2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
11 #include "string-list.h"
17 #include "run-command.h"
20 #include "tree-walk.h"
22 #include "pack-revindex.h"
23 #include "sha1-lookup.h"
24 #include "bulk-checkin.h"
25 #include "repository.h"
26 #include "replace-object.h"
27 #include "streaming.h"
30 #include "mergesort.h"
33 #include "object-store.h"
34 #include "promisor-remote.h"
36 /* The maximum size for an object header. */
37 #define MAX_HEADER_LEN 32
40 #define EMPTY_TREE_SHA1_BIN_LITERAL \
41 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
42 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
43 #define EMPTY_TREE_SHA256_BIN_LITERAL \
44 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
45 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
46 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
49 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
50 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
51 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
52 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
53 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
54 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
55 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
58 const struct object_id null_oid;
59 static const struct object_id empty_tree_oid = {
60 EMPTY_TREE_SHA1_BIN_LITERAL
62 static const struct object_id empty_blob_oid = {
63 EMPTY_BLOB_SHA1_BIN_LITERAL
65 static const struct object_id empty_tree_oid_sha256 = {
66 EMPTY_TREE_SHA256_BIN_LITERAL
68 static const struct object_id empty_blob_oid_sha256 = {
69 EMPTY_BLOB_SHA256_BIN_LITERAL
72 static void git_hash_sha1_init(git_hash_ctx *ctx)
74 git_SHA1_Init(&ctx->sha1);
77 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
79 git_SHA1_Update(&ctx->sha1, data, len);
82 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
84 git_SHA1_Final(hash, &ctx->sha1);
88 static void git_hash_sha256_init(git_hash_ctx *ctx)
90 git_SHA256_Init(&ctx->sha256);
93 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
95 git_SHA256_Update(&ctx->sha256, data, len);
98 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
100 git_SHA256_Final(hash, &ctx->sha256);
103 static void git_hash_unknown_init(git_hash_ctx *ctx)
105 BUG("trying to init unknown hash");
108 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
110 BUG("trying to update unknown hash");
113 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
115 BUG("trying to finalize unknown hash");
118 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
125 git_hash_unknown_init,
126 git_hash_unknown_update,
127 git_hash_unknown_final,
133 /* "sha1", big-endian */
139 git_hash_sha1_update,
146 /* "s256", big-endian */
151 git_hash_sha256_init,
152 git_hash_sha256_update,
153 git_hash_sha256_final,
154 &empty_tree_oid_sha256,
155 &empty_blob_oid_sha256,
159 const char *empty_tree_oid_hex(void)
161 static char buf[GIT_MAX_HEXSZ + 1];
162 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
165 const char *empty_blob_oid_hex(void)
167 static char buf[GIT_MAX_HEXSZ + 1];
168 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
171 int hash_algo_by_name(const char *name)
175 return GIT_HASH_UNKNOWN;
176 for (i = 1; i < GIT_HASH_NALGOS; i++)
177 if (!strcmp(name, hash_algos[i].name))
179 return GIT_HASH_UNKNOWN;
182 int hash_algo_by_id(uint32_t format_id)
185 for (i = 1; i < GIT_HASH_NALGOS; i++)
186 if (format_id == hash_algos[i].format_id)
188 return GIT_HASH_UNKNOWN;
191 int hash_algo_by_length(int len)
194 for (i = 1; i < GIT_HASH_NALGOS; i++)
195 if (len == hash_algos[i].rawsz)
197 return GIT_HASH_UNKNOWN;
201 * This is meant to hold a *small* number of objects that you would
202 * want read_object_file() to be able to return, but yet you do not want
203 * to write them into the object store (e.g. a browse-only
206 static struct cached_object {
207 struct object_id oid;
208 enum object_type type;
212 static int cached_object_nr, cached_object_alloc;
214 static struct cached_object empty_tree = {
215 { EMPTY_TREE_SHA1_BIN_LITERAL },
221 static struct cached_object *find_cached_object(const struct object_id *oid)
224 struct cached_object *co = cached_objects;
226 for (i = 0; i < cached_object_nr; i++, co++) {
227 if (oideq(&co->oid, oid))
230 if (oideq(oid, the_hash_algo->empty_tree))
236 static int get_conv_flags(unsigned flags)
238 if (flags & HASH_RENORMALIZE)
239 return CONV_EOL_RENORMALIZE;
240 else if (flags & HASH_WRITE_OBJECT)
241 return global_conv_flags_eol | CONV_WRITE_OBJECT;
247 int mkdir_in_gitdir(const char *path)
249 if (mkdir(path, 0777)) {
250 int saved_errno = errno;
252 struct strbuf sb = STRBUF_INIT;
257 * Are we looking at a path in a symlinked worktree
258 * whose original repository does not yet have it?
259 * e.g. .git/rr-cache pointing at its original
260 * repository in which the user hasn't performed any
261 * conflict resolution yet?
263 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
264 strbuf_readlink(&sb, path, st.st_size) ||
265 !is_absolute_path(sb.buf) ||
266 mkdir(sb.buf, 0777)) {
273 return adjust_shared_perm(path);
276 enum scld_error safe_create_leading_directories(char *path)
278 char *next_component = path + offset_1st_component(path);
279 enum scld_error ret = SCLD_OK;
281 while (ret == SCLD_OK && next_component) {
283 char *slash = next_component, slash_character;
285 while (*slash && !is_dir_sep(*slash))
291 next_component = slash + 1;
292 while (is_dir_sep(*next_component))
294 if (!*next_component)
297 slash_character = *slash;
299 if (!stat(path, &st)) {
301 if (!S_ISDIR(st.st_mode)) {
305 } else if (mkdir(path, 0777)) {
306 if (errno == EEXIST &&
307 !stat(path, &st) && S_ISDIR(st.st_mode))
308 ; /* somebody created it since we checked */
309 else if (errno == ENOENT)
311 * Either mkdir() failed because
312 * somebody just pruned the containing
313 * directory, or stat() failed because
314 * the file that was in our way was
315 * just removed. Either way, inform
316 * the caller that it might be worth
322 } else if (adjust_shared_perm(path)) {
325 *slash = slash_character;
330 enum scld_error safe_create_leading_directories_const(const char *path)
333 /* path points to cache entries, so xstrdup before messing with it */
334 char *buf = xstrdup(path);
335 enum scld_error result = safe_create_leading_directories(buf);
343 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
346 * The number of times we will try to remove empty directories
347 * in the way of path. This is only 1 because if another
348 * process is racily creating directories that conflict with
349 * us, we don't want to fight against them.
