builtin-add.c: restructure the code for maintainability
[git/gitweb-caching.git] / read-cache.c
blob6833af6cf1ab4628849d1d677ccc9aced28aab91
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"
11 #include "tree.h"
12 #include "commit.h"
13 #include "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
17 /* Index extensions.
19 * The first letter should be 'A'..'Z' for extensions that are not
20 * necessary for a correct operation (i.e. optimization data).
21 * When new extensions are added that _needs_ to be understood in
22 * order to correctly interpret the index file, pick character that
23 * is outside the range, to cause the reader to abort.
26 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
27 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
29 struct index_state the_index;
31 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
33 istate->cache[nr] = ce;
34 add_name_hash(istate, ce);
37 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
39 struct cache_entry *old = istate->cache[nr];
41 remove_name_hash(old);
42 set_index_entry(istate, nr, ce);
43 istate->cache_changed = 1;
47 * This only updates the "non-critical" parts of the directory
48 * cache, ie the parts that aren't tracked by GIT, and only used
49 * to validate the cache.
51 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
53 ce->ce_ctime = st->st_ctime;
54 ce->ce_mtime = st->st_mtime;
55 ce->ce_dev = st->st_dev;
56 ce->ce_ino = st->st_ino;
57 ce->ce_uid = st->st_uid;
58 ce->ce_gid = st->st_gid;
59 ce->ce_size = st->st_size;
61 if (assume_unchanged)
62 ce->ce_flags |= CE_VALID;
64 if (S_ISREG(st->st_mode))
65 ce_mark_uptodate(ce);
68 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
70 int match = -1;
71 int fd = open(ce->name, O_RDONLY);
73 if (fd >= 0) {
74 unsigned char sha1[20];
75 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
76 match = hashcmp(sha1, ce->sha1);
77 /* index_fd() closed the file descriptor already */
79 return match;
82 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
84 int match = -1;
85 char *target;
86 void *buffer;
87 unsigned long size;
88 enum object_type type;
89 int len;
91 target = xmalloc(expected_size);
92 len = readlink(ce->name, target, expected_size);
93 if (len != expected_size) {
94 free(target);
95 return -1;
97 buffer = read_sha1_file(ce->sha1, &type, &size);
98 if (!buffer) {
99 free(target);
100 return -1;
102 if (size == expected_size)
103 match = memcmp(buffer, target, size);
104 free(buffer);
105 free(target);
106 return match;
109 static int ce_compare_gitlink(struct cache_entry *ce)
111 unsigned char sha1[20];
114 * We don't actually require that the .git directory
115 * under GITLINK directory be a valid git directory. It
116 * might even be missing (in case nobody populated that
117 * sub-project).
119 * If so, we consider it always to match.
121 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
122 return 0;
123 return hashcmp(sha1, ce->sha1);
126 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
128 switch (st->st_mode & S_IFMT) {
129 case S_IFREG:
130 if (ce_compare_data(ce, st))
131 return DATA_CHANGED;
132 break;
133 case S_IFLNK:
134 if (ce_compare_link(ce, xsize_t(st->st_size)))
135 return DATA_CHANGED;
136 break;
137 case S_IFDIR:
138 if (S_ISGITLINK(ce->ce_mode))
139 return 0;
140 default:
141 return TYPE_CHANGED;
143 return 0;
146 static int is_empty_blob_sha1(const unsigned char *sha1)
148 static const unsigned char empty_blob_sha1[20] = {
149 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
150 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
153 return !hashcmp(sha1, empty_blob_sha1);
156 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
158 unsigned int changed = 0;
160 if (ce->ce_flags & CE_REMOVE)
161 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
163 switch (ce->ce_mode & S_IFMT) {
164 case S_IFREG:
165 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
166 /* We consider only the owner x bit to be relevant for
167 * "mode changes"
169 if (trust_executable_bit &&
170 (0100 & (ce->ce_mode ^ st->st_mode)))
171 changed |= MODE_CHANGED;
172 break;
173 case S_IFLNK:
174 if (!S_ISLNK(st->st_mode) &&
175 (has_symlinks || !S_ISREG(st->st_mode)))
176 changed |= TYPE_CHANGED;
177 break;
178 case S_IFGITLINK:
179 if (!S_ISDIR(st->st_mode))
180 changed |= TYPE_CHANGED;
181 else if (ce_compare_gitlink(ce))
182 changed |= DATA_CHANGED;
183 return changed;
184 default:
185 die("internal error: ce_mode is %o", ce->ce_mode);
187 if (ce->ce_mtime != (unsigned int) st->st_mtime)
188 changed |= MTIME_CHANGED;
189 if (ce->ce_ctime != (unsigned int) st->st_ctime)
190 changed |= CTIME_CHANGED;
192 if (ce->ce_uid != (unsigned int) st->st_uid ||
193 ce->ce_gid != (unsigned int) st->st_gid)
194 changed |= OWNER_CHANGED;
195 if (ce->ce_ino != (unsigned int) st->st_ino)
196 changed |= INODE_CHANGED;
198 #ifdef USE_STDEV
200 * st_dev breaks on network filesystems where different
201 * clients will have different views of what "device"
202 * the filesystem is on
204 if (ce->ce_dev != (unsigned int) st->st_dev)
205 changed |= INODE_CHANGED;
206 #endif
208 if (ce->ce_size != (unsigned int) st->st_size)
209 changed |= DATA_CHANGED;
211 /* Racily smudged entry? */
212 if (!ce->ce_size) {
213 if (!is_empty_blob_sha1(ce->sha1))
214 changed |= DATA_CHANGED;
217 return changed;
220 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
222 return (!S_ISGITLINK(ce->ce_mode) &&
223 istate->timestamp &&
224 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
227 int ie_match_stat(const struct index_state *istate,
228 struct cache_entry *ce, struct stat *st,
229 unsigned int options)
231 unsigned int changed;
232 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
233 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
236 * If it's marked as always valid in the index, it's
237 * valid whatever the checked-out copy says.
