Documentation/config.txt: simplify boolean description in the syntax section
[git/mjg.git] / read-cache.c
blob4fa208b662e0dfd888c8bfa94addaee71244f4f0
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"
17 #include "utf8.h"
19 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
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 set_index_entry(istate, nr, ce);
52 istate->cache_changed = 1;
55 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
57 struct cache_entry *old = istate->cache[nr], *new;
58 int namelen = strlen(new_name);
60 new = xmalloc(cache_entry_size(namelen));
61 copy_cache_entry(new, old);
62 new->ce_flags &= ~CE_STATE_MASK;
63 new->ce_namelen = namelen;
64 memcpy(new->name, new_name, namelen + 1);
66 cache_tree_invalidate_path(istate->cache_tree, old->name);
67 remove_index_entry_at(istate, nr);
68 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
71 void fill_stat_data(struct stat_data *sd, struct stat *st)
73 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
74 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
75 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
76 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
77 sd->sd_dev = st->st_dev;
78 sd->sd_ino = st->st_ino;
79 sd->sd_uid = st->st_uid;
80 sd->sd_gid = st->st_gid;
81 sd->sd_size = st->st_size;
84 int match_stat_data(const struct stat_data *sd, struct stat *st)
86 int changed = 0;
88 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
89 changed |= MTIME_CHANGED;
90 if (trust_ctime && check_stat &&
91 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
92 changed |= CTIME_CHANGED;
94 #ifdef USE_NSEC
95 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
96 changed |= MTIME_CHANGED;
97 if (trust_ctime && check_stat &&
98 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
99 changed |= CTIME_CHANGED;
100 #endif
102 if (check_stat) {
103 if (sd->sd_uid != (unsigned int) st->st_uid ||
104 sd->sd_gid != (unsigned int) st->st_gid)
105 changed |= OWNER_CHANGED;
106 if (sd->sd_ino != (unsigned int) st->st_ino)
107 changed |= INODE_CHANGED;
110 #ifdef USE_STDEV
112 * st_dev breaks on network filesystems where different
113 * clients will have different views of what "device"
114 * the filesystem is on
116 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
117 changed |= INODE_CHANGED;
118 #endif
120 if (sd->sd_size != (unsigned int) st->st_size)
121 changed |= DATA_CHANGED;
123 return changed;
127 * This only updates the "non-critical" parts of the directory
128 * cache, ie the parts that aren't tracked by GIT, and only used
129 * to validate the cache.
131 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
133 fill_stat_data(&ce->ce_stat_data, st);
135 if (assume_unchanged)
136 ce->ce_flags |= CE_VALID;
138 if (S_ISREG(st->st_mode))
139 ce_mark_uptodate(ce);
142 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
144 int match = -1;
145 int fd = open(ce->name, O_RDONLY);
147 if (fd >= 0) {
148 unsigned char sha1[20];
149 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
150 match = hashcmp(sha1, ce->sha1);
151 /* index_fd() closed the file descriptor already */
153 return match;
156 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
158 int match = -1;
159 void *buffer;
160 unsigned long size;
161 enum object_type type;
162 struct strbuf sb = STRBUF_INIT;
164 if (strbuf_readlink(&sb, ce->name, expected_size))
165 return -1;
167 buffer = read_sha1_file(ce->sha1, &type, &size);
168 if (buffer) {
169 if (size == sb.len)
170 match = memcmp(buffer, sb.buf, size);
171 free(buffer);
173 strbuf_release(&sb);
174 return match;
177 static int ce_compare_gitlink(const struct cache_entry *ce)
179 unsigned char sha1[20];
182 * We don't actually require that the .git directory
183 * under GITLINK directory be a valid git directory. It
184 * might even be missing (in case nobody populated that
185 * sub-project).
187 * If so, we consider it always to match.
189 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
190 return 0;
191 return hashcmp(sha1, ce->sha1);
194 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
196 switch (st->st_mode & S_IFMT) {
197 case S_IFREG:
198 if (ce_compare_data(ce, st))
199 return DATA_CHANGED;
200 break;
201 case S_IFLNK:
202 if (ce_compare_link(ce, xsize_t(st->st_size)))
203 return DATA_CHANGED;
204 break;
205 case S_IFDIR:
206 if (S_ISGITLINK(ce->ce_mode))
207 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
208 default:
209 return TYPE_CHANGED;
211 return 0;
214 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
216 unsigned int changed = 0;
218 if (ce->ce_flags & CE_REMOVE)
219 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
221 switch (ce->ce_mode & S_IFMT) {
222 case S_IFREG:
223 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
224 /* We consider only the owner x bit to be relevant for
225 * "mode changes"
227 if (trust_executable_bit &&
228 (0100 & (ce->ce_mode ^ st->st_mode)))
229 changed |= MODE_CHANGED;
230 break;
231 case S_IFLNK:
232 if (!S_ISLNK(st->st_mode) &&
233 (has_symlinks || !S_ISREG(st->st_mode)))
234 changed |= TYPE_CHANGED;
235 break;
236 case S_IFGITLINK:
237 /* We ignore most of the st_xxx fields for gitlinks */
238 if (!S_ISDIR(st->st_mode))
239 changed |= TYPE_CHANGED;
240 else if (ce_compare_gitlink(ce))
241 changed |= DATA_CHANGED;
242 return changed;
243 default:
244 die("internal error: ce_mode is %o", ce->ce_mode);
247 changed |= match_stat_data(&ce->ce_stat_data, st);
249 /* Racily smudged entry? */
250 if (!ce->ce_stat_data.sd_size) {
251 if (!is_empty_blob_sha1(ce->sha1))
252 changed |= DATA_CHANGED;
255 return changed;
258 static int is_racy_timestamp(const struct index_state *istate,
259 const struct cache_entry *ce)
261 return (!S_ISGITLINK(ce->ce_mode) &&
262 istate->timestamp.sec &&
263 #ifdef USE_NSEC
264 /* nanosecond timestamped files can also be racy! */
265 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
266 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
267 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
268 #else
269 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
270 #endif
274 int ie_match_stat(const struct index_state *istate,
275 const struct cache_entry *ce, struct stat *st,
276 unsigned int options)
278 unsigned int changed;
279 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
280 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
281 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
284 * If it's marked as always valid in the index, it's
285 * valid whatever the checked-out copy says.
287 * skip-worktree has the same effect with higher precedence
289 if (!ignore_skip_worktree && ce_skip_worktree(ce))
290 return 0;
291 if (!ignore_valid && (ce->ce_flags & CE_VALID))
292 return 0;
295 * Intent-to-add entries have not been added, so the index entry
296 * by definition never matches what is in the work tree until it
297 * actually gets added.
