t1700: test shared index file expiration
[git/git.git] / read-cache.c
CommitLineData
8bc9a0c7
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
4aab5b46 6#define NO_THE_INDEX_COMPATIBILITY_MACROS
e83c5163 7#include "cache.h"
f6ecc62d 8#include "tempfile.h"
697cc8ef 9#include "lockfile.h"
bad68ec9 10#include "cache-tree.h"
f35a6d3b 11#include "refs.h"
d616813d 12#include "dir.h"
041aee31
JH
13#include "tree.h"
14#include "commit.h"
39425819 15#include "blob.h"
cfc5789a 16#include "resolve-undo.h"
6c9cd161
JH
17#include "strbuf.h"
18#include "varint.h"
5fc2fc8f 19#include "split-index.h"
a42643aa 20#include "utf8.h"
bad68ec9 21
b60e188c
TG
22/* Mask for the name length in ce_flags in the on-disk index */
23
24#define CE_NAMEMASK (0x0fff)
25
bad68ec9
JH
26/* Index extensions.
27 *
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
33 */
34
35#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
b659b49b 37#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
5fc2fc8f 38#define CACHE_EXT_LINK 0x6c696e6b /* "link" */
83c094ad 39#define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
5fc2fc8f
NTND
40
41/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
e0cf0d7d 42#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
c18b80a0 43 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
1bbb3dba 44 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
e83c5163 45
228e94f9 46struct index_state the_index;
626f35c8 47static const char *alternate_index_output;
8fd2cb40 48
9cb76b8c
JH
49static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
50{
51 istate->cache[nr] = ce;
96872bc2 52 add_name_hash(istate, ce);
9cb76b8c
JH
53}
54
cf558704
LT
55static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
56{
57 struct cache_entry *old = istate->cache[nr];
58
078a58e8 59 replace_index_entry_in_base(istate, old, ce);
2092678c 60 remove_name_hash(istate, old);
5699d17e 61 free(old);
a22c6371 62 set_index_entry(istate, nr, ce);
078a58e8 63 ce->ce_flags |= CE_UPDATE_IN_BASE;
e636a7b4 64 istate->cache_changed |= CE_ENTRY_CHANGED;
cf558704
LT
65}
66
81dc2307
PB
67void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
68{
69 struct cache_entry *old = istate->cache[nr], *new;
70 int namelen = strlen(new_name);
71
72 new = xmalloc(cache_entry_size(namelen));
73 copy_cache_entry(new, old);
419a597f 74 new->ce_flags &= ~CE_HASHED;
b60e188c 75 new->ce_namelen = namelen;
5fc2fc8f 76 new->index = 0;
81dc2307
PB
77 memcpy(new->name, new_name, namelen + 1);
78
a5400efe 79 cache_tree_invalidate_path(istate, old->name);
e931371a 80 untracked_cache_remove_from_index(istate, old->name);
81dc2307
PB
81 remove_index_entry_at(istate, nr);
82 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
83}
84
c21d39d7
MH
85void fill_stat_data(struct stat_data *sd, struct stat *st)
86{
87 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
88 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
89 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
90 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
91 sd->sd_dev = st->st_dev;
92 sd->sd_ino = st->st_ino;
93 sd->sd_uid = st->st_uid;
94 sd->sd_gid = st->st_gid;
95 sd->sd_size = st->st_size;
96}
97
98int match_stat_data(const struct stat_data *sd, struct stat *st)
99{
100 int changed = 0;
101
102 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
103 changed |= MTIME_CHANGED;
104 if (trust_ctime && check_stat &&
105 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
106 changed |= CTIME_CHANGED;
107
108#ifdef USE_NSEC
109 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
110 changed |= MTIME_CHANGED;
111 if (trust_ctime && check_stat &&
112 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
113 changed |= CTIME_CHANGED;
114#endif
115
116 if (check_stat) {
117 if (sd->sd_uid != (unsigned int) st->st_uid ||
118 sd->sd_gid != (unsigned int) st->st_gid)
119 changed |= OWNER_CHANGED;
120 if (sd->sd_ino != (unsigned int) st->st_ino)
121 changed |= INODE_CHANGED;
122 }
123
124#ifdef USE_STDEV
125 /*
126 * st_dev breaks on network filesystems where different
127 * clients will have different views of what "device"
128 * the filesystem is on
129 */
130 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
131 changed |= INODE_CHANGED;
132#endif
133
134 if (sd->sd_size != (unsigned int) st->st_size)
135 changed |= DATA_CHANGED;
136
137 return changed;
138}
139
415e96c8
JH
140/*
141 * This only updates the "non-critical" parts of the directory
142 * cache, ie the parts that aren't tracked by GIT, and only used
143 * to validate the cache.
144 */
145void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
146{
c21d39d7 147 fill_stat_data(&ce->ce_stat_data, st);
5f73076c
JH
148
149 if (assume_unchanged)
7a51ed66 150 ce->ce_flags |= CE_VALID;
eadb5831
JH
151
152 if (S_ISREG(st->st_mode))
153 ce_mark_uptodate(ce);
415e96c8
JH
154}
155
21a6b9fa 156static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
157{
158 int match = -1;
a0a6cb96
LS
159 static int cloexec = O_CLOEXEC;
160 int fd = open(ce->name, O_RDONLY | cloexec);
161
162 if ((cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
163 /* Try again w/o O_CLOEXEC: the kernel might not support it */
164 cloexec &= ~O_CLOEXEC;
165 fd = open(ce->name, O_RDONLY | cloexec);
166 }
29e4d363
JH
167
168 if (fd >= 0) {
169 unsigned char sha1[20];
c4ce46fc 170 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
99d1a986 171 match = hashcmp(sha1, ce->oid.hash);
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
99d1a986 188 buffer = read_sha1_file(ce->oid.hash, &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
LT
199{
200 unsigned char sha1[20];
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 */
210 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
211 return 0;
99d1a986 212 return hashcmp(sha1, ce->oid.hash);
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;
29e4d363
JH
229 default:
230 return TYPE_CHANGED;
231 }
232 return 0;
233}
234
21a6b9fa 235static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
734aab75
LT
236{
237 unsigned int changed = 0;
238
7a51ed66
LT
239 if (ce->ce_flags & CE_REMOVE)
240 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
241
242 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
243 case S_IFREG:
244 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
245 /* We consider only the owner x bit to be relevant for
246 * "mode changes"
247 */
248 if (trust_executable_bit &&
7a51ed66 249 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 250 changed |= MODE_CHANGED;
8ae0a8c5
KS
251 break;
252 case S_IFLNK:
78a8d641
JS
253 if (!S_ISLNK(st->st_mode) &&
254 (has_symlinks || !S_ISREG(st->st_mode)))
255 changed |= TYPE_CHANGED;
8ae0a8c5 256 break;
302b9282 257 case S_IFGITLINK:
c70115b4 258 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
259 if (!S_ISDIR(st->st_mode))
260 changed |= TYPE_CHANGED;
261 else if (ce_compare_gitlink(ce))
262 changed |= DATA_CHANGED;
a8ee75bc 263 return changed;
8ae0a8c5 264 default:
7a51ed66 265 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 266 }
2cb45e95 267
c21d39d7 268 changed |= match_stat_data(&ce->ce_stat_data, st);
b0391890 269
f49c2c22 270 /* Racily smudged entry? */
c21d39d7 271 if (!ce->ce_stat_data.sd_size) {
99d1a986 272 if (!is_empty_blob_sha1(ce->oid.hash))
f49c2c22
LT
273 changed |= DATA_CHANGED;
274 }
275
407c8eb0
JH
276 return changed;
277}
278
2bb4cda1
NTND
279static int is_racy_stat(const struct index_state *istate,
280 const struct stat_data *sd)
6d91da6d 281{
2bb4cda1 282 return (istate->timestamp.sec &&
fba2f38a
KB
283#ifdef USE_NSEC
284 /* nanosecond timestamped files can also be racy! */
2bb4cda1
NTND
285 (istate->timestamp.sec < sd->sd_mtime.sec ||
286 (istate->timestamp.sec == sd->sd_mtime.sec &&
287 istate->timestamp.nsec <= sd->sd_mtime.nsec))
fba2f38a 288#else
2bb4cda1 289 istate->timestamp.sec <= sd->sd_mtime.sec
fba2f38a 290#endif
2bb4cda1
NTND
291 );
292}
293
294static int is_racy_timestamp(const struct index_state *istate,
295 const struct cache_entry *ce)
296{
297 return (!S_ISGITLINK(ce->ce_mode) &&
298 is_racy_stat(istate, &ce->ce_stat_data));
6d91da6d
JH
299}
300
ed4efab1
NTND
301int match_stat_data_racy(const struct index_state *istate,
302 const struct stat_data *sd, struct stat *st)
303{
304 if (is_racy_stat(istate, sd))
305 return MTIME_CHANGED;
306 return match_stat_data(sd, st);
6d91da6d
JH
307}
308
d1f128b0 309int ie_match_stat(const struct index_state *istate,
21a6b9fa 310 const struct cache_entry *ce, struct stat *st,
4bd5b7da 311 unsigned int options)
407c8eb0 312{
5f73076c 313 unsigned int changed;
4bd5b7da 314 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 315 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
4bd5b7da 316 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
5f73076c
JH
317
318 /*
319 * If it's marked as always valid in the index, it's
320 * valid whatever the checked-out copy says.
56cac48c
NTND
321 *
322 * skip-worktree has the same effect with higher precedence
5f73076c 323 */
56cac48c
NTND
324 if (!ignore_skip_worktree && ce_skip_worktree(ce))
325 return 0;
7a51ed66 326 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c
JH
327 return 0;
328
331fcb59
JH
329 /*
330 * Intent-to-add entries have not been added, so the index entry
331 * by definition never matches what is in the work tree until it
332 * actually gets added.
333 */
895ff3b2 334 if (ce_intent_to_add(ce))
331fcb59
JH
335 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
336
5f73076c 337 changed = ce_match_stat_basic(ce, st);
407c8eb0 338
29e4d363
JH
339 /*
340 * Within 1 second of this sequence:
341 * echo xyzzy >file && git-update-index --add file
342 * running this command:
343 * echo frotz >file
344 * would give a falsely clean cache entry. The mtime and
345 * length match the cache, and other stat fields do not change.
346 *
347 * We could detect this at update-index time (the cache entry
348 * being registered/updated records the same time as "now")
349 * and delay the return from git-update-index, but that would
350 * effectively mean we can make at most one commit per second,
351 * which is not acceptable. Instead, we check cache entries
352 * whose mtime are the same as the index file timestamp more
5f73076c 353 * carefully than others.
