dir.h: move struct exclude declaration to top level
[git.git] / read-cache.c
blob6a45966ec499b47efbc76719046b53633e8839a1
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 "blob.h"
14 #include "resolve-undo.h"
15 #include "strbuf.h"
16 #include "varint.h"
18 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
19 unsigned int options);
21 /* Mask for the name length in ce_flags in the on-disk index */
23 #define CE_NAMEMASK (0x0fff)
25 /* Index extensions.
27 * The first letter should be 'A'..'Z' for extensions that are not
28 * necessary for a correct operation (i.e. optimization data).
29 * When new extensions are added that _needs_ to be understood in
30 * order to correctly interpret the index file, pick character that
31 * is outside the range, to cause the reader to abort.
34 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
35 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
36 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 struct index_state the_index;
40 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
42 istate->cache[nr] = ce;
43 add_name_hash(istate, ce);
46 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
48 struct cache_entry *old = istate->cache[nr];
50 remove_name_hash(istate, old);
51 free(old);
52 set_index_entry(istate, nr, ce);
53 istate->cache_changed = 1;
56 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
58 struct cache_entry *old = istate->cache[nr], *new;
59 int namelen = strlen(new_name);
61 new = xmalloc(cache_entry_size(namelen));
62 copy_cache_entry(new, old);
63 new->ce_flags &= ~CE_HASHED;
64 new->ce_namelen = namelen;
65 memcpy(new->name, new_name, namelen + 1);
67 cache_tree_invalidate_path(istate->cache_tree, old->name);
68 remove_index_entry_at(istate, nr);
69 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
72 void fill_stat_data(struct stat_data *sd, struct stat *st)
74 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
75 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
76 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
77 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
78 sd->sd_dev = st->st_dev;
79 sd->sd_ino = st->st_ino;
80 sd->sd_uid = st->st_uid;
81 sd->sd_gid = st->st_gid;
82 sd->sd_size = st->st_size;
85 int match_stat_data(const struct stat_data *sd, struct stat *st)
87 int changed = 0;
89 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
90 changed |= MTIME_CHANGED;
91 if (trust_ctime && check_stat &&
92 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
93 changed |= CTIME_CHANGED;
95 #ifdef USE_NSEC
96 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
97 changed |= MTIME_CHANGED;
98 if (trust_ctime && check_stat &&
99 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
100 changed |= CTIME_CHANGED;
101 #endif
103 if (check_stat) {
104 if (sd->sd_uid != (unsigned int) st->st_uid ||
105 sd->sd_gid != (unsigned int) st->st_gid)
106 changed |= OWNER_CHANGED;
107 if (sd->sd_ino != (unsigned int) st->st_ino)
108 changed |= INODE_CHANGED;
111 #ifdef USE_STDEV
113 * st_dev breaks on network filesystems where different
114 * clients will have different views of what "device"
115 * the filesystem is on
117 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
118 changed |= INODE_CHANGED;
119 #endif
121 if (sd->sd_size != (unsigned int) st->st_size)
122 changed |= DATA_CHANGED;
124 return changed;
128 * This only updates the "non-critical" parts of the directory
129 * cache, ie the parts that aren't tracked by GIT, and only used
130 * to validate the cache.
132 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
134 fill_stat_data(&ce->ce_stat_data, st);
136 if (assume_unchanged)
137 ce->ce_flags |= CE_VALID;
139 if (S_ISREG(st->st_mode))
140 ce_mark_uptodate(ce);
143 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
145 int match = -1;
146 int fd = open(ce->name, O_RDONLY);
148 if (fd >= 0) {
149 unsigned char sha1[20];
150 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
151 match = hashcmp(sha1, ce->sha1);
152 /* index_fd() closed the file descriptor already */
154 return match;
157 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
159 int match = -1;
160 void *buffer;
161 unsigned long size;
162 enum object_type type;
163 struct strbuf sb = STRBUF_INIT;
165 if (strbuf_readlink(&sb, ce->name, expected_size))
166 return -1;
168 buffer = read_sha1_file(ce->sha1, &type, &size);
169 if (buffer) {
170 if (size == sb.len)
171 match = memcmp(buffer, sb.buf, size);
172 free(buffer);
174 strbuf_release(&sb);
175 return match;
178 static int ce_compare_gitlink(const struct cache_entry *ce)
180 unsigned char sha1[20];
183 * We don't actually require that the .git directory
184 * under GITLINK directory be a valid git directory. It
185 * might even be missing (in case nobody populated that
186 * sub-project).
188 * If so, we consider it always to match.
190 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
191 return 0;
192 return hashcmp(sha1, ce->sha1);
195 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
197 switch (st->st_mode & S_IFMT) {
198 case S_IFREG:
199 if (ce_compare_data(ce, st))
200 return DATA_CHANGED;
201 break;
202 case S_IFLNK:
203 if (ce_compare_link(ce, xsize_t(st->st_size)))
204 return DATA_CHANGED;
205 break;
206 case S_IFDIR:
207 if (S_ISGITLINK(ce->ce_mode))
208 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
209 default:
210 return TYPE_CHANGED;
212 return 0;
215 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
217 unsigned int changed = 0;
219 if (ce->ce_flags & CE_REMOVE)
220 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
222 switch (ce->ce_mode & S_IFMT) {
223 case S_IFREG:
224 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
225 /* We consider only the owner x bit to be relevant for
226 * "mode changes"
228 if (trust_executable_bit &&
229 (0100 & (ce->ce_mode ^ st->st_mode)))
230 changed |= MODE_CHANGED;
231 break;
232 case S_IFLNK:
233 if (!S_ISLNK(st->st_mode) &&
234 (has_symlinks || !S_ISREG(st->st_mode)))
235 changed |= TYPE_CHANGED;
236 break;
237 case S_IFGITLINK:
238 /* We ignore most of the st_xxx fields for gitlinks */
239 if (!S_ISDIR(st->st_mode))
240 changed |= TYPE_CHANGED;
241 else if (ce_compare_gitlink(ce))
242 changed |= DATA_CHANGED;
243 return changed;
244 default:
245 die("internal error: ce_mode is %o", ce->ce_mode);
248 changed |= match_stat_data(&ce->ce_stat_data, st);
250 /* Racily smudged entry? */
251 if (!ce->ce_stat_data.sd_size) {
252 if (!is_empty_blob_sha1(ce->sha1))
253 changed |= DATA_CHANGED;
256 return changed;
259 static int is_racy_timestamp(const struct index_state *istate,
260 const struct cache_entry *ce)
262 return (!S_ISGITLINK(ce->ce_mode) &&
263 istate->timestamp.sec &&
264 #ifdef USE_NSEC
265 /* nanosecond timestamped files can also be racy! */
266 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
267 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
268 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
269 #else
270 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
271 #endif
275 int ie_match_stat(const struct index_state *istate,
276 const struct cache_entry *ce, struct stat *st,
277 unsigned int options)
279 unsigned int changed;
280 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
281 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
282 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
285 * If it's marked as always valid in the index, it's
286 * valid whatever the checked-out copy says.
