]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * GIT - The information manager from hell | |
3 | * | |
4 | * Copyright (C) Linus Torvalds, 2005 | |
5 | */ | |
6 | #define NO_THE_INDEX_COMPATIBILITY_MACROS | |
7 | #include "cache.h" | |
8 | #include "config.h" | |
9 | #include "diff.h" | |
10 | #include "diffcore.h" | |
11 | #include "tempfile.h" | |
12 | #include "lockfile.h" | |
13 | #include "cache-tree.h" | |
14 | #include "refs.h" | |
15 | #include "dir.h" | |
16 | #include "object-store.h" | |
17 | #include "tree.h" | |
18 | #include "commit.h" | |
19 | #include "blob.h" | |
20 | #include "resolve-undo.h" | |
21 | #include "strbuf.h" | |
22 | #include "varint.h" | |
23 | #include "split-index.h" | |
24 | #include "utf8.h" | |
25 | #include "fsmonitor.h" | |
26 | #include "thread-utils.h" | |
27 | #include "progress.h" | |
28 | ||
29 | /* Mask for the name length in ce_flags in the on-disk index */ | |
30 | ||
31 | #define CE_NAMEMASK (0x0fff) | |
32 | ||
33 | /* Index extensions. | |
34 | * | |
35 | * The first letter should be 'A'..'Z' for extensions that are not | |
36 | * necessary for a correct operation (i.e. optimization data). | |
37 | * When new extensions are added that _needs_ to be understood in | |
38 | * order to correctly interpret the index file, pick character that | |
39 | * is outside the range, to cause the reader to abort. | |
40 | */ | |
41 | ||
42 | #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) ) | |
43 | #define CACHE_EXT_TREE 0x54524545 /* "TREE" */ | |
44 | #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */ | |
45 | #define CACHE_EXT_LINK 0x6c696e6b /* "link" */ | |
46 | #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */ | |
47 | #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */ | |
48 | #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */ | |
49 | #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */ | |
50 | ||
51 | /* changes that can be kept in $GIT_DIR/index (basically all extensions) */ | |
52 | #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \ | |
53 | CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \ | |
54 | SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED) | |
55 | ||
56 | ||
57 | /* | |
58 | * This is an estimate of the pathname length in the index. We use | |
59 | * this for V4 index files to guess the un-deltafied size of the index | |
60 | * in memory because of pathname deltafication. This is not required | |
61 | * for V2/V3 index formats because their pathnames are not compressed. | |
62 | * If the initial amount of memory set aside is not sufficient, the | |
63 | * mem pool will allocate extra memory. | |
64 | */ | |
65 | #define CACHE_ENTRY_PATH_LENGTH 80 | |
66 | ||
67 | static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len) | |
68 | { | |
69 | struct cache_entry *ce; | |
70 | ce = mem_pool_alloc(mem_pool, cache_entry_size(len)); | |
71 | ce->mem_pool_allocated = 1; | |
72 | return ce; | |
73 | } | |
74 | ||
75 | static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len) | |
76 | { | |
77 | struct cache_entry * ce; | |
78 | ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len)); | |
79 | ce->mem_pool_allocated = 1; | |
80 | return ce; | |
81 | } | |
82 | ||
83 | static struct mem_pool *find_mem_pool(struct index_state *istate) | |
84 | { | |
85 | struct mem_pool **pool_ptr; | |
86 | ||
87 | if (istate->split_index && istate->split_index->base) | |
88 | pool_ptr = &istate->split_index->base->ce_mem_pool; | |
89 | else | |
90 | pool_ptr = &istate->ce_mem_pool; | |
91 | ||
92 | if (!*pool_ptr) | |
93 | mem_pool_init(pool_ptr, 0); | |
94 | ||
95 | return *pool_ptr; | |
96 | } | |
97 | ||
98 | struct index_state the_index; | |
99 | static const char *alternate_index_output; | |
100 | ||
101 | static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) | |
102 | { | |
103 | istate->cache[nr] = ce; | |
104 | add_name_hash(istate, ce); | |
105 | } | |
106 | ||
107 | static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) | |
108 | { | |
109 | struct cache_entry *old = istate->cache[nr]; | |
110 | ||
111 | replace_index_entry_in_base(istate, old, ce); | |
112 | remove_name_hash(istate, old); | |
113 | discard_cache_entry(old); | |
114 | ce->ce_flags &= ~CE_HASHED; | |
115 | set_index_entry(istate, nr, ce); | |
116 | ce->ce_flags |= CE_UPDATE_IN_BASE; | |
117 | mark_fsmonitor_invalid(istate, ce); | |
118 | istate->cache_changed |= CE_ENTRY_CHANGED; | |
119 | } | |
120 | ||
121 | void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name) | |
122 | { | |
123 | struct cache_entry *old_entry = istate->cache[nr], *new_entry; | |
124 | int namelen = strlen(new_name); | |
125 | ||
126 | new_entry = make_empty_cache_entry(istate, namelen); | |
127 | copy_cache_entry(new_entry, old_entry); | |
128 | new_entry->ce_flags &= ~CE_HASHED; | |
129 | new_entry->ce_namelen = namelen; | |
130 | new_entry->index = 0; | |
131 | memcpy(new_entry->name, new_name, namelen + 1); | |
132 | ||
133 | cache_tree_invalidate_path(istate, old_entry->name); | |
134 | untracked_cache_remove_from_index(istate, old_entry->name); | |
135 | remove_index_entry_at(istate, nr); | |
136 | add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); | |
137 | } | |
138 | ||
139 | void fill_stat_data(struct stat_data *sd, struct stat *st) | |
140 | { | |
141 | sd->sd_ctime.sec = (unsigned int)st->st_ctime; | |
142 | sd->sd_mtime.sec = (unsigned int)st->st_mtime; | |
143 | sd->sd_ctime.nsec = ST_CTIME_NSEC(*st); | |
144 | sd->sd_mtime.nsec = ST_MTIME_NSEC(*st); | |
145 | sd->sd_dev = st->st_dev; | |
146 | sd->sd_ino = st->st_ino; | |
147 | sd->sd_uid = st->st_uid; | |
148 | sd->sd_gid = st->st_gid; | |
149 | sd->sd_size = st->st_size; | |
150 | } | |
151 | ||
152 | int match_stat_data(const struct stat_data *sd, struct stat *st) | |
153 | { | |
154 | int changed = 0; | |
155 | ||
156 | if (sd->sd_mtime.sec != (unsigned int)st->st_mtime) | |
157 | changed |= MTIME_CHANGED; | |
158 | if (trust_ctime && check_stat && | |
159 | sd->sd_ctime.sec != (unsigned int)st->st_ctime) | |
160 | changed |= CTIME_CHANGED; | |
161 | ||
162 | #ifdef USE_NSEC | |
163 | if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st)) | |
164 | changed |= MTIME_CHANGED; | |
165 | if (trust_ctime && check_stat && | |
166 | sd->sd_ctime.nsec != ST_CTIME_NSEC(*st)) | |
167 | changed |= CTIME_CHANGED; | |
168 | #endif | |
169 | ||
170 | if (check_stat) { | |
171 | if (sd->sd_uid != (unsigned int) st->st_uid || | |
172 | sd->sd_gid != (unsigned int) st->st_gid) | |
173 | changed |= OWNER_CHANGED; | |
174 | if (sd->sd_ino != (unsigned int) st->st_ino) | |
175 | changed |= INODE_CHANGED; | |
176 | } | |
177 | ||
178 | #ifdef USE_STDEV | |
179 | /* | |
180 | * st_dev breaks on network filesystems where different | |
181 | * clients will have different views of what "device" | |
182 | * the filesystem is on | |
183 | */ | |
184 | if (check_stat && sd->sd_dev != (unsigned int) st->st_dev) | |
185 | changed |= INODE_CHANGED; | |
186 | #endif | |
187 | ||
188 | if (sd->sd_size != (unsigned int) st->st_size) | |
189 | changed |= DATA_CHANGED; | |
190 | ||
191 | return changed; | |
192 | } | |
193 | ||
194 | /* | |
195 | * This only updates the "non-critical" parts of the directory | |
196 | * cache, ie the parts that aren't tracked by GIT, and only used | |
197 | * to validate the cache. | |
198 | */ | |
199 | void fill_stat_cache_info(struct cache_entry *ce, struct stat *st) | |
200 | { | |
201 | fill_stat_data(&ce->ce_stat_data, st); | |
202 | ||
203 | if (assume_unchanged) | |
204 | ce->ce_flags |= CE_VALID; | |
205 | ||
206 | if (S_ISREG(st->st_mode)) { | |
207 | ce_mark_uptodate(ce); | |
208 | mark_fsmonitor_valid(ce); | |
209 | } | |
210 | } | |
211 | ||
212 | static int ce_compare_data(struct index_state *istate, | |
213 | const struct cache_entry *ce, | |
214 | struct stat *st) | |
215 | { | |
216 | int match = -1; | |
217 | int fd = git_open_cloexec(ce->name, O_RDONLY); | |
218 | ||
219 | if (fd >= 0) { | |
220 | struct object_id oid; | |
221 | if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0)) | |
222 | match = !oideq(&oid, &ce->oid); | |
223 | /* index_fd() closed the file descriptor already */ | |
224 | } | |
225 | return match; | |
226 | } | |
227 | ||
228 | static int ce_compare_link(const struct cache_entry *ce, size_t expected_size) | |
229 | { | |
230 | int match = -1; | |
231 | void *buffer; | |
232 | unsigned long size; | |
233 | enum object_type type; | |
234 | struct strbuf sb = STRBUF_INIT; | |
235 | ||
236 | if (strbuf_readlink(&sb, ce->name, expected_size)) | |
237 | return -1; | |
238 | ||
239 | buffer = read_object_file(&ce->oid, &type, &size); | |
240 | if (buffer) { | |
241 | if (size == sb.len) | |
242 | match = memcmp(buffer, sb.buf, size); | |
243 | free(buffer); | |
244 | } | |
245 | strbuf_release(&sb); | |
246 | return match; | |
247 | } | |
248 | ||
249 | static int ce_compare_gitlink(const struct cache_entry *ce) | |
250 | { | |
251 | struct object_id oid; | |
252 | ||
253 | /* | |
254 | * We don't actually require that the .git directory | |
255 | * under GITLINK directory be a valid git directory. It | |
256 | * might even be missing (in case nobody populated that | |
257 | * sub-project). | |
258 | * | |
259 | * If so, we consider it always to match. | |
260 | */ | |
261 | if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0) | |
262 | return 0; | |
263 | return !oideq(&oid, &ce->oid); | |
264 | } | |
265 | ||
266 | static int ce_modified_check_fs(struct index_state *istate, | |
267 | const struct cache_entry *ce, | |
268 | struct stat *st) | |
269 | { | |
270 | switch (st->st_mode & S_IFMT) { | |
271 | case S_IFREG: | |
272 | if (ce_compare_data(istate, ce, st)) | |
273 | return DATA_CHANGED; | |
274 | break; | |
275 | case S_IFLNK: | |
276 | if (ce_compare_link(ce, xsize_t(st->st_size))) | |
277 | return DATA_CHANGED; | |
278 | break; | |
279 | case S_IFDIR: | |
280 | if (S_ISGITLINK(ce->ce_mode)) | |
281 | return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; | |
282 | /* else fallthrough */ | |
283 | default: | |
284 | return TYPE_CHANGED; | |
285 | } | |
286 | return 0; | |
287 | } | |
288 | ||
289 | static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st) | |
290 | { | |
291 | unsigned int changed = 0; | |
292 | ||
293 | if (ce->ce_flags & CE_REMOVE) | |
294 | return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; | |
295 | ||
296 | switch (ce->ce_mode & S_IFMT) { | |
297 | case S_IFREG: | |
298 | changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; | |
299 | /* We consider only the owner x bit to be relevant for | |
300 | * "mode changes" | |
301 | */ | |
302 | if (trust_executable_bit && | |
303 | (0100 & (ce->ce_mode ^ st->st_mode))) | |
304 | changed |= MODE_CHANGED; | |
305 | break; | |
306 | case S_IFLNK: | |
307 | if (!S_ISLNK(st->st_mode) && | |
308 | (has_symlinks || !S_ISREG(st->st_mode))) | |
309 | changed |= TYPE_CHANGED; | |
310 | break; | |
311 | case S_IFGITLINK: | |
312 | /* We ignore most of the st_xxx fields for gitlinks */ | |
313 | if (!S_ISDIR(st->st_mode)) | |
314 | changed |= TYPE_CHANGED; | |
315 | else if (ce_compare_gitlink(ce)) | |
316 | changed |= DATA_CHANGED; | |
317 | return changed; | |
318 | default: | |
319 | die("internal error: ce_mode is %o", ce->ce_mode); | |
320 | } | |
321 | ||
322 | changed |= match_stat_data(&ce->ce_stat_data, st); | |
323 | ||
324 | /* Racily smudged entry? */ | |
325 | if (!ce->ce_stat_data.sd_size) { | |
326 | if (!is_empty_blob_sha1(ce->oid.hash)) | |
327 | changed |= DATA_CHANGED; | |
328 | } | |
329 | ||
330 | return changed; | |
331 | } | |
332 | ||
333 | static int is_racy_stat(const struct index_state *istate, | |
334 | const struct stat_data *sd) | |
335 | { | |
336 | return (istate->timestamp.sec && | |
337 | #ifdef USE_NSEC | |
338 | /* nanosecond timestamped files can also be racy! */ | |
339 | (istate->timestamp.sec < sd->sd_mtime.sec || | |
340 | (istate->timestamp.sec == sd->sd_mtime.sec && | |
341 | istate->timestamp.nsec <= sd->sd_mtime.nsec)) | |
342 | #else | |
343 | istate->timestamp.sec <= sd->sd_mtime.sec | |
344 | #endif | |
345 | ); | |
346 | } | |
347 | ||
348 | int is_racy_timestamp(const struct index_state *istate, | |
349 | const struct cache_entry *ce) | |
350 | { | |
351 | return (!S_ISGITLINK(ce->ce_mode) && | |
352 | is_racy_stat(istate, &ce->ce_stat_data)); | |
353 | } | |
354 | ||
355 | int match_stat_data_racy(const struct index_state *istate, | |
356 | const struct stat_data *sd, struct stat *st) | |
357 | { | |
358 | if (is_racy_stat(istate, sd)) | |
359 | return MTIME_CHANGED; | |
360 | return match_stat_data(sd, st); | |
361 | } | |
362 | ||
363 | int ie_match_stat(struct index_state *istate, | |
364 | const struct cache_entry *ce, struct stat *st, | |
365 | unsigned int options) | |
366 | { | |
367 | unsigned int changed; | |
368 | int ignore_valid = options & CE_MATCH_IGNORE_VALID; | |
369 | int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; | |
370 | int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; | |
371 | int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR; | |
372 | ||
373 | if (!ignore_fsmonitor) | |
374 | refresh_fsmonitor(istate); | |
375 | /* | |
376 | * If it's marked as always valid in the index, it's | |
377 | * valid whatever the checked-out copy says. | |
378 | * | |
379 | * skip-worktree has the same effect with higher precedence | |
380 | */ | |
381 | if (!ignore_skip_worktree && ce_skip_worktree(ce)) | |
382 | return 0; | |
383 | if (!ignore_valid && (ce->ce_flags & CE_VALID)) | |
384 | return 0; | |
385 | if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) | |
386 | return 0; | |
387 | ||
388 | /* | |
389 | * Intent-to-add entries have not been added, so the index entry | |
390 | * by definition never matches what is in the work tree until it | |
391 | * actually gets added. | |
392 | */ | |
393 | if (ce_intent_to_add(ce)) | |
394 | return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; | |
395 | ||
396 | changed = ce_match_stat_basic(ce, st); | |
397 | ||
398 | /* | |
399 | * Within 1 second of this sequence: | |
400 | * echo xyzzy >file && git-update-index --add file | |
401 | * running this command: | |
402 | * echo frotz >file | |
403 | * would give a falsely clean cache entry. The mtime and | |
404 | * length match the cache, and other stat fields do not change. | |
405 | * | |
406 | * We could detect this at update-index time (the cache entry | |
407 | * being registered/updated records the same time as "now") | |
408 | * and delay the return from git-update-index, but that would | |
409 | * effectively mean we can make at most one commit per second, | |
410 | * which is not acceptable. Instead, we check cache entries | |
411 | * whose mtime are the same as the index file timestamp more | |
412 | * carefully than others. | |
413 | */ | |
414 | if (!changed && is_racy_timestamp(istate, ce)) { | |
415 | if (assume_racy_is_modified) | |
416 | changed |= DATA_CHANGED; | |
417 | else | |
418 | changed |= ce_modified_check_fs(istate, ce, st); | |
419 | } | |
420 | ||
421 | return changed; | |
422 | } | |
423 | ||
424 | int ie_modified(struct index_state *istate, | |
425 | const struct cache_entry *ce, | |
426 | struct stat *st, unsigned int options) | |
427 | { | |
428 | int changed, changed_fs; | |
429 | ||
430 | changed = ie_match_stat(istate, ce, st, options); | |
431 | if (!changed) | |
432 | return 0; | |
433 | /* | |
434 | * If the mode or type has changed, there's no point in trying | |
435 | * to refresh the entry - it's not going to match | |
436 | */ | |
437 | if (changed & (MODE_CHANGED | TYPE_CHANGED)) | |
438 | return changed; | |
439 | ||
440 | /* | |
441 | * Immediately after read-tree or update-index --cacheinfo, | |
442 | * the length field is zero, as we have never even read the | |
