lstat_cache(): print a warning if doing ping-pong between cache types
[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"
bad68ec9 8#include "cache-tree.h"
f35a6d3b 9#include "refs.h"
d616813d 10#include "dir.h"
041aee31
JH
11#include "tree.h"
12#include "commit.h"
13#include "diff.h"
14#include "diffcore.h"
15#include "revision.h"
39425819 16#include "blob.h"
bad68ec9
JH
17
18/* Index extensions.
19 *
20 * The first letter should be 'A'..'Z' for extensions that are not
21 * necessary for a correct operation (i.e. optimization data).
22 * When new extensions are added that _needs_ to be understood in
23 * order to correctly interpret the index file, pick character that
24 * is outside the range, to cause the reader to abort.
25 */
26
27#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
28#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
e83c5163 29
228e94f9 30struct index_state the_index;
8fd2cb40 31
9cb76b8c
JH
32static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
33{
34 istate->cache[nr] = ce;
96872bc2 35 add_name_hash(istate, ce);
9cb76b8c
JH
36}
37
cf558704
LT
38static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
39{
40 struct cache_entry *old = istate->cache[nr];
41
96872bc2 42 remove_name_hash(old);
a22c6371 43 set_index_entry(istate, nr, ce);
cf558704
LT
44 istate->cache_changed = 1;
45}
46
81dc2307
PB
47void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
48{
49 struct cache_entry *old = istate->cache[nr], *new;
50 int namelen = strlen(new_name);
51
52 new = xmalloc(cache_entry_size(namelen));
53 copy_cache_entry(new, old);
54 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
55 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
56 memcpy(new->name, new_name, namelen + 1);
57
58 cache_tree_invalidate_path(istate->cache_tree, old->name);
59 remove_index_entry_at(istate, nr);
60 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
61}
62
415e96c8
JH
63/*
64 * This only updates the "non-critical" parts of the directory
65 * cache, ie the parts that aren't tracked by GIT, and only used
66 * to validate the cache.
67 */
68void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
69{
7a51ed66
LT
70 ce->ce_ctime = st->st_ctime;
71 ce->ce_mtime = st->st_mtime;
72 ce->ce_dev = st->st_dev;
73 ce->ce_ino = st->st_ino;
74 ce->ce_uid = st->st_uid;
75 ce->ce_gid = st->st_gid;
76 ce->ce_size = st->st_size;
5f73076c
JH
77
78 if (assume_unchanged)
7a51ed66 79 ce->ce_flags |= CE_VALID;
eadb5831
JH
80
81 if (S_ISREG(st->st_mode))
82 ce_mark_uptodate(ce);
415e96c8
JH
83}
84
29e4d363
JH
85static int ce_compare_data(struct cache_entry *ce, struct stat *st)
86{
87 int match = -1;
88 int fd = open(ce->name, O_RDONLY);
89
90 if (fd >= 0) {
91 unsigned char sha1[20];
53bca91a 92 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
a89fccd2 93 match = hashcmp(sha1, ce->sha1);
7f8508e8 94 /* index_fd() closed the file descriptor already */
29e4d363
JH
95 }
96 return match;
97}
98
dc49cd76 99static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
29e4d363
JH
100{
101 int match = -1;
29e4d363
JH
102 void *buffer;
103 unsigned long size;
21666f1a 104 enum object_type type;
a60272b3 105 struct strbuf sb = STRBUF_INIT;
29e4d363 106
a60272b3 107 if (strbuf_readlink(&sb, ce->name, expected_size))
29e4d363 108 return -1;
a60272b3 109
21666f1a 110 buffer = read_sha1_file(ce->sha1, &type, &size);
a60272b3
LT
111 if (buffer) {
112 if (size == sb.len)
113 match = memcmp(buffer, sb.buf, size);
114 free(buffer);
29e4d363 115 }
a60272b3 116 strbuf_release(&sb);
29e4d363
JH
117 return match;
118}
119
f35a6d3b
LT
120static int ce_compare_gitlink(struct cache_entry *ce)
121{
122 unsigned char sha1[20];
123
124 /*
125 * We don't actually require that the .git directory
302b9282 126 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
127 * might even be missing (in case nobody populated that
128 * sub-project).
129 *
130 * If so, we consider it always to match.
131 */
132 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
133 return 0;
134 return hashcmp(sha1, ce->sha1);
135}
136
29e4d363
JH
137static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
138{
139 switch (st->st_mode & S_IFMT) {
140 case S_IFREG:
141 if (ce_compare_data(ce, st))
142 return DATA_CHANGED;
143 break;
144 case S_IFLNK:
dc49cd76 145 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
146 return DATA_CHANGED;
147 break;
a8ee75bc 148 case S_IFDIR:
7a51ed66 149 if (S_ISGITLINK(ce->ce_mode))
c70115b4 150 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
29e4d363
JH
151 default:
152 return TYPE_CHANGED;
153 }
154 return 0;
155}
156
f55527f8 157int is_empty_blob_sha1(const unsigned char *sha1)
f49c2c22
LT
158{
159 static const unsigned char empty_blob_sha1[20] = {
160 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
161 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
162 };
163
164 return !hashcmp(sha1, empty_blob_sha1);
165}
166
407c8eb0 167static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
734aab75
LT
168{
169 unsigned int changed = 0;
170
7a51ed66
LT
171 if (ce->ce_flags & CE_REMOVE)
172 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
173
174 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
175 case S_IFREG:
176 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
177 /* We consider only the owner x bit to be relevant for
178 * "mode changes"
179 */
180 if (trust_executable_bit &&
7a51ed66 181 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 182 changed |= MODE_CHANGED;
8ae0a8c5
KS
183 break;
184 case S_IFLNK:
78a8d641
JS
185 if (!S_ISLNK(st->st_mode) &&
186 (has_symlinks || !S_ISREG(st->st_mode)))
187 changed |= TYPE_CHANGED;
8ae0a8c5 188 break;
302b9282 189 case S_IFGITLINK:
c70115b4 190 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
191 if (!S_ISDIR(st->st_mode))
192 changed |= TYPE_CHANGED;
193 else if (ce_compare_gitlink(ce))
194 changed |= DATA_CHANGED;
a8ee75bc 195 return changed;
8ae0a8c5 196 default:
7a51ed66 197 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 198 }
7a51ed66 199 if (ce->ce_mtime != (unsigned int) st->st_mtime)
ccc4feb5 200 changed |= MTIME_CHANGED;
1ce4790b 201 if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime)
734aab75 202 changed |= CTIME_CHANGED;
ccc4feb5 203
7a51ed66
LT
204 if (ce->ce_uid != (unsigned int) st->st_uid ||
205 ce->ce_gid != (unsigned int) st->st_gid)
734aab75 206 changed |= OWNER_CHANGED;
7a51ed66 207 if (ce->ce_ino != (unsigned int) st->st_ino)
734aab75 208 changed |= INODE_CHANGED;
2cb45e95
LT
209
210#ifdef USE_STDEV
211 /*
212 * st_dev breaks on network filesystems where different
213 * clients will have different views of what "device"
214 * the filesystem is on
215 */
7a51ed66 216 if (ce->ce_dev != (unsigned int) st->st_dev)
2cb45e95
LT
217 changed |= INODE_CHANGED;
218#endif
219
7a51ed66 220 if (ce->ce_size != (unsigned int) st->st_size)
734aab75 221 changed |= DATA_CHANGED;
b0391890 222
f49c2c22
LT
223 /* Racily smudged entry? */
224 if (!ce->ce_size) {
225 if (!is_empty_blob_sha1(ce->sha1))
226 changed |= DATA_CHANGED;
227 }
228
407c8eb0
JH
229 return changed;
230}
231
d1f128b0 232static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
6d91da6d 233{
050288d5
JH
234 return (!S_ISGITLINK(ce->ce_mode) &&
235 istate->timestamp &&
6d91da6d
JH
236 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
237}
238
d1f128b0 239int ie_match_stat(const struct index_state *istate,
4bd5b7da
JH
240 struct cache_entry *ce, struct stat *st,
241 unsigned int options)
407c8eb0 242{
5f73076c 243 unsigned int changed;
4bd5b7da
JH
244 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
245 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
5f73076c
JH
246
247 /*
248 * If it's marked as always valid in the index, it's
249 * valid whatever the checked-out copy says.
