Document LF appearing in shallow command during send-pack/receive-pack
[git/git-svn.git] / read-cache.c
blob4b4effd64b842173d6a06656935436515c755242
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "cache-tree.h"
9 #include "refs.h"
10 #include "dir.h"
11 #include "tree.h"
12 #include "commit.h"
13 #include "blob.h"
14 #include "resolve-undo.h"
15 #include "strbuf.h"
16 #include "varint.h"
18 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
19 unsigned int options);
21 /* Mask for the name length in ce_flags in the on-disk index */
23 #define CE_NAMEMASK (0x0fff)
25 /* Index extensions.
27 * The first letter should be 'A'..'Z' for extensions that are not
28 * necessary for a correct operation (i.e. optimization data).
29 * When new extensions are added that _needs_ to be understood in
30 * order to correctly interpret the index file, pick character that
31 * is outside the range, to cause the reader to abort.
34 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
35 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
36 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 struct index_state the_index;
40 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
42 istate->cache[nr] = ce;
43 add_name_hash(istate, ce);
46 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
48 struct cache_entry *old = istate->cache[nr];
50 remove_name_hash(istate, old);
51 set_index_entry(istate, nr, ce);
52 istate->cache_changed = 1;
55 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
57 struct cache_entry *old = istate->cache[nr], *new;
58 int namelen = strlen(new_name);
60 new = xmalloc(cache_entry_size(namelen));
61 copy_cache_entry(new, old);
62 new->ce_flags &= ~CE_STATE_MASK;
63 new->ce_namelen = namelen;
64 memcpy(new->name, new_name, namelen + 1);
66 cache_tree_invalidate_path(istate->cache_tree, old->name);
67 remove_index_entry_at(istate, nr);
68 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
71 void fill_stat_data(struct stat_data *sd, struct stat *st)
73 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
74 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
75 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
76 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
77 sd->sd_dev = st->st_dev;
78 sd->sd_ino = st->st_ino;
79 sd->sd_uid = st->st_uid;
80 sd->sd_gid = st->st_gid;
81 sd->sd_size = st->st_size;
84 int match_stat_data(const struct stat_data *sd, struct stat *st)
86 int changed = 0;
88 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
89 changed |= MTIME_CHANGED;
90 if (trust_ctime && check_stat &&
91 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
92 changed |= CTIME_CHANGED;
94 #ifdef USE_NSEC
95 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
96 changed |= MTIME_CHANGED;
97 if (trust_ctime && check_stat &&
98 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
99 changed |= CTIME_CHANGED;
100 #endif
102 if (check_stat) {
103 if (sd->sd_uid != (unsigned int) st->st_uid ||
104 sd->sd_gid != (unsigned int) st->st_gid)
105 changed |= OWNER_CHANGED;
106 if (sd->sd_ino != (unsigned int) st->st_ino)
107 changed |= INODE_CHANGED;
110 #ifdef USE_STDEV
112 * st_dev breaks on network filesystems where different
113 * clients will have different views of what "device"
114 * the filesystem is on
116 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
117 changed |= INODE_CHANGED;
118 #endif
120 if (sd->sd_size != (unsigned int) st->st_size)
121 changed |= DATA_CHANGED;
123 return changed;
127 * This only updates the "non-critical" parts of the directory
128 * cache, ie the parts that aren't tracked by GIT, and only used
129 * to validate the cache.
131 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
133 fill_stat_data(&ce->ce_stat_data, st);
135 if (assume_unchanged)
136 ce->ce_flags |= CE_VALID;
138 if (S_ISREG(st->st_mode))
139 ce_mark_uptodate(ce);
142 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
144 int match = -1;
145 int fd = open(ce->name, O_RDONLY);
147 if (fd >= 0) {
148 unsigned char sha1[20];
149 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
150 match = hashcmp(sha1, ce->sha1);
151 /* index_fd() closed the file descriptor already */
153 return match;
156 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
158 int match = -1;
159 void *buffer;
160 unsigned long size;
161 enum object_type type;
162 struct strbuf sb = STRBUF_INIT;
164 if (strbuf_readlink(&sb, ce->name, expected_size))
165 return -1;
167 buffer = read_sha1_file(ce->sha1, &type, &size);
168 if (buffer) {
169 if (size == sb.len)
170 match = memcmp(buffer, sb.buf, size);
171 free(buffer);
173 strbuf_release(&sb);
174 return match;
177 static int ce_compare_gitlink(const struct cache_entry *ce)
179 unsigned char sha1[20];
182 * We don't actually require that the .git directory
183 * under GITLINK directory be a valid git directory. It
184 * might even be missing (in case nobody populated that
185 * sub-project).
187 * If so, we consider it always to match.
189 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
190 return 0;
191 return hashcmp(sha1, ce->sha1);
194 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
196 switch (st->st_mode & S_IFMT) {
197 case S_IFREG:
198 if (ce_compare_data(ce, st))
199 return DATA_CHANGED;
200 break;
201 case S_IFLNK:
202 if (ce_compare_link(ce, xsize_t(st->st_size)))
203 return DATA_CHANGED;
204 break;
205 case S_IFDIR:
206 if (S_ISGITLINK(ce->ce_mode))
207 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
208 default:
209 return TYPE_CHANGED;
211 return 0;
214 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
216 unsigned int changed = 0;
218 if (ce->ce_flags & CE_REMOVE)
219 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
221 switch (ce->ce_mode & S_IFMT) {
222 case S_IFREG:
223 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
224 /* We consider only the owner x bit to be relevant for
225 * "mode changes"
227 if (trust_executable_bit &&
228 (0100 & (ce->ce_mode ^ st->st_mode)))
229 changed |= MODE_CHANGED;
230 break;
231 case S_IFLNK:
232 if (!S_ISLNK(st->st_mode) &&
233 (has_symlinks || !S_ISREG(st->st_mode)))
234 changed |= TYPE_CHANGED;
235 break;
236 case S_IFGITLINK:
237 /* We ignore most of the st_xxx fields for gitlinks */
238 if (!S_ISDIR(st->st_mode))
239 changed |= TYPE_CHANGED;
240 else if (ce_compare_gitlink(ce))
241 changed |= DATA_CHANGED;
242 return changed;
243 default:
244 die("internal error: ce_mode is %o", ce->ce_mode);
247 changed |= match_stat_data(&ce->ce_stat_data, st);
249 /* Racily smudged entry? */
250 if (!ce->ce_stat_data.sd_size) {
251 if (!is_empty_blob_sha1(ce->sha1))
252 changed |= DATA_CHANGED;
255 return changed;
258 static int is_racy_timestamp(const struct index_state *istate,
259 const struct cache_entry *ce)
261 return (!S_ISGITLINK(ce->ce_mode) &&
262 istate->timestamp.sec &&
263 #ifdef USE_NSEC
264 /* nanosecond timestamped files can also be racy! */
265 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
266 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
267 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
268 #else
269 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
270 #endif
274 int ie_match_stat(const struct index_state *istate,
275 const struct cache_entry *ce, struct stat *st,
276 unsigned int options)
278 unsigned int changed;
279 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
280 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
281 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
284 * If it's marked as always valid in the index, it's
285 * valid whatever the checked-out copy says.
