read-cache: speed up add_index_entry during checkout
[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?
913 */
4aab5b46
JH
914static int has_dir_name(struct index_state *istate,
915 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
916{
917 int retval = 0;
b155725d 918 int stage = ce_stage(ce);
12676608
LT
919 const char *name = ce->name;
920 const char *slash = name + ce_namelen(ce);
0f1e4f04 921
12676608
LT
922 for (;;) {
923 int len;
0f1e4f04 924
12676608
LT
925 for (;;) {
926 if (*--slash == '/')
927 break;
928 if (slash <= ce->name)
929 return retval;
930 }
931 len = slash - name;
0f1e4f04 932
b60e188c 933 pos = index_name_stage_pos(istate, name, len, stage);
12676608 934 if (pos >= 0) {
21cd8d00
JH
935 /*
936 * Found one, but not so fast. This could
937 * be a marker that says "I was here, but
938 * I am being removed". Such an entry is
939 * not a part of the resulting tree, and
940 * it is Ok to have a directory at the same
941 * path.
942 */
077c48df 943 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
944 retval = -1;
945 if (!ok_to_replace)
946 break;
4aab5b46 947 remove_index_entry_at(istate, pos);
21cd8d00
JH
948 continue;
949 }
12676608 950 }
21cd8d00
JH
951 else
952 pos = -pos-1;
12676608
LT
953
954 /*
955 * Trivial optimization: if we find an entry that
956 * already matches the sub-directory, then we know
b155725d 957 * we're ok, and we can exit.
12676608 958 */
4aab5b46
JH
959 while (pos < istate->cache_nr) {
960 struct cache_entry *p = istate->cache[pos];
b155725d
JH
961 if ((ce_namelen(p) <= len) ||
962 (p->name[len] != '/') ||
963 memcmp(p->name, name, len))
964 break; /* not our subdirectory */
077c48df
JH
965 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
966 /*
967 * p is at the same stage as our entry, and
b155725d
JH
968 * is a subdirectory of what we are looking
969 * at, so we cannot have conflicts at our
970 * level or anything shorter.
971 */
972 return retval;
973 pos++;
192268c1 974 }
0f1e4f04 975 }
12676608
LT
976 return retval;
977}
978
979/* We may be in a situation where we already have path/file and path
980 * is being added, or we already have path and path/file is being
981 * added. Either one would result in a nonsense tree that has path
982 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 983 *
12676608
LT
984 * If ok-to-replace is specified, we remove the conflicting entries
985 * from the cache so the caller should recompute the insert position.
986 * When this happens, we return non-zero.
987 */
4aab5b46
JH
988static int check_file_directory_conflict(struct index_state *istate,
989 const struct cache_entry *ce,
990 int pos, int ok_to_replace)
12676608 991{
21cd8d00
JH
992 int retval;
993
994 /*
995 * When ce is an "I am going away" entry, we allow it to be added
996 */
7a51ed66 997 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
998 return 0;
999
12676608
LT
1000 /*
1001 * We check if the path is a sub-path of a subsequent pathname
1002 * first, since removing those will not change the position
21cd8d00 1003 * in the array.
12676608 1004 */
4aab5b46 1005 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 1006
12676608
LT
1007 /*
1008 * Then check if the path might have a clashing sub-directory
1009 * before it.
1010 */
4aab5b46 1011 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
1012}
1013
af3785dc 1014static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
1015{
1016 int pos;
192268c1
JH
1017 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1018 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 1019 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 1020 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 1021
ce7c614b
NTND
1022 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1023 cache_tree_invalidate_path(istate, ce->name);
e5494631
JH
1024
1025 /*
1026 * If this entry's path sorts after the last entry in the index,
1027 * we can avoid searching for it.
1028 */
1029 if (istate->cache_nr > 0 &&
1030 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1031 pos = -istate->cache_nr - 1;
1032 else
1033 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
197ee8c9 1034
3e09cdfd 1035 /* existing match? Just replace it. */
76e7f4ec 1036 if (pos >= 0) {
39425819
JH
1037 if (!new_only)
1038 replace_index_entry(istate, pos, ce);
197ee8c9
LT
1039 return 0;
1040 }
76e7f4ec 1041 pos = -pos-1;
197ee8c9 1042
ffcc9ba7
NTND
1043 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1044 untracked_cache_add_to_index(istate, ce->name);
e931371a 1045
7b937ca3
LT
1046 /*
1047 * Inserting a merged entry ("stage 0") into the index
1048 * will always replace all non-merged entries..
1049 */
4aab5b46
JH
1050 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1051 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 1052 ok_to_add = 1;
4aab5b46 1053 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
1054 break;
1055 }
1056 }
1057
121481ab
LT
1058 if (!ok_to_add)
1059 return -1;
8dcf39c4 1060 if (!verify_path(ce->name))
7e7abea9 1061 return error("Invalid path '%s'", ce->name);
121481ab 1062
3e09cdfd 1063 if (!skip_df_check &&
4aab5b46 1064 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 1065 if (!ok_to_replace)
4aab5b46
JH
1066 return error("'%s' appears as both a file and as a directory",
1067 ce->name);
b60e188c 1068 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
192268c1
JH
1069 pos = -pos-1;
1070 }
af3785dc
JH
1071 return pos + 1;
1072}
1073
1074int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1075{
1076 int pos;
1077
1078 if (option & ADD_CACHE_JUST_APPEND)
1079 pos = istate->cache_nr;
1080 else {
1081 int ret;
1082 ret = add_index_entry_with_check(istate, ce, option);
1083 if (ret <= 0)
1084 return ret;
1085 pos = ret - 1;
1086 }
0f1e4f04 1087
197ee8c9 1088 /* Make sure the array is big enough .. */
999f5660 1089 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
197ee8c9
LT
1090
1091 /* Add it in.. */
4aab5b46 1092 istate->cache_nr++;
af3785dc 1093 if (istate->cache_nr > pos + 1)
4aab5b46
JH
1094 memmove(istate->cache + pos + 1,
1095 istate->cache + pos,
1096 (istate->cache_nr - pos - 1) * sizeof(ce));
cf558704 1097 set_index_entry(istate, pos, ce);
e636a7b4 1098 istate->cache_changed |= CE_ENTRY_ADDED;
197ee8c9
LT
1099 return 0;
1100}
1101
405e5b2f
LT
1102/*
1103 * "refresh" does not calculate a new sha1 file or bring the
1104 * cache up-to-date for mode/content changes. But what it
1105 * _does_ do is to "re-match" the stat information of a file
1106 * with the cache, so that you can refresh the cache for a
1107 * file that hasn't been changed but where the stat entry is
1108 * out of date.
1109 *
1110 * For example, you'd want to do this after doing a "git-read-tree",
1111 * to link up the stat cache details with the proper files.
