pack-objects: turn off bitmaps when we split packs
[git.git] / read-cache.c
blob7f5645e74546e459efdb584dbf63e1fd75857317
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "cache-tree.h"
9 #include "refs.h"
10 #include "dir.h"
11 #include "tree.h"
12 #include "commit.h"
13 #include "blob.h"
14 #include "resolve-undo.h"
15 #include "strbuf.h"
16 #include "varint.h"
18 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
19 unsigned int options);
21 /* Mask for the name length in ce_flags in the on-disk index */
23 #define CE_NAMEMASK (0x0fff)
25 /* Index extensions.
27 * The first letter should be 'A'..'Z' for extensions that are not
28 * necessary for a correct operation (i.e. optimization data).
29 * When new extensions are added that _needs_ to be understood in
30 * order to correctly interpret the index file, pick character that
31 * is outside the range, to cause the reader to abort.
34 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
35 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
36 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 struct index_state the_index;
40 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
42 istate->cache[nr] = ce;
43 add_name_hash(istate, ce);
46 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
48 struct cache_entry *old = istate->cache[nr];
50 remove_name_hash(istate, old);
51 free(old);
52 set_index_entry(istate, nr, ce);
53 istate->cache_changed = 1;
56 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
58 struct cache_entry *old = istate->cache[nr], *new;
59 int namelen = strlen(new_name);
61 new = xmalloc(cache_entry_size(namelen));
62 copy_cache_entry(new, old);
63 new->ce_flags &= ~CE_HASHED;
64 new->ce_namelen = namelen;
65 memcpy(new->name, new_name, namelen + 1);
67 cache_tree_invalidate_path(istate->cache_tree, old->name);
68 remove_index_entry_at(istate, nr);
69 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
72 void fill_stat_data(struct stat_data *sd, struct stat *st)
74 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
75 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
76 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
77 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
78 sd->sd_dev = st->st_dev;
79 sd->sd_ino = st->st_ino;
80 sd->sd_uid = st->st_uid;
81 sd->sd_gid = st->st_gid;
82 sd->sd_size = st->st_size;
85 int match_stat_data(const struct stat_data *sd, struct stat *st)
87 int changed = 0;
89 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
90 changed |= MTIME_CHANGED;
91 if (trust_ctime && check_stat &&
92 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
93 changed |= CTIME_CHANGED;
95 #ifdef USE_NSEC
96 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
97 changed |= MTIME_CHANGED;
98 if (trust_ctime && check_stat &&
99 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
100 changed |= CTIME_CHANGED;
101 #endif
103 if (check_stat) {
104 if (sd->sd_uid != (unsigned int) st->st_uid ||
105 sd->sd_gid != (unsigned int) st->st_gid)
106 changed |= OWNER_CHANGED;
107 if (sd->sd_ino != (unsigned int) st->st_ino)
108 changed |= INODE_CHANGED;
111 #ifdef USE_STDEV
113 * st_dev breaks on network filesystems where different
114 * clients will have different views of what "device"
115 * the filesystem is on
117 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
118 changed |= INODE_CHANGED;
119 #endif
121 if (sd->sd_size != (unsigned int) st->st_size)
122 changed |= DATA_CHANGED;
124 return changed;
128 * This only updates the "non-critical" parts of the directory
129 * cache, ie the parts that aren't tracked by GIT, and only used
130 * to validate the cache.
132 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
134 fill_stat_data(&ce->ce_stat_data, st);
136 if (assume_unchanged)
137 ce->ce_flags |= CE_VALID;
139 if (S_ISREG(st->st_mode))
140 ce_mark_uptodate(ce);
143 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
145 int match = -1;
146 int fd = open(ce->name, O_RDONLY);
148 if (fd >= 0) {
149 unsigned char sha1[20];
150 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
151 match = hashcmp(sha1, ce->sha1);
152 /* index_fd() closed the file descriptor already */
154 return match;
157 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
159 int match = -1;
160 void *buffer;
161 unsigned long size;
162 enum object_type type;
163 struct strbuf sb = STRBUF_INIT;
165 if (strbuf_readlink(&sb, ce->name, expected_size))
166 return -1;
168 buffer = read_sha1_file(ce->sha1, &type, &size);
169 if (buffer) {
170 if (size == sb.len)
171 match = memcmp(buffer, sb.buf, size);
172 free(buffer);
174 strbuf_release(&sb);
175 return match;
178 static int ce_compare_gitlink(const struct cache_entry *ce)
180 unsigned char sha1[20];
183 * We don't actually require that the .git directory
184 * under GITLINK directory be a valid git directory. It
185 * might even be missing (in case nobody populated that
186 * sub-project).
188 * If so, we consider it always to match.
190 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
191 return 0;
192 return hashcmp(sha1, ce->sha1);
195 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
197 switch (st->st_mode & S_IFMT) {
198 case S_IFREG:
199 if (ce_compare_data(ce, st))
200 return DATA_CHANGED;
201 break;
202 case S_IFLNK:
203 if (ce_compare_link(ce, xsize_t(st->st_size)))
204 return DATA_CHANGED;
205 break;
206 case S_IFDIR:
207 if (S_ISGITLINK(ce->ce_mode))
208 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
209 default:
210 return TYPE_CHANGED;
212 return 0;
215 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
217 unsigned int changed = 0;
219 if (ce->ce_flags & CE_REMOVE)
220 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
222 switch (ce->ce_mode & S_IFMT) {
223 case S_IFREG:
224 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
225 /* We consider only the owner x bit to be relevant for
226 * "mode changes"
228 if (trust_executable_bit &&
229 (0100 & (ce->ce_mode ^ st->st_mode)))
230 changed |= MODE_CHANGED;
231 break;
232 case S_IFLNK:
233 if (!S_ISLNK(st->st_mode) &&
234 (has_symlinks || !S_ISREG(st->st_mode)))
235 changed |= TYPE_CHANGED;
236 break;
237 case S_IFGITLINK:
238 /* We ignore most of the st_xxx fields for gitlinks */
239 if (!S_ISDIR(st->st_mode))
240 changed |= TYPE_CHANGED;
241 else if (ce_compare_gitlink(ce))
242 changed |= DATA_CHANGED;
243 return changed;
244 default:
245 die("internal error: ce_mode is %o", ce->ce_mode);
248 changed |= match_stat_data(&ce->ce_stat_data, st);
250 /* Racily smudged entry? */
251 if (!ce->ce_stat_data.sd_size) {
252 if (!is_empty_blob_sha1(ce->sha1))
253 changed |= DATA_CHANGED;
256 return changed;
259 static int is_racy_timestamp(const struct index_state *istate,
260 const struct cache_entry *ce)
262 return (!S_ISGITLINK(ce->ce_mode) &&
263 istate->timestamp.sec &&
264 #ifdef USE_NSEC
265 /* nanosecond timestamped files can also be racy! */
266 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
267 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
268 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
269 #else
270 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
271 #endif
275 int ie_match_stat(const struct index_state *istate,
276 const struct cache_entry *ce, struct stat *st,
277 unsigned int options)
279 unsigned int changed;
280 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
281 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
282 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
285 * If it's marked as always valid in the index, it's
286 * valid whatever the checked-out copy says.
