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 const unsigned char null_sha1
[GIT_MAX_RAWSZ
];
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
42 static void git_hash_sha1_init(git_hash_ctx
*ctx
)
44 git_SHA1_Init(&ctx
->sha1
);
47 static void git_hash_sha1_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
49 git_SHA1_Update(&ctx
->sha1
, data
, len
);
52 static void git_hash_sha1_final(unsigned char *hash
, git_hash_ctx
*ctx
)
54 git_SHA1_Final(hash
, &ctx
->sha1
);
57 static void git_hash_unknown_init(git_hash_ctx
*ctx
)
59 die("trying to init unknown hash");
62 static void git_hash_unknown_update(git_hash_ctx
*ctx
, const void *data
, size_t len
)
64 die("trying to update unknown hash");
67 static void git_hash_unknown_final(unsigned char *hash
, git_hash_ctx
*ctx
)
69 die("trying to finalize unknown hash");
72 const struct git_hash_algo hash_algos
[GIT_HASH_NALGOS
] = {
78 git_hash_unknown_init
,
79 git_hash_unknown_update
,
80 git_hash_unknown_final
,
86 /* "sha1", big-endian */
99 * This is meant to hold a *small* number of objects that you would
100 * want read_sha1_file() to be able to return, but yet you do not want
101 * to write them into the object store (e.g. a browse-only
104 static struct cached_object
{
105 unsigned char sha1
[20];
106 enum object_type type
;
110 static int cached_object_nr
, cached_object_alloc
;
112 static struct cached_object empty_tree
= {
113 EMPTY_TREE_SHA1_BIN_LITERAL
,
119 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
122 struct cached_object
*co
= cached_objects
;
124 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
125 if (!hashcmp(co
->sha1
, sha1
))
128 if (!hashcmp(sha1
, empty_tree
.sha1
))
134 static int get_conv_flags(unsigned flags
)
136 if (flags
& HASH_RENORMALIZE
)
137 return CONV_EOL_RENORMALIZE
;
138 else if (flags
& HASH_WRITE_OBJECT
)
139 return global_conv_flags_eol
;
145 int mkdir_in_gitdir(const char *path
)
147 if (mkdir(path
, 0777)) {
148 int saved_errno
= errno
;
150 struct strbuf sb
= STRBUF_INIT
;
155 * Are we looking at a path in a symlinked worktree
156 * whose original repository does not yet have it?
157 * e.g. .git/rr-cache pointing at its original
158 * repository in which the user hasn't performed any
159 * conflict resolution yet?
161 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
162 strbuf_readlink(&sb
, path
, st
.st_size
) ||
163 !is_absolute_path(sb
.buf
) ||
164 mkdir(sb
.buf
, 0777)) {
171 return adjust_shared_perm(path
);
174 enum scld_error
safe_create_leading_directories(char *path
)
176 char *next_component
= path
+ offset_1st_component(path
);
177 enum scld_error ret
= SCLD_OK
;
179 while (ret
== SCLD_OK
&& next_component
) {
181 char *slash
= next_component
, slash_character
;
183 while (*slash
&& !is_dir_sep(*slash
))
189 next_component
= slash
+ 1;
190 while (is_dir_sep(*next_component
))
192 if (!*next_component
)
195 slash_character
= *slash
;
197 if (!stat(path
, &st
)) {
199 if (!S_ISDIR(st
.st_mode
)) {
203 } else if (mkdir(path
, 0777)) {
204 if (errno
== EEXIST
&&
205 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
206 ; /* somebody created it since we checked */
207 else if (errno
== ENOENT
)
209 * Either mkdir() failed because
210 * somebody just pruned the containing
211 * directory, or stat() failed because
212 * the file that was in our way was
213 * just removed. Either way, inform
214 * the caller that it might be worth
220 } else if (adjust_shared_perm(path
)) {
223 *slash
= slash_character
;
228 enum scld_error
safe_create_leading_directories_const(const char *path
)
231 /* path points to cache entries, so xstrdup before messing with it */
232 char *buf
= xstrdup(path
);
233 enum scld_error result
= safe_create_leading_directories(buf
);
241 int raceproof_create_file(const char *path
, create_file_fn fn
, void *cb
)
244 * The number of times we will try to remove empty directories
245 * in the way of path. This is only 1 because if another
246 * process is racily creating directories that conflict with
247 * us, we don't want to fight against them.
249 int remove_directories_remaining
= 1;
252 * The number of times that we will try to create the
253 * directories containing path. We are willing to attempt this
254 * more than once, because another process could be trying to
255 * clean up empty directories at the same time as we are
256 * trying to create them.
258 int create_directories_remaining
= 3;
260 /* A scratch copy of path, filled lazily if we need it: */
261 struct strbuf path_copy
= STRBUF_INIT
;
274 if (errno
== EISDIR
&& remove_directories_remaining
-- > 0) {
276 * A directory is in the way. Maybe it is empty; try
280 strbuf_addstr(&path_copy
, path
);
282 if (!remove_dir_recursively(&path_copy
, REMOVE_DIR_EMPTY_ONLY
))
284 } else if (errno
== ENOENT
&& create_directories_remaining
-- > 0) {
286 * Maybe the containing directory didn't exist, or
287 * maybe it was just deleted by a process that is
288 * racing with us to clean up empty directories. Try
291 enum scld_error scld_result
;
294 strbuf_addstr(&path_copy
, path
);
297 scld_result
= safe_create_leading_directories(path_copy
.buf
);
298 if (scld_result
== SCLD_OK
)
300 } while (scld_result
== SCLD_VANISHED
&& create_directories_remaining
-- > 0);
304 strbuf_release(&path_copy
);
309 static void fill_sha1_path(struct strbuf
*buf
, const unsigned char *sha1
)
312 for (i
= 0; i
< 20; i
++) {
313 static char hex
[] = "0123456789abcdef";
314 unsigned int val
= sha1
[i
];
315 strbuf_addch(buf
, hex
[val
>> 4]);
316 strbuf_addch(buf
, hex
[val
& 0xf]);
318 strbuf_addch(buf
, '/');
322 void sha1_file_name(struct strbuf
*buf
, const unsigned char *sha1
)
324 strbuf_addstr(buf
, get_object_directory());
325 strbuf_addch(buf
, '/');
326 fill_sha1_path(buf
, sha1
);
329 struct strbuf
*alt_scratch_buf(struct alternate_object_database
*alt
)
331 strbuf_setlen(&alt
->scratch
, alt
->base_len
);
332 return &alt
->scratch
;
335 static const char *alt_sha1_path(struct alternate_object_database
*alt
,
336 const unsigned char *sha1
)
338 struct strbuf
*buf
= alt_scratch_buf(alt
);
339 fill_sha1_path(buf
, sha1
);
343 struct alternate_object_database
*alt_odb_list
;
344 static struct alternate_object_database
**alt_odb_tail
;
347 * Return non-zero iff the path is usable as an alternate object database.
