Don't try and percent-escape existing percent escapes in git-svn URIs
[git/dscho.git] / read-cache.c
bloba92b25b59bf0e096942bca126542a1ea411b525b
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(const 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(const 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(const 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;
355 * df_name_compare() is identical to base_name_compare(), except it
356 * compares conflicting directory/file entries as equal. Note that
357 * while a directory name compares as equal to a regular file, they
358 * then individually compare _differently_ to a filename that has
359 * a dot after the basename (because '\0' < '.' < '/').
361 * This is used by routines that want to traverse the git namespace
362 * but then handle conflicting entries together when possible.
364 int df_name_compare(const char *name1, int len1, int mode1,
365 const char *name2, int len2, int mode2)
367 int len = len1 < len2 ? len1 : len2, cmp;
368 unsigned char c1, c2;
370 cmp = memcmp(name1, name2, len);
371 if (cmp)
372 return cmp;
373 /* Directories and files compare equal (same length, same name) */
374 if (len1 == len2)
375 return 0;
376 c1 = name1[len];
377 if (!c1 && S_ISDIR(mode1))
378 c1 = '/';
379 c2 = name2[len];
380 if (!c2 && S_ISDIR(mode2))
381 c2 = '/';
382 if (c1 == '/' && !c2)
383 return 0;
384 if (c2 == '/' && !c1)
385 return 0;
386 return c1 - c2;
389 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
391 int len1 = flags1 & CE_NAMEMASK;
392 int len2 = flags2 & CE_NAMEMASK;
393 int len = len1 < len2 ? len1 : len2;
394 int cmp;
396 cmp = memcmp(name1, name2, len);
397 if (cmp)
398 return cmp;
399 if (len1 < len2)
400 return -1;
401 if (len1 > len2)
402 return 1;
404 /* Compare stages */
405 flags1 &= CE_STAGEMASK;
406 flags2 &= CE_STAGEMASK;
408 if (flags1 < flags2)
409 return -1;
410 if (flags1 > flags2)
411 return 1;
412 return 0;
415 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
417 int first, last;
419 first = 0;
420 last = istate->cache_nr;
421 while (last > first) {
422 int next = (last + first) >> 1;
423 struct cache_entry *ce = istate->cache[next];
424 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
425 if (!cmp)
426 return next;
427 if (cmp < 0) {
428 last = next;
429 continue;
431 first = next+1;
433 return -first-1;
436 /* Remove entry, return true if there are more entries to go.. */
437 int remove_index_entry_at(struct index_state *istate, int pos)
439 struct cache_entry *ce = istate->cache[pos];
441 remove_index_entry(ce);
442 istate->cache_changed = 1;
443 istate->cache_nr--;
444 if (pos >= istate->cache_nr)
445 return 0;
446 memmove(istate->cache + pos,
447 istate->cache + pos + 1,
448 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
449 return 1;
452 int remove_file_from_index(struct index_state *istate, const char *path)
454 int pos = index_name_pos(istate, path, strlen(path));
455 if (pos < 0)
456 pos = -pos-1;
457 cache_tree_invalidate_path(istate->cache_tree, path);
458 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
459 remove_index_entry_at(istate, pos);
460 return 0;
463 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
465 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
468 static int index_name_pos_also_unmerged(struct index_state *istate,
469 const char *path, int namelen)
471 int pos = index_name_pos(istate, path, namelen);
472 struct cache_entry *ce;
474 if (pos >= 0)
475 return pos;
477 /* maybe unmerged? */
478 pos = -1 - pos;
479 if (pos >= istate->cache_nr ||
480 compare_name((ce = istate->cache[pos]), path, namelen))
481 return -1;
483 /* order of preference: stage 2, 1, 3 */
484 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
485 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
486 !compare_name(ce, path, namelen))
487 pos++;
488 return pos;
491 int add_file_to_index(struct index_state *istate, const char *path, int verbose)
493 int size, namelen, pos;
494 struct stat st;
495 struct cache_entry *ce;
496 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
498 if (lstat(path, &st))
499 die("%s: unable to stat (%s)", path, strerror(errno));
501 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
502 die("%s: can only add regular files, symbolic links or git-directories", path);
504 namelen = strlen(path);
505 if (S_ISDIR(st.st_mode)) {
506 while (namelen && path[namelen-1] == '/')
507 namelen--;
509 size = cache_entry_size(namelen);
510 ce = xcalloc(1, size);
511 memcpy(ce->name, path, namelen);
512 ce->ce_flags = namelen;
513 fill_stat_cache_info(ce, &st);
515 if (trust_executable_bit && has_symlinks)
516 ce->ce_mode = create_ce_mode(st.st_mode);
517 else {
518 /* If there is an existing entry, pick the mode bits and type
519 * from it, otherwise assume unexecutable regular file.
