read-cache: teach make_cache_entry to take object_id
[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"
b2141fc1 8#include "config.h"
f6ecc62d 9#include "tempfile.h"
697cc8ef 10#include "lockfile.h"
bad68ec9 11#include "cache-tree.h"
f35a6d3b 12#include "refs.h"
d616813d 13#include "dir.h"
041aee31
JH
14#include "tree.h"
15#include "commit.h"
39425819 16#include "blob.h"
cfc5789a 17#include "resolve-undo.h"
6c9cd161
JH
18#include "strbuf.h"
19#include "varint.h"
5fc2fc8f 20#include "split-index.h"
a42643aa 21#include "utf8.h"
883e248b 22#include "fsmonitor.h"
bad68ec9 23
b60e188c
TG
24/* Mask for the name length in ce_flags in the on-disk index */
25
26#define CE_NAMEMASK (0x0fff)
27
bad68ec9
JH
28/* Index extensions.
29 *
30 * The first letter should be 'A'..'Z' for extensions that are not
31 * necessary for a correct operation (i.e. optimization data).
32 * When new extensions are added that _needs_ to be understood in
33 * order to correctly interpret the index file, pick character that
34 * is outside the range, to cause the reader to abort.
35 */
36
37#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
38#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
b659b49b 39#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
5fc2fc8f 40#define CACHE_EXT_LINK 0x6c696e6b /* "link" */
83c094ad 41#define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
883e248b 42#define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
5fc2fc8f
NTND
43
44/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
e0cf0d7d 45#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
c18b80a0 46 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
883e248b 47 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
e83c5163 48
228e94f9 49struct index_state the_index;
626f35c8 50static const char *alternate_index_output;
8fd2cb40 51
9cb76b8c
JH
52static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
53{
54 istate->cache[nr] = ce;
96872bc2 55 add_name_hash(istate, ce);
9cb76b8c
JH
56}
57
cf558704
LT
58static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
59{
60 struct cache_entry *old = istate->cache[nr];
61
078a58e8 62 replace_index_entry_in_base(istate, old, ce);
2092678c 63 remove_name_hash(istate, old);
5699d17e 64 free(old);
0e267b7a 65 ce->ce_flags &= ~CE_HASHED;
a22c6371 66 set_index_entry(istate, nr, ce);
078a58e8 67 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 68 mark_fsmonitor_invalid(istate, ce);
e636a7b4 69 istate->cache_changed |= CE_ENTRY_CHANGED;
cf558704
LT
70}
71
81dc2307
PB
72void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
73{
285c2e25 74 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
81dc2307
PB
75 int namelen = strlen(new_name);
76
285c2e25
BW
77 new_entry = xmalloc(cache_entry_size(namelen));
78 copy_cache_entry(new_entry, old_entry);
79 new_entry->ce_flags &= ~CE_HASHED;
80 new_entry->ce_namelen = namelen;
81 new_entry->index = 0;
82 memcpy(new_entry->name, new_name, namelen + 1);
81dc2307 83
285c2e25
BW
84 cache_tree_invalidate_path(istate, old_entry->name);
85 untracked_cache_remove_from_index(istate, old_entry->name);
81dc2307 86 remove_index_entry_at(istate, nr);
285c2e25 87 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
81dc2307
PB
88}
89
c21d39d7
MH
90void fill_stat_data(struct stat_data *sd, struct stat *st)
91{
92 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
93 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
94 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
95 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
96 sd->sd_dev = st->st_dev;
97 sd->sd_ino = st->st_ino;
98 sd->sd_uid = st->st_uid;
99 sd->sd_gid = st->st_gid;
100 sd->sd_size = st->st_size;
101}
102
103int match_stat_data(const struct stat_data *sd, struct stat *st)
104{
105 int changed = 0;
106
107 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
108 changed |= MTIME_CHANGED;
109 if (trust_ctime && check_stat &&
110 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
111 changed |= CTIME_CHANGED;
112
113#ifdef USE_NSEC
114 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
115 changed |= MTIME_CHANGED;
116 if (trust_ctime && check_stat &&
117 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
118 changed |= CTIME_CHANGED;
119#endif
120
121 if (check_stat) {
122 if (sd->sd_uid != (unsigned int) st->st_uid ||
123 sd->sd_gid != (unsigned int) st->st_gid)
124 changed |= OWNER_CHANGED;
125 if (sd->sd_ino != (unsigned int) st->st_ino)
126 changed |= INODE_CHANGED;
127 }
128
129#ifdef USE_STDEV
130 /*
131 * st_dev breaks on network filesystems where different
132 * clients will have different views of what "device"
133 * the filesystem is on
134 */
135 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
136 changed |= INODE_CHANGED;
137#endif
138
139 if (sd->sd_size != (unsigned int) st->st_size)
140 changed |= DATA_CHANGED;
141
142 return changed;
143}
144
415e96c8
JH
145/*
146 * This only updates the "non-critical" parts of the directory
147 * cache, ie the parts that aren't tracked by GIT, and only used
148 * to validate the cache.
149 */
150void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
151{
c21d39d7 152 fill_stat_data(&ce->ce_stat_data, st);
5f73076c
JH
153
154 if (assume_unchanged)
7a51ed66 155 ce->ce_flags |= CE_VALID;
eadb5831 156
883e248b 157 if (S_ISREG(st->st_mode)) {
eadb5831 158 ce_mark_uptodate(ce);
883e248b
BP
159 mark_fsmonitor_valid(ce);
160 }
415e96c8
JH
161}
162
21a6b9fa 163static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
164{
165 int match = -1;
1b8ac5ea 166 int fd = git_open_cloexec(ce->name, O_RDONLY);
29e4d363
JH
167
168 if (fd >= 0) {
bebfecb9 169 struct object_id oid;
e3506559 170 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
bebfecb9 171 match = oidcmp(&oid, &ce->oid);
7f8508e8 172 /* index_fd() closed the file descriptor already */
29e4d363
JH
173 }
174 return match;
175}
176
21a6b9fa 177static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
29e4d363
JH
178{
179 int match = -1;
29e4d363
JH
180 void *buffer;
181 unsigned long size;
21666f1a 182 enum object_type type;
a60272b3 183 struct strbuf sb = STRBUF_INIT;
29e4d363 184
a60272b3 185 if (strbuf_readlink(&sb, ce->name, expected_size))
29e4d363 186 return -1;
a60272b3 187
b4f5aca4 188 buffer = read_object_file(&ce->oid, &type, &size);
a60272b3
LT
189 if (buffer) {
190 if (size == sb.len)
191 match = memcmp(buffer, sb.buf, size);
192 free(buffer);
29e4d363 193 }
a60272b3 194 strbuf_release(&sb);
29e4d363
JH
195 return match;
196}
197
21a6b9fa 198static int ce_compare_gitlink(const struct cache_entry *ce)
f35a6d3b 199{
1053fe82 200 struct object_id oid;
f35a6d3b
LT
201
202 /*
203 * We don't actually require that the .git directory
302b9282 204 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
205 * might even be missing (in case nobody populated that
206 * sub-project).
207 *
208 * If so, we consider it always to match.
209 */
a98e6101 210 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
f35a6d3b 211 return 0;
1053fe82 212 return oidcmp(&oid, &ce->oid);
f35a6d3b
LT
213}
214
21a6b9fa 215static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
216{
217 switch (st->st_mode & S_IFMT) {
218 case S_IFREG:
219 if (ce_compare_data(ce, st))
220 return DATA_CHANGED;
221 break;
222 case S_IFLNK:
dc49cd76 223 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
224 return DATA_CHANGED;
225 break;
a8ee75bc 226 case S_IFDIR:
7a51ed66 227 if (S_ISGITLINK(ce->ce_mode))
c70115b4 228 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
1cf01a34 229 /* else fallthrough */
29e4d363
JH
230 default:
231 return TYPE_CHANGED;
232 }
233 return 0;
234}
235
21a6b9fa 236static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
734aab75
LT
237{
238 unsigned int changed = 0;
239
7a51ed66
LT
240 if (ce->ce_flags & CE_REMOVE)
241 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
242
243 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
244 case S_IFREG:
245 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
246 /* We consider only the owner x bit to be relevant for
247 * "mode changes"
248 */
249 if (trust_executable_bit &&
7a51ed66 250 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 251 changed |= MODE_CHANGED;
8ae0a8c5
KS
252 break;
253 case S_IFLNK:
78a8d641
JS
254 if (!S_ISLNK(st->st_mode) &&
255 (has_symlinks || !S_ISREG(st->st_mode)))
256 changed |= TYPE_CHANGED;
8ae0a8c5 257 break;
302b9282 258 case S_IFGITLINK:
c70115b4 259 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
260 if (!S_ISDIR(st->st_mode))
261 changed |= TYPE_CHANGED;
262 else if (ce_compare_gitlink(ce))
263 changed |= DATA_CHANGED;
a8ee75bc 264 return changed;
8ae0a8c5 265 default:
7a51ed66 266 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 267 }
2cb45e95 268
c21d39d7 269 changed |= match_stat_data(&ce->ce_stat_data, st);
b0391890 270
f49c2c22 271 /* Racily smudged entry? */
c21d39d7 272 if (!ce->ce_stat_data.sd_size) {
99d1a986 273 if (!is_empty_blob_sha1(ce->oid.hash))
f49c2c22
LT
274 changed |= DATA_CHANGED;
275 }
276
407c8eb0
JH
277 return changed;
278}
279
2bb4cda1
NTND
280static int is_racy_stat(const struct index_state *istate,
281 const struct stat_data *sd)
6d91da6d 282{
2bb4cda1 283 return (istate->timestamp.sec &&
fba2f38a
KB
284#ifdef USE_NSEC
285 /* nanosecond timestamped files can also be racy! */
2bb4cda1
NTND
286 (istate->timestamp.sec < sd->sd_mtime.sec ||
287 (istate->timestamp.sec == sd->sd_mtime.sec &&
288 istate->timestamp.nsec <= sd->sd_mtime.nsec))
fba2f38a 289#else
2bb4cda1 290 istate->timestamp.sec <= sd->sd_mtime.sec
fba2f38a 291#endif
2bb4cda1
NTND
292 );
293}
294
295static int is_racy_timestamp(const struct index_state *istate,
296 const struct cache_entry *ce)
297{
298 return (!S_ISGITLINK(ce->ce_mode) &&
299 is_racy_stat(istate, &ce->ce_stat_data));
6d91da6d
JH
300}
301
ed4efab1
NTND
302int match_stat_data_racy(const struct index_state *istate,
303 const struct stat_data *sd, struct stat *st)
304{
305 if (is_racy_stat(istate, sd))
306 return MTIME_CHANGED;
307 return match_stat_data(sd, st);
6d91da6d
JH
308}
309
883e248b 310int ie_match_stat(struct index_state *istate,
21a6b9fa 311 const struct cache_entry *ce, struct stat *st,
4bd5b7da 312 unsigned int options)
407c8eb0 313{
5f73076c 314 unsigned int changed;
4bd5b7da 315 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 316 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
4bd5b7da 317 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
883e248b 318 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
5f73076c 319
883e248b
BP
320 if (!ignore_fsmonitor)
321 refresh_fsmonitor(istate);
5f73076c
JH
322 /*
323 * If it's marked as always valid in the index, it's
324 * valid whatever the checked-out copy says.
56cac48c
NTND
325 *
326 * skip-worktree has the same effect with higher precedence
5f73076c 327 */
56cac48c
NTND
328 if (!ignore_skip_worktree && ce_skip_worktree(ce))
329 return 0;
7a51ed66 330 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c 331 return 0;
883e248b
BP
332 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
333 return 0;
5f73076c 334
331fcb59
JH
335 /*
336 * Intent-to-add entries have not been added, so the index entry
337 * by definition never matches what is in the work tree until it
338 * actually gets added.
339 */
895ff3b2 340 if (ce_intent_to_add(ce))
331fcb59
JH
341 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
342
5f73076c 343 changed = ce_match_stat_basic(ce, st);
407c8eb0 344
29e4d363
JH
345 /*
346 * Within 1 second of this sequence:
347 * echo xyzzy >file && git-update-index --add file
348 * running this command:
349 * echo frotz >file
350 * would give a falsely clean cache entry. The mtime and
351 * length match the cache, and other stat fields do not change.
352 *
353 * We could detect this at update-index time (the cache entry
354 * being registered/updated records the same time as "now")
355 * and delay the return from git-update-index, but that would
356 * effectively mean we can make at most one commit per second,
357 * which is not acceptable. Instead, we check cache entries
358 * whose mtime are the same as the index file timestamp more
5f73076c 359 * carefully than others.