299 if (ce->ce_flags & CE_INTENT_TO_ADD)
300 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
302 changed = ce_match_stat_basic(ce, st);
305 * Within 1 second of this sequence:
306 * echo xyzzy >file && git-update-index --add file
307 * running this command:
308 * echo frotz >file
309 * would give a falsely clean cache entry. The mtime and
310 * length match the cache, and other stat fields do not change.
312 * We could detect this at update-index time (the cache entry
313 * being registered/updated records the same time as "now")
314 * and delay the return from git-update-index, but that would
315 * effectively mean we can make at most one commit per second,
316 * which is not acceptable. Instead, we check cache entries
317 * whose mtime are the same as the index file timestamp more
318 * carefully than others.
320 if (!changed && is_racy_timestamp(istate, ce)) {
321 if (assume_racy_is_modified)
322 changed |= DATA_CHANGED;
323 else
324 changed |= ce_modified_check_fs(ce, st);
327 return changed;
330 int ie_modified(const struct index_state *istate,
331 const struct cache_entry *ce,
332 struct stat *st, unsigned int options)
334 int changed, changed_fs;
336 changed = ie_match_stat(istate, ce, st, options);
337 if (!changed)
338 return 0;
340 * If the mode or type has changed, there's no point in trying
341 * to refresh the entry - it's not going to match
343 if (changed & (MODE_CHANGED | TYPE_CHANGED))
344 return changed;
347 * Immediately after read-tree or update-index --cacheinfo,
348 * the length field is zero, as we have never even read the
349 * lstat(2) information once, and we cannot trust DATA_CHANGED
350 * returned by ie_match_stat() which in turn was returned by
351 * ce_match_stat_basic() to signal that the filesize of the
352 * blob changed. We have to actually go to the filesystem to
353 * see if the contents match, and if so, should answer "unchanged".
355 * The logic does not apply to gitlinks, as ce_match_stat_basic()
356 * already has checked the actual HEAD from the filesystem in the
357 * subproject. If ie_match_stat() already said it is different,
358 * then we know it is.
360 if ((changed & DATA_CHANGED) &&
361 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
362 return changed;
364 changed_fs = ce_modified_check_fs(ce, st);
365 if (changed_fs)
366 return changed | changed_fs;
367 return 0;
370 int base_name_compare(const char *name1, int len1, int mode1,
371 const char *name2, int len2, int mode2)
373 unsigned char c1, c2;
374 int len = len1 < len2 ? len1 : len2;
375 int cmp;
377 cmp = memcmp(name1, name2, len);
378 if (cmp)
379 return cmp;
380 c1 = name1[len];
381 c2 = name2[len];
382 if (!c1 && S_ISDIR(mode1))
383 c1 = '/';
384 if (!c2 && S_ISDIR(mode2))
385 c2 = '/';
386 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
390 * df_name_compare() is identical to base_name_compare(), except it
391 * compares conflicting directory/file entries as equal. Note that
392 * while a directory name compares as equal to a regular file, they
393 * then individually compare _differently_ to a filename that has
394 * a dot after the basename (because '\0' < '.' < '/').
396 * This is used by routines that want to traverse the git namespace
397 * but then handle conflicting entries together when possible.
399 int df_name_compare(const char *name1, int len1, int mode1,
400 const char *name2, int len2, int mode2)
402 int len = len1 < len2 ? len1 : len2, cmp;
403 unsigned char c1, c2;
405 cmp = memcmp(name1, name2, len);
406 if (cmp)
407 return cmp;
408 /* Directories and files compare equal (same length, same name) */
409 if (len1 == len2)
410 return 0;
411 c1 = name1[len];
412 if (!c1 && S_ISDIR(mode1))
413 c1 = '/';
414 c2 = name2[len];
415 if (!c2 && S_ISDIR(mode2))
416 c2 = '/';
417 if (c1 == '/' && !c2)
418 return 0;
419 if (c2 == '/' && !c1)
420 return 0;
421 return c1 - c2;
424 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
426 int len = len1 < len2 ? len1 : len2;
427 int cmp;
429 cmp = memcmp(name1, name2, len);
430 if (cmp)
431 return cmp;
432 if (len1 < len2)
433 return -1;
434 if (len1 > len2)
435 return 1;
437 if (stage1 < stage2)
438 return -1;
439 if (stage1 > stage2)
440 return 1;
441 return 0;
444 int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
446 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
449 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
451 int first, last;
453 first = 0;
454 last = istate->cache_nr;
455 while (last > first) {
456 int next = (last + first) >> 1;
457 struct cache_entry *ce = istate->cache[next];
458 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
459 if (!cmp)
460 return next;
461 if (cmp < 0) {
462 last = next;
463 continue;
465 first = next+1;
467 return -first-1;
470 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
472 return index_name_stage_pos(istate, name, namelen, 0);
475 /* Remove entry, return true if there are more entries to go.. */
476 int remove_index_entry_at(struct index_state *istate, int pos)
478 struct cache_entry *ce = istate->cache[pos];
480 record_resolve_undo(istate, ce);
481 remove_name_hash(istate, ce);
482 istate->cache_changed = 1;
483 istate->cache_nr--;
484 if (pos >= istate->cache_nr)
485 return 0;
486 memmove(istate->cache + pos,
487 istate->cache + pos + 1,
488 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
489 return 1;
493 * Remove all cache entries marked for removal, that is where
494 * CE_REMOVE is set in ce_flags. This is much more effective than
495 * calling remove_index_entry_at() for each entry to be removed.
497 void remove_marked_cache_entries(struct index_state *istate)
499 struct cache_entry **ce_array = istate->cache;
500 unsigned int i, j;
502 for (i = j = 0; i < istate->cache_nr; i++) {
503 if (ce_array[i]->ce_flags & CE_REMOVE)
504 remove_name_hash(istate, ce_array[i]);
505 else
506 ce_array[j++] = ce_array[i];
508 istate->cache_changed = 1;
509 istate->cache_nr = j;
512 int remove_file_from_index(struct index_state *istate, const char *path)
514 int pos = index_name_pos(istate, path, strlen(path));
515 if (pos < 0)
516 pos = -pos-1;
517 cache_tree_invalidate_path(istate->cache_tree, path);
518 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
519 remove_index_entry_at(istate, pos);
520 return 0;
523 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
525 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
528 static int index_name_pos_also_unmerged(struct index_state *istate,
529 const char *path, int namelen)
531 int pos = index_name_pos(istate, path, namelen);
532 struct cache_entry *ce;
534 if (pos >= 0)
535 return pos;
537 /* maybe unmerged? */
538 pos = -1 - pos;
539 if (pos >= istate->cache_nr ||
540 compare_name((ce = istate->cache[pos]), path, namelen))
541 return -1;
543 /* order of preference: stage 2, 1, 3 */
544 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
545 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
546 !compare_name(ce, path, namelen))
547 pos++;
548 return pos;
551 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
553 int len = ce_namelen(ce);
554 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
558 * If we add a filename that aliases in the cache, we will use the
559 * name that we already have - but we don't want to update the same
560 * alias twice, because that implies that there were actually two
561 * different files with aliasing names!