521 struct cache_entry *ent;
522 int pos = index_name_pos_also_unmerged(istate, path, namelen);
524 ent = (0 <= pos) ? istate->cache[pos] : NULL;
525 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
528 pos = index_name_pos(istate, ce->name, namelen);
529 if (0 <= pos &&
530 !ce_stage(istate->cache[pos]) &&
531 !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) {
532 /* Nothing changed, really */
533 free(ce);
534 ce_mark_uptodate(istate->cache[pos]);
535 return 0;
538 if (index_path(ce->sha1, path, &st, 1))
539 die("unable to index file %s", path);
540 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
541 die("unable to add %s to index",path);
542 if (verbose)
543 printf("add '%s'\n", path);
544 return 0;
547 struct cache_entry *make_cache_entry(unsigned int mode,
548 const unsigned char *sha1, const char *path, int stage,
549 int refresh)
551 int size, len;
552 struct cache_entry *ce;
554 if (!verify_path(path))
555 return NULL;
557 len = strlen(path);
558 size = cache_entry_size(len);
559 ce = xcalloc(1, size);
561 hashcpy(ce->sha1, sha1);
562 memcpy(ce->name, path, len);
563 ce->ce_flags = create_ce_flags(len, stage);
564 ce->ce_mode = create_ce_mode(mode);
566 if (refresh)
567 return refresh_cache_entry(ce, 0);
569 return ce;
572 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
574 int len = ce_namelen(a);
575 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
578 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
580 const char *match, *name;
581 int len;
583 if (!pathspec)
584 return 1;
586 len = ce_namelen(ce);
587 name = ce->name;
588 while ((match = *pathspec++) != NULL) {
589 int matchlen = strlen(match);
590 if (matchlen > len)
591 continue;
592 if (memcmp(name, match, matchlen))
593 continue;
594 if (matchlen && name[matchlen-1] == '/')
595 return 1;
596 if (name[matchlen] == '/' || !name[matchlen])
597 return 1;
598 if (!matchlen)
599 return 1;
601 return 0;
605 * We fundamentally don't like some paths: we don't want
606 * dot or dot-dot anywhere, and for obvious reasons don't
607 * want to recurse into ".git" either.
609 * Also, we don't want double slashes or slashes at the
610 * end that can make pathnames ambiguous.
612 static int verify_dotfile(const char *rest)
615 * The first character was '.', but that
616 * has already been discarded, we now test
617 * the rest.
619 switch (*rest) {
620 /* "." is not allowed */
621 case '\0': case '/':
622 return 0;
625 * ".git" followed by NUL or slash is bad. This
626 * shares the path end test with the ".." case.
628 case 'g':
629 if (rest[1] != 'i')
630 break;
631 if (rest[2] != 't')
632 break;
633 rest += 2;
634 /* fallthrough */
635 case '.':
636 if (rest[1] == '\0' || rest[1] == '/')
637 return 0;
639 return 1;
642 int verify_path(const char *path)
644 char c;
646 goto inside;
647 for (;;) {
648 if (!c)
649 return 1;
650 if (c == '/') {
651 inside:
652 c = *path++;
653 switch (c) {
654 default:
655 continue;
656 case '/': case '\0':
657 break;
658 case '.':
659 if (verify_dotfile(path))
660 continue;
662 return 0;
664 c = *path++;
669 * Do we have another file that has the beginning components being a
670 * proper superset of the name we're trying to add?