443 | * lstat(2) information once, and we cannot trust DATA_CHANGED | |
444 | * returned by ie_match_stat() which in turn was returned by | |
445 | * ce_match_stat_basic() to signal that the filesize of the | |
446 | * blob changed. We have to actually go to the filesystem to | |
447 | * see if the contents match, and if so, should answer "unchanged". | |
448 | * | |
449 | * The logic does not apply to gitlinks, as ce_match_stat_basic() | |
450 | * already has checked the actual HEAD from the filesystem in the | |
451 | * subproject. If ie_match_stat() already said it is different, | |
452 | * then we know it is. | |
453 | */ | |
454 | if ((changed & DATA_CHANGED) && | |
455 | (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0)) | |
456 | return changed; | |
457 | ||
458 | changed_fs = ce_modified_check_fs(istate, ce, st); | |
459 | if (changed_fs) | |
460 | return changed | changed_fs; | |
461 | return 0; | |
462 | } | |
463 | ||
464 | int base_name_compare(const char *name1, int len1, int mode1, | |
465 | const char *name2, int len2, int mode2) | |
466 | { | |
467 | unsigned char c1, c2; | |
468 | int len = len1 < len2 ? len1 : len2; | |
469 | int cmp; | |
470 | ||
471 | cmp = memcmp(name1, name2, len); | |
472 | if (cmp) | |
473 | return cmp; | |
474 | c1 = name1[len]; | |
475 | c2 = name2[len]; | |
476 | if (!c1 && S_ISDIR(mode1)) | |
477 | c1 = '/'; | |
478 | if (!c2 && S_ISDIR(mode2)) | |
479 | c2 = '/'; | |
480 | return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; | |
481 | } | |
482 | ||
483 | /* | |
484 | * df_name_compare() is identical to base_name_compare(), except it | |
485 | * compares conflicting directory/file entries as equal. Note that | |
486 | * while a directory name compares as equal to a regular file, they | |
487 | * then individually compare _differently_ to a filename that has | |
488 | * a dot after the basename (because '\0' < '.' < '/'). | |
489 | * | |
490 | * This is used by routines that want to traverse the git namespace | |
491 | * but then handle conflicting entries together when possible. | |
492 | */ | |
493 | int df_name_compare(const char *name1, int len1, int mode1, | |
494 | const char *name2, int len2, int mode2) | |
495 | { | |
496 | int len = len1 < len2 ? len1 : len2, cmp; | |
497 | unsigned char c1, c2; | |
498 | ||
499 | cmp = memcmp(name1, name2, len); | |
500 | if (cmp) | |
501 | return cmp; | |
502 | /* Directories and files compare equal (same length, same name) */ | |
503 | if (len1 == len2) | |
504 | return 0; | |
505 | c1 = name1[len]; | |
506 | if (!c1 && S_ISDIR(mode1)) | |
507 | c1 = '/'; | |
508 | c2 = name2[len]; | |
509 | if (!c2 && S_ISDIR(mode2)) | |
510 | c2 = '/'; | |
511 | if (c1 == '/' && !c2) | |
512 | return 0; | |
513 | if (c2 == '/' && !c1) | |
514 | return 0; | |
515 | return c1 - c2; | |
516 | } | |
517 | ||
518 | int name_compare(const char *name1, size_t len1, const char *name2, size_t len2) | |
519 | { | |
520 | size_t min_len = (len1 < len2) ? len1 : len2; | |
521 | int cmp = memcmp(name1, name2, min_len); | |
522 | if (cmp) | |
523 | return cmp; | |
524 | if (len1 < len2) | |
525 | return -1; | |
526 | if (len1 > len2) | |
527 | return 1; | |
528 | return 0; | |
529 | } | |
530 | ||
531 | int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2) | |
532 | { | |
533 | int cmp; | |
534 | ||
535 | cmp = name_compare(name1, len1, name2, len2); | |
536 | if (cmp) | |
537 | return cmp; | |
538 | ||
539 | if (stage1 < stage2) | |
540 | return -1; | |
541 | if (stage1 > stage2) | |
542 | return 1; | |
543 | return 0; | |
544 | } | |
545 | ||
546 | static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage) | |
547 | { | |
548 | int first, last; | |
549 | ||
550 | first = 0; | |
551 | last = istate->cache_nr; | |
552 | while (last > first) { | |
553 | int next = (last + first) >> 1; | |
554 | struct cache_entry *ce = istate->cache[next]; | |
555 | int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce)); | |
556 | if (!cmp) | |
557 | return next; | |
558 | if (cmp < 0) { | |
559 | last = next; | |
560 | continue; | |
561 | } | |
562 | first = next+1; | |
563 | } | |
564 | return -first-1; | |
565 | } | |
566 | ||
567 | int index_name_pos(const struct index_state *istate, const char *name, int namelen) | |
568 | { | |
569 | return index_name_stage_pos(istate, name, namelen, 0); | |
570 | } | |
571 | ||
572 | int remove_index_entry_at(struct index_state *istate, int pos) | |
573 | { | |
574 | struct cache_entry *ce = istate->cache[pos]; | |
575 | ||
576 | record_resolve_undo(istate, ce); | |
577 | remove_name_hash(istate, ce); | |
578 | save_or_free_index_entry(istate, ce); | |
579 | istate->cache_changed |= CE_ENTRY_REMOVED; | |
580 | istate->cache_nr--; | |
581 | if (pos >= istate->cache_nr) | |
582 | return 0; | |
583 | MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1, | |
584 | istate->cache_nr - pos); | |
585 | return 1; | |
586 | } | |
587 | ||
588 | /* | |
589 | * Remove all cache entries marked for removal, that is where | |
590 | * CE_REMOVE is set in ce_flags. This is much more effective than | |
591 | * calling remove_index_entry_at() for each entry to be removed. | |
592 | */ | |
593 | void remove_marked_cache_entries(struct index_state *istate) | |
594 | { | |
595 | struct cache_entry **ce_array = istate->cache; | |
596 | unsigned int i, j; | |
597 | ||
598 | for (i = j = 0; i < istate->cache_nr; i++) { | |
599 | if (ce_array[i]->ce_flags & CE_REMOVE) { | |
600 | remove_name_hash(istate, ce_array[i]); | |
601 | save_or_free_index_entry(istate, ce_array[i]); | |
602 | } | |
603 | else | |
604 | ce_array[j++] = ce_array[i]; | |
605 | } | |
606 | if (j == istate->cache_nr) | |
607 | return; | |
608 | istate->cache_changed |= CE_ENTRY_REMOVED; | |
609 | istate->cache_nr = j; | |
610 | } | |
611 | ||
612 | int remove_file_from_index(struct index_state *istate, const char *path) | |
613 | { | |
614 | int pos = index_name_pos(istate, path, strlen(path)); | |
615 | if (pos < 0) | |
616 | pos = -pos-1; | |
617 | cache_tree_invalidate_path(istate, path); | |
618 | untracked_cache_remove_from_index(istate, path); | |
619 | while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) | |
620 | remove_index_entry_at(istate, pos); | |
621 | return 0; | |
622 | } | |
623 | ||
624 | static int compare_name(struct cache_entry *ce, const char *path, int namelen) | |
625 | { | |
626 | return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); | |
627 | } | |
628 | ||
629 | static int index_name_pos_also_unmerged(struct index_state *istate, | |
630 | const char *path, int namelen) | |
631 | { | |
632 | int pos = index_name_pos(istate, path, namelen); | |
633 | struct cache_entry *ce; | |
634 | ||
635 | if (pos >= 0) | |
636 | return pos; | |
637 | ||
638 | /* maybe unmerged? */ | |
639 | pos = -1 - pos; | |
640 | if (pos >= istate->cache_nr || | |
641 | compare_name((ce = istate->cache[pos]), path, namelen)) | |
642 | return -1; | |
643 | ||
644 | /* order of preference: stage 2, 1, 3 */ | |
645 | if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && | |
646 | ce_stage((ce = istate->cache[pos + 1])) == 2 && | |
647 | !compare_name(ce, path, namelen)) | |
648 | pos++; | |
649 | return pos; | |
650 | } | |
651 | ||
652 | static int different_name(struct cache_entry *ce, struct cache_entry *alias) | |
653 | { | |
654 | int len = ce_namelen(ce); | |
655 | return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); | |
656 | } | |
657 | ||
658 | /* | |
659 | * If we add a filename that aliases in the cache, we will use the | |
660 | * name that we already have - but we don't want to update the same | |
661 | * alias twice, because that implies that there were actually two | |
662 | * different files with aliasing names! | |
663 | * | |
664 | * So we use the CE_ADDED flag to verify that the alias was an old | |
665 | * one before we accept it as | |
666 | */ | |
667 | static struct cache_entry *create_alias_ce(struct index_state *istate, | |
668 | struct cache_entry *ce, | |
669 | struct cache_entry *alias) | |
670 | { | |
671 | int len; | |
672 | struct cache_entry *new_entry; | |
673 | ||
674 | if (alias->ce_flags & CE_ADDED) | |
675 | die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); | |
676 | ||
677 | /* Ok, create the new entry using the name of the existing alias */ | |
678 | len = ce_namelen(alias); | |
679 | new_entry = make_empty_cache_entry(istate, len); | |
680 | memcpy(new_entry->name, alias->name, len); | |
681 | copy_cache_entry(new_entry, ce); | |
682 | save_or_free_index_entry(istate, ce); | |
683 | return new_entry; | |
684 | } | |
685 | ||
686 | void set_object_name_for_intent_to_add_entry(struct cache_entry *ce) | |
687 | { | |
688 | struct object_id oid; | |
689 | if (write_object_file("", 0, blob_type, &oid)) | |
690 | die("cannot create an empty blob in the object database"); | |
691 | oidcpy(&ce->oid, &oid); | |
692 | } | |
693 | ||
694 | int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) | |
695 | { | |
696 | int namelen, was_same; | |
697 | mode_t st_mode = st->st_mode; | |
698 | struct cache_entry *ce, *alias = NULL; | |
699 | unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY; | |
700 | int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); | |
701 | int pretend = flags & ADD_CACHE_PRETEND; | |
702 | int intent_only = flags & ADD_CACHE_INTENT; | |
703 | int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| | |
704 | (intent_only ? ADD_CACHE_NEW_ONLY : 0)); | |
705 | int newflags = HASH_WRITE_OBJECT; | |
706 | ||
707 | if (flags & HASH_RENORMALIZE) | |
708 | newflags |= HASH_RENORMALIZE; | |
709 | ||
710 | if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) | |
711 | return error("%s: can only add regular files, symbolic links or git-directories", path); | |
712 | ||
713 | namelen = strlen(path); | |
714 | if (S_ISDIR(st_mode)) { | |
715 | while (namelen && path[namelen-1] == '/') | |
716 | namelen--; | |
717 | } | |
718 | ce = make_empty_cache_entry(istate, namelen); | |
719 | memcpy(ce->name, path, namelen); | |
720 | ce->ce_namelen = namelen; | |
721 | if (!intent_only) | |
722 | fill_stat_cache_info(ce, st); | |
723 | else | |
724 | ce->ce_flags |= CE_INTENT_TO_ADD; | |
725 | ||
726 | ||
727 | if (trust_executable_bit && has_symlinks) { | |
728 | ce->ce_mode = create_ce_mode(st_mode); | |
729 | } else { | |
730 | /* If there is an existing entry, pick the mode bits and type | |
731 | * from it, otherwise assume unexecutable regular file. | |
732 | */ | |
733 | struct cache_entry *ent; | |
734 | int pos = index_name_pos_also_unmerged(istate, path, namelen); | |
735 | ||
736 | ent = (0 <= pos) ? istate->cache[pos] : NULL; | |
737 | ce->ce_mode = ce_mode_from_stat(ent, st_mode); | |
738 | } | |
739 | ||
740 | /* When core.ignorecase=true, determine if a directory of the same name but differing | |
741 | * case already exists within the Git repository. If it does, ensure the directory | |
742 | * case of the file being added to the repository matches (is folded into) the existing | |
743 | * entry's directory case. | |
744 | */ | |
745 | if (ignore_case) { | |
746 | adjust_dirname_case(istate, ce->name); | |
747 | } | |
748 | if (!(flags & HASH_RENORMALIZE)) { | |
749 | alias = index_file_exists(istate, ce->name, | |
750 | ce_namelen(ce), ignore_case); | |
751 | if (alias && | |
752 | !ce_stage(alias) && | |
753 | !ie_match_stat(istate, alias, st, ce_option)) { | |
754 | /* Nothing changed, really */ | |
755 | if (!S_ISGITLINK(alias->ce_mode)) | |
756 | ce_mark_uptodate(alias); | |
757 | alias->ce_flags |= CE_ADDED; | |
758 | ||
759 | discard_cache_entry(ce); | |
760 | return 0; | |
761 | } | |
762 | } | |
763 | if (!intent_only) { | |
764 | if (index_path(istate, &ce->oid, path, st, newflags)) { | |
765 | discard_cache_entry(ce); | |
766 | return error("unable to index file %s", path); | |
767 | } | |
768 | } else | |
769 | set_object_name_for_intent_to_add_entry(ce); | |
770 | ||
771 | if (ignore_case && alias && different_name(ce, alias)) | |
772 | ce = create_alias_ce(istate, ce, alias); | |
773 | ce->ce_flags |= CE_ADDED; | |
774 | ||
775 | /* It was suspected to be racily clean, but it turns out to be Ok */ | |
776 | was_same = (alias && | |
777 | !ce_stage(alias) && | |
778 | oideq(&alias->oid, &ce->oid) && | |
779 | ce->ce_mode == alias->ce_mode); | |
780 | ||
781 | if (pretend) | |
782 | discard_cache_entry(ce); | |
783 | else if (add_index_entry(istate, ce, add_option)) { | |
784 | discard_cache_entry(ce); | |
785 | return error("unable to add %s to index", path); | |
786 | } | |
787 | if (verbose && !was_same) | |
788 | printf("add '%s'\n", path); | |
789 | return 0; | |
790 | } | |
791 | ||
792 | int add_file_to_index(struct index_state *istate, const char *path, int flags) | |
793 | { | |
794 | struct stat st; | |
795 | if (lstat(path, &st)) | |
796 | die_errno("unable to stat '%s'", path); | |
797 | return add_to_index(istate, path, &st, flags); | |
798 | } | |
799 | ||
800 | struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len) | |
801 | { | |
802 | return mem_pool__ce_calloc(find_mem_pool(istate), len); | |
803 | } | |
804 | ||
805 | struct cache_entry *make_empty_transient_cache_entry(size_t len) | |
806 | { | |
807 | return xcalloc(1, cache_entry_size(len)); | |
808 | } | |
809 | ||
810 | struct cache_entry *make_cache_entry(struct index_state *istate, | |
811 | unsigned int mode, | |
812 | const struct object_id *oid, | |
813 | const char *path, | |
814 | int stage, | |
815 | unsigned int refresh_options) | |
816 | { | |
817 | struct cache_entry *ce, *ret; | |
818 | int len; | |
819 | ||
820 | if (!verify_path(path, mode)) { | |
821 | error("Invalid path '%s'", path); | |
822 | return NULL; | |
823 | } | |
824 | ||
825 | len = strlen(path); | |
826 | ce = make_empty_cache_entry(istate, len); | |
827 | ||
828 | oidcpy(&ce->oid, oid); | |
829 | memcpy(ce->name, path, len); | |
830 | ce->ce_flags = create_ce_flags(stage); | |
831 | ce->ce_namelen = len; | |
832 | ce->ce_mode = create_ce_mode(mode); | |
833 | ||
834 | ret = refresh_cache_entry(istate, ce, refresh_options); | |
835 | if (ret != ce) | |
836 | discard_cache_entry(ce); | |
837 | return ret; | |
838 | } | |
839 | ||
840 | struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid, | |
841 | const char *path, int stage) | |
842 | { | |
843 | struct cache_entry *ce; | |
844 | int len; | |
845 | ||
846 | if (!verify_path(path, mode)) { | |
847 | error("Invalid path '%s'", path); | |
848 | return NULL; | |
849 | } | |
850 | ||
851 | len = strlen(path); | |
852 | ce = make_empty_transient_cache_entry(len); | |
853 | ||
854 | oidcpy(&ce->oid, oid); | |
855 | memcpy(ce->name, path, len); | |
856 | ce->ce_flags = create_ce_flags(stage); | |
857 | ce->ce_namelen = len; | |
858 | ce->ce_mode = create_ce_mode(mode); | |
859 | ||
860 | return ce; | |
861 | } | |
862 | ||
863 | /* | |
864 | * Chmod an index entry with either +x or -x. | |
865 | * | |
866 | * Returns -1 if the chmod for the particular cache entry failed (if it's | |
867 | * not a regular file), -2 if an invalid flip argument is passed in, 0 | |
868 | * otherwise. | |
869 | */ | |
870 | int chmod_index_entry(struct index_state *istate, struct cache_entry *ce, | |
871 | char flip) | |
872 | { | |
873 | if (!S_ISREG(ce->ce_mode)) | |
874 | return -1; | |
875 | switch (flip) { | |
876 | case '+': | |
877 | ce->ce_mode |= 0111; | |
878 | break; | |
879 | case '-': | |
880 | ce->ce_mode &= ~0111; | |
881 | break; | |
882 | default: | |
883 | return -2; | |
884 | } | |
885 | cache_tree_invalidate_path(istate, ce->name); | |
886 | ce->ce_flags |= CE_UPDATE_IN_BASE; | |
887 | mark_fsmonitor_invalid(istate, ce); | |
888 | istate->cache_changed |= CE_ENTRY_CHANGED; | |
889 | ||
890 | return 0; | |
891 | } | |
892 | ||
893 | int ce_same_name(const struct cache_entry *a, const struct cache_entry *b) | |
894 | { | |
