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 "repository.h"
26 #include "streaming.h"
29 #include "mergesort.h"
32 #include "fetch-object.h"
33 #include "object-store.h"
35 const unsigned char null_sha1
[GIT_MAX_RAWSZ
];
36 const struct object_id null_oid
;
37 const struct object_id empty_tree_oid
= {
38 EMPTY_TREE_SHA1_BIN_LITERAL
40 const struct object_id empty_blob_oid
= {
41 EMPTY_BLOB_SHA1_BIN_LITERAL
44 static void git_hash_sha1_init(void *ctx
)
46 git_SHA1_Init((git_SHA_CTX
*)ctx
);
49 static void git_hash_sha1_update(void *ctx
, const void *data
, size_t len
)
51 git_SHA1_Update((git_SHA_CTX
*)ctx
, data
, len
);
54 static void git_hash_sha1_final(unsigned char *hash
, void *ctx
)
56 git_SHA1_Final(hash
, (git_SHA_CTX
*)ctx
);
59 static void git_hash_unknown_init(void *ctx
)
61 die("trying to init unknown hash");
64 static void git_hash_unknown_update(void *ctx
, const void *data
, size_t len
)
66 die("trying to update unknown hash");
69 static void git_hash_unknown_final(unsigned char *hash
, void *ctx
)
71 die("trying to finalize unknown hash");
74 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 */
103 * This is meant to hold a *small* number of objects that you would
104 * want read_sha1_file() to be able to return, but yet you do not want
105 * to write them into the object store (e.g. a browse-only
108 static struct cached_object
{
109 unsigned char sha1
[20];
110 enum object_type type
;
114 static int cached_object_nr
, cached_object_alloc
;
116 static struct cached_object empty_tree
= {
117 EMPTY_TREE_SHA1_BIN_LITERAL
,
123 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
126 struct cached_object
*co
= cached_objects
;
128 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
129 if (!hashcmp(co
->sha1
, sha1
))
132 if (!hashcmp(sha1
, empty_tree
.sha1
))
138 static int get_conv_flags(unsigned flags
)
140 if (flags
& HASH_RENORMALIZE
)
141 return CONV_EOL_RENORMALIZE
;
142 else if (flags
& HASH_WRITE_OBJECT
)
143 return global_conv_flags_eol
;
149 int mkdir_in_gitdir(const char *path
)
151 if (mkdir(path
, 0777)) {
152 int saved_errno
= errno
;
154 struct strbuf sb
= STRBUF_INIT
;
159 * Are we looking at a path in a symlinked worktree
160 * whose original repository does not yet have it?
161 * e.g. .git/rr-cache pointing at its original
162 * repository in which the user hasn't performed any
163 * conflict resolution yet?
165 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
166 strbuf_readlink(&sb
, path
, st
.st_size
) ||
167 !is_absolute_path(sb
.buf
) ||
168 mkdir(sb
.buf
, 0777)) {
175 return adjust_shared_perm(path
);
178 enum scld_error
safe_create_leading_directories(char *path
)
180 char *next_component
= path
+ offset_1st_component(path
);
181 enum scld_error ret
= SCLD_OK
;
183 while (ret
== SCLD_OK
&& next_component
) {
185 char *slash
= next_component
, slash_character
;
187 while (*slash
&& !is_dir_sep(*slash
))
193 next_component
= slash
+ 1;
194 while (is_dir_sep(*next_component
))
196 if (!*next_component
)
199 slash_character
= *slash
;
201 if (!stat(path
, &st
)) {
203 if (!S_ISDIR(st
.st_mode
)) {
207 } else if (mkdir(path
, 0777)) {
208 if (errno
== EEXIST
&&
209 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
210 ; /* somebody created it since we checked */
211 else if (errno
== ENOENT
)
213 * Either mkdir() failed because
214 * somebody just pruned the containing
215 * directory, or stat() failed because
216 * the file that was in our way was
217 * just removed. Either way, inform
218 * the caller that it might be worth
224 } else if (adjust_shared_perm(path
)) {
227 *slash
= slash_character
;
232 enum scld_error
safe_create_leading_directories_const(const char *path
)
235 /* path points to cache entries, so xstrdup before messing with it */
236 char *buf
= xstrdup(path
);
237 enum scld_error result
= safe_create_leading_directories(buf
);
245 int raceproof_create_file(const char *path
, create_file_fn fn
, void *cb
)
248 * The number of times we will try to remove empty directories
249 * in the way of path. This is only 1 because if another
250 * process is racily creating directories that conflict with
251 * us, we don't want to fight against them.
253 int remove_directories_remaining
= 1;
256 * The number of times that we will try to create the
257 * directories containing path. We are willing to attempt this
258 * more than once, because another process could be trying to
259 * clean up empty directories at the same time as we are
260 * trying to create them.
