promisor-remote: skip move_to_tail when no-op
[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 "object-store.h"
34 #include "promisor-remote.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 int hash_algo_by_length(int len)
193 {
194 int i;
195 for (i = 1; i < GIT_HASH_NALGOS; i++)
196 if (len == hash_algos[i].rawsz)
197 return i;
198 return GIT_HASH_UNKNOWN;
199 }
200
201 /*
202 * This is meant to hold a *small* number of objects that you would
203 * want read_object_file() to be able to return, but yet you do not want
204 * to write them into the object store (e.g. a browse-only
205 * application).
206 */
207 static struct cached_object {
208 struct object_id oid;
209 enum object_type type;
210 void *buf;
211 unsigned long size;
212 } *cached_objects;
213 static int cached_object_nr, cached_object_alloc;
214
215 static struct cached_object empty_tree = {
216 { EMPTY_TREE_SHA1_BIN_LITERAL },
217 OBJ_TREE,
218 "",
219 0
220 };
221
222 static struct cached_object *find_cached_object(const struct object_id *oid)
223 {
224 int i;
225 struct cached_object *co = cached_objects;
226
227 for (i = 0; i < cached_object_nr; i++, co++) {
228 if (oideq(&co->oid, oid))
229 return co;
230 }
231 if (oideq(oid, the_hash_algo->empty_tree))
232 return &empty_tree;
233 return NULL;
234 }
235
236
237 static int get_conv_flags(unsigned flags)
238 {
239 if (flags & HASH_RENORMALIZE)
240 return CONV_EOL_RENORMALIZE;
241 else if (flags & HASH_WRITE_OBJECT)
242 return global_conv_flags_eol | CONV_WRITE_OBJECT;
243 else
244 return 0;
245 }
246
247
248 int mkdir_in_gitdir(const char *path)
249 {
250 if (mkdir(path, 0777)) {
251 int saved_errno = errno;
252 struct stat st;
253 struct strbuf sb = STRBUF_INIT;
254
255 if (errno != EEXIST)
256 return -1;
257 /*
258 * Are we looking at a path in a symlinked worktree
259 * whose original repository does not yet have it?
260 * e.g. .git/rr-cache pointing at its original
261 * repository in which the user hasn't performed any
262 * conflict resolution yet?
263 */
264 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
265 strbuf_readlink(&sb, path, st.st_size) ||
266 !is_absolute_path(sb.buf) ||
267 mkdir(sb.buf, 0777)) {
268 strbuf_release(&sb);
269 errno = saved_errno;
270 return -1;
271 }
272 strbuf_release(&sb);
273 }
274 return adjust_shared_perm(path);
275 }
276
277 enum scld_error safe_create_leading_directories(char *path)
278 {
279 char *next_component = path + offset_1st_component(path);
280 enum scld_error ret = SCLD_OK;
281
282 while (ret == SCLD_OK && next_component) {
283 struct stat st;
284 char *slash = next_component, slash_character;
285
286 while (*slash && !is_dir_sep(*slash))
287 slash++;
288
289 if (!*slash)
290 break;
291
292 next_component = slash + 1;
293 while (is_dir_sep(*next_component))
294 next_component++;
295 if (!*next_component)
296 break;
297
298 slash_character = *slash;
299 *slash = '\0';
300 if (!stat(path, &st)) {
301 /* path exists */
302 if (!S_ISDIR(st.st_mode)) {
303 errno = ENOTDIR;
304 ret = SCLD_EXISTS;
305 }
306 } else if (mkdir(path, 0777)) {
307 if (errno == EEXIST &&
308 !stat(path, &st) && S_ISDIR(st.st_mode))
309 ; /* somebody created it since we checked */
310 else if (errno == ENOENT)
311 /*
312 * Either mkdir() failed because
313 * somebody just pruned the containing
314 * directory, or stat() failed because
315 * the file that was in our way was
316 * just removed. Either way, inform
317 * the caller that it might be worth
318 * trying again:
319 */
320 ret = SCLD_VANISHED;
321 else
322 ret = SCLD_FAILED;
323 } else if (adjust_shared_perm(path)) {
324 ret = SCLD_PERMS;
325 }
326 *slash = slash_character;
327 }
328 return ret;
329 }
330
331 enum scld_error safe_create_leading_directories_const(const char *path)
332 {
333 int save_errno;
334 /* path points to cache entries, so xstrdup before messing with it */
335 char *buf = xstrdup(path);
336 enum scld_error result = safe_create_leading_directories(buf);
337
338 save_errno = errno;
339 free(buf);
340 errno = save_errno;
341 return result;
342 }
343
344 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
345 {
346 /*
347 * The number of times we will try to remove empty directories
348 * in the way of path. This is only 1 because if another
349 * process is racily creating directories that conflict with
350 * us, we don't want to fight against them.
351 */
352 int remove_directories_remaining = 1;
353
354 /*
355 * The number of times that we will try to create the
356 * directories containing path. We are willing to attempt this
357 * more than once, because another process could be trying to
358 * clean up empty directories at the same time as we are
359 * trying to create them.
360 */
361 int create_directories_remaining = 3;
362
363 /* A scratch copy of path, filled lazily if we need it: */
364 struct strbuf path_copy = STRBUF_INIT;
365
366 int ret, save_errno;
367
368 /* Sanity check: */
369 assert(*path);
370
371 retry_fn:
372 ret = fn(path, cb);
373 save_errno = errno;
374 if (!ret)
375 goto out;
376
377 if (errno == EISDIR && remove_directories_remaining-- > 0) {
378 /*
379 * A directory is in the way. Maybe it is empty; try
380 * to remove it:
381 */
382 if (!path_copy.len)
383 strbuf_addstr(&path_copy, path);
384
385 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
386 goto retry_fn;
387 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
388 /*
389 * Maybe the containing directory didn't exist, or
390 * maybe it was just deleted by a process that is
391 * racing with us to clean up empty directories. Try
392 * to create it:
393 */
394 enum scld_error scld_result;
395
396 if (!path_copy.len)
397 strbuf_addstr(&path_copy, path);
398
399 do {
400 scld_result = safe_create_leading_directories(path_copy.buf);
401 if (scld_result == SCLD_OK)
402 goto retry_fn;
403 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
404 }
405
406 out:
407 strbuf_release(&path_copy);
408 errno = save_errno;
409 return ret;
410 }
411
412 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
413 {
414 int i;
415 for (i = 0; i < the_hash_algo->rawsz; i++) {
416 static char hex[] = "0123456789abcdef";
417 unsigned int val = oid->hash[i];
418 strbuf_addch(buf, hex[val >> 4]);
419 strbuf_addch(buf, hex[val & 0xf]);
420 if (!i)
421 strbuf_addch(buf, '/');
422 }
423 }
424
425 static const char *odb_loose_path(struct object_directory *odb,
426 struct strbuf *buf,
427 const struct object_id *oid)
428 {
429 strbuf_reset(buf);
430 strbuf_addstr(buf, odb->path);
431 strbuf_addch(buf, '/');
432 fill_loose_path(buf, oid);
433 return buf->buf;
434 }
435
436 const char *loose_object_path(struct repository *r, struct strbuf *buf,
437 const struct object_id *oid)
438 {
439 return odb_loose_path(r->objects->odb, buf, oid);
440 }
441
442 /*
443 * Return non-zero iff the path is usable as an alternate object database.
444 */
445 static int alt_odb_usable(struct raw_object_store *o,
446 struct strbuf *path,
447 const char *normalized_objdir)
448 {
449 struct object_directory *odb;
450
451 /* Detect cases where alternate disappeared */
452 if (!is_directory(path->buf)) {
453 error(_("object directory %s does not exist; "
454 "check .git/objects/info/alternates"),
455 path->buf);
456 return 0;
457 }
458
459 /*
460 * Prevent the common mistake of listing the same
461 * thing twice, or object directory itself.
462 */
463 for (odb = o->odb; odb; odb = odb->next) {
464 if (!fspathcmp(path->buf, odb->path))
465 return 0;
466 }
467 if (!fspathcmp(path->buf, normalized_objdir))
468 return 0;
469
470 return 1;
471 }
472
473 /*
474 * Prepare alternate object database registry.
