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");
723 * With an in-core object data in "map", rehash it to make sure the
724 * object name actually matches "sha1" to detect object corruption.
725 * With "map" == NULL, try reading the object named with "sha1" using
726 * the streaming interface and rehash it to do the same.
728 int check_sha1_signature(const unsigned char *sha1
, void *map
,
729 unsigned long size
, const char *type
)
731 unsigned char real_sha1
[20];
732 enum object_type obj_type
;
733 struct git_istream
*st
;
739 hash_sha1_file(map
, size
, type
, real_sha1
);
740 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
743 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
747 /* Generate the header */
748 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
752 git_SHA1_Update(&c
, hdr
, hdrlen
);
755 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
763 git_SHA1_Update(&c
, buf
, readlen
);
765 git_SHA1_Final(real_sha1
, &c
);
767 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
770 int git_open_cloexec(const char *name
, int flags
)
773 static int o_cloexec
= O_CLOEXEC
;
775 fd
= open(name
, flags
| o_cloexec
);
776 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
777 /* Try again w/o O_CLOEXEC: the kernel might not support it */
778 o_cloexec
&= ~O_CLOEXEC
;
779 fd
= open(name
, flags
| o_cloexec
);
782 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
784 static int fd_cloexec
= FD_CLOEXEC
;
786 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
787 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
788 int flags
= fcntl(fd
, F_GETFD
);
789 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
798 * Find "sha1" as a loose object in the local repository or in an alternate.
799 * Returns 0 on success, negative on failure.
801 * The "path" out-parameter will give the path of the object we found (if any).
802 * Note that it may point to static storage and is only valid until another
803 * call to sha1_file_name(), etc.
805 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
,
808 struct alternate_object_database
*alt
;
810 *path
= sha1_file_name(sha1
);
811 if (!lstat(*path
, st
))
816 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
817 *path
= alt_sha1_path(alt
, sha1
);
818 if (!lstat(*path
, st
))
826 * Like stat_sha1_file(), but actually open the object and return the
827 * descriptor. See the caveats on the "path" parameter above.
829 static int open_sha1_file(const unsigned char *sha1
, const char **path
)
832 struct alternate_object_database
*alt
;
833 int most_interesting_errno
;
835 *path
= sha1_file_name(sha1
);
836 fd
= git_open(*path
);
839 most_interesting_errno
= errno
;
842 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
843 *path
= alt_sha1_path(alt
, sha1
);
844 fd
= git_open(*path
);
847 if (most_interesting_errno
== ENOENT
)
848 most_interesting_errno
= errno
;
850 errno
= most_interesting_errno
;
855 * Map the loose object at "path" if it is not NULL, or the path found by
856 * searching for a loose object named "sha1".
858 static void *map_sha1_file_1(const char *path
,
859 const unsigned char *sha1
,
868 fd
= open_sha1_file(sha1
, &path
);
873 if (!fstat(fd
, &st
)) {
874 *size
= xsize_t(st
.st_size
);
876 /* mmap() is forbidden on empty files */
877 error("object file %s is empty", path
);
880 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
887 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
889 return map_sha1_file_1(NULL
, sha1
, size
);
892 static int unpack_sha1_short_header(git_zstream
*stream
,
893 unsigned char *map
, unsigned long mapsize
,
894 void *buffer
, unsigned long bufsiz
)
896 /* Get the data stream */
897 memset(stream
, 0, sizeof(*stream
));
898 stream
->next_in
= map
;
899 stream
->avail_in
= mapsize
;
900 stream
->next_out
= buffer
;
901 stream
->avail_out
= bufsiz
;
903 git_inflate_init(stream
);
904 return git_inflate(stream
, 0);
907 int unpack_sha1_header(git_zstream
*stream
,
908 unsigned char *map
, unsigned long mapsize
,
909 void *buffer
, unsigned long bufsiz
)
911 int status
= unpack_sha1_short_header(stream
, map
, mapsize
,
917 /* Make sure we have the terminating NUL */
918 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
923 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
924 unsigned long mapsize
, void *buffer
,
925 unsigned long bufsiz
, struct strbuf
*header
)
929 status
= unpack_sha1_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
934 * Check if entire header is unpacked in the first iteration.
936 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
940 * buffer[0..bufsiz] was not large enough. Copy the partial
941 * result out to header, and then append the result of further
942 * reading the stream.
