2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
11 #include "string-list.h"
17 #include "run-command.h"
20 #include "tree-walk.h"
22 #include "pack-revindex.h"
23 #include "sha1-lookup.h"
24 #include "bulk-checkin.h"
25 #include "streaming.h"
29 #include "mergesort.h"
33 #define SZ_FMT PRIuMAX
34 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
36 const unsigned char null_sha1
[20];
37 const struct object_id null_oid
;
38 const struct object_id empty_tree_oid
= {
39 EMPTY_TREE_SHA1_BIN_LITERAL
41 const struct object_id empty_blob_oid
= {
42 EMPTY_BLOB_SHA1_BIN_LITERAL
46 * This is meant to hold a *small* number of objects that you would
47 * want read_sha1_file() to be able to return, but yet you do not want
48 * to write them into the object store (e.g. a browse-only
51 static struct cached_object
{
52 unsigned char sha1
[20];
53 enum object_type type
;
57 static int cached_object_nr
, cached_object_alloc
;
59 static struct cached_object empty_tree
= {
60 EMPTY_TREE_SHA1_BIN_LITERAL
,
66 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
69 struct cached_object
*co
= cached_objects
;
71 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
72 if (!hashcmp(co
->sha1
, sha1
))
75 if (!hashcmp(sha1
, empty_tree
.sha1
))
80 int mkdir_in_gitdir(const char *path
)
82 if (mkdir(path
, 0777)) {
83 int saved_errno
= errno
;
85 struct strbuf sb
= STRBUF_INIT
;
90 * Are we looking at a path in a symlinked worktree
91 * whose original repository does not yet have it?
92 * e.g. .git/rr-cache pointing at its original
93 * repository in which the user hasn't performed any
94 * conflict resolution yet?
96 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
97 strbuf_readlink(&sb
, path
, st
.st_size
) ||
98 !is_absolute_path(sb
.buf
) ||
99 mkdir(sb
.buf
, 0777)) {
106 return adjust_shared_perm(path
);
109 enum scld_error
safe_create_leading_directories(char *path
)
111 char *next_component
= path
+ offset_1st_component(path
);
112 enum scld_error ret
= SCLD_OK
;
114 while (ret
== SCLD_OK
&& next_component
) {
116 char *slash
= next_component
, slash_character
;
118 while (*slash
&& !is_dir_sep(*slash
))
124 next_component
= slash
+ 1;
125 while (is_dir_sep(*next_component
))
127 if (!*next_component
)
130 slash_character
= *slash
;
132 if (!stat(path
, &st
)) {
134 if (!S_ISDIR(st
.st_mode
)) {
138 } else if (mkdir(path
, 0777)) {
139 if (errno
== EEXIST
&&
140 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
141 ; /* somebody created it since we checked */
142 else if (errno
== ENOENT
)
144 * Either mkdir() failed because
145 * somebody just pruned the containing
146 * directory, or stat() failed because
147 * the file that was in our way was
148 * just removed. Either way, inform
149 * the caller that it might be worth
155 } else if (adjust_shared_perm(path
)) {
158 *slash
= slash_character
;
163 enum scld_error
safe_create_leading_directories_const(const char *path
)
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf
= xstrdup(path
);
168 enum scld_error result
= safe_create_leading_directories(buf
);
176 int raceproof_create_file(const char *path
, create_file_fn fn
, void *cb
)
179 * The number of times we will try to remove empty directories
180 * in the way of path. This is only 1 because if another
181 * process is racily creating directories that conflict with
182 * us, we don't want to fight against them.
184 int remove_directories_remaining
= 1;
187 * The number of times that we will try to create the
188 * directories containing path. We are willing to attempt this
189 * more than once, because another process could be trying to
190 * clean up empty directories at the same time as we are
191 * trying to create them.
193 int create_directories_remaining
= 3;
195 /* A scratch copy of path, filled lazily if we need it: */
196 struct strbuf path_copy
= STRBUF_INIT
;
209 if (errno
== EISDIR
&& remove_directories_remaining
-- > 0) {
211 * A directory is in the way. Maybe it is empty; try
215 strbuf_addstr(&path_copy
, path
);
217 if (!remove_dir_recursively(&path_copy
, REMOVE_DIR_EMPTY_ONLY
))
219 } else if (errno
== ENOENT
&& create_directories_remaining
-- > 0) {
221 * Maybe the containing directory didn't exist, or
222 * maybe it was just deleted by a process that is
223 * racing with us to clean up empty directories. Try
226 enum scld_error scld_result
;
229 strbuf_addstr(&path_copy
, path
);
232 scld_result
= safe_create_leading_directories(path_copy
.buf
);
233 if (scld_result
== SCLD_OK
)
235 } while (scld_result
== SCLD_VANISHED
&& create_directories_remaining
-- > 0);
239 strbuf_release(&path_copy
);
244 static void fill_sha1_path(struct strbuf
*buf
, const unsigned char *sha1
)
247 for (i
= 0; i
< 20; i
++) {
248 static char hex
[] = "0123456789abcdef";
249 unsigned int val
= sha1
[i
];
250 strbuf_addch(buf
, hex
[val
>> 4]);
251 strbuf_addch(buf
, hex
[val
& 0xf]);
253 strbuf_addch(buf
, '/');
257 const char *sha1_file_name(const unsigned char *sha1
)
259 static struct strbuf buf
= STRBUF_INIT
;
262 strbuf_addf(&buf
, "%s/", get_object_directory());
264 fill_sha1_path(&buf
, sha1
);
268 struct strbuf
*alt_scratch_buf(struct alternate_object_database
*alt
)
270 strbuf_setlen(&alt
->scratch
, alt
->base_len
);
271 return &alt
->scratch
;
274 static const char *alt_sha1_path(struct alternate_object_database
*alt
,
275 const unsigned char *sha1
)
277 struct strbuf
*buf
= alt_scratch_buf(alt
);
278 fill_sha1_path(buf
, sha1
);
282 struct alternate_object_database
*alt_odb_list
;
283 static struct alternate_object_database
**alt_odb_tail
;
286 * Return non-zero iff the path is usable as an alternate object database.
288 static int alt_odb_usable(struct strbuf
*path
, const char *normalized_objdir
)
290 struct alternate_object_database
*alt
;
292 /* Detect cases where alternate disappeared */
293 if (!is_directory(path
->buf
)) {
294 error("object directory %s does not exist; "
295 "check .git/objects/info/alternates.",
301 * Prevent the common mistake of listing the same
302 * thing twice, or object directory itself.
304 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
305 if (!fspathcmp(path
->buf
, alt
->path
))
308 if (!fspathcmp(path
->buf
, normalized_objdir
))
315 * Prepare alternate object database registry.
317 * The variable alt_odb_list points at the list of struct
318 * alternate_object_database. The elements on this list come from
319 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
320 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
321 * whose contents is similar to that environment variable but can be
322 * LF separated. Its base points at a statically allocated buffer that
323 * contains "/the/directory/corresponding/to/.git/objects/...", while
324 * its name points just after the slash at the end of ".git/objects/"
325 * in the example above, and has enough space to hold 40-byte hex
326 * SHA1, an extra slash for the first level indirection, and the
329 static void read_info_alternates(const char * relative_base
, int depth
);
330 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
331 int depth
, const char *normalized_objdir
)
333 struct alternate_object_database
*ent
;
334 struct strbuf pathbuf
= STRBUF_INIT
;
336 if (!is_absolute_path(entry
) && relative_base
) {
337 strbuf_realpath(&pathbuf
, relative_base
, 1);
338 strbuf_addch(&pathbuf
, '/');
340 strbuf_addstr(&pathbuf
, entry
);
342 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
343 error("unable to normalize alternate object path: %s",
345 strbuf_release(&pathbuf
);
350 * The trailing slash after the directory name is given by
351 * this function at the end. Remove duplicates.
353 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
354 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
356 if (!alt_odb_usable(&pathbuf
, normalized_objdir
)) {
357 strbuf_release(&pathbuf
);
361 ent
= alloc_alt_odb(pathbuf
.buf
);
363 /* add the alternate entry */
365 alt_odb_tail
= &(ent
->next
);
368 /* recursively add alternates */
369 read_info_alternates(pathbuf
.buf
, depth
+ 1);
371 strbuf_release(&pathbuf
);
375 static const char *parse_alt_odb_entry(const char *string
,
383 if (*string
== '#') {
384 /* comment; consume up to next separator */
385 end
= strchrnul(string
, sep
);
386 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
388 * quoted path; unquote_c_style has copied the
389 * data for us and set "end". Broken quoting (e.g.,
390 * an entry that doesn't end with a quote) falls
391 * back to the unquoted case below.
394 /* normal, unquoted path */
395 end
= strchrnul(string
, sep
);
396 strbuf_add(out
, string
, end
- string
);
404 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
405 const char *relative_base
, int depth
)
407 struct strbuf objdirbuf
= STRBUF_INIT
;
408 struct strbuf entry
= STRBUF_INIT
;
411 error("%s: ignoring alternate object stores, nesting too deep.",
416 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
417 if (strbuf_normalize_path(&objdirbuf
) < 0)
418 die("unable to normalize object directory: %s",
422 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
425 link_alt_odb_entry(entry
.buf
, relative_base
, depth
, objdirbuf
.buf
);
427 strbuf_release(&entry
);
428 strbuf_release(&objdirbuf
);
431 static void read_info_alternates(const char * relative_base
, int depth
)
439 path
= xstrfmt("%s/info/alternates", relative_base
);
444 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
448 mapsz
= xsize_t(st
.st_size
);
449 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
452 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
457 struct alternate_object_database
*alloc_alt_odb(const char *dir
)
459 struct alternate_object_database
*ent
;
461 FLEX_ALLOC_STR(ent
, path
, dir
);
462 strbuf_init(&ent
->scratch
, 0);
463 strbuf_addf(&ent
->scratch
, "%s/", dir
);
464 ent
->base_len
= ent
->scratch
.len
;
469 void add_to_alternates_file(const char *reference
)
471 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
472 char *alts
= git_pathdup("objects/info/alternates");
475 hold_lock_file_for_update(lock
, alts
, LOCK_DIE_ON_ERROR
);
476 out
= fdopen_lock_file(lock
, "w");
478 die_errno("unable to fdopen alternates lockfile");
480 in
= fopen(alts
, "r");
482 struct strbuf line
= STRBUF_INIT
;
485 while (strbuf_getline(&line
, in
) != EOF
) {
486 if (!strcmp(reference
, line
.buf
)) {
490 fprintf_or_die(out
, "%s\n", line
.buf
);
493 strbuf_release(&line
);
497 rollback_lock_file(lock
);
501 else if (errno
!= ENOENT
)
502 die_errno("unable to read alternates file");
505 fprintf_or_die(out
, "%s\n", reference
);
506 if (commit_lock_file(lock
))
507 die_errno("unable to move new alternates file into place");
509 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
514 void add_to_alternates_memory(const char *reference
)
517 * Make sure alternates are initialized, or else our entry may be
518 * overwritten when they are.