563 * So we use the CE_ADDED flag to verify that the alias was an old
564 * one before we accept it as
566 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
568 int len;
569 struct cache_entry *new;
571 if (alias->ce_flags & CE_ADDED)
572 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
574 /* Ok, create the new entry using the name of the existing alias */
575 len = ce_namelen(alias);
576 new = xcalloc(1, cache_entry_size(len));
577 memcpy(new->name, alias->name, len);
578 copy_cache_entry(new, ce);
579 free(ce);
580 return new;
583 static void record_intent_to_add(struct cache_entry *ce)
585 unsigned char sha1[20];
586 if (write_sha1_file("", 0, blob_type, sha1))
587 die("cannot create an empty blob in the object database");
588 hashcpy(ce->sha1, sha1);
591 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
593 int size, namelen, was_same;
594 mode_t st_mode = st->st_mode;
595 struct cache_entry *ce, *alias;
596 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
597 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
598 int pretend = flags & ADD_CACHE_PRETEND;
599 int intent_only = flags & ADD_CACHE_INTENT;
600 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
601 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
603 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
604 return error("%s: can only add regular files, symbolic links or git-directories", path);
606 namelen = strlen(path);
607 if (S_ISDIR(st_mode)) {
608 while (namelen && path[namelen-1] == '/')
609 namelen--;
611 size = cache_entry_size(namelen);
612 ce = xcalloc(1, size);
613 memcpy(ce->name, path, namelen);
614 ce->ce_namelen = namelen;
615 if (!intent_only)
616 fill_stat_cache_info(ce, st);
617 else
618 ce->ce_flags |= CE_INTENT_TO_ADD;
620 if (trust_executable_bit && has_symlinks)
621 ce->ce_mode = create_ce_mode(st_mode);
622 else {
623 /* If there is an existing entry, pick the mode bits and type
624 * from it, otherwise assume unexecutable regular file.
626 struct cache_entry *ent;
627 int pos = index_name_pos_also_unmerged(istate, path, namelen);
629 ent = (0 <= pos) ? istate->cache[pos] : NULL;
630 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
633 /* When core.ignorecase=true, determine if a directory of the same name but differing
634 * case already exists within the Git repository. If it does, ensure the directory
635 * case of the file being added to the repository matches (is folded into) the existing
636 * entry's directory case.
638 if (ignore_case) {
639 const char *startPtr = ce->name;
640 const char *ptr = startPtr;
641 while (*ptr) {
642 while (*ptr && *ptr != '/')
643 ++ptr;
644 if (*ptr == '/') {
645 struct cache_entry *foundce;
646 ++ptr;
647 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
648 if (foundce) {
649 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
650 startPtr = ptr;
656 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
657 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
658 /* Nothing changed, really */
659 free(ce);
660 if (!S_ISGITLINK(alias->ce_mode))
661 ce_mark_uptodate(alias);
662 alias->ce_flags |= CE_ADDED;
663 return 0;
665 if (!intent_only) {
666 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
667 return error("unable to index file %s", path);
668 } else
669 record_intent_to_add(ce);
671 if (ignore_case && alias && different_name(ce, alias))
672 ce = create_alias_ce(ce, alias);
673 ce->ce_flags |= CE_ADDED;
675 /* It was suspected to be racily clean, but it turns out to be Ok */
676 was_same = (alias &&
677 !ce_stage(alias) &&
678 !hashcmp(alias->sha1, ce->sha1) &&
679 ce->ce_mode == alias->ce_mode);
681 if (pretend)
683 else if (add_index_entry(istate, ce, add_option))
684 return error("unable to add %s to index",path);
685 if (verbose && !was_same)
686 printf("add '%s'\n", path);
687 return 0;
690 int add_file_to_index(struct index_state *istate, const char *path, int flags)
692 struct stat st;
693 if (lstat(path, &st))
694 die_errno("unable to stat '%s'", path);
695 return add_to_index(istate, path, &st, flags);
698 struct cache_entry *make_cache_entry(unsigned int mode,
699 const unsigned char *sha1, const char *path, int stage,
700 int refresh)
702 int size, len;
703 struct cache_entry *ce;
705 if (!verify_path(path)) {
706 error("Invalid path '%s'", path);
707 return NULL;
710 len = strlen(path);
711 size = cache_entry_size(len);
712 ce = xcalloc(1, size);
714 hashcpy(ce->sha1, sha1);
715 memcpy(ce->name, path, len);
716 ce->ce_flags = create_ce_flags(stage);
717 ce->ce_namelen = len;
718 ce->ce_mode = create_ce_mode(mode);
720 if (refresh)
721 return refresh_cache_entry(ce, 0);
723 return ce;
726 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
728 int len = ce_namelen(a);
729 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
732 int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec)
734 return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
738 * We fundamentally don't like some paths: we don't want
739 * dot or dot-dot anywhere, and for obvious reasons don't
740 * want to recurse into ".git" either.
742 * Also, we don't want double slashes or slashes at the
743 * end that can make pathnames ambiguous.
745 static int verify_dotfile(const char *rest)
748 * The first character was '.', but that
749 * has already been discarded, we now test
750 * the rest.
753 /* "." is not allowed */
754 if (*rest == '\0' || is_dir_sep(*rest))
755 return 0;
757 switch (*rest) {
759 * ".git" followed by NUL or slash is bad. This
760 * shares the path end test with the ".." case.
762 case 'g':
763 case 'G':
764 if (rest[1] != 'i' && rest[1] != 'I')
765 break;
766 if (rest[2] != 't' && rest[2] != 'T')
767 break;
768 rest += 2;
769 /* fallthrough */
770 case '.':
771 if (rest[1] == '\0' || is_dir_sep(rest[1]))
772 return 0;
774 return 1;
777 int verify_path(const char *path)
779 char c;
781 if (has_dos_drive_prefix(path))
782 return 0;
784 goto inside;
785 for (;;) {
786 if (!c)
787 return 1;
788 if (is_dir_sep(c)) {
789 inside:
790 if (protect_hfs && is_hfs_dotgit(path))
791 return 0;
792 if (protect_ntfs && is_ntfs_dotgit(path))
793 return 0;
794 c = *path++;
795 if ((c == '.' && !verify_dotfile(path)) ||
796 is_dir_sep(c) || c == '\0')
797 return 0;
799 c = *path++;
804 * Do we have another file that has the beginning components being a
805 * proper superset of the name we're trying to add?
