]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * GIT - The information manager from hell | |
3 | * | |
4 | * Copyright (C) Linus Torvalds, 2005 | |
5 | * | |
6 | * This handles basic git sha1 object files - packing, unpacking, | |
7 | * creation etc. | |
8 | */ | |
9 | #include "cache.h" | |
10 | #include "config.h" | |
11 | #include "string-list.h" | |
12 | #include "lockfile.h" | |
13 | #include "delta.h" | |
14 | #include "pack.h" | |
15 | #include "blob.h" | |
16 | #include "commit.h" | |
17 | #include "run-command.h" | |
18 | #include "tag.h" | |
19 | #include "tree.h" | |
20 | #include "tree-walk.h" | |
21 | #include "refs.h" | |
22 | #include "pack-revindex.h" | |
23 | #include "sha1-lookup.h" | |
24 | #include "bulk-checkin.h" | |
25 | #include "streaming.h" | |
26 | #include "dir.h" | |
27 | #include "list.h" | |
28 | #include "mergesort.h" | |
29 | #include "quote.h" | |
30 | #include "packfile.h" | |
31 | #include "fetch-object.h" | |
32 | ||
33 | const unsigned char null_sha1[GIT_MAX_RAWSZ]; | |
34 | const struct object_id null_oid; | |
35 | const struct object_id empty_tree_oid = { | |
36 | EMPTY_TREE_SHA1_BIN_LITERAL | |
37 | }; | |
38 | const struct object_id empty_blob_oid = { | |
39 | EMPTY_BLOB_SHA1_BIN_LITERAL | |
40 | }; | |
41 | ||
42 | static void git_hash_sha1_init(git_hash_ctx *ctx) | |
43 | { | |
44 | git_SHA1_Init(&ctx->sha1); | |
45 | } | |
46 | ||
47 | static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len) | |
48 | { | |
49 | git_SHA1_Update(&ctx->sha1, data, len); | |
50 | } | |
51 | ||
52 | static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx) | |
53 | { | |
54 | git_SHA1_Final(hash, &ctx->sha1); | |
55 | } | |
56 | ||
57 | static void git_hash_unknown_init(git_hash_ctx *ctx) | |
58 | { | |
59 | die("trying to init unknown hash"); | |
60 | } | |
61 | ||
62 | static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len) | |
63 | { | |
64 | die("trying to update unknown hash"); | |
65 | } | |
66 | ||
67 | static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx) | |
68 | { | |
69 | die("trying to finalize unknown hash"); | |
70 | } | |
71 | ||
72 | const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = { | |
73 | { | |
74 | NULL, | |
75 | 0x00000000, | |
76 | 0, | |
77 | 0, | |
78 | git_hash_unknown_init, | |
79 | git_hash_unknown_update, | |
80 | git_hash_unknown_final, | |
81 | NULL, | |
82 | NULL, | |
83 | }, | |
84 | { | |
85 | "sha-1", | |
86 | /* "sha1", big-endian */ | |
87 | 0x73686131, | |
88 | GIT_SHA1_RAWSZ, | |
89 | GIT_SHA1_HEXSZ, | |
90 | git_hash_sha1_init, | |
91 | git_hash_sha1_update, | |
92 | git_hash_sha1_final, | |
93 | &empty_tree_oid, | |
94 | &empty_blob_oid, | |
95 | }, | |
96 | }; | |
97 | ||
98 | /* | |
99 | * This is meant to hold a *small* number of objects that you would | |
100 | * want read_sha1_file() to be able to return, but yet you do not want | |
101 | * to write them into the object store (e.g. a browse-only | |
102 | * application). | |
103 | */ | |
104 | static struct cached_object { | |
105 | unsigned char sha1[20]; | |
106 | enum object_type type; | |
107 | void *buf; | |
108 | unsigned long size; | |
109 | } *cached_objects; | |
110 | static int cached_object_nr, cached_object_alloc; | |
111 | ||
112 | static struct cached_object empty_tree = { | |
113 | EMPTY_TREE_SHA1_BIN_LITERAL, | |
114 | OBJ_TREE, | |
115 | "", | |
116 | 0 | |
117 | }; | |
118 | ||
119 | static struct cached_object *find_cached_object(const unsigned char *sha1) | |
120 | { | |
121 | int i; | |
122 | struct cached_object *co = cached_objects; | |
123 | ||
124 | for (i = 0; i < cached_object_nr; i++, co++) { | |
125 | if (!hashcmp(co->sha1, sha1)) | |
126 | return co; | |
127 | } | |
128 | if (!hashcmp(sha1, empty_tree.sha1)) | |
129 | return &empty_tree; | |
130 | return NULL; | |
131 | } | |
132 | ||
133 | ||
134 | static int get_conv_flags(unsigned flags) | |
135 | { | |
136 | if (flags & HASH_RENORMALIZE) | |
137 | return CONV_EOL_RENORMALIZE; | |
138 | else if (flags & HASH_WRITE_OBJECT) | |
139 | return global_conv_flags_eol; | |
140 | else | |
141 | return 0; | |
142 | } | |
143 | ||
144 | ||
145 | int mkdir_in_gitdir(const char *path) | |
146 | { | |
147 | if (mkdir(path, 0777)) { | |
148 | int saved_errno = errno; | |
149 | struct stat st; | |
150 | struct strbuf sb = STRBUF_INIT; | |
151 | ||
152 | if (errno != EEXIST) | |
153 | return -1; | |
154 | /* | |
155 | * Are we looking at a path in a symlinked worktree | |
156 | * whose original repository does not yet have it? | |
157 | * e.g. .git/rr-cache pointing at its original | |
158 | * repository in which the user hasn't performed any | |
159 | * conflict resolution yet? | |
160 | */ | |
161 | if (lstat(path, &st) || !S_ISLNK(st.st_mode) || | |
162 | strbuf_readlink(&sb, path, st.st_size) || | |
163 | !is_absolute_path(sb.buf) || | |
164 | mkdir(sb.buf, 0777)) { | |
165 | strbuf_release(&sb); | |
166 | errno = saved_errno; | |
167 | return -1; | |
168 | } | |
169 | strbuf_release(&sb); | |
170 | } | |
171 | return adjust_shared_perm(path); | |
172 | } | |
173 | ||
174 | enum scld_error safe_create_leading_directories(char *path) | |
175 | { | |
176 | char *next_component = path + offset_1st_component(path); | |
177 | enum scld_error ret = SCLD_OK; | |
178 | ||
179 | while (ret == SCLD_OK && next_component) { | |
180 | struct stat st; | |
181 | char *slash = next_component, slash_character; | |
182 | ||
183 | while (*slash && !is_dir_sep(*slash)) | |
184 | slash++; | |
185 | ||
186 | if (!*slash) | |
187 | break; | |
188 | ||
189 | next_component = slash + 1; | |
190 | while (is_dir_sep(*next_component)) | |
191 | next_component++; | |
192 | if (!*next_component) | |
193 | break; | |
194 | ||
195 | slash_character = *slash; | |
196 | *slash = '\0'; | |
197 | if (!stat(path, &st)) { | |
198 | /* path exists */ | |
199 | if (!S_ISDIR(st.st_mode)) { | |
200 | errno = ENOTDIR; | |
201 | ret = SCLD_EXISTS; | |
202 | } | |
203 | } else if (mkdir(path, 0777)) { | |
204 | if (errno == EEXIST && | |
205 | !stat(path, &st) && S_ISDIR(st.st_mode)) | |
206 | ; /* somebody created it since we checked */ | |
207 | else if (errno == ENOENT) | |
208 | /* | |
209 | * Either mkdir() failed because | |
210 | * somebody just pruned the containing | |
211 | * directory, or stat() failed because | |
212 | * the file that was in our way was | |
213 | * just removed. Either way, inform | |
214 | * the caller that it might be worth | |
215 | * trying again: | |
216 | */ | |
217 | ret = SCLD_VANISHED; | |
218 | else | |
219 | ret = SCLD_FAILED; | |
220 | } else if (adjust_shared_perm(path)) { | |
221 | ret = SCLD_PERMS; | |
222 | } | |
223 | *slash = slash_character; | |
224 | } | |
225 | return ret; | |
226 | } | |
227 | ||
228 | enum scld_error safe_create_leading_directories_const(const char *path) | |
229 | { | |
230 | int save_errno; | |
231 | /* path points to cache entries, so xstrdup before messing with it */ | |
232 | char *buf = xstrdup(path); | |
233 | enum scld_error result = safe_create_leading_directories(buf); | |
234 | ||
235 | save_errno = errno; | |
236 | free(buf); | |
237 | errno = save_errno; | |
238 | return result; | |
239 | } | |
240 | ||
241 | int raceproof_create_file(const char *path, create_file_fn fn, void *cb) | |
242 | { | |
243 | /* | |
244 | * The number of times we will try to remove empty directories | |
245 | * in the way of path. This is only 1 because if another | |
246 | * process is racily creating directories that conflict with | |
247 | * us, we don't want to fight against them. | |
248 | */ | |
249 | int remove_directories_remaining = 1; | |
250 | ||
251 | /* | |
252 | * The number of times that we will try to create the | |
253 | * directories containing path. We are willing to attempt this | |
254 | * more than once, because another process could be trying to | |
255 | * clean up empty directories at the same time as we are | |
256 | * trying to create them. | |
257 | */ | |
258 | int create_directories_remaining = 3; | |
259 | ||
260 | /* A scratch copy of path, filled lazily if we need it: */ | |
261 | struct strbuf path_copy = STRBUF_INIT; | |
262 | ||
263 | int ret, save_errno; | |
264 | ||
265 | /* Sanity check: */ | |
266 | assert(*path); | |
267 | ||
268 | retry_fn: | |
269 | ret = fn(path, cb); | |
270 | save_errno = errno; | |
271 | if (!ret) | |
272 | goto out; | |
273 | ||
274 | if (errno == EISDIR && remove_directories_remaining-- > 0) { | |
275 | /* | |
276 | * A directory is in the way. Maybe it is empty; try | |
277 | * to remove it: | |
278 | */ | |
279 | if (!path_copy.len) | |
280 | strbuf_addstr(&path_copy, path); | |
281 | ||
282 | if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY)) | |
283 | goto retry_fn; | |
284 | } else if (errno == ENOENT && create_directories_remaining-- > 0) { | |
285 | /* | |
286 | * Maybe the containing directory didn't exist, or | |
287 | * maybe it was just deleted by a process that is | |
288 | * racing with us to clean up empty directories. Try | |
289 | * to create it: | |
290 | */ | |
291 | enum scld_error scld_result; | |
292 | ||
293 | if (!path_copy.len) | |
294 | strbuf_addstr(&path_copy, path); | |
295 | ||
296 | do { | |
297 | scld_result = safe_create_leading_directories(path_copy.buf); | |
298 | if (scld_result == SCLD_OK) | |
299 | goto retry_fn; | |