29e4d363 354 */
6d91da6d 355 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
356 if (assume_racy_is_modified)
357 changed |= DATA_CHANGED;
358 else
359 changed |= ce_modified_check_fs(ce, st);
360 }
b0391890 361
29e4d363 362 return changed;
b0391890
JH
363}
364
d1f128b0 365int ie_modified(const struct index_state *istate,
21a6b9fa
RS
366 const struct cache_entry *ce,
367 struct stat *st, unsigned int options)
b0391890 368{
29e4d363 369 int changed, changed_fs;
4bd5b7da
JH
370
371 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
372 if (!changed)
373 return 0;
b0391890
JH
374 /*
375 * If the mode or type has changed, there's no point in trying
376 * to refresh the entry - it's not going to match
377 */
378 if (changed & (MODE_CHANGED | TYPE_CHANGED))
379 return changed;
380
c70115b4
JH
381 /*
382 * Immediately after read-tree or update-index --cacheinfo,
383 * the length field is zero, as we have never even read the
384 * lstat(2) information once, and we cannot trust DATA_CHANGED
385 * returned by ie_match_stat() which in turn was returned by
386 * ce_match_stat_basic() to signal that the filesize of the
387 * blob changed. We have to actually go to the filesystem to
388 * see if the contents match, and if so, should answer "unchanged".
389 *
390 * The logic does not apply to gitlinks, as ce_match_stat_basic()
391 * already has checked the actual HEAD from the filesystem in the
392 * subproject. If ie_match_stat() already said it is different,
393 * then we know it is.
b0391890 394 */
c70115b4 395 if ((changed & DATA_CHANGED) &&
c21d39d7 396 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
b0391890
JH
397 return changed;
398
29e4d363
JH
399 changed_fs = ce_modified_check_fs(ce, st);
400 if (changed_fs)
401 return changed | changed_fs;
b0391890
JH
402 return 0;
403}
404
958ba6c9
LT
405int base_name_compare(const char *name1, int len1, int mode1,
406 const char *name2, int len2, int mode2)
407{
408 unsigned char c1, c2;
409 int len = len1 < len2 ? len1 : len2;
410 int cmp;
411
412 cmp = memcmp(name1, name2, len);
413 if (cmp)
414 return cmp;
415 c1 = name1[len];
416 c2 = name2[len];
1833a925 417 if (!c1 && S_ISDIR(mode1))
958ba6c9 418 c1 = '/';
1833a925 419 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
420 c2 = '/';
421 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
422}
423
0ab9e1e8
LT
424/*
425 * df_name_compare() is identical to base_name_compare(), except it
426 * compares conflicting directory/file entries as equal. Note that
427 * while a directory name compares as equal to a regular file, they
428 * then individually compare _differently_ to a filename that has
429 * a dot after the basename (because '\0' < '.' < '/').
430 *
431 * This is used by routines that want to traverse the git namespace
432 * but then handle conflicting entries together when possible.
433 */
434int df_name_compare(const char *name1, int len1, int mode1,
435 const char *name2, int len2, int mode2)
436{
437 int len = len1 < len2 ? len1 : len2, cmp;
438 unsigned char c1, c2;
439
440 cmp = memcmp(name1, name2, len);
441 if (cmp)
442 return cmp;
443 /* Directories and files compare equal (same length, same name) */
444 if (len1 == len2)
445 return 0;
446 c1 = name1[len];
447 if (!c1 && S_ISDIR(mode1))
448 c1 = '/';
449 c2 = name2[len];
450 if (!c2 && S_ISDIR(mode2))
451 c2 = '/';
452 if (c1 == '/' && !c2)
453 return 0;
454 if (c2 == '/' && !c1)
455 return 0;
456 return c1 - c2;
457}
458
ccdd4a0f 459int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
eb38c22f 460{
ccdd4a0f
JM
461 size_t min_len = (len1 < len2) ? len1 : len2;
462 int cmp = memcmp(name1, name2, min_len);
eb38c22f
LT
463 if (cmp)
464 return cmp;
465 if (len1 < len2)
466 return -1;
467 if (len1 > len2)
468 return 1;
ccdd4a0f
JM
469 return 0;
470}
471
472int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
473{
474 int cmp;
475
476 cmp = name_compare(name1, len1, name2, len2);
477 if (cmp)
478 return cmp;
5f73076c 479
b60e188c 480 if (stage1 < stage2)
95fd5bf8 481 return -1;
b60e188c 482 if (stage1 > stage2)
95fd5bf8 483 return 1;
eb38c22f
LT
484 return 0;
485}
486
357e9c69 487static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
eb38c22f
LT
488{
489 int first, last;
490
491 first = 0;
4aab5b46 492 last = istate->cache_nr;
eb38c22f
LT
493 while (last > first) {
494 int next = (last + first) >> 1;
4aab5b46 495 struct cache_entry *ce = istate->cache[next];
b60e188c 496 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
eb38c22f 497 if (!cmp)
76e7f4ec 498 return next;
eb38c22f
LT
499 if (cmp < 0) {
500 last = next;
501 continue;
502 }
503 first = next+1;
504 }
76e7f4ec 505 return -first-1;
eb38c22f
LT
506}
507
b60e188c
TG
508int index_name_pos(const struct index_state *istate, const char *name, int namelen)
509{
510 return index_name_stage_pos(istate, name, namelen, 0);
511}
512
7b937ca3 513/* Remove entry, return true if there are more entries to go.. */
4aab5b46 514int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 515{
cf558704
LT
516 struct cache_entry *ce = istate->cache[pos];
517
cfc5789a 518 record_resolve_undo(istate, ce);
2092678c 519 remove_name_hash(istate, ce);
045113a5 520 save_or_free_index_entry(istate, ce);
e636a7b4 521 istate->cache_changed |= CE_ENTRY_REMOVED;
4aab5b46
JH
522 istate->cache_nr--;
523 if (pos >= istate->cache_nr)
7b937ca3 524 return 0;
4aab5b46
JH
525 memmove(istate->cache + pos,
526 istate->cache + pos + 1,
527 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
7b937ca3
LT
528 return 1;
529}
530
36419c8e 531/*
98e023de 532 * Remove all cache entries marked for removal, that is where
36419c8e
KB
533 * CE_REMOVE is set in ce_flags. This is much more effective than
534 * calling remove_index_entry_at() for each entry to be removed.
535 */
536void remove_marked_cache_entries(struct index_state *istate)
537{
538 struct cache_entry **ce_array = istate->cache;
539 unsigned int i, j;
540
541 for (i = j = 0; i < istate->cache_nr; i++) {
5699d17e 542 if (ce_array[i]->ce_flags & CE_REMOVE) {
2092678c 543 remove_name_hash(istate, ce_array[i]);
045113a5 544 save_or_free_index_entry(istate, ce_array[i]);
5699d17e 545 }
36419c8e
KB
546 else
547 ce_array[j++] = ce_array[i];
548 }
ad837d9e
NTND
549 if (j == istate->cache_nr)
550 return;
e636a7b4 551 istate->cache_changed |= CE_ENTRY_REMOVED;
36419c8e
KB
552 istate->cache_nr = j;
553}
554
4aab5b46 555int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 556{
4aab5b46 557 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
558 if (pos < 0)
559 pos = -pos-1;
a5400efe 560 cache_tree_invalidate_path(istate, path);
e931371a 561 untracked_cache_remove_from_index(istate, path);
4aab5b46
JH
562 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
563 remove_index_entry_at(istate, pos);
197ee8c9
LT
564 return 0;
565}
566
20314271
JS
567static int compare_name(struct cache_entry *ce, const char *path, int namelen)
568{
569 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
570}
571
572static int index_name_pos_also_unmerged(struct index_state *istate,
573 const char *path, int namelen)
574{
575 int pos = index_name_pos(istate, path, namelen);
576 struct cache_entry *ce;
577
578 if (pos >= 0)
579 return pos;
580
581 /* maybe unmerged? */
582 pos = -1 - pos;
583 if (pos >= istate->cache_nr ||
584 compare_name((ce = istate->cache[pos]), path, namelen))
585 return -1;
586
587 /* order of preference: stage 2, 1, 3 */
588 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
589 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
590 !compare_name(ce, path, namelen))
591 pos++;
592 return pos;
593}
594
1102952b
LT
595static int different_name(struct cache_entry *ce, struct cache_entry *alias)
596{
597 int len = ce_namelen(ce);
598 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
599}
600
601/*
602 * If we add a filename that aliases in the cache, we will use the
603 * name that we already have - but we don't want to update the same
604 * alias twice, because that implies that there were actually two
605 * different files with aliasing names!
606 *
607 * So we use the CE_ADDED flag to verify that the alias was an old
608 * one before we accept it as
609 */
045113a5
NTND
610static struct cache_entry *create_alias_ce(struct index_state *istate,
611 struct cache_entry *ce,
612 struct cache_entry *alias)
1102952b
LT
613{
614 int len;
615 struct cache_entry *new;
616
617 if (alias->ce_flags & CE_ADDED)
618 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
619
620 /* Ok, create the new entry using the name of the existing alias */
621 len = ce_namelen(alias);
622 new = xcalloc(1, cache_entry_size(len));
623 memcpy(new->name, alias->name, len);
624 copy_cache_entry(new, ce);
045113a5 625 save_or_free_index_entry(istate, ce);
1102952b
LT
626 return new;
627}
628
b4b313f9 629void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
39425819
JH
630{
631 unsigned char sha1[20];
632 if (write_sha1_file("", 0, blob_type, sha1))
633 die("cannot create an empty blob in the object database");
99d1a986 634 hashcpy(ce->oid.hash, sha1);
39425819
JH
635}
636
610d55af 637int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 638{
38ed1d89 639 int size, namelen, was_same;
d177cab0 640 mode_t st_mode = st->st_mode;
6835550d 641 struct cache_entry *ce, *alias;
56cac48c 642 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
643 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
644 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
645 int intent_only = flags & ADD_CACHE_INTENT;
646 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
647 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
11be42a4 648
d177cab0 649 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
960b8ad1 650 return error("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
651
652 namelen = strlen(path);
d177cab0 653 if (S_ISDIR(st_mode)) {
09595258
LT
654 while (namelen && path[namelen-1] == '/')
655 namelen--;
656 }
11be42a4
JS
657 size = cache_entry_size(namelen);
658 ce = xcalloc(1, size);
659 memcpy(ce->name, path, namelen);
b60e188c 660 ce->ce_namelen = namelen;
39425819
JH
661 if (!intent_only)
662 fill_stat_cache_info(ce, st);
388b2acd
JH
663 else
664 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 665
610d55af
TG
666
667 if (trust_executable_bit && has_symlinks) {
d177cab0 668 ce->ce_mode = create_ce_mode(st_mode);
610d55af 669 } else {
78a8d641
JS
670 /* If there is an existing entry, pick the mode bits and type
671 * from it, otherwise assume unexecutable regular file.
11be42a4 672 */
185c975f 673 struct cache_entry *ent;
20314271 674 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 675
4aab5b46 676 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 677 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
678 }
679
dc1ae704
JJ
680 /* When core.ignorecase=true, determine if a directory of the same name but differing
681 * case already exists within the Git repository. If it does, ensure the directory
682 * case of the file being added to the repository matches (is folded into) the existing
683 * entry's directory case.