262 int create_directories_remaining
= 3;
264 /* A scratch copy of path, filled lazily if we need it: */
265 struct strbuf path_copy
= STRBUF_INIT
;
278 if (errno
== EISDIR
&& remove_directories_remaining
-- > 0) {
280 * A directory is in the way. Maybe it is empty; try
284 strbuf_addstr(&path_copy
, path
);
286 if (!remove_dir_recursively(&path_copy
, REMOVE_DIR_EMPTY_ONLY
))
288 } else if (errno
== ENOENT
&& create_directories_remaining
-- > 0) {
290 * Maybe the containing directory didn't exist, or
291 * maybe it was just deleted by a process that is
292 * racing with us to clean up empty directories. Try
295 enum scld_error scld_result
;
298 strbuf_addstr(&path_copy
, path
);
301 scld_result
= safe_create_leading_directories(path_copy
.buf
);
302 if (scld_result
== SCLD_OK
)
304 } while (scld_result
== SCLD_VANISHED
&& create_directories_remaining
-- > 0);
308 strbuf_release(&path_copy
);
313 static void fill_sha1_path(struct strbuf
*buf
, const unsigned char *sha1
)
316 for (i
= 0; i
< 20; i
++) {
317 static char hex
[] = "0123456789abcdef";
318 unsigned int val
= sha1
[i
];
319 strbuf_addch(buf
, hex
[val
>> 4]);
320 strbuf_addch(buf
, hex
[val
& 0xf]);
322 strbuf_addch(buf
, '/');
326 void sha1_file_name(struct strbuf
*buf
, const unsigned char *sha1
)
328 strbuf_addstr(buf
, get_object_directory());
329 strbuf_addch(buf
, '/');
330 fill_sha1_path(buf
, sha1
);
333 struct strbuf
*alt_scratch_buf(struct alternate_object_database
*alt
)
335 strbuf_setlen(&alt
->scratch
, alt
->base_len
);
336 return &alt
->scratch
;
339 static const char *alt_sha1_path(struct alternate_object_database
*alt
,
340 const unsigned char *sha1
)
342 struct strbuf
*buf
= alt_scratch_buf(alt
);
343 fill_sha1_path(buf
, sha1
);
348 * Return non-zero iff the path is usable as an alternate object database.
350 static int alt_odb_usable(struct strbuf
*path
, const char *normalized_objdir
)
352 struct alternate_object_database
*alt
;
354 /* Detect cases where alternate disappeared */
355 if (!is_directory(path
->buf
)) {
356 error("object directory %s does not exist; "
357 "check .git/objects/info/alternates.",
363 * Prevent the common mistake of listing the same
364 * thing twice, or object directory itself.
366 for (alt
= the_repository
->objects
->alt_odb_list
; alt
; alt
= alt
->next
) {
367 if (!fspathcmp(path
->buf
, alt
->path
))
370 if (!fspathcmp(path
->buf
, normalized_objdir
))
377 * Prepare alternate object database registry.
379 * The variable alt_odb_list points at the list of struct
380 * alternate_object_database. The elements on this list come from
381 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
382 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
383 * whose contents is similar to that environment variable but can be
384 * LF separated. Its base points at a statically allocated buffer that
385 * contains "/the/directory/corresponding/to/.git/objects/...", while
386 * its name points just after the slash at the end of ".git/objects/"
387 * in the example above, and has enough space to hold 40-byte hex
388 * SHA1, an extra slash for the first level indirection, and the
391 static void read_info_alternates(const char * relative_base
, int depth
);
392 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
393 int depth
, const char *normalized_objdir
)
395 struct alternate_object_database
*ent
;
396 struct strbuf pathbuf
= STRBUF_INIT
;
398 if (!is_absolute_path(entry
) && relative_base
) {
399 strbuf_realpath(&pathbuf
, relative_base
, 1);
400 strbuf_addch(&pathbuf
, '/');
402 strbuf_addstr(&pathbuf
, entry
);
404 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
405 error("unable to normalize alternate object path: %s",
407 strbuf_release(&pathbuf
);
412 * The trailing slash after the directory name is given by
413 * this function at the end. Remove duplicates.
415 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
416 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
418 if (!alt_odb_usable(&pathbuf
, normalized_objdir
)) {
419 strbuf_release(&pathbuf
);
423 ent
= alloc_alt_odb(pathbuf
.buf
);
425 /* add the alternate entry */
426 *the_repository
->objects
->alt_odb_tail
= ent
;
427 the_repository
->objects
->alt_odb_tail
= &(ent
->next
);
430 /* recursively add alternates */
431 read_info_alternates(pathbuf
.buf
, depth
+ 1);
433 strbuf_release(&pathbuf
);
437 static const char *parse_alt_odb_entry(const char *string
,
445 if (*string
== '#') {
446 /* comment; consume up to next separator */
447 end
= strchrnul(string
, sep
);
448 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
450 * quoted path; unquote_c_style has copied the
451 * data for us and set "end". Broken quoting (e.g.,
452 * an entry that doesn't end with a quote) falls
453 * back to the unquoted case below.
