2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
16 #include "tree-walk.h"
18 #include "pack-revindex.h"
19 #include "sha1-lookup.h"
22 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
23 #define O_NOATIME 01000000
31 static unsigned long sz_fmt(size_t s
) { return (unsigned long)s
; }
34 static size_t sz_fmt(size_t s
) { return s
; }
37 const unsigned char null_sha1
[20];
39 static int git_open_noatime(const char *name
, struct packed_git
*p
);
42 * This is meant to hold a *small* number of objects that you would
43 * want read_sha1_file() to be able to return, but yet you do not want
44 * to write them into the object store (e.g. a browse-only
47 static struct cached_object
{
48 unsigned char sha1
[20];
49 enum object_type type
;
53 static int cached_object_nr
, cached_object_alloc
;
55 static struct cached_object empty_tree
= {
56 EMPTY_TREE_SHA1_BIN_LITERAL
,
62 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
65 struct cached_object
*co
= cached_objects
;
67 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
68 if (!hashcmp(co
->sha1
, sha1
))
71 if (!hashcmp(sha1
, empty_tree
.sha1
))
76 int safe_create_leading_directories(char *path
)
78 char *pos
= path
+ offset_1st_component(path
);
82 pos
= strchr(pos
, '/');
90 if (!stat(path
, &st
)) {
92 if (!S_ISDIR(st
.st_mode
)) {
97 else if (mkdir(path
, 0777)) {
101 else if (adjust_shared_perm(path
)) {
110 int safe_create_leading_directories_const(const char *path
)
112 /* path points to cache entries, so xstrdup before messing with it */
113 char *buf
= xstrdup(path
);
114 int result
= safe_create_leading_directories(buf
);
119 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
122 for (i
= 0; i
< 20; i
++) {
123 static char hex
[] = "0123456789abcdef";
124 unsigned int val
= sha1
[i
];
125 char *pos
= pathbuf
+ i
*2 + (i
> 0);
126 *pos
++ = hex
[val
>> 4];
127 *pos
= hex
[val
& 0xf];
132 * NOTE! This returns a statically allocated buffer, so you have to be
133 * careful about using it. Do an "xstrdup()" if you need to save the
136 * Also note that this returns the location for creating. Reading
137 * SHA1 file can happen from any alternate directory listed in the
138 * DB_ENVIRONMENT environment variable if it is not found in
139 * the primary object database.
141 char *sha1_file_name(const unsigned char *sha1
)
143 static char buf
[PATH_MAX
];
147 objdir
= get_object_directory();
148 len
= strlen(objdir
);
150 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
151 if (len
+ 43 > PATH_MAX
)
152 die("insanely long object directory %s", objdir
);
153 memcpy(buf
, objdir
, len
);
157 fill_sha1_path(buf
+ len
+ 1, sha1
);
161 static char *sha1_get_pack_name(const unsigned char *sha1
,
162 char **name
, char **base
, const char *which
)
164 static const char hex
[] = "0123456789abcdef";
169 const char *sha1_file_directory
= get_object_directory();
170 int len
= strlen(sha1_file_directory
);
171 *base
= xmalloc(len
+ 60);
172 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
173 sha1_file_directory
, which
);
174 *name
= *base
+ len
+ 11;
179 for (i
= 0; i
< 20; i
++) {
180 unsigned int val
= *sha1
++;
181 *buf
++ = hex
[val
>> 4];
182 *buf
++ = hex
[val
& 0xf];
188 char *sha1_pack_name(const unsigned char *sha1
)
190 static char *name
, *base
;
192 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
195 char *sha1_pack_index_name(const unsigned char *sha1
)
197 static char *name
, *base
;
199 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
202 struct alternate_object_database
*alt_odb_list
;
203 static struct alternate_object_database
**alt_odb_tail
;
205 static void read_info_alternates(const char * alternates
, int depth
);
208 * Prepare alternate object database registry.
