read-cache: speed up has_dir_name (part 1)
[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"
f6ecc62d 8#include "tempfile.h"
697cc8ef 9#include "lockfile.h"
bad68ec9 10#include "cache-tree.h"
f35a6d3b 11#include "refs.h"
d616813d 12#include "dir.h"
041aee31
JH
13#include "tree.h"
14#include "commit.h"
39425819 15#include "blob.h"
cfc5789a 16#include "resolve-undo.h"
6c9cd161
JH
17#include "strbuf.h"
18#include "varint.h"
5fc2fc8f 19#include "split-index.h"
a42643aa 20#include "utf8.h"
bad68ec9 21
b60e188c
TG
22/* Mask for the name length in ce_flags in the on-disk index */
23
24#define CE_NAMEMASK (0x0fff)
25
bad68ec9
JH
26/* Index extensions.
27 *
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
33 */
34
35#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
b659b49b 37#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
5fc2fc8f 38#define CACHE_EXT_LINK 0x6c696e6b /* "link" */
83c094ad 39#define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
5fc2fc8f
NTND
40
41/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
e0cf0d7d 42#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
c18b80a0 43 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
1bbb3dba 44 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
e83c5163 45
228e94f9 46struct index_state the_index;
626f35c8 47static const char *alternate_index_output;
8fd2cb40 48
9cb76b8c
JH
49static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
50{
51 istate->cache[nr] = ce;
96872bc2 52 add_name_hash(istate, ce);
9cb76b8c
JH
53}
54
cf558704
LT
55static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
56{
57 struct cache_entry *old = istate->cache[nr];
58
078a58e8 59 replace_index_entry_in_base(istate, old, ce);
2092678c 60 remove_name_hash(istate, old);
5699d17e 61 free(old);
a22c6371 62 set_index_entry(istate, nr, ce);
078a58e8 63 ce->ce_flags |= CE_UPDATE_IN_BASE;
e636a7b4 64 istate->cache_changed |= CE_ENTRY_CHANGED;
cf558704
LT
65}
66
81dc2307
PB
67void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
68{
69 struct cache_entry *old = istate->cache[nr], *new;
70 int namelen = strlen(new_name);
71
72 new = xmalloc(cache_entry_size(namelen));
73 copy_cache_entry(new, old);
419a597f 74 new->ce_flags &= ~CE_HASHED;
b60e188c 75 new->ce_namelen = namelen;
5fc2fc8f 76 new->index = 0;
81dc2307
PB
77 memcpy(new->name, new_name, namelen + 1);
78
a5400efe 79 cache_tree_invalidate_path(istate, old->name);
e931371a 80 untracked_cache_remove_from_index(istate, old->name);
81dc2307
PB
81 remove_index_entry_at(istate, nr);
82 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
83}
84
c21d39d7
MH
85void fill_stat_data(struct stat_data *sd, struct stat *st)
86{
87 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
88 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
89 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
90 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
91 sd->sd_dev = st->st_dev;
92 sd->sd_ino = st->st_ino;
93 sd->sd_uid = st->st_uid;
94 sd->sd_gid = st->st_gid;
95 sd->sd_size = st->st_size;
96}
97
98int match_stat_data(const struct stat_data *sd, struct stat *st)
99{
100 int changed = 0;
101
102 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
103 changed |= MTIME_CHANGED;
104 if (trust_ctime && check_stat &&
105 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
106 changed |= CTIME_CHANGED;
107
108#ifdef USE_NSEC
109 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
110 changed |= MTIME_CHANGED;
111 if (trust_ctime && check_stat &&
112 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
113 changed |= CTIME_CHANGED;
114#endif
115
116 if (check_stat) {
117 if (sd->sd_uid != (unsigned int) st->st_uid ||
118 sd->sd_gid != (unsigned int) st->st_gid)
119 changed |= OWNER_CHANGED;
120 if (sd->sd_ino != (unsigned int) st->st_ino)
121 changed |= INODE_CHANGED;
122 }
123
124#ifdef USE_STDEV
125 /*
126 * st_dev breaks on network filesystems where different
127 * clients will have different views of what "device"
128 * the filesystem is on
129 */
130 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
131 changed |= INODE_CHANGED;
132#endif
133
134 if (sd->sd_size != (unsigned int) st->st_size)
135 changed |= DATA_CHANGED;
136
137 return changed;
138}
139
415e96c8
JH
140/*
141 * This only updates the "non-critical" parts of the directory
142 * cache, ie the parts that aren't tracked by GIT, and only used
143 * to validate the cache.
144 */
145void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
146{
c21d39d7 147 fill_stat_data(&ce->ce_stat_data, st);
5f73076c
JH
148
149 if (assume_unchanged)
7a51ed66 150 ce->ce_flags |= CE_VALID;
eadb5831
JH
151
152 if (S_ISREG(st->st_mode))
153 ce_mark_uptodate(ce);
415e96c8
JH
154}
155
21a6b9fa 156static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
157{
158 int match = -1;
1b8ac5ea 159 int fd = git_open_cloexec(ce->name, O_RDONLY);
29e4d363
JH
160
161 if (fd >= 0) {
162 unsigned char sha1[20];
c4ce46fc 163 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
99d1a986 164 match = hashcmp(sha1, ce->oid.hash);
7f8508e8 165 /* index_fd() closed the file descriptor already */
29e4d363
JH
166 }
167 return match;
168}
169
21a6b9fa 170static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
29e4d363
JH
171{
172 int match = -1;
29e4d363
JH
173 void *buffer;
174 unsigned long size;
21666f1a 175 enum object_type type;
a60272b3 176 struct strbuf sb = STRBUF_INIT;
29e4d363 177
a60272b3 178 if (strbuf_readlink(&sb, ce->name, expected_size))
29e4d363 179 return -1;
a60272b3 180
99d1a986 181 buffer = read_sha1_file(ce->oid.hash, &type, &size);
a60272b3
LT
182 if (buffer) {
183 if (size == sb.len)
184 match = memcmp(buffer, sb.buf, size);
185 free(buffer);
29e4d363 186 }
a60272b3 187 strbuf_release(&sb);
29e4d363
JH
188 return match;
189}
190
21a6b9fa 191static int ce_compare_gitlink(const struct cache_entry *ce)
f35a6d3b
LT
192{
193 unsigned char sha1[20];
194
195 /*
196 * We don't actually require that the .git directory
302b9282 197 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
198 * might even be missing (in case nobody populated that
199 * sub-project).
200 *
201 * If so, we consider it always to match.
202 */
203 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
204 return 0;
99d1a986 205 return hashcmp(sha1, ce->oid.hash);
f35a6d3b
LT
206}
207
21a6b9fa 208static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
209{
210 switch (st->st_mode & S_IFMT) {
211 case S_IFREG:
212 if (ce_compare_data(ce, st))
213 return DATA_CHANGED;
214 break;
215 case S_IFLNK:
dc49cd76 216 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
217 return DATA_CHANGED;
218 break;
a8ee75bc 219 case S_IFDIR:
7a51ed66 220 if (S_ISGITLINK(ce->ce_mode))
c70115b4 221 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
29e4d363
JH
222 default:
223 return TYPE_CHANGED;
224 }
225 return 0;
226}
227
21a6b9fa 228static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
734aab75
LT
229{
230 unsigned int changed = 0;
231
7a51ed66
LT
232 if (ce->ce_flags & CE_REMOVE)
233 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
234
235 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
236 case S_IFREG:
237 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
238 /* We consider only the owner x bit to be relevant for
239 * "mode changes"
240 */
241 if (trust_executable_bit &&
7a51ed66 242 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 243 changed |= MODE_CHANGED;
8ae0a8c5
KS
244 break;
245 case S_IFLNK:
78a8d641
JS
246 if (!S_ISLNK(st->st_mode) &&
247 (has_symlinks || !S_ISREG(st->st_mode)))
248 changed |= TYPE_CHANGED;
8ae0a8c5 249 break;
302b9282 250 case S_IFGITLINK:
c70115b4 251 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
252 if (!S_ISDIR(st->st_mode))
253 changed |= TYPE_CHANGED;
254 else if (ce_compare_gitlink(ce))
255 changed |= DATA_CHANGED;
a8ee75bc 256 return changed;
8ae0a8c5 257 default:
7a51ed66 258 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 259 }
2cb45e95 260
c21d39d7 261 changed |= match_stat_data(&ce->ce_stat_data, st);
b0391890 262
f49c2c22 263 /* Racily smudged entry? */
c21d39d7 264 if (!ce->ce_stat_data.sd_size) {
99d1a986 265 if (!is_empty_blob_sha1(ce->oid.hash))
f49c2c22
LT
266 changed |= DATA_CHANGED;
267 }
268
407c8eb0
JH
269 return changed;
270}
271
2bb4cda1
NTND
272static int is_racy_stat(const struct index_state *istate,
273 const struct stat_data *sd)
6d91da6d 274{
2bb4cda1 275 return (istate->timestamp.sec &&
fba2f38a
KB
276#ifdef USE_NSEC
277 /* nanosecond timestamped files can also be racy! */
2bb4cda1
NTND
278 (istate->timestamp.sec < sd->sd_mtime.sec ||
279 (istate->timestamp.sec == sd->sd_mtime.sec &&
280 istate->timestamp.nsec <= sd->sd_mtime.nsec))
fba2f38a 281#else
2bb4cda1 282 istate->timestamp.sec <= sd->sd_mtime.sec
fba2f38a 283#endif
2bb4cda1
NTND
284 );
285}
286
287static int is_racy_timestamp(const struct index_state *istate,
288 const struct cache_entry *ce)
289{
290 return (!S_ISGITLINK(ce->ce_mode) &&
291 is_racy_stat(istate, &ce->ce_stat_data));
6d91da6d
JH
292}
293
ed4efab1
NTND
294int match_stat_data_racy(const struct index_state *istate,
295 const struct stat_data *sd, struct stat *st)
296{
297 if (is_racy_stat(istate, sd))
298 return MTIME_CHANGED;
299 return match_stat_data(sd, st);
6d91da6d
JH
300}
301
d1f128b0 302int ie_match_stat(const struct index_state *istate,
21a6b9fa 303 const struct cache_entry *ce, struct stat *st,
4bd5b7da 304 unsigned int options)
407c8eb0 305{
5f73076c 306 unsigned int changed;
4bd5b7da 307 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 308 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
4bd5b7da 309 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
5f73076c
JH
310
311 /*
312 * If it's marked as always valid in the index, it's
313 * valid whatever the checked-out copy says.
56cac48c
NTND
314 *
315 * skip-worktree has the same effect with higher precedence
5f73076c 316 */
56cac48c
NTND
317 if (!ignore_skip_worktree && ce_skip_worktree(ce))
318 return 0;
7a51ed66 319 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c
JH
320 return 0;
321
331fcb59
JH
322 /*
323 * Intent-to-add entries have not been added, so the index entry
324 * by definition never matches what is in the work tree until it
325 * actually gets added.
326 */
895ff3b2 327 if (ce_intent_to_add(ce))
331fcb59
JH
328 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
329
5f73076c 330 changed = ce_match_stat_basic(ce, st);
407c8eb0 331
29e4d363
JH
332 /*
333 * Within 1 second of this sequence:
334 * echo xyzzy >file && git-update-index --add file
335 * running this command:
336 * echo frotz >file
337 * would give a falsely clean cache entry. The mtime and
338 * length match the cache, and other stat fields do not change.
