2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git object files - packing, unpacking,
9 #include "git-compat-util.h"
15 #include "string-list.h"
21 #include "run-command.h"
24 #include "tree-walk.h"
26 #include "pack-revindex.h"
27 #include "hash-lookup.h"
28 #include "bulk-checkin.h"
29 #include "repository.h"
30 #include "replace-object.h"
31 #include "streaming.h"
34 #include "mergesort.h"
37 #include "object-store.h"
38 #include "promisor-remote.h"
39 #include "submodule.h"
42 /* The maximum size for an object header. */
43 #define MAX_HEADER_LEN 32
46 #define EMPTY_TREE_SHA1_BIN_LITERAL \
47 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
48 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
49 #define EMPTY_TREE_SHA256_BIN_LITERAL \
50 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
51 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
52 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
55 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
56 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
57 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
58 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
59 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
60 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
61 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
64 static const struct object_id empty_tree_oid
= {
65 .hash
= EMPTY_TREE_SHA1_BIN_LITERAL
,
66 .algo
= GIT_HASH_SHA1
,
68 static const struct object_id empty_blob_oid
= {
69 .hash
= EMPTY_BLOB_SHA1_BIN_LITERAL
,
70 .algo
= GIT_HASH_SHA1
,
72 static const struct object_id null_oid_sha1
= {
74 .algo
= GIT_HASH_SHA1
,
76 static const struct object_id empty_tree_oid_sha256
= {
77 .hash
= EMPTY_TREE_SHA256_BIN_LITERAL
,
78 .algo
= GIT_HASH_SHA256
,
80 static const struct object_id empty_blob_oid_sha256
= {
81 .hash
= EMPTY_BLOB_SHA256_BIN_LITERAL
,
82 .algo
= GIT_HASH_SHA256
,
84 static const struct object_id null_oid_sha256
= {
86 .algo
= GIT_HASH_SHA256
,
89 static void git_hash_sha1_init(git_hash_ctx
*ctx
)
91 git_SHA1_Init(&ctx
->sha1
);
94 static void git_hash_sha1_clone(git_hash_ctx
*dst
, const git_hash_ctx
*src
)
96 git_SHA1_Clone(&dst
->sha1
, &src
->sha1
);
99 static void git_hash_sha1_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
101 git_SHA1_Update(&ctx
->sha1
, data
, len
);
104 static void git_hash_sha1_final(unsigned char *hash
, git_hash_ctx
*ctx
)
106 git_SHA1_Final(hash
, &ctx
->sha1
);
109 static void git_hash_sha1_final_oid(struct object_id
*oid
, git_hash_ctx
*ctx
)
111 git_SHA1_Final(oid
->hash
, &ctx
->sha1
);
112 memset(oid
->hash
+ GIT_SHA1_RAWSZ
, 0, GIT_MAX_RAWSZ
- GIT_SHA1_RAWSZ
);
113 oid
->algo
= GIT_HASH_SHA1
;
117 static void git_hash_sha256_init(git_hash_ctx
*ctx
)
119 git_SHA256_Init(&ctx
->sha256
);
122 static void git_hash_sha256_clone(git_hash_ctx
*dst
, const git_hash_ctx
*src
)
124 git_SHA256_Clone(&dst
->sha256
, &src
->sha256
);
127 static void git_hash_sha256_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
129 git_SHA256_Update(&ctx
->sha256
, data
, len
);
132 static void git_hash_sha256_final(unsigned char *hash
, git_hash_ctx
*ctx
)
134 git_SHA256_Final(hash
, &ctx
->sha256
);
137 static void git_hash_sha256_final_oid(struct object_id
*oid
, git_hash_ctx
*ctx
)
139 git_SHA256_Final(oid
->hash
, &ctx
->sha256
);
141 * This currently does nothing, so the compiler should optimize it out,
142 * but keep it in case we extend the hash size again.
144 memset(oid
->hash
+ GIT_SHA256_RAWSZ
, 0, GIT_MAX_RAWSZ
- GIT_SHA256_RAWSZ
);
145 oid
->algo
= GIT_HASH_SHA256
;
148 static void git_hash_unknown_init(git_hash_ctx
*ctx UNUSED
)
150 BUG("trying to init unknown hash");
153 static void git_hash_unknown_clone(git_hash_ctx
*dst UNUSED
,
154 const git_hash_ctx
*src UNUSED
)
156 BUG("trying to clone unknown hash");
159 static void git_hash_unknown_update(git_hash_ctx
*ctx UNUSED
,
160 const void *data UNUSED
,
163 BUG("trying to update unknown hash");
166 static void git_hash_unknown_final(unsigned char *hash UNUSED
,
167 git_hash_ctx
*ctx UNUSED
)
169 BUG("trying to finalize unknown hash");
172 static void git_hash_unknown_final_oid(struct object_id
*oid UNUSED
,
173 git_hash_ctx
*ctx UNUSED
)
175 BUG("trying to finalize unknown hash");
178 const struct git_hash_algo hash_algos
[GIT_HASH_NALGOS
] = {
181 .format_id
= 0x00000000,
185 .init_fn
= git_hash_unknown_init
,
186 .clone_fn
= git_hash_unknown_clone
,
187 .update_fn
= git_hash_unknown_update
,
188 .final_fn
= git_hash_unknown_final
,
189 .final_oid_fn
= git_hash_unknown_final_oid
,
196 .format_id
= GIT_SHA1_FORMAT_ID
,
197 .rawsz
= GIT_SHA1_RAWSZ
,
198 .hexsz
= GIT_SHA1_HEXSZ
,
199 .blksz
= GIT_SHA1_BLKSZ
,
200 .init_fn
= git_hash_sha1_init
,
201 .clone_fn
= git_hash_sha1_clone
,
202 .update_fn
= git_hash_sha1_update
,
203 .final_fn
= git_hash_sha1_final
,
204 .final_oid_fn
= git_hash_sha1_final_oid
,
205 .empty_tree
= &empty_tree_oid
,
206 .empty_blob
= &empty_blob_oid
,
207 .null_oid
= &null_oid_sha1
,
211 .format_id
= GIT_SHA256_FORMAT_ID
,
212 .rawsz
= GIT_SHA256_RAWSZ
,
213 .hexsz
= GIT_SHA256_HEXSZ
,
214 .blksz
= GIT_SHA256_BLKSZ
,
215 .init_fn
= git_hash_sha256_init
,
216 .clone_fn
= git_hash_sha256_clone
,
217 .update_fn
= git_hash_sha256_update
,
218 .final_fn
= git_hash_sha256_final
,
219 .final_oid_fn
= git_hash_sha256_final_oid
,
220 .empty_tree
= &empty_tree_oid_sha256
,
221 .empty_blob
= &empty_blob_oid_sha256
,
222 .null_oid
= &null_oid_sha256
,
226 const struct object_id
*null_oid(void)
228 return the_hash_algo
->null_oid
;
231 const char *empty_tree_oid_hex(void)
233 static char buf
[GIT_MAX_HEXSZ
+ 1];
234 return oid_to_hex_r(buf
, the_hash_algo
->empty_tree
);
237 const char *empty_blob_oid_hex(void)
239 static char buf
[GIT_MAX_HEXSZ
+ 1];
240 return oid_to_hex_r(buf
, the_hash_algo
->empty_blob
);
243 int hash_algo_by_name(const char *name
)
247 return GIT_HASH_UNKNOWN
;
248 for (i
= 1; i
< GIT_HASH_NALGOS
; i
++)
249 if (!strcmp(name
, hash_algos
[i
].name
))
251 return GIT_HASH_UNKNOWN
;
254 int hash_algo_by_id(uint32_t format_id
)
257 for (i
= 1; i
< GIT_HASH_NALGOS
; i
++)
258 if (format_id
== hash_algos
[i
].format_id
)
260 return GIT_HASH_UNKNOWN
;
263 int hash_algo_by_length(int len
)
266 for (i
= 1; i
< GIT_HASH_NALGOS
; i
++)
267 if (len
== hash_algos
[i
].rawsz
)
269 return GIT_HASH_UNKNOWN
;
273 * This is meant to hold a *small* number of objects that you would
274 * want read_object_file() to be able to return, but yet you do not want
275 * to write them into the object store (e.g. a browse-only
278 static struct cached_object
{
279 struct object_id oid
;
280 enum object_type type
;
284 static int cached_object_nr
, cached_object_alloc
;
286 static struct cached_object empty_tree
= {
288 .hash
= EMPTY_TREE_SHA1_BIN_LITERAL
,
294 static struct cached_object
*find_cached_object(const struct object_id
*oid
)
297 struct cached_object
*co
= cached_objects
;
299 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
300 if (oideq(&co
->oid
, oid
))
303 if (oideq(oid
, the_hash_algo
->empty_tree
))
309 static int get_conv_flags(unsigned flags
)
311 if (flags
& HASH_RENORMALIZE
)
312 return CONV_EOL_RENORMALIZE
;
313 else if (flags
& HASH_WRITE_OBJECT
)
314 return global_conv_flags_eol
| CONV_WRITE_OBJECT
;
320 int mkdir_in_gitdir(const char *path
)
322 if (mkdir(path
, 0777)) {
323 int saved_errno
= errno
;
325 struct strbuf sb
= STRBUF_INIT
;
330 * Are we looking at a path in a symlinked worktree
331 * whose original repository does not yet have it?
332 * e.g. .git/rr-cache pointing at its original
333 * repository in which the user hasn't performed any
334 * conflict resolution yet?
