read-cache: load cache extensions on a worker thread
[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"
cffbfad5
EN
9#include "diff.h"
10#include "diffcore.h"
f6ecc62d 11#include "tempfile.h"
697cc8ef 12#include "lockfile.h"
bad68ec9 13#include "cache-tree.h"
f35a6d3b 14#include "refs.h"
d616813d 15#include "dir.h"
cbd53a21 16#include "object-store.h"
041aee31
JH
17#include "tree.h"
18#include "commit.h"
39425819 19#include "blob.h"
cfc5789a 20#include "resolve-undo.h"
6c9cd161
JH
21#include "strbuf.h"
22#include "varint.h"
5fc2fc8f 23#include "split-index.h"
a42643aa 24#include "utf8.h"
883e248b 25#include "fsmonitor.h"
abb4bb83 26#include "thread-utils.h"
bad68ec9 27
b60e188c
TG
28/* Mask for the name length in ce_flags in the on-disk index */
29
30#define CE_NAMEMASK (0x0fff)
31
bad68ec9
JH
32/* Index extensions.
33 *
34 * The first letter should be 'A'..'Z' for extensions that are not
35 * necessary for a correct operation (i.e. optimization data).
36 * When new extensions are added that _needs_ to be understood in
37 * order to correctly interpret the index file, pick character that
38 * is outside the range, to cause the reader to abort.
39 */
40
41#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
42#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
b659b49b 43#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
5fc2fc8f 44#define CACHE_EXT_LINK 0x6c696e6b /* "link" */
83c094ad 45#define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
883e248b 46#define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
3b1d9e04 47#define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
5fc2fc8f
NTND
48
49/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
e0cf0d7d 50#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
c18b80a0 51 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
883e248b 52 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
e83c5163 53
8e72d675
JM
54
55/*
56 * This is an estimate of the pathname length in the index. We use
57 * this for V4 index files to guess the un-deltafied size of the index
58 * in memory because of pathname deltafication. This is not required
59 * for V2/V3 index formats because their pathnames are not compressed.
60 * If the initial amount of memory set aside is not sufficient, the
61 * mem pool will allocate extra memory.
62 */
63#define CACHE_ENTRY_PATH_LENGTH 80
64
65static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
66{
67 struct cache_entry *ce;
68 ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
69 ce->mem_pool_allocated = 1;
70 return ce;
71}
72
73static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
74{
75 struct cache_entry * ce;
76 ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
77 ce->mem_pool_allocated = 1;
78 return ce;
79}
80
81static struct mem_pool *find_mem_pool(struct index_state *istate)
82{
83 struct mem_pool **pool_ptr;
84
85 if (istate->split_index && istate->split_index->base)
86 pool_ptr = &istate->split_index->base->ce_mem_pool;
87 else
88 pool_ptr = &istate->ce_mem_pool;
89
90 if (!*pool_ptr)
91 mem_pool_init(pool_ptr, 0);
92
93 return *pool_ptr;
94}
95
228e94f9 96struct index_state the_index;
626f35c8 97static const char *alternate_index_output;
8fd2cb40 98
9cb76b8c
JH
99static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
100{
101 istate->cache[nr] = ce;
96872bc2 102 add_name_hash(istate, ce);
9cb76b8c
JH
103}
104
cf558704
LT
105static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
106{
107 struct cache_entry *old = istate->cache[nr];
108
078a58e8 109 replace_index_entry_in_base(istate, old, ce);
2092678c 110 remove_name_hash(istate, old);
a849735b 111 discard_cache_entry(old);
0e267b7a 112 ce->ce_flags &= ~CE_HASHED;
a22c6371 113 set_index_entry(istate, nr, ce);
078a58e8 114 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 115 mark_fsmonitor_invalid(istate, ce);
e636a7b4 116 istate->cache_changed |= CE_ENTRY_CHANGED;
cf558704
LT
117}
118
81dc2307
PB
119void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
120{
285c2e25 121 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
81dc2307
PB
122 int namelen = strlen(new_name);
123
a849735b 124 new_entry = make_empty_cache_entry(istate, namelen);
285c2e25
BW
125 copy_cache_entry(new_entry, old_entry);
126 new_entry->ce_flags &= ~CE_HASHED;
127 new_entry->ce_namelen = namelen;
128 new_entry->index = 0;
129 memcpy(new_entry->name, new_name, namelen + 1);
81dc2307 130
285c2e25
BW
131 cache_tree_invalidate_path(istate, old_entry->name);
132 untracked_cache_remove_from_index(istate, old_entry->name);
81dc2307 133 remove_index_entry_at(istate, nr);
285c2e25 134 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
81dc2307
PB
135}
136
c21d39d7
MH
137void fill_stat_data(struct stat_data *sd, struct stat *st)
138{
139 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
140 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
141 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
142 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
143 sd->sd_dev = st->st_dev;
144 sd->sd_ino = st->st_ino;
145 sd->sd_uid = st->st_uid;
146 sd->sd_gid = st->st_gid;
147 sd->sd_size = st->st_size;
148}
149
150int match_stat_data(const struct stat_data *sd, struct stat *st)
151{
152 int changed = 0;
153
154 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
155 changed |= MTIME_CHANGED;
156 if (trust_ctime && check_stat &&
157 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
158 changed |= CTIME_CHANGED;
159
160#ifdef USE_NSEC
161 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
162 changed |= MTIME_CHANGED;
163 if (trust_ctime && check_stat &&
164 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
165 changed |= CTIME_CHANGED;
166#endif
167
168 if (check_stat) {
169 if (sd->sd_uid != (unsigned int) st->st_uid ||
170 sd->sd_gid != (unsigned int) st->st_gid)
171 changed |= OWNER_CHANGED;
172 if (sd->sd_ino != (unsigned int) st->st_ino)
173 changed |= INODE_CHANGED;
174 }
175
176#ifdef USE_STDEV
177 /*
178 * st_dev breaks on network filesystems where different
179 * clients will have different views of what "device"
180 * the filesystem is on
181 */
182 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
183 changed |= INODE_CHANGED;
184#endif
185
186 if (sd->sd_size != (unsigned int) st->st_size)
187 changed |= DATA_CHANGED;
188
189 return changed;
190}
191
415e96c8
JH
192/*
193 * This only updates the "non-critical" parts of the directory
194 * cache, ie the parts that aren't tracked by GIT, and only used
195 * to validate the cache.
196 */
197void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
198{
c21d39d7 199 fill_stat_data(&ce->ce_stat_data, st);
5f73076c
JH
200
201 if (assume_unchanged)
7a51ed66 202 ce->ce_flags |= CE_VALID;
eadb5831 203
883e248b 204 if (S_ISREG(st->st_mode)) {
eadb5831 205 ce_mark_uptodate(ce);
883e248b
BP
206 mark_fsmonitor_valid(ce);
207 }
415e96c8
JH
208}
209
21a6b9fa 210static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
211{
212 int match = -1;
1b8ac5ea 213 int fd = git_open_cloexec(ce->name, O_RDONLY);
29e4d363
JH
214
215 if (fd >= 0) {
bebfecb9 216 struct object_id oid;
e3506559 217 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
6a29d7b7 218 match = !oideq(&oid, &ce->oid);
7f8508e8 219 /* index_fd() closed the file descriptor already */
29e4d363
JH
220 }
221 return match;
222}
223
21a6b9fa 224static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
29e4d363
JH
225{
226 int match = -1;
29e4d363
JH
227 void *buffer;
228 unsigned long size;
21666f1a 229 enum object_type type;
a60272b3 230 struct strbuf sb = STRBUF_INIT;
29e4d363 231
a60272b3 232 if (strbuf_readlink(&sb, ce->name, expected_size))
29e4d363 233 return -1;
a60272b3 234
b4f5aca4 235 buffer = read_object_file(&ce->oid, &type, &size);
a60272b3
LT
236 if (buffer) {
237 if (size == sb.len)
238 match = memcmp(buffer, sb.buf, size);
239 free(buffer);
29e4d363 240 }
a60272b3 241 strbuf_release(&sb);
29e4d363
JH
242 return match;
243}
244
21a6b9fa 245static int ce_compare_gitlink(const struct cache_entry *ce)
f35a6d3b 246{
1053fe82 247 struct object_id oid;
f35a6d3b
LT
248
249 /*
250 * We don't actually require that the .git directory
302b9282 251 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
252 * might even be missing (in case nobody populated that
253 * sub-project).
254 *
255 * If so, we consider it always to match.
256 */
a98e6101 257 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
f35a6d3b 258 return 0;
6a29d7b7 259 return !oideq(&oid, &ce->oid);
f35a6d3b
LT
260}
261
21a6b9fa 262static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
263{
264 switch (st->st_mode & S_IFMT) {
265 case S_IFREG:
266 if (ce_compare_data(ce, st))
267 return DATA_CHANGED;
268 break;
269 case S_IFLNK:
dc49cd76 270 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
271 return DATA_CHANGED;
272 break;
a8ee75bc 273 case S_IFDIR:
7a51ed66 274 if (S_ISGITLINK(ce->ce_mode))
c70115b4 275 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
1cf01a34 276 /* else fallthrough */
29e4d363
JH
277 default:
278 return TYPE_CHANGED;
279 }
280 return 0;
281}
282
21a6b9fa 283static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
734aab75
LT
284{
285 unsigned int changed = 0;
286
7a51ed66
LT
287 if (ce->ce_flags & CE_REMOVE)
288 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
289
290 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
291 case S_IFREG:
292 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
293 /* We consider only the owner x bit to be relevant for
294 * "mode changes"
295 */
296 if (trust_executable_bit &&
7a51ed66 297 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 298 changed |= MODE_CHANGED;
8ae0a8c5
KS
299 break;
300 case S_IFLNK:
78a8d641
JS
301 if (!S_ISLNK(st->st_mode) &&
302 (has_symlinks || !S_ISREG(st->st_mode)))
303 changed |= TYPE_CHANGED;
8ae0a8c5 304 break;
302b9282 305 case S_IFGITLINK:
c70115b4 306 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
307 if (!S_ISDIR(st->st_mode))
308 changed |= TYPE_CHANGED;
309 else if (ce_compare_gitlink(ce))
310 changed |= DATA_CHANGED;
a8ee75bc 311 return changed;
8ae0a8c5 312 default:
7a51ed66 313 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 314 }
2cb45e95 315
c21d39d7 316 changed |= match_stat_data(&ce->ce_stat_data, st);
b0391890 317
f49c2c22 318 /* Racily smudged entry? */
c21d39d7 319 if (!ce->ce_stat_data.sd_size) {
99d1a986 320 if (!is_empty_blob_sha1(ce->oid.hash))
f49c2c22
LT
321 changed |= DATA_CHANGED;
322 }
323
407c8eb0
JH
324 return changed;
325}
326
2bb4cda1
NTND
327static int is_racy_stat(const struct index_state *istate,
328 const struct stat_data *sd)
6d91da6d 329{
2bb4cda1 330 return (istate->timestamp.sec &&
fba2f38a
KB
331#ifdef USE_NSEC
332 /* nanosecond timestamped files can also be racy! */
2bb4cda1
NTND
333 (istate->timestamp.sec < sd->sd_mtime.sec ||
334 (istate->timestamp.sec == sd->sd_mtime.sec &&
335 istate->timestamp.nsec <= sd->sd_mtime.nsec))
fba2f38a 336#else
2bb4cda1 337 istate->timestamp.sec <= sd->sd_mtime.sec
fba2f38a 338#endif
2bb4cda1
NTND
339 );
340}
341
342static int is_racy_timestamp(const struct index_state *istate,
343 const struct cache_entry *ce)
344{
345 return (!S_ISGITLINK(ce->ce_mode) &&
346 is_racy_stat(istate, &ce->ce_stat_data));
6d91da6d
JH
347}
348
ed4efab1
NTND
349int match_stat_data_racy(const struct index_state *istate,
350 const struct stat_data *sd, struct stat *st)
351{
352 if (is_racy_stat(istate, sd))
353 return MTIME_CHANGED;
354 return match_stat_data(sd, st);
6d91da6d
JH
355}
356
883e248b 357int ie_match_stat(struct index_state *istate,
21a6b9fa 358 const struct cache_entry *ce, struct stat *st,
4bd5b7da 359 unsigned int options)
407c8eb0 360{
5f73076c 361 unsigned int changed;
4bd5b7da 362 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 363 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
4bd5b7da 364 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
883e248b 365 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
5f73076c 366
883e248b
BP
367 if (!ignore_fsmonitor)
368 refresh_fsmonitor(istate);
5f73076c
JH
369 /*
370 * If it's marked as always valid in the index, it's
371 * valid whatever the checked-out copy says.
56cac48c
NTND
372 *
373 * skip-worktree has the same effect with higher precedence
5f73076c 374 */
56cac48c
NTND
375 if (!ignore_skip_worktree && ce_skip_worktree(ce))
376 return 0;
7a51ed66 377 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c 378 return 0;
883e248b
BP
379 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
380 return 0;
5f73076c 381
331fcb59
JH
382 /*
383 * Intent-to-add entries have not been added, so the index entry
384 * by definition never matches what is in the work tree until it
385 * actually gets added.
386 */
895ff3b2 387 if (ce_intent_to_add(ce))
331fcb59
JH
388 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
389
5f73076c 390 changed = ce_match_stat_basic(ce, st);
407c8eb0 391
29e4d363
JH
392 /*
393 * Within 1 second of this sequence:
394 * echo xyzzy >file && git-update-index --add file
395 * running this command:
396 * echo frotz >file
397 * would give a falsely clean cache entry. The mtime and
398 * length match the cache, and other stat fields do not change.
399 *
400 * We could detect this at update-index time (the cache entry
401 * being registered/updated records the same time as "now")
402 * and delay the return from git-update-index, but that would
403 * effectively mean we can make at most one commit per second,
404 * which is not acceptable. Instead, we check cache entries
405 * whose mtime are the same as the index file timestamp more
5f73076c 406 * carefully than others.
