Sync with 2.33.8
[git/debian.git] / object-file.c
blobc3d866a287e03ad0ae5de97ab6a352887c92e233
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,
1020 struct object_id *real_oidp)
1022 struct object_id tmp;
1023 struct object_id *real_oid = real_oidp ? real_oidp : &tmp;
1024 enum object_type obj_type;
1025 struct git_istream *st;
1026 git_hash_ctx c;
1027 char hdr[MAX_HEADER_LEN];
1028 int hdrlen;
1030 if (map) {
1031 hash_object_file(r->hash_algo, map, size, type, real_oid);
1032 return !oideq(oid, real_oid) ? -1 : 0;
1035 st = open_istream(r, oid, &obj_type, &size, NULL);
1036 if (!st)
1037 return -1;
1039 /* Generate the header */
1040 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1042 /* Sha1.. */
1043 r->hash_algo->init_fn(&c);
1044 r->hash_algo->update_fn(&c, hdr, hdrlen);
1045 for (;;) {
1046 char buf[1024 * 16];
1047 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1049 if (readlen < 0) {
1050 close_istream(st);
1051 return -1;
1053 if (!readlen)
1054 break;
1055 r->hash_algo->update_fn(&c, buf, readlen);
1057 r->hash_algo->final_oid_fn(real_oid, &c);
1058 close_istream(st);
1059 return !oideq(oid, real_oid) ? -1 : 0;
1062 int git_open_cloexec(const char *name, int flags)
1064 int fd;
1065 static int o_cloexec = O_CLOEXEC;
1067 fd = open(name, flags | o_cloexec);
1068 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1069 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1070 o_cloexec &= ~O_CLOEXEC;
1071 fd = open(name, flags | o_cloexec);
1074 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1076 static int fd_cloexec = FD_CLOEXEC;
1078 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1079 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1080 int flags = fcntl(fd, F_GETFD);
1081 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1082 fd_cloexec = 0;
1085 #endif
1086 return fd;
1090 * Find "oid" as a loose object in the local repository or in an alternate.
1091 * Returns 0 on success, negative on failure.
1093 * The "path" out-parameter will give the path of the object we found (if any).
1094 * Note that it may point to static storage and is only valid until another
1095 * call to stat_loose_object().
1097 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1098 struct stat *st, const char **path)
1100 struct object_directory *odb;
1101 static struct strbuf buf = STRBUF_INIT;
1103 prepare_alt_odb(r);
1104 for (odb = r->objects->odb; odb; odb = odb->next) {
1105 *path = odb_loose_path(odb, &buf, oid);
1106 if (!lstat(*path, st))
1107 return 0;
1110 return -1;
1114 * Like stat_loose_object(), but actually open the object and return the
1115 * descriptor. See the caveats on the "path" parameter above.
1117 static int open_loose_object(struct repository *r,
1118 const struct object_id *oid, const char **path)
1120 int fd;
1121 struct object_directory *odb;
1122 int most_interesting_errno = ENOENT;
1123 static struct strbuf buf = STRBUF_INIT;
1125 prepare_alt_odb(r);
1126 for (odb = r->objects->odb; odb; odb = odb->next) {
1127 *path = odb_loose_path(odb, &buf, oid);
1128 fd = git_open(*path);
1129 if (fd >= 0)
1130 return fd;
1132 if (most_interesting_errno == ENOENT)
1133 most_interesting_errno = errno;
1135 errno = most_interesting_errno;
1136 return -1;
1139 static int quick_has_loose(struct repository *r,
1140 const struct object_id *oid)
1142 struct object_directory *odb;
1144 prepare_alt_odb(r);
1145 for (odb = r->objects->odb; odb; odb = odb->next) {
1146 if (oidtree_contains(odb_loose_cache(odb, oid), oid))
1147 return 1;
1149 return 0;
1153 * Map the loose object at "path" if it is not NULL, or the path found by
1154 * searching for a loose object named "oid".
1156 static void *map_loose_object_1(struct repository *r, const char *path,
1157 const struct object_id *oid, unsigned long *size)
1159 void *map;
1160 int fd;
1162 if (path)
1163 fd = git_open(path);
1164 else
1165 fd = open_loose_object(r, oid, &path);
1166 map = NULL;
1167 if (fd >= 0) {
1168 struct stat st;
1170 if (!fstat(fd, &st)) {
1171 *size = xsize_t(st.st_size);
1172 if (!*size) {
1173 /* mmap() is forbidden on empty files */
1174 error(_("object file %s is empty"), path);
1175 close(fd);
1176 return NULL;
1178 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1180 close(fd);
1182 return map;
1185 void *map_loose_object(struct repository *r,
1186 const struct object_id *oid,
1187 unsigned long *size)
1189 return map_loose_object_1(r, NULL, oid, size);
1192 enum unpack_loose_header_result unpack_loose_header(git_zstream *stream,
1193 unsigned char *map,
1194 unsigned long mapsize,
1195 void *buffer,
1196 unsigned long bufsiz,
1197 struct strbuf *header)
1199 int status;
1201 /* Get the data stream */
1202 memset(stream, 0, sizeof(*stream));
1203 stream->next_in = map;
1204 stream->avail_in = mapsize;
1205 stream->next_out = buffer;
1206 stream->avail_out = bufsiz;
1208 git_inflate_init(stream);
1209 obj_read_unlock();
1210 status = git_inflate(stream, 0);
1211 obj_read_lock();
1212 if (status < Z_OK)
1213 return ULHR_BAD;
1216 * Check if entire header is unpacked in the first iteration.