300 | } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0); | |
301 | } | |
302 | ||
303 | out: | |
304 | strbuf_release(&path_copy); | |
305 | errno = save_errno; | |
306 | return ret; | |
307 | } | |
308 | ||
309 | static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1) | |
310 | { | |
311 | int i; | |
312 | for (i = 0; i < 20; i++) { | |
313 | static char hex[] = "0123456789abcdef"; | |
314 | unsigned int val = sha1[i]; | |
315 | strbuf_addch(buf, hex[val >> 4]); | |
316 | strbuf_addch(buf, hex[val & 0xf]); | |
317 | if (!i) | |
318 | strbuf_addch(buf, '/'); | |
319 | } | |
320 | } | |
321 | ||
322 | void sha1_file_name(struct strbuf *buf, const unsigned char *sha1) | |
323 | { | |
324 | strbuf_addstr(buf, get_object_directory()); | |
325 | strbuf_addch(buf, '/'); | |
326 | fill_sha1_path(buf, sha1); | |
327 | } | |
328 | ||
329 | struct strbuf *alt_scratch_buf(struct alternate_object_database *alt) | |
330 | { | |
331 | strbuf_setlen(&alt->scratch, alt->base_len); | |
332 | return &alt->scratch; | |
333 | } | |
334 | ||
335 | static const char *alt_sha1_path(struct alternate_object_database *alt, | |
336 | const unsigned char *sha1) | |
337 | { | |
338 | struct strbuf *buf = alt_scratch_buf(alt); | |
339 | fill_sha1_path(buf, sha1); | |
340 | return buf->buf; | |
341 | } | |
342 | ||
343 | struct alternate_object_database *alt_odb_list; | |
344 | static struct alternate_object_database **alt_odb_tail; | |
345 | ||
346 | /* | |
347 | * Return non-zero iff the path is usable as an alternate object database. | |
348 | */ | |
349 | static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir) | |
350 | { | |
351 | struct alternate_object_database *alt; | |
352 | ||
353 | /* Detect cases where alternate disappeared */ | |
354 | if (!is_directory(path->buf)) { | |
355 | error("object directory %s does not exist; " | |
356 | "check .git/objects/info/alternates.", | |
357 | path->buf); | |
358 | return 0; | |
359 | } | |
360 | ||
361 | /* | |
362 | * Prevent the common mistake of listing the same | |
363 | * thing twice, or object directory itself. | |
364 | */ | |
365 | for (alt = alt_odb_list; alt; alt = alt->next) { | |
366 | if (!fspathcmp(path->buf, alt->path)) | |
367 | return 0; | |
368 | } | |
369 | if (!fspathcmp(path->buf, normalized_objdir)) | |
370 | return 0; | |
371 | ||
372 | return 1; | |
373 | } | |
374 | ||
375 | /* | |
376 | * Prepare alternate object database registry. | |
377 | * | |
378 | * The variable alt_odb_list points at the list of struct | |
379 | * alternate_object_database. The elements on this list come from | |
380 | * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT | |
381 | * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates, | |
382 | * whose contents is similar to that environment variable but can be | |
383 | * LF separated. Its base points at a statically allocated buffer that | |
384 | * contains "/the/directory/corresponding/to/.git/objects/...", while | |
385 | * its name points just after the slash at the end of ".git/objects/" | |
386 | * in the example above, and has enough space to hold 40-byte hex | |
387 | * SHA1, an extra slash for the first level indirection, and the | |
388 | * terminating NUL. | |
389 | */ | |
390 | static void read_info_alternates(const char * relative_base, int depth); | |
391 | static int link_alt_odb_entry(const char *entry, const char *relative_base, | |
392 | int depth, const char *normalized_objdir) | |
393 | { | |
394 | struct alternate_object_database *ent; | |
395 | struct strbuf pathbuf = STRBUF_INIT; | |
396 | ||
397 | if (!is_absolute_path(entry) && relative_base) { | |
398 | strbuf_realpath(&pathbuf, relative_base, 1); | |
399 | strbuf_addch(&pathbuf, '/'); | |
400 | } | |
401 | strbuf_addstr(&pathbuf, entry); | |
402 | ||
403 | if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) { | |
404 | error("unable to normalize alternate object path: %s", | |
405 | pathbuf.buf); | |
406 | strbuf_release(&pathbuf); | |
407 | return -1; | |
408 | } | |
409 | ||
410 | /* | |
411 | * The trailing slash after the directory name is given by | |
412 | * this function at the end. Remove duplicates. | |
413 | */ | |
414 | while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/') | |
415 | strbuf_setlen(&pathbuf, pathbuf.len - 1); | |
416 | ||
417 | if (!alt_odb_usable(&pathbuf, normalized_objdir)) { | |
418 | strbuf_release(&pathbuf); | |
419 | return -1; | |
420 | } | |
421 | ||
422 | ent = alloc_alt_odb(pathbuf.buf); | |
423 | ||
424 | /* add the alternate entry */ | |
425 | *alt_odb_tail = ent; | |
426 | alt_odb_tail = &(ent->next); | |
427 | ent->next = NULL; | |
428 | ||
429 | /* recursively add alternates */ | |
430 | read_info_alternates(pathbuf.buf, depth + 1); | |
431 | ||
432 | strbuf_release(&pathbuf); | |
433 | return 0; | |
434 | } | |
435 | ||
436 | static const char *parse_alt_odb_entry(const char *string, | |
437 | int sep, | |
438 | struct strbuf *out) | |
439 | { | |
440 | const char *end; | |
441 | ||
442 | strbuf_reset(out); | |
443 | ||
444 | if (*string == '#') { | |
445 | /* comment; consume up to next separator */ | |
446 | end = strchrnul(string, sep); | |
447 | } else if (*string == '"' && !unquote_c_style(out, string, &end)) { | |
448 | /* | |
449 | * quoted path; unquote_c_style has copied the | |
450 | * data for us and set "end". Broken quoting (e.g., | |
451 | * an entry that doesn't end with a quote) falls | |
452 | * back to the unquoted case below. | |
453 | */ | |
454 | } else { | |
455 | /* normal, unquoted path */ | |
456 | end = strchrnul(string, sep); | |
457 | strbuf_add(out, string, end - string); | |
458 | } | |
459 | ||
460 | if (*end) | |
461 | end++; | |
462 | return end; | |
463 | } | |
464 | ||
465 | static void link_alt_odb_entries(const char *alt, int sep, | |
466 | const char *relative_base, int depth) | |
467 | { | |
468 | struct strbuf objdirbuf = STRBUF_INIT; | |
469 | struct strbuf entry = STRBUF_INIT; | |
470 | ||
471 | if (!alt || !*alt) | |
472 | return; | |
473 | ||
474 | if (depth > 5) { | |
475 | error("%s: ignoring alternate object stores, nesting too deep.", | |
476 | relative_base); | |
477 | return; | |
478 | } | |
479 | ||
480 | strbuf_add_absolute_path(&objdirbuf, get_object_directory()); | |
481 | if (strbuf_normalize_path(&objdirbuf) < 0) | |
482 | die("unable to normalize object directory: %s", | |
483 | objdirbuf.buf); | |
484 | ||
485 | while (*alt) { | |
486 | alt = parse_alt_odb_entry(alt, sep, &entry); | |
487 | if (!entry.len) | |
488 | continue; | |
489 | link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf); | |
490 | } | |
491 | strbuf_release(&entry); | |
492 | strbuf_release(&objdirbuf); | |
493 | } | |
494 | ||
495 | static void read_info_alternates(const char * relative_base, int depth) | |
496 | { | |
497 | char *path; | |
498 | struct strbuf buf = STRBUF_INIT; | |
499 | ||
500 | path = xstrfmt("%s/info/alternates", relative_base); | |
501 | if (strbuf_read_file(&buf, path, 1024) < 0) { | |
502 | warn_on_fopen_errors(path); | |
503 | free(path); | |
504 | return; | |
505 | } | |
506 | ||
507 | link_alt_odb_entries(buf.buf, '\n', relative_base, depth); | |
508 | strbuf_release(&buf); | |
509 | free(path); | |
510 | } | |
511 | ||
512 | struct alternate_object_database *alloc_alt_odb(const char *dir) | |
513 | { | |
514 | struct alternate_object_database *ent; | |
515 | ||
516 | FLEX_ALLOC_STR(ent, path, dir); | |
517 | strbuf_init(&ent->scratch, 0); | |
518 | strbuf_addf(&ent->scratch, "%s/", dir); | |
519 | ent->base_len = ent->scratch.len; | |
520 | ||
521 | return ent; | |
522 | } | |
523 | ||
524 | void add_to_alternates_file(const char *reference) | |
525 | { | |
526 | struct lock_file lock = LOCK_INIT; | |
527 | char *alts = git_pathdup("objects/info/alternates"); | |
528 | FILE *in, *out; | |
529 | int found = 0; | |
530 | ||
531 | hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR); | |
532 | out = fdopen_lock_file(&lock, "w"); | |
533 | if (!out) | |
534 | die_errno("unable to fdopen alternates lockfile"); | |
535 | ||
536 | in = fopen(alts, "r"); | |
537 | if (in) { | |
538 | struct strbuf line = STRBUF_INIT; | |
539 | ||
540 | while (strbuf_getline(&line, in) != EOF) { | |
541 | if (!strcmp(reference, line.buf)) { | |
542 | found = 1; | |
543 | break; | |
544 | } | |
545 | fprintf_or_die(out, "%s\n", line.buf); | |
546 | } | |
547 | ||
548 | strbuf_release(&line); | |
549 | fclose(in); | |
550 | } | |
551 | else if (errno != ENOENT) | |
552 | die_errno("unable to read alternates file"); | |
553 | ||
554 | if (found) { | |
555 | rollback_lock_file(&lock); | |
556 | } else { | |
557 | fprintf_or_die(out, "%s\n", reference); | |
558 | if (commit_lock_file(&lock)) | |
559 | die_errno("unable to move new alternates file into place"); | |
560 | if (alt_odb_tail) | |
561 | link_alt_odb_entries(reference, '\n', NULL, 0); | |
562 | } | |
563 | free(alts); | |
564 | } | |
565 | ||
566 | void add_to_alternates_memory(const char *reference) | |
567 | { | |
568 | /* | |
569 | * Make sure alternates are initialized, or else our entry may be | |
570 | * overwritten when they are. | |
571 | */ | |
572 | prepare_alt_odb(); | |
573 | ||
574 | link_alt_odb_entries(reference, '\n', NULL, 0); | |
575 | } | |
576 | ||
577 | /* | |
578 | * Compute the exact path an alternate is at and returns it. In case of | |
