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"
28 #include "mergesort.h"
31 #include "fetch-object.h"
33 /* The maximum size for an object header. */
34 #define MAX_HEADER_LEN 32
36 const unsigned char null_sha1
[GIT_MAX_RAWSZ
];
37 const struct object_id null_oid
;
38 const struct object_id empty_tree_oid
= {
39 EMPTY_TREE_SHA1_BIN_LITERAL
41 const struct object_id empty_blob_oid
= {
42 EMPTY_BLOB_SHA1_BIN_LITERAL
45 static void git_hash_sha1_init(git_hash_ctx
*ctx
)
47 git_SHA1_Init(&ctx
->sha1
);
50 static void git_hash_sha1_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
52 git_SHA1_Update(&ctx
->sha1
, data
, len
);
55 static void git_hash_sha1_final(unsigned char *hash
, git_hash_ctx
*ctx
)
57 git_SHA1_Final(hash
, &ctx
->sha1
);
60 static void git_hash_unknown_init(git_hash_ctx
*ctx
)
62 die("trying to init unknown hash");
65 static void git_hash_unknown_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
67 die("trying to update unknown hash");
70 static void git_hash_unknown_final(unsigned char *hash
, git_hash_ctx
*ctx
)
72 die("trying to finalize unknown hash");
75 const struct git_hash_algo hash_algos
[GIT_HASH_NALGOS
] = {
81 git_hash_unknown_init
,
82 git_hash_unknown_update
,
83 git_hash_unknown_final
,
89 /* "sha1", big-endian */
102 * This is meant to hold a *small* number of objects that you would
103 * want read_sha1_file() to be able to return, but yet you do not want
104 * to write them into the object store (e.g. a browse-only
107 static struct cached_object
{
108 unsigned char sha1
[20];
109 enum object_type type
;
113 static int cached_object_nr
, cached_object_alloc
;
115 static struct cached_object empty_tree
= {
116 EMPTY_TREE_SHA1_BIN_LITERAL
,
122 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
125 struct cached_object
*co
= cached_objects
;
127 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
128 if (!hashcmp(co
->sha1
, sha1
))
131 if (!hashcmp(sha1
, empty_tree
.sha1
))
137 static int get_conv_flags(unsigned flags
)
139 if (flags
& HASH_RENORMALIZE
)
140 return CONV_EOL_RENORMALIZE
;
141 else if (flags
& HASH_WRITE_OBJECT
)
142 return global_conv_flags_eol
;
148 int mkdir_in_gitdir(const char *path
)
150 if (mkdir(path
, 0777)) {
151 int saved_errno
= errno
;
153 struct strbuf sb
= STRBUF_INIT
;
158 * Are we looking at a path in a symlinked worktree
159 * whose original repository does not yet have it?
160 * e.g. .git/rr-cache pointing at its original
161 * repository in which the user hasn't performed any
162 * conflict resolution yet?
164 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
165 strbuf_readlink(&sb
, path
, st
.st_size
) ||
166 !is_absolute_path(sb
.buf
) ||
167 mkdir(sb
.buf
, 0777)) {
174 return adjust_shared_perm(path
);
177 enum scld_error
safe_create_leading_directories(char *path
)
179 char *next_component
= path
+ offset_1st_component(path
);
180 enum scld_error ret
= SCLD_OK
;
182 while (ret
== SCLD_OK
&& next_component
) {
184 char *slash
= next_component
, slash_character
;
186 while (*slash
&& !is_dir_sep(*slash
))
192 next_component
= slash
+ 1;
193 while (is_dir_sep(*next_component
))
195 if (!*next_component
)
198 slash_character
= *slash
;
200 if (!stat(path
, &st
)) {
202 if (!S_ISDIR(st
.st_mode
)) {
206 } else if (mkdir(path
, 0777)) {
207 if (errno
== EEXIST
&&
208 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
209 ; /* somebody created it since we checked */
210 else if (errno
== ENOENT
)
212 * Either mkdir() failed because
213 * somebody just pruned the containing
214 * directory, or stat() failed because
215 * the file that was in our way was
216 * just removed. Either way, inform
217 * the caller that it might be worth
223 } else if (adjust_shared_perm(path
)) {
226 *slash
= slash_character
;
231 enum scld_error
safe_create_leading_directories_const(const char *path
)
234 /* path points to cache entries, so xstrdup before messing with it */
235 char *buf
= xstrdup(path
);
236 enum scld_error result
= safe_create_leading_directories(buf
);
244 int raceproof_create_file(const char *path
, create_file_fn fn
, void *cb
)
247 * The number of times we will try to remove empty directories
248 * in the way of path. This is only 1 because if another
249 * process is racily creating directories that conflict with
250 * us, we don't want to fight against them.
252 int remove_directories_remaining
= 1;
255 * The number of times that we will try to create the
256 * directories containing path. We are willing to attempt this
257 * more than once, because another process could be trying to
258 * clean up empty directories at the same time as we are
259 * trying to create them.
261 int create_directories_remaining
= 3;
263 /* A scratch copy of path, filled lazily if we need it: */
264 struct strbuf path_copy
= STRBUF_INIT
;
277 if (errno
== EISDIR
&& remove_directories_remaining
-- > 0) {
279 * A directory is in the way. Maybe it is empty; try
283 strbuf_addstr(&path_copy
, path
);
285 if (!remove_dir_recursively(&path_copy
, REMOVE_DIR_EMPTY_ONLY
))
287 } else if (errno
== ENOENT
&& create_directories_remaining
-- > 0) {
289 * Maybe the containing directory didn't exist, or
290 * maybe it was just deleted by a process that is
291 * racing with us to clean up empty directories. Try
294 enum scld_error scld_result
;
297 strbuf_addstr(&path_copy
, path
);
300 scld_result
= safe_create_leading_directories(path_copy
.buf
);
301 if (scld_result
== SCLD_OK
)
303 } while (scld_result
== SCLD_VANISHED
&& create_directories_remaining
-- > 0);
307 strbuf_release(&path_copy
);
312 static void fill_sha1_path(struct strbuf
*buf
, const unsigned char *sha1
)
315 for (i
= 0; i
< 20; i
++) {
316 static char hex
[] = "0123456789abcdef";
317 unsigned int val
= sha1
[i
];
318 strbuf_addch(buf
, hex
[val
>> 4]);
319 strbuf_addch(buf
, hex
[val
& 0xf]);
321 strbuf_addch(buf
, '/');
325 void sha1_file_name(struct strbuf
*buf
, const unsigned char *sha1
)
327 strbuf_addstr(buf
, get_object_directory());
328 strbuf_addch(buf
, '/');
329 fill_sha1_path(buf
, sha1
);
332 struct strbuf
*alt_scratch_buf(struct alternate_object_database
*alt
)
334 strbuf_setlen(&alt
->scratch
, alt
->base_len
);
335 return &alt
->scratch
;
338 static const char *alt_sha1_path(struct alternate_object_database
*alt
,
339 const unsigned char *sha1
)
341 struct strbuf
*buf
= alt_scratch_buf(alt
);
342 fill_sha1_path(buf
, sha1
);
346 struct alternate_object_database
*alt_odb_list
;
347 static struct alternate_object_database
**alt_odb_tail
;
350 * Return non-zero iff the path is usable as an alternate object database.