944 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
945 stream
->next_out
= buffer
;
946 stream
->avail_out
= bufsiz
;
949 status
= git_inflate(stream
, 0);
950 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
951 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
953 stream
->next_out
= buffer
;
954 stream
->avail_out
= bufsiz
;
955 } while (status
!= Z_STREAM_END
);
959 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
961 int bytes
= strlen(buffer
) + 1;
962 unsigned char *buf
= xmallocz(size
);
966 n
= stream
->total_out
- bytes
;
969 memcpy(buf
, (char *) buffer
+ bytes
, n
);
973 * The above condition must be (bytes <= size), not
974 * (bytes < size). In other words, even though we
975 * expect no more output and set avail_out to zero,
976 * the input zlib stream may have bytes that express
977 * "this concludes the stream", and we *do* want to
980 * Otherwise we would not be able to test that we
981 * consumed all the input to reach the expected size;
982 * we also want to check that zlib tells us that all
983 * went well with status == Z_STREAM_END at the end.
985 stream
->next_out
= buf
+ bytes
;
986 stream
->avail_out
= size
- bytes
;
987 while (status
== Z_OK
)
988 status
= git_inflate(stream
, Z_FINISH
);
990 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
991 git_inflate_end(stream
);
996 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
997 else if (stream
->avail_in
)
998 error("garbage at end of loose object '%s'",
1005 * We used to just use "sscanf()", but that's actually way
1006 * too permissive for what we want to check. So do an anal
1007 * object header parse by hand.
1009 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1012 const char *type_buf
= hdr
;
1014 int type
, type_len
= 0;
1017 * The type can be of any size but is followed by
1029 type
= type_from_string_gently(type_buf
, type_len
, 1);
1031 strbuf_add(oi
->typename
, type_buf
, type_len
);
1033 * Set type to 0 if its an unknown object and
1034 * we're obtaining the type using '--allow-unknown-type'
1037 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
) && (type
< 0))
1040 die("invalid object type");
1045 * The length must follow immediately, and be in canonical
1046 * decimal format (ie "010" is not valid).
1048 size
= *hdr
++ - '0';
1053 unsigned long c
= *hdr
- '0';
1057 size
= size
* 10 + c
;
1065 * The length must be followed by a zero byte
1067 return *hdr
? -1 : type
;
1070 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1072 struct object_info oi
= OBJECT_INFO_INIT
;
1075 return parse_sha1_header_extended(hdr
, &oi
, 0);
1078 off_t
find_pack_entry_one(const unsigned char *sha1
,
1079 struct packed_git
*p
)
1081 const uint32_t *level1_ofs
= p
->index_data
;
1082 const unsigned char *index
= p
->index_data
;
1083 unsigned hi
, lo
, stride
;
1084 static int debug_lookup
= -1;
1086 if (debug_lookup
< 0)
1087 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1090 if (open_pack_index(p
))
1092 level1_ofs
= p
->index_data
;
1093 index
= p
->index_data
;
1095 if (p
->index_version
> 1) {
1100 hi
= ntohl(level1_ofs
[*sha1
]);
1101 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1102 if (p
->index_version
> 1) {
1110 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1111 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1114 unsigned mi
= (lo
+ hi
) / 2;
1115 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1118 printf("lo %u hi %u rg %u mi %u\n",
1119 lo
, hi
, hi
- lo
, mi
);
1121 return nth_packed_object_offset(p
, mi
);
1130 int is_pack_valid(struct packed_git
*p
)
1132 /* An already open pack is known to be valid. */
1133 if (p
->pack_fd
!= -1)
1136 /* If the pack has one window completely covering the
1137 * file size, the pack is known to be valid even if
1138 * the descriptor is not currently open.
1141 struct pack_window
*w
= p
->windows
;
1143 if (!w
->offset
&& w
->len
== p
->pack_size
)
1147 /* Force the pack to open to prove its valid. */
1148 return !open_packed_git(p
);
1151 static int fill_pack_entry(const unsigned char *sha1
,
1152 struct pack_entry
*e
,
1153 struct packed_git
*p
)
1157 if (p
->num_bad_objects
) {
1159 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1160 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1164 offset
= find_pack_entry_one(sha1
, p
);
1169 * We are about to tell the caller where they can locate the
1170 * requested object. We better make sure the packfile is
1171 * still here and can be accessed before supplying that
1172 * answer, as it may have been deleted since the index was
1175 if (!is_pack_valid(p
))
1179 hashcpy(e
->sha1
, sha1
);
1184 * Iff a pack file contains the object named by sha1, return true and
1185 * store its location to e.
