trace2:data: add trace2 hook classification
[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 */
e83c5163 6#include "cache.h"
b2141fc1 7#include "config.h"
cffbfad5
EN
8#include "diff.h"
9#include "diffcore.h"
f6ecc62d 10#include "tempfile.h"
697cc8ef 11#include "lockfile.h"
bad68ec9 12#include "cache-tree.h"
f35a6d3b 13#include "refs.h"
d616813d 14#include "dir.h"
cbd53a21 15#include "object-store.h"
041aee31
JH
16#include "tree.h"
17#include "commit.h"
39425819 18#include "blob.h"
cfc5789a 19#include "resolve-undo.h"
6c9cd161
JH
20#include "strbuf.h"
21#include "varint.h"
5fc2fc8f 22#include "split-index.h"
a42643aa 23#include "utf8.h"
883e248b 24#include "fsmonitor.h"
abb4bb83 25#include "thread-utils.h"
ae9af122 26#include "progress.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" */
3255089a 48#define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
5fc2fc8f
NTND
49
50/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
e0cf0d7d 51#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
c18b80a0 52 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
883e248b 53 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
e83c5163 54
8e72d675
JM
55
56/*
57 * This is an estimate of the pathname length in the index. We use
58 * this for V4 index files to guess the un-deltafied size of the index
59 * in memory because of pathname deltafication. This is not required
60 * for V2/V3 index formats because their pathnames are not compressed.
61 * If the initial amount of memory set aside is not sufficient, the
62 * mem pool will allocate extra memory.
63 */
64#define CACHE_ENTRY_PATH_LENGTH 80
65
66static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
67{
68 struct cache_entry *ce;
69 ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
70 ce->mem_pool_allocated = 1;
71 return ce;
72}
73
74static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
75{
76 struct cache_entry * ce;
77 ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
78 ce->mem_pool_allocated = 1;
79 return ce;
80}
81
82static struct mem_pool *find_mem_pool(struct index_state *istate)
83{
84 struct mem_pool **pool_ptr;
85
86 if (istate->split_index && istate->split_index->base)
87 pool_ptr = &istate->split_index->base->ce_mem_pool;
88 else
89 pool_ptr = &istate->ce_mem_pool;
90
91 if (!*pool_ptr)
92 mem_pool_init(pool_ptr, 0);
93
94 return *pool_ptr;
95}
96
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
58bf2a4c
NTND
210static int ce_compare_data(struct index_state *istate,
211 const struct cache_entry *ce,
212 struct stat *st)
29e4d363
JH
213{
214 int match = -1;
1b8ac5ea 215 int fd = git_open_cloexec(ce->name, O_RDONLY);
29e4d363
JH
216
217 if (fd >= 0) {
bebfecb9 218 struct object_id oid;
58bf2a4c 219 if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
6a29d7b7 220 match = !oideq(&oid, &ce->oid);
7f8508e8 221 /* index_fd() closed the file descriptor already */
29e4d363
JH
222 }
223 return match;
224}
225
21a6b9fa 226static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
29e4d363
JH
227{
228 int match = -1;
29e4d363
JH
229 void *buffer;
230 unsigned long size;
21666f1a 231 enum object_type type;
a60272b3 232 struct strbuf sb = STRBUF_INIT;
29e4d363 233
a60272b3 234 if (strbuf_readlink(&sb, ce->name, expected_size))
29e4d363 235 return -1;
a60272b3 236
b4f5aca4 237 buffer = read_object_file(&ce->oid, &type, &size);
a60272b3
LT
238 if (buffer) {
239 if (size == sb.len)
240 match = memcmp(buffer, sb.buf, size);
241 free(buffer);
29e4d363 242 }
a60272b3 243 strbuf_release(&sb);
29e4d363
JH
244 return match;
245}
246
21a6b9fa 247static int ce_compare_gitlink(const struct cache_entry *ce)
f35a6d3b 248{
1053fe82 249 struct object_id oid;
f35a6d3b
LT
250
251 /*
252 * We don't actually require that the .git directory
302b9282 253 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
254 * might even be missing (in case nobody populated that
255 * sub-project).
256 *
257 * If so, we consider it always to match.
258 */
a98e6101 259 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
f35a6d3b 260 return 0;
6a29d7b7 261 return !oideq(&oid, &ce->oid);
f35a6d3b
LT
262}
263
58bf2a4c
NTND
264static int ce_modified_check_fs(struct index_state *istate,
265 const struct cache_entry *ce,
266 struct stat *st)
29e4d363
JH
267{
268 switch (st->st_mode & S_IFMT) {
269 case S_IFREG:
58bf2a4c 270 if (ce_compare_data(istate, ce, st))
29e4d363
JH
271 return DATA_CHANGED;
272 break;
273 case S_IFLNK:
dc49cd76 274 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
275 return DATA_CHANGED;
276 break;
a8ee75bc 277 case S_IFDIR:
7a51ed66 278 if (S_ISGITLINK(ce->ce_mode))
c70115b4 279 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
1cf01a34 280 /* else fallthrough */
29e4d363
JH
281 default:
282 return TYPE_CHANGED;
283 }
284 return 0;
285}
286
21a6b9fa 287static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
734aab75
LT
288{
289 unsigned int changed = 0;
290
7a51ed66
LT
291 if (ce->ce_flags & CE_REMOVE)
292 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
293
294 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
295 case S_IFREG:
296 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
297 /* We consider only the owner x bit to be relevant for
298 * "mode changes"
299 */
300 if (trust_executable_bit &&
7a51ed66 301 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 302 changed |= MODE_CHANGED;
8ae0a8c5
KS
303 break;
304 case S_IFLNK:
78a8d641
JS
305 if (!S_ISLNK(st->st_mode) &&
306 (has_symlinks || !S_ISREG(st->st_mode)))
307 changed |= TYPE_CHANGED;
8ae0a8c5 308 break;
302b9282 309 case S_IFGITLINK:
c70115b4 310 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
311 if (!S_ISDIR(st->st_mode))
312 changed |= TYPE_CHANGED;
313 else if (ce_compare_gitlink(ce))
314 changed |= DATA_CHANGED;
a8ee75bc 315 return changed;
8ae0a8c5 316 default:
391408e5 317 BUG("unsupported ce_mode: %o", ce->ce_mode);
8ae0a8c5 318 }
2cb45e95 319
c21d39d7 320 changed |= match_stat_data(&ce->ce_stat_data, st);
b0391890 321
f49c2c22 322 /* Racily smudged entry? */
c21d39d7 323 if (!ce->ce_stat_data.sd_size) {
99d1a986 324 if (!is_empty_blob_sha1(ce->oid.hash))
f49c2c22
LT
325 changed |= DATA_CHANGED;
326 }
327
407c8eb0
JH
328 return changed;
329}
330
2bb4cda1
NTND
331static int is_racy_stat(const struct index_state *istate,
332 const struct stat_data *sd)
6d91da6d 333{
2bb4cda1 334 return (istate->timestamp.sec &&
fba2f38a
KB
335#ifdef USE_NSEC
336 /* nanosecond timestamped files can also be racy! */
2bb4cda1
NTND
337 (istate->timestamp.sec < sd->sd_mtime.sec ||
338 (istate->timestamp.sec == sd->sd_mtime.sec &&
339 istate->timestamp.nsec <= sd->sd_mtime.nsec))
fba2f38a 340#else
2bb4cda1 341 istate->timestamp.sec <= sd->sd_mtime.sec
fba2f38a 342#endif
2bb4cda1
NTND
343 );
344}
345
5581a019 346int is_racy_timestamp(const struct index_state *istate,
2bb4cda1
NTND
347 const struct cache_entry *ce)
348{
349 return (!S_ISGITLINK(ce->ce_mode) &&
350 is_racy_stat(istate, &ce->ce_stat_data));
6d91da6d
JH
351}
352
ed4efab1
NTND
353int match_stat_data_racy(const struct index_state *istate,
354 const struct stat_data *sd, struct stat *st)
355{
356 if (is_racy_stat(istate, sd))
357 return MTIME_CHANGED;
358 return match_stat_data(sd, st);
6d91da6d
JH
359}
360
883e248b 361int ie_match_stat(struct index_state *istate,
21a6b9fa 362 const struct cache_entry *ce, struct stat *st,
4bd5b7da 363 unsigned int options)
407c8eb0 364{
5f73076c 365 unsigned int changed;
4bd5b7da 366 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 367 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
4bd5b7da 368 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
883e248b 369 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
5f73076c 370
883e248b
BP
371 if (!ignore_fsmonitor)
372 refresh_fsmonitor(istate);
5f73076c
JH
373 /*
374 * If it's marked as always valid in the index, it's
375 * valid whatever the checked-out copy says.
56cac48c
NTND
376 *
377 * skip-worktree has the same effect with higher precedence
5f73076c 378 */
56cac48c
NTND
379 if (!ignore_skip_worktree && ce_skip_worktree(ce))
380 return 0;
7a51ed66 381 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c 382 return 0;
883e248b
BP
383 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
384 return 0;
5f73076c 385
331fcb59
JH
386 /*
387 * Intent-to-add entries have not been added, so the index entry
388 * by definition never matches what is in the work tree until it
389 * actually gets added.
390 */
895ff3b2 391 if (ce_intent_to_add(ce))
331fcb59
JH
392 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
393
5f73076c 394 changed = ce_match_stat_basic(ce, st);
407c8eb0 395
29e4d363
JH
396 /*
397 * Within 1 second of this sequence:
398 * echo xyzzy >file && git-update-index --add file
399 * running this command:
400 * echo frotz >file
401 * would give a falsely clean cache entry. The mtime and
402 * length match the cache, and other stat fields do not change.
403 *
404 * We could detect this at update-index time (the cache entry
405 * being registered/updated records the same time as "now")
406 * and delay the return from git-update-index, but that would
407 * effectively mean we can make at most one commit per second,
408 * which is not acceptable. Instead, we check cache entries
409 * whose mtime are the same as the index file timestamp more
5f73076c 410 * carefully than others.
29e4d363 411 */
6d91da6d 412 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
413 if (assume_racy_is_modified)
414 changed |= DATA_CHANGED;
415 else
58bf2a4c 416 changed |= ce_modified_check_fs(istate, ce, st);
42f77406 417 }
b0391890 418
29e4d363 419 return changed;
b0391890
JH
420}
421
883e248b 422int ie_modified(struct index_state *istate,
21a6b9fa
RS
423 const struct cache_entry *ce,
424 struct stat *st, unsigned int options)
b0391890 425{
29e4d363 426 int changed, changed_fs;
4bd5b7da
JH
427
428 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
429 if (!changed)
430 return 0;
b0391890
JH
431 /*
432 * If the mode or type has changed, there's no point in trying
433 * to refresh the entry - it's not going to match
434 */
435 if (changed & (MODE_CHANGED | TYPE_CHANGED))
436 return changed;
437
c70115b4
JH
438 /*
439 * Immediately after read-tree or update-index --cacheinfo,
440 * the length field is zero, as we have never even read the
441 * lstat(2) information once, and we cannot trust DATA_CHANGED
442 * returned by ie_match_stat() which in turn was returned by
443 * ce_match_stat_basic() to signal that the filesize of the
444 * blob changed. We have to actually go to the filesystem to
445 * see if the contents match, and if so, should answer "unchanged".
446 *
447 * The logic does not apply to gitlinks, as ce_match_stat_basic()
448 * already has checked the actual HEAD from the filesystem in the
449 * subproject. If ie_match_stat() already said it is different,
450 * then we know it is.
b0391890 451 */
c70115b4 452 if ((changed & DATA_CHANGED) &&
c21d39d7 453 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
b0391890
JH
454 return changed;
455
58bf2a4c 456 changed_fs = ce_modified_check_fs(istate, ce, st);
29e4d363
JH
457 if (changed_fs)
458 return changed | changed_fs;
b0391890
JH
459 return 0;
460}
461
958ba6c9
LT
462int base_name_compare(const char *name1, int len1, int mode1,
463 const char *name2, int len2, int mode2)
464{
465 unsigned char c1, c2;
466 int len = len1 < len2 ? len1 : len2;
467 int cmp;
468
469 cmp = memcmp(name1, name2, len);
470 if (cmp)
471 return cmp;
472 c1 = name1[len];
473 c2 = name2[len];
1833a925 474 if (!c1 && S_ISDIR(mode1))
958ba6c9 475 c1 = '/';
1833a925 476 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
477 c2 = '/';
478 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
479}
480
0ab9e1e8
LT
481/*
482 * df_name_compare() is identical to base_name_compare(), except it
483 * compares conflicting directory/file entries as equal. Note that
484 * while a directory name compares as equal to a regular file, they
485 * then individually compare _differently_ to a filename that has
486 * a dot after the basename (because '\0' < '.' < '/').
487 *
488 * This is used by routines that want to traverse the git namespace
489 * but then handle conflicting entries together when possible.
490 */
491int df_name_compare(const char *name1, int len1, int mode1,
492 const char *name2, int len2, int mode2)
493{
494 int len = len1 < len2 ? len1 : len2, cmp;
495 unsigned char c1, c2;
496
497 cmp = memcmp(name1, name2, len);
498 if (cmp)
499 return cmp;
500 /* Directories and files compare equal (same length, same name) */
501 if (len1 == len2)
502 return 0;
503 c1 = name1[len];
504 if (!c1 && S_ISDIR(mode1))
505 c1 = '/';
506 c2 = name2[len];
507 if (!c2 && S_ISDIR(mode2))
508 c2 = '/';
509 if (c1 == '/' && !c2)
510 return 0;
511 if (c2 == '/' && !c1)
512 return 0;
513 return c1 - c2;
514}
515
ccdd4a0f 516int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
eb38c22f 517{
ccdd4a0f
JM
518 size_t min_len = (len1 < len2) ? len1 : len2;
519 int cmp = memcmp(name1, name2, min_len);
eb38c22f
LT
520 if (cmp)
521 return cmp;
522 if (len1 < len2)
523 return -1;
524 if (len1 > len2)
525 return 1;
ccdd4a0f
JM
526 return 0;
527}
528
529int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
530{
531 int cmp;
532
533 cmp = name_compare(name1, len1, name2, len2);
534 if (cmp)
535 return cmp;
5f73076c 536
b60e188c 537 if (stage1 < stage2)
95fd5bf8 538 return -1;
b60e188c 539 if (stage1 > stage2)
95fd5bf8 540 return 1;
eb38c22f
LT
541 return 0;
542}
543
357e9c69 544static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
eb38c22f
LT
545{
546 int first, last;
547
548 first = 0;
4aab5b46 549 last = istate->cache_nr;
eb38c22f
LT
550 while (last > first) {
551 int next = (last + first) >> 1;
4aab5b46 552 struct cache_entry *ce = istate->cache[next];
b60e188c 553 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
eb38c22f 554 if (!cmp)
76e7f4ec 555 return next;
eb38c22f
LT
556 if (cmp < 0) {
557 last = next;
558 continue;
559 }
560 first = next+1;
561 }
76e7f4ec 562 return -first-1;
eb38c22f
LT
563}
564
b60e188c
TG
565int index_name_pos(const struct index_state *istate, const char *name, int namelen)
566{
567 return index_name_stage_pos(istate, name, namelen, 0);
568}
569
4aab5b46 570int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 571{
cf558704
LT
572 struct cache_entry *ce = istate->cache[pos];
573
cfc5789a 574 record_resolve_undo(istate, ce);
2092678c 575 remove_name_hash(istate, ce);
045113a5 576 save_or_free_index_entry(istate, ce);
e636a7b4 577 istate->cache_changed |= CE_ENTRY_REMOVED;
4aab5b46
JH
578 istate->cache_nr--;
579 if (pos >= istate->cache_nr)
7b937ca3 580 return 0;
f331ab9d
RS
581 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
582 istate->cache_nr - pos);
7b937ca3
LT
583 return 1;
584}
585
36419c8e 586/*
98e023de 587 * Remove all cache entries marked for removal, that is where
36419c8e
KB
588 * CE_REMOVE is set in ce_flags. This is much more effective than
589 * calling remove_index_entry_at() for each entry to be removed.
