ieot: add Index Entry Offset Table (IEOT) extension
[git/git.git] / read-cache.c
CommitLineData
8bc9a0c7
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
4aab5b46 6#define NO_THE_INDEX_COMPATIBILITY_MACROS
e83c5163 7#include "cache.h"
b2141fc1 8#include "config.h"
cffbfad5
EN
9#include "diff.h"
10#include "diffcore.h"
f6ecc62d 11#include "tempfile.h"
697cc8ef 12#include "lockfile.h"
bad68ec9 13#include "cache-tree.h"
f35a6d3b 14#include "refs.h"
d616813d 15#include "dir.h"
cbd53a21 16#include "object-store.h"
041aee31
JH
17#include "tree.h"
18#include "commit.h"
39425819 19#include "blob.h"
cfc5789a 20#include "resolve-undo.h"
6c9cd161
JH
21#include "strbuf.h"
22#include "varint.h"
5fc2fc8f 23#include "split-index.h"
a42643aa 24#include "utf8.h"
883e248b 25#include "fsmonitor.h"
abb4bb83 26#include "thread-utils.h"
bad68ec9 27
b60e188c
TG
28/* Mask for the name length in ce_flags in the on-disk index */
29
30#define CE_NAMEMASK (0x0fff)
31
bad68ec9
JH
32/* Index extensions.
33 *
34 * The first letter should be 'A'..'Z' for extensions that are not
35 * necessary for a correct operation (i.e. optimization data).
36 * When new extensions are added that _needs_ to be understood in
37 * order to correctly interpret the index file, pick character that
38 * is outside the range, to cause the reader to abort.
39 */
40
41#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
42#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
b659b49b 43#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
5fc2fc8f 44#define CACHE_EXT_LINK 0x6c696e6b /* "link" */
83c094ad 45#define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
883e248b 46#define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
3b1d9e04 47#define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
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
228e94f9 97struct index_state the_index;
626f35c8 98static const char *alternate_index_output;
8fd2cb40 99
9cb76b8c
JH
100static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
101{
102 istate->cache[nr] = ce;
96872bc2 103 add_name_hash(istate, ce);
9cb76b8c
JH
104}
105
cf558704
LT
106static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
107{
108 struct cache_entry *old = istate->cache[nr];
109
078a58e8 110 replace_index_entry_in_base(istate, old, ce);
2092678c 111 remove_name_hash(istate, old);
a849735b 112 discard_cache_entry(old);
0e267b7a 113 ce->ce_flags &= ~CE_HASHED;
a22c6371 114 set_index_entry(istate, nr, ce);
078a58e8 115 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 116 mark_fsmonitor_invalid(istate, ce);
e636a7b4 117 istate->cache_changed |= CE_ENTRY_CHANGED;
cf558704
LT
118}
119
81dc2307
PB
120void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
121{
285c2e25 122 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
81dc2307
PB
123 int namelen = strlen(new_name);
124
a849735b 125 new_entry = make_empty_cache_entry(istate, namelen);
285c2e25
BW
126 copy_cache_entry(new_entry, old_entry);
127 new_entry->ce_flags &= ~CE_HASHED;
128 new_entry->ce_namelen = namelen;
129 new_entry->index = 0;
130 memcpy(new_entry->name, new_name, namelen + 1);
81dc2307 131
285c2e25
BW
132 cache_tree_invalidate_path(istate, old_entry->name);
133 untracked_cache_remove_from_index(istate, old_entry->name);
81dc2307 134 remove_index_entry_at(istate, nr);
285c2e25 135 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
81dc2307
PB
136}
137
c21d39d7
MH
138void fill_stat_data(struct stat_data *sd, struct stat *st)
139{
140 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
141 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
142 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
143 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
144 sd->sd_dev = st->st_dev;
145 sd->sd_ino = st->st_ino;
146 sd->sd_uid = st->st_uid;
147 sd->sd_gid = st->st_gid;
148 sd->sd_size = st->st_size;
149}
150
151int match_stat_data(const struct stat_data *sd, struct stat *st)
152{
153 int changed = 0;
154
155 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
156 changed |= MTIME_CHANGED;
157 if (trust_ctime && check_stat &&
158 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
159 changed |= CTIME_CHANGED;
160
161#ifdef USE_NSEC
162 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
163 changed |= MTIME_CHANGED;
164 if (trust_ctime && check_stat &&
165 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
166 changed |= CTIME_CHANGED;
167#endif
168
169 if (check_stat) {
170 if (sd->sd_uid != (unsigned int) st->st_uid ||
171 sd->sd_gid != (unsigned int) st->st_gid)
172 changed |= OWNER_CHANGED;
173 if (sd->sd_ino != (unsigned int) st->st_ino)
174 changed |= INODE_CHANGED;
175 }
176
177#ifdef USE_STDEV
178 /*
179 * st_dev breaks on network filesystems where different
180 * clients will have different views of what "device"
181 * the filesystem is on
182 */
183 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
184 changed |= INODE_CHANGED;
185#endif
186
187 if (sd->sd_size != (unsigned int) st->st_size)
188 changed |= DATA_CHANGED;
189
190 return changed;
191}
192
415e96c8
JH
193/*
194 * This only updates the "non-critical" parts of the directory
195 * cache, ie the parts that aren't tracked by GIT, and only used
196 * to validate the cache.
197 */
198void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
199{
c21d39d7 200 fill_stat_data(&ce->ce_stat_data, st);
5f73076c
JH
201
202 if (assume_unchanged)
7a51ed66 203 ce->ce_flags |= CE_VALID;
eadb5831 204
883e248b 205 if (S_ISREG(st->st_mode)) {
eadb5831 206 ce_mark_uptodate(ce);
883e248b
BP
207 mark_fsmonitor_valid(ce);
208 }
415e96c8
JH
209}
210
21a6b9fa 211static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
212{
213 int match = -1;
1b8ac5ea 214 int fd = git_open_cloexec(ce->name, O_RDONLY);
29e4d363
JH
215
216 if (fd >= 0) {
bebfecb9 217 struct object_id oid;
e3506559 218 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
6a29d7b7 219 match = !oideq(&oid, &ce->oid);
7f8508e8 220 /* index_fd() closed the file descriptor already */
29e4d363
JH
221 }
222 return match;
223}
224
21a6b9fa 225static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
29e4d363
JH
226{
227 int match = -1;
29e4d363
JH
228 void *buffer;
229 unsigned long size;
21666f1a 230 enum object_type type;
a60272b3 231 struct strbuf sb = STRBUF_INIT;
29e4d363 232
a60272b3 233 if (strbuf_readlink(&sb, ce->name, expected_size))
29e4d363 234 return -1;
a60272b3 235
b4f5aca4 236 buffer = read_object_file(&ce->oid, &type, &size);
a60272b3
LT
237 if (buffer) {
238 if (size == sb.len)
239 match = memcmp(buffer, sb.buf, size);
240 free(buffer);
29e4d363 241 }
a60272b3 242 strbuf_release(&sb);
29e4d363
JH
243 return match;
244}
245
21a6b9fa 246static int ce_compare_gitlink(const struct cache_entry *ce)
f35a6d3b 247{
1053fe82 248 struct object_id oid;
f35a6d3b
LT
249
250 /*
251 * We don't actually require that the .git directory
302b9282 252 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
253 * might even be missing (in case nobody populated that
254 * sub-project).
255 *
256 * If so, we consider it always to match.
257 */
a98e6101 258 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
f35a6d3b 259 return 0;
6a29d7b7 260 return !oideq(&oid, &ce->oid);
f35a6d3b
LT
261}
262
21a6b9fa 263static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
29e4d363
JH
264{
265 switch (st->st_mode & S_IFMT) {
266 case S_IFREG:
267 if (ce_compare_data(ce, st))
268 return DATA_CHANGED;
269 break;
270 case S_IFLNK:
dc49cd76 271 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
272 return DATA_CHANGED;
273 break;
a8ee75bc 274 case S_IFDIR:
7a51ed66 275 if (S_ISGITLINK(ce->ce_mode))
c70115b4 276 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
1cf01a34 277 /* else fallthrough */
29e4d363
JH
278 default:
279 return TYPE_CHANGED;
280 }
281 return 0;
282}
283
21a6b9fa 284static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
734aab75
LT
285{
286 unsigned int changed = 0;
287
7a51ed66
LT
288 if (ce->ce_flags & CE_REMOVE)
289 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
290
291 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
292 case S_IFREG:
293 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
294 /* We consider only the owner x bit to be relevant for
295 * "mode changes"
296 */
297 if (trust_executable_bit &&
7a51ed66 298 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 299 changed |= MODE_CHANGED;
8ae0a8c5
KS
300 break;
301 case S_IFLNK:
78a8d641
JS
302 if (!S_ISLNK(st->st_mode) &&
303 (has_symlinks || !S_ISREG(st->st_mode)))
304 changed |= TYPE_CHANGED;
8ae0a8c5 305 break;
302b9282 306 case S_IFGITLINK:
c70115b4 307 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
308 if (!S_ISDIR(st->st_mode))
309 changed |= TYPE_CHANGED;
310 else if (ce_compare_gitlink(ce))
311 changed |= DATA_CHANGED;
a8ee75bc 312 return changed;
8ae0a8c5 313 default:
7a51ed66 314 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 315 }
2cb45e95 316
c21d39d7 317 changed |= match_stat_data(&ce->ce_stat_data, st);
b0391890 318
f49c2c22 319 /* Racily smudged entry? */
c21d39d7 320 if (!ce->ce_stat_data.sd_size) {
99d1a986 321 if (!is_empty_blob_sha1(ce->oid.hash))
f49c2c22
LT
322 changed |= DATA_CHANGED;
323 }
324
407c8eb0
JH
325 return changed;
326}
327
2bb4cda1
NTND
328static int is_racy_stat(const struct index_state *istate,
329 const struct stat_data *sd)
6d91da6d 330{
2bb4cda1 331 return (istate->timestamp.sec &&
fba2f38a
KB
332#ifdef USE_NSEC
333 /* nanosecond timestamped files can also be racy! */
2bb4cda1
NTND
334 (istate->timestamp.sec < sd->sd_mtime.sec ||
335 (istate->timestamp.sec == sd->sd_mtime.sec &&
336 istate->timestamp.nsec <= sd->sd_mtime.nsec))
fba2f38a 337#else
2bb4cda1 338 istate->timestamp.sec <= sd->sd_mtime.sec
fba2f38a 339#endif
2bb4cda1
NTND
340 );
341}
342
343static int is_racy_timestamp(const struct index_state *istate,
344 const struct cache_entry *ce)
345{
346 return (!S_ISGITLINK(ce->ce_mode) &&
347 is_racy_stat(istate, &ce->ce_stat_data));
6d91da6d
JH
348}
349
ed4efab1
NTND
350int match_stat_data_racy(const struct index_state *istate,
351 const struct stat_data *sd, struct stat *st)
352{
353 if (is_racy_stat(istate, sd))
354 return MTIME_CHANGED;
355 return match_stat_data(sd, st);
6d91da6d
JH
356}
357
883e248b 358int ie_match_stat(struct index_state *istate,
21a6b9fa 359 const struct cache_entry *ce, struct stat *st,
4bd5b7da 360 unsigned int options)
407c8eb0 361{
5f73076c 362 unsigned int changed;
4bd5b7da 363 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 364 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
4bd5b7da 365 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
883e248b 366 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
5f73076c 367
883e248b
BP
368 if (!ignore_fsmonitor)
369 refresh_fsmonitor(istate);
5f73076c
JH
370 /*
371 * If it's marked as always valid in the index, it's
372 * valid whatever the checked-out copy says.
56cac48c
NTND
373 *
374 * skip-worktree has the same effect with higher precedence
5f73076c 375 */
56cac48c
NTND
376 if (!ignore_skip_worktree && ce_skip_worktree(ce))
377 return 0;
7a51ed66 378 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c 379 return 0;
883e248b
BP
380 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
381 return 0;
5f73076c 382
331fcb59
JH
383 /*
384 * Intent-to-add entries have not been added, so the index entry
385 * by definition never matches what is in the work tree until it
386 * actually gets added.
387 */
895ff3b2 388 if (ce_intent_to_add(ce))
331fcb59
JH
389 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
390
5f73076c 391 changed = ce_match_stat_basic(ce, st);
407c8eb0 392
29e4d363
JH
393 /*
394 * Within 1 second of this sequence:
395 * echo xyzzy >file && git-update-index --add file
396 * running this command:
397 * echo frotz >file
398 * would give a falsely clean cache entry. The mtime and
399 * length match the cache, and other stat fields do not change.
400 *
401 * We could detect this at update-index time (the cache entry
402 * being registered/updated records the same time as "now")
403 * and delay the return from git-update-index, but that would
404 * effectively mean we can make at most one commit per second,
405 * which is not acceptable. Instead, we check cache entries
406 * whose mtime are the same as the index file timestamp more
5f73076c 407 * carefully than others.
29e4d363 408 */
6d91da6d 409 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
410 if (assume_racy_is_modified)
411 changed |= DATA_CHANGED;
412 else
413 changed |= ce_modified_check_fs(ce, st);
414 }
b0391890 415
29e4d363 416 return changed;
b0391890
JH
417}
418
883e248b 419int ie_modified(struct index_state *istate,
21a6b9fa
RS
420 const struct cache_entry *ce,
421 struct stat *st, unsigned int options)
b0391890 422{
29e4d363 423 int changed, changed_fs;
4bd5b7da
JH
424
425 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
426 if (!changed)
427 return 0;
b0391890
JH
428 /*
429 * If the mode or type has changed, there's no point in trying
430 * to refresh the entry - it's not going to match
431 */
432 if (changed & (MODE_CHANGED | TYPE_CHANGED))
433 return changed;
434
c70115b4
JH
435 /*
436 * Immediately after read-tree or update-index --cacheinfo,
437 * the length field is zero, as we have never even read the
438 * lstat(2) information once, and we cannot trust DATA_CHANGED
439 * returned by ie_match_stat() which in turn was returned by
440 * ce_match_stat_basic() to signal that the filesize of the
441 * blob changed. We have to actually go to the filesystem to
442 * see if the contents match, and if so, should answer "unchanged".