349 static int alt_odb_usable(struct strbuf
*path
, const char *normalized_objdir
)
351 struct alternate_object_database
*alt
;
353 /* Detect cases where alternate disappeared */
354 if (!is_directory(path
->buf
)) {
355 error("object directory %s does not exist; "
356 "check .git/objects/info/alternates.",
362 * Prevent the common mistake of listing the same
363 * thing twice, or object directory itself.
365 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
366 if (!fspathcmp(path
->buf
, alt
->path
))
369 if (!fspathcmp(path
->buf
, normalized_objdir
))
376 * Prepare alternate object database registry.
378 * The variable alt_odb_list points at the list of struct
379 * alternate_object_database. The elements on this list come from
380 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
381 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
382 * whose contents is similar to that environment variable but can be
383 * LF separated. Its base points at a statically allocated buffer that
384 * contains "/the/directory/corresponding/to/.git/objects/...", while
385 * its name points just after the slash at the end of ".git/objects/"
386 * in the example above, and has enough space to hold 40-byte hex
387 * SHA1, an extra slash for the first level indirection, and the
390 static void read_info_alternates(const char * relative_base
, int depth
);
391 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
392 int depth
, const char *normalized_objdir
)
394 struct alternate_object_database
*ent
;
395 struct strbuf pathbuf
= STRBUF_INIT
;
397 if (!is_absolute_path(entry
) && relative_base
) {
398 strbuf_realpath(&pathbuf
, relative_base
, 1);
399 strbuf_addch(&pathbuf
, '/');
401 strbuf_addstr(&pathbuf
, entry
);
403 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
404 error("unable to normalize alternate object path: %s",
406 strbuf_release(&pathbuf
);
411 * The trailing slash after the directory name is given by
412 * this function at the end. Remove duplicates.
414 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
415 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
417 if (!alt_odb_usable(&pathbuf
, normalized_objdir
)) {
418 strbuf_release(&pathbuf
);
422 ent
= alloc_alt_odb(pathbuf
.buf
);
424 /* add the alternate entry */
426 alt_odb_tail
= &(ent
->next
);
429 /* recursively add alternates */
430 read_info_alternates(pathbuf
.buf
, depth
+ 1);
432 strbuf_release(&pathbuf
);
436 static const char *parse_alt_odb_entry(const char *string
,
444 if (*string
== '#') {
445 /* comment; consume up to next separator */
446 end
= strchrnul(string
, sep
);
447 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
449 * quoted path; unquote_c_style has copied the
450 * data for us and set "end". Broken quoting (e.g.,
451 * an entry that doesn't end with a quote) falls
452 * back to the unquoted case below.
455 /* normal, unquoted path */
456 end
= strchrnul(string
, sep
);
457 strbuf_add(out
, string
, end
- string
);
465 static void link_alt_odb_entries(const char *alt
, int sep
,
466 const char *relative_base
, int depth
)
468 struct strbuf objdirbuf
= STRBUF_INIT
;
469 struct strbuf entry
= STRBUF_INIT
;
475 error("%s: ignoring alternate object stores, nesting too deep.",
480 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
481 if (strbuf_normalize_path(&objdirbuf
) < 0)
482 die("unable to normalize object directory: %s",
486 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
489 link_alt_odb_entry(entry
.buf
, relative_base
, depth
, objdirbuf
.buf
);
491 strbuf_release(&entry
);
492 strbuf_release(&objdirbuf
);
495 static void read_info_alternates(const char * relative_base
, int depth
)
498 struct strbuf buf
= STRBUF_INIT
;
500 path
= xstrfmt("%s/info/alternates", relative_base
);
501 if (strbuf_read_file(&buf
, path
, 1024) < 0) {
502 warn_on_fopen_errors(path
);
507 link_alt_odb_entries(buf
.buf
, '\n', relative_base
, depth
);
508 strbuf_release(&buf
);
512 struct alternate_object_database
*alloc_alt_odb(const char *dir
)
514 struct alternate_object_database
*ent
;
516 FLEX_ALLOC_STR(ent
, path
, dir
);
517 strbuf_init(&ent
->scratch
, 0);
518 strbuf_addf(&ent
->scratch
, "%s/", dir
);
519 ent
->base_len
= ent
->scratch
.len
;
524 void add_to_alternates_file(const char *reference
)
526 struct lock_file lock
= LOCK_INIT
;
527 char *alts
= git_pathdup("objects/info/alternates");
531 hold_lock_file_for_update(&lock
, alts
, LOCK_DIE_ON_ERROR
);
532 out
= fdopen_lock_file(&lock
, "w");
534 die_errno("unable to fdopen alternates lockfile");
536 in
= fopen(alts
, "r");
538 struct strbuf line
= STRBUF_INIT
;
540 while (strbuf_getline(&line
, in
) != EOF
) {
541 if (!strcmp(reference
, line
.buf
)) {
545 fprintf_or_die(out
, "%s\n", line
.buf
);
548 strbuf_release(&line
);
551 else if (errno
!= ENOENT
)
552 die_errno("unable to read alternates file");
555 rollback_lock_file(&lock
);
557 fprintf_or_die(out
, "%s\n", reference
);
558 if (commit_lock_file(&lock
))
559 die_errno("unable to move new alternates file into place");
561 link_alt_odb_entries(reference
, '\n', NULL
, 0);
566 void add_to_alternates_memory(const char *reference
)
569 * Make sure alternates are initialized, or else our entry may be
570 * overwritten when they are.
574 link_alt_odb_entries(reference
, '\n', NULL
, 0);
578 * Compute the exact path an alternate is at and returns it. In case of
579 * error NULL is returned and the human readable error is added to `err`
580 * `path` may be relative and should point to $GITDIR.
581 * `err` must not be null.
