Fix builtin checkout crashing when given an invalid path
[git/jrn.git] / read-cache.c
blob657f0c5894c65831b80ceee54d161d0beac1d733
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "cache-tree.h"
9 #include "refs.h"
10 #include "dir.h"
12 /* Index extensions.
14 * The first letter should be 'A'..'Z' for extensions that are not
15 * necessary for a correct operation (i.e. optimization data).
16 * When new extensions are added that _needs_ to be understood in
17 * order to correctly interpret the index file, pick character that
18 * is outside the range, to cause the reader to abort.
21 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
22 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
24 struct index_state the_index;
26 static unsigned int hash_name(const char *name, int namelen)
28 unsigned int hash = 0x123;
30 do {
31 unsigned char c = *name++;
32 hash = hash*101 + c;
33 } while (--namelen);
34 return hash;
37 static void hash_index_entry(struct index_state *istate, struct cache_entry *ce)
39 void **pos;
40 unsigned int hash;
42 if (ce->ce_flags & CE_HASHED)
43 return;
44 ce->ce_flags |= CE_HASHED;
45 ce->next = NULL;
46 hash = hash_name(ce->name, ce_namelen(ce));
47 pos = insert_hash(hash, ce, &istate->name_hash);
48 if (pos) {
49 ce->next = *pos;
50 *pos = ce;
54 static void lazy_init_name_hash(struct index_state *istate)
56 int nr;
58 if (istate->name_hash_initialized)
59 return;
60 for (nr = 0; nr < istate->cache_nr; nr++)
61 hash_index_entry(istate, istate->cache[nr]);
62 istate->name_hash_initialized = 1;
65 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
67 ce->ce_flags &= ~CE_UNHASHED;
68 istate->cache[nr] = ce;
69 if (istate->name_hash_initialized)
70 hash_index_entry(istate, ce);
73 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
75 struct cache_entry *old = istate->cache[nr];
77 remove_index_entry(old);
78 set_index_entry(istate, nr, ce);
79 istate->cache_changed = 1;
82 int index_name_exists(struct index_state *istate, const char *name, int namelen)
84 unsigned int hash = hash_name(name, namelen);
85 struct cache_entry *ce;
87 lazy_init_name_hash(istate);
88 ce = lookup_hash(hash, &istate->name_hash);
90 while (ce) {
91 if (!(ce->ce_flags & CE_UNHASHED)) {
92 if (!cache_name_compare(name, namelen, ce->name, ce->ce_flags))
93 return 1;
95 ce = ce->next;
97 return 0;
101 * This only updates the "non-critical" parts of the directory
102 * cache, ie the parts that aren't tracked by GIT, and only used
103 * to validate the cache.
105 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
107 ce->ce_ctime = st->st_ctime;
108 ce->ce_mtime = st->st_mtime;
109 ce->ce_dev = st->st_dev;
110 ce->ce_ino = st->st_ino;
111 ce->ce_uid = st->st_uid;
112 ce->ce_gid = st->st_gid;
113 ce->ce_size = st->st_size;
115 if (assume_unchanged)
116 ce->ce_flags |= CE_VALID;
118 if (S_ISREG(st->st_mode))
119 ce_mark_uptodate(ce);
122 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
124 int match = -1;
125 int fd = open(ce->name, O_RDONLY);
127 if (fd >= 0) {
128 unsigned char sha1[20];
129 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
130 match = hashcmp(sha1, ce->sha1);
131 /* index_fd() closed the file descriptor already */
133 return match;
136 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
138 int match = -1;
139 char *target;
140 void *buffer;
141 unsigned long size;
142 enum object_type type;
143 int len;
145 target = xmalloc(expected_size);
146 len = readlink(ce->name, target, expected_size);
147 if (len != expected_size) {
148 free(target);
149 return -1;
151 buffer = read_sha1_file(ce->sha1, &type, &size);
152 if (!buffer) {
153 free(target);
154 return -1;
156 if (size == expected_size)
157 match = memcmp(buffer, target, size);
158 free(buffer);
159 free(target);
160 return match;
163 static int ce_compare_gitlink(struct cache_entry *ce)
165 unsigned char sha1[20];
168 * We don't actually require that the .git directory
169 * under GITLINK directory be a valid git directory. It
170 * might even be missing (in case nobody populated that
171 * sub-project).
173 * If so, we consider it always to match.