475 *
476 * The variable alt_odb_list points at the list of struct
477 * object_directory. The elements on this list come from
478 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
479 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
480 * whose contents is similar to that environment variable but can be
481 * LF separated. Its base points at a statically allocated buffer that
482 * contains "/the/directory/corresponding/to/.git/objects/...", while
483 * its name points just after the slash at the end of ".git/objects/"
484 * in the example above, and has enough space to hold 40-byte hex
485 * SHA1, an extra slash for the first level indirection, and the
486 * terminating NUL.
487 */
488 static void read_info_alternates(struct repository *r,
489 const char *relative_base,
490 int depth);
491 static int link_alt_odb_entry(struct repository *r, const char *entry,
492 const char *relative_base, int depth, const char *normalized_objdir)
493 {
494 struct object_directory *ent;
495 struct strbuf pathbuf = STRBUF_INIT;
496
497 if (!is_absolute_path(entry) && relative_base) {
498 strbuf_realpath(&pathbuf, relative_base, 1);
499 strbuf_addch(&pathbuf, '/');
500 }
501 strbuf_addstr(&pathbuf, entry);
502
503 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
504 error(_("unable to normalize alternate object path: %s"),
505 pathbuf.buf);
506 strbuf_release(&pathbuf);
507 return -1;
508 }
509
510 /*
511 * The trailing slash after the directory name is given by
512 * this function at the end. Remove duplicates.
513 */
514 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
515 strbuf_setlen(&pathbuf, pathbuf.len - 1);
516
517 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
518 strbuf_release(&pathbuf);
519 return -1;
520 }
521
522 ent = xcalloc(1, sizeof(*ent));
523 ent->path = xstrdup(pathbuf.buf);
524
525 /* add the alternate entry */
526 *r->objects->odb_tail = ent;
527 r->objects->odb_tail = &(ent->next);
528 ent->next = NULL;
529
530 /* recursively add alternates */
531 read_info_alternates(r, pathbuf.buf, depth + 1);
532
533 strbuf_release(&pathbuf);
534 return 0;
535 }
536
537 static const char *parse_alt_odb_entry(const char *string,
538 int sep,
539 struct strbuf *out)
540 {
541 const char *end;
542
543 strbuf_reset(out);
544
545 if (*string == '#') {
546 /* comment; consume up to next separator */
547 end = strchrnul(string, sep);
548 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
549 /*
550 * quoted path; unquote_c_style has copied the
551 * data for us and set "end". Broken quoting (e.g.,
552 * an entry that doesn't end with a quote) falls
553 * back to the unquoted case below.
554 */
555 } else {
556 /* normal, unquoted path */
557 end = strchrnul(string, sep);
558 strbuf_add(out, string, end - string);
559 }
560
561 if (*end)
562 end++;
563 return end;
564 }
565
566 static void link_alt_odb_entries(struct repository *r, const char *alt,
567 int sep, const char *relative_base, int depth)
568 {
569 struct strbuf objdirbuf = STRBUF_INIT;
570 struct strbuf entry = STRBUF_INIT;
571
572 if (!alt || !*alt)
573 return;
574
575 if (depth > 5) {
576 error(_("%s: ignoring alternate object stores, nesting too deep"),
577 relative_base);
578 return;
579 }
580
581 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
582 if (strbuf_normalize_path(&objdirbuf) < 0)
583 die(_("unable to normalize object directory: %s"),
584 objdirbuf.buf);
585
586 while (*alt) {
587 alt = parse_alt_odb_entry(alt, sep, &entry);
588 if (!entry.len)
589 continue;
590 link_alt_odb_entry(r, entry.buf,
591 relative_base, depth, objdirbuf.buf);
592 }
593 strbuf_release(&entry);
594 strbuf_release(&objdirbuf);
595 }
596
597 static void read_info_alternates(struct repository *r,
598 const char *relative_base,
599 int depth)
600 {
601 char *path;
602 struct strbuf buf = STRBUF_INIT;
603
604 path = xstrfmt("%s/info/alternates", relative_base);
605 if (strbuf_read_file(&buf, path, 1024) < 0) {
606 warn_on_fopen_errors(path);
607 free(path);
608 return;
609 }
610
611 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
612 strbuf_release(&buf);
613 free(path);
614 }
615
616 void add_to_alternates_file(const char *reference)
617 {
618 struct lock_file lock = LOCK_INIT;
619 char *alts = git_pathdup("objects/info/alternates");
620 FILE *in, *out;
621 int found = 0;
622
623 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
624 out = fdopen_lock_file(&lock, "w");
625 if (!out)
626 die_errno(_("unable to fdopen alternates lockfile"));
627
628 in = fopen(alts, "r");
629 if (in) {
630 struct strbuf line = STRBUF_INIT;
631
632 while (strbuf_getline(&line, in) != EOF) {
633 if (!strcmp(reference, line.buf)) {
634 found = 1;
635 break;
636 }
637 fprintf_or_die(out, "%s\n", line.buf);
638 }
639
640 strbuf_release(&line);
641 fclose(in);
642 }
643 else if (errno != ENOENT)
644 die_errno(_("unable to read alternates file"));
645
646 if (found) {
647 rollback_lock_file(&lock);
648 } else {
649 fprintf_or_die(out, "%s\n", reference);
650 if (commit_lock_file(&lock))
651 die_errno(_("unable to move new alternates file into place"));
652 if (the_repository->objects->loaded_alternates)
653 link_alt_odb_entries(the_repository, reference,
654 '\n', NULL, 0);
655 }
656 free(alts);
657 }
658
659 void add_to_alternates_memory(const char *reference)
660 {
661 /*
662 * Make sure alternates are initialized, or else our entry may be
663 * overwritten when they are.
664 */
665 prepare_alt_odb(the_repository);
666
667 link_alt_odb_entries(the_repository, reference,
668 '\n', NULL, 0);
669 }
670
671 /*
672 * Compute the exact path an alternate is at and returns it. In case of
673 * error NULL is returned and the human readable error is added to `err`
674 * `path` may be relative and should point to $GIT_DIR.
675 * `err` must not be null.
676 */
677 char *compute_alternate_path(const char *path, struct strbuf *err)
678 {
679 char *ref_git = NULL;
680 const char *repo, *ref_git_s;
681 int seen_error = 0;
682
683 ref_git_s = real_path_if_valid(path);
684 if (!ref_git_s) {
685 seen_error = 1;
686 strbuf_addf(err, _("path '%s' does not exist"), path);
687 goto out;
688 } else
689 /*
690 * Beware: read_gitfile(), real_path() and mkpath()
691 * return static buffer
692 */
693 ref_git = xstrdup(ref_git_s);
694
695 repo = read_gitfile(ref_git);
696 if (!repo)
697 repo = read_gitfile(mkpath("%s/.git", ref_git));
698 if (repo) {
699 free(ref_git);
700 ref_git = xstrdup(repo);
701 }
702
703 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
704 char *ref_git_git = mkpathdup("%s/.git", ref_git);
705 free(ref_git);
706 ref_git = ref_git_git;
707 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
708 struct strbuf sb = STRBUF_INIT;
709 seen_error = 1;
710 if (get_common_dir(&sb, ref_git)) {
711 strbuf_addf(err,
712 _("reference repository '%s' as a linked "
713 "checkout is not supported yet."),
714 path);
715 goto out;
716 }
717
718 strbuf_addf(err, _("reference repository '%s' is not a "
719 "local repository."), path);
720 goto out;
721 }
722
723 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
724 strbuf_addf(err, _("reference repository '%s' is shallow"),
725 path);
726 seen_error = 1;
727 goto out;
728 }
729
730 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
731 strbuf_addf(err,
732 _("reference repository '%s' is grafted"),
733 path);
734 seen_error = 1;
735 goto out;
736 }
737
738 out:
739 if (seen_error) {
740 FREE_AND_NULL(ref_git);
741 }
742
743 return ref_git;
744 }
745
746 int foreach_alt_odb(alt_odb_fn fn, void *cb)
747 {
748 struct object_directory *ent;
749 int r = 0;
750
751 prepare_alt_odb(the_repository);
752 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
753 r = fn(ent, cb);
754 if (r)
755 break;
756 }
757 return r;
758 }
759
760 void prepare_alt_odb(struct repository *r)
761 {
762 if (r->objects->loaded_alternates)
763 return;
764
765 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
766
767 read_info_alternates(r, r->objects->odb->path, 0);
768 r->objects->loaded_alternates = 1;
769 }
770
771 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
772 static int freshen_file(const char *fn)
773 {
774 struct utimbuf t;
775 t.actime = t.modtime = time(NULL);
776 return !utime(fn, &t);
777 }
778
779 /*
780 * All of the check_and_freshen functions return 1 if the file exists and was
781 * freshened (if freshening was requested), 0 otherwise. If they return
782 * 0, you should not assume that it is safe to skip a write of the object (it
783 * either does not exist on disk, or has a stale mtime and may be subject to
784 * pruning).
