![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
1 /*
2 * GIT - The information manager from hell
3 *
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 "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16 #include "blob.h"
18 /* Index extensions.
19 *
20 * The first letter should be 'A'..'Z' for extensions that are not
21 * necessary for a correct operation (i.e. optimization data).
22 * When new extensions are added that _needs_ to be understood in
23 * order to correctly interpret the index file, pick character that
24 * is outside the range, to cause the reader to abort.
25 */
27 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
28 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
30 struct index_state the_index;
32 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
33 {
34 istate->cache[nr] = ce;
35 add_name_hash(istate, ce);
36 }
38 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
39 {
40 struct cache_entry *old = istate->cache[nr];
42 remove_name_hash(old);
43 set_index_entry(istate, nr, ce);
44 istate->cache_changed = 1;
45 }
47 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
48 {
49 struct cache_entry *old = istate->cache[nr], *new;
50 int namelen = strlen(new_name);
52 new = xmalloc(cache_entry_size(namelen));
53 copy_cache_entry(new, old);
54 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
55 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
56 memcpy(new->name, new_name, namelen + 1);
58 cache_tree_invalidate_path(istate->cache_tree, old->name);
59 remove_index_entry_at(istate, nr);
60 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
61 }
63 /*
64 * This only updates the "non-critical" parts of the directory
65 * cache, ie the parts that aren't tracked by GIT, and only used
66 * to validate the cache.
67 */
68 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
69 {
70 ce->ce_ctime.sec = (unsigned int)st->st_ctime;
71 ce->ce_mtime.sec = (unsigned int)st->st_mtime;
72 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
73 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
74 ce->ce_dev = st->st_dev;
75 ce->ce_ino = st->st_ino;
76 ce->ce_uid = st->st_uid;
77 ce->ce_gid = st->st_gid;
78 ce->ce_size = st->st_size;
80 if (assume_unchanged)
81 ce->ce_flags |= CE_VALID;
83 if (S_ISREG(st->st_mode))
84 ce_mark_uptodate(ce);
85 }
87 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
88 {
89 int match = -1;
90 int fd = open(ce->name, O_RDONLY);
92 if (fd >= 0) {
93 unsigned char sha1[20];
94 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
95 match = hashcmp(sha1, ce->sha1);
96 /* index_fd() closed the file descriptor already */
97 }
98 return match;
99 }
101 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
102 {
103 int match = -1;
104 void *buffer;
105 unsigned long size;
106 enum object_type type;
107 struct strbuf sb = STRBUF_INIT;
109 if (strbuf_readlink(&sb, ce->name, expected_size))
110 return -1;
112 buffer = read_sha1_file(ce->sha1, &type, &size);
113 if (buffer) {
114 if (size == sb.len)
115 match = memcmp(buffer, sb.buf, size);
116 free(buffer);
117 }
118 strbuf_release(&sb);
119 return match;
120 }
122 static int ce_compare_gitlink(struct cache_entry *ce)
123 {
124 unsigned char sha1[20];
126 /*
127 * We don't actually require that the .git directory
128 * under GITLINK directory be a valid git directory. It
129 * might even be missing (in case nobody populated that
130 * sub-project).
131 *
132 * If so, we consider it always to match.
133 */
134 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
135 return 0;
136 return hashcmp(sha1, ce->sha1);
137 }
139 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
140 {
141 switch (st->st_mode & S_IFMT) {
142 case S_IFREG:
143 if (ce_compare_data(ce, st))
144 return DATA_CHANGED;
145 break;
146 case S_IFLNK:
147 if (ce_compare_link(ce, xsize_t(st->st_size)))
148 return DATA_CHANGED;
149 break;
150 case S_IFDIR:
151 if (S_ISGITLINK(ce->ce_mode))
152 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
153 default:
154 return TYPE_CHANGED;
155 }
156 return 0;
157 }
159 int is_empty_blob_sha1(const unsigned char *sha1)
160 {
161 static const unsigned char empty_blob_sha1[20] = {
162 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
163 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
164 };
166 return !hashcmp(sha1, empty_blob_sha1);
167 }
169 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
170 {
171 unsigned int changed = 0;
173 if (ce->ce_flags & CE_REMOVE)
174 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
176 switch (ce->ce_mode & S_IFMT) {
177 case S_IFREG:
178 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
179 /* We consider only the owner x bit to be relevant for
180 * "mode changes"
181 */
182 if (trust_executable_bit &&
183 (0100 & (ce->ce_mode ^ st->st_mode)))
184 changed |= MODE_CHANGED;
185 break;
186 case S_IFLNK:
187 if (!S_ISLNK(st->st_mode) &&
188 (has_symlinks || !S_ISREG(st->st_mode)))
189 changed |= TYPE_CHANGED;
190 break;
191 case S_IFGITLINK:
192 /* We ignore most of the st_xxx fields for gitlinks */
193 if (!S_ISDIR(st->st_mode))
194 changed |= TYPE_CHANGED;
195 else if (ce_compare_gitlink(ce))
196 changed |= DATA_CHANGED;
197 return changed;
198 default:
199 die("internal error: ce_mode is %o", ce->ce_mode);
200 }
201 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
202 changed |= MTIME_CHANGED;
203 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
204 changed |= CTIME_CHANGED;
206 #ifdef USE_NSEC
207 if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
208 changed |= MTIME_CHANGED;
209 if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
210 changed |= CTIME_CHANGED;
211 #endif
213 if (ce->ce_uid != (unsigned int) st->st_uid ||
214 ce->ce_gid != (unsigned int) st->st_gid)
215 changed |= OWNER_CHANGED;
216 if (ce->ce_ino != (unsigned int) st->st_ino)
217 changed |= INODE_CHANGED;
219 #ifdef USE_STDEV
220 /*
221 * st_dev breaks on network filesystems where different
222 * clients will have different views of what "device"
223 * the filesystem is on
224 */
225 if (ce->ce_dev != (unsigned int) st->st_dev)
226 changed |= INODE_CHANGED;
227 #endif
229 if (ce->ce_size != (unsigned int) st->st_size)
230 changed |= DATA_CHANGED;
232 /* Racily smudged entry? */
233 if (!ce->ce_size) {
234 if (!is_empty_blob_sha1(ce->sha1))
235 changed |= DATA_CHANGED;
236 }
238 return changed;
239 }
241 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
242 {
243 return (!S_ISGITLINK(ce->ce_mode) &&
244 istate->timestamp.sec &&
245 #ifdef USE_NSEC
246 /* nanosecond timestamped files can also be racy! */
247 (istate->timestamp.sec < ce->ce_mtime.sec ||
248 (istate->timestamp.sec == ce->ce_mtime.sec &&
249 istate->timestamp.nsec <= ce->ce_mtime.nsec))
250 #else
251 istate->timestamp.sec <= ce->ce_mtime.sec
252 #endif
253 );
254 }
256 int ie_match_stat(const struct index_state *istate,
257 struct cache_entry *ce, struct stat *st,
258 unsigned int options)
259 {
260 unsigned int changed;
261 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
262 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
264 /*
265 * If it's marked as always valid in the index, it's
266 * valid whatever the checked-out copy says.