175 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
176 return 0;
177 return hashcmp(sha1, ce->sha1);
180 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
182 switch (st->st_mode & S_IFMT) {
183 case S_IFREG:
184 if (ce_compare_data(ce, st))
185 return DATA_CHANGED;
186 break;
187 case S_IFLNK:
188 if (ce_compare_link(ce, xsize_t(st->st_size)))
189 return DATA_CHANGED;
190 break;
191 case S_IFDIR:
192 if (S_ISGITLINK(ce->ce_mode))
193 return 0;
194 default:
195 return TYPE_CHANGED;
197 return 0;
200 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
202 unsigned int changed = 0;
204 if (ce->ce_flags & CE_REMOVE)
205 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
207 switch (ce->ce_mode & S_IFMT) {
208 case S_IFREG:
209 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
210 /* We consider only the owner x bit to be relevant for
211 * "mode changes"
213 if (trust_executable_bit &&
214 (0100 & (ce->ce_mode ^ st->st_mode)))
215 changed |= MODE_CHANGED;
216 break;
217 case S_IFLNK:
218 if (!S_ISLNK(st->st_mode) &&
219 (has_symlinks || !S_ISREG(st->st_mode)))
220 changed |= TYPE_CHANGED;
221 break;
222 case S_IFGITLINK:
223 if (!S_ISDIR(st->st_mode))
224 changed |= TYPE_CHANGED;
225 else if (ce_compare_gitlink(ce))
226 changed |= DATA_CHANGED;
227 return changed;
228 default:
229 die("internal error: ce_mode is %o", ce->ce_mode);
231 if (ce->ce_mtime != (unsigned int) st->st_mtime)
232 changed |= MTIME_CHANGED;
233 if (ce->ce_ctime != (unsigned int) st->st_ctime)
234 changed |= CTIME_CHANGED;
236 if (ce->ce_uid != (unsigned int) st->st_uid ||
237 ce->ce_gid != (unsigned int) st->st_gid)
238 changed |= OWNER_CHANGED;
239 if (ce->ce_ino != (unsigned int) st->st_ino)
240 changed |= INODE_CHANGED;
242 #ifdef USE_STDEV
244 * st_dev breaks on network filesystems where different
245 * clients will have different views of what "device"
246 * the filesystem is on
248 if (ce->ce_dev != (unsigned int) st->st_dev)
249 changed |= INODE_CHANGED;
250 #endif
252 if (ce->ce_size != (unsigned int) st->st_size)
253 changed |= DATA_CHANGED;
255 return changed;
258 static int is_racy_timestamp(struct index_state *istate, struct cache_entry *ce)
260 return (istate->timestamp &&
261 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
264 int ie_match_stat(struct index_state *istate,
265 struct cache_entry *ce, struct stat *st,
266 unsigned int options)
268 unsigned int changed;
269 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
270 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
273 * If it's marked as always valid in the index, it's
274 * valid whatever the checked-out copy says.
276 if (!ignore_valid && (ce->ce_flags & CE_VALID))
277 return 0;
279 changed = ce_match_stat_basic(ce, st);
282 * Within 1 second of this sequence:
283 * echo xyzzy >file && git-update-index --add file
284 * running this command:
285 * echo frotz >file
286 * would give a falsely clean cache entry. The mtime and
287 * length match the cache, and other stat fields do not change.
289 * We could detect this at update-index time (the cache entry
290 * being registered/updated records the same time as "now")
291 * and delay the return from git-update-index, but that would
292 * effectively mean we can make at most one commit per second,
293 * which is not acceptable. Instead, we check cache entries
294 * whose mtime are the same as the index file timestamp more
295 * carefully than others.
297 if (!changed && is_racy_timestamp(istate, ce)) {
298 if (assume_racy_is_modified)
299 changed |= DATA_CHANGED;
300 else
301 changed |= ce_modified_check_fs(ce, st);
304 return changed;
307 int ie_modified(struct index_state *istate,
308 struct cache_entry *ce, struct stat *st, unsigned int options)
310 int changed, changed_fs;
312 changed = ie_match_stat(istate, ce, st, options);
313 if (!changed)
314 return 0;
316 * If the mode or type has changed, there's no point in trying
317 * to refresh the entry - it's not going to match
319 if (changed & (MODE_CHANGED | TYPE_CHANGED))
320 return changed;
322 /* Immediately after read-tree or update-index --cacheinfo,
323 * the length field is zero. For other cases the ce_size
324 * should match the SHA1 recorded in the index entry.