239 if (!ignore_valid && (ce->ce_flags & CE_VALID))
240 return 0;
242 changed = ce_match_stat_basic(ce, st);
245 * Within 1 second of this sequence:
246 * echo xyzzy >file && git-update-index --add file
247 * running this command:
248 * echo frotz >file
249 * would give a falsely clean cache entry. The mtime and
250 * length match the cache, and other stat fields do not change.
252 * We could detect this at update-index time (the cache entry
253 * being registered/updated records the same time as "now")
254 * and delay the return from git-update-index, but that would
255 * effectively mean we can make at most one commit per second,
256 * which is not acceptable. Instead, we check cache entries
257 * whose mtime are the same as the index file timestamp more
258 * carefully than others.
260 if (!changed && is_racy_timestamp(istate, ce)) {
261 if (assume_racy_is_modified)
262 changed |= DATA_CHANGED;
263 else
264 changed |= ce_modified_check_fs(ce, st);
267 return changed;
270 int ie_modified(const struct index_state *istate,
271 struct cache_entry *ce, struct stat *st, unsigned int options)
273 int changed, changed_fs;
275 changed = ie_match_stat(istate, ce, st, options);
276 if (!changed)
277 return 0;
279 * If the mode or type has changed, there's no point in trying
280 * to refresh the entry - it's not going to match
282 if (changed & (MODE_CHANGED | TYPE_CHANGED))
283 return changed;
285 /* Immediately after read-tree or update-index --cacheinfo,
286 * the length field is zero. For other cases the ce_size
287 * should match the SHA1 recorded in the index entry.
289 if ((changed & DATA_CHANGED) && ce->ce_size != 0)
290 return changed;
292 changed_fs = ce_modified_check_fs(ce, st);
293 if (changed_fs)
294 return changed | changed_fs;
295 return 0;
298 int base_name_compare(const char *name1, int len1, int mode1,
299 const char *name2, int len2, int mode2)
301 unsigned char c1, c2;
302 int len = len1 < len2 ? len1 : len2;
303 int cmp;
305 cmp = memcmp(name1, name2, len);
306 if (cmp)
307 return cmp;
308 c1 = name1[len];
309 c2 = name2[len];
310 if (!c1 && S_ISDIR(mode1))
311 c1 = '/';
312 if (!c2 && S_ISDIR(mode2))
313 c2 = '/';
314 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
318 * df_name_compare() is identical to base_name_compare(), except it
319 * compares conflicting directory/file entries as equal. Note that
320 * while a directory name compares as equal to a regular file, they
321 * then individually compare _differently_ to a filename that has
322 * a dot after the basename (because '\0' < '.' < '/').
324 * This is used by routines that want to traverse the git namespace
325 * but then handle conflicting entries together when possible.
