verify_path(): simplify check at the directory boundary
[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"
39425819 13#include "blob.h"
cfc5789a 14#include "resolve-undo.h"
bad68ec9 15
87b29e5a
JH
16static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
17
bad68ec9
JH
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" */
b659b49b 29#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
e83c5163 30
228e94f9 31struct index_state the_index;
8fd2cb40 32
9cb76b8c
JH
33static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
34{
35 istate->cache[nr] = ce;
96872bc2 36 add_name_hash(istate, ce);
9cb76b8c
JH
37}
38
cf558704
LT
39static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
40{
41 struct cache_entry *old = istate->cache[nr];
42
96872bc2 43 remove_name_hash(old);
a22c6371 44 set_index_entry(istate, nr, ce);
cf558704
LT
45 istate->cache_changed = 1;
46}
47
81dc2307
PB
48void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
49{
50 struct cache_entry *old = istate->cache[nr], *new;
51 int namelen = strlen(new_name);
52
53 new = xmalloc(cache_entry_size(namelen));
54 copy_cache_entry(new, old);
55 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
56 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
57 memcpy(new->name, new_name, namelen + 1);
58
59 cache_tree_invalidate_path(istate->cache_tree, old->name);
60 remove_index_entry_at(istate, nr);
61 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
62}
63
415e96c8
JH
64/*
65 * This only updates the "non-critical" parts of the directory
66 * cache, ie the parts that aren't tracked by GIT, and only used
67 * to validate the cache.
68 */
69void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
70{
fba2f38a
KB
71 ce->ce_ctime.sec = (unsigned int)st->st_ctime;
72 ce->ce_mtime.sec = (unsigned int)st->st_mtime;
c06ff490
KB
73 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
74 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
7a51ed66
LT
75 ce->ce_dev = st->st_dev;
76 ce->ce_ino = st->st_ino;
77 ce->ce_uid = st->st_uid;
78 ce->ce_gid = st->st_gid;
79 ce->ce_size = st->st_size;
5f73076c
JH
80
81 if (assume_unchanged)
7a51ed66 82 ce->ce_flags |= CE_VALID;
eadb5831
JH
83
84 if (S_ISREG(st->st_mode))
85 ce_mark_uptodate(ce);
415e96c8
JH
86}
87
29e4d363
JH
88static int ce_compare_data(struct cache_entry *ce, struct stat *st)
89{
90 int match = -1;
91 int fd = open(ce->name, O_RDONLY);
92
93 if (fd >= 0) {
94 unsigned char sha1[20];
53bca91a 95 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
a89fccd2 96 match = hashcmp(sha1, ce->sha1);
7f8508e8 97 /* index_fd() closed the file descriptor already */
29e4d363
JH
98 }
99 return match;
100}
101
dc49cd76 102static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
29e4d363
JH
103{
104 int match = -1;
29e4d363
JH
105 void *buffer;
106 unsigned long size;
21666f1a 107 enum object_type type;
a60272b3 108 struct strbuf sb = STRBUF_INIT;
29e4d363 109
a60272b3 110 if (strbuf_readlink(&sb, ce->name, expected_size))
29e4d363 111 return -1;
a60272b3 112
21666f1a 113 buffer = read_sha1_file(ce->sha1, &type, &size);
a60272b3
LT
114 if (buffer) {
115 if (size == sb.len)
116 match = memcmp(buffer, sb.buf, size);
117 free(buffer);
29e4d363 118 }
a60272b3 119 strbuf_release(&sb);
29e4d363
JH
120 return match;
121}
122
f35a6d3b
LT
123static int ce_compare_gitlink(struct cache_entry *ce)
124{
125 unsigned char sha1[20];
126
127 /*
128 * We don't actually require that the .git directory
302b9282 129 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
130 * might even be missing (in case nobody populated that
131 * sub-project).
132 *
133 * If so, we consider it always to match.
134 */
135 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
136 return 0;
137 return hashcmp(sha1, ce->sha1);
138}
139
29e4d363
JH
140static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
141{
142 switch (st->st_mode & S_IFMT) {
143 case S_IFREG:
144 if (ce_compare_data(ce, st))
145 return DATA_CHANGED;
146 break;
147 case S_IFLNK:
dc49cd76 148 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
149 return DATA_CHANGED;
150 break;
a8ee75bc 151 case S_IFDIR:
7a51ed66 152 if (S_ISGITLINK(ce->ce_mode))
c70115b4 153 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
29e4d363
JH
154 default:
155 return TYPE_CHANGED;
156 }
157 return 0;
158}
159
87b29e5a 160static int is_empty_blob_sha1(const unsigned char *sha1)
f49c2c22
LT
161{
162 static const unsigned char empty_blob_sha1[20] = {
163 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
164 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
165 };
166
167 return !hashcmp(sha1, empty_blob_sha1);
168}
169
407c8eb0 170static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
734aab75
LT
171{
172 unsigned int changed = 0;
173
7a51ed66
LT
174 if (ce->ce_flags & CE_REMOVE)
175 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
176
177 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
178 case S_IFREG:
179 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
180 /* We consider only the owner x bit to be relevant for
181 * "mode changes"
182 */
183 if (trust_executable_bit &&
7a51ed66 184 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 185 changed |= MODE_CHANGED;
8ae0a8c5
KS
186 break;
187 case S_IFLNK:
78a8d641
JS
188 if (!S_ISLNK(st->st_mode) &&
189 (has_symlinks || !S_ISREG(st->st_mode)))
190 changed |= TYPE_CHANGED;
8ae0a8c5 191 break;
302b9282 192 case S_IFGITLINK:
c70115b4 193 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
194 if (!S_ISDIR(st->st_mode))
195 changed |= TYPE_CHANGED;
196 else if (ce_compare_gitlink(ce))
197 changed |= DATA_CHANGED;
a8ee75bc 198 return changed;
8ae0a8c5 199 default:
7a51ed66 200 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 201 }
fba2f38a 202 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
ccc4feb5 203 changed |= MTIME_CHANGED;
fba2f38a 204 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
734aab75 205 changed |= CTIME_CHANGED;
ccc4feb5 206
fba2f38a 207#ifdef USE_NSEC
110c46a9 208 if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
fba2f38a 209 changed |= MTIME_CHANGED;
110c46a9 210 if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
fba2f38a
KB
211 changed |= CTIME_CHANGED;
212#endif
213
7a51ed66
LT
214 if (ce->ce_uid != (unsigned int) st->st_uid ||
215 ce->ce_gid != (unsigned int) st->st_gid)
734aab75 216 changed |= OWNER_CHANGED;
7a51ed66 217 if (ce->ce_ino != (unsigned int) st->st_ino)
734aab75 218 changed |= INODE_CHANGED;
2cb45e95
LT
219
220#ifdef USE_STDEV
221 /*
222 * st_dev breaks on network filesystems where different
223 * clients will have different views of what "device"
224 * the filesystem is on
225 */
7a51ed66 226 if (ce->ce_dev != (unsigned int) st->st_dev)
2cb45e95
LT
227 changed |= INODE_CHANGED;
228#endif
229
7a51ed66 230 if (ce->ce_size != (unsigned int) st->st_size)
734aab75 231 changed |= DATA_CHANGED;
b0391890 232
f49c2c22
LT
233 /* Racily smudged entry? */
234 if (!ce->ce_size) {
235 if (!is_empty_blob_sha1(ce->sha1))
236 changed |= DATA_CHANGED;
237 }
238
407c8eb0
JH
239 return changed;
240}
241
d1f128b0 242static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
6d91da6d 243{
050288d5 244 return (!S_ISGITLINK(ce->ce_mode) &&
fba2f38a
KB
245 istate->timestamp.sec &&
246#ifdef USE_NSEC
247 /* nanosecond timestamped files can also be racy! */
248 (istate->timestamp.sec < ce->ce_mtime.sec ||
249 (istate->timestamp.sec == ce->ce_mtime.sec &&
250 istate->timestamp.nsec <= ce->ce_mtime.nsec))
251#else
252 istate->timestamp.sec <= ce->ce_mtime.sec
253#endif
254 );
6d91da6d
JH
255}
256
d1f128b0 257int ie_match_stat(const struct index_state *istate,
4bd5b7da
JH
258 struct cache_entry *ce, struct stat *st,
259 unsigned int options)
407c8eb0 260{
5f73076c 261 unsigned int changed;
4bd5b7da 262 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 263 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
4bd5b7da 264 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
5f73076c
JH
265
266 /*
267 * If it's marked as always valid in the index, it's
268 * valid whatever the checked-out copy says.
