ad02649124bb817360d6cd1ab9b354fd3243cdd7
[git/git.git] / sha1-file.c
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 "repository.h"
26 #include "replace-object.h"
27 #include "streaming.h"
28 #include "dir.h"
29 #include "list.h"
30 #include "mergesort.h"
31 #include "quote.h"
32 #include "packfile.h"
33 #include "fetch-object.h"
34 #include "object-store.h"
35
36 /* The maximum size for an object header. */
37 #define MAX_HEADER_LEN 32
38
39
40 #define EMPTY_TREE_SHA1_BIN_LITERAL \
41 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
42 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
43 #define EMPTY_TREE_SHA256_BIN_LITERAL \
44 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
45 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
46 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
47 "\x53\x21"
48
49 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
50 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
51 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
52 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
53 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
54 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
55 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
56 "\x18\x13"
57
58 const unsigned char null_sha1[GIT_MAX_RAWSZ];
59 const struct object_id null_oid;
60 static const struct object_id empty_tree_oid = {
61 EMPTY_TREE_SHA1_BIN_LITERAL
62 };
63 static const struct object_id empty_blob_oid = {
64 EMPTY_BLOB_SHA1_BIN_LITERAL
65 };
66 static const struct object_id empty_tree_oid_sha256 = {
67 EMPTY_TREE_SHA256_BIN_LITERAL
68 };
69 static const struct object_id empty_blob_oid_sha256 = {
70 EMPTY_BLOB_SHA256_BIN_LITERAL
71 };
72
73 static void git_hash_sha1_init(git_hash_ctx *ctx)
74 {
75 git_SHA1_Init(&ctx->sha1);
76 }
77
78 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
79 {
80 git_SHA1_Update(&ctx->sha1, data, len);
81 }
82
83 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
84 {
85 git_SHA1_Final(hash, &ctx->sha1);
86 }
87
88
89 static void git_hash_sha256_init(git_hash_ctx *ctx)
90 {
91 git_SHA256_Init(&ctx->sha256);
92 }
93
94 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
95 {
96 git_SHA256_Update(&ctx->sha256, data, len);
97 }
98
99 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
100 {
101 git_SHA256_Final(hash, &ctx->sha256);
102 }
103
104 static void git_hash_unknown_init(git_hash_ctx *ctx)
105 {
106 BUG("trying to init unknown hash");
107 }
108
109 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
110 {
111 BUG("trying to update unknown hash");
112 }
113
114 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
115 {
116 BUG("trying to finalize unknown hash");
117 }
118
119 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
120 {
121 NULL,
122 0x00000000,
123 0,
124 0,
125 0,
126 git_hash_unknown_init,
127 git_hash_unknown_update,
128 git_hash_unknown_final,
129 NULL,
130 NULL,
131 },
132 {
133 "sha1",
134 /* "sha1", big-endian */
135 0x73686131,
136 GIT_SHA1_RAWSZ,
137 GIT_SHA1_HEXSZ,
138 GIT_SHA1_BLKSZ,
139 git_hash_sha1_init,
140 git_hash_sha1_update,
141 git_hash_sha1_final,
142 &empty_tree_oid,
143 &empty_blob_oid,
144 },
145 {
146 "sha256",
147 /* "s256", big-endian */
148 0x73323536,
149 GIT_SHA256_RAWSZ,
150 GIT_SHA256_HEXSZ,
151 GIT_SHA256_BLKSZ,
152 git_hash_sha256_init,
153 git_hash_sha256_update,
154 git_hash_sha256_final,
155 &empty_tree_oid_sha256,
156 &empty_blob_oid_sha256,
157 }
158 };
159
160 const char *empty_tree_oid_hex(void)
161 {
162 static char buf[GIT_MAX_HEXSZ + 1];
163 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
164 }
165
166 const char *empty_blob_oid_hex(void)
167 {
168 static char buf[GIT_MAX_HEXSZ + 1];
169 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
170 }
171
172 int hash_algo_by_name(const char *name)
173 {
174 int i;
175 if (!name)
176 return GIT_HASH_UNKNOWN;
177 for (i = 1; i < GIT_HASH_NALGOS; i++)
178 if (!strcmp(name, hash_algos[i].name))
179 return i;
180 return GIT_HASH_UNKNOWN;
181 }
182
183 int hash_algo_by_id(uint32_t format_id)
184 {
185 int i;
186 for (i = 1; i < GIT_HASH_NALGOS; i++)
187 if (format_id == hash_algos[i].format_id)
188 return i;
189 return GIT_HASH_UNKNOWN;
190 }
191
192
193 /*
194 * This is meant to hold a *small* number of objects that you would
195 * want read_object_file() to be able to return, but yet you do not want
196 * to write them into the object store (e.g. a browse-only
197 * application).
198 */
199 static struct cached_object {
200 struct object_id oid;
201 enum object_type type;
202 void *buf;
203 unsigned long size;
204 } *cached_objects;
205 static int cached_object_nr, cached_object_alloc;
206
207 static struct cached_object empty_tree = {
208 { EMPTY_TREE_SHA1_BIN_LITERAL },
209 OBJ_TREE,
210 "",
211 0
212 };
213
214 static struct cached_object *find_cached_object(const struct object_id *oid)
215 {
216 int i;
217 struct cached_object *co = cached_objects;
218
219 for (i = 0; i < cached_object_nr; i++, co++) {
220 if (oideq(&co->oid, oid))
221 return co;
222 }
223 if (oideq(oid, the_hash_algo->empty_tree))
224 return &empty_tree;
225 return NULL;
226 }
227
228
229 static int get_conv_flags(unsigned flags)
230 {
231 if (flags & HASH_RENORMALIZE)
232 return CONV_EOL_RENORMALIZE;
233 else if (flags & HASH_WRITE_OBJECT)
234 return global_conv_flags_eol | CONV_WRITE_OBJECT;
235 else
236 return 0;
237 }
238
239
240 int mkdir_in_gitdir(const char *path)
241 {
242 if (mkdir(path, 0777)) {
243 int saved_errno = errno;
244 struct stat st;
245 struct strbuf sb = STRBUF_INIT;
246
247 if (errno != EEXIST)
248 return -1;
249 /*
250 * Are we looking at a path in a symlinked worktree
251 * whose original repository does not yet have it?
252 * e.g. .git/rr-cache pointing at its original
253 * repository in which the user hasn't performed any
254 * conflict resolution yet?
255 */
256 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
257 strbuf_readlink(&sb, path, st.st_size) ||
258 !is_absolute_path(sb.buf) ||
259 mkdir(sb.buf, 0777)) {
260 strbuf_release(&sb);
261 errno = saved_errno;
262 return -1;
263 }
264 strbuf_release(&sb);
265 }
266 return adjust_shared_perm(path);
267 }
268
269 enum scld_error safe_create_leading_directories(char *path)
270 {
271 char *next_component = path + offset_1st_component(path);
272 enum scld_error ret = SCLD_OK;
273
274 while (ret == SCLD_OK && next_component) {
275 struct stat st;
276 char *slash = next_component, slash_character;
277
278 while (*slash && !is_dir_sep(*slash))
279 slash++;
280
281 if (!*slash)
282 break;
283
284 next_component = slash + 1;
285 while (is_dir_sep(*next_component))
286 next_component++;
287 if (!*next_component)
288 break;
289
290 slash_character = *slash;
291 *slash = '\0';
292 if (!stat(path, &st)) {
293 /* path exists */
294 if (!S_ISDIR(st.st_mode)) {
295 errno = ENOTDIR;
296 ret = SCLD_EXISTS;
297 }
298 } else if (mkdir(path, 0777)) {
299 if (errno == EEXIST &&
300 !stat(path, &st) && S_ISDIR(st.st_mode))
301 ; /* somebody created it since we checked */
302 else if (errno == ENOENT)
303 /*
304 * Either mkdir() failed because
305 * somebody just pruned the containing
306 * directory, or stat() failed because
307 * the file that was in our way was
308 * just removed. Either way, inform
309 * the caller that it might be worth
310 * trying again:
311 */
312 ret = SCLD_VANISHED;
313 else
314 ret = SCLD_FAILED;
315 } else if (adjust_shared_perm(path)) {
316 ret = SCLD_PERMS;
317 }
318 *slash = slash_character;
319 }
320 return ret;
321 }
322
323 enum scld_error safe_create_leading_directories_const(const char *path)
324 {
325 int save_errno;
326 /* path points to cache entries, so xstrdup before messing with it */
327 char *buf = xstrdup(path);
328 enum scld_error result = safe_create_leading_directories(buf);
329
330 save_errno = errno;
331 free(buf);
332 errno = save_errno;
333 return result;
334 }
335
336 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
337 {
338 /*
339 * The number of times we will try to remove empty directories
340 * in the way of path. This is only 1 because if another
341 * process is racily creating directories that conflict with
342 * us, we don't want to fight against them.