326 if ((changed & DATA_CHANGED) && ce->ce_size != 0)
327 return changed;
329 changed_fs = ce_modified_check_fs(ce, st);
330 if (changed_fs)
331 return changed | changed_fs;
332 return 0;
335 int base_name_compare(const char *name1, int len1, int mode1,
336 const char *name2, int len2, int mode2)
338 unsigned char c1, c2;
339 int len = len1 < len2 ? len1 : len2;
340 int cmp;
342 cmp = memcmp(name1, name2, len);
343 if (cmp)
344 return cmp;
345 c1 = name1[len];
346 c2 = name2[len];
347 if (!c1 && S_ISDIR(mode1))
348 c1 = '/';
349 if (!c2 && S_ISDIR(mode2))
350 c2 = '/';
351 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
354 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
356 int len1 = flags1 & CE_NAMEMASK;
357 int len2 = flags2 & CE_NAMEMASK;
358 int len = len1 < len2 ? len1 : len2;
359 int cmp;
361 cmp = memcmp(name1, name2, len);
362 if (cmp)
363 return cmp;
364 if (len1 < len2)
365 return -1;
366 if (len1 > len2)
367 return 1;
369 /* Compare stages */
370 flags1 &= CE_STAGEMASK;
371 flags2 &= CE_STAGEMASK;
373 if (flags1 < flags2)
374 return -1;
375 if (flags1 > flags2)
376 return 1;
377 return 0;
380 int index_name_pos(struct index_state *istate, const char *name, int namelen)
382 int first, last;
384 first = 0;
385 last = istate->cache_nr;
386 while (last > first) {
387 int next = (last + first) >> 1;
388 struct cache_entry *ce = istate->cache[next];
389 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
390 if (!cmp)
391 return next;
392 if (cmp < 0) {
393 last = next;
394 continue;
396 first = next+1;
398 return -first-1;
401 /* Remove entry, return true if there are more entries to go.. */
402 int remove_index_entry_at(struct index_state *istate, int pos)
404 struct cache_entry *ce = istate->cache[pos];
406 remove_index_entry(ce);
407 istate->cache_changed = 1;
408 istate->cache_nr--;
409 if (pos >= istate->cache_nr)
410 return 0;
411 memmove(istate->cache + pos,
412 istate->cache + pos + 1,
413 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
414 return 1;
417 int remove_file_from_index(struct index_state *istate, const char *path)
419 int pos = index_name_pos(istate, path, strlen(path));
420 if (pos < 0)
421 pos = -pos-1;
422 cache_tree_invalidate_path(istate->cache_tree, path);
423 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
424 remove_index_entry_at(istate, pos);
425 return 0;
428 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
430 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
433 static int index_name_pos_also_unmerged(struct index_state *istate,
434 const char *path, int namelen)
436 int pos = index_name_pos(istate, path, namelen);
437 struct cache_entry *ce;
439 if (pos >= 0)
440 return pos;
442 /* maybe unmerged? */
443 pos = -1 - pos;
444 if (pos >= istate->cache_nr ||
445 compare_name((ce = istate->cache[pos]), path, namelen))
446 return -1;
448 /* order of preference: stage 2, 1, 3 */
449 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
450 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
451 !compare_name(ce, path, namelen))
452 pos++;
453 return pos;
456 int add_file_to_index(struct index_state *istate, const char *path, int verbose)
458 int size, namelen, pos;
459 struct stat st;
460 struct cache_entry *ce;
461 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
463 if (lstat(path, &st))
464 die("%s: unable to stat (%s)", path, strerror(errno));
466 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
467 die("%s: can only add regular files, symbolic links or git-directories", path);
469 namelen = strlen(path);
470 if (S_ISDIR(st.st_mode)) {
471 while (namelen && path[namelen-1] == '/')
472 namelen--;
474 size = cache_entry_size(namelen);
475 ce = xcalloc(1, size);
476 memcpy(ce->name, path, namelen);
477 ce->ce_flags = namelen;
478 fill_stat_cache_info(ce, &st);
480 if (trust_executable_bit && has_symlinks)
481 ce->ce_mode = create_ce_mode(st.st_mode);
482 else {
483 /* If there is an existing entry, pick the mode bits and type
484 * from it, otherwise assume unexecutable regular file.
