Merge branch 'py/diff-submodule'
[git/git.git] / read-cache.c
CommitLineData
8bc9a0c7
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
4aab5b46 6#define NO_THE_INDEX_COMPATIBILITY_MACROS
e83c5163 7#include "cache.h"
bad68ec9 8#include "cache-tree.h"
f35a6d3b 9#include "refs.h"
d616813d 10#include "dir.h"
bad68ec9
JH
11
12/* Index extensions.
13 *
14 * The first letter should be 'A'..'Z' for extensions that are not
15 * necessary for a correct operation (i.e. optimization data).
16 * When new extensions are added that _needs_ to be understood in
17 * order to correctly interpret the index file, pick character that
18 * is outside the range, to cause the reader to abort.
19 */
20
21#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
22#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
e83c5163 23
228e94f9 24struct index_state the_index;
8fd2cb40 25
9cb76b8c
JH
26static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
27{
28 istate->cache[nr] = ce;
96872bc2 29 add_name_hash(istate, ce);
9cb76b8c
JH
30}
31
cf558704
LT
32static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
33{
34 struct cache_entry *old = istate->cache[nr];
35
96872bc2 36 remove_name_hash(old);
a22c6371 37 set_index_entry(istate, nr, ce);
cf558704
LT
38 istate->cache_changed = 1;
39}
40
415e96c8
JH
41/*
42 * This only updates the "non-critical" parts of the directory
43 * cache, ie the parts that aren't tracked by GIT, and only used
44 * to validate the cache.
45 */
46void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
47{
7a51ed66
LT
48 ce->ce_ctime = st->st_ctime;
49 ce->ce_mtime = st->st_mtime;
50 ce->ce_dev = st->st_dev;
51 ce->ce_ino = st->st_ino;
52 ce->ce_uid = st->st_uid;
53 ce->ce_gid = st->st_gid;
54 ce->ce_size = st->st_size;
5f73076c
JH
55
56 if (assume_unchanged)
7a51ed66 57 ce->ce_flags |= CE_VALID;
eadb5831
JH
58
59 if (S_ISREG(st->st_mode))
60 ce_mark_uptodate(ce);
415e96c8
JH
61}
62
29e4d363
JH
63static int ce_compare_data(struct cache_entry *ce, struct stat *st)
64{
65 int match = -1;
66 int fd = open(ce->name, O_RDONLY);
67
68 if (fd >= 0) {
69 unsigned char sha1[20];
53bca91a 70 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
a89fccd2 71 match = hashcmp(sha1, ce->sha1);
7f8508e8 72 /* index_fd() closed the file descriptor already */
29e4d363
JH
73 }
74 return match;
75}
76
dc49cd76 77static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
29e4d363
JH
78{
79 int match = -1;
80 char *target;
81 void *buffer;
82 unsigned long size;
21666f1a 83 enum object_type type;
29e4d363
JH
84 int len;
85
86 target = xmalloc(expected_size);
87 len = readlink(ce->name, target, expected_size);
88 if (len != expected_size) {
89 free(target);
90 return -1;
91 }
21666f1a 92 buffer = read_sha1_file(ce->sha1, &type, &size);
29e4d363
JH
93 if (!buffer) {
94 free(target);
95 return -1;
96 }
97 if (size == expected_size)
98 match = memcmp(buffer, target, size);
99 free(buffer);
100 free(target);
101 return match;
102}
103
f35a6d3b
LT
104static int ce_compare_gitlink(struct cache_entry *ce)
105{
106 unsigned char sha1[20];
107
108 /*
109 * We don't actually require that the .git directory
302b9282 110 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
111 * might even be missing (in case nobody populated that
112 * sub-project).
113 *
114 * If so, we consider it always to match.
115 */
116 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
117 return 0;
118 return hashcmp(sha1, ce->sha1);
119}
120
29e4d363
JH
121static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
122{
123 switch (st->st_mode & S_IFMT) {
124 case S_IFREG:
125 if (ce_compare_data(ce, st))
126 return DATA_CHANGED;
127 break;
128 case S_IFLNK:
dc49cd76 129 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
130 return DATA_CHANGED;
131 break;
a8ee75bc 132 case S_IFDIR:
7a51ed66 133 if (S_ISGITLINK(ce->ce_mode))
a8ee75bc 134 return 0;
29e4d363
JH
135 default:
136 return TYPE_CHANGED;
137 }
138 return 0;
139}
140
407c8eb0 141static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
734aab75
LT
142{
143 unsigned int changed = 0;
144
7a51ed66
LT
145 if (ce->ce_flags & CE_REMOVE)
146 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
147
148 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
149 case S_IFREG:
150 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
151 /* We consider only the owner x bit to be relevant for
152 * "mode changes"
153 */
154 if (trust_executable_bit &&
7a51ed66 155 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 156 changed |= MODE_CHANGED;
8ae0a8c5
KS
157 break;
158 case S_IFLNK:
78a8d641
JS
159 if (!S_ISLNK(st->st_mode) &&
160 (has_symlinks || !S_ISREG(st->st_mode)))
161 changed |= TYPE_CHANGED;
8ae0a8c5 162 break;
302b9282 163 case S_IFGITLINK:
f35a6d3b
LT
164 if (!S_ISDIR(st->st_mode))
165 changed |= TYPE_CHANGED;
166 else if (ce_compare_gitlink(ce))
167 changed |= DATA_CHANGED;
a8ee75bc 168 return changed;
8ae0a8c5 169 default:
7a51ed66 170 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 171 }
7a51ed66 172 if (ce->ce_mtime != (unsigned int) st->st_mtime)
ccc4feb5 173 changed |= MTIME_CHANGED;
7a51ed66 174 if (ce->ce_ctime != (unsigned int) st->st_ctime)
734aab75 175 changed |= CTIME_CHANGED;
ccc4feb5 176
7a51ed66
LT
177 if (ce->ce_uid != (unsigned int) st->st_uid ||
178 ce->ce_gid != (unsigned int) st->st_gid)
734aab75 179 changed |= OWNER_CHANGED;
7a51ed66 180 if (ce->ce_ino != (unsigned int) st->st_ino)
734aab75 181 changed |= INODE_CHANGED;
2cb45e95
LT
182
183#ifdef USE_STDEV
184 /*
185 * st_dev breaks on network filesystems where different
186 * clients will have different views of what "device"
187 * the filesystem is on
188 */
7a51ed66 189 if (ce->ce_dev != (unsigned int) st->st_dev)
2cb45e95
LT
190 changed |= INODE_CHANGED;
191#endif
192
7a51ed66 193 if (ce->ce_size != (unsigned int) st->st_size)
734aab75 194 changed |= DATA_CHANGED;
b0391890 195
407c8eb0
JH
196 return changed;
197}
198
d1f128b0 199static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
6d91da6d 200{
050288d5
JH
201 return (!S_ISGITLINK(ce->ce_mode) &&
202 istate->timestamp &&
6d91da6d
JH
203 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
204}
205
d1f128b0 206int ie_match_stat(const struct index_state *istate,
4bd5b7da
JH
207 struct cache_entry *ce, struct stat *st,
208 unsigned int options)
407c8eb0 209{
5f73076c 210 unsigned int changed;
4bd5b7da
JH
211 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
212 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
5f73076c
JH
213
214 /*
215 * If it's marked as always valid in the index, it's
216 * valid whatever the checked-out copy says.