250 */
7a51ed66 251 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c
JH
252 return 0;
253
331fcb59
JH
254 /*
255 * Intent-to-add entries have not been added, so the index entry
256 * by definition never matches what is in the work tree until it
257 * actually gets added.
258 */
259 if (ce->ce_flags & CE_INTENT_TO_ADD)
260 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
261
5f73076c 262 changed = ce_match_stat_basic(ce, st);
407c8eb0 263
29e4d363
JH
264 /*
265 * Within 1 second of this sequence:
266 * echo xyzzy >file && git-update-index --add file
267 * running this command:
268 * echo frotz >file
269 * would give a falsely clean cache entry. The mtime and
270 * length match the cache, and other stat fields do not change.
271 *
272 * We could detect this at update-index time (the cache entry
273 * being registered/updated records the same time as "now")
274 * and delay the return from git-update-index, but that would
275 * effectively mean we can make at most one commit per second,
276 * which is not acceptable. Instead, we check cache entries
277 * whose mtime are the same as the index file timestamp more
5f73076c 278 * carefully than others.
29e4d363 279 */
6d91da6d 280 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
281 if (assume_racy_is_modified)
282 changed |= DATA_CHANGED;
283 else
284 changed |= ce_modified_check_fs(ce, st);
285 }
b0391890 286
29e4d363 287 return changed;
b0391890
JH
288}
289
d1f128b0 290int ie_modified(const struct index_state *istate,
4bd5b7da 291 struct cache_entry *ce, struct stat *st, unsigned int options)
b0391890 292{
29e4d363 293 int changed, changed_fs;
4bd5b7da
JH
294
295 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
296 if (!changed)
297 return 0;
b0391890
JH
298 /*
299 * If the mode or type has changed, there's no point in trying
300 * to refresh the entry - it's not going to match
301 */
302 if (changed & (MODE_CHANGED | TYPE_CHANGED))
303 return changed;
304
c70115b4
JH
305 /*
306 * Immediately after read-tree or update-index --cacheinfo,
307 * the length field is zero, as we have never even read the
308 * lstat(2) information once, and we cannot trust DATA_CHANGED
309 * returned by ie_match_stat() which in turn was returned by
310 * ce_match_stat_basic() to signal that the filesize of the
311 * blob changed. We have to actually go to the filesystem to
312 * see if the contents match, and if so, should answer "unchanged".
313 *
314 * The logic does not apply to gitlinks, as ce_match_stat_basic()
315 * already has checked the actual HEAD from the filesystem in the
316 * subproject. If ie_match_stat() already said it is different,
317 * then we know it is.
b0391890 318 */
c70115b4
JH
319 if ((changed & DATA_CHANGED) &&
320 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
b0391890
JH
321 return changed;
322
29e4d363
JH
323 changed_fs = ce_modified_check_fs(ce, st);
324 if (changed_fs)
325 return changed | changed_fs;
b0391890
JH
326 return 0;
327}
328
958ba6c9
LT
329int base_name_compare(const char *name1, int len1, int mode1,
330 const char *name2, int len2, int mode2)
331{
332 unsigned char c1, c2;
333 int len = len1 < len2 ? len1 : len2;
334 int cmp;
335
336 cmp = memcmp(name1, name2, len);
337 if (cmp)
338 return cmp;
339 c1 = name1[len];
340 c2 = name2[len];
1833a925 341 if (!c1 && S_ISDIR(mode1))
958ba6c9 342 c1 = '/';
1833a925 343 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
344 c2 = '/';
345 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
346}
347
0ab9e1e8
LT
348/*
349 * df_name_compare() is identical to base_name_compare(), except it
350 * compares conflicting directory/file entries as equal. Note that
351 * while a directory name compares as equal to a regular file, they
352 * then individually compare _differently_ to a filename that has
353 * a dot after the basename (because '\0' < '.' < '/').
354 *
355 * This is used by routines that want to traverse the git namespace
356 * but then handle conflicting entries together when possible.
357 */
358int df_name_compare(const char *name1, int len1, int mode1,
359 const char *name2, int len2, int mode2)
360{
361 int len = len1 < len2 ? len1 : len2, cmp;
362 unsigned char c1, c2;
363
364 cmp = memcmp(name1, name2, len);
365 if (cmp)
366 return cmp;
367 /* Directories and files compare equal (same length, same name) */
368 if (len1 == len2)
369 return 0;
370 c1 = name1[len];
371 if (!c1 && S_ISDIR(mode1))
372 c1 = '/';
373 c2 = name2[len];
374 if (!c2 && S_ISDIR(mode2))
375 c2 = '/';
376 if (c1 == '/' && !c2)
377 return 0;
378 if (c2 == '/' && !c1)
379 return 0;
380 return c1 - c2;
381}
382
95fd5bf8 383int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
eb38c22f 384{
95fd5bf8
LT
385 int len1 = flags1 & CE_NAMEMASK;
386 int len2 = flags2 & CE_NAMEMASK;
eb38c22f
LT
387 int len = len1 < len2 ? len1 : len2;
388 int cmp;
389
390 cmp = memcmp(name1, name2, len);
391 if (cmp)
392 return cmp;
393 if (len1 < len2)
394 return -1;
395 if (len1 > len2)
396 return 1;
5f73076c 397
7b80be15
JH
398 /* Compare stages */
399 flags1 &= CE_STAGEMASK;
400 flags2 &= CE_STAGEMASK;
5f73076c 401
95fd5bf8
LT
402 if (flags1 < flags2)
403 return -1;
404 if (flags1 > flags2)
405 return 1;
eb38c22f
LT
406 return 0;
407}
408
d1f128b0 409int index_name_pos(const struct index_state *istate, const char *name, int namelen)
eb38c22f
LT
410{
411 int first, last;
412
413 first = 0;
4aab5b46 414 last = istate->cache_nr;
eb38c22f
LT
415 while (last > first) {
416 int next = (last + first) >> 1;
4aab5b46 417 struct cache_entry *ce = istate->cache[next];
7a51ed66 418 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
eb38c22f 419 if (!cmp)
76e7f4ec 420 return next;
eb38c22f
LT
421 if (cmp < 0) {
422 last = next;
423 continue;
424 }
425 first = next+1;
426 }
76e7f4ec 427 return -first-1;
eb38c22f
LT
428}
429
7b937ca3 430/* Remove entry, return true if there are more entries to go.. */
4aab5b46 431int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 432{
cf558704
LT
433 struct cache_entry *ce = istate->cache[pos];
434
96872bc2 435 remove_name_hash(ce);
4aab5b46
JH
436 istate->cache_changed = 1;
437 istate->cache_nr--;
438 if (pos >= istate->cache_nr)
7b937ca3 439 return 0;
4aab5b46
JH
440 memmove(istate->cache + pos,
441 istate->cache + pos + 1,
442 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
7b937ca3
LT
443 return 1;
444}
445
4aab5b46 446int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 447{
4aab5b46 448 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
449 if (pos < 0)
450 pos = -pos-1;
09d5dc32 451 cache_tree_invalidate_path(istate->cache_tree, path);
4aab5b46
JH
452 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
453 remove_index_entry_at(istate, pos);
197ee8c9
LT
454 return 0;
455}
456
20314271
JS
457static int compare_name(struct cache_entry *ce, const char *path, int namelen)
458{
459 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
460}
461
462static int index_name_pos_also_unmerged(struct index_state *istate,
463 const char *path, int namelen)
464{
465 int pos = index_name_pos(istate, path, namelen);
466 struct cache_entry *ce;
467
468 if (pos >= 0)
469 return pos;
470
471 /* maybe unmerged? */
472 pos = -1 - pos;
473 if (pos >= istate->cache_nr ||
474 compare_name((ce = istate->cache[pos]), path, namelen))
475 return -1;
476
477 /* order of preference: stage 2, 1, 3 */
478 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
479 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
480 !compare_name(ce, path, namelen))
481 pos++;
482 return pos;
483}
484
1102952b
LT
485static int different_name(struct cache_entry *ce, struct cache_entry *alias)
486{
487 int len = ce_namelen(ce);
488 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
489}
490
491/*
492 * If we add a filename that aliases in the cache, we will use the
493 * name that we already have - but we don't want to update the same
494 * alias twice, because that implies that there were actually two
495 * different files with aliasing names!
