2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git 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 "hash-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"
35 #include "submodule.h"
37 /* The maximum size for an object header. */
38 #define MAX_HEADER_LEN 32
41 #define EMPTY_TREE_SHA1_BIN_LITERAL \
42 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
43 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
44 #define EMPTY_TREE_SHA256_BIN_LITERAL \
45 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
46 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
47 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
50 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
51 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
52 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
53 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
54 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
55 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
56 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
59 static const struct object_id empty_tree_oid
= {
60 .hash
= EMPTY_TREE_SHA1_BIN_LITERAL
,
61 .algo
= GIT_HASH_SHA1
,
63 static const struct object_id empty_blob_oid
= {
64 .hash
= EMPTY_BLOB_SHA1_BIN_LITERAL
,
65 .algo
= GIT_HASH_SHA1
,
67 static const struct object_id null_oid_sha1
= {
69 .algo
= GIT_HASH_SHA1
,
71 static const struct object_id empty_tree_oid_sha256
= {
72 .hash
= EMPTY_TREE_SHA256_BIN_LITERAL
,
73 .algo
= GIT_HASH_SHA256
,
75 static const struct object_id empty_blob_oid_sha256
= {
76 .hash
= EMPTY_BLOB_SHA256_BIN_LITERAL
,
77 .algo
= GIT_HASH_SHA256
,
79 static const struct object_id null_oid_sha256
= {
81 .algo
= GIT_HASH_SHA256
,
84 static void git_hash_sha1_init(git_hash_ctx
*ctx
)
86 git_SHA1_Init(&ctx
->sha1
);
89 static void git_hash_sha1_clone(git_hash_ctx
*dst
, const git_hash_ctx
*src
)
91 git_SHA1_Clone(&dst
->sha1
, &src
->sha1
);
94 static void git_hash_sha1_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
96 git_SHA1_Update(&ctx
->sha1
, data
, len
);
99 static void git_hash_sha1_final(unsigned char *hash
, git_hash_ctx
*ctx
)
101 git_SHA1_Final(hash
, &ctx
->sha1
);
104 static void git_hash_sha1_final_oid(struct object_id
*oid
, git_hash_ctx
*ctx
)
106 git_SHA1_Final(oid
->hash
, &ctx
->sha1
);
107 memset(oid
->hash
+ GIT_SHA1_RAWSZ
, 0, GIT_MAX_RAWSZ
- GIT_SHA1_RAWSZ
);
108 oid
->algo
= GIT_HASH_SHA1
;
112 static void git_hash_sha256_init(git_hash_ctx
*ctx
)
114 git_SHA256_Init(&ctx
->sha256
);
117 static void git_hash_sha256_clone(git_hash_ctx
*dst
, const git_hash_ctx
*src
)
119 git_SHA256_Clone(&dst
->sha256
, &src
->sha256
);
122 static void git_hash_sha256_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
124 git_SHA256_Update(&ctx
->sha256
, data
, len
);
127 static void git_hash_sha256_final(unsigned char *hash
, git_hash_ctx
*ctx
)
129 git_SHA256_Final(hash
, &ctx
->sha256
);
132 static void git_hash_sha256_final_oid(struct object_id
*oid
, git_hash_ctx
*ctx
)
134 git_SHA256_Final(oid
->hash
, &ctx
->sha256
);
136 * This currently does nothing, so the compiler should optimize it out,
137 * but keep it in case we extend the hash size again.
139 memset(oid
->hash
+ GIT_SHA256_RAWSZ
, 0, GIT_MAX_RAWSZ
- GIT_SHA256_RAWSZ
);
140 oid
->algo
= GIT_HASH_SHA256
;
143 static void git_hash_unknown_init(git_hash_ctx
*ctx
)
145 BUG("trying to init unknown hash");
148 static void git_hash_unknown_clone(git_hash_ctx
*dst
, const git_hash_ctx
*src
)
150 BUG("trying to clone unknown hash");
153 static void git_hash_unknown_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
155 BUG("trying to update unknown hash");
158 static void git_hash_unknown_final(unsigned char *hash
, git_hash_ctx
*ctx
)
160 BUG("trying to finalize unknown hash");
163 static void git_hash_unknown_final_oid(struct object_id
*oid
, git_hash_ctx
*ctx
)
165 BUG("trying to finalize unknown hash");
168 const struct git_hash_algo hash_algos
[GIT_HASH_NALGOS
] = {
175 git_hash_unknown_init
,
176 git_hash_unknown_clone
,
177 git_hash_unknown_update
,
178 git_hash_unknown_final
,
179 git_hash_unknown_final_oid
,
192 git_hash_sha1_update
,
194 git_hash_sha1_final_oid
,
201 GIT_SHA256_FORMAT_ID
,
205 git_hash_sha256_init
,
206 git_hash_sha256_clone
,
207 git_hash_sha256_update
,
208 git_hash_sha256_final
,
209 git_hash_sha256_final_oid
,
210 &empty_tree_oid_sha256
,
211 &empty_blob_oid_sha256
,
216 const struct object_id
*null_oid(void)
218 return the_hash_algo
->null_oid
;
221 const char *empty_tree_oid_hex(void)
223 static char buf
[GIT_MAX_HEXSZ
+ 1];
224 return oid_to_hex_r(buf
, the_hash_algo
->empty_tree
);
227 const char *empty_blob_oid_hex(void)
229 static char buf
[GIT_MAX_HEXSZ
+ 1];
230 return oid_to_hex_r(buf
, the_hash_algo
->empty_blob
);
233 int hash_algo_by_name(const char *name
)
237 return GIT_HASH_UNKNOWN
;
238 for (i
= 1; i
< GIT_HASH_NALGOS
; i
++)
239 if (!strcmp(name
, hash_algos
[i
].name
))
241 return GIT_HASH_UNKNOWN
;
244 int hash_algo_by_id(uint32_t format_id
)
247 for (i
= 1; i
< GIT_HASH_NALGOS
; i
++)
248 if (format_id
== hash_algos
[i
].format_id
)
250 return GIT_HASH_UNKNOWN
;
253 int hash_algo_by_length(int len
)
256 for (i
= 1; i
< GIT_HASH_NALGOS
; i
++)
257 if (len
== hash_algos
[i
].rawsz
)
259 return GIT_HASH_UNKNOWN
;
263 * This is meant to hold a *small* number of objects that you would
264 * want read_object_file() to be able to return, but yet you do not want
265 * to write them into the object store (e.g. a browse-only
268 static struct cached_object
{
269 struct object_id oid
;
270 enum object_type type
;
274 static int cached_object_nr
, cached_object_alloc
;
276 static struct cached_object empty_tree
= {
277 { EMPTY_TREE_SHA1_BIN_LITERAL
},
283 static struct cached_object
*find_cached_object(const struct object_id
*oid
)
286 struct cached_object
*co
= cached_objects
;
288 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
289 if (oideq(&co
->oid
, oid
))
292 if (oideq(oid
, the_hash_algo
->empty_tree
))
298 static int get_conv_flags(unsigned flags
)
300 if (flags
& HASH_RENORMALIZE
)
301 return CONV_EOL_RENORMALIZE
;
302 else if (flags
& HASH_WRITE_OBJECT
)
303 return global_conv_flags_eol
| CONV_WRITE_OBJECT
;
309 int mkdir_in_gitdir(const char *path
)
311 if (mkdir(path
, 0777)) {
312 int saved_errno
= errno
;
314 struct strbuf sb
= STRBUF_INIT
;
319 * Are we looking at a path in a symlinked worktree
320 * whose original repository does not yet have it?
321 * e.g. .git/rr-cache pointing at its original
322 * repository in which the user hasn't performed any
323 * conflict resolution yet?