486 struct cache_entry *ent;
487 int pos = index_name_pos_also_unmerged(istate, path, namelen);
489 ent = (0 <= pos) ? istate->cache[pos] : NULL;
490 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
493 pos = index_name_pos(istate, ce->name, namelen);
494 if (0 <= pos &&
495 !ce_stage(istate->cache[pos]) &&
496 !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) {
497 /* Nothing changed, really */
498 free(ce);
499 ce_mark_uptodate(istate->cache[pos]);
500 return 0;
503 if (index_path(ce->sha1, path, &st, 1))
504 die("unable to index file %s", path);
505 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
506 die("unable to add %s to index",path);
507 if (verbose)
508 printf("add '%s'\n", path);
509 return 0;
512 struct cache_entry *make_cache_entry(unsigned int mode,
513 const unsigned char *sha1, const char *path, int stage,
514 int refresh)
516 int size, len;
517 struct cache_entry *ce;
519 if (!verify_path(path))
520 return NULL;
522 len = strlen(path);
523 size = cache_entry_size(len);
524 ce = xcalloc(1, size);
526 hashcpy(ce->sha1, sha1);
527 memcpy(ce->name, path, len);
528 ce->ce_flags = create_ce_flags(len, stage);
529 ce->ce_mode = create_ce_mode(mode);
531 if (refresh)
532 return refresh_cache_entry(ce, 0);
534 return ce;
537 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
539 int len = ce_namelen(a);
540 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
543 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
545 const char *match, *name;
546 int len;
548 if (!pathspec)
549 return 1;
551 len = ce_namelen(ce);
552 name = ce->name;
553 while ((match = *pathspec++) != NULL) {
554 int matchlen = strlen(match);
555 if (matchlen > len)
556 continue;
557 if (memcmp(name, match, matchlen))
558 continue;
559 if (matchlen && name[matchlen-1] == '/')
560 return 1;
561 if (name[matchlen] == '/' || !name[matchlen])
562 return 1;
563 if (!matchlen)
564 return 1;
566 return 0;
570 * We fundamentally don't like some paths: we don't want
571 * dot or dot-dot anywhere, and for obvious reasons don't
572 * want to recurse into ".git" either.
574 * Also, we don't want double slashes or slashes at the
575 * end that can make pathnames ambiguous.
577 static int verify_dotfile(const char *rest)
580 * The first character was '.', but that
581 * has already been discarded, we now test
582 * the rest.
584 switch (*rest) {
585 /* "." is not allowed */
586 case '\0': case '/':
587 return 0;
590 * ".git" followed by NUL or slash is bad. This
591 * shares the path end test with the ".." case.
593 case 'g':
594 if (rest[1] != 'i')
595 break;
596 if (rest[2] != 't')
597 break;
598 rest += 2;
599 /* fallthrough */
600 case '.':
601 if (rest[1] == '\0' || rest[1] == '/')
602 return 0;
604 return 1;
607 int verify_path(const char *path)
609 char c;
611 goto inside;
612 for (;;) {
613 if (!c)
614 return 1;
615 if (c == '/') {
616 inside:
617 c = *path++;
618 switch (c) {
619 default:
620 continue;
621 case '/': case '\0':
622 break;
623 case '.':
624 if (verify_dotfile(path))
625 continue;
627 return 0;
629 c = *path++;
634 * Do we have another file that has the beginning components being a
635 * proper superset of the name we're trying to add?
637 static int has_file_name(struct index_state *istate,
638 const struct cache_entry *ce, int pos, int ok_to_replace)
640 int retval = 0;
641 int len = ce_namelen(ce);
642 int stage = ce_stage(ce);
643 const char *name = ce->name;
645 while (pos < istate->cache_nr) {
646 struct cache_entry *p = istate->cache[pos++];
648 if (len >= ce_namelen(p))
649 break;
650 if (memcmp(name, p->name, len))
651 break;
652 if (ce_stage(p) != stage)
653 continue;
654 if (p->name[len] != '/')
655 continue;
656 if (p->ce_flags & CE_REMOVE)
657 continue;
658 retval = -1;
659 if (!ok_to_replace)
660 break;
661 remove_index_entry_at(istate, --pos);
663 return retval;
667 * Do we have another file with a pathname that is a proper
668 * subset of the name we're trying to add?
670 static int has_dir_name(struct index_state *istate,
671 const struct cache_entry *ce, int pos, int ok_to_replace)
673 int retval = 0;
674 int stage = ce_stage(ce);
675 const char *name = ce->name;
676 const char *slash = name + ce_namelen(ce);
678 for (;;) {
679 int len;
681 for (;;) {
682 if (*--slash == '/')
683 break;
684 if (slash <= ce->name)
685 return retval;
687 len = slash - name;
689 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
690 if (pos >= 0) {
692 * Found one, but not so fast. This could
693 * be a marker that says "I was here, but
694 * I am being removed". Such an entry is
695 * not a part of the resulting tree, and
696 * it is Ok to have a directory at the same
697 * path.
699 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
700 retval = -1;
701 if (!ok_to_replace)
702 break;
703 remove_index_entry_at(istate, pos);
704 continue;
707 else
708 pos = -pos-1;
711 * Trivial optimization: if we find an entry that
712 * already matches the sub-directory, then we know
713 * we're ok, and we can exit.