56cac48c
NTND
269 *
270 * skip-worktree has the same effect with higher precedence
5f73076c 271 */
56cac48c
NTND
272 if (!ignore_skip_worktree && ce_skip_worktree(ce))
273 return 0;
7a51ed66 274 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c
JH
275 return 0;
276
331fcb59
JH
277 /*
278 * Intent-to-add entries have not been added, so the index entry
279 * by definition never matches what is in the work tree until it
280 * actually gets added.
281 */
282 if (ce->ce_flags & CE_INTENT_TO_ADD)
283 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
284
5f73076c 285 changed = ce_match_stat_basic(ce, st);
407c8eb0 286
29e4d363
JH
287 /*
288 * Within 1 second of this sequence:
289 * echo xyzzy >file && git-update-index --add file
290 * running this command:
291 * echo frotz >file
292 * would give a falsely clean cache entry. The mtime and
293 * length match the cache, and other stat fields do not change.
294 *
295 * We could detect this at update-index time (the cache entry
296 * being registered/updated records the same time as "now")
297 * and delay the return from git-update-index, but that would
298 * effectively mean we can make at most one commit per second,
299 * which is not acceptable. Instead, we check cache entries
300 * whose mtime are the same as the index file timestamp more
5f73076c 301 * carefully than others.
29e4d363 302 */
6d91da6d 303 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
304 if (assume_racy_is_modified)
305 changed |= DATA_CHANGED;
306 else
307 changed |= ce_modified_check_fs(ce, st);
308 }
b0391890 309
29e4d363 310 return changed;
b0391890
JH
311}
312
d1f128b0 313int ie_modified(const struct index_state *istate,
4bd5b7da 314 struct cache_entry *ce, struct stat *st, unsigned int options)
b0391890 315{
29e4d363 316 int changed, changed_fs;
4bd5b7da
JH
317
318 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
319 if (!changed)
320 return 0;
b0391890
JH
321 /*
322 * If the mode or type has changed, there's no point in trying
323 * to refresh the entry - it's not going to match
324 */
325 if (changed & (MODE_CHANGED | TYPE_CHANGED))
326 return changed;
327
c70115b4
JH
328 /*
329 * Immediately after read-tree or update-index --cacheinfo,
330 * the length field is zero, as we have never even read the
331 * lstat(2) information once, and we cannot trust DATA_CHANGED
332 * returned by ie_match_stat() which in turn was returned by
333 * ce_match_stat_basic() to signal that the filesize of the
334 * blob changed. We have to actually go to the filesystem to
335 * see if the contents match, and if so, should answer "unchanged".
336 *
337 * The logic does not apply to gitlinks, as ce_match_stat_basic()
338 * already has checked the actual HEAD from the filesystem in the
339 * subproject. If ie_match_stat() already said it is different,
340 * then we know it is.
b0391890 341 */
c70115b4
JH
342 if ((changed & DATA_CHANGED) &&
343 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
b0391890
JH
344 return changed;
345
29e4d363
JH
346 changed_fs = ce_modified_check_fs(ce, st);
347 if (changed_fs)
348 return changed | changed_fs;
b0391890
JH
349 return 0;
350}
351
958ba6c9
LT
352int base_name_compare(const char *name1, int len1, int mode1,
353 const char *name2, int len2, int mode2)
354{
355 unsigned char c1, c2;
356 int len = len1 < len2 ? len1 : len2;
357 int cmp;
358
359 cmp = memcmp(name1, name2, len);
360 if (cmp)
361 return cmp;
362 c1 = name1[len];
363 c2 = name2[len];
1833a925 364 if (!c1 && S_ISDIR(mode1))
958ba6c9 365 c1 = '/';
1833a925 366 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
367 c2 = '/';
368 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
369}
370
0ab9e1e8
LT
371/*
372 * df_name_compare() is identical to base_name_compare(), except it
373 * compares conflicting directory/file entries as equal. Note that
374 * while a directory name compares as equal to a regular file, they
375 * then individually compare _differently_ to a filename that has
376 * a dot after the basename (because '\0' < '.' < '/').
377 *
378 * This is used by routines that want to traverse the git namespace
379 * but then handle conflicting entries together when possible.
380 */
381int df_name_compare(const char *name1, int len1, int mode1,
382 const char *name2, int len2, int mode2)
383{
384 int len = len1 < len2 ? len1 : len2, cmp;
385 unsigned char c1, c2;
386
387 cmp = memcmp(name1, name2, len);
388 if (cmp)
389 return cmp;
390 /* Directories and files compare equal (same length, same name) */
391 if (len1 == len2)
392 return 0;
393 c1 = name1[len];
394 if (!c1 && S_ISDIR(mode1))
395 c1 = '/';
396 c2 = name2[len];
397 if (!c2 && S_ISDIR(mode2))
398 c2 = '/';
399 if (c1 == '/' && !c2)
400 return 0;
401 if (c2 == '/' && !c1)
402 return 0;
403 return c1 - c2;
404}
405
95fd5bf8 406int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
eb38c22f 407{
95fd5bf8
LT
408 int len1 = flags1 & CE_NAMEMASK;
409 int len2 = flags2 & CE_NAMEMASK;
eb38c22f
LT
410 int len = len1 < len2 ? len1 : len2;
411 int cmp;
412
413 cmp = memcmp(name1, name2, len);
414 if (cmp)
415 return cmp;
416 if (len1 < len2)
417 return -1;
418 if (len1 > len2)
419 return 1;
5f73076c 420
7b80be15
JH
421 /* Compare stages */
422 flags1 &= CE_STAGEMASK;
423 flags2 &= CE_STAGEMASK;
5f73076c 424
95fd5bf8
LT
425 if (flags1 < flags2)
426 return -1;
427 if (flags1 > flags2)
428 return 1;
eb38c22f
LT
429 return 0;
430}
431
d1f128b0 432int index_name_pos(const struct index_state *istate, const char *name, int namelen)
eb38c22f
LT
433{
434 int first, last;
435
436 first = 0;
4aab5b46 437 last = istate->cache_nr;
eb38c22f
LT
438 while (last > first) {
439 int next = (last + first) >> 1;
4aab5b46 440 struct cache_entry *ce = istate->cache[next];
7a51ed66 441 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
eb38c22f 442 if (!cmp)
76e7f4ec 443 return next;
eb38c22f
LT
444 if (cmp < 0) {
445 last = next;
446 continue;
447 }
448 first = next+1;
449 }
76e7f4ec 450 return -first-1;
eb38c22f
LT
451}
452
7b937ca3 453/* Remove entry, return true if there are more entries to go.. */
4aab5b46 454int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 455{
cf558704
LT
456 struct cache_entry *ce = istate->cache[pos];
457
cfc5789a 458 record_resolve_undo(istate, ce);
96872bc2 459 remove_name_hash(ce);
4aab5b46
JH
460 istate->cache_changed = 1;
461 istate->cache_nr--;
462 if (pos >= istate->cache_nr)
7b937ca3 463 return 0;
4aab5b46
JH
464 memmove(istate->cache + pos,
465 istate->cache + pos + 1,
466 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
7b937ca3
LT
467 return 1;
468}
469
36419c8e
KB
470/*
471 * Remove all cache ententries marked for removal, that is where
472 * CE_REMOVE is set in ce_flags. This is much more effective than
473 * calling remove_index_entry_at() for each entry to be removed.