579 | * error NULL is returned and the human readable error is added to `err` | |
580 | * `path` may be relative and should point to $GITDIR. | |
581 | * `err` must not be null. | |
582 | */ | |
583 | char *compute_alternate_path(const char *path, struct strbuf *err) | |
584 | { | |
585 | char *ref_git = NULL; | |
586 | const char *repo, *ref_git_s; | |
587 | int seen_error = 0; | |
588 | ||
589 | ref_git_s = real_path_if_valid(path); | |
590 | if (!ref_git_s) { | |
591 | seen_error = 1; | |
592 | strbuf_addf(err, _("path '%s' does not exist"), path); | |
593 | goto out; | |
594 | } else | |
595 | /* | |
596 | * Beware: read_gitfile(), real_path() and mkpath() | |
597 | * return static buffer | |
598 | */ | |
599 | ref_git = xstrdup(ref_git_s); | |
600 | ||
601 | repo = read_gitfile(ref_git); | |
602 | if (!repo) | |
603 | repo = read_gitfile(mkpath("%s/.git", ref_git)); | |
604 | if (repo) { | |
605 | free(ref_git); | |
606 | ref_git = xstrdup(repo); | |
607 | } | |
608 | ||
609 | if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) { | |
610 | char *ref_git_git = mkpathdup("%s/.git", ref_git); | |
611 | free(ref_git); | |
612 | ref_git = ref_git_git; | |
613 | } else if (!is_directory(mkpath("%s/objects", ref_git))) { | |
614 | struct strbuf sb = STRBUF_INIT; | |
615 | seen_error = 1; | |
616 | if (get_common_dir(&sb, ref_git)) { | |
617 | strbuf_addf(err, | |
618 | _("reference repository '%s' as a linked " | |
619 | "checkout is not supported yet."), | |
620 | path); | |
621 | goto out; | |
622 | } | |
623 | ||
624 | strbuf_addf(err, _("reference repository '%s' is not a " | |
625 | "local repository."), path); | |
626 | goto out; | |
627 | } | |
628 | ||
629 | if (!access(mkpath("%s/shallow", ref_git), F_OK)) { | |
630 | strbuf_addf(err, _("reference repository '%s' is shallow"), | |
631 | path); | |
632 | seen_error = 1; | |
633 | goto out; | |
634 | } | |
635 | ||
636 | if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) { | |
637 | strbuf_addf(err, | |
638 | _("reference repository '%s' is grafted"), | |
639 | path); | |
640 | seen_error = 1; | |
641 | goto out; | |
642 | } | |
643 | ||
644 | out: | |
645 | if (seen_error) { | |
646 | FREE_AND_NULL(ref_git); | |
647 | } | |
648 | ||
649 | return ref_git; | |
650 | } | |
651 | ||
652 | int foreach_alt_odb(alt_odb_fn fn, void *cb) | |
653 | { | |
654 | struct alternate_object_database *ent; | |
655 | int r = 0; | |
656 | ||
657 | prepare_alt_odb(); | |
658 | for (ent = alt_odb_list; ent; ent = ent->next) { | |
659 | r = fn(ent, cb); | |
660 | if (r) | |
661 | break; | |
662 | } | |
663 | return r; | |
664 | } | |
665 | ||
666 | void prepare_alt_odb(void) | |
667 | { | |
668 | const char *alt; | |
669 | ||
670 | if (alt_odb_tail) | |
671 | return; | |
672 | ||
673 | alt = getenv(ALTERNATE_DB_ENVIRONMENT); | |
674 | ||
675 | alt_odb_tail = &alt_odb_list; | |
676 | link_alt_odb_entries(alt, PATH_SEP, NULL, 0); | |
677 | ||
678 | read_info_alternates(get_object_directory(), 0); | |
679 | } | |
680 | ||
681 | /* Returns 1 if we have successfully freshened the file, 0 otherwise. */ | |
682 | static int freshen_file(const char *fn) | |
683 | { | |
684 | struct utimbuf t; | |
685 | t.actime = t.modtime = time(NULL); | |
686 | return !utime(fn, &t); | |
687 | } | |
688 | ||
689 | /* | |
690 | * All of the check_and_freshen functions return 1 if the file exists and was | |
691 | * freshened (if freshening was requested), 0 otherwise. If they return | |
692 | * 0, you should not assume that it is safe to skip a write of the object (it | |
693 | * either does not exist on disk, or has a stale mtime and may be subject to | |
694 | * pruning). | |
695 | */ | |
696 | int check_and_freshen_file(const char *fn, int freshen) | |
697 | { | |
698 | if (access(fn, F_OK)) | |
699 | return 0; | |
700 | if (freshen && !freshen_file(fn)) | |
701 | return 0; | |
702 | return 1; | |
703 | } | |
704 | ||
705 | static int check_and_freshen_local(const unsigned char *sha1, int freshen) | |
706 | { | |
707 | static struct strbuf buf = STRBUF_INIT; | |
708 | ||
709 | strbuf_reset(&buf); | |
710 | sha1_file_name(&buf, sha1); | |
711 | ||
712 | return check_and_freshen_file(buf.buf, freshen); | |
713 | } | |
714 | ||
715 | static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen) | |
716 | { | |
717 | struct alternate_object_database *alt; | |
718 | prepare_alt_odb(); | |
719 | for (alt = alt_odb_list; alt; alt = alt->next) { | |
720 | const char *path = alt_sha1_path(alt, sha1); | |
721 | if (check_and_freshen_file(path, freshen)) | |
722 | return 1; | |
723 | } | |
724 | return 0; | |
725 | } | |
726 | ||
727 | static int check_and_freshen(const unsigned char *sha1, int freshen) | |
728 | { | |
729 | return check_and_freshen_local(sha1, freshen) || | |
730 | check_and_freshen_nonlocal(sha1, freshen); | |
731 | } | |
732 | ||
733 | int has_loose_object_nonlocal(const unsigned char *sha1) | |
734 | { | |
735 | return check_and_freshen_nonlocal(sha1, 0); | |
736 | } | |
737 | ||
738 | static int has_loose_object(const unsigned char *sha1) | |
739 | { | |
740 | return check_and_freshen(sha1, 0); | |
741 | } | |
742 | ||
743 | static void mmap_limit_check(size_t length) | |
744 | { | |
745 | static size_t limit = 0; | |
746 | if (!limit) { | |
747 | limit = git_env_ulong("GIT_MMAP_LIMIT", 0); | |
748 | if (!limit) | |
749 | limit = SIZE_MAX; | |
750 | } | |
751 | if (length > limit) | |
752 | die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX, | |
753 | (uintmax_t)length, (uintmax_t)limit); | |
754 | } | |
755 | ||
756 | void *xmmap_gently(void *start, size_t length, | |
757 | int prot, int flags, int fd, off_t offset) | |
758 | { | |
759 | void *ret; | |
760 | ||
761 | mmap_limit_check(length); | |
762 | ret = mmap(start, length, prot, flags, fd, offset); | |
763 | if (ret == MAP_FAILED) { | |
764 | if (!length) | |
765 | return NULL; | |
766 | release_pack_memory(length); | |
767 | ret = mmap(start, length, prot, flags, fd, offset); | |
768 | } | |
769 | return ret; | |
770 | } | |
771 | ||
772 | void *xmmap(void *start, size_t length, | |
773 | int prot, int flags, int fd, off_t offset) | |
774 | { | |
775 | void *ret = xmmap_gently(start, length, prot, flags, fd, offset); | |
776 | if (ret == MAP_FAILED) | |
777 | die_errno("mmap failed"); | |
778 | return ret; | |
779 | } | |
780 | ||
781 | /* | |
782 | * With an in-core object data in "map", rehash it to make sure the | |
783 | * object name actually matches "sha1" to detect object corruption. | |
784 | * With "map" == NULL, try reading the object named with "sha1" using | |
785 | * the streaming interface and rehash it to do the same. | |
786 | */ | |
787 | int check_sha1_signature(const unsigned char *sha1, void *map, | |
788 | unsigned long size, const char *type) | |
789 | { | |
790 | struct object_id real_oid; | |
791 | enum object_type obj_type; | |
792 | struct git_istream *st; | |
793 | git_hash_ctx c; | |
794 | char hdr[32]; | |
795 | int hdrlen; | |
796 | ||
797 | if (map) { | |
798 | hash_object_file(map, size, type, &real_oid); | |
799 | return hashcmp(sha1, real_oid.hash) ? -1 : 0; | |
800 | } | |
801 | ||
802 | st = open_istream(sha1, &obj_type, &size, NULL); | |
803 | if (!st) | |
804 | return -1; | |
805 | ||
806 | /* Generate the header */ | |
807 | hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1; | |
808 | ||
809 | /* Sha1.. */ | |
810 | the_hash_algo->init_fn(&c); | |
811 | the_hash_algo->update_fn(&c, hdr, hdrlen); | |
812 | for (;;) { | |
813 | char buf[1024 * 16]; | |
814 | ssize_t readlen = read_istream(st, buf, sizeof(buf)); | |
815 | ||
816 | if (readlen < 0) { | |
817 | close_istream(st); | |
818 | return -1; | |
819 | } | |
820 | if (!readlen) | |
821 | break; | |
822 | the_hash_algo->update_fn(&c, buf, readlen); | |
823 | } | |
824 | the_hash_algo->final_fn(real_oid.hash, &c); | |
825 | close_istream(st); | |
826 | return hashcmp(sha1, real_oid.hash) ? -1 : 0; | |
827 | } | |
828 | ||
829 | int git_open_cloexec(const char *name, int flags) | |
830 | { | |
831 | int fd; | |
832 | static int o_cloexec = O_CLOEXEC; | |
833 | ||
834 | fd = open(name, flags | o_cloexec); | |
835 | if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) { | |
836 | /* Try again w/o O_CLOEXEC: the kernel might not support it */ | |
837 | o_cloexec &= ~O_CLOEXEC; | |
838 | fd = open(name, flags | o_cloexec); | |
839 | } | |
840 | ||
841 | #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC) | |
842 | { | |
843 | static int fd_cloexec = FD_CLOEXEC; | |
844 | ||
845 | if (!o_cloexec && 0 <= fd && fd_cloexec) { | |
846 | /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */ | |
847 | int flags = fcntl(fd, F_GETFD); | |
848 | if (fcntl(fd, F_SETFD, flags | fd_cloexec)) | |
849 | fd_cloexec = 0; | |
850 | } | |
851 | } | |
852 | #endif | |
853 | return fd; | |
854 | } | |
855 | ||
856 | /* | |
857 | * Find "sha1" as a loose object in the local repository or in an alternate. | |
858 | * Returns 0 on success, negative on failure. | |
859 | * | |
860 | * The "path" out-parameter will give the path of the object we found (if any). | |
861 | * Note that it may point to static storage and is only valid until another | |
862 | * call to sha1_file_name(), etc. | |
863 | */ | |
864 | static int stat_sha1_file(const unsigned char *sha1, struct stat *st, | |
