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 raw_object_store
*o
,
352 const char *normalized_objdir
)
354 struct alternate_object_database
*alt
;
356 /* Detect cases where alternate disappeared */
357 if (!is_directory(path
->buf
)) {
358 error("object directory %s does not exist; "
359 "check .git/objects/info/alternates.",
365 * Prevent the common mistake of listing the same
366 * thing twice, or object directory itself.
368 for (alt
= o
->alt_odb_list
; alt
; alt
= alt
->next
) {
369 if (!fspathcmp(path
->buf
, alt
->path
))
372 if (!fspathcmp(path
->buf
, normalized_objdir
))
379 * Prepare alternate object database registry.
381 * The variable alt_odb_list points at the list of struct
382 * alternate_object_database. The elements on this list come from
383 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
384 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
385 * whose contents is similar to that environment variable but can be
386 * LF separated. Its base points at a statically allocated buffer that
387 * contains "/the/directory/corresponding/to/.git/objects/...", while
388 * its name points just after the slash at the end of ".git/objects/"
389 * in the example above, and has enough space to hold 40-byte hex
390 * SHA1, an extra slash for the first level indirection, and the
393 static void read_info_alternates(const char * relative_base
, int depth
);
394 #define link_alt_odb_entry(r, e, rb, d, n) link_alt_odb_entry_##r(e, rb, d, n)
395 static int link_alt_odb_entry_the_repository(const char *entry
,
396 const char *relative_base
, int depth
, const char *normalized_objdir
)
398 struct alternate_object_database
*ent
;
399 struct strbuf pathbuf
= STRBUF_INIT
;
401 if (!is_absolute_path(entry
) && relative_base
) {
402 strbuf_realpath(&pathbuf
, relative_base
, 1);
403 strbuf_addch(&pathbuf
, '/');
405 strbuf_addstr(&pathbuf
, entry
);
407 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
408 error("unable to normalize alternate object path: %s",
410 strbuf_release(&pathbuf
);
415 * The trailing slash after the directory name is given by
416 * this function at the end. Remove duplicates.
418 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
419 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
421 if (!alt_odb_usable(the_repository
->objects
, &pathbuf
, normalized_objdir
)) {
422 strbuf_release(&pathbuf
);
426 ent
= alloc_alt_odb(pathbuf
.buf
);
428 /* add the alternate entry */
429 *the_repository
->objects
->alt_odb_tail
= ent
;
430 the_repository
->objects
->alt_odb_tail
= &(ent
->next
);
433 /* recursively add alternates */
434 read_info_alternates(pathbuf
.buf
, depth
+ 1);
436 strbuf_release(&pathbuf
);
440 static const char *parse_alt_odb_entry(const char *string
,
448 if (*string
== '#') {
449 /* comment; consume up to next separator */
450 end
= strchrnul(string
, sep
);
451 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
453 * quoted path; unquote_c_style has copied the
454 * data for us and set "end". Broken quoting (e.g.,
455 * an entry that doesn't end with a quote) falls
456 * back to the unquoted case below.
459 /* normal, unquoted path */
460 end
= strchrnul(string
, sep
);
461 strbuf_add(out
, string
, end
- string
);
469 static void link_alt_odb_entries(const char *alt
, int sep
,
470 const char *relative_base
, int depth
)
472 struct strbuf objdirbuf
= STRBUF_INIT
;
473 struct strbuf entry
= STRBUF_INIT
;
479 error("%s: ignoring alternate object stores, nesting too deep.",
484 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
485 if (strbuf_normalize_path(&objdirbuf
) < 0)
486 die("unable to normalize object directory: %s",
490 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
493 link_alt_odb_entry(the_repository
, entry
.buf
,
494 relative_base
, depth
, objdirbuf
.buf
);
496 strbuf_release(&entry
);
497 strbuf_release(&objdirbuf
);
500 static void read_info_alternates(const char * relative_base
, int depth
)
503 struct strbuf buf
= STRBUF_INIT
;
505 path
= xstrfmt("%s/info/alternates", relative_base
);
506 if (strbuf_read_file(&buf
, path
, 1024) < 0) {
507 warn_on_fopen_errors(path
);
512 link_alt_odb_entries(buf
.buf
, '\n', relative_base
, depth
);
513 strbuf_release(&buf
);
517 struct alternate_object_database
*alloc_alt_odb(const char *dir
)
519 struct alternate_object_database
*ent
;
521 FLEX_ALLOC_STR(ent
, path
, dir
);
522 strbuf_init(&ent
->scratch
, 0);
523 strbuf_addf(&ent
->scratch
, "%s/", dir
);
524 ent
->base_len
= ent
->scratch
.len
;
529 void add_to_alternates_file(const char *reference
)
531 struct lock_file lock
= LOCK_INIT
;
532 char *alts
= git_pathdup("objects/info/alternates");
536 hold_lock_file_for_update(&lock
, alts
, LOCK_DIE_ON_ERROR
);
537 out
= fdopen_lock_file(&lock
, "w");
539 die_errno("unable to fdopen alternates lockfile");
541 in
= fopen(alts
, "r");
543 struct strbuf line
= STRBUF_INIT
;
545 while (strbuf_getline(&line
, in
) != EOF
) {
546 if (!strcmp(reference
, line
.buf
)) {
550 fprintf_or_die(out
, "%s\n", line
.buf
);
553 strbuf_release(&line
);
556 else if (errno
!= ENOENT
)
557 die_errno("unable to read alternates file");
560 rollback_lock_file(&lock
);
562 fprintf_or_die(out
, "%s\n", reference
);
563 if (commit_lock_file(&lock
))
564 die_errno("unable to move new alternates file into place");
565 if (the_repository
->objects
->alt_odb_tail
)
566 link_alt_odb_entries(reference
, '\n', NULL
, 0);
571 void add_to_alternates_memory(const char *reference
)
574 * Make sure alternates are initialized, or else our entry may be
575 * overwritten when they are.