287 * skip-worktree has the same effect with higher precedence
289 if (!ignore_skip_worktree && ce_skip_worktree(ce))
290 return 0;
291 if (!ignore_valid && (ce->ce_flags & CE_VALID))
292 return 0;
295 * Intent-to-add entries have not been added, so the index entry
296 * by definition never matches what is in the work tree until it
297 * actually gets added.
299 if (ce->ce_flags & CE_INTENT_TO_ADD)
300 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
302 changed = ce_match_stat_basic(ce, st);
305 * Within 1 second of this sequence:
306 * echo xyzzy >file && git-update-index --add file
307 * running this command:
308 * echo frotz >file
309 * would give a falsely clean cache entry. The mtime and
310 * length match the cache, and other stat fields do not change.
312 * We could detect this at update-index time (the cache entry
313 * being registered/updated records the same time as "now")
314 * and delay the return from git-update-index, but that would
315 * effectively mean we can make at most one commit per second,
316 * which is not acceptable. Instead, we check cache entries
317 * whose mtime are the same as the index file timestamp more
318 * carefully than others.
320 if (!changed && is_racy_timestamp(istate, ce)) {
321 if (assume_racy_is_modified)
322 changed |= DATA_CHANGED;
323 else
324 changed |= ce_modified_check_fs(ce, st);
327 return changed;
330 int ie_modified(const struct index_state *istate,
331 const struct cache_entry *ce,
332 struct stat *st, unsigned int options)
334 int changed, changed_fs;
336 changed = ie_match_stat(istate, ce, st, options);
337 if (!changed)
338 return 0;
340 * If the mode or type has changed, there's no point in trying
341 * to refresh the entry - it's not going to match
343 if (changed & (MODE_CHANGED | TYPE_CHANGED))
344 return changed;
347 * Immediately after read-tree or update-index --cacheinfo,
348 * the length field is zero, as we have never even read the
349 * lstat(2) information once, and we cannot trust DATA_CHANGED
350 * returned by ie_match_stat() which in turn was returned by
351 * ce_match_stat_basic() to signal that the filesize of the
352 * blob changed. We have to actually go to the filesystem to
353 * see if the contents match, and if so, should answer "unchanged".
355 * The logic does not apply to gitlinks, as ce_match_stat_basic()
356 * already has checked the actual HEAD from the filesystem in the
357 * subproject. If ie_match_stat() already said it is different,
358 * then we know it is.
360 if ((changed & DATA_CHANGED) &&
361 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
362 return changed;
364 changed_fs = ce_modified_check_fs(ce, st);
365 if (changed_fs)
366 return changed | changed_fs;
367 return 0;
370 int base_name_compare(const char *name1, int len1, int mode1,
371 const char *name2, int len2, int mode2)
373 unsigned char c1, c2;
374 int len = len1 < len2 ? len1 : len2;
375 int cmp;
377 cmp = memcmp(name1, name2, len);
378 if (cmp)
379 return cmp;
380 c1 = name1[len];
381 c2 = name2[len];
382 if (!c1 && S_ISDIR(mode1))
383 c1 = '/';
384 if (!c2 && S_ISDIR(mode2))
385 c2 = '/';
386 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
390 * df_name_compare() is identical to base_name_compare(), except it
391 * compares conflicting directory/file entries as equal. Note that
392 * while a directory name compares as equal to a regular file, they
393 * then individually compare _differently_ to a filename that has
394 * a dot after the basename (because '\0' < '.' < '/').
396 * This is used by routines that want to traverse the git namespace
397 * but then handle conflicting entries together when possible.
399 int df_name_compare(const char *name1, int len1, int mode1,
400 const char *name2, int len2, int mode2)
402 int len = len1 < len2 ? len1 : len2, cmp;
403 unsigned char c1, c2;
405 cmp = memcmp(name1, name2, len);
406 if (cmp)
407 return cmp;
408 /* Directories and files compare equal (same length, same name) */
409 if (len1 == len2)
410 return 0;
411 c1 = name1[len];
412 if (!c1 && S_ISDIR(mode1))
413 c1 = '/';
414 c2 = name2[len];
415 if (!c2 && S_ISDIR(mode2))
416 c2 = '/';
417 if (c1 == '/' && !c2)
418 return 0;
419 if (c2 == '/' && !c1)
420 return 0;
421 return c1 - c2;
424 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
426 int len = len1 < len2 ? len1 : len2;
427 int cmp;
429 cmp = memcmp(name1, name2, len);
430 if (cmp)
431 return cmp;
432 if (len1 < len2)
433 return -1;
434 if (len1 > len2)
435 return 1;
437 if (stage1 < stage2)
438 return -1;
439 if (stage1 > stage2)
440 return 1;
441 return 0;
444 int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
446 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
449 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
451 int first, last;
453 first = 0;
454 last = istate->cache_nr;
455 while (last > first) {
456 int next = (last + first) >> 1;
457 struct cache_entry *ce = istate->cache[next];
458 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
459 if (!cmp)
460 return next;
461 if (cmp < 0) {
462 last = next;
463 continue;
465 first = next+1;
467 return -first-1;
470 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
472 return index_name_stage_pos(istate, name, namelen, 0);
475 /* Remove entry, return true if there are more entries to go.. */
476 int remove_index_entry_at(struct index_state *istate, int pos)
478 struct cache_entry *ce = istate->cache[pos];
480 record_resolve_undo(istate, ce);
481 remove_name_hash(istate, ce);
482 istate->cache_changed = 1;
483 istate->cache_nr--;
484 if (pos >= istate->cache_nr)
485 return 0;
486 memmove(istate->cache + pos,
487 istate->cache + pos + 1,
488 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
489 return 1;
493 * Remove all cache entries marked for removal, that is where
494 * CE_REMOVE is set in ce_flags. This is much more effective than
495 * calling remove_index_entry_at() for each entry to be removed.