325 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
326 strbuf_readlink(&sb
, path
, st
.st_size
) ||
327 !is_absolute_path(sb
.buf
) ||
328 mkdir(sb
.buf
, 0777)) {
335 return adjust_shared_perm(path
);
338 static enum scld_error
safe_create_leading_directories_1(char *path
, int share
)
340 char *next_component
= path
+ offset_1st_component(path
);
341 enum scld_error ret
= SCLD_OK
;
343 while (ret
== SCLD_OK
&& next_component
) {
345 char *slash
= next_component
, slash_character
;
347 while (*slash
&& !is_dir_sep(*slash
))
353 next_component
= slash
+ 1;
354 while (is_dir_sep(*next_component
))
356 if (!*next_component
)
359 slash_character
= *slash
;
361 if (!stat(path
, &st
)) {
363 if (!S_ISDIR(st
.st_mode
)) {
367 } else if (mkdir(path
, 0777)) {
368 if (errno
== EEXIST
&&
369 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
370 ; /* somebody created it since we checked */
371 else if (errno
== ENOENT
)
373 * Either mkdir() failed because
374 * somebody just pruned the containing
375 * directory, or stat() failed because
376 * the file that was in our way was
377 * just removed. Either way, inform
378 * the caller that it might be worth
384 } else if (share
&& adjust_shared_perm(path
)) {
387 *slash
= slash_character
;
392 enum scld_error
safe_create_leading_directories(char *path
)
394 return safe_create_leading_directories_1(path
, 1);
397 enum scld_error
safe_create_leading_directories_no_share(char *path
)
399 return safe_create_leading_directories_1(path
, 0);
402 enum scld_error
safe_create_leading_directories_const(const char *path
)
405 /* path points to cache entries, so xstrdup before messing with it */
406 char *buf
= xstrdup(path
);
407 enum scld_error result
= safe_create_leading_directories(buf
);
415 static void fill_loose_path(struct strbuf
*buf
, const struct object_id
*oid
)
418 for (i
= 0; i
< the_hash_algo
->rawsz
; i
++) {
419 static char hex
[] = "0123456789abcdef";
420 unsigned int val
= oid
->hash
[i
];
421 strbuf_addch(buf
, hex
[val
>> 4]);
422 strbuf_addch(buf
, hex
[val
& 0xf]);
424 strbuf_addch(buf
, '/');
428 static const char *odb_loose_path(struct object_directory
*odb
,
430 const struct object_id
*oid
)
433 strbuf_addstr(buf
, odb
->path
);
434 strbuf_addch(buf
, '/');
435 fill_loose_path(buf
, oid
);
439 const char *loose_object_path(struct repository
*r
, struct strbuf
*buf
,
440 const struct object_id
*oid
)
442 return odb_loose_path(r
->objects
->odb
, buf
, oid
);
446 * Return non-zero iff the path is usable as an alternate object database.
448 static int alt_odb_usable(struct raw_object_store
*o
,
450 const char *normalized_objdir
, khiter_t
*pos
)
454 /* Detect cases where alternate disappeared */
455 if (!is_directory(path
->buf
)) {
456 error(_("object directory %s does not exist; "
457 "check .git/objects/info/alternates"),
463 * Prevent the common mistake of listing the same
464 * thing twice, or object directory itself.
466 if (!o
->odb_by_path
) {
469 o
->odb_by_path
= kh_init_odb_path_map();
470 assert(!o
->odb
->next
);
471 p
= kh_put_odb_path_map(o
->odb_by_path
, o
->odb
->path
, &r
);
472 assert(r
== 1); /* never used */
473 kh_value(o
->odb_by_path
, p
) = o
->odb
;
475 if (fspatheq(path
->buf
, normalized_objdir
))
477 *pos
= kh_put_odb_path_map(o
->odb_by_path
, path
->buf
, &r
);
478 /* r: 0 = exists, 1 = never used, 2 = deleted */
479 return r
== 0 ? 0 : 1;
483 * Prepare alternate object database registry.
485 * The variable alt_odb_list points at the list of struct
486 * object_directory. The elements on this list come from
487 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
488 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
489 * whose contents is similar to that environment variable but can be
490 * LF separated. Its base points at a statically allocated buffer that
491 * contains "/the/directory/corresponding/to/.git/objects/...", while
492 * its name points just after the slash at the end of ".git/objects/"
493 * in the example above, and has enough space to hold all hex characters
494 * of the object ID, an extra slash for the first level indirection, and
495 * the terminating NUL.
497 static void read_info_alternates(struct repository
*r
,
498 const char *relative_base
,
500 static int link_alt_odb_entry(struct repository
*r
, const struct strbuf
*entry
,
501 const char *relative_base
, int depth
, const char *normalized_objdir
)
503 struct object_directory
*ent
;
504 struct strbuf pathbuf
= STRBUF_INIT
;
507 if (!is_absolute_path(entry
->buf
) && relative_base
) {
508 strbuf_realpath(&pathbuf
, relative_base
, 1);
509 strbuf_addch(&pathbuf
, '/');
511 strbuf_addbuf(&pathbuf
, entry
);
513 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
514 error(_("unable to normalize alternate object path: %s"),
516 strbuf_release(&pathbuf
);
521 * The trailing slash after the directory name is given by
522 * this function at the end. Remove duplicates.
524 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
525 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
527 if (!alt_odb_usable(r
->objects
, &pathbuf
, normalized_objdir
, &pos
)) {
528 strbuf_release(&pathbuf
);
532 CALLOC_ARRAY(ent
, 1);
533 /* pathbuf.buf is already in r->objects->odb_by_path */
534 ent
->path
= strbuf_detach(&pathbuf
, NULL
);
536 /* add the alternate entry */
537 *r
->objects
->odb_tail
= ent
;
538 r
->objects
->odb_tail
= &(ent
->next
);
540 assert(r
->objects
->odb_by_path
);
541 kh_value(r
->objects
->odb_by_path
, pos
) = ent
;
543 /* recursively add alternates */
544 read_info_alternates(r
, ent
->path
, depth
+ 1);
549 static const char *parse_alt_odb_entry(const char *string
,
557 if (*string
== '#') {
558 /* comment; consume up to next separator */
559 end
= strchrnul(string
, sep
);
560 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
562 * quoted path; unquote_c_style has copied the
563 * data for us and set "end". Broken quoting (e.g.,
564 * an entry that doesn't end with a quote) falls
565 * back to the unquoted case below.
568 /* normal, unquoted path */
569 end
= strchrnul(string
, sep
);
570 strbuf_add(out
, string
, end
- string
);
578 static void link_alt_odb_entries(struct repository
*r
, const char *alt
,
579 int sep
, const char *relative_base
, int depth
)
581 struct strbuf objdirbuf
= STRBUF_INIT
;
582 struct strbuf entry
= STRBUF_INIT
;
588 error(_("%s: ignoring alternate object stores, nesting too deep"),
593 strbuf_add_absolute_path(&objdirbuf
, r
->objects
->odb
->path
);
594 if (strbuf_normalize_path(&objdirbuf
) < 0)
595 die(_("unable to normalize object directory: %s"),
599 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
602 link_alt_odb_entry(r
, &entry
,
603 relative_base
, depth
, objdirbuf
.buf
);
605 strbuf_release(&entry
);
606 strbuf_release(&objdirbuf
);
609 static void read_info_alternates(struct repository
*r
,
610 const char *relative_base
,
614 struct strbuf buf
= STRBUF_INIT
;
616 path
= xstrfmt("%s/info/alternates", relative_base
);
617 if (strbuf_read_file(&buf
, path
, 1024) < 0) {
618 warn_on_fopen_errors(path
);
623 link_alt_odb_entries(r
, buf
.buf
, '\n', relative_base
, depth
);
624 strbuf_release(&buf
);
628 void add_to_alternates_file(const char *reference
)
630 struct lock_file lock
= LOCK_INIT
;
631 char *alts
= git_pathdup("objects/info/alternates");
635 hold_lock_file_for_update(&lock
, alts
, LOCK_DIE_ON_ERROR
);
636 out
= fdopen_lock_file(&lock
, "w");
638 die_errno(_("unable to fdopen alternates lockfile"));
640 in
= fopen(alts
, "r");
642 struct strbuf line
= STRBUF_INIT
;
644 while (strbuf_getline(&line
, in
) != EOF
) {
645 if (!strcmp(reference
, line
.buf
)) {
649 fprintf_or_die(out
, "%s\n", line
.buf
);
652 strbuf_release(&line
);
655 else if (errno
!= ENOENT
)
656 die_errno(_("unable to read alternates file"));
659 rollback_lock_file(&lock
);
661 fprintf_or_die(out
, "%s\n", reference
);
662 if (commit_lock_file(&lock
))
663 die_errno(_("unable to move new alternates file into place"));
664 if (the_repository
->objects
->loaded_alternates
)
665 link_alt_odb_entries(the_repository
, reference
,
671 void add_to_alternates_memory(const char *reference
)
674 * Make sure alternates are initialized, or else our entry may be
675 * overwritten when they are.
677 prepare_alt_odb(the_repository
);
679 link_alt_odb_entries(the_repository
, reference
,
683 struct object_directory
*set_temporary_primary_odb(const char *dir
, int will_destroy
)
685 struct object_directory
*new_odb
;
688 * Make sure alternates are initialized, or else our entry may be
689 * overwritten when they are.