474 */
475void remove_marked_cache_entries(struct index_state *istate)
476{
477 struct cache_entry **ce_array = istate->cache;
478 unsigned int i, j;
479
480 for (i = j = 0; i < istate->cache_nr; i++) {
481 if (ce_array[i]->ce_flags & CE_REMOVE)
482 remove_name_hash(ce_array[i]);
483 else
484 ce_array[j++] = ce_array[i];
485 }
486 istate->cache_changed = 1;
487 istate->cache_nr = j;
488}
489
4aab5b46 490int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 491{
4aab5b46 492 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
493 if (pos < 0)
494 pos = -pos-1;
09d5dc32 495 cache_tree_invalidate_path(istate->cache_tree, path);
4aab5b46
JH
496 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
497 remove_index_entry_at(istate, pos);
197ee8c9
LT
498 return 0;
499}
500
20314271
JS
501static int compare_name(struct cache_entry *ce, const char *path, int namelen)
502{
503 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
504}
505
506static int index_name_pos_also_unmerged(struct index_state *istate,
507 const char *path, int namelen)
508{
509 int pos = index_name_pos(istate, path, namelen);
510 struct cache_entry *ce;
511
512 if (pos >= 0)
513 return pos;
514
515 /* maybe unmerged? */
516 pos = -1 - pos;
517 if (pos >= istate->cache_nr ||
518 compare_name((ce = istate->cache[pos]), path, namelen))
519 return -1;
520
521 /* order of preference: stage 2, 1, 3 */
522 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
523 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
524 !compare_name(ce, path, namelen))
525 pos++;
526 return pos;
527}
528
1102952b
LT
529static int different_name(struct cache_entry *ce, struct cache_entry *alias)
530{
531 int len = ce_namelen(ce);
532 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
533}
534
535/*
536 * If we add a filename that aliases in the cache, we will use the
537 * name that we already have - but we don't want to update the same
538 * alias twice, because that implies that there were actually two
539 * different files with aliasing names!
540 *
541 * So we use the CE_ADDED flag to verify that the alias was an old
542 * one before we accept it as
543 */
544static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
545{
546 int len;
547 struct cache_entry *new;
548
549 if (alias->ce_flags & CE_ADDED)
550 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
551
552 /* Ok, create the new entry using the name of the existing alias */
553 len = ce_namelen(alias);
554 new = xcalloc(1, cache_entry_size(len));
555 memcpy(new->name, alias->name, len);
556 copy_cache_entry(new, ce);
557 free(ce);
558 return new;
559}
560
39425819
JH
561static void record_intent_to_add(struct cache_entry *ce)
562{
563 unsigned char sha1[20];
564 if (write_sha1_file("", 0, blob_type, sha1))
565 die("cannot create an empty blob in the object database");
566 hashcpy(ce->sha1, sha1);
567}
568
38ed1d89 569int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 570{
38ed1d89 571 int size, namelen, was_same;
d177cab0 572 mode_t st_mode = st->st_mode;
6835550d 573 struct cache_entry *ce, *alias;
56cac48c 574 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
575 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
576 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
577 int intent_only = flags & ADD_CACHE_INTENT;
578 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
579 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
11be42a4 580
d177cab0 581 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
960b8ad1 582 return error("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
583
584 namelen = strlen(path);
d177cab0 585 if (S_ISDIR(st_mode)) {
09595258
LT
586 while (namelen && path[namelen-1] == '/')
587 namelen--;
588 }
11be42a4
JS
589 size = cache_entry_size(namelen);
590 ce = xcalloc(1, size);
591 memcpy(ce->name, path, namelen);
7a51ed66 592 ce->ce_flags = namelen;
39425819
JH
593 if (!intent_only)
594 fill_stat_cache_info(ce, st);
388b2acd
JH
595 else
596 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 597
78a8d641 598 if (trust_executable_bit && has_symlinks)
d177cab0 599 ce->ce_mode = create_ce_mode(st_mode);
185c975f 600 else {
78a8d641
JS
601 /* If there is an existing entry, pick the mode bits and type
602 * from it, otherwise assume unexecutable regular file.
11be42a4 603 */
185c975f 604 struct cache_entry *ent;
20314271 605 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 606
4aab5b46 607 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 608 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
609 }
610
dc1ae704
JJ
611 /* When core.ignorecase=true, determine if a directory of the same name but differing
612 * case already exists within the Git repository. If it does, ensure the directory
613 * case of the file being added to the repository matches (is folded into) the existing
614 * entry's directory case.
615 */
616 if (ignore_case) {
617 const char *startPtr = ce->name;
618 const char *ptr = startPtr;
619 while (*ptr) {
620 while (*ptr && *ptr != '/')
621 ++ptr;
622 if (*ptr == '/') {
623 struct cache_entry *foundce;
624 ++ptr;
625 foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case);
626 if (foundce) {
627 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
628 startPtr = ptr;
629 }
630 }
631 }
632 }
633
6835550d 634 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
d177cab0 635 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
0781b8a9
JH
636 /* Nothing changed, really */
637 free(ce);
125fd984
JH
638 if (!S_ISGITLINK(alias->ce_mode))
639 ce_mark_uptodate(alias);
1102952b 640 alias->ce_flags |= CE_ADDED;
0781b8a9
JH
641 return 0;
642 }
39425819
JH
643 if (!intent_only) {
644 if (index_path(ce->sha1, path, st, 1))
645 return error("unable to index file %s", path);
646 } else
647 record_intent_to_add(ce);
648
1102952b
LT
649 if (ignore_case && alias && different_name(ce, alias))
650 ce = create_alias_ce(ce, alias);
651 ce->ce_flags |= CE_ADDED;
38ed1d89 652
3bf0dd1f 653 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
654 was_same = (alias &&
655 !ce_stage(alias) &&
656 !hashcmp(alias->sha1, ce->sha1) &&
657 ce->ce_mode == alias->ce_mode);
658
659 if (pretend)
660 ;
39425819 661 else if (add_index_entry(istate, ce, add_option))
960b8ad1 662 return error("unable to add %s to index",path);
38ed1d89 663 if (verbose && !was_same)
11be42a4 664 printf("add '%s'\n", path);
11be42a4
JS
665 return 0;
666}
667
38ed1d89 668int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
669{
670 struct stat st;
671 if (lstat(path, &st))
d824cbba 672 die_errno("unable to stat '%s'", path);
38ed1d89 673 return add_to_index(istate, path, &st, flags);
d177cab0
LT
674}
675
6640f881
CR
676struct cache_entry *make_cache_entry(unsigned int mode,
677 const unsigned char *sha1, const char *path, int stage,
678 int refresh)
679{
680 int size, len;
681 struct cache_entry *ce;
682
7e7abea9
DP
683 if (!verify_path(path)) {
684 error("Invalid path '%s'", path);
6640f881 685 return NULL;
7e7abea9 686 }
6640f881
CR
687
688 len = strlen(path);
689 size = cache_entry_size(len);
690 ce = xcalloc(1, size);
691
692 hashcpy(ce->sha1, sha1);
693 memcpy(ce->name, path, len);
694 ce->ce_flags = create_ce_flags(len, stage);
695 ce->ce_mode = create_ce_mode(mode);
696
697 if (refresh)
698 return refresh_cache_entry(ce, 0);
699
700 return ce;
701}
702
dbbce55b 703int ce_same_name(struct cache_entry *a, struct cache_entry *b)
7b937ca3
LT
704{
705 int len = ce_namelen(a);
706 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
707}
708
c0fd1f51
LT
709int ce_path_match(const struct cache_entry *ce, const char **pathspec)
710{
711 const char *match, *name;
712 int len;
713
714 if (!pathspec)
715 return 1;
716
717 len = ce_namelen(ce);
718 name = ce->name;
719 while ((match = *pathspec++) != NULL) {
720 int matchlen = strlen(match);
721 if (matchlen > len)
722 continue;
723 if (memcmp(name, match, matchlen))
724 continue;
725 if (matchlen && name[matchlen-1] == '/')
726 return 1;
727 if (name[matchlen] == '/' || !name[matchlen])
728 return 1;
f332726e
LT
729 if (!matchlen)
730 return 1;
c0fd1f51
LT
731 }
732 return 0;
733}
734
8dcf39c4
LT
735/*
736 * We fundamentally don't like some paths: we don't want
737 * dot or dot-dot anywhere, and for obvious reasons don't
738 * want to recurse into ".git" either.