583 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
585 char *ref_git
= NULL
;
586 const char *repo
, *ref_git_s
;
589 ref_git_s
= real_path_if_valid(path
);
592 strbuf_addf(err
, _("path '%s' does not exist"), path
);
596 * Beware: read_gitfile(), real_path() and mkpath()
597 * return static buffer
599 ref_git
= xstrdup(ref_git_s
);
601 repo
= read_gitfile(ref_git
);
603 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
606 ref_git
= xstrdup(repo
);
609 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
610 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
612 ref_git
= ref_git_git
;
613 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
614 struct strbuf sb
= STRBUF_INIT
;
616 if (get_common_dir(&sb
, ref_git
)) {
618 _("reference repository '%s' as a linked "
619 "checkout is not supported yet."),
624 strbuf_addf(err
, _("reference repository '%s' is not a "
625 "local repository."), path
);
629 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
630 strbuf_addf(err
, _("reference repository '%s' is shallow"),
636 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
638 _("reference repository '%s' is grafted"),
646 FREE_AND_NULL(ref_git
);
652 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
654 struct alternate_object_database
*ent
;
658 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
666 void prepare_alt_odb(void)
673 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
675 alt_odb_tail
= &alt_odb_list
;
676 link_alt_odb_entries(alt
, PATH_SEP
, NULL
, 0);
678 read_info_alternates(get_object_directory(), 0);
681 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
682 static int freshen_file(const char *fn
)
685 t
.actime
= t
.modtime
= time(NULL
);
686 return !utime(fn
, &t
);
690 * All of the check_and_freshen functions return 1 if the file exists and was
691 * freshened (if freshening was requested), 0 otherwise. If they return
692 * 0, you should not assume that it is safe to skip a write of the object (it
693 * either does not exist on disk, or has a stale mtime and may be subject to
696 int check_and_freshen_file(const char *fn
, int freshen
)
698 if (access(fn
, F_OK
))
700 if (freshen
&& !freshen_file(fn
))
705 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
707 static struct strbuf buf
= STRBUF_INIT
;
710 sha1_file_name(&buf
, sha1
);
712 return check_and_freshen_file(buf
.buf
, freshen
);
715 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
717 struct alternate_object_database
*alt
;
719 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
720 const char *path
= alt_sha1_path(alt
, sha1
);
721 if (check_and_freshen_file(path
, freshen
))
727 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
729 return check_and_freshen_local(sha1
, freshen
) ||
730 check_and_freshen_nonlocal(sha1
, freshen
);
733 int has_loose_object_nonlocal(const unsigned char *sha1
)
735 return check_and_freshen_nonlocal(sha1
, 0);
738 static int has_loose_object(const unsigned char *sha1
)
740 return check_and_freshen(sha1
, 0);
743 static void mmap_limit_check(size_t length
)
745 static size_t limit
= 0;
747 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
752 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
753 (uintmax_t)length
, (uintmax_t)limit
);
756 void *xmmap_gently(void *start
, size_t length
,
757 int prot
, int flags
, int fd
, off_t offset
)
761 mmap_limit_check(length
);
762 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
763 if (ret
== MAP_FAILED
) {
766 release_pack_memory(length
);
767 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
772 void *xmmap(void *start
, size_t length
,
773 int prot
, int flags
, int fd
, off_t offset
)
775 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
776 if (ret
== MAP_FAILED
)
777 die_errno("mmap failed");
782 * With an in-core object data in "map", rehash it to make sure the
783 * object name actually matches "sha1" to detect object corruption.
784 * With "map" == NULL, try reading the object named with "sha1" using
785 * the streaming interface and rehash it to do the same.
787 int check_sha1_signature(const unsigned char *sha1
, void *map
,
788 unsigned long size
, const char *type
)
790 struct object_id real_oid
;
791 enum object_type obj_type
;
792 struct git_istream
*st
;
798 hash_object_file(map
, size
, type
, &real_oid
);
799 return hashcmp(sha1
, real_oid
.hash
) ? -1 : 0;
802 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
806 /* Generate the header */
807 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", type_name(obj_type
), size
) + 1;
810 the_hash_algo
->init_fn(&c
);
811 the_hash_algo
->update_fn(&c
, hdr
, hdrlen
);
814 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
822 the_hash_algo
->update_fn(&c
, buf
, readlen
);
824 the_hash_algo
->final_fn(real_oid
.hash
, &c
);
826 return hashcmp(sha1
, real_oid
.hash
) ? -1 : 0;
829 int git_open_cloexec(const char *name
, int flags
)
832 static int o_cloexec
= O_CLOEXEC
;
834 fd
= open(name
, flags
| o_cloexec
);
835 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
836 /* Try again w/o O_CLOEXEC: the kernel might not support it */
837 o_cloexec
&= ~O_CLOEXEC
;
838 fd
= open(name
, flags
| o_cloexec
);
841 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
843 static int fd_cloexec
= FD_CLOEXEC
;
845 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
846 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
847 int flags
= fcntl(fd
, F_GETFD
);
848 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
857 * Find "sha1" as a loose object in the local repository or in an alternate.
858 * Returns 0 on success, negative on failure.
860 * The "path" out-parameter will give the path of the object we found (if any).
861 * Note that it may point to static storage and is only valid until another
862 * call to sha1_file_name(), etc.
864 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
,
867 struct alternate_object_database
*alt
;
868 static struct strbuf buf
= STRBUF_INIT
;
871 sha1_file_name(&buf
, sha1
);
874 if (!lstat(*path
, st
))
879 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
880 *path
= alt_sha1_path(alt
, sha1
);
881 if (!lstat(*path
, st
))
889 * Like stat_sha1_file(), but actually open the object and return the
890 * descriptor. See the caveats on the "path" parameter above.
892 static int open_sha1_file(const unsigned char *sha1
, const char **path
)
895 struct alternate_object_database
*alt
;
896 int most_interesting_errno
;
897 static struct strbuf buf
= STRBUF_INIT
;
900 sha1_file_name(&buf
, sha1
);
903 fd
= git_open(*path
);
906 most_interesting_errno
= errno
;
909 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
910 *path
= alt_sha1_path(alt
, sha1
);
911 fd
= git_open(*path
);
914 if (most_interesting_errno
== ENOENT
)
915 most_interesting_errno
= errno
;
917 errno
= most_interesting_errno
;
922 * Map the loose object at "path" if it is not NULL, or the path found by
923 * searching for a loose object named "sha1".