684 */
685 if (ignore_case) {
41284eb0 686 adjust_dirname_case(istate, ce->name);
dc1ae704
JJ
687 }
688
ebbd7439 689 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
d177cab0 690 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
0781b8a9 691 /* Nothing changed, really */
125fd984
JH
692 if (!S_ISGITLINK(alias->ce_mode))
693 ce_mark_uptodate(alias);
1102952b 694 alias->ce_flags |= CE_ADDED;
2d9426b0
SB
695
696 free(ce);
0781b8a9
JH
697 return 0;
698 }
39425819 699 if (!intent_only) {
99d1a986 700 if (index_path(ce->oid.hash, path, st, HASH_WRITE_OBJECT)) {
2d9426b0 701 free(ce);
39425819 702 return error("unable to index file %s", path);
2d9426b0 703 }
39425819 704 } else
b4b313f9 705 set_object_name_for_intent_to_add_entry(ce);
39425819 706
1102952b 707 if (ignore_case && alias && different_name(ce, alias))
045113a5 708 ce = create_alias_ce(istate, ce, alias);
1102952b 709 ce->ce_flags |= CE_ADDED;
38ed1d89 710
3bf0dd1f 711 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
712 was_same = (alias &&
713 !ce_stage(alias) &&
99d1a986 714 !oidcmp(&alias->oid, &ce->oid) &&
38ed1d89
JH
715 ce->ce_mode == alias->ce_mode);
716
717 if (pretend)
067178ed
JH
718 free(ce);
719 else if (add_index_entry(istate, ce, add_option)) {
720 free(ce);
721 return error("unable to add %s to index", path);
722 }
38ed1d89 723 if (verbose && !was_same)
11be42a4 724 printf("add '%s'\n", path);
11be42a4
JS
725 return 0;
726}
727
610d55af 728int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
729{
730 struct stat st;
731 if (lstat(path, &st))
d824cbba 732 die_errno("unable to stat '%s'", path);
610d55af 733 return add_to_index(istate, path, &st, flags);
d177cab0
LT
734}
735
6640f881
CR
736struct cache_entry *make_cache_entry(unsigned int mode,
737 const unsigned char *sha1, const char *path, int stage,
25762726 738 unsigned int refresh_options)
6640f881
CR
739{
740 int size, len;
bc1c2caa 741 struct cache_entry *ce, *ret;
6640f881 742
7e7abea9
DP
743 if (!verify_path(path)) {
744 error("Invalid path '%s'", path);
6640f881 745 return NULL;
7e7abea9 746 }
6640f881
CR
747
748 len = strlen(path);
749 size = cache_entry_size(len);
750 ce = xcalloc(1, size);
751
99d1a986 752 hashcpy(ce->oid.hash, sha1);
6640f881 753 memcpy(ce->name, path, len);
b60e188c
TG
754 ce->ce_flags = create_ce_flags(stage);
755 ce->ce_namelen = len;
6640f881
CR
756 ce->ce_mode = create_ce_mode(mode);
757
bc1c2caa 758 ret = refresh_cache_entry(ce, refresh_options);
915e44c6 759 if (ret != ce)
bc1c2caa 760 free(ce);
915e44c6 761 return ret;
6640f881
CR
762}
763
d9d70966
TG
764/*
765 * Chmod an index entry with either +x or -x.
766 *
767 * Returns -1 if the chmod for the particular cache entry failed (if it's
768 * not a regular file), -2 if an invalid flip argument is passed in, 0
769 * otherwise.
770 */
771int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
772 char flip)
773{
774 if (!S_ISREG(ce->ce_mode))
775 return -1;
776 switch (flip) {
777 case '+':
778 ce->ce_mode |= 0111;
779 break;
780 case '-':
781 ce->ce_mode &= ~0111;
782 break;
783 default:
784 return -2;
785 }
786 cache_tree_invalidate_path(istate, ce->name);
787 ce->ce_flags |= CE_UPDATE_IN_BASE;
788 istate->cache_changed |= CE_ENTRY_CHANGED;
789
790 return 0;
791}
792
9c5e6c80 793int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
7b937ca3
LT
794{
795 int len = ce_namelen(a);
796 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
797}
798
8dcf39c4
LT
799/*
800 * We fundamentally don't like some paths: we don't want
801 * dot or dot-dot anywhere, and for obvious reasons don't
802 * want to recurse into ".git" either.
803 *
804 * Also, we don't want double slashes or slashes at the
805 * end that can make pathnames ambiguous.
806 */
807static int verify_dotfile(const char *rest)
808{
809 /*
810 * The first character was '.', but that
811 * has already been discarded, we now test
812 * the rest.
813 */
e0f530ff 814
8dcf39c4 815 /* "." is not allowed */
e0f530ff 816 if (*rest == '\0' || is_dir_sep(*rest))
8dcf39c4
LT
817 return 0;
818
e0f530ff 819 switch (*rest) {
8dcf39c4
LT
820 /*
821 * ".git" followed by NUL or slash is bad. This
822 * shares the path end test with the ".." case.
823 */
824 case 'g':
cc2fc7c2
JK
825 case 'G':
826 if (rest[1] != 'i' && rest[1] != 'I')
8dcf39c4 827 break;
cc2fc7c2 828 if (rest[2] != 't' && rest[2] != 'T')
8dcf39c4
LT
829 break;
830 rest += 2;
831 /* fallthrough */
832 case '.':
e0f530ff 833 if (rest[1] == '\0' || is_dir_sep(rest[1]))
8dcf39c4
LT
834 return 0;
835 }
836 return 1;
837}
838
839int verify_path(const char *path)
840{
841 char c;
842
56948cb6
EFL
843 if (has_dos_drive_prefix(path))
844 return 0;
845
8dcf39c4
LT
846 goto inside;
847 for (;;) {
848 if (!c)
849 return 1;
56948cb6 850 if (is_dir_sep(c)) {
8dcf39c4 851inside:
a42643aa
JK
852 if (protect_hfs && is_hfs_dotgit(path))
853 return 0;
2b4c6efc
JS
854 if (protect_ntfs && is_ntfs_dotgit(path))
855 return 0;
8dcf39c4 856 c = *path++;
3bdf09c7
JH
857 if ((c == '.' && !verify_dotfile(path)) ||
858 is_dir_sep(c) || c == '\0')
859 return 0;
8dcf39c4
LT
860 }
861 c = *path++;
862 }
863}
864
12676608
LT
865/*
866 * Do we have another file that has the beginning components being a
867 * proper superset of the name we're trying to add?
0f1e4f04 868 */
4aab5b46
JH
869static int has_file_name(struct index_state *istate,
870 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 871{
12676608
LT
872 int retval = 0;
873 int len = ce_namelen(ce);
b155725d 874 int stage = ce_stage(ce);
12676608 875 const char *name = ce->name;
0f1e4f04 876
4aab5b46
JH
877 while (pos < istate->cache_nr) {
878 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 879
12676608 880 if (len >= ce_namelen(p))
0f1e4f04 881 break;
12676608
LT
882 if (memcmp(name, p->name, len))
883 break;
b155725d
JH
884 if (ce_stage(p) != stage)
885 continue;
12676608
LT
886 if (p->name[len] != '/')
887 continue;
7a51ed66 888 if (p->ce_flags & CE_REMOVE)
21cd8d00 889 continue;
12676608
LT
890 retval = -1;
891 if (!ok_to_replace)
892 break;
4aab5b46 893 remove_index_entry_at(istate, --pos);
0f1e4f04 894 }
12676608
LT
895 return retval;
896}
0f1e4f04 897
12676608
LT
898/*
899 * Do we have another file with a pathname that is a proper
900 * subset of the name we're trying to add?
901 */
4aab5b46
JH
902static int has_dir_name(struct index_state *istate,
903 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
904{
905 int retval = 0;
b155725d 906 int stage = ce_stage(ce);
12676608
LT
907 const char *name = ce->name;
908 const char *slash = name + ce_namelen(ce);
0f1e4f04 909
12676608
LT
910 for (;;) {
911 int len;
0f1e4f04 912
12676608
LT
913 for (;;) {
914 if (*--slash == '/')
915 break;
916 if (slash <= ce->name)
917 return retval;
918 }
919 len = slash - name;
0f1e4f04 920
b60e188c 921 pos = index_name_stage_pos(istate, name, len, stage);
12676608 922 if (pos >= 0) {
21cd8d00
JH
923 /*
924 * Found one, but not so fast. This could
925 * be a marker that says "I was here, but
926 * I am being removed". Such an entry is
927 * not a part of the resulting tree, and
928 * it is Ok to have a directory at the same
929 * path.
930 */
077c48df 931 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
932 retval = -1;
933 if (!ok_to_replace)
934 break;
4aab5b46 935 remove_index_entry_at(istate, pos);
21cd8d00
JH
936 continue;
937 }
12676608 938 }
21cd8d00
JH
939 else
940 pos = -pos-1;
12676608
LT
941
942 /*
943 * Trivial optimization: if we find an entry that
944 * already matches the sub-directory, then we know
b155725d 945 * we're ok, and we can exit.
12676608 946 */
4aab5b46
JH
947 while (pos < istate->cache_nr) {
948 struct cache_entry *p = istate->cache[pos];
b155725d
JH
949 if ((ce_namelen(p) <= len) ||
950 (p->name[len] != '/') ||
951 memcmp(p->name, name, len))
952 break; /* not our subdirectory */
077c48df
JH
953 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
954 /*
955 * p is at the same stage as our entry, and
b155725d
JH
956 * is a subdirectory of what we are looking
957 * at, so we cannot have conflicts at our
958 * level or anything shorter.
959 */
960 return retval;
961 pos++;
192268c1 962 }
0f1e4f04 963 }
12676608
LT
964 return retval;
965}
966
967/* We may be in a situation where we already have path/file and path
968 * is being added, or we already have path and path/file is being
969 * added. Either one would result in a nonsense tree that has path
970 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 971 *
12676608
LT
972 * If ok-to-replace is specified, we remove the conflicting entries
973 * from the cache so the caller should recompute the insert position.
974 * When this happens, we return non-zero.
975 */
4aab5b46
JH
976static int check_file_directory_conflict(struct index_state *istate,
977 const struct cache_entry *ce,
978 int pos, int ok_to_replace)
12676608 979{
21cd8d00
JH
980 int retval;
981
982 /*
983 * When ce is an "I am going away" entry, we allow it to be added
984 */
7a51ed66 985 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
986 return 0;
987
12676608
LT
988 /*
989 * We check if the path is a sub-path of a subsequent pathname
990 * first, since removing those will not change the position
21cd8d00 991 * in the array.
12676608 992 */
4aab5b46 993 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 994
12676608
LT
995 /*
996 * Then check if the path might have a clashing sub-directory
997 * before it.