785 */
786 int check_and_freshen_file(const char *fn, int freshen)
787 {
788 if (access(fn, F_OK))
789 return 0;
790 if (freshen && !freshen_file(fn))
791 return 0;
792 return 1;
793 }
794
795 static int check_and_freshen_odb(struct object_directory *odb,
796 const struct object_id *oid,
797 int freshen)
798 {
799 static struct strbuf path = STRBUF_INIT;
800 odb_loose_path(odb, &path, oid);
801 return check_and_freshen_file(path.buf, freshen);
802 }
803
804 static int check_and_freshen_local(const struct object_id *oid, int freshen)
805 {
806 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
807 }
808
809 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
810 {
811 struct object_directory *odb;
812
813 prepare_alt_odb(the_repository);
814 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
815 if (check_and_freshen_odb(odb, oid, freshen))
816 return 1;
817 }
818 return 0;
819 }
820
821 static int check_and_freshen(const struct object_id *oid, int freshen)
822 {
823 return check_and_freshen_local(oid, freshen) ||
824 check_and_freshen_nonlocal(oid, freshen);
825 }
826
827 int has_loose_object_nonlocal(const struct object_id *oid)
828 {
829 return check_and_freshen_nonlocal(oid, 0);
830 }
831
832 static int has_loose_object(const struct object_id *oid)
833 {
834 return check_and_freshen(oid, 0);
835 }
836
837 static void mmap_limit_check(size_t length)
838 {
839 static size_t limit = 0;
840 if (!limit) {
841 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
842 if (!limit)
843 limit = SIZE_MAX;
844 }
845 if (length > limit)
846 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
847 (uintmax_t)length, (uintmax_t)limit);
848 }
849
850 void *xmmap_gently(void *start, size_t length,
851 int prot, int flags, int fd, off_t offset)
852 {
853 void *ret;
854
855 mmap_limit_check(length);
856 ret = mmap(start, length, prot, flags, fd, offset);
857 if (ret == MAP_FAILED) {
858 if (!length)
859 return NULL;
860 release_pack_memory(length);
861 ret = mmap(start, length, prot, flags, fd, offset);
862 }
863 return ret;
864 }
865
866 void *xmmap(void *start, size_t length,
867 int prot, int flags, int fd, off_t offset)
868 {
869 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
870 if (ret == MAP_FAILED)
871 die_errno(_("mmap failed"));
872 return ret;
873 }
874
875 /*
876 * With an in-core object data in "map", rehash it to make sure the
877 * object name actually matches "oid" to detect object corruption.
878 * With "map" == NULL, try reading the object named with "oid" using
879 * the streaming interface and rehash it to do the same.
880 */
881 int check_object_signature(const struct object_id *oid, void *map,
882 unsigned long size, const char *type)
883 {
884 struct object_id real_oid;
885 enum object_type obj_type;
886 struct git_istream *st;
887 git_hash_ctx c;
888 char hdr[MAX_HEADER_LEN];
889 int hdrlen;
890
891 if (map) {
892 hash_object_file(map, size, type, &real_oid);
893 return !oideq(oid, &real_oid) ? -1 : 0;
894 }
895
896 st = open_istream(oid, &obj_type, &size, NULL);
897 if (!st)
898 return -1;
899
900 /* Generate the header */
901 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
902
903 /* Sha1.. */
904 the_hash_algo->init_fn(&c);
905 the_hash_algo->update_fn(&c, hdr, hdrlen);
906 for (;;) {
907 char buf[1024 * 16];
908 ssize_t readlen = read_istream(st, buf, sizeof(buf));
909
910 if (readlen < 0) {
911 close_istream(st);
912 return -1;
913 }
914 if (!readlen)
915 break;
916 the_hash_algo->update_fn(&c, buf, readlen);
917 }
918 the_hash_algo->final_fn(real_oid.hash, &c);
919 close_istream(st);
920 return !oideq(oid, &real_oid) ? -1 : 0;
921 }
922
923 int git_open_cloexec(const char *name, int flags)
924 {
925 int fd;
926 static int o_cloexec = O_CLOEXEC;
927
928 fd = open(name, flags | o_cloexec);
929 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
930 /* Try again w/o O_CLOEXEC: the kernel might not support it */
931 o_cloexec &= ~O_CLOEXEC;
932 fd = open(name, flags | o_cloexec);
933 }
934
935 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
936 {
937 static int fd_cloexec = FD_CLOEXEC;
938
939 if (!o_cloexec && 0 <= fd && fd_cloexec) {
940 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
941 int flags = fcntl(fd, F_GETFD);
942 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
943 fd_cloexec = 0;
944 }
945 }
946 #endif
947 return fd;
948 }
949
950 /*
951 * Find "oid" as a loose object in the local repository or in an alternate.
952 * Returns 0 on success, negative on failure.
953 *
954 * The "path" out-parameter will give the path of the object we found (if any).
955 * Note that it may point to static storage and is only valid until another
956 * call to stat_loose_object().
957 */
958 static int stat_loose_object(struct repository *r, const struct object_id *oid,
959 struct stat *st, const char **path)
960 {
961 struct object_directory *odb;
962 static struct strbuf buf = STRBUF_INIT;
963
964 prepare_alt_odb(r);
965 for (odb = r->objects->odb; odb; odb = odb->next) {
966 *path = odb_loose_path(odb, &buf, oid);
967 if (!lstat(*path, st))
968 return 0;
969 }
970
971 return -1;
972 }
973
974 /*
975 * Like stat_loose_object(), but actually open the object and return the
976 * descriptor. See the caveats on the "path" parameter above.
977 */
978 static int open_loose_object(struct repository *r,
979 const struct object_id *oid, const char **path)
980 {
981 int fd;
982 struct object_directory *odb;
983 int most_interesting_errno = ENOENT;
984 static struct strbuf buf = STRBUF_INIT;
985
986 prepare_alt_odb(r);
987 for (odb = r->objects->odb; odb; odb = odb->next) {
988 *path = odb_loose_path(odb, &buf, oid);
989 fd = git_open(*path);
990 if (fd >= 0)
991 return fd;
992
993 if (most_interesting_errno == ENOENT)
994 most_interesting_errno = errno;
995 }
996 errno = most_interesting_errno;
997 return -1;
998 }
999
1000 static int quick_has_loose(struct repository *r,
1001 const struct object_id *oid)
1002 {
1003 struct object_directory *odb;
1004
1005 prepare_alt_odb(r);
1006 for (odb = r->objects->odb; odb; odb = odb->next) {
1007 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1008 return 1;
1009 }
1010 return 0;
1011 }
1012
1013 /*
1014 * Map the loose object at "path" if it is not NULL, or the path found by
1015 * searching for a loose object named "oid".