210 * The variable alt_odb_list points at the list of struct
211 * alternate_object_database. The elements on this list come from
212 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
213 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
214 * whose contents is similar to that environment variable but can be
215 * LF separated. Its base points at a statically allocated buffer that
216 * contains "/the/directory/corresponding/to/.git/objects/...", while
217 * its name points just after the slash at the end of ".git/objects/"
218 * in the example above, and has enough space to hold 40-byte hex
219 * SHA1, an extra slash for the first level indirection, and the
222 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
224 const char *objdir
= get_object_directory();
225 struct alternate_object_database
*ent
;
226 struct alternate_object_database
*alt
;
227 /* 43 = 40-byte + 2 '/' + terminating NUL */
229 int entlen
= pfxlen
+ 43;
232 if (!is_absolute_path(entry
) && relative_base
) {
233 /* Relative alt-odb */
235 base_len
= strlen(relative_base
) + 1;
239 ent
= xmalloc(sizeof(*ent
) + entlen
);
241 if (!is_absolute_path(entry
) && relative_base
) {
242 memcpy(ent
->base
, relative_base
, base_len
- 1);
243 ent
->base
[base_len
- 1] = '/';
244 memcpy(ent
->base
+ base_len
, entry
, len
);
247 memcpy(ent
->base
, entry
, pfxlen
);
249 ent
->name
= ent
->base
+ pfxlen
+ 1;
250 ent
->base
[pfxlen
+ 3] = '/';
251 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
253 /* Detect cases where alternate disappeared */
254 if (!is_directory(ent
->base
)) {
255 error("object directory %s does not exist; "
256 "check .git/objects/info/alternates.",
262 /* Prevent the common mistake of listing the same
263 * thing twice, or object directory itself.
265 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
266 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
271 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
276 /* add the alternate entry */
278 alt_odb_tail
= &(ent
->next
);
281 /* recursively add alternates */
282 read_info_alternates(ent
->base
, depth
+ 1);
284 ent
->base
[pfxlen
] = '/';
289 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
290 const char *relative_base
, int depth
)
292 const char *cp
, *last
;
295 error("%s: ignoring alternate object stores, nesting too deep.",
303 if (cp
< ep
&& *cp
== '#') {
304 while (cp
< ep
&& *cp
!= sep
)
309 while (cp
< ep
&& *cp
!= sep
)
312 if (!is_absolute_path(last
) && depth
) {
313 error("%s: ignoring relative alternate object store %s",
314 relative_base
, last
);
316 link_alt_odb_entry(last
, cp
- last
,
317 relative_base
, depth
);
320 while (cp
< ep
&& *cp
== sep
)
326 static void read_info_alternates(const char * relative_base
, int depth
)
331 const char alt_file_name
[] = "info/alternates";
332 /* Given that relative_base is no longer than PATH_MAX,
333 ensure that "path" has enough space to append "/", the
334 file name, "info/alternates", and a trailing NUL. */
335 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
338 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
339 fd
= git_open_noatime(path
, NULL
);
342 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
346 mapsz
= xsize_t(st
.st_size
);
347 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
350 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
355 void add_to_alternates_file(const char *reference
)
357 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
358 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
359 char *alt
= mkpath("%s/objects\n", reference
);
360 write_or_die(fd
, alt
, strlen(alt
));
361 if (commit_lock_file(lock
))
362 die("could not close alternates file");
364 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
367 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
369 struct alternate_object_database
*ent
;
372 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
377 void prepare_alt_odb(void)
384 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
387 alt_odb_tail
= &alt_odb_list
;
388 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
390 read_info_alternates(get_object_directory(), 0);
393 static int has_loose_object_local(const unsigned char *sha1
)
395 char *name
= sha1_file_name(sha1
);
396 return !access(name
, F_OK
);
399 int has_loose_object_nonlocal(const unsigned char *sha1
)
401 struct alternate_object_database
*alt
;
403 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
404 fill_sha1_path(alt
->name
, sha1
);
405 if (!access(alt
->base
, F_OK
))
411 static int has_loose_object(const unsigned char *sha1
)
413 return has_loose_object_local(sha1
) ||
414 has_loose_object_nonlocal(sha1
);
