Start preparing for 1.8.4.2
[alt-git.git] / read-cache.c
blobc3d5e3543fae75c81a7a0a00c48bb0076675a2f1
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, int really);
20 /* Mask for the name length in ce_flags in the on-disk index */
22 #define CE_NAMEMASK (0x0fff)
24 /* Index extensions.
26 * The first letter should be 'A'..'Z' for extensions that are not
27 * necessary for a correct operation (i.e. optimization data).
28 * When new extensions are added that _needs_ to be understood in
29 * order to correctly interpret the index file, pick character that
30 * is outside the range, to cause the reader to abort.
33 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
34 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
35 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
37 struct index_state the_index;
39 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
41 istate->cache[nr] = ce;
42 add_name_hash(istate, ce);
45 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
47 struct cache_entry *old = istate->cache[nr];
49 remove_name_hash(istate, old);
50 set_index_entry(istate, nr, ce);
51 istate->cache_changed = 1;
54 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
56 struct cache_entry *old = istate->cache[nr], *new;
57 int namelen = strlen(new_name);
59 new = xmalloc(cache_entry_size(namelen));
60 copy_cache_entry(new, old);
61 new->ce_flags &= ~CE_STATE_MASK;
62 new->ce_namelen = namelen;
63 memcpy(new->name, new_name, namelen + 1);
65 cache_tree_invalidate_path(istate->cache_tree, old->name);
66 remove_index_entry_at(istate, nr);
67 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
70 void fill_stat_data(struct stat_data *sd, struct stat *st)
72 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
73 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
74 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
75 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
76 sd->sd_dev = st->st_dev;
77 sd->sd_ino = st->st_ino;
78 sd->sd_uid = st->st_uid;
79 sd->sd_gid = st->st_gid;
80 sd->sd_size = st->st_size;
83 int match_stat_data(const struct stat_data *sd, struct stat *st)
85 int changed = 0;
87 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
88 changed |= MTIME_CHANGED;
89 if (trust_ctime && check_stat &&
90 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
91 changed |= CTIME_CHANGED;
93 #ifdef USE_NSEC
94 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
95 changed |= MTIME_CHANGED;
96 if (trust_ctime && check_stat &&
97 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
98 changed |= CTIME_CHANGED;
99 #endif
101 if (check_stat) {
102 if (sd->sd_uid != (unsigned int) st->st_uid ||
103 sd->sd_gid != (unsigned int) st->st_gid)
104 changed |= OWNER_CHANGED;
105 if (sd->sd_ino != (unsigned int) st->st_ino)
106 changed |= INODE_CHANGED;
109 #ifdef USE_STDEV
111 * st_dev breaks on network filesystems where different
112 * clients will have different views of what "device"
113 * the filesystem is on
115 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
116 changed |= INODE_CHANGED;
117 #endif
119 if (sd->sd_size != (unsigned int) st->st_size)
120 changed |= DATA_CHANGED;
122 return changed;
126 * This only updates the "non-critical" parts of the directory
127 * cache, ie the parts that aren't tracked by GIT, and only used
128 * to validate the cache.
130 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
132 fill_stat_data(&ce->ce_stat_data, st);
134 if (assume_unchanged)
135 ce->ce_flags |= CE_VALID;
137 if (S_ISREG(st->st_mode))
138 ce_mark_uptodate(ce);
141 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
143 int match = -1;
144 int fd = open(ce->name, O_RDONLY);
146 if (fd >= 0) {
147 unsigned char sha1[20];
148 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
149 match = hashcmp(sha1, ce->sha1);
150 /* index_fd() closed the file descriptor already */
152 return match;
155 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
157 int match = -1;
158 void *buffer;
159 unsigned long size;
160 enum object_type type;
161 struct strbuf sb = STRBUF_INIT;
163 if (strbuf_readlink(&sb, ce->name, expected_size))
164 return -1;
166 buffer = read_sha1_file(ce->sha1, &type, &size);
167 if (buffer) {
168 if (size == sb.len)
169 match = memcmp(buffer, sb.buf, size);
170 free(buffer);
172 strbuf_release(&sb);
173 return match;
176 static int ce_compare_gitlink(const struct cache_entry *ce)
178 unsigned char sha1[20];
181 * We don't actually require that the .git directory
182 * under GITLINK directory be a valid git directory. It
183 * might even be missing (in case nobody populated that
184 * sub-project).
186 * If so, we consider it always to match.
188 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
189 return 0;
190 return hashcmp(sha1, ce->sha1);
193 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
195 switch (st->st_mode & S_IFMT) {
196 case S_IFREG:
197 if (ce_compare_data(ce, st))
198 return DATA_CHANGED;
199 break;
200 case S_IFLNK:
201 if (ce_compare_link(ce, xsize_t(st->st_size)))
202 return DATA_CHANGED;
203 break;
204 case S_IFDIR:
205 if (S_ISGITLINK(ce->ce_mode))
206 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
207 default:
208 return TYPE_CHANGED;
210 return 0;
213 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
215 unsigned int changed = 0;
217 if (ce->ce_flags & CE_REMOVE)
218 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
220 switch (ce->ce_mode & S_IFMT) {
221 case S_IFREG:
222 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
223 /* We consider only the owner x bit to be relevant for
224 * "mode changes"
226 if (trust_executable_bit &&
227 (0100 & (ce->ce_mode ^ st->st_mode)))
228 changed |= MODE_CHANGED;
229 break;
230 case S_IFLNK:
231 if (!S_ISLNK(st->st_mode) &&
232 (has_symlinks || !S_ISREG(st->st_mode)))
233 changed |= TYPE_CHANGED;
234 break;
235 case S_IFGITLINK:
236 /* We ignore most of the st_xxx fields for gitlinks */
237 if (!S_ISDIR(st->st_mode))
238 changed |= TYPE_CHANGED;
239 else if (ce_compare_gitlink(ce))
240 changed |= DATA_CHANGED;
241 return changed;
242 default:
243 die("internal error: ce_mode is %o", ce->ce_mode);
246 changed |= match_stat_data(&ce->ce_stat_data, st);
248 /* Racily smudged entry? */
249 if (!ce->ce_stat_data.sd_size) {
250 if (!is_empty_blob_sha1(ce->sha1))
251 changed |= DATA_CHANGED;
254 return changed;
257 static int is_racy_timestamp(const struct index_state *istate,
258 const struct cache_entry *ce)
260 return (!S_ISGITLINK(ce->ce_mode) &&
261 istate->timestamp.sec &&
262 #ifdef USE_NSEC
263 /* nanosecond timestamped files can also be racy! */
264 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
265 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
266 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
267 #else
268 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
269 #endif
273 int ie_match_stat(const struct index_state *istate,
274 const struct cache_entry *ce, struct stat *st,
275 unsigned int options)
277 unsigned int changed;
278 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
279 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
280 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
283 * If it's marked as always valid in the index, it's
284 * valid whatever the checked-out copy says.