336 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
337 strbuf_readlink(&sb
, path
, st
.st_size
) ||
338 !is_absolute_path(sb
.buf
) ||
339 mkdir(sb
.buf
, 0777)) {
346 return adjust_shared_perm(path
);
349 static enum scld_error
safe_create_leading_directories_1(char *path
, int share
)
351 char *next_component
= path
+ offset_1st_component(path
);
352 enum scld_error ret
= SCLD_OK
;
354 while (ret
== SCLD_OK
&& next_component
) {
356 char *slash
= next_component
, slash_character
;
358 while (*slash
&& !is_dir_sep(*slash
))
364 next_component
= slash
+ 1;
365 while (is_dir_sep(*next_component
))
367 if (!*next_component
)
370 slash_character
= *slash
;
372 if (!stat(path
, &st
)) {
374 if (!S_ISDIR(st
.st_mode
)) {
378 } else if (mkdir(path
, 0777)) {
379 if (errno
== EEXIST
&&
380 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
381 ; /* somebody created it since we checked */
382 else if (errno
== ENOENT
)
384 * Either mkdir() failed because
385 * somebody just pruned the containing
386 * directory, or stat() failed because
387 * the file that was in our way was
388 * just removed. Either way, inform
389 * the caller that it might be worth
395 } else if (share
&& adjust_shared_perm(path
)) {
398 *slash
= slash_character
;
403 enum scld_error
safe_create_leading_directories(char *path
)
405 return safe_create_leading_directories_1(path
, 1);
408 enum scld_error
safe_create_leading_directories_no_share(char *path
)
410 return safe_create_leading_directories_1(path
, 0);
413 enum scld_error
safe_create_leading_directories_const(const char *path
)
416 /* path points to cache entries, so xstrdup before messing with it */
417 char *buf
= xstrdup(path
);
418 enum scld_error result
= safe_create_leading_directories(buf
);
426 static void fill_loose_path(struct strbuf
*buf
, const struct object_id
*oid
)
429 for (i
= 0; i
< the_hash_algo
->rawsz
; i
++) {
430 static char hex
[] = "0123456789abcdef";
431 unsigned int val
= oid
->hash
[i
];
432 strbuf_addch(buf
, hex
[val
>> 4]);
433 strbuf_addch(buf
, hex
[val
& 0xf]);
435 strbuf_addch(buf
, '/');
439 static const char *odb_loose_path(struct object_directory
*odb
,
441 const struct object_id
*oid
)
444 strbuf_addstr(buf
, odb
->path
);
445 strbuf_addch(buf
, '/');
446 fill_loose_path(buf
, oid
);
450 const char *loose_object_path(struct repository
*r
, struct strbuf
*buf
,
451 const struct object_id
*oid
)
453 return odb_loose_path(r
->objects
->odb
, buf
, oid
);
457 * Return non-zero iff the path is usable as an alternate object database.
459 static int alt_odb_usable(struct raw_object_store
*o
,
461 const char *normalized_objdir
, khiter_t
*pos
)
465 /* Detect cases where alternate disappeared */
466 if (!is_directory(path
->buf
)) {
467 error(_("object directory %s does not exist; "
468 "check .git/objects/info/alternates"),
474 * Prevent the common mistake of listing the same
475 * thing twice, or object directory itself.
477 if (!o
->odb_by_path
) {
480 o
->odb_by_path
= kh_init_odb_path_map();
481 assert(!o
->odb
->next
);
482 p
= kh_put_odb_path_map(o
->odb_by_path
, o
->odb
->path
, &r
);
483 assert(r
== 1); /* never used */
484 kh_value(o
->odb_by_path
, p
) = o
->odb
;
486 if (fspatheq(path
->buf
, normalized_objdir
))
488 *pos
= kh_put_odb_path_map(o
->odb_by_path
, path
->buf
, &r
);
489 /* r: 0 = exists, 1 = never used, 2 = deleted */
490 return r
== 0 ? 0 : 1;
494 * Prepare alternate object database registry.
496 * The variable alt_odb_list points at the list of struct
497 * object_directory. The elements on this list come from
498 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
499 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
500 * whose contents is similar to that environment variable but can be
501 * LF separated. Its base points at a statically allocated buffer that
502 * contains "/the/directory/corresponding/to/.git/objects/...", while
503 * its name points just after the slash at the end of ".git/objects/"
504 * in the example above, and has enough space to hold all hex characters
505 * of the object ID, an extra slash for the first level indirection, and
506 * the terminating NUL.
508 static void read_info_alternates(struct repository
*r
,
509 const char *relative_base
,
511 static int link_alt_odb_entry(struct repository
*r
, const struct strbuf
*entry
,
512 const char *relative_base
, int depth
, const char *normalized_objdir
)
514 struct object_directory
*ent
;
515 struct strbuf pathbuf
= STRBUF_INIT
;
516 struct strbuf tmp
= STRBUF_INIT
;
520 if (!is_absolute_path(entry
->buf
) && relative_base
) {
521 strbuf_realpath(&pathbuf
, relative_base
, 1);
522 strbuf_addch(&pathbuf
, '/');
524 strbuf_addbuf(&pathbuf
, entry
);
526 if (!strbuf_realpath(&tmp
, pathbuf
.buf
, 0)) {
527 error(_("unable to normalize alternate object path: %s"),
531 strbuf_swap(&pathbuf
, &tmp
);
534 * The trailing slash after the directory name is given by
535 * this function at the end. Remove duplicates.
537 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
538 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
540 if (!alt_odb_usable(r
->objects
, &pathbuf
, normalized_objdir
, &pos
))
543 CALLOC_ARRAY(ent
, 1);
544 /* pathbuf.buf is already in r->objects->odb_by_path */
545 ent
->path
= strbuf_detach(&pathbuf
, NULL
);
547 /* add the alternate entry */
548 *r
->objects
->odb_tail
= ent
;
549 r
->objects
->odb_tail
= &(ent
->next
);
551 assert(r
->objects
->odb_by_path
);
552 kh_value(r
->objects
->odb_by_path
, pos
) = ent
;
554 /* recursively add alternates */
555 read_info_alternates(r
, ent
->path
, depth
+ 1);
558 strbuf_release(&tmp
);
559 strbuf_release(&pathbuf
);
563 static const char *parse_alt_odb_entry(const char *string
,
571 if (*string
== '#') {
572 /* comment; consume up to next separator */
573 end
= strchrnul(string
, sep
);
574 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
576 * quoted path; unquote_c_style has copied the
577 * data for us and set "end". Broken quoting (e.g.,
578 * an entry that doesn't end with a quote) falls
579 * back to the unquoted case below.
582 /* normal, unquoted path */
583 end
= strchrnul(string
, sep
);
584 strbuf_add(out
, string
, end
- string
);
592 static void link_alt_odb_entries(struct repository
*r
, const char *alt
,
593 int sep
, const char *relative_base
, int depth
)
595 struct strbuf objdirbuf
= STRBUF_INIT
;
596 struct strbuf entry
= STRBUF_INIT
;
602 error(_("%s: ignoring alternate object stores, nesting too deep"),
607 strbuf_realpath(&objdirbuf
, r
->objects
->odb
->path
, 1);
610 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
613 link_alt_odb_entry(r
, &entry
,
614 relative_base
, depth
, objdirbuf
.buf
);
616 strbuf_release(&entry
);
617 strbuf_release(&objdirbuf
);
620 static void read_info_alternates(struct repository
*r
,
621 const char *relative_base
,
625 struct strbuf buf
= STRBUF_INIT
;
627 path
= xstrfmt("%s/info/alternates", relative_base
);
628 if (strbuf_read_file(&buf
, path
, 1024) < 0) {
629 warn_on_fopen_errors(path
);
634 link_alt_odb_entries(r
, buf
.buf
, '\n', relative_base
, depth
);
635 strbuf_release(&buf
);
639 void add_to_alternates_file(const char *reference
)
641 struct lock_file lock
= LOCK_INIT
;
642 char *alts
= git_pathdup("objects/info/alternates");
646 hold_lock_file_for_update(&lock
, alts
, LOCK_DIE_ON_ERROR
);
647 out
= fdopen_lock_file(&lock
, "w");
649 die_errno(_("unable to fdopen alternates lockfile"));
651 in
= fopen(alts
, "r");
653 struct strbuf line
= STRBUF_INIT
;
655 while (strbuf_getline(&line
, in
) != EOF
) {
656 if (!strcmp(reference
, line
.buf
)) {
660 fprintf_or_die(out
, "%s\n", line
.buf
);
663 strbuf_release(&line
);
666 else if (errno
!= ENOENT
)
667 die_errno(_("unable to read alternates file"));
670 rollback_lock_file(&lock
);
672 fprintf_or_die(out
, "%s\n", reference
);
673 if (commit_lock_file(&lock
))
674 die_errno(_("unable to move new alternates file into place"));
675 if (the_repository
->objects
->loaded_alternates
)
676 link_alt_odb_entries(the_repository
, reference
,
682 void add_to_alternates_memory(const char *reference
)
685 * Make sure alternates are initialized, or else our entry may be
686 * overwritten when they are.
688 prepare_alt_odb(the_repository
);
690 link_alt_odb_entries(the_repository
, reference
,
694 struct object_directory
*set_temporary_primary_odb(const char *dir
, int will_destroy
)
696 struct object_directory
*new_odb
;
699 * Make sure alternates are initialized, or else our entry may be
700 * overwritten when they are.
