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
;
679 const char *repo
, *ref_git_s
;
682 ref_git_s
= real_path_if_valid(path
);
685 strbuf_addf(err
, _("path '%s' does not exist"), path
);
689 * Beware: read_gitfile(), real_path() and mkpath()
690 * return static buffer
692 ref_git
= xstrdup(ref_git_s
);
694 repo
= read_gitfile(ref_git
);
696 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
699 ref_git
= xstrdup(repo
);
702 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
703 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
705 ref_git
= ref_git_git
;
706 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
707 struct strbuf sb
= STRBUF_INIT
;
709 if (get_common_dir(&sb
, ref_git
)) {
711 _("reference repository '%s' as a linked "
712 "checkout is not supported yet."),
717 strbuf_addf(err
, _("reference repository '%s' is not a "
718 "local repository."), path
);
722 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
723 strbuf_addf(err
, _("reference repository '%s' is shallow"),
729 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
731 _("reference repository '%s' is grafted"),
739 FREE_AND_NULL(ref_git
);
745 static void fill_alternate_refs_command(struct child_process
*cmd
,
746 const char *repo_path
)
750 if (!git_config_get_value("core.alternateRefsCommand", &value
)) {
753 argv_array_push(&cmd
->args
, value
);
754 argv_array_push(&cmd
->args
, repo_path
);
758 argv_array_pushf(&cmd
->args
, "--git-dir=%s", repo_path
);
759 argv_array_push(&cmd
->args
, "for-each-ref");
760 argv_array_push(&cmd
->args
, "--format=%(objectname)");
762 if (!git_config_get_value("core.alternateRefsPrefixes", &value
)) {
763 argv_array_push(&cmd
->args
, "--");
764 argv_array_split(&cmd
->args
, value
);
768 cmd
->env
= local_repo_env
;
772 static void read_alternate_refs(const char *path
,
773 alternate_ref_fn
*cb
,
776 struct child_process cmd
= CHILD_PROCESS_INIT
;
777 struct strbuf line
= STRBUF_INIT
;
780 fill_alternate_refs_command(&cmd
, path
);
782 if (start_command(&cmd
))
785 fh
= xfdopen(cmd
.out
, "r");
786 while (strbuf_getline_lf(&line
, fh
) != EOF
) {
787 struct object_id oid
;
790 if (parse_oid_hex(line
.buf
, &oid
, &p
) || *p
) {
791 warning(_("invalid line while parsing alternate refs: %s"),
800 finish_command(&cmd
);
801 strbuf_release(&line
);
804 struct alternate_refs_data
{
805 alternate_ref_fn
*fn
;
809 static int refs_from_alternate_cb(struct object_directory
*e
,
812 struct strbuf path
= STRBUF_INIT
;
814 struct alternate_refs_data
*cb
= data
;
816 if (!strbuf_realpath(&path
, e
->path
, 0))
818 if (!strbuf_strip_suffix(&path
, "/objects"))
822 /* Is this a git repository with refs? */
823 strbuf_addstr(&path
, "/refs");
824 if (!is_directory(path
.buf
))
826 strbuf_setlen(&path
, base_len
);
828 read_alternate_refs(path
.buf
, cb
->fn
, cb
->data
);
831 strbuf_release(&path
);
835 void for_each_alternate_ref(alternate_ref_fn fn
, void *data
)
837 struct alternate_refs_data cb
;
840 foreach_alt_odb(refs_from_alternate_cb
, &cb
);
843 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
845 struct object_directory
*ent
;
848 prepare_alt_odb(the_repository
);
849 for (ent
= the_repository
->objects
->odb
->next
; ent
; ent
= ent
->next
) {
857 void prepare_alt_odb(struct repository
*r
)
859 if (r
->objects
->loaded_alternates
)
862 link_alt_odb_entries(r
, r
->objects
->alternate_db
, PATH_SEP
, NULL
, 0);
864 read_info_alternates(r
, r
->objects
->odb
->path
, 0);
865 r
->objects
->loaded_alternates
= 1;
868 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
869 static int freshen_file(const char *fn
)
872 t
.actime
= t
.modtime
= time(NULL
);
873 return !utime(fn
, &t
);
877 * All of the check_and_freshen functions return 1 if the file exists and was
878 * freshened (if freshening was requested), 0 otherwise. If they return
879 * 0, you should not assume that it is safe to skip a write of the object (it
880 * either does not exist on disk, or has a stale mtime and may be subject to
883 int check_and_freshen_file(const char *fn
, int freshen
)
885 if (access(fn
, F_OK
))
887 if (freshen
&& !freshen_file(fn
))
892 static int check_and_freshen_odb(struct object_directory
*odb
,
893 const struct object_id
*oid
,
896 static struct strbuf path
= STRBUF_INIT
;
897 odb_loose_path(odb
, &path
, oid
);
898 return check_and_freshen_file(path
.buf
, freshen
);
901 static int check_and_freshen_local(const struct object_id
*oid
, int freshen
)
903 return check_and_freshen_odb(the_repository
->objects
->odb
, oid
, freshen
);
906 static int check_and_freshen_nonlocal(const struct object_id
*oid
, int freshen
)
908 struct object_directory
*odb
;
910 prepare_alt_odb(the_repository
);
911 for (odb
= the_repository
->objects
->odb
->next
; odb
; odb
= odb
->next
) {
912 if (check_and_freshen_odb(odb
, oid
, freshen
))
918 static int check_and_freshen(const struct object_id
*oid
, int freshen
)
920 return check_and_freshen_local(oid
, freshen
) ||
921 check_and_freshen_nonlocal(oid
, freshen
);
924 int has_loose_object_nonlocal(const struct object_id
*oid
)
926 return check_and_freshen_nonlocal(oid
, 0);
929 static int has_loose_object(const struct object_id
*oid
)
931 return check_and_freshen(oid
, 0);
934 static void mmap_limit_check(size_t length
)
936 static size_t limit
= 0;
938 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
943 die(_("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
),
944 (uintmax_t)length
, (uintmax_t)limit
);
947 void *xmmap_gently(void *start
, size_t length
,
948 int prot
, int flags
, int fd
, off_t offset
)
952 mmap_limit_check(length
);
953 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
954 if (ret
== MAP_FAILED
&& !length
)
959 void *xmmap(void *start
, size_t length
,
960 int prot
, int flags
, int fd
, off_t offset
)
962 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
963 if (ret
== MAP_FAILED
)
964 die_errno(_("mmap failed"));
969 * With an in-core object data in "map", rehash it to make sure the
970 * object name actually matches "oid" to detect object corruption.