925 static void *map_sha1_file_1(const char *path
,
926 const unsigned char *sha1
,
935 fd
= open_sha1_file(sha1
, &path
);
940 if (!fstat(fd
, &st
)) {
941 *size
= xsize_t(st
.st_size
);
943 /* mmap() is forbidden on empty files */
944 error("object file %s is empty", path
);
947 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
954 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
956 return map_sha1_file_1(NULL
, sha1
, size
);
959 static int unpack_sha1_short_header(git_zstream
*stream
,
960 unsigned char *map
, unsigned long mapsize
,
961 void *buffer
, unsigned long bufsiz
)
963 /* Get the data stream */
964 memset(stream
, 0, sizeof(*stream
));
965 stream
->next_in
= map
;
966 stream
->avail_in
= mapsize
;
967 stream
->next_out
= buffer
;
968 stream
->avail_out
= bufsiz
;
970 git_inflate_init(stream
);
971 return git_inflate(stream
, 0);
974 int unpack_sha1_header(git_zstream
*stream
,
975 unsigned char *map
, unsigned long mapsize
,
976 void *buffer
, unsigned long bufsiz
)
978 int status
= unpack_sha1_short_header(stream
, map
, mapsize
,
984 /* Make sure we have the terminating NUL */
985 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
990 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
991 unsigned long mapsize
, void *buffer
,
992 unsigned long bufsiz
, struct strbuf
*header
)
996 status
= unpack_sha1_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1001 * Check if entire header is unpacked in the first iteration.
1003 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1007 * buffer[0..bufsiz] was not large enough. Copy the partial
1008 * result out to header, and then append the result of further
1009 * reading the stream.
1011 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1012 stream
->next_out
= buffer
;
1013 stream
->avail_out
= bufsiz
;
1016 status
= git_inflate(stream
, 0);
1017 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1018 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1020 stream
->next_out
= buffer
;
1021 stream
->avail_out
= bufsiz
;
1022 } while (status
!= Z_STREAM_END
);
1026 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1028 int bytes
= strlen(buffer
) + 1;
1029 unsigned char *buf
= xmallocz(size
);
1033 n
= stream
->total_out
- bytes
;
1036 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1038 if (bytes
<= size
) {
1040 * The above condition must be (bytes <= size), not
1041 * (bytes < size). In other words, even though we
1042 * expect no more output and set avail_out to zero,
1043 * the input zlib stream may have bytes that express
1044 * "this concludes the stream", and we *do* want to
1047 * Otherwise we would not be able to test that we
1048 * consumed all the input to reach the expected size;
1049 * we also want to check that zlib tells us that all
1050 * went well with status == Z_STREAM_END at the end.
1052 stream
->next_out
= buf
+ bytes
;
1053 stream
->avail_out
= size
- bytes
;
1054 while (status
== Z_OK
)
1055 status
= git_inflate(stream
, Z_FINISH
);
1057 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1058 git_inflate_end(stream
);
1063 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1064 else if (stream
->avail_in
)
1065 error("garbage at end of loose object '%s'",
1072 * We used to just use "sscanf()", but that's actually way
1073 * too permissive for what we want to check. So do an anal
1074 * object header parse by hand.
1076 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1079 const char *type_buf
= hdr
;
1081 int type
, type_len
= 0;
1084 * The type can be of any size but is followed by
1096 type
= type_from_string_gently(type_buf
, type_len
, 1);
1098 strbuf_add(oi
->type_name
, type_buf
, type_len
);
1100 * Set type to 0 if its an unknown object and
1101 * we're obtaining the type using '--allow-unknown-type'
1104 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
) && (type
< 0))
1107 die("invalid object type");
1112 * The length must follow immediately, and be in canonical
1113 * decimal format (ie "010" is not valid).
1115 size
= *hdr
++ - '0';
1120 unsigned long c
= *hdr
- '0';
1124 size
= size
* 10 + c
;
1132 * The length must be followed by a zero byte
1134 return *hdr
? -1 : type
;
1137 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1139 struct object_info oi
= OBJECT_INFO_INIT
;
1142 return parse_sha1_header_extended(hdr
, &oi
, 0);
1145 static int sha1_loose_object_info(const unsigned char *sha1
,
1146 struct object_info
*oi
,
1150 unsigned long mapsize
;
1154 struct strbuf hdrbuf
= STRBUF_INIT
;
1155 unsigned long size_scratch
;
1157 if (oi
->delta_base_sha1
)
1158 hashclr(oi
->delta_base_sha1
);
1161 * If we don't care about type or size, then we don't
1162 * need to look inside the object at all. Note that we
1163 * do not optimize out the stat call, even if the
1164 * caller doesn't care about the disk-size, since our
1165 * return value implicitly indicates whether the
1166 * object even exists.
1168 if (!oi
->typep
&& !oi
->type_name
&& !oi
->sizep
&& !oi
->contentp
) {
1171 if (stat_sha1_file(sha1
, &st
, &path
) < 0)
1174 *oi
->disk_sizep
= st
.st_size
;
1178 map
= map_sha1_file(sha1
, &mapsize
);
1183 oi
->sizep
= &size_scratch
;
1186 *oi
->disk_sizep
= mapsize
;
1187 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
)) {
1188 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
1189 status
= error("unable to unpack %s header with --allow-unknown-type",
1191 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1192 status
= error("unable to unpack %s header",
1196 else if (hdrbuf
.len
) {
1197 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
1198 status
= error("unable to parse %s header with --allow-unknown-type",
1200 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
1201 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1203 if (status
>= 0 && oi
->contentp
) {
1204 *oi
->contentp
= unpack_sha1_rest(&stream
, hdr
,
1206 if (!*oi
->contentp
) {
1207 git_inflate_end(&stream
);
1211 git_inflate_end(&stream
);
1213 munmap(map
, mapsize
);
1214 if (status
&& oi
->typep
)
1215 *oi
->typep
= status
;
1216 if (oi
->sizep
== &size_scratch
)
1218 strbuf_release(&hdrbuf
);
1219 oi
->whence
= OI_LOOSE
;
1220 return (status
< 0) ? status
: 0;
1223 int fetch_if_missing
= 1;
1225 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
1227 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
1228 struct pack_entry e
;
1230 const unsigned char *real
= (flags
& OBJECT_INFO_LOOKUP_REPLACE
) ?