522 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
526 * Compute the exact path an alternate is at and returns it. In case of
527 * error NULL is returned and the human readable error is added to `err`
528 * `path` may be relative and should point to $GITDIR.
529 * `err` must not be null.
531 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
533 char *ref_git
= NULL
;
534 const char *repo
, *ref_git_s
;
537 ref_git_s
= real_path_if_valid(path
);
540 strbuf_addf(err
, _("path '%s' does not exist"), path
);
544 * Beware: read_gitfile(), real_path() and mkpath()
545 * return static buffer
547 ref_git
= xstrdup(ref_git_s
);
549 repo
= read_gitfile(ref_git
);
551 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
554 ref_git
= xstrdup(repo
);
557 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
558 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
560 ref_git
= ref_git_git
;
561 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
562 struct strbuf sb
= STRBUF_INIT
;
564 if (get_common_dir(&sb
, ref_git
)) {
566 _("reference repository '%s' as a linked "
567 "checkout is not supported yet."),
572 strbuf_addf(err
, _("reference repository '%s' is not a "
573 "local repository."), path
);
577 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
578 strbuf_addf(err
, _("reference repository '%s' is shallow"),
584 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
586 _("reference repository '%s' is grafted"),
594 FREE_AND_NULL(ref_git
);
600 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
602 struct alternate_object_database
*ent
;
606 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
614 void prepare_alt_odb(void)
621 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
624 alt_odb_tail
= &alt_odb_list
;
625 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
627 read_info_alternates(get_object_directory(), 0);
630 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
631 static int freshen_file(const char *fn
)
634 t
.actime
= t
.modtime
= time(NULL
);
635 return !utime(fn
, &t
);
639 * All of the check_and_freshen functions return 1 if the file exists and was
640 * freshened (if freshening was requested), 0 otherwise. If they return
641 * 0, you should not assume that it is safe to skip a write of the object (it
642 * either does not exist on disk, or has a stale mtime and may be subject to
645 int check_and_freshen_file(const char *fn
, int freshen
)
647 if (access(fn
, F_OK
))
649 if (freshen
&& !freshen_file(fn
))
654 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
656 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
659 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
661 struct alternate_object_database
*alt
;
663 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
664 const char *path
= alt_sha1_path(alt
, sha1
);
665 if (check_and_freshen_file(path
, freshen
))
671 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
673 return check_and_freshen_local(sha1
, freshen
) ||
674 check_and_freshen_nonlocal(sha1
, freshen
);
677 int has_loose_object_nonlocal(const unsigned char *sha1
)
679 return check_and_freshen_nonlocal(sha1
, 0);
682 static int has_loose_object(const unsigned char *sha1
)
684 return check_and_freshen(sha1
, 0);
687 static unsigned int pack_used_ctr
;
688 static unsigned int pack_mmap_calls
;
689 static unsigned int peak_pack_open_windows
;
690 static unsigned int pack_open_windows
;
691 static unsigned int pack_open_fds
;
692 static unsigned int pack_max_fds
;
693 static size_t peak_pack_mapped
;
694 static size_t pack_mapped
;
695 struct packed_git
*packed_git
;
697 static struct mru packed_git_mru_storage
;
698 struct mru
*packed_git_mru
= &packed_git_mru_storage
;
700 void pack_report(void)
703 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
704 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
705 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
706 sz_fmt(getpagesize()),
707 sz_fmt(packed_git_window_size
),
708 sz_fmt(packed_git_limit
));
710 "pack_report: pack_used_ctr = %10u\n"
711 "pack_report: pack_mmap_calls = %10u\n"
712 "pack_report: pack_open_windows = %10u / %10u\n"
713 "pack_report: pack_mapped = "
714 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
717 pack_open_windows
, peak_pack_open_windows
,
718 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
722 * Open and mmap the index file at path, perform a couple of
723 * consistency checks, then record its information to p. Return 0 on
726 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
729 struct pack_idx_header
*hdr
;
731 uint32_t version
, nr
, i
, *index
;
732 int fd
= git_open(path
);
737 if (fstat(fd
, &st
)) {
741 idx_size
= xsize_t(st
.st_size
);
742 if (idx_size
< 4 * 256 + 20 + 20) {
744 return error("index file %s is too small", path
);
746 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
750 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
751 version
= ntohl(hdr
->idx_version
);
752 if (version
< 2 || version
> 2) {
753 munmap(idx_map
, idx_size
);
754 return error("index file %s is version %"PRIu32
755 " and is not supported by this binary"
756 " (try upgrading GIT to a newer version)",
765 index
+= 2; /* skip index header */
766 for (i
= 0; i
< 256; i
++) {
767 uint32_t n
= ntohl(index
[i
]);
769 munmap(idx_map
, idx_size
);
770 return error("non-monotonic index %s", path
);
778 * - 256 index entries 4 bytes each
779 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
780 * - 20-byte SHA1 of the packfile
781 * - 20-byte SHA1 file checksum
783 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
784 munmap(idx_map
, idx_size
);
785 return error("wrong index v1 file size in %s", path
);
787 } else if (version
== 2) {
790 * - 8 bytes of header
791 * - 256 index entries 4 bytes each
792 * - 20-byte sha1 entry * nr
793 * - 4-byte crc entry * nr
794 * - 4-byte offset entry * nr
795 * - 20-byte SHA1 of the packfile
796 * - 20-byte SHA1 file checksum
797 * And after the 4-byte offset table might be a
798 * variable sized table containing 8-byte entries
799 * for offsets larger than 2^31.
801 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
802 unsigned long max_size
= min_size
;
804 max_size
+= (nr
- 1)*8;
805 if (idx_size
< min_size
|| idx_size
> max_size
) {
806 munmap(idx_map
, idx_size
);
807 return error("wrong index v2 file size in %s", path
);
809 if (idx_size
!= min_size
&&
811 * make sure we can deal with large pack offsets.
812 * 31-bit signed offset won't be enough, neither
813 * 32-bit unsigned one will be.
815 (sizeof(off_t
) <= 4)) {
816 munmap(idx_map
, idx_size
);
817 return error("pack too large for current definition of off_t in %s", path
);
821 p
->index_version
= version
;
822 p
->index_data
= idx_map
;
823 p
->index_size
= idx_size
;
828 int open_pack_index(struct packed_git
*p
)
837 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
838 die("BUG: pack_name does not end in .pack");
839 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
840 ret
= check_packed_git_idx(idx_name
, p
);
845 static void scan_windows(struct packed_git
*p
,
846 struct packed_git
**lru_p
,
847 struct pack_window
**lru_w
,
848 struct pack_window
**lru_l
)
850 struct pack_window
*w
, *w_l
;
852 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
854 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
864 static int unuse_one_window(struct packed_git
*current
)
866 struct packed_git
*p
, *lru_p
= NULL
;
867 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
870 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
871 for (p
= packed_git
; p
; p
= p
->next
)
872 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
874 munmap(lru_w
->base
, lru_w
->len
);
875 pack_mapped
-= lru_w
->len
;
877 lru_l
->next
= lru_w
->next
;
879 lru_p
->windows
= lru_w
->next
;
887 void release_pack_memory(size_t need
)
889 size_t cur
= pack_mapped
;
890 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
894 static void mmap_limit_check(size_t length
)
896 static size_t limit
= 0;
898 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
903 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
904 (uintmax_t)length
, (uintmax_t)limit
);
907 void *xmmap_gently(void *start
, size_t length
,
908 int prot
, int flags
, int fd
, off_t offset
)
912 mmap_limit_check(length
);
913 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
914 if (ret
== MAP_FAILED
) {
917 release_pack_memory(length
);
918 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
923 void *xmmap(void *start
, size_t length
,
924 int prot
, int flags
, int fd
, off_t offset
)
926 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
927 if (ret
== MAP_FAILED
)
928 die_errno("mmap failed");
932 void close_pack_windows(struct packed_git
*p
)
935 struct pack_window
*w
= p
->windows
;
938 die("pack '%s' still has open windows to it",
940 munmap(w
->base
, w
->len
);
941 pack_mapped
-= w
->len
;
943 p
->windows
= w
->next
;
948 static int close_pack_fd(struct packed_git
*p
)
960 static void close_pack(struct packed_git
*p
)
962 close_pack_windows(p
);
967 void close_all_packs(void)
969 struct packed_git
*p
;
971 for (p
= packed_git
; p
; p
= p
->next
)
973 die("BUG: want to close pack marked 'do-not-close'");
980 * The LRU pack is the one with the oldest MRU window, preferring packs
981 * with no used windows, or the oldest mtime if it has no windows allocated.
983 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
985 struct pack_window
*w
, *this_mru_w
;
986 int has_windows_inuse
= 0;
989 * Reject this pack if it has windows and the previously selected
990 * one does not. If this pack does not have windows, reject
991 * it if the pack file is newer than the previously selected one.
993 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
996 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
998 * Reject this pack if any of its windows are in use,
999 * but the previously selected pack did not have any
1000 * inuse windows. Otherwise, record that this pack
1001 * has windows in use.
1004 if (*accept_windows_inuse
)
1005 has_windows_inuse
= 1;
1010 if (w
->last_used
> this_mru_w
->last_used
)
1014 * Reject this pack if it has windows that have been
1015 * used more recently than the previously selected pack.
1016 * If the previously selected pack had windows inuse and
1017 * we have not encountered a window in this pack that is
1018 * inuse, skip this check since we prefer a pack with no
1019 * inuse windows to one that has inuse windows.
1021 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
1022 this_mru_w
->last_used
> (*mru_w
)->last_used
)
1029 *mru_w
= this_mru_w
;
1031 *accept_windows_inuse
= has_windows_inuse
;
1034 static int close_one_pack(void)
1036 struct packed_git
*p
, *lru_p
= NULL
;
1037 struct pack_window
*mru_w
= NULL
;
1038 int accept_windows_inuse
= 1;
1040 for (p
= packed_git
; p
; p
= p
->next
) {
1041 if (p
->pack_fd
== -1)
1043 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
1047 return close_pack_fd(lru_p
);
1052 void unuse_pack(struct pack_window
**w_cursor
)
1054 struct pack_window
*w
= *w_cursor
;
1061 void close_pack_index(struct packed_git
*p
)
1063 if (p
->index_data
) {
1064 munmap((void *)p
->index_data
, p
->index_size
);
1065 p
->index_data
= NULL
;
1069 static unsigned int get_max_fd_limit(void)
1071 #ifdef RLIMIT_NOFILE
1075 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
1076 return lim
.rlim_cur
;
1082 long open_max
= sysconf(_SC_OPEN_MAX
);
1086 * Otherwise, we got -1 for one of the two
1089 * (1) sysconf() did not understand _SC_OPEN_MAX
1090 * and signaled an error with -1; or
1091 * (2) sysconf() said there is no limit.