971 * With "map" == NULL, try reading the object named with "oid" using
972 * the streaming interface and rehash it to do the same.
974 int check_object_signature(const struct object_id
*oid
, void *map
,
975 unsigned long size
, const char *type
)
977 struct object_id real_oid
;
978 enum object_type obj_type
;
979 struct git_istream
*st
;
981 char hdr
[MAX_HEADER_LEN
];
985 hash_object_file(map
, size
, type
, &real_oid
);
986 return !oideq(oid
, &real_oid
) ? -1 : 0;
989 st
= open_istream(the_repository
, oid
, &obj_type
, &size
, NULL
);
993 /* Generate the header */
994 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %"PRIuMAX
, type_name(obj_type
), (uintmax_t)size
) + 1;
997 the_hash_algo
->init_fn(&c
);
998 the_hash_algo
->update_fn(&c
, hdr
, hdrlen
);
1000 char buf
[1024 * 16];
1001 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1009 the_hash_algo
->update_fn(&c
, buf
, readlen
);
1011 the_hash_algo
->final_fn(real_oid
.hash
, &c
);
1013 return !oideq(oid
, &real_oid
) ? -1 : 0;
1016 int git_open_cloexec(const char *name
, int flags
)
1019 static int o_cloexec
= O_CLOEXEC
;
1021 fd
= open(name
, flags
| o_cloexec
);
1022 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
1023 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1024 o_cloexec
&= ~O_CLOEXEC
;
1025 fd
= open(name
, flags
| o_cloexec
);
1028 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1030 static int fd_cloexec
= FD_CLOEXEC
;
1032 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
1033 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1034 int flags
= fcntl(fd
, F_GETFD
);
1035 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
1044 * Find "oid" as a loose object in the local repository or in an alternate.
1045 * Returns 0 on success, negative on failure.
1047 * The "path" out-parameter will give the path of the object we found (if any).
1048 * Note that it may point to static storage and is only valid until another
1049 * call to stat_loose_object().
1051 static int stat_loose_object(struct repository
*r
, const struct object_id
*oid
,
1052 struct stat
*st
, const char **path
)
1054 struct object_directory
*odb
;
1055 static struct strbuf buf
= STRBUF_INIT
;
1058 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1059 *path
= odb_loose_path(odb
, &buf
, oid
);
1060 if (!lstat(*path
, st
))
1068 * Like stat_loose_object(), but actually open the object and return the
1069 * descriptor. See the caveats on the "path" parameter above.
1071 static int open_loose_object(struct repository
*r
,
1072 const struct object_id
*oid
, const char **path
)
1075 struct object_directory
*odb
;
1076 int most_interesting_errno
= ENOENT
;
1077 static struct strbuf buf
= STRBUF_INIT
;
1080 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1081 *path
= odb_loose_path(odb
, &buf
, oid
);
1082 fd
= git_open(*path
);
1086 if (most_interesting_errno
== ENOENT
)
1087 most_interesting_errno
= errno
;
1089 errno
= most_interesting_errno
;
1093 static int quick_has_loose(struct repository
*r
,
1094 const struct object_id
*oid
)
1096 struct object_directory
*odb
;
1099 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1100 if (oid_array_lookup(odb_loose_cache(odb
, oid
), oid
) >= 0)
1107 * Map the loose object at "path" if it is not NULL, or the path found by
1108 * searching for a loose object named "oid".
1110 static void *map_loose_object_1(struct repository
*r
, const char *path
,
1111 const struct object_id
*oid
, unsigned long *size
)
1117 fd
= git_open(path
);
1119 fd
= open_loose_object(r
, oid
, &path
);
1124 if (!fstat(fd
, &st
)) {
1125 *size
= xsize_t(st
.st_size
);
1127 /* mmap() is forbidden on empty files */
1128 error(_("object file %s is empty"), path
);
1132 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1139 void *map_loose_object(struct repository
*r
,
1140 const struct object_id
*oid
,
1141 unsigned long *size
)
1143 return map_loose_object_1(r
, NULL
, oid
, size
);
1146 static int unpack_loose_short_header(git_zstream
*stream
,
1147 unsigned char *map
, unsigned long mapsize
,
1148 void *buffer
, unsigned long bufsiz
)
1150 /* Get the data stream */
1151 memset(stream
, 0, sizeof(*stream
));
1152 stream
->next_in
= map
;
1153 stream
->avail_in
= mapsize
;
1154 stream
->next_out
= buffer
;
1155 stream
->avail_out
= bufsiz
;
1157 git_inflate_init(stream
);
1158 return git_inflate(stream
, 0);
1161 int unpack_loose_header(git_zstream
*stream
,
1162 unsigned char *map
, unsigned long mapsize
,
1163 void *buffer
, unsigned long bufsiz
)
1165 int status
= unpack_loose_short_header(stream
, map
, mapsize
,
1171 /* Make sure we have the terminating NUL */
1172 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1177 static int unpack_loose_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1178 unsigned long mapsize
, void *buffer
,
1179 unsigned long bufsiz
, struct strbuf
*header
)
1183 status
= unpack_loose_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1188 * Check if entire header is unpacked in the first iteration.