807 static int has_file_name(struct index_state *istate,
808 const struct cache_entry *ce, int pos, int ok_to_replace)
810 int retval = 0;
811 int len = ce_namelen(ce);
812 int stage = ce_stage(ce);
813 const char *name = ce->name;
815 while (pos < istate->cache_nr) {
816 struct cache_entry *p = istate->cache[pos++];
818 if (len >= ce_namelen(p))
819 break;
820 if (memcmp(name, p->name, len))
821 break;
822 if (ce_stage(p) != stage)
823 continue;
824 if (p->name[len] != '/')
825 continue;
826 if (p->ce_flags & CE_REMOVE)
827 continue;
828 retval = -1;
829 if (!ok_to_replace)
830 break;
831 remove_index_entry_at(istate, --pos);
833 return retval;
837 * Do we have another file with a pathname that is a proper
838 * subset of the name we're trying to add?
840 static int has_dir_name(struct index_state *istate,
841 const struct cache_entry *ce, int pos, int ok_to_replace)
843 int retval = 0;
844 int stage = ce_stage(ce);
845 const char *name = ce->name;
846 const char *slash = name + ce_namelen(ce);
848 for (;;) {
849 int len;
851 for (;;) {
852 if (*--slash == '/')
853 break;
854 if (slash <= ce->name)
855 return retval;
857 len = slash - name;
859 pos = index_name_stage_pos(istate, name, len, stage);
860 if (pos >= 0) {
862 * Found one, but not so fast. This could
863 * be a marker that says "I was here, but
864 * I am being removed". Such an entry is
865 * not a part of the resulting tree, and
866 * it is Ok to have a directory at the same
867 * path.
869 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
870 retval = -1;
871 if (!ok_to_replace)
872 break;
873 remove_index_entry_at(istate, pos);
874 continue;
877 else
878 pos = -pos-1;
881 * Trivial optimization: if we find an entry that
882 * already matches the sub-directory, then we know
883 * we're ok, and we can exit.
885 while (pos < istate->cache_nr) {
886 struct cache_entry *p = istate->cache[pos];
887 if ((ce_namelen(p) <= len) ||
888 (p->name[len] != '/') ||
889 memcmp(p->name, name, len))
890 break; /* not our subdirectory */
891 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
893 * p is at the same stage as our entry, and
894 * is a subdirectory of what we are looking
895 * at, so we cannot have conflicts at our
896 * level or anything shorter.
898 return retval;
899 pos++;
902 return retval;
905 /* We may be in a situation where we already have path/file and path
906 * is being added, or we already have path and path/file is being
907 * added. Either one would result in a nonsense tree that has path
908 * twice when git-write-tree tries to write it out. Prevent it.
910 * If ok-to-replace is specified, we remove the conflicting entries
911 * from the cache so the caller should recompute the insert position.
912 * When this happens, we return non-zero.
914 static int check_file_directory_conflict(struct index_state *istate,
915 const struct cache_entry *ce,
916 int pos, int ok_to_replace)
918 int retval;
921 * When ce is an "I am going away" entry, we allow it to be added
923 if (ce->ce_flags & CE_REMOVE)
924 return 0;
927 * We check if the path is a sub-path of a subsequent pathname
928 * first, since removing those will not change the position
929 * in the array.
931 retval = has_file_name(istate, ce, pos, ok_to_replace);
934 * Then check if the path might have a clashing sub-directory
935 * before it.
937 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
940 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
942 int pos;
943 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
944 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
945 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
946 int new_only = option & ADD_CACHE_NEW_ONLY;
948 cache_tree_invalidate_path(istate->cache_tree, ce->name);
949 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
951 /* existing match? Just replace it. */
952 if (pos >= 0) {
953 if (!new_only)
954 replace_index_entry(istate, pos, ce);
955 return 0;
957 pos = -pos-1;
960 * Inserting a merged entry ("stage 0") into the index
961 * will always replace all non-merged entries..
963 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
964 while (ce_same_name(istate->cache[pos], ce)) {
965 ok_to_add = 1;
966 if (!remove_index_entry_at(istate, pos))
967 break;
971 if (!ok_to_add)
972 return -1;
973 if (!verify_path(ce->name))
974 return error("Invalid path '%s'", ce->name);
976 if (!skip_df_check &&
977 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
978 if (!ok_to_replace)
979 return error("'%s' appears as both a file and as a directory",
980 ce->name);
981 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
982 pos = -pos-1;
984 return pos + 1;
987 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
989 int pos;
991 if (option & ADD_CACHE_JUST_APPEND)
992 pos = istate->cache_nr;
993 else {
994 int ret;
995 ret = add_index_entry_with_check(istate, ce, option);
996 if (ret <= 0)
997 return ret;
998 pos = ret - 1;
1001 /* Make sure the array is big enough .. */
1002 if (istate->cache_nr == istate->cache_alloc) {
1003 istate->cache_alloc = alloc_nr(istate->cache_alloc);
1004 istate->cache = xrealloc(istate->cache,
1005 istate->cache_alloc * sizeof(*istate->cache));
1008 /* Add it in.. */
1009 istate->cache_nr++;
1010 if (istate->cache_nr > pos + 1)
1011 memmove(istate->cache + pos + 1,
1012 istate->cache + pos,
1013 (istate->cache_nr - pos - 1) * sizeof(ce));
1014 set_index_entry(istate, pos, ce);
1015 istate->cache_changed = 1;
1016 return 0;
1020 * "refresh" does not calculate a new sha1 file or bring the
1021 * cache up-to-date for mode/content changes. But what it
1022 * _does_ do is to "re-match" the stat information of a file
1023 * with the cache, so that you can refresh the cache for a
1024 * file that hasn't been changed but where the stat entry is
1025 * out of date.
1027 * For example, you'd want to do this after doing a "git-read-tree",
1028 * to link up the stat cache details with the proper files.
1030 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1031 struct cache_entry *ce,
1032 unsigned int options, int *err,
1033 int *changed_ret)
1035 struct stat st;
1036 struct cache_entry *updated;
1037 int changed, size;
1038 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1039 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1041 if (ce_uptodate(ce))
1042 return ce;
1045 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1046 * that the change to the work tree does not matter and told
1047 * us not to worry.
1049 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1050 ce_mark_uptodate(ce);
1051 return ce;
1053 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1054 ce_mark_uptodate(ce);
1055 return ce;
1058 if (lstat(ce->name, &st) < 0) {
1059 if (err)
1060 *err = errno;
1061 return NULL;
1064 changed = ie_match_stat(istate, ce, &st, options);
1065 if (changed_ret)
1066 *changed_ret = changed;
1067 if (!changed) {
1069 * The path is unchanged. If we were told to ignore
1070 * valid bit, then we did the actual stat check and
1071 * found that the entry is unmodified. If the entry
1072 * is not marked VALID, this is the place to mark it
1073 * valid again, under "assume unchanged" mode.