288 * skip-worktree has the same effect with higher precedence
290 if (!ignore_skip_worktree && ce_skip_worktree(ce))
291 return 0;
292 if (!ignore_valid && (ce->ce_flags & CE_VALID))
293 return 0;
296 * Intent-to-add entries have not been added, so the index entry
297 * by definition never matches what is in the work tree until it
298 * actually gets added.
300 if (ce->ce_flags & CE_INTENT_TO_ADD)
301 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
303 changed = ce_match_stat_basic(ce, st);
306 * Within 1 second of this sequence:
307 * echo xyzzy >file && git-update-index --add file
308 * running this command:
309 * echo frotz >file
310 * would give a falsely clean cache entry. The mtime and
311 * length match the cache, and other stat fields do not change.
313 * We could detect this at update-index time (the cache entry
314 * being registered/updated records the same time as "now")
315 * and delay the return from git-update-index, but that would
316 * effectively mean we can make at most one commit per second,
317 * which is not acceptable. Instead, we check cache entries
318 * whose mtime are the same as the index file timestamp more
319 * carefully than others.
321 if (!changed && is_racy_timestamp(istate, ce)) {
322 if (assume_racy_is_modified)
323 changed |= DATA_CHANGED;
324 else
325 changed |= ce_modified_check_fs(ce, st);
328 return changed;
331 int ie_modified(const struct index_state *istate,
332 const struct cache_entry *ce,
333 struct stat *st, unsigned int options)
335 int changed, changed_fs;
337 changed = ie_match_stat(istate, ce, st, options);
338 if (!changed)
339 return 0;
341 * If the mode or type has changed, there's no point in trying
342 * to refresh the entry - it's not going to match
344 if (changed & (MODE_CHANGED | TYPE_CHANGED))
345 return changed;
348 * Immediately after read-tree or update-index --cacheinfo,
349 * the length field is zero, as we have never even read the
350 * lstat(2) information once, and we cannot trust DATA_CHANGED
351 * returned by ie_match_stat() which in turn was returned by
352 * ce_match_stat_basic() to signal that the filesize of the
353 * blob changed. We have to actually go to the filesystem to
354 * see if the contents match, and if so, should answer "unchanged".
356 * The logic does not apply to gitlinks, as ce_match_stat_basic()
357 * already has checked the actual HEAD from the filesystem in the
358 * subproject. If ie_match_stat() already said it is different,
359 * then we know it is.
361 if ((changed & DATA_CHANGED) &&
362 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
363 return changed;
365 changed_fs = ce_modified_check_fs(ce, st);
366 if (changed_fs)
367 return changed | changed_fs;
368 return 0;
371 int base_name_compare(const char *name1, int len1, int mode1,
372 const char *name2, int len2, int mode2)
374 unsigned char c1, c2;
375 int len = len1 < len2 ? len1 : len2;
376 int cmp;
378 cmp = memcmp(name1, name2, len);
379 if (cmp)
380 return cmp;
381 c1 = name1[len];
382 c2 = name2[len];
383 if (!c1 && S_ISDIR(mode1))
384 c1 = '/';
385 if (!c2 && S_ISDIR(mode2))
386 c2 = '/';
387 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
391 * df_name_compare() is identical to base_name_compare(), except it
392 * compares conflicting directory/file entries as equal. Note that
393 * while a directory name compares as equal to a regular file, they
394 * then individually compare _differently_ to a filename that has
395 * a dot after the basename (because '\0' < '.' < '/').
397 * This is used by routines that want to traverse the git namespace
398 * but then handle conflicting entries together when possible.
400 int df_name_compare(const char *name1, int len1, int mode1,
401 const char *name2, int len2, int mode2)
403 int len = len1 < len2 ? len1 : len2, cmp;
404 unsigned char c1, c2;
406 cmp = memcmp(name1, name2, len);
407 if (cmp)
408 return cmp;
409 /* Directories and files compare equal (same length, same name) */
410 if (len1 == len2)
411 return 0;
412 c1 = name1[len];
413 if (!c1 && S_ISDIR(mode1))
414 c1 = '/';
415 c2 = name2[len];
416 if (!c2 && S_ISDIR(mode2))
417 c2 = '/';
418 if (c1 == '/' && !c2)
419 return 0;
420 if (c2 == '/' && !c1)
421 return 0;
422 return c1 - c2;
425 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
427 size_t min_len = (len1 < len2) ? len1 : len2;
428 int cmp = memcmp(name1, name2, min_len);
429 if (cmp)
430 return cmp;
431 if (len1 < len2)
432 return -1;
433 if (len1 > len2)
434 return 1;
435 return 0;
438 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
440 int cmp;
442 cmp = name_compare(name1, len1, name2, len2);
443 if (cmp)
444 return cmp;
446 if (stage1 < stage2)
447 return -1;
448 if (stage1 > stage2)
449 return 1;
450 return 0;
453 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
455 int first, last;
457 first = 0;
458 last = istate->cache_nr;
459 while (last > first) {
460 int next = (last + first) >> 1;
461 struct cache_entry *ce = istate->cache[next];
462 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
463 if (!cmp)
464 return next;
465 if (cmp < 0) {
466 last = next;
467 continue;
469 first = next+1;
471 return -first-1;
474 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
476 return index_name_stage_pos(istate, name, namelen, 0);
479 /* Remove entry, return true if there are more entries to go.. */
480 int remove_index_entry_at(struct index_state *istate, int pos)
482 struct cache_entry *ce = istate->cache[pos];
484 record_resolve_undo(istate, ce);
485 remove_name_hash(istate, ce);
486 free(ce);
487 istate->cache_changed = 1;
488 istate->cache_nr--;
489 if (pos >= istate->cache_nr)
490 return 0;
491 memmove(istate->cache + pos,
492 istate->cache + pos + 1,
493 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
494 return 1;
498 * Remove all cache entries marked for removal, that is where
499 * CE_REMOVE is set in ce_flags. This is much more effective than
500 * calling remove_index_entry_at() for each entry to be removed.