715 while (pos < istate->cache_nr) {
716 struct cache_entry *p = istate->cache[pos];
717 if ((ce_namelen(p) <= len) ||
718 (p->name[len] != '/') ||
719 memcmp(p->name, name, len))
720 break; /* not our subdirectory */
721 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
723 * p is at the same stage as our entry, and
724 * is a subdirectory of what we are looking
725 * at, so we cannot have conflicts at our
726 * level or anything shorter.
728 return retval;
729 pos++;
732 return retval;
735 /* We may be in a situation where we already have path/file and path
736 * is being added, or we already have path and path/file is being
737 * added. Either one would result in a nonsense tree that has path
738 * twice when git-write-tree tries to write it out. Prevent it.
740 * If ok-to-replace is specified, we remove the conflicting entries
741 * from the cache so the caller should recompute the insert position.
742 * When this happens, we return non-zero.
744 static int check_file_directory_conflict(struct index_state *istate,
745 const struct cache_entry *ce,
746 int pos, int ok_to_replace)
748 int retval;
751 * When ce is an "I am going away" entry, we allow it to be added
753 if (ce->ce_flags & CE_REMOVE)
754 return 0;
757 * We check if the path is a sub-path of a subsequent pathname
758 * first, since removing those will not change the position
759 * in the array.
761 retval = has_file_name(istate, ce, pos, ok_to_replace);
764 * Then check if the path might have a clashing sub-directory
765 * before it.
767 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
770 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
772 int pos;
773 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
774 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
775 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
777 cache_tree_invalidate_path(istate->cache_tree, ce->name);
778 pos = index_name_pos(istate, ce->name, ce->ce_flags);
780 /* existing match? Just replace it. */
781 if (pos >= 0) {
782 replace_index_entry(istate, pos, ce);
783 return 0;
785 pos = -pos-1;
788 * Inserting a merged entry ("stage 0") into the index
789 * will always replace all non-merged entries..
791 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
792 while (ce_same_name(istate->cache[pos], ce)) {
793 ok_to_add = 1;
794 if (!remove_index_entry_at(istate, pos))
795 break;
799 if (!ok_to_add)
800 return -1;
801 if (!verify_path(ce->name))
802 return -1;
804 if (!skip_df_check &&
805 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
806 if (!ok_to_replace)
807 return error("'%s' appears as both a file and as a directory",
808 ce->name);
809 pos = index_name_pos(istate, ce->name, ce->ce_flags);
810 pos = -pos-1;
812 return pos + 1;
815 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
817 int pos;
819 if (option & ADD_CACHE_JUST_APPEND)
820 pos = istate->cache_nr;
821 else {
822 int ret;
823 ret = add_index_entry_with_check(istate, ce, option);
824 if (ret <= 0)
825 return ret;
826 pos = ret - 1;
829 /* Make sure the array is big enough .. */
830 if (istate->cache_nr == istate->cache_alloc) {
831 istate->cache_alloc = alloc_nr(istate->cache_alloc);
832 istate->cache = xrealloc(istate->cache,
833 istate->cache_alloc * sizeof(struct cache_entry *));
836 /* Add it in.. */
837 istate->cache_nr++;
838 if (istate->cache_nr > pos + 1)
839 memmove(istate->cache + pos + 1,
840 istate->cache + pos,
841 (istate->cache_nr - pos - 1) * sizeof(ce));
842 set_index_entry(istate, pos, ce);
843 istate->cache_changed = 1;
844 return 0;
848 * "refresh" does not calculate a new sha1 file or bring the
849 * cache up-to-date for mode/content changes. But what it
850 * _does_ do is to "re-match" the stat information of a file
851 * with the cache, so that you can refresh the cache for a
852 * file that hasn't been changed but where the stat entry is
853 * out of date.
855 * For example, you'd want to do this after doing a "git-read-tree",
856 * to link up the stat cache details with the proper files.
858 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
859 struct cache_entry *ce,
860 unsigned int options, int *err)
862 struct stat st;
863 struct cache_entry *updated;
864 int changed, size;
865 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
867 if (ce_uptodate(ce))
868 return ce;
870 if (lstat(ce->name, &st) < 0) {
871 if (err)
872 *err = errno;
873 return NULL;
876 changed = ie_match_stat(istate, ce, &st, options);
877 if (!changed) {
879 * The path is unchanged. If we were told to ignore
880 * valid bit, then we did the actual stat check and
881 * found that the entry is unmodified. If the entry
882 * is not marked VALID, this is the place to mark it
883 * valid again, under "assume unchanged" mode.
885 if (ignore_valid && assume_unchanged &&
886 !(ce->ce_flags & CE_VALID))
887 ; /* mark this one VALID again */
888 else {
890 * We do not mark the index itself "modified"
891 * because CE_UPTODATE flag is in-core only;
892 * we are not going to write this change out.
