dumb-http: do not pass NULL path to parse_pack_index
[git.git] / read-cache.c
blob36a6f73fedc2fdb7c5dc2e1fb73ca5c5c76979c0
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,
20 unsigned int options);
22 /* Mask for the name length in ce_flags in the on-disk index */
24 #define CE_NAMEMASK (0x0fff)
26 /* Index extensions.
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
35 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
37 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
39 struct index_state the_index;
41 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
43 istate->cache[nr] = ce;
44 add_name_hash(istate, ce);
47 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
49 struct cache_entry *old = istate->cache[nr];
51 remove_name_hash(istate, old);
52 free(old);
53 set_index_entry(istate, nr, ce);
54 istate->cache_changed = 1;
57 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
59 struct cache_entry *old = istate->cache[nr], *new;
60 int namelen = strlen(new_name);
62 new = xmalloc(cache_entry_size(namelen));
63 copy_cache_entry(new, old);
64 new->ce_flags &= ~CE_HASHED;
65 new->ce_namelen = namelen;
66 memcpy(new->name, new_name, namelen + 1);
68 cache_tree_invalidate_path(istate->cache_tree, old->name);
69 remove_index_entry_at(istate, nr);
70 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
73 void fill_stat_data(struct stat_data *sd, struct stat *st)
75 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
76 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
77 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
78 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
79 sd->sd_dev = st->st_dev;
80 sd->sd_ino = st->st_ino;
81 sd->sd_uid = st->st_uid;
82 sd->sd_gid = st->st_gid;
83 sd->sd_size = st->st_size;
86 int match_stat_data(const struct stat_data *sd, struct stat *st)
88 int changed = 0;
90 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
91 changed |= MTIME_CHANGED;
92 if (trust_ctime && check_stat &&
93 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
94 changed |= CTIME_CHANGED;
96 #ifdef USE_NSEC
97 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
98 changed |= MTIME_CHANGED;
99 if (trust_ctime && check_stat &&
100 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
101 changed |= CTIME_CHANGED;
102 #endif
104 if (check_stat) {
105 if (sd->sd_uid != (unsigned int) st->st_uid ||
106 sd->sd_gid != (unsigned int) st->st_gid)
107 changed |= OWNER_CHANGED;
108 if (sd->sd_ino != (unsigned int) st->st_ino)
109 changed |= INODE_CHANGED;
112 #ifdef USE_STDEV
114 * st_dev breaks on network filesystems where different
115 * clients will have different views of what "device"
116 * the filesystem is on
118 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
119 changed |= INODE_CHANGED;
120 #endif
122 if (sd->sd_size != (unsigned int) st->st_size)
123 changed |= DATA_CHANGED;
125 return changed;
129 * This only updates the "non-critical" parts of the directory
130 * cache, ie the parts that aren't tracked by GIT, and only used
131 * to validate the cache.
133 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
135 fill_stat_data(&ce->ce_stat_data, st);
137 if (assume_unchanged)
138 ce->ce_flags |= CE_VALID;
140 if (S_ISREG(st->st_mode))
141 ce_mark_uptodate(ce);
144 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
146 int match = -1;
147 int fd = open(ce->name, O_RDONLY);
149 if (fd >= 0) {
150 unsigned char sha1[20];
151 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
152 match = hashcmp(sha1, ce->sha1);
153 /* index_fd() closed the file descriptor already */
155 return match;
158 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
160 int match = -1;
161 void *buffer;
162 unsigned long size;
163 enum object_type type;
164 struct strbuf sb = STRBUF_INIT;
166 if (strbuf_readlink(&sb, ce->name, expected_size))
167 return -1;
169 buffer = read_sha1_file(ce->sha1, &type, &size);
170 if (buffer) {
171 if (size == sb.len)
172 match = memcmp(buffer, sb.buf, size);
173 free(buffer);
175 strbuf_release(&sb);
176 return match;
179 static int ce_compare_gitlink(const struct cache_entry *ce)
181 unsigned char sha1[20];
184 * We don't actually require that the .git directory
185 * under GITLINK directory be a valid git directory. It
186 * might even be missing (in case nobody populated that
187 * sub-project).
189 * If so, we consider it always to match.
191 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
192 return 0;
193 return hashcmp(sha1, ce->sha1);
196 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
198 switch (st->st_mode & S_IFMT) {
199 case S_IFREG:
200 if (ce_compare_data(ce, st))
201 return DATA_CHANGED;
202 break;
203 case S_IFLNK:
204 if (ce_compare_link(ce, xsize_t(st->st_size)))
205 return DATA_CHANGED;
206 break;
207 case S_IFDIR:
208 if (S_ISGITLINK(ce->ce_mode))
209 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
210 default:
211 return TYPE_CHANGED;
213 return 0;
216 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
218 unsigned int changed = 0;
220 if (ce->ce_flags & CE_REMOVE)
221 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
223 switch (ce->ce_mode & S_IFMT) {
224 case S_IFREG:
225 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
226 /* We consider only the owner x bit to be relevant for
227 * "mode changes"
229 if (trust_executable_bit &&
230 (0100 & (ce->ce_mode ^ st->st_mode)))
231 changed |= MODE_CHANGED;
232 break;
233 case S_IFLNK:
234 if (!S_ISLNK(st->st_mode) &&
235 (has_symlinks || !S_ISREG(st->st_mode)))
236 changed |= TYPE_CHANGED;
237 break;
238 case S_IFGITLINK:
239 /* We ignore most of the st_xxx fields for gitlinks */
240 if (!S_ISDIR(st->st_mode))
241 changed |= TYPE_CHANGED;
242 else if (ce_compare_gitlink(ce))
243 changed |= DATA_CHANGED;
244 return changed;
245 default:
246 die("internal error: ce_mode is %o", ce->ce_mode);
249 changed |= match_stat_data(&ce->ce_stat_data, st);
251 /* Racily smudged entry? */
252 if (!ce->ce_stat_data.sd_size) {
253 if (!is_empty_blob_sha1(ce->sha1))
254 changed |= DATA_CHANGED;
257 return changed;
260 static int is_racy_timestamp(const struct index_state *istate,
261 const struct cache_entry *ce)
263 return (!S_ISGITLINK(ce->ce_mode) &&
264 istate->timestamp.sec &&
265 #ifdef USE_NSEC
266 /* nanosecond timestamped files can also be racy! */
267 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
268 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
269 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
270 #else
271 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
272 #endif
276 int ie_match_stat(const struct index_state *istate,
277 const struct cache_entry *ce, struct stat *st,
278 unsigned int options)
280 unsigned int changed;
281 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
282 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
283 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
286 * If it's marked as always valid in the index, it's
287 * valid whatever the checked-out copy says.
289 * skip-worktree has the same effect with higher precedence
291 if (!ignore_skip_worktree && ce_skip_worktree(ce))
292 return 0;
293 if (!ignore_valid && (ce->ce_flags & CE_VALID))
294 return 0;
297 * Intent-to-add entries have not been added, so the index entry
298 * by definition never matches what is in the work tree until it
299 * actually gets added.
