reset: make the reminder output consistent with "checkout"
[git/git.git] / read-cache.c
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"
17
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 */
26
27 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
28 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
29
30 struct index_state the_index;
31
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 }
37
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];
41
42 remove_name_hash(old);
43 set_index_entry(istate, nr, ce);
44 istate->cache_changed = 1;
45 }
46
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);
51
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);
57
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 }
62
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;
79
80 if (assume_unchanged)
81 ce->ce_flags |= CE_VALID;
82
83 if (S_ISREG(st->st_mode))
84 ce_mark_uptodate(ce);
85 }
86
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);
91
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 }
100
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;
108
109 if (strbuf_readlink(&sb, ce->name, expected_size))
110 return -1;
111
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 }
121
122 static int ce_compare_gitlink(struct cache_entry *ce)
123 {
124 unsigned char sha1[20];
125
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 }
138
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 }
158
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 };
165
166 return !hashcmp(sha1, empty_blob_sha1);
167 }
168
169 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
170 {
171 unsigned int changed = 0;
172
173 if (ce->ce_flags & CE_REMOVE)
174 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
175
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;
205
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
212
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;
218
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
228
229 if (ce->ce_size != (unsigned int) st->st_size)
230 changed |= DATA_CHANGED;
231
232 /* Racily smudged entry? */
233 if (!ce->ce_size) {
234 if (!is_empty_blob_sha1(ce->sha1))
235 changed |= DATA_CHANGED;
236 }
237
238 return changed;
239 }
240
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 }
255
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;
263
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;
270
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;
278
279 changed = ce_match_stat_basic(ce, st);
280
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 }
303
304 return changed;
305 }
306
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;
311
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;
321
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;
339
340 changed_fs = ce_modified_check_fs(ce, st);
341 if (changed_fs)
342 return changed | changed_fs;
343 return 0;
344 }
345
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;
352
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 }
364
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;
380
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 }
399
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;
406
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;
414
415 /* Compare stages */
416 flags1 &= CE_STAGEMASK;
417 flags2 &= CE_STAGEMASK;
418
419 if (flags1 < flags2)
420 return -1;
421 if (flags1 > flags2)
422 return 1;
423 return 0;
424 }
425
426 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
427 {
428 int first, last;
429
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 }
446
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];
451
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 }
462
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;
472
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 }
482
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 }
493
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 }
498
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;
504
505 if (pos >= 0)
506 return pos;
507
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;
513
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 }
521
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 }
527
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;
541
542 if (alias->ce_flags & CE_ADDED)
543 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
544
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 }
553
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 }
561
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));
573
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);
576
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;
590
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);
599
600 ent = (0 <= pos) ? istate->cache[pos] : NULL;
601 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
602 }
603
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);
617
618 if (ignore_case && alias && different_name(ce, alias))
619 ce = create_alias_ce(ce, alias);
620 ce->ce_flags |= CE_ADDED;
621
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);
627
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 }
636
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_errno("unable to stat '%s'", path);
642 return add_to_index(istate, path, &st, flags);
643 }
644
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;
651
652 if (!verify_path(path)) {
653 error("Invalid path '%s'", path);
654 return NULL;
655 }
656
657 len = strlen(path);
658 size = cache_entry_size(len);
659 ce = xcalloc(1, size);
660
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);
665
666 if (refresh)
667 return refresh_cache_entry(ce, 0);
668
669 return ce;
670 }
671
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 }
677
678 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
679 {
680 const char *match, *name;
681 int len;
682
683 if (!pathspec)
684 return 1;
685
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 }
703
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;
723
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 }
741
742 int verify_path(const char *path)
743 {
744 char c;
745
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 }
767
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;
779
780 while (pos < istate->cache_nr) {
781 struct cache_entry *p = istate->cache[pos++];
782
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 }
800
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);
812
813 for (;;) {
814 int len;
815
816 for (;;) {
817 if (*--slash == '/')
818 break;
819 if (slash <= ce->name)