352 static int alt_odb_usable(struct strbuf
*path
, const char *normalized_objdir
)
354 struct alternate_object_database
*alt
;
356 /* Detect cases where alternate disappeared */
357 if (!is_directory(path
->buf
)) {
358 error("object directory %s does not exist; "
359 "check .git/objects/info/alternates.",
365 * Prevent the common mistake of listing the same
366 * thing twice, or object directory itself.
368 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
369 if (!fspathcmp(path
->buf
, alt
->path
))
372 if (!fspathcmp(path
->buf
, normalized_objdir
))
379 * Prepare alternate object database registry.
381 * The variable alt_odb_list points at the list of struct
382 * alternate_object_database. The elements on this list come from
383 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
384 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
385 * whose contents is similar to that environment variable but can be
386 * LF separated. Its base points at a statically allocated buffer that
387 * contains "/the/directory/corresponding/to/.git/objects/...", while
388 * its name points just after the slash at the end of ".git/objects/"
389 * in the example above, and has enough space to hold 40-byte hex
390 * SHA1, an extra slash for the first level indirection, and the
393 static void read_info_alternates(const char * relative_base
, int depth
);
394 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
395 int depth
, const char *normalized_objdir
)
397 struct alternate_object_database
*ent
;
398 struct strbuf pathbuf
= STRBUF_INIT
;
400 if (!is_absolute_path(entry
) && relative_base
) {
401 strbuf_realpath(&pathbuf
, relative_base
, 1);
402 strbuf_addch(&pathbuf
, '/');
404 strbuf_addstr(&pathbuf
, entry
);
406 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
407 error("unable to normalize alternate object path: %s",
409 strbuf_release(&pathbuf
);
414 * The trailing slash after the directory name is given by
415 * this function at the end. Remove duplicates.
417 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
418 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
420 if (!alt_odb_usable(&pathbuf
, normalized_objdir
)) {
421 strbuf_release(&pathbuf
);
425 ent
= alloc_alt_odb(pathbuf
.buf
);
427 /* add the alternate entry */
429 alt_odb_tail
= &(ent
->next
);
432 /* recursively add alternates */
433 read_info_alternates(pathbuf
.buf
, depth
+ 1);
435 strbuf_release(&pathbuf
);
439 static const char *parse_alt_odb_entry(const char *string
,
447 if (*string
== '#') {
448 /* comment; consume up to next separator */
449 end
= strchrnul(string
, sep
);
450 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
452 * quoted path; unquote_c_style has copied the
453 * data for us and set "end". Broken quoting (e.g.,
454 * an entry that doesn't end with a quote) falls
455 * back to the unquoted case below.
458 /* normal, unquoted path */
459 end
= strchrnul(string
, sep
);
460 strbuf_add(out
, string
, end
- string
);
468 static void link_alt_odb_entries(const char *alt
, int sep
,
469 const char *relative_base
, int depth
)
471 struct strbuf objdirbuf
= STRBUF_INIT
;
472 struct strbuf entry
= STRBUF_INIT
;
478 error("%s: ignoring alternate object stores, nesting too deep.",
483 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
484 if (strbuf_normalize_path(&objdirbuf
) < 0)
485 die("unable to normalize object directory: %s",
489 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
492 link_alt_odb_entry(entry
.buf
, relative_base
, depth
, objdirbuf
.buf
);
494 strbuf_release(&entry
);
495 strbuf_release(&objdirbuf
);
498 static void read_info_alternates(const char * relative_base
, int depth
)
501 struct strbuf buf
= STRBUF_INIT
;
503 path
= xstrfmt("%s/info/alternates", relative_base
);
504 if (strbuf_read_file(&buf
, path
, 1024) < 0) {
505 warn_on_fopen_errors(path
);
510 link_alt_odb_entries(buf
.buf
, '\n', relative_base
, depth
);
511 strbuf_release(&buf
);
515 struct alternate_object_database
*alloc_alt_odb(const char *dir
)
517 struct alternate_object_database
*ent
;
519 FLEX_ALLOC_STR(ent
, path
, dir
);
520 strbuf_init(&ent
->scratch
, 0);
521 strbuf_addf(&ent
->scratch
, "%s/", dir
);
522 ent
->base_len
= ent
->scratch
.len
;
527 void add_to_alternates_file(const char *reference
)
529 struct lock_file lock
= LOCK_INIT
;
530 char *alts
= git_pathdup("objects/info/alternates");
534 hold_lock_file_for_update(&lock
, alts
, LOCK_DIE_ON_ERROR
);
535 out
= fdopen_lock_file(&lock
, "w");
537 die_errno("unable to fdopen alternates lockfile");
539 in
= fopen(alts
, "r");
541 struct strbuf line
= STRBUF_INIT
;
543 while (strbuf_getline(&line
, in
) != EOF
) {
544 if (!strcmp(reference
, line
.buf
)) {
548 fprintf_or_die(out
, "%s\n", line
.buf
);
551 strbuf_release(&line
);
554 else if (errno
!= ENOENT
)
555 die_errno("unable to read alternates file");
558 rollback_lock_file(&lock
);
560 fprintf_or_die(out
, "%s\n", reference
);
561 if (commit_lock_file(&lock
))
562 die_errno("unable to move new alternates file into place");
564 link_alt_odb_entries(reference
, '\n', NULL
, 0);
569 void add_to_alternates_memory(const char *reference
)
572 * Make sure alternates are initialized, or else our entry may be
573 * overwritten when they are.
577 link_alt_odb_entries(reference
, '\n', NULL
, 0);
581 * Compute the exact path an alternate is at and returns it. In case of
582 * error NULL is returned and the human readable error is added to `err`
583 * `path` may be relative and should point to $GITDIR.
584 * `err` must not be null.