1016 */
1017 static void *map_loose_object_1(struct repository *r, const char *path,
1018 const struct object_id *oid, unsigned long *size)
1019 {
1020 void *map;
1021 int fd;
1022
1023 if (path)
1024 fd = git_open(path);
1025 else
1026 fd = open_loose_object(r, oid, &path);
1027 map = NULL;
1028 if (fd >= 0) {
1029 struct stat st;
1030
1031 if (!fstat(fd, &st)) {
1032 *size = xsize_t(st.st_size);
1033 if (!*size) {
1034 /* mmap() is forbidden on empty files */
1035 error(_("object file %s is empty"), path);
1036 close(fd);
1037 return NULL;
1038 }
1039 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1040 }
1041 close(fd);
1042 }
1043 return map;
1044 }
1045
1046 void *map_loose_object(struct repository *r,
1047 const struct object_id *oid,
1048 unsigned long *size)
1049 {
1050 return map_loose_object_1(r, NULL, oid, size);
1051 }
1052
1053 static int unpack_loose_short_header(git_zstream *stream,
1054 unsigned char *map, unsigned long mapsize,
1055 void *buffer, unsigned long bufsiz)
1056 {
1057 /* Get the data stream */
1058 memset(stream, 0, sizeof(*stream));
1059 stream->next_in = map;
1060 stream->avail_in = mapsize;
1061 stream->next_out = buffer;
1062 stream->avail_out = bufsiz;
1063
1064 git_inflate_init(stream);
1065 return git_inflate(stream, 0);
1066 }
1067
1068 int unpack_loose_header(git_zstream *stream,
1069 unsigned char *map, unsigned long mapsize,
1070 void *buffer, unsigned long bufsiz)
1071 {
1072 int status = unpack_loose_short_header(stream, map, mapsize,
1073 buffer, bufsiz);
1074
1075 if (status < Z_OK)
1076 return status;
1077
1078 /* Make sure we have the terminating NUL */
1079 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1080 return -1;
1081 return 0;
1082 }
1083
1084 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1085 unsigned long mapsize, void *buffer,
1086 unsigned long bufsiz, struct strbuf *header)
1087 {
1088 int status;
1089
1090 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1091 if (status < Z_OK)
1092 return -1;
1093
1094 /*
1095 * Check if entire header is unpacked in the first iteration.
1096 */
1097 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1098 return 0;
1099
1100 /*
1101 * buffer[0..bufsiz] was not large enough. Copy the partial
1102 * result out to header, and then append the result of further
1103 * reading the stream.
1104 */
1105 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1106 stream->next_out = buffer;
1107 stream->avail_out = bufsiz;
1108
1109 do {
1110 status = git_inflate(stream, 0);
1111 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1112 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1113 return 0;
1114 stream->next_out = buffer;
1115 stream->avail_out = bufsiz;
1116 } while (status != Z_STREAM_END);
1117 return -1;
1118 }
1119
1120 static void *unpack_loose_rest(git_zstream *stream,
1121 void *buffer, unsigned long size,
1122 const struct object_id *oid)
1123 {
1124 int bytes = strlen(buffer) + 1;
1125 unsigned char *buf = xmallocz(size);
1126 unsigned long n;
1127 int status = Z_OK;
1128
1129 n = stream->total_out - bytes;
1130 if (n > size)
1131 n = size;
1132 memcpy(buf, (char *) buffer + bytes, n);
1133 bytes = n;
1134 if (bytes <= size) {
1135 /*
1136 * The above condition must be (bytes <= size), not
1137 * (bytes < size). In other words, even though we
1138 * expect no more output and set avail_out to zero,
1139 * the input zlib stream may have bytes that express
1140 * "this concludes the stream", and we *do* want to
1141 * eat that input.
1142 *
1143 * Otherwise we would not be able to test that we
1144 * consumed all the input to reach the expected size;
1145 * we also want to check that zlib tells us that all
1146 * went well with status == Z_STREAM_END at the end.
1147 */
1148 stream->next_out = buf + bytes;
1149 stream->avail_out = size - bytes;
1150 while (status == Z_OK)
1151 status = git_inflate(stream, Z_FINISH);
1152 }
1153 if (status == Z_STREAM_END && !stream->avail_in) {
1154 git_inflate_end(stream);
1155 return buf;
1156 }
1157
1158 if (status < 0)
1159 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1160 else if (stream->avail_in)
1161 error(_("garbage at end of loose object '%s'"),
1162 oid_to_hex(oid));
1163 free(buf);
1164 return NULL;
1165 }
1166
1167 /*
1168 * We used to just use "sscanf()", but that's actually way
1169 * too permissive for what we want to check. So do an anal
1170 * object header parse by hand.
1171 */
1172 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1173 unsigned int flags)
1174 {
1175 const char *type_buf = hdr;
1176 unsigned long size;
1177 int type, type_len = 0;
1178
1179 /*
1180 * The type can be of any size but is followed by
1181 * a space.
1182 */
1183 for (;;) {
1184 char c = *hdr++;
1185 if (!c)
1186 return -1;
1187 if (c == ' ')
1188 break;
1189 type_len++;
1190 }
1191
1192 type = type_from_string_gently(type_buf, type_len, 1);
1193 if (oi->type_name)
1194 strbuf_add(oi->type_name, type_buf, type_len);
1195 /*
1196 * Set type to 0 if its an unknown object and
1197 * we're obtaining the type using '--allow-unknown-type'
1198 * option.
1199 */
1200 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1201 type = 0;
1202 else if (type < 0)
1203 die(_("invalid object type"));
1204 if (oi->typep)
1205 *oi->typep = type;
1206
1207 /*
1208 * The length must follow immediately, and be in canonical
1209 * decimal format (ie "010" is not valid).
1210 */
1211 size = *hdr++ - '0';
1212 if (size > 9)
1213 return -1;
1214 if (size) {
1215 for (;;) {
1216 unsigned long c = *hdr - '0';
1217 if (c > 9)
1218 break;
1219 hdr++;
1220 size = size * 10 + c;
1221 }
1222 }
1223
1224 if (oi->sizep)
1225 *oi->sizep = size;
1226
1227 /*
1228 * The length must be followed by a zero byte
1229 */
1230 return *hdr ? -1 : type;
1231 }
1232
1233 int parse_loose_header(const char *hdr, unsigned long *sizep)
1234 {
1235 struct object_info oi = OBJECT_INFO_INIT;
1236
1237 oi.sizep = sizep;
1238 return parse_loose_header_extended(hdr, &oi, 0);
1239 }
1240
1241 static int loose_object_info(struct repository *r,
1242 const struct object_id *oid,
1243 struct object_info *oi, int flags)
1244 {
1245 int status = 0;
1246 unsigned long mapsize;
1247 void *map;
1248 git_zstream stream;
1249 char hdr[MAX_HEADER_LEN];
1250 struct strbuf hdrbuf = STRBUF_INIT;
1251 unsigned long size_scratch;
1252
1253 if (oi->delta_base_sha1)
1254 hashclr(oi->delta_base_sha1);
1255
1256 /*
1257 * If we don't care about type or size, then we don't
1258 * need to look inside the object at all. Note that we
1259 * do not optimize out the stat call, even if the
1260 * caller doesn't care about the disk-size, since our
1261 * return value implicitly indicates whether the
1262 * object even exists.