702 prepare_alt_odb(the_repository
);
705 * Make a new primary odb and link the old primary ODB in as an
708 new_odb
= xcalloc(1, sizeof(*new_odb
));
709 new_odb
->path
= xstrdup(dir
);
712 * Disable ref updates while a temporary odb is active, since
713 * the objects in the database may roll back.
715 new_odb
->disable_ref_updates
= 1;
716 new_odb
->will_destroy
= will_destroy
;
717 new_odb
->next
= the_repository
->objects
->odb
;
718 the_repository
->objects
->odb
= new_odb
;
719 return new_odb
->next
;
722 void restore_primary_odb(struct object_directory
*restore_odb
, const char *old_path
)
724 struct object_directory
*cur_odb
= the_repository
->objects
->odb
;
726 if (strcmp(old_path
, cur_odb
->path
))
727 BUG("expected %s as primary object store; found %s",
728 old_path
, cur_odb
->path
);
730 if (cur_odb
->next
!= restore_odb
)
731 BUG("we expect the old primary object store to be the first alternate");
733 the_repository
->objects
->odb
= restore_odb
;
734 free_object_directory(cur_odb
);
738 * Compute the exact path an alternate is at and returns it. In case of
739 * error NULL is returned and the human readable error is added to `err`
740 * `path` may be relative and should point to $GIT_DIR.
741 * `err` must not be null.
743 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
745 char *ref_git
= NULL
;
749 ref_git
= real_pathdup(path
, 0);
752 strbuf_addf(err
, _("path '%s' does not exist"), path
);
756 repo
= read_gitfile(ref_git
);
758 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
761 ref_git
= xstrdup(repo
);
764 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
765 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
767 ref_git
= ref_git_git
;
768 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
769 struct strbuf sb
= STRBUF_INIT
;
771 if (get_common_dir(&sb
, ref_git
)) {
773 _("reference repository '%s' as a linked "
774 "checkout is not supported yet."),
779 strbuf_addf(err
, _("reference repository '%s' is not a "
780 "local repository."), path
);
784 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
785 strbuf_addf(err
, _("reference repository '%s' is shallow"),
791 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
793 _("reference repository '%s' is grafted"),
801 FREE_AND_NULL(ref_git
);
807 struct object_directory
*find_odb(struct repository
*r
, const char *obj_dir
)
809 struct object_directory
*odb
;
810 char *obj_dir_real
= real_pathdup(obj_dir
, 1);
811 struct strbuf odb_path_real
= STRBUF_INIT
;
814 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
815 strbuf_realpath(&odb_path_real
, odb
->path
, 1);
816 if (!strcmp(obj_dir_real
, odb_path_real
.buf
))
821 strbuf_release(&odb_path_real
);
824 die(_("could not find object directory matching %s"), obj_dir
);
828 static void fill_alternate_refs_command(struct child_process
*cmd
,
829 const char *repo_path
)
833 if (!git_config_get_value("core.alternateRefsCommand", &value
)) {
836 strvec_push(&cmd
->args
, value
);
837 strvec_push(&cmd
->args
, repo_path
);
841 strvec_pushf(&cmd
->args
, "--git-dir=%s", repo_path
);
842 strvec_push(&cmd
->args
, "for-each-ref");
843 strvec_push(&cmd
->args
, "--format=%(objectname)");
845 if (!git_config_get_value("core.alternateRefsPrefixes", &value
)) {
846 strvec_push(&cmd
->args
, "--");
847 strvec_split(&cmd
->args
, value
);
851 strvec_pushv(&cmd
->env
, (const char **)local_repo_env
);
855 static void read_alternate_refs(const char *path
,
856 alternate_ref_fn
*cb
,
859 struct child_process cmd
= CHILD_PROCESS_INIT
;
860 struct strbuf line
= STRBUF_INIT
;
863 fill_alternate_refs_command(&cmd
, path
);
865 if (start_command(&cmd
))
868 fh
= xfdopen(cmd
.out
, "r");
869 while (strbuf_getline_lf(&line
, fh
) != EOF
) {
870 struct object_id oid
;
873 if (parse_oid_hex(line
.buf
, &oid
, &p
) || *p
) {
874 warning(_("invalid line while parsing alternate refs: %s"),
883 finish_command(&cmd
);
884 strbuf_release(&line
);
887 struct alternate_refs_data
{
888 alternate_ref_fn
*fn
;
892 static int refs_from_alternate_cb(struct object_directory
*e
,
895 struct strbuf path
= STRBUF_INIT
;
897 struct alternate_refs_data
*cb
= data
;
899 if (!strbuf_realpath(&path
, e
->path
, 0))
901 if (!strbuf_strip_suffix(&path
, "/objects"))
905 /* Is this a git repository with refs? */
906 strbuf_addstr(&path
, "/refs");
907 if (!is_directory(path
.buf
))
909 strbuf_setlen(&path
, base_len
);
911 read_alternate_refs(path
.buf
, cb
->fn
, cb
->data
);
914 strbuf_release(&path
);
918 void for_each_alternate_ref(alternate_ref_fn fn
, void *data
)
920 struct alternate_refs_data cb
;
923 foreach_alt_odb(refs_from_alternate_cb
, &cb
);
926 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
928 struct object_directory
*ent
;
931 prepare_alt_odb(the_repository
);
932 for (ent
= the_repository
->objects
->odb
->next
; ent
; ent
= ent
->next
) {
940 void prepare_alt_odb(struct repository
*r
)
942 if (r
->objects
->loaded_alternates
)
945 link_alt_odb_entries(r
, r
->objects
->alternate_db
, PATH_SEP
, NULL
, 0);
947 read_info_alternates(r
, r
->objects
->odb
->path
, 0);
948 r
->objects
->loaded_alternates
= 1;
951 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
952 static int freshen_file(const char *fn
)
954 return !utime(fn
, NULL
);
958 * All of the check_and_freshen functions return 1 if the file exists and was
959 * freshened (if freshening was requested), 0 otherwise. If they return
960 * 0, you should not assume that it is safe to skip a write of the object (it
961 * either does not exist on disk, or has a stale mtime and may be subject to
964 int check_and_freshen_file(const char *fn
, int freshen
)
966 if (access(fn
, F_OK
))
968 if (freshen
&& !freshen_file(fn
))
973 static int check_and_freshen_odb(struct object_directory
*odb
,
974 const struct object_id
*oid
,
977 static struct strbuf path
= STRBUF_INIT
;
978 odb_loose_path(odb
, &path
, oid
);
979 return check_and_freshen_file(path
.buf
, freshen
);
982 static int check_and_freshen_local(const struct object_id
*oid
, int freshen
)
984 return check_and_freshen_odb(the_repository
->objects
->odb
, oid
, freshen
);
987 static int check_and_freshen_nonlocal(const struct object_id
*oid
, int freshen
)
989 struct object_directory
*odb
;
991 prepare_alt_odb(the_repository
);
992 for (odb
= the_repository
->objects
->odb
->next
; odb
; odb
= odb
->next
) {
993 if (check_and_freshen_odb(odb
, oid
, freshen
))
999 static int check_and_freshen(const struct object_id
*oid
, int freshen
)
1001 return check_and_freshen_local(oid
, freshen
) ||
1002 check_and_freshen_nonlocal(oid
, freshen
);
1005 int has_loose_object_nonlocal(const struct object_id
*oid
)
1007 return check_and_freshen_nonlocal(oid
, 0);
1010 int has_loose_object(const struct object_id
*oid
)
1012 return check_and_freshen(oid
, 0);
1015 static void mmap_limit_check(size_t length
)
1017 static size_t limit
= 0;
1019 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
1024 die(_("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
),
1025 (uintmax_t)length
, (uintmax_t)limit
);
1028 void *xmmap_gently(void *start
, size_t length
,
1029 int prot
, int flags
, int fd
, off_t offset
)
1033 mmap_limit_check(length
);
1034 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
1035 if (ret
== MAP_FAILED
&& !length
)
1040 const char *mmap_os_err(void)
1042 static const char blank
[] = "";
1043 #if defined(__linux__)
1044 if (errno
== ENOMEM
) {
1045 /* this continues an existing error message: */
1046 static const char enomem
[] =
1047 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
1050 #endif /* OS-specific bits */
1054 void *xmmap(void *start
, size_t length
,
1055 int prot
, int flags
, int fd
, off_t offset
)
1057 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
1058 if (ret
== MAP_FAILED
)
1059 die_errno(_("mmap failed%s"), mmap_os_err());
1063 static int format_object_header_literally(char *str
, size_t size
,
1064 const char *type
, size_t objsize
)
1066 return xsnprintf(str
, size
, "%s %"PRIuMAX
, type
, (uintmax_t)objsize
) + 1;
1069 int format_object_header(char *str
, size_t size
, enum object_type type
,
1072 const char *name
= type_name(type
);
1075 BUG("could not get a type name for 'enum object_type' value %d", type
);
1077 return format_object_header_literally(str
, size
, name
, objsize
);
1080 int check_object_signature(struct repository
*r
, const struct object_id
*oid
,
1081 void *buf
, unsigned long size
,
1082 enum object_type type
)
1084 struct object_id real_oid
;
1086 hash_object_file(r
->hash_algo
, buf
, size
, type
, &real_oid
);
1088 return !oideq(oid
, &real_oid
) ? -1 : 0;
1091 int stream_object_signature(struct repository
*r
, const struct object_id
*oid
)
1093 struct object_id real_oid
;
1095 enum object_type obj_type
;
1096 struct git_istream
*st
;
1098 char hdr
[MAX_HEADER_LEN
];
1101 st
= open_istream(r
, oid
, &obj_type
, &size
, NULL
);
1105 /* Generate the header */
1106 hdrlen
= format_object_header(hdr
, sizeof(hdr
), obj_type
, size
);
1109 r
->hash_algo
->init_fn(&c
);
1110 r
->hash_algo
->update_fn(&c
, hdr
, hdrlen
);
1112 char buf
[1024 * 16];
1113 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1121 r
->hash_algo
->update_fn(&c
, buf
, readlen
);
1123 r
->hash_algo
->final_oid_fn(&real_oid
, &c
);
1125 return !oideq(oid
, &real_oid
) ? -1 : 0;
1128 int git_open_cloexec(const char *name
, int flags
)
1131 static int o_cloexec
= O_CLOEXEC
;
1133 fd
= open(name
, flags
| o_cloexec
);
1134 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
1135 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1136 o_cloexec
&= ~O_CLOEXEC
;
1137 fd
= open(name
, flags
| o_cloexec
);
1140 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1142 static int fd_cloexec
= FD_CLOEXEC
;
1144 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
1145 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1146 int flags
= fcntl(fd
, F_GETFD
);
1147 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
1156 * Find "oid" as a loose object in the local repository or in an alternate.
