builtin/pack-objects.c: don't leak memory via arguments
[git/debian.git] / object-file.c
blob112d9b4badcfdeb2667aea1f2ea07c3adb6ea8e7
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git 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 "hash-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 #include "submodule.h"
37 /* The maximum size for an object header. */
38 #define MAX_HEADER_LEN 32
41 #define EMPTY_TREE_SHA1_BIN_LITERAL \
42 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
43 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
44 #define EMPTY_TREE_SHA256_BIN_LITERAL \
45 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
46 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
47 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
48 "\x53\x21"
50 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
51 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
52 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
53 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
54 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
55 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
56 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
57 "\x18\x13"
59 static const struct object_id empty_tree_oid = {
60 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
61 .algo = GIT_HASH_SHA1,
63 static const struct object_id empty_blob_oid = {
64 .hash = EMPTY_BLOB_SHA1_BIN_LITERAL,
65 .algo = GIT_HASH_SHA1,
67 static const struct object_id null_oid_sha1 = {
68 .hash = {0},
69 .algo = GIT_HASH_SHA1,
71 static const struct object_id empty_tree_oid_sha256 = {
72 .hash = EMPTY_TREE_SHA256_BIN_LITERAL,
73 .algo = GIT_HASH_SHA256,
75 static const struct object_id empty_blob_oid_sha256 = {
76 .hash = EMPTY_BLOB_SHA256_BIN_LITERAL,
77 .algo = GIT_HASH_SHA256,
79 static const struct object_id null_oid_sha256 = {
80 .hash = {0},
81 .algo = GIT_HASH_SHA256,
84 static void git_hash_sha1_init(git_hash_ctx *ctx)
86 git_SHA1_Init(&ctx->sha1);
89 static void git_hash_sha1_clone(git_hash_ctx *dst, const git_hash_ctx *src)
91 git_SHA1_Clone(&dst->sha1, &src->sha1);
94 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
96 git_SHA1_Update(&ctx->sha1, data, len);
99 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
101 git_SHA1_Final(hash, &ctx->sha1);
104 static void git_hash_sha1_final_oid(struct object_id *oid, git_hash_ctx *ctx)
106 git_SHA1_Final(oid->hash, &ctx->sha1);
107 memset(oid->hash + GIT_SHA1_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA1_RAWSZ);
108 oid->algo = GIT_HASH_SHA1;
112 static void git_hash_sha256_init(git_hash_ctx *ctx)
114 git_SHA256_Init(&ctx->sha256);
117 static void git_hash_sha256_clone(git_hash_ctx *dst, const git_hash_ctx *src)
119 git_SHA256_Clone(&dst->sha256, &src->sha256);
122 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
124 git_SHA256_Update(&ctx->sha256, data, len);
127 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
129 git_SHA256_Final(hash, &ctx->sha256);
132 static void git_hash_sha256_final_oid(struct object_id *oid, git_hash_ctx *ctx)
134 git_SHA256_Final(oid->hash, &ctx->sha256);
136 * This currently does nothing, so the compiler should optimize it out,
137 * but keep it in case we extend the hash size again.
139 memset(oid->hash + GIT_SHA256_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA256_RAWSZ);
140 oid->algo = GIT_HASH_SHA256;
143 static void git_hash_unknown_init(git_hash_ctx *ctx)
145 BUG("trying to init unknown hash");
148 static void git_hash_unknown_clone(git_hash_ctx *dst, const git_hash_ctx *src)
150 BUG("trying to clone unknown hash");
153 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
155 BUG("trying to update unknown hash");
158 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
160 BUG("trying to finalize unknown hash");
163 static void git_hash_unknown_final_oid(struct object_id *oid, git_hash_ctx *ctx)
165 BUG("trying to finalize unknown hash");
169 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
171 NULL,
172 0x00000000,
176 git_hash_unknown_init,
177 git_hash_unknown_clone,
178 git_hash_unknown_update,
179 git_hash_unknown_final,
180 git_hash_unknown_final_oid,
181 NULL,
182 NULL,
183 NULL,
186 "sha1",
187 /* "sha1", big-endian */
188 0x73686131,
189 GIT_SHA1_RAWSZ,
190 GIT_SHA1_HEXSZ,
191 GIT_SHA1_BLKSZ,
192 git_hash_sha1_init,
193 git_hash_sha1_clone,
194 git_hash_sha1_update,
195 git_hash_sha1_final,
196 git_hash_sha1_final_oid,
197 &empty_tree_oid,
198 &empty_blob_oid,
199 &null_oid_sha1,
202 "sha256",
203 /* "s256", big-endian */
204 0x73323536,
205 GIT_SHA256_RAWSZ,
206 GIT_SHA256_HEXSZ,
207 GIT_SHA256_BLKSZ,
208 git_hash_sha256_init,
209 git_hash_sha256_clone,
210 git_hash_sha256_update,
211 git_hash_sha256_final,
212 git_hash_sha256_final_oid,
213 &empty_tree_oid_sha256,
214 &empty_blob_oid_sha256,
215 &null_oid_sha256,
219 const struct object_id *null_oid(void)
221 return the_hash_algo->null_oid;
224 const char *empty_tree_oid_hex(void)
226 static char buf[GIT_MAX_HEXSZ + 1];
227 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
230 const char *empty_blob_oid_hex(void)
232 static char buf[GIT_MAX_HEXSZ + 1];
233 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
236 int hash_algo_by_name(const char *name)
238 int i;
239 if (!name)
240 return GIT_HASH_UNKNOWN;
241 for (i = 1; i < GIT_HASH_NALGOS; i++)
242 if (!strcmp(name, hash_algos[i].name))
243 return i;
244 return GIT_HASH_UNKNOWN;
247 int hash_algo_by_id(uint32_t format_id)
249 int i;
250 for (i = 1; i < GIT_HASH_NALGOS; i++)
251 if (format_id == hash_algos[i].format_id)
252 return i;
253 return GIT_HASH_UNKNOWN;
256 int hash_algo_by_length(int len)
258 int i;
259 for (i = 1; i < GIT_HASH_NALGOS; i++)
260 if (len == hash_algos[i].rawsz)
261 return i;
262 return GIT_HASH_UNKNOWN;
266 * This is meant to hold a *small* number of objects that you would
267 * want read_object_file() to be able to return, but yet you do not want
268 * to write them into the object store (e.g. a browse-only
269 * application).
271 static struct cached_object {
272 struct object_id oid;
273 enum object_type type;
274 void *buf;
275 unsigned long size;
276 } *cached_objects;
277 static int cached_object_nr, cached_object_alloc;
279 static struct cached_object empty_tree = {
280 { EMPTY_TREE_SHA1_BIN_LITERAL },
281 OBJ_TREE,
286 static struct cached_object *find_cached_object(const struct object_id *oid)
288 int i;
289 struct cached_object *co = cached_objects;
291 for (i = 0; i < cached_object_nr; i++, co++) {
292 if (oideq(&co->oid, oid))
293 return co;
295 if (oideq(oid, the_hash_algo->empty_tree))
296 return &empty_tree;
297 return NULL;
301 static int get_conv_flags(unsigned flags)
303 if (flags & HASH_RENORMALIZE)
304 return CONV_EOL_RENORMALIZE;
305 else if (flags & HASH_WRITE_OBJECT)
306 return global_conv_flags_eol | CONV_WRITE_OBJECT;
307 else
308 return 0;
312 int mkdir_in_gitdir(const char *path)
314 if (mkdir(path, 0777)) {
315 int saved_errno = errno;
316 struct stat st;
317 struct strbuf sb = STRBUF_INIT;
319 if (errno != EEXIST)
320 return -1;
322 * Are we looking at a path in a symlinked worktree
323 * whose original repository does not yet have it?
324 * e.g. .git/rr-cache pointing at its original
325 * repository in which the user hasn't performed any
326 * conflict resolution yet?