998 */
4aab5b46 999 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
1000}
1001
af3785dc 1002static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
1003{
1004 int pos;
192268c1
JH
1005 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1006 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 1007 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 1008 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 1009
ce7c614b
NTND
1010 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1011 cache_tree_invalidate_path(istate, ce->name);
b60e188c 1012 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
197ee8c9 1013
3e09cdfd 1014 /* existing match? Just replace it. */
76e7f4ec 1015 if (pos >= 0) {
39425819
JH
1016 if (!new_only)
1017 replace_index_entry(istate, pos, ce);
197ee8c9
LT
1018 return 0;
1019 }
76e7f4ec 1020 pos = -pos-1;
197ee8c9 1021
ffcc9ba7
NTND
1022 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1023 untracked_cache_add_to_index(istate, ce->name);
e931371a 1024
7b937ca3
LT
1025 /*
1026 * Inserting a merged entry ("stage 0") into the index
1027 * will always replace all non-merged entries..
1028 */
4aab5b46
JH
1029 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1030 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 1031 ok_to_add = 1;
4aab5b46 1032 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
1033 break;
1034 }
1035 }
1036
121481ab
LT
1037 if (!ok_to_add)
1038 return -1;
8dcf39c4 1039 if (!verify_path(ce->name))
7e7abea9 1040 return error("Invalid path '%s'", ce->name);
121481ab 1041
3e09cdfd 1042 if (!skip_df_check &&
4aab5b46 1043 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 1044 if (!ok_to_replace)
4aab5b46
JH
1045 return error("'%s' appears as both a file and as a directory",
1046 ce->name);
b60e188c 1047 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
192268c1
JH
1048 pos = -pos-1;
1049 }
af3785dc
JH
1050 return pos + 1;
1051}
1052
1053int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1054{
1055 int pos;
1056
1057 if (option & ADD_CACHE_JUST_APPEND)
1058 pos = istate->cache_nr;
1059 else {
1060 int ret;
1061 ret = add_index_entry_with_check(istate, ce, option);
1062 if (ret <= 0)
1063 return ret;
1064 pos = ret - 1;
1065 }
0f1e4f04 1066
197ee8c9 1067 /* Make sure the array is big enough .. */
999f5660 1068 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
197ee8c9
LT
1069
1070 /* Add it in.. */
4aab5b46 1071 istate->cache_nr++;
af3785dc 1072 if (istate->cache_nr > pos + 1)
4aab5b46
JH
1073 memmove(istate->cache + pos + 1,
1074 istate->cache + pos,
1075 (istate->cache_nr - pos - 1) * sizeof(ce));
cf558704 1076 set_index_entry(istate, pos, ce);
e636a7b4 1077 istate->cache_changed |= CE_ENTRY_ADDED;
197ee8c9
LT
1078 return 0;
1079}
1080
405e5b2f
LT
1081/*
1082 * "refresh" does not calculate a new sha1 file or bring the
1083 * cache up-to-date for mode/content changes. But what it
1084 * _does_ do is to "re-match" the stat information of a file
1085 * with the cache, so that you can refresh the cache for a
1086 * file that hasn't been changed but where the stat entry is
1087 * out of date.
1088 *
1089 * For example, you'd want to do this after doing a "git-read-tree",
1090 * to link up the stat cache details with the proper files.
1091 */
4aab5b46 1092static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da 1093 struct cache_entry *ce,
d05e6970
JK
1094 unsigned int options, int *err,
1095 int *changed_ret)
405e5b2f
LT
1096{
1097 struct stat st;
1098 struct cache_entry *updated;
1099 int changed, size;
25762726 1100 int refresh = options & CE_MATCH_REFRESH;
4bd5b7da 1101 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1102 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
2e2e7ec1 1103 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
405e5b2f 1104
25762726 1105 if (!refresh || ce_uptodate(ce))
eadb5831
JH
1106 return ce;
1107
aa9349d4 1108 /*
56cac48c
NTND
1109 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1110 * that the change to the work tree does not matter and told
1111 * us not to worry.
aa9349d4 1112 */
56cac48c
NTND
1113 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1114 ce_mark_uptodate(ce);
1115 return ce;
1116 }
aa9349d4
MSO
1117 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1118 ce_mark_uptodate(ce);
1119 return ce;
1120 }
1121
ccad42d4
RS
1122 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1123 if (ignore_missing)
1124 return ce;
1125 if (err)
1126 *err = ENOENT;
1127 return NULL;
1128 }
1129
8fd2cb40 1130 if (lstat(ce->name, &st) < 0) {
2e2e7ec1
BK
1131 if (ignore_missing && errno == ENOENT)
1132 return ce;
ec0cc704
JH
1133 if (err)
1134 *err = errno;
8fd2cb40
JS
1135 return NULL;
1136 }
405e5b2f 1137
4bd5b7da 1138 changed = ie_match_stat(istate, ce, &st, options);
d05e6970
JK
1139 if (changed_ret)
1140 *changed_ret = changed;
405e5b2f 1141 if (!changed) {
4bd5b7da
JH
1142 /*
1143 * The path is unchanged. If we were told to ignore
1144 * valid bit, then we did the actual stat check and
1145 * found that the entry is unmodified. If the entry
1146 * is not marked VALID, this is the place to mark it
1147 * valid again, under "assume unchanged" mode.
1148 */
1149 if (ignore_valid && assume_unchanged &&
7a51ed66 1150 !(ce->ce_flags & CE_VALID))
405e5b2f 1151 ; /* mark this one VALID again */
eadb5831
JH
1152 else {
1153 /*
1154 * We do not mark the index itself "modified"
1155 * because CE_UPTODATE flag is in-core only;
1156 * we are not going to write this change out.
1157 */
125fd984
JH
1158 if (!S_ISGITLINK(ce->ce_mode))
1159 ce_mark_uptodate(ce);
8fd2cb40 1160 return ce;
eadb5831 1161 }
405e5b2f
LT
1162 }
1163
4bd5b7da 1164 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1165 if (err)
1166 *err = EINVAL;
8fd2cb40
JS
1167 return NULL;
1168 }
405e5b2f
LT
1169
1170 size = ce_size(ce);
1171 updated = xmalloc(size);
1172 memcpy(updated, ce, size);
1173 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1174 /*
1175 * If ignore_valid is not set, we should leave CE_VALID bit
1176 * alone. Otherwise, paths marked with --no-assume-unchanged
1177 * (i.e. things to be edited) will reacquire CE_VALID bit
1178 * automatically, which is not really what we want.
405e5b2f 1179 */
4bd5b7da 1180 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1181 !(ce->ce_flags & CE_VALID))
1182 updated->ce_flags &= ~CE_VALID;
405e5b2f 1183
e636a7b4 1184 /* istate->cache_changed is updated in the caller */
405e5b2f
LT
1185 return updated;
1186}
1187
3deffc52 1188static void show_file(const char * fmt, const char * name, int in_porcelain,
046613c5 1189 int * first, const char *header_msg)
3deffc52
MM
1190{
1191 if (in_porcelain && *first && header_msg) {
1192 printf("%s\n", header_msg);
cd2b8ae9 1193 *first = 0;
3deffc52
MM
1194 }
1195 printf(fmt, name);
1196}
1197
9b2d6149
NTND
1198int refresh_index(struct index_state *istate, unsigned int flags,
1199 const struct pathspec *pathspec,
046613c5 1200 char *seen, const char *header_msg)
405e5b2f
LT
1201{
1202 int i;
1203 int has_errors = 0;
1204 int really = (flags & REFRESH_REALLY) != 0;
1205 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1206 int quiet = (flags & REFRESH_QUIET) != 0;
1207 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1208 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1209 int first = 1;
1210 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
25762726
BK
1211 unsigned int options = (CE_MATCH_REFRESH |
1212 (really ? CE_MATCH_IGNORE_VALID : 0) |
2e2e7ec1 1213 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
4bd4e730 1214 const char *modified_fmt;
73b7eae6
JK
1215 const char *deleted_fmt;
1216 const char *typechange_fmt;
1217 const char *added_fmt;
4bd4e730 1218 const char *unmerged_fmt;
405e5b2f 1219
4bd4e730 1220 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
73b7eae6
JK
1221 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1222 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1223 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
4bd4e730 1224 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
4aab5b46 1225 for (i = 0; i < istate->cache_nr; i++) {
405e5b2f 1226 struct cache_entry *ce, *new;
ec0cc704 1227 int cache_errno = 0;
73b7eae6 1228 int changed = 0;
3d1f148c 1229 int filtered = 0;
ec0cc704 1230
4aab5b46 1231 ce = istate->cache[i];
5fdeacb0
JS
1232 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1233 continue;
1234
429bb40a 1235 if (pathspec && !ce_path_match(ce, pathspec, seen))
3d1f148c
JH
1236 filtered = 1;
1237
405e5b2f 1238 if (ce_stage(ce)) {
4aab5b46
JH
1239 while ((i < istate->cache_nr) &&
1240 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1241 i++;
1242 i--;
1243 if (allow_unmerged)
1244 continue;
3d1f148c
JH
1245 if (!filtered)
1246 show_file(unmerged_fmt, ce->name, in_porcelain,
1247 &first, header_msg);
405e5b2f
LT
1248 has_errors = 1;
1249 continue;
1250 }
1251
3d1f148c 1252 if (filtered)
d616813d
AJ
1253 continue;
1254
73b7eae6 1255 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
8fd2cb40 1256 if (new == ce)
405e5b2f 1257 continue;
8fd2cb40 1258 if (!new) {
73b7eae6
JK
1259 const char *fmt;
1260
8fd2cb40 1261 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1262 /* If we are doing --really-refresh that
1263 * means the index is not valid anymore.
1264 */
7a51ed66 1265 ce->ce_flags &= ~CE_VALID;
078a58e8 1266 ce->ce_flags |= CE_UPDATE_IN_BASE;
e636a7b4 1267 istate->cache_changed |= CE_ENTRY_CHANGED;
405e5b2f
LT
1268 }
1269 if (quiet)
1270 continue;
73b7eae6
JK
1271
1272 if (cache_errno == ENOENT)
1273 fmt = deleted_fmt;
895ff3b2 1274 else if (ce_intent_to_add(ce))
73b7eae6
JK
1275 fmt = added_fmt; /* must be before other checks */
1276 else if (changed & TYPE_CHANGED)
1277 fmt = typechange_fmt;
1278 else
1279 fmt = modified_fmt;
1280 show_file(fmt,
1281 ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1282 has_errors = 1;
1283 continue;
1284 }
cf558704
LT
1285
1286 replace_index_entry(istate, i, new);
405e5b2f
LT
1287 }
1288 return has_errors;
1289}
1290
1335d76e 1291struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
25762726 1292 unsigned int options)
ec0cc704 1293{
25762726 1294 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
ec0cc704
JH
1295}
1296
db3b313c
JH
1297
1298/*****************************************************************
1299 * Index File I/O
1300 *****************************************************************/
1301
9d227781
JH
1302#define INDEX_FORMAT_DEFAULT 3
1303
136347d7
TG
1304static unsigned int get_index_format_default(void)
1305{
1306 char *envversion = getenv("GIT_INDEX_VERSION");
3c09d684 1307 char *endp;
b27a5720 1308 int value;
3c09d684
TG
1309 unsigned int version = INDEX_FORMAT_DEFAULT;
1310
136347d7 1311 if (!envversion) {
b27a5720
TA
1312 if (!git_config_get_int("index.version", &value))
1313 version = value;
3c09d684
TG
1314 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1315 warning(_("index.version set, but the value is invalid.\n"
136347d7 1316 "Using version %i"), INDEX_FORMAT_DEFAULT);
3c09d684 1317 return INDEX_FORMAT_DEFAULT;
136347d7
TG
1318 }
1319 return version;
1320 }
3c09d684
TG
1321
1322 version = strtoul(envversion, &endp, 10);
1323 if (*endp ||
1324 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1325 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1326 "Using version %i"), INDEX_FORMAT_DEFAULT);
1327 version = INDEX_FORMAT_DEFAULT;
1328 }
1329 return version;
136347d7
TG
1330}
1331
db3b313c
JH
1332/*
1333 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1334 * Again - this is just a (very strong in practice) heuristic that
1335 * the inode hasn't changed.