1157 * Returns 0 on success, negative on failure.
1159 * The "path" out-parameter will give the path of the object we found (if any).
1160 * Note that it may point to static storage and is only valid until another
1161 * call to stat_loose_object().
1163 static int stat_loose_object(struct repository
*r
, const struct object_id
*oid
,
1164 struct stat
*st
, const char **path
)
1166 struct object_directory
*odb
;
1167 static struct strbuf buf
= STRBUF_INIT
;
1170 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1171 *path
= odb_loose_path(odb
, &buf
, oid
);
1172 if (!lstat(*path
, st
))
1180 * Like stat_loose_object(), but actually open the object and return the
1181 * descriptor. See the caveats on the "path" parameter above.
1183 static int open_loose_object(struct repository
*r
,
1184 const struct object_id
*oid
, const char **path
)
1187 struct object_directory
*odb
;
1188 int most_interesting_errno
= ENOENT
;
1189 static struct strbuf buf
= STRBUF_INIT
;
1192 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1193 *path
= odb_loose_path(odb
, &buf
, oid
);
1194 fd
= git_open(*path
);
1198 if (most_interesting_errno
== ENOENT
)
1199 most_interesting_errno
= errno
;
1201 errno
= most_interesting_errno
;
1205 static int quick_has_loose(struct repository
*r
,
1206 const struct object_id
*oid
)
1208 struct object_directory
*odb
;
1211 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1212 if (oidtree_contains(odb_loose_cache(odb
, oid
), oid
))
1219 * Map and close the given loose object fd. The path argument is used for
1222 static void *map_fd(int fd
, const char *path
, unsigned long *size
)
1227 if (!fstat(fd
, &st
)) {
1228 *size
= xsize_t(st
.st_size
);
1230 /* mmap() is forbidden on empty files */
1231 error(_("object file %s is empty"), path
);
1235 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1241 void *map_loose_object(struct repository
*r
,
1242 const struct object_id
*oid
,
1243 unsigned long *size
)
1246 int fd
= open_loose_object(r
, oid
, &p
);
1250 return map_fd(fd
, p
, size
);
1253 enum unpack_loose_header_result
unpack_loose_header(git_zstream
*stream
,
1255 unsigned long mapsize
,
1257 unsigned long bufsiz
,
1258 struct strbuf
*header
)
1262 /* Get the data stream */
1263 memset(stream
, 0, sizeof(*stream
));
1264 stream
->next_in
= map
;
1265 stream
->avail_in
= mapsize
;
1266 stream
->next_out
= buffer
;
1267 stream
->avail_out
= bufsiz
;
1269 git_inflate_init(stream
);
1271 status
= git_inflate(stream
, 0);
1277 * Check if entire header is unpacked in the first iteration.
1279 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1283 * We have a header longer than MAX_HEADER_LEN. The "header"
1284 * here is only non-NULL when we run "cat-file
1285 * --allow-unknown-type".
1288 return ULHR_TOO_LONG
;
1291 * buffer[0..bufsiz] was not large enough. Copy the partial
1292 * result out to header, and then append the result of further
1293 * reading the stream.
1295 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1296 stream
->next_out
= buffer
;
1297 stream
->avail_out
= bufsiz
;
1301 status
= git_inflate(stream
, 0);
1303 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1304 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1306 stream
->next_out
= buffer
;
1307 stream
->avail_out
= bufsiz
;
1308 } while (status
!= Z_STREAM_END
);
1309 return ULHR_TOO_LONG
;
1312 static void *unpack_loose_rest(git_zstream
*stream
,
1313 void *buffer
, unsigned long size
,
1314 const struct object_id
*oid
)
1316 int bytes
= strlen(buffer
) + 1;
1317 unsigned char *buf
= xmallocz(size
);
1321 n
= stream
->total_out
- bytes
;
1324 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1326 if (bytes
<= size
) {
1328 * The above condition must be (bytes <= size), not
1329 * (bytes < size). In other words, even though we
1330 * expect no more output and set avail_out to zero,
1331 * the input zlib stream may have bytes that express
1332 * "this concludes the stream", and we *do* want to
1335 * Otherwise we would not be able to test that we
1336 * consumed all the input to reach the expected size;
1337 * we also want to check that zlib tells us that all
1338 * went well with status == Z_STREAM_END at the end.
1340 stream
->next_out
= buf
+ bytes
;
1341 stream
->avail_out
= size
- bytes
;
1342 while (status
== Z_OK
) {
1344 status
= git_inflate(stream
, Z_FINISH
);
1348 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1349 git_inflate_end(stream
);
1354 error(_("corrupt loose object '%s'"), oid_to_hex(oid
));
1355 else if (stream
->avail_in
)
1356 error(_("garbage at end of loose object '%s'"),
1363 * We used to just use "sscanf()", but that's actually way
1364 * too permissive for what we want to check. So do an anal
1365 * object header parse by hand.
1367 int parse_loose_header(const char *hdr
, struct object_info
*oi
)
1369 const char *type_buf
= hdr
;
1371 int type
, type_len
= 0;
1374 * The type can be of any size but is followed by
1386 type
= type_from_string_gently(type_buf
, type_len
, 1);
1388 strbuf_add(oi
->type_name
, type_buf
, type_len
);
1393 * The length must follow immediately, and be in canonical
1394 * decimal format (ie "010" is not valid).
1396 size
= *hdr
++ - '0';
1401 unsigned long c
= *hdr
- '0';
1405 size
= st_add(st_mult(size
, 10), c
);
1410 *oi
->sizep
= cast_size_t_to_ulong(size
);
1413 * The length must be followed by a zero byte
1419 * The format is valid, but the type may still be bogus. The
1420 * Caller needs to check its oi->typep.
1425 static int loose_object_info(struct repository
*r
,
1426 const struct object_id
*oid
,
1427 struct object_info
*oi
, int flags
)
1431 unsigned long mapsize
;
1435 char hdr
[MAX_HEADER_LEN
];
1436 struct strbuf hdrbuf
= STRBUF_INIT
;
1437 unsigned long size_scratch
;
1438 enum object_type type_scratch
;
1439 int allow_unknown
= flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
;
1441 if (oi
->delta_base_oid
)
1442 oidclr(oi
->delta_base_oid
);
1445 * If we don't care about type or size, then we don't
1446 * need to look inside the object at all. Note that we
1447 * do not optimize out the stat call, even if the
1448 * caller doesn't care about the disk-size, since our
1449 * return value implicitly indicates whether the
1450 * object even exists.