328 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
329 strbuf_readlink(&sb, path, st.st_size) ||
330 !is_absolute_path(sb.buf) ||
331 mkdir(sb.buf, 0777)) {
332 strbuf_release(&sb);
333 errno = saved_errno;
334 return -1;
336 strbuf_release(&sb);
338 return adjust_shared_perm(path);
341 static enum scld_error safe_create_leading_directories_1(char *path, int share)
343 char *next_component = path + offset_1st_component(path);
344 enum scld_error ret = SCLD_OK;
346 while (ret == SCLD_OK && next_component) {
347 struct stat st;
348 char *slash = next_component, slash_character;
350 while (*slash && !is_dir_sep(*slash))
351 slash++;
353 if (!*slash)
354 break;
356 next_component = slash + 1;
357 while (is_dir_sep(*next_component))
358 next_component++;
359 if (!*next_component)
360 break;
362 slash_character = *slash;
363 *slash = '\0';
364 if (!stat(path, &st)) {
365 /* path exists */
366 if (!S_ISDIR(st.st_mode)) {
367 errno = ENOTDIR;
368 ret = SCLD_EXISTS;
370 } else if (mkdir(path, 0777)) {
371 if (errno == EEXIST &&
372 !stat(path, &st) && S_ISDIR(st.st_mode))
373 ; /* somebody created it since we checked */
374 else if (errno == ENOENT)
376 * Either mkdir() failed because
377 * somebody just pruned the containing
378 * directory, or stat() failed because
379 * the file that was in our way was
380 * just removed. Either way, inform
381 * the caller that it might be worth
382 * trying again:
384 ret = SCLD_VANISHED;
385 else
386 ret = SCLD_FAILED;
387 } else if (share && adjust_shared_perm(path)) {
388 ret = SCLD_PERMS;
390 *slash = slash_character;
392 return ret;
395 enum scld_error safe_create_leading_directories(char *path)
397 return safe_create_leading_directories_1(path, 1);
400 enum scld_error safe_create_leading_directories_no_share(char *path)
402 return safe_create_leading_directories_1(path, 0);
405 enum scld_error safe_create_leading_directories_const(const char *path)
407 int save_errno;
408 /* path points to cache entries, so xstrdup before messing with it */
409 char *buf = xstrdup(path);
410 enum scld_error result = safe_create_leading_directories(buf);
412 save_errno = errno;
413 free(buf);
414 errno = save_errno;
415 return result;
418 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
420 int i;
421 for (i = 0; i < the_hash_algo->rawsz; i++) {
422 static char hex[] = "0123456789abcdef";
423 unsigned int val = oid->hash[i];
424 strbuf_addch(buf, hex[val >> 4]);
425 strbuf_addch(buf, hex[val & 0xf]);
426 if (!i)
427 strbuf_addch(buf, '/');
431 static const char *odb_loose_path(struct object_directory *odb,
432 struct strbuf *buf,
433 const struct object_id *oid)
435 strbuf_reset(buf);
436 strbuf_addstr(buf, odb->path);
437 strbuf_addch(buf, '/');
438 fill_loose_path(buf, oid);
439 return buf->buf;
442 const char *loose_object_path(struct repository *r, struct strbuf *buf,
443 const struct object_id *oid)
445 return odb_loose_path(r->objects->odb, buf, oid);
449 * Return non-zero iff the path is usable as an alternate object database.
451 static int alt_odb_usable(struct raw_object_store *o,
452 struct strbuf *path,
453 const char *normalized_objdir, khiter_t *pos)
455 int r;
457 /* Detect cases where alternate disappeared */
458 if (!is_directory(path->buf)) {
459 error(_("object directory %s does not exist; "
460 "check .git/objects/info/alternates"),
461 path->buf);
462 return 0;
466 * Prevent the common mistake of listing the same
467 * thing twice, or object directory itself.
469 if (!o->odb_by_path) {
470 khiter_t p;
472 o->odb_by_path = kh_init_odb_path_map();
473 assert(!o->odb->next);
474 p = kh_put_odb_path_map(o->odb_by_path, o->odb->path, &r);
475 assert(r == 1); /* never used */
476 kh_value(o->odb_by_path, p) = o->odb;
478 if (fspatheq(path->buf, normalized_objdir))
479 return 0;
480 *pos = kh_put_odb_path_map(o->odb_by_path, path->buf, &r);
481 /* r: 0 = exists, 1 = never used, 2 = deleted */
482 return r == 0 ? 0 : 1;
486 * Prepare alternate object database registry.
488 * The variable alt_odb_list points at the list of struct
489 * object_directory. The elements on this list come from
490 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
491 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
492 * whose contents is similar to that environment variable but can be
493 * LF separated. Its base points at a statically allocated buffer that
494 * contains "/the/directory/corresponding/to/.git/objects/...", while
495 * its name points just after the slash at the end of ".git/objects/"
496 * in the example above, and has enough space to hold all hex characters
497 * of the object ID, an extra slash for the first level indirection, and
498 * the terminating NUL.
500 static void read_info_alternates(struct repository *r,
501 const char *relative_base,
502 int depth);
503 static int link_alt_odb_entry(struct repository *r, const struct strbuf *entry,
504 const char *relative_base, int depth, const char *normalized_objdir)
506 struct object_directory *ent;
507 struct strbuf pathbuf = STRBUF_INIT;
508 khiter_t pos;
510 if (!is_absolute_path(entry->buf) && relative_base) {
511 strbuf_realpath(&pathbuf, relative_base, 1);
512 strbuf_addch(&pathbuf, '/');
514 strbuf_addbuf(&pathbuf, entry);
516 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
517 error(_("unable to normalize alternate object path: %s"),
518 pathbuf.buf);
519 strbuf_release(&pathbuf);
520 return -1;
524 * The trailing slash after the directory name is given by
525 * this function at the end. Remove duplicates.
527 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
528 strbuf_setlen(&pathbuf, pathbuf.len - 1);
530 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir, &pos)) {
531 strbuf_release(&pathbuf);
532 return -1;
535 CALLOC_ARRAY(ent, 1);
536 /* pathbuf.buf is already in r->objects->odb_by_path */
537 ent->path = strbuf_detach(&pathbuf, NULL);
539 /* add the alternate entry */
540 *r->objects->odb_tail = ent;
541 r->objects->odb_tail = &(ent->next);
542 ent->next = NULL;
543 assert(r->objects->odb_by_path);
544 kh_value(r->objects->odb_by_path, pos) = ent;
546 /* recursively add alternates */
547 read_info_alternates(r, ent->path, depth + 1);
549 return 0;
552 static const char *parse_alt_odb_entry(const char *string,
553 int sep,
554 struct strbuf *out)
556 const char *end;
558 strbuf_reset(out);
560 if (*string == '#') {
561 /* comment; consume up to next separator */
562 end = strchrnul(string, sep);
563 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
565 * quoted path; unquote_c_style has copied the
566 * data for us and set "end". Broken quoting (e.g.,
567 * an entry that doesn't end with a quote) falls
568 * back to the unquoted case below.
570 } else {
571 /* normal, unquoted path */
572 end = strchrnul(string, sep);
573 strbuf_add(out, string, end - string);
576 if (*end)
577 end++;
578 return end;
581 static void link_alt_odb_entries(struct repository *r, const char *alt,
582 int sep, const char *relative_base, int depth)
584 struct strbuf objdirbuf = STRBUF_INIT;
585 struct strbuf entry = STRBUF_INIT;
587 if (!alt || !*alt)
588 return;
590 if (depth > 5) {
591 error(_("%s: ignoring alternate object stores, nesting too deep"),
592 relative_base);
593 return;
596 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
597 if (strbuf_normalize_path(&objdirbuf) < 0)
598 die(_("unable to normalize object directory: %s"),
599 objdirbuf.buf);
601 while (*alt) {
602 alt = parse_alt_odb_entry(alt, sep, &entry);
603 if (!entry.len)
604 continue;
605 link_alt_odb_entry(r, &entry,
606 relative_base, depth, objdirbuf.buf);
608 strbuf_release(&entry);
609 strbuf_release(&objdirbuf);
612 static void read_info_alternates(struct repository *r,
613 const char *relative_base,
614 int depth)
616 char *path;
617 struct strbuf buf = STRBUF_INIT;
619 path = xstrfmt("%s/info/alternates", relative_base);
620 if (strbuf_read_file(&buf, path, 1024) < 0) {
621 warn_on_fopen_errors(path);
622 free(path);
623 return;
626 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
627 strbuf_release(&buf);
628 free(path);
631 void add_to_alternates_file(const char *reference)
633 struct lock_file lock = LOCK_INIT;
634 char *alts = git_pathdup("objects/info/alternates");
635 FILE *in, *out;
636 int found = 0;
638 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
639 out = fdopen_lock_file(&lock, "w");
640 if (!out)
641 die_errno(_("unable to fdopen alternates lockfile"));
643 in = fopen(alts, "r");
644 if (in) {
645 struct strbuf line = STRBUF_INIT;
647 while (strbuf_getline(&line, in) != EOF) {
648 if (!strcmp(reference, line.buf)) {
649 found = 1;
650 break;
652 fprintf_or_die(out, "%s\n", line.buf);
655 strbuf_release(&line);
656 fclose(in);
658 else if (errno != ENOENT)
659 die_errno(_("unable to read alternates file"));
661 if (found) {
662 rollback_lock_file(&lock);
663 } else {
664 fprintf_or_die(out, "%s\n", reference);
665 if (commit_lock_file(&lock))
666 die_errno(_("unable to move new alternates file into place"));
667 if (the_repository->objects->loaded_alternates)
668 link_alt_odb_entries(the_repository, reference,
669 '\n', NULL, 0);
671 free(alts);
674 void add_to_alternates_memory(const char *reference)
677 * Make sure alternates are initialized, or else our entry may be
678 * overwritten when they are.