895 | int len = ce_namelen(a); | |
896 | return ce_namelen(b) == len && !memcmp(a->name, b->name, len); | |
897 | } | |
898 | ||
899 | /* | |
900 | * We fundamentally don't like some paths: we don't want | |
901 | * dot or dot-dot anywhere, and for obvious reasons don't | |
902 | * want to recurse into ".git" either. | |
903 | * | |
904 | * Also, we don't want double slashes or slashes at the | |
905 | * end that can make pathnames ambiguous. | |
906 | */ | |
907 | static int verify_dotfile(const char *rest, unsigned mode) | |
908 | { | |
909 | /* | |
910 | * The first character was '.', but that | |
911 | * has already been discarded, we now test | |
912 | * the rest. | |
913 | */ | |
914 | ||
915 | /* "." is not allowed */ | |
916 | if (*rest == '\0' || is_dir_sep(*rest)) | |
917 | return 0; | |
918 | ||
919 | switch (*rest) { | |
920 | /* | |
921 | * ".git" followed by NUL or slash is bad. Note that we match | |
922 | * case-insensitively here, even if ignore_case is not set. | |
923 | * This outlaws ".GIT" everywhere out of an abundance of caution, | |
924 | * since there's really no good reason to allow it. | |
925 | * | |
926 | * Once we've seen ".git", we can also find ".gitmodules", etc (also | |
927 | * case-insensitively). | |
928 | */ | |
929 | case 'g': | |
930 | case 'G': | |
931 | if (rest[1] != 'i' && rest[1] != 'I') | |
932 | break; | |
933 | if (rest[2] != 't' && rest[2] != 'T') | |
934 | break; | |
935 | if (rest[3] == '\0' || is_dir_sep(rest[3])) | |
936 | return 0; | |
937 | if (S_ISLNK(mode)) { | |
938 | rest += 3; | |
939 | if (skip_iprefix(rest, "modules", &rest) && | |
940 | (*rest == '\0' || is_dir_sep(*rest))) | |
941 | return 0; | |
942 | } | |
943 | break; | |
944 | case '.': | |
945 | if (rest[1] == '\0' || is_dir_sep(rest[1])) | |
946 | return 0; | |
947 | } | |
948 | return 1; | |
949 | } | |
950 | ||
951 | int verify_path(const char *path, unsigned mode) | |
952 | { | |
953 | char c; | |
954 | ||
955 | if (has_dos_drive_prefix(path)) | |
956 | return 0; | |
957 | ||
958 | goto inside; | |
959 | for (;;) { | |
960 | if (!c) | |
961 | return 1; | |
962 | if (is_dir_sep(c)) { | |
963 | inside: | |
964 | if (protect_hfs) { | |
965 | if (is_hfs_dotgit(path)) | |
966 | return 0; | |
967 | if (S_ISLNK(mode)) { | |
968 | if (is_hfs_dotgitmodules(path)) | |
969 | return 0; | |
970 | } | |
971 | } | |
972 | if (protect_ntfs) { | |
973 | if (is_ntfs_dotgit(path)) | |
974 | return 0; | |
975 | if (S_ISLNK(mode)) { | |
976 | if (is_ntfs_dotgitmodules(path)) | |
977 | return 0; | |
978 | } | |
979 | } | |
980 | ||
981 | c = *path++; | |
982 | if ((c == '.' && !verify_dotfile(path, mode)) || | |
983 | is_dir_sep(c) || c == '\0') | |
984 | return 0; | |
985 | } | |
986 | c = *path++; | |
987 | } | |
988 | } | |
989 | ||
990 | /* | |
991 | * Do we have another file that has the beginning components being a | |
992 | * proper superset of the name we're trying to add? | |
993 | */ | |
994 | static int has_file_name(struct index_state *istate, | |
995 | const struct cache_entry *ce, int pos, int ok_to_replace) | |
996 | { | |
997 | int retval = 0; | |
998 | int len = ce_namelen(ce); | |
999 | int stage = ce_stage(ce); | |
1000 | const char *name = ce->name; | |
1001 | ||
1002 | while (pos < istate->cache_nr) { | |
1003 | struct cache_entry *p = istate->cache[pos++]; | |
1004 | ||
1005 | if (len >= ce_namelen(p)) | |
1006 | break; | |
1007 | if (memcmp(name, p->name, len)) | |
1008 | break; | |
1009 | if (ce_stage(p) != stage) | |
1010 | continue; | |
1011 | if (p->name[len] != '/') | |
1012 | continue; | |
1013 | if (p->ce_flags & CE_REMOVE) | |
1014 | continue; | |
1015 | retval = -1; | |
1016 | if (!ok_to_replace) | |
1017 | break; | |
1018 | remove_index_entry_at(istate, --pos); | |
1019 | } | |
1020 | return retval; | |
1021 | } | |
1022 | ||
1023 | ||
1024 | /* | |
1025 | * Like strcmp(), but also return the offset of the first change. | |
1026 | * If strings are equal, return the length. | |
1027 | */ | |
1028 | int strcmp_offset(const char *s1, const char *s2, size_t *first_change) | |
1029 | { | |
1030 | size_t k; | |
1031 | ||
1032 | if (!first_change) | |
1033 | return strcmp(s1, s2); | |
1034 | ||
1035 | for (k = 0; s1[k] == s2[k]; k++) | |
1036 | if (s1[k] == '\0') | |
1037 | break; | |
1038 | ||
1039 | *first_change = k; | |
1040 | return (unsigned char)s1[k] - (unsigned char)s2[k]; | |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * Do we have another file with a pathname that is a proper | |
1045 | * subset of the name we're trying to add? | |
1046 | * | |
1047 | * That is, is there another file in the index with a path | |
1048 | * that matches a sub-directory in the given entry? | |
1049 | */ | |
1050 | static int has_dir_name(struct index_state *istate, | |
1051 | const struct cache_entry *ce, int pos, int ok_to_replace) | |
1052 | { | |
1053 | int retval = 0; | |
1054 | int stage = ce_stage(ce); | |
1055 | const char *name = ce->name; | |
1056 | const char *slash = name + ce_namelen(ce); | |
1057 | size_t len_eq_last; | |
1058 | int cmp_last = 0; | |
1059 | ||
1060 | /* | |
1061 | * We are frequently called during an iteration on a sorted | |
1062 | * list of pathnames and while building a new index. Therefore, | |
1063 | * there is a high probability that this entry will eventually | |
1064 | * be appended to the index, rather than inserted in the middle. | |
1065 | * If we can confirm that, we can avoid binary searches on the | |
1066 | * components of the pathname. | |
1067 | * | |
1068 | * Compare the entry's full path with the last path in the index. | |
1069 | */ | |
1070 | if (istate->cache_nr > 0) { | |
1071 | cmp_last = strcmp_offset(name, | |
1072 | istate->cache[istate->cache_nr - 1]->name, | |
1073 | &len_eq_last); | |
1074 | if (cmp_last > 0) { | |
1075 | if (len_eq_last == 0) { | |
1076 | /* | |
1077 | * The entry sorts AFTER the last one in the | |
1078 | * index and their paths have no common prefix, | |
1079 | * so there cannot be a F/D conflict. | |
1080 | */ | |
1081 | return retval; | |
1082 | } else { | |
1083 | /* | |
1084 | * The entry sorts AFTER the last one in the | |
1085 | * index, but has a common prefix. Fall through | |
1086 | * to the loop below to disect the entry's path | |
1087 | * and see where the difference is. | |
1088 | */ | |
1089 | } | |
1090 | } else if (cmp_last == 0) { | |
1091 | /* | |
1092 | * The entry exactly matches the last one in the | |
1093 | * index, but because of multiple stage and CE_REMOVE | |
1094 | * items, we fall through and let the regular search | |
1095 | * code handle it. | |
1096 | */ | |
1097 | } | |
1098 | } | |
1099 | ||
1100 | for (;;) { | |
1101 | size_t len; | |
1102 | ||
1103 | for (;;) { | |
1104 | if (*--slash == '/') | |
1105 | break; | |
1106 | if (slash <= ce->name) | |
1107 | return retval; | |
1108 | } | |
1109 | len = slash - name; | |
1110 | ||
1111 | if (cmp_last > 0) { | |
1112 | /* | |
1113 | * (len + 1) is a directory boundary (including | |
1114 | * the trailing slash). And since the loop is | |
1115 | * decrementing "slash", the first iteration is | |
1116 | * the longest directory prefix; subsequent | |
1117 | * iterations consider parent directories. | |
1118 | */ | |
1119 | ||
1120 | if (len + 1 <= len_eq_last) { | |
1121 | /* | |
1122 | * The directory prefix (including the trailing | |
1123 | * slash) also appears as a prefix in the last | |
1124 | * entry, so the remainder cannot collide (because | |
1125 | * strcmp said the whole path was greater). | |
1126 | * | |
1127 | * EQ: last: xxx/A | |
1128 | * this: xxx/B | |
1129 | * | |
1130 | * LT: last: xxx/file_A | |
1131 | * this: xxx/file_B | |
1132 | */ | |
1133 | return retval; | |
1134 | } | |
1135 | ||
1136 | if (len > len_eq_last) { | |
1137 | /* | |
1138 | * This part of the directory prefix (excluding | |
1139 | * the trailing slash) is longer than the known | |
1140 | * equal portions, so this sub-directory cannot | |
1141 | * collide with a file. | |
1142 | * | |
1143 | * GT: last: xxxA | |
1144 | * this: xxxB/file | |
1145 | */ | |
1146 | return retval; | |
1147 | } | |
1148 | ||
1149 | if (istate->cache_nr > 0 && | |
1150 | ce_namelen(istate->cache[istate->cache_nr - 1]) > len) { | |
1151 | /* | |
1152 | * The directory prefix lines up with part of | |
1153 | * a longer file or directory name, but sorts | |
1154 | * after it, so this sub-directory cannot | |
1155 | * collide with a file. | |
1156 | * | |
1157 | * last: xxx/yy-file (because '-' sorts before '/') | |
1158 | * this: xxx/yy/abc | |
1159 | */ | |
1160 | return retval; | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * This is a possible collision. Fall through and | |
1165 | * let the regular search code handle it. | |
1166 | * | |
1167 | * last: xxx | |
1168 | * this: xxx/file | |
1169 | */ | |
1170 | } | |
1171 | ||
1172 | pos = index_name_stage_pos(istate, name, len, stage); | |
1173 | if (pos >= 0) { | |
1174 | /* | |
1175 | * Found one, but not so fast. This could | |
1176 | * be a marker that says "I was here, but | |
1177 | * I am being removed". Such an entry is | |
1178 | * not a part of the resulting tree, and | |
1179 | * it is Ok to have a directory at the same | |
1180 | * path. | |
1181 | */ | |
1182 | if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { | |
1183 | retval = -1; | |
1184 | if (!ok_to_replace) | |
1185 | break; | |
1186 | remove_index_entry_at(istate, pos); | |
1187 | continue; | |
1188 | } | |
1189 | } | |
1190 | else | |
1191 | pos = -pos-1; | |
1192 | ||
1193 | /* | |
1194 | * Trivial optimization: if we find an entry that | |
1195 | * already matches the sub-directory, then we know | |
1196 | * we're ok, and we can exit. | |
1197 | */ | |
1198 | while (pos < istate->cache_nr) { | |
1199 | struct cache_entry *p = istate->cache[pos]; | |
1200 | if ((ce_namelen(p) <= len) || | |
1201 | (p->name[len] != '/') || | |
1202 | memcmp(p->name, name, len)) | |
1203 | break; /* not our subdirectory */ | |
1204 | if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) | |
1205 | /* | |
1206 | * p is at the same stage as our entry, and | |
1207 | * is a subdirectory of what we are looking | |
1208 | * at, so we cannot have conflicts at our | |
1209 | * level or anything shorter. | |
1210 | */ | |
1211 | return retval; | |
1212 | pos++; | |
1213 | } | |
1214 | } | |
1215 | return retval; | |
1216 | } | |
1217 | ||
1218 | /* We may be in a situation where we already have path/file and path | |
1219 | * is being added, or we already have path and path/file is being | |
1220 | * added. Either one would result in a nonsense tree that has path | |
1221 | * twice when git-write-tree tries to write it out. Prevent it. | |
1222 | * | |
1223 | * If ok-to-replace is specified, we remove the conflicting entries | |
1224 | * from the cache so the caller should recompute the insert position. | |
1225 | * When this happens, we return non-zero. | |
1226 | */ | |
1227 | static int check_file_directory_conflict(struct index_state *istate, | |
1228 | const struct cache_entry *ce, | |
1229 | int pos, int ok_to_replace) | |
1230 | { | |
1231 | int retval; | |
1232 | ||
1233 | /* | |
1234 | * When ce is an "I am going away" entry, we allow it to be added | |
1235 | */ | |
1236 | if (ce->ce_flags & CE_REMOVE) | |
1237 | return 0; | |
1238 | ||
1239 | /* | |
1240 | * We check if the path is a sub-path of a subsequent pathname | |
1241 | * first, since removing those will not change the position | |
1242 | * in the array. | |
1243 | */ | |
1244 | retval = has_file_name(istate, ce, pos, ok_to_replace); | |
1245 | ||
1246 | /* | |
1247 | * Then check if the path might have a clashing sub-directory | |
1248 | * before it. | |
1249 | */ | |
1250 | return retval + has_dir_name(istate, ce, pos, ok_to_replace); | |
1251 | } | |
1252 | ||
1253 | static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) | |
1254 | { | |
1255 | int pos; | |
1256 | int ok_to_add = option & ADD_CACHE_OK_TO_ADD; | |
1257 | int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; | |
1258 | int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; | |
1259 | int new_only = option & ADD_CACHE_NEW_ONLY; | |
1260 | ||
1261 | if (!(option & ADD_CACHE_KEEP_CACHE_TREE)) | |
1262 | cache_tree_invalidate_path(istate, ce->name); | |
1263 | ||
1264 | /* | |
1265 | * If this entry's path sorts after the last entry in the index, | |
1266 | * we can avoid searching for it. | |
1267 | */ | |
1268 | if (istate->cache_nr > 0 && | |
1269 | strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0) | |
1270 | pos = -istate->cache_nr - 1; | |
1271 | else | |
1272 | pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce)); | |
1273 | ||
1274 | /* existing match? Just replace it. */ | |
1275 | if (pos >= 0) { | |
1276 | if (!new_only) | |
1277 | replace_index_entry(istate, pos, ce); | |
1278 | return 0; | |
1279 | } | |
1280 | pos = -pos-1; | |
1281 | ||
1282 | if (!(option & ADD_CACHE_KEEP_CACHE_TREE)) | |
1283 | untracked_cache_add_to_index(istate, ce->name); | |
1284 | ||
1285 | /* | |
1286 | * Inserting a merged entry ("stage 0") into the index | |
1287 | * will always replace all non-merged entries.. | |
1288 | */ | |
1289 | if (pos < istate->cache_nr && ce_stage(ce) == 0) { | |
1290 | while (ce_same_name(istate->cache[pos], ce)) { | |
1291 | ok_to_add = 1; | |
1292 | if (!remove_index_entry_at(istate, pos)) | |
1293 | break; | |
1294 | } | |
1295 | } | |
1296 | ||
1297 | if (!ok_to_add) | |
1298 | return -1; | |
1299 | if (!verify_path(ce->name, ce->ce_mode)) | |
1300 | return error("Invalid path '%s'", ce->name); | |
1301 | ||
1302 | if (!skip_df_check && | |
1303 | check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { | |
1304 | if (!ok_to_replace) | |
1305 | return error("'%s' appears as both a file and as a directory", | |
1306 | ce->name); | |
1307 | pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce)); | |
1308 | pos = -pos-1; | |
1309 | } | |
1310 | return pos + 1; | |
1311 | } | |
1312 | ||
1313 | int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) | |
1314 | { | |
1315 | int pos; | |
1316 | ||
1317 | if (option & ADD_CACHE_JUST_APPEND) | |
1318 | pos = istate->cache_nr; | |
1319 | else { | |
1320 | int ret; | |
1321 | ret = add_index_entry_with_check(istate, ce, option); | |
1322 | if (ret <= 0) | |
1323 | return ret; | |
1324 | pos = ret - 1; | |
1325 | } | |
1326 | ||
1327 | /* Make sure the array is big enough .. */ | |
1328 | ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc); | |
1329 | ||
1330 | /* Add it in.. */ | |
1331 | istate->cache_nr++; | |
1332 | if (istate->cache_nr > pos + 1) | |
1333 | MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos, | |
1334 | istate->cache_nr - pos - 1); | |
1335 | set_index_entry(istate, pos, ce); | |
1336 | istate->cache_changed |= CE_ENTRY_ADDED; | |
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * "refresh" does not calculate a new sha1 file or bring the | |
1342 | * cache up-to-date for mode/content changes. But what it | |
1343 | * _does_ do is to "re-match" the stat information of a file | |
1344 | * with the cache, so that you can refresh the cache for a | |
1345 | * file that hasn't been changed but where the stat entry is | |
1346 | * out of date. | |
1347 | * | |
1348 | * For example, you'd want to do this after doing a "git-read-tree", | |
1349 | * to link up the stat cache details with the proper files. | |
1350 | */ | |
1351 | static struct cache_entry *refresh_cache_ent(struct index_state *istate, | |
1352 | struct cache_entry *ce, | |
1353 | unsigned int options, int *err, | |
1354 | int *changed_ret) | |
1355 | { | |
1356 | struct stat st; | |
1357 | struct cache_entry *updated; | |
1358 | int changed; | |
1359 | int refresh = options & CE_MATCH_REFRESH; | |
1360 | int ignore_valid = options & CE_MATCH_IGNORE_VALID; | |
1361 | int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; | |