417 static unsigned int pack_used_ctr
;
418 static unsigned int pack_mmap_calls
;
419 static unsigned int peak_pack_open_windows
;
420 static unsigned int pack_open_windows
;
421 static size_t peak_pack_mapped
;
422 static size_t pack_mapped
;
423 struct packed_git
*packed_git
;
425 void pack_report(void)
428 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
429 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
430 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
431 sz_fmt(getpagesize()),
432 sz_fmt(packed_git_window_size
),
433 sz_fmt(packed_git_limit
));
435 "pack_report: pack_used_ctr = %10u\n"
436 "pack_report: pack_mmap_calls = %10u\n"
437 "pack_report: pack_open_windows = %10u / %10u\n"
438 "pack_report: pack_mapped = "
439 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
442 pack_open_windows
, peak_pack_open_windows
,
443 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
446 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
449 struct pack_idx_header
*hdr
;
451 uint32_t version
, nr
, i
, *index
;
452 int fd
= git_open_noatime(path
, p
);
457 if (fstat(fd
, &st
)) {
461 idx_size
= xsize_t(st
.st_size
);
462 if (idx_size
< 4 * 256 + 20 + 20) {
464 return error("index file %s is too small", path
);
466 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
470 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
471 version
= ntohl(hdr
->idx_version
);
472 if (version
< 2 || version
> 2) {
473 munmap(idx_map
, idx_size
);
474 return error("index file %s is version %"PRIu32
475 " and is not supported by this binary"
476 " (try upgrading GIT to a newer version)",
485 index
+= 2; /* skip index header */
486 for (i
= 0; i
< 256; i
++) {
487 uint32_t n
= ntohl(index
[i
]);
489 munmap(idx_map
, idx_size
);
490 return error("non-monotonic index %s", path
);
498 * - 256 index entries 4 bytes each
499 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
500 * - 20-byte SHA1 of the packfile
501 * - 20-byte SHA1 file checksum
503 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
504 munmap(idx_map
, idx_size
);
505 return error("wrong index v1 file size in %s", path
);
507 } else if (version
== 2) {
510 * - 8 bytes of header
511 * - 256 index entries 4 bytes each
512 * - 20-byte sha1 entry * nr
513 * - 4-byte crc entry * nr
514 * - 4-byte offset entry * nr
515 * - 20-byte SHA1 of the packfile
516 * - 20-byte SHA1 file checksum
517 * And after the 4-byte offset table might be a
518 * variable sized table containing 8-byte entries
519 * for offsets larger than 2^31.
521 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
522 unsigned long max_size
= min_size
;
524 max_size
+= (nr
- 1)*8;
525 if (idx_size
< min_size
|| idx_size
> max_size
) {
526 munmap(idx_map
, idx_size
);
527 return error("wrong index v2 file size in %s", path
);
529 if (idx_size
!= min_size
&&
531 * make sure we can deal with large pack offsets.
532 * 31-bit signed offset won't be enough, neither
533 * 32-bit unsigned one will be.
535 (sizeof(off_t
) <= 4)) {
536 munmap(idx_map
, idx_size
);
537 return error("pack too large for current definition of off_t in %s", path
);
541 p
->index_version
= version
;
542 p
->index_data
= idx_map
;
543 p
->index_size
= idx_size
;
548 int open_pack_index(struct packed_git
*p
)
556 idx_name
= xstrdup(p
->pack_name
);
557 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
558 ret
= check_packed_git_idx(idx_name
, p
);
563 static void scan_windows(struct packed_git
*p
,
564 struct packed_git
**lru_p
,
565 struct pack_window
**lru_w
,
566 struct pack_window
**lru_l
)
568 struct pack_window
*w
, *w_l
;
570 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
572 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
582 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
584 struct packed_git
*p
, *lru_p
= NULL
;
585 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
588 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
589 for (p
= packed_git
; p
; p
= p
->next
)
590 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
592 munmap(lru_w
->base
, lru_w
->len
);
593 pack_mapped
-= lru_w
->len
;
595 lru_l
->next
= lru_w
->next
;
597 lru_p
->windows
= lru_w
->next
;
598 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
599 close(lru_p
->pack_fd
);
610 void release_pack_memory(size_t need
, int fd
)
612 size_t cur
= pack_mapped
;
613 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
617 void *xmmap(void *start
, size_t length
,
618 int prot
, int flags
, int fd
, off_t offset
)
620 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
621 if (ret
== MAP_FAILED
) {
624 release_pack_memory(length
, fd
);
625 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
626 if (ret
== MAP