1093 * We _could_ clear errno before calling sysconf() to
1094 * tell these two cases apart and return a huge number
1095 * in the latter case to let the caller cap it to a
1096 * value that is not so selfish, but letting the
1097 * fallback OPEN_MAX codepath take care of these cases
1106 return 1; /* see the caller ;-) */
1111 * Do not call this directly as this leaks p->pack_fd on error return;
1112 * call open_packed_git() instead.
1114 static int open_packed_git_1(struct packed_git
*p
)
1117 struct pack_header hdr
;
1118 unsigned char sha1
[20];
1119 unsigned char *idx_sha1
;
1122 if (!p
->index_data
&& open_pack_index(p
))
1123 return error("packfile %s index unavailable", p
->pack_name
);
1125 if (!pack_max_fds
) {
1126 unsigned int max_fds
= get_max_fd_limit();
1128 /* Save 3 for stdin/stdout/stderr, 22 for work */
1130 pack_max_fds
= max_fds
- 25;
1135 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
1138 p
->pack_fd
= git_open(p
->pack_name
);
1139 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
1143 /* If we created the struct before we had the pack we lack size. */
1144 if (!p
->pack_size
) {
1145 if (!S_ISREG(st
.st_mode
))
1146 return error("packfile %s not a regular file", p
->pack_name
);
1147 p
->pack_size
= st
.st_size
;
1148 } else if (p
->pack_size
!= st
.st_size
)
1149 return error("packfile %s size changed", p
->pack_name
);
1151 /* We leave these file descriptors open with sliding mmap;
1152 * there is no point keeping them open across exec(), though.
1154 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
1156 return error("cannot determine file descriptor flags");
1157 fd_flag
|= FD_CLOEXEC
;
1158 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
1159 return error("cannot set FD_CLOEXEC");
1161 /* Verify we recognize this pack file format. */
1162 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
1163 return error("file %s is far too short to be a packfile", p
->pack_name
);
1164 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
1165 return error("file %s is not a GIT packfile", p
->pack_name
);
1166 if (!pack_version_ok(hdr
.hdr_version
))
1167 return error("packfile %s is version %"PRIu32
" and not"
1168 " supported (try upgrading GIT to a newer version)",
1169 p
->pack_name
, ntohl(hdr
.hdr_version
));
1171 /* Verify the pack matches its index. */
1172 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
1173 return error("packfile %s claims to have %"PRIu32
" objects"
1174 " while index indicates %"PRIu32
" objects",
1175 p
->pack_name
, ntohl(hdr
.hdr_entries
),
1177 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
1178 return error("end of packfile %s is unavailable", p
->pack_name
);
1179 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
1180 return error("packfile %s signature is unavailable", p
->pack_name
);
1181 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
1182 if (hashcmp(sha1
, idx_sha1
))
1183 return error("packfile %s does not match index", p
->pack_name
);
1187 static int open_packed_git(struct packed_git
*p
)
1189 if (!open_packed_git_1(p
))
1195 static int in_window(struct pack_window
*win
, off_t offset
)
1197 /* We must promise at least 20 bytes (one hash) after the
1198 * offset is available from this window, otherwise the offset
1199 * is not actually in this window and a different window (which
1200 * has that one hash excess) must be used. This is to support
1201 * the object header and delta base parsing routines below.
1203 off_t win_off
= win
->offset
;
1204 return win_off
<= offset
1205 && (offset
+ 20) <= (win_off
+ win
->len
);
1208 unsigned char *use_pack(struct packed_git
*p
,
1209 struct pack_window
**w_cursor
,
1211 unsigned long *left
)
1213 struct pack_window
*win
= *w_cursor
;
1215 /* Since packfiles end in a hash of their content and it's
1216 * pointless to ask for an offset into the middle of that
1217 * hash, and the in_window function above wouldn't match
1218 * don't allow an offset too close to the end of the file.
1220 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
1221 die("packfile %s cannot be accessed", p
->pack_name
);
1222 if (offset
> (p
->pack_size
- 20))
1223 die("offset beyond end of packfile (truncated pack?)");
1225 die(_("offset before end of packfile (broken .idx?)"));
1227 if (!win
|| !in_window(win
, offset
)) {
1230 for (win
= p
->windows
; win
; win
= win
->next
) {
1231 if (in_window(win
, offset
))
1235 size_t window_align
= packed_git_window_size
/ 2;
1238 if (p
->pack_fd
== -1 && open_packed_git(p
))
1239 die("packfile %s cannot be accessed", p
->pack_name
);
1241 win
= xcalloc(1, sizeof(*win
));
1242 win
->offset
= (offset
/ window_align
) * window_align
;
1243 len
= p
->pack_size
- win
->offset
;
1244 if (len
> packed_git_window_size
)
1245 len
= packed_git_window_size
;
1246 win
->len
= (size_t)len
;
1247 pack_mapped
+= win
->len
;
1248 while (packed_git_limit
< pack_mapped
1249 && unuse_one_window(p
))
1251 win
->base
= xmmap(NULL
, win
->len
,
1252 PROT_READ
, MAP_PRIVATE
,
1253 p
->pack_fd
, win
->offset
);
1254 if (win
->base
== MAP_FAILED
)
1255 die_errno("packfile %s cannot be mapped",
1257 if (!win
->offset
&& win
->len
== p
->pack_size
1258 && !p
->do_not_close
)
1261 pack_open_windows
++;
1262 if (pack_mapped
> peak_pack_mapped
)
1263 peak_pack_mapped
= pack_mapped
;
1264 if (pack_open_windows
> peak_pack_open_windows
)
1265 peak_pack_open_windows
= pack_open_windows
;
1266 win
->next
= p
->windows
;
1270 if (win
!= *w_cursor
) {
1271 win
->last_used
= pack_used_ctr
++;
1275 offset
-= win
->offset
;
1277 *left
= win
->len
- xsize_t(offset
);
1278 return win
->base
+ offset
;
1281 static struct packed_git
*alloc_packed_git(int extra
)
1283 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
1284 memset(p
, 0, sizeof(*p
));
1289 static void try_to_free_pack_memory(size_t size
)
1291 release_pack_memory(size
);
1294 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
1296 static int have_set_try_to_free_routine
;
1299 struct packed_git
*p
;
1301 if (!have_set_try_to_free_routine
) {
1302 have_set_try_to_free_routine
= 1;
1303 set_try_to_free_routine(try_to_free_pack_memory
);
1307 * Make sure a corresponding .pack file exists and that
1308 * the index looks sane.
1310 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
1314 * ".pack" is long enough to hold any suffix we're adding (and
1315 * the use xsnprintf double-checks that)
1317 alloc
= st_add3(path_len
, strlen(".pack"), 1);
1318 p
= alloc_packed_git(alloc
);
1319 memcpy(p
->pack_name
, path
, path_len
);
1321 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
1322 if (!access(p
->pack_name
, F_OK
))
1325 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
1326 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1331 /* ok, it looks sane as far as we can check without
1332 * actually mapping the pack file.
1334 p
->pack_size
= st
.st_size
;
1335 p
->pack_local
= local
;
1336 p
->mtime
= st
.st_mtime
;
1337 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1342 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1344 const char *path
= sha1_pack_name(sha1
);
1345 size_t alloc
= st_add(strlen(path
), 1);
1346 struct packed_git
*p
= alloc_packed_git(alloc
);
1348 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
1349 hashcpy(p
->sha1
, sha1
);
1350 if (check_packed_git_idx(idx_path
, p
)) {
1358 void install_packed_git(struct packed_git
*pack
)
1360 if (pack
->pack_fd
!= -1)
1363 pack
->next
= packed_git
;
1367 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
1369 static void report_helper(const struct string_list
*list
,
1370 int seen_bits
, int first
, int last
)
1372 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
1375 for (; first
< last
; first
++)
1376 report_garbage(seen_bits
, list
->items
[first
].string
);
1379 static void report_pack_garbage(struct string_list
*list
)
1381 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1383 if (!report_garbage
)
1386 string_list_sort(list
);
1388 for (i
= 0; i
< list
->nr
; i
++) {
1389 const char *path
= list
->items
[i
].string
;
1390 if (baselen
!= -1 &&
1391 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1392 report_helper(list
, seen_bits
, first
, i
);
1396 if (baselen
== -1) {
1397 const char *dot
= strrchr(path
, '.');
1399 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
1402 baselen
= dot
- path
+ 1;
1405 if (!strcmp(path
+ baselen
, "pack"))
1407 else if (!strcmp(path
+ baselen
, "idx"))
1410 report_helper(list
, seen_bits
, first
, list
->nr
);
1413 static void prepare_packed_git_one(char *objdir
, int local
)
1415 struct strbuf path
= STRBUF_INIT
;
1419 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1421 strbuf_addstr(&path
, objdir
);
1422 strbuf_addstr(&path
, "/pack");
1423 dir
= opendir(path
.buf
);
1425 if (errno
!= ENOENT
)
1426 error_errno("unable to open object pack directory: %s",
1428 strbuf_release(&path
);
1431 strbuf_addch(&path
, '/');
1432 dirnamelen
= path
.len
;
1433 while ((de
= readdir(dir
)) != NULL
) {
1434 struct packed_git
*p
;
1437 if (is_dot_or_dotdot(de
->d_name
))
1440 strbuf_setlen(&path
, dirnamelen
);
1441 strbuf_addstr(&path
, de
->d_name
);
1443 base_len
= path
.len
;
1444 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1445 /* Don't reopen a pack we already have. */
1446 for (p
= packed_git
; p
; p
= p
->next
) {
1448 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1450 !memcmp(p
->pack_name
, path
.buf
, len
))
1455 * See if it really is a valid .idx file with
1456 * corresponding .pack file that we can map.
1458 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1459 install_packed_git(p
);
1462 if (!report_garbage
)
1465 if (ends_with(de
->d_name
, ".idx") ||
1466 ends_with(de
->d_name
, ".pack") ||
1467 ends_with(de
->d_name
, ".bitmap") ||
1468 ends_with(de
->d_name
, ".keep"))
1469 string_list_append(&garbage
, path
.buf
);
1471 report_garbage(PACKDIR_FILE_GARBAGE
, path
.buf
);
1474 report_pack_garbage(&garbage
);
1475 string_list_clear(&garbage
, 0);
1476 strbuf_release(&path
);
1479 static int approximate_object_count_valid
;
1482 * Give a fast, rough count of the number of objects in the repository. This
1483 * ignores loose objects completely. If you have a lot of them, then either
1484 * you should repack because your performance will be awful, or they are
1485 * all unreachable objects about to be pruned, in which case they're not really
1486 * interesting as a measure of repo size in the first place.