286 * skip-worktree has the same effect with higher precedence
288 if (!ignore_skip_worktree && ce_skip_worktree(ce))
289 return 0;
290 if (!ignore_valid && (ce->ce_flags & CE_VALID))
291 return 0;
294 * Intent-to-add entries have not been added, so the index entry
295 * by definition never matches what is in the work tree until it
296 * actually gets added.
298 if (ce->ce_flags & CE_INTENT_TO_ADD)
299 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
301 changed = ce_match_stat_basic(ce, st);
304 * Within 1 second of this sequence:
305 * echo xyzzy >file && git-update-index --add file
306 * running this command:
307 * echo frotz >file
308 * would give a falsely clean cache entry. The mtime and
309 * length match the cache, and other stat fields do not change.
311 * We could detect this at update-index time (the cache entry
312 * being registered/updated records the same time as "now")
313 * and delay the return from git-update-index, but that would
314 * effectively mean we can make at most one commit per second,
315 * which is not acceptable. Instead, we check cache entries
316 * whose mtime are the same as the index file timestamp more
317 * carefully than others.
319 if (!changed && is_racy_timestamp(istate, ce)) {
320 if (assume_racy_is_modified)
321 changed |= DATA_CHANGED;
322 else
323 changed |= ce_modified_check_fs(ce, st);
326 return changed;
329 int ie_modified(const struct index_state *istate,
330 const struct cache_entry *ce,
331 struct stat *st, unsigned int options)
333 int changed, changed_fs;
335 changed = ie_match_stat(istate, ce, st, options);
336 if (!changed)
337 return 0;
339 * If the mode or type has changed, there's no point in trying
340 * to refresh the entry - it's not going to match
342 if (changed & (MODE_CHANGED | TYPE_CHANGED))
343 return changed;
346 * Immediately after read-tree or update-index --cacheinfo,
347 * the length field is zero, as we have never even read the
348 * lstat(2) information once, and we cannot trust DATA_CHANGED
349 * returned by ie_match_stat() which in turn was returned by
350 * ce_match_stat_basic() to signal that the filesize of the
351 * blob changed. We have to actually go to the filesystem to
352 * see if the contents match, and if so, should answer "unchanged".
354 * The logic does not apply to gitlinks, as ce_match_stat_basic()
355 * already has checked the actual HEAD from the filesystem in the
356 * subproject. If ie_match_stat() already said it is different,
357 * then we know it is.
359 if ((changed & DATA_CHANGED) &&
360 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
361 return changed;
363 changed_fs = ce_modified_check_fs(ce, st);
364 if (changed_fs)
365 return changed | changed_fs;
366 return 0;
369 int base_name_compare(const char *name1, int len1, int mode1,
370 const char *name2, int len2, int mode2)
372 unsigned char c1, c2;
373 int len = len1 < len2 ? len1 : len2;
374 int cmp;
376 cmp = memcmp(name1, name2, len);
377 if (cmp)
378 return cmp;
379 c1 = name1[len];
380 c2 = name2[len];
381 if (!c1 && S_ISDIR(mode1))
382 c1 = '/';
383 if (!c2 && S_ISDIR(mode2))
384 c2 = '/';
385 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
389 * df_name_compare() is identical to base_name_compare(), except it
390 * compares conflicting directory/file entries as equal. Note that
391 * while a directory name compares as equal to a regular file, they
392 * then individually compare _differently_ to a filename that has
393 * a dot after the basename (because '\0' < '.' < '/').
395 * This is used by routines that want to traverse the git namespace
396 * but then handle conflicting entries together when possible.