217 */
7a51ed66 218 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c
JH
219 return 0;
220
221 changed = ce_match_stat_basic(ce, st);
407c8eb0 222
29e4d363
JH
223 /*
224 * Within 1 second of this sequence:
225 * echo xyzzy >file && git-update-index --add file
226 * running this command:
227 * echo frotz >file
228 * would give a falsely clean cache entry. The mtime and
229 * length match the cache, and other stat fields do not change.
230 *
231 * We could detect this at update-index time (the cache entry
232 * being registered/updated records the same time as "now")
233 * and delay the return from git-update-index, but that would
234 * effectively mean we can make at most one commit per second,
235 * which is not acceptable. Instead, we check cache entries
236 * whose mtime are the same as the index file timestamp more
5f73076c 237 * carefully than others.
29e4d363 238 */
6d91da6d 239 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
240 if (assume_racy_is_modified)
241 changed |= DATA_CHANGED;
242 else
243 changed |= ce_modified_check_fs(ce, st);
244 }
b0391890 245
29e4d363 246 return changed;
b0391890
JH
247}
248
d1f128b0 249int ie_modified(const struct index_state *istate,
4bd5b7da 250 struct cache_entry *ce, struct stat *st, unsigned int options)
b0391890 251{
29e4d363 252 int changed, changed_fs;
4bd5b7da
JH
253
254 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
255 if (!changed)
256 return 0;
b0391890
JH
257 /*
258 * If the mode or type has changed, there's no point in trying
259 * to refresh the entry - it's not going to match
260 */
261 if (changed & (MODE_CHANGED | TYPE_CHANGED))
262 return changed;
263
264 /* Immediately after read-tree or update-index --cacheinfo,
265 * the length field is zero. For other cases the ce_size
266 * should match the SHA1 recorded in the index entry.
267 */
7a51ed66 268 if ((changed & DATA_CHANGED) && ce->ce_size != 0)
b0391890
JH
269 return changed;
270
29e4d363
JH
271 changed_fs = ce_modified_check_fs(ce, st);
272 if (changed_fs)
273 return changed | changed_fs;
b0391890
JH
274 return 0;
275}
276
958ba6c9
LT
277int base_name_compare(const char *name1, int len1, int mode1,
278 const char *name2, int len2, int mode2)
279{
280 unsigned char c1, c2;
281 int len = len1 < len2 ? len1 : len2;
282 int cmp;
283
284 cmp = memcmp(name1, name2, len);
285 if (cmp)
286 return cmp;
287 c1 = name1[len];
288 c2 = name2[len];
1833a925 289 if (!c1 && S_ISDIR(mode1))
958ba6c9 290 c1 = '/';
1833a925 291 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
292 c2 = '/';
293 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
294}
295
0ab9e1e8
LT
296/*
297 * df_name_compare() is identical to base_name_compare(), except it
298 * compares conflicting directory/file entries as equal. Note that
299 * while a directory name compares as equal to a regular file, they
300 * then individually compare _differently_ to a filename that has
301 * a dot after the basename (because '\0' < '.' < '/').
302 *
303 * This is used by routines that want to traverse the git namespace
304 * but then handle conflicting entries together when possible.