343 */
344 int remove_directories_remaining = 1;
345
346 /*
347 * The number of times that we will try to create the
348 * directories containing path. We are willing to attempt this
349 * more than once, because another process could be trying to
350 * clean up empty directories at the same time as we are
351 * trying to create them.
352 */
353 int create_directories_remaining = 3;
354
355 /* A scratch copy of path, filled lazily if we need it: */
356 struct strbuf path_copy = STRBUF_INIT;
357
358 int ret, save_errno;
359
360 /* Sanity check: */
361 assert(*path);
362
363 retry_fn:
364 ret = fn(path, cb);
365 save_errno = errno;
366 if (!ret)
367 goto out;
368
369 if (errno == EISDIR && remove_directories_remaining-- > 0) {
370 /*
371 * A directory is in the way. Maybe it is empty; try
372 * to remove it:
373 */
374 if (!path_copy.len)
375 strbuf_addstr(&path_copy, path);
376
377 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
378 goto retry_fn;
379 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
380 /*
381 * Maybe the containing directory didn't exist, or
382 * maybe it was just deleted by a process that is
383 * racing with us to clean up empty directories. Try
384 * to create it:
385 */
386 enum scld_error scld_result;
387
388 if (!path_copy.len)
389 strbuf_addstr(&path_copy, path);
390
391 do {
392 scld_result = safe_create_leading_directories(path_copy.buf);
393 if (scld_result == SCLD_OK)
394 goto retry_fn;
395 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
396 }
397
398 out:
399 strbuf_release(&path_copy);
400 errno = save_errno;
401 return ret;
402 }
403
404 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
405 {
406 int i;
407 for (i = 0; i < the_hash_algo->rawsz; i++) {
408 static char hex[] = "0123456789abcdef";
409 unsigned int val = oid->hash[i];
410 strbuf_addch(buf, hex[val >> 4]);
411 strbuf_addch(buf, hex[val & 0xf]);
412 if (!i)
413 strbuf_addch(buf, '/');
414 }
415 }
416
417 static const char *odb_loose_path(struct object_directory *odb,
418 struct strbuf *buf,
419 const struct object_id *oid)
420 {
421 strbuf_reset(buf);
422 strbuf_addstr(buf, odb->path);
423 strbuf_addch(buf, '/');
424 fill_loose_path(buf, oid);
425 return buf->buf;
426 }
427
428 const char *loose_object_path(struct repository *r, struct strbuf *buf,
429 const struct object_id *oid)
430 {
431 return odb_loose_path(r->objects->odb, buf, oid);
432 }
433
434 /*
435 * Return non-zero iff the path is usable as an alternate object database.
436 */
437 static int alt_odb_usable(struct raw_object_store *o,
438 struct strbuf *path,
439 const char *normalized_objdir)
440 {
441 struct object_directory *odb;
442
443 /* Detect cases where alternate disappeared */
444 if (!is_directory(path->buf)) {
445 error(_("object directory %s does not exist; "
446 "check .git/objects/info/alternates"),
447 path->buf);
448 return 0;
449 }
450
451 /*
452 * Prevent the common mistake of listing the same
453 * thing twice, or object directory itself.
454 */
455 for (odb = o->odb; odb; odb = odb->next) {
456 if (!fspathcmp(path->buf, odb->path))
457 return 0;
458 }
459 if (!fspathcmp(path->buf, normalized_objdir))
460 return 0;
461
462 return 1;
463 }
464
465 /*
466 * Prepare alternate object database registry.
467 *
468 * The variable alt_odb_list points at the list of struct
469 * object_directory. The elements on this list come from
470 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
471 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
472 * whose contents is similar to that environment variable but can be
473 * LF separated. Its base points at a statically allocated buffer that
474 * contains "/the/directory/corresponding/to/.git/objects/...", while
475 * its name points just after the slash at the end of ".git/objects/"
476 * in the example above, and has enough space to hold 40-byte hex
477 * SHA1, an extra slash for the first level indirection, and the
478 * terminating NUL.
479 */
480 static void read_info_alternates(struct repository *r,
481 const char *relative_base,
482 int depth);
483 static int link_alt_odb_entry(struct repository *r, const char *entry,
484 const char *relative_base, int depth, const char *normalized_objdir)
485 {
486 struct object_directory *ent;
487 struct strbuf pathbuf = STRBUF_INIT;
488
489 if (!is_absolute_path(entry) && relative_base) {
490 strbuf_realpath(&pathbuf, relative_base, 1);
491 strbuf_addch(&pathbuf, '/');
492 }
493 strbuf_addstr(&pathbuf, entry);
494
495 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
496 error(_("unable to normalize alternate object path: %s"),
497 pathbuf.buf);
498 strbuf_release(&pathbuf);
499 return -1;
500 }
501
502 /*
503 * The trailing slash after the directory name is given by
504 * this function at the end. Remove duplicates.
505 */
506 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
507 strbuf_setlen(&pathbuf, pathbuf.len - 1);
508
509 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
510 strbuf_release(&pathbuf);
511 return -1;
512 }
513
514 ent = xcalloc(1, sizeof(*ent));
515 ent->path = xstrdup(pathbuf.buf);
516
517 /* add the alternate entry */
518 *r->objects->odb_tail = ent;
519 r->objects->odb_tail = &(ent->next);
520 ent->next = NULL;
521
522 /* recursively add alternates */
523 read_info_alternates(r, pathbuf.buf, depth + 1);
524
525 strbuf_release(&pathbuf);
526 return 0;
527 }
528
529 static const char *parse_alt_odb_entry(const char *string,
530 int sep,
531 struct strbuf *out)
532 {
533 const char *end;
534
535 strbuf_reset(out);
536
537 if (*string == '#') {
538 /* comment; consume up to next separator */
539 end = strchrnul(string, sep);
540 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
541 /*
542 * quoted path; unquote_c_style has copied the
543 * data for us and set "end". Broken quoting (e.g.,
544 * an entry that doesn't end with a quote) falls
545 * back to the unquoted case below.
546 */
547 } else {
548 /* normal, unquoted path */
549 end = strchrnul(string, sep);
550 strbuf_add(out, string, end - string);
551 }
552
553 if (*end)
554 end++;
555 return end;
556 }
557
558 static void link_alt_odb_entries(struct repository *r, const char *alt,
559 int sep, const char *relative_base, int depth)
560 {
561 struct strbuf objdirbuf = STRBUF_INIT;
562 struct strbuf entry = STRBUF_INIT;
563
564 if (!alt || !*alt)
565 return;
566
567 if (depth > 5) {
568 error(_("%s: ignoring alternate object stores, nesting too deep"),
569 relative_base);
570 return;
571 }
572
573 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
574 if (strbuf_normalize_path(&objdirbuf) < 0)
575 die(_("unable to normalize object directory: %s"),
576 objdirbuf.buf);
577
578 while (*alt) {
579 alt = parse_alt_odb_entry(alt, sep, &entry);
580 if (!entry.len)
581 continue;
582 link_alt_odb_entry(r, entry.buf,
583 relative_base, depth, objdirbuf.buf);
584 }
585 strbuf_release(&entry);
586 strbuf_release(&objdirbuf);
587 }
588
589 static void read_info_alternates(struct repository *r,
590 const char *relative_base,
591 int depth)
592 {
593 char *path;
594 struct strbuf buf = STRBUF_INIT;
595
596 path = xstrfmt("%s/info/alternates", relative_base);
597 if (strbuf_read_file(&buf, path, 1024) < 0) {
598 warn_on_fopen_errors(path);
599 free(path);
600 return;
601 }
602
603 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
604 strbuf_release(&buf);
605 free(path);
606 }
607
608 void add_to_alternates_file(const char *reference)
609 {
610 struct lock_file lock = LOCK_INIT;
611 char *alts = git_pathdup("objects/info/alternates");
612 FILE *in, *out;
613 int found = 0;
614
615 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
616 out = fdopen_lock_file(&lock, "w");
617 if (!out)
618 die_errno(_("unable to fdopen alternates lockfile"));
619
620 in = fopen(alts, "r");
621 if (in) {
622 struct strbuf line = STRBUF_INIT;
623
624 while (strbuf_getline(&line, in) != EOF) {
625 if (!strcmp(reference, line.buf)) {
626 found = 1;
627 break;
628 }
629 fprintf_or_die(out, "%s\n", line.buf);
630 }
631
632 strbuf_release(&line);
633 fclose(in);
634 }
635 else if (errno != ENOENT)
636 die_errno(_("unable to read alternates file"));
637
638 if (found) {
639 rollback_lock_file(&lock);
640 } else {
641 fprintf_or_die(out, "%s\n", reference);
642 if (commit_lock_file(&lock))
643 die_errno(_("unable to move new alternates file into place"));
644 if (the_repository->objects->loaded_alternates)
645 link_alt_odb_entries(the_repository, reference,
646 '\n', NULL, 0);
647 }
648 free(alts);
649 }
650
651 void add_to_alternates_memory(const char *reference)
652 {
653 /*
654 * Make sure alternates are initialized, or else our entry may be
655 * overwritten when they are.