327 int df_name_compare(const char *name1, int len1, int mode1,
328 const char *name2, int len2, int mode2)
330 int len = len1 < len2 ? len1 : len2, cmp;
331 unsigned char c1, c2;
333 cmp = memcmp(name1, name2, len);
334 if (cmp)
335 return cmp;
336 /* Directories and files compare equal (same length, same name) */
337 if (len1 == len2)
338 return 0;
339 c1 = name1[len];
340 if (!c1 && S_ISDIR(mode1))
341 c1 = '/';
342 c2 = name2[len];
343 if (!c2 && S_ISDIR(mode2))
344 c2 = '/';
345 if (c1 == '/' && !c2)
346 return 0;
347 if (c2 == '/' && !c1)
348 return 0;
349 return c1 - c2;
352 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
354 int len1 = flags1 & CE_NAMEMASK;
355 int len2 = flags2 & CE_NAMEMASK;
356 int len = len1 < len2 ? len1 : len2;
357 int cmp;
359 cmp = memcmp(name1, name2, len);
360 if (cmp)
361 return cmp;
362 if (len1 < len2)
363 return -1;
364 if (len1 > len2)
365 return 1;
367 /* Compare stages */
368 flags1 &= CE_STAGEMASK;
369 flags2 &= CE_STAGEMASK;
371 if (flags1 < flags2)
372 return -1;
373 if (flags1 > flags2)
374 return 1;
375 return 0;
378 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
380 int first, last;
382 first = 0;
383 last = istate->cache_nr;
384 while (last > first) {
385 int next = (last + first) >> 1;
386 struct cache_entry *ce = istate->cache[next];
387 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
388 if (!cmp)
389 return next;
390 if (cmp < 0) {
391 last = next;
392 continue;
394 first = next+1;
396 return -first-1;
399 /* Remove entry, return true if there are more entries to go.. */
400 int remove_index_entry_at(struct index_state *istate, int pos)
402 struct cache_entry *ce = istate->cache[pos];
404 remove_name_hash(ce);
405 istate->cache_changed = 1;
406 istate->cache_nr--;
407 if (pos >= istate->cache_nr)
408 return 0;
409 memmove(istate->cache + pos,
410 istate->cache + pos + 1,
411 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
412 return 1;
415 int remove_file_from_index(struct index_state *istate, const char *path)
417 int pos = index_name_pos(istate, path, strlen(path));
418 if (pos < 0)
419 pos = -pos-1;
420 cache_tree_invalidate_path(istate->cache_tree, path);
421 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
422 remove_index_entry_at(istate, pos);
423 return 0;
426 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
428 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
431 static int index_name_pos_also_unmerged(struct index_state *istate,
432 const char *path, int namelen)
434 int pos = index_name_pos(istate, path, namelen);
435 struct cache_entry *ce;
437 if (pos >= 0)
438 return pos;
440 /* maybe unmerged? */
441 pos = -1 - pos;
442 if (pos >= istate->cache_nr ||
443 compare_name((ce = istate->cache[pos]), path, namelen))
444 return -1;
446 /* order of preference: stage 2, 1, 3 */
447 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
448 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
449 !compare_name(ce, path, namelen))
450 pos++;
451 return pos;
454 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
456 int len = ce_namelen(ce);
457 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
461 * If we add a filename that aliases in the cache, we will use the
462 * name that we already have - but we don't want to update the same
463 * alias twice, because that implies that there were actually two
464 * different files with aliasing names!
466 * So we use the CE_ADDED flag to verify that the alias was an old
467 * one before we accept it as
469 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
471 int len;
472 struct cache_entry *new;
474 if (alias->ce_flags & CE_ADDED)
475 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
477 /* Ok, create the new entry using the name of the existing alias */
478 len = ce_namelen(alias);
479 new = xcalloc(1, cache_entry_size(len));
480 memcpy(new->name, alias->name, len);
481 copy_cache_entry(new, ce);
482 free(ce);
483 return new;
486 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
488 int size, namelen, was_same;
489 mode_t st_mode = st->st_mode;
490 struct cache_entry *ce, *alias;
491 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
492 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
493 int pretend = flags & ADD_CACHE_PRETEND;
495 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
496 return error("%s: can only add regular files, symbolic links or git-directories", path);
498 namelen = strlen(path);
499 if (S_ISDIR(st_mode)) {
500 while (namelen && path[namelen-1] == '/')
501 namelen--;
503 size = cache_entry_size(namelen);
504 ce = xcalloc(1, size);
505 memcpy(ce->name, path, namelen);
506 ce->ce_flags = namelen;
507 fill_stat_cache_info(ce, st);
509 if (trust_executable_bit && has_symlinks)
510 ce->ce_mode = create_ce_mode(st_mode);
511 else {
512 /* If there is an existing entry, pick the mode bits and type
513 * from it, otherwise assume unexecutable regular file.
515 struct cache_entry *ent;
516 int pos = index_name_pos_also_unmerged(istate, path, namelen);
518 ent = (0 <= pos) ? istate->cache[pos] : NULL;
519 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
522 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
523 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
524 /* Nothing changed, really */
525 free(ce);
526 ce_mark_uptodate(alias);
527 alias->ce_flags |= CE_ADDED;
528 return 0;
530 if (index_path(ce->sha1, path, st, 1))
531 return error("unable to index file %s", path);
532 if (ignore_case && alias && different_name(ce, alias))
533 ce = create_alias_ce(ce, alias);
534 ce->ce_flags |= CE_ADDED;
536 /* It was suspected to be racily clean, but it turns out to be Ok */
537 was_same = (alias &&
538 !ce_stage(alias) &&
539 !hashcmp(alias->sha1, ce->sha1) &&
540 ce->ce_mode == alias->ce_mode);
542 if (pretend)
544 else if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
545 return error("unable to add %s to index",path);
546 if (verbose && !was_same)
547 printf("add '%s'\n", path);
548 return 0;
551 int add_file_to_index(struct index_state *istate, const char *path, int flags)
553 struct stat st;
554 if (lstat(path, &st))
555 die("%s: unable to stat (%s)", path, strerror(errno));
556 return add_to_index(istate, path, &st, flags);
559 struct cache_entry *make_cache_entry(unsigned int mode,
560 const unsigned char *sha1, const char *path, int stage,
561 int refresh)
563 int size, len;
564 struct cache_entry *ce;
566 if (!verify_path(path))
567 return NULL;
569 len = strlen(path);
570 size = cache_entry_size(len);
571 ce = xcalloc(1, size);
573 hashcpy(ce->sha1, sha1);
574 memcpy(ce->name, path, len);
575 ce->ce_flags = create_ce_flags(len, stage);
576 ce->ce_mode = create_ce_mode(mode);
578 if (refresh)
579 return refresh_cache_entry(ce, 0);
581 return ce;
584 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
586 int len = ce_namelen(a);
587 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
590 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
592 const char *match, *name;
593 int len;
595 if (!pathspec)
596 return 1;
598 len = ce_namelen(ce);
599 name = ce->name;
600 while ((match = *pathspec++) != NULL) {
601 int matchlen = strlen(match);
602 if (matchlen > len)
603 continue;
604 if (memcmp(name, match, matchlen))
605 continue;
606 if (matchlen && name[matchlen-1] == '/')
607 return 1;
608 if (name[matchlen] == '/' || !name[matchlen])
609 return 1;
610 if (!matchlen)
611 return 1;
613 return 0;
617 * We fundamentally don't like some paths: we don't want
618 * dot or dot-dot anywhere, and for obvious reasons don't
619 * want to recurse into ".git" either.