691 prepare_alt_odb(the_repository
);
694 * Make a new primary odb and link the old primary ODB in as an
697 new_odb
= xcalloc(1, sizeof(*new_odb
));
698 new_odb
->path
= xstrdup(dir
);
701 * Disable ref updates while a temporary odb is active, since
702 * the objects in the database may roll back.
704 new_odb
->disable_ref_updates
= 1;
705 new_odb
->will_destroy
= will_destroy
;
706 new_odb
->next
= the_repository
->objects
->odb
;
707 the_repository
->objects
->odb
= new_odb
;
708 return new_odb
->next
;
711 void restore_primary_odb(struct object_directory
*restore_odb
, const char *old_path
)
713 struct object_directory
*cur_odb
= the_repository
->objects
->odb
;
715 if (strcmp(old_path
, cur_odb
->path
))
716 BUG("expected %s as primary object store; found %s",
717 old_path
, cur_odb
->path
);
719 if (cur_odb
->next
!= restore_odb
)
720 BUG("we expect the old primary object store to be the first alternate");
722 the_repository
->objects
->odb
= restore_odb
;
723 free_object_directory(cur_odb
);
727 * Compute the exact path an alternate is at and returns it. In case of
728 * error NULL is returned and the human readable error is added to `err`
729 * `path` may be relative and should point to $GIT_DIR.
730 * `err` must not be null.
732 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
734 char *ref_git
= NULL
;
738 ref_git
= real_pathdup(path
, 0);
741 strbuf_addf(err
, _("path '%s' does not exist"), path
);
745 repo
= read_gitfile(ref_git
);
747 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
750 ref_git
= xstrdup(repo
);
753 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
754 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
756 ref_git
= ref_git_git
;
757 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
758 struct strbuf sb
= STRBUF_INIT
;
760 if (get_common_dir(&sb
, ref_git
)) {
762 _("reference repository '%s' as a linked "
763 "checkout is not supported yet."),
768 strbuf_addf(err
, _("reference repository '%s' is not a "
769 "local repository."), path
);
773 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
774 strbuf_addf(err
, _("reference repository '%s' is shallow"),
780 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
782 _("reference repository '%s' is grafted"),
790 FREE_AND_NULL(ref_git
);
796 struct object_directory
*find_odb(struct repository
*r
, const char *obj_dir
)
798 struct object_directory
*odb
;
799 char *obj_dir_real
= real_pathdup(obj_dir
, 1);
800 struct strbuf odb_path_real
= STRBUF_INIT
;
803 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
804 strbuf_realpath(&odb_path_real
, odb
->path
, 1);
805 if (!strcmp(obj_dir_real
, odb_path_real
.buf
))
810 strbuf_release(&odb_path_real
);
813 die(_("could not find object directory matching %s"), obj_dir
);
817 static void fill_alternate_refs_command(struct child_process
*cmd
,
818 const char *repo_path
)
822 if (!git_config_get_value("core.alternateRefsCommand", &value
)) {
825 strvec_push(&cmd
->args
, value
);
826 strvec_push(&cmd
->args
, repo_path
);
830 strvec_pushf(&cmd
->args
, "--git-dir=%s", repo_path
);
831 strvec_push(&cmd
->args
, "for-each-ref");
832 strvec_push(&cmd
->args
, "--format=%(objectname)");
834 if (!git_config_get_value("core.alternateRefsPrefixes", &value
)) {
835 strvec_push(&cmd
->args
, "--");
836 strvec_split(&cmd
->args
, value
);
840 strvec_pushv(&cmd
->env_array
, (const char **)local_repo_env
);
844 static void read_alternate_refs(const char *path
,
845 alternate_ref_fn
*cb
,
848 struct child_process cmd
= CHILD_PROCESS_INIT
;
849 struct strbuf line
= STRBUF_INIT
;
852 fill_alternate_refs_command(&cmd
, path
);
854 if (start_command(&cmd
))
857 fh
= xfdopen(cmd
.out
, "r");
858 while (strbuf_getline_lf(&line
, fh
) != EOF
) {
859 struct object_id oid
;
862 if (parse_oid_hex(line
.buf
, &oid
, &p
) || *p
) {
863 warning(_("invalid line while parsing alternate refs: %s"),
872 finish_command(&cmd
);
873 strbuf_release(&line
);
876 struct alternate_refs_data
{
877 alternate_ref_fn
*fn
;
881 static int refs_from_alternate_cb(struct object_directory
*e
,
884 struct strbuf path
= STRBUF_INIT
;
886 struct alternate_refs_data
*cb
= data
;
888 if (!strbuf_realpath(&path
, e
->path
, 0))
890 if (!strbuf_strip_suffix(&path
, "/objects"))
894 /* Is this a git repository with refs? */
895 strbuf_addstr(&path
, "/refs");
896 if (!is_directory(path
.buf
))
898 strbuf_setlen(&path
, base_len
);
900 read_alternate_refs(path
.buf
, cb
->fn
, cb
->data
);
903 strbuf_release(&path
);
907 void for_each_alternate_ref(alternate_ref_fn fn
, void *data
)
909 struct alternate_refs_data cb
;
912 foreach_alt_odb(refs_from_alternate_cb
, &cb
);
915 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
917 struct object_directory
*ent
;
920 prepare_alt_odb(the_repository
);
921 for (ent
= the_repository
->objects
->odb
->next
; ent
; ent
= ent
->next
) {
929 void prepare_alt_odb(struct repository
*r
)
931 if (r
->objects
->loaded_alternates
)
934 link_alt_odb_entries(r
, r
->objects
->alternate_db
, PATH_SEP
, NULL
, 0);
936 read_info_alternates(r
, r
->objects
->odb
->path
, 0);
937 r
->objects
->loaded_alternates
= 1;
940 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
941 static int freshen_file(const char *fn
)
943 return !utime(fn
, NULL
);
947 * All of the check_and_freshen functions return 1 if the file exists and was
948 * freshened (if freshening was requested), 0 otherwise. If they return
949 * 0, you should not assume that it is safe to skip a write of the object (it
950 * either does not exist on disk, or has a stale mtime and may be subject to
953 int check_and_freshen_file(const char *fn
, int freshen
)
955 if (access(fn
, F_OK
))
957 if (freshen
&& !freshen_file(fn
))
962 static int check_and_freshen_odb(struct object_directory
*odb
,
963 const struct object_id
*oid
,
966 static struct strbuf path
= STRBUF_INIT
;
967 odb_loose_path(odb
, &path
, oid
);
968 return check_and_freshen_file(path
.buf
, freshen
);
971 static int check_and_freshen_local(const struct object_id
*oid
, int freshen
)
973 return check_and_freshen_odb(the_repository
->objects
->odb
, oid
, freshen
);
976 static int check_and_freshen_nonlocal(const struct object_id
*oid
, int freshen
)
978 struct object_directory
*odb
;
980 prepare_alt_odb(the_repository
);
981 for (odb
= the_repository
->objects
->odb
->next
; odb
; odb
= odb
->next
) {
982 if (check_and_freshen_odb(odb
, oid
, freshen
))
988 static int check_and_freshen(const struct object_id
*oid
, int freshen
)
990 return check_and_freshen_local(oid
, freshen
) ||
991 check_and_freshen_nonlocal(oid
, freshen
);
994 int has_loose_object_nonlocal(const struct object_id
*oid
)
996 return check_and_freshen_nonlocal(oid
, 0);
999 static int has_loose_object(const struct object_id
*oid
)
1001 return check_and_freshen(oid
, 0);
1004 static void mmap_limit_check(size_t length
)
1006 static size_t limit
= 0;
1008 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
1013 die(_("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
),
1014 (uintmax_t)length
, (uintmax_t)limit
);
1017 void *xmmap_gently(void *start
, size_t length
,
1018 int prot
, int flags
, int fd
, off_t offset
)
1022 mmap_limit_check(length
);
1023 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
1024 if (ret
== MAP_FAILED
&& !length
)
1029 const char *mmap_os_err(void)
1031 static const char blank
[] = "";
1032 #if defined(__linux__)
1033 if (errno
== ENOMEM
) {
1034 /* this continues an existing error message: */
1035 static const char enomem
[] =
1036 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
1039 #endif /* OS-specific bits */
1043 void *xmmap(void *start
, size_t length
,
1044 int prot
, int flags
, int fd
, off_t offset
)
1046 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
1047 if (ret
== MAP_FAILED
)
1048 die_errno(_("mmap failed%s"), mmap_os_err());
1053 * With an in-core object data in "map", rehash it to make sure the
1054 * object name actually matches "oid" to detect object corruption.