496 *
497 * So we use the CE_ADDED flag to verify that the alias was an old
498 * one before we accept it as
499 */
500static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
501{
502 int len;
503 struct cache_entry *new;
504
505 if (alias->ce_flags & CE_ADDED)
506 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
507
508 /* Ok, create the new entry using the name of the existing alias */
509 len = ce_namelen(alias);
510 new = xcalloc(1, cache_entry_size(len));
511 memcpy(new->name, alias->name, len);
512 copy_cache_entry(new, ce);
513 free(ce);
514 return new;
515}
516
39425819
JH
517static void record_intent_to_add(struct cache_entry *ce)
518{
519 unsigned char sha1[20];
520 if (write_sha1_file("", 0, blob_type, sha1))
521 die("cannot create an empty blob in the object database");
522 hashcpy(ce->sha1, sha1);
523}
524
38ed1d89 525int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 526{
38ed1d89 527 int size, namelen, was_same;
d177cab0 528 mode_t st_mode = st->st_mode;
6835550d 529 struct cache_entry *ce, *alias;
fb63d7f8 530 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
531 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
532 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
533 int intent_only = flags & ADD_CACHE_INTENT;
534 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
535 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
11be42a4 536
d177cab0 537 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
960b8ad1 538 return error("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
539
540 namelen = strlen(path);
d177cab0 541 if (S_ISDIR(st_mode)) {
09595258
LT
542 while (namelen && path[namelen-1] == '/')
543 namelen--;
544 }
11be42a4
JS
545 size = cache_entry_size(namelen);
546 ce = xcalloc(1, size);
547 memcpy(ce->name, path, namelen);
7a51ed66 548 ce->ce_flags = namelen;
39425819
JH
549 if (!intent_only)
550 fill_stat_cache_info(ce, st);
388b2acd
JH
551 else
552 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 553
78a8d641 554 if (trust_executable_bit && has_symlinks)
d177cab0 555 ce->ce_mode = create_ce_mode(st_mode);
185c975f 556 else {
78a8d641
JS
557 /* If there is an existing entry, pick the mode bits and type
558 * from it, otherwise assume unexecutable regular file.
11be42a4 559 */
185c975f 560 struct cache_entry *ent;
20314271 561 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 562
4aab5b46 563 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 564 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
565 }
566
6835550d 567 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
d177cab0 568 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
0781b8a9
JH
569 /* Nothing changed, really */
570 free(ce);
6835550d 571 ce_mark_uptodate(alias);
1102952b 572 alias->ce_flags |= CE_ADDED;
0781b8a9
JH
573 return 0;
574 }
39425819
JH
575 if (!intent_only) {
576 if (index_path(ce->sha1, path, st, 1))
577 return error("unable to index file %s", path);
578 } else
579 record_intent_to_add(ce);
580
1102952b
LT
581 if (ignore_case && alias && different_name(ce, alias))
582 ce = create_alias_ce(ce, alias);
583 ce->ce_flags |= CE_ADDED;
38ed1d89 584
3bf0dd1f 585 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
586 was_same = (alias &&
587 !ce_stage(alias) &&
588 !hashcmp(alias->sha1, ce->sha1) &&
589 ce->ce_mode == alias->ce_mode);
590
591 if (pretend)
592 ;
39425819 593 else if (add_index_entry(istate, ce, add_option))
960b8ad1 594 return error("unable to add %s to index",path);
38ed1d89 595 if (verbose && !was_same)
11be42a4 596 printf("add '%s'\n", path);
11be42a4
JS
597 return 0;
598}
599
38ed1d89 600int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
601{
602 struct stat st;
603 if (lstat(path, &st))
604 die("%s: unable to stat (%s)", path, strerror(errno));
38ed1d89 605 return add_to_index(istate, path, &st, flags);
d177cab0
LT
606}
607
6640f881
CR
608struct cache_entry *make_cache_entry(unsigned int mode,
609 const unsigned char *sha1, const char *path, int stage,
610 int refresh)
611{
612 int size, len;
613 struct cache_entry *ce;
614
7e7abea9
DP
615 if (!verify_path(path)) {
616 error("Invalid path '%s'", path);
6640f881 617 return NULL;
7e7abea9 618 }
6640f881
CR
619
620 len = strlen(path);
621 size = cache_entry_size(len);
622 ce = xcalloc(1, size);
623
624 hashcpy(ce->sha1, sha1);
625 memcpy(ce->name, path, len);
626 ce->ce_flags = create_ce_flags(len, stage);
627 ce->ce_mode = create_ce_mode(mode);
628
629 if (refresh)
630 return refresh_cache_entry(ce, 0);
631
632 return ce;
633}
634
dbbce55b 635int ce_same_name(struct cache_entry *a, struct cache_entry *b)
7b937ca3
LT
636{
637 int len = ce_namelen(a);
638 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
639}
640
c0fd1f51
LT
641int ce_path_match(const struct cache_entry *ce, const char **pathspec)
642{
643 const char *match, *name;
644 int len;
645
646 if (!pathspec)
647 return 1;
648
649 len = ce_namelen(ce);
650 name = ce->name;
651 while ((match = *pathspec++) != NULL) {
652 int matchlen = strlen(match);
653 if (matchlen > len)
654 continue;
655 if (memcmp(name, match, matchlen))
656 continue;
657 if (matchlen && name[matchlen-1] == '/')
658 return 1;
659 if (name[matchlen] == '/' || !name[matchlen])
660 return 1;
f332726e
LT
661 if (!matchlen)
662 return 1;
c0fd1f51
LT
663 }
664 return 0;
665}
666
8dcf39c4
LT
667/*
668 * We fundamentally don't like some paths: we don't want
669 * dot or dot-dot anywhere, and for obvious reasons don't
670 * want to recurse into ".git" either.