680 prepare_alt_odb(the_repository);
682 link_alt_odb_entries(the_repository, reference,
683 '\n', NULL, 0);
687 * Compute the exact path an alternate is at and returns it. In case of
688 * error NULL is returned and the human readable error is added to `err`
689 * `path` may be relative and should point to $GIT_DIR.
690 * `err` must not be null.
692 char *compute_alternate_path(const char *path, struct strbuf *err)
694 char *ref_git = NULL;
695 const char *repo;
696 int seen_error = 0;
698 ref_git = real_pathdup(path, 0);
699 if (!ref_git) {
700 seen_error = 1;
701 strbuf_addf(err, _("path '%s' does not exist"), path);
702 goto out;
705 repo = read_gitfile(ref_git);
706 if (!repo)
707 repo = read_gitfile(mkpath("%s/.git", ref_git));
708 if (repo) {
709 free(ref_git);
710 ref_git = xstrdup(repo);
713 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
714 char *ref_git_git = mkpathdup("%s/.git", ref_git);
715 free(ref_git);
716 ref_git = ref_git_git;
717 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
718 struct strbuf sb = STRBUF_INIT;
719 seen_error = 1;
720 if (get_common_dir(&sb, ref_git)) {
721 strbuf_addf(err,
722 _("reference repository '%s' as a linked "
723 "checkout is not supported yet."),
724 path);
725 goto out;
728 strbuf_addf(err, _("reference repository '%s' is not a "
729 "local repository."), path);
730 goto out;
733 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
734 strbuf_addf(err, _("reference repository '%s' is shallow"),
735 path);
736 seen_error = 1;
737 goto out;
740 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
741 strbuf_addf(err,
742 _("reference repository '%s' is grafted"),
743 path);
744 seen_error = 1;
745 goto out;
748 out:
749 if (seen_error) {
750 FREE_AND_NULL(ref_git);
753 return ref_git;
756 struct object_directory *find_odb(struct repository *r, const char *obj_dir)
758 struct object_directory *odb;
759 char *obj_dir_real = real_pathdup(obj_dir, 1);
760 struct strbuf odb_path_real = STRBUF_INIT;
762 prepare_alt_odb(r);
763 for (odb = r->objects->odb; odb; odb = odb->next) {
764 strbuf_realpath(&odb_path_real, odb->path, 1);
765 if (!strcmp(obj_dir_real, odb_path_real.buf))
766 break;
769 free(obj_dir_real);
770 strbuf_release(&odb_path_real);
772 if (!odb)
773 die(_("could not find object directory matching %s"), obj_dir);
774 return odb;
777 static void fill_alternate_refs_command(struct child_process *cmd,
778 const char *repo_path)
780 const char *value;
782 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
783 cmd->use_shell = 1;
785 strvec_push(&cmd->args, value);
786 strvec_push(&cmd->args, repo_path);
787 } else {
788 cmd->git_cmd = 1;
790 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
791 strvec_push(&cmd->args, "for-each-ref");
792 strvec_push(&cmd->args, "--format=%(objectname)");
794 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
795 strvec_push(&cmd->args, "--");
796 strvec_split(&cmd->args, value);
800 cmd->env = local_repo_env;
801 cmd->out = -1;
804 static void read_alternate_refs(const char *path,
805 alternate_ref_fn *cb,
806 void *data)
808 struct child_process cmd = CHILD_PROCESS_INIT;
809 struct strbuf line = STRBUF_INIT;
810 FILE *fh;
812 fill_alternate_refs_command(&cmd, path);
814 if (start_command(&cmd))
815 return;
817 fh = xfdopen(cmd.out, "r");
818 while (strbuf_getline_lf(&line, fh) != EOF) {
819 struct object_id oid;
820 const char *p;
822 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
823 warning(_("invalid line while parsing alternate refs: %s"),
824 line.buf);
825 break;
828 cb(&oid, data);
831 fclose(fh);
832 finish_command(&cmd);
833 strbuf_release(&line);
836 struct alternate_refs_data {
837 alternate_ref_fn *fn;
838 void *data;
841 static int refs_from_alternate_cb(struct object_directory *e,
842 void *data)
844 struct strbuf path = STRBUF_INIT;
845 size_t base_len;
846 struct alternate_refs_data *cb = data;
848 if (!strbuf_realpath(&path, e->path, 0))
849 goto out;
850 if (!strbuf_strip_suffix(&path, "/objects"))
851 goto out;
852 base_len = path.len;
854 /* Is this a git repository with refs? */
855 strbuf_addstr(&path, "/refs");
856 if (!is_directory(path.buf))
857 goto out;
858 strbuf_setlen(&path, base_len);
860 read_alternate_refs(path.buf, cb->fn, cb->data);
862 out:
863 strbuf_release(&path);
864 return 0;
867 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
869 struct alternate_refs_data cb;
870 cb.fn = fn;
871 cb.data = data;
872 foreach_alt_odb(refs_from_alternate_cb, &cb);
875 int foreach_alt_odb(alt_odb_fn fn, void *cb)
877 struct object_directory *ent;
878 int r = 0;
880 prepare_alt_odb(the_repository);
881 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
882 r = fn(ent, cb);
883 if (r)
884 break;
886 return r;
889 void prepare_alt_odb(struct repository *r)
891 if (r->objects->loaded_alternates)
892 return;
894 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
896 read_info_alternates(r, r->objects->odb->path, 0);
897 r->objects->loaded_alternates = 1;
900 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
901 static int freshen_file(const char *fn)
903 return !utime(fn, NULL);
907 * All of the check_and_freshen functions return 1 if the file exists and was
908 * freshened (if freshening was requested), 0 otherwise. If they return
909 * 0, you should not assume that it is safe to skip a write of the object (it
910 * either does not exist on disk, or has a stale mtime and may be subject to
911 * pruning).
913 int check_and_freshen_file(const char *fn, int freshen)
915 if (access(fn, F_OK))
916 return 0;
917 if (freshen && !freshen_file(fn))
918 return 0;
919 return 1;
922 static int check_and_freshen_odb(struct object_directory *odb,
923 const struct object_id *oid,
924 int freshen)
926 static struct strbuf path = STRBUF_INIT;
927 odb_loose_path(odb, &path, oid);
928 return check_and_freshen_file(path.buf, freshen);
931 static int check_and_freshen_local(const struct object_id *oid, int freshen)
933 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
936 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
938 struct object_directory *odb;
940 prepare_alt_odb(the_repository);
941 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
942 if (check_and_freshen_odb(odb, oid, freshen))
943 return 1;
945 return 0;
948 static int check_and_freshen(const struct object_id *oid, int freshen)
950 return check_and_freshen_local(oid, freshen) ||
951 check_and_freshen_nonlocal(oid, freshen);
954 int has_loose_object_nonlocal(const struct object_id *oid)
956 return check_and_freshen_nonlocal(oid, 0);
959 static int has_loose_object(const struct object_id *oid)
961 return check_and_freshen(oid, 0);
964 static void mmap_limit_check(size_t length)
966 static size_t limit = 0;
967 if (!limit) {
968 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
969 if (!limit)
970 limit = SIZE_MAX;
972 if (length > limit)
973 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
974 (uintmax_t)length, (uintmax_t)limit);
977 void *xmmap_gently(void *start, size_t length,
978 int prot, int flags, int fd, off_t offset)
980 void *ret;
982 mmap_limit_check(length);
983 ret = mmap(start, length, prot, flags, fd, offset);
984 if (ret == MAP_FAILED && !length)
985 ret = NULL;
986 return ret;
989 const char *mmap_os_err(void)
991 static const char blank[] = "";
992 #if defined(__linux__)
993 if (errno == ENOMEM) {
994 /* this continues an existing error message: */
995 static const char enomem[] =
996 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
997 return enomem;
999 #endif /* OS-specific bits */
1000 return blank;
1003 void *xmmap(void *start, size_t length,
1004 int prot, int flags, int fd, off_t offset)
1006 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1007 if (ret == MAP_FAILED)
1008 die_errno(_("mmap failed%s"), mmap_os_err());
1009 return ret;
1013 * With an in-core object data in "map", rehash it to make sure the
1014 * object name actually matches "oid" to detect object corruption.
1015 * With "map" == NULL, try reading the object named with "oid" using
1016 * the streaming interface and rehash it to do the same.