820 return retval;
821 }
822 len = slash - name;
823
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;
844
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 }
869
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;
884
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;
890
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);
897
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 }
904
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;
912
913 cache_tree_invalidate_path(istate->cache_tree, ce->name);
914 pos = index_name_pos(istate, ce->name, ce->ce_flags);
915
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;
923
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 }
935
936 if (!ok_to_add)
937 return -1;
938 if (!verify_path(ce->name))
939 return error("Invalid path '%s'", ce->name);
940
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 }
951
952 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
953 {
954 int pos;
955
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 }
965
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 }
972
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 }
983
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;
1003
1004 if (ce_uptodate(ce))
1005 return ce;
1006
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 }
1015
1016 if (lstat(ce->name, &st) < 0) {
1017 if (err)
1018 *err = errno;
1019 return NULL;
1020 }
1021
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 }
1044
1045 if (ie_modified(istate, ce, &st, options)) {
1046 if (err)
1047 *err = EINVAL;
1048 return NULL;
1049 }
1050
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;
1064
1065 return updated;
1066 }
1067
1068 static void show_file(const char * fmt, const char * name, int in_porcelain,
1069 int * first, char *header_msg)
1070 {
1071 if (in_porcelain && *first && header_msg) {
1072 printf("%s\n", header_msg);
1073 *first=0;
1074 }
1075 printf(fmt, name);
1076 }
1077
1078 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1079 char *seen, char *header_msg)
1080 {
1081 int i;
1082 int has_errors = 0;
1083 int really = (flags & REFRESH_REALLY) != 0;
1084 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1085 int quiet = (flags & REFRESH_QUIET) != 0;
1086 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1087 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1088 int first = 1;
1089 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1090 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1091 const char *needs_update_fmt;
1092 const char *needs_merge_fmt;
1093
1094 needs_update_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1095 needs_merge_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1096 for (i = 0; i < istate->cache_nr; i++) {
1097 struct cache_entry *ce, *new;
1098 int cache_errno = 0;
1099
1100 ce = istate->cache[i];
1101 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1102 continue;
1103
1104 if (ce_stage(ce)) {
1105 while ((i < istate->cache_nr) &&
1106 ! strcmp(istate->cache[i]->name, ce->name))
1107 i++;
1108 i--;
1109 if (allow_unmerged)
1110 continue;
1111 show_file(needs_merge_fmt, ce->name, in_porcelain, &first, header_msg);
1112 has_errors = 1;
1113 continue;
1114 }
1115
1116 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1117 continue;
1118
1119 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1120 if (new == ce)
1121 continue;
1122 if (!new) {
1123 if (not_new && cache_errno == ENOENT)
1124 continue;
1125 if (really && cache_errno == EINVAL) {
1126 /* If we are doing --really-refresh that
1127 * means the index is not valid anymore.
1128 */
1129 ce->ce_flags &= ~CE_VALID;
1130 istate->cache_changed = 1;
1131 }
1132 if (quiet)
1133 continue;
1134 show_file(needs_update_fmt, ce->name, in_porcelain, &first, header_msg);
1135 has_errors = 1;
1136 continue;
1137 }
1138
1139 replace_index_entry(istate, i, new);
1140 }
1141 return has_errors;
1142 }
1143
1144 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1145 {
1146 return refresh_cache_ent(&the_index, ce, really, NULL);
1147 }
1148
1149 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1150 {
1151 git_SHA_CTX c;
1152 unsigned char sha1[20];
1153
1154 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1155 return error("bad signature");
1156 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1157 return error("bad index version");
1158 git_SHA1_Init(&c);
1159 git_SHA1_Update(&c, hdr, size - 20);
1160 git_SHA1_Final(sha1, &c);
1161 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1162 return error("bad index file sha1 signature");
1163 return 0;
1164 }
1165
1166 static int read_index_extension(struct index_state *istate,
1167 const char *ext, void *data, unsigned long sz)
1168 {
1169 switch (CACHE_EXT(ext)) {
1170 case CACHE_EXT_TREE:
1171 istate->cache_tree = cache_tree_read(data, sz);
1172 break;
1173 default:
1174 if (*ext < 'A' || 'Z' < *ext)
1175 return error("index uses %.4s extension, which we do not understand",
1176 ext);
1177 fprintf(stderr, "ignoring %.4s extension\n", ext);
1178 break;
1179 }
1180 return 0;
1181 }
1182
1183 int read_index(struct index_state *istate)
1184 {
1185 return read_index_from(istate, get_index_file());
1186 }
1187
1188 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1189 {
1190 size_t len;
1191 const char *name;
1192
1193 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1194 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1195 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1196 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1197 ce->ce_dev = ntohl(ondisk->dev);
1198 ce->ce_ino = ntohl(ondisk->ino);
1199 ce->ce_mode = ntohl(ondisk->mode);
1200 ce->ce_uid = ntohl(ondisk->uid);
1201 ce->ce_gid = ntohl(ondisk->gid);
1202 ce->ce_size = ntohl(ondisk->size);
1203 /* On-disk flags are just 16 bits */
1204 ce->ce_flags = ntohs(ondisk->flags);
1205
1206 hashcpy(ce->sha1, ondisk->sha1);
1207
1208 len = ce->ce_flags & CE_NAMEMASK;
1209
1210 if (ce->ce_flags & CE_EXTENDED) {
1211 struct ondisk_cache_entry_extended *ondisk2;
1212 int extended_flags;
1213 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1214 extended_flags = ntohs(ondisk2->flags2) << 16;
1215 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1216 if (extended_flags & ~CE_EXTENDED_FLAGS)
1217 die("Unknown index entry format %08x", extended_flags);
1218 ce->ce_flags |= extended_flags;
1219 name = ondisk2->name;
1220 }
1221 else
1222 name = ondisk->name;
1223
1224 if (len == CE_NAMEMASK)
1225 len = strlen(name);
1226 /*
1227 * NEEDSWORK: If the original index is crafted, this copy could
1228 * go unchecked.