865 | const char **path) | |
866 | { | |
867 | struct alternate_object_database *alt; | |
868 | static struct strbuf buf = STRBUF_INIT; | |
869 | ||
870 | strbuf_reset(&buf); | |
871 | sha1_file_name(&buf, sha1); | |
872 | *path = buf.buf; | |
873 | ||
874 | if (!lstat(*path, st)) | |
875 | return 0; | |
876 | ||
877 | prepare_alt_odb(); | |
878 | errno = ENOENT; | |
879 | for (alt = alt_odb_list; alt; alt = alt->next) { | |
880 | *path = alt_sha1_path(alt, sha1); | |
881 | if (!lstat(*path, st)) | |
882 | return 0; | |
883 | } | |
884 | ||
885 | return -1; | |
886 | } | |
887 | ||
888 | /* | |
889 | * Like stat_sha1_file(), but actually open the object and return the | |
890 | * descriptor. See the caveats on the "path" parameter above. | |
891 | */ | |
892 | static int open_sha1_file(const unsigned char *sha1, const char **path) | |
893 | { | |
894 | int fd; | |
895 | struct alternate_object_database *alt; | |
896 | int most_interesting_errno; | |
897 | static struct strbuf buf = STRBUF_INIT; | |
898 | ||
899 | strbuf_reset(&buf); | |
900 | sha1_file_name(&buf, sha1); | |
901 | *path = buf.buf; | |
902 | ||
903 | fd = git_open(*path); | |
904 | if (fd >= 0) | |
905 | return fd; | |
906 | most_interesting_errno = errno; | |
907 | ||
908 | prepare_alt_odb(); | |
909 | for (alt = alt_odb_list; alt; alt = alt->next) { | |
910 | *path = alt_sha1_path(alt, sha1); | |
911 | fd = git_open(*path); | |
912 | if (fd >= 0) | |
913 | return fd; | |
914 | if (most_interesting_errno == ENOENT) | |
915 | most_interesting_errno = errno; | |
916 | } | |
917 | errno = most_interesting_errno; | |
918 | return -1; | |
919 | } | |
920 | ||
921 | /* | |
922 | * Map the loose object at "path" if it is not NULL, or the path found by | |
923 | * searching for a loose object named "sha1". | |
924 | */ | |
925 | static void *map_sha1_file_1(const char *path, | |
926 | const unsigned char *sha1, | |
927 | unsigned long *size) | |
928 | { | |
929 | void *map; | |
930 | int fd; | |
931 | ||
932 | if (path) | |
933 | fd = git_open(path); | |
934 | else | |
935 | fd = open_sha1_file(sha1, &path); | |
936 | map = NULL; | |
937 | if (fd >= 0) { | |
938 | struct stat st; | |
939 | ||
940 | if (!fstat(fd, &st)) { | |
941 | *size = xsize_t(st.st_size); | |
942 | if (!*size) { | |
943 | /* mmap() is forbidden on empty files */ | |
944 | error("object file %s is empty", path); | |
945 | return NULL; | |
946 | } | |
947 | map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0); | |
948 | } | |
949 | close(fd); | |
950 | } | |
951 | return map; | |
952 | } | |
953 | ||
954 | void *map_sha1_file(const unsigned char *sha1, unsigned long *size) | |
955 | { | |
956 | return map_sha1_file_1(NULL, sha1, size); | |
957 | } | |
958 | ||
959 | static int unpack_sha1_short_header(git_zstream *stream, | |
960 | unsigned char *map, unsigned long mapsize, | |
961 | void *buffer, unsigned long bufsiz) | |
962 | { | |
963 | /* Get the data stream */ | |
964 | memset(stream, 0, sizeof(*stream)); | |
965 | stream->next_in = map; | |
966 | stream->avail_in = mapsize; | |
967 | stream->next_out = buffer; | |
968 | stream->avail_out = bufsiz; | |
969 | ||
970 | git_inflate_init(stream); | |
971 | return git_inflate(stream, 0); | |
972 | } | |
973 | ||
974 | int unpack_sha1_header(git_zstream *stream, | |
975 | unsigned char *map, unsigned long mapsize, | |
976 | void *buffer, unsigned long bufsiz) | |
977 | { | |
978 | int status = unpack_sha1_short_header(stream, map, mapsize, | |
979 | buffer, bufsiz); | |
980 | ||
981 | if (status < Z_OK) | |
982 | return status; | |
983 | ||
984 | /* Make sure we have the terminating NUL */ | |
985 | if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer)) | |
986 | return -1; | |
987 | return 0; | |
988 | } | |
989 | ||
990 | static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map, | |
991 | unsigned long mapsize, void *buffer, | |
992 | unsigned long bufsiz, struct strbuf *header) | |
993 | { | |
994 | int status; | |
995 | ||
996 | status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz); | |
997 | if (status < Z_OK) | |
998 | return -1; | |
999 | ||
1000 | /* | |
1001 | * Check if entire header is unpacked in the first iteration. | |
1002 | */ | |
1003 | if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer)) | |
1004 | return 0; | |
1005 | ||
1006 | /* | |
1007 | * buffer[0..bufsiz] was not large enough. Copy the partial | |
1008 | * result out to header, and then append the result of further | |
1009 | * reading the stream. | |
1010 | */ | |
1011 | strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer); | |
1012 | stream->next_out = buffer; | |
1013 | stream->avail_out = bufsiz; | |
1014 | ||
1015 | do { | |
1016 | status = git_inflate(stream, 0); | |
1017 | strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer); | |
1018 | if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer)) | |
1019 | return 0; | |
1020 | stream->next_out = buffer; | |
1021 | stream->avail_out = bufsiz; | |
1022 | } while (status != Z_STREAM_END); | |
1023 | return -1; | |
1024 | } | |
1025 | ||
1026 | static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1) | |
1027 | { | |
1028 | int bytes = strlen(buffer) + 1; | |
1029 | unsigned char *buf = xmallocz(size); | |
1030 | unsigned long n; | |
1031 | int status = Z_OK; | |
1032 | ||
1033 | n = stream->total_out - bytes; | |
1034 | if (n > size) | |
1035 | n = size; | |
1036 | memcpy(buf, (char *) buffer + bytes, n); | |
1037 | bytes = n; | |
1038 | if (bytes <= size) { | |
1039 | /* | |
1040 | * The above condition must be (bytes <= size), not | |
1041 | * (bytes < size). In other words, even though we | |
1042 | * expect no more output and set avail_out to zero, | |
1043 | * the input zlib stream may have bytes that express | |
1044 | * "this concludes the stream", and we *do* want to | |
1045 | * eat that input. | |
1046 | * | |
1047 | * Otherwise we would not be able to test that we | |
1048 | * consumed all the input to reach the expected size; | |
1049 | * we also want to check that zlib tells us that all | |
1050 | * went well with status == Z_STREAM_END at the end. | |
1051 | */ | |
1052 | stream->next_out = buf + bytes; | |
1053 | stream->avail_out = size - bytes; | |
1054 | while (status == Z_OK) | |
1055 | status = git_inflate(stream, Z_FINISH); | |
1056 | } | |
1057 | if (status == Z_STREAM_END && !stream->avail_in) { | |
1058 | git_inflate_end(stream); | |
1059 | return buf; | |
1060 | } | |
1061 | ||
1062 | if (status < 0) | |
1063 | error("corrupt loose object '%s'", sha1_to_hex(sha1)); | |
1064 | else if (stream->avail_in) | |
1065 | error("garbage at end of loose object '%s'", | |
1066 | sha1_to_hex(sha1)); | |
1067 | free(buf); | |
1068 | return NULL; | |
1069 | } | |
1070 | ||
1071 | /* | |
1072 | * We used to just use "sscanf()", but that's actually way | |
1073 | * too permissive for what we want to check. So do an anal | |
1074 | * object header parse by hand. | |
1075 | */ | |
1076 | static int parse_sha1_header_extended(const char *hdr, struct object_info *oi, | |
1077 | unsigned int flags) | |
1078 | { | |
1079 | const char *type_buf = hdr; | |
1080 | unsigned long size; | |
1081 | int type, type_len = 0; | |
1082 | ||
1083 | /* | |
1084 | * The type can be of any size but is followed by | |
1085 | * a space. | |
1086 | */ | |
1087 | for (;;) { | |
1088 | char c = *hdr++; | |
1089 | if (!c) | |
1090 | return -1; | |
1091 | if (c == ' ') | |
1092 | break; | |
1093 | type_len++; | |
1094 | } | |
1095 | ||
1096 | type = type_from_string_gently(type_buf, type_len, 1); | |
1097 | if (oi->type_name) | |
1098 | strbuf_add(oi->type_name, type_buf, type_len); | |
1099 | /* | |
1100 | * Set type to 0 if its an unknown object and | |
1101 | * we're obtaining the type using '--allow-unknown-type' | |
1102 | * option. | |
1103 | */ | |
1104 | if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0)) | |
1105 | type = 0; | |
1106 | else if (type < 0) | |
1107 | die("invalid object type"); | |
1108 | if (oi->typep) | |
1109 | *oi->typep = type; | |
1110 | ||
1111 | /* | |
1112 | * The length must follow immediately, and be in canonical | |
1113 | * decimal format (ie "010" is not valid). | |
1114 | */ | |
1115 | size = *hdr++ - '0'; | |
1116 | if (size > 9) | |
1117 | return -1; | |
1118 | if (size) { | |
1119 | for (;;) { | |
1120 | unsigned long c = *hdr - '0'; | |
1121 | if (c > 9) | |
1122 | break; | |
1123 | hdr++; | |
1124 | size = size * 10 + c; | |
1125 | } | |
1126 | } | |
1127 | ||
1128 | if (oi->sizep) | |
1129 | *oi->sizep = size; | |
1130 | ||
1131 | /* | |
1132 | * The length must be followed by a zero byte | |
1133 | */ | |
1134 | return *hdr ? -1 : type; | |
1135 | } | |
1136 | ||
1137 | int parse_sha1_header(const char *hdr, unsigned long *sizep) | |
1138 | { | |
1139 | struct object_info oi = OBJECT_INFO_INIT; | |
1140 | ||
1141 | oi.sizep = sizep; | |
1142 | return parse_sha1_header_extended(hdr, &oi, 0); | |
1143 | } | |
1144 | ||
1145 | static int sha1_loose_object_info(const unsigned char *sha1, | |
1146 | struct object_info *oi, | |
1147 | int flags) | |
1148 | { | |
1149 | int status = 0; | |
1150 | unsigned long mapsize; | |
1151 | void *map; | |
1152 | git_zstream stream; | |
1153 | char hdr[32]; | |
1154 | struct strbuf hdrbuf = STRBUF_INIT; | |