301 if (ce->ce_flags & CE_INTENT_TO_ADD)
302 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
304 changed = ce_match_stat_basic(ce, st);
307 * Within 1 second of this sequence:
308 * echo xyzzy >file && git-update-index --add file
309 * running this command:
310 * echo frotz >file
311 * would give a falsely clean cache entry. The mtime and
312 * length match the cache, and other stat fields do not change.
314 * We could detect this at update-index time (the cache entry
315 * being registered/updated records the same time as "now")
316 * and delay the return from git-update-index, but that would
317 * effectively mean we can make at most one commit per second,
318 * which is not acceptable. Instead, we check cache entries
319 * whose mtime are the same as the index file timestamp more
320 * carefully than others.
322 if (!changed && is_racy_timestamp(istate, ce)) {
323 if (assume_racy_is_modified)
324 changed |= DATA_CHANGED;
325 else
326 changed |= ce_modified_check_fs(ce, st);
329 return changed;
332 int ie_modified(const struct index_state *istate,
333 const struct cache_entry *ce,
334 struct stat *st, unsigned int options)
336 int changed, changed_fs;
338 changed = ie_match_stat(istate, ce, st, options);
339 if (!changed)
340 return 0;
342 * If the mode or type has changed, there's no point in trying
343 * to refresh the entry - it's not going to match
345 if (changed & (MODE_CHANGED | TYPE_CHANGED))
346 return changed;
349 * Immediately after read-tree or update-index --cacheinfo,
350 * the length field is zero, as we have never even read the
351 * lstat(2) information once, and we cannot trust DATA_CHANGED
352 * returned by ie_match_stat() which in turn was returned by
353 * ce_match_stat_basic() to signal that the filesize of the
354 * blob changed. We have to actually go to the filesystem to
355 * see if the contents match, and if so, should answer "unchanged".
357 * The logic does not apply to gitlinks, as ce_match_stat_basic()
358 * already has checked the actual HEAD from the filesystem in the
359 * subproject. If ie_match_stat() already said it is different,
360 * then we know it is.
362 if ((changed & DATA_CHANGED) &&
363 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
364 return changed;
366 changed_fs = ce_modified_check_fs(ce, st);
367 if (changed_fs)
368 return changed | changed_fs;
369 return 0;
372 int base_name_compare(const char *name1, int len1, int mode1,
373 const char *name2, int len2, int mode2)
375 unsigned char c1, c2;
376 int len = len1 < len2 ? len1 : len2;
377 int cmp;
379 cmp = memcmp(name1, name2, len);
380 if (cmp)
381 return cmp;
382 c1 = name1[len];
383 c2 = name2[len];
384 if (!c1 && S_ISDIR(mode1))
385 c1 = '/';
386 if (!c2 && S_ISDIR(mode2))
387 c2 = '/';
388 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
392 * df_name_compare() is identical to base_name_compare(), except it
393 * compares conflicting directory/file entries as equal. Note that
394 * while a directory name compares as equal to a regular file, they
395 * then individually compare _differently_ to a filename that has
396 * a dot after the basename (because '\0' < '.' < '/').
398 * This is used by routines that want to traverse the git namespace
399 * but then handle conflicting entries together when possible.
401 int df_name_compare(const char *name1, int len1, int mode1,
402 const char *name2, int len2, int mode2)
404 int len = len1 < len2 ? len1 : len2, cmp;
405 unsigned char c1, c2;
407 cmp = memcmp(name1, name2, len);
408 if (cmp)
409 return cmp;
410 /* Directories and files compare equal (same length, same name) */
411 if (len1 == len2)
412 return 0;
413 c1 = name1[len];
414 if (!c1 && S_ISDIR(mode1))
415 c1 = '/';
416 c2 = name2[len];
417 if (!c2 && S_ISDIR(mode2))
418 c2 = '/';
419 if (c1 == '/' && !c2)
420 return 0;
421 if (c2 == '/' && !c1)
422 return 0;
423 return c1 - c2;
426 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
428 int len = len1 < len2 ? len1 : len2;
429 int cmp;
431 cmp = memcmp(name1, name2, len);
432 if (cmp)
433 return cmp;
434 if (len1 < len2)
435 return -1;
436 if (len1 > len2)
437 return 1;
439 if (stage1 < stage2)
440 return -1;
441 if (stage1 > stage2)
442 return 1;
443 return 0;
446 int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
448 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
451 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
453 int first, last;
455 first = 0;
456 last = istate->cache_nr;
457 while (last > first) {
458 int next = (last + first) >> 1;
459 struct cache_entry *ce = istate->cache[next];
460 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
461 if (!cmp)
462 return next;
463 if (cmp < 0) {
464 last = next;
465 continue;
467 first = next+1;
469 return -first-1;
472 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
474 return index_name_stage_pos(istate, name, namelen, 0);
477 /* Remove entry, return true if there are more entries to go.. */
478 int remove_index_entry_at(struct index_state *istate, int pos)
480 struct cache_entry *ce = istate->cache[pos];
482 record_resolve_undo(istate, ce);
483 remove_name_hash(istate, ce);
484 free(ce);
485 istate->cache_changed = 1;
486 istate->cache_nr--;
487 if (pos >= istate->cache_nr)
488 return 0;
489 memmove(istate->cache + pos,
490 istate->cache + pos + 1,
491 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
492 return 1;
496 * Remove all cache entries marked for removal, that is where
497 * CE_REMOVE is set in ce_flags. This is much more effective than
498 * calling remove_index_entry_at() for each entry to be removed.
500 void remove_marked_cache_entries(struct index_state *istate)
502 struct cache_entry **ce_array = istate->cache;
503 unsigned int i, j;
505 for (i = j = 0; i < istate->cache_nr; i++) {
506 if (ce_array[i]->ce_flags & CE_REMOVE) {
507 remove_name_hash(istate, ce_array[i]);
508 free(ce_array[i]);
510 else
511 ce_array[j++] = ce_array[i];
513 istate->cache_changed = 1;
514 istate->cache_nr = j;
517 int remove_file_from_index(struct index_state *istate, const char *path)
519 int pos = index_name_pos(istate, path, strlen(path));
520 if (pos < 0)
521 pos = -pos-1;
522 cache_tree_invalidate_path(istate->cache_tree, path);
523 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
524 remove_index_entry_at(istate, pos);
525 return 0;
528 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
530 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
533 static int index_name_pos_also_unmerged(struct index_state *istate,
534 const char *path, int namelen)
536 int pos = index_name_pos(istate, path, namelen);
537 struct cache_entry *ce;
539 if (pos >= 0)
540 return pos;
542 /* maybe unmerged? */
543 pos = -1 - pos;
544 if (pos >= istate->cache_nr ||
545 compare_name((ce = istate->cache[pos]), path, namelen))
546 return -1;
548 /* order of preference: stage 2, 1, 3 */
549 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
550 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
551 !compare_name(ce, path, namelen))
552 pos++;
553 return pos;
556 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
558 int len = ce_namelen(ce);
559 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
563 * If we add a filename that aliases in the cache, we will use the
564 * name that we already have - but we don't want to update the same
565 * alias twice, because that implies that there were actually two
566 * different files with aliasing names!