1112 */
4aab5b46 1113static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da 1114 struct cache_entry *ce,
d05e6970
JK
1115 unsigned int options, int *err,
1116 int *changed_ret)
405e5b2f
LT
1117{
1118 struct stat st;
1119 struct cache_entry *updated;
1120 int changed, size;
25762726 1121 int refresh = options & CE_MATCH_REFRESH;
4bd5b7da 1122 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1123 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
2e2e7ec1 1124 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
405e5b2f 1125
25762726 1126 if (!refresh || ce_uptodate(ce))
eadb5831
JH
1127 return ce;
1128
aa9349d4 1129 /*
56cac48c
NTND
1130 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1131 * that the change to the work tree does not matter and told
1132 * us not to worry.
aa9349d4 1133 */
56cac48c
NTND
1134 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1135 ce_mark_uptodate(ce);
1136 return ce;
1137 }
aa9349d4
MSO
1138 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1139 ce_mark_uptodate(ce);
1140 return ce;
1141 }
1142
ccad42d4
RS
1143 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1144 if (ignore_missing)
1145 return ce;
1146 if (err)
1147 *err = ENOENT;
1148 return NULL;
1149 }
1150
8fd2cb40 1151 if (lstat(ce->name, &st) < 0) {
2e2e7ec1
BK
1152 if (ignore_missing && errno == ENOENT)
1153 return ce;
ec0cc704
JH
1154 if (err)
1155 *err = errno;
8fd2cb40
JS
1156 return NULL;
1157 }
405e5b2f 1158
4bd5b7da 1159 changed = ie_match_stat(istate, ce, &st, options);
d05e6970
JK
1160 if (changed_ret)
1161 *changed_ret = changed;
405e5b2f 1162 if (!changed) {
4bd5b7da
JH
1163 /*
1164 * The path is unchanged. If we were told to ignore
1165 * valid bit, then we did the actual stat check and
1166 * found that the entry is unmodified. If the entry
1167 * is not marked VALID, this is the place to mark it
1168 * valid again, under "assume unchanged" mode.
1169 */
1170 if (ignore_valid && assume_unchanged &&
7a51ed66 1171 !(ce->ce_flags & CE_VALID))
405e5b2f 1172 ; /* mark this one VALID again */
eadb5831
JH
1173 else {
1174 /*
1175 * We do not mark the index itself "modified"
1176 * because CE_UPTODATE flag is in-core only;
1177 * we are not going to write this change out.
1178 */
125fd984
JH
1179 if (!S_ISGITLINK(ce->ce_mode))
1180 ce_mark_uptodate(ce);
8fd2cb40 1181 return ce;
eadb5831 1182 }
405e5b2f
LT
1183 }
1184
4bd5b7da 1185 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1186 if (err)
1187 *err = EINVAL;
8fd2cb40
JS
1188 return NULL;
1189 }
405e5b2f
LT
1190
1191 size = ce_size(ce);
1192 updated = xmalloc(size);
1193 memcpy(updated, ce, size);
1194 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1195 /*
1196 * If ignore_valid is not set, we should leave CE_VALID bit
1197 * alone. Otherwise, paths marked with --no-assume-unchanged
1198 * (i.e. things to be edited) will reacquire CE_VALID bit
1199 * automatically, which is not really what we want.
405e5b2f 1200 */
4bd5b7da 1201 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1202 !(ce->ce_flags & CE_VALID))
1203 updated->ce_flags &= ~CE_VALID;
405e5b2f 1204
e636a7b4 1205 /* istate->cache_changed is updated in the caller */
405e5b2f
LT
1206 return updated;
1207}
1208
3deffc52 1209static void show_file(const char * fmt, const char * name, int in_porcelain,
046613c5 1210 int * first, const char *header_msg)
3deffc52
MM
1211{
1212 if (in_porcelain && *first && header_msg) {
1213 printf("%s\n", header_msg);
cd2b8ae9 1214 *first = 0;
3deffc52
MM
1215 }
1216 printf(fmt, name);
1217}
1218
9b2d6149
NTND
1219int refresh_index(struct index_state *istate, unsigned int flags,
1220 const struct pathspec *pathspec,
046613c5 1221 char *seen, const char *header_msg)
405e5b2f
LT
1222{
1223 int i;
1224 int has_errors = 0;
1225 int really = (flags & REFRESH_REALLY) != 0;
1226 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1227 int quiet = (flags & REFRESH_QUIET) != 0;
1228 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1229 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1230 int first = 1;
1231 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
25762726
BK
1232 unsigned int options = (CE_MATCH_REFRESH |
1233 (really ? CE_MATCH_IGNORE_VALID : 0) |
2e2e7ec1 1234 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
4bd4e730 1235 const char *modified_fmt;
73b7eae6
JK
1236 const char *deleted_fmt;
1237 const char *typechange_fmt;
1238 const char *added_fmt;
4bd4e730 1239 const char *unmerged_fmt;
405e5b2f 1240
4bd4e730 1241 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
73b7eae6
JK
1242 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1243 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1244 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
4bd4e730 1245 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
4aab5b46 1246 for (i = 0; i < istate->cache_nr; i++) {
405e5b2f 1247 struct cache_entry *ce, *new;
ec0cc704 1248 int cache_errno = 0;
73b7eae6 1249 int changed = 0;
3d1f148c 1250 int filtered = 0;
ec0cc704 1251
4aab5b46 1252 ce = istate->cache[i];
5fdeacb0
JS
1253 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1254 continue;
1255
429bb40a 1256 if (pathspec && !ce_path_match(ce, pathspec, seen))
3d1f148c
JH
1257 filtered = 1;
1258
405e5b2f 1259 if (ce_stage(ce)) {
4aab5b46
JH
1260 while ((i < istate->cache_nr) &&
1261 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1262 i++;
1263 i--;
1264 if (allow_unmerged)
1265 continue;
3d1f148c
JH
1266 if (!filtered)
1267 show_file(unmerged_fmt, ce->name, in_porcelain,
1268 &first, header_msg);
405e5b2f
LT
1269 has_errors = 1;
1270 continue;
1271 }
1272
3d1f148c 1273 if (filtered)
d616813d
AJ
1274 continue;
1275
73b7eae6 1276 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
8fd2cb40 1277 if (new == ce)
405e5b2f 1278 continue;
8fd2cb40 1279 if (!new) {
73b7eae6
JK
1280 const char *fmt;
1281
8fd2cb40 1282 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1283 /* If we are doing --really-refresh that
1284 * means the index is not valid anymore.
1285 */
7a51ed66 1286 ce->ce_flags &= ~CE_VALID;
078a58e8 1287 ce->ce_flags |= CE_UPDATE_IN_BASE;
e636a7b4 1288 istate->cache_changed |= CE_ENTRY_CHANGED;
405e5b2f
LT
1289 }
1290 if (quiet)
1291 continue;
73b7eae6
JK
1292
1293 if (cache_errno == ENOENT)
1294 fmt = deleted_fmt;
895ff3b2 1295 else if (ce_intent_to_add(ce))
73b7eae6
JK
1296 fmt = added_fmt; /* must be before other checks */
1297 else if (changed & TYPE_CHANGED)
1298 fmt = typechange_fmt;
1299 else
1300 fmt = modified_fmt;
1301 show_file(fmt,
1302 ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1303 has_errors = 1;
1304 continue;
1305 }
cf558704
LT
1306
1307 replace_index_entry(istate, i, new);
405e5b2f
LT
1308 }
1309 return has_errors;
1310}
1311
1335d76e 1312struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
25762726 1313 unsigned int options)
ec0cc704 1314{
25762726 1315 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
ec0cc704
JH
1316}
1317
db3b313c
JH
1318
1319/*****************************************************************
1320 * Index File I/O
1321 *****************************************************************/
1322
9d227781
JH
1323#define INDEX_FORMAT_DEFAULT 3
1324
136347d7
TG
1325static unsigned int get_index_format_default(void)
1326{
1327 char *envversion = getenv("GIT_INDEX_VERSION");
3c09d684 1328 char *endp;
b27a5720 1329 int value;
3c09d684
TG
1330 unsigned int version = INDEX_FORMAT_DEFAULT;
1331
136347d7 1332 if (!envversion) {
b27a5720
TA
1333 if (!git_config_get_int("index.version", &value))
1334 version = value;
3c09d684
TG
1335 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1336 warning(_("index.version set, but the value is invalid.\n"
136347d7 1337 "Using version %i"), INDEX_FORMAT_DEFAULT);
3c09d684 1338 return INDEX_FORMAT_DEFAULT;
136347d7
TG
1339 }
1340 return version;
1341 }
3c09d684
TG
1342
1343 version = strtoul(envversion, &endp, 10);
1344 if (*endp ||
1345 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1346 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1347 "Using version %i"), INDEX_FORMAT_DEFAULT);
1348 version = INDEX_FORMAT_DEFAULT;
1349 }
1350 return version;
136347d7
TG
1351}
1352
db3b313c
JH
1353/*
1354 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1355 * Again - this is just a (very strong in practice) heuristic that
1356 * the inode hasn't changed.