1075 if (ignore_valid && assume_unchanged &&
1076 !(ce->ce_flags & CE_VALID))
1077 ; /* mark this one VALID again */
1078 else {
1080 * We do not mark the index itself "modified"
1081 * because CE_UPTODATE flag is in-core only;
1082 * we are not going to write this change out.
1084 if (!S_ISGITLINK(ce->ce_mode))
1085 ce_mark_uptodate(ce);
1086 return ce;
1090 if (ie_modified(istate, ce, &st, options)) {
1091 if (err)
1092 *err = EINVAL;
1093 return NULL;
1096 size = ce_size(ce);
1097 updated = xmalloc(size);
1098 memcpy(updated, ce, size);
1099 fill_stat_cache_info(updated, &st);
1101 * If ignore_valid is not set, we should leave CE_VALID bit
1102 * alone. Otherwise, paths marked with --no-assume-unchanged
1103 * (i.e. things to be edited) will reacquire CE_VALID bit
1104 * automatically, which is not really what we want.
1106 if (!ignore_valid && assume_unchanged &&
1107 !(ce->ce_flags & CE_VALID))
1108 updated->ce_flags &= ~CE_VALID;
1110 return updated;
1113 static void show_file(const char * fmt, const char * name, int in_porcelain,
1114 int * first, const char *header_msg)
1116 if (in_porcelain && *first && header_msg) {
1117 printf("%s\n", header_msg);
1118 *first = 0;
1120 printf(fmt, name);
1123 int refresh_index(struct index_state *istate, unsigned int flags,
1124 const struct pathspec *pathspec,
1125 char *seen, const char *header_msg)
1127 int i;
1128 int has_errors = 0;
1129 int really = (flags & REFRESH_REALLY) != 0;
1130 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1131 int quiet = (flags & REFRESH_QUIET) != 0;
1132 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1133 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1134 int first = 1;
1135 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1136 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1137 const char *modified_fmt;
1138 const char *deleted_fmt;
1139 const char *typechange_fmt;
1140 const char *added_fmt;
1141 const char *unmerged_fmt;
1143 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1144 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1145 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1146 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1147 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1148 for (i = 0; i < istate->cache_nr; i++) {
1149 struct cache_entry *ce, *new;
1150 int cache_errno = 0;
1151 int changed = 0;
1152 int filtered = 0;
1154 ce = istate->cache[i];
1155 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1156 continue;
1158 if (pathspec &&
1159 !match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, seen))
1160 filtered = 1;
1162 if (ce_stage(ce)) {
1163 while ((i < istate->cache_nr) &&
1164 ! strcmp(istate->cache[i]->name, ce->name))
1165 i++;
1166 i--;
1167 if (allow_unmerged)
1168 continue;
1169 if (!filtered)
1170 show_file(unmerged_fmt, ce->name, in_porcelain,
1171 &first, header_msg);
1172 has_errors = 1;
1173 continue;
1176 if (filtered)
1177 continue;
1179 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1180 if (new == ce)
1181 continue;
1182 if (!new) {
1183 const char *fmt;
1185 if (not_new && cache_errno == ENOENT)
1186 continue;
1187 if (really && cache_errno == EINVAL) {
1188 /* If we are doing --really-refresh that
1189 * means the index is not valid anymore.
1191 ce->ce_flags &= ~CE_VALID;
1192 istate->cache_changed = 1;
1194 if (quiet)
1195 continue;
1197 if (cache_errno == ENOENT)
1198 fmt = deleted_fmt;
1199 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1200 fmt = added_fmt; /* must be before other checks */
1201 else if (changed & TYPE_CHANGED)
1202 fmt = typechange_fmt;
1203 else
1204 fmt = modified_fmt;
1205 show_file(fmt,
1206 ce->name, in_porcelain, &first, header_msg);
1207 has_errors = 1;
1208 continue;
1211 replace_index_entry(istate, i, new);
1213 return has_errors;
1216 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1218 return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
1222 /*****************************************************************
1223 * Index File I/O
1224 *****************************************************************/
1226 #define INDEX_FORMAT_DEFAULT 3
1229 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1230 * Again - this is just a (very strong in practice) heuristic that
1231 * the inode hasn't changed.
1233 * We save the fields in big-endian order to allow using the
1234 * index file over NFS transparently.
1236 struct ondisk_cache_entry {
1237 struct cache_time ctime;
1238 struct cache_time mtime;
1239 uint32_t dev;
1240 uint32_t ino;
1241 uint32_t mode;
1242 uint32_t uid;
1243 uint32_t gid;
1244 uint32_t size;
1245 unsigned char sha1[20];
1246 uint16_t flags;
1247 char name[FLEX_ARRAY]; /* more */
1251 * This struct is used when CE_EXTENDED bit is 1
1252 * The struct must match ondisk_cache_entry exactly from
1253 * ctime till flags
1255 struct ondisk_cache_entry_extended {
1256 struct cache_time ctime;
1257 struct cache_time mtime;
1258 uint32_t dev;
1259 uint32_t ino;
1260 uint32_t mode;
1261 uint32_t uid;
1262 uint32_t gid;
1263 uint32_t size;
1264 unsigned char sha1[20];
1265 uint16_t flags;
1266 uint16_t flags2;
1267 char name[FLEX_ARRAY]; /* more */
1270 /* These are only used for v3 or lower */
1271 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1272 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1273 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1274 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1275 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1276 ondisk_cache_entry_size(ce_namelen(ce)))
1278 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1280 git_SHA_CTX c;
1281 unsigned char sha1[20];
1282 int hdr_version;
1284 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1285 return error("bad signature");
1286 hdr_version = ntohl(hdr->hdr_version);
1287 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1288 return error("bad index version %d", hdr_version);
1289 git_SHA1_Init(&c);
1290 git_SHA1_Update(&c, hdr, size - 20);
1291 git_SHA1_Final(sha1, &c);
1292 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1293 return error("bad index file sha1 signature");
1294 return 0;
1297 static int read_index_extension(struct index_state *istate,