29e4d363 360 */
6d91da6d 361 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
362 if (assume_racy_is_modified)
363 changed |= DATA_CHANGED;
364 else
365 changed |= ce_modified_check_fs(ce, st);
366 }
b0391890 367
29e4d363 368 return changed;
b0391890
JH
369}
370
883e248b 371int ie_modified(struct index_state *istate,
21a6b9fa
RS
372 const struct cache_entry *ce,
373 struct stat *st, unsigned int options)
b0391890 374{
29e4d363 375 int changed, changed_fs;
4bd5b7da
JH
376
377 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
378 if (!changed)
379 return 0;
b0391890
JH
380 /*
381 * If the mode or type has changed, there's no point in trying
382 * to refresh the entry - it's not going to match
383 */
384 if (changed & (MODE_CHANGED | TYPE_CHANGED))
385 return changed;
386
c70115b4
JH
387 /*
388 * Immediately after read-tree or update-index --cacheinfo,
389 * the length field is zero, as we have never even read the
390 * lstat(2) information once, and we cannot trust DATA_CHANGED
391 * returned by ie_match_stat() which in turn was returned by
392 * ce_match_stat_basic() to signal that the filesize of the
393 * blob changed. We have to actually go to the filesystem to
394 * see if the contents match, and if so, should answer "unchanged".
395 *
396 * The logic does not apply to gitlinks, as ce_match_stat_basic()
397 * already has checked the actual HEAD from the filesystem in the
398 * subproject. If ie_match_stat() already said it is different,
399 * then we know it is.
b0391890 400 */
c70115b4 401 if ((changed & DATA_CHANGED) &&
c21d39d7 402 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
b0391890
JH
403 return changed;
404
29e4d363
JH
405 changed_fs = ce_modified_check_fs(ce, st);
406 if (changed_fs)
407 return changed | changed_fs;
b0391890
JH
408 return 0;
409}
410
958ba6c9
LT
411int base_name_compare(const char *name1, int len1, int mode1,
412 const char *name2, int len2, int mode2)
413{
414 unsigned char c1, c2;
415 int len = len1 < len2 ? len1 : len2;
416 int cmp;
417
418 cmp = memcmp(name1, name2, len);
419 if (cmp)
420 return cmp;
421 c1 = name1[len];
422 c2 = name2[len];
1833a925 423 if (!c1 && S_ISDIR(mode1))
958ba6c9 424 c1 = '/';
1833a925 425 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
426 c2 = '/';
427 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
428}
429
0ab9e1e8
LT
430/*
431 * df_name_compare() is identical to base_name_compare(), except it
432 * compares conflicting directory/file entries as equal. Note that
433 * while a directory name compares as equal to a regular file, they
434 * then individually compare _differently_ to a filename that has
435 * a dot after the basename (because '\0' < '.' < '/').
436 *
437 * This is used by routines that want to traverse the git namespace
438 * but then handle conflicting entries together when possible.
439 */
440int df_name_compare(const char *name1, int len1, int mode1,
441 const char *name2, int len2, int mode2)
442{
443 int len = len1 < len2 ? len1 : len2, cmp;
444 unsigned char c1, c2;
445
446 cmp = memcmp(name1, name2, len);
447 if (cmp)
448 return cmp;
449 /* Directories and files compare equal (same length, same name) */
450 if (len1 == len2)
451 return 0;
452 c1 = name1[len];
453 if (!c1 && S_ISDIR(mode1))
454 c1 = '/';
455 c2 = name2[len];
456 if (!c2 && S_ISDIR(mode2))
457 c2 = '/';
458 if (c1 == '/' && !c2)
459 return 0;
460 if (c2 == '/' && !c1)
461 return 0;
462 return c1 - c2;
463}
464
ccdd4a0f 465int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
eb38c22f 466{
ccdd4a0f
JM
467 size_t min_len = (len1 < len2) ? len1 : len2;
468 int cmp = memcmp(name1, name2, min_len);
eb38c22f
LT
469 if (cmp)
470 return cmp;
471 if (len1 < len2)
472 return -1;
473 if (len1 > len2)
474 return 1;
ccdd4a0f
JM
475 return 0;
476}
477
478int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
479{
480 int cmp;
481
482 cmp = name_compare(name1, len1, name2, len2);
483 if (cmp)
484 return cmp;
5f73076c 485
b60e188c 486 if (stage1 < stage2)
95fd5bf8 487 return -1;
b60e188c 488 if (stage1 > stage2)
95fd5bf8 489 return 1;
eb38c22f
LT
490 return 0;
491}
492
357e9c69 493static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
eb38c22f
LT
494{
495 int first, last;
496
497 first = 0;
4aab5b46 498 last = istate->cache_nr;
eb38c22f
LT
499 while (last > first) {
500 int next = (last + first) >> 1;
4aab5b46 501 struct cache_entry *ce = istate->cache[next];
b60e188c 502 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
eb38c22f 503 if (!cmp)
76e7f4ec 504 return next;
eb38c22f
LT
505 if (cmp < 0) {
506 last = next;
507 continue;
508 }
509 first = next+1;
510 }
76e7f4ec 511 return -first-1;
eb38c22f
LT
512}
513
b60e188c
TG
514int index_name_pos(const struct index_state *istate, const char *name, int namelen)
515{
516 return index_name_stage_pos(istate, name, namelen, 0);
517}
518
4aab5b46 519int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 520{
cf558704
LT
521 struct cache_entry *ce = istate->cache[pos];
522
cfc5789a 523 record_resolve_undo(istate, ce);
2092678c 524 remove_name_hash(istate, ce);
045113a5 525 save_or_free_index_entry(istate, ce);
e636a7b4 526 istate->cache_changed |= CE_ENTRY_REMOVED;
4aab5b46
JH
527 istate->cache_nr--;
528 if (pos >= istate->cache_nr)
7b937ca3 529 return 0;
f331ab9d
RS
530 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
531 istate->cache_nr - pos);
7b937ca3
LT
532 return 1;
533}
534
36419c8e 535/*
98e023de 536 * Remove all cache entries marked for removal, that is where
36419c8e
KB
537 * CE_REMOVE is set in ce_flags. This is much more effective than
538 * calling remove_index_entry_at() for each entry to be removed.
539 */
540void remove_marked_cache_entries(struct index_state *istate)
541{
542 struct cache_entry **ce_array = istate->cache;
543 unsigned int i, j;
544
545 for (i = j = 0; i < istate->cache_nr; i++) {
5699d17e 546 if (ce_array[i]->ce_flags & CE_REMOVE) {
2092678c 547 remove_name_hash(istate, ce_array[i]);
045113a5 548 save_or_free_index_entry(istate, ce_array[i]);
5699d17e 549 }
36419c8e
KB
550 else
551 ce_array[j++] = ce_array[i];
552 }
ad837d9e
NTND
553 if (j == istate->cache_nr)
554 return;
e636a7b4 555 istate->cache_changed |= CE_ENTRY_REMOVED;
36419c8e
KB
556 istate->cache_nr = j;
557}
558
4aab5b46 559int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 560{
4aab5b46 561 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
562 if (pos < 0)
563 pos = -pos-1;
a5400efe 564 cache_tree_invalidate_path(istate, path);
e931371a 565 untracked_cache_remove_from_index(istate, path);
4aab5b46
JH
566 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
567 remove_index_entry_at(istate, pos);
197ee8c9
LT
568 return 0;
569}
570
20314271
JS
571static int compare_name(struct cache_entry *ce, const char *path, int namelen)
572{
573 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
574}
575
576static int index_name_pos_also_unmerged(struct index_state *istate,
577 const char *path, int namelen)
578{
579 int pos = index_name_pos(istate, path, namelen);
580 struct cache_entry *ce;
581
582 if (pos >= 0)
583 return pos;
584
585 /* maybe unmerged? */
586 pos = -1 - pos;
587 if (pos >= istate->cache_nr ||
588 compare_name((ce = istate->cache[pos]), path, namelen))
589 return -1;
590
591 /* order of preference: stage 2, 1, 3 */
592 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
593 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
594 !compare_name(ce, path, namelen))
595 pos++;
596 return pos;
597}
598
1102952b
LT
599static int different_name(struct cache_entry *ce, struct cache_entry *alias)
600{
601 int len = ce_namelen(ce);
602 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
603}
604
605/*
606 * If we add a filename that aliases in the cache, we will use the
607 * name that we already have - but we don't want to update the same
608 * alias twice, because that implies that there were actually two
609 * different files with aliasing names!
610 *
611 * So we use the CE_ADDED flag to verify that the alias was an old
612 * one before we accept it as
613 */
045113a5
NTND
614static struct cache_entry *create_alias_ce(struct index_state *istate,
615 struct cache_entry *ce,
616 struct cache_entry *alias)
1102952b
LT
617{
618 int len;
285c2e25 619 struct cache_entry *new_entry;
1102952b
LT
620
621 if (alias->ce_flags & CE_ADDED)
622 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
623
624 /* Ok, create the new entry using the name of the existing alias */
625 len = ce_namelen(alias);
285c2e25
BW
626 new_entry = xcalloc(1, cache_entry_size(len));
627 memcpy(new_entry->name, alias->name, len);
628 copy_cache_entry(new_entry, ce);
045113a5 629 save_or_free_index_entry(istate, ce);
285c2e25 630 return new_entry;
1102952b
LT
631}
632
b4b313f9 633void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
39425819 634{
a09c985e
PO
635 struct object_id oid;
636 if (write_object_file("", 0, blob_type, &oid))
39425819 637 die("cannot create an empty blob in the object database");
a09c985e 638 oidcpy(&ce->oid, &oid);
39425819
JH
639}
640
610d55af 641int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 642{
38ed1d89 643 int size, namelen, was_same;
d177cab0 644 mode_t st_mode = st->st_mode;
9472935d 645 struct cache_entry *ce, *alias = NULL;
56cac48c 646 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
647 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
648 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
649 int intent_only = flags & ADD_CACHE_INTENT;
650 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
651 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
9472935d
TB
652 int newflags = HASH_WRITE_OBJECT;
653
654 if (flags & HASH_RENORMALIZE)
655 newflags |= HASH_RENORMALIZE;
11be42a4 656
d177cab0 657 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
960b8ad1 658 return error("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
659
660 namelen = strlen(path);
d177cab0 661 if (S_ISDIR(st_mode)) {
09595258
LT
662 while (namelen && path[namelen-1] == '/')
663 namelen--;
664 }
11be42a4
JS
665 size = cache_entry_size(namelen);
666 ce = xcalloc(1, size);
667 memcpy(ce->name, path, namelen);
b60e188c 668 ce->ce_namelen = namelen;
39425819
JH
669 if (!intent_only)
670 fill_stat_cache_info(ce, st);
388b2acd
JH
671 else
672 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 673
610d55af
TG
674
675 if (trust_executable_bit && has_symlinks) {
d177cab0 676 ce->ce_mode = create_ce_mode(st_mode);
610d55af 677 } else {
78a8d641
JS
678 /* If there is an existing entry, pick the mode bits and type
679 * from it, otherwise assume unexecutable regular file.
11be42a4 680 */
185c975f 681 struct cache_entry *ent;
20314271 682 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 683
4aab5b46 684 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 685 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
686 }
687
dc1ae704
JJ
688 /* When core.ignorecase=true, determine if a directory of the same name but differing
689 * case already exists within the Git repository. If it does, ensure the directory
690 * case of the file being added to the repository matches (is folded into) the existing
691 * entry's directory case.