351 int remove_directories_remaining = 1;
354 * The number of times that we will try to create the
355 * directories containing path. We are willing to attempt this
356 * more than once, because another process could be trying to
357 * clean up empty directories at the same time as we are
358 * trying to create them.
360 int create_directories_remaining = 3;
362 /* A scratch copy of path, filled lazily if we need it: */
363 struct strbuf path_copy = STRBUF_INIT;
376 if (errno == EISDIR && remove_directories_remaining-- > 0) {
378 * A directory is in the way. Maybe it is empty; try
382 strbuf_addstr(&path_copy, path);
384 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
386 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
388 * Maybe the containing directory didn't exist, or
389 * maybe it was just deleted by a process that is
390 * racing with us to clean up empty directories. Try
393 enum scld_error scld_result;
396 strbuf_addstr(&path_copy, path);
399 scld_result = safe_create_leading_directories(path_copy.buf);
400 if (scld_result == SCLD_OK)
402 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
406 strbuf_release(&path_copy);
411 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
414 for (i = 0; i < the_hash_algo->rawsz; i++) {
415 static char hex[] = "0123456789abcdef";
416 unsigned int val = oid->hash[i];
417 strbuf_addch(buf, hex[val >> 4]);
418 strbuf_addch(buf, hex[val & 0xf]);
420 strbuf_addch(buf, '/');
424 static const char *odb_loose_path(struct object_directory *odb,
426 const struct object_id *oid)
429 strbuf_addstr(buf, odb->path);
430 strbuf_addch(buf, '/');
431 fill_loose_path(buf, oid);
435 const char *loose_object_path(struct repository *r, struct strbuf *buf,
436 const struct object_id *oid)
438 return odb_loose_path(r->objects->odb, buf, oid);
442 * Return non-zero iff the path is usable as an alternate object database.
444 static int alt_odb_usable(struct raw_object_store *o,
446 const char *normalized_objdir)
448 struct object_directory *odb;
450 /* Detect cases where alternate disappeared */
451 if (!is_directory(path->buf)) {
452 error(_("object directory %s does not exist; "
453 "check .git/objects/info/alternates"),
459 * Prevent the common mistake of listing the same
460 * thing twice, or object directory itself.
462 for (odb = o->odb; odb; odb = odb->next) {
463 if (!fspathcmp(path->buf, odb->path))
466 if (!fspathcmp(path->buf, normalized_objdir))
473 * Prepare alternate object database registry.
475 * The variable alt_odb_list points at the list of struct
476 * object_directory. The elements on this list come from
477 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
478 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
479 * whose contents is similar to that environment variable but can be
480 * LF separated. Its base points at a statically allocated buffer that
481 * contains "/the/directory/corresponding/to/.git/objects/...", while
482 * its name points just after the slash at the end of ".git/objects/"
483 * in the example above, and has enough space to hold 40-byte hex
484 * SHA1, an extra slash for the first level indirection, and the
487 static void read_info_alternates(struct repository *r,
488 const char *relative_base,
490 static int link_alt_odb_entry(struct repository *r, const char *entry,
491 const char *relative_base, int depth, const char *normalized_objdir)
493 struct object_directory *ent;
494 struct strbuf pathbuf = STRBUF_INIT;
496 if (!is_absolute_path(entry) && relative_base) {
497 strbuf_realpath(&pathbuf, relative_base, 1);
498 strbuf_addch(&pathbuf, '/');
500 strbuf_addstr(&pathbuf, entry);
502 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
503 error(_("unable to normalize alternate object path: %s"),
505 strbuf_release(&pathbuf);
510 * The trailing slash after the directory name is given by
511 * this function at the end. Remove duplicates.
513 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
514 strbuf_setlen(&pathbuf, pathbuf.len - 1);
516 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
517 strbuf_release(&pathbuf);
521 ent = xcalloc(1, sizeof(*ent));
522 ent->path = xstrdup(pathbuf.buf);
524 /* add the alternate entry */
525 *r->objects->odb_tail = ent;
526 r->objects->odb_tail = &(ent->next);
529 /* recursively add alternates */
530 read_info_alternates(r, pathbuf.buf, depth + 1);
532 strbuf_release(&pathbuf);
536 static const char *parse_alt_odb_entry(const char *string,
544 if (*string == '#') {
545 /* comment; consume up to next separator */
546 end = strchrnul(string, sep);
547 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
549 * quoted path; unquote_c_style has copied the
550 * data for us and set "end". Broken quoting (e.g.,
551 * an entry that doesn't end with a quote) falls
552 * back to the unquoted case below.
555 /* normal, unquoted path */
556 end = strchrnul(string, sep);
557 strbuf_add(out, string, end - string);
565 static void link_alt_odb_entries(struct repository *r, const char *alt,
566 int sep, const char *relative_base, int depth)
568 struct strbuf objdirbuf = STRBUF_INIT;
569 struct strbuf entry = STRBUF_INIT;
575 error(_("%s: ignoring alternate object stores, nesting too deep"),
580 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
581 if (strbuf_normalize_path(&objdirbuf) < 0)
582 die(_("unable to normalize object directory: %s"),
586 alt = parse_alt_odb_entry(alt, sep, &entry);
589 link_alt_odb_entry(r, entry.buf,
590 relative_base, depth, objdirbuf.buf);
592 strbuf_release(&entry);
593 strbuf_release(&objdirbuf);
596 static void read_info_alternates(struct repository *r,
597 const char *relative_base,
601 struct strbuf buf = STRBUF_INIT;
603 path = xstrfmt("%s/info/alternates", relative_base);
604 if (strbuf_read_file(&buf, path, 1024) < 0) {
605 warn_on_fopen_errors(path);
610 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
611 strbuf_release(&buf);
615 void add_to_alternates_file(const char *reference)
617 struct lock_file lock = LOCK_INIT;
618 char *alts = git_pathdup("objects/info/alternates");
622 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
623 out = fdopen_lock_file(&lock, "w");
625 die_errno(_("unable to fdopen alternates lockfile"));
627 in = fopen(alts, "r");
629 struct strbuf line = STRBUF_INIT;
631 while (strbuf_getline(&line, in) != EOF) {
632 if (!strcmp(reference, line.buf)) {
636 fprintf_or_die(out, "%s\n", line.buf);
639 strbuf_release(&line);
642 else if (errno != ENOENT)
643 die_errno(_("unable to read alternates file"));
646 rollback_lock_file(&lock);
648 fprintf_or_die(out, "%s\n", reference);
649 if (commit_lock_file(&lock))
650 die_errno(_("unable to move new alternates file into place"));
651 if (the_repository->objects->loaded_alternates)
652 link_alt_odb_entries(the_repository, reference,
658 void add_to_alternates_memory(const char *reference)
661 * Make sure alternates are initialized, or else our entry may be
662 * overwritten when they are.