443 *
444 * The logic does not apply to gitlinks, as ce_match_stat_basic()
445 * already has checked the actual HEAD from the filesystem in the
446 * subproject. If ie_match_stat() already said it is different,
447 * then we know it is.
b0391890 448 */
c70115b4 449 if ((changed & DATA_CHANGED) &&
c21d39d7 450 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
b0391890
JH
451 return changed;
452
29e4d363
JH
453 changed_fs = ce_modified_check_fs(ce, st);
454 if (changed_fs)
455 return changed | changed_fs;
b0391890
JH
456 return 0;
457}
458
958ba6c9
LT
459int base_name_compare(const char *name1, int len1, int mode1,
460 const char *name2, int len2, int mode2)
461{
462 unsigned char c1, c2;
463 int len = len1 < len2 ? len1 : len2;
464 int cmp;
465
466 cmp = memcmp(name1, name2, len);
467 if (cmp)
468 return cmp;
469 c1 = name1[len];
470 c2 = name2[len];
1833a925 471 if (!c1 && S_ISDIR(mode1))
958ba6c9 472 c1 = '/';
1833a925 473 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
474 c2 = '/';
475 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
476}
477
0ab9e1e8
LT
478/*
479 * df_name_compare() is identical to base_name_compare(), except it
480 * compares conflicting directory/file entries as equal. Note that
481 * while a directory name compares as equal to a regular file, they
482 * then individually compare _differently_ to a filename that has
483 * a dot after the basename (because '\0' < '.' < '/').
484 *
485 * This is used by routines that want to traverse the git namespace
486 * but then handle conflicting entries together when possible.
487 */
488int df_name_compare(const char *name1, int len1, int mode1,
489 const char *name2, int len2, int mode2)
490{
491 int len = len1 < len2 ? len1 : len2, cmp;
492 unsigned char c1, c2;
493
494 cmp = memcmp(name1, name2, len);
495 if (cmp)
496 return cmp;
497 /* Directories and files compare equal (same length, same name) */
498 if (len1 == len2)
499 return 0;
500 c1 = name1[len];
501 if (!c1 && S_ISDIR(mode1))
502 c1 = '/';
503 c2 = name2[len];
504 if (!c2 && S_ISDIR(mode2))
505 c2 = '/';
506 if (c1 == '/' && !c2)
507 return 0;
508 if (c2 == '/' && !c1)
509 return 0;
510 return c1 - c2;
511}
512
ccdd4a0f 513int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
eb38c22f 514{
ccdd4a0f
JM
515 size_t min_len = (len1 < len2) ? len1 : len2;
516 int cmp = memcmp(name1, name2, min_len);
eb38c22f
LT
517 if (cmp)
518 return cmp;
519 if (len1 < len2)
520 return -1;
521 if (len1 > len2)
522 return 1;
ccdd4a0f
JM
523 return 0;
524}
525
526int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
527{
528 int cmp;
529
530 cmp = name_compare(name1, len1, name2, len2);
531 if (cmp)
532 return cmp;
5f73076c 533
b60e188c 534 if (stage1 < stage2)
95fd5bf8 535 return -1;
b60e188c 536 if (stage1 > stage2)
95fd5bf8 537 return 1;
eb38c22f
LT
538 return 0;
539}
540
357e9c69 541static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
eb38c22f
LT
542{
543 int first, last;
544
545 first = 0;
4aab5b46 546 last = istate->cache_nr;
eb38c22f
LT
547 while (last > first) {
548 int next = (last + first) >> 1;
4aab5b46 549 struct cache_entry *ce = istate->cache[next];
b60e188c 550 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
eb38c22f 551 if (!cmp)
76e7f4ec 552 return next;
eb38c22f
LT
553 if (cmp < 0) {
554 last = next;
555 continue;
556 }
557 first = next+1;
558 }
76e7f4ec 559 return -first-1;
eb38c22f
LT
560}
561
b60e188c
TG
562int index_name_pos(const struct index_state *istate, const char *name, int namelen)
563{
564 return index_name_stage_pos(istate, name, namelen, 0);
565}
566
4aab5b46 567int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 568{
cf558704
LT
569 struct cache_entry *ce = istate->cache[pos];
570
cfc5789a 571 record_resolve_undo(istate, ce);
2092678c 572 remove_name_hash(istate, ce);
045113a5 573 save_or_free_index_entry(istate, ce);
e636a7b4 574 istate->cache_changed |= CE_ENTRY_REMOVED;
4aab5b46
JH
575 istate->cache_nr--;
576 if (pos >= istate->cache_nr)
7b937ca3 577 return 0;
f331ab9d
RS
578 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
579 istate->cache_nr - pos);
7b937ca3
LT
580 return 1;
581}
582
36419c8e 583/*
98e023de 584 * Remove all cache entries marked for removal, that is where
36419c8e
KB
585 * CE_REMOVE is set in ce_flags. This is much more effective than
586 * calling remove_index_entry_at() for each entry to be removed.
587 */
588void remove_marked_cache_entries(struct index_state *istate)
589{
590 struct cache_entry **ce_array = istate->cache;
591 unsigned int i, j;
592
593 for (i = j = 0; i < istate->cache_nr; i++) {
5699d17e 594 if (ce_array[i]->ce_flags & CE_REMOVE) {
2092678c 595 remove_name_hash(istate, ce_array[i]);
045113a5 596 save_or_free_index_entry(istate, ce_array[i]);
5699d17e 597 }
36419c8e
KB
598 else
599 ce_array[j++] = ce_array[i];
600 }
ad837d9e
NTND
601 if (j == istate->cache_nr)
602 return;
e636a7b4 603 istate->cache_changed |= CE_ENTRY_REMOVED;
36419c8e
KB
604 istate->cache_nr = j;
605}
606
4aab5b46 607int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 608{
4aab5b46 609 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
610 if (pos < 0)
611 pos = -pos-1;
a5400efe 612 cache_tree_invalidate_path(istate, path);
e931371a 613 untracked_cache_remove_from_index(istate, path);
4aab5b46
JH
614 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
615 remove_index_entry_at(istate, pos);
197ee8c9
LT
616 return 0;
617}
618
20314271
JS
619static int compare_name(struct cache_entry *ce, const char *path, int namelen)
620{
621 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
622}
623
624static int index_name_pos_also_unmerged(struct index_state *istate,
625 const char *path, int namelen)
626{
627 int pos = index_name_pos(istate, path, namelen);
628 struct cache_entry *ce;
629
630 if (pos >= 0)
631 return pos;
632
633 /* maybe unmerged? */
634 pos = -1 - pos;
635 if (pos >= istate->cache_nr ||
636 compare_name((ce = istate->cache[pos]), path, namelen))
637 return -1;
638
639 /* order of preference: stage 2, 1, 3 */
640 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
641 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
642 !compare_name(ce, path, namelen))
643 pos++;
644 return pos;
645}
646
1102952b
LT
647static int different_name(struct cache_entry *ce, struct cache_entry *alias)
648{
649 int len = ce_namelen(ce);
650 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
651}
652
653/*
654 * If we add a filename that aliases in the cache, we will use the
655 * name that we already have - but we don't want to update the same
656 * alias twice, because that implies that there were actually two
657 * different files with aliasing names!
658 *
659 * So we use the CE_ADDED flag to verify that the alias was an old
660 * one before we accept it as
661 */
045113a5
NTND
662static struct cache_entry *create_alias_ce(struct index_state *istate,
663 struct cache_entry *ce,
664 struct cache_entry *alias)
1102952b
LT
665{
666 int len;
285c2e25 667 struct cache_entry *new_entry;
1102952b
LT
668
669 if (alias->ce_flags & CE_ADDED)
670 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
671
672 /* Ok, create the new entry using the name of the existing alias */
673 len = ce_namelen(alias);
a849735b 674 new_entry = make_empty_cache_entry(istate, len);
285c2e25
BW
675 memcpy(new_entry->name, alias->name, len);
676 copy_cache_entry(new_entry, ce);
045113a5 677 save_or_free_index_entry(istate, ce);
285c2e25 678 return new_entry;
1102952b
LT
679}
680
b4b313f9 681void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
39425819 682{
a09c985e
PO
683 struct object_id oid;
684 if (write_object_file("", 0, blob_type, &oid))
39425819 685 die("cannot create an empty blob in the object database");
a09c985e 686 oidcpy(&ce->oid, &oid);
39425819
JH
687}
688
610d55af 689int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 690{
a849735b 691 int namelen, was_same;
d177cab0 692 mode_t st_mode = st->st_mode;
9472935d 693 struct cache_entry *ce, *alias = NULL;
56cac48c 694 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
695 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
696 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
697 int intent_only = flags & ADD_CACHE_INTENT;
698 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
699 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
9472935d
TB
700 int newflags = HASH_WRITE_OBJECT;
701
702 if (flags & HASH_RENORMALIZE)
703 newflags |= HASH_RENORMALIZE;
11be42a4 704
d177cab0 705 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
960b8ad1 706 return error("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
707
708 namelen = strlen(path);
d177cab0 709 if (S_ISDIR(st_mode)) {
09595258
LT
710 while (namelen && path[namelen-1] == '/')
711 namelen--;
712 }
a849735b 713 ce = make_empty_cache_entry(istate, namelen);
11be42a4 714 memcpy(ce->name, path, namelen);
b60e188c 715 ce->ce_namelen = namelen;
39425819
JH
716 if (!intent_only)
717 fill_stat_cache_info(ce, st);
388b2acd
JH
718 else
719 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 720
610d55af
TG
721
722 if (trust_executable_bit && has_symlinks) {
d177cab0 723 ce->ce_mode = create_ce_mode(st_mode);
610d55af 724 } else {
78a8d641
JS
725 /* If there is an existing entry, pick the mode bits and type
726 * from it, otherwise assume unexecutable regular file.
11be42a4 727 */
185c975f 728 struct cache_entry *ent;
20314271 729 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 730
4aab5b46 731 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 732 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
733 }
734
dc1ae704
JJ
735 /* When core.ignorecase=true, determine if a directory of the same name but differing
736 * case already exists within the Git repository. If it does, ensure the directory
737 * case of the file being added to the repository matches (is folded into) the existing
738 * entry's directory case.
739 */
740 if (ignore_case) {
41284eb0 741 adjust_dirname_case(istate, ce->name);
dc1ae704 742 }
9472935d
TB
743 if (!(flags & HASH_RENORMALIZE)) {
744 alias = index_file_exists(istate, ce->name,
745 ce_namelen(ce), ignore_case);
746 if (alias &&
747 !ce_stage(alias) &&
748 !ie_match_stat(istate, alias, st, ce_option)) {
749 /* Nothing changed, really */
750 if (!S_ISGITLINK(alias->ce_mode))
751 ce_mark_uptodate(alias);
752 alias->ce_flags |= CE_ADDED;
dc1ae704 753
a849735b 754 discard_cache_entry(ce);
9472935d
TB
755 return 0;
756 }
0781b8a9 757 }
39425819 758 if (!intent_only) {
9472935d 759 if (index_path(&ce->oid, path, st, newflags)) {
a849735b 760 discard_cache_entry(ce);
39425819 761 return error("unable to index file %s", path);
2d9426b0 762 }
39425819 763 } else
b4b313f9 764 set_object_name_for_intent_to_add_entry(ce);
39425819 765
1102952b 766 if (ignore_case && alias && different_name(ce, alias))
045113a5 767 ce = create_alias_ce(istate, ce, alias);
1102952b 768 ce->ce_flags |= CE_ADDED;
38ed1d89 769
3bf0dd1f 770 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
771 was_same = (alias &&
772 !ce_stage(alias) &&
4a7e27e9 773 oideq(&alias->oid, &ce->oid) &&
38ed1d89
JH
774 ce->ce_mode == alias->ce_mode);
775
776 if (pretend)
a849735b 777 discard_cache_entry(ce);
067178ed 778 else if (add_index_entry(istate, ce, add_option)) {
a849735b 779 discard_cache_entry(ce);
067178ed
JH
780 return error("unable to add %s to index", path);
781 }
38ed1d89 782 if (verbose && !was_same)
11be42a4 783 printf("add '%s'\n", path);
11be42a4
JS
784 return 0;
785}
786
610d55af 787int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
788{
789 struct stat st;
790 if (lstat(path, &st))
d824cbba 791 die_errno("unable to stat '%s'", path);
610d55af 792 return add_to_index(istate, path, &st, flags);
d177cab0
LT
793}
794
a849735b
JM
795struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
796{
8e72d675 797 return mem_pool__ce_calloc(find_mem_pool(istate), len);
a849735b
JM
798}
799
800struct cache_entry *make_empty_transient_cache_entry(size_t len)
801{
802 return xcalloc(1, cache_entry_size(len));
803}
804
805struct cache_entry *make_cache_entry(struct index_state *istate,
806 unsigned int mode,
825ed4d9
JM
807 const struct object_id *oid,
808 const char *path,
809 int stage,
810 unsigned int refresh_options)
6640f881 811{
bc1c2caa 812 struct cache_entry *ce, *ret;
a849735b 813 int len;
6640f881 814
10ecfa76 815 if (!verify_path(path, mode)) {
7e7abea9 816 error("Invalid path '%s'", path);
6640f881 817 return NULL;
7e7abea9 818 }
6640f881
CR
819
820 len = strlen(path);
a849735b 821 ce = make_empty_cache_entry(istate, len);
6640f881 822
825ed4d9 823 oidcpy(&ce->oid, oid);
6640f881 824 memcpy(ce->name, path, len);
b60e188c
TG
825 ce->ce_flags = create_ce_flags(stage);
826 ce->ce_namelen = len;
6640f881
CR
827 ce->ce_mode = create_ce_mode(mode);
828
768d7965 829 ret = refresh_cache_entry(&the_index, ce, refresh_options);
915e44c6 830 if (ret != ce)
a849735b 831 discard_cache_entry(ce);
915e44c6 832 return ret;
6640f881
CR
833}
834
a849735b
JM
835struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
836 const char *path, int stage)
837{
838 struct cache_entry *ce;
839 int len;
840
841 if (!verify_path(path, mode)) {
842 error("Invalid path '%s'", path);
843 return NULL;
844 }
845
846 len = strlen(path);
847 ce = make_empty_transient_cache_entry(len);
848
849 oidcpy(&ce->oid, oid);
850 memcpy(ce->name, path, len);
851 ce->ce_flags = create_ce_flags(stage);
852 ce->ce_namelen = len;
853 ce->ce_mode = create_ce_mode(mode);
854
855 return ce;
856}
857
d9d70966
TG
858/*
859 * Chmod an index entry with either +x or -x.
860 *
861 * Returns -1 if the chmod for the particular cache entry failed (if it's
862 * not a regular file), -2 if an invalid flip argument is passed in, 0
863 * otherwise.
864 */
865int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
866 char flip)
867{
868 if (!S_ISREG(ce->ce_mode))
869 return -1;
870 switch (flip) {
871 case '+':
872 ce->ce_mode |= 0111;
873 break;
874 case '-':
875 ce->ce_mode &= ~0111;
876 break;
877 default:
878 return -2;
879 }
880 cache_tree_invalidate_path(istate, ce->name);
881 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 882 mark_fsmonitor_invalid(istate, ce);
d9d70966
TG
883 istate->cache_changed |= CE_ENTRY_CHANGED;
884
885 return 0;
886}
887
9c5e6c80 888int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
7b937ca3
LT
889{
890 int len = ce_namelen(a);
891 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
892}
893
8dcf39c4
LT
894/*
895 * We fundamentally don't like some paths: we don't want
896 * dot or dot-dot anywhere, and for obvious reasons don't
897 * want to recurse into ".git" either.