672 static int has_file_name(struct index_state *istate,
673 const struct cache_entry *ce, int pos, int ok_to_replace)
675 int retval = 0;
676 int len = ce_namelen(ce);
677 int stage = ce_stage(ce);
678 const char *name = ce->name;
680 while (pos < istate->cache_nr) {
681 struct cache_entry *p = istate->cache[pos++];
683 if (len >= ce_namelen(p))
684 break;
685 if (memcmp(name, p->name, len))
686 break;
687 if (ce_stage(p) != stage)
688 continue;
689 if (p->name[len] != '/')
690 continue;
691 if (p->ce_flags & CE_REMOVE)
692 continue;
693 retval = -1;
694 if (!ok_to_replace)
695 break;
696 remove_index_entry_at(istate, --pos);
698 return retval;
702 * Do we have another file with a pathname that is a proper
703 * subset of the name we're trying to add?
705 static int has_dir_name(struct index_state *istate,
706 const struct cache_entry *ce, int pos, int ok_to_replace)
708 int retval = 0;
709 int stage = ce_stage(ce);
710 const char *name = ce->name;
711 const char *slash = name + ce_namelen(ce);
713 for (;;) {
714 int len;
716 for (;;) {
717 if (*--slash == '/')
718 break;
719 if (slash <= ce->name)
720 return retval;
722 len = slash - name;
724 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
725 if (pos >= 0) {
727 * Found one, but not so fast. This could
728 * be a marker that says "I was here, but
729 * I am being removed". Such an entry is
730 * not a part of the resulting tree, and
731 * it is Ok to have a directory at the same
732 * path.
734 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
735 retval = -1;
736 if (!ok_to_replace)
737 break;
738 remove_index_entry_at(istate, pos);
739 continue;
742 else
743 pos = -pos-1;
746 * Trivial optimization: if we find an entry that
747 * already matches the sub-directory, then we know
748 * we're ok, and we can exit.
750 while (pos < istate->cache_nr) {
751 struct cache_entry *p = istate->cache[pos];
752 if ((ce_namelen(p) <= len) ||
753 (p->name[len] != '/') ||
754 memcmp(p->name, name, len))
755 break; /* not our subdirectory */
756 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
758 * p is at the same stage as our entry, and
759 * is a subdirectory of what we are looking
760 * at, so we cannot have conflicts at our
761 * level or anything shorter.
763 return retval;
764 pos++;
767 return retval;
770 /* We may be in a situation where we already have path/file and path
771 * is being added, or we already have path and path/file is being
772 * added. Either one would result in a nonsense tree that has path
773 * twice when git-write-tree tries to write it out. Prevent it.
775 * If ok-to-replace is specified, we remove the conflicting entries
776 * from the cache so the caller should recompute the insert position.
777 * When this happens, we return non-zero.
779 static int check_file_directory_conflict(struct index_state *istate,
780 const struct cache_entry *ce,
781 int pos, int ok_to_replace)
783 int retval;
786 * When ce is an "I am going away" entry, we allow it to be added
788 if (ce->ce_flags & CE_REMOVE)
789 return 0;
792 * We check if the path is a sub-path of a subsequent pathname
793 * first, since removing those will not change the position
794 * in the array.
796 retval = has_file_name(istate, ce, pos, ok_to_replace);
799 * Then check if the path might have a clashing sub-directory
800 * before it.
802 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
805 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
807 int pos;
808 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
809 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
810 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
812 cache_tree_invalidate_path(istate->cache_tree, ce->name);
813 pos = index_name_pos(istate, ce->name, ce->ce_flags);
815 /* existing match? Just replace it. */
816 if (pos >= 0) {
817 replace_index_entry(istate, pos, ce);
818 return 0;
820 pos = -pos-1;
823 * Inserting a merged entry ("stage 0") into the index
824 * will always replace all non-merged entries..