664 prepare_alt_odb(the_repository);
666 link_alt_odb_entries(the_repository, reference,
671 * Compute the exact path an alternate is at and returns it. In case of
672 * error NULL is returned and the human readable error is added to `err`
673 * `path` may be relative and should point to $GIT_DIR.
674 * `err` must not be null.
676 char *compute_alternate_path(const char *path, struct strbuf *err)
678 char *ref_git = NULL;
682 ref_git = real_pathdup(path, 0);
685 strbuf_addf(err, _("path '%s' does not exist"), path);
689 repo = read_gitfile(ref_git);
691 repo = read_gitfile(mkpath("%s/.git", ref_git));
694 ref_git = xstrdup(repo);
697 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
698 char *ref_git_git = mkpathdup("%s/.git", ref_git);
700 ref_git = ref_git_git;
701 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
702 struct strbuf sb = STRBUF_INIT;
704 if (get_common_dir(&sb, ref_git)) {
706 _("reference repository '%s' as a linked "
707 "checkout is not supported yet."),
712 strbuf_addf(err, _("reference repository '%s' is not a "
713 "local repository."), path);
717 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
718 strbuf_addf(err, _("reference repository '%s' is shallow"),
724 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
726 _("reference repository '%s' is grafted"),
734 FREE_AND_NULL(ref_git);
740 static void fill_alternate_refs_command(struct child_process *cmd,
741 const char *repo_path)
745 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
748 argv_array_push(&cmd->args, value);
749 argv_array_push(&cmd->args, repo_path);
753 argv_array_pushf(&cmd->args, "--git-dir=%s", repo_path);
754 argv_array_push(&cmd->args, "for-each-ref");
755 argv_array_push(&cmd->args, "--format=%(objectname)");
757 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
758 argv_array_push(&cmd->args, "--");
759 argv_array_split(&cmd->args, value);
763 cmd->env = local_repo_env;
767 static void read_alternate_refs(const char *path,
768 alternate_ref_fn *cb,
771 struct child_process cmd = CHILD_PROCESS_INIT;
772 struct strbuf line = STRBUF_INIT;
775 fill_alternate_refs_command(&cmd, path);
777 if (start_command(&cmd))
780 fh = xfdopen(cmd.out, "r");
781 while (strbuf_getline_lf(&line, fh) != EOF) {
782 struct object_id oid;
785 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
786 warning(_("invalid line while parsing alternate refs: %s"),
795 finish_command(&cmd);
796 strbuf_release(&line);
799 struct alternate_refs_data {
800 alternate_ref_fn *fn;
804 static int refs_from_alternate_cb(struct object_directory *e,
807 struct strbuf path = STRBUF_INIT;
809 struct alternate_refs_data *cb = data;
811 if (!strbuf_realpath(&path, e->path, 0))
813 if (!strbuf_strip_suffix(&path, "/objects"))
817 /* Is this a git repository with refs? */
818 strbuf_addstr(&path, "/refs");
819 if (!is_directory(path.buf))
821 strbuf_setlen(&path, base_len);
823 read_alternate_refs(path.buf, cb->fn, cb->data);
826 strbuf_release(&path);
830 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
832 struct alternate_refs_data cb;
835 foreach_alt_odb(refs_from_alternate_cb, &cb);
838 int foreach_alt_odb(alt_odb_fn fn, void *cb)
840 struct object_directory *ent;
843 prepare_alt_odb(the_repository);
844 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
852 void prepare_alt_odb(struct repository *r)
854 if (r->objects->loaded_alternates)
857 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
859 read_info_alternates(r, r->objects->odb->path, 0);
860 r->objects->loaded_alternates = 1;
863 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
864 static int freshen_file(const char *fn)
867 t.actime = t.modtime = time(NULL);
868 return !utime(fn, &t);
872 * All of the check_and_freshen functions return 1 if the file exists and was
873 * freshened (if freshening was requested), 0 otherwise. If they return
874 * 0, you should not assume that it is safe to skip a write of the object (it
875 * either does not exist on disk, or has a stale mtime and may be subject to
878 int check_and_freshen_file(const char *fn, int freshen)
880 if (access(fn, F_OK))
882 if (freshen && !freshen_file(fn))
887 static int check_and_freshen_odb(struct object_directory *odb,
888 const struct object_id *oid,
891 static struct strbuf path = STRBUF_INIT;
892 odb_loose_path(odb, &path, oid);
893 return check_and_freshen_file(path.buf, freshen);
896 static int check_and_freshen_local(const struct object_id *oid, int freshen)
898 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
901 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
903 struct object_directory *odb;
905 prepare_alt_odb(the_repository);
906 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
907 if (check_and_freshen_odb(odb, oid, freshen))
913 static int check_and_freshen(const struct object_id *oid, int freshen)
915 return check_and_freshen_local(oid, freshen) ||
916 check_and_freshen_nonlocal(oid, freshen);
919 int has_loose_object_nonlocal(const struct object_id *oid)
921 return check_and_freshen_nonlocal(oid, 0);
924 static int has_loose_object(const struct object_id *oid)
926 return check_and_freshen(oid, 0);
929 static void mmap_limit_check(size_t length)
931 static size_t limit = 0;
933 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
938 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
939 (uintmax_t)length, (uintmax_t)limit);
942 void *xmmap_gently(void *start, size_t length,
943 int prot, int flags, int fd, off_t offset)
947 mmap_limit_check(length);
948 ret = mmap(start, length, prot, flags, fd, offset);
949 if (ret == MAP_FAILED && !length)
954 void *xmmap(void *start, size_t length,
955 int prot, int flags, int fd, off_t offset)
957 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
958 if (ret == MAP_FAILED)
959 die_errno(_("mmap failed"));
964 * With an in-core object data in "map", rehash it to make sure the
965 * object name actually matches "oid" to detect object corruption.
966 * With "map" == NULL, try reading the object named with "oid" using
967 * the streaming interface and rehash it to do the same.