898 *
899 * Also, we don't want double slashes or slashes at the
900 * end that can make pathnames ambiguous.
901 */
10ecfa76 902static int verify_dotfile(const char *rest, unsigned mode)
8dcf39c4
LT
903{
904 /*
905 * The first character was '.', but that
906 * has already been discarded, we now test
907 * the rest.
908 */
e0f530ff 909
8dcf39c4 910 /* "." is not allowed */
e0f530ff 911 if (*rest == '\0' || is_dir_sep(*rest))
8dcf39c4
LT
912 return 0;
913
e0f530ff 914 switch (*rest) {
8dcf39c4 915 /*
641084b6
JK
916 * ".git" followed by NUL or slash is bad. Note that we match
917 * case-insensitively here, even if ignore_case is not set.
918 * This outlaws ".GIT" everywhere out of an abundance of caution,
919 * since there's really no good reason to allow it.
10ecfa76
JK
920 *
921 * Once we've seen ".git", we can also find ".gitmodules", etc (also
922 * case-insensitively).
8dcf39c4
LT
923 */
924 case 'g':
cc2fc7c2
JK
925 case 'G':
926 if (rest[1] != 'i' && rest[1] != 'I')
8dcf39c4 927 break;
cc2fc7c2 928 if (rest[2] != 't' && rest[2] != 'T')
8dcf39c4 929 break;
e19e5e66
JK
930 if (rest[3] == '\0' || is_dir_sep(rest[3]))
931 return 0;
10ecfa76
JK
932 if (S_ISLNK(mode)) {
933 rest += 3;
934 if (skip_iprefix(rest, "modules", &rest) &&
935 (*rest == '\0' || is_dir_sep(*rest)))
936 return 0;
937 }
e19e5e66 938 break;
8dcf39c4 939 case '.':
e0f530ff 940 if (rest[1] == '\0' || is_dir_sep(rest[1]))
8dcf39c4
LT
941 return 0;
942 }
943 return 1;
944}
945
10ecfa76 946int verify_path(const char *path, unsigned mode)
8dcf39c4
LT
947{
948 char c;
949
56948cb6
EFL
950 if (has_dos_drive_prefix(path))
951 return 0;
952
8dcf39c4
LT
953 goto inside;
954 for (;;) {
955 if (!c)
956 return 1;
56948cb6 957 if (is_dir_sep(c)) {
8dcf39c4 958inside:
10ecfa76
JK
959 if (protect_hfs) {
960 if (is_hfs_dotgit(path))
961 return 0;
962 if (S_ISLNK(mode)) {
963 if (is_hfs_dotgitmodules(path))
964 return 0;
965 }
966 }
967 if (protect_ntfs) {
968 if (is_ntfs_dotgit(path))
969 return 0;
970 if (S_ISLNK(mode)) {
971 if (is_ntfs_dotgitmodules(path))
972 return 0;
973 }
974 }
975
8dcf39c4 976 c = *path++;
10ecfa76 977 if ((c == '.' && !verify_dotfile(path, mode)) ||
3bdf09c7
JH
978 is_dir_sep(c) || c == '\0')
979 return 0;
8dcf39c4
LT
980 }
981 c = *path++;
982 }
983}
984
12676608
LT
985/*
986 * Do we have another file that has the beginning components being a
987 * proper superset of the name we're trying to add?
0f1e4f04 988 */
4aab5b46
JH
989static int has_file_name(struct index_state *istate,
990 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 991{
12676608
LT
992 int retval = 0;
993 int len = ce_namelen(ce);
b155725d 994 int stage = ce_stage(ce);
12676608 995 const char *name = ce->name;
0f1e4f04 996
4aab5b46
JH
997 while (pos < istate->cache_nr) {
998 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 999
12676608 1000 if (len >= ce_namelen(p))
0f1e4f04 1001 break;
12676608
LT
1002 if (memcmp(name, p->name, len))
1003 break;
b155725d
JH
1004 if (ce_stage(p) != stage)
1005 continue;
12676608
LT
1006 if (p->name[len] != '/')
1007 continue;
7a51ed66 1008 if (p->ce_flags & CE_REMOVE)
21cd8d00 1009 continue;
12676608
LT
1010 retval = -1;
1011 if (!ok_to_replace)
1012 break;
4aab5b46 1013 remove_index_entry_at(istate, --pos);
0f1e4f04 1014 }
12676608
LT
1015 return retval;
1016}
0f1e4f04 1017
a6db3fbb
JH
1018
1019/*
1020 * Like strcmp(), but also return the offset of the first change.
1021 * If strings are equal, return the length.
1022 */
1023int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1024{
1025 size_t k;
1026
1027 if (!first_change)
1028 return strcmp(s1, s2);
1029
1030 for (k = 0; s1[k] == s2[k]; k++)
1031 if (s1[k] == '\0')
1032 break;
1033
1034 *first_change = k;
1035 return (unsigned char)s1[k] - (unsigned char)s2[k];
1036}
1037
12676608
LT
1038/*
1039 * Do we have another file with a pathname that is a proper
1040 * subset of the name we're trying to add?
06b6d81b
JH
1041 *
1042 * That is, is there another file in the index with a path
1043 * that matches a sub-directory in the given entry?
12676608 1044 */
4aab5b46
JH
1045static int has_dir_name(struct index_state *istate,
1046 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
1047{
1048 int retval = 0;
b155725d 1049 int stage = ce_stage(ce);
12676608
LT
1050 const char *name = ce->name;
1051 const char *slash = name + ce_namelen(ce);
06b6d81b
JH
1052 size_t len_eq_last;
1053 int cmp_last = 0;
1054
1055 /*
1056 * We are frequently called during an iteration on a sorted
1057 * list of pathnames and while building a new index. Therefore,
1058 * there is a high probability that this entry will eventually
1059 * be appended to the index, rather than inserted in the middle.
1060 * If we can confirm that, we can avoid binary searches on the
1061 * components of the pathname.
1062 *
1063 * Compare the entry's full path with the last path in the index.
1064 */
1065 if (istate->cache_nr > 0) {
1066 cmp_last = strcmp_offset(name,
1067 istate->cache[istate->cache_nr - 1]->name,
1068 &len_eq_last);
1069 if (cmp_last > 0) {
1070 if (len_eq_last == 0) {
1071 /*
1072 * The entry sorts AFTER the last one in the
1073 * index and their paths have no common prefix,
1074 * so there cannot be a F/D conflict.
1075 */
1076 return retval;
1077 } else {
1078 /*
1079 * The entry sorts AFTER the last one in the
1080 * index, but has a common prefix. Fall through
1081 * to the loop below to disect the entry's path
1082 * and see where the difference is.
1083 */
1084 }
1085 } else if (cmp_last == 0) {
1086 /*
1087 * The entry exactly matches the last one in the
1088 * index, but because of multiple stage and CE_REMOVE
1089 * items, we fall through and let the regular search
1090 * code handle it.
1091 */
1092 }
1093 }
0f1e4f04 1094
12676608 1095 for (;;) {
b986df5c 1096 size_t len;
0f1e4f04 1097
12676608
LT
1098 for (;;) {
1099 if (*--slash == '/')
1100 break;
1101 if (slash <= ce->name)
1102 return retval;
1103 }
1104 len = slash - name;
0f1e4f04 1105
b986df5c
JH
1106 if (cmp_last > 0) {
1107 /*
1108 * (len + 1) is a directory boundary (including
1109 * the trailing slash). And since the loop is
1110 * decrementing "slash", the first iteration is
1111 * the longest directory prefix; subsequent
1112 * iterations consider parent directories.
1113 */
1114
1115 if (len + 1 <= len_eq_last) {
1116 /*
1117 * The directory prefix (including the trailing
1118 * slash) also appears as a prefix in the last
1119 * entry, so the remainder cannot collide (because
1120 * strcmp said the whole path was greater).
1121 *
1122 * EQ: last: xxx/A
1123 * this: xxx/B
1124 *
1125 * LT: last: xxx/file_A
1126 * this: xxx/file_B
1127 */
1128 return retval;
1129 }
1130
1131 if (len > len_eq_last) {
1132 /*
1133 * This part of the directory prefix (excluding
1134 * the trailing slash) is longer than the known
1135 * equal portions, so this sub-directory cannot
1136 * collide with a file.
1137 *
1138 * GT: last: xxxA
1139 * this: xxxB/file
1140 */
1141 return retval;
1142 }
1143
1144 if (istate->cache_nr > 0 &&
1145 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1146 /*
1147 * The directory prefix lines up with part of
1148 * a longer file or directory name, but sorts
1149 * after it, so this sub-directory cannot
1150 * collide with a file.
1151 *
1152 * last: xxx/yy-file (because '-' sorts before '/')
1153 * this: xxx/yy/abc
1154 */
1155 return retval;
1156 }
1157
1158 /*
1159 * This is a possible collision. Fall through and
1160 * let the regular search code handle it.
1161 *
1162 * last: xxx
1163 * this: xxx/file
1164 */
1165 }
1166
b60e188c 1167 pos = index_name_stage_pos(istate, name, len, stage);
12676608 1168 if (pos >= 0) {
21cd8d00
JH
1169 /*
1170 * Found one, but not so fast. This could
1171 * be a marker that says "I was here, but
1172 * I am being removed". Such an entry is
1173 * not a part of the resulting tree, and
1174 * it is Ok to have a directory at the same
1175 * path.
1176 */
077c48df 1177 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
1178 retval = -1;
1179 if (!ok_to_replace)
1180 break;
4aab5b46 1181 remove_index_entry_at(istate, pos);
21cd8d00
JH
1182 continue;
1183 }
12676608 1184 }
21cd8d00
JH
1185 else
1186 pos = -pos-1;
12676608
LT
1187
1188 /*
1189 * Trivial optimization: if we find an entry that
1190 * already matches the sub-directory, then we know
b155725d 1191 * we're ok, and we can exit.
12676608 1192 */
4aab5b46
JH
1193 while (pos < istate->cache_nr) {
1194 struct cache_entry *p = istate->cache[pos];
b155725d
JH
1195 if ((ce_namelen(p) <= len) ||
1196 (p->name[len] != '/') ||
1197 memcmp(p->name, name, len))
1198 break; /* not our subdirectory */
077c48df
JH
1199 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1200 /*
1201 * p is at the same stage as our entry, and
b155725d
JH
1202 * is a subdirectory of what we are looking
1203 * at, so we cannot have conflicts at our
1204 * level or anything shorter.
1205 */
1206 return retval;
1207 pos++;
192268c1 1208 }
0f1e4f04 1209 }
12676608
LT
1210 return retval;
1211}
1212
1213/* We may be in a situation where we already have path/file and path
1214 * is being added, or we already have path and path/file is being
1215 * added. Either one would result in a nonsense tree that has path
1216 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 1217 *
12676608
LT
1218 * If ok-to-replace is specified, we remove the conflicting entries
1219 * from the cache so the caller should recompute the insert position.
1220 * When this happens, we return non-zero.
1221 */
4aab5b46
JH
1222static int check_file_directory_conflict(struct index_state *istate,
1223 const struct cache_entry *ce,
1224 int pos, int ok_to_replace)
12676608 1225{
21cd8d00
JH
1226 int retval;
1227
1228 /*
1229 * When ce is an "I am going away" entry, we allow it to be added
1230 */
7a51ed66 1231 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
1232 return 0;
1233
12676608
LT
1234 /*
1235 * We check if the path is a sub-path of a subsequent pathname
1236 * first, since removing those will not change the position
21cd8d00 1237 * in the array.
12676608 1238 */
4aab5b46 1239 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 1240
12676608
LT
1241 /*
1242 * Then check if the path might have a clashing sub-directory
1243 * before it.
1244 */
4aab5b46 1245 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
1246}
1247
af3785dc 1248static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
1249{
1250 int pos;
192268c1
JH
1251 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1252 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 1253 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 1254 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 1255
ce7c614b
NTND
1256 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1257 cache_tree_invalidate_path(istate, ce->name);
e5494631
JH
1258
1259 /*
1260 * If this entry's path sorts after the last entry in the index,
1261 * we can avoid searching for it.
1262 */
1263 if (istate->cache_nr > 0 &&
1264 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1265 pos = -istate->cache_nr - 1;
1266 else
1267 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
197ee8c9 1268
3e09cdfd 1269 /* existing match? Just replace it. */
76e7f4ec 1270 if (pos >= 0) {
39425819
JH
1271 if (!new_only)
1272 replace_index_entry(istate, pos, ce);
197ee8c9
LT
1273 return 0;
1274 }
76e7f4ec 1275 pos = -pos-1;
197ee8c9 1276
ffcc9ba7
NTND
1277 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1278 untracked_cache_add_to_index(istate, ce->name);
e931371a 1279
7b937ca3
LT
1280 /*
1281 * Inserting a merged entry ("stage 0") into the index
1282 * will always replace all non-merged entries..
1283 */
4aab5b46
JH
1284 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1285 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 1286 ok_to_add = 1;
4aab5b46 1287 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
1288 break;
1289 }
1290 }
1291
121481ab
LT
1292 if (!ok_to_add)
1293 return -1;
10ecfa76 1294 if (!verify_path(ce->name, ce->ce_mode))
7e7abea9 1295 return error("Invalid path '%s'", ce->name);
121481ab 1296
3e09cdfd 1297 if (!skip_df_check &&
4aab5b46 1298 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 1299 if (!ok_to_replace)
4aab5b46
JH
1300 return error("'%s' appears as both a file and as a directory",
1301 ce->name);
b60e188c 1302 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
192268c1
JH
1303 pos = -pos-1;
1304 }
af3785dc
JH
1305 return pos + 1;
1306}
1307
1308int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1309{
1310 int pos;
1311
1312 if (option & ADD_CACHE_JUST_APPEND)
1313 pos = istate->cache_nr;
1314 else {
1315 int ret;
1316 ret = add_index_entry_with_check(istate, ce, option);
1317 if (ret <= 0)
1318 return ret;
1319 pos = ret - 1;
1320 }
0f1e4f04 1321
197ee8c9 1322 /* Make sure the array is big enough .. */
999f5660 1323 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
197ee8c9
LT
1324
1325 /* Add it in.. */
4aab5b46 1326 istate->cache_nr++;
af3785dc 1327 if (istate->cache_nr > pos + 1)
f919ffeb
SG
1328 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1329 istate->cache_nr - pos - 1);
cf558704 1330 set_index_entry(istate, pos, ce);
e636a7b4 1331 istate->cache_changed |= CE_ENTRY_ADDED;
197ee8c9
LT
1332 return 0;
1333}
1334
405e5b2f
LT
1335/*
1336 * "refresh" does not calculate a new sha1 file or bring the
1337 * cache up-to-date for mode/content changes. But what it
1338 * _does_ do is to "re-match" the stat information of a file
1339 * with the cache, so that you can refresh the cache for a
1340 * file that hasn't been changed but where the stat entry is
1341 * out of date.