1488 unsigned long approximate_object_count(void)
1490 static unsigned long count
;
1491 if (!approximate_object_count_valid
) {
1492 struct packed_git
*p
;
1494 prepare_packed_git();
1496 for (p
= packed_git
; p
; p
= p
->next
) {
1497 if (open_pack_index(p
))
1499 count
+= p
->num_objects
;
1505 static void *get_next_packed_git(const void *p
)
1507 return ((const struct packed_git
*)p
)->next
;
1510 static void set_next_packed_git(void *p
, void *next
)
1512 ((struct packed_git
*)p
)->next
= next
;
1515 static int sort_pack(const void *a_
, const void *b_
)
1517 const struct packed_git
*a
= a_
;
1518 const struct packed_git
*b
= b_
;
1522 * Local packs tend to contain objects specific to our
1523 * variant of the project than remote ones. In addition,
1524 * remote ones could be on a network mounted filesystem.
1525 * Favor local ones for these reasons.
1527 st
= a
->pack_local
- b
->pack_local
;
1532 * Younger packs tend to contain more recent objects,
1533 * and more recent objects tend to get accessed more
1536 if (a
->mtime
< b
->mtime
)
1538 else if (a
->mtime
== b
->mtime
)
1543 static void rearrange_packed_git(void)
1545 packed_git
= llist_mergesort(packed_git
, get_next_packed_git
,
1546 set_next_packed_git
, sort_pack
);
1549 static void prepare_packed_git_mru(void)
1551 struct packed_git
*p
;
1553 mru_clear(packed_git_mru
);
1554 for (p
= packed_git
; p
; p
= p
->next
)
1555 mru_append(packed_git_mru
, p
);
1558 static int prepare_packed_git_run_once
= 0;
1559 void prepare_packed_git(void)
1561 struct alternate_object_database
*alt
;
1563 if (prepare_packed_git_run_once
)
1565 prepare_packed_git_one(get_object_directory(), 1);
1567 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
)
1568 prepare_packed_git_one(alt
->path
, 0);
1569 rearrange_packed_git();
1570 prepare_packed_git_mru();
1571 prepare_packed_git_run_once
= 1;
1574 void reprepare_packed_git(void)
1576 approximate_object_count_valid
= 0;
1577 prepare_packed_git_run_once
= 0;
1578 prepare_packed_git();
1581 static void mark_bad_packed_object(struct packed_git
*p
,
1582 const unsigned char *sha1
)
1585 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1586 if (!hashcmp(sha1
, p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* i
))
1588 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1589 st_mult(GIT_MAX_RAWSZ
,
1590 st_add(p
->num_bad_objects
, 1)));
1591 hashcpy(p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* p
->num_bad_objects
, sha1
);
1592 p
->num_bad_objects
++;
1595 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1597 struct packed_git
*p
;
1600 for (p
= packed_git
; p
; p
= p
->next
)
1601 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1602 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1608 * With an in-core object data in "map", rehash it to make sure the
1609 * object name actually matches "sha1" to detect object corruption.
1610 * With "map" == NULL, try reading the object named with "sha1" using
1611 * the streaming interface and rehash it to do the same.
1613 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1614 unsigned long size
, const char *type
)
1616 unsigned char real_sha1
[20];
1617 enum object_type obj_type
;
1618 struct git_istream
*st
;
1624 hash_sha1_file(map
, size
, type
, real_sha1
);
1625 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1628 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1632 /* Generate the header */
1633 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
1637 git_SHA1_Update(&c
, hdr
, hdrlen
);
1639 char buf
[1024 * 16];
1640 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1648 git_SHA1_Update(&c
, buf
, readlen
);
1650 git_SHA1_Final(real_sha1
, &c
);
1652 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1655 int git_open_cloexec(const char *name
, int flags
)
1658 static int o_cloexec
= O_CLOEXEC
;
1660 fd
= open(name
, flags
| o_cloexec
);
1661 if ((o_cloexec
& O_CLOEXEC
) && fd
< 0 && errno
== EINVAL
) {
1662 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1663 o_cloexec
&= ~O_CLOEXEC
;
1664 fd
= open(name
, flags
| o_cloexec
);
1667 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1669 static int fd_cloexec
= FD_CLOEXEC
;
1671 if (!o_cloexec
&& 0 <= fd
&& fd_cloexec
) {
1672 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1673 int flags
= fcntl(fd
, F_GETFD
);
1674 if (fcntl(fd
, F_SETFD
, flags
| fd_cloexec
))
1683 * Find "sha1" as a loose object in the local repository or in an alternate.
1684 * Returns 0 on success, negative on failure.
1686 * The "path" out-parameter will give the path of the object we found (if any).
1687 * Note that it may point to static storage and is only valid until another
1688 * call to sha1_file_name(), etc.
1690 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
,
1693 struct alternate_object_database
*alt
;
1695 *path
= sha1_file_name(sha1
);
1696 if (!lstat(*path
, st
))
1701 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1702 *path
= alt_sha1_path(alt
, sha1
);
1703 if (!lstat(*path
, st
))
1711 * Like stat_sha1_file(), but actually open the object and return the
1712 * descriptor. See the caveats on the "path" parameter above.
1714 static int open_sha1_file(const unsigned char *sha1
, const char **path
)
1717 struct alternate_object_database
*alt
;
1718 int most_interesting_errno
;
1720 *path
= sha1_file_name(sha1
);
1721 fd
= git_open(*path
);
1724 most_interesting_errno
= errno
;
1727 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1728 *path
= alt_sha1_path(alt
, sha1
);
1729 fd
= git_open(*path
);
1732 if (most_interesting_errno
== ENOENT
)
1733 most_interesting_errno
= errno
;
1735 errno
= most_interesting_errno
;
1740 * Map the loose object at "path" if it is not NULL, or the path found by
1741 * searching for a loose object named "sha1".
1743 static void *map_sha1_file_1(const char *path
,
1744 const unsigned char *sha1
,
1745 unsigned long *size
)
1751 fd
= git_open(path
);
1753 fd
= open_sha1_file(sha1
, &path
);
1758 if (!fstat(fd
, &st
)) {
1759 *size
= xsize_t(st
.st_size
);
1761 /* mmap() is forbidden on empty files */
1762 error("object file %s is empty", path
);
1765 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1772 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1774 return map_sha1_file_1(NULL
, sha1
, size
);
1777 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1778 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1781 unsigned long size
, c
;
1782 unsigned long used
= 0;
1785 *type
= (c
>> 4) & 7;
1789 if (len
<= used
|| bitsizeof(long) <= shift
) {
1790 error("bad object header");
1795 size
+= (c
& 0x7f) << shift
;
1802 static int unpack_sha1_short_header(git_zstream
*stream
,
1803 unsigned char *map
, unsigned long mapsize
,
1804 void *buffer
, unsigned long bufsiz
)
1806 /* Get the data stream */
1807 memset(stream
, 0, sizeof(*stream
));
1808 stream
->next_in
= map
;
1809 stream
->avail_in
= mapsize
;
1810 stream
->next_out
= buffer
;
1811 stream
->avail_out
= bufsiz
;
1813 git_inflate_init(stream
);
1814 return git_inflate(stream
, 0);
1817 int unpack_sha1_header(git_zstream
*stream
,
1818 unsigned char *map
, unsigned long mapsize
,
1819 void *buffer
, unsigned long bufsiz
)
1821 int status
= unpack_sha1_short_header(stream
, map
, mapsize
,
1827 /* Make sure we have the terminating NUL */
1828 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1833 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1834 unsigned long mapsize
, void *buffer
,
1835 unsigned long bufsiz
, struct strbuf
*header
)
1839 status
= unpack_sha1_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1844 * Check if entire header is unpacked in the first iteration.
1846 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1850 * buffer[0..bufsiz] was not large enough. Copy the partial
1851 * result out to header, and then append the result of further
1852 * reading the stream.
1854 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1855 stream
->next_out
= buffer
;
1856 stream
->avail_out
= bufsiz
;
1859 status
= git_inflate(stream
, 0);
1860 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1861 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1863 stream
->next_out
= buffer
;
1864 stream
->avail_out
= bufsiz
;
1865 } while (status
!= Z_STREAM_END
);
1869 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1871 int bytes
= strlen(buffer
) + 1;
1872 unsigned char *buf
= xmallocz(size
);
1876 n
= stream
->total_out
- bytes
;
1879 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1881 if (bytes
<= size
) {
1883 * The above condition must be (bytes <= size), not
1884 * (bytes < size). In other words, even though we
1885 * expect no more output and set avail_out to zero,
1886 * the input zlib stream may have bytes that express
1887 * "this concludes the stream", and we *do* want to
1890 * Otherwise we would not be able to test that we
1891 * consumed all the input to reach the expected size;
1892 * we also want to check that zlib tells us that all
1893 * went well with status == Z_STREAM_END at the end.
1895 stream
->next_out
= buf
+ bytes
;
1896 stream
->avail_out
= size
- bytes
;
1897 while (status
== Z_OK
)
1898 status
= git_inflate(stream
, Z_FINISH
);
1900 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1901 git_inflate_end(stream
);
1906 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1907 else if (stream
->avail_in
)
1908 error("garbage at end of loose object '%s'",
1915 * We used to just use "sscanf()", but that's actually way
1916 * too permissive for what we want to check. So do an anal
1917 * object header parse by hand.
1919 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1922 const char *type_buf
= hdr
;
1924 int type
, type_len
= 0;
1927 * The type can be of any size but is followed by
1939 type
= type_from_string_gently(type_buf
, type_len
, 1);
1941 strbuf_add(oi
->typename
, type_buf
, type_len
);
1943 * Set type to 0 if its an unknown object and
1944 * we're obtaining the type using '--allow-unknown-type'
1947 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
) && (type
< 0))
1950 die("invalid object type");
1955 * The length must follow immediately, and be in canonical
1956 * decimal format (ie "010" is not valid).