590 */
591void remove_marked_cache_entries(struct index_state *istate)
592{
593 struct cache_entry **ce_array = istate->cache;
594 unsigned int i, j;
595
596 for (i = j = 0; i < istate->cache_nr; i++) {
5699d17e 597 if (ce_array[i]->ce_flags & CE_REMOVE) {
2092678c 598 remove_name_hash(istate, ce_array[i]);
045113a5 599 save_or_free_index_entry(istate, ce_array[i]);
5699d17e 600 }
36419c8e
KB
601 else
602 ce_array[j++] = ce_array[i];
603 }
ad837d9e
NTND
604 if (j == istate->cache_nr)
605 return;
e636a7b4 606 istate->cache_changed |= CE_ENTRY_REMOVED;
36419c8e
KB
607 istate->cache_nr = j;
608}
609
4aab5b46 610int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 611{
4aab5b46 612 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
613 if (pos < 0)
614 pos = -pos-1;
a5400efe 615 cache_tree_invalidate_path(istate, path);
e931371a 616 untracked_cache_remove_from_index(istate, path);
4aab5b46
JH
617 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
618 remove_index_entry_at(istate, pos);
197ee8c9
LT
619 return 0;
620}
621
20314271
JS
622static int compare_name(struct cache_entry *ce, const char *path, int namelen)
623{
624 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
625}
626
627static int index_name_pos_also_unmerged(struct index_state *istate,
628 const char *path, int namelen)
629{
630 int pos = index_name_pos(istate, path, namelen);
631 struct cache_entry *ce;
632
633 if (pos >= 0)
634 return pos;
635
636 /* maybe unmerged? */
637 pos = -1 - pos;
638 if (pos >= istate->cache_nr ||
639 compare_name((ce = istate->cache[pos]), path, namelen))
640 return -1;
641
642 /* order of preference: stage 2, 1, 3 */
643 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
644 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
645 !compare_name(ce, path, namelen))
646 pos++;
647 return pos;
648}
649
1102952b
LT
650static int different_name(struct cache_entry *ce, struct cache_entry *alias)
651{
652 int len = ce_namelen(ce);
653 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
654}
655
656/*
657 * If we add a filename that aliases in the cache, we will use the
658 * name that we already have - but we don't want to update the same
659 * alias twice, because that implies that there were actually two
660 * different files with aliasing names!
661 *
662 * So we use the CE_ADDED flag to verify that the alias was an old
663 * one before we accept it as
664 */
045113a5
NTND
665static struct cache_entry *create_alias_ce(struct index_state *istate,
666 struct cache_entry *ce,
667 struct cache_entry *alias)
1102952b
LT
668{
669 int len;
285c2e25 670 struct cache_entry *new_entry;
1102952b
LT
671
672 if (alias->ce_flags & CE_ADDED)
9d0a9e90
NTND
673 die(_("will not add file alias '%s' ('%s' already exists in index)"),
674 ce->name, alias->name);
1102952b
LT
675
676 /* Ok, create the new entry using the name of the existing alias */
677 len = ce_namelen(alias);
a849735b 678 new_entry = make_empty_cache_entry(istate, len);
285c2e25
BW
679 memcpy(new_entry->name, alias->name, len);
680 copy_cache_entry(new_entry, ce);
045113a5 681 save_or_free_index_entry(istate, ce);
285c2e25 682 return new_entry;
1102952b
LT
683}
684
b4b313f9 685void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
39425819 686{
a09c985e
PO
687 struct object_id oid;
688 if (write_object_file("", 0, blob_type, &oid))
9d0a9e90 689 die(_("cannot create an empty blob in the object database"));
a09c985e 690 oidcpy(&ce->oid, &oid);
39425819
JH
691}
692
610d55af 693int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 694{
a849735b 695 int namelen, was_same;
d177cab0 696 mode_t st_mode = st->st_mode;
9472935d 697 struct cache_entry *ce, *alias = NULL;
56cac48c 698 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
699 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
700 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
701 int intent_only = flags & ADD_CACHE_INTENT;
702 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
703 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
9e5da3d0 704 int hash_flags = HASH_WRITE_OBJECT;
9472935d 705
9e5da3d0
JK
706 if (flags & ADD_CACHE_RENORMALIZE)
707 hash_flags |= HASH_RENORMALIZE;
11be42a4 708
d177cab0 709 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
9d0a9e90 710 return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
11be42a4
JS
711
712 namelen = strlen(path);
d177cab0 713 if (S_ISDIR(st_mode)) {
09595258
LT
714 while (namelen && path[namelen-1] == '/')
715 namelen--;
716 }
a849735b 717 ce = make_empty_cache_entry(istate, namelen);
11be42a4 718 memcpy(ce->name, path, namelen);
b60e188c 719 ce->ce_namelen = namelen;
39425819
JH
720 if (!intent_only)
721 fill_stat_cache_info(ce, st);
388b2acd
JH
722 else
723 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 724
610d55af
TG
725
726 if (trust_executable_bit && has_symlinks) {
d177cab0 727 ce->ce_mode = create_ce_mode(st_mode);
610d55af 728 } else {
78a8d641
JS
729 /* If there is an existing entry, pick the mode bits and type
730 * from it, otherwise assume unexecutable regular file.
11be42a4 731 */
185c975f 732 struct cache_entry *ent;
20314271 733 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 734
4aab5b46 735 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 736 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
737 }
738
dc1ae704
JJ
739 /* When core.ignorecase=true, determine if a directory of the same name but differing
740 * case already exists within the Git repository. If it does, ensure the directory
741 * case of the file being added to the repository matches (is folded into) the existing
742 * entry's directory case.
743 */
744 if (ignore_case) {
41284eb0 745 adjust_dirname_case(istate, ce->name);
dc1ae704 746 }
e2c2a375 747 if (!(flags & ADD_CACHE_RENORMALIZE)) {
9472935d
TB
748 alias = index_file_exists(istate, ce->name,
749 ce_namelen(ce), ignore_case);
750 if (alias &&
751 !ce_stage(alias) &&
752 !ie_match_stat(istate, alias, st, ce_option)) {
753 /* Nothing changed, really */
754 if (!S_ISGITLINK(alias->ce_mode))
755 ce_mark_uptodate(alias);
756 alias->ce_flags |= CE_ADDED;
dc1ae704 757
a849735b 758 discard_cache_entry(ce);
9472935d
TB
759 return 0;
760 }
0781b8a9 761 }
39425819 762 if (!intent_only) {
1c418243 763 if (index_path(istate, &ce->oid, path, st, hash_flags)) {
a849735b 764 discard_cache_entry(ce);
9d0a9e90 765 return error(_("unable to index file '%s'"), path);
2d9426b0 766 }
39425819 767 } else
b4b313f9 768 set_object_name_for_intent_to_add_entry(ce);
39425819 769
1102952b 770 if (ignore_case && alias && different_name(ce, alias))
045113a5 771 ce = create_alias_ce(istate, ce, alias);
1102952b 772 ce->ce_flags |= CE_ADDED;
38ed1d89 773
3bf0dd1f 774 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
775 was_same = (alias &&
776 !ce_stage(alias) &&
4a7e27e9 777 oideq(&alias->oid, &ce->oid) &&
38ed1d89
JH
778 ce->ce_mode == alias->ce_mode);
779
780 if (pretend)
a849735b 781 discard_cache_entry(ce);
067178ed 782 else if (add_index_entry(istate, ce, add_option)) {
a849735b 783 discard_cache_entry(ce);
9d0a9e90 784 return error(_("unable to add '%s' to index"), path);
067178ed 785 }
38ed1d89 786 if (verbose && !was_same)
11be42a4 787 printf("add '%s'\n", path);
11be42a4
JS
788 return 0;
789}
790
610d55af 791int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
792{
793 struct stat st;
794 if (lstat(path, &st))
9d0a9e90 795 die_errno(_("unable to stat '%s'"), path);
610d55af 796 return add_to_index(istate, path, &st, flags);
d177cab0
LT
797}
798
a849735b
JM
799struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
800{
8e72d675 801 return mem_pool__ce_calloc(find_mem_pool(istate), len);
a849735b
JM
802}
803
804struct cache_entry *make_empty_transient_cache_entry(size_t len)
805{
806 return xcalloc(1, cache_entry_size(len));
807}
808
809struct cache_entry *make_cache_entry(struct index_state *istate,
810 unsigned int mode,
825ed4d9
JM
811 const struct object_id *oid,
812 const char *path,
813 int stage,
814 unsigned int refresh_options)
6640f881 815{
bc1c2caa 816 struct cache_entry *ce, *ret;
a849735b 817 int len;
6640f881 818
10ecfa76 819 if (!verify_path(path, mode)) {
9d0a9e90 820 error(_("invalid path '%s'"), path);
6640f881 821 return NULL;
7e7abea9 822 }
6640f881
CR
823
824 len = strlen(path);
a849735b 825 ce = make_empty_cache_entry(istate, len);
6640f881 826
825ed4d9 827 oidcpy(&ce->oid, oid);
6640f881 828 memcpy(ce->name, path, len);
b60e188c
TG
829 ce->ce_flags = create_ce_flags(stage);
830 ce->ce_namelen = len;
6640f881
CR
831 ce->ce_mode = create_ce_mode(mode);
832
d7b665c3 833 ret = refresh_cache_entry(istate, ce, refresh_options);
915e44c6 834 if (ret != ce)
a849735b 835 discard_cache_entry(ce);
915e44c6 836 return ret;
6640f881
CR
837}
838
a849735b
JM
839struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
840 const char *path, int stage)
841{
842 struct cache_entry *ce;
843 int len;
844
845 if (!verify_path(path, mode)) {
9d0a9e90 846 error(_("invalid path '%s'"), path);
a849735b
JM
847 return NULL;
848 }
849
850 len = strlen(path);
851 ce = make_empty_transient_cache_entry(len);
852
853 oidcpy(&ce->oid, oid);
854 memcpy(ce->name, path, len);
855 ce->ce_flags = create_ce_flags(stage);
856 ce->ce_namelen = len;
857 ce->ce_mode = create_ce_mode(mode);
858
859 return ce;
860}
861
d9d70966
TG
862/*
863 * Chmod an index entry with either +x or -x.
864 *
865 * Returns -1 if the chmod for the particular cache entry failed (if it's
866 * not a regular file), -2 if an invalid flip argument is passed in, 0
867 * otherwise.
868 */
869int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
870 char flip)
871{
872 if (!S_ISREG(ce->ce_mode))
873 return -1;
874 switch (flip) {
875 case '+':
876 ce->ce_mode |= 0111;
877 break;
878 case '-':
879 ce->ce_mode &= ~0111;
880 break;
881 default:
882 return -2;
883 }
884 cache_tree_invalidate_path(istate, ce->name);
885 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 886 mark_fsmonitor_invalid(istate, ce);
d9d70966
TG
887 istate->cache_changed |= CE_ENTRY_CHANGED;
888
889 return 0;
890}
891
9c5e6c80 892int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
7b937ca3
LT
893{
894 int len = ce_namelen(a);
895 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
896}
897
8dcf39c4
LT
898/*
899 * We fundamentally don't like some paths: we don't want
900 * dot or dot-dot anywhere, and for obvious reasons don't
901 * want to recurse into ".git" either.
902 *
903 * Also, we don't want double slashes or slashes at the
904 * end that can make pathnames ambiguous.
905 */
10ecfa76 906static int verify_dotfile(const char *rest, unsigned mode)
8dcf39c4
LT
907{
908 /*
909 * The first character was '.', but that
910 * has already been discarded, we now test
911 * the rest.
912 */
e0f530ff 913
8dcf39c4 914 /* "." is not allowed */
e0f530ff 915 if (*rest == '\0' || is_dir_sep(*rest))
8dcf39c4
LT
916 return 0;
917
e0f530ff 918 switch (*rest) {
8dcf39c4 919 /*
641084b6
JK
920 * ".git" followed by NUL or slash is bad. Note that we match
921 * case-insensitively here, even if ignore_case is not set.
922 * This outlaws ".GIT" everywhere out of an abundance of caution,
923 * since there's really no good reason to allow it.
10ecfa76
JK
924 *
925 * Once we've seen ".git", we can also find ".gitmodules", etc (also
926 * case-insensitively).
8dcf39c4
LT
927 */
928 case 'g':
cc2fc7c2
JK
929 case 'G':
930 if (rest[1] != 'i' && rest[1] != 'I')
8dcf39c4 931 break;
cc2fc7c2 932 if (rest[2] != 't' && rest[2] != 'T')
8dcf39c4 933 break;
e19e5e66
JK
934 if (rest[3] == '\0' || is_dir_sep(rest[3]))
935 return 0;
10ecfa76
JK
936 if (S_ISLNK(mode)) {
937 rest += 3;
938 if (skip_iprefix(rest, "modules", &rest) &&
939 (*rest == '\0' || is_dir_sep(*rest)))
940 return 0;
941 }
e19e5e66 942 break;
8dcf39c4 943 case '.':
e0f530ff 944 if (rest[1] == '\0' || is_dir_sep(rest[1]))
8dcf39c4
LT
945 return 0;
946 }
947 return 1;
948}
949
10ecfa76 950int verify_path(const char *path, unsigned mode)
8dcf39c4
LT
951{
952 char c;
953
56948cb6
EFL
954 if (has_dos_drive_prefix(path))
955 return 0;
956
8dcf39c4
LT
957 goto inside;
958 for (;;) {
959 if (!c)
960 return 1;
56948cb6 961 if (is_dir_sep(c)) {
8dcf39c4 962inside:
10ecfa76
JK
963 if (protect_hfs) {
964 if (is_hfs_dotgit(path))
965 return 0;
966 if (S_ISLNK(mode)) {
967 if (is_hfs_dotgitmodules(path))
968 return 0;
969 }
970 }
971 if (protect_ntfs) {
972 if (is_ntfs_dotgit(path))
973 return 0;
974 if (S_ISLNK(mode)) {
975 if (is_ntfs_dotgitmodules(path))
976 return 0;
977 }
978 }
979
8dcf39c4 980 c = *path++;
10ecfa76 981 if ((c == '.' && !verify_dotfile(path, mode)) ||
3bdf09c7
JH
982 is_dir_sep(c) || c == '\0')
983 return 0;
8dcf39c4
LT
984 }
985 c = *path++;
986 }
987}
988
12676608
LT
989/*
990 * Do we have another file that has the beginning components being a
991 * proper superset of the name we're trying to add?