267 */
268 if (!ignore_valid && (ce->ce_flags & CE_VALID))
269 return 0;
271 /*
272 * Intent-to-add entries have not been added, so the index entry
273 * by definition never matches what is in the work tree until it
274 * actually gets added.
275 */
276 if (ce->ce_flags & CE_INTENT_TO_ADD)
277 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
279 changed = ce_match_stat_basic(ce, st);
281 /*
282 * Within 1 second of this sequence:
283 * echo xyzzy >file && git-update-index --add file
284 * running this command:
285 * echo frotz >file
286 * would give a falsely clean cache entry. The mtime and
287 * length match the cache, and other stat fields do not change.
288 *
289 * We could detect this at update-index time (the cache entry
290 * being registered/updated records the same time as "now")
291 * and delay the return from git-update-index, but that would
292 * effectively mean we can make at most one commit per second,
293 * which is not acceptable. Instead, we check cache entries
294 * whose mtime are the same as the index file timestamp more
295 * carefully than others.
296 */
297 if (!changed && is_racy_timestamp(istate, ce)) {
298 if (assume_racy_is_modified)
299 changed |= DATA_CHANGED;
300 else
301 changed |= ce_modified_check_fs(ce, st);
302 }
304 return changed;
305 }
307 int ie_modified(const struct index_state *istate,
308 struct cache_entry *ce, struct stat *st, unsigned int options)
309 {
310 int changed, changed_fs;
312 changed = ie_match_stat(istate, ce, st, options);
313 if (!changed)
314 return 0;
315 /*
316 * If the mode or type has changed, there's no point in trying
317 * to refresh the entry - it's not going to match
318 */
319 if (changed & (MODE_CHANGED | TYPE_CHANGED))
320 return changed;
322 /*
323 * Immediately after read-tree or update-index --cacheinfo,
324 * the length field is zero, as we have never even read the
325 * lstat(2) information once, and we cannot trust DATA_CHANGED
326 * returned by ie_match_stat() which in turn was returned by
327 * ce_match_stat_basic() to signal that the filesize of the
328 * blob changed. We have to actually go to the filesystem to
329 * see if the contents match, and if so, should answer "unchanged".
330 *
331 * The logic does not apply to gitlinks, as ce_match_stat_basic()
332 * already has checked the actual HEAD from the filesystem in the
333 * subproject. If ie_match_stat() already said it is different,
334 * then we know it is.
335 */
336 if ((changed & DATA_CHANGED) &&
337 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
338 return changed;
340 changed_fs = ce_modified_check_fs(ce, st);
341 if (changed_fs)
342 return changed | changed_fs;
343 return 0;
344 }
346 int base_name_compare(const char *name1, int len1, int mode1,
347 const char *name2, int len2, int mode2)
348 {
349 unsigned char c1, c2;
350 int len = len1 < len2 ? len1 : len2;
351 int cmp;
353 cmp = memcmp(name1, name2, len);
354 if (cmp)
355 return cmp;
356 c1 = name1[len];
357 c2 = name2[len];
358 if (!c1 && S_ISDIR(mode1))
359 c1 = '/';
360 if (!c2 && S_ISDIR(mode2))
361 c2 = '/';
362 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
363 }
365 /*
366 * df_name_compare() is identical to base_name_compare(), except it
367 * compares conflicting directory/file entries as equal. Note that
368 * while a directory name compares as equal to a regular file, they
369 * then individually compare _differently_ to a filename that has
370 * a dot after the basename (because '\0' < '.' < '/').
371 *
372 * This is used by routines that want to traverse the git namespace
373 * but then handle conflicting entries together when possible.
374 */
375 int df_name_compare(const char *name1, int len1, int mode1,
376 const char *name2, int len2, int mode2)
377 {
378 int len = len1 < len2 ? len1 : len2, cmp;
379 unsigned char c1, c2;
381 cmp = memcmp(name1, name2, len);
382 if (cmp)
383 return cmp;
384 /* Directories and files compare equal (same length, same name) */
385 if (len1 == len2)
386 return 0;
387 c1 = name1[len];
388 if (!c1 && S_ISDIR(mode1))
389 c1 = '/';
390 c2 = name2[len];
391 if (!c2 && S_ISDIR(mode2))
392 c2 = '/';
393 if (c1 == '/' && !c2)
394 return 0;
395 if (c2 == '/' && !c1)
396 return 0;
397 return c1 - c2;
398 }
400 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
401 {
402 int len1 = flags1 & CE_NAMEMASK;
403 int len2 = flags2 & CE_NAMEMASK;
404 int len = len1 < len2 ? len1 : len2;
405 int cmp;
407 cmp = memcmp(name1, name2, len);
408 if (cmp)
409 return cmp;
410 if (len1 < len2)
411 return -1;
412 if (len1 > len2)
413 return 1;
415 /* Compare stages */
416 flags1 &= CE_STAGEMASK;
417 flags2 &= CE_STAGEMASK;
419 if (flags1 < flags2)
420 return -1;
421 if (flags1 > flags2)
422 return 1;
423 return 0;
424 }
426 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
427 {
428 int first, last;
430 first = 0;
431 last = istate->cache_nr;
432 while (last > first) {
433 int next = (last + first) >> 1;
434 struct cache_entry *ce = istate->cache[next];
435 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
436 if (!cmp)
437 return next;
438 if (cmp < 0) {
439 last = next;
440 continue;
441 }
442 first = next+1;
443 }
444 return -first-1;
445 }
447 /* Remove entry, return true if there are more entries to go.. */
448 int remove_index_entry_at(struct index_state *istate, int pos)
449 {
450 struct cache_entry *ce = istate->cache[pos];
452 remove_name_hash(ce);
453 istate->cache_changed = 1;
454 istate->cache_nr--;
455 if (pos >= istate->cache_nr)
456 return 0;
457 memmove(istate->cache + pos,
458 istate->cache + pos + 1,
459 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
460 return 1;
461 }
463 /*
464 * Remove all cache ententries marked for removal, that is where
465 * CE_REMOVE is set in ce_flags. This is much more effective than
466 * calling remove_index_entry_at() for each entry to be removed.