739 *
740 * Also, we don't want double slashes or slashes at the
741 * end that can make pathnames ambiguous.
742 */
743static int verify_dotfile(const char *rest)
744{
745 /*
746 * The first character was '.', but that
747 * has already been discarded, we now test
748 * the rest.
749 */
750 switch (*rest) {
751 /* "." is not allowed */
752 case '\0': case '/':
753 return 0;
754
755 /*
756 * ".git" followed by NUL or slash is bad. This
757 * shares the path end test with the ".." case.
758 */
759 case 'g':
760 if (rest[1] != 'i')
761 break;
762 if (rest[2] != 't')
763 break;
764 rest += 2;
765 /* fallthrough */
766 case '.':
767 if (rest[1] == '\0' || rest[1] == '/')
768 return 0;
769 }
770 return 1;
771}
772
773int verify_path(const char *path)
774{
775 char c;
776
56948cb6
EFL
777 if (has_dos_drive_prefix(path))
778 return 0;
779
8dcf39c4
LT
780 goto inside;
781 for (;;) {
782 if (!c)
783 return 1;
56948cb6 784 if (is_dir_sep(c)) {
8dcf39c4
LT
785inside:
786 c = *path++;
3bdf09c7
JH
787 if ((c == '.' && !verify_dotfile(path)) ||
788 is_dir_sep(c) || c == '\0')
789 return 0;
8dcf39c4
LT
790 }
791 c = *path++;
792 }
793}
794
12676608
LT
795/*
796 * Do we have another file that has the beginning components being a
797 * proper superset of the name we're trying to add?
0f1e4f04 798 */
4aab5b46
JH
799static int has_file_name(struct index_state *istate,
800 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 801{
12676608
LT
802 int retval = 0;
803 int len = ce_namelen(ce);
b155725d 804 int stage = ce_stage(ce);
12676608 805 const char *name = ce->name;
0f1e4f04 806
4aab5b46
JH
807 while (pos < istate->cache_nr) {
808 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 809
12676608 810 if (len >= ce_namelen(p))
0f1e4f04 811 break;
12676608
LT
812 if (memcmp(name, p->name, len))
813 break;
b155725d
JH
814 if (ce_stage(p) != stage)
815 continue;
12676608
LT
816 if (p->name[len] != '/')
817 continue;
7a51ed66 818 if (p->ce_flags & CE_REMOVE)
21cd8d00 819 continue;
12676608
LT
820 retval = -1;
821 if (!ok_to_replace)
822 break;
4aab5b46 823 remove_index_entry_at(istate, --pos);
0f1e4f04 824 }
12676608
LT
825 return retval;
826}
0f1e4f04 827
12676608
LT
828/*
829 * Do we have another file with a pathname that is a proper
830 * subset of the name we're trying to add?
831 */
4aab5b46
JH
832static int has_dir_name(struct index_state *istate,
833 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
834{
835 int retval = 0;
b155725d 836 int stage = ce_stage(ce);
12676608
LT
837 const char *name = ce->name;
838 const char *slash = name + ce_namelen(ce);
0f1e4f04 839
12676608
LT
840 for (;;) {
841 int len;
0f1e4f04 842
12676608
LT
843 for (;;) {
844 if (*--slash == '/')
845 break;
846 if (slash <= ce->name)
847 return retval;
848 }
849 len = slash - name;
0f1e4f04 850
7a51ed66 851 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
12676608 852 if (pos >= 0) {
21cd8d00
JH
853 /*
854 * Found one, but not so fast. This could
855 * be a marker that says "I was here, but
856 * I am being removed". Such an entry is
857 * not a part of the resulting tree, and
858 * it is Ok to have a directory at the same
859 * path.
860 */
077c48df 861 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
862 retval = -1;
863 if (!ok_to_replace)
864 break;
4aab5b46 865 remove_index_entry_at(istate, pos);
21cd8d00
JH
866 continue;
867 }
12676608 868 }
21cd8d00
JH
869 else
870 pos = -pos-1;
12676608
LT
871
872 /*
873 * Trivial optimization: if we find an entry that
874 * already matches the sub-directory, then we know
b155725d 875 * we're ok, and we can exit.
12676608 876 */
4aab5b46
JH
877 while (pos < istate->cache_nr) {
878 struct cache_entry *p = istate->cache[pos];
b155725d
JH
879 if ((ce_namelen(p) <= len) ||
880 (p->name[len] != '/') ||
881 memcmp(p->name, name, len))
882 break; /* not our subdirectory */
077c48df
JH
883 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
884 /*
885 * p is at the same stage as our entry, and
b155725d
JH
886 * is a subdirectory of what we are looking
887 * at, so we cannot have conflicts at our
888 * level or anything shorter.
889 */
890 return retval;
891 pos++;
192268c1 892 }
0f1e4f04 893 }
12676608
LT
894 return retval;
895}
896
897/* We may be in a situation where we already have path/file and path
898 * is being added, or we already have path and path/file is being
899 * added. Either one would result in a nonsense tree that has path
900 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 901 *
12676608
LT
902 * If ok-to-replace is specified, we remove the conflicting entries
903 * from the cache so the caller should recompute the insert position.
904 * When this happens, we return non-zero.
905 */
4aab5b46
JH
906static int check_file_directory_conflict(struct index_state *istate,
907 const struct cache_entry *ce,
908 int pos, int ok_to_replace)
12676608 909{
21cd8d00
JH
910 int retval;
911
912 /*
913 * When ce is an "I am going away" entry, we allow it to be added
914 */
7a51ed66 915 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
916 return 0;
917
12676608
LT
918 /*
919 * We check if the path is a sub-path of a subsequent pathname
920 * first, since removing those will not change the position
21cd8d00 921 * in the array.
12676608 922 */
4aab5b46 923 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 924
12676608
LT
925 /*
926 * Then check if the path might have a clashing sub-directory
927 * before it.