671 *
672 * Also, we don't want double slashes or slashes at the
673 * end that can make pathnames ambiguous.
674 */
675static int verify_dotfile(const char *rest)
676{
677 /*
678 * The first character was '.', but that
679 * has already been discarded, we now test
680 * the rest.
681 */
682 switch (*rest) {
683 /* "." is not allowed */
684 case '\0': case '/':
685 return 0;
686
687 /*
688 * ".git" followed by NUL or slash is bad. This
689 * shares the path end test with the ".." case.
690 */
691 case 'g':
692 if (rest[1] != 'i')
693 break;
694 if (rest[2] != 't')
695 break;
696 rest += 2;
697 /* fallthrough */
698 case '.':
699 if (rest[1] == '\0' || rest[1] == '/')
700 return 0;
701 }
702 return 1;
703}
704
705int verify_path(const char *path)
706{
707 char c;
708
709 goto inside;
710 for (;;) {
711 if (!c)
712 return 1;
713 if (c == '/') {
714inside:
715 c = *path++;
716 switch (c) {
717 default:
718 continue;
719 case '/': case '\0':
720 break;
721 case '.':
722 if (verify_dotfile(path))
723 continue;
724 }
725 return 0;
726 }
727 c = *path++;
728 }
729}
730
12676608
LT
731/*
732 * Do we have another file that has the beginning components being a
733 * proper superset of the name we're trying to add?
0f1e4f04 734 */
4aab5b46
JH
735static int has_file_name(struct index_state *istate,
736 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 737{
12676608
LT
738 int retval = 0;
739 int len = ce_namelen(ce);
b155725d 740 int stage = ce_stage(ce);
12676608 741 const char *name = ce->name;
0f1e4f04 742
4aab5b46
JH
743 while (pos < istate->cache_nr) {
744 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 745
12676608 746 if (len >= ce_namelen(p))
0f1e4f04 747 break;
12676608
LT
748 if (memcmp(name, p->name, len))
749 break;
b155725d
JH
750 if (ce_stage(p) != stage)
751 continue;
12676608
LT
752 if (p->name[len] != '/')
753 continue;
7a51ed66 754 if (p->ce_flags & CE_REMOVE)
21cd8d00 755 continue;
12676608
LT
756 retval = -1;
757 if (!ok_to_replace)
758 break;
4aab5b46 759 remove_index_entry_at(istate, --pos);
0f1e4f04 760 }
12676608
LT
761 return retval;
762}
0f1e4f04 763
12676608
LT
764/*
765 * Do we have another file with a pathname that is a proper
766 * subset of the name we're trying to add?
767 */
4aab5b46
JH
768static int has_dir_name(struct index_state *istate,
769 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
770{
771 int retval = 0;
b155725d 772 int stage = ce_stage(ce);
12676608
LT
773 const char *name = ce->name;
774 const char *slash = name + ce_namelen(ce);
0f1e4f04 775
12676608
LT
776 for (;;) {
777 int len;
0f1e4f04 778
12676608
LT
779 for (;;) {
780 if (*--slash == '/')
781 break;
782 if (slash <= ce->name)
783 return retval;
784 }
785 len = slash - name;
0f1e4f04 786
7a51ed66 787 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
12676608 788 if (pos >= 0) {
21cd8d00
JH
789 /*
790 * Found one, but not so fast. This could
791 * be a marker that says "I was here, but
792 * I am being removed". Such an entry is
793 * not a part of the resulting tree, and
794 * it is Ok to have a directory at the same
795 * path.
796 */
077c48df 797 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
798 retval = -1;
799 if (!ok_to_replace)
800 break;
4aab5b46 801 remove_index_entry_at(istate, pos);
21cd8d00
JH
802 continue;
803 }
12676608 804 }
21cd8d00
JH
805 else
806 pos = -pos-1;
12676608
LT
807
808 /*
809 * Trivial optimization: if we find an entry that
810 * already matches the sub-directory, then we know
b155725d 811 * we're ok, and we can exit.
12676608 812 */
4aab5b46
JH
813 while (pos < istate->cache_nr) {
814 struct cache_entry *p = istate->cache[pos];
b155725d
JH
815 if ((ce_namelen(p) <= len) ||
816 (p->name[len] != '/') ||
817 memcmp(p->name, name, len))
818 break; /* not our subdirectory */
077c48df
JH
819 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
820 /*
821 * p is at the same stage as our entry, and
b155725d
JH
822 * is a subdirectory of what we are looking
823 * at, so we cannot have conflicts at our
824 * level or anything shorter.
825 */
826 return retval;
827 pos++;
192268c1 828 }
0f1e4f04 829 }
12676608
LT
830 return retval;
831}
832
833/* We may be in a situation where we already have path/file and path
834 * is being added, or we already have path and path/file is being
835 * added. Either one would result in a nonsense tree that has path
836 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 837 *
12676608
LT
838 * If ok-to-replace is specified, we remove the conflicting entries
839 * from the cache so the caller should recompute the insert position.
840 * When this happens, we return non-zero.
841 */
4aab5b46
JH
842static int check_file_directory_conflict(struct index_state *istate,
843 const struct cache_entry *ce,
844 int pos, int ok_to_replace)
12676608 845{
21cd8d00
JH
846 int retval;
847
848 /*
849 * When ce is an "I am going away" entry, we allow it to be added
850 */
7a51ed66 851 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
852 return 0;
853
12676608
LT
854 /*
855 * We check if the path is a sub-path of a subsequent pathname
856 * first, since removing those will not change the position
21cd8d00 857 * in the array.
12676608 858 */
4aab5b46 859 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 860
12676608
LT
861 /*
862 * Then check if the path might have a clashing sub-directory
863 * before it.
864 */
4aab5b46 865 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
866}
867
af3785dc 868static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
869{
870 int pos;
192268c1
JH
871 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
872 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 873 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 874 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 875
09d5dc32 876 cache_tree_invalidate_path(istate->cache_tree, ce->name);
7a51ed66 877 pos = index_name_pos(istate, ce->name, ce->ce_flags);
197ee8c9 878
3e09cdfd 879 /* existing match? Just replace it. */
76e7f4ec 880 if (pos >= 0) {
39425819
JH
881 if (!new_only)
882 replace_index_entry(istate, pos, ce);
197ee8c9
LT
883 return 0;
884 }
76e7f4ec 885 pos = -pos-1;
197ee8c9 886
7b937ca3
LT
887 /*
888 * Inserting a merged entry ("stage 0") into the index
889 * will always replace all non-merged entries..