467 */
468 void remove_marked_cache_entries(struct index_state *istate)
469 {
470 struct cache_entry **ce_array = istate->cache;
471 unsigned int i, j;
473 for (i = j = 0; i < istate->cache_nr; i++) {
474 if (ce_array[i]->ce_flags & CE_REMOVE)
475 remove_name_hash(ce_array[i]);
476 else
477 ce_array[j++] = ce_array[i];
478 }
479 istate->cache_changed = 1;
480 istate->cache_nr = j;
481 }
483 int remove_file_from_index(struct index_state *istate, const char *path)
484 {
485 int pos = index_name_pos(istate, path, strlen(path));
486 if (pos < 0)
487 pos = -pos-1;
488 cache_tree_invalidate_path(istate->cache_tree, path);
489 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
490 remove_index_entry_at(istate, pos);
491 return 0;
492 }
494 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
495 {
496 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
497 }
499 static int index_name_pos_also_unmerged(struct index_state *istate,
500 const char *path, int namelen)
501 {
502 int pos = index_name_pos(istate, path, namelen);
503 struct cache_entry *ce;
505 if (pos >= 0)
506 return pos;
508 /* maybe unmerged? */
509 pos = -1 - pos;
510 if (pos >= istate->cache_nr ||
511 compare_name((ce = istate->cache[pos]), path, namelen))
512 return -1;
514 /* order of preference: stage 2, 1, 3 */
515 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
516 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
517 !compare_name(ce, path, namelen))
518 pos++;
519 return pos;
520 }
522 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
523 {
524 int len = ce_namelen(ce);
525 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
526 }
528 /*
529 * If we add a filename that aliases in the cache, we will use the
530 * name that we already have - but we don't want to update the same
531 * alias twice, because that implies that there were actually two
532 * different files with aliasing names!
533 *
534 * So we use the CE_ADDED flag to verify that the alias was an old
535 * one before we accept it as
536 */
537 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
538 {
539 int len;
540 struct cache_entry *new;
542 if (alias->ce_flags & CE_ADDED)
543 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
545 /* Ok, create the new entry using the name of the existing alias */
546 len = ce_namelen(alias);
547 new = xcalloc(1, cache_entry_size(len));
548 memcpy(new->name, alias->name, len);
549 copy_cache_entry(new, ce);
550 free(ce);
551 return new;
552 }
554 static void record_intent_to_add(struct cache_entry *ce)
555 {
556 unsigned char sha1[20];
557 if (write_sha1_file("", 0, blob_type, sha1))
558 die("cannot create an empty blob in the object database");
559 hashcpy(ce->sha1, sha1);
560 }
562 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
563 {
564 int size, namelen, was_same;
565 mode_t st_mode = st->st_mode;
566 struct cache_entry *ce, *alias;
567 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
568 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
569 int pretend = flags & ADD_CACHE_PRETEND;
570 int intent_only = flags & ADD_CACHE_INTENT;
571 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
572 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
574 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
575 return error("%s: can only add regular files, symbolic links or git-directories", path);
577 namelen = strlen(path);
578 if (S_ISDIR(st_mode)) {
579 while (namelen && path[namelen-1] == '/')
580 namelen--;
581 }
582 size = cache_entry_size(namelen);
583 ce = xcalloc(1, size);
584 memcpy(ce->name, path, namelen);
585 ce->ce_flags = namelen;
586 if (!intent_only)
587 fill_stat_cache_info(ce, st);
588 else
589 ce->ce_flags |= CE_INTENT_TO_ADD;
591 if (trust_executable_bit && has_symlinks)
592 ce->ce_mode = create_ce_mode(st_mode);
593 else {
594 /* If there is an existing entry, pick the mode bits and type
595 * from it, otherwise assume unexecutable regular file.
596 */
597 struct cache_entry *ent;
598 int pos = index_name_pos_also_unmerged(istate, path, namelen);
600 ent = (0 <= pos) ? istate->cache[pos] : NULL;
601 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
602 }
604 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
605 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
606 /* Nothing changed, really */
607 free(ce);
608 ce_mark_uptodate(alias);
609 alias->ce_flags |= CE_ADDED;
610 return 0;
611 }
612 if (!intent_only) {
613 if (index_path(ce->sha1, path, st, 1))
614 return error("unable to index file %s", path);
615 } else
616 record_intent_to_add(ce);
618 if (ignore_case && alias && different_name(ce, alias))
619 ce = create_alias_ce(ce, alias);
620 ce->ce_flags |= CE_ADDED;
622 /* It was suspected to be racily clean, but it turns out to be Ok */
623 was_same = (alias &&
624 !ce_stage(alias) &&
625 !hashcmp(alias->sha1, ce->sha1) &&
626 ce->ce_mode == alias->ce_mode);
628 if (pretend)
629 ;
630 else if (add_index_entry(istate, ce, add_option))
631 return error("unable to add %s to index",path);
632 if (verbose && !was_same)
633 printf("add '%s'\n", path);
634 return 0;
635 }
637 int add_file_to_index(struct index_state *istate, const char *path, int flags)
638 {
639 struct stat st;
640 if (lstat(path, &st))
641 die("%s: unable to stat (%s)", path, strerror(errno));
642 return add_to_index(istate, path, &st, flags);
643 }
645 struct cache_entry *make_cache_entry(unsigned int mode,
646 const unsigned char *sha1, const char *path, int stage,
647 int refresh)
648 {
649 int size, len;
650 struct cache_entry *ce;
652 if (!verify_path(path)) {
653 error("Invalid path '%s'", path);
654 return NULL;
655 }
657 len = strlen(path);
658 size = cache_entry_size(len);
659 ce = xcalloc(1, size);
661 hashcpy(ce->sha1, sha1);
662 memcpy(ce->name, path, len);
663 ce->ce_flags = create_ce_flags(len, stage);
664 ce->ce_mode = create_ce_mode(mode);
666 if (refresh)
667 return refresh_cache_entry(ce, 0);
669 return ce;
670 }
672 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
673 {
674 int len = ce_namelen(a);
675 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
676 }
678 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
679 {
680 const char *match, *name;
681 int len;
683 if (!pathspec)
684 return 1;
686 len = ce_namelen(ce);
687 name = ce->name;
688 while ((match = *pathspec++) != NULL) {
689 int matchlen = strlen(match);
690 if (matchlen > len)
691 continue;
692 if (memcmp(name, match, matchlen))
693 continue;
694 if (matchlen && name[matchlen-1] == '/')
695 return 1;
696 if (name[matchlen] == '/' || !name[matchlen])
697 return 1;
698 if (!matchlen)
699 return 1;
700 }
701 return 0;
702 }
704 /*
705 * We fundamentally don't like some paths: we don't want
706 * dot or dot-dot anywhere, and for obvious reasons don't
707 * want to recurse into ".git" either.