1229 */
1230 memcpy(ce->name, name, len + 1);
1231 }
1232
1233 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1234 {
1235 long per_entry;
1236
1237 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1238
1239 /*
1240 * Alignment can cause differences. This should be "alignof", but
1241 * since that's a gcc'ism, just use the size of a pointer.
1242 */
1243 per_entry += sizeof(void *);
1244 return ondisk_size + entries*per_entry;
1245 }
1246
1247 /* remember to discard_cache() before reading a different cache! */
1248 int read_index_from(struct index_state *istate, const char *path)
1249 {
1250 int fd, i;
1251 struct stat st;
1252 unsigned long src_offset, dst_offset;
1253 struct cache_header *hdr;
1254 void *mmap;
1255 size_t mmap_size;
1256
1257 errno = EBUSY;
1258 if (istate->initialized)
1259 return istate->cache_nr;
1260
1261 errno = ENOENT;
1262 istate->timestamp.sec = 0;
1263 istate->timestamp.nsec = 0;
1264 fd = open(path, O_RDONLY);
1265 if (fd < 0) {
1266 if (errno == ENOENT)
1267 return 0;
1268 die_errno("index file open failed");
1269 }
1270
1271 if (fstat(fd, &st))
1272 die_errno("cannot stat the open index");
1273
1274 errno = EINVAL;
1275 mmap_size = xsize_t(st.st_size);
1276 if (mmap_size < sizeof(struct cache_header) + 20)
1277 die("index file smaller than expected");
1278
1279 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1280 close(fd);
1281 if (mmap == MAP_FAILED)
1282 die_errno("unable to map index file");
1283
1284 hdr = mmap;
1285 if (verify_hdr(hdr, mmap_size) < 0)
1286 goto unmap;
1287
1288 istate->cache_nr = ntohl(hdr->hdr_entries);
1289 istate->cache_alloc = alloc_nr(istate->cache_nr);
1290 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1291
1292 /*
1293 * The disk format is actually larger than the in-memory format,
1294 * due to space for nsec etc, so even though the in-memory one
1295 * has room for a few more flags, we can allocate using the same
1296 * index size
1297 */
1298 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1299 istate->initialized = 1;
1300
1301 src_offset = sizeof(*hdr);
1302 dst_offset = 0;
1303 for (i = 0; i < istate->cache_nr; i++) {
1304 struct ondisk_cache_entry *disk_ce;
1305 struct cache_entry *ce;
1306
1307 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1308 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1309 convert_from_disk(disk_ce, ce);
1310 set_index_entry(istate, i, ce);
1311
1312 src_offset += ondisk_ce_size(ce);
1313 dst_offset += ce_size(ce);
1314 }
1315 istate->timestamp.sec = st.st_mtime;
1316 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1317
1318 while (src_offset <= mmap_size - 20 - 8) {
1319 /* After an array of active_nr index entries,
1320 * there can be arbitrary number of extended
1321 * sections, each of which is prefixed with
1322 * extension name (4-byte) and section length
1323 * in 4-byte network byte order.
