Interested in helping open source friends on HP-UX?
[git/mjg.git] / read-cache.c
blob9cff715d6b669ecf05f861bc553b61d20d04191d
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 "lockfile.h"
9 #include "cache-tree.h"
10 #include "refs.h"
11 #include "dir.h"
12 #include "tree.h"
13 #include "commit.h"
14 #include "blob.h"
15 #include "resolve-undo.h"
16 #include "strbuf.h"
17 #include "varint.h"
18 #include "split-index.h"
19 #include "sigchain.h"
20 #include "utf8.h"
22 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
23 unsigned int options);
25 /* Mask for the name length in ce_flags in the on-disk index */
27 #define CE_NAMEMASK (0x0fff)
29 /* Index extensions.
31 * The first letter should be 'A'..'Z' for extensions that are not
32 * necessary for a correct operation (i.e. optimization data).
33 * When new extensions are added that _needs_ to be understood in
34 * order to correctly interpret the index file, pick character that
35 * is outside the range, to cause the reader to abort.
38 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
39 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
40 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
41 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
43 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
44 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
45 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
46 SPLIT_INDEX_ORDERED)
48 struct index_state the_index;
49 static const char *alternate_index_output;
51 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
53 istate->cache[nr] = ce;
54 add_name_hash(istate, ce);
57 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
59 struct cache_entry *old = istate->cache[nr];
61 replace_index_entry_in_base(istate, old, ce);
62 remove_name_hash(istate, old);
63 free(old);
64 set_index_entry(istate, nr, ce);
65 ce->ce_flags |= CE_UPDATE_IN_BASE;
66 istate->cache_changed |= CE_ENTRY_CHANGED;
69 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
71 struct cache_entry *old = istate->cache[nr], *new;
72 int namelen = strlen(new_name);
74 new = xmalloc(cache_entry_size(namelen));
75 copy_cache_entry(new, old);
76 new->ce_flags &= ~CE_HASHED;
77 new->ce_namelen = namelen;
78 new->index = 0;
79 memcpy(new->name, new_name, namelen + 1);
81 cache_tree_invalidate_path(istate, old->name);
82 remove_index_entry_at(istate, nr);
83 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
86 void fill_stat_data(struct stat_data *sd, struct stat *st)
88 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
89 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
90 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
91 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
92 sd->sd_dev = st->st_dev;
93 sd->sd_ino = st->st_ino;
94 sd->sd_uid = st->st_uid;
95 sd->sd_gid = st->st_gid;
96 sd->sd_size = st->st_size;
99 int match_stat_data(const struct stat_data *sd, struct stat *st)
101 int changed = 0;
103 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
104 changed |= MTIME_CHANGED;
105 if (trust_ctime && check_stat &&
106 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
107 changed |= CTIME_CHANGED;
109 #ifdef USE_NSEC
110 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
111 changed |= MTIME_CHANGED;
112 if (trust_ctime && check_stat &&
113 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
114 changed |= CTIME_CHANGED;
115 #endif
117 if (check_stat) {
118 if (sd->sd_uid != (unsigned int) st->st_uid ||
119 sd->sd_gid != (unsigned int) st->st_gid)
120 changed |= OWNER_CHANGED;
121 if (sd->sd_ino != (unsigned int) st->st_ino)
122 changed |= INODE_CHANGED;
125 #ifdef USE_STDEV
127 * st_dev breaks on network filesystems where different
128 * clients will have different views of what "device"
129 * the filesystem is on
131 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
132 changed |= INODE_CHANGED;
133 #endif
135 if (sd->sd_size != (unsigned int) st->st_size)
136 changed |= DATA_CHANGED;
138 return changed;
142 * This only updates the "non-critical" parts of the directory
143 * cache, ie the parts that aren't tracked by GIT, and only used
144 * to validate the cache.
146 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
148 fill_stat_data(&ce->ce_stat_data, st);
150 if (assume_unchanged)
151 ce->ce_flags |= CE_VALID;
153 if (S_ISREG(st->st_mode))
154 ce_mark_uptodate(ce);
157 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
159 int match = -1;
160 int fd = open(ce->name, O_RDONLY);
162 if (fd >= 0) {
163 unsigned char sha1[20];
164 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
165 match = hashcmp(sha1, ce->sha1);
166 /* index_fd() closed the file descriptor already */
168 return match;
171 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
173 int match = -1;
174 void *buffer;
175 unsigned long size;
176 enum object_type type;
177 struct strbuf sb = STRBUF_INIT;
179 if (strbuf_readlink(&sb, ce->name, expected_size))
180 return -1;
182 buffer = read_sha1_file(ce->sha1, &type, &size);
183 if (buffer) {
184 if (size == sb.len)
185 match = memcmp(buffer, sb.buf, size);
186 free(buffer);
188 strbuf_release(&sb);
189 return match;
192 static int ce_compare_gitlink(const struct cache_entry *ce)
194 unsigned char sha1[20];
197 * We don't actually require that the .git directory
198 * under GITLINK directory be a valid git directory. It
199 * might even be missing (in case nobody populated that
200 * sub-project).
202 * If so, we consider it always to match.
204 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
205 return 0;
206 return hashcmp(sha1, ce->sha1);
209 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
211 switch (st->st_mode & S_IFMT) {
212 case S_IFREG:
213 if (ce_compare_data(ce, st))
214 return DATA_CHANGED;
215 break;
216 case S_IFLNK:
217 if (ce_compare_link(ce, xsize_t(st->st_size)))
218 return DATA_CHANGED;
219 break;
220 case S_IFDIR:
221 if (S_ISGITLINK(ce->ce_mode))
222 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
223 default:
224 return TYPE_CHANGED;
226 return 0;
229 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
231 unsigned int changed = 0;
233 if (ce->ce_flags & CE_REMOVE)
234 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
236 switch (ce->ce_mode & S_IFMT) {
237 case S_IFREG:
238 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
239 /* We consider only the owner x bit to be relevant for
240 * "mode changes"
242 if (trust_executable_bit &&
243 (0100 & (ce->ce_mode ^ st->st_mode)))
244 changed |= MODE_CHANGED;
245 break;
246 case S_IFLNK:
247 if (!S_ISLNK(st->st_mode) &&
248 (has_symlinks || !S_ISREG(st->st_mode)))
249 changed |= TYPE_CHANGED;
250 break;
251 case S_IFGITLINK:
252 /* We ignore most of the st_xxx fields for gitlinks */
253 if (!S_ISDIR(st->st_mode))
254 changed |= TYPE_CHANGED;
255 else if (ce_compare_gitlink(ce))
256 changed |= DATA_CHANGED;
257 return changed;
258 default:
259 die("internal error: ce_mode is %o", ce->ce_mode);
262 changed |= match_stat_data(&ce->ce_stat_data, st);
264 /* Racily smudged entry? */
265 if (!ce->ce_stat_data.sd_size) {
266 if (!is_empty_blob_sha1(ce->sha1))
267 changed |= DATA_CHANGED;
270 return changed;
273 static int is_racy_timestamp(const struct index_state *istate,
274 const struct cache_entry *ce)
276 return (!S_ISGITLINK(ce->ce_mode) &&
277 istate->timestamp.sec &&
278 #ifdef USE_NSEC
279 /* nanosecond timestamped files can also be racy! */
280 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
281 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
282 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
283 #else
284 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
285 #endif
289 int ie_match_stat(const struct index_state *istate,
290 const struct cache_entry *ce, struct stat *st,
291 unsigned int options)
293 unsigned int changed;
294 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
295 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
296 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
299 * If it's marked as always valid in the index, it's
300 * valid whatever the checked-out copy says.
302 * skip-worktree has the same effect with higher precedence
304 if (!ignore_skip_worktree && ce_skip_worktree(ce))
305 return 0;
306 if (!ignore_valid && (ce->ce_flags & CE_VALID))
307 return 0;
310 * Intent-to-add entries have not been added, so the index entry
311 * by definition never matches what is in the work tree until it
312 * actually gets added.
