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