2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s
) { return (unsigned long)s
; }
33 static size_t sz_fmt(size_t s
) { return s
; }
36 const unsigned char null_sha1
[20];
38 static int git_open_noatime(const char *name
, struct packed_git
*p
);
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
46 static struct cached_object
{
47 unsigned char sha1
[20];
48 enum object_type type
;
52 static int cached_object_nr
, cached_object_alloc
;
54 static struct cached_object empty_tree
= {
55 EMPTY_TREE_SHA1_BIN_LITERAL
,
61 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
64 struct cached_object
*co
= cached_objects
;
66 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
67 if (!hashcmp(co
->sha1
, sha1
))
70 if (!hashcmp(sha1
, empty_tree
.sha1
))
75 int safe_create_leading_directories(char *path
)
77 char *pos
= path
+ offset_1st_component(path
);
81 pos
= strchr(pos
, '/');
89 if (!stat(path
, &st
)) {
91 if (!S_ISDIR(st
.st_mode
)) {
96 else if (mkdir(path
, 0777)) {
100 else if (adjust_shared_perm(path
)) {
109 int safe_create_leading_directories_const(const char *path
)
111 /* path points to cache entries, so xstrdup before messing with it */
112 char *buf
= xstrdup(path
);
113 int result
= safe_create_leading_directories(buf
);
118 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
121 for (i
= 0; i
< 20; i
++) {
122 static char hex
[] = "0123456789abcdef";
123 unsigned int val
= sha1
[i
];
124 char *pos
= pathbuf
+ i
*2 + (i
> 0);
125 *pos
++ = hex
[val
>> 4];
126 *pos
= hex
[val
& 0xf];
131 * NOTE! This returns a statically allocated buffer, so you have to be
132 * careful about using it. Do an "xstrdup()" if you need to save the
135 * Also note that this returns the location for creating. Reading
136 * SHA1 file can happen from any alternate directory listed in the
137 * DB_ENVIRONMENT environment variable if it is not found in
138 * the primary object database.
140 char *sha1_file_name(const unsigned char *sha1
)
142 static char buf
[PATH_MAX
];
146 objdir
= get_object_directory();
147 len
= strlen(objdir
);
149 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
150 if (len
+ 43 > PATH_MAX
)
151 die("insanely long object directory %s", objdir
);
152 memcpy(buf
, objdir
, len
);
156 fill_sha1_path(buf
+ len
+ 1, sha1
);
160 static char *sha1_get_pack_name(const unsigned char *sha1
,
161 char **name
, char **base
, const char *which
)
163 static const char hex
[] = "0123456789abcdef";
168 const char *sha1_file_directory
= get_object_directory();
169 int len
= strlen(sha1_file_directory
);
170 *base
= xmalloc(len
+ 60);
171 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
172 sha1_file_directory
, which
);
173 *name
= *base
+ len
+ 11;
178 for (i
= 0; i
< 20; i
++) {
179 unsigned int val
= *sha1
++;
180 *buf
++ = hex
[val
>> 4];
181 *buf
++ = hex
[val
& 0xf];
187 char *sha1_pack_name(const unsigned char *sha1
)
189 static char *name
, *base
;
191 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
194 char *sha1_pack_index_name(const unsigned char *sha1
)
196 static char *name
, *base
;
198 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
201 struct alternate_object_database
*alt_odb_list
;
202 static struct alternate_object_database
**alt_odb_tail
;
204 static void read_info_alternates(const char * alternates
, int depth
);
207 * Prepare alternate object database registry.
209 * The variable alt_odb_list points at the list of struct
210 * alternate_object_database. The elements on this list come from
211 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
212 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
213 * whose contents is similar to that environment variable but can be
214 * LF separated. Its base points at a statically allocated buffer that
215 * contains "/the/directory/corresponding/to/.git/objects/...", while
216 * its name points just after the slash at the end of ".git/objects/"
217 * in the example above, and has enough space to hold 40-byte hex
218 * SHA1, an extra slash for the first level indirection, and the
221 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
223 const char *objdir
= get_object_directory();
224 struct alternate_object_database
*ent
;
225 struct alternate_object_database
*alt
;
226 /* 43 = 40-byte + 2 '/' + terminating NUL */
228 int entlen
= pfxlen
+ 43;
231 if (!is_absolute_path(entry
) && relative_base
) {
232 /* Relative alt-odb */
234 base_len
= strlen(relative_base
) + 1;
238 ent
= xmalloc(sizeof(*ent
) + entlen
);
240 if (!is_absolute_path(entry
) && relative_base
) {
241 memcpy(ent
->base
, relative_base
, base_len
- 1);
242 ent
->base
[base_len
- 1] = '/';
243 memcpy(ent
->base
+ base_len
, entry
, len
);
246 memcpy(ent
->base
, entry
, pfxlen
);
248 ent
->name
= ent
->base
+ pfxlen
+ 1;
249 ent
->base
[pfxlen
+ 3] = '/';
250 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
252 /* Detect cases where alternate disappeared */
253 if (!is_directory(ent
->base
)) {
254 error("object directory %s does not exist; "
255 "check .git/objects/info/alternates.",
261 /* Prevent the common mistake of listing the same
262 * thing twice, or object directory itself.