0f1e4f04 992 */
4aab5b46
JH
993static int has_file_name(struct index_state *istate,
994 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 995{
12676608
LT
996 int retval = 0;
997 int len = ce_namelen(ce);
b155725d 998 int stage = ce_stage(ce);
12676608 999 const char *name = ce->name;
0f1e4f04 1000
4aab5b46
JH
1001 while (pos < istate->cache_nr) {
1002 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 1003
12676608 1004 if (len >= ce_namelen(p))
0f1e4f04 1005 break;
12676608
LT
1006 if (memcmp(name, p->name, len))
1007 break;
b155725d
JH
1008 if (ce_stage(p) != stage)
1009 continue;
12676608
LT
1010 if (p->name[len] != '/')
1011 continue;
7a51ed66 1012 if (p->ce_flags & CE_REMOVE)
21cd8d00 1013 continue;
12676608
LT
1014 retval = -1;
1015 if (!ok_to_replace)
1016 break;
4aab5b46 1017 remove_index_entry_at(istate, --pos);
0f1e4f04 1018 }
12676608
LT
1019 return retval;
1020}
0f1e4f04 1021
a6db3fbb
JH
1022
1023/*
1024 * Like strcmp(), but also return the offset of the first change.
1025 * If strings are equal, return the length.
1026 */
1027int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1028{
1029 size_t k;
1030
1031 if (!first_change)
1032 return strcmp(s1, s2);
1033
1034 for (k = 0; s1[k] == s2[k]; k++)
1035 if (s1[k] == '\0')
1036 break;
1037
1038 *first_change = k;
1039 return (unsigned char)s1[k] - (unsigned char)s2[k];
1040}
1041
12676608
LT
1042/*
1043 * Do we have another file with a pathname that is a proper
1044 * subset of the name we're trying to add?
06b6d81b
JH
1045 *
1046 * That is, is there another file in the index with a path
1047 * that matches a sub-directory in the given entry?
12676608 1048 */
4aab5b46
JH
1049static int has_dir_name(struct index_state *istate,
1050 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
1051{
1052 int retval = 0;
b155725d 1053 int stage = ce_stage(ce);
12676608
LT
1054 const char *name = ce->name;
1055 const char *slash = name + ce_namelen(ce);
06b6d81b
JH
1056 size_t len_eq_last;
1057 int cmp_last = 0;
1058
1059 /*
1060 * We are frequently called during an iteration on a sorted
1061 * list of pathnames and while building a new index. Therefore,
1062 * there is a high probability that this entry will eventually
1063 * be appended to the index, rather than inserted in the middle.
1064 * If we can confirm that, we can avoid binary searches on the
1065 * components of the pathname.
1066 *
1067 * Compare the entry's full path with the last path in the index.
1068 */
1069 if (istate->cache_nr > 0) {
1070 cmp_last = strcmp_offset(name,
1071 istate->cache[istate->cache_nr - 1]->name,
1072 &len_eq_last);
1073 if (cmp_last > 0) {
1074 if (len_eq_last == 0) {
1075 /*
1076 * The entry sorts AFTER the last one in the
1077 * index and their paths have no common prefix,
1078 * so there cannot be a F/D conflict.
1079 */
1080 return retval;
1081 } else {
1082 /*
1083 * The entry sorts AFTER the last one in the
1084 * index, but has a common prefix. Fall through
1085 * to the loop below to disect the entry's path
1086 * and see where the difference is.
1087 */
1088 }
1089 } else if (cmp_last == 0) {
1090 /*
1091 * The entry exactly matches the last one in the
1092 * index, but because of multiple stage and CE_REMOVE
1093 * items, we fall through and let the regular search
1094 * code handle it.
1095 */
1096 }
1097 }
0f1e4f04 1098
12676608 1099 for (;;) {
b986df5c 1100 size_t len;
0f1e4f04 1101
12676608
LT
1102 for (;;) {
1103 if (*--slash == '/')
1104 break;
1105 if (slash <= ce->name)
1106 return retval;
1107 }
1108 len = slash - name;
0f1e4f04 1109
b986df5c
JH
1110 if (cmp_last > 0) {
1111 /*
1112 * (len + 1) is a directory boundary (including
1113 * the trailing slash). And since the loop is
1114 * decrementing "slash", the first iteration is
1115 * the longest directory prefix; subsequent
1116 * iterations consider parent directories.
1117 */
1118
1119 if (len + 1 <= len_eq_last) {
1120 /*
1121 * The directory prefix (including the trailing
1122 * slash) also appears as a prefix in the last
1123 * entry, so the remainder cannot collide (because
1124 * strcmp said the whole path was greater).
1125 *
1126 * EQ: last: xxx/A
1127 * this: xxx/B
1128 *
1129 * LT: last: xxx/file_A
1130 * this: xxx/file_B
1131 */
1132 return retval;
1133 }
1134
1135 if (len > len_eq_last) {
1136 /*
1137 * This part of the directory prefix (excluding
1138 * the trailing slash) is longer than the known
1139 * equal portions, so this sub-directory cannot
1140 * collide with a file.
1141 *
1142 * GT: last: xxxA
1143 * this: xxxB/file
1144 */
1145 return retval;
1146 }
1147
1148 if (istate->cache_nr > 0 &&
1149 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1150 /*
1151 * The directory prefix lines up with part of
1152 * a longer file or directory name, but sorts
1153 * after it, so this sub-directory cannot
1154 * collide with a file.
1155 *
1156 * last: xxx/yy-file (because '-' sorts before '/')
1157 * this: xxx/yy/abc
1158 */
1159 return retval;
1160 }
1161
1162 /*
1163 * This is a possible collision. Fall through and
1164 * let the regular search code handle it.
1165 *
1166 * last: xxx
1167 * this: xxx/file
1168 */
1169 }
1170
b60e188c 1171 pos = index_name_stage_pos(istate, name, len, stage);
12676608 1172 if (pos >= 0) {
21cd8d00
JH
1173 /*
1174 * Found one, but not so fast. This could
1175 * be a marker that says "I was here, but
1176 * I am being removed". Such an entry is
1177 * not a part of the resulting tree, and
1178 * it is Ok to have a directory at the same
1179 * path.
1180 */
077c48df 1181 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
1182 retval = -1;
1183 if (!ok_to_replace)
1184 break;
4aab5b46 1185 remove_index_entry_at(istate, pos);
21cd8d00
JH
1186 continue;
1187 }
12676608 1188 }
21cd8d00
JH
1189 else
1190 pos = -pos-1;
12676608
LT
1191
1192 /*
1193 * Trivial optimization: if we find an entry that
1194 * already matches the sub-directory, then we know
b155725d 1195 * we're ok, and we can exit.
12676608 1196 */
4aab5b46
JH
1197 while (pos < istate->cache_nr) {
1198 struct cache_entry *p = istate->cache[pos];
b155725d
JH
1199 if ((ce_namelen(p) <= len) ||
1200 (p->name[len] != '/') ||
1201 memcmp(p->name, name, len))
1202 break; /* not our subdirectory */
077c48df
JH
1203 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1204 /*
1205 * p is at the same stage as our entry, and
b155725d
JH
1206 * is a subdirectory of what we are looking
1207 * at, so we cannot have conflicts at our
1208 * level or anything shorter.
1209 */
1210 return retval;
1211 pos++;
192268c1 1212 }
0f1e4f04 1213 }
12676608
LT
1214 return retval;
1215}
1216
1217/* We may be in a situation where we already have path/file and path
1218 * is being added, or we already have path and path/file is being
1219 * added. Either one would result in a nonsense tree that has path
1220 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 1221 *
12676608
LT
1222 * If ok-to-replace is specified, we remove the conflicting entries
1223 * from the cache so the caller should recompute the insert position.
1224 * When this happens, we return non-zero.
1225 */
4aab5b46
JH
1226static int check_file_directory_conflict(struct index_state *istate,
1227 const struct cache_entry *ce,
1228 int pos, int ok_to_replace)
12676608 1229{
21cd8d00
JH
1230 int retval;
1231
1232 /*
1233 * When ce is an "I am going away" entry, we allow it to be added
1234 */
7a51ed66 1235 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
1236 return 0;
1237
12676608
LT
1238 /*
1239 * We check if the path is a sub-path of a subsequent pathname
1240 * first, since removing those will not change the position
21cd8d00 1241 * in the array.
12676608 1242 */
4aab5b46 1243 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 1244
12676608
LT
1245 /*
1246 * Then check if the path might have a clashing sub-directory
1247 * before it.
1248 */
4aab5b46 1249 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
1250}
1251
af3785dc 1252static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
1253{
1254 int pos;
192268c1
JH
1255 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1256 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 1257 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 1258 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 1259
ce7c614b
NTND
1260 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1261 cache_tree_invalidate_path(istate, ce->name);
e5494631
JH
1262
1263 /*
1264 * If this entry's path sorts after the last entry in the index,
1265 * we can avoid searching for it.
1266 */
1267 if (istate->cache_nr > 0 &&
1268 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1269 pos = -istate->cache_nr - 1;
1270 else
1271 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
197ee8c9 1272
3e09cdfd 1273 /* existing match? Just replace it. */
76e7f4ec 1274 if (pos >= 0) {
39425819
JH
1275 if (!new_only)
1276 replace_index_entry(istate, pos, ce);
197ee8c9
LT
1277 return 0;
1278 }
76e7f4ec 1279 pos = -pos-1;
197ee8c9 1280
ffcc9ba7
NTND
1281 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1282 untracked_cache_add_to_index(istate, ce->name);
e931371a 1283
7b937ca3
LT
1284 /*
1285 * Inserting a merged entry ("stage 0") into the index
1286 * will always replace all non-merged entries..
1287 */
4aab5b46
JH
1288 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1289 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 1290 ok_to_add = 1;
4aab5b46 1291 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
1292 break;
1293 }
1294 }
1295
121481ab
LT
1296 if (!ok_to_add)
1297 return -1;
10ecfa76 1298 if (!verify_path(ce->name, ce->ce_mode))
9d0a9e90 1299 return error(_("invalid path '%s'"), ce->name);
121481ab 1300
3e09cdfd 1301 if (!skip_df_check &&
4aab5b46 1302 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 1303 if (!ok_to_replace)
9d0a9e90 1304 return error(_("'%s' appears as both a file and as a directory"),
4aab5b46 1305 ce->name);
b60e188c 1306 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
192268c1
JH
1307 pos = -pos-1;
1308 }
af3785dc
JH
1309 return pos + 1;
1310}
1311
1312int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1313{
1314 int pos;
1315
1316 if (option & ADD_CACHE_JUST_APPEND)
1317 pos = istate->cache_nr;
1318 else {
1319 int ret;
1320 ret = add_index_entry_with_check(istate, ce, option);
1321 if (ret <= 0)
1322 return ret;
1323 pos = ret - 1;
1324 }
0f1e4f04 1325
197ee8c9 1326 /* Make sure the array is big enough .. */
999f5660 1327 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
197ee8c9
LT
1328
1329 /* Add it in.. */
4aab5b46 1330 istate->cache_nr++;
af3785dc 1331 if (istate->cache_nr > pos + 1)
f919ffeb
SG
1332 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1333 istate->cache_nr - pos - 1);
cf558704 1334 set_index_entry(istate, pos, ce);
e636a7b4 1335 istate->cache_changed |= CE_ENTRY_ADDED;
197ee8c9
LT
1336 return 0;
1337}
1338
405e5b2f
LT
1339/*
1340 * "refresh" does not calculate a new sha1 file or bring the
1341 * cache up-to-date for mode/content changes. But what it
1342 * _does_ do is to "re-match" the stat information of a file
1343 * with the cache, so that you can refresh the cache for a
1344 * file that hasn't been changed but where the stat entry is
1345 * out of date.
1346 *
1347 * For example, you'd want to do this after doing a "git-read-tree",
1348 * to link up the stat cache details with the proper files.
1349 */
4aab5b46 1350static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da 1351 struct cache_entry *ce,
d05e6970
JK
1352 unsigned int options, int *err,
1353 int *changed_ret)
405e5b2f
LT
1354{
1355 struct stat st;
1356 struct cache_entry *updated;
a849735b 1357 int changed;
25762726 1358 int refresh = options & CE_MATCH_REFRESH;
4bd5b7da 1359 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1360 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
2e2e7ec1 1361 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
883e248b 1362 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
405e5b2f 1363
25762726 1364 if (!refresh || ce_uptodate(ce))
eadb5831
JH
1365 return ce;
1366
883e248b
BP
1367 if (!ignore_fsmonitor)
1368 refresh_fsmonitor(istate);
aa9349d4 1369 /*
56cac48c
NTND
1370 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1371 * that the change to the work tree does not matter and told
1372 * us not to worry.
aa9349d4 1373 */
56cac48c
NTND
1374 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1375 ce_mark_uptodate(ce);
1376 return ce;
1377 }
aa9349d4
MSO
1378 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1379 ce_mark_uptodate(ce);
1380 return ce;
1381 }
883e248b
BP
1382 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1383 ce_mark_uptodate(ce);
1384 return ce;
1385 }
aa9349d4 1386
ccad42d4
RS
1387 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1388 if (ignore_missing)
1389 return ce;
1390 if (err)
1391 *err = ENOENT;
1392 return NULL;
1393 }
1394
8fd2cb40 1395 if (lstat(ce->name, &st) < 0) {
2e2e7ec1
BK
1396 if (ignore_missing && errno == ENOENT)
1397 return ce;
ec0cc704
JH
1398 if (err)
1399 *err = errno;
8fd2cb40
JS
1400 return NULL;
1401 }
405e5b2f 1402
4bd5b7da 1403 changed = ie_match_stat(istate, ce, &st, options);
d05e6970
JK
1404 if (changed_ret)
1405 *changed_ret = changed;
405e5b2f 1406 if (!changed) {
4bd5b7da
JH
1407 /*
1408 * The path is unchanged. If we were told to ignore
1409 * valid bit, then we did the actual stat check and
1410 * found that the entry is unmodified. If the entry
1411 * is not marked VALID, this is the place to mark it
1412 * valid again, under "assume unchanged" mode.
1413 */
1414 if (ignore_valid && assume_unchanged &&
7a51ed66 1415 !(ce->ce_flags & CE_VALID))
405e5b2f 1416 ; /* mark this one VALID again */
eadb5831
JH
1417 else {
1418 /*
1419 * We do not mark the index itself "modified"
1420 * because CE_UPTODATE flag is in-core only;
1421 * we are not going to write this change out.
1422 */
883e248b 1423 if (!S_ISGITLINK(ce->ce_mode)) {
125fd984 1424 ce_mark_uptodate(ce);
883e248b
BP
1425 mark_fsmonitor_valid(ce);
1426 }
8fd2cb40 1427 return ce;
eadb5831 1428 }
405e5b2f
LT
1429 }
1430
4bd5b7da 1431 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1432 if (err)
1433 *err = EINVAL;
8fd2cb40
JS
1434 return NULL;
1435 }
405e5b2f 1436
a849735b 1437 updated = make_empty_cache_entry(istate, ce_namelen(ce));
0e267b7a
BP
1438 copy_cache_entry(updated, ce);
1439 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
405e5b2f 1440 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1441 /*
1442 * If ignore_valid is not set, we should leave CE_VALID bit
1443 * alone. Otherwise, paths marked with --no-assume-unchanged
1444 * (i.e. things to be edited) will reacquire CE_VALID bit
1445 * automatically, which is not really what we want.