890 */
4aab5b46
JH
891 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
892 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 893 ok_to_add = 1;
4aab5b46 894 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
895 break;
896 }
897 }
898
121481ab
LT
899 if (!ok_to_add)
900 return -1;
8dcf39c4 901 if (!verify_path(ce->name))
7e7abea9 902 return error("Invalid path '%s'", ce->name);
121481ab 903
3e09cdfd 904 if (!skip_df_check &&
4aab5b46 905 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 906 if (!ok_to_replace)
4aab5b46
JH
907 return error("'%s' appears as both a file and as a directory",
908 ce->name);
7a51ed66 909 pos = index_name_pos(istate, ce->name, ce->ce_flags);
192268c1
JH
910 pos = -pos-1;
911 }
af3785dc
JH
912 return pos + 1;
913}
914
915int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
916{
917 int pos;
918
919 if (option & ADD_CACHE_JUST_APPEND)
920 pos = istate->cache_nr;
921 else {
922 int ret;
923 ret = add_index_entry_with_check(istate, ce, option);
924 if (ret <= 0)
925 return ret;
926 pos = ret - 1;
927 }
0f1e4f04 928
197ee8c9 929 /* Make sure the array is big enough .. */
4aab5b46
JH
930 if (istate->cache_nr == istate->cache_alloc) {
931 istate->cache_alloc = alloc_nr(istate->cache_alloc);
932 istate->cache = xrealloc(istate->cache,
933 istate->cache_alloc * sizeof(struct cache_entry *));
197ee8c9
LT
934 }
935
936 /* Add it in.. */
4aab5b46 937 istate->cache_nr++;
af3785dc 938 if (istate->cache_nr > pos + 1)
4aab5b46
JH
939 memmove(istate->cache + pos + 1,
940 istate->cache + pos,
941 (istate->cache_nr - pos - 1) * sizeof(ce));
cf558704 942 set_index_entry(istate, pos, ce);
4aab5b46 943 istate->cache_changed = 1;
197ee8c9
LT
944 return 0;
945}
946
405e5b2f
LT
947/*
948 * "refresh" does not calculate a new sha1 file or bring the
949 * cache up-to-date for mode/content changes. But what it
950 * _does_ do is to "re-match" the stat information of a file
951 * with the cache, so that you can refresh the cache for a
952 * file that hasn't been changed but where the stat entry is
953 * out of date.
954 *
955 * For example, you'd want to do this after doing a "git-read-tree",
956 * to link up the stat cache details with the proper files.
957 */
4aab5b46 958static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da
JH
959 struct cache_entry *ce,
960 unsigned int options, int *err)
405e5b2f
LT
961{
962 struct stat st;
963 struct cache_entry *updated;
964 int changed, size;
4bd5b7da 965 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
405e5b2f 966
eadb5831
JH
967 if (ce_uptodate(ce))
968 return ce;
969
aa9349d4
MSO
970 /*
971 * CE_VALID means the user promised us that the change to
972 * the work tree does not matter and told us not to worry.
973 */
974 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
975 ce_mark_uptodate(ce);
976 return ce;
977 }
978
8fd2cb40 979 if (lstat(ce->name, &st) < 0) {
ec0cc704
JH
980 if (err)
981 *err = errno;
8fd2cb40
JS
982 return NULL;
983 }
405e5b2f 984
4bd5b7da 985 changed = ie_match_stat(istate, ce, &st, options);
405e5b2f 986 if (!changed) {
4bd5b7da
JH
987 /*
988 * The path is unchanged. If we were told to ignore
989 * valid bit, then we did the actual stat check and
990 * found that the entry is unmodified. If the entry
991 * is not marked VALID, this is the place to mark it
992 * valid again, under "assume unchanged" mode.
993 */
994 if (ignore_valid && assume_unchanged &&
7a51ed66 995 !(ce->ce_flags & CE_VALID))
405e5b2f 996 ; /* mark this one VALID again */
eadb5831
JH
997 else {
998 /*
999 * We do not mark the index itself "modified"
1000 * because CE_UPTODATE flag is in-core only;
1001 * we are not going to write this change out.
1002 */
1003 ce_mark_uptodate(ce);
8fd2cb40 1004 return ce;
eadb5831 1005 }
405e5b2f
LT
1006 }
1007
4bd5b7da 1008 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1009 if (err)
1010 *err = EINVAL;
8fd2cb40
JS
1011 return NULL;
1012 }
405e5b2f
LT
1013
1014 size = ce_size(ce);
1015 updated = xmalloc(size);
1016 memcpy(updated, ce, size);
1017 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1018 /*
1019 * If ignore_valid is not set, we should leave CE_VALID bit
1020 * alone. Otherwise, paths marked with --no-assume-unchanged
1021 * (i.e. things to be edited) will reacquire CE_VALID bit
1022 * automatically, which is not really what we want.
405e5b2f 1023 */
4bd5b7da 1024 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1025 !(ce->ce_flags & CE_VALID))
1026 updated->ce_flags &= ~CE_VALID;
405e5b2f
LT
1027
1028 return updated;
1029}
1030
d616813d 1031int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
405e5b2f
LT
1032{
1033 int i;
1034 int has_errors = 0;
1035 int really = (flags & REFRESH_REALLY) != 0;
1036 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1037 int quiet = (flags & REFRESH_QUIET) != 0;
1038 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1039 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
4bd5b7da 1040 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
d14e7407 1041 const char *needs_update_message;
405e5b2f 1042
d14e7407
JH
1043 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
1044 ? "locally modified" : "needs update");
4aab5b46 1045 for (i = 0; i < istate->cache_nr; i++) {
405e5b2f 1046 struct cache_entry *ce, *new;
ec0cc704
JH
1047 int cache_errno = 0;
1048
4aab5b46 1049 ce = istate->cache[i];
5fdeacb0
JS
1050 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1051 continue;
1052
405e5b2f 1053 if (ce_stage(ce)) {
4aab5b46
JH
1054 while ((i < istate->cache_nr) &&
1055 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1056 i++;
1057 i--;
1058 if (allow_unmerged)
1059 continue;
1060 printf("%s: needs merge\n", ce->name);
1061 has_errors = 1;
1062 continue;
1063 }
1064
d616813d
AJ
1065 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1066 continue;
1067
4bd5b7da 1068 new = refresh_cache_ent(istate, ce, options, &cache_errno);
8fd2cb40 1069 if (new == ce)
405e5b2f 1070 continue;
8fd2cb40
JS
1071 if (!new) {
1072 if (not_new && cache_errno == ENOENT)
405e5b2f 1073 continue;
8fd2cb40 1074 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1075 /* If we are doing --really-refresh that
1076 * means the index is not valid anymore.