579 link_alt_odb_entries(reference
, '\n', NULL
, 0);
583 * Compute the exact path an alternate is at and returns it. In case of
584 * error NULL is returned and the human readable error is added to `err`
585 * `path` may be relative and should point to $GITDIR.
586 * `err` must not be null.
588 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
590 char *ref_git
= NULL
;
591 const char *repo
, *ref_git_s
;
594 ref_git_s
= real_path_if_valid(path
);
597 strbuf_addf(err
, _("path '%s' does not exist"), path
);
601 * Beware: read_gitfile(), real_path() and mkpath()
602 * return static buffer
604 ref_git
= xstrdup(ref_git_s
);
606 repo
= read_gitfile(ref_git
);
608 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
611 ref_git
= xstrdup(repo
);
614 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
615 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
617 ref_git
= ref_git_git
;
618 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
619 struct strbuf sb
= STRBUF_INIT
;
621 if (get_common_dir(&sb
, ref_git
)) {
623 _("reference repository '%s' as a linked "
624 "checkout is not supported yet."),
629 strbuf_addf(err
, _("reference repository '%s' is not a "
630 "local repository."), path
);
634 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
635 strbuf_addf(err
, _("reference repository '%s' is shallow"),
641 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
643 _("reference repository '%s' is grafted"),
651 FREE_AND_NULL(ref_git
);
657 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
659 struct alternate_object_database
*ent
;
663 for (ent
= the_repository
->objects
->alt_odb_list
; ent
; ent
= ent
->next
) {
671 void prepare_alt_odb(void)
673 if (the_repository
->objects
->alt_odb_tail
)
676 the_repository
->objects
->alt_odb_tail
=
677 &the_repository
->objects
->alt_odb_list
;
678 link_alt_odb_entries(the_repository
->objects
->alternate_db
,
681 read_info_alternates(get_object_directory(), 0);
684 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
685 static int freshen_file(const char *fn
)
688 t
.actime
= t
.modtime
= time(NULL
);
689 return !utime(fn
, &t
);
693 * All of the check_and_freshen functions return 1 if the file exists and was
694 * freshened (if freshening was requested), 0 otherwise. If they return
695 * 0, you should not assume that it is safe to skip a write of the object (it
696 * either does not exist on disk, or has a stale mtime and may be subject to
699 int check_and_freshen_file(const char *fn
, int freshen
)
701 if (access(fn
, F_OK
))
703 if (freshen
&& !freshen_file(fn
))
708 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
710 static struct strbuf buf
= STRBUF_INIT
;
713 sha1_file_name(&buf
, sha1
);
715 return check_and_freshen_file(buf
.buf
, freshen
);
718 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
720 struct alternate_object_database
*alt
;
722 for (alt
= the_repository
->objects
->alt_odb_list
; alt
; alt
= alt
->next
) {
723 const char *path
= alt_sha1_path(alt
, sha1
);
724 if (check_and_freshen_file(path
, freshen
))
730 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
732 return check_and_freshen_local(sha1
, freshen
) ||
733 check_and_freshen_nonlocal(sha1
, freshen
);
736 int has_loose_object_nonlocal(const unsigned char *sha1
)
738 return check_and_freshen_nonlocal(sha1
, 0);
741 static int has_loose_object(const unsigned char *sha1
)
743 return check_and_freshen(sha1
, 0);
746 static void mmap_limit_check(size_t length
)
748 static size_t limit
= 0;
750 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
755 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
756 (uintmax_t)length
, (uintmax_t)limit
);
759 void *xmmap_gently(void *start
, size_t length
,
760 int prot
, int flags
, int fd
, off_t offset
)
764 mmap_limit_check(length
);
765 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
766 if (ret
== MAP_FAILED
) {
769 release_pack_memory(length
);
770 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
775 void *xmmap(void *start
, size_t length
,
776 int prot
, int flags
, int fd
, off_t offset
)
778 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
779 if (ret
== MAP_FAILED
)
780 die_errno("mmap failed");
785 * With an in-core object data in "map", rehash it to make sure the
786 * object name actually matches "sha1" to detect object corruption.
787 * With "map" == NULL, try reading the object named with "sha1" using
788 * the streaming interface and rehash it to do the same.
790 int check_sha1_signature(const unsigned char *sha1
, void *map
,
791 unsigned long size
, const char *type
)
793 unsigned char real_sha1
[20];
794 enum object_type obj_type
;
795 struct git_istream
*st
;
801 hash_sha1_file(map
, size
, type
, real_sha1
);
802 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
805 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
809 /* Generate the header */
810 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
814 git_SHA1_Update(&c
, hdr
, hdrlen
);
817 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
825 git_SHA1_Update(&c
, buf
, readlen
);
827 git_SHA1_Final(real_sha1
, &c
);
829 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
832 int git_open_cloexec(const char *name
, int flags
)
835 static int o_cloexec
= O_CLOEXEC
;
837 fd
= open(name
, flags
| o_cloexec
);
838 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
839 /* Try again w/o O_CLOEXEC: the kernel might not support it */
840 o_cloexec
&= ~O_CLOEXEC
;
841 fd
= open(name
, flags
| o_cloexec
);
844 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
846 static int fd_cloexec
= FD_CLOEXEC
;
848 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
849 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
850 int flags
= fcntl(fd
, F_GETFD
);
851 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
860 * Find "sha1" as a loose object in the local repository or in an alternate.