1055 * With "map" == NULL, try reading the object named with "oid" using
1056 * the streaming interface and rehash it to do the same.
1058 int check_object_signature(struct repository
*r
, const struct object_id
*oid
,
1059 void *map
, unsigned long size
, const char *type
,
1060 struct object_id
*real_oidp
)
1062 struct object_id tmp
;
1063 struct object_id
*real_oid
= real_oidp
? real_oidp
: &tmp
;
1064 enum object_type obj_type
;
1065 struct git_istream
*st
;
1067 char hdr
[MAX_HEADER_LEN
];
1071 hash_object_file(r
->hash_algo
, map
, size
, type
, real_oid
);
1072 return !oideq(oid
, real_oid
) ? -1 : 0;
1075 st
= open_istream(r
, oid
, &obj_type
, &size
, NULL
);
1079 /* Generate the header */
1080 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %"PRIuMAX
, type_name(obj_type
), (uintmax_t)size
) + 1;
1083 r
->hash_algo
->init_fn(&c
);
1084 r
->hash_algo
->update_fn(&c
, hdr
, hdrlen
);
1086 char buf
[1024 * 16];
1087 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1095 r
->hash_algo
->update_fn(&c
, buf
, readlen
);
1097 r
->hash_algo
->final_oid_fn(real_oid
, &c
);
1099 return !oideq(oid
, real_oid
) ? -1 : 0;
1102 int git_open_cloexec(const char *name
, int flags
)
1105 static int o_cloexec
= O_CLOEXEC
;
1107 fd
= open(name
, flags
| o_cloexec
);
1108 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
1109 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1110 o_cloexec
&= ~O_CLOEXEC
;
1111 fd
= open(name
, flags
| o_cloexec
);
1114 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1116 static int fd_cloexec
= FD_CLOEXEC
;
1118 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
1119 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1120 int flags
= fcntl(fd
, F_GETFD
);
1121 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
1130 * Find "oid" as a loose object in the local repository or in an alternate.
1131 * Returns 0 on success, negative on failure.
1133 * The "path" out-parameter will give the path of the object we found (if any).
1134 * Note that it may point to static storage and is only valid until another
1135 * call to stat_loose_object().
1137 static int stat_loose_object(struct repository
*r
, const struct object_id
*oid
,
1138 struct stat
*st
, const char **path
)
1140 struct object_directory
*odb
;
1141 static struct strbuf buf
= STRBUF_INIT
;
1144 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1145 *path
= odb_loose_path(odb
, &buf
, oid
);
1146 if (!lstat(*path
, st
))
1154 * Like stat_loose_object(), but actually open the object and return the
1155 * descriptor. See the caveats on the "path" parameter above.
1157 static int open_loose_object(struct repository
*r
,
1158 const struct object_id
*oid
, const char **path
)
1161 struct object_directory
*odb
;
1162 int most_interesting_errno
= ENOENT
;
1163 static struct strbuf buf
= STRBUF_INIT
;
1166 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1167 *path
= odb_loose_path(odb
, &buf
, oid
);
1168 fd
= git_open(*path
);
1172 if (most_interesting_errno
== ENOENT
)
1173 most_interesting_errno
= errno
;
1175 errno
= most_interesting_errno
;
1179 static int quick_has_loose(struct repository
*r
,
1180 const struct object_id
*oid
)
1182 struct object_directory
*odb
;
1185 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
1186 if (oidtree_contains(odb_loose_cache(odb
, oid
), oid
))
1193 * Map the loose object at "path" if it is not NULL, or the path found by
1194 * searching for a loose object named "oid".
1196 static void *map_loose_object_1(struct repository
*r
, const char *path
,
1197 const struct object_id
*oid
, unsigned long *size
)
1203 fd
= git_open(path
);
1205 fd
= open_loose_object(r
, oid
, &path
);
1210 if (!fstat(fd
, &st
)) {
1211 *size
= xsize_t(st
.st_size
);
1213 /* mmap() is forbidden on empty files */
1214 error(_("object file %s is empty"), path
);
1218 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1225 void *map_loose_object(struct repository
*r
,
1226 const struct object_id
*oid
,
1227 unsigned long *size
)
1229 return map_loose_object_1(r
, NULL
, oid
, size
);
1232 enum unpack_loose_header_result
unpack_loose_header(git_zstream
*stream
,
1234 unsigned long mapsize
,
1236 unsigned long bufsiz
,
1237 struct strbuf
*header
)
1241 /* Get the data stream */
1242 memset(stream
, 0, sizeof(*stream
));
1243 stream
->next_in
= map
;
1244 stream
->avail_in
= mapsize
;
1245 stream
->next_out
= buffer
;
1246 stream
->avail_out
= bufsiz
;
1248 git_inflate_init(stream
);
1250 status
= git_inflate(stream
, 0);
1256 * Check if entire header is unpacked in the first iteration.
1258 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1262 * We have a header longer than MAX_HEADER_LEN. The "header"
1263 * here is only non-NULL when we run "cat-file
1264 * --allow-unknown-type".
1267 return ULHR_TOO_LONG
;
1270 * buffer[0..bufsiz] was not large enough. Copy the partial
1271 * result out to header, and then append the result of further
1272 * reading the stream.
1274 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1275 stream
->next_out
= buffer
;
1276 stream
->avail_out
= bufsiz
;
1280 status
= git_inflate(stream
, 0);
1282 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1283 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1285 stream
->next_out
= buffer
;
1286 stream
->avail_out
= bufsiz
;
1287 } while (status
!= Z_STREAM_END
);
1288 return ULHR_TOO_LONG
;
1291 static void *unpack_loose_rest(git_zstream
*stream
,
1292 void *buffer
, unsigned long size
,
1293 const struct object_id
*oid
)
1295 int bytes
= strlen(buffer
) + 1;
1296 unsigned char *buf
= xmallocz(size
);
1300 n
= stream
->total_out
- bytes
;
1303 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1305 if (bytes
<= size
) {
1307 * The above condition must be (bytes <= size), not
1308 * (bytes < size). In other words, even though we
1309 * expect no more output and set avail_out to zero,
1310 * the input zlib stream may have bytes that express
1311 * "this concludes the stream", and we *do* want to
1314 * Otherwise we would not be able to test that we
1315 * consumed all the input to reach the expected size;
1316 * we also want to check that zlib tells us that all
1317 * went well with status == Z_STREAM_END at the end.
1319 stream
->next_out
= buf
+ bytes
;
1320 stream
->avail_out
= size
- bytes
;
1321 while (status
== Z_OK
) {
1323 status
= git_inflate(stream
, Z_FINISH
);
1327 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1328 git_inflate_end(stream
);
1333 error(_("corrupt loose object '%s'"), oid_to_hex(oid
));
1334 else if (stream
->avail_in
)
1335 error(_("garbage at end of loose object '%s'"),
1342 * We used to just use "sscanf()", but that's actually way
1343 * too permissive for what we want to check. So do an anal
1344 * object header parse by hand.
1346 int parse_loose_header(const char *hdr
, struct object_info
*oi
)
1348 const char *type_buf
= hdr
;
1350 int type
, type_len
= 0;
1353 * The type can be of any size but is followed by
1365 type
= type_from_string_gently(type_buf
, type_len
, 1);
1367 strbuf_add(oi
->type_name
, type_buf
, type_len
);
1372 * The length must follow immediately, and be in canonical
1373 * decimal format (ie "010" is not valid).
1375 size
= *hdr
++ - '0';
1380 unsigned long c
= *hdr
- '0';
1384 size
= st_add(st_mult(size
, 10), c
);
1389 *oi
->sizep
= cast_size_t_to_ulong(size
);
1392 * The length must be followed by a zero byte
1398 * The format is valid, but the type may still be bogus. The
1399 * Caller needs to check its oi->typep.