1218 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1219 return ULHR_OK;
1222 * We have a header longer than MAX_HEADER_LEN. The "header"
1223 * here is only non-NULL when we run "cat-file
1224 * --allow-unknown-type".
1226 if (!header)
1227 return ULHR_TOO_LONG;
1230 * buffer[0..bufsiz] was not large enough. Copy the partial
1231 * result out to header, and then append the result of further
1232 * reading the stream.
1234 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1235 stream->next_out = buffer;
1236 stream->avail_out = bufsiz;
1238 do {
1239 obj_read_unlock();
1240 status = git_inflate(stream, 0);
1241 obj_read_lock();
1242 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1243 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1244 return 0;
1245 stream->next_out = buffer;
1246 stream->avail_out = bufsiz;
1247 } while (status != Z_STREAM_END);
1248 return ULHR_TOO_LONG;
1251 static void *unpack_loose_rest(git_zstream *stream,
1252 void *buffer, unsigned long size,
1253 const struct object_id *oid)
1255 int bytes = strlen(buffer) + 1;
1256 unsigned char *buf = xmallocz(size);
1257 unsigned long n;
1258 int status = Z_OK;
1260 n = stream->total_out - bytes;
1261 if (n > size)
1262 n = size;
1263 memcpy(buf, (char *) buffer + bytes, n);
1264 bytes = n;
1265 if (bytes <= size) {
1267 * The above condition must be (bytes <= size), not
1268 * (bytes < size). In other words, even though we
1269 * expect no more output and set avail_out to zero,
1270 * the input zlib stream may have bytes that express
1271 * "this concludes the stream", and we *do* want to
1272 * eat that input.
1274 * Otherwise we would not be able to test that we
1275 * consumed all the input to reach the expected size;
1276 * we also want to check that zlib tells us that all
1277 * went well with status == Z_STREAM_END at the end.
1279 stream->next_out = buf + bytes;
1280 stream->avail_out = size - bytes;
1281 while (status == Z_OK) {
1282 obj_read_unlock();
1283 status = git_inflate(stream, Z_FINISH);
1284 obj_read_lock();
1287 if (status == Z_STREAM_END && !stream->avail_in) {
1288 git_inflate_end(stream);
1289 return buf;
1292 if (status < 0)
1293 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1294 else if (stream->avail_in)
1295 error(_("garbage at end of loose object '%s'"),
1296 oid_to_hex(oid));
1297 free(buf);
1298 return NULL;
1302 * We used to just use "sscanf()", but that's actually way
1303 * too permissive for what we want to check. So do an anal
1304 * object header parse by hand.
1306 int parse_loose_header(const char *hdr, struct object_info *oi)
1308 const char *type_buf = hdr;
1309 unsigned long size;
1310 int type, type_len = 0;
1313 * The type can be of any size but is followed by
1314 * a space.
1316 for (;;) {
1317 char c = *hdr++;
1318 if (!c)
1319 return -1;
1320 if (c == ' ')
1321 break;
1322 type_len++;
1325 type = type_from_string_gently(type_buf, type_len, 1);
1326 if (oi->type_name)
1327 strbuf_add(oi->type_name, type_buf, type_len);
1328 if (oi->typep)
1329 *oi->typep = type;
1332 * The length must follow immediately, and be in canonical
1333 * decimal format (ie "010" is not valid).
1335 size = *hdr++ - '0';
1336 if (size > 9)
1337 return -1;
1338 if (size) {
1339 for (;;) {
1340 unsigned long c = *hdr - '0';
1341 if (c > 9)
1342 break;
1343 hdr++;
1344 size = size * 10 + c;
1348 if (oi->sizep)
1349 *oi->sizep = size;
1352 * The length must be followed by a zero byte
1354 if (*hdr)
1355 return -1;
1358 * The format is valid, but the type may still be bogus. The
1359 * Caller needs to check its oi->typep.
1361 return 0;
1364 static int loose_object_info(struct repository *r,
1365 const struct object_id *oid,
1366 struct object_info *oi, int flags)
1368 int status = 0;
1369 unsigned long mapsize;
1370 void *map;
1371 git_zstream stream;
1372 char hdr[MAX_HEADER_LEN];
1373 struct strbuf hdrbuf = STRBUF_INIT;
1374 unsigned long size_scratch;
1375 enum object_type type_scratch;
1376 int allow_unknown = flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
1378 if (oi->delta_base_oid)
1379 oidclr(oi->delta_base_oid);
1382 * If we don't care about type or size, then we don't
1383 * need to look inside the object at all. Note that we
1384 * do not optimize out the stat call, even if the
1385 * caller doesn't care about the disk-size, since our
1386 * return value implicitly indicates whether the
1387 * object even exists.