1018 int check_object_signature(struct repository *r, const struct object_id *oid,
1019 void *map, unsigned long size, const char *type)
1021 struct object_id real_oid;
1022 enum object_type obj_type;
1023 struct git_istream *st;
1024 git_hash_ctx c;
1025 char hdr[MAX_HEADER_LEN];
1026 int hdrlen;
1028 if (map) {
1029 hash_object_file(r->hash_algo, map, size, type, &real_oid);
1030 return !oideq(oid, &real_oid) ? -1 : 0;
1033 st = open_istream(r, oid, &obj_type, &size, NULL);
1034 if (!st)
1035 return -1;
1037 /* Generate the header */
1038 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1040 /* Sha1.. */
1041 r->hash_algo->init_fn(&c);
1042 r->hash_algo->update_fn(&c, hdr, hdrlen);
1043 for (;;) {
1044 char buf[1024 * 16];
1045 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1047 if (readlen < 0) {
1048 close_istream(st);
1049 return -1;
1051 if (!readlen)
1052 break;
1053 r->hash_algo->update_fn(&c, buf, readlen);
1055 r->hash_algo->final_oid_fn(&real_oid, &c);
1056 close_istream(st);
1057 return !oideq(oid, &real_oid) ? -1 : 0;
1060 int git_open_cloexec(const char *name, int flags)
1062 int fd;
1063 static int o_cloexec = O_CLOEXEC;
1065 fd = open(name, flags | o_cloexec);
1066 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1067 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1068 o_cloexec &= ~O_CLOEXEC;
1069 fd = open(name, flags | o_cloexec);
1072 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1074 static int fd_cloexec = FD_CLOEXEC;
1076 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1077 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1078 int flags = fcntl(fd, F_GETFD);
1079 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1080 fd_cloexec = 0;
1083 #endif
1084 return fd;
1088 * Find "oid" as a loose object in the local repository or in an alternate.
1089 * Returns 0 on success, negative on failure.
1091 * The "path" out-parameter will give the path of the object we found (if any).
1092 * Note that it may point to static storage and is only valid until another
1093 * call to stat_loose_object().
1095 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1096 struct stat *st, const char **path)
1098 struct object_directory *odb;
1099 static struct strbuf buf = STRBUF_INIT;
1101 prepare_alt_odb(r);
1102 for (odb = r->objects->odb; odb; odb = odb->next) {
1103 *path = odb_loose_path(odb, &buf, oid);
1104 if (!lstat(*path, st))
1105 return 0;
1108 return -1;
1112 * Like stat_loose_object(), but actually open the object and return the
1113 * descriptor. See the caveats on the "path" parameter above.
1115 static int open_loose_object(struct repository *r,
1116 const struct object_id *oid, const char **path)
1118 int fd;
1119 struct object_directory *odb;
1120 int most_interesting_errno = ENOENT;
1121 static struct strbuf buf = STRBUF_INIT;
1123 prepare_alt_odb(r);
1124 for (odb = r->objects->odb; odb; odb = odb->next) {
1125 *path = odb_loose_path(odb, &buf, oid);
1126 fd = git_open(*path);
1127 if (fd >= 0)
1128 return fd;
1130 if (most_interesting_errno == ENOENT)
1131 most_interesting_errno = errno;
1133 errno = most_interesting_errno;
1134 return -1;
1137 static int quick_has_loose(struct repository *r,
1138 const struct object_id *oid)
1140 struct object_directory *odb;
1142 prepare_alt_odb(r);
1143 for (odb = r->objects->odb; odb; odb = odb->next) {
1144 if (oidtree_contains(odb_loose_cache(odb, oid), oid))
1145 return 1;
1147 return 0;
1151 * Map the loose object at "path" if it is not NULL, or the path found by
1152 * searching for a loose object named "oid".
1154 static void *map_loose_object_1(struct repository *r, const char *path,
1155 const struct object_id *oid, unsigned long *size)
1157 void *map;
1158 int fd;
1160 if (path)
1161 fd = git_open(path);
1162 else
1163 fd = open_loose_object(r, oid, &path);
1164 map = NULL;
1165 if (fd >= 0) {
1166 struct stat st;
1168 if (!fstat(fd, &st)) {
1169 *size = xsize_t(st.st_size);
1170 if (!*size) {
1171 /* mmap() is forbidden on empty files */
1172 error(_("object file %s is empty"), path);
1173 close(fd);
1174 return NULL;
1176 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1178 close(fd);
1180 return map;
1183 void *map_loose_object(struct repository *r,
1184 const struct object_id *oid,
1185 unsigned long *size)
1187 return map_loose_object_1(r, NULL, oid, size);
1190 static int unpack_loose_short_header(git_zstream *stream,
1191 unsigned char *map, unsigned long mapsize,
1192 void *buffer, unsigned long bufsiz)
1194 int ret;
1196 /* Get the data stream */
1197 memset(stream, 0, sizeof(*stream));
1198 stream->next_in = map;
1199 stream->avail_in = mapsize;
1200 stream->next_out = buffer;
1201 stream->avail_out = bufsiz;
1203 git_inflate_init(stream);
1204 obj_read_unlock();
1205 ret = git_inflate(stream, 0);
1206 obj_read_lock();
1208 return ret;
1211 int unpack_loose_header(git_zstream *stream,
1212 unsigned char *map, unsigned long mapsize,
1213 void *buffer, unsigned long bufsiz)
1215 int status = unpack_loose_short_header(stream, map, mapsize,
1216 buffer, bufsiz);
1218 if (status < Z_OK)
1219 return status;
1221 /* Make sure we have the terminating NUL */
1222 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1223 return -1;
1224 return 0;
1227 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1228 unsigned long mapsize, void *buffer,
1229 unsigned long bufsiz, struct strbuf *header)
1231 int status;
1233 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1234 if (status < Z_OK)
1235 return -1;
1238 * Check if entire header is unpacked in the first iteration.
1240 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1241 return 0;
1244 * buffer[0..bufsiz] was not large enough. Copy the partial
1245 * result out to header, and then append the result of further
1246 * reading the stream.
1248 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1249 stream->next_out = buffer;
1250 stream->avail_out = bufsiz;
1252 do {
1253 obj_read_unlock();
1254 status = git_inflate(stream, 0);
1255 obj_read_lock();
1256 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1257 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1258 return 0;
1259 stream->next_out = buffer;
1260 stream->avail_out = bufsiz;
1261 } while (status != Z_STREAM_END);
1262 return -1;
1265 static void *unpack_loose_rest(git_zstream *stream,
1266 void *buffer, unsigned long size,
1267 const struct object_id *oid)
1269 int bytes = strlen(buffer) + 1;
1270 unsigned char *buf = xmallocz(size);
1271 unsigned long n;
1272 int status = Z_OK;
1274 n = stream->total_out - bytes;
1275 if (n > size)
1276 n = size;
1277 memcpy(buf, (char *) buffer + bytes, n);
1278 bytes = n;
1279 if (bytes <= size) {
1281 * The above condition must be (bytes <= size), not
1282 * (bytes < size). In other words, even though we
1283 * expect no more output and set avail_out to zero,
1284 * the input zlib stream may have bytes that express
1285 * "this concludes the stream", and we *do* want to
1286 * eat that input.
1288 * Otherwise we would not be able to test that we
1289 * consumed all the input to reach the expected size;
1290 * we also want to check that zlib tells us that all
1291 * went well with status == Z_STREAM_END at the end.
1293 stream->next_out = buf + bytes;
1294 stream->avail_out = size - bytes;
1295 while (status == Z_OK) {
1296 obj_read_unlock();
1297 status = git_inflate(stream, Z_FINISH);
1298 obj_read_lock();
1301 if (status == Z_STREAM_END && !stream->avail_in) {
1302 git_inflate_end(stream);
1303 return buf;
1306 if (status < 0)
1307 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1308 else if (stream->avail_in)
1309 error(_("garbage at end of loose object '%s'"),
1310 oid_to_hex(oid));
1311 free(buf);
1312 return NULL;
1316 * We used to just use "sscanf()", but that's actually way
1317 * too permissive for what we want to check. So do an anal
1318 * object header parse by hand.
1320 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1321 unsigned int flags)
1323 const char *type_buf = hdr;
1324 unsigned long size;
1325 int type, type_len = 0;
1328 * The type can be of any size but is followed by
1329 * a space.
1331 for (;;) {
1332 char c = *hdr++;
1333 if (!c)
1334 return -1;
1335 if (c == ' ')
1336 break;
1337 type_len++;
1340 type = type_from_string_gently(type_buf, type_len, 1);
1341 if (oi->type_name)
1342 strbuf_add(oi->type_name, type_buf, type_len);
1344 * Set type to 0 if its an unknown object and
1345 * we're obtaining the type using '--allow-unknown-type'
1346 * option.