826 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
827 while (ce_same_name(istate->cache[pos], ce)) {
828 ok_to_add = 1;
829 if (!remove_index_entry_at(istate, pos))
830 break;
834 if (!ok_to_add)
835 return -1;
836 if (!verify_path(ce->name))
837 return -1;
839 if (!skip_df_check &&
840 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
841 if (!ok_to_replace)
842 return error("'%s' appears as both a file and as a directory",
843 ce->name);
844 pos = index_name_pos(istate, ce->name, ce->ce_flags);
845 pos = -pos-1;
847 return pos + 1;
850 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
852 int pos;
854 if (option & ADD_CACHE_JUST_APPEND)
855 pos = istate->cache_nr;
856 else {
857 int ret;
858 ret = add_index_entry_with_check(istate, ce, option);
859 if (ret <= 0)
860 return ret;
861 pos = ret - 1;
864 /* Make sure the array is big enough .. */
865 if (istate->cache_nr == istate->cache_alloc) {
866 istate->cache_alloc = alloc_nr(istate->cache_alloc);
867 istate->cache = xrealloc(istate->cache,
868 istate->cache_alloc * sizeof(struct cache_entry *));
871 /* Add it in.. */
872 istate->cache_nr++;
873 if (istate->cache_nr > pos + 1)
874 memmove(istate->cache + pos + 1,
875 istate->cache + pos,
876 (istate->cache_nr - pos - 1) * sizeof(ce));
877 set_index_entry(istate, pos, ce);
878 istate->cache_changed = 1;
879 return 0;
883 * "refresh" does not calculate a new sha1 file or bring the
884 * cache up-to-date for mode/content changes. But what it
885 * _does_ do is to "re-match" the stat information of a file
886 * with the cache, so that you can refresh the cache for a
887 * file that hasn't been changed but where the stat entry is
888 * out of date.
890 * For example, you'd want to do this after doing a "git-read-tree",
891 * to link up the stat cache details with the proper files.
893 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
894 struct cache_entry *ce,
895 unsigned int options, int *err)
897 struct stat st;
898 struct cache_entry *updated;
899 int changed, size;
900 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
902 if (ce_uptodate(ce))
903 return ce;
905 if (lstat(ce->name, &st) < 0) {
906 if (err)
907 *err = errno;
908 return NULL;
911 changed = ie_match_stat(istate, ce, &st, options);
912 if (!changed) {
914 * The path is unchanged. If we were told to ignore
915 * valid bit, then we did the actual stat check and
916 * found that the entry is unmodified. If the entry
917 * is not marked VALID, this is the place to mark it
918 * valid again, under "assume unchanged" mode.
920 if (ignore_valid && assume_unchanged &&
921 !(ce->ce_flags & CE_VALID))
922 ; /* mark this one VALID again */
923 else {
925 * We do not mark the index itself "modified"
926 * because CE_UPTODATE flag is in-core only;
927 * we are not going to write this change out.
929 ce_mark_uptodate(ce);
930 return ce;
934 if (ie_modified(istate, ce, &st, options)) {
935 if (err)
936 *err = EINVAL;
937 return NULL;
940 size = ce_size(ce);
941 updated = xmalloc(size);
942 memcpy(updated, ce, size);
943 fill_stat_cache_info(updated, &st);
945 * If ignore_valid is not set, we should leave CE_VALID bit
946 * alone. Otherwise, paths marked with --no-assume-unchanged
947 * (i.e. things to be edited) will reacquire CE_VALID bit
948 * automatically, which is not really what we want.
950 if (!ignore_valid && assume_unchanged &&
951 !(ce->ce_flags & CE_VALID))
952 updated->ce_flags &= ~CE_VALID;
954 return updated;
957 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
959 int i;
960 int has_errors = 0;
961 int really = (flags & REFRESH_REALLY) != 0;
962 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
963 int quiet = (flags & REFRESH_QUIET) != 0;
964 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
965 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
967 for (i = 0; i < istate->cache_nr; i++) {
968 struct cache_entry *ce, *new;
969 int cache_errno = 0;
971 ce = istate->cache[i];
972 if (ce_stage(ce)) {
973 while ((i < istate->cache_nr) &&
974 ! strcmp(istate->cache[i]->name, ce->name))
975 i++;
976 i--;
977 if (allow_unmerged)
978 continue;
979 printf("%s: needs merge\n", ce->name);
980 has_errors = 1;
981 continue;
984 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
985 continue;
987 new = refresh_cache_ent(istate, ce, options, &cache_errno);
988 if (new == ce)
989 continue;
990 if (!new) {
991 if (not_new && cache_errno == ENOENT)
992 continue;
993 if (really && cache_errno == EINVAL) {
994 /* If we are doing --really-refresh that
995 * means the index is not valid anymore.