692 */
693 if (ignore_case) {
41284eb0 694 adjust_dirname_case(istate, ce->name);
dc1ae704 695 }
9472935d
TB
696 if (!(flags & HASH_RENORMALIZE)) {
697 alias = index_file_exists(istate, ce->name,
698 ce_namelen(ce), ignore_case);
699 if (alias &&
700 !ce_stage(alias) &&
701 !ie_match_stat(istate, alias, st, ce_option)) {
702 /* Nothing changed, really */
703 if (!S_ISGITLINK(alias->ce_mode))
704 ce_mark_uptodate(alias);
705 alias->ce_flags |= CE_ADDED;
dc1ae704 706
9472935d
TB
707 free(ce);
708 return 0;
709 }
0781b8a9 710 }
39425819 711 if (!intent_only) {
9472935d 712 if (index_path(&ce->oid, path, st, newflags)) {
2d9426b0 713 free(ce);
39425819 714 return error("unable to index file %s", path);
2d9426b0 715 }
39425819 716 } else
b4b313f9 717 set_object_name_for_intent_to_add_entry(ce);
39425819 718
1102952b 719 if (ignore_case && alias && different_name(ce, alias))
045113a5 720 ce = create_alias_ce(istate, ce, alias);
1102952b 721 ce->ce_flags |= CE_ADDED;
38ed1d89 722
3bf0dd1f 723 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
724 was_same = (alias &&
725 !ce_stage(alias) &&
99d1a986 726 !oidcmp(&alias->oid, &ce->oid) &&
38ed1d89
JH
727 ce->ce_mode == alias->ce_mode);
728
729 if (pretend)
067178ed
JH
730 free(ce);
731 else if (add_index_entry(istate, ce, add_option)) {
732 free(ce);
733 return error("unable to add %s to index", path);
734 }
38ed1d89 735 if (verbose && !was_same)
11be42a4 736 printf("add '%s'\n", path);
11be42a4
JS
737 return 0;
738}
739
610d55af 740int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
741{
742 struct stat st;
743 if (lstat(path, &st))
d824cbba 744 die_errno("unable to stat '%s'", path);
610d55af 745 return add_to_index(istate, path, &st, flags);
d177cab0
LT
746}
747
6640f881 748struct cache_entry *make_cache_entry(unsigned int mode,
825ed4d9
JM
749 const struct object_id *oid,
750 const char *path,
751 int stage,
752 unsigned int refresh_options)
6640f881
CR
753{
754 int size, len;
bc1c2caa 755 struct cache_entry *ce, *ret;
6640f881 756
10ecfa76 757 if (!verify_path(path, mode)) {
7e7abea9 758 error("Invalid path '%s'", path);
6640f881 759 return NULL;
7e7abea9 760 }
6640f881
CR
761
762 len = strlen(path);
763 size = cache_entry_size(len);
764 ce = xcalloc(1, size);
765
825ed4d9 766 oidcpy(&ce->oid, oid);
6640f881 767 memcpy(ce->name, path, len);
b60e188c
TG
768 ce->ce_flags = create_ce_flags(stage);
769 ce->ce_namelen = len;
6640f881
CR
770 ce->ce_mode = create_ce_mode(mode);
771
768d7965 772 ret = refresh_cache_entry(&the_index, ce, refresh_options);
915e44c6 773 if (ret != ce)
bc1c2caa 774 free(ce);
915e44c6 775 return ret;
6640f881
CR
776}
777
d9d70966
TG
778/*
779 * Chmod an index entry with either +x or -x.
780 *
781 * Returns -1 if the chmod for the particular cache entry failed (if it's
782 * not a regular file), -2 if an invalid flip argument is passed in, 0
783 * otherwise.
784 */
785int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
786 char flip)
787{
788 if (!S_ISREG(ce->ce_mode))
789 return -1;
790 switch (flip) {
791 case '+':
792 ce->ce_mode |= 0111;
793 break;
794 case '-':
795 ce->ce_mode &= ~0111;
796 break;
797 default:
798 return -2;
799 }
800 cache_tree_invalidate_path(istate, ce->name);
801 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 802 mark_fsmonitor_invalid(istate, ce);
d9d70966
TG
803 istate->cache_changed |= CE_ENTRY_CHANGED;
804
805 return 0;
806}
807
9c5e6c80 808int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
7b937ca3
LT
809{
810 int len = ce_namelen(a);
811 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
812}
813
8dcf39c4
LT
814/*
815 * We fundamentally don't like some paths: we don't want
816 * dot or dot-dot anywhere, and for obvious reasons don't
817 * want to recurse into ".git" either.
818 *
819 * Also, we don't want double slashes or slashes at the
820 * end that can make pathnames ambiguous.
821 */
10ecfa76 822static int verify_dotfile(const char *rest, unsigned mode)
8dcf39c4
LT
823{
824 /*
825 * The first character was '.', but that
826 * has already been discarded, we now test
827 * the rest.
828 */
e0f530ff 829
8dcf39c4 830 /* "." is not allowed */
e0f530ff 831 if (*rest == '\0' || is_dir_sep(*rest))
8dcf39c4
LT
832 return 0;
833
e0f530ff 834 switch (*rest) {
8dcf39c4 835 /*
641084b6
JK
836 * ".git" followed by NUL or slash is bad. Note that we match
837 * case-insensitively here, even if ignore_case is not set.
838 * This outlaws ".GIT" everywhere out of an abundance of caution,
839 * since there's really no good reason to allow it.
10ecfa76
JK
840 *
841 * Once we've seen ".git", we can also find ".gitmodules", etc (also
842 * case-insensitively).
8dcf39c4
LT
843 */
844 case 'g':
cc2fc7c2
JK
845 case 'G':
846 if (rest[1] != 'i' && rest[1] != 'I')
8dcf39c4 847 break;
cc2fc7c2 848 if (rest[2] != 't' && rest[2] != 'T')
8dcf39c4 849 break;
e19e5e66
JK
850 if (rest[3] == '\0' || is_dir_sep(rest[3]))
851 return 0;
10ecfa76
JK
852 if (S_ISLNK(mode)) {
853 rest += 3;
854 if (skip_iprefix(rest, "modules", &rest) &&
855 (*rest == '\0' || is_dir_sep(*rest)))
856 return 0;
857 }
e19e5e66 858 break;
8dcf39c4 859 case '.':
e0f530ff 860 if (rest[1] == '\0' || is_dir_sep(rest[1]))
8dcf39c4
LT
861 return 0;
862 }
863 return 1;
864}
865
10ecfa76 866int verify_path(const char *path, unsigned mode)
8dcf39c4
LT
867{
868 char c;
869
56948cb6
EFL
870 if (has_dos_drive_prefix(path))
871 return 0;
872
8dcf39c4
LT
873 goto inside;
874 for (;;) {
875 if (!c)
876 return 1;
56948cb6 877 if (is_dir_sep(c)) {
8dcf39c4 878inside:
10ecfa76
JK
879 if (protect_hfs) {
880 if (is_hfs_dotgit(path))
881 return 0;
882 if (S_ISLNK(mode)) {
883 if (is_hfs_dotgitmodules(path))
884 return 0;
885 }
886 }
887 if (protect_ntfs) {
888 if (is_ntfs_dotgit(path))
889 return 0;
890 if (S_ISLNK(mode)) {
891 if (is_ntfs_dotgitmodules(path))
892 return 0;
893 }
894 }
895
8dcf39c4 896 c = *path++;
10ecfa76 897 if ((c == '.' && !verify_dotfile(path, mode)) ||
3bdf09c7
JH
898 is_dir_sep(c) || c == '\0')
899 return 0;
8dcf39c4
LT
900 }
901 c = *path++;
902 }
903}
904
12676608
LT
905/*
906 * Do we have another file that has the beginning components being a
907 * proper superset of the name we're trying to add?
0f1e4f04 908 */
4aab5b46
JH
909static int has_file_name(struct index_state *istate,
910 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 911{
12676608
LT
912 int retval = 0;
913 int len = ce_namelen(ce);
b155725d 914 int stage = ce_stage(ce);
12676608 915 const char *name = ce->name;
0f1e4f04 916
4aab5b46
JH
917 while (pos < istate->cache_nr) {
918 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 919
12676608 920 if (len >= ce_namelen(p))
0f1e4f04 921 break;
12676608
LT
922 if (memcmp(name, p->name, len))
923 break;
b155725d
JH
924 if (ce_stage(p) != stage)
925 continue;
12676608
LT
926 if (p->name[len] != '/')
927 continue;
7a51ed66 928 if (p->ce_flags & CE_REMOVE)
21cd8d00 929 continue;
12676608
LT
930 retval = -1;
931 if (!ok_to_replace)
932 break;
4aab5b46 933 remove_index_entry_at(istate, --pos);
0f1e4f04 934 }
12676608
LT
935 return retval;
936}
0f1e4f04 937
a6db3fbb
JH
938
939/*
940 * Like strcmp(), but also return the offset of the first change.
941 * If strings are equal, return the length.
942 */
943int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
944{
945 size_t k;
946
947 if (!first_change)
948 return strcmp(s1, s2);
949
950 for (k = 0; s1[k] == s2[k]; k++)
951 if (s1[k] == '\0')
952 break;
953
954 *first_change = k;
955 return (unsigned char)s1[k] - (unsigned char)s2[k];
956}
957
12676608
LT
958/*
959 * Do we have another file with a pathname that is a proper
960 * subset of the name we're trying to add?
06b6d81b
JH
961 *
962 * That is, is there another file in the index with a path
963 * that matches a sub-directory in the given entry?
12676608 964 */
4aab5b46
JH
965static int has_dir_name(struct index_state *istate,
966 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
967{
968 int retval = 0;
b155725d 969 int stage = ce_stage(ce);
12676608
LT
970 const char *name = ce->name;
971 const char *slash = name + ce_namelen(ce);
06b6d81b
JH
972 size_t len_eq_last;
973 int cmp_last = 0;
974
975 /*
976 * We are frequently called during an iteration on a sorted
977 * list of pathnames and while building a new index. Therefore,
978 * there is a high probability that this entry will eventually
979 * be appended to the index, rather than inserted in the middle.
980 * If we can confirm that, we can avoid binary searches on the
981 * components of the pathname.
982 *
983 * Compare the entry's full path with the last path in the index.
984 */
985 if (istate->cache_nr > 0) {
986 cmp_last = strcmp_offset(name,
987 istate->cache[istate->cache_nr - 1]->name,
988 &len_eq_last);
989 if (cmp_last > 0) {
990 if (len_eq_last == 0) {
991 /*
992 * The entry sorts AFTER the last one in the
993 * index and their paths have no common prefix,
994 * so there cannot be a F/D conflict.
995 */
996 return retval;
997 } else {
998 /*
999 * The entry sorts AFTER the last one in the
1000 * index, but has a common prefix. Fall through
1001 * to the loop below to disect the entry's path
1002 * and see where the difference is.
1003 */
1004 }
1005 } else if (cmp_last == 0) {
1006 /*
1007 * The entry exactly matches the last one in the
1008 * index, but because of multiple stage and CE_REMOVE
1009 * items, we fall through and let the regular search
1010 * code handle it.
1011 */
1012 }
1013 }
0f1e4f04 1014
12676608 1015 for (;;) {
b986df5c 1016 size_t len;
0f1e4f04 1017
12676608
LT
1018 for (;;) {
1019 if (*--slash == '/')
1020 break;
1021 if (slash <= ce->name)
1022 return retval;
1023 }
1024 len = slash - name;
0f1e4f04 1025
b986df5c
JH
1026 if (cmp_last > 0) {
1027 /*
1028 * (len + 1) is a directory boundary (including
1029 * the trailing slash). And since the loop is
1030 * decrementing "slash", the first iteration is
1031 * the longest directory prefix; subsequent
1032 * iterations consider parent directories.
1033 */
1034
1035 if (len + 1 <= len_eq_last) {
1036 /*
1037 * The directory prefix (including the trailing
1038 * slash) also appears as a prefix in the last
1039 * entry, so the remainder cannot collide (because
1040 * strcmp said the whole path was greater).
1041 *
1042 * EQ: last: xxx/A
1043 * this: xxx/B
1044 *
1045 * LT: last: xxx/file_A
1046 * this: xxx/file_B
1047 */
1048 return retval;
1049 }
1050
1051 if (len > len_eq_last) {
1052 /*
1053 * This part of the directory prefix (excluding
1054 * the trailing slash) is longer than the known
1055 * equal portions, so this sub-directory cannot
1056 * collide with a file.
1057 *
1058 * GT: last: xxxA
1059 * this: xxxB/file
1060 */
1061 return retval;
1062 }
1063
1064 if (istate->cache_nr > 0 &&
1065 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1066 /*
1067 * The directory prefix lines up with part of
1068 * a longer file or directory name, but sorts
1069 * after it, so this sub-directory cannot
1070 * collide with a file.
1071 *
1072 * last: xxx/yy-file (because '-' sorts before '/')
1073 * this: xxx/yy/abc
1074 */
1075 return retval;
1076 }
1077
1078 /*
1079 * This is a possible collision. Fall through and
1080 * let the regular search code handle it.