339 *
340 * We could detect this at update-index time (the cache entry
341 * being registered/updated records the same time as "now")
342 * and delay the return from git-update-index, but that would
343 * effectively mean we can make at most one commit per second,
344 * which is not acceptable. Instead, we check cache entries
345 * whose mtime are the same as the index file timestamp more
5f73076c 346 * carefully than others.
29e4d363 347 */
6d91da6d 348 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
349 if (assume_racy_is_modified)
350 changed |= DATA_CHANGED;
351 else
352 changed |= ce_modified_check_fs(ce, st);
353 }
b0391890 354
29e4d363 355 return changed;
b0391890
JH
356}
357
d1f128b0 358int ie_modified(const struct index_state *istate,
21a6b9fa
RS
359 const struct cache_entry *ce,
360 struct stat *st, unsigned int options)
b0391890 361{
29e4d363 362 int changed, changed_fs;
4bd5b7da
JH
363
364 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
365 if (!changed)
366 return 0;
b0391890
JH
367 /*
368 * If the mode or type has changed, there's no point in trying
369 * to refresh the entry - it's not going to match
370 */
371 if (changed & (MODE_CHANGED | TYPE_CHANGED))
372 return changed;
373
c70115b4
JH
374 /*
375 * Immediately after read-tree or update-index --cacheinfo,
376 * the length field is zero, as we have never even read the
377 * lstat(2) information once, and we cannot trust DATA_CHANGED
378 * returned by ie_match_stat() which in turn was returned by
379 * ce_match_stat_basic() to signal that the filesize of the
380 * blob changed. We have to actually go to the filesystem to
381 * see if the contents match, and if so, should answer "unchanged".
382 *
383 * The logic does not apply to gitlinks, as ce_match_stat_basic()
384 * already has checked the actual HEAD from the filesystem in the
385 * subproject. If ie_match_stat() already said it is different,
386 * then we know it is.
b0391890 387 */
c70115b4 388 if ((changed & DATA_CHANGED) &&
c21d39d7 389 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
b0391890
JH
390 return changed;
391
29e4d363
JH
392 changed_fs = ce_modified_check_fs(ce, st);
393 if (changed_fs)
394 return changed | changed_fs;
b0391890
JH
395 return 0;
396}
397
958ba6c9
LT
398int base_name_compare(const char *name1, int len1, int mode1,
399 const char *name2, int len2, int mode2)
400{
401 unsigned char c1, c2;
402 int len = len1 < len2 ? len1 : len2;
403 int cmp;
404
405 cmp = memcmp(name1, name2, len);
406 if (cmp)
407 return cmp;
408 c1 = name1[len];
409 c2 = name2[len];
1833a925 410 if (!c1 && S_ISDIR(mode1))
958ba6c9 411 c1 = '/';
1833a925 412 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
413 c2 = '/';
414 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
415}
416
0ab9e1e8
LT
417/*
418 * df_name_compare() is identical to base_name_compare(), except it
419 * compares conflicting directory/file entries as equal. Note that
420 * while a directory name compares as equal to a regular file, they
421 * then individually compare _differently_ to a filename that has
422 * a dot after the basename (because '\0' < '.' < '/').
423 *
424 * This is used by routines that want to traverse the git namespace
425 * but then handle conflicting entries together when possible.
426 */
427int df_name_compare(const char *name1, int len1, int mode1,
428 const char *name2, int len2, int mode2)
429{
430 int len = len1 < len2 ? len1 : len2, cmp;
431 unsigned char c1, c2;
432
433 cmp = memcmp(name1, name2, len);
434 if (cmp)
435 return cmp;
436 /* Directories and files compare equal (same length, same name) */
437 if (len1 == len2)
438 return 0;
439 c1 = name1[len];
440 if (!c1 && S_ISDIR(mode1))
441 c1 = '/';
442 c2 = name2[len];
443 if (!c2 && S_ISDIR(mode2))
444 c2 = '/';
445 if (c1 == '/' && !c2)
446 return 0;
447 if (c2 == '/' && !c1)
448 return 0;
449 return c1 - c2;
450}
451
ccdd4a0f 452int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
eb38c22f 453{
ccdd4a0f
JM
454 size_t min_len = (len1 < len2) ? len1 : len2;
455 int cmp = memcmp(name1, name2, min_len);
eb38c22f
LT
456 if (cmp)
457 return cmp;
458 if (len1 < len2)
459 return -1;
460 if (len1 > len2)
461 return 1;
ccdd4a0f
JM
462 return 0;
463}
464
465int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
466{
467 int cmp;
468
469 cmp = name_compare(name1, len1, name2, len2);
470 if (cmp)
471 return cmp;
5f73076c 472
b60e188c 473 if (stage1 < stage2)
95fd5bf8 474 return -1;
b60e188c 475 if (stage1 > stage2)
95fd5bf8 476 return 1;
eb38c22f
LT
477 return 0;
478}
479
357e9c69 480static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
eb38c22f
LT
481{
482 int first, last;
483
484 first = 0;
4aab5b46 485 last = istate->cache_nr;
eb38c22f
LT
486 while (last > first) {
487 int next = (last + first) >> 1;
4aab5b46 488 struct cache_entry *ce = istate->cache[next];
b60e188c 489 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
eb38c22f 490 if (!cmp)
76e7f4ec 491 return next;
eb38c22f
LT
492 if (cmp < 0) {
493 last = next;
494 continue;
495 }
496 first = next+1;
497 }
76e7f4ec 498 return -first-1;
eb38c22f
LT
499}
500
b60e188c
TG
501int index_name_pos(const struct index_state *istate, const char *name, int namelen)
502{
503 return index_name_stage_pos(istate, name, namelen, 0);
504}
505
4aab5b46 506int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 507{
cf558704
LT
508 struct cache_entry *ce = istate->cache[pos];
509
cfc5789a 510 record_resolve_undo(istate, ce);
2092678c 511 remove_name_hash(istate, ce);
045113a5 512 save_or_free_index_entry(istate, ce);
e636a7b4 513 istate->cache_changed |= CE_ENTRY_REMOVED;
4aab5b46
JH
514 istate->cache_nr--;
515 if (pos >= istate->cache_nr)
7b937ca3 516 return 0;
4aab5b46
JH
517 memmove(istate->cache + pos,
518 istate->cache + pos + 1,
519 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
7b937ca3
LT
520 return 1;
521}
522
36419c8e 523/*
98e023de 524 * Remove all cache entries marked for removal, that is where
36419c8e
KB
525 * CE_REMOVE is set in ce_flags. This is much more effective than
526 * calling remove_index_entry_at() for each entry to be removed.
527 */
528void remove_marked_cache_entries(struct index_state *istate)
529{
530 struct cache_entry **ce_array = istate->cache;
531 unsigned int i, j;
532
533 for (i = j = 0; i < istate->cache_nr; i++) {
5699d17e 534 if (ce_array[i]->ce_flags & CE_REMOVE) {
2092678c 535 remove_name_hash(istate, ce_array[i]);
045113a5 536 save_or_free_index_entry(istate, ce_array[i]);
5699d17e 537 }
36419c8e
KB
538 else
539 ce_array[j++] = ce_array[i];
540 }
ad837d9e
NTND
541 if (j == istate->cache_nr)
542 return;
e636a7b4 543 istate->cache_changed |= CE_ENTRY_REMOVED;
36419c8e
KB
544 istate->cache_nr = j;
545}
546
4aab5b46 547int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 548{
4aab5b46 549 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
550 if (pos < 0)
551 pos = -pos-1;
a5400efe 552 cache_tree_invalidate_path(istate, path);
e931371a 553 untracked_cache_remove_from_index(istate, path);
4aab5b46
JH
554 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
555 remove_index_entry_at(istate, pos);
197ee8c9
LT
556 return 0;
557}
558
20314271
JS
559static int compare_name(struct cache_entry *ce, const char *path, int namelen)
560{
561 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
562}
563
564static int index_name_pos_also_unmerged(struct index_state *istate,
565 const char *path, int namelen)
566{
567 int pos = index_name_pos(istate, path, namelen);
568 struct cache_entry *ce;
569
570 if (pos >= 0)
571 return pos;
572
573 /* maybe unmerged? */
574 pos = -1 - pos;
575 if (pos >= istate->cache_nr ||
576 compare_name((ce = istate->cache[pos]), path, namelen))
577 return -1;
578
579 /* order of preference: stage 2, 1, 3 */
580 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
581 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
582 !compare_name(ce, path, namelen))
583 pos++;
584 return pos;
585}
586
1102952b
LT
587static int different_name(struct cache_entry *ce, struct cache_entry *alias)
588{
589 int len = ce_namelen(ce);
590 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
591}
592
593/*
594 * If we add a filename that aliases in the cache, we will use the
595 * name that we already have - but we don't want to update the same
596 * alias twice, because that implies that there were actually two
597 * different files with aliasing names!
598 *
599 * So we use the CE_ADDED flag to verify that the alias was an old
600 * one before we accept it as
601 */
045113a5
NTND
602static struct cache_entry *create_alias_ce(struct index_state *istate,
603 struct cache_entry *ce,
604 struct cache_entry *alias)
1102952b
LT
605{
606 int len;
607 struct cache_entry *new;
608
609 if (alias->ce_flags & CE_ADDED)
610 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
611
612 /* Ok, create the new entry using the name of the existing alias */
613 len = ce_namelen(alias);
614 new = xcalloc(1, cache_entry_size(len));
615 memcpy(new->name, alias->name, len);
616 copy_cache_entry(new, ce);
045113a5 617 save_or_free_index_entry(istate, ce);
1102952b
LT
618 return new;
619}
620
b4b313f9 621void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
39425819
JH
622{
623 unsigned char sha1[20];
624 if (write_sha1_file("", 0, blob_type, sha1))
625 die("cannot create an empty blob in the object database");
99d1a986 626 hashcpy(ce->oid.hash, sha1);
39425819
JH
627}
628
610d55af 629int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 630{
38ed1d89 631 int size, namelen, was_same;
d177cab0 632 mode_t st_mode = st->st_mode;
6835550d 633 struct cache_entry *ce, *alias;
56cac48c 634 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
635 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
636 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
637 int intent_only = flags & ADD_CACHE_INTENT;
638 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
639 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
11be42a4 640
d177cab0 641 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
960b8ad1 642 return error("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
643
644 namelen = strlen(path);
d177cab0 645 if (S_ISDIR(st_mode)) {
09595258
LT
646 while (namelen && path[namelen-1] == '/')
647 namelen--;
648 }
11be42a4
JS
649 size = cache_entry_size(namelen);
650 ce = xcalloc(1, size);
651 memcpy(ce->name, path, namelen);
b60e188c 652 ce->ce_namelen = namelen;
39425819
JH
653 if (!intent_only)
654 fill_stat_cache_info(ce, st);
388b2acd
JH
655 else
656 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 657
610d55af
TG
658
659 if (trust_executable_bit && has_symlinks) {
d177cab0 660 ce->ce_mode = create_ce_mode(st_mode);
610d55af 661 } else {
78a8d641
JS
662 /* If there is an existing entry, pick the mode bits and type
663 * from it, otherwise assume unexecutable regular file.