29e4d363 407 */
6d91da6d 408 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
409 if (assume_racy_is_modified)
410 changed |= DATA_CHANGED;
411 else
412 changed |= ce_modified_check_fs(ce, st);
413 }
b0391890 414
29e4d363 415 return changed;
b0391890
JH
416}
417
883e248b 418int ie_modified(struct index_state *istate,
21a6b9fa
RS
419 const struct cache_entry *ce,
420 struct stat *st, unsigned int options)
b0391890 421{
29e4d363 422 int changed, changed_fs;
4bd5b7da
JH
423
424 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
425 if (!changed)
426 return 0;
b0391890
JH
427 /*
428 * If the mode or type has changed, there's no point in trying
429 * to refresh the entry - it's not going to match
430 */
431 if (changed & (MODE_CHANGED | TYPE_CHANGED))
432 return changed;
433
c70115b4
JH
434 /*
435 * Immediately after read-tree or update-index --cacheinfo,
436 * the length field is zero, as we have never even read the
437 * lstat(2) information once, and we cannot trust DATA_CHANGED
438 * returned by ie_match_stat() which in turn was returned by
439 * ce_match_stat_basic() to signal that the filesize of the
440 * blob changed. We have to actually go to the filesystem to
441 * see if the contents match, and if so, should answer "unchanged".
442 *
443 * The logic does not apply to gitlinks, as ce_match_stat_basic()
444 * already has checked the actual HEAD from the filesystem in the
445 * subproject. If ie_match_stat() already said it is different,
446 * then we know it is.
b0391890 447 */
c70115b4 448 if ((changed & DATA_CHANGED) &&
c21d39d7 449 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
b0391890
JH
450 return changed;
451
29e4d363
JH
452 changed_fs = ce_modified_check_fs(ce, st);
453 if (changed_fs)
454 return changed | changed_fs;
b0391890
JH
455 return 0;
456}
457
958ba6c9
LT
458int base_name_compare(const char *name1, int len1, int mode1,
459 const char *name2, int len2, int mode2)
460{
461 unsigned char c1, c2;
462 int len = len1 < len2 ? len1 : len2;
463 int cmp;
464
465 cmp = memcmp(name1, name2, len);
466 if (cmp)
467 return cmp;
468 c1 = name1[len];
469 c2 = name2[len];
1833a925 470 if (!c1 && S_ISDIR(mode1))
958ba6c9 471 c1 = '/';
1833a925 472 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
473 c2 = '/';
474 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
475}
476
0ab9e1e8
LT
477/*
478 * df_name_compare() is identical to base_name_compare(), except it
479 * compares conflicting directory/file entries as equal. Note that
480 * while a directory name compares as equal to a regular file, they
481 * then individually compare _differently_ to a filename that has
482 * a dot after the basename (because '\0' < '.' < '/').
483 *
484 * This is used by routines that want to traverse the git namespace
485 * but then handle conflicting entries together when possible.
486 */
487int df_name_compare(const char *name1, int len1, int mode1,
488 const char *name2, int len2, int mode2)
489{
490 int len = len1 < len2 ? len1 : len2, cmp;
491 unsigned char c1, c2;
492
493 cmp = memcmp(name1, name2, len);
494 if (cmp)
495 return cmp;
496 /* Directories and files compare equal (same length, same name) */
497 if (len1 == len2)
498 return 0;
499 c1 = name1[len];
500 if (!c1 && S_ISDIR(mode1))
501 c1 = '/';
502 c2 = name2[len];
503 if (!c2 && S_ISDIR(mode2))
504 c2 = '/';
505 if (c1 == '/' && !c2)
506 return 0;
507 if (c2 == '/' && !c1)
508 return 0;
509 return c1 - c2;
510}
511
ccdd4a0f 512int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
eb38c22f 513{
ccdd4a0f
JM
514 size_t min_len = (len1 < len2) ? len1 : len2;
515 int cmp = memcmp(name1, name2, min_len);
eb38c22f
LT
516 if (cmp)
517 return cmp;
518 if (len1 < len2)
519 return -1;
520 if (len1 > len2)
521 return 1;
ccdd4a0f
JM
522 return 0;
523}
524
525int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
526{
527 int cmp;
528
529 cmp = name_compare(name1, len1, name2, len2);
530 if (cmp)
531 return cmp;
5f73076c 532
b60e188c 533 if (stage1 < stage2)
95fd5bf8 534 return -1;
b60e188c 535 if (stage1 > stage2)
95fd5bf8 536 return 1;
eb38c22f
LT
537 return 0;
538}
539
357e9c69 540static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
eb38c22f
LT
541{
542 int first, last;
543
544 first = 0;
4aab5b46 545 last = istate->cache_nr;
eb38c22f
LT
546 while (last > first) {
547 int next = (last + first) >> 1;
4aab5b46 548 struct cache_entry *ce = istate->cache[next];
b60e188c 549 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
eb38c22f 550 if (!cmp)
76e7f4ec 551 return next;
eb38c22f
LT
552 if (cmp < 0) {
553 last = next;
554 continue;
555 }
556 first = next+1;
557 }
76e7f4ec 558 return -first-1;
eb38c22f
LT
559}
560
b60e188c
TG
561int index_name_pos(const struct index_state *istate, const char *name, int namelen)
562{
563 return index_name_stage_pos(istate, name, namelen, 0);
564}
565
4aab5b46 566int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 567{
cf558704
LT
568 struct cache_entry *ce = istate->cache[pos];
569
cfc5789a 570 record_resolve_undo(istate, ce);
2092678c 571 remove_name_hash(istate, ce);
045113a5 572 save_or_free_index_entry(istate, ce);
e636a7b4 573 istate->cache_changed |= CE_ENTRY_REMOVED;
4aab5b46
JH
574 istate->cache_nr--;
575 if (pos >= istate->cache_nr)
7b937ca3 576 return 0;
f331ab9d
RS
577 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
578 istate->cache_nr - pos);
7b937ca3
LT
579 return 1;
580}
581
36419c8e 582/*
98e023de 583 * Remove all cache entries marked for removal, that is where
36419c8e
KB
584 * CE_REMOVE is set in ce_flags. This is much more effective than
585 * calling remove_index_entry_at() for each entry to be removed.
586 */
587void remove_marked_cache_entries(struct index_state *istate)
588{
589 struct cache_entry **ce_array = istate->cache;
590 unsigned int i, j;
591
592 for (i = j = 0; i < istate->cache_nr; i++) {
5699d17e 593 if (ce_array[i]->ce_flags & CE_REMOVE) {
2092678c 594 remove_name_hash(istate, ce_array[i]);
045113a5 595 save_or_free_index_entry(istate, ce_array[i]);
5699d17e 596 }
36419c8e
KB
597 else
598 ce_array[j++] = ce_array[i];
599 }
ad837d9e
NTND
600 if (j == istate->cache_nr)
601 return;
e636a7b4 602 istate->cache_changed |= CE_ENTRY_REMOVED;
36419c8e
KB
603 istate->cache_nr = j;
604}
605
4aab5b46 606int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 607{
4aab5b46 608 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
609 if (pos < 0)
610 pos = -pos-1;
a5400efe 611 cache_tree_invalidate_path(istate, path);
e931371a 612 untracked_cache_remove_from_index(istate, path);
4aab5b46
JH
613 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
614 remove_index_entry_at(istate, pos);
197ee8c9
LT
615 return 0;
616}
617
20314271
JS
618static int compare_name(struct cache_entry *ce, const char *path, int namelen)
619{
620 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
621}
622
623static int index_name_pos_also_unmerged(struct index_state *istate,
624 const char *path, int namelen)
625{
626 int pos = index_name_pos(istate, path, namelen);
627 struct cache_entry *ce;
628
629 if (pos >= 0)
630 return pos;
631
632 /* maybe unmerged? */
633 pos = -1 - pos;
634 if (pos >= istate->cache_nr ||
635 compare_name((ce = istate->cache[pos]), path, namelen))
636 return -1;
637
638 /* order of preference: stage 2, 1, 3 */
639 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
640 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
641 !compare_name(ce, path, namelen))
642 pos++;
643 return pos;
644}
645
1102952b
LT
646static int different_name(struct cache_entry *ce, struct cache_entry *alias)
647{
648 int len = ce_namelen(ce);
649 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
650}
651
652/*
653 * If we add a filename that aliases in the cache, we will use the
654 * name that we already have - but we don't want to update the same
655 * alias twice, because that implies that there were actually two
656 * different files with aliasing names!
657 *
658 * So we use the CE_ADDED flag to verify that the alias was an old
659 * one before we accept it as
660 */
045113a5
NTND
661static struct cache_entry *create_alias_ce(struct index_state *istate,
662 struct cache_entry *ce,
663 struct cache_entry *alias)
1102952b
LT
664{
665 int len;
285c2e25 666 struct cache_entry *new_entry;
1102952b
LT
667
668 if (alias->ce_flags & CE_ADDED)
669 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
670
671 /* Ok, create the new entry using the name of the existing alias */
672 len = ce_namelen(alias);
a849735b 673 new_entry = make_empty_cache_entry(istate, len);
285c2e25
BW
674 memcpy(new_entry->name, alias->name, len);
675 copy_cache_entry(new_entry, ce);
045113a5 676 save_or_free_index_entry(istate, ce);
285c2e25 677 return new_entry;
1102952b
LT
678}
679
b4b313f9 680void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
39425819 681{
a09c985e
PO
682 struct object_id oid;
683 if (write_object_file("", 0, blob_type, &oid))
39425819 684 die("cannot create an empty blob in the object database");
a09c985e 685 oidcpy(&ce->oid, &oid);
39425819
JH
686}
687
610d55af 688int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 689{
a849735b 690 int namelen, was_same;
d177cab0 691 mode_t st_mode = st->st_mode;
9472935d 692 struct cache_entry *ce, *alias = NULL;
56cac48c 693 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
694 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
695 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
696 int intent_only = flags & ADD_CACHE_INTENT;
697 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
698 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
9472935d
TB
699 int newflags = HASH_WRITE_OBJECT;
700
701 if (flags & HASH_RENORMALIZE)
702 newflags |= HASH_RENORMALIZE;
11be42a4 703
d177cab0 704 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
960b8ad1 705 return error("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
706
707 namelen = strlen(path);
d177cab0 708 if (S_ISDIR(st_mode)) {
09595258
LT
709 while (namelen && path[namelen-1] == '/')
710 namelen--;
711 }
a849735b 712 ce = make_empty_cache_entry(istate, namelen);
11be42a4 713 memcpy(ce->name, path, namelen);
b60e188c 714 ce->ce_namelen = namelen;
39425819
JH
715 if (!intent_only)
716 fill_stat_cache_info(ce, st);
388b2acd
JH
717 else
718 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 719
610d55af
TG
720
721 if (trust_executable_bit && has_symlinks) {
d177cab0 722 ce->ce_mode = create_ce_mode(st_mode);
610d55af 723 } else {
78a8d641
JS
724 /* If there is an existing entry, pick the mode bits and type
725 * from it, otherwise assume unexecutable regular file.
11be42a4 726 */
185c975f 727 struct cache_entry *ent;
20314271 728 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 729
4aab5b46 730 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 731 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
732 }
733
dc1ae704
JJ
734 /* When core.ignorecase=true, determine if a directory of the same name but differing
735 * case already exists within the Git repository. If it does, ensure the directory
736 * case of the file being added to the repository matches (is folded into) the existing
737 * entry's directory case.
738 */
739 if (ignore_case) {
41284eb0 740 adjust_dirname_case(istate, ce->name);
dc1ae704 741 }
9472935d
TB
742 if (!(flags & HASH_RENORMALIZE)) {
743 alias = index_file_exists(istate, ce->name,
744 ce_namelen(ce), ignore_case);
745 if (alias &&
746 !ce_stage(alias) &&
747 !ie_match_stat(istate, alias, st, ce_option)) {
748 /* Nothing changed, really */
749 if (!S_ISGITLINK(alias->ce_mode))
750 ce_mark_uptodate(alias);
751 alias->ce_flags |= CE_ADDED;
dc1ae704 752
a849735b 753 discard_cache_entry(ce);
9472935d
TB
754 return 0;
755 }
0781b8a9 756 }
39425819 757 if (!intent_only) {
9472935d 758 if (index_path(&ce->oid, path, st, newflags)) {
a849735b 759 discard_cache_entry(ce);
39425819 760 return error("unable to index file %s", path);
2d9426b0 761 }
39425819 762 } else
b4b313f9 763 set_object_name_for_intent_to_add_entry(ce);
39425819 764
1102952b 765 if (ignore_case && alias && different_name(ce, alias))
045113a5 766 ce = create_alias_ce(istate, ce, alias);
1102952b 767 ce->ce_flags |= CE_ADDED;
38ed1d89 768
3bf0dd1f 769 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
770 was_same = (alias &&
771 !ce_stage(alias) &&
4a7e27e9 772 oideq(&alias->oid, &ce->oid) &&
38ed1d89
JH
773 ce->ce_mode == alias->ce_mode);
774
775 if (pretend)
a849735b 776 discard_cache_entry(ce);
067178ed 777 else if (add_index_entry(istate, ce, add_option)) {
a849735b 778 discard_cache_entry(ce);
067178ed
JH
779 return error("unable to add %s to index", path);
780 }
38ed1d89 781 if (verbose && !was_same)
11be42a4 782 printf("add '%s'\n", path);
11be42a4
JS
783 return 0;
784}
785
610d55af 786int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
787{
788 struct stat st;
789 if (lstat(path, &st))
d824cbba 790 die_errno("unable to stat '%s'", path);
610d55af 791 return add_to_index(istate, path, &st, flags);
d177cab0
LT
792}
793
a849735b
JM
794struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
795{
8e72d675 796 return mem_pool__ce_calloc(find_mem_pool(istate), len);
a849735b
JM
797}
798
799struct cache_entry *make_empty_transient_cache_entry(size_t len)
800{
801 return xcalloc(1, cache_entry_size(len));
802}
803
804struct cache_entry *make_cache_entry(struct index_state *istate,
805 unsigned int mode,
825ed4d9
JM
806 const struct object_id *oid,
807 const char *path,
808 int stage,
809 unsigned int refresh_options)
6640f881 810{
bc1c2caa 811 struct cache_entry *ce, *ret;
a849735b 812 int len;
6640f881 813
10ecfa76 814 if (!verify_path(path, mode)) {
7e7abea9 815 error("Invalid path '%s'", path);
6640f881 816 return NULL;
7e7abea9 817 }
6640f881
CR
818
819 len = strlen(path);
a849735b 820 ce = make_empty_cache_entry(istate, len);
6640f881 821
825ed4d9 822 oidcpy(&ce->oid, oid);
6640f881 823 memcpy(ce->name, path, len);
b60e188c
TG
824 ce->ce_flags = create_ce_flags(stage);
825 ce->ce_namelen = len;
6640f881
CR
826 ce->ce_mode = create_ce_mode(mode);
827
768d7965 828 ret = refresh_cache_entry(&the_index, ce, refresh_options);
915e44c6 829 if (ret != ce)
a849735b 830 discard_cache_entry(ce);
915e44c6 831 return ret;
6640f881
CR
832}
833
a849735b
JM
834struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
835 const char *path, int stage)
836{
837 struct cache_entry *ce;
838 int len;
839
840 if (!verify_path(path, mode)) {
841 error("Invalid path '%s'", path);
842 return NULL;
843 }
844
845 len = strlen(path);
846 ce = make_empty_transient_cache_entry(len);
847
848 oidcpy(&ce->oid, oid);
849 memcpy(ce->name, path, len);
850 ce->ce_flags = create_ce_flags(stage);
851 ce->ce_namelen = len;
852 ce->ce_mode = create_ce_mode(mode);
853
854 return ce;
855}
856
d9d70966
TG
857/*
858 * Chmod an index entry with either +x or -x.