1081 *
1082 * last: xxx
1083 * this: xxx/file
1084 */
1085 }
1086
b60e188c 1087 pos = index_name_stage_pos(istate, name, len, stage);
12676608 1088 if (pos >= 0) {
21cd8d00
JH
1089 /*
1090 * Found one, but not so fast. This could
1091 * be a marker that says "I was here, but
1092 * I am being removed". Such an entry is
1093 * not a part of the resulting tree, and
1094 * it is Ok to have a directory at the same
1095 * path.
1096 */
077c48df 1097 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
1098 retval = -1;
1099 if (!ok_to_replace)
1100 break;
4aab5b46 1101 remove_index_entry_at(istate, pos);
21cd8d00
JH
1102 continue;
1103 }
12676608 1104 }
21cd8d00
JH
1105 else
1106 pos = -pos-1;
12676608
LT
1107
1108 /*
1109 * Trivial optimization: if we find an entry that
1110 * already matches the sub-directory, then we know
b155725d 1111 * we're ok, and we can exit.
12676608 1112 */
4aab5b46
JH
1113 while (pos < istate->cache_nr) {
1114 struct cache_entry *p = istate->cache[pos];
b155725d
JH
1115 if ((ce_namelen(p) <= len) ||
1116 (p->name[len] != '/') ||
1117 memcmp(p->name, name, len))
1118 break; /* not our subdirectory */
077c48df
JH
1119 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1120 /*
1121 * p is at the same stage as our entry, and
b155725d
JH
1122 * is a subdirectory of what we are looking
1123 * at, so we cannot have conflicts at our
1124 * level or anything shorter.
1125 */
1126 return retval;
1127 pos++;
192268c1 1128 }
0f1e4f04 1129 }
12676608
LT
1130 return retval;
1131}
1132
1133/* We may be in a situation where we already have path/file and path
1134 * is being added, or we already have path and path/file is being
1135 * added. Either one would result in a nonsense tree that has path
1136 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 1137 *
12676608
LT
1138 * If ok-to-replace is specified, we remove the conflicting entries
1139 * from the cache so the caller should recompute the insert position.
1140 * When this happens, we return non-zero.
1141 */
4aab5b46
JH
1142static int check_file_directory_conflict(struct index_state *istate,
1143 const struct cache_entry *ce,
1144 int pos, int ok_to_replace)
12676608 1145{
21cd8d00
JH
1146 int retval;
1147
1148 /*
1149 * When ce is an "I am going away" entry, we allow it to be added
1150 */
7a51ed66 1151 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
1152 return 0;
1153
12676608
LT
1154 /*
1155 * We check if the path is a sub-path of a subsequent pathname
1156 * first, since removing those will not change the position
21cd8d00 1157 * in the array.
12676608 1158 */
4aab5b46 1159 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 1160
12676608
LT
1161 /*
1162 * Then check if the path might have a clashing sub-directory
1163 * before it.
1164 */
4aab5b46 1165 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
1166}
1167
af3785dc 1168static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
1169{
1170 int pos;
192268c1
JH
1171 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1172 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 1173 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 1174 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 1175
ce7c614b
NTND
1176 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1177 cache_tree_invalidate_path(istate, ce->name);
e5494631
JH
1178
1179 /*
1180 * If this entry's path sorts after the last entry in the index,
1181 * we can avoid searching for it.
1182 */
1183 if (istate->cache_nr > 0 &&
1184 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1185 pos = -istate->cache_nr - 1;
1186 else
1187 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
197ee8c9 1188
3e09cdfd 1189 /* existing match? Just replace it. */
76e7f4ec 1190 if (pos >= 0) {
39425819
JH
1191 if (!new_only)
1192 replace_index_entry(istate, pos, ce);
197ee8c9
LT
1193 return 0;
1194 }
76e7f4ec 1195 pos = -pos-1;
197ee8c9 1196
ffcc9ba7
NTND
1197 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1198 untracked_cache_add_to_index(istate, ce->name);
e931371a 1199
7b937ca3
LT
1200 /*
1201 * Inserting a merged entry ("stage 0") into the index
1202 * will always replace all non-merged entries..
1203 */
4aab5b46
JH
1204 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1205 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 1206 ok_to_add = 1;
4aab5b46 1207 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
1208 break;
1209 }
1210 }
1211
121481ab
LT
1212 if (!ok_to_add)
1213 return -1;
10ecfa76 1214 if (!verify_path(ce->name, ce->ce_mode))
7e7abea9 1215 return error("Invalid path '%s'", ce->name);
121481ab 1216
3e09cdfd 1217 if (!skip_df_check &&
4aab5b46 1218 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 1219 if (!ok_to_replace)
4aab5b46
JH
1220 return error("'%s' appears as both a file and as a directory",
1221 ce->name);
b60e188c 1222 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
192268c1
JH
1223 pos = -pos-1;
1224 }
af3785dc
JH
1225 return pos + 1;
1226}
1227
1228int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1229{
1230 int pos;
1231
1232 if (option & ADD_CACHE_JUST_APPEND)
1233 pos = istate->cache_nr;
1234 else {
1235 int ret;
1236 ret = add_index_entry_with_check(istate, ce, option);
1237 if (ret <= 0)
1238 return ret;
1239 pos = ret - 1;
1240 }
0f1e4f04 1241
197ee8c9 1242 /* Make sure the array is big enough .. */
999f5660 1243 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
197ee8c9
LT
1244
1245 /* Add it in.. */
4aab5b46 1246 istate->cache_nr++;
af3785dc 1247 if (istate->cache_nr > pos + 1)
f919ffeb
SG
1248 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1249 istate->cache_nr - pos - 1);
cf558704 1250 set_index_entry(istate, pos, ce);
e636a7b4 1251 istate->cache_changed |= CE_ENTRY_ADDED;
197ee8c9
LT
1252 return 0;
1253}
1254
405e5b2f
LT
1255/*
1256 * "refresh" does not calculate a new sha1 file or bring the
1257 * cache up-to-date for mode/content changes. But what it
1258 * _does_ do is to "re-match" the stat information of a file
1259 * with the cache, so that you can refresh the cache for a
1260 * file that hasn't been changed but where the stat entry is
1261 * out of date.
1262 *
1263 * For example, you'd want to do this after doing a "git-read-tree",
1264 * to link up the stat cache details with the proper files.
1265 */
4aab5b46 1266static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da 1267 struct cache_entry *ce,
d05e6970
JK
1268 unsigned int options, int *err,
1269 int *changed_ret)
405e5b2f
LT
1270{
1271 struct stat st;
1272 struct cache_entry *updated;
1273 int changed, size;
25762726 1274 int refresh = options & CE_MATCH_REFRESH;
4bd5b7da 1275 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1276 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
2e2e7ec1 1277 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
883e248b 1278 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
405e5b2f 1279
25762726 1280 if (!refresh || ce_uptodate(ce))
eadb5831
JH
1281 return ce;
1282
883e248b
BP
1283 if (!ignore_fsmonitor)
1284 refresh_fsmonitor(istate);
aa9349d4 1285 /*
56cac48c
NTND
1286 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1287 * that the change to the work tree does not matter and told
1288 * us not to worry.
aa9349d4 1289 */
56cac48c
NTND
1290 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1291 ce_mark_uptodate(ce);
1292 return ce;
1293 }
aa9349d4
MSO
1294 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1295 ce_mark_uptodate(ce);
1296 return ce;
1297 }
883e248b
BP
1298 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1299 ce_mark_uptodate(ce);
1300 return ce;
1301 }
aa9349d4 1302
ccad42d4
RS
1303 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1304 if (ignore_missing)
1305 return ce;
1306 if (err)
1307 *err = ENOENT;
1308 return NULL;
1309 }
1310
8fd2cb40 1311 if (lstat(ce->name, &st) < 0) {
2e2e7ec1
BK
1312 if (ignore_missing && errno == ENOENT)
1313 return ce;
ec0cc704
JH
1314 if (err)
1315 *err = errno;
8fd2cb40
JS
1316 return NULL;
1317 }
405e5b2f 1318
4bd5b7da 1319 changed = ie_match_stat(istate, ce, &st, options);
d05e6970
JK
1320 if (changed_ret)
1321 *changed_ret = changed;
405e5b2f 1322 if (!changed) {
4bd5b7da
JH
1323 /*
1324 * The path is unchanged. If we were told to ignore
1325 * valid bit, then we did the actual stat check and
1326 * found that the entry is unmodified. If the entry
1327 * is not marked VALID, this is the place to mark it
1328 * valid again, under "assume unchanged" mode.
1329 */
1330 if (ignore_valid && assume_unchanged &&
7a51ed66 1331 !(ce->ce_flags & CE_VALID))
405e5b2f 1332 ; /* mark this one VALID again */
eadb5831
JH
1333 else {
1334 /*
1335 * We do not mark the index itself "modified"
1336 * because CE_UPTODATE flag is in-core only;
1337 * we are not going to write this change out.
1338 */
883e248b 1339 if (!S_ISGITLINK(ce->ce_mode)) {
125fd984 1340 ce_mark_uptodate(ce);
883e248b
BP
1341 mark_fsmonitor_valid(ce);
1342 }
8fd2cb40 1343 return ce;
eadb5831 1344 }
405e5b2f
LT
1345 }
1346
4bd5b7da 1347 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1348 if (err)
1349 *err = EINVAL;
8fd2cb40
JS
1350 return NULL;
1351 }
405e5b2f
LT
1352
1353 size = ce_size(ce);
1354 updated = xmalloc(size);
0e267b7a
BP
1355 copy_cache_entry(updated, ce);
1356 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
405e5b2f 1357 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1358 /*
1359 * If ignore_valid is not set, we should leave CE_VALID bit
1360 * alone. Otherwise, paths marked with --no-assume-unchanged
1361 * (i.e. things to be edited) will reacquire CE_VALID bit
1362 * automatically, which is not really what we want.
405e5b2f 1363 */
4bd5b7da 1364 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1365 !(ce->ce_flags & CE_VALID))
1366 updated->ce_flags &= ~CE_VALID;
405e5b2f 1367
e636a7b4 1368 /* istate->cache_changed is updated in the caller */
405e5b2f
LT
1369 return updated;
1370}
1371
3deffc52 1372static void show_file(const char * fmt, const char * name, int in_porcelain,
046613c5 1373 int * first, const char *header_msg)
3deffc52
MM
1374{
1375 if (in_porcelain && *first && header_msg) {
1376 printf("%s\n", header_msg);
cd2b8ae9 1377 *first = 0;
3deffc52
MM
1378 }
1379 printf(fmt, name);
1380}
1381
9b2d6149
NTND
1382int refresh_index(struct index_state *istate, unsigned int flags,
1383 const struct pathspec *pathspec,
046613c5 1384 char *seen, const char *header_msg)
405e5b2f
LT
1385{
1386 int i;
1387 int has_errors = 0;
1388 int really = (flags & REFRESH_REALLY) != 0;
1389 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1390 int quiet = (flags & REFRESH_QUIET) != 0;
1391 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1392 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1393 int first = 1;
1394 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
25762726
BK
1395 unsigned int options = (CE_MATCH_REFRESH |
1396 (really ? CE_MATCH_IGNORE_VALID : 0) |
2e2e7ec1 1397 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
4bd4e730 1398 const char *modified_fmt;
73b7eae6
JK
1399 const char *deleted_fmt;
1400 const char *typechange_fmt;
1401 const char *added_fmt;
4bd4e730 1402 const char *unmerged_fmt;
ca54d9ba 1403 uint64_t start = getnanotime();
405e5b2f 1404
4bd4e730 1405 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
73b7eae6
JK
1406 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1407 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1408 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
4bd4e730 1409 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
4aab5b46 1410 for (i = 0; i < istate->cache_nr; i++) {
285c2e25 1411 struct cache_entry *ce, *new_entry;
ec0cc704 1412 int cache_errno = 0;
73b7eae6 1413 int changed = 0;
3d1f148c 1414 int filtered = 0;
ec0cc704 1415
4aab5b46 1416 ce = istate->cache[i];
5fdeacb0
JS
1417 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1418 continue;
1419
429bb40a 1420 if (pathspec && !ce_path_match(ce, pathspec, seen))
3d1f148c
JH
1421 filtered = 1;
1422
405e5b2f 1423 if (ce_stage(ce)) {
4aab5b46
JH
1424 while ((i < istate->cache_nr) &&
1425 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1426 i++;
1427 i--;
1428 if (allow_unmerged)
1429 continue;
3d1f148c
JH
1430 if (!filtered)
1431 show_file(unmerged_fmt, ce->name, in_porcelain,
1432 &first, header_msg);
405e5b2f
LT
1433 has_errors = 1;
1434 continue;
1435 }
1436
3d1f148c 1437 if (filtered)
d616813d
AJ
1438 continue;
1439
285c2e25
BW
1440 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1441 if (new_entry == ce)
405e5b2f 1442 continue;
285c2e25 1443 if (!new_entry) {
73b7eae6
JK
1444 const char *fmt;
1445
8fd2cb40 1446 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1447 /* If we are doing --really-refresh that
1448 * means the index is not valid anymore.