894 ce_mark_uptodate(ce);
895 return ce;
899 if (ie_modified(istate, ce, &st, options)) {
900 if (err)
901 *err = EINVAL;
902 return NULL;
905 size = ce_size(ce);
906 updated = xmalloc(size);
907 memcpy(updated, ce, size);
908 fill_stat_cache_info(updated, &st);
910 * If ignore_valid is not set, we should leave CE_VALID bit
911 * alone. Otherwise, paths marked with --no-assume-unchanged
912 * (i.e. things to be edited) will reacquire CE_VALID bit
913 * automatically, which is not really what we want.
915 if (!ignore_valid && assume_unchanged &&
916 !(ce->ce_flags & CE_VALID))
917 updated->ce_flags &= ~CE_VALID;
919 return updated;
922 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
924 int i;
925 int has_errors = 0;
926 int really = (flags & REFRESH_REALLY) != 0;
927 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
928 int quiet = (flags & REFRESH_QUIET) != 0;
929 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
930 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
932 for (i = 0; i < istate->cache_nr; i++) {
933 struct cache_entry *ce, *new;
934 int cache_errno = 0;
936 ce = istate->cache[i];
937 if (ce_stage(ce)) {
938 while ((i < istate->cache_nr) &&
939 ! strcmp(istate->cache[i]->name, ce->name))
940 i++;
941 i--;
942 if (allow_unmerged)
943 continue;
944 printf("%s: needs merge\n", ce->name);
945 has_errors = 1;
946 continue;
949 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
950 continue;
952 new = refresh_cache_ent(istate, ce, options, &cache_errno);
953 if (new == ce)
954 continue;
955 if (!new) {
956 if (not_new && cache_errno == ENOENT)
957 continue;
958 if (really && cache_errno == EINVAL) {
959 /* If we are doing --really-refresh that
960 * means the index is not valid anymore.
962 ce->ce_flags &= ~CE_VALID;
963 istate->cache_changed = 1;
965 if (quiet)
966 continue;
967 printf("%s: needs update\n", ce->name);
968 has_errors = 1;
969 continue;
972 replace_index_entry(istate, i, new);
974 return has_errors;
977 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
979 return refresh_cache_ent(&the_index, ce, really, NULL);
982 static int verify_hdr(struct cache_header *hdr, unsigned long size)
984 SHA_CTX c;
985 unsigned char sha1[20];
987 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
988 return error("bad signature");
989 if (hdr->hdr_version != htonl(2))
990 return error("bad index version");
991 SHA1_Init(&c);
992 SHA1_Update(&c, hdr, size - 20);
993 SHA1_Final(sha1, &c);
994 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
995 return error("bad index file sha1 signature");
996 return 0;
999 static int read_index_extension(struct index_state *istate,
1000 const char *ext, void *data, unsigned long sz)
1002 switch (CACHE_EXT(ext)) {
1003 case CACHE_EXT_TREE:
1004 istate->cache_tree = cache_tree_read(data, sz);
1005 break;
1006 default:
1007 if (*ext < 'A' || 'Z' < *ext)
1008 return error("index uses %.4s extension, which we do not understand",
1009 ext);
1010 fprintf(stderr, "ignoring %.4s extension\n", ext);
1011 break;
1013 return 0;
1016 int read_index(struct index_state *istate)
1018 return read_index_from(istate, get_index_file());
1021 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1023 size_t len;
1025 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1026 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1027 ce->ce_dev = ntohl(ondisk->dev);
1028 ce->ce_ino = ntohl(ondisk->ino);
1029 ce->ce_mode = ntohl(ondisk->mode);
1030 ce->ce_uid = ntohl(ondisk->uid);
1031 ce->ce_gid = ntohl(ondisk->gid);
1032 ce->ce_size = ntohl(ondisk->size);
1033 /* On-disk flags are just 16 bits */
1034 ce->ce_flags = ntohs(ondisk->flags);
1035 hashcpy(ce->sha1, ondisk->sha1);
1037 len = ce->ce_flags & CE_NAMEMASK;
1038 if (len == CE_NAMEMASK)
1039 len = strlen(ondisk->name);
1041 * NEEDSWORK: If the original index is crafted, this copy could
1042 * go unchecked.
1044 memcpy(ce->name, ondisk->name, len + 1);
1047 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1049 long per_entry;
1051 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1054 * Alignment can cause differences. This should be "alignof", but
1055 * since that's a gcc'ism, just use the size of a pointer.