497 void remove_marked_cache_entries(struct index_state *istate)
499 struct cache_entry **ce_array = istate->cache;
500 unsigned int i, j;
502 for (i = j = 0; i < istate->cache_nr; i++) {
503 if (ce_array[i]->ce_flags & CE_REMOVE)
504 remove_name_hash(istate, ce_array[i]);
505 else
506 ce_array[j++] = ce_array[i];
508 istate->cache_changed = 1;
509 istate->cache_nr = j;
512 int remove_file_from_index(struct index_state *istate, const char *path)
514 int pos = index_name_pos(istate, path, strlen(path));
515 if (pos < 0)
516 pos = -pos-1;
517 cache_tree_invalidate_path(istate->cache_tree, path);
518 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
519 remove_index_entry_at(istate, pos);
520 return 0;
523 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
525 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
528 static int index_name_pos_also_unmerged(struct index_state *istate,
529 const char *path, int namelen)
531 int pos = index_name_pos(istate, path, namelen);
532 struct cache_entry *ce;
534 if (pos >= 0)
535 return pos;
537 /* maybe unmerged? */
538 pos = -1 - pos;
539 if (pos >= istate->cache_nr ||
540 compare_name((ce = istate->cache[pos]), path, namelen))
541 return -1;
543 /* order of preference: stage 2, 1, 3 */
544 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
545 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
546 !compare_name(ce, path, namelen))
547 pos++;
548 return pos;
551 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
553 int len = ce_namelen(ce);
554 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
558 * If we add a filename that aliases in the cache, we will use the
559 * name that we already have - but we don't want to update the same
560 * alias twice, because that implies that there were actually two
561 * different files with aliasing names!
563 * So we use the CE_ADDED flag to verify that the alias was an old
564 * one before we accept it as
566 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
568 int len;
569 struct cache_entry *new;
571 if (alias->ce_flags & CE_ADDED)
572 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
574 /* Ok, create the new entry using the name of the existing alias */
575 len = ce_namelen(alias);
576 new = xcalloc(1, cache_entry_size(len));
577 memcpy(new->name, alias->name, len);
578 copy_cache_entry(new, ce);
579 free(ce);
580 return new;
583 static void record_intent_to_add(struct cache_entry *ce)
585 unsigned char sha1[20];
586 if (write_sha1_file("", 0, blob_type, sha1))
587 die("cannot create an empty blob in the object database");
588 hashcpy(ce->sha1, sha1);
591 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
593 int size, namelen, was_same;
594 mode_t st_mode = st->st_mode;
595 struct cache_entry *ce, *alias;
596 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
597 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
598 int pretend = flags & ADD_CACHE_PRETEND;
599 int intent_only = flags & ADD_CACHE_INTENT;
600 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
601 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
603 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
604 return error("%s: can only add regular files, symbolic links or git-directories", path);
606 namelen = strlen(path);
607 if (S_ISDIR(st_mode)) {
608 while (namelen && path[namelen-1] == '/')
609 namelen--;
611 size = cache_entry_size(namelen);
612 ce = xcalloc(1, size);
613 memcpy(ce->name, path, namelen);
614 ce->ce_namelen = namelen;
615 if (!intent_only)
616 fill_stat_cache_info(ce, st);
617 else
618 ce->ce_flags |= CE_INTENT_TO_ADD;
620 if (trust_executable_bit && has_symlinks)
621 ce->ce_mode = create_ce_mode(st_mode);
622 else {
623 /* If there is an existing entry, pick the mode bits and type
624 * from it, otherwise assume unexecutable regular file.
626 struct cache_entry *ent;
627 int pos = index_name_pos_also_unmerged(istate, path, namelen);
629 ent = (0 <= pos) ? istate->cache[pos] : NULL;
630 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
633 /* When core.ignorecase=true, determine if a directory of the same name but differing
634 * case already exists within the Git repository. If it does, ensure the directory
635 * case of the file being added to the repository matches (is folded into) the existing
636 * entry's directory case.
638 if (ignore_case) {
639 const char *startPtr = ce->name;
640 const char *ptr = startPtr;
641 while (*ptr) {
642 while (*ptr && *ptr != '/')
643 ++ptr;
644 if (*ptr == '/') {
645 struct cache_entry *foundce;
646 ++ptr;
647 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
648 if (foundce) {
649 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
650 startPtr = ptr;
656 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
657 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
658 /* Nothing changed, really */
659 free(ce);
660 if (!S_ISGITLINK(alias->ce_mode))
661 ce_mark_uptodate(alias);
662 alias->ce_flags |= CE_ADDED;
663 return 0;
665 if (!intent_only) {
666 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
667 return error("unable to index file %s", path);
668 } else
669 record_intent_to_add(ce);
671 if (ignore_case && alias && different_name(ce, alias))
672 ce = create_alias_ce(ce, alias);
673 ce->ce_flags |= CE_ADDED;
675 /* It was suspected to be racily clean, but it turns out to be Ok */
676 was_same = (alias &&
677 !ce_stage(alias) &&
678 !hashcmp(alias->sha1, ce->sha1) &&
679 ce->ce_mode == alias->ce_mode);
681 if (pretend)
683 else if (add_index_entry(istate, ce, add_option))
684 return error("unable to add %s to index",path);
685 if (verbose && !was_same)
686 printf("add '%s'\n", path);
687 return 0;
690 int add_file_to_index(struct index_state *istate, const char *path, int flags)
692 struct stat st;
693 if (lstat(path, &st))
694 die_errno("unable to stat '%s'", path);
695 return add_to_index(istate, path, &st, flags);
698 struct cache_entry *make_cache_entry(unsigned int mode,
699 const unsigned char *sha1, const char *path, int stage,
700 unsigned int refresh_options)
702 int size, len;
703 struct cache_entry *ce;
705 if (!verify_path(path)) {
706 error("Invalid path '%s'", path);
707 return NULL;
710 len = strlen(path);
711 size = cache_entry_size(len);
712 ce = xcalloc(1, size);
714 hashcpy(ce->sha1, sha1);
715 memcpy(ce->name, path, len);
716 ce->ce_flags = create_ce_flags(stage);
717 ce->ce_namelen = len;
718 ce->ce_mode = create_ce_mode(mode);
720 return refresh_cache_entry(ce, refresh_options);
723 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
725 int len = ce_namelen(a);
726 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
730 * We fundamentally don't like some paths: we don't want
731 * dot or dot-dot anywhere, and for obvious reasons don't
732 * want to recurse into ".git" either.
734 * Also, we don't want double slashes or slashes at the
735 * end that can make pathnames ambiguous.
737 static int verify_dotfile(const char *rest)
740 * The first character was '.', but that
741 * has already been discarded, we now test
742 * the rest.
745 /* "." is not allowed */
746 if (*rest == '\0' || is_dir_sep(*rest))
747 return 0;
749 switch (*rest) {
751 * ".git" followed by NUL or slash is bad. This
752 * shares the path end test with the ".." case.
754 case 'g':
755 if (rest[1] != 'i')
756 break;
757 if (rest[2] != 't')
758 break;
759 rest += 2;
760 /* fallthrough */
761 case '.':
762 if (rest[1] == '\0' || is_dir_sep(rest[1]))
763 return 0;
765 return 1;
768 int verify_path(const char *path)
770 char c;
772 if (has_dos_drive_prefix(path))
773 return 0;
775 goto inside;
776 for (;;) {
777 if (!c)
778 return 1;
779 if (is_dir_sep(c)) {
780 inside:
781 c = *path++;
782 if ((c == '.' && !verify_dotfile(path)) ||
783 is_dir_sep(c) || c == '\0')
784 return 0;
786 c = *path++;
791 * Do we have another file that has the beginning components being a
792 * proper superset of the name we're trying to add?