1187 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1189 struct mru_entry
*p
;
1191 prepare_packed_git();
1195 for (p
= packed_git_mru
->head
; p
; p
= p
->next
) {
1196 if (fill_pack_entry(sha1
, e
, p
->item
)) {
1197 mru_mark(packed_git_mru
, p
);
1204 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1205 struct packed_git
*packs
)
1207 struct packed_git
*p
;
1209 for (p
= packs
; p
; p
= p
->next
) {
1210 if (find_pack_entry_one(sha1
, p
))
1217 static int sha1_loose_object_info(const unsigned char *sha1
,
1218 struct object_info
*oi
,
1222 unsigned long mapsize
;
1226 struct strbuf hdrbuf
= STRBUF_INIT
;
1227 unsigned long size_scratch
;
1229 if (oi
->delta_base_sha1
)
1230 hashclr(oi
->delta_base_sha1
);
1233 * If we don't care about type or size, then we don't
1234 * need to look inside the object at all. Note that we
1235 * do not optimize out the stat call, even if the
1236 * caller doesn't care about the disk-size, since our
1237 * return value implicitly indicates whether the
1238 * object even exists.
1240 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
&& !oi
->contentp
) {
1243 if (stat_sha1_file(sha1
, &st
, &path
) < 0)
1246 *oi
->disk_sizep
= st
.st_size
;
1250 map
= map_sha1_file(sha1
, &mapsize
);
1255 oi
->sizep
= &size_scratch
;
1258 *oi
->disk_sizep
= mapsize
;
1259 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
)) {
1260 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
1261 status
= error("unable to unpack %s header with --allow-unknown-type",
1263 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1264 status
= error("unable to unpack %s header",
1268 else if (hdrbuf
.len
) {
1269 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
1270 status
= error("unable to parse %s header with --allow-unknown-type",
1272 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
1273 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1275 if (status
>= 0 && oi
->contentp
)
1276 *oi
->contentp
= unpack_sha1_rest(&stream
, hdr
,
1279 git_inflate_end(&stream
);
1281 munmap(map
, mapsize
);
1282 if (status
&& oi
->typep
)
1283 *oi
->typep
= status
;
1284 if (oi
->sizep
== &size_scratch
)
1286 strbuf_release(&hdrbuf
);
1287 oi
->whence
= OI_LOOSE
;
1288 return (status
< 0) ? status
: 0;
1291 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
1293 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
1294 struct pack_entry e
;
1296 const unsigned char *real
= (flags
& OBJECT_INFO_LOOKUP_REPLACE
) ?
1297 lookup_replace_object(sha1
) :
1303 if (!(flags
& OBJECT_INFO_SKIP_CACHED
)) {
1304 struct cached_object
*co
= find_cached_object(real
);
1307 *(oi
->typep
) = co
->type
;
1309 *(oi
->sizep
) = co
->size
;
1311 *(oi
->disk_sizep
) = 0;
1312 if (oi
->delta_base_sha1
)
1313 hashclr(oi
->delta_base_sha1
);
1315 strbuf_addstr(oi
->typename
, typename(co
->type
));
1317 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
1318 oi
->whence
= OI_CACHED
;
1323 if (!find_pack_entry(real
, &e
)) {
1324 /* Most likely it's a loose object. */
1325 if (!sha1_loose_object_info(real
, oi
, flags
))
1328 /* Not a loose object; someone else may have just packed it. */
1329 if (flags
& OBJECT_INFO_QUICK
) {
1332 reprepare_packed_git();
1333 if (!find_pack_entry(real
, &e
))
1338 if (oi
== &blank_oi
)
1340 * We know that the caller doesn't actually need the
1341 * information below, so return early.