398 int df_name_compare(const char *name1, int len1, int mode1,
399 const char *name2, int len2, int mode2)
401 int len = len1 < len2 ? len1 : len2, cmp;
402 unsigned char c1, c2;
404 cmp = memcmp(name1, name2, len);
405 if (cmp)
406 return cmp;
407 /* Directories and files compare equal (same length, same name) */
408 if (len1 == len2)
409 return 0;
410 c1 = name1[len];
411 if (!c1 && S_ISDIR(mode1))
412 c1 = '/';
413 c2 = name2[len];
414 if (!c2 && S_ISDIR(mode2))
415 c2 = '/';
416 if (c1 == '/' && !c2)
417 return 0;
418 if (c2 == '/' && !c1)
419 return 0;
420 return c1 - c2;
423 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
425 int len = len1 < len2 ? len1 : len2;
426 int cmp;
428 cmp = memcmp(name1, name2, len);
429 if (cmp)
430 return cmp;
431 if (len1 < len2)
432 return -1;
433 if (len1 > len2)
434 return 1;
436 if (stage1 < stage2)
437 return -1;
438 if (stage1 > stage2)
439 return 1;
440 return 0;
443 int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
445 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
448 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
450 int first, last;
452 first = 0;
453 last = istate->cache_nr;
454 while (last > first) {
455 int next = (last + first) >> 1;
456 struct cache_entry *ce = istate->cache[next];
457 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
458 if (!cmp)
459 return next;
460 if (cmp < 0) {
461 last = next;
462 continue;
464 first = next+1;
466 return -first-1;
469 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
471 return index_name_stage_pos(istate, name, namelen, 0);
474 /* Remove entry, return true if there are more entries to go.. */
475 int remove_index_entry_at(struct index_state *istate, int pos)
477 struct cache_entry *ce = istate->cache[pos];
479 record_resolve_undo(istate, ce);
480 remove_name_hash(istate, ce);
481 istate->cache_changed = 1;
482 istate->cache_nr--;
483 if (pos >= istate->cache_nr)
484 return 0;
485 memmove(istate->cache + pos,
486 istate->cache + pos + 1,
487 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
488 return 1;
492 * Remove all cache entries marked for removal, that is where
493 * CE_REMOVE is set in ce_flags. This is much more effective than
494 * calling remove_index_entry_at() for each entry to be removed.
496 void remove_marked_cache_entries(struct index_state *istate)
498 struct cache_entry **ce_array = istate->cache;
499 unsigned int i, j;
501 for (i = j = 0; i < istate->cache_nr; i++) {
502 if (ce_array[i]->ce_flags & CE_REMOVE)
503 remove_name_hash(istate, ce_array[i]);
504 else
505 ce_array[j++] = ce_array[i];
507 istate->cache_changed = 1;
508 istate->cache_nr = j;
511 int remove_file_from_index(struct index_state *istate, const char *path)
513 int pos = index_name_pos(istate, path, strlen(path));
514 if (pos < 0)
515 pos = -pos-1;
516 cache_tree_invalidate_path(istate->cache_tree, path);
517 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
518 remove_index_entry_at(istate, pos);
519 return 0;
522 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
524 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
527 static int index_name_pos_also_unmerged(struct index_state *istate,
528 const char *path, int namelen)
530 int pos = index_name_pos(istate, path, namelen);
531 struct cache_entry *ce;
533 if (pos >= 0)
534 return pos;
536 /* maybe unmerged? */
537 pos = -1 - pos;
538 if (pos >= istate->cache_nr ||
539 compare_name((ce = istate->cache[pos]), path, namelen))
540 return -1;
542 /* order of preference: stage 2, 1, 3 */
543 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
544 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
545 !compare_name(ce, path, namelen))
546 pos++;
547 return pos;
550 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
552 int len = ce_namelen(ce);
553 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
557 * If we add a filename that aliases in the cache, we will use the
558 * name that we already have - but we don't want to update the same
559 * alias twice, because that implies that there were actually two
560 * different files with aliasing names!
562 * So we use the CE_ADDED flag to verify that the alias was an old
563 * one before we accept it as
565 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
567 int len;
568 struct cache_entry *new;
570 if (alias->ce_flags & CE_ADDED)
571 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
573 /* Ok, create the new entry using the name of the existing alias */
574 len = ce_namelen(alias);
575 new = xcalloc(1, cache_entry_size(len));
576 memcpy(new->name, alias->name, len);
577 copy_cache_entry(new, ce);
578 free(ce);
579 return new;
582 static void record_intent_to_add(struct cache_entry *ce)
584 unsigned char sha1[20];
585 if (write_sha1_file("", 0, blob_type, sha1))
586 die("cannot create an empty blob in the object database");
587 hashcpy(ce->sha1, sha1);
590 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
592 int size, namelen, was_same;
593 mode_t st_mode = st->st_mode;
594 struct cache_entry *ce, *alias;
595 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
596 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
597 int pretend = flags & ADD_CACHE_PRETEND;
598 int intent_only = flags & ADD_CACHE_INTENT;
599 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
600 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
602 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
603 return error("%s: can only add regular files, symbolic links or git-directories", path);
605 namelen = strlen(path);
606 if (S_ISDIR(st_mode)) {
607 while (namelen && path[namelen-1] == '/')
608 namelen--;
610 size = cache_entry_size(namelen);
611 ce = xcalloc(1, size);
612 memcpy(ce->name, path, namelen);
613 ce->ce_namelen = namelen;
614 if (!intent_only)
615 fill_stat_cache_info(ce, st);
616 else
617 ce->ce_flags |= CE_INTENT_TO_ADD;
619 if (trust_executable_bit && has_symlinks)
620 ce->ce_mode = create_ce_mode(st_mode);
621 else {
622 /* If there is an existing entry, pick the mode bits and type
623 * from it, otherwise assume unexecutable regular file.
625 struct cache_entry *ent;
626 int pos = index_name_pos_also_unmerged(istate, path, namelen);
628 ent = (0 <= pos) ? istate->cache[pos] : NULL;
629 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
632 /* When core.ignorecase=true, determine if a directory of the same name but differing
633 * case already exists within the Git repository. If it does, ensure the directory
634 * case of the file being added to the repository matches (is folded into) the existing
635 * entry's directory case.