305 */
306int df_name_compare(const char *name1, int len1, int mode1,
307 const char *name2, int len2, int mode2)
308{
309 int len = len1 < len2 ? len1 : len2, cmp;
310 unsigned char c1, c2;
311
312 cmp = memcmp(name1, name2, len);
313 if (cmp)
314 return cmp;
315 /* Directories and files compare equal (same length, same name) */
316 if (len1 == len2)
317 return 0;
318 c1 = name1[len];
319 if (!c1 && S_ISDIR(mode1))
320 c1 = '/';
321 c2 = name2[len];
322 if (!c2 && S_ISDIR(mode2))
323 c2 = '/';
324 if (c1 == '/' && !c2)
325 return 0;
326 if (c2 == '/' && !c1)
327 return 0;
328 return c1 - c2;
329}
330
95fd5bf8 331int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
eb38c22f 332{
95fd5bf8
LT
333 int len1 = flags1 & CE_NAMEMASK;
334 int len2 = flags2 & CE_NAMEMASK;
eb38c22f
LT
335 int len = len1 < len2 ? len1 : len2;
336 int cmp;
337
338 cmp = memcmp(name1, name2, len);
339 if (cmp)
340 return cmp;
341 if (len1 < len2)
342 return -1;
343 if (len1 > len2)
344 return 1;
5f73076c 345
7b80be15
JH
346 /* Compare stages */
347 flags1 &= CE_STAGEMASK;
348 flags2 &= CE_STAGEMASK;
5f73076c 349
95fd5bf8
LT
350 if (flags1 < flags2)
351 return -1;
352 if (flags1 > flags2)
353 return 1;
eb38c22f
LT
354 return 0;
355}
356
d1f128b0 357int index_name_pos(const struct index_state *istate, const char *name, int namelen)
eb38c22f
LT
358{
359 int first, last;
360
361 first = 0;
4aab5b46 362 last = istate->cache_nr;
eb38c22f
LT
363 while (last > first) {
364 int next = (last + first) >> 1;
4aab5b46 365 struct cache_entry *ce = istate->cache[next];
7a51ed66 366 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
eb38c22f 367 if (!cmp)
76e7f4ec 368 return next;
eb38c22f
LT
369 if (cmp < 0) {
370 last = next;
371 continue;
372 }
373 first = next+1;
374 }
76e7f4ec 375 return -first-1;
eb38c22f
LT
376}
377
7b937ca3 378/* Remove entry, return true if there are more entries to go.. */
4aab5b46 379int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 380{
cf558704
LT
381 struct cache_entry *ce = istate->cache[pos];
382
96872bc2 383 remove_name_hash(ce);
4aab5b46
JH
384 istate->cache_changed = 1;
385 istate->cache_nr--;
386 if (pos >= istate->cache_nr)
7b937ca3 387 return 0;
4aab5b46
JH
388 memmove(istate->cache + pos,
389 istate->cache + pos + 1,
390 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
7b937ca3
LT
391 return 1;
392}
393
4aab5b46 394int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 395{
4aab5b46 396 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
397 if (pos < 0)
398 pos = -pos-1;
09d5dc32 399 cache_tree_invalidate_path(istate->cache_tree, path);
4aab5b46
JH
400 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
401 remove_index_entry_at(istate, pos);
197ee8c9
LT
402 return 0;
403}
404
20314271
JS
405static int compare_name(struct cache_entry *ce, const char *path, int namelen)
406{
407 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
408}
409
410static int index_name_pos_also_unmerged(struct index_state *istate,
411 const char *path, int namelen)
412{
413 int pos = index_name_pos(istate, path, namelen);
414 struct cache_entry *ce;
415
416 if (pos >= 0)
417 return pos;
418
419 /* maybe unmerged? */
420 pos = -1 - pos;
421 if (pos >= istate->cache_nr ||
422 compare_name((ce = istate->cache[pos]), path, namelen))
423 return -1;
424
425 /* order of preference: stage 2, 1, 3 */
426 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
427 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
428 !compare_name(ce, path, namelen))
429 pos++;
430 return pos;
431}
432
1102952b
LT
433static int different_name(struct cache_entry *ce, struct cache_entry *alias)
434{
435 int len = ce_namelen(ce);
436 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
437}
438
439/*
440 * If we add a filename that aliases in the cache, we will use the
441 * name that we already have - but we don't want to update the same
442 * alias twice, because that implies that there were actually two
443 * different files with aliasing names!
444 *
445 * So we use the CE_ADDED flag to verify that the alias was an old
446 * one before we accept it as
447 */
448static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
449{
450 int len;
451 struct cache_entry *new;
452
453 if (alias->ce_flags & CE_ADDED)
454 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
455
456 /* Ok, create the new entry using the name of the existing alias */
457 len = ce_namelen(alias);
458 new = xcalloc(1, cache_entry_size(len));
459 memcpy(new->name, alias->name, len);
460 copy_cache_entry(new, ce);
461 free(ce);
462 return new;
463}
464
4aab5b46 465int add_file_to_index(struct index_state *istate, const char *path, int verbose)
11be42a4 466{
6835550d 467 int size, namelen;
11be42a4 468 struct stat st;
6835550d 469 struct cache_entry *ce, *alias;
fb63d7f8 470 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
11be42a4
JS
471
472 if (lstat(path, &st))
473 die("%s: unable to stat (%s)", path, strerror(errno));
474
f35a6d3b
LT
475 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
476 die("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
477
478 namelen = strlen(path);
09595258
LT
479 if (S_ISDIR(st.st_mode)) {
480 while (namelen && path[namelen-1] == '/')
481 namelen--;
482 }
11be42a4
JS
483 size = cache_entry_size(namelen);
484 ce = xcalloc(1, size);
485 memcpy(ce->name, path, namelen);
7a51ed66 486 ce->ce_flags = namelen;
11be42a4
JS
487 fill_stat_cache_info(ce, &st);
488
78a8d641 489 if (trust_executable_bit && has_symlinks)
185c975f
JH
490 ce->ce_mode = create_ce_mode(st.st_mode);
491 else {
78a8d641
JS
492 /* If there is an existing entry, pick the mode bits and type
493 * from it, otherwise assume unexecutable regular file.