1452 if (!oi
->typep
&& !oi
->type_name
&& !oi
->sizep
&& !oi
->contentp
) {
1454 if (!oi
->disk_sizep
&& (flags
& OBJECT_INFO_QUICK
))
1455 return quick_has_loose(r
, oid
) ? 0 : -1;
1456 if (stat_loose_object(r
, oid
, &st
, &path
) < 0)
1459 *oi
->disk_sizep
= st
.st_size
;
1463 fd
= open_loose_object(r
, oid
, &path
);
1465 if (errno
!= ENOENT
)
1466 error_errno(_("unable to open loose object %s"), oid_to_hex(oid
));
1469 map
= map_fd(fd
, path
, &mapsize
);
1474 oi
->sizep
= &size_scratch
;
1476 oi
->typep
= &type_scratch
;
1479 *oi
->disk_sizep
= mapsize
;
1481 switch (unpack_loose_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
),
1482 allow_unknown
? &hdrbuf
: NULL
)) {
1484 if (parse_loose_header(hdrbuf
.len
? hdrbuf
.buf
: hdr
, oi
) < 0)
1485 status
= error(_("unable to parse %s header"), oid_to_hex(oid
));
1486 else if (!allow_unknown
&& *oi
->typep
< 0)
1487 die(_("invalid object type"));
1491 *oi
->contentp
= unpack_loose_rest(&stream
, hdr
, *oi
->sizep
, oid
);
1498 status
= error(_("unable to unpack %s header"),
1502 status
= error(_("header for %s too long, exceeds %d bytes"),
1503 oid_to_hex(oid
), MAX_HEADER_LEN
);
1507 if (status
&& (flags
& OBJECT_INFO_DIE_IF_CORRUPT
))
1508 die(_("loose object %s (stored in %s) is corrupt"),
1509 oid_to_hex(oid
), path
);
1511 git_inflate_end(&stream
);
1513 munmap(map
, mapsize
);
1514 if (oi
->sizep
== &size_scratch
)
1516 strbuf_release(&hdrbuf
);
1517 if (oi
->typep
== &type_scratch
)
1519 oi
->whence
= OI_LOOSE
;
1523 int obj_read_use_lock
= 0;
1524 pthread_mutex_t obj_read_mutex
;
1526 void enable_obj_read_lock(void)
1528 if (obj_read_use_lock
)
1531 obj_read_use_lock
= 1;
1532 init_recursive_mutex(&obj_read_mutex
);
1535 void disable_obj_read_lock(void)
1537 if (!obj_read_use_lock
)
1540 obj_read_use_lock
= 0;
1541 pthread_mutex_destroy(&obj_read_mutex
);
1544 int fetch_if_missing
= 1;
1546 static int do_oid_object_info_extended(struct repository
*r
,
1547 const struct object_id
*oid
,
1548 struct object_info
*oi
, unsigned flags
)
1550 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
1551 struct cached_object
*co
;
1552 struct pack_entry e
;
1554 const struct object_id
*real
= oid
;
1555 int already_retried
= 0;
1558 if (flags
& OBJECT_INFO_LOOKUP_REPLACE
)
1559 real
= lookup_replace_object(r
, oid
);
1561 if (is_null_oid(real
))
1567 co
= find_cached_object(real
);
1570 *(oi
->typep
) = co
->type
;
1572 *(oi
->sizep
) = co
->size
;
1574 *(oi
->disk_sizep
) = 0;
1575 if (oi
->delta_base_oid
)
1576 oidclr(oi
->delta_base_oid
);
1578 strbuf_addstr(oi
->type_name
, type_name(co
->type
));
1580 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
1581 oi
->whence
= OI_CACHED
;
1586 if (find_pack_entry(r
, real
, &e
))
1589 /* Most likely it's a loose object. */
1590 if (!loose_object_info(r
, real
, oi
, flags
))
1593 /* Not a loose object; someone else may have just packed it. */
1594 if (!(flags
& OBJECT_INFO_QUICK
)) {
1595 reprepare_packed_git(r
);
1596 if (find_pack_entry(r
, real
, &e
))
1601 * If r is the_repository, this might be an attempt at
1602 * accessing a submodule object as if it were in the_repository
1603 * (having called add_submodule_odb() on that submodule's ODB).
1604 * If any such ODBs exist, register them and try again.
1606 if (r
== the_repository
&&
1607 register_all_submodule_odb_as_alternates())
1608 /* We added some alternates; retry */
1611 /* Check if it is a missing object */
1612 if (fetch_if_missing
&& repo_has_promisor_remote(r
) &&
1614 !(flags
& OBJECT_INFO_SKIP_FETCH_OBJECT
)) {
1615 promisor_remote_get_direct(r
, real
, 1);
1616 already_retried
= 1;
1620 if (flags
& OBJECT_INFO_DIE_IF_CORRUPT
) {
1621 const struct packed_git
*p
;
1622 if ((flags
& OBJECT_INFO_LOOKUP_REPLACE
) && !oideq(real
, oid
))
1623 die(_("replacement %s not found for %s"),
1624 oid_to_hex(real
), oid_to_hex(oid
));
1625 if ((p
= has_packed_and_bad(r
, real
)))
1626 die(_("packed object %s (stored in %s) is corrupt"),
1627 oid_to_hex(real
), p
->pack_name
);
1632 if (oi
== &blank_oi
)
1634 * We know that the caller doesn't actually need the
1635 * information below, so return early.
1638 rtype
= packed_object_info(r
, e
.p
, e
.offset
, oi
);
1640 mark_bad_packed_object(e
.p
, real
);
1641 return do_oid_object_info_extended(r
, real
, oi
, 0);
1642 } else if (oi
->whence
== OI_PACKED
) {
1643 oi
->u
.packed
.offset
= e
.offset
;
1644 oi
->u
.packed
.pack
= e
.p
;
1645 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
1646 rtype
== OBJ_OFS_DELTA
);
1652 int oid_object_info_extended(struct repository
*r
, const struct object_id
*oid
,
1653 struct object_info
*oi
, unsigned flags
)
1657 ret
= do_oid_object_info_extended(r
, oid
, oi
, flags
);
1663 /* returns enum object_type or negative */
1664 int oid_object_info(struct repository
*r
,
1665 const struct object_id
*oid
,
1666 unsigned long *sizep
)
1668 enum object_type type
;
1669 struct object_info oi
= OBJECT_INFO_INIT
;
1673 if (oid_object_info_extended(r
, oid
, &oi
,
1674 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
1679 int pretend_object_file(void *buf
, unsigned long len
, enum object_type type
,
1680 struct object_id
*oid
)
1682 struct cached_object
*co
;
1684 hash_object_file(the_hash_algo
, buf
, len
, type
, oid
);
1685 if (has_object_file_with_flags(oid
, OBJECT_INFO_QUICK
| OBJECT_INFO_SKIP_FETCH_OBJECT
) ||
1686 find_cached_object(oid
))
1688 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
1689 co
= &cached_objects
[cached_object_nr
++];
1692 co
->buf
= xmalloc(len
);
1693 memcpy(co
->buf
, buf
, len
);
1694 oidcpy(&co
->oid
, oid
);
1699 * This function dies on corrupt objects; the callers who want to
1700 * deal with them should arrange to call oid_object_info_extended() and give
1701 * error messages themselves.
1703 void *repo_read_object_file(struct repository
*r
,
1704 const struct object_id
*oid
,
1705 enum object_type
*type
,
1706 unsigned long *size
)
1708 struct object_info oi
= OBJECT_INFO_INIT
;
1709 unsigned flags
= OBJECT_INFO_DIE_IF_CORRUPT
| OBJECT_INFO_LOOKUP_REPLACE
;
1714 oi
.contentp
= &data
;
1715 if (oid_object_info_extended(r
, oid
, &oi
, flags
))
1721 void *read_object_with_reference(struct repository
*r
,
1722 const struct object_id
*oid
,
1723 enum object_type required_type
,
1724 unsigned long *size
,
1725 struct object_id
*actual_oid_return
)
1727 enum object_type type
;
1729 unsigned long isize
;
1730 struct object_id actual_oid
;
1732 oidcpy(&actual_oid
, oid
);
1734 int ref_length
= -1;
1735 const char *ref_type
= NULL
;
1737 buffer
= repo_read_object_file(r
, &actual_oid
, &type
, &isize
);
1740 if (type
== required_type
) {
1742 if (actual_oid_return
)
1743 oidcpy(actual_oid_return
, &actual_oid
);
1746 /* Handle references */
1747 else if (type
== OBJ_COMMIT
)
1749 else if (type
== OBJ_TAG
)
1750 ref_type
= "object ";
1755 ref_length
= strlen(ref_type
);
1757 if (ref_length
+ the_hash_algo
->hexsz
> isize
||
1758 memcmp(buffer
, ref_type
, ref_length
) ||
1759 get_oid_hex((char *) buffer
+ ref_length
, &actual_oid
)) {
1764 /* Now we have the ID of the referred-to object in
1765 * actual_oid. Check again. */
1769 static void hash_object_body(const struct git_hash_algo
*algo
, git_hash_ctx
*c
,
1770 const void *buf
, unsigned long len
,
1771 struct object_id
*oid
,
1772 char *hdr
, int *hdrlen
)
1775 algo
->update_fn(c
, hdr
, *hdrlen
);
1776 algo
->update_fn(c
, buf
, len
);
1777 algo
->final_oid_fn(oid
, c
);
1780 static void write_object_file_prepare(const struct git_hash_algo
*algo
,
1781 const void *buf
, unsigned long len
,
1782 enum object_type type
, struct object_id
*oid
,
1783 char *hdr
, int *hdrlen
)
1787 /* Generate the header */
1788 *hdrlen
= format_object_header(hdr
, *hdrlen
, type
, len
);
1791 hash_object_body(algo
, &c
, buf
, len
, oid
, hdr
, hdrlen
);
1794 static void write_object_file_prepare_literally(const struct git_hash_algo
*algo
,
1795 const void *buf
, unsigned long len
,
1796 const char *type
, struct object_id
*oid
,
1797 char *hdr
, int *hdrlen
)
1801 *hdrlen
= format_object_header_literally(hdr
, *hdrlen
, type
, len
);
1802 hash_object_body(algo
, &c
, buf
, len
, oid
, hdr
, hdrlen
);
1806 * Move the just written object into its final resting place.
1808 int finalize_object_file(const char *tmpfile
, const char *filename
)
1812 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
1814 else if (link(tmpfile
, filename
))
1818 * Coda hack - coda doesn't like cross-directory links,
1819 * so we fall back to a rename, which will mean that it
1820 * won't be able to check collisions, but that's not a
1823 * The same holds for FAT formatted media.