656 */
657 prepare_alt_odb(the_repository);
658
659 link_alt_odb_entries(the_repository, reference,
660 '\n', NULL, 0);
661 }
662
663 /*
664 * Compute the exact path an alternate is at and returns it. In case of
665 * error NULL is returned and the human readable error is added to `err`
666 * `path` may be relative and should point to $GIT_DIR.
667 * `err` must not be null.
668 */
669 char *compute_alternate_path(const char *path, struct strbuf *err)
670 {
671 char *ref_git = NULL;
672 const char *repo, *ref_git_s;
673 int seen_error = 0;
674
675 ref_git_s = real_path_if_valid(path);
676 if (!ref_git_s) {
677 seen_error = 1;
678 strbuf_addf(err, _("path '%s' does not exist"), path);
679 goto out;
680 } else
681 /*
682 * Beware: read_gitfile(), real_path() and mkpath()
683 * return static buffer
684 */
685 ref_git = xstrdup(ref_git_s);
686
687 repo = read_gitfile(ref_git);
688 if (!repo)
689 repo = read_gitfile(mkpath("%s/.git", ref_git));
690 if (repo) {
691 free(ref_git);
692 ref_git = xstrdup(repo);
693 }
694
695 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
696 char *ref_git_git = mkpathdup("%s/.git", ref_git);
697 free(ref_git);
698 ref_git = ref_git_git;
699 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
700 struct strbuf sb = STRBUF_INIT;
701 seen_error = 1;
702 if (get_common_dir(&sb, ref_git)) {
703 strbuf_addf(err,
704 _("reference repository '%s' as a linked "
705 "checkout is not supported yet."),
706 path);
707 goto out;
708 }
709
710 strbuf_addf(err, _("reference repository '%s' is not a "
711 "local repository."), path);
712 goto out;
713 }
714
715 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
716 strbuf_addf(err, _("reference repository '%s' is shallow"),
717 path);
718 seen_error = 1;
719 goto out;
720 }
721
722 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
723 strbuf_addf(err,
724 _("reference repository '%s' is grafted"),
725 path);
726 seen_error = 1;
727 goto out;
728 }
729
730 out:
731 if (seen_error) {
732 FREE_AND_NULL(ref_git);
733 }
734
735 return ref_git;
736 }
737
738 int foreach_alt_odb(alt_odb_fn fn, void *cb)
739 {
740 struct object_directory *ent;
741 int r = 0;
742
743 prepare_alt_odb(the_repository);
744 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
745 r = fn(ent, cb);
746 if (r)
747 break;
748 }
749 return r;
750 }
751
752 void prepare_alt_odb(struct repository *r)
753 {
754 if (r->objects->loaded_alternates)
755 return;
756
757 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
758
759 read_info_alternates(r, r->objects->odb->path, 0);
760 r->objects->loaded_alternates = 1;
761 }
762
763 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
764 static int freshen_file(const char *fn)
765 {
766 struct utimbuf t;
767 t.actime = t.modtime = time(NULL);
768 return !utime(fn, &t);
769 }
770
771 /*
772 * All of the check_and_freshen functions return 1 if the file exists and was
773 * freshened (if freshening was requested), 0 otherwise. If they return
774 * 0, you should not assume that it is safe to skip a write of the object (it
775 * either does not exist on disk, or has a stale mtime and may be subject to
776 * pruning).
777 */
778 int check_and_freshen_file(const char *fn, int freshen)
779 {
780 if (access(fn, F_OK))
781 return 0;
782 if (freshen && !freshen_file(fn))
783 return 0;
784 return 1;
785 }
786
787 static int check_and_freshen_odb(struct object_directory *odb,
788 const struct object_id *oid,
789 int freshen)
790 {
791 static struct strbuf path = STRBUF_INIT;
792 odb_loose_path(odb, &path, oid);
793 return check_and_freshen_file(path.buf, freshen);
794 }
795
796 static int check_and_freshen_local(const struct object_id *oid, int freshen)
797 {
798 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
799 }
800
801 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
802 {
803 struct object_directory *odb;
804
805 prepare_alt_odb(the_repository);
806 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
807 if (check_and_freshen_odb(odb, oid, freshen))
808 return 1;
809 }
810 return 0;
811 }
812
813 static int check_and_freshen(const struct object_id *oid, int freshen)
814 {
815 return check_and_freshen_local(oid, freshen) ||
816 check_and_freshen_nonlocal(oid, freshen);
817 }
818
819 int has_loose_object_nonlocal(const struct object_id *oid)
820 {
821 return check_and_freshen_nonlocal(oid, 0);
822 }
823
824 static int has_loose_object(const struct object_id *oid)
825 {
826 return check_and_freshen(oid, 0);
827 }
828
829 static void mmap_limit_check(size_t length)
830 {
831 static size_t limit = 0;
832 if (!limit) {
833 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
834 if (!limit)
835 limit = SIZE_MAX;
836 }
837 if (length > limit)
838 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
839 (uintmax_t)length, (uintmax_t)limit);
840 }
841
842 void *xmmap_gently(void *start, size_t length,
843 int prot, int flags, int fd, off_t offset)
844 {
845 void *ret;
846
847 mmap_limit_check(length);
848 ret = mmap(start, length, prot, flags, fd, offset);
849 if (ret == MAP_FAILED) {
850 if (!length)
851 return NULL;
852 release_pack_memory(length);
853 ret = mmap(start, length, prot, flags, fd, offset);
854 }
855 return ret;
856 }
857
858 void *xmmap(void *start, size_t length,
859 int prot, int flags, int fd, off_t offset)
860 {
861 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
862 if (ret == MAP_FAILED)
863 die_errno(_("mmap failed"));
864 return ret;
865 }
866
867 /*
868 * With an in-core object data in "map", rehash it to make sure the
869 * object name actually matches "oid" to detect object corruption.
870 * With "map" == NULL, try reading the object named with "oid" using
871 * the streaming interface and rehash it to do the same.
872 */
873 int check_object_signature(const struct object_id *oid, void *map,
874 unsigned long size, const char *type)
875 {
876 struct object_id real_oid;
877 enum object_type obj_type;
878 struct git_istream *st;
879 git_hash_ctx c;
880 char hdr[MAX_HEADER_LEN];
881 int hdrlen;
882
883 if (map) {
884 hash_object_file(map, size, type, &real_oid);
885 return !oideq(oid, &real_oid) ? -1 : 0;
886 }
887
888 st = open_istream(oid, &obj_type, &size, NULL);
889 if (!st)
890 return -1;
891
892 /* Generate the header */
893 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
894
895 /* Sha1.. */
896 the_hash_algo->init_fn(&c);
897 the_hash_algo->update_fn(&c, hdr, hdrlen);
898 for (;;) {
899 char buf[1024 * 16];
900 ssize_t readlen = read_istream(st, buf, sizeof(buf));
901
902 if (readlen < 0) {
903 close_istream(st);
904 return -1;
905 }
906 if (!readlen)
907 break;
908 the_hash_algo->update_fn(&c, buf, readlen);
909 }
910 the_hash_algo->final_fn(real_oid.hash, &c);
911 close_istream(st);
912 return !oideq(oid, &real_oid) ? -1 : 0;
913 }
914
915 int git_open_cloexec(const char *name, int flags)
916 {
917 int fd;
918 static int o_cloexec = O_CLOEXEC;
919
920 fd = open(name, flags | o_cloexec);
921 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
922 /* Try again w/o O_CLOEXEC: the kernel might not support it */
923 o_cloexec &= ~O_CLOEXEC;
924 fd = open(name, flags | o_cloexec);
925 }
926
927 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
928 {
929 static int fd_cloexec = FD_CLOEXEC;
930
931 if (!o_cloexec && 0 <= fd && fd_cloexec) {
932 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
933 int flags = fcntl(fd, F_GETFD);
934 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
935 fd_cloexec = 0;
936 }
937 }
938 #endif
939 return fd;
940 }
941
942 /*
943 * Find "oid" as a loose object in the local repository or in an alternate.
944 * Returns 0 on success, negative on failure.
945 *
946 * The "path" out-parameter will give the path of the object we found (if any).
947 * Note that it may point to static storage and is only valid until another
948 * call to stat_loose_object().