1389 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1390 const char *path;
1391 struct stat st;
1392 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1393 return quick_has_loose(r, oid) ? 0 : -1;
1394 if (stat_loose_object(r, oid, &st, &path) < 0)
1395 return -1;
1396 if (oi->disk_sizep)
1397 *oi->disk_sizep = st.st_size;
1398 return 0;
1401 map = map_loose_object(r, oid, &mapsize);
1402 if (!map)
1403 return -1;
1405 if (!oi->sizep)
1406 oi->sizep = &size_scratch;
1407 if (!oi->typep)
1408 oi->typep = &type_scratch;
1410 if (oi->disk_sizep)
1411 *oi->disk_sizep = mapsize;
1413 switch (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
1414 allow_unknown ? &hdrbuf : NULL)) {
1415 case ULHR_OK:
1416 if (parse_loose_header(hdrbuf.len ? hdrbuf.buf : hdr, oi) < 0)
1417 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1418 else if (!allow_unknown && *oi->typep < 0)
1419 die(_("invalid object type"));
1421 if (!oi->contentp)
1422 break;
1423 *oi->contentp = unpack_loose_rest(&stream, hdr, *oi->sizep, oid);
1424 if (*oi->contentp)
1425 goto cleanup;
1427 status = -1;
1428 break;
1429 case ULHR_BAD:
1430 status = error(_("unable to unpack %s header"),
1431 oid_to_hex(oid));
1432 break;
1433 case ULHR_TOO_LONG:
1434 status = error(_("header for %s too long, exceeds %d bytes"),
1435 oid_to_hex(oid), MAX_HEADER_LEN);
1436 break;
1439 git_inflate_end(&stream);
1440 cleanup:
1441 munmap(map, mapsize);
1442 if (oi->sizep == &size_scratch)
1443 oi->sizep = NULL;
1444 strbuf_release(&hdrbuf);
1445 if (oi->typep == &type_scratch)
1446 oi->typep = NULL;
1447 oi->whence = OI_LOOSE;
1448 return status;
1451 int obj_read_use_lock = 0;
1452 pthread_mutex_t obj_read_mutex;
1454 void enable_obj_read_lock(void)
1456 if (obj_read_use_lock)
1457 return;
1459 obj_read_use_lock = 1;
1460 init_recursive_mutex(&obj_read_mutex);
1463 void disable_obj_read_lock(void)
1465 if (!obj_read_use_lock)
1466 return;
1468 obj_read_use_lock = 0;
1469 pthread_mutex_destroy(&obj_read_mutex);
1472 int fetch_if_missing = 1;
1474 static int do_oid_object_info_extended(struct repository *r,
1475 const struct object_id *oid,
1476 struct object_info *oi, unsigned flags)
1478 static struct object_info blank_oi = OBJECT_INFO_INIT;
1479 struct cached_object *co;
1480 struct pack_entry e;
1481 int rtype;
1482 const struct object_id *real = oid;
1483 int already_retried = 0;
1486 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1487 real = lookup_replace_object(r, oid);
1489 if (is_null_oid(real))
1490 return -1;
1492 if (!oi)
1493 oi = &blank_oi;
1495 co = find_cached_object(real);
1496 if (co) {
1497 if (oi->typep)
1498 *(oi->typep) = co->type;
1499 if (oi->sizep)
1500 *(oi->sizep) = co->size;
1501 if (oi->disk_sizep)
1502 *(oi->disk_sizep) = 0;
1503 if (oi->delta_base_oid)
1504 oidclr(oi->delta_base_oid);
1505 if (oi->type_name)
1506 strbuf_addstr(oi->type_name, type_name(co->type));
1507 if (oi->contentp)
1508 *oi->contentp = xmemdupz(co->buf, co->size);
1509 oi->whence = OI_CACHED;
1510 return 0;
1513 while (1) {
1514 if (find_pack_entry(r, real, &e))
1515 break;
1517 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1518 return -1;
1520 /* Most likely it's a loose object. */
1521 if (!loose_object_info(r, real, oi, flags))
1522 return 0;
1524 /* Not a loose object; someone else may have just packed it. */
1525 if (!(flags & OBJECT_INFO_QUICK)) {
1526 reprepare_packed_git(r);
1527 if (find_pack_entry(r, real, &e))
1528 break;
1532 * If r is the_repository, this might be an attempt at
1533 * accessing a submodule object as if it were in the_repository
1534 * (having called add_submodule_odb() on that submodule's ODB).
1535 * If any such ODBs exist, register them and try again.
1537 if (r == the_repository &&
1538 register_all_submodule_odb_as_alternates())
1539 /* We added some alternates; retry */
1540 continue;
1542 /* Check if it is a missing object */
1543 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1544 !already_retried &&
1545 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1547 * TODO Investigate checking promisor_remote_get_direct()
1548 * TODO return value and stopping on error here.
1550 promisor_remote_get_direct(r, real, 1);
1551 already_retried = 1;
1552 continue;
1555 return -1;
1558 if (oi == &blank_oi)
1560 * We know that the caller doesn't actually need the
1561 * information below, so return early.
1563 return 0;
1564 rtype = packed_object_info(r, e.p, e.offset, oi);
1565 if (rtype < 0) {
1566 mark_bad_packed_object(e.p, real);
1567 return do_oid_object_info_extended(r, real, oi, 0);
1568 } else if (oi->whence == OI_PACKED) {
1569 oi->u.packed.offset = e.offset;
1570 oi->u.packed.pack = e.p;
1571 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1572 rtype == OBJ_OFS_DELTA);
1575 return 0;
1578 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1579 struct object_info *oi, unsigned flags)
1581 int ret;
1582 obj_read_lock();
1583 ret = do_oid_object_info_extended(r, oid, oi, flags);
1584 obj_read_unlock();
1585 return ret;
1589 /* returns enum object_type or negative */
1590 int oid_object_info(struct repository *r,
1591 const struct object_id *oid,
1592 unsigned long *sizep)
1594 enum object_type type;
1595 struct object_info oi = OBJECT_INFO_INIT;
1597 oi.typep = &type;
1598 oi.sizep = sizep;
1599 if (oid_object_info_extended(r, oid, &oi,
1600 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1601 return -1;
1602 return type;
1605 static void *read_object(struct repository *r,
1606 const struct object_id *oid, enum object_type *type,
1607 unsigned long *size)
1609 struct object_info oi = OBJECT_INFO_INIT;
1610 void *content;
1611 oi.typep = type;
1612 oi.sizep = size;
1613 oi.contentp = &content;
1615 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1616 return NULL;
1617 return content;
1620 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1621 struct object_id *oid)
1623 struct cached_object *co;
1625 hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1626 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1627 find_cached_object(oid))
1628 return 0;
1629 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1630 co = &cached_objects[cached_object_nr++];
1631 co->size = len;
1632 co->type = type;
1633 co->buf = xmalloc(len);
1634 memcpy(co->buf, buf, len);
1635 oidcpy(&co->oid, oid);
1636 return 0;
1640 * This function dies on corrupt objects; the callers who want to
1641 * deal with them should arrange to call read_object() and give error
1642 * messages themselves.