288 * skip-worktree has the same effect with higher precedence
290 if (!ignore_skip_worktree && ce_skip_worktree(ce))
291 return 0;
292 if (!ignore_valid && (ce->ce_flags & CE_VALID))
293 return 0;
296 * Intent-to-add entries have not been added, so the index entry
297 * by definition never matches what is in the work tree until it
298 * actually gets added.
300 if (ce->ce_flags & CE_INTENT_TO_ADD)
301 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
303 changed = ce_match_stat_basic(ce, st);
306 * Within 1 second of this sequence:
307 * echo xyzzy >file && git-update-index --add file
308 * running this command:
309 * echo frotz >file
310 * would give a falsely clean cache entry. The mtime and
311 * length match the cache, and other stat fields do not change.
313 * We could detect this at update-index time (the cache entry
314 * being registered/updated records the same time as "now")
315 * and delay the return from git-update-index, but that would
316 * effectively mean we can make at most one commit per second,
317 * which is not acceptable. Instead, we check cache entries
318 * whose mtime are the same as the index file timestamp more
319 * carefully than others.
321 if (!changed && is_racy_timestamp(istate, ce)) {
322 if (assume_racy_is_modified)
323 changed |= DATA_CHANGED;
324 else
325 changed |= ce_modified_check_fs(ce, st);
328 return changed;
331 int ie_modified(const struct index_state *istate,
332 const struct cache_entry *ce,
333 struct stat *st, unsigned int options)
335 int changed, changed_fs;
337 changed = ie_match_stat(istate, ce, st, options);
338 if (!changed)
339 return 0;
341 * If the mode or type has changed, there's no point in trying
342 * to refresh the entry - it's not going to match
344 if (changed & (MODE_CHANGED | TYPE_CHANGED))
345 return changed;
348 * Immediately after read-tree or update-index --cacheinfo,
349 * the length field is zero, as we have never even read the
350 * lstat(2) information once, and we cannot trust DATA_CHANGED
351 * returned by ie_match_stat() which in turn was returned by
352 * ce_match_stat_basic() to signal that the filesize of the
353 * blob changed. We have to actually go to the filesystem to
354 * see if the contents match, and if so, should answer "unchanged".
356 * The logic does not apply to gitlinks, as ce_match_stat_basic()
357 * already has checked the actual HEAD from the filesystem in the
358 * subproject. If ie_match_stat() already said it is different,
359 * then we know it is.
361 if ((changed & DATA_CHANGED) &&
362 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
363 return changed;
365 changed_fs = ce_modified_check_fs(ce, st);
366 if (changed_fs)
367 return changed | changed_fs;
368 return 0;
371 int base_name_compare(const char *name1, int len1, int mode1,
372 const char *name2, int len2, int mode2)
374 unsigned char c1, c2;
375 int len = len1 < len2 ? len1 : len2;
376 int cmp;
378 cmp = memcmp(name1, name2, len);
379 if (cmp)
380 return cmp;
381 c1 = name1[len];
382 c2 = name2[len];
383 if (!c1 && S_ISDIR(mode1))
384 c1 = '/';
385 if (!c2 && S_ISDIR(mode2))
386 c2 = '/';
387 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
391 * df_name_compare() is identical to base_name_compare(), except it
392 * compares conflicting directory/file entries as equal. Note that
393 * while a directory name compares as equal to a regular file, they
394 * then individually compare _differently_ to a filename that has
395 * a dot after the basename (because '\0' < '.' < '/').
397 * This is used by routines that want to traverse the git namespace
398 * but then handle conflicting entries together when possible.
400 int df_name_compare(const char *name1, int len1, int mode1,
401 const char *name2, int len2, int mode2)
403 int len = len1 < len2 ? len1 : len2, cmp;
404 unsigned char c1, c2;
406 cmp = memcmp(name1, name2, len);
407 if (cmp)
408 return cmp;
409 /* Directories and files compare equal (same length, same name) */
410 if (len1 == len2)
411 return 0;
412 c1 = name1[len];
413 if (!c1 && S_ISDIR(mode1))
414 c1 = '/';
415 c2 = name2[len];
416 if (!c2 && S_ISDIR(mode2))
417 c2 = '/';
418 if (c1 == '/' && !c2)
419 return 0;
420 if (c2 == '/' && !c1)
421 return 0;
422 return c1 - c2;
425 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
427 int len = len1 < len2 ? len1 : len2;
428 int cmp;
430 cmp = memcmp(name1, name2, len);
431 if (cmp)
432 return cmp;
433 if (len1 < len2)
434 return -1;
435 if (len1 > len2)
436 return 1;
438 if (stage1 < stage2)
439 return -1;
440 if (stage1 > stage2)
441 return 1;
442 return 0;
445 int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
447 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
450 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
452 int first, last;
454 first = 0;
455 last = istate->cache_nr;
456 while (last > first) {
457 int next = (last + first) >> 1;
458 struct cache_entry *ce = istate->cache[next];
459 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
460 if (!cmp)
461 return next;
462 if (cmp < 0) {
463 last = next;
464 continue;
466 first = next+1;
468 return -first-1;
471 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
473 return index_name_stage_pos(istate, name, namelen, 0);
476 /* Remove entry, return true if there are more entries to go.. */
477 int remove_index_entry_at(struct index_state *istate, int pos)
479 struct cache_entry *ce = istate->cache[pos];
481 record_resolve_undo(istate, ce);
482 remove_name_hash(istate, ce);
483 free(ce);
484 istate->cache_changed = 1;
485 istate->cache_nr--;
486 if (pos >= istate->cache_nr)
487 return 0;
488 memmove(istate->cache + pos,
489 istate->cache + pos + 1,
490 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
491 return 1;
495 * Remove all cache entries marked for removal, that is where
496 * CE_REMOVE is set in ce_flags. This is much more effective than
497 * calling remove_index_entry_at() for each entry to be removed.