1231 lookup_replace_object(sha1
) :
1233 int already_retried
= 0;
1235 if (is_null_sha1(real
))
1241 if (!(flags
& OBJECT_INFO_SKIP_CACHED
)) {
1242 struct cached_object
*co
= find_cached_object(real
);
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
, &e
))
1265 /* Most likely it's a loose object. */
1266 if (!sha1_loose_object_info(real
, oi
, flags
))
1269 /* Not a loose object; someone else may have just packed it. */
1270 reprepare_packed_git();
1271 if (find_pack_entry(real
, &e
))
1274 /* Check if it is a missing object */
1275 if (fetch_if_missing
&& repository_format_partial_clone
&&
1278 * TODO Investigate haveing fetch_object() return
1279 * TODO error/success and stopping the music here.
1281 fetch_object(repository_format_partial_clone
, real
);
1282 already_retried
= 1;
1289 if (oi
== &blank_oi
)
1291 * We know that the caller doesn't actually need the
1292 * information below, so return early.
1295 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
1297 mark_bad_packed_object(e
.p
, real
);
1298 return sha1_object_info_extended(real
, oi
, 0);
1299 } else if (oi
->whence
== OI_PACKED
) {
1300 oi
->u
.packed
.offset
= e
.offset
;
1301 oi
->u
.packed
.pack
= e
.p
;
1302 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
1303 rtype
== OBJ_OFS_DELTA
);
1309 /* returns enum object_type or negative */
1310 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1312 enum object_type type
;
1313 struct object_info oi
= OBJECT_INFO_INIT
;
1317 if (sha1_object_info_extended(sha1
, &oi
,
1318 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
1323 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1324 unsigned long *size
)
1326 struct object_info oi
= OBJECT_INFO_INIT
;
1330 oi
.contentp
= &content
;
1332 if (sha1_object_info_extended(sha1
, &oi
, 0) < 0)
1337 int pretend_object_file(void *buf
, unsigned long len
, enum object_type type
,
1338 struct object_id
*oid
)
1340 struct cached_object
*co
;
1342 hash_object_file(buf
, len
, type_name(type
), oid
);
1343 if (has_sha1_file(oid
->hash
) || find_cached_object(oid
->hash
))
1345 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
1346 co
= &cached_objects
[cached_object_nr
++];
1349 co
->buf
= xmalloc(len
);
1350 memcpy(co
->buf
, buf
, len
);
1351 hashcpy(co
->sha1
, oid
->hash
);
1356 * This function dies on corrupt objects; the callers who want to
1357 * deal with them should arrange to call read_object() and give error
1358 * messages themselves.
1360 void *read_sha1_file_extended(const unsigned char *sha1
,
1361 enum object_type
*type
,
1362 unsigned long *size
,
1366 const struct packed_git
*p
;
1369 const unsigned char *repl
= lookup_replace
? lookup_replace_object(sha1
)
1373 data
= read_object(repl
, type
, size
);
1377 if (errno
&& errno
!= ENOENT
)
1378 die_errno("failed to read object %s", sha1_to_hex(sha1
));
1380 /* die if we replaced an object with one that does not exist */
1382 die("replacement %s not found for %s",
1383 sha1_to_hex(repl
), sha1_to_hex(sha1
));
1385 if (!stat_sha1_file(repl
, &st
, &path
))
1386 die("loose object %s (stored in %s) is corrupt",
1387 sha1_to_hex(repl
), path
);
1389 if ((p
= has_packed_and_bad(repl
)) != NULL
)
1390 die("packed object %s (stored in %s) is corrupt",
1391 sha1_to_hex(repl
), p
->pack_name
);
1396 void *read_object_with_reference(const unsigned char *sha1
,
1397 const char *required_type_name
,
1398 unsigned long *size
,
1399 unsigned char *actual_sha1_return
)
1401 enum object_type type
, required_type
;
1403 unsigned long isize
;
1404 unsigned char actual_sha1
[20];
1406 required_type
= type_from_string(required_type_name
);
1407 hashcpy(actual_sha1
, sha1
);
1409 int ref_length
= -1;
1410 const char *ref_type
= NULL
;
1412 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1415 if (type
== required_type
) {
1417 if (actual_sha1_return
)
1418 hashcpy(actual_sha1_return
, actual_sha1
);
1421 /* Handle references */
1422 else if (type
== OBJ_COMMIT
)
1424 else if (type
== OBJ_TAG
)
1425 ref_type
= "object ";
1430 ref_length
= strlen(ref_type
);
1432 if (ref_length
+ 40 > isize
||
1433 memcmp(buffer
, ref_type
, ref_length
) ||
1434 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1439 /* Now we have the ID of the referred-to object in
1440 * actual_sha1. Check again. */
1444 static void write_object_file_prepare(const void *buf
, unsigned long len
,
1445 const char *type
, struct object_id
*oid
,
1446 char *hdr
, int *hdrlen
)
1450 /* Generate the header */
1451 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
1454 the_hash_algo
->init_fn(&c
);
1455 the_hash_algo
->update_fn(&c
, hdr
, *hdrlen
);
1456 the_hash_algo
->update_fn(&c
, buf
, len
);
1457 the_hash_algo
->final_fn(oid
->hash
, &c
);
1461 * Move the just written object into its final resting place.
1463 int finalize_object_file(const char *tmpfile
, const char *filename
)
1467 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
1469 else if (link(tmpfile
, filename
))
1473 * Coda hack - coda doesn't like cross-directory links,
1474 * so we fall back to a rename, which will mean that it
1475 * won't be able to check collisions, but that's not a
1478 * The same holds for FAT formatted media.
1480 * When this succeeds, we just return. We have nothing
1483 if (ret
&& ret
!= EEXIST
) {
1485 if (!rename(tmpfile
, filename
))
1489 unlink_or_warn(tmpfile
);
1491 if (ret
!= EEXIST
) {
1492 return error_errno("unable to write sha1 filename %s", filename
);
1494 /* FIXME!!! Collision check here ? */
1498 if (adjust_shared_perm(filename
))
1499 return error("unable to set permission to '%s'", filename
);
1503 static int write_buffer(int fd
, const void *buf
, size_t len
)
1505 if (write_in_full(fd
, buf
, len
) < 0)
1506 return error_errno("file write error");
1510 int hash_object_file(const void *buf
, unsigned long len
, const char *type
,
1511 struct object_id
*oid
)
1514 int hdrlen
= sizeof(hdr
);
1515 write_object_file_prepare(buf
, len
, type
, oid
, hdr
, &hdrlen
);
1519 /* Finalize a file on disk, and close it. */
1520 static void close_sha1_file(int fd
)
1522 if (fsync_object_files
)
1523 fsync_or_die(fd
, "sha1 file");
1525 die_errno("error when closing sha1 file");
1528 /* Size of directory component, including the ending '/' */
1529 static inline int directory_size(const char *filename
)
1531 const char *s
= strrchr(filename
, '/');
1534 return s
- filename
+ 1;
1538 * This creates a temporary file in the same directory as the final
1541 * We want to avoid cross-directory filename renames, because those
1542 * can have problems on various filesystems (FAT, NFS, Coda).