1449 */
7a51ed66 1450 ce->ce_flags &= ~CE_VALID;
078a58e8 1451 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 1452 mark_fsmonitor_invalid(istate, ce);
e636a7b4 1453 istate->cache_changed |= CE_ENTRY_CHANGED;
405e5b2f
LT
1454 }
1455 if (quiet)
1456 continue;
73b7eae6
JK
1457
1458 if (cache_errno == ENOENT)
1459 fmt = deleted_fmt;
895ff3b2 1460 else if (ce_intent_to_add(ce))
73b7eae6
JK
1461 fmt = added_fmt; /* must be before other checks */
1462 else if (changed & TYPE_CHANGED)
1463 fmt = typechange_fmt;
1464 else
1465 fmt = modified_fmt;
1466 show_file(fmt,
1467 ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1468 has_errors = 1;
1469 continue;
1470 }
cf558704 1471
285c2e25 1472 replace_index_entry(istate, i, new_entry);
405e5b2f 1473 }
ca54d9ba 1474 trace_performance_since(start, "refresh index");
405e5b2f
LT
1475 return has_errors;
1476}
1477
768d7965
JM
1478struct cache_entry *refresh_cache_entry(struct index_state *istate,
1479 struct cache_entry *ce,
1480 unsigned int options)
ec0cc704 1481{
768d7965 1482 return refresh_cache_ent(istate, ce, options, NULL, NULL);
ec0cc704
JH
1483}
1484
db3b313c
JH
1485
1486/*****************************************************************
1487 * Index File I/O
1488 *****************************************************************/
1489
9d227781
JH
1490#define INDEX_FORMAT_DEFAULT 3
1491
136347d7
TG
1492static unsigned int get_index_format_default(void)
1493{
1494 char *envversion = getenv("GIT_INDEX_VERSION");
3c09d684 1495 char *endp;
b27a5720 1496 int value;
3c09d684
TG
1497 unsigned int version = INDEX_FORMAT_DEFAULT;
1498
136347d7 1499 if (!envversion) {
b27a5720
TA
1500 if (!git_config_get_int("index.version", &value))
1501 version = value;
3c09d684
TG
1502 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1503 warning(_("index.version set, but the value is invalid.\n"
136347d7 1504 "Using version %i"), INDEX_FORMAT_DEFAULT);
3c09d684 1505 return INDEX_FORMAT_DEFAULT;
136347d7
TG
1506 }
1507 return version;
1508 }
3c09d684
TG
1509
1510 version = strtoul(envversion, &endp, 10);
1511 if (*endp ||
1512 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1513 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1514 "Using version %i"), INDEX_FORMAT_DEFAULT);
1515 version = INDEX_FORMAT_DEFAULT;
1516 }
1517 return version;
136347d7
TG
1518}
1519
db3b313c
JH
1520/*
1521 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1522 * Again - this is just a (very strong in practice) heuristic that
1523 * the inode hasn't changed.
1524 *
1525 * We save the fields in big-endian order to allow using the
1526 * index file over NFS transparently.
1527 */
1528struct ondisk_cache_entry {
1529 struct cache_time ctime;
1530 struct cache_time mtime;
7800c1eb
TG
1531 uint32_t dev;
1532 uint32_t ino;
1533 uint32_t mode;
1534 uint32_t uid;
1535 uint32_t gid;
1536 uint32_t size;
db3b313c 1537 unsigned char sha1[20];
7800c1eb 1538 uint16_t flags;
db3b313c
JH
1539 char name[FLEX_ARRAY]; /* more */
1540};
1541
1542/*
1543 * This struct is used when CE_EXTENDED bit is 1
1544 * The struct must match ondisk_cache_entry exactly from
1545 * ctime till flags
1546 */
1547struct ondisk_cache_entry_extended {
1548 struct cache_time ctime;
1549 struct cache_time mtime;
7800c1eb
TG
1550 uint32_t dev;
1551 uint32_t ino;
1552 uint32_t mode;
1553 uint32_t uid;
1554 uint32_t gid;
1555 uint32_t size;
db3b313c 1556 unsigned char sha1[20];
7800c1eb
TG
1557 uint16_t flags;
1558 uint16_t flags2;
db3b313c
JH
1559 char name[FLEX_ARRAY]; /* more */
1560};
1561
6c9cd161 1562/* These are only used for v3 or lower */
ce012deb 1563#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
db3b313c
JH
1564#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1565#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1566#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1567#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1568 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1569 ondisk_cache_entry_size(ce_namelen(ce)))
1570
a33fc72f
JH
1571/* Allow fsck to force verification of the index checksum. */
1572int verify_index_checksum;
1573
00ec50e5
BP
1574/* Allow fsck to force verification of the cache entry order. */
1575int verify_ce_order;
1576
bad68ec9 1577static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163 1578{
aab61359 1579 git_hash_ctx c;
1580 unsigned char hash[GIT_MAX_RAWSZ];
0136bac9 1581 int hdr_version;
e83c5163 1582
ccc4feb5 1583 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1584 return error("bad signature");
0136bac9 1585 hdr_version = ntohl(hdr->hdr_version);
b82a7b5b 1586 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
0136bac9 1587 return error("bad index version %d", hdr_version);
a33fc72f
JH
1588
1589 if (!verify_index_checksum)
1590 return 0;
1591
aab61359 1592 the_hash_algo->init_fn(&c);
1593 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1594 the_hash_algo->final_fn(hash, &c);
1595 if (hashcmp(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
ca9be054 1596 return error("bad index file sha1 signature");
e83c5163
LT
1597 return 0;
1598}
1599
4aab5b46
JH
1600static int read_index_extension(struct index_state *istate,
1601 const char *ext, void *data, unsigned long sz)
bad68ec9
JH
1602{
1603 switch (CACHE_EXT(ext)) {
1604 case CACHE_EXT_TREE:
4aab5b46 1605 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1606 break;
cfc5789a
JH
1607 case CACHE_EXT_RESOLVE_UNDO:
1608 istate->resolve_undo = resolve_undo_read(data, sz);
1609 break;
5fc2fc8f
NTND
1610 case CACHE_EXT_LINK:
1611 if (read_link_extension(istate, data, sz))
1612 return -1;
1613 break;
f9e6c649
NTND
1614 case CACHE_EXT_UNTRACKED:
1615 istate->untracked = read_untracked_extension(data, sz);
1616 break;
883e248b
BP
1617 case CACHE_EXT_FSMONITOR:
1618 read_fsmonitor_extension(istate, data, sz);
1619 break;
bad68ec9
JH
1620 default:
1621 if (*ext < 'A' || 'Z' < *ext)
1622 return error("index uses %.4s extension, which we do not understand",
1623 ext);
1624 fprintf(stderr, "ignoring %.4s extension\n", ext);
1625 break;
1626 }
1627 return 0;
1628}
1629
b3e83cc7 1630int hold_locked_index(struct lock_file *lk, int lock_flags)
216aab1e 1631{
b3e83cc7 1632 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
216aab1e
MH
1633}
1634
4aab5b46 1635int read_index(struct index_state *istate)
8fd2cb40 1636{
a125a223 1637 return read_index_from(istate, get_index_file(), get_git_dir());
8fd2cb40
JS
1638}
1639
3fc22b53
JH
1640static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1641 unsigned int flags,
1642 const char *name,
1643 size_t len)
1644{
1645 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1646
c3d8da57
JK
1647 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1648 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1649 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1650 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1651 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1652 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1653 ce->ce_mode = get_be32(&ondisk->mode);
1654 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1655 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1656 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
b60e188c
TG
1657 ce->ce_flags = flags & ~CE_NAMEMASK;
1658 ce->ce_namelen = len;
5fc2fc8f 1659 ce->index = 0;
99d1a986 1660 hashcpy(ce->oid.hash, ondisk->sha1);
3fc22b53
JH
1661 memcpy(ce->name, name, len);
1662 ce->name[len] = '\0';
1663 return ce;
1664}
1665
6c9cd161
JH
1666/*
1667 * Adjacent cache entries tend to share the leading paths, so it makes
1668 * sense to only store the differences in later entries. In the v4
1669 * on-disk format of the index, each on-disk cache entry stores the
1670 * number of bytes to be stripped from the end of the previous name,
1671 * and the bytes to append to the result, to come up with its name.
1672 */
1673static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1674{
1675 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1676 size_t len = decode_varint(&cp);
1677
1678 if (name->len < len)
1679 die("malformed name field in the index");
1680 strbuf_remove(name, name->len - len, len);
1681 for (ep = cp; *ep; ep++)
1682 ; /* find the end */
1683 strbuf_add(name, cp, ep - cp);
1684 return (const char *)ep + 1 - cp_;
1685}
1686
936f53d0 1687static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
6c9cd161
JH
1688 unsigned long *ent_size,
1689 struct strbuf *previous_name)
7a51ed66 1690{
debed2a6 1691 struct cache_entry *ce;
7fec10b7 1692 size_t len;
06aaaa0b 1693 const char *name;
debed2a6 1694 unsigned int flags;
7fec10b7 1695
7a51ed66 1696 /* On-disk flags are just 16 bits */
c3d8da57 1697 flags = get_be16(&ondisk->flags);
debed2a6 1698 len = flags & CE_NAMEMASK;
7fec10b7 1699
debed2a6 1700 if (flags & CE_EXTENDED) {
06aaaa0b
NTND
1701 struct ondisk_cache_entry_extended *ondisk2;
1702 int extended_flags;
1703 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
c3d8da57 1704 extended_flags = get_be16(&ondisk2->flags2) << 16;
06aaaa0b
NTND
1705 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1706 if (extended_flags & ~CE_EXTENDED_FLAGS)
1707 die("Unknown index entry format %08x", extended_flags);
debed2a6 1708 flags |= extended_flags;
06aaaa0b
NTND
1709 name = ondisk2->name;
1710 }
1711 else
1712 name = ondisk->name;
1713
6c9cd161
JH
1714 if (!previous_name) {
1715 /* v3 and earlier */
1716 if (len == CE_NAMEMASK)
1717 len = strlen(name);
1718 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1719
1720 *ent_size = ondisk_ce_size(ce);
1721 } else {
1722 unsigned long consumed;
1723 consumed = expand_name_field(previous_name, name);
1724 ce = cache_entry_from_ondisk(ondisk, flags,
1725 previous_name->buf,
1726 previous_name->len);
1727
1728 *ent_size = (name - ((char *)ondisk)) + consumed;
1729 }
debed2a6 1730 return ce;
cf558704
LT
1731}
1732
03f15a79 1733static void check_ce_order(struct index_state *istate)
15999d0b 1734{
03f15a79
TG
1735 unsigned int i;
1736
00ec50e5
BP
1737 if (!verify_ce_order)
1738 return;
1739
03f15a79
TG
1740 for (i = 1; i < istate->cache_nr; i++) {
1741 struct cache_entry *ce = istate->cache[i - 1];
1742 struct cache_entry *next_ce = istate->cache[i];
1743 int name_compare = strcmp(ce->name, next_ce->name);
1744
1745 if (0 < name_compare)
1746 die("unordered stage entries in index");
1747 if (!name_compare) {
1748 if (!ce_stage(ce))
1749 die("multiple stage entries for merged file '%s'",
1750 ce->name);
1751 if (ce_stage(ce) > ce_stage(next_ce))
1752 die("unordered stage entries for '%s'",
1753 ce->name);
1754 }
15999d0b
JSP
1755 }
1756}
1757
435ec090
CC
1758static void tweak_untracked_cache(struct index_state *istate)
1759{
1760 switch (git_config_get_untracked_cache()) {
1761 case -1: /* keep: do nothing */
1762 break;
1763 case 0: /* false */
1764 remove_untracked_cache(istate);
1765 break;
1766 case 1: /* true */
1767 add_untracked_cache(istate);
1768 break;
1769 default: /* unknown value: do nothing */
1770 break;
1771 }
1772}
1773
43925312
CC
1774static void tweak_split_index(struct index_state *istate)
1775{
1776 switch (git_config_get_split_index()) {
1777 case -1: /* unset: do nothing */
1778 break;
1779 case 0: /* false */
1780 remove_split_index(istate);
1781 break;
1782 case 1: /* true */
1783 add_split_index(istate);
1784 break;
1785 default: /* unknown value: do nothing */
1786 break;
1787 }
1788}
1789
435ec090
CC
1790static void post_read_index_from(struct index_state *istate)
1791{
1792 check_ce_order(istate);