11be42a4 664 */
185c975f 665 struct cache_entry *ent;
20314271 666 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 667
4aab5b46 668 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 669 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
670 }
671
dc1ae704
JJ
672 /* When core.ignorecase=true, determine if a directory of the same name but differing
673 * case already exists within the Git repository. If it does, ensure the directory
674 * case of the file being added to the repository matches (is folded into) the existing
675 * entry's directory case.
676 */
677 if (ignore_case) {
41284eb0 678 adjust_dirname_case(istate, ce->name);
dc1ae704
JJ
679 }
680
ebbd7439 681 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
d177cab0 682 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
0781b8a9 683 /* Nothing changed, really */
125fd984
JH
684 if (!S_ISGITLINK(alias->ce_mode))
685 ce_mark_uptodate(alias);
1102952b 686 alias->ce_flags |= CE_ADDED;
2d9426b0
SB
687
688 free(ce);
0781b8a9
JH
689 return 0;
690 }
39425819 691 if (!intent_only) {
99d1a986 692 if (index_path(ce->oid.hash, path, st, HASH_WRITE_OBJECT)) {
2d9426b0 693 free(ce);
39425819 694 return error("unable to index file %s", path);
2d9426b0 695 }
39425819 696 } else
b4b313f9 697 set_object_name_for_intent_to_add_entry(ce);
39425819 698
1102952b 699 if (ignore_case && alias && different_name(ce, alias))
045113a5 700 ce = create_alias_ce(istate, ce, alias);
1102952b 701 ce->ce_flags |= CE_ADDED;
38ed1d89 702
3bf0dd1f 703 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
704 was_same = (alias &&
705 !ce_stage(alias) &&
99d1a986 706 !oidcmp(&alias->oid, &ce->oid) &&
38ed1d89
JH
707 ce->ce_mode == alias->ce_mode);
708
709 if (pretend)
067178ed
JH
710 free(ce);
711 else if (add_index_entry(istate, ce, add_option)) {
712 free(ce);
713 return error("unable to add %s to index", path);
714 }
38ed1d89 715 if (verbose && !was_same)
11be42a4 716 printf("add '%s'\n", path);
11be42a4
JS
717 return 0;
718}
719
610d55af 720int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
721{
722 struct stat st;
723 if (lstat(path, &st))
d824cbba 724 die_errno("unable to stat '%s'", path);
610d55af 725 return add_to_index(istate, path, &st, flags);
d177cab0
LT
726}
727
6640f881
CR
728struct cache_entry *make_cache_entry(unsigned int mode,
729 const unsigned char *sha1, const char *path, int stage,
25762726 730 unsigned int refresh_options)
6640f881
CR
731{
732 int size, len;
bc1c2caa 733 struct cache_entry *ce, *ret;
6640f881 734
7e7abea9
DP
735 if (!verify_path(path)) {
736 error("Invalid path '%s'", path);
6640f881 737 return NULL;
7e7abea9 738 }
6640f881
CR
739
740 len = strlen(path);
741 size = cache_entry_size(len);
742 ce = xcalloc(1, size);
743
99d1a986 744 hashcpy(ce->oid.hash, sha1);
6640f881 745 memcpy(ce->name, path, len);
b60e188c
TG
746 ce->ce_flags = create_ce_flags(stage);
747 ce->ce_namelen = len;
6640f881
CR
748 ce->ce_mode = create_ce_mode(mode);
749
bc1c2caa 750 ret = refresh_cache_entry(ce, refresh_options);
915e44c6 751 if (ret != ce)
bc1c2caa 752 free(ce);
915e44c6 753 return ret;
6640f881
CR
754}
755
d9d70966
TG
756/*
757 * Chmod an index entry with either +x or -x.
758 *
759 * Returns -1 if the chmod for the particular cache entry failed (if it's
760 * not a regular file), -2 if an invalid flip argument is passed in, 0
761 * otherwise.
762 */
763int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
764 char flip)
765{
766 if (!S_ISREG(ce->ce_mode))
767 return -1;
768 switch (flip) {
769 case '+':
770 ce->ce_mode |= 0111;
771 break;
772 case '-':
773 ce->ce_mode &= ~0111;
774 break;
775 default:
776 return -2;
777 }
778 cache_tree_invalidate_path(istate, ce->name);
779 ce->ce_flags |= CE_UPDATE_IN_BASE;
780 istate->cache_changed |= CE_ENTRY_CHANGED;
781
782 return 0;
783}
784
9c5e6c80 785int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
7b937ca3
LT
786{
787 int len = ce_namelen(a);
788 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
789}
790
8dcf39c4
LT
791/*
792 * We fundamentally don't like some paths: we don't want
793 * dot or dot-dot anywhere, and for obvious reasons don't
794 * want to recurse into ".git" either.
795 *
796 * Also, we don't want double slashes or slashes at the
797 * end that can make pathnames ambiguous.
798 */
799static int verify_dotfile(const char *rest)
800{
801 /*
802 * The first character was '.', but that
803 * has already been discarded, we now test
804 * the rest.
805 */
e0f530ff 806
8dcf39c4 807 /* "." is not allowed */
e0f530ff 808 if (*rest == '\0' || is_dir_sep(*rest))
8dcf39c4
LT
809 return 0;
810
e0f530ff 811 switch (*rest) {
8dcf39c4
LT
812 /*
813 * ".git" followed by NUL or slash is bad. This
814 * shares the path end test with the ".." case.
815 */
816 case 'g':
cc2fc7c2
JK
817 case 'G':
818 if (rest[1] != 'i' && rest[1] != 'I')
8dcf39c4 819 break;
cc2fc7c2 820 if (rest[2] != 't' && rest[2] != 'T')
8dcf39c4
LT
821 break;
822 rest += 2;
823 /* fallthrough */
824 case '.':
e0f530ff 825 if (rest[1] == '\0' || is_dir_sep(rest[1]))
8dcf39c4
LT
826 return 0;
827 }
828 return 1;
829}
830
831int verify_path(const char *path)
832{
833 char c;
834
56948cb6
EFL
835 if (has_dos_drive_prefix(path))
836 return 0;
837
8dcf39c4
LT
838 goto inside;
839 for (;;) {
840 if (!c)
841 return 1;
56948cb6 842 if (is_dir_sep(c)) {
8dcf39c4 843inside:
a42643aa
JK
844 if (protect_hfs && is_hfs_dotgit(path))
845 return 0;
2b4c6efc
JS
846 if (protect_ntfs && is_ntfs_dotgit(path))
847 return 0;
8dcf39c4 848 c = *path++;
3bdf09c7
JH
849 if ((c == '.' && !verify_dotfile(path)) ||
850 is_dir_sep(c) || c == '\0')
851 return 0;
8dcf39c4
LT
852 }
853 c = *path++;
854 }
855}
856
12676608
LT
857/*
858 * Do we have another file that has the beginning components being a
859 * proper superset of the name we're trying to add?
0f1e4f04 860 */
4aab5b46
JH
861static int has_file_name(struct index_state *istate,
862 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 863{
12676608
LT
864 int retval = 0;
865 int len = ce_namelen(ce);
b155725d 866 int stage = ce_stage(ce);
12676608 867 const char *name = ce->name;
0f1e4f04 868
4aab5b46
JH
869 while (pos < istate->cache_nr) {
870 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 871
12676608 872 if (len >= ce_namelen(p))
0f1e4f04 873 break;
12676608
LT
874 if (memcmp(name, p->name, len))
875 break;
b155725d
JH
876 if (ce_stage(p) != stage)
877 continue;
12676608
LT
878 if (p->name[len] != '/')
879 continue;
7a51ed66 880 if (p->ce_flags & CE_REMOVE)
21cd8d00 881 continue;
12676608
LT
882 retval = -1;
883 if (!ok_to_replace)
884 break;
4aab5b46 885 remove_index_entry_at(istate, --pos);
0f1e4f04 886 }
12676608
LT
887 return retval;
888}
0f1e4f04 889
a6db3fbb
JH
890
891/*
892 * Like strcmp(), but also return the offset of the first change.
893 * If strings are equal, return the length.
894 */
895int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
896{
897 size_t k;
898
899 if (!first_change)
900 return strcmp(s1, s2);
901
902 for (k = 0; s1[k] == s2[k]; k++)
903 if (s1[k] == '\0')
904 break;
905
906 *first_change = k;
907 return (unsigned char)s1[k] - (unsigned char)s2[k];
908}
909
12676608
LT
910/*
911 * Do we have another file with a pathname that is a proper
912 * subset of the name we're trying to add?
06b6d81b
JH
913 *
914 * That is, is there another file in the index with a path
915 * that matches a sub-directory in the given entry?
12676608 916 */
4aab5b46
JH
917static int has_dir_name(struct index_state *istate,
918 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
919{
920 int retval = 0;
b155725d 921 int stage = ce_stage(ce);
12676608
LT
922 const char *name = ce->name;
923 const char *slash = name + ce_namelen(ce);
06b6d81b
JH
924 size_t len_eq_last;
925 int cmp_last = 0;
926
927 /*
928 * We are frequently called during an iteration on a sorted
929 * list of pathnames and while building a new index. Therefore,
930 * there is a high probability that this entry will eventually
931 * be appended to the index, rather than inserted in the middle.
932 * If we can confirm that, we can avoid binary searches on the
933 * components of the pathname.
934 *
935 * Compare the entry's full path with the last path in the index.
936 */
937 if (istate->cache_nr > 0) {
938 cmp_last = strcmp_offset(name,
939 istate->cache[istate->cache_nr - 1]->name,
940 &len_eq_last);
941 if (cmp_last > 0) {
942 if (len_eq_last == 0) {
943 /*
944 * The entry sorts AFTER the last one in the
945 * index and their paths have no common prefix,
946 * so there cannot be a F/D conflict.
947 */
948 return retval;
949 } else {
950 /*
951 * The entry sorts AFTER the last one in the
952 * index, but has a common prefix. Fall through
953 * to the loop below to disect the entry's path
954 * and see where the difference is.
955 */
956 }
957 } else if (cmp_last == 0) {
958 /*
959 * The entry exactly matches the last one in the
960 * index, but because of multiple stage and CE_REMOVE
961 * items, we fall through and let the regular search
962 * code handle it.
963 */
964 }
965 }
0f1e4f04 966
12676608
LT
967 for (;;) {
968 int len;
0f1e4f04 969
12676608
LT
970 for (;;) {
971 if (*--slash == '/')
972 break;
973 if (slash <= ce->name)
974 return retval;
975 }
976 len = slash - name;
0f1e4f04 977
b60e188c 978 pos = index_name_stage_pos(istate, name, len, stage);
12676608 979 if (pos >= 0) {
21cd8d00
JH
980 /*
981 * Found one, but not so fast. This could
982 * be a marker that says "I was here, but
983 * I am being removed". Such an entry is
984 * not a part of the resulting tree, and
985 * it is Ok to have a directory at the same
986 * path.
987 */
077c48df 988 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
989 retval = -1;
990 if (!ok_to_replace)
991 break;
4aab5b46 992 remove_index_entry_at(istate, pos);
21cd8d00
JH
993 continue;
994 }
12676608 995 }
21cd8d00
JH
996 else
997 pos = -pos-1;
12676608
LT
998
999 /*
1000 * Trivial optimization: if we find an entry that
1001 * already matches the sub-directory, then we know
b155725d 1002 * we're ok, and we can exit.