1357 *
1358 * We save the fields in big-endian order to allow using the
1359 * index file over NFS transparently.
1360 */
1361struct ondisk_cache_entry {
1362 struct cache_time ctime;
1363 struct cache_time mtime;
7800c1eb
TG
1364 uint32_t dev;
1365 uint32_t ino;
1366 uint32_t mode;
1367 uint32_t uid;
1368 uint32_t gid;
1369 uint32_t size;
db3b313c 1370 unsigned char sha1[20];
7800c1eb 1371 uint16_t flags;
db3b313c
JH
1372 char name[FLEX_ARRAY]; /* more */
1373};
1374
1375/*
1376 * This struct is used when CE_EXTENDED bit is 1
1377 * The struct must match ondisk_cache_entry exactly from
1378 * ctime till flags
1379 */
1380struct ondisk_cache_entry_extended {
1381 struct cache_time ctime;
1382 struct cache_time mtime;
7800c1eb
TG
1383 uint32_t dev;
1384 uint32_t ino;
1385 uint32_t mode;
1386 uint32_t uid;
1387 uint32_t gid;
1388 uint32_t size;
db3b313c 1389 unsigned char sha1[20];
7800c1eb
TG
1390 uint16_t flags;
1391 uint16_t flags2;
db3b313c
JH
1392 char name[FLEX_ARRAY]; /* more */
1393};
1394
6c9cd161 1395/* These are only used for v3 or lower */
db3b313c
JH
1396#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1397#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1398#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1399#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1400 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1401 ondisk_cache_entry_size(ce_namelen(ce)))
1402
bad68ec9 1403static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163 1404{
9126f009 1405 git_SHA_CTX c;
bad68ec9 1406 unsigned char sha1[20];
0136bac9 1407 int hdr_version;
e83c5163 1408
ccc4feb5 1409 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1410 return error("bad signature");
0136bac9 1411 hdr_version = ntohl(hdr->hdr_version);
b82a7b5b 1412 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
0136bac9 1413 return error("bad index version %d", hdr_version);
9126f009
NP
1414 git_SHA1_Init(&c);
1415 git_SHA1_Update(&c, hdr, size - 20);
1416 git_SHA1_Final(sha1, &c);
a89fccd2 1417 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
ca9be054 1418 return error("bad index file sha1 signature");
e83c5163
LT
1419 return 0;
1420}
1421
4aab5b46
JH
1422static int read_index_extension(struct index_state *istate,
1423 const char *ext, void *data, unsigned long sz)
bad68ec9
JH
1424{
1425 switch (CACHE_EXT(ext)) {
1426 case CACHE_EXT_TREE:
4aab5b46 1427 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1428 break;
cfc5789a
JH
1429 case CACHE_EXT_RESOLVE_UNDO:
1430 istate->resolve_undo = resolve_undo_read(data, sz);
1431 break;
5fc2fc8f
NTND
1432 case CACHE_EXT_LINK:
1433 if (read_link_extension(istate, data, sz))
1434 return -1;
1435 break;
f9e6c649
NTND
1436 case CACHE_EXT_UNTRACKED:
1437 istate->untracked = read_untracked_extension(data, sz);
1438 break;
bad68ec9
JH
1439 default:
1440 if (*ext < 'A' || 'Z' < *ext)
1441 return error("index uses %.4s extension, which we do not understand",
1442 ext);
1443 fprintf(stderr, "ignoring %.4s extension\n", ext);
1444 break;
1445 }
1446 return 0;
1447}
1448
b3e83cc7 1449int hold_locked_index(struct lock_file *lk, int lock_flags)
216aab1e 1450{
b3e83cc7 1451 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
216aab1e
MH
1452}
1453
4aab5b46 1454int read_index(struct index_state *istate)
8fd2cb40 1455{
4aab5b46 1456 return read_index_from(istate, get_index_file());
8fd2cb40
JS
1457}
1458
3fc22b53
JH
1459static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1460 unsigned int flags,
1461 const char *name,
1462 size_t len)
1463{
1464 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1465
c3d8da57
JK
1466 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1467 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1468 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1469 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1470 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1471 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1472 ce->ce_mode = get_be32(&ondisk->mode);
1473 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1474 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1475 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
b60e188c
TG
1476 ce->ce_flags = flags & ~CE_NAMEMASK;
1477 ce->ce_namelen = len;
5fc2fc8f 1478 ce->index = 0;
99d1a986 1479 hashcpy(ce->oid.hash, ondisk->sha1);
3fc22b53
JH
1480 memcpy(ce->name, name, len);
1481 ce->name[len] = '\0';
1482 return ce;
1483}
1484
6c9cd161
JH
1485/*
1486 * Adjacent cache entries tend to share the leading paths, so it makes
1487 * sense to only store the differences in later entries. In the v4
1488 * on-disk format of the index, each on-disk cache entry stores the
1489 * number of bytes to be stripped from the end of the previous name,
1490 * and the bytes to append to the result, to come up with its name.