586 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
588 char *ref_git
= NULL
;
589 const char *repo
, *ref_git_s
;
592 ref_git_s
= real_path_if_valid(path
);
595 strbuf_addf(err
, _("path '%s' does not exist"), path
);
599 * Beware: read_gitfile(), real_path() and mkpath()
600 * return static buffer
602 ref_git
= xstrdup(ref_git_s
);
604 repo
= read_gitfile(ref_git
);
606 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
609 ref_git
= xstrdup(repo
);
612 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
613 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
615 ref_git
= ref_git_git
;
616 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
617 struct strbuf sb
= STRBUF_INIT
;
619 if (get_common_dir(&sb
, ref_git
)) {
621 _("reference repository '%s' as a linked "
622 "checkout is not supported yet."),
627 strbuf_addf(err
, _("reference repository '%s' is not a "
628 "local repository."), path
);
632 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
633 strbuf_addf(err
, _("reference repository '%s' is shallow"),
639 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
641 _("reference repository '%s' is grafted"),
649 FREE_AND_NULL(ref_git
);
655 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
657 struct alternate_object_database
*ent
;
661 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
669 void prepare_alt_odb(void)
674 alt_odb_tail
= &alt_odb_list
;
675 link_alt_odb_entries(the_repository
->alternate_db
, PATH_SEP
, NULL
, 0);
677 read_info_alternates(get_object_directory(), 0);
680 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
681 static int freshen_file(const char *fn
)
684 t
.actime
= t
.modtime
= time(NULL
);
685 return !utime(fn
, &t
);
689 * All of the check_and_freshen functions return 1 if the file exists and was
690 * freshened (if freshening was requested), 0 otherwise. If they return
691 * 0, you should not assume that it is safe to skip a write of the object (it
692 * either does not exist on disk, or has a stale mtime and may be subject to
695 int check_and_freshen_file(const char *fn
, int freshen
)
697 if (access(fn
, F_OK
))
699 if (freshen
&& !freshen_file(fn
))
704 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
706 static struct strbuf buf
= STRBUF_INIT
;
709 sha1_file_name(&buf
, sha1
);
711 return check_and_freshen_file(buf
.buf
, freshen
);
714 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
716 struct alternate_object_database
*alt
;
718 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
719 const char *path
= alt_sha1_path(alt
, sha1
);
720 if (check_and_freshen_file(path
, freshen
))
726 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
728 return check_and_freshen_local(sha1
, freshen
) ||
729 check_and_freshen_nonlocal(sha1
, freshen
);
732 int has_loose_object_nonlocal(const unsigned char *sha1
)
734 return check_and_freshen_nonlocal(sha1
, 0);
737 static int has_loose_object(const unsigned char *sha1
)
739 return check_and_freshen(sha1
, 0);
742 static void mmap_limit_check(size_t length
)
744 static size_t limit
= 0;
746 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
751 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
752 (uintmax_t)length
, (uintmax_t)limit
);
755 void *xmmap_gently(void *start
, size_t length
,
756 int prot
, int flags
, int fd
, off_t offset
)
760 mmap_limit_check(length
);
761 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
762 if (ret
== MAP_FAILED
) {
765 release_pack_memory(length
);
766 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
771 void *xmmap(void *start
, size_t length
,
772 int prot
, int flags
, int fd
, off_t offset
)
774 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
775 if (ret
== MAP_FAILED
)
776 die_errno("mmap failed");
781 * With an in-core object data in "map", rehash it to make sure the
782 * object name actually matches "sha1" to detect object corruption.
783 * With "map" == NULL, try reading the object named with "sha1" using
784 * the streaming interface and rehash it to do the same.
786 int check_object_signature(const struct object_id
*oid
, void *map
,
787 unsigned long size
, const char *type
)
789 struct object_id real_oid
;
790 enum object_type obj_type
;
791 struct git_istream
*st
;
793 char hdr
[MAX_HEADER_LEN
];
797 hash_object_file(map
, size
, type
, &real_oid
);
798 return oidcmp(oid
, &real_oid
) ? -1 : 0;
801 st
= open_istream(oid
, &obj_type
, &size
, NULL
);
805 /* Generate the header */
806 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", type_name(obj_type
), size
) + 1;
809 the_hash_algo
->init_fn(&c
);
810 the_hash_algo
->update_fn(&c
, hdr
, hdrlen
);
813 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
821 the_hash_algo
->update_fn(&c
, buf
, readlen
);
823 the_hash_algo
->final_fn(real_oid
.hash
, &c
);
825 return oidcmp(oid
, &real_oid
) ? -1 : 0;
828 int git_open_cloexec(const char *name
, int flags
)
831 static int o_cloexec
= O_CLOEXEC
;
833 fd
= open(name
, flags
| o_cloexec
);
834 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
835 /* Try again w/o O_CLOEXEC: the kernel might not support it */
836 o_cloexec
&= ~O_CLOEXEC
;
837 fd
= open(name
, flags
| o_cloexec
);
840 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
842 static int fd_cloexec
= FD_CLOEXEC
;
844 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
845 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
846 int flags
= fcntl(fd
, F_GETFD
);
847 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
856 * Find "sha1" as a loose object in the local repository or in an alternate.
857 * Returns 0 on success, negative on failure.
859 * The "path" out-parameter will give the path of the object we found (if any).
860 * Note that it may point to static storage and is only valid until another
861 * call to sha1_file_name(), etc.
863 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
,
866 struct alternate_object_database
*alt
;
867 static struct strbuf buf
= STRBUF_INIT
;
870 sha1_file_name(&buf
, sha1
);
873 if (!lstat(*path
, st
))
878 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
879 *path
= alt_sha1_path(alt
, sha1
);
880 if (!lstat(*path
, st
))
888 * Like stat_sha1_file(), but actually open the object and return the
889 * descriptor. See the caveats on the "path" parameter above.
891 static int open_sha1_file(const unsigned char *sha1
, const char **path
)
894 struct alternate_object_database
*alt
;
895 int most_interesting_errno
;
896 static struct strbuf buf
= STRBUF_INIT
;
899 sha1_file_name(&buf
, sha1
);
902 fd
= git_open(*path
);
905 most_interesting_errno
= errno
;
908 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
909 *path
= alt_sha1_path(alt
, sha1
);
910 fd
= git_open(*path
);
913 if (most_interesting_errno
== ENOENT
)
914 most_interesting_errno
= errno
;
916 errno
= most_interesting_errno
;
921 * Map the loose object at "path" if it is not NULL, or the path found by
922 * searching for a loose object named "sha1".