568 * So we use the CE_ADDED flag to verify that the alias was an old
569 * one before we accept it as
571 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
573 int len;
574 struct cache_entry *new;
576 if (alias->ce_flags & CE_ADDED)
577 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
579 /* Ok, create the new entry using the name of the existing alias */
580 len = ce_namelen(alias);
581 new = xcalloc(1, cache_entry_size(len));
582 memcpy(new->name, alias->name, len);
583 copy_cache_entry(new, ce);
584 free(ce);
585 return new;
588 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
590 unsigned char sha1[20];
591 if (write_sha1_file("", 0, blob_type, sha1))
592 die("cannot create an empty blob in the object database");
593 hashcpy(ce->sha1, sha1);
596 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
598 int size, namelen, was_same;
599 mode_t st_mode = st->st_mode;
600 struct cache_entry *ce, *alias;
601 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
602 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
603 int pretend = flags & ADD_CACHE_PRETEND;
604 int intent_only = flags & ADD_CACHE_INTENT;
605 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
606 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
608 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
609 return error("%s: can only add regular files, symbolic links or git-directories", path);
611 namelen = strlen(path);
612 if (S_ISDIR(st_mode)) {
613 while (namelen && path[namelen-1] == '/')
614 namelen--;
616 size = cache_entry_size(namelen);
617 ce = xcalloc(1, size);
618 memcpy(ce->name, path, namelen);
619 ce->ce_namelen = namelen;
620 if (!intent_only)
621 fill_stat_cache_info(ce, st);
622 else
623 ce->ce_flags |= CE_INTENT_TO_ADD;
625 if (trust_executable_bit && has_symlinks)
626 ce->ce_mode = create_ce_mode(st_mode);
627 else {
628 /* If there is an existing entry, pick the mode bits and type
629 * from it, otherwise assume unexecutable regular file.
631 struct cache_entry *ent;
632 int pos = index_name_pos_also_unmerged(istate, path, namelen);
634 ent = (0 <= pos) ? istate->cache[pos] : NULL;
635 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
638 /* When core.ignorecase=true, determine if a directory of the same name but differing
639 * case already exists within the Git repository. If it does, ensure the directory
640 * case of the file being added to the repository matches (is folded into) the existing
641 * entry's directory case.
643 if (ignore_case) {
644 const char *startPtr = ce->name;
645 const char *ptr = startPtr;
646 while (*ptr) {
647 while (*ptr && *ptr != '/')
648 ++ptr;
649 if (*ptr == '/') {
650 struct cache_entry *foundce;
651 ++ptr;
652 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
653 if (foundce) {
654 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
655 startPtr = ptr;
661 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
662 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
663 /* Nothing changed, really */
664 free(ce);
665 if (!S_ISGITLINK(alias->ce_mode))
666 ce_mark_uptodate(alias);
667 alias->ce_flags |= CE_ADDED;
668 return 0;
670 if (!intent_only) {
671 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
672 return error("unable to index file %s", path);
673 } else
674 set_object_name_for_intent_to_add_entry(ce);
676 if (ignore_case && alias && different_name(ce, alias))
677 ce = create_alias_ce(ce, alias);
678 ce->ce_flags |= CE_ADDED;
680 /* It was suspected to be racily clean, but it turns out to be Ok */
681 was_same = (alias &&
682 !ce_stage(alias) &&
683 !hashcmp(alias->sha1, ce->sha1) &&
684 ce->ce_mode == alias->ce_mode);
686 if (pretend)
688 else if (add_index_entry(istate, ce, add_option))
689 return error("unable to add %s to index",path);
690 if (verbose && !was_same)
691 printf("add '%s'\n", path);
692 return 0;
695 int add_file_to_index(struct index_state *istate, const char *path, int flags)
697 struct stat st;
698 if (lstat(path, &st))
699 die_errno("unable to stat '%s'", path);
700 return add_to_index(istate, path, &st, flags);
703 struct cache_entry *make_cache_entry(unsigned int mode,
704 const unsigned char *sha1, const char *path, int stage,
705 unsigned int refresh_options)
707 int size, len;
708 struct cache_entry *ce;
710 if (!verify_path(path)) {
711 error("Invalid path '%s'", path);
712 return NULL;
715 len = strlen(path);
716 size = cache_entry_size(len);
717 ce = xcalloc(1, size);
719 hashcpy(ce->sha1, sha1);
720 memcpy(ce->name, path, len);
721 ce->ce_flags = create_ce_flags(stage);
722 ce->ce_namelen = len;
723 ce->ce_mode = create_ce_mode(mode);
725 return refresh_cache_entry(ce, refresh_options);
728 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
730 int len = ce_namelen(a);
731 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
735 * We fundamentally don't like some paths: we don't want
736 * dot or dot-dot anywhere, and for obvious reasons don't
737 * want to recurse into ".git" either.
739 * Also, we don't want double slashes or slashes at the
740 * end that can make pathnames ambiguous.
742 static int verify_dotfile(const char *rest)
745 * The first character was '.', but that
746 * has already been discarded, we now test
747 * the rest.
750 /* "." is not allowed */
751 if (*rest == '\0' || is_dir_sep(*rest))
752 return 0;
754 switch (*rest) {
756 * ".git" followed by NUL or slash is bad. This
757 * shares the path end test with the ".." case.
759 case 'g':
760 case 'G':
761 if (rest[1] != 'i' && rest[1] != 'I')
762 break;
763 if (rest[2] != 't' && rest[2] != 'T')
764 break;
765 rest += 2;
766 /* fallthrough */
767 case '.':
768 if (rest[1] == '\0' || is_dir_sep(rest[1]))
769 return 0;
771 return 1;
774 int verify_path(const char *path)
776 char c;
778 if (has_dos_drive_prefix(path))
779 return 0;
781 goto inside;
782 for (;;) {
783 if (!c)
784 return 1;
785 if (is_dir_sep(c)) {
786 inside:
787 if (protect_hfs && is_hfs_dotgit(path))
788 return 0;
789 if (protect_ntfs && is_ntfs_dotgit(path))
790 return 0;
791 c = *path++;
792 if ((c == '.' && !verify_dotfile(path)) ||
793 is_dir_sep(c) || c == '\0')
794 return 0;
796 c = *path++;
801 * Do we have another file that has the beginning components being a
802 * proper superset of the name we're trying to add?