1190 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1194 * buffer[0..bufsiz] was not large enough. Copy the partial
1195 * result out to header, and then append the result of further
1196 * reading the stream.
1198 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1199 stream
->next_out
= buffer
;
1200 stream
->avail_out
= bufsiz
;
1203 status
= git_inflate(stream
, 0);
1204 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1205 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1207 stream
->next_out
= buffer
;
1208 stream
->avail_out
= bufsiz
;
1209 } while (status
!= Z_STREAM_END
);
1213 static void *unpack_loose_rest(git_zstream
*stream
,
1214 void *buffer
, unsigned long size
,
1215 const struct object_id
*oid
)
1217 int bytes
= strlen(buffer
) + 1;
1218 unsigned char *buf
= xmallocz(size
);
1222 n
= stream
->total_out
- bytes
;
1225 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1227 if (bytes
<= size
) {
1229 * The above condition must be (bytes <= size), not
1230 * (bytes < size). In other words, even though we
1231 * expect no more output and set avail_out to zero,
1232 * the input zlib stream may have bytes that express
1233 * "this concludes the stream", and we *do* want to
1236 * Otherwise we would not be able to test that we
1237 * consumed all the input to reach the expected size;
1238 * we also want to check that zlib tells us that all
1239 * went well with status == Z_STREAM_END at the end.
1241 stream
->next_out
= buf
+ bytes
;
1242 stream
->avail_out
= size
- bytes
;
1243 while (status
== Z_OK
)
1244 status
= git_inflate(stream
, Z_FINISH
);
1246 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1247 git_inflate_end(stream
);
1252 error(_("corrupt loose object '%s'"), oid_to_hex(oid
));
1253 else if (stream
->avail_in
)
1254 error(_("garbage at end of loose object '%s'"),
1261 * We used to just use "sscanf()", but that's actually way
1262 * too permissive for what we want to check. So do an anal
1263 * object header parse by hand.
1265 static int parse_loose_header_extended(const char *hdr
, struct object_info
*oi
,
1268 const char *type_buf
= hdr
;
1270 int type
, type_len
= 0;
1273 * The type can be of any size but is followed by
1285 type
= type_from_string_gently(type_buf
, type_len
, 1);
1287 strbuf_add(oi
->type_name
, type_buf
, type_len
);
1289 * Set type to 0 if its an unknown object and
1290 * we're obtaining the type using '--allow-unknown-type'
1293 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
) && (type
< 0))
1296 die(_("invalid object type"));
1301 * The length must follow immediately, and be in canonical
1302 * decimal format (ie "010" is not valid).
1304 size
= *hdr
++ - '0';
1309 unsigned long c
= *hdr
- '0';
1313 size
= size
* 10 + c
;
1321 * The length must be followed by a zero byte
1323 return *hdr
? -1 : type
;
1326 int parse_loose_header(const char *hdr
, unsigned long *sizep
)
1328 struct object_info oi
= OBJECT_INFO_INIT
;
1331 return parse_loose_header_extended(hdr
, &oi
, 0);
1334 static int loose_object_info(struct repository
*r
,
1335 const struct object_id
*oid
,
1336 struct object_info
*oi
, int flags
)
1339 unsigned long mapsize
;
1342 char hdr
[MAX_HEADER_LEN
];
1343 struct strbuf hdrbuf
= STRBUF_INIT
;
1344 unsigned long size_scratch
;
1346 if (oi
->delta_base_sha1
)
1347 hashclr(oi
->delta_base_sha1
);
1350 * If we don't care about type or size, then we don't
1351 * need to look inside the object at all. Note that we
1352 * do not optimize out the stat call, even if the
1353 * caller doesn't care about the disk-size, since our
1354 * return value implicitly indicates whether the
1355 * object even exists.
1357 if (!oi
->typep
&& !oi
->type_name
&& !oi
->sizep
&& !oi
->contentp
) {
1360 if (!oi
->disk_sizep
&& (flags
& OBJECT_INFO_QUICK
))
1361 return quick_has_loose(r
, oid
) ? 0 : -1;
1362 if (stat_loose_object(r
, oid
, &st
, &path
) < 0)
1365 *oi
->disk_sizep
= st
.st_size
;
1369 map
= map_loose_object(r
, oid
, &mapsize
);
1374 oi
->sizep
= &size_scratch
;
1377 *oi
->disk_sizep
= mapsize
;
1378 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
)) {
1379 if (unpack_loose_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
1380 status
= error(_("unable to unpack %s header with --allow-unknown-type"),
1382 } else if (unpack_loose_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1383 status
= error(_("unable to unpack %s header"),
1387 else if (hdrbuf
.len
) {
1388 if ((status
= parse_loose_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
1389 status
= error(_("unable to parse %s header with --allow-unknown-type"),
1391 } else if ((status
= parse_loose_header_extended(hdr
, oi
, flags
)) < 0)
1392 status
= error(_("unable to parse %s header"), oid_to_hex(oid
));
1394 if (status
>= 0 && oi
->contentp
) {
1395 *oi
->contentp
= unpack_loose_rest(&stream
, hdr
,
1397 if (!*oi
->contentp
) {
1398 git_inflate_end(&stream
);
1402 git_inflate_end(&stream
);
1404 munmap(map
, mapsize
);
1405 if (status
&& oi
->typep
)
1406 *oi
->typep
= status
;
1407 if (oi
->sizep
== &size_scratch
)
1409 strbuf_release(&hdrbuf
);
1410 oi
->whence
= OI_LOOSE
;
1411 return (status
< 0) ? status
: 0;
1414 int fetch_if_missing
= 1;
1416 int oid_object_info_extended(struct repository
*r
, const struct object_id
*oid
,
1417 struct object_info
*oi
, unsigned flags
)
1419 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
1420 struct pack_entry e
;
1422 const struct object_id
*real
= oid
;
1423 int already_retried
= 0;
1425 if (flags
& OBJECT_INFO_LOOKUP_REPLACE
)
1426 real
= lookup_replace_object(r
, oid
);
1428 if (is_null_oid(real
))
1434 if (!(flags
& OBJECT_INFO_SKIP_CACHED
)) {
1435 struct cached_object
*co
= find_cached_object(real
);
1438 *(oi
->typep
) = co
->type
;
1440 *(oi
->sizep
) = co
->size
;
1442 *(oi
->disk_sizep
) = 0;
1443 if (oi
->delta_base_sha1
)
1444 hashclr(oi
->delta_base_sha1
);
1446 strbuf_addstr(oi
->type_name
, type_name(co
->type
));
1448 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
1449 oi
->whence
= OI_CACHED
;
1455 if (find_pack_entry(r
, real
, &e
))
1458 if (flags
& OBJECT_INFO_IGNORE_LOOSE
)
1461 /* Most likely it's a loose object. */
1462 if (!loose_object_info(r
, real
, oi
, flags
))
1465 /* Not a loose object; someone else may have just packed it. */
1466 if (!(flags
& OBJECT_INFO_QUICK
)) {
1467 reprepare_packed_git(r
);
1468 if (find_pack_entry(r
, real
, &e
))
1472 /* Check if it is a missing object */
1473 if (fetch_if_missing
&& has_promisor_remote() &&
1474 !already_retried
&& r
== the_repository
&&
1475 !(flags
& OBJECT_INFO_SKIP_FETCH_OBJECT
)) {
1477 * TODO Investigate checking promisor_remote_get_direct()
1478 * TODO return value and stopping on error here.