924 static void *map_sha1_file_1(const char *path
,
925 const unsigned char *sha1
,
934 fd
= open_sha1_file(sha1
, &path
);
939 if (!fstat(fd
, &st
)) {
940 *size
= xsize_t(st
.st_size
);
942 /* mmap() is forbidden on empty files */
943 error("object file %s is empty", path
);
946 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
953 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
955 return map_sha1_file_1(NULL
, sha1
, size
);
958 static int unpack_sha1_short_header(git_zstream
*stream
,
959 unsigned char *map
, unsigned long mapsize
,
960 void *buffer
, unsigned long bufsiz
)
962 /* Get the data stream */
963 memset(stream
, 0, sizeof(*stream
));
964 stream
->next_in
= map
;
965 stream
->avail_in
= mapsize
;
966 stream
->next_out
= buffer
;
967 stream
->avail_out
= bufsiz
;
969 git_inflate_init(stream
);
970 return git_inflate(stream
, 0);
973 int unpack_sha1_header(git_zstream
*stream
,
974 unsigned char *map
, unsigned long mapsize
,
975 void *buffer
, unsigned long bufsiz
)
977 int status
= unpack_sha1_short_header(stream
, map
, mapsize
,
983 /* Make sure we have the terminating NUL */
984 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
989 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
990 unsigned long mapsize
, void *buffer
,
991 unsigned long bufsiz
, struct strbuf
*header
)
995 status
= unpack_sha1_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1000 * Check if entire header is unpacked in the first iteration.
1002 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1006 * buffer[0..bufsiz] was not large enough. Copy the partial
1007 * result out to header, and then append the result of further
1008 * reading the stream.
1010 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1011 stream
->next_out
= buffer
;
1012 stream
->avail_out
= bufsiz
;
1015 status
= git_inflate(stream
, 0);
1016 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1017 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1019 stream
->next_out
= buffer
;
1020 stream
->avail_out
= bufsiz
;
1021 } while (status
!= Z_STREAM_END
);
1025 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1027 int bytes
= strlen(buffer
) + 1;
1028 unsigned char *buf
= xmallocz(size
);
1032 n
= stream
->total_out
- bytes
;
1035 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1037 if (bytes
<= size
) {
1039 * The above condition must be (bytes <= size), not
1040 * (bytes < size). In other words, even though we
1041 * expect no more output and set avail_out to zero,
1042 * the input zlib stream may have bytes that express
1043 * "this concludes the stream", and we *do* want to
1046 * Otherwise we would not be able to test that we
1047 * consumed all the input to reach the expected size;
1048 * we also want to check that zlib tells us that all
1049 * went well with status == Z_STREAM_END at the end.
1051 stream
->next_out
= buf
+ bytes
;
1052 stream
->avail_out
= size
- bytes
;
1053 while (status
== Z_OK
)
1054 status
= git_inflate(stream
, Z_FINISH
);
1056 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1057 git_inflate_end(stream
);
1062 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1063 else if (stream
->avail_in
)
1064 error("garbage at end of loose object '%s'",
1071 * We used to just use "sscanf()", but that's actually way
1072 * too permissive for what we want to check. So do an anal
1073 * object header parse by hand.
1075 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1078 const char *type_buf
= hdr
;
1080 int type
, type_len
= 0;
1083 * The type can be of any size but is followed by
1095 type
= type_from_string_gently(type_buf
, type_len
, 1);
1097 strbuf_add(oi
->type_name
, type_buf
, type_len
);
1099 * Set type to 0 if its an unknown object and
1100 * we're obtaining the type using '--allow-unknown-type'
1103 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
) && (type
< 0))
1106 die("invalid object type");
1111 * The length must follow immediately, and be in canonical
1112 * decimal format (ie "010" is not valid).
1114 size
= *hdr
++ - '0';
1119 unsigned long c
= *hdr
- '0';
1123 size
= size
* 10 + c
;
1131 * The length must be followed by a zero byte
1133 return *hdr
? -1 : type
;
1136 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1138 struct object_info oi
= OBJECT_INFO_INIT
;
1141 return parse_sha1_header_extended(hdr
, &oi
, 0);
1144 static int sha1_loose_object_info(const unsigned char *sha1
,
1145 struct object_info
*oi
,
1149 unsigned long mapsize
;
1152 char hdr
[MAX_HEADER_LEN
];
1153 struct strbuf hdrbuf
= STRBUF_INIT
;
1154 unsigned long size_scratch
;
1156 if (oi
->delta_base_sha1
)
1157 hashclr(oi
->delta_base_sha1
);
1160 * If we don't care about type or size, then we don't
1161 * need to look inside the object at all. Note that we
1162 * do not optimize out the stat call, even if the
1163 * caller doesn't care about the disk-size, since our
1164 * return value implicitly indicates whether the
1165 * object even exists.
1167 if (!oi
->typep
&& !oi
->type_name
&& !oi
->sizep
&& !oi
->contentp
) {
1170 if (stat_sha1_file(sha1
, &st
, &path
) < 0)
1173 *oi
->disk_sizep
= st
.st_size
;
1177 map
= map_sha1_file(sha1
, &mapsize
);
1182 oi
->sizep
= &size_scratch
;
1185 *oi
->disk_sizep
= mapsize
;
1186 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
)) {
1187 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
1188 status
= error("unable to unpack %s header with --allow-unknown-type",
1190 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1191 status
= error("unable to unpack %s header",
1195 else if (hdrbuf
.len
) {
1196 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
1197 status
= error("unable to parse %s header with --allow-unknown-type",
1199 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
1200 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1202 if (status
>= 0 && oi
->contentp
) {
1203 *oi
->contentp
= unpack_sha1_rest(&stream
, hdr
,
1205 if (!*oi
->contentp
) {
1206 git_inflate_end(&stream
);
1210 git_inflate_end(&stream
);
1212 munmap(map
, mapsize
);
1213 if (status
&& oi
->typep
)
1214 *oi
->typep
= status
;
1215 if (oi
->sizep
== &size_scratch
)
1217 strbuf_release(&hdrbuf
);
1218 oi
->whence
= OI_LOOSE
;
1219 return (status
< 0) ? status
: 0;
1222 int fetch_if_missing
= 1;
1224 int oid_object_info_extended(const struct object_id
*oid
, struct object_info
*oi
, unsigned flags
)
1226 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
1227 struct pack_entry e
;
1229 const struct object_id
*real
= oid
;
1230 int already_retried
= 0;
1232 if (flags
& OBJECT_INFO_LOOKUP_REPLACE
)
1233 real
= lookup_replace_object(oid
);
1235 if (is_null_oid(real
))
1241 if (!(flags
& OBJECT_INFO_SKIP_CACHED
)) {
1242 struct cached_object
*co
= find_cached_object(real
->hash
);
1245 *(oi
->typep
) = co
->type
;
1247 *(oi
->sizep
) = co
->size
;
1249 *(oi
->disk_sizep
) = 0;
1250 if (oi
->delta_base_sha1
)
1251 hashclr(oi
->delta_base_sha1
);
1253 strbuf_addstr(oi
->type_name
, type_name(co
->type
));
1255 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
1256 oi
->whence
= OI_CACHED
;
1262 if (find_pack_entry(real
->hash
, &e
))
1265 if (flags
& OBJECT_INFO_IGNORE_LOOSE
)
1268 /* Most likely it's a loose object. */
1269 if (!sha1_loose_object_info(real
->hash
, oi
, flags
))
1272 /* Not a loose object; someone else may have just packed it. */
1273 if (!(flags
& OBJECT_INFO_QUICK
)) {
1274 reprepare_packed_git();
1275 if (find_pack_entry(real
->hash
, &e
))
1279 /* Check if it is a missing object */
1280 if (fetch_if_missing
&& repository_format_partial_clone
&&
1283 * TODO Investigate haveing fetch_object() return
1284 * TODO error/success and stopping the music here.