405e5b2f 1446 */
4bd5b7da 1447 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1448 !(ce->ce_flags & CE_VALID))
1449 updated->ce_flags &= ~CE_VALID;
405e5b2f 1450
e636a7b4 1451 /* istate->cache_changed is updated in the caller */
405e5b2f
LT
1452 return updated;
1453}
1454
3deffc52 1455static void show_file(const char * fmt, const char * name, int in_porcelain,
046613c5 1456 int * first, const char *header_msg)
3deffc52
MM
1457{
1458 if (in_porcelain && *first && header_msg) {
1459 printf("%s\n", header_msg);
cd2b8ae9 1460 *first = 0;
3deffc52
MM
1461 }
1462 printf(fmt, name);
1463}
1464
9b2d6149
NTND
1465int refresh_index(struct index_state *istate, unsigned int flags,
1466 const struct pathspec *pathspec,
046613c5 1467 char *seen, const char *header_msg)
405e5b2f
LT
1468{
1469 int i;
1470 int has_errors = 0;
1471 int really = (flags & REFRESH_REALLY) != 0;
1472 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1473 int quiet = (flags & REFRESH_QUIET) != 0;
1474 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1475 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1476 int first = 1;
1477 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
25762726
BK
1478 unsigned int options = (CE_MATCH_REFRESH |
1479 (really ? CE_MATCH_IGNORE_VALID : 0) |
2e2e7ec1 1480 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
4bd4e730 1481 const char *modified_fmt;
73b7eae6
JK
1482 const char *deleted_fmt;
1483 const char *typechange_fmt;
1484 const char *added_fmt;
4bd4e730 1485 const char *unmerged_fmt;
ae9af122
NTND
1486 struct progress *progress = NULL;
1487
1488 if (flags & REFRESH_PROGRESS && isatty(2))
1489 progress = start_delayed_progress(_("Refresh index"),
1490 istate->cache_nr);
405e5b2f 1491
c46c406a 1492 trace_performance_enter();
a71806a7
NTND
1493 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1494 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1495 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1496 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1497 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
99ce720c
BP
1498 /*
1499 * Use the multi-threaded preload_index() to refresh most of the
1500 * cache entries quickly then in the single threaded loop below,
1501 * we only have to do the special cases that are left.
1502 */
1503 preload_index(istate, pathspec, 0);
4aab5b46 1504 for (i = 0; i < istate->cache_nr; i++) {
285c2e25 1505 struct cache_entry *ce, *new_entry;
ec0cc704 1506 int cache_errno = 0;
73b7eae6 1507 int changed = 0;
3d1f148c 1508 int filtered = 0;
ec0cc704 1509
4aab5b46 1510 ce = istate->cache[i];
5fdeacb0
JS
1511 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1512 continue;
1513
d7b665c3 1514 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
3d1f148c
JH
1515 filtered = 1;
1516
405e5b2f 1517 if (ce_stage(ce)) {
4aab5b46
JH
1518 while ((i < istate->cache_nr) &&
1519 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1520 i++;
1521 i--;
1522 if (allow_unmerged)
1523 continue;
3d1f148c
JH
1524 if (!filtered)
1525 show_file(unmerged_fmt, ce->name, in_porcelain,
1526 &first, header_msg);
405e5b2f
LT
1527 has_errors = 1;
1528 continue;
1529 }
1530
3d1f148c 1531 if (filtered)
d616813d
AJ
1532 continue;
1533
285c2e25
BW
1534 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1535 if (new_entry == ce)
405e5b2f 1536 continue;
ae9af122
NTND
1537 if (progress)
1538 display_progress(progress, i);
285c2e25 1539 if (!new_entry) {
73b7eae6
JK
1540 const char *fmt;
1541
8fd2cb40 1542 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1543 /* If we are doing --really-refresh that
1544 * means the index is not valid anymore.
1545 */
7a51ed66 1546 ce->ce_flags &= ~CE_VALID;
078a58e8 1547 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 1548 mark_fsmonitor_invalid(istate, ce);
e636a7b4 1549 istate->cache_changed |= CE_ENTRY_CHANGED;
405e5b2f
LT
1550 }
1551 if (quiet)
1552 continue;
73b7eae6
JK
1553
1554 if (cache_errno == ENOENT)
1555 fmt = deleted_fmt;
895ff3b2 1556 else if (ce_intent_to_add(ce))
73b7eae6
JK
1557 fmt = added_fmt; /* must be before other checks */
1558 else if (changed & TYPE_CHANGED)
1559 fmt = typechange_fmt;
1560 else
1561 fmt = modified_fmt;
1562 show_file(fmt,
1563 ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1564 has_errors = 1;
1565 continue;
1566 }
cf558704 1567
285c2e25 1568 replace_index_entry(istate, i, new_entry);
405e5b2f 1569 }
ae9af122
NTND
1570 if (progress) {
1571 display_progress(progress, istate->cache_nr);
1572 stop_progress(&progress);
1573 }
c46c406a 1574 trace_performance_leave("refresh index");
405e5b2f
LT
1575 return has_errors;
1576}
1577
768d7965
JM
1578struct cache_entry *refresh_cache_entry(struct index_state *istate,
1579 struct cache_entry *ce,
1580 unsigned int options)
ec0cc704 1581{
768d7965 1582 return refresh_cache_ent(istate, ce, options, NULL, NULL);
ec0cc704
JH
1583}
1584
db3b313c
JH
1585
1586/*****************************************************************
1587 * Index File I/O
1588 *****************************************************************/
1589
9d227781
JH
1590#define INDEX_FORMAT_DEFAULT 3
1591
136347d7
TG
1592static unsigned int get_index_format_default(void)
1593{
1594 char *envversion = getenv("GIT_INDEX_VERSION");
3c09d684 1595 char *endp;
b27a5720 1596 int value;
3c09d684
TG
1597 unsigned int version = INDEX_FORMAT_DEFAULT;
1598
136347d7 1599 if (!envversion) {
b27a5720
TA
1600 if (!git_config_get_int("index.version", &value))
1601 version = value;
3c09d684
TG
1602 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1603 warning(_("index.version set, but the value is invalid.\n"
136347d7 1604 "Using version %i"), INDEX_FORMAT_DEFAULT);
3c09d684 1605 return INDEX_FORMAT_DEFAULT;
136347d7
TG
1606 }
1607 return version;
1608 }
3c09d684
TG
1609
1610 version = strtoul(envversion, &endp, 10);
1611 if (*endp ||
1612 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1613 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1614 "Using version %i"), INDEX_FORMAT_DEFAULT);
1615 version = INDEX_FORMAT_DEFAULT;
1616 }
1617 return version;
136347d7
TG
1618}
1619
db3b313c
JH
1620/*
1621 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1622 * Again - this is just a (very strong in practice) heuristic that
1623 * the inode hasn't changed.
1624 *
1625 * We save the fields in big-endian order to allow using the
1626 * index file over NFS transparently.
1627 */
1628struct ondisk_cache_entry {
1629 struct cache_time ctime;
1630 struct cache_time mtime;
7800c1eb
TG
1631 uint32_t dev;
1632 uint32_t ino;
1633 uint32_t mode;
1634 uint32_t uid;
1635 uint32_t gid;
1636 uint32_t size;
db3b313c 1637 unsigned char sha1[20];
7800c1eb 1638 uint16_t flags;
db3b313c
JH
1639 char name[FLEX_ARRAY]; /* more */
1640};
1641
1642/*
1643 * This struct is used when CE_EXTENDED bit is 1
1644 * The struct must match ondisk_cache_entry exactly from
1645 * ctime till flags
1646 */
1647struct ondisk_cache_entry_extended {
1648 struct cache_time ctime;
1649 struct cache_time mtime;
7800c1eb
TG
1650 uint32_t dev;
1651 uint32_t ino;
1652 uint32_t mode;
1653 uint32_t uid;
1654 uint32_t gid;
1655 uint32_t size;
db3b313c 1656 unsigned char sha1[20];
7800c1eb
TG
1657 uint16_t flags;
1658 uint16_t flags2;
db3b313c
JH
1659 char name[FLEX_ARRAY]; /* more */
1660};
1661
6c9cd161 1662/* These are only used for v3 or lower */
ce012deb 1663#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
db3b313c
JH
1664#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1665#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1666#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1667#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1668 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1669 ondisk_cache_entry_size(ce_namelen(ce)))
1670
a33fc72f
JH
1671/* Allow fsck to force verification of the index checksum. */
1672int verify_index_checksum;
1673
00ec50e5
BP
1674/* Allow fsck to force verification of the cache entry order. */
1675int verify_ce_order;
1676
371ed0de 1677static int verify_hdr(const struct cache_header *hdr, unsigned long size)
e83c5163 1678{
aab61359 1679 git_hash_ctx c;
1680 unsigned char hash[GIT_MAX_RAWSZ];
0136bac9 1681 int hdr_version;
e83c5163 1682
ccc4feb5 1683 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
9d0a9e90 1684 return error(_("bad signature 0x%08x"), hdr->hdr_signature);
0136bac9 1685 hdr_version = ntohl(hdr->hdr_version);
b82a7b5b 1686 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
9d0a9e90 1687 return error(_("bad index version %d"), hdr_version);
a33fc72f
JH
1688
1689 if (!verify_index_checksum)
1690 return 0;
1691
aab61359 1692 the_hash_algo->init_fn(&c);
1693 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1694 the_hash_algo->final_fn(hash, &c);
67947c34 1695 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
9d0a9e90 1696 return error(_("bad index file sha1 signature"));
e83c5163
LT
1697 return 0;
1698}
1699
4aab5b46 1700static int read_index_extension(struct index_state *istate,
371ed0de 1701 const char *ext, const char *data, unsigned long sz)
bad68ec9
JH
1702{
1703 switch (CACHE_EXT(ext)) {
1704 case CACHE_EXT_TREE:
4aab5b46 1705 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1706 break;
cfc5789a
JH
1707 case CACHE_EXT_RESOLVE_UNDO:
1708 istate->resolve_undo = resolve_undo_read(data, sz);
1709 break;
5fc2fc8f
NTND
1710 case CACHE_EXT_LINK:
1711 if (read_link_extension(istate, data, sz))
1712 return -1;
1713 break;
f9e6c649
NTND
1714 case CACHE_EXT_UNTRACKED:
1715 istate->untracked = read_untracked_extension(data, sz);
1716 break;
883e248b
BP
1717 case CACHE_EXT_FSMONITOR:
1718 read_fsmonitor_extension(istate, data, sz);
1719 break;
3b1d9e04 1720 case CACHE_EXT_ENDOFINDEXENTRIES:
3255089a 1721 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
3b1d9e04
BP
1722 /* already handled in do_read_index() */
1723 break;
bad68ec9
JH
1724 default:
1725 if (*ext < 'A' || 'Z' < *ext)
9d0a9e90 1726 return error(_("index uses %.4s extension, which we do not understand"),
bad68ec9 1727 ext);
9d0a9e90 1728 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
bad68ec9
JH
1729 break;
1730 }
1731 return 0;
1732}
1733
77ff1127
BP
1734static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1735 unsigned int version,
a849735b 1736 struct ondisk_cache_entry *ondisk,
6c9cd161 1737 unsigned long *ent_size,
252d079c 1738 const struct cache_entry *previous_ce)
7a51ed66 1739{
debed2a6 1740 struct cache_entry *ce;
7fec10b7 1741 size_t len;
06aaaa0b 1742 const char *name;
debed2a6 1743 unsigned int flags;
f5c4a9af 1744 size_t copy_len = 0;
252d079c
NTND
1745 /*
1746 * Adjacent cache entries tend to share the leading paths, so it makes
1747 * sense to only store the differences in later entries. In the v4
1748 * on-disk format of the index, each on-disk cache entry stores the
1749 * number of bytes to be stripped from the end of the previous name,
1750 * and the bytes to append to the result, to come up with its name.
1751 */
77ff1127 1752 int expand_name_field = version == 4;
7fec10b7 1753
7a51ed66 1754 /* On-disk flags are just 16 bits */
c3d8da57 1755 flags = get_be16(&ondisk->flags);
debed2a6 1756 len = flags & CE_NAMEMASK;
7fec10b7 1757
debed2a6 1758 if (flags & CE_EXTENDED) {
06aaaa0b
NTND
1759 struct ondisk_cache_entry_extended *ondisk2;
1760 int extended_flags;
1761 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
c3d8da57 1762 extended_flags = get_be16(&ondisk2->flags2) << 16;
06aaaa0b
NTND
1763 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1764 if (extended_flags & ~CE_EXTENDED_FLAGS)
9d0a9e90 1765 die(_("unknown index entry format 0x%08x"), extended_flags);
debed2a6 1766 flags |= extended_flags;
06aaaa0b
NTND
1767 name = ondisk2->name;
1768 }
1769 else
1770 name = ondisk->name;
1771
252d079c
NTND
1772 if (expand_name_field) {
1773 const unsigned char *cp = (const unsigned char *)name;
1774 size_t strip_len, previous_len;
6c9cd161 1775
77ff1127 1776 /* If we're at the begining of a block, ignore the previous name */
252d079c 1777 strip_len = decode_varint(&cp);
77ff1127
BP
1778 if (previous_ce) {
1779 previous_len = previous_ce->ce_namelen;
1780 if (previous_len < strip_len)
252d079c 1781 die(_("malformed name field in the index, near path '%s'"),
77ff1127
BP
1782 previous_ce->name);
1783 copy_len = previous_len - strip_len;
252d079c 1784 }
252d079c
NTND
1785 name = (const char *)cp;
1786 }
6c9cd161 1787
252d079c
NTND
1788 if (len == CE_NAMEMASK) {
1789 len = strlen(name);
1790 if (expand_name_field)
1791 len += copy_len;
1792 }
6c9cd161 1793
77ff1127 1794 ce = mem_pool__ce_alloc(ce_mem_pool, len);
252d079c
NTND
1795
1796 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1797 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1798 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1799 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1800 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1801 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1802 ce->ce_mode = get_be32(&ondisk->mode);
1803 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1804 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1805 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1806 ce->ce_flags = flags & ~CE_NAMEMASK;
1807 ce->ce_namelen = len;
1808 ce->index = 0;
1809 hashcpy(ce->oid.hash, ondisk->sha1);
6c9cd161 1810
252d079c
NTND
1811 if (expand_name_field) {
1812 if (copy_len)
1813 memcpy(ce->name, previous_ce->name, copy_len);
1814 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1815 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1816 } else {
1817 memcpy(ce->name, name, len + 1);
1818 *ent_size = ondisk_ce_size(ce);
6c9cd161 1819 }
debed2a6 1820 return ce;
cf558704
LT
1821}
1822
03f15a79 1823static void check_ce_order(struct index_state *istate)
15999d0b 1824{
03f15a79
TG
1825 unsigned int i;
1826
00ec50e5
BP
1827 if (!verify_ce_order)
1828 return;
1829
03f15a79
TG
1830 for (i = 1; i < istate->cache_nr; i++) {
1831 struct cache_entry *ce = istate->cache[i - 1];
1832 struct cache_entry *next_ce = istate->cache[i];
1833 int name_compare = strcmp(ce->name, next_ce->name);
1834
1835 if (0 < name_compare)
9d0a9e90 1836 die(_("unordered stage entries in index"));
03f15a79
TG
1837 if (!name_compare) {
1838 if (!ce_stage(ce))
9d0a9e90 1839 die(_("multiple stage entries for merged file '%s'"),
03f15a79
TG
1840 ce->name);
1841 if (ce_stage(ce) > ce_stage(next_ce))
9d0a9e90 1842 die(_("unordered stage entries for '%s'"),
03f15a79
TG
1843 ce->name);
1844 }
15999d0b
JSP
1845 }
1846}
1847
435ec090
CC
1848static void tweak_untracked_cache(struct index_state *istate)
1849{
1850 switch (git_config_get_untracked_cache()) {
1851 case -1: /* keep: do nothing */
1852 break;
1853 case 0: /* false */
1854 remove_untracked_cache(istate);
1855 break;
1856 case 1: /* true */
1857 add_untracked_cache(istate);
1858 break;
1859 default: /* unknown value: do nothing */
1860 break;
1861 }
1862}
1863
43925312
CC
1864static void tweak_split_index(struct index_state *istate)
1865{
1866 switch (git_config_get_split_index()) {
1867 case -1: /* unset: do nothing */
1868 break;
1869 case 0: /* false */
1870 remove_split_index(istate);
1871 break;
1872 case 1: /* true */
1873 add_split_index(istate);
1874 break;
1875 default: /* unknown value: do nothing */
1876 break;
1877 }
1878}
1879
435ec090
CC
1880static void post_read_index_from(struct index_state *istate)
1881{
1882 check_ce_order(istate);
1883 tweak_untracked_cache(istate);
43925312 1884 tweak_split_index(istate);
883e248b 1885 tweak_fsmonitor(istate);
435ec090
CC
1886}
1887
8e72d675
JM
1888static size_t estimate_cache_size_from_compressed(unsigned int entries)
1889{
1890 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1891}
1892
1893static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1894{
1895 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1896
1897 /*
1898 * Account for potential alignment differences.