456 /* normal, unquoted path */
457 end
= strchrnul(string
, sep
);
458 strbuf_add(out
, string
, end
- string
);
466 static void link_alt_odb_entries(const char *alt
, int sep
,
467 const char *relative_base
, int depth
)
469 struct strbuf objdirbuf
= STRBUF_INIT
;
470 struct strbuf entry
= STRBUF_INIT
;
476 error("%s: ignoring alternate object stores, nesting too deep.",
481 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
482 if (strbuf_normalize_path(&objdirbuf
) < 0)
483 die("unable to normalize object directory: %s",
487 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
490 link_alt_odb_entry(entry
.buf
, relative_base
, depth
, objdirbuf
.buf
);
492 strbuf_release(&entry
);
493 strbuf_release(&objdirbuf
);
496 static void read_info_alternates(const char * relative_base
, int depth
)
499 struct strbuf buf
= STRBUF_INIT
;
501 path
= xstrfmt("%s/info/alternates", relative_base
);
502 if (strbuf_read_file(&buf
, path
, 1024) < 0) {
503 warn_on_fopen_errors(path
);
508 link_alt_odb_entries(buf
.buf
, '\n', relative_base
, depth
);
509 strbuf_release(&buf
);
513 struct alternate_object_database
*alloc_alt_odb(const char *dir
)
515 struct alternate_object_database
*ent
;
517 FLEX_ALLOC_STR(ent
, path
, dir
);
518 strbuf_init(&ent
->scratch
, 0);
519 strbuf_addf(&ent
->scratch
, "%s/", dir
);
520 ent
->base_len
= ent
->scratch
.len
;
525 void add_to_alternates_file(const char *reference
)
527 struct lock_file lock
= LOCK_INIT
;
528 char *alts
= git_pathdup("objects/info/alternates");
532 hold_lock_file_for_update(&lock
, alts
, LOCK_DIE_ON_ERROR
);
533 out
= fdopen_lock_file(&lock
, "w");
535 die_errno("unable to fdopen alternates lockfile");
537 in
= fopen(alts
, "r");
539 struct strbuf line
= STRBUF_INIT
;
541 while (strbuf_getline(&line
, in
) != EOF
) {
542 if (!strcmp(reference
, line
.buf
)) {
546 fprintf_or_die(out
, "%s\n", line
.buf
);
549 strbuf_release(&line
);
552 else if (errno
!= ENOENT
)
553 die_errno("unable to read alternates file");
556 rollback_lock_file(&lock
);
558 fprintf_or_die(out
, "%s\n", reference
);
559 if (commit_lock_file(&lock
))
560 die_errno("unable to move new alternates file into place");
561 if (the_repository
->objects
->alt_odb_tail
)
562 link_alt_odb_entries(reference
, '\n', NULL
, 0);
567 void add_to_alternates_memory(const char *reference
)
570 * Make sure alternates are initialized, or else our entry may be
571 * overwritten when they are.
575 link_alt_odb_entries(reference
, '\n', NULL
, 0);
579 * Compute the exact path an alternate is at and returns it. In case of
580 * error NULL is returned and the human readable error is added to `err`
581 * `path` may be relative and should point to $GITDIR.
582 * `err` must not be null.
584 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
586 char *ref_git
= NULL
;
587 const char *repo
, *ref_git_s
;
590 ref_git_s
= real_path_if_valid(path
);
593 strbuf_addf(err
, _("path '%s' does not exist"), path
);
597 * Beware: read_gitfile(), real_path() and mkpath()
598 * return static buffer
600 ref_git
= xstrdup(ref_git_s
);
602 repo
= read_gitfile(ref_git
);
604 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
607 ref_git
= xstrdup(repo
);
610 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
611 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
613 ref_git
= ref_git_git
;
614 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
615 struct strbuf sb
= STRBUF_INIT
;
617 if (get_common_dir(&sb
, ref_git
)) {
619 _("reference repository '%s' as a linked "
620 "checkout is not supported yet."),
625 strbuf_addf(err
, _("reference repository '%s' is not a "
626 "local repository."), path
);
630 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
631 strbuf_addf(err
, _("reference repository '%s' is shallow"),
637 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
639 _("reference repository '%s' is grafted"),
647 FREE_AND_NULL(ref_git
);
653 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
655 struct alternate_object_database
*ent
;
659 for (ent
= the_repository
->objects
->alt_odb_list
; ent
; ent
= ent
->next
) {
667 void prepare_alt_odb(void)
669 if (the_repository
->objects
->alt_odb_tail
)
672 the_repository
->objects
->alt_odb_tail
=
673 &the_repository
->objects
->alt_odb_list
;
674 link_alt_odb_entries(the_repository
->objects
->alternate_db
,
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
= the_repository
->objects
->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_sha1_signature(const unsigned char *sha1
, void *map
,
787 unsigned long size
, const char *type
)
789 unsigned char real_sha1
[20];
790 enum object_type obj_type
;
791 struct git_istream
*st
;
797 hash_sha1_file(map
, size
, type
, real_sha1
);
798 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
801 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
805 /* Generate the header */
806 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
810 git_SHA1_Update(&c
, hdr
, hdrlen
);
813 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
821 git_SHA1_Update(&c
, buf
, readlen
);
823 git_SHA1_Final(real_sha1
, &c
);
825 return hashcmp(sha1
, real_sha1
) ? -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
= the_repository
->objects
->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
= the_repository
->objects
->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
->typename
, 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
;
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
->typename
&& !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 sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
1226 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
1227 struct pack_entry e
;
1229 const unsigned char *real
= (flags
& OBJECT_INFO_LOOKUP_REPLACE
) ?