928 */
4aab5b46 929 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
930}
931
af3785dc 932static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
933{
934 int pos;
192268c1
JH
935 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
936 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 937 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 938 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 939
09d5dc32 940 cache_tree_invalidate_path(istate->cache_tree, ce->name);
7a51ed66 941 pos = index_name_pos(istate, ce->name, ce->ce_flags);
197ee8c9 942
3e09cdfd 943 /* existing match? Just replace it. */
76e7f4ec 944 if (pos >= 0) {
39425819
JH
945 if (!new_only)
946 replace_index_entry(istate, pos, ce);
197ee8c9
LT
947 return 0;
948 }
76e7f4ec 949 pos = -pos-1;
197ee8c9 950
7b937ca3
LT
951 /*
952 * Inserting a merged entry ("stage 0") into the index
953 * will always replace all non-merged entries..
954 */
4aab5b46
JH
955 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
956 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 957 ok_to_add = 1;
4aab5b46 958 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
959 break;
960 }
961 }
962
121481ab
LT
963 if (!ok_to_add)
964 return -1;
8dcf39c4 965 if (!verify_path(ce->name))
7e7abea9 966 return error("Invalid path '%s'", ce->name);
121481ab 967
3e09cdfd 968 if (!skip_df_check &&
4aab5b46 969 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 970 if (!ok_to_replace)
4aab5b46
JH
971 return error("'%s' appears as both a file and as a directory",
972 ce->name);
7a51ed66 973 pos = index_name_pos(istate, ce->name, ce->ce_flags);
192268c1
JH
974 pos = -pos-1;
975 }
af3785dc
JH
976 return pos + 1;
977}
978
979int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
980{
981 int pos;
982
983 if (option & ADD_CACHE_JUST_APPEND)
984 pos = istate->cache_nr;
985 else {
986 int ret;
987 ret = add_index_entry_with_check(istate, ce, option);
988 if (ret <= 0)
989 return ret;
990 pos = ret - 1;
991 }
0f1e4f04 992
197ee8c9 993 /* Make sure the array is big enough .. */
4aab5b46
JH
994 if (istate->cache_nr == istate->cache_alloc) {
995 istate->cache_alloc = alloc_nr(istate->cache_alloc);
996 istate->cache = xrealloc(istate->cache,
997 istate->cache_alloc * sizeof(struct cache_entry *));
197ee8c9
LT
998 }
999
1000 /* Add it in.. */
4aab5b46 1001 istate->cache_nr++;
af3785dc 1002 if (istate->cache_nr > pos + 1)
4aab5b46
JH
1003 memmove(istate->cache + pos + 1,
1004 istate->cache + pos,
1005 (istate->cache_nr - pos - 1) * sizeof(ce));
cf558704 1006 set_index_entry(istate, pos, ce);
4aab5b46 1007 istate->cache_changed = 1;
197ee8c9
LT
1008 return 0;
1009}
1010
405e5b2f
LT
1011/*
1012 * "refresh" does not calculate a new sha1 file or bring the
1013 * cache up-to-date for mode/content changes. But what it
1014 * _does_ do is to "re-match" the stat information of a file
1015 * with the cache, so that you can refresh the cache for a
1016 * file that hasn't been changed but where the stat entry is
1017 * out of date.
1018 *
1019 * For example, you'd want to do this after doing a "git-read-tree",
1020 * to link up the stat cache details with the proper files.
1021 */
4aab5b46 1022static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da
JH
1023 struct cache_entry *ce,
1024 unsigned int options, int *err)
405e5b2f
LT
1025{
1026 struct stat st;
1027 struct cache_entry *updated;
1028 int changed, size;
4bd5b7da 1029 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1030 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
405e5b2f 1031
eadb5831
JH
1032 if (ce_uptodate(ce))
1033 return ce;
1034
aa9349d4 1035 /*
56cac48c
NTND
1036 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1037 * that the change to the work tree does not matter and told
1038 * us not to worry.
aa9349d4 1039 */
56cac48c
NTND
1040 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1041 ce_mark_uptodate(ce);
1042 return ce;
1043 }
aa9349d4
MSO
1044 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1045 ce_mark_uptodate(ce);
1046 return ce;
1047 }
1048
8fd2cb40 1049 if (lstat(ce->name, &st) < 0) {
ec0cc704
JH
1050 if (err)
1051 *err = errno;
8fd2cb40
JS
1052 return NULL;
1053 }
405e5b2f 1054
4bd5b7da 1055 changed = ie_match_stat(istate, ce, &st, options);
405e5b2f 1056 if (!changed) {
4bd5b7da
JH
1057 /*
1058 * The path is unchanged. If we were told to ignore
1059 * valid bit, then we did the actual stat check and
1060 * found that the entry is unmodified. If the entry
1061 * is not marked VALID, this is the place to mark it
1062 * valid again, under "assume unchanged" mode.
1063 */
1064 if (ignore_valid && assume_unchanged &&
7a51ed66 1065 !(ce->ce_flags & CE_VALID))
405e5b2f 1066 ; /* mark this one VALID again */
eadb5831
JH
1067 else {
1068 /*
1069 * We do not mark the index itself "modified"
1070 * because CE_UPTODATE flag is in-core only;
1071 * we are not going to write this change out.
1072 */
125fd984
JH
1073 if (!S_ISGITLINK(ce->ce_mode))
1074 ce_mark_uptodate(ce);
8fd2cb40 1075 return ce;
eadb5831 1076 }
405e5b2f
LT
1077 }
1078
4bd5b7da 1079 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1080 if (err)
1081 *err = EINVAL;
8fd2cb40
JS
1082 return NULL;
1083 }
405e5b2f
LT
1084
1085 size = ce_size(ce);
1086 updated = xmalloc(size);
1087 memcpy(updated, ce, size);
1088 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1089 /*
1090 * If ignore_valid is not set, we should leave CE_VALID bit
1091 * alone. Otherwise, paths marked with --no-assume-unchanged
1092 * (i.e. things to be edited) will reacquire CE_VALID bit
1093 * automatically, which is not really what we want.
405e5b2f 1094 */
4bd5b7da 1095 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1096 !(ce->ce_flags & CE_VALID))
1097 updated->ce_flags &= ~CE_VALID;
405e5b2f
LT
1098
1099 return updated;
1100}
1101
3deffc52
MM
1102static void show_file(const char * fmt, const char * name, int in_porcelain,
1103 int * first, char *header_msg)
1104{
1105 if (in_porcelain && *first && header_msg) {
1106 printf("%s\n", header_msg);
1107 *first=0;
1108 }
1109 printf(fmt, name);
1110}
1111
1112int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1113 char *seen, char *header_msg)
405e5b2f
LT
1114{
1115 int i;
1116 int has_errors = 0;
1117 int really = (flags & REFRESH_REALLY) != 0;
1118 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1119 int quiet = (flags & REFRESH_QUIET) != 0;
1120 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1121 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1122 int first = 1;
1123 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
4bd5b7da 1124 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
3deffc52
MM
1125 const char *needs_update_fmt;
1126 const char *needs_merge_fmt;
405e5b2f 1127
3deffc52
MM
1128 needs_update_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1129 needs_merge_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
4aab5b46 1130 for (i = 0; i < istate->cache_nr; i++) {
405e5b2f 1131 struct cache_entry *ce, *new;
ec0cc704
JH
1132 int cache_errno = 0;
1133
4aab5b46 1134 ce = istate->cache[i];
5fdeacb0
JS
1135 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1136 continue;
1137
405e5b2f 1138 if (ce_stage(ce)) {
4aab5b46
JH
1139 while ((i < istate->cache_nr) &&
1140 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1141 i++;
1142 i--;
1143 if (allow_unmerged)
1144 continue;
3deffc52 1145 show_file(needs_merge_fmt, ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1146 has_errors = 1;
1147 continue;
1148 }
1149
d616813d
AJ
1150 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1151 continue;
1152
4bd5b7da 1153 new = refresh_cache_ent(istate, ce, options, &cache_errno);
8fd2cb40 1154 if (new == ce)
405e5b2f 1155 continue;
8fd2cb40
JS
1156 if (!new) {
1157 if (not_new && cache_errno == ENOENT)
405e5b2f 1158 continue;
8fd2cb40 1159 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1160 /* If we are doing --really-refresh that
1161 * means the index is not valid anymore.