1324 */
1325 unsigned long extsize;
1326 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1327 extsize = ntohl(extsize);
1328 if (read_index_extension(istate,
1329 (const char *) mmap + src_offset,
1330 (char *) mmap + src_offset + 8,
1331 extsize) < 0)
1332 goto unmap;
1333 src_offset += 8;
1334 src_offset += extsize;
1335 }
1336 munmap(mmap, mmap_size);
1337 return istate->cache_nr;
1338
1339 unmap:
1340 munmap(mmap, mmap_size);
1341 errno = EINVAL;
1342 die("index file corrupt");
1343 }
1344
1345 int is_index_unborn(struct index_state *istate)
1346 {
1347 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
1348 }
1349
1350 int discard_index(struct index_state *istate)
1351 {
1352 istate->cache_nr = 0;
1353 istate->cache_changed = 0;
1354 istate->timestamp.sec = 0;
1355 istate->timestamp.nsec = 0;
1356 istate->name_hash_initialized = 0;
1357 free_hash(&istate->name_hash);
1358 cache_tree_free(&(istate->cache_tree));
1359 free(istate->alloc);
1360 istate->alloc = NULL;
1361 istate->initialized = 0;
1362
1363 /* no need to throw away allocated active_cache */
1364 return 0;
1365 }
1366
1367 int unmerged_index(const struct index_state *istate)
1368 {
1369 int i;
1370 for (i = 0; i < istate->cache_nr; i++) {
1371 if (ce_stage(istate->cache[i]))
1372 return 1;
1373 }
1374 return 0;
1375 }
1376
1377 #define WRITE_BUFFER_SIZE 8192
1378 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1379 static unsigned long write_buffer_len;
1380
1381 static int ce_write_flush(git_SHA_CTX *context, int fd)
1382 {
1383 unsigned int buffered = write_buffer_len;
1384 if (buffered) {
1385 git_SHA1_Update(context, write_buffer, buffered);
1386 if (write_in_full(fd, write_buffer, buffered) != buffered)
1387 return -1;
1388 write_buffer_len = 0;
1389 }
1390 return 0;
1391 }
1392
1393 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1394 {
1395 while (len) {
1396 unsigned int buffered = write_buffer_len;
1397 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1398 if (partial > len)
1399 partial = len;
1400 memcpy(write_buffer + buffered, data, partial);
1401 buffered += partial;
1402 if (buffered == WRITE_BUFFER_SIZE) {
1403 write_buffer_len = buffered;
1404 if (ce_write_flush(context, fd))
1405 return -1;
1406 buffered = 0;
1407 }
1408 write_buffer_len = buffered;
1409 len -= partial;
1410 data = (char *) data + partial;
1411 }
1412 return 0;
1413 }
1414
1415 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1416 unsigned int ext, unsigned int sz)
1417 {
1418 ext = htonl(ext);
1419 sz = htonl(sz);
1420 return ((ce_write(context, fd, &ext, 4) < 0) ||
1421 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1422 }
1423
1424 static int ce_flush(git_SHA_CTX *context, int fd)
1425 {
1426 unsigned int left = write_buffer_len;
1427
1428 if (left) {
1429 write_buffer_len = 0;
1430 git_SHA1_Update(context, write_buffer, left);
1431 }
1432
1433 /* Flush first if not enough space for SHA1 signature */
1434 if (left + 20 > WRITE_BUFFER_SIZE) {
1435 if (write_in_full(fd, write_buffer, left) != left)
1436 return -1;
1437 left = 0;
1438 }
1439
1440 /* Append the SHA1 signature at the end */
1441 git_SHA1_Final(write_buffer + left, context);
1442 left += 20;
1443 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1444 }
1445
1446 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1447 {
1448 /*
1449 * The only thing we care about in this function is to smudge the
1450 * falsely clean entry due to touch-update-touch race, so we leave
1451 * everything else as they are. We are called for entries whose
1452 * ce_mtime match the index file mtime.