794 static int has_file_name(struct index_state *istate,
795 const struct cache_entry *ce, int pos, int ok_to_replace)
797 int retval = 0;
798 int len = ce_namelen(ce);
799 int stage = ce_stage(ce);
800 const char *name = ce->name;
802 while (pos < istate->cache_nr) {
803 struct cache_entry *p = istate->cache[pos++];
805 if (len >= ce_namelen(p))
806 break;
807 if (memcmp(name, p->name, len))
808 break;
809 if (ce_stage(p) != stage)
810 continue;
811 if (p->name[len] != '/')
812 continue;
813 if (p->ce_flags & CE_REMOVE)
814 continue;
815 retval = -1;
816 if (!ok_to_replace)
817 break;
818 remove_index_entry_at(istate, --pos);
820 return retval;
824 * Do we have another file with a pathname that is a proper
825 * subset of the name we're trying to add?
827 static int has_dir_name(struct index_state *istate,
828 const struct cache_entry *ce, int pos, int ok_to_replace)
830 int retval = 0;
831 int stage = ce_stage(ce);
832 const char *name = ce->name;
833 const char *slash = name + ce_namelen(ce);
835 for (;;) {
836 int len;
838 for (;;) {
839 if (*--slash == '/')
840 break;
841 if (slash <= ce->name)
842 return retval;
844 len = slash - name;
846 pos = index_name_stage_pos(istate, name, len, stage);
847 if (pos >= 0) {
849 * Found one, but not so fast. This could
850 * be a marker that says "I was here, but
851 * I am being removed". Such an entry is
852 * not a part of the resulting tree, and
853 * it is Ok to have a directory at the same
854 * path.
856 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
857 retval = -1;
858 if (!ok_to_replace)
859 break;
860 remove_index_entry_at(istate, pos);
861 continue;
864 else
865 pos = -pos-1;
868 * Trivial optimization: if we find an entry that
869 * already matches the sub-directory, then we know
870 * we're ok, and we can exit.
872 while (pos < istate->cache_nr) {
873 struct cache_entry *p = istate->cache[pos];
874 if ((ce_namelen(p) <= len) ||
875 (p->name[len] != '/') ||
876 memcmp(p->name, name, len))
877 break; /* not our subdirectory */
878 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
880 * p is at the same stage as our entry, and
881 * is a subdirectory of what we are looking
882 * at, so we cannot have conflicts at our
883 * level or anything shorter.
885 return retval;
886 pos++;
889 return retval;
892 /* We may be in a situation where we already have path/file and path
893 * is being added, or we already have path and path/file is being
894 * added. Either one would result in a nonsense tree that has path
895 * twice when git-write-tree tries to write it out. Prevent it.
897 * If ok-to-replace is specified, we remove the conflicting entries
898 * from the cache so the caller should recompute the insert position.
899 * When this happens, we return non-zero.
901 static int check_file_directory_conflict(struct index_state *istate,
902 const struct cache_entry *ce,
903 int pos, int ok_to_replace)
905 int retval;
908 * When ce is an "I am going away" entry, we allow it to be added
910 if (ce->ce_flags & CE_REMOVE)
911 return 0;
914 * We check if the path is a sub-path of a subsequent pathname
915 * first, since removing those will not change the position
916 * in the array.
918 retval = has_file_name(istate, ce, pos, ok_to_replace);
921 * Then check if the path might have a clashing sub-directory
922 * before it.
924 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
927 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
929 int pos;
930 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
931 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
932 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
933 int new_only = option & ADD_CACHE_NEW_ONLY;
935 cache_tree_invalidate_path(istate->cache_tree, ce->name);
936 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
938 /* existing match? Just replace it. */
939 if (pos >= 0) {
940 if (!new_only)
941 replace_index_entry(istate, pos, ce);
942 return 0;
944 pos = -pos-1;
947 * Inserting a merged entry ("stage 0") into the index
948 * will always replace all non-merged entries..
950 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
951 while (ce_same_name(istate->cache[pos], ce)) {
952 ok_to_add = 1;
953 if (!remove_index_entry_at(istate, pos))
954 break;
958 if (!ok_to_add)
959 return -1;
960 if (!verify_path(ce->name))
961 return error("Invalid path '%s'", ce->name);
963 if (!skip_df_check &&
964 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
965 if (!ok_to_replace)
966 return error("'%s' appears as both a file and as a directory",
967 ce->name);
968 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
969 pos = -pos-1;
971 return pos + 1;
974 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
976 int pos;
978 if (option & ADD_CACHE_JUST_APPEND)
979 pos = istate->cache_nr;
980 else {
981 int ret;
982 ret = add_index_entry_with_check(istate, ce, option);
983 if (ret <= 0)
984 return ret;
985 pos = ret - 1;
988 /* Make sure the array is big enough .. */
989 if (istate->cache_nr == istate->cache_alloc) {
990 istate->cache_alloc = alloc_nr(istate->cache_alloc);
991 istate->cache = xrealloc(istate->cache,
992 istate->cache_alloc * sizeof(*istate->cache));
995 /* Add it in.. */
996 istate->cache_nr++;
997 if (istate->cache_nr > pos + 1)
998 memmove(istate->cache + pos + 1,
999 istate->cache + pos,
1000 (istate->cache_nr - pos - 1) * sizeof(ce));
1001 set_index_entry(istate, pos, ce);
1002 istate->cache_changed = 1;
1003 return 0;
1007 * "refresh" does not calculate a new sha1 file or bring the
1008 * cache up-to-date for mode/content changes. But what it
1009 * _does_ do is to "re-match" the stat information of a file
1010 * with the cache, so that you can refresh the cache for a
1011 * file that hasn't been changed but where the stat entry is
1012 * out of date.
1014 * For example, you'd want to do this after doing a "git-read-tree",
1015 * to link up the stat cache details with the proper files.
1017 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1018 struct cache_entry *ce,
1019 unsigned int options, int *err,
1020 int *changed_ret)
1022 struct stat st;
1023 struct cache_entry *updated;
1024 int changed, size;
1025 int refresh = options & CE_MATCH_REFRESH;
1026 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1027 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1028 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1030 if (!refresh || ce_uptodate(ce))
1031 return ce;
1034 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1035 * that the change to the work tree does not matter and told
1036 * us not to worry.
1038 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1039 ce_mark_uptodate(ce);
1040 return ce;
1042 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1043 ce_mark_uptodate(ce);
1044 return ce;
1047 if (lstat(ce->name, &st) < 0) {
1048 if (ignore_missing && errno == ENOENT)
1049 return ce;
1050 if (err)
1051 *err = errno;
1052 return NULL;
1055 changed = ie_match_stat(istate, ce, &st, options);
1056 if (changed_ret)
1057 *changed_ret = changed;
1058 if (!changed) {
1060 * The path is unchanged. If we were told to ignore
1061 * valid bit, then we did the actual stat check and
1062 * found that the entry is unmodified. If the entry
1063 * is not marked VALID, this is the place to mark it
1064 * valid again, under "assume unchanged" mode.