637 if (ignore_case) {
638 const char *startPtr = ce->name;
639 const char *ptr = startPtr;
640 while (*ptr) {
641 while (*ptr && *ptr != '/')
642 ++ptr;
643 if (*ptr == '/') {
644 struct cache_entry *foundce;
645 ++ptr;
646 foundce = index_name_exists(istate, ce->name, ptr - ce->name, ignore_case);
647 if (foundce) {
648 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
649 startPtr = ptr;
655 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
656 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
657 /* Nothing changed, really */
658 free(ce);
659 if (!S_ISGITLINK(alias->ce_mode))
660 ce_mark_uptodate(alias);
661 alias->ce_flags |= CE_ADDED;
662 return 0;
664 if (!intent_only) {
665 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
666 return error("unable to index file %s", path);
667 } else
668 record_intent_to_add(ce);
670 if (ignore_case && alias && different_name(ce, alias))
671 ce = create_alias_ce(ce, alias);
672 ce->ce_flags |= CE_ADDED;
674 /* It was suspected to be racily clean, but it turns out to be Ok */
675 was_same = (alias &&
676 !ce_stage(alias) &&
677 !hashcmp(alias->sha1, ce->sha1) &&
678 ce->ce_mode == alias->ce_mode);
680 if (pretend)
682 else if (add_index_entry(istate, ce, add_option))
683 return error("unable to add %s to index",path);
684 if (verbose && !was_same)
685 printf("add '%s'\n", path);
686 return 0;
689 int add_file_to_index(struct index_state *istate, const char *path, int flags)
691 struct stat st;
692 if (lstat(path, &st))
693 die_errno("unable to stat '%s'", path);
694 return add_to_index(istate, path, &st, flags);
697 struct cache_entry *make_cache_entry(unsigned int mode,
698 const unsigned char *sha1, const char *path, int stage,
699 int refresh)
701 int size, len;
702 struct cache_entry *ce;
704 if (!verify_path(path)) {
705 error("Invalid path '%s'", path);
706 return NULL;
709 len = strlen(path);
710 size = cache_entry_size(len);
711 ce = xcalloc(1, size);
713 hashcpy(ce->sha1, sha1);
714 memcpy(ce->name, path, len);
715 ce->ce_flags = create_ce_flags(stage);
716 ce->ce_namelen = len;
717 ce->ce_mode = create_ce_mode(mode);
719 if (refresh)
720 return refresh_cache_entry(ce, 0);
722 return ce;
725 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
727 int len = ce_namelen(a);
728 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
731 int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec)
733 return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
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 if (istate->cache_nr == istate->cache_alloc) {
997 istate->cache_alloc = alloc_nr(istate->cache_alloc);
998 istate->cache = xrealloc(istate->cache,
999 istate->cache_alloc * sizeof(*istate->cache));
1002 /* Add it in.. */
1003 istate->cache_nr++;
1004 if (istate->cache_nr > pos + 1)
1005 memmove(istate->cache + pos + 1,
1006 istate->cache + pos,
1007 (istate->cache_nr - pos - 1) * sizeof(ce));
1008 set_index_entry(istate, pos, ce);
1009 istate->cache_changed = 1;
1010 return 0;
1014 * "refresh" does not calculate a new sha1 file or bring the
1015 * cache up-to-date for mode/content changes. But what it
1016 * _does_ do is to "re-match" the stat information of a file
1017 * with the cache, so that you can refresh the cache for a
1018 * file that hasn't been changed but where the stat entry is
1019 * out of date.
1021 * For example, you'd want to do this after doing a "git-read-tree",
1022 * to link up the stat cache details with the proper files.
1024 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1025 struct cache_entry *ce,
1026 unsigned int options, int *err,
1027 int *changed_ret)
1029 struct stat st;
1030 struct cache_entry *updated;
1031 int changed, size;
1032 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1033 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1035 if (ce_uptodate(ce))
1036 return ce;
1039 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1040 * that the change to the work tree does not matter and told
1041 * us not to worry.
1043 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1044 ce_mark_uptodate(ce);
1045 return ce;
1047 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1048 ce_mark_uptodate(ce);
1049 return ce;
1052 if (lstat(ce->name, &st) < 0) {
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, const char **pathspec,
1118 char *seen, const char *header_msg)
1120 int i;
1121 int has_errors = 0;
1122 int really = (flags & REFRESH_REALLY) != 0;
1123 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1124 int quiet = (flags & REFRESH_QUIET) != 0;
1125 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1126 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1127 int first = 1;
1128 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1129 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1130 const char *modified_fmt;
1131 const char *deleted_fmt;
1132 const char *typechange_fmt;
1133 const char *added_fmt;
1134 const char *unmerged_fmt;
1136 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1137 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1138 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1139 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1140 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1141 for (i = 0; i < istate->cache_nr; i++) {
1142 struct cache_entry *ce, *new;
1143 int cache_errno = 0;
1144 int changed = 0;
1145 int filtered = 0;
1147 ce = istate->cache[i];
1148 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1149 continue;
1151 if (pathspec &&
1152 !match_pathspec(pathspec, ce->name, ce_namelen(ce), 0, seen))
1153 filtered = 1;
1155 if (ce_stage(ce)) {
1156 while ((i < istate->cache_nr) &&
1157 ! strcmp(istate->cache[i]->name, ce->name))
1158 i++;
1159 i--;
1160 if (allow_unmerged)
1161 continue;
1162 if (!filtered)
1163 show_file(unmerged_fmt, ce->name, in_porcelain,
1164 &first, header_msg);
1165 has_errors = 1;
1166 continue;
1169 if (filtered)
1170 continue;
1172 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1173 if (new == ce)
1174 continue;
1175 if (!new) {
1176 const char *fmt;
1178 if (not_new && cache_errno == ENOENT)
1179 continue;
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, int really)
1211 return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
1215 /*****************************************************************
1216 * Index File I/O
1217 *****************************************************************/
1219 #define INDEX_FORMAT_DEFAULT 3
1222 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1223 * Again - this is just a (very strong in practice) heuristic that
1224 * the inode hasn't changed.
1226 * We save the fields in big-endian order to allow using the
1227 * index file over NFS transparently.