949 */
950 static int stat_loose_object(struct repository *r, const struct object_id *oid,
951 struct stat *st, const char **path)
952 {
953 struct object_directory *odb;
954 static struct strbuf buf = STRBUF_INIT;
955
956 prepare_alt_odb(r);
957 for (odb = r->objects->odb; odb; odb = odb->next) {
958 *path = odb_loose_path(odb, &buf, oid);
959 if (!lstat(*path, st))
960 return 0;
961 }
962
963 return -1;
964 }
965
966 /*
967 * Like stat_loose_object(), but actually open the object and return the
968 * descriptor. See the caveats on the "path" parameter above.
969 */
970 static int open_loose_object(struct repository *r,
971 const struct object_id *oid, const char **path)
972 {
973 int fd;
974 struct object_directory *odb;
975 int most_interesting_errno = ENOENT;
976 static struct strbuf buf = STRBUF_INIT;
977
978 prepare_alt_odb(r);
979 for (odb = r->objects->odb; odb; odb = odb->next) {
980 *path = odb_loose_path(odb, &buf, oid);
981 fd = git_open(*path);
982 if (fd >= 0)
983 return fd;
984
985 if (most_interesting_errno == ENOENT)
986 most_interesting_errno = errno;
987 }
988 errno = most_interesting_errno;
989 return -1;
990 }
991
992 static int quick_has_loose(struct repository *r,
993 const struct object_id *oid)
994 {
995 struct object_directory *odb;
996
997 prepare_alt_odb(r);
998 for (odb = r->objects->odb; odb; odb = odb->next) {
999 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1000 return 1;
1001 }
1002 return 0;
1003 }
1004
1005 /*
1006 * Map the loose object at "path" if it is not NULL, or the path found by
1007 * searching for a loose object named "oid".
1008 */
1009 static void *map_loose_object_1(struct repository *r, const char *path,
1010 const struct object_id *oid, unsigned long *size)
1011 {
1012 void *map;
1013 int fd;
1014
1015 if (path)
1016 fd = git_open(path);
1017 else
1018 fd = open_loose_object(r, oid, &path);
1019 map = NULL;
1020 if (fd >= 0) {
1021 struct stat st;
1022
1023 if (!fstat(fd, &st)) {
1024 *size = xsize_t(st.st_size);
1025 if (!*size) {
1026 /* mmap() is forbidden on empty files */
1027 error(_("object file %s is empty"), path);
1028 close(fd);
1029 return NULL;
1030 }
1031 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1032 }
1033 close(fd);
1034 }
1035 return map;
1036 }
1037
1038 void *map_loose_object(struct repository *r,
1039 const struct object_id *oid,
1040 unsigned long *size)
1041 {
1042 return map_loose_object_1(r, NULL, oid, size);
1043 }
1044
1045 static int unpack_loose_short_header(git_zstream *stream,
1046 unsigned char *map, unsigned long mapsize,
1047 void *buffer, unsigned long bufsiz)
1048 {
1049 /* Get the data stream */
1050 memset(stream, 0, sizeof(*stream));
1051 stream->next_in = map;
1052 stream->avail_in = mapsize;
1053 stream->next_out = buffer;
1054 stream->avail_out = bufsiz;
1055
1056 git_inflate_init(stream);
1057 return git_inflate(stream, 0);
1058 }
1059
1060 int unpack_loose_header(git_zstream *stream,
1061 unsigned char *map, unsigned long mapsize,
1062 void *buffer, unsigned long bufsiz)
1063 {
1064 int status = unpack_loose_short_header(stream, map, mapsize,
1065 buffer, bufsiz);
1066
1067 if (status < Z_OK)
1068 return status;
1069
1070 /* Make sure we have the terminating NUL */
1071 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1072 return -1;
1073 return 0;
1074 }
1075
1076 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1077 unsigned long mapsize, void *buffer,
1078 unsigned long bufsiz, struct strbuf *header)
1079 {
1080 int status;
1081
1082 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1083 if (status < Z_OK)
1084 return -1;
1085
1086 /*
1087 * Check if entire header is unpacked in the first iteration.
1088 */
1089 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1090 return 0;
1091
1092 /*
1093 * buffer[0..bufsiz] was not large enough. Copy the partial
1094 * result out to header, and then append the result of further
1095 * reading the stream.
1096 */
1097 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1098 stream->next_out = buffer;
1099 stream->avail_out = bufsiz;
1100
1101 do {
1102 status = git_inflate(stream, 0);
1103 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1104 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1105 return 0;
1106 stream->next_out = buffer;
1107 stream->avail_out = bufsiz;
1108 } while (status != Z_STREAM_END);
1109 return -1;
1110 }
1111
1112 static void *unpack_loose_rest(git_zstream *stream,
1113 void *buffer, unsigned long size,
1114 const struct object_id *oid)
1115 {
1116 int bytes = strlen(buffer) + 1;
1117 unsigned char *buf = xmallocz(size);
1118 unsigned long n;
1119 int status = Z_OK;
1120
1121 n = stream->total_out - bytes;
1122 if (n > size)
1123 n = size;
1124 memcpy(buf, (char *) buffer + bytes, n);
1125 bytes = n;
1126 if (bytes <= size) {
1127 /*
1128 * The above condition must be (bytes <= size), not
1129 * (bytes < size). In other words, even though we
1130 * expect no more output and set avail_out to zero,
1131 * the input zlib stream may have bytes that express
1132 * "this concludes the stream", and we *do* want to
1133 * eat that input.
1134 *
1135 * Otherwise we would not be able to test that we
1136 * consumed all the input to reach the expected size;
1137 * we also want to check that zlib tells us that all
1138 * went well with status == Z_STREAM_END at the end.
1139 */
1140 stream->next_out = buf + bytes;
1141 stream->avail_out = size - bytes;
1142 while (status == Z_OK)
1143 status = git_inflate(stream, Z_FINISH);
1144 }
1145 if (status == Z_STREAM_END && !stream->avail_in) {
1146 git_inflate_end(stream);
1147 return buf;
1148 }
1149
1150 if (status < 0)
1151 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1152 else if (stream->avail_in)
1153 error(_("garbage at end of loose object '%s'"),
1154 oid_to_hex(oid));
1155 free(buf);
1156 return NULL;
1157 }
1158
1159 /*
1160 * We used to just use "sscanf()", but that's actually way
1161 * too permissive for what we want to check. So do an anal
1162 * object header parse by hand.
1163 */
1164 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1165 unsigned int flags)
1166 {
1167 const char *type_buf = hdr;
1168 unsigned long size;
1169 int type, type_len = 0;
1170
1171 /*
1172 * The type can be of any size but is followed by
1173 * a space.
1174 */
1175 for (;;) {
1176 char c = *hdr++;
1177 if (!c)
1178 return -1;
1179 if (c == ' ')
1180 break;
1181 type_len++;
1182 }
1183
1184 type = type_from_string_gently(type_buf, type_len, 1);
1185 if (oi->type_name)
1186 strbuf_add(oi->type_name, type_buf, type_len);
1187 /*
1188 * Set type to 0 if its an unknown object and
1189 * we're obtaining the type using '--allow-unknown-type'
1190 * option.
1191 */
1192 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1193 type = 0;
1194 else if (type < 0)
1195 die(_("invalid object type"));
1196 if (oi->typep)
1197 *oi->typep = type;
1198
1199 /*
1200 * The length must follow immediately, and be in canonical
1201 * decimal format (ie "010" is not valid).
1202 */
1203 size = *hdr++ - '0';
1204 if (size > 9)
1205 return -1;
1206 if (size) {
1207 for (;;) {
1208 unsigned long c = *hdr - '0';
1209 if (c > 9)
1210 break;
1211 hdr++;
1212 size = size * 10 + c;
1213 }
1214 }
1215
1216 if (oi->sizep)
1217 *oi->sizep = size;
1218
1219 /*
1220 * The length must be followed by a zero byte
1221 */
1222 return *hdr ? -1 : type;
1223 }
1224
1225 int parse_loose_header(const char *hdr, unsigned long *sizep)
1226 {
1227 struct object_info oi = OBJECT_INFO_INIT;
1228
1229 oi.sizep = sizep;
1230 return parse_loose_header_extended(hdr, &oi, 0);
1231 }
1232
1233 static int loose_object_info(struct repository *r,
1234 const struct object_id *oid,
1235 struct object_info *oi, int flags)
1236 {
1237 int status = 0;
1238 unsigned long mapsize;
1239 void *map;
1240 git_zstream stream;
1241 char hdr[MAX_HEADER_LEN];
1242 struct strbuf hdrbuf = STRBUF_INIT;
1243 unsigned long size_scratch;
1244
1245 if (oi->delta_base_sha1)
1246 hashclr(oi->delta_base_sha1);
1247
1248 /*
1249 * If we don't care about type or size, then we don't
1250 * need to look inside the object at all. Note that we
1251 * do not optimize out the stat call, even if the
1252 * caller doesn't care about the disk-size, since our
1253 * return value implicitly indicates whether the
1254 * object even exists.