1286 fetch_object(repository_format_partial_clone
, real
->hash
);
1287 already_retried
= 1;
1294 if (oi
== &blank_oi
)
1296 * We know that the caller doesn't actually need the
1297 * information below, so return early.
1300 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
1302 mark_bad_packed_object(e
.p
, real
->hash
);
1303 return oid_object_info_extended(real
, oi
, 0);
1304 } else if (oi
->whence
== OI_PACKED
) {
1305 oi
->u
.packed
.offset
= e
.offset
;
1306 oi
->u
.packed
.pack
= e
.p
;
1307 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
1308 rtype
== OBJ_OFS_DELTA
);
1314 /* returns enum object_type or negative */
1315 int oid_object_info(const struct object_id
*oid
, unsigned long *sizep
)
1317 enum object_type type
;
1318 struct object_info oi
= OBJECT_INFO_INIT
;
1322 if (oid_object_info_extended(oid
, &oi
,
1323 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
1328 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1329 unsigned long *size
)
1331 struct object_id oid
;
1332 struct object_info oi
= OBJECT_INFO_INIT
;
1336 oi
.contentp
= &content
;
1338 hashcpy(oid
.hash
, sha1
);
1340 if (oid_object_info_extended(&oid
, &oi
, 0) < 0)
1345 int pretend_object_file(void *buf
, unsigned long len
, enum object_type type
,
1346 struct object_id
*oid
)
1348 struct cached_object
*co
;
1350 hash_object_file(buf
, len
, type_name(type
), oid
);
1351 if (has_sha1_file(oid
->hash
) || find_cached_object(oid
->hash
))
1353 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
1354 co
= &cached_objects
[cached_object_nr
++];
1357 co
->buf
= xmalloc(len
);
1358 memcpy(co
->buf
, buf
, len
);
1359 hashcpy(co
->sha1
, oid
->hash
);
1364 * This function dies on corrupt objects; the callers who want to
1365 * deal with them should arrange to call read_object() and give error
1366 * messages themselves.
1368 void *read_object_file_extended(const struct object_id
*oid
,
1369 enum object_type
*type
,
1370 unsigned long *size
,
1374 const struct packed_git
*p
;
1377 const struct object_id
*repl
= lookup_replace
? lookup_replace_object(oid
)
1381 data
= read_object(repl
->hash
, type
, size
);
1385 if (errno
&& errno
!= ENOENT
)
1386 die_errno("failed to read object %s", oid_to_hex(oid
));
1388 /* die if we replaced an object with one that does not exist */
1390 die("replacement %s not found for %s",
1391 oid_to_hex(repl
), oid_to_hex(oid
));
1393 if (!stat_sha1_file(repl
->hash
, &st
, &path
))
1394 die("loose object %s (stored in %s) is corrupt",
1395 oid_to_hex(repl
), path
);
1397 if ((p
= has_packed_and_bad(repl
->hash
)) != NULL
)
1398 die("packed object %s (stored in %s) is corrupt",
1399 oid_to_hex(repl
), p
->pack_name
);
1404 void *read_object_with_reference(const struct object_id
*oid
,
1405 const char *required_type_name
,
1406 unsigned long *size
,
1407 struct object_id
*actual_oid_return
)
1409 enum object_type type
, required_type
;
1411 unsigned long isize
;
1412 struct object_id actual_oid
;
1414 required_type
= type_from_string(required_type_name
);
1415 oidcpy(&actual_oid
, oid
);
1417 int ref_length
= -1;
1418 const char *ref_type
= NULL
;
1420 buffer
= read_object_file(&actual_oid
, &type
, &isize
);
1423 if (type
== required_type
) {
1425 if (actual_oid_return
)
1426 oidcpy(actual_oid_return
, &actual_oid
);
1429 /* Handle references */
1430 else if (type
== OBJ_COMMIT
)
1432 else if (type
== OBJ_TAG
)
1433 ref_type
= "object ";
1438 ref_length
= strlen(ref_type
);
1440 if (ref_length
+ GIT_SHA1_HEXSZ
> isize
||
1441 memcmp(buffer
, ref_type
, ref_length
) ||
1442 get_oid_hex((char *) buffer
+ ref_length
, &actual_oid
)) {
1447 /* Now we have the ID of the referred-to object in
1448 * actual_oid. Check again. */
1452 static void write_object_file_prepare(const void *buf
, unsigned long len
,
1453 const char *type
, struct object_id
*oid
,
1454 char *hdr
, int *hdrlen
)
1458 /* Generate the header */
1459 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
1462 the_hash_algo
->init_fn(&c
);
1463 the_hash_algo
->update_fn(&c
, hdr
, *hdrlen
);
1464 the_hash_algo
->update_fn(&c
, buf
, len
);
1465 the_hash_algo
->final_fn(oid
->hash
, &c
);
1469 * Move the just written object into its final resting place.
1471 int finalize_object_file(const char *tmpfile
, const char *filename
)
1475 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
1477 else if (link(tmpfile
, filename
))
1481 * Coda hack - coda doesn't like cross-directory links,
1482 * so we fall back to a rename, which will mean that it
1483 * won't be able to check collisions, but that's not a
1486 * The same holds for FAT formatted media.