1345 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
1347 mark_bad_packed_object(e
.p
, real
);
1348 return sha1_object_info_extended(real
, oi
, 0);
1349 } else if (oi
->whence
== OI_PACKED
) {
1350 oi
->u
.packed
.offset
= e
.offset
;
1351 oi
->u
.packed
.pack
= e
.p
;
1352 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
1353 rtype
== OBJ_OFS_DELTA
);
1359 /* returns enum object_type or negative */
1360 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1362 enum object_type type
;
1363 struct object_info oi
= OBJECT_INFO_INIT
;
1367 if (sha1_object_info_extended(sha1
, &oi
,
1368 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
1373 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1374 unsigned long *size
)
1376 struct object_info oi
= OBJECT_INFO_INIT
;
1380 oi
.contentp
= &content
;
1382 if (sha1_object_info_extended(sha1
, &oi
, 0) < 0)
1387 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1388 unsigned char *sha1
)
1390 struct cached_object
*co
;
1392 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1393 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1395 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
1396 co
= &cached_objects
[cached_object_nr
++];
1399 co
->buf
= xmalloc(len
);
1400 memcpy(co
->buf
, buf
, len
);
1401 hashcpy(co
->sha1
, sha1
);
1406 * This function dies on corrupt objects; the callers who want to
1407 * deal with them should arrange to call read_object() and give error
1408 * messages themselves.
1410 void *read_sha1_file_extended(const unsigned char *sha1
,
1411 enum object_type
*type
,
1412 unsigned long *size
,
1416 const struct packed_git
*p
;
1419 const unsigned char *repl
= lookup_replace
? lookup_replace_object(sha1
)
1423 data
= read_object(repl
, type
, size
);
1427 if (errno
&& errno
!= ENOENT
)
1428 die_errno("failed to read object %s", sha1_to_hex(sha1
));
1430 /* die if we replaced an object with one that does not exist */
1432 die("replacement %s not found for %s",
1433 sha1_to_hex(repl
), sha1_to_hex(sha1
));
1435 if (!stat_sha1_file(repl
, &st
, &path
))
1436 die("loose object %s (stored in %s) is corrupt",
1437 sha1_to_hex(repl
), path
);
1439 if ((p
= has_packed_and_bad(repl
)) != NULL
)
1440 die("packed object %s (stored in %s) is corrupt",
1441 sha1_to_hex(repl
), p
->pack_name
);
1446 void *read_object_with_reference(const unsigned char *sha1
,
1447 const char *required_type_name
,
1448 unsigned long *size
,
1449 unsigned char *actual_sha1_return
)
1451 enum object_type type
, required_type
;
1453 unsigned long isize
;
1454 unsigned char actual_sha1
[20];
1456 required_type
= type_from_string(required_type_name
);
1457 hashcpy(actual_sha1
, sha1
);
1459 int ref_length
= -1;
1460 const char *ref_type
= NULL
;
1462 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1465 if (type
== required_type
) {
1467 if (actual_sha1_return
)
1468 hashcpy(actual_sha1_return
, actual_sha1
);
1471 /* Handle references */
1472 else if (type
== OBJ_COMMIT
)
1474 else if (type
== OBJ_TAG
)
1475 ref_type
= "object ";
1480 ref_length
= strlen(ref_type
);
1482 if (ref_length
+ 40 > isize
||
1483 memcmp(buffer
, ref_type
, ref_length
) ||
1484 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1489 /* Now we have the ID of the referred-to object in
1490 * actual_sha1. Check again. */
1494 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
1495 const char *type
, unsigned char *sha1
,
1496 char *hdr
, int *hdrlen
)
1500 /* Generate the header */
1501 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
1505 git_SHA1_Update(&c
, hdr
, *hdrlen
);
1506 git_SHA1_Update(&c
, buf
, len
);
1507 git_SHA1_Final(sha1
, &c
);
1511 * Move the just written object into its final resting place.
1513 int finalize_object_file(const char *tmpfile
, const char *filename
)
1517 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
1519 else if (link(tmpfile
, filename
))
1523 * Coda hack - coda doesn't like cross-directory links,
1524 * so we fall back to a rename, which will mean that it
1525 * won't be able to check collisions, but that's not a
1528 * The same holds for FAT formatted media.