314 if (ce->ce_flags & CE_INTENT_TO_ADD)
315 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
317 changed = ce_match_stat_basic(ce, st);
320 * Within 1 second of this sequence:
321 * echo xyzzy >file && git-update-index --add file
322 * running this command:
323 * echo frotz >file
324 * would give a falsely clean cache entry. The mtime and
325 * length match the cache, and other stat fields do not change.
327 * We could detect this at update-index time (the cache entry
328 * being registered/updated records the same time as "now")
329 * and delay the return from git-update-index, but that would
330 * effectively mean we can make at most one commit per second,
331 * which is not acceptable. Instead, we check cache entries
332 * whose mtime are the same as the index file timestamp more
333 * carefully than others.
335 if (!changed && is_racy_timestamp(istate, ce)) {
336 if (assume_racy_is_modified)
337 changed |= DATA_CHANGED;
338 else
339 changed |= ce_modified_check_fs(ce, st);
342 return changed;
345 int ie_modified(const struct index_state *istate,
346 const struct cache_entry *ce,
347 struct stat *st, unsigned int options)
349 int changed, changed_fs;
351 changed = ie_match_stat(istate, ce, st, options);
352 if (!changed)
353 return 0;
355 * If the mode or type has changed, there's no point in trying
356 * to refresh the entry - it's not going to match
358 if (changed & (MODE_CHANGED | TYPE_CHANGED))
359 return changed;
362 * Immediately after read-tree or update-index --cacheinfo,
363 * the length field is zero, as we have never even read the
364 * lstat(2) information once, and we cannot trust DATA_CHANGED
365 * returned by ie_match_stat() which in turn was returned by
366 * ce_match_stat_basic() to signal that the filesize of the
367 * blob changed. We have to actually go to the filesystem to
368 * see if the contents match, and if so, should answer "unchanged".
370 * The logic does not apply to gitlinks, as ce_match_stat_basic()
371 * already has checked the actual HEAD from the filesystem in the
372 * subproject. If ie_match_stat() already said it is different,
373 * then we know it is.
375 if ((changed & DATA_CHANGED) &&
376 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
377 return changed;
379 changed_fs = ce_modified_check_fs(ce, st);
380 if (changed_fs)
381 return changed | changed_fs;
382 return 0;
385 int base_name_compare(const char *name1, int len1, int mode1,
386 const char *name2, int len2, int mode2)
388 unsigned char c1, c2;
389 int len = len1 < len2 ? len1 : len2;
390 int cmp;
392 cmp = memcmp(name1, name2, len);
393 if (cmp)
394 return cmp;
395 c1 = name1[len];
396 c2 = name2[len];
397 if (!c1 && S_ISDIR(mode1))
398 c1 = '/';
399 if (!c2 && S_ISDIR(mode2))
400 c2 = '/';
401 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
405 * df_name_compare() is identical to base_name_compare(), except it
406 * compares conflicting directory/file entries as equal. Note that
407 * while a directory name compares as equal to a regular file, they
408 * then individually compare _differently_ to a filename that has
409 * a dot after the basename (because '\0' < '.' < '/').
411 * This is used by routines that want to traverse the git namespace
412 * but then handle conflicting entries together when possible.
414 int df_name_compare(const char *name1, int len1, int mode1,
415 const char *name2, int len2, int mode2)
417 int len = len1 < len2 ? len1 : len2, cmp;
418 unsigned char c1, c2;
420 cmp = memcmp(name1, name2, len);
421 if (cmp)
422 return cmp;
423 /* Directories and files compare equal (same length, same name) */
424 if (len1 == len2)
425 return 0;
426 c1 = name1[len];
427 if (!c1 && S_ISDIR(mode1))
428 c1 = '/';
429 c2 = name2[len];
430 if (!c2 && S_ISDIR(mode2))
431 c2 = '/';
432 if (c1 == '/' && !c2)
433 return 0;
434 if (c2 == '/' && !c1)
435 return 0;
436 return c1 - c2;
439 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
441 size_t min_len = (len1 < len2) ? len1 : len2;
442 int cmp = memcmp(name1, name2, min_len);
443 if (cmp)
444 return cmp;
445 if (len1 < len2)
446 return -1;
447 if (len1 > len2)
448 return 1;
449 return 0;
452 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
454 int cmp;
456 cmp = name_compare(name1, len1, name2, len2);
457 if (cmp)
458 return cmp;
460 if (stage1 < stage2)
461 return -1;
462 if (stage1 > stage2)
463 return 1;
464 return 0;
467 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
469 int first, last;
471 first = 0;
472 last = istate->cache_nr;
473 while (last > first) {
474 int next = (last + first) >> 1;
475 struct cache_entry *ce = istate->cache[next];
476 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
477 if (!cmp)
478 return next;
479 if (cmp < 0) {
480 last = next;
481 continue;
483 first = next+1;
485 return -first-1;
488 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
490 return index_name_stage_pos(istate, name, namelen, 0);
493 /* Remove entry, return true if there are more entries to go.. */
494 int remove_index_entry_at(struct index_state *istate, int pos)
496 struct cache_entry *ce = istate->cache[pos];
498 record_resolve_undo(istate, ce);
499 remove_name_hash(istate, ce);
500 save_or_free_index_entry(istate, ce);
501 istate->cache_changed |= CE_ENTRY_REMOVED;
502 istate->cache_nr--;
503 if (pos >= istate->cache_nr)
504 return 0;
505 memmove(istate->cache + pos,
506 istate->cache + pos + 1,
507 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
508 return 1;
512 * Remove all cache entries marked for removal, that is where
513 * CE_REMOVE is set in ce_flags. This is much more effective than
514 * calling remove_index_entry_at() for each entry to be removed.
516 void remove_marked_cache_entries(struct index_state *istate)
518 struct cache_entry **ce_array = istate->cache;
519 unsigned int i, j;
521 for (i = j = 0; i < istate->cache_nr; i++) {
522 if (ce_array[i]->ce_flags & CE_REMOVE) {
523 remove_name_hash(istate, ce_array[i]);
524 save_or_free_index_entry(istate, ce_array[i]);
526 else
527 ce_array[j++] = ce_array[i];
529 if (j == istate->cache_nr)
530 return;
531 istate->cache_changed |= CE_ENTRY_REMOVED;
532 istate->cache_nr = j;
535 int remove_file_from_index(struct index_state *istate, const char *path)
537 int pos = index_name_pos(istate, path, strlen(path));
538 if (pos < 0)
539 pos = -pos-1;
540 cache_tree_invalidate_path(istate, path);
541 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
542 remove_index_entry_at(istate, pos);
543 return 0;
546 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
548 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
551 static int index_name_pos_also_unmerged(struct index_state *istate,
552 const char *path, int namelen)
554 int pos = index_name_pos(istate, path, namelen);
555 struct cache_entry *ce;
557 if (pos >= 0)
558 return pos;
560 /* maybe unmerged? */
561 pos = -1 - pos;
562 if (pos >= istate->cache_nr ||
563 compare_name((ce = istate->cache[pos]), path, namelen))
564 return -1;
566 /* order of preference: stage 2, 1, 3 */
567 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
568 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
569 !compare_name(ce, path, namelen))
570 pos++;
571 return pos;
574 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
576 int len = ce_namelen(ce);
577 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
581 * If we add a filename that aliases in the cache, we will use the
582 * name that we already have - but we don't want to update the same
583 * alias twice, because that implies that there were actually two
584 * different files with aliasing names!