861 * Returns 0 on success, negative on failure.
863 * The "path" out-parameter will give the path of the object we found (if any).
864 * Note that it may point to static storage and is only valid until another
865 * call to sha1_file_name(), etc.
867 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
,
870 struct alternate_object_database
*alt
;
871 static struct strbuf buf
= STRBUF_INIT
;
874 sha1_file_name(&buf
, sha1
);
877 if (!lstat(*path
, st
))
882 for (alt
= the_repository
->objects
->alt_odb_list
; alt
; alt
= alt
->next
) {
883 *path
= alt_sha1_path(alt
, sha1
);
884 if (!lstat(*path
, st
))
892 * Like stat_sha1_file(), but actually open the object and return the
893 * descriptor. See the caveats on the "path" parameter above.
895 static int open_sha1_file(const unsigned char *sha1
, const char **path
)
898 struct alternate_object_database
*alt
;
899 int most_interesting_errno
;
900 static struct strbuf buf
= STRBUF_INIT
;
903 sha1_file_name(&buf
, sha1
);
906 fd
= git_open(*path
);
909 most_interesting_errno
= errno
;
912 for (alt
= the_repository
->objects
->alt_odb_list
; alt
; alt
= alt
->next
) {
913 *path
= alt_sha1_path(alt
, sha1
);
914 fd
= git_open(*path
);
917 if (most_interesting_errno
== ENOENT
)
918 most_interesting_errno
= errno
;
920 errno
= most_interesting_errno
;
925 * Map the loose object at "path" if it is not NULL, or the path found by
926 * searching for a loose object named "sha1".
928 static void *map_sha1_file_1(const char *path
,
929 const unsigned char *sha1
,
938 fd
= open_sha1_file(sha1
, &path
);
943 if (!fstat(fd
, &st
)) {
944 *size
= xsize_t(st
.st_size
);
946 /* mmap() is forbidden on empty files */
947 error("object file %s is empty", path
);
950 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
957 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
959 return map_sha1_file_1(NULL
, sha1
, size
);
962 static int unpack_sha1_short_header(git_zstream
*stream
,
963 unsigned char *map
, unsigned long mapsize
,
964 void *buffer
, unsigned long bufsiz
)
966 /* Get the data stream */
967 memset(stream
, 0, sizeof(*stream
));
968 stream
->next_in
= map
;
969 stream
->avail_in
= mapsize
;
970 stream
->next_out
= buffer
;
971 stream
->avail_out
= bufsiz
;
973 git_inflate_init(stream
);
974 return git_inflate(stream
, 0);
977 int unpack_sha1_header(git_zstream
*stream
,
978 unsigned char *map
, unsigned long mapsize
,
979 void *buffer
, unsigned long bufsiz
)
981 int status
= unpack_sha1_short_header(stream
, map
, mapsize
,
987 /* Make sure we have the terminating NUL */
988 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
993 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
994 unsigned long mapsize
, void *buffer
,
995 unsigned long bufsiz
, struct strbuf
*header
)
999 status
= unpack_sha1_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1004 * Check if entire header is unpacked in the first iteration.
1006 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1010 * buffer[0..bufsiz] was not large enough. Copy the partial
1011 * result out to header, and then append the result of further
1012 * reading the stream.
1014 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1015 stream
->next_out
= buffer
;
1016 stream
->avail_out
= bufsiz
;
1019 status
= git_inflate(stream
, 0);
1020 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1021 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1023 stream
->next_out
= buffer
;
1024 stream
->avail_out
= bufsiz
;
1025 } while (status
!= Z_STREAM_END
);
1029 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1031 int bytes
= strlen(buffer
) + 1;
1032 unsigned char *buf
= xmallocz(size
);
1036 n
= stream
->total_out
- bytes
;
1039 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1041 if (bytes
<= size
) {
1043 * The above condition must be (bytes <= size), not
1044 * (bytes < size). In other words, even though we
1045 * expect no more output and set avail_out to zero,
1046 * the input zlib stream may have bytes that express
1047 * "this concludes the stream", and we *do* want to
1050 * Otherwise we would not be able to test that we
1051 * consumed all the input to reach the expected size;
1052 * we also want to check that zlib tells us that all
1053 * went well with status == Z_STREAM_END at the end.
1055 stream
->next_out
= buf
+ bytes
;
1056 stream
->avail_out
= size
- bytes
;
1057 while (status
== Z_OK
)
1058 status
= git_inflate(stream
, Z_FINISH
);
1060 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1061 git_inflate_end(stream
);
1066 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1067 else if (stream
->avail_in
)
1068 error("garbage at end of loose object '%s'",
1075 * We used to just use "sscanf()", but that's actually way
1076 * too permissive for what we want to check. So do an anal
1077 * object header parse by hand.
1079 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1082 const char *type_buf
= hdr
;
1084 int type
, type_len
= 0;
1087 * The type can be of any size but is followed by
1099 type
= type_from_string_gently(type_buf
, type_len
, 1);
1101 strbuf_add(oi
->typename
, type_buf
, type_len
);
1103 * Set type to 0 if its an unknown object and
1104 * we're obtaining the type using '--allow-unknown-type'
1107 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
) && (type
< 0))
1110 die("invalid object type");
1115 * The length must follow immediately, and be in canonical
1116 * decimal format (ie "010" is not valid).