969 int check_object_signature(struct repository *r, const struct object_id *oid,
970 void *map, unsigned long size, const char *type)
972 struct object_id real_oid;
973 enum object_type obj_type;
974 struct git_istream *st;
976 char hdr[MAX_HEADER_LEN];
980 hash_object_file(r->hash_algo, map, size, type, &real_oid);
981 return !oideq(oid, &real_oid) ? -1 : 0;
984 st = open_istream(r, oid, &obj_type, &size, NULL);
988 /* Generate the header */
989 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
992 r->hash_algo->init_fn(&c);
993 r->hash_algo->update_fn(&c, hdr, hdrlen);
996 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1004 r->hash_algo->update_fn(&c, buf, readlen);
1006 r->hash_algo->final_fn(real_oid.hash, &c);
1008 return !oideq(oid, &real_oid) ? -1 : 0;
1011 int git_open_cloexec(const char *name, int flags)
1014 static int o_cloexec = O_CLOEXEC;
1016 fd = open(name, flags | o_cloexec);
1017 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1018 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1019 o_cloexec &= ~O_CLOEXEC;
1020 fd = open(name, flags | o_cloexec);
1023 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1025 static int fd_cloexec = FD_CLOEXEC;
1027 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1028 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1029 int flags = fcntl(fd, F_GETFD);
1030 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1039 * Find "oid" as a loose object in the local repository or in an alternate.
1040 * Returns 0 on success, negative on failure.
1042 * The "path" out-parameter will give the path of the object we found (if any).
1043 * Note that it may point to static storage and is only valid until another
1044 * call to stat_loose_object().
1046 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1047 struct stat *st, const char **path)
1049 struct object_directory *odb;
1050 static struct strbuf buf = STRBUF_INIT;
1053 for (odb = r->objects->odb; odb; odb = odb->next) {
1054 *path = odb_loose_path(odb, &buf, oid);
1055 if (!lstat(*path, st))
1063 * Like stat_loose_object(), but actually open the object and return the
1064 * descriptor. See the caveats on the "path" parameter above.
1066 static int open_loose_object(struct repository *r,
1067 const struct object_id *oid, const char **path)
1070 struct object_directory *odb;
1071 int most_interesting_errno = ENOENT;
1072 static struct strbuf buf = STRBUF_INIT;
1075 for (odb = r->objects->odb; odb; odb = odb->next) {
1076 *path = odb_loose_path(odb, &buf, oid);
1077 fd = git_open(*path);
1081 if (most_interesting_errno == ENOENT)
1082 most_interesting_errno = errno;
1084 errno = most_interesting_errno;
1088 static int quick_has_loose(struct repository *r,
1089 const struct object_id *oid)
1091 struct object_directory *odb;
1094 for (odb = r->objects->odb; odb; odb = odb->next) {
1095 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1102 * Map the loose object at "path" if it is not NULL, or the path found by
1103 * searching for a loose object named "oid".
1105 static void *map_loose_object_1(struct repository *r, const char *path,
1106 const struct object_id *oid, unsigned long *size)
1112 fd = git_open(path);
1114 fd = open_loose_object(r, oid, &path);
1119 if (!fstat(fd, &st)) {
1120 *size = xsize_t(st.st_size);
1122 /* mmap() is forbidden on empty files */
1123 error(_("object file %s is empty"), path);
1127 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1134 void *map_loose_object(struct repository *r,
1135 const struct object_id *oid,
1136 unsigned long *size)
1138 return map_loose_object_1(r, NULL, oid, size);
1141 static int unpack_loose_short_header(git_zstream *stream,
1142 unsigned char *map, unsigned long mapsize,
1143 void *buffer, unsigned long bufsiz)
1147 /* Get the data stream */
1148 memset(stream, 0, sizeof(*stream));
1149 stream->next_in = map;
1150 stream->avail_in = mapsize;
1151 stream->next_out = buffer;
1152 stream->avail_out = bufsiz;
1154 git_inflate_init(stream);
1156 ret = git_inflate(stream, 0);
1162 int unpack_loose_header(git_zstream *stream,
1163 unsigned char *map, unsigned long mapsize,
1164 void *buffer, unsigned long bufsiz)
1166 int status = unpack_loose_short_header(stream, map, mapsize,
1172 /* Make sure we have the terminating NUL */
1173 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1178 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1179 unsigned long mapsize, void *buffer,
1180 unsigned long bufsiz, struct strbuf *header)
1184 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1189 * Check if entire header is unpacked in the first iteration.
1191 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1195 * buffer[0..bufsiz] was not large enough. Copy the partial
1196 * result out to header, and then append the result of further
1197 * reading the stream.
1199 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1200 stream->next_out = buffer;
1201 stream->avail_out = bufsiz;
1205 status = git_inflate(stream, 0);
1207 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1208 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1210 stream->next_out = buffer;
1211 stream->avail_out = bufsiz;
1212 } while (status != Z_STREAM_END);
1216 static void *unpack_loose_rest(git_zstream *stream,
1217 void *buffer, unsigned long size,
1218 const struct object_id *oid)
1220 int bytes = strlen(buffer) + 1;
1221 unsigned char *buf = xmallocz(size);
1225 n = stream->total_out - bytes;
1228 memcpy(buf, (char *) buffer + bytes, n);
1230 if (bytes <= size) {
1232 * The above condition must be (bytes <= size), not
1233 * (bytes < size). In other words, even though we
1234 * expect no more output and set avail_out to zero,
1235 * the input zlib stream may have bytes that express
1236 * "this concludes the stream", and we *do* want to
1239 * Otherwise we would not be able to test that we
1240 * consumed all the input to reach the expected size;
1241 * we also want to check that zlib tells us that all
1242 * went well with status == Z_STREAM_END at the end.
1244 stream->next_out = buf + bytes;
1245 stream->avail_out = size - bytes;
1246 while (status == Z_OK) {
1248 status = git_inflate(stream, Z_FINISH);
1252 if (status == Z_STREAM_END && !stream->avail_in) {
1253 git_inflate_end(stream);
1258 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1259 else if (stream->avail_in)
1260 error(_("garbage at end of loose object '%s'"),
1267 * We used to just use "sscanf()", but that's actually way
1268 * too permissive for what we want to check. So do an anal
1269 * object header parse by hand.
1271 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1274 const char *type_buf = hdr;
1276 int type, type_len = 0;
1279 * The type can be of any size but is followed by
1291 type = type_from_string_gently(type_buf, type_len, 1);
1293 strbuf_add(oi->type_name, type_buf, type_len);
1295 * Set type to 0 if its an unknown object and
1296 * we're obtaining the type using '--allow-unknown-type'
1299 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1302 die(_("invalid object type"));
1307 * The length must follow immediately, and be in canonical
1308 * decimal format (ie "010" is not valid).