499 void remove_marked_cache_entries(struct index_state *istate)
501 struct cache_entry **ce_array = istate->cache;
502 unsigned int i, j;
504 for (i = j = 0; i < istate->cache_nr; i++) {
505 if (ce_array[i]->ce_flags & CE_REMOVE) {
506 remove_name_hash(istate, ce_array[i]);
507 free(ce_array[i]);
509 else
510 ce_array[j++] = ce_array[i];
512 istate->cache_changed = 1;
513 istate->cache_nr = j;
516 int remove_file_from_index(struct index_state *istate, const char *path)
518 int pos = index_name_pos(istate, path, strlen(path));
519 if (pos < 0)
520 pos = -pos-1;
521 cache_tree_invalidate_path(istate->cache_tree, path);
522 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
523 remove_index_entry_at(istate, pos);
524 return 0;
527 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
529 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
532 static int index_name_pos_also_unmerged(struct index_state *istate,
533 const char *path, int namelen)
535 int pos = index_name_pos(istate, path, namelen);
536 struct cache_entry *ce;
538 if (pos >= 0)
539 return pos;
541 /* maybe unmerged? */
542 pos = -1 - pos;
543 if (pos >= istate->cache_nr ||
544 compare_name((ce = istate->cache[pos]), path, namelen))
545 return -1;
547 /* order of preference: stage 2, 1, 3 */
548 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
549 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
550 !compare_name(ce, path, namelen))
551 pos++;
552 return pos;
555 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
557 int len = ce_namelen(ce);
558 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
562 * If we add a filename that aliases in the cache, we will use the
563 * name that we already have - but we don't want to update the same
564 * alias twice, because that implies that there were actually two
565 * different files with aliasing names!
567 * So we use the CE_ADDED flag to verify that the alias was an old
568 * one before we accept it as
570 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
572 int len;
573 struct cache_entry *new;
575 if (alias->ce_flags & CE_ADDED)
576 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
578 /* Ok, create the new entry using the name of the existing alias */
579 len = ce_namelen(alias);
580 new = xcalloc(1, cache_entry_size(len));
581 memcpy(new->name, alias->name, len);
582 copy_cache_entry(new, ce);
583 free(ce);
584 return new;
587 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
589 unsigned char sha1[20];
590 if (write_sha1_file("", 0, blob_type, sha1))
591 die("cannot create an empty blob in the object database");
592 hashcpy(ce->sha1, sha1);
595 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
597 int size, namelen, was_same;
598 mode_t st_mode = st->st_mode;
599 struct cache_entry *ce, *alias;
600 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
601 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
602 int pretend = flags & ADD_CACHE_PRETEND;
603 int intent_only = flags & ADD_CACHE_INTENT;
604 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
605 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
607 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
608 return error("%s: can only add regular files, symbolic links or git-directories", path);
610 namelen = strlen(path);
611 if (S_ISDIR(st_mode)) {
612 while (namelen && path[namelen-1] == '/')
613 namelen--;
615 size = cache_entry_size(namelen);
616 ce = xcalloc(1, size);
617 memcpy(ce->name, path, namelen);
618 ce->ce_namelen = namelen;
619 if (!intent_only)
620 fill_stat_cache_info(ce, st);
621 else
622 ce->ce_flags |= CE_INTENT_TO_ADD;
624 if (trust_executable_bit && has_symlinks)
625 ce->ce_mode = create_ce_mode(st_mode);
626 else {
627 /* If there is an existing entry, pick the mode bits and type
628 * from it, otherwise assume unexecutable regular file.
630 struct cache_entry *ent;
631 int pos = index_name_pos_also_unmerged(istate, path, namelen);
633 ent = (0 <= pos) ? istate->cache[pos] : NULL;
634 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
637 /* When core.ignorecase=true, determine if a directory of the same name but differing
638 * case already exists within the Git repository. If it does, ensure the directory
639 * case of the file being added to the repository matches (is folded into) the existing
640 * entry's directory case.
642 if (ignore_case) {
643 const char *startPtr = ce->name;
644 const char *ptr = startPtr;
645 while (*ptr) {
646 while (*ptr && *ptr != '/')
647 ++ptr;
648 if (*ptr == '/') {
649 struct cache_entry *foundce;
650 ++ptr;
651 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
652 if (foundce) {
653 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
654 startPtr = ptr;
660 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
661 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
662 /* Nothing changed, really */
663 free(ce);
664 if (!S_ISGITLINK(alias->ce_mode))
665 ce_mark_uptodate(alias);
666 alias->ce_flags |= CE_ADDED;
667 return 0;
669 if (!intent_only) {
670 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
671 return error("unable to index file %s", path);
672 } else
673 set_object_name_for_intent_to_add_entry(ce);
675 if (ignore_case && alias && different_name(ce, alias))
676 ce = create_alias_ce(ce, alias);
677 ce->ce_flags |= CE_ADDED;
679 /* It was suspected to be racily clean, but it turns out to be Ok */
680 was_same = (alias &&
681 !ce_stage(alias) &&
682 !hashcmp(alias->sha1, ce->sha1) &&
683 ce->ce_mode == alias->ce_mode);
685 if (pretend)
687 else if (add_index_entry(istate, ce, add_option))
688 return error("unable to add %s to index",path);
689 if (verbose && !was_same)
690 printf("add '%s'\n", path);
691 return 0;
694 int add_file_to_index(struct index_state *istate, const char *path, int flags)
696 struct stat st;
697 if (lstat(path, &st))
698 die_errno("unable to stat '%s'", path);
699 return add_to_index(istate, path, &st, flags);
702 struct cache_entry *make_cache_entry(unsigned int mode,
703 const unsigned char *sha1, const char *path, int stage,
704 unsigned int refresh_options)
706 int size, len;
707 struct cache_entry *ce;
709 if (!verify_path(path)) {
710 error("Invalid path '%s'", path);
711 return NULL;
714 len = strlen(path);
715 size = cache_entry_size(len);
716 ce = xcalloc(1, size);
718 hashcpy(ce->sha1, sha1);
719 memcpy(ce->name, path, len);
720 ce->ce_flags = create_ce_flags(stage);
721 ce->ce_namelen = len;
722 ce->ce_mode = create_ce_mode(mode);
724 return refresh_cache_entry(ce, refresh_options);
727 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
729 int len = ce_namelen(a);
730 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
734 * We fundamentally don't like some paths: we don't want
735 * dot or dot-dot anywhere, and for obvious reasons don't
736 * want to recurse into ".git" either.
738 * Also, we don't want double slashes or slashes at the
739 * end that can make pathnames ambiguous.
741 static int verify_dotfile(const char *rest)
744 * The first character was '.', but that
745 * has already been discarded, we now test
746 * the rest.
749 /* "." is not allowed */
750 if (*rest == '\0' || is_dir_sep(*rest))
751 return 0;
753 switch (*rest) {
755 * ".git" followed by NUL or slash is bad. This
756 * shares the path end test with the ".." case.
758 case 'g':
759 if (rest[1] != 'i')
760 break;
761 if (rest[2] != 't')
762 break;
763 rest += 2;
764 /* fallthrough */
765 case '.':
766 if (rest[1] == '\0' || is_dir_sep(rest[1]))
767 return 0;
769 return 1;
772 int verify_path(const char *path)
774 char c;
776 if (has_dos_drive_prefix(path))
777 return 0;
779 goto inside;
780 for (;;) {
781 if (!c)
782 return 1;
783 if (is_dir_sep(c)) {
784 inside:
785 c = *path++;
786 if ((c == '.' && !verify_dotfile(path)) ||
787 is_dir_sep(c) || c == '\0')
788 return 0;
790 c = *path++;
795 * Do we have another file that has the beginning components being a
796 * proper superset of the name we're trying to add?