1118 size
= *hdr
++ - '0';
1123 unsigned long c
= *hdr
- '0';
1127 size
= size
* 10 + c
;
1135 * The length must be followed by a zero byte
1137 return *hdr
? -1 : type
;
1140 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1142 struct object_info oi
= OBJECT_INFO_INIT
;
1145 return parse_sha1_header_extended(hdr
, &oi
, 0);
1148 static int sha1_loose_object_info(const unsigned char *sha1
,
1149 struct object_info
*oi
,
1153 unsigned long mapsize
;
1157 struct strbuf hdrbuf
= STRBUF_INIT
;
1158 unsigned long size_scratch
;
1160 if (oi
->delta_base_sha1
)
1161 hashclr(oi
->delta_base_sha1
);
1164 * If we don't care about type or size, then we don't
1165 * need to look inside the object at all. Note that we
1166 * do not optimize out the stat call, even if the
1167 * caller doesn't care about the disk-size, since our
1168 * return value implicitly indicates whether the
1169 * object even exists.
1171 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
&& !oi
->contentp
) {
1174 if (stat_sha1_file(sha1
, &st
, &path
) < 0)
1177 *oi
->disk_sizep
= st
.st_size
;
1181 map
= map_sha1_file(sha1
, &mapsize
);
1186 oi
->sizep
= &size_scratch
;
1189 *oi
->disk_sizep
= mapsize
;
1190 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
)) {
1191 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
1192 status
= error("unable to unpack %s header with --allow-unknown-type",
1194 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1195 status
= error("unable to unpack %s header",
1199 else if (hdrbuf
.len
) {
1200 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
1201 status
= error("unable to parse %s header with --allow-unknown-type",
1203 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
1204 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1206 if (status
>= 0 && oi
->contentp
) {
1207 *oi
->contentp
= unpack_sha1_rest(&stream
, hdr
,
1209 if (!*oi
->contentp
) {
1210 git_inflate_end(&stream
);
1214 git_inflate_end(&stream
);
1216 munmap(map
, mapsize
);
1217 if (status
&& oi
->typep
)
1218 *oi
->typep
= status
;
1219 if (oi
->sizep
== &size_scratch
)
1221 strbuf_release(&hdrbuf
);
1222 oi
->whence
= OI_LOOSE
;
1223 return (status
< 0) ? status
: 0;
1226 int fetch_if_missing
= 1;
1228 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
1230 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
1231 struct pack_entry e
;
1233 const unsigned char *real
= (flags
& OBJECT_INFO_LOOKUP_REPLACE
) ?
1234 lookup_replace_object(sha1
) :
1236 int already_retried
= 0;
1238 if (is_null_sha1(real
))
1244 if (!(flags
& OBJECT_INFO_SKIP_CACHED
)) {
1245 struct cached_object
*co
= find_cached_object(real
);
1248 *(oi
->typep
) = co
->type
;
1250 *(oi
->sizep
) = co
->size
;
1252 *(oi
->disk_sizep
) = 0;
1253 if (oi
->delta_base_sha1
)
1254 hashclr(oi
->delta_base_sha1
);
1256 strbuf_addstr(oi
->typename
, typename(co
->type
));
1258 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
1259 oi
->whence
= OI_CACHED
;
1265 if (find_pack_entry(real
, &e
))
1268 /* Most likely it's a loose object. */
1269 if (!sha1_loose_object_info(real
, oi
, flags
))
1272 /* Not a loose object; someone else may have just packed it. */
1273 reprepare_packed_git();
1274 if (find_pack_entry(real
, &e
))
1277 /* Check if it is a missing object */
1278 if (fetch_if_missing
&& repository_format_partial_clone
&&
1281 * TODO Investigate haveing fetch_object() return
1282 * TODO error/success and stopping the music here.
1284 fetch_object(repository_format_partial_clone
, real
);
1285 already_retried
= 1;
1292 if (oi
== &blank_oi
)
1294 * We know that the caller doesn't actually need the
1295 * information below, so return early.
1298 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
1300 mark_bad_packed_object(e
.p
, real
);
1301 return sha1_object_info_extended(real
, oi
, 0);
1302 } else if (oi
->whence
== OI_PACKED
) {
1303 oi
->u
.packed
.offset
= e
.offset
;
1304 oi
->u
.packed
.pack
= e
.p
;
1305 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
1306 rtype
== OBJ_OFS_DELTA
);
1312 /* returns enum object_type or negative */
1313 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1315 enum object_type type
;
1316 struct object_info oi
= OBJECT_INFO_INIT
;
1320 if (sha1_object_info_extended(sha1
, &oi
,
1321 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
1326 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1327 unsigned long *size
)
1329 struct object_info oi
= OBJECT_INFO_INIT
;
1333 oi
.contentp
= &content
;
1335 if (sha1_object_info_extended(sha1
, &oi
, 0) < 0)
1340 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1341 unsigned char *sha1
)
1343 struct cached_object
*co
;
1345 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1346 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1348 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
1349 co
= &cached_objects
[cached_object_nr
++];
1352 co
->buf
= xmalloc(len
);
1353 memcpy(co
->buf
, buf
, len
);
1354 hashcpy(co
->sha1
, sha1
);
1359 * This function dies on corrupt objects; the callers who want to
1360 * deal with them should arrange to call read_object() and give error
1361 * messages themselves.