1348 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1349 type = 0;
1350 else if (type < 0)
1351 die(_("invalid object type"));
1352 if (oi->typep)
1353 *oi->typep = type;
1356 * The length must follow immediately, and be in canonical
1357 * decimal format (ie "010" is not valid).
1359 size = *hdr++ - '0';
1360 if (size > 9)
1361 return -1;
1362 if (size) {
1363 for (;;) {
1364 unsigned long c = *hdr - '0';
1365 if (c > 9)
1366 break;
1367 hdr++;
1368 size = size * 10 + c;
1372 if (oi->sizep)
1373 *oi->sizep = size;
1376 * The length must be followed by a zero byte
1378 return *hdr ? -1 : type;
1381 int parse_loose_header(const char *hdr, unsigned long *sizep)
1383 struct object_info oi = OBJECT_INFO_INIT;
1385 oi.sizep = sizep;
1386 return parse_loose_header_extended(hdr, &oi, 0);
1389 static int loose_object_info(struct repository *r,
1390 const struct object_id *oid,
1391 struct object_info *oi, int flags)
1393 int status = 0;
1394 unsigned long mapsize;
1395 void *map;
1396 git_zstream stream;
1397 char hdr[MAX_HEADER_LEN];
1398 struct strbuf hdrbuf = STRBUF_INIT;
1399 unsigned long size_scratch;
1401 if (oi->delta_base_oid)
1402 oidclr(oi->delta_base_oid);
1405 * If we don't care about type or size, then we don't
1406 * need to look inside the object at all. Note that we
1407 * do not optimize out the stat call, even if the
1408 * caller doesn't care about the disk-size, since our
1409 * return value implicitly indicates whether the
1410 * object even exists.
1412 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1413 const char *path;
1414 struct stat st;
1415 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1416 return quick_has_loose(r, oid) ? 0 : -1;
1417 if (stat_loose_object(r, oid, &st, &path) < 0)
1418 return -1;
1419 if (oi->disk_sizep)
1420 *oi->disk_sizep = st.st_size;
1421 return 0;
1424 map = map_loose_object(r, oid, &mapsize);
1425 if (!map)
1426 return -1;
1428 if (!oi->sizep)
1429 oi->sizep = &size_scratch;
1431 if (oi->disk_sizep)
1432 *oi->disk_sizep = mapsize;
1433 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1434 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1435 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1436 oid_to_hex(oid));
1437 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1438 status = error(_("unable to unpack %s header"),
1439 oid_to_hex(oid));
1440 if (status < 0)
1441 ; /* Do nothing */
1442 else if (hdrbuf.len) {
1443 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1444 status = error(_("unable to parse %s header with --allow-unknown-type"),
1445 oid_to_hex(oid));
1446 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1447 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1449 if (status >= 0 && oi->contentp) {
1450 *oi->contentp = unpack_loose_rest(&stream, hdr,
1451 *oi->sizep, oid);
1452 if (!*oi->contentp) {
1453 git_inflate_end(&stream);
1454 status = -1;
1456 } else
1457 git_inflate_end(&stream);
1459 munmap(map, mapsize);
1460 if (status && oi->typep)
1461 *oi->typep = status;
1462 if (oi->sizep == &size_scratch)
1463 oi->sizep = NULL;
1464 strbuf_release(&hdrbuf);
1465 oi->whence = OI_LOOSE;
1466 return (status < 0) ? status : 0;
1469 int obj_read_use_lock = 0;
1470 pthread_mutex_t obj_read_mutex;
1472 void enable_obj_read_lock(void)
1474 if (obj_read_use_lock)
1475 return;
1477 obj_read_use_lock = 1;
1478 init_recursive_mutex(&obj_read_mutex);
1481 void disable_obj_read_lock(void)
1483 if (!obj_read_use_lock)
1484 return;
1486 obj_read_use_lock = 0;
1487 pthread_mutex_destroy(&obj_read_mutex);
1490 int fetch_if_missing = 1;
1492 static int do_oid_object_info_extended(struct repository *r,
1493 const struct object_id *oid,
1494 struct object_info *oi, unsigned flags)
1496 static struct object_info blank_oi = OBJECT_INFO_INIT;
1497 struct cached_object *co;
1498 struct pack_entry e;
1499 int rtype;
1500 const struct object_id *real = oid;
1501 int already_retried = 0;
1504 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1505 real = lookup_replace_object(r, oid);
1507 if (is_null_oid(real))
1508 return -1;
1510 if (!oi)
1511 oi = &blank_oi;
1513 co = find_cached_object(real);
1514 if (co) {
1515 if (oi->typep)
1516 *(oi->typep) = co->type;
1517 if (oi->sizep)
1518 *(oi->sizep) = co->size;
1519 if (oi->disk_sizep)
1520 *(oi->disk_sizep) = 0;
1521 if (oi->delta_base_oid)
1522 oidclr(oi->delta_base_oid);
1523 if (oi->type_name)
1524 strbuf_addstr(oi->type_name, type_name(co->type));
1525 if (oi->contentp)
1526 *oi->contentp = xmemdupz(co->buf, co->size);
1527 oi->whence = OI_CACHED;
1528 return 0;
1531 while (1) {
1532 if (find_pack_entry(r, real, &e))
1533 break;
1535 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1536 return -1;
1538 /* Most likely it's a loose object. */
1539 if (!loose_object_info(r, real, oi, flags))
1540 return 0;
1542 /* Not a loose object; someone else may have just packed it. */
1543 if (!(flags & OBJECT_INFO_QUICK)) {
1544 reprepare_packed_git(r);
1545 if (find_pack_entry(r, real, &e))
1546 break;
1549 if (register_all_submodule_odb_as_alternates())
1550 /* We added some alternates; retry */
1551 continue;
1553 /* Check if it is a missing object */
1554 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1555 !already_retried &&
1556 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1558 * TODO Investigate checking promisor_remote_get_direct()
1559 * TODO return value and stopping on error here.
1561 promisor_remote_get_direct(r, real, 1);
1562 already_retried = 1;
1563 continue;
1566 return -1;
1569 if (oi == &blank_oi)
1571 * We know that the caller doesn't actually need the
1572 * information below, so return early.
1574 return 0;
1575 rtype = packed_object_info(r, e.p, e.offset, oi);
1576 if (rtype < 0) {
1577 mark_bad_packed_object(e.p, real);
1578 return do_oid_object_info_extended(r, real, oi, 0);
1579 } else if (oi->whence == OI_PACKED) {
1580 oi->u.packed.offset = e.offset;
1581 oi->u.packed.pack = e.p;
1582 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1583 rtype == OBJ_OFS_DELTA);
1586 return 0;
1589 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1590 struct object_info *oi, unsigned flags)
1592 int ret;
1593 obj_read_lock();
1594 ret = do_oid_object_info_extended(r, oid, oi, flags);
1595 obj_read_unlock();
1596 return ret;
1600 /* returns enum object_type or negative */
1601 int oid_object_info(struct repository *r,
1602 const struct object_id *oid,
1603 unsigned long *sizep)
1605 enum object_type type;
1606 struct object_info oi = OBJECT_INFO_INIT;
1608 oi.typep = &type;
1609 oi.sizep = sizep;
1610 if (oid_object_info_extended(r, oid, &oi,
1611 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1612 return -1;
1613 return type;
1616 static void *read_object(struct repository *r,
1617 const struct object_id *oid, enum object_type *type,
1618 unsigned long *size)
1620 struct object_info oi = OBJECT_INFO_INIT;
1621 void *content;
1622 oi.typep = type;
1623 oi.sizep = size;
1624 oi.contentp = &content;
1626 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1627 return NULL;
1628 return content;
1631 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1632 struct object_id *oid)
1634 struct cached_object *co;
1636 hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1637 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1638 find_cached_object(oid))
1639 return 0;
1640 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1641 co = &cached_objects[cached_object_nr++];
1642 co->size = len;
1643 co->type = type;
1644 co->buf = xmalloc(len);
1645 memcpy(co->buf, buf, len);
1646 oidcpy(&co->oid, oid);
1647 return 0;
1651 * This function dies on corrupt objects; the callers who want to
1652 * deal with them should arrange to call read_object() and give error
1653 * messages themselves.
1655 void *read_object_file_extended(struct repository *r,
1656 const struct object_id *oid,
1657 enum object_type *type,
1658 unsigned long *size,
1659 int lookup_replace)
1661 void *data;
1662 const struct packed_git *p;
1663 const char *path;
1664 struct stat st;
1665 const struct object_id *repl = lookup_replace ?