1057 per_entry += sizeof(void *);
1058 return ondisk_size + entries*per_entry;
1061 /* remember to discard_cache() before reading a different cache! */
1062 int read_index_from(struct index_state *istate, const char *path)
1064 int fd, i;
1065 struct stat st;
1066 unsigned long src_offset, dst_offset;
1067 struct cache_header *hdr;
1068 void *mmap;
1069 size_t mmap_size;
1071 errno = EBUSY;
1072 if (istate->alloc)
1073 return istate->cache_nr;
1075 errno = ENOENT;
1076 istate->timestamp = 0;
1077 fd = open(path, O_RDONLY);
1078 if (fd < 0) {
1079 if (errno == ENOENT)
1080 return 0;
1081 die("index file open failed (%s)", strerror(errno));
1084 if (fstat(fd, &st))
1085 die("cannot stat the open index (%s)", strerror(errno));
1087 errno = EINVAL;
1088 mmap_size = xsize_t(st.st_size);
1089 if (mmap_size < sizeof(struct cache_header) + 20)
1090 die("index file smaller than expected");
1092 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1093 close(fd);
1094 if (mmap == MAP_FAILED)
1095 die("unable to map index file");
1097 hdr = mmap;
1098 if (verify_hdr(hdr, mmap_size) < 0)
1099 goto unmap;
1101 istate->cache_nr = ntohl(hdr->hdr_entries);
1102 istate->cache_alloc = alloc_nr(istate->cache_nr);
1103 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1106 * The disk format is actually larger than the in-memory format,
1107 * due to space for nsec etc, so even though the in-memory one
1108 * has room for a few more flags, we can allocate using the same
1109 * index size
1111 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1113 src_offset = sizeof(*hdr);
1114 dst_offset = 0;
1115 for (i = 0; i < istate->cache_nr; i++) {
1116 struct ondisk_cache_entry *disk_ce;
1117 struct cache_entry *ce;
1119 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1120 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1121 convert_from_disk(disk_ce, ce);
1122 set_index_entry(istate, i, ce);
1124 src_offset += ondisk_ce_size(ce);
1125 dst_offset += ce_size(ce);
1127 istate->timestamp = st.st_mtime;
1128 while (src_offset <= mmap_size - 20 - 8) {
1129 /* After an array of active_nr index entries,
1130 * there can be arbitrary number of extended
1131 * sections, each of which is prefixed with
1132 * extension name (4-byte) and section length
1133 * in 4-byte network byte order.
1135 unsigned long extsize;
1136 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1137 extsize = ntohl(extsize);
1138 if (read_index_extension(istate,
1139 (const char *) mmap + src_offset,
1140 (char *) mmap + src_offset + 8,
1141 extsize) < 0)
1142 goto unmap;
1143 src_offset += 8;
1144 src_offset += extsize;
1146 munmap(mmap, mmap_size);
1147 return istate->cache_nr;
1149 unmap:
1150 munmap(mmap, mmap_size);
1151 errno = EINVAL;
1152 die("index file corrupt");
1155 int discard_index(struct index_state *istate)
1157 istate->cache_nr = 0;
1158 istate->cache_changed = 0;
1159 istate->timestamp = 0;
1160 free_hash(&istate->name_hash);
1161 cache_tree_free(&(istate->cache_tree));
1162 free(istate->alloc);
1163 istate->alloc = NULL;
1165 /* no need to throw away allocated active_cache */
1166 return 0;
1169 int unmerged_index(struct index_state *istate)
1171 int i;
1172 for (i = 0; i < istate->cache_nr; i++) {
1173 if (ce_stage(istate->cache[i]))
1174 return 1;
1176 return 0;
1179 #define WRITE_BUFFER_SIZE 8192
1180 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1181 static unsigned long write_buffer_len;
1183 static int ce_write_flush(SHA_CTX *context, int fd)
1185 unsigned int buffered = write_buffer_len;
1186 if (buffered) {
1187 SHA1_Update(context, write_buffer, buffered);
1188 if (write_in_full(fd, write_buffer, buffered) != buffered)
1189 return -1;
1190 write_buffer_len = 0;
1192 return 0;
1195 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1197 while (len) {
1198 unsigned int buffered = write_buffer_len;
1199 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1200 if (partial > len)
1201 partial = len;
1202 memcpy(write_buffer + buffered, data, partial);
1203 buffered += partial;
1204 if (buffered == WRITE_BUFFER_SIZE) {
1205 write_buffer_len = buffered;
1206 if (ce_write_flush(context, fd))
1207 return -1;
1208 buffered = 0;
1210 write_buffer_len = buffered;
1211 len -= partial;
1212 data = (char *) data + partial;
1214 return 0;
1217 static int write_index_ext_header(SHA_CTX *context, int fd,
1218 unsigned int ext, unsigned int sz)
1220 ext = htonl(ext);
1221 sz = htonl(sz);
1222 return ((ce_write(context, fd, &ext, 4) < 0) ||
1223 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1226 static int ce_flush(SHA_CTX *context, int fd)
1228 unsigned int left = write_buffer_len;
1230 if (left) {
1231 write_buffer_len = 0;
1232 SHA1_Update(context, write_buffer, left);
1235 /* Flush first if not enough space for SHA1 signature */
1236 if (left + 20 > WRITE_BUFFER_SIZE) {
1237 if (write_in_full(fd, write_buffer, left) != left)
1238 return -1;
1239 left = 0;
1242 /* Append the SHA1 signature at the end */
1243 SHA1_Final(write_buffer + left, context);
1244 left += 20;
1245 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1248 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1251 * The only thing we care about in this function is to smudge the
1252 * falsely clean entry due to touch-update-touch race, so we leave
1253 * everything else as they are. We are called for entries whose
1254 * ce_mtime match the index file mtime.