586 * So we use the CE_ADDED flag to verify that the alias was an old
587 * one before we accept it as
589 static struct cache_entry *create_alias_ce(struct index_state *istate,
590 struct cache_entry *ce,
591 struct cache_entry *alias)
593 int len;
594 struct cache_entry *new;
596 if (alias->ce_flags & CE_ADDED)
597 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
599 /* Ok, create the new entry using the name of the existing alias */
600 len = ce_namelen(alias);
601 new = xcalloc(1, cache_entry_size(len));
602 memcpy(new->name, alias->name, len);
603 copy_cache_entry(new, ce);
604 save_or_free_index_entry(istate, ce);
605 return new;
608 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
610 unsigned char sha1[20];
611 if (write_sha1_file("", 0, blob_type, sha1))
612 die("cannot create an empty blob in the object database");
613 hashcpy(ce->sha1, sha1);
616 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
618 int size, namelen, was_same;
619 mode_t st_mode = st->st_mode;
620 struct cache_entry *ce, *alias;
621 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
622 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
623 int pretend = flags & ADD_CACHE_PRETEND;
624 int intent_only = flags & ADD_CACHE_INTENT;
625 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
626 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
628 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
629 return error("%s: can only add regular files, symbolic links or git-directories", path);
631 namelen = strlen(path);
632 if (S_ISDIR(st_mode)) {
633 while (namelen && path[namelen-1] == '/')
634 namelen--;
636 size = cache_entry_size(namelen);
637 ce = xcalloc(1, size);
638 memcpy(ce->name, path, namelen);
639 ce->ce_namelen = namelen;
640 if (!intent_only)
641 fill_stat_cache_info(ce, st);
642 else
643 ce->ce_flags |= CE_INTENT_TO_ADD;
645 if (trust_executable_bit && has_symlinks)
646 ce->ce_mode = create_ce_mode(st_mode);
647 else {
648 /* If there is an existing entry, pick the mode bits and type
649 * from it, otherwise assume unexecutable regular file.
651 struct cache_entry *ent;
652 int pos = index_name_pos_also_unmerged(istate, path, namelen);
654 ent = (0 <= pos) ? istate->cache[pos] : NULL;
655 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
658 /* When core.ignorecase=true, determine if a directory of the same name but differing
659 * case already exists within the Git repository. If it does, ensure the directory
660 * case of the file being added to the repository matches (is folded into) the existing
661 * entry's directory case.
663 if (ignore_case) {
664 const char *startPtr = ce->name;
665 const char *ptr = startPtr;
666 while (*ptr) {
667 while (*ptr && *ptr != '/')
668 ++ptr;
669 if (*ptr == '/') {
670 struct cache_entry *foundce;
671 ++ptr;
672 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
673 if (foundce) {
674 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
675 startPtr = ptr;
681 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
682 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
683 /* Nothing changed, really */
684 free(ce);
685 if (!S_ISGITLINK(alias->ce_mode))
686 ce_mark_uptodate(alias);
687 alias->ce_flags |= CE_ADDED;
688 return 0;
690 if (!intent_only) {
691 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
692 return error("unable to index file %s", path);
693 } else
694 set_object_name_for_intent_to_add_entry(ce);
696 if (ignore_case && alias && different_name(ce, alias))
697 ce = create_alias_ce(istate, ce, alias);
698 ce->ce_flags |= CE_ADDED;
700 /* It was suspected to be racily clean, but it turns out to be Ok */
701 was_same = (alias &&
702 !ce_stage(alias) &&
703 !hashcmp(alias->sha1, ce->sha1) &&
704 ce->ce_mode == alias->ce_mode);
706 if (pretend)
708 else if (add_index_entry(istate, ce, add_option))
709 return error("unable to add %s to index",path);
710 if (verbose && !was_same)
711 printf("add '%s'\n", path);
712 return 0;
715 int add_file_to_index(struct index_state *istate, const char *path, int flags)
717 struct stat st;
718 if (lstat(path, &st))
719 die_errno("unable to stat '%s'", path);
720 return add_to_index(istate, path, &st, flags);
723 struct cache_entry *make_cache_entry(unsigned int mode,
724 const unsigned char *sha1, const char *path, int stage,
725 unsigned int refresh_options)
727 int size, len;
728 struct cache_entry *ce;
730 if (!verify_path(path)) {
731 error("Invalid path '%s'", path);
732 return NULL;
735 len = strlen(path);
736 size = cache_entry_size(len);
737 ce = xcalloc(1, size);
739 hashcpy(ce->sha1, sha1);
740 memcpy(ce->name, path, len);
741 ce->ce_flags = create_ce_flags(stage);
742 ce->ce_namelen = len;
743 ce->ce_mode = create_ce_mode(mode);
745 return refresh_cache_entry(ce, refresh_options);
748 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
750 int len = ce_namelen(a);
751 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
755 * We fundamentally don't like some paths: we don't want
756 * dot or dot-dot anywhere, and for obvious reasons don't
757 * want to recurse into ".git" either.
759 * Also, we don't want double slashes or slashes at the
760 * end that can make pathnames ambiguous.
762 static int verify_dotfile(const char *rest)
765 * The first character was '.', but that
766 * has already been discarded, we now test
767 * the rest.
770 /* "." is not allowed */
771 if (*rest == '\0' || is_dir_sep(*rest))
772 return 0;
774 switch (*rest) {
776 * ".git" followed by NUL or slash is bad. This
777 * shares the path end test with the ".." case.
779 case 'g':
780 case 'G':
781 if (rest[1] != 'i' && rest[1] != 'I')
782 break;
783 if (rest[2] != 't' && rest[2] != 'T')
784 break;
785 rest += 2;
786 /* fallthrough */
787 case '.':
788 if (rest[1] == '\0' || is_dir_sep(rest[1]))
789 return 0;
791 return 1;
794 int verify_path(const char *path)
796 char c;
798 if (has_dos_drive_prefix(path))
799 return 0;
801 goto inside;
802 for (;;) {
803 if (!c)
804 return 1;
805 if (is_dir_sep(c)) {
806 inside:
807 if (protect_hfs && is_hfs_dotgit(path))
808 return 0;
809 if (protect_ntfs && is_ntfs_dotgit(path))
810 return 0;
811 c = *path++;
812 if ((c == '.' && !verify_dotfile(path)) ||
813 is_dir_sep(c) || c == '\0')
814 return 0;
816 c = *path++;
821 * Do we have another file that has the beginning components being a
822 * proper superset of the name we're trying to add?
824 static int has_file_name(struct index_state *istate,
825 const struct cache_entry *ce, int pos, int ok_to_replace)
827 int retval = 0;
828 int len = ce_namelen(ce);
829 int stage = ce_stage(ce);
830 const char *name = ce->name;
832 while (pos < istate->cache_nr) {
833 struct cache_entry *p = istate->cache[pos++];
835 if (len >= ce_namelen(p))
836 break;
837 if (memcmp(name, p->name, len))
838 break;
839 if (ce_stage(p) != stage)
840 continue;
841 if (p->name[len] != '/')
842 continue;
843 if (p->ce_flags & CE_REMOVE)
844 continue;
845 retval = -1;
846 if (!ok_to_replace)
847 break;
848 remove_index_entry_at(istate, --pos);
850 return retval;
854 * Do we have another file with a pathname that is a proper
855 * subset of the name we're trying to add?
857 static int has_dir_name(struct index_state *istate,
858 const struct cache_entry *ce, int pos, int ok_to_replace)
860 int retval = 0;
861 int stage = ce_stage(ce);
862 const char *name = ce->name;
863 const char *slash = name + ce_namelen(ce);
865 for (;;) {
866 int len;
868 for (;;) {
869 if (*--slash == '/')
870 break;
871 if (slash <= ce->name)
872 return retval;
874 len = slash - name;
876 pos = index_name_stage_pos(istate, name, len, stage);
877 if (pos >= 0) {
879 * Found one, but not so fast. This could
880 * be a marker that says "I was here, but
881 * I am being removed". Such an entry is
882 * not a part of the resulting tree, and
883 * it is Ok to have a directory at the same
884 * path.