1077 */
7a51ed66 1078 ce->ce_flags &= ~CE_VALID;
4aab5b46 1079 istate->cache_changed = 1;
405e5b2f
LT
1080 }
1081 if (quiet)
1082 continue;
d14e7407 1083 printf("%s: %s\n", ce->name, needs_update_message);
405e5b2f
LT
1084 has_errors = 1;
1085 continue;
1086 }
cf558704
LT
1087
1088 replace_index_entry(istate, i, new);
405e5b2f
LT
1089 }
1090 return has_errors;
1091}
1092
ec0cc704
JH
1093struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1094{
4aab5b46 1095 return refresh_cache_ent(&the_index, ce, really, NULL);
ec0cc704
JH
1096}
1097
bad68ec9 1098static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163 1099{
9126f009 1100 git_SHA_CTX c;
bad68ec9 1101 unsigned char sha1[20];
e83c5163 1102
ccc4feb5 1103 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1104 return error("bad signature");
06aaaa0b 1105 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
ca9be054 1106 return error("bad index version");
9126f009
NP
1107 git_SHA1_Init(&c);
1108 git_SHA1_Update(&c, hdr, size - 20);
1109 git_SHA1_Final(sha1, &c);
a89fccd2 1110 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
ca9be054 1111 return error("bad index file sha1 signature");
e83c5163
LT
1112 return 0;
1113}
1114
4aab5b46
JH
1115static int read_index_extension(struct index_state *istate,
1116 const char *ext, void *data, unsigned long sz)
bad68ec9
JH
1117{
1118 switch (CACHE_EXT(ext)) {
1119 case CACHE_EXT_TREE:
4aab5b46 1120 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9
JH
1121 break;
1122 default:
1123 if (*ext < 'A' || 'Z' < *ext)
1124 return error("index uses %.4s extension, which we do not understand",
1125 ext);
1126 fprintf(stderr, "ignoring %.4s extension\n", ext);
1127 break;
1128 }
1129 return 0;
1130}
1131
4aab5b46 1132int read_index(struct index_state *istate)
8fd2cb40 1133{
4aab5b46 1134 return read_index_from(istate, get_index_file());
8fd2cb40
JS
1135}
1136
7a51ed66
LT
1137static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1138{
7fec10b7 1139 size_t len;
06aaaa0b 1140 const char *name;
7fec10b7 1141
7a51ed66
LT
1142 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1143 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1144 ce->ce_dev = ntohl(ondisk->dev);
1145 ce->ce_ino = ntohl(ondisk->ino);
1146 ce->ce_mode = ntohl(ondisk->mode);
1147 ce->ce_uid = ntohl(ondisk->uid);
1148 ce->ce_gid = ntohl(ondisk->gid);
1149 ce->ce_size = ntohl(ondisk->size);
1150 /* On-disk flags are just 16 bits */
1151 ce->ce_flags = ntohs(ondisk->flags);
16ce2e4c 1152
7a51ed66 1153 hashcpy(ce->sha1, ondisk->sha1);
7fec10b7
JH
1154
1155 len = ce->ce_flags & CE_NAMEMASK;
06aaaa0b
NTND
1156
1157 if (ce->ce_flags & CE_EXTENDED) {
1158 struct ondisk_cache_entry_extended *ondisk2;
1159 int extended_flags;
1160 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1161 extended_flags = ntohs(ondisk2->flags2) << 16;
1162 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1163 if (extended_flags & ~CE_EXTENDED_FLAGS)
1164 die("Unknown index entry format %08x", extended_flags);
1165 ce->ce_flags |= extended_flags;
1166 name = ondisk2->name;
1167 }
1168 else
1169 name = ondisk->name;
1170
7fec10b7 1171 if (len == CE_NAMEMASK)
06aaaa0b 1172 len = strlen(name);
7fec10b7
JH
1173 /*
1174 * NEEDSWORK: If the original index is crafted, this copy could
1175 * go unchecked.
1176 */
06aaaa0b 1177 memcpy(ce->name, name, len + 1);
7a51ed66
LT
1178}
1179
cf558704
LT
1180static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1181{
1182 long per_entry;
1183
1184 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1185
1186 /*
1187 * Alignment can cause differences. This should be "alignof", but
1188 * since that's a gcc'ism, just use the size of a pointer.
1189 */
1190 per_entry += sizeof(void *);
1191 return ondisk_size + entries*per_entry;
1192}
1193
8fd2cb40 1194/* remember to discard_cache() before reading a different cache! */
4aab5b46 1195int read_index_from(struct index_state *istate, const char *path)
e83c5163
LT
1196{
1197 int fd, i;
1198 struct stat st;
7a51ed66 1199 unsigned long src_offset, dst_offset;
e83c5163 1200 struct cache_header *hdr;
7a51ed66
LT
1201 void *mmap;
1202 size_t mmap_size;
e83c5163
LT
1203
1204 errno = EBUSY;
913e0e99 1205 if (istate->initialized)
4aab5b46 1206 return istate->cache_nr;
5d1a5c02 1207
e83c5163 1208 errno = ENOENT;
4aab5b46 1209 istate->timestamp = 0;
8fd2cb40 1210 fd = open(path, O_RDONLY);
5d1a5c02
LT
1211 if (fd < 0) {
1212 if (errno == ENOENT)
1213 return 0;
1214 die("index file open failed (%s)", strerror(errno));
1215 }
e83c5163 1216
3511a377 1217 if (fstat(fd, &st))
5fe5c830 1218 die("cannot stat the open index (%s)", strerror(errno));
3511a377
LFC
1219
1220 errno = EINVAL;
7a51ed66
LT
1221 mmap_size = xsize_t(st.st_size);
1222 if (mmap_size < sizeof(struct cache_header) + 20)
3511a377
LFC
1223 die("index file smaller than expected");
1224
7a51ed66 1225 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
e83c5163 1226 close(fd);
7a51ed66
LT
1227 if (mmap == MAP_FAILED)
1228 die("unable to map index file");
e83c5163 1229
7a51ed66
LT
1230 hdr = mmap;
1231 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1232 goto unmap;
1233
4aab5b46
JH
1234 istate->cache_nr = ntohl(hdr->hdr_entries);
1235 istate->cache_alloc = alloc_nr(istate->cache_nr);
1236 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
e83c5163 1237
7a51ed66
LT
1238 /*
1239 * The disk format is actually larger than the in-memory format,
1240 * due to space for nsec etc, so even though the in-memory one
1241 * has room for a few more flags, we can allocate using the same
1242 * index size
1243 */
cf558704 1244 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
913e0e99 1245 istate->initialized = 1;
7a51ed66
LT
1246
1247 src_offset = sizeof(*hdr);
1248 dst_offset = 0;
4aab5b46 1249 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 1250 struct ondisk_cache_entry *disk_ce;
4aab5b46
JH
1251 struct cache_entry *ce;
1252
7a51ed66
LT
1253 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1254 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1255 convert_from_disk(disk_ce, ce);
cf558704 1256 set_index_entry(istate, i, ce);
7a51ed66
LT
1257
1258 src_offset += ondisk_ce_size(ce);
1259 dst_offset += ce_size(ce);
e83c5163 1260 }
4aab5b46 1261 istate->timestamp = st.st_mtime;
7a51ed66 1262 while (src_offset <= mmap_size - 20 - 8) {
bad68ec9
JH
1263 /* After an array of active_nr index entries,
1264 * there can be arbitrary number of extended
1265 * sections, each of which is prefixed with
1266 * extension name (4-byte) and section length
1267 * in 4-byte network byte order.