1255 */
1256 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1257 const char *path;
1258 struct stat st;
1259 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1260 return quick_has_loose(r, oid) ? 0 : -1;
1261 if (stat_loose_object(r, oid, &st, &path) < 0)
1262 return -1;
1263 if (oi->disk_sizep)
1264 *oi->disk_sizep = st.st_size;
1265 return 0;
1266 }
1267
1268 map = map_loose_object(r, oid, &mapsize);
1269 if (!map)
1270 return -1;
1271
1272 if (!oi->sizep)
1273 oi->sizep = &size_scratch;
1274
1275 if (oi->disk_sizep)
1276 *oi->disk_sizep = mapsize;
1277 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1278 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1279 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1280 oid_to_hex(oid));
1281 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1282 status = error(_("unable to unpack %s header"),
1283 oid_to_hex(oid));
1284 if (status < 0)
1285 ; /* Do nothing */
1286 else if (hdrbuf.len) {
1287 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1288 status = error(_("unable to parse %s header with --allow-unknown-type"),
1289 oid_to_hex(oid));
1290 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1291 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1292
1293 if (status >= 0 && oi->contentp) {
1294 *oi->contentp = unpack_loose_rest(&stream, hdr,
1295 *oi->sizep, oid);
1296 if (!*oi->contentp) {
1297 git_inflate_end(&stream);
1298 status = -1;
1299 }
1300 } else
1301 git_inflate_end(&stream);
1302
1303 munmap(map, mapsize);
1304 if (status && oi->typep)
1305 *oi->typep = status;
1306 if (oi->sizep == &size_scratch)
1307 oi->sizep = NULL;
1308 strbuf_release(&hdrbuf);
1309 oi->whence = OI_LOOSE;
1310 return (status < 0) ? status : 0;
1311 }
1312
1313 int fetch_if_missing = 1;
1314
1315 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1316 struct object_info *oi, unsigned flags)
1317 {
1318 static struct object_info blank_oi = OBJECT_INFO_INIT;
1319 struct pack_entry e;
1320 int rtype;
1321 const struct object_id *real = oid;
1322 int already_retried = 0;
1323
1324 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1325 real = lookup_replace_object(r, oid);
1326
1327 if (is_null_oid(real))
1328 return -1;
1329
1330 if (!oi)
1331 oi = &blank_oi;
1332
1333 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1334 struct cached_object *co = find_cached_object(real);
1335 if (co) {
1336 if (oi->typep)
1337 *(oi->typep) = co->type;
1338 if (oi->sizep)
1339 *(oi->sizep) = co->size;
1340 if (oi->disk_sizep)
1341 *(oi->disk_sizep) = 0;
1342 if (oi->delta_base_sha1)
1343 hashclr(oi->delta_base_sha1);
1344 if (oi->type_name)
1345 strbuf_addstr(oi->type_name, type_name(co->type));
1346 if (oi->contentp)
1347 *oi->contentp = xmemdupz(co->buf, co->size);
1348 oi->whence = OI_CACHED;
1349 return 0;
1350 }
1351 }
1352
1353 while (1) {
1354 if (find_pack_entry(r, real, &e))
1355 break;
1356
1357 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1358 return -1;
1359
1360 /* Most likely it's a loose object. */
1361 if (!loose_object_info(r, real, oi, flags))
1362 return 0;
1363
1364 /* Not a loose object; someone else may have just packed it. */
1365 if (!(flags & OBJECT_INFO_QUICK)) {
1366 reprepare_packed_git(r);
1367 if (find_pack_entry(r, real, &e))
1368 break;
1369 }
1370
1371 /* Check if it is a missing object */
1372 if (fetch_if_missing && repository_format_partial_clone &&
1373 !already_retried && r == the_repository &&
1374 !(flags & OBJECT_INFO_FOR_PREFETCH)) {
1375 /*
1376 * TODO Investigate having fetch_object() return
1377 * TODO error/success and stopping the music here.
1378 * TODO Pass a repository struct through fetch_object,
1379 * such that arbitrary repositories work.
1380 */
1381 fetch_objects(repository_format_partial_clone, real, 1);
1382 already_retried = 1;
1383 continue;
1384 }
1385
1386 return -1;
1387 }
1388
1389 if (oi == &blank_oi)
1390 /*
1391 * We know that the caller doesn't actually need the
1392 * information below, so return early.
1393 */
1394 return 0;
1395 rtype = packed_object_info(r, e.p, e.offset, oi);
1396 if (rtype < 0) {
1397 mark_bad_packed_object(e.p, real->hash);
1398 return oid_object_info_extended(r, real, oi, 0);
1399 } else if (oi->whence == OI_PACKED) {
1400 oi->u.packed.offset = e.offset;
1401 oi->u.packed.pack = e.p;
1402 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1403 rtype == OBJ_OFS_DELTA);
1404 }
1405
1406 return 0;
1407 }
1408
1409 /* returns enum object_type or negative */
1410 int oid_object_info(struct repository *r,
1411 const struct object_id *oid,
1412 unsigned long *sizep)
1413 {
1414 enum object_type type;
1415 struct object_info oi = OBJECT_INFO_INIT;
1416
1417 oi.typep = &type;
1418 oi.sizep = sizep;
1419 if (oid_object_info_extended(r, oid, &oi,
1420 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1421 return -1;
1422 return type;
1423 }
1424
1425 static void *read_object(struct repository *r,
1426 const struct object_id *oid, enum object_type *type,
1427 unsigned long *size)
1428 {
1429 struct object_info oi = OBJECT_INFO_INIT;
1430 void *content;
1431 oi.typep = type;
1432 oi.sizep = size;
1433 oi.contentp = &content;
1434
1435 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1436 return NULL;
1437 return content;
1438 }
1439
1440 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1441 struct object_id *oid)
1442 {
1443 struct cached_object *co;
1444
1445 hash_object_file(buf, len, type_name(type), oid);
1446 if (has_object_file(oid) || find_cached_object(oid))
1447 return 0;
1448 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1449 co = &cached_objects[cached_object_nr++];
1450 co->size = len;
1451 co->type = type;
1452 co->buf = xmalloc(len);
1453 memcpy(co->buf, buf, len);
1454 oidcpy(&co->oid, oid);
1455 return 0;
1456 }
1457
1458 /*
1459 * This function dies on corrupt objects; the callers who want to
1460 * deal with them should arrange to call read_object() and give error
1461 * messages themselves.
1462 */
1463 void *read_object_file_extended(struct repository *r,
1464 const struct object_id *oid,
1465 enum object_type *type,
1466 unsigned long *size,
1467 int lookup_replace)
1468 {
1469 void *data;
1470 const struct packed_git *p;
1471 const char *path;
1472 struct stat st;
1473 const struct object_id *repl = lookup_replace ?
1474 lookup_replace_object(r, oid) : oid;
1475
1476 errno = 0;
1477 data = read_object(r, repl, type, size);
1478 if (data)
1479 return data;
1480
1481 if (errno && errno != ENOENT)
1482 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1483
1484 /* die if we replaced an object with one that does not exist */
1485 if (repl != oid)
1486 die(_("replacement %s not found for %s"),
1487 oid_to_hex(repl), oid_to_hex(oid));
1488
1489 if (!stat_loose_object(r, repl, &st, &path))
1490 die(_("loose object %s (stored in %s) is corrupt"),
1491 oid_to_hex(repl), path);
1492
1493 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1494 die(_("packed object %s (stored in %s) is corrupt"),
1495 oid_to_hex(repl), p->pack_name);
1496
1497 return NULL;
1498 }
1499
1500 void *read_object_with_reference(const struct object_id *oid,
1501 const char *required_type_name,
1502 unsigned long *size,
1503 struct object_id *actual_oid_return)
1504 {
1505 enum object_type type, required_type;
1506 void *buffer;
1507 unsigned long isize;
1508 struct object_id actual_oid;
1509
1510 required_type = type_from_string(required_type_name);
1511 oidcpy(&actual_oid, oid);
1512 while (1) {
1513 int ref_length = -1;
1514 const char *ref_type = NULL;
1515
1516 buffer = read_object_file(&actual_oid, &type, &isize);
1517 if (!buffer)
1518 return NULL;
1519 if (type == required_type) {
1520 *size = isize;
1521 if (actual_oid_return)
1522 oidcpy(actual_oid_return, &actual_oid);
1523 return buffer;
1524 }
1525 /* Handle references */
1526 else if (type == OBJ_COMMIT)
1527 ref_type = "tree ";
1528 else if (type == OBJ_TAG)
1529 ref_type = "object ";
1530 else {
1531 free(buffer);
1532 return NULL;
1533 }
1534 ref_length = strlen(ref_type);
1535
1536 if (ref_length + the_hash_algo->hexsz > isize ||
1537 memcmp(buffer, ref_type, ref_length) ||
1538 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1539 free(buffer);
1540 return NULL;
1541 }
1542 free(buffer);
1543 /* Now we have the ID of the referred-to object in
1544 * actual_oid. Check again. */
1545 }
1546 }
1547
1548 static void write_object_file_prepare(const void *buf, unsigned long len,
1549 const char *type, struct object_id *oid,
1550 char *hdr, int *hdrlen)
1551 {
1552 git_hash_ctx c;
1553
1554 /* Generate the header */
1555 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1556
1557 /* Sha1.. */
1558 the_hash_algo->init_fn(&c);
1559 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1560 the_hash_algo->update_fn(&c, buf, len);
1561 the_hash_algo->final_fn(oid->hash, &c);
1562 }
1563
1564 /*
1565 * Move the just written object into its final resting place.