859 *
860 * Returns -1 if the chmod for the particular cache entry failed (if it's
861 * not a regular file), -2 if an invalid flip argument is passed in, 0
862 * otherwise.
863 */
864int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
865 char flip)
866{
867 if (!S_ISREG(ce->ce_mode))
868 return -1;
869 switch (flip) {
870 case '+':
871 ce->ce_mode |= 0111;
872 break;
873 case '-':
874 ce->ce_mode &= ~0111;
875 break;
876 default:
877 return -2;
878 }
879 cache_tree_invalidate_path(istate, ce->name);
880 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 881 mark_fsmonitor_invalid(istate, ce);
d9d70966
TG
882 istate->cache_changed |= CE_ENTRY_CHANGED;
883
884 return 0;
885}
886
9c5e6c80 887int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
7b937ca3
LT
888{
889 int len = ce_namelen(a);
890 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
891}
892
8dcf39c4
LT
893/*
894 * We fundamentally don't like some paths: we don't want
895 * dot or dot-dot anywhere, and for obvious reasons don't
896 * want to recurse into ".git" either.
897 *
898 * Also, we don't want double slashes or slashes at the
899 * end that can make pathnames ambiguous.
900 */
10ecfa76 901static int verify_dotfile(const char *rest, unsigned mode)
8dcf39c4
LT
902{
903 /*
904 * The first character was '.', but that
905 * has already been discarded, we now test
906 * the rest.
907 */
e0f530ff 908
8dcf39c4 909 /* "." is not allowed */
e0f530ff 910 if (*rest == '\0' || is_dir_sep(*rest))
8dcf39c4
LT
911 return 0;
912
e0f530ff 913 switch (*rest) {
8dcf39c4 914 /*
641084b6
JK
915 * ".git" followed by NUL or slash is bad. Note that we match
916 * case-insensitively here, even if ignore_case is not set.
917 * This outlaws ".GIT" everywhere out of an abundance of caution,
918 * since there's really no good reason to allow it.
10ecfa76
JK
919 *
920 * Once we've seen ".git", we can also find ".gitmodules", etc (also
921 * case-insensitively).
8dcf39c4
LT
922 */
923 case 'g':
cc2fc7c2
JK
924 case 'G':
925 if (rest[1] != 'i' && rest[1] != 'I')
8dcf39c4 926 break;
cc2fc7c2 927 if (rest[2] != 't' && rest[2] != 'T')
8dcf39c4 928 break;
e19e5e66
JK
929 if (rest[3] == '\0' || is_dir_sep(rest[3]))
930 return 0;
10ecfa76
JK
931 if (S_ISLNK(mode)) {
932 rest += 3;
933 if (skip_iprefix(rest, "modules", &rest) &&
934 (*rest == '\0' || is_dir_sep(*rest)))
935 return 0;
936 }
e19e5e66 937 break;
8dcf39c4 938 case '.':
e0f530ff 939 if (rest[1] == '\0' || is_dir_sep(rest[1]))
8dcf39c4
LT
940 return 0;
941 }
942 return 1;
943}
944
10ecfa76 945int verify_path(const char *path, unsigned mode)
8dcf39c4
LT
946{
947 char c;
948
56948cb6
EFL
949 if (has_dos_drive_prefix(path))
950 return 0;
951
8dcf39c4
LT
952 goto inside;
953 for (;;) {
954 if (!c)
955 return 1;
56948cb6 956 if (is_dir_sep(c)) {
8dcf39c4 957inside:
10ecfa76
JK
958 if (protect_hfs) {
959 if (is_hfs_dotgit(path))
960 return 0;
961 if (S_ISLNK(mode)) {
962 if (is_hfs_dotgitmodules(path))
963 return 0;
964 }
965 }
966 if (protect_ntfs) {
967 if (is_ntfs_dotgit(path))
968 return 0;
969 if (S_ISLNK(mode)) {
970 if (is_ntfs_dotgitmodules(path))
971 return 0;
972 }
973 }
974
8dcf39c4 975 c = *path++;
10ecfa76 976 if ((c == '.' && !verify_dotfile(path, mode)) ||
3bdf09c7
JH
977 is_dir_sep(c) || c == '\0')
978 return 0;
8dcf39c4
LT
979 }
980 c = *path++;
981 }
982}
983
12676608
LT
984/*
985 * Do we have another file that has the beginning components being a
986 * proper superset of the name we're trying to add?
0f1e4f04 987 */
4aab5b46
JH
988static int has_file_name(struct index_state *istate,
989 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 990{
12676608
LT
991 int retval = 0;
992 int len = ce_namelen(ce);
b155725d 993 int stage = ce_stage(ce);
12676608 994 const char *name = ce->name;
0f1e4f04 995
4aab5b46
JH
996 while (pos < istate->cache_nr) {
997 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 998
12676608 999 if (len >= ce_namelen(p))
0f1e4f04 1000 break;
12676608
LT
1001 if (memcmp(name, p->name, len))
1002 break;
b155725d
JH
1003 if (ce_stage(p) != stage)
1004 continue;
12676608
LT
1005 if (p->name[len] != '/')
1006 continue;
7a51ed66 1007 if (p->ce_flags & CE_REMOVE)
21cd8d00 1008 continue;
12676608
LT
1009 retval = -1;
1010 if (!ok_to_replace)
1011 break;
4aab5b46 1012 remove_index_entry_at(istate, --pos);
0f1e4f04 1013 }
12676608
LT
1014 return retval;
1015}
0f1e4f04 1016
a6db3fbb
JH
1017
1018/*
1019 * Like strcmp(), but also return the offset of the first change.
1020 * If strings are equal, return the length.
1021 */
1022int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1023{
1024 size_t k;
1025
1026 if (!first_change)
1027 return strcmp(s1, s2);
1028
1029 for (k = 0; s1[k] == s2[k]; k++)
1030 if (s1[k] == '\0')
1031 break;
1032
1033 *first_change = k;
1034 return (unsigned char)s1[k] - (unsigned char)s2[k];
1035}
1036
12676608
LT
1037/*
1038 * Do we have another file with a pathname that is a proper
1039 * subset of the name we're trying to add?
06b6d81b
JH
1040 *
1041 * That is, is there another file in the index with a path
1042 * that matches a sub-directory in the given entry?
12676608 1043 */
4aab5b46
JH
1044static int has_dir_name(struct index_state *istate,
1045 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
1046{
1047 int retval = 0;
b155725d 1048 int stage = ce_stage(ce);
12676608
LT
1049 const char *name = ce->name;
1050 const char *slash = name + ce_namelen(ce);
06b6d81b
JH
1051 size_t len_eq_last;
1052 int cmp_last = 0;
1053
1054 /*
1055 * We are frequently called during an iteration on a sorted
1056 * list of pathnames and while building a new index. Therefore,
1057 * there is a high probability that this entry will eventually
1058 * be appended to the index, rather than inserted in the middle.
1059 * If we can confirm that, we can avoid binary searches on the
1060 * components of the pathname.
1061 *
1062 * Compare the entry's full path with the last path in the index.
1063 */
1064 if (istate->cache_nr > 0) {
1065 cmp_last = strcmp_offset(name,
1066 istate->cache[istate->cache_nr - 1]->name,
1067 &len_eq_last);
1068 if (cmp_last > 0) {
1069 if (len_eq_last == 0) {
1070 /*
1071 * The entry sorts AFTER the last one in the
1072 * index and their paths have no common prefix,
1073 * so there cannot be a F/D conflict.
1074 */
1075 return retval;
1076 } else {
1077 /*
1078 * The entry sorts AFTER the last one in the
1079 * index, but has a common prefix. Fall through
1080 * to the loop below to disect the entry's path
1081 * and see where the difference is.
1082 */
1083 }
1084 } else if (cmp_last == 0) {
1085 /*
1086 * The entry exactly matches the last one in the
1087 * index, but because of multiple stage and CE_REMOVE
1088 * items, we fall through and let the regular search
1089 * code handle it.
1090 */
1091 }
1092 }
0f1e4f04 1093
12676608 1094 for (;;) {
b986df5c 1095 size_t len;
0f1e4f04 1096
12676608
LT
1097 for (;;) {
1098 if (*--slash == '/')
1099 break;
1100 if (slash <= ce->name)
1101 return retval;
1102 }
1103 len = slash - name;
0f1e4f04 1104
b986df5c
JH
1105 if (cmp_last > 0) {
1106 /*
1107 * (len + 1) is a directory boundary (including
1108 * the trailing slash). And since the loop is
1109 * decrementing "slash", the first iteration is
1110 * the longest directory prefix; subsequent
1111 * iterations consider parent directories.
1112 */
1113
1114 if (len + 1 <= len_eq_last) {
1115 /*
1116 * The directory prefix (including the trailing
1117 * slash) also appears as a prefix in the last
1118 * entry, so the remainder cannot collide (because
1119 * strcmp said the whole path was greater).
1120 *
1121 * EQ: last: xxx/A
1122 * this: xxx/B
1123 *
1124 * LT: last: xxx/file_A
1125 * this: xxx/file_B
1126 */
1127 return retval;
1128 }
1129
1130 if (len > len_eq_last) {
1131 /*
1132 * This part of the directory prefix (excluding
1133 * the trailing slash) is longer than the known
1134 * equal portions, so this sub-directory cannot
1135 * collide with a file.
1136 *
1137 * GT: last: xxxA
1138 * this: xxxB/file
1139 */
1140 return retval;
1141 }
1142
1143 if (istate->cache_nr > 0 &&
1144 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1145 /*
1146 * The directory prefix lines up with part of
1147 * a longer file or directory name, but sorts
1148 * after it, so this sub-directory cannot
1149 * collide with a file.
1150 *
1151 * last: xxx/yy-file (because '-' sorts before '/')
1152 * this: xxx/yy/abc
1153 */
1154 return retval;
1155 }
1156
1157 /*
1158 * This is a possible collision. Fall through and
1159 * let the regular search code handle it.
1160 *
1161 * last: xxx
1162 * this: xxx/file
1163 */
1164 }
1165
b60e188c 1166 pos = index_name_stage_pos(istate, name, len, stage);
12676608 1167 if (pos >= 0) {
21cd8d00
JH
1168 /*
1169 * Found one, but not so fast. This could
1170 * be a marker that says "I was here, but
1171 * I am being removed". Such an entry is
1172 * not a part of the resulting tree, and
1173 * it is Ok to have a directory at the same
1174 * path.
1175 */
077c48df 1176 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
1177 retval = -1;
1178 if (!ok_to_replace)
1179 break;
4aab5b46 1180 remove_index_entry_at(istate, pos);
21cd8d00
JH
1181 continue;
1182 }
12676608 1183 }
21cd8d00
JH
1184 else
1185 pos = -pos-1;
12676608
LT
1186
1187 /*
1188 * Trivial optimization: if we find an entry that
1189 * already matches the sub-directory, then we know
b155725d 1190 * we're ok, and we can exit.
12676608 1191 */
4aab5b46
JH
1192 while (pos < istate->cache_nr) {
1193 struct cache_entry *p = istate->cache[pos];
b155725d
JH
1194 if ((ce_namelen(p) <= len) ||
1195 (p->name[len] != '/') ||
1196 memcmp(p->name, name, len))
1197 break; /* not our subdirectory */
077c48df
JH
1198 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1199 /*
1200 * p is at the same stage as our entry, and
b155725d
JH
1201 * is a subdirectory of what we are looking
1202 * at, so we cannot have conflicts at our
1203 * level or anything shorter.
1204 */
1205 return retval;
1206 pos++;
192268c1 1207 }
0f1e4f04 1208 }
12676608
LT
1209 return retval;
1210}
1211
1212/* We may be in a situation where we already have path/file and path
1213 * is being added, or we already have path and path/file is being
1214 * added. Either one would result in a nonsense tree that has path
1215 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 1216 *
12676608
LT
1217 * If ok-to-replace is specified, we remove the conflicting entries
1218 * from the cache so the caller should recompute the insert position.
1219 * When this happens, we return non-zero.
1220 */
4aab5b46
JH
1221static int check_file_directory_conflict(struct index_state *istate,
1222 const struct cache_entry *ce,
1223 int pos, int ok_to_replace)
12676608 1224{
21cd8d00
JH
1225 int retval;
1226
1227 /*
1228 * When ce is an "I am going away" entry, we allow it to be added
1229 */
7a51ed66 1230 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
1231 return 0;
1232
12676608
LT
1233 /*
1234 * We check if the path is a sub-path of a subsequent pathname
1235 * first, since removing those will not change the position
21cd8d00 1236 * in the array.