1229 struct ondisk_cache_entry {
1230 struct cache_time ctime;
1231 struct cache_time mtime;
1232 unsigned int dev;
1233 unsigned int ino;
1234 unsigned int mode;
1235 unsigned int uid;
1236 unsigned int gid;
1237 unsigned int size;
1238 unsigned char sha1[20];
1239 unsigned short flags;
1240 char name[FLEX_ARRAY]; /* more */
1244 * This struct is used when CE_EXTENDED bit is 1
1245 * The struct must match ondisk_cache_entry exactly from
1246 * ctime till flags
1248 struct ondisk_cache_entry_extended {
1249 struct cache_time ctime;
1250 struct cache_time mtime;
1251 unsigned int dev;
1252 unsigned int ino;
1253 unsigned int mode;
1254 unsigned int uid;
1255 unsigned int gid;
1256 unsigned int size;
1257 unsigned char sha1[20];
1258 unsigned short flags;
1259 unsigned short flags2;
1260 char name[FLEX_ARRAY]; /* more */
1263 /* These are only used for v3 or lower */
1264 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1265 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1266 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1267 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1268 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1269 ondisk_cache_entry_size(ce_namelen(ce)))
1271 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1273 git_SHA_CTX c;
1274 unsigned char sha1[20];
1275 int hdr_version;
1277 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1278 return error("bad signature");
1279 hdr_version = ntohl(hdr->hdr_version);
1280 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1281 return error("bad index version %d", hdr_version);
1282 git_SHA1_Init(&c);
1283 git_SHA1_Update(&c, hdr, size - 20);
1284 git_SHA1_Final(sha1, &c);
1285 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1286 return error("bad index file sha1 signature");
1287 return 0;
1290 static int read_index_extension(struct index_state *istate,
1291 const char *ext, void *data, unsigned long sz)
1293 switch (CACHE_EXT(ext)) {
1294 case CACHE_EXT_TREE:
1295 istate->cache_tree = cache_tree_read(data, sz);
1296 break;
1297 case CACHE_EXT_RESOLVE_UNDO:
1298 istate->resolve_undo = resolve_undo_read(data, sz);
1299 break;
1300 default:
1301 if (*ext < 'A' || 'Z' < *ext)
1302 return error("index uses %.4s extension, which we do not understand",
1303 ext);
1304 fprintf(stderr, "ignoring %.4s extension\n", ext);
1305 break;
1307 return 0;
1310 int read_index(struct index_state *istate)
1312 return read_index_from(istate, get_index_file());
1315 #ifndef NEEDS_ALIGNED_ACCESS
1316 #define ntoh_s(var) ntohs(var)
1317 #define ntoh_l(var) ntohl(var)
1318 #else
1319 static inline uint16_t ntoh_s_force_align(void *p)
1321 uint16_t x;
1322 memcpy(&x, p, sizeof(x));
1323 return ntohs(x);
1325 static inline uint32_t ntoh_l_force_align(void *p)
1327 uint32_t x;
1328 memcpy(&x, p, sizeof(x));
1329 return ntohl(x);
1331 #define ntoh_s(var) ntoh_s_force_align(&(var))
1332 #define ntoh_l(var) ntoh_l_force_align(&(var))
1333 #endif
1335 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1336 unsigned int flags,
1337 const char *name,
1338 size_t len)
1340 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1342 ce->ce_stat_data.sd_ctime.sec = ntoh_l(ondisk->ctime.sec);
1343 ce->ce_stat_data.sd_mtime.sec = ntoh_l(ondisk->mtime.sec);
1344 ce->ce_stat_data.sd_ctime.nsec = ntoh_l(ondisk->ctime.nsec);
1345 ce->ce_stat_data.sd_mtime.nsec = ntoh_l(ondisk->mtime.nsec);
1346 ce->ce_stat_data.sd_dev = ntoh_l(ondisk->dev);
1347 ce->ce_stat_data.sd_ino = ntoh_l(ondisk->ino);
1348 ce->ce_mode = ntoh_l(ondisk->mode);
1349 ce->ce_stat_data.sd_uid = ntoh_l(ondisk->uid);
1350 ce->ce_stat_data.sd_gid = ntoh_l(ondisk->gid);
1351 ce->ce_stat_data.sd_size = ntoh_l(ondisk->size);
1352 ce->ce_flags = flags & ~CE_NAMEMASK;
1353 ce->ce_namelen = len;
1354 hashcpy(ce->sha1, ondisk->sha1);
1355 memcpy(ce->name, name, len);
1356 ce->name[len] = '\0';
1357 return ce;
1361 * Adjacent cache entries tend to share the leading paths, so it makes
1362 * sense to only store the differences in later entries. In the v4
1363 * on-disk format of the index, each on-disk cache entry stores the
1364 * number of bytes to be stripped from the end of the previous name,
1365 * and the bytes to append to the result, to come up with its name.