1066 if (ignore_valid && assume_unchanged &&
1067 !(ce->ce_flags & CE_VALID))
1068 ; /* mark this one VALID again */
1069 else {
1071 * We do not mark the index itself "modified"
1072 * because CE_UPTODATE flag is in-core only;
1073 * we are not going to write this change out.
1075 if (!S_ISGITLINK(ce->ce_mode))
1076 ce_mark_uptodate(ce);
1077 return ce;
1081 if (ie_modified(istate, ce, &st, options)) {
1082 if (err)
1083 *err = EINVAL;
1084 return NULL;
1087 size = ce_size(ce);
1088 updated = xmalloc(size);
1089 memcpy(updated, ce, size);
1090 fill_stat_cache_info(updated, &st);
1092 * If ignore_valid is not set, we should leave CE_VALID bit
1093 * alone. Otherwise, paths marked with --no-assume-unchanged
1094 * (i.e. things to be edited) will reacquire CE_VALID bit
1095 * automatically, which is not really what we want.
1097 if (!ignore_valid && assume_unchanged &&
1098 !(ce->ce_flags & CE_VALID))
1099 updated->ce_flags &= ~CE_VALID;
1101 return updated;
1104 static void show_file(const char * fmt, const char * name, int in_porcelain,
1105 int * first, const char *header_msg)
1107 if (in_porcelain && *first && header_msg) {
1108 printf("%s\n", header_msg);
1109 *first = 0;
1111 printf(fmt, name);
1114 int refresh_index(struct index_state *istate, unsigned int flags,
1115 const struct pathspec *pathspec,
1116 char *seen, const char *header_msg)
1118 int i;
1119 int has_errors = 0;
1120 int really = (flags & REFRESH_REALLY) != 0;
1121 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1122 int quiet = (flags & REFRESH_QUIET) != 0;
1123 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1124 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1125 int first = 1;
1126 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1127 unsigned int options = (CE_MATCH_REFRESH |
1128 (really ? CE_MATCH_IGNORE_VALID : 0) |
1129 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1130 const char *modified_fmt;
1131 const char *deleted_fmt;
1132 const char *typechange_fmt;
1133 const char *added_fmt;
1134 const char *unmerged_fmt;
1136 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1137 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1138 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1139 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1140 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1141 for (i = 0; i < istate->cache_nr; i++) {
1142 struct cache_entry *ce, *new;
1143 int cache_errno = 0;
1144 int changed = 0;
1145 int filtered = 0;
1147 ce = istate->cache[i];
1148 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1149 continue;
1151 if (pathspec && !ce_path_match(ce, pathspec, seen))
1152 filtered = 1;
1154 if (ce_stage(ce)) {
1155 while ((i < istate->cache_nr) &&
1156 ! strcmp(istate->cache[i]->name, ce->name))
1157 i++;
1158 i--;
1159 if (allow_unmerged)
1160 continue;
1161 if (!filtered)
1162 show_file(unmerged_fmt, ce->name, in_porcelain,
1163 &first, header_msg);
1164 has_errors = 1;
1165 continue;
1168 if (filtered)
1169 continue;
1171 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1172 if (new == ce)
1173 continue;
1174 if (!new) {
1175 const char *fmt;
1177 if (really && cache_errno == EINVAL) {
1178 /* If we are doing --really-refresh that
1179 * means the index is not valid anymore.
1181 ce->ce_flags &= ~CE_VALID;
1182 istate->cache_changed = 1;
1184 if (quiet)
1185 continue;
1187 if (cache_errno == ENOENT)
1188 fmt = deleted_fmt;
1189 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1190 fmt = added_fmt; /* must be before other checks */
1191 else if (changed & TYPE_CHANGED)
1192 fmt = typechange_fmt;
1193 else
1194 fmt = modified_fmt;
1195 show_file(fmt,
1196 ce->name, in_porcelain, &first, header_msg);
1197 has_errors = 1;
1198 continue;
1201 replace_index_entry(istate, i, new);
1203 return has_errors;
1206 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1207 unsigned int options)
1209 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1213 /*****************************************************************
1214 * Index File I/O
1215 *****************************************************************/
1217 #define INDEX_FORMAT_DEFAULT 3
1220 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1221 * Again - this is just a (very strong in practice) heuristic that
1222 * the inode hasn't changed.
1224 * We save the fields in big-endian order to allow using the
1225 * index file over NFS transparently.
1227 struct ondisk_cache_entry {
1228 struct cache_time ctime;
1229 struct cache_time mtime;
1230 uint32_t dev;
1231 uint32_t ino;
1232 uint32_t mode;
1233 uint32_t uid;
1234 uint32_t gid;
1235 uint32_t size;
1236 unsigned char sha1[20];
1237 uint16_t flags;
1238 char name[FLEX_ARRAY]; /* more */
1242 * This struct is used when CE_EXTENDED bit is 1
1243 * The struct must match ondisk_cache_entry exactly from
1244 * ctime till flags
1246 struct ondisk_cache_entry_extended {
1247 struct cache_time ctime;
1248 struct cache_time mtime;
1249 uint32_t dev;
1250 uint32_t ino;
1251 uint32_t mode;
1252 uint32_t uid;
1253 uint32_t gid;
1254 uint32_t size;
1255 unsigned char sha1[20];
1256 uint16_t flags;
1257 uint16_t flags2;
1258 char name[FLEX_ARRAY]; /* more */
1261 /* These are only used for v3 or lower */
1262 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1263 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1264 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1265 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1266 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1267 ondisk_cache_entry_size(ce_namelen(ce)))
1269 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1271 git_SHA_CTX c;
1272 unsigned char sha1[20];
1273 int hdr_version;
1275 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1276 return error("bad signature");
1277 hdr_version = ntohl(hdr->hdr_version);
1278 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1279 return error("bad index version %d", hdr_version);
1280 git_SHA1_Init(&c);
1281 git_SHA1_Update(&c, hdr, size - 20);
1282 git_SHA1_Final(sha1, &c);
1283 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1284 return error("bad index file sha1 signature");
1285 return 0;
1288 static int read_index_extension(struct index_state *istate,