1230 lookup_replace_object(sha1
) :
1232 int already_retried
= 0;
1234 if (is_null_sha1(real
))
1240 if (!(flags
& OBJECT_INFO_SKIP_CACHED
)) {
1241 struct cached_object
*co
= find_cached_object(real
);
1244 *(oi
->typep
) = co
->type
;
1246 *(oi
->sizep
) = co
->size
;
1248 *(oi
->disk_sizep
) = 0;
1249 if (oi
->delta_base_sha1
)
1250 hashclr(oi
->delta_base_sha1
);
1252 strbuf_addstr(oi
->typename
, typename(co
->type
));
1254 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
1255 oi
->whence
= OI_CACHED
;
1261 if (find_pack_entry(real
, &e
))
1264 /* Most likely it's a loose object. */
1265 if (!sha1_loose_object_info(real
, oi
, flags
))
1268 /* Not a loose object; someone else may have just packed it. */
1269 reprepare_packed_git();
1270 if (find_pack_entry(real
, &e
))
1273 /* Check if it is a missing object */
1274 if (fetch_if_missing
&& repository_format_partial_clone
&&
1277 * TODO Investigate haveing fetch_object() return
1278 * TODO error/success and stopping the music here.
1280 fetch_object(repository_format_partial_clone
, real
);
1281 already_retried
= 1;
1288 if (oi
== &blank_oi
)
1290 * We know that the caller doesn't actually need the
1291 * information below, so return early.
1294 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
1296 mark_bad_packed_object(e
.p
, real
);
1297 return sha1_object_info_extended(real
, oi
, 0);
1298 } else if (oi
->whence
== OI_PACKED
) {
1299 oi
->u
.packed
.offset
= e
.offset
;
1300 oi
->u
.packed
.pack
= e
.p
;
1301 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
1302 rtype
== OBJ_OFS_DELTA
);
1308 /* returns enum object_type or negative */
1309 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1311 enum object_type type
;
1312 struct object_info oi
= OBJECT_INFO_INIT
;
1316 if (sha1_object_info_extended(sha1
, &oi
,
1317 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
1322 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1323 unsigned long *size
)
1325 struct object_info oi
= OBJECT_INFO_INIT
;
1329 oi
.contentp
= &content
;
1331 if (sha1_object_info_extended(sha1
, &oi
, 0) < 0)
1336 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1337 unsigned char *sha1
)
1339 struct cached_object
*co
;
1341 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1342 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1344 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
1345 co
= &cached_objects
[cached_object_nr
++];
1348 co
->buf
= xmalloc(len
);
1349 memcpy(co
->buf
, buf
, len
);
1350 hashcpy(co
->sha1
, sha1
);
1355 * This function dies on corrupt objects; the callers who want to
1356 * deal with them should arrange to call read_object() and give error
1357 * messages themselves.
1359 void *read_sha1_file_extended(const unsigned char *sha1
,
1360 enum object_type
*type
,
1361 unsigned long *size
,
1365 const struct packed_git
*p
;
1368 const unsigned char *repl
= lookup_replace
? lookup_replace_object(sha1
)
1372 data
= read_object(repl
, type
, size
);
1376 if (errno
&& errno
!= ENOENT
)
1377 die_errno("failed to read object %s", sha1_to_hex(sha1
));
1379 /* die if we replaced an object with one that does not exist */
1381 die("replacement %s not found for %s",
1382 sha1_to_hex(repl
), sha1_to_hex(sha1
));
1384 if (!stat_sha1_file(repl
, &st
, &path
))
1385 die("loose object %s (stored in %s) is corrupt",
1386 sha1_to_hex(repl
), path
);
1388 if ((p
= has_packed_and_bad(repl
)) != NULL
)
1389 die("packed object %s (stored in %s) is corrupt",
1390 sha1_to_hex(repl
), p
->pack_name
);
1395 void *read_object_with_reference(const unsigned char *sha1
,
1396 const char *required_type_name
,
1397 unsigned long *size
,
1398 unsigned char *actual_sha1_return
)
1400 enum object_type type
, required_type
;
1402 unsigned long isize
;
1403 unsigned char actual_sha1
[20];
1405 required_type
= type_from_string(required_type_name
);
1406 hashcpy(actual_sha1
, sha1
);
1408 int ref_length
= -1;
1409 const char *ref_type
= NULL
;
1411 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1414 if (type
== required_type
) {
1416 if (actual_sha1_return
)
1417 hashcpy(actual_sha1_return
, actual_sha1
);
1420 /* Handle references */
1421 else if (type
== OBJ_COMMIT
)
1423 else if (type
== OBJ_TAG
)
1424 ref_type
= "object ";
1429 ref_length
= strlen(ref_type
);
1431 if (ref_length
+ 40 > isize
||
1432 memcmp(buffer
, ref_type
, ref_length
) ||
1433 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1438 /* Now we have the ID of the referred-to object in
1439 * actual_sha1. Check again. */
1443 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
1444 const char *type
, unsigned char *sha1
,
1445 char *hdr
, int *hdrlen
)
1449 /* Generate the header */
1450 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
1454 git_SHA1_Update(&c
, hdr
, *hdrlen
);
1455 git_SHA1_Update(&c
, buf
, len
);
1456 git_SHA1_Final(sha1
, &c
);
1460 * Move the just written object into its final resting place.