12676608 1237 */
4aab5b46 1238 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 1239
12676608
LT
1240 /*
1241 * Then check if the path might have a clashing sub-directory
1242 * before it.
1243 */
4aab5b46 1244 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
1245}
1246
af3785dc 1247static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
1248{
1249 int pos;
192268c1
JH
1250 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1251 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 1252 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 1253 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 1254
ce7c614b
NTND
1255 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1256 cache_tree_invalidate_path(istate, ce->name);
e5494631
JH
1257
1258 /*
1259 * If this entry's path sorts after the last entry in the index,
1260 * we can avoid searching for it.
1261 */
1262 if (istate->cache_nr > 0 &&
1263 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1264 pos = -istate->cache_nr - 1;
1265 else
1266 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
197ee8c9 1267
3e09cdfd 1268 /* existing match? Just replace it. */
76e7f4ec 1269 if (pos >= 0) {
39425819
JH
1270 if (!new_only)
1271 replace_index_entry(istate, pos, ce);
197ee8c9
LT
1272 return 0;
1273 }
76e7f4ec 1274 pos = -pos-1;
197ee8c9 1275
ffcc9ba7
NTND
1276 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1277 untracked_cache_add_to_index(istate, ce->name);
e931371a 1278
7b937ca3
LT
1279 /*
1280 * Inserting a merged entry ("stage 0") into the index
1281 * will always replace all non-merged entries..
1282 */
4aab5b46
JH
1283 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1284 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 1285 ok_to_add = 1;
4aab5b46 1286 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
1287 break;
1288 }
1289 }
1290
121481ab
LT
1291 if (!ok_to_add)
1292 return -1;
10ecfa76 1293 if (!verify_path(ce->name, ce->ce_mode))
7e7abea9 1294 return error("Invalid path '%s'", ce->name);
121481ab 1295
3e09cdfd 1296 if (!skip_df_check &&
4aab5b46 1297 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 1298 if (!ok_to_replace)
4aab5b46
JH
1299 return error("'%s' appears as both a file and as a directory",
1300 ce->name);
b60e188c 1301 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
192268c1
JH
1302 pos = -pos-1;
1303 }
af3785dc
JH
1304 return pos + 1;
1305}
1306
1307int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1308{
1309 int pos;
1310
1311 if (option & ADD_CACHE_JUST_APPEND)
1312 pos = istate->cache_nr;
1313 else {
1314 int ret;
1315 ret = add_index_entry_with_check(istate, ce, option);
1316 if (ret <= 0)
1317 return ret;
1318 pos = ret - 1;
1319 }
0f1e4f04 1320
197ee8c9 1321 /* Make sure the array is big enough .. */
999f5660 1322 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
197ee8c9
LT
1323
1324 /* Add it in.. */
4aab5b46 1325 istate->cache_nr++;
af3785dc 1326 if (istate->cache_nr > pos + 1)
f919ffeb
SG
1327 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1328 istate->cache_nr - pos - 1);
cf558704 1329 set_index_entry(istate, pos, ce);
e636a7b4 1330 istate->cache_changed |= CE_ENTRY_ADDED;
197ee8c9
LT
1331 return 0;
1332}
1333
405e5b2f
LT
1334/*
1335 * "refresh" does not calculate a new sha1 file or bring the
1336 * cache up-to-date for mode/content changes. But what it
1337 * _does_ do is to "re-match" the stat information of a file
1338 * with the cache, so that you can refresh the cache for a
1339 * file that hasn't been changed but where the stat entry is
1340 * out of date.
1341 *
1342 * For example, you'd want to do this after doing a "git-read-tree",
1343 * to link up the stat cache details with the proper files.
1344 */
4aab5b46 1345static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da 1346 struct cache_entry *ce,
d05e6970
JK
1347 unsigned int options, int *err,
1348 int *changed_ret)
405e5b2f
LT
1349{
1350 struct stat st;
1351 struct cache_entry *updated;
a849735b 1352 int changed;
25762726 1353 int refresh = options & CE_MATCH_REFRESH;
4bd5b7da 1354 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1355 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
2e2e7ec1 1356 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
883e248b 1357 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
405e5b2f 1358
25762726 1359 if (!refresh || ce_uptodate(ce))
eadb5831
JH
1360 return ce;
1361
883e248b
BP
1362 if (!ignore_fsmonitor)
1363 refresh_fsmonitor(istate);
aa9349d4 1364 /*
56cac48c
NTND
1365 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1366 * that the change to the work tree does not matter and told
1367 * us not to worry.
aa9349d4 1368 */
56cac48c
NTND
1369 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1370 ce_mark_uptodate(ce);
1371 return ce;
1372 }
aa9349d4
MSO
1373 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1374 ce_mark_uptodate(ce);
1375 return ce;
1376 }
883e248b
BP
1377 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1378 ce_mark_uptodate(ce);
1379 return ce;
1380 }
aa9349d4 1381
ccad42d4
RS
1382 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1383 if (ignore_missing)
1384 return ce;
1385 if (err)
1386 *err = ENOENT;
1387 return NULL;
1388 }
1389
8fd2cb40 1390 if (lstat(ce->name, &st) < 0) {
2e2e7ec1
BK
1391 if (ignore_missing && errno == ENOENT)
1392 return ce;
ec0cc704
JH
1393 if (err)
1394 *err = errno;
8fd2cb40
JS
1395 return NULL;
1396 }
405e5b2f 1397
4bd5b7da 1398 changed = ie_match_stat(istate, ce, &st, options);
d05e6970
JK
1399 if (changed_ret)
1400 *changed_ret = changed;
405e5b2f 1401 if (!changed) {
4bd5b7da
JH
1402 /*
1403 * The path is unchanged. If we were told to ignore
1404 * valid bit, then we did the actual stat check and
1405 * found that the entry is unmodified. If the entry
1406 * is not marked VALID, this is the place to mark it
1407 * valid again, under "assume unchanged" mode.
1408 */
1409 if (ignore_valid && assume_unchanged &&
7a51ed66 1410 !(ce->ce_flags & CE_VALID))
405e5b2f 1411 ; /* mark this one VALID again */
eadb5831
JH
1412 else {
1413 /*
1414 * We do not mark the index itself "modified"
1415 * because CE_UPTODATE flag is in-core only;
1416 * we are not going to write this change out.
1417 */
883e248b 1418 if (!S_ISGITLINK(ce->ce_mode)) {
125fd984 1419 ce_mark_uptodate(ce);
883e248b
BP
1420 mark_fsmonitor_valid(ce);
1421 }
8fd2cb40 1422 return ce;
eadb5831 1423 }
405e5b2f
LT
1424 }
1425
4bd5b7da 1426 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1427 if (err)
1428 *err = EINVAL;
8fd2cb40
JS
1429 return NULL;
1430 }
405e5b2f 1431
a849735b 1432 updated = make_empty_cache_entry(istate, ce_namelen(ce));
0e267b7a
BP
1433 copy_cache_entry(updated, ce);
1434 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
405e5b2f 1435 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1436 /*
1437 * If ignore_valid is not set, we should leave CE_VALID bit
1438 * alone. Otherwise, paths marked with --no-assume-unchanged
1439 * (i.e. things to be edited) will reacquire CE_VALID bit
1440 * automatically, which is not really what we want.
405e5b2f 1441 */
4bd5b7da 1442 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1443 !(ce->ce_flags & CE_VALID))
1444 updated->ce_flags &= ~CE_VALID;
405e5b2f 1445
e636a7b4 1446 /* istate->cache_changed is updated in the caller */
405e5b2f
LT
1447 return updated;
1448}
1449
3deffc52 1450static void show_file(const char * fmt, const char * name, int in_porcelain,
046613c5 1451 int * first, const char *header_msg)
3deffc52
MM
1452{
1453 if (in_porcelain && *first && header_msg) {
1454 printf("%s\n", header_msg);
cd2b8ae9 1455 *first = 0;
3deffc52
MM
1456 }
1457 printf(fmt, name);
1458}
1459
9b2d6149
NTND
1460int refresh_index(struct index_state *istate, unsigned int flags,
1461 const struct pathspec *pathspec,
046613c5 1462 char *seen, const char *header_msg)
405e5b2f
LT
1463{
1464 int i;
1465 int has_errors = 0;
1466 int really = (flags & REFRESH_REALLY) != 0;
1467 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1468 int quiet = (flags & REFRESH_QUIET) != 0;
1469 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1470 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1471 int first = 1;
1472 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
25762726
BK
1473 unsigned int options = (CE_MATCH_REFRESH |
1474 (really ? CE_MATCH_IGNORE_VALID : 0) |
2e2e7ec1 1475 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
4bd4e730 1476 const char *modified_fmt;
73b7eae6
JK
1477 const char *deleted_fmt;
1478 const char *typechange_fmt;
1479 const char *added_fmt;
4bd4e730 1480 const char *unmerged_fmt;
405e5b2f 1481
c46c406a 1482 trace_performance_enter();
4bd4e730 1483 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
73b7eae6
JK
1484 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1485 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1486 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
4bd4e730 1487 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
4aab5b46 1488 for (i = 0; i < istate->cache_nr; i++) {
285c2e25 1489 struct cache_entry *ce, *new_entry;
ec0cc704 1490 int cache_errno = 0;
73b7eae6 1491 int changed = 0;
3d1f148c 1492 int filtered = 0;
ec0cc704 1493
4aab5b46 1494 ce = istate->cache[i];
5fdeacb0
JS
1495 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1496 continue;
1497
6d2df284 1498 if (pathspec && !ce_path_match(&the_index, ce, pathspec, seen))
3d1f148c
JH
1499 filtered = 1;
1500
405e5b2f 1501 if (ce_stage(ce)) {
4aab5b46
JH
1502 while ((i < istate->cache_nr) &&
1503 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1504 i++;
1505 i--;
1506 if (allow_unmerged)
1507 continue;
3d1f148c
JH
1508 if (!filtered)
1509 show_file(unmerged_fmt, ce->name, in_porcelain,
1510 &first, header_msg);
405e5b2f
LT
1511 has_errors = 1;
1512 continue;
1513 }
1514
3d1f148c 1515 if (filtered)
d616813d
AJ
1516 continue;
1517
285c2e25
BW
1518 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1519 if (new_entry == ce)
405e5b2f 1520 continue;
285c2e25 1521 if (!new_entry) {
73b7eae6
JK
1522 const char *fmt;
1523
8fd2cb40 1524 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1525 /* If we are doing --really-refresh that
1526 * means the index is not valid anymore.
1527 */
7a51ed66 1528 ce->ce_flags &= ~CE_VALID;
078a58e8 1529 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 1530 mark_fsmonitor_invalid(istate, ce);
e636a7b4 1531 istate->cache_changed |= CE_ENTRY_CHANGED;
405e5b2f
LT
1532 }
1533 if (quiet)
1534 continue;
73b7eae6
JK
1535
1536 if (cache_errno == ENOENT)
1537 fmt = deleted_fmt;
895ff3b2 1538 else if (ce_intent_to_add(ce))
73b7eae6
JK
1539 fmt = added_fmt; /* must be before other checks */
1540 else if (changed & TYPE_CHANGED)
1541 fmt = typechange_fmt;
1542 else
1543 fmt = modified_fmt;
1544 show_file(fmt,
1545 ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1546 has_errors = 1;
1547 continue;
1548 }
cf558704 1549
285c2e25 1550 replace_index_entry(istate, i, new_entry);
405e5b2f 1551 }
c46c406a 1552 trace_performance_leave("refresh index");
405e5b2f
LT
1553 return has_errors;
1554}
1555
768d7965
JM
1556struct cache_entry *refresh_cache_entry(struct index_state *istate,
1557 struct cache_entry *ce,
1558 unsigned int options)
ec0cc704 1559{
768d7965 1560 return refresh_cache_ent(istate, ce, options, NULL, NULL);
ec0cc704
JH
1561}
1562
db3b313c
JH
1563
1564/*****************************************************************
1565 * Index File I/O
1566 *****************************************************************/
1567
9d227781
JH
1568#define INDEX_FORMAT_DEFAULT 3
1569
136347d7
TG
1570static unsigned int get_index_format_default(void)
1571{
1572 char *envversion = getenv("GIT_INDEX_VERSION");
3c09d684 1573 char *endp;
b27a5720 1574 int value;
3c09d684
TG
1575 unsigned int version = INDEX_FORMAT_DEFAULT;
1576
136347d7 1577 if (!envversion) {
b27a5720
TA
1578 if (!git_config_get_int("index.version", &value))
1579 version = value;
3c09d684
TG
1580 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1581 warning(_("index.version set, but the value is invalid.\n"
136347d7 1582 "Using version %i"), INDEX_FORMAT_DEFAULT);
3c09d684 1583 return INDEX_FORMAT_DEFAULT;
136347d7
TG
1584 }
1585 return version;
1586 }
3c09d684
TG
1587
1588 version = strtoul(envversion, &endp, 10);
1589 if (*endp ||
1590 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1591 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1592 "Using version %i"), INDEX_FORMAT_DEFAULT);
1593 version = INDEX_FORMAT_DEFAULT;
1594 }
1595 return version;
136347d7
TG
1596}
1597
db3b313c
JH
1598/*
1599 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1600 * Again - this is just a (very strong in practice) heuristic that
1601 * the inode hasn't changed.
1602 *
1603 * We save the fields in big-endian order to allow using the
1604 * index file over NFS transparently.