1342 *
1343 * For example, you'd want to do this after doing a "git-read-tree",
1344 * to link up the stat cache details with the proper files.
1345 */
4aab5b46 1346static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da 1347 struct cache_entry *ce,
d05e6970
JK
1348 unsigned int options, int *err,
1349 int *changed_ret)
405e5b2f
LT
1350{
1351 struct stat st;
1352 struct cache_entry *updated;
a849735b 1353 int changed;
25762726 1354 int refresh = options & CE_MATCH_REFRESH;
4bd5b7da 1355 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1356 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
2e2e7ec1 1357 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
883e248b 1358 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
405e5b2f 1359
25762726 1360 if (!refresh || ce_uptodate(ce))
eadb5831
JH
1361 return ce;
1362
883e248b
BP
1363 if (!ignore_fsmonitor)
1364 refresh_fsmonitor(istate);
aa9349d4 1365 /*
56cac48c
NTND
1366 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1367 * that the change to the work tree does not matter and told
1368 * us not to worry.
aa9349d4 1369 */
56cac48c
NTND
1370 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1371 ce_mark_uptodate(ce);
1372 return ce;
1373 }
aa9349d4
MSO
1374 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1375 ce_mark_uptodate(ce);
1376 return ce;
1377 }
883e248b
BP
1378 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1379 ce_mark_uptodate(ce);
1380 return ce;
1381 }
aa9349d4 1382
ccad42d4
RS
1383 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1384 if (ignore_missing)
1385 return ce;
1386 if (err)
1387 *err = ENOENT;
1388 return NULL;
1389 }
1390
8fd2cb40 1391 if (lstat(ce->name, &st) < 0) {
2e2e7ec1
BK
1392 if (ignore_missing && errno == ENOENT)
1393 return ce;
ec0cc704
JH
1394 if (err)
1395 *err = errno;
8fd2cb40
JS
1396 return NULL;
1397 }
405e5b2f 1398
4bd5b7da 1399 changed = ie_match_stat(istate, ce, &st, options);
d05e6970
JK
1400 if (changed_ret)
1401 *changed_ret = changed;
405e5b2f 1402 if (!changed) {
4bd5b7da
JH
1403 /*
1404 * The path is unchanged. If we were told to ignore
1405 * valid bit, then we did the actual stat check and
1406 * found that the entry is unmodified. If the entry
1407 * is not marked VALID, this is the place to mark it
1408 * valid again, under "assume unchanged" mode.
1409 */
1410 if (ignore_valid && assume_unchanged &&
7a51ed66 1411 !(ce->ce_flags & CE_VALID))
405e5b2f 1412 ; /* mark this one VALID again */
eadb5831
JH
1413 else {
1414 /*
1415 * We do not mark the index itself "modified"
1416 * because CE_UPTODATE flag is in-core only;
1417 * we are not going to write this change out.
1418 */
883e248b 1419 if (!S_ISGITLINK(ce->ce_mode)) {
125fd984 1420 ce_mark_uptodate(ce);
883e248b
BP
1421 mark_fsmonitor_valid(ce);
1422 }
8fd2cb40 1423 return ce;
eadb5831 1424 }
405e5b2f
LT
1425 }
1426
4bd5b7da 1427 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1428 if (err)
1429 *err = EINVAL;
8fd2cb40
JS
1430 return NULL;
1431 }
405e5b2f 1432
a849735b 1433 updated = make_empty_cache_entry(istate, ce_namelen(ce));
0e267b7a
BP
1434 copy_cache_entry(updated, ce);
1435 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
405e5b2f 1436 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1437 /*
1438 * If ignore_valid is not set, we should leave CE_VALID bit
1439 * alone. Otherwise, paths marked with --no-assume-unchanged
1440 * (i.e. things to be edited) will reacquire CE_VALID bit
1441 * automatically, which is not really what we want.
405e5b2f 1442 */
4bd5b7da 1443 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1444 !(ce->ce_flags & CE_VALID))
1445 updated->ce_flags &= ~CE_VALID;
405e5b2f 1446
e636a7b4 1447 /* istate->cache_changed is updated in the caller */
405e5b2f
LT
1448 return updated;
1449}
1450
3deffc52 1451static void show_file(const char * fmt, const char * name, int in_porcelain,
046613c5 1452 int * first, const char *header_msg)
3deffc52
MM
1453{
1454 if (in_porcelain && *first && header_msg) {
1455 printf("%s\n", header_msg);
cd2b8ae9 1456 *first = 0;
3deffc52
MM
1457 }
1458 printf(fmt, name);
1459}
1460
9b2d6149
NTND
1461int refresh_index(struct index_state *istate, unsigned int flags,
1462 const struct pathspec *pathspec,
046613c5 1463 char *seen, const char *header_msg)
405e5b2f
LT
1464{
1465 int i;
1466 int has_errors = 0;
1467 int really = (flags & REFRESH_REALLY) != 0;
1468 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1469 int quiet = (flags & REFRESH_QUIET) != 0;
1470 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1471 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1472 int first = 1;
1473 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
25762726
BK
1474 unsigned int options = (CE_MATCH_REFRESH |
1475 (really ? CE_MATCH_IGNORE_VALID : 0) |
2e2e7ec1 1476 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
4bd4e730 1477 const char *modified_fmt;
73b7eae6
JK
1478 const char *deleted_fmt;
1479 const char *typechange_fmt;
1480 const char *added_fmt;
4bd4e730 1481 const char *unmerged_fmt;
405e5b2f 1482
c46c406a 1483 trace_performance_enter();
4bd4e730 1484 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
73b7eae6
JK
1485 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1486 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1487 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
4bd4e730 1488 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
4aab5b46 1489 for (i = 0; i < istate->cache_nr; i++) {
285c2e25 1490 struct cache_entry *ce, *new_entry;
ec0cc704 1491 int cache_errno = 0;
73b7eae6 1492 int changed = 0;
3d1f148c 1493 int filtered = 0;
ec0cc704 1494
4aab5b46 1495 ce = istate->cache[i];
5fdeacb0
JS
1496 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1497 continue;
1498
6d2df284 1499 if (pathspec && !ce_path_match(&the_index, ce, pathspec, seen))
3d1f148c
JH
1500 filtered = 1;
1501
405e5b2f 1502 if (ce_stage(ce)) {
4aab5b46
JH
1503 while ((i < istate->cache_nr) &&
1504 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1505 i++;
1506 i--;
1507 if (allow_unmerged)
1508 continue;
3d1f148c
JH
1509 if (!filtered)
1510 show_file(unmerged_fmt, ce->name, in_porcelain,
1511 &first, header_msg);
405e5b2f
LT
1512 has_errors = 1;
1513 continue;
1514 }
1515
3d1f148c 1516 if (filtered)
d616813d
AJ
1517 continue;
1518
285c2e25
BW
1519 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1520 if (new_entry == ce)
405e5b2f 1521 continue;
285c2e25 1522 if (!new_entry) {
73b7eae6
JK
1523 const char *fmt;
1524
8fd2cb40 1525 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1526 /* If we are doing --really-refresh that
1527 * means the index is not valid anymore.
1528 */
7a51ed66 1529 ce->ce_flags &= ~CE_VALID;
078a58e8 1530 ce->ce_flags |= CE_UPDATE_IN_BASE;
883e248b 1531 mark_fsmonitor_invalid(istate, ce);
e636a7b4 1532 istate->cache_changed |= CE_ENTRY_CHANGED;
405e5b2f
LT
1533 }
1534 if (quiet)
1535 continue;
73b7eae6
JK
1536
1537 if (cache_errno == ENOENT)
1538 fmt = deleted_fmt;
895ff3b2 1539 else if (ce_intent_to_add(ce))
73b7eae6
JK
1540 fmt = added_fmt; /* must be before other checks */
1541 else if (changed & TYPE_CHANGED)
1542 fmt = typechange_fmt;
1543 else
1544 fmt = modified_fmt;
1545 show_file(fmt,
1546 ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1547 has_errors = 1;
1548 continue;
1549 }
cf558704 1550
285c2e25 1551 replace_index_entry(istate, i, new_entry);
405e5b2f 1552 }
c46c406a 1553 trace_performance_leave("refresh index");
405e5b2f
LT
1554 return has_errors;
1555}
1556
768d7965
JM
1557struct cache_entry *refresh_cache_entry(struct index_state *istate,
1558 struct cache_entry *ce,
1559 unsigned int options)
ec0cc704 1560{
768d7965 1561 return refresh_cache_ent(istate, ce, options, NULL, NULL);
ec0cc704
JH
1562}
1563
db3b313c
JH
1564
1565/*****************************************************************
1566 * Index File I/O
1567 *****************************************************************/
1568
9d227781
JH
1569#define INDEX_FORMAT_DEFAULT 3
1570
136347d7
TG
1571static unsigned int get_index_format_default(void)
1572{
1573 char *envversion = getenv("GIT_INDEX_VERSION");
3c09d684 1574 char *endp;
b27a5720 1575 int value;
3c09d684
TG
1576 unsigned int version = INDEX_FORMAT_DEFAULT;
1577
136347d7 1578 if (!envversion) {
b27a5720
TA
1579 if (!git_config_get_int("index.version", &value))
1580 version = value;
3c09d684
TG
1581 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1582 warning(_("index.version set, but the value is invalid.\n"
136347d7 1583 "Using version %i"), INDEX_FORMAT_DEFAULT);
3c09d684 1584 return INDEX_FORMAT_DEFAULT;
136347d7
TG
1585 }
1586 return version;
1587 }
3c09d684
TG
1588
1589 version = strtoul(envversion, &endp, 10);
1590 if (*endp ||
1591 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1592 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1593 "Using version %i"), INDEX_FORMAT_DEFAULT);
1594 version = INDEX_FORMAT_DEFAULT;
1595 }
1596 return version;
136347d7
TG
1597}
1598
db3b313c
JH
1599/*
1600 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1601 * Again - this is just a (very strong in practice) heuristic that
1602 * the inode hasn't changed.
1603 *
1604 * We save the fields in big-endian order to allow using the
1605 * index file over NFS transparently.
1606 */
1607struct ondisk_cache_entry {
1608 struct cache_time ctime;
1609 struct cache_time mtime;
7800c1eb
TG
1610 uint32_t dev;
1611 uint32_t ino;
1612 uint32_t mode;
1613 uint32_t uid;
1614 uint32_t gid;
1615 uint32_t size;
db3b313c 1616 unsigned char sha1[20];
7800c1eb 1617 uint16_t flags;
db3b313c
JH
1618 char name[FLEX_ARRAY]; /* more */
1619};
1620
1621/*
1622 * This struct is used when CE_EXTENDED bit is 1
1623 * The struct must match ondisk_cache_entry exactly from
1624 * ctime till flags
1625 */
1626struct ondisk_cache_entry_extended {
1627 struct cache_time ctime;
1628 struct cache_time mtime;
7800c1eb
TG
1629 uint32_t dev;
1630 uint32_t ino;
1631 uint32_t mode;
1632 uint32_t uid;
1633 uint32_t gid;
1634 uint32_t size;
db3b313c 1635 unsigned char sha1[20];
7800c1eb
TG
1636 uint16_t flags;
1637 uint16_t flags2;
db3b313c
JH
1638 char name[FLEX_ARRAY]; /* more */
1639};
1640
6c9cd161 1641/* These are only used for v3 or lower */
ce012deb 1642#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
db3b313c
JH
1643#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1644#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1645#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1646#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1647 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1648 ondisk_cache_entry_size(ce_namelen(ce)))
1649
a33fc72f
JH
1650/* Allow fsck to force verification of the index checksum. */
1651int verify_index_checksum;
1652
00ec50e5
BP
1653/* Allow fsck to force verification of the cache entry order. */
1654int verify_ce_order;
1655
371ed0de 1656static int verify_hdr(const struct cache_header *hdr, unsigned long size)
e83c5163 1657{
aab61359 1658 git_hash_ctx c;
1659 unsigned char hash[GIT_MAX_RAWSZ];
0136bac9 1660 int hdr_version;
e83c5163 1661
ccc4feb5 1662 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1663 return error("bad signature");
0136bac9 1664 hdr_version = ntohl(hdr->hdr_version);
b82a7b5b 1665 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
0136bac9 1666 return error("bad index version %d", hdr_version);
a33fc72f
JH
1667
1668 if (!verify_index_checksum)
1669 return 0;
1670
aab61359 1671 the_hash_algo->init_fn(&c);
1672 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1673 the_hash_algo->final_fn(hash, &c);
67947c34 1674 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
ca9be054 1675 return error("bad index file sha1 signature");
e83c5163
LT
1676 return 0;
1677}
1678
4aab5b46 1679static int read_index_extension(struct index_state *istate,
371ed0de 1680 const char *ext, const char *data, unsigned long sz)
bad68ec9
JH
1681{
1682 switch (CACHE_EXT(ext)) {
1683 case CACHE_EXT_TREE:
4aab5b46 1684 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1685 break;
cfc5789a
JH
1686 case CACHE_EXT_RESOLVE_UNDO:
1687 istate->resolve_undo = resolve_undo_read(data, sz);
1688 break;
5fc2fc8f
NTND
1689 case CACHE_EXT_LINK:
1690 if (read_link_extension(istate, data, sz))
1691 return -1;
1692 break;
f9e6c649
NTND
1693 case CACHE_EXT_UNTRACKED:
1694 istate->untracked = read_untracked_extension(data, sz);
1695 break;
883e248b
BP
1696 case CACHE_EXT_FSMONITOR:
1697 read_fsmonitor_extension(istate, data, sz);
1698 break;
3b1d9e04 1699 case CACHE_EXT_ENDOFINDEXENTRIES:
3255089a 1700 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
3b1d9e04
BP
1701 /* already handled in do_read_index() */
1702 break;
bad68ec9
JH
1703 default:
1704 if (*ext < 'A' || 'Z' < *ext)
1705 return error("index uses %.4s extension, which we do not understand",
1706 ext);
1707 fprintf(stderr, "ignoring %.4s extension\n", ext);
1708 break;
1709 }
1710 return 0;
1711}
1712
b3e83cc7 1713int hold_locked_index(struct lock_file *lk, int lock_flags)
216aab1e 1714{
b3e83cc7 1715 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
216aab1e
MH
1716}
1717
4aab5b46 1718int read_index(struct index_state *istate)
8fd2cb40 1719{
a125a223 1720 return read_index_from(istate, get_index_file(), get_git_dir());
8fd2cb40
JS
1721}
1722
252d079c 1723static struct cache_entry *create_from_disk(struct index_state *istate,
a849735b 1724 struct ondisk_cache_entry *ondisk,
6c9cd161 1725 unsigned long *ent_size,
252d079c 1726 const struct cache_entry *previous_ce)
7a51ed66 1727{
debed2a6 1728 struct cache_entry *ce;
7fec10b7 1729 size_t len;
06aaaa0b 1730 const char *name;
debed2a6 1731 unsigned int flags;
252d079c
NTND
1732 size_t copy_len;
1733 /*
1734 * Adjacent cache entries tend to share the leading paths, so it makes
1735 * sense to only store the differences in later entries. In the v4
1736 * on-disk format of the index, each on-disk cache entry stores the
1737 * number of bytes to be stripped from the end of the previous name,
1738 * and the bytes to append to the result, to come up with its name.