11be42a4 494 */
185c975f 495 struct cache_entry *ent;
20314271 496 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 497
4aab5b46 498 ent = (0 <= pos) ? istate->cache[pos] : NULL;
185c975f 499 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
11be42a4
JS
500 }
501
6835550d
LT
502 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
503 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, &st, ce_option)) {
0781b8a9
JH
504 /* Nothing changed, really */
505 free(ce);
6835550d 506 ce_mark_uptodate(alias);
1102952b 507 alias->ce_flags |= CE_ADDED;
0781b8a9
JH
508 return 0;
509 }
11be42a4
JS
510 if (index_path(ce->sha1, path, &st, 1))
511 die("unable to index file %s", path);
1102952b
LT
512 if (ignore_case && alias && different_name(ce, alias))
513 ce = create_alias_ce(ce, alias);
514 ce->ce_flags |= CE_ADDED;
4aab5b46 515 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
11be42a4
JS
516 die("unable to add %s to index",path);
517 if (verbose)
518 printf("add '%s'\n", path);
11be42a4
JS
519 return 0;
520}
521
6640f881
CR
522struct cache_entry *make_cache_entry(unsigned int mode,
523 const unsigned char *sha1, const char *path, int stage,
524 int refresh)
525{
526 int size, len;
527 struct cache_entry *ce;
528
529 if (!verify_path(path))
530 return NULL;
531
532 len = strlen(path);
533 size = cache_entry_size(len);
534 ce = xcalloc(1, size);
535
536 hashcpy(ce->sha1, sha1);
537 memcpy(ce->name, path, len);
538 ce->ce_flags = create_ce_flags(len, stage);
539 ce->ce_mode = create_ce_mode(mode);
540
541 if (refresh)
542 return refresh_cache_entry(ce, 0);
543
544 return ce;
545}
546
dbbce55b 547int ce_same_name(struct cache_entry *a, struct cache_entry *b)
7b937ca3
LT
548{
549 int len = ce_namelen(a);
550 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
551}
552
c0fd1f51
LT
553int ce_path_match(const struct cache_entry *ce, const char **pathspec)
554{
555 const char *match, *name;
556 int len;
557
558 if (!pathspec)
559 return 1;
560
561 len = ce_namelen(ce);
562 name = ce->name;
563 while ((match = *pathspec++) != NULL) {
564 int matchlen = strlen(match);
565 if (matchlen > len)
566 continue;
567 if (memcmp(name, match, matchlen))
568 continue;
569 if (matchlen && name[matchlen-1] == '/')
570 return 1;
571 if (name[matchlen] == '/' || !name[matchlen])
572 return 1;
f332726e
LT
573 if (!matchlen)
574 return 1;
c0fd1f51
LT
575 }
576 return 0;
577}
578
8dcf39c4
LT
579/*
580 * We fundamentally don't like some paths: we don't want
581 * dot or dot-dot anywhere, and for obvious reasons don't
582 * want to recurse into ".git" either.
583 *
584 * Also, we don't want double slashes or slashes at the
585 * end that can make pathnames ambiguous.
586 */
587static int verify_dotfile(const char *rest)
588{
589 /*
590 * The first character was '.', but that
591 * has already been discarded, we now test
592 * the rest.
593 */
594 switch (*rest) {
595 /* "." is not allowed */
596 case '\0': case '/':
597 return 0;
598
599 /*
600 * ".git" followed by NUL or slash is bad. This
601 * shares the path end test with the ".." case.
602 */
603 case 'g':
604 if (rest[1] != 'i')
605 break;
606 if (rest[2] != 't')
607 break;
608 rest += 2;
609 /* fallthrough */
610 case '.':
611 if (rest[1] == '\0' || rest[1] == '/')
612 return 0;
613 }
614 return 1;
615}
616
617int verify_path(const char *path)
618{
619 char c;
620
621 goto inside;
622 for (;;) {
623 if (!c)
624 return 1;
625 if (c == '/') {
626inside:
627 c = *path++;
628 switch (c) {
629 default:
630 continue;
631 case '/': case '\0':
632 break;
633 case '.':
634 if (verify_dotfile(path))
635 continue;
636 }
637 return 0;
638 }
639 c = *path++;
640 }
641}
642
12676608
LT
643/*
644 * Do we have another file that has the beginning components being a
645 * proper superset of the name we're trying to add?
0f1e4f04 646 */
4aab5b46
JH
647static int has_file_name(struct index_state *istate,
648 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 649{
12676608
LT
650 int retval = 0;
651 int len = ce_namelen(ce);
b155725d 652 int stage = ce_stage(ce);
12676608 653 const char *name = ce->name;
0f1e4f04 654
4aab5b46
JH
655 while (pos < istate->cache_nr) {
656 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 657
12676608 658 if (len >= ce_namelen(p))
0f1e4f04 659 break;
12676608
LT
660 if (memcmp(name, p->name, len))
661 break;
b155725d
JH
662 if (ce_stage(p) != stage)
663 continue;
12676608
LT
664 if (p->name[len] != '/')
665 continue;
7a51ed66 666 if (p->ce_flags & CE_REMOVE)
21cd8d00 667 continue;
12676608
LT
668 retval = -1;
669 if (!ok_to_replace)
670 break;
4aab5b46 671 remove_index_entry_at(istate, --pos);
0f1e4f04 672 }
12676608
LT
673 return retval;
674}
0f1e4f04 675
12676608
LT
676/*
677 * Do we have another file with a pathname that is a proper
678 * subset of the name we're trying to add?
679 */
4aab5b46
JH
680static int has_dir_name(struct index_state *istate,
681 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
682{
683 int retval = 0;
b155725d 684 int stage = ce_stage(ce);
12676608
LT
685 const char *name = ce->name;
686 const char *slash = name + ce_namelen(ce);
0f1e4f04 687
12676608
LT
688 for (;;) {
689 int len;
0f1e4f04 690
12676608
LT
691 for (;;) {
692 if (*--slash == '/')
693 break;
694 if (slash <= ce->name)
695 return retval;
696 }
697 len = slash - name;
0f1e4f04 698
7a51ed66 699 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
12676608 700 if (pos >= 0) {
21cd8d00
JH
701 /*
702 * Found one, but not so fast. This could
703 * be a marker that says "I was here, but
704 * I am being removed". Such an entry is
705 * not a part of the resulting tree, and
706 * it is Ok to have a directory at the same
707 * path.
708 */
077c48df 709 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
710 retval = -1;
711 if (!ok_to_replace)
712 break;
4aab5b46 713 remove_index_entry_at(istate, pos);
21cd8d00
JH
714 continue;
715 }
12676608 716 }
21cd8d00
JH
717 else
718 pos = -pos-1;
12676608
LT
719
720 /*
721 * Trivial optimization: if we find an entry that
722 * already matches the sub-directory, then we know
b155725d 723 * we're ok, and we can exit.