1256 struct stat st;
1258 if (lstat(ce->name, &st) < 0)
1259 return;
1260 if (ce_match_stat_basic(ce, &st))
1261 return;
1262 if (ce_modified_check_fs(ce, &st)) {
1263 /* This is "racily clean"; smudge it. Note that this
1264 * is a tricky code. At first glance, it may appear
1265 * that it can break with this sequence:
1267 * $ echo xyzzy >frotz
1268 * $ git-update-index --add frotz
1269 * $ : >frotz
1270 * $ sleep 3
1271 * $ echo filfre >nitfol
1272 * $ git-update-index --add nitfol
1274 * but it does not. When the second update-index runs,
1275 * it notices that the entry "frotz" has the same timestamp
1276 * as index, and if we were to smudge it by resetting its
1277 * size to zero here, then the object name recorded
1278 * in index is the 6-byte file but the cached stat information
1279 * becomes zero --- which would then match what we would
1280 * obtain from the filesystem next time we stat("frotz").
1282 * However, the second update-index, before calling
1283 * this function, notices that the cached size is 6
1284 * bytes and what is on the filesystem is an empty
1285 * file, and never calls us, so the cached size information
1286 * for "frotz" stays 6 which does not match the filesystem.
1288 ce->ce_size = 0;
1292 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1294 int size = ondisk_ce_size(ce);
1295 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1297 ondisk->ctime.sec = htonl(ce->ce_ctime);
1298 ondisk->ctime.nsec = 0;
1299 ondisk->mtime.sec = htonl(ce->ce_mtime);
1300 ondisk->mtime.nsec = 0;
1301 ondisk->dev = htonl(ce->ce_dev);
1302 ondisk->ino = htonl(ce->ce_ino);
1303 ondisk->mode = htonl(ce->ce_mode);
1304 ondisk->uid = htonl(ce->ce_uid);
1305 ondisk->gid = htonl(ce->ce_gid);
1306 ondisk->size = htonl(ce->ce_size);
1307 hashcpy(ondisk->sha1, ce->sha1);
1308 ondisk->flags = htons(ce->ce_flags);
1309 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1311 return ce_write(c, fd, ondisk, size);
1314 int write_index(struct index_state *istate, int newfd)
1316 SHA_CTX c;
1317 struct cache_header hdr;
1318 int i, err, removed;
1319 struct cache_entry **cache = istate->cache;
1320 int entries = istate->cache_nr;
1322 for (i = removed = 0; i < entries; i++)
1323 if (cache[i]->ce_flags & CE_REMOVE)
1324 removed++;
1326 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1327 hdr.hdr_version = htonl(2);
1328 hdr.hdr_entries = htonl(entries - removed);
1330 SHA1_Init(&c);
1331 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1332 return -1;
1334 for (i = 0; i < entries; i++) {
1335 struct cache_entry *ce = cache[i];
1336 if (ce->ce_flags & CE_REMOVE)
1337 continue;
1338 if (is_racy_timestamp(istate, ce))
1339 ce_smudge_racily_clean_entry(ce);
1340 if (ce_write_entry(&c, newfd, ce) < 0)
1341 return -1;
1344 /* Write extension data here */
1345 if (istate->cache_tree) {
1346 struct strbuf sb;
1348 strbuf_init(&sb, 0);
1349 cache_tree_write(&sb, istate->cache_tree);
1350 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1351 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1352 strbuf_release(&sb);
1353 if (err)
1354 return -1;
1356 return ce_flush(&c, newfd);