1739 */
1740 int expand_name_field = istate->version == 4;
7fec10b7 1741
7a51ed66 1742 /* On-disk flags are just 16 bits */
c3d8da57 1743 flags = get_be16(&ondisk->flags);
debed2a6 1744 len = flags & CE_NAMEMASK;
7fec10b7 1745
debed2a6 1746 if (flags & CE_EXTENDED) {
06aaaa0b
NTND
1747 struct ondisk_cache_entry_extended *ondisk2;
1748 int extended_flags;
1749 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
c3d8da57 1750 extended_flags = get_be16(&ondisk2->flags2) << 16;
06aaaa0b
NTND
1751 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1752 if (extended_flags & ~CE_EXTENDED_FLAGS)
1753 die("Unknown index entry format %08x", extended_flags);
debed2a6 1754 flags |= extended_flags;
06aaaa0b
NTND
1755 name = ondisk2->name;
1756 }
1757 else
1758 name = ondisk->name;
1759
252d079c
NTND
1760 if (expand_name_field) {
1761 const unsigned char *cp = (const unsigned char *)name;
1762 size_t strip_len, previous_len;
6c9cd161 1763
252d079c
NTND
1764 previous_len = previous_ce ? previous_ce->ce_namelen : 0;
1765 strip_len = decode_varint(&cp);
1766 if (previous_len < strip_len) {
1767 if (previous_ce)
1768 die(_("malformed name field in the index, near path '%s'"),
1769 previous_ce->name);
1770 else
1771 die(_("malformed name field in the index in the first path"));
1772 }
1773 copy_len = previous_len - strip_len;
1774 name = (const char *)cp;
1775 }
1776
1777 if (len == CE_NAMEMASK) {
1778 len = strlen(name);
1779 if (expand_name_field)
1780 len += copy_len;
1781 }
1782
1783 ce = mem_pool__ce_alloc(istate->ce_mem_pool, len);
1784
1785 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1786 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1787 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1788 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1789 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1790 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1791 ce->ce_mode = get_be32(&ondisk->mode);
1792 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1793 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1794 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1795 ce->ce_flags = flags & ~CE_NAMEMASK;
1796 ce->ce_namelen = len;
1797 ce->index = 0;
1798 hashcpy(ce->oid.hash, ondisk->sha1);
6c9cd161 1799
252d079c
NTND
1800 if (expand_name_field) {
1801 if (copy_len)
1802 memcpy(ce->name, previous_ce->name, copy_len);
1803 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1804 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1805 } else {
1806 memcpy(ce->name, name, len + 1);
1807 *ent_size = ondisk_ce_size(ce);
6c9cd161 1808 }
debed2a6 1809 return ce;
cf558704
LT
1810}
1811
03f15a79 1812static void check_ce_order(struct index_state *istate)
15999d0b 1813{
03f15a79
TG
1814 unsigned int i;
1815
00ec50e5
BP
1816 if (!verify_ce_order)
1817 return;
1818
03f15a79
TG
1819 for (i = 1; i < istate->cache_nr; i++) {
1820 struct cache_entry *ce = istate->cache[i - 1];
1821 struct cache_entry *next_ce = istate->cache[i];
1822 int name_compare = strcmp(ce->name, next_ce->name);
1823
1824 if (0 < name_compare)
1825 die("unordered stage entries in index");
1826 if (!name_compare) {
1827 if (!ce_stage(ce))
1828 die("multiple stage entries for merged file '%s'",
1829 ce->name);
1830 if (ce_stage(ce) > ce_stage(next_ce))
1831 die("unordered stage entries for '%s'",
1832 ce->name);
1833 }
15999d0b
JSP
1834 }
1835}
1836
435ec090
CC
1837static void tweak_untracked_cache(struct index_state *istate)
1838{
1839 switch (git_config_get_untracked_cache()) {
1840 case -1: /* keep: do nothing */
1841 break;
1842 case 0: /* false */
1843 remove_untracked_cache(istate);
1844 break;
1845 case 1: /* true */
1846 add_untracked_cache(istate);
1847 break;
1848 default: /* unknown value: do nothing */
1849 break;
1850 }
1851}
1852
43925312
CC
1853static void tweak_split_index(struct index_state *istate)
1854{
1855 switch (git_config_get_split_index()) {
1856 case -1: /* unset: do nothing */
1857 break;
1858 case 0: /* false */
1859 remove_split_index(istate);
1860 break;
1861 case 1: /* true */
1862 add_split_index(istate);
1863 break;
1864 default: /* unknown value: do nothing */
1865 break;
1866 }
1867}
1868
435ec090
CC
1869static void post_read_index_from(struct index_state *istate)
1870{
1871 check_ce_order(istate);
1872 tweak_untracked_cache(istate);
43925312 1873 tweak_split_index(istate);
883e248b 1874 tweak_fsmonitor(istate);
435ec090
CC
1875}
1876
8e72d675
JM
1877static size_t estimate_cache_size_from_compressed(unsigned int entries)
1878{
1879 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1880}
1881
1882static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1883{
1884 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1885
1886 /*
1887 * Account for potential alignment differences.
1888 */
1889 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1890 return ondisk_size + entries * per_entry;
1891}
1892
3255089a
BP
1893struct index_entry_offset
1894{
1895 /* starting byte offset into index file, count of index entries in this block */
1896 int offset, nr;
1897};
1898
1899struct index_entry_offset_table
1900{
1901 int nr;
1902 struct index_entry_offset entries[FLEX_ARRAY];
1903};
1904
1905#ifndef NO_PTHREADS
1906static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1907static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1908#endif
1909
3b1d9e04
BP
1910static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1911static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1912
abb4bb83
BP
1913struct load_index_extensions
1914{
1915#ifndef NO_PTHREADS
1916 pthread_t pthread;
1917#endif
1918 struct index_state *istate;
1919 const char *mmap;
1920 size_t mmap_size;
1921 unsigned long src_offset;
1922};
1923
1924static void *load_index_extensions(void *_data)
1925{
1926 struct load_index_extensions *p = _data;
1927 unsigned long src_offset = p->src_offset;
1928
1929 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1930 /* After an array of active_nr index entries,
1931 * there can be arbitrary number of extended
1932 * sections, each of which is prefixed with
1933 * extension name (4-byte) and section length
1934 * in 4-byte network byte order.
1935 */
1936 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1937 if (read_index_extension(p->istate,
1938 p->mmap + src_offset,
1939 p->mmap + src_offset + 8,
1940 extsize) < 0) {
1941 munmap((void *)p->mmap, p->mmap_size);
1942 die(_("index file corrupt"));
1943 }
1944 src_offset += 8;
1945 src_offset += extsize;
1946 }
1947
1948 return NULL;
1949}
1950
3255089a
BP
1951/*
1952 * Mostly randomly chosen maximum thread counts: we
1953 * cap the parallelism to online_cpus() threads, and we want
1954 * to have at least 10000 cache entries per thread for it to
1955 * be worth starting a thread.
1956 */
1957
1958#define THREAD_COST (10000)
1959
8fd2cb40 1960/* remember to discard_cache() before reading a different cache! */
3e52f70b 1961int do_read_index(struct index_state *istate, const char *path, int must_exist)
e83c5163
LT
1962{
1963 int fd, i;
1964 struct stat st;
debed2a6 1965 unsigned long src_offset;
371ed0de
BP
1966 const struct cache_header *hdr;
1967 const char *mmap;
7a51ed66 1968 size_t mmap_size;
252d079c 1969 const struct cache_entry *previous_ce = NULL;
abb4bb83
BP
1970 struct load_index_extensions p;
1971 size_t extension_offset = 0;
1972#ifndef NO_PTHREADS
1973 int nr_threads;
1974#endif
e83c5163 1975
913e0e99 1976 if (istate->initialized)
4aab5b46 1977 return istate->cache_nr;
5d1a5c02 1978
fba2f38a
KB
1979 istate->timestamp.sec = 0;
1980 istate->timestamp.nsec = 0;
8fd2cb40 1981 fd = open(path, O_RDONLY);
5d1a5c02 1982 if (fd < 0) {
5fc2fc8f 1983 if (!must_exist && errno == ENOENT)
5d1a5c02 1984 return 0;
5fc2fc8f 1985 die_errno("%s: index file open failed", path);
5d1a5c02 1986 }
e83c5163 1987
3511a377 1988 if (fstat(fd, &st))
d824cbba 1989 die_errno("cannot stat the open index");
3511a377 1990
7a51ed66 1991 mmap_size = xsize_t(st.st_size);
aab61359 1992 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
3511a377
LFC
1993 die("index file smaller than expected");
1994
a1293ef7 1995 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
7a51ed66 1996 if (mmap == MAP_FAILED)
0721c314 1997 die_errno("unable to map index file");
57d84f8d 1998 close(fd);
e83c5163 1999
371ed0de 2000 hdr = (const struct cache_header *)mmap;
7a51ed66 2001 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
2002 goto unmap;
2003
75691ea3 2004 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
9d227781 2005 istate->version = ntohl(hdr->hdr_version);
4aab5b46
JH
2006 istate->cache_nr = ntohl(hdr->hdr_entries);
2007 istate->cache_alloc = alloc_nr(istate->cache_nr);
c4aa3167 2008 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
913e0e99 2009 istate->initialized = 1;
7a51ed66 2010
abb4bb83
BP
2011 p.istate = istate;
2012 p.mmap = mmap;
2013 p.mmap_size = mmap_size;
2014
2015#ifndef NO_PTHREADS
2016 nr_threads = git_config_get_index_threads();
2017 if (!nr_threads)
2018 nr_threads = online_cpus();
2019
2020 if (nr_threads > 1) {
2021 extension_offset = read_eoie_extension(mmap, mmap_size);
2022 if (extension_offset) {
2023 int err;
2024
2025 p.src_offset = extension_offset;
2026 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2027 if (err)
2028 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2029
2030 nr_threads--;
2031 }
2032 }
2033#endif
2034
8e72d675 2035 if (istate->version == 4) {
8e72d675
JM
2036 mem_pool_init(&istate->ce_mem_pool,
2037 estimate_cache_size_from_compressed(istate->cache_nr));
2038 } else {
8e72d675
JM
2039 mem_pool_init(&istate->ce_mem_pool,
2040 estimate_cache_size(mmap_size, istate->cache_nr));
2041 }
6c9cd161 2042
7a51ed66 2043 src_offset = sizeof(*hdr);
4aab5b46 2044 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 2045 struct ondisk_cache_entry *disk_ce;
4aab5b46 2046 struct cache_entry *ce;
936f53d0 2047 unsigned long consumed;
4aab5b46 2048
371ed0de 2049 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
252d079c 2050 ce = create_from_disk(istate, disk_ce, &consumed, previous_ce);
cf558704 2051 set_index_entry(istate, i, ce);
7a51ed66 2052
936f53d0 2053 src_offset += consumed;
252d079c 2054 previous_ce = ce;
e83c5163 2055 }
fba2f38a 2056 istate->timestamp.sec = st.st_mtime;
c06ff490 2057 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 2058
abb4bb83
BP
2059 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2060#ifndef NO_PTHREADS
2061 if (extension_offset) {
2062 int ret = pthread_join(p.pthread, NULL);
2063 if (ret)
2064 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2065 }
2066#endif
2067 if (!extension_offset) {
2068 p.src_offset = src_offset;
2069 load_index_extensions(&p);
bad68ec9 2070 }
371ed0de 2071 munmap((void *)mmap, mmap_size);
4aab5b46 2072 return istate->cache_nr;
e83c5163
LT
2073
2074unmap:
371ed0de 2075 munmap((void *)mmap, mmap_size);
5d1a5c02 2076 die("index file corrupt");
e83c5163
LT
2077}
2078
0d59ffb4
CC
2079/*
2080 * Signal that the shared index is used by updating its mtime.
2081 *
2082 * This way, shared index can be removed if they have not been used
2083 * for some time.
2084 */
a125a223 2085static void freshen_shared_index(const char *shared_index, int warn)
0d59ffb4 2086{
0d59ffb4
CC
2087 if (!check_and_freshen_file(shared_index, 1) && warn)
2088 warning("could not freshen shared index '%s'", shared_index);
2089}
2090
a125a223
TG
2091int read_index_from(struct index_state *istate, const char *path,
2092 const char *gitdir)
5fc2fc8f
NTND
2093{
2094 struct split_index *split_index;
2095 int ret;
2182abd9 2096 char *base_oid_hex;
a125a223 2097 char *base_path;
5fc2fc8f
NTND
2098
2099 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2100 if (istate->initialized)
2101 return istate->cache_nr;
2102
c46c406a 2103 trace_performance_enter();
5fc2fc8f 2104 ret = do_read_index(istate, path, 0);
c46c406a 2105 trace_performance_leave("read cache %s", path);
435ec090 2106
5fc2fc8f 2107 split_index = istate->split_index;
2182abd9 2108 if (!split_index || is_null_oid(&split_index->base_oid)) {
435ec090 2109 post_read_index_from(istate);
5fc2fc8f 2110 return ret;
03f15a79 2111 }
5fc2fc8f 2112
c46c406a 2113 trace_performance_enter();
5fc2fc8f
NTND
2114 if (split_index->base)
2115 discard_index(split_index->base);
2116 else
2117 split_index->base = xcalloc(1, sizeof(*split_index->base));
de6ae5f9 2118
2182abd9 2119 base_oid_hex = oid_to_hex(&split_index->base_oid);
2120 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
de6ae5f9 2121 ret = do_read_index(split_index->base, base_path, 1);
9001dc2a 2122 if (!oideq(&split_index->base_oid, &split_index->base->oid))
5fc2fc8f 2123 die("broken index, expect %s in %s, got %s",
2182abd9 2124 base_oid_hex, base_path,
75691ea3 2125 oid_to_hex(&split_index->base->oid));
de6ae5f9 2126
a125a223 2127 freshen_shared_index(base_path, 0);
5fc2fc8f 2128 merge_base_index(istate);
435ec090 2129 post_read_index_from(istate);
a125a223 2130 free(base_path);
c46c406a 2131 trace_performance_leave("read cache %s", base_path);
5fc2fc8f
NTND
2132 return ret;
2133}
2134
fa7b3c2f
JH
2135int is_index_unborn(struct index_state *istate)
2136{
debed2a6 2137 return (!istate->cache_nr && !istate->timestamp.sec);
fa7b3c2f
JH
2138}
2139
4aab5b46 2140int discard_index(struct index_state *istate)
6d297f81 2141{
8e72d675
JM
2142 /*
2143 * Cache entries in istate->cache[] should have been allocated
2144 * from the memory pool associated with this index, or from an
2145 * associated split_index. There is no need to free individual
8616a2d0
JM
2146 * cache entries. validate_cache_entries can detect when this
2147 * assertion does not hold.