1605 */
1606struct ondisk_cache_entry {
1607 struct cache_time ctime;
1608 struct cache_time mtime;
7800c1eb
TG
1609 uint32_t dev;
1610 uint32_t ino;
1611 uint32_t mode;
1612 uint32_t uid;
1613 uint32_t gid;
1614 uint32_t size;
db3b313c 1615 unsigned char sha1[20];
7800c1eb 1616 uint16_t flags;
db3b313c
JH
1617 char name[FLEX_ARRAY]; /* more */
1618};
1619
1620/*
1621 * This struct is used when CE_EXTENDED bit is 1
1622 * The struct must match ondisk_cache_entry exactly from
1623 * ctime till flags
1624 */
1625struct ondisk_cache_entry_extended {
1626 struct cache_time ctime;
1627 struct cache_time mtime;
7800c1eb
TG
1628 uint32_t dev;
1629 uint32_t ino;
1630 uint32_t mode;
1631 uint32_t uid;
1632 uint32_t gid;
1633 uint32_t size;
db3b313c 1634 unsigned char sha1[20];
7800c1eb
TG
1635 uint16_t flags;
1636 uint16_t flags2;
db3b313c
JH
1637 char name[FLEX_ARRAY]; /* more */
1638};
1639
6c9cd161 1640/* These are only used for v3 or lower */
ce012deb 1641#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
db3b313c
JH
1642#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1643#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1644#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1645#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1646 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1647 ondisk_cache_entry_size(ce_namelen(ce)))
1648
a33fc72f
JH
1649/* Allow fsck to force verification of the index checksum. */
1650int verify_index_checksum;
1651
00ec50e5
BP
1652/* Allow fsck to force verification of the cache entry order. */
1653int verify_ce_order;
1654
371ed0de 1655static int verify_hdr(const struct cache_header *hdr, unsigned long size)
e83c5163 1656{
aab61359 1657 git_hash_ctx c;
1658 unsigned char hash[GIT_MAX_RAWSZ];
0136bac9 1659 int hdr_version;
e83c5163 1660
ccc4feb5 1661 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1662 return error("bad signature");
0136bac9 1663 hdr_version = ntohl(hdr->hdr_version);
b82a7b5b 1664 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
0136bac9 1665 return error("bad index version %d", hdr_version);
a33fc72f
JH
1666
1667 if (!verify_index_checksum)
1668 return 0;
1669
aab61359 1670 the_hash_algo->init_fn(&c);
1671 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1672 the_hash_algo->final_fn(hash, &c);
67947c34 1673 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
ca9be054 1674 return error("bad index file sha1 signature");
e83c5163
LT
1675 return 0;
1676}
1677
4aab5b46 1678static int read_index_extension(struct index_state *istate,
371ed0de 1679 const char *ext, const char *data, unsigned long sz)
bad68ec9
JH
1680{
1681 switch (CACHE_EXT(ext)) {
1682 case CACHE_EXT_TREE:
4aab5b46 1683 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1684 break;
cfc5789a
JH
1685 case CACHE_EXT_RESOLVE_UNDO:
1686 istate->resolve_undo = resolve_undo_read(data, sz);
1687 break;
5fc2fc8f
NTND
1688 case CACHE_EXT_LINK:
1689 if (read_link_extension(istate, data, sz))
1690 return -1;
1691 break;
f9e6c649
NTND
1692 case CACHE_EXT_UNTRACKED:
1693 istate->untracked = read_untracked_extension(data, sz);
1694 break;
883e248b
BP
1695 case CACHE_EXT_FSMONITOR:
1696 read_fsmonitor_extension(istate, data, sz);
1697 break;
3b1d9e04
BP
1698 case CACHE_EXT_ENDOFINDEXENTRIES:
1699 /* already handled in do_read_index() */
1700 break;
bad68ec9
JH
1701 default:
1702 if (*ext < 'A' || 'Z' < *ext)
1703 return error("index uses %.4s extension, which we do not understand",
1704 ext);
1705 fprintf(stderr, "ignoring %.4s extension\n", ext);
1706 break;
1707 }
1708 return 0;
1709}
1710
b3e83cc7 1711int hold_locked_index(struct lock_file *lk, int lock_flags)
216aab1e 1712{
b3e83cc7 1713 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
216aab1e
MH
1714}
1715
4aab5b46 1716int read_index(struct index_state *istate)
8fd2cb40 1717{
a125a223 1718 return read_index_from(istate, get_index_file(), get_git_dir());
8fd2cb40
JS
1719}
1720
252d079c 1721static struct cache_entry *create_from_disk(struct index_state *istate,
a849735b 1722 struct ondisk_cache_entry *ondisk,
6c9cd161 1723 unsigned long *ent_size,
252d079c 1724 const struct cache_entry *previous_ce)
7a51ed66 1725{
debed2a6 1726 struct cache_entry *ce;
7fec10b7 1727 size_t len;
06aaaa0b 1728 const char *name;
debed2a6 1729 unsigned int flags;
252d079c
NTND
1730 size_t copy_len;
1731 /*
1732 * Adjacent cache entries tend to share the leading paths, so it makes
1733 * sense to only store the differences in later entries. In the v4
1734 * on-disk format of the index, each on-disk cache entry stores the
1735 * number of bytes to be stripped from the end of the previous name,
1736 * and the bytes to append to the result, to come up with its name.
1737 */
1738 int expand_name_field = istate->version == 4;
7fec10b7 1739
7a51ed66 1740 /* On-disk flags are just 16 bits */
c3d8da57 1741 flags = get_be16(&ondisk->flags);
debed2a6 1742 len = flags & CE_NAMEMASK;
7fec10b7 1743
debed2a6 1744 if (flags & CE_EXTENDED) {
06aaaa0b
NTND
1745 struct ondisk_cache_entry_extended *ondisk2;
1746 int extended_flags;
1747 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
c3d8da57 1748 extended_flags = get_be16(&ondisk2->flags2) << 16;
06aaaa0b
NTND
1749 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1750 if (extended_flags & ~CE_EXTENDED_FLAGS)
1751 die("Unknown index entry format %08x", extended_flags);
debed2a6 1752 flags |= extended_flags;
06aaaa0b
NTND
1753 name = ondisk2->name;
1754 }
1755 else
1756 name = ondisk->name;
1757
252d079c
NTND
1758 if (expand_name_field) {
1759 const unsigned char *cp = (const unsigned char *)name;
1760 size_t strip_len, previous_len;
6c9cd161 1761
252d079c
NTND
1762 previous_len = previous_ce ? previous_ce->ce_namelen : 0;
1763 strip_len = decode_varint(&cp);
1764 if (previous_len < strip_len) {
1765 if (previous_ce)
1766 die(_("malformed name field in the index, near path '%s'"),
1767 previous_ce->name);
1768 else
1769 die(_("malformed name field in the index in the first path"));
1770 }
1771 copy_len = previous_len - strip_len;
1772 name = (const char *)cp;
1773 }
1774
1775 if (len == CE_NAMEMASK) {
1776 len = strlen(name);
1777 if (expand_name_field)
1778 len += copy_len;
1779 }
1780
1781 ce = mem_pool__ce_alloc(istate->ce_mem_pool, len);
1782
1783 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1784 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1785 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1786 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1787 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1788 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1789 ce->ce_mode = get_be32(&ondisk->mode);
1790 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1791 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1792 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1793 ce->ce_flags = flags & ~CE_NAMEMASK;
1794 ce->ce_namelen = len;
1795 ce->index = 0;
1796 hashcpy(ce->oid.hash, ondisk->sha1);
6c9cd161 1797
252d079c
NTND
1798 if (expand_name_field) {
1799 if (copy_len)
1800 memcpy(ce->name, previous_ce->name, copy_len);
1801 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1802 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1803 } else {
1804 memcpy(ce->name, name, len + 1);
1805 *ent_size = ondisk_ce_size(ce);
6c9cd161 1806 }
debed2a6 1807 return ce;
cf558704
LT
1808}
1809
03f15a79 1810static void check_ce_order(struct index_state *istate)
15999d0b 1811{
03f15a79
TG
1812 unsigned int i;
1813
00ec50e5
BP
1814 if (!verify_ce_order)
1815 return;
1816
03f15a79
TG
1817 for (i = 1; i < istate->cache_nr; i++) {
1818 struct cache_entry *ce = istate->cache[i - 1];
1819 struct cache_entry *next_ce = istate->cache[i];
1820 int name_compare = strcmp(ce->name, next_ce->name);
1821
1822 if (0 < name_compare)
1823 die("unordered stage entries in index");
1824 if (!name_compare) {
1825 if (!ce_stage(ce))
1826 die("multiple stage entries for merged file '%s'",
1827 ce->name);
1828 if (ce_stage(ce) > ce_stage(next_ce))
1829 die("unordered stage entries for '%s'",
1830 ce->name);
1831 }
15999d0b
JSP
1832 }
1833}
1834
435ec090
CC
1835static void tweak_untracked_cache(struct index_state *istate)
1836{
1837 switch (git_config_get_untracked_cache()) {
1838 case -1: /* keep: do nothing */
1839 break;
1840 case 0: /* false */
1841 remove_untracked_cache(istate);
1842 break;
1843 case 1: /* true */
1844 add_untracked_cache(istate);
1845 break;
1846 default: /* unknown value: do nothing */
1847 break;
1848 }
1849}
1850
43925312
CC
1851static void tweak_split_index(struct index_state *istate)
1852{
1853 switch (git_config_get_split_index()) {
1854 case -1: /* unset: do nothing */
1855 break;
1856 case 0: /* false */
1857 remove_split_index(istate);
1858 break;
1859 case 1: /* true */
1860 add_split_index(istate);
1861 break;
1862 default: /* unknown value: do nothing */
1863 break;
1864 }
1865}
1866
435ec090
CC
1867static void post_read_index_from(struct index_state *istate)
1868{
1869 check_ce_order(istate);
1870 tweak_untracked_cache(istate);
43925312 1871 tweak_split_index(istate);
883e248b 1872 tweak_fsmonitor(istate);
435ec090
CC
1873}
1874
8e72d675
JM
1875static size_t estimate_cache_size_from_compressed(unsigned int entries)
1876{
1877 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1878}
1879
1880static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1881{
1882 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1883
1884 /*
1885 * Account for potential alignment differences.
1886 */
1887 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1888 return ondisk_size + entries * per_entry;
1889}
1890
3b1d9e04
BP
1891static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1892static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1893
abb4bb83
BP
1894struct load_index_extensions
1895{
1896#ifndef NO_PTHREADS
1897 pthread_t pthread;
1898#endif
1899 struct index_state *istate;
1900 const char *mmap;
1901 size_t mmap_size;
1902 unsigned long src_offset;
1903};
1904
1905static void *load_index_extensions(void *_data)
1906{
1907 struct load_index_extensions *p = _data;
1908 unsigned long src_offset = p->src_offset;
1909
1910 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1911 /* After an array of active_nr index entries,
1912 * there can be arbitrary number of extended
1913 * sections, each of which is prefixed with
1914 * extension name (4-byte) and section length
1915 * in 4-byte network byte order.
1916 */
1917 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1918 if (read_index_extension(p->istate,
1919 p->mmap + src_offset,
1920 p->mmap + src_offset + 8,
1921 extsize) < 0) {
1922 munmap((void *)p->mmap, p->mmap_size);
1923 die(_("index file corrupt"));
1924 }
1925 src_offset += 8;
1926 src_offset += extsize;
1927 }
1928
1929 return NULL;
1930}
1931
8fd2cb40 1932/* remember to discard_cache() before reading a different cache! */
3e52f70b 1933int do_read_index(struct index_state *istate, const char *path, int must_exist)
e83c5163
LT
1934{
1935 int fd, i;
1936 struct stat st;
debed2a6 1937 unsigned long src_offset;
371ed0de
BP
1938 const struct cache_header *hdr;
1939 const char *mmap;
7a51ed66 1940 size_t mmap_size;
252d079c 1941 const struct cache_entry *previous_ce = NULL;
abb4bb83
BP
1942 struct load_index_extensions p;
1943 size_t extension_offset = 0;
1944#ifndef NO_PTHREADS
1945 int nr_threads;
1946#endif
e83c5163 1947
913e0e99 1948 if (istate->initialized)
4aab5b46 1949 return istate->cache_nr;
5d1a5c02 1950
fba2f38a
KB
1951 istate->timestamp.sec = 0;
1952 istate->timestamp.nsec = 0;
8fd2cb40 1953 fd = open(path, O_RDONLY);
5d1a5c02 1954 if (fd < 0) {
5fc2fc8f 1955 if (!must_exist && errno == ENOENT)
5d1a5c02 1956 return 0;
5fc2fc8f 1957 die_errno("%s: index file open failed", path);
5d1a5c02 1958 }
e83c5163 1959
3511a377 1960 if (fstat(fd, &st))
d824cbba 1961 die_errno("cannot stat the open index");
3511a377 1962
7a51ed66 1963 mmap_size = xsize_t(st.st_size);
aab61359 1964 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
3511a377
LFC
1965 die("index file smaller than expected");
1966
a1293ef7 1967 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
7a51ed66 1968 if (mmap == MAP_FAILED)
0721c314 1969 die_errno("unable to map index file");
57d84f8d 1970 close(fd);
e83c5163 1971
371ed0de 1972 hdr = (const struct cache_header *)mmap;
7a51ed66 1973 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1974 goto unmap;
1975
75691ea3 1976 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
9d227781 1977 istate->version = ntohl(hdr->hdr_version);
4aab5b46
JH
1978 istate->cache_nr = ntohl(hdr->hdr_entries);
1979 istate->cache_alloc = alloc_nr(istate->cache_nr);
c4aa3167 1980 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
913e0e99 1981 istate->initialized = 1;
7a51ed66 1982
abb4bb83
BP
1983 p.istate = istate;
1984 p.mmap = mmap;
1985 p.mmap_size = mmap_size;
1986
1987#ifndef NO_PTHREADS
1988 nr_threads = git_config_get_index_threads();
1989 if (!nr_threads)
1990 nr_threads = online_cpus();
1991
1992 if (nr_threads > 1) {
1993 extension_offset = read_eoie_extension(mmap, mmap_size);
1994 if (extension_offset) {
1995 int err;
1996
1997 p.src_offset = extension_offset;
1998 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
1999 if (err)
2000 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2001
2002 nr_threads--;
2003 }
2004 }
2005#endif
2006
8e72d675 2007 if (istate->version == 4) {
8e72d675
JM
2008 mem_pool_init(&istate->ce_mem_pool,
2009 estimate_cache_size_from_compressed(istate->cache_nr));
2010 } else {
8e72d675
JM
2011 mem_pool_init(&istate->ce_mem_pool,
2012 estimate_cache_size(mmap_size, istate->cache_nr));
2013 }
6c9cd161 2014
7a51ed66 2015 src_offset = sizeof(*hdr);
4aab5b46 2016 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 2017 struct ondisk_cache_entry *disk_ce;
4aab5b46 2018 struct cache_entry *ce;
936f53d0 2019 unsigned long consumed;
4aab5b46 2020
371ed0de 2021 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
252d079c 2022 ce = create_from_disk(istate, disk_ce, &consumed, previous_ce);
cf558704 2023 set_index_entry(istate, i, ce);
7a51ed66 2024
936f53d0 2025 src_offset += consumed;
252d079c 2026 previous_ce = ce;
e83c5163 2027 }
fba2f38a 2028 istate->timestamp.sec = st.st_mtime;
c06ff490 2029 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 2030
abb4bb83
BP
2031 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2032#ifndef NO_PTHREADS
2033 if (extension_offset) {
2034 int ret = pthread_join(p.pthread, NULL);
2035 if (ret)
2036 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2037 }
2038#endif
2039 if (!extension_offset) {
2040 p.src_offset = src_offset;
2041 load_index_extensions(&p);
bad68ec9 2042 }
371ed0de 2043 munmap((void *)mmap, mmap_size);
4aab5b46 2044 return istate->cache_nr;
e83c5163
LT
2045
2046unmap:
371ed0de 2047 munmap((void *)mmap, mmap_size);
5d1a5c02 2048 die("index file corrupt");
e83c5163
LT
2049}
2050
0d59ffb4
CC
2051/*
2052 * Signal that the shared index is used by updating its mtime.