1644 void *read_object_file_extended(struct repository *r,
1645 const struct object_id *oid,
1646 enum object_type *type,
1647 unsigned long *size,
1648 int lookup_replace)
1650 void *data;
1651 const struct packed_git *p;
1652 const char *path;
1653 struct stat st;
1654 const struct object_id *repl = lookup_replace ?
1655 lookup_replace_object(r, oid) : oid;
1657 errno = 0;
1658 data = read_object(r, repl, type, size);
1659 if (data)
1660 return data;
1662 obj_read_lock();
1663 if (errno && errno != ENOENT)
1664 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1666 /* die if we replaced an object with one that does not exist */
1667 if (repl != oid)
1668 die(_("replacement %s not found for %s"),
1669 oid_to_hex(repl), oid_to_hex(oid));
1671 if (!stat_loose_object(r, repl, &st, &path))
1672 die(_("loose object %s (stored in %s) is corrupt"),
1673 oid_to_hex(repl), path);
1675 if ((p = has_packed_and_bad(r, repl)) != NULL)
1676 die(_("packed object %s (stored in %s) is corrupt"),
1677 oid_to_hex(repl), p->pack_name);
1678 obj_read_unlock();
1680 return NULL;
1683 void *read_object_with_reference(struct repository *r,
1684 const struct object_id *oid,
1685 const char *required_type_name,
1686 unsigned long *size,
1687 struct object_id *actual_oid_return)
1689 enum object_type type, required_type;
1690 void *buffer;
1691 unsigned long isize;
1692 struct object_id actual_oid;
1694 required_type = type_from_string(required_type_name);
1695 oidcpy(&actual_oid, oid);
1696 while (1) {
1697 int ref_length = -1;
1698 const char *ref_type = NULL;
1700 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1701 if (!buffer)
1702 return NULL;
1703 if (type == required_type) {
1704 *size = isize;
1705 if (actual_oid_return)
1706 oidcpy(actual_oid_return, &actual_oid);
1707 return buffer;
1709 /* Handle references */
1710 else if (type == OBJ_COMMIT)
1711 ref_type = "tree ";
1712 else if (type == OBJ_TAG)
1713 ref_type = "object ";
1714 else {
1715 free(buffer);
1716 return NULL;
1718 ref_length = strlen(ref_type);
1720 if (ref_length + the_hash_algo->hexsz > isize ||
1721 memcmp(buffer, ref_type, ref_length) ||
1722 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1723 free(buffer);
1724 return NULL;
1726 free(buffer);
1727 /* Now we have the ID of the referred-to object in
1728 * actual_oid. Check again. */
1732 static void write_object_file_prepare(const struct git_hash_algo *algo,
1733 const void *buf, unsigned long len,
1734 const char *type, struct object_id *oid,
1735 char *hdr, int *hdrlen)
1737 git_hash_ctx c;
1739 /* Generate the header */
1740 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1742 /* Sha1.. */
1743 algo->init_fn(&c);
1744 algo->update_fn(&c, hdr, *hdrlen);
1745 algo->update_fn(&c, buf, len);
1746 algo->final_oid_fn(oid, &c);
1750 * Move the just written object into its final resting place.
1752 int finalize_object_file(const char *tmpfile, const char *filename)
1754 int ret = 0;
1756 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1757 goto try_rename;
1758 else if (link(tmpfile, filename))
1759 ret = errno;
1762 * Coda hack - coda doesn't like cross-directory links,
1763 * so we fall back to a rename, which will mean that it
1764 * won't be able to check collisions, but that's not a
1765 * big deal.
1767 * The same holds for FAT formatted media.
1769 * When this succeeds, we just return. We have nothing
1770 * left to unlink.
1772 if (ret && ret != EEXIST) {
1773 try_rename:
1774 if (!rename(tmpfile, filename))
1775 goto out;
1776 ret = errno;
1778 unlink_or_warn(tmpfile);
1779 if (ret) {
1780 if (ret != EEXIST) {
1781 return error_errno(_("unable to write file %s"), filename);
1783 /* FIXME!!! Collision check here ? */
1786 out:
1787 if (adjust_shared_perm(filename))
1788 return error(_("unable to set permission to '%s'"), filename);
1789 return 0;
1792 static int write_buffer(int fd, const void *buf, size_t len)
1794 if (write_in_full(fd, buf, len) < 0)
1795 return error_errno(_("file write error"));
1796 return 0;
1799 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1800 unsigned long len, const char *type,
1801 struct object_id *oid)
1803 char hdr[MAX_HEADER_LEN];
1804 int hdrlen = sizeof(hdr);
1805 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1806 return 0;
1809 /* Finalize a file on disk, and close it. */
1810 static void close_loose_object(int fd)
1812 if (fsync_object_files)
1813 fsync_or_die(fd, "loose object file");
1814 if (close(fd) != 0)
1815 die_errno(_("error when closing loose object file"));
1818 /* Size of directory component, including the ending '/' */
1819 static inline int directory_size(const char *filename)
1821 const char *s = strrchr(filename, '/');
1822 if (!s)
1823 return 0;
1824 return s - filename + 1;
1828 * This creates a temporary file in the same directory as the final
1829 * 'filename'
1831 * We want to avoid cross-directory filename renames, because those
1832 * can have problems on various filesystems (FAT, NFS, Coda).