621 * Also, we don't want double slashes or slashes at the
622 * end that can make pathnames ambiguous.
624 static int verify_dotfile(const char *rest)
627 * The first character was '.', but that
628 * has already been discarded, we now test
629 * the rest.
631 switch (*rest) {
632 /* "." is not allowed */
633 case '\0': case '/':
634 return 0;
637 * ".git" followed by NUL or slash is bad. This
638 * shares the path end test with the ".." case.
640 case 'g':
641 if (rest[1] != 'i')
642 break;
643 if (rest[2] != 't')
644 break;
645 rest += 2;
646 /* fallthrough */
647 case '.':
648 if (rest[1] == '\0' || rest[1] == '/')
649 return 0;
651 return 1;
654 int verify_path(const char *path)
656 char c;
658 goto inside;
659 for (;;) {
660 if (!c)
661 return 1;
662 if (c == '/') {
663 inside:
664 c = *path++;
665 switch (c) {
666 default:
667 continue;
668 case '/': case '\0':
669 break;
670 case '.':
671 if (verify_dotfile(path))
672 continue;
674 return 0;
676 c = *path++;
681 * Do we have another file that has the beginning components being a
682 * proper superset of the name we're trying to add?
684 static int has_file_name(struct index_state *istate,
685 const struct cache_entry *ce, int pos, int ok_to_replace)
687 int retval = 0;
688 int len = ce_namelen(ce);
689 int stage = ce_stage(ce);
690 const char *name = ce->name;
692 while (pos < istate->cache_nr) {
693 struct cache_entry *p = istate->cache[pos++];
695 if (len >= ce_namelen(p))
696 break;
697 if (memcmp(name, p->name, len))
698 break;
699 if (ce_stage(p) != stage)
700 continue;
701 if (p->name[len] != '/')
702 continue;
703 if (p->ce_flags & CE_REMOVE)
704 continue;
705 retval = -1;
706 if (!ok_to_replace)
707 break;
708 remove_index_entry_at(istate, --pos);
710 return retval;
714 * Do we have another file with a pathname that is a proper
715 * subset of the name we're trying to add?
717 static int has_dir_name(struct index_state *istate,
718 const struct cache_entry *ce, int pos, int ok_to_replace)
720 int retval = 0;
721 int stage = ce_stage(ce);
722 const char *name = ce->name;
723 const char *slash = name + ce_namelen(ce);
725 for (;;) {
726 int len;
728 for (;;) {
729 if (*--slash == '/')
730 break;
731 if (slash <= ce->name)
732 return retval;
734 len = slash - name;
736 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
737 if (pos >= 0) {
739 * Found one, but not so fast. This could
740 * be a marker that says "I was here, but
741 * I am being removed". Such an entry is
742 * not a part of the resulting tree, and
743 * it is Ok to have a directory at the same
744 * path.
746 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
747 retval = -1;
748 if (!ok_to_replace)
749 break;
750 remove_index_entry_at(istate, pos);
751 continue;
754 else
755 pos = -pos-1;
758 * Trivial optimization: if we find an entry that
759 * already matches the sub-directory, then we know
760 * we're ok, and we can exit.
762 while (pos < istate->cache_nr) {
763 struct cache_entry *p = istate->cache[pos];
764 if ((ce_namelen(p) <= len) ||
765 (p->name[len] != '/') ||
766 memcmp(p->name, name, len))
767 break; /* not our subdirectory */
768 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
770 * p is at the same stage as our entry, and
771 * is a subdirectory of what we are looking
772 * at, so we cannot have conflicts at our
773 * level or anything shorter.
775 return retval;
776 pos++;
779 return retval;
782 /* We may be in a situation where we already have path/file and path
783 * is being added, or we already have path and path/file is being
784 * added. Either one would result in a nonsense tree that has path
785 * twice when git-write-tree tries to write it out. Prevent it.
787 * If ok-to-replace is specified, we remove the conflicting entries
788 * from the cache so the caller should recompute the insert position.
789 * When this happens, we return non-zero.