1155 | unsigned long size_scratch; | |
1156 | ||
1157 | if (oi->delta_base_sha1) | |
1158 | hashclr(oi->delta_base_sha1); | |
1159 | ||
1160 | /* | |
1161 | * If we don't care about type or size, then we don't | |
1162 | * need to look inside the object at all. Note that we | |
1163 | * do not optimize out the stat call, even if the | |
1164 | * caller doesn't care about the disk-size, since our | |
1165 | * return value implicitly indicates whether the | |
1166 | * object even exists. | |
1167 | */ | |
1168 | if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) { | |
1169 | const char *path; | |
1170 | struct stat st; | |
1171 | if (stat_sha1_file(sha1, &st, &path) < 0) | |
1172 | return -1; | |
1173 | if (oi->disk_sizep) | |
1174 | *oi->disk_sizep = st.st_size; | |
1175 | return 0; | |
1176 | } | |
1177 | ||
1178 | map = map_sha1_file(sha1, &mapsize); | |
1179 | if (!map) | |
1180 | return -1; | |
1181 | ||
1182 | if (!oi->sizep) | |
1183 | oi->sizep = &size_scratch; | |
1184 | ||
1185 | if (oi->disk_sizep) | |
1186 | *oi->disk_sizep = mapsize; | |
1187 | if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) { | |
1188 | if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0) | |
1189 | status = error("unable to unpack %s header with --allow-unknown-type", | |
1190 | sha1_to_hex(sha1)); | |
1191 | } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) | |
1192 | status = error("unable to unpack %s header", | |
1193 | sha1_to_hex(sha1)); | |
1194 | if (status < 0) | |
1195 | ; /* Do nothing */ | |
1196 | else if (hdrbuf.len) { | |
1197 | if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0) | |
1198 | status = error("unable to parse %s header with --allow-unknown-type", | |
1199 | sha1_to_hex(sha1)); | |
1200 | } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0) | |
1201 | status = error("unable to parse %s header", sha1_to_hex(sha1)); | |
1202 | ||
1203 | if (status >= 0 && oi->contentp) { | |
1204 | *oi->contentp = unpack_sha1_rest(&stream, hdr, | |
1205 | *oi->sizep, sha1); | |
1206 | if (!*oi->contentp) { | |
1207 | git_inflate_end(&stream); | |
1208 | status = -1; | |
1209 | } | |
1210 | } else | |
1211 | git_inflate_end(&stream); | |
1212 | ||
1213 | munmap(map, mapsize); | |
1214 | if (status && oi->typep) | |
1215 | *oi->typep = status; | |
1216 | if (oi->sizep == &size_scratch) | |
1217 | oi->sizep = NULL; | |
1218 | strbuf_release(&hdrbuf); | |
1219 | oi->whence = OI_LOOSE; | |
1220 | return (status < 0) ? status : 0; | |
1221 | } | |
1222 | ||
1223 | int fetch_if_missing = 1; | |
1224 | ||
1225 | int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags) | |
1226 | { | |
1227 | static struct object_info blank_oi = OBJECT_INFO_INIT; | |
1228 | struct pack_entry e; | |
1229 | int rtype; | |
1230 | const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ? | |
1231 | lookup_replace_object(sha1) : | |
1232 | sha1; | |
1233 | int already_retried = 0; | |
1234 | ||
1235 | if (is_null_sha1(real)) | |
1236 | return -1; | |
1237 | ||
1238 | if (!oi) | |
1239 | oi = &blank_oi; | |
1240 | ||
1241 | if (!(flags & OBJECT_INFO_SKIP_CACHED)) { | |
1242 | struct cached_object *co = find_cached_object(real); | |
1243 | if (co) { | |
1244 | if (oi->typep) | |
1245 | *(oi->typep) = co->type; | |
1246 | if (oi->sizep) | |
1247 | *(oi->sizep) = co->size; | |
1248 | if (oi->disk_sizep) | |
1249 | *(oi->disk_sizep) = 0; | |
1250 | if (oi->delta_base_sha1) | |
1251 | hashclr(oi->delta_base_sha1); | |
1252 | if (oi->type_name) | |
1253 | strbuf_addstr(oi->type_name, type_name(co->type)); | |
1254 | if (oi->contentp) | |
1255 | *oi->contentp = xmemdupz(co->buf, co->size); | |
1256 | oi->whence = OI_CACHED; | |
1257 | return 0; | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | while (1) { | |
1262 | if (find_pack_entry(real, &e)) | |
1263 | break; | |
1264 | ||
1265 | /* Most likely it's a loose object. */ | |
1266 | if (!sha1_loose_object_info(real, oi, flags)) | |
1267 | return 0; | |
1268 | ||
1269 | /* Not a loose object; someone else may have just packed it. */ | |
1270 | if (!(flags & OBJECT_INFO_QUICK)) { | |
1271 | reprepare_packed_git(); | |
1272 | if (find_pack_entry(real, &e)) | |
1273 | break; | |
1274 | } | |
1275 | ||
1276 | /* Check if it is a missing object */ | |
1277 | if (fetch_if_missing && repository_format_partial_clone && | |
1278 | !already_retried) { | |
1279 | /* | |
1280 | * TODO Investigate haveing fetch_object() return | |
1281 | * TODO error/success and stopping the music here. | |
1282 | */ | |
1283 | fetch_object(repository_format_partial_clone, real); | |
1284 | already_retried = 1; | |
1285 | continue; | |
1286 | } | |
1287 | ||
1288 | return -1; | |
1289 | } | |
1290 | ||
1291 | if (oi == &blank_oi) | |
1292 | /* | |
1293 | * We know that the caller doesn't actually need the | |
1294 | * information below, so return early. | |
1295 | */ | |
1296 | return 0; | |
1297 | rtype = packed_object_info(e.p, e.offset, oi); | |
1298 | if (rtype < 0) { | |
1299 | mark_bad_packed_object(e.p, real); | |
1300 | return sha1_object_info_extended(real, oi, 0); | |
1301 | } else if (oi->whence == OI_PACKED) { | |
1302 | oi->u.packed.offset = e.offset; | |
1303 | oi->u.packed.pack = e.p; | |
1304 | oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA || | |
1305 | rtype == OBJ_OFS_DELTA); | |
1306 | } | |
1307 | ||
1308 | return 0; | |
1309 | } | |
1310 | ||
1311 | /* returns enum object_type or negative */ | |
1312 | int sha1_object_info(const unsigned char *sha1, unsigned long *sizep) | |
1313 | { | |
1314 | enum object_type type; | |
1315 | struct object_info oi = OBJECT_INFO_INIT; | |
1316 | ||
1317 | oi.typep = &type; | |
1318 | oi.sizep = sizep; | |
1319 | if (sha1_object_info_extended(sha1, &oi, | |
1320 | OBJECT_INFO_LOOKUP_REPLACE) < 0) | |
1321 | return -1; | |
1322 | return type; | |
1323 | } | |
1324 | ||
1325 | static void *read_object(const unsigned char *sha1, enum object_type *type, | |
1326 | unsigned long *size) | |
1327 | { | |
1328 | struct object_info oi = OBJECT_INFO_INIT; | |
1329 | void *content; | |
1330 | oi.typep = type; | |
1331 | oi.sizep = size; | |
1332 | oi.contentp = &content; | |
1333 | ||
1334 | if (sha1_object_info_extended(sha1, &oi, 0) < 0) | |
1335 | return NULL; | |
1336 | return content; | |
1337 | } | |
1338 | ||
1339 | int pretend_object_file(void *buf, unsigned long len, enum object_type type, | |
1340 | struct object_id *oid) | |
1341 | { | |
1342 | struct cached_object *co; | |
1343 | ||
1344 | hash_object_file(buf, len, type_name(type), oid); | |
1345 | if (has_sha1_file(oid->hash) || find_cached_object(oid->hash)) | |
1346 | return 0; | |
1347 | ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc); | |
1348 | co = &cached_objects[cached_object_nr++]; | |
1349 | co->size = len; | |
1350 | co->type = type; | |
1351 | co->buf = xmalloc(len); | |
1352 | memcpy(co->buf, buf, len); | |
1353 | hashcpy(co->sha1, oid->hash); | |
1354 | return 0; | |
1355 | } | |
1356 | ||
1357 | /* | |
1358 | * This function dies on corrupt objects; the callers who want to | |
1359 | * deal with them should arrange to call read_object() and give error | |
1360 | * messages themselves. | |
1361 | */ | |
1362 | void *read_sha1_file_extended(const unsigned char *sha1, | |
1363 | enum object_type *type, | |
1364 | unsigned long *size, | |
1365 | int lookup_replace) | |
1366 | { | |
1367 | void *data; | |
1368 | const struct packed_git *p; | |
1369 | const char *path; | |
1370 | struct stat st; | |
1371 | const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1) | |
1372 | : sha1; | |
1373 | ||
1374 | errno = 0; | |
1375 | data = read_object(repl, type, size); | |
1376 | if (data) | |
1377 | return data; | |
1378 | ||
1379 | if (errno && errno != ENOENT) | |
1380 | die_errno("failed to read object %s", sha1_to_hex(sha1)); | |
1381 | ||
1382 | /* die if we replaced an object with one that does not exist */ | |
1383 | if (repl != sha1) | |
1384 | die("replacement %s not found for %s", | |
1385 | sha1_to_hex(repl), sha1_to_hex(sha1)); | |
1386 | ||
1387 | if (!stat_sha1_file(repl, &st, &path)) | |
1388 | die("loose object %s (stored in %s) is corrupt", | |
1389 | sha1_to_hex(repl), path); | |
1390 | ||
1391 | if ((p = has_packed_and_bad(repl)) != NULL) | |
1392 | die("packed object %s (stored in %s) is corrupt", | |
1393 | sha1_to_hex(repl), p->pack_name); | |
1394 | ||
1395 | return NULL; | |
1396 | } | |
1397 | ||
1398 | void *read_object_with_reference(const unsigned char *sha1, | |
1399 | const char *required_type_name, | |
1400 | unsigned long *size, | |
1401 | unsigned char *actual_sha1_return) | |
1402 | { | |
1403 | enum object_type type, required_type; | |
1404 | void *buffer; | |
1405 | unsigned long isize; | |
1406 | unsigned char actual_sha1[20]; | |
1407 | ||
1408 | required_type = type_from_string(required_type_name); | |
1409 | hashcpy(actual_sha1, sha1); | |
1410 | while (1) { | |
1411 | int ref_length = -1; | |
1412 | const char *ref_type = NULL; | |
1413 | ||
1414 | buffer = read_sha1_file(actual_sha1, &type, &isize); | |
1415 | if (!buffer) | |
1416 | return NULL; | |
1417 | if (type == required_type) { | |
1418 | *size = isize; | |
1419 | if (actual_sha1_return) | |
1420 | hashcpy(actual_sha1_return, actual_sha1); | |
1421 | return buffer; | |
1422 | } | |
1423 | /* Handle references */ | |