708 *
709 * Also, we don't want double slashes or slashes at the
710 * end that can make pathnames ambiguous.
711 */
712 static int verify_dotfile(const char *rest)
713 {
714 /*
715 * The first character was '.', but that
716 * has already been discarded, we now test
717 * the rest.
718 */
719 switch (*rest) {
720 /* "." is not allowed */
721 case '\0': case '/':
722 return 0;
724 /*
725 * ".git" followed by NUL or slash is bad. This
726 * shares the path end test with the ".." case.
727 */
728 case 'g':
729 if (rest[1] != 'i')
730 break;
731 if (rest[2] != 't')
732 break;
733 rest += 2;
734 /* fallthrough */
735 case '.':
736 if (rest[1] == '\0' || rest[1] == '/')
737 return 0;
738 }
739 return 1;
740 }
742 int verify_path(const char *path)
743 {
744 char c;
746 goto inside;
747 for (;;) {
748 if (!c)
749 return 1;
750 if (c == '/') {
751 inside:
752 c = *path++;
753 switch (c) {
754 default:
755 continue;
756 case '/': case '\0':
757 break;
758 case '.':
759 if (verify_dotfile(path))
760 continue;
761 }
762 return 0;
763 }
764 c = *path++;
765 }
766 }
768 /*
769 * Do we have another file that has the beginning components being a
770 * proper superset of the name we're trying to add?
771 */
772 static int has_file_name(struct index_state *istate,
773 const struct cache_entry *ce, int pos, int ok_to_replace)
774 {
775 int retval = 0;
776 int len = ce_namelen(ce);
777 int stage = ce_stage(ce);
778 const char *name = ce->name;
780 while (pos < istate->cache_nr) {
781 struct cache_entry *p = istate->cache[pos++];
783 if (len >= ce_namelen(p))
784 break;
785 if (memcmp(name, p->name, len))
786 break;
787 if (ce_stage(p) != stage)
788 continue;
789 if (p->name[len] != '/')
790 continue;
791 if (p->ce_flags & CE_REMOVE)
792 continue;
793 retval = -1;
794 if (!ok_to_replace)
795 break;
796 remove_index_entry_at(istate, --pos);
797 }
798 return retval;
799 }
801 /*
802 * Do we have another file with a pathname that is a proper
803 * subset of the name we're trying to add?
804 */
805 static int has_dir_name(struct index_state *istate,
806 const struct cache_entry *ce, int pos, int ok_to_replace)
807 {
808 int retval = 0;
809 int stage = ce_stage(ce);
810 const char *name = ce->name;
811 const char *slash = name + ce_namelen(ce);
813 for (;;) {
814 int len;
816 for (;;) {
817 if (*--slash == '/')
818 break;
819 if (slash <= ce->name)
820 return retval;
821 }
822 len = slash - name;
824 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
825 if (pos >= 0) {
826 /*
827 * Found one, but not so fast. This could
828 * be a marker that says "I was here, but
829 * I am being removed". Such an entry is
830 * not a part of the resulting tree, and
831 * it is Ok to have a directory at the same
832 * path.
833 */
834 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
835 retval = -1;
836 if (!ok_to_replace)
837 break;
838 remove_index_entry_at(istate, pos);
839 continue;
840 }
841 }
842 else
843 pos = -pos-1;
845 /*
846 * Trivial optimization: if we find an entry that
847 * already matches the sub-directory, then we know
848 * we're ok, and we can exit.
849 */
850 while (pos < istate->cache_nr) {
851 struct cache_entry *p = istate->cache[pos];
852 if ((ce_namelen(p) <= len) ||
853 (p->name[len] != '/') ||
854 memcmp(p->name, name, len))
855 break; /* not our subdirectory */
856 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
857 /*
858 * p is at the same stage as our entry, and
859 * is a subdirectory of what we are looking
860 * at, so we cannot have conflicts at our
861 * level or anything shorter.
862 */
863 return retval;
864 pos++;
865 }
866 }
867 return retval;
868 }
870 /* We may be in a situation where we already have path/file and path
871 * is being added, or we already have path and path/file is being
872 * added. Either one would result in a nonsense tree that has path
873 * twice when git-write-tree tries to write it out. Prevent it.
874 *
875 * If ok-to-replace is specified, we remove the conflicting entries
876 * from the cache so the caller should recompute the insert position.
877 * When this happens, we return non-zero.
878 */
879 static int check_file_directory_conflict(struct index_state *istate,
880 const struct cache_entry *ce,
881 int pos, int ok_to_replace)
882 {
883 int retval;
885 /*
886 * When ce is an "I am going away" entry, we allow it to be added
887 */
888 if (ce->ce_flags & CE_REMOVE)
889 return 0;
891 /*
892 * We check if the path is a sub-path of a subsequent pathname
893 * first, since removing those will not change the position
894 * in the array.
895 */
896 retval = has_file_name(istate, ce, pos, ok_to_replace);
898 /*
899 * Then check if the path might have a clashing sub-directory
900 * before it.