1958 size
= *hdr
++ - '0';
1963 unsigned long c
= *hdr
- '0';
1967 size
= size
* 10 + c
;
1975 * The length must be followed by a zero byte
1977 return *hdr
? -1 : type
;
1980 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1982 struct object_info oi
= OBJECT_INFO_INIT
;
1985 return parse_sha1_header_extended(hdr
, &oi
, 0);
1988 unsigned long get_size_from_delta(struct packed_git
*p
,
1989 struct pack_window
**w_curs
,
1992 const unsigned char *data
;
1993 unsigned char delta_head
[20], *in
;
1997 memset(&stream
, 0, sizeof(stream
));
1998 stream
.next_out
= delta_head
;
1999 stream
.avail_out
= sizeof(delta_head
);
2001 git_inflate_init(&stream
);
2003 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
2004 stream
.next_in
= in
;
2005 st
= git_inflate(&stream
, Z_FINISH
);
2006 curpos
+= stream
.next_in
- in
;
2007 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
2008 stream
.total_out
< sizeof(delta_head
));
2009 git_inflate_end(&stream
);
2010 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
2011 error("delta data unpack-initial failed");
2015 /* Examine the initial part of the delta to figure out
2020 /* ignore base size */
2021 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
2023 /* Read the result size */
2024 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
2027 static off_t
get_delta_base(struct packed_git
*p
,
2028 struct pack_window
**w_curs
,
2030 enum object_type type
,
2031 off_t delta_obj_offset
)
2033 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
2036 /* use_pack() assured us we have [base_info, base_info + 20)
2037 * as a range that we can look at without walking off the
2038 * end of the mapped window. Its actually the hash size
2039 * that is assured. An OFS_DELTA longer than the hash size
2040 * is stupid, as then a REF_DELTA would be smaller to store.
2042 if (type
== OBJ_OFS_DELTA
) {
2044 unsigned char c
= base_info
[used
++];
2045 base_offset
= c
& 127;
2048 if (!base_offset
|| MSB(base_offset
, 7))
2049 return 0; /* overflow */
2050 c
= base_info
[used
++];
2051 base_offset
= (base_offset
<< 7) + (c
& 127);
2053 base_offset
= delta_obj_offset
- base_offset
;
2054 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
2055 return 0; /* out of bound */
2057 } else if (type
== OBJ_REF_DELTA
) {
2058 /* The base entry _must_ be in the same pack */
2059 base_offset
= find_pack_entry_one(base_info
, p
);
2062 die("I am totally screwed");
2067 * Like get_delta_base above, but we return the sha1 instead of the pack
2068 * offset. This means it is cheaper for REF deltas (we do not have to do
2069 * the final object lookup), but more expensive for OFS deltas (we
2070 * have to load the revidx to convert the offset back into a sha1).
2072 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
2073 struct pack_window
**w_curs
,
2075 enum object_type type
,
2076 off_t delta_obj_offset
)
2078 if (type
== OBJ_REF_DELTA
) {
2079 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
2081 } else if (type
== OBJ_OFS_DELTA
) {
2082 struct revindex_entry
*revidx
;
2083 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
2084 type
, delta_obj_offset
);
2089 revidx
= find_pack_revindex(p
, base_offset
);
2093 return nth_packed_object_sha1(p
, revidx
->nr
);
2098 int unpack_object_header(struct packed_git
*p
,
2099 struct pack_window
**w_curs
,
2101 unsigned long *sizep
)
2103 unsigned char *base
;
2106 enum object_type type
;
2108 /* use_pack() assures us we have [base, base + 20) available
2109 * as a range that we can look at. (Its actually the hash
2110 * size that is assured.) With our object header encoding
2111 * the maximum deflated object size is 2^137, which is just
2112 * insane, so we know won't exceed what we have been given.
2114 base
= use_pack(p
, w_curs
, *curpos
, &left
);
2115 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
2124 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
2127 struct revindex_entry
*revidx
;
2128 const unsigned char *sha1
;
2129 revidx
= find_pack_revindex(p
, obj_offset
);
2132 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2133 mark_bad_packed_object(p
, sha1
);
2134 type
= sha1_object_info(sha1
, NULL
);
2135 if (type
<= OBJ_NONE
)
2140 #define POI_STACK_PREALLOC 64
2142 static enum object_type
packed_to_object_type(struct packed_git
*p
,
2144 enum object_type type
,
2145 struct pack_window
**w_curs
,
2148 off_t small_poi_stack
[POI_STACK_PREALLOC
];
2149 off_t
*poi_stack
= small_poi_stack
;
2150 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
2152 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2155 /* Push the object we're going to leave behind */
2156 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
2157 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
2158 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
2159 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
2161 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
2163 poi_stack
[poi_stack_nr
++] = obj_offset
;
2164 /* If parsing the base offset fails, just unwind */
2165 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
2168 curpos
= obj_offset
= base_offset
;
2169 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
2170 if (type
<= OBJ_NONE
) {
2171 /* If getting the base itself fails, we first
2172 * retry the base, otherwise unwind */
2173 type
= retry_bad_packed_offset(p
, base_offset
);
2174 if (type
> OBJ_NONE
)
2188 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2189 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2194 if (poi_stack
!= small_poi_stack
)
2199 while (poi_stack_nr
) {
2200 obj_offset
= poi_stack
[--poi_stack_nr
];
2201 type
= retry_bad_packed_offset(p
, obj_offset
);
2202 if (type
> OBJ_NONE
)
2209 static struct hashmap delta_base_cache
;
2210 static size_t delta_base_cached
;
2212 static LIST_HEAD(delta_base_cache_lru
);
2214 struct delta_base_cache_key
{
2215 struct packed_git
*p
;
2219 struct delta_base_cache_entry
{
2220 struct hashmap hash
;
2221 struct delta_base_cache_key key
;
2222 struct list_head lru
;
2225 enum object_type type
;
2228 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
2232 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
2233 hash
+= (hash
>> 8) + (hash
>> 16);
2237 static struct delta_base_cache_entry
*
2238 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
2240 struct hashmap_entry entry
;
2241 struct delta_base_cache_key key
;
2243 if (!delta_base_cache
.cmpfn
)
2246 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
2248 key
.base_offset
= base_offset
;
2249 return hashmap_get(&delta_base_cache
, &entry
, &key
);
2252 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
2253 const struct delta_base_cache_key
*b
)
2255 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
2258 static int delta_base_cache_hash_cmp(const void *unused_cmp_data
,
2259 const void *va
, const void *vb
,
2262 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
2263 const struct delta_base_cache_key
*key
= vkey
;
2265 return !delta_base_cache_key_eq(&a
->key
, key
);
2267 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
2270 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2272 return !!get_delta_base_cache_entry(p
, base_offset
);
2276 * Remove the entry from the cache, but do _not_ free the associated
2277 * entry data. The caller takes ownership of the "data" buffer, and
2278 * should copy out any fields it wants before detaching.
2280 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2282 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
2283 list_del(&ent
->lru
);
2284 delta_base_cached
-= ent
->size
;
2288 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2289 unsigned long *base_size
, enum object_type
*type
)
2291 struct delta_base_cache_entry
*ent
;
2293 ent
= get_delta_base_cache_entry(p
, base_offset
);
2295 return unpack_entry(p
, base_offset
, type
, base_size
);
2300 *base_size
= ent
->size
;
2301 return xmemdupz(ent
->data
, ent
->size
);
2304 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2307 detach_delta_base_cache_entry(ent
);
2310 void clear_delta_base_cache(void)
2312 struct list_head
*lru
, *tmp
;
2313 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
2314 struct delta_base_cache_entry
*entry
=
2315 list_entry(lru
, struct delta_base_cache_entry
, lru
);
2316 release_delta_base_cache(entry
);
2320 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2321 void *base
, unsigned long base_size
, enum object_type type
)
2323 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
2324 struct list_head
*lru
, *tmp
;
2326 delta_base_cached
+= base_size
;
2328 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
2329 struct delta_base_cache_entry
*f
=
2330 list_entry(lru
, struct delta_base_cache_entry
, lru
);
2331 if (delta_base_cached
<= delta_base_cache_limit
)
2333 release_delta_base_cache(f
);
2337 ent
->key
.base_offset
= base_offset
;
2340 ent
->size
= base_size
;
2341 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
2343 if (!delta_base_cache
.cmpfn
)
2344 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, NULL
, 0);
2345 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
2346 hashmap_add(&delta_base_cache
, ent
);
2349 int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
2350 struct object_info
*oi
)
2352 struct pack_window
*w_curs
= NULL
;
2354 off_t curpos
= obj_offset
;
2355 enum object_type type
;
2358 * We always get the representation type, but only convert it to
2359 * a "real" type later if the caller is interested.
2362 *oi
->contentp
= cache_or_unpack_entry(p
, obj_offset
, oi
->sizep
,
2367 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2370 if (!oi
->contentp
&& oi
->sizep
) {
2371 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2372 off_t tmp_pos
= curpos
;
2373 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
2379 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
2380 if (*oi
->sizep
== 0) {
2389 if (oi
->disk_sizep
) {
2390 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2391 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
2394 if (oi
->typep
|| oi
->typename
) {
2395 enum object_type ptot
;
2396 ptot
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
,
2401 const char *tn
= typename(ptot
);
2403 strbuf_addstr(oi
->typename
, tn
);
2411 if (oi
->delta_base_sha1
) {
2412 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2413 const unsigned char *base
;
2415 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
2422 hashcpy(oi
->delta_base_sha1
, base
);
2424 hashclr(oi
->delta_base_sha1
);
2427 oi
->whence
= in_delta_base_cache(p
, obj_offset
) ? OI_DBCACHED
:
2431 unuse_pack(&w_curs
);
2435 static void *unpack_compressed_entry(struct packed_git
*p
,
2436 struct pack_window
**w_curs
,
2442 unsigned char *buffer
, *in
;
2444 buffer
= xmallocz_gently(size
);
2447 memset(&stream
, 0, sizeof(stream
));
2448 stream
.next_out
= buffer
;
2449 stream
.avail_out
= size
+ 1;
2451 git_inflate_init(&stream
);
2453 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
2454 stream
.next_in
= in
;
2455 st
= git_inflate(&stream
, Z_FINISH
);
2456 if (!stream
.avail_out
)
2457 break; /* the payload is larger than it should be */
2458 curpos
+= stream
.next_in
- in
;
2459 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
2460 git_inflate_end(&stream
);
2461 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
2469 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2470 unsigned long *size
);
2472 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2474 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2475 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2476 p
->pack_name
, (uintmax_t)obj_offset
);
2479 int do_check_packed_object_crc
;
2481 #define UNPACK_ENTRY_STACK_PREALLOC 64
2482 struct unpack_entry_stack_ent
{
2488 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2489 enum object_type
*final_type
, unsigned long *final_size
)
2491 struct pack_window
*w_curs
= NULL
;
2492 off_t curpos
= obj_offset
;
2495 enum object_type type
;
2496 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2497 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2498 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2499 int base_from_cache
= 0;
2501 write_pack_access_log(p
, obj_offset
);
2503 /* PHASE 1: drill down to the innermost base object */
2507 struct delta_base_cache_entry
*ent
;
2509 ent
= get_delta_base_cache_entry(p
, curpos
);
2514 detach_delta_base_cache_entry(ent
);
2515 base_from_cache
= 1;
2519 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2520 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2521 off_t len
= revidx
[1].offset
- obj_offset
;
2522 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2523 const unsigned char *sha1
=
2524 nth_packed_object_sha1(p
, revidx
->nr
);
2525 error("bad packed object CRC for %s",
2527 mark_bad_packed_object(p
, sha1
);
2533 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2534 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2537 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2539 error("failed to validate delta base reference "
2540 "at offset %"PRIuMAX
" from %s",
2541 (uintmax_t)curpos
, p
->pack_name
);
2542 /* bail to phase 2, in hopes of recovery */
2547 /* push object, proceed to base */
2548 if (delta_stack_nr
>= delta_stack_alloc
2549 && delta_stack
== small_delta_stack
) {
2550 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2551 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
2552 memcpy(delta_stack
, small_delta_stack
,
2553 sizeof(*delta_stack
)*delta_stack_nr
);
2555 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2557 i
= delta_stack_nr
++;
2558 delta_stack
[i
].obj_offset
= obj_offset
;
2559 delta_stack
[i
].curpos
= curpos
;
2560 delta_stack
[i
].size
= size
;
2562 curpos
= obj_offset
= base_offset
;
2565 /* PHASE 2: handle the base */
2570 die("BUG: unpack_entry: left loop at a valid delta");
2576 if (!base_from_cache
)
2577 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2581 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2582 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2585 /* PHASE 3: apply deltas in order */
2588 * 'data' holds the base data, or NULL if there was corruption
2590 while (delta_stack_nr
) {
2593 void *external_base
= NULL
;
2594 unsigned long delta_size
, base_size
= size
;
2600 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2604 * We're probably in deep shit, but let's try to fetch
2605 * the required base anyway from another pack or loose.