1268 */
1269 unsigned long extsize;
7a51ed66 1270 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
bad68ec9 1271 extsize = ntohl(extsize);
4aab5b46 1272 if (read_index_extension(istate,
7a51ed66
LT
1273 (const char *) mmap + src_offset,
1274 (char *) mmap + src_offset + 8,
1d7f171c 1275 extsize) < 0)
bad68ec9 1276 goto unmap;
7a51ed66
LT
1277 src_offset += 8;
1278 src_offset += extsize;
bad68ec9 1279 }
7a51ed66 1280 munmap(mmap, mmap_size);
4aab5b46 1281 return istate->cache_nr;
e83c5163
LT
1282
1283unmap:
7a51ed66 1284 munmap(mmap, mmap_size);
e83c5163 1285 errno = EINVAL;
5d1a5c02 1286 die("index file corrupt");
e83c5163
LT
1287}
1288
fa7b3c2f
JH
1289int is_index_unborn(struct index_state *istate)
1290{
1291 return (!istate->cache_nr && !istate->alloc && !istate->timestamp);
1292}
1293
4aab5b46 1294int discard_index(struct index_state *istate)
6d297f81 1295{
4aab5b46
JH
1296 istate->cache_nr = 0;
1297 istate->cache_changed = 0;
1298 istate->timestamp = 0;
64ca23af 1299 istate->name_hash_initialized = 0;
cf558704 1300 free_hash(&istate->name_hash);
4aab5b46 1301 cache_tree_free(&(istate->cache_tree));
7a51ed66
LT
1302 free(istate->alloc);
1303 istate->alloc = NULL;
913e0e99 1304 istate->initialized = 0;
6d297f81
JS
1305
1306 /* no need to throw away allocated active_cache */
7a51ed66 1307 return 0;
6d297f81
JS
1308}
1309
d1f128b0 1310int unmerged_index(const struct index_state *istate)
94a5728c
DB
1311{
1312 int i;
1313 for (i = 0; i < istate->cache_nr; i++) {
1314 if (ce_stage(istate->cache[i]))
1315 return 1;
1316 }
1317 return 0;
1318}
1319
4990aadc 1320#define WRITE_BUFFER_SIZE 8192
bf0f910d 1321static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
1322static unsigned long write_buffer_len;
1323
9126f009 1324static int ce_write_flush(git_SHA_CTX *context, int fd)
6015c28b
JH
1325{
1326 unsigned int buffered = write_buffer_len;
1327 if (buffered) {
9126f009 1328 git_SHA1_Update(context, write_buffer, buffered);
93822c22 1329 if (write_in_full(fd, write_buffer, buffered) != buffered)
6015c28b
JH
1330 return -1;
1331 write_buffer_len = 0;
1332 }
1333 return 0;
1334}
1335
9126f009 1336static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
1337{
1338 while (len) {
1339 unsigned int buffered = write_buffer_len;
1340 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1341 if (partial > len)
1342 partial = len;
1343 memcpy(write_buffer + buffered, data, partial);
1344 buffered += partial;
1345 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
1346 write_buffer_len = buffered;
1347 if (ce_write_flush(context, fd))
4990aadc
LT
1348 return -1;
1349 buffered = 0;
1350 }
1351 write_buffer_len = buffered;
1352 len -= partial;
1d7f171c 1353 data = (char *) data + partial;
a6080a0a
JH
1354 }
1355 return 0;
4990aadc
LT
1356}
1357
9126f009 1358static int write_index_ext_header(git_SHA_CTX *context, int fd,
ac58c7b1 1359 unsigned int ext, unsigned int sz)
bad68ec9
JH
1360{
1361 ext = htonl(ext);
1362 sz = htonl(sz);
968a1d65
DR
1363 return ((ce_write(context, fd, &ext, 4) < 0) ||
1364 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
1365}
1366
9126f009 1367static int ce_flush(git_SHA_CTX *context, int fd)
4990aadc
LT
1368{
1369 unsigned int left = write_buffer_len;
ca9be054 1370
4990aadc
LT
1371 if (left) {
1372 write_buffer_len = 0;
9126f009 1373 git_SHA1_Update(context, write_buffer, left);
4990aadc 1374 }
ca9be054 1375
2c865d9a
QH
1376 /* Flush first if not enough space for SHA1 signature */
1377 if (left + 20 > WRITE_BUFFER_SIZE) {
93822c22 1378 if (write_in_full(fd, write_buffer, left) != left)
2c865d9a
QH
1379 return -1;
1380 left = 0;
1381 }
1382
ca9be054 1383 /* Append the SHA1 signature at the end */
9126f009 1384 git_SHA1_Final(write_buffer + left, context);
ca9be054 1385 left += 20;
93822c22 1386 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
4990aadc
LT
1387}
1388
407c8eb0
JH
1389static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1390{
1391 /*
1392 * The only thing we care about in this function is to smudge the
1393 * falsely clean entry due to touch-update-touch race, so we leave
1394 * everything else as they are. We are called for entries whose
1395 * ce_mtime match the index file mtime.
c70115b4
JH
1396 *
1397 * Note that this actually does not do much for gitlinks, for
1398 * which ce_match_stat_basic() always goes to the actual
1399 * contents. The caller checks with is_racy_timestamp() which
1400 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
1401 */
1402 struct stat st;
1403
1404 if (lstat(ce->name, &st) < 0)
1405 return;
1406 if (ce_match_stat_basic(ce, &st))
1407 return;
1408 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
1409 /* This is "racily clean"; smudge it. Note that this
1410 * is a tricky code. At first glance, it may appear
1411 * that it can break with this sequence:
1412 *
1413 * $ echo xyzzy >frotz
1414 * $ git-update-index --add frotz
1415 * $ : >frotz
1416 * $ sleep 3
1417 * $ echo filfre >nitfol
1418 * $ git-update-index --add nitfol
1419 *
b7e58b17 1420 * but it does not. When the second update-index runs,
4b3511b0
JH
1421 * it notices that the entry "frotz" has the same timestamp
1422 * as index, and if we were to smudge it by resetting its
1423 * size to zero here, then the object name recorded
1424 * in index is the 6-byte file but the cached stat information
1425 * becomes zero --- which would then match what we would
a6080a0a 1426 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
1427 *
1428 * However, the second update-index, before calling
1429 * this function, notices that the cached size is 6
1430 * bytes and what is on the filesystem is an empty
1431 * file, and never calls us, so the cached size information
1432 * for "frotz" stays 6 which does not match the filesystem.