1544 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
1546 int fd
, dirlen
= directory_size(filename
);
1549 strbuf_add(tmp
, filename
, dirlen
);
1550 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
1551 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1552 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
1554 * Make sure the directory exists; note that the contents
1555 * of the buffer are undefined after mkstemp returns an
1556 * error, so we have to rewrite the whole buffer from
1560 strbuf_add(tmp
, filename
, dirlen
- 1);
1561 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
1563 if (adjust_shared_perm(tmp
->buf
))
1567 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
1568 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1573 static int write_loose_object(const struct object_id
*oid
, char *hdr
,
1574 int hdrlen
, const void *buf
, unsigned long len
,
1578 unsigned char compressed
[4096];
1581 struct object_id parano_oid
;
1582 static struct strbuf tmp_file
= STRBUF_INIT
;
1583 static struct strbuf filename
= STRBUF_INIT
;
1585 strbuf_reset(&filename
);
1586 sha1_file_name(&filename
, oid
->hash
);
1588 fd
= create_tmpfile(&tmp_file
, filename
.buf
);
1590 if (errno
== EACCES
)
1591 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1593 return error_errno("unable to create temporary file");
1597 git_deflate_init(&stream
, zlib_compression_level
);
1598 stream
.next_out
= compressed
;
1599 stream
.avail_out
= sizeof(compressed
);
1600 the_hash_algo
->init_fn(&c
);
1602 /* First header.. */
1603 stream
.next_in
= (unsigned char *)hdr
;
1604 stream
.avail_in
= hdrlen
;
1605 while (git_deflate(&stream
, 0) == Z_OK
)
1607 the_hash_algo
->update_fn(&c
, hdr
, hdrlen
);
1609 /* Then the data itself.. */
1610 stream
.next_in
= (void *)buf
;
1611 stream
.avail_in
= len
;
1613 unsigned char *in0
= stream
.next_in
;
1614 ret
= git_deflate(&stream
, Z_FINISH
);
1615 the_hash_algo
->update_fn(&c
, in0
, stream
.next_in
- in0
);
1616 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
1617 die("unable to write sha1 file");
1618 stream
.next_out
= compressed
;
1619 stream
.avail_out
= sizeof(compressed
);
1620 } while (ret
== Z_OK
);
1622 if (ret
!= Z_STREAM_END
)
1623 die("unable to deflate new object %s (%d)", oid_to_hex(oid
),
1625 ret
= git_deflate_end_gently(&stream
);
1627 die("deflateEnd on object %s failed (%d)", oid_to_hex(oid
),
1629 the_hash_algo
->final_fn(parano_oid
.hash
, &c
);
1630 if (oidcmp(oid
, ¶no_oid
) != 0)
1631 die("confused by unstable object source data for %s",
1634 close_sha1_file(fd
);
1639 utb
.modtime
= mtime
;
1640 if (utime(tmp_file
.buf
, &utb
) < 0)
1641 warning_errno("failed utime() on %s", tmp_file
.buf
);
1644 return finalize_object_file(tmp_file
.buf
, filename
.buf
);
1647 static int freshen_loose_object(const unsigned char *sha1
)
1649 return check_and_freshen(sha1
, 1);
1652 static int freshen_packed_object(const unsigned char *sha1
)
1654 struct pack_entry e
;
1655 if (!find_pack_entry(sha1
, &e
))
1659 if (!freshen_file(e
.p
->pack_name
))
1665 int write_object_file(const void *buf
, unsigned long len
, const char *type
,
1666 struct object_id
*oid
)
1669 int hdrlen
= sizeof(hdr
);
1671 /* Normally if we have it in the pack then we do not bother writing
1672 * it out into .git/objects/??/?{38} file.
1674 write_object_file_prepare(buf
, len
, type
, oid
, hdr
, &hdrlen
);
1675 if (freshen_packed_object(oid
->hash
) || freshen_loose_object(oid
->hash
))
1677 return write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, 0);
1680 int hash_object_file_literally(const void *buf
, unsigned long len
,
1681 const char *type
, struct object_id
*oid
,
1685 int hdrlen
, status
= 0;
1687 /* type string, SP, %lu of the length plus NUL must fit this */
1688 hdrlen
= strlen(type
) + 32;
1689 header
= xmalloc(hdrlen
);
1690 write_object_file_prepare(buf
, len
, type
, oid
, header
, &hdrlen
);
1692 if (!(flags
& HASH_WRITE_OBJECT
))
1694 if (freshen_packed_object(oid
->hash
) || freshen_loose_object(oid
->hash
))
1696 status
= write_loose_object(oid
, header
, hdrlen
, buf
, len
, 0);
1703 int force_object_loose(const struct object_id
*oid
, time_t mtime
)
1707 enum object_type type
;
1712 if (has_loose_object(oid
->hash
))
1714 buf
= read_object(oid
->hash
, &type
, &len
);
1716 return error("cannot read sha1_file for %s", oid_to_hex(oid
));
1717 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", type_name(type
), len
) + 1;
1718 ret
= write_loose_object(oid
, hdr
, hdrlen
, buf
, len
, mtime
);
1724 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
1726 if (!startup_info