1530 * When this succeeds, we just return. We have nothing
1533 if (ret
&& ret
!= EEXIST
) {
1535 if (!rename(tmpfile
, filename
))
1539 unlink_or_warn(tmpfile
);
1541 if (ret
!= EEXIST
) {
1542 return error_errno("unable to write sha1 filename %s", filename
);
1544 /* FIXME!!! Collision check here ? */
1548 if (adjust_shared_perm(filename
))
1549 return error("unable to set permission to '%s'", filename
);
1553 static int write_buffer(int fd
, const void *buf
, size_t len
)
1555 if (write_in_full(fd
, buf
, len
) < 0)
1556 return error_errno("file write error");
1560 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
1561 unsigned char *sha1
)
1564 int hdrlen
= sizeof(hdr
);
1565 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
1569 /* Finalize a file on disk, and close it. */
1570 static void close_sha1_file(int fd
)
1572 if (fsync_object_files
)
1573 fsync_or_die(fd
, "sha1 file");
1575 die_errno("error when closing sha1 file");
1578 /* Size of directory component, including the ending '/' */
1579 static inline int directory_size(const char *filename
)
1581 const char *s
= strrchr(filename
, '/');
1584 return s
- filename
+ 1;
1588 * This creates a temporary file in the same directory as the final
1591 * We want to avoid cross-directory filename renames, because those
1592 * can have problems on various filesystems (FAT, NFS, Coda).
1594 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
1596 int fd
, dirlen
= directory_size(filename
);
1599 strbuf_add(tmp
, filename
, dirlen
);
1600 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
1601 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1602 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
1604 * Make sure the directory exists; note that the contents
1605 * of the buffer are undefined after mkstemp returns an
1606 * error, so we have to rewrite the whole buffer from
1610 strbuf_add(tmp
, filename
, dirlen
- 1);
1611 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
1613 if (adjust_shared_perm(tmp
->buf
))
1617 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
1618 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1623 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
1624 const void *buf
, unsigned long len
, time_t mtime
)
1627 unsigned char compressed
[4096];
1630 unsigned char parano_sha1
[20];
1631 static struct strbuf tmp_file
= STRBUF_INIT
;
1632 const char *filename
= sha1_file_name(sha1
);
1634 fd
= create_tmpfile(&tmp_file
, filename
);
1636 if (errno
== EACCES
)
1637 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1639 return error_errno("unable to create temporary file");
1643 git_deflate_init(&stream
, zlib_compression_level
);
1644 stream
.next_out
= compressed
;
1645 stream
.avail_out
= sizeof(compressed
);
1648 /* First header.. */
1649 stream
.next_in
= (unsigned char *)hdr
;
1650 stream
.avail_in
= hdrlen
;
1651 while (git_deflate(&stream
, 0) == Z_OK
)
1653 git_SHA1_Update(&c
, hdr
, hdrlen
);
1655 /* Then the data itself.. */
1656 stream
.next_in
= (void *)buf
;
1657 stream
.avail_in
= len
;
1659 unsigned char *in0
= stream
.next_in
;
1660 ret
= git_deflate(&stream
, Z_FINISH
);
1661 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
1662 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
1663 die("unable to write sha1 file");
1664 stream
.next_out
= compressed
;
1665 stream
.avail_out
= sizeof(compressed
);
1666 } while (ret
== Z_OK
);
1668 if (ret
!= Z_STREAM_END
)
1669 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
1670 ret
= git_deflate_end_gently(&stream
);
1672 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
1673 git_SHA1_Final(parano_sha1
, &c
);
1674 if (hashcmp(sha1
, parano_sha1
) != 0)
1675 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
1677 close_sha1_file(fd
);
1682 utb
.modtime
= mtime
;
1683 if (utime(tmp_file
.buf
, &utb
) < 0)
1684 warning_errno("failed utime() on %s", tmp_file
.buf
);
1687 return finalize_object_file(tmp_file
.buf
, filename
);
1690 static int freshen_loose_object(const unsigned char *sha1
)
1692 return check_and_freshen(sha1
, 1);
1695 static int freshen_packed_object(const unsigned char *sha1
)
1697 struct pack_entry e
;
1698 if (!find_pack_entry(sha1
, &e
))
1702 if (!freshen_file(e
.p
->pack_name
))
1708 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
1711 int hdrlen
= sizeof(hdr
);
1713 /* Normally if we have it in the pack then we do not bother writing
1714 * it out into .git/objects/??/?{38} file.