1899 */
1900 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1901 return ondisk_size + entries * per_entry;
1902}
1903
3255089a
BP
1904struct index_entry_offset
1905{
1906 /* starting byte offset into index file, count of index entries in this block */
1907 int offset, nr;
1908};
1909
1910struct index_entry_offset_table
1911{
1912 int nr;
1913 struct index_entry_offset entries[FLEX_ARRAY];
1914};
1915
3255089a
BP
1916static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1917static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
3255089a 1918
3b1d9e04
BP
1919static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1920static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1921
abb4bb83
BP
1922struct load_index_extensions
1923{
abb4bb83 1924 pthread_t pthread;
abb4bb83
BP
1925 struct index_state *istate;
1926 const char *mmap;
1927 size_t mmap_size;
1928 unsigned long src_offset;
1929};
1930
1931static void *load_index_extensions(void *_data)
1932{
1933 struct load_index_extensions *p = _data;
1934 unsigned long src_offset = p->src_offset;
1935
1936 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1937 /* After an array of active_nr index entries,
1938 * there can be arbitrary number of extended
1939 * sections, each of which is prefixed with
1940 * extension name (4-byte) and section length
1941 * in 4-byte network byte order.
1942 */
1943 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1944 if (read_index_extension(p->istate,
1945 p->mmap + src_offset,
1946 p->mmap + src_offset + 8,
1947 extsize) < 0) {
1948 munmap((void *)p->mmap, p->mmap_size);
1949 die(_("index file corrupt"));
1950 }
1951 src_offset += 8;
1952 src_offset += extsize;
1953 }
1954
1955 return NULL;
1956}
1957
77ff1127
BP
1958/*
1959 * A helper function that will load the specified range of cache entries
1960 * from the memory mapped file and add them to the given index.
1961 */
1962static unsigned long load_cache_entry_block(struct index_state *istate,
1963 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1964 unsigned long start_offset, const struct cache_entry *previous_ce)
1965{
1966 int i;
1967 unsigned long src_offset = start_offset;
1968
1969 for (i = offset; i < offset + nr; i++) {
1970 struct ondisk_cache_entry *disk_ce;
1971 struct cache_entry *ce;
1972 unsigned long consumed;
1973
1974 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1975 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1976 set_index_entry(istate, i, ce);
1977
1978 src_offset += consumed;
1979 previous_ce = ce;
1980 }
1981 return src_offset - start_offset;
1982}
1983
1984static unsigned long load_all_cache_entries(struct index_state *istate,
1985 const char *mmap, size_t mmap_size, unsigned long src_offset)
1986{
1987 unsigned long consumed;
1988
1989 if (istate->version == 4) {
1990 mem_pool_init(&istate->ce_mem_pool,
1991 estimate_cache_size_from_compressed(istate->cache_nr));
1992 } else {
1993 mem_pool_init(&istate->ce_mem_pool,
1994 estimate_cache_size(mmap_size, istate->cache_nr));
1995 }
1996
1997 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
1998 0, istate->cache_nr, mmap, src_offset, NULL);
1999 return consumed;
2000}
2001
3255089a
BP
2002/*
2003 * Mostly randomly chosen maximum thread counts: we
2004 * cap the parallelism to online_cpus() threads, and we want
2005 * to have at least 10000 cache entries per thread for it to
2006 * be worth starting a thread.
2007 */
2008
2009#define THREAD_COST (10000)
2010
77ff1127
BP
2011struct load_cache_entries_thread_data
2012{
2013 pthread_t pthread;
2014 struct index_state *istate;
2015 struct mem_pool *ce_mem_pool;
2016 int offset;
2017 const char *mmap;
2018 struct index_entry_offset_table *ieot;
2019 int ieot_start; /* starting index into the ieot array */
2020 int ieot_blocks; /* count of ieot entries to process */
2021 unsigned long consumed; /* return # of bytes in index file processed */
2022};
2023
2024/*
2025 * A thread proc to run the load_cache_entries() computation
2026 * across multiple background threads.
2027 */
2028static void *load_cache_entries_thread(void *_data)
2029{
2030 struct load_cache_entries_thread_data *p = _data;
2031 int i;
2032
2033 /* iterate across all ieot blocks assigned to this thread */
2034 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2035 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2036 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2037 p->offset += p->ieot->entries[i].nr;
2038 }
2039 return NULL;
2040}
2041
2042static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2043 unsigned long src_offset, int nr_threads, struct index_entry_offset_table *ieot)
2044{
2045 int i, offset, ieot_blocks, ieot_start, err;
2046 struct load_cache_entries_thread_data *data;
2047 unsigned long consumed = 0;
2048
2049 /* a little sanity checking */
2050 if (istate->name_hash_initialized)
2051 BUG("the name hash isn't thread safe");
2052
2053 mem_pool_init(&istate->ce_mem_pool, 0);
2054
2055 /* ensure we have no more threads than we have blocks to process */
2056 if (nr_threads > ieot->nr)
2057 nr_threads = ieot->nr;
2058 data = xcalloc(nr_threads, sizeof(*data));
2059
2060 offset = ieot_start = 0;
2061 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2062 for (i = 0; i < nr_threads; i++) {
2063 struct load_cache_entries_thread_data *p = &data[i];
2064 int nr, j;
2065
2066 if (ieot_start + ieot_blocks > ieot->nr)
2067 ieot_blocks = ieot->nr - ieot_start;
2068
2069 p->istate = istate;
2070 p->offset = offset;
2071 p->mmap = mmap;
2072 p->ieot = ieot;
2073 p->ieot_start = ieot_start;
2074 p->ieot_blocks = ieot_blocks;
2075
2076 /* create a mem_pool for each thread */
2077 nr = 0;
2078 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2079 nr += p->ieot->entries[j].nr;
2080 if (istate->version == 4) {
2081 mem_pool_init(&p->ce_mem_pool,
2082 estimate_cache_size_from_compressed(nr));
2083 } else {
2084 mem_pool_init(&p->ce_mem_pool,
2085 estimate_cache_size(mmap_size, nr));
2086 }
2087
2088 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2089 if (err)
2090 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2091
2092 /* increment by the number of cache entries in the ieot block being processed */
2093 for (j = 0; j < ieot_blocks; j++)
2094 offset += ieot->entries[ieot_start + j].nr;
2095 ieot_start += ieot_blocks;
2096 }
2097
2098 for (i = 0; i < nr_threads; i++) {
2099 struct load_cache_entries_thread_data *p = &data[i];
2100
2101 err = pthread_join(p->pthread, NULL);
2102 if (err)
2103 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2104 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2105 consumed += p->consumed;
2106 }
2107
2108 free(data);
2109
2110 return consumed;
2111}
77ff1127 2112
8fd2cb40 2113/* remember to discard_cache() before reading a different cache! */
3e52f70b 2114int do_read_index(struct index_state *istate, const char *path, int must_exist)
e83c5163 2115{
77ff1127 2116 int fd;
e83c5163 2117 struct stat st;
debed2a6 2118 unsigned long src_offset;
371ed0de
BP
2119 const struct cache_header *hdr;
2120 const char *mmap;
7a51ed66 2121 size_t mmap_size;
abb4bb83
BP
2122 struct load_index_extensions p;
2123 size_t extension_offset = 0;
77ff1127
BP
2124 int nr_threads, cpus;
2125 struct index_entry_offset_table *ieot = NULL;
e83c5163 2126
913e0e99 2127 if (istate->initialized)
4aab5b46 2128 return istate->cache_nr;
5d1a5c02 2129
fba2f38a
KB
2130 istate->timestamp.sec = 0;
2131 istate->timestamp.nsec = 0;
8fd2cb40 2132 fd = open(path, O_RDONLY);
5d1a5c02 2133 if (fd < 0) {
5fc2fc8f 2134 if (!must_exist && errno == ENOENT)
5d1a5c02 2135 return 0;
9d0a9e90 2136 die_errno(_("%s: index file open failed"), path);
5d1a5c02 2137 }
e83c5163 2138
3511a377 2139 if (fstat(fd, &st))
9d0a9e90 2140 die_errno(_("%s: cannot stat the open index"), path);
3511a377 2141
7a51ed66 2142 mmap_size = xsize_t(st.st_size);
aab61359 2143 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
9d0a9e90 2144 die(_("%s: index file smaller than expected"), path);
3511a377 2145
a1293ef7 2146 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
7a51ed66 2147 if (mmap == MAP_FAILED)
9d0a9e90 2148 die_errno(_("%s: unable to map index file"), path);
57d84f8d 2149 close(fd);
e83c5163 2150
371ed0de 2151 hdr = (const struct cache_header *)mmap;
7a51ed66 2152 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
2153 goto unmap;
2154
75691ea3 2155 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
9d227781 2156 istate->version = ntohl(hdr->hdr_version);
4aab5b46
JH
2157 istate->cache_nr = ntohl(hdr->hdr_entries);
2158 istate->cache_alloc = alloc_nr(istate->cache_nr);
c4aa3167 2159 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
913e0e99 2160 istate->initialized = 1;
7a51ed66 2161
abb4bb83
BP
2162 p.istate = istate;
2163 p.mmap = mmap;
2164 p.mmap_size = mmap_size;
6c9cd161 2165
7a51ed66 2166 src_offset = sizeof(*hdr);
4aab5b46 2167
2a9dedef
JN
2168 if (git_config_get_index_threads(&nr_threads))
2169 nr_threads = 1;
7a51ed66 2170
77ff1127
BP
2171 /* TODO: does creating more threads than cores help? */
2172 if (!nr_threads) {
2173 nr_threads = istate->cache_nr / THREAD_COST;
2174 cpus = online_cpus();
2175 if (nr_threads > cpus)
2176 nr_threads = cpus;
e83c5163 2177 }
abb4bb83 2178
88168b9b
NTND
2179 if (!HAVE_THREADS)
2180 nr_threads = 1;
2181
abb4bb83
BP
2182 if (nr_threads > 1) {
2183 extension_offset = read_eoie_extension(mmap, mmap_size);
2184 if (extension_offset) {
2185 int err;
2186
2187 p.src_offset = extension_offset;
2188 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2189 if (err)
2190 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2191
2192 nr_threads--;
2193 }
2194 }
abb4bb83 2195
77ff1127
BP
2196 /*
2197 * Locate and read the index entry offset table so that we can use it
2198 * to multi-thread the reading of the cache entries.
2199 */
2200 if (extension_offset && nr_threads > 1)
2201 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2202
2203 if (ieot) {
2204 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, src_offset, nr_threads, ieot);
2205 free(ieot);
8e72d675 2206 } else {
77ff1127 2207 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
8e72d675 2208 }
6c9cd161 2209
fba2f38a 2210 istate->timestamp.sec = st.st_mtime;
c06ff490 2211 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 2212
abb4bb83 2213 /* if we created a thread, join it otherwise load the extensions on the primary thread */
abb4bb83
BP
2214 if (extension_offset) {
2215 int ret = pthread_join(p.pthread, NULL);
2216 if (ret)
2217 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
88168b9b 2218 } else {
abb4bb83
BP
2219 p.src_offset = src_offset;
2220 load_index_extensions(&p);
bad68ec9 2221 }
371ed0de 2222 munmap((void *)mmap, mmap_size);
4aab5b46 2223 return istate->cache_nr;
e83c5163
LT
2224
2225unmap:
371ed0de 2226 munmap((void *)mmap, mmap_size);
9d0a9e90 2227 die(_("index file corrupt"));
e83c5163
LT
2228}
2229
0d59ffb4
CC
2230/*
2231 * Signal that the shared index is used by updating its mtime.
2232 *
2233 * This way, shared index can be removed if they have not been used
2234 * for some time.
2235 */
a125a223 2236static void freshen_shared_index(const char *shared_index, int warn)
0d59ffb4 2237{
0d59ffb4 2238 if (!check_and_freshen_file(shared_index, 1) && warn)
9d0a9e90 2239 warning(_("could not freshen shared index '%s'"), shared_index);
0d59ffb4
CC
2240}
2241
a125a223
TG
2242int read_index_from(struct index_state *istate, const char *path,
2243 const char *gitdir)
5fc2fc8f
NTND
2244{
2245 struct split_index *split_index;
2246 int ret;
2182abd9 2247 char *base_oid_hex;
a125a223 2248 char *base_path;
5fc2fc8f
NTND
2249
2250 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2251 if (istate->initialized)
2252 return istate->cache_nr;
2253
c46c406a 2254 trace_performance_enter();
5fc2fc8f 2255 ret = do_read_index(istate, path, 0);
c46c406a 2256 trace_performance_leave("read cache %s", path);
435ec090 2257
5fc2fc8f 2258 split_index = istate->split_index;
2182abd9 2259 if (!split_index || is_null_oid(&split_index->base_oid)) {
435ec090 2260 post_read_index_from(istate);
5fc2fc8f 2261 return ret;
03f15a79 2262 }
5fc2fc8f 2263
c46c406a 2264 trace_performance_enter();
5fc2fc8f
NTND
2265 if (split_index->base)
2266 discard_index(split_index->base);
2267 else
2268 split_index->base = xcalloc(1, sizeof(*split_index->base));
de6ae5f9 2269
2182abd9 2270 base_oid_hex = oid_to_hex(&split_index->base_oid);
2271 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
de6ae5f9 2272 ret = do_read_index(split_index->base, base_path, 1);
9001dc2a 2273 if (!oideq(&split_index->base_oid, &split_index->base->oid))
9d0a9e90 2274 die(_("broken index, expect %s in %s, got %s"),
2182abd9 2275 base_oid_hex, base_path,
75691ea3 2276 oid_to_hex(&split_index->base->oid));
de6ae5f9 2277
a125a223 2278 freshen_shared_index(base_path, 0);
5fc2fc8f 2279 merge_base_index(istate);
435ec090 2280 post_read_index_from(istate);
c46c406a 2281 trace_performance_leave("read cache %s", base_path);
b42ad7d5 2282 free(base_path);
5fc2fc8f
NTND
2283 return ret;
2284}
2285
fa7b3c2f
JH
2286int is_index_unborn(struct index_state *istate)
2287{
debed2a6 2288 return (!istate->cache_nr && !istate->timestamp.sec);
fa7b3c2f
JH
2289}
2290
4aab5b46 2291int discard_index(struct index_state *istate)
6d297f81 2292{
8e72d675
JM
2293 /*
2294 * Cache entries in istate->cache[] should have been allocated
2295 * from the memory pool associated with this index, or from an
2296 * associated split_index. There is no need to free individual
8616a2d0
JM
2297 * cache entries. validate_cache_entries can detect when this
2298 * assertion does not hold.