1310 size = *hdr++ - '0';
1315 unsigned long c = *hdr - '0';
1319 size = size * 10 + c;
1327 * The length must be followed by a zero byte
1329 return *hdr ? -1 : type;
1332 int parse_loose_header(const char *hdr, unsigned long *sizep)
1334 struct object_info oi = OBJECT_INFO_INIT;
1337 return parse_loose_header_extended(hdr, &oi, 0);
1340 static int loose_object_info(struct repository *r,
1341 const struct object_id *oid,
1342 struct object_info *oi, int flags)
1345 unsigned long mapsize;
1348 char hdr[MAX_HEADER_LEN];
1349 struct strbuf hdrbuf = STRBUF_INIT;
1350 unsigned long size_scratch;
1352 if (oi->delta_base_oid)
1353 oidclr(oi->delta_base_oid);
1356 * If we don't care about type or size, then we don't
1357 * need to look inside the object at all. Note that we
1358 * do not optimize out the stat call, even if the
1359 * caller doesn't care about the disk-size, since our
1360 * return value implicitly indicates whether the
1361 * object even exists.
1363 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1366 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1367 return quick_has_loose(r, oid) ? 0 : -1;
1368 if (stat_loose_object(r, oid, &st, &path) < 0)
1371 *oi->disk_sizep = st.st_size;
1375 map = map_loose_object(r, oid, &mapsize);
1380 oi->sizep = &size_scratch;
1383 *oi->disk_sizep = mapsize;
1384 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1385 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1386 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1388 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1389 status = error(_("unable to unpack %s header"),
1393 else if (hdrbuf.len) {
1394 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1395 status = error(_("unable to parse %s header with --allow-unknown-type"),
1397 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1398 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1400 if (status >= 0 && oi->contentp) {
1401 *oi->contentp = unpack_loose_rest(&stream, hdr,
1403 if (!*oi->contentp) {
1404 git_inflate_end(&stream);
1408 git_inflate_end(&stream);
1410 munmap(map, mapsize);
1411 if (status && oi->typep)
1412 *oi->typep = status;
1413 if (oi->sizep == &size_scratch)
1415 strbuf_release(&hdrbuf);
1416 oi->whence = OI_LOOSE;
1417 return (status < 0) ? status : 0;
1420 int obj_read_use_lock = 0;
1421 pthread_mutex_t obj_read_mutex;
1423 void enable_obj_read_lock(void)
1425 if (obj_read_use_lock)
1428 obj_read_use_lock = 1;
1429 init_recursive_mutex(&obj_read_mutex);
1432 void disable_obj_read_lock(void)
1434 if (!obj_read_use_lock)
1437 obj_read_use_lock = 0;
1438 pthread_mutex_destroy(&obj_read_mutex);
1441 int fetch_if_missing = 1;
1443 static int do_oid_object_info_extended(struct repository *r,
1444 const struct object_id *oid,
1445 struct object_info *oi, unsigned flags)
1447 static struct object_info blank_oi = OBJECT_INFO_INIT;
1448 struct cached_object *co;
1449 struct pack_entry e;
1451 const struct object_id *real = oid;
1452 int already_retried = 0;
1455 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1456 real = lookup_replace_object(r, oid);
1458 if (is_null_oid(real))
1464 co = find_cached_object(real);
1467 *(oi->typep) = co->type;
1469 *(oi->sizep) = co->size;
1471 *(oi->disk_sizep) = 0;
1472 if (oi->delta_base_oid)
1473 oidclr(oi->delta_base_oid);
1475 strbuf_addstr(oi->type_name, type_name(co->type));
1477 *oi->contentp = xmemdupz(co->buf, co->size);
1478 oi->whence = OI_CACHED;
1483 if (find_pack_entry(r, real, &e))
1486 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1489 /* Most likely it's a loose object. */
1490 if (!loose_object_info(r, real, oi, flags))
1493 /* Not a loose object; someone else may have just packed it. */
1494 if (!(flags & OBJECT_INFO_QUICK)) {
1495 reprepare_packed_git(r);
1496 if (find_pack_entry(r, real, &e))
1500 /* Check if it is a missing object */
1501 if (fetch_if_missing && has_promisor_remote() &&
1502 !already_retried && r == the_repository &&
1503 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1505 * TODO Investigate checking promisor_remote_get_direct()
1506 * TODO return value and stopping on error here.
1507 * TODO Pass a repository struct through
1508 * promisor_remote_get_direct(), such that arbitrary
1509 * repositories work.
1511 promisor_remote_get_direct(r, real, 1);
1512 already_retried = 1;
1519 if (oi == &blank_oi)
1521 * We know that the caller doesn't actually need the
1522 * information below, so return early.
1525 rtype = packed_object_info(r, e.p, e.offset, oi);
1527 mark_bad_packed_object(e.p, real->hash);
1528 return do_oid_object_info_extended(r, real, oi, 0);
1529 } else if (oi->whence == OI_PACKED) {
1530 oi->u.packed.offset = e.offset;
1531 oi->u.packed.pack = e.p;
1532 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1533 rtype == OBJ_OFS_DELTA);
1539 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1540 struct object_info *oi, unsigned flags)
1544 ret = do_oid_object_info_extended(r, oid, oi, flags);
1550 /* returns enum object_type or negative */
1551 int oid_object_info(struct repository *r,
1552 const struct object_id *oid,
1553 unsigned long *sizep)
1555 enum object_type type;
1556 struct object_info oi = OBJECT_INFO_INIT;
1560 if (oid_object_info_extended(r, oid, &oi,
1561 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1566 static void *read_object(struct repository *r,
1567 const struct object_id *oid, enum object_type *type,
1568 unsigned long *size)
1570 struct object_info oi = OBJECT_INFO_INIT;
1574 oi.contentp = &content;
1576 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1581 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1582 struct object_id *oid)
1584 struct cached_object *co;
1586 hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1587 if (has_object_file(oid) || find_cached_object(oid))
1589 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1590 co = &cached_objects[cached_object_nr++];
1593 co->buf = xmalloc(len);
1594 memcpy(co->buf, buf, len);
1595 oidcpy(&co->oid, oid);
1600 * This function dies on corrupt objects; the callers who want to
1601 * deal with them should arrange to call read_object() and give error
1602 * messages themselves.
1604 void *read_object_file_extended(struct repository *r,
1605 const struct object_id *oid,
1606 enum object_type *type,
1607 unsigned long *size,
1611 const struct packed_git *p;
1614 const struct object_id *repl = lookup_replace ?
