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