12676608 1003 */
4aab5b46
JH
1004 while (pos < istate->cache_nr) {
1005 struct cache_entry *p = istate->cache[pos];
b155725d
JH
1006 if ((ce_namelen(p) <= len) ||
1007 (p->name[len] != '/') ||
1008 memcmp(p->name, name, len))
1009 break; /* not our subdirectory */
077c48df
JH
1010 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1011 /*
1012 * p is at the same stage as our entry, and
b155725d
JH
1013 * is a subdirectory of what we are looking
1014 * at, so we cannot have conflicts at our
1015 * level or anything shorter.
1016 */
1017 return retval;
1018 pos++;
192268c1 1019 }
0f1e4f04 1020 }
12676608
LT
1021 return retval;
1022}
1023
1024/* We may be in a situation where we already have path/file and path
1025 * is being added, or we already have path and path/file is being
1026 * added. Either one would result in a nonsense tree that has path
1027 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 1028 *
12676608
LT
1029 * If ok-to-replace is specified, we remove the conflicting entries
1030 * from the cache so the caller should recompute the insert position.
1031 * When this happens, we return non-zero.
1032 */
4aab5b46
JH
1033static int check_file_directory_conflict(struct index_state *istate,
1034 const struct cache_entry *ce,
1035 int pos, int ok_to_replace)
12676608 1036{
21cd8d00
JH
1037 int retval;
1038
1039 /*
1040 * When ce is an "I am going away" entry, we allow it to be added
1041 */
7a51ed66 1042 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
1043 return 0;
1044
12676608
LT
1045 /*
1046 * We check if the path is a sub-path of a subsequent pathname
1047 * first, since removing those will not change the position
21cd8d00 1048 * in the array.
12676608 1049 */
4aab5b46 1050 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 1051
12676608
LT
1052 /*
1053 * Then check if the path might have a clashing sub-directory
1054 * before it.
1055 */
4aab5b46 1056 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
1057}
1058
af3785dc 1059static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
1060{
1061 int pos;
192268c1
JH
1062 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1063 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 1064 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 1065 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 1066
ce7c614b
NTND
1067 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1068 cache_tree_invalidate_path(istate, ce->name);
e5494631
JH
1069
1070 /*
1071 * If this entry's path sorts after the last entry in the index,
1072 * we can avoid searching for it.
1073 */
1074 if (istate->cache_nr > 0 &&
1075 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1076 pos = -istate->cache_nr - 1;
1077 else
1078 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
197ee8c9 1079
3e09cdfd 1080 /* existing match? Just replace it. */
76e7f4ec 1081 if (pos >= 0) {
39425819
JH
1082 if (!new_only)
1083 replace_index_entry(istate, pos, ce);
197ee8c9
LT
1084 return 0;
1085 }
76e7f4ec 1086 pos = -pos-1;
197ee8c9 1087
ffcc9ba7
NTND
1088 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1089 untracked_cache_add_to_index(istate, ce->name);
e931371a 1090
7b937ca3
LT
1091 /*
1092 * Inserting a merged entry ("stage 0") into the index
1093 * will always replace all non-merged entries..
1094 */
4aab5b46
JH
1095 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1096 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 1097 ok_to_add = 1;
4aab5b46 1098 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
1099 break;
1100 }
1101 }
1102
121481ab
LT
1103 if (!ok_to_add)
1104 return -1;
8dcf39c4 1105 if (!verify_path(ce->name))
7e7abea9 1106 return error("Invalid path '%s'", ce->name);
121481ab 1107
3e09cdfd 1108 if (!skip_df_check &&
4aab5b46 1109 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 1110 if (!ok_to_replace)
4aab5b46
JH
1111 return error("'%s' appears as both a file and as a directory",
1112 ce->name);
b60e188c 1113 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
192268c1
JH
1114 pos = -pos-1;
1115 }
af3785dc
JH
1116 return pos + 1;
1117}
1118
1119int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1120{
1121 int pos;
1122
1123 if (option & ADD_CACHE_JUST_APPEND)
1124 pos = istate->cache_nr;
1125 else {
1126 int ret;
1127 ret = add_index_entry_with_check(istate, ce, option);
1128 if (ret <= 0)
1129 return ret;
1130 pos = ret - 1;
1131 }
0f1e4f04 1132
197ee8c9 1133 /* Make sure the array is big enough .. */
999f5660 1134 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
197ee8c9
LT
1135
1136 /* Add it in.. */
4aab5b46 1137 istate->cache_nr++;
af3785dc 1138 if (istate->cache_nr > pos + 1)
4aab5b46
JH
1139 memmove(istate->cache + pos + 1,
1140 istate->cache + pos,
1141 (istate->cache_nr - pos - 1) * sizeof(ce));
cf558704 1142 set_index_entry(istate, pos, ce);
e636a7b4 1143 istate->cache_changed |= CE_ENTRY_ADDED;
197ee8c9
LT
1144 return 0;
1145}
1146
405e5b2f
LT
1147/*
1148 * "refresh" does not calculate a new sha1 file or bring the
1149 * cache up-to-date for mode/content changes. But what it
1150 * _does_ do is to "re-match" the stat information of a file
1151 * with the cache, so that you can refresh the cache for a
1152 * file that hasn't been changed but where the stat entry is
1153 * out of date.
1154 *
1155 * For example, you'd want to do this after doing a "git-read-tree",
1156 * to link up the stat cache details with the proper files.
1157 */
4aab5b46 1158static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da 1159 struct cache_entry *ce,
d05e6970
JK
1160 unsigned int options, int *err,
1161 int *changed_ret)
405e5b2f
LT
1162{
1163 struct stat st;
1164 struct cache_entry *updated;
1165 int changed, size;
25762726 1166 int refresh = options & CE_MATCH_REFRESH;
4bd5b7da 1167 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1168 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
2e2e7ec1 1169 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
405e5b2f 1170
25762726 1171 if (!refresh || ce_uptodate(ce))
eadb5831
JH
1172 return ce;
1173
aa9349d4 1174 /*
56cac48c
NTND
1175 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1176 * that the change to the work tree does not matter and told
1177 * us not to worry.
aa9349d4 1178 */
56cac48c
NTND
1179 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1180 ce_mark_uptodate(ce);
1181 return ce;
1182 }
aa9349d4
MSO
1183 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1184 ce_mark_uptodate(ce);
1185 return ce;
1186 }
1187
ccad42d4
RS
1188 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1189 if (ignore_missing)
1190 return ce;
1191 if (err)
1192 *err = ENOENT;
1193 return NULL;
1194 }
1195
8fd2cb40 1196 if (lstat(ce->name, &st) < 0) {
2e2e7ec1
BK
1197 if (ignore_missing && errno == ENOENT)
1198 return ce;
ec0cc704
JH
1199 if (err)
1200 *err = errno;
8fd2cb40
JS
1201 return NULL;
1202 }
405e5b2f 1203
4bd5b7da 1204 changed = ie_match_stat(istate, ce, &st, options);
d05e6970
JK
1205 if (changed_ret)
1206 *changed_ret = changed;
405e5b2f 1207 if (!changed) {
4bd5b7da
JH
1208 /*
1209 * The path is unchanged. If we were told to ignore
1210 * valid bit, then we did the actual stat check and
1211 * found that the entry is unmodified. If the entry
1212 * is not marked VALID, this is the place to mark it
1213 * valid again, under "assume unchanged" mode.
1214 */
1215 if (ignore_valid && assume_unchanged &&
7a51ed66 1216 !(ce->ce_flags & CE_VALID))
405e5b2f 1217 ; /* mark this one VALID again */
eadb5831
JH
1218 else {
1219 /*
1220 * We do not mark the index itself "modified"
1221 * because CE_UPTODATE flag is in-core only;
1222 * we are not going to write this change out.
1223 */
125fd984
JH
1224 if (!S_ISGITLINK(ce->ce_mode))
1225 ce_mark_uptodate(ce);
8fd2cb40 1226 return ce;
eadb5831 1227 }
405e5b2f
LT
1228 }
1229
4bd5b7da 1230 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1231 if (err)
1232 *err = EINVAL;
8fd2cb40
JS
1233 return NULL;
1234 }
405e5b2f
LT
1235
1236 size = ce_size(ce);
1237 updated = xmalloc(size);
1238 memcpy(updated, ce, size);
1239 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1240 /*
1241 * If ignore_valid is not set, we should leave CE_VALID bit
1242 * alone. Otherwise, paths marked with --no-assume-unchanged
1243 * (i.e. things to be edited) will reacquire CE_VALID bit
1244 * automatically, which is not really what we want.
405e5b2f 1245 */
4bd5b7da 1246 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1247 !(ce->ce_flags & CE_VALID))
1248 updated->ce_flags &= ~CE_VALID;
405e5b2f 1249
e636a7b4 1250 /* istate->cache_changed is updated in the caller */
405e5b2f
LT
1251 return updated;
1252}
1253
3deffc52 1254static void show_file(const char * fmt, const char * name, int in_porcelain,
046613c5 1255 int * first, const char *header_msg)
3deffc52
MM
1256{
1257 if (in_porcelain && *first && header_msg) {
1258 printf("%s\n", header_msg);
cd2b8ae9 1259 *first = 0;
3deffc52
MM
1260 }
1261 printf(fmt, name);
1262}
1263
9b2d6149
NTND
1264int refresh_index(struct index_state *istate, unsigned int flags,
1265 const struct pathspec *pathspec,
046613c5 1266 char *seen, const char *header_msg)
405e5b2f
LT
1267{
1268 int i;
1269 int has_errors = 0;
1270 int really = (flags & REFRESH_REALLY) != 0;
1271 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1272 int quiet = (flags & REFRESH_QUIET) != 0;
1273 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1274 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1275 int first = 1;
1276 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
25762726
BK
1277 unsigned int options = (CE_MATCH_REFRESH |
1278 (really ? CE_MATCH_IGNORE_VALID : 0) |
2e2e7ec1 1279 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
4bd4e730 1280 const char *modified_fmt;
73b7eae6
JK
1281 const char *deleted_fmt;
1282 const char *typechange_fmt;
1283 const char *added_fmt;
4bd4e730 1284 const char *unmerged_fmt;
405e5b2f 1285
4bd4e730 1286 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
73b7eae6
JK
1287 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1288 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1289 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
4bd4e730 1290 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
4aab5b46 1291 for (i = 0; i < istate->cache_nr; i++) {
405e5b2f 1292 struct cache_entry *ce, *new;
ec0cc704 1293 int cache_errno = 0;
73b7eae6 1294 int changed = 0;
3d1f148c 1295 int filtered = 0;
ec0cc704 1296
4aab5b46 1297 ce = istate->cache[i];
5fdeacb0
JS
1298 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1299 continue;
1300
429bb40a 1301 if (pathspec && !ce_path_match(ce, pathspec, seen))
3d1f148c
JH
1302 filtered = 1;
1303
405e5b2f 1304 if (ce_stage(ce)) {
4aab5b46
JH
1305 while ((i < istate->cache_nr) &&
1306 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1307 i++;
1308 i--;
1309 if (allow_unmerged)
1310 continue;
3d1f148c
JH
1311 if (!filtered)
1312 show_file(unmerged_fmt, ce->name, in_porcelain,
1313 &first, header_msg);
405e5b2f
LT
1314 has_errors = 1;
1315 continue;
1316 }
1317
3d1f148c 1318 if (filtered)
d616813d
AJ
1319 continue;
1320
73b7eae6 1321 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
8fd2cb40 1322 if (new == ce)
405e5b2f 1323 continue;
8fd2cb40 1324 if (!new) {
73b7eae6
JK
1325 const char *fmt;
1326
8fd2cb40 1327 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1328 /* If we are doing --really-refresh that
1329 * means the index is not valid anymore.