1433 */
7a51ed66 1434 ce->ce_size = 0;
407c8eb0
JH
1435 }
1436}
1437
9126f009 1438static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
7a51ed66
LT
1439{
1440 int size = ondisk_ce_size(ce);
1441 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
06aaaa0b 1442 char *name;
7a51ed66
LT
1443
1444 ondisk->ctime.sec = htonl(ce->ce_ctime);
1445 ondisk->ctime.nsec = 0;
1446 ondisk->mtime.sec = htonl(ce->ce_mtime);
1447 ondisk->mtime.nsec = 0;
1448 ondisk->dev = htonl(ce->ce_dev);
1449 ondisk->ino = htonl(ce->ce_ino);
1450 ondisk->mode = htonl(ce->ce_mode);
1451 ondisk->uid = htonl(ce->ce_uid);
1452 ondisk->gid = htonl(ce->ce_gid);
1453 ondisk->size = htonl(ce->ce_size);
1454 hashcpy(ondisk->sha1, ce->sha1);
1455 ondisk->flags = htons(ce->ce_flags);
06aaaa0b
NTND
1456 if (ce->ce_flags & CE_EXTENDED) {
1457 struct ondisk_cache_entry_extended *ondisk2;
1458 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1459 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1460 name = ondisk2->name;
1461 }
1462 else
1463 name = ondisk->name;
1464 memcpy(name, ce->name, ce_namelen(ce));
7a51ed66
LT
1465
1466 return ce_write(c, fd, ondisk, size);
1467}
1468
d1f128b0 1469int write_index(const struct index_state *istate, int newfd)
197ee8c9 1470{
9126f009 1471 git_SHA_CTX c;
197ee8c9 1472 struct cache_header hdr;
06aaaa0b 1473 int i, err, removed, extended;
4aab5b46
JH
1474 struct cache_entry **cache = istate->cache;
1475 int entries = istate->cache_nr;
025a0709 1476
06aaaa0b 1477 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 1478 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 1479 removed++;
197ee8c9 1480
06aaaa0b
NTND
1481 /* reduce extended entries if possible */
1482 cache[i]->ce_flags &= ~CE_EXTENDED;
1483 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1484 extended++;
1485 cache[i]->ce_flags |= CE_EXTENDED;
1486 }
1487 }
1488
ccc4feb5 1489 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
06aaaa0b
NTND
1490 /* for extended format, increase version so older git won't try to read it */
1491 hdr.hdr_version = htonl(extended ? 3 : 2);
025a0709 1492 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 1493
9126f009 1494 git_SHA1_Init(&c);
ca9be054 1495 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
1496 return -1;
1497
1498 for (i = 0; i < entries; i++) {
1499 struct cache_entry *ce = cache[i];
7a51ed66 1500 if (ce->ce_flags & CE_REMOVE)
aa16021e 1501 continue;
e06c43c7 1502 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 1503 ce_smudge_racily_clean_entry(ce);
7a51ed66 1504 if (ce_write_entry(&c, newfd, ce) < 0)
197ee8c9
LT
1505 return -1;
1506 }
1af1c2b6 1507
bad68ec9 1508 /* Write extension data here */
4aab5b46 1509 if (istate->cache_tree) {
f285a2d7 1510 struct strbuf sb = STRBUF_INIT;
1dffb8fa 1511
1dffb8fa
PH
1512 cache_tree_write(&sb, istate->cache_tree);
1513 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1514 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1515 strbuf_release(&sb);
1516 if (err)
bad68ec9 1517 return -1;
bad68ec9
JH
1518 }
1519 return ce_flush(&c, newfd);
197ee8c9 1520}
e46bbcf6
MV
1521
1522/*
1523 * Read the index file that is potentially unmerged into given
63e8dc5b 1524 * index_state, dropping any unmerged entries. Returns true if
e46bbcf6
MV
1525 * the index is unmerged. Callers who want to refuse to work
1526 * from an unmerged state can call this and check its return value,
1527 * instead of calling read_cache().
1528 */
1529int read_index_unmerged(struct index_state *istate)
1530{
1531 int i;
d1a43f2a 1532 int unmerged = 0;
e46bbcf6
MV
1533
1534 read_index(istate);
e46bbcf6
MV
1535 for (i = 0; i < istate->cache_nr; i++) {
1536 struct cache_entry *ce = istate->cache[i];
d1a43f2a
JH
1537 struct cache_entry *new_ce;
1538 int size, len;
1539
1540 if (!ce_stage(ce))
e46bbcf6 1541 continue;
d1a43f2a
JH
1542 unmerged = 1;
1543 len = strlen(ce->name);
1544 size = cache_entry_size(len);
1545 new_ce = xcalloc(1, size);
1546 hashcpy(new_ce->sha1, ce->sha1);
1547 memcpy(new_ce->name, ce->name, len);
1548 new_ce->ce_flags = create_ce_flags(len, 0);
1549 new_ce->ce_mode = ce->ce_mode;
1550 if (add_index_entry(istate, new_ce, 0))
1551 return error("%s: cannot drop to stage #0",
1552 ce->name);
1553 i = index_name_pos(istate, new_ce->name, len);
e46bbcf6 1554 }
d1a43f2a 1555 return unmerged;
e46bbcf6 1556}
041aee31
JH
1557
1558struct update_callback_data
1559{
1560 int flags;
1561 int add_errors;
1562};
1563
1564static void update_callback(struct diff_queue_struct *q,
1565 struct diff_options *opt, void *cbdata)
1566{
1567 int i;
1568 struct update_callback_data *data = cbdata;
1569
1570 for (i = 0; i < q->nr; i++) {
1571 struct diff_filepair *p = q->queue[i];
1572 const char *path = p->one->path;
1573 switch (p->status) {
1574 default:
1575 die("unexpected diff status %c", p->status);
1576 case DIFF_STATUS_UNMERGED:
4cc8d6c6
JH
1577 /*
1578 * ADD_CACHE_IGNORE_REMOVAL is unset if "git
1579 * add -u" is calling us, In such a case, a
1580 * missing work tree file needs to be removed
1581 * if there is an unmerged entry at stage #2,
1582 * but such a diff record is followed by
1583 * another with DIFF_STATUS_DELETED (and if
1584 * there is no stage #2, we won't see DELETED
1585 * nor MODIFIED). We can simply continue
1586 * either way.
1587 */
1588 if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL))
1589 continue;
1590 /*
1591 * Otherwise, it is "git add path" is asking
1592 * to explicitly add it; we fall through. A
1593 * missing work tree file is an error and is
1594 * caught by add_file_to_index() in such a
1595 * case.
1596 */
041aee31
JH
1597 case DIFF_STATUS_MODIFIED:
1598 case DIFF_STATUS_TYPE_CHANGED:
1599 if (add_file_to_index(&the_index, path, data->flags)) {
1600 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1601 die("updating files failed");
1602 data->add_errors++;
1603 }
1604 break;
1605 case DIFF_STATUS_DELETED:
1606 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1607 break;
1608 if (!(data->flags & ADD_CACHE_PRETEND))
1609 remove_file_from_index(&the_index, path);
1610 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1611 printf("remove '%s'\n", path);
1612 break;
1613 }
1614 }
1615}
1616
1617int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1618{
1619 struct update_callback_data data;
1620 struct rev_info rev;
1621 init_revisions(&rev, prefix);
1622 setup_revisions(0, NULL, &rev, NULL);
1623 rev.prune_data = pathspec;
1624 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1625 rev.diffopt.format_callback = update_callback;
1626 data.flags = flags;
1627 data.add_errors = 0;
1628 rev.diffopt.format_callback_data = &data;
1629 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1630 return !!data.add_errors;
1631}
1632
98fa4738
JK
1633/*
1634 * Returns 1 if the path is an "other" path with respect to
1635 * the index; that is, the path is not mentioned in the index at all,
1636 * either as a file, a directory with some files in the index,
1637 * or as an unmerged entry.
1638 *
1639 * We helpfully remove a trailing "/" from directories so that
1640 * the output of read_directory can be used as-is.
1641 */
1642int index_name_is_other(const struct index_state *istate, const char *name,
1643 int namelen)
1644{
1645 int pos;
1646 if (namelen && name[namelen - 1] == '/')
1647 namelen--;
1648 pos = index_name_pos(istate, name, namelen);
1649 if (0 <= pos)
1650 return 0; /* exact match */
1651 pos = -pos - 1;
1652 if (pos < istate->cache_nr) {
1653 struct cache_entry *ce = istate->cache[pos];
1654 if (ce_namelen(ce) == namelen &&
1655 !memcmp(ce->name, name, namelen))
1656 return 0; /* Yup, this one exists unmerged */
1657 }
1658 return 1;
1659}