1263 */
1264 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1265 const char *path;
1266 struct stat st;
1267 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1268 return quick_has_loose(r, oid) ? 0 : -1;
1269 if (stat_loose_object(r, oid, &st, &path) < 0)
1270 return -1;
1271 if (oi->disk_sizep)
1272 *oi->disk_sizep = st.st_size;
1273 return 0;
1274 }
1275
1276 map = map_loose_object(r, oid, &mapsize);
1277 if (!map)
1278 return -1;
1279
1280 if (!oi->sizep)
1281 oi->sizep = &size_scratch;
1282
1283 if (oi->disk_sizep)
1284 *oi->disk_sizep = mapsize;
1285 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1286 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1287 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1288 oid_to_hex(oid));
1289 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1290 status = error(_("unable to unpack %s header"),
1291 oid_to_hex(oid));
1292 if (status < 0)
1293 ; /* Do nothing */
1294 else if (hdrbuf.len) {
1295 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1296 status = error(_("unable to parse %s header with --allow-unknown-type"),
1297 oid_to_hex(oid));
1298 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1299 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1300
1301 if (status >= 0 && oi->contentp) {
1302 *oi->contentp = unpack_loose_rest(&stream, hdr,
1303 *oi->sizep, oid);
1304 if (!*oi->contentp) {
1305 git_inflate_end(&stream);
1306 status = -1;
1307 }
1308 } else
1309 git_inflate_end(&stream);
1310
1311 munmap(map, mapsize);
1312 if (status && oi->typep)
1313 *oi->typep = status;
1314 if (oi->sizep == &size_scratch)
1315 oi->sizep = NULL;
1316 strbuf_release(&hdrbuf);
1317 oi->whence = OI_LOOSE;
1318 return (status < 0) ? status : 0;
1319 }
1320
1321 int fetch_if_missing = 1;
1322
1323 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1324 struct object_info *oi, unsigned flags)
1325 {
1326 static struct object_info blank_oi = OBJECT_INFO_INIT;
1327 struct pack_entry e;
1328 int rtype;
1329 const struct object_id *real = oid;
1330 int already_retried = 0;
1331
1332 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1333 real = lookup_replace_object(r, oid);
1334
1335 if (is_null_oid(real))
1336 return -1;
1337
1338 if (!oi)
1339 oi = &blank_oi;
1340
1341 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1342 struct cached_object *co = find_cached_object(real);
1343 if (co) {
1344 if (oi->typep)
1345 *(oi->typep) = co->type;
1346 if (oi->sizep)
1347 *(oi->sizep) = co->size;
1348 if (oi->disk_sizep)
1349 *(oi->disk_sizep) = 0;
1350 if (oi->delta_base_sha1)
1351 hashclr(oi->delta_base_sha1);
1352 if (oi->type_name)
1353 strbuf_addstr(oi->type_name, type_name(co->type));
1354 if (oi->contentp)
1355 *oi->contentp = xmemdupz(co->buf, co->size);
1356 oi->whence = OI_CACHED;
1357 return 0;
1358 }
1359 }
1360
1361 while (1) {
1362 if (find_pack_entry(r, real, &e))
1363 break;
1364
1365 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1366 return -1;
1367
1368 /* Most likely it's a loose object. */
1369 if (!loose_object_info(r, real, oi, flags))
1370 return 0;
1371
1372 /* Not a loose object; someone else may have just packed it. */
1373 if (!(flags & OBJECT_INFO_QUICK)) {
1374 reprepare_packed_git(r);
1375 if (find_pack_entry(r, real, &e))
1376 break;
1377 }
1378
1379 /* Check if it is a missing object */
1380 if (fetch_if_missing && has_promisor_remote() &&
1381 !already_retried && r == the_repository &&
1382 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1383 /*
1384 * TODO Investigate checking promisor_remote_get_direct()
1385 * TODO return value and stopping on error here.
1386 * TODO Pass a repository struct through
1387 * promisor_remote_get_direct(), such that arbitrary
1388 * repositories work.
1389 */
1390 promisor_remote_get_direct(r, real, 1);
1391 already_retried = 1;
1392 continue;
1393 }
1394
1395 return -1;
1396 }
1397
1398 if (oi == &blank_oi)
1399 /*
1400 * We know that the caller doesn't actually need the
1401 * information below, so return early.
1402 */
1403 return 0;
1404 rtype = packed_object_info(r, e.p, e.offset, oi);
1405 if (rtype < 0) {
1406 mark_bad_packed_object(e.p, real->hash);
1407 return oid_object_info_extended(r, real, oi, 0);
1408 } else if (oi->whence == OI_PACKED) {
1409 oi->u.packed.offset = e.offset;
1410 oi->u.packed.pack = e.p;
1411 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1412 rtype == OBJ_OFS_DELTA);
1413 }
1414
1415 return 0;
1416 }
1417
1418 /* returns enum object_type or negative */
1419 int oid_object_info(struct repository *r,
1420 const struct object_id *oid,
1421 unsigned long *sizep)
1422 {
1423 enum object_type type;
1424 struct object_info oi = OBJECT_INFO_INIT;
1425
1426 oi.typep = &type;
1427 oi.sizep = sizep;
1428 if (oid_object_info_extended(r, oid, &oi,
1429 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1430 return -1;
1431 return type;
1432 }
1433
1434 static void *read_object(struct repository *r,
1435 const struct object_id *oid, enum object_type *type,
1436 unsigned long *size)
1437 {
1438 struct object_info oi = OBJECT_INFO_INIT;
1439 void *content;
1440 oi.typep = type;
1441 oi.sizep = size;
1442 oi.contentp = &content;
1443
1444 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1445 return NULL;
1446 return content;
1447 }
1448
1449 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1450 struct object_id *oid)
1451 {
1452 struct cached_object *co;
1453
1454 hash_object_file(buf, len, type_name(type), oid);
1455 if (has_object_file(oid) || find_cached_object(oid))
1456 return 0;
1457 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1458 co = &cached_objects[cached_object_nr++];
1459 co->size = len;
1460 co->type = type;
1461 co->buf = xmalloc(len);
1462 memcpy(co->buf, buf, len);
1463 oidcpy(&co->oid, oid);
1464 return 0;
1465 }
1466
1467 /*
1468 * This function dies on corrupt objects; the callers who want to
1469 * deal with them should arrange to call read_object() and give error
1470 * messages themselves.
1471 */
1472 void *read_object_file_extended(struct repository *r,
1473 const struct object_id *oid,
1474 enum object_type *type,
1475 unsigned long *size,
1476 int lookup_replace)
1477 {
1478 void *data;
1479 const struct packed_git *p;
1480 const char *path;
1481 struct stat st;
1482 const struct object_id *repl = lookup_replace ?
1483 lookup_replace_object(r, oid) : oid;
1484
1485 errno = 0;
1486 data = read_object(r, repl, type, size);
1487 if (data)
1488 return data;
1489
1490 if (errno && errno != ENOENT)
1491 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1492
1493 /* die if we replaced an object with one that does not exist */
1494 if (repl != oid)
1495 die(_("replacement %s not found for %s"),
1496 oid_to_hex(repl), oid_to_hex(oid));
1497
1498 if (!stat_loose_object(r, repl, &st, &path))
1499 die(_("loose object %s (stored in %s) is corrupt"),
1500 oid_to_hex(repl), path);
1501
1502 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1503 die(_("packed object %s (stored in %s) is corrupt"),
1504 oid_to_hex(repl), p->pack_name);
1505
1506 return NULL;
1507 }
1508
1509 void *read_object_with_reference(const struct object_id *oid,
1510 const char *required_type_name,
1511 unsigned long *size,
1512 struct object_id *actual_oid_return)
1513 {
1514 enum object_type type, required_type;
1515 void *buffer;
1516 unsigned long isize;
1517 struct object_id actual_oid;
1518
1519 required_type = type_from_string(required_type_name);
1520 oidcpy(&actual_oid, oid);
1521 while (1) {
1522 int ref_length = -1;
1523 const char *ref_type = NULL;
1524
1525 buffer = read_object_file(&actual_oid, &type, &isize);
1526 if (!buffer)
1527 return NULL;
1528 if (type == required_type) {
1529 *size = isize;
1530 if (actual_oid_return)
1531 oidcpy(actual_oid_return, &actual_oid);
1532 return buffer;
1533 }
1534 /* Handle references */
1535 else if (type == OBJ_COMMIT)
1536 ref_type = "tree ";
1537 else if (type == OBJ_TAG)
1538 ref_type = "object ";
1539 else {
1540 free(buffer);
1541 return NULL;
1542 }
1543 ref_length = strlen(ref_type);
1544
1545 if (ref_length + the_hash_algo->hexsz > isize ||
1546 memcmp(buffer, ref_type, ref_length) ||
1547 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1548 free(buffer);
1549 return NULL;
1550 }
1551 free(buffer);
1552 /* Now we have the ID of the referred-to object in
1553 * actual_oid. Check again. */
1554 }
1555 }
1556
1557 static void write_object_file_prepare(const void *buf, unsigned long len,
1558 const char *type, struct object_id *oid,
1559 char *hdr, int *hdrlen)
1560 {
1561 git_hash_ctx c;
1562
1563 /* Generate the header */
1564 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1565
1566 /* Sha1.. */
1567 the_hash_algo->init_fn(&c);
1568 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1569 the_hash_algo->update_fn(&c, buf, len);
1570 the_hash_algo->final_fn(oid->hash, &c);
1571 }
1572
1573 /*
1574 * Move the just written object into its final resting place.
1575 */
1576 int finalize_object_file(const char *tmpfile, const char *filename)
1577 {
1578 int ret = 0;
1579
1580 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1581 goto try_rename;
1582 else if (link(tmpfile, filename))
1583 ret = errno;
1584
1585 /*
1586 * Coda hack - coda doesn't like cross-directory links,
1587 * so we fall back to a rename, which will mean that it
1588 * won't be able to check collisions, but that's not a
1589 * big deal.