1424 | else if (type == OBJ_COMMIT) | |
1425 | ref_type = "tree "; | |
1426 | else if (type == OBJ_TAG) | |
1427 | ref_type = "object "; | |
1428 | else { | |
1429 | free(buffer); | |
1430 | return NULL; | |
1431 | } | |
1432 | ref_length = strlen(ref_type); | |
1433 | ||
1434 | if (ref_length + 40 > isize || | |
1435 | memcmp(buffer, ref_type, ref_length) || | |
1436 | get_sha1_hex((char *) buffer + ref_length, actual_sha1)) { | |
1437 | free(buffer); | |
1438 | return NULL; | |
1439 | } | |
1440 | free(buffer); | |
1441 | /* Now we have the ID of the referred-to object in | |
1442 | * actual_sha1. Check again. */ | |
1443 | } | |
1444 | } | |
1445 | ||
1446 | static void write_object_file_prepare(const void *buf, unsigned long len, | |
1447 | const char *type, struct object_id *oid, | |
1448 | char *hdr, int *hdrlen) | |
1449 | { | |
1450 | git_hash_ctx c; | |
1451 | ||
1452 | /* Generate the header */ | |
1453 | *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1; | |
1454 | ||
1455 | /* Sha1.. */ | |
1456 | the_hash_algo->init_fn(&c); | |
1457 | the_hash_algo->update_fn(&c, hdr, *hdrlen); | |
1458 | the_hash_algo->update_fn(&c, buf, len); | |
1459 | the_hash_algo->final_fn(oid->hash, &c); | |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * Move the just written object into its final resting place. | |
1464 | */ | |
1465 | int finalize_object_file(const char *tmpfile, const char *filename) | |
1466 | { | |
1467 | int ret = 0; | |
1468 | ||
1469 | if (object_creation_mode == OBJECT_CREATION_USES_RENAMES) | |
1470 | goto try_rename; | |
1471 | else if (link(tmpfile, filename)) | |
1472 | ret = errno; | |
1473 | ||
1474 | /* | |
1475 | * Coda hack - coda doesn't like cross-directory links, | |
1476 | * so we fall back to a rename, which will mean that it | |
1477 | * won't be able to check collisions, but that's not a | |
1478 | * big deal. | |
1479 | * | |
1480 | * The same holds for FAT formatted media. | |
1481 | * | |
1482 | * When this succeeds, we just return. We have nothing | |
1483 | * left to unlink. | |
1484 | */ | |
1485 | if (ret && ret != EEXIST) { | |
1486 | try_rename: | |
1487 | if (!rename(tmpfile, filename)) | |
1488 | goto out; | |
1489 | ret = errno; | |
1490 | } | |
1491 | unlink_or_warn(tmpfile); | |
1492 | if (ret) { | |
1493 | if (ret != EEXIST) { | |
1494 | return error_errno("unable to write sha1 filename %s", filename); | |
1495 | } | |
1496 | /* FIXME!!! Collision check here ? */ | |
1497 | } | |
1498 | ||
1499 | out: | |
1500 | if (adjust_shared_perm(filename)) | |
1501 | return error("unable to set permission to '%s'", filename); | |
1502 | return 0; | |
1503 | } | |
1504 | ||
1505 | static int write_buffer(int fd, const void *buf, size_t len) | |
1506 | { | |
1507 | if (write_in_full(fd, buf, len) < 0) | |
1508 | return error_errno("file write error"); | |
1509 | return 0; | |
1510 | } | |
1511 | ||
1512 | int hash_object_file(const void *buf, unsigned long len, const char *type, | |
1513 | struct object_id *oid) | |
1514 | { | |
1515 | char hdr[32]; | |
1516 | int hdrlen = sizeof(hdr); | |
1517 | write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen); | |
1518 | return 0; | |
1519 | } | |
1520 | ||
1521 | /* Finalize a file on disk, and close it. */ | |
1522 | static void close_sha1_file(int fd) | |
1523 | { | |
1524 | if (fsync_object_files) | |
1525 | fsync_or_die(fd, "sha1 file"); | |
1526 | if (close(fd) != 0) | |
1527 | die_errno("error when closing sha1 file"); | |
1528 | } | |
1529 | ||
1530 | /* Size of directory component, including the ending '/' */ | |
1531 | static inline int directory_size(const char *filename) | |
1532 | { | |
1533 | const char *s = strrchr(filename, '/'); | |
1534 | if (!s) | |
1535 | return 0; | |
1536 | return s - filename + 1; | |
1537 | } | |
1538 | ||
1539 | /* | |
1540 | * This creates a temporary file in the same directory as the final | |
1541 | * 'filename' | |
1542 | * | |
1543 | * We want to avoid cross-directory filename renames, because those | |
1544 | * can have problems on various filesystems (FAT, NFS, Coda). | |
1545 | */ | |
1546 | static int create_tmpfile(struct strbuf *tmp, const char *filename) | |
1547 | { | |
1548 | int fd, dirlen = directory_size(filename); | |
1549 | ||
1550 | strbuf_reset(tmp); | |
1551 | strbuf_add(tmp, filename, dirlen); | |
1552 | strbuf_addstr(tmp, "tmp_obj_XXXXXX"); | |
1553 | fd = git_mkstemp_mode(tmp->buf, 0444); | |
1554 | if (fd < 0 && dirlen && errno == ENOENT) { | |
1555 | /* | |
1556 | * Make sure the directory exists; note that the contents | |
1557 | * of the buffer are undefined after mkstemp returns an | |
1558 | * error, so we have to rewrite the whole buffer from | |
1559 | * scratch. | |
1560 | */ | |
1561 | strbuf_reset(tmp); | |
1562 | strbuf_add(tmp, filename, dirlen - 1); | |
1563 | if (mkdir(tmp->buf, 0777) && errno != EEXIST) | |
1564 | return -1; | |
1565 | if (adjust_shared_perm(tmp->buf)) | |
1566 | return -1; | |
1567 | ||
1568 | /* Try again */ | |
1569 | strbuf_addstr(tmp, "/tmp_obj_XXXXXX"); | |
1570 | fd = git_mkstemp_mode(tmp->buf, 0444); | |
1571 | } | |
1572 | return fd; | |
1573 | } | |
1574 | ||
1575 | static int write_loose_object(const struct object_id *oid, char *hdr, | |
1576 | int hdrlen, const void *buf, unsigned long len, | |
1577 | time_t mtime) | |
1578 | { | |
1579 | int fd, ret; | |
1580 | unsigned char compressed[4096]; | |
1581 | git_zstream stream; | |
1582 | git_hash_ctx c; | |
1583 | struct object_id parano_oid; | |
1584 | static struct strbuf tmp_file = STRBUF_INIT; | |
1585 | static struct strbuf filename = STRBUF_INIT; | |
1586 | ||
1587 | strbuf_reset(&filename); | |
1588 | sha1_file_name(&filename, oid->hash); | |
1589 | ||
1590 | fd = create_tmpfile(&tmp_file, filename.buf); | |
1591 | if (fd < 0) { | |
1592 | if (errno == EACCES) | |
1593 | return error("insufficient permission for adding an object to repository database %s", get_object_directory()); | |
1594 | else | |
1595 | return error_errno("unable to create temporary file"); | |
1596 | } | |
1597 | ||
1598 | /* Set it up */ | |
1599 | git_deflate_init(&stream, zlib_compression_level); | |
1600 | stream.next_out = compressed; | |
1601 | stream.avail_out = sizeof(compressed); | |
1602 | the_hash_algo->init_fn(&c); | |
1603 | ||
1604 | /* First header.. */ | |
1605 | stream.next_in = (unsigned char *)hdr; | |
1606 | stream.avail_in = hdrlen; | |
1607 | while (git_deflate(&stream, 0) == Z_OK) | |
1608 | ; /* nothing */ | |
1609 | the_hash_algo->update_fn(&c, hdr, hdrlen); | |
1610 | ||
1611 | /* Then the data itself.. */ | |
1612 | stream.next_in = (void *)buf; | |
1613 | stream.avail_in = len; | |
1614 | do { | |
1615 | unsigned char *in0 = stream.next_in; | |
1616 | ret = git_deflate(&stream, Z_FINISH); | |
1617 | the_hash_algo->update_fn(&c, in0, stream.next_in - in0); | |
1618 | if (write_buffer(fd, compressed, stream.next_out - compressed) < 0) | |
1619 | die("unable to write sha1 file"); | |
1620 | stream.next_out = compressed; | |
1621 | stream.avail_out = sizeof(compressed); | |
1622 | } while (ret == Z_OK); | |
1623 | ||
1624 | if (ret != Z_STREAM_END) | |
1625 | die("unable to deflate new object %s (%d)", oid_to_hex(oid), | |
1626 | ret); | |
1627 | ret = git_deflate_end_gently(&stream); | |
1628 | if (ret != Z_OK) | |
1629 | die("deflateEnd on object %s failed (%d)", oid_to_hex(oid), | |
1630 | ret); | |
1631 | the_hash_algo->final_fn(parano_oid.hash, &c); | |
1632 | if (oidcmp(oid, ¶no_oid) != 0) | |
1633 | die("confused by unstable object source data for %s", | |
1634 | oid_to_hex(oid)); | |
1635 | ||
1636 | close_sha1_file(fd); | |
1637 | ||
1638 | if (mtime) { | |
1639 | struct utimbuf utb; | |
1640 | utb.actime = mtime; | |
1641 | utb.modtime = mtime; | |
1642 | if (utime(tmp_file.buf, &utb) < 0) | |
1643 | warning_errno("failed utime() on %s", tmp_file.buf); | |
1644 | } | |
1645 | ||
1646 | return finalize_object_file(tmp_file.buf, filename.buf); | |
1647 | } | |
1648 | ||
1649 | static int freshen_loose_object(const unsigned char *sha1) | |
1650 | { | |
1651 | return check_and_freshen(sha1, 1); | |
1652 | } | |
1653 | ||
1654 | static int freshen_packed_object(const unsigned char *sha1) | |
1655 | { | |
1656 | struct pack_entry e; | |
1657 | if (!find_pack_entry(sha1, &e)) | |
1658 | return 0; | |
1659 | if (e.p->freshened) | |
1660 | return 1; | |
1661 | if (!freshen_file(e.p->pack_name)) | |
1662 | return 0; | |
1663 | e.p->freshened = 1; | |
1664 | return 1; | |
1665 | } | |
1666 | ||
1667 | int write_object_file(const void *buf, unsigned long len, const char *type, | |
1668 | struct object_id *oid) | |
1669 | { | |
1670 | char hdr[32]; | |
1671 | int hdrlen = sizeof(hdr); | |
1672 | ||
1673 | /* Normally if we have it in the pack then we do not bother writing | |
1674 | * it out into .git/objects/??/?{38} file. | |
1675 | */ | |
1676 | write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen); | |
1677 | if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash)) | |
1678 | return 0; | |
1679 | return write_loose_object(oid, hdr, hdrlen, buf, len, 0); | |
1680 | } | |
1681 | ||
1682 | int hash_object_file_literally(const void *buf, unsigned long len, | |
1683 | const char *type, struct object_id *oid, | |
1684 | unsigned flags) | |
1685 | { | |
1686 | char *header; | |
1687 | int hdrlen, status = 0; | |
1688 | ||