1363 void *read_sha1_file_extended(const unsigned char *sha1
,
1364 enum object_type
*type
,
1365 unsigned long *size
,
1369 const struct packed_git
*p
;
1372 const unsigned char *repl
= lookup_replace
? lookup_replace_object(sha1
)
1376 data
= read_object(repl
, type
, size
);
1380 if (errno
&& errno
!= ENOENT
)
1381 die_errno("failed to read object %s", sha1_to_hex(sha1
));
1383 /* die if we replaced an object with one that does not exist */
1385 die("replacement %s not found for %s",
1386 sha1_to_hex(repl
), sha1_to_hex(sha1
));
1388 if (!stat_sha1_file(repl
, &st
, &path
))
1389 die("loose object %s (stored in %s) is corrupt",
1390 sha1_to_hex(repl
), path
);
1392 if ((p
= has_packed_and_bad(repl
)) != NULL
)
1393 die("packed object %s (stored in %s) is corrupt",
1394 sha1_to_hex(repl
), p
->pack_name
);
1399 void *read_object_with_reference(const unsigned char *sha1
,
1400 const char *required_type_name
,
1401 unsigned long *size
,
1402 unsigned char *actual_sha1_return
)
1404 enum object_type type
, required_type
;
1406 unsigned long isize
;
1407 unsigned char actual_sha1
[20];
1409 required_type
= type_from_string(required_type_name
);
1410 hashcpy(actual_sha1
, sha1
);
1412 int ref_length
= -1;
1413 const char *ref_type
= NULL
;
1415 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1418 if (type
== required_type
) {
1420 if (actual_sha1_return
)
1421 hashcpy(actual_sha1_return
, actual_sha1
);
1424 /* Handle references */
1425 else if (type
== OBJ_COMMIT
)
1427 else if (type
== OBJ_TAG
)
1428 ref_type
= "object ";
1433 ref_length
= strlen(ref_type
);
1435 if (ref_length
+ 40 > isize
||
1436 memcmp(buffer
, ref_type
, ref_length
) ||
1437 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1442 /* Now we have the ID of the referred-to object in
1443 * actual_sha1. Check again. */
1447 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
1448 const char *type
, unsigned char *sha1
,
1449 char *hdr
, int *hdrlen
)
1453 /* Generate the header */
1454 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
1458 git_SHA1_Update(&c
, hdr
, *hdrlen
);
1459 git_SHA1_Update(&c
, buf
, len
);
1460 git_SHA1_Final(sha1
, &c
);
1464 * Move the just written object into its final resting place.
1466 int finalize_object_file(const char *tmpfile
, const char *filename
)
1470 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
1472 else if (link(tmpfile
, filename
))
1476 * Coda hack - coda doesn't like cross-directory links,
1477 * so we fall back to a rename, which will mean that it
1478 * won't be able to check collisions, but that's not a
1481 * The same holds for FAT formatted media.
1483 * When this succeeds, we just return. We have nothing
1486 if (ret
&& ret
!= EEXIST
) {
1488 if (!rename(tmpfile
, filename
))
1492 unlink_or_warn(tmpfile
);
1494 if (ret
!= EEXIST
) {
1495 return error_errno("unable to write sha1 filename %s", filename
);
1497 /* FIXME!!! Collision check here ? */
1501 if (adjust_shared_perm(filename
))
1502 return error("unable to set permission to '%s'", filename
);
1506 static int write_buffer(int fd
, const void *buf
, size_t len
)
1508 if (write_in_full(fd
, buf
, len
) < 0)
1509 return error_errno("file write error");
1513 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
1514 unsigned char *sha1
)
1517 int hdrlen
= sizeof(hdr
);
1518 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
1522 /* Finalize a file on disk, and close it. */
1523 static void close_sha1_file(int fd
)
1525 if (fsync_object_files
)
1526 fsync_or_die(fd
, "sha1 file");
1528 die_errno("error when closing sha1 file");
1531 /* Size of directory component, including the ending '/' */
1532 static inline int directory_size(const char *filename
)
1534 const char *s
= strrchr(filename
, '/');
1537 return s
- filename
+ 1;
1541 * This creates a temporary file in the same directory as the final
1544 * We want to avoid cross-directory filename renames, because those
1545 * can have problems on various filesystems (FAT, NFS, Coda).