1330 */
7a51ed66 1331 ce->ce_flags &= ~CE_VALID;
078a58e8 1332 ce->ce_flags |= CE_UPDATE_IN_BASE;
e636a7b4 1333 istate->cache_changed |= CE_ENTRY_CHANGED;
405e5b2f
LT
1334 }
1335 if (quiet)
1336 continue;
73b7eae6
JK
1337
1338 if (cache_errno == ENOENT)
1339 fmt = deleted_fmt;
895ff3b2 1340 else if (ce_intent_to_add(ce))
73b7eae6
JK
1341 fmt = added_fmt; /* must be before other checks */
1342 else if (changed & TYPE_CHANGED)
1343 fmt = typechange_fmt;
1344 else
1345 fmt = modified_fmt;
1346 show_file(fmt,
1347 ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1348 has_errors = 1;
1349 continue;
1350 }
cf558704
LT
1351
1352 replace_index_entry(istate, i, new);
405e5b2f
LT
1353 }
1354 return has_errors;
1355}
1356
1335d76e 1357struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
25762726 1358 unsigned int options)
ec0cc704 1359{
25762726 1360 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
ec0cc704
JH
1361}
1362
db3b313c
JH
1363
1364/*****************************************************************
1365 * Index File I/O
1366 *****************************************************************/
1367
9d227781
JH
1368#define INDEX_FORMAT_DEFAULT 3
1369
136347d7
TG
1370static unsigned int get_index_format_default(void)
1371{
1372 char *envversion = getenv("GIT_INDEX_VERSION");
3c09d684 1373 char *endp;
b27a5720 1374 int value;
3c09d684
TG
1375 unsigned int version = INDEX_FORMAT_DEFAULT;
1376
136347d7 1377 if (!envversion) {
b27a5720
TA
1378 if (!git_config_get_int("index.version", &value))
1379 version = value;
3c09d684
TG
1380 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1381 warning(_("index.version set, but the value is invalid.\n"
136347d7 1382 "Using version %i"), INDEX_FORMAT_DEFAULT);
3c09d684 1383 return INDEX_FORMAT_DEFAULT;
136347d7
TG
1384 }
1385 return version;
1386 }
3c09d684
TG
1387
1388 version = strtoul(envversion, &endp, 10);
1389 if (*endp ||
1390 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1391 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1392 "Using version %i"), INDEX_FORMAT_DEFAULT);
1393 version = INDEX_FORMAT_DEFAULT;
1394 }
1395 return version;
136347d7
TG
1396}
1397
db3b313c
JH
1398/*
1399 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1400 * Again - this is just a (very strong in practice) heuristic that
1401 * the inode hasn't changed.
1402 *
1403 * We save the fields in big-endian order to allow using the
1404 * index file over NFS transparently.
1405 */
1406struct ondisk_cache_entry {
1407 struct cache_time ctime;
1408 struct cache_time mtime;
7800c1eb
TG
1409 uint32_t dev;
1410 uint32_t ino;
1411 uint32_t mode;
1412 uint32_t uid;
1413 uint32_t gid;
1414 uint32_t size;
db3b313c 1415 unsigned char sha1[20];
7800c1eb 1416 uint16_t flags;
db3b313c
JH
1417 char name[FLEX_ARRAY]; /* more */
1418};
1419
1420/*
1421 * This struct is used when CE_EXTENDED bit is 1
1422 * The struct must match ondisk_cache_entry exactly from
1423 * ctime till flags
1424 */
1425struct ondisk_cache_entry_extended {
1426 struct cache_time ctime;
1427 struct cache_time mtime;
7800c1eb
TG
1428 uint32_t dev;
1429 uint32_t ino;
1430 uint32_t mode;
1431 uint32_t uid;
1432 uint32_t gid;
1433 uint32_t size;
db3b313c 1434 unsigned char sha1[20];
7800c1eb
TG
1435 uint16_t flags;
1436 uint16_t flags2;
db3b313c
JH
1437 char name[FLEX_ARRAY]; /* more */
1438};
1439
6c9cd161 1440/* These are only used for v3 or lower */
db3b313c
JH
1441#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1442#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1443#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1444#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1445 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1446 ondisk_cache_entry_size(ce_namelen(ce)))
1447
bad68ec9 1448static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163 1449{
9126f009 1450 git_SHA_CTX c;
bad68ec9 1451 unsigned char sha1[20];
0136bac9 1452 int hdr_version;
e83c5163 1453
ccc4feb5 1454 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1455 return error("bad signature");
0136bac9 1456 hdr_version = ntohl(hdr->hdr_version);
b82a7b5b 1457 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
0136bac9 1458 return error("bad index version %d", hdr_version);
9126f009
NP
1459 git_SHA1_Init(&c);
1460 git_SHA1_Update(&c, hdr, size - 20);
1461 git_SHA1_Final(sha1, &c);
a89fccd2 1462 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
ca9be054 1463 return error("bad index file sha1 signature");
e83c5163
LT
1464 return 0;
1465}
1466
4aab5b46
JH
1467static int read_index_extension(struct index_state *istate,
1468 const char *ext, void *data, unsigned long sz)
bad68ec9
JH
1469{
1470 switch (CACHE_EXT(ext)) {
1471 case CACHE_EXT_TREE:
4aab5b46 1472 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1473 break;
cfc5789a
JH
1474 case CACHE_EXT_RESOLVE_UNDO:
1475 istate->resolve_undo = resolve_undo_read(data, sz);
1476 break;
5fc2fc8f
NTND
1477 case CACHE_EXT_LINK:
1478 if (read_link_extension(istate, data, sz))
1479 return -1;
1480 break;
f9e6c649
NTND
1481 case CACHE_EXT_UNTRACKED:
1482 istate->untracked = read_untracked_extension(data, sz);
1483 break;
bad68ec9
JH
1484 default:
1485 if (*ext < 'A' || 'Z' < *ext)
1486 return error("index uses %.4s extension, which we do not understand",
1487 ext);
1488 fprintf(stderr, "ignoring %.4s extension\n", ext);
1489 break;
1490 }
1491 return 0;
1492}
1493
b3e83cc7 1494int hold_locked_index(struct lock_file *lk, int lock_flags)
216aab1e 1495{
b3e83cc7 1496 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
216aab1e
MH
1497}
1498
4aab5b46 1499int read_index(struct index_state *istate)
8fd2cb40 1500{
4aab5b46 1501 return read_index_from(istate, get_index_file());
8fd2cb40
JS
1502}
1503
3fc22b53
JH
1504static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1505 unsigned int flags,
1506 const char *name,
1507 size_t len)
1508{
1509 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1510
c3d8da57
JK
1511 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1512 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1513 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1514 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1515 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1516 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1517 ce->ce_mode = get_be32(&ondisk->mode);
1518 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1519 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1520 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
b60e188c
TG
1521 ce->ce_flags = flags & ~CE_NAMEMASK;
1522 ce->ce_namelen = len;
5fc2fc8f 1523 ce->index = 0;
99d1a986 1524 hashcpy(ce->oid.hash, ondisk->sha1);
3fc22b53
JH
1525 memcpy(ce->name, name, len);
1526 ce->name[len] = '\0';
1527 return ce;
1528}
1529
6c9cd161
JH
1530/*
1531 * Adjacent cache entries tend to share the leading paths, so it makes
1532 * sense to only store the differences in later entries. In the v4
1533 * on-disk format of the index, each on-disk cache entry stores the
1534 * number of bytes to be stripped from the end of the previous name,
1535 * and the bytes to append to the result, to come up with its name.
1536 */
1537static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1538{
1539 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1540 size_t len = decode_varint(&cp);
1541
1542 if (name->len < len)
1543 die("malformed name field in the index");
1544 strbuf_remove(name, name->len - len, len);
1545 for (ep = cp; *ep; ep++)
1546 ; /* find the end */
1547 strbuf_add(name, cp, ep - cp);
1548 return (const char *)ep + 1 - cp_;
1549}
1550
936f53d0 1551static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
6c9cd161
JH
1552 unsigned long *ent_size,
1553 struct strbuf *previous_name)
7a51ed66 1554{
debed2a6 1555 struct cache_entry *ce;
7fec10b7 1556 size_t len;
06aaaa0b 1557 const char *name;
debed2a6 1558 unsigned int flags;
7fec10b7 1559
7a51ed66 1560 /* On-disk flags are just 16 bits */
c3d8da57 1561 flags = get_be16(&ondisk->flags);
debed2a6 1562 len = flags & CE_NAMEMASK;
7fec10b7 1563
debed2a6 1564 if (flags & CE_EXTENDED) {
06aaaa0b
NTND
1565 struct ondisk_cache_entry_extended *ondisk2;
1566 int extended_flags;
1567 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
c3d8da57 1568 extended_flags = get_be16(&ondisk2->flags2) << 16;
06aaaa0b
NTND
1569 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1570 if (extended_flags & ~CE_EXTENDED_FLAGS)
1571 die("Unknown index entry format %08x", extended_flags);
debed2a6 1572 flags |= extended_flags;
06aaaa0b
NTND
1573 name = ondisk2->name;
1574 }
1575 else
1576 name = ondisk->name;
1577
6c9cd161
JH
1578 if (!previous_name) {
1579 /* v3 and earlier */
1580 if (len == CE_NAMEMASK)
1581 len = strlen(name);
1582 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1583
1584 *ent_size = ondisk_ce_size(ce);
1585 } else {
1586 unsigned long consumed;
1587 consumed = expand_name_field(previous_name, name);
1588 ce = cache_entry_from_ondisk(ondisk, flags,
1589 previous_name->buf,
1590 previous_name->len);
1591
1592 *ent_size = (name - ((char *)ondisk)) + consumed;
1593 }
debed2a6 1594 return ce;
cf558704
LT
1595}
1596
03f15a79 1597static void check_ce_order(struct index_state *istate)
15999d0b 1598{
03f15a79
TG
1599 unsigned int i;
1600
1601 for (i = 1; i < istate->cache_nr; i++) {
1602 struct cache_entry *ce = istate->cache[i - 1];
1603 struct cache_entry *next_ce = istate->cache[i];
1604 int name_compare = strcmp(ce->name, next_ce->name);
1605
1606 if (0 < name_compare)
1607 die("unordered stage entries in index");
1608 if (!name_compare) {
1609 if (!ce_stage(ce))
1610 die("multiple stage entries for merged file '%s'",
1611 ce->name);
1612 if (ce_stage(ce) > ce_stage(next_ce))
1613 die("unordered stage entries for '%s'",
1614 ce->name);
1615 }
15999d0b
JSP
1616 }
1617}
1618
435ec090
CC
1619static void tweak_untracked_cache(struct index_state *istate)
1620{
1621 switch (git_config_get_untracked_cache()) {
1622 case -1: /* keep: do nothing */
1623 break;
1624 case 0: /* false */
1625 remove_untracked_cache(istate);
1626 break;
1627 case 1: /* true */
1628 add_untracked_cache(istate);
1629 break;