1404 static int loose_object_info(struct repository
*r
,
1405 const struct object_id
*oid
,
1406 struct object_info
*oi
, int flags
)
1409 unsigned long mapsize
;
1412 char hdr
[MAX_HEADER_LEN
];
1413 struct strbuf hdrbuf
= STRBUF_INIT
;
1414 unsigned long size_scratch
;
1415 enum object_type type_scratch
;
1416 int allow_unknown
= flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
;
1418 if (oi
->delta_base_oid
)
1419 oidclr(oi
->delta_base_oid
);
1422 * If we don't care about type or size, then we don't
1423 * need to look inside the object at all. Note that we
1424 * do not optimize out the stat call, even if the
1425 * caller doesn't care about the disk-size, since our
1426 * return value implicitly indicates whether the
1427 * object even exists.
1429 if (!oi
->typep
&& !oi
->type_name
&& !oi
->sizep
&& !oi
->contentp
) {
1432 if (!oi
->disk_sizep
&& (flags
& OBJECT_INFO_QUICK
))
1433 return quick_has_loose(r
, oid
) ? 0 : -1;
1434 if (stat_loose_object(r
, oid
, &st
, &path
) < 0)
1437 *oi
->disk_sizep
= st
.st_size
;
1441 map
= map_loose_object(r
, oid
, &mapsize
);
1446 oi
->sizep
= &size_scratch
;
1448 oi
->typep
= &type_scratch
;
1451 *oi
->disk_sizep
= mapsize
;
1453 switch (unpack_loose_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
),
1454 allow_unknown
? &hdrbuf
: NULL
)) {
1456 if (parse_loose_header(hdrbuf
.len
? hdrbuf
.buf
: hdr
, oi
) < 0)
1457 status
= error(_("unable to parse %s header"), oid_to_hex(oid
));
1458 else if (!allow_unknown
&& *oi
->typep
< 0)
1459 die(_("invalid object type"));
1463 *oi
->contentp
= unpack_loose_rest(&stream
, hdr
, *oi
->sizep
, oid
);
1470 status
= error(_("unable to unpack %s header"),
1474 status
= error(_("header for %s too long, exceeds %d bytes"),
1475 oid_to_hex(oid
), MAX_HEADER_LEN
);
1479 git_inflate_end(&stream
);
1481 munmap(map
, mapsize
);
1482 if (oi
->sizep
== &size_scratch
)
1484 strbuf_release(&hdrbuf
);
1485 if (oi
->typep
== &type_scratch
)
1487 oi
->whence
= OI_LOOSE
;
1491 int obj_read_use_lock
= 0;
1492 pthread_mutex_t obj_read_mutex
;
1494 void enable_obj_read_lock(void)
1496 if (obj_read_use_lock
)
1499 obj_read_use_lock
= 1;
1500 init_recursive_mutex(&obj_read_mutex
);
1503 void disable_obj_read_lock(void)
1505 if (!obj_read_use_lock
)
1508 obj_read_use_lock
= 0;
1509 pthread_mutex_destroy(&obj_read_mutex
);
1512 int fetch_if_missing
= 1;
1514 static int do_oid_object_info_extended(struct repository
*r
,
1515 const struct object_id
*oid
,
1516 struct object_info
*oi
, unsigned flags
)
1518 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
1519 struct cached_object
*co
;
1520 struct pack_entry e
;
1522 const struct object_id
*real
= oid
;
1523 int already_retried
= 0;
1526 if (flags
& OBJECT_INFO_LOOKUP_REPLACE
)
1527 real
= lookup_replace_object(r
, oid
);
1529 if (is_null_oid(real
))
1535 co
= find_cached_object(real
);
1538 *(oi
->typep
) = co
->type
;
1540 *(oi
->sizep
) = co
->size
;
1542 *(oi
->disk_sizep
) = 0;
1543 if (oi
->delta_base_oid
)
1544 oidclr(oi
->delta_base_oid
);
1546 strbuf_addstr(oi
->type_name
, type_name(co
->type
));
1548 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
1549 oi
->whence
= OI_CACHED
;
1554 if (find_pack_entry(r
, real
, &e
))
1557 if (flags
& OBJECT_INFO_IGNORE_LOOSE
)
1560 /* Most likely it's a loose object. */
1561 if (!loose_object_info(r
, real
, oi
, flags
))
1564 /* Not a loose object; someone else may have just packed it. */
1565 if (!(flags
& OBJECT_INFO_QUICK
)) {
1566 reprepare_packed_git(r
);
1567 if (find_pack_entry(r
, real
, &e
))
1572 * If r is the_repository, this might be an attempt at
1573 * accessing a submodule object as if it were in the_repository
1574 * (having called add_submodule_odb() on that submodule's ODB).
1575 * If any such ODBs exist, register them and try again.
1577 if (r
== the_repository
&&
1578 register_all_submodule_odb_as_alternates())
1579 /* We added some alternates; retry */
1582 /* Check if it is a missing object */
1583 if (fetch_if_missing
&& repo_has_promisor_remote(r
) &&
1585 !(flags
& OBJECT_INFO_SKIP_FETCH_OBJECT
)) {
1587 * TODO Investigate checking promisor_remote_get_direct()
1588 * TODO return value and stopping on error here.
1590 promisor_remote_get_direct(r
, real
, 1);
1591 already_retried
= 1;
1598 if (oi
== &blank_oi
)
1600 * We know that the caller doesn't actually need the
1601 * information below, so return early.
1604 rtype
= packed_object_info(r
, e
.p
, e
.offset
, oi
);
1606 mark_bad_packed_object(e
.p
, real
);
1607 return do_oid_object_info_extended(r
, real
, oi
, 0);
1608 } else if (oi
->whence
== OI_PACKED
) {
1609 oi
->u
.packed
.offset
= e
.offset
;
1610 oi
->u
.packed
.pack
= e
.p
;
1611 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
1612 rtype
== OBJ_OFS_DELTA
);
1618 int oid_object_info_extended(struct repository
*r
, const struct object_id
*oid
,
1619 struct object_info
*oi
, unsigned flags
)
1623 ret
= do_oid_object_info_extended(r
, oid
, oi
, flags
);
1629 /* returns enum object_type or negative */
1630 int oid_object_info(struct repository
*r
,
1631 const struct object_id
*oid
,
1632 unsigned long *sizep
)
1634 enum object_type type
;
1635 struct object_info oi
= OBJECT_INFO_INIT
;
1639 if (oid_object_info_extended(r
, oid
, &oi
,
1640 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
1645 static void *read_object(struct repository
*r
,
1646 const struct object_id
*oid
, enum object_type
*type
,
1647 unsigned long *size
)
1649 struct object_info oi
= OBJECT_INFO_INIT
;
1653 oi
.contentp
= &content
;
1655 if (oid_object_info_extended(r
, oid
, &oi
, 0) < 0)
1660 int pretend_object_file(void *buf
, unsigned long len
, enum object_type type
,
1661 struct object_id
*oid
)
1663 struct cached_object
*co
;
1665 hash_object_file(the_hash_algo
, buf
, len
, type_name(type
), oid
);
1666 if (has_object_file_with_flags(oid
, OBJECT_INFO_QUICK
| OBJECT_INFO_SKIP_FETCH_OBJECT
) ||
1667 find_cached_object(oid
))
1669 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
1670 co
= &cached_objects
[cached_object_nr
++];
1673 co
->buf
= xmalloc(len
);
1674 memcpy(co
->buf
, buf
, len
);
1675 oidcpy(&co
->oid
, oid
);
1680 * This function dies on corrupt objects; the callers who want to
1681 * deal with them should arrange to call read_object() and give error
1682 * messages themselves.
1684 void *read_object_file_extended(struct repository
*r
,
1685 const struct object_id
*oid
,
1686 enum object_type
*type
,
1687 unsigned long *size
,
1691 const struct packed_git
*p
;
1694 const struct object_id
*repl
= lookup_replace
?