1491 */
1492static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1493{
1494 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1495 size_t len = decode_varint(&cp);
1496
1497 if (name->len < len)
1498 die("malformed name field in the index");
1499 strbuf_remove(name, name->len - len, len);
1500 for (ep = cp; *ep; ep++)
1501 ; /* find the end */
1502 strbuf_add(name, cp, ep - cp);
1503 return (const char *)ep + 1 - cp_;
1504}
1505
936f53d0 1506static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
6c9cd161
JH
1507 unsigned long *ent_size,
1508 struct strbuf *previous_name)
7a51ed66 1509{
debed2a6 1510 struct cache_entry *ce;
7fec10b7 1511 size_t len;
06aaaa0b 1512 const char *name;
debed2a6 1513 unsigned int flags;
7fec10b7 1514
7a51ed66 1515 /* On-disk flags are just 16 bits */
c3d8da57 1516 flags = get_be16(&ondisk->flags);
debed2a6 1517 len = flags & CE_NAMEMASK;
7fec10b7 1518
debed2a6 1519 if (flags & CE_EXTENDED) {
06aaaa0b
NTND
1520 struct ondisk_cache_entry_extended *ondisk2;
1521 int extended_flags;
1522 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
c3d8da57 1523 extended_flags = get_be16(&ondisk2->flags2) << 16;
06aaaa0b
NTND
1524 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1525 if (extended_flags & ~CE_EXTENDED_FLAGS)
1526 die("Unknown index entry format %08x", extended_flags);
debed2a6 1527 flags |= extended_flags;
06aaaa0b
NTND
1528 name = ondisk2->name;
1529 }
1530 else
1531 name = ondisk->name;
1532
6c9cd161
JH
1533 if (!previous_name) {
1534 /* v3 and earlier */
1535 if (len == CE_NAMEMASK)
1536 len = strlen(name);
1537 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1538
1539 *ent_size = ondisk_ce_size(ce);
1540 } else {
1541 unsigned long consumed;
1542 consumed = expand_name_field(previous_name, name);
1543 ce = cache_entry_from_ondisk(ondisk, flags,
1544 previous_name->buf,
1545 previous_name->len);
1546
1547 *ent_size = (name - ((char *)ondisk)) + consumed;
1548 }
debed2a6 1549 return ce;
cf558704
LT
1550}
1551
03f15a79 1552static void check_ce_order(struct index_state *istate)
15999d0b 1553{
03f15a79
TG
1554 unsigned int i;
1555
1556 for (i = 1; i < istate->cache_nr; i++) {
1557 struct cache_entry *ce = istate->cache[i - 1];
1558 struct cache_entry *next_ce = istate->cache[i];
1559 int name_compare = strcmp(ce->name, next_ce->name);
1560
1561 if (0 < name_compare)
1562 die("unordered stage entries in index");
1563 if (!name_compare) {
1564 if (!ce_stage(ce))
1565 die("multiple stage entries for merged file '%s'",
1566 ce->name);
1567 if (ce_stage(ce) > ce_stage(next_ce))
1568 die("unordered stage entries for '%s'",
1569 ce->name);
1570 }
15999d0b
JSP
1571 }
1572}
1573
435ec090
CC
1574static void tweak_untracked_cache(struct index_state *istate)
1575{
1576 switch (git_config_get_untracked_cache()) {
1577 case -1: /* keep: do nothing */
1578 break;
1579 case 0: /* false */
1580 remove_untracked_cache(istate);
1581 break;
1582 case 1: /* true */
1583 add_untracked_cache(istate);
1584 break;
1585 default: /* unknown value: do nothing */
1586 break;
1587 }
1588}
1589
43925312
CC
1590static void tweak_split_index(struct index_state *istate)
1591{
1592 switch (git_config_get_split_index()) {
1593 case -1: /* unset: do nothing */
1594 break;
1595 case 0: /* false */
1596 remove_split_index(istate);
1597 break;
1598 case 1: /* true */
1599 add_split_index(istate);
1600 break;
1601 default: /* unknown value: do nothing */
1602 break;
1603 }
1604}
1605
435ec090
CC
1606static void post_read_index_from(struct index_state *istate)
1607{
1608 check_ce_order(istate);
1609 tweak_untracked_cache(istate);
43925312 1610 tweak_split_index(istate);
435ec090
CC
1611}
1612
8fd2cb40 1613/* remember to discard_cache() before reading a different cache! */
3e52f70b 1614int do_read_index(struct index_state *istate, const char *path, int must_exist)
e83c5163
LT
1615{
1616 int fd, i;
1617 struct stat st;
debed2a6 1618 unsigned long src_offset;
e83c5163 1619 struct cache_header *hdr;
7a51ed66
LT
1620 void *mmap;
1621 size_t mmap_size;
6c9cd161 1622 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
e83c5163 1623
913e0e99 1624 if (istate->initialized)
4aab5b46 1625 return istate->cache_nr;
5d1a5c02 1626
fba2f38a
KB
1627 istate->timestamp.sec = 0;
1628 istate->timestamp.nsec = 0;
8fd2cb40 1629 fd = open(path, O_RDONLY);
5d1a5c02 1630 if (fd < 0) {
5fc2fc8f 1631 if (!must_exist && errno == ENOENT)
5d1a5c02 1632 return 0;
5fc2fc8f 1633 die_errno("%s: index file open failed", path);
5d1a5c02 1634 }
e83c5163 1635
3511a377 1636 if (fstat(fd, &st))
d824cbba 1637 die_errno("cannot stat the open index");
3511a377 1638
7a51ed66
LT
1639 mmap_size = xsize_t(st.st_size);
1640 if (mmap_size < sizeof(struct cache_header) + 20)
3511a377
LFC
1641 die("index file smaller than expected");
1642
a1293ef7 1643 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
7a51ed66 1644 if (mmap == MAP_FAILED)
0721c314 1645 die_errno("unable to map index file");
57d84f8d 1646 close(fd);
e83c5163 1647
7a51ed66
LT
1648 hdr = mmap;
1649 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1650 goto unmap;
1651
e93021b2 1652 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
9d227781 1653 istate->version = ntohl(hdr->hdr_version);
4aab5b46
JH
1654 istate->cache_nr = ntohl(hdr->hdr_entries);
1655 istate->cache_alloc = alloc_nr(istate->cache_nr);
c4aa3167 1656 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
913e0e99 1657 istate->initialized = 1;
7a51ed66 1658
9d227781 1659 if (istate->version == 4)
6c9cd161
JH
1660 previous_name = &previous_name_buf;
1661 else
1662 previous_name = NULL;
1663
7a51ed66 1664 src_offset = sizeof(*hdr);
4aab5b46 1665 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 1666 struct ondisk_cache_entry *disk_ce;
4aab5b46 1667 struct cache_entry *ce;
936f53d0 1668 unsigned long consumed;
4aab5b46 1669
7a51ed66 1670 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
6c9cd161 1671 ce = create_from_disk(disk_ce, &consumed, previous_name);
cf558704 1672 set_index_entry(istate, i, ce);
7a51ed66 1673
936f53d0 1674 src_offset += consumed;
e83c5163 1675 }
6c9cd161 1676 strbuf_release(&previous_name_buf);
fba2f38a 1677 istate->timestamp.sec = st.st_mtime;
c06ff490 1678 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 1679
7a51ed66 1680 while (src_offset <= mmap_size - 20 - 8) {
bad68ec9
JH
1681 /* After an array of active_nr index entries,
1682 * there can be arbitrary number of extended
1683 * sections, each of which is prefixed with
1684 * extension name (4-byte) and section length
1685 * in 4-byte network byte order.
1686 */
07cc8eca 1687 uint32_t extsize;
7a51ed66 1688 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
bad68ec9 1689 extsize = ntohl(extsize);
4aab5b46 1690 if (read_index_extension(istate,
7a51ed66
LT
1691 (const char *) mmap + src_offset,
1692 (char *) mmap + src_offset + 8,
1d7f171c 1693 extsize) < 0)
bad68ec9 1694 goto unmap;
7a51ed66
LT
1695 src_offset += 8;
1696 src_offset += extsize;
bad68ec9 1697 }
7a51ed66 1698 munmap(mmap, mmap_size);
4aab5b46 1699 return istate->cache_nr;
e83c5163
LT
1700
1701unmap:
7a51ed66 1702 munmap(mmap, mmap_size);
5d1a5c02 1703 die("index file corrupt");
e83c5163
LT
1704}
1705
0d59ffb4
CC
1706/*
1707 * Signal that the shared index is used by updating its mtime.
1708 *
1709 * This way, shared index can be removed if they have not been used
1710 * for some time.