886 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
887 retval = -1;
888 if (!ok_to_replace)
889 break;
890 remove_index_entry_at(istate, pos);
891 continue;
894 else
895 pos = -pos-1;
898 * Trivial optimization: if we find an entry that
899 * already matches the sub-directory, then we know
900 * we're ok, and we can exit.
902 while (pos < istate->cache_nr) {
903 struct cache_entry *p = istate->cache[pos];
904 if ((ce_namelen(p) <= len) ||
905 (p->name[len] != '/') ||
906 memcmp(p->name, name, len))
907 break; /* not our subdirectory */
908 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
910 * p is at the same stage as our entry, and
911 * is a subdirectory of what we are looking
912 * at, so we cannot have conflicts at our
913 * level or anything shorter.
915 return retval;
916 pos++;
919 return retval;
922 /* We may be in a situation where we already have path/file and path
923 * is being added, or we already have path and path/file is being
924 * added. Either one would result in a nonsense tree that has path
925 * twice when git-write-tree tries to write it out. Prevent it.
927 * If ok-to-replace is specified, we remove the conflicting entries
928 * from the cache so the caller should recompute the insert position.
929 * When this happens, we return non-zero.
931 static int check_file_directory_conflict(struct index_state *istate,
932 const struct cache_entry *ce,
933 int pos, int ok_to_replace)
935 int retval;
938 * When ce is an "I am going away" entry, we allow it to be added
940 if (ce->ce_flags & CE_REMOVE)
941 return 0;
944 * We check if the path is a sub-path of a subsequent pathname
945 * first, since removing those will not change the position
946 * in the array.
948 retval = has_file_name(istate, ce, pos, ok_to_replace);
951 * Then check if the path might have a clashing sub-directory
952 * before it.
954 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
957 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
959 int pos;
960 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
961 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
962 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
963 int new_only = option & ADD_CACHE_NEW_ONLY;
965 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
966 cache_tree_invalidate_path(istate, ce->name);
967 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
969 /* existing match? Just replace it. */
970 if (pos >= 0) {
971 if (!new_only)
972 replace_index_entry(istate, pos, ce);
973 return 0;
975 pos = -pos-1;
978 * Inserting a merged entry ("stage 0") into the index
979 * will always replace all non-merged entries..
981 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
982 while (ce_same_name(istate->cache[pos], ce)) {
983 ok_to_add = 1;
984 if (!remove_index_entry_at(istate, pos))
985 break;
989 if (!ok_to_add)
990 return -1;
991 if (!verify_path(ce->name))
992 return error("Invalid path '%s'", ce->name);
994 if (!skip_df_check &&
995 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
996 if (!ok_to_replace)
997 return error("'%s' appears as both a file and as a directory",
998 ce->name);
999 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1000 pos = -pos-1;
1002 return pos + 1;
1005 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1007 int pos;
1009 if (option & ADD_CACHE_JUST_APPEND)
1010 pos = istate->cache_nr;
1011 else {
1012 int ret;
1013 ret = add_index_entry_with_check(istate, ce, option);
1014 if (ret <= 0)
1015 return ret;
1016 pos = ret - 1;
1019 /* Make sure the array is big enough .. */
1020 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1022 /* Add it in.. */
1023 istate->cache_nr++;
1024 if (istate->cache_nr > pos + 1)
1025 memmove(istate->cache + pos + 1,
1026 istate->cache + pos,
1027 (istate->cache_nr - pos - 1) * sizeof(ce));
1028 set_index_entry(istate, pos, ce);
1029 istate->cache_changed |= CE_ENTRY_ADDED;
1030 return 0;
1034 * "refresh" does not calculate a new sha1 file or bring the
1035 * cache up-to-date for mode/content changes. But what it
1036 * _does_ do is to "re-match" the stat information of a file
1037 * with the cache, so that you can refresh the cache for a
1038 * file that hasn't been changed but where the stat entry is
1039 * out of date.
1041 * For example, you'd want to do this after doing a "git-read-tree",
1042 * to link up the stat cache details with the proper files.
1044 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1045 struct cache_entry *ce,
1046 unsigned int options, int *err,
1047 int *changed_ret)
1049 struct stat st;
1050 struct cache_entry *updated;
1051 int changed, size;
1052 int refresh = options & CE_MATCH_REFRESH;
1053 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1054 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1055 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1057 if (!refresh || ce_uptodate(ce))
1058 return ce;
1061 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1062 * that the change to the work tree does not matter and told
1063 * us not to worry.
1065 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1066 ce_mark_uptodate(ce);
1067 return ce;
1069 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1070 ce_mark_uptodate(ce);
1071 return ce;
1074 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1075 if (ignore_missing)
1076 return ce;
1077 if (err)
1078 *err = ENOENT;
1079 return NULL;
1082 if (lstat(ce->name, &st) < 0) {
1083 if (ignore_missing && errno == ENOENT)
1084 return ce;
1085 if (err)
1086 *err = errno;
1087 return NULL;
1090 changed = ie_match_stat(istate, ce, &st, options);
1091 if (changed_ret)
1092 *changed_ret = changed;
1093 if (!changed) {
1095 * The path is unchanged. If we were told to ignore
1096 * valid bit, then we did the actual stat check and
1097 * found that the entry is unmodified. If the entry
1098 * is not marked VALID, this is the place to mark it
1099 * valid again, under "assume unchanged" mode.
1101 if (ignore_valid && assume_unchanged &&
1102 !(ce->ce_flags & CE_VALID))
1103 ; /* mark this one VALID again */
1104 else {
1106 * We do not mark the index itself "modified"
1107 * because CE_UPTODATE flag is in-core only;
1108 * we are not going to write this change out.
1110 if (!S_ISGITLINK(ce->ce_mode))
1111 ce_mark_uptodate(ce);
1112 return ce;
1116 if (ie_modified(istate, ce, &st, options)) {
1117 if (err)
1118 *err = EINVAL;
1119 return NULL;
1122 size = ce_size(ce);
1123 updated = xmalloc(size);
1124 memcpy(updated, ce, size);
1125 fill_stat_cache_info(updated, &st);
1127 * If ignore_valid is not set, we should leave CE_VALID bit
1128 * alone. Otherwise, paths marked with --no-assume-unchanged
1129 * (i.e. things to be edited) will reacquire CE_VALID bit
1130 * automatically, which is not really what we want.
1132 if (!ignore_valid && assume_unchanged &&
1133 !(ce->ce_flags & CE_VALID))
1134 updated->ce_flags &= ~CE_VALID;
1136 /* istate->cache_changed is updated in the caller */
1137 return updated;
1140 static void show_file(const char * fmt, const char * name, int in_porcelain,
1141 int * first, const char *header_msg)
1143 if (in_porcelain && *first && header_msg) {
1144 printf("%s\n", header_msg);
1145 *first = 0;
1147 printf(fmt, name);
1150 int refresh_index(struct index_state *istate, unsigned int flags,
1151 const struct pathspec *pathspec,
1152 char *seen, const char *header_msg)
1154 int i;
1155 int has_errors = 0;
1156 int really = (flags & REFRESH_REALLY) != 0;
1157 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1158 int quiet = (flags & REFRESH_QUIET) != 0;
1159 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1160 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1161 int first = 1;
1162 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1163 unsigned int options = (CE_MATCH_REFRESH |
1164 (really ? CE_MATCH_IGNORE_VALID : 0) |
1165 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1166 const char *modified_fmt;
1167 const char *deleted_fmt;
1168 const char *typechange_fmt;
1169 const char *added_fmt;
1170 const char *unmerged_fmt;
1172 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1173 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1174 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1175 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1176 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1177 for (i = 0; i < istate->cache_nr; i++) {
1178 struct cache_entry *ce, *new;
1179 int cache_errno = 0;
1180 int changed = 0;
1181 int filtered = 0;
1183 ce = istate->cache[i];
1184 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1185 continue;
1187 if (pathspec && !ce_path_match(ce, pathspec, seen))
1188 filtered = 1;
1190 if (ce_stage(ce)) {
1191 while ((i < istate->cache_nr) &&
1192 ! strcmp(istate->cache[i]->name, ce->name))
1193 i++;
1194 i--;
1195 if (allow_unmerged)
1196 continue;
1197 if (!filtered)
1198 show_file(unmerged_fmt, ce->name, in_porcelain,
1199 &first, header_msg);
1200 has_errors = 1;
1201 continue;
1204 if (filtered)
1205 continue;
1207 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1208 if (new == ce)
1209 continue;
1210 if (!new) {
1211 const char *fmt;
1213 if (really && cache_errno == EINVAL) {
1214 /* If we are doing --really-refresh that
1215 * means the index is not valid anymore.
