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 "streaming.h"
29 #include "mergesort.h"
33 const unsigned char null_sha1
[20];
34 const struct object_id null_oid
;
35 const struct object_id empty_tree_oid
= {
36 EMPTY_TREE_SHA1_BIN_LITERAL
38 const struct object_id empty_blob_oid
= {
39 EMPTY_BLOB_SHA1_BIN_LITERAL
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
48 static struct cached_object
{
49 unsigned char sha1
[20];
50 enum object_type type
;
54 static int cached_object_nr
, cached_object_alloc
;
56 static struct cached_object empty_tree
= {
57 EMPTY_TREE_SHA1_BIN_LITERAL
,
63 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
66 struct cached_object
*co
= cached_objects
;
68 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
69 if (!hashcmp(co
->sha1
, sha1
))
72 if (!hashcmp(sha1
, empty_tree
.sha1
))
77 int mkdir_in_gitdir(const char *path
)
79 if (mkdir(path
, 0777)) {
80 int saved_errno
= errno
;
82 struct strbuf sb
= STRBUF_INIT
;
87 * Are we looking at a path in a symlinked worktree
88 * whose original repository does not yet have it?
89 * e.g. .git/rr-cache pointing at its original
90 * repository in which the user hasn't performed any
91 * conflict resolution yet?
93 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
94 strbuf_readlink(&sb
, path
, st
.st_size
) ||
95 !is_absolute_path(sb
.buf
) ||
96 mkdir(sb
.buf
, 0777)) {
103 return adjust_shared_perm(path
);
106 enum scld_error
safe_create_leading_directories(char *path
)
108 char *next_component
= path
+ offset_1st_component(path
);
109 enum scld_error ret
= SCLD_OK
;
111 while (ret
== SCLD_OK
&& next_component
) {
113 char *slash
= next_component
, slash_character
;
115 while (*slash
&& !is_dir_sep(*slash
))
121 next_component
= slash
+ 1;
122 while (is_dir_sep(*next_component
))
124 if (!*next_component
)
127 slash_character
= *slash
;
129 if (!stat(path
, &st
)) {
131 if (!S_ISDIR(st
.st_mode
)) {
135 } else if (mkdir(path
, 0777)) {
136 if (errno
== EEXIST
&&
137 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
138 ; /* somebody created it since we checked */
139 else if (errno
== ENOENT
)
141 * Either mkdir() failed because
142 * somebody just pruned the containing
143 * directory, or stat() failed because
144 * the file that was in our way was
145 * just removed. Either way, inform
146 * the caller that it might be worth
152 } else if (adjust_shared_perm(path
)) {
155 *slash
= slash_character
;
160 enum scld_error
safe_create_leading_directories_const(const char *path
)
163 /* path points to cache entries, so xstrdup before messing with it */
164 char *buf
= xstrdup(path
);
165 enum scld_error result
= safe_create_leading_directories(buf
);
173 int raceproof_create_file(const char *path
, create_file_fn fn
, void *cb
)
176 * The number of times we will try to remove empty directories
177 * in the way of path. This is only 1 because if another
178 * process is racily creating directories that conflict with
179 * us, we don't want to fight against them.
181 int remove_directories_remaining
= 1;
184 * The number of times that we will try to create the
185 * directories containing path. We are willing to attempt this
186 * more than once, because another process could be trying to
187 * clean up empty directories at the same time as we are
188 * trying to create them.
190 int create_directories_remaining
= 3;
192 /* A scratch copy of path, filled lazily if we need it: */
193 struct strbuf path_copy
= STRBUF_INIT
;
206 if (errno
== EISDIR
&& remove_directories_remaining
-- > 0) {
208 * A directory is in the way. Maybe it is empty; try
212 strbuf_addstr(&path_copy
, path
);
214 if (!remove_dir_recursively(&path_copy
, REMOVE_DIR_EMPTY_ONLY
))
216 } else if (errno
== ENOENT
&& create_directories_remaining
-- > 0) {
218 * Maybe the containing directory didn't exist, or
219 * maybe it was just deleted by a process that is
220 * racing with us to clean up empty directories. Try
223 enum scld_error scld_result
;
226 strbuf_addstr(&path_copy
, path
);
229 scld_result
= safe_create_leading_directories(path_copy
.buf
);
230 if (scld_result
== SCLD_OK
)
232 } while (scld_result
== SCLD_VANISHED
&& create_directories_remaining
-- > 0);
236 strbuf_release(&path_copy
);
241 static void fill_sha1_path(struct strbuf
*buf
, const unsigned char *sha1
)
244 for (i
= 0; i
< 20; i
++) {
245 static char hex
[] = "0123456789abcdef";
246 unsigned int val
= sha1
[i
];
247 strbuf_addch(buf
, hex
[val
>> 4]);
248 strbuf_addch(buf
, hex
[val
& 0xf]);
250 strbuf_addch(buf
, '/');
254 const char *sha1_file_name(const unsigned char *sha1
)
256 static struct strbuf buf
= STRBUF_INIT
;
259 strbuf_addf(&buf
, "%s/", get_object_directory());
261 fill_sha1_path(&buf
, sha1
);
265 struct strbuf
*alt_scratch_buf(struct alternate_object_database
*alt
)
267 strbuf_setlen(&alt
->scratch
, alt
->base_len
);
268 return &alt
->scratch
;
271 static const char *alt_sha1_path(struct alternate_object_database
*alt
,
272 const unsigned char *sha1
)
274 struct strbuf
*buf
= alt_scratch_buf(alt
);
275 fill_sha1_path(buf
, sha1
);
279 struct alternate_object_database
*alt_odb_list
;
280 static struct alternate_object_database
**alt_odb_tail
;
283 * Return non-zero iff the path is usable as an alternate object database.
285 static int alt_odb_usable(struct strbuf
*path
, const char *normalized_objdir
)
287 struct alternate_object_database
*alt
;
289 /* Detect cases where alternate disappeared */
290 if (!is_directory(path
->buf
)) {
291 error("object directory %s does not exist; "
292 "check .git/objects/info/alternates.",
298 * Prevent the common mistake of listing the same
299 * thing twice, or object directory itself.
301 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
302 if (!fspathcmp(path
->buf
, alt
->path
))
305 if (!fspathcmp(path
->buf
, normalized_objdir
))
312 * Prepare alternate object database registry.
314 * The variable alt_odb_list points at the list of struct
315 * alternate_object_database. The elements on this list come from
316 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
317 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
318 * whose contents is similar to that environment variable but can be
319 * LF separated. Its base points at a statically allocated buffer that
320 * contains "/the/directory/corresponding/to/.git/objects/...", while
321 * its name points just after the slash at the end of ".git/objects/"
322 * in the example above, and has enough space to hold 40-byte hex
323 * SHA1, an extra slash for the first level indirection, and the
326 static void read_info_alternates(const char * relative_base
, int depth
);
327 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
328 int depth
, const char *normalized_objdir
)
330 struct alternate_object_database
*ent
;
331 struct strbuf pathbuf
= STRBUF_INIT
;
333 if (!is_absolute_path(entry
) && relative_base
) {
334 strbuf_realpath(&pathbuf
, relative_base
, 1);
335 strbuf_addch(&pathbuf
, '/');
337 strbuf_addstr(&pathbuf
, entry
);
339 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
340 error("unable to normalize alternate object path: %s",
342 strbuf_release(&pathbuf
);
347 * The trailing slash after the directory name is given by
348 * this function at the end. Remove duplicates.
350 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
351 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
353 if (!alt_odb_usable(&pathbuf
, normalized_objdir
)) {
354 strbuf_release(&pathbuf
);
358 ent
= alloc_alt_odb(pathbuf
.buf
);
360 /* add the alternate entry */
362 alt_odb_tail
= &(ent
->next
);
365 /* recursively add alternates */
366 read_info_alternates(pathbuf
.buf
, depth
+ 1);
368 strbuf_release(&pathbuf
);
372 static const char *parse_alt_odb_entry(const char *string
,
380 if (*string
== '#') {
381 /* comment; consume up to next separator */
382 end
= strchrnul(string
, sep
);
383 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
385 * quoted path; unquote_c_style has copied the
386 * data for us and set "end". Broken quoting (e.g.,
387 * an entry that doesn't end with a quote) falls
388 * back to the unquoted case below.