1336 *
1337 * We save the fields in big-endian order to allow using the
1338 * index file over NFS transparently.
1339 */
1340struct ondisk_cache_entry {
1341 struct cache_time ctime;
1342 struct cache_time mtime;
7800c1eb
TG
1343 uint32_t dev;
1344 uint32_t ino;
1345 uint32_t mode;
1346 uint32_t uid;
1347 uint32_t gid;
1348 uint32_t size;
db3b313c 1349 unsigned char sha1[20];
7800c1eb 1350 uint16_t flags;
db3b313c
JH
1351 char name[FLEX_ARRAY]; /* more */
1352};
1353
1354/*
1355 * This struct is used when CE_EXTENDED bit is 1
1356 * The struct must match ondisk_cache_entry exactly from
1357 * ctime till flags
1358 */
1359struct ondisk_cache_entry_extended {
1360 struct cache_time ctime;
1361 struct cache_time mtime;
7800c1eb
TG
1362 uint32_t dev;
1363 uint32_t ino;
1364 uint32_t mode;
1365 uint32_t uid;
1366 uint32_t gid;
1367 uint32_t size;
db3b313c 1368 unsigned char sha1[20];
7800c1eb
TG
1369 uint16_t flags;
1370 uint16_t flags2;
db3b313c
JH
1371 char name[FLEX_ARRAY]; /* more */
1372};
1373
6c9cd161 1374/* These are only used for v3 or lower */
db3b313c
JH
1375#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1376#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1377#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1378#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1379 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1380 ondisk_cache_entry_size(ce_namelen(ce)))
1381
bad68ec9 1382static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163 1383{
9126f009 1384 git_SHA_CTX c;
bad68ec9 1385 unsigned char sha1[20];
0136bac9 1386 int hdr_version;
e83c5163 1387
ccc4feb5 1388 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1389 return error("bad signature");
0136bac9 1390 hdr_version = ntohl(hdr->hdr_version);
b82a7b5b 1391 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
0136bac9 1392 return error("bad index version %d", hdr_version);
9126f009
NP
1393 git_SHA1_Init(&c);
1394 git_SHA1_Update(&c, hdr, size - 20);
1395 git_SHA1_Final(sha1, &c);
a89fccd2 1396 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
ca9be054 1397 return error("bad index file sha1 signature");
e83c5163
LT
1398 return 0;
1399}
1400
4aab5b46
JH
1401static int read_index_extension(struct index_state *istate,
1402 const char *ext, void *data, unsigned long sz)
bad68ec9
JH
1403{
1404 switch (CACHE_EXT(ext)) {
1405 case CACHE_EXT_TREE:
4aab5b46 1406 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1407 break;
cfc5789a
JH
1408 case CACHE_EXT_RESOLVE_UNDO:
1409 istate->resolve_undo = resolve_undo_read(data, sz);
1410 break;
5fc2fc8f
NTND
1411 case CACHE_EXT_LINK:
1412 if (read_link_extension(istate, data, sz))
1413 return -1;
1414 break;
f9e6c649
NTND
1415 case CACHE_EXT_UNTRACKED:
1416 istate->untracked = read_untracked_extension(data, sz);
1417 break;
bad68ec9
JH
1418 default:
1419 if (*ext < 'A' || 'Z' < *ext)
1420 return error("index uses %.4s extension, which we do not understand",
1421 ext);
1422 fprintf(stderr, "ignoring %.4s extension\n", ext);
1423 break;
1424 }
1425 return 0;
1426}
1427
b3e83cc7 1428int hold_locked_index(struct lock_file *lk, int lock_flags)
216aab1e 1429{
b3e83cc7 1430 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
216aab1e
MH
1431}
1432
4aab5b46 1433int read_index(struct index_state *istate)
8fd2cb40 1434{
4aab5b46 1435 return read_index_from(istate, get_index_file());
8fd2cb40
JS
1436}
1437
3fc22b53
JH
1438static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1439 unsigned int flags,
1440 const char *name,
1441 size_t len)
1442{
1443 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1444
c3d8da57
JK
1445 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1446 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1447 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1448 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1449 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1450 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1451 ce->ce_mode = get_be32(&ondisk->mode);
1452 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1453 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1454 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
b60e188c
TG
1455 ce->ce_flags = flags & ~CE_NAMEMASK;
1456 ce->ce_namelen = len;
5fc2fc8f 1457 ce->index = 0;
99d1a986 1458 hashcpy(ce->oid.hash, ondisk->sha1);
3fc22b53
JH
1459 memcpy(ce->name, name, len);
1460 ce->name[len] = '\0';
1461 return ce;
1462}
1463
6c9cd161
JH
1464/*
1465 * Adjacent cache entries tend to share the leading paths, so it makes
1466 * sense to only store the differences in later entries. In the v4
1467 * on-disk format of the index, each on-disk cache entry stores the
1468 * number of bytes to be stripped from the end of the previous name,
1469 * and the bytes to append to the result, to come up with its name.
1470 */
1471static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1472{
1473 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1474 size_t len = decode_varint(&cp);
1475
1476 if (name->len < len)
1477 die("malformed name field in the index");
1478 strbuf_remove(name, name->len - len, len);
1479 for (ep = cp; *ep; ep++)
1480 ; /* find the end */
1481 strbuf_add(name, cp, ep - cp);
1482 return (const char *)ep + 1 - cp_;
1483}
1484
936f53d0 1485static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
6c9cd161
JH
1486 unsigned long *ent_size,
1487 struct strbuf *previous_name)
7a51ed66 1488{
debed2a6 1489 struct cache_entry *ce;
7fec10b7 1490 size_t len;
06aaaa0b 1491 const char *name;
debed2a6 1492 unsigned int flags;
7fec10b7 1493
7a51ed66 1494 /* On-disk flags are just 16 bits */
c3d8da57 1495 flags = get_be16(&ondisk->flags);
debed2a6 1496 len = flags & CE_NAMEMASK;
7fec10b7 1497
debed2a6 1498 if (flags & CE_EXTENDED) {
06aaaa0b
NTND
1499 struct ondisk_cache_entry_extended *ondisk2;
1500 int extended_flags;
1501 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
c3d8da57 1502 extended_flags = get_be16(&ondisk2->flags2) << 16;
06aaaa0b
NTND
1503 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1504 if (extended_flags & ~CE_EXTENDED_FLAGS)
1505 die("Unknown index entry format %08x", extended_flags);
debed2a6 1506 flags |= extended_flags;
06aaaa0b
NTND
1507 name = ondisk2->name;
1508 }
1509 else
1510 name = ondisk->name;
1511
6c9cd161
JH
1512 if (!previous_name) {
1513 /* v3 and earlier */
1514 if (len == CE_NAMEMASK)
1515 len = strlen(name);
1516 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1517
1518 *ent_size = ondisk_ce_size(ce);
1519 } else {
1520 unsigned long consumed;
1521 consumed = expand_name_field(previous_name, name);
1522 ce = cache_entry_from_ondisk(ondisk, flags,
1523 previous_name->buf,
1524 previous_name->len);
1525
1526 *ent_size = (name - ((char *)ondisk)) + consumed;
1527 }
debed2a6 1528 return ce;
cf558704
LT
1529}
1530
03f15a79 1531static void check_ce_order(struct index_state *istate)
15999d0b 1532{
03f15a79
TG
1533 unsigned int i;
1534
1535 for (i = 1; i < istate->cache_nr; i++) {
1536 struct cache_entry *ce = istate->cache[i - 1];
1537 struct cache_entry *next_ce = istate->cache[i];
1538 int name_compare = strcmp(ce->name, next_ce->name);
1539
1540 if (0 < name_compare)
1541 die("unordered stage entries in index");
1542 if (!name_compare) {
1543 if (!ce_stage(ce))
1544 die("multiple stage entries for merged file '%s'",
1545 ce->name);
1546 if (ce_stage(ce) > ce_stage(next_ce))
1547 die("unordered stage entries for '%s'",
1548 ce->name);
1549 }
15999d0b
JSP
1550 }
1551}
1552
435ec090
CC
1553static void tweak_untracked_cache(struct index_state *istate)
1554{
1555 switch (git_config_get_untracked_cache()) {
1556 case -1: /* keep: do nothing */
1557 break;
1558 case 0: /* false */
1559 remove_untracked_cache(istate);
1560 break;
1561 case 1: /* true */
1562 add_untracked_cache(istate);
1563 break;
1564 default: /* unknown value: do nothing */
1565 break;
1566 }
1567}
1568
43925312
CC
1569static void tweak_split_index(struct index_state *istate)
1570{
1571 switch (git_config_get_split_index()) {
1572 case -1: /* unset: do nothing */
1573 break;
1574 case 0: /* false */
1575 remove_split_index(istate);
1576 break;
1577 case 1: /* true */
1578 add_split_index(istate);
1579 break;
1580 default: /* unknown value: do nothing */
1581 break;
1582 }
1583}
1584
435ec090
CC
1585static void post_read_index_from(struct index_state *istate)
1586{
1587 check_ce_order(istate);
1588 tweak_untracked_cache(istate);
43925312 1589 tweak_split_index(istate);
435ec090
CC
1590}
1591
8fd2cb40 1592/* remember to discard_cache() before reading a different cache! */
3e52f70b 1593int do_read_index(struct index_state *istate, const char *path, int must_exist)
e83c5163
LT
1594{
1595 int fd, i;
1596 struct stat st;
debed2a6 1597 unsigned long src_offset;
e83c5163 1598 struct cache_header *hdr;
7a51ed66
LT
1599 void *mmap;
1600 size_t mmap_size;
6c9cd161 1601 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
e83c5163 1602
913e0e99 1603 if (istate->initialized)
4aab5b46 1604 return istate->cache_nr;
5d1a5c02 1605
fba2f38a
KB
1606 istate->timestamp.sec = 0;
1607 istate->timestamp.nsec = 0;
8fd2cb40 1608 fd = open(path, O_RDONLY);
5d1a5c02 1609 if (fd < 0) {
5fc2fc8f 1610 if (!must_exist && errno == ENOENT)
5d1a5c02 1611 return 0;
5fc2fc8f 1612 die_errno("%s: index file open failed", path);
5d1a5c02 1613 }
e83c5163 1614
3511a377 1615 if (fstat(fd, &st))
d824cbba 1616 die_errno("cannot stat the open index");
3511a377 1617
7a51ed66
LT
1618 mmap_size = xsize_t(st.st_size);
1619 if (mmap_size < sizeof(struct cache_header) + 20)
3511a377
LFC
1620 die("index file smaller than expected");
1621
a1293ef7 1622 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
7a51ed66 1623 if (mmap == MAP_FAILED)
0721c314 1624 die_errno("unable to map index file");
57d84f8d 1625 close(fd);
e83c5163 1626
7a51ed66
LT
1627 hdr = mmap;
1628 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1629 goto unmap;
1630
e93021b2 1631 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
9d227781 1632 istate->version = ntohl(hdr->hdr_version);
4aab5b46
JH
1633 istate->cache_nr = ntohl(hdr->hdr_entries);
1634 istate->cache_alloc = alloc_nr(istate->cache_nr);
c4aa3167 1635 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
913e0e99 1636 istate->initialized = 1;
7a51ed66 1637
9d227781 1638 if (istate->version == 4)
6c9cd161
JH
1639 previous_name = &previous_name_buf;
1640 else
1641 previous_name = NULL;
1642
7a51ed66 1643 src_offset = sizeof(*hdr);
4aab5b46 1644 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 1645 struct ondisk_cache_entry *disk_ce;
4aab5b46 1646 struct cache_entry *ce;
936f53d0 1647 unsigned long consumed;
4aab5b46 1648
7a51ed66 1649 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
6c9cd161 1650 ce = create_from_disk(disk_ce, &consumed, previous_name);
cf558704 1651 set_index_entry(istate, i, ce);
7a51ed66 1652
936f53d0 1653 src_offset += consumed;
e83c5163 1654 }
6c9cd161 1655 strbuf_release(&previous_name_buf);
fba2f38a 1656 istate->timestamp.sec = st.st_mtime;
c06ff490 1657 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 1658
7a51ed66 1659 while (src_offset <= mmap_size - 20 - 8) {
bad68ec9
JH
1660 /* After an array of active_nr index entries,
1661 * there can be arbitrary number of extended
1662 * sections, each of which is prefixed with
1663 * extension name (4-byte) and section length
1664 * in 4-byte network byte order.