8e72d675 2299 */
8616a2d0 2300 validate_cache_entries(istate);
debed2a6 2301
cfc5789a 2302 resolve_undo_clear_index(istate);
4aab5b46
JH
2303 istate->cache_nr = 0;
2304 istate->cache_changed = 0;
fba2f38a
KB
2305 istate->timestamp.sec = 0;
2306 istate->timestamp.nsec = 0;
2092678c 2307 free_name_hash(istate);
4aab5b46 2308 cache_tree_free(&(istate->cache_tree));
913e0e99 2309 istate->initialized = 0;
6a83d902 2310 FREE_AND_NULL(istate->cache);
a0fc4db0 2311 istate->cache_alloc = 0;
5fc2fc8f 2312 discard_split_index(istate);
f9e6c649
NTND
2313 free_untracked_cache(istate->untracked);
2314 istate->untracked = NULL;
8e72d675
JM
2315
2316 if (istate->ce_mem_pool) {
8616a2d0 2317 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
8e72d675
JM
2318 istate->ce_mem_pool = NULL;
2319 }
2320
7a51ed66 2321 return 0;
6d297f81
JS
2322}
2323
8616a2d0
JM
2324/*
2325 * Validate the cache entries of this index.
2326 * All cache entries associated with this index
2327 * should have been allocated by the memory pool
2328 * associated with this index, or by a referenced
2329 * split index.
2330 */
2331void validate_cache_entries(const struct index_state *istate)
2332{
2333 int i;
2334
2335 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2336 return;
2337
2338 for (i = 0; i < istate->cache_nr; i++) {
2339 if (!istate) {
391408e5 2340 BUG("cache entry is not allocated from expected memory pool");
8616a2d0
JM
2341 } else if (!istate->ce_mem_pool ||
2342 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2343 if (!istate->split_index ||
2344 !istate->split_index->base ||
2345 !istate->split_index->base->ce_mem_pool ||
2346 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
391408e5 2347 BUG("cache entry is not allocated from expected memory pool");
8616a2d0
JM
2348 }
2349 }
2350 }
2351
2352 if (istate->split_index)
2353 validate_cache_entries(istate->split_index->base);
2354}
2355
d1f128b0 2356int unmerged_index(const struct index_state *istate)
94a5728c
DB
2357{
2358 int i;
2359 for (i = 0; i < istate->cache_nr; i++) {
2360 if (ce_stage(istate->cache[i]))
2361 return 1;
2362 }
2363 return 0;
2364}
2365
150fe065
NTND
2366int repo_index_has_changes(struct repository *repo,
2367 struct tree *tree,
2368 struct strbuf *sb)
cffbfad5 2369{
150fe065 2370 struct index_state *istate = repo->index;
e1f8694f 2371 struct object_id cmp;
cffbfad5
EN
2372 int i;
2373
e1f8694f
EN
2374 if (tree)
2375 cmp = tree->object.oid;
2376 if (tree || !get_oid_tree("HEAD", &cmp)) {
cffbfad5
EN
2377 struct diff_options opt;
2378
150fe065 2379 repo_diff_setup(repo, &opt);
cffbfad5
EN
2380 opt.flags.exit_with_status = 1;
2381 if (!sb)
2382 opt.flags.quick = 1;
e1f8694f 2383 do_diff_cache(&cmp, &opt);
cffbfad5
EN
2384 diffcore_std(&opt);
2385 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2386 if (i)
2387 strbuf_addch(sb, ' ');
2388 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2389 }
2390 diff_flush(&opt);
2391 return opt.flags.has_changes != 0;
2392 } else {
1b9fbefb 2393 for (i = 0; sb && i < istate->cache_nr; i++) {
cffbfad5
EN
2394 if (i)
2395 strbuf_addch(sb, ' ');
1b9fbefb 2396 strbuf_addstr(sb, istate->cache[i]->name);
cffbfad5 2397 }
1b9fbefb 2398 return !!istate->cache_nr;
cffbfad5
EN
2399 }
2400}
2401
4990aadc 2402#define WRITE_BUFFER_SIZE 8192
bf0f910d 2403static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
2404static unsigned long write_buffer_len;
2405
aab61359 2406static int ce_write_flush(git_hash_ctx *context, int fd)
6015c28b
JH
2407{
2408 unsigned int buffered = write_buffer_len;
2409 if (buffered) {
aab61359 2410 the_hash_algo->update_fn(context, write_buffer, buffered);
06f46f23 2411 if (write_in_full(fd, write_buffer, buffered) < 0)
6015c28b
JH
2412 return -1;
2413 write_buffer_len = 0;
2414 }
2415 return 0;
2416}
2417
aab61359 2418static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
4990aadc
LT
2419{
2420 while (len) {
2421 unsigned int buffered = write_buffer_len;
2422 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2423 if (partial > len)
2424 partial = len;
2425 memcpy(write_buffer + buffered, data, partial);
2426 buffered += partial;
2427 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
2428 write_buffer_len = buffered;
2429 if (ce_write_flush(context, fd))
4990aadc
LT
2430 return -1;
2431 buffered = 0;
2432 }
2433 write_buffer_len = buffered;
2434 len -= partial;
1d7f171c 2435 data = (char *) data + partial;
a6080a0a
JH
2436 }
2437 return 0;
4990aadc
LT
2438}
2439
3b1d9e04
BP
2440static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2441 int fd, unsigned int ext, unsigned int sz)
bad68ec9
JH
2442{
2443 ext = htonl(ext);
2444 sz = htonl(sz);
3b1d9e04
BP
2445 if (eoie_context) {
2446 the_hash_algo->update_fn(eoie_context, &ext, 4);
2447 the_hash_algo->update_fn(eoie_context, &sz, 4);
2448 }
968a1d65
DR
2449 return ((ce_write(context, fd, &ext, 4) < 0) ||
2450 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
2451}
2452
aab61359 2453static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
4990aadc
LT
2454{
2455 unsigned int left = write_buffer_len;
ca9be054 2456
4990aadc
LT
2457 if (left) {
2458 write_buffer_len = 0;
aab61359 2459 the_hash_algo->update_fn(context, write_buffer, left);
4990aadc 2460 }
ca9be054 2461
aab61359 2462 /* Flush first if not enough space for hash signature */
2463 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
06f46f23 2464 if (write_in_full(fd, write_buffer, left) < 0)
2c865d9a
QH
2465 return -1;
2466 left = 0;
2467 }
2468
aab61359 2469 /* Append the hash signature at the end */
2470 the_hash_algo->final_fn(write_buffer + left, context);
2471 hashcpy(hash, write_buffer + left);
2472 left += the_hash_algo->rawsz;
06f46f23 2473 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
4990aadc
LT
2474}
2475
58bf2a4c
NTND
2476static void ce_smudge_racily_clean_entry(struct index_state *istate,
2477 struct cache_entry *ce)
407c8eb0
JH
2478{
2479 /*
2480 * The only thing we care about in this function is to smudge the
2481 * falsely clean entry due to touch-update-touch race, so we leave
2482 * everything else as they are. We are called for entries whose
c21d39d7 2483 * ce_stat_data.sd_mtime match the index file mtime.
c70115b4
JH
2484 *
2485 * Note that this actually does not do much for gitlinks, for
2486 * which ce_match_stat_basic() always goes to the actual
2487 * contents. The caller checks with is_racy_timestamp() which
2488 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
2489 */
2490 struct stat st;
2491
2492 if (lstat(ce->name, &st) < 0)
2493 return;
2494 if (ce_match_stat_basic(ce, &st))
2495 return;
58bf2a4c 2496 if (ce_modified_check_fs(istate, ce, &st)) {
4b3511b0
JH
2497 /* This is "racily clean"; smudge it. Note that this
2498 * is a tricky code. At first glance, it may appear
2499 * that it can break with this sequence:
2500 *
2501 * $ echo xyzzy >frotz
2502 * $ git-update-index --add frotz
2503 * $ : >frotz
2504 * $ sleep 3
2505 * $ echo filfre >nitfol
2506 * $ git-update-index --add nitfol
2507 *
b7e58b17 2508 * but it does not. When the second update-index runs,
4b3511b0
JH
2509 * it notices that the entry "frotz" has the same timestamp
2510 * as index, and if we were to smudge it by resetting its
2511 * size to zero here, then the object name recorded
2512 * in index is the 6-byte file but the cached stat information
2513 * becomes zero --- which would then match what we would
a6080a0a 2514 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
2515 *
2516 * However, the second update-index, before calling
2517 * this function, notices that the cached size is 6
2518 * bytes and what is on the filesystem is an empty
2519 * file, and never calls us, so the cached size information
2520 * for "frotz" stays 6 which does not match the filesystem.
2521 */
c21d39d7 2522 ce->ce_stat_data.sd_size = 0;
407c8eb0
JH
2523 }
2524}
2525
f136f7bf 2526/* Copy miscellaneous fields but not the name */
ce012deb 2527static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
f136f7bf 2528 struct cache_entry *ce)
7a51ed66 2529{
b60e188c
TG
2530 short flags;
2531
c21d39d7
MH
2532 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2533 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2534 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2535 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2536 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2537 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
7a51ed66 2538 ondisk->mode = htonl(ce->ce_mode);
c21d39d7
MH
2539 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2540 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2541 ondisk->size = htonl(ce->ce_stat_data.sd_size);
99d1a986 2542 hashcpy(ondisk->sha1, ce->oid.hash);
b60e188c 2543
ce51bf09 2544 flags = ce->ce_flags & ~CE_NAMEMASK;
b60e188c
TG
2545 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2546 ondisk->flags = htons(flags);
06aaaa0b
NTND
2547 if (ce->ce_flags & CE_EXTENDED) {
2548 struct ondisk_cache_entry_extended *ondisk2;
2549 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2550 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
f136f7bf
JH
2551 }
2552}
2553
aab61359 2554static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
ce012deb 2555 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
f136f7bf 2556{
9d227781 2557 int size;
f136f7bf 2558 int result;
00a4b035
RJ
2559 unsigned int saved_namelen;
2560 int stripped_name = 0;
ce012deb 2561 static unsigned char padding[8] = { 0x00 };
f136f7bf 2562
b3c96fb1
NTND
2563 if (ce->ce_flags & CE_STRIP_NAME) {
2564 saved_namelen = ce_namelen(ce);
2565 ce->ce_namelen = 0;
00a4b035 2566 stripped_name = 1;
b3c96fb1
NTND
2567 }
2568
ce012deb
KW
2569 if (ce->ce_flags & CE_EXTENDED)
2570 size = offsetof(struct ondisk_cache_entry_extended, name);
2571 else
2572 size = offsetof(struct ondisk_cache_entry, name);
2573
9d227781 2574 if (!previous_name) {
ce012deb
KW
2575 int len = ce_namelen(ce);
2576 copy_cache_entry_to_ondisk(ondisk, ce);
2577 result = ce_write(c, fd, ondisk, size);
2578 if (!result)
2579 result = ce_write(c, fd, ce->name, len);
2580 if (!result)
2581 result = ce_write(c, fd, padding, align_padding_size(size, len));
9d227781
JH
2582 } else {
2583 int common, to_remove, prefix_size;
2584 unsigned char to_remove_vi[16];
2585 for (common = 0;
2586 (ce->name[common] &&
2587 common < previous_name->len &&
2588 ce->name[common] == previous_name->buf[common]);
2589 common++)
2590 ; /* still matching */
2591 to_remove = previous_name->len - common;
2592 prefix_size = encode_varint(to_remove, to_remove_vi);
2593
ce012deb
KW
2594 copy_cache_entry_to_ondisk(ondisk, ce);
2595 result = ce_write(c, fd, ondisk, size);
2596 if (!result)
2597 result = ce_write(c, fd, to_remove_vi, prefix_size);
2598 if (!result)
0b90b881
TG
2599 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2600 if (!result)
2601 result = ce_write(c, fd, padding, 1);
9d227781
JH
2602
2603 strbuf_splice(previous_name, common, to_remove,
2604 ce->name + common, ce_namelen(ce) - common);
06aaaa0b 2605 }
00a4b035 2606 if (stripped_name) {
b3c96fb1
NTND
2607 ce->ce_namelen = saved_namelen;
2608 ce->ce_flags &= ~CE_STRIP_NAME;
2609 }
7a51ed66 2610
59efba64 2611 return result;
7a51ed66
LT
2612}
2613
426ddeea
YM
2614/*
2615 * This function verifies if index_state has the correct sha1 of the
2616 * index file. Don't die if we have any other failure, just return 0.
2617 */
2618static int verify_index_from(const struct index_state *istate, const char *path)
2619{
2620 int fd;
2621 ssize_t n;
2622 struct stat st;
aab61359 2623 unsigned char hash[GIT_MAX_RAWSZ];
426ddeea
YM
2624
2625 if (!istate->initialized)
2626 return 0;
2627
2628 fd = open(path, O_RDONLY);
2629 if (fd < 0)
2630 return 0;
2631
2632 if (fstat(fd, &st))
2633 goto out;
2634
aab61359 2635 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
426ddeea
YM
2636 goto out;
2637
aab61359 2638 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2639 if (n != the_hash_algo->rawsz)
426ddeea
YM
2640 goto out;
2641
67947c34 2642 if (!hasheq(istate->oid.hash, hash))
426ddeea
YM
2643 goto out;
2644
2645 close(fd);
2646 return 1;
2647
2648out:
2649 close(fd);
2650 return 0;
2651}
2652
1b0d968b 2653static int repo_verify_index(struct repository *repo)
426ddeea 2654{
1b0d968b 2655 return verify_index_from(repo->index, repo->index_file);
426ddeea
YM
2656}
2657
483fbe2b
JH
2658static int has_racy_timestamp(struct index_state *istate)
2659{
2660 int entries = istate->cache_nr;
2661 int i;
2662
2663 for (i = 0; i < entries; i++) {
2664 struct cache_entry *ce = istate->cache[i];
2665 if (is_racy_timestamp(istate, ce))
2666 return 1;
2667 }
2668 return 0;
2669}
2670
1b0d968b
NTND
2671void repo_update_index_if_able(struct repository *repo,
2672 struct lock_file *lockfile)
ccdc4ec3 2673{
1b0d968b
NTND
2674 if ((repo->index->cache_changed ||
2675 has_racy_timestamp(repo->index)) &&
2676 repo_verify_index(repo))
2677 write_locked_index(repo->index, lockfile, COMMIT_LOCK);
b74c90fb
2678 else
2679 rollback_lock_file(lockfile);
ccdc4ec3
JH
2680}
2681
d8465500
JN
2682static int record_eoie(void)
2683{
2684 int val;
2685
2686 if (!git_config_get_bool("index.recordendofindexentries", &val))
2687 return val;
2a9dedef
JN
2688
2689 /*
2690 * As a convenience, the end of index entries extension
2691 * used for threading is written by default if the user
2692 * explicitly requested threaded index reads.