791 static int check_file_directory_conflict(struct index_state *istate,
792 const struct cache_entry *ce,
793 int pos, int ok_to_replace)
795 int retval;
798 * When ce is an "I am going away" entry, we allow it to be added
800 if (ce->ce_flags & CE_REMOVE)
801 return 0;
804 * We check if the path is a sub-path of a subsequent pathname
805 * first, since removing those will not change the position
806 * in the array.
808 retval = has_file_name(istate, ce, pos, ok_to_replace);
811 * Then check if the path might have a clashing sub-directory
812 * before it.
814 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
817 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
819 int pos;
820 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
821 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
822 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
824 cache_tree_invalidate_path(istate->cache_tree, ce->name);
825 pos = index_name_pos(istate, ce->name, ce->ce_flags);
827 /* existing match? Just replace it. */
828 if (pos >= 0) {
829 replace_index_entry(istate, pos, ce);
830 return 0;
832 pos = -pos-1;
835 * Inserting a merged entry ("stage 0") into the index
836 * will always replace all non-merged entries..
838 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
839 while (ce_same_name(istate->cache[pos], ce)) {
840 ok_to_add = 1;
841 if (!remove_index_entry_at(istate, pos))
842 break;
846 if (!ok_to_add)
847 return -1;
848 if (!verify_path(ce->name))
849 return -1;
851 if (!skip_df_check &&
852 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
853 if (!ok_to_replace)
854 return error("'%s' appears as both a file and as a directory",
855 ce->name);
856 pos = index_name_pos(istate, ce->name, ce->ce_flags);
857 pos = -pos-1;
859 return pos + 1;
862 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
864 int pos;
866 if (option & ADD_CACHE_JUST_APPEND)
867 pos = istate->cache_nr;
868 else {
869 int ret;
870 ret = add_index_entry_with_check(istate, ce, option);
871 if (ret <= 0)
872 return ret;
873 pos = ret - 1;
876 /* Make sure the array is big enough .. */
877 if (istate->cache_nr == istate->cache_alloc) {
878 istate->cache_alloc = alloc_nr(istate->cache_alloc);
879 istate->cache = xrealloc(istate->cache,
880 istate->cache_alloc * sizeof(struct cache_entry *));
883 /* Add it in.. */
884 istate->cache_nr++;
885 if (istate->cache_nr > pos + 1)
886 memmove(istate->cache + pos + 1,
887 istate->cache + pos,
888 (istate->cache_nr - pos - 1) * sizeof(ce));
889 set_index_entry(istate, pos, ce);
890 istate->cache_changed = 1;
891 return 0;
895 * "refresh" does not calculate a new sha1 file or bring the
896 * cache up-to-date for mode/content changes. But what it
897 * _does_ do is to "re-match" the stat information of a file
898 * with the cache, so that you can refresh the cache for a
899 * file that hasn't been changed but where the stat entry is
900 * out of date.
902 * For example, you'd want to do this after doing a "git-read-tree",
903 * to link up the stat cache details with the proper files.
905 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
906 struct cache_entry *ce,
907 unsigned int options, int *err)
909 struct stat st;
910 struct cache_entry *updated;
911 int changed, size;
912 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
914 if (ce_uptodate(ce))
915 return ce;
918 * CE_VALID means the user promised us that the change to
919 * the work tree does not matter and told us not to worry.
921 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
922 ce_mark_uptodate(ce);
923 return ce;
926 if (lstat(ce->name, &st) < 0) {
927 if (err)
928 *err = errno;
929 return NULL;
932 changed = ie_match_stat(istate, ce, &st, options);
933 if (!changed) {
935 * The path is unchanged. If we were told to ignore
936 * valid bit, then we did the actual stat check and
937 * found that the entry is unmodified. If the entry
938 * is not marked VALID, this is the place to mark it
939 * valid again, under "assume unchanged" mode.
941 if (ignore_valid && assume_unchanged &&
942 !(ce->ce_flags & CE_VALID))
943 ; /* mark this one VALID again */
944 else {
946 * We do not mark the index itself "modified"
947 * because CE_UPTODATE flag is in-core only;
948 * we are not going to write this change out.
950 ce_mark_uptodate(ce);
951 return ce;
955 if (ie_modified(istate, ce, &st, options)) {
956 if (err)
957 *err = EINVAL;
958 return NULL;
961 size = ce_size(ce);
962 updated = xmalloc(size);
963 memcpy(updated, ce, size);
964 fill_stat_cache_info(updated, &st);
966 * If ignore_valid is not set, we should leave CE_VALID bit
967 * alone. Otherwise, paths marked with --no-assume-unchanged
968 * (i.e. things to be edited) will reacquire CE_VALID bit
969 * automatically, which is not really what we want.