997 ce->ce_flags &= ~CE_VALID;
998 istate->cache_changed = 1;
1000 if (quiet)
1001 continue;
1002 printf("%s: needs update\n", ce->name);
1003 has_errors = 1;
1004 continue;
1007 replace_index_entry(istate, i, new);
1009 return has_errors;
1012 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1014 return refresh_cache_ent(&the_index, ce, really, NULL);
1017 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1019 SHA_CTX c;
1020 unsigned char sha1[20];
1022 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1023 return error("bad signature");
1024 if (hdr->hdr_version != htonl(2))
1025 return error("bad index version");
1026 SHA1_Init(&c);
1027 SHA1_Update(&c, hdr, size - 20);
1028 SHA1_Final(sha1, &c);
1029 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1030 return error("bad index file sha1 signature");
1031 return 0;
1034 static int read_index_extension(struct index_state *istate,
1035 const char *ext, void *data, unsigned long sz)
1037 switch (CACHE_EXT(ext)) {
1038 case CACHE_EXT_TREE:
1039 istate->cache_tree = cache_tree_read(data, sz);
1040 break;
1041 default:
1042 if (*ext < 'A' || 'Z' < *ext)
1043 return error("index uses %.4s extension, which we do not understand",
1044 ext);
1045 fprintf(stderr, "ignoring %.4s extension\n", ext);
1046 break;
1048 return 0;
1051 int read_index(struct index_state *istate)
1053 return read_index_from(istate, get_index_file());
1056 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1058 size_t len;
1060 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1061 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1062 ce->ce_dev = ntohl(ondisk->dev);
1063 ce->ce_ino = ntohl(ondisk->ino);
1064 ce->ce_mode = ntohl(ondisk->mode);
1065 ce->ce_uid = ntohl(ondisk->uid);
1066 ce->ce_gid = ntohl(ondisk->gid);
1067 ce->ce_size = ntohl(ondisk->size);
1068 /* On-disk flags are just 16 bits */
1069 ce->ce_flags = ntohs(ondisk->flags);
1070 hashcpy(ce->sha1, ondisk->sha1);
1072 len = ce->ce_flags & CE_NAMEMASK;
1073 if (len == CE_NAMEMASK)
1074 len = strlen(ondisk->name);
1076 * NEEDSWORK: If the original index is crafted, this copy could
1077 * go unchecked.
1079 memcpy(ce->name, ondisk->name, len + 1);
1082 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1084 long per_entry;
1086 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1089 * Alignment can cause differences. This should be "alignof", but
1090 * since that's a gcc'ism, just use the size of a pointer.