1665 */
07cc8eca 1666 uint32_t extsize;
7a51ed66 1667 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
bad68ec9 1668 extsize = ntohl(extsize);
4aab5b46 1669 if (read_index_extension(istate,
7a51ed66
LT
1670 (const char *) mmap + src_offset,
1671 (char *) mmap + src_offset + 8,
1d7f171c 1672 extsize) < 0)
bad68ec9 1673 goto unmap;
7a51ed66
LT
1674 src_offset += 8;
1675 src_offset += extsize;
bad68ec9 1676 }
7a51ed66 1677 munmap(mmap, mmap_size);
4aab5b46 1678 return istate->cache_nr;
e83c5163
LT
1679
1680unmap:
7a51ed66 1681 munmap(mmap, mmap_size);
5d1a5c02 1682 die("index file corrupt");
e83c5163
LT
1683}
1684
0d59ffb4
CC
1685/*
1686 * Signal that the shared index is used by updating its mtime.
1687 *
1688 * This way, shared index can be removed if they have not been used
1689 * for some time.
1690 */
1691static void freshen_shared_index(char *base_sha1_hex, int warn)
1692{
1693 const char *shared_index = git_path("sharedindex.%s", base_sha1_hex);
1694 if (!check_and_freshen_file(shared_index, 1) && warn)
1695 warning("could not freshen shared index '%s'", shared_index);
1696}
1697
5fc2fc8f
NTND
1698int read_index_from(struct index_state *istate, const char *path)
1699{
1700 struct split_index *split_index;
1701 int ret;
1702
1703 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1704 if (istate->initialized)
1705 return istate->cache_nr;
1706
1707 ret = do_read_index(istate, path, 0);
435ec090 1708
5fc2fc8f 1709 split_index = istate->split_index;
03f15a79 1710 if (!split_index || is_null_sha1(split_index->base_sha1)) {
435ec090 1711 post_read_index_from(istate);
5fc2fc8f 1712 return ret;
03f15a79 1713 }
5fc2fc8f
NTND
1714
1715 if (split_index->base)
1716 discard_index(split_index->base);
1717 else
1718 split_index->base = xcalloc(1, sizeof(*split_index->base));
1719 ret = do_read_index(split_index->base,
1720 git_path("sharedindex.%s",
1721 sha1_to_hex(split_index->base_sha1)), 1);
1722 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1723 die("broken index, expect %s in %s, got %s",
1724 sha1_to_hex(split_index->base_sha1),
1725 git_path("sharedindex.%s",
6bea53c1 1726 sha1_to_hex(split_index->base_sha1)),
5fc2fc8f
NTND
1727 sha1_to_hex(split_index->base->sha1));
1728 merge_base_index(istate);
435ec090 1729 post_read_index_from(istate);
5fc2fc8f
NTND
1730 return ret;
1731}
1732
fa7b3c2f
JH
1733int is_index_unborn(struct index_state *istate)
1734{
debed2a6 1735 return (!istate->cache_nr && !istate->timestamp.sec);
fa7b3c2f
JH
1736}
1737
4aab5b46 1738int discard_index(struct index_state *istate)
6d297f81 1739{
debed2a6
RS
1740 int i;
1741
5fc2fc8f
NTND
1742 for (i = 0; i < istate->cache_nr; i++) {
1743 if (istate->cache[i]->index &&
1744 istate->split_index &&
1745 istate->split_index->base &&
1746 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1747 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1748 continue;
debed2a6 1749 free(istate->cache[i]);
5fc2fc8f 1750 }
cfc5789a 1751 resolve_undo_clear_index(istate);
4aab5b46
JH
1752 istate->cache_nr = 0;
1753 istate->cache_changed = 0;
fba2f38a
KB
1754 istate->timestamp.sec = 0;
1755 istate->timestamp.nsec = 0;
2092678c 1756 free_name_hash(istate);
4aab5b46 1757 cache_tree_free(&(istate->cache_tree));
913e0e99 1758 istate->initialized = 0;
a0fc4db0
RS
1759 free(istate->cache);
1760 istate->cache = NULL;
1761 istate->cache_alloc = 0;
5fc2fc8f 1762 discard_split_index(istate);
f9e6c649
NTND
1763 free_untracked_cache(istate->untracked);
1764 istate->untracked = NULL;
7a51ed66 1765 return 0;
6d297f81
JS
1766}
1767
d1f128b0 1768int unmerged_index(const struct index_state *istate)
94a5728c
DB
1769{
1770 int i;
1771 for (i = 0; i < istate->cache_nr; i++) {
1772 if (ce_stage(istate->cache[i]))
1773 return 1;
1774 }
1775 return 0;
1776}
1777
4990aadc 1778#define WRITE_BUFFER_SIZE 8192
bf0f910d 1779static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
1780static unsigned long write_buffer_len;
1781
9126f009 1782static int ce_write_flush(git_SHA_CTX *context, int fd)
6015c28b
JH
1783{
1784 unsigned int buffered = write_buffer_len;
1785 if (buffered) {
9126f009 1786 git_SHA1_Update(context, write_buffer, buffered);
93822c22 1787 if (write_in_full(fd, write_buffer, buffered) != buffered)
6015c28b
JH
1788 return -1;
1789 write_buffer_len = 0;
1790 }
1791 return 0;
1792}
1793
9126f009 1794static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
1795{
1796 while (len) {
1797 unsigned int buffered = write_buffer_len;
1798 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1799 if (partial > len)
1800 partial = len;
1801 memcpy(write_buffer + buffered, data, partial);
1802 buffered += partial;
1803 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
1804 write_buffer_len = buffered;
1805 if (ce_write_flush(context, fd))
4990aadc
LT
1806 return -1;
1807 buffered = 0;
1808 }
1809 write_buffer_len = buffered;
1810 len -= partial;
1d7f171c 1811 data = (char *) data + partial;
a6080a0a
JH
1812 }
1813 return 0;
4990aadc
LT
1814}
1815
9126f009 1816static int write_index_ext_header(git_SHA_CTX *context, int fd,
ac58c7b1 1817 unsigned int ext, unsigned int sz)
bad68ec9
JH
1818{
1819 ext = htonl(ext);
1820 sz = htonl(sz);
968a1d65
DR
1821 return ((ce_write(context, fd, &ext, 4) < 0) ||
1822 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
1823}
1824
e93021b2 1825static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
4990aadc
LT
1826{
1827 unsigned int left = write_buffer_len;
ca9be054 1828
4990aadc
LT
1829 if (left) {
1830 write_buffer_len = 0;
9126f009 1831 git_SHA1_Update(context, write_buffer, left);
4990aadc 1832 }
ca9be054 1833
2c865d9a
QH
1834 /* Flush first if not enough space for SHA1 signature */
1835 if (left + 20 > WRITE_BUFFER_SIZE) {
93822c22 1836 if (write_in_full(fd, write_buffer, left) != left)
2c865d9a
QH
1837 return -1;
1838 left = 0;
1839 }
1840
ca9be054 1841 /* Append the SHA1 signature at the end */
9126f009 1842 git_SHA1_Final(write_buffer + left, context);
e93021b2 1843 hashcpy(sha1, write_buffer + left);
ca9be054 1844 left += 20;
93822c22 1845 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
4990aadc
LT
1846}
1847
407c8eb0
JH
1848static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1849{
1850 /*
1851 * The only thing we care about in this function is to smudge the
1852 * falsely clean entry due to touch-update-touch race, so we leave
1853 * everything else as they are. We are called for entries whose
c21d39d7 1854 * ce_stat_data.sd_mtime match the index file mtime.
c70115b4
JH
1855 *
1856 * Note that this actually does not do much for gitlinks, for
1857 * which ce_match_stat_basic() always goes to the actual
1858 * contents. The caller checks with is_racy_timestamp() which
1859 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
1860 */
1861 struct stat st;
1862
1863 if (lstat(ce->name, &st) < 0)
1864 return;
1865 if (ce_match_stat_basic(ce, &st))
1866 return;
1867 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
1868 /* This is "racily clean"; smudge it. Note that this
1869 * is a tricky code. At first glance, it may appear
1870 * that it can break with this sequence:
1871 *
1872 * $ echo xyzzy >frotz
1873 * $ git-update-index --add frotz
1874 * $ : >frotz
1875 * $ sleep 3
1876 * $ echo filfre >nitfol
1877 * $ git-update-index --add nitfol
1878 *
b7e58b17 1879 * but it does not. When the second update-index runs,
4b3511b0
JH
1880 * it notices that the entry "frotz" has the same timestamp
1881 * as index, and if we were to smudge it by resetting its
1882 * size to zero here, then the object name recorded
1883 * in index is the 6-byte file but the cached stat information
1884 * becomes zero --- which would then match what we would
a6080a0a 1885 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
1886 *
1887 * However, the second update-index, before calling
1888 * this function, notices that the cached size is 6
1889 * bytes and what is on the filesystem is an empty
1890 * file, and never calls us, so the cached size information
1891 * for "frotz" stays 6 which does not match the filesystem.