1666 lookup_replace_object(r, oid) : oid;
1668 errno = 0;
1669 data = read_object(r, repl, type, size);
1670 if (data)
1671 return data;
1673 obj_read_lock();
1674 if (errno && errno != ENOENT)
1675 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1677 /* die if we replaced an object with one that does not exist */
1678 if (repl != oid)
1679 die(_("replacement %s not found for %s"),
1680 oid_to_hex(repl), oid_to_hex(oid));
1682 if (!stat_loose_object(r, repl, &st, &path))
1683 die(_("loose object %s (stored in %s) is corrupt"),
1684 oid_to_hex(repl), path);
1686 if ((p = has_packed_and_bad(r, repl)) != NULL)
1687 die(_("packed object %s (stored in %s) is corrupt"),
1688 oid_to_hex(repl), p->pack_name);
1689 obj_read_unlock();
1691 return NULL;
1694 void *read_object_with_reference(struct repository *r,
1695 const struct object_id *oid,
1696 const char *required_type_name,
1697 unsigned long *size,
1698 struct object_id *actual_oid_return)
1700 enum object_type type, required_type;
1701 void *buffer;
1702 unsigned long isize;
1703 struct object_id actual_oid;
1705 required_type = type_from_string(required_type_name);
1706 oidcpy(&actual_oid, oid);
1707 while (1) {
1708 int ref_length = -1;
1709 const char *ref_type = NULL;
1711 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1712 if (!buffer)
1713 return NULL;
1714 if (type == required_type) {
1715 *size = isize;
1716 if (actual_oid_return)
1717 oidcpy(actual_oid_return, &actual_oid);
1718 return buffer;
1720 /* Handle references */
1721 else if (type == OBJ_COMMIT)
1722 ref_type = "tree ";
1723 else if (type == OBJ_TAG)
1724 ref_type = "object ";
1725 else {
1726 free(buffer);
1727 return NULL;
1729 ref_length = strlen(ref_type);
1731 if (ref_length + the_hash_algo->hexsz > isize ||
1732 memcmp(buffer, ref_type, ref_length) ||
1733 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1734 free(buffer);
1735 return NULL;
1737 free(buffer);
1738 /* Now we have the ID of the referred-to object in
1739 * actual_oid. Check again. */
1743 static void write_object_file_prepare(const struct git_hash_algo *algo,
1744 const void *buf, unsigned long len,
1745 const char *type, struct object_id *oid,
1746 char *hdr, int *hdrlen)
1748 git_hash_ctx c;
1750 /* Generate the header */
1751 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1753 /* Sha1.. */
1754 algo->init_fn(&c);
1755 algo->update_fn(&c, hdr, *hdrlen);
1756 algo->update_fn(&c, buf, len);
1757 algo->final_oid_fn(oid, &c);
1761 * Move the just written object into its final resting place.
1763 int finalize_object_file(const char *tmpfile, const char *filename)
1765 int ret = 0;
1767 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1768 goto try_rename;
1769 else if (link(tmpfile, filename))
1770 ret = errno;
1773 * Coda hack - coda doesn't like cross-directory links,
1774 * so we fall back to a rename, which will mean that it
1775 * won't be able to check collisions, but that's not a
1776 * big deal.
1778 * The same holds for FAT formatted media.
1780 * When this succeeds, we just return. We have nothing
1781 * left to unlink.
1783 if (ret && ret != EEXIST) {
1784 try_rename:
1785 if (!rename(tmpfile, filename))
1786 goto out;
1787 ret = errno;
1789 unlink_or_warn(tmpfile);
1790 if (ret) {
1791 if (ret != EEXIST) {
1792 return error_errno(_("unable to write file %s"), filename);
1794 /* FIXME!!! Collision check here ? */
1797 out:
1798 if (adjust_shared_perm(filename))
1799 return error(_("unable to set permission to '%s'"), filename);
1800 return 0;
1803 static int write_buffer(int fd, const void *buf, size_t len)
1805 if (write_in_full(fd, buf, len) < 0)
1806 return error_errno(_("file write error"));
1807 return 0;
1810 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1811 unsigned long len, const char *type,
1812 struct object_id *oid)
1814 char hdr[MAX_HEADER_LEN];
1815 int hdrlen = sizeof(hdr);
1816 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1817 return 0;
1820 /* Finalize a file on disk, and close it. */
1821 static void close_loose_object(int fd)
1823 if (fsync_object_files)
1824 fsync_or_die(fd, "loose object file");
1825 if (close(fd) != 0)
1826 die_errno(_("error when closing loose object file"));
1829 /* Size of directory component, including the ending '/' */
1830 static inline int directory_size(const char *filename)
1832 const char *s = strrchr(filename, '/');
1833 if (!s)
1834 return 0;
1835 return s - filename + 1;
1839 * This creates a temporary file in the same directory as the final
1840 * 'filename'
1842 * We want to avoid cross-directory filename renames, because those
1843 * can have problems on various filesystems (FAT, NFS, Coda).
1845 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1847 int fd, dirlen = directory_size(filename);
1849 strbuf_reset(tmp);
1850 strbuf_add(tmp, filename, dirlen);
1851 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1852 fd = git_mkstemp_mode(tmp->buf, 0444);
1853 if (fd < 0 && dirlen && errno == ENOENT) {
1855 * Make sure the directory exists; note that the contents
1856 * of the buffer are undefined after mkstemp returns an
1857 * error, so we have to rewrite the whole buffer from
1858 * scratch.
1860 strbuf_reset(tmp);
1861 strbuf_add(tmp, filename, dirlen - 1);
1862 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1863 return -1;
1864 if (adjust_shared_perm(tmp->buf))
1865 return -1;
1867 /* Try again */
1868 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1869 fd = git_mkstemp_mode(tmp->buf, 0444);
1871 return fd;
1874 static int write_loose_object(const struct object_id *oid, char *hdr,
1875 int hdrlen, const void *buf, unsigned long len,
1876 time_t mtime)
1878 int fd, ret;
1879 unsigned char compressed[4096];
1880 git_zstream stream;
1881 git_hash_ctx c;
1882 struct object_id parano_oid;
1883 static struct strbuf tmp_file = STRBUF_INIT;
1884 static struct strbuf filename = STRBUF_INIT;
1886 loose_object_path(the_repository, &filename, oid);
1888 fd = create_tmpfile(&tmp_file, filename.buf);
1889 if (fd < 0) {
1890 if (errno == EACCES)
1891 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1892 else
1893 return error_errno(_("unable to create temporary file"));
1896 /* Set it up */
1897 git_deflate_init(&stream, zlib_compression_level);
1898 stream.next_out = compressed;
1899 stream.avail_out = sizeof(compressed);
1900 the_hash_algo->init_fn(&c);
1902 /* First header.. */
1903 stream.next_in = (unsigned char *)hdr;
1904 stream.avail_in = hdrlen;
1905 while (git_deflate(&stream, 0) == Z_OK)
1906 ; /* nothing */
1907 the_hash_algo->update_fn(&c, hdr, hdrlen);
1909 /* Then the data itself.. */
1910 stream.next_in = (void *)buf;
1911 stream.avail_in = len;
1912 do {
1913 unsigned char *in0 = stream.next_in;
1914 ret = git_deflate(&stream, Z_FINISH);
1915 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1916 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1917 die(_("unable to write loose object file"));
1918 stream.next_out = compressed;
1919 stream.avail_out = sizeof(compressed);
1920 } while (ret == Z_OK);
1922 if (ret != Z_STREAM_END)
1923 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1924 ret);
1925 ret = git_deflate_end_gently(&stream);
1926 if (ret != Z_OK)
1927 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1928 ret);
1929 the_hash_algo->final_oid_fn(&parano_oid, &c);
1930 if (!oideq(oid, &parano_oid))
1931 die(_("confused by unstable object source data for %s"),
1932 oid_to_hex(oid));
1934 close_loose_object(fd);
1936 if (mtime) {
1937 struct utimbuf utb;
1938 utb.actime = mtime;
1939 utb.modtime = mtime;
1940 if (utime(tmp_file.buf, &utb) < 0)
1941 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1944 return finalize_object_file(tmp_file.buf, filename.buf);
1947 static int freshen_loose_object(const struct object_id *oid)
1949 return check_and_freshen(oid, 1);
1952 static int freshen_packed_object(const struct object_id *oid)
1954 struct pack_entry e;
1955 if (!find_pack_entry(the_repository, oid, &e))
1956 return 0;
1957 if (e.p->freshened)
1958 return 1;
1959 if (!freshen_file(e.p->pack_name))
1960 return 0;
1961 e.p->freshened = 1;
1962 return 1;
1965 int write_object_file(const void *buf, unsigned long len, const char *type,
1966 struct object_id *oid)
1968 char hdr[MAX_HEADER_LEN];
1969 int hdrlen = sizeof(hdr);
1971 /* Normally if we have it in the pack then we do not bother writing
1972 * it out into .git/objects/??/?{38} file.