1488 * When this succeeds, we just return. We have nothing
1491 if (ret
&& ret
!= EEXIST
) {
1493 if (!rename(tmpfile
, filename
))
1497 unlink_or_warn(tmpfile
);
1499 if (ret
!= EEXIST
) {
1500 return error_errno("unable to write sha1 filename %s", filename
);
1502 /* FIXME!!! Collision check here ? */
1506 if (adjust_shared_perm(filename
))
1507 return error("unable to set permission to '%s'", filename
);
1511 static int write_buffer(int fd
, const void *buf
, size_t len
)
1513 if (write_in_full(fd
, buf
, len
) < 0)
1514 return error_errno("file write error");
1518 int hash_object_file(const void *buf
, unsigned long len
, const char *type
,
1519 struct object_id
*oid
)
1521 char hdr
[MAX_HEADER_LEN
];
1522 int hdrlen
= sizeof(hdr
);
1523 write_object_file_prepare(buf
, len
, type
, oid
, hdr
, &hdrlen
);
1527 /* Finalize a file on disk, and close it. */
1528 static void close_sha1_file(int fd
)
1530 if (fsync_object_files
)
1531 fsync_or_die(fd
, "sha1 file");
1533 die_errno("error when closing sha1 file");
1536 /* Size of directory component, including the ending '/' */
1537 static inline int directory_size(const char *filename
)
1539 const char *s
= strrchr(filename
, '/');
1542 return s
- filename
+ 1;
1546 * This creates a temporary file in the same directory as the final
1549 * We want to avoid cross-directory filename renames, because those
1550 * can have problems on various filesystems (FAT, NFS, Coda).
1552 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
1554 int fd
, dirlen
= directory_size(filename
);
1557 strbuf_add(tmp
, filename
, dirlen
);
1558 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
1559 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1560 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
1562 * Make sure the directory exists; note that the contents
1563 * of the buffer are undefined after mkstemp returns an
1564 * error, so we have to rewrite the whole buffer from
1568 strbuf_add(tmp
, filename
, dirlen
- 1);
1569 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
1571 if (adjust_shared_perm(tmp
->buf
))
1575 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
1576 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1581 static int write_loose_object(const struct object_id
*oid
, char *hdr
,
1582 int hdrlen
, const void *buf
, unsigned long len
,
1586 unsigned char compressed
[4096];
1589 struct object_id parano_oid
;
1590 static struct strbuf tmp_file
= STRBUF_INIT
;
1591 static struct strbuf filename
= STRBUF_INIT
;
1593 strbuf_reset(&filename
);
1594 sha1_file_name(&filename
, oid
->hash
);
1596 fd
= create_tmpfile(&tmp_file
, filename
.buf
);
1598 if (errno
== EACCES
)
1599 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1601 return error_errno("unable to create temporary file");
1605 git_deflate_init(&stream
, zlib_compression_level
);
1606 stream
.next_out
= compressed
;
1607 stream
.avail_out
= sizeof(compressed
);
1608 the_hash_algo
->init_fn(&c
);
1610 /* First header.. */
1611 stream
.next_in
= (unsigned char *)hdr
;
1612 stream
.avail_in
= hdrlen
;
1613 while (git_deflate(&stream
, 0) == Z_OK
)
1615 the_hash_algo
->update_fn(&c
, hdr
, hdrlen
);
1617 /* Then the data itself.. */
1618 stream
.next_in
= (void *)buf
;
1619 stream
.avail_in
= len
;
1621 unsigned char *in0
= stream
.next_in
;
1622 ret
= git_deflate(&stream
, Z_FINISH
);
1623 the_hash_algo
->update_fn(&c
, in0
, stream
.next_in
- in0
);
1624 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
1625 die("unable to write sha1 file");
1626 stream
.next_out
= compressed
;
1627 stream
.avail_out
= sizeof(compressed
);
1628 } while (ret
== Z_OK
);
1630 if (ret
!= Z_STREAM_END
)
1631 die("unable to deflate new object %s (%d)", oid_to_hex(oid
),
1633 ret
= git_deflate_end_gently(&stream
);
1635 die("deflateEnd on object %s failed (%d)", oid_to_hex(oid
),
1637 the_hash_algo
->final_fn(parano_oid
.hash
, &c
);
1638 if (oidcmp(oid
, ¶no_oid
) != 0)
1639 die("confused by unstable object source data for %s",
1642 close_sha1_file(fd
);
1647 utb
.modtime
= mtime
;
1648 if (utime(tmp_file
.buf
, &utb
) < 0)
1649 warning_errno("failed utime() on %s", tmp_file
.buf
);
1652 return finalize_object_file(tmp_file
.buf
, filename
.buf
);
1655 static int freshen_loose_object(const unsigned char *sha1
)
1657 return check_and_freshen(sha1
, 1);
1660 static int freshen_packed_object(const unsigned char *sha1
)
1662 struct pack_entry e
;
1663 if (!find_pack_entry(sha1
, &e
))
1667 if (!freshen_file(e
.p
->pack_name
))
1673 int write_object_file(const void *buf
, unsigned long len
, const char *type
,
1674 struct object_id
*oid
)
1676 char hdr
[MAX_HEADER_LEN
];
1677 int hdrlen
= sizeof(hdr
);
1679 /* Normally if we have it in the pack then we do not bother writing
1680 * it out into .git/objects/??/?{38} file.