1615 lookup_replace_object(r, oid) : oid;
1618 data = read_object(r, repl, type, size);
1623 if (errno && errno != ENOENT)
1624 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1626 /* die if we replaced an object with one that does not exist */
1628 die(_("replacement %s not found for %s"),
1629 oid_to_hex(repl), oid_to_hex(oid));
1631 if (!stat_loose_object(r, repl, &st, &path))
1632 die(_("loose object %s (stored in %s) is corrupt"),
1633 oid_to_hex(repl), path);
1635 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1636 die(_("packed object %s (stored in %s) is corrupt"),
1637 oid_to_hex(repl), p->pack_name);
1643 void *read_object_with_reference(struct repository *r,
1644 const struct object_id *oid,
1645 const char *required_type_name,
1646 unsigned long *size,
1647 struct object_id *actual_oid_return)
1649 enum object_type type, required_type;
1651 unsigned long isize;
1652 struct object_id actual_oid;
1654 required_type = type_from_string(required_type_name);
1655 oidcpy(&actual_oid, oid);
1657 int ref_length = -1;
1658 const char *ref_type = NULL;
1660 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1663 if (type == required_type) {
1665 if (actual_oid_return)
1666 oidcpy(actual_oid_return, &actual_oid);
1669 /* Handle references */
1670 else if (type == OBJ_COMMIT)
1672 else if (type == OBJ_TAG)
1673 ref_type = "object ";
1678 ref_length = strlen(ref_type);
1680 if (ref_length + the_hash_algo->hexsz > isize ||
1681 memcmp(buffer, ref_type, ref_length) ||
1682 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1687 /* Now we have the ID of the referred-to object in
1688 * actual_oid. Check again. */
1692 static void write_object_file_prepare(const struct git_hash_algo *algo,
1693 const void *buf, unsigned long len,
1694 const char *type, struct object_id *oid,
1695 char *hdr, int *hdrlen)
1699 /* Generate the header */
1700 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1704 algo->update_fn(&c, hdr, *hdrlen);
1705 algo->update_fn(&c, buf, len);
1706 algo->final_fn(oid->hash, &c);
1710 * Move the just written object into its final resting place.
1712 int finalize_object_file(const char *tmpfile, const char *filename)
1716 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1718 else if (link(tmpfile, filename))
1722 * Coda hack - coda doesn't like cross-directory links,
1723 * so we fall back to a rename, which will mean that it
1724 * won't be able to check collisions, but that's not a
1727 * The same holds for FAT formatted media.
1729 * When this succeeds, we just return. We have nothing
1732 if (ret && ret != EEXIST) {
1734 if (!rename(tmpfile, filename))
1738 unlink_or_warn(tmpfile);
1740 if (ret != EEXIST) {
1741 return error_errno(_("unable to write file %s"), filename);
1743 /* FIXME!!! Collision check here ? */
1747 if (adjust_shared_perm(filename))
1748 return error(_("unable to set permission to '%s'"), filename);
1752 static int write_buffer(int fd, const void *buf, size_t len)
1754 if (write_in_full(fd, buf, len) < 0)
1755 return error_errno(_("file write error"));
1759 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1760 unsigned long len, const char *type,
1761 struct object_id *oid)
1763 char hdr[MAX_HEADER_LEN];
1764 int hdrlen = sizeof(hdr);
1765 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1769 /* Finalize a file on disk, and close it. */
1770 static void close_loose_object(int fd)
1772 if (fsync_object_files)
1773 fsync_or_die(fd, "loose object file");
1775 die_errno(_("error when closing loose object file"));
1778 /* Size of directory component, including the ending '/' */
1779 static inline int directory_size(const char *filename)
1781 const char *s = strrchr(filename, '/');
1784 return s - filename + 1;
1788 * This creates a temporary file in the same directory as the final
1791 * We want to avoid cross-directory filename renames, because those
1792 * can have problems on various filesystems (FAT, NFS, Coda).
1794 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1796 int fd, dirlen = directory_size(filename);
1799 strbuf_add(tmp, filename, dirlen);
1800 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1801 fd = git_mkstemp_mode(tmp->buf, 0444);
1802 if (fd < 0 && dirlen && errno == ENOENT) {
1804 * Make sure the directory exists; note that the contents
1805 * of the buffer are undefined after mkstemp returns an
1806 * error, so we have to rewrite the whole buffer from
1810 strbuf_add(tmp, filename, dirlen - 1);
1811 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1813 if (adjust_shared_perm(tmp->buf))
1817 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1818 fd = git_mkstemp_mode(tmp->buf, 0444);
1823 static int write_loose_object(const struct object_id *oid, char *hdr,
1824 int hdrlen, const void *buf, unsigned long len,
1828 unsigned char compressed[4096];
1831 struct object_id parano_oid;
1832 static struct strbuf tmp_file = STRBUF_INIT;
1833 static struct strbuf filename = STRBUF_INIT;
1835 loose_object_path(the_repository, &filename, oid);
1837 fd = create_tmpfile(&tmp_file, filename.buf);
1839 if (errno == EACCES)
1840 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1842 return error_errno(_("unable to create temporary file"));
1846 git_deflate_init(&stream, zlib_compression_level);
1847 stream.next_out = compressed;
1848 stream.avail_out = sizeof(compressed);
1849 the_hash_algo->init_fn(&c);
1851 /* First header.. */
1852 stream.next_in = (unsigned char *)hdr;
1853 stream.avail_in = hdrlen;
1854 while (git_deflate(&stream, 0) == Z_OK)
1856 the_hash_algo->update_fn(&c, hdr, hdrlen);
1858 /* Then the data itself.. */
1859 stream.next_in = (void *)buf;
1860 stream.avail_in = len;
1862 unsigned char *in0 = stream.next_in;
1863 ret = git_deflate(&stream, Z_FINISH);
1864 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1865 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1866 die(_("unable to write loose object file"));
1867 stream.next_out = compressed;
1868 stream.avail_out = sizeof(compressed);
1869 } while (ret == Z_OK);
1871 if (ret != Z_STREAM_END)
1872 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1874 ret = git_deflate_end_gently(&stream);
1876 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1878 the_hash_algo->final_fn(parano_oid.hash, &c);
1879 if (!oideq(oid, ¶no_oid))
1880 die(_("confused by unstable object source data for %s"),
1883 close_loose_object(fd);
1888 utb.modtime = mtime;
1889 if (utime(tmp_file.buf, &utb) < 0)
1890 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1893 return finalize_object_file(tmp_file.buf, filename.buf);
1896 static int freshen_loose_object(const struct object_id *oid)
1898 return check_and_freshen(oid, 1);
1901 static int freshen_packed_object(const struct object_id *oid)
1903 struct pack_entry e;
1904 if (!find_pack_entry(the_repository, oid, &e))
1908 if (!freshen_file(e.p->pack_name))
1914 int write_object_file(const void *buf, unsigned long len, const char *type,
1915 struct object_id *oid)
1917 char hdr[MAX_HEADER_LEN];
1918 int hdrlen = sizeof(hdr);
1920 /* Normally if we have it in the pack then we do not bother writing
1921 * it out into .git/objects/??/?{38} file.