798 static int has_file_name(struct index_state *istate,
799 const struct cache_entry *ce, int pos, int ok_to_replace)
801 int retval = 0;
802 int len = ce_namelen(ce);
803 int stage = ce_stage(ce);
804 const char *name = ce->name;
806 while (pos < istate->cache_nr) {
807 struct cache_entry *p = istate->cache[pos++];
809 if (len >= ce_namelen(p))
810 break;
811 if (memcmp(name, p->name, len))
812 break;
813 if (ce_stage(p) != stage)
814 continue;
815 if (p->name[len] != '/')
816 continue;
817 if (p->ce_flags & CE_REMOVE)
818 continue;
819 retval = -1;
820 if (!ok_to_replace)
821 break;
822 remove_index_entry_at(istate, --pos);
824 return retval;
828 * Do we have another file with a pathname that is a proper
829 * subset of the name we're trying to add?
831 static int has_dir_name(struct index_state *istate,
832 const struct cache_entry *ce, int pos, int ok_to_replace)
834 int retval = 0;
835 int stage = ce_stage(ce);
836 const char *name = ce->name;
837 const char *slash = name + ce_namelen(ce);
839 for (;;) {
840 int len;
842 for (;;) {
843 if (*--slash == '/')
844 break;
845 if (slash <= ce->name)
846 return retval;
848 len = slash - name;
850 pos = index_name_stage_pos(istate, name, len, stage);
851 if (pos >= 0) {
853 * Found one, but not so fast. This could
854 * be a marker that says "I was here, but
855 * I am being removed". Such an entry is
856 * not a part of the resulting tree, and
857 * it is Ok to have a directory at the same
858 * path.
860 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
861 retval = -1;
862 if (!ok_to_replace)
863 break;
864 remove_index_entry_at(istate, pos);
865 continue;
868 else
869 pos = -pos-1;
872 * Trivial optimization: if we find an entry that
873 * already matches the sub-directory, then we know
874 * we're ok, and we can exit.
876 while (pos < istate->cache_nr) {
877 struct cache_entry *p = istate->cache[pos];
878 if ((ce_namelen(p) <= len) ||
879 (p->name[len] != '/') ||
880 memcmp(p->name, name, len))
881 break; /* not our subdirectory */
882 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
884 * p is at the same stage as our entry, and
885 * is a subdirectory of what we are looking
886 * at, so we cannot have conflicts at our
887 * level or anything shorter.
889 return retval;
890 pos++;
893 return retval;
896 /* We may be in a situation where we already have path/file and path
897 * is being added, or we already have path and path/file is being
898 * added. Either one would result in a nonsense tree that has path
899 * twice when git-write-tree tries to write it out. Prevent it.
901 * If ok-to-replace is specified, we remove the conflicting entries
902 * from the cache so the caller should recompute the insert position.
903 * When this happens, we return non-zero.
905 static int check_file_directory_conflict(struct index_state *istate,
906 const struct cache_entry *ce,
907 int pos, int ok_to_replace)
909 int retval;
912 * When ce is an "I am going away" entry, we allow it to be added
914 if (ce->ce_flags & CE_REMOVE)
915 return 0;
918 * We check if the path is a sub-path of a subsequent pathname
919 * first, since removing those will not change the position
920 * in the array.
922 retval = has_file_name(istate, ce, pos, ok_to_replace);
925 * Then check if the path might have a clashing sub-directory
926 * before it.
928 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
931 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
933 int pos;
934 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
935 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
936 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
937 int new_only = option & ADD_CACHE_NEW_ONLY;
939 cache_tree_invalidate_path(istate->cache_tree, ce->name);
940 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
942 /* existing match? Just replace it. */
943 if (pos >= 0) {
944 if (!new_only)
945 replace_index_entry(istate, pos, ce);
946 return 0;
948 pos = -pos-1;
951 * Inserting a merged entry ("stage 0") into the index
952 * will always replace all non-merged entries..
954 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
955 while (ce_same_name(istate->cache[pos], ce)) {
956 ok_to_add = 1;
957 if (!remove_index_entry_at(istate, pos))
958 break;
962 if (!ok_to_add)
963 return -1;
964 if (!verify_path(ce->name))
965 return error("Invalid path '%s'", ce->name);
967 if (!skip_df_check &&
968 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
969 if (!ok_to_replace)
970 return error("'%s' appears as both a file and as a directory",
971 ce->name);
972 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
973 pos = -pos-1;
975 return pos + 1;
978 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
980 int pos;
982 if (option & ADD_CACHE_JUST_APPEND)
983 pos = istate->cache_nr;
984 else {
985 int ret;
986 ret = add_index_entry_with_check(istate, ce, option);
987 if (ret <= 0)
988 return ret;
989 pos = ret - 1;
992 /* Make sure the array is big enough .. */
993 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
995 /* Add it in.. */
996 istate->cache_nr++;
997 if (istate->cache_nr > pos + 1)
998 memmove(istate->cache + pos + 1,
999 istate->cache + pos,
1000 (istate->cache_nr - pos - 1) * sizeof(ce));
1001 set_index_entry(istate, pos, ce);
1002 istate->cache_changed = 1;
1003 return 0;
1007 * "refresh" does not calculate a new sha1 file or bring the
1008 * cache up-to-date for mode/content changes. But what it
1009 * _does_ do is to "re-match" the stat information of a file
1010 * with the cache, so that you can refresh the cache for a
1011 * file that hasn't been changed but where the stat entry is
1012 * out of date.
1014 * For example, you'd want to do this after doing a "git-read-tree",
1015 * to link up the stat cache details with the proper files.
1017 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1018 struct cache_entry *ce,
1019 unsigned int options, int *err,
1020 int *changed_ret)
1022 struct stat st;
1023 struct cache_entry *updated;
1024 int changed, size;
1025 int refresh = options & CE_MATCH_REFRESH;
1026 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1027 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1028 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1030 if (!refresh || ce_uptodate(ce))
1031 return ce;
1034 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1035 * that the change to the work tree does not matter and told
1036 * us not to worry.
1038 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1039 ce_mark_uptodate(ce);
1040 return ce;
1042 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1043 ce_mark_uptodate(ce);
1044 return ce;
1047 if (lstat(ce->name, &st) < 0) {
1048 if (ignore_missing && errno == ENOENT)
1049 return ce;
1050 if (err)
1051 *err = errno;
1052 return NULL;
1055 changed = ie_match_stat(istate, ce, &st, options);
1056 if (changed_ret)
1057 *changed_ret = changed;
1058 if (!changed) {
1060 * The path is unchanged. If we were told to ignore
1061 * valid bit, then we did the actual stat check and
1062 * found that the entry is unmodified. If the entry
1063 * is not marked VALID, this is the place to mark it
1064 * valid again, under "assume unchanged" mode.