1892 */
c21d39d7 1893 ce->ce_stat_data.sd_size = 0;
407c8eb0
JH
1894 }
1895}
1896
f136f7bf
JH
1897/* Copy miscellaneous fields but not the name */
1898static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1899 struct cache_entry *ce)
7a51ed66 1900{
b60e188c
TG
1901 short flags;
1902
c21d39d7
MH
1903 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1904 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1905 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1906 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1907 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1908 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
7a51ed66 1909 ondisk->mode = htonl(ce->ce_mode);
c21d39d7
MH
1910 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1911 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1912 ondisk->size = htonl(ce->ce_stat_data.sd_size);
99d1a986 1913 hashcpy(ondisk->sha1, ce->oid.hash);
b60e188c 1914
ce51bf09 1915 flags = ce->ce_flags & ~CE_NAMEMASK;
b60e188c
TG
1916 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1917 ondisk->flags = htons(flags);
06aaaa0b
NTND
1918 if (ce->ce_flags & CE_EXTENDED) {
1919 struct ondisk_cache_entry_extended *ondisk2;
1920 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1921 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
f136f7bf 1922 return ondisk2->name;
06aaaa0b 1923 }
f136f7bf
JH
1924 else {
1925 return ondisk->name;
1926 }
1927}
1928
9d227781
JH
1929static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1930 struct strbuf *previous_name)
f136f7bf 1931{
9d227781
JH
1932 int size;
1933 struct ondisk_cache_entry *ondisk;
b3c96fb1 1934 int saved_namelen = saved_namelen; /* compiler workaround */
f136f7bf
JH
1935 char *name;
1936 int result;
1937
b3c96fb1
NTND
1938 if (ce->ce_flags & CE_STRIP_NAME) {
1939 saved_namelen = ce_namelen(ce);
1940 ce->ce_namelen = 0;
1941 }
1942
9d227781
JH
1943 if (!previous_name) {
1944 size = ondisk_ce_size(ce);
1945 ondisk = xcalloc(1, size);
1946 name = copy_cache_entry_to_ondisk(ondisk, ce);
1947 memcpy(name, ce->name, ce_namelen(ce));
1948 } else {
1949 int common, to_remove, prefix_size;
1950 unsigned char to_remove_vi[16];
1951 for (common = 0;
1952 (ce->name[common] &&
1953 common < previous_name->len &&
1954 ce->name[common] == previous_name->buf[common]);
1955 common++)
1956 ; /* still matching */
1957 to_remove = previous_name->len - common;
1958 prefix_size = encode_varint(to_remove, to_remove_vi);
1959
1960 if (ce->ce_flags & CE_EXTENDED)
1961 size = offsetof(struct ondisk_cache_entry_extended, name);
1962 else
1963 size = offsetof(struct ondisk_cache_entry, name);
1964 size += prefix_size + (ce_namelen(ce) - common + 1);
1965
1966 ondisk = xcalloc(1, size);
1967 name = copy_cache_entry_to_ondisk(ondisk, ce);
1968 memcpy(name, to_remove_vi, prefix_size);
1969 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1970
1971 strbuf_splice(previous_name, common, to_remove,
1972 ce->name + common, ce_namelen(ce) - common);
06aaaa0b 1973 }
b3c96fb1
NTND
1974 if (ce->ce_flags & CE_STRIP_NAME) {
1975 ce->ce_namelen = saved_namelen;
1976 ce->ce_flags &= ~CE_STRIP_NAME;
1977 }
7a51ed66 1978
59efba64
JN
1979 result = ce_write(c, fd, ondisk, size);
1980 free(ondisk);
1981 return result;
7a51ed66
LT
1982}
1983
426ddeea
YM
1984/*
1985 * This function verifies if index_state has the correct sha1 of the
1986 * index file. Don't die if we have any other failure, just return 0.
1987 */
1988static int verify_index_from(const struct index_state *istate, const char *path)
1989{
1990 int fd;
1991 ssize_t n;
1992 struct stat st;
1993 unsigned char sha1[20];
1994
1995 if (!istate->initialized)
1996 return 0;
1997
1998 fd = open(path, O_RDONLY);
1999 if (fd < 0)
2000 return 0;
2001
2002 if (fstat(fd, &st))
2003 goto out;
2004
2005 if (st.st_size < sizeof(struct cache_header) + 20)
2006 goto out;
2007
2008 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2009 if (n != 20)
2010 goto out;
2011
2012 if (hashcmp(istate->sha1, sha1))
2013 goto out;
2014
2015 close(fd);
2016 return 1;
2017
2018out:
2019 close(fd);
2020 return 0;
2021}
2022
2023static int verify_index(const struct index_state *istate)
2024{
2025 return verify_index_from(istate, get_index_file());
2026}
2027
483fbe2b
JH
2028static int has_racy_timestamp(struct index_state *istate)
2029{
2030 int entries = istate->cache_nr;
2031 int i;
2032
2033 for (i = 0; i < entries; i++) {
2034 struct cache_entry *ce = istate->cache[i];
2035 if (is_racy_timestamp(istate, ce))
2036 return 1;
2037 }
2038 return 0;
2039}
2040
ccdc4ec3 2041/*
98e023de 2042 * Opportunistically update the index but do not complain if we can't
ccdc4ec3
JH
2043 */
2044void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2045{
483fbe2b 2046 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
788cef81 2047 verify_index(istate) &&
03b86647 2048 write_locked_index(istate, lockfile, COMMIT_LOCK))
ccdc4ec3
JH
2049 rollback_lock_file(lockfile);
2050}
2051
c18b80a0
NTND
2052static int do_write_index(struct index_state *istate, int newfd,
2053 int strip_extensions)
197ee8c9 2054{
9126f009 2055 git_SHA_CTX c;
197ee8c9 2056 struct cache_header hdr;
9d227781 2057 int i, err, removed, extended, hdr_version;
4aab5b46
JH
2058 struct cache_entry **cache = istate->cache;
2059 int entries = istate->cache_nr;
e1afca4f 2060 struct stat st;
9d227781 2061 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
025a0709 2062
06aaaa0b 2063 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 2064 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 2065 removed++;
197ee8c9 2066
06aaaa0b
NTND
2067 /* reduce extended entries if possible */
2068 cache[i]->ce_flags &= ~CE_EXTENDED;
2069 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2070 extended++;
2071 cache[i]->ce_flags |= CE_EXTENDED;
2072 }
2073 }
2074
d6e3c181 2075 if (!istate->version) {
136347d7 2076 istate->version = get_index_format_default();
d6e3c181
NTND
2077 if (getenv("GIT_TEST_SPLIT_INDEX"))
2078 init_split_index(istate);
2079 }
9d227781
JH
2080
2081 /* demote version 3 to version 2 when the latter suffices */
2082 if (istate->version == 3 || istate->version == 2)
2083 istate->version = extended ? 3 : 2;
2084
2085 hdr_version = istate->version;
2086
ccc4feb5 2087 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
9d227781 2088 hdr.hdr_version = htonl(hdr_version);
025a0709 2089 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 2090
9126f009 2091 git_SHA1_Init(&c);
ca9be054 2092 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
2093 return -1;
2094
9d227781 2095 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
197ee8c9
LT
2096 for (i = 0; i < entries; i++) {
2097 struct cache_entry *ce = cache[i];
7a51ed66 2098 if (ce->ce_flags & CE_REMOVE)
aa16021e 2099 continue;
e06c43c7 2100 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 2101 ce_smudge_racily_clean_entry(ce);
99d1a986 2102 if (is_null_oid(&ce->oid)) {
83bd7437
JK
2103 static const char msg[] = "cache entry has null sha1: %s";
2104 static int allow = -1;
2105
2106 if (allow < 0)
2107 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2108 if (allow)
2109 warning(msg, ce->name);
2110 else
2111 return error(msg, ce->name);
2112 }
9d227781 2113 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
197ee8c9
LT
2114 return -1;
2115 }
9d227781 2116 strbuf_release(&previous_name_buf);
1af1c2b6 2117
bad68ec9 2118 /* Write extension data here */
c18b80a0 2119 if (!strip_extensions && istate->split_index) {
5fc2fc8f
NTND
2120 struct strbuf sb = STRBUF_INIT;
2121
2122 err = write_link_extension(&sb, istate) < 0 ||
2123 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2124 sb.len) < 0 ||
2125 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2126 strbuf_release(&sb);
2127 if (err)
2128 return -1;
2129 }
c18b80a0 2130 if (!strip_extensions && istate->cache_tree) {
f285a2d7 2131 struct strbuf sb = STRBUF_INIT;
1dffb8fa 2132
1dffb8fa
PH
2133 cache_tree_write(&sb, istate->cache_tree);
2134 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2135 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2136 strbuf_release(&sb);
2137 if (err)
bad68ec9 2138 return -1;
bad68ec9 2139 }
c18b80a0 2140 if (!strip_extensions && istate->resolve_undo) {
cfc5789a
JH
2141 struct strbuf sb = STRBUF_INIT;
2142
2143 resolve_undo_write(&sb, istate->resolve_undo);
2144 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2145 sb.len) < 0
2146 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2147 strbuf_release(&sb);
2148 if (err)
2149 return -1;
2150 }
83c094ad
NTND
2151 if (!strip_extensions && istate->untracked) {
2152 struct strbuf sb = STRBUF_INIT;
2153
2154 write_untracked_extension(&sb, istate->untracked);
2155 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2156 sb.len) < 0 ||
2157 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2158 strbuf_release(&sb);
2159 if (err)
2160 return -1;
2161 }
e1afca4f 2162
e93021b2 2163 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
e1afca4f 2164 return -1;
5bcf109c
KB
2165 istate->timestamp.sec = (unsigned int)st.st_mtime;
2166 istate->timestamp.nsec = ST_MTIME_NSEC(st);
e1afca4f 2167 return 0;
197ee8c9 2168}
e46bbcf6 2169
626f35c8
NTND
2170void set_alternate_index_output(const char *name)
2171{
2172 alternate_index_output = name;
2173}
2174
2175static int commit_locked_index(struct lock_file *lk)
2176{
751baced
MH
2177 if (alternate_index_output)
2178 return commit_lock_file_to(lk, alternate_index_output);
2179 else
626f35c8 2180 return commit_lock_file(lk);