391 /* normal, unquoted path */
392 end
= strchrnul(string
, sep
);
393 strbuf_add(out
, string
, end
- string
);
401 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
402 const char *relative_base
, int depth
)
404 struct strbuf objdirbuf
= STRBUF_INIT
;
405 struct strbuf entry
= STRBUF_INIT
;
408 error("%s: ignoring alternate object stores, nesting too deep.",
413 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
414 if (strbuf_normalize_path(&objdirbuf
) < 0)
415 die("unable to normalize object directory: %s",
419 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
422 link_alt_odb_entry(entry
.buf
, relative_base
, depth
, objdirbuf
.buf
);
424 strbuf_release(&entry
);
425 strbuf_release(&objdirbuf
);
428 static void read_info_alternates(const char * relative_base
, int depth
)
436 path
= xstrfmt("%s/info/alternates", relative_base
);
441 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
445 mapsz
= xsize_t(st
.st_size
);
446 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
449 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
454 struct alternate_object_database
*alloc_alt_odb(const char *dir
)
456 struct alternate_object_database
*ent
;
458 FLEX_ALLOC_STR(ent
, path
, dir
);
459 strbuf_init(&ent
->scratch
, 0);
460 strbuf_addf(&ent
->scratch
, "%s/", dir
);
461 ent
->base_len
= ent
->scratch
.len
;
466 void add_to_alternates_file(const char *reference
)
468 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
469 char *alts
= git_pathdup("objects/info/alternates");
472 hold_lock_file_for_update(lock
, alts
, LOCK_DIE_ON_ERROR
);
473 out
= fdopen_lock_file(lock
, "w");
475 die_errno("unable to fdopen alternates lockfile");
477 in
= fopen(alts
, "r");
479 struct strbuf line
= STRBUF_INIT
;
482 while (strbuf_getline(&line
, in
) != EOF
) {
483 if (!strcmp(reference
, line
.buf
)) {
487 fprintf_or_die(out
, "%s\n", line
.buf
);
490 strbuf_release(&line
);
494 rollback_lock_file(lock
);
498 else if (errno
!= ENOENT
)
499 die_errno("unable to read alternates file");
502 fprintf_or_die(out
, "%s\n", reference
);
503 if (commit_lock_file(lock
))
504 die_errno("unable to move new alternates file into place");
506 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
511 void add_to_alternates_memory(const char *reference
)
514 * Make sure alternates are initialized, or else our entry may be
515 * overwritten when they are.
519 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
523 * Compute the exact path an alternate is at and returns it. In case of
524 * error NULL is returned and the human readable error is added to `err`
525 * `path` may be relative and should point to $GITDIR.
526 * `err` must not be null.
528 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
530 char *ref_git
= NULL
;
531 const char *repo
, *ref_git_s
;
534 ref_git_s
= real_path_if_valid(path
);
537 strbuf_addf(err
, _("path '%s' does not exist"), path
);
541 * Beware: read_gitfile(), real_path() and mkpath()
542 * return static buffer
544 ref_git
= xstrdup(ref_git_s
);
546 repo
= read_gitfile(ref_git
);
548 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
551 ref_git
= xstrdup(repo
);
554 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
555 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
557 ref_git
= ref_git_git
;
558 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
559 struct strbuf sb
= STRBUF_INIT
;
561 if (get_common_dir(&sb
, ref_git
)) {
563 _("reference repository '%s' as a linked "
564 "checkout is not supported yet."),
569 strbuf_addf(err
, _("reference repository '%s' is not a "
570 "local repository."), path
);
574 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
575 strbuf_addf(err
, _("reference repository '%s' is shallow"),
581 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
583 _("reference repository '%s' is grafted"),
591 FREE_AND_NULL(ref_git
);
597 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
599 struct alternate_object_database
*ent
;
603 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
611 void prepare_alt_odb(void)
618 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
621 alt_odb_tail
= &alt_odb_list
;
622 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
624 read_info_alternates(get_object_directory(), 0);
627 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
628 static int freshen_file(const char *fn
)
631 t
.actime
= t
.modtime
= time(NULL
);
632 return !utime(fn
, &t
);
636 * All of the check_and_freshen functions return 1 if the file exists and was
637 * freshened (if freshening was requested), 0 otherwise. If they return
638 * 0, you should not assume that it is safe to skip a write of the object (it
639 * either does not exist on disk, or has a stale mtime and may be subject to
642 int check_and_freshen_file(const char *fn
, int freshen
)
644 if (access(fn
, F_OK
))
646 if (freshen
&& !freshen_file(fn
))
651 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
653 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
656 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
658 struct alternate_object_database
*alt
;
660 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
661 const char *path
= alt_sha1_path(alt
, sha1
);
662 if (check_and_freshen_file(path
, freshen
))
668 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
670 return check_and_freshen_local(sha1
, freshen
) ||
671 check_and_freshen_nonlocal(sha1
, freshen
);
674 int has_loose_object_nonlocal(const unsigned char *sha1
)
676 return check_and_freshen_nonlocal(sha1
, 0);
679 static int has_loose_object(const unsigned char *sha1
)
681 return check_and_freshen(sha1
, 0);
684 static void mmap_limit_check(size_t length
)
686 static size_t limit
= 0;
688 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
693 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
694 (uintmax_t)length
, (uintmax_t)limit
);
697 void *xmmap_gently(void *start
, size_t length
,
698 int prot
, int flags
, int fd
, off_t offset
)
702 mmap_limit_check(length
);
703 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
704 if (ret
== MAP_FAILED
) {
707 release_pack_memory(length
);
708 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
713 void *xmmap(void *start
, size_t length
,
714 int prot
, int flags
, int fd
, off_t offset
)
716 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
717 if (ret
== MAP_FAILED
)
718 die_errno("mmap failed");
722 void install_packed_git(struct packed_git
*pack
)
724 if (pack
->pack_fd
!= -1)
727 pack
->next
= packed_git
;
731 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
733 static void report_helper(const struct string_list
*list
,
734 int seen_bits
, int first
, int last
)
736 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
739 for (; first
< last
; first
++)
740 report_garbage(seen_bits
, list
->items
[first
].string
);
743 static void report_pack_garbage(struct string_list
*list
)
745 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
750 string_list_sort(list
);
752 for (i
= 0; i
< list
->nr
; i
++) {
753 const char *path
= list
->items
[i
].string
;
755 strncmp(path
, list
->items
[first
].string
, baselen
)) {
756 report_helper(list
, seen_bits
, first
, i
);
761 const char *dot
= strrchr(path
, '.');
763 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
766 baselen
= dot
- path
+ 1;
769 if (!strcmp(path
+ baselen
, "pack"))
771 else if (!strcmp(path
+ baselen
, "idx"))
774 report_helper(list
, seen_bits
, first
, list
->nr
);
777 static void prepare_packed_git_one(char *objdir
, int local
)
779 struct strbuf path
= STRBUF_INIT
;
783 struct string_list garbage
= STRING_LIST_INIT_DUP
;
785 strbuf_addstr(&path
, objdir
);
786 strbuf_addstr(&path
, "/pack");
787 dir
= opendir(path
.buf
);
790 error_errno("unable to open object pack directory: %s",
792 strbuf_release(&path
);
795 strbuf_addch(&path
, '/');
796 dirnamelen
= path
.len
;
797 while ((de
= readdir(dir
)) != NULL
) {
798 struct packed_git
*p
;
801 if (is_dot_or_dotdot(de
->d_name
))
804 strbuf_setlen(&path
, dirnamelen
);
805 strbuf_addstr(&path
, de
->d_name
);
808 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
809 /* Don't reopen a pack we already have. */
810 for (p
= packed_git
; p
; p
= p
->next
) {
812 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
814 !memcmp(p
->pack_name
, path
.buf
, len
))
819 * See if it really is a valid .idx file with
820 * corresponding .pack file that we can map.
822 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
823 install_packed_git(p
);
829 if (ends_with(de
->d_name
, ".idx") ||
830 ends_with(de
->d_name
, ".pack") ||
831 ends_with(de
->d_name
, ".bitmap") ||
832 ends_with(de
->d_name
, ".keep"))
833 string_list_append(&garbage
, path
.buf
);
835 report_garbage(PACKDIR_FILE_GARBAGE
, path
.buf
);
838 report_pack_garbage(&garbage
);
839 string_list_clear(&garbage
, 0);
840 strbuf_release(&path
);
843 static int approximate_object_count_valid
;
846 * Give a fast, rough count of the number of objects in the repository. This
847 * ignores loose objects completely. If you have a lot of them, then either
848 * you should repack because your performance will be awful, or they are
849 * all unreachable objects about to be pruned, in which case they're not really
850 * interesting as a measure of repo size in the first place.