1162 */
7a51ed66 1163 ce->ce_flags &= ~CE_VALID;
4aab5b46 1164 istate->cache_changed = 1;
405e5b2f
LT
1165 }
1166 if (quiet)
1167 continue;
3deffc52 1168 show_file(needs_update_fmt, ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1169 has_errors = 1;
1170 continue;
1171 }
cf558704
LT
1172
1173 replace_index_entry(istate, i, new);
405e5b2f
LT
1174 }
1175 return has_errors;
1176}
1177
87b29e5a 1178static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
ec0cc704 1179{
4aab5b46 1180 return refresh_cache_ent(&the_index, ce, really, NULL);
ec0cc704
JH
1181}
1182
bad68ec9 1183static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163 1184{
9126f009 1185 git_SHA_CTX c;
bad68ec9 1186 unsigned char sha1[20];
e83c5163 1187
ccc4feb5 1188 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1189 return error("bad signature");
06aaaa0b 1190 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
ca9be054 1191 return error("bad index version");
9126f009
NP
1192 git_SHA1_Init(&c);
1193 git_SHA1_Update(&c, hdr, size - 20);
1194 git_SHA1_Final(sha1, &c);
a89fccd2 1195 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
ca9be054 1196 return error("bad index file sha1 signature");
e83c5163
LT
1197 return 0;
1198}
1199
4aab5b46
JH
1200static int read_index_extension(struct index_state *istate,
1201 const char *ext, void *data, unsigned long sz)
bad68ec9
JH
1202{
1203 switch (CACHE_EXT(ext)) {
1204 case CACHE_EXT_TREE:
4aab5b46 1205 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1206 break;
cfc5789a
JH
1207 case CACHE_EXT_RESOLVE_UNDO:
1208 istate->resolve_undo = resolve_undo_read(data, sz);
1209 break;
bad68ec9
JH
1210 default:
1211 if (*ext < 'A' || 'Z' < *ext)
1212 return error("index uses %.4s extension, which we do not understand",
1213 ext);
1214 fprintf(stderr, "ignoring %.4s extension\n", ext);
1215 break;
1216 }
1217 return 0;
1218}
1219
4aab5b46 1220int read_index(struct index_state *istate)
8fd2cb40 1221{
4aab5b46 1222 return read_index_from(istate, get_index_file());
8fd2cb40
JS
1223}
1224
7a51ed66
LT
1225static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1226{
7fec10b7 1227 size_t len;
06aaaa0b 1228 const char *name;
7fec10b7 1229
fba2f38a
KB
1230 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1231 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
fba2f38a
KB
1232 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1233 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
7a51ed66
LT
1234 ce->ce_dev = ntohl(ondisk->dev);
1235 ce->ce_ino = ntohl(ondisk->ino);
1236 ce->ce_mode = ntohl(ondisk->mode);
1237 ce->ce_uid = ntohl(ondisk->uid);
1238 ce->ce_gid = ntohl(ondisk->gid);
1239 ce->ce_size = ntohl(ondisk->size);
1240 /* On-disk flags are just 16 bits */
1241 ce->ce_flags = ntohs(ondisk->flags);
16ce2e4c 1242
7a51ed66 1243 hashcpy(ce->sha1, ondisk->sha1);
7fec10b7
JH
1244
1245 len = ce->ce_flags & CE_NAMEMASK;
06aaaa0b
NTND
1246
1247 if (ce->ce_flags & CE_EXTENDED) {
1248 struct ondisk_cache_entry_extended *ondisk2;
1249 int extended_flags;
1250 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1251 extended_flags = ntohs(ondisk2->flags2) << 16;
1252 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1253 if (extended_flags & ~CE_EXTENDED_FLAGS)
1254 die("Unknown index entry format %08x", extended_flags);
1255 ce->ce_flags |= extended_flags;
1256 name = ondisk2->name;
1257 }
1258 else
1259 name = ondisk->name;
1260
7fec10b7 1261 if (len == CE_NAMEMASK)
06aaaa0b 1262 len = strlen(name);
7fec10b7
JH
1263 /*
1264 * NEEDSWORK: If the original index is crafted, this copy could
1265 * go unchecked.
1266 */
06aaaa0b 1267 memcpy(ce->name, name, len + 1);
7a51ed66
LT
1268}
1269
cf558704
LT
1270static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1271{
1272 long per_entry;
1273
1274 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1275
1276 /*
1277 * Alignment can cause differences. This should be "alignof", but
1278 * since that's a gcc'ism, just use the size of a pointer.
1279 */
1280 per_entry += sizeof(void *);
1281 return ondisk_size + entries*per_entry;
1282}
1283
8fd2cb40 1284/* remember to discard_cache() before reading a different cache! */
4aab5b46 1285int read_index_from(struct index_state *istate, const char *path)
e83c5163
LT
1286{
1287 int fd, i;
1288 struct stat st;
7a51ed66 1289 unsigned long src_offset, dst_offset;
e83c5163 1290 struct cache_header *hdr;
7a51ed66
LT
1291 void *mmap;
1292 size_t mmap_size;
e83c5163
LT
1293
1294 errno = EBUSY;
913e0e99 1295 if (istate->initialized)
4aab5b46 1296 return istate->cache_nr;
5d1a5c02 1297
e83c5163 1298 errno = ENOENT;
fba2f38a
KB
1299 istate->timestamp.sec = 0;
1300 istate->timestamp.nsec = 0;
8fd2cb40 1301 fd = open(path, O_RDONLY);
5d1a5c02
LT
1302 if (fd < 0) {
1303 if (errno == ENOENT)
1304 return 0;
d824cbba 1305 die_errno("index file open failed");
5d1a5c02 1306 }
e83c5163 1307
3511a377 1308 if (fstat(fd, &st))
d824cbba 1309 die_errno("cannot stat the open index");
3511a377
LFC
1310
1311 errno = EINVAL;
7a51ed66
LT
1312 mmap_size = xsize_t(st.st_size);
1313 if (mmap_size < sizeof(struct cache_header) + 20)
3511a377
LFC
1314 die("index file smaller than expected");
1315
7a51ed66 1316 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
e83c5163 1317 close(fd);
7a51ed66 1318 if (mmap == MAP_FAILED)
0721c314 1319 die_errno("unable to map index file");
e83c5163 1320
7a51ed66
LT
1321 hdr = mmap;
1322 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1323 goto unmap;
1324
4aab5b46
JH
1325 istate->cache_nr = ntohl(hdr->hdr_entries);
1326 istate->cache_alloc = alloc_nr(istate->cache_nr);
1327 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
e83c5163 1328
7a51ed66
LT
1329 /*
1330 * The disk format is actually larger than the in-memory format,
1331 * due to space for nsec etc, so even though the in-memory one
1332 * has room for a few more flags, we can allocate using the same
1333 * index size
1334 */
cf558704 1335 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
913e0e99 1336 istate->initialized = 1;
7a51ed66
LT
1337
1338 src_offset = sizeof(*hdr);
1339 dst_offset = 0;
4aab5b46 1340 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 1341 struct ondisk_cache_entry *disk_ce;
4aab5b46
JH
1342 struct cache_entry *ce;
1343
7a51ed66
LT
1344 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1345 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1346 convert_from_disk(disk_ce, ce);
cf558704 1347 set_index_entry(istate, i, ce);
7a51ed66
LT
1348
1349 src_offset += ondisk_ce_size(ce);
1350 dst_offset += ce_size(ce);
e83c5163 1351 }
fba2f38a 1352 istate->timestamp.sec = st.st_mtime;
c06ff490 1353 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 1354
7a51ed66 1355 while (src_offset <= mmap_size - 20 - 8) {
bad68ec9
JH
1356 /* After an array of active_nr index entries,
1357 * there can be arbitrary number of extended
1358 * sections, each of which is prefixed with
1359 * extension name (4-byte) and section length
1360 * in 4-byte network byte order.