901 */
902 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
903 }
905 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
906 {
907 int pos;
908 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
909 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
910 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
911 int new_only = option & ADD_CACHE_NEW_ONLY;
913 cache_tree_invalidate_path(istate->cache_tree, ce->name);
914 pos = index_name_pos(istate, ce->name, ce->ce_flags);
916 /* existing match? Just replace it. */
917 if (pos >= 0) {
918 if (!new_only)
919 replace_index_entry(istate, pos, ce);
920 return 0;
921 }
922 pos = -pos-1;
924 /*
925 * Inserting a merged entry ("stage 0") into the index
926 * will always replace all non-merged entries..
927 */
928 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
929 while (ce_same_name(istate->cache[pos], ce)) {
930 ok_to_add = 1;
931 if (!remove_index_entry_at(istate, pos))
932 break;
933 }
934 }
936 if (!ok_to_add)
937 return -1;
938 if (!verify_path(ce->name))
939 return error("Invalid path '%s'", ce->name);
941 if (!skip_df_check &&
942 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
943 if (!ok_to_replace)
944 return error("'%s' appears as both a file and as a directory",
945 ce->name);
946 pos = index_name_pos(istate, ce->name, ce->ce_flags);
947 pos = -pos-1;
948 }
949 return pos + 1;
950 }
952 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
953 {
954 int pos;
956 if (option & ADD_CACHE_JUST_APPEND)
957 pos = istate->cache_nr;
958 else {
959 int ret;
960 ret = add_index_entry_with_check(istate, ce, option);
961 if (ret <= 0)
962 return ret;
963 pos = ret - 1;
964 }
966 /* Make sure the array is big enough .. */
967 if (istate->cache_nr == istate->cache_alloc) {
968 istate->cache_alloc = alloc_nr(istate->cache_alloc);
969 istate->cache = xrealloc(istate->cache,
970 istate->cache_alloc * sizeof(struct cache_entry *));
971 }
973 /* Add it in.. */
974 istate->cache_nr++;
975 if (istate->cache_nr > pos + 1)
976 memmove(istate->cache + pos + 1,
977 istate->cache + pos,
978 (istate->cache_nr - pos - 1) * sizeof(ce));
979 set_index_entry(istate, pos, ce);
980 istate->cache_changed = 1;
981 return 0;
982 }
984 /*
985 * "refresh" does not calculate a new sha1 file or bring the
986 * cache up-to-date for mode/content changes. But what it
987 * _does_ do is to "re-match" the stat information of a file
988 * with the cache, so that you can refresh the cache for a
989 * file that hasn't been changed but where the stat entry is
990 * out of date.
991 *
992 * For example, you'd want to do this after doing a "git-read-tree",
993 * to link up the stat cache details with the proper files.
994 */
995 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
996 struct cache_entry *ce,
997 unsigned int options, int *err)
998 {
999 struct stat st;
1000 struct cache_entry *updated;
1001 int changed, size;
1002 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1004 if (ce_uptodate(ce))
1005 return ce;
1007 /*
1008 * CE_VALID means the user promised us that the change to
1009 * the work tree does not matter and told us not to worry.
1010 */
1011 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1012 ce_mark_uptodate(ce);
1013 return ce;
1014 }
1016 if (lstat(ce->name, &st) < 0) {
1017 if (err)
1018 *err = errno;
1019 return NULL;
1020 }
1022 changed = ie_match_stat(istate, ce, &st, options);
1023 if (!changed) {
1024 /*
1025 * The path is unchanged. If we were told to ignore
1026 * valid bit, then we did the actual stat check and
1027 * found that the entry is unmodified. If the entry
1028 * is not marked VALID, this is the place to mark it
1029 * valid again, under "assume unchanged" mode.
1030 */
1031 if (ignore_valid && assume_unchanged &&
1032 !(ce->ce_flags & CE_VALID))
1033 ; /* mark this one VALID again */
1034 else {
1035 /*
1036 * We do not mark the index itself "modified"
1037 * because CE_UPTODATE flag is in-core only;
1038 * we are not going to write this change out.
1039 */
1040 ce_mark_uptodate(ce);
1041 return ce;
1042 }
1043 }
1045 if (ie_modified(istate, ce, &st, options)) {
1046 if (err)
1047 *err = EINVAL;
1048 return NULL;
1049 }
1051 size = ce_size(ce);
1052 updated = xmalloc(size);
1053 memcpy(updated, ce, size);
1054 fill_stat_cache_info(updated, &st);
1055 /*
1056 * If ignore_valid is not set, we should leave CE_VALID bit
1057 * alone. Otherwise, paths marked with --no-assume-unchanged
1058 * (i.e. things to be edited) will reacquire CE_VALID bit
1059 * automatically, which is not really what we want.
1060 */
1061 if (!ignore_valid && assume_unchanged &&
1062 !(ce->ce_flags & CE_VALID))
1063 updated->ce_flags &= ~CE_VALID;
1065 return updated;
1066 }
1068 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
1069 {
1070 int i;
1071 int has_errors = 0;
1072 int really = (flags & REFRESH_REALLY) != 0;
1073 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1074 int quiet = (flags & REFRESH_QUIET) != 0;
1075 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1076 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1077 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1078 const char *needs_update_message;
1080 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
1081 ? "locally modified" : "needs update");
1082 for (i = 0; i < istate->cache_nr; i++) {
1083 struct cache_entry *ce, *new;
1084 int cache_errno = 0;
1086 ce = istate->cache[i];
1087 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1088 continue;
1090 if (ce_stage(ce)) {
1091 while ((i < istate->cache_nr) &&
1092 ! strcmp(istate->cache[i]->name, ce->name))
1093 i++;
1094 i--;
1095 if (allow_unmerged)
1096 continue;
1097 printf("%s: needs merge\n", ce->name);
1098 has_errors = 1;
1099 continue;
1100 }
1102 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1103 continue;
1105 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1106 if (new == ce)
1107 continue;
1108 if (!new) {
1109 if (not_new && cache_errno == ENOENT)
1110 continue;
1111 if (really && cache_errno == EINVAL) {
1112 /* If we are doing --really-refresh that
1113 * means the index is not valid anymore.