1695 lookup_replace_object(r
, oid
) : oid
;
1698 data
= read_object(r
, repl
, type
, size
);
1703 if (errno
&& errno
!= ENOENT
)
1704 die_errno(_("failed to read object %s"), oid_to_hex(oid
));
1706 /* die if we replaced an object with one that does not exist */
1708 die(_("replacement %s not found for %s"),
1709 oid_to_hex(repl
), oid_to_hex(oid
));
1711 if (!stat_loose_object(r
, repl
, &st
, &path
))
1712 die(_("loose object %s (stored in %s) is corrupt"),
1713 oid_to_hex(repl
), path
);
1715 if ((p
= has_packed_and_bad(r
, repl
)) != NULL
)
1716 die(_("packed object %s (stored in %s) is corrupt"),
1717 oid_to_hex(repl
), p
->pack_name
);
1723 void *read_object_with_reference(struct repository
*r
,
1724 const struct object_id
*oid
,
1725 const char *required_type_name
,
1726 unsigned long *size
,
1727 struct object_id
*actual_oid_return
)
1729 enum object_type type
, required_type
;
1731 unsigned long isize
;
1732 struct object_id actual_oid
;
1734 required_type
= type_from_string(required_type_name
);
1735 oidcpy(&actual_oid
, oid
);
1737 int ref_length
= -1;
1738 const char *ref_type
= NULL
;
1740 buffer
= repo_read_object_file(r
, &actual_oid
, &type
, &isize
);
1743 if (type
== required_type
) {
1745 if (actual_oid_return
)
1746 oidcpy(actual_oid_return
, &actual_oid
);
1749 /* Handle references */
1750 else if (type
== OBJ_COMMIT
)
1752 else if (type
== OBJ_TAG
)
1753 ref_type
= "object ";
1758 ref_length
= strlen(ref_type
);
1760 if (ref_length
+ the_hash_algo
->hexsz
> isize
||
1761 memcmp(buffer
, ref_type
, ref_length
) ||
1762 get_oid_hex((char *) buffer
+ ref_length
, &actual_oid
)) {
1767 /* Now we have the ID of the referred-to object in
1768 * actual_oid. Check again. */
1772 static void write_object_file_prepare(const struct git_hash_algo
*algo
,
1773 const void *buf
, unsigned long len
,
1774 const char *type
, struct object_id
*oid
,
1775 char *hdr
, int *hdrlen
)
1779 /* Generate the header */
1780 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %"PRIuMAX
, type
, (uintmax_t)len
)+1;
1784 algo
->update_fn(&c
, hdr
, *hdrlen
);
1785 algo
->update_fn(&c
, buf
, len
);
1786 algo
->final_oid_fn(oid
, &c
);
1790 * Move the just written object into its final resting place.
1792 int finalize_object_file(const char *tmpfile
, const char *filename
)
1796 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
1798 else if (link(tmpfile
, filename
))
1802 * Coda hack - coda doesn't like cross-directory links,
1803 * so we fall back to a rename, which will mean that it
1804 * won't be able to check collisions, but that's not a
1807 * The same holds for FAT formatted media.
1809 * When this succeeds, we just return. We have nothing
1812 if (ret
&& ret
!= EEXIST
) {
1814 if (!rename(tmpfile
, filename
))
1818 unlink_or_warn(tmpfile
);
1820 if (ret
!= EEXIST
) {
1821 return error_errno(_("unable to write file %s"), filename
);
1823 /* FIXME!!! Collision check here ? */
1827 if (adjust_shared_perm(filename
))
1828 return error(_("unable to set permission to '%s'"), filename
);
1832 static int write_buffer(int fd
, const void *buf
, size_t len
)
1834 if (write_in_full(fd
, buf
, len
) < 0)
1835 return error_errno(_("file write error"));
1839 int hash_object_file(const struct git_hash_algo
*algo
, const void *buf
,
1840 unsigned long len
, const char *type
,
1841 struct object_id
*oid
)
1843 char hdr
[MAX_HEADER_LEN
];
1844 int hdrlen
= sizeof(hdr
);
1845 write_object_file_prepare(algo
, buf
, len
, type
, oid
, hdr
, &hdrlen
);
1849 /* Finalize a file on disk, and close it. */
1850 static void close_loose_object(int fd
)
1852 if (!the_repository
->objects
->odb
->will_destroy
) {
1853 if (fsync_object_files
)
1854 fsync_or_die(fd
, "loose object file");
1858 die_errno(_("error when closing loose object file"));
1861 /* Size of directory component, including the ending '/' */
1862 static inline int directory_size(const char *filename
)
1864 const char *s
= strrchr(filename
, '/');
1867 return s
- filename
+ 1;
1871 * This creates a temporary file in the same directory as the final
1874 * We want to avoid cross-directory filename renames, because those
1875 * can have problems on various filesystems (FAT, NFS, Coda).
1877 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
1879 int fd
, dirlen
= directory_size(filename
);
1882 strbuf_add(tmp
, filename
, dirlen
);
1883 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
1884 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1885 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
1887 * Make sure the directory exists; note that the contents
1888 * of the buffer are undefined after mkstemp returns an
1889 * error, so we have to rewrite the whole buffer from
1893 strbuf_add(tmp
, filename
, dirlen
- 1);
1894 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
1896 if (adjust_shared_perm(tmp
->buf
))
1900 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
1901 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1906 static int write_loose_object(const struct object_id
*oid
, char *hdr
,
1907 int hdrlen
, const void *buf
, unsigned long len
,
1908 time_t mtime
, unsigned flags
)
1911 unsigned char compressed
[4096];
1914 struct object_id parano_oid
;
1915 static struct strbuf tmp_file
= STRBUF_INIT
;
1916 static struct strbuf filename
= STRBUF_INIT
;
1918 loose_object_path(the_repository
, &filename
, oid
);
1920 fd
= create_tmpfile(&tmp_file
, filename
.buf
);
1922 if (flags
& HASH_SILENT
)
1924 else if (errno
== EACCES
)
1925 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1927 return error_errno(_("unable to create temporary file"));
1931 git_deflate_init(&stream
, zlib_compression_level
);
1932 stream
.next_out
= compressed
;
1933 stream
.avail_out
= sizeof(compressed
);
1934 the_hash_algo
->init_fn(&c
);
1936 /* First header.. */
1937 stream
.next_in
= (unsigned char *)hdr
;
1938 stream
.avail_in
= hdrlen
;
1939 while (git_deflate(&stream
, 0) == Z_OK
)
1941 the_hash_algo
->update_fn(&c
, hdr
, hdrlen
);
1943 /* Then the data itself.. */
1944 stream
.next_in
= (void *)buf
;
1945 stream
.avail_in
= len
;
1947 unsigned char *in0
= stream
.next_in
;
1948 ret
= git_deflate(&stream
, Z_FINISH
);
1949 the_hash_algo
->update_fn(&c
, in0
, stream
.next_in
- in0
);
1950 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
1951 die(_("unable to write loose object file"));
1952 stream
.next_out
= compressed
;
1953 stream
.avail_out
= sizeof(compressed
);
1954 } while (ret
== Z_OK
);
1956 if (ret
!= Z_STREAM_END
)
1957 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid
),
1959 ret
= git_deflate_end_gently(&stream
);
1961 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid
),
1963 the_hash_algo
->final_oid_fn(¶no_oid
, &c
);
1964 if (!oideq(oid
, ¶no_oid
))
1965 die(_("confused by unstable object source data for %s"),
1968 close_loose_object(fd
);
1973 utb
.modtime
= mtime
;
1974 if (utime(tmp_file
.buf
, &utb
) < 0 &&
1975 !(flags
& HASH_SILENT
))
1976 warning_errno(_("failed utime() on %s"), tmp_file
.buf
);
1979 return finalize_object_file(tmp_file
.buf
, filename
.buf
);
1982 static int freshen_loose_object(const struct object_id
*oid
)
1984 return check_and_freshen(oid
, 1);
1987 static int freshen_packed_object(const struct object_id
*oid
)
1989 struct pack_entry e
;
1990 if (!find_pack_entry(the_repository
, oid
, &e
))
1994 if (!freshen_file(e
.p
->pack_name
))
2000 int write_object_file_flags(const void *buf
, unsigned long len
,
2001 const char *type
, struct object_id
*oid
,
2004 char hdr
[MAX_HEADER_LEN
];
2005 int hdrlen
= sizeof(hdr
);
2007 /* Normally if we have it in the pack then we do not bother writing
2008 * it out into .git/objects/??/?{38} file.