1462 int finalize_object_file(const char *tmpfile
, const char *filename
)
1466 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
1468 else if (link(tmpfile
, filename
))
1472 * Coda hack - coda doesn't like cross-directory links,
1473 * so we fall back to a rename, which will mean that it
1474 * won't be able to check collisions, but that's not a
1477 * The same holds for FAT formatted media.
1479 * When this succeeds, we just return. We have nothing
1482 if (ret
&& ret
!= EEXIST
) {
1484 if (!rename(tmpfile
, filename
))
1488 unlink_or_warn(tmpfile
);
1490 if (ret
!= EEXIST
) {
1491 return error_errno("unable to write sha1 filename %s", filename
);
1493 /* FIXME!!! Collision check here ? */
1497 if (adjust_shared_perm(filename
))
1498 return error("unable to set permission to '%s'", filename
);
1502 static int write_buffer(int fd
, const void *buf
, size_t len
)
1504 if (write_in_full(fd
, buf
, len
) < 0)
1505 return error_errno("file write error");
1509 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
1510 unsigned char *sha1
)
1513 int hdrlen
= sizeof(hdr
);
1514 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
1518 /* Finalize a file on disk, and close it. */
1519 static void close_sha1_file(int fd
)
1521 if (fsync_object_files
)
1522 fsync_or_die(fd
, "sha1 file");
1524 die_errno("error when closing sha1 file");
1527 /* Size of directory component, including the ending '/' */
1528 static inline int directory_size(const char *filename
)
1530 const char *s
= strrchr(filename
, '/');
1533 return s
- filename
+ 1;
1537 * This creates a temporary file in the same directory as the final
1540 * We want to avoid cross-directory filename renames, because those
1541 * can have problems on various filesystems (FAT, NFS, Coda).
1543 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
1545 int fd
, dirlen
= directory_size(filename
);
1548 strbuf_add(tmp
, filename
, dirlen
);
1549 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
1550 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1551 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
1553 * Make sure the directory exists; note that the contents
1554 * of the buffer are undefined after mkstemp returns an
1555 * error, so we have to rewrite the whole buffer from
1559 strbuf_add(tmp
, filename
, dirlen
- 1);
1560 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
1562 if (adjust_shared_perm(tmp
->buf
))
1566 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
1567 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1572 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
1573 const void *buf
, unsigned long len
, time_t mtime
)
1576 unsigned char compressed
[4096];
1579 unsigned char parano_sha1
[20];
1580 static struct strbuf tmp_file
= STRBUF_INIT
;
1581 static struct strbuf filename
= STRBUF_INIT
;
1583 strbuf_reset(&filename
);
1584 sha1_file_name(&filename
, sha1
);
1586 fd
= create_tmpfile(&tmp_file
, filename
.buf
);
1588 if (errno
== EACCES
)
1589 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1591 return error_errno("unable to create temporary file");
1595 git_deflate_init(&stream
, zlib_compression_level
);
1596 stream
.next_out
= compressed
;
1597 stream
.avail_out
= sizeof(compressed
);
1600 /* First header.. */
1601 stream
.next_in
= (unsigned char *)hdr
;
1602 stream
.avail_in
= hdrlen
;
1603 while (git_deflate(&stream
, 0) == Z_OK
)
1605 git_SHA1_Update(&c
, hdr
, hdrlen
);
1607 /* Then the data itself.. */
1608 stream
.next_in
= (void *)buf
;
1609 stream
.avail_in
= len
;
1611 unsigned char *in0
= stream
.next_in
;
1612 ret
= git_deflate(&stream
, Z_FINISH
);
1613 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
1614 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
1615 die("unable to write sha1 file");
1616 stream
.next_out
= compressed
;
1617 stream
.avail_out
= sizeof(compressed
);
1618 } while (ret
== Z_OK
);
1620 if (ret
!= Z_STREAM_END
)
1621 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
1622 ret
= git_deflate_end_gently(&stream
);
1624 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
1625 git_SHA1_Final(parano_sha1
, &c
);
1626 if (hashcmp(sha1
, parano_sha1
) != 0)
1627 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
1629 close_sha1_file(fd
);
1634 utb
.modtime
= mtime
;
1635 if (utime(tmp_file
.buf
, &utb
) < 0)
1636 warning_errno("failed utime() on %s", tmp_file
.buf
);
1639 return finalize_object_file(tmp_file
.buf
, filename
.buf
);
1642 static int freshen_loose_object(const unsigned char *sha1
)
1644 return check_and_freshen(sha1
, 1);
1647 static int freshen_packed_object(const unsigned char *sha1
)
1649 struct pack_entry e
;
1650 if (!find_pack_entry(sha1
, &e
))
1654 if (!freshen_file(e
.p
->pack_name
))
1660 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
1663 int hdrlen
= sizeof(hdr
);
1665 /* Normally if we have it in the pack then we do not bother writing
1666 * it out into .git/objects/??/?{38} file.