1630 default: /* unknown value: do nothing */
1631 break;
1632 }
1633}
1634
43925312
CC
1635static void tweak_split_index(struct index_state *istate)
1636{
1637 switch (git_config_get_split_index()) {
1638 case -1: /* unset: do nothing */
1639 break;
1640 case 0: /* false */
1641 remove_split_index(istate);
1642 break;
1643 case 1: /* true */
1644 add_split_index(istate);
1645 break;
1646 default: /* unknown value: do nothing */
1647 break;
1648 }
1649}
1650
435ec090
CC
1651static void post_read_index_from(struct index_state *istate)
1652{
1653 check_ce_order(istate);
1654 tweak_untracked_cache(istate);
43925312 1655 tweak_split_index(istate);
435ec090
CC
1656}
1657
8fd2cb40 1658/* remember to discard_cache() before reading a different cache! */
3e52f70b 1659int do_read_index(struct index_state *istate, const char *path, int must_exist)
e83c5163
LT
1660{
1661 int fd, i;
1662 struct stat st;
debed2a6 1663 unsigned long src_offset;
e83c5163 1664 struct cache_header *hdr;
7a51ed66
LT
1665 void *mmap;
1666 size_t mmap_size;
6c9cd161 1667 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
e83c5163 1668
913e0e99 1669 if (istate->initialized)
4aab5b46 1670 return istate->cache_nr;
5d1a5c02 1671
fba2f38a
KB
1672 istate->timestamp.sec = 0;
1673 istate->timestamp.nsec = 0;
8fd2cb40 1674 fd = open(path, O_RDONLY);
5d1a5c02 1675 if (fd < 0) {
5fc2fc8f 1676 if (!must_exist && errno == ENOENT)
5d1a5c02 1677 return 0;
5fc2fc8f 1678 die_errno("%s: index file open failed", path);
5d1a5c02 1679 }
e83c5163 1680
3511a377 1681 if (fstat(fd, &st))
d824cbba 1682 die_errno("cannot stat the open index");
3511a377 1683
7a51ed66
LT
1684 mmap_size = xsize_t(st.st_size);
1685 if (mmap_size < sizeof(struct cache_header) + 20)
3511a377
LFC
1686 die("index file smaller than expected");
1687
a1293ef7 1688 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
7a51ed66 1689 if (mmap == MAP_FAILED)
0721c314 1690 die_errno("unable to map index file");
57d84f8d 1691 close(fd);
e83c5163 1692
7a51ed66
LT
1693 hdr = mmap;
1694 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1695 goto unmap;
1696
e93021b2 1697 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
9d227781 1698 istate->version = ntohl(hdr->hdr_version);
4aab5b46
JH
1699 istate->cache_nr = ntohl(hdr->hdr_entries);
1700 istate->cache_alloc = alloc_nr(istate->cache_nr);
c4aa3167 1701 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
913e0e99 1702 istate->initialized = 1;
7a51ed66 1703
9d227781 1704 if (istate->version == 4)
6c9cd161
JH
1705 previous_name = &previous_name_buf;
1706 else
1707 previous_name = NULL;
1708
7a51ed66 1709 src_offset = sizeof(*hdr);
4aab5b46 1710 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 1711 struct ondisk_cache_entry *disk_ce;
4aab5b46 1712 struct cache_entry *ce;
936f53d0 1713 unsigned long consumed;
4aab5b46 1714
7a51ed66 1715 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
6c9cd161 1716 ce = create_from_disk(disk_ce, &consumed, previous_name);
cf558704 1717 set_index_entry(istate, i, ce);
7a51ed66 1718
936f53d0 1719 src_offset += consumed;
e83c5163 1720 }
6c9cd161 1721 strbuf_release(&previous_name_buf);
fba2f38a 1722 istate->timestamp.sec = st.st_mtime;
c06ff490 1723 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 1724
7a51ed66 1725 while (src_offset <= mmap_size - 20 - 8) {
bad68ec9
JH
1726 /* After an array of active_nr index entries,
1727 * there can be arbitrary number of extended
1728 * sections, each of which is prefixed with
1729 * extension name (4-byte) and section length
1730 * in 4-byte network byte order.
1731 */
07cc8eca 1732 uint32_t extsize;
7a51ed66 1733 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
bad68ec9 1734 extsize = ntohl(extsize);
4aab5b46 1735 if (read_index_extension(istate,
7a51ed66
LT
1736 (const char *) mmap + src_offset,
1737 (char *) mmap + src_offset + 8,
1d7f171c 1738 extsize) < 0)
bad68ec9 1739 goto unmap;
7a51ed66
LT
1740 src_offset += 8;
1741 src_offset += extsize;
bad68ec9 1742 }
7a51ed66 1743 munmap(mmap, mmap_size);
4aab5b46 1744 return istate->cache_nr;
e83c5163
LT
1745
1746unmap:
7a51ed66 1747 munmap(mmap, mmap_size);
5d1a5c02 1748 die("index file corrupt");
e83c5163
LT
1749}
1750
0d59ffb4
CC
1751/*
1752 * Signal that the shared index is used by updating its mtime.
1753 *
1754 * This way, shared index can be removed if they have not been used
1755 * for some time.
1756 */
1757static void freshen_shared_index(char *base_sha1_hex, int warn)
1758{
1759 const char *shared_index = git_path("sharedindex.%s", base_sha1_hex);
1760 if (!check_and_freshen_file(shared_index, 1) && warn)
1761 warning("could not freshen shared index '%s'", shared_index);
1762}
1763
5fc2fc8f
NTND
1764int read_index_from(struct index_state *istate, const char *path)
1765{
1766 struct split_index *split_index;
1767 int ret;
de6ae5f9
CC
1768 char *base_sha1_hex;
1769 const char *base_path;
5fc2fc8f
NTND
1770
1771 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1772 if (istate->initialized)
1773 return istate->cache_nr;
1774
1775 ret = do_read_index(istate, path, 0);
435ec090 1776
5fc2fc8f 1777 split_index = istate->split_index;
03f15a79 1778 if (!split_index || is_null_sha1(split_index->base_sha1)) {
435ec090 1779 post_read_index_from(istate);
5fc2fc8f 1780 return ret;
03f15a79 1781 }
5fc2fc8f
NTND
1782
1783 if (split_index->base)
1784 discard_index(split_index->base);
1785 else
1786 split_index->base = xcalloc(1, sizeof(*split_index->base));
de6ae5f9
CC
1787
1788 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1789 base_path = git_path("sharedindex.%s", base_sha1_hex);
1790 ret = do_read_index(split_index->base, base_path, 1);
5fc2fc8f
NTND
1791 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1792 die("broken index, expect %s in %s, got %s",
de6ae5f9 1793 base_sha1_hex, base_path,
5fc2fc8f 1794 sha1_to_hex(split_index->base->sha1));
de6ae5f9 1795
c3a00825 1796 freshen_shared_index(base_sha1_hex, 0);
5fc2fc8f 1797 merge_base_index(istate);
435ec090 1798 post_read_index_from(istate);
5fc2fc8f
NTND
1799 return ret;
1800}
1801
fa7b3c2f
JH
1802int is_index_unborn(struct index_state *istate)
1803{
debed2a6 1804 return (!istate->cache_nr && !istate->timestamp.sec);
fa7b3c2f
JH
1805}
1806
4aab5b46 1807int discard_index(struct index_state *istate)
6d297f81 1808{
debed2a6
RS
1809 int i;
1810
5fc2fc8f
NTND
1811 for (i = 0; i < istate->cache_nr; i++) {
1812 if (istate->cache[i]->index &&
1813 istate->split_index &&
1814 istate->split_index->base &&
1815 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1816 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1817 continue;
debed2a6 1818 free(istate->cache[i]);
5fc2fc8f 1819 }
cfc5789a 1820 resolve_undo_clear_index(istate);
4aab5b46
JH
1821 istate->cache_nr = 0;
1822 istate->cache_changed = 0;
fba2f38a
KB
1823 istate->timestamp.sec = 0;
1824 istate->timestamp.nsec = 0;
2092678c 1825 free_name_hash(istate);
4aab5b46 1826 cache_tree_free(&(istate->cache_tree));
913e0e99 1827 istate->initialized = 0;
a0fc4db0
RS
1828 free(istate->cache);
1829 istate->cache = NULL;
1830 istate->cache_alloc = 0;
5fc2fc8f 1831 discard_split_index(istate);
f9e6c649
NTND
1832 free_untracked_cache(istate->untracked);
1833 istate->untracked = NULL;
7a51ed66 1834 return 0;
6d297f81
JS
1835}
1836
d1f128b0 1837int unmerged_index(const struct index_state *istate)
94a5728c
DB
1838{
1839 int i;
1840 for (i = 0; i < istate->cache_nr; i++) {
1841 if (ce_stage(istate->cache[i]))
1842 return 1;
1843 }
1844 return 0;
1845}
1846
4990aadc 1847#define WRITE_BUFFER_SIZE 8192
bf0f910d 1848static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
1849static unsigned long write_buffer_len;
1850
9126f009 1851static int ce_write_flush(git_SHA_CTX *context, int fd)
6015c28b
JH
1852{
1853 unsigned int buffered = write_buffer_len;
1854 if (buffered) {
9126f009 1855 git_SHA1_Update(context, write_buffer, buffered);
93822c22 1856 if (write_in_full(fd, write_buffer, buffered) != buffered)
6015c28b
JH
1857 return -1;
1858 write_buffer_len = 0;
1859 }
1860 return 0;
1861}
1862
9126f009 1863static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
1864{
1865 while (len) {
1866 unsigned int buffered = write_buffer_len;
1867 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1868 if (partial > len)
1869 partial = len;
1870 memcpy(write_buffer + buffered, data, partial);
1871 buffered += partial;
1872 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
1873 write_buffer_len = buffered;
1874 if (ce_write_flush(context, fd))
4990aadc
LT
1875 return -1;
1876 buffered = 0;
1877 }
1878 write_buffer_len = buffered;
1879 len -= partial;
1d7f171c 1880 data = (char *) data + partial;
a6080a0a
JH
1881 }
1882 return 0;
4990aadc
LT
1883}
1884
9126f009 1885static int write_index_ext_header(git_SHA_CTX *context, int fd,
ac58c7b1 1886 unsigned int ext, unsigned int sz)
bad68ec9
JH
1887{
1888 ext = htonl(ext);
1889 sz = htonl(sz);
968a1d65
DR
1890 return ((ce_write(context, fd, &ext, 4) < 0) ||
1891 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
1892}
1893
e93021b2 1894static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
4990aadc
LT
1895{
1896 unsigned int left = write_buffer_len;
ca9be054 1897
4990aadc
LT
1898 if (left) {
1899 write_buffer_len = 0;
9126f009 1900 git_SHA1_Update(context, write_buffer, left);
4990aadc 1901 }
ca9be054 1902
2c865d9a
QH
1903 /* Flush first if not enough space for SHA1 signature */
1904 if (left + 20 > WRITE_BUFFER_SIZE) {
93822c22 1905 if (write_in_full(fd, write_buffer, left) != left)
2c865d9a
QH
1906 return -1;
1907 left = 0;
1908 }
1909
ca9be054 1910 /* Append the SHA1 signature at the end */
9126f009 1911 git_SHA1_Final(write_buffer + left, context);
e93021b2 1912 hashcpy(sha1, write_buffer + left);
ca9be054 1913 left += 20;
93822c22 1914 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
4990aadc
LT
1915}
1916
407c8eb0
JH
1917static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1918{
1919 /*
1920 * The only thing we care about in this function is to smudge the
1921 * falsely clean entry due to touch-update-touch race, so we leave
1922 * everything else as they are. We are called for entries whose
c21d39d7 1923 * ce_stat_data.sd_mtime match the index file mtime.
