitmap(istate);
2562
2563 if (!si || alternate_index_output ||
2564 (istate->cache_changed & ~EXTMASK)) {
2565 if (si)
2566 hashclr(si->base_sha1);
2567 ret = do_write_locked_index(istate, lock, flags);
2568 goto out;
2569 }
2570
2571 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2572 int v = si->base_sha1[0];
2573 if ((v & 15) < 6)
2574 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2575 }
2576 if (too_many_not_shared_entries(istate))
2577 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2578
2579 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2580
2581 if (new_shared_index) {
2582 struct tempfile *temp;
2583 int saved_errno;
2584
2585 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2586 if (!temp) {
2587 hashclr(si->base_sha1);
2588 ret = do_write_locked_index(istate, lock, flags);
2589 goto out;
2590 }
2591 ret = write_shared_index(istate, &temp);
2592
2593 saved_errno = errno;
2594 if (is_tempfile_active(temp))
2595 delete_tempfile(&temp);
2596 errno = saved_errno;
2597
2598 if (ret)
2599 goto out;
2600 }
2601
2602 ret = write_split_index(istate, lock, flags);
2603
2604 /* Freshen the shared index only if the split-index was written */
2605 if (!ret && !new_shared_index) {
2606 const char *shared_index = git_path("sharedindex.%s",
2607 sha1_to_hex(si->base_sha1));
2608 freshen_shared_index(shared_index, 1);
2609 }
2610
2611 out:
2612 if (flags & COMMIT_LOCK)
2613 rollback_lock_file(lock);
2614 return ret;
2615 }
2616
2617 /*
2618 * Read the index file that is potentially unmerged into given
2619 * index_state, dropping any unmerged entries. Returns true if
2620 * the index is unmerged. Callers who want to refuse to work
2621 * from an unmerged state can call this and check its return value,
2622 * instead of calling read_cache().
2623 */
2624 int read_index_unmerged(struct index_state *istate)
2625 {
2626 int i;
2627 int unmerged = 0;
2628
2629 read_index(istate);
2630 for (i = 0; i < istate->cache_nr; i++) {
2631 struct cache_entry *ce = istate->cache[i];
2632 struct cache_entry *new_ce;
2633 int size, len;
2634
2635 if (!ce_stage(ce))
2636 continue;
2637 unmerged = 1;
2638 len = ce_namelen(ce);
2639 size = cache_entry_size(len);
2640 new_ce = xcalloc(1, size);
2641 memcpy(new_ce->name, ce->name, len);
2642 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2643 new_ce->ce_namelen = len;
2644 new_ce->ce_mode = ce->ce_mode;
2645 if (add_index_entry(istate, new_ce, 0))
2646 return error("%s: cannot drop to stage #0",
2647 new_ce->name);
2648 }
2649 return unmerged;
2650 }
2651
2652 /*
2653 * Returns 1 if the path is an "other" path with respect to
2654 * the index; that is, the path is not mentioned in the index at all,
2655 * either as a file, a directory with some files in the index,
2656 * or as an unmerged entry.
2657 *
2658 * We helpfully remove a trailing "/" from directories so that
2659 * the output of read_directory can be used as-is.
2660 */
2661 int index_name_is_other(const struct index_state *istate, const char *name,
2662 int namelen)
2663 {
2664 int pos;
2665 if (namelen && name[namelen - 1] == '/')
2666 namelen--;
2667 pos = index_name_pos(istate, name, namelen);
2668 if (0 <= pos)
2669 return 0; /* exact match */
2670 pos = -pos - 1;
2671 if (pos < istate->cache_nr) {
2672 struct cache_entry *ce = istate->cache[pos];
2673 if (ce_namelen(ce) == namelen &&
2674 !memcmp(ce->name, name, namelen))
2675 return 0; /* Yup, this one exists unmerged */
2676 }
2677 return 1;
2678 }
2679
2680 void *read_blob_data_from_index(const struct index_state *istate,
2681 const char *path, unsigned long *size)
2682 {
2683 int pos, len;
2684 unsigned long sz;
2685 enum object_type type;
2686 void *data;
2687
2688 len = strlen(path);
2689 pos = index_name_pos(istate, path, len);
2690 if (pos < 0) {
2691 /*
2692 * We might be in the middle of a merge, in which
2693 * case we would read stage #2 (ours).
2694 */
2695 int i;
2696 for (i = -pos - 1;
2697 (pos < 0 && i < istate->cache_nr &&
2698 !strcmp(istate->cache[i]->name, path));
2699 i++)
2700 if (ce_stage(istate->cache[i]) == 2)
2701 pos = i;
2702 }
2703 if (pos < 0)
2704 return NULL;
2705 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2706 if (!data || type != OBJ_BLOB) {
2707 free(data);
2708 return NULL;
2709 }
2710 if (size)
2711 *size = sz;
2712 return data;
2713 }
2714
2715 void stat_validity_clear(struct stat_validity *sv)
2716 {
2717 FREE_AND_NULL(sv->sd);
2718 }
2719
2720 int stat_validity_check(struct stat_validity *sv, const char *path)
2721 {
2722 struct stat st;
2723
2724 if (stat(path, &st) < 0)
2725 return sv->sd == NULL;
2726 if (!sv->sd)
2727 return 0;
2728 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2729 }
2730
2731 void stat_validity_update(struct stat_validity *sv, int fd)
2732 {
2733 struct stat st;
2734
2735 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2736 stat_validity_clear(sv);
2737 else {
2738 if (!sv->sd)
2739 sv->sd = xcalloc(1, sizeof(struct stat_data));
2740 fill_stat_data(sv->sd, &st);
2741 }
2742 }
2743
2744 void move_index_extensions(struct index_state *dst, struct index_state *src)
2745 {
2746 dst->untracked = src->untracked;
2747 src->untracked = NULL;
2748 }