264 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
265 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
270 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
275 /* add the alternate entry */
277 alt_odb_tail
= &(ent
->next
);
280 /* recursively add alternates */
281 read_info_alternates(ent
->base
, depth
+ 1);
283 ent
->base
[pfxlen
] = '/';
288 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
289 const char *relative_base
, int depth
)
291 const char *cp
, *last
;
294 error("%s: ignoring alternate object stores, nesting too deep.",
302 if (cp
< ep
&& *cp
== '#') {
303 while (cp
< ep
&& *cp
!= sep
)
308 while (cp
< ep
&& *cp
!= sep
)
311 if (!is_absolute_path(last
) && depth
) {
312 error("%s: ignoring relative alternate object store %s",
313 relative_base
, last
);
315 link_alt_odb_entry(last
, cp
- last
,
316 relative_base
, depth
);
319 while (cp
< ep
&& *cp
== sep
)
325 static void read_info_alternates(const char * relative_base
, int depth
)
330 const char alt_file_name
[] = "info/alternates";
331 /* Given that relative_base is no longer than PATH_MAX,
332 ensure that "path" has enough space to append "/", the
333 file name, "info/alternates", and a trailing NUL. */
334 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
337 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
338 fd
= git_open_noatime(path
, NULL
);
341 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
345 mapsz
= xsize_t(st
.st_size
);
346 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
349 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
354 void add_to_alternates_file(const char *reference
)
356 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
357 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
358 char *alt
= mkpath("%s/objects\n", reference
);
359 write_or_die(fd
, alt
, strlen(alt
));
360 if (commit_lock_file(lock
))
361 die("could not close alternates file");
363 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
366 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
368 struct alternate_object_database
*ent
;
371 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
376 void prepare_alt_odb(void)
383 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
386 alt_odb_tail
= &alt_odb_list
;
387 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
389 read_info_alternates(get_object_directory(), 0);
392 static int has_loose_object_local(const unsigned char *sha1
)
394 char *name
= sha1_file_name(sha1
);
395 return !access(name
, F_OK
);
398 int has_loose_object_nonlocal(const unsigned char *sha1
)
400 struct alternate_object_database
*alt
;
402 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
403 fill_sha1_path(alt
->name
, sha1
);
404 if (!access(alt
->base
, F_OK
))
410 static int has_loose_object(const unsigned char *sha1
)
412 return has_loose_object_local(sha1
) ||
413 has_loose_object_nonlocal(sha1
);
416 static unsigned int pack_used_ctr
;
417 static unsigned int pack_mmap_calls
;
418 static unsigned int peak_pack_open_windows
;
419 static unsigned int pack_open_windows
;
420 static size_t peak_pack_mapped
;
421 static size_t pack_mapped
;
422 struct packed_git
*packed_git
;
424 void pack_report(void)
427 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
428 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
429 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
430 sz_fmt(getpagesize()),
431 sz_fmt(packed_git_window_size
),
432 sz_fmt(packed_git_limit
));
434 "pack_report: pack_used_ctr = %10u\n"
435 "pack_report: pack_mmap_calls = %10u\n"
436 "pack_report: pack_open_windows = %10u / %10u\n"
437 "pack_report: pack_mapped = "
438 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
441 pack_open_windows
, peak_pack_open_windows
,
442 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
445 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
448 struct pack_idx_header
*hdr
;
450 uint32_t version
, nr
, i
, *index
;
451 int fd
= git_open_noatime(path
, p
);
456 if (fstat(fd
, &st
)) {
460 idx_size
= xsize_t(st
.st_size
);
461 if (idx_size
< 4 * 256 + 20 + 20) {
463 return error("index file %s is too small", path
);
465 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
469 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
470 version
= ntohl(hdr
->idx_version
);
471 if (version
< 2 || version
> 2) {
472 munmap(idx_map
, idx_size
);
473 return error("index file %s is version %"PRIu32
474 " and is not supported by this binary"
475 " (try upgrading GIT to a newer version)",
484 index
+= 2; /* skip index header */
485 for (i
= 0; i
< 256; i
++) {
486 uint32_t n
= ntohl(index
[i
]);
488 munmap(idx_map
, idx_size
);
489 return error("non-monotonic index %s", path
);
497 * - 256 index entries 4 bytes each
498 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
499 * - 20-byte SHA1 of the packfile
500 * - 20-byte SHA1 file checksum
502 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
503 munmap(idx_map
, idx_size
);
504 return error("wrong index v1 file size in %s", path
);
506 } else if (version
== 2) {
509 * - 8 bytes of header
510 * - 256 index entries 4 bytes each
511 * - 20-byte sha1 entry * nr
512 * - 4-byte crc entry * nr
513 * - 4-byte offset entry * nr
514 * - 20-byte SHA1 of the packfile
515 * - 20-byte SHA1 file checksum
516 * And after the 4-byte offset table might be a
517 * variable sized table containing 8-byte entries
518 * for offsets larger than 2^31.
520 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
521 unsigned long max_size
= min_size
;
523 max_size
+= (nr
- 1)*8;
524 if (idx_size
< min_size
|| idx_size
> max_size
) {
525 munmap(idx_map
, idx_size
);
526 return error("wrong index v2 file size in %s", path
);
528 if (idx_size
!= min_size
&&
530 * make sure we can deal with large pack offsets.
531 * 31-bit signed offset won't be enough, neither
532 * 32-bit unsigned one will be.
534 (sizeof(off_t
) <= 4)) {
535 munmap(idx_map
, idx_size
);
536 return error("pack too large for current definition of off_t in %s", path
);
540 p
->index_version
= version
;
541 p
->index_data
= idx_map
;
542 p
->index_size
= idx_size
;
547 int open_pack_index(struct packed_git
*p
)
555 idx_name
= xstrdup(p
->pack_name
);
556 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
557 ret
= check_packed_git_idx(idx_name
, p
);
562 static void scan_windows(struct packed_git
*p
,
563 struct packed_git
**lru_p
,
564 struct pack_window
**lru_w
,
565 struct pack_window
**lru_l
)
567 struct pack_window
*w
, *w_l
;
569 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
571 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
581 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
583 struct packed_git
*p
, *lru_p
= NULL
;
584 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
587 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
588 for (p
= packed_git
; p
; p
= p
->next
)
589 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
591 munmap(lru_w
->base
, lru_w
->len
);
592 pack_mapped
-= lru_w
->len
;
594 lru_l
->next
= lru_w
->next
;
596 lru_p
->windows
= lru_w
->next
;
597 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
598 close(lru_p
->pack_fd
);
609 void release_pack_memory(size_t need
, int fd
)
611 size_t cur
= pack_mapped
;
612 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
616 void *xmmap(void *start
, size_t length
,
617 int prot
, int flags
, int fd
, off_t offset
)
619 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
620 if (ret
== MAP_FAILED
) {
623 release_pack_memory(length
, fd
);
624 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
625 if (ret
== MAP_FAILED
)
626 die_errno("Out of memory? mmap failed");
631 void close_pack_windows(struct packed_git
*p
)
634 struct pack_window
*w
= p
->windows
;
637 die("pack '%s' still has open windows to it",
639 munmap(w
->base
, w
->len
);
640 pack_mapped
-= w
->len
;
642 p
->windows
= w
->next
;
647 void unuse_pack(struct pack_window
**w_cursor
)
649 struct pack_window
*w
= *w_cursor
;
656 void close_pack_index(struct packed_git
*p
)
659 munmap((void *)p
->index_data
, p
->index_size
);
660 p
->index_data
= NULL
;
665 * This is used by git-repack in case a newly created pack happens to
666 * contain the same set of objects as an existing one. In that case
667 * the resulting file might be different even if its name would be the
668 * same. It is best to close any reference to the old pack before it is
669 * replaced on disk. Of course no index pointers nor windows for given pack
670 * must subsist at this point. If ever objects from this pack are requested
671 * again, the new version of the pack will be reinitialized through
672 * reprepare_packed_git().
674 void free_pack_by_name(const char *pack_name
)
676 struct packed_git
*p
, **pp
= &packed_git
;
680 if (strcmp(pack_name
, p
->pack_name
) == 0) {
681 clear_delta_base_cache();
682 close_pack_windows(p
);
683 if (p
->pack_fd
!= -1)
686 free(p
->bad_object_sha1
);
696 * Do not call this directly as this leaks p->pack_fd on error return;
697 * call open_packed_git() instead.