1092 per_entry += sizeof(void *);
1093 return ondisk_size + entries*per_entry;
1096 /* remember to discard_cache() before reading a different cache! */
1097 int read_index_from(struct index_state *istate, const char *path)
1099 int fd, i;
1100 struct stat st;
1101 unsigned long src_offset, dst_offset;
1102 struct cache_header *hdr;
1103 void *mmap;
1104 size_t mmap_size;
1106 errno = EBUSY;
1107 if (istate->alloc)
1108 return istate->cache_nr;
1110 errno = ENOENT;
1111 istate->timestamp = 0;
1112 fd = open(path, O_RDONLY);
1113 if (fd < 0) {
1114 if (errno == ENOENT)
1115 return 0;
1116 die("index file open failed (%s)", strerror(errno));
1119 if (fstat(fd, &st))
1120 die("cannot stat the open index (%s)", strerror(errno));
1122 errno = EINVAL;
1123 mmap_size = xsize_t(st.st_size);
1124 if (mmap_size < sizeof(struct cache_header) + 20)
1125 die("index file smaller than expected");
1127 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1128 close(fd);
1129 if (mmap == MAP_FAILED)
1130 die("unable to map index file");
1132 hdr = mmap;
1133 if (verify_hdr(hdr, mmap_size) < 0)
1134 goto unmap;
1136 istate->cache_nr = ntohl(hdr->hdr_entries);
1137 istate->cache_alloc = alloc_nr(istate->cache_nr);
1138 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1141 * The disk format is actually larger than the in-memory format,
1142 * due to space for nsec etc, so even though the in-memory one
1143 * has room for a few more flags, we can allocate using the same
1144 * index size
1146 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1148 src_offset = sizeof(*hdr);
1149 dst_offset = 0;
1150 for (i = 0; i < istate->cache_nr; i++) {
1151 struct ondisk_cache_entry *disk_ce;
1152 struct cache_entry *ce;
1154 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1155 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1156 convert_from_disk(disk_ce, ce);
1157 set_index_entry(istate, i, ce);
1159 src_offset += ondisk_ce_size(ce);
1160 dst_offset += ce_size(ce);
1162 istate->timestamp = st.st_mtime;
1163 while (src_offset <= mmap_size - 20 - 8) {
1164 /* After an array of active_nr index entries,
1165 * there can be arbitrary number of extended
1166 * sections, each of which is prefixed with
1167 * extension name (4-byte) and section length
1168 * in 4-byte network byte order.
1170 unsigned long extsize;
1171 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1172 extsize = ntohl(extsize);
1173 if (read_index_extension(istate,
1174 (const char *) mmap + src_offset,
1175 (char *) mmap + src_offset + 8,
1176 extsize) < 0)
1177 goto unmap;
1178 src_offset += 8;
1179 src_offset += extsize;
1181 munmap(mmap, mmap_size);
1182 return istate->cache_nr;
1184 unmap:
1185 munmap(mmap, mmap_size);
1186 errno = EINVAL;
1187 die("index file corrupt");
1190 int discard_index(struct index_state *istate)
1192 istate->cache_nr = 0;
1193 istate->cache_changed = 0;
1194 istate->timestamp = 0;
1195 free_hash(&istate->name_hash);
1196 cache_tree_free(&(istate->cache_tree));
1197 free(istate->alloc);
1198 istate->alloc = NULL;
1200 /* no need to throw away allocated active_cache */
1201 return 0;
1204 int unmerged_index(const struct index_state *istate)
1206 int i;
1207 for (i = 0; i < istate->cache_nr; i++) {
1208 if (ce_stage(istate->cache[i]))
1209 return 1;
1211 return 0;
1214 #define WRITE_BUFFER_SIZE 8192
1215 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1216 static unsigned long write_buffer_len;
1218 static int ce_write_flush(SHA_CTX *context, int fd)
1220 unsigned int buffered = write_buffer_len;
1221 if (buffered) {
1222 SHA1_Update(context, write_buffer, buffered);
1223 if (write_in_full(fd, write_buffer, buffered) != buffered)
1224 return -1;
1225 write_buffer_len = 0;
1227 return 0;
1230 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1232 while (len) {
1233 unsigned int buffered = write_buffer_len;
1234 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1235 if (partial > len)
1236 partial = len;
1237 memcpy(write_buffer + buffered, data, partial);
1238 buffered += partial;
1239 if (buffered == WRITE_BUFFER_SIZE) {
1240 write_buffer_len = buffered;
1241 if (ce_write_flush(context, fd))
1242 return -1;
1243 buffered = 0;
1245 write_buffer_len = buffered;
1246 len -= partial;
1247 data = (char *) data + partial;
1249 return 0;
1252 static int write_index_ext_header(SHA_CTX *context, int fd,
1253 unsigned int ext, unsigned int sz)
1255 ext = htonl(ext);
1256 sz = htonl(sz);
1257 return ((ce_write(context, fd, &ext, 4) < 0) ||
1258 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1261 static int ce_flush(SHA_CTX *context, int fd)
1263 unsigned int left = write_buffer_len;
1265 if (left) {
1266 write_buffer_len = 0;
1267 SHA1_Update(context, write_buffer, left);
1270 /* Flush first if not enough space for SHA1 signature */
1271 if (left + 20 > WRITE_BUFFER_SIZE) {
1272 if (write_in_full(fd, write_buffer, left) != left)
1273 return -1;
1274 left = 0;
1277 /* Append the SHA1 signature at the end */
1278 SHA1_Final(write_buffer + left, context);
1279 left += 20;
1280 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1283 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1286 * The only thing we care about in this function is to smudge the
1287 * falsely clean entry due to touch-update-touch race, so we leave
1288 * everything else as they are. We are called for entries whose
1289 * ce_mtime match the index file mtime.