1682 write_object_file_prepare(buf
, len
, type
, oid
, hdr
, &hdrlen
);
1683 if (freshen_packed_object(oid
->hash
) || freshen_loose_object(oid
->hash
))
1685 return write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, 0);
1688 int hash_object_file_literally(const void *buf
, unsigned long len
,
1689 const char *type
, struct object_id
*oid
,
1693 int hdrlen
, status
= 0;
1695 /* type string, SP, %lu of the length plus NUL must fit this */
1696 hdrlen
= strlen(type
) + MAX_HEADER_LEN
;
1697 header
= xmalloc(hdrlen
);
1698 write_object_file_prepare(buf
, len
, type
, oid
, header
, &hdrlen
);
1700 if (!(flags
& HASH_WRITE_OBJECT
))
1702 if (freshen_packed_object(oid
->hash
) || freshen_loose_object(oid
->hash
))
1704 status
= write_loose_object(oid
, header
, hdrlen
, buf
, len
, 0);
1711 int force_object_loose(const struct object_id
*oid
, time_t mtime
)
1715 enum object_type type
;
1716 char hdr
[MAX_HEADER_LEN
];
1720 if (has_loose_object(oid
->hash
))
1722 buf
= read_object(oid
->hash
, &type
, &len
);
1724 return error("cannot read sha1_file for %s", oid_to_hex(oid
));
1725 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", type_name(type
), len
) + 1;
1726 ret
= write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, mtime
);
1732 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
1734 struct object_id oid
;
1735 if (!startup_info
->have_repository
)
1737 hashcpy(oid
.hash
, sha1
);
1738 return oid_object_info_extended(&oid
, NULL
,
1739 flags
| OBJECT_INFO_SKIP_CACHED
) >= 0;
1742 int has_object_file(const struct object_id
*oid
)
1744 return has_sha1_file(oid
->hash
);
1747 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
1749 return has_sha1_file_with_flags(oid
->hash
, flags
);
1752 static void check_tree(const void *buf
, size_t size
)
1754 struct tree_desc desc
;
1755 struct name_entry entry
;
1757 init_tree_desc(&desc
, buf
, size
);
1758 while (tree_entry(&desc
, &entry
))
1760 * tree_entry() will die() on malformed entries */
1764 static void check_commit(const void *buf
, size_t size
)
1767 memset(&c
, 0, sizeof(c
));
1768 if (parse_commit_buffer(&c
, buf
, size
))
1769 die("corrupt commit");
1772 static void check_tag(const void *buf
, size_t size
)
1775 memset(&t
, 0, sizeof(t
));
1776 if (parse_tag_buffer(&t
, buf
, size
))
1780 static int index_mem(struct object_id
*oid
, void *buf
, size_t size
,
1781 enum object_type type
,
1782 const char *path
, unsigned flags
)
1784 int ret
, re_allocated
= 0;
1785 int write_object
= flags
& HASH_WRITE_OBJECT
;
1791 * Convert blobs to git internal format
1793 if ((type
== OBJ_BLOB
) && path
) {
1794 struct strbuf nbuf
= STRBUF_INIT
;
1795 if (convert_to_git(&the_index
, path
, buf
, size
, &nbuf
,
1796 get_conv_flags(flags
))) {
1797 buf
= strbuf_detach(&nbuf
, &size
);
1801 if (flags
& HASH_FORMAT_CHECK
) {
1802 if (type
== OBJ_TREE
)
1803 check_tree(buf
, size
);
1804 if (type
== OBJ_COMMIT
)
1805 check_commit(buf
, size
);
1806 if (type
== OBJ_TAG
)
1807 check_tag(buf
, size
);
1811 ret
= write_object_file(buf
, size
, type_name(type
), oid
);
1813 ret
= hash_object_file(buf
, size
, type_name(type
), oid
);
1819 static int index_stream_convert_blob(struct object_id
*oid
, int fd
,
1820 const char *path
, unsigned flags
)
1823 const int write_object
= flags
& HASH_WRITE_OBJECT
;
1824 struct strbuf sbuf
= STRBUF_INIT
;
1827 assert(would_convert_to_git_filter_fd(path
));
1829 convert_to_git_filter_fd(&the_index
, path
, fd
, &sbuf
,
1830 get_conv_flags(flags
));
1833 ret
= write_object_file(sbuf
.buf
, sbuf
.len
, type_name(OBJ_BLOB
),
1836 ret
= hash_object_file(sbuf
.buf
, sbuf
.len
, type_name(OBJ_BLOB
),
1838 strbuf_release(&sbuf
);
1842 static int index_pipe(struct object_id
*oid
, int fd
, enum object_type type
,
1843 const char *path
, unsigned flags
)
1845 struct strbuf sbuf
= STRBUF_INIT
;
1848 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
1849 ret
= index_mem(oid
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
1852 strbuf_release(&sbuf
);
1856 #define SMALL_FILE_SIZE (32*1024)
1858 static int index_core(struct object_id
*oid
, int fd
, size_t size
,
1859 enum object_type type
, const char *path
,
1865 ret
= index_mem(oid
, "", size
, type
, path
, flags
);
1866 } else if (size
<= SMALL_FILE_SIZE
) {
1867 char *buf
= xmalloc(size
);
1868 ssize_t read_result
= read_in_full(fd
, buf
, size
);
1869 if (read_result
< 0)
1870 ret
= error_errno("read error while indexing %s",
1871 path
? path
: "<unknown>");
1872 else if (read_result
!= size
)
1873 ret
= error("short read while indexing %s",
1874 path
? path
: "<unknown>");
1876 ret
= index_mem(oid
, buf
, size
, type
, path
, flags
);
1879 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1880 ret
= index_mem(oid
, buf
, size
, type
, path
, flags
);
1887 * This creates one packfile per large blob unless bulk-checkin
1888 * machinery is "plugged".
1890 * This also bypasses the usual "convert-to-git" dance, and that is on
1891 * purpose. We could write a streaming version of the converting
1892 * functions and insert that before feeding the data to fast-import
1893 * (or equivalent in-core API described above). However, that is
1894 * somewhat complicated, as we do not know the size of the filter
1895 * result, which we need to know beforehand when writing a git object.
1896 * Since the primary motivation for trying to stream from the working
1897 * tree file and to avoid mmaping it in core is to deal with large
1898 * binary blobs, they generally do not want to get any conversion, and
1899 * callers should avoid this code path when filters are requested.
1901 static int index_stream(struct object_id
*oid
, int fd
, size_t size
,
1902 enum object_type type
, const char *path
,
1905 return index_bulk_checkin(oid
, fd
, size
, type
, path
, flags
);
1908 int index_fd(struct object_id
*oid
, int fd
, struct stat
*st
,
1909 enum object_type type
, const char *path
, unsigned flags
)
1914 * Call xsize_t() only when needed to avoid potentially unnecessary
1915 * die() for large files.