699 static int open_packed_git_1(struct packed_git
*p
)
702 struct pack_header hdr
;
703 unsigned char sha1
[20];
704 unsigned char *idx_sha1
;
707 if (!p
->index_data
&& open_pack_index(p
))
708 return error("packfile %s index unavailable", p
->pack_name
);
710 p
->pack_fd
= git_open_noatime(p
->pack_name
, p
);
711 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
714 /* If we created the struct before we had the pack we lack size. */
716 if (!S_ISREG(st
.st_mode
))
717 return error("packfile %s not a regular file", p
->pack_name
);
718 p
->pack_size
= st
.st_size
;
719 } else if (p
->pack_size
!= st
.st_size
)
720 return error("packfile %s size changed", p
->pack_name
);
722 /* We leave these file descriptors open with sliding mmap;
723 * there is no point keeping them open across exec(), though.
725 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
727 return error("cannot determine file descriptor flags");
728 fd_flag
|= FD_CLOEXEC
;
729 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
730 return error("cannot set FD_CLOEXEC");
732 /* Verify we recognize this pack file format. */
733 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
734 return error("file %s is far too short to be a packfile", p
->pack_name
);
735 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
736 return error("file %s is not a GIT packfile", p
->pack_name
);
737 if (!pack_version_ok(hdr
.hdr_version
))
738 return error("packfile %s is version %"PRIu32
" and not"
739 " supported (try upgrading GIT to a newer version)",
740 p
->pack_name
, ntohl(hdr
.hdr_version
));
742 /* Verify the pack matches its index. */
743 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
744 return error("packfile %s claims to have %"PRIu32
" objects"
745 " while index indicates %"PRIu32
" objects",
746 p
->pack_name
, ntohl(hdr
.hdr_entries
),
748 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
749 return error("end of packfile %s is unavailable", p
->pack_name
);
750 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
751 return error("packfile %s signature is unavailable", p
->pack_name
);
752 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
753 if (hashcmp(sha1
, idx_sha1
))
754 return error("packfile %s does not match index", p
->pack_name
);
758 static int open_packed_git(struct packed_git
*p
)
760 if (!open_packed_git_1(p
))
762 if (p
->pack_fd
!= -1) {
769 static int in_window(struct pack_window
*win
, off_t offset
)
771 /* We must promise at least 20 bytes (one hash) after the
772 * offset is available from this window, otherwise the offset
773 * is not actually in this window and a different window (which
774 * has that one hash excess) must be used. This is to support
775 * the object header and delta base parsing routines below.
777 off_t win_off
= win
->offset
;
778 return win_off
<= offset
779 && (offset
+ 20) <= (win_off
+ win
->len
);
782 unsigned char *use_pack(struct packed_git
*p
,
783 struct pack_window
**w_cursor
,
787 struct pack_window
*win
= *w_cursor
;
789 if (p
->pack_fd
== -1 && open_packed_git(p
))
790 die("packfile %s cannot be accessed", p
->pack_name
);
792 /* Since packfiles end in a hash of their content and it's
793 * pointless to ask for an offset into the middle of that
794 * hash, and the in_window function above wouldn't match
795 * don't allow an offset too close to the end of the file.
797 if (offset
> (p
->pack_size
- 20))
798 die("offset beyond end of packfile (truncated pack?)");
800 if (!win
|| !in_window(win
, offset
)) {
803 for (win
= p
->windows
; win
; win
= win
->next
) {
804 if (in_window(win
, offset
))
808 size_t window_align
= packed_git_window_size
/ 2;
810 win
= xcalloc(1, sizeof(*win
));
811 win
->offset
= (offset
/ window_align
) * window_align
;
812 len
= p
->pack_size
- win
->offset
;
813 if (len
> packed_git_window_size
)
814 len
= packed_git_window_size
;
815 win
->len
= (size_t)len
;
816 pack_mapped
+= win
->len
;
817 while (packed_git_limit
< pack_mapped
818 && unuse_one_window(p
, p
->pack_fd
))
820 win
->base
= xmmap(NULL
, win
->len
,
821 PROT_READ
, MAP_PRIVATE
,
822 p
->pack_fd
, win
->offset
);
823 if (win
->base
== MAP_FAILED
)
824 die("packfile %s cannot be mapped: %s",
829 if (pack_mapped
> peak_pack_mapped
)
830 peak_pack_mapped
= pack_mapped
;
831 if (pack_open_windows
> peak_pack_open_windows
)
832 peak_pack_open_windows
= pack_open_windows
;
833 win
->next
= p
->windows
;
837 if (win
!= *w_cursor
) {
838 win
->last_used
= pack_used_ctr
++;
842 offset
-= win
->offset
;
844 *left
= win
->len
- xsize_t(offset
);
845 return win
->base
+ offset
;
848 static struct packed_git
*alloc_packed_git(int extra
)
850 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
851 memset(p
, 0, sizeof(*p
));
856 static void try_to_free_pack_memory(size_t size
)
858 release_pack_memory(size
, -1);
861 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
863 static int have_set_try_to_free_routine
;
865 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
867 if (!have_set_try_to_free_routine
) {
868 have_set_try_to_free_routine
= 1;
869 set_try_to_free_routine(try_to_free_pack_memory
);
873 * Make sure a corresponding .pack file exists and that
874 * the index looks sane.
876 path_len
-= strlen(".idx");
881 memcpy(p
->pack_name
, path
, path_len
);
883 strcpy(p
->pack_name
+ path_len
, ".keep");
884 if (!access(p
->pack_name
, F_OK
))
887 strcpy(p
->pack_name
+ path_len
, ".pack");
888 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
893 /* ok, it looks sane as far as we can check without
894 * actually mapping the pack file.
896 p
->pack_size
= st
.st_size
;
897 p
->pack_local
= local
;
898 p
->mtime
= st
.st_mtime
;
899 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
904 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
906 const char *path
= sha1_pack_name(sha1
);
907 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
909 strcpy(p
->pack_name
, path
);
910 hashcpy(p
->sha1
, sha1
);
911 if (check_packed_git_idx(idx_path
, p
)) {
919 void install_packed_git(struct packed_git
*pack
)
921 pack
->next
= packed_git
;
925 static void prepare_packed_git_one(char *objdir
, int local
)
927 /* Ensure that this buffer is large enough so that we can
928 append "/pack/" without clobbering the stack even if
929 strlen(objdir) were PATH_MAX. */
930 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
935 sprintf(path
, "%s/pack", objdir
);
938 while (!dir
&& errno
== EMFILE
&& unuse_one_window(NULL
, -1))
942 error("unable to open object pack directory: %s: %s",
943 path
, strerror(errno
));
947 while ((de
= readdir(dir
)) != NULL
) {
948 int namelen
= strlen(de
->d_name
);
949 struct packed_git
*p
;
951 if (!has_extension(de
->d_name
, ".idx"))
954 if (len
+ namelen
+ 1 > sizeof(path
))
957 /* Don't reopen a pack we already have. */
958 strcpy(path
+ len
, de
->d_name
);
959 for (p
= packed_git
; p
; p
= p
->next
) {
960 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
965 /* See if it really is a valid .idx file with corresponding
966 * .pack file that we can map.