2606 * This is costly but should happen only in the presence
2607 * of a corrupted pack, and is better than failing outright.
2609 struct revindex_entry
*revidx
;
2610 const unsigned char *base_sha1
;
2611 revidx
= find_pack_revindex(p
, obj_offset
);
2613 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2614 error("failed to read delta base object %s"
2615 " at offset %"PRIuMAX
" from %s",
2616 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2618 mark_bad_packed_object(p
, base_sha1
);
2619 base
= read_object(base_sha1
, &type
, &base_size
);
2620 external_base
= base
;
2624 i
= --delta_stack_nr
;
2625 obj_offset
= delta_stack
[i
].obj_offset
;
2626 curpos
= delta_stack
[i
].curpos
;
2627 delta_size
= delta_stack
[i
].size
;
2632 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2635 error("failed to unpack compressed delta "
2636 "at offset %"PRIuMAX
" from %s",
2637 (uintmax_t)curpos
, p
->pack_name
);
2639 free(external_base
);
2643 data
= patch_delta(base
, base_size
,
2644 delta_data
, delta_size
,
2648 * We could not apply the delta; warn the user, but keep going.
2649 * Our failure will be noticed either in the next iteration of
2650 * the loop, or if this is the final delta, in the caller when
2651 * we return NULL. Those code paths will take care of making
2652 * a more explicit warning and retrying with another copy of
2656 error("failed to apply delta");
2659 free(external_base
);
2668 unuse_pack(&w_curs
);
2670 if (delta_stack
!= small_delta_stack
)
2676 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2679 const unsigned char *index
= p
->index_data
;
2681 if (open_pack_index(p
))
2683 index
= p
->index_data
;
2685 if (n
>= p
->num_objects
)
2688 if (p
->index_version
== 1) {
2689 return index
+ 24 * n
+ 4;
2692 return index
+ 20 * n
;
2696 const struct object_id
*nth_packed_object_oid(struct object_id
*oid
,
2697 struct packed_git
*p
,
2700 const unsigned char *hash
= nth_packed_object_sha1(p
, n
);
2703 hashcpy(oid
->hash
, hash
);
2707 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
2709 const unsigned char *ptr
= vptr
;
2710 const unsigned char *start
= p
->index_data
;
2711 const unsigned char *end
= start
+ p
->index_size
;
2713 die(_("offset before start of pack index for %s (corrupt index?)"),
2715 /* No need to check for underflow; .idx files must be at least 8 bytes */
2717 die(_("offset beyond end of pack index for %s (truncated index?)"),
2721 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2723 const unsigned char *index
= p
->index_data
;
2725 if (p
->index_version
== 1) {
2726 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2729 index
+= 8 + p
->num_objects
* (20 + 4);
2730 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2731 if (!(off
& 0x80000000))
2733 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2734 check_pack_index_ptr(p
, index
);
2735 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2736 ntohl(*((uint32_t *)(index
+ 4)));
2740 off_t
find_pack_entry_one(const unsigned char *sha1
,
2741 struct packed_git
*p
)
2743 const uint32_t *level1_ofs
= p
->index_data
;
2744 const unsigned char *index
= p
->index_data
;
2745 unsigned hi
, lo
, stride
;
2746 static int debug_lookup
= -1;
2748 if (debug_lookup
< 0)
2749 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2752 if (open_pack_index(p
))
2754 level1_ofs
= p
->index_data
;
2755 index
= p
->index_data
;
2757 if (p
->index_version
> 1) {
2762 hi
= ntohl(level1_ofs
[*sha1
]);
2763 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2764 if (p
->index_version
> 1) {
2772 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2773 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2776 unsigned mi
= (lo
+ hi
) / 2;
2777 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2780 printf("lo %u hi %u rg %u mi %u\n",
2781 lo
, hi
, hi
- lo
, mi
);
2783 return nth_packed_object_offset(p
, mi
);
2792 int is_pack_valid(struct packed_git
*p
)
2794 /* An already open pack is known to be valid. */
2795 if (p
->pack_fd
!= -1)
2798 /* If the pack has one window completely covering the
2799 * file size, the pack is known to be valid even if
2800 * the descriptor is not currently open.
2803 struct pack_window
*w
= p
->windows
;
2805 if (!w
->offset
&& w
->len
== p
->pack_size
)
2809 /* Force the pack to open to prove its valid. */
2810 return !open_packed_git(p
);
2813 static int fill_pack_entry(const unsigned char *sha1
,
2814 struct pack_entry
*e
,
2815 struct packed_git
*p
)
2819 if (p
->num_bad_objects
) {
2821 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2822 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2826 offset
= find_pack_entry_one(sha1
, p
);
2831 * We are about to tell the caller where they can locate the
2832 * requested object. We better make sure the packfile is
2833 * still here and can be accessed before supplying that
2834 * answer, as it may have been deleted since the index was
2837 if (!is_pack_valid(p
))
2841 hashcpy(e
->sha1
, sha1
);
2846 * Iff a pack file contains the object named by sha1, return true and
2847 * store its location to e.
2849 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2851 struct mru_entry
*p
;
2853 prepare_packed_git();
2857 for (p
= packed_git_mru
->head
; p
; p
= p
->next
) {
2858 if (fill_pack_entry(sha1
, e
, p
->item
)) {
2859 mru_mark(packed_git_mru
, p
);
2866 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2867 struct packed_git
*packs
)
2869 struct packed_git
*p
;
2871 for (p
= packs
; p
; p
= p
->next
) {
2872 if (find_pack_entry_one(sha1
, p
))
2879 static int sha1_loose_object_info(const unsigned char *sha1
,
2880 struct object_info
*oi
,
2884 unsigned long mapsize
;
2888 struct strbuf hdrbuf
= STRBUF_INIT
;
2889 unsigned long size_scratch
;
2891 if (oi
->delta_base_sha1
)
2892 hashclr(oi
->delta_base_sha1
);
2895 * If we don't care about type or size, then we don't
2896 * need to look inside the object at all. Note that we
2897 * do not optimize out the stat call, even if the
2898 * caller doesn't care about the disk-size, since our
2899 * return value implicitly indicates whether the
2900 * object even exists.
2902 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
&& !oi
->contentp
) {
2905 if (stat_sha1_file(sha1
, &st
, &path
) < 0)
2908 *oi
->disk_sizep
= st
.st_size
;
2912 map
= map_sha1_file(sha1
, &mapsize
);
2917 oi
->sizep
= &size_scratch
;
2920 *oi
->disk_sizep
= mapsize
;
2921 if ((flags
& OBJECT_INFO_ALLOW_UNKNOWN_TYPE
)) {
2922 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
2923 status
= error("unable to unpack %s header with --allow-unknown-type",
2925 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2926 status
= error("unable to unpack %s header",
2930 else if (hdrbuf
.len
) {
2931 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
2932 status
= error("unable to parse %s header with --allow-unknown-type",
2934 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
2935 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2937 if (status
>= 0 && oi
->contentp
)
2938 *oi
->contentp
= unpack_sha1_rest(&stream
, hdr
,
2941 git_inflate_end(&stream
);
2943 munmap(map
, mapsize
);
2944 if (status
&& oi
->typep
)
2945 *oi
->typep
= status
;
2946 if (oi
->sizep
== &size_scratch
)
2948 strbuf_release(&hdrbuf
);
2949 oi
->whence
= OI_LOOSE
;
2950 return (status
< 0) ? status
: 0;
2953 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2955 static struct object_info blank_oi
= OBJECT_INFO_INIT
;
2956 struct pack_entry e
;
2958 const unsigned char *real
= (flags
& OBJECT_INFO_LOOKUP_REPLACE
) ?
2959 lookup_replace_object(sha1
) :
2965 if (!(flags
& OBJECT_INFO_SKIP_CACHED
)) {
2966 struct cached_object
*co
= find_cached_object(real
);
2969 *(oi
->typep
) = co
->type
;
2971 *(oi
->sizep
) = co
->size
;
2973 *(oi
->disk_sizep
) = 0;
2974 if (oi
->delta_base_sha1
)
2975 hashclr(oi
->delta_base_sha1
);
2977 strbuf_addstr(oi
->typename
, typename(co
->type
));
2979 *oi
->contentp
= xmemdupz(co
->buf
, co
->size
);
2980 oi
->whence
= OI_CACHED
;
2985 if (!find_pack_entry(real
, &e
)) {
2986 /* Most likely it's a loose object. */
2987 if (!sha1_loose_object_info(real
, oi
, flags
))
2990 /* Not a loose object; someone else may have just packed it. */
2991 if (flags
& OBJECT_INFO_QUICK
) {
2994 reprepare_packed_git();
2995 if (!find_pack_entry(real
, &e
))
3000 if (oi
== &blank_oi
)
3002 * We know that the caller doesn't actually need the
3003 * information below, so return early.