1825 * When this succeeds, we just return. We have nothing
1828 if (ret
&& ret
!= EEXIST
) {
1830 if (!rename(tmpfile
, filename
))
1834 unlink_or_warn(tmpfile
);
1836 if (ret
!= EEXIST
) {
1837 return error_errno(_("unable to write file %s"), filename
);
1839 /* FIXME!!! Collision check here ? */
1843 if (adjust_shared_perm(filename
))
1844 return error(_("unable to set permission to '%s'"), filename
);
1848 static void hash_object_file_literally(const struct git_hash_algo
*algo
,
1849 const void *buf
, unsigned long len
,
1850 const char *type
, struct object_id
*oid
)
1852 char hdr
[MAX_HEADER_LEN
];
1853 int hdrlen
= sizeof(hdr
);
1855 write_object_file_prepare_literally(algo
, buf
, len
, type
, oid
, hdr
, &hdrlen
);
1858 void hash_object_file(const struct git_hash_algo
*algo
, const void *buf
,
1859 unsigned long len
, enum object_type type
,
1860 struct object_id
*oid
)
1862 hash_object_file_literally(algo
, buf
, len
, type_name(type
), oid
);
1865 /* Finalize a file on disk, and close it. */
1866 static void close_loose_object(int fd
, const char *filename
)
1868 if (the_repository
->objects
->odb
->will_destroy
)
1871 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT
))
1872 fsync_loose_object_bulk_checkin(fd
, filename
);
1873 else if (fsync_object_files
> 0)
1874 fsync_or_die(fd
, filename
);
1876 fsync_component_or_die(FSYNC_COMPONENT_LOOSE_OBJECT
, fd
,
1881 die_errno(_("error when closing loose object file"));
1884 /* Size of directory component, including the ending '/' */
1885 static inline int directory_size(const char *filename
)
1887 const char *s
= strrchr(filename
, '/');
1890 return s
- filename
+ 1;
1894 * This creates a temporary file in the same directory as the final
1897 * We want to avoid cross-directory filename renames, because those
1898 * can have problems on various filesystems (FAT, NFS, Coda).
1900 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
1902 int fd
, dirlen
= directory_size(filename
);
1905 strbuf_add(tmp
, filename
, dirlen
);
1906 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
1907 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1908 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
1910 * Make sure the directory exists; note that the contents
1911 * of the buffer are undefined after mkstemp returns an
1912 * error, so we have to rewrite the whole buffer from
1916 strbuf_add(tmp
, filename
, dirlen
- 1);
1917 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
1919 if (adjust_shared_perm(tmp
->buf
))
1923 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
1924 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1930 * Common steps for loose object writers to start writing loose
1933 * - Create tmpfile for the loose object.
1934 * - Setup zlib stream for compression.
1935 * - Start to feed header to zlib stream.
1937 * Returns a "fd", which should later be provided to
1938 * end_loose_object_common().
1940 static int start_loose_object_common(struct strbuf
*tmp_file
,
1941 const char *filename
, unsigned flags
,
1942 git_zstream
*stream
,
1943 unsigned char *buf
, size_t buflen
,
1945 char *hdr
, int hdrlen
)
1949 fd
= create_tmpfile(tmp_file
, filename
);
1951 if (flags
& HASH_SILENT
)
1953 else if (errno
== EACCES
)
1954 return error(_("insufficient permission for adding "
1955 "an object to repository database %s"),
1956 get_object_directory());
1959 _("unable to create temporary file"));
1962 /* Setup zlib stream for compression */
1963 git_deflate_init(stream
, zlib_compression_level
);
1964 stream
->next_out
= buf
;
1965 stream
->avail_out
= buflen
;
1966 the_hash_algo
->init_fn(c
);
1968 /* Start to feed header to zlib stream */
1969 stream
->next_in
= (unsigned char *)hdr
;
1970 stream
->avail_in
= hdrlen
;
1971 while (git_deflate(stream
, 0) == Z_OK
)
1973 the_hash_algo
->update_fn(c
, hdr
, hdrlen
);
1979 * Common steps for the inner git_deflate() loop for writing loose
1980 * objects. Returns what git_deflate() returns.
1982 static int write_loose_object_common(git_hash_ctx
*c
,
1983 git_zstream
*stream
, const int flush
,
1984 unsigned char *in0
, const int fd
,
1985 unsigned char *compressed
,
1986 const size_t compressed_len
)
1990 ret
= git_deflate(stream
, flush
? Z_FINISH
: 0);
1991 the_hash_algo
->update_fn(c
, in0
, stream
->next_in
- in0
);
1992 if (write_in_full(fd
, compressed
, stream
->next_out
- compressed
) < 0)
1993 die_errno(_("unable to write loose object file"));
1994 stream
->next_out
= compressed
;
1995 stream
->avail_out
= compressed_len
;
2001 * Common steps for loose object writers to end writing loose objects:
2003 * - End the compression of zlib stream.
2004 * - Get the calculated oid to "oid".
2006 static int end_loose_object_common(git_hash_ctx
*c
, git_zstream
*stream
,
2007 struct object_id
*oid
)
2011 ret
= git_deflate_end_gently(stream
);
2014 the_hash_algo
->final_oid_fn(oid
, c
);
2019 static int write_loose_object(const struct object_id
*oid
, char *hdr
,
2020 int hdrlen
, const void *buf
, unsigned long len
,
2021 time_t mtime
, unsigned flags
)
2024 unsigned char compressed
[4096];
2027 struct object_id parano_oid
;
2028 static struct strbuf tmp_file
= STRBUF_INIT
;
2029 static struct strbuf filename
= STRBUF_INIT
;
2031 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT
))
2032 prepare_loose_object_bulk_checkin();
2034 loose_object_path(the_repository
, &filename
, oid
);
2036 fd
= start_loose_object_common(&tmp_file
, filename
.buf
, flags
,
2037 &stream
, compressed
, sizeof(compressed
),
2042 /* Then the data itself.. */
2043 stream
.next_in
= (void *)buf
;
2044 stream
.avail_in
= len
;
2046 unsigned char *in0
= stream
.next_in
;
2048 ret
= write_loose_object_common(&c
, &stream
, 1, in0
, fd
,
2049 compressed
, sizeof(compressed
));
2050 } while (ret
== Z_OK
);
2052 if (ret
!= Z_STREAM_END
)
2053 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid
),
2055 ret
= end_loose_object_common(&c
, &stream
, ¶no_oid
);
2057 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid
),
2059 if (!oideq(oid
, ¶no_oid
))
2060 die(_("confused by unstable object source data for %s"),
2063 close_loose_object(fd
, tmp_file
.buf
);
2068 utb
.modtime
= mtime
;
2069 if (utime(tmp_file
.buf
, &utb
) < 0 &&
2070 !(flags
& HASH_SILENT
))
2071 warning_errno(_("failed utime() on %s"), tmp_file
.buf
);
2074 return finalize_object_file(tmp_file
.buf
, filename
.buf
);
2077 static int freshen_loose_object(const struct object_id
*oid
)
2079 return check_and_freshen(oid
, 1);
2082 static int freshen_packed_object(const struct object_id
*oid
)
2084 struct pack_entry e
;
2085 if (!find_pack_entry(the_repository
, oid
, &e
))
2091 if (!freshen_file(e
.p
->pack_name
))
2097 int stream_loose_object(struct input_stream
*in_stream
, size_t len
,
2098 struct object_id
*oid
)
2100 int fd
, ret
, err
= 0, flush
= 0;
2101 unsigned char compressed
[4096];
2104 struct strbuf tmp_file
= STRBUF_INIT
;
2105 struct strbuf filename
= STRBUF_INIT
;
2107 char hdr
[MAX_HEADER_LEN
];
2110 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT
))
2111 prepare_loose_object_bulk_checkin();
2113 /* Since oid is not determined, save tmp file to odb path. */
2114 strbuf_addf(&filename
, "%s/", get_object_directory());
2115 hdrlen
= format_object_header(hdr
, sizeof(hdr
), OBJ_BLOB
, len
);
2118 * Common steps for write_loose_object and stream_loose_object to
2119 * start writing loose objects:
2121 * - Create tmpfile for the loose object.
2122 * - Setup zlib stream for compression.
2123 * - Start to feed header to zlib stream.
2125 fd
= start_loose_object_common(&tmp_file
, filename
.buf
, 0,
2126 &stream
, compressed
, sizeof(compressed
),
2133 /* Then the data itself.. */
2135 unsigned char *in0
= stream
.next_in
;
2137 if (!stream
.avail_in
&& !in_stream
->is_finished
) {
2138 const void *in
= in_stream
->read(in_stream
, &stream
.avail_in
);
2139 stream
.next_in
= (void *)in
;
2140 in0
= (unsigned char *)in
;
2141 /* All data has been read. */
2142 if (in_stream
->is_finished
)
2145 ret
= write_loose_object_common(&c
, &stream
, flush
, in0
, fd
,
2146 compressed
, sizeof(compressed
));
2148 * Unlike write_loose_object(), we do not have the entire
2149 * buffer. If we get Z_BUF_ERROR due to too few input bytes,
2150 * then we'll replenish them in the next input_stream->read()
2151 * call when we loop.
2153 } while (ret
== Z_OK
|| ret
== Z_BUF_ERROR
);
2155 if (stream
.total_in
!= len
+ hdrlen
)
2156 die(_("write stream object %ld != %"PRIuMAX
), stream
.total_in
,
2157 (uintmax_t)len
+ hdrlen
);
2160 * Common steps for write_loose_object and stream_loose_object to
2161 * end writing loose oject:
2163 * - End the compression of zlib stream.
2164 * - Get the calculated oid.