1716 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
1717 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
1719 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
1722 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
1723 unsigned char *sha1
, unsigned flags
)
1726 int hdrlen
, status
= 0;
1728 /* type string, SP, %lu of the length plus NUL must fit this */
1729 hdrlen
= strlen(type
) + 32;
1730 header
= xmalloc(hdrlen
);
1731 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
1733 if (!(flags
& HASH_WRITE_OBJECT
))
1735 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
1737 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
1744 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
1748 enum object_type type
;
1753 if (has_loose_object(sha1
))
1755 buf
= read_object(sha1
, &type
, &len
);
1757 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
1758 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
1759 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
1765 int has_pack_index(const unsigned char *sha1
)
1768 if (stat(sha1_pack_index_name(sha1
), &st
))
1773 int has_sha1_pack(const unsigned char *sha1
)
1775 struct pack_entry e
;
1776 return find_pack_entry(sha1
, &e
);
1779 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
1781 if (!startup_info
->have_repository
)
1783 return sha1_object_info_extended(sha1
, NULL
,
1784 flags
| OBJECT_INFO_SKIP_CACHED
) >= 0;
1787 int has_object_file(const struct object_id
*oid
)
1789 return has_sha1_file(oid
->hash
);
1792 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
1794 return has_sha1_file_with_flags(oid
->hash
, flags
);
1797 static void check_tree(const void *buf
, size_t size
)
1799 struct tree_desc desc
;
1800 struct name_entry entry
;
1802 init_tree_desc(&desc
, buf
, size
);
1803 while (tree_entry(&desc
, &entry
))
1805 * tree_entry() will die() on malformed entries */
1809 static void check_commit(const void *buf
, size_t size
)
1812 memset(&c
, 0, sizeof(c
));
1813 if (parse_commit_buffer(&c
, buf
, size
))
1814 die("corrupt commit");
1817 static void check_tag(const void *buf
, size_t size
)
1820 memset(&t
, 0, sizeof(t
));
1821 if (parse_tag_buffer(&t
, buf
, size
))
1825 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
1826 enum object_type type
,
1827 const char *path
, unsigned flags
)
1829 int ret
, re_allocated
= 0;
1830 int write_object
= flags
& HASH_WRITE_OBJECT
;
1836 * Convert blobs to git internal format
1838 if ((type
== OBJ_BLOB
) && path
) {
1839 struct strbuf nbuf
= STRBUF_INIT
;
1840 if (convert_to_git(&the_index
, path
, buf
, size
, &nbuf
,
1841 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
1842 buf
= strbuf_detach(&nbuf
, &size
);
1846 if (flags
& HASH_FORMAT_CHECK
) {
1847 if (type
== OBJ_TREE
)
1848 check_tree(buf
, size
);
1849 if (type
== OBJ_COMMIT
)
1850 check_commit(buf
, size
);
1851 if (type
== OBJ_TAG
)
1852 check_tag(buf
, size
);
1856 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
1858 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
1864 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
1865 const char *path
, unsigned flags
)
1868 const int write_object
= flags
& HASH_WRITE_OBJECT
;
1869 struct strbuf sbuf
= STRBUF_INIT
;
1872 assert(would_convert_to_git_filter_fd(path
));
1874 convert_to_git_filter_fd(&the_index
, path
, fd
, &sbuf
,
1875 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
1878 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
1881 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
1883 strbuf_release(&sbuf
);
1887 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
1888 const char *path
, unsigned flags
)
1890 struct strbuf sbuf
= STRBUF_INIT
;
1893 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
1894 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
1897 strbuf_release(&sbuf
);
1901 #define SMALL_FILE_SIZE (32*1024)
1903 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
1904 enum object_type type
, const char *path
,
1910 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
1911 } else if (size
<= SMALL_FILE_SIZE
) {
1912 char *buf
= xmalloc(size
);
1913 if (size
== read_in_full(fd
, buf
, size
))
1914 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
1916 ret
= error_errno("short read");
1919 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1920 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
1927 * This creates one packfile per large blob unless bulk-checkin
1928 * machinery is "plugged".
1930 * This also bypasses the usual "convert-to-git" dance, and that is on
1931 * purpose. We could write a streaming version of the converting
1932 * functions and insert that before feeding the data to fast-import
1933 * (or equivalent in-core API described above). However, that is
1934 * somewhat complicated, as we do not know the size of the filter
1935 * result, which we need to know beforehand when writing a git object.
1936 * Since the primary motivation for trying to stream from the working
1937 * tree file and to avoid mmaping it in core is to deal with large
1938 * binary blobs, they generally do not want to get any conversion, and
1939 * callers should avoid this code path when filters are requested.