852 unsigned long approximate_object_count(void)
854 static unsigned long count
;
855 if (!approximate_object_count_valid
) {
856 struct packed_git
*p
;
858 prepare_packed_git();
860 for (p
= packed_git
; p
; p
= p
->next
) {
861 if (open_pack_index(p
))
863 count
+= p
->num_objects
;
869 static void *get_next_packed_git(const void *p
)
871 return ((const struct packed_git
*)p
)->next
;
874 static void set_next_packed_git(void *p
, void *next
)
876 ((struct packed_git
*)p
)->next
= next
;
879 static int sort_pack(const void *a_
, const void *b_
)
881 const struct packed_git
*a
= a_
;
882 const struct packed_git
*b
= b_
;
886 * Local packs tend to contain objects specific to our
887 * variant of the project than remote ones. In addition,
888 * remote ones could be on a network mounted filesystem.
889 * Favor local ones for these reasons.
891 st
= a
->pack_local
- b
->pack_local
;
896 * Younger packs tend to contain more recent objects,
897 * and more recent objects tend to get accessed more
900 if (a
->mtime
< b
->mtime
)
902 else if (a
->mtime
== b
->mtime
)
907 static void rearrange_packed_git(void)
909 packed_git
= llist_mergesort(packed_git
, get_next_packed_git
,
910 set_next_packed_git
, sort_pack
);
913 static void prepare_packed_git_mru(void)
915 struct packed_git
*p
;
917 mru_clear(packed_git_mru
);
918 for (p
= packed_git
; p
; p
= p
->next
)
919 mru_append(packed_git_mru
, p
);
922 static int prepare_packed_git_run_once
= 0;
923 void prepare_packed_git(void)
925 struct alternate_object_database
*alt
;
927 if (prepare_packed_git_run_once
)
929 prepare_packed_git_one(get_object_directory(), 1);
931 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
)
932 prepare_packed_git_one(alt
->path
, 0);
933 rearrange_packed_git();
934 prepare_packed_git_mru();
935 prepare_packed_git_run_once
= 1;
938 void reprepare_packed_git(void)
940 approximate_object_count_valid
= 0;
941 prepare_packed_git_run_once
= 0;
942 prepare_packed_git();
945 static void mark_bad_packed_object(struct packed_git
*p
,
946 const unsigned char *sha1
)
949 for (i
= 0; i
< p
->num_bad_objects
; i
++)
950 if (!hashcmp(sha1
, p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* i
))
952 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
953 st_mult(GIT_MAX_RAWSZ
,
954 st_add(p
->num_bad_objects
, 1)));
955 hashcpy(p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* p
->num_bad_objects
, sha1
);
956 p
->num_bad_objects
++;
959 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
961 struct packed_git
*p
;
964 for (p
= packed_git
; p
; p
= p
->next
)
965 for (i
= 0; i
< p
->num_bad_objects
; i
++)
966 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
972 * With an in-core object data in "map", rehash it to make sure the
973 * object name actually matches "sha1" to detect object corruption.
974 * With "map" == NULL, try reading the object named with "sha1" using
975 * the streaming interface and rehash it to do the same.
977 int check_sha1_signature(const unsigned char *sha1
, void *map
,
978 unsigned long size
, const char *type
)
980 unsigned char real_sha1
[20];
981 enum object_type obj_type
;
982 struct git_istream
*st
;
988 hash_sha1_file(map
, size
, type
, real_sha1
);
989 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
992 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
996 /* Generate the header */
997 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
1001 git_SHA1_Update(&c
, hdr
, hdrlen
);
1003 char buf
[1024 * 16];
1004 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1012 git_SHA1_Update(&c
, buf
, readlen
);
1014 git_SHA1_Final(real_sha1
, &c
);
1016 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1019 int git_open_cloexec(const char *name
, int flags
)
1022 static int o_cloexec
= O_CLOEXEC
;
1024 fd
= open(name
, flags
| o_cloexec
);
1025 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
1026 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1027 o_cloexec
&= ~O_CLOEXEC
;
1028 fd
= open(name
, flags
| o_cloexec
);
1031 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1033 static int fd_cloexec
= FD_CLOEXEC
;
1035 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
1036 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1037 int flags
= fcntl(fd
, F_GETFD
);
1038 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
1047 * Find "sha1" as a loose object in the local repository or in an alternate.
1048 * Returns 0 on success, negative on failure.
1050 * The "path" out-parameter will give the path of the object we found (if any).
1051 * Note that it may point to static storage and is only valid until another
1052 * call to sha1_file_name(), etc.
1054 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
,
1057 struct alternate_object_database
*alt
;
1059 *path
= sha1_file_name(sha1
);
1060 if (!lstat(*path
, st
))
1065 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1066 *path
= alt_sha1_path(alt
, sha1
);
1067 if (!lstat(*path
, st
))
1075 * Like stat_sha1_file(), but actually open the object and return the
1076 * descriptor. See the caveats on the "path" parameter above.
1078 static int open_sha1_file(const unsigned char *sha1
, const char **path
)
1081 struct alternate_object_database
*alt
;
1082 int most_interesting_errno
;
1084 *path
= sha1_file_name(sha1
);
1085 fd
= git_open(*path
);
1088 most_interesting_errno
= errno
;
1091 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1092 *path
= alt_sha1_path(alt
, sha1
);
1093 fd
= git_open(*path
);
1096 if (most_interesting_errno
== ENOENT
)
1097 most_interesting_errno
= errno
;
1099 errno
= most_interesting_errno
;
1104 * Map the loose object at "path" if it is not NULL, or the path found by
1105 * searching for a loose object named "sha1".
1107 static void *map_sha1_file_1(const char *path
,
1108 const unsigned char *sha1
,
1109 unsigned long *size
)
1115 fd
= git_open(path
);
1117 fd
= open_sha1_file(sha1
, &path
);
1122 if (!fstat(fd
, &st
)) {
1123 *size
= xsize_t(st
.st_size
);
1125 /* mmap() is forbidden on empty files */
1126 error("object file %s is empty", path
);
1129 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1136 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1138 return map_sha1_file_1(NULL
, sha1
, size
);
1141 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1142 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1145 unsigned long size
, c
;
1146 unsigned long used
= 0;
1149 *type
= (c
>> 4) & 7;
1153 if (len
<= used
|| bitsizeof(long) <= shift
) {
1154 error("bad object header");
1159 size
+= (c
& 0x7f) << shift
;
1166 static int unpack_sha1_short_header(git_zstream
*stream
,
1167 unsigned char *map
, unsigned long mapsize
,
1168 void *buffer
, unsigned long bufsiz
)
1170 /* Get the data stream */
1171 memset(stream
, 0, sizeof(*stream
));
1172 stream
->next_in
= map
;
1173 stream
->avail_in
= mapsize
;
1174 stream
->next_out
= buffer
;
1175 stream
->avail_out
= bufsiz
;
1177 git_inflate_init(stream
);
1178 return git_inflate(stream
, 0);
1181 int unpack_sha1_header(git_zstream
*stream
,
1182 unsigned char *map
, unsigned long mapsize
,
1183 void *buffer
, unsigned long bufsiz
)
1185 int status
= unpack_sha1_short_header(stream
, map
, mapsize
,
1191 /* Make sure we have the terminating NUL */
1192 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1197 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1198 unsigned long mapsize
, void *buffer
,
1199 unsigned long bufsiz
, struct strbuf
*header
)
1203 status
= unpack_sha1_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1208 * Check if entire header is unpacked in the first iteration.
1210 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1214 * buffer[0..bufsiz] was not large enough. Copy the partial
1215 * result out to header, and then append the result of further
1216 * reading the stream.
1218 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1219 stream
->next_out
= buffer
;
1220 stream
->avail_out
= bufsiz
;
1223 status
= git_inflate(stream
, 0);
1224 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1225 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1227 stream
->next_out
= buffer
;
1228 stream
->avail_out
= bufsiz
;
1229 } while (status
!= Z_STREAM_END
);
1233 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1235 int bytes
= strlen(buffer
) + 1;
1236 unsigned char *buf
= xmallocz(size
);
1240 n
= stream
->total_out
- bytes
;
1243 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1245 if (bytes
<= size
) {
1247 * The above condition must be (bytes <= size), not
1248 * (bytes < size). In other words, even though we
1249 * expect no more output and set avail_out to zero,
1250 * the input zlib stream may have bytes that express
1251 * "this concludes the stream", and we *do* want to
1254 * Otherwise we would not be able to test that we
1255 * consumed all the input to reach the expected size;
1256 * we also want to check that zlib tells us that all
1257 * went well with status == Z_STREAM_END at the end.
1259 stream
->next_out
= buf
+ bytes
;
1260 stream
->avail_out
= size
- bytes
;
1261 while (status
== Z_OK
)
1262 status
= git_inflate(stream
, Z_FINISH
);
1264 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1265 git_inflate_end(stream
);
1270 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1271 else if (stream
->avail_in
)
1272 error("garbage at end of loose object '%s'",
1279 * We used to just use "sscanf()", but that's actually way
1280 * too permissive for what we want to check. So do an anal
1281 * object header parse by hand.