1114 */
1115 ce->ce_flags &= ~CE_VALID;
1116 istate->cache_changed = 1;
1117 }
1118 if (quiet)
1119 continue;
1120 printf("%s: %s\n", ce->name, needs_update_message);
1121 has_errors = 1;
1122 continue;
1123 }
1125 replace_index_entry(istate, i, new);
1126 }
1127 return has_errors;
1128 }
1130 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1131 {
1132 return refresh_cache_ent(&the_index, ce, really, NULL);
1133 }
1135 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1136 {
1137 git_SHA_CTX c;
1138 unsigned char sha1[20];
1140 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1141 return error("bad signature");
1142 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1143 return error("bad index version");
1144 git_SHA1_Init(&c);
1145 git_SHA1_Update(&c, hdr, size - 20);
1146 git_SHA1_Final(sha1, &c);
1147 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1148 return error("bad index file sha1 signature");
1149 return 0;
1150 }
1152 static int read_index_extension(struct index_state *istate,
1153 const char *ext, void *data, unsigned long sz)
1154 {
1155 switch (CACHE_EXT(ext)) {
1156 case CACHE_EXT_TREE:
1157 istate->cache_tree = cache_tree_read(data, sz);
1158 break;
1159 default:
1160 if (*ext < 'A' || 'Z' < *ext)
1161 return error("index uses %.4s extension, which we do not understand",
1162 ext);
1163 fprintf(stderr, "ignoring %.4s extension\n", ext);
1164 break;
1165 }
1166 return 0;
1167 }
1169 int read_index(struct index_state *istate)
1170 {
1171 return read_index_from(istate, get_index_file());
1172 }
1174 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1175 {
1176 size_t len;
1177 const char *name;
1179 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1180 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1181 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1182 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1183 ce->ce_dev = ntohl(ondisk->dev);
1184 ce->ce_ino = ntohl(ondisk->ino);
1185 ce->ce_mode = ntohl(ondisk->mode);
1186 ce->ce_uid = ntohl(ondisk->uid);
1187 ce->ce_gid = ntohl(ondisk->gid);
1188 ce->ce_size = ntohl(ondisk->size);
1189 /* On-disk flags are just 16 bits */
1190 ce->ce_flags = ntohs(ondisk->flags);
1192 hashcpy(ce->sha1, ondisk->sha1);
1194 len = ce->ce_flags & CE_NAMEMASK;
1196 if (ce->ce_flags & CE_EXTENDED) {
1197 struct ondisk_cache_entry_extended *ondisk2;
1198 int extended_flags;
1199 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1200 extended_flags = ntohs(ondisk2->flags2) << 16;
1201 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1202 if (extended_flags & ~CE_EXTENDED_FLAGS)
1203 die("Unknown index entry format %08x", extended_flags);
1204 ce->ce_flags |= extended_flags;
1205 name = ondisk2->name;
1206 }
1207 else
1208 name = ondisk->name;
1210 if (len == CE_NAMEMASK)
1211 len = strlen(name);
1212 /*
1213 * NEEDSWORK: If the original index is crafted, this copy could
1214 * go unchecked.
1215 */
1216 memcpy(ce->name, name, len + 1);
1217 }
1219 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1220 {
1221 long per_entry;
1223 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1225 /*
1226 * Alignment can cause differences. This should be "alignof", but
1227 * since that's a gcc'ism, just use the size of a pointer.
1228 */
1229 per_entry += sizeof(void *);
1230 return ondisk_size + entries*per_entry;
1231 }
1233 /* remember to discard_cache() before reading a different cache! */
1234 int read_index_from(struct index_state *istate, const char *path)
1235 {
1236 int fd, i;
1237 struct stat st;
1238 unsigned long src_offset, dst_offset;
1239 struct cache_header *hdr;
1240 void *mmap;
1241 size_t mmap_size;
1243 errno = EBUSY;
1244 if (istate->initialized)
1245 return istate->cache_nr;
1247 errno = ENOENT;
1248 istate->timestamp.sec = 0;
1249 istate->timestamp.nsec = 0;
1250 fd = open(path, O_RDONLY);
1251 if (fd < 0) {
1252 if (errno == ENOENT)
1253 return 0;
1254 die("index file open failed (%s)", strerror(errno));
1255 }
1257 if (fstat(fd, &st))
1258 die("cannot stat the open index (%s)", strerror(errno));
1260 errno = EINVAL;
1261 mmap_size = xsize_t(st.st_size);
1262 if (mmap_size < sizeof(struct cache_header) + 20)
1263 die("index file smaller than expected");
1265 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1266 close(fd);
1267 if (mmap == MAP_FAILED)
1268 die("unable to map index file");
1270 hdr = mmap;
1271 if (verify_hdr(hdr, mmap_size) < 0)
1272 goto unmap;
1274 istate->cache_nr = ntohl(hdr->hdr_entries);
1275 istate->cache_alloc = alloc_nr(istate->cache_nr);
1276 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1278 /*
1279 * The disk format is actually larger than the in-memory format,
1280 * due to space for nsec etc, so even though the in-memory one
1281 * has room for a few more flags, we can allocate using the same
1282 * index size
1283 */
1284 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1285 istate->initialized = 1;
1287 src_offset = sizeof(*hdr);
1288 dst_offset = 0;
1289 for (i = 0; i < istate->cache_nr; i++) {
1290 struct ondisk_cache_entry *disk_ce;
1291 struct cache_entry *ce;
1293 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1294 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1295 convert_from_disk(disk_ce, ce);
1296 set_index_entry(istate, i, ce);
1298 src_offset += ondisk_ce_size(ce);
1299 dst_offset += ce_size(ce);
1300 }
1301 istate->timestamp.sec = st.st_mtime;
1302 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1304 while (src_offset <= mmap_size - 20 - 8) {
1305 /* After an array of active_nr index entries,
1306 * there can be arbitrary number of extended
1307 * sections, each of which is prefixed with
1308 * extension name (4-byte) and section length
1309 * in 4-byte network byte order.