1590 *
1591 * The same holds for FAT formatted media.
1592 *
1593 * When this succeeds, we just return. We have nothing
1594 * left to unlink.
1595 */
1596 if (ret && ret != EEXIST) {
1597 try_rename:
1598 if (!rename(tmpfile, filename))
1599 goto out;
1600 ret = errno;
1601 }
1602 unlink_or_warn(tmpfile);
1603 if (ret) {
1604 if (ret != EEXIST) {
1605 return error_errno(_("unable to write file %s"), filename);
1606 }
1607 /* FIXME!!! Collision check here ? */
1608 }
1609
1610 out:
1611 if (adjust_shared_perm(filename))
1612 return error(_("unable to set permission to '%s'"), filename);
1613 return 0;
1614 }
1615
1616 static int write_buffer(int fd, const void *buf, size_t len)
1617 {
1618 if (write_in_full(fd, buf, len) < 0)
1619 return error_errno(_("file write error"));
1620 return 0;
1621 }
1622
1623 int hash_object_file(const void *buf, unsigned long len, const char *type,
1624 struct object_id *oid)
1625 {
1626 char hdr[MAX_HEADER_LEN];
1627 int hdrlen = sizeof(hdr);
1628 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1629 return 0;
1630 }
1631
1632 /* Finalize a file on disk, and close it. */
1633 static void close_loose_object(int fd)
1634 {
1635 if (fsync_object_files)
1636 fsync_or_die(fd, "loose object file");
1637 if (close(fd) != 0)
1638 die_errno(_("error when closing loose object file"));
1639 }
1640
1641 /* Size of directory component, including the ending '/' */
1642 static inline int directory_size(const char *filename)
1643 {
1644 const char *s = strrchr(filename, '/');
1645 if (!s)
1646 return 0;
1647 return s - filename + 1;
1648 }
1649
1650 /*
1651 * This creates a temporary file in the same directory as the final
1652 * 'filename'
1653 *
1654 * We want to avoid cross-directory filename renames, because those
1655 * can have problems on various filesystems (FAT, NFS, Coda).
1656 */
1657 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1658 {
1659 int fd, dirlen = directory_size(filename);
1660
1661 strbuf_reset(tmp);
1662 strbuf_add(tmp, filename, dirlen);
1663 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1664 fd = git_mkstemp_mode(tmp->buf, 0444);
1665 if (fd < 0 && dirlen && errno == ENOENT) {
1666 /*
1667 * Make sure the directory exists; note that the contents
1668 * of the buffer are undefined after mkstemp returns an
1669 * error, so we have to rewrite the whole buffer from
1670 * scratch.
1671 */
1672 strbuf_reset(tmp);
1673 strbuf_add(tmp, filename, dirlen - 1);
1674 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1675 return -1;
1676 if (adjust_shared_perm(tmp->buf))
1677 return -1;
1678
1679 /* Try again */
1680 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1681 fd = git_mkstemp_mode(tmp->buf, 0444);
1682 }
1683 return fd;
1684 }
1685
1686 static int write_loose_object(const struct object_id *oid, char *hdr,
1687 int hdrlen, const void *buf, unsigned long len,
1688 time_t mtime)
1689 {
1690 int fd, ret;
1691 unsigned char compressed[4096];
1692 git_zstream stream;
1693 git_hash_ctx c;
1694 struct object_id parano_oid;
1695 static struct strbuf tmp_file = STRBUF_INIT;
1696 static struct strbuf filename = STRBUF_INIT;
1697
1698 loose_object_path(the_repository, &filename, oid);
1699
1700 fd = create_tmpfile(&tmp_file, filename.buf);
1701 if (fd < 0) {
1702 if (errno == EACCES)
1703 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1704 else
1705 return error_errno(_("unable to create temporary file"));
1706 }
1707
1708 /* Set it up */
1709 git_deflate_init(&stream, zlib_compression_level);
1710 stream.next_out = compressed;
1711 stream.avail_out = sizeof(compressed);
1712 the_hash_algo->init_fn(&c);
1713
1714 /* First header.. */
1715 stream.next_in = (unsigned char *)hdr;
1716 stream.avail_in = hdrlen;
1717 while (git_deflate(&stream, 0) == Z_OK)
1718 ; /* nothing */
1719 the_hash_algo->update_fn(&c, hdr, hdrlen);
1720
1721 /* Then the data itself.. */
1722 stream.next_in = (void *)buf;
1723 stream.avail_in = len;
1724 do {
1725 unsigned char *in0 = stream.next_in;
1726 ret = git_deflate(&stream, Z_FINISH);
1727 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1728 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1729 die(_("unable to write loose object file"));
1730 stream.next_out = compressed;
1731 stream.avail_out = sizeof(compressed);
1732 } while (ret == Z_OK);
1733
1734 if (ret != Z_STREAM_END)
1735 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1736 ret);
1737 ret = git_deflate_end_gently(&stream);
1738 if (ret != Z_OK)
1739 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1740 ret);
1741 the_hash_algo->final_fn(parano_oid.hash, &c);
1742 if (!oideq(oid, &parano_oid))
1743 die(_("confused by unstable object source data for %s"),
1744 oid_to_hex(oid));
1745
1746 close_loose_object(fd);
1747
1748 if (mtime) {
1749 struct utimbuf utb;
1750 utb.actime = mtime;
1751 utb.modtime = mtime;
1752 if (utime(tmp_file.buf, &utb) < 0)
1753 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1754 }
1755
1756 return finalize_object_file(tmp_file.buf, filename.buf);
1757 }
1758
1759 static int freshen_loose_object(const struct object_id *oid)
1760 {
1761 return check_and_freshen(oid, 1);
1762 }
1763
1764 static int freshen_packed_object(const struct object_id *oid)
1765 {
1766 struct pack_entry e;
1767 if (!find_pack_entry(the_repository, oid, &e))
1768 return 0;
1769 if (e.p->freshened)
1770 return 1;
1771 if (!freshen_file(e.p->pack_name))
1772 return 0;
1773 e.p->freshened = 1;
1774 return 1;
1775 }
1776
1777 int write_object_file(const void *buf, unsigned long len, const char *type,
1778 struct object_id *oid)
1779 {
1780 char hdr[MAX_HEADER_LEN];
1781 int hdrlen = sizeof(hdr);
1782
1783 /* Normally if we have it in the pack then we do not bother writing
1784 * it out into .git/objects/??/?{38} file.