2053 *
2054 * This way, shared index can be removed if they have not been used
2055 * for some time.
2056 */
a125a223 2057static void freshen_shared_index(const char *shared_index, int warn)
0d59ffb4 2058{
0d59ffb4
CC
2059 if (!check_and_freshen_file(shared_index, 1) && warn)
2060 warning("could not freshen shared index '%s'", shared_index);
2061}
2062
a125a223
TG
2063int read_index_from(struct index_state *istate, const char *path,
2064 const char *gitdir)
5fc2fc8f
NTND
2065{
2066 struct split_index *split_index;
2067 int ret;
2182abd9 2068 char *base_oid_hex;
a125a223 2069 char *base_path;
5fc2fc8f
NTND
2070
2071 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2072 if (istate->initialized)
2073 return istate->cache_nr;
2074
c46c406a 2075 trace_performance_enter();
5fc2fc8f 2076 ret = do_read_index(istate, path, 0);
c46c406a 2077 trace_performance_leave("read cache %s", path);
435ec090 2078
5fc2fc8f 2079 split_index = istate->split_index;
2182abd9 2080 if (!split_index || is_null_oid(&split_index->base_oid)) {
435ec090 2081 post_read_index_from(istate);
5fc2fc8f 2082 return ret;
03f15a79 2083 }
5fc2fc8f 2084
c46c406a 2085 trace_performance_enter();
5fc2fc8f
NTND
2086 if (split_index->base)
2087 discard_index(split_index->base);
2088 else
2089 split_index->base = xcalloc(1, sizeof(*split_index->base));
de6ae5f9 2090
2182abd9 2091 base_oid_hex = oid_to_hex(&split_index->base_oid);
2092 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
de6ae5f9 2093 ret = do_read_index(split_index->base, base_path, 1);
9001dc2a 2094 if (!oideq(&split_index->base_oid, &split_index->base->oid))
5fc2fc8f 2095 die("broken index, expect %s in %s, got %s",
2182abd9 2096 base_oid_hex, base_path,
75691ea3 2097 oid_to_hex(&split_index->base->oid));
de6ae5f9 2098
a125a223 2099 freshen_shared_index(base_path, 0);
5fc2fc8f 2100 merge_base_index(istate);
435ec090 2101 post_read_index_from(istate);
a125a223 2102 free(base_path);
c46c406a 2103 trace_performance_leave("read cache %s", base_path);
5fc2fc8f
NTND
2104 return ret;
2105}
2106
fa7b3c2f
JH
2107int is_index_unborn(struct index_state *istate)
2108{
debed2a6 2109 return (!istate->cache_nr && !istate->timestamp.sec);
fa7b3c2f
JH
2110}
2111
4aab5b46 2112int discard_index(struct index_state *istate)
6d297f81 2113{
8e72d675
JM
2114 /*
2115 * Cache entries in istate->cache[] should have been allocated
2116 * from the memory pool associated with this index, or from an
2117 * associated split_index. There is no need to free individual
8616a2d0
JM
2118 * cache entries. validate_cache_entries can detect when this
2119 * assertion does not hold.
8e72d675 2120 */
8616a2d0 2121 validate_cache_entries(istate);
debed2a6 2122
cfc5789a 2123 resolve_undo_clear_index(istate);
4aab5b46
JH
2124 istate->cache_nr = 0;
2125 istate->cache_changed = 0;
fba2f38a
KB
2126 istate->timestamp.sec = 0;
2127 istate->timestamp.nsec = 0;
2092678c 2128 free_name_hash(istate);
4aab5b46 2129 cache_tree_free(&(istate->cache_tree));
913e0e99 2130 istate->initialized = 0;
6a83d902 2131 FREE_AND_NULL(istate->cache);
a0fc4db0 2132 istate->cache_alloc = 0;
5fc2fc8f 2133 discard_split_index(istate);
f9e6c649
NTND
2134 free_untracked_cache(istate->untracked);
2135 istate->untracked = NULL;
8e72d675
JM
2136
2137 if (istate->ce_mem_pool) {
8616a2d0 2138 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
8e72d675
JM
2139 istate->ce_mem_pool = NULL;
2140 }
2141
7a51ed66 2142 return 0;
6d297f81
JS
2143}
2144
8616a2d0
JM
2145/*
2146 * Validate the cache entries of this index.
2147 * All cache entries associated with this index
2148 * should have been allocated by the memory pool
2149 * associated with this index, or by a referenced
2150 * split index.
2151 */
2152void validate_cache_entries(const struct index_state *istate)
2153{
2154 int i;
2155
2156 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2157 return;
2158
2159 for (i = 0; i < istate->cache_nr; i++) {
2160 if (!istate) {
2161 die("internal error: cache entry is not allocated from expected memory pool");
2162 } else if (!istate->ce_mem_pool ||
2163 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2164 if (!istate->split_index ||
2165 !istate->split_index->base ||
2166 !istate->split_index->base->ce_mem_pool ||
2167 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2168 die("internal error: cache entry is not allocated from expected memory pool");
2169 }
2170 }
2171 }
2172
2173 if (istate->split_index)
2174 validate_cache_entries(istate->split_index->base);
2175}
2176
d1f128b0 2177int unmerged_index(const struct index_state *istate)
94a5728c
DB
2178{
2179 int i;
2180 for (i = 0; i < istate->cache_nr; i++) {
2181 if (ce_stage(istate->cache[i]))
2182 return 1;
2183 }
2184 return 0;
2185}
2186
e1f8694f
EN
2187int index_has_changes(const struct index_state *istate,
2188 struct tree *tree,
2189 struct strbuf *sb)
cffbfad5 2190{
e1f8694f 2191 struct object_id cmp;
cffbfad5
EN
2192 int i;
2193
1b9fbefb
EN
2194 if (istate != &the_index) {
2195 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2196 }
e1f8694f
EN
2197 if (tree)
2198 cmp = tree->object.oid;
2199 if (tree || !get_oid_tree("HEAD", &cmp)) {
cffbfad5
EN
2200 struct diff_options opt;
2201
2202 diff_setup(&opt);
2203 opt.flags.exit_with_status = 1;
2204 if (!sb)
2205 opt.flags.quick = 1;
e1f8694f 2206 do_diff_cache(&cmp, &opt);
cffbfad5
EN
2207 diffcore_std(&opt);
2208 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2209 if (i)
2210 strbuf_addch(sb, ' ');
2211 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2212 }
2213 diff_flush(&opt);
2214 return opt.flags.has_changes != 0;
2215 } else {
1b9fbefb 2216 for (i = 0; sb && i < istate->cache_nr; i++) {
cffbfad5
EN
2217 if (i)
2218 strbuf_addch(sb, ' ');
1b9fbefb 2219 strbuf_addstr(sb, istate->cache[i]->name);
cffbfad5 2220 }
1b9fbefb 2221 return !!istate->cache_nr;
cffbfad5
EN
2222 }
2223}
2224
4990aadc 2225#define WRITE_BUFFER_SIZE 8192
bf0f910d 2226static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
2227static unsigned long write_buffer_len;
2228
aab61359 2229static int ce_write_flush(git_hash_ctx *context, int fd)
6015c28b
JH
2230{
2231 unsigned int buffered = write_buffer_len;
2232 if (buffered) {
aab61359 2233 the_hash_algo->update_fn(context, write_buffer, buffered);
06f46f23 2234 if (write_in_full(fd, write_buffer, buffered) < 0)
6015c28b
JH
2235 return -1;
2236 write_buffer_len = 0;
2237 }
2238 return 0;
2239}
2240
aab61359 2241static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
4990aadc
LT
2242{
2243 while (len) {
2244 unsigned int buffered = write_buffer_len;
2245 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2246 if (partial > len)
2247 partial = len;
2248 memcpy(write_buffer + buffered, data, partial);
2249 buffered += partial;
2250 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
2251 write_buffer_len = buffered;
2252 if (ce_write_flush(context, fd))
4990aadc
LT
2253 return -1;
2254 buffered = 0;
2255 }
2256 write_buffer_len = buffered;
2257 len -= partial;
1d7f171c 2258 data = (char *) data + partial;
a6080a0a
JH
2259 }
2260 return 0;
4990aadc
LT
2261}
2262
3b1d9e04
BP
2263static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2264 int fd, unsigned int ext, unsigned int sz)
bad68ec9
JH
2265{
2266 ext = htonl(ext);
2267 sz = htonl(sz);
3b1d9e04
BP
2268 if (eoie_context) {
2269 the_hash_algo->update_fn(eoie_context, &ext, 4);
2270 the_hash_algo->update_fn(eoie_context, &sz, 4);
2271 }
968a1d65
DR
2272 return ((ce_write(context, fd, &ext, 4) < 0) ||
2273 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
2274}
2275
aab61359 2276static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
4990aadc
LT
2277{
2278 unsigned int left = write_buffer_len;
ca9be054 2279
4990aadc
LT
2280 if (left) {
2281 write_buffer_len = 0;
aab61359 2282 the_hash_algo->update_fn(context, write_buffer, left);
4990aadc 2283 }
ca9be054 2284
aab61359 2285 /* Flush first if not enough space for hash signature */
2286 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
06f46f23 2287 if (write_in_full(fd, write_buffer, left) < 0)
2c865d9a
QH
2288 return -1;
2289 left = 0;
2290 }
2291
aab61359 2292 /* Append the hash signature at the end */
2293 the_hash_algo->final_fn(write_buffer + left, context);
2294 hashcpy(hash, write_buffer + left);
2295 left += the_hash_algo->rawsz;
06f46f23 2296 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
4990aadc
LT
2297}
2298
407c8eb0
JH
2299static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2300{
2301 /*
2302 * The only thing we care about in this function is to smudge the
2303 * falsely clean entry due to touch-update-touch race, so we leave
2304 * everything else as they are. We are called for entries whose
c21d39d7 2305 * ce_stat_data.sd_mtime match the index file mtime.
c70115b4
JH
2306 *
2307 * Note that this actually does not do much for gitlinks, for
2308 * which ce_match_stat_basic() always goes to the actual
2309 * contents. The caller checks with is_racy_timestamp() which
2310 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
2311 */
2312 struct stat st;
2313
2314 if (lstat(ce->name, &st) < 0)
2315 return;
2316 if (ce_match_stat_basic(ce, &st))
2317 return;
2318 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
2319 /* This is "racily clean"; smudge it. Note that this
2320 * is a tricky code. At first glance, it may appear
2321 * that it can break with this sequence:
2322 *
2323 * $ echo xyzzy >frotz
2324 * $ git-update-index --add frotz
2325 * $ : >frotz
2326 * $ sleep 3
2327 * $ echo filfre >nitfol
2328 * $ git-update-index --add nitfol
2329 *
b7e58b17 2330 * but it does not. When the second update-index runs,
4b3511b0
JH
2331 * it notices that the entry "frotz" has the same timestamp
2332 * as index, and if we were to smudge it by resetting its
2333 * size to zero here, then the object name recorded
2334 * in index is the 6-byte file but the cached stat information
2335 * becomes zero --- which would then match what we would
a6080a0a 2336 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
2337 *
2338 * However, the second update-index, before calling
2339 * this function, notices that the cached size is 6
2340 * bytes and what is on the filesystem is an empty
2341 * file, and never calls us, so the cached size information
2342 * for "frotz" stays 6 which does not match the filesystem.