968 p
= add_packed_git(path
, len
+ namelen
, local
);
971 install_packed_git(p
);
976 static int sort_pack(const void *a_
, const void *b_
)
978 struct packed_git
*a
= *((struct packed_git
**)a_
);
979 struct packed_git
*b
= *((struct packed_git
**)b_
);
983 * Local packs tend to contain objects specific to our
984 * variant of the project than remote ones. In addition,
985 * remote ones could be on a network mounted filesystem.
986 * Favor local ones for these reasons.
988 st
= a
->pack_local
- b
->pack_local
;
993 * Younger packs tend to contain more recent objects,
994 * and more recent objects tend to get accessed more
997 if (a
->mtime
< b
->mtime
)
999 else if (a
->mtime
== b
->mtime
)
1004 static void rearrange_packed_git(void)
1006 struct packed_git
**ary
, *p
;
1009 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1014 /* prepare an array of packed_git for easier sorting */
1015 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1016 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1019 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1021 /* link them back again */
1022 for (i
= 0; i
< n
- 1; i
++)
1023 ary
[i
]->next
= ary
[i
+ 1];
1024 ary
[n
- 1]->next
= NULL
;
1025 packed_git
= ary
[0];
1030 static int prepare_packed_git_run_once
= 0;
1031 void prepare_packed_git(void)
1033 struct alternate_object_database
*alt
;
1035 if (prepare_packed_git_run_once
)
1037 prepare_packed_git_one(get_object_directory(), 1);
1039 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1041 prepare_packed_git_one(alt
->base
, 0);
1042 alt
->name
[-1] = '/';
1044 rearrange_packed_git();
1045 prepare_packed_git_run_once
= 1;
1048 void reprepare_packed_git(void)
1051 prepare_packed_git_run_once
= 0;
1052 prepare_packed_git();
1055 static void mark_bad_packed_object(struct packed_git
*p
,
1056 const unsigned char *sha1
)
1059 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1060 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1062 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1063 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1064 p
->num_bad_objects
++;
1067 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1069 struct packed_git
*p
;
1072 for (p
= packed_git
; p
; p
= p
->next
)
1073 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1074 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1079 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1081 unsigned char real_sha1
[20];
1082 hash_sha1_file(map
, size
, type
, real_sha1
);
1083 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1086 static int git_open_noatime(const char *name
, struct packed_git
*p
)
1088 static int sha1_file_open_flag
= O_NOATIME
;
1091 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1095 /* Might the failure be insufficient file descriptors? */
1096 if (errno
== EMFILE
) {
1097 if (unuse_one_window(p
, -1))
1103 /* Might the failure be due to O_NOATIME? */
1104 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1105 sha1_file_open_flag
= 0;
1113 static int open_sha1_file(const unsigned char *sha1
)
1116 char *name
= sha1_file_name(sha1
);
1117 struct alternate_object_database
*alt
;
1119 fd
= git_open_noatime(name
, NULL
);
1125 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1127 fill_sha1_path(name
, sha1
);
1128 fd
= git_open_noatime(alt
->base
, NULL
);
1135 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1140 fd
= open_sha1_file(sha1
);
1145 if (!fstat(fd
, &st
)) {
1146 *size
= xsize_t(st
.st_size
);
1147 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1154 static int legacy_loose_object(unsigned char *map
)
1159 * Is it a zlib-compressed buffer? If so, the first byte
1160 * must be 0x78 (15-bit window size, deflated), and the
1161 * first 16-bit word is evenly divisible by 31
1163 word
= (map
[0] << 8) + map
[1];
1164 if (map
[0] == 0x78 && !(word
% 31))
1170 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1171 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1174 unsigned long size
, c
;
1175 unsigned long used
= 0;
1178 *type
= (c
>> 4) & 7;
1182 if (len
<= used
|| bitsizeof(long) <= shift
) {
1183 error("bad object header");
1187 size
+= (c
& 0x7f) << shift
;
1194 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1196 unsigned long size
, used
;
1197 static const char valid_loose_object_type
[8] = {
1199 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1200 0, /* "delta" and others are invalid in a loose object */
1202 enum object_type type
;
1204 /* Get the data stream */
1205 memset(stream
, 0, sizeof(*stream
));
1206 stream
->next_in
= map
;
1207 stream
->avail_in
= mapsize
;
1208 stream
->next_out
= buffer
;
1209 stream
->avail_out
= bufsiz
;
1211 if (legacy_loose_object(map
)) {
1212 git_inflate_init(stream
);
1213 return git_inflate(stream
, 0);
1218 * There used to be a second loose object header format which
1219 * was meant to mimic the in-pack format, allowing for direct
1220 * copy of the object data. This format turned up not to be
1221 * really worth it and we don't write it any longer. But we
1222 * can still read it.
1224 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1225 if (!used
|| !valid_loose_object_type
[type
])
1230 /* Set up the stream for the rest.. */
1231 stream
->next_in
= map
;
1232 stream
->avail_in
= mapsize
;
1233 git_inflate_init(stream
);
1235 /* And generate the fake traditional header */
1236 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1237 typename(type
), size
);
1241 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1243 int bytes
= strlen(buffer
) + 1;
1244 unsigned char *buf
= xmallocz(size
);
1248 n
= stream
->total_out
- bytes
;
1251 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1253 if (bytes
<= size
) {
1255 * The above condition must be (bytes <= size), not
1256 * (bytes < size). In other words, even though we
1257 * expect no more output and set avail_out to zer0,
1258 * the input zlib stream may have bytes that express
1259 * "this concludes the stream", and we *do* want to
1262 * Otherwise we would not be able to test that we
1263 * consumed all the input to reach the expected size;
1264 * we also want to check that zlib tells us that all
1265 * went well with status == Z_STREAM_END at the end.
1267 stream
->next_out
= buf
+ bytes
;
1268 stream
->avail_out
= size
- bytes
;
1269 while (status
== Z_OK
)
1270 status
= git_inflate(stream
, Z_FINISH
);
1272 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1273 git_inflate_end(stream
);
1278 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1279 else if (stream
->avail_in
)
1280 error("garbage at end of loose object '%s'",
1287 * We used to just use "sscanf()", but that's actually way
1288 * too permissive for what we want to check. So do an anal
1289 * object header parse by hand.