1566 */
1567 int finalize_object_file(const char *tmpfile, const char *filename)
1568 {
1569 int ret = 0;
1570
1571 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1572 goto try_rename;
1573 else if (link(tmpfile, filename))
1574 ret = errno;
1575
1576 /*
1577 * Coda hack - coda doesn't like cross-directory links,
1578 * so we fall back to a rename, which will mean that it
1579 * won't be able to check collisions, but that's not a
1580 * big deal.
1581 *
1582 * The same holds for FAT formatted media.
1583 *
1584 * When this succeeds, we just return. We have nothing
1585 * left to unlink.
1586 */
1587 if (ret && ret != EEXIST) {
1588 try_rename:
1589 if (!rename(tmpfile, filename))
1590 goto out;
1591 ret = errno;
1592 }
1593 unlink_or_warn(tmpfile);
1594 if (ret) {
1595 if (ret != EEXIST) {
1596 return error_errno(_("unable to write file %s"), filename);
1597 }
1598 /* FIXME!!! Collision check here ? */
1599 }
1600
1601 out:
1602 if (adjust_shared_perm(filename))
1603 return error(_("unable to set permission to '%s'"), filename);
1604 return 0;
1605 }
1606
1607 static int write_buffer(int fd, const void *buf, size_t len)
1608 {
1609 if (write_in_full(fd, buf, len) < 0)
1610 return error_errno(_("file write error"));
1611 return 0;
1612 }
1613
1614 int hash_object_file(const void *buf, unsigned long len, const char *type,
1615 struct object_id *oid)
1616 {
1617 char hdr[MAX_HEADER_LEN];
1618 int hdrlen = sizeof(hdr);
1619 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1620 return 0;
1621 }
1622
1623 /* Finalize a file on disk, and close it. */
1624 static void close_loose_object(int fd)
1625 {
1626 if (fsync_object_files)
1627 fsync_or_die(fd, "loose object file");
1628 if (close(fd) != 0)
1629 die_errno(_("error when closing loose object file"));
1630 }
1631
1632 /* Size of directory component, including the ending '/' */
1633 static inline int directory_size(const char *filename)
1634 {
1635 const char *s = strrchr(filename, '/');
1636 if (!s)
1637 return 0;
1638 return s - filename + 1;
1639 }
1640
1641 /*
1642 * This creates a temporary file in the same directory as the final
1643 * 'filename'
1644 *
1645 * We want to avoid cross-directory filename renames, because those
1646 * can have problems on various filesystems (FAT, NFS, Coda).
1647 */
1648 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1649 {
1650 int fd, dirlen = directory_size(filename);
1651
1652 strbuf_reset(tmp);
1653 strbuf_add(tmp, filename, dirlen);
1654 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1655 fd = git_mkstemp_mode(tmp->buf, 0444);
1656 if (fd < 0 && dirlen && errno == ENOENT) {
1657 /*
1658 * Make sure the directory exists; note that the contents
1659 * of the buffer are undefined after mkstemp returns an
1660 * error, so we have to rewrite the whole buffer from
1661 * scratch.
1662 */
1663 strbuf_reset(tmp);
1664 strbuf_add(tmp, filename, dirlen - 1);
1665 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1666 return -1;
1667 if (adjust_shared_perm(tmp->buf))
1668 return -1;
1669
1670 /* Try again */
1671 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1672 fd = git_mkstemp_mode(tmp->buf, 0444);
1673 }
1674 return fd;
1675 }
1676
1677 static int write_loose_object(const struct object_id *oid, char *hdr,
1678 int hdrlen, const void *buf, unsigned long len,
1679 time_t mtime)
1680 {
1681 int fd, ret;
1682 unsigned char compressed[4096];
1683 git_zstream stream;
1684 git_hash_ctx c;
1685 struct object_id parano_oid;
1686 static struct strbuf tmp_file = STRBUF_INIT;
1687 static struct strbuf filename = STRBUF_INIT;
1688
1689 loose_object_path(the_repository, &filename, oid);
1690
1691 fd = create_tmpfile(&tmp_file, filename.buf);
1692 if (fd < 0) {
1693 if (errno == EACCES)
1694 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1695 else
1696 return error_errno(_("unable to create temporary file"));
1697 }
1698
1699 /* Set it up */
1700 git_deflate_init(&stream, zlib_compression_level);
1701 stream.next_out = compressed;
1702 stream.avail_out = sizeof(compressed);
1703 the_hash_algo->init_fn(&c);
1704
1705 /* First header.. */
1706 stream.next_in = (unsigned char *)hdr;
1707 stream.avail_in = hdrlen;
1708 while (git_deflate(&stream, 0) == Z_OK)
1709 ; /* nothing */
1710 the_hash_algo->update_fn(&c, hdr, hdrlen);
1711
1712 /* Then the data itself.. */
1713 stream.next_in = (void *)buf;
1714 stream.avail_in = len;
1715 do {
1716 unsigned char *in0 = stream.next_in;
1717 ret = git_deflate(&stream, Z_FINISH);
1718 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1719 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1720 die(_("unable to write loose object file"));
1721 stream.next_out = compressed;
1722 stream.avail_out = sizeof(compressed);
1723 } while (ret == Z_OK);
1724
1725 if (ret != Z_STREAM_END)
1726 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1727 ret);
1728 ret = git_deflate_end_gently(&stream);
1729 if (ret != Z_OK)
1730 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1731 ret);
1732 the_hash_algo->final_fn(parano_oid.hash, &c);
1733 if (!oideq(oid, &parano_oid))
1734 die(_("confused by unstable object source data for %s"),
1735 oid_to_hex(oid));
1736
1737 close_loose_object(fd);
1738
1739 if (mtime) {
1740 struct utimbuf utb;
1741 utb.actime = mtime;
1742 utb.modtime = mtime;
1743 if (utime(tmp_file.buf, &utb) < 0)
1744 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1745 }
1746
1747 return finalize_object_file(tmp_file.buf, filename.buf);
1748 }
1749
1750 static int freshen_loose_object(const struct object_id *oid)
1751 {
1752 return check_and_freshen(oid, 1);
1753 }
1754
1755 static int freshen_packed_object(const struct object_id *oid)
1756 {
1757 struct pack_entry e;
1758 if (!find_pack_entry(the_repository, oid, &e))
1759 return 0;
1760 if (e.p->freshened)
1761 return 1;
1762 if (!freshen_file(e.p->pack_name))
1763 return 0;
1764 e.p->freshened = 1;
1765 return 1;
1766 }
1767
1768 int write_object_file(const void *buf, unsigned long len, const char *type,
1769 struct object_id *oid)
1770 {
1771 char hdr[MAX_HEADER_LEN];
1772 int hdrlen = sizeof(hdr);
1773
1774 /* Normally if we have it in the pack then we do not bother writing
1775 * it out into .git/objects/??/?{38} file.