1298 const char *ext, void *data, unsigned long sz)
1300 switch (CACHE_EXT(ext)) {
1301 case CACHE_EXT_TREE:
1302 istate->cache_tree = cache_tree_read(data, sz);
1303 break;
1304 case CACHE_EXT_RESOLVE_UNDO:
1305 istate->resolve_undo = resolve_undo_read(data, sz);
1306 break;
1307 default:
1308 if (*ext < 'A' || 'Z' < *ext)
1309 return error("index uses %.4s extension, which we do not understand",
1310 ext);
1311 fprintf(stderr, "ignoring %.4s extension\n", ext);
1312 break;
1314 return 0;
1317 int read_index(struct index_state *istate)
1319 return read_index_from(istate, get_index_file());
1322 #ifndef NEEDS_ALIGNED_ACCESS
1323 #define ntoh_s(var) ntohs(var)
1324 #define ntoh_l(var) ntohl(var)
1325 #else
1326 static inline uint16_t ntoh_s_force_align(void *p)
1328 uint16_t x;
1329 memcpy(&x, p, sizeof(x));
1330 return ntohs(x);
1332 static inline uint32_t ntoh_l_force_align(void *p)
1334 uint32_t x;
1335 memcpy(&x, p, sizeof(x));
1336 return ntohl(x);
1338 #define ntoh_s(var) ntoh_s_force_align(&(var))
1339 #define ntoh_l(var) ntoh_l_force_align(&(var))
1340 #endif
1342 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1343 unsigned int flags,
1344 const char *name,
1345 size_t len)
1347 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1349 ce->ce_stat_data.sd_ctime.sec = ntoh_l(ondisk->ctime.sec);
1350 ce->ce_stat_data.sd_mtime.sec = ntoh_l(ondisk->mtime.sec);
1351 ce->ce_stat_data.sd_ctime.nsec = ntoh_l(ondisk->ctime.nsec);
1352 ce->ce_stat_data.sd_mtime.nsec = ntoh_l(ondisk->mtime.nsec);
1353 ce->ce_stat_data.sd_dev = ntoh_l(ondisk->dev);
1354 ce->ce_stat_data.sd_ino = ntoh_l(ondisk->ino);
1355 ce->ce_mode = ntoh_l(ondisk->mode);
1356 ce->ce_stat_data.sd_uid = ntoh_l(ondisk->uid);
1357 ce->ce_stat_data.sd_gid = ntoh_l(ondisk->gid);
1358 ce->ce_stat_data.sd_size = ntoh_l(ondisk->size);
1359 ce->ce_flags = flags & ~CE_NAMEMASK;
1360 ce->ce_namelen = len;
1361 hashcpy(ce->sha1, ondisk->sha1);
1362 memcpy(ce->name, name, len);
1363 ce->name[len] = '\0';
1364 return ce;
1368 * Adjacent cache entries tend to share the leading paths, so it makes
1369 * sense to only store the differences in later entries. In the v4
1370 * on-disk format of the index, each on-disk cache entry stores the
1371 * number of bytes to be stripped from the end of the previous name,
1372 * and the bytes to append to the result, to come up with its name.
1374 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1376 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1377 size_t len = decode_varint(&cp);
1379 if (name->len < len)
1380 die("malformed name field in the index");
1381 strbuf_remove(name, name->len - len, len);
1382 for (ep = cp; *ep; ep++)
1383 ; /* find the end */
1384 strbuf_add(name, cp, ep - cp);
1385 return (const char *)ep + 1 - cp_;
1388 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1389 unsigned long *ent_size,
1390 struct strbuf *previous_name)
1392 struct cache_entry *ce;
1393 size_t len;
1394 const char *name;
1395 unsigned int flags;
1397 /* On-disk flags are just 16 bits */
1398 flags = ntoh_s(ondisk->flags);
1399 len = flags & CE_NAMEMASK;
1401 if (flags & CE_EXTENDED) {
1402 struct ondisk_cache_entry_extended *ondisk2;
1403 int extended_flags;
1404 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1405 extended_flags = ntoh_s(ondisk2->flags2) << 16;
1406 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1407 if (extended_flags & ~CE_EXTENDED_FLAGS)
1408 die("Unknown index entry format %08x", extended_flags);
1409 flags |= extended_flags;
1410 name = ondisk2->name;
1412 else
1413 name = ondisk->name;
1415 if (!previous_name) {
1416 /* v3 and earlier */
1417 if (len == CE_NAMEMASK)
1418 len = strlen(name);
1419 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1421 *ent_size = ondisk_ce_size(ce);
1422 } else {
1423 unsigned long consumed;
1424 consumed = expand_name_field(previous_name, name);
1425 ce = cache_entry_from_ondisk(ondisk, flags,
1426 previous_name->buf,
1427 previous_name->len);
1429 *ent_size = (name - ((char *)ondisk)) + consumed;
1431 return ce;
1434 /* remember to discard_cache() before reading a different cache! */
1435 int read_index_from(struct index_state *istate, const char *path)
1437 int fd, i;
1438 struct stat st;
1439 unsigned long src_offset;
1440 struct cache_header *hdr;
1441 void *mmap;
1442 size_t mmap_size;
1443 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1445 if (istate->initialized)
1446 return istate->cache_nr;
1448 istate->timestamp.sec = 0;
1449 istate->timestamp.nsec = 0;
1450 fd = open(path, O_RDONLY);
1451 if (fd < 0) {
1452 if (errno == ENOENT)
1453 return 0;
1454 die_errno("index file open failed");
1457 if (fstat(fd, &st))
1458 die_errno("cannot stat the open index");
1460 mmap_size = xsize_t(st.st_size);
1461 if (mmap_size < sizeof(struct cache_header) + 20)
1462 die("index file smaller than expected");
1464 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1465 if (mmap == MAP_FAILED)
1466 die_errno("unable to map index file");
1467 close(fd);
1469 hdr = mmap;
1470 if (verify_hdr(hdr, mmap_size) < 0)
1471 goto unmap;
1473 istate->version = ntohl(hdr->hdr_version);
1474 istate->cache_nr = ntohl(hdr->hdr_entries);
1475 istate->cache_alloc = alloc_nr(istate->cache_nr);
1476 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1477 istate->initialized = 1;
1479 if (istate->version == 4)
1480 previous_name = &previous_name_buf;
1481 else
1482 previous_name = NULL;
1484 src_offset = sizeof(*hdr);
1485 for (i = 0; i < istate->cache_nr; i++) {
1486 struct ondisk_cache_entry *disk_ce;
1487 struct cache_entry *ce;
1488 unsigned long consumed;
1490 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1491 ce = create_from_disk(disk_ce, &consumed, previous_name);
1492 set_index_entry(istate, i, ce);
1494 src_offset += consumed;
1496 strbuf_release(&previous_name_buf);
1497 istate->timestamp.sec = st.st_mtime;
1498 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1500 while (src_offset <= mmap_size - 20 - 8) {
1501 /* After an array of active_nr index entries,
1502 * there can be arbitrary number of extended
1503 * sections, each of which is prefixed with
1504 * extension name (4-byte) and section length
1505 * in 4-byte network byte order.