1291 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1298 * The type can be at most ten bytes (including the
1299 * terminating '\0' that we add), and is followed by
1308 if (i
>= sizeof(type
))
1314 * The length must follow immediately, and be in canonical
1315 * decimal format (ie "010" is not valid).
1317 size
= *hdr
++ - '0';
1322 unsigned long c
= *hdr
- '0';
1326 size
= size
* 10 + c
;
1332 * The length must be followed by a zero byte
1334 return *hdr
? -1 : type_from_string(type
);
1337 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1343 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1344 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1347 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1350 unsigned long get_size_from_delta(struct packed_git
*p
,
1351 struct pack_window
**w_curs
,
1354 const unsigned char *data
;
1355 unsigned char delta_head
[20], *in
;
1359 memset(&stream
, 0, sizeof(stream
));
1360 stream
.next_out
= delta_head
;
1361 stream
.avail_out
= sizeof(delta_head
);
1363 git_inflate_init(&stream
);
1365 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1366 stream
.next_in
= in
;
1367 st
= git_inflate(&stream
, Z_FINISH
);
1368 curpos
+= stream
.next_in
- in
;
1369 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1370 stream
.total_out
< sizeof(delta_head
));
1371 git_inflate_end(&stream
);
1372 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1373 error("delta data unpack-initial failed");
1377 /* Examine the initial part of the delta to figure out
1382 /* ignore base size */
1383 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1385 /* Read the result size */
1386 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1389 static off_t
get_delta_base(struct packed_git
*p
,
1390 struct pack_window
**w_curs
,
1392 enum object_type type
,
1393 off_t delta_obj_offset
)
1395 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1398 /* use_pack() assured us we have [base_info, base_info + 20)
1399 * as a range that we can look at without walking off the
1400 * end of the mapped window. Its actually the hash size
1401 * that is assured. An OFS_DELTA longer than the hash size
1402 * is stupid, as then a REF_DELTA would be smaller to store.
1404 if (type
== OBJ_OFS_DELTA
) {
1406 unsigned char c
= base_info
[used
++];
1407 base_offset
= c
& 127;
1410 if (!base_offset
|| MSB(base_offset
, 7))
1411 return 0; /* overflow */
1412 c
= base_info
[used
++];
1413 base_offset
= (base_offset
<< 7) + (c
& 127);
1415 base_offset
= delta_obj_offset
- base_offset
;
1416 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1417 return 0; /* out of bound */
1419 } else if (type
== OBJ_REF_DELTA
) {
1420 /* The base entry _must_ be in the same pack */
1421 base_offset
= find_pack_entry_one(base_info
, p
);
1424 die("I am totally screwed");
1428 /* forward declaration for a mutually recursive function */
1429 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1430 unsigned long *sizep
);
1432 static int packed_delta_info(struct packed_git
*p
,
1433 struct pack_window
**w_curs
,
1435 enum object_type type
,
1437 unsigned long *sizep
)
1441 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1444 type
= packed_object_info(p
, base_offset
, NULL
);
1445 if (type
<= OBJ_NONE
) {
1446 struct revindex_entry
*revidx
;
1447 const unsigned char *base_sha1
;
1448 revidx
= find_pack_revindex(p
, base_offset
);
1451 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1452 mark_bad_packed_object(p
, base_sha1
);
1453 type
= sha1_object_info(base_sha1
, NULL
);
1454 if (type
<= OBJ_NONE
)
1458 /* We choose to only get the type of the base object and
1459 * ignore potentially corrupt pack file that expects the delta
1460 * based on a base with a wrong size. This saves tons of
1464 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1472 static int unpack_object_header(struct packed_git
*p
,
1473 struct pack_window
**w_curs
,
1475 unsigned long *sizep
)
1477 unsigned char *base
;
1480 enum object_type type
;
1482 /* use_pack() assures us we have [base, base + 20) available
1483 * as a range that we can look at at. (Its actually the hash
1484 * size that is assured.) With our object header encoding
1485 * the maximum deflated object size is 2^137, which is just
1486 * insane, so we know won't exceed what we have been given.
1488 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1489 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1498 const char *packed_object_info_detail(struct packed_git
*p
,
1500 unsigned long *size
,
1501 unsigned long *store_size
,
1502 unsigned int *delta_chain_length
,
1503 unsigned char *base_sha1
)
1505 struct pack_window
*w_curs
= NULL
;
1507 unsigned long dummy
;
1508 unsigned char *next_sha1
;
1509 enum object_type type
;
1510 struct revindex_entry
*revidx
;
1512 *delta_chain_length
= 0;
1513 curpos
= obj_offset
;
1514 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1516 revidx
= find_pack_revindex(p
, obj_offset
);
1517 *store_size
= revidx
[1].offset
- obj_offset
;
1522 die("pack %s contains unknown object type %d",
1523 p
->pack_name
, type
);
1528 unuse_pack(&w_curs
);
1529 return typename(type
);
1531 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1533 die("pack %s contains bad delta base reference of type %s",
1534 p
->pack_name
, typename(type
));
1535 if (*delta_chain_length
== 0) {
1536 revidx
= find_pack_revindex(p
, obj_offset
);
1537 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1541 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1542 if (*delta_chain_length
== 0)
1543 hashcpy(base_sha1
, next_sha1
);
1544 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1547 (*delta_chain_length
)++;
1548 curpos
= obj_offset
;
1549 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1553 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1554 unsigned long *sizep
)
1556 struct pack_window
*w_curs
= NULL
;
1558 off_t curpos
= obj_offset
;
1559 enum object_type type
;
1561 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1566 type
= packed_delta_info(p
, &w_curs
, curpos
,
1567 type
, obj_offset
, sizep
);
1577 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1578 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1581 unuse_pack(&w_curs
);
1585 static void *unpack_compressed_entry(struct packed_git
*p
,
1586 struct pack_window
**w_curs
,
1592 unsigned char *buffer
, *in
;
1594 buffer
= xmallocz(size
);
1595 memset(&stream
, 0, sizeof(stream
));
1596 stream
.next_out
= buffer
;
1597 stream
.avail_out
= size
+ 1;
1599 git_inflate_init(&stream
);
1601 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1602 stream
.next_in
= in
;
1603 st
= git_inflate(&stream
, Z_FINISH
);
1604 if (!stream
.avail_out
)
1605 break; /* the payload is larger than it should be */
1606 curpos
+= stream
.next_in
- in
;
1607 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1608 git_inflate_end(&stream
);
1609 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1617 #define MAX_DELTA_CACHE (256)
1619 static size_t delta_base_cached
;
1621 static struct delta_base_cache_lru_list
{
1622 struct delta_base_cache_lru_list
*prev
;
1623 struct delta_base_cache_lru_list
*next
;
1624 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1626 static struct delta_base_cache_entry
{
1627 struct delta_base_cache_lru_list lru
;
1629 struct packed_git
*p
;
1632 enum object_type type
;
1633 } delta_base_cache
[MAX_DELTA_CACHE
];
1635 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1639 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1640 hash
+= (hash
>> 8) + (hash
>> 16);
1641 return hash
% MAX_DELTA_CACHE
;
1644 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1645 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1648 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1649 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1652 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1653 return unpack_entry(p
, base_offset
, type
, base_size
);
1657 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1658 ent
->lru
.prev
->next
= ent
->lru
.next
;
1659 delta_base_cached
-= ent
->size
;
1661 ret
= xmemdupz(ent
->data
, ent
->size
);
1664 *base_size
= ent
->size
;
1668 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1673 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1674 ent
->lru
.prev
->next
= ent
->lru
.next
;