1668 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
1669 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
1671 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
1674 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
1675 struct object_id
*oid
, unsigned flags
)
1678 int hdrlen
, status
= 0;
1680 /* type string, SP, %lu of the length plus NUL must fit this */
1681 hdrlen
= strlen(type
) + 32;
1682 header
= xmalloc(hdrlen
);
1683 write_sha1_file_prepare(buf
, len
, type
, oid
->hash
, header
, &hdrlen
);
1685 if (!(flags
& HASH_WRITE_OBJECT
))
1687 if (freshen_packed_object(oid
->hash
) || freshen_loose_object(oid
->hash
))
1689 status
= write_loose_object(oid
->hash
, header
, hdrlen
, buf
, len
, 0);
1696 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
1700 enum object_type type
;
1705 if (has_loose_object(sha1
))
1707 buf
= read_object(sha1
, &type
, &len
);
1709 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
1710 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
1711 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
1717 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
1719 if (!startup_info
->have_repository
)
1721 return sha1_object_info_extended(sha1
, NULL
,
1722 flags
| OBJECT_INFO_SKIP_CACHED
) >= 0;
1725 int has_object_file(const struct object_id
*oid
)
1727 return has_sha1_file(oid
->hash
);
1730 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
1732 return has_sha1_file_with_flags(oid
->hash
, flags
);
1735 static void check_tree(const void *buf
, size_t size
)
1737 struct tree_desc desc
;
1738 struct name_entry entry
;
1740 init_tree_desc(&desc
, buf
, size
);
1741 while (tree_entry(&desc
, &entry
))
1743 * tree_entry() will die() on malformed entries */
1747 static void check_commit(const void *buf
, size_t size
)
1750 memset(&c
, 0, sizeof(c
));
1751 if (parse_commit_buffer(&c
, buf
, size
))
1752 die("corrupt commit");
1755 static void check_tag(const void *buf
, size_t size
)
1758 memset(&t
, 0, sizeof(t
));
1759 if (parse_tag_buffer(&t
, buf
, size
))
1763 static int index_mem(struct object_id
*oid
, void *buf
, size_t size
,
1764 enum object_type type
,
1765 const char *path
, unsigned flags
)
1767 int ret
, re_allocated
= 0;
1768 int write_object
= flags
& HASH_WRITE_OBJECT
;
1774 * Convert blobs to git internal format
1776 if ((type
== OBJ_BLOB
) && path
) {
1777 struct strbuf nbuf
= STRBUF_INIT
;
1778 if (convert_to_git(&the_index
, path
, buf
, size
, &nbuf
,
1779 get_conv_flags(flags
))) {
1780 buf
= strbuf_detach(&nbuf
, &size
);
1784 if (flags
& HASH_FORMAT_CHECK
) {
1785 if (type
== OBJ_TREE
)
1786 check_tree(buf
, size
);
1787 if (type
== OBJ_COMMIT
)
1788 check_commit(buf
, size
);
1789 if (type
== OBJ_TAG
)
1790 check_tag(buf
, size
);
1794 ret
= write_sha1_file(buf
, size
, typename(type
), oid
->hash
);
1796 ret
= hash_sha1_file(buf
, size
, typename(type
), oid
->hash
);
1802 static int index_stream_convert_blob(struct object_id
*oid
, int fd
,
1803 const char *path
, unsigned flags
)
1806 const int write_object
= flags
& HASH_WRITE_OBJECT
;
1807 struct strbuf sbuf
= STRBUF_INIT
;
1810 assert(would_convert_to_git_filter_fd(path
));
1812 convert_to_git_filter_fd(&the_index
, path
, fd
, &sbuf
,
1813 get_conv_flags(flags
));
1816 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
1819 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
1821 strbuf_release(&sbuf
);
1825 static int index_pipe(struct object_id
*oid
, int fd
, enum object_type type
,
1826 const char *path
, unsigned flags
)
1828 struct strbuf sbuf
= STRBUF_INIT
;
1831 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
1832 ret
= index_mem(oid
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
1835 strbuf_release(&sbuf
);
1839 #define SMALL_FILE_SIZE (32*1024)
1841 static int index_core(struct object_id
*oid
, int fd
, size_t size
,
1842 enum object_type type
, const char *path
,
1848 ret
= index_mem(oid
, "", size
, type
, path
, flags
);
1849 } else if (size
<= SMALL_FILE_SIZE
) {
1850 char *buf
= xmalloc(size
);
1851 ssize_t read_result
= read_in_full(fd
, buf
, size
);
1852 if (read_result
< 0)
1853 ret
= error_errno("read error while indexing %s",
1854 path
? path
: "<unknown>");
1855 else if (read_result
!= size
)
1856 ret
= error("short read while indexing %s",
1857 path
? path
: "<unknown>");
1859 ret
= index_mem(oid
, buf
, size
, type
, path
, flags
);
1862 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1863 ret
= index_mem(oid
, buf
, size
, type
, path
, flags
);
1870 * This creates one packfile per large blob unless bulk-checkin
1871 * machinery is "plugged".