1283 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1286 const char *type_buf
= hdr
;
1288 int type
, type_len
= 0;
1291 * The type can be of any size but is followed by
1303 type
= type_from_string_gently(type_buf
, type_len
, 1);
1305 strbuf_add(oi
->typename
, type_buf
, type_len
);
1307 * Set type to 0 if its an unknown object and
1308 * we're obtaining the type using '--allow-unknown-type'
1311 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
) && (type
< 0))
1314 die("invalid object type");
1319 * The length must follow immediately, and be in canonical
1320 * decimal format (ie "010" is not valid).
1322 size
= *hdr
++ - '0';
1327 unsigned long c
= *hdr
- '0';
1331 size
= size
* 10 + c
;
1339 * The length must be followed by a zero byte
1341 return *hdr
? -1 : type
;
1344 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1346 struct object_info oi
= OBJECT_INFO_INIT
;
1349 return parse_sha1_header_extended(hdr
, &oi
, 0);
1352 unsigned long get_size_from_delta(struct packed_git
*p
,
1353 struct pack_window
**w_curs
,
1356 const unsigned char *data
;
1357 unsigned char delta_head
[20], *in
;
1361 memset(&stream
, 0, sizeof(stream
));
1362 stream
.next_out
= delta_head
;
1363 stream
.avail_out
= sizeof(delta_head
);
1365 git_inflate_init(&stream
);
1367 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1368 stream
.next_in
= in
;
1369 st
= git_inflate(&stream
, Z_FINISH
);
1370 curpos
+= stream
.next_in
- in
;
1371 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1372 stream
.total_out
< sizeof(delta_head
));
1373 git_inflate_end(&stream
);
1374 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1375 error("delta data unpack-initial failed");
1379 /* Examine the initial part of the delta to figure out
1384 /* ignore base size */
1385 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1387 /* Read the result size */
1388 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1391 static off_t
get_delta_base(struct packed_git
*p
,
1392 struct pack_window
**w_curs
,
1394 enum object_type type
,
1395 off_t delta_obj_offset
)
1397 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1400 /* use_pack() assured us we have [base_info, base_info + 20)
1401 * as a range that we can look at without walking off the
1402 * end of the mapped window. Its actually the hash size
1403 * that is assured. An OFS_DELTA longer than the hash size
1404 * is stupid, as then a REF_DELTA would be smaller to store.
1406 if (type
== OBJ_OFS_DELTA
) {
1408 unsigned char c
= base_info
[used
++];
1409 base_offset
= c
& 127;
1412 if (!base_offset
|| MSB(base_offset
, 7))
1413 return 0; /* overflow */
1414 c
= base_info
[used
++];
1415 base_offset
= (base_offset
<< 7) + (c
& 127);
1417 base_offset
= delta_obj_offset
- base_offset
;
1418 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1419 return 0; /* out of bound */
1421 } else if (type
== OBJ_REF_DELTA
) {
1422 /* The base entry _must_ be in the same pack */
1423 base_offset
= find_pack_entry_one(base_info
, p
);
1426 die("I am totally screwed");
1431 * Like get_delta_base above, but we return the sha1 instead of the pack
1432 * offset. This means it is cheaper for REF deltas (we do not have to do
1433 * the final object lookup), but more expensive for OFS deltas (we
1434 * have to load the revidx to convert the offset back into a sha1).
1436 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1437 struct pack_window
**w_curs
,
1439 enum object_type type
,
1440 off_t delta_obj_offset
)
1442 if (type
== OBJ_REF_DELTA
) {
1443 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1445 } else if (type
== OBJ_OFS_DELTA
) {
1446 struct revindex_entry
*revidx
;
1447 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1448 type
, delta_obj_offset
);
1453 revidx
= find_pack_revindex(p
, base_offset
);
1457 return nth_packed_object_sha1(p
, revidx
->nr
);
1462 int unpack_object_header(struct packed_git
*p
,
1463 struct pack_window
**w_curs
,
1465 unsigned long *sizep
)
1467 unsigned char *base
;
1470 enum object_type type
;
1472 /* use_pack() assures us we have [base, base + 20) available
1473 * as a range that we can look at. (Its actually the hash
1474 * size that is assured.) With our object header encoding
1475 * the maximum deflated object size is 2^137, which is just
1476 * insane, so we know won't exceed what we have been given.
1478 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1479 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1488 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1491 struct revindex_entry
*revidx
;
1492 const unsigned char *sha1
;
1493 revidx
= find_pack_revindex(p
, obj_offset
);
1496 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1497 mark_bad_packed_object(p
, sha1
);
1498 type
= sha1_object_info(sha1
, NULL
);
1499 if (type
<= OBJ_NONE
)
1504 #define POI_STACK_PREALLOC 64
1506 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1508 enum object_type type
,
1509 struct pack_window
**w_curs
,
1512 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1513 off_t
*poi_stack
= small_poi_stack
;
1514 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1516 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1519 /* Push the object we're going to leave behind */
1520 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1521 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1522 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
1523 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1525 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1527 poi_stack
[poi_stack_nr
++] = obj_offset
;
1528 /* If parsing the base offset fails, just unwind */
1529 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1532 curpos
= obj_offset
= base_offset
;
1533 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1534 if (type
<= OBJ_NONE
) {
1535 /* If getting the base itself fails, we first
1536 * retry the base, otherwise unwind */
1537 type
= retry_bad_packed_offset(p
, base_offset
);
1538 if (type
> OBJ_NONE
)
1552 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1553 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1558 if (poi_stack
!= small_poi_stack
)
1563 while (poi_stack_nr
) {
1564 obj_offset
= poi_stack
[--poi_stack_nr
];
1565 type
= retry_bad_packed_offset(p
, obj_offset
);
1566 if (type
> OBJ_NONE
)
1573 static struct hashmap delta_base_cache
;
1574 static size_t delta_base_cached
;
1576 static LIST_HEAD(delta_base_cache_lru
);
1578 struct delta_base_cache_key
{
1579 struct packed_git
*p
;
1583 struct delta_base_cache_entry
{
1584 struct hashmap hash
;
1585 struct delta_base_cache_key key
;
1586 struct list_head lru
;
1589 enum object_type type
;
1592 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1596 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
1597 hash
+= (hash
>> 8) + (hash
>> 16);
1601 static struct delta_base_cache_entry
*
1602 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1604 struct hashmap_entry entry
;
1605 struct delta_base_cache_key key
;
1607 if (!delta_base_cache
.cmpfn
)
1610 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
1612 key
.base_offset
= base_offset
;
1613 return hashmap_get(&delta_base_cache
, &entry
, &key
);
1616 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
1617 const struct delta_base_cache_key
*b
)
1619 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
1622 static int delta_base_cache_hash_cmp(const void *unused_cmp_data
,
1623 const void *va
, const void *vb
,
1626 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
1627 const struct delta_base_cache_key
*key
= vkey
;
1629 return !delta_base_cache_key_eq(&a
->key
, key
);
1631 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
1634 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1636 return !!get_delta_base_cache_entry(p
, base_offset
);
1640 * Remove the entry from the cache, but do _not_ free the associated
1641 * entry data. The caller takes ownership of the "data" buffer, and
1642 * should copy out any fields it wants before detaching.
1644 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1646 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
1647 list_del(&ent
->lru
);
1648 delta_base_cached
-= ent
->size
;
1652 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1653 unsigned long *base_size
, enum object_type
*type
)
1655 struct delta_base_cache_entry
*ent
;
1657 ent
= get_delta_base_cache_entry(p
, base_offset
);
1659 return unpack_entry(p
, base_offset
, type
, base_size
);
1664 *base_size
= ent
->size
;
1665 return xmemdupz(ent
->data
, ent
->size
);
1668 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1671 detach_delta_base_cache_entry(ent
);
1674 void clear_delta_base_cache(void)
1676 struct list_head
*lru
, *tmp
;
1677 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1678 struct delta_base_cache_entry
*entry
=
1679 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1680 release_delta_base_cache(entry
);
1684 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1685 void *base
, unsigned long base_size
, enum object_type type
)
1687 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
1688 struct list_head
*lru
, *tmp
;
1690 delta_base_cached
+= base_size
;
1692 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1693 struct delta_base_cache_entry
*f
=
1694 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1695 if (delta_base_cached
<= delta_base_cache_limit
)
1697 release_delta_base_cache(f
);
1701 ent
->key
.base_offset
= base_offset
;
1704 ent
->size
= base_size
;
1705 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
1707 if (!delta_base_cache
.cmpfn
)
1708 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, NULL
, 0);
1709 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
1710 hashmap_add(&delta_base_cache
, ent
);
1713 int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1714 struct object_info
*oi
)
1716 struct pack_window
*w_curs
= NULL
;
1718 off_t curpos
= obj_offset
;
1719 enum object_type type
;
1722 * We always get the representation type, but only convert it to
1723 * a "real" type later if the caller is interested.