8e72d675 2148 */
8616a2d0 2149 validate_cache_entries(istate);
debed2a6 2150
cfc5789a 2151 resolve_undo_clear_index(istate);
4aab5b46
JH
2152 istate->cache_nr = 0;
2153 istate->cache_changed = 0;
fba2f38a
KB
2154 istate->timestamp.sec = 0;
2155 istate->timestamp.nsec = 0;
2092678c 2156 free_name_hash(istate);
4aab5b46 2157 cache_tree_free(&(istate->cache_tree));
913e0e99 2158 istate->initialized = 0;
6a83d902 2159 FREE_AND_NULL(istate->cache);
a0fc4db0 2160 istate->cache_alloc = 0;
5fc2fc8f 2161 discard_split_index(istate);
f9e6c649
NTND
2162 free_untracked_cache(istate->untracked);
2163 istate->untracked = NULL;
8e72d675
JM
2164
2165 if (istate->ce_mem_pool) {
8616a2d0 2166 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
8e72d675
JM
2167 istate->ce_mem_pool = NULL;
2168 }
2169
7a51ed66 2170 return 0;
6d297f81
JS
2171}
2172
8616a2d0
JM
2173/*
2174 * Validate the cache entries of this index.
2175 * All cache entries associated with this index
2176 * should have been allocated by the memory pool
2177 * associated with this index, or by a referenced
2178 * split index.
2179 */
2180void validate_cache_entries(const struct index_state *istate)
2181{
2182 int i;
2183
2184 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2185 return;
2186
2187 for (i = 0; i < istate->cache_nr; i++) {
2188 if (!istate) {
2189 die("internal error: cache entry is not allocated from expected memory pool");
2190 } else if (!istate->ce_mem_pool ||
2191 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2192 if (!istate->split_index ||
2193 !istate->split_index->base ||
2194 !istate->split_index->base->ce_mem_pool ||
2195 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2196 die("internal error: cache entry is not allocated from expected memory pool");
2197 }
2198 }
2199 }
2200
2201 if (istate->split_index)
2202 validate_cache_entries(istate->split_index->base);
2203}
2204
d1f128b0 2205int unmerged_index(const struct index_state *istate)
94a5728c
DB
2206{
2207 int i;
2208 for (i = 0; i < istate->cache_nr; i++) {
2209 if (ce_stage(istate->cache[i]))
2210 return 1;
2211 }
2212 return 0;
2213}
2214
e1f8694f
EN
2215int index_has_changes(const struct index_state *istate,
2216 struct tree *tree,
2217 struct strbuf *sb)
cffbfad5 2218{
e1f8694f 2219 struct object_id cmp;
cffbfad5
EN
2220 int i;
2221
1b9fbefb
EN
2222 if (istate != &the_index) {
2223 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2224 }
e1f8694f
EN
2225 if (tree)
2226 cmp = tree->object.oid;
2227 if (tree || !get_oid_tree("HEAD", &cmp)) {
cffbfad5
EN
2228 struct diff_options opt;
2229
2230 diff_setup(&opt);
2231 opt.flags.exit_with_status = 1;
2232 if (!sb)
2233 opt.flags.quick = 1;
e1f8694f 2234 do_diff_cache(&cmp, &opt);
cffbfad5
EN
2235 diffcore_std(&opt);
2236 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2237 if (i)
2238 strbuf_addch(sb, ' ');
2239 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2240 }
2241 diff_flush(&opt);
2242 return opt.flags.has_changes != 0;
2243 } else {
1b9fbefb 2244 for (i = 0; sb && i < istate->cache_nr; i++) {
cffbfad5
EN
2245 if (i)
2246 strbuf_addch(sb, ' ');
1b9fbefb 2247 strbuf_addstr(sb, istate->cache[i]->name);
cffbfad5 2248 }
1b9fbefb 2249 return !!istate->cache_nr;
cffbfad5
EN
2250 }
2251}
2252
4990aadc 2253#define WRITE_BUFFER_SIZE 8192
bf0f910d 2254static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
2255static unsigned long write_buffer_len;
2256
aab61359 2257static int ce_write_flush(git_hash_ctx *context, int fd)
6015c28b
JH
2258{
2259 unsigned int buffered = write_buffer_len;
2260 if (buffered) {
aab61359 2261 the_hash_algo->update_fn(context, write_buffer, buffered);
06f46f23 2262 if (write_in_full(fd, write_buffer, buffered) < 0)
6015c28b
JH
2263 return -1;
2264 write_buffer_len = 0;
2265 }
2266 return 0;
2267}
2268
aab61359 2269static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
4990aadc
LT
2270{
2271 while (len) {
2272 unsigned int buffered = write_buffer_len;
2273 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2274 if (partial > len)
2275 partial = len;
2276 memcpy(write_buffer + buffered, data, partial);
2277 buffered += partial;
2278 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
2279 write_buffer_len = buffered;
2280 if (ce_write_flush(context, fd))
4990aadc
LT
2281 return -1;
2282 buffered = 0;
2283 }
2284 write_buffer_len = buffered;
2285 len -= partial;
1d7f171c 2286 data = (char *) data + partial;
a6080a0a
JH
2287 }
2288 return 0;
4990aadc
LT
2289}
2290
3b1d9e04
BP
2291static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2292 int fd, unsigned int ext, unsigned int sz)
bad68ec9
JH
2293{
2294 ext = htonl(ext);
2295 sz = htonl(sz);
3b1d9e04
BP
2296 if (eoie_context) {
2297 the_hash_algo->update_fn(eoie_context, &ext, 4);
2298 the_hash_algo->update_fn(eoie_context, &sz, 4);
2299 }
968a1d65
DR
2300 return ((ce_write(context, fd, &ext, 4) < 0) ||
2301 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
2302}
2303
aab61359 2304static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
4990aadc
LT
2305{
2306 unsigned int left = write_buffer_len;
ca9be054 2307
4990aadc
LT
2308 if (left) {
2309 write_buffer_len = 0;
aab61359 2310 the_hash_algo->update_fn(context, write_buffer, left);
4990aadc 2311 }
ca9be054 2312
aab61359 2313 /* Flush first if not enough space for hash signature */
2314 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
06f46f23 2315 if (write_in_full(fd, write_buffer, left) < 0)
2c865d9a
QH
2316 return -1;
2317 left = 0;
2318 }
2319
aab61359 2320 /* Append the hash signature at the end */
2321 the_hash_algo->final_fn(write_buffer + left, context);
2322 hashcpy(hash, write_buffer + left);
2323 left += the_hash_algo->rawsz;
06f46f23 2324 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
4990aadc
LT
2325}
2326
407c8eb0
JH
2327static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2328{
2329 /*
2330 * The only thing we care about in this function is to smudge the
2331 * falsely clean entry due to touch-update-touch race, so we leave
2332 * everything else as they are. We are called for entries whose
c21d39d7 2333 * ce_stat_data.sd_mtime match the index file mtime.
c70115b4
JH
2334 *
2335 * Note that this actually does not do much for gitlinks, for
2336 * which ce_match_stat_basic() always goes to the actual
2337 * contents. The caller checks with is_racy_timestamp() which
2338 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
2339 */
2340 struct stat st;
2341
2342 if (lstat(ce->name, &st) < 0)
2343 return;
2344 if (ce_match_stat_basic(ce, &st))
2345 return;
2346 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
2347 /* This is "racily clean"; smudge it. Note that this
2348 * is a tricky code. At first glance, it may appear
2349 * that it can break with this sequence:
2350 *
2351 * $ echo xyzzy >frotz
2352 * $ git-update-index --add frotz
2353 * $ : >frotz
2354 * $ sleep 3
2355 * $ echo filfre >nitfol
2356 * $ git-update-index --add nitfol
2357 *
b7e58b17 2358 * but it does not. When the second update-index runs,
4b3511b0
JH
2359 * it notices that the entry "frotz" has the same timestamp
2360 * as index, and if we were to smudge it by resetting its
2361 * size to zero here, then the object name recorded
2362 * in index is the 6-byte file but the cached stat information
2363 * becomes zero --- which would then match what we would
a6080a0a 2364 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
2365 *
2366 * However, the second update-index, before calling
2367 * this function, notices that the cached size is 6
2368 * bytes and what is on the filesystem is an empty
2369 * file, and never calls us, so the cached size information
2370 * for "frotz" stays 6 which does not match the filesystem.
2371 */
c21d39d7 2372 ce->ce_stat_data.sd_size = 0;
407c8eb0
JH
2373 }
2374}
2375
f136f7bf 2376/* Copy miscellaneous fields but not the name */
ce012deb 2377static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
f136f7bf 2378 struct cache_entry *ce)
7a51ed66 2379{
b60e188c
TG
2380 short flags;
2381
c21d39d7
MH
2382 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2383 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2384 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2385 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2386 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2387 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
7a51ed66 2388 ondisk->mode = htonl(ce->ce_mode);
c21d39d7
MH
2389 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2390 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2391 ondisk->size = htonl(ce->ce_stat_data.sd_size);
99d1a986 2392 hashcpy(ondisk->sha1, ce->oid.hash);
b60e188c 2393
ce51bf09 2394 flags = ce->ce_flags & ~CE_NAMEMASK;
b60e188c
TG
2395 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2396 ondisk->flags = htons(flags);
06aaaa0b
NTND
2397 if (ce->ce_flags & CE_EXTENDED) {
2398 struct ondisk_cache_entry_extended *ondisk2;
2399 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2400 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
f136f7bf
JH
2401 }
2402}
2403
aab61359 2404static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
ce012deb 2405 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
f136f7bf 2406{
9d227781 2407 int size;
f136f7bf 2408 int result;
00a4b035
RJ
2409 unsigned int saved_namelen;
2410 int stripped_name = 0;
ce012deb 2411 static unsigned char padding[8] = { 0x00 };
f136f7bf 2412
b3c96fb1
NTND
2413 if (ce->ce_flags & CE_STRIP_NAME) {
2414 saved_namelen = ce_namelen(ce);
2415 ce->ce_namelen = 0;
00a4b035 2416 stripped_name = 1;
b3c96fb1
NTND
2417 }
2418
ce012deb
KW
2419 if (ce->ce_flags & CE_EXTENDED)
2420 size = offsetof(struct ondisk_cache_entry_extended, name);
2421 else
2422 size = offsetof(struct ondisk_cache_entry, name);
2423
9d227781 2424 if (!previous_name) {
ce012deb
KW
2425 int len = ce_namelen(ce);
2426 copy_cache_entry_to_ondisk(ondisk, ce);
2427 result = ce_write(c, fd, ondisk, size);
2428 if (!result)
2429 result = ce_write(c, fd, ce->name, len);
2430 if (!result)
2431 result = ce_write(c, fd, padding, align_padding_size(size, len));
9d227781
JH
2432 } else {
2433 int common, to_remove, prefix_size;
2434 unsigned char to_remove_vi[16];
2435 for (common = 0;
2436 (ce->name[common] &&
2437 common < previous_name->len &&
2438 ce->name[common] == previous_name->buf[common]);
2439 common++)
2440 ; /* still matching */
2441 to_remove = previous_name->len - common;
2442 prefix_size = encode_varint(to_remove, to_remove_vi);
2443
ce012deb
KW
2444 copy_cache_entry_to_ondisk(ondisk, ce);
2445 result = ce_write(c, fd, ondisk, size);
2446 if (!result)
2447 result = ce_write(c, fd, to_remove_vi, prefix_size);
2448 if (!result)
0b90b881
TG
2449 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2450 if (!result)
2451 result = ce_write(c, fd, padding, 1);
9d227781
JH
2452
2453 strbuf_splice(previous_name, common, to_remove,
2454 ce->name + common, ce_namelen(ce) - common);
06aaaa0b 2455 }
00a4b035 2456 if (stripped_name) {
b3c96fb1
NTND
2457 ce->ce_namelen = saved_namelen;
2458 ce->ce_flags &= ~CE_STRIP_NAME;
2459 }
7a51ed66 2460
59efba64 2461 return result;
7a51ed66
LT
2462}
2463
426ddeea
YM
2464/*
2465 * This function verifies if index_state has the correct sha1 of the
2466 * index file. Don't die if we have any other failure, just return 0.
2467 */
2468static int verify_index_from(const struct index_state *istate, const char *path)
2469{
2470 int fd;
2471 ssize_t n;
2472 struct stat st;
aab61359 2473 unsigned char hash[GIT_MAX_RAWSZ];
426ddeea
YM
2474
2475 if (!istate->initialized)
2476 return 0;
2477
2478 fd = open(path, O_RDONLY);
2479 if (fd < 0)
2480 return 0;
2481
2482 if (fstat(fd, &st))
2483 goto out;
2484
aab61359 2485 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
426ddeea
YM
2486 goto out;
2487
aab61359 2488 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2489 if (n != the_hash_algo->rawsz)
426ddeea
YM
2490 goto out;
2491
67947c34 2492 if (!hasheq(istate->oid.hash, hash))
426ddeea
YM
2493 goto out;
2494
2495 close(fd);
2496 return 1;
2497
2498out:
2499 close(fd);
2500 return 0;
2501}
2502
2503static int verify_index(const struct index_state *istate)
2504{
2505 return verify_index_from(istate, get_index_file());
2506}
2507
483fbe2b
JH
2508static int has_racy_timestamp(struct index_state *istate)
2509{
2510 int entries = istate->cache_nr;
2511 int i;
2512
2513 for (i = 0; i < entries; i++) {
2514 struct cache_entry *ce = istate->cache[i];
2515 if (is_racy_timestamp(istate, ce))
2516 return 1;
2517 }
2518 return 0;
2519}
2520
ccdc4ec3
JH
2521void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2522{
483fbe2b 2523 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
df60cf57
2524 verify_index(istate))
2525 write_locked_index(istate, lockfile, COMMIT_LOCK);
b74c90fb
2526 else
2527 rollback_lock_file(lockfile);
ccdc4ec3
JH
2528}
2529
812d6b00
2530/*
2531 * On success, `tempfile` is closed. If it is the temporary file
2532 * of a `struct lock_file`, we will therefore effectively perform
2533 * a 'close_lock_file_gently()`. Since that is an implementation
2534 * detail of lockfiles, callers of `do_write_index()` should not
2535 * rely on it.