1793 tweak_untracked_cache(istate);
43925312 1794 tweak_split_index(istate);
883e248b 1795 tweak_fsmonitor(istate);
435ec090
CC
1796}
1797
8fd2cb40 1798/* remember to discard_cache() before reading a different cache! */
3e52f70b 1799int do_read_index(struct index_state *istate, const char *path, int must_exist)
e83c5163
LT
1800{
1801 int fd, i;
1802 struct stat st;
debed2a6 1803 unsigned long src_offset;
e83c5163 1804 struct cache_header *hdr;
7a51ed66
LT
1805 void *mmap;
1806 size_t mmap_size;
6c9cd161 1807 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
e83c5163 1808
913e0e99 1809 if (istate->initialized)
4aab5b46 1810 return istate->cache_nr;
5d1a5c02 1811
fba2f38a
KB
1812 istate->timestamp.sec = 0;
1813 istate->timestamp.nsec = 0;
8fd2cb40 1814 fd = open(path, O_RDONLY);
5d1a5c02 1815 if (fd < 0) {
5fc2fc8f 1816 if (!must_exist && errno == ENOENT)
5d1a5c02 1817 return 0;
5fc2fc8f 1818 die_errno("%s: index file open failed", path);
5d1a5c02 1819 }
e83c5163 1820
3511a377 1821 if (fstat(fd, &st))
d824cbba 1822 die_errno("cannot stat the open index");
3511a377 1823
7a51ed66 1824 mmap_size = xsize_t(st.st_size);
aab61359 1825 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
3511a377
LFC
1826 die("index file smaller than expected");
1827
a1293ef7 1828 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
7a51ed66 1829 if (mmap == MAP_FAILED)
0721c314 1830 die_errno("unable to map index file");
57d84f8d 1831 close(fd);
e83c5163 1832
7a51ed66
LT
1833 hdr = mmap;
1834 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1835 goto unmap;
1836
75691ea3 1837 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
9d227781 1838 istate->version = ntohl(hdr->hdr_version);
4aab5b46
JH
1839 istate->cache_nr = ntohl(hdr->hdr_entries);
1840 istate->cache_alloc = alloc_nr(istate->cache_nr);
c4aa3167 1841 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
913e0e99 1842 istate->initialized = 1;
7a51ed66 1843
9d227781 1844 if (istate->version == 4)
6c9cd161
JH
1845 previous_name = &previous_name_buf;
1846 else
1847 previous_name = NULL;
1848
7a51ed66 1849 src_offset = sizeof(*hdr);
4aab5b46 1850 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 1851 struct ondisk_cache_entry *disk_ce;
4aab5b46 1852 struct cache_entry *ce;
936f53d0 1853 unsigned long consumed;
4aab5b46 1854
7a51ed66 1855 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
6c9cd161 1856 ce = create_from_disk(disk_ce, &consumed, previous_name);
cf558704 1857 set_index_entry(istate, i, ce);
7a51ed66 1858
936f53d0 1859 src_offset += consumed;
e83c5163 1860 }
6c9cd161 1861 strbuf_release(&previous_name_buf);
fba2f38a 1862 istate->timestamp.sec = st.st_mtime;
c06ff490 1863 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 1864
aab61359 1865 while (src_offset <= mmap_size - the_hash_algo->rawsz - 8) {
bad68ec9
JH
1866 /* After an array of active_nr index entries,
1867 * there can be arbitrary number of extended
1868 * sections, each of which is prefixed with
1869 * extension name (4-byte) and section length
1870 * in 4-byte network byte order.
1871 */
07cc8eca 1872 uint32_t extsize;
7a51ed66 1873 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
bad68ec9 1874 extsize = ntohl(extsize);
4aab5b46 1875 if (read_index_extension(istate,
7a51ed66
LT
1876 (const char *) mmap + src_offset,
1877 (char *) mmap + src_offset + 8,
1d7f171c 1878 extsize) < 0)
bad68ec9 1879 goto unmap;
7a51ed66
LT
1880 src_offset += 8;
1881 src_offset += extsize;
bad68ec9 1882 }
7a51ed66 1883 munmap(mmap, mmap_size);
4aab5b46 1884 return istate->cache_nr;
e83c5163
LT
1885
1886unmap:
7a51ed66 1887 munmap(mmap, mmap_size);
5d1a5c02 1888 die("index file corrupt");
e83c5163
LT
1889}
1890
0d59ffb4
CC
1891/*
1892 * Signal that the shared index is used by updating its mtime.
1893 *
1894 * This way, shared index can be removed if they have not been used
1895 * for some time.
1896 */
a125a223 1897static void freshen_shared_index(const char *shared_index, int warn)
0d59ffb4 1898{
0d59ffb4
CC
1899 if (!check_and_freshen_file(shared_index, 1) && warn)
1900 warning("could not freshen shared index '%s'", shared_index);
1901}
1902
a125a223
TG
1903int read_index_from(struct index_state *istate, const char *path,
1904 const char *gitdir)
5fc2fc8f 1905{
ca54d9ba 1906 uint64_t start = getnanotime();
5fc2fc8f
NTND
1907 struct split_index *split_index;
1908 int ret;
2182abd9 1909 char *base_oid_hex;
a125a223 1910 char *base_path;
5fc2fc8f
NTND
1911
1912 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1913 if (istate->initialized)
1914 return istate->cache_nr;
1915
1916 ret = do_read_index(istate, path, 0);
ca54d9ba 1917 trace_performance_since(start, "read cache %s", path);
435ec090 1918
5fc2fc8f 1919 split_index = istate->split_index;
2182abd9 1920 if (!split_index || is_null_oid(&split_index->base_oid)) {
435ec090 1921 post_read_index_from(istate);
5fc2fc8f 1922 return ret;
03f15a79 1923 }
5fc2fc8f
NTND
1924
1925 if (split_index->base)
1926 discard_index(split_index->base);
1927 else
1928 split_index->base = xcalloc(1, sizeof(*split_index->base));
de6ae5f9 1929
2182abd9 1930 base_oid_hex = oid_to_hex(&split_index->base_oid);
1931 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
de6ae5f9 1932 ret = do_read_index(split_index->base, base_path, 1);
75691ea3 1933 if (oidcmp(&split_index->base_oid, &split_index->base->oid))
5fc2fc8f 1934 die("broken index, expect %s in %s, got %s",
2182abd9 1935 base_oid_hex, base_path,
75691ea3 1936 oid_to_hex(&split_index->base->oid));
de6ae5f9 1937
a125a223 1938 freshen_shared_index(base_path, 0);
5fc2fc8f 1939 merge_base_index(istate);
435ec090 1940 post_read_index_from(istate);
ca54d9ba 1941 trace_performance_since(start, "read cache %s", base_path);
a125a223 1942 free(base_path);
5fc2fc8f
NTND
1943 return ret;
1944}
1945
fa7b3c2f
JH
1946int is_index_unborn(struct index_state *istate)
1947{
debed2a6 1948 return (!istate->cache_nr && !istate->timestamp.sec);
fa7b3c2f
JH
1949}
1950
4aab5b46 1951int discard_index(struct index_state *istate)
6d297f81 1952{
debed2a6
RS
1953 int i;
1954
64719b11
JH
1955 for (i = 0; i < istate->cache_nr; i++) {
1956 if (istate->cache[i]->index &&
1957 istate->split_index &&
1958 istate->split_index->base &&
1959 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1960 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1961 continue;
debed2a6 1962 free(istate->cache[i]);
64719b11 1963 }
cfc5789a 1964 resolve_undo_clear_index(istate);
4aab5b46
JH
1965 istate->cache_nr = 0;
1966 istate->cache_changed = 0;
fba2f38a
KB
1967 istate->timestamp.sec = 0;
1968 istate->timestamp.nsec = 0;
2092678c 1969 free_name_hash(istate);
4aab5b46 1970 cache_tree_free(&(istate->cache_tree));
913e0e99 1971 istate->initialized = 0;
6a83d902 1972 FREE_AND_NULL(istate->cache);
a0fc4db0 1973 istate->cache_alloc = 0;
5fc2fc8f 1974 discard_split_index(istate);
f9e6c649
NTND
1975 free_untracked_cache(istate->untracked);
1976 istate->untracked = NULL;
7a51ed66 1977 return 0;
6d297f81
JS
1978}
1979
d1f128b0 1980int unmerged_index(const struct index_state *istate)
94a5728c
DB
1981{
1982 int i;
1983 for (i = 0; i < istate->cache_nr; i++) {
1984 if (ce_stage(istate->cache[i]))
1985 return 1;
1986 }
1987 return 0;
1988}
1989
4990aadc 1990#define WRITE_BUFFER_SIZE 8192
bf0f910d 1991static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
1992static unsigned long write_buffer_len;
1993
aab61359 1994static int ce_write_flush(git_hash_ctx *context, int fd)
6015c28b
JH
1995{
1996 unsigned int buffered = write_buffer_len;
1997 if (buffered) {
aab61359 1998 the_hash_algo->update_fn(context, write_buffer, buffered);
06f46f23 1999 if (write_in_full(fd, write_buffer, buffered) < 0)
6015c28b
JH
2000 return -1;
2001 write_buffer_len = 0;
2002 }
2003 return 0;
2004}
2005
aab61359 2006static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
4990aadc
LT
2007{
2008 while (len) {
2009 unsigned int buffered = write_buffer_len;
2010 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2011 if (partial > len)
2012 partial = len;
2013 memcpy(write_buffer + buffered, data, partial);
2014 buffered += partial;
2015 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
2016 write_buffer_len = buffered;
2017 if (ce_write_flush(context, fd))
4990aadc
LT
2018 return -1;
2019 buffered = 0;
2020 }
2021 write_buffer_len = buffered;
2022 len -= partial;
1d7f171c 2023 data = (char *) data + partial;
a6080a0a
JH
2024 }
2025 return 0;
4990aadc
LT
2026}
2027
aab61359 2028static int write_index_ext_header(git_hash_ctx *context, int fd,
ac58c7b1 2029 unsigned int ext, unsigned int sz)
bad68ec9
JH
2030{
2031 ext = htonl(ext);
2032 sz = htonl(sz);
968a1d65
DR
2033 return ((ce_write(context, fd, &ext, 4) < 0) ||
2034 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
2035}
2036
aab61359 2037static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
4990aadc
LT
2038{
2039 unsigned int left = write_buffer_len;
ca9be054 2040
4990aadc
LT
2041 if (left) {
2042 write_buffer_len = 0;
aab61359 2043 the_hash_algo->update_fn(context, write_buffer, left);
4990aadc 2044 }
ca9be054 2045
aab61359 2046 /* Flush first if not enough space for hash signature */
2047 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
06f46f23 2048 if (write_in_full(fd, write_buffer, left) < 0)
2c865d9a
QH
2049 return -1;
2050 left = 0;
2051 }
2052
aab61359 2053 /* Append the hash signature at the end */
2054 the_hash_algo->final_fn(write_buffer + left, context);
2055 hashcpy(hash, write_buffer + left);
2056 left += the_hash_algo->rawsz;
06f46f23 2057 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
4990aadc
LT
2058}
2059
407c8eb0
JH
2060static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2061{
2062 /*
2063 * The only thing we care about in this function is to smudge the
2064 * falsely clean entry due to touch-update-touch race, so we leave
2065 * everything else as they are. We are called for entries whose
c21d39d7 2066 * ce_stat_data.sd_mtime match the index file mtime.
c70115b4
JH
2067 *
2068 * Note that this actually does not do much for gitlinks, for
2069 * which ce_match_stat_basic() always goes to the actual
2070 * contents. The caller checks with is_racy_timestamp() which
2071 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
2072 */
2073 struct stat st;
2074
2075 if (lstat(ce->name, &st) < 0)
2076 return;
2077 if (ce_match_stat_basic(ce, &st))
2078 return;
2079 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
2080 /* This is "racily clean"; smudge it. Note that this
2081 * is a tricky code. At first glance, it may appear
2082 * that it can break with this sequence:
2083 *
2084 * $ echo xyzzy >frotz
2085 * $ git-update-index --add frotz
2086 * $ : >frotz
2087 * $ sleep 3
2088 * $ echo filfre >nitfol
2089 * $ git-update-index --add nitfol
2090 *
b7e58b17 2091 * but it does not. When the second update-index runs,
4b3511b0
JH
2092 * it notices that the entry "frotz" has the same timestamp
2093 * as index, and if we were to smudge it by resetting its
2094 * size to zero here, then the object name recorded
2095 * in index is the 6-byte file but the cached stat information
2096 * becomes zero --- which would then match what we would
a6080a0a 2097 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
2098 *
2099 * However, the second update-index, before calling
2100 * this function, notices that the cached size is 6
2101 * bytes and what is on the filesystem is an empty
2102 * file, and never calls us, so the cached size information
2103 * for "frotz" stays 6 which does not match the filesystem.