c70115b4
JH
1924 *
1925 * Note that this actually does not do much for gitlinks, for
1926 * which ce_match_stat_basic() always goes to the actual
1927 * contents. The caller checks with is_racy_timestamp() which
1928 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
1929 */
1930 struct stat st;
1931
1932 if (lstat(ce->name, &st) < 0)
1933 return;
1934 if (ce_match_stat_basic(ce, &st))
1935 return;
1936 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
1937 /* This is "racily clean"; smudge it. Note that this
1938 * is a tricky code. At first glance, it may appear
1939 * that it can break with this sequence:
1940 *
1941 * $ echo xyzzy >frotz
1942 * $ git-update-index --add frotz
1943 * $ : >frotz
1944 * $ sleep 3
1945 * $ echo filfre >nitfol
1946 * $ git-update-index --add nitfol
1947 *
b7e58b17 1948 * but it does not. When the second update-index runs,
4b3511b0
JH
1949 * it notices that the entry "frotz" has the same timestamp
1950 * as index, and if we were to smudge it by resetting its
1951 * size to zero here, then the object name recorded
1952 * in index is the 6-byte file but the cached stat information
1953 * becomes zero --- which would then match what we would
a6080a0a 1954 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
1955 *
1956 * However, the second update-index, before calling
1957 * this function, notices that the cached size is 6
1958 * bytes and what is on the filesystem is an empty
1959 * file, and never calls us, so the cached size information
1960 * for "frotz" stays 6 which does not match the filesystem.
1961 */
c21d39d7 1962 ce->ce_stat_data.sd_size = 0;
407c8eb0
JH
1963 }
1964}
1965
f136f7bf
JH
1966/* Copy miscellaneous fields but not the name */
1967static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1968 struct cache_entry *ce)
7a51ed66 1969{
b60e188c
TG
1970 short flags;
1971
c21d39d7
MH
1972 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1973 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1974 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1975 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1976 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1977 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
7a51ed66 1978 ondisk->mode = htonl(ce->ce_mode);
c21d39d7
MH
1979 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1980 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1981 ondisk->size = htonl(ce->ce_stat_data.sd_size);
99d1a986 1982 hashcpy(ondisk->sha1, ce->oid.hash);
b60e188c 1983
ce51bf09 1984 flags = ce->ce_flags & ~CE_NAMEMASK;
b60e188c
TG
1985 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1986 ondisk->flags = htons(flags);
06aaaa0b
NTND
1987 if (ce->ce_flags & CE_EXTENDED) {
1988 struct ondisk_cache_entry_extended *ondisk2;
1989 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1990 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
f136f7bf 1991 return ondisk2->name;
06aaaa0b 1992 }
f136f7bf
JH
1993 else {
1994 return ondisk->name;
1995 }
1996}
1997
9d227781
JH
1998static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1999 struct strbuf *previous_name)
f136f7bf 2000{
9d227781
JH
2001 int size;
2002 struct ondisk_cache_entry *ondisk;
b3c96fb1 2003 int saved_namelen = saved_namelen; /* compiler workaround */
f136f7bf
JH
2004 char *name;
2005 int result;
2006
b3c96fb1
NTND
2007 if (ce->ce_flags & CE_STRIP_NAME) {
2008 saved_namelen = ce_namelen(ce);
2009 ce->ce_namelen = 0;
2010 }
2011
9d227781
JH
2012 if (!previous_name) {
2013 size = ondisk_ce_size(ce);
2014 ondisk = xcalloc(1, size);
2015 name = copy_cache_entry_to_ondisk(ondisk, ce);
2016 memcpy(name, ce->name, ce_namelen(ce));
2017 } else {
2018 int common, to_remove, prefix_size;
2019 unsigned char to_remove_vi[16];
2020 for (common = 0;
2021 (ce->name[common] &&
2022 common < previous_name->len &&
2023 ce->name[common] == previous_name->buf[common]);
2024 common++)
2025 ; /* still matching */
2026 to_remove = previous_name->len - common;
2027 prefix_size = encode_varint(to_remove, to_remove_vi);
2028
2029 if (ce->ce_flags & CE_EXTENDED)
2030 size = offsetof(struct ondisk_cache_entry_extended, name);
2031 else
2032 size = offsetof(struct ondisk_cache_entry, name);
2033 size += prefix_size + (ce_namelen(ce) - common + 1);
2034
2035 ondisk = xcalloc(1, size);
2036 name = copy_cache_entry_to_ondisk(ondisk, ce);
2037 memcpy(name, to_remove_vi, prefix_size);
2038 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
2039
2040 strbuf_splice(previous_name, common, to_remove,
2041 ce->name + common, ce_namelen(ce) - common);
06aaaa0b 2042 }
b3c96fb1
NTND
2043 if (ce->ce_flags & CE_STRIP_NAME) {
2044 ce->ce_namelen = saved_namelen;
2045 ce->ce_flags &= ~CE_STRIP_NAME;
2046 }
7a51ed66 2047
59efba64
JN
2048 result = ce_write(c, fd, ondisk, size);
2049 free(ondisk);
2050 return result;
7a51ed66
LT
2051}
2052
426ddeea
YM
2053/*
2054 * This function verifies if index_state has the correct sha1 of the
2055 * index file. Don't die if we have any other failure, just return 0.
2056 */
2057static int verify_index_from(const struct index_state *istate, const char *path)
2058{
2059 int fd;
2060 ssize_t n;
2061 struct stat st;
2062 unsigned char sha1[20];
2063
2064 if (!istate->initialized)
2065 return 0;
2066
2067 fd = open(path, O_RDONLY);
2068 if (fd < 0)
2069 return 0;
2070
2071 if (fstat(fd, &st))
2072 goto out;
2073
2074 if (st.st_size < sizeof(struct cache_header) + 20)
2075 goto out;
2076
2077 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2078 if (n != 20)
2079 goto out;
2080
2081 if (hashcmp(istate->sha1, sha1))
2082 goto out;
2083
2084 close(fd);
2085 return 1;
2086
2087out:
2088 close(fd);
2089 return 0;
2090}
2091
2092static int verify_index(const struct index_state *istate)
2093{
2094 return verify_index_from(istate, get_index_file());
2095}
2096
483fbe2b
JH
2097static int has_racy_timestamp(struct index_state *istate)
2098{
2099 int entries = istate->cache_nr;
2100 int i;
2101
2102 for (i = 0; i < entries; i++) {
2103 struct cache_entry *ce = istate->cache[i];
2104 if (is_racy_timestamp(istate, ce))
2105 return 1;
2106 }
2107 return 0;
2108}
2109
ccdc4ec3 2110/*
98e023de 2111 * Opportunistically update the index but do not complain if we can't
ccdc4ec3
JH
2112 */
2113void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2114{
483fbe2b 2115 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
788cef81 2116 verify_index(istate) &&
03b86647 2117 write_locked_index(istate, lockfile, COMMIT_LOCK))
ccdc4ec3
JH
2118 rollback_lock_file(lockfile);
2119}
2120
c18b80a0
NTND
2121static int do_write_index(struct index_state *istate, int newfd,
2122 int strip_extensions)
197ee8c9 2123{
9126f009 2124 git_SHA_CTX c;
197ee8c9 2125 struct cache_header hdr;
9d227781 2126 int i, err, removed, extended, hdr_version;
4aab5b46
JH
2127 struct cache_entry **cache = istate->cache;
2128 int entries = istate->cache_nr;
e1afca4f 2129 struct stat st;
9d227781 2130 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
025a0709 2131
06aaaa0b 2132 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 2133 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 2134 removed++;
197ee8c9 2135
06aaaa0b
NTND
2136 /* reduce extended entries if possible */
2137 cache[i]->ce_flags &= ~CE_EXTENDED;
2138 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2139 extended++;
2140 cache[i]->ce_flags |= CE_EXTENDED;
2141 }
2142 }
2143
d6e3c181 2144 if (!istate->version) {
136347d7 2145 istate->version = get_index_format_default();
d6e3c181
NTND
2146 if (getenv("GIT_TEST_SPLIT_INDEX"))
2147 init_split_index(istate);
2148 }
9d227781
JH
2149
2150 /* demote version 3 to version 2 when the latter suffices */
2151 if (istate->version == 3 || istate->version == 2)
2152 istate->version = extended ? 3 : 2;
2153
2154 hdr_version = istate->version;
2155
ccc4feb5 2156 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
9d227781 2157 hdr.hdr_version = htonl(hdr_version);
025a0709 2158 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 2159
9126f009 2160 git_SHA1_Init(&c);
ca9be054 2161 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
2162 return -1;
2163
9d227781 2164 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
197ee8c9
LT
2165 for (i = 0; i < entries; i++) {
2166 struct cache_entry *ce = cache[i];
7a51ed66 2167 if (ce->ce_flags & CE_REMOVE)
aa16021e 2168 continue;
e06c43c7 2169 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 2170 ce_smudge_racily_clean_entry(ce);
99d1a986 2171 if (is_null_oid(&ce->oid)) {
83bd7437
JK
2172 static const char msg[] = "cache entry has null sha1: %s";
2173 static int allow = -1;
2174
2175 if (allow < 0)
2176 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2177 if (allow)
2178 warning(msg, ce->name);
2179 else
2180 return error(msg, ce->name);
2181 }
9d227781 2182 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
197ee8c9
LT
2183 return -1;
2184 }
9d227781 2185 strbuf_release(&previous_name_buf);
1af1c2b6 2186
bad68ec9 2187 /* Write extension data here */
c18b80a0 2188 if (!strip_extensions && istate->split_index) {
5fc2fc8f
NTND
2189 struct strbuf sb = STRBUF_INIT;
2190
2191 err = write_link_extension(&sb, istate) < 0 ||
2192 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2193 sb.len) < 0 ||
2194 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2195 strbuf_release(&sb);
2196 if (err)
2197 return -1;
2198 }
c18b80a0 2199 if (!strip_extensions && istate->cache_tree) {
f285a2d7 2200 struct strbuf sb = STRBUF_INIT;
1dffb8fa 2201
1dffb8fa
PH
2202 cache_tree_write(&sb, istate->cache_tree);
2203 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2204 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2205 strbuf_release(&sb);
2206 if (err)
bad68ec9 2207 return -1;
bad68ec9 2208 }
c18b80a0 2209 if (!strip_extensions && istate->resolve_undo) {
cfc5789a
JH
2210 struct strbuf sb = STRBUF_INIT;
2211
2212 resolve_undo_write(&sb, istate->resolve_undo);
2213 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2214 sb.len) < 0
2215 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2216 strbuf_release(&sb);
2217 if (err)
2218 return -1;
2219 }
83c094ad
NTND
2220 if (!strip_extensions && istate->untracked) {
2221 struct strbuf sb = STRBUF_INIT;
2222
2223 write_untracked_extension(&sb, istate->untracked);
2224 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2225 sb.len) < 0 ||