3007 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
3009 mark_bad_packed_object(e
.p
, real
);
3010 return sha1_object_info_extended(real
, oi
, 0);
3011 } else if (oi
->whence
== OI_PACKED
) {
3012 oi
->u
.packed
.offset
= e
.offset
;
3013 oi
->u
.packed
.pack
= e
.p
;
3014 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
3015 rtype
== OBJ_OFS_DELTA
);
3021 /* returns enum object_type or negative */
3022 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
3024 enum object_type type
;
3025 struct object_info oi
= OBJECT_INFO_INIT
;
3029 if (sha1_object_info_extended(sha1
, &oi
,
3030 OBJECT_INFO_LOOKUP_REPLACE
) < 0)
3035 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
3036 unsigned char *sha1
)
3038 struct cached_object
*co
;
3040 hash_sha1_file(buf
, len
, typename(type
), sha1
);
3041 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
3043 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
3044 co
= &cached_objects
[cached_object_nr
++];
3047 co
->buf
= xmalloc(len
);
3048 memcpy(co
->buf
, buf
, len
);
3049 hashcpy(co
->sha1
, sha1
);
3053 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
3054 unsigned long *size
)
3056 struct object_info oi
= OBJECT_INFO_INIT
;
3060 oi
.contentp
= &content
;
3062 if (sha1_object_info_extended(sha1
, &oi
, 0) < 0)
3068 * This function dies on corrupt objects; the callers who want to
3069 * deal with them should arrange to call read_object() and give error
3070 * messages themselves.
3072 void *read_sha1_file_extended(const unsigned char *sha1
,
3073 enum object_type
*type
,
3074 unsigned long *size
,
3078 const struct packed_git
*p
;
3081 const unsigned char *repl
= lookup_replace
? lookup_replace_object(sha1
)
3085 data
= read_object(repl
, type
, size
);
3089 if (errno
&& errno
!= ENOENT
)
3090 die_errno("failed to read object %s", sha1_to_hex(sha1
));
3092 /* die if we replaced an object with one that does not exist */
3094 die("replacement %s not found for %s",
3095 sha1_to_hex(repl
), sha1_to_hex(sha1
));
3097 if (!stat_sha1_file(repl
, &st
, &path
))
3098 die("loose object %s (stored in %s) is corrupt",
3099 sha1_to_hex(repl
), path
);
3101 if ((p
= has_packed_and_bad(repl
)) != NULL
)
3102 die("packed object %s (stored in %s) is corrupt",
3103 sha1_to_hex(repl
), p
->pack_name
);
3108 void *read_object_with_reference(const unsigned char *sha1
,
3109 const char *required_type_name
,
3110 unsigned long *size
,
3111 unsigned char *actual_sha1_return
)
3113 enum object_type type
, required_type
;
3115 unsigned long isize
;
3116 unsigned char actual_sha1
[20];
3118 required_type
= type_from_string(required_type_name
);
3119 hashcpy(actual_sha1
, sha1
);
3121 int ref_length
= -1;
3122 const char *ref_type
= NULL
;
3124 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
3127 if (type
== required_type
) {
3129 if (actual_sha1_return
)
3130 hashcpy(actual_sha1_return
, actual_sha1
);
3133 /* Handle references */
3134 else if (type
== OBJ_COMMIT
)
3136 else if (type
== OBJ_TAG
)
3137 ref_type
= "object ";
3142 ref_length
= strlen(ref_type
);
3144 if (ref_length
+ 40 > isize
||
3145 memcmp(buffer
, ref_type
, ref_length
) ||
3146 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
3151 /* Now we have the ID of the referred-to object in
3152 * actual_sha1. Check again. */
3156 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
3157 const char *type
, unsigned char *sha1
,
3158 char *hdr
, int *hdrlen
)
3162 /* Generate the header */
3163 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
3167 git_SHA1_Update(&c
, hdr
, *hdrlen
);
3168 git_SHA1_Update(&c
, buf
, len
);
3169 git_SHA1_Final(sha1
, &c
);
3173 * Move the just written object into its final resting place.
3175 int finalize_object_file(const char *tmpfile
, const char *filename
)
3179 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
3181 else if (link(tmpfile
, filename
))
3185 * Coda hack - coda doesn't like cross-directory links,
3186 * so we fall back to a rename, which will mean that it
3187 * won't be able to check collisions, but that's not a
3190 * The same holds for FAT formatted media.
3192 * When this succeeds, we just return. We have nothing
3195 if (ret
&& ret
!= EEXIST
) {
3197 if (!rename(tmpfile
, filename
))
3201 unlink_or_warn(tmpfile
);
3203 if (ret
!= EEXIST
) {
3204 return error_errno("unable to write sha1 filename %s", filename
);
3206 /* FIXME!!! Collision check here ? */
3210 if (adjust_shared_perm(filename
))
3211 return error("unable to set permission to '%s'", filename
);
3215 static int write_buffer(int fd
, const void *buf
, size_t len
)
3217 if (write_in_full(fd
, buf
, len
) < 0)
3218 return error_errno("file write error");
3222 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
3223 unsigned char *sha1
)
3226 int hdrlen
= sizeof(hdr
);
3227 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3231 /* Finalize a file on disk, and close it. */
3232 static void close_sha1_file(int fd
)
3234 if (fsync_object_files
)
3235 fsync_or_die(fd
, "sha1 file");
3237 die_errno("error when closing sha1 file");
3240 /* Size of directory component, including the ending '/' */
3241 static inline int directory_size(const char *filename
)
3243 const char *s
= strrchr(filename
, '/');
3246 return s
- filename
+ 1;
3250 * This creates a temporary file in the same directory as the final
3253 * We want to avoid cross-directory filename renames, because those
3254 * can have problems on various filesystems (FAT, NFS, Coda).
3256 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
3258 int fd
, dirlen
= directory_size(filename
);
3261 strbuf_add(tmp
, filename
, dirlen
);
3262 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
3263 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3264 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
3266 * Make sure the directory exists; note that the contents
3267 * of the buffer are undefined after mkstemp returns an
3268 * error, so we have to rewrite the whole buffer from
3272 strbuf_add(tmp
, filename
, dirlen
- 1);
3273 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
3275 if (adjust_shared_perm(tmp
->buf
))
3279 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
3280 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3285 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
3286 const void *buf
, unsigned long len
, time_t mtime
)
3289 unsigned char compressed
[4096];
3292 unsigned char parano_sha1
[20];
3293 static struct strbuf tmp_file
= STRBUF_INIT
;
3294 const char *filename
= sha1_file_name(sha1
);
3296 fd
= create_tmpfile(&tmp_file
, filename
);
3298 if (errno
== EACCES
)
3299 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3301 return error_errno("unable to create temporary file");
3305 git_deflate_init(&stream
, zlib_compression_level
);
3306 stream
.next_out
= compressed
;
3307 stream
.avail_out
= sizeof(compressed
);
3310 /* First header.. */
3311 stream
.next_in
= (unsigned char *)hdr
;
3312 stream
.avail_in
= hdrlen
;
3313 while (git_deflate(&stream
, 0) == Z_OK
)
3315 git_SHA1_Update(&c
, hdr
, hdrlen
);
3317 /* Then the data itself.. */
3318 stream
.next_in
= (void *)buf
;
3319 stream
.avail_in
= len
;
3321 unsigned char *in0
= stream
.next_in
;
3322 ret
= git_deflate(&stream
, Z_FINISH
);
3323 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
3324 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
3325 die("unable to write sha1 file");
3326 stream
.next_out
= compressed
;
3327 stream
.avail_out
= sizeof(compressed
);
3328 } while (ret
== Z_OK
);
3330 if (ret
!= Z_STREAM_END
)
3331 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
3332 ret
= git_deflate_end_gently(&stream
);
3334 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
3335 git_SHA1_Final(parano_sha1
, &c
);
3336 if (hashcmp(sha1
, parano_sha1
) != 0)
3337 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
3339 close_sha1_file(fd
);
3344 utb
.modtime
= mtime
;
3345 if (utime(tmp_file
.buf
, &utb
) < 0)
3346 warning_errno("failed utime() on %s", tmp_file
.buf
);
3349 return finalize_object_file(tmp_file
.buf
, filename
);
3352 static int freshen_loose_object(const unsigned char *sha1
)
3354 return check_and_freshen(sha1
, 1);
3357 static int freshen_packed_object(const unsigned char *sha1
)
3359 struct pack_entry e
;
3360 if (!find_pack_entry(sha1
, &e
))
3364 if (!freshen_file(e
.p
->pack_name
))
3370 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
3373 int hdrlen
= sizeof(hdr
);
3375 /* Normally if we have it in the pack then we do not bother writing
3376 * it out into .git/objects/??/?{38} file.
3378 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3379 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3381 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
3384 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
3385 unsigned char *sha1
, unsigned flags
)
3388 int hdrlen
, status
= 0;
3390 /* type string, SP, %lu of the length plus NUL must fit this */
3391 hdrlen
= strlen(type
) + 32;
3392 header
= xmalloc(hdrlen
);
3393 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
3395 if (!(flags
& HASH_WRITE_OBJECT
))
3397 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3399 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
3406 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
3410 enum object_type type
;
3415 if (has_loose_object(sha1
))
3417 buf
= read_object(sha1
, &type
, &len
);
3419 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3420 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
3421 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3427 int has_pack_index(const unsigned char *sha1
)
3430 if (stat(sha1_pack_index_name(sha1
), &st
))
3435 int has_sha1_pack(const unsigned char *sha1
)
3437 struct pack_entry e
;
3438 return find_pack_entry(sha1
, &e
);
3441 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
3443 if (!startup_info
->have_repository
)
3445 return sha1_object_info_extended(sha1
, NULL
,
3446 flags
| OBJECT_INFO_SKIP_CACHED
) >= 0;
3449 int has_object_file(const struct object_id
*oid
)
3451 return has_sha1_file(oid
->hash
);
3454 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
3456 return has_sha1_file_with_flags(oid
->hash
, flags
);
3459 static void check_tree(const void *buf
, size_t size
)
3461 struct tree_desc desc
;
3462 struct name_entry entry
;
3464 init_tree_desc(&desc
, buf
, size
);
3465 while (tree_entry(&desc
, &entry
))
3467 * tree_entry() will die() on malformed entries */
3471 static void check_commit(const void *buf
, size_t size
)
3474 memset(&c
, 0, sizeof(c
));
3475 if (parse_commit_buffer(&c
, buf
, size
))
3476 die("corrupt commit");
3479 static void check_tag(const void *buf
, size_t size
)
3482 memset(&t
, 0, sizeof(t
));
3483 if (parse_tag_buffer(&t
, buf
, size
))
3487 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3488 enum object_type type
,
3489 const char *path
, unsigned flags
)
3491 int ret
, re_allocated
= 0;
3492 int write_object
= flags
& HASH_WRITE_OBJECT
;
3498 * Convert blobs to git internal format
3500 if ((type
== OBJ_BLOB
) && path
) {
3501 struct strbuf nbuf
= STRBUF_INIT
;
3502 if (convert_to_git(&the_index
, path
, buf
, size
, &nbuf
,
3503 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3504 buf
= strbuf_detach(&nbuf
, &size
);
3508 if (flags
& HASH_FORMAT_CHECK
) {
3509 if (type
== OBJ_TREE
)
3510 check_tree(buf
, size
);
3511 if (type
== OBJ_COMMIT
)
3512 check_commit(buf
, size
);
3513 if (type
== OBJ_TAG
)
3514 check_tag(buf
, size
);
3518 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3520 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3526 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
3527 const char *path
, unsigned flags
)
3530 const int write_object
= flags
& HASH_WRITE_OBJECT
;
3531 struct strbuf sbuf
= STRBUF_INIT
;
3534 assert(would_convert_to_git_filter_fd(path
));
3536 convert_to_git_filter_fd(&the_index
, path
, fd
, &sbuf
,
3537 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
3540 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3543 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3545 strbuf_release(&sbuf
);
3549 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3550 const char *path
, unsigned flags
)
3552 struct strbuf sbuf
= STRBUF_INIT
;
3555 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3556 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3559 strbuf_release(&sbuf
);
3563 #define SMALL_FILE_SIZE (32*1024)
3565 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3566 enum object_type type
, const char *path
,
3572 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
3573 } else if (size
<= SMALL_FILE_SIZE
) {
3574 char *buf
= xmalloc(size
);
3575 if (size
== read_in_full(fd
, buf
, size
))
3576 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3578 ret
= error_errno("short read");
3581 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3582 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3589 * This creates one packfile per large blob unless bulk-checkin
3590 * machinery is "plugged".