1362 | int ignore_missing = options & CE_MATCH_IGNORE_MISSING; | |
1363 | int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR; | |
1364 | ||
1365 | if (!refresh || ce_uptodate(ce)) | |
1366 | return ce; | |
1367 | ||
1368 | if (!ignore_fsmonitor) | |
1369 | refresh_fsmonitor(istate); | |
1370 | /* | |
1371 | * CE_VALID or CE_SKIP_WORKTREE means the user promised us | |
1372 | * that the change to the work tree does not matter and told | |
1373 | * us not to worry. | |
1374 | */ | |
1375 | if (!ignore_skip_worktree && ce_skip_worktree(ce)) { | |
1376 | ce_mark_uptodate(ce); | |
1377 | return ce; | |
1378 | } | |
1379 | if (!ignore_valid && (ce->ce_flags & CE_VALID)) { | |
1380 | ce_mark_uptodate(ce); | |
1381 | return ce; | |
1382 | } | |
1383 | if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) { | |
1384 | ce_mark_uptodate(ce); | |
1385 | return ce; | |
1386 | } | |
1387 | ||
1388 | if (has_symlink_leading_path(ce->name, ce_namelen(ce))) { | |
1389 | if (ignore_missing) | |
1390 | return ce; | |
1391 | if (err) | |
1392 | *err = ENOENT; | |
1393 | return NULL; | |
1394 | } | |
1395 | ||
1396 | if (lstat(ce->name, &st) < 0) { | |
1397 | if (ignore_missing && errno == ENOENT) | |
1398 | return ce; | |
1399 | if (err) | |
1400 | *err = errno; | |
1401 | return NULL; | |
1402 | } | |
1403 | ||
1404 | changed = ie_match_stat(istate, ce, &st, options); | |
1405 | if (changed_ret) | |
1406 | *changed_ret = changed; | |
1407 | if (!changed) { | |
1408 | /* | |
1409 | * The path is unchanged. If we were told to ignore | |
1410 | * valid bit, then we did the actual stat check and | |
1411 | * found that the entry is unmodified. If the entry | |
1412 | * is not marked VALID, this is the place to mark it | |
1413 | * valid again, under "assume unchanged" mode. | |
1414 | */ | |
1415 | if (ignore_valid && assume_unchanged && | |
1416 | !(ce->ce_flags & CE_VALID)) | |
1417 | ; /* mark this one VALID again */ | |
1418 | else { | |
1419 | /* | |
1420 | * We do not mark the index itself "modified" | |
1421 | * because CE_UPTODATE flag is in-core only; | |
1422 | * we are not going to write this change out. | |
1423 | */ | |
1424 | if (!S_ISGITLINK(ce->ce_mode)) { | |
1425 | ce_mark_uptodate(ce); | |
1426 | mark_fsmonitor_valid(ce); | |
1427 | } | |
1428 | return ce; | |
1429 | } | |
1430 | } | |
1431 | ||
1432 | if (ie_modified(istate, ce, &st, options)) { | |
1433 | if (err) | |
1434 | *err = EINVAL; | |
1435 | return NULL; | |
1436 | } | |
1437 | ||
1438 | updated = make_empty_cache_entry(istate, ce_namelen(ce)); | |
1439 | copy_cache_entry(updated, ce); | |
1440 | memcpy(updated->name, ce->name, ce->ce_namelen + 1); | |
1441 | fill_stat_cache_info(updated, &st); | |
1442 | /* | |
1443 | * If ignore_valid is not set, we should leave CE_VALID bit | |
1444 | * alone. Otherwise, paths marked with --no-assume-unchanged | |
1445 | * (i.e. things to be edited) will reacquire CE_VALID bit | |
1446 | * automatically, which is not really what we want. | |
1447 | */ | |
1448 | if (!ignore_valid && assume_unchanged && | |
1449 | !(ce->ce_flags & CE_VALID)) | |
1450 | updated->ce_flags &= ~CE_VALID; | |
1451 | ||
1452 | /* istate->cache_changed is updated in the caller */ | |
1453 | return updated; | |
1454 | } | |
1455 | ||
1456 | static void show_file(const char * fmt, const char * name, int in_porcelain, | |
1457 | int * first, const char *header_msg) | |
1458 | { | |
1459 | if (in_porcelain && *first && header_msg) { | |
1460 | printf("%s\n", header_msg); | |
1461 | *first = 0; | |
1462 | } | |
1463 | printf(fmt, name); | |
1464 | } | |
1465 | ||
1466 | int refresh_index(struct index_state *istate, unsigned int flags, | |
1467 | const struct pathspec *pathspec, | |
1468 | char *seen, const char *header_msg) | |
1469 | { | |
1470 | int i; | |
1471 | int has_errors = 0; | |
1472 | int really = (flags & REFRESH_REALLY) != 0; | |
1473 | int allow_unmerged = (flags & REFRESH_UNMERGED) != 0; | |
1474 | int quiet = (flags & REFRESH_QUIET) != 0; | |
1475 | int not_new = (flags & REFRESH_IGNORE_MISSING) != 0; | |
1476 | int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0; | |
1477 | int first = 1; | |
1478 | int in_porcelain = (flags & REFRESH_IN_PORCELAIN); | |
1479 | unsigned int options = (CE_MATCH_REFRESH | | |
1480 | (really ? CE_MATCH_IGNORE_VALID : 0) | | |
1481 | (not_new ? CE_MATCH_IGNORE_MISSING : 0)); | |
1482 | const char *modified_fmt; | |
1483 | const char *deleted_fmt; | |
1484 | const char *typechange_fmt; | |
1485 | const char *added_fmt; | |
1486 | const char *unmerged_fmt; | |
1487 | struct progress *progress = NULL; | |
1488 | ||
1489 | if (flags & REFRESH_PROGRESS && isatty(2)) | |
1490 | progress = start_delayed_progress(_("Refresh index"), | |
1491 | istate->cache_nr); | |
1492 | ||
1493 | trace_performance_enter(); | |
1494 | modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n"); | |
1495 | deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n"); | |
1496 | typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n"); | |
1497 | added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n"); | |
1498 | unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n"); | |
1499 | /* | |
1500 | * Use the multi-threaded preload_index() to refresh most of the | |
1501 | * cache entries quickly then in the single threaded loop below, | |
1502 | * we only have to do the special cases that are left. | |
1503 | */ | |
1504 | preload_index(istate, pathspec, 0); | |
1505 | for (i = 0; i < istate->cache_nr; i++) { | |
1506 | struct cache_entry *ce, *new_entry; | |
1507 | int cache_errno = 0; | |
1508 | int changed = 0; | |
1509 | int filtered = 0; | |
1510 | ||
1511 | ce = istate->cache[i]; | |
1512 | if (ignore_submodules && S_ISGITLINK(ce->ce_mode)) | |
1513 | continue; | |
1514 | ||
1515 | if (pathspec && !ce_path_match(istate, ce, pathspec, seen)) | |
1516 | filtered = 1; | |
1517 | ||
1518 | if (ce_stage(ce)) { | |
1519 | while ((i < istate->cache_nr) && | |
1520 | ! strcmp(istate->cache[i]->name, ce->name)) | |
1521 | i++; | |
1522 | i--; | |
1523 | if (allow_unmerged) | |
1524 | continue; | |
1525 | if (!filtered) | |
1526 | show_file(unmerged_fmt, ce->name, in_porcelain, | |
1527 | &first, header_msg); | |
1528 | has_errors = 1; | |
1529 | continue; | |
1530 | } | |
1531 | ||
1532 | if (filtered) | |
1533 | continue; | |
1534 | ||
1535 | new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed); | |
1536 | if (new_entry == ce) | |
1537 | continue; | |
1538 | if (progress) | |
1539 | display_progress(progress, i); | |
1540 | if (!new_entry) { | |
1541 | const char *fmt; | |
1542 | ||
1543 | if (really && cache_errno == EINVAL) { | |
1544 | /* If we are doing --really-refresh that | |
1545 | * means the index is not valid anymore. | |
1546 | */ | |
1547 | ce->ce_flags &= ~CE_VALID; | |
1548 | ce->ce_flags |= CE_UPDATE_IN_BASE; | |
1549 | mark_fsmonitor_invalid(istate, ce); | |
1550 | istate->cache_changed |= CE_ENTRY_CHANGED; | |
1551 | } | |
1552 | if (quiet) | |
1553 | continue; | |
1554 | ||
1555 | if (cache_errno == ENOENT) | |
1556 | fmt = deleted_fmt; | |
1557 | else if (ce_intent_to_add(ce)) | |
1558 | fmt = added_fmt; /* must be before other checks */ | |
1559 | else if (changed & TYPE_CHANGED) | |
1560 | fmt = typechange_fmt; | |
1561 | else | |
1562 | fmt = modified_fmt; | |
1563 | show_file(fmt, | |
1564 | ce->name, in_porcelain, &first, header_msg); | |
1565 | has_errors = 1; | |
1566 | continue; | |
1567 | } | |
1568 | ||
1569 | replace_index_entry(istate, i, new_entry); | |
1570 | } | |
1571 | if (progress) { | |
1572 | display_progress(progress, istate->cache_nr); | |
1573 | stop_progress(&progress); | |
1574 | } | |
1575 | trace_performance_leave("refresh index"); | |
1576 | return has_errors; | |
1577 | } | |
1578 | ||
1579 | struct cache_entry *refresh_cache_entry(struct index_state *istate, | |
1580 | struct cache_entry *ce, | |
1581 | unsigned int options) | |
1582 | { | |
1583 | return refresh_cache_ent(istate, ce, options, NULL, NULL); | |
1584 | } | |
1585 | ||
1586 | ||
1587 | /***************************************************************** | |
1588 | * Index File I/O | |
1589 | *****************************************************************/ | |
1590 | ||
1591 | #define INDEX_FORMAT_DEFAULT 3 | |
1592 | ||
1593 | static unsigned int get_index_format_default(void) | |
1594 | { | |
1595 | char *envversion = getenv("GIT_INDEX_VERSION"); | |
1596 | char *endp; | |
1597 | int value; | |
1598 | unsigned int version = INDEX_FORMAT_DEFAULT; | |
1599 | ||
1600 | if (!envversion) { | |
1601 | if (!git_config_get_int("index.version", &value)) | |
1602 | version = value; | |
1603 | if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) { | |
1604 | warning(_("index.version set, but the value is invalid.\n" | |
1605 | "Using version %i"), INDEX_FORMAT_DEFAULT); | |
1606 | return INDEX_FORMAT_DEFAULT; | |
1607 | } | |
1608 | return version; | |
1609 | } | |
1610 | ||
1611 | version = strtoul(envversion, &endp, 10); | |
1612 | if (*endp || | |
1613 | version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) { | |
1614 | warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n" | |
1615 | "Using version %i"), INDEX_FORMAT_DEFAULT); | |
1616 | version = INDEX_FORMAT_DEFAULT; | |
1617 | } | |
1618 | return version; | |
1619 | } | |
1620 | ||
1621 | /* | |
1622 | * dev/ino/uid/gid/size are also just tracked to the low 32 bits | |
1623 | * Again - this is just a (very strong in practice) heuristic that | |
1624 | * the inode hasn't changed. | |
1625 | * | |
1626 | * We save the fields in big-endian order to allow using the | |
1627 | * index file over NFS transparently. | |
1628 | */ | |
1629 | struct ondisk_cache_entry { | |
1630 | struct cache_time ctime; | |
1631 | struct cache_time mtime; | |
1632 | uint32_t dev; | |
1633 | uint32_t ino; | |
1634 | uint32_t mode; | |
1635 | uint32_t uid; | |
1636 | uint32_t gid; | |
1637 | uint32_t size; | |
1638 | unsigned char sha1[20]; | |
1639 | uint16_t flags; | |
1640 | char name[FLEX_ARRAY]; /* more */ | |
1641 | }; | |
1642 | ||
1643 | /* | |
1644 | * This struct is used when CE_EXTENDED bit is 1 | |
1645 | * The struct must match ondisk_cache_entry exactly from | |
1646 | * ctime till flags | |
1647 | */ | |
1648 | struct ondisk_cache_entry_extended { | |
1649 | struct cache_time ctime; | |
1650 | struct cache_time mtime; | |
1651 | uint32_t dev; | |
1652 | uint32_t ino; | |
1653 | uint32_t mode; | |
1654 | uint32_t uid; | |
1655 | uint32_t gid; | |
1656 | uint32_t size; | |
1657 | unsigned char sha1[20]; | |
1658 | uint16_t flags; | |
1659 | uint16_t flags2; | |
1660 | char name[FLEX_ARRAY]; /* more */ | |
1661 | }; | |
1662 | ||
1663 | /* These are only used for v3 or lower */ | |
1664 | #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len) | |
1665 | #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7) | |
1666 | #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len) | |
1667 | #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len) | |
1668 | #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \ | |
1669 | ondisk_cache_entry_extended_size(ce_namelen(ce)) : \ | |
1670 | ondisk_cache_entry_size(ce_namelen(ce))) | |
1671 | ||
1672 | /* Allow fsck to force verification of the index checksum. */ | |
1673 | int verify_index_checksum; | |
1674 | ||
1675 | /* Allow fsck to force verification of the cache entry order. */ | |
1676 | int verify_ce_order; | |
1677 | ||
1678 | static int verify_hdr(const struct cache_header *hdr, unsigned long size) | |
1679 | { | |
1680 | git_hash_ctx c; | |
1681 | unsigned char hash[GIT_MAX_RAWSZ]; | |
1682 | int hdr_version; | |
1683 | ||
1684 | if (hdr->hdr_signature != htonl(CACHE_SIGNATURE)) | |
1685 | return error("bad signature"); | |
1686 | hdr_version = ntohl(hdr->hdr_version); | |
1687 | if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version) | |
1688 | return error("bad index version %d", hdr_version); | |
1689 | ||
1690 | if (!verify_index_checksum) | |
1691 | return 0; | |
1692 | ||
1693 | the_hash_algo->init_fn(&c); | |
1694 | the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz); | |
1695 | the_hash_algo->final_fn(hash, &c); | |
1696 | if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz)) | |
1697 | return error("bad index file sha1 signature"); | |
1698 | return 0; | |
1699 | } | |
1700 | ||
1701 | static int read_index_extension(struct index_state *istate, | |
1702 | const char *ext, const char *data, unsigned long sz) | |
1703 | { | |
1704 | switch (CACHE_EXT(ext)) { | |
1705 | case CACHE_EXT_TREE: | |
1706 | istate->cache_tree = cache_tree_read(data, sz); | |
1707 | break; | |
1708 | case CACHE_EXT_RESOLVE_UNDO: | |
1709 | istate->resolve_undo = resolve_undo_read(data, sz); | |
1710 | break; | |
1711 | case CACHE_EXT_LINK: | |
1712 | if (read_link_extension(istate, data, sz)) | |
1713 | return -1; | |
1714 | break; | |
1715 | case CACHE_EXT_UNTRACKED: | |
1716 | istate->untracked = read_untracked_extension(data, sz); | |
1717 | break; | |
1718 | case CACHE_EXT_FSMONITOR: | |
1719 | read_fsmonitor_extension(istate, data, sz); | |
1720 | break; | |
1721 | case CACHE_EXT_ENDOFINDEXENTRIES: | |
1722 | case CACHE_EXT_INDEXENTRYOFFSETTABLE: | |
1723 | /* already handled in do_read_index() */ | |
1724 | break; | |
1725 | default: | |
1726 | if (*ext < 'A' || 'Z' < *ext) | |
1727 | return error("index uses %.4s extension, which we do not understand", | |
1728 | ext); | |
1729 | fprintf(stderr, "ignoring %.4s extension\n", ext); | |
1730 | break; | |
1731 | } | |
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | int hold_locked_index(struct lock_file *lk, int lock_flags) | |
1736 | { | |
1737 | return hold_lock_file_for_update(lk, get_index_file(), lock_flags); | |
1738 | } | |
1739 | ||
1740 | int read_index(struct index_state *istate) | |
1741 | { | |
1742 | return read_index_from(istate, get_index_file(), get_git_dir()); | |
1743 | } | |
1744 | ||
1745 | static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool, | |
1746 | unsigned int version, | |
1747 | struct ondisk_cache_entry *ondisk, | |
1748 | unsigned long *ent_size, | |
1749 | const struct cache_entry *previous_ce) | |
1750 | { | |
1751 | struct cache_entry *ce; | |
1752 | size_t len; | |
1753 | const char *name; | |
1754 | unsigned int flags; | |
1755 | size_t copy_len; | |
1756 | /* | |
1757 | * Adjacent cache entries tend to share the leading paths, so it makes | |
1758 | * sense to only store the differences in later entries. In the v4 | |
1759 | * on-disk format of the index, each on-disk cache entry stores the | |
1760 | * number of bytes to be stripped from the end of the previous name, | |
1761 | * and the bytes to append to the result, to come up with its name. | |
1762 | */ | |
1763 | int expand_name_field = version == 4; | |
1764 | ||
1765 | /* On-disk flags are just 16 bits */ | |
1766 | flags = get_be16(&ondisk->flags); | |
1767 | len = flags & CE_NAMEMASK; | |
1768 | ||
1769 | if (flags & CE_EXTENDED) { | |
1770 | struct ondisk_cache_entry_extended *ondisk2; | |
1771 | int extended_flags; | |
1772 | ondisk2 = (struct ondisk_cache_entry_extended *)ondisk; | |
1773 | extended_flags = get_be16(&ondisk2->flags2) << 16; | |
1774 | /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */ | |
1775 | if (extended_flags & ~CE_EXTENDED_FLAGS) | |
1776 | die("Unknown index entry format %08x", extended_flags); | |
1777 | flags |= extended_flags; | |
1778 | name = ondisk2->name; | |
1779 | } | |
1780 | else | |
1781 | name = ondisk->name; | |
1782 | ||
1783 | if (expand_name_field) { | |
1784 | const unsigned char *cp = (const unsigned char *)name; | |
1785 | size_t strip_len, previous_len; | |
1786 | ||
1787 | /* If we're at the begining of a block, ignore the previous name */ | |
1788 | strip_len = decode_varint(&cp); | |
1789 | if (previous_ce) { | |