1453 *
1454 * Note that this actually does not do much for gitlinks, for
1455 * which ce_match_stat_basic() always goes to the actual
1456 * contents. The caller checks with is_racy_timestamp() which
1457 * always says "no" for gitlinks, so we are not called for them ;-)
1458 */
1459 struct stat st;
1460
1461 if (lstat(ce->name, &st) < 0)
1462 return;
1463 if (ce_match_stat_basic(ce, &st))
1464 return;
1465 if (ce_modified_check_fs(ce, &st)) {
1466 /* This is "racily clean"; smudge it. Note that this
1467 * is a tricky code. At first glance, it may appear
1468 * that it can break with this sequence:
1469 *
1470 * $ echo xyzzy >frotz
1471 * $ git-update-index --add frotz
1472 * $ : >frotz
1473 * $ sleep 3
1474 * $ echo filfre >nitfol
1475 * $ git-update-index --add nitfol
1476 *
1477 * but it does not. When the second update-index runs,
1478 * it notices that the entry "frotz" has the same timestamp
1479 * as index, and if we were to smudge it by resetting its
1480 * size to zero here, then the object name recorded
1481 * in index is the 6-byte file but the cached stat information
1482 * becomes zero --- which would then match what we would
1483 * obtain from the filesystem next time we stat("frotz").
1484 *
1485 * However, the second update-index, before calling
1486 * this function, notices that the cached size is 6
1487 * bytes and what is on the filesystem is an empty
1488 * file, and never calls us, so the cached size information
1489 * for "frotz" stays 6 which does not match the filesystem.
1490 */
1491 ce->ce_size = 0;
1492 }
1493 }
1494
1495 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1496 {
1497 int size = ondisk_ce_size(ce);
1498 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1499 char *name;
1500
1501 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1502 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1503 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1504 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1505 ondisk->dev = htonl(ce->ce_dev);
1506 ondisk->ino = htonl(ce->ce_ino);
1507 ondisk->mode = htonl(ce->ce_mode);
1508 ondisk->uid = htonl(ce->ce_uid);
1509 ondisk->gid = htonl(ce->ce_gid);
1510 ondisk->size = htonl(ce->ce_size);
1511 hashcpy(ondisk->sha1, ce->sha1);
1512 ondisk->flags = htons(ce->ce_flags);
1513 if (ce->ce_flags & CE_EXTENDED) {
1514 struct ondisk_cache_entry_extended *ondisk2;
1515 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1516 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1517 name = ondisk2->name;
1518 }
1519 else
1520 name = ondisk->name;
1521 memcpy(name, ce->name, ce_namelen(ce));
1522
1523 return ce_write(c, fd, ondisk, size);
1524 }
1525
1526 int write_index(struct index_state *istate, int newfd)
1527 {
1528 git_SHA_CTX c;
1529 struct cache_header hdr;
1530 int i, err, removed, extended;
1531 struct cache_entry **cache = istate->cache;
1532 int entries = istate->cache_nr;
1533 struct stat st;
1534
1535 for (i = removed = extended = 0; i < entries; i++) {
1536 if (cache[i]->ce_flags & CE_REMOVE)
1537 removed++;
1538
1539 /* reduce extended entries if possible */
1540 cache[i]->ce_flags &= ~CE_EXTENDED;
1541 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1542 extended++;
1543 cache[i]->ce_flags |= CE_EXTENDED;
1544 }
1545 }
1546
1547 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1548 /* for extended format, increase version so older git won't try to read it */
1549 hdr.hdr_version = htonl(extended ? 3 : 2);
1550 hdr.hdr_entries = htonl(entries - removed);
1551
1552 git_SHA1_Init(&c);
1553 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1554 return -1;
1555
1556 for (i = 0; i < entries; i++) {
1557 struct cache_entry *ce = cache[i];
1558 if (ce->ce_flags & CE_REMOVE)
1559 continue;
1560 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1561 ce_smudge_racily_clean_entry(ce);
1562 if (ce_write_entry(&c, newfd, ce) < 0)
1563 return -1;
1564 }
1565
1566 /* Write extension data here */
1567 if (istate->cache_tree) {
1568 struct strbuf sb = STRBUF_INIT;
1569
1570 cache_tree_write(&sb, istate->cache_tree);
1571 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1572 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1573 strbuf_release(&sb);
1574 if (err)
1575 return -1;
1576 }
1577
1578 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1579 return -1;
1580 istate->timestamp.sec = (unsigned int)st.st_mtime;
1581 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1582 return 0;
1583 }
1584
1585 /*
1586 * Read the index file that is potentially unmerged into given
1587 * index_state, dropping any unmerged entries. Returns true if
1588 * the index is unmerged. Callers who want to refuse to work
1589 * from an unmerged state can call this and check its return value,
1590 * instead of calling read_cache().