2536 */
9f41c7a6 2537static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
c18b80a0 2538 int strip_extensions)
197ee8c9 2539{
ca54d9ba 2540 uint64_t start = getnanotime();
9f41c7a6 2541 int newfd = tempfile->fd;
3b1d9e04 2542 git_hash_ctx c, eoie_c;
197ee8c9 2543 struct cache_header hdr;
b50386c7 2544 int i, err = 0, removed, extended, hdr_version;
4aab5b46
JH
2545 struct cache_entry **cache = istate->cache;
2546 int entries = istate->cache_nr;
e1afca4f 2547 struct stat st;
ce012deb 2548 struct ondisk_cache_entry_extended ondisk;
9d227781 2549 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
4bddd983 2550 int drop_cache_tree = istate->drop_cache_tree;
3b1d9e04 2551 off_t offset;
3255089a
BP
2552 int ieot_blocks = 1;
2553 struct index_entry_offset_table *ieot = NULL;
2554 int nr, nr_threads;
025a0709 2555
06aaaa0b 2556 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 2557 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 2558 removed++;
197ee8c9 2559
06aaaa0b
NTND
2560 /* reduce extended entries if possible */
2561 cache[i]->ce_flags &= ~CE_EXTENDED;
2562 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2563 extended++;
2564 cache[i]->ce_flags |= CE_EXTENDED;
2565 }
2566 }
2567
d6e3c181 2568 if (!istate->version) {
136347d7 2569 istate->version = get_index_format_default();
4c2db938 2570 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
d6e3c181
NTND
2571 init_split_index(istate);
2572 }
9d227781
JH
2573
2574 /* demote version 3 to version 2 when the latter suffices */
2575 if (istate->version == 3 || istate->version == 2)
2576 istate->version = extended ? 3 : 2;
2577
2578 hdr_version = istate->version;
2579
ccc4feb5 2580 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
9d227781 2581 hdr.hdr_version = htonl(hdr_version);
025a0709 2582 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 2583
aab61359 2584 the_hash_algo->init_fn(&c);
ca9be054 2585 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
2586 return -1;
2587
3255089a
BP
2588#ifndef NO_PTHREADS
2589 nr_threads = git_config_get_index_threads();
2590 if (nr_threads != 1) {
2591 int ieot_blocks, cpus;
2592
2593 /*
2594 * ensure default number of ieot blocks maps evenly to the
2595 * default number of threads that will process them leaving
2596 * room for the thread to load the index extensions.
2597 */
2598 if (!nr_threads) {
2599 ieot_blocks = istate->cache_nr / THREAD_COST;
2600 cpus = online_cpus();
2601 if (ieot_blocks > cpus - 1)
2602 ieot_blocks = cpus - 1;
2603 } else {
2604 ieot_blocks = nr_threads;
2605 }
2606
2607 /*
2608 * no reason to write out the IEOT extension if we don't
2609 * have enough blocks to utilize multi-threading
2610 */
2611 if (ieot_blocks > 1) {
2612 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2613 + (ieot_blocks * sizeof(struct index_entry_offset)));
2614 ieot_blocks = DIV_ROUND_UP(entries, ieot_blocks);
2615 }
2616 }
2617#endif
2618
3b1d9e04 2619 offset = lseek(newfd, 0, SEEK_CUR);
3255089a
BP
2620 if (offset < 0) {
2621 free(ieot);
3b1d9e04 2622 return -1;
3255089a 2623 }
3b1d9e04 2624 offset += write_buffer_len;
3255089a 2625 nr = 0;
9d227781 2626 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
ce012deb 2627
197ee8c9
LT
2628 for (i = 0; i < entries; i++) {
2629 struct cache_entry *ce = cache[i];
7a51ed66 2630 if (ce->ce_flags & CE_REMOVE)
aa16021e 2631 continue;
e06c43c7 2632 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 2633 ce_smudge_racily_clean_entry(ce);
99d1a986 2634 if (is_null_oid(&ce->oid)) {
83bd7437
JK
2635 static const char msg[] = "cache entry has null sha1: %s";
2636 static int allow = -1;
2637
2638 if (allow < 0)
2639 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2640 if (allow)
2641 warning(msg, ce->name);
2642 else
b50386c7 2643 err = error(msg, ce->name);
a96d3cc3
JK
2644
2645 drop_cache_tree = 1;
83bd7437 2646 }
3255089a
BP
2647 if (ieot && i && (i % ieot_blocks == 0)) {
2648 ieot->entries[ieot->nr].nr = nr;
2649 ieot->entries[ieot->nr].offset = offset;
2650 ieot->nr++;
2651 /*
2652 * If we have a V4 index, set the first byte to an invalid
2653 * character to ensure there is nothing common with the previous
2654 * entry
2655 */
2656 if (previous_name)
2657 previous_name->buf[0] = 0;
2658 nr = 0;
2659 offset = lseek(newfd, 0, SEEK_CUR);
2660 if (offset < 0) {
2661 free(ieot);
2662 return -1;
2663 }
2664 offset += write_buffer_len;
2665 }
ce012deb 2666 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
b50386c7
KW
2667 err = -1;
2668
2669 if (err)
2670 break;
3255089a
BP
2671 nr++;
2672 }
2673 if (ieot && nr) {
2674 ieot->entries[ieot->nr].nr = nr;
2675 ieot->entries[ieot->nr].offset = offset;
2676 ieot->nr++;
197ee8c9 2677 }
9d227781 2678 strbuf_release(&previous_name_buf);
1af1c2b6 2679
3255089a
BP
2680 if (err) {
2681 free(ieot);
b50386c7 2682 return err;
3255089a 2683 }
b50386c7 2684
bad68ec9 2685 /* Write extension data here */
3b1d9e04 2686 offset = lseek(newfd, 0, SEEK_CUR);
3255089a
BP
2687 if (offset < 0) {
2688 free(ieot);
3b1d9e04 2689 return -1;
3255089a 2690 }
3b1d9e04
BP
2691 offset += write_buffer_len;
2692 the_hash_algo->init_fn(&eoie_c);
2693
3255089a
BP
2694 /*
2695 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2696 * can minimize the number of extensions we have to scan through to
2697 * find it during load. Write it out regardless of the
2698 * strip_extensions parameter as we need it when loading the shared
2699 * index.
2700 */
2701#ifndef NO_PTHREADS
2702 if (ieot) {
2703 struct strbuf sb = STRBUF_INIT;
2704
2705 write_ieot_extension(&sb, ieot);
2706 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2707 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2708 strbuf_release(&sb);
2709 free(ieot);
2710 if (err)
2711 return -1;
2712 }
2713#endif
2714
c18b80a0 2715 if (!strip_extensions && istate->split_index) {
5fc2fc8f
NTND
2716 struct strbuf sb = STRBUF_INIT;
2717
2718 err = write_link_extension(&sb, istate) < 0 ||
3b1d9e04 2719 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
5fc2fc8f
NTND
2720 sb.len) < 0 ||
2721 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2722 strbuf_release(&sb);
2723 if (err)
2724 return -1;
2725 }
a96d3cc3 2726 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
f285a2d7 2727 struct strbuf sb = STRBUF_INIT;
1dffb8fa 2728
1dffb8fa 2729 cache_tree_write(&sb, istate->cache_tree);
3b1d9e04 2730 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
1dffb8fa
PH
2731 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2732 strbuf_release(&sb);
2733 if (err)
bad68ec9 2734 return -1;
bad68ec9 2735 }
c18b80a0 2736 if (!strip_extensions && istate->resolve_undo) {
cfc5789a
JH
2737 struct strbuf sb = STRBUF_INIT;
2738
2739 resolve_undo_write(&sb, istate->resolve_undo);
3b1d9e04 2740 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
cfc5789a
JH
2741 sb.len) < 0
2742 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2743 strbuf_release(&sb);
2744 if (err)
2745 return -1;
2746 }
83c094ad
NTND
2747 if (!strip_extensions && istate->untracked) {
2748 struct strbuf sb = STRBUF_INIT;
2749
2750 write_untracked_extension(&sb, istate->untracked);
3b1d9e04 2751 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
83c094ad
NTND
2752 sb.len) < 0 ||
2753 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2754 strbuf_release(&sb);
2755 if (err)
2756 return -1;
2757 }
883e248b
BP
2758 if (!strip_extensions && istate->fsmonitor_last_update) {
2759 struct strbuf sb = STRBUF_INIT;
2760
2761 write_fsmonitor_extension(&sb, istate);
3b1d9e04
BP
2762 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2763 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2764 strbuf_release(&sb);
2765 if (err)
2766 return -1;
2767 }
2768
2769 /*
2770 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2771 * so that it can be found and processed before all the index entries are
2772 * read. Write it out regardless of the strip_extensions parameter as we need it
2773 * when loading the shared index.
2774 */
2775 if (offset) {
2776 struct strbuf sb = STRBUF_INIT;
2777
2778 write_eoie_extension(&sb, &eoie_c, offset);
2779 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
883e248b
BP
2780 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2781 strbuf_release(&sb);
2782 if (err)
2783 return -1;
2784 }
e1afca4f 2785
75691ea3 2786 if (ce_flush(&c, newfd, istate->oid.hash))
9f41c7a6 2787 return -1;
49bd0fc2
JK
2788 if (close_tempfile_gently(tempfile)) {
2789 error(_("could not close '%s'"), tempfile->filename.buf);
49bd0fc2
JK
2790 return -1;
2791 }
9f41c7a6 2792 if (stat(tempfile->filename.buf, &st))
e1afca4f 2793 return -1;
5bcf109c
KB
2794 istate->timestamp.sec = (unsigned int)st.st_mtime;
2795 istate->timestamp.nsec = ST_MTIME_NSEC(st);
ca54d9ba 2796 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
e1afca4f 2797 return 0;
197ee8c9 2798}
e46bbcf6 2799
626f35c8
NTND
2800void set_alternate_index_output(const char *name)
2801{
2802 alternate_index_output = name;
2803}
2804
2805static int commit_locked_index(struct lock_file *lk)
2806{
751baced
MH
2807 if (alternate_index_output)
2808 return commit_lock_file_to(lk, alternate_index_output);
2809 else
626f35c8 2810 return commit_lock_file(lk);
626f35c8
NTND
2811}
2812
03b86647
NTND
2813static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2814 unsigned flags)
2815{
076aa2cb 2816 int ret = do_write_index(istate, lock->tempfile, 0);
03b86647
NTND
2817 if (ret)
2818 return ret;
03b86647
NTND
2819 if (flags & COMMIT_LOCK)
2820 return commit_locked_index(lock);
812d6b00 2821 return close_lock_file_gently(lock);
03b86647
NTND
2822}
2823
5fc2fc8f
NTND
2824static int write_split_index(struct index_state *istate,
2825 struct lock_file *lock,
2826 unsigned flags)
2827{
2828 int ret;
2829 prepare_to_write_split_index(istate);
2830 ret = do_write_locked_index(istate, lock, flags);
2831 finish_writing_split_index(istate);
2832 return ret;
2833}
2834
b9683722
CC
2835static const char *shared_index_expire = "2.weeks.ago";
2836
2837static unsigned long get_shared_index_expire_date(void)
2838{
2839 static unsigned long shared_index_expire_date;
2840 static int shared_index_expire_date_prepared;
2841
2842 if (!shared_index_expire_date_prepared) {
2843 git_config_get_expiry("splitindex.sharedindexexpire",
2844 &shared_index_expire);
2845 shared_index_expire_date = approxidate(shared_index_expire);
2846 shared_index_expire_date_prepared = 1;
2847 }
2848
2849 return shared_index_expire_date;
2850}
2851
2852static int should_delete_shared_index(const char *shared_index_path)
2853{
2854 struct stat st;
2855 unsigned long expiration;
2856
2857 /* Check timestamp */
2858 expiration = get_shared_index_expire_date();
2859 if (!expiration)
2860 return 0;
2861 if (stat(shared_index_path, &st))
78bde923 2862 return error_errno(_("could not stat '%s'"), shared_index_path);
b9683722
CC
2863 if (st.st_mtime > expiration)
2864 return 0;
2865
2866 return 1;
2867}
2868
2869static int clean_shared_index_files(const char *current_hex)
2870{
2871 struct dirent *de;
2872 DIR *dir = opendir(get_git_dir());
2873
2874 if (!dir)
2875 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2876
2877 while ((de = readdir(dir)) != NULL) {
2878 const char *sha1_hex;
2879 const char *shared_index_path;
2880 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2881 continue;
2882 if (!strcmp(sha1_hex, current_hex))
2883 continue;
2884 shared_index_path = git_path("%s", de->d_name);
2885 if (should_delete_shared_index(shared_index_path) > 0 &&
2886 unlink(shared_index_path))
2887 warning_errno(_("unable to unlink: %s"), shared_index_path);
2888 }
2889 closedir(dir);
2890
2891 return 0;
2892}
2893
a0a96756 2894static int write_shared_index(struct index_state *istate,
59f9d2dd 2895 struct tempfile **temp)
c18b80a0
NTND
2896{
2897 struct split_index *si = istate->split_index;
076aa2cb 2898 int ret;
c18b80a0 2899
c18b80a0 2900 move_cache_to_base_index(istate);
7db2d08c 2901 ret = do_write_index(si->base, *temp, 1);
59f9d2dd 2902 if (ret)
c18b80a0 2903 return ret;
7db2d08c 2904 ret = adjust_shared_perm(get_tempfile_path(*temp));
df801f3f 2905 if (ret) {
7db2d08c 2906 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
df801f3f
CC
2907 return ret;
2908 }
7db2d08c 2909 ret = rename_tempfile(temp,
75691ea3 2910 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
b9683722 2911 if (!ret) {
75691ea3 2912 oidcpy(&si->base_oid, &si->base->oid);
2913 clean_shared_index_files(oid_to_hex(&si->base->oid));
b9683722
CC
2914 }
2915
c18b80a0
NTND
2916 return ret;
2917}
2918
e6a1dd77
CC
2919static const int default_max_percent_split_change = 20;
2920
2921static int too_many_not_shared_entries(struct index_state *istate)
2922{
2923 int i, not_shared = 0;
2924 int max_split = git_config_get_max_percent_split_change();
2925
2926 switch (max_split) {
2927 case -1:
2928 /* not or badly configured: use the default value */
2929 max_split = default_max_percent_split_change;
2930 break;
2931 case 0:
2932 return 1; /* 0% means always write a new shared index */
2933 case 100:
2934 return 0; /* 100% means never write a new shared index */
2935 default:
2936 break; /* just use the configured value */
2937 }
2938
2939 /* Count not shared entries */
2940 for (i = 0; i < istate->cache_nr; i++) {
2941 struct cache_entry *ce = istate->cache[i];
2942 if (!ce->index)
2943 not_shared++;
2944 }
2945
2946 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2947}
2948
03b86647
NTND
2949int write_locked_index(struct index_state *istate, struct lock_file *lock,
2950 unsigned flags)
2951{
0d59ffb4 2952 int new_shared_index, ret;
5fc2fc8f
NTND
2953 struct split_index *si = istate->split_index;
2954
4592e608
NTND
2955 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
2956 cache_tree_verify(istate);
2957
61000814
2958 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
2959 if (flags & COMMIT_LOCK)
2960 rollback_lock_file(lock);
2961 return 0;
2962 }
2963
3bd28eb2
AV
2964 if (istate->fsmonitor_last_update)
2965 fill_fsmonitor_bitmap(istate);
2966
5165dd59
NTND
2967 if (!si || alternate_index_output ||
2968 (istate->cache_changed & ~EXTMASK)) {
5fc2fc8f 2969 if (si)
2182abd9 2970 oidclr(&si->base_oid);
df60cf57
2971 ret = do_write_locked_index(istate, lock, flags);
2972 goto out;
5fc2fc8f
NTND
2973 }
2974
4c2db938 2975 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2182abd9 2976 int v = si->base_oid.hash[0];
d6e3c181
NTND
2977 if ((v & 15) < 6)
2978 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2979 }
e6a1dd77
CC
2980 if (too_many_not_shared_entries(istate))
2981 istate->cache_changed |= SPLIT_INDEX_ORDERED;
0d59ffb4
CC
2982
2983 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2984
2985 if (new_shared_index) {
59f9d2dd
NTND
2986 struct tempfile *temp;
2987 int saved_errno;
2988
2989 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2990 if (!temp) {
2182abd9 2991 oidclr(&si->base_oid);
59f9d2dd 2992 ret = do_write_locked_index(istate, lock, flags);
ef5b3a6c
NTND
2993 goto out;
2994 }
2995 ret = write_shared_index(istate, &temp);
59f9d2dd
NTND
2996
2997 saved_errno = errno;
2998 if (is_tempfile_active(temp))
2999 delete_tempfile(&temp);
3000 errno = saved_errno;
3001
c18b80a0 3002 if (ret)
df60cf57 3003 goto out;
c18b80a0
NTND
3004 }
3005
0d59ffb4
CC
3006 ret = write_split_index(istate, lock, flags);
3007
3008 /* Freshen the shared index only if the split-index was written */
a125a223
TG
3009 if (!ret && !new_shared_index) {
3010 const char *shared_index = git_path("sharedindex.%s",
2182abd9 3011 oid_to_hex(&si->base_oid));
a125a223
TG
3012 freshen_shared_index(shared_index, 1);
3013 }
0d59ffb4 3014
df60cf57
3015out:
3016 if (flags & COMMIT_LOCK)
3017 rollback_lock_file(lock);
0d59ffb4 3018 return ret;
03b86647
NTND
3019}
3020
e46bbcf6
MV
3021/*
3022 * Read the index file that is potentially unmerged into given
ad376204
EN
3023 * index_state, dropping any unmerged entries to stage #0 (potentially
3024 * resulting in a path appearing as both a file and a directory in the
3025 * index; the caller is responsible to clear out the extra entries
3026 * before writing the index to a tree). Returns true if the index is
3027 * unmerged. Callers who want to refuse to work from an unmerged
3028 * state can call this and check its return value, instead of calling
3029 * read_cache().