971 if (!ignore_valid && assume_unchanged &&
972 !(ce->ce_flags & CE_VALID))
973 updated->ce_flags &= ~CE_VALID;
975 return updated;
978 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
980 int i;
981 int has_errors = 0;
982 int really = (flags & REFRESH_REALLY) != 0;
983 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
984 int quiet = (flags & REFRESH_QUIET) != 0;
985 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
986 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
987 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
988 const char *needs_update_message;
990 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
991 ? "locally modified" : "needs update");
992 for (i = 0; i < istate->cache_nr; i++) {
993 struct cache_entry *ce, *new;
994 int cache_errno = 0;
996 ce = istate->cache[i];
997 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
998 continue;
1000 if (ce_stage(ce)) {
1001 while ((i < istate->cache_nr) &&
1002 ! strcmp(istate->cache[i]->name, ce->name))
1003 i++;
1004 i--;
1005 if (allow_unmerged)
1006 continue;
1007 printf("%s: needs merge\n", ce->name);
1008 has_errors = 1;
1009 continue;
1012 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1013 continue;
1015 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1016 if (new == ce)
1017 continue;
1018 if (!new) {
1019 if (not_new && cache_errno == ENOENT)
1020 continue;
1021 if (really && cache_errno == EINVAL) {
1022 /* If we are doing --really-refresh that
1023 * means the index is not valid anymore.
1025 ce->ce_flags &= ~CE_VALID;
1026 istate->cache_changed = 1;
1028 if (quiet)
1029 continue;
1030 printf("%s: %s\n", ce->name, needs_update_message);
1031 has_errors = 1;
1032 continue;
1035 replace_index_entry(istate, i, new);
1037 return has_errors;
1040 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1042 return refresh_cache_ent(&the_index, ce, really, NULL);
1045 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1047 SHA_CTX c;
1048 unsigned char sha1[20];
1050 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1051 return error("bad signature");
1052 if (hdr->hdr_version != htonl(2))
1053 return error("bad index version");
1054 SHA1_Init(&c);
1055 SHA1_Update(&c, hdr, size - 20);
1056 SHA1_Final(sha1, &c);
1057 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1058 return error("bad index file sha1 signature");
1059 return 0;
1062 static int read_index_extension(struct index_state *istate,
1063 const char *ext, void *data, unsigned long sz)
1065 switch (CACHE_EXT(ext)) {
1066 case CACHE_EXT_TREE:
1067 istate->cache_tree = cache_tree_read(data, sz);
1068 break;
1069 default:
1070 if (*ext < 'A' || 'Z' < *ext)
1071 return error("index uses %.4s extension, which we do not understand",
1072 ext);
1073 fprintf(stderr, "ignoring %.4s extension\n", ext);
1074 break;
1076 return 0;
1079 int read_index(struct index_state *istate)
1081 return read_index_from(istate, get_index_file());
1084 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1086 size_t len;
1088 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1089 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1090 ce->ce_dev = ntohl(ondisk->dev);
1091 ce->ce_ino = ntohl(ondisk->ino);
1092 ce->ce_mode = ntohl(ondisk->mode);
1093 ce->ce_uid = ntohl(ondisk->uid);
1094 ce->ce_gid = ntohl(ondisk->gid);
1095 ce->ce_size = ntohl(ondisk->size);
1096 /* On-disk flags are just 16 bits */
1097 ce->ce_flags = ntohs(ondisk->flags);
1098 hashcpy(ce->sha1, ondisk->sha1);
1100 len = ce->ce_flags & CE_NAMEMASK;
1101 if (len == CE_NAMEMASK)
1102 len = strlen(ondisk->name);
1104 * NEEDSWORK: If the original index is crafted, this copy could
1105 * go unchecked.
1107 memcpy(ce->name, ondisk->name, len + 1);
1110 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1112 long per_entry;
1114 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1117 * Alignment can cause differences. This should be "alignof", but
1118 * since that's a gcc'ism, just use the size of a pointer.
1120 per_entry += sizeof(void *);
1121 return ondisk_size + entries*per_entry;
1124 /* remember to discard_cache() before reading a different cache! */
1125 int read_index_from(struct index_state *istate, const char *path)
1127 int fd, i;
1128 struct stat st;
1129 unsigned long src_offset, dst_offset;
1130 struct cache_header *hdr;
1131 void *mmap;
1132 size_t mmap_size;
1134 errno = EBUSY;
1135 if (istate->alloc)
1136 return istate->cache_nr;
1138 errno = ENOENT;
1139 istate->timestamp = 0;
1140 fd = open(path, O_RDONLY);
1141 if (fd < 0) {
1142 if (errno == ENOENT)
1143 return 0;
1144 die("index file open failed (%s)", strerror(errno));
1147 if (fstat(fd, &st))
1148 die("cannot stat the open index (%s)", strerror(errno));
1150 errno = EINVAL;
1151 mmap_size = xsize_t(st.st_size);
1152 if (mmap_size < sizeof(struct cache_header) + 20)
1153 die("index file smaller than expected");
1155 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1156 close(fd);
1157 if (mmap == MAP_FAILED)
1158 die("unable to map index file");
1160 hdr = mmap;
1161 if (verify_hdr(hdr, mmap_size) < 0)
1162 goto unmap;
1164 istate->cache_nr = ntohl(hdr->hdr_entries);
1165 istate->cache_alloc = alloc_nr(istate->cache_nr);
1166 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1169 * The disk format is actually larger than the in-memory format,
1170 * due to space for nsec etc, so even though the in-memory one
1171 * has room for a few more flags, we can allocate using the same
1172 * index size
1174 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1176 src_offset = sizeof(*hdr);
1177 dst_offset = 0;
1178 for (i = 0; i < istate->cache_nr; i++) {
1179 struct ondisk_cache_entry *disk_ce;
1180 struct cache_entry *ce;
1182 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1183 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1184 convert_from_disk(disk_ce, ce);
1185 set_index_entry(istate, i, ce);
1187 src_offset += ondisk_ce_size(ce);
1188 dst_offset += ce_size(ce);
1190 istate->timestamp = st.st_mtime;
1191 while (src_offset <= mmap_size - 20 - 8) {
1192 /* After an array of active_nr index entries,
1193 * there can be arbitrary number of extended
1194 * sections, each of which is prefixed with
1195 * extension name (4-byte) and section length
1196 * in 4-byte network byte order.