1776 */
1777 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1778 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1779 return 0;
1780 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1781 }
1782
1783 int hash_object_file_literally(const void *buf, unsigned long len,
1784 const char *type, struct object_id *oid,
1785 unsigned flags)
1786 {
1787 char *header;
1788 int hdrlen, status = 0;
1789
1790 /* type string, SP, %lu of the length plus NUL must fit this */
1791 hdrlen = strlen(type) + MAX_HEADER_LEN;
1792 header = xmalloc(hdrlen);
1793 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1794
1795 if (!(flags & HASH_WRITE_OBJECT))
1796 goto cleanup;
1797 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1798 goto cleanup;
1799 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1800
1801 cleanup:
1802 free(header);
1803 return status;
1804 }
1805
1806 int force_object_loose(const struct object_id *oid, time_t mtime)
1807 {
1808 void *buf;
1809 unsigned long len;
1810 enum object_type type;
1811 char hdr[MAX_HEADER_LEN];
1812 int hdrlen;
1813 int ret;
1814
1815 if (has_loose_object(oid))
1816 return 0;
1817 buf = read_object(the_repository, oid, &type, &len);
1818 if (!buf)
1819 return error(_("cannot read object for %s"), oid_to_hex(oid));
1820 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1821 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1822 free(buf);
1823
1824 return ret;
1825 }
1826
1827 int repo_has_object_file_with_flags(struct repository *r,
1828 const struct object_id *oid, int flags)
1829 {
1830 if (!startup_info->have_repository)
1831 return 0;
1832 return oid_object_info_extended(r, oid, NULL,
1833 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1834 }
1835
1836 int repo_has_object_file(struct repository *r,
1837 const struct object_id *oid)
1838 {
1839 return repo_has_object_file_with_flags(r, oid, 0);
1840 }
1841
1842 static void check_tree(const void *buf, size_t size)
1843 {
1844 struct tree_desc desc;
1845 struct name_entry entry;
1846
1847 init_tree_desc(&desc, buf, size);
1848 while (tree_entry(&desc, &entry))
1849 /* do nothing
1850 * tree_entry() will die() on malformed entries */
1851 ;
1852 }
1853
1854 static void check_commit(const void *buf, size_t size)
1855 {
1856 struct commit c;
1857 memset(&c, 0, sizeof(c));
1858 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1859 die(_("corrupt commit"));
1860 }
1861
1862 static void check_tag(const void *buf, size_t size)
1863 {
1864 struct tag t;
1865 memset(&t, 0, sizeof(t));
1866 if (parse_tag_buffer(the_repository, &t, buf, size))
1867 die(_("corrupt tag"));
1868 }
1869
1870 static int index_mem(struct index_state *istate,
1871 struct object_id *oid, void *buf, size_t size,
1872 enum object_type type,
1873 const char *path, unsigned flags)
1874 {
1875 int ret, re_allocated = 0;
1876 int write_object = flags & HASH_WRITE_OBJECT;
1877
1878 if (!type)
1879 type = OBJ_BLOB;
1880
1881 /*
1882 * Convert blobs to git internal format
1883 */
1884 if ((type == OBJ_BLOB) && path) {
1885 struct strbuf nbuf = STRBUF_INIT;
1886 if (convert_to_git(istate, path, buf, size, &nbuf,
1887 get_conv_flags(flags))) {
1888 buf = strbuf_detach(&nbuf, &size);
1889 re_allocated = 1;
1890 }
1891 }
1892 if (flags & HASH_FORMAT_CHECK) {
1893 if (type == OBJ_TREE)
1894 check_tree(buf, size);
1895 if (type == OBJ_COMMIT)
1896 check_commit(buf, size);
1897 if (type == OBJ_TAG)
1898 check_tag(buf, size);
1899 }
1900
1901 if (write_object)
1902 ret = write_object_file(buf, size, type_name(type), oid);
1903 else
1904 ret = hash_object_file(buf, size, type_name(type), oid);
1905 if (re_allocated)
1906 free(buf);
1907 return ret;
1908 }
1909
1910 static int index_stream_convert_blob(struct index_state *istate,
1911 struct object_id *oid,
1912 int fd,
1913 const char *path,
1914 unsigned flags)
1915 {
1916 int ret;
1917 const int write_object = flags & HASH_WRITE_OBJECT;
1918 struct strbuf sbuf = STRBUF_INIT;
1919
1920 assert(path);
1921 assert(would_convert_to_git_filter_fd(istate, path));
1922
1923 convert_to_git_filter_fd(istate, path, fd, &sbuf,
1924 get_conv_flags(flags));
1925
1926 if (write_object)
1927 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1928 oid);
1929 else
1930 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1931 oid);
1932 strbuf_release(&sbuf);
1933 return ret;
1934 }
1935
1936 static int index_pipe(struct index_state *istate, struct object_id *oid,
1937 int fd, enum object_type type,
1938 const char *path, unsigned flags)
1939 {
1940 struct strbuf sbuf = STRBUF_INIT;
1941 int ret;
1942
1943 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1944 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
1945 else
1946 ret = -1;
1947 strbuf_release(&sbuf);
1948 return ret;
1949 }
1950
1951 #define SMALL_FILE_SIZE (32*1024)
1952
1953 static int index_core(struct index_state *istate,
1954 struct object_id *oid, int fd, size_t size,
1955 enum object_type type, const char *path,
1956 unsigned flags)
1957 {
1958 int ret;
1959
1960 if (!size) {
1961 ret = index_mem(istate, oid, "", size, type, path, flags);
1962 } else if (size <= SMALL_FILE_SIZE) {
1963 char *buf = xmalloc(size);
1964 ssize_t read_result = read_in_full(fd, buf, size);
1965 if (read_result < 0)
1966 ret = error_errno(_("read error while indexing %s"),
1967 path ? path : "<unknown>");
1968 else if (read_result != size)
1969 ret = error(_("short read while indexing %s"),
1970 path ? path : "<unknown>");
1971 else
1972 ret = index_mem(istate, oid, buf, size, type, path, flags);
1973 free(buf);
1974 } else {
1975 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1976 ret = index_mem(istate, oid, buf, size, type, path, flags);
1977 munmap(buf, size);
1978 }
1979 return ret;
1980 }
1981
1982 /*
1983 * This creates one packfile per large blob unless bulk-checkin
1984 * machinery is "plugged".
1985 *
1986 * This also bypasses the usual "convert-to-git" dance, and that is on
1987 * purpose. We could write a streaming version of the converting
1988 * functions and insert that before feeding the data to fast-import
1989 * (or equivalent in-core API described above). However, that is
1990 * somewhat complicated, as we do not know the size of the filter
1991 * result, which we need to know beforehand when writing a git object.
1992 * Since the primary motivation for trying to stream from the working
1993 * tree file and to avoid mmaping it in core is to deal with large
1994 * binary blobs, they generally do not want to get any conversion, and
1995 * callers should avoid this code path when filters are requested.
1996 */
1997 static int index_stream(struct object_id *oid, int fd, size_t size,
1998 enum object_type type, const char *path,
1999 unsigned flags)
2000 {
2001 return index_bulk_checkin(oid, fd, size, type, path, flags);
2002 }
2003
2004 int index_fd(struct index_state *istate, struct object_id *oid,
2005 int fd, struct stat *st,
2006 enum object_type type, const char *path, unsigned flags)
2007 {
2008 int ret;
2009
2010 /*
2011 * Call xsize_t() only when needed to avoid potentially unnecessary
2012 * die() for large files.