12676608 724 */
4aab5b46
JH
725 while (pos < istate->cache_nr) {
726 struct cache_entry *p = istate->cache[pos];
b155725d
JH
727 if ((ce_namelen(p) <= len) ||
728 (p->name[len] != '/') ||
729 memcmp(p->name, name, len))
730 break; /* not our subdirectory */
077c48df
JH
731 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
732 /*
733 * p is at the same stage as our entry, and
b155725d
JH
734 * is a subdirectory of what we are looking
735 * at, so we cannot have conflicts at our
736 * level or anything shorter.
737 */
738 return retval;
739 pos++;
192268c1 740 }
0f1e4f04 741 }
12676608
LT
742 return retval;
743}
744
745/* We may be in a situation where we already have path/file and path
746 * is being added, or we already have path and path/file is being
747 * added. Either one would result in a nonsense tree that has path
748 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 749 *
12676608
LT
750 * If ok-to-replace is specified, we remove the conflicting entries
751 * from the cache so the caller should recompute the insert position.
752 * When this happens, we return non-zero.
753 */
4aab5b46
JH
754static int check_file_directory_conflict(struct index_state *istate,
755 const struct cache_entry *ce,
756 int pos, int ok_to_replace)
12676608 757{
21cd8d00
JH
758 int retval;
759
760 /*
761 * When ce is an "I am going away" entry, we allow it to be added
762 */
7a51ed66 763 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
764 return 0;
765
12676608
LT
766 /*
767 * We check if the path is a sub-path of a subsequent pathname
768 * first, since removing those will not change the position
21cd8d00 769 * in the array.
12676608 770 */
4aab5b46 771 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 772
12676608
LT
773 /*
774 * Then check if the path might have a clashing sub-directory
775 * before it.
776 */
4aab5b46 777 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
778}
779
af3785dc 780static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
781{
782 int pos;
192268c1
JH
783 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
784 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 785 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
5f73076c 786
09d5dc32 787 cache_tree_invalidate_path(istate->cache_tree, ce->name);
7a51ed66 788 pos = index_name_pos(istate, ce->name, ce->ce_flags);
197ee8c9 789
3e09cdfd 790 /* existing match? Just replace it. */
76e7f4ec 791 if (pos >= 0) {
cf558704 792 replace_index_entry(istate, pos, ce);
197ee8c9
LT
793 return 0;
794 }
76e7f4ec 795 pos = -pos-1;
197ee8c9 796
7b937ca3
LT
797 /*
798 * Inserting a merged entry ("stage 0") into the index
799 * will always replace all non-merged entries..
800 */
4aab5b46
JH
801 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
802 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 803 ok_to_add = 1;
4aab5b46 804 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
805 break;
806 }
807 }
808
121481ab
LT
809 if (!ok_to_add)
810 return -1;
8dcf39c4
LT
811 if (!verify_path(ce->name))
812 return -1;
121481ab 813
3e09cdfd 814 if (!skip_df_check &&
4aab5b46 815 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 816 if (!ok_to_replace)
4aab5b46
JH
817 return error("'%s' appears as both a file and as a directory",
818 ce->name);
7a51ed66 819 pos = index_name_pos(istate, ce->name, ce->ce_flags);
192268c1
JH
820 pos = -pos-1;
821 }
af3785dc
JH
822 return pos + 1;
823}
824
825int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
826{
827 int pos;
828
829 if (option & ADD_CACHE_JUST_APPEND)
830 pos = istate->cache_nr;
831 else {
832 int ret;
833 ret = add_index_entry_with_check(istate, ce, option);
834 if (ret <= 0)
835 return ret;
836 pos = ret - 1;
837 }
0f1e4f04 838
197ee8c9 839 /* Make sure the array is big enough .. */
4aab5b46
JH
840 if (istate->cache_nr == istate->cache_alloc) {
841 istate->cache_alloc = alloc_nr(istate->cache_alloc);
842 istate->cache = xrealloc(istate->cache,
843 istate->cache_alloc * sizeof(struct cache_entry *));
197ee8c9
LT
844 }
845
846 /* Add it in.. */
4aab5b46 847 istate->cache_nr++;
af3785dc 848 if (istate->cache_nr > pos + 1)
4aab5b46
JH
849 memmove(istate->cache + pos + 1,
850 istate->cache + pos,
851 (istate->cache_nr - pos - 1) * sizeof(ce));
cf558704 852 set_index_entry(istate, pos, ce);
4aab5b46 853 istate->cache_changed = 1;
197ee8c9
LT
854 return 0;
855}
856
405e5b2f
LT
857/*
858 * "refresh" does not calculate a new sha1 file or bring the
859 * cache up-to-date for mode/content changes. But what it
860 * _does_ do is to "re-match" the stat information of a file
861 * with the cache, so that you can refresh the cache for a
862 * file that hasn't been changed but where the stat entry is
863 * out of date.
864 *
865 * For example, you'd want to do this after doing a "git-read-tree",
866 * to link up the stat cache details with the proper files.
867 */
4aab5b46 868static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da
JH
869 struct cache_entry *ce,
870 unsigned int options, int *err)
405e5b2f
LT
871{
872 struct stat st;
873 struct cache_entry *updated;
874 int changed, size;
4bd5b7da 875 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
405e5b2f 876
eadb5831
JH
877 if (ce_uptodate(ce))
878 return ce;
879
8fd2cb40 880 if (lstat(ce->name, &st) < 0) {
ec0cc704
JH
881 if (err)
882 *err = errno;
8fd2cb40
JS
883 return NULL;
884 }
405e5b2f 885
4bd5b7da 886 changed = ie_match_stat(istate, ce, &st, options);
405e5b2f 887 if (!changed) {
4bd5b7da
JH
888 /*
889 * The path is unchanged. If we were told to ignore
890 * valid bit, then we did the actual stat check and
891 * found that the entry is unmodified. If the entry
892 * is not marked VALID, this is the place to mark it
893 * valid again, under "assume unchanged" mode.
894 */
895 if (ignore_valid && assume_unchanged &&
7a51ed66 896 !(ce->ce_flags & CE_VALID))
405e5b2f 897 ; /* mark this one VALID again */
eadb5831
JH
898 else {
899 /*
900 * We do not mark the index itself "modified"
901 * because CE_UPTODATE flag is in-core only;
902 * we are not going to write this change out.