1198 unsigned long extsize;
1199 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1200 extsize = ntohl(extsize);
1201 if (read_index_extension(istate,
1202 (const char *) mmap + src_offset,
1203 (char *) mmap + src_offset + 8,
1204 extsize) < 0)
1205 goto unmap;
1206 src_offset += 8;
1207 src_offset += extsize;
1209 munmap(mmap, mmap_size);
1210 return istate->cache_nr;
1212 unmap:
1213 munmap(mmap, mmap_size);
1214 errno = EINVAL;
1215 die("index file corrupt");
1218 int discard_index(struct index_state *istate)
1220 istate->cache_nr = 0;
1221 istate->cache_changed = 0;
1222 istate->timestamp = 0;
1223 free_hash(&istate->name_hash);
1224 cache_tree_free(&(istate->cache_tree));
1225 free(istate->alloc);
1226 istate->alloc = NULL;
1228 /* no need to throw away allocated active_cache */
1229 return 0;
1232 int unmerged_index(const struct index_state *istate)
1234 int i;
1235 for (i = 0; i < istate->cache_nr; i++) {
1236 if (ce_stage(istate->cache[i]))
1237 return 1;
1239 return 0;
1242 #define WRITE_BUFFER_SIZE 8192
1243 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1244 static unsigned long write_buffer_len;
1246 static int ce_write_flush(SHA_CTX *context, int fd)
1248 unsigned int buffered = write_buffer_len;
1249 if (buffered) {
1250 SHA1_Update(context, write_buffer, buffered);
1251 if (write_in_full(fd, write_buffer, buffered) != buffered)
1252 return -1;
1253 write_buffer_len = 0;
1255 return 0;
1258 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1260 while (len) {
1261 unsigned int buffered = write_buffer_len;
1262 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1263 if (partial > len)
1264 partial = len;
1265 memcpy(write_buffer + buffered, data, partial);
1266 buffered += partial;
1267 if (buffered == WRITE_BUFFER_SIZE) {
1268 write_buffer_len = buffered;
1269 if (ce_write_flush(context, fd))
1270 return -1;
1271 buffered = 0;
1273 write_buffer_len = buffered;
1274 len -= partial;
1275 data = (char *) data + partial;
1277 return 0;
1280 static int write_index_ext_header(SHA_CTX *context, int fd,
1281 unsigned int ext, unsigned int sz)
1283 ext = htonl(ext);
1284 sz = htonl(sz);
1285 return ((ce_write(context, fd, &ext, 4) < 0) ||
1286 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1289 static int ce_flush(SHA_CTX *context, int fd)
1291 unsigned int left = write_buffer_len;
1293 if (left) {
1294 write_buffer_len = 0;
1295 SHA1_Update(context, write_buffer, left);
1298 /* Flush first if not enough space for SHA1 signature */
1299 if (left + 20 > WRITE_BUFFER_SIZE) {
1300 if (write_in_full(fd, write_buffer, left) != left)
1301 return -1;
1302 left = 0;
1305 /* Append the SHA1 signature at the end */
1306 SHA1_Final(write_buffer + left, context);
1307 left += 20;
1308 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1311 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1314 * The only thing we care about in this function is to smudge the
1315 * falsely clean entry due to touch-update-touch race, so we leave
1316 * everything else as they are. We are called for entries whose
1317 * ce_mtime match the index file mtime.
1319 struct stat st;
1321 if (lstat(ce->name, &st) < 0)
1322 return;
1323 if (ce_match_stat_basic(ce, &st))
1324 return;
1325 if (ce_modified_check_fs(ce, &st)) {
1326 /* This is "racily clean"; smudge it. Note that this
1327 * is a tricky code. At first glance, it may appear
1328 * that it can break with this sequence:
1330 * $ echo xyzzy >frotz
1331 * $ git-update-index --add frotz
1332 * $ : >frotz
1333 * $ sleep 3
1334 * $ echo filfre >nitfol
1335 * $ git-update-index --add nitfol
1337 * but it does not. When the second update-index runs,
1338 * it notices that the entry "frotz" has the same timestamp
1339 * as index, and if we were to smudge it by resetting its
1340 * size to zero here, then the object name recorded
1341 * in index is the 6-byte file but the cached stat information
1342 * becomes zero --- which would then match what we would
1343 * obtain from the filesystem next time we stat("frotz").