1689 | /* type string, SP, %lu of the length plus NUL must fit this */ | |
1690 | hdrlen = strlen(type) + 32; | |
1691 | header = xmalloc(hdrlen); | |
1692 | write_object_file_prepare(buf, len, type, oid, header, &hdrlen); | |
1693 | ||
1694 | if (!(flags & HASH_WRITE_OBJECT)) | |
1695 | goto cleanup; | |
1696 | if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash)) | |
1697 | goto cleanup; | |
1698 | status = write_loose_object(oid, header, hdrlen, buf, len, 0); | |
1699 | ||
1700 | cleanup: | |
1701 | free(header); | |
1702 | return status; | |
1703 | } | |
1704 | ||
1705 | int force_object_loose(const struct object_id *oid, time_t mtime) | |
1706 | { | |
1707 | void *buf; | |
1708 | unsigned long len; | |
1709 | enum object_type type; | |
1710 | char hdr[32]; | |
1711 | int hdrlen; | |
1712 | int ret; | |
1713 | ||
1714 | if (has_loose_object(oid->hash)) | |
1715 | return 0; | |
1716 | buf = read_object(oid->hash, &type, &len); | |
1717 | if (!buf) | |
1718 | return error("cannot read sha1_file for %s", oid_to_hex(oid)); | |
1719 | hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1; | |
1720 | ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime); | |
1721 | free(buf); | |
1722 | ||
1723 | return ret; | |
1724 | } | |
1725 | ||
1726 | int has_sha1_file_with_flags(const unsigned char *sha1, int flags) | |
1727 | { | |
1728 | if (!startup_info->have_repository) | |
1729 | return 0; | |
1730 | return sha1_object_info_extended(sha1, NULL, | |
1731 | flags | OBJECT_INFO_SKIP_CACHED) >= 0; | |
1732 | } | |
1733 | ||
1734 | int has_object_file(const struct object_id *oid) | |
1735 | { | |
1736 | return has_sha1_file(oid->hash); | |
1737 | } | |
1738 | ||
1739 | int has_object_file_with_flags(const struct object_id *oid, int flags) | |
1740 | { | |
1741 | return has_sha1_file_with_flags(oid->hash, flags); | |
1742 | } | |
1743 | ||
1744 | static void check_tree(const void *buf, size_t size) | |
1745 | { | |
1746 | struct tree_desc desc; | |
1747 | struct name_entry entry; | |
1748 | ||
1749 | init_tree_desc(&desc, buf, size); | |
1750 | while (tree_entry(&desc, &entry)) | |
1751 | /* do nothing | |
1752 | * tree_entry() will die() on malformed entries */ | |
1753 | ; | |
1754 | } | |
1755 | ||
1756 | static void check_commit(const void *buf, size_t size) | |
1757 | { | |
1758 | struct commit c; | |
1759 | memset(&c, 0, sizeof(c)); | |
1760 | if (parse_commit_buffer(&c, buf, size)) | |
1761 | die("corrupt commit"); | |
1762 | } | |
1763 | ||
1764 | static void check_tag(const void *buf, size_t size) | |
1765 | { | |
1766 | struct tag t; | |
1767 | memset(&t, 0, sizeof(t)); | |
1768 | if (parse_tag_buffer(&t, buf, size)) | |
1769 | die("corrupt tag"); | |
1770 | } | |
1771 | ||
1772 | static int index_mem(struct object_id *oid, void *buf, size_t size, | |
1773 | enum object_type type, | |
1774 | const char *path, unsigned flags) | |
1775 | { | |
1776 | int ret, re_allocated = 0; | |
1777 | int write_object = flags & HASH_WRITE_OBJECT; | |
1778 | ||
1779 | if (!type) | |
1780 | type = OBJ_BLOB; | |
1781 | ||
1782 | /* | |
1783 | * Convert blobs to git internal format | |
1784 | */ | |
1785 | if ((type == OBJ_BLOB) && path) { | |
1786 | struct strbuf nbuf = STRBUF_INIT; | |
1787 | if (convert_to_git(&the_index, path, buf, size, &nbuf, | |
1788 | get_conv_flags(flags))) { | |
1789 | buf = strbuf_detach(&nbuf, &size); | |
1790 | re_allocated = 1; | |
1791 | } | |
1792 | } | |
1793 | if (flags & HASH_FORMAT_CHECK) { | |
1794 | if (type == OBJ_TREE) | |
1795 | check_tree(buf, size); | |
1796 | if (type == OBJ_COMMIT) | |
1797 | check_commit(buf, size); | |
1798 | if (type == OBJ_TAG) | |
1799 | check_tag(buf, size); | |
1800 | } | |
1801 | ||
1802 | if (write_object) | |
1803 | ret = write_object_file(buf, size, type_name(type), oid); | |
1804 | else | |
1805 | ret = hash_object_file(buf, size, type_name(type), oid); | |
1806 | if (re_allocated) | |
1807 | free(buf); | |
1808 | return ret; | |
1809 | } | |
1810 | ||
1811 | static int index_stream_convert_blob(struct object_id *oid, int fd, | |
1812 | const char *path, unsigned flags) | |
1813 | { | |
1814 | int ret; | |
1815 | const int write_object = flags & HASH_WRITE_OBJECT; | |
1816 | struct strbuf sbuf = STRBUF_INIT; | |
1817 | ||
1818 | assert(path); | |
1819 | assert(would_convert_to_git_filter_fd(path)); | |
1820 | ||
1821 | convert_to_git_filter_fd(&the_index, path, fd, &sbuf, | |
1822 | get_conv_flags(flags)); | |
1823 | ||
1824 | if (write_object) | |
1825 | ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB), | |
1826 | oid); | |
1827 | else | |
1828 | ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB), | |
1829 | oid); | |
1830 | strbuf_release(&sbuf); | |
1831 | return ret; | |
1832 | } | |
1833 | ||
1834 | static int index_pipe(struct object_id *oid, int fd, enum object_type type, | |
1835 | const char *path, unsigned flags) | |
1836 | { | |
1837 | struct strbuf sbuf = STRBUF_INIT; | |
1838 | int ret; | |
1839 | ||
1840 | if (strbuf_read(&sbuf, fd, 4096) >= 0) | |
1841 | ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags); | |
1842 | else | |
1843 | ret = -1; | |
1844 | strbuf_release(&sbuf); | |
1845 | return ret; | |
1846 | } | |
1847 | ||
1848 | #define SMALL_FILE_SIZE (32*1024) | |
1849 | ||
1850 | static int index_core(struct object_id *oid, int fd, size_t size, | |
1851 | enum object_type type, const char *path, | |
1852 | unsigned flags) | |
1853 | { | |
1854 | int ret; | |
1855 | ||
1856 | if (!size) { | |
1857 | ret = index_mem(oid, "", size, type, path, flags); | |
1858 | } else if (size <= SMALL_FILE_SIZE) { | |
1859 | char *buf = xmalloc(size); | |
1860 | ssize_t read_result = read_in_full(fd, buf, size); | |
1861 | if (read_result < 0) | |
1862 | ret = error_errno("read error while indexing %s", | |
1863 | path ? path : "<unknown>"); | |
1864 | else if (read_result != size) | |
1865 | ret = error("short read while indexing %s", | |
1866 | path ? path : "<unknown>"); | |
1867 | else | |
1868 | ret = index_mem(oid, buf, size, type, path, flags); | |
1869 | free(buf); | |
1870 | } else { | |
1871 | void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0); | |
1872 | ret = index_mem(oid, buf, size, type, path, flags); | |
1873 | munmap(buf, size); | |
1874 | } | |
1875 | return ret; | |
1876 | } | |
1877 | ||
1878 | /* | |
1879 | * This creates one packfile per large blob unless bulk-checkin | |
1880 | * machinery is "plugged". | |
1881 | * | |
1882 | * This also bypasses the usual "convert-to-git" dance, and that is on | |
1883 | * purpose. We could write a streaming version of the converting | |
1884 | * functions and insert that before feeding the data to fast-import | |
1885 | * (or equivalent in-core API described above). However, that is | |
1886 | * somewhat complicated, as we do not know the size of the filter | |
1887 | * result, which we need to know beforehand when writing a git object. | |
1888 | * Since the primary motivation for trying to stream from the working | |
1889 | * tree file and to avoid mmaping it in core is to deal with large | |
1890 | * binary blobs, they generally do not want to get any conversion, and | |
1891 | * callers should avoid this code path when filters are requested. | |
1892 | */ | |
1893 | static int index_stream(struct object_id *oid, int fd, size_t size, | |
1894 | enum object_type type, const char *path, | |
1895 | unsigned flags) | |
1896 | { | |
1897 | return index_bulk_checkin(oid->hash, fd, size, type, path, flags); | |
1898 | } | |
1899 | ||
1900 | int index_fd(struct object_id *oid, int fd, struct stat *st, | |
1901 | enum object_type type, const char *path, unsigned flags) | |
1902 | { | |
1903 | int ret; | |
1904 | ||
1905 | /* | |
1906 | * Call xsize_t() only when needed to avoid potentially unnecessary | |
1907 | * die() for large files. | |
1908 | */ | |
1909 | if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path)) | |
1910 | ret = index_stream_convert_blob(oid, fd, path, flags); | |
1911 | else if (!S_ISREG(st->st_mode)) | |
1912 | ret = index_pipe(oid, fd, type, path, flags); | |
1913 | else if (st->st_size <= big_file_threshold || type != OBJ_BLOB || | |
1914 | (path && would_convert_to_git(&the_index, path))) | |
1915 | ret = index_core(oid, fd, xsize_t(st->st_size), type, path, | |
1916 | flags); | |
1917 | else | |
1918 | ret = index_stream(oid, fd, xsize_t(st->st_size), type, path, | |
1919 | flags); | |
1920 | close(fd); | |
1921 | return ret; | |
1922 | } | |
1923 | ||
1924 | int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags) | |
1925 | { | |
1926 | int fd; | |
1927 | struct strbuf sb = STRBUF_INIT; | |
1928 | int rc = 0; | |
1929 | ||
1930 | switch (st->st_mode & S_IFMT) { | |
1931 | case S_IFREG: | |
1932 | fd = open(path, O_RDONLY); | |
1933 | if (fd < 0) | |
1934 | return error_errno("open(\"%s\")", path); | |
1935 | if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0) | |
1936 | return error("%s: failed to insert into database", | |
1937 | path); | |
1938 | break; | |
1939 | case S_IFLNK: | |
1940 | if (strbuf_readlink(&sb, path, st->st_size)) | |
1941 | return error_errno("readlink(\"%s\")", path); | |
1942 | if (!(flags & HASH_WRITE_OBJECT)) | |
1943 | hash_object_file(sb.buf, sb.len, blob_type, oid); | |
1944 | else if (write_object_file(sb.buf, sb.len, blob_type, oid)) | |