1591 */
1592 int read_index_unmerged(struct index_state *istate)
1593 {
1594 int i;
1595 int unmerged = 0;
1596
1597 read_index(istate);
1598 for (i = 0; i < istate->cache_nr; i++) {
1599 struct cache_entry *ce = istate->cache[i];
1600 struct cache_entry *new_ce;
1601 int size, len;
1602
1603 if (!ce_stage(ce))
1604 continue;
1605 unmerged = 1;
1606 len = strlen(ce->name);
1607 size = cache_entry_size(len);
1608 new_ce = xcalloc(1, size);
1609 hashcpy(new_ce->sha1, ce->sha1);
1610 memcpy(new_ce->name, ce->name, len);
1611 new_ce->ce_flags = create_ce_flags(len, 0);
1612 new_ce->ce_mode = ce->ce_mode;
1613 if (add_index_entry(istate, new_ce, 0))
1614 return error("%s: cannot drop to stage #0",
1615 ce->name);
1616 i = index_name_pos(istate, new_ce->name, len);
1617 }
1618 return unmerged;
1619 }
1620
1621 struct update_callback_data
1622 {
1623 int flags;
1624 int add_errors;
1625 };
1626
1627 static void update_callback(struct diff_queue_struct *q,
1628 struct diff_options *opt, void *cbdata)
1629 {
1630 int i;
1631 struct update_callback_data *data = cbdata;
1632
1633 for (i = 0; i < q->nr; i++) {
1634 struct diff_filepair *p = q->queue[i];
1635 const char *path = p->one->path;
1636 switch (p->status) {
1637 default:
1638 die("unexpected diff status %c", p->status);
1639 case DIFF_STATUS_UNMERGED:
1640 /*
1641 * ADD_CACHE_IGNORE_REMOVAL is unset if "git
1642 * add -u" is calling us, In such a case, a
1643 * missing work tree file needs to be removed
1644 * if there is an unmerged entry at stage #2,
1645 * but such a diff record is followed by
1646 * another with DIFF_STATUS_DELETED (and if
1647 * there is no stage #2, we won't see DELETED
1648 * nor MODIFIED). We can simply continue
1649 * either way.
1650 */
1651 if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL))
1652 continue;
1653 /*
1654 * Otherwise, it is "git add path" is asking
1655 * to explicitly add it; we fall through. A
1656 * missing work tree file is an error and is
1657 * caught by add_file_to_index() in such a
1658 * case.
1659 */
1660 case DIFF_STATUS_MODIFIED:
1661 case DIFF_STATUS_TYPE_CHANGED:
1662 if (add_file_to_index(&the_index, path, data->flags)) {
1663 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1664 die("updating files failed");
1665 data->add_errors++;
1666 }
1667 break;
1668 case DIFF_STATUS_DELETED:
1669 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1670 break;
1671 if (!(data->flags & ADD_CACHE_PRETEND))
1672 remove_file_from_index(&the_index, path);
1673 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1674 printf("remove '%s'\n", path);
1675 break;
1676 }
1677 }
1678 }
1679
1680 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1681 {
1682 struct update_callback_data data;
1683 struct rev_info rev;
1684 init_revisions(&rev, prefix);
1685 setup_revisions(0, NULL, &rev, NULL);
1686 rev.prune_data = pathspec;
1687 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1688 rev.diffopt.format_callback = update_callback;
1689 data.flags = flags;
1690 data.add_errors = 0;
1691 rev.diffopt.format_callback_data = &data;
1692 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1693 return !!data.add_errors;
1694 }
1695
1696 /*
1697 * Returns 1 if the path is an "other" path with respect to
1698 * the index; that is, the path is not mentioned in the index at all,
1699 * either as a file, a directory with some files in the index,
1700 * or as an unmerged entry.
1701 *
1702 * We helpfully remove a trailing "/" from directories so that
1703 * the output of read_directory can be used as-is.
1704 */
1705 int index_name_is_other(const struct index_state *istate, const char *name,
1706 int namelen)
1707 {
1708 int pos;
1709 if (namelen && name[namelen - 1] == '/')
1710 namelen--;
1711 pos = index_name_pos(istate, name, namelen);
1712 if (0 <= pos)
1713 return 0; /* exact match */
1714 pos = -pos - 1;
1715 if (pos < istate->cache_nr) {
1716 struct cache_entry *ce = istate->cache[pos];
1717 if (ce_namelen(ce) == namelen &&
1718 !memcmp(ce->name, name, namelen))
1719 return 0; /* Yup, this one exists unmerged */
1720 }
1721 return 1;
1722 }