804 static int has_file_name(struct index_state *istate,
805 const struct cache_entry *ce, int pos, int ok_to_replace)
807 int retval = 0;
808 int len = ce_namelen(ce);
809 int stage = ce_stage(ce);
810 const char *name = ce->name;
812 while (pos < istate->cache_nr) {
813 struct cache_entry *p = istate->cache[pos++];
815 if (len >= ce_namelen(p))
816 break;
817 if (memcmp(name, p->name, len))
818 break;
819 if (ce_stage(p) != stage)
820 continue;
821 if (p->name[len] != '/')
822 continue;
823 if (p->ce_flags & CE_REMOVE)
824 continue;
825 retval = -1;
826 if (!ok_to_replace)
827 break;
828 remove_index_entry_at(istate, --pos);
830 return retval;
834 * Do we have another file with a pathname that is a proper
835 * subset of the name we're trying to add?
837 static int has_dir_name(struct index_state *istate,
838 const struct cache_entry *ce, int pos, int ok_to_replace)
840 int retval = 0;
841 int stage = ce_stage(ce);
842 const char *name = ce->name;
843 const char *slash = name + ce_namelen(ce);
845 for (;;) {
846 int len;
848 for (;;) {
849 if (*--slash == '/')
850 break;
851 if (slash <= ce->name)
852 return retval;
854 len = slash - name;
856 pos = index_name_stage_pos(istate, name, len, stage);
857 if (pos >= 0) {
859 * Found one, but not so fast. This could
860 * be a marker that says "I was here, but
861 * I am being removed". Such an entry is
862 * not a part of the resulting tree, and
863 * it is Ok to have a directory at the same
864 * path.
866 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
867 retval = -1;
868 if (!ok_to_replace)
869 break;
870 remove_index_entry_at(istate, pos);
871 continue;
874 else
875 pos = -pos-1;
878 * Trivial optimization: if we find an entry that
879 * already matches the sub-directory, then we know
880 * we're ok, and we can exit.
882 while (pos < istate->cache_nr) {
883 struct cache_entry *p = istate->cache[pos];
884 if ((ce_namelen(p) <= len) ||
885 (p->name[len] != '/') ||
886 memcmp(p->name, name, len))
887 break; /* not our subdirectory */
888 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
890 * p is at the same stage as our entry, and
891 * is a subdirectory of what we are looking
892 * at, so we cannot have conflicts at our
893 * level or anything shorter.
895 return retval;
896 pos++;
899 return retval;
902 /* We may be in a situation where we already have path/file and path
903 * is being added, or we already have path and path/file is being
904 * added. Either one would result in a nonsense tree that has path
905 * twice when git-write-tree tries to write it out. Prevent it.
907 * If ok-to-replace is specified, we remove the conflicting entries
908 * from the cache so the caller should recompute the insert position.
909 * When this happens, we return non-zero.
911 static int check_file_directory_conflict(struct index_state *istate,
912 const struct cache_entry *ce,
913 int pos, int ok_to_replace)
915 int retval;
918 * When ce is an "I am going away" entry, we allow it to be added
920 if (ce->ce_flags & CE_REMOVE)
921 return 0;
924 * We check if the path is a sub-path of a subsequent pathname
925 * first, since removing those will not change the position
926 * in the array.
928 retval = has_file_name(istate, ce, pos, ok_to_replace);
931 * Then check if the path might have a clashing sub-directory
932 * before it.
934 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
937 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
939 int pos;
940 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
941 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
942 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
943 int new_only = option & ADD_CACHE_NEW_ONLY;
945 cache_tree_invalidate_path(istate->cache_tree, ce->name);
946 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
948 /* existing match? Just replace it. */
949 if (pos >= 0) {
950 if (!new_only)
951 replace_index_entry(istate, pos, ce);
952 return 0;
954 pos = -pos-1;
957 * Inserting a merged entry ("stage 0") into the index
958 * will always replace all non-merged entries..
960 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
961 while (ce_same_name(istate->cache[pos], ce)) {
962 ok_to_add = 1;
963 if (!remove_index_entry_at(istate, pos))
964 break;
968 if (!ok_to_add)
969 return -1;
970 if (!verify_path(ce->name))
971 return error("Invalid path '%s'", ce->name);
973 if (!skip_df_check &&
974 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
975 if (!ok_to_replace)
976 return error("'%s' appears as both a file and as a directory",
977 ce->name);
978 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
979 pos = -pos-1;
981 return pos + 1;
984 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
986 int pos;
988 if (option & ADD_CACHE_JUST_APPEND)
989 pos = istate->cache_nr;
990 else {
991 int ret;
992 ret = add_index_entry_with_check(istate, ce, option);
993 if (ret <= 0)
994 return ret;
995 pos = ret - 1;
998 /* Make sure the array is big enough .. */
999 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1001 /* Add it in.. */
1002 istate->cache_nr++;
1003 if (istate->cache_nr > pos + 1)
1004 memmove(istate->cache + pos + 1,
1005 istate->cache + pos,
1006 (istate->cache_nr - pos - 1) * sizeof(ce));
1007 set_index_entry(istate, pos, ce);
1008 istate->cache_changed = 1;
1009 return 0;
1013 * "refresh" does not calculate a new sha1 file or bring the
1014 * cache up-to-date for mode/content changes. But what it
1015 * _does_ do is to "re-match" the stat information of a file
1016 * with the cache, so that you can refresh the cache for a
1017 * file that hasn't been changed but where the stat entry is
1018 * out of date.
1020 * For example, you'd want to do this after doing a "git-read-tree",
1021 * to link up the stat cache details with the proper files.
1023 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1024 struct cache_entry *ce,
1025 unsigned int options, int *err,
1026 int *changed_ret)
1028 struct stat st;
1029 struct cache_entry *updated;
1030 int changed, size;
1031 int refresh = options & CE_MATCH_REFRESH;
1032 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1033 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1034 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1036 if (!refresh || ce_uptodate(ce))
1037 return ce;
1040 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1041 * that the change to the work tree does not matter and told
1042 * us not to worry.
1044 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1045 ce_mark_uptodate(ce);
1046 return ce;
1048 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1049 ce_mark_uptodate(ce);
1050 return ce;
1053 if (lstat(ce->name, &st) < 0) {
1054 if (ignore_missing && errno == ENOENT)
1055 return ce;
1056 if (err)
1057 *err = errno;
1058 return NULL;
1061 changed = ie_match_stat(istate, ce, &st, options);
1062 if (changed_ret)
1063 *changed_ret = changed;
1064 if (!changed) {
1066 * The path is unchanged. If we were told to ignore
1067 * valid bit, then we did the actual stat check and
1068 * found that the entry is unmodified. If the entry
1069 * is not marked VALID, this is the place to mark it
1070 * valid again, under "assume unchanged" mode.
1072 if (ignore_valid && assume_unchanged &&
1073 !(ce->ce_flags & CE_VALID))
1074 ; /* mark this one VALID again */
1075 else {
1077 * We do not mark the index itself "modified"
1078 * because CE_UPTODATE flag is in-core only;
1079 * we are not going to write this change out.