1066 if (ignore_valid && assume_unchanged &&
1067 !(ce->ce_flags & CE_VALID))
1068 ; /* mark this one VALID again */
1069 else {
1071 * We do not mark the index itself "modified"
1072 * because CE_UPTODATE flag is in-core only;
1073 * we are not going to write this change out.
1075 if (!S_ISGITLINK(ce->ce_mode))
1076 ce_mark_uptodate(ce);
1077 return ce;
1081 if (ie_modified(istate, ce, &st, options)) {
1082 if (err)
1083 *err = EINVAL;
1084 return NULL;
1087 size = ce_size(ce);
1088 updated = xmalloc(size);
1089 memcpy(updated, ce, size);
1090 fill_stat_cache_info(updated, &st);
1092 * If ignore_valid is not set, we should leave CE_VALID bit
1093 * alone. Otherwise, paths marked with --no-assume-unchanged
1094 * (i.e. things to be edited) will reacquire CE_VALID bit
1095 * automatically, which is not really what we want.
1097 if (!ignore_valid && assume_unchanged &&
1098 !(ce->ce_flags & CE_VALID))
1099 updated->ce_flags &= ~CE_VALID;
1101 return updated;
1104 static void show_file(const char * fmt, const char * name, int in_porcelain,
1105 int * first, const char *header_msg)
1107 if (in_porcelain && *first && header_msg) {
1108 printf("%s\n", header_msg);
1109 *first = 0;
1111 printf(fmt, name);
1114 int refresh_index(struct index_state *istate, unsigned int flags,
1115 const struct pathspec *pathspec,
1116 char *seen, const char *header_msg)
1118 int i;
1119 int has_errors = 0;
1120 int really = (flags & REFRESH_REALLY) != 0;
1121 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1122 int quiet = (flags & REFRESH_QUIET) != 0;
1123 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1124 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1125 int first = 1;
1126 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1127 unsigned int options = (CE_MATCH_REFRESH |
1128 (really ? CE_MATCH_IGNORE_VALID : 0) |
1129 (not_new ? CE_MATCH_IGNORE_MISSING : 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 && !ce_path_match(ce, pathspec, seen))
1152 filtered = 1;
1154 if (ce_stage(ce)) {
1155 while ((i < istate->cache_nr) &&
1156 ! strcmp(istate->cache[i]->name, ce->name))
1157 i++;
1158 i--;
1159 if (allow_unmerged)
1160 continue;
1161 if (!filtered)
1162 show_file(unmerged_fmt, ce->name, in_porcelain,
1163 &first, header_msg);
1164 has_errors = 1;
1165 continue;
1168 if (filtered)
1169 continue;
1171 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1172 if (new == ce)
1173 continue;
1174 if (!new) {
1175 const char *fmt;
1177 if (really && cache_errno == EINVAL) {
1178 /* If we are doing --really-refresh that
1179 * means the index is not valid anymore.
1181 ce->ce_flags &= ~CE_VALID;
1182 istate->cache_changed = 1;
1184 if (quiet)
1185 continue;
1187 if (cache_errno == ENOENT)
1188 fmt = deleted_fmt;
1189 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1190 fmt = added_fmt; /* must be before other checks */
1191 else if (changed & TYPE_CHANGED)
1192 fmt = typechange_fmt;
1193 else
1194 fmt = modified_fmt;
1195 show_file(fmt,
1196 ce->name, in_porcelain, &first, header_msg);
1197 has_errors = 1;
1198 continue;
1201 replace_index_entry(istate, i, new);
1203 return has_errors;
1206 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1207 unsigned int options)
1209 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1213 /*****************************************************************
1214 * Index File I/O
1215 *****************************************************************/
1217 #define INDEX_FORMAT_DEFAULT 3
1219 static int index_format_config(const char *var, const char *value, void *cb)
1221 unsigned int *version = cb;
1222 if (!strcmp(var, "index.version")) {
1223 *version = git_config_int(var, value);
1224 return 0;
1226 return 1;
1229 static unsigned int get_index_format_default(void)
1231 char *envversion = getenv("GIT_INDEX_VERSION");
1232 char *endp;
1233 unsigned int version = INDEX_FORMAT_DEFAULT;
1235 if (!envversion) {
1236 git_config(index_format_config, &version);
1237 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1238 warning(_("index.version set, but the value is invalid.\n"
1239 "Using version %i"), INDEX_FORMAT_DEFAULT);
1240 return INDEX_FORMAT_DEFAULT;
1242 return version;
1245 version = strtoul(envversion, &endp, 10);
1246 if (*endp ||
1247 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1248 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1249 "Using version %i"), INDEX_FORMAT_DEFAULT);
1250 version = INDEX_FORMAT_DEFAULT;
1252 return version;
1256 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1257 * Again - this is just a (very strong in practice) heuristic that
1258 * the inode hasn't changed.
1260 * We save the fields in big-endian order to allow using the
1261 * index file over NFS transparently.
1263 struct ondisk_cache_entry {
1264 struct cache_time ctime;
1265 struct cache_time mtime;
1266 uint32_t dev;
1267 uint32_t ino;
1268 uint32_t mode;
1269 uint32_t uid;
1270 uint32_t gid;
1271 uint32_t size;
1272 unsigned char sha1[20];
1273 uint16_t flags;
1274 char name[FLEX_ARRAY]; /* more */
1278 * This struct is used when CE_EXTENDED bit is 1
1279 * The struct must match ondisk_cache_entry exactly from
1280 * ctime till flags
1282 struct ondisk_cache_entry_extended {
1283 struct cache_time ctime;
1284 struct cache_time mtime;
1285 uint32_t dev;
1286 uint32_t ino;
1287 uint32_t mode;
1288 uint32_t uid;
1289 uint32_t gid;
1290 uint32_t size;
1291 unsigned char sha1[20];
1292 uint16_t flags;
1293 uint16_t flags2;
1294 char name[FLEX_ARRAY]; /* more */
1297 /* These are only used for v3 or lower */
1298 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1299 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1300 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1301 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1302 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1303 ondisk_cache_entry_size(ce_namelen(ce)))
1305 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1307 git_SHA_CTX c;
1308 unsigned char sha1[20];
1309 int hdr_version;
1311 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1312 return error("bad signature");
1313 hdr_version = ntohl(hdr->hdr_version);
1314 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1315 return error("bad index version %d", hdr_version);
1316 git_SHA1_Init(&c);
1317 git_SHA1_Update(&c, hdr, size - 20);
1318 git_SHA1_Final(sha1, &c);
1319 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1320 return error("bad index file sha1 signature");
1321 return 0;
1324 static int read_index_extension(struct index_state *istate,
1325 const char *ext, void *data, unsigned long sz)
1327 switch (CACHE_EXT(ext)) {
1328 case CACHE_EXT_TREE:
1329 istate->cache_tree = cache_tree_read(data, sz);
1330 break;
1331 case CACHE_EXT_RESOLVE_UNDO:
1332 istate->resolve_undo = resolve_undo_read(data, sz);
1333 break;
1334 default:
1335 if (*ext < 'A' || 'Z' < *ext)
1336 return error("index uses %.4s extension, which we do not understand",
1337 ext);
1338 fprintf(stderr, "ignoring %.4s extension\n", ext);
1339 break;
1341 return 0;
1344 int read_index(struct index_state *istate)
1346 return read_index_from(istate, get_index_file());
1349 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1350 unsigned int flags,
1351 const char *name,
1352 size_t len)
1354 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1356 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1357 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1358 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1359 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1360 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1361 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1362 ce->ce_mode = get_be32(&ondisk->mode);
1363 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1364 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1365 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1366 ce->ce_flags = flags & ~CE_NAMEMASK;
1367 ce->ce_namelen = len;
1368 hashcpy(ce->sha1, ondisk->sha1);
1369 memcpy(ce->name, name, len);
1370 ce->name[len] = '\0';
1371 return ce;
1375 * Adjacent cache entries tend to share the leading paths, so it makes
1376 * sense to only store the differences in later entries. In the v4
1377 * on-disk format of the index, each on-disk cache entry stores the
1378 * number of bytes to be stripped from the end of the previous name,
1379 * and the bytes to append to the result, to come up with its name.