1941 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
1942 enum object_type type
, const char *path
,
1945 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
1948 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
1949 enum object_type type
, const char *path
, unsigned flags
)
1954 * Call xsize_t() only when needed to avoid potentially unnecessary
1955 * die() for large files.
1957 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
1958 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
1959 else if (!S_ISREG(st
->st_mode
))
1960 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
1961 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
1962 (path
&& would_convert_to_git(&the_index
, path
)))
1963 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
1966 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
1972 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
1975 struct strbuf sb
= STRBUF_INIT
;
1977 switch (st
->st_mode
& S_IFMT
) {
1979 fd
= open(path
, O_RDONLY
);
1981 return error_errno("open(\"%s\")", path
);
1982 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
1983 return error("%s: failed to insert into database",
1987 if (strbuf_readlink(&sb
, path
, st
->st_size
))
1988 return error_errno("readlink(\"%s\")", path
);
1989 if (!(flags
& HASH_WRITE_OBJECT
))
1990 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
1991 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
1992 return error("%s: failed to insert into database",
1994 strbuf_release(&sb
);
1997 return resolve_gitlink_ref(path
, "HEAD", sha1
);
1999 return error("%s: unsupported file type", path
);
2004 int read_pack_header(int fd
, struct pack_header
*header
)
2006 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2007 /* "eof before pack header was fully read" */
2008 return PH_ERROR_EOF
;
2010 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2011 /* "protocol error (pack signature mismatch detected)" */
2012 return PH_ERROR_PACK_SIGNATURE
;
2013 if (!pack_version_ok(header
->hdr_version
))
2014 /* "protocol error (pack version unsupported)" */
2015 return PH_ERROR_PROTOCOL
;
2019 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2021 enum object_type type
= sha1_object_info(sha1
, NULL
);
2023 die("%s is not a valid object", sha1_to_hex(sha1
));
2025 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
2029 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
2030 struct strbuf
*path
,
2031 each_loose_object_fn obj_cb
,
2032 each_loose_cruft_fn cruft_cb
,
2033 each_loose_subdir_fn subdir_cb
,
2036 size_t origlen
, baselen
;
2041 if (subdir_nr
> 0xff)
2042 BUG("invalid loose object subdirectory: %x", subdir_nr
);
2044 origlen
= path
->len
;
2045 strbuf_complete(path
, '/');
2046 strbuf_addf(path
, "%02x", subdir_nr
);
2047 baselen
= path
->len
;
2049 dir
= opendir(path
->buf
);
2051 if (errno
!= ENOENT
)
2052 r
= error_errno("unable to open %s", path
->buf
);
2053 strbuf_setlen(path
, origlen
);
2057 while ((de
= readdir(dir
))) {
2058 if (is_dot_or_dotdot(de
->d_name
))
2061 strbuf_setlen(path
, baselen
);
2062 strbuf_addf(path
, "/%s", de
->d_name
);
2064 if (strlen(de
->d_name
) == GIT_SHA1_HEXSZ
- 2) {
2065 char hex
[GIT_MAX_HEXSZ
+1];
2066 struct object_id oid
;
2068 xsnprintf(hex
, sizeof(hex
), "%02x%s",
2069 subdir_nr
, de
->d_name
);
2070 if (!get_oid_hex(hex
, &oid
)) {
2072 r
= obj_cb(&oid
, path
->buf
, data
);
2081 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
2088 strbuf_setlen(path
, baselen
);
2089 if (!r
&& subdir_cb
)
2090 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
2092 strbuf_setlen(path
, origlen
);
2097 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
2098 each_loose_object_fn obj_cb
,
2099 each_loose_cruft_fn cruft_cb
,
2100 each_loose_subdir_fn subdir_cb
,
2106 for (i
= 0; i
< 256; i
++) {
2107 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
2116 int for_each_loose_file_in_objdir(const char *path
,
2117 each_loose_object_fn obj_cb
,