2104 */
c21d39d7 2105 ce->ce_stat_data.sd_size = 0;
407c8eb0
JH
2106 }
2107}
2108
f136f7bf 2109/* Copy miscellaneous fields but not the name */
ce012deb 2110static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
f136f7bf 2111 struct cache_entry *ce)
7a51ed66 2112{
b60e188c
TG
2113 short flags;
2114
c21d39d7
MH
2115 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2116 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2117 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2118 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2119 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2120 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
7a51ed66 2121 ondisk->mode = htonl(ce->ce_mode);
c21d39d7
MH
2122 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2123 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2124 ondisk->size = htonl(ce->ce_stat_data.sd_size);
99d1a986 2125 hashcpy(ondisk->sha1, ce->oid.hash);
b60e188c 2126
ce51bf09 2127 flags = ce->ce_flags & ~CE_NAMEMASK;
b60e188c
TG
2128 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2129 ondisk->flags = htons(flags);
06aaaa0b
NTND
2130 if (ce->ce_flags & CE_EXTENDED) {
2131 struct ondisk_cache_entry_extended *ondisk2;
2132 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2133 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
f136f7bf
JH
2134 }
2135}
2136
aab61359 2137static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
ce012deb 2138 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
f136f7bf 2139{
9d227781 2140 int size;
f136f7bf 2141 int result;
00a4b035
RJ
2142 unsigned int saved_namelen;
2143 int stripped_name = 0;
ce012deb 2144 static unsigned char padding[8] = { 0x00 };
f136f7bf 2145
b3c96fb1
NTND
2146 if (ce->ce_flags & CE_STRIP_NAME) {
2147 saved_namelen = ce_namelen(ce);
2148 ce->ce_namelen = 0;
00a4b035 2149 stripped_name = 1;
b3c96fb1
NTND
2150 }
2151
ce012deb
KW
2152 if (ce->ce_flags & CE_EXTENDED)
2153 size = offsetof(struct ondisk_cache_entry_extended, name);
2154 else
2155 size = offsetof(struct ondisk_cache_entry, name);
2156
9d227781 2157 if (!previous_name) {
ce012deb
KW
2158 int len = ce_namelen(ce);
2159 copy_cache_entry_to_ondisk(ondisk, ce);
2160 result = ce_write(c, fd, ondisk, size);
2161 if (!result)
2162 result = ce_write(c, fd, ce->name, len);
2163 if (!result)
2164 result = ce_write(c, fd, padding, align_padding_size(size, len));
9d227781
JH
2165 } else {
2166 int common, to_remove, prefix_size;
2167 unsigned char to_remove_vi[16];
2168 for (common = 0;
2169 (ce->name[common] &&
2170 common < previous_name->len &&
2171 ce->name[common] == previous_name->buf[common]);
2172 common++)
2173 ; /* still matching */
2174 to_remove = previous_name->len - common;
2175 prefix_size = encode_varint(to_remove, to_remove_vi);
2176
ce012deb
KW
2177 copy_cache_entry_to_ondisk(ondisk, ce);
2178 result = ce_write(c, fd, ondisk, size);
2179 if (!result)
2180 result = ce_write(c, fd, to_remove_vi, prefix_size);
2181 if (!result)
0b90b881
TG
2182 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2183 if (!result)
2184 result = ce_write(c, fd, padding, 1);
9d227781
JH
2185
2186 strbuf_splice(previous_name, common, to_remove,
2187 ce->name + common, ce_namelen(ce) - common);
06aaaa0b 2188 }
00a4b035 2189 if (stripped_name) {
b3c96fb1
NTND
2190 ce->ce_namelen = saved_namelen;
2191 ce->ce_flags &= ~CE_STRIP_NAME;
2192 }
7a51ed66 2193
59efba64 2194 return result;
7a51ed66
LT
2195}
2196
426ddeea
YM
2197/*
2198 * This function verifies if index_state has the correct sha1 of the
2199 * index file. Don't die if we have any other failure, just return 0.
2200 */
2201static int verify_index_from(const struct index_state *istate, const char *path)
2202{
2203 int fd;
2204 ssize_t n;
2205 struct stat st;
aab61359 2206 unsigned char hash[GIT_MAX_RAWSZ];
426ddeea
YM
2207
2208 if (!istate->initialized)
2209 return 0;
2210
2211 fd = open(path, O_RDONLY);
2212 if (fd < 0)
2213 return 0;
2214
2215 if (fstat(fd, &st))
2216 goto out;
2217
aab61359 2218 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
426ddeea
YM
2219 goto out;
2220
aab61359 2221 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2222 if (n != the_hash_algo->rawsz)
426ddeea
YM
2223 goto out;
2224
75691ea3 2225 if (hashcmp(istate->oid.hash, hash))
426ddeea
YM
2226 goto out;
2227
2228 close(fd);
2229 return 1;
2230
2231out:
2232 close(fd);
2233 return 0;
2234}
2235
2236static int verify_index(const struct index_state *istate)
2237{
2238 return verify_index_from(istate, get_index_file());
2239}
2240
483fbe2b
JH
2241static int has_racy_timestamp(struct index_state *istate)
2242{
2243 int entries = istate->cache_nr;
2244 int i;
2245
2246 for (i = 0; i < entries; i++) {
2247 struct cache_entry *ce = istate->cache[i];
2248 if (is_racy_timestamp(istate, ce))
2249 return 1;
2250 }
2251 return 0;
2252}
2253
ccdc4ec3
JH
2254void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2255{
483fbe2b 2256 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
df60cf57
2257 verify_index(istate))
2258 write_locked_index(istate, lockfile, COMMIT_LOCK);
b74c90fb
2259 else
2260 rollback_lock_file(lockfile);
ccdc4ec3
JH
2261}
2262
812d6b00
2263/*
2264 * On success, `tempfile` is closed. If it is the temporary file
2265 * of a `struct lock_file`, we will therefore effectively perform
2266 * a 'close_lock_file_gently()`. Since that is an implementation
2267 * detail of lockfiles, callers of `do_write_index()` should not
2268 * rely on it.
2269 */
9f41c7a6 2270static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
c18b80a0 2271 int strip_extensions)
197ee8c9 2272{
ca54d9ba 2273 uint64_t start = getnanotime();
9f41c7a6 2274 int newfd = tempfile->fd;
aab61359 2275 git_hash_ctx c;
197ee8c9 2276 struct cache_header hdr;
b50386c7 2277 int i, err = 0, removed, extended, hdr_version;
4aab5b46
JH
2278 struct cache_entry **cache = istate->cache;
2279 int entries = istate->cache_nr;
e1afca4f 2280 struct stat st;
ce012deb 2281 struct ondisk_cache_entry_extended ondisk;
9d227781 2282 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
4bddd983 2283 int drop_cache_tree = istate->drop_cache_tree;
025a0709 2284
06aaaa0b 2285 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 2286 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 2287 removed++;
197ee8c9 2288
06aaaa0b
NTND
2289 /* reduce extended entries if possible */
2290 cache[i]->ce_flags &= ~CE_EXTENDED;
2291 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2292 extended++;
2293 cache[i]->ce_flags |= CE_EXTENDED;
2294 }
2295 }
2296
d6e3c181 2297 if (!istate->version) {
136347d7 2298 istate->version = get_index_format_default();
4c2db938 2299 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
d6e3c181
NTND
2300 init_split_index(istate);
2301 }
9d227781
JH
2302
2303 /* demote version 3 to version 2 when the latter suffices */
2304 if (istate->version == 3 || istate->version == 2)
2305 istate->version = extended ? 3 : 2;
2306
2307 hdr_version = istate->version;
2308
ccc4feb5 2309 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
9d227781 2310 hdr.hdr_version = htonl(hdr_version);
025a0709 2311 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 2312
aab61359 2313 the_hash_algo->init_fn(&c);
ca9be054 2314 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
2315 return -1;
2316
9d227781 2317 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
ce012deb 2318
197ee8c9
LT
2319 for (i = 0; i < entries; i++) {
2320 struct cache_entry *ce = cache[i];
7a51ed66 2321 if (ce->ce_flags & CE_REMOVE)
aa16021e 2322 continue;
e06c43c7 2323 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 2324 ce_smudge_racily_clean_entry(ce);
99d1a986 2325 if (is_null_oid(&ce->oid)) {
83bd7437
JK
2326 static const char msg[] = "cache entry has null sha1: %s";
2327 static int allow = -1;
2328
2329 if (allow < 0)
2330 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2331 if (allow)
2332 warning(msg, ce->name);
2333 else
b50386c7 2334 err = error(msg, ce->name);
a96d3cc3
JK
2335
2336 drop_cache_tree = 1;
83bd7437 2337 }
ce012deb 2338 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
b50386c7
KW
2339 err = -1;
2340
2341 if (err)
2342 break;
197ee8c9 2343 }
9d227781 2344 strbuf_release(&previous_name_buf);
1af1c2b6 2345
b50386c7
KW
2346 if (err)
2347 return err;
2348
bad68ec9 2349 /* Write extension data here */
c18b80a0 2350 if (!strip_extensions && istate->split_index) {
5fc2fc8f
NTND
2351 struct strbuf sb = STRBUF_INIT;
2352
2353 err = write_link_extension(&sb, istate) < 0 ||
2354 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2355 sb.len) < 0 ||
2356 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2357 strbuf_release(&sb);
2358 if (err)
2359 return -1;
2360 }
a96d3cc3 2361 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
f285a2d7 2362 struct strbuf sb = STRBUF_INIT;
1dffb8fa 2363
1dffb8fa
PH
2364 cache_tree_write(&sb, istate->cache_tree);
2365 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2366 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2367 strbuf_release(&sb);
2368 if (err)
bad68ec9 2369 return -1;
bad68ec9 2370 }
c18b80a0 2371 if (!strip_extensions && istate->resolve_undo) {
cfc5789a
JH
2372 struct strbuf sb = STRBUF_INIT;
2373
2374 resolve_undo_write(&sb, istate->resolve_undo);
2375 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2376 sb.len) < 0
2377 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2378 strbuf_release(&sb);
2379 if (err)
2380 return -1;
2381 }
83c094ad
NTND
2382 if (!strip_extensions && istate->untracked) {
2383 struct strbuf sb = STRBUF_INIT;
2384
2385 write_untracked_extension(&sb, istate->untracked);
2386 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2387 sb.len) < 0 ||
2388 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2389 strbuf_release(&sb);
2390 if (err)
2391 return -1;
2392 }
883e248b
BP
2393 if (!strip_extensions && istate->fsmonitor_last_update) {
2394 struct strbuf sb = STRBUF_INIT;
2395
2396 write_fsmonitor_extension(&sb, istate);
2397 err = write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2398 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2399 strbuf_release(&sb);
2400 if (err)
2401 return -1;
2402 }
e1afca4f 2403
75691ea3 2404 if (ce_flush(&c, newfd, istate->oid.hash))
9f41c7a6 2405 return -1;
49bd0fc2
JK
2406 if (close_tempfile_gently(tempfile)) {
2407 error(_("could not close '%s'"), tempfile->filename.buf);
49bd0fc2
JK
2408 return -1;
2409 }
9f41c7a6 2410 if (stat(tempfile->filename.buf, &st))
e1afca4f 2411 return -1;
5bcf109c
KB
2412 istate->timestamp.sec = (unsigned int)st.st_mtime;
2413 istate->timestamp.nsec = ST_MTIME_NSEC(st);
ca54d9ba 2414 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
e1afca4f 2415 return 0;
197ee8c9 2416}
e46bbcf6 2417
626f35c8
NTND