1367 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1369 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1370 size_t len = decode_varint(&cp);
1372 if (name->len < len)
1373 die("malformed name field in the index");
1374 strbuf_remove(name, name->len - len, len);
1375 for (ep = cp; *ep; ep++)
1376 ; /* find the end */
1377 strbuf_add(name, cp, ep - cp);
1378 return (const char *)ep + 1 - cp_;
1381 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1382 unsigned long *ent_size,
1383 struct strbuf *previous_name)
1385 struct cache_entry *ce;
1386 size_t len;
1387 const char *name;
1388 unsigned int flags;
1390 /* On-disk flags are just 16 bits */
1391 flags = ntoh_s(ondisk->flags);
1392 len = flags & CE_NAMEMASK;
1394 if (flags & CE_EXTENDED) {
1395 struct ondisk_cache_entry_extended *ondisk2;
1396 int extended_flags;
1397 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1398 extended_flags = ntoh_s(ondisk2->flags2) << 16;
1399 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1400 if (extended_flags & ~CE_EXTENDED_FLAGS)
1401 die("Unknown index entry format %08x", extended_flags);
1402 flags |= extended_flags;
1403 name = ondisk2->name;
1405 else
1406 name = ondisk->name;
1408 if (!previous_name) {
1409 /* v3 and earlier */
1410 if (len == CE_NAMEMASK)
1411 len = strlen(name);
1412 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1414 *ent_size = ondisk_ce_size(ce);
1415 } else {
1416 unsigned long consumed;
1417 consumed = expand_name_field(previous_name, name);
1418 ce = cache_entry_from_ondisk(ondisk, flags,
1419 previous_name->buf,
1420 previous_name->len);
1422 *ent_size = (name - ((char *)ondisk)) + consumed;
1424 return ce;
1427 /* remember to discard_cache() before reading a different cache! */
1428 int read_index_from(struct index_state *istate, const char *path)
1430 int fd, i;
1431 struct stat st;
1432 unsigned long src_offset;
1433 struct cache_header *hdr;
1434 void *mmap;
1435 size_t mmap_size;
1436 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1438 if (istate->initialized)
1439 return istate->cache_nr;
1441 istate->timestamp.sec = 0;
1442 istate->timestamp.nsec = 0;
1443 fd = open(path, O_RDONLY);
1444 if (fd < 0) {
1445 if (errno == ENOENT)
1446 return 0;
1447 die_errno("index file open failed");
1450 if (fstat(fd, &st))
1451 die_errno("cannot stat the open index");
1453 mmap_size = xsize_t(st.st_size);
1454 if (mmap_size < sizeof(struct cache_header) + 20)
1455 die("index file smaller than expected");
1457 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1458 if (mmap == MAP_FAILED)
1459 die_errno("unable to map index file");
1460 close(fd);
1462 hdr = mmap;
1463 if (verify_hdr(hdr, mmap_size) < 0)
1464 goto unmap;
1466 istate->version = ntohl(hdr->hdr_version);
1467 istate->cache_nr = ntohl(hdr->hdr_entries);
1468 istate->cache_alloc = alloc_nr(istate->cache_nr);
1469 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1470 istate->initialized = 1;
1472 if (istate->version == 4)
1473 previous_name = &previous_name_buf;
1474 else
1475 previous_name = NULL;
1477 src_offset = sizeof(*hdr);
1478 for (i = 0; i < istate->cache_nr; i++) {
1479 struct ondisk_cache_entry *disk_ce;
1480 struct cache_entry *ce;
1481 unsigned long consumed;
1483 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1484 ce = create_from_disk(disk_ce, &consumed, previous_name);
1485 set_index_entry(istate, i, ce);
1487 src_offset += consumed;
1489 strbuf_release(&previous_name_buf);
1490 istate->timestamp.sec = st.st_mtime;
1491 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1493 while (src_offset <= mmap_size - 20 - 8) {
1494 /* After an array of active_nr index entries,
1495 * there can be arbitrary number of extended
1496 * sections, each of which is prefixed with
1497 * extension name (4-byte) and section length
1498 * in 4-byte network byte order.
1500 uint32_t extsize;
1501 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1502 extsize = ntohl(extsize);
1503 if (read_index_extension(istate,
1504 (const char *) mmap + src_offset,
1505 (char *) mmap + src_offset + 8,
1506 extsize) < 0)
1507 goto unmap;
1508 src_offset += 8;
1509 src_offset += extsize;
1511 munmap(mmap, mmap_size);
1512 return istate->cache_nr;
1514 unmap:
1515 munmap(mmap, mmap_size);
1516 die("index file corrupt");
1519 int is_index_unborn(struct index_state *istate)
1521 return (!istate->cache_nr && !istate->timestamp.sec);
1524 int discard_index(struct index_state *istate)
1526 int i;
1528 for (i = 0; i < istate->cache_nr; i++)
1529 free(istate->cache[i]);
1530 resolve_undo_clear_index(istate);
1531 istate->cache_nr = 0;
1532 istate->cache_changed = 0;
1533 istate->timestamp.sec = 0;
1534 istate->timestamp.nsec = 0;
1535 free_name_hash(istate);
1536 cache_tree_free(&(istate->cache_tree));
1537 istate->initialized = 0;
1538 free(istate->cache);
1539 istate->cache = NULL;
1540 istate->cache_alloc = 0;
1541 return 0;
1544 int unmerged_index(const struct index_state *istate)
1546 int i;
1547 for (i = 0; i < istate->cache_nr; i++) {
1548 if (ce_stage(istate->cache[i]))
1549 return 1;
1551 return 0;
1554 #define WRITE_BUFFER_SIZE 8192
1555 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1556 static unsigned long write_buffer_len;
1558 static int ce_write_flush(git_SHA_CTX *context, int fd)
1560 unsigned int buffered = write_buffer_len;
1561 if (buffered) {
1562 git_SHA1_Update(context, write_buffer, buffered);
1563 if (write_in_full(fd, write_buffer, buffered) != buffered)
1564 return -1;
1565 write_buffer_len = 0;
1567 return 0;
1570 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1572 while (len) {
1573 unsigned int buffered = write_buffer_len;
1574 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1575 if (partial > len)
1576 partial = len;
1577 memcpy(write_buffer + buffered, data, partial);
1578 buffered += partial;
1579 if (buffered == WRITE_BUFFER_SIZE) {
1580 write_buffer_len = buffered;
1581 if (ce_write_flush(context, fd))
1582 return -1;
1583 buffered = 0;
1585 write_buffer_len = buffered;
1586 len -= partial;
1587 data = (char *) data + partial;
1589 return 0;
1592 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1593 unsigned int ext, unsigned int sz)
1595 ext = htonl(ext);
1596 sz = htonl(sz);
1597 return ((ce_write(context, fd, &ext, 4) < 0) ||
1598 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1601 static int ce_flush(git_SHA_CTX *context, int fd)
1603 unsigned int left = write_buffer_len;
1605 if (left) {
1606 write_buffer_len = 0;
1607 git_SHA1_Update(context, write_buffer, left);
1610 /* Flush first if not enough space for SHA1 signature */
1611 if (left + 20 > WRITE_BUFFER_SIZE) {
1612 if (write_in_full(fd, write_buffer, left) != left)
1613 return -1;
1614 left = 0;
1617 /* Append the SHA1 signature at the end */
1618 git_SHA1_Final(write_buffer + left, context);
1619 left += 20;
1620 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1623 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1626 * The only thing we care about in this function is to smudge the
1627 * falsely clean entry due to touch-update-touch race, so we leave
1628 * everything else as they are. We are called for entries whose
1629 * ce_stat_data.sd_mtime match the index file mtime.