2118 each_loose_cruft_fn cruft_cb
,
2119 each_loose_subdir_fn subdir_cb
,
2122 struct strbuf buf
= STRBUF_INIT
;
2125 strbuf_addstr(&buf
, path
);
2126 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
2128 strbuf_release(&buf
);
2133 struct loose_alt_odb_data
{
2134 each_loose_object_fn
*cb
;
2138 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
2141 struct loose_alt_odb_data
*data
= vdata
;
2142 struct strbuf buf
= STRBUF_INIT
;
2145 strbuf_addstr(&buf
, alt
->path
);
2146 r
= for_each_loose_file_in_objdir_buf(&buf
,
2147 data
->cb
, NULL
, NULL
,
2149 strbuf_release(&buf
);
2153 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
2155 struct loose_alt_odb_data alt
;
2158 r
= for_each_loose_file_in_objdir(get_object_directory(),
2159 cb
, NULL
, NULL
, data
);
2163 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
2168 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
2171 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
2176 for (i
= 0; i
< p
->num_objects
; i
++) {
2177 struct object_id oid
;
2179 if (!nth_packed_object_oid(&oid
, p
, i
))
2180 return error("unable to get sha1 of object %u in %s",
2183 r
= cb(&oid
, p
, i
, data
);
2190 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
2192 struct packed_git
*p
;
2194 int pack_errors
= 0;
2196 prepare_packed_git();
2197 for (p
= packed_git
; p
; p
= p
->next
) {
2198 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
2200 if (open_pack_index(p
)) {
2204 r
= for_each_object_in_pack(p
, cb
, data
);
2208 return r
? r
: pack_errors
;
2211 static int check_stream_sha1(git_zstream
*stream
,
2215 const unsigned char *expected_sha1
)
2218 unsigned char real_sha1
[GIT_MAX_RAWSZ
];
2219 unsigned char buf
[4096];
2220 unsigned long total_read
;
2224 git_SHA1_Update(&c
, hdr
, stream
->total_out
);
2227 * We already read some bytes into hdr, but the ones up to the NUL
2228 * do not count against the object's content size.
2230 total_read
= stream
->total_out
- strlen(hdr
) - 1;
2233 * This size comparison must be "<=" to read the final zlib packets;
2234 * see the comment in unpack_sha1_rest for details.
2236 while (total_read
<= size
&&
2237 (status
== Z_OK
|| status
== Z_BUF_ERROR
)) {
2238 stream
->next_out
= buf
;
2239 stream
->avail_out
= sizeof(buf
);
2240 if (size
- total_read
< stream
->avail_out
)
2241 stream
->avail_out
= size
- total_read
;
2242 status
= git_inflate(stream
, Z_FINISH
);
2243 git_SHA1_Update(&c
, buf
, stream
->next_out
- buf
);
2244 total_read
+= stream
->next_out
- buf
;
2246 git_inflate_end(stream
);
2248 if (status
!= Z_STREAM_END
) {
2249 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1
));
2252 if (stream
->avail_in
) {
2253 error("garbage at end of loose object '%s'",
2254 sha1_to_hex(expected_sha1
));
2258 git_SHA1_Final(real_sha1
, &c
);
2259 if (hashcmp(expected_sha1
, real_sha1
)) {
2260 error("sha1 mismatch for %s (expected %s)", path
,
2261 sha1_to_hex(expected_sha1
));
2268 int read_loose_object(const char *path
,
2269 const unsigned char *expected_sha1
,
2270 enum object_type
*type
,
2271 unsigned long *size
,
2276 unsigned long mapsize
;
2282 map
= map_sha1_file_1(path
, NULL
, &mapsize
);
2284 error_errno("unable to mmap %s", path
);
2288 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0) {
2289 error("unable to unpack header of %s", path
);
2293 *type
= parse_sha1_header(hdr
, size
);
2295 error("unable to parse header of %s", path
);
2296 git_inflate_end(&stream
);
2300 if (*type
== OBJ_BLOB
) {
2301 if (check_stream_sha1(&stream
, hdr
, *size
, path
, expected_sha1
) < 0)
2304 *contents
= unpack_sha1_rest(&stream
, hdr
, *size
, expected_sha1
);
2306 error("unable to unpack contents of %s", path
);
2307 git_inflate_end(&stream
);
2310 if (check_sha1_signature(expected_sha1
, *contents
,
2311 *size
, typename(*type
))) {
2312 error("sha1 mismatch for %s (expected %s)", path
,
2313 sha1_to_hex(expected_sha1
));
2319 ret
= 0; /* everything checks out */
2323 munmap(map
, mapsize
);