e46bbcf6
MV
3030 */
3031int read_index_unmerged(struct index_state *istate)
3032{
3033 int i;
d1a43f2a 3034 int unmerged = 0;
e46bbcf6
MV
3035
3036 read_index(istate);
e46bbcf6
MV
3037 for (i = 0; i < istate->cache_nr; i++) {
3038 struct cache_entry *ce = istate->cache[i];
d1a43f2a 3039 struct cache_entry *new_ce;
a849735b 3040 int len;
d1a43f2a
JH
3041
3042 if (!ce_stage(ce))
e46bbcf6 3043 continue;
d1a43f2a 3044 unmerged = 1;
68c4f6a5 3045 len = ce_namelen(ce);
a849735b 3046 new_ce = make_empty_cache_entry(istate, len);
d1a43f2a 3047 memcpy(new_ce->name, ce->name, len);
b60e188c
TG
3048 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3049 new_ce->ce_namelen = len;
d1a43f2a 3050 new_ce->ce_mode = ce->ce_mode;
ad376204 3051 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
d1a43f2a 3052 return error("%s: cannot drop to stage #0",
5699d17e 3053 new_ce->name);
e46bbcf6 3054 }
d1a43f2a 3055 return unmerged;
e46bbcf6 3056}
041aee31 3057
98fa4738
JK
3058/*
3059 * Returns 1 if the path is an "other" path with respect to
3060 * the index; that is, the path is not mentioned in the index at all,
3061 * either as a file, a directory with some files in the index,
3062 * or as an unmerged entry.
3063 *
3064 * We helpfully remove a trailing "/" from directories so that
3065 * the output of read_directory can be used as-is.
3066 */
3067int index_name_is_other(const struct index_state *istate, const char *name,
3068 int namelen)
3069{
3070 int pos;
3071 if (namelen && name[namelen - 1] == '/')
3072 namelen--;
3073 pos = index_name_pos(istate, name, namelen);
3074 if (0 <= pos)
3075 return 0; /* exact match */
3076 pos = -pos - 1;
3077 if (pos < istate->cache_nr) {
3078 struct cache_entry *ce = istate->cache[pos];
3079 if (ce_namelen(ce) == namelen &&
3080 !memcmp(ce->name, name, namelen))
3081 return 0; /* Yup, this one exists unmerged */
3082 }
3083 return 1;
3084}
29fb37b2 3085
87542508
BW
3086void *read_blob_data_from_index(const struct index_state *istate,
3087 const char *path, unsigned long *size)
29fb37b2
LF
3088{
3089 int pos, len;
3090 unsigned long sz;
3091 enum object_type type;
3092 void *data;
3093
3094 len = strlen(path);
3095 pos = index_name_pos(istate, path, len);
3096 if (pos < 0) {
3097 /*
3098 * We might be in the middle of a merge, in which
3099 * case we would read stage #2 (ours).
3100 */
3101 int i;
3102 for (i = -pos - 1;
3103 (pos < 0 && i < istate->cache_nr &&
3104 !strcmp(istate->cache[i]->name, path));
3105 i++)
3106 if (ce_stage(istate->cache[i]) == 2)
3107 pos = i;
3108 }
3109 if (pos < 0)
3110 return NULL;
b4f5aca4 3111 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
29fb37b2
LF
3112 if (!data || type != OBJ_BLOB) {
3113 free(data);
3114 return NULL;
3115 }
ff366825
LF
3116 if (size)
3117 *size = sz;
29fb37b2
LF
3118 return data;
3119}
38612532
MH
3120
3121void stat_validity_clear(struct stat_validity *sv)
3122{
6a83d902 3123 FREE_AND_NULL(sv->sd);
38612532
MH
3124}
3125
3126int stat_validity_check(struct stat_validity *sv, const char *path)
3127{
3128 struct stat st;
3129
3130 if (stat(path, &st) < 0)
3131 return sv->sd == NULL;
3132 if (!sv->sd)
3133 return 0;
3134 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3135}
3136
3137void stat_validity_update(struct stat_validity *sv, int fd)
3138{
3139 struct stat st;
3140
3141 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3142 stat_validity_clear(sv);
3143 else {
3144 if (!sv->sd)
3145 sv->sd = xcalloc(1, sizeof(struct stat_data));
3146 fill_stat_data(sv->sd, &st);
3147 }
3148}
edf3b905
DT
3149
3150void move_index_extensions(struct index_state *dst, struct index_state *src)
3151{
3152 dst->untracked = src->untracked;
3153 src->untracked = NULL;
836ef2b6
NTND
3154 dst->cache_tree = src->cache_tree;
3155 src->cache_tree = NULL;
edf3b905 3156}
a849735b 3157
8e72d675
JM
3158struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3159 struct index_state *istate)
3160{
3161 unsigned int size = ce_size(ce);
3162 int mem_pool_allocated;
3163 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3164 mem_pool_allocated = new_entry->mem_pool_allocated;
3165
3166 memcpy(new_entry, ce, size);
3167 new_entry->mem_pool_allocated = mem_pool_allocated;
3168 return new_entry;
3169}
3170
a849735b
JM
3171void discard_cache_entry(struct cache_entry *ce)
3172{
8616a2d0
JM
3173 if (ce && should_validate_cache_entries())
3174 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3175
8e72d675
JM
3176 if (ce && ce->mem_pool_allocated)
3177 return;
3178
a849735b
JM
3179 free(ce);
3180}
8616a2d0
JM
3181
3182int should_validate_cache_entries(void)
3183{
3184 static int validate_index_cache_entries = -1;
3185
3186 if (validate_index_cache_entries < 0) {
3187 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3188 validate_index_cache_entries = 1;
3189 else
3190 validate_index_cache_entries = 0;
3191 }
3192
3193 return validate_index_cache_entries;
3194}
3b1d9e04
BP
3195
3196#define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3197#define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3198
3199static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3200{
3201 /*
3202 * The end of index entries (EOIE) extension is guaranteed to be last
3203 * so that it can be found by scanning backwards from the EOF.
3204 *
3205 * "EOIE"
3206 * <4-byte length>
3207 * <4-byte offset>
3208 * <20-byte hash>
3209 */
3210 const char *index, *eoie;
3211 uint32_t extsize;
3212 size_t offset, src_offset;
3213 unsigned char hash[GIT_MAX_RAWSZ];
3214 git_hash_ctx c;
3215
3216 /* ensure we have an index big enough to contain an EOIE extension */
3217 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3218 return 0;
3219
3220 /* validate the extension signature */
3221 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3222 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3223 return 0;
3224 index += sizeof(uint32_t);
3225
3226 /* validate the extension size */
3227 extsize = get_be32(index);
3228 if (extsize != EOIE_SIZE)
3229 return 0;
3230 index += sizeof(uint32_t);
3231
3232 /*
3233 * Validate the offset we're going to look for the first extension
3234 * signature is after the index header and before the eoie extension.
3235 */
3236 offset = get_be32(index);
3237 if (mmap + offset < mmap + sizeof(struct cache_header))
3238 return 0;
3239 if (mmap + offset >= eoie)
3240 return 0;
3241 index += sizeof(uint32_t);
3242
3243 /*
3244 * The hash is computed over extension types and their sizes (but not
3245 * their contents). E.g. if we have "TREE" extension that is N-bytes
3246 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3247 * then the hash would be:
3248 *
3249 * SHA-1("TREE" + <binary representation of N> +
3250 * "REUC" + <binary representation of M>)
3251 */
3252 src_offset = offset;
3253 the_hash_algo->init_fn(&c);
3254 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3255 /* After an array of active_nr index entries,
3256 * there can be arbitrary number of extended
3257 * sections, each of which is prefixed with
3258 * extension name (4-byte) and section length
3259 * in 4-byte network byte order.
3260 */
3261 uint32_t extsize;
3262 memcpy(&extsize, mmap + src_offset + 4, 4);
3263 extsize = ntohl(extsize);
3264
3265 /* verify the extension size isn't so large it will wrap around */
3266 if (src_offset + 8 + extsize < src_offset)
3267 return 0;
3268
3269 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3270
3271 src_offset += 8;
3272 src_offset += extsize;
3273 }
3274 the_hash_algo->final_fn(hash, &c);
3275 if (!hasheq(hash, (const unsigned char *)index))
3276 return 0;
3277
3278 /* Validate that the extension offsets returned us back to the eoie extension. */
3279 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3280 return 0;
3281
3282 return offset;
3283}
3284
3285static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3286{
3287 uint32_t buffer;
3288 unsigned char hash[GIT_MAX_RAWSZ];
3289
3290 /* offset */
3291 put_be32(&buffer, offset);
3292 strbuf_add(sb, &buffer, sizeof(uint32_t));
3293
3294 /* hash */
3295 the_hash_algo->final_fn(hash, eoie_context);
3296 strbuf_add(sb, hash, the_hash_algo->rawsz);
3297}
3255089a
BP
3298
3299#ifndef NO_PTHREADS
3300#define IEOT_VERSION (1)
3301
3302static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3303{
3304 const char *index = NULL;
3305 uint32_t extsize, ext_version;
3306 struct index_entry_offset_table *ieot;
3307 int i, nr;
3308
3309 /* find the IEOT extension */
3310 if (!offset)
3311 return NULL;
3312 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3313 extsize = get_be32(mmap + offset + 4);
3314 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3315 index = mmap + offset + 4 + 4;
3316 break;
3317 }
3318 offset += 8;
3319 offset += extsize;
3320 }
3321 if (!index)
3322 return NULL;
3323
3324 /* validate the version is IEOT_VERSION */
3325 ext_version = get_be32(index);
3326 if (ext_version != IEOT_VERSION) {
3327 error("invalid IEOT version %d", ext_version);
3328 return NULL;
3329 }
3330 index += sizeof(uint32_t);
3331
3332 /* extension size - version bytes / bytes per entry */
3333 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3334 if (!nr) {
3335 error("invalid number of IEOT entries %d", nr);
3336 return NULL;
3337 }
3338 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3339 + (nr * sizeof(struct index_entry_offset)));
3340 ieot->nr = nr;
3341 for (i = 0; i < nr; i++) {
3342 ieot->entries[i].offset = get_be32(index);
3343 index += sizeof(uint32_t);
3344 ieot->entries[i].nr = get_be32(index);
3345 index += sizeof(uint32_t);
3346 }
3347
3348 return ieot;
3349}
3350
3351static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3352{
3353 uint32_t buffer;
3354 int i;
3355
3356 /* version */
3357 put_be32(&buffer, IEOT_VERSION);
3358 strbuf_add(sb, &buffer, sizeof(uint32_t));
3359
3360 /* ieot */
3361 for (i = 0; i < ieot->nr; i++) {
3362
3363 /* offset */
3364 put_be32(&buffer, ieot->entries[i].offset);
3365 strbuf_add(sb, &buffer, sizeof(uint32_t));
3366
3367 /* count */
3368 put_be32(&buffer, ieot->entries[i].nr);
3369 strbuf_add(sb, &buffer, sizeof(uint32_t));
3370 }
3371}
3372#endif