1923 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1925 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1927 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1930 int hash_object_file_literally(const void *buf, unsigned long len,
1931 const char *type, struct object_id *oid,
1935 int hdrlen, status = 0;
1937 /* type string, SP, %lu of the length plus NUL must fit this */
1938 hdrlen = strlen(type) + MAX_HEADER_LEN;
1939 header = xmalloc(hdrlen);
1940 write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1943 if (!(flags & HASH_WRITE_OBJECT))
1945 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1947 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1954 int force_object_loose(const struct object_id *oid, time_t mtime)
1958 enum object_type type;
1959 char hdr[MAX_HEADER_LEN];
1963 if (has_loose_object(oid))
1965 buf = read_object(the_repository, oid, &type, &len);
1967 return error(_("cannot read object for %s"), oid_to_hex(oid));
1968 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1969 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1975 int repo_has_object_file_with_flags(struct repository *r,
1976 const struct object_id *oid, int flags)
1978 if (!startup_info->have_repository)
1980 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
1983 int repo_has_object_file(struct repository *r,
1984 const struct object_id *oid)
1986 return repo_has_object_file_with_flags(r, oid, 0);
1989 static void check_tree(const void *buf, size_t size)
1991 struct tree_desc desc;
1992 struct name_entry entry;
1994 init_tree_desc(&desc, buf, size);
1995 while (tree_entry(&desc, &entry))
1997 * tree_entry() will die() on malformed entries */
2001 static void check_commit(const void *buf, size_t size)
2004 memset(&c, 0, sizeof(c));
2005 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2006 die(_("corrupt commit"));
2009 static void check_tag(const void *buf, size_t size)
2012 memset(&t, 0, sizeof(t));
2013 if (parse_tag_buffer(the_repository, &t, buf, size))
2014 die(_("corrupt tag"));
2017 static int index_mem(struct index_state *istate,
2018 struct object_id *oid, void *buf, size_t size,
2019 enum object_type type,
2020 const char *path, unsigned flags)
2022 int ret, re_allocated = 0;
2023 int write_object = flags & HASH_WRITE_OBJECT;
2029 * Convert blobs to git internal format
2031 if ((type == OBJ_BLOB) && path) {
2032 struct strbuf nbuf = STRBUF_INIT;
2033 if (convert_to_git(istate, path, buf, size, &nbuf,
2034 get_conv_flags(flags))) {
2035 buf = strbuf_detach(&nbuf, &size);
2039 if (flags & HASH_FORMAT_CHECK) {
2040 if (type == OBJ_TREE)
2041 check_tree(buf, size);
2042 if (type == OBJ_COMMIT)
2043 check_commit(buf, size);
2044 if (type == OBJ_TAG)
2045 check_tag(buf, size);
2049 ret = write_object_file(buf, size, type_name(type), oid);
2051 ret = hash_object_file(the_hash_algo, buf, size,
2052 type_name(type), oid);
2058 static int index_stream_convert_blob(struct index_state *istate,
2059 struct object_id *oid,
2065 const int write_object = flags & HASH_WRITE_OBJECT;
2066 struct strbuf sbuf = STRBUF_INIT;
2069 assert(would_convert_to_git_filter_fd(istate, path));
2071 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2072 get_conv_flags(flags));
2075 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2078 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2079 type_name(OBJ_BLOB), oid);
2080 strbuf_release(&sbuf);
2084 static int index_pipe(struct index_state *istate, struct object_id *oid,
2085 int fd, enum object_type type,
2086 const char *path, unsigned flags)
2088 struct strbuf sbuf = STRBUF_INIT;
2091 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2092 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2095 strbuf_release(&sbuf);
2099 #define SMALL_FILE_SIZE (32*1024)
2101 static int index_core(struct index_state *istate,
2102 struct object_id *oid, int fd, size_t size,
2103 enum object_type type, const char *path,
2109 ret = index_mem(istate, oid, "", size, type, path, flags);
2110 } else if (size <= SMALL_FILE_SIZE) {
2111 char *buf = xmalloc(size);
2112 ssize_t read_result = read_in_full(fd, buf, size);
2113 if (read_result < 0)
2114 ret = error_errno(_("read error while indexing %s"),
2115 path ? path : "<unknown>");
2116 else if (read_result != size)
2117 ret = error(_("short read while indexing %s"),
2118 path ? path : "<unknown>");
2120 ret = index_mem(istate, oid, buf, size, type, path, flags);
2123 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2124 ret = index_mem(istate, oid, buf, size, type, path, flags);
2131 * This creates one packfile per large blob unless bulk-checkin
2132 * machinery is "plugged".
2134 * This also bypasses the usual "convert-to-git" dance, and that is on
2135 * purpose. We could write a streaming version of the converting
2136 * functions and insert that before feeding the data to fast-import
2137 * (or equivalent in-core API described above). However, that is
2138 * somewhat complicated, as we do not know the size of the filter
2139 * result, which we need to know beforehand when writing a git object.
2140 * Since the primary motivation for trying to stream from the working
2141 * tree file and to avoid mmaping it in core is to deal with large
2142 * binary blobs, they generally do not want to get any conversion, and
2143 * callers should avoid this code path when filters are requested.
2145 static int index_stream(struct object_id *oid, int fd, size_t size,
2146 enum object_type type, const char *path,
2149 return index_bulk_checkin(oid, fd, size, type, path, flags);
2152 int index_fd(struct index_state *istate, struct object_id *oid,
2153 int fd, struct stat *st,
2154 enum object_type type, const char *path, unsigned flags)
2159 * Call xsize_t() only when needed to avoid potentially unnecessary
2160 * die() for large files.