1507 uint32_t extsize;
1508 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1509 extsize = ntohl(extsize);
1510 if (read_index_extension(istate,
1511 (const char *) mmap + src_offset,
1512 (char *) mmap + src_offset + 8,
1513 extsize) < 0)
1514 goto unmap;
1515 src_offset += 8;
1516 src_offset += extsize;
1518 munmap(mmap, mmap_size);
1519 return istate->cache_nr;
1521 unmap:
1522 munmap(mmap, mmap_size);
1523 die("index file corrupt");
1526 int is_index_unborn(struct index_state *istate)
1528 return (!istate->cache_nr && !istate->timestamp.sec);
1531 int discard_index(struct index_state *istate)
1533 int i;
1535 for (i = 0; i < istate->cache_nr; i++)
1536 free(istate->cache[i]);
1537 resolve_undo_clear_index(istate);
1538 istate->cache_nr = 0;
1539 istate->cache_changed = 0;
1540 istate->timestamp.sec = 0;
1541 istate->timestamp.nsec = 0;
1542 free_name_hash(istate);
1543 cache_tree_free(&(istate->cache_tree));
1544 istate->initialized = 0;
1545 free(istate->cache);
1546 istate->cache = NULL;
1547 istate->cache_alloc = 0;
1548 return 0;
1551 int unmerged_index(const struct index_state *istate)
1553 int i;
1554 for (i = 0; i < istate->cache_nr; i++) {
1555 if (ce_stage(istate->cache[i]))
1556 return 1;
1558 return 0;
1561 #define WRITE_BUFFER_SIZE 8192
1562 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1563 static unsigned long write_buffer_len;
1565 static int ce_write_flush(git_SHA_CTX *context, int fd)
1567 unsigned int buffered = write_buffer_len;
1568 if (buffered) {
1569 git_SHA1_Update(context, write_buffer, buffered);
1570 if (write_in_full(fd, write_buffer, buffered) != buffered)
1571 return -1;
1572 write_buffer_len = 0;
1574 return 0;
1577 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1579 while (len) {
1580 unsigned int buffered = write_buffer_len;
1581 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1582 if (partial > len)
1583 partial = len;
1584 memcpy(write_buffer + buffered, data, partial);
1585 buffered += partial;
1586 if (buffered == WRITE_BUFFER_SIZE) {
1587 write_buffer_len = buffered;
1588 if (ce_write_flush(context, fd))
1589 return -1;
1590 buffered = 0;
1592 write_buffer_len = buffered;
1593 len -= partial;
1594 data = (char *) data + partial;
1596 return 0;
1599 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1600 unsigned int ext, unsigned int sz)
1602 ext = htonl(ext);
1603 sz = htonl(sz);
1604 return ((ce_write(context, fd, &ext, 4) < 0) ||
1605 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1608 static int ce_flush(git_SHA_CTX *context, int fd)
1610 unsigned int left = write_buffer_len;
1612 if (left) {
1613 write_buffer_len = 0;
1614 git_SHA1_Update(context, write_buffer, left);
1617 /* Flush first if not enough space for SHA1 signature */
1618 if (left + 20 > WRITE_BUFFER_SIZE) {
1619 if (write_in_full(fd, write_buffer, left) != left)
1620 return -1;
1621 left = 0;
1624 /* Append the SHA1 signature at the end */
1625 git_SHA1_Final(write_buffer + left, context);
1626 left += 20;
1627 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1630 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1633 * The only thing we care about in this function is to smudge the
1634 * falsely clean entry due to touch-update-touch race, so we leave
1635 * everything else as they are. We are called for entries whose
1636 * ce_stat_data.sd_mtime match the index file mtime.
1638 * Note that this actually does not do much for gitlinks, for
1639 * which ce_match_stat_basic() always goes to the actual
1640 * contents. The caller checks with is_racy_timestamp() which
1641 * always says "no" for gitlinks, so we are not called for them ;-)
1643 struct stat st;
1645 if (lstat(ce->name, &st) < 0)
1646 return;
1647 if (ce_match_stat_basic(ce, &st))
1648 return;
1649 if (ce_modified_check_fs(ce, &st)) {
1650 /* This is "racily clean"; smudge it. Note that this
1651 * is a tricky code. At first glance, it may appear
1652 * that it can break with this sequence:
1654 * $ echo xyzzy >frotz
1655 * $ git-update-index --add frotz
1656 * $ : >frotz
1657 * $ sleep 3
1658 * $ echo filfre >nitfol
1659 * $ git-update-index --add nitfol
1661 * but it does not. When the second update-index runs,
1662 * it notices that the entry "frotz" has the same timestamp
1663 * as index, and if we were to smudge it by resetting its
1664 * size to zero here, then the object name recorded
1665 * in index is the 6-byte file but the cached stat information
1666 * becomes zero --- which would then match what we would
1667 * obtain from the filesystem next time we stat("frotz").
1669 * However, the second update-index, before calling
1670 * this function, notices that the cached size is 6
1671 * bytes and what is on the filesystem is an empty
1672 * file, and never calls us, so the cached size information
1673 * for "frotz" stays 6 which does not match the filesystem.
1675 ce->ce_stat_data.sd_size = 0;
1679 /* Copy miscellaneous fields but not the name */
1680 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1681 struct cache_entry *ce)
1683 short flags;
1685 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1686 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1687 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1688 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1689 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1690 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1691 ondisk->mode = htonl(ce->ce_mode);
1692 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1693 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1694 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1695 hashcpy(ondisk->sha1, ce->sha1);
1697 flags = ce->ce_flags;
1698 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1699 ondisk->flags = htons(flags);
1700 if (ce->ce_flags & CE_EXTENDED) {
1701 struct ondisk_cache_entry_extended *ondisk2;
1702 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1703 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1704 return ondisk2->name;
1706 else {
1707 return ondisk->name;
1711 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1712 struct strbuf *previous_name)
1714 int size;
1715 struct ondisk_cache_entry *ondisk;
1716 char *name;
1717 int result;
1719 if (!previous_name) {
1720 size = ondisk_ce_size(ce);
1721 ondisk = xcalloc(1, size);
1722 name = copy_cache_entry_to_ondisk(ondisk, ce);
1723 memcpy(name, ce->name, ce_namelen(ce));
1724 } else {
1725 int common, to_remove, prefix_size;
1726 unsigned char to_remove_vi[16];
1727 for (common = 0;
1728 (ce->name[common] &&
1729 common < previous_name->len &&
1730 ce->name[common] == previous_name->buf[common]);
1731 common++)
1732 ; /* still matching */
1733 to_remove = previous_name->len - common;
1734 prefix_size = encode_varint(to_remove, to_remove_vi);
1736 if (ce->ce_flags & CE_EXTENDED)
1737 size = offsetof(struct ondisk_cache_entry_extended, name);
1738 else
1739 size = offsetof(struct ondisk_cache_entry, name);
1740 size += prefix_size + (ce_namelen(ce) - common + 1);