1974 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1975 &hdrlen);
1976 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1977 return 0;
1978 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1981 int hash_object_file_literally(const void *buf, unsigned long len,
1982 const char *type, struct object_id *oid,
1983 unsigned flags)
1985 char *header;
1986 int hdrlen, status = 0;
1988 /* type string, SP, %lu of the length plus NUL must fit this */
1989 hdrlen = strlen(type) + MAX_HEADER_LEN;
1990 header = xmalloc(hdrlen);
1991 write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1992 &hdrlen);
1994 if (!(flags & HASH_WRITE_OBJECT))
1995 goto cleanup;
1996 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1997 goto cleanup;
1998 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
2000 cleanup:
2001 free(header);
2002 return status;
2005 int force_object_loose(const struct object_id *oid, time_t mtime)
2007 void *buf;
2008 unsigned long len;
2009 enum object_type type;
2010 char hdr[MAX_HEADER_LEN];
2011 int hdrlen;
2012 int ret;
2014 if (has_loose_object(oid))
2015 return 0;
2016 buf = read_object(the_repository, oid, &type, &len);
2017 if (!buf)
2018 return error(_("cannot read object for %s"), oid_to_hex(oid));
2019 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
2020 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
2021 free(buf);
2023 return ret;
2026 int has_object(struct repository *r, const struct object_id *oid,
2027 unsigned flags)
2029 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2030 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2031 (quick ? OBJECT_INFO_QUICK : 0);
2033 if (!startup_info->have_repository)
2034 return 0;
2035 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2038 int repo_has_object_file_with_flags(struct repository *r,
2039 const struct object_id *oid, int flags)
2041 if (!startup_info->have_repository)
2042 return 0;
2043 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2046 int repo_has_object_file(struct repository *r,
2047 const struct object_id *oid)
2049 return repo_has_object_file_with_flags(r, oid, 0);
2052 static void check_tree(const void *buf, size_t size)
2054 struct tree_desc desc;
2055 struct name_entry entry;
2057 init_tree_desc(&desc, buf, size);
2058 while (tree_entry(&desc, &entry))
2059 /* do nothing
2060 * tree_entry() will die() on malformed entries */
2064 static void check_commit(const void *buf, size_t size)
2066 struct commit c;
2067 memset(&c, 0, sizeof(c));
2068 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2069 die(_("corrupt commit"));
2072 static void check_tag(const void *buf, size_t size)
2074 struct tag t;
2075 memset(&t, 0, sizeof(t));
2076 if (parse_tag_buffer(the_repository, &t, buf, size))
2077 die(_("corrupt tag"));
2080 static int index_mem(struct index_state *istate,
2081 struct object_id *oid, void *buf, size_t size,
2082 enum object_type type,
2083 const char *path, unsigned flags)
2085 int ret, re_allocated = 0;
2086 int write_object = flags & HASH_WRITE_OBJECT;
2088 if (!type)
2089 type = OBJ_BLOB;
2092 * Convert blobs to git internal format
2094 if ((type == OBJ_BLOB) && path) {
2095 struct strbuf nbuf = STRBUF_INIT;
2096 if (convert_to_git(istate, path, buf, size, &nbuf,
2097 get_conv_flags(flags))) {
2098 buf = strbuf_detach(&nbuf, &size);
2099 re_allocated = 1;
2102 if (flags & HASH_FORMAT_CHECK) {
2103 if (type == OBJ_TREE)
2104 check_tree(buf, size);
2105 if (type == OBJ_COMMIT)
2106 check_commit(buf, size);
2107 if (type == OBJ_TAG)
2108 check_tag(buf, size);
2111 if (write_object)
2112 ret = write_object_file(buf, size, type_name(type), oid);
2113 else
2114 ret = hash_object_file(the_hash_algo, buf, size,
2115 type_name(type), oid);
2116 if (re_allocated)
2117 free(buf);
2118 return ret;
2121 static int index_stream_convert_blob(struct index_state *istate,
2122 struct object_id *oid,
2123 int fd,
2124 const char *path,
2125 unsigned flags)
2127 int ret;
2128 const int write_object = flags & HASH_WRITE_OBJECT;
2129 struct strbuf sbuf = STRBUF_INIT;
2131 assert(path);
2132 assert(would_convert_to_git_filter_fd(istate, path));
2134 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2135 get_conv_flags(flags));
2137 if (write_object)
2138 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2139 oid);
2140 else
2141 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2142 type_name(OBJ_BLOB), oid);
2143 strbuf_release(&sbuf);
2144 return ret;
2147 static int index_pipe(struct index_state *istate, struct object_id *oid,
2148 int fd, enum object_type type,
2149 const char *path, unsigned flags)
2151 struct strbuf sbuf = STRBUF_INIT;
2152 int ret;
2154 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2155 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2156 else
2157 ret = -1;
2158 strbuf_release(&sbuf);
2159 return ret;
2162 #define SMALL_FILE_SIZE (32*1024)
2164 static int index_core(struct index_state *istate,
2165 struct object_id *oid, int fd, size_t size,
2166 enum object_type type, const char *path,
2167 unsigned flags)
2169 int ret;
2171 if (!size) {
2172 ret = index_mem(istate, oid, "", size, type, path, flags);
2173 } else if (size <= SMALL_FILE_SIZE) {
2174 char *buf = xmalloc(size);
2175 ssize_t read_result = read_in_full(fd, buf, size);
2176 if (read_result < 0)
2177 ret = error_errno(_("read error while indexing %s"),
2178 path ? path : "<unknown>");
2179 else if (read_result != size)
2180 ret = error(_("short read while indexing %s"),
2181 path ? path : "<unknown>");
2182 else
2183 ret = index_mem(istate, oid, buf, size, type, path, flags);
2184 free(buf);
2185 } else {
2186 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2187 ret = index_mem(istate, oid, buf, size, type, path, flags);
2188 munmap(buf, size);
2190 return ret;
2194 * This creates one packfile per large blob unless bulk-checkin
2195 * machinery is "plugged".
2197 * This also bypasses the usual "convert-to-git" dance, and that is on
2198 * purpose. We could write a streaming version of the converting
2199 * functions and insert that before feeding the data to fast-import
2200 * (or equivalent in-core API described above). However, that is
2201 * somewhat complicated, as we do not know the size of the filter
2202 * result, which we need to know beforehand when writing a git object.
2203 * Since the primary motivation for trying to stream from the working
2204 * tree file and to avoid mmaping it in core is to deal with large
2205 * binary blobs, they generally do not want to get any conversion, and
2206 * callers should avoid this code path when filters are requested.
2208 static int index_stream(struct object_id *oid, int fd, size_t size,
2209 enum object_type type, const char *path,
2210 unsigned flags)
2212 return index_bulk_checkin(oid, fd, size, type, path, flags);
2215 int index_fd(struct index_state *istate, struct object_id *oid,
2216 int fd, struct stat *st,
2217 enum object_type type, const char *path, unsigned flags)
2219 int ret;
2222 * Call xsize_t() only when needed to avoid potentially unnecessary
2223 * die() for large files.