2010 write_object_file_prepare(the_hash_algo
, buf
, len
, type
, oid
, hdr
,
2012 if (freshen_packed_object(oid
) || freshen_loose_object(oid
))
2014 return write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, 0, flags
);
2017 int hash_object_file_literally(const void *buf
, unsigned long len
,
2018 const char *type
, struct object_id
*oid
,
2022 int hdrlen
, status
= 0;
2024 /* type string, SP, %lu of the length plus NUL must fit this */
2025 hdrlen
= strlen(type
) + MAX_HEADER_LEN
;
2026 header
= xmalloc(hdrlen
);
2027 write_object_file_prepare(the_hash_algo
, buf
, len
, type
, oid
, header
,
2030 if (!(flags
& HASH_WRITE_OBJECT
))
2032 if (freshen_packed_object(oid
) || freshen_loose_object(oid
))
2034 status
= write_loose_object(oid
, header
, hdrlen
, buf
, len
, 0, 0);
2041 int force_object_loose(const struct object_id
*oid
, time_t mtime
)
2045 enum object_type type
;
2046 char hdr
[MAX_HEADER_LEN
];
2050 if (has_loose_object(oid
))
2052 buf
= read_object(the_repository
, oid
, &type
, &len
);
2054 return error(_("cannot read object for %s"), oid_to_hex(oid
));
2055 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %"PRIuMAX
, type_name(type
), (uintmax_t)len
) + 1;
2056 ret
= write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, mtime
, 0);
2062 int has_object(struct repository
*r
, const struct object_id
*oid
,
2065 int quick
= !(flags
& HAS_OBJECT_RECHECK_PACKED
);
2066 unsigned object_info_flags
= OBJECT_INFO_SKIP_FETCH_OBJECT
|
2067 (quick
? OBJECT_INFO_QUICK
: 0);
2069 if (!startup_info
->have_repository
)
2071 return oid_object_info_extended(r
, oid
, NULL
, object_info_flags
) >= 0;
2074 int repo_has_object_file_with_flags(struct repository
*r
,
2075 const struct object_id
*oid
, int flags
)
2077 if (!startup_info
->have_repository
)
2079 return oid_object_info_extended(r
, oid
, NULL
, flags
) >= 0;
2082 int repo_has_object_file(struct repository
*r
,
2083 const struct object_id
*oid
)
2085 return repo_has_object_file_with_flags(r
, oid
, 0);
2088 static void check_tree(const void *buf
, size_t size
)
2090 struct tree_desc desc
;
2091 struct name_entry entry
;
2093 init_tree_desc(&desc
, buf
, size
);
2094 while (tree_entry(&desc
, &entry
))
2096 * tree_entry() will die() on malformed entries */
2100 static void check_commit(const void *buf
, size_t size
)
2103 memset(&c
, 0, sizeof(c
));
2104 if (parse_commit_buffer(the_repository
, &c
, buf
, size
, 0))
2105 die(_("corrupt commit"));
2108 static void check_tag(const void *buf
, size_t size
)
2111 memset(&t
, 0, sizeof(t
));
2112 if (parse_tag_buffer(the_repository
, &t
, buf
, size
))
2113 die(_("corrupt tag"));
2116 static int index_mem(struct index_state
*istate
,
2117 struct object_id
*oid
, void *buf
, size_t size
,
2118 enum object_type type
,
2119 const char *path
, unsigned flags
)
2121 int ret
, re_allocated
= 0;
2122 int write_object
= flags
& HASH_WRITE_OBJECT
;
2128 * Convert blobs to git internal format
2130 if ((type
== OBJ_BLOB
) && path
) {
2131 struct strbuf nbuf
= STRBUF_INIT
;
2132 if (convert_to_git(istate
, path
, buf
, size
, &nbuf
,
2133 get_conv_flags(flags
))) {
2134 buf
= strbuf_detach(&nbuf
, &size
);
2138 if (flags
& HASH_FORMAT_CHECK
) {
2139 if (type
== OBJ_TREE
)
2140 check_tree(buf
, size
);
2141 if (type
== OBJ_COMMIT
)
2142 check_commit(buf
, size
);
2143 if (type
== OBJ_TAG
)
2144 check_tag(buf
, size
);
2148 ret
= write_object_file(buf
, size
, type_name(type
), oid
);
2150 ret
= hash_object_file(the_hash_algo
, buf
, size
,
2151 type_name(type
), oid
);
2157 static int index_stream_convert_blob(struct index_state
*istate
,
2158 struct object_id
*oid
,
2164 const int write_object
= flags
& HASH_WRITE_OBJECT
;
2165 struct strbuf sbuf
= STRBUF_INIT
;
2168 assert(would_convert_to_git_filter_fd(istate
, path
));
2170 convert_to_git_filter_fd(istate
, path
, fd
, &sbuf
,
2171 get_conv_flags(flags
));
2174 ret
= write_object_file(sbuf
.buf
, sbuf
.len
, type_name(OBJ_BLOB
),
2177 ret
= hash_object_file(the_hash_algo
, sbuf
.buf
, sbuf
.len
,
2178 type_name(OBJ_BLOB
), oid
);
2179 strbuf_release(&sbuf
);
2183 static int index_pipe(struct index_state
*istate
, struct object_id
*oid
,
2184 int fd
, enum object_type type
,
2185 const char *path
, unsigned flags
)
2187 struct strbuf sbuf
= STRBUF_INIT
;
2190 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2191 ret
= index_mem(istate
, oid
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
2194 strbuf_release(&sbuf
);
2198 #define SMALL_FILE_SIZE (32*1024)
2200 static int index_core(struct index_state
*istate
,
2201 struct object_id
*oid
, int fd
, size_t size
,
2202 enum object_type type
, const char *path
,
2208 ret
= index_mem(istate
, oid
, "", size
, type
, path
, flags
);
2209 } else if (size
<= SMALL_FILE_SIZE
) {
2210 char *buf
= xmalloc(size
);
2211 ssize_t read_result
= read_in_full(fd
, buf
, size
);
2212 if (read_result
< 0)
2213 ret
= error_errno(_("read error while indexing %s"),
2214 path
? path
: "<unknown>");
2215 else if (read_result
!= size
)
2216 ret
= error(_("short read while indexing %s"),
2217 path
? path
: "<unknown>");
2219 ret
= index_mem(istate
, oid
, buf
, size
, type
, path
, flags
);
2222 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2223 ret
= index_mem(istate
, oid
, buf
, size
, type
, path
, flags
);
2230 * This creates one packfile per large blob unless bulk-checkin
2231 * machinery is "plugged".
2233 * This also bypasses the usual "convert-to-git" dance, and that is on
2234 * purpose. We could write a streaming version of the converting
2235 * functions and insert that before feeding the data to fast-import
2236 * (or equivalent in-core API described above). However, that is
2237 * somewhat complicated, as we do not know the size of the filter
2238 * result, which we need to know beforehand when writing a git object.
2239 * Since the primary motivation for trying to stream from the working
2240 * tree file and to avoid mmaping it in core is to deal with large
2241 * binary blobs, they generally do not want to get any conversion, and
2242 * callers should avoid this code path when filters are requested.
2244 static int index_stream(struct object_id
*oid
, int fd
, size_t size
,
2245 enum object_type type
, const char *path
,
2248 return index_bulk_checkin(oid
, fd
, size
, type
, path
, flags
);
2251 int index_fd(struct index_state
*istate
, struct object_id
*oid
,
2252 int fd
, struct stat
*st
,
2253 enum object_type type
, const char *path
, unsigned flags
)
2258 * Call xsize_t() only when needed to avoid potentially unnecessary
2259 * die() for large files.