1711 */
1712static void freshen_shared_index(char *base_sha1_hex, int warn)
1713{
1714 const char *shared_index = git_path("sharedindex.%s", base_sha1_hex);
1715 if (!check_and_freshen_file(shared_index, 1) && warn)
1716 warning("could not freshen shared index '%s'", shared_index);
1717}
1718
5fc2fc8f
NTND
1719int read_index_from(struct index_state *istate, const char *path)
1720{
1721 struct split_index *split_index;
1722 int ret;
de6ae5f9
CC
1723 char *base_sha1_hex;
1724 const char *base_path;
5fc2fc8f
NTND
1725
1726 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1727 if (istate->initialized)
1728 return istate->cache_nr;
1729
1730 ret = do_read_index(istate, path, 0);
435ec090 1731
5fc2fc8f 1732 split_index = istate->split_index;
03f15a79 1733 if (!split_index || is_null_sha1(split_index->base_sha1)) {
435ec090 1734 post_read_index_from(istate);
5fc2fc8f 1735 return ret;
03f15a79 1736 }
5fc2fc8f
NTND
1737
1738 if (split_index->base)
1739 discard_index(split_index->base);
1740 else
1741 split_index->base = xcalloc(1, sizeof(*split_index->base));
de6ae5f9
CC
1742
1743 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1744 base_path = git_path("sharedindex.%s", base_sha1_hex);
1745 ret = do_read_index(split_index->base, base_path, 1);
5fc2fc8f
NTND
1746 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1747 die("broken index, expect %s in %s, got %s",
de6ae5f9 1748 base_sha1_hex, base_path,
5fc2fc8f 1749 sha1_to_hex(split_index->base->sha1));
de6ae5f9 1750
c3a00825 1751 freshen_shared_index(base_sha1_hex, 0);
5fc2fc8f 1752 merge_base_index(istate);
435ec090 1753 post_read_index_from(istate);
5fc2fc8f
NTND
1754 return ret;
1755}
1756
fa7b3c2f
JH
1757int is_index_unborn(struct index_state *istate)
1758{
debed2a6 1759 return (!istate->cache_nr && !istate->timestamp.sec);
fa7b3c2f
JH
1760}
1761
4aab5b46 1762int discard_index(struct index_state *istate)
6d297f81 1763{
debed2a6
RS
1764 int i;
1765
5fc2fc8f
NTND
1766 for (i = 0; i < istate->cache_nr; i++) {
1767 if (istate->cache[i]->index &&
1768 istate->split_index &&
1769 istate->split_index->base &&
1770 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1771 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1772 continue;
debed2a6 1773 free(istate->cache[i]);
5fc2fc8f 1774 }
cfc5789a 1775 resolve_undo_clear_index(istate);
4aab5b46
JH
1776 istate->cache_nr = 0;
1777 istate->cache_changed = 0;
fba2f38a
KB
1778 istate->timestamp.sec = 0;
1779 istate->timestamp.nsec = 0;
2092678c 1780 free_name_hash(istate);
4aab5b46 1781 cache_tree_free(&(istate->cache_tree));
913e0e99 1782 istate->initialized = 0;
a0fc4db0
RS
1783 free(istate->cache);
1784 istate->cache = NULL;
1785 istate->cache_alloc = 0;
5fc2fc8f 1786 discard_split_index(istate);
f9e6c649
NTND
1787 free_untracked_cache(istate->untracked);
1788 istate->untracked = NULL;
7a51ed66 1789 return 0;
6d297f81
JS
1790}
1791
d1f128b0 1792int unmerged_index(const struct index_state *istate)
94a5728c
DB
1793{
1794 int i;
1795 for (i = 0; i < istate->cache_nr; i++) {
1796 if (ce_stage(istate->cache[i]))
1797 return 1;
1798 }
1799 return 0;
1800}
1801
4990aadc 1802#define WRITE_BUFFER_SIZE 8192
bf0f910d 1803static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
1804static unsigned long write_buffer_len;
1805
9126f009 1806static int ce_write_flush(git_SHA_CTX *context, int fd)
6015c28b
JH
1807{
1808 unsigned int buffered = write_buffer_len;
1809 if (buffered) {
9126f009 1810 git_SHA1_Update(context, write_buffer, buffered);
93822c22 1811 if (write_in_full(fd, write_buffer, buffered) != buffered)
6015c28b
JH
1812 return -1;
1813 write_buffer_len = 0;
1814 }
1815 return 0;
1816}
1817
9126f009 1818static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
1819{
1820 while (len) {
1821 unsigned int buffered = write_buffer_len;
1822 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1823 if (partial > len)
1824 partial = len;
1825 memcpy(write_buffer + buffered, data, partial);
1826 buffered += partial;
1827 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
1828 write_buffer_len = buffered;
1829 if (ce_write_flush(context, fd))
4990aadc
LT
1830 return -1;
1831 buffered = 0;
1832 }
1833 write_buffer_len = buffered;
1834 len -= partial;
1d7f171c 1835 data = (char *) data + partial;
a6080a0a
JH
1836 }
1837 return 0;
4990aadc
LT
1838}
1839
9126f009 1840static int write_index_ext_header(git_SHA_CTX *context, int fd,
ac58c7b1 1841 unsigned int ext, unsigned int sz)
bad68ec9
JH
1842{
1843 ext = htonl(ext);
1844 sz = htonl(sz);
968a1d65
DR
1845 return ((ce_write(context, fd, &ext, 4) < 0) ||
1846 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
1847}
1848
e93021b2 1849static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
4990aadc
LT
1850{
1851 unsigned int left = write_buffer_len;
ca9be054 1852
4990aadc
LT
1853 if (left) {
1854 write_buffer_len = 0;
9126f009 1855 git_SHA1_Update(context, write_buffer, left);
4990aadc 1856 }
ca9be054 1857
2c865d9a
QH
1858 /* Flush first if not enough space for SHA1 signature */
1859 if (left + 20 > WRITE_BUFFER_SIZE) {
93822c22 1860 if (write_in_full(fd, write_buffer, left) != left)
2c865d9a
QH
1861 return -1;
1862 left = 0;
1863 }
1864
ca9be054 1865 /* Append the SHA1 signature at the end */
9126f009 1866 git_SHA1_Final(write_buffer + left, context);
e93021b2 1867 hashcpy(sha1, write_buffer + left);
ca9be054 1868 left += 20;
93822c22 1869 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
4990aadc
LT
1870}
1871
407c8eb0
JH
1872static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1873{
1874 /*
1875 * The only thing we care about in this function is to smudge the
1876 * falsely clean entry due to touch-update-touch race, so we leave
1877 * everything else as they are. We are called for entries whose
c21d39d7 1878 * ce_stat_data.sd_mtime match the index file mtime.
c70115b4
JH
1879 *
1880 * Note that this actually does not do much for gitlinks, for
1881 * which ce_match_stat_basic() always goes to the actual
1882 * contents. The caller checks with is_racy_timestamp() which
1883 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
1884 */
1885 struct stat st;
1886
1887 if (lstat(ce->name, &st) < 0)
1888 return;
1889 if (ce_match_stat_basic(ce, &st))
1890 return;
1891 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
1892 /* This is "racily clean"; smudge it. Note that this
1893 * is a tricky code. At first glance, it may appear
1894 * that it can break with this sequence:
1895 *
1896 * $ echo xyzzy >frotz
1897 * $ git-update-index --add frotz
1898 * $ : >frotz
1899 * $ sleep 3
1900 * $ echo filfre >nitfol
1901 * $ git-update-index --add nitfol
1902 *
b7e58b17 1903 * but it does not. When the second update-index runs,
4b3511b0
JH
1904 * it notices that the entry "frotz" has the same timestamp
1905 * as index, and if we were to smudge it by resetting its
1906 * size to zero here, then the object name recorded
1907 * in index is the 6-byte file but the cached stat information
1908 * becomes zero --- which would then match what we would
a6080a0a 1909 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
1910 *
1911 * However, the second update-index, before calling
1912 * this function, notices that the cached size is 6
1913 * bytes and what is on the filesystem is an empty
1914 * file, and never calls us, so the cached size information
1915 * for "frotz" stays 6 which does not match the filesystem.