2166 if (ret
!= Z_STREAM_END
)
2167 die(_("unable to stream deflate new object (%d)"), ret
);
2168 ret
= end_loose_object_common(&c
, &stream
, oid
);
2170 die(_("deflateEnd on stream object failed (%d)"), ret
);
2171 close_loose_object(fd
, tmp_file
.buf
);
2173 if (freshen_packed_object(oid
) || freshen_loose_object(oid
)) {
2174 unlink_or_warn(tmp_file
.buf
);
2178 loose_object_path(the_repository
, &filename
, oid
);
2180 /* We finally know the object path, and create the missing dir. */
2181 dirlen
= directory_size(filename
.buf
);
2183 struct strbuf dir
= STRBUF_INIT
;
2184 strbuf_add(&dir
, filename
.buf
, dirlen
);
2186 if (mkdir_in_gitdir(dir
.buf
) && errno
!= EEXIST
) {
2187 err
= error_errno(_("unable to create directory %s"), dir
.buf
);
2188 strbuf_release(&dir
);
2191 strbuf_release(&dir
);
2194 err
= finalize_object_file(tmp_file
.buf
, filename
.buf
);
2196 strbuf_release(&tmp_file
);
2197 strbuf_release(&filename
);
2201 int write_object_file_flags(const void *buf
, unsigned long len
,
2202 enum object_type type
, struct object_id
*oid
,
2205 char hdr
[MAX_HEADER_LEN
];
2206 int hdrlen
= sizeof(hdr
);
2208 /* Normally if we have it in the pack then we do not bother writing
2209 * it out into .git/objects/??/?{38} file.
2211 write_object_file_prepare(the_hash_algo
, buf
, len
, type
, oid
, hdr
,
2213 if (freshen_packed_object(oid
) || freshen_loose_object(oid
))
2215 return write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, 0, flags
);
2218 int write_object_file_literally(const void *buf
, unsigned long len
,
2219 const char *type
, struct object_id
*oid
,
2223 int hdrlen
, status
= 0;
2225 /* type string, SP, %lu of the length plus NUL must fit this */
2226 hdrlen
= strlen(type
) + MAX_HEADER_LEN
;
2227 header
= xmalloc(hdrlen
);
2228 write_object_file_prepare_literally(the_hash_algo
, buf
, len
, type
,
2229 oid
, header
, &hdrlen
);
2231 if (!(flags
& HASH_WRITE_OBJECT
))
2233 if (freshen_packed_object(oid
) || freshen_loose_object(oid
))
2235 status
= write_loose_object(oid
, header
, hdrlen
, buf
, len
, 0, 0);
2242 int force_object_loose(const struct object_id
*oid
, time_t mtime
)
2246 struct object_info oi
= OBJECT_INFO_INIT
;
2247 enum object_type type
;
2248 char hdr
[MAX_HEADER_LEN
];
2252 if (has_loose_object(oid
))
2257 if (oid_object_info_extended(the_repository
, oid
, &oi
, 0))
2258 return error(_("cannot read object for %s"), oid_to_hex(oid
));
2259 hdrlen
= format_object_header(hdr
, sizeof(hdr
), type
, len
);
2260 ret
= write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, mtime
, 0);
2266 int has_object(struct repository
*r
, const struct object_id
*oid
,
2269 int quick
= !(flags
& HAS_OBJECT_RECHECK_PACKED
);
2270 unsigned object_info_flags
= OBJECT_INFO_SKIP_FETCH_OBJECT
|
2271 (quick
? OBJECT_INFO_QUICK
: 0);
2273 if (!startup_info
->have_repository
)
2275 return oid_object_info_extended(r
, oid
, NULL
, object_info_flags
) >= 0;
2278 int repo_has_object_file_with_flags(struct repository
*r
,
2279 const struct object_id
*oid
, int flags
)
2281 if (!startup_info
->have_repository
)
2283 return oid_object_info_extended(r
, oid
, NULL
, flags
) >= 0;
2286 int repo_has_object_file(struct repository
*r
,
2287 const struct object_id
*oid
)
2289 return repo_has_object_file_with_flags(r
, oid
, 0);
2293 * We can't use the normal fsck_error_function() for index_mem(),
2294 * because we don't yet have a valid oid for it to report. Instead,
2295 * report the minimal fsck error here, and rely on the caller to
2296 * give more context.
2298 static int hash_format_check_report(struct fsck_options
*opts
,
2299 const struct object_id
*oid
,
2300 enum object_type object_type
,
2301 enum fsck_msg_type msg_type
,
2302 enum fsck_msg_id msg_id
,
2303 const char *message
)
2305 error(_("object fails fsck: %s"), message
);
2309 static int index_mem(struct index_state
*istate
,
2310 struct object_id
*oid
, void *buf
, size_t size
,
2311 enum object_type type
,
2312 const char *path
, unsigned flags
)
2315 int re_allocated
= 0;
2316 int write_object
= flags
& HASH_WRITE_OBJECT
;
2322 * Convert blobs to git internal format
2324 if ((type
== OBJ_BLOB
) && path
) {
2325 struct strbuf nbuf
= STRBUF_INIT
;
2326 if (convert_to_git(istate
, path
, buf
, size
, &nbuf
,
2327 get_conv_flags(flags
))) {
2328 buf
= strbuf_detach(&nbuf
, &size
);
2332 if (flags
& HASH_FORMAT_CHECK
) {
2333 struct fsck_options opts
= FSCK_OPTIONS_DEFAULT
;
2336 opts
.error_func
= hash_format_check_report
;
2337 if (fsck_buffer(null_oid(), type
, buf
, size
, &opts
))
2338 die(_("refusing to create malformed object"));
2343 ret
= write_object_file(buf
, size
, type
, oid
);
2345 hash_object_file(the_hash_algo
, buf
, size
, type
, oid
);
2351 static int index_stream_convert_blob(struct index_state
*istate
,
2352 struct object_id
*oid
,
2358 const int write_object
= flags
& HASH_WRITE_OBJECT
;
2359 struct strbuf sbuf
= STRBUF_INIT
;
2362 assert(would_convert_to_git_filter_fd(istate
, path
));
2364 convert_to_git_filter_fd(istate
, path
, fd
, &sbuf
,
2365 get_conv_flags(flags
));
2368 ret
= write_object_file(sbuf
.buf
, sbuf
.len
, OBJ_BLOB
,
2371 hash_object_file(the_hash_algo
, sbuf
.buf
, sbuf
.len
, OBJ_BLOB
,
2373 strbuf_release(&sbuf
);
2377 static int index_pipe(struct index_state
*istate
, struct object_id
*oid
,
2378 int fd
, enum object_type type
,
2379 const char *path
, unsigned flags
)
2381 struct strbuf sbuf
= STRBUF_INIT
;
2384 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2385 ret
= index_mem(istate
, oid
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
2388 strbuf_release(&sbuf
);
2392 #define SMALL_FILE_SIZE (32*1024)
2394 static int index_core(struct index_state
*istate
,
2395 struct object_id
*oid
, int fd
, size_t size
,
2396 enum object_type type
, const char *path
,
2402 ret
= index_mem(istate
, oid
, "", size
, type
, path
, flags
);
2403 } else if (size
<= SMALL_FILE_SIZE
) {
2404 char *buf
= xmalloc(size
);
2405 ssize_t read_result
= read_in_full(fd
, buf
, size
);
2406 if (read_result
< 0)
2407 ret
= error_errno(_("read error while indexing %s"),
2408 path
? path
: "<unknown>");
2409 else if (read_result
!= size
)
2410 ret
= error(_("short read while indexing %s"),
2411 path
? path
: "<unknown>");
2413 ret
= index_mem(istate
, oid
, buf
, size
, type
, path
, flags
);
2416 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2417 ret
= index_mem(istate
, oid
, buf
, size
, type
, path
, flags
);
2424 * This creates one packfile per large blob unless bulk-checkin
2425 * machinery is "plugged".
2427 * This also bypasses the usual "convert-to-git" dance, and that is on
2428 * purpose. We could write a streaming version of the converting
2429 * functions and insert that before feeding the data to fast-import
2430 * (or equivalent in-core API described above). However, that is
2431 * somewhat complicated, as we do not know the size of the filter
2432 * result, which we need to know beforehand when writing a git object.
2433 * Since the primary motivation for trying to stream from the working
2434 * tree file and to avoid mmaping it in core is to deal with large
2435 * binary blobs, they generally do not want to get any conversion, and
2436 * callers should avoid this code path when filters are requested.
2438 static int index_stream(struct object_id
*oid
, int fd
, size_t size
,
2439 enum object_type type
, const char *path
,
2442 return index_bulk_checkin(oid
, fd
, size
, type
, path
, flags
);
2445 int index_fd(struct index_state
*istate
, struct object_id
*oid
,
2446 int fd
, struct stat
*st
,
2447 enum object_type type
, const char *path
, unsigned flags
)
2452 * Call xsize_t() only when needed to avoid potentially unnecessary
2453 * die() for large files.