1361 */
07cc8eca 1362 uint32_t extsize;
7a51ed66 1363 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
bad68ec9 1364 extsize = ntohl(extsize);
4aab5b46 1365 if (read_index_extension(istate,
7a51ed66
LT
1366 (const char *) mmap + src_offset,
1367 (char *) mmap + src_offset + 8,
1d7f171c 1368 extsize) < 0)
bad68ec9 1369 goto unmap;
7a51ed66
LT
1370 src_offset += 8;
1371 src_offset += extsize;
bad68ec9 1372 }
7a51ed66 1373 munmap(mmap, mmap_size);
4aab5b46 1374 return istate->cache_nr;
e83c5163
LT
1375
1376unmap:
7a51ed66 1377 munmap(mmap, mmap_size);
e83c5163 1378 errno = EINVAL;
5d1a5c02 1379 die("index file corrupt");
e83c5163
LT
1380}
1381
fa7b3c2f
JH
1382int is_index_unborn(struct index_state *istate)
1383{
fba2f38a 1384 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
fa7b3c2f
JH
1385}
1386
4aab5b46 1387int discard_index(struct index_state *istate)
6d297f81 1388{
cfc5789a 1389 resolve_undo_clear_index(istate);
4aab5b46
JH
1390 istate->cache_nr = 0;
1391 istate->cache_changed = 0;
fba2f38a
KB
1392 istate->timestamp.sec = 0;
1393 istate->timestamp.nsec = 0;
64ca23af 1394 istate->name_hash_initialized = 0;
cf558704 1395 free_hash(&istate->name_hash);
4aab5b46 1396 cache_tree_free(&(istate->cache_tree));
7a51ed66
LT
1397 free(istate->alloc);
1398 istate->alloc = NULL;
913e0e99 1399 istate->initialized = 0;
6d297f81
JS
1400
1401 /* no need to throw away allocated active_cache */
7a51ed66 1402 return 0;
6d297f81
JS
1403}
1404
d1f128b0 1405int unmerged_index(const struct index_state *istate)
94a5728c
DB
1406{
1407 int i;
1408 for (i = 0; i < istate->cache_nr; i++) {
1409 if (ce_stage(istate->cache[i]))
1410 return 1;
1411 }
1412 return 0;
1413}
1414
4990aadc 1415#define WRITE_BUFFER_SIZE 8192
bf0f910d 1416static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
1417static unsigned long write_buffer_len;
1418
9126f009 1419static int ce_write_flush(git_SHA_CTX *context, int fd)
6015c28b
JH
1420{
1421 unsigned int buffered = write_buffer_len;
1422 if (buffered) {
9126f009 1423 git_SHA1_Update(context, write_buffer, buffered);
93822c22 1424 if (write_in_full(fd, write_buffer, buffered) != buffered)
6015c28b
JH
1425 return -1;
1426 write_buffer_len = 0;
1427 }
1428 return 0;
1429}
1430
9126f009 1431static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
1432{
1433 while (len) {
1434 unsigned int buffered = write_buffer_len;
1435 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1436 if (partial > len)
1437 partial = len;
1438 memcpy(write_buffer + buffered, data, partial);
1439 buffered += partial;
1440 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
1441 write_buffer_len = buffered;
1442 if (ce_write_flush(context, fd))
4990aadc
LT
1443 return -1;
1444 buffered = 0;
1445 }
1446 write_buffer_len = buffered;
1447 len -= partial;
1d7f171c 1448 data = (char *) data + partial;
a6080a0a
JH
1449 }
1450 return 0;
4990aadc
LT
1451}
1452
9126f009 1453static int write_index_ext_header(git_SHA_CTX *context, int fd,
ac58c7b1 1454 unsigned int ext, unsigned int sz)
bad68ec9
JH
1455{
1456 ext = htonl(ext);
1457 sz = htonl(sz);
968a1d65
DR
1458 return ((ce_write(context, fd, &ext, 4) < 0) ||
1459 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
1460}
1461
9126f009 1462static int ce_flush(git_SHA_CTX *context, int fd)
4990aadc
LT
1463{
1464 unsigned int left = write_buffer_len;
ca9be054 1465
4990aadc
LT
1466 if (left) {
1467 write_buffer_len = 0;
9126f009 1468 git_SHA1_Update(context, write_buffer, left);
4990aadc 1469 }
ca9be054 1470
2c865d9a
QH
1471 /* Flush first if not enough space for SHA1 signature */
1472 if (left + 20 > WRITE_BUFFER_SIZE) {
93822c22 1473 if (write_in_full(fd, write_buffer, left) != left)
2c865d9a
QH
1474 return -1;
1475 left = 0;
1476 }
1477
ca9be054 1478 /* Append the SHA1 signature at the end */
9126f009 1479 git_SHA1_Final(write_buffer + left, context);
ca9be054 1480 left += 20;
93822c22 1481 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
4990aadc
LT
1482}
1483
407c8eb0
JH
1484static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1485{
1486 /*
1487 * The only thing we care about in this function is to smudge the
1488 * falsely clean entry due to touch-update-touch race, so we leave
1489 * everything else as they are. We are called for entries whose
1490 * ce_mtime match the index file mtime.
c70115b4
JH
1491 *
1492 * Note that this actually does not do much for gitlinks, for
1493 * which ce_match_stat_basic() always goes to the actual
1494 * contents. The caller checks with is_racy_timestamp() which
1495 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
1496 */
1497 struct stat st;
1498
1499 if (lstat(ce->name, &st) < 0)
1500 return;
1501 if (ce_match_stat_basic(ce, &st))
1502 return;
1503 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
1504 /* This is "racily clean"; smudge it. Note that this
1505 * is a tricky code. At first glance, it may appear
1506 * that it can break with this sequence:
1507 *
1508 * $ echo xyzzy >frotz
1509 * $ git-update-index --add frotz
1510 * $ : >frotz
1511 * $ sleep 3
1512 * $ echo filfre >nitfol
1513 * $ git-update-index --add nitfol
1514 *
b7e58b17 1515 * but it does not. When the second update-index runs,
4b3511b0
JH
1516 * it notices that the entry "frotz" has the same timestamp
1517 * as index, and if we were to smudge it by resetting its
1518 * size to zero here, then the object name recorded
1519 * in index is the 6-byte file but the cached stat information
1520 * becomes zero --- which would then match what we would
a6080a0a 1521 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
1522 *
1523 * However, the second update-index, before calling
1524 * this function, notices that the cached size is 6
1525 * bytes and what is on the filesystem is an empty
1526 * file, and never calls us, so the cached size information
1527 * for "frotz" stays 6 which does not match the filesystem.