1790 | previous_len = previous_ce->ce_namelen; | |
1791 | if (previous_len < strip_len) | |
1792 | die(_("malformed name field in the index, near path '%s'"), | |
1793 | previous_ce->name); | |
1794 | copy_len = previous_len - strip_len; | |
1795 | } else { | |
1796 | copy_len = 0; | |
1797 | } | |
1798 | name = (const char *)cp; | |
1799 | } | |
1800 | ||
1801 | if (len == CE_NAMEMASK) { | |
1802 | len = strlen(name); | |
1803 | if (expand_name_field) | |
1804 | len += copy_len; | |
1805 | } | |
1806 | ||
1807 | ce = mem_pool__ce_alloc(ce_mem_pool, len); | |
1808 | ||
1809 | ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec); | |
1810 | ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec); | |
1811 | ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec); | |
1812 | ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec); | |
1813 | ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev); | |
1814 | ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino); | |
1815 | ce->ce_mode = get_be32(&ondisk->mode); | |
1816 | ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid); | |
1817 | ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid); | |
1818 | ce->ce_stat_data.sd_size = get_be32(&ondisk->size); | |
1819 | ce->ce_flags = flags & ~CE_NAMEMASK; | |
1820 | ce->ce_namelen = len; | |
1821 | ce->index = 0; | |
1822 | hashcpy(ce->oid.hash, ondisk->sha1); | |
1823 | ||
1824 | if (expand_name_field) { | |
1825 | if (copy_len) | |
1826 | memcpy(ce->name, previous_ce->name, copy_len); | |
1827 | memcpy(ce->name + copy_len, name, len + 1 - copy_len); | |
1828 | *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len; | |
1829 | } else { | |
1830 | memcpy(ce->name, name, len + 1); | |
1831 | *ent_size = ondisk_ce_size(ce); | |
1832 | } | |
1833 | return ce; | |
1834 | } | |
1835 | ||
1836 | static void check_ce_order(struct index_state *istate) | |
1837 | { | |
1838 | unsigned int i; | |
1839 | ||
1840 | if (!verify_ce_order) | |
1841 | return; | |
1842 | ||
1843 | for (i = 1; i < istate->cache_nr; i++) { | |
1844 | struct cache_entry *ce = istate->cache[i - 1]; | |
1845 | struct cache_entry *next_ce = istate->cache[i]; | |
1846 | int name_compare = strcmp(ce->name, next_ce->name); | |
1847 | ||
1848 | if (0 < name_compare) | |
1849 | die("unordered stage entries in index"); | |
1850 | if (!name_compare) { | |
1851 | if (!ce_stage(ce)) | |
1852 | die("multiple stage entries for merged file '%s'", | |
1853 | ce->name); | |
1854 | if (ce_stage(ce) > ce_stage(next_ce)) | |
1855 | die("unordered stage entries for '%s'", | |
1856 | ce->name); | |
1857 | } | |
1858 | } | |
1859 | } | |
1860 | ||
1861 | static void tweak_untracked_cache(struct index_state *istate) | |
1862 | { | |
1863 | switch (git_config_get_untracked_cache()) { | |
1864 | case -1: /* keep: do nothing */ | |
1865 | break; | |
1866 | case 0: /* false */ | |
1867 | remove_untracked_cache(istate); | |
1868 | break; | |
1869 | case 1: /* true */ | |
1870 | add_untracked_cache(istate); | |
1871 | break; | |
1872 | default: /* unknown value: do nothing */ | |
1873 | break; | |
1874 | } | |
1875 | } | |
1876 | ||
1877 | static void tweak_split_index(struct index_state *istate) | |
1878 | { | |
1879 | switch (git_config_get_split_index()) { | |
1880 | case -1: /* unset: do nothing */ | |
1881 | break; | |
1882 | case 0: /* false */ | |
1883 | remove_split_index(istate); | |
1884 | break; | |
1885 | case 1: /* true */ | |
1886 | add_split_index(istate); | |
1887 | break; | |
1888 | default: /* unknown value: do nothing */ | |
1889 | break; | |
1890 | } | |
1891 | } | |
1892 | ||
1893 | static void post_read_index_from(struct index_state *istate) | |
1894 | { | |
1895 | check_ce_order(istate); | |
1896 | tweak_untracked_cache(istate); | |
1897 | tweak_split_index(istate); | |
1898 | tweak_fsmonitor(istate); | |
1899 | } | |
1900 | ||
1901 | static size_t estimate_cache_size_from_compressed(unsigned int entries) | |
1902 | { | |
1903 | return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH); | |
1904 | } | |
1905 | ||
1906 | static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries) | |
1907 | { | |
1908 | long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry); | |
1909 | ||
1910 | /* | |
1911 | * Account for potential alignment differences. | |
1912 | */ | |
1913 | per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry)); | |
1914 | return ondisk_size + entries * per_entry; | |
1915 | } | |
1916 | ||
1917 | struct index_entry_offset | |
1918 | { | |
1919 | /* starting byte offset into index file, count of index entries in this block */ | |
1920 | int offset, nr; | |
1921 | }; | |
1922 | ||
1923 | struct index_entry_offset_table | |
1924 | { | |
1925 | int nr; | |
1926 | struct index_entry_offset entries[FLEX_ARRAY]; | |
1927 | }; | |
1928 | ||
1929 | #ifndef NO_PTHREADS | |
1930 | static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset); | |
1931 | static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot); | |
1932 | #endif | |
1933 | ||
1934 | static size_t read_eoie_extension(const char *mmap, size_t mmap_size); | |
1935 | static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset); | |
1936 | ||
1937 | struct load_index_extensions | |
1938 | { | |
1939 | #ifndef NO_PTHREADS | |
1940 | pthread_t pthread; | |
1941 | #endif | |
1942 | struct index_state *istate; | |
1943 | const char *mmap; | |
1944 | size_t mmap_size; | |
1945 | unsigned long src_offset; | |
1946 | }; | |
1947 | ||
1948 | static void *load_index_extensions(void *_data) | |
1949 | { | |
1950 | struct load_index_extensions *p = _data; | |
1951 | unsigned long src_offset = p->src_offset; | |
1952 | ||
1953 | while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) { | |
1954 | /* After an array of active_nr index entries, | |
1955 | * there can be arbitrary number of extended | |
1956 | * sections, each of which is prefixed with | |
1957 | * extension name (4-byte) and section length | |
1958 | * in 4-byte network byte order. | |
1959 | */ | |
1960 | uint32_t extsize = get_be32(p->mmap + src_offset + 4); | |
1961 | if (read_index_extension(p->istate, | |
1962 | p->mmap + src_offset, | |
1963 | p->mmap + src_offset + 8, | |
1964 | extsize) < 0) { | |
1965 | munmap((void *)p->mmap, p->mmap_size); | |
1966 | die(_("index file corrupt")); | |
1967 | } | |
1968 | src_offset += 8; | |
1969 | src_offset += extsize; | |
1970 | } | |
1971 | ||
1972 | return NULL; | |
1973 | } | |
1974 | ||
1975 | /* | |
1976 | * A helper function that will load the specified range of cache entries | |
1977 | * from the memory mapped file and add them to the given index. | |
1978 | */ | |
1979 | static unsigned long load_cache_entry_block(struct index_state *istate, | |
1980 | struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap, | |
1981 | unsigned long start_offset, const struct cache_entry *previous_ce) | |
1982 | { | |
1983 | int i; | |
1984 | unsigned long src_offset = start_offset; | |
1985 | ||
1986 | for (i = offset; i < offset + nr; i++) { | |
1987 | struct ondisk_cache_entry *disk_ce; | |
1988 | struct cache_entry *ce; | |
1989 | unsigned long consumed; | |
1990 | ||
1991 | disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset); | |
1992 | ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce); | |
1993 | set_index_entry(istate, i, ce); | |
1994 | ||
1995 | src_offset += consumed; | |
1996 | previous_ce = ce; | |
1997 | } | |
1998 | return src_offset - start_offset; | |
1999 | } | |
2000 | ||
2001 | static unsigned long load_all_cache_entries(struct index_state *istate, | |
2002 | const char *mmap, size_t mmap_size, unsigned long src_offset) | |
2003 | { | |
2004 | unsigned long consumed; | |
2005 | ||
2006 | if (istate->version == 4) { | |
2007 | mem_pool_init(&istate->ce_mem_pool, | |
2008 | estimate_cache_size_from_compressed(istate->cache_nr)); | |
2009 | } else { | |
2010 | mem_pool_init(&istate->ce_mem_pool, | |
2011 | estimate_cache_size(mmap_size, istate->cache_nr)); | |
2012 | } | |
2013 | ||
2014 | consumed = load_cache_entry_block(istate, istate->ce_mem_pool, | |
2015 | 0, istate->cache_nr, mmap, src_offset, NULL); | |
2016 | return consumed; | |
2017 | } | |
2018 | ||
2019 | #ifndef NO_PTHREADS | |
2020 | ||
2021 | /* | |
2022 | * Mostly randomly chosen maximum thread counts: we | |
2023 | * cap the parallelism to online_cpus() threads, and we want | |
2024 | * to have at least 10000 cache entries per thread for it to | |
2025 | * be worth starting a thread. | |
2026 | */ | |
2027 | ||
2028 | #define THREAD_COST (10000) | |
2029 | ||
2030 | struct load_cache_entries_thread_data | |
2031 | { | |
2032 | pthread_t pthread; | |
2033 | struct index_state *istate; | |
2034 | struct mem_pool *ce_mem_pool; | |
2035 | int offset; | |
2036 | const char *mmap; | |
2037 | struct index_entry_offset_table *ieot; | |
2038 | int ieot_start; /* starting index into the ieot array */ | |
2039 | int ieot_blocks; /* count of ieot entries to process */ | |
2040 | unsigned long consumed; /* return # of bytes in index file processed */ | |
2041 | }; | |
2042 | ||
2043 | /* | |
2044 | * A thread proc to run the load_cache_entries() computation | |
2045 | * across multiple background threads. | |
2046 | */ | |
2047 | static void *load_cache_entries_thread(void *_data) | |
2048 | { | |
2049 | struct load_cache_entries_thread_data *p = _data; | |
2050 | int i; | |
2051 | ||
2052 | /* iterate across all ieot blocks assigned to this thread */ | |
2053 | for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) { | |
2054 | p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool, | |
2055 | p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL); | |
2056 | p->offset += p->ieot->entries[i].nr; | |
2057 | } | |
2058 | return NULL; | |
2059 | } | |
2060 | ||
2061 | static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size, | |
2062 | unsigned long src_offset, int nr_threads, struct index_entry_offset_table *ieot) | |
2063 | { | |
2064 | int i, offset, ieot_blocks, ieot_start, err; | |
2065 | struct load_cache_entries_thread_data *data; | |
2066 | unsigned long consumed = 0; | |
2067 | ||
2068 | /* a little sanity checking */ | |
2069 | if (istate->name_hash_initialized) | |
2070 | BUG("the name hash isn't thread safe"); | |
2071 | ||
2072 | mem_pool_init(&istate->ce_mem_pool, 0); | |
2073 | ||
2074 | /* ensure we have no more threads than we have blocks to process */ | |
2075 | if (nr_threads > ieot->nr) | |
2076 | nr_threads = ieot->nr; | |
2077 | data = xcalloc(nr_threads, sizeof(*data)); | |
2078 | ||
2079 | offset = ieot_start = 0; | |
2080 | ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads); | |
2081 | for (i = 0; i < nr_threads; i++) { | |
2082 | struct load_cache_entries_thread_data *p = &data[i]; | |
2083 | int nr, j; | |
2084 | ||
2085 | if (ieot_start + ieot_blocks > ieot->nr) | |
2086 | ieot_blocks = ieot->nr - ieot_start; | |
2087 | ||
2088 | p->istate = istate; | |
2089 | p->offset = offset; | |
2090 | p->mmap = mmap; | |
2091 | p->ieot = ieot; | |
2092 | p->ieot_start = ieot_start; | |
2093 | p->ieot_blocks = ieot_blocks; | |
2094 | ||
2095 | /* create a mem_pool for each thread */ | |
2096 | nr = 0; | |
2097 | for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++) | |
2098 | nr += p->ieot->entries[j].nr; | |
2099 | if (istate->version == 4) { | |
2100 | mem_pool_init(&p->ce_mem_pool, | |
2101 | estimate_cache_size_from_compressed(nr)); | |
2102 | } else { | |
2103 | mem_pool_init(&p->ce_mem_pool, | |
2104 | estimate_cache_size(mmap_size, nr)); | |
2105 | } | |
2106 | ||
2107 | err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p); | |
2108 | if (err) | |
2109 | die(_("unable to create load_cache_entries thread: %s"), strerror(err)); | |
2110 | ||
2111 | /* increment by the number of cache entries in the ieot block being processed */ | |
2112 | for (j = 0; j < ieot_blocks; j++) | |
2113 | offset += ieot->entries[ieot_start + j].nr; | |
2114 | ieot_start += ieot_blocks; | |
2115 | } | |
2116 | ||
2117 | for (i = 0; i < nr_threads; i++) { | |
2118 | struct load_cache_entries_thread_data *p = &data[i]; | |
2119 | ||
2120 | err = pthread_join(p->pthread, NULL); | |
2121 | if (err) | |
2122 | die(_("unable to join load_cache_entries thread: %s"), strerror(err)); | |
2123 | mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool); | |
2124 | consumed += p->consumed; | |
2125 | } | |
2126 | ||
2127 | free(data); | |
2128 | ||
2129 | return consumed; | |
2130 | } | |
2131 | #endif | |
2132 | ||
2133 | /* remember to discard_cache() before reading a different cache! */ | |
2134 | int do_read_index(struct index_state *istate, const char *path, int must_exist) | |
2135 | { | |
2136 | int fd; | |
2137 | struct stat st; | |
2138 | unsigned long src_offset; | |
2139 | const struct cache_header *hdr; | |
2140 | const char *mmap; | |
2141 | size_t mmap_size; | |
2142 | struct load_index_extensions p; | |
2143 | size_t extension_offset = 0; | |
2144 | #ifndef NO_PTHREADS | |
2145 | int nr_threads, cpus; | |
2146 | struct index_entry_offset_table *ieot = NULL; | |
2147 | #endif | |
2148 | ||
2149 | if (istate->initialized) | |
2150 | return istate->cache_nr; | |
2151 | ||
2152 | istate->timestamp.sec = 0; | |
2153 | istate->timestamp.nsec = 0; | |
2154 | fd = open(path, O_RDONLY); | |
2155 | if (fd < 0) { | |
2156 | if (!must_exist && errno == ENOENT) | |
2157 | return 0; | |
2158 | die_errno("%s: index file open failed", path); | |
2159 | } | |
2160 | ||
2161 | if (fstat(fd, &st)) | |
2162 | die_errno("cannot stat the open index"); | |
2163 | ||
2164 | mmap_size = xsize_t(st.st_size); | |
2165 | if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz) | |
2166 | die("index file smaller than expected"); | |
2167 | ||
2168 | mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
2169 | if (mmap == MAP_FAILED) | |
2170 | die_errno("unable to map index file"); | |
2171 | close(fd); | |
2172 | ||
2173 | hdr = (const struct cache_header *)mmap; | |
2174 | if (verify_hdr(hdr, mmap_size) < 0) | |
2175 | goto unmap; | |
2176 | ||
2177 | hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz); | |
2178 | istate->version = ntohl(hdr->hdr_version); | |
2179 | istate->cache_nr = ntohl(hdr->hdr_entries); | |
2180 | istate->cache_alloc = alloc_nr(istate->cache_nr); | |
2181 | istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache)); | |
2182 | istate->initialized = 1; | |
2183 | ||
2184 | p.istate = istate; | |
2185 | p.mmap = mmap; | |
2186 | p.mmap_size = mmap_size; | |
2187 | ||
2188 | src_offset = sizeof(*hdr); | |
2189 | ||
2190 | #ifndef NO_PTHREADS | |
2191 | nr_threads = git_config_get_index_threads(); | |
2192 | ||
2193 | /* TODO: does creating more threads than cores help? */ | |
2194 | if (!nr_threads) { | |
2195 | nr_threads = istate->cache_nr / THREAD_COST; | |
2196 | cpus = online_cpus(); | |
2197 | if (nr_threads > cpus) | |
2198 | nr_threads = cpus; | |
2199 | } | |
2200 | ||
2201 | if (nr_threads > 1) { | |
2202 | extension_offset = read_eoie_extension(mmap, mmap_size); | |
2203 | if (extension_offset) { | |
2204 | int err; | |
2205 | ||
2206 | p.src_offset = extension_offset; | |
2207 | err = pthread_create(&p.pthread, NULL, load_index_extensions, &p); | |
2208 | if (err) | |
2209 | die(_("unable to create load_index_extensions thread: %s"), strerror(err)); | |
2210 | ||
2211 | nr_threads--; | |
2212 | } | |
2213 | } | |
2214 | ||
2215 | /* | |
2216 | * Locate and read the index entry offset table so that we can use it | |
2217 | * to multi-thread the reading of the cache entries. | |
2218 | */ | |
2219 | if (extension_offset && nr_threads > 1) | |
2220 | ieot = read_ieot_extension(mmap, mmap_size, extension_offset); | |
2221 | ||
2222 | if (ieot) { | |
2223 | src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, src_offset, nr_threads, ieot); | |
2224 | free(ieot); | |
2225 | } else { | |