1547 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
1549 int fd
, dirlen
= directory_size(filename
);
1552 strbuf_add(tmp
, filename
, dirlen
);
1553 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
1554 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1555 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
1557 * Make sure the directory exists; note that the contents
1558 * of the buffer are undefined after mkstemp returns an
1559 * error, so we have to rewrite the whole buffer from
1563 strbuf_add(tmp
, filename
, dirlen
- 1);
1564 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
1566 if (adjust_shared_perm(tmp
->buf
))
1570 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
1571 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
1576 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
1577 const void *buf
, unsigned long len
, time_t mtime
)
1580 unsigned char compressed
[4096];
1583 unsigned char parano_sha1
[20];
1584 static struct strbuf tmp_file
= STRBUF_INIT
;
1585 static struct strbuf filename
= STRBUF_INIT
;
1587 strbuf_reset(&filename
);
1588 sha1_file_name(&filename
, sha1
);
1590 fd
= create_tmpfile(&tmp_file
, filename
.buf
);
1592 if (errno
== EACCES
)
1593 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1595 return error_errno("unable to create temporary file");
1599 git_deflate_init(&stream
, zlib_compression_level
);
1600 stream
.next_out
= compressed
;
1601 stream
.avail_out
= sizeof(compressed
);
1604 /* First header.. */
1605 stream
.next_in
= (unsigned char *)hdr
;
1606 stream
.avail_in
= hdrlen
;
1607 while (git_deflate(&stream
, 0) == Z_OK
)
1609 git_SHA1_Update(&c
, hdr
, hdrlen
);
1611 /* Then the data itself.. */
1612 stream
.next_in
= (void *)buf
;
1613 stream
.avail_in
= len
;
1615 unsigned char *in0
= stream
.next_in
;
1616 ret
= git_deflate(&stream
, Z_FINISH
);
1617 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
1618 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
1619 die("unable to write sha1 file");
1620 stream
.next_out
= compressed
;
1621 stream
.avail_out
= sizeof(compressed
);
1622 } while (ret
== Z_OK
);
1624 if (ret
!= Z_STREAM_END
)
1625 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
1626 ret
= git_deflate_end_gently(&stream
);
1628 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
1629 git_SHA1_Final(parano_sha1
, &c
);
1630 if (hashcmp(sha1
, parano_sha1
) != 0)
1631 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
1633 close_sha1_file(fd
);
1638 utb
.modtime
= mtime
;
1639 if (utime(tmp_file
.buf
, &utb
) < 0)
1640 warning_errno("failed utime() on %s", tmp_file
.buf
);
1643 return finalize_object_file(tmp_file
.buf
, filename
.buf
);
1646 static int freshen_loose_object(const unsigned char *sha1
)
1648 return check_and_freshen(sha1
, 1);
1651 static int freshen_packed_object(const unsigned char *sha1
)
1653 struct pack_entry e
;
1654 if (!find_pack_entry(sha1
, &e
))
1658 if (!freshen_file(e
.p
->pack_name
))
1664 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
1667 int hdrlen
= sizeof(hdr
);
1669 /* Normally if we have it in the pack then we do not bother writing
1670 * it out into .git/objects/??/?{38} file.
1672 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
1673 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
1675 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
1678 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
1679 struct object_id
*oid
, unsigned flags
)
1682 int hdrlen
, status
= 0;
1684 /* type string, SP, %lu of the length plus NUL must fit this */
1685 hdrlen
= strlen(type
) + 32;
1686 header
= xmalloc(hdrlen
);
1687 write_sha1_file_prepare(buf
, len
, type
, oid
->hash
, header
, &hdrlen
);
1689 if (!(flags
& HASH_WRITE_OBJECT
))
1691 if (freshen_packed_object(oid
->hash
) || freshen_loose_object(oid
->hash
))
1693 status
= write_loose_object(oid
->hash
, header
, hdrlen
, buf
, len
, 0);
1700 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
1704 enum object_type type
;
1709 if (has_loose_object(sha1
))
1711 buf
= read_object(sha1
, &type
, &len
);
1713 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
1714 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
1715 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
1721 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
1723 if (!startup_info
->have_repository
)
1725 return sha1_object_info_extended(sha1
, NULL
,
1726 flags
| OBJECT_INFO_SKIP_CACHED
) >= 0;
1729 int has_object_file(const struct object_id
*oid
)
1731 return has_sha1_file(oid
->hash
);
1734 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
1736 return has_sha1_file_with_flags(oid
->hash
, flags
);
1739 static void check_tree(const void *buf
, size_t size
)
1741 struct tree_desc desc
;
1742 struct name_entry entry
;
1744 init_tree_desc(&desc
, buf
, size
);
1745 while (tree_entry(&desc
, &entry
))
1747 * tree_entry() will die() on malformed entries */
1751 static void check_commit(const void *buf
, size_t size
)
1754 memset(&c
, 0, sizeof(c
));
1755 if (parse_commit_buffer(&c
, buf
, size
))
1756 die("corrupt commit");
1759 static void check_tag(const void *buf
, size_t size
)
1762 memset(&t
, 0, sizeof(t
));
1763 if (parse_tag_buffer(&t
, buf
, size
))
1767 static int index_mem(struct object_id
*oid
, void *buf
, size_t size
,
1768 enum object_type type
,