626f35c8
NTND
2181}
2182
03b86647
NTND
2183static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2184 unsigned flags)
2185{
c99a4c2d 2186 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
03b86647
NTND
2187 if (ret)
2188 return ret;
2189 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2190 (COMMIT_LOCK | CLOSE_LOCK));
2191 if (flags & COMMIT_LOCK)
2192 return commit_locked_index(lock);
2193 else if (flags & CLOSE_LOCK)
2194 return close_lock_file(lock);
2195 else
2196 return ret;
2197}
2198
5fc2fc8f
NTND
2199static int write_split_index(struct index_state *istate,
2200 struct lock_file *lock,
2201 unsigned flags)
2202{
2203 int ret;
2204 prepare_to_write_split_index(istate);
2205 ret = do_write_locked_index(istate, lock, flags);
2206 finish_writing_split_index(istate);
2207 return ret;
2208}
2209
b9683722
CC
2210static const char *shared_index_expire = "2.weeks.ago";
2211
2212static unsigned long get_shared_index_expire_date(void)
2213{
2214 static unsigned long shared_index_expire_date;
2215 static int shared_index_expire_date_prepared;
2216
2217 if (!shared_index_expire_date_prepared) {
2218 git_config_get_expiry("splitindex.sharedindexexpire",
2219 &shared_index_expire);
2220 shared_index_expire_date = approxidate(shared_index_expire);
2221 shared_index_expire_date_prepared = 1;
2222 }
2223
2224 return shared_index_expire_date;
2225}
2226
2227static int should_delete_shared_index(const char *shared_index_path)
2228{
2229 struct stat st;
2230 unsigned long expiration;
2231
2232 /* Check timestamp */
2233 expiration = get_shared_index_expire_date();
2234 if (!expiration)
2235 return 0;
2236 if (stat(shared_index_path, &st))
2237 return error_errno(_("could not stat '%s"), shared_index_path);
2238 if (st.st_mtime > expiration)
2239 return 0;
2240
2241 return 1;
2242}
2243
2244static int clean_shared_index_files(const char *current_hex)
2245{
2246 struct dirent *de;
2247 DIR *dir = opendir(get_git_dir());
2248
2249 if (!dir)
2250 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2251
2252 while ((de = readdir(dir)) != NULL) {
2253 const char *sha1_hex;
2254 const char *shared_index_path;
2255 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2256 continue;
2257 if (!strcmp(sha1_hex, current_hex))
2258 continue;
2259 shared_index_path = git_path("%s", de->d_name);
2260 if (should_delete_shared_index(shared_index_path) > 0 &&
2261 unlink(shared_index_path))
2262 warning_errno(_("unable to unlink: %s"), shared_index_path);
2263 }
2264 closedir(dir);
2265
2266 return 0;
2267}
2268
f6ecc62d 2269static struct tempfile temporary_sharedindex;
c18b80a0 2270
a0a96756
NTND
2271static int write_shared_index(struct index_state *istate,
2272 struct lock_file *lock, unsigned flags)
c18b80a0
NTND
2273{
2274 struct split_index *si = istate->split_index;
c18b80a0
NTND
2275 int fd, ret;
2276
f6ecc62d 2277 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
a0a96756 2278 if (fd < 0) {
a0a96756
NTND
2279 hashclr(si->base_sha1);
2280 return do_write_locked_index(istate, lock, flags);
2281 }
c18b80a0
NTND
2282 move_cache_to_base_index(istate);
2283 ret = do_write_index(si->base, fd, 1);
c18b80a0 2284 if (ret) {
f6ecc62d 2285 delete_tempfile(&temporary_sharedindex);
c18b80a0
NTND
2286 return ret;
2287 }
f6ecc62d
MH
2288 ret = rename_tempfile(&temporary_sharedindex,
2289 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
b9683722 2290 if (!ret) {
c18b80a0 2291 hashcpy(si->base_sha1, si->base->sha1);
b9683722
CC
2292 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2293 }
2294
c18b80a0
NTND
2295 return ret;
2296}
2297
e6a1dd77
CC
2298static const int default_max_percent_split_change = 20;
2299
2300static int too_many_not_shared_entries(struct index_state *istate)
2301{
2302 int i, not_shared = 0;
2303 int max_split = git_config_get_max_percent_split_change();
2304
2305 switch (max_split) {
2306 case -1:
2307 /* not or badly configured: use the default value */
2308 max_split = default_max_percent_split_change;
2309 break;
2310 case 0:
2311 return 1; /* 0% means always write a new shared index */
2312 case 100:
2313 return 0; /* 100% means never write a new shared index */
2314 default:
2315 break; /* just use the configured value */
2316 }
2317
2318 /* Count not shared entries */
2319 for (i = 0; i < istate->cache_nr; i++) {
2320 struct cache_entry *ce = istate->cache[i];
2321 if (!ce->index)
2322 not_shared++;
2323 }
2324
2325 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2326}
2327
03b86647
NTND
2328int write_locked_index(struct index_state *istate, struct lock_file *lock,
2329 unsigned flags)
2330{
0d59ffb4 2331 int new_shared_index, ret;
5fc2fc8f
NTND
2332 struct split_index *si = istate->split_index;
2333
5165dd59
NTND
2334 if (!si || alternate_index_output ||
2335 (istate->cache_changed & ~EXTMASK)) {
5fc2fc8f
NTND
2336 if (si)
2337 hashclr(si->base_sha1);
2338 return do_write_locked_index(istate, lock, flags);
2339 }
2340
d6e3c181
NTND
2341 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2342 int v = si->base_sha1[0];
2343 if ((v & 15) < 6)
2344 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2345 }
e6a1dd77
CC
2346 if (too_many_not_shared_entries(istate))
2347 istate->cache_changed |= SPLIT_INDEX_ORDERED;
0d59ffb4
CC
2348
2349 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2350
2351 if (new_shared_index) {
2352 ret = write_shared_index(istate, lock, flags);
c18b80a0
NTND
2353 if (ret)
2354 return ret;
2355 }
2356
0d59ffb4
CC
2357 ret = write_split_index(istate, lock, flags);
2358
2359 /* Freshen the shared index only if the split-index was written */
2360 if (!ret && !new_shared_index)
2361 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2362
2363 return ret;
03b86647
NTND
2364}
2365
e46bbcf6
MV
2366/*
2367 * Read the index file that is potentially unmerged into given
63e8dc5b 2368 * index_state, dropping any unmerged entries. Returns true if
e46bbcf6
MV
2369 * the index is unmerged. Callers who want to refuse to work
2370 * from an unmerged state can call this and check its return value,
2371 * instead of calling read_cache().
2372 */
2373int read_index_unmerged(struct index_state *istate)
2374{
2375 int i;
d1a43f2a 2376 int unmerged = 0;
e46bbcf6
MV
2377
2378 read_index(istate);
e46bbcf6
MV
2379 for (i = 0; i < istate->cache_nr; i++) {
2380 struct cache_entry *ce = istate->cache[i];
d1a43f2a
JH
2381 struct cache_entry *new_ce;
2382 int size, len;
2383
2384 if (!ce_stage(ce))
e46bbcf6 2385 continue;
d1a43f2a 2386 unmerged = 1;
68c4f6a5 2387 len = ce_namelen(ce);
d1a43f2a
JH
2388 size = cache_entry_size(len);
2389 new_ce = xcalloc(1, size);
d1a43f2a 2390 memcpy(new_ce->name, ce->name, len);
b60e188c
TG
2391 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2392 new_ce->ce_namelen = len;
d1a43f2a
JH
2393 new_ce->ce_mode = ce->ce_mode;
2394 if (add_index_entry(istate, new_ce, 0))
2395 return error("%s: cannot drop to stage #0",
5699d17e 2396 new_ce->name);
e46bbcf6 2397 }
d1a43f2a 2398 return unmerged;
e46bbcf6 2399}
041aee31 2400
98fa4738
JK
2401/*
2402 * Returns 1 if the path is an "other" path with respect to
2403 * the index; that is, the path is not mentioned in the index at all,
2404 * either as a file, a directory with some files in the index,
2405 * or as an unmerged entry.
2406 *
2407 * We helpfully remove a trailing "/" from directories so that
2408 * the output of read_directory can be used as-is.
2409 */
2410int index_name_is_other(const struct index_state *istate, const char *name,
2411 int namelen)
2412{
2413 int pos;
2414 if (namelen && name[namelen - 1] == '/')
2415 namelen--;
2416 pos = index_name_pos(istate, name, namelen);
2417 if (0 <= pos)
2418 return 0; /* exact match */
2419 pos = -pos - 1;
2420 if (pos < istate->cache_nr) {
2421 struct cache_entry *ce = istate->cache[pos];
2422 if (ce_namelen(ce) == namelen &&
2423 !memcmp(ce->name, name, namelen))
2424 return 0; /* Yup, this one exists unmerged */
2425 }
2426 return 1;
2427}
29fb37b2 2428
ff366825 2429void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
29fb37b2
LF
2430{
2431 int pos, len;
2432 unsigned long sz;
2433 enum object_type type;
2434 void *data;
2435
2436 len = strlen(path);
2437 pos = index_name_pos(istate, path, len);
2438 if (pos < 0) {
2439 /*
2440 * We might be in the middle of a merge, in which
2441 * case we would read stage #2 (ours).
2442 */
2443 int i;
2444 for (i = -pos - 1;
2445 (pos < 0 && i < istate->cache_nr &&
2446 !strcmp(istate->cache[i]->name, path));
2447 i++)
2448 if (ce_stage(istate->cache[i]) == 2)
2449 pos = i;
2450 }
2451 if (pos < 0)
2452 return NULL;
99d1a986 2453 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
29fb37b2
LF
2454 if (!data || type != OBJ_BLOB) {
2455 free(data);
2456 return NULL;
2457 }
ff366825
LF
2458 if (size)
2459 *size = sz;
29fb37b2
LF
2460 return data;
2461}
38612532
MH
2462
2463void stat_validity_clear(struct stat_validity *sv)
2464{
2465 free(sv->sd);
2466 sv->sd = NULL;
2467}
2468
2469int stat_validity_check(struct stat_validity *sv, const char *path)
2470{
2471 struct stat st;
2472
2473 if (stat(path, &st) < 0)
2474 return sv->sd == NULL;
2475 if (!sv->sd)
2476 return 0;
2477 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2478}
2479
2480void stat_validity_update(struct stat_validity *sv, int fd)
2481{
2482 struct stat st;
2483
2484 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2485 stat_validity_clear(sv);
2486 else {
2487 if (!sv->sd)
2488 sv->sd = xcalloc(1, sizeof(struct stat_data));
2489 fill_stat_data(sv->sd, &st);
2490 }
2491}