1479 * TODO Pass a repository struct through
1480 * promisor_remote_get_direct(), such that arbitrary
1481 * repositories work.
1483 promisor_remote_get_direct(r
, real
, 1);
1484 already_retried
= 1;
1491 if (oi
== &blank_oi
)
1493 * We know that the caller doesn't actually need the
1494 * information below, so return early.
1497 rtype
= packed_object_info(r
, e
.p
, e
.offset
, oi
);
1499 mark_bad_packed_object(e
.p
, real
->hash
);
1500 return oid_object_info_extended(r
, real
, oi
, 0);
1501 } else if (oi
->whence
== OI_PACKED
) {
1502 oi
->u
.packed
.offset
= e
.offset
;
1503 oi
->u
.packed
.pack
= e
.p
;
1504 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
1505 rtype
== OBJ_OFS_DELTA
);
1511 /* returns enum object_type or negative */
1512 int oid_object_info(struct repository
*r
,
1513 const struct object_id
*oid
,
1514 unsigned long *sizep
)
1516 enum object_type type
;
1517 struct object_info oi
= OBJECT_INFO_INIT
;
1521 if (oid_object_info_extended(r
, oid
, &oi
,
1522 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
1527 static void *read_object(struct repository
*r
,
1528 const struct object_id
*oid
, enum object_type
*type
,
1529 unsigned long *size
)
1531 struct object_info oi
= OBJECT_INFO_INIT
;
1535 oi
.contentp
= &content
;
1537 if (oid_object_info_extended(r
, oid
, &oi
, 0) < 0)
1542 int pretend_object_file(void *buf
, unsigned long len
, enum object_type type
,
1543 struct object_id
*oid
)
1545 struct cached_object
*co
;
1547 hash_object_file(buf
, len
, type_name(type
), oid
);
1548 if (has_object_file(oid
) || find_cached_object(oid
))
1550 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
1551 co
= &cached_objects
[cached_object_nr
++];
1554 co
->buf
= xmalloc(len
);
1555 memcpy(co
->buf
, buf
, len
);
1556 oidcpy(&co
->oid
, oid
);
1561 * This function dies on corrupt objects; the callers who want to
1562 * deal with them should arrange to call read_object() and give error
1563 * messages themselves.
1565 void *read_object_file_extended(struct repository
*r
,
1566 const struct object_id
*oid
,
1567 enum object_type
*type
,
1568 unsigned long *size
,
1572 const struct packed_git
*p
;
1575 const struct object_id
*repl
= lookup_replace
?
1576 lookup_replace_object(r
, oid
) : oid
;
1579 data
= read_object(r
, repl
, type
, size
);
1583 if (errno
&& errno
!= ENOENT
)
1584 die_errno(_("failed to read object %s"), oid_to_hex(oid
));
1586 /* die if we replaced an object with one that does not exist */
1588 die(_("replacement %s not found for %s"),
1589 oid_to_hex(repl
), oid_to_hex(oid
));
1591 if (!stat_loose_object(r
, repl
, &st
, &path
))
1592 die(_("loose object %s (stored in %s) is corrupt"),
1593 oid_to_hex(repl
), path
);
1595 if ((p
= has_packed_and_bad(r
, repl
->hash
)) != NULL
)
1596 die(_("packed object %s (stored in %s) is corrupt"),
1597 oid_to_hex(repl
), p
->pack_name
);
1602 void *read_object_with_reference(struct repository
*r
,
1603 const struct object_id
*oid
,
1604 const char *required_type_name
,
1605 unsigned long *size
,
1606 struct object_id
*actual_oid_return
)
1608 enum object_type type
, required_type
;
1610 unsigned long isize
;
1611 struct object_id actual_oid
;
1613 required_type
= type_from_string(required_type_name
);
1614 oidcpy(&actual_oid
, oid
);
1616 int ref_length
= -1;
1617 const char *ref_type
= NULL
;
1619 buffer
= repo_read_object_file(r
, &actual_oid
, &type
, &isize
);
1622 if (type
== required_type
) {
1624 if (actual_oid_return
)
1625 oidcpy(actual_oid_return
, &actual_oid
);
1628 /* Handle references */
1629 else if (type
== OBJ_COMMIT
)
1631 else if (type
== OBJ_TAG
)
1632 ref_type
= "object ";
1637 ref_length
= strlen(ref_type
);
1639 if (ref_length
+ the_hash_algo
->hexsz
> isize
||
1640 memcmp(buffer
, ref_type
, ref_length
) ||
1641 get_oid_hex((char *) buffer
+ ref_length
, &actual_oid
)) {
1646 /* Now we have the ID of the referred-to object in
1647 * actual_oid. Check again. */
1651 static void write_object_file_prepare(const struct git_hash_algo
*algo
,
1652 const void *buf
, unsigned long len
,
1653 const char *type
, struct object_id
*oid
,
1654 char *hdr
, int *hdrlen
)
1658 /* Generate the header */
1659 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %"PRIuMAX
, type
, (uintmax_t)len
)+1;
1663 algo
->update_fn(&c
, hdr
, *hdrlen
);
1664 algo
->update_fn(&c
, buf
, len
);
1665 algo
->final_fn(oid
->hash
, &c
);
1669 * Move the just written object into its final resting place.