502 void remove_marked_cache_entries(struct index_state *istate)
504 struct cache_entry **ce_array = istate->cache;
505 unsigned int i, j;
507 for (i = j = 0; i < istate->cache_nr; i++) {
508 if (ce_array[i]->ce_flags & CE_REMOVE) {
509 remove_name_hash(istate, ce_array[i]);
510 free(ce_array[i]);
512 else
513 ce_array[j++] = ce_array[i];
515 istate->cache_changed = 1;
516 istate->cache_nr = j;
519 int remove_file_from_index(struct index_state *istate, const char *path)
521 int pos = index_name_pos(istate, path, strlen(path));
522 if (pos < 0)
523 pos = -pos-1;
524 cache_tree_invalidate_path(istate->cache_tree, path);
525 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
526 remove_index_entry_at(istate, pos);
527 return 0;
530 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
532 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
535 static int index_name_pos_also_unmerged(struct index_state *istate,
536 const char *path, int namelen)
538 int pos = index_name_pos(istate, path, namelen);
539 struct cache_entry *ce;
541 if (pos >= 0)
542 return pos;
544 /* maybe unmerged? */
545 pos = -1 - pos;
546 if (pos >= istate->cache_nr ||
547 compare_name((ce = istate->cache[pos]), path, namelen))
548 return -1;
550 /* order of preference: stage 2, 1, 3 */
551 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
552 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
553 !compare_name(ce, path, namelen))
554 pos++;
555 return pos;
558 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
560 int len = ce_namelen(ce);
561 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
565 * If we add a filename that aliases in the cache, we will use the
566 * name that we already have - but we don't want to update the same
567 * alias twice, because that implies that there were actually two
568 * different files with aliasing names!
570 * So we use the CE_ADDED flag to verify that the alias was an old
571 * one before we accept it as
573 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
575 int len;
576 struct cache_entry *new;
578 if (alias->ce_flags & CE_ADDED)
579 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
581 /* Ok, create the new entry using the name of the existing alias */
582 len = ce_namelen(alias);
583 new = xcalloc(1, cache_entry_size(len));
584 memcpy(new->name, alias->name, len);
585 copy_cache_entry(new, ce);
586 free(ce);
587 return new;
590 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
592 unsigned char sha1[20];
593 if (write_sha1_file("", 0, blob_type, sha1))
594 die("cannot create an empty blob in the object database");
595 hashcpy(ce->sha1, sha1);
598 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
600 int size, namelen, was_same;
601 mode_t st_mode = st->st_mode;
602 struct cache_entry *ce, *alias;
603 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
604 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
605 int pretend = flags & ADD_CACHE_PRETEND;
606 int intent_only = flags & ADD_CACHE_INTENT;
607 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
608 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
610 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
611 return error("%s: can only add regular files, symbolic links or git-directories", path);
613 namelen = strlen(path);
614 if (S_ISDIR(st_mode)) {
615 while (namelen && path[namelen-1] == '/')
616 namelen--;
618 size = cache_entry_size(namelen);
619 ce = xcalloc(1, size);
620 memcpy(ce->name, path, namelen);
621 ce->ce_namelen = namelen;
622 if (!intent_only)
623 fill_stat_cache_info(ce, st);
624 else
625 ce->ce_flags |= CE_INTENT_TO_ADD;
627 if (trust_executable_bit && has_symlinks)
628 ce->ce_mode = create_ce_mode(st_mode);
629 else {
630 /* If there is an existing entry, pick the mode bits and type
631 * from it, otherwise assume unexecutable regular file.
633 struct cache_entry *ent;
634 int pos = index_name_pos_also_unmerged(istate, path, namelen);
636 ent = (0 <= pos) ? istate->cache[pos] : NULL;
637 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
640 /* When core.ignorecase=true, determine if a directory of the same name but differing
641 * case already exists within the Git repository. If it does, ensure the directory
642 * case of the file being added to the repository matches (is folded into) the existing
643 * entry's directory case.
645 if (ignore_case) {
646 const char *startPtr = ce->name;
647 const char *ptr = startPtr;
648 while (*ptr) {
649 while (*ptr && *ptr != '/')
650 ++ptr;
651 if (*ptr == '/') {
652 struct cache_entry *foundce;
653 ++ptr;
654 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
655 if (foundce) {
656 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
657 startPtr = ptr;
663 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
664 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
665 /* Nothing changed, really */
666 free(ce);
667 if (!S_ISGITLINK(alias->ce_mode))
668 ce_mark_uptodate(alias);
669 alias->ce_flags |= CE_ADDED;
670 return 0;
672 if (!intent_only) {
673 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
674 return error("unable to index file %s", path);
675 } else
676 set_object_name_for_intent_to_add_entry(ce);
678 if (ignore_case && alias && different_name(ce, alias))
679 ce = create_alias_ce(ce, alias);
680 ce->ce_flags |= CE_ADDED;
682 /* It was suspected to be racily clean, but it turns out to be Ok */
683 was_same = (alias &&
684 !ce_stage(alias) &&
685 !hashcmp(alias->sha1, ce->sha1) &&
686 ce->ce_mode == alias->ce_mode);
688 if (pretend)
690 else if (add_index_entry(istate, ce, add_option))
691 return error("unable to add %s to index",path);
692 if (verbose && !was_same)
693 printf("add '%s'\n", path);
694 return 0;
697 int add_file_to_index(struct index_state *istate, const char *path, int flags)
699 struct stat st;
700 if (lstat(path, &st))
701 die_errno("unable to stat '%s'", path);
702 return add_to_index(istate, path, &st, flags);
705 struct cache_entry *make_cache_entry(unsigned int mode,
706 const unsigned char *sha1, const char *path, int stage,
707 unsigned int refresh_options)
709 int size, len;
710 struct cache_entry *ce;
712 if (!verify_path(path)) {
713 error("Invalid path '%s'", path);
714 return NULL;
717 len = strlen(path);
718 size = cache_entry_size(len);
719 ce = xcalloc(1, size);
721 hashcpy(ce->sha1, sha1);
722 memcpy(ce->name, path, len);
723 ce->ce_flags = create_ce_flags(stage);
724 ce->ce_namelen = len;
725 ce->ce_mode = create_ce_mode(mode);
727 return refresh_cache_entry(ce, refresh_options);
730 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
732 int len = ce_namelen(a);
733 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
737 * We fundamentally don't like some paths: we don't want
738 * dot or dot-dot anywhere, and for obvious reasons don't
739 * want to recurse into ".git" either.
741 * Also, we don't want double slashes or slashes at the
742 * end that can make pathnames ambiguous.
744 static int verify_dotfile(const char *rest)
747 * The first character was '.', but that
748 * has already been discarded, we now test
749 * the rest.
752 /* "." is not allowed */
753 if (*rest == '\0' || is_dir_sep(*rest))
754 return 0;
756 switch (*rest) {
758 * ".git" followed by NUL or slash is bad. This
759 * shares the path end test with the ".." case.
761 case 'g':
762 if (rest[1] != 'i')
763 break;
764 if (rest[2] != 't')
765 break;
766 rest += 2;
767 /* fallthrough */
768 case '.':
769 if (rest[1] == '\0' || is_dir_sep(rest[1]))
770 return 0;
772 return 1;
775 int verify_path(const char *path)
777 char c;
779 if (has_dos_drive_prefix(path))
780 return 0;
782 goto inside;
783 for (;;) {
784 if (!c)
785 return 1;
786 if (is_dir_sep(c)) {
787 inside:
788 c = *path++;
789 if ((c == '.' && !verify_dotfile(path)) ||
790 is_dir_sep(c) || c == '\0')
791 return 0;
793 c = *path++;
798 * Do we have another file that has the beginning components being a
799 * proper superset of the name we're trying to add?