2693 */
2694 return !git_config_get_index_threads(&val) && val != 1;
d8465500
JN
2695}
2696
42916054
JN
2697static int record_ieot(void)
2698{
2699 int val;
2700
2701 if (!git_config_get_bool("index.recordoffsettable", &val))
2702 return val;
2a9dedef
JN
2703
2704 /*
2705 * As a convenience, the offset table used for threading is
2706 * written by default if the user explicitly requested
2707 * threaded index reads.
2708 */
2709 return !git_config_get_index_threads(&val) && val != 1;
42916054
JN
2710}
2711
812d6b00
2712/*
2713 * On success, `tempfile` is closed. If it is the temporary file
2714 * of a `struct lock_file`, we will therefore effectively perform
2715 * a 'close_lock_file_gently()`. Since that is an implementation
2716 * detail of lockfiles, callers of `do_write_index()` should not
2717 * rely on it.
2718 */
9f41c7a6 2719static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
c18b80a0 2720 int strip_extensions)
197ee8c9 2721{
ca54d9ba 2722 uint64_t start = getnanotime();
9f41c7a6 2723 int newfd = tempfile->fd;
3b1d9e04 2724 git_hash_ctx c, eoie_c;
197ee8c9 2725 struct cache_header hdr;
b50386c7 2726 int i, err = 0, removed, extended, hdr_version;
4aab5b46
JH
2727 struct cache_entry **cache = istate->cache;
2728 int entries = istate->cache_nr;
e1afca4f 2729 struct stat st;
ce012deb 2730 struct ondisk_cache_entry_extended ondisk;
9d227781 2731 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
4bddd983 2732 int drop_cache_tree = istate->drop_cache_tree;
3b1d9e04 2733 off_t offset;
77ff1127 2734 int ieot_entries = 1;
3255089a
BP
2735 struct index_entry_offset_table *ieot = NULL;
2736 int nr, nr_threads;
025a0709 2737
06aaaa0b 2738 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 2739 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 2740 removed++;
197ee8c9 2741
06aaaa0b
NTND
2742 /* reduce extended entries if possible */
2743 cache[i]->ce_flags &= ~CE_EXTENDED;
2744 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2745 extended++;
2746 cache[i]->ce_flags |= CE_EXTENDED;
2747 }
2748 }
2749
d6e3c181 2750 if (!istate->version) {
136347d7 2751 istate->version = get_index_format_default();
4c2db938 2752 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
d6e3c181
NTND
2753 init_split_index(istate);
2754 }
9d227781
JH
2755
2756 /* demote version 3 to version 2 when the latter suffices */
2757 if (istate->version == 3 || istate->version == 2)
2758 istate->version = extended ? 3 : 2;
2759
2760 hdr_version = istate->version;
2761
ccc4feb5 2762 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
9d227781 2763 hdr.hdr_version = htonl(hdr_version);
025a0709 2764 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 2765
aab61359 2766 the_hash_algo->init_fn(&c);
ca9be054 2767 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
2768 return -1;
2769
2a9dedef 2770 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))
88168b9b 2771 nr_threads = 1;
62e5ee81 2772
42916054 2773 if (nr_threads != 1 && record_ieot()) {
3255089a
BP
2774 int ieot_blocks, cpus;
2775
2776 /*
2777 * ensure default number of ieot blocks maps evenly to the
2778 * default number of threads that will process them leaving
2779 * room for the thread to load the index extensions.
2780 */
2781 if (!nr_threads) {
2782 ieot_blocks = istate->cache_nr / THREAD_COST;
2783 cpus = online_cpus();
2784 if (ieot_blocks > cpus - 1)
2785 ieot_blocks = cpus - 1;
2786 } else {
2787 ieot_blocks = nr_threads;
77ff1127
BP
2788 if (ieot_blocks > istate->cache_nr)
2789 ieot_blocks = istate->cache_nr;
3255089a
BP
2790 }
2791
2792 /*
2793 * no reason to write out the IEOT extension if we don't
2794 * have enough blocks to utilize multi-threading
2795 */
2796 if (ieot_blocks > 1) {
2797 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2798 + (ieot_blocks * sizeof(struct index_entry_offset)));
77ff1127 2799 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
3255089a
BP
2800 }
2801 }
3255089a 2802
3b1d9e04 2803 offset = lseek(newfd, 0, SEEK_CUR);
3255089a
BP
2804 if (offset < 0) {
2805 free(ieot);
3b1d9e04 2806 return -1;
3255089a 2807 }
3b1d9e04 2808 offset += write_buffer_len;
3255089a 2809 nr = 0;
9d227781 2810 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
ce012deb 2811
197ee8c9
LT
2812 for (i = 0; i < entries; i++) {
2813 struct cache_entry *ce = cache[i];
7a51ed66 2814 if (ce->ce_flags & CE_REMOVE)
aa16021e 2815 continue;
e06c43c7 2816 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
58bf2a4c 2817 ce_smudge_racily_clean_entry(istate, ce);
99d1a986 2818 if (is_null_oid(&ce->oid)) {
83bd7437
JK
2819 static const char msg[] = "cache entry has null sha1: %s";
2820 static int allow = -1;
2821
2822 if (allow < 0)
2823 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2824 if (allow)
2825 warning(msg, ce->name);
2826 else
b50386c7 2827 err = error(msg, ce->name);
a96d3cc3
JK
2828
2829 drop_cache_tree = 1;
83bd7437 2830 }
77ff1127 2831 if (ieot && i && (i % ieot_entries == 0)) {
3255089a
BP
2832 ieot->entries[ieot->nr].nr = nr;
2833 ieot->entries[ieot->nr].offset = offset;
2834 ieot->nr++;
2835 /*
2836 * If we have a V4 index, set the first byte to an invalid
2837 * character to ensure there is nothing common with the previous
2838 * entry
2839 */
2840 if (previous_name)
2841 previous_name->buf[0] = 0;
2842 nr = 0;
2843 offset = lseek(newfd, 0, SEEK_CUR);
2844 if (offset < 0) {
2845 free(ieot);
2846 return -1;
2847 }
2848 offset += write_buffer_len;
2849 }
ce012deb 2850 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
b50386c7
KW
2851 err = -1;
2852
2853 if (err)
2854 break;
3255089a
BP
2855 nr++;
2856 }
2857 if (ieot && nr) {
2858 ieot->entries[ieot->nr].nr = nr;
2859 ieot->entries[ieot->nr].offset = offset;
2860 ieot->nr++;
197ee8c9 2861 }
9d227781 2862 strbuf_release(&previous_name_buf);
1af1c2b6 2863
3255089a
BP
2864 if (err) {
2865 free(ieot);
b50386c7 2866 return err;
3255089a 2867 }
b50386c7 2868
bad68ec9 2869 /* Write extension data here */
3b1d9e04 2870 offset = lseek(newfd, 0, SEEK_CUR);
3255089a
BP
2871 if (offset < 0) {
2872 free(ieot);
3b1d9e04 2873 return -1;
3255089a 2874 }
3b1d9e04
BP
2875 offset += write_buffer_len;
2876 the_hash_algo->init_fn(&eoie_c);
2877
3255089a
BP
2878 /*
2879 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2880 * can minimize the number of extensions we have to scan through to
2881 * find it during load. Write it out regardless of the
2882 * strip_extensions parameter as we need it when loading the shared
2883 * index.
2884 */
3255089a
BP
2885 if (ieot) {
2886 struct strbuf sb = STRBUF_INIT;
2887
2888 write_ieot_extension(&sb, ieot);
2889 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2890 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2891 strbuf_release(&sb);
2892 free(ieot);
2893 if (err)
2894 return -1;
2895 }
3255089a 2896
c18b80a0 2897 if (!strip_extensions && istate->split_index) {
5fc2fc8f
NTND
2898 struct strbuf sb = STRBUF_INIT;
2899
2900 err = write_link_extension(&sb, istate) < 0 ||
3b1d9e04 2901 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
5fc2fc8f
NTND
2902 sb.len) < 0 ||
2903 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2904 strbuf_release(&sb);
2905 if (err)
2906 return -1;
2907 }
a96d3cc3 2908 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
f285a2d7 2909 struct strbuf sb = STRBUF_INIT;
1dffb8fa 2910
1dffb8fa 2911 cache_tree_write(&sb, istate->cache_tree);
3b1d9e04 2912 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
1dffb8fa
PH
2913 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2914 strbuf_release(&sb);
2915 if (err)
bad68ec9 2916 return -1;
bad68ec9 2917 }
c18b80a0 2918 if (!strip_extensions && istate->resolve_undo) {
cfc5789a
JH
2919 struct strbuf sb = STRBUF_INIT;
2920
2921 resolve_undo_write(&sb, istate->resolve_undo);
3b1d9e04 2922 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
cfc5789a
JH
2923 sb.len) < 0
2924 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2925 strbuf_release(&sb);
2926 if (err)
2927 return -1;
2928 }
83c094ad
NTND
2929 if (!strip_extensions && istate->untracked) {
2930 struct strbuf sb = STRBUF_INIT;
2931
2932 write_untracked_extension(&sb, istate->untracked);
3b1d9e04 2933 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
83c094ad
NTND
2934 sb.len) < 0 ||
2935 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2936 strbuf_release(&sb);
2937 if (err)
2938 return -1;
2939 }
883e248b
BP
2940 if (!strip_extensions && istate->fsmonitor_last_update) {
2941 struct strbuf sb = STRBUF_INIT;
2942
2943 write_fsmonitor_extension(&sb, istate);
3b1d9e04
BP
2944 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2945 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2946 strbuf_release(&sb);
2947 if (err)
2948 return -1;
2949 }
2950
2951 /*
2952 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2953 * so that it can be found and processed before all the index entries are
2954 * read. Write it out regardless of the strip_extensions parameter as we need it
2955 * when loading the shared index.