2343 */
c21d39d7 2344 ce->ce_stat_data.sd_size = 0;
407c8eb0
JH
2345 }
2346}
2347
f136f7bf 2348/* Copy miscellaneous fields but not the name */
ce012deb 2349static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
f136f7bf 2350 struct cache_entry *ce)
7a51ed66 2351{
b60e188c
TG
2352 short flags;
2353
c21d39d7
MH
2354 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2355 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2356 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2357 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2358 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2359 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
7a51ed66 2360 ondisk->mode = htonl(ce->ce_mode);
c21d39d7
MH
2361 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2362 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2363 ondisk->size = htonl(ce->ce_stat_data.sd_size);
99d1a986 2364 hashcpy(ondisk->sha1, ce->oid.hash);
b60e188c 2365
ce51bf09 2366 flags = ce->ce_flags & ~CE_NAMEMASK;
b60e188c
TG
2367 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2368 ondisk->flags = htons(flags);
06aaaa0b
NTND
2369 if (ce->ce_flags & CE_EXTENDED) {
2370 struct ondisk_cache_entry_extended *ondisk2;
2371 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2372 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
f136f7bf
JH
2373 }
2374}
2375
aab61359 2376static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
ce012deb 2377 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
f136f7bf 2378{
9d227781 2379 int size;
f136f7bf 2380 int result;
00a4b035
RJ
2381 unsigned int saved_namelen;
2382 int stripped_name = 0;
ce012deb 2383 static unsigned char padding[8] = { 0x00 };
f136f7bf 2384
b3c96fb1
NTND
2385 if (ce->ce_flags & CE_STRIP_NAME) {
2386 saved_namelen = ce_namelen(ce);
2387 ce->ce_namelen = 0;
00a4b035 2388 stripped_name = 1;
b3c96fb1
NTND
2389 }
2390
ce012deb
KW
2391 if (ce->ce_flags & CE_EXTENDED)
2392 size = offsetof(struct ondisk_cache_entry_extended, name);
2393 else
2394 size = offsetof(struct ondisk_cache_entry, name);
2395
9d227781 2396 if (!previous_name) {
ce012deb
KW
2397 int len = ce_namelen(ce);
2398 copy_cache_entry_to_ondisk(ondisk, ce);
2399 result = ce_write(c, fd, ondisk, size);
2400 if (!result)
2401 result = ce_write(c, fd, ce->name, len);
2402 if (!result)
2403 result = ce_write(c, fd, padding, align_padding_size(size, len));
9d227781
JH
2404 } else {
2405 int common, to_remove, prefix_size;
2406 unsigned char to_remove_vi[16];
2407 for (common = 0;
2408 (ce->name[common] &&
2409 common < previous_name->len &&
2410 ce->name[common] == previous_name->buf[common]);
2411 common++)
2412 ; /* still matching */
2413 to_remove = previous_name->len - common;
2414 prefix_size = encode_varint(to_remove, to_remove_vi);
2415
ce012deb
KW
2416 copy_cache_entry_to_ondisk(ondisk, ce);
2417 result = ce_write(c, fd, ondisk, size);
2418 if (!result)
2419 result = ce_write(c, fd, to_remove_vi, prefix_size);
2420 if (!result)
0b90b881
TG
2421 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2422 if (!result)
2423 result = ce_write(c, fd, padding, 1);
9d227781
JH
2424
2425 strbuf_splice(previous_name, common, to_remove,
2426 ce->name + common, ce_namelen(ce) - common);
06aaaa0b 2427 }
00a4b035 2428 if (stripped_name) {
b3c96fb1
NTND
2429 ce->ce_namelen = saved_namelen;
2430 ce->ce_flags &= ~CE_STRIP_NAME;
2431 }
7a51ed66 2432
59efba64 2433 return result;
7a51ed66
LT
2434}
2435
426ddeea
YM
2436/*
2437 * This function verifies if index_state has the correct sha1 of the
2438 * index file. Don't die if we have any other failure, just return 0.
2439 */
2440static int verify_index_from(const struct index_state *istate, const char *path)
2441{
2442 int fd;
2443 ssize_t n;
2444 struct stat st;
aab61359 2445 unsigned char hash[GIT_MAX_RAWSZ];
426ddeea
YM
2446
2447 if (!istate->initialized)
2448 return 0;
2449
2450 fd = open(path, O_RDONLY);
2451 if (fd < 0)
2452 return 0;
2453
2454 if (fstat(fd, &st))
2455 goto out;
2456
aab61359 2457 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
426ddeea
YM
2458 goto out;
2459
aab61359 2460 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2461 if (n != the_hash_algo->rawsz)
426ddeea
YM
2462 goto out;
2463
67947c34 2464 if (!hasheq(istate->oid.hash, hash))
426ddeea
YM
2465 goto out;
2466
2467 close(fd);
2468 return 1;
2469
2470out:
2471 close(fd);
2472 return 0;
2473}
2474
2475static int verify_index(const struct index_state *istate)
2476{
2477 return verify_index_from(istate, get_index_file());
2478}
2479
483fbe2b
JH
2480static int has_racy_timestamp(struct index_state *istate)
2481{
2482 int entries = istate->cache_nr;
2483 int i;
2484
2485 for (i = 0; i < entries; i++) {
2486 struct cache_entry *ce = istate->cache[i];
2487 if (is_racy_timestamp(istate, ce))
2488 return 1;
2489 }
2490 return 0;
2491}
2492
ccdc4ec3
JH
2493void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2494{
483fbe2b 2495 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
df60cf57
2496 verify_index(istate))
2497 write_locked_index(istate, lockfile, COMMIT_LOCK);
b74c90fb
2498 else
2499 rollback_lock_file(lockfile);
ccdc4ec3
JH
2500}
2501
812d6b00
2502/*
2503 * On success, `tempfile` is closed. If it is the temporary file
2504 * of a `struct lock_file`, we will therefore effectively perform
2505 * a 'close_lock_file_gently()`. Since that is an implementation
2506 * detail of lockfiles, callers of `do_write_index()` should not
2507 * rely on it.
2508 */
9f41c7a6 2509static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
c18b80a0 2510 int strip_extensions)
197ee8c9 2511{
ca54d9ba 2512 uint64_t start = getnanotime();
9f41c7a6 2513 int newfd = tempfile->fd;
3b1d9e04 2514 git_hash_ctx c, eoie_c;
197ee8c9 2515 struct cache_header hdr;
b50386c7 2516 int i, err = 0, removed, extended, hdr_version;
4aab5b46
JH
2517 struct cache_entry **cache = istate->cache;
2518 int entries = istate->cache_nr;
e1afca4f 2519 struct stat st;
ce012deb 2520 struct ondisk_cache_entry_extended ondisk;
9d227781 2521 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
4bddd983 2522 int drop_cache_tree = istate->drop_cache_tree;
3b1d9e04 2523 off_t offset;
025a0709 2524
06aaaa0b 2525 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 2526 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 2527 removed++;
197ee8c9 2528
06aaaa0b
NTND
2529 /* reduce extended entries if possible */
2530 cache[i]->ce_flags &= ~CE_EXTENDED;
2531 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2532 extended++;
2533 cache[i]->ce_flags |= CE_EXTENDED;
2534 }
2535 }
2536
d6e3c181 2537 if (!istate->version) {
136347d7 2538 istate->version = get_index_format_default();
4c2db938 2539 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
d6e3c181
NTND
2540 init_split_index(istate);
2541 }
9d227781
JH
2542
2543 /* demote version 3 to version 2 when the latter suffices */
2544 if (istate->version == 3 || istate->version == 2)
2545 istate->version = extended ? 3 : 2;
2546
2547 hdr_version = istate->version;
2548
ccc4feb5 2549 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
9d227781 2550 hdr.hdr_version = htonl(hdr_version);
025a0709 2551 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 2552
aab61359 2553 the_hash_algo->init_fn(&c);
ca9be054 2554 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
2555 return -1;
2556
3b1d9e04
BP
2557 offset = lseek(newfd, 0, SEEK_CUR);
2558 if (offset < 0)
2559 return -1;
2560 offset += write_buffer_len;
9d227781 2561 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
ce012deb 2562
197ee8c9
LT
2563 for (i = 0; i < entries; i++) {
2564 struct cache_entry *ce = cache[i];
7a51ed66 2565 if (ce->ce_flags & CE_REMOVE)
aa16021e 2566 continue;
e06c43c7 2567 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 2568 ce_smudge_racily_clean_entry(ce);
99d1a986 2569 if (is_null_oid(&ce->oid)) {
83bd7437
JK
2570 static const char msg[] = "cache entry has null sha1: %s";
2571 static int allow = -1;
2572
2573 if (allow < 0)
2574 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2575 if (allow)
2576 warning(msg, ce->name);
2577 else
b50386c7 2578 err = error(msg, ce->name);
a96d3cc3
JK
2579
2580 drop_cache_tree = 1;
83bd7437 2581 }
ce012deb 2582 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
b50386c7
KW
2583 err = -1;
2584
2585 if (err)
2586 break;
197ee8c9 2587 }
9d227781 2588 strbuf_release(&previous_name_buf);
1af1c2b6 2589
b50386c7
KW
2590 if (err)
2591 return err;
2592
bad68ec9 2593 /* Write extension data here */
3b1d9e04
BP
2594 offset = lseek(newfd, 0, SEEK_CUR);
2595 if (offset < 0)
2596 return -1;
2597 offset += write_buffer_len;
2598 the_hash_algo->init_fn(&eoie_c);
2599
c18b80a0 2600 if (!strip_extensions && istate->split_index) {
5fc2fc8f
NTND
2601 struct strbuf sb = STRBUF_INIT;
2602
2603 err = write_link_extension(&sb, istate) < 0 ||
3b1d9e04 2604 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
5fc2fc8f
NTND
2605 sb.len) < 0 ||
2606 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2607 strbuf_release(&sb);
2608 if (err)
2609 return -1;
2610 }
a96d3cc3 2611 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
f285a2d7 2612 struct strbuf sb = STRBUF_INIT;
1dffb8fa 2613
1dffb8fa 2614 cache_tree_write(&sb, istate->cache_tree);
3b1d9e04 2615 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
1dffb8fa
PH
2616 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2617 strbuf_release(&sb);
2618 if (err)
bad68ec9 2619 return -1;
bad68ec9 2620 }
c18b80a0 2621 if (!strip_extensions && istate->resolve_undo) {
cfc5789a
JH
2622 struct strbuf sb = STRBUF_INIT;
2623
2624 resolve_undo_write(&sb, istate->resolve_undo);
3b1d9e04 2625 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
cfc5789a
JH
2626 sb.len) < 0
2627 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2628 strbuf_release(&sb);
2629 if (err)
2630 return -1;
2631 }
83c094ad
NTND
2632 if (!strip_extensions && istate->untracked) {
2633 struct strbuf sb = STRBUF_INIT;
2634
2635 write_untracked_extension(&sb, istate->untracked);
3b1d9e04 2636 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
83c094ad
NTND
2637 sb.len) < 0 ||
2638 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2639 strbuf_release(&sb);
2640 if (err)
2641 return -1;
2642 }
883e248b
BP
2643 if (!strip_extensions && istate->fsmonitor_last_update) {
2644 struct strbuf sb = STRBUF_INIT;
2645
2646 write_fsmonitor_extension(&sb, istate);
3b1d9e04
BP
2647 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2648 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2649 strbuf_release(&sb);
2650 if (err)
2651 return -1;
2652 }
2653
2654 /*
2655 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2656 * so that it can be found and processed before all the index entries are
2657 * read. Write it out regardless of the strip_extensions parameter as we need it
2658 * when loading the shared index.