1310 */
1311 unsigned long extsize;
1312 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1313 extsize = ntohl(extsize);
1314 if (read_index_extension(istate,
1315 (const char *) mmap + src_offset,
1316 (char *) mmap + src_offset + 8,
1317 extsize) < 0)
1318 goto unmap;
1319 src_offset += 8;
1320 src_offset += extsize;
1321 }
1322 munmap(mmap, mmap_size);
1323 return istate->cache_nr;
1325 unmap:
1326 munmap(mmap, mmap_size);
1327 errno = EINVAL;
1328 die("index file corrupt");
1329 }
1331 int is_index_unborn(struct index_state *istate)
1332 {
1333 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
1334 }
1336 int discard_index(struct index_state *istate)
1337 {
1338 istate->cache_nr = 0;
1339 istate->cache_changed = 0;
1340 istate->timestamp.sec = 0;
1341 istate->timestamp.nsec = 0;
1342 istate->name_hash_initialized = 0;
1343 free_hash(&istate->name_hash);
1344 cache_tree_free(&(istate->cache_tree));
1345 free(istate->alloc);
1346 istate->alloc = NULL;
1347 istate->initialized = 0;
1349 /* no need to throw away allocated active_cache */
1350 return 0;
1351 }
1353 int unmerged_index(const struct index_state *istate)
1354 {
1355 int i;
1356 for (i = 0; i < istate->cache_nr; i++) {
1357 if (ce_stage(istate->cache[i]))
1358 return 1;
1359 }
1360 return 0;
1361 }
1363 #define WRITE_BUFFER_SIZE 8192
1364 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1365 static unsigned long write_buffer_len;
1367 static int ce_write_flush(git_SHA_CTX *context, int fd)
1368 {
1369 unsigned int buffered = write_buffer_len;
1370 if (buffered) {
1371 git_SHA1_Update(context, write_buffer, buffered);
1372 if (write_in_full(fd, write_buffer, buffered) != buffered)
1373 return -1;
1374 write_buffer_len = 0;
1375 }
1376 return 0;
1377 }
1379 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1380 {
1381 while (len) {
1382 unsigned int buffered = write_buffer_len;
1383 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1384 if (partial > len)
1385 partial = len;
1386 memcpy(write_buffer + buffered, data, partial);
1387 buffered += partial;
1388 if (buffered == WRITE_BUFFER_SIZE) {
1389 write_buffer_len = buffered;
1390 if (ce_write_flush(context, fd))
1391 return -1;
1392 buffered = 0;
1393 }
1394 write_buffer_len = buffered;
1395 len -= partial;
1396 data = (char *) data + partial;
1397 }
1398 return 0;
1399 }
1401 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1402 unsigned int ext, unsigned int sz)
1403 {
1404 ext = htonl(ext);
1405 sz = htonl(sz);
1406 return ((ce_write(context, fd, &ext, 4) < 0) ||
1407 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1408 }
1410 static int ce_flush(git_SHA_CTX *context, int fd)
1411 {
1412 unsigned int left = write_buffer_len;
1414 if (left) {
1415 write_buffer_len = 0;
1416 git_SHA1_Update(context, write_buffer, left);
1417 }
1419 /* Flush first if not enough space for SHA1 signature */
1420 if (left + 20 > WRITE_BUFFER_SIZE) {
1421 if (write_in_full(fd, write_buffer, left) != left)
1422 return -1;
1423 left = 0;
1424 }
1426 /* Append the SHA1 signature at the end */
1427 git_SHA1_Final(write_buffer + left, context);
1428 left += 20;
1429 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1430 }
1432 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1433 {
1434 /*
1435 * The only thing we care about in this function is to smudge the
1436 * falsely clean entry due to touch-update-touch race, so we leave
1437 * everything else as they are. We are called for entries whose
1438 * ce_mtime match the index file mtime.
1439 *
1440 * Note that this actually does not do much for gitlinks, for
1441 * which ce_match_stat_basic() always goes to the actual
1442 * contents. The caller checks with is_racy_timestamp() which
1443 * always says "no" for gitlinks, so we are not called for them ;-)
1444 */
1445 struct stat st;
1447 if (lstat(ce->name, &st) < 0)
1448 return;
1449 if (ce_match_stat_basic(ce, &st))
1450 return;
1451 if (ce_modified_check_fs(ce, &st)) {
1452 /* This is "racily clean"; smudge it. Note that this
1453 * is a tricky code. At first glance, it may appear
1454 * that it can break with this sequence:
1455 *
1456 * $ echo xyzzy >frotz
1457 * $ git-update-index --add frotz
1458 * $ : >frotz
1459 * $ sleep 3
1460 * $ echo filfre >nitfol
1461 * $ git-update-index --add nitfol
1462 *
1463 * but it does not. When the second update-index runs,
1464 * it notices that the entry "frotz" has the same timestamp
1465 * as index, and if we were to smudge it by resetting its
1466 * size to zero here, then the object name recorded
1467 * in index is the 6-byte file but the cached stat information
1468 * becomes zero --- which would then match what we would
1469 * obtain from the filesystem next time we stat("frotz").
1470 *
1471 * However, the second update-index, before calling
1472 * this function, notices that the cached size is 6
1473 * bytes and what is on the filesystem is an empty
1474 * file, and never calls us, so the cached size information
1475 * for "frotz" stays 6 which does not match the filesystem.