1834 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1836 int fd, dirlen = directory_size(filename);
1838 strbuf_reset(tmp);
1839 strbuf_add(tmp, filename, dirlen);
1840 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1841 fd = git_mkstemp_mode(tmp->buf, 0444);
1842 if (fd < 0 && dirlen && errno == ENOENT) {
1844 * Make sure the directory exists; note that the contents
1845 * of the buffer are undefined after mkstemp returns an
1846 * error, so we have to rewrite the whole buffer from
1847 * scratch.
1849 strbuf_reset(tmp);
1850 strbuf_add(tmp, filename, dirlen - 1);
1851 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1852 return -1;
1853 if (adjust_shared_perm(tmp->buf))
1854 return -1;
1856 /* Try again */
1857 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1858 fd = git_mkstemp_mode(tmp->buf, 0444);
1860 return fd;
1863 static int write_loose_object(const struct object_id *oid, char *hdr,
1864 int hdrlen, const void *buf, unsigned long len,
1865 time_t mtime, unsigned flags)
1867 int fd, ret;
1868 unsigned char compressed[4096];
1869 git_zstream stream;
1870 git_hash_ctx c;
1871 struct object_id parano_oid;
1872 static struct strbuf tmp_file = STRBUF_INIT;
1873 static struct strbuf filename = STRBUF_INIT;
1875 loose_object_path(the_repository, &filename, oid);
1877 fd = create_tmpfile(&tmp_file, filename.buf);
1878 if (fd < 0) {
1879 if (flags & HASH_SILENT)
1880 return -1;
1881 else if (errno == EACCES)
1882 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1883 else
1884 return error_errno(_("unable to create temporary file"));
1887 /* Set it up */
1888 git_deflate_init(&stream, zlib_compression_level);
1889 stream.next_out = compressed;
1890 stream.avail_out = sizeof(compressed);
1891 the_hash_algo->init_fn(&c);
1893 /* First header.. */
1894 stream.next_in = (unsigned char *)hdr;
1895 stream.avail_in = hdrlen;
1896 while (git_deflate(&stream, 0) == Z_OK)
1897 ; /* nothing */
1898 the_hash_algo->update_fn(&c, hdr, hdrlen);
1900 /* Then the data itself.. */
1901 stream.next_in = (void *)buf;
1902 stream.avail_in = len;
1903 do {
1904 unsigned char *in0 = stream.next_in;
1905 ret = git_deflate(&stream, Z_FINISH);
1906 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1907 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1908 die(_("unable to write loose object file"));
1909 stream.next_out = compressed;
1910 stream.avail_out = sizeof(compressed);
1911 } while (ret == Z_OK);
1913 if (ret != Z_STREAM_END)
1914 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1915 ret);
1916 ret = git_deflate_end_gently(&stream);
1917 if (ret != Z_OK)
1918 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1919 ret);
1920 the_hash_algo->final_oid_fn(&parano_oid, &c);
1921 if (!oideq(oid, &parano_oid))
1922 die(_("confused by unstable object source data for %s"),
1923 oid_to_hex(oid));
1925 close_loose_object(fd);
1927 if (mtime) {
1928 struct utimbuf utb;
1929 utb.actime = mtime;
1930 utb.modtime = mtime;
1931 if (utime(tmp_file.buf, &utb) < 0 &&
1932 !(flags & HASH_SILENT))
1933 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1936 return finalize_object_file(tmp_file.buf, filename.buf);
1939 static int freshen_loose_object(const struct object_id *oid)
1941 return check_and_freshen(oid, 1);
1944 static int freshen_packed_object(const struct object_id *oid)
1946 struct pack_entry e;
1947 if (!find_pack_entry(the_repository, oid, &e))
1948 return 0;
1949 if (e.p->freshened)
1950 return 1;
1951 if (!freshen_file(e.p->pack_name))
1952 return 0;
1953 e.p->freshened = 1;
1954 return 1;
1957 int write_object_file_flags(const void *buf, unsigned long len,
1958 const char *type, struct object_id *oid,
1959 unsigned flags)
1961 char hdr[MAX_HEADER_LEN];
1962 int hdrlen = sizeof(hdr);
1964 /* Normally if we have it in the pack then we do not bother writing
1965 * it out into .git/objects/??/?{38} file.