1381 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1383 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1384 size_t len = decode_varint(&cp);
1386 if (name->len < len)
1387 die("malformed name field in the index");
1388 strbuf_remove(name, name->len - len, len);
1389 for (ep = cp; *ep; ep++)
1390 ; /* find the end */
1391 strbuf_add(name, cp, ep - cp);
1392 return (const char *)ep + 1 - cp_;
1395 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1396 unsigned long *ent_size,
1397 struct strbuf *previous_name)
1399 struct cache_entry *ce;
1400 size_t len;
1401 const char *name;
1402 unsigned int flags;
1404 /* On-disk flags are just 16 bits */
1405 flags = get_be16(&ondisk->flags);
1406 len = flags & CE_NAMEMASK;
1408 if (flags & CE_EXTENDED) {
1409 struct ondisk_cache_entry_extended *ondisk2;
1410 int extended_flags;
1411 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1412 extended_flags = get_be16(&ondisk2->flags2) << 16;
1413 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1414 if (extended_flags & ~CE_EXTENDED_FLAGS)
1415 die("Unknown index entry format %08x", extended_flags);
1416 flags |= extended_flags;
1417 name = ondisk2->name;
1419 else
1420 name = ondisk->name;
1422 if (!previous_name) {
1423 /* v3 and earlier */
1424 if (len == CE_NAMEMASK)
1425 len = strlen(name);
1426 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1428 *ent_size = ondisk_ce_size(ce);
1429 } else {
1430 unsigned long consumed;
1431 consumed = expand_name_field(previous_name, name);
1432 ce = cache_entry_from_ondisk(ondisk, flags,
1433 previous_name->buf,
1434 previous_name->len);
1436 *ent_size = (name - ((char *)ondisk)) + consumed;
1438 return ce;
1441 /* remember to discard_cache() before reading a different cache! */
1442 int read_index_from(struct index_state *istate, const char *path)
1444 int fd, i;
1445 struct stat st;
1446 unsigned long src_offset;
1447 struct cache_header *hdr;
1448 void *mmap;
1449 size_t mmap_size;
1450 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1452 if (istate->initialized)
1453 return istate->cache_nr;
1455 istate->timestamp.sec = 0;
1456 istate->timestamp.nsec = 0;
1457 fd = open(path, O_RDONLY);
1458 if (fd < 0) {
1459 if (errno == ENOENT)
1460 return 0;
1461 die_errno("index file open failed");
1464 if (fstat(fd, &st))
1465 die_errno("cannot stat the open index");
1467 mmap_size = xsize_t(st.st_size);
1468 if (mmap_size < sizeof(struct cache_header) + 20)
1469 die("index file smaller than expected");
1471 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1472 if (mmap == MAP_FAILED)
1473 die_errno("unable to map index file");
1474 close(fd);
1476 hdr = mmap;
1477 if (verify_hdr(hdr, mmap_size) < 0)
1478 goto unmap;
1480 hashcpy(istate->sha1, (unsigned char *)hdr + mmap_size - 20);
1481 istate->version = ntohl(hdr->hdr_version);
1482 istate->cache_nr = ntohl(hdr->hdr_entries);
1483 istate->cache_alloc = alloc_nr(istate->cache_nr);
1484 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1485 istate->initialized = 1;
1487 if (istate->version == 4)
1488 previous_name = &previous_name_buf;
1489 else
1490 previous_name = NULL;
1492 src_offset = sizeof(*hdr);
1493 for (i = 0; i < istate->cache_nr; i++) {
1494 struct ondisk_cache_entry *disk_ce;
1495 struct cache_entry *ce;
1496 unsigned long consumed;
1498 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1499 ce = create_from_disk(disk_ce, &consumed, previous_name);
1500 set_index_entry(istate, i, ce);
1502 src_offset += consumed;
1504 strbuf_release(&previous_name_buf);
1505 istate->timestamp.sec = st.st_mtime;
1506 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1508 while (src_offset <= mmap_size - 20 - 8) {
1509 /* After an array of active_nr index entries,
1510 * there can be arbitrary number of extended
1511 * sections, each of which is prefixed with
1512 * extension name (4-byte) and section length
1513 * in 4-byte network byte order.