1345 * However, the second update-index, before calling
1346 * this function, notices that the cached size is 6
1347 * bytes and what is on the filesystem is an empty
1348 * file, and never calls us, so the cached size information
1349 * for "frotz" stays 6 which does not match the filesystem.
1351 ce->ce_size = 0;
1355 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1357 int size = ondisk_ce_size(ce);
1358 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1360 ondisk->ctime.sec = htonl(ce->ce_ctime);
1361 ondisk->ctime.nsec = 0;
1362 ondisk->mtime.sec = htonl(ce->ce_mtime);
1363 ondisk->mtime.nsec = 0;
1364 ondisk->dev = htonl(ce->ce_dev);
1365 ondisk->ino = htonl(ce->ce_ino);
1366 ondisk->mode = htonl(ce->ce_mode);
1367 ondisk->uid = htonl(ce->ce_uid);
1368 ondisk->gid = htonl(ce->ce_gid);
1369 ondisk->size = htonl(ce->ce_size);
1370 hashcpy(ondisk->sha1, ce->sha1);
1371 ondisk->flags = htons(ce->ce_flags);
1372 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1374 return ce_write(c, fd, ondisk, size);
1377 int write_index(const struct index_state *istate, int newfd)
1379 SHA_CTX c;
1380 struct cache_header hdr;
1381 int i, err, removed;
1382 struct cache_entry **cache = istate->cache;
1383 int entries = istate->cache_nr;
1385 for (i = removed = 0; i < entries; i++)
1386 if (cache[i]->ce_flags & CE_REMOVE)
1387 removed++;
1389 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1390 hdr.hdr_version = htonl(2);
1391 hdr.hdr_entries = htonl(entries - removed);
1393 SHA1_Init(&c);
1394 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1395 return -1;
1397 for (i = 0; i < entries; i++) {
1398 struct cache_entry *ce = cache[i];
1399 if (ce->ce_flags & CE_REMOVE)
1400 continue;
1401 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1402 ce_smudge_racily_clean_entry(ce);
1403 if (ce_write_entry(&c, newfd, ce) < 0)
1404 return -1;
1407 /* Write extension data here */
1408 if (istate->cache_tree) {
1409 struct strbuf sb;
1411 strbuf_init(&sb, 0);
1412 cache_tree_write(&sb, istate->cache_tree);
1413 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1414 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1415 strbuf_release(&sb);
1416 if (err)
1417 return -1;
1419 return ce_flush(&c, newfd);
1423 * Read the index file that is potentially unmerged into given
1424 * index_state, dropping any unmerged entries. Returns true is
1425 * the index is unmerged. Callers who want to refuse to work
1426 * from an unmerged state can call this and check its return value,
1427 * instead of calling read_cache().
1429 int read_index_unmerged(struct index_state *istate)
1431 int i;
1432 struct cache_entry **dst;
1433 struct cache_entry *last = NULL;
1435 read_index(istate);
1436 dst = istate->cache;
1437 for (i = 0; i < istate->cache_nr; i++) {
1438 struct cache_entry *ce = istate->cache[i];
1439 if (ce_stage(ce)) {
1440 remove_name_hash(ce);
1441 if (last && !strcmp(ce->name, last->name))
1442 continue;
1443 cache_tree_invalidate_path(istate->cache_tree, ce->name);
1444 last = ce;
1445 continue;
1447 *dst++ = ce;
1449 istate->cache_nr = dst - istate->cache;
1450 return !!last;
1453 struct update_callback_data
1455 int flags;
1456 int add_errors;
1459 static void update_callback(struct diff_queue_struct *q,
1460 struct diff_options *opt, void *cbdata)
1462 int i;
1463 struct update_callback_data *data = cbdata;
1465 for (i = 0; i < q->nr; i++) {
1466 struct diff_filepair *p = q->queue[i];
1467 const char *path = p->one->path;
1468 switch (p->status) {
1469 default:
1470 die("unexpected diff status %c", p->status);
1471 case DIFF_STATUS_UNMERGED:
1472 case DIFF_STATUS_MODIFIED:
1473 case DIFF_STATUS_TYPE_CHANGED:
1474 if (add_file_to_index(&the_index, path, data->flags)) {
1475 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1476 die("updating files failed");
1477 data->add_errors++;
1479 break;
1480 case DIFF_STATUS_DELETED:
1481 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1482 break;
1483 if (!(data->flags & ADD_CACHE_PRETEND))
1484 remove_file_from_index(&the_index, path);
1485 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1486 printf("remove '%s'\n", path);
1487 break;
1492 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1494 struct update_callback_data data;
1495 struct rev_info rev;
1496 init_revisions(&rev, prefix);
1497 setup_revisions(0, NULL, &rev, NULL);
1498 rev.prune_data = pathspec;
1499 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1500 rev.diffopt.format_callback = update_callback;
1501 data.flags = flags;
1502 data.add_errors = 0;
1503 rev.diffopt.format_callback_data = &data;
1504 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1505 return !!data.add_errors;