1967 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1968 &hdrlen);
1969 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1970 return 0;
1971 return write_loose_object(oid, hdr, hdrlen, buf, len, 0, flags);
1974 int hash_object_file_literally(const void *buf, unsigned long len,
1975 const char *type, struct object_id *oid,
1976 unsigned flags)
1978 char *header;
1979 int hdrlen, status = 0;
1981 /* type string, SP, %lu of the length plus NUL must fit this */
1982 hdrlen = strlen(type) + MAX_HEADER_LEN;
1983 header = xmalloc(hdrlen);
1984 write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1985 &hdrlen);
1987 if (!(flags & HASH_WRITE_OBJECT))
1988 goto cleanup;
1989 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1990 goto cleanup;
1991 status = write_loose_object(oid, header, hdrlen, buf, len, 0, 0);
1993 cleanup:
1994 free(header);
1995 return status;
1998 int force_object_loose(const struct object_id *oid, time_t mtime)
2000 void *buf;
2001 unsigned long len;
2002 enum object_type type;
2003 char hdr[MAX_HEADER_LEN];
2004 int hdrlen;
2005 int ret;
2007 if (has_loose_object(oid))
2008 return 0;
2009 buf = read_object(the_repository, oid, &type, &len);
2010 if (!buf)
2011 return error(_("cannot read object for %s"), oid_to_hex(oid));
2012 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
2013 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime, 0);
2014 free(buf);
2016 return ret;
2019 int has_object(struct repository *r, const struct object_id *oid,
2020 unsigned flags)
2022 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2023 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2024 (quick ? OBJECT_INFO_QUICK : 0);
2026 if (!startup_info->have_repository)
2027 return 0;
2028 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2031 int repo_has_object_file_with_flags(struct repository *r,
2032 const struct object_id *oid, int flags)
2034 if (!startup_info->have_repository)
2035 return 0;
2036 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2039 int repo_has_object_file(struct repository *r,
2040 const struct object_id *oid)
2042 return repo_has_object_file_with_flags(r, oid, 0);
2045 static void check_tree(const void *buf, size_t size)
2047 struct tree_desc desc;
2048 struct name_entry entry;
2050 init_tree_desc(&desc, buf, size);
2051 while (tree_entry(&desc, &entry))
2052 /* do nothing
2053 * tree_entry() will die() on malformed entries */
2057 static void check_commit(const void *buf, size_t size)
2059 struct commit c;
2060 memset(&c, 0, sizeof(c));
2061 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2062 die(_("corrupt commit"));
2065 static void check_tag(const void *buf, size_t size)
2067 struct tag t;
2068 memset(&t, 0, sizeof(t));
2069 if (parse_tag_buffer(the_repository, &t, buf, size))
2070 die(_("corrupt tag"));
2073 static int index_mem(struct index_state *istate,
2074 struct object_id *oid, void *buf, size_t size,
2075 enum object_type type,
2076 const char *path, unsigned flags)
2078 int ret, re_allocated = 0;
2079 int write_object = flags & HASH_WRITE_OBJECT;
2081 if (!type)
2082 type = OBJ_BLOB;
2085 * Convert blobs to git internal format
2087 if ((type == OBJ_BLOB) && path) {
2088 struct strbuf nbuf = STRBUF_INIT;
2089 if (convert_to_git(istate, path, buf, size, &nbuf,
2090 get_conv_flags(flags))) {
2091 buf = strbuf_detach(&nbuf, &size);
2092 re_allocated = 1;
2095 if (flags & HASH_FORMAT_CHECK) {
2096 if (type == OBJ_TREE)
2097 check_tree(buf, size);
2098 if (type == OBJ_COMMIT)
2099 check_commit(buf, size);
2100 if (type == OBJ_TAG)
2101 check_tag(buf, size);
2104 if (write_object)
2105 ret = write_object_file(buf, size, type_name(type), oid);
2106 else
2107 ret = hash_object_file(the_hash_algo, buf, size,
2108 type_name(type), oid);
2109 if (re_allocated)
2110 free(buf);
2111 return ret;
2114 static int index_stream_convert_blob(struct index_state *istate,
2115 struct object_id *oid,
2116 int fd,
2117 const char *path,
2118 unsigned flags)
2120 int ret;
2121 const int write_object = flags & HASH_WRITE_OBJECT;
2122 struct strbuf sbuf = STRBUF_INIT;
2124 assert(path);
2125 assert(would_convert_to_git_filter_fd(istate, path));
2127 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2128 get_conv_flags(flags));
2130 if (write_object)
2131 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2132 oid);
2133 else
2134 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2135 type_name(OBJ_BLOB), oid);
2136 strbuf_release(&sbuf);
2137 return ret;
2140 static int index_pipe(struct index_state *istate, struct object_id *oid,
2141 int fd, enum object_type type,
2142 const char *path, unsigned flags)
2144 struct strbuf sbuf = STRBUF_INIT;
2145 int ret;
2147 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2148 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2149 else
2150 ret = -1;
2151 strbuf_release(&sbuf);
2152 return ret;
2155 #define SMALL_FILE_SIZE (32*1024)
2157 static int index_core(struct index_state *istate,
2158 struct object_id *oid, int fd, size_t size,
2159 enum object_type type, const char *path,
2160 unsigned flags)
2162 int ret;
2164 if (!size) {
2165 ret = index_mem(istate, oid, "", size, type, path, flags);
2166 } else if (size <= SMALL_FILE_SIZE) {
2167 char *buf = xmalloc(size);
2168 ssize_t read_result = read_in_full(fd, buf, size);
2169 if (read_result < 0)
2170 ret = error_errno(_("read error while indexing %s"),
2171 path ? path : "<unknown>");
2172 else if (read_result != size)
2173 ret = error(_("short read while indexing %s"),
2174 path ? path : "<unknown>");
2175 else
2176 ret = index_mem(istate, oid, buf, size, type, path, flags);
2177 free(buf);
2178 } else {
2179 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2180 ret = index_mem(istate, oid, buf, size, type, path, flags);
2181 munmap(buf, size);
2183 return ret;
2187 * This creates one packfile per large blob unless bulk-checkin
2188 * machinery is "plugged".
2190 * This also bypasses the usual "convert-to-git" dance, and that is on
2191 * purpose. We could write a streaming version of the converting
2192 * functions and insert that before feeding the data to fast-import
2193 * (or equivalent in-core API described above). However, that is
2194 * somewhat complicated, as we do not know the size of the filter
2195 * result, which we need to know beforehand when writing a git object.