1081 if (!S_ISGITLINK(ce->ce_mode))
1082 ce_mark_uptodate(ce);
1083 return ce;
1087 if (ie_modified(istate, ce, &st, options)) {
1088 if (err)
1089 *err = EINVAL;
1090 return NULL;
1093 size = ce_size(ce);
1094 updated = xmalloc(size);
1095 memcpy(updated, ce, size);
1096 fill_stat_cache_info(updated, &st);
1098 * If ignore_valid is not set, we should leave CE_VALID bit
1099 * alone. Otherwise, paths marked with --no-assume-unchanged
1100 * (i.e. things to be edited) will reacquire CE_VALID bit
1101 * automatically, which is not really what we want.
1103 if (!ignore_valid && assume_unchanged &&
1104 !(ce->ce_flags & CE_VALID))
1105 updated->ce_flags &= ~CE_VALID;
1107 return updated;
1110 static void show_file(const char * fmt, const char * name, int in_porcelain,
1111 int * first, const char *header_msg)
1113 if (in_porcelain && *first && header_msg) {
1114 printf("%s\n", header_msg);
1115 *first = 0;
1117 printf(fmt, name);
1120 int refresh_index(struct index_state *istate, unsigned int flags,
1121 const struct pathspec *pathspec,
1122 char *seen, const char *header_msg)
1124 int i;
1125 int has_errors = 0;
1126 int really = (flags & REFRESH_REALLY) != 0;
1127 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1128 int quiet = (flags & REFRESH_QUIET) != 0;
1129 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1130 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1131 int first = 1;
1132 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1133 unsigned int options = (CE_MATCH_REFRESH |
1134 (really ? CE_MATCH_IGNORE_VALID : 0) |
1135 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1136 const char *modified_fmt;
1137 const char *deleted_fmt;
1138 const char *typechange_fmt;
1139 const char *added_fmt;
1140 const char *unmerged_fmt;
1142 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1143 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1144 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1145 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1146 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1147 for (i = 0; i < istate->cache_nr; i++) {
1148 struct cache_entry *ce, *new;
1149 int cache_errno = 0;
1150 int changed = 0;
1151 int filtered = 0;
1153 ce = istate->cache[i];
1154 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1155 continue;
1157 if (pathspec && !ce_path_match(ce, pathspec, seen))
1158 filtered = 1;
1160 if (ce_stage(ce)) {
1161 while ((i < istate->cache_nr) &&
1162 ! strcmp(istate->cache[i]->name, ce->name))
1163 i++;
1164 i--;
1165 if (allow_unmerged)
1166 continue;
1167 if (!filtered)
1168 show_file(unmerged_fmt, ce->name, in_porcelain,
1169 &first, header_msg);
1170 has_errors = 1;
1171 continue;
1174 if (filtered)
1175 continue;
1177 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1178 if (new == ce)
1179 continue;
1180 if (!new) {
1181 const char *fmt;
1183 if (really && cache_errno == EINVAL) {
1184 /* If we are doing --really-refresh that
1185 * means the index is not valid anymore.
1187 ce->ce_flags &= ~CE_VALID;
1188 istate->cache_changed = 1;
1190 if (quiet)
1191 continue;
1193 if (cache_errno == ENOENT)
1194 fmt = deleted_fmt;
1195 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1196 fmt = added_fmt; /* must be before other checks */
1197 else if (changed & TYPE_CHANGED)
1198 fmt = typechange_fmt;
1199 else
1200 fmt = modified_fmt;
1201 show_file(fmt,
1202 ce->name, in_porcelain, &first, header_msg);
1203 has_errors = 1;
1204 continue;
1207 replace_index_entry(istate, i, new);
1209 return has_errors;
1212 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1213 unsigned int options)
1215 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1219 /*****************************************************************
1220 * Index File I/O
1221 *****************************************************************/
1223 #define INDEX_FORMAT_DEFAULT 3
1225 static int index_format_config(const char *var, const char *value, void *cb)
1227 unsigned int *version = cb;
1228 if (!strcmp(var, "index.version")) {
1229 *version = git_config_int(var, value);
1230 return 0;
1232 return 1;
1235 static unsigned int get_index_format_default(void)
1237 char *envversion = getenv("GIT_INDEX_VERSION");
1238 char *endp;
1239 unsigned int version = INDEX_FORMAT_DEFAULT;
1241 if (!envversion) {
1242 git_config(index_format_config, &version);
1243 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1244 warning(_("index.version set, but the value is invalid.\n"
1245 "Using version %i"), INDEX_FORMAT_DEFAULT);
1246 return INDEX_FORMAT_DEFAULT;
1248 return version;
1251 version = strtoul(envversion, &endp, 10);
1252 if (*endp ||
1253 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1254 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1255 "Using version %i"), INDEX_FORMAT_DEFAULT);
1256 version = INDEX_FORMAT_DEFAULT;
1258 return version;
1262 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1263 * Again - this is just a (very strong in practice) heuristic that
1264 * the inode hasn't changed.
1266 * We save the fields in big-endian order to allow using the
1267 * index file over NFS transparently.
1269 struct ondisk_cache_entry {
1270 struct cache_time ctime;
1271 struct cache_time mtime;
1272 uint32_t dev;
1273 uint32_t ino;
1274 uint32_t mode;
1275 uint32_t uid;
1276 uint32_t gid;
1277 uint32_t size;
1278 unsigned char sha1[20];
1279 uint16_t flags;
1280 char name[FLEX_ARRAY]; /* more */
1284 * This struct is used when CE_EXTENDED bit is 1
1285 * The struct must match ondisk_cache_entry exactly from
1286 * ctime till flags
1288 struct ondisk_cache_entry_extended {
1289 struct cache_time ctime;
1290 struct cache_time mtime;
1291 uint32_t dev;
1292 uint32_t ino;
1293 uint32_t mode;
1294 uint32_t uid;
1295 uint32_t gid;
1296 uint32_t size;
1297 unsigned char sha1[20];
1298 uint16_t flags;
1299 uint16_t flags2;
1300 char name[FLEX_ARRAY]; /* more */
1303 /* These are only used for v3 or lower */
1304 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1305 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1306 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1307 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1308 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1309 ondisk_cache_entry_size(ce_namelen(ce)))
1311 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1313 git_SHA_CTX c;
1314 unsigned char sha1[20];
1315 int hdr_version;
1317 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1318 return error("bad signature");
1319 hdr_version = ntohl(hdr->hdr_version);
1320 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1321 return error("bad index version %d", hdr_version);
1322 git_SHA1_Init(&c);
1323 git_SHA1_Update(&c, hdr, size - 20);
1324 git_SHA1_Final(sha1, &c);
1325 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1326 return error("bad index file sha1 signature");
1327 return 0;
1330 static int read_index_extension(struct index_state *istate,
1331 const char *ext, void *data, unsigned long sz)
1333 switch (CACHE_EXT(ext)) {
1334 case CACHE_EXT_TREE:
1335 istate->cache_tree = cache_tree_read(data, sz);
1336 break;
1337 case CACHE_EXT_RESOLVE_UNDO:
1338 istate->resolve_undo = resolve_undo_read(data, sz);
1339 break;
1340 default:
1341 if (*ext < 'A' || 'Z' < *ext)
1342 return error("index uses %.4s extension, which we do not understand",
1343 ext);
1344 fprintf(stderr, "ignoring %.4s extension\n", ext);
1345 break;
1347 return 0;
1350 int read_index(struct index_state *istate)
1352 return read_index_from(istate, get_index_file());
1355 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1356 unsigned int flags,
1357 const char *name,
1358 size_t len)
1360 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1362 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1363 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1364 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1365 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1366 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1367 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1368 ce->ce_mode = get_be32(&ondisk->mode);
1369 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1370 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1371 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1372 ce->ce_flags = flags & ~CE_NAMEMASK;
1373 ce->ce_namelen = len;
1374 hashcpy(ce->sha1, ondisk->sha1);
1375 memcpy(ce->name, name, len);
1376 ce->name[len] = '\0';
1377 return ce;
1381 * Adjacent cache entries tend to share the leading paths, so it makes
1382 * sense to only store the differences in later entries. In the v4
1383 * on-disk format of the index, each on-disk cache entry stores the
1384 * number of bytes to be stripped from the end of the previous name,
1385 * and the bytes to append to the result, to come up with its name.