1528 */
7a51ed66 1529 ce->ce_size = 0;
407c8eb0
JH
1530 }
1531}
1532
9126f009 1533static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
7a51ed66
LT
1534{
1535 int size = ondisk_ce_size(ce);
1536 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
06aaaa0b 1537 char *name;
59efba64 1538 int result;
7a51ed66 1539
fba2f38a
KB
1540 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1541 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
fba2f38a
KB
1542 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1543 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
7a51ed66
LT
1544 ondisk->dev = htonl(ce->ce_dev);
1545 ondisk->ino = htonl(ce->ce_ino);
1546 ondisk->mode = htonl(ce->ce_mode);
1547 ondisk->uid = htonl(ce->ce_uid);
1548 ondisk->gid = htonl(ce->ce_gid);
1549 ondisk->size = htonl(ce->ce_size);
1550 hashcpy(ondisk->sha1, ce->sha1);
1551 ondisk->flags = htons(ce->ce_flags);
06aaaa0b
NTND
1552 if (ce->ce_flags & CE_EXTENDED) {
1553 struct ondisk_cache_entry_extended *ondisk2;
1554 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1555 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1556 name = ondisk2->name;
1557 }
1558 else
1559 name = ondisk->name;
1560 memcpy(name, ce->name, ce_namelen(ce));
7a51ed66 1561
59efba64
JN
1562 result = ce_write(c, fd, ondisk, size);
1563 free(ondisk);
1564 return result;
7a51ed66
LT
1565}
1566
483fbe2b
JH
1567static int has_racy_timestamp(struct index_state *istate)
1568{
1569 int entries = istate->cache_nr;
1570 int i;
1571
1572 for (i = 0; i < entries; i++) {
1573 struct cache_entry *ce = istate->cache[i];
1574 if (is_racy_timestamp(istate, ce))
1575 return 1;
1576 }
1577 return 0;
1578}
1579
ccdc4ec3
JH
1580/*
1581 * Opportunisticly update the index but do not complain if we can't
1582 */
1583void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1584{
483fbe2b 1585 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
ccdc4ec3
JH
1586 !write_index(istate, lockfile->fd))
1587 commit_locked_index(lockfile);
1588 else
1589 rollback_lock_file(lockfile);
1590}
1591
e1afca4f 1592int write_index(struct index_state *istate, int newfd)
197ee8c9 1593{
9126f009 1594 git_SHA_CTX c;
197ee8c9 1595 struct cache_header hdr;
06aaaa0b 1596 int i, err, removed, extended;
4aab5b46
JH
1597 struct cache_entry **cache = istate->cache;
1598 int entries = istate->cache_nr;
e1afca4f 1599 struct stat st;
025a0709 1600
06aaaa0b 1601 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 1602 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 1603 removed++;
197ee8c9 1604
06aaaa0b
NTND
1605 /* reduce extended entries if possible */
1606 cache[i]->ce_flags &= ~CE_EXTENDED;
1607 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1608 extended++;
1609 cache[i]->ce_flags |= CE_EXTENDED;
1610 }
1611 }
1612
ccc4feb5 1613 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
06aaaa0b
NTND
1614 /* for extended format, increase version so older git won't try to read it */
1615 hdr.hdr_version = htonl(extended ? 3 : 2);
025a0709 1616 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 1617
9126f009 1618 git_SHA1_Init(&c);
ca9be054 1619 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
1620 return -1;
1621
1622 for (i = 0; i < entries; i++) {
1623 struct cache_entry *ce = cache[i];
7a51ed66 1624 if (ce->ce_flags & CE_REMOVE)
aa16021e 1625 continue;
e06c43c7 1626 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 1627 ce_smudge_racily_clean_entry(ce);
7a51ed66 1628 if (ce_write_entry(&c, newfd, ce) < 0)
197ee8c9
LT
1629 return -1;
1630 }
1af1c2b6 1631
bad68ec9 1632 /* Write extension data here */
4aab5b46 1633 if (istate->cache_tree) {
f285a2d7 1634 struct strbuf sb = STRBUF_INIT;
1dffb8fa 1635
1dffb8fa
PH
1636 cache_tree_write(&sb, istate->cache_tree);
1637 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1638 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1639 strbuf_release(&sb);
1640 if (err)
bad68ec9 1641 return -1;
bad68ec9 1642 }
cfc5789a
JH
1643 if (istate->resolve_undo) {
1644 struct strbuf sb = STRBUF_INIT;
1645
1646 resolve_undo_write(&sb, istate->resolve_undo);
1647 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1648 sb.len) < 0
1649 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1650 strbuf_release(&sb);
1651 if (err)
1652 return -1;
1653 }
e1afca4f
KB
1654
1655 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1656 return -1;
5bcf109c
KB
1657 istate->timestamp.sec = (unsigned int)st.st_mtime;
1658 istate->timestamp.nsec = ST_MTIME_NSEC(st);
e1afca4f 1659 return 0;
197ee8c9 1660}
e46bbcf6
MV
1661
1662/*
1663 * Read the index file that is potentially unmerged into given
63e8dc5b 1664 * index_state, dropping any unmerged entries. Returns true if
e46bbcf6
MV
1665 * the index is unmerged. Callers who want to refuse to work
1666 * from an unmerged state can call this and check its return value,
1667 * instead of calling read_cache().
1668 */
1669int read_index_unmerged(struct index_state *istate)
1670{
1671 int i;
d1a43f2a 1672 int unmerged = 0;
e46bbcf6
MV
1673
1674 read_index(istate);
e46bbcf6
MV
1675 for (i = 0; i < istate->cache_nr; i++) {
1676 struct cache_entry *ce = istate->cache[i];
d1a43f2a
JH
1677 struct cache_entry *new_ce;
1678 int size, len;
1679
1680 if (!ce_stage(ce))
e46bbcf6 1681 continue;
d1a43f2a
JH
1682 unmerged = 1;
1683 len = strlen(ce->name);
1684 size = cache_entry_size(len);
1685 new_ce = xcalloc(1, size);
d1a43f2a 1686 memcpy(new_ce->name, ce->name, len);
e11d7b59 1687 new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;
d1a43f2a
JH
1688 new_ce->ce_mode = ce->ce_mode;
1689 if (add_index_entry(istate, new_ce, 0))
1690 return error("%s: cannot drop to stage #0",
1691 ce->name);
1692 i = index_name_pos(istate, new_ce->name, len);
e46bbcf6 1693 }
d1a43f2a 1694 return unmerged;
e46bbcf6 1695}
041aee31 1696
98fa4738
JK
1697/*
1698 * Returns 1 if the path is an "other" path with respect to
1699 * the index; that is, the path is not mentioned in the index at all,
1700 * either as a file, a directory with some files in the index,
1701 * or as an unmerged entry.
1702 *
1703 * We helpfully remove a trailing "/" from directories so that
1704 * the output of read_directory can be used as-is.
1705 */
1706int index_name_is_other(const struct index_state *istate, const char *name,
1707 int namelen)
1708{
1709 int pos;
1710 if (namelen && name[namelen - 1] == '/')
1711 namelen--;
1712 pos = index_name_pos(istate, name, namelen);
1713 if (0 <= pos)
1714 return 0; /* exact match */
1715 pos = -pos - 1;
1716 if (pos < istate->cache_nr) {
1717 struct cache_entry *ce = istate->cache[pos];
1718 if (ce_namelen(ce) == namelen &&
1719 !memcmp(ce->name, name, namelen))
1720 return 0; /* Yup, this one exists unmerged */
1721 }
1722 return 1;
1723}