1671 int finalize_object_file(const char *tmpfile
, const char *filename
)
1675 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
1677 else if (link(tmpfile
, filename
))
1681 * Coda hack - coda doesn't like cross-directory links,
1682 * so we fall back to a rename, which will mean that it
1683 * won't be able to check collisions, but that's not a
1686 * The same holds for FAT formatted media.
1688 * When this succeeds, we just return. We have nothing
1691 if (ret
&& ret
!= EEXIST
) {
1693 if (!rename(tmpfile
, filename
))
1697 unlink_or_warn(tmpfile
);
1699 if (ret
!= EEXIST
) {
1700 return error_errno(_("unable to write file %s"), filename
);
1702 /* FIXME!!! Collision check here ? */
1706 if (adjust_shared_perm(filename
))
1707 return error(_("unable to set permission to '%s'"), filename
);
1711 static int write_buffer(int fd
, const void *buf
, size_t len
)
1713 if (write_in_full(fd
, buf
, len
) < 0)
1714 return error_errno(_("file write error"));
1718 int hash_object_file(const void *buf
, unsigned long len
, const char *type
,
1719 struct object_id
*oid
)
1721 char hdr
[MAX_HEADER_LEN
];
1722 int hdrlen
= sizeof(hdr
);
1723 write_object_file_prepare(the_hash_algo
, buf
, len
, type
, oid
, hdr
,
1728 /* Finalize a file on disk, and close it. */
1729 static void close_loose_object(int fd
)
1731 if (fsync_object_files
)
1732 fsync_or_die(fd
, "loose object file");
1734 die_errno(_("error when closing loose object file"));
1737 /* Size of directory component, including the ending '/' */
1738 static inline int directory_size(const char *filename
)
1740 const char *s
= strrchr(filename
, '/');
1743 return s
- filename
+ 1;
1747 * This creates a temporary file in the same directory as the final
1750 * We want to avoid cross-directory filename renames, because those
1751 * can have problems on various filesystems (FAT, NFS, Coda).
1753 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
1755 int fd
, dirlen
= directory_size(filename
);
1758 strbuf_add(tmp
, filename
, dirlen
);
1759 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
1760 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1761 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
1763 * Make sure the directory exists; note that the contents
1764 * of the buffer are undefined after mkstemp returns an
1765 * error, so we have to rewrite the whole buffer from
1769 strbuf_add(tmp
, filename
, dirlen
- 1);
1770 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
1772 if (adjust_shared_perm(tmp
->buf
))
1776 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
1777 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1782 static int write_loose_object(const struct object_id
*oid
, char *hdr
,
1783 int hdrlen
, const void *buf
, unsigned long len
,
1787 unsigned char compressed
[4096];
1790 struct object_id parano_oid
;
1791 static struct strbuf tmp_file
= STRBUF_INIT
;
1792 static struct strbuf filename
= STRBUF_INIT
;
1794 loose_object_path(the_repository
, &filename
, oid
);
1796 fd
= create_tmpfile(&tmp_file
, filename
.buf
);
1798 if (errno
== EACCES
)
1799 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1801 return error_errno(_("unable to create temporary file"));
1805 git_deflate_init(&stream
, zlib_compression_level
);
1806 stream
.next_out
= compressed
;
1807 stream
.avail_out
= sizeof(compressed
);
1808 the_hash_algo
->init_fn(&c
);
1810 /* First header.. */
1811 stream
.next_in
= (unsigned char *)hdr
;
1812 stream
.avail_in
= hdrlen
;
1813 while (git_deflate(&stream
, 0) == Z_OK
)
1815 the_hash_algo
->update_fn(&c
, hdr
, hdrlen
);
1817 /* Then the data itself.. */
1818 stream
.next_in
= (void *)buf
;
1819 stream
.avail_in
= len
;
1821 unsigned char *in0
= stream
.next_in
;
1822 ret
= git_deflate(&stream
, Z_FINISH
);
1823 the_hash_algo
->update_fn(&c
, in0
, stream
.next_in
- in0
);
1824 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
1825 die(_("unable to write loose object file"));
1826 stream
.next_out
= compressed
;
1827 stream
.avail_out
= sizeof(compressed
);
1828 } while (ret
== Z_OK
);
1830 if (ret
!= Z_STREAM_END
)
1831 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid
),
1833 ret
= git_deflate_end_gently(&stream
);
1835 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid
),
1837 the_hash_algo
->final_fn(parano_oid
.hash
, &c
);
1838 if (!oideq(oid
, ¶no_oid
))
1839 die(_("confused by unstable object source data for %s"),
1842 close_loose_object(fd
);
1847 utb
.modtime
= mtime
;
1848 if (utime(tmp_file
.buf
, &utb
) < 0)
1849 warning_errno(_("failed utime() on %s"), tmp_file
.buf
);
1852 return finalize_object_file(tmp_file
.buf
, filename
.buf
);
1855 static int freshen_loose_object(const struct object_id
*oid
)
1857 return check_and_freshen(oid
, 1);
1860 static int freshen_packed_object(const struct object_id
*oid
)
1862 struct pack_entry e
;
1863 if (!find_pack_entry(the_repository
, oid
, &e
))
1867 if (!freshen_file(e
.p
->pack_name
))
1873 int write_object_file(const void *buf
, unsigned long len
, const char *type
,
1874 struct object_id
*oid
)
1876 char hdr
[MAX_HEADER_LEN
];
1877 int hdrlen
= sizeof(hdr
);
1879 /* Normally if we have it in the pack then we do not bother writing
1880 * it out into .git/objects/??/?{38} file.