1289 const char *ext, void *data, unsigned long sz)
1291 switch (CACHE_EXT(ext)) {
1292 case CACHE_EXT_TREE:
1293 istate->cache_tree = cache_tree_read(data, sz);
1294 break;
1295 case CACHE_EXT_RESOLVE_UNDO:
1296 istate->resolve_undo = resolve_undo_read(data, sz);
1297 break;
1298 default:
1299 if (*ext < 'A' || 'Z' < *ext)
1300 return error("index uses %.4s extension, which we do not understand",
1301 ext);
1302 fprintf(stderr, "ignoring %.4s extension\n", ext);
1303 break;
1305 return 0;
1308 int read_index(struct index_state *istate)
1310 return read_index_from(istate, get_index_file());
1313 #ifndef NEEDS_ALIGNED_ACCESS
1314 #define ntoh_s(var) ntohs(var)
1315 #define ntoh_l(var) ntohl(var)
1316 #else
1317 static inline uint16_t ntoh_s_force_align(void *p)
1319 uint16_t x;
1320 memcpy(&x, p, sizeof(x));
1321 return ntohs(x);
1323 static inline uint32_t ntoh_l_force_align(void *p)
1325 uint32_t x;
1326 memcpy(&x, p, sizeof(x));
1327 return ntohl(x);
1329 #define ntoh_s(var) ntoh_s_force_align(&(var))
1330 #define ntoh_l(var) ntoh_l_force_align(&(var))
1331 #endif
1333 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1334 unsigned int flags,
1335 const char *name,
1336 size_t len)
1338 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1340 ce->ce_stat_data.sd_ctime.sec = ntoh_l(ondisk->ctime.sec);
1341 ce->ce_stat_data.sd_mtime.sec = ntoh_l(ondisk->mtime.sec);
1342 ce->ce_stat_data.sd_ctime.nsec = ntoh_l(ondisk->ctime.nsec);
1343 ce->ce_stat_data.sd_mtime.nsec = ntoh_l(ondisk->mtime.nsec);
1344 ce->ce_stat_data.sd_dev = ntoh_l(ondisk->dev);
1345 ce->ce_stat_data.sd_ino = ntoh_l(ondisk->ino);
1346 ce->ce_mode = ntoh_l(ondisk->mode);
1347 ce->ce_stat_data.sd_uid = ntoh_l(ondisk->uid);
1348 ce->ce_stat_data.sd_gid = ntoh_l(ondisk->gid);
1349 ce->ce_stat_data.sd_size = ntoh_l(ondisk->size);
1350 ce->ce_flags = flags & ~CE_NAMEMASK;
1351 ce->ce_namelen = len;
1352 hashcpy(ce->sha1, ondisk->sha1);
1353 memcpy(ce->name, name, len);
1354 ce->name[len] = '\0';
1355 return ce;
1359 * Adjacent cache entries tend to share the leading paths, so it makes
1360 * sense to only store the differences in later entries. In the v4
1361 * on-disk format of the index, each on-disk cache entry stores the
1362 * number of bytes to be stripped from the end of the previous name,
1363 * and the bytes to append to the result, to come up with its name.
1365 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1367 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1368 size_t len = decode_varint(&cp);
1370 if (name->len < len)
1371 die("malformed name field in the index");
1372 strbuf_remove(name, name->len - len, len);
1373 for (ep = cp; *ep; ep++)
1374 ; /* find the end */
1375 strbuf_add(name, cp, ep - cp);
1376 return (const char *)ep + 1 - cp_;
1379 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1380 unsigned long *ent_size,
1381 struct strbuf *previous_name)
1383 struct cache_entry *ce;
1384 size_t len;
1385 const char *name;
1386 unsigned int flags;
1388 /* On-disk flags are just 16 bits */
1389 flags = ntoh_s(ondisk->flags);
1390 len = flags & CE_NAMEMASK;
1392 if (flags & CE_EXTENDED) {
1393 struct ondisk_cache_entry_extended *ondisk2;
1394 int extended_flags;
1395 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1396 extended_flags = ntoh_s(ondisk2->flags2) << 16;
1397 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1398 if (extended_flags & ~CE_EXTENDED_FLAGS)
1399 die("Unknown index entry format %08x", extended_flags);
1400 flags |= extended_flags;
1401 name = ondisk2->name;
1403 else
1404 name = ondisk->name;
1406 if (!previous_name) {
1407 /* v3 and earlier */
1408 if (len == CE_NAMEMASK)
1409 len = strlen(name);
1410 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1412 *ent_size = ondisk_ce_size(ce);
1413 } else {
1414 unsigned long consumed;
1415 consumed = expand_name_field(previous_name, name);
1416 ce = cache_entry_from_ondisk(ondisk, flags,
1417 previous_name->buf,
1418 previous_name->len);
1420 *ent_size = (name - ((char *)ondisk)) + consumed;
1422 return ce;
1425 /* remember to discard_cache() before reading a different cache! */
1426 int read_index_from(struct index_state *istate, const char *path)
1428 int fd, i;
1429 struct stat st;
1430 unsigned long src_offset;
1431 struct cache_header *hdr;
1432 void *mmap;
1433 size_t mmap_size;
1434 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1436 if (istate->initialized)
1437 return istate->cache_nr;
1439 istate->timestamp.sec = 0;
1440 istate->timestamp.nsec = 0;
1441 fd = open(path, O_RDONLY);
1442 if (fd < 0) {
1443 if (errno == ENOENT)
1444 return 0;
1445 die_errno("index file open failed");
1448 if (fstat(fd, &st))
1449 die_errno("cannot stat the open index");
1451 mmap_size = xsize_t(st.st_size);
1452 if (mmap_size < sizeof(struct cache_header) + 20)
1453 die("index file smaller than expected");
1455 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1456 if (mmap == MAP_FAILED)
1457 die_errno("unable to map index file");
1458 close(fd);
1460 hdr = mmap;
1461 if (verify_hdr(hdr, mmap_size) < 0)
1462 goto unmap;
1464 istate->version = ntohl(hdr->hdr_version);
1465 istate->cache_nr = ntohl(hdr->hdr_entries);
1466 istate->cache_alloc = alloc_nr(istate->cache_nr);
1467 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1468 istate->initialized = 1;
1470 if (istate->version == 4)
1471 previous_name = &previous_name_buf;
1472 else
1473 previous_name = NULL;
1475 src_offset = sizeof(*hdr);
1476 for (i = 0; i < istate->cache_nr; i++) {
1477 struct ondisk_cache_entry *disk_ce;
1478 struct cache_entry *ce;
1479 unsigned long consumed;
1481 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1482 ce = create_from_disk(disk_ce, &consumed, previous_name);
1483 set_index_entry(istate, i, ce);
1485 src_offset += consumed;
1487 strbuf_release(&previous_name_buf);
1488 istate->timestamp.sec = st.st_mtime;
1489 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1491 while (src_offset <= mmap_size - 20 - 8) {
1492 /* After an array of active_nr index entries,
1493 * there can be arbitrary number of extended
1494 * sections, each of which is prefixed with
1495 * extension name (4-byte) and section length
1496 * in 4-byte network byte order.