1873 * This also bypasses the usual "convert-to-git" dance, and that is on
1874 * purpose. We could write a streaming version of the converting
1875 * functions and insert that before feeding the data to fast-import
1876 * (or equivalent in-core API described above). However, that is
1877 * somewhat complicated, as we do not know the size of the filter
1878 * result, which we need to know beforehand when writing a git object.
1879 * Since the primary motivation for trying to stream from the working
1880 * tree file and to avoid mmaping it in core is to deal with large
1881 * binary blobs, they generally do not want to get any conversion, and
1882 * callers should avoid this code path when filters are requested.
1884 static int index_stream(struct object_id
*oid
, int fd
, size_t size
,
1885 enum object_type type
, const char *path
,
1888 return index_bulk_checkin(oid
->hash
, fd
, size
, type
, path
, flags
);
1891 int index_fd(struct object_id
*oid
, int fd
, struct stat
*st
,
1892 enum object_type type
, const char *path
, unsigned flags
)
1897 * Call xsize_t() only when needed to avoid potentially unnecessary
1898 * die() for large files.
1900 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
1901 ret
= index_stream_convert_blob(oid
, fd
, path
, flags
);
1902 else if (!S_ISREG(st
->st_mode
))
1903 ret
= index_pipe(oid
, fd
, type
, path
, flags
);
1904 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
1905 (path
&& would_convert_to_git(&the_index
, path
)))
1906 ret
= index_core(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
1909 ret
= index_stream(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
1915 int index_path(struct object_id
*oid
, const char *path
, struct stat
*st
, unsigned flags
)
1918 struct strbuf sb
= STRBUF_INIT
;
1921 switch (st
->st_mode
& S_IFMT
) {
1923 fd
= open(path
, O_RDONLY
);
1925 return error_errno("open(\"%s\")", path
);
1926 if (index_fd(oid
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
1927 return error("%s: failed to insert into database",
1931 if (strbuf_readlink(&sb
, path
, st
->st_size
))
1932 return error_errno("readlink(\"%s\")", path
);
1933 if (!(flags
& HASH_WRITE_OBJECT
))
1934 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, oid
->hash
);
1935 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, oid
->hash
))
1936 rc
= error("%s: failed to insert into database", path
);
1937 strbuf_release(&sb
);
1940 return resolve_gitlink_ref(path
, "HEAD", oid
);
1942 return error("%s: unsupported file type", path
);
1947 int read_pack_header(int fd
, struct pack_header
*header
)
1949 if (read_in_full(fd
, header
, sizeof(*header
)) != sizeof(*header
))
1950 /* "eof before pack header was fully read" */
1951 return PH_ERROR_EOF
;
1953 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
1954 /* "protocol error (pack signature mismatch detected)" */
1955 return PH_ERROR_PACK_SIGNATURE
;
1956 if (!pack_version_ok(header
->hdr_version
))
1957 /* "protocol error (pack version unsupported)" */
1958 return PH_ERROR_PROTOCOL
;
1962 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
1964 enum object_type type
= sha1_object_info(sha1
, NULL
);
1966 die("%s is not a valid object", sha1_to_hex(sha1
));
1968 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
1972 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
1973 struct strbuf
*path
,
1974 each_loose_object_fn obj_cb
,
1975 each_loose_cruft_fn cruft_cb
,
1976 each_loose_subdir_fn subdir_cb
,
1979 size_t origlen
, baselen
;
1983 struct object_id oid
;
1985 if (subdir_nr
> 0xff)
1986 BUG("invalid loose object subdirectory: %x", subdir_nr
);
1988 origlen
= path
->len
;
1989 strbuf_complete(path
, '/');
1990 strbuf_addf(path
, "%02x", subdir_nr
);
1992 dir
= opendir(path
->buf
);
1994 if (errno
!= ENOENT
)
1995 r
= error_errno("unable to open %s", path
->buf
);
1996 strbuf_setlen(path
, origlen
);
2000 oid
.hash
[0] = subdir_nr
;
2001 strbuf_addch(path
, '/');
2002 baselen
= path
->len
;
2004 while ((de
= readdir(dir
))) {
2006 if (is_dot_or_dotdot(de
->d_name
))
2009 namelen
= strlen(de
->d_name
);
2010 strbuf_setlen(path
, baselen
);