1882 write_object_file_prepare(the_hash_algo
, buf
, len
, type
, oid
, hdr
,
1884 if (freshen_packed_object(oid
) || freshen_loose_object(oid
))
1886 return write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, 0);
1889 int hash_object_file_literally(const void *buf
, unsigned long len
,
1890 const char *type
, struct object_id
*oid
,
1894 int hdrlen
, status
= 0;
1896 /* type string, SP, %lu of the length plus NUL must fit this */
1897 hdrlen
= strlen(type
) + MAX_HEADER_LEN
;
1898 header
= xmalloc(hdrlen
);
1899 write_object_file_prepare(the_hash_algo
, buf
, len
, type
, oid
, header
,
1902 if (!(flags
& HASH_WRITE_OBJECT
))
1904 if (freshen_packed_object(oid
) || freshen_loose_object(oid
))
1906 status
= write_loose_object(oid
, header
, hdrlen
, buf
, len
, 0);
1913 int force_object_loose(const struct object_id
*oid
, time_t mtime
)
1917 enum object_type type
;
1918 char hdr
[MAX_HEADER_LEN
];
1922 if (has_loose_object(oid
))
1924 buf
= read_object(the_repository
, oid
, &type
, &len
);
1926 return error(_("cannot read object for %s"), oid_to_hex(oid
));
1927 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %"PRIuMAX
, type_name(type
), (uintmax_t)len
) + 1;
1928 ret
= write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, mtime
);
1934 int repo_has_object_file_with_flags(struct repository
*r
,
1935 const struct object_id
*oid
, int flags
)
1937 if (!startup_info
->have_repository
)
1939 return oid_object_info_extended(r
, oid
, NULL
,
1940 flags
| OBJECT_INFO_SKIP_CACHED
) >= 0;
1943 int repo_has_object_file(struct repository
*r
,
1944 const struct object_id
*oid
)
1946 return repo_has_object_file_with_flags(r
, oid
, 0);
1949 static void check_tree(const void *buf
, size_t size
)
1951 struct tree_desc desc
;
1952 struct name_entry entry
;
1954 init_tree_desc(&desc
, buf
, size
);
1955 while (tree_entry(&desc
, &entry
))
1957 * tree_entry() will die() on malformed entries */
1961 static void check_commit(const void *buf
, size_t size
)
1964 memset(&c
, 0, sizeof(c
));
1965 if (parse_commit_buffer(the_repository
, &c
, buf
, size
, 0))
1966 die(_("corrupt commit"));
1969 static void check_tag(const void *buf
, size_t size
)
1972 memset(&t
, 0, sizeof(t
));
1973 if (parse_tag_buffer(the_repository
, &t
, buf
, size
))
1974 die(_("corrupt tag"));
1977 static int index_mem(struct index_state
*istate
,
1978 struct object_id
*oid
, void *buf
, size_t size
,
1979 enum object_type type
,
1980 const char *path
, unsigned flags
)
1982 int ret
, re_allocated
= 0;
1983 int write_object
= flags
& HASH_WRITE_OBJECT
;
1989 * Convert blobs to git internal format
1991 if ((type
== OBJ_BLOB
) && path
) {
1992 struct strbuf nbuf
= STRBUF_INIT
;
1993 if (convert_to_git(istate
, path
, buf
, size
, &nbuf
,
1994 get_conv_flags(flags
))) {
1995 buf
= strbuf_detach(&nbuf
, &size
);
1999 if (flags
& HASH_FORMAT_CHECK
) {
2000 if (type
== OBJ_TREE
)
2001 check_tree(buf
, size
);
2002 if (type
== OBJ_COMMIT
)
2003 check_commit(buf
, size
);
2004 if (type
== OBJ_TAG
)
2005 check_tag(buf
, size
);
2009 ret
= write_object_file(buf
, size
, type_name(type
), oid
);
2011 ret
= hash_object_file(buf
, size
, type_name(type
), oid
);
2017 static int index_stream_convert_blob(struct index_state
*istate
,
2018 struct object_id
*oid
,
2024 const int write_object
= flags
& HASH_WRITE_OBJECT
;
2025 struct strbuf sbuf
= STRBUF_INIT
;
2028 assert(would_convert_to_git_filter_fd(istate
, path
));
2030 convert_to_git_filter_fd(istate
, path
, fd
, &sbuf
,
2031 get_conv_flags(flags
));
2034 ret
= write_object_file(sbuf
.buf
, sbuf
.len
, type_name(OBJ_BLOB
),
2037 ret
= hash_object_file(sbuf
.buf
, sbuf
.len
, type_name(OBJ_BLOB
),
2039 strbuf_release(&sbuf
);
2043 static int index_pipe(struct index_state
*istate
, struct object_id
*oid
,
2044 int fd
, enum object_type type
,
2045 const char *path
, unsigned flags
)
2047 struct strbuf sbuf
= STRBUF_INIT
;
2050 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2051 ret
= index_mem(istate
, oid
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
2054 strbuf_release(&sbuf
);
2058 #define SMALL_FILE_SIZE (32*1024)
2060 static int index_core(struct index_state
*istate
,
2061 struct object_id
*oid
, int fd
, size_t size
,
2062 enum object_type type
, const char *path
,
2068 ret
= index_mem(istate
, oid
, "", size
, type
, path
, flags
);
2069 } else if (size
<= SMALL_FILE_SIZE
) {
2070 char *buf
= xmalloc(size
);
2071 ssize_t read_result
= read_in_full(fd
, buf
, size
);
2072 if (read_result
< 0)
2073 ret
= error_errno(_("read error while indexing %s"),
2074 path
? path
: "<unknown>");
2075 else if (read_result
!= size
)
2076 ret
= error(_("short read while indexing %s"),
2077 path
? path
: "<unknown>");
2079 ret
= index_mem(istate
, oid
, buf
, size
, type
, path
, flags
);
2082 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2083 ret
= index_mem(istate
, oid
, buf
, size
, type
, path
, flags
);
2090 * This creates one packfile per large blob unless bulk-checkin
2091 * machinery is "plugged".