1498 uint32_t extsize;
1499 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1500 extsize = ntohl(extsize);
1501 if (read_index_extension(istate,
1502 (const char *) mmap + src_offset,
1503 (char *) mmap + src_offset + 8,
1504 extsize) < 0)
1505 goto unmap;
1506 src_offset += 8;
1507 src_offset += extsize;
1509 munmap(mmap, mmap_size);
1510 return istate->cache_nr;
1512 unmap:
1513 munmap(mmap, mmap_size);
1514 die("index file corrupt");
1517 int is_index_unborn(struct index_state *istate)
1519 return (!istate->cache_nr && !istate->timestamp.sec);
1522 int discard_index(struct index_state *istate)
1524 int i;
1526 for (i = 0; i < istate->cache_nr; i++)
1527 free(istate->cache[i]);
1528 resolve_undo_clear_index(istate);
1529 istate->cache_nr = 0;
1530 istate->cache_changed = 0;
1531 istate->timestamp.sec = 0;
1532 istate->timestamp.nsec = 0;
1533 free_name_hash(istate);
1534 cache_tree_free(&(istate->cache_tree));
1535 istate->initialized = 0;
1536 free(istate->cache);
1537 istate->cache = NULL;
1538 istate->cache_alloc = 0;
1539 return 0;
1542 int unmerged_index(const struct index_state *istate)
1544 int i;
1545 for (i = 0; i < istate->cache_nr; i++) {
1546 if (ce_stage(istate->cache[i]))
1547 return 1;
1549 return 0;
1552 #define WRITE_BUFFER_SIZE 8192
1553 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1554 static unsigned long write_buffer_len;
1556 static int ce_write_flush(git_SHA_CTX *context, int fd)
1558 unsigned int buffered = write_buffer_len;
1559 if (buffered) {
1560 git_SHA1_Update(context, write_buffer, buffered);
1561 if (write_in_full(fd, write_buffer, buffered) != buffered)
1562 return -1;
1563 write_buffer_len = 0;
1565 return 0;
1568 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1570 while (len) {
1571 unsigned int buffered = write_buffer_len;
1572 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1573 if (partial > len)
1574 partial = len;
1575 memcpy(write_buffer + buffered, data, partial);
1576 buffered += partial;
1577 if (buffered == WRITE_BUFFER_SIZE) {
1578 write_buffer_len = buffered;
1579 if (ce_write_flush(context, fd))
1580 return -1;
1581 buffered = 0;
1583 write_buffer_len = buffered;
1584 len -= partial;
1585 data = (char *) data + partial;
1587 return 0;
1590 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1591 unsigned int ext, unsigned int sz)
1593 ext = htonl(ext);
1594 sz = htonl(sz);
1595 return ((ce_write(context, fd, &ext, 4) < 0) ||
1596 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1599 static int ce_flush(git_SHA_CTX *context, int fd)
1601 unsigned int left = write_buffer_len;
1603 if (left) {
1604 write_buffer_len = 0;
1605 git_SHA1_Update(context, write_buffer, left);
1608 /* Flush first if not enough space for SHA1 signature */
1609 if (left + 20 > WRITE_BUFFER_SIZE) {
1610 if (write_in_full(fd, write_buffer, left) != left)
1611 return -1;
1612 left = 0;
1615 /* Append the SHA1 signature at the end */
1616 git_SHA1_Final(write_buffer + left, context);
1617 left += 20;
1618 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1621 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1624 * The only thing we care about in this function is to smudge the
1625 * falsely clean entry due to touch-update-touch race, so we leave
1626 * everything else as they are. We are called for entries whose
1627 * ce_stat_data.sd_mtime match the index file mtime.
1629 * Note that this actually does not do much for gitlinks, for
1630 * which ce_match_stat_basic() always goes to the actual
1631 * contents. The caller checks with is_racy_timestamp() which
1632 * always says "no" for gitlinks, so we are not called for them ;-)
1634 struct stat st;
1636 if (lstat(ce->name, &st) < 0)
1637 return;
1638 if (ce_match_stat_basic(ce, &st))
1639 return;
1640 if (ce_modified_check_fs(ce, &st)) {
1641 /* This is "racily clean"; smudge it. Note that this
1642 * is a tricky code. At first glance, it may appear
1643 * that it can break with this sequence:
1645 * $ echo xyzzy >frotz
1646 * $ git-update-index --add frotz
1647 * $ : >frotz
1648 * $ sleep 3
1649 * $ echo filfre >nitfol
1650 * $ git-update-index --add nitfol
1652 * but it does not. When the second update-index runs,
1653 * it notices that the entry "frotz" has the same timestamp
1654 * as index, and if we were to smudge it by resetting its
1655 * size to zero here, then the object name recorded
1656 * in index is the 6-byte file but the cached stat information
1657 * becomes zero --- which would then match what we would
1658 * obtain from the filesystem next time we stat("frotz").
1660 * However, the second update-index, before calling
1661 * this function, notices that the cached size is 6
1662 * bytes and what is on the filesystem is an empty
1663 * file, and never calls us, so the cached size information
1664 * for "frotz" stays 6 which does not match the filesystem.
1666 ce->ce_stat_data.sd_size = 0;
1670 /* Copy miscellaneous fields but not the name */
1671 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1672 struct cache_entry *ce)
1674 short flags;
1676 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1677 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1678 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1679 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1680 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1681 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1682 ondisk->mode = htonl(ce->ce_mode);
1683 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1684 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1685 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1686 hashcpy(ondisk->sha1, ce->sha1);
1688 flags = ce->ce_flags;
1689 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1690 ondisk->flags = htons(flags);
1691 if (ce->ce_flags & CE_EXTENDED) {
1692 struct ondisk_cache_entry_extended *ondisk2;
1693 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1694 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1695 return ondisk2->name;
1697 else {
1698 return ondisk->name;
1702 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1703 struct strbuf *previous_name)
1705 int size;
1706 struct ondisk_cache_entry *ondisk;
1707 char *name;
1708 int result;
1710 if (!previous_name) {
1711 size = ondisk_ce_size(ce);
1712 ondisk = xcalloc(1, size);
1713 name = copy_cache_entry_to_ondisk(ondisk, ce);
1714 memcpy(name, ce->name, ce_namelen(ce));
1715 } else {
1716 int common, to_remove, prefix_size;
1717 unsigned char to_remove_vi[16];
1718 for (common = 0;
1719 (ce->name[common] &&
1720 common < previous_name->len &&
1721 ce->name[common] == previous_name->buf[common]);
1722 common++)
1723 ; /* still matching */
1724 to_remove = previous_name->len - common;
1725 prefix_size = encode_varint(to_remove, to_remove_vi);
1727 if (ce->ce_flags & CE_EXTENDED)
1728 size = offsetof(struct ondisk_cache_entry_extended, name);
1729 else
1730 size = offsetof(struct ondisk_cache_entry, name);
1731 size += prefix_size + (ce_namelen(ce) - common + 1);