3592 * This also bypasses the usual "convert-to-git" dance, and that is on
3593 * purpose. We could write a streaming version of the converting
3594 * functions and insert that before feeding the data to fast-import
3595 * (or equivalent in-core API described above). However, that is
3596 * somewhat complicated, as we do not know the size of the filter
3597 * result, which we need to know beforehand when writing a git object.
3598 * Since the primary motivation for trying to stream from the working
3599 * tree file and to avoid mmaping it in core is to deal with large
3600 * binary blobs, they generally do not want to get any conversion, and
3601 * callers should avoid this code path when filters are requested.
3603 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3604 enum object_type type
, const char *path
,
3607 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3610 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3611 enum object_type type
, const char *path
, unsigned flags
)
3616 * Call xsize_t() only when needed to avoid potentially unnecessary
3617 * die() for large files.
3619 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
3620 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
3621 else if (!S_ISREG(st
->st_mode
))
3622 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3623 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3624 (path
&& would_convert_to_git(&the_index
, path
)))
3625 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3628 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3634 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3637 struct strbuf sb
= STRBUF_INIT
;
3639 switch (st
->st_mode
& S_IFMT
) {
3641 fd
= open(path
, O_RDONLY
);
3643 return error_errno("open(\"%s\")", path
);
3644 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3645 return error("%s: failed to insert into database",
3649 if (strbuf_readlink(&sb
, path
, st
->st_size
))
3650 return error_errno("readlink(\"%s\")", path
);
3651 if (!(flags
& HASH_WRITE_OBJECT
))
3652 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3653 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3654 return error("%s: failed to insert into database",
3656 strbuf_release(&sb
);
3659 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3661 return error("%s: unsupported file type", path
);
3666 int read_pack_header(int fd
, struct pack_header
*header
)
3668 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3669 /* "eof before pack header was fully read" */
3670 return PH_ERROR_EOF
;
3672 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3673 /* "protocol error (pack signature mismatch detected)" */
3674 return PH_ERROR_PACK_SIGNATURE
;
3675 if (!pack_version_ok(header
->hdr_version
))
3676 /* "protocol error (pack version unsupported)" */
3677 return PH_ERROR_PROTOCOL
;
3681 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3683 enum object_type type
= sha1_object_info(sha1
, NULL
);
3685 die("%s is not a valid object", sha1_to_hex(sha1
));
3687 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3691 int for_each_file_in_obj_subdir(unsigned int subdir_nr
,
3692 struct strbuf
*path
,
3693 each_loose_object_fn obj_cb
,
3694 each_loose_cruft_fn cruft_cb
,
3695 each_loose_subdir_fn subdir_cb
,
3698 size_t origlen
, baselen
;
3703 if (subdir_nr
> 0xff)
3704 BUG("invalid loose object subdirectory: %x", subdir_nr
);
3706 origlen
= path
->len
;
3707 strbuf_complete(path
, '/');
3708 strbuf_addf(path
, "%02x", subdir_nr
);
3709 baselen
= path
->len
;
3711 dir
= opendir(path
->buf
);
3713 if (errno
!= ENOENT
)
3714 r
= error_errno("unable to open %s", path
->buf
);
3715 strbuf_setlen(path
, origlen
);
3719 while ((de
= readdir(dir
))) {
3720 if (is_dot_or_dotdot(de
->d_name
))
3723 strbuf_setlen(path
, baselen
);
3724 strbuf_addf(path
, "/%s", de
->d_name
);
3726 if (strlen(de
->d_name
) == GIT_SHA1_HEXSZ
- 2) {
3727 char hex
[GIT_MAX_HEXSZ
+1];
3728 struct object_id oid
;
3730 xsnprintf(hex
, sizeof(hex
), "%02x%s",
3731 subdir_nr
, de
->d_name
);
3732 if (!get_oid_hex(hex
, &oid
)) {
3734 r
= obj_cb(&oid
, path
->buf
, data
);
3743 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3750 strbuf_setlen(path
, baselen
);
3751 if (!r
&& subdir_cb
)
3752 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3754 strbuf_setlen(path
, origlen
);
3759 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3760 each_loose_object_fn obj_cb
,
3761 each_loose_cruft_fn cruft_cb
,
3762 each_loose_subdir_fn subdir_cb
,
3768 for (i
= 0; i
< 256; i
++) {
3769 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3778 int for_each_loose_file_in_objdir(const char *path
,
3779 each_loose_object_fn obj_cb
,
3780 each_loose_cruft_fn cruft_cb
,
3781 each_loose_subdir_fn subdir_cb
,
3784 struct strbuf buf
= STRBUF_INIT
;
3787 strbuf_addstr(&buf
, path
);
3788 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3790 strbuf_release(&buf
);
3795 struct loose_alt_odb_data
{
3796 each_loose_object_fn
*cb
;
3800 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3803 struct loose_alt_odb_data
*data
= vdata
;
3804 struct strbuf buf
= STRBUF_INIT
;
3807 strbuf_addstr(&buf
, alt
->path
);
3808 r
= for_each_loose_file_in_objdir_buf(&buf
,
3809 data
->cb
, NULL
, NULL
,
3811 strbuf_release(&buf
);
3815 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3817 struct loose_alt_odb_data alt
;
3820 r
= for_each_loose_file_in_objdir(get_object_directory(),
3821 cb
, NULL
, NULL
, data
);
3825 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3830 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3833 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3838 for (i
= 0; i
< p
->num_objects
; i
++) {
3839 struct object_id oid
;
3841 if (!nth_packed_object_oid(&oid
, p
, i
))
3842 return error("unable to get sha1 of object %u in %s",
3845 r
= cb(&oid
, p
, i
, data
);
3852 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3854 struct packed_git
*p
;
3856 int pack_errors
= 0;
3858 prepare_packed_git();
3859 for (p
= packed_git
; p
; p
= p
->next
) {
3860 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3862 if (open_pack_index(p
)) {
3866 r
= for_each_object_in_pack(p
, cb
, data
);
3870 return r
? r
: pack_errors
;
3873 static int check_stream_sha1(git_zstream
*stream
,
3877 const unsigned char *expected_sha1
)
3880 unsigned char real_sha1
[GIT_MAX_RAWSZ
];
3881 unsigned char buf
[4096];
3882 unsigned long total_read
;
3886 git_SHA1_Update(&c
, hdr
, stream
->total_out
);
3889 * We already read some bytes into hdr, but the ones up to the NUL
3890 * do not count against the object's content size.
3892 total_read
= stream
->total_out
- strlen(hdr
) - 1;
3895 * This size comparison must be "<=" to read the final zlib packets;
3896 * see the comment in unpack_sha1_rest for details.
3898 while (total_read
<= size
&&
3899 (status
== Z_OK
|| status
== Z_BUF_ERROR
)) {
3900 stream
->next_out
= buf
;
3901 stream
->avail_out
= sizeof(buf
);
3902 if (size
- total_read
< stream
->avail_out
)
3903 stream
->avail_out
= size
- total_read
;
3904 status
= git_inflate(stream
, Z_FINISH
);
3905 git_SHA1_Update(&c
, buf
, stream
->next_out
- buf
);
3906 total_read
+= stream
->next_out
- buf
;
3908 git_inflate_end(stream
);
3910 if (status
!= Z_STREAM_END
) {
3911 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1
));
3914 if (stream
->avail_in
) {
3915 error("garbage at end of loose object '%s'",
3916 sha1_to_hex(expected_sha1
));
3920 git_SHA1_Final(real_sha1
, &c
);
3921 if (hashcmp(expected_sha1
, real_sha1
)) {
3922 error("sha1 mismatch for %s (expected %s)", path
,
3923 sha1_to_hex(expected_sha1
));
3930 int read_loose_object(const char *path
,
3931 const unsigned char *expected_sha1
,
3932 enum object_type
*type
,
3933 unsigned long *size
,
3938 unsigned long mapsize
;
3944 map
= map_sha1_file_1(path
, NULL
, &mapsize
);
3946 error_errno("unable to mmap %s", path
);
3950 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0) {
3951 error("unable to unpack header of %s", path
);
3955 *type
= parse_sha1_header(hdr
, size
);
3957 error("unable to parse header of %s", path
);
3958 git_inflate_end(&stream
);
3962 if (*type
== OBJ_BLOB
) {
3963 if (check_stream_sha1(&stream
, hdr
, *size
, path
, expected_sha1
) < 0)
3966 *contents
= unpack_sha1_rest(&stream
, hdr
, *size
, expected_sha1
);
3968 error("unable to unpack contents of %s", path
);
3969 git_inflate_end(&stream
);
3972 if (check_sha1_signature(expected_sha1
, *contents
,
3973 *size
, typename(*type
))) {
3974 error("sha1 mismatch for %s (expected %s)", path
,
3975 sha1_to_hex(expected_sha1
));
3981 ret
= 0; /* everything checks out */
3985 munmap(map
, mapsize
);