1916 */
c21d39d7 1917 ce->ce_stat_data.sd_size = 0;
407c8eb0
JH
1918 }
1919}
1920
f136f7bf
JH
1921/* Copy miscellaneous fields but not the name */
1922static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1923 struct cache_entry *ce)
7a51ed66 1924{
b60e188c
TG
1925 short flags;
1926
c21d39d7
MH
1927 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1928 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1929 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1930 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1931 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1932 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
7a51ed66 1933 ondisk->mode = htonl(ce->ce_mode);
c21d39d7
MH
1934 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1935 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1936 ondisk->size = htonl(ce->ce_stat_data.sd_size);
99d1a986 1937 hashcpy(ondisk->sha1, ce->oid.hash);
b60e188c 1938
ce51bf09 1939 flags = ce->ce_flags & ~CE_NAMEMASK;
b60e188c
TG
1940 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1941 ondisk->flags = htons(flags);
06aaaa0b
NTND
1942 if (ce->ce_flags & CE_EXTENDED) {
1943 struct ondisk_cache_entry_extended *ondisk2;
1944 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1945 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
f136f7bf 1946 return ondisk2->name;
06aaaa0b 1947 }
f136f7bf
JH
1948 else {
1949 return ondisk->name;
1950 }
1951}
1952
9d227781
JH
1953static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1954 struct strbuf *previous_name)
f136f7bf 1955{
9d227781
JH
1956 int size;
1957 struct ondisk_cache_entry *ondisk;
b3c96fb1 1958 int saved_namelen = saved_namelen; /* compiler workaround */
f136f7bf
JH
1959 char *name;
1960 int result;
1961
b3c96fb1
NTND
1962 if (ce->ce_flags & CE_STRIP_NAME) {
1963 saved_namelen = ce_namelen(ce);
1964 ce->ce_namelen = 0;
1965 }
1966
9d227781
JH
1967 if (!previous_name) {
1968 size = ondisk_ce_size(ce);
1969 ondisk = xcalloc(1, size);
1970 name = copy_cache_entry_to_ondisk(ondisk, ce);
1971 memcpy(name, ce->name, ce_namelen(ce));
1972 } else {
1973 int common, to_remove, prefix_size;
1974 unsigned char to_remove_vi[16];
1975 for (common = 0;
1976 (ce->name[common] &&
1977 common < previous_name->len &&
1978 ce->name[common] == previous_name->buf[common]);
1979 common++)
1980 ; /* still matching */
1981 to_remove = previous_name->len - common;
1982 prefix_size = encode_varint(to_remove, to_remove_vi);
1983
1984 if (ce->ce_flags & CE_EXTENDED)
1985 size = offsetof(struct ondisk_cache_entry_extended, name);
1986 else
1987 size = offsetof(struct ondisk_cache_entry, name);
1988 size += prefix_size + (ce_namelen(ce) - common + 1);
1989
1990 ondisk = xcalloc(1, size);
1991 name = copy_cache_entry_to_ondisk(ondisk, ce);
1992 memcpy(name, to_remove_vi, prefix_size);
1993 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1994
1995 strbuf_splice(previous_name, common, to_remove,
1996 ce->name + common, ce_namelen(ce) - common);
06aaaa0b 1997 }
b3c96fb1
NTND
1998 if (ce->ce_flags & CE_STRIP_NAME) {
1999 ce->ce_namelen = saved_namelen;
2000 ce->ce_flags &= ~CE_STRIP_NAME;
2001 }
7a51ed66 2002
59efba64
JN
2003 result = ce_write(c, fd, ondisk, size);
2004 free(ondisk);
2005 return result;
7a51ed66
LT
2006}
2007
426ddeea
YM
2008/*
2009 * This function verifies if index_state has the correct sha1 of the
2010 * index file. Don't die if we have any other failure, just return 0.
2011 */
2012static int verify_index_from(const struct index_state *istate, const char *path)
2013{
2014 int fd;
2015 ssize_t n;
2016 struct stat st;
2017 unsigned char sha1[20];
2018
2019 if (!istate->initialized)
2020 return 0;
2021
2022 fd = open(path, O_RDONLY);
2023 if (fd < 0)
2024 return 0;
2025
2026 if (fstat(fd, &st))
2027 goto out;
2028
2029 if (st.st_size < sizeof(struct cache_header) + 20)
2030 goto out;
2031
2032 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2033 if (n != 20)
2034 goto out;
2035
2036 if (hashcmp(istate->sha1, sha1))
2037 goto out;
2038
2039 close(fd);
2040 return 1;
2041
2042out:
2043 close(fd);
2044 return 0;
2045}
2046
2047static int verify_index(const struct index_state *istate)
2048{
2049 return verify_index_from(istate, get_index_file());
2050}
2051
483fbe2b
JH
2052static int has_racy_timestamp(struct index_state *istate)
2053{
2054 int entries = istate->cache_nr;
2055 int i;
2056
2057 for (i = 0; i < entries; i++) {
2058 struct cache_entry *ce = istate->cache[i];
2059 if (is_racy_timestamp(istate, ce))
2060 return 1;
2061 }
2062 return 0;
2063}
2064
ccdc4ec3 2065/*
98e023de 2066 * Opportunistically update the index but do not complain if we can't
ccdc4ec3
JH
2067 */
2068void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2069{
483fbe2b 2070 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
788cef81 2071 verify_index(istate) &&
03b86647 2072 write_locked_index(istate, lockfile, COMMIT_LOCK))
ccdc4ec3
JH
2073 rollback_lock_file(lockfile);
2074}
2075
c18b80a0
NTND
2076static int do_write_index(struct index_state *istate, int newfd,
2077 int strip_extensions)
197ee8c9 2078{
9126f009 2079 git_SHA_CTX c;
197ee8c9 2080 struct cache_header hdr;
9d227781 2081 int i, err, removed, extended, hdr_version;
4aab5b46
JH
2082 struct cache_entry **cache = istate->cache;
2083 int entries = istate->cache_nr;
e1afca4f 2084 struct stat st;
9d227781 2085 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
025a0709 2086
06aaaa0b 2087 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 2088 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 2089 removed++;
197ee8c9 2090
06aaaa0b
NTND
2091 /* reduce extended entries if possible */
2092 cache[i]->ce_flags &= ~CE_EXTENDED;
2093 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2094 extended++;
2095 cache[i]->ce_flags |= CE_EXTENDED;
2096 }
2097 }
2098
d6e3c181 2099 if (!istate->version) {
136347d7 2100 istate->version = get_index_format_default();
d6e3c181
NTND
2101 if (getenv("GIT_TEST_SPLIT_INDEX"))
2102 init_split_index(istate);
2103 }
9d227781
JH
2104
2105 /* demote version 3 to version 2 when the latter suffices */
2106 if (istate->version == 3 || istate->version == 2)
2107 istate->version = extended ? 3 : 2;
2108
2109 hdr_version = istate->version;
2110
ccc4feb5 2111 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
9d227781 2112 hdr.hdr_version = htonl(hdr_version);