2225 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2226 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2227 else if (!S_ISREG(st->st_mode))
2228 ret = index_pipe(istate, oid, fd, type, path, flags);
2229 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2230 (path && would_convert_to_git(istate, path)))
2231 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2232 type, path, flags);
2233 else
2234 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2235 flags);
2236 close(fd);
2237 return ret;
2240 int index_path(struct index_state *istate, struct object_id *oid,
2241 const char *path, struct stat *st, unsigned flags)
2243 int fd;
2244 struct strbuf sb = STRBUF_INIT;
2245 int rc = 0;
2247 switch (st->st_mode & S_IFMT) {
2248 case S_IFREG:
2249 fd = open(path, O_RDONLY);
2250 if (fd < 0)
2251 return error_errno("open(\"%s\")", path);
2252 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2253 return error(_("%s: failed to insert into database"),
2254 path);
2255 break;
2256 case S_IFLNK:
2257 if (strbuf_readlink(&sb, path, st->st_size))
2258 return error_errno("readlink(\"%s\")", path);
2259 if (!(flags & HASH_WRITE_OBJECT))
2260 hash_object_file(the_hash_algo, sb.buf, sb.len,
2261 blob_type, oid);
2262 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2263 rc = error(_("%s: failed to insert into database"), path);
2264 strbuf_release(&sb);
2265 break;
2266 case S_IFDIR:
2267 return resolve_gitlink_ref(path, "HEAD", oid);
2268 default:
2269 return error(_("%s: unsupported file type"), path);
2271 return rc;
2274 int read_pack_header(int fd, struct pack_header *header)
2276 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2277 /* "eof before pack header was fully read" */
2278 return PH_ERROR_EOF;
2280 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2281 /* "protocol error (pack signature mismatch detected)" */
2282 return PH_ERROR_PACK_SIGNATURE;
2283 if (!pack_version_ok(header->hdr_version))
2284 /* "protocol error (pack version unsupported)" */
2285 return PH_ERROR_PROTOCOL;
2286 return 0;
2289 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2291 enum object_type type = oid_object_info(the_repository, oid, NULL);
2292 if (type < 0)
2293 die(_("%s is not a valid object"), oid_to_hex(oid));
2294 if (type != expect)
2295 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2296 type_name(expect));
2299 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2300 struct strbuf *path,
2301 each_loose_object_fn obj_cb,
2302 each_loose_cruft_fn cruft_cb,
2303 each_loose_subdir_fn subdir_cb,
2304 void *data)
2306 size_t origlen, baselen;
2307 DIR *dir;
2308 struct dirent *de;
2309 int r = 0;
2310 struct object_id oid;
2312 if (subdir_nr > 0xff)
2313 BUG("invalid loose object subdirectory: %x", subdir_nr);
2315 origlen = path->len;
2316 strbuf_complete(path, '/');
2317 strbuf_addf(path, "%02x", subdir_nr);
2319 dir = opendir(path->buf);
2320 if (!dir) {
2321 if (errno != ENOENT)
2322 r = error_errno(_("unable to open %s"), path->buf);
2323 strbuf_setlen(path, origlen);
2324 return r;
2327 oid.hash[0] = subdir_nr;
2328 strbuf_addch(path, '/');
2329 baselen = path->len;
2331 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2332 size_t namelen;
2334 namelen = strlen(de->d_name);
2335 strbuf_setlen(path, baselen);
2336 strbuf_add(path, de->d_name, namelen);
2337 if (namelen == the_hash_algo->hexsz - 2 &&
2338 !hex_to_bytes(oid.hash + 1, de->d_name,
2339 the_hash_algo->rawsz - 1)) {
2340 oid_set_algo(&oid, the_hash_algo);
2341 if (obj_cb) {
2342 r = obj_cb(&oid, path->buf, data);
2343 if (r)
2344 break;
2346 continue;
2349 if (cruft_cb) {
2350 r = cruft_cb(de->d_name, path->buf, data);
2351 if (r)
2352 break;
2355 closedir(dir);
2357 strbuf_setlen(path, baselen - 1);
2358 if (!r && subdir_cb)
2359 r = subdir_cb(subdir_nr, path->buf, data);
2361 strbuf_setlen(path, origlen);
2363 return r;
2366 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2367 each_loose_object_fn obj_cb,
2368 each_loose_cruft_fn cruft_cb,
2369 each_loose_subdir_fn subdir_cb,
2370 void *data)
2372 int r = 0;
2373 int i;
2375 for (i = 0; i < 256; i++) {
2376 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2377 subdir_cb, data);
2378 if (r)
2379 break;
2382 return r;
2385 int for_each_loose_file_in_objdir(const char *path,
2386 each_loose_object_fn obj_cb,
2387 each_loose_cruft_fn cruft_cb,
2388 each_loose_subdir_fn subdir_cb,
2389 void *data)
2391 struct strbuf buf = STRBUF_INIT;
2392 int r;
2394 strbuf_addstr(&buf, path);
2395 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2396 subdir_cb, data);
2397 strbuf_release(&buf);
2399 return r;
2402 int for_each_loose_object(each_loose_object_fn cb, void *data,
2403 enum for_each_object_flags flags)
2405 struct object_directory *odb;
2407 prepare_alt_odb(the_repository);
2408 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2409 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2410 NULL, data);
2411 if (r)
2412 return r;
2414 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2415 break;
2418 return 0;
2421 static int append_loose_object(const struct object_id *oid, const char *path,
2422 void *data)
2424 oidtree_insert(data, oid);
2425 return 0;
2428 struct oidtree *odb_loose_cache(struct object_directory *odb,
2429 const struct object_id *oid)
2431 int subdir_nr = oid->hash[0];
2432 struct strbuf buf = STRBUF_INIT;
2433 size_t word_bits = bitsizeof(odb->loose_objects_subdir_seen[0]);
2434 size_t word_index = subdir_nr / word_bits;
2435 size_t mask = 1u << (subdir_nr % word_bits);
2436 uint32_t *bitmap;
2438 if (subdir_nr < 0 ||
2439 subdir_nr >= bitsizeof(odb->loose_objects_subdir_seen))
2440 BUG("subdir_nr out of range");
2442 bitmap = &odb->loose_objects_subdir_seen[word_index];
2443 if (*bitmap & mask)
2444 return odb->loose_objects_cache;
2445 if (!odb->loose_objects_cache) {
2446 ALLOC_ARRAY(odb->loose_objects_cache, 1);
2447 oidtree_init(odb->loose_objects_cache);
2449 strbuf_addstr(&buf, odb->path);
2450 for_each_file_in_obj_subdir(subdir_nr, &buf,
2451 append_loose_object,
2452 NULL, NULL,
2453 odb->loose_objects_cache);
2454 *bitmap |= mask;
2455 strbuf_release(&buf);
2456 return odb->loose_objects_cache;
2459 void odb_clear_loose_cache(struct object_directory *odb)
2461 oidtree_clear(odb->loose_objects_cache);
2462 FREE_AND_NULL(odb->loose_objects_cache);
2463 memset(&odb->loose_objects_subdir_seen, 0,
2464 sizeof(odb->loose_objects_subdir_seen));
2467 static int check_stream_oid(git_zstream *stream,
2468 const char *hdr,
2469 unsigned long size,
2470 const char *path,
2471 const struct object_id *expected_oid)
2473 git_hash_ctx c;
2474 struct object_id real_oid;
2475 unsigned char buf[4096];
2476 unsigned long total_read;
2477 int status = Z_OK;
2479 the_hash_algo->init_fn(&c);
2480 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2483 * We already read some bytes into hdr, but the ones up to the NUL
2484 * do not count against the object's content size.
2486 total_read = stream->total_out - strlen(hdr) - 1;
2489 * This size comparison must be "<=" to read the final zlib packets;
2490 * see the comment in unpack_loose_rest for details.
2492 while (total_read <= size &&
2493 (status == Z_OK ||
2494 (status == Z_BUF_ERROR && !stream->avail_out))) {
2495 stream->next_out = buf;
2496 stream->avail_out = sizeof(buf);
2497 if (size - total_read < stream->avail_out)
2498 stream->avail_out = size - total_read;
2499 status = git_inflate(stream, Z_FINISH);
2500 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2501 total_read += stream->next_out - buf;
2503 git_inflate_end(stream);
2505 if (status != Z_STREAM_END) {
2506 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2507 return -1;
2509 if (stream->avail_in) {
2510 error(_("garbage at end of loose object '%s'"),
2511 oid_to_hex(expected_oid));
2512 return -1;
2515 the_hash_algo->final_oid_fn(&real_oid, &c);
2516 if (!oideq(expected_oid, &real_oid)) {
2517 error(_("hash mismatch for %s (expected %s)"), path,
2518 oid_to_hex(expected_oid));
2519 return -1;
2522 return 0;
2525 int read_loose_object(const char *path,
2526 const struct object_id *expected_oid,
2527 enum object_type *type,
2528 unsigned long *size,
2529 void **contents)
2531 int ret = -1;
2532 void *map = NULL;
2533 unsigned long mapsize;
2534 git_zstream stream;
2535 char hdr[MAX_HEADER_LEN];
2537 *contents = NULL;
2539 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2540 if (!map) {
2541 error_errno(_("unable to mmap %s"), path);
2542 goto out;
2545 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2546 error(_("unable to unpack header of %s"), path);
2547 goto out;
2550 *type = parse_loose_header(hdr, size);
2551 if (*type < 0) {
2552 error(_("unable to parse header of %s"), path);
2553 git_inflate_end(&stream);
2554 goto out;
2557 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2558 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2559 goto out;
2560 } else {
2561 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2562 if (!*contents) {
2563 error(_("unable to unpack contents of %s"), path);
2564 git_inflate_end(&stream);
2565 goto out;
2567 if (check_object_signature(the_repository, expected_oid,
2568 *contents, *size,
2569 type_name(*type))) {
2570 error(_("hash mismatch for %s (expected %s)"), path,
2571 oid_to_hex(expected_oid));
2572 free(*contents);
2573 goto out;
2577 ret = 0; /* everything checks out */
2579 out:
2580 if (map)
2581 munmap(map, mapsize);
2582 return ret;