2093 * This also bypasses the usual "convert-to-git" dance, and that is on
2094 * purpose. We could write a streaming version of the converting
2095 * functions and insert that before feeding the data to fast-import
2096 * (or equivalent in-core API described above). However, that is
2097 * somewhat complicated, as we do not know the size of the filter
2098 * result, which we need to know beforehand when writing a git object.
2099 * Since the primary motivation for trying to stream from the working
2100 * tree file and to avoid mmaping it in core is to deal with large
2101 * binary blobs, they generally do not want to get any conversion, and
2102 * callers should avoid this code path when filters are requested.
2104 static int index_stream(struct object_id
*oid
, int fd
, size_t size
,
2105 enum object_type type
, const char *path
,
2108 return index_bulk_checkin(oid
, fd
, size
, type
, path
, flags
);
2111 int index_fd(struct index_state
*istate
, struct object_id
*oid
,
2112 int fd
, struct stat
*st
,
2113 enum object_type type
, const char *path
, unsigned flags
)
2118 * Call xsize_t() only when needed to avoid potentially unnecessary
2119 * die() for large files.
2121 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(istate
, path
))
2122 ret
= index_stream_convert_blob(istate
, oid
, fd
, path
, flags
);
2123 else if (!S_ISREG(st
->st_mode
))
2124 ret
= index_pipe(istate
, oid
, fd
, type
, path
, flags
);
2125 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
2126 (path
&& would_convert_to_git(istate
, path
)))
2127 ret
= index_core(istate
, oid
, fd
, xsize_t(st
->st_size
),
2130 ret
= index_stream(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
2136 int index_path(struct index_state
*istate
, struct object_id
*oid
,
2137 const char *path
, struct stat
*st
, unsigned flags
)
2140 struct strbuf sb
= STRBUF_INIT
;
2143 switch (st
->st_mode
& S_IFMT
) {
2145 fd
= open(path
, O_RDONLY
);
2147 return error_errno("open(\"%s\")", path
);
2148 if (index_fd(istate
, oid
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
2149 return error(_("%s: failed to insert into database"),
2153 if (strbuf_readlink(&sb
, path
, st
->st_size
))
2154 return error_errno("readlink(\"%s\")", path
);
2155 if (!(flags
& HASH_WRITE_OBJECT
))
2156 hash_object_file(sb
.buf
, sb
.len
, blob_type
, oid
);
2157 else if (write_object_file(sb
.buf
, sb
.len
, blob_type
, oid
))
2158 rc
= error(_("%s: failed to insert into database"), path
);
2159 strbuf_release(&sb
);
2162 return resolve_gitlink_ref(path
, "HEAD", oid
);
2164 return error(_("%s: unsupported file type"), path
);
2169 int read_pack_header(int fd
, struct pack_header
*header
)
2171 if (read_in_full(fd
, header
, sizeof(*header
)) != sizeof(*header
))
2172 /* "eof before pack header was fully read" */
2173 return PH_ERROR_EOF
;
2175 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2176 /* "protocol error (pack signature mismatch detected)" */
2177 return PH_ERROR_PACK_SIGNATURE
;
2178 if (!pack_version_ok(header
->hdr_version
))
2179 /* "protocol error (pack version unsupported)" */
2180 return PH_ERROR_PROTOCOL
;
2184 void assert_oid_type(const struct object_id
*oid
, enum object_type expect
)
2186 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
2188 die(_("%s is not a valid object"), oid_to_hex(oid
));
2190 die(_("%s is not a valid '%s' object"), oid_to_hex(oid
),
2194 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
2195 struct strbuf
*path
,
2196 each_loose_object_fn obj_cb
,
2197 each_loose_cruft_fn cruft_cb
,
2198 each_loose_subdir_fn subdir_cb
,
2201 size_t origlen
, baselen
;
2205 struct object_id oid
;
2207 if (subdir_nr
> 0xff)
2208 BUG("invalid loose object subdirectory: %x", subdir_nr
);
2210 origlen
= path
->len
;
2211 strbuf_complete(path
, '/');
2212 strbuf_addf(path
, "%02x", subdir_nr
);
2214 dir
= opendir(path
->buf
);
2216 if (errno
!= ENOENT
)
2217 r
= error_errno(_("unable to open %s"), path
->buf
);
2218 strbuf_setlen(path
, origlen
);
2222 oid
.hash
[0] = subdir_nr
;
2223 strbuf_addch(path
, '/');
2224 baselen
= path
->len
;
2226 while ((de
= readdir(dir
))) {
2228 if (is_dot_or_dotdot(de
->d_name
))
2231 namelen
= strlen(de
->d_name
);
2232 strbuf_setlen(path
, baselen
);
2233 strbuf_add(path
, de
->d_name
, namelen
);
2234 if (namelen
== the_hash_algo
->hexsz
- 2 &&
2235 !hex_to_bytes(oid
.hash
+ 1, de
->d_name
,
2236 the_hash_algo
->rawsz
- 1)) {
2238 r
= obj_cb(&oid
, path
->buf
, data
);
2246 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
2253 strbuf_setlen(path
, baselen
- 1);
2254 if (!r
&& subdir_cb
)
2255 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
2257 strbuf_setlen(path
, origlen
);