2226 | src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset); | |
2227 | } | |
2228 | #else | |
2229 | src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset); | |
2230 | #endif | |
2231 | ||
2232 | istate->timestamp.sec = st.st_mtime; | |
2233 | istate->timestamp.nsec = ST_MTIME_NSEC(st); | |
2234 | ||
2235 | /* if we created a thread, join it otherwise load the extensions on the primary thread */ | |
2236 | #ifndef NO_PTHREADS | |
2237 | if (extension_offset) { | |
2238 | int ret = pthread_join(p.pthread, NULL); | |
2239 | if (ret) | |
2240 | die(_("unable to join load_index_extensions thread: %s"), strerror(ret)); | |
2241 | } | |
2242 | #endif | |
2243 | if (!extension_offset) { | |
2244 | p.src_offset = src_offset; | |
2245 | load_index_extensions(&p); | |
2246 | } | |
2247 | munmap((void *)mmap, mmap_size); | |
2248 | return istate->cache_nr; | |
2249 | ||
2250 | unmap: | |
2251 | munmap((void *)mmap, mmap_size); | |
2252 | die("index file corrupt"); | |
2253 | } | |
2254 | ||
2255 | /* | |
2256 | * Signal that the shared index is used by updating its mtime. | |
2257 | * | |
2258 | * This way, shared index can be removed if they have not been used | |
2259 | * for some time. | |
2260 | */ | |
2261 | static void freshen_shared_index(const char *shared_index, int warn) | |
2262 | { | |
2263 | if (!check_and_freshen_file(shared_index, 1) && warn) | |
2264 | warning("could not freshen shared index '%s'", shared_index); | |
2265 | } | |
2266 | ||
2267 | int read_index_from(struct index_state *istate, const char *path, | |
2268 | const char *gitdir) | |
2269 | { | |
2270 | struct split_index *split_index; | |
2271 | int ret; | |
2272 | char *base_oid_hex; | |
2273 | char *base_path; | |
2274 | ||
2275 | /* istate->initialized covers both .git/index and .git/sharedindex.xxx */ | |
2276 | if (istate->initialized) | |
2277 | return istate->cache_nr; | |
2278 | ||
2279 | trace_performance_enter(); | |
2280 | ret = do_read_index(istate, path, 0); | |
2281 | trace_performance_leave("read cache %s", path); | |
2282 | ||
2283 | split_index = istate->split_index; | |
2284 | if (!split_index || is_null_oid(&split_index->base_oid)) { | |
2285 | post_read_index_from(istate); | |
2286 | return ret; | |
2287 | } | |
2288 | ||
2289 | trace_performance_enter(); | |
2290 | if (split_index->base) | |
2291 | discard_index(split_index->base); | |
2292 | else | |
2293 | split_index->base = xcalloc(1, sizeof(*split_index->base)); | |
2294 | ||
2295 | base_oid_hex = oid_to_hex(&split_index->base_oid); | |
2296 | base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex); | |
2297 | ret = do_read_index(split_index->base, base_path, 1); | |
2298 | if (!oideq(&split_index->base_oid, &split_index->base->oid)) | |
2299 | die("broken index, expect %s in %s, got %s", | |
2300 | base_oid_hex, base_path, | |
2301 | oid_to_hex(&split_index->base->oid)); | |
2302 | ||
2303 | freshen_shared_index(base_path, 0); | |
2304 | merge_base_index(istate); | |
2305 | post_read_index_from(istate); | |
2306 | trace_performance_leave("read cache %s", base_path); | |
2307 | free(base_path); | |
2308 | return ret; | |
2309 | } | |
2310 | ||
2311 | int is_index_unborn(struct index_state *istate) | |
2312 | { | |
2313 | return (!istate->cache_nr && !istate->timestamp.sec); | |
2314 | } | |
2315 | ||
2316 | int discard_index(struct index_state *istate) | |
2317 | { | |
2318 | /* | |
2319 | * Cache entries in istate->cache[] should have been allocated | |
2320 | * from the memory pool associated with this index, or from an | |
2321 | * associated split_index. There is no need to free individual | |
2322 | * cache entries. validate_cache_entries can detect when this | |
2323 | * assertion does not hold. | |
2324 | */ | |
2325 | validate_cache_entries(istate); | |
2326 | ||
2327 | resolve_undo_clear_index(istate); | |
2328 | istate->cache_nr = 0; | |
2329 | istate->cache_changed = 0; | |
2330 | istate->timestamp.sec = 0; | |
2331 | istate->timestamp.nsec = 0; | |
2332 | free_name_hash(istate); | |
2333 | cache_tree_free(&(istate->cache_tree)); | |
2334 | istate->initialized = 0; | |
2335 | FREE_AND_NULL(istate->cache); | |
2336 | istate->cache_alloc = 0; | |
2337 | discard_split_index(istate); | |
2338 | free_untracked_cache(istate->untracked); | |
2339 | istate->untracked = NULL; | |
2340 | ||
2341 | if (istate->ce_mem_pool) { | |
2342 | mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries()); | |
2343 | istate->ce_mem_pool = NULL; | |
2344 | } | |
2345 | ||
2346 | return 0; | |
2347 | } | |
2348 | ||
2349 | /* | |
2350 | * Validate the cache entries of this index. | |
2351 | * All cache entries associated with this index | |
2352 | * should have been allocated by the memory pool | |
2353 | * associated with this index, or by a referenced | |
2354 | * split index. | |
2355 | */ | |
2356 | void validate_cache_entries(const struct index_state *istate) | |
2357 | { | |
2358 | int i; | |
2359 | ||
2360 | if (!should_validate_cache_entries() ||!istate || !istate->initialized) | |
2361 | return; | |
2362 | ||
2363 | for (i = 0; i < istate->cache_nr; i++) { | |
2364 | if (!istate) { | |
2365 | die("internal error: cache entry is not allocated from expected memory pool"); | |
2366 | } else if (!istate->ce_mem_pool || | |
2367 | !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) { | |
2368 | if (!istate->split_index || | |
2369 | !istate->split_index->base || | |
2370 | !istate->split_index->base->ce_mem_pool || | |
2371 | !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) { | |
2372 | die("internal error: cache entry is not allocated from expected memory pool"); | |
2373 | } | |
2374 | } | |
2375 | } | |
2376 | ||
2377 | if (istate->split_index) | |
2378 | validate_cache_entries(istate->split_index->base); | |
2379 | } | |
2380 | ||
2381 | int unmerged_index(const struct index_state *istate) | |
2382 | { | |
2383 | int i; | |
2384 | for (i = 0; i < istate->cache_nr; i++) { | |
2385 | if (ce_stage(istate->cache[i])) | |
2386 | return 1; | |
2387 | } | |
2388 | return 0; | |
2389 | } | |
2390 | ||
2391 | int index_has_changes(struct index_state *istate, | |
2392 | struct tree *tree, | |
2393 | struct strbuf *sb) | |
2394 | { | |
2395 | struct object_id cmp; | |
2396 | int i; | |
2397 | ||
2398 | if (istate != &the_index) { | |
2399 | BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first."); | |
2400 | } | |
2401 | if (tree) | |
2402 | cmp = tree->object.oid; | |
2403 | if (tree || !get_oid_tree("HEAD", &cmp)) { | |
2404 | struct diff_options opt; | |
2405 | ||
2406 | repo_diff_setup(the_repository, &opt); | |
2407 | opt.flags.exit_with_status = 1; | |
2408 | if (!sb) | |
2409 | opt.flags.quick = 1; | |
2410 | do_diff_cache(&cmp, &opt); | |
2411 | diffcore_std(&opt); | |
2412 | for (i = 0; sb && i < diff_queued_diff.nr; i++) { | |
2413 | if (i) | |
2414 | strbuf_addch(sb, ' '); | |
2415 | strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path); | |
2416 | } | |
2417 | diff_flush(&opt); | |
2418 | return opt.flags.has_changes != 0; | |
2419 | } else { | |
2420 | for (i = 0; sb && i < istate->cache_nr; i++) { | |
2421 | if (i) | |
2422 | strbuf_addch(sb, ' '); | |
2423 | strbuf_addstr(sb, istate->cache[i]->name); | |
2424 | } | |
2425 | return !!istate->cache_nr; | |
2426 | } | |
2427 | } | |
2428 | ||
2429 | #define WRITE_BUFFER_SIZE 8192 | |
2430 | static unsigned char write_buffer[WRITE_BUFFER_SIZE]; | |
2431 | static unsigned long write_buffer_len; | |
2432 | ||
2433 | static int ce_write_flush(git_hash_ctx *context, int fd) | |
2434 | { | |
2435 | unsigned int buffered = write_buffer_len; | |
2436 | if (buffered) { | |
2437 | the_hash_algo->update_fn(context, write_buffer, buffered); | |
2438 | if (write_in_full(fd, write_buffer, buffered) < 0) | |
2439 | return -1; | |
2440 | write_buffer_len = 0; | |
2441 | } | |
2442 | return 0; | |
2443 | } | |
2444 | ||
2445 | static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len) | |
2446 | { | |
2447 | while (len) { | |
2448 | unsigned int buffered = write_buffer_len; | |
2449 | unsigned int partial = WRITE_BUFFER_SIZE - buffered; | |
2450 | if (partial > len) | |
2451 | partial = len; | |
2452 | memcpy(write_buffer + buffered, data, partial); | |
2453 | buffered += partial; | |
2454 | if (buffered == WRITE_BUFFER_SIZE) { | |
2455 | write_buffer_len = buffered; | |
2456 | if (ce_write_flush(context, fd)) | |
2457 | return -1; | |
2458 | buffered = 0; | |
2459 | } | |
2460 | write_buffer_len = buffered; | |
2461 | len -= partial; | |
2462 | data = (char *) data + partial; | |
2463 | } | |
2464 | return 0; | |
2465 | } | |
2466 | ||
2467 | static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context, | |
2468 | int fd, unsigned int ext, unsigned int sz) | |
2469 | { | |
2470 | ext = htonl(ext); | |
2471 | sz = htonl(sz); | |
2472 | if (eoie_context) { | |
2473 | the_hash_algo->update_fn(eoie_context, &ext, 4); | |
2474 | the_hash_algo->update_fn(eoie_context, &sz, 4); | |
2475 | } | |
2476 | return ((ce_write(context, fd, &ext, 4) < 0) || | |
2477 | (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0; | |
2478 | } | |
2479 | ||
2480 | static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash) | |
2481 | { | |
2482 | unsigned int left = write_buffer_len; | |
2483 | ||
2484 | if (left) { | |
2485 | write_buffer_len = 0; | |
2486 | the_hash_algo->update_fn(context, write_buffer, left); | |
2487 | } | |
2488 | ||
2489 | /* Flush first if not enough space for hash signature */ | |
2490 | if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) { | |
2491 | if (write_in_full(fd, write_buffer, left) < 0) | |
2492 | return -1; | |
2493 | left = 0; | |
2494 | } | |
2495 | ||
2496 | /* Append the hash signature at the end */ | |
2497 | the_hash_algo->final_fn(write_buffer + left, context); | |
2498 | hashcpy(hash, write_buffer + left); | |
2499 | left += the_hash_algo->rawsz; | |
2500 | return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0; | |
2501 | } | |
2502 | ||
2503 | static void ce_smudge_racily_clean_entry(struct index_state *istate, | |
2504 | struct cache_entry *ce) | |
2505 | { | |
2506 | /* | |
2507 | * The only thing we care about in this function is to smudge the | |
2508 | * falsely clean entry due to touch-update-touch race, so we leave | |
2509 | * everything else as they are. We are called for entries whose | |
2510 | * ce_stat_data.sd_mtime match the index file mtime. | |
2511 | * | |
2512 | * Note that this actually does not do much for gitlinks, for | |
2513 | * which ce_match_stat_basic() always goes to the actual | |
2514 | * contents. The caller checks with is_racy_timestamp() which | |
2515 | * always says "no" for gitlinks, so we are not called for them ;-) | |
2516 | */ | |
2517 | struct stat st; | |
2518 | ||
2519 | if (lstat(ce->name, &st) < 0) | |
2520 | return; | |
2521 | if (ce_match_stat_basic(ce, &st)) | |
2522 | return; | |
2523 | if (ce_modified_check_fs(istate, ce, &st)) { | |
2524 | /* This is "racily clean"; smudge it. Note that this | |
2525 | * is a tricky code. At first glance, it may appear | |
2526 | * that it can break with this sequence: | |
2527 | * | |
2528 | * $ echo xyzzy >frotz | |
2529 | * $ git-update-index --add frotz | |
2530 | * $ : >frotz | |
2531 | * $ sleep 3 | |
2532 | * $ echo filfre >nitfol | |
2533 | * $ git-update-index --add nitfol | |
2534 | * | |
2535 | * but it does not. When the second update-index runs, | |
2536 | * it notices that the entry "frotz" has the same timestamp | |
2537 | * as index, and if we were to smudge it by resetting its | |
2538 | * size to zero here, then the object name recorded | |
2539 | * in index is the 6-byte file but the cached stat information | |
2540 | * becomes zero --- which would then match what we would | |
2541 | * obtain from the filesystem next time we stat("frotz"). | |
2542 | * | |
2543 | * However, the second update-index, before calling | |
2544 | * this function, notices that the cached size is 6 | |
2545 | * bytes and what is on the filesystem is an empty | |
2546 | * file, and never calls us, so the cached size information | |
2547 | * for "frotz" stays 6 which does not match the filesystem. | |
2548 | */ | |
2549 | ce->ce_stat_data.sd_size = 0; | |
2550 | } | |
2551 | } | |
2552 | ||
2553 | /* Copy miscellaneous fields but not the name */ | |
2554 | static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk, | |
2555 | struct cache_entry *ce) | |
2556 | { | |
2557 | short flags; | |
2558 | ||
2559 | ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec); | |
2560 | ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec); | |
2561 | ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec); | |
2562 | ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec); | |
2563 | ondisk->dev = htonl(ce->ce_stat_data.sd_dev); | |
2564 | ondisk->ino = htonl(ce->ce_stat_data.sd_ino); | |
2565 | ondisk->mode = htonl(ce->ce_mode); | |
2566 | ondisk->uid = htonl(ce->ce_stat_data.sd_uid); | |
2567 | ondisk->gid = htonl(ce->ce_stat_data.sd_gid); | |
2568 | ondisk->size = htonl(ce->ce_stat_data.sd_size); | |
2569 | hashcpy(ondisk->sha1, ce->oid.hash); | |
2570 | ||
2571 | flags = ce->ce_flags & ~CE_NAMEMASK; | |
2572 | flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce)); | |
2573 | ondisk->flags = htons(flags); | |
2574 | if (ce->ce_flags & CE_EXTENDED) { | |
2575 | struct ondisk_cache_entry_extended *ondisk2; | |
2576 | ondisk2 = (struct ondisk_cache_entry_extended *)ondisk; | |
2577 | ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16); | |
2578 | } | |
2579 | } | |
2580 | ||
2581 | static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce, | |
2582 | struct strbuf *previous_name, struct ondisk_cache_entry *ondisk) | |
2583 | { | |
2584 | int size; | |
2585 | int result; | |
2586 | unsigned int saved_namelen; | |
2587 | int stripped_name = 0; | |
2588 | static unsigned char padding[8] = { 0x00 }; | |
2589 | ||
2590 | if (ce->ce_flags & CE_STRIP_NAME) { | |
2591 | saved_namelen = ce_namelen(ce); | |
2592 | ce->ce_namelen = 0; | |
2593 | stripped_name = 1; | |
2594 | } | |
2595 | ||
2596 | if (ce->ce_flags & CE_EXTENDED) | |
2597 | size = offsetof(struct ondisk_cache_entry_extended, name); | |
2598 | else | |
2599 | size = offsetof(struct ondisk_cache_entry, name); | |
2600 | ||
2601 | if (!previous_name) { | |
2602 | int len = ce_namelen(ce); | |
2603 | copy_cache_entry_to_ondisk(ondisk, ce); | |
2604 | result = ce_write(c, fd, ondisk, size); | |
2605 | if (!result) | |
2606 | result = ce_write(c, fd, ce->name, len); | |
2607 | if (!result) | |
2608 | result = ce_write(c, fd, padding, align_padding_size(size, len)); | |
2609 | } else { | |
2610 | int common, to_remove, prefix_size; | |
2611 | unsigned char to_remove_vi[16]; | |
2612 | for (common = 0; | |
2613 | (ce->name[common] && | |
2614 | common < previous_name->len && | |
2615 | ce->name[common] == previous_name->buf[common]); | |
2616 | common++) | |
2617 | ; /* still matching */ | |
2618 | to_remove = previous_name->len - common; | |
2619 | prefix_size = encode_varint(to_remove, to_remove_vi); | |
2620 | ||
2621 | copy_cache_entry_to_ondisk(ondisk, ce); | |
2622 | result = ce_write(c, fd, ondisk, size); | |
2623 | if (!result) | |
2624 | result = ce_write(c, fd, to_remove_vi, prefix_size); | |
2625 | if (!result) | |
2626 | result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common); | |
2627 | if (!result) | |
2628 | result = ce_write(c, fd, padding, 1); | |
2629 | ||
2630 | strbuf_splice(previous_name, common, to_remove, | |
2631 | ce->name + common, ce_namelen(ce) - common); | |
2632 | } | |
2633 | if (stripped_name) { | |
2634 | ce->ce_namelen = saved_namelen; | |
2635 | ce->ce_flags &= ~CE_STRIP_NAME; | |
2636 | } | |
2637 | ||
2638 | return result; | |
2639 | } | |
2640 | ||
2641 | /* | |
2642 | * This function verifies if index_state has the correct sha1 of the | |