1217 ce->ce_flags &= ~CE_VALID;
1218 ce->ce_flags |= CE_UPDATE_IN_BASE;
1219 istate->cache_changed |= CE_ENTRY_CHANGED;
1221 if (quiet)
1222 continue;
1224 if (cache_errno == ENOENT)
1225 fmt = deleted_fmt;
1226 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1227 fmt = added_fmt; /* must be before other checks */
1228 else if (changed & TYPE_CHANGED)
1229 fmt = typechange_fmt;
1230 else
1231 fmt = modified_fmt;
1232 show_file(fmt,
1233 ce->name, in_porcelain, &first, header_msg);
1234 has_errors = 1;
1235 continue;
1238 replace_index_entry(istate, i, new);
1240 return has_errors;
1243 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1244 unsigned int options)
1246 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1250 /*****************************************************************
1251 * Index File I/O
1252 *****************************************************************/
1254 #define INDEX_FORMAT_DEFAULT 3
1256 static unsigned int get_index_format_default(void)
1258 char *envversion = getenv("GIT_INDEX_VERSION");
1259 char *endp;
1260 int value;
1261 unsigned int version = INDEX_FORMAT_DEFAULT;
1263 if (!envversion) {
1264 if (!git_config_get_int("index.version", &value))
1265 version = value;
1266 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1267 warning(_("index.version set, but the value is invalid.\n"
1268 "Using version %i"), INDEX_FORMAT_DEFAULT);
1269 return INDEX_FORMAT_DEFAULT;
1271 return version;
1274 version = strtoul(envversion, &endp, 10);
1275 if (*endp ||
1276 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1277 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1278 "Using version %i"), INDEX_FORMAT_DEFAULT);
1279 version = INDEX_FORMAT_DEFAULT;
1281 return version;
1285 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1286 * Again - this is just a (very strong in practice) heuristic that
1287 * the inode hasn't changed.
1289 * We save the fields in big-endian order to allow using the
1290 * index file over NFS transparently.
1292 struct ondisk_cache_entry {
1293 struct cache_time ctime;
1294 struct cache_time mtime;
1295 uint32_t dev;
1296 uint32_t ino;
1297 uint32_t mode;
1298 uint32_t uid;
1299 uint32_t gid;
1300 uint32_t size;
1301 unsigned char sha1[20];
1302 uint16_t flags;
1303 char name[FLEX_ARRAY]; /* more */
1307 * This struct is used when CE_EXTENDED bit is 1
1308 * The struct must match ondisk_cache_entry exactly from
1309 * ctime till flags
1311 struct ondisk_cache_entry_extended {
1312 struct cache_time ctime;
1313 struct cache_time mtime;
1314 uint32_t dev;
1315 uint32_t ino;
1316 uint32_t mode;
1317 uint32_t uid;
1318 uint32_t gid;
1319 uint32_t size;
1320 unsigned char sha1[20];
1321 uint16_t flags;
1322 uint16_t flags2;
1323 char name[FLEX_ARRAY]; /* more */
1326 /* These are only used for v3 or lower */
1327 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1328 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1329 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1330 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1331 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1332 ondisk_cache_entry_size(ce_namelen(ce)))
1334 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1336 git_SHA_CTX c;
1337 unsigned char sha1[20];
1338 int hdr_version;
1340 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1341 return error("bad signature");
1342 hdr_version = ntohl(hdr->hdr_version);
1343 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1344 return error("bad index version %d", hdr_version);
1345 git_SHA1_Init(&c);
1346 git_SHA1_Update(&c, hdr, size - 20);
1347 git_SHA1_Final(sha1, &c);
1348 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1349 return error("bad index file sha1 signature");
1350 return 0;
1353 static int read_index_extension(struct index_state *istate,
1354 const char *ext, void *data, unsigned long sz)
1356 switch (CACHE_EXT(ext)) {
1357 case CACHE_EXT_TREE:
1358 istate->cache_tree = cache_tree_read(data, sz);
1359 break;
1360 case CACHE_EXT_RESOLVE_UNDO:
1361 istate->resolve_undo = resolve_undo_read(data, sz);
1362 break;
1363 case CACHE_EXT_LINK:
1364 if (read_link_extension(istate, data, sz))
1365 return -1;
1366 break;
1367 default:
1368 if (*ext < 'A' || 'Z' < *ext)
1369 return error("index uses %.4s extension, which we do not understand",
1370 ext);
1371 fprintf(stderr, "ignoring %.4s extension\n", ext);
1372 break;
1374 return 0;
1377 int hold_locked_index(struct lock_file *lk, int die_on_error)
1379 return hold_lock_file_for_update(lk, get_index_file(),
1380 die_on_error
1381 ? LOCK_DIE_ON_ERROR
1382 : 0);
1385 int read_index(struct index_state *istate)
1387 return read_index_from(istate, get_index_file());
1390 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1391 unsigned int flags,
1392 const char *name,
1393 size_t len)
1395 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1397 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1398 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1399 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1400 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1401 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1402 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1403 ce->ce_mode = get_be32(&ondisk->mode);
1404 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1405 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1406 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1407 ce->ce_flags = flags & ~CE_NAMEMASK;
1408 ce->ce_namelen = len;
1409 ce->index = 0;
1410 hashcpy(ce->sha1, ondisk->sha1);
1411 memcpy(ce->name, name, len);
1412 ce->name[len] = '\0';
1413 return ce;
1417 * Adjacent cache entries tend to share the leading paths, so it makes
1418 * sense to only store the differences in later entries. In the v4
1419 * on-disk format of the index, each on-disk cache entry stores the
1420 * number of bytes to be stripped from the end of the previous name,
1421 * and the bytes to append to the result, to come up with its name.