1742 ondisk = xcalloc(1, size);
1743 name = copy_cache_entry_to_ondisk(ondisk, ce);
1744 memcpy(name, to_remove_vi, prefix_size);
1745 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1747 strbuf_splice(previous_name, common, to_remove,
1748 ce->name + common, ce_namelen(ce) - common);
1751 result = ce_write(c, fd, ondisk, size);
1752 free(ondisk);
1753 return result;
1756 static int has_racy_timestamp(struct index_state *istate)
1758 int entries = istate->cache_nr;
1759 int i;
1761 for (i = 0; i < entries; i++) {
1762 struct cache_entry *ce = istate->cache[i];
1763 if (is_racy_timestamp(istate, ce))
1764 return 1;
1766 return 0;
1770 * Opportunistically update the index but do not complain if we can't
1772 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1774 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1775 !write_index(istate, lockfile->fd))
1776 commit_locked_index(lockfile);
1777 else
1778 rollback_lock_file(lockfile);
1781 int write_index(struct index_state *istate, int newfd)
1783 git_SHA_CTX c;
1784 struct cache_header hdr;
1785 int i, err, removed, extended, hdr_version;
1786 struct cache_entry **cache = istate->cache;
1787 int entries = istate->cache_nr;
1788 struct stat st;
1789 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1791 for (i = removed = extended = 0; i < entries; i++) {
1792 if (cache[i]->ce_flags & CE_REMOVE)
1793 removed++;
1795 /* reduce extended entries if possible */
1796 cache[i]->ce_flags &= ~CE_EXTENDED;
1797 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1798 extended++;
1799 cache[i]->ce_flags |= CE_EXTENDED;
1803 if (!istate->version)
1804 istate->version = INDEX_FORMAT_DEFAULT;
1806 /* demote version 3 to version 2 when the latter suffices */
1807 if (istate->version == 3 || istate->version == 2)
1808 istate->version = extended ? 3 : 2;
1810 hdr_version = istate->version;
1812 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1813 hdr.hdr_version = htonl(hdr_version);
1814 hdr.hdr_entries = htonl(entries - removed);
1816 git_SHA1_Init(&c);
1817 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1818 return -1;
1820 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1821 for (i = 0; i < entries; i++) {
1822 struct cache_entry *ce = cache[i];
1823 if (ce->ce_flags & CE_REMOVE)
1824 continue;
1825 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1826 ce_smudge_racily_clean_entry(ce);
1827 if (is_null_sha1(ce->sha1)) {
1828 static const char msg[] = "cache entry has null sha1: %s";
1829 static int allow = -1;
1831 if (allow < 0)
1832 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1833 if (allow)
1834 warning(msg, ce->name);
1835 else
1836 return error(msg, ce->name);
1838 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1839 return -1;
1841 strbuf_release(&previous_name_buf);
1843 /* Write extension data here */
1844 if (istate->cache_tree) {
1845 struct strbuf sb = STRBUF_INIT;
1847 cache_tree_write(&sb, istate->cache_tree);
1848 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1849 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1850 strbuf_release(&sb);
1851 if (err)
1852 return -1;
1854 if (istate->resolve_undo) {
1855 struct strbuf sb = STRBUF_INIT;
1857 resolve_undo_write(&sb, istate->resolve_undo);
1858 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1859 sb.len) < 0
1860 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1861 strbuf_release(&sb);
1862 if (err)
1863 return -1;
1866 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1867 return -1;
1868 istate->timestamp.sec = (unsigned int)st.st_mtime;
1869 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1870 return 0;
1874 * Read the index file that is potentially unmerged into given
1875 * index_state, dropping any unmerged entries. Returns true if
1876 * the index is unmerged. Callers who want to refuse to work
1877 * from an unmerged state can call this and check its return value,
1878 * instead of calling read_cache().
1880 int read_index_unmerged(struct index_state *istate)
1882 int i;
1883 int unmerged = 0;
1885 read_index(istate);
1886 for (i = 0; i < istate->cache_nr; i++) {
1887 struct cache_entry *ce = istate->cache[i];
1888 struct cache_entry *new_ce;
1889 int size, len;
1891 if (!ce_stage(ce))
1892 continue;
1893 unmerged = 1;
1894 len = ce_namelen(ce);
1895 size = cache_entry_size(len);
1896 new_ce = xcalloc(1, size);
1897 memcpy(new_ce->name, ce->name, len);
1898 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1899 new_ce->ce_namelen = len;
1900 new_ce->ce_mode = ce->ce_mode;
1901 if (add_index_entry(istate, new_ce, 0))
1902 return error("%s: cannot drop to stage #0",
1903 ce->name);
1904 i = index_name_pos(istate, new_ce->name, len);
1906 return unmerged;
1910 * Returns 1 if the path is an "other" path with respect to
1911 * the index; that is, the path is not mentioned in the index at all,
1912 * either as a file, a directory with some files in the index,
1913 * or as an unmerged entry.
1915 * We helpfully remove a trailing "/" from directories so that
1916 * the output of read_directory can be used as-is.
1918 int index_name_is_other(const struct index_state *istate, const char *name,
1919 int namelen)
1921 int pos;
1922 if (namelen && name[namelen - 1] == '/')
1923 namelen--;
1924 pos = index_name_pos(istate, name, namelen);
1925 if (0 <= pos)
1926 return 0; /* exact match */
1927 pos = -pos - 1;
1928 if (pos < istate->cache_nr) {
1929 struct cache_entry *ce = istate->cache[pos];
1930 if (ce_namelen(ce) == namelen &&
1931 !memcmp(ce->name, name, namelen))
1932 return 0; /* Yup, this one exists unmerged */
1934 return 1;
1937 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
1939 int pos, len;
1940 unsigned long sz;
1941 enum object_type type;
1942 void *data;
1944 len = strlen(path);
1945 pos = index_name_pos(istate, path, len);
1946 if (pos < 0) {
1948 * We might be in the middle of a merge, in which
1949 * case we would read stage #2 (ours).
1951 int i;
1952 for (i = -pos - 1;
1953 (pos < 0 && i < istate->cache_nr &&
1954 !strcmp(istate->cache[i]->name, path));
1955 i++)
1956 if (ce_stage(istate->cache[i]) == 2)
1957 pos = i;
1959 if (pos < 0)
1960 return NULL;
1961 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
1962 if (!data || type != OBJ_BLOB) {
1963 free(data);
1964 return NULL;
1966 if (size)
1967 *size = sz;
1968 return data;
1971 void stat_validity_clear(struct stat_validity *sv)
1973 free(sv->sd);
1974 sv->sd = NULL;
1977 int stat_validity_check(struct stat_validity *sv, const char *path)
1979 struct stat st;
1981 if (stat(path, &st) < 0)
1982 return sv->sd == NULL;
1983 if (!sv->sd)
1984 return 0;
1985 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
1988 void stat_validity_update(struct stat_validity *sv, int fd)
1990 struct stat st;
1992 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
1993 stat_validity_clear(sv);
1994 else {
1995 if (!sv->sd)
1996 sv->sd = xcalloc(1, sizeof(struct stat_data));
1997 fill_stat_data(sv->sd, &st);