1945 | rc = error("%s: failed to insert into database", path); | |
1946 | strbuf_release(&sb); | |
1947 | break; | |
1948 | case S_IFDIR: | |
1949 | return resolve_gitlink_ref(path, "HEAD", oid); | |
1950 | default: | |
1951 | return error("%s: unsupported file type", path); | |
1952 | } | |
1953 | return rc; | |
1954 | } | |
1955 | ||
1956 | int read_pack_header(int fd, struct pack_header *header) | |
1957 | { | |
1958 | if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header)) | |
1959 | /* "eof before pack header was fully read" */ | |
1960 | return PH_ERROR_EOF; | |
1961 | ||
1962 | if (header->hdr_signature != htonl(PACK_SIGNATURE)) | |
1963 | /* "protocol error (pack signature mismatch detected)" */ | |
1964 | return PH_ERROR_PACK_SIGNATURE; | |
1965 | if (!pack_version_ok(header->hdr_version)) | |
1966 | /* "protocol error (pack version unsupported)" */ | |
1967 | return PH_ERROR_PROTOCOL; | |
1968 | return 0; | |
1969 | } | |
1970 | ||
1971 | void assert_sha1_type(const unsigned char *sha1, enum object_type expect) | |
1972 | { | |
1973 | enum object_type type = sha1_object_info(sha1, NULL); | |
1974 | if (type < 0) | |
1975 | die("%s is not a valid object", sha1_to_hex(sha1)); | |
1976 | if (type != expect) | |
1977 | die("%s is not a valid '%s' object", sha1_to_hex(sha1), | |
1978 | type_name(expect)); | |
1979 | } | |
1980 | ||
1981 | int for_each_file_in_obj_subdir(unsigned int subdir_nr, | |
1982 | struct strbuf *path, | |
1983 | each_loose_object_fn obj_cb, | |
1984 | each_loose_cruft_fn cruft_cb, | |
1985 | each_loose_subdir_fn subdir_cb, | |
1986 | void *data) | |
1987 | { | |
1988 | size_t origlen, baselen; | |
1989 | DIR *dir; | |
1990 | struct dirent *de; | |
1991 | int r = 0; | |
1992 | struct object_id oid; | |
1993 | ||
1994 | if (subdir_nr > 0xff) | |
1995 | BUG("invalid loose object subdirectory: %x", subdir_nr); | |
1996 | ||
1997 | origlen = path->len; | |
1998 | strbuf_complete(path, '/'); | |
1999 | strbuf_addf(path, "%02x", subdir_nr); | |
2000 | ||
2001 | dir = opendir(path->buf); | |
2002 | if (!dir) { | |
2003 | if (errno != ENOENT) | |
2004 | r = error_errno("unable to open %s", path->buf); | |
2005 | strbuf_setlen(path, origlen); | |
2006 | return r; | |
2007 | } | |
2008 | ||
2009 | oid.hash[0] = subdir_nr; | |
2010 | strbuf_addch(path, '/'); | |
2011 | baselen = path->len; | |
2012 | ||
2013 | while ((de = readdir(dir))) { | |
2014 | size_t namelen; | |
2015 | if (is_dot_or_dotdot(de->d_name)) | |
2016 | continue; | |
2017 | ||
2018 | namelen = strlen(de->d_name); | |
2019 | strbuf_setlen(path, baselen); | |
2020 | strbuf_add(path, de->d_name, namelen); | |
2021 | if (namelen == GIT_SHA1_HEXSZ - 2 && | |
2022 | !hex_to_bytes(oid.hash + 1, de->d_name, | |
2023 | GIT_SHA1_RAWSZ - 1)) { | |
2024 | if (obj_cb) { | |
2025 | r = obj_cb(&oid, path->buf, data); | |
2026 | if (r) | |
2027 | break; | |
2028 | } | |
2029 | continue; | |
2030 | } | |
2031 | ||
2032 | if (cruft_cb) { | |
2033 | r = cruft_cb(de->d_name, path->buf, data); | |
2034 | if (r) | |
2035 | break; | |
2036 | } | |
2037 | } | |
2038 | closedir(dir); | |
2039 | ||
2040 | strbuf_setlen(path, baselen - 1); | |
2041 | if (!r && subdir_cb) | |
2042 | r = subdir_cb(subdir_nr, path->buf, data); | |
2043 | ||
2044 | strbuf_setlen(path, origlen); | |
2045 | ||
2046 | return r; | |
2047 | } | |
2048 | ||
2049 | int for_each_loose_file_in_objdir_buf(struct strbuf *path, | |
2050 | each_loose_object_fn obj_cb, | |
2051 | each_loose_cruft_fn cruft_cb, | |
2052 | each_loose_subdir_fn subdir_cb, | |
2053 | void *data) | |
2054 | { | |
2055 | int r = 0; | |
2056 | int i; | |
2057 | ||
2058 | for (i = 0; i < 256; i++) { | |
2059 | r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb, | |
2060 | subdir_cb, data); | |
2061 | if (r) | |
2062 | break; | |
2063 | } | |
2064 | ||
2065 | return r; | |
2066 | } | |
2067 | ||
2068 | int for_each_loose_file_in_objdir(const char *path, | |
2069 | each_loose_object_fn obj_cb, | |
2070 | each_loose_cruft_fn cruft_cb, | |
2071 | each_loose_subdir_fn subdir_cb, | |
2072 | void *data) | |
2073 | { | |
2074 | struct strbuf buf = STRBUF_INIT; | |
2075 | int r; | |
2076 | ||
2077 | strbuf_addstr(&buf, path); | |
2078 | r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb, | |
2079 | subdir_cb, data); | |
2080 | strbuf_release(&buf); | |
2081 | ||
2082 | return r; | |
2083 | } | |
2084 | ||
2085 | struct loose_alt_odb_data { | |
2086 | each_loose_object_fn *cb; | |
2087 | void *data; | |
2088 | }; | |
2089 | ||
2090 | static int loose_from_alt_odb(struct alternate_object_database *alt, | |
2091 | void *vdata) | |
2092 | { | |
2093 | struct loose_alt_odb_data *data = vdata; | |
2094 | struct strbuf buf = STRBUF_INIT; | |
2095 | int r; | |
2096 | ||
2097 | strbuf_addstr(&buf, alt->path); | |
2098 | r = for_each_loose_file_in_objdir_buf(&buf, | |
2099 | data->cb, NULL, NULL, | |
2100 | data->data); | |
2101 | strbuf_release(&buf); | |
2102 | return r; | |
2103 | } | |
2104 | ||
2105 | int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags) | |
2106 | { | |
2107 | struct loose_alt_odb_data alt; | |
2108 | int r; | |
2109 | ||
2110 | r = for_each_loose_file_in_objdir(get_object_directory(), | |
2111 | cb, NULL, NULL, data); | |
2112 | if (r) | |
2113 | return r; | |
2114 | ||
2115 | if (flags & FOR_EACH_OBJECT_LOCAL_ONLY) | |
2116 | return 0; | |
2117 | ||
2118 | alt.cb = cb; | |
2119 | alt.data = data; | |
2120 | return foreach_alt_odb(loose_from_alt_odb, &alt); | |
2121 | } | |
2122 | ||
2123 | static int check_stream_sha1(git_zstream *stream, | |
2124 | const char *hdr, | |
2125 | unsigned long size, | |
2126 | const char *path, | |
2127 | const unsigned char *expected_sha1) | |
2128 | { | |
2129 | git_hash_ctx c; | |
2130 | unsigned char real_sha1[GIT_MAX_RAWSZ]; | |
2131 | unsigned char buf[4096]; | |
2132 | unsigned long total_read; | |
2133 | int status = Z_OK; | |
2134 | ||
2135 | the_hash_algo->init_fn(&c); | |
2136 | the_hash_algo->update_fn(&c, hdr, stream->total_out); | |
2137 | ||
2138 | /* | |
2139 | * We already read some bytes into hdr, but the ones up to the NUL | |
2140 | * do not count against the object's content size. | |
2141 | */ | |
2142 | total_read = stream->total_out - strlen(hdr) - 1; | |
2143 | ||
2144 | /* | |
2145 | * This size comparison must be "<=" to read the final zlib packets; | |
2146 | * see the comment in unpack_sha1_rest for details. | |
2147 | */ | |
2148 | while (total_read <= size && | |
2149 | (status == Z_OK || status == Z_BUF_ERROR)) { | |
2150 | stream->next_out = buf; | |
2151 | stream->avail_out = sizeof(buf); | |
2152 | if (size - total_read < stream->avail_out) | |
2153 | stream->avail_out = size - total_read; | |
2154 | status = git_inflate(stream, Z_FINISH); | |
2155 | the_hash_algo->update_fn(&c, buf, stream->next_out - buf); | |
2156 | total_read += stream->next_out - buf; | |
2157 | } | |
2158 | git_inflate_end(stream); | |
2159 | ||
2160 | if (status != Z_STREAM_END) { | |
2161 | error("corrupt loose object '%s'", sha1_to_hex(expected_sha1)); | |
2162 | return -1; | |
2163 | } | |
2164 | if (stream->avail_in) { | |
2165 | error("garbage at end of loose object '%s'", | |
2166 | sha1_to_hex(expected_sha1)); | |
2167 | return -1; | |
2168 | } | |
2169 | ||
2170 | the_hash_algo->final_fn(real_sha1, &c); | |
2171 | if (hashcmp(expected_sha1, real_sha1)) { | |
2172 | error("sha1 mismatch for %s (expected %s)", path, | |
2173 | sha1_to_hex(expected_sha1)); | |
2174 | return -1; | |
2175 | } | |
2176 | ||
2177 | return 0; | |
2178 | } | |
2179 | ||
2180 | int read_loose_object(const char *path, | |
2181 | const unsigned char *expected_sha1, | |
2182 | enum object_type *type, | |
2183 | unsigned long *size, | |
2184 | void **contents) | |
2185 | { | |
2186 | int ret = -1; | |
2187 | void *map = NULL; | |
2188 | unsigned long mapsize; | |
2189 | git_zstream stream; | |
2190 | char hdr[32]; | |
2191 | ||
2192 | *contents = NULL; | |
2193 | ||
2194 | map = map_sha1_file_1(path, NULL, &mapsize); | |
2195 | if (!map) { | |
2196 | error_errno("unable to mmap %s", path); | |
2197 | goto out; | |
2198 | } | |
2199 | ||
2200 | if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) { | |
2201 | error("unable to unpack header of %s", path); | |
2202 | goto out; | |
2203 | } | |
2204 | ||
2205 | *type = parse_sha1_header(hdr, size); | |
2206 | if (*type < 0) { | |
2207 | error("unable to parse header of %s", path); | |
2208 | git_inflate_end(&stream); | |
2209 | goto out; | |
2210 | } | |
2211 | ||
2212 | if (*type == OBJ_BLOB) { | |
2213 | if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0) | |
2214 | goto out; | |
2215 | } else { | |
2216 | *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1); | |
2217 | if (!*contents) { | |
2218 | error("unable to unpack contents of %s", path); | |
2219 | git_inflate_end(&stream); | |
2220 | goto out; | |
2221 | } | |
2222 | if (check_sha1_signature(expected_sha1, *contents, | |
2223 | *size, type_name(*type))) { | |
2224 | error("sha1 mismatch for %s (expected %s)", path, | |
2225 | sha1_to_hex(expected_sha1)); | |
2226 | free(*contents); | |
2227 | goto out; | |
2228 | } | |
2229 | } | |
2230 | ||
2231 | ret = 0; /* everything checks out */ | |
2232 | ||
2233 | out: | |
2234 | if (map) | |
2235 | munmap(map, mapsize); | |
2236 | return ret; | |
2237 | } |