2196 * Since the primary motivation for trying to stream from the working
2197 * tree file and to avoid mmaping it in core is to deal with large
2198 * binary blobs, they generally do not want to get any conversion, and
2199 * callers should avoid this code path when filters are requested.
2201 static int index_stream(struct object_id *oid, int fd, size_t size,
2202 enum object_type type, const char *path,
2203 unsigned flags)
2205 return index_bulk_checkin(oid, fd, size, type, path, flags);
2208 int index_fd(struct index_state *istate, struct object_id *oid,
2209 int fd, struct stat *st,
2210 enum object_type type, const char *path, unsigned flags)
2212 int ret;
2215 * Call xsize_t() only when needed to avoid potentially unnecessary
2216 * die() for large files.
2218 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2219 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2220 else if (!S_ISREG(st->st_mode))
2221 ret = index_pipe(istate, oid, fd, type, path, flags);
2222 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2223 (path && would_convert_to_git(istate, path)))
2224 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2225 type, path, flags);
2226 else
2227 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2228 flags);
2229 close(fd);
2230 return ret;
2233 int index_path(struct index_state *istate, struct object_id *oid,
2234 const char *path, struct stat *st, unsigned flags)
2236 int fd;
2237 struct strbuf sb = STRBUF_INIT;
2238 int rc = 0;
2240 switch (st->st_mode & S_IFMT) {
2241 case S_IFREG:
2242 fd = open(path, O_RDONLY);
2243 if (fd < 0)
2244 return error_errno("open(\"%s\")", path);
2245 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2246 return error(_("%s: failed to insert into database"),
2247 path);
2248 break;
2249 case S_IFLNK:
2250 if (strbuf_readlink(&sb, path, st->st_size))
2251 return error_errno("readlink(\"%s\")", path);
2252 if (!(flags & HASH_WRITE_OBJECT))
2253 hash_object_file(the_hash_algo, sb.buf, sb.len,
2254 blob_type, oid);
2255 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2256 rc = error(_("%s: failed to insert into database"), path);
2257 strbuf_release(&sb);
2258 break;
2259 case S_IFDIR:
2260 return resolve_gitlink_ref(path, "HEAD", oid);
2261 default:
2262 return error(_("%s: unsupported file type"), path);
2264 return rc;
2267 int read_pack_header(int fd, struct pack_header *header)
2269 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2270 /* "eof before pack header was fully read" */
2271 return PH_ERROR_EOF;
2273 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2274 /* "protocol error (pack signature mismatch detected)" */
2275 return PH_ERROR_PACK_SIGNATURE;
2276 if (!pack_version_ok(header->hdr_version))
2277 /* "protocol error (pack version unsupported)" */
2278 return PH_ERROR_PROTOCOL;
2279 return 0;
2282 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2284 enum object_type type = oid_object_info(the_repository, oid, NULL);
2285 if (type < 0)
2286 die(_("%s is not a valid object"), oid_to_hex(oid));
2287 if (type != expect)
2288 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2289 type_name(expect));
2292 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2293 struct strbuf *path,
2294 each_loose_object_fn obj_cb,
2295 each_loose_cruft_fn cruft_cb,
2296 each_loose_subdir_fn subdir_cb,
2297 void *data)
2299 size_t origlen, baselen;
2300 DIR *dir;
2301 struct dirent *de;
2302 int r = 0;
2303 struct object_id oid;
2305 if (subdir_nr > 0xff)
2306 BUG("invalid loose object subdirectory: %x", subdir_nr);
2308 origlen = path->len;
2309 strbuf_complete(path, '/');
2310 strbuf_addf(path, "%02x", subdir_nr);
2312 dir = opendir(path->buf);
2313 if (!dir) {
2314 if (errno != ENOENT)
2315 r = error_errno(_("unable to open %s"), path->buf);
2316 strbuf_setlen(path, origlen);
2317 return r;
2320 oid.hash[0] = subdir_nr;
2321 strbuf_addch(path, '/');
2322 baselen = path->len;
2324 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2325 size_t namelen;
2327 namelen = strlen(de->d_name);
2328 strbuf_setlen(path, baselen);
2329 strbuf_add(path, de->d_name, namelen);
2330 if (namelen == the_hash_algo->hexsz - 2 &&
2331 !hex_to_bytes(oid.hash + 1, de->d_name,
2332 the_hash_algo->rawsz - 1)) {
2333 oid_set_algo(&oid, the_hash_algo);
2334 if (obj_cb) {
2335 r = obj_cb(&oid, path->buf, data);
2336 if (r)
2337 break;
2339 continue;
2342 if (cruft_cb) {
2343 r = cruft_cb(de->d_name, path->buf, data);
2344 if (r)
2345 break;
2348 closedir(dir);
2350 strbuf_setlen(path, baselen - 1);
2351 if (!r && subdir_cb)
2352 r = subdir_cb(subdir_nr, path->buf, data);
2354 strbuf_setlen(path, origlen);
2356 return r;
2359 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2360 each_loose_object_fn obj_cb,
2361 each_loose_cruft_fn cruft_cb,
2362 each_loose_subdir_fn subdir_cb,
2363 void *data)
2365 int r = 0;
2366 int i;
2368 for (i = 0; i < 256; i++) {
2369 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2370 subdir_cb, data);
2371 if (r)
2372 break;
2375 return r;
2378 int for_each_loose_file_in_objdir(const char *path,