2226 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2227 strbuf_release(&sb);
2228 if (err)
2229 return -1;
2230 }
e1afca4f 2231
e93021b2 2232 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
e1afca4f 2233 return -1;
5bcf109c
KB
2234 istate->timestamp.sec = (unsigned int)st.st_mtime;
2235 istate->timestamp.nsec = ST_MTIME_NSEC(st);
e1afca4f 2236 return 0;
197ee8c9 2237}
e46bbcf6 2238
626f35c8
NTND
2239void set_alternate_index_output(const char *name)
2240{
2241 alternate_index_output = name;
2242}
2243
2244static int commit_locked_index(struct lock_file *lk)
2245{
751baced
MH
2246 if (alternate_index_output)
2247 return commit_lock_file_to(lk, alternate_index_output);
2248 else
626f35c8 2249 return commit_lock_file(lk);
626f35c8
NTND
2250}
2251
03b86647
NTND
2252static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2253 unsigned flags)
2254{
c99a4c2d 2255 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
03b86647
NTND
2256 if (ret)
2257 return ret;
2258 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2259 (COMMIT_LOCK | CLOSE_LOCK));
2260 if (flags & COMMIT_LOCK)
2261 return commit_locked_index(lock);
2262 else if (flags & CLOSE_LOCK)
2263 return close_lock_file(lock);
2264 else
2265 return ret;
2266}
2267
5fc2fc8f
NTND
2268static int write_split_index(struct index_state *istate,
2269 struct lock_file *lock,
2270 unsigned flags)
2271{
2272 int ret;
2273 prepare_to_write_split_index(istate);
2274 ret = do_write_locked_index(istate, lock, flags);
2275 finish_writing_split_index(istate);
2276 return ret;
2277}
2278
b9683722
CC
2279static const char *shared_index_expire = "2.weeks.ago";
2280
2281static unsigned long get_shared_index_expire_date(void)
2282{
2283 static unsigned long shared_index_expire_date;
2284 static int shared_index_expire_date_prepared;
2285
2286 if (!shared_index_expire_date_prepared) {
2287 git_config_get_expiry("splitindex.sharedindexexpire",
2288 &shared_index_expire);
2289 shared_index_expire_date = approxidate(shared_index_expire);
2290 shared_index_expire_date_prepared = 1;
2291 }
2292
2293 return shared_index_expire_date;
2294}
2295
2296static int should_delete_shared_index(const char *shared_index_path)
2297{
2298 struct stat st;
2299 unsigned long expiration;
2300
2301 /* Check timestamp */
2302 expiration = get_shared_index_expire_date();
2303 if (!expiration)
2304 return 0;
2305 if (stat(shared_index_path, &st))
2306 return error_errno(_("could not stat '%s"), shared_index_path);
2307 if (st.st_mtime > expiration)
2308 return 0;
2309
2310 return 1;
2311}
2312
2313static int clean_shared_index_files(const char *current_hex)
2314{
2315 struct dirent *de;
2316 DIR *dir = opendir(get_git_dir());
2317
2318 if (!dir)
2319 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2320
2321 while ((de = readdir(dir)) != NULL) {
2322 const char *sha1_hex;
2323 const char *shared_index_path;
2324 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2325 continue;
2326 if (!strcmp(sha1_hex, current_hex))
2327 continue;
2328 shared_index_path = git_path("%s", de->d_name);
2329 if (should_delete_shared_index(shared_index_path) > 0 &&
2330 unlink(shared_index_path))
2331 warning_errno(_("unable to unlink: %s"), shared_index_path);
2332 }
2333 closedir(dir);
2334
2335 return 0;
2336}
2337
f6ecc62d 2338static struct tempfile temporary_sharedindex;
c18b80a0 2339
a0a96756
NTND
2340static int write_shared_index(struct index_state *istate,
2341 struct lock_file *lock, unsigned flags)
c18b80a0
NTND
2342{
2343 struct split_index *si = istate->split_index;
c18b80a0
NTND
2344 int fd, ret;
2345
f6ecc62d 2346 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
a0a96756 2347 if (fd < 0) {
a0a96756
NTND
2348 hashclr(si->base_sha1);
2349 return do_write_locked_index(istate, lock, flags);
2350 }
c18b80a0
NTND
2351 move_cache_to_base_index(istate);
2352 ret = do_write_index(si->base, fd, 1);
c18b80a0 2353 if (ret) {
f6ecc62d 2354 delete_tempfile(&temporary_sharedindex);
c18b80a0
NTND
2355 return ret;
2356 }
f6ecc62d
MH
2357 ret = rename_tempfile(&temporary_sharedindex,
2358 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
b9683722 2359 if (!ret) {
c18b80a0 2360 hashcpy(si->base_sha1, si->base->sha1);
b9683722
CC
2361 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2362 }
2363
c18b80a0
NTND
2364 return ret;
2365}
2366
e6a1dd77
CC
2367static const int default_max_percent_split_change = 20;
2368
2369static int too_many_not_shared_entries(struct index_state *istate)
2370{
2371 int i, not_shared = 0;
2372 int max_split = git_config_get_max_percent_split_change();
2373
2374 switch (max_split) {
2375 case -1:
2376 /* not or badly configured: use the default value */
2377 max_split = default_max_percent_split_change;
2378 break;
2379 case 0:
2380 return 1; /* 0% means always write a new shared index */
2381 case 100:
2382 return 0; /* 100% means never write a new shared index */
2383 default:
2384 break; /* just use the configured value */
2385 }
2386
2387 /* Count not shared entries */
2388 for (i = 0; i < istate->cache_nr; i++) {
2389 struct cache_entry *ce = istate->cache[i];
2390 if (!ce->index)
2391 not_shared++;
2392 }
2393
2394 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2395}
2396
03b86647
NTND
2397int write_locked_index(struct index_state *istate, struct lock_file *lock,
2398 unsigned flags)
2399{
0d59ffb4 2400 int new_shared_index, ret;
5fc2fc8f
NTND
2401 struct split_index *si = istate->split_index;
2402
5165dd59
NTND
2403 if (!si || alternate_index_output ||
2404 (istate->cache_changed & ~EXTMASK)) {
5fc2fc8f
NTND
2405 if (si)
2406 hashclr(si->base_sha1);
2407 return do_write_locked_index(istate, lock, flags);
2408 }
2409
d6e3c181
NTND
2410 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2411 int v = si->base_sha1[0];
2412 if ((v & 15) < 6)
2413 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2414 }
e6a1dd77
CC
2415 if (too_many_not_shared_entries(istate))
2416 istate->cache_changed |= SPLIT_INDEX_ORDERED;
0d59ffb4
CC
2417
2418 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2419
2420 if (new_shared_index) {
2421 ret = write_shared_index(istate, lock, flags);
c18b80a0
NTND
2422 if (ret)
2423 return ret;
2424 }
2425
0d59ffb4
CC
2426 ret = write_split_index(istate, lock, flags);
2427
2428 /* Freshen the shared index only if the split-index was written */
2429 if (!ret && !new_shared_index)
2430 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2431
2432 return ret;
03b86647
NTND
2433}
2434
e46bbcf6
MV
2435/*
2436 * Read the index file that is potentially unmerged into given
63e8dc5b 2437 * index_state, dropping any unmerged entries. Returns true if
e46bbcf6
MV
2438 * the index is unmerged. Callers who want to refuse to work
2439 * from an unmerged state can call this and check its return value,
2440 * instead of calling read_cache().
2441 */
2442int read_index_unmerged(struct index_state *istate)
2443{
2444 int i;
d1a43f2a 2445 int unmerged = 0;
e46bbcf6
MV
2446
2447 read_index(istate);
e46bbcf6
MV
2448 for (i = 0; i < istate->cache_nr; i++) {
2449 struct cache_entry *ce = istate->cache[i];
d1a43f2a
JH
2450 struct cache_entry *new_ce;
2451 int size, len;
2452
2453 if (!ce_stage(ce))
e46bbcf6 2454 continue;
d1a43f2a 2455 unmerged = 1;
68c4f6a5 2456 len = ce_namelen(ce);
d1a43f2a
JH
2457 size = cache_entry_size(len);
2458 new_ce = xcalloc(1, size);
d1a43f2a 2459 memcpy(new_ce->name, ce->name, len);
b60e188c
TG
2460 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2461 new_ce->ce_namelen = len;
d1a43f2a
JH
2462 new_ce->ce_mode = ce->ce_mode;
2463 if (add_index_entry(istate, new_ce, 0))
2464 return error("%s: cannot drop to stage #0",
5699d17e 2465 new_ce->name);
e46bbcf6 2466 }
d1a43f2a 2467 return unmerged;
e46bbcf6 2468}
041aee31 2469
98fa4738
JK
2470/*
2471 * Returns 1 if the path is an "other" path with respect to
2472 * the index; that is, the path is not mentioned in the index at all,
2473 * either as a file, a directory with some files in the index,
2474 * or as an unmerged entry.
2475 *
2476 * We helpfully remove a trailing "/" from directories so that
2477 * the output of read_directory can be used as-is.
2478 */
2479int index_name_is_other(const struct index_state *istate, const char *name,
2480 int namelen)
2481{
2482 int pos;
2483 if (namelen && name[namelen - 1] == '/')
2484 namelen--;
2485 pos = index_name_pos(istate, name, namelen);
2486 if (0 <= pos)
2487 return 0; /* exact match */
2488 pos = -pos - 1;
2489 if (pos < istate->cache_nr) {
2490 struct cache_entry *ce = istate->cache[pos];
2491 if (ce_namelen(ce) == namelen &&
2492 !memcmp(ce->name, name, namelen))
2493 return 0; /* Yup, this one exists unmerged */
2494 }
2495 return 1;
2496}
29fb37b2 2497
87542508
BW
2498void *read_blob_data_from_index(const struct index_state *istate,
2499 const char *path, unsigned long *size)
29fb37b2
LF
2500{
2501 int pos, len;
2502 unsigned long sz;
2503 enum object_type type;
2504 void *data;
2505
2506 len = strlen(path);
2507 pos = index_name_pos(istate, path, len);
2508 if (pos < 0) {
2509 /*
2510 * We might be in the middle of a merge, in which
2511 * case we would read stage #2 (ours).
2512 */
2513 int i;
2514 for (i = -pos - 1;
2515 (pos < 0 && i < istate->cache_nr &&
2516 !strcmp(istate->cache[i]->name, path));
2517 i++)
2518 if (ce_stage(istate->cache[i]) == 2)
2519 pos = i;
2520 }
2521 if (pos < 0)
2522 return NULL;
99d1a986 2523 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
29fb37b2
LF
2524 if (!data || type != OBJ_BLOB) {
2525 free(data);
2526 return NULL;
2527 }
ff366825
LF
2528 if (size)
2529 *size = sz;
29fb37b2
LF
2530 return data;
2531}
38612532
MH
2532
2533void stat_validity_clear(struct stat_validity *sv)
2534{
2535 free(sv->sd);
2536 sv->sd = NULL;
2537}
2538
2539int stat_validity_check(struct stat_validity *sv, const char *path)
2540{
2541 struct stat st;
2542
2543 if (stat(path, &st) < 0)
2544 return sv->sd == NULL;
2545 if (!sv->sd)
2546 return 0;
2547 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2548}
2549
2550void stat_validity_update(struct stat_validity *sv, int fd)
2551{
2552 struct stat st;
2553
2554 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2555 stat_validity_clear(sv);
2556 else {
2557 if (!sv->sd)
2558 sv->sd = xcalloc(1, sizeof(struct stat_data));
2559 fill_stat_data(sv->sd, &st);
2560 }
2561}