1726 *oi
->contentp
= cache_or_unpack_entry(p
, obj_offset
, oi
->sizep
,
1731 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1734 if (!oi
->contentp
&& oi
->sizep
) {
1735 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1736 off_t tmp_pos
= curpos
;
1737 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1743 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1744 if (*oi
->sizep
== 0) {
1753 if (oi
->disk_sizep
) {
1754 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1755 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1758 if (oi
->typep
|| oi
->typename
) {
1759 enum object_type ptot
;
1760 ptot
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
,
1765 const char *tn
= typename(ptot
);
1767 strbuf_addstr(oi
->typename
, tn
);
1775 if (oi
->delta_base_sha1
) {
1776 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1777 const unsigned char *base
;
1779 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1786 hashcpy(oi
->delta_base_sha1
, base
);
1788 hashclr(oi
->delta_base_sha1
);
1791 oi
->whence
= in_delta_base_cache(p
, obj_offset
) ? OI_DBCACHED
:
1795 unuse_pack(&w_curs
);
1799 static void *unpack_compressed_entry(struct packed_git
*p
,
1800 struct pack_window
**w_curs
,
1806 unsigned char *buffer
, *in
;
1808 buffer
= xmallocz_gently(size
);
1811 memset(&stream
, 0, sizeof(stream
));
1812 stream
.next_out
= buffer
;
1813 stream
.avail_out
= size
+ 1;
1815 git_inflate_init(&stream
);
1817 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1818 stream
.next_in
= in
;
1819 st
= git_inflate(&stream
, Z_FINISH
);
1820 if (!stream
.avail_out
)
1821 break; /* the payload is larger than it should be */
1822 curpos
+= stream
.next_in
- in
;
1823 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1824 git_inflate_end(&stream
);
1825 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1833 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1834 unsigned long *size
);
1836 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1838 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
1839 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
1840 p
->pack_name
, (uintmax_t)obj_offset
);
1843 int do_check_packed_object_crc
;
1845 #define UNPACK_ENTRY_STACK_PREALLOC 64
1846 struct unpack_entry_stack_ent
{
1852 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1853 enum object_type
*final_type
, unsigned long *final_size
)
1855 struct pack_window
*w_curs
= NULL
;
1856 off_t curpos
= obj_offset
;
1859 enum object_type type
;
1860 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
1861 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
1862 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
1863 int base_from_cache
= 0;
1865 write_pack_access_log(p
, obj_offset
);
1867 /* PHASE 1: drill down to the innermost base object */
1871 struct delta_base_cache_entry
*ent
;
1873 ent
= get_delta_base_cache_entry(p
, curpos
);
1878 detach_delta_base_cache_entry(ent
);
1879 base_from_cache
= 1;
1883 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1884 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1885 off_t len
= revidx
[1].offset
- obj_offset
;
1886 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1887 const unsigned char *sha1
=
1888 nth_packed_object_sha1(p
, revidx
->nr
);
1889 error("bad packed object CRC for %s",
1891 mark_bad_packed_object(p
, sha1
);
1897 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1898 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
1901 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1903 error("failed to validate delta base reference "
1904 "at offset %"PRIuMAX
" from %s",
1905 (uintmax_t)curpos
, p
->pack_name
);
1906 /* bail to phase 2, in hopes of recovery */
1911 /* push object, proceed to base */
1912 if (delta_stack_nr
>= delta_stack_alloc
1913 && delta_stack
== small_delta_stack
) {
1914 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
1915 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
1916 memcpy(delta_stack
, small_delta_stack
,
1917 sizeof(*delta_stack
)*delta_stack_nr
);
1919 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
1921 i
= delta_stack_nr
++;
1922 delta_stack
[i
].obj_offset
= obj_offset
;
1923 delta_stack
[i
].curpos
= curpos
;
1924 delta_stack
[i
].size
= size
;
1926 curpos
= obj_offset
= base_offset
;
1929 /* PHASE 2: handle the base */
1934 die("BUG: unpack_entry: left loop at a valid delta");
1940 if (!base_from_cache
)
1941 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
1945 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1946 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1949 /* PHASE 3: apply deltas in order */
1952 * 'data' holds the base data, or NULL if there was corruption
1954 while (delta_stack_nr
) {
1957 void *external_base
= NULL
;
1958 unsigned long delta_size
, base_size
= size
;
1964 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
1968 * We're probably in deep shit, but let's try to fetch
1969 * the required base anyway from another pack or loose.
1970 * This is costly but should happen only in the presence
1971 * of a corrupted pack, and is better than failing outright.
1973 struct revindex_entry
*revidx
;
1974 const unsigned char *base_sha1
;
1975 revidx
= find_pack_revindex(p
, obj_offset
);
1977 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1978 error("failed to read delta base object %s"
1979 " at offset %"PRIuMAX
" from %s",
1980 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
1982 mark_bad_packed_object(p
, base_sha1
);
1983 base
= read_object(base_sha1
, &type
, &base_size
);
1984 external_base
= base
;
1988 i
= --delta_stack_nr
;
1989 obj_offset
= delta_stack
[i
].obj_offset
;
1990 curpos
= delta_stack
[i
].curpos
;
1991 delta_size
= delta_stack
[i
].size
;
1996 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
1999 error("failed to unpack compressed delta "
2000 "at offset %"PRIuMAX
" from %s",
2001 (uintmax_t)curpos
, p
->pack_name
);
2003 free(external_base
);
2007 data
= patch_delta(base
, base_size
,
2008 delta_data
, delta_size
,
2012 * We could not apply the delta; warn the user, but keep going.
2013 * Our failure will be noticed either in the next iteration of
2014 * the loop, or if this is the final delta, in the caller when
2015 * we return NULL. Those code paths will take care of making
2016 * a more explicit warning and retrying with another copy of
2020 error("failed to apply delta");
2023 free(external_base
);
2032 unuse_pack(&w_curs
);
2034 if (delta_stack
!= small_delta_stack
)
2040 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2043 const unsigned char *index
= p
->index_data
;
2045 if (open_pack_index(p
))
2047 index
= p
->index_data
;
2049 if (n
>= p
->num_objects
)
2052 if (p
->index_version
== 1) {
2053 return index
+ 24 * n
+ 4;
2056 return index
+ 20 * n
;
2060 const struct object_id
*nth_packed_object_oid(struct object_id
*oid
,
2061 struct packed_git
*p
,
2064 const unsigned char *hash
= nth_packed_object_sha1(p
, n
);
2067 hashcpy(oid
->hash
, hash
);
2071 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
2073 const unsigned char *ptr
= vptr
;
2074 const unsigned char *start
= p
->index_data
;
2075 const unsigned char *end
= start
+ p
->index_size
;
2077 die(_("offset before start of pack index for %s (corrupt index?)"),
2079 /* No need to check for underflow; .idx files must be at least 8 bytes */
2081 die(_("offset beyond end of pack index for %s (truncated index?)"),
2085 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2087 const unsigned char *index
= p
->index_data
;
2089 if (p
->index_version
== 1) {
2090 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2093 index
+= 8 + p
->num_objects
* (20 + 4);
2094 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2095 if (!(off
& 0x80000000))
2097 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2098 check_pack_index_ptr(p
, index
);
2099 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2100 ntohl(*((uint32_t *)(index
+ 4)));
2104 off_t
find_pack_entry_one(const unsigned char *sha1
,
2105 struct packed_git
*p
)
2107 const uint32_t *level1_ofs
= p
->index_data
;
2108 const unsigned char *index
= p
->index_data
;
2109 unsigned hi
, lo
, stride
;
2110 static int debug_lookup
= -1;
2112 if (debug_lookup
< 0)
2113 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2116 if (open_pack_index(p
))
2118 level1_ofs
= p
->index_data
;
2119 index
= p
->index_data
;
2121 if (p
->index_version
> 1) {
2126 hi
= ntohl(level1_ofs
[*sha1
]);
2127 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2128 if (p
->index_version
> 1) {
2136 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2137 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2140 unsigned mi
= (lo
+ hi
) / 2;
2141 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2144 printf("lo %u hi %u rg %u mi %u\n",
2145 lo
, hi
, hi
- lo
, mi
);
2147 return nth_packed_object_offset(p
, mi
);
2156 int is_pack_valid(struct packed_git
*p
)
2158 /* An already open pack is known to be valid. */
2159 if (p
->pack_fd
!= -1)
2162 /* If the pack has one window completely covering the
2163 * file size, the pack is known to be valid even if
2164 * the descriptor is not currently open.