->have_repository
)
1728 return sha1_object_info_extended(sha1
, NULL
,
1729 flags
| OBJECT_INFO_SKIP_CACHED
) >= 0;
1732 int has_object_file(const struct object_id
*oid
)
1734 return has_sha1_file(oid
->hash
);
1737 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
1739 return has_sha1_file_with_flags(oid
->hash
, flags
);
1742 static void check_tree(const void *buf
, size_t size
)
1744 struct tree_desc desc
;
1745 struct name_entry entry
;
1747 init_tree_desc(&desc
, buf
, size
);
1748 while (tree_entry(&desc
, &entry
))
1750 * tree_entry() will die() on malformed entries */
1754 static void check_commit(const void *buf
, size_t size
)
1757 memset(&c
, 0, sizeof(c
));
1758 if (parse_commit_buffer(&c
, buf
, size
))
1759 die("corrupt commit");
1762 static void check_tag(const void *buf
, size_t size
)
1765 memset(&t
, 0, sizeof(t
));
1766 if (parse_tag_buffer(&t
, buf
, size
))
1770 static int index_mem(struct object_id
*oid
, void *buf
, size_t size
,
1771 enum object_type type
,
1772 const char *path
, unsigned flags
)
1774 int ret
, re_allocated
= 0;
1775 int write_object
= flags
& HASH_WRITE_OBJECT
;
1781 * Convert blobs to git internal format
1783 if ((type
== OBJ_BLOB
) && path
) {
1784 struct strbuf nbuf
= STRBUF_INIT
;
1785 if (convert_to_git(&the_index
, path
, buf
, size
, &nbuf
,
1786 get_conv_flags(flags
))) {
1787 buf
= strbuf_detach(&nbuf
, &size
);
1791 if (flags
& HASH_FORMAT_CHECK
) {
1792 if (type
== OBJ_TREE
)
1793 check_tree(buf
, size
);
1794 if (type
== OBJ_COMMIT
)
1795 check_commit(buf
, size
);
1796 if (type
== OBJ_TAG
)
1797 check_tag(buf
, size
);
1801 ret
= write_object_file(buf
, size
, type_name(type
), oid
);
1803 ret
= hash_object_file(buf
, size
, type_name(type
), oid
);
1809 static int index_stream_convert_blob(struct object_id
*oid
, int fd
,
1810 const char *path
, unsigned flags
)
1813 const int write_object
= flags
& HASH_WRITE_OBJECT
;
1814 struct strbuf sbuf
= STRBUF_INIT
;
1817 assert(would_convert_to_git_filter_fd(path
));
1819 convert_to_git_filter_fd(&the_index
, path
, fd
, &sbuf
,
1820 get_conv_flags(flags
));
1823 ret
= write_object_file(sbuf
.buf
, sbuf
.len
, type_name(OBJ_BLOB
),
1826 ret
= hash_object_file(sbuf
.buf
, sbuf
.len
, type_name(OBJ_BLOB
),
1828 strbuf_release(&sbuf
);
1832 static int index_pipe(struct object_id
*oid
, int fd
, enum object_type type
,
1833 const char *path
, unsigned flags
)
1835 struct strbuf sbuf
= STRBUF_INIT
;
1838 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
1839 ret
= index_mem(oid
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
1842 strbuf_release(&sbuf
);
1846 #define SMALL_FILE_SIZE (32*1024)
1848 static int index_core(struct object_id
*oid
, int fd
, size_t size
,
1849 enum object_type type
, const char *path
,
1855 ret
= index_mem(oid
, "", size
, type
, path
, flags
);
1856 } else if (size
<= SMALL_FILE_SIZE
) {
1857 char *buf
= xmalloc(size
);
1858 ssize_t read_result
= read_in_full(fd
, buf
, size
);
1859 if (read_result
< 0)
1860 ret
= error_errno("read error while indexing %s",
1861 path
? path
: "<unknown>");
1862 else if (read_result
!= size
)
1863 ret
= error("short read while indexing %s",
1864 path
? path
: "<unknown>");
1866 ret
= index_mem(oid
, buf
, size
, type
, path
, flags
);
1869 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1870 ret
= index_mem(oid
, buf
, size
, type
, path
, flags
);
1877 * This creates one packfile per large blob unless bulk-checkin
1878 * machinery is "plugged".
1880 * This also bypasses the usual "convert-to-git" dance, and that is on
1881 * purpose. We could write a streaming version of the converting
1882 * functions and insert that before feeding the data to fast-import
1883 * (or equivalent in-core API described above). However, that is
1884 * somewhat complicated, as we do not know the size of the filter
1885 * result, which we need to know beforehand when writing a git object.
1886 * Since the primary motivation for trying to stream from the working
1887 * tree file and to avoid mmaping it in core is to deal with large
1888 * binary blobs, they generally do not want to get any conversion, and
1889 * callers should avoid this code path when filters are requested.
1891 static int index_stream(struct object_id
*oid
, int fd
, size_t size
,
1892 enum object_type type
, const char *path
,
1895 return index_bulk_checkin(oid
->hash
, fd
, size
, type
, path
, flags
);
1898 int index_fd(struct object_id
*oid
, int fd
, struct stat
*st
,
1899 enum object_type type
, const char *path
, unsigned flags
)
1904 * Call xsize_t() only when needed to avoid potentially unnecessary
1905 * die() for large files.