1423 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1425 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1426 size_t len = decode_varint(&cp);
1428 if (name->len < len)
1429 die("malformed name field in the index");
1430 strbuf_remove(name, name->len - len, len);
1431 for (ep = cp; *ep; ep++)
1432 ; /* find the end */
1433 strbuf_add(name, cp, ep - cp);
1434 return (const char *)ep + 1 - cp_;
1437 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1438 unsigned long *ent_size,
1439 struct strbuf *previous_name)
1441 struct cache_entry *ce;
1442 size_t len;
1443 const char *name;
1444 unsigned int flags;
1446 /* On-disk flags are just 16 bits */
1447 flags = get_be16(&ondisk->flags);
1448 len = flags & CE_NAMEMASK;
1450 if (flags & CE_EXTENDED) {
1451 struct ondisk_cache_entry_extended *ondisk2;
1452 int extended_flags;
1453 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1454 extended_flags = get_be16(&ondisk2->flags2) << 16;
1455 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1456 if (extended_flags & ~CE_EXTENDED_FLAGS)
1457 die("Unknown index entry format %08x", extended_flags);
1458 flags |= extended_flags;
1459 name = ondisk2->name;
1461 else
1462 name = ondisk->name;
1464 if (!previous_name) {
1465 /* v3 and earlier */
1466 if (len == CE_NAMEMASK)
1467 len = strlen(name);
1468 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1470 *ent_size = ondisk_ce_size(ce);
1471 } else {
1472 unsigned long consumed;
1473 consumed = expand_name_field(previous_name, name);
1474 ce = cache_entry_from_ondisk(ondisk, flags,
1475 previous_name->buf,
1476 previous_name->len);
1478 *ent_size = (name - ((char *)ondisk)) + consumed;
1480 return ce;
1483 static void check_ce_order(struct cache_entry *ce, struct cache_entry *next_ce)
1485 int name_compare = strcmp(ce->name, next_ce->name);
1486 if (0 < name_compare)
1487 die("unordered stage entries in index");
1488 if (!name_compare) {
1489 if (!ce_stage(ce))
1490 die("multiple stage entries for merged file '%s'",
1491 ce->name);
1492 if (ce_stage(ce) > ce_stage(next_ce))
1493 die("unordered stage entries for '%s'",
1494 ce->name);
1498 /* remember to discard_cache() before reading a different cache! */
1499 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1501 int fd, i;
1502 struct stat st;
1503 unsigned long src_offset;
1504 struct cache_header *hdr;
1505 void *mmap;
1506 size_t mmap_size;
1507 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1509 if (istate->initialized)
1510 return istate->cache_nr;
1512 istate->timestamp.sec = 0;
1513 istate->timestamp.nsec = 0;
1514 fd = open(path, O_RDONLY);
1515 if (fd < 0) {
1516 if (!must_exist && errno == ENOENT)
1517 return 0;
1518 die_errno("%s: index file open failed", path);
1521 if (fstat(fd, &st))
1522 die_errno("cannot stat the open index");
1524 mmap_size = xsize_t(st.st_size);
1525 if (mmap_size < sizeof(struct cache_header) + 20)
1526 die("index file smaller than expected");
1528 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1529 if (mmap == MAP_FAILED)
1530 die_errno("unable to map index file");
1531 close(fd);
1533 hdr = mmap;
1534 if (verify_hdr(hdr, mmap_size) < 0)
1535 goto unmap;
1537 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1538 istate->version = ntohl(hdr->hdr_version);
1539 istate->cache_nr = ntohl(hdr->hdr_entries);
1540 istate->cache_alloc = alloc_nr(istate->cache_nr);
1541 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1542 istate->initialized = 1;
1544 if (istate->version == 4)
1545 previous_name = &previous_name_buf;
1546 else
1547 previous_name = NULL;
1549 src_offset = sizeof(*hdr);
1550 for (i = 0; i < istate->cache_nr; i++) {
1551 struct ondisk_cache_entry *disk_ce;
1552 struct cache_entry *ce;
1553 unsigned long consumed;
1555 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1556 ce = create_from_disk(disk_ce, &consumed, previous_name);
1557 set_index_entry(istate, i, ce);
1559 if (i > 0)
1560 check_ce_order(istate->cache[i - 1], ce);
1562 src_offset += consumed;
1564 strbuf_release(&previous_name_buf);
1565 istate->timestamp.sec = st.st_mtime;
1566 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1568 while (src_offset <= mmap_size - 20 - 8) {
1569 /* After an array of active_nr index entries,
1570 * there can be arbitrary number of extended
1571 * sections, each of which is prefixed with
1572 * extension name (4-byte) and section length
1573 * in 4-byte network byte order.
1575 uint32_t extsize;
1576 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1577 extsize = ntohl(extsize);
1578 if (read_index_extension(istate,
1579 (const char *) mmap + src_offset,
1580 (char *) mmap + src_offset + 8,
1581 extsize) < 0)
1582 goto unmap;
1583 src_offset += 8;
1584 src_offset += extsize;
1586 munmap(mmap, mmap_size);
1587 return istate->cache_nr;
1589 unmap:
1590 munmap(mmap, mmap_size);
1591 die("index file corrupt");
1594 int read_index_from(struct index_state *istate, const char *path)
1596 struct split_index *split_index;
1597 int ret;
1599 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1600 if (istate->initialized)
1601 return istate->cache_nr;
1603 ret = do_read_index(istate, path, 0);
1604 split_index = istate->split_index;
1605 if (!split_index)
1606 return ret;
1608 if (is_null_sha1(split_index->base_sha1))
1609 return ret;
1611 if (split_index->base)
1612 discard_index(split_index->base);
1613 else
1614 split_index->base = xcalloc(1, sizeof(*split_index->base));
1615 ret = do_read_index(split_index->base,
1616 git_path("sharedindex.%s",
1617 sha1_to_hex(split_index->base_sha1)), 1);
1618 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1619 die("broken index, expect %s in %s, got %s",
1620 sha1_to_hex(split_index->base_sha1),
1621 git_path("sharedindex.%s",
1622 sha1_to_hex(split_index->base_sha1)),
1623 sha1_to_hex(split_index->base->sha1));
1624 merge_base_index(istate);
1625 return ret;
1628 int is_index_unborn(struct index_state *istate)
1630 return (!istate->cache_nr && !istate->timestamp.sec);
1633 int discard_index(struct index_state *istate)
1635 int i;
1637 for (i = 0; i < istate->cache_nr; i++) {
1638 if (istate->cache[i]->index &&
1639 istate->split_index &&
1640 istate->split_index->base &&
1641 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1642 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1643 continue;
1644 free(istate->cache[i]);
1646 resolve_undo_clear_index(istate);
1647 istate->cache_nr = 0;
1648 istate->cache_changed = 0;
1649 istate->timestamp.sec = 0;
1650 istate->timestamp.nsec = 0;
1651 free_name_hash(istate);
1652 cache_tree_free(&(istate->cache_tree));
1653 istate->initialized = 0;
1654 free(istate->cache);
1655 istate->cache = NULL;
1656 istate->cache_alloc = 0;
1657 discard_split_index(istate);
1658 return 0;
1661 int unmerged_index(const struct index_state *istate)
1663 int i;
1664 for (i = 0; i < istate->cache_nr; i++) {
1665 if (ce_stage(istate->cache[i]))
1666 return 1;
1668 return 0;
1671 #define WRITE_BUFFER_SIZE 8192
1672 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1673 static unsigned long write_buffer_len;
1675 static int ce_write_flush(git_SHA_CTX *context, int fd)
1677 unsigned int buffered = write_buffer_len;
1678 if (buffered) {
1679 git_SHA1_Update(context, write_buffer, buffered);
1680 if (write_in_full(fd, write_buffer, buffered) != buffered)
1681 return -1;
1682 write_buffer_len = 0;
1684 return 0;
1687 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1689 while (len) {
1690 unsigned int buffered = write_buffer_len;
1691 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1692 if (partial > len)
1693 partial = len;
1694 memcpy(write_buffer + buffered, data, partial);
1695 buffered += partial;
1696 if (buffered == WRITE_BUFFER_SIZE) {
1697 write_buffer_len = buffered;
1698 if (ce_write_flush(context, fd))
1699 return -1;
1700 buffered = 0;
1702 write_buffer_len = buffered;
1703 len -= partial;
1704 data = (char *) data + partial;
1706 return 0;
1709 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1710 unsigned int ext, unsigned int sz)
1712 ext = htonl(ext);
1713 sz = htonl(sz);
1714 return ((ce_write(context, fd, &ext, 4) < 0) ||
1715 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1718 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1720 unsigned int left = write_buffer_len;
1722 if (left) {
1723 write_buffer_len = 0;
1724 git_SHA1_Update(context, write_buffer, left);
1727 /* Flush first if not enough space for SHA1 signature */
1728 if (left + 20 > WRITE_BUFFER_SIZE) {
1729 if (write_in_full(fd, write_buffer, left) != left)
1730 return -1;
1731 left = 0;
1734 /* Append the SHA1 signature at the end */
1735 git_SHA1_Final(write_buffer + left, context);
1736 hashcpy(sha1, write_buffer + left);
1737 left += 20;
1738 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1741 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1744 * The only thing we care about in this function is to smudge the
1745 * falsely clean entry due to touch-update-touch race, so we leave
1746 * everything else as they are. We are called for entries whose
1747 * ce_stat_data.sd_mtime match the index file mtime.