2167 struct pack_window
*w
= p
->windows
;
2169 if (!w
->offset
&& w
->len
== p
->pack_size
)
2173 /* Force the pack to open to prove its valid. */
2174 return !open_packed_git(p
);
2177 static int fill_pack_entry(const unsigned char *sha1
,
2178 struct pack_entry
*e
,
2179 struct packed_git
*p
)
2183 if (p
->num_bad_objects
) {
2185 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2186 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2190 offset
= find_pack_entry_one(sha1
, p
);
2195 * We are about to tell the caller where they can locate the
2196 * requested object. We better make sure the packfile is
2197 * still here and can be accessed before supplying that
2198 * answer, as it may have been deleted since the index was
2201 if (!is_pack_valid(p
))
2205 hashcpy(e
->sha1
, sha1
);
2210 * Iff a pack file contains the object named by sha1, return true and
2211 * store its location to e.
2213 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2215 struct mru_entry
*p
;
2217 prepare_packed_git();
2221 for (p
= packed_git_mru
->head
; p
; p
= p
->next
) {
2222 if (fill_pack_entry(sha1
, e
, p
->item
)) {
2223 mru_mark(packed_git_mru
, p
);
2230 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2231 struct packed_git
*packs
)
2233 struct packed_git
*p
;
2235 for (p
= packs
; p
; p
= p
->next
) {
2236 if (find_pack_entry_one(sha1
, p
))
2243 static int sha1_loose_object_info(const unsigned char *sha1
,
2244 struct object_info
*oi
,
2248 unsigned long mapsize
;
2252 struct strbuf hdrbuf
= STRBUF_INIT
;
2253 unsigned long size_scratch
;
2255 if (oi
->delta_base_sha1
)
2256 hashclr(oi
->delta_base_sha1
);
2259 * If we don't care about type or size, then we don't
2260 * need to look inside the object at all. Note that we
2261 * do not optimize out the stat call, even if the
2262 * caller doesn't care about the disk-size, since our
2263 * return value implicitly indicates whether the
2264 * object even exists.
2266 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
&& !oi
->contentp
) {
2269 if (stat_sha1_file(sha1
, &st
, &path
) < 0)
2272 *oi
->disk_sizep
= st
.st_size
;
2276 map
= map_sha1_file(sha1
, &mapsize
);
2281 oi
->sizep
= &size_scratch
;
2284 *oi
->disk_sizep
= mapsize
;
2285 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
)) {
2286 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
2287 status
= error("unable to unpack %s header with --allow-unknown-type",
2289 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2290 status
= error("unable to unpack %s header",
2294 else if (hdrbuf
.len
) {
2295 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
2296 status
= error("unable to parse %s header with --allow-unknown-type",
2298 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
2299 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2301 if (status
>= 0 && oi
->contentp
)
2302 *oi
->contentp
= unpack_sha1_rest(&stream
, hdr
,
2305 git_inflate_end(&stream
);
2307 munmap(map
, mapsize
);
2308 if (status
&& oi
->typep
)
2309 *oi
->typep
= status
;
2310 if (oi
->sizep
== &size_scratch
)
2312 strbuf_release(&hdrbuf
);
2313 oi
->whence
= OI_LOOSE
;
2314 return (status
< 0) ? status
: 0;
2317 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2319 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
2320 struct pack_entry e
;
2322 const unsigned char *real
= (flags
& OBJECT_INFO_LOOKUP_REPLACE
) ?
2323 lookup_replace_object(sha1
) :
2329 if (!(flags
& OBJECT_INFO_SKIP_CACHED
)) {
2330 struct cached_object
*co
= find_cached_object(real
);
2333 *(oi
->typep
) = co
->type
;
2335 *(oi
->sizep
) = co
->size
;
2337 *(oi
->disk_sizep
) = 0;
2338 if (oi
->delta_base_sha1
)
2339 hashclr(oi
->delta_base_sha1
);
2341 strbuf_addstr(oi
->typename
, typename(co
->type
));
2343 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
2344 oi
->whence
= OI_CACHED
;
2349 if (!find_pack_entry(real
, &e
)) {
2350 /* Most likely it's a loose object. */
2351 if (!sha1_loose_object_info(real
, oi
, flags
))
2354 /* Not a loose object; someone else may have just packed it. */
2355 if (flags
& OBJECT_INFO_QUICK
) {
2358 reprepare_packed_git();
2359 if (!find_pack_entry(real
, &e
))
2364 if (oi
== &blank_oi
)
2366 * We know that the caller doesn't actually need the
2367 * information below, so return early.
2371 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2373 mark_bad_packed_object(e
.p
, real
);
2374 return sha1_object_info_extended(real
, oi
, 0);
2375 } else if (oi
->whence
== OI_PACKED
) {
2376 oi
->u
.packed
.offset
= e
.offset
;
2377 oi
->u
.packed
.pack
= e
.p
;
2378 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2379 rtype
== OBJ_OFS_DELTA
);
2385 /* returns enum object_type or negative */
2386 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2388 enum object_type type
;
2389 struct object_info oi
= OBJECT_INFO_INIT
;
2393 if (sha1_object_info_extended(sha1
, &oi
,
2394 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
2399 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2400 unsigned char *sha1
)
2402 struct cached_object
*co
;
2404 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2405 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2407 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2408 co
= &cached_objects
[cached_object_nr
++];
2411 co
->buf
= xmalloc(len
);
2412 memcpy(co
->buf
, buf
, len
);
2413 hashcpy(co
->sha1
, sha1
);
2417 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2418 unsigned long *size
)
2420 struct object_info oi
= OBJECT_INFO_INIT
;
2424 oi
.contentp
= &content
;
2426 if (sha1_object_info_extended(sha1
, &oi
, 0) < 0)
2432 * This function dies on corrupt objects; the callers who want to
2433 * deal with them should arrange to call read_object() and give error
2434 * messages themselves.
2436 void *read_sha1_file_extended(const unsigned char *sha1
,
2437 enum object_type
*type
,
2438 unsigned long *size
,
2442 const struct packed_git
*p
;
2445 const unsigned char *repl
= lookup_replace
? lookup_replace_object(sha1
)
2449 data
= read_object(repl
, type
, size
);
2453 if (errno
&& errno
!= ENOENT
)
2454 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2456 /* die if we replaced an object with one that does not exist */
2458 die("replacement %s not found for %s",
2459 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2461 if (!stat_sha1_file(repl
, &st
, &path
))
2462 die("loose object %s (stored in %s) is corrupt",
2463 sha1_to_hex(repl
), path
);
2465 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2466 die("packed object %s (stored in %s) is corrupt",
2467 sha1_to_hex(repl
), p
->pack_name
);
2472 void *read_object_with_reference(const unsigned char *sha1
,
2473 const char *required_type_name
,
2474 unsigned long *size
,
2475 unsigned char *actual_sha1_return
)
2477 enum object_type type
, required_type
;
2479 unsigned long isize
;
2480 unsigned char actual_sha1
[20];
2482 required_type
= type_from_string(required_type_name
);
2483 hashcpy(actual_sha1
, sha1
);
2485 int ref_length
= -1;
2486 const char *ref_type
= NULL
;
2488 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2491 if (type
== required_type
) {
2493 if (actual_sha1_return
)
2494 hashcpy(actual_sha1_return
, actual_sha1
);
2497 /* Handle references */
2498 else if (type
== OBJ_COMMIT
)
2500 else if (type
== OBJ_TAG
)
2501 ref_type
= "object ";
2506 ref_length
= strlen(ref_type
);
2508 if (ref_length
+ 40 > isize
||
2509 memcmp(buffer
, ref_type
, ref_length
) ||
2510 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2515 /* Now we have the ID of the referred-to object in
2516 * actual_sha1. Check again. */
2520 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2521 const char *type
, unsigned char *sha1
,
2522 char *hdr
, int *hdrlen
)
2526 /* Generate the header */
2527 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
2531 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2532 git_SHA1_Update(&c
, buf
, len
);
2533 git_SHA1_Final(sha1
, &c
);
2537 * Move the just written object into its final resting place.
2539 int finalize_object_file(const char *tmpfile
, const char *filename
)
2543 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2545 else if (link(tmpfile
, filename
))
2549 * Coda hack - coda doesn't like cross-directory links,
2550 * so we fall back to a rename, which will mean that it
2551 * won't be able to check collisions, but that's not a
2554 * The same holds for FAT formatted media.