1769 const char *path
, unsigned flags
)
1771 int ret
, re_allocated
= 0;
1772 int write_object
= flags
& HASH_WRITE_OBJECT
;
1778 * Convert blobs to git internal format
1780 if ((type
== OBJ_BLOB
) && path
) {
1781 struct strbuf nbuf
= STRBUF_INIT
;
1782 if (convert_to_git(&the_index
, path
, buf
, size
, &nbuf
,
1783 get_conv_flags(flags
))) {
1784 buf
= strbuf_detach(&nbuf
, &size
);
1788 if (flags
& HASH_FORMAT_CHECK
) {
1789 if (type
== OBJ_TREE
)
1790 check_tree(buf
, size
);
1791 if (type
== OBJ_COMMIT
)
1792 check_commit(buf
, size
);
1793 if (type
== OBJ_TAG
)
1794 check_tag(buf
, size
);
1798 ret
= write_sha1_file(buf
, size
, typename(type
), oid
->hash
);
1800 ret
= hash_sha1_file(buf
, size
, typename(type
), oid
->hash
);
1806 static int index_stream_convert_blob(struct object_id
*oid
, int fd
,
1807 const char *path
, unsigned flags
)
1810 const int write_object
= flags
& HASH_WRITE_OBJECT
;
1811 struct strbuf sbuf
= STRBUF_INIT
;
1814 assert(would_convert_to_git_filter_fd(path
));
1816 convert_to_git_filter_fd(&the_index
, path
, fd
, &sbuf
,
1817 get_conv_flags(flags
));
1820 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
1823 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
1825 strbuf_release(&sbuf
);
1829 static int index_pipe(struct object_id
*oid
, int fd
, enum object_type type
,
1830 const char *path
, unsigned flags
)
1832 struct strbuf sbuf
= STRBUF_INIT
;
1835 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
1836 ret
= index_mem(oid
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
1839 strbuf_release(&sbuf
);
1843 #define SMALL_FILE_SIZE (32*1024)
1845 static int index_core(struct object_id
*oid
, int fd
, size_t size
,
1846 enum object_type type
, const char *path
,
1852 ret
= index_mem(oid
, "", size
, type
, path
, flags
);
1853 } else if (size
<= SMALL_FILE_SIZE
) {
1854 char *buf
= xmalloc(size
);
1855 ssize_t read_result
= read_in_full(fd
, buf
, size
);
1856 if (read_result
< 0)
1857 ret
= error_errno("read error while indexing %s",
1858 path
? path
: "<unknown>");
1859 else if (read_result
!= size
)
1860 ret
= error("short read while indexing %s",
1861 path
? path
: "<unknown>");
1863 ret
= index_mem(oid
, buf
, size
, type
, path
, flags
);
1866 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1867 ret
= index_mem(oid
, buf
, size
, type
, path
, flags
);
1874 * This creates one packfile per large blob unless bulk-checkin
1875 * machinery is "plugged".
1877 * This also bypasses the usual "convert-to-git" dance, and that is on
1878 * purpose. We could write a streaming version of the converting
1879 * functions and insert that before feeding the data to fast-import
1880 * (or equivalent in-core API described above). However, that is
1881 * somewhat complicated, as we do not know the size of the filter
1882 * result, which we need to know beforehand when writing a git object.
1883 * Since the primary motivation for trying to stream from the working
1884 * tree file and to avoid mmaping it in core is to deal with large
1885 * binary blobs, they generally do not want to get any conversion, and
1886 * callers should avoid this code path when filters are requested.
1888 static int index_stream(struct object_id
*oid
, int fd
, size_t size
,
1889 enum object_type type
, const char *path
,
1892 return index_bulk_checkin(oid
->hash
, fd
, size
, type
, path
, flags
);
1895 int index_fd(struct object_id
*oid
, int fd
, struct stat
*st
,
1896 enum object_type type
, const char *path
, unsigned flags
)
1901 * Call xsize_t() only when needed to avoid potentially unnecessary
1902 * die() for large files.
1904 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
1905 ret
= index_stream_convert_blob(oid
, fd
, path
, flags
);
1906 else if (!S_ISREG(st
->st_mode
))
1907 ret
= index_pipe(oid
, fd
, type
, path
, flags
);
1908 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
1909 (path
&& would_convert_to_git(&the_index
, path
)))
1910 ret
= index_core(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
1913 ret
= index_stream(oid
, fd
, xsize_t(st
->st_size
), type
, path
,
1919 int index_path(struct object_id
*oid
, const char *path
, struct stat
*st
, unsigned flags
)
1922 struct strbuf sb
= STRBUF_INIT
;
1925 switch (st
->st_mode
& S_IFMT
) {
1927 fd
= open(path
, O_RDONLY
);
1929 return error_errno("open(\"%s\")", path
);
1930 if (index_fd(oid
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
1931 return error("%s: failed to insert into database",
1935 if (strbuf_readlink(&sb
, path
, st
->st_size
))
1936 return error_errno("readlink(\"%s\")", path
);
1937 if (!(flags
& HASH_WRITE_OBJECT
))
1938 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, oid
->hash
);
1939 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, oid
->hash
))
1940 rc
= error("%s: failed to insert into database", path
);
1941 strbuf_release(&sb
);
1944 return resolve_gitlink_ref(path
, "HEAD", oid
);
1946 return error("%s: unsupported file type", path
);
1951 int read_pack_header(int fd
, struct pack_header
*header
)
1953 if (read_in_full(fd
, header
, sizeof(*header
)) != sizeof(*header
))
1954 /* "eof before pack header was fully read" */
1955 return PH_ERROR_EOF
;
1957 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
1958 /* "protocol error (pack signature mismatch detected)" */
1959 return PH_ERROR_PACK_SIGNATURE
;