1387 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1389 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1390 size_t len = decode_varint(&cp);
1392 if (name->len < len)
1393 die("malformed name field in the index");
1394 strbuf_remove(name, name->len - len, len);
1395 for (ep = cp; *ep; ep++)
1396 ; /* find the end */
1397 strbuf_add(name, cp, ep - cp);
1398 return (const char *)ep + 1 - cp_;
1401 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1402 unsigned long *ent_size,
1403 struct strbuf *previous_name)
1405 struct cache_entry *ce;
1406 size_t len;
1407 const char *name;
1408 unsigned int flags;
1410 /* On-disk flags are just 16 bits */
1411 flags = get_be16(&ondisk->flags);
1412 len = flags & CE_NAMEMASK;
1414 if (flags & CE_EXTENDED) {
1415 struct ondisk_cache_entry_extended *ondisk2;
1416 int extended_flags;
1417 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1418 extended_flags = get_be16(&ondisk2->flags2) << 16;
1419 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1420 if (extended_flags & ~CE_EXTENDED_FLAGS)
1421 die("Unknown index entry format %08x", extended_flags);
1422 flags |= extended_flags;
1423 name = ondisk2->name;
1425 else
1426 name = ondisk->name;
1428 if (!previous_name) {
1429 /* v3 and earlier */
1430 if (len == CE_NAMEMASK)
1431 len = strlen(name);
1432 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1434 *ent_size = ondisk_ce_size(ce);
1435 } else {
1436 unsigned long consumed;
1437 consumed = expand_name_field(previous_name, name);
1438 ce = cache_entry_from_ondisk(ondisk, flags,
1439 previous_name->buf,
1440 previous_name->len);
1442 *ent_size = (name - ((char *)ondisk)) + consumed;
1444 return ce;
1447 /* remember to discard_cache() before reading a different cache! */
1448 int read_index_from(struct index_state *istate, const char *path)
1450 int fd, i;
1451 struct stat st;
1452 unsigned long src_offset;
1453 struct cache_header *hdr;
1454 void *mmap;
1455 size_t mmap_size;
1456 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1458 if (istate->initialized)
1459 return istate->cache_nr;
1461 istate->timestamp.sec = 0;
1462 istate->timestamp.nsec = 0;
1463 fd = open(path, O_RDONLY);
1464 if (fd < 0) {
1465 if (errno == ENOENT)
1466 return 0;
1467 die_errno("index file open failed");
1470 if (fstat(fd, &st))
1471 die_errno("cannot stat the open index");
1473 mmap_size = xsize_t(st.st_size);
1474 if (mmap_size < sizeof(struct cache_header) + 20)
1475 die("index file smaller than expected");
1477 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1478 if (mmap == MAP_FAILED)
1479 die_errno("unable to map index file");
1480 close(fd);
1482 hdr = mmap;
1483 if (verify_hdr(hdr, mmap_size) < 0)
1484 goto unmap;
1486 hashcpy(istate->sha1, (unsigned char *)hdr + mmap_size - 20);
1487 istate->version = ntohl(hdr->hdr_version);
1488 istate->cache_nr = ntohl(hdr->hdr_entries);
1489 istate->cache_alloc = alloc_nr(istate->cache_nr);
1490 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1491 istate->initialized = 1;
1493 if (istate->version == 4)
1494 previous_name = &previous_name_buf;
1495 else
1496 previous_name = NULL;
1498 src_offset = sizeof(*hdr);
1499 for (i = 0; i < istate->cache_nr; i++) {
1500 struct ondisk_cache_entry *disk_ce;
1501 struct cache_entry *ce;
1502 unsigned long consumed;
1504 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1505 ce = create_from_disk(disk_ce, &consumed, previous_name);
1506 set_index_entry(istate, i, ce);
1508 src_offset += consumed;
1510 strbuf_release(&previous_name_buf);
1511 istate->timestamp.sec = st.st_mtime;
1512 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1514 while (src_offset <= mmap_size - 20 - 8) {
1515 /* After an array of active_nr index entries,
1516 * there can be arbitrary number of extended
1517 * sections, each of which is prefixed with
1518 * extension name (4-byte) and section length
1519 * in 4-byte network byte order.