1515 uint32_t extsize;
1516 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1517 extsize = ntohl(extsize);
1518 if (read_index_extension(istate,
1519 (const char *) mmap + src_offset,
1520 (char *) mmap + src_offset + 8,
1521 extsize) < 0)
1522 goto unmap;
1523 src_offset += 8;
1524 src_offset += extsize;
1526 munmap(mmap, mmap_size);
1527 return istate->cache_nr;
1529 unmap:
1530 munmap(mmap, mmap_size);
1531 die("index file corrupt");
1534 int is_index_unborn(struct index_state *istate)
1536 return (!istate->cache_nr && !istate->timestamp.sec);
1539 int discard_index(struct index_state *istate)
1541 int i;
1543 for (i = 0; i < istate->cache_nr; i++)
1544 free(istate->cache[i]);
1545 resolve_undo_clear_index(istate);
1546 istate->cache_nr = 0;
1547 istate->cache_changed = 0;
1548 istate->timestamp.sec = 0;
1549 istate->timestamp.nsec = 0;
1550 free_name_hash(istate);
1551 cache_tree_free(&(istate->cache_tree));
1552 istate->initialized = 0;
1553 free(istate->cache);
1554 istate->cache = NULL;
1555 istate->cache_alloc = 0;
1556 return 0;
1559 int unmerged_index(const struct index_state *istate)
1561 int i;
1562 for (i = 0; i < istate->cache_nr; i++) {
1563 if (ce_stage(istate->cache[i]))
1564 return 1;
1566 return 0;
1569 #define WRITE_BUFFER_SIZE 8192
1570 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1571 static unsigned long write_buffer_len;
1573 static int ce_write_flush(git_SHA_CTX *context, int fd)
1575 unsigned int buffered = write_buffer_len;
1576 if (buffered) {
1577 git_SHA1_Update(context, write_buffer, buffered);
1578 if (write_in_full(fd, write_buffer, buffered) != buffered)
1579 return -1;
1580 write_buffer_len = 0;
1582 return 0;
1585 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1587 while (len) {
1588 unsigned int buffered = write_buffer_len;
1589 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1590 if (partial > len)
1591 partial = len;
1592 memcpy(write_buffer + buffered, data, partial);
1593 buffered += partial;
1594 if (buffered == WRITE_BUFFER_SIZE) {
1595 write_buffer_len = buffered;
1596 if (ce_write_flush(context, fd))
1597 return -1;
1598 buffered = 0;
1600 write_buffer_len = buffered;
1601 len -= partial;
1602 data = (char *) data + partial;
1604 return 0;
1607 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1608 unsigned int ext, unsigned int sz)
1610 ext = htonl(ext);
1611 sz = htonl(sz);
1612 return ((ce_write(context, fd, &ext, 4) < 0) ||
1613 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1616 static int ce_flush(git_SHA_CTX *context, int fd)
1618 unsigned int left = write_buffer_len;
1620 if (left) {
1621 write_buffer_len = 0;
1622 git_SHA1_Update(context, write_buffer, left);
1625 /* Flush first if not enough space for SHA1 signature */
1626 if (left + 20 > WRITE_BUFFER_SIZE) {
1627 if (write_in_full(fd, write_buffer, left) != left)
1628 return -1;
1629 left = 0;
1632 /* Append the SHA1 signature at the end */
1633 git_SHA1_Final(write_buffer + left, context);
1634 left += 20;
1635 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1638 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1641 * The only thing we care about in this function is to smudge the
1642 * falsely clean entry due to touch-update-touch race, so we leave
1643 * everything else as they are. We are called for entries whose
1644 * ce_stat_data.sd_mtime match the index file mtime.
1646 * Note that this actually does not do much for gitlinks, for
1647 * which ce_match_stat_basic() always goes to the actual
1648 * contents. The caller checks with is_racy_timestamp() which
1649 * always says "no" for gitlinks, so we are not called for them ;-)
1651 struct stat st;
1653 if (lstat(ce->name, &st) < 0)
1654 return;
1655 if (ce_match_stat_basic(ce, &st))
1656 return;
1657 if (ce_modified_check_fs(ce, &st)) {
1658 /* This is "racily clean"; smudge it. Note that this
1659 * is a tricky code. At first glance, it may appear
1660 * that it can break with this sequence:
1662 * $ echo xyzzy >frotz
1663 * $ git-update-index --add frotz
1664 * $ : >frotz
1665 * $ sleep 3
1666 * $ echo filfre >nitfol
1667 * $ git-update-index --add nitfol
1669 * but it does not. When the second update-index runs,
1670 * it notices that the entry "frotz" has the same timestamp
1671 * as index, and if we were to smudge it by resetting its
1672 * size to zero here, then the object name recorded
1673 * in index is the 6-byte file but the cached stat information
1674 * becomes zero --- which would then match what we would
1675 * obtain from the filesystem next time we stat("frotz").
1677 * However, the second update-index, before calling
1678 * this function, notices that the cached size is 6
1679 * bytes and what is on the filesystem is an empty
1680 * file, and never calls us, so the cached size information
1681 * for "frotz" stays 6 which does not match the filesystem.
1683 ce->ce_stat_data.sd_size = 0;
1687 /* Copy miscellaneous fields but not the name */
1688 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1689 struct cache_entry *ce)
1691 short flags;
1693 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1694 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1695 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1696 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1697 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1698 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1699 ondisk->mode = htonl(ce->ce_mode);
1700 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1701 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1702 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1703 hashcpy(ondisk->sha1, ce->sha1);
1705 flags = ce->ce_flags;
1706 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1707 ondisk->flags = htons(flags);
1708 if (ce->ce_flags & CE_EXTENDED) {
1709 struct ondisk_cache_entry_extended *ondisk2;
1710 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1711 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1712 return ondisk2->name;
1714 else {
1715 return ondisk->name;
1719 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1720 struct strbuf *previous_name)
1722 int size;
1723 struct ondisk_cache_entry *ondisk;
1724 char *name;
1725 int result;
1727 if (!previous_name) {
1728 size = ondisk_ce_size(ce);
1729 ondisk = xcalloc(1, size);
1730 name = copy_cache_entry_to_ondisk(ondisk, ce);
1731 memcpy(name, ce->name, ce_namelen(ce));
1732 } else {
1733 int common, to_remove, prefix_size;
1734 unsigned char to_remove_vi[16];
1735 for (common = 0;
1736 (ce->name[common] &&
1737 common < previous_name->len &&
1738 ce->name[common] == previous_name->buf[common]);
1739 common++)
1740 ; /* still matching */
1741 to_remove = previous_name->len - common;
1742 prefix_size = encode_varint(to_remove, to_remove_vi);
1744 if (ce->ce_flags & CE_EXTENDED)
1745 size = offsetof(struct ondisk_cache_entry_extended, name);
1746 else
1747 size = offsetof(struct ondisk_cache_entry, name);
1748 size += prefix_size + (ce_namelen(ce) - common + 1);
1750 ondisk = xcalloc(1, size);
1751 name = copy_cache_entry_to_ondisk(ondisk, ce);
1752 memcpy(name, to_remove_vi, prefix_size);
1753 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1755 strbuf_splice(previous_name, common, to_remove,
1756 ce->name + common, ce_namelen(ce) - common);
1759 result = ce_write(c, fd, ondisk, size);
1760 free(ondisk);
1761 return result;
1765 * This function verifies if index_state has the correct sha1 of the
1766 * index file. Don't die if we have any other failure, just return 0.