1785 */
1786 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1787 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1788 return 0;
1789 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1790 }
1791
1792 int hash_object_file_literally(const void *buf, unsigned long len,
1793 const char *type, struct object_id *oid,
1794 unsigned flags)
1795 {
1796 char *header;
1797 int hdrlen, status = 0;
1798
1799 /* type string, SP, %lu of the length plus NUL must fit this */
1800 hdrlen = strlen(type) + MAX_HEADER_LEN;
1801 header = xmalloc(hdrlen);
1802 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1803
1804 if (!(flags & HASH_WRITE_OBJECT))
1805 goto cleanup;
1806 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1807 goto cleanup;
1808 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1809
1810 cleanup:
1811 free(header);
1812 return status;
1813 }
1814
1815 int force_object_loose(const struct object_id *oid, time_t mtime)
1816 {
1817 void *buf;
1818 unsigned long len;
1819 enum object_type type;
1820 char hdr[MAX_HEADER_LEN];
1821 int hdrlen;
1822 int ret;
1823
1824 if (has_loose_object(oid))
1825 return 0;
1826 buf = read_object(the_repository, oid, &type, &len);
1827 if (!buf)
1828 return error(_("cannot read object for %s"), oid_to_hex(oid));
1829 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1830 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1831 free(buf);
1832
1833 return ret;
1834 }
1835
1836 int repo_has_object_file_with_flags(struct repository *r,
1837 const struct object_id *oid, int flags)
1838 {
1839 if (!startup_info->have_repository)
1840 return 0;
1841 return oid_object_info_extended(r, oid, NULL,
1842 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1843 }
1844
1845 int repo_has_object_file(struct repository *r,
1846 const struct object_id *oid)
1847 {
1848 return repo_has_object_file_with_flags(r, oid, 0);
1849 }
1850
1851 static void check_tree(const void *buf, size_t size)
1852 {
1853 struct tree_desc desc;
1854 struct name_entry entry;
1855
1856 init_tree_desc(&desc, buf, size);
1857 while (tree_entry(&desc, &entry))
1858 /* do nothing
1859 * tree_entry() will die() on malformed entries */
1860 ;
1861 }
1862
1863 static void check_commit(const void *buf, size_t size)
1864 {
1865 struct commit c;
1866 memset(&c, 0, sizeof(c));
1867 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1868 die(_("corrupt commit"));
1869 }
1870
1871 static void check_tag(const void *buf, size_t size)
1872 {
1873 struct tag t;
1874 memset(&t, 0, sizeof(t));
1875 if (parse_tag_buffer(the_repository, &t, buf, size))
1876 die(_("corrupt tag"));
1877 }
1878
1879 static int index_mem(struct index_state *istate,
1880 struct object_id *oid, void *buf, size_t size,
1881 enum object_type type,
1882 const char *path, unsigned flags)
1883 {
1884 int ret, re_allocated = 0;
1885 int write_object = flags & HASH_WRITE_OBJECT;
1886
1887 if (!type)
1888 type = OBJ_BLOB;
1889
1890 /*
1891 * Convert blobs to git internal format
1892 */
1893 if ((type == OBJ_BLOB) && path) {
1894 struct strbuf nbuf = STRBUF_INIT;
1895 if (convert_to_git(istate, path, buf, size, &nbuf,
1896 get_conv_flags(flags))) {
1897 buf = strbuf_detach(&nbuf, &size);
1898 re_allocated = 1;
1899 }
1900 }
1901 if (flags & HASH_FORMAT_CHECK) {
1902 if (type == OBJ_TREE)
1903 check_tree(buf, size);
1904 if (type == OBJ_COMMIT)
1905 check_commit(buf, size);
1906 if (type == OBJ_TAG)
1907 check_tag(buf, size);
1908 }
1909
1910 if (write_object)
1911 ret = write_object_file(buf, size, type_name(type), oid);
1912 else
1913 ret = hash_object_file(buf, size, type_name(type), oid);
1914 if (re_allocated)
1915 free(buf);
1916 return ret;
1917 }
1918
1919 static int index_stream_convert_blob(struct index_state *istate,
1920 struct object_id *oid,
1921 int fd,
1922 const char *path,
1923 unsigned flags)
1924 {
1925 int ret;
1926 const int write_object = flags & HASH_WRITE_OBJECT;
1927 struct strbuf sbuf = STRBUF_INIT;
1928
1929 assert(path);
1930 assert(would_convert_to_git_filter_fd(istate, path));
1931
1932 convert_to_git_filter_fd(istate, path, fd, &sbuf,
1933 get_conv_flags(flags));
1934
1935 if (write_object)
1936 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1937 oid);
1938 else
1939 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1940 oid);
1941 strbuf_release(&sbuf);
1942 return ret;
1943 }
1944
1945 static int index_pipe(struct index_state *istate, struct object_id *oid,
1946 int fd, enum object_type type,
1947 const char *path, unsigned flags)
1948 {
1949 struct strbuf sbuf = STRBUF_INIT;
1950 int ret;
1951
1952 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1953 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
1954 else
1955 ret = -1;
1956 strbuf_release(&sbuf);
1957 return ret;
1958 }
1959
1960 #define SMALL_FILE_SIZE (32*1024)
1961
1962 static int index_core(struct index_state *istate,
1963 struct object_id *oid, int fd, size_t size,
1964 enum object_type type, const char *path,
1965 unsigned flags)
1966 {
1967 int ret;
1968
1969 if (!size) {
1970 ret = index_mem(istate, oid, "", size, type, path, flags);
1971 } else if (size <= SMALL_FILE_SIZE) {
1972 char *buf = xmalloc(size);
1973 ssize_t read_result = read_in_full(fd, buf, size);
1974 if (read_result < 0)
1975 ret = error_errno(_("read error while indexing %s"),
1976 path ? path : "<unknown>");
1977 else if (read_result != size)
1978 ret = error(_("short read while indexing %s"),
1979 path ? path : "<unknown>");
1980 else
1981 ret = index_mem(istate, oid, buf, size, type, path, flags);
1982 free(buf);
1983 } else {
1984 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1985 ret = index_mem(istate, oid, buf, size, type, path, flags);
1986 munmap(buf, size);
1987 }
1988 return ret;
1989 }
1990
1991 /*
1992 * This creates one packfile per large blob unless bulk-checkin
1993 * machinery is "plugged".
1994 *
1995 * This also bypasses the usual "convert-to-git" dance, and that is on
1996 * purpose. We could write a streaming version of the converting
1997 * functions and insert that before feeding the data to fast-import
1998 * (or equivalent in-core API described above). However, that is
1999 * somewhat complicated, as we do not know the size of the filter
2000 * result, which we need to know beforehand when writing a git object.
2001 * Since the primary motivation for trying to stream from the working
2002 * tree file and to avoid mmaping it in core is to deal with large
2003 * binary blobs, they generally do not want to get any conversion, and
2004 * callers should avoid this code path when filters are requested.
2005 */
2006 static int index_stream(struct object_id *oid, int fd, size_t size,
2007 enum object_type type, const char *path,
2008 unsigned flags)
2009 {
2010 return index_bulk_checkin(oid, fd, size, type, path, flags);
2011 }
2012
2013 int index_fd(struct index_state *istate, struct object_id *oid,
2014 int fd, struct stat *st,
2015 enum object_type type, const char *path, unsigned flags)
2016 {
2017 int ret;
2018
2019 /*
2020 * Call xsize_t() only when needed to avoid potentially unnecessary
2021 * die() for large files.
2022 */
2023 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2024 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2025 else if (!S_ISREG(st->st_mode))
2026 ret = index_pipe(istate, oid, fd, type, path, flags);
2027 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2028 (path && would_convert_to_git(istate, path)))
2029 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2030 type, path, flags);
2031 else
2032 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2033 flags);
2034 close(fd);
2035 return ret;
2036 }
2037
2038 int index_path(struct index_state *istate, struct object_id *oid,
2039 const char *path, struct stat *st, unsigned flags)
2040 {
2041 int fd;
2042 struct strbuf sb = STRBUF_INIT;
2043 int rc = 0;
2044
2045 switch (st->st_mode & S_IFMT) {
2046 case S_IFREG:
2047 fd = open(path, O_RDONLY);
2048 if (fd < 0)
2049 return error_errno("open(\"%s\")", path);
2050 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2051 return error(_("%s: failed to insert into database"),
2052 path);
2053 break;
2054 case S_IFLNK:
2055 if (strbuf_readlink(&sb, path, st->st_size))
2056 return error_errno("readlink(\"%s\")", path);
2057 if (!(flags & HASH_WRITE_OBJECT))
2058 hash_object_file(sb.buf, sb.len, blob_type, oid);
2059 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2060 rc = error(_("%s: failed to insert into database"), path);