2379 each_loose_object_fn obj_cb,
2380 each_loose_cruft_fn cruft_cb,
2381 each_loose_subdir_fn subdir_cb,
2382 void *data)
2384 struct strbuf buf = STRBUF_INIT;
2385 int r;
2387 strbuf_addstr(&buf, path);
2388 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2389 subdir_cb, data);
2390 strbuf_release(&buf);
2392 return r;
2395 int for_each_loose_object(each_loose_object_fn cb, void *data,
2396 enum for_each_object_flags flags)
2398 struct object_directory *odb;
2400 prepare_alt_odb(the_repository);
2401 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2402 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2403 NULL, data);
2404 if (r)
2405 return r;
2407 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2408 break;
2411 return 0;
2414 static int append_loose_object(const struct object_id *oid, const char *path,
2415 void *data)
2417 oidtree_insert(data, oid);
2418 return 0;
2421 struct oidtree *odb_loose_cache(struct object_directory *odb,
2422 const struct object_id *oid)
2424 int subdir_nr = oid->hash[0];
2425 struct strbuf buf = STRBUF_INIT;
2426 size_t word_bits = bitsizeof(odb->loose_objects_subdir_seen[0]);
2427 size_t word_index = subdir_nr / word_bits;
2428 size_t mask = 1u << (subdir_nr % word_bits);
2429 uint32_t *bitmap;
2431 if (subdir_nr < 0 ||
2432 subdir_nr >= bitsizeof(odb->loose_objects_subdir_seen))
2433 BUG("subdir_nr out of range");
2435 bitmap = &odb->loose_objects_subdir_seen[word_index];
2436 if (*bitmap & mask)
2437 return odb->loose_objects_cache;
2438 if (!odb->loose_objects_cache) {
2439 ALLOC_ARRAY(odb->loose_objects_cache, 1);
2440 oidtree_init(odb->loose_objects_cache);
2442 strbuf_addstr(&buf, odb->path);
2443 for_each_file_in_obj_subdir(subdir_nr, &buf,
2444 append_loose_object,
2445 NULL, NULL,
2446 odb->loose_objects_cache);
2447 *bitmap |= mask;
2448 strbuf_release(&buf);
2449 return odb->loose_objects_cache;
2452 void odb_clear_loose_cache(struct object_directory *odb)
2454 oidtree_clear(odb->loose_objects_cache);
2455 FREE_AND_NULL(odb->loose_objects_cache);
2456 memset(&odb->loose_objects_subdir_seen, 0,
2457 sizeof(odb->loose_objects_subdir_seen));
2460 static int check_stream_oid(git_zstream *stream,
2461 const char *hdr,
2462 unsigned long size,
2463 const char *path,
2464 const struct object_id *expected_oid)
2466 git_hash_ctx c;
2467 struct object_id real_oid;
2468 unsigned char buf[4096];
2469 unsigned long total_read;
2470 int status = Z_OK;
2472 the_hash_algo->init_fn(&c);
2473 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2476 * We already read some bytes into hdr, but the ones up to the NUL
2477 * do not count against the object's content size.
2479 total_read = stream->total_out - strlen(hdr) - 1;
2482 * This size comparison must be "<=" to read the final zlib packets;
2483 * see the comment in unpack_loose_rest for details.
2485 while (total_read <= size &&
2486 (status == Z_OK ||
2487 (status == Z_BUF_ERROR && !stream->avail_out))) {
2488 stream->next_out = buf;
2489 stream->avail_out = sizeof(buf);
2490 if (size - total_read < stream->avail_out)
2491 stream->avail_out = size - total_read;
2492 status = git_inflate(stream, Z_FINISH);
2493 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2494 total_read += stream->next_out - buf;
2496 git_inflate_end(stream);
2498 if (status != Z_STREAM_END) {
2499 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2500 return -1;
2502 if (stream->avail_in) {
2503 error(_("garbage at end of loose object '%s'"),
2504 oid_to_hex(expected_oid));
2505 return -1;
2508 the_hash_algo->final_oid_fn(&real_oid, &c);
2509 if (!oideq(expected_oid, &real_oid)) {
2510 error(_("hash mismatch for %s (expected %s)"), path,
2511 oid_to_hex(expected_oid));
2512 return -1;
2515 return 0;
2518 int read_loose_object(const char *path,
2519 const struct object_id *expected_oid,
2520 struct object_id *real_oid,
2521 void **contents,
2522 struct object_info *oi)
2524 int ret = -1;
2525 void *map = NULL;
2526 unsigned long mapsize;
2527 git_zstream stream;
2528 char hdr[MAX_HEADER_LEN];
2529 unsigned long *size = oi->sizep;
2531 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2532 if (!map) {
2533 error_errno(_("unable to mmap %s"), path);
2534 goto out;
2537 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
2538 NULL) < 0) {
2539 error(_("unable to unpack header of %s"), path);
2540 goto out;
2543 if (parse_loose_header(hdr, oi) < 0) {
2544 error(_("unable to parse header of %s"), path);
2545 git_inflate_end(&stream);
2546 goto out;
2549 if (*oi->typep == OBJ_BLOB && *size > big_file_threshold) {
2550 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2551 goto out;
2552 } else {
2553 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2554 if (!*contents) {
2555 error(_("unable to unpack contents of %s"), path);
2556 git_inflate_end(&stream);
2557 goto out;
2559 if (check_object_signature(the_repository, expected_oid,
2560 *contents, *size,
2561 oi->type_name->buf, real_oid))
2562 goto out;
2565 ret = 0; /* everything checks out */
2567 out:
2568 if (map)
2569 munmap(map, mapsize);
2570 return ret;