025a0709 2113 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 2114
9126f009 2115 git_SHA1_Init(&c);
ca9be054 2116 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
2117 return -1;
2118
9d227781 2119 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
197ee8c9
LT
2120 for (i = 0; i < entries; i++) {
2121 struct cache_entry *ce = cache[i];
7a51ed66 2122 if (ce->ce_flags & CE_REMOVE)
aa16021e 2123 continue;
e06c43c7 2124 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 2125 ce_smudge_racily_clean_entry(ce);
99d1a986 2126 if (is_null_oid(&ce->oid)) {
83bd7437
JK
2127 static const char msg[] = "cache entry has null sha1: %s";
2128 static int allow = -1;
2129
2130 if (allow < 0)
2131 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2132 if (allow)
2133 warning(msg, ce->name);
2134 else
2135 return error(msg, ce->name);
2136 }
9d227781 2137 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
197ee8c9
LT
2138 return -1;
2139 }
9d227781 2140 strbuf_release(&previous_name_buf);
1af1c2b6 2141
bad68ec9 2142 /* Write extension data here */
c18b80a0 2143 if (!strip_extensions && istate->split_index) {
5fc2fc8f
NTND
2144 struct strbuf sb = STRBUF_INIT;
2145
2146 err = write_link_extension(&sb, istate) < 0 ||
2147 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2148 sb.len) < 0 ||
2149 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2150 strbuf_release(&sb);
2151 if (err)
2152 return -1;
2153 }
c18b80a0 2154 if (!strip_extensions && istate->cache_tree) {
f285a2d7 2155 struct strbuf sb = STRBUF_INIT;
1dffb8fa 2156
1dffb8fa
PH
2157 cache_tree_write(&sb, istate->cache_tree);
2158 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2159 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2160 strbuf_release(&sb);
2161 if (err)
bad68ec9 2162 return -1;
bad68ec9 2163 }
c18b80a0 2164 if (!strip_extensions && istate->resolve_undo) {
cfc5789a
JH
2165 struct strbuf sb = STRBUF_INIT;
2166
2167 resolve_undo_write(&sb, istate->resolve_undo);
2168 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2169 sb.len) < 0
2170 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2171 strbuf_release(&sb);
2172 if (err)
2173 return -1;
2174 }
83c094ad
NTND
2175 if (!strip_extensions && istate->untracked) {
2176 struct strbuf sb = STRBUF_INIT;
2177
2178 write_untracked_extension(&sb, istate->untracked);
2179 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2180 sb.len) < 0 ||
2181 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2182 strbuf_release(&sb);
2183 if (err)
2184 return -1;
2185 }
e1afca4f 2186
e93021b2 2187 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
e1afca4f 2188 return -1;
5bcf109c
KB
2189 istate->timestamp.sec = (unsigned int)st.st_mtime;
2190 istate->timestamp.nsec = ST_MTIME_NSEC(st);
e1afca4f 2191 return 0;
197ee8c9 2192}
e46bbcf6 2193
626f35c8
NTND
2194void set_alternate_index_output(const char *name)
2195{
2196 alternate_index_output = name;
2197}
2198
2199static int commit_locked_index(struct lock_file *lk)
2200{
751baced
MH
2201 if (alternate_index_output)
2202 return commit_lock_file_to(lk, alternate_index_output);
2203 else
626f35c8 2204 return commit_lock_file(lk);
626f35c8
NTND
2205}
2206
03b86647
NTND
2207static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2208 unsigned flags)
2209{
c99a4c2d 2210 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
03b86647
NTND
2211 if (ret)
2212 return ret;
2213 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2214 (COMMIT_LOCK | CLOSE_LOCK));
2215 if (flags & COMMIT_LOCK)
2216 return commit_locked_index(lock);
2217 else if (flags & CLOSE_LOCK)
2218 return close_lock_file(lock);
2219 else
2220 return ret;
2221}
2222
5fc2fc8f
NTND
2223static int write_split_index(struct index_state *istate,
2224 struct lock_file *lock,
2225 unsigned flags)
2226{
2227 int ret;
2228 prepare_to_write_split_index(istate);
2229 ret = do_write_locked_index(istate, lock, flags);
2230 finish_writing_split_index(istate);
2231 return ret;
2232}
2233
b9683722
CC
2234static const char *shared_index_expire = "2.weeks.ago";
2235
2236static unsigned long get_shared_index_expire_date(void)
2237{
2238 static unsigned long shared_index_expire_date;
2239 static int shared_index_expire_date_prepared;
2240
2241 if (!shared_index_expire_date_prepared) {
2242 git_config_get_expiry("splitindex.sharedindexexpire",
2243 &shared_index_expire);
2244 shared_index_expire_date = approxidate(shared_index_expire);
2245 shared_index_expire_date_prepared = 1;
2246 }
2247
2248 return shared_index_expire_date;
2249}
2250
2251static int should_delete_shared_index(const char *shared_index_path)
2252{
2253 struct stat st;
2254 unsigned long expiration;
2255
2256 /* Check timestamp */
2257 expiration = get_shared_index_expire_date();
2258 if (!expiration)
2259 return 0;
2260 if (stat(shared_index_path, &st))
2261 return error_errno(_("could not stat '%s"), shared_index_path);
2262 if (st.st_mtime > expiration)
2263 return 0;
2264
2265 return 1;
2266}
2267
2268static int clean_shared_index_files(const char *current_hex)
2269{
2270 struct dirent *de;
2271 DIR *dir = opendir(get_git_dir());
2272
2273 if (!dir)
2274 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2275
2276 while ((de = readdir(dir)) != NULL) {
2277 const char *sha1_hex;
2278 const char *shared_index_path;
2279 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2280 continue;
2281 if (!strcmp(sha1_hex, current_hex))
2282 continue;
2283 shared_index_path = git_path("%s", de->d_name);
2284 if (should_delete_shared_index(shared_index_path) > 0 &&
2285 unlink(shared_index_path))
2286 warning_errno(_("unable to unlink: %s"), shared_index_path);
2287 }
2288 closedir(dir);
2289
2290 return 0;
2291}
2292
f6ecc62d 2293static struct tempfile temporary_sharedindex;
c18b80a0 2294
a0a96756
NTND
2295static int write_shared_index(struct index_state *istate,
2296 struct lock_file *lock, unsigned flags)
c18b80a0
NTND
2297{
2298 struct split_index *si = istate->split_index;
c18b80a0
NTND
2299 int fd, ret;
2300
f6ecc62d 2301 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
a0a96756 2302 if (fd < 0) {
a0a96756
NTND
2303 hashclr(si->base_sha1);
2304 return do_write_locked_index(istate, lock, flags);
2305 }
c18b80a0
NTND
2306 move_cache_to_base_index(istate);
2307 ret = do_write_index(si->base, fd, 1);
c18b80a0 2308 if (ret) {
f6ecc62d 2309 delete_tempfile(&temporary_sharedindex);