903 */
904 ce_mark_uptodate(ce);
8fd2cb40 905 return ce;
eadb5831 906 }
405e5b2f
LT
907 }
908
4bd5b7da 909 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
910 if (err)
911 *err = EINVAL;
8fd2cb40
JS
912 return NULL;
913 }
405e5b2f
LT
914
915 size = ce_size(ce);
916 updated = xmalloc(size);
917 memcpy(updated, ce, size);
918 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
919 /*
920 * If ignore_valid is not set, we should leave CE_VALID bit
921 * alone. Otherwise, paths marked with --no-assume-unchanged
922 * (i.e. things to be edited) will reacquire CE_VALID bit
923 * automatically, which is not really what we want.
405e5b2f 924 */
4bd5b7da 925 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
926 !(ce->ce_flags & CE_VALID))
927 updated->ce_flags &= ~CE_VALID;
405e5b2f
LT
928
929 return updated;
930}
931
d616813d 932int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
405e5b2f
LT
933{
934 int i;
935 int has_errors = 0;
936 int really = (flags & REFRESH_REALLY) != 0;
937 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
938 int quiet = (flags & REFRESH_QUIET) != 0;
939 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
4bd5b7da 940 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
405e5b2f 941
4aab5b46 942 for (i = 0; i < istate->cache_nr; i++) {
405e5b2f 943 struct cache_entry *ce, *new;
ec0cc704
JH
944 int cache_errno = 0;
945
4aab5b46 946 ce = istate->cache[i];
405e5b2f 947 if (ce_stage(ce)) {
4aab5b46
JH
948 while ((i < istate->cache_nr) &&
949 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
950 i++;
951 i--;
952 if (allow_unmerged)
953 continue;
954 printf("%s: needs merge\n", ce->name);
955 has_errors = 1;
956 continue;
957 }
958
d616813d
AJ
959 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
960 continue;
961
4bd5b7da 962 new = refresh_cache_ent(istate, ce, options, &cache_errno);
8fd2cb40 963 if (new == ce)
405e5b2f 964 continue;
8fd2cb40
JS
965 if (!new) {
966 if (not_new && cache_errno == ENOENT)
405e5b2f 967 continue;
8fd2cb40 968 if (really && cache_errno == EINVAL) {
405e5b2f
LT
969 /* If we are doing --really-refresh that
970 * means the index is not valid anymore.
971 */
7a51ed66 972 ce->ce_flags &= ~CE_VALID;
4aab5b46 973 istate->cache_changed = 1;
405e5b2f
LT
974 }
975 if (quiet)
976 continue;
977 printf("%s: needs update\n", ce->name);
978 has_errors = 1;
979 continue;
980 }
cf558704
LT
981
982 replace_index_entry(istate, i, new);
405e5b2f
LT
983 }
984 return has_errors;
985}
986
ec0cc704
JH
987struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
988{
4aab5b46 989 return refresh_cache_ent(&the_index, ce, really, NULL);
ec0cc704
JH
990}
991
bad68ec9 992static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163
LT
993{
994 SHA_CTX c;
bad68ec9 995 unsigned char sha1[20];
e83c5163 996
ccc4feb5 997 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 998 return error("bad signature");
ca9be054
LT
999 if (hdr->hdr_version != htonl(2))
1000 return error("bad index version");
e83c5163 1001 SHA1_Init(&c);
ca9be054 1002 SHA1_Update(&c, hdr, size - 20);
e83c5163 1003 SHA1_Final(sha1, &c);
a89fccd2 1004 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
ca9be054 1005 return error("bad index file sha1 signature");
e83c5163
LT
1006 return 0;
1007}
1008
4aab5b46
JH
1009static int read_index_extension(struct index_state *istate,
1010 const char *ext, void *data, unsigned long sz)
bad68ec9
JH
1011{
1012 switch (CACHE_EXT(ext)) {
1013 case CACHE_EXT_TREE:
4aab5b46 1014 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9
JH
1015 break;
1016 default:
1017 if (*ext < 'A' || 'Z' < *ext)
1018 return error("index uses %.4s extension, which we do not understand",
1019 ext);
1020 fprintf(stderr, "ignoring %.4s extension\n", ext);
1021 break;
1022 }
1023 return 0;
1024}
1025
4aab5b46 1026int read_index(struct index_state *istate)
8fd2cb40 1027{
4aab5b46 1028 return read_index_from(istate, get_index_file());
8fd2cb40
JS
1029}
1030
7a51ed66
LT
1031static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1032{
7fec10b7
JH
1033 size_t len;
1034
7a51ed66
LT
1035 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1036 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1037 ce->ce_dev = ntohl(ondisk->dev);
1038 ce->ce_ino = ntohl(ondisk->ino);
1039 ce->ce_mode = ntohl(ondisk->mode);
1040 ce->ce_uid = ntohl(ondisk->uid);
1041 ce->ce_gid = ntohl(ondisk->gid);
1042 ce->ce_size = ntohl(ondisk->size);
1043 /* On-disk flags are just 16 bits */
1044 ce->ce_flags = ntohs(ondisk->flags);
1045 hashcpy(ce->sha1, ondisk->sha1);
7fec10b7
JH
1046
1047 len = ce->ce_flags & CE_NAMEMASK;
1048 if (len == CE_NAMEMASK)
1049 len = strlen(ondisk->name);
1050 /*
1051 * NEEDSWORK: If the original index is crafted, this copy could
1052 * go unchecked.
1053 */
1054 memcpy(ce->name, ondisk->name, len + 1);
7a51ed66
LT
1055}
1056
cf558704
LT
1057static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1058{
1059 long per_entry;
1060
1061 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1062
1063 /*
1064 * Alignment can cause differences. This should be "alignof", but
1065 * since that's a gcc'ism, just use the size of a pointer.