1749 * Note that this actually does not do much for gitlinks, for
1750 * which ce_match_stat_basic() always goes to the actual
1751 * contents. The caller checks with is_racy_timestamp() which
1752 * always says "no" for gitlinks, so we are not called for them ;-)
1754 struct stat st;
1756 if (lstat(ce->name, &st) < 0)
1757 return;
1758 if (ce_match_stat_basic(ce, &st))
1759 return;
1760 if (ce_modified_check_fs(ce, &st)) {
1761 /* This is "racily clean"; smudge it. Note that this
1762 * is a tricky code. At first glance, it may appear
1763 * that it can break with this sequence:
1765 * $ echo xyzzy >frotz
1766 * $ git-update-index --add frotz
1767 * $ : >frotz
1768 * $ sleep 3
1769 * $ echo filfre >nitfol
1770 * $ git-update-index --add nitfol
1772 * but it does not. When the second update-index runs,
1773 * it notices that the entry "frotz" has the same timestamp
1774 * as index, and if we were to smudge it by resetting its
1775 * size to zero here, then the object name recorded
1776 * in index is the 6-byte file but the cached stat information
1777 * becomes zero --- which would then match what we would
1778 * obtain from the filesystem next time we stat("frotz").
1780 * However, the second update-index, before calling
1781 * this function, notices that the cached size is 6
1782 * bytes and what is on the filesystem is an empty
1783 * file, and never calls us, so the cached size information
1784 * for "frotz" stays 6 which does not match the filesystem.
1786 ce->ce_stat_data.sd_size = 0;
1790 /* Copy miscellaneous fields but not the name */
1791 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1792 struct cache_entry *ce)
1794 short flags;
1796 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1797 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1798 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1799 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1800 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1801 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1802 ondisk->mode = htonl(ce->ce_mode);
1803 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1804 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1805 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1806 hashcpy(ondisk->sha1, ce->sha1);
1808 flags = ce->ce_flags & ~CE_NAMEMASK;
1809 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1810 ondisk->flags = htons(flags);
1811 if (ce->ce_flags & CE_EXTENDED) {
1812 struct ondisk_cache_entry_extended *ondisk2;
1813 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1814 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1815 return ondisk2->name;
1817 else {
1818 return ondisk->name;
1822 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1823 struct strbuf *previous_name)
1825 int size;
1826 struct ondisk_cache_entry *ondisk;
1827 int saved_namelen = saved_namelen; /* compiler workaround */
1828 char *name;
1829 int result;
1831 if (ce->ce_flags & CE_STRIP_NAME) {
1832 saved_namelen = ce_namelen(ce);
1833 ce->ce_namelen = 0;
1836 if (!previous_name) {
1837 size = ondisk_ce_size(ce);
1838 ondisk = xcalloc(1, size);
1839 name = copy_cache_entry_to_ondisk(ondisk, ce);
1840 memcpy(name, ce->name, ce_namelen(ce));
1841 } else {
1842 int common, to_remove, prefix_size;
1843 unsigned char to_remove_vi[16];
1844 for (common = 0;
1845 (ce->name[common] &&
1846 common < previous_name->len &&
1847 ce->name[common] == previous_name->buf[common]);
1848 common++)
1849 ; /* still matching */
1850 to_remove = previous_name->len - common;
1851 prefix_size = encode_varint(to_remove, to_remove_vi);
1853 if (ce->ce_flags & CE_EXTENDED)
1854 size = offsetof(struct ondisk_cache_entry_extended, name);
1855 else
1856 size = offsetof(struct ondisk_cache_entry, name);
1857 size += prefix_size + (ce_namelen(ce) - common + 1);
1859 ondisk = xcalloc(1, size);
1860 name = copy_cache_entry_to_ondisk(ondisk, ce);
1861 memcpy(name, to_remove_vi, prefix_size);
1862 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1864 strbuf_splice(previous_name, common, to_remove,
1865 ce->name + common, ce_namelen(ce) - common);
1867 if (ce->ce_flags & CE_STRIP_NAME) {
1868 ce->ce_namelen = saved_namelen;
1869 ce->ce_flags &= ~CE_STRIP_NAME;
1872 result = ce_write(c, fd, ondisk, size);
1873 free(ondisk);
1874 return result;
1878 * This function verifies if index_state has the correct sha1 of the
1879 * index file. Don't die if we have any other failure, just return 0.
1881 static int verify_index_from(const struct index_state *istate, const char *path)
1883 int fd;
1884 ssize_t n;
1885 struct stat st;
1886 unsigned char sha1[20];
1888 if (!istate->initialized)
1889 return 0;
1891 fd = open(path, O_RDONLY);
1892 if (fd < 0)
1893 return 0;
1895 if (fstat(fd, &st))
1896 goto out;
1898 if (st.st_size < sizeof(struct cache_header) + 20)
1899 goto out;
1901 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1902 if (n != 20)
1903 goto out;
1905 if (hashcmp(istate->sha1, sha1))
1906 goto out;
1908 close(fd);
1909 return 1;
1911 out:
1912 close(fd);
1913 return 0;
1916 static int verify_index(const struct index_state *istate)
1918 return verify_index_from(istate, get_index_file());
1921 static int has_racy_timestamp(struct index_state *istate)
1923 int entries = istate->cache_nr;
1924 int i;
1926 for (i = 0; i < entries; i++) {
1927 struct cache_entry *ce = istate->cache[i];
1928 if (is_racy_timestamp(istate, ce))
1929 return 1;
1931 return 0;
1935 * Opportunistically update the index but do not complain if we can't
1937 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1939 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1940 verify_index(istate) &&
1941 write_locked_index(istate, lockfile, COMMIT_LOCK))
1942 rollback_lock_file(lockfile);
1945 static int do_write_index(struct index_state *istate, int newfd,
1946 int strip_extensions)
1948 git_SHA_CTX c;
1949 struct cache_header hdr;
1950 int i, err, removed, extended, hdr_version;
1951 struct cache_entry **cache = istate->cache;
1952 int entries = istate->cache_nr;
1953 struct stat st;
1954 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1956 for (i = removed = extended = 0; i < entries; i++) {
1957 if (cache[i]->ce_flags & CE_REMOVE)
1958 removed++;
1960 /* reduce extended entries if possible */
1961 cache[i]->ce_flags &= ~CE_EXTENDED;
1962 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1963 extended++;
1964 cache[i]->ce_flags |= CE_EXTENDED;
1968 if (!istate->version) {
1969 istate->version = get_index_format_default();
1970 if (getenv("GIT_TEST_SPLIT_INDEX"))
1971 init_split_index(istate);
1974 /* demote version 3 to version 2 when the latter suffices */
1975 if (istate->version == 3 || istate->version == 2)
1976 istate->version = extended ? 3 : 2;
1978 hdr_version = istate->version;
1980 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1981 hdr.hdr_version = htonl(hdr_version);
1982 hdr.hdr_entries = htonl(entries - removed);
1984 git_SHA1_Init(&c);
1985 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1986 return -1;
1988 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1989 for (i = 0; i < entries; i++) {
1990 struct cache_entry *ce = cache[i];
1991 if (ce->ce_flags & CE_REMOVE)
1992 continue;
1993 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1994 ce_smudge_racily_clean_entry(ce);
1995 if (is_null_sha1(ce->sha1)) {
1996 static const char msg[] = "cache entry has null sha1: %s";
1997 static int allow = -1;
1999 if (allow < 0)
2000 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2001 if (allow)
2002 warning(msg, ce->name);
2003 else
2004 return error(msg, ce->name);
2006 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2007 return -1;