2455 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(istate
, path
))
2456 ret
= index_stream_convert_blob(istate
, oid
, fd
, path
, flags
);
2457 else if (!S_ISREG(st
->st_mode
))
2458 ret
= index_pipe(istate
, oid
, fd
, type
, path
, flags
);
2459 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
2460 (path
&& would_convert_to_git(istate
, path
)))
2461 ret
= index_core(istate
, oid
, fd
, xsize_t(st
->st_size
),
2464 ret
= index_stream(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
2470 int index_path(struct index_state
*istate
, struct object_id
*oid
,
2471 const char *path
, struct stat
*st
, unsigned flags
)
2474 struct strbuf sb
= STRBUF_INIT
;
2477 switch (st
->st_mode
& S_IFMT
) {
2479 fd
= open(path
, O_RDONLY
);
2481 return error_errno("open(\"%s\")", path
);
2482 if (index_fd(istate
, oid
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
2483 return error(_("%s: failed to insert into database"),
2487 if (strbuf_readlink(&sb
, path
, st
->st_size
))
2488 return error_errno("readlink(\"%s\")", path
);
2489 if (!(flags
& HASH_WRITE_OBJECT
))
2490 hash_object_file(the_hash_algo
, sb
.buf
, sb
.len
,
2492 else if (write_object_file(sb
.buf
, sb
.len
, OBJ_BLOB
, oid
))
2493 rc
= error(_("%s: failed to insert into database"), path
);
2494 strbuf_release(&sb
);
2497 return resolve_gitlink_ref(path
, "HEAD", oid
);
2499 return error(_("%s: unsupported file type"), path
);
2504 int read_pack_header(int fd
, struct pack_header
*header
)
2506 if (read_in_full(fd
, header
, sizeof(*header
)) != sizeof(*header
))
2507 /* "eof before pack header was fully read" */
2508 return PH_ERROR_EOF
;
2510 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2511 /* "protocol error (pack signature mismatch detected)" */
2512 return PH_ERROR_PACK_SIGNATURE
;
2513 if (!pack_version_ok(header
->hdr_version
))
2514 /* "protocol error (pack version unsupported)" */
2515 return PH_ERROR_PROTOCOL
;
2519 void assert_oid_type(const struct object_id
*oid
, enum object_type expect
)
2521 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
2523 die(_("%s is not a valid object"), oid_to_hex(oid
));
2525 die(_("%s is not a valid '%s' object"), oid_to_hex(oid
),
2529 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
2530 struct strbuf
*path
,
2531 each_loose_object_fn obj_cb
,
2532 each_loose_cruft_fn cruft_cb
,
2533 each_loose_subdir_fn subdir_cb
,
2536 size_t origlen
, baselen
;
2540 struct object_id oid
;
2542 if (subdir_nr
> 0xff)
2543 BUG("invalid loose object subdirectory: %x", subdir_nr
);
2545 origlen
= path
->len
;
2546 strbuf_complete(path
, '/');
2547 strbuf_addf(path
, "%02x", subdir_nr
);
2549 dir
= opendir(path
->buf
);
2551 if (errno
!= ENOENT
)
2552 r
= error_errno(_("unable to open %s"), path
->buf
);
2553 strbuf_setlen(path
, origlen
);
2557 oid
.hash
[0] = subdir_nr
;
2558 strbuf_addch(path
, '/');
2559 baselen
= path
->len
;
2561 while ((de
= readdir_skip_dot_and_dotdot(dir
))) {
2564 namelen
= strlen(de
->d_name
);
2565 strbuf_setlen(path
, baselen
);
2566 strbuf_add(path
, de
->d_name
, namelen
);
2567 if (namelen
== the_hash_algo
->hexsz
- 2 &&
2568 !hex_to_bytes(oid
.hash
+ 1, de
->d_name
,
2569 the_hash_algo
->rawsz
- 1)) {
2570 oid_set_algo(&oid
, the_hash_algo
);
2572 r
= obj_cb(&oid
, path
->buf
, data
);
2580 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
2587 strbuf_setlen(path
, baselen
- 1);
2588 if (!r
&& subdir_cb
)
2589 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
2591 strbuf_setlen(path
, origlen
);
2596 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
2597 each_loose_object_fn obj_cb
,
2598 each_loose_cruft_fn cruft_cb
,
2599 each_loose_subdir_fn subdir_cb
,
2605 for (i
= 0; i
< 256; i
++) {
2606 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
2615 int for_each_loose_file_in_objdir(const char *path
,
2616 each_loose_object_fn obj_cb
,
2617 each_loose_cruft_fn cruft_cb
,
2618 each_loose_subdir_fn subdir_cb
,
2621 struct strbuf buf
= STRBUF_INIT
;
2624 strbuf_addstr(&buf
, path
);
2625 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
2627 strbuf_release(&buf
);
2632 int for_each_loose_object(each_loose_object_fn cb
, void *data
,
2633 enum for_each_object_flags flags
)
2635 struct object_directory
*odb
;
2637 prepare_alt_odb(the_repository
);
2638 for (odb
= the_repository
->objects
->odb
; odb
; odb
= odb
->next
) {
2639 int r
= for_each_loose_file_in_objdir(odb
->path
, cb
, NULL
,
2644 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
2651 static int append_loose_object(const struct object_id
*oid
,
2652 const char *path UNUSED
,
2655 oidtree_insert(data
, oid
);
2659 struct oidtree
*odb_loose_cache(struct object_directory
*odb
,
2660 const struct object_id
*oid
)
2662 int subdir_nr
= oid
->hash
[0];
2663 struct strbuf buf
= STRBUF_INIT
;
2664 size_t word_bits
= bitsizeof(odb
->loose_objects_subdir_seen
[0]);
2665 size_t word_index
= subdir_nr
/ word_bits
;
2666 size_t mask
= (size_t)1u << (subdir_nr
% word_bits
);
2669 if (subdir_nr
< 0 ||
2670 subdir_nr
>= bitsizeof(odb
->loose_objects_subdir_seen
))
2671 BUG("subdir_nr out of range");
2673 bitmap
= &odb
->loose_objects_subdir_seen
[word_index
];
2675 return odb
->loose_objects_cache
;
2676 if (!odb
->loose_objects_cache
) {
2677 ALLOC_ARRAY(odb
->loose_objects_cache
, 1);
2678 oidtree_init(odb
->loose_objects_cache
);
2680 strbuf_addstr(&buf
, odb
->path
);
2681 for_each_file_in_obj_subdir(subdir_nr
, &buf
,
2682 append_loose_object
,
2684 odb
->loose_objects_cache
);
2686 strbuf_release(&buf
);
2687 return odb
->loose_objects_cache
;
2690 void odb_clear_loose_cache(struct object_directory
*odb
)
2692 oidtree_clear(odb
->loose_objects_cache
);
2693 FREE_AND_NULL(odb
->loose_objects_cache
);
2694 memset(&odb
->loose_objects_subdir_seen
, 0,
2695 sizeof(odb
->loose_objects_subdir_seen
));
2698 static int check_stream_oid(git_zstream
*stream
,
2702 const struct object_id
*expected_oid
)
2705 struct object_id real_oid
;
2706 unsigned char buf
[4096];
2707 unsigned long total_read
;
2710 the_hash_algo
->init_fn(&c
);
2711 the_hash_algo
->update_fn(&c
, hdr
, stream
->total_out
);
2714 * We already read some bytes into hdr, but the ones up to the NUL
2715 * do not count against the object's content size.
2717 total_read
= stream
->total_out
- strlen(hdr
) - 1;
2720 * This size comparison must be "<=" to read the final zlib packets;
2721 * see the comment in unpack_loose_rest for details.
2723 while (total_read
<= size
&&
2725 (status
== Z_BUF_ERROR
&& !stream
->avail_out
))) {
2726 stream
->next_out
= buf
;
2727 stream
->avail_out
= sizeof(buf
);
2728 if (size
- total_read
< stream
->avail_out
)
2729 stream
->avail_out
= size
- total_read
;
2730 status
= git_inflate(stream
, Z_FINISH
);
2731 the_hash_algo
->update_fn(&c
, buf
, stream
->next_out
- buf
);
2732 total_read
+= stream
->next_out
- buf
;
2734 git_inflate_end(stream
);
2736 if (status
!= Z_STREAM_END
) {
2737 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid
));
2740 if (stream
->avail_in
) {
2741 error(_("garbage at end of loose object '%s'"),
2742 oid_to_hex(expected_oid
));
2746 the_hash_algo
->final_oid_fn(&real_oid
, &c
);
2747 if (!oideq(expected_oid
, &real_oid
)) {
2748 error(_("hash mismatch for %s (expected %s)"), path
,
2749 oid_to_hex(expected_oid
));
2756 int read_loose_object(const char *path
,
2757 const struct object_id
*expected_oid
,
2758 struct object_id
*real_oid
,
2760 struct object_info
*oi
)
2765 unsigned long mapsize
;
2767 char hdr
[MAX_HEADER_LEN
];
2768 unsigned long *size
= oi
->sizep
;
2770 fd
= git_open(path
);
2772 map
= map_fd(fd
, path
, &mapsize
);
2774 error_errno(_("unable to mmap %s"), path
);
2778 if (unpack_loose_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
),
2780 error(_("unable to unpack header of %s"), path
);
2784 if (parse_loose_header(hdr
, oi
) < 0) {
2785 error(_("unable to parse header of %s"), path
);
2786 git_inflate_end(&stream
);
2790 if (*oi
->typep
== OBJ_BLOB
&& *size
> big_file_threshold
) {
2791 if (check_stream_oid(&stream
, hdr
, *size
, path
, expected_oid
) < 0)
2794 *contents
= unpack_loose_rest(&stream
, hdr
, *size
, expected_oid
);
2796 error(_("unable to unpack contents of %s"), path
);
2797 git_inflate_end(&stream
);
2800 hash_object_file_literally(the_repository
->hash_algo
,
2802 oi
->type_name
->buf
, real_oid
);
2803 if (!oideq(expected_oid
, real_oid
))
2807 ret
= 0; /* everything checks out */
2811 munmap(map
, mapsize
);