c18b80a0
NTND
2310 return ret;
2311 }
f6ecc62d
MH
2312 ret = rename_tempfile(&temporary_sharedindex,
2313 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
b9683722 2314 if (!ret) {
c18b80a0 2315 hashcpy(si->base_sha1, si->base->sha1);
b9683722
CC
2316 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2317 }
2318
c18b80a0
NTND
2319 return ret;
2320}
2321
e6a1dd77
CC
2322static const int default_max_percent_split_change = 20;
2323
2324static int too_many_not_shared_entries(struct index_state *istate)
2325{
2326 int i, not_shared = 0;
2327 int max_split = git_config_get_max_percent_split_change();
2328
2329 switch (max_split) {
2330 case -1:
2331 /* not or badly configured: use the default value */
2332 max_split = default_max_percent_split_change;
2333 break;
2334 case 0:
2335 return 1; /* 0% means always write a new shared index */
2336 case 100:
2337 return 0; /* 100% means never write a new shared index */
2338 default:
2339 break; /* just use the configured value */
2340 }
2341
2342 /* Count not shared entries */
2343 for (i = 0; i < istate->cache_nr; i++) {
2344 struct cache_entry *ce = istate->cache[i];
2345 if (!ce->index)
2346 not_shared++;
2347 }
2348
2349 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2350}
2351
03b86647
NTND
2352int write_locked_index(struct index_state *istate, struct lock_file *lock,
2353 unsigned flags)
2354{
0d59ffb4 2355 int new_shared_index, ret;
5fc2fc8f
NTND
2356 struct split_index *si = istate->split_index;
2357
5165dd59
NTND
2358 if (!si || alternate_index_output ||
2359 (istate->cache_changed & ~EXTMASK)) {
5fc2fc8f
NTND
2360 if (si)
2361 hashclr(si->base_sha1);
2362 return do_write_locked_index(istate, lock, flags);
2363 }
2364
d6e3c181
NTND
2365 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2366 int v = si->base_sha1[0];
2367 if ((v & 15) < 6)
2368 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2369 }
e6a1dd77
CC
2370 if (too_many_not_shared_entries(istate))
2371 istate->cache_changed |= SPLIT_INDEX_ORDERED;
0d59ffb4
CC
2372
2373 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2374
2375 if (new_shared_index) {
2376 ret = write_shared_index(istate, lock, flags);
c18b80a0
NTND
2377 if (ret)
2378 return ret;
2379 }
2380
0d59ffb4
CC
2381 ret = write_split_index(istate, lock, flags);
2382
2383 /* Freshen the shared index only if the split-index was written */
2384 if (!ret && !new_shared_index)
2385 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2386
2387 return ret;
03b86647
NTND
2388}
2389
e46bbcf6
MV
2390/*
2391 * Read the index file that is potentially unmerged into given
63e8dc5b 2392 * index_state, dropping any unmerged entries. Returns true if
e46bbcf6
MV
2393 * the index is unmerged. Callers who want to refuse to work
2394 * from an unmerged state can call this and check its return value,
2395 * instead of calling read_cache().
2396 */
2397int read_index_unmerged(struct index_state *istate)
2398{
2399 int i;
d1a43f2a 2400 int unmerged = 0;
e46bbcf6
MV
2401
2402 read_index(istate);
e46bbcf6
MV
2403 for (i = 0; i < istate->cache_nr; i++) {
2404 struct cache_entry *ce = istate->cache[i];
d1a43f2a
JH
2405 struct cache_entry *new_ce;
2406 int size, len;
2407
2408 if (!ce_stage(ce))
e46bbcf6 2409 continue;
d1a43f2a 2410 unmerged = 1;
68c4f6a5 2411 len = ce_namelen(ce);
d1a43f2a
JH
2412 size = cache_entry_size(len);
2413 new_ce = xcalloc(1, size);
d1a43f2a 2414 memcpy(new_ce->name, ce->name, len);
b60e188c
TG
2415 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2416 new_ce->ce_namelen = len;
d1a43f2a
JH
2417 new_ce->ce_mode = ce->ce_mode;
2418 if (add_index_entry(istate, new_ce, 0))
2419 return error("%s: cannot drop to stage #0",
5699d17e 2420 new_ce->name);
e46bbcf6 2421 }
d1a43f2a 2422 return unmerged;
e46bbcf6 2423}
041aee31 2424
98fa4738
JK
2425/*
2426 * Returns 1 if the path is an "other" path with respect to
2427 * the index; that is, the path is not mentioned in the index at all,
2428 * either as a file, a directory with some files in the index,
2429 * or as an unmerged entry.
2430 *
2431 * We helpfully remove a trailing "/" from directories so that
2432 * the output of read_directory can be used as-is.
2433 */
2434int index_name_is_other(const struct index_state *istate, const char *name,
2435 int namelen)
2436{
2437 int pos;
2438 if (namelen && name[namelen - 1] == '/')
2439 namelen--;
2440 pos = index_name_pos(istate, name, namelen);
2441 if (0 <= pos)
2442 return 0; /* exact match */
2443 pos = -pos - 1;
2444 if (pos < istate->cache_nr) {
2445 struct cache_entry *ce = istate->cache[pos];
2446 if (ce_namelen(ce) == namelen &&
2447 !memcmp(ce->name, name, namelen))
2448 return 0; /* Yup, this one exists unmerged */
2449 }
2450 return 1;
2451}
29fb37b2 2452
87542508
BW
2453void *read_blob_data_from_index(const struct index_state *istate,
2454 const char *path, unsigned long *size)
29fb37b2
LF
2455{
2456 int pos, len;
2457 unsigned long sz;
2458 enum object_type type;
2459 void *data;
2460
2461 len = strlen(path);
2462 pos = index_name_pos(istate, path, len);
2463 if (pos < 0) {
2464 /*
2465 * We might be in the middle of a merge, in which
2466 * case we would read stage #2 (ours).
2467 */
2468 int i;
2469 for (i = -pos - 1;
2470 (pos < 0 && i < istate->cache_nr &&
2471 !strcmp(istate->cache[i]->name, path));
2472 i++)
2473 if (ce_stage(istate->cache[i]) == 2)
2474 pos = i;
2475 }
2476 if (pos < 0)
2477 return NULL;
99d1a986 2478 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
29fb37b2
LF
2479 if (!data || type != OBJ_BLOB) {
2480 free(data);
2481 return NULL;
2482 }
ff366825
LF
2483 if (size)
2484 *size = sz;
29fb37b2
LF
2485 return data;
2486}
38612532
MH
2487
2488void stat_validity_clear(struct stat_validity *sv)
2489{
2490 free(sv->sd);
2491 sv->sd = NULL;
2492}
2493
2494int stat_validity_check(struct stat_validity *sv, const char *path)
2495{
2496 struct stat st;
2497
2498 if (stat(path, &st) < 0)
2499 return sv->sd == NULL;
2500 if (!sv->sd)
2501 return 0;
2502 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2503}
2504
2505void stat_validity_update(struct stat_validity *sv, int fd)
2506{
2507 struct stat st;
2508
2509 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2510 stat_validity_clear(sv);
2511 else {
2512 if (!sv->sd)
2513 sv->sd = xcalloc(1, sizeof(struct stat_data));
2514 fill_stat_data(sv->sd, &st);
2515 }
2516}