2061 strbuf_release(&sb);
2062 break;
2063 case S_IFDIR:
2064 return resolve_gitlink_ref(path, "HEAD", oid);
2065 default:
2066 return error(_("%s: unsupported file type"), path);
2067 }
2068 return rc;
2069 }
2070
2071 int read_pack_header(int fd, struct pack_header *header)
2072 {
2073 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2074 /* "eof before pack header was fully read" */
2075 return PH_ERROR_EOF;
2076
2077 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2078 /* "protocol error (pack signature mismatch detected)" */
2079 return PH_ERROR_PACK_SIGNATURE;
2080 if (!pack_version_ok(header->hdr_version))
2081 /* "protocol error (pack version unsupported)" */
2082 return PH_ERROR_PROTOCOL;
2083 return 0;
2084 }
2085
2086 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2087 {
2088 enum object_type type = oid_object_info(the_repository, oid, NULL);
2089 if (type < 0)
2090 die(_("%s is not a valid object"), oid_to_hex(oid));
2091 if (type != expect)
2092 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2093 type_name(expect));
2094 }
2095
2096 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2097 struct strbuf *path,
2098 each_loose_object_fn obj_cb,
2099 each_loose_cruft_fn cruft_cb,
2100 each_loose_subdir_fn subdir_cb,
2101 void *data)
2102 {
2103 size_t origlen, baselen;
2104 DIR *dir;
2105 struct dirent *de;
2106 int r = 0;
2107 struct object_id oid;
2108
2109 if (subdir_nr > 0xff)
2110 BUG("invalid loose object subdirectory: %x", subdir_nr);
2111
2112 origlen = path->len;
2113 strbuf_complete(path, '/');
2114 strbuf_addf(path, "%02x", subdir_nr);
2115
2116 dir = opendir(path->buf);
2117 if (!dir) {
2118 if (errno != ENOENT)
2119 r = error_errno(_("unable to open %s"), path->buf);
2120 strbuf_setlen(path, origlen);
2121 return r;
2122 }
2123
2124 oid.hash[0] = subdir_nr;
2125 strbuf_addch(path, '/');
2126 baselen = path->len;
2127
2128 while ((de = readdir(dir))) {
2129 size_t namelen;
2130 if (is_dot_or_dotdot(de->d_name))
2131 continue;
2132
2133 namelen = strlen(de->d_name);
2134 strbuf_setlen(path, baselen);
2135 strbuf_add(path, de->d_name, namelen);
2136 if (namelen == the_hash_algo->hexsz - 2 &&
2137 !hex_to_bytes(oid.hash + 1, de->d_name,
2138 the_hash_algo->rawsz - 1)) {
2139 if (obj_cb) {
2140 r = obj_cb(&oid, path->buf, data);
2141 if (r)
2142 break;
2143 }
2144 continue;
2145 }
2146
2147 if (cruft_cb) {
2148 r = cruft_cb(de->d_name, path->buf, data);
2149 if (r)
2150 break;
2151 }
2152 }
2153 closedir(dir);
2154
2155 strbuf_setlen(path, baselen - 1);
2156 if (!r && subdir_cb)
2157 r = subdir_cb(subdir_nr, path->buf, data);
2158
2159 strbuf_setlen(path, origlen);
2160
2161 return r;
2162 }
2163
2164 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2165 each_loose_object_fn obj_cb,
2166 each_loose_cruft_fn cruft_cb,
2167 each_loose_subdir_fn subdir_cb,
2168 void *data)
2169 {
2170 int r = 0;
2171 int i;
2172
2173 for (i = 0; i < 256; i++) {
2174 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2175 subdir_cb, data);
2176 if (r)
2177 break;
2178 }
2179
2180 return r;
2181 }
2182
2183 int for_each_loose_file_in_objdir(const char *path,
2184 each_loose_object_fn obj_cb,
2185 each_loose_cruft_fn cruft_cb,
2186 each_loose_subdir_fn subdir_cb,
2187 void *data)
2188 {
2189 struct strbuf buf = STRBUF_INIT;
2190 int r;
2191
2192 strbuf_addstr(&buf, path);
2193 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2194 subdir_cb, data);
2195 strbuf_release(&buf);
2196
2197 return r;
2198 }
2199
2200 int for_each_loose_object(each_loose_object_fn cb, void *data,
2201 enum for_each_object_flags flags)
2202 {
2203 struct object_directory *odb;
2204
2205 prepare_alt_odb(the_repository);
2206 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2207 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2208 NULL, data);
2209 if (r)
2210 return r;
2211
2212 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2213 break;
2214 }
2215
2216 return 0;
2217 }
2218
2219 static int append_loose_object(const struct object_id *oid, const char *path,
2220 void *data)
2221 {
2222 oid_array_append(data, oid);
2223 return 0;
2224 }
2225
2226 struct oid_array *odb_loose_cache(struct object_directory *odb,
2227 const struct object_id *oid)
2228 {
2229 int subdir_nr = oid->hash[0];
2230 struct strbuf buf = STRBUF_INIT;
2231
2232 if (subdir_nr < 0 ||
2233 subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2234 BUG("subdir_nr out of range");
2235
2236 if (odb->loose_objects_subdir_seen[subdir_nr])
2237 return &odb->loose_objects_cache[subdir_nr];
2238
2239 strbuf_addstr(&buf, odb->path);
2240 for_each_file_in_obj_subdir(subdir_nr, &buf,
2241 append_loose_object,
2242 NULL, NULL,
2243 &odb->loose_objects_cache[subdir_nr]);
2244 odb->loose_objects_subdir_seen[subdir_nr] = 1;
2245 strbuf_release(&buf);
2246 return &odb->loose_objects_cache[subdir_nr];
2247 }
2248
2249 void odb_clear_loose_cache(struct object_directory *odb)
2250 {
2251 int i;
2252
2253 for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2254 oid_array_clear(&odb->loose_objects_cache[i]);
2255 memset(&odb->loose_objects_subdir_seen, 0,
2256 sizeof(odb->loose_objects_subdir_seen));
2257 }
2258
2259 static int check_stream_oid(git_zstream *stream,
2260 const char *hdr,
2261 unsigned long size,
2262 const char *path,
2263 const struct object_id *expected_oid)
2264 {
2265 git_hash_ctx c;
2266 struct object_id real_oid;
2267 unsigned char buf[4096];
2268 unsigned long total_read;
2269 int status = Z_OK;
2270
2271 the_hash_algo->init_fn(&c);
2272 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2273
2274 /*
2275 * We already read some bytes into hdr, but the ones up to the NUL
2276 * do not count against the object's content size.
2277 */
2278 total_read = stream->total_out - strlen(hdr) - 1;
2279
2280 /*
2281 * This size comparison must be "<=" to read the final zlib packets;
2282 * see the comment in unpack_loose_rest for details.
2283 */
2284 while (total_read <= size &&
2285 (status == Z_OK ||
2286 (status == Z_BUF_ERROR && !stream->avail_out))) {
2287 stream->next_out = buf;
2288 stream->avail_out = sizeof(buf);
2289 if (size - total_read < stream->avail_out)
2290 stream->avail_out = size - total_read;
2291 status = git_inflate(stream, Z_FINISH);
2292 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2293 total_read += stream->next_out - buf;
2294 }
2295 git_inflate_end(stream);
2296
2297 if (status != Z_STREAM_END) {
2298 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2299 return -1;
2300 }
2301 if (stream->avail_in) {
2302 error(_("garbage at end of loose object '%s'"),
2303 oid_to_hex(expected_oid));
2304 return -1;
2305 }
2306
2307 the_hash_algo->final_fn(real_oid.hash, &c);
2308 if (!oideq(expected_oid, &real_oid)) {
2309 error(_("hash mismatch for %s (expected %s)"), path,
2310 oid_to_hex(expected_oid));
2311 return -1;
2312 }
2313
2314 return 0;
2315 }
2316
2317 int read_loose_object(const char *path,
2318 const struct object_id *expected_oid,
2319 enum object_type *type,
2320 unsigned long *size,
2321 void **contents)
2322 {
2323 int ret = -1;
2324 void *map = NULL;
2325 unsigned long mapsize;
2326 git_zstream stream;
2327 char hdr[MAX_HEADER_LEN];
2328
2329 *contents = NULL;
2330
2331 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2332 if (!map) {
2333 error_errno(_("unable to mmap %s"), path);
2334 goto out;
2335 }
2336
2337 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2338 error(_("unable to unpack header of %s"), path);
2339 goto out;
2340 }
2341
2342 *type = parse_loose_header(hdr, size);
2343 if (*type < 0) {
2344 error(_("unable to parse header of %s"), path);
2345 git_inflate_end(&stream);
2346 goto out;
2347 }
2348
2349 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2350 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2351 goto out;
2352 } else {
2353 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2354 if (!*contents) {
2355 error(_("unable to unpack contents of %s"), path);
2356 git_inflate_end(&stream);
2357 goto out;
2358 }
2359 if (check_object_signature(expected_oid, *contents,
2360 *size, type_name(*type))) {
2361 error(_("hash mismatch for %s (expected %s)"), path,
2362 oid_to_hex(expected_oid));
2363 free(*contents);
2364 goto out;
2365 }
2366 }
2367
2368 ret = 0; /* everything checks out */
2369
2370 out:
2371 if (map)
2372 munmap(map, mapsize);
2373 return ret;
2374 }