2643 | * index file. Don't die if we have any other failure, just return 0. | |
2644 | */ | |
2645 | static int verify_index_from(const struct index_state *istate, const char *path) | |
2646 | { | |
2647 | int fd; | |
2648 | ssize_t n; | |
2649 | struct stat st; | |
2650 | unsigned char hash[GIT_MAX_RAWSZ]; | |
2651 | ||
2652 | if (!istate->initialized) | |
2653 | return 0; | |
2654 | ||
2655 | fd = open(path, O_RDONLY); | |
2656 | if (fd < 0) | |
2657 | return 0; | |
2658 | ||
2659 | if (fstat(fd, &st)) | |
2660 | goto out; | |
2661 | ||
2662 | if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz) | |
2663 | goto out; | |
2664 | ||
2665 | n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz); | |
2666 | if (n != the_hash_algo->rawsz) | |
2667 | goto out; | |
2668 | ||
2669 | if (!hasheq(istate->oid.hash, hash)) | |
2670 | goto out; | |
2671 | ||
2672 | close(fd); | |
2673 | return 1; | |
2674 | ||
2675 | out: | |
2676 | close(fd); | |
2677 | return 0; | |
2678 | } | |
2679 | ||
2680 | static int verify_index(const struct index_state *istate) | |
2681 | { | |
2682 | return verify_index_from(istate, get_index_file()); | |
2683 | } | |
2684 | ||
2685 | static int has_racy_timestamp(struct index_state *istate) | |
2686 | { | |
2687 | int entries = istate->cache_nr; | |
2688 | int i; | |
2689 | ||
2690 | for (i = 0; i < entries; i++) { | |
2691 | struct cache_entry *ce = istate->cache[i]; | |
2692 | if (is_racy_timestamp(istate, ce)) | |
2693 | return 1; | |
2694 | } | |
2695 | return 0; | |
2696 | } | |
2697 | ||
2698 | void update_index_if_able(struct index_state *istate, struct lock_file *lockfile) | |
2699 | { | |
2700 | if ((istate->cache_changed || has_racy_timestamp(istate)) && | |
2701 | verify_index(istate)) | |
2702 | write_locked_index(istate, lockfile, COMMIT_LOCK); | |
2703 | else | |
2704 | rollback_lock_file(lockfile); | |
2705 | } | |
2706 | ||
2707 | /* | |
2708 | * On success, `tempfile` is closed. If it is the temporary file | |
2709 | * of a `struct lock_file`, we will therefore effectively perform | |
2710 | * a 'close_lock_file_gently()`. Since that is an implementation | |
2711 | * detail of lockfiles, callers of `do_write_index()` should not | |
2712 | * rely on it. | |
2713 | */ | |
2714 | static int do_write_index(struct index_state *istate, struct tempfile *tempfile, | |
2715 | int strip_extensions) | |
2716 | { | |
2717 | uint64_t start = getnanotime(); | |
2718 | int newfd = tempfile->fd; | |
2719 | git_hash_ctx c, eoie_c; | |
2720 | struct cache_header hdr; | |
2721 | int i, err = 0, removed, extended, hdr_version; | |
2722 | struct cache_entry **cache = istate->cache; | |
2723 | int entries = istate->cache_nr; | |
2724 | struct stat st; | |
2725 | struct ondisk_cache_entry_extended ondisk; | |
2726 | struct strbuf previous_name_buf = STRBUF_INIT, *previous_name; | |
2727 | int drop_cache_tree = istate->drop_cache_tree; | |
2728 | off_t offset; | |
2729 | int ieot_entries = 1; | |
2730 | struct index_entry_offset_table *ieot = NULL; | |
2731 | int nr, nr_threads; | |
2732 | ||
2733 | for (i = removed = extended = 0; i < entries; i++) { | |
2734 | if (cache[i]->ce_flags & CE_REMOVE) | |
2735 | removed++; | |
2736 | ||
2737 | /* reduce extended entries if possible */ | |
2738 | cache[i]->ce_flags &= ~CE_EXTENDED; | |
2739 | if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) { | |
2740 | extended++; | |
2741 | cache[i]->ce_flags |= CE_EXTENDED; | |
2742 | } | |
2743 | } | |
2744 | ||
2745 | if (!istate->version) { | |
2746 | istate->version = get_index_format_default(); | |
2747 | if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) | |
2748 | init_split_index(istate); | |
2749 | } | |
2750 | ||
2751 | /* demote version 3 to version 2 when the latter suffices */ | |
2752 | if (istate->version == 3 || istate->version == 2) | |
2753 | istate->version = extended ? 3 : 2; | |
2754 | ||
2755 | hdr_version = istate->version; | |
2756 | ||
2757 | hdr.hdr_signature = htonl(CACHE_SIGNATURE); | |
2758 | hdr.hdr_version = htonl(hdr_version); | |
2759 | hdr.hdr_entries = htonl(entries - removed); | |
2760 | ||
2761 | the_hash_algo->init_fn(&c); | |
2762 | if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0) | |
2763 | return -1; | |
2764 | ||
2765 | #ifndef NO_PTHREADS | |
2766 | nr_threads = git_config_get_index_threads(); | |
2767 | if (nr_threads != 1) { | |
2768 | int ieot_blocks, cpus; | |
2769 | ||
2770 | /* | |
2771 | * ensure default number of ieot blocks maps evenly to the | |
2772 | * default number of threads that will process them leaving | |
2773 | * room for the thread to load the index extensions. | |
2774 | */ | |
2775 | if (!nr_threads) { | |
2776 | ieot_blocks = istate->cache_nr / THREAD_COST; | |
2777 | cpus = online_cpus(); | |
2778 | if (ieot_blocks > cpus - 1) | |
2779 | ieot_blocks = cpus - 1; | |
2780 | } else { | |
2781 | ieot_blocks = nr_threads; | |
2782 | if (ieot_blocks > istate->cache_nr) | |
2783 | ieot_blocks = istate->cache_nr; | |
2784 | } | |
2785 | ||
2786 | /* | |
2787 | * no reason to write out the IEOT extension if we don't | |
2788 | * have enough blocks to utilize multi-threading | |
2789 | */ | |
2790 | if (ieot_blocks > 1) { | |
2791 | ieot = xcalloc(1, sizeof(struct index_entry_offset_table) | |
2792 | + (ieot_blocks * sizeof(struct index_entry_offset))); | |
2793 | ieot_entries = DIV_ROUND_UP(entries, ieot_blocks); | |
2794 | } | |
2795 | } | |
2796 | #endif | |
2797 | ||
2798 | offset = lseek(newfd, 0, SEEK_CUR); | |
2799 | if (offset < 0) { | |
2800 | free(ieot); | |
2801 | return -1; | |
2802 | } | |
2803 | offset += write_buffer_len; | |
2804 | nr = 0; | |
2805 | previous_name = (hdr_version == 4) ? &previous_name_buf : NULL; | |
2806 | ||
2807 | for (i = 0; i < entries; i++) { | |
2808 | struct cache_entry *ce = cache[i]; | |
2809 | if (ce->ce_flags & CE_REMOVE) | |
2810 | continue; | |
2811 | if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce)) | |
2812 | ce_smudge_racily_clean_entry(istate, ce); | |
2813 | if (is_null_oid(&ce->oid)) { | |
2814 | static const char msg[] = "cache entry has null sha1: %s"; | |
2815 | static int allow = -1; | |
2816 | ||
2817 | if (allow < 0) | |
2818 | allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0); | |
2819 | if (allow) | |
2820 | warning(msg, ce->name); | |
2821 | else | |
2822 | err = error(msg, ce->name); | |
2823 | ||
2824 | drop_cache_tree = 1; | |
2825 | } | |
2826 | if (ieot && i && (i % ieot_entries == 0)) { | |
2827 | ieot->entries[ieot->nr].nr = nr; | |
2828 | ieot->entries[ieot->nr].offset = offset; | |
2829 | ieot->nr++; | |
2830 | /* | |
2831 | * If we have a V4 index, set the first byte to an invalid | |
2832 | * character to ensure there is nothing common with the previous | |
2833 | * entry | |
2834 | */ | |
2835 | if (previous_name) | |
2836 | previous_name->buf[0] = 0; | |
2837 | nr = 0; | |
2838 | offset = lseek(newfd, 0, SEEK_CUR); | |
2839 | if (offset < 0) { | |
2840 | free(ieot); | |
2841 | return -1; | |
2842 | } | |
2843 | offset += write_buffer_len; | |
2844 | } | |
2845 | if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0) | |
2846 | err = -1; | |
2847 | ||
2848 | if (err) | |
2849 | break; | |
2850 | nr++; | |
2851 | } | |
2852 | if (ieot && nr) { | |
2853 | ieot->entries[ieot->nr].nr = nr; | |
2854 | ieot->entries[ieot->nr].offset = offset; | |
2855 | ieot->nr++; | |
2856 | } | |
2857 | strbuf_release(&previous_name_buf); | |
2858 | ||
2859 | if (err) { | |
2860 | free(ieot); | |
2861 | return err; | |
2862 | } | |
2863 | ||
2864 | /* Write extension data here */ | |
2865 | offset = lseek(newfd, 0, SEEK_CUR); | |
2866 | if (offset < 0) { | |
2867 | free(ieot); | |
2868 | return -1; | |
2869 | } | |
2870 | offset += write_buffer_len; | |
2871 | the_hash_algo->init_fn(&eoie_c); | |
2872 | ||
2873 | /* | |
2874 | * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we | |
2875 | * can minimize the number of extensions we have to scan through to | |
2876 | * find it during load. Write it out regardless of the | |
2877 | * strip_extensions parameter as we need it when loading the shared | |
2878 | * index. | |
2879 | */ | |
2880 | #ifndef NO_PTHREADS | |
2881 | if (ieot) { | |
2882 | struct strbuf sb = STRBUF_INIT; | |
2883 | ||
2884 | write_ieot_extension(&sb, ieot); | |
2885 | err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0 | |
2886 | || ce_write(&c, newfd, sb.buf, sb.len) < 0; | |
2887 | strbuf_release(&sb); | |
2888 | free(ieot); | |
2889 | if (err) | |
2890 | return -1; | |
2891 | } | |
2892 | #endif | |
2893 | ||
2894 | if (!strip_extensions && istate->split_index) { | |
2895 | struct strbuf sb = STRBUF_INIT; | |
2896 | ||
2897 | err = write_link_extension(&sb, istate) < 0 || | |
2898 | write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK, | |
2899 | sb.len) < 0 || | |
2900 | ce_write(&c, newfd, sb.buf, sb.len) < 0; | |
2901 | strbuf_release(&sb); | |
2902 | if (err) | |
2903 | return -1; | |
2904 | } | |
2905 | if (!strip_extensions && !drop_cache_tree && istate->cache_tree) { | |
2906 | struct strbuf sb = STRBUF_INIT; | |
2907 | ||
2908 | cache_tree_write(&sb, istate->cache_tree); | |
2909 | err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0 | |
2910 | || ce_write(&c, newfd, sb.buf, sb.len) < 0; | |
2911 | strbuf_release(&sb); | |
2912 | if (err) | |
2913 | return -1; | |
2914 | } | |
2915 | if (!strip_extensions && istate->resolve_undo) { | |
2916 | struct strbuf sb = STRBUF_INIT; | |
2917 | ||
2918 | resolve_undo_write(&sb, istate->resolve_undo); | |
2919 | err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO, | |
2920 | sb.len) < 0 | |
2921 | || ce_write(&c, newfd, sb.buf, sb.len) < 0; | |
2922 | strbuf_release(&sb); | |
2923 | if (err) | |
2924 | return -1; | |
2925 | } | |
2926 | if (!strip_extensions && istate->untracked) { | |
2927 | struct strbuf sb = STRBUF_INIT; | |
2928 | ||
2929 | write_untracked_extension(&sb, istate->untracked); | |
2930 | err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED, | |
2931 | sb.len) < 0 || | |
2932 | ce_write(&c, newfd, sb.buf, sb.len) < 0; | |
2933 | strbuf_release(&sb); | |
2934 | if (err) | |
2935 | return -1; | |
2936 | } | |
2937 | if (!strip_extensions && istate->fsmonitor_last_update) { | |
2938 | struct strbuf sb = STRBUF_INIT; | |
2939 | ||
2940 | write_fsmonitor_extension(&sb, istate); | |
2941 | err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0 | |
2942 | || ce_write(&c, newfd, sb.buf, sb.len) < 0; | |
2943 | strbuf_release(&sb); | |
2944 | if (err) | |
2945 | return -1; | |
2946 | } | |
2947 | ||
2948 | /* | |
2949 | * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1 | |
2950 | * so that it can be found and processed before all the index entries are | |
2951 | * read. Write it out regardless of the strip_extensions parameter as we need it | |
2952 | * when loading the shared index. | |
2953 | */ | |
2954 | if (offset) { | |
2955 | struct strbuf sb = STRBUF_INIT; | |
2956 | ||
2957 | write_eoie_extension(&sb, &eoie_c, offset); | |
2958 | err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0 | |
2959 | || ce_write(&c, newfd, sb.buf, sb.len) < 0; | |
2960 | strbuf_release(&sb); | |
2961 | if (err) | |
2962 | return -1; | |
2963 | } | |
2964 | ||
2965 | if (ce_flush(&c, newfd, istate->oid.hash)) | |
2966 | return -1; | |
2967 | if (close_tempfile_gently(tempfile)) { | |
2968 | error(_("could not close '%s'"), tempfile->filename.buf); | |
2969 | return -1; | |
2970 | } | |
2971 | if (stat(tempfile->filename.buf, &st)) | |
2972 | return -1; | |
2973 | istate->timestamp.sec = (unsigned int)st.st_mtime; | |
2974 | istate->timestamp.nsec = ST_MTIME_NSEC(st); | |
2975 | trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed); | |
2976 | return 0; | |
2977 | } | |
2978 | ||
2979 | void set_alternate_index_output(const char *name) | |
2980 | { | |
2981 | alternate_index_output = name; | |
2982 | } | |
2983 | ||
2984 | static int commit_locked_index(struct lock_file *lk) | |
2985 | { | |
2986 | if (alternate_index_output) | |
2987 | return commit_lock_file_to(lk, alternate_index_output); | |
2988 | else | |
2989 | return commit_lock_file(lk); | |
2990 | } | |
2991 | ||
2992 | static int do_write_locked_index(struct index_state *istate, struct lock_file *lock, | |
2993 | unsigned flags) | |
2994 | { | |
2995 | int ret = do_write_index(istate, lock->tempfile, 0); | |
2996 | if (ret) | |
2997 | return ret; | |
2998 | if (flags & COMMIT_LOCK) | |
2999 | return commit_locked_index(lock); | |
3000 | return close_lock_file_gently(lock); | |
3001 | } | |
3002 | ||
3003 | static int write_split_index(struct index_state *istate, | |
3004 | struct lock_file *lock, | |
3005 | unsigned flags) | |
3006 | { | |
3007 | int ret; | |
3008 | prepare_to_write_split_index(istate); | |
3009 | ret = do_write_locked_index(istate, lock, flags); | |
3010 | finish_writing_split_index(istate); | |
3011 | return ret; | |
3012 | } | |
3013 | ||
3014 | static const char *shared_index_expire = "2.weeks.ago"; | |
3015 | ||
3016 | static unsigned long get_shared_index_expire_date(void) | |
3017 | { | |
3018 | static unsigned long shared_index_expire_date; | |
3019 | static int shared_index_expire_date_prepared; | |
3020 | ||
3021 | if (!shared_index_expire_date_prepared) { | |
3022 | git_config_get_expiry("splitindex.sharedindexexpire", | |
3023 | &shared_index_expire); | |
3024 | shared_index_expire_date = approxidate(shared_index_expire); | |
3025 | shared_index_expire_date_prepared = 1; | |
3026 | } | |
3027 | ||
3028 | return shared_index_expire_date; | |
3029 | } | |
3030 | ||
3031 | static int should_delete_shared_index(const char *shared_index_path) | |
3032 | { | |
3033 | struct stat st; | |
3034 | unsigned long expiration; | |
3035 | ||
3036 | /* Check timestamp */ | |
3037 | expiration = get_shared_index_expire_date(); | |
3038 | if (!expiration) | |
3039 | return 0; | |
3040 | if (stat(shared_index_path, &st)) | |
3041 | return error_errno(_("could not stat '%s'"), shared_index_path); | |
3042 | if (st.st_mtime > expiration) | |
3043 | return 0; | |
3044 | ||
3045 | return 1; | |
3046 | } | |
3047 | ||
3048 | static int clean_shared_index_files(const char *current_hex) | |
3049 | { | |
3050 | struct dirent *de; | |
3051 | DIR *dir = opendir(get_git_dir()); | |
3052 | ||
3053 | if (!dir) | |
3054 | return error_errno(_("unable to open git dir: %s"), get_git_dir()); | |
3055 | ||
3056 | while ((de = readdir(dir)) != NULL) { | |
3057 | const char *sha1_hex; | |
3058 | const char *shared_index_path; | |
3059 | if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex)) | |
3060 | continue; | |
3061 | if (!strcmp(sha1_hex, current_hex)) | |
3062 | continue; | |
3063 | shared_index_path = git_path("%s", de->d_name); | |
3064 | if (should_delete_shared_index(shared_index_path) > 0 && | |
3065 | unlink(shared_index_path)) | |
3066 | warning_errno(_("unable to unlink: %s"), shared_index_path); | |
3067 | } | |
3068 | closedir(dir); | |
3069 | ||
3070 | return 0; | |
3071 | } | |
3072 | ||
3073 | static int write_shared_index(struct index_state *istate, | |
3074 | struct tempfile **temp) | |
3075 | { | |
3076 | struct split_index *si = istate->split_index; | |
3077 | int ret; | |
3078 | ||
3079 | move_cache_to_base_index(istate); | |
3080 | ret = do_write_index(si->base, *temp, 1); | |
3081 | if (ret) | |
3082 | return ret; | |
3083 | ret = adjust_shared_perm(get_tempfile_path(*temp)); | |
3084 | if (ret) { | |
3085 | error("cannot fix permission bits on %s", get_tempfile_path(*temp)); | |
3086 | return ret; | |
3087 | } | |
3088 | ret = rename_tempfile(temp, | |
3089 | git_path("sharedindex.%s", oid_to_hex(&si->base->oid))); | |
3090 | if (!ret) { | |
3091 | oidcpy(&si->base_oid, &si->base->oid); | |
3092 | clean_shared_index_files(oid_to_hex(&si->base->oid)); | |
3093 | } | |
3094 | ||
3095 | return ret; | |
3096 | } | |
3097 | ||
3098 | static const int default_max_percent_split_change = 20; | |