2556 * When this succeeds, we just return. We have nothing
2559 if (ret
&& ret
!= EEXIST
) {
2561 if (!rename(tmpfile
, filename
))
2565 unlink_or_warn(tmpfile
);
2567 if (ret
!= EEXIST
) {
2568 return error_errno("unable to write sha1 filename %s", filename
);
2570 /* FIXME!!! Collision check here ? */
2574 if (adjust_shared_perm(filename
))
2575 return error("unable to set permission to '%s'", filename
);
2579 static int write_buffer(int fd
, const void *buf
, size_t len
)
2581 if (write_in_full(fd
, buf
, len
) < 0)
2582 return error_errno("file write error");
2586 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2587 unsigned char *sha1
)
2590 int hdrlen
= sizeof(hdr
);
2591 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2595 /* Finalize a file on disk, and close it. */
2596 static void close_sha1_file(int fd
)
2598 if (fsync_object_files
)
2599 fsync_or_die(fd
, "sha1 file");
2601 die_errno("error when closing sha1 file");
2604 /* Size of directory component, including the ending '/' */
2605 static inline int directory_size(const char *filename
)
2607 const char *s
= strrchr(filename
, '/');
2610 return s
- filename
+ 1;
2614 * This creates a temporary file in the same directory as the final
2617 * We want to avoid cross-directory filename renames, because those
2618 * can have problems on various filesystems (FAT, NFS, Coda).
2620 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
2622 int fd
, dirlen
= directory_size(filename
);
2625 strbuf_add(tmp
, filename
, dirlen
);
2626 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
2627 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
2628 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2630 * Make sure the directory exists; note that the contents
2631 * of the buffer are undefined after mkstemp returns an
2632 * error, so we have to rewrite the whole buffer from
2636 strbuf_add(tmp
, filename
, dirlen
- 1);
2637 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
2639 if (adjust_shared_perm(tmp
->buf
))
2643 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
2644 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
2649 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2650 const void *buf
, unsigned long len
, time_t mtime
)
2653 unsigned char compressed
[4096];
2656 unsigned char parano_sha1
[20];
2657 static struct strbuf tmp_file
= STRBUF_INIT
;
2658 const char *filename
= sha1_file_name(sha1
);
2660 fd
= create_tmpfile(&tmp_file
, filename
);
2662 if (errno
== EACCES
)
2663 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2665 return error_errno("unable to create temporary file");
2669 git_deflate_init(&stream
, zlib_compression_level
);
2670 stream
.next_out
= compressed
;
2671 stream
.avail_out
= sizeof(compressed
);
2674 /* First header.. */
2675 stream
.next_in
= (unsigned char *)hdr
;
2676 stream
.avail_in
= hdrlen
;
2677 while (git_deflate(&stream
, 0) == Z_OK
)
2679 git_SHA1_Update(&c
, hdr
, hdrlen
);
2681 /* Then the data itself.. */
2682 stream
.next_in
= (void *)buf
;
2683 stream
.avail_in
= len
;
2685 unsigned char *in0
= stream
.next_in
;
2686 ret
= git_deflate(&stream
, Z_FINISH
);
2687 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2688 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2689 die("unable to write sha1 file");
2690 stream
.next_out
= compressed
;
2691 stream
.avail_out
= sizeof(compressed
);
2692 } while (ret
== Z_OK
);
2694 if (ret
!= Z_STREAM_END
)
2695 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2696 ret
= git_deflate_end_gently(&stream
);
2698 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2699 git_SHA1_Final(parano_sha1
, &c
);
2700 if (hashcmp(sha1
, parano_sha1
) != 0)
2701 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2703 close_sha1_file(fd
);
2708 utb
.modtime
= mtime
;
2709 if (utime(tmp_file
.buf
, &utb
) < 0)
2710 warning_errno("failed utime() on %s", tmp_file
.buf
);
2713 return finalize_object_file(tmp_file
.buf
, filename
);
2716 static int freshen_loose_object(const unsigned char *sha1
)
2718 return check_and_freshen(sha1
, 1);
2721 static int freshen_packed_object(const unsigned char *sha1
)
2723 struct pack_entry e
;
2724 if (!find_pack_entry(sha1
, &e
))
2728 if (!freshen_file(e
.p
->pack_name
))
2734 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
2737 int hdrlen
= sizeof(hdr
);
2739 /* Normally if we have it in the pack then we do not bother writing
2740 * it out into .git/objects/??/?{38} file.
2742 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2743 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
2745 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2748 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
2749 unsigned char *sha1
, unsigned flags
)
2752 int hdrlen
, status
= 0;
2754 /* type string, SP, %lu of the length plus NUL must fit this */
2755 hdrlen
= strlen(type
) + 32;
2756 header
= xmalloc(hdrlen
);
2757 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
2759 if (!(flags
& HASH_WRITE_OBJECT
))
2761 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
2763 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
2770 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2774 enum object_type type
;
2779 if (has_loose_object(sha1
))
2781 buf
= read_object(sha1
, &type
, &len
);
2783 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2784 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
2785 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2791 int has_pack_index(const unsigned char *sha1
)
2794 if (stat(sha1_pack_index_name(sha1
), &st
))
2799 int has_sha1_pack(const unsigned char *sha1
)
2801 struct pack_entry e
;
2802 return find_pack_entry(sha1
, &e
);
2805 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
2807 if (!startup_info
->have_repository
)
2809 return sha1_object_info_extended(sha1
, NULL
,
2810 flags
| OBJECT_INFO_SKIP_CACHED
) >= 0;
2813 int has_object_file(const struct object_id
*oid
)
2815 return has_sha1_file(oid
->hash
);
2818 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
2820 return has_sha1_file_with_flags(oid
->hash
, flags
);
2823 static void check_tree(const void *buf
, size_t size
)
2825 struct tree_desc desc
;
2826 struct name_entry entry
;
2828 init_tree_desc(&desc
, buf
, size
);
2829 while (tree_entry(&desc
, &entry
))
2831 * tree_entry() will die() on malformed entries */
2835 static void check_commit(const void *buf
, size_t size
)
2838 memset(&c
, 0, sizeof(c
));
2839 if (parse_commit_buffer(&c
, buf
, size
))
2840 die("corrupt commit");
2843 static void check_tag(const void *buf
, size_t size
)
2846 memset(&t
, 0, sizeof(t
));
2847 if (parse_tag_buffer(&t
, buf
, size
))
2851 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2852 enum object_type type
,
2853 const char *path
, unsigned flags
)
2855 int ret
, re_allocated
= 0;
2856 int write_object
= flags
& HASH_WRITE_OBJECT
;
2862 * Convert blobs to git internal format
2864 if ((type
== OBJ_BLOB
) && path
) {
2865 struct strbuf nbuf
= STRBUF_INIT
;
2866 if (convert_to_git(&the_index
, path
, buf
, size
, &nbuf
,
2867 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
2868 buf
= strbuf_detach(&nbuf
, &size
);
2872 if (flags
& HASH_FORMAT_CHECK
) {
2873 if (type
== OBJ_TREE
)
2874 check_tree(buf
, size
);
2875 if (type
== OBJ_COMMIT
)
2876 check_commit(buf
, size
);
2877 if (type
== OBJ_TAG
)
2878 check_tag(buf
, size
);
2882 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2884 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2890 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
2891 const char *path
, unsigned flags
)
2894 const int write_object
= flags
& HASH_WRITE_OBJECT
;
2895 struct strbuf sbuf
= STRBUF_INIT
;
2898 assert(would_convert_to_git_filter_fd(path
));
2900 convert_to_git_filter_fd(&the_index
, path
, fd
, &sbuf
,
2901 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
2904 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
2907 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
2909 strbuf_release(&sbuf
);
2913 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
2914 const char *path
, unsigned flags
)
2916 struct strbuf sbuf
= STRBUF_INIT
;
2919 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2920 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
2923 strbuf_release(&sbuf
);
2927 #define SMALL_FILE_SIZE (32*1024)
2929 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
2930 enum object_type type
, const char *path
,
2936 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
2937 } else if (size
<= SMALL_FILE_SIZE
) {
2938 char *buf
= xmalloc(size
);
2939 if (size
== read_in_full(fd
, buf
, size
))
2940 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
2942 ret
= error_errno("short read");
2945 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2946 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
2953 * This creates one packfile per large blob unless bulk-checkin
2954 * machinery is "plugged".
2956 * This also bypasses the usual "convert-to-git" dance, and that is on
2957 * purpose. We could write a streaming version of the converting
2958 * functions and insert that before feeding the data to fast-import
2959 * (or equivalent in-core API described above). However, that is
2960 * somewhat complicated, as we do not know the size of the filter
2961 * result, which we need to know beforehand when writing a git object.