2011 strbuf_add(path
, de
->d_name
, namelen
);
2012 if (namelen
== GIT_SHA1_HEXSZ
- 2 &&
2013 !hex_to_bytes(oid
.hash
+ 1, de
->d_name
,
2014 GIT_SHA1_RAWSZ
- 1)) {
2016 r
= obj_cb(&oid
, path
->buf
, data
);
2024 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
2031 strbuf_setlen(path
, baselen
- 1);
2032 if (!r
&& subdir_cb
)
2033 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
2035 strbuf_setlen(path
, origlen
);
2040 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
2041 each_loose_object_fn obj_cb
,
2042 each_loose_cruft_fn cruft_cb
,
2043 each_loose_subdir_fn subdir_cb
,
2049 for (i
= 0; i
< 256; i
++) {
2050 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
2059 int for_each_loose_file_in_objdir(const char *path
,
2060 each_loose_object_fn obj_cb
,
2061 each_loose_cruft_fn cruft_cb
,
2062 each_loose_subdir_fn subdir_cb
,
2065 struct strbuf buf
= STRBUF_INIT
;
2068 strbuf_addstr(&buf
, path
);
2069 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
2071 strbuf_release(&buf
);
2076 struct loose_alt_odb_data
{
2077 each_loose_object_fn
*cb
;
2081 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
2084 struct loose_alt_odb_data
*data
= vdata
;
2085 struct strbuf buf
= STRBUF_INIT
;
2088 strbuf_addstr(&buf
, alt
->path
);
2089 r
= for_each_loose_file_in_objdir_buf(&buf
,
2090 data
->cb
, NULL
, NULL
,
2092 strbuf_release(&buf
);
2096 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
2098 struct loose_alt_odb_data alt
;
2101 r
= for_each_loose_file_in_objdir(get_object_directory(),
2102 cb
, NULL
, NULL
, data
);
2106 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
2111 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
2114 static int check_stream_sha1(git_zstream
*stream
,
2118 const unsigned char *expected_sha1
)
2121 unsigned char real_sha1
[GIT_MAX_RAWSZ
];
2122 unsigned char buf
[4096];
2123 unsigned long total_read
;
2127 git_SHA1_Update(&c
, hdr
, stream
->total_out
);
2130 * We already read some bytes into hdr, but the ones up to the NUL
2131 * do not count against the object's content size.
2133 total_read
= stream
->total_out
- strlen(hdr
) - 1;
2136 * This size comparison must be "<=" to read the final zlib packets;
2137 * see the comment in unpack_sha1_rest for details.
2139 while (total_read
<= size
&&
2140 (status
== Z_OK
|| status
== Z_BUF_ERROR
)) {
2141 stream
->next_out
= buf
;
2142 stream
->avail_out
= sizeof(buf
);
2143 if (size
- total_read
< stream
->avail_out
)
2144 stream
->avail_out
= size
- total_read
;
2145 status
= git_inflate(stream
, Z_FINISH
);
2146 git_SHA1_Update(&c
, buf
, stream
->next_out
- buf
);
2147 total_read
+= stream
->next_out
- buf
;
2149 git_inflate_end(stream
);
2151 if (status
!= Z_STREAM_END
) {
2152 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1
));
2155 if (stream
->avail_in
) {
2156 error("garbage at end of loose object '%s'",
2157 sha1_to_hex(expected_sha1
));
2161 git_SHA1_Final(real_sha1
, &c
);
2162 if (hashcmp(expected_sha1
, real_sha1
)) {
2163 error("sha1 mismatch for %s (expected %s)", path
,
2164 sha1_to_hex(expected_sha1
));
2171 int read_loose_object(const char *path
,
2172 const unsigned char *expected_sha1
,
2173 enum object_type
*type
,
2174 unsigned long *size
,
2179 unsigned long mapsize
;
2185 map
= map_sha1_file_1(path
, NULL
, &mapsize
);
2187 error_errno("unable to mmap %s", path
);
2191 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0) {
2192 error("unable to unpack header of %s", path
);
2196 *type
= parse_sha1_header(hdr
, size
);
2198 error("unable to parse header of %s", path
);
2199 git_inflate_end(&stream
);
2203 if (*type
== OBJ_BLOB
) {
2204 if (check_stream_sha1(&stream
, hdr
, *size
, path
, expected_sha1
) < 0)
2207 *contents
= unpack_sha1_rest(&stream
, hdr
, *size
, expected_sha1
);
2209 error("unable to unpack contents of %s", path
);
2210 git_inflate_end(&stream
);
2213 if (check_sha1_signature(expected_sha1
, *contents
,
2214 *size
, typename(*type
))) {
2215 error("sha1 mismatch for %s (expected %s)", path
,
2216 sha1_to_hex(expected_sha1
));
2222 ret
= 0; /* everything checks out */
2226 munmap(map
, mapsize
);