2659 */
2660 if (offset) {
2661 struct strbuf sb = STRBUF_INIT;
2662
2663 write_eoie_extension(&sb, &eoie_c, offset);
2664 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
883e248b
BP
2665 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2666 strbuf_release(&sb);
2667 if (err)
2668 return -1;
2669 }
e1afca4f 2670
75691ea3 2671 if (ce_flush(&c, newfd, istate->oid.hash))
9f41c7a6 2672 return -1;
49bd0fc2
JK
2673 if (close_tempfile_gently(tempfile)) {
2674 error(_("could not close '%s'"), tempfile->filename.buf);
49bd0fc2
JK
2675 return -1;
2676 }
9f41c7a6 2677 if (stat(tempfile->filename.buf, &st))
e1afca4f 2678 return -1;
5bcf109c
KB
2679 istate->timestamp.sec = (unsigned int)st.st_mtime;
2680 istate->timestamp.nsec = ST_MTIME_NSEC(st);
ca54d9ba 2681 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
e1afca4f 2682 return 0;
197ee8c9 2683}
e46bbcf6 2684
626f35c8
NTND
2685void set_alternate_index_output(const char *name)
2686{
2687 alternate_index_output = name;
2688}
2689
2690static int commit_locked_index(struct lock_file *lk)
2691{
751baced
MH
2692 if (alternate_index_output)
2693 return commit_lock_file_to(lk, alternate_index_output);
2694 else
626f35c8 2695 return commit_lock_file(lk);
626f35c8
NTND
2696}
2697
03b86647
NTND
2698static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2699 unsigned flags)
2700{
076aa2cb 2701 int ret = do_write_index(istate, lock->tempfile, 0);
03b86647
NTND
2702 if (ret)
2703 return ret;
03b86647
NTND
2704 if (flags & COMMIT_LOCK)
2705 return commit_locked_index(lock);
812d6b00 2706 return close_lock_file_gently(lock);
03b86647
NTND
2707}
2708
5fc2fc8f
NTND
2709static int write_split_index(struct index_state *istate,
2710 struct lock_file *lock,
2711 unsigned flags)
2712{
2713 int ret;
2714 prepare_to_write_split_index(istate);
2715 ret = do_write_locked_index(istate, lock, flags);
2716 finish_writing_split_index(istate);
2717 return ret;
2718}
2719
b9683722
CC
2720static const char *shared_index_expire = "2.weeks.ago";
2721
2722static unsigned long get_shared_index_expire_date(void)
2723{
2724 static unsigned long shared_index_expire_date;
2725 static int shared_index_expire_date_prepared;
2726
2727 if (!shared_index_expire_date_prepared) {
2728 git_config_get_expiry("splitindex.sharedindexexpire",
2729 &shared_index_expire);
2730 shared_index_expire_date = approxidate(shared_index_expire);
2731 shared_index_expire_date_prepared = 1;
2732 }
2733
2734 return shared_index_expire_date;
2735}
2736
2737static int should_delete_shared_index(const char *shared_index_path)
2738{
2739 struct stat st;
2740 unsigned long expiration;
2741
2742 /* Check timestamp */
2743 expiration = get_shared_index_expire_date();
2744 if (!expiration)
2745 return 0;
2746 if (stat(shared_index_path, &st))
78bde923 2747 return error_errno(_("could not stat '%s'"), shared_index_path);
b9683722
CC
2748 if (st.st_mtime > expiration)
2749 return 0;
2750
2751 return 1;
2752}
2753
2754static int clean_shared_index_files(const char *current_hex)
2755{
2756 struct dirent *de;
2757 DIR *dir = opendir(get_git_dir());
2758
2759 if (!dir)
2760 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2761
2762 while ((de = readdir(dir)) != NULL) {
2763 const char *sha1_hex;
2764 const char *shared_index_path;
2765 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2766 continue;
2767 if (!strcmp(sha1_hex, current_hex))
2768 continue;
2769 shared_index_path = git_path("%s", de->d_name);
2770 if (should_delete_shared_index(shared_index_path) > 0 &&
2771 unlink(shared_index_path))
2772 warning_errno(_("unable to unlink: %s"), shared_index_path);
2773 }
2774 closedir(dir);
2775
2776 return 0;
2777}
2778
a0a96756 2779static int write_shared_index(struct index_state *istate,
59f9d2dd 2780 struct tempfile **temp)
c18b80a0
NTND
2781{
2782 struct split_index *si = istate->split_index;
076aa2cb 2783 int ret;
c18b80a0 2784
c18b80a0 2785 move_cache_to_base_index(istate);
7db2d08c 2786 ret = do_write_index(si->base, *temp, 1);
59f9d2dd 2787 if (ret)
c18b80a0 2788 return ret;
7db2d08c 2789 ret = adjust_shared_perm(get_tempfile_path(*temp));
df801f3f 2790 if (ret) {
7db2d08c 2791 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
df801f3f
CC
2792 return ret;
2793 }
7db2d08c 2794 ret = rename_tempfile(temp,
75691ea3 2795 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
b9683722 2796 if (!ret) {
75691ea3 2797 oidcpy(&si->base_oid, &si->base->oid);
2798 clean_shared_index_files(oid_to_hex(&si->base->oid));
b9683722
CC
2799 }
2800
c18b80a0
NTND
2801 return ret;
2802}
2803
e6a1dd77
CC
2804static const int default_max_percent_split_change = 20;
2805
2806static int too_many_not_shared_entries(struct index_state *istate)
2807{
2808 int i, not_shared = 0;
2809 int max_split = git_config_get_max_percent_split_change();
2810
2811 switch (max_split) {
2812 case -1:
2813 /* not or badly configured: use the default value */
2814 max_split = default_max_percent_split_change;
2815 break;
2816 case 0:
2817 return 1; /* 0% means always write a new shared index */
2818 case 100:
2819 return 0; /* 100% means never write a new shared index */
2820 default:
2821 break; /* just use the configured value */
2822 }
2823
2824 /* Count not shared entries */
2825 for (i = 0; i < istate->cache_nr; i++) {
2826 struct cache_entry *ce = istate->cache[i];
2827 if (!ce->index)
2828 not_shared++;
2829 }
2830
2831 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2832}
2833
03b86647
NTND
2834int write_locked_index(struct index_state *istate, struct lock_file *lock,
2835 unsigned flags)
2836{
0d59ffb4 2837 int new_shared_index, ret;
5fc2fc8f
NTND
2838 struct split_index *si = istate->split_index;
2839
4592e608
NTND
2840 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
2841 cache_tree_verify(istate);
2842
61000814
2843 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
2844 if (flags & COMMIT_LOCK)
2845 rollback_lock_file(lock);
2846 return 0;
2847 }
2848
3bd28eb2
AV
2849 if (istate->fsmonitor_last_update)
2850 fill_fsmonitor_bitmap(istate);
2851
5165dd59
NTND
2852 if (!si || alternate_index_output ||
2853 (istate->cache_changed & ~EXTMASK)) {
5fc2fc8f 2854 if (si)
2182abd9 2855 oidclr(&si->base_oid);
df60cf57
2856 ret = do_write_locked_index(istate, lock, flags);
2857 goto out;
5fc2fc8f
NTND
2858 }
2859
4c2db938 2860 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2182abd9 2861 int v = si->base_oid.hash[0];
d6e3c181
NTND
2862 if ((v & 15) < 6)
2863 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2864 }
e6a1dd77
CC
2865 if (too_many_not_shared_entries(istate))
2866 istate->cache_changed |= SPLIT_INDEX_ORDERED;
0d59ffb4
CC
2867
2868 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2869
2870 if (new_shared_index) {
59f9d2dd
NTND
2871 struct tempfile *temp;
2872 int saved_errno;
2873
2874 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2875 if (!temp) {
2182abd9 2876 oidclr(&si->base_oid);
59f9d2dd 2877 ret = do_write_locked_index(istate, lock, flags);
ef5b3a6c
NTND
2878 goto out;
2879 }
2880 ret = write_shared_index(istate, &temp);
59f9d2dd
NTND
2881
2882 saved_errno = errno;
2883 if (is_tempfile_active(temp))
2884 delete_tempfile(&temp);
2885 errno = saved_errno;
2886
c18b80a0 2887 if (ret)
df60cf57 2888 goto out;
c18b80a0
NTND
2889 }
2890
0d59ffb4
CC
2891 ret = write_split_index(istate, lock, flags);
2892
2893 /* Freshen the shared index only if the split-index was written */
a125a223
TG
2894 if (!ret && !new_shared_index) {
2895 const char *shared_index = git_path("sharedindex.%s",
2182abd9 2896 oid_to_hex(&si->base_oid));
a125a223
TG
2897 freshen_shared_index(shared_index, 1);
2898 }
0d59ffb4 2899
df60cf57
2900out:
2901 if (flags & COMMIT_LOCK)
2902 rollback_lock_file(lock);
0d59ffb4 2903 return ret;
03b86647
NTND
2904}
2905
e46bbcf6
MV
2906/*
2907 * Read the index file that is potentially unmerged into given
ad376204
EN
2908 * index_state, dropping any unmerged entries to stage #0 (potentially
2909 * resulting in a path appearing as both a file and a directory in the
2910 * index; the caller is responsible to clear out the extra entries
2911 * before writing the index to a tree). Returns true if the index is
2912 * unmerged. Callers who want to refuse to work from an unmerged
2913 * state can call this and check its return value, instead of calling
2914 * read_cache().
e46bbcf6
MV
2915 */
2916int read_index_unmerged(struct index_state *istate)
2917{
2918 int i;
d1a43f2a 2919 int unmerged = 0;
e46bbcf6
MV
2920
2921 read_index(istate);
e46bbcf6
MV
2922 for (i = 0; i < istate->cache_nr; i++) {
2923 struct cache_entry *ce = istate->cache[i];
d1a43f2a 2924 struct cache_entry *new_ce;
a849735b 2925 int len;
d1a43f2a
JH
2926
2927 if (!ce_stage(ce))
e46bbcf6 2928 continue;
d1a43f2a 2929 unmerged = 1;
68c4f6a5 2930 len = ce_namelen(ce);
a849735b 2931 new_ce = make_empty_cache_entry(istate, len);
d1a43f2a 2932 memcpy(new_ce->name, ce->name, len);
b60e188c
TG
2933 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2934 new_ce->ce_namelen = len;
d1a43f2a 2935 new_ce->ce_mode = ce->ce_mode;
ad376204 2936 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
d1a43f2a 2937 return error("%s: cannot drop to stage #0",
5699d17e 2938 new_ce->name);
e46bbcf6 2939 }
d1a43f2a 2940 return unmerged;
e46bbcf6 2941}
041aee31 2942
98fa4738
JK
2943/*
2944 * Returns 1 if the path is an "other" path with respect to
2945 * the index; that is, the path is not mentioned in the index at all,
2946 * either as a file, a directory with some files in the index,
2947 * or as an unmerged entry.
2948 *
2949 * We helpfully remove a trailing "/" from directories so that
2950 * the output of read_directory can be used as-is.
2951 */
2952int index_name_is_other(const struct index_state *istate, const char *name,
2953 int namelen)
2954{
2955 int pos;
2956 if (namelen && name[namelen - 1] == '/')
2957 namelen--;
2958 pos = index_name_pos(istate, name, namelen);
2959 if (0 <= pos)
2960 return 0; /* exact match */
2961 pos = -pos - 1;
2962 if (pos < istate->cache_nr) {
2963 struct cache_entry *ce = istate->cache[pos];
2964 if (ce_namelen(ce) == namelen &&
2965 !memcmp(ce->name, name, namelen))
2966 return 0; /* Yup, this one exists unmerged */
2967 }
2968 return 1;
2969}
29fb37b2 2970
87542508
BW
2971void *read_blob_data_from_index(const struct index_state *istate,
2972 const char *path, unsigned long *size)
29fb37b2
LF
2973{
2974 int pos, len;
2975 unsigned long sz;
2976 enum object_type type;
2977 void *data;
2978
2979 len = strlen(path);
2980 pos = index_name_pos(istate, path, len);
2981 if (pos < 0) {
2982 /*
2983 * We might be in the middle of a merge, in which
2984 * case we would read stage #2 (ours).
2985 */
2986 int i;
2987 for (i = -pos - 1;
2988 (pos < 0 && i < istate->cache_nr &&
2989 !strcmp(istate->cache[i]->name, path));
2990 i++)
2991 if (ce_stage(istate->cache[i]) == 2)
2992 pos = i;
2993 }
2994 if (pos < 0)
2995 return NULL;
b4f5aca4 2996 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
29fb37b2
LF
2997 if (!data || type != OBJ_BLOB) {
2998 free(data);
2999 return NULL;
3000 }
ff366825
LF
3001 if (size)
3002 *size = sz;
29fb37b2
LF
3003 return data;
3004}
38612532
MH
3005
3006void stat_validity_clear(struct stat_validity *sv)
3007{
6a83d902 3008 FREE_AND_NULL(sv->sd);
38612532
MH
3009}
3010
3011int stat_validity_check(struct stat_validity *sv, const char *path)
3012{
3013 struct stat st;
3014
3015 if (stat(path, &st) < 0)
3016 return sv->sd == NULL;
3017 if (!sv->sd)
3018 return 0;
3019 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3020}
3021
3022void stat_validity_update(struct stat_validity *sv, int fd)
3023{
3024 struct stat st;
3025
3026 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3027 stat_validity_clear(sv);
3028 else {
3029 if (!sv->sd)
3030 sv->sd = xcalloc(1, sizeof(struct stat_data));
3031 fill_stat_data(sv->sd, &st);
3032 }
3033}
edf3b905
DT
3034
3035void move_index_extensions(struct index_state *dst, struct index_state *src)
3036{
3037 dst->untracked = src->untracked;
3038 src->untracked = NULL;
836ef2b6
NTND
3039 dst->cache_tree = src->cache_tree;
3040 src->cache_tree = NULL;
edf3b905 3041}
a849735b 3042
8e72d675
JM
3043struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3044 struct index_state *istate)
3045{
3046 unsigned int size = ce_size(ce);
3047 int mem_pool_allocated;
3048 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3049 mem_pool_allocated = new_entry->mem_pool_allocated;
3050
3051 memcpy(new_entry, ce, size);
3052 new_entry->mem_pool_allocated = mem_pool_allocated;
3053 return new_entry;
3054}
3055
a849735b
JM
3056void discard_cache_entry(struct cache_entry *ce)
3057{
8616a2d0
JM
3058 if (ce && should_validate_cache_entries())
3059 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3060
8e72d675
JM
3061 if (ce && ce->mem_pool_allocated)
3062 return;
3063
a849735b
JM
3064 free(ce);
3065}
8616a2d0
JM
3066
3067int should_validate_cache_entries(void)
3068{
3069 static int validate_index_cache_entries = -1;
3070
3071 if (validate_index_cache_entries < 0) {
3072 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3073 validate_index_cache_entries = 1;
3074 else
3075 validate_index_cache_entries = 0;
3076 }
3077
3078 return validate_index_cache_entries;
3079}
3b1d9e04
BP
3080
3081#define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3082#define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3083
3084static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3085{
3086 /*
3087 * The end of index entries (EOIE) extension is guaranteed to be last
3088 * so that it can be found by scanning backwards from the EOF.
3089 *
3090 * "EOIE"
3091 * <4-byte length>
3092 * <4-byte offset>
3093 * <20-byte hash>
3094 */
3095 const char *index, *eoie;
3096 uint32_t extsize;
3097 size_t offset, src_offset;
3098 unsigned char hash[GIT_MAX_RAWSZ];
3099 git_hash_ctx c;
3100
3101 /* ensure we have an index big enough to contain an EOIE extension */
3102 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3103 return 0;
3104
3105 /* validate the extension signature */
3106 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3107 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3108 return 0;
3109 index += sizeof(uint32_t);
3110
3111 /* validate the extension size */
3112 extsize = get_be32(index);
3113 if (extsize != EOIE_SIZE)
3114 return 0;
3115 index += sizeof(uint32_t);
3116
3117 /*
3118 * Validate the offset we're going to look for the first extension
3119 * signature is after the index header and before the eoie extension.
3120 */
3121 offset = get_be32(index);
3122 if (mmap + offset < mmap + sizeof(struct cache_header))
3123 return 0;
3124 if (mmap + offset >= eoie)
3125 return 0;
3126 index += sizeof(uint32_t);
3127
3128 /*
3129 * The hash is computed over extension types and their sizes (but not
3130 * their contents). E.g. if we have "TREE" extension that is N-bytes
3131 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3132 * then the hash would be:
3133 *
3134 * SHA-1("TREE" + <binary representation of N> +
3135 * "REUC" + <binary representation of M>)
3136 */
3137 src_offset = offset;
3138 the_hash_algo->init_fn(&c);
3139 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3140 /* After an array of active_nr index entries,
3141 * there can be arbitrary number of extended
3142 * sections, each of which is prefixed with
3143 * extension name (4-byte) and section length
3144 * in 4-byte network byte order.
3145 */
3146 uint32_t extsize;
3147 memcpy(&extsize, mmap + src_offset + 4, 4);
3148 extsize = ntohl(extsize);
3149
3150 /* verify the extension size isn't so large it will wrap around */
3151 if (src_offset + 8 + extsize < src_offset)
3152 return 0;
3153
3154 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3155
3156 src_offset += 8;
3157 src_offset += extsize;
3158 }
3159 the_hash_algo->final_fn(hash, &c);
3160 if (!hasheq(hash, (const unsigned char *)index))
3161 return 0;
3162
3163 /* Validate that the extension offsets returned us back to the eoie extension. */
3164 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3165 return 0;
3166
3167 return offset;
3168}
3169
3170static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3171{
3172 uint32_t buffer;
3173 unsigned char hash[GIT_MAX_RAWSZ];
3174
3175 /* offset */
3176 put_be32(&buffer, offset);
3177 strbuf_add(sb, &buffer, sizeof(uint32_t));
3178
3179 /* hash */
3180 the_hash_algo->final_fn(hash, eoie_context);
3181 strbuf_add(sb, hash, the_hash_algo->rawsz);
3182}