1066 */
1067 per_entry += sizeof(void *);
1068 return ondisk_size + entries*per_entry;
1069}
1070
8fd2cb40 1071/* remember to discard_cache() before reading a different cache! */
4aab5b46 1072int read_index_from(struct index_state *istate, const char *path)
e83c5163
LT
1073{
1074 int fd, i;
1075 struct stat st;
7a51ed66 1076 unsigned long src_offset, dst_offset;
e83c5163 1077 struct cache_header *hdr;
7a51ed66
LT
1078 void *mmap;
1079 size_t mmap_size;
e83c5163
LT
1080
1081 errno = EBUSY;
7a51ed66 1082 if (istate->alloc)
4aab5b46 1083 return istate->cache_nr;
5d1a5c02 1084
e83c5163 1085 errno = ENOENT;
4aab5b46 1086 istate->timestamp = 0;
8fd2cb40 1087 fd = open(path, O_RDONLY);
5d1a5c02
LT
1088 if (fd < 0) {
1089 if (errno == ENOENT)
1090 return 0;
1091 die("index file open failed (%s)", strerror(errno));
1092 }
e83c5163 1093
3511a377 1094 if (fstat(fd, &st))
5fe5c830 1095 die("cannot stat the open index (%s)", strerror(errno));
3511a377
LFC
1096
1097 errno = EINVAL;
7a51ed66
LT
1098 mmap_size = xsize_t(st.st_size);
1099 if (mmap_size < sizeof(struct cache_header) + 20)
3511a377
LFC
1100 die("index file smaller than expected");
1101
7a51ed66 1102 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
e83c5163 1103 close(fd);
7a51ed66
LT
1104 if (mmap == MAP_FAILED)
1105 die("unable to map index file");
e83c5163 1106
7a51ed66
LT
1107 hdr = mmap;
1108 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1109 goto unmap;
1110
4aab5b46
JH
1111 istate->cache_nr = ntohl(hdr->hdr_entries);
1112 istate->cache_alloc = alloc_nr(istate->cache_nr);
1113 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
e83c5163 1114
7a51ed66
LT
1115 /*
1116 * The disk format is actually larger than the in-memory format,
1117 * due to space for nsec etc, so even though the in-memory one
1118 * has room for a few more flags, we can allocate using the same
1119 * index size
1120 */
cf558704 1121 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
7a51ed66
LT
1122
1123 src_offset = sizeof(*hdr);
1124 dst_offset = 0;
4aab5b46 1125 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 1126 struct ondisk_cache_entry *disk_ce;
4aab5b46
JH
1127 struct cache_entry *ce;
1128
7a51ed66
LT
1129 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1130 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1131 convert_from_disk(disk_ce, ce);
cf558704 1132 set_index_entry(istate, i, ce);
7a51ed66
LT
1133
1134 src_offset += ondisk_ce_size(ce);
1135 dst_offset += ce_size(ce);
e83c5163 1136 }
4aab5b46 1137 istate->timestamp = st.st_mtime;
7a51ed66 1138 while (src_offset <= mmap_size - 20 - 8) {
bad68ec9
JH
1139 /* After an array of active_nr index entries,
1140 * there can be arbitrary number of extended
1141 * sections, each of which is prefixed with
1142 * extension name (4-byte) and section length
1143 * in 4-byte network byte order.
1144 */
1145 unsigned long extsize;
7a51ed66 1146 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
bad68ec9 1147 extsize = ntohl(extsize);
4aab5b46 1148 if (read_index_extension(istate,
7a51ed66
LT
1149 (const char *) mmap + src_offset,
1150 (char *) mmap + src_offset + 8,
1d7f171c 1151 extsize) < 0)
bad68ec9 1152 goto unmap;
7a51ed66
LT
1153 src_offset += 8;
1154 src_offset += extsize;
bad68ec9 1155 }
7a51ed66 1156 munmap(mmap, mmap_size);
4aab5b46 1157 return istate->cache_nr;
e83c5163
LT
1158
1159unmap:
7a51ed66 1160 munmap(mmap, mmap_size);
e83c5163 1161 errno = EINVAL;
5d1a5c02 1162 die("index file corrupt");
e83c5163
LT
1163}
1164
4aab5b46 1165int discard_index(struct index_state *istate)
6d297f81 1166{
4aab5b46
JH
1167 istate->cache_nr = 0;
1168 istate->cache_changed = 0;
1169 istate->timestamp = 0;
cf558704 1170 free_hash(&istate->name_hash);
4aab5b46 1171 cache_tree_free(&(istate->cache_tree));
7a51ed66
LT
1172 free(istate->alloc);
1173 istate->alloc = NULL;
6d297f81
JS
1174
1175 /* no need to throw away allocated active_cache */
7a51ed66 1176 return 0;
6d297f81
JS
1177}
1178
d1f128b0 1179int unmerged_index(const struct index_state *istate)
94a5728c
DB
1180{
1181 int i;
1182 for (i = 0; i < istate->cache_nr; i++) {
1183 if (ce_stage(istate->cache[i]))
1184 return 1;
1185 }
1186 return 0;
1187}
1188
4990aadc 1189#define WRITE_BUFFER_SIZE 8192
bf0f910d 1190static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
1191static unsigned long write_buffer_len;
1192
6015c28b
JH
1193static int ce_write_flush(SHA_CTX *context, int fd)
1194{
1195 unsigned int buffered = write_buffer_len;
1196 if (buffered) {
1197 SHA1_Update(context, write_buffer, buffered);
93822c22 1198 if (write_in_full(fd, write_buffer, buffered) != buffered)
6015c28b
JH
1199 return -1;
1200 write_buffer_len = 0;
1201 }
1202 return 0;
1203}
1204
ca9be054 1205static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
1206{
1207 while (len) {
1208 unsigned int buffered = write_buffer_len;
1209 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1210 if (partial > len)
1211 partial = len;
1212 memcpy(write_buffer + buffered, data, partial);
1213 buffered += partial;
1214 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
1215 write_buffer_len = buffered;
1216 if (ce_write_flush(context, fd))
4990aadc
LT
1217 return -1;
1218 buffered = 0;