2956 */
d8465500 2957 if (offset && record_eoie()) {
3b1d9e04
BP
2958 struct strbuf sb = STRBUF_INIT;
2959
2960 write_eoie_extension(&sb, &eoie_c, offset);
2961 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
883e248b
BP
2962 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2963 strbuf_release(&sb);
2964 if (err)
2965 return -1;
2966 }
e1afca4f 2967
75691ea3 2968 if (ce_flush(&c, newfd, istate->oid.hash))
9f41c7a6 2969 return -1;
49bd0fc2
JK
2970 if (close_tempfile_gently(tempfile)) {
2971 error(_("could not close '%s'"), tempfile->filename.buf);
49bd0fc2
JK
2972 return -1;
2973 }
9f41c7a6 2974 if (stat(tempfile->filename.buf, &st))
e1afca4f 2975 return -1;
5bcf109c
KB
2976 istate->timestamp.sec = (unsigned int)st.st_mtime;
2977 istate->timestamp.nsec = ST_MTIME_NSEC(st);
ca54d9ba 2978 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
e1afca4f 2979 return 0;
197ee8c9 2980}
e46bbcf6 2981
626f35c8
NTND
2982void set_alternate_index_output(const char *name)
2983{
2984 alternate_index_output = name;
2985}
2986
2987static int commit_locked_index(struct lock_file *lk)
2988{
751baced
MH
2989 if (alternate_index_output)
2990 return commit_lock_file_to(lk, alternate_index_output);
2991 else
626f35c8 2992 return commit_lock_file(lk);
626f35c8
NTND
2993}
2994
03b86647
NTND
2995static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2996 unsigned flags)
2997{
076aa2cb 2998 int ret = do_write_index(istate, lock->tempfile, 0);
03b86647
NTND
2999 if (ret)
3000 return ret;
03b86647
NTND
3001 if (flags & COMMIT_LOCK)
3002 return commit_locked_index(lock);
812d6b00 3003 return close_lock_file_gently(lock);
03b86647
NTND
3004}
3005
5fc2fc8f
NTND
3006static int write_split_index(struct index_state *istate,
3007 struct lock_file *lock,
3008 unsigned flags)
3009{
3010 int ret;
3011 prepare_to_write_split_index(istate);
3012 ret = do_write_locked_index(istate, lock, flags);
3013 finish_writing_split_index(istate);
3014 return ret;
3015}
3016
b9683722
CC
3017static const char *shared_index_expire = "2.weeks.ago";
3018
3019static unsigned long get_shared_index_expire_date(void)
3020{
3021 static unsigned long shared_index_expire_date;
3022 static int shared_index_expire_date_prepared;
3023
3024 if (!shared_index_expire_date_prepared) {
3025 git_config_get_expiry("splitindex.sharedindexexpire",
3026 &shared_index_expire);
3027 shared_index_expire_date = approxidate(shared_index_expire);
3028 shared_index_expire_date_prepared = 1;
3029 }
3030
3031 return shared_index_expire_date;
3032}
3033
3034static int should_delete_shared_index(const char *shared_index_path)
3035{
3036 struct stat st;
3037 unsigned long expiration;
3038
3039 /* Check timestamp */
3040 expiration = get_shared_index_expire_date();
3041 if (!expiration)
3042 return 0;
3043 if (stat(shared_index_path, &st))
78bde923 3044 return error_errno(_("could not stat '%s'"), shared_index_path);
b9683722
CC
3045 if (st.st_mtime > expiration)
3046 return 0;
3047
3048 return 1;
3049}
3050
3051static int clean_shared_index_files(const char *current_hex)
3052{
3053 struct dirent *de;
3054 DIR *dir = opendir(get_git_dir());
3055
3056 if (!dir)
3057 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3058
3059 while ((de = readdir(dir)) != NULL) {
3060 const char *sha1_hex;
3061 const char *shared_index_path;
3062 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3063 continue;
3064 if (!strcmp(sha1_hex, current_hex))
3065 continue;
3066 shared_index_path = git_path("%s", de->d_name);
3067 if (should_delete_shared_index(shared_index_path) > 0 &&
3068 unlink(shared_index_path))
3069 warning_errno(_("unable to unlink: %s"), shared_index_path);
3070 }
3071 closedir(dir);
3072
3073 return 0;
3074}
3075
a0a96756 3076static int write_shared_index(struct index_state *istate,
59f9d2dd 3077 struct tempfile **temp)
c18b80a0
NTND
3078{
3079 struct split_index *si = istate->split_index;
076aa2cb 3080 int ret;
c18b80a0 3081
c18b80a0 3082 move_cache_to_base_index(istate);
7db2d08c 3083 ret = do_write_index(si->base, *temp, 1);
59f9d2dd 3084 if (ret)
c18b80a0 3085 return ret;
7db2d08c 3086 ret = adjust_shared_perm(get_tempfile_path(*temp));
df801f3f 3087 if (ret) {
9d0a9e90 3088 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
df801f3f
CC
3089 return ret;
3090 }
7db2d08c 3091 ret = rename_tempfile(temp,
75691ea3 3092 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
b9683722 3093 if (!ret) {
75691ea3 3094 oidcpy(&si->base_oid, &si->base->oid);
3095 clean_shared_index_files(oid_to_hex(&si->base->oid));
b9683722
CC
3096 }
3097
c18b80a0
NTND
3098 return ret;
3099}
3100
e6a1dd77
CC
3101static const int default_max_percent_split_change = 20;
3102
3103static int too_many_not_shared_entries(struct index_state *istate)
3104{
3105 int i, not_shared = 0;
3106 int max_split = git_config_get_max_percent_split_change();
3107
3108 switch (max_split) {
3109 case -1:
3110 /* not or badly configured: use the default value */
3111 max_split = default_max_percent_split_change;
3112 break;
3113 case 0:
3114 return 1; /* 0% means always write a new shared index */
3115 case 100:
3116 return 0; /* 100% means never write a new shared index */
3117 default:
3118 break; /* just use the configured value */
3119 }
3120
3121 /* Count not shared entries */
3122 for (i = 0; i < istate->cache_nr; i++) {
3123 struct cache_entry *ce = istate->cache[i];
3124 if (!ce->index)
3125 not_shared++;
3126 }
3127
3128 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3129}
3130
03b86647
NTND
3131int write_locked_index(struct index_state *istate, struct lock_file *lock,
3132 unsigned flags)
3133{
0d59ffb4 3134 int new_shared_index, ret;
5fc2fc8f
NTND
3135 struct split_index *si = istate->split_index;
3136
4592e608 3137 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
c207e9e1 3138 cache_tree_verify(the_repository, istate);
4592e608 3139
61000814
3140 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3141 if (flags & COMMIT_LOCK)
3142 rollback_lock_file(lock);
3143 return 0;
3144 }
3145
3bd28eb2
AV
3146 if (istate->fsmonitor_last_update)
3147 fill_fsmonitor_bitmap(istate);
3148
5165dd59
NTND
3149 if (!si || alternate_index_output ||
3150 (istate->cache_changed & ~EXTMASK)) {
5fc2fc8f 3151 if (si)
2182abd9 3152 oidclr(&si->base_oid);
df60cf57
3153 ret = do_write_locked_index(istate, lock, flags);
3154 goto out;
5fc2fc8f
NTND
3155 }
3156
4c2db938 3157 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2182abd9 3158 int v = si->base_oid.hash[0];
d6e3c181
NTND
3159 if ((v & 15) < 6)
3160 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3161 }
e6a1dd77
CC
3162 if (too_many_not_shared_entries(istate))
3163 istate->cache_changed |= SPLIT_INDEX_ORDERED;
0d59ffb4
CC
3164
3165 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3166
3167 if (new_shared_index) {
59f9d2dd
NTND
3168 struct tempfile *temp;
3169 int saved_errno;
3170
c9d6c788
ÆAB
3171 /* Same initial permissions as the main .git/index file */
3172 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
59f9d2dd 3173 if (!temp) {
2182abd9 3174 oidclr(&si->base_oid);
59f9d2dd 3175 ret = do_write_locked_index(istate, lock, flags);
ef5b3a6c
NTND
3176 goto out;
3177 }
3178 ret = write_shared_index(istate, &temp);
59f9d2dd
NTND
3179
3180 saved_errno = errno;
3181 if (is_tempfile_active(temp))
3182 delete_tempfile(&temp);
3183 errno = saved_errno;
3184
c18b80a0 3185 if (ret)
df60cf57 3186 goto out;
c18b80a0
NTND
3187 }
3188
0d59ffb4
CC
3189 ret = write_split_index(istate, lock, flags);
3190
3191 /* Freshen the shared index only if the split-index was written */
a125a223
TG
3192 if (!ret && !new_shared_index) {
3193 const char *shared_index = git_path("sharedindex.%s",
2182abd9 3194 oid_to_hex(&si->base_oid));
a125a223
TG
3195 freshen_shared_index(shared_index, 1);
3196 }
0d59ffb4 3197
df60cf57
3198out:
3199 if (flags & COMMIT_LOCK)
3200 rollback_lock_file(lock);
0d59ffb4 3201 return ret;
03b86647
NTND
3202}
3203
e46bbcf6
MV
3204/*
3205 * Read the index file that is potentially unmerged into given
ad376204
EN
3206 * index_state, dropping any unmerged entries to stage #0 (potentially
3207 * resulting in a path appearing as both a file and a directory in the
3208 * index; the caller is responsible to clear out the extra entries
3209 * before writing the index to a tree). Returns true if the index is
3210 * unmerged. Callers who want to refuse to work from an unmerged
3211 * state can call this and check its return value, instead of calling
3212 * read_cache().
e46bbcf6 3213 */
e1ff0a32 3214int repo_read_index_unmerged(struct repository *repo)
e46bbcf6 3215{
e1ff0a32 3216 struct index_state *istate;
e46bbcf6 3217 int i;
d1a43f2a 3218 int unmerged = 0;
e46bbcf6 3219
e1ff0a32
NTND
3220 repo_read_index(repo);
3221 istate = repo->index;
e46bbcf6
MV
3222 for (i = 0; i < istate->cache_nr; i++) {
3223 struct cache_entry *ce = istate->cache[i];
d1a43f2a 3224 struct cache_entry *new_ce;
a849735b 3225 int len;
d1a43f2a
JH
3226
3227 if (!ce_stage(ce))
e46bbcf6 3228 continue;
d1a43f2a 3229 unmerged = 1;
68c4f6a5 3230 len = ce_namelen(ce);
a849735b 3231 new_ce = make_empty_cache_entry(istate, len);
d1a43f2a 3232 memcpy(new_ce->name, ce->name, len);
b60e188c
TG
3233 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3234 new_ce->ce_namelen = len;
d1a43f2a 3235 new_ce->ce_mode = ce->ce_mode;
ad376204 3236 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
9d0a9e90 3237 return error(_("%s: cannot drop to stage #0"),
5699d17e 3238 new_ce->name);
e46bbcf6 3239 }
d1a43f2a 3240 return unmerged;
e46bbcf6 3241}
041aee31 3242
98fa4738
JK
3243/*
3244 * Returns 1 if the path is an "other" path with respect to
3245 * the index; that is, the path is not mentioned in the index at all,
3246 * either as a file, a directory with some files in the index,
3247 * or as an unmerged entry.
3248 *
3249 * We helpfully remove a trailing "/" from directories so that
3250 * the output of read_directory can be used as-is.
3251 */
3252int index_name_is_other(const struct index_state *istate, const char *name,
3253 int namelen)
3254{
3255 int pos;
3256 if (namelen && name[namelen - 1] == '/')
3257 namelen--;
3258 pos = index_name_pos(istate, name, namelen);
3259 if (0 <= pos)
3260 return 0; /* exact match */
3261 pos = -pos - 1;
3262 if (pos < istate->cache_nr) {
3263 struct cache_entry *ce = istate->cache[pos];
3264 if (ce_namelen(ce) == namelen &&
3265 !memcmp(ce->name, name, namelen))
3266 return 0; /* Yup, this one exists unmerged */
3267 }
3268 return 1;
3269}
29fb37b2 3270
87542508
BW
3271void *read_blob_data_from_index(const struct index_state *istate,
3272 const char *path, unsigned long *size)
29fb37b2
LF
3273{
3274 int pos, len;
3275 unsigned long sz;
3276 enum object_type type;
3277 void *data;
3278
3279 len = strlen(path);
3280 pos = index_name_pos(istate, path, len);
3281 if (pos < 0) {
3282 /*
3283 * We might be in the middle of a merge, in which
3284 * case we would read stage #2 (ours).
3285 */
3286 int i;
3287 for (i = -pos - 1;
3288 (pos < 0 && i < istate->cache_nr &&
3289 !strcmp(istate->cache[i]->name, path));
3290 i++)
3291 if (ce_stage(istate->cache[i]) == 2)
3292 pos = i;
3293 }
3294 if (pos < 0)
3295 return NULL;
b4f5aca4 3296 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
29fb37b2
LF
3297 if (!data || type != OBJ_BLOB) {
3298 free(data);
3299 return NULL;
3300 }
ff366825
LF
3301 if (size)
3302 *size = sz;
29fb37b2
LF
3303 return data;
3304}
38612532
MH
3305
3306void stat_validity_clear(struct stat_validity *sv)
3307{
6a83d902 3308 FREE_AND_NULL(sv->sd);
38612532
MH
3309}
3310
3311int stat_validity_check(struct stat_validity *sv, const char *path)
3312{
3313 struct stat st;
3314
3315 if (stat(path, &st) < 0)
3316 return sv->sd == NULL;
3317 if (!sv->sd)
3318 return 0;
3319 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3320}
3321
3322void stat_validity_update(struct stat_validity *sv, int fd)
3323{
3324 struct stat st;
3325
3326 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3327 stat_validity_clear(sv);
3328 else {
3329 if (!sv->sd)
3330 sv->sd = xcalloc(1, sizeof(struct stat_data));
3331 fill_stat_data(sv->sd, &st);
3332 }
3333}
edf3b905
DT
3334
3335void move_index_extensions(struct index_state *dst, struct index_state *src)
3336{
3337 dst->untracked = src->untracked;
3338 src->untracked = NULL;
836ef2b6
NTND
3339 dst->cache_tree = src->cache_tree;
3340 src->cache_tree = NULL;
edf3b905 3341}
a849735b 3342
8e72d675
JM
3343struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3344 struct index_state *istate)
3345{
3346 unsigned int size = ce_size(ce);
3347 int mem_pool_allocated;
3348 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3349 mem_pool_allocated = new_entry->mem_pool_allocated;
3350
3351 memcpy(new_entry, ce, size);
3352 new_entry->mem_pool_allocated = mem_pool_allocated;
3353 return new_entry;
3354}
3355
a849735b
JM
3356void discard_cache_entry(struct cache_entry *ce)
3357{
8616a2d0
JM
3358 if (ce && should_validate_cache_entries())
3359 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3360
8e72d675
JM
3361 if (ce && ce->mem_pool_allocated)
3362 return;
3363
a849735b
JM
3364 free(ce);
3365}
8616a2d0
JM
3366
3367int should_validate_cache_entries(void)
3368{
3369 static int validate_index_cache_entries = -1;
3370
3371 if (validate_index_cache_entries < 0) {
3372 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3373 validate_index_cache_entries = 1;
3374 else
3375 validate_index_cache_entries = 0;
3376 }
3377
3378 return validate_index_cache_entries;
3379}
3b1d9e04
BP
3380
3381#define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3382#define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3383
3384static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3385{
3386 /*
3387 * The end of index entries (EOIE) extension is guaranteed to be last
3388 * so that it can be found by scanning backwards from the EOF.
3389 *
3390 * "EOIE"
3391 * <4-byte length>
3392 * <4-byte offset>
3393 * <20-byte hash>
3394 */
3395 const char *index, *eoie;
3396 uint32_t extsize;
3397 size_t offset, src_offset;
3398 unsigned char hash[GIT_MAX_RAWSZ];
3399 git_hash_ctx c;
3400
3401 /* ensure we have an index big enough to contain an EOIE extension */
3402 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3403 return 0;
3404
3405 /* validate the extension signature */
3406 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3407 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3408 return 0;
3409 index += sizeof(uint32_t);
3410
3411 /* validate the extension size */
3412 extsize = get_be32(index);
3413 if (extsize != EOIE_SIZE)
3414 return 0;
3415 index += sizeof(uint32_t);
3416
3417 /*
3418 * Validate the offset we're going to look for the first extension
3419 * signature is after the index header and before the eoie extension.
3420 */
3421 offset = get_be32(index);
3422 if (mmap + offset < mmap + sizeof(struct cache_header))
3423 return 0;
3424 if (mmap + offset >= eoie)
3425 return 0;
3426 index += sizeof(uint32_t);
3427
3428 /*
3429 * The hash is computed over extension types and their sizes (but not
3430 * their contents). E.g. if we have "TREE" extension that is N-bytes
3431 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3432 * then the hash would be:
3433 *
3434 * SHA-1("TREE" + <binary representation of N> +
3435 * "REUC" + <binary representation of M>)
3436 */
3437 src_offset = offset;
3438 the_hash_algo->init_fn(&c);
3439 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3440 /* After an array of active_nr index entries,
3441 * there can be arbitrary number of extended
3442 * sections, each of which is prefixed with
3443 * extension name (4-byte) and section length
3444 * in 4-byte network byte order.
3445 */
3446 uint32_t extsize;
3447 memcpy(&extsize, mmap + src_offset + 4, 4);
3448 extsize = ntohl(extsize);
3449
3450 /* verify the extension size isn't so large it will wrap around */
3451 if (src_offset + 8 + extsize < src_offset)
3452 return 0;
3453
3454 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3455
3456 src_offset += 8;
3457 src_offset += extsize;
3458 }
3459 the_hash_algo->final_fn(hash, &c);
3460 if (!hasheq(hash, (const unsigned char *)index))
3461 return 0;
3462
3463 /* Validate that the extension offsets returned us back to the eoie extension. */
3464 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3465 return 0;
3466
3467 return offset;
3468}
3469
3470static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3471{
3472 uint32_t buffer;
3473 unsigned char hash[GIT_MAX_RAWSZ];
3474
3475 /* offset */
3476 put_be32(&buffer, offset);
3477 strbuf_add(sb, &buffer, sizeof(uint32_t));
3478
3479 /* hash */
3480 the_hash_algo->final_fn(hash, eoie_context);
3481 strbuf_add(sb, hash, the_hash_algo->rawsz);
3482}
3255089a 3483
3255089a
BP
3484#define IEOT_VERSION (1)
3485
3486static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3487{
ec36c42a
NTND
3488 const char *index = NULL;
3489 uint32_t extsize, ext_version;
3490 struct index_entry_offset_table *ieot;
3491 int i, nr;
3492
3493 /* find the IEOT extension */
3494 if (!offset)
3495 return NULL;
3496 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3497 extsize = get_be32(mmap + offset + 4);
3498 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3499 index = mmap + offset + 4 + 4;
3500 break;
3501 }
3502 offset += 8;
3503 offset += extsize;
3504 }
3505 if (!index)
3506 return NULL;
3507
3508 /* validate the version is IEOT_VERSION */
3509 ext_version = get_be32(index);
3510 if (ext_version != IEOT_VERSION) {
3511 error("invalid IEOT version %d", ext_version);
3512 return NULL;
3513 }
3514 index += sizeof(uint32_t);
3515
3516 /* extension size - version bytes / bytes per entry */
3517 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3518 if (!nr) {
3519 error("invalid number of IEOT entries %d", nr);
3520 return NULL;
3521 }
3522 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3523 + (nr * sizeof(struct index_entry_offset)));