1521 uint32_t extsize;
1522 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1523 extsize = ntohl(extsize);
1524 if (read_index_extension(istate,
1525 (const char *) mmap + src_offset,
1526 (char *) mmap + src_offset + 8,
1527 extsize) < 0)
1528 goto unmap;
1529 src_offset += 8;
1530 src_offset += extsize;
1532 munmap(mmap, mmap_size);
1533 return istate->cache_nr;
1535 unmap:
1536 munmap(mmap, mmap_size);
1537 die("index file corrupt");
1540 int is_index_unborn(struct index_state *istate)
1542 return (!istate->cache_nr && !istate->timestamp.sec);
1545 int discard_index(struct index_state *istate)
1547 int i;
1549 for (i = 0; i < istate->cache_nr; i++)
1550 free(istate->cache[i]);
1551 resolve_undo_clear_index(istate);
1552 istate->cache_nr = 0;
1553 istate->cache_changed = 0;
1554 istate->timestamp.sec = 0;
1555 istate->timestamp.nsec = 0;
1556 free_name_hash(istate);
1557 cache_tree_free(&(istate->cache_tree));
1558 istate->initialized = 0;
1559 free(istate->cache);
1560 istate->cache = NULL;
1561 istate->cache_alloc = 0;
1562 return 0;
1565 int unmerged_index(const struct index_state *istate)
1567 int i;
1568 for (i = 0; i < istate->cache_nr; i++) {
1569 if (ce_stage(istate->cache[i]))
1570 return 1;
1572 return 0;
1575 #define WRITE_BUFFER_SIZE 8192
1576 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1577 static unsigned long write_buffer_len;
1579 static int ce_write_flush(git_SHA_CTX *context, int fd)
1581 unsigned int buffered = write_buffer_len;
1582 if (buffered) {
1583 git_SHA1_Update(context, write_buffer, buffered);
1584 if (write_in_full(fd, write_buffer, buffered) != buffered)
1585 return -1;
1586 write_buffer_len = 0;
1588 return 0;
1591 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1593 while (len) {
1594 unsigned int buffered = write_buffer_len;
1595 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1596 if (partial > len)
1597 partial = len;
1598 memcpy(write_buffer + buffered, data, partial);
1599 buffered += partial;
1600 if (buffered == WRITE_BUFFER_SIZE) {
1601 write_buffer_len = buffered;
1602 if (ce_write_flush(context, fd))
1603 return -1;
1604 buffered = 0;
1606 write_buffer_len = buffered;
1607 len -= partial;
1608 data = (char *) data + partial;
1610 return 0;
1613 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1614 unsigned int ext, unsigned int sz)
1616 ext = htonl(ext);
1617 sz = htonl(sz);
1618 return ((ce_write(context, fd, &ext, 4) < 0) ||
1619 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1622 static int ce_flush(git_SHA_CTX *context, int fd)
1624 unsigned int left = write_buffer_len;
1626 if (left) {
1627 write_buffer_len = 0;
1628 git_SHA1_Update(context, write_buffer, left);
1631 /* Flush first if not enough space for SHA1 signature */
1632 if (left + 20 > WRITE_BUFFER_SIZE) {
1633 if (write_in_full(fd, write_buffer, left) != left)
1634 return -1;
1635 left = 0;
1638 /* Append the SHA1 signature at the end */
1639 git_SHA1_Final(write_buffer + left, context);
1640 left += 20;
1641 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1644 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1647 * The only thing we care about in this function is to smudge the
1648 * falsely clean entry due to touch-update-touch race, so we leave
1649 * everything else as they are. We are called for entries whose
1650 * ce_stat_data.sd_mtime match the index file mtime.
1652 * Note that this actually does not do much for gitlinks, for
1653 * which ce_match_stat_basic() always goes to the actual
1654 * contents. The caller checks with is_racy_timestamp() which
1655 * always says "no" for gitlinks, so we are not called for them ;-)
1657 struct stat st;
1659 if (lstat(ce->name, &st) < 0)
1660 return;
1661 if (ce_match_stat_basic(ce, &st))
1662 return;
1663 if (ce_modified_check_fs(ce, &st)) {
1664 /* This is "racily clean"; smudge it. Note that this
1665 * is a tricky code. At first glance, it may appear
1666 * that it can break with this sequence:
1668 * $ echo xyzzy >frotz
1669 * $ git-update-index --add frotz
1670 * $ : >frotz
1671 * $ sleep 3
1672 * $ echo filfre >nitfol
1673 * $ git-update-index --add nitfol
1675 * but it does not. When the second update-index runs,
1676 * it notices that the entry "frotz" has the same timestamp
1677 * as index, and if we were to smudge it by resetting its
1678 * size to zero here, then the object name recorded
1679 * in index is the 6-byte file but the cached stat information
1680 * becomes zero --- which would then match what we would
1681 * obtain from the filesystem next time we stat("frotz").
1683 * However, the second update-index, before calling
1684 * this function, notices that the cached size is 6
1685 * bytes and what is on the filesystem is an empty
1686 * file, and never calls us, so the cached size information
1687 * for "frotz" stays 6 which does not match the filesystem.
1689 ce->ce_stat_data.sd_size = 0;
1693 /* Copy miscellaneous fields but not the name */
1694 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1695 struct cache_entry *ce)
1697 short flags;
1699 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1700 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1701 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1702 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1703 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1704 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1705 ondisk->mode = htonl(ce->ce_mode);
1706 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1707 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1708 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1709 hashcpy(ondisk->sha1, ce->sha1);
1711 flags = ce->ce_flags;
1712 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1713 ondisk->flags = htons(flags);
1714 if (ce->ce_flags & CE_EXTENDED) {
1715 struct ondisk_cache_entry_extended *ondisk2;
1716 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1717 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1718 return ondisk2->name;
1720 else {
1721 return ondisk->name;
1725 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1726 struct strbuf *previous_name)
1728 int size;
1729 struct ondisk_cache_entry *ondisk;
1730 char *name;
1731 int result;
1733 if (!previous_name) {
1734 size = ondisk_ce_size(ce);
1735 ondisk = xcalloc(1, size);
1736 name = copy_cache_entry_to_ondisk(ondisk, ce);
1737 memcpy(name, ce->name, ce_namelen(ce));
1738 } else {
1739 int common, to_remove, prefix_size;
1740 unsigned char to_remove_vi[16];
1741 for (common = 0;
1742 (ce->name[common] &&
1743 common < previous_name->len &&
1744 ce->name[common] == previous_name->buf[common]);
1745 common++)
1746 ; /* still matching */
1747 to_remove = previous_name->len - common;
1748 prefix_size = encode_varint(to_remove, to_remove_vi);
1750 if (ce->ce_flags & CE_EXTENDED)
1751 size = offsetof(struct ondisk_cache_entry_extended, name);
1752 else
1753 size = offsetof(struct ondisk_cache_entry, name);
1754 size += prefix_size + (ce_namelen(ce) - common + 1);
1756 ondisk = xcalloc(1, size);
1757 name = copy_cache_entry_to_ondisk(ondisk, ce);
1758 memcpy(name, to_remove_vi, prefix_size);
1759 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1761 strbuf_splice(previous_name, common, to_remove,
1762 ce->name + common, ce_namelen(ce) - common);
1765 result = ce_write(c, fd, ondisk, size);
1766 free(ondisk);
1767 return result;
1771 * This function verifies if index_state has the correct sha1 of the
1772 * index file. Don't die if we have any other failure, just return 0.