1675 delta_base_cached
-= ent
->size
;
1679 void clear_delta_base_cache(void)
1682 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1683 release_delta_base_cache(&delta_base_cache
[p
]);
1686 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1687 void *base
, unsigned long base_size
, enum object_type type
)
1689 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1690 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1691 struct delta_base_cache_lru_list
*lru
;
1693 release_delta_base_cache(ent
);
1694 delta_base_cached
+= base_size
;
1696 for (lru
= delta_base_cache_lru
.next
;
1697 delta_base_cached
> delta_base_cache_limit
1698 && lru
!= &delta_base_cache_lru
;
1700 struct delta_base_cache_entry
*f
= (void *)lru
;
1701 if (f
->type
== OBJ_BLOB
)
1702 release_delta_base_cache(f
);
1704 for (lru
= delta_base_cache_lru
.next
;
1705 delta_base_cached
> delta_base_cache_limit
1706 && lru
!= &delta_base_cache_lru
;
1708 struct delta_base_cache_entry
*f
= (void *)lru
;
1709 release_delta_base_cache(f
);
1713 ent
->base_offset
= base_offset
;
1716 ent
->size
= base_size
;
1717 ent
->lru
.next
= &delta_base_cache_lru
;
1718 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1719 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1720 delta_base_cache_lru
.prev
= &ent
->lru
;
1723 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1724 unsigned long *size
);
1726 static void *unpack_delta_entry(struct packed_git
*p
,
1727 struct pack_window
**w_curs
,
1729 unsigned long delta_size
,
1731 enum object_type
*type
,
1732 unsigned long *sizep
)
1734 void *delta_data
, *result
, *base
;
1735 unsigned long base_size
;
1738 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1740 error("failed to validate delta base reference "
1741 "at offset %"PRIuMAX
" from %s",
1742 (uintmax_t)curpos
, p
->pack_name
);
1746 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1749 * We're probably in deep shit, but let's try to fetch
1750 * the required base anyway from another pack or loose.
1751 * This is costly but should happen only in the presence
1752 * of a corrupted pack, and is better than failing outright.
1754 struct revindex_entry
*revidx
;
1755 const unsigned char *base_sha1
;
1756 revidx
= find_pack_revindex(p
, base_offset
);
1759 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1760 error("failed to read delta base object %s"
1761 " at offset %"PRIuMAX
" from %s",
1762 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1764 mark_bad_packed_object(p
, base_sha1
);
1765 base
= read_object(base_sha1
, type
, &base_size
);
1770 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1772 error("failed to unpack compressed delta "
1773 "at offset %"PRIuMAX
" from %s",
1774 (uintmax_t)curpos
, p
->pack_name
);
1778 result
= patch_delta(base
, base_size
,
1779 delta_data
, delta_size
,
1782 die("failed to apply delta");
1784 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1788 int do_check_packed_object_crc
;
1790 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1791 enum object_type
*type
, unsigned long *sizep
)
1793 struct pack_window
*w_curs
= NULL
;
1794 off_t curpos
= obj_offset
;
1797 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1798 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1799 unsigned long len
= revidx
[1].offset
- obj_offset
;
1800 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1801 const unsigned char *sha1
=
1802 nth_packed_object_sha1(p
, revidx
->nr
);
1803 error("bad packed object CRC for %s",
1805 mark_bad_packed_object(p
, sha1
);
1806 unuse_pack(&w_curs
);
1811 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1815 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1816 obj_offset
, type
, sizep
);
1822 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1826 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1827 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1829 unuse_pack(&w_curs
);
1833 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1836 const unsigned char *index
= p
->index_data
;
1838 if (open_pack_index(p
))
1840 index
= p
->index_data
;
1842 if (n
>= p
->num_objects
)
1845 if (p
->index_version
== 1) {
1846 return index
+ 24 * n
+ 4;
1849 return index
+ 20 * n
;
1853 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1855 const unsigned char *index
= p
->index_data
;
1857 if (p
->index_version
== 1) {
1858 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1861 index
+= 8 + p
->num_objects
* (20 + 4);
1862 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1863 if (!(off
& 0x80000000))
1865 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1866 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1867 ntohl(*((uint32_t *)(index
+ 4)));
1871 off_t
find_pack_entry_one(const unsigned char *sha1
,
1872 struct packed_git
*p
)
1874 const uint32_t *level1_ofs
= p
->index_data
;
1875 const unsigned char *index
= p
->index_data
;
1876 unsigned hi
, lo
, stride
;
1877 static int use_lookup
= -1;
1878 static int debug_lookup
= -1;
1880 if (debug_lookup
< 0)
1881 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1884 if (open_pack_index(p
))
1886 level1_ofs
= p
->index_data
;
1887 index
= p
->index_data
;
1889 if (p
->index_version
> 1) {
1894 hi
= ntohl(level1_ofs
[*sha1
]);
1895 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1896 if (p
->index_version
> 1) {
1904 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1905 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1908 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1910 int pos
= sha1_entry_pos(index
, stride
, 0,
1911 lo
, hi
, p
->num_objects
, sha1
);
1914 return nth_packed_object_offset(p
, pos
);
1918 unsigned mi
= (lo
+ hi
) / 2;
1919 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1922 printf("lo %u hi %u rg %u mi %u\n",
1923 lo
, hi
, hi
- lo
, mi
);
1925 return nth_packed_object_offset(p
, mi
);
1934 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1936 static struct packed_git
*last_found
= (void *)1;
1937 struct packed_git
*p
;
1940 prepare_packed_git();
1943 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1946 if (p
->num_bad_objects
) {
1948 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1949 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1953 offset
= find_pack_entry_one(sha1
, p
);
1956 * We are about to tell the caller where they can
1957 * locate the requested object. We better make
1958 * sure the packfile is still here and can be
1959 * accessed before supplying that answer, as
1960 * it may have been deleted since the index
1963 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1964 error("packfile %s cannot be accessed", p
->pack_name
);
1969 hashcpy(e
->sha1
, sha1
);
1975 if (p
== last_found
)
1979 if (p
== last_found
)
1985 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1986 struct packed_git
*packs
)
1988 struct packed_git
*p
;
1990 for (p
= packs
; p
; p
= p
->next
) {
1991 if (find_pack_entry_one(sha1
, p
))
1998 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2001 unsigned long mapsize
, size
;
2006 map
= map_sha1_file(sha1
, &mapsize
);
2008 return error("unable to find %s", sha1_to_hex(sha1
));
2009 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2010 status
= error("unable to unpack %s header",
2012 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2013 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2016 git_inflate_end(&stream
);
2017 munmap(map
, mapsize
);
2021 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2023 struct cached_object
*co
;
2024 struct pack_entry e
;
2027 co
= find_cached_object(sha1
);
2034 if (!find_pack_entry(sha1
, &e
)) {
2035 /* Most likely it's a loose object. */
2036 status
= sha1_loose_object_info(sha1
, sizep
);
2040 /* Not a loose object; someone else may have just packed it. */
2041 reprepare_packed_git();
2042 if (!find_pack_entry(sha1
, &e
))
2046 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
2048 mark_bad_packed_object(e
.p
, sha1
);
2049 status
= sha1_object_info(sha1
, sizep
);
2055 static void *read_packed_sha1(const unsigned char *sha1
,
2056 enum object_type
*type
, unsigned long *size
)
2058 struct pack_entry e
;
2061 if (!find_pack_entry(sha1
, &e
))
2063 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2066 * We're probably in deep shit, but let's try to fetch
2067 * the required object anyway from another pack or loose.