1917 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
1918 ret
= index_stream_convert_blob(oid
, fd
, path
, flags
);
1919 else if (!S_ISREG(st
->st_mode
))
1920 ret
= index_pipe(oid
, fd
, type
, path
, flags
);
1921 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
1922 (path
&& would_convert_to_git(&the_index
, path
)))
1923 ret
= index_core(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
1926 ret
= index_stream(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
1932 int index_path(struct object_id
*oid
, const char *path
, struct stat
*st
, unsigned flags
)
1935 struct strbuf sb
= STRBUF_INIT
;
1938 switch (st
->st_mode
& S_IFMT
) {
1940 fd
= open(path
, O_RDONLY
);
1942 return error_errno("open(\"%s\")", path
);
1943 if (index_fd(oid
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
1944 return error("%s: failed to insert into database",
1948 if (strbuf_readlink(&sb
, path
, st
->st_size
))
1949 return error_errno("readlink(\"%s\")", path
);
1950 if (!(flags
& HASH_WRITE_OBJECT
))
1951 hash_object_file(sb
.buf
, sb
.len
, blob_type
, oid
);
1952 else if (write_object_file(sb
.buf
, sb
.len
, blob_type
, oid
))
1953 rc
= error("%s: failed to insert into database", path
);
1954 strbuf_release(&sb
);
1957 return resolve_gitlink_ref(path
, "HEAD", oid
);
1959 return error("%s: unsupported file type", path
);
1964 int read_pack_header(int fd
, struct pack_header
*header
)
1966 if (read_in_full(fd
, header
, sizeof(*header
)) != sizeof(*header
))
1967 /* "eof before pack header was fully read" */
1968 return PH_ERROR_EOF
;
1970 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
1971 /* "protocol error (pack signature mismatch detected)" */
1972 return PH_ERROR_PACK_SIGNATURE
;
1973 if (!pack_version_ok(header
->hdr_version
))
1974 /* "protocol error (pack version unsupported)" */
1975 return PH_ERROR_PROTOCOL
;
1979 void assert_oid_type(const struct object_id
*oid
, enum object_type expect
)
1981 enum object_type type
= oid_object_info(oid
, NULL
);
1983 die("%s is not a valid object", oid_to_hex(oid
));
1985 die("%s is not a valid '%s' object", oid_to_hex(oid
),
1989 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
1990 struct strbuf
*path
,
1991 each_loose_object_fn obj_cb
,
1992 each_loose_cruft_fn cruft_cb
,
1993 each_loose_subdir_fn subdir_cb
,
1996 size_t origlen
, baselen
;
2000 struct object_id oid
;
2002 if (subdir_nr
> 0xff)
2003 BUG("invalid loose object subdirectory: %x", subdir_nr
);
2005 origlen
= path
->len
;
2006 strbuf_complete(path
, '/');
2007 strbuf_addf(path
, "%02x", subdir_nr
);
2009 dir
= opendir(path
->buf
);
2011 if (errno
!= ENOENT
)
2012 r
= error_errno("unable to open %s", path
->buf
);
2013 strbuf_setlen(path
, origlen
);
2017 oid
.hash
[0] = subdir_nr
;
2018 strbuf_addch(path
, '/');
2019 baselen
= path
->len
;
2021 while ((de
= readdir(dir
))) {
2023 if (is_dot_or_dotdot(de
->d_name
))
2026 namelen
= strlen(de
->d_name
);
2027 strbuf_setlen(path
, baselen
);
2028 strbuf_add(path
, de
->d_name
, namelen
);
2029 if (namelen
== GIT_SHA1_HEXSZ
- 2 &&
2030 !hex_to_bytes(oid
.hash
+ 1, de
->d_name
,
2031 GIT_SHA1_RAWSZ
- 1)) {
2033 r
= obj_cb(&oid
, path
->buf
, data
);
2041 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
2048 strbuf_setlen(path
, baselen
- 1);
2049 if (!r
&& subdir_cb
)
2050 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
2052 strbuf_setlen(path
, origlen
);
2057 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
2058 each_loose_object_fn obj_cb
,
2059 each_loose_cruft_fn cruft_cb
,
2060 each_loose_subdir_fn subdir_cb
,
2066 for (i
= 0; i
< 256; i
++) {
2067 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
2076 int for_each_loose_file_in_objdir(const char *path
,
2077 each_loose_object_fn obj_cb
,
2078 each_loose_cruft_fn cruft_cb
,
2079 each_loose_subdir_fn subdir_cb
,
2082 struct strbuf buf
= STRBUF_INIT
;
2085 strbuf_addstr(&buf
, path
);
2086 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
2088 strbuf_release(&buf
);
2093 struct loose_alt_odb_data
{
2094 each_loose_object_fn
*cb
;
2098 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
2101 struct loose_alt_odb_data
*data
= vdata
;
2102 struct strbuf buf
= STRBUF_INIT
;
2105 strbuf_addstr(&buf
, alt
->path
);
2106 r
= for_each_loose_file_in_objdir_buf(&buf
,
2107 data
->cb
, NULL
, NULL
,
2109 strbuf_release(&buf
);
2113 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
2115 struct loose_alt_odb_data alt
;
2118 r
= for_each_loose_file_in_objdir(get_object_directory(),
2119 cb
, NULL
, NULL
, data
);
2123 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
2128 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
2131 static int check_stream_sha1(git_zstream
*stream
,
2135 const unsigned char *expected_sha1
)
2138 unsigned char real_sha1
[GIT_MAX_RAWSZ
];
2139 unsigned char buf
[4096];
2140 unsigned long total_read
;
2143 the_hash_algo
->init_fn(&c
);
2144 the_hash_algo
->update_fn(&c
, hdr
, stream
->total_out
);
2147 * We already read some bytes into hdr, but the ones up to the NUL
2148 * do not count against the object's content size.
2150 total_read
= stream
->total_out
- strlen(hdr
) - 1;
2153 * This size comparison must be "<=" to read the final zlib packets;
2154 * see the comment in unpack_sha1_rest for details.
2156 while (total_read
<= size
&&
2157 (status
== Z_OK
|| status
== Z_BUF_ERROR
)) {
2158 stream
->next_out
= buf
;
2159 stream
->avail_out
= sizeof(buf
);
2160 if (size
- total_read
< stream
->avail_out
)
2161 stream
->avail_out
= size
- total_read
;
2162 status
= git_inflate(stream
, Z_FINISH
);
2163 the_hash_algo
->update_fn(&c
, buf
, stream
->next_out
- buf
);
2164 total_read
+= stream
->next_out
- buf
;
2166 git_inflate_end(stream
);
2168 if (status
!= Z_STREAM_END
) {
2169 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1
));
2172 if (stream
->avail_in
) {
2173 error("garbage at end of loose object '%s'",
2174 sha1_to_hex(expected_sha1
));
2178 the_hash_algo
->final_fn(real_sha1
, &c
);
2179 if (hashcmp(expected_sha1
, real_sha1
)) {
2180 error("sha1 mismatch for %s (expected %s)", path
,
2181 sha1_to_hex(expected_sha1
));
2188 int read_loose_object(const char *path
,
2189 const struct object_id
*expected_oid
,
2190 enum object_type
*type
,
2191 unsigned long *size
,
2196 unsigned long mapsize
;
2198 char hdr
[MAX_HEADER_LEN
];
2202 map
= map_sha1_file_1(path
, NULL
, &mapsize
);
2204 error_errno("unable to mmap %s", path
);
2208 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0) {
2209 error("unable to unpack header of %s", path
);
2213 *type
= parse_sha1_header(hdr
, size
);
2215 error("unable to parse header of %s", path
);
2216 git_inflate_end(&stream
);
2220 if (*type
== OBJ_BLOB
) {
2221 if (check_stream_sha1(&stream
, hdr
, *size
, path
, expected_oid
->hash
) < 0)
2224 *contents
= unpack_sha1_rest(&stream
, hdr
, *size
, expected_oid
->hash
);
2226 error("unable to unpack contents of %s", path
);
2227 git_inflate_end(&stream
);
2230 if (check_object_signature(expected_oid
, *contents
,
2231 *size
, type_name(*type
))) {
2232 error("sha1 mismatch for %s (expected %s)", path
,
2233 oid_to_hex(expected_oid
));
2239 ret
= 0; /* everything checks out */
2243 munmap(map
, mapsize
);