1774 static int verify_index_from(const struct index_state *istate, const char *path)
1776 int fd;
1777 ssize_t n;
1778 struct stat st;
1779 unsigned char sha1[20];
1781 if (!istate->initialized)
1782 return 0;
1784 fd = open(path, O_RDONLY);
1785 if (fd < 0)
1786 return 0;
1788 if (fstat(fd, &st))
1789 goto out;
1791 if (st.st_size < sizeof(struct cache_header) + 20)
1792 goto out;
1794 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1795 if (n != 20)
1796 goto out;
1798 if (hashcmp(istate->sha1, sha1))
1799 goto out;
1801 close(fd);
1802 return 1;
1804 out:
1805 close(fd);
1806 return 0;
1809 static int verify_index(const struct index_state *istate)
1811 return verify_index_from(istate, get_index_file());
1814 static int has_racy_timestamp(struct index_state *istate)
1816 int entries = istate->cache_nr;
1817 int i;
1819 for (i = 0; i < entries; i++) {
1820 struct cache_entry *ce = istate->cache[i];
1821 if (is_racy_timestamp(istate, ce))
1822 return 1;
1824 return 0;
1828 * Opportunistically update the index but do not complain if we can't
1830 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1832 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1833 verify_index(istate) && !write_index(istate, lockfile->fd))
1834 commit_locked_index(lockfile);
1835 else
1836 rollback_lock_file(lockfile);
1839 int write_index(struct index_state *istate, int newfd)
1841 git_SHA_CTX c;
1842 struct cache_header hdr;
1843 int i, err, removed, extended, hdr_version;
1844 struct cache_entry **cache = istate->cache;
1845 int entries = istate->cache_nr;
1846 struct stat st;
1847 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1849 for (i = removed = extended = 0; i < entries; i++) {
1850 if (cache[i]->ce_flags & CE_REMOVE)
1851 removed++;
1853 /* reduce extended entries if possible */
1854 cache[i]->ce_flags &= ~CE_EXTENDED;
1855 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1856 extended++;
1857 cache[i]->ce_flags |= CE_EXTENDED;
1861 if (!istate->version)
1862 istate->version = get_index_format_default();
1864 /* demote version 3 to version 2 when the latter suffices */
1865 if (istate->version == 3 || istate->version == 2)
1866 istate->version = extended ? 3 : 2;
1868 hdr_version = istate->version;
1870 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1871 hdr.hdr_version = htonl(hdr_version);
1872 hdr.hdr_entries = htonl(entries - removed);
1874 git_SHA1_Init(&c);
1875 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1876 return -1;
1878 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1879 for (i = 0; i < entries; i++) {
1880 struct cache_entry *ce = cache[i];
1881 if (ce->ce_flags & CE_REMOVE)
1882 continue;
1883 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1884 ce_smudge_racily_clean_entry(ce);
1885 if (is_null_sha1(ce->sha1)) {
1886 static const char msg[] = "cache entry has null sha1: %s";
1887 static int allow = -1;
1889 if (allow < 0)
1890 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1891 if (allow)
1892 warning(msg, ce->name);
1893 else
1894 return error(msg, ce->name);
1896 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1897 return -1;
1899 strbuf_release(&previous_name_buf);
1901 /* Write extension data here */
1902 if (istate->cache_tree) {
1903 struct strbuf sb = STRBUF_INIT;
1905 cache_tree_write(&sb, istate->cache_tree);
1906 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1907 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1908 strbuf_release(&sb);
1909 if (err)
1910 return -1;
1912 if (istate->resolve_undo) {
1913 struct strbuf sb = STRBUF_INIT;
1915 resolve_undo_write(&sb, istate->resolve_undo);
1916 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1917 sb.len) < 0
1918 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1919 strbuf_release(&sb);
1920 if (err)
1921 return -1;
1924 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1925 return -1;
1926 istate->timestamp.sec = (unsigned int)st.st_mtime;
1927 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1928 return 0;
1932 * Read the index file that is potentially unmerged into given
1933 * index_state, dropping any unmerged entries. Returns true if
1934 * the index is unmerged. Callers who want to refuse to work
1935 * from an unmerged state can call this and check its return value,
1936 * instead of calling read_cache().
1938 int read_index_unmerged(struct index_state *istate)
1940 int i;
1941 int unmerged = 0;
1943 read_index(istate);
1944 for (i = 0; i < istate->cache_nr; i++) {
1945 struct cache_entry *ce = istate->cache[i];
1946 struct cache_entry *new_ce;
1947 int size, len;
1949 if (!ce_stage(ce))
1950 continue;
1951 unmerged = 1;
1952 len = ce_namelen(ce);
1953 size = cache_entry_size(len);
1954 new_ce = xcalloc(1, size);
1955 memcpy(new_ce->name, ce->name, len);
1956 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1957 new_ce->ce_namelen = len;
1958 new_ce->ce_mode = ce->ce_mode;
1959 if (add_index_entry(istate, new_ce, 0))
1960 return error("%s: cannot drop to stage #0",
1961 new_ce->name);
1962 i = index_name_pos(istate, new_ce->name, len);
1964 return unmerged;
1968 * Returns 1 if the path is an "other" path with respect to
1969 * the index; that is, the path is not mentioned in the index at all,
1970 * either as a file, a directory with some files in the index,
1971 * or as an unmerged entry.
1973 * We helpfully remove a trailing "/" from directories so that
1974 * the output of read_directory can be used as-is.
1976 int index_name_is_other(const struct index_state *istate, const char *name,
1977 int namelen)
1979 int pos;
1980 if (namelen && name[namelen - 1] == '/')
1981 namelen--;
1982 pos = index_name_pos(istate, name, namelen);
1983 if (0 <= pos)
1984 return 0; /* exact match */
1985 pos = -pos - 1;
1986 if (pos < istate->cache_nr) {
1987 struct cache_entry *ce = istate->cache[pos];
1988 if (ce_namelen(ce) == namelen &&
1989 !memcmp(ce->name, name, namelen))
1990 return 0; /* Yup, this one exists unmerged */
1992 return 1;
1995 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
1997 int pos, len;
1998 unsigned long sz;
1999 enum object_type type;
2000 void *data;
2002 len = strlen(path);
2003 pos = index_name_pos(istate, path, len);
2004 if (pos < 0) {
2006 * We might be in the middle of a merge, in which
2007 * case we would read stage #2 (ours).
2009 int i;
2010 for (i = -pos - 1;
2011 (pos < 0 && i < istate->cache_nr &&
2012 !strcmp(istate->cache[i]->name, path));
2013 i++)
2014 if (ce_stage(istate->cache[i]) == 2)
2015 pos = i;
2017 if (pos < 0)
2018 return NULL;
2019 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2020 if (!data || type != OBJ_BLOB) {
2021 free(data);
2022 return NULL;
2024 if (size)
2025 *size = sz;
2026 return data;
2029 void stat_validity_clear(struct stat_validity *sv)
2031 free(sv->sd);
2032 sv->sd = NULL;
2035 int stat_validity_check(struct stat_validity *sv, const char *path)
2037 struct stat st;
2039 if (stat(path, &st) < 0)
2040 return sv->sd == NULL;
2041 if (!sv->sd)
2042 return 0;
2043 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2046 void stat_validity_update(struct stat_validity *sv, int fd)
2048 struct stat st;
2050 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2051 stat_validity_clear(sv);
2052 else {
2053 if (!sv->sd)
2054 sv->sd = xcalloc(1, sizeof(struct stat_data));
2055 fill_stat_data(sv->sd, &st);