2262 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
2263 each_loose_object_fn obj_cb
,
2264 each_loose_cruft_fn cruft_cb
,
2265 each_loose_subdir_fn subdir_cb
,
2271 for (i
= 0; i
< 256; i
++) {
2272 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
2281 int for_each_loose_file_in_objdir(const char *path
,
2282 each_loose_object_fn obj_cb
,
2283 each_loose_cruft_fn cruft_cb
,
2284 each_loose_subdir_fn subdir_cb
,
2287 struct strbuf buf
= STRBUF_INIT
;
2290 strbuf_addstr(&buf
, path
);
2291 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
2293 strbuf_release(&buf
);
2298 int for_each_loose_object(each_loose_object_fn cb
, void *data
,
2299 enum for_each_object_flags flags
)
2301 struct object_directory
*odb
;
2303 prepare_alt_odb(the_repository
);
2304 for (odb
= the_repository
->objects
->odb
; odb
; odb
= odb
->next
) {
2305 int r
= for_each_loose_file_in_objdir(odb
->path
, cb
, NULL
,
2310 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
2317 static int append_loose_object(const struct object_id
*oid
, const char *path
,
2320 oid_array_append(data
, oid
);
2324 struct oid_array
*odb_loose_cache(struct object_directory
*odb
,
2325 const struct object_id
*oid
)
2327 int subdir_nr
= oid
->hash
[0];
2328 struct strbuf buf
= STRBUF_INIT
;
2330 if (subdir_nr
< 0 ||
2331 subdir_nr
>= ARRAY_SIZE(odb
->loose_objects_subdir_seen
))
2332 BUG("subdir_nr out of range");
2334 if (odb
->loose_objects_subdir_seen
[subdir_nr
])
2335 return &odb
->loose_objects_cache
[subdir_nr
];
2337 strbuf_addstr(&buf
, odb
->path
);
2338 for_each_file_in_obj_subdir(subdir_nr
, &buf
,
2339 append_loose_object
,
2341 &odb
->loose_objects_cache
[subdir_nr
]);
2342 odb
->loose_objects_subdir_seen
[subdir_nr
] = 1;
2343 strbuf_release(&buf
);
2344 return &odb
->loose_objects_cache
[subdir_nr
];
2347 void odb_clear_loose_cache(struct object_directory
*odb
)
2351 for (i
= 0; i
< ARRAY_SIZE(odb
->loose_objects_cache
); i
++)
2352 oid_array_clear(&odb
->loose_objects_cache
[i
]);
2353 memset(&odb
->loose_objects_subdir_seen
, 0,
2354 sizeof(odb
->loose_objects_subdir_seen
));
2357 static int check_stream_oid(git_zstream
*stream
,
2361 const struct object_id
*expected_oid
)
2364 struct object_id real_oid
;
2365 unsigned char buf
[4096];
2366 unsigned long total_read
;
2369 the_hash_algo
->init_fn(&c
);
2370 the_hash_algo
->update_fn(&c
, hdr
, stream
->total_out
);
2373 * We already read some bytes into hdr, but the ones up to the NUL
2374 * do not count against the object's content size.
2376 total_read
= stream
->total_out
- strlen(hdr
) - 1;
2379 * This size comparison must be "<=" to read the final zlib packets;
2380 * see the comment in unpack_loose_rest for details.
2382 while (total_read
<= size
&&
2384 (status
== Z_BUF_ERROR
&& !stream
->avail_out
))) {
2385 stream
->next_out
= buf
;
2386 stream
->avail_out
= sizeof(buf
);
2387 if (size
- total_read
< stream
->avail_out
)
2388 stream
->avail_out
= size
- total_read
;
2389 status
= git_inflate(stream
, Z_FINISH
);
2390 the_hash_algo
->update_fn(&c
, buf
, stream
->next_out
- buf
);
2391 total_read
+= stream
->next_out
- buf
;
2393 git_inflate_end(stream
);
2395 if (status
!= Z_STREAM_END
) {
2396 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid
));
2399 if (stream
->avail_in
) {
2400 error(_("garbage at end of loose object '%s'"),
2401 oid_to_hex(expected_oid
));
2405 the_hash_algo
->final_fn(real_oid
.hash
, &c
);
2406 if (!oideq(expected_oid
, &real_oid
)) {
2407 error(_("hash mismatch for %s (expected %s)"), path
,
2408 oid_to_hex(expected_oid
));
2415 int read_loose_object(const char *path
,
2416 const struct object_id
*expected_oid
,
2417 enum object_type
*type
,
2418 unsigned long *size
,
2423 unsigned long mapsize
;
2425 char hdr
[MAX_HEADER_LEN
];
2429 map
= map_loose_object_1(the_repository
, path
, NULL
, &mapsize
);
2431 error_errno(_("unable to mmap %s"), path
);
2435 if (unpack_loose_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0) {
2436 error(_("unable to unpack header of %s"), path
);
2440 *type
= parse_loose_header(hdr
, size
);
2442 error(_("unable to parse header of %s"), path
);
2443 git_inflate_end(&stream
);
2447 if (*type
== OBJ_BLOB
&& *size
> big_file_threshold
) {
2448 if (check_stream_oid(&stream
, hdr
, *size
, path
, expected_oid
) < 0)
2451 *contents
= unpack_loose_rest(&stream
, hdr
, *size
, expected_oid
);
2453 error(_("unable to unpack contents of %s"), path
);
2454 git_inflate_end(&stream
);
2457 if (check_object_signature(expected_oid
, *contents
,
2458 *size
, type_name(*type
))) {
2459 error(_("hash mismatch for %s (expected %s)"), path
,
2460 oid_to_hex(expected_oid
));
2466 ret
= 0; /* everything checks out */
2470 munmap(map
, mapsize
);