801 static int has_file_name(struct index_state *istate,
802 const struct cache_entry *ce, int pos, int ok_to_replace)
804 int retval = 0;
805 int len = ce_namelen(ce);
806 int stage = ce_stage(ce);
807 const char *name = ce->name;
809 while (pos < istate->cache_nr) {
810 struct cache_entry *p = istate->cache[pos++];
812 if (len >= ce_namelen(p))
813 break;
814 if (memcmp(name, p->name, len))
815 break;
816 if (ce_stage(p) != stage)
817 continue;
818 if (p->name[len] != '/')
819 continue;
820 if (p->ce_flags & CE_REMOVE)
821 continue;
822 retval = -1;
823 if (!ok_to_replace)
824 break;
825 remove_index_entry_at(istate, --pos);
827 return retval;
831 * Do we have another file with a pathname that is a proper
832 * subset of the name we're trying to add?
834 static int has_dir_name(struct index_state *istate,
835 const struct cache_entry *ce, int pos, int ok_to_replace)
837 int retval = 0;
838 int stage = ce_stage(ce);
839 const char *name = ce->name;
840 const char *slash = name + ce_namelen(ce);
842 for (;;) {
843 int len;
845 for (;;) {
846 if (*--slash == '/')
847 break;
848 if (slash <= ce->name)
849 return retval;
851 len = slash - name;
853 pos = index_name_stage_pos(istate, name, len, stage);
854 if (pos >= 0) {
856 * Found one, but not so fast. This could
857 * be a marker that says "I was here, but
858 * I am being removed". Such an entry is
859 * not a part of the resulting tree, and
860 * it is Ok to have a directory at the same
861 * path.
863 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
864 retval = -1;
865 if (!ok_to_replace)
866 break;
867 remove_index_entry_at(istate, pos);
868 continue;
871 else
872 pos = -pos-1;
875 * Trivial optimization: if we find an entry that
876 * already matches the sub-directory, then we know
877 * we're ok, and we can exit.
879 while (pos < istate->cache_nr) {
880 struct cache_entry *p = istate->cache[pos];
881 if ((ce_namelen(p) <= len) ||
882 (p->name[len] != '/') ||
883 memcmp(p->name, name, len))
884 break; /* not our subdirectory */
885 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
887 * p is at the same stage as our entry, and
888 * is a subdirectory of what we are looking
889 * at, so we cannot have conflicts at our
890 * level or anything shorter.
892 return retval;
893 pos++;
896 return retval;
899 /* We may be in a situation where we already have path/file and path
900 * is being added, or we already have path and path/file is being
901 * added. Either one would result in a nonsense tree that has path
902 * twice when git-write-tree tries to write it out. Prevent it.
904 * If ok-to-replace is specified, we remove the conflicting entries
905 * from the cache so the caller should recompute the insert position.
906 * When this happens, we return non-zero.
908 static int check_file_directory_conflict(struct index_state *istate,
909 const struct cache_entry *ce,
910 int pos, int ok_to_replace)
912 int retval;
915 * When ce is an "I am going away" entry, we allow it to be added
917 if (ce->ce_flags & CE_REMOVE)
918 return 0;
921 * We check if the path is a sub-path of a subsequent pathname
922 * first, since removing those will not change the position
923 * in the array.
925 retval = has_file_name(istate, ce, pos, ok_to_replace);
928 * Then check if the path might have a clashing sub-directory
929 * before it.
931 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
934 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
936 int pos;
937 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
938 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
939 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
940 int new_only = option & ADD_CACHE_NEW_ONLY;
942 cache_tree_invalidate_path(istate->cache_tree, ce->name);
943 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
945 /* existing match? Just replace it. */
946 if (pos >= 0) {
947 if (!new_only)
948 replace_index_entry(istate, pos, ce);
949 return 0;
951 pos = -pos-1;
954 * Inserting a merged entry ("stage 0") into the index
955 * will always replace all non-merged entries..
957 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
958 while (ce_same_name(istate->cache[pos], ce)) {
959 ok_to_add = 1;
960 if (!remove_index_entry_at(istate, pos))
961 break;
965 if (!ok_to_add)
966 return -1;
967 if (!verify_path(ce->name))
968 return error("Invalid path '%s'", ce->name);
970 if (!skip_df_check &&
971 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
972 if (!ok_to_replace)
973 return error("'%s' appears as both a file and as a directory",
974 ce->name);
975 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
976 pos = -pos-1;
978 return pos + 1;
981 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
983 int pos;
985 if (option & ADD_CACHE_JUST_APPEND)
986 pos = istate->cache_nr;
987 else {
988 int ret;
989 ret = add_index_entry_with_check(istate, ce, option);
990 if (ret <= 0)
991 return ret;
992 pos = ret - 1;
995 /* Make sure the array is big enough .. */
996 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
998 /* Add it in.. */
999 istate->cache_nr++;
1000 if (istate->cache_nr > pos + 1)
1001 memmove(istate->cache + pos + 1,
1002 istate->cache + pos,
1003 (istate->cache_nr - pos - 1) * sizeof(ce));
1004 set_index_entry(istate, pos, ce);
1005 istate->cache_changed = 1;
1006 return 0;
1010 * "refresh" does not calculate a new sha1 file or bring the
1011 * cache up-to-date for mode/content changes. But what it
1012 * _does_ do is to "re-match" the stat information of a file
1013 * with the cache, so that you can refresh the cache for a
1014 * file that hasn't been changed but where the stat entry is
1015 * out of date.
1017 * For example, you'd want to do this after doing a "git-read-tree",
1018 * to link up the stat cache details with the proper files.
1020 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1021 struct cache_entry *ce,
1022 unsigned int options, int *err,
1023 int *changed_ret)
1025 struct stat st;
1026 struct cache_entry *updated;
1027 int changed, size;
1028 int refresh = options & CE_MATCH_REFRESH;
1029 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1030 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1031 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1033 if (!refresh || ce_uptodate(ce))
1034 return ce;
1037 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1038 * that the change to the work tree does not matter and told
1039 * us not to worry.
1041 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1042 ce_mark_uptodate(ce);
1043 return ce;
1045 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1046 ce_mark_uptodate(ce);
1047 return ce;
1050 if (lstat(ce->name, &st) < 0) {
1051 if (ignore_missing && errno == ENOENT)
1052 return ce;
1053 if (err)
1054 *err = errno;
1055 return NULL;
1058 changed = ie_match_stat(istate, ce, &st, options);
1059 if (changed_ret)
1060 *changed_ret = changed;
1061 if (!changed) {
1063 * The path is unchanged. If we were told to ignore
1064 * valid bit, then we did the actual stat check and
1065 * found that the entry is unmodified. If the entry
1066 * is not marked VALID, this is the place to mark it
1067 * valid again, under "assume unchanged" mode.