2013 */
2014 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2015 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2016 else if (!S_ISREG(st->st_mode))
2017 ret = index_pipe(istate, oid, fd, type, path, flags);
2018 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2019 (path && would_convert_to_git(istate, path)))
2020 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2021 type, path, flags);
2022 else
2023 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2024 flags);
2025 close(fd);
2026 return ret;
2027 }
2028
2029 int index_path(struct index_state *istate, struct object_id *oid,
2030 const char *path, struct stat *st, unsigned flags)
2031 {
2032 int fd;
2033 struct strbuf sb = STRBUF_INIT;
2034 int rc = 0;
2035
2036 switch (st->st_mode & S_IFMT) {
2037 case S_IFREG:
2038 fd = open(path, O_RDONLY);
2039 if (fd < 0)
2040 return error_errno("open(\"%s\")", path);
2041 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2042 return error(_("%s: failed to insert into database"),
2043 path);
2044 break;
2045 case S_IFLNK:
2046 if (strbuf_readlink(&sb, path, st->st_size))
2047 return error_errno("readlink(\"%s\")", path);
2048 if (!(flags & HASH_WRITE_OBJECT))
2049 hash_object_file(sb.buf, sb.len, blob_type, oid);
2050 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2051 rc = error(_("%s: failed to insert into database"), path);
2052 strbuf_release(&sb);
2053 break;
2054 case S_IFDIR:
2055 return resolve_gitlink_ref(path, "HEAD", oid);
2056 default:
2057 return error(_("%s: unsupported file type"), path);
2058 }
2059 return rc;
2060 }
2061
2062 int read_pack_header(int fd, struct pack_header *header)
2063 {
2064 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2065 /* "eof before pack header was fully read" */
2066 return PH_ERROR_EOF;
2067
2068 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2069 /* "protocol error (pack signature mismatch detected)" */
2070 return PH_ERROR_PACK_SIGNATURE;
2071 if (!pack_version_ok(header->hdr_version))
2072 /* "protocol error (pack version unsupported)" */
2073 return PH_ERROR_PROTOCOL;
2074 return 0;
2075 }
2076
2077 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2078 {
2079 enum object_type type = oid_object_info(the_repository, oid, NULL);
2080 if (type < 0)
2081 die(_("%s is not a valid object"), oid_to_hex(oid));
2082 if (type != expect)
2083 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2084 type_name(expect));
2085 }
2086
2087 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2088 struct strbuf *path,
2089 each_loose_object_fn obj_cb,
2090 each_loose_cruft_fn cruft_cb,
2091 each_loose_subdir_fn subdir_cb,
2092 void *data)
2093 {
2094 size_t origlen, baselen;
2095 DIR *dir;
2096 struct dirent *de;
2097 int r = 0;
2098 struct object_id oid;
2099
2100 if (subdir_nr > 0xff)
2101 BUG("invalid loose object subdirectory: %x", subdir_nr);
2102
2103 origlen = path->len;
2104 strbuf_complete(path, '/');
2105 strbuf_addf(path, "%02x", subdir_nr);
2106
2107 dir = opendir(path->buf);
2108 if (!dir) {
2109 if (errno != ENOENT)
2110 r = error_errno(_("unable to open %s"), path->buf);
2111 strbuf_setlen(path, origlen);
2112 return r;
2113 }
2114
2115 oid.hash[0] = subdir_nr;
2116 strbuf_addch(path, '/');
2117 baselen = path->len;
2118
2119 while ((de = readdir(dir))) {
2120 size_t namelen;
2121 if (is_dot_or_dotdot(de->d_name))
2122 continue;
2123
2124 namelen = strlen(de->d_name);
2125 strbuf_setlen(path, baselen);
2126 strbuf_add(path, de->d_name, namelen);
2127 if (namelen == the_hash_algo->hexsz - 2 &&
2128 !hex_to_bytes(oid.hash + 1, de->d_name,
2129 the_hash_algo->rawsz - 1)) {
2130 if (obj_cb) {
2131 r = obj_cb(&oid, path->buf, data);
2132 if (r)
2133 break;
2134 }
2135 continue;
2136 }
2137
2138 if (cruft_cb) {
2139 r = cruft_cb(de->d_name, path->buf, data);
2140 if (r)
2141 break;
2142 }
2143 }
2144 closedir(dir);
2145
2146 strbuf_setlen(path, baselen - 1);
2147 if (!r && subdir_cb)
2148 r = subdir_cb(subdir_nr, path->buf, data);
2149
2150 strbuf_setlen(path, origlen);
2151
2152 return r;
2153 }
2154
2155 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2156 each_loose_object_fn obj_cb,
2157 each_loose_cruft_fn cruft_cb,
2158 each_loose_subdir_fn subdir_cb,
2159 void *data)
2160 {
2161 int r = 0;
2162 int i;
2163
2164 for (i = 0; i < 256; i++) {
2165 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2166 subdir_cb, data);
2167 if (r)
2168 break;
2169 }
2170
2171 return r;
2172 }
2173
2174 int for_each_loose_file_in_objdir(const char *path,
2175 each_loose_object_fn obj_cb,
2176 each_loose_cruft_fn cruft_cb,
2177 each_loose_subdir_fn subdir_cb,
2178 void *data)
2179 {
2180 struct strbuf buf = STRBUF_INIT;
2181 int r;
2182
2183 strbuf_addstr(&buf, path);
2184 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2185 subdir_cb, data);
2186 strbuf_release(&buf);
2187
2188 return r;
2189 }
2190
2191 int for_each_loose_object(each_loose_object_fn cb, void *data,
2192 enum for_each_object_flags flags)
2193 {
2194 struct object_directory *odb;
2195
2196 prepare_alt_odb(the_repository);
2197 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2198 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2199 NULL, data);
2200 if (r)
2201 return r;
2202
2203 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2204 break;
2205 }
2206
2207 return 0;
2208 }
2209
2210 static int append_loose_object(const struct object_id *oid, const char *path,
2211 void *data)
2212 {
2213 oid_array_append(data, oid);
2214 return 0;
2215 }
2216
2217 struct oid_array *odb_loose_cache(struct object_directory *odb,
2218 const struct object_id *oid)
2219 {
2220 int subdir_nr = oid->hash[0];
2221 struct strbuf buf = STRBUF_INIT;
2222
2223 if (subdir_nr < 0 ||
2224 subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2225 BUG("subdir_nr out of range");
2226
2227 if (odb->loose_objects_subdir_seen[subdir_nr])
2228 return &odb->loose_objects_cache[subdir_nr];
2229
2230 strbuf_addstr(&buf, odb->path);
2231 for_each_file_in_obj_subdir(subdir_nr, &buf,
2232 append_loose_object,
2233 NULL, NULL,
2234 &odb->loose_objects_cache[subdir_nr]);
2235 odb->loose_objects_subdir_seen[subdir_nr] = 1;
2236 strbuf_release(&buf);
2237 return &odb->loose_objects_cache[subdir_nr];
2238 }
2239
2240 void odb_clear_loose_cache(struct object_directory *odb)
2241 {
2242 int i;
2243
2244 for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2245 oid_array_clear(&odb->loose_objects_cache[i]);
2246 memset(&odb->loose_objects_subdir_seen, 0,
2247 sizeof(odb->loose_objects_subdir_seen));
2248 }
2249
2250 static int check_stream_oid(git_zstream *stream,
2251 const char *hdr,
2252 unsigned long size,
2253 const char *path,
2254 const struct object_id *expected_oid)
2255 {
2256 git_hash_ctx c;
2257 struct object_id real_oid;
2258 unsigned char buf[4096];
2259 unsigned long total_read;
2260 int status = Z_OK;
2261
2262 the_hash_algo->init_fn(&c);
2263 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2264
2265 /*
2266 * We already read some bytes into hdr, but the ones up to the NUL
2267 * do not count against the object's content size.
2268 */
2269 total_read = stream->total_out - strlen(hdr) - 1;
2270
2271 /*
2272 * This size comparison must be "<=" to read the final zlib packets;
2273 * see the comment in unpack_loose_rest for details.
2274 */
2275 while (total_read <= size &&
2276 (status == Z_OK ||
2277 (status == Z_BUF_ERROR && !stream->avail_out))) {
2278 stream->next_out = buf;
2279 stream->avail_out = sizeof(buf);
2280 if (size - total_read < stream->avail_out)
2281 stream->avail_out = size - total_read;
2282 status = git_inflate(stream, Z_FINISH);
2283 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2284 total_read += stream->next_out - buf;
2285 }
2286 git_inflate_end(stream);
2287
2288 if (status != Z_STREAM_END) {
2289 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2290 return -1;
2291 }
2292 if (stream->avail_in) {
2293 error(_("garbage at end of loose object '%s'"),
2294 oid_to_hex(expected_oid));
2295 return -1;
2296 }
2297
2298 the_hash_algo->final_fn(real_oid.hash, &c);
2299 if (!oideq(expected_oid, &real_oid)) {
2300 error(_("hash mismatch for %s (expected %s)"), path,
2301 oid_to_hex(expected_oid));
2302 return -1;
2303 }
2304
2305 return 0;
2306 }
2307
2308 int read_loose_object(const char *path,
2309 const struct object_id *expected_oid,
2310 enum object_type *type,
2311 unsigned long *size,
2312 void **contents)
2313 {
2314 int ret = -1;
2315 void *map = NULL;
2316 unsigned long mapsize;
2317 git_zstream stream;
2318 char hdr[MAX_HEADER_LEN];
2319
2320 *contents = NULL;
2321
2322 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2323 if (!map) {
2324 error_errno(_("unable to mmap %s"), path);
2325 goto out;
2326 }
2327
2328 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2329 error(_("unable to unpack header of %s"), path);
2330 goto out;
2331 }
2332
2333 *type = parse_loose_header(hdr, size);
2334 if (*type < 0) {
2335 error(_("unable to parse header of %s"), path);
2336 git_inflate_end(&stream);
2337 goto out;
2338 }
2339
2340 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2341 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2342 goto out;
2343 } else {
2344 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2345 if (!*contents) {
2346 error(_("unable to unpack contents of %s"), path);
2347 git_inflate_end(&stream);
2348 goto out;
2349 }
2350 if (check_object_signature(expected_oid, *contents,
2351 *size, type_name(*type))) {
2352 error(_("hash mismatch for %s (expected %s)"), path,
2353 oid_to_hex(expected_oid));
2354 free(*contents);
2355 goto out;
2356 }
2357 }
2358
2359 ret = 0; /* everything checks out */
2360
2361 out:
2362 if (map)
2363 munmap(map, mapsize);
2364 return ret;
2365 }