2068 * This should happen only in the presence of a corrupted
2069 * pack, and is better than failing outright.
2071 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2072 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2073 mark_bad_packed_object(e
.p
, sha1
);
2074 data
= read_object(sha1
, type
, size
);
2079 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2080 unsigned char *sha1
)
2082 struct cached_object
*co
;
2084 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2085 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2087 if (cached_object_alloc
<= cached_object_nr
) {
2088 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2089 cached_objects
= xrealloc(cached_objects
,
2090 sizeof(*cached_objects
) *
2091 cached_object_alloc
);
2093 co
= &cached_objects
[cached_object_nr
++];
2096 co
->buf
= xmalloc(len
);
2097 memcpy(co
->buf
, buf
, len
);
2098 hashcpy(co
->sha1
, sha1
);
2102 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2103 unsigned long *size
)
2105 unsigned long mapsize
;
2107 struct cached_object
*co
;
2109 co
= find_cached_object(sha1
);
2113 return xmemdupz(co
->buf
, co
->size
);
2116 buf
= read_packed_sha1(sha1
, type
, size
);
2119 map
= map_sha1_file(sha1
, &mapsize
);
2121 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2122 munmap(map
, mapsize
);
2125 reprepare_packed_git();
2126 return read_packed_sha1(sha1
, type
, size
);
2130 * This function dies on corrupt objects; the callers who want to
2131 * deal with them should arrange to call read_object() and give error
2132 * messages themselves.
2134 void *read_sha1_file_repl(const unsigned char *sha1
,
2135 enum object_type
*type
,
2136 unsigned long *size
,
2137 const unsigned char **replacement
)
2139 const unsigned char *repl
= lookup_replace_object(sha1
);
2142 const struct packed_git
*p
;
2145 data
= read_object(repl
, type
, size
);
2148 *replacement
= repl
;
2152 if (errno
&& errno
!= ENOENT
)
2153 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2155 /* die if we replaced an object with one that does not exist */
2157 die("replacement %s not found for %s",
2158 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2160 if (has_loose_object(repl
)) {
2161 path
= sha1_file_name(sha1
);
2162 die("loose object %s (stored in %s) is corrupt",
2163 sha1_to_hex(repl
), path
);
2166 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2167 die("packed object %s (stored in %s) is corrupt",
2168 sha1_to_hex(repl
), p
->pack_name
);
2173 void *read_object_with_reference(const unsigned char *sha1
,
2174 const char *required_type_name
,
2175 unsigned long *size
,
2176 unsigned char *actual_sha1_return
)
2178 enum object_type type
, required_type
;
2180 unsigned long isize
;
2181 unsigned char actual_sha1
[20];
2183 required_type
= type_from_string(required_type_name
);
2184 hashcpy(actual_sha1
, sha1
);
2186 int ref_length
= -1;
2187 const char *ref_type
= NULL
;
2189 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2192 if (type
== required_type
) {
2194 if (actual_sha1_return
)
2195 hashcpy(actual_sha1_return
, actual_sha1
);
2198 /* Handle references */
2199 else if (type
== OBJ_COMMIT
)
2201 else if (type
== OBJ_TAG
)
2202 ref_type
= "object ";
2207 ref_length
= strlen(ref_type
);
2209 if (ref_length
+ 40 > isize
||
2210 memcmp(buffer
, ref_type
, ref_length
) ||
2211 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2216 /* Now we have the ID of the referred-to object in
2217 * actual_sha1. Check again. */
2221 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2222 const char *type
, unsigned char *sha1
,
2223 char *hdr
, int *hdrlen
)
2227 /* Generate the header */
2228 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2232 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2233 git_SHA1_Update(&c
, buf
, len
);
2234 git_SHA1_Final(sha1
, &c
);
2238 * Move the just written object into its final resting place.
2239 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2240 * "moving" is only a part of what it does, when no patch between
2241 * master to pu changes the call sites of this function.
2243 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2247 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2249 else if (link(tmpfile
, filename
))
2253 * Coda hack - coda doesn't like cross-directory links,
2254 * so we fall back to a rename, which will mean that it
2255 * won't be able to check collisions, but that's not a
2258 * The same holds for FAT formatted media.