1631 * Note that this actually does not do much for gitlinks, for
1632 * which ce_match_stat_basic() always goes to the actual
1633 * contents. The caller checks with is_racy_timestamp() which
1634 * always says "no" for gitlinks, so we are not called for them ;-)
1636 struct stat st;
1638 if (lstat(ce->name, &st) < 0)
1639 return;
1640 if (ce_match_stat_basic(ce, &st))
1641 return;
1642 if (ce_modified_check_fs(ce, &st)) {
1643 /* This is "racily clean"; smudge it. Note that this
1644 * is a tricky code. At first glance, it may appear
1645 * that it can break with this sequence:
1647 * $ echo xyzzy >frotz
1648 * $ git-update-index --add frotz
1649 * $ : >frotz
1650 * $ sleep 3
1651 * $ echo filfre >nitfol
1652 * $ git-update-index --add nitfol
1654 * but it does not. When the second update-index runs,
1655 * it notices that the entry "frotz" has the same timestamp
1656 * as index, and if we were to smudge it by resetting its
1657 * size to zero here, then the object name recorded
1658 * in index is the 6-byte file but the cached stat information
1659 * becomes zero --- which would then match what we would
1660 * obtain from the filesystem next time we stat("frotz").
1662 * However, the second update-index, before calling
1663 * this function, notices that the cached size is 6
1664 * bytes and what is on the filesystem is an empty
1665 * file, and never calls us, so the cached size information
1666 * for "frotz" stays 6 which does not match the filesystem.
1668 ce->ce_stat_data.sd_size = 0;
1672 /* Copy miscellaneous fields but not the name */
1673 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1674 struct cache_entry *ce)
1676 short flags;
1678 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1679 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1680 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1681 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1682 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1683 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1684 ondisk->mode = htonl(ce->ce_mode);
1685 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1686 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1687 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1688 hashcpy(ondisk->sha1, ce->sha1);
1690 flags = ce->ce_flags;
1691 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1692 ondisk->flags = htons(flags);
1693 if (ce->ce_flags & CE_EXTENDED) {
1694 struct ondisk_cache_entry_extended *ondisk2;
1695 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1696 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1697 return ondisk2->name;
1699 else {
1700 return ondisk->name;
1704 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1705 struct strbuf *previous_name)
1707 int size;
1708 struct ondisk_cache_entry *ondisk;
1709 char *name;
1710 int result;
1712 if (!previous_name) {
1713 size = ondisk_ce_size(ce);
1714 ondisk = xcalloc(1, size);
1715 name = copy_cache_entry_to_ondisk(ondisk, ce);
1716 memcpy(name, ce->name, ce_namelen(ce));
1717 } else {
1718 int common, to_remove, prefix_size;
1719 unsigned char to_remove_vi[16];
1720 for (common = 0;
1721 (ce->name[common] &&
1722 common < previous_name->len &&
1723 ce->name[common] == previous_name->buf[common]);
1724 common++)
1725 ; /* still matching */
1726 to_remove = previous_name->len - common;
1727 prefix_size = encode_varint(to_remove, to_remove_vi);
1729 if (ce->ce_flags & CE_EXTENDED)
1730 size = offsetof(struct ondisk_cache_entry_extended, name);
1731 else
1732 size = offsetof(struct ondisk_cache_entry, name);
1733 size += prefix_size + (ce_namelen(ce) - common + 1);
1735 ondisk = xcalloc(1, size);
1736 name = copy_cache_entry_to_ondisk(ondisk, ce);
1737 memcpy(name, to_remove_vi, prefix_size);
1738 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1740 strbuf_splice(previous_name, common, to_remove,
1741 ce->name + common, ce_namelen(ce) - common);
1744 result = ce_write(c, fd, ondisk, size);
1745 free(ondisk);
1746 return result;
1749 static int has_racy_timestamp(struct index_state *istate)
1751 int entries = istate->cache_nr;
1752 int i;
1754 for (i = 0; i < entries; i++) {
1755 struct cache_entry *ce = istate->cache[i];
1756 if (is_racy_timestamp(istate, ce))
1757 return 1;
1759 return 0;
1763 * Opportunistically update the index but do not complain if we can't
1765 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1767 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1768 !write_index(istate, lockfile->fd))
1769 commit_locked_index(lockfile);
1770 else
1771 rollback_lock_file(lockfile);
1774 int write_index(struct index_state *istate, int newfd)
1776 git_SHA_CTX c;
1777 struct cache_header hdr;
1778 int i, err, removed, extended, hdr_version;
1779 struct cache_entry **cache = istate->cache;
1780 int entries = istate->cache_nr;
1781 struct stat st;
1782 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1784 for (i = removed = extended = 0; i < entries; i++) {
1785 if (cache[i]->ce_flags & CE_REMOVE)
1786 removed++;
1788 /* reduce extended entries if possible */
1789 cache[i]->ce_flags &= ~CE_EXTENDED;
1790 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1791 extended++;