2162 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2163 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2164 else if (!S_ISREG(st->st_mode))
2165 ret = index_pipe(istate, oid, fd, type, path, flags);
2166 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2167 (path && would_convert_to_git(istate, path)))
2168 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2171 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2177 int index_path(struct index_state *istate, struct object_id *oid,
2178 const char *path, struct stat *st, unsigned flags)
2181 struct strbuf sb = STRBUF_INIT;
2184 switch (st->st_mode & S_IFMT) {
2186 fd = open(path, O_RDONLY);
2188 return error_errno("open(\"%s\")", path);
2189 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2190 return error(_("%s: failed to insert into database"),
2194 if (strbuf_readlink(&sb, path, st->st_size))
2195 return error_errno("readlink(\"%s\")", path);
2196 if (!(flags & HASH_WRITE_OBJECT))
2197 hash_object_file(the_hash_algo, sb.buf, sb.len,
2199 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2200 rc = error(_("%s: failed to insert into database"), path);
2201 strbuf_release(&sb);
2204 return resolve_gitlink_ref(path, "HEAD", oid);
2206 return error(_("%s: unsupported file type"), path);
2211 int read_pack_header(int fd, struct pack_header *header)
2213 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2214 /* "eof before pack header was fully read" */
2215 return PH_ERROR_EOF;
2217 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2218 /* "protocol error (pack signature mismatch detected)" */
2219 return PH_ERROR_PACK_SIGNATURE;
2220 if (!pack_version_ok(header->hdr_version))
2221 /* "protocol error (pack version unsupported)" */
2222 return PH_ERROR_PROTOCOL;
2226 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2228 enum object_type type = oid_object_info(the_repository, oid, NULL);
2230 die(_("%s is not a valid object"), oid_to_hex(oid));
2232 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2236 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2237 struct strbuf *path,
2238 each_loose_object_fn obj_cb,
2239 each_loose_cruft_fn cruft_cb,
2240 each_loose_subdir_fn subdir_cb,
2243 size_t origlen, baselen;
2247 struct object_id oid;
2249 if (subdir_nr > 0xff)
2250 BUG("invalid loose object subdirectory: %x", subdir_nr);
2252 origlen = path->len;
2253 strbuf_complete(path, '/');
2254 strbuf_addf(path, "%02x", subdir_nr);
2256 dir = opendir(path->buf);
2258 if (errno != ENOENT)
2259 r = error_errno(_("unable to open %s"), path->buf);
2260 strbuf_setlen(path, origlen);
2264 oid.hash[0] = subdir_nr;
2265 strbuf_addch(path, '/');
2266 baselen = path->len;
2268 while ((de = readdir(dir))) {
2270 if (is_dot_or_dotdot(de->d_name))
2273 namelen = strlen(de->d_name);
2274 strbuf_setlen(path, baselen);
2275 strbuf_add(path, de->d_name, namelen);
2276 if (namelen == the_hash_algo->hexsz - 2 &&
2277 !hex_to_bytes(oid.hash + 1, de->d_name,
2278 the_hash_algo->rawsz - 1)) {
2280 r = obj_cb(&oid, path->buf, data);
2288 r = cruft_cb(de->d_name, path->buf, data);
2295 strbuf_setlen(path, baselen - 1);
2296 if (!r && subdir_cb)
2297 r = subdir_cb(subdir_nr, path->buf, data);
2299 strbuf_setlen(path, origlen);
2304 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2305 each_loose_object_fn obj_cb,
2306 each_loose_cruft_fn cruft_cb,
2307 each_loose_subdir_fn subdir_cb,
2313 for (i = 0; i < 256; i++) {
2314 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2323 int for_each_loose_file_in_objdir(const char *path,
2324 each_loose_object_fn obj_cb,
2325 each_loose_cruft_fn cruft_cb,
2326 each_loose_subdir_fn subdir_cb,
2329 struct strbuf buf = STRBUF_INIT;
2332 strbuf_addstr(&buf, path);
2333 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2335 strbuf_release(&buf);
2340 int for_each_loose_object(each_loose_object_fn cb, void *data,
2341 enum for_each_object_flags flags)
2343 struct object_directory *odb;
2345 prepare_alt_odb(the_repository);
2346 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2347 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2352 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2359 static int append_loose_object(const struct object_id *oid, const char *path,
2362 oid_array_append(data, oid);
2366 struct oid_array *odb_loose_cache(struct object_directory *odb,
2367 const struct object_id *oid)
2369 int subdir_nr = oid->hash[0];
2370 struct strbuf buf = STRBUF_INIT;
2372 if (subdir_nr < 0 ||
2373 subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2374 BUG("subdir_nr out of range");
2376 if (odb->loose_objects_subdir_seen[subdir_nr])
2377 return &odb->loose_objects_cache[subdir_nr];
2379 strbuf_addstr(&buf, odb->path);
2380 for_each_file_in_obj_subdir(subdir_nr, &buf,
2381 append_loose_object,
2383 &odb->loose_objects_cache[subdir_nr]);
2384 odb->loose_objects_subdir_seen[subdir_nr] = 1;
2385 strbuf_release(&buf);
2386 return &odb->loose_objects_cache[subdir_nr];
2389 void odb_clear_loose_cache(struct object_directory *odb)
2393 for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2394 oid_array_clear(&odb->loose_objects_cache[i]);
2395 memset(&odb->loose_objects_subdir_seen, 0,
2396 sizeof(odb->loose_objects_subdir_seen));
2399 static int check_stream_oid(git_zstream *stream,
2403 const struct object_id *expected_oid)
2406 struct object_id real_oid;
2407 unsigned char buf[4096];
2408 unsigned long total_read;
2411 the_hash_algo->init_fn(&c);
2412 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2415 * We already read some bytes into hdr, but the ones up to the NUL
2416 * do not count against the object's content size.
2418 total_read = stream->total_out - strlen(hdr) - 1;
2421 * This size comparison must be "<=" to read the final zlib packets;
2422 * see the comment in unpack_loose_rest for details.
2424 while (total_read <= size &&
2426 (status == Z_BUF_ERROR && !stream->avail_out))) {
2427 stream->next_out = buf;
2428 stream->avail_out = sizeof(buf);
2429 if (size - total_read < stream->avail_out)
2430 stream->avail_out = size - total_read;
2431 status = git_inflate(stream, Z_FINISH);
2432 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2433 total_read += stream->next_out - buf;
2435 git_inflate_end(stream);
2437 if (status != Z_STREAM_END) {
2438 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2441 if (stream->avail_in) {
2442 error(_("garbage at end of loose object '%s'"),
2443 oid_to_hex(expected_oid));
2447 the_hash_algo->final_fn(real_oid.hash, &c);
2448 if (!oideq(expected_oid, &real_oid)) {
2449 error(_("hash mismatch for %s (expected %s)"), path,
2450 oid_to_hex(expected_oid));
2457 int read_loose_object(const char *path,
2458 const struct object_id *expected_oid,
2459 enum object_type *type,
2460 unsigned long *size,
2465 unsigned long mapsize;
2467 char hdr[MAX_HEADER_LEN];
2471 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2473 error_errno(_("unable to mmap %s"), path);
2477 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2478 error(_("unable to unpack header of %s"), path);
2482 *type = parse_loose_header(hdr, size);
2484 error(_("unable to parse header of %s"), path);
2485 git_inflate_end(&stream);
2489 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2490 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2493 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2495 error(_("unable to unpack contents of %s"), path);
2496 git_inflate_end(&stream);
2499 if (check_object_signature(the_repository, expected_oid,
2501 type_name(*type))) {
2502 error(_("hash mismatch for %s (expected %s)"), path,
2503 oid_to_hex(expected_oid));
2509 ret = 0; /* everything checks out */
2513 munmap(map, mapsize);