1219 }
1220 write_buffer_len = buffered;
1221 len -= partial;
1d7f171c 1222 data = (char *) data + partial;
a6080a0a
JH
1223 }
1224 return 0;
4990aadc
LT
1225}
1226
bad68ec9 1227static int write_index_ext_header(SHA_CTX *context, int fd,
ac58c7b1 1228 unsigned int ext, unsigned int sz)
bad68ec9
JH
1229{
1230 ext = htonl(ext);
1231 sz = htonl(sz);
968a1d65
DR
1232 return ((ce_write(context, fd, &ext, 4) < 0) ||
1233 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
1234}
1235
1236static int ce_flush(SHA_CTX *context, int fd)
4990aadc
LT
1237{
1238 unsigned int left = write_buffer_len;
ca9be054 1239
4990aadc
LT
1240 if (left) {
1241 write_buffer_len = 0;
ca9be054 1242 SHA1_Update(context, write_buffer, left);
4990aadc 1243 }
ca9be054 1244
2c865d9a
QH
1245 /* Flush first if not enough space for SHA1 signature */
1246 if (left + 20 > WRITE_BUFFER_SIZE) {
93822c22 1247 if (write_in_full(fd, write_buffer, left) != left)
2c865d9a
QH
1248 return -1;
1249 left = 0;
1250 }
1251
ca9be054 1252 /* Append the SHA1 signature at the end */
bad68ec9 1253 SHA1_Final(write_buffer + left, context);
ca9be054 1254 left += 20;
93822c22 1255 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
4990aadc
LT
1256}
1257
407c8eb0
JH
1258static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1259{
1260 /*
1261 * The only thing we care about in this function is to smudge the
1262 * falsely clean entry due to touch-update-touch race, so we leave
1263 * everything else as they are. We are called for entries whose
1264 * ce_mtime match the index file mtime.
1265 */
1266 struct stat st;
1267
1268 if (lstat(ce->name, &st) < 0)
1269 return;
1270 if (ce_match_stat_basic(ce, &st))
1271 return;
1272 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
1273 /* This is "racily clean"; smudge it. Note that this
1274 * is a tricky code. At first glance, it may appear
1275 * that it can break with this sequence:
1276 *
1277 * $ echo xyzzy >frotz
1278 * $ git-update-index --add frotz
1279 * $ : >frotz
1280 * $ sleep 3
1281 * $ echo filfre >nitfol
1282 * $ git-update-index --add nitfol
1283 *
b7e58b17 1284 * but it does not. When the second update-index runs,
4b3511b0
JH
1285 * it notices that the entry "frotz" has the same timestamp
1286 * as index, and if we were to smudge it by resetting its
1287 * size to zero here, then the object name recorded
1288 * in index is the 6-byte file but the cached stat information
1289 * becomes zero --- which would then match what we would
a6080a0a 1290 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
1291 *
1292 * However, the second update-index, before calling
1293 * this function, notices that the cached size is 6
1294 * bytes and what is on the filesystem is an empty
1295 * file, and never calls us, so the cached size information
1296 * for "frotz" stays 6 which does not match the filesystem.
1297 */
7a51ed66 1298 ce->ce_size = 0;
407c8eb0
JH
1299 }
1300}
1301
7a51ed66
LT
1302static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1303{
1304 int size = ondisk_ce_size(ce);
1305 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1306
1307 ondisk->ctime.sec = htonl(ce->ce_ctime);
1308 ondisk->ctime.nsec = 0;
1309 ondisk->mtime.sec = htonl(ce->ce_mtime);
1310 ondisk->mtime.nsec = 0;
1311 ondisk->dev = htonl(ce->ce_dev);
1312 ondisk->ino = htonl(ce->ce_ino);
1313 ondisk->mode = htonl(ce->ce_mode);
1314 ondisk->uid = htonl(ce->ce_uid);
1315 ondisk->gid = htonl(ce->ce_gid);
1316 ondisk->size = htonl(ce->ce_size);
1317 hashcpy(ondisk->sha1, ce->sha1);
1318 ondisk->flags = htons(ce->ce_flags);
1319 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1320
1321 return ce_write(c, fd, ondisk, size);
1322}
1323
d1f128b0 1324int write_index(const struct index_state *istate, int newfd)
197ee8c9
LT
1325{
1326 SHA_CTX c;
1327 struct cache_header hdr;
1dffb8fa 1328 int i, err, removed;
4aab5b46
JH
1329 struct cache_entry **cache = istate->cache;
1330 int entries = istate->cache_nr;
025a0709
JH
1331
1332 for (i = removed = 0; i < entries; i++)
7a51ed66 1333 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 1334 removed++;
197ee8c9 1335
ccc4feb5 1336 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
ca9be054 1337 hdr.hdr_version = htonl(2);
025a0709 1338 hdr.hdr_entries = htonl(entries - removed);
197ee8c9
LT
1339
1340 SHA1_Init(&c);
ca9be054 1341 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
1342 return -1;
1343
1344 for (i = 0; i < entries; i++) {
1345 struct cache_entry *ce = cache[i];
7a51ed66 1346 if (ce->ce_flags & CE_REMOVE)
aa16021e 1347 continue;
e06c43c7 1348 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 1349 ce_smudge_racily_clean_entry(ce);
7a51ed66 1350 if (ce_write_entry(&c, newfd, ce) < 0)
197ee8c9
LT
1351 return -1;
1352 }
1af1c2b6 1353
bad68ec9 1354 /* Write extension data here */
4aab5b46 1355 if (istate->cache_tree) {
1dffb8fa
PH
1356 struct strbuf sb;
1357
1358 strbuf_init(&sb, 0);
1359 cache_tree_write(&sb, istate->cache_tree);
1360 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1361 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1362 strbuf_release(&sb);
1363 if (err)
bad68ec9 1364 return -1;
bad68ec9
JH
1365 }
1366 return ce_flush(&c, newfd);
197ee8c9 1367}