3099 | ||
3100 | static int too_many_not_shared_entries(struct index_state *istate) | |
3101 | { | |
3102 | int i, not_shared = 0; | |
3103 | int max_split = git_config_get_max_percent_split_change(); | |
3104 | ||
3105 | switch (max_split) { | |
3106 | case -1: | |
3107 | /* not or badly configured: use the default value */ | |
3108 | max_split = default_max_percent_split_change; | |
3109 | break; | |
3110 | case 0: | |
3111 | return 1; /* 0% means always write a new shared index */ | |
3112 | case 100: | |
3113 | return 0; /* 100% means never write a new shared index */ | |
3114 | default: | |
3115 | break; /* just use the configured value */ | |
3116 | } | |
3117 | ||
3118 | /* Count not shared entries */ | |
3119 | for (i = 0; i < istate->cache_nr; i++) { | |
3120 | struct cache_entry *ce = istate->cache[i]; | |
3121 | if (!ce->index) | |
3122 | not_shared++; | |
3123 | } | |
3124 | ||
3125 | return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100; | |
3126 | } | |
3127 | ||
3128 | int write_locked_index(struct index_state *istate, struct lock_file *lock, | |
3129 | unsigned flags) | |
3130 | { | |
3131 | int new_shared_index, ret; | |
3132 | struct split_index *si = istate->split_index; | |
3133 | ||
3134 | if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0)) | |
3135 | cache_tree_verify(istate); | |
3136 | ||
3137 | if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) { | |
3138 | if (flags & COMMIT_LOCK) | |
3139 | rollback_lock_file(lock); | |
3140 | return 0; | |
3141 | } | |
3142 | ||
3143 | if (istate->fsmonitor_last_update) | |
3144 | fill_fsmonitor_bitmap(istate); | |
3145 | ||
3146 | if (!si || alternate_index_output || | |
3147 | (istate->cache_changed & ~EXTMASK)) { | |
3148 | if (si) | |
3149 | oidclr(&si->base_oid); | |
3150 | ret = do_write_locked_index(istate, lock, flags); | |
3151 | goto out; | |
3152 | } | |
3153 | ||
3154 | if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) { | |
3155 | int v = si->base_oid.hash[0]; | |
3156 | if ((v & 15) < 6) | |
3157 | istate->cache_changed |= SPLIT_INDEX_ORDERED; | |
3158 | } | |
3159 | if (too_many_not_shared_entries(istate)) | |
3160 | istate->cache_changed |= SPLIT_INDEX_ORDERED; | |
3161 | ||
3162 | new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED; | |
3163 | ||
3164 | if (new_shared_index) { | |
3165 | struct tempfile *temp; | |
3166 | int saved_errno; | |
3167 | ||
3168 | temp = mks_tempfile(git_path("sharedindex_XXXXXX")); | |
3169 | if (!temp) { | |
3170 | oidclr(&si->base_oid); | |
3171 | ret = do_write_locked_index(istate, lock, flags); | |
3172 | goto out; | |
3173 | } | |
3174 | ret = write_shared_index(istate, &temp); | |
3175 | ||
3176 | saved_errno = errno; | |
3177 | if (is_tempfile_active(temp)) | |
3178 | delete_tempfile(&temp); | |
3179 | errno = saved_errno; | |
3180 | ||
3181 | if (ret) | |
3182 | goto out; | |
3183 | } | |
3184 | ||
3185 | ret = write_split_index(istate, lock, flags); | |
3186 | ||
3187 | /* Freshen the shared index only if the split-index was written */ | |
3188 | if (!ret && !new_shared_index) { | |
3189 | const char *shared_index = git_path("sharedindex.%s", | |
3190 | oid_to_hex(&si->base_oid)); | |
3191 | freshen_shared_index(shared_index, 1); | |
3192 | } | |
3193 | ||
3194 | out: | |
3195 | if (flags & COMMIT_LOCK) | |
3196 | rollback_lock_file(lock); | |
3197 | return ret; | |
3198 | } | |
3199 | ||
3200 | /* | |
3201 | * Read the index file that is potentially unmerged into given | |
3202 | * index_state, dropping any unmerged entries to stage #0 (potentially | |
3203 | * resulting in a path appearing as both a file and a directory in the | |
3204 | * index; the caller is responsible to clear out the extra entries | |
3205 | * before writing the index to a tree). Returns true if the index is | |
3206 | * unmerged. Callers who want to refuse to work from an unmerged | |
3207 | * state can call this and check its return value, instead of calling | |
3208 | * read_cache(). | |
3209 | */ | |
3210 | int read_index_unmerged(struct index_state *istate) | |
3211 | { | |
3212 | int i; | |
3213 | int unmerged = 0; | |
3214 | ||
3215 | read_index(istate); | |
3216 | for (i = 0; i < istate->cache_nr; i++) { | |
3217 | struct cache_entry *ce = istate->cache[i]; | |
3218 | struct cache_entry *new_ce; | |
3219 | int len; | |
3220 | ||
3221 | if (!ce_stage(ce)) | |
3222 | continue; | |
3223 | unmerged = 1; | |
3224 | len = ce_namelen(ce); | |
3225 | new_ce = make_empty_cache_entry(istate, len); | |
3226 | memcpy(new_ce->name, ce->name, len); | |
3227 | new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED; | |
3228 | new_ce->ce_namelen = len; | |
3229 | new_ce->ce_mode = ce->ce_mode; | |
3230 | if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK)) | |
3231 | return error("%s: cannot drop to stage #0", | |
3232 | new_ce->name); | |
3233 | } | |
3234 | return unmerged; | |
3235 | } | |
3236 | ||
3237 | /* | |
3238 | * Returns 1 if the path is an "other" path with respect to | |
3239 | * the index; that is, the path is not mentioned in the index at all, | |
3240 | * either as a file, a directory with some files in the index, | |
3241 | * or as an unmerged entry. | |
3242 | * | |
3243 | * We helpfully remove a trailing "/" from directories so that | |
3244 | * the output of read_directory can be used as-is. | |
3245 | */ | |
3246 | int index_name_is_other(const struct index_state *istate, const char *name, | |
3247 | int namelen) | |
3248 | { | |
3249 | int pos; | |
3250 | if (namelen && name[namelen - 1] == '/') | |
3251 | namelen--; | |
3252 | pos = index_name_pos(istate, name, namelen); | |
3253 | if (0 <= pos) | |
3254 | return 0; /* exact match */ | |
3255 | pos = -pos - 1; | |
3256 | if (pos < istate->cache_nr) { | |
3257 | struct cache_entry *ce = istate->cache[pos]; | |
3258 | if (ce_namelen(ce) == namelen && | |
3259 | !memcmp(ce->name, name, namelen)) | |
3260 | return 0; /* Yup, this one exists unmerged */ | |
3261 | } | |
3262 | return 1; | |
3263 | } | |
3264 | ||
3265 | void *read_blob_data_from_index(const struct index_state *istate, | |
3266 | const char *path, unsigned long *size) | |
3267 | { | |
3268 | int pos, len; | |
3269 | unsigned long sz; | |
3270 | enum object_type type; | |
3271 | void *data; | |
3272 | ||
3273 | len = strlen(path); | |
3274 | pos = index_name_pos(istate, path, len); | |
3275 | if (pos < 0) { | |
3276 | /* | |
3277 | * We might be in the middle of a merge, in which | |
3278 | * case we would read stage #2 (ours). | |
3279 | */ | |
3280 | int i; | |
3281 | for (i = -pos - 1; | |
3282 | (pos < 0 && i < istate->cache_nr && | |
3283 | !strcmp(istate->cache[i]->name, path)); | |
3284 | i++) | |
3285 | if (ce_stage(istate->cache[i]) == 2) | |
3286 | pos = i; | |
3287 | } | |
3288 | if (pos < 0) | |
3289 | return NULL; | |
3290 | data = read_object_file(&istate->cache[pos]->oid, &type, &sz); | |
3291 | if (!data || type != OBJ_BLOB) { | |
3292 | free(data); | |
3293 | return NULL; | |
3294 | } | |
3295 | if (size) | |
3296 | *size = sz; | |
3297 | return data; | |
3298 | } | |
3299 | ||
3300 | void stat_validity_clear(struct stat_validity *sv) | |
3301 | { | |
3302 | FREE_AND_NULL(sv->sd); | |
3303 | } | |
3304 | ||
3305 | int stat_validity_check(struct stat_validity *sv, const char *path) | |
3306 | { | |
3307 | struct stat st; | |
3308 | ||
3309 | if (stat(path, &st) < 0) | |
3310 | return sv->sd == NULL; | |
3311 | if (!sv->sd) | |
3312 | return 0; | |
3313 | return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st); | |
3314 | } | |
3315 | ||
3316 | void stat_validity_update(struct stat_validity *sv, int fd) | |
3317 | { | |
3318 | struct stat st; | |
3319 | ||
3320 | if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode)) | |
3321 | stat_validity_clear(sv); | |
3322 | else { | |
3323 | if (!sv->sd) | |
3324 | sv->sd = xcalloc(1, sizeof(struct stat_data)); | |
3325 | fill_stat_data(sv->sd, &st); | |
3326 | } | |
3327 | } | |
3328 | ||
3329 | void move_index_extensions(struct index_state *dst, struct index_state *src) | |
3330 | { | |
3331 | dst->untracked = src->untracked; | |
3332 | src->untracked = NULL; | |
3333 | dst->cache_tree = src->cache_tree; | |
3334 | src->cache_tree = NULL; | |
3335 | } | |
3336 | ||
3337 | struct cache_entry *dup_cache_entry(const struct cache_entry *ce, | |
3338 | struct index_state *istate) | |
3339 | { | |
3340 | unsigned int size = ce_size(ce); | |
3341 | int mem_pool_allocated; | |
3342 | struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce)); | |
3343 | mem_pool_allocated = new_entry->mem_pool_allocated; | |
3344 | ||
3345 | memcpy(new_entry, ce, size); | |
3346 | new_entry->mem_pool_allocated = mem_pool_allocated; | |
3347 | return new_entry; | |
3348 | } | |
3349 | ||
3350 | void discard_cache_entry(struct cache_entry *ce) | |
3351 | { | |
3352 | if (ce && should_validate_cache_entries()) | |
3353 | memset(ce, 0xCD, cache_entry_size(ce->ce_namelen)); | |
3354 | ||
3355 | if (ce && ce->mem_pool_allocated) | |
3356 | return; | |
3357 | ||
3358 | free(ce); | |
3359 | } | |
3360 | ||
3361 | int should_validate_cache_entries(void) | |
3362 | { | |
3363 | static int validate_index_cache_entries = -1; | |
3364 | ||
3365 | if (validate_index_cache_entries < 0) { | |
3366 | if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES")) | |
3367 | validate_index_cache_entries = 1; | |
3368 | else | |
3369 | validate_index_cache_entries = 0; | |
3370 | } | |
3371 | ||
3372 | return validate_index_cache_entries; | |
3373 | } | |
3374 | ||
3375 | #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */ | |
3376 | #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */ | |
3377 | ||
3378 | static size_t read_eoie_extension(const char *mmap, size_t mmap_size) | |
3379 | { | |
3380 | /* | |
3381 | * The end of index entries (EOIE) extension is guaranteed to be last | |
3382 | * so that it can be found by scanning backwards from the EOF. | |
3383 | * | |
3384 | * "EOIE" | |
3385 | * <4-byte length> | |
3386 | * <4-byte offset> | |
3387 | * <20-byte hash> | |
3388 | */ | |
3389 | const char *index, *eoie; | |
3390 | uint32_t extsize; | |
3391 | size_t offset, src_offset; | |
3392 | unsigned char hash[GIT_MAX_RAWSZ]; | |
3393 | git_hash_ctx c; | |
3394 | ||
3395 | /* ensure we have an index big enough to contain an EOIE extension */ | |
3396 | if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz) | |
3397 | return 0; | |
3398 | ||
3399 | /* validate the extension signature */ | |
3400 | index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz; | |
3401 | if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES) | |
3402 | return 0; | |
3403 | index += sizeof(uint32_t); | |
3404 | ||
3405 | /* validate the extension size */ | |
3406 | extsize = get_be32(index); | |
3407 | if (extsize != EOIE_SIZE) | |
3408 | return 0; | |
3409 | index += sizeof(uint32_t); | |
3410 | ||
3411 | /* | |
3412 | * Validate the offset we're going to look for the first extension | |
3413 | * signature is after the index header and before the eoie extension. | |
3414 | */ | |
3415 | offset = get_be32(index); | |
3416 | if (mmap + offset < mmap + sizeof(struct cache_header)) | |
3417 | return 0; | |
3418 | if (mmap + offset >= eoie) | |
3419 | return 0; | |
3420 | index += sizeof(uint32_t); | |
3421 | ||
3422 | /* | |
3423 | * The hash is computed over extension types and their sizes (but not | |
3424 | * their contents). E.g. if we have "TREE" extension that is N-bytes | |
3425 | * long, "REUC" extension that is M-bytes long, followed by "EOIE", | |
3426 | * then the hash would be: | |
3427 | * | |
3428 | * SHA-1("TREE" + <binary representation of N> + | |
3429 | * "REUC" + <binary representation of M>) | |
3430 | */ | |
3431 | src_offset = offset; | |
3432 | the_hash_algo->init_fn(&c); | |
3433 | while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) { | |
3434 | /* After an array of active_nr index entries, | |
3435 | * there can be arbitrary number of extended | |
3436 | * sections, each of which is prefixed with | |
3437 | * extension name (4-byte) and section length | |
3438 | * in 4-byte network byte order. | |
3439 | */ | |
3440 | uint32_t extsize; | |
3441 | memcpy(&extsize, mmap + src_offset + 4, 4); | |
3442 | extsize = ntohl(extsize); | |
3443 | ||
3444 | /* verify the extension size isn't so large it will wrap around */ | |
3445 | if (src_offset + 8 + extsize < src_offset) | |
3446 | return 0; | |
3447 | ||
3448 | the_hash_algo->update_fn(&c, mmap + src_offset, 8); | |
3449 | ||
3450 | src_offset += 8; | |
3451 | src_offset += extsize; | |
3452 | } | |
3453 | the_hash_algo->final_fn(hash, &c); | |
3454 | if (!hasheq(hash, (const unsigned char *)index)) | |
3455 | return 0; | |
3456 | ||
3457 | /* Validate that the extension offsets returned us back to the eoie extension. */ | |
3458 | if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) | |
3459 | return 0; | |
3460 | ||
3461 | return offset; | |
3462 | } | |
3463 | ||
3464 | static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset) | |
3465 | { | |
3466 | uint32_t buffer; | |
3467 | unsigned char hash[GIT_MAX_RAWSZ]; | |
3468 | ||
3469 | /* offset */ | |
3470 | put_be32(&buffer, offset); | |
3471 | strbuf_add(sb, &buffer, sizeof(uint32_t)); | |
3472 | ||
3473 | /* hash */ | |
3474 | the_hash_algo->final_fn(hash, eoie_context); | |
3475 | strbuf_add(sb, hash, the_hash_algo->rawsz); | |
3476 | } | |
3477 | ||
3478 | #ifndef NO_PTHREADS | |
3479 | #define IEOT_VERSION (1) | |
3480 | ||
3481 | static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset) | |
3482 | { | |
3483 | const char *index = NULL; | |
3484 | uint32_t extsize, ext_version; | |
3485 | struct index_entry_offset_table *ieot; | |
3486 | int i, nr; | |
3487 | ||
3488 | /* find the IEOT extension */ | |
3489 | if (!offset) | |
3490 | return NULL; | |
3491 | while (offset <= mmap_size - the_hash_algo->rawsz - 8) { | |
3492 | extsize = get_be32(mmap + offset + 4); | |
3493 | if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) { | |
3494 | index = mmap + offset + 4 + 4; | |
3495 | break; | |
3496 | } | |
3497 | offset += 8; | |
3498 | offset += extsize; | |
3499 | } | |
3500 | if (!index) | |
3501 | return NULL; | |
3502 | ||
3503 | /* validate the version is IEOT_VERSION */ | |
3504 | ext_version = get_be32(index); | |
3505 | if (ext_version != IEOT_VERSION) { | |
3506 | error("invalid IEOT version %d", ext_version); | |
3507 | return NULL; | |
3508 | } | |
3509 | index += sizeof(uint32_t); | |
3510 | ||
3511 | /* extension size - version bytes / bytes per entry */ | |
3512 | nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t)); | |
3513 | if (!nr) { | |
3514 | error("invalid number of IEOT entries %d", nr); | |
3515 | return NULL; | |
3516 | } | |
3517 | ieot = xmalloc(sizeof(struct index_entry_offset_table) | |
3518 | + (nr * sizeof(struct index_entry_offset))); | |
3519 | ieot->nr = nr; | |
3520 | for (i = 0; i < nr; i++) { | |
3521 | ieot->entries[i].offset = get_be32(index); | |
3522 | index += sizeof(uint32_t); | |
3523 | ieot->entries[i].nr = get_be32(index); | |
3524 | index += sizeof(uint32_t); | |
3525 | } | |
3526 | ||
3527 | return ieot; | |
3528 | } | |
3529 | ||
3530 | static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot) | |
3531 | { | |
3532 | uint32_t buffer; | |
3533 | int i; | |
3534 | ||
3535 | /* version */ | |
3536 | put_be32(&buffer, IEOT_VERSION); | |
3537 | strbuf_add(sb, &buffer, sizeof(uint32_t)); | |
3538 | ||
3539 | /* ieot */ | |
3540 | for (i = 0; i < ieot->nr; i++) { | |
3541 | ||
3542 | /* offset */ | |
3543 | put_be32(&buffer, ieot->entries[i].offset); | |
3544 | strbuf_add(sb, &buffer, sizeof(uint32_t)); | |
3545 | ||
3546 | /* count */ | |
3547 | put_be32(&buffer, ieot->entries[i].nr); | |
3548 | strbuf_add(sb, &buffer, sizeof(uint32_t)); | |
3549 | } | |
3550 | } | |
3551 | #endif |