2418void set_alternate_index_output(const char *name)
2419{
2420 alternate_index_output = name;
2421}
2422
2423static int commit_locked_index(struct lock_file *lk)
2424{
751baced
MH
2425 if (alternate_index_output)
2426 return commit_lock_file_to(lk, alternate_index_output);
2427 else
626f35c8 2428 return commit_lock_file(lk);
626f35c8
NTND
2429}
2430
03b86647
NTND
2431static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2432 unsigned flags)
2433{
076aa2cb 2434 int ret = do_write_index(istate, lock->tempfile, 0);
03b86647
NTND
2435 if (ret)
2436 return ret;
03b86647
NTND
2437 if (flags & COMMIT_LOCK)
2438 return commit_locked_index(lock);
812d6b00 2439 return close_lock_file_gently(lock);
03b86647
NTND
2440}
2441
5fc2fc8f
NTND
2442static int write_split_index(struct index_state *istate,
2443 struct lock_file *lock,
2444 unsigned flags)
2445{
2446 int ret;
2447 prepare_to_write_split_index(istate);
2448 ret = do_write_locked_index(istate, lock, flags);
2449 finish_writing_split_index(istate);
2450 return ret;
2451}
2452
b9683722
CC
2453static const char *shared_index_expire = "2.weeks.ago";
2454
2455static unsigned long get_shared_index_expire_date(void)
2456{
2457 static unsigned long shared_index_expire_date;
2458 static int shared_index_expire_date_prepared;
2459
2460 if (!shared_index_expire_date_prepared) {
2461 git_config_get_expiry("splitindex.sharedindexexpire",
2462 &shared_index_expire);
2463 shared_index_expire_date = approxidate(shared_index_expire);
2464 shared_index_expire_date_prepared = 1;
2465 }
2466
2467 return shared_index_expire_date;
2468}
2469
2470static int should_delete_shared_index(const char *shared_index_path)
2471{
2472 struct stat st;
2473 unsigned long expiration;
2474
2475 /* Check timestamp */
2476 expiration = get_shared_index_expire_date();
2477 if (!expiration)
2478 return 0;
2479 if (stat(shared_index_path, &st))
78bde923 2480 return error_errno(_("could not stat '%s'"), shared_index_path);
b9683722
CC
2481 if (st.st_mtime > expiration)
2482 return 0;
2483
2484 return 1;
2485}
2486
2487static int clean_shared_index_files(const char *current_hex)
2488{
2489 struct dirent *de;
2490 DIR *dir = opendir(get_git_dir());
2491
2492 if (!dir)
2493 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2494
2495 while ((de = readdir(dir)) != NULL) {
2496 const char *sha1_hex;
2497 const char *shared_index_path;
2498 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2499 continue;
2500 if (!strcmp(sha1_hex, current_hex))
2501 continue;
2502 shared_index_path = git_path("%s", de->d_name);
2503 if (should_delete_shared_index(shared_index_path) > 0 &&
2504 unlink(shared_index_path))
2505 warning_errno(_("unable to unlink: %s"), shared_index_path);
2506 }
2507 closedir(dir);
2508
2509 return 0;
2510}
2511
a0a96756 2512static int write_shared_index(struct index_state *istate,
59f9d2dd 2513 struct tempfile **temp)
c18b80a0
NTND
2514{
2515 struct split_index *si = istate->split_index;
076aa2cb 2516 int ret;
c18b80a0 2517
c18b80a0 2518 move_cache_to_base_index(istate);
7db2d08c 2519 ret = do_write_index(si->base, *temp, 1);
59f9d2dd 2520 if (ret)
c18b80a0 2521 return ret;
7db2d08c 2522 ret = adjust_shared_perm(get_tempfile_path(*temp));
df801f3f 2523 if (ret) {
7db2d08c 2524 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
df801f3f
CC
2525 return ret;
2526 }
7db2d08c 2527 ret = rename_tempfile(temp,
75691ea3 2528 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
b9683722 2529 if (!ret) {
75691ea3 2530 oidcpy(&si->base_oid, &si->base->oid);
2531 clean_shared_index_files(oid_to_hex(&si->base->oid));
b9683722
CC
2532 }
2533
c18b80a0
NTND
2534 return ret;
2535}
2536
e6a1dd77
CC
2537static const int default_max_percent_split_change = 20;
2538
2539static int too_many_not_shared_entries(struct index_state *istate)
2540{
2541 int i, not_shared = 0;
2542 int max_split = git_config_get_max_percent_split_change();
2543
2544 switch (max_split) {
2545 case -1:
2546 /* not or badly configured: use the default value */
2547 max_split = default_max_percent_split_change;
2548 break;
2549 case 0:
2550 return 1; /* 0% means always write a new shared index */
2551 case 100:
2552 return 0; /* 100% means never write a new shared index */
2553 default:
2554 break; /* just use the configured value */
2555 }
2556
2557 /* Count not shared entries */
2558 for (i = 0; i < istate->cache_nr; i++) {
2559 struct cache_entry *ce = istate->cache[i];
2560 if (!ce->index)
2561 not_shared++;
2562 }
2563
2564 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2565}
2566
03b86647
NTND
2567int write_locked_index(struct index_state *istate, struct lock_file *lock,
2568 unsigned flags)
2569{
0d59ffb4 2570 int new_shared_index, ret;
5fc2fc8f
NTND
2571 struct split_index *si = istate->split_index;
2572
61000814
2573 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
2574 if (flags & COMMIT_LOCK)
2575 rollback_lock_file(lock);
2576 return 0;
2577 }
2578
3bd28eb2
AV
2579 if (istate->fsmonitor_last_update)
2580 fill_fsmonitor_bitmap(istate);
2581
5165dd59
NTND
2582 if (!si || alternate_index_output ||
2583 (istate->cache_changed & ~EXTMASK)) {
5fc2fc8f 2584 if (si)
2182abd9 2585 oidclr(&si->base_oid);
df60cf57
2586 ret = do_write_locked_index(istate, lock, flags);
2587 goto out;
5fc2fc8f
NTND
2588 }
2589
4c2db938 2590 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2182abd9 2591 int v = si->base_oid.hash[0];
d6e3c181
NTND
2592 if ((v & 15) < 6)
2593 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2594 }
e6a1dd77
CC
2595 if (too_many_not_shared_entries(istate))
2596 istate->cache_changed |= SPLIT_INDEX_ORDERED;
0d59ffb4
CC
2597
2598 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2599
2600 if (new_shared_index) {
59f9d2dd
NTND
2601 struct tempfile *temp;
2602 int saved_errno;
2603
2604 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2605 if (!temp) {
2182abd9 2606 oidclr(&si->base_oid);
59f9d2dd 2607 ret = do_write_locked_index(istate, lock, flags);
ef5b3a6c
NTND
2608 goto out;
2609 }
2610 ret = write_shared_index(istate, &temp);
59f9d2dd
NTND
2611
2612 saved_errno = errno;
2613 if (is_tempfile_active(temp))
2614 delete_tempfile(&temp);
2615 errno = saved_errno;
2616
c18b80a0 2617 if (ret)
df60cf57 2618 goto out;
c18b80a0
NTND
2619 }
2620
0d59ffb4
CC
2621 ret = write_split_index(istate, lock, flags);
2622
2623 /* Freshen the shared index only if the split-index was written */
a125a223
TG
2624 if (!ret && !new_shared_index) {
2625 const char *shared_index = git_path("sharedindex.%s",
2182abd9 2626 oid_to_hex(&si->base_oid));
a125a223
TG
2627 freshen_shared_index(shared_index, 1);
2628 }
0d59ffb4 2629
df60cf57
2630out:
2631 if (flags & COMMIT_LOCK)
2632 rollback_lock_file(lock);
0d59ffb4 2633 return ret;
03b86647
NTND
2634}
2635
e46bbcf6
MV
2636/*
2637 * Read the index file that is potentially unmerged into given
63e8dc5b 2638 * index_state, dropping any unmerged entries. Returns true if
e46bbcf6
MV
2639 * the index is unmerged. Callers who want to refuse to work
2640 * from an unmerged state can call this and check its return value,
2641 * instead of calling read_cache().
2642 */
2643int read_index_unmerged(struct index_state *istate)
2644{
2645 int i;
d1a43f2a 2646 int unmerged = 0;
e46bbcf6
MV
2647
2648 read_index(istate);
e46bbcf6
MV
2649 for (i = 0; i < istate->cache_nr; i++) {
2650 struct cache_entry *ce = istate->cache[i];
d1a43f2a
JH
2651 struct cache_entry *new_ce;
2652 int size, len;
2653
2654 if (!ce_stage(ce))
e46bbcf6 2655 continue;
d1a43f2a 2656 unmerged = 1;
68c4f6a5 2657 len = ce_namelen(ce);
d1a43f2a
JH
2658 size = cache_entry_size(len);
2659 new_ce = xcalloc(1, size);
d1a43f2a 2660 memcpy(new_ce->name, ce->name, len);
b60e188c
TG
2661 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2662 new_ce->ce_namelen = len;
d1a43f2a
JH
2663 new_ce->ce_mode = ce->ce_mode;
2664 if (add_index_entry(istate, new_ce, 0))
2665 return error("%s: cannot drop to stage #0",
5699d17e 2666 new_ce->name);
e46bbcf6 2667 }
d1a43f2a 2668 return unmerged;
e46bbcf6 2669}
041aee31 2670
98fa4738
JK
2671/*
2672 * Returns 1 if the path is an "other" path with respect to
2673 * the index; that is, the path is not mentioned in the index at all,
2674 * either as a file, a directory with some files in the index,
2675 * or as an unmerged entry.
2676 *
2677 * We helpfully remove a trailing "/" from directories so that
2678 * the output of read_directory can be used as-is.
2679 */
2680int index_name_is_other(const struct index_state *istate, const char *name,
2681 int namelen)
2682{
2683 int pos;
2684 if (namelen && name[namelen - 1] == '/')
2685 namelen--;
2686 pos = index_name_pos(istate, name, namelen);
2687 if (0 <= pos)
2688 return 0; /* exact match */
2689 pos = -pos - 1;
2690 if (pos < istate->cache_nr) {
2691 struct cache_entry *ce = istate->cache[pos];
2692 if (ce_namelen(ce) == namelen &&
2693 !memcmp(ce->name, name, namelen))
2694 return 0; /* Yup, this one exists unmerged */
2695 }
2696 return 1;
2697}
29fb37b2 2698
87542508
BW
2699void *read_blob_data_from_index(const struct index_state *istate,
2700 const char *path, unsigned long *size)
29fb37b2
LF
2701{
2702 int pos, len;
2703 unsigned long sz;
2704 enum object_type type;
2705 void *data;
2706
2707 len = strlen(path);
2708 pos = index_name_pos(istate, path, len);
2709 if (pos < 0) {
2710 /*
2711 * We might be in the middle of a merge, in which
2712 * case we would read stage #2 (ours).
2713 */
2714 int i;
2715 for (i = -pos - 1;
2716 (pos < 0 && i < istate->cache_nr &&
2717 !strcmp(istate->cache[i]->name, path));
2718 i++)
2719 if (ce_stage(istate->cache[i]) == 2)
2720 pos = i;
2721 }
2722 if (pos < 0)
2723 return NULL;
b4f5aca4 2724 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
29fb37b2
LF
2725 if (!data || type != OBJ_BLOB) {
2726 free(data);
2727 return NULL;
2728 }
ff366825
LF
2729 if (size)
2730 *size = sz;
29fb37b2
LF
2731 return data;
2732}
38612532
MH
2733
2734void stat_validity_clear(struct stat_validity *sv)
2735{
6a83d902 2736 FREE_AND_NULL(sv->sd);
38612532
MH
2737}
2738
2739int stat_validity_check(struct stat_validity *sv, const char *path)
2740{
2741 struct stat st;
2742
2743 if (stat(path, &st) < 0)
2744 return sv->sd == NULL;
2745 if (!sv->sd)
2746 return 0;
2747 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2748}
2749
2750void stat_validity_update(struct stat_validity *sv, int fd)
2751{
2752 struct stat st;
2753
2754 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2755 stat_validity_clear(sv);
2756 else {
2757 if (!sv->sd)
2758 sv->sd = xcalloc(1, sizeof(struct stat_data));
2759 fill_stat_data(sv->sd, &st);
2760 }
2761}
edf3b905
DT
2762
2763void move_index_extensions(struct index_state *dst, struct index_state *src)
2764{
2765 dst->untracked = src->untracked;
2766 src->untracked = NULL;
2767}