1907 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
1908 ret
= index_stream_convert_blob(oid
, fd
, path
, flags
);
1909 else if (!S_ISREG(st
->st_mode
))
1910 ret
= index_pipe(oid
, fd
, type
, path
, flags
);
1911 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
1912 (path
&& would_convert_to_git(&the_index
, path
)))
1913 ret
= index_core(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
1916 ret
= index_stream(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
1922 int index_path(struct object_id
*oid
, const char *path
, struct stat
*st
, unsigned flags
)
1925 struct strbuf sb
= STRBUF_INIT
;
1928 switch (st
->st_mode
& S_IFMT
) {
1930 fd
= open(path
, O_RDONLY
);
1932 return error_errno("open(\"%s\")", path
);
1933 if (index_fd(oid
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
1934 return error("%s: failed to insert into database",
1938 if (strbuf_readlink(&sb
, path
, st
->st_size
))
1939 return error_errno("readlink(\"%s\")", path
);
1940 if (!(flags
& HASH_WRITE_OBJECT
))
1941 hash_object_file(sb
.buf
, sb
.len
, blob_type
, oid
);
1942 else if (write_object_file(sb
.buf
, sb
.len
, blob_type
, oid
))
1943 rc
= error("%s: failed to insert into database", path
);
1944 strbuf_release(&sb
);
1947 return resolve_gitlink_ref(path
, "HEAD", oid
);
1949 return error("%s: unsupported file type", path
);
1954 int read_pack_header(int fd
, struct pack_header
*header
)
1956 if (read_in_full(fd
, header
, sizeof(*header
)) != sizeof(*header
))
1957 /* "eof before pack header was fully read" */
1958 return PH_ERROR_EOF
;
1960 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
1961 /* "protocol error (pack signature mismatch detected)" */
1962 return PH_ERROR_PACK_SIGNATURE
;
1963 if (!pack_version_ok(header
->hdr_version
))
1964 /* "protocol error (pack version unsupported)" */
1965 return PH_ERROR_PROTOCOL
;
1969 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
1971 enum object_type type
= sha1_object_info(sha1
, NULL
);
1973 die("%s is not a valid object", sha1_to_hex(sha1
));
1975 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
1979 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
1980 struct strbuf
*path
,
1981 each_loose_object_fn obj_cb
,
1982 each_loose_cruft_fn cruft_cb
,
1983 each_loose_subdir_fn subdir_cb
,
1986 size_t origlen
, baselen
;
1990 struct object_id oid
;
1992 if (subdir_nr
> 0xff)
1993 BUG("invalid loose object subdirectory: %x", subdir_nr
);
1995 origlen
= path
->len
;
1996 strbuf_complete(path
, '/');
1997 strbuf_addf(path
, "%02x", subdir_nr
);
1999 dir
= opendir(path
->buf
);
2001 if (errno
!= ENOENT
)
2002 r
= error_errno("unable to open %s", path
->buf
);
2003 strbuf_setlen(path
, origlen
);
2007 oid
.hash
[0] = subdir_nr
;
2008 strbuf_addch(path
, '/');
2009 baselen
= path
->len
;
2011 while ((de
= readdir(dir
))) {
2013 if (is_dot_or_dotdot(de
->d_name
))
2016 namelen
= strlen(de
->d_name
);
2017 strbuf_setlen(path
, baselen
);
2018 strbuf_add(path
, de
->d_name
, namelen
);
2019 if (namelen
== GIT_SHA1_HEXSZ
- 2 &&
2020 !hex_to_bytes(oid
.hash
+ 1, de
->d_name
,
2021 GIT_SHA1_RAWSZ
- 1)) {
2023 r
= obj_cb(&oid
, path
->buf
, data
);
2031 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
2038 strbuf_setlen(path
, baselen
- 1);
2039 if (!r
&& subdir_cb
)
2040 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
2042 strbuf_setlen(path
, origlen
);
2047 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
2048 each_loose_object_fn obj_cb
,
2049 each_loose_cruft_fn cruft_cb
,
2050 each_loose_subdir_fn subdir_cb
,
2056 for (i
= 0; i
< 256; i
++) {
2057 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
2066 int for_each_loose_file_in_objdir(const char *path
,
2067 each_loose_object_fn obj_cb
,
2068 each_loose_cruft_fn cruft_cb
,
2069 each_loose_subdir_fn subdir_cb
,
2072 struct strbuf buf
= STRBUF_INIT
;
2075 strbuf_addstr(&buf
, path
);
2076 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
2078 strbuf_release(&buf
);
2083 struct loose_alt_odb_data
{
2084 each_loose_object_fn
*cb
;
2088 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
2091 struct loose_alt_odb_data
*data
= vdata
;
2092 struct strbuf buf
= STRBUF_INIT
;
2095 strbuf_addstr(&buf
, alt
->path
);
2096 r
= for_each_loose_file_in_objdir_buf(&buf
,
2097 data
->cb
, NULL
, NULL
,
2099 strbuf_release(&buf
);
2103 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
2105 struct loose_alt_odb_data alt
;
2108 r
= for_each_loose_file_in_objdir(get_object_directory(),
2109 cb
, NULL
, NULL
, data
);
2113 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
2118 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
2121 static int check_stream_sha1(git_zstream
*stream
,
2125 const unsigned char *expected_sha1
)
2128 unsigned char real_sha1
[GIT_MAX_RAWSZ
];
2129 unsigned char buf
[4096];
2130 unsigned long total_read
;
2133 the_hash_algo
->init_fn(&c
);
2134 the_hash_algo
->update_fn(&c
, hdr
, stream
->total_out
);
2137 * We already read some bytes into hdr, but the ones up to the NUL
2138 * do not count against the object's content size.
2140 total_read
= stream
->total_out
- strlen(hdr
) - 1;
2143 * This size comparison must be "<=" to read the final zlib packets;
2144 * see the comment in unpack_sha1_rest for details.
2146 while (total_read
<= size
&&
2147 (status
== Z_OK
|| status
== Z_BUF_ERROR
)) {
2148 stream
->next_out
= buf
;
2149 stream
->avail_out
= sizeof(buf
);
2150 if (size
- total_read
< stream
->avail_out
)
2151 stream
->avail_out
= size
- total_read
;
2152 status
= git_inflate(stream
, Z_FINISH
);
2153 the_hash_algo
->update_fn(&c
, buf
, stream
->next_out
- buf
);
2154 total_read
+= stream
->next_out
- buf
;
2156 git_inflate_end(stream
);
2158 if (status
!= Z_STREAM_END
) {
2159 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1
));
2162 if (stream
->avail_in
) {
2163 error("garbage at end of loose object '%s'",
2164 sha1_to_hex(expected_sha1
));
2168 the_hash_algo
->final_fn(real_sha1
, &c
);
2169 if (hashcmp(expected_sha1
, real_sha1
)) {
2170 error("sha1 mismatch for %s (expected %s)", path
,
2171 sha1_to_hex(expected_sha1
));
2178 int read_loose_object(const char *path
,
2179 const unsigned char *expected_sha1
,
2180 enum object_type
*type
,
2181 unsigned long *size
,
2186 unsigned long mapsize
;
2192 map
= map_sha1_file_1(path
, NULL
, &mapsize
);
2194 error_errno("unable to mmap %s", path
);
2198 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0) {
2199 error("unable to unpack header of %s", path
);
2203 *type
= parse_sha1_header(hdr
, size
);
2205 error("unable to parse header of %s", path
);
2206 git_inflate_end(&stream
);
2210 if (*type
== OBJ_BLOB
) {
2211 if (check_stream_sha1(&stream
, hdr
, *size
, path
, expected_sha1
) < 0)
2214 *contents
= unpack_sha1_rest(&stream
, hdr
, *size
, expected_sha1
);
2216 error("unable to unpack contents of %s", path
);
2217 git_inflate_end(&stream
);
2220 if (check_sha1_signature(expected_sha1
, *contents
,
2221 *size
, type_name(*type
))) {
2222 error("sha1 mismatch for %s (expected %s)", path
,
2223 sha1_to_hex(expected_sha1
));
2229 ret
= 0; /* everything checks out */
2233 munmap(map
, mapsize
);