2962 * Since the primary motivation for trying to stream from the working
2963 * tree file and to avoid mmaping it in core is to deal with large
2964 * binary blobs, they generally do not want to get any conversion, and
2965 * callers should avoid this code path when filters are requested.
2967 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
2968 enum object_type type
, const char *path
,
2971 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
2974 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
2975 enum object_type type
, const char *path
, unsigned flags
)
2980 * Call xsize_t() only when needed to avoid potentially unnecessary
2981 * die() for large files.
2983 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
2984 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
2985 else if (!S_ISREG(st
->st_mode
))
2986 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
2987 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
2988 (path
&& would_convert_to_git(&the_index
, path
)))
2989 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
2992 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
2998 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3001 struct strbuf sb
= STRBUF_INIT
;
3003 switch (st
->st_mode
& S_IFMT
) {
3005 fd
= open(path
, O_RDONLY
);
3007 return error_errno("open(\"%s\")", path
);
3008 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3009 return error("%s: failed to insert into database",
3013 if (strbuf_readlink(&sb
, path
, st
->st_size
))
3014 return error_errno("readlink(\"%s\")", path
);
3015 if (!(flags
& HASH_WRITE_OBJECT
))
3016 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3017 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3018 return error("%s: failed to insert into database",
3020 strbuf_release(&sb
);
3023 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3025 return error("%s: unsupported file type", path
);
3030 int read_pack_header(int fd
, struct pack_header
*header
)
3032 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3033 /* "eof before pack header was fully read" */
3034 return PH_ERROR_EOF
;
3036 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3037 /* "protocol error (pack signature mismatch detected)" */
3038 return PH_ERROR_PACK_SIGNATURE
;
3039 if (!pack_version_ok(header
->hdr_version
))
3040 /* "protocol error (pack version unsupported)" */
3041 return PH_ERROR_PROTOCOL
;
3045 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3047 enum object_type type
= sha1_object_info(sha1
, NULL
);
3049 die("%s is not a valid object", sha1_to_hex(sha1
));
3051 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3055 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
3056 struct strbuf
*path
,
3057 each_loose_object_fn obj_cb
,
3058 each_loose_cruft_fn cruft_cb
,
3059 each_loose_subdir_fn subdir_cb
,
3062 size_t origlen
, baselen
;
3067 if (subdir_nr
> 0xff)
3068 BUG("invalid loose object subdirectory: %x", subdir_nr
);
3070 origlen
= path
->len
;
3071 strbuf_complete(path
, '/');
3072 strbuf_addf(path
, "%02x", subdir_nr
);
3073 baselen
= path
->len
;
3075 dir
= opendir(path
->buf
);
3077 if (errno
!= ENOENT
)
3078 r
= error_errno("unable to open %s", path
->buf
);
3079 strbuf_setlen(path
, origlen
);
3083 while ((de
= readdir(dir
))) {
3084 if (is_dot_or_dotdot(de
->d_name
))
3087 strbuf_setlen(path
, baselen
);
3088 strbuf_addf(path
, "/%s", de
->d_name
);
3090 if (strlen(de
->d_name
) == GIT_SHA1_HEXSZ
- 2) {
3091 char hex
[GIT_MAX_HEXSZ
+1];
3092 struct object_id oid
;
3094 xsnprintf(hex
, sizeof(hex
), "%02x%s",
3095 subdir_nr
, de
->d_name
);
3096 if (!get_oid_hex(hex
, &oid
)) {
3098 r
= obj_cb(&oid
, path
->buf
, data
);
3107 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3114 strbuf_setlen(path
, baselen
);
3115 if (!r
&& subdir_cb
)
3116 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3118 strbuf_setlen(path
, origlen
);
3123 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3124 each_loose_object_fn obj_cb
,
3125 each_loose_cruft_fn cruft_cb
,
3126 each_loose_subdir_fn subdir_cb
,
3132 for (i
= 0; i
< 256; i
++) {
3133 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3142 int for_each_loose_file_in_objdir(const char *path
,
3143 each_loose_object_fn obj_cb
,
3144 each_loose_cruft_fn cruft_cb
,
3145 each_loose_subdir_fn subdir_cb
,
3148 struct strbuf buf
= STRBUF_INIT
;
3151 strbuf_addstr(&buf
, path
);
3152 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3154 strbuf_release(&buf
);
3159 struct loose_alt_odb_data
{
3160 each_loose_object_fn
*cb
;
3164 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3167 struct loose_alt_odb_data
*data
= vdata
;
3168 struct strbuf buf
= STRBUF_INIT
;
3171 strbuf_addstr(&buf
, alt
->path
);
3172 r
= for_each_loose_file_in_objdir_buf(&buf
,
3173 data
->cb
, NULL
, NULL
,
3175 strbuf_release(&buf
);
3179 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3181 struct loose_alt_odb_data alt
;
3184 r
= for_each_loose_file_in_objdir(get_object_directory(),
3185 cb
, NULL
, NULL
, data
);
3189 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3194 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3197 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3202 for (i
= 0; i
< p
->num_objects
; i
++) {
3203 struct object_id oid
;
3205 if (!nth_packed_object_oid(&oid
, p
, i
))
3206 return error("unable to get sha1 of object %u in %s",
3209 r
= cb(&oid
, p
, i
, data
);
3216 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3218 struct packed_git
*p
;
3220 int pack_errors
= 0;
3222 prepare_packed_git();
3223 for (p
= packed_git
; p
; p
= p
->next
) {
3224 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3226 if (open_pack_index(p
)) {
3230 r
= for_each_object_in_pack(p
, cb
, data
);
3234 return r
? r
: pack_errors
;
3237 static int check_stream_sha1(git_zstream
*stream
,
3241 const unsigned char *expected_sha1
)
3244 unsigned char real_sha1
[GIT_MAX_RAWSZ
];
3245 unsigned char buf
[4096];
3246 unsigned long total_read
;
3250 git_SHA1_Update(&c
, hdr
, stream
->total_out
);
3253 * We already read some bytes into hdr, but the ones up to the NUL
3254 * do not count against the object's content size.
3256 total_read
= stream
->total_out
- strlen(hdr
) - 1;
3259 * This size comparison must be "<=" to read the final zlib packets;
3260 * see the comment in unpack_sha1_rest for details.
3262 while (total_read
<= size
&&
3263 (status
== Z_OK
|| status
== Z_BUF_ERROR
)) {
3264 stream
->next_out
= buf
;
3265 stream
->avail_out
= sizeof(buf
);
3266 if (size
- total_read
< stream
->avail_out
)
3267 stream
->avail_out
= size
- total_read
;
3268 status
= git_inflate(stream
, Z_FINISH
);
3269 git_SHA1_Update(&c
, buf
, stream
->next_out
- buf
);
3270 total_read
+= stream
->next_out
- buf
;
3272 git_inflate_end(stream
);
3274 if (status
!= Z_STREAM_END
) {
3275 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1
));
3278 if (stream
->avail_in
) {
3279 error("garbage at end of loose object '%s'",
3280 sha1_to_hex(expected_sha1
));
3284 git_SHA1_Final(real_sha1
, &c
);
3285 if (hashcmp(expected_sha1
, real_sha1
)) {
3286 error("sha1 mismatch for %s (expected %s)", path
,
3287 sha1_to_hex(expected_sha1
));
3294 int read_loose_object(const char *path
,
3295 const unsigned char *expected_sha1
,
3296 enum object_type
*type
,
3297 unsigned long *size
,
3302 unsigned long mapsize
;
3308 map
= map_sha1_file_1(path
, NULL
, &mapsize
);
3310 error_errno("unable to mmap %s", path
);
3314 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0) {
3315 error("unable to unpack header of %s", path
);
3319 *type
= parse_sha1_header(hdr
, size
);
3321 error("unable to parse header of %s", path
);
3322 git_inflate_end(&stream
);
3326 if (*type
== OBJ_BLOB
) {
3327 if (check_stream_sha1(&stream
, hdr
, *size
, path
, expected_sha1
) < 0)
3330 *contents
= unpack_sha1_rest(&stream
, hdr
, *size
, expected_sha1
);
3332 error("unable to unpack contents of %s", path
);
3333 git_inflate_end(&stream
);
3336 if (check_sha1_signature(expected_sha1
, *contents
,
3337 *size
, typename(*type
))) {
3338 error("sha1 mismatch for %s (expected %s)", path
,
3339 sha1_to_hex(expected_sha1
));
3345 ret
= 0; /* everything checks out */
3349 munmap(map
, mapsize
);