2260 * When this succeeds, we just return. We have nothing
2263 if (ret
&& ret
!= EEXIST
) {
2265 if (!rename(tmpfile
, filename
))
2269 unlink_or_warn(tmpfile
);
2271 if (ret
!= EEXIST
) {
2272 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2274 /* FIXME!!! Collision check here ? */
2278 if (adjust_shared_perm(filename
))
2279 return error("unable to set permission to '%s'", filename
);
2283 static int write_buffer(int fd
, const void *buf
, size_t len
)
2285 if (write_in_full(fd
, buf
, len
) < 0)
2286 return error("file write error (%s)", strerror(errno
));
2290 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2291 unsigned char *sha1
)
2295 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2299 /* Finalize a file on disk, and close it. */
2300 static void close_sha1_file(int fd
)
2302 if (fsync_object_files
)
2303 fsync_or_die(fd
, "sha1 file");
2305 die_errno("error when closing sha1 file");
2308 /* Size of directory component, including the ending '/' */
2309 static inline int directory_size(const char *filename
)
2311 const char *s
= strrchr(filename
, '/');
2314 return s
- filename
+ 1;
2318 * This creates a temporary file in the same directory as the final
2321 * We want to avoid cross-directory filename renames, because those
2322 * can have problems on various filesystems (FAT, NFS, Coda).
2324 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2326 int fd
, dirlen
= directory_size(filename
);
2328 if (dirlen
+ 20 > bufsiz
) {
2329 errno
= ENAMETOOLONG
;
2332 memcpy(buffer
, filename
, dirlen
);
2333 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2334 fd
= git_mkstemp_mode(buffer
, 0444);
2335 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2336 /* Make sure the directory exists */
2337 memcpy(buffer
, filename
, dirlen
);
2338 buffer
[dirlen
-1] = 0;
2339 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2343 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2344 fd
= git_mkstemp_mode(buffer
, 0444);
2349 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2350 const void *buf
, unsigned long len
, time_t mtime
)
2353 unsigned char compressed
[4096];
2356 unsigned char parano_sha1
[20];
2358 static char tmpfile
[PATH_MAX
];
2360 filename
= sha1_file_name(sha1
);
2361 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2362 while (fd
< 0 && errno
== EMFILE
&& unuse_one_window(NULL
, -1))
2363 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2365 if (errno
== EACCES
)
2366 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2368 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2372 memset(&stream
, 0, sizeof(stream
));
2373 deflateInit(&stream
, zlib_compression_level
);
2374 stream
.next_out
= compressed
;
2375 stream
.avail_out
= sizeof(compressed
);
2378 /* First header.. */
2379 stream
.next_in
= (unsigned char *)hdr
;
2380 stream
.avail_in
= hdrlen
;
2381 while (deflate(&stream
, 0) == Z_OK
)
2383 git_SHA1_Update(&c
, hdr
, hdrlen
);
2385 /* Then the data itself.. */
2386 stream
.next_in
= (void *)buf
;
2387 stream
.avail_in
= len
;
2389 unsigned char *in0
= stream
.next_in
;
2390 ret
= deflate(&stream
, Z_FINISH
);
2391 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2392 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2393 die("unable to write sha1 file");
2394 stream
.next_out
= compressed
;
2395 stream
.avail_out
= sizeof(compressed
);
2396 } while (ret
== Z_OK
);
2398 if (ret
!= Z_STREAM_END
)
2399 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2400 ret
= deflateEnd(&stream
);
2402 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2403 git_SHA1_Final(parano_sha1
, &c
);
2404 if (hashcmp(sha1
, parano_sha1
) != 0)
2405 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2407 close_sha1_file(fd
);
2412 utb
.modtime
= mtime
;
2413 if (utime(tmpfile
, &utb
) < 0)
2414 warning("failed utime() on %s: %s",
2415 tmpfile
, strerror(errno
));
2418 return move_temp_to_file(tmpfile
, filename
);
2421 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2423 unsigned char sha1
[20];
2427 /* Normally if we have it in the pack then we do not bother writing
2428 * it out into .git/objects/??/?{38} file.
2430 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2432 hashcpy(returnsha1
, sha1
);
2433 if (has_sha1_file(sha1
))
2435 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2438 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2442 enum object_type type
;
2447 if (has_loose_object(sha1
))
2449 buf
= read_packed_sha1(sha1
, &type
, &len
);
2451 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2452 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2453 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2459 int has_pack_index(const unsigned char *sha1
)
2462 if (stat(sha1_pack_index_name(sha1
), &st
))
2467 int has_sha1_pack(const unsigned char *sha1
)
2469 struct pack_entry e
;
2470 return find_pack_entry(sha1
, &e
);
2473 int has_sha1_file(const unsigned char *sha1
)
2475 struct pack_entry e
;
2477 if (find_pack_entry(sha1
, &e
))
2479 return has_loose_object(sha1
);
2482 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2483 int write_object
, enum object_type type
, const char *path
)
2485 int ret
, re_allocated
= 0;
2491 * Convert blobs to git internal format
2493 if ((type
== OBJ_BLOB
) && path
) {
2494 struct strbuf nbuf
= STRBUF_INIT
;
2495 if (convert_to_git(path
, buf
, size
, &nbuf
,
2496 write_object
? safe_crlf
: 0)) {
2497 buf
= strbuf_detach(&nbuf
, &size
);
2503 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2505 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2511 #define SMALL_FILE_SIZE (32*1024)
2513 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2514 enum object_type type
, const char *path
)
2517 size_t size
= xsize_t(st
->st_size
);
2519 if (!S_ISREG(st
->st_mode
)) {
2520 struct strbuf sbuf
= STRBUF_INIT
;
2521 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2522 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2526 strbuf_release(&sbuf
);
2528 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2529 } else if (size
<= SMALL_FILE_SIZE
) {
2530 char *buf
= xmalloc(size
);
2531 if (size
== read_in_full(fd
, buf
, size
))
2532 ret
= index_mem(sha1
, buf
, size
, write_object
, type
,
2535 ret
= error("short read %s", strerror(errno
));
2538 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2539 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2546 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2549 struct strbuf sb
= STRBUF_INIT
;
2551 switch (st
->st_mode
& S_IFMT
) {
2553 fd
= open(path
, O_RDONLY
);
2555 return error("open(\"%s\"): %s", path
,
2557 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2558 return error("%s: failed to insert into database",
2562 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2563 char *errstr
= strerror(errno
);
2564 return error("readlink(\"%s\"): %s", path
,
2568 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2569 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2570 return error("%s: failed to insert into database",
2572 strbuf_release(&sb
);
2575 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2577 return error("%s: unsupported file type", path
);
2582 int read_pack_header(int fd
, struct pack_header
*header
)
2584 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2585 /* "eof before pack header was fully read" */
2586 return PH_ERROR_EOF
;
2588 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2589 /* "protocol error (pack signature mismatch detected)" */
2590 return PH_ERROR_PACK_SIGNATURE
;
2591 if (!pack_version_ok(header
->hdr_version
))
2592 /* "protocol error (pack version unsupported)" */
2593 return PH_ERROR_PROTOCOL
;
2597 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2599 enum object_type type
= sha1_object_info(sha1
, NULL
);
2601 die("%s is not a valid object", sha1_to_hex(sha1
));
2603 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),