1476 */
1477 ce->ce_size = 0;
1478 }
1479 }
1481 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1482 {
1483 int size = ondisk_ce_size(ce);
1484 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1485 char *name;
1487 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1488 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1489 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1490 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1491 ondisk->dev = htonl(ce->ce_dev);
1492 ondisk->ino = htonl(ce->ce_ino);
1493 ondisk->mode = htonl(ce->ce_mode);
1494 ondisk->uid = htonl(ce->ce_uid);
1495 ondisk->gid = htonl(ce->ce_gid);
1496 ondisk->size = htonl(ce->ce_size);
1497 hashcpy(ondisk->sha1, ce->sha1);
1498 ondisk->flags = htons(ce->ce_flags);
1499 if (ce->ce_flags & CE_EXTENDED) {
1500 struct ondisk_cache_entry_extended *ondisk2;
1501 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1502 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1503 name = ondisk2->name;
1504 }
1505 else
1506 name = ondisk->name;
1507 memcpy(name, ce->name, ce_namelen(ce));
1509 return ce_write(c, fd, ondisk, size);
1510 }
1512 int write_index(struct index_state *istate, int newfd)
1513 {
1514 git_SHA_CTX c;
1515 struct cache_header hdr;
1516 int i, err, removed, extended;
1517 struct cache_entry **cache = istate->cache;
1518 int entries = istate->cache_nr;
1519 struct stat st;
1521 for (i = removed = extended = 0; i < entries; i++) {
1522 if (cache[i]->ce_flags & CE_REMOVE)
1523 removed++;
1525 /* reduce extended entries if possible */
1526 cache[i]->ce_flags &= ~CE_EXTENDED;
1527 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1528 extended++;
1529 cache[i]->ce_flags |= CE_EXTENDED;
1530 }
1531 }
1533 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1534 /* for extended format, increase version so older git won't try to read it */
1535 hdr.hdr_version = htonl(extended ? 3 : 2);
1536 hdr.hdr_entries = htonl(entries - removed);
1538 git_SHA1_Init(&c);
1539 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1540 return -1;
1542 for (i = 0; i < entries; i++) {
1543 struct cache_entry *ce = cache[i];
1544 if (ce->ce_flags & CE_REMOVE)
1545 continue;
1546 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1547 ce_smudge_racily_clean_entry(ce);
1548 if (ce_write_entry(&c, newfd, ce) < 0)
1549 return -1;
1550 }
1552 /* Write extension data here */
1553 if (istate->cache_tree) {
1554 struct strbuf sb = STRBUF_INIT;
1556 cache_tree_write(&sb, istate->cache_tree);
1557 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1558 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1559 strbuf_release(&sb);
1560 if (err)
1561 return -1;
1562 }
1564 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1565 return -1;
1566 istate->timestamp.sec = (unsigned int)st.st_mtime;
1567 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1568 return 0;
1569 }
1571 /*
1572 * Read the index file that is potentially unmerged into given
1573 * index_state, dropping any unmerged entries. Returns true if
1574 * the index is unmerged. Callers who want to refuse to work
1575 * from an unmerged state can call this and check its return value,
1576 * instead of calling read_cache().
1577 */
1578 int read_index_unmerged(struct index_state *istate)
1579 {
1580 int i;
1581 int unmerged = 0;
1583 read_index(istate);
1584 for (i = 0; i < istate->cache_nr; i++) {
1585 struct cache_entry *ce = istate->cache[i];
1586 struct cache_entry *new_ce;
1587 int size, len;
1589 if (!ce_stage(ce))
1590 continue;
1591 unmerged = 1;
1592 len = strlen(ce->name);
1593 size = cache_entry_size(len);
1594 new_ce = xcalloc(1, size);
1595 hashcpy(new_ce->sha1, ce->sha1);
1596 memcpy(new_ce->name, ce->name, len);
1597 new_ce->ce_flags = create_ce_flags(len, 0);
1598 new_ce->ce_mode = ce->ce_mode;
1599 if (add_index_entry(istate, new_ce, 0))
1600 return error("%s: cannot drop to stage #0",
1601 ce->name);
1602 i = index_name_pos(istate, new_ce->name, len);
1603 }
1604 return unmerged;
1605 }
1607 struct update_callback_data
1608 {
1609 int flags;
1610 int add_errors;
1611 };
1613 static void update_callback(struct diff_queue_struct *q,
1614 struct diff_options *opt, void *cbdata)
1615 {
1616 int i;
1617 struct update_callback_data *data = cbdata;
1619 for (i = 0; i < q->nr; i++) {
1620 struct diff_filepair *p = q->queue[i];
1621 const char *path = p->one->path;
1622 switch (p->status) {
1623 default:
1624 die("unexpected diff status %c", p->status);
1625 case DIFF_STATUS_UNMERGED:
1626 /*
1627 * ADD_CACHE_IGNORE_REMOVAL is unset if "git
1628 * add -u" is calling us, In such a case, a
1629 * missing work tree file needs to be removed
1630 * if there is an unmerged entry at stage #2,
1631 * but such a diff record is followed by
1632 * another with DIFF_STATUS_DELETED (and if
1633 * there is no stage #2, we won't see DELETED
1634 * nor MODIFIED). We can simply continue
1635 * either way.
1636 */
1637 if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL))
1638 continue;
1639 /*
1640 * Otherwise, it is "git add path" is asking
1641 * to explicitly add it; we fall through. A
1642 * missing work tree file is an error and is
1643 * caught by add_file_to_index() in such a
1644 * case.
1645 */
1646 case DIFF_STATUS_MODIFIED:
1647 case DIFF_STATUS_TYPE_CHANGED:
1648 if (add_file_to_index(&the_index, path, data->flags)) {
1649 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1650 die("updating files failed");
1651 data->add_errors++;
1652 }
1653 break;
1654 case DIFF_STATUS_DELETED:
1655 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1656 break;
1657 if (!(data->flags & ADD_CACHE_PRETEND))
1658 remove_file_from_index(&the_index, path);
1659 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1660 printf("remove '%s'\n", path);
1661 break;
1662 }
1663 }
1664 }
1666 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1667 {
1668 struct update_callback_data data;
1669 struct rev_info rev;
1670 init_revisions(&rev, prefix);
1671 setup_revisions(0, NULL, &rev, NULL);
1672 rev.prune_data = pathspec;
1673 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1674 rev.diffopt.format_callback = update_callback;
1675 data.flags = flags;
1676 data.add_errors = 0;
1677 rev.diffopt.format_callback_data = &data;
1678 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1679 return !!data.add_errors;
1680 }
1682 /*
1683 * Returns 1 if the path is an "other" path with respect to
1684 * the index; that is, the path is not mentioned in the index at all,
1685 * either as a file, a directory with some files in the index,
1686 * or as an unmerged entry.
1687 *
1688 * We helpfully remove a trailing "/" from directories so that
1689 * the output of read_directory can be used as-is.
1690 */
1691 int index_name_is_other(const struct index_state *istate, const char *name,
1692 int namelen)
1693 {
1694 int pos;
1695 if (namelen && name[namelen - 1] == '/')
1696 namelen--;
1697 pos = index_name_pos(istate, name, namelen);
1698 if (0 <= pos)
1699 return 0; /* exact match */
1700 pos = -pos - 1;
1701 if (pos < istate->cache_nr) {
1702 struct cache_entry *ce = istate->cache[pos];
1703 if (ce_namelen(ce) == namelen &&
1704 !memcmp(ce->name, name, namelen))
1705 return 0; /* Yup, this one exists unmerged */
1706 }
1707 return 1;
1708 }