2261 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(istate
, path
))
2262 ret
= index_stream_convert_blob(istate
, oid
, fd
, path
, flags
);
2263 else if (!S_ISREG(st
->st_mode
))
2264 ret
= index_pipe(istate
, oid
, fd
, type
, path
, flags
);
2265 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
2266 (path
&& would_convert_to_git(istate
, path
)))
2267 ret
= index_core(istate
, oid
, fd
, xsize_t(st
->st_size
),
2270 ret
= index_stream(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
2276 int index_path(struct index_state
*istate
, struct object_id
*oid
,
2277 const char *path
, struct stat
*st
, unsigned flags
)
2280 struct strbuf sb
= STRBUF_INIT
;
2283 switch (st
->st_mode
& S_IFMT
) {
2285 fd
= open(path
, O_RDONLY
);
2287 return error_errno("open(\"%s\")", path
);
2288 if (index_fd(istate
, oid
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
2289 return error(_("%s: failed to insert into database"),
2293 if (strbuf_readlink(&sb
, path
, st
->st_size
))
2294 return error_errno("readlink(\"%s\")", path
);
2295 if (!(flags
& HASH_WRITE_OBJECT
))
2296 hash_object_file(the_hash_algo
, sb
.buf
, sb
.len
,
2298 else if (write_object_file(sb
.buf
, sb
.len
, blob_type
, oid
))
2299 rc
= error(_("%s: failed to insert into database"), path
);
2300 strbuf_release(&sb
);
2303 return resolve_gitlink_ref(path
, "HEAD", oid
);
2305 return error(_("%s: unsupported file type"), path
);
2310 int read_pack_header(int fd
, struct pack_header
*header
)
2312 if (read_in_full(fd
, header
, sizeof(*header
)) != sizeof(*header
))
2313 /* "eof before pack header was fully read" */
2314 return PH_ERROR_EOF
;
2316 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2317 /* "protocol error (pack signature mismatch detected)" */
2318 return PH_ERROR_PACK_SIGNATURE
;
2319 if (!pack_version_ok(header
->hdr_version
))
2320 /* "protocol error (pack version unsupported)" */
2321 return PH_ERROR_PROTOCOL
;
2325 void assert_oid_type(const struct object_id
*oid
, enum object_type expect
)
2327 enum object_type type
= oid_object_info(the_repository
, oid
, NULL
);
2329 die(_("%s is not a valid object"), oid_to_hex(oid
));
2331 die(_("%s is not a valid '%s' object"), oid_to_hex(oid
),
2335 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
2336 struct strbuf
*path
,
2337 each_loose_object_fn obj_cb
,
2338 each_loose_cruft_fn cruft_cb
,
2339 each_loose_subdir_fn subdir_cb
,
2342 size_t origlen
, baselen
;
2346 struct object_id oid
;
2348 if (subdir_nr
> 0xff)
2349 BUG("invalid loose object subdirectory: %x", subdir_nr
);
2351 origlen
= path
->len
;
2352 strbuf_complete(path
, '/');
2353 strbuf_addf(path
, "%02x", subdir_nr
);
2355 dir
= opendir(path
->buf
);
2357 if (errno
!= ENOENT
)
2358 r
= error_errno(_("unable to open %s"), path
->buf
);
2359 strbuf_setlen(path
, origlen
);
2363 oid
.hash
[0] = subdir_nr
;
2364 strbuf_addch(path
, '/');
2365 baselen
= path
->len
;
2367 while ((de
= readdir_skip_dot_and_dotdot(dir
))) {
2370 namelen
= strlen(de
->d_name
);
2371 strbuf_setlen(path
, baselen
);
2372 strbuf_add(path
, de
->d_name
, namelen
);
2373 if (namelen
== the_hash_algo
->hexsz
- 2 &&
2374 !hex_to_bytes(oid
.hash
+ 1, de
->d_name
,
2375 the_hash_algo
->rawsz
- 1)) {
2376 oid_set_algo(&oid
, the_hash_algo
);
2378 r
= obj_cb(&oid
, path
->buf
, data
);
2386 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
2393 strbuf_setlen(path
, baselen
- 1);
2394 if (!r
&& subdir_cb
)
2395 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
2397 strbuf_setlen(path
, origlen
);
2402 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
2403 each_loose_object_fn obj_cb
,
2404 each_loose_cruft_fn cruft_cb
,
2405 each_loose_subdir_fn subdir_cb
,
2411 for (i
= 0; i
< 256; i
++) {
2412 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
2421 int for_each_loose_file_in_objdir(const char *path
,
2422 each_loose_object_fn obj_cb
,
2423 each_loose_cruft_fn cruft_cb
,
2424 each_loose_subdir_fn subdir_cb
,
2427 struct strbuf buf
= STRBUF_INIT
;
2430 strbuf_addstr(&buf
, path
);
2431 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
2433 strbuf_release(&buf
);
2438 int for_each_loose_object(each_loose_object_fn cb
, void *data
,
2439 enum for_each_object_flags flags
)
2441 struct object_directory
*odb
;
2443 prepare_alt_odb(the_repository
);
2444 for (odb
= the_repository
->objects
->odb
; odb
; odb
= odb
->next
) {
2445 int r
= for_each_loose_file_in_objdir(odb
->path
, cb
, NULL
,
2450 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
2457 static int append_loose_object(const struct object_id
*oid
, const char *path
,
2460 oidtree_insert(data
, oid
);
2464 struct oidtree
*odb_loose_cache(struct object_directory
*odb
,
2465 const struct object_id
*oid
)
2467 int subdir_nr
= oid
->hash
[0];
2468 struct strbuf buf
= STRBUF_INIT
;
2469 size_t word_bits
= bitsizeof(odb
->loose_objects_subdir_seen
[0]);
2470 size_t word_index
= subdir_nr
/ word_bits
;
2471 size_t mask
= (size_t)1u << (subdir_nr
% word_bits
);
2474 if (subdir_nr
< 0 ||
2475 subdir_nr
>= bitsizeof(odb
->loose_objects_subdir_seen
))
2476 BUG("subdir_nr out of range");
2478 bitmap
= &odb
->loose_objects_subdir_seen
[word_index
];
2480 return odb
->loose_objects_cache
;
2481 if (!odb
->loose_objects_cache
) {
2482 ALLOC_ARRAY(odb
->loose_objects_cache
, 1);
2483 oidtree_init(odb
->loose_objects_cache
);
2485 strbuf_addstr(&buf
, odb
->path
);
2486 for_each_file_in_obj_subdir(subdir_nr
, &buf
,
2487 append_loose_object
,
2489 odb
->loose_objects_cache
);
2491 strbuf_release(&buf
);
2492 return odb
->loose_objects_cache
;
2495 void odb_clear_loose_cache(struct object_directory
*odb
)
2497 oidtree_clear(odb
->loose_objects_cache
);
2498 FREE_AND_NULL(odb
->loose_objects_cache
);
2499 memset(&odb
->loose_objects_subdir_seen
, 0,
2500 sizeof(odb
->loose_objects_subdir_seen
));
2503 static int check_stream_oid(git_zstream
*stream
,
2507 const struct object_id
*expected_oid
)
2510 struct object_id real_oid
;
2511 unsigned char buf
[4096];
2512 unsigned long total_read
;
2515 the_hash_algo
->init_fn(&c
);
2516 the_hash_algo
->update_fn(&c
, hdr
, stream
->total_out
);
2519 * We already read some bytes into hdr, but the ones up to the NUL
2520 * do not count against the object's content size.
2522 total_read
= stream
->total_out
- strlen(hdr
) - 1;
2525 * This size comparison must be "<=" to read the final zlib packets;
2526 * see the comment in unpack_loose_rest for details.
2528 while (total_read
<= size
&&
2530 (status
== Z_BUF_ERROR
&& !stream
->avail_out
))) {
2531 stream
->next_out
= buf
;
2532 stream
->avail_out
= sizeof(buf
);
2533 if (size
- total_read
< stream
->avail_out
)
2534 stream
->avail_out
= size
- total_read
;
2535 status
= git_inflate(stream
, Z_FINISH
);
2536 the_hash_algo
->update_fn(&c
, buf
, stream
->next_out
- buf
);
2537 total_read
+= stream
->next_out
- buf
;
2539 git_inflate_end(stream
);
2541 if (status
!= Z_STREAM_END
) {
2542 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid
));
2545 if (stream
->avail_in
) {
2546 error(_("garbage at end of loose object '%s'"),
2547 oid_to_hex(expected_oid
));
2551 the_hash_algo
->final_oid_fn(&real_oid
, &c
);
2552 if (!oideq(expected_oid
, &real_oid
)) {
2553 error(_("hash mismatch for %s (expected %s)"), path
,
2554 oid_to_hex(expected_oid
));
2561 int read_loose_object(const char *path
,
2562 const struct object_id
*expected_oid
,
2563 struct object_id
*real_oid
,
2565 struct object_info
*oi
)
2569 unsigned long mapsize
;
2571 char hdr
[MAX_HEADER_LEN
];
2572 unsigned long *size
= oi
->sizep
;
2574 map
= map_loose_object_1(the_repository
, path
, NULL
, &mapsize
);
2576 error_errno(_("unable to mmap %s"), path
);
2580 if (unpack_loose_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
),
2582 error(_("unable to unpack header of %s"), path
);
2586 if (parse_loose_header(hdr
, oi
) < 0) {
2587 error(_("unable to parse header of %s"), path
);
2588 git_inflate_end(&stream
);
2592 if (*oi
->typep
== OBJ_BLOB
&& *size
> big_file_threshold
) {
2593 if (check_stream_oid(&stream
, hdr
, *size
, path
, expected_oid
) < 0)
2596 *contents
= unpack_loose_rest(&stream
, hdr
, *size
, expected_oid
);
2598 error(_("unable to unpack contents of %s"), path
);
2599 git_inflate_end(&stream
);
2602 if (check_object_signature(the_repository
, expected_oid
,
2604 oi
->type_name
->buf
, real_oid
))
2608 ret
= 0; /* everything checks out */
2612 munmap(map
, mapsize
);