1768 static int verify_index_from(const struct index_state *istate, const char *path)
1770 int fd;
1771 ssize_t n;
1772 struct stat st;
1773 unsigned char sha1[20];
1775 if (!istate->initialized)
1776 return 0;
1778 fd = open(path, O_RDONLY);
1779 if (fd < 0)
1780 return 0;
1782 if (fstat(fd, &st))
1783 goto out;
1785 if (st.st_size < sizeof(struct cache_header) + 20)
1786 goto out;
1788 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1789 if (n != 20)
1790 goto out;
1792 if (hashcmp(istate->sha1, sha1))
1793 goto out;
1795 close(fd);
1796 return 1;
1798 out:
1799 close(fd);
1800 return 0;
1803 static int verify_index(const struct index_state *istate)
1805 return verify_index_from(istate, get_index_file());
1808 static int has_racy_timestamp(struct index_state *istate)
1810 int entries = istate->cache_nr;
1811 int i;
1813 for (i = 0; i < entries; i++) {
1814 struct cache_entry *ce = istate->cache[i];
1815 if (is_racy_timestamp(istate, ce))
1816 return 1;
1818 return 0;
1822 * Opportunistically update the index but do not complain if we can't
1824 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1826 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1827 verify_index(istate) && !write_index(istate, lockfile->fd))
1828 commit_locked_index(lockfile);
1829 else
1830 rollback_lock_file(lockfile);
1833 int write_index(struct index_state *istate, int newfd)
1835 git_SHA_CTX c;
1836 struct cache_header hdr;
1837 int i, err, removed, extended, hdr_version;
1838 struct cache_entry **cache = istate->cache;
1839 int entries = istate->cache_nr;
1840 struct stat st;
1841 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1843 for (i = removed = extended = 0; i < entries; i++) {
1844 if (cache[i]->ce_flags & CE_REMOVE)
1845 removed++;
1847 /* reduce extended entries if possible */
1848 cache[i]->ce_flags &= ~CE_EXTENDED;
1849 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1850 extended++;
1851 cache[i]->ce_flags |= CE_EXTENDED;
1855 if (!istate->version)
1856 istate->version = get_index_format_default();
1858 /* demote version 3 to version 2 when the latter suffices */
1859 if (istate->version == 3 || istate->version == 2)
1860 istate->version = extended ? 3 : 2;
1862 hdr_version = istate->version;
1864 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1865 hdr.hdr_version = htonl(hdr_version);
1866 hdr.hdr_entries = htonl(entries - removed);
1868 git_SHA1_Init(&c);
1869 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1870 return -1;
1872 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1873 for (i = 0; i < entries; i++) {
1874 struct cache_entry *ce = cache[i];
1875 if (ce->ce_flags & CE_REMOVE)
1876 continue;
1877 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1878 ce_smudge_racily_clean_entry(ce);
1879 if (is_null_sha1(ce->sha1)) {
1880 static const char msg[] = "cache entry has null sha1: %s";
1881 static int allow = -1;
1883 if (allow < 0)
1884 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1885 if (allow)
1886 warning(msg, ce->name);
1887 else
1888 return error(msg, ce->name);
1890 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1891 return -1;
1893 strbuf_release(&previous_name_buf);
1895 /* Write extension data here */
1896 if (istate->cache_tree) {
1897 struct strbuf sb = STRBUF_INIT;
1899 cache_tree_write(&sb, istate->cache_tree);
1900 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1901 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1902 strbuf_release(&sb);
1903 if (err)
1904 return -1;
1906 if (istate->resolve_undo) {
1907 struct strbuf sb = STRBUF_INIT;
1909 resolve_undo_write(&sb, istate->resolve_undo);
1910 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1911 sb.len) < 0
1912 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1913 strbuf_release(&sb);
1914 if (err)
1915 return -1;
1918 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1919 return -1;
1920 istate->timestamp.sec = (unsigned int)st.st_mtime;
1921 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1922 return 0;
1926 * Read the index file that is potentially unmerged into given
1927 * index_state, dropping any unmerged entries. Returns true if
1928 * the index is unmerged. Callers who want to refuse to work
1929 * from an unmerged state can call this and check its return value,
1930 * instead of calling read_cache().
1932 int read_index_unmerged(struct index_state *istate)
1934 int i;
1935 int unmerged = 0;
1937 read_index(istate);
1938 for (i = 0; i < istate->cache_nr; i++) {
1939 struct cache_entry *ce = istate->cache[i];
1940 struct cache_entry *new_ce;
1941 int size, len;
1943 if (!ce_stage(ce))
1944 continue;
1945 unmerged = 1;
1946 len = ce_namelen(ce);
1947 size = cache_entry_size(len);
1948 new_ce = xcalloc(1, size);
1949 memcpy(new_ce->name, ce->name, len);
1950 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1951 new_ce->ce_namelen = len;
1952 new_ce->ce_mode = ce->ce_mode;
1953 if (add_index_entry(istate, new_ce, 0))
1954 return error("%s: cannot drop to stage #0",
1955 new_ce->name);
1956 i = index_name_pos(istate, new_ce->name, len);
1958 return unmerged;
1962 * Returns 1 if the path is an "other" path with respect to
1963 * the index; that is, the path is not mentioned in the index at all,
1964 * either as a file, a directory with some files in the index,
1965 * or as an unmerged entry.
1967 * We helpfully remove a trailing "/" from directories so that
1968 * the output of read_directory can be used as-is.
1970 int index_name_is_other(const struct index_state *istate, const char *name,
1971 int namelen)
1973 int pos;
1974 if (namelen && name[namelen - 1] == '/')
1975 namelen--;
1976 pos = index_name_pos(istate, name, namelen);
1977 if (0 <= pos)
1978 return 0; /* exact match */
1979 pos = -pos - 1;
1980 if (pos < istate->cache_nr) {
1981 struct cache_entry *ce = istate->cache[pos];
1982 if (ce_namelen(ce) == namelen &&
1983 !memcmp(ce->name, name, namelen))
1984 return 0; /* Yup, this one exists unmerged */
1986 return 1;
1989 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
1991 int pos, len;
1992 unsigned long sz;
1993 enum object_type type;
1994 void *data;
1996 len = strlen(path);
1997 pos = index_name_pos(istate, path, len);
1998 if (pos < 0) {
2000 * We might be in the middle of a merge, in which
2001 * case we would read stage #2 (ours).
2003 int i;
2004 for (i = -pos - 1;
2005 (pos < 0 && i < istate->cache_nr &&
2006 !strcmp(istate->cache[i]->name, path));
2007 i++)
2008 if (ce_stage(istate->cache[i]) == 2)
2009 pos = i;
2011 if (pos < 0)
2012 return NULL;
2013 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2014 if (!data || type != OBJ_BLOB) {
2015 free(data);
2016 return NULL;
2018 if (size)
2019 *size = sz;
2020 return data;
2023 void stat_validity_clear(struct stat_validity *sv)
2025 free(sv->sd);
2026 sv->sd = NULL;
2029 int stat_validity_check(struct stat_validity *sv, const char *path)
2031 struct stat st;
2033 if (stat(path, &st) < 0)
2034 return sv->sd == NULL;
2035 if (!sv->sd)
2036 return 0;
2037 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2040 void stat_validity_update(struct stat_validity *sv, int fd)
2042 struct stat st;
2044 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2045 stat_validity_clear(sv);
2046 else {
2047 if (!sv->sd)
2048 sv->sd = xcalloc(1, sizeof(struct stat_data));
2049 fill_stat_data(sv->sd, &st);