1291 struct stat st;
1293 if (lstat(ce->name, &st) < 0)
1294 return;
1295 if (ce_match_stat_basic(ce, &st))
1296 return;
1297 if (ce_modified_check_fs(ce, &st)) {
1298 /* This is "racily clean"; smudge it. Note that this
1299 * is a tricky code. At first glance, it may appear
1300 * that it can break with this sequence:
1302 * $ echo xyzzy >frotz
1303 * $ git-update-index --add frotz
1304 * $ : >frotz
1305 * $ sleep 3
1306 * $ echo filfre >nitfol
1307 * $ git-update-index --add nitfol
1309 * but it does not. When the second update-index runs,
1310 * it notices that the entry "frotz" has the same timestamp
1311 * as index, and if we were to smudge it by resetting its
1312 * size to zero here, then the object name recorded
1313 * in index is the 6-byte file but the cached stat information
1314 * becomes zero --- which would then match what we would
1315 * obtain from the filesystem next time we stat("frotz").
1317 * However, the second update-index, before calling
1318 * this function, notices that the cached size is 6
1319 * bytes and what is on the filesystem is an empty
1320 * file, and never calls us, so the cached size information
1321 * for "frotz" stays 6 which does not match the filesystem.
1323 ce->ce_size = 0;
1327 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1329 int size = ondisk_ce_size(ce);
1330 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1332 ondisk->ctime.sec = htonl(ce->ce_ctime);
1333 ondisk->ctime.nsec = 0;
1334 ondisk->mtime.sec = htonl(ce->ce_mtime);
1335 ondisk->mtime.nsec = 0;
1336 ondisk->dev = htonl(ce->ce_dev);
1337 ondisk->ino = htonl(ce->ce_ino);
1338 ondisk->mode = htonl(ce->ce_mode);
1339 ondisk->uid = htonl(ce->ce_uid);
1340 ondisk->gid = htonl(ce->ce_gid);
1341 ondisk->size = htonl(ce->ce_size);
1342 hashcpy(ondisk->sha1, ce->sha1);
1343 ondisk->flags = htons(ce->ce_flags);
1344 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1346 return ce_write(c, fd, ondisk, size);
1349 int write_index(const struct index_state *istate, int newfd)
1351 SHA_CTX c;
1352 struct cache_header hdr;
1353 int i, err, removed;
1354 struct cache_entry **cache = istate->cache;
1355 int entries = istate->cache_nr;
1357 for (i = removed = 0; i < entries; i++)
1358 if (cache[i]->ce_flags & CE_REMOVE)
1359 removed++;
1361 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1362 hdr.hdr_version = htonl(2);
1363 hdr.hdr_entries = htonl(entries - removed);
1365 SHA1_Init(&c);
1366 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1367 return -1;
1369 for (i = 0; i < entries; i++) {
1370 struct cache_entry *ce = cache[i];
1371 if (ce->ce_flags & CE_REMOVE)
1372 continue;
1373 if (is_racy_timestamp(istate, ce))
1374 ce_smudge_racily_clean_entry(ce);
1375 if (ce_write_entry(&c, newfd, ce) < 0)
1376 return -1;
1379 /* Write extension data here */
1380 if (istate->cache_tree) {
1381 struct strbuf sb;
1383 strbuf_init(&sb, 0);
1384 cache_tree_write(&sb, istate->cache_tree);
1385 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1386 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1387 strbuf_release(&sb);
1388 if (err)
1389 return -1;
1391 return ce_flush(&c, newfd);