1792 cache[i]->ce_flags |= CE_EXTENDED;
1796 if (!istate->version)
1797 istate->version = INDEX_FORMAT_DEFAULT;
1799 /* demote version 3 to version 2 when the latter suffices */
1800 if (istate->version == 3 || istate->version == 2)
1801 istate->version = extended ? 3 : 2;
1803 hdr_version = istate->version;
1805 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1806 hdr.hdr_version = htonl(hdr_version);
1807 hdr.hdr_entries = htonl(entries - removed);
1809 git_SHA1_Init(&c);
1810 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1811 return -1;
1813 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1814 for (i = 0; i < entries; i++) {
1815 struct cache_entry *ce = cache[i];
1816 if (ce->ce_flags & CE_REMOVE)
1817 continue;
1818 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1819 ce_smudge_racily_clean_entry(ce);
1820 if (is_null_sha1(ce->sha1))
1821 return error("cache entry has null sha1: %s", ce->name);
1822 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1823 return -1;
1825 strbuf_release(&previous_name_buf);
1827 /* Write extension data here */
1828 if (istate->cache_tree) {
1829 struct strbuf sb = STRBUF_INIT;
1831 cache_tree_write(&sb, istate->cache_tree);
1832 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1833 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1834 strbuf_release(&sb);
1835 if (err)
1836 return -1;
1838 if (istate->resolve_undo) {
1839 struct strbuf sb = STRBUF_INIT;
1841 resolve_undo_write(&sb, istate->resolve_undo);
1842 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1843 sb.len) < 0
1844 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1845 strbuf_release(&sb);
1846 if (err)
1847 return -1;
1850 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1851 return -1;
1852 istate->timestamp.sec = (unsigned int)st.st_mtime;
1853 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1854 return 0;
1858 * Read the index file that is potentially unmerged into given
1859 * index_state, dropping any unmerged entries. Returns true if
1860 * the index is unmerged. Callers who want to refuse to work
1861 * from an unmerged state can call this and check its return value,
1862 * instead of calling read_cache().
1864 int read_index_unmerged(struct index_state *istate)
1866 int i;
1867 int unmerged = 0;
1869 read_index(istate);
1870 for (i = 0; i < istate->cache_nr; i++) {
1871 struct cache_entry *ce = istate->cache[i];
1872 struct cache_entry *new_ce;
1873 int size, len;
1875 if (!ce_stage(ce))
1876 continue;
1877 unmerged = 1;
1878 len = ce_namelen(ce);
1879 size = cache_entry_size(len);
1880 new_ce = xcalloc(1, size);
1881 memcpy(new_ce->name, ce->name, len);
1882 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1883 new_ce->ce_namelen = len;
1884 new_ce->ce_mode = ce->ce_mode;
1885 if (add_index_entry(istate, new_ce, 0))
1886 return error("%s: cannot drop to stage #0",
1887 ce->name);
1888 i = index_name_pos(istate, new_ce->name, len);
1890 return unmerged;
1894 * Returns 1 if the path is an "other" path with respect to
1895 * the index; that is, the path is not mentioned in the index at all,
1896 * either as a file, a directory with some files in the index,
1897 * or as an unmerged entry.
1899 * We helpfully remove a trailing "/" from directories so that
1900 * the output of read_directory can be used as-is.
1902 int index_name_is_other(const struct index_state *istate, const char *name,
1903 int namelen)
1905 int pos;
1906 if (namelen && name[namelen - 1] == '/')
1907 namelen--;
1908 pos = index_name_pos(istate, name, namelen);
1909 if (0 <= pos)
1910 return 0; /* exact match */
1911 pos = -pos - 1;
1912 if (pos < istate->cache_nr) {
1913 struct cache_entry *ce = istate->cache[pos];
1914 if (ce_namelen(ce) == namelen &&
1915 !memcmp(ce->name, name, namelen))
1916 return 0; /* Yup, this one exists unmerged */
1918 return 1;
1921 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
1923 int pos, len;
1924 unsigned long sz;
1925 enum object_type type;
1926 void *data;
1928 len = strlen(path);
1929 pos = index_name_pos(istate, path, len);
1930 if (pos < 0) {
1932 * We might be in the middle of a merge, in which
1933 * case we would read stage #2 (ours).
1935 int i;
1936 for (i = -pos - 1;
1937 (pos < 0 && i < istate->cache_nr &&
1938 !strcmp(istate->cache[i]->name, path));
1939 i++)
1940 if (ce_stage(istate->cache[i]) == 2)
1941 pos = i;
1943 if (pos < 0)
1944 return NULL;
1945 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
1946 if (!data || type != OBJ_BLOB) {
1947 free(data);
1948 return NULL;
1950 if (size)
1951 *size = sz;
1952 return data;
1955 void stat_validity_clear(struct stat_validity *sv)
1957 free(sv->sd);
1958 sv->sd = NULL;
1961 int stat_validity_check(struct stat_validity *sv, const char *path)
1963 struct stat st;
1965 if (stat(path, &st) < 0)
1966 return sv->sd == NULL;
1967 if (!sv->sd)
1968 return 0;
1969 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
1972 void stat_validity_update(struct stat_validity *sv, int fd)
1974 struct stat st;
1976 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
1977 stat_validity_clear(sv);
1978 else {
1979 if (!sv->sd)
1980 sv->sd = xcalloc(1, sizeof(struct stat_data));
1981 fill_stat_data(sv->sd, &st);