1960 if (!pack_version_ok(header
->hdr_version
))
1961 /* "protocol error (pack version unsupported)" */
1962 return PH_ERROR_PROTOCOL
;
1966 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
1968 enum object_type type
= sha1_object_info(sha1
, NULL
);
1970 die("%s is not a valid object", sha1_to_hex(sha1
));
1972 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
1976 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
1977 struct strbuf
*path
,
1978 each_loose_object_fn obj_cb
,
1979 each_loose_cruft_fn cruft_cb
,
1980 each_loose_subdir_fn subdir_cb
,
1983 size_t origlen
, baselen
;
1987 struct object_id oid
;
1989 if (subdir_nr
> 0xff)
1990 BUG("invalid loose object subdirectory: %x", subdir_nr
);
1992 origlen
= path
->len
;
1993 strbuf_complete(path
, '/');
1994 strbuf_addf(path
, "%02x", subdir_nr
);
1996 dir
= opendir(path
->buf
);
1998 if (errno
!= ENOENT
)
1999 r
= error_errno("unable to open %s", path
->buf
);
2000 strbuf_setlen(path
, origlen
);
2004 oid
.hash
[0] = subdir_nr
;
2005 strbuf_addch(path
, '/');
2006 baselen
= path
->len
;
2008 while ((de
= readdir(dir
))) {
2010 if (is_dot_or_dotdot(de
->d_name
))
2013 namelen
= strlen(de
->d_name
);
2014 strbuf_setlen(path
, baselen
);
2015 strbuf_add(path
, de
->d_name
, namelen
);
2016 if (namelen
== GIT_SHA1_HEXSZ
- 2 &&
2017 !hex_to_bytes(oid
.hash
+ 1, de
->d_name
,
2018 GIT_SHA1_RAWSZ
- 1)) {
2020 r
= obj_cb(&oid
, path
->buf
, data
);
2028 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
2035 strbuf_setlen(path
, baselen
- 1);
2036 if (!r
&& subdir_cb
)
2037 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
2039 strbuf_setlen(path
, origlen
);
2044 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
2045 each_loose_object_fn obj_cb
,
2046 each_loose_cruft_fn cruft_cb
,
2047 each_loose_subdir_fn subdir_cb
,
2053 for (i
= 0; i
< 256; i
++) {
2054 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
2063 int for_each_loose_file_in_objdir(const char *path
,
2064 each_loose_object_fn obj_cb
,
2065 each_loose_cruft_fn cruft_cb
,
2066 each_loose_subdir_fn subdir_cb
,
2069 struct strbuf buf
= STRBUF_INIT
;
2072 strbuf_addstr(&buf
, path
);
2073 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
2075 strbuf_release(&buf
);
2080 struct loose_alt_odb_data
{
2081 each_loose_object_fn
*cb
;
2085 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
2088 struct loose_alt_odb_data
*data
= vdata
;
2089 struct strbuf buf
= STRBUF_INIT
;
2092 strbuf_addstr(&buf
, alt
->path
);
2093 r
= for_each_loose_file_in_objdir_buf(&buf
,
2094 data
->cb
, NULL
, NULL
,
2096 strbuf_release(&buf
);
2100 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
2102 struct loose_alt_odb_data alt
;
2105 r
= for_each_loose_file_in_objdir(get_object_directory(),
2106 cb
, NULL
, NULL
, data
);
2110 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
2115 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
2118 static int check_stream_sha1(git_zstream
*stream
,
2122 const unsigned char *expected_sha1
)
2125 unsigned char real_sha1
[GIT_MAX_RAWSZ
];
2126 unsigned char buf
[4096];
2127 unsigned long total_read
;
2131 git_SHA1_Update(&c
, hdr
, stream
->total_out
);
2134 * We already read some bytes into hdr, but the ones up to the NUL
2135 * do not count against the object's content size.
2137 total_read
= stream
->total_out
- strlen(hdr
) - 1;
2140 * This size comparison must be "<=" to read the final zlib packets;
2141 * see the comment in unpack_sha1_rest for details.
2143 while (total_read
<= size
&&
2144 (status
== Z_OK
|| status
== Z_BUF_ERROR
)) {
2145 stream
->next_out
= buf
;
2146 stream
->avail_out
= sizeof(buf
);
2147 if (size
- total_read
< stream
->avail_out
)
2148 stream
->avail_out
= size
- total_read
;
2149 status
= git_inflate(stream
, Z_FINISH
);
2150 git_SHA1_Update(&c
, buf
, stream
->next_out
- buf
);
2151 total_read
+= stream
->next_out
- buf
;
2153 git_inflate_end(stream
);
2155 if (status
!= Z_STREAM_END
) {
2156 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1
));
2159 if (stream
->avail_in
) {
2160 error("garbage at end of loose object '%s'",
2161 sha1_to_hex(expected_sha1
));
2165 git_SHA1_Final(real_sha1
, &c
);
2166 if (hashcmp(expected_sha1
, real_sha1
)) {
2167 error("sha1 mismatch for %s (expected %s)", path
,
2168 sha1_to_hex(expected_sha1
));
2175 int read_loose_object(const char *path
,
2176 const unsigned char *expected_sha1
,
2177 enum object_type
*type
,
2178 unsigned long *size
,
2183 unsigned long mapsize
;
2189 map
= map_sha1_file_1(path
, NULL
, &mapsize
);
2191 error_errno("unable to mmap %s", path
);
2195 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0) {
2196 error("unable to unpack header of %s", path
);
2200 *type
= parse_sha1_header(hdr
, size
);
2202 error("unable to parse header of %s", path
);
2203 git_inflate_end(&stream
);
2207 if (*type
== OBJ_BLOB
) {
2208 if (check_stream_sha1(&stream
, hdr
, *size
, path
, expected_sha1
) < 0)
2211 *contents
= unpack_sha1_rest(&stream
, hdr
, *size
, expected_sha1
);
2213 error("unable to unpack contents of %s", path
);
2214 git_inflate_end(&stream
);
2217 if (check_sha1_signature(expected_sha1
, *contents
,
2218 *size
, typename(*type
))) {
2219 error("sha1 mismatch for %s (expected %s)", path
,
2220 sha1_to_hex(expected_sha1
));
2226 ret
= 0; /* everything checks out */
2230 munmap(map
, mapsize
);