1733 ondisk = xcalloc(1, size);
1734 name = copy_cache_entry_to_ondisk(ondisk, ce);
1735 memcpy(name, to_remove_vi, prefix_size);
1736 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1738 strbuf_splice(previous_name, common, to_remove,
1739 ce->name + common, ce_namelen(ce) - common);
1742 result = ce_write(c, fd, ondisk, size);
1743 free(ondisk);
1744 return result;
1747 static int has_racy_timestamp(struct index_state *istate)
1749 int entries = istate->cache_nr;
1750 int i;
1752 for (i = 0; i < entries; i++) {
1753 struct cache_entry *ce = istate->cache[i];
1754 if (is_racy_timestamp(istate, ce))
1755 return 1;
1757 return 0;
1761 * Opportunistically update the index but do not complain if we can't
1763 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1765 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1766 !write_index(istate, lockfile->fd))
1767 commit_locked_index(lockfile);
1768 else
1769 rollback_lock_file(lockfile);
1772 int write_index(struct index_state *istate, int newfd)
1774 git_SHA_CTX c;
1775 struct cache_header hdr;
1776 int i, err, removed, extended, hdr_version;
1777 struct cache_entry **cache = istate->cache;
1778 int entries = istate->cache_nr;
1779 struct stat st;
1780 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1782 for (i = removed = extended = 0; i < entries; i++) {
1783 if (cache[i]->ce_flags & CE_REMOVE)
1784 removed++;
1786 /* reduce extended entries if possible */
1787 cache[i]->ce_flags &= ~CE_EXTENDED;
1788 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1789 extended++;
1790 cache[i]->ce_flags |= CE_EXTENDED;
1794 if (!istate->version)
1795 istate->version = INDEX_FORMAT_DEFAULT;
1797 /* demote version 3 to version 2 when the latter suffices */
1798 if (istate->version == 3 || istate->version == 2)
1799 istate->version = extended ? 3 : 2;
1801 hdr_version = istate->version;
1803 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1804 hdr.hdr_version = htonl(hdr_version);
1805 hdr.hdr_entries = htonl(entries - removed);
1807 git_SHA1_Init(&c);
1808 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1809 return -1;
1811 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1812 for (i = 0; i < entries; i++) {
1813 struct cache_entry *ce = cache[i];
1814 if (ce->ce_flags & CE_REMOVE)
1815 continue;
1816 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1817 ce_smudge_racily_clean_entry(ce);
1818 if (is_null_sha1(ce->sha1)) {
1819 static const char msg[] = "cache entry has null sha1: %s";
1820 static int allow = -1;
1822 if (allow < 0)
1823 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1824 if (allow)
1825 warning(msg, ce->name);
1826 else
1827 return error(msg, ce->name);
1829 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1830 return -1;
1832 strbuf_release(&previous_name_buf);
1834 /* Write extension data here */
1835 if (istate->cache_tree) {
1836 struct strbuf sb = STRBUF_INIT;
1838 cache_tree_write(&sb, istate->cache_tree);
1839 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1840 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1841 strbuf_release(&sb);
1842 if (err)
1843 return -1;
1845 if (istate->resolve_undo) {
1846 struct strbuf sb = STRBUF_INIT;
1848 resolve_undo_write(&sb, istate->resolve_undo);
1849 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1850 sb.len) < 0
1851 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1852 strbuf_release(&sb);
1853 if (err)
1854 return -1;
1857 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1858 return -1;
1859 istate->timestamp.sec = (unsigned int)st.st_mtime;
1860 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1861 return 0;
1865 * Read the index file that is potentially unmerged into given
1866 * index_state, dropping any unmerged entries. Returns true if
1867 * the index is unmerged. Callers who want to refuse to work
1868 * from an unmerged state can call this and check its return value,
1869 * instead of calling read_cache().
1871 int read_index_unmerged(struct index_state *istate)
1873 int i;
1874 int unmerged = 0;
1876 read_index(istate);
1877 for (i = 0; i < istate->cache_nr; i++) {
1878 struct cache_entry *ce = istate->cache[i];
1879 struct cache_entry *new_ce;
1880 int size, len;
1882 if (!ce_stage(ce))
1883 continue;
1884 unmerged = 1;
1885 len = ce_namelen(ce);
1886 size = cache_entry_size(len);
1887 new_ce = xcalloc(1, size);
1888 memcpy(new_ce->name, ce->name, len);
1889 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1890 new_ce->ce_namelen = len;
1891 new_ce->ce_mode = ce->ce_mode;
1892 if (add_index_entry(istate, new_ce, 0))
1893 return error("%s: cannot drop to stage #0",
1894 ce->name);
1895 i = index_name_pos(istate, new_ce->name, len);
1897 return unmerged;
1901 * Returns 1 if the path is an "other" path with respect to
1902 * the index; that is, the path is not mentioned in the index at all,
1903 * either as a file, a directory with some files in the index,
1904 * or as an unmerged entry.
1906 * We helpfully remove a trailing "/" from directories so that
1907 * the output of read_directory can be used as-is.
1909 int index_name_is_other(const struct index_state *istate, const char *name,
1910 int namelen)
1912 int pos;
1913 if (namelen && name[namelen - 1] == '/')
1914 namelen--;
1915 pos = index_name_pos(istate, name, namelen);
1916 if (0 <= pos)
1917 return 0; /* exact match */
1918 pos = -pos - 1;
1919 if (pos < istate->cache_nr) {
1920 struct cache_entry *ce = istate->cache[pos];
1921 if (ce_namelen(ce) == namelen &&
1922 !memcmp(ce->name, name, namelen))
1923 return 0; /* Yup, this one exists unmerged */
1925 return 1;
1928 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
1930 int pos, len;
1931 unsigned long sz;
1932 enum object_type type;
1933 void *data;
1935 len = strlen(path);
1936 pos = index_name_pos(istate, path, len);
1937 if (pos < 0) {
1939 * We might be in the middle of a merge, in which
1940 * case we would read stage #2 (ours).
1942 int i;
1943 for (i = -pos - 1;
1944 (pos < 0 && i < istate->cache_nr &&
1945 !strcmp(istate->cache[i]->name, path));
1946 i++)
1947 if (ce_stage(istate->cache[i]) == 2)
1948 pos = i;
1950 if (pos < 0)
1951 return NULL;
1952 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
1953 if (!data || type != OBJ_BLOB) {
1954 free(data);
1955 return NULL;
1957 if (size)
1958 *size = sz;
1959 return data;
1962 void stat_validity_clear(struct stat_validity *sv)
1964 free(sv->sd);
1965 sv->sd = NULL;
1968 int stat_validity_check(struct stat_validity *sv, const char *path)
1970 struct stat st;
1972 if (stat(path, &st) < 0)
1973 return sv->sd == NULL;
1974 if (!sv->sd)
1975 return 0;
1976 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
1979 void stat_validity_update(struct stat_validity *sv, int fd)
1981 struct stat st;
1983 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
1984 stat_validity_clear(sv);
1985 else {
1986 if (!sv->sd)
1987 sv->sd = xcalloc(1, sizeof(struct stat_data));
1988 fill_stat_data(sv->sd, &st);