2009 strbuf_release(&previous_name_buf);
2011 /* Write extension data here */
2012 if (!strip_extensions && istate->split_index) {
2013 struct strbuf sb = STRBUF_INIT;
2015 err = write_link_extension(&sb, istate) < 0 ||
2016 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2017 sb.len) < 0 ||
2018 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2019 strbuf_release(&sb);
2020 if (err)
2021 return -1;
2023 if (!strip_extensions && istate->cache_tree) {
2024 struct strbuf sb = STRBUF_INIT;
2026 cache_tree_write(&sb, istate->cache_tree);
2027 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2028 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2029 strbuf_release(&sb);
2030 if (err)
2031 return -1;
2033 if (!strip_extensions && istate->resolve_undo) {
2034 struct strbuf sb = STRBUF_INIT;
2036 resolve_undo_write(&sb, istate->resolve_undo);
2037 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2038 sb.len) < 0
2039 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2040 strbuf_release(&sb);
2041 if (err)
2042 return -1;
2045 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2046 return -1;
2047 istate->timestamp.sec = (unsigned int)st.st_mtime;
2048 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2049 return 0;
2052 void set_alternate_index_output(const char *name)
2054 alternate_index_output = name;
2057 static int commit_locked_index(struct lock_file *lk)
2059 if (alternate_index_output)
2060 return commit_lock_file_to(lk, alternate_index_output);
2061 else
2062 return commit_lock_file(lk);
2065 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2066 unsigned flags)
2068 int ret = do_write_index(istate, lock->fd, 0);
2069 if (ret)
2070 return ret;
2071 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2072 (COMMIT_LOCK | CLOSE_LOCK));
2073 if (flags & COMMIT_LOCK)
2074 return commit_locked_index(lock);
2075 else if (flags & CLOSE_LOCK)
2076 return close_lock_file(lock);
2077 else
2078 return ret;
2081 static int write_split_index(struct index_state *istate,
2082 struct lock_file *lock,
2083 unsigned flags)
2085 int ret;
2086 prepare_to_write_split_index(istate);
2087 ret = do_write_locked_index(istate, lock, flags);
2088 finish_writing_split_index(istate);
2089 return ret;
2092 static char *temporary_sharedindex;
2094 static void remove_temporary_sharedindex(void)
2096 if (temporary_sharedindex) {
2097 unlink_or_warn(temporary_sharedindex);
2098 free(temporary_sharedindex);
2099 temporary_sharedindex = NULL;
2103 static void remove_temporary_sharedindex_on_signal(int signo)
2105 remove_temporary_sharedindex();
2106 sigchain_pop(signo);
2107 raise(signo);
2110 static int write_shared_index(struct index_state *istate,
2111 struct lock_file *lock, unsigned flags)
2113 struct split_index *si = istate->split_index;
2114 static int installed_handler;
2115 int fd, ret;
2117 temporary_sharedindex = git_pathdup("sharedindex_XXXXXX");
2118 fd = mkstemp(temporary_sharedindex);
2119 if (fd < 0) {
2120 free(temporary_sharedindex);
2121 temporary_sharedindex = NULL;
2122 hashclr(si->base_sha1);
2123 return do_write_locked_index(istate, lock, flags);
2125 if (!installed_handler) {
2126 atexit(remove_temporary_sharedindex);
2127 sigchain_push_common(remove_temporary_sharedindex_on_signal);
2129 move_cache_to_base_index(istate);
2130 ret = do_write_index(si->base, fd, 1);
2131 close(fd);
2132 if (ret) {
2133 remove_temporary_sharedindex();
2134 return ret;
2136 ret = rename(temporary_sharedindex,
2137 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2138 free(temporary_sharedindex);
2139 temporary_sharedindex = NULL;
2140 if (!ret)
2141 hashcpy(si->base_sha1, si->base->sha1);
2142 return ret;
2145 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2146 unsigned flags)
2148 struct split_index *si = istate->split_index;
2150 if (!si || alternate_index_output ||
2151 (istate->cache_changed & ~EXTMASK)) {
2152 if (si)
2153 hashclr(si->base_sha1);
2154 return do_write_locked_index(istate, lock, flags);
2157 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2158 int v = si->base_sha1[0];
2159 if ((v & 15) < 6)
2160 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2162 if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2163 int ret = write_shared_index(istate, lock, flags);
2164 if (ret)
2165 return ret;
2168 return write_split_index(istate, lock, flags);
2172 * Read the index file that is potentially unmerged into given
2173 * index_state, dropping any unmerged entries. Returns true if
2174 * the index is unmerged. Callers who want to refuse to work
2175 * from an unmerged state can call this and check its return value,
2176 * instead of calling read_cache().
2178 int read_index_unmerged(struct index_state *istate)
2180 int i;
2181 int unmerged = 0;
2183 read_index(istate);
2184 for (i = 0; i < istate->cache_nr; i++) {
2185 struct cache_entry *ce = istate->cache[i];
2186 struct cache_entry *new_ce;
2187 int size, len;
2189 if (!ce_stage(ce))
2190 continue;
2191 unmerged = 1;
2192 len = ce_namelen(ce);
2193 size = cache_entry_size(len);
2194 new_ce = xcalloc(1, size);
2195 memcpy(new_ce->name, ce->name, len);
2196 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2197 new_ce->ce_namelen = len;
2198 new_ce->ce_mode = ce->ce_mode;
2199 if (add_index_entry(istate, new_ce, 0))
2200 return error("%s: cannot drop to stage #0",
2201 new_ce->name);
2203 return unmerged;
2207 * Returns 1 if the path is an "other" path with respect to
2208 * the index; that is, the path is not mentioned in the index at all,
2209 * either as a file, a directory with some files in the index,
2210 * or as an unmerged entry.
2212 * We helpfully remove a trailing "/" from directories so that
2213 * the output of read_directory can be used as-is.
2215 int index_name_is_other(const struct index_state *istate, const char *name,
2216 int namelen)
2218 int pos;
2219 if (namelen && name[namelen - 1] == '/')
2220 namelen--;
2221 pos = index_name_pos(istate, name, namelen);
2222 if (0 <= pos)
2223 return 0; /* exact match */
2224 pos = -pos - 1;
2225 if (pos < istate->cache_nr) {
2226 struct cache_entry *ce = istate->cache[pos];
2227 if (ce_namelen(ce) == namelen &&
2228 !memcmp(ce->name, name, namelen))
2229 return 0; /* Yup, this one exists unmerged */
2231 return 1;
2234 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2236 int pos, len;
2237 unsigned long sz;
2238 enum object_type type;
2239 void *data;
2241 len = strlen(path);
2242 pos = index_name_pos(istate, path, len);
2243 if (pos < 0) {
2245 * We might be in the middle of a merge, in which
2246 * case we would read stage #2 (ours).
2248 int i;
2249 for (i = -pos - 1;
2250 (pos < 0 && i < istate->cache_nr &&
2251 !strcmp(istate->cache[i]->name, path));
2252 i++)
2253 if (ce_stage(istate->cache[i]) == 2)
2254 pos = i;
2256 if (pos < 0)
2257 return NULL;
2258 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2259 if (!data || type != OBJ_BLOB) {
2260 free(data);
2261 return NULL;
2263 if (size)
2264 *size = sz;
2265 return data;
2268 void stat_validity_clear(struct stat_validity *sv)
2270 free(sv->sd);
2271 sv->sd = NULL;
2274 int stat_validity_check(struct stat_validity *sv, const char *path)
2276 struct stat st;
2278 if (stat(path, &st) < 0)
2279 return sv->sd == NULL;
2280 if (!sv->sd)
2281 return 0;
2282 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2285 void stat_validity_update(struct stat_validity *sv, int fd)
2287 struct stat st;
2289 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2290 stat_validity_clear(sv);
2291 else {
2292 if (!sv->sd)
2293 sv->sd = xcalloc(1, sizeof(struct stat_data));
2294 fill_stat_data(sv->sd, &st);