1069 if (ignore_valid && assume_unchanged &&
1070 !(ce->ce_flags & CE_VALID))
1071 ; /* mark this one VALID again */
1072 else {
1074 * We do not mark the index itself "modified"
1075 * because CE_UPTODATE flag is in-core only;
1076 * we are not going to write this change out.
1078 if (!S_ISGITLINK(ce->ce_mode))
1079 ce_mark_uptodate(ce);
1080 return ce;
1084 if (ie_modified(istate, ce, &st, options)) {
1085 if (err)
1086 *err = EINVAL;
1087 return NULL;
1090 size = ce_size(ce);
1091 updated = xmalloc(size);
1092 memcpy(updated, ce, size);
1093 fill_stat_cache_info(updated, &st);
1095 * If ignore_valid is not set, we should leave CE_VALID bit
1096 * alone. Otherwise, paths marked with --no-assume-unchanged
1097 * (i.e. things to be edited) will reacquire CE_VALID bit
1098 * automatically, which is not really what we want.
1100 if (!ignore_valid && assume_unchanged &&
1101 !(ce->ce_flags & CE_VALID))
1102 updated->ce_flags &= ~CE_VALID;
1104 return updated;
1107 static void show_file(const char * fmt, const char * name, int in_porcelain,
1108 int * first, const char *header_msg)
1110 if (in_porcelain && *first && header_msg) {
1111 printf("%s\n", header_msg);
1112 *first = 0;
1114 printf(fmt, name);
1117 int refresh_index(struct index_state *istate, unsigned int flags,
1118 const struct pathspec *pathspec,
1119 char *seen, const char *header_msg)
1121 int i;
1122 int has_errors = 0;
1123 int really = (flags & REFRESH_REALLY) != 0;
1124 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1125 int quiet = (flags & REFRESH_QUIET) != 0;
1126 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1127 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1128 int first = 1;
1129 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1130 unsigned int options = (CE_MATCH_REFRESH |
1131 (really ? CE_MATCH_IGNORE_VALID : 0) |
1132 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1133 const char *modified_fmt;
1134 const char *deleted_fmt;
1135 const char *typechange_fmt;
1136 const char *added_fmt;
1137 const char *unmerged_fmt;
1139 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1140 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1141 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1142 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1143 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1144 for (i = 0; i < istate->cache_nr; i++) {
1145 struct cache_entry *ce, *new;
1146 int cache_errno = 0;
1147 int changed = 0;
1148 int filtered = 0;
1150 ce = istate->cache[i];
1151 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1152 continue;
1154 if (pathspec && !ce_path_match(ce, pathspec, seen))
1155 filtered = 1;
1157 if (ce_stage(ce)) {
1158 while ((i < istate->cache_nr) &&
1159 ! strcmp(istate->cache[i]->name, ce->name))
1160 i++;
1161 i--;
1162 if (allow_unmerged)
1163 continue;
1164 if (!filtered)
1165 show_file(unmerged_fmt, ce->name, in_porcelain,
1166 &first, header_msg);
1167 has_errors = 1;
1168 continue;
1171 if (filtered)
1172 continue;
1174 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1175 if (new == ce)
1176 continue;
1177 if (!new) {
1178 const char *fmt;
1180 if (really && cache_errno == EINVAL) {
1181 /* If we are doing --really-refresh that
1182 * means the index is not valid anymore.
1184 ce->ce_flags &= ~CE_VALID;
1185 istate->cache_changed = 1;
1187 if (quiet)
1188 continue;
1190 if (cache_errno == ENOENT)
1191 fmt = deleted_fmt;
1192 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1193 fmt = added_fmt; /* must be before other checks */
1194 else if (changed & TYPE_CHANGED)
1195 fmt = typechange_fmt;
1196 else
1197 fmt = modified_fmt;
1198 show_file(fmt,
1199 ce->name, in_porcelain, &first, header_msg);
1200 has_errors = 1;
1201 continue;
1204 replace_index_entry(istate, i, new);
1206 return has_errors;
1209 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1210 unsigned int options)
1212 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1216 /*****************************************************************
1217 * Index File I/O
1218 *****************************************************************/
1220 #define INDEX_FORMAT_DEFAULT 3
1222 static int index_format_config(const char *var, const char *value, void *cb)
1224 unsigned int *version = cb;
1225 if (!strcmp(var, "index.version")) {
1226 *version = git_config_int(var, value);
1227 return 0;
1229 return 1;
1232 static unsigned int get_index_format_default(void)
1234 char *envversion = getenv("GIT_INDEX_VERSION");
1235 char *endp;
1236 unsigned int version = INDEX_FORMAT_DEFAULT;
1238 if (!envversion) {
1239 git_config(index_format_config, &version);
1240 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1241 warning(_("index.version set, but the value is invalid.\n"
1242 "Using version %i"), INDEX_FORMAT_DEFAULT);
1243 return INDEX_FORMAT_DEFAULT;
1245 return version;
1248 version = strtoul(envversion, &endp, 10);
1249 if (*endp ||
1250 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1251 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1252 "Using version %i"), INDEX_FORMAT_DEFAULT);
1253 version = INDEX_FORMAT_DEFAULT;
1255 return version;
1259 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1260 * Again - this is just a (very strong in practice) heuristic that
1261 * the inode hasn't changed.
1263 * We save the fields in big-endian order to allow using the
1264 * index file over NFS transparently.
1266 struct ondisk_cache_entry {
1267 struct cache_time ctime;
1268 struct cache_time mtime;
1269 uint32_t dev;
1270 uint32_t ino;
1271 uint32_t mode;
1272 uint32_t uid;
1273 uint32_t gid;
1274 uint32_t size;
1275 unsigned char sha1[20];
1276 uint16_t flags;
1277 char name[FLEX_ARRAY]; /* more */
1281 * This struct is used when CE_EXTENDED bit is 1
1282 * The struct must match ondisk_cache_entry exactly from
1283 * ctime till flags
1285 struct ondisk_cache_entry_extended {
1286 struct cache_time ctime;
1287 struct cache_time mtime;
1288 uint32_t dev;
1289 uint32_t ino;
1290 uint32_t mode;
1291 uint32_t uid;
1292 uint32_t gid;
1293 uint32_t size;
1294 unsigned char sha1[20];
1295 uint16_t flags;
1296 uint16_t flags2;
1297 char name[FLEX_ARRAY]; /* more */
1300 /* These are only used for v3 or lower */
1301 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1302 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1303 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1304 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1305 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1306 ondisk_cache_entry_size(ce_namelen(ce)))
1308 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1310 git_SHA_CTX c;
1311 unsigned char sha1[20];
1312 int hdr_version;
1314 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1315 return error("bad signature");
1316 hdr_version = ntohl(hdr->hdr_version);
1317 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1318 return error("bad index version %d", hdr_version);
1319 git_SHA1_Init(&c);
1320 git_SHA1_Update(&c, hdr, size - 20);
1321 git_SHA1_Final(sha1, &c);
1322 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1323 return error("bad index file sha1 signature");
1324 return 0;
1327 static int read_index_extension(struct index_state *istate,
1328 const char *ext, void *data, unsigned long sz)
1330 switch (CACHE_EXT(ext)) {
1331 case CACHE_EXT_TREE:
1332 istate->cache_tree = cache_tree_read(data, sz);
1333 break;
1334 case CACHE_EXT_RESOLVE_UNDO:
1335 istate->resolve_undo = resolve_undo_read(data, sz);
1336 break;
1337 default:
1338 if (*ext < 'A' || 'Z' < *ext)
1339 return error("index uses %.4s extension, which we do not understand",
1340 ext);
1341 fprintf(stderr, "ignoring %.4s extension\n", ext);
1342 break;
1344 return 0;
1347 int read_index(struct index_state *istate)
1349 return read_index_from(istate, get_index_file());
1352 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1353 unsigned int flags,
1354 const char *name,
1355 size_t len)
1357 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1359 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1360 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1361 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1362 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1363 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1364 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1365 ce->ce_mode = get_be32(&ondisk->mode);
1366 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1367 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1368 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1369 ce->ce_flags = flags & ~CE_NAMEMASK;
1370 ce->ce_namelen = len;
1371 hashcpy(ce->sha1, ondisk->sha1);
1372 memcpy(ce->name, name, len);
1373 ce->name[len] = '\0';
1374 return ce;
1378 * Adjacent cache entries tend to share the leading paths, so it makes
1379 * sense to only store the differences in later entries. In the v4
1380 * on-disk format of the index, each on-disk cache entry stores the
1381 * number of bytes to be stripped from the end of the previous name,
1382 * and the bytes to append to the result, to come up with its name.
1384 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1386 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1387 size_t len = decode_varint(&cp);
1389 if (name->len < len)
1390 die("malformed name field in the index");
1391 strbuf_remove(name, name->len - len, len);
1392 for (ep = cp; *ep; ep++)
1393 ; /* find the end */
1394 strbuf_add(name, cp, ep - cp);
1395 return (const char *)ep + 1 - cp_;
1398 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1399 unsigned long *ent_size,
1400 struct strbuf *previous_name)
1402 struct cache_entry *ce;
1403 size_t len;
1404 const char *name;
1405 unsigned int flags;
1407 /* On-disk flags are just 16 bits */
1408 flags = get_be16(&ondisk->flags);
1409 len = flags & CE_NAMEMASK;
1411 if (flags & CE_EXTENDED) {
1412 struct ondisk_cache_entry_extended *ondisk2;
1413 int extended_flags;
1414 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1415 extended_flags = get_be16(&ondisk2->flags2) << 16;
1416 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1417 if (extended_flags & ~CE_EXTENDED_FLAGS)
1418 die("Unknown index entry format %08x", extended_flags);
1419 flags |= extended_flags;
1420 name = ondisk2->name;
1422 else
1423 name = ondisk->name;
1425 if (!previous_name) {
1426 /* v3 and earlier */
1427 if (len == CE_NAMEMASK)
1428 len = strlen(name);
1429 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1431 *ent_size = ondisk_ce_size(ce);
1432 } else {
1433 unsigned long consumed;
1434 consumed = expand_name_field(previous_name, name);
1435 ce = cache_entry_from_ondisk(ondisk, flags,
1436 previous_name->buf,
1437 previous_name->len);
1439 *ent_size = (name - ((char *)ondisk)) + consumed;
1441 return ce;
1444 /* remember to discard_cache() before reading a different cache! */
1445 int read_index_from(struct index_state *istate, const char *path)
1447 int fd, i;
1448 struct stat st;
1449 unsigned long src_offset;
1450 struct cache_header *hdr;
1451 void *mmap;
1452 size_t mmap_size;
1453 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1455 if (istate->initialized)
1456 return istate->cache_nr;
1458 istate->timestamp.sec = 0;
1459 istate->timestamp.nsec = 0;
1460 fd = open(path, O_RDONLY);
1461 if (fd < 0) {
1462 if (errno == ENOENT)
1463 return 0;
1464 die_errno("index file open failed");
1467 if (fstat(fd, &st))
1468 die_errno("cannot stat the open index");
1470 mmap_size = xsize_t(st.st_size);
1471 if (mmap_size < sizeof(struct cache_header) + 20)
1472 die("index file smaller than expected");
1474 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1475 if (mmap == MAP_FAILED)
1476 die_errno("unable to map index file");
1477 close(fd);
1479 hdr = mmap;
1480 if (verify_hdr(hdr, mmap_size) < 0)
1481 goto unmap;
1483 hashcpy(istate->sha1, (unsigned char *)hdr + mmap_size - 20);
1484 istate->version = ntohl(hdr->hdr_version);
1485 istate->cache_nr = ntohl(hdr->hdr_entries);
1486 istate->cache_alloc = alloc_nr(istate->cache_nr);
1487 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1488 istate->initialized = 1;
1490 if (istate->version == 4)
1491 previous_name = &previous_name_buf;
1492 else
1493 previous_name = NULL;
1495 src_offset = sizeof(*hdr);
1496 for (i = 0; i < istate->cache_nr; i++) {
1497 struct ondisk_cache_entry *disk_ce;
1498 struct cache_entry *ce;
1499 unsigned long consumed;
1501 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1502 ce = create_from_disk(disk_ce, &consumed, previous_name);
1503 set_index_entry(istate, i, ce);
1505 src_offset += consumed;
1507 strbuf_release(&previous_name_buf);
1508 istate->timestamp.sec = st.st_mtime;
1509 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1511 while (src_offset <= mmap_size - 20 - 8) {
1512 /* After an array of active_nr index entries,
1513 * there can be arbitrary number of extended
1514 * sections, each of which is prefixed with
1515 * extension name (4-byte) and section length
1516 * in 4-byte network byte order.
1518 uint32_t extsize;
1519 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1520 extsize = ntohl(extsize);
1521 if (read_index_extension(istate,
1522 (const char *) mmap + src_offset,
1523 (char *) mmap + src_offset + 8,
1524 extsize) < 0)
1525 goto unmap;
1526 src_offset += 8;
1527 src_offset += extsize;
1529 munmap(mmap, mmap_size);
1530 return istate->cache_nr;
1532 unmap:
1533 munmap(mmap, mmap_size);
1534 die("index file corrupt");
1537 int is_index_unborn(struct index_state *istate)
1539 return (!istate->cache_nr && !istate->timestamp.sec);
1542 int discard_index(struct index_state *istate)
1544 int i;
1546 for (i = 0; i < istate->cache_nr; i++)
1547 free(istate->cache[i]);
1548 resolve_undo_clear_index(istate);
1549 istate->cache_nr = 0;
1550 istate->cache_changed = 0;
1551 istate->timestamp.sec = 0;
1552 istate->timestamp.nsec = 0;
1553 free_name_hash(istate);
1554 cache_tree_free(&(istate->cache_tree));
1555 istate->initialized = 0;
1556 free(istate->cache);
1557 istate->cache = NULL;
1558 istate->cache_alloc = 0;
1559 return 0;
1562 int unmerged_index(const struct index_state *istate)
1564 int i;
1565 for (i = 0; i < istate->cache_nr; i++) {
1566 if (ce_stage(istate->cache[i]))
1567 return 1;
1569 return 0;
1572 #define WRITE_BUFFER_SIZE 8192
1573 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1574 static unsigned long write_buffer_len;
1576 static int ce_write_flush(git_SHA_CTX *context, int fd)
1578 unsigned int buffered = write_buffer_len;
1579 if (buffered) {
1580 git_SHA1_Update(context, write_buffer, buffered);
1581 if (write_in_full(fd, write_buffer, buffered) != buffered)
1582 return -1;
1583 write_buffer_len = 0;
1585 return 0;
1588 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1590 while (len) {
1591 unsigned int buffered = write_buffer_len;
1592 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1593 if (partial > len)
1594 partial = len;
1595 memcpy(write_buffer + buffered, data, partial);
1596 buffered += partial;
1597 if (buffered == WRITE_BUFFER_SIZE) {
1598 write_buffer_len = buffered;
1599 if (ce_write_flush(context, fd))
1600 return -1;
1601 buffered = 0;
1603 write_buffer_len = buffered;
1604 len -= partial;
1605 data = (char *) data + partial;
1607 return 0;
1610 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1611 unsigned int ext, unsigned int sz)
1613 ext = htonl(ext);
1614 sz = htonl(sz);
1615 return ((ce_write(context, fd, &ext, 4) < 0) ||
1616 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1619 static int ce_flush(git_SHA_CTX *context, int fd)
1621 unsigned int left = write_buffer_len;
1623 if (left) {
1624 write_buffer_len = 0;
1625 git_SHA1_Update(context, write_buffer, left);
1628 /* Flush first if not enough space for SHA1 signature */
1629 if (left + 20 > WRITE_BUFFER_SIZE) {
1630 if (write_in_full(fd, write_buffer, left) != left)
1631 return -1;
1632 left = 0;
1635 /* Append the SHA1 signature at the end */
1636 git_SHA1_Final(write_buffer + left, context);
1637 left += 20;
1638 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1641 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1644 * The only thing we care about in this function is to smudge the
1645 * falsely clean entry due to touch-update-touch race, so we leave
1646 * everything else as they are. We are called for entries whose
1647 * ce_stat_data.sd_mtime match the index file mtime.
1649 * Note that this actually does not do much for gitlinks, for
1650 * which ce_match_stat_basic() always goes to the actual
1651 * contents. The caller checks with is_racy_timestamp() which
1652 * always says "no" for gitlinks, so we are not called for them ;-)
1654 struct stat st;
1656 if (lstat(ce->name, &st) < 0)
1657 return;
1658 if (ce_match_stat_basic(ce, &st))
1659 return;
1660 if (ce_modified_check_fs(ce, &st)) {
1661 /* This is "racily clean"; smudge it. Note that this
1662 * is a tricky code. At first glance, it may appear
1663 * that it can break with this sequence:
1665 * $ echo xyzzy >frotz
1666 * $ git-update-index --add frotz
1667 * $ : >frotz
1668 * $ sleep 3
1669 * $ echo filfre >nitfol
1670 * $ git-update-index --add nitfol
1672 * but it does not. When the second update-index runs,
1673 * it notices that the entry "frotz" has the same timestamp
1674 * as index, and if we were to smudge it by resetting its
1675 * size to zero here, then the object name recorded
1676 * in index is the 6-byte file but the cached stat information
1677 * becomes zero --- which would then match what we would
1678 * obtain from the filesystem next time we stat("frotz").
1680 * However, the second update-index, before calling
1681 * this function, notices that the cached size is 6
1682 * bytes and what is on the filesystem is an empty
1683 * file, and never calls us, so the cached size information
1684 * for "frotz" stays 6 which does not match the filesystem.
1686 ce->ce_stat_data.sd_size = 0;
1690 /* Copy miscellaneous fields but not the name */
1691 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1692 struct cache_entry *ce)
1694 short flags;
1696 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1697 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1698 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1699 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1700 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1701 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1702 ondisk->mode = htonl(ce->ce_mode);
1703 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1704 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1705 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1706 hashcpy(ondisk->sha1, ce->sha1);
1708 flags = ce->ce_flags;
1709 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1710 ondisk->flags = htons(flags);
1711 if (ce->ce_flags & CE_EXTENDED) {
1712 struct ondisk_cache_entry_extended *ondisk2;
1713 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1714 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1715 return ondisk2->name;
1717 else {
1718 return ondisk->name;
1722 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1723 struct strbuf *previous_name)
1725 int size;
1726 struct ondisk_cache_entry *ondisk;
1727 char *name;
1728 int result;
1730 if (!previous_name) {
1731 size = ondisk_ce_size(ce);
1732 ondisk = xcalloc(1, size);
1733 name = copy_cache_entry_to_ondisk(ondisk, ce);
1734 memcpy(name, ce->name, ce_namelen(ce));
1735 } else {
1736 int common, to_remove, prefix_size;
1737 unsigned char to_remove_vi[16];
1738 for (common = 0;
1739 (ce->name[common] &&
1740 common < previous_name->len &&
1741 ce->name[common] == previous_name->buf[common]);
1742 common++)
1743 ; /* still matching */
1744 to_remove = previous_name->len - common;
1745 prefix_size = encode_varint(to_remove, to_remove_vi);
1747 if (ce->ce_flags & CE_EXTENDED)
1748 size = offsetof(struct ondisk_cache_entry_extended, name);
1749 else
1750 size = offsetof(struct ondisk_cache_entry, name);
1751 size += prefix_size + (ce_namelen(ce) - common + 1);
1753 ondisk = xcalloc(1, size);
1754 name = copy_cache_entry_to_ondisk(ondisk, ce);
1755 memcpy(name, to_remove_vi, prefix_size);
1756 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1758 strbuf_splice(previous_name, common, to_remove,
1759 ce->name + common, ce_namelen(ce) - common);
1762 result = ce_write(c, fd, ondisk, size);
1763 free(ondisk);
1764 return result;
1768 * This function verifies if index_state has the correct sha1 of the
1769 * index file. Don't die if we have any other failure, just return 0.
1771 static int verify_index_from(const struct index_state *istate, const char *path)
1773 int fd;
1774 ssize_t n;
1775 struct stat st;
1776 unsigned char sha1[20];
1778 if (!istate->initialized)
1779 return 0;
1781 fd = open(path, O_RDONLY);
1782 if (fd < 0)
1783 return 0;
1785 if (fstat(fd, &st))
1786 goto out;
1788 if (st.st_size < sizeof(struct cache_header) + 20)
1789 goto out;
1791 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1792 if (n != 20)
1793 goto out;
1795 if (hashcmp(istate->sha1, sha1))
1796 goto out;
1798 close(fd);
1799 return 1;
1801 out:
1802 close(fd);
1803 return 0;
1806 static int verify_index(const struct index_state *istate)
1808 return verify_index_from(istate, get_index_file());
1811 static int has_racy_timestamp(struct index_state *istate)
1813 int entries = istate->cache_nr;
1814 int i;
1816 for (i = 0; i < entries; i++) {
1817 struct cache_entry *ce = istate->cache[i];
1818 if (is_racy_timestamp(istate, ce))
1819 return 1;
1821 return 0;
1825 * Opportunistically update the index but do not complain if we can't
1827 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1829 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1830 verify_index(istate) && !write_index(istate, lockfile->fd))
1831 commit_locked_index(lockfile);
1832 else
1833 rollback_lock_file(lockfile);
1836 int write_index(struct index_state *istate, int newfd)
1838 git_SHA_CTX c;
1839 struct cache_header hdr;
1840 int i, err, removed, extended, hdr_version;
1841 struct cache_entry **cache = istate->cache;
1842 int entries = istate->cache_nr;
1843 struct stat st;
1844 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1846 for (i = removed = extended = 0; i < entries; i++) {
1847 if (cache[i]->ce_flags & CE_REMOVE)
1848 removed++;
1850 /* reduce extended entries if possible */
1851 cache[i]->ce_flags &= ~CE_EXTENDED;
1852 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1853 extended++;
1854 cache[i]->ce_flags |= CE_EXTENDED;
1858 if (!istate->version)
1859 istate->version = get_index_format_default();
1861 /* demote version 3 to version 2 when the latter suffices */
1862 if (istate->version == 3 || istate->version == 2)
1863 istate->version = extended ? 3 : 2;
1865 hdr_version = istate->version;
1867 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1868 hdr.hdr_version = htonl(hdr_version);
1869 hdr.hdr_entries = htonl(entries - removed);
1871 git_SHA1_Init(&c);
1872 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1873 return -1;
1875 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1876 for (i = 0; i < entries; i++) {
1877 struct cache_entry *ce = cache[i];
1878 if (ce->ce_flags & CE_REMOVE)
1879 continue;
1880 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1881 ce_smudge_racily_clean_entry(ce);
1882 if (is_null_sha1(ce->sha1)) {
1883 static const char msg[] = "cache entry has null sha1: %s";
1884 static int allow = -1;
1886 if (allow < 0)
1887 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1888 if (allow)
1889 warning(msg, ce->name);
1890 else
1891 return error(msg, ce->name);
1893 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1894 return -1;
1896 strbuf_release(&previous_name_buf);
1898 /* Write extension data here */
1899 if (istate->cache_tree) {
1900 struct strbuf sb = STRBUF_INIT;
1902 cache_tree_write(&sb, istate->cache_tree);
1903 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1904 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1905 strbuf_release(&sb);
1906 if (err)
1907 return -1;
1909 if (istate->resolve_undo) {
1910 struct strbuf sb = STRBUF_INIT;
1912 resolve_undo_write(&sb, istate->resolve_undo);
1913 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1914 sb.len) < 0
1915 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1916 strbuf_release(&sb);
1917 if (err)
1918 return -1;
1921 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1922 return -1;
1923 istate->timestamp.sec = (unsigned int)st.st_mtime;
1924 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1925 return 0;
1929 * Read the index file that is potentially unmerged into given
1930 * index_state, dropping any unmerged entries. Returns true if
1931 * the index is unmerged. Callers who want to refuse to work
1932 * from an unmerged state can call this and check its return value,
1933 * instead of calling read_cache().
1935 int read_index_unmerged(struct index_state *istate)
1937 int i;
1938 int unmerged = 0;
1940 read_index(istate);
1941 for (i = 0; i < istate->cache_nr; i++) {
1942 struct cache_entry *ce = istate->cache[i];
1943 struct cache_entry *new_ce;
1944 int size, len;
1946 if (!ce_stage(ce))
1947 continue;
1948 unmerged = 1;
1949 len = ce_namelen(ce);
1950 size = cache_entry_size(len);
1951 new_ce = xcalloc(1, size);
1952 memcpy(new_ce->name, ce->name, len);
1953 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1954 new_ce->ce_namelen = len;
1955 new_ce->ce_mode = ce->ce_mode;
1956 if (add_index_entry(istate, new_ce, 0))
1957 return error("%s: cannot drop to stage #0",
1958 new_ce->name);
1959 i = index_name_pos(istate, new_ce->name, len);
1961 return unmerged;
1965 * Returns 1 if the path is an "other" path with respect to
1966 * the index; that is, the path is not mentioned in the index at all,
1967 * either as a file, a directory with some files in the index,
1968 * or as an unmerged entry.
1970 * We helpfully remove a trailing "/" from directories so that
1971 * the output of read_directory can be used as-is.
1973 int index_name_is_other(const struct index_state *istate, const char *name,
1974 int namelen)
1976 int pos;
1977 if (namelen && name[namelen - 1] == '/')
1978 namelen--;
1979 pos = index_name_pos(istate, name, namelen);
1980 if (0 <= pos)
1981 return 0; /* exact match */
1982 pos = -pos - 1;
1983 if (pos < istate->cache_nr) {
1984 struct cache_entry *ce = istate->cache[pos];
1985 if (ce_namelen(ce) == namelen &&
1986 !memcmp(ce->name, name, namelen))
1987 return 0; /* Yup, this one exists unmerged */
1989 return 1;
1992 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
1994 int pos, len;
1995 unsigned long sz;
1996 enum object_type type;
1997 void *data;
1999 len = strlen(path);
2000 pos = index_name_pos(istate, path, len);
2001 if (pos < 0) {
2003 * We might be in the middle of a merge, in which
2004 * case we would read stage #2 (ours).
2006 int i;
2007 for (i = -pos - 1;
2008 (pos < 0 && i < istate->cache_nr &&
2009 !strcmp(istate->cache[i]->name, path));
2010 i++)
2011 if (ce_stage(istate->cache[i]) == 2)
2012 pos = i;
2014 if (pos < 0)
2015 return NULL;
2016 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2017 if (!data || type != OBJ_BLOB) {
2018 free(data);
2019 return NULL;
2021 if (size)
2022 *size = sz;
2023 return data;
2026 void stat_validity_clear(struct stat_validity *sv)
2028 free(sv->sd);
2029 sv->sd = NULL;
2032 int stat_validity_check(struct stat_validity *sv, const char *path)
2034 struct stat st;
2036 if (stat(path, &st) < 0)
2037 return sv->sd == NULL;
2038 if (!sv->sd)
2039 return 0;
2040 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2043 void stat_validity_update(struct stat_validity *sv, int fd)
2045 struct stat st;
2047 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2048 stat_validity_clear(sv);
2049 else {
2050 if (!sv->sd)
2051 sv->sd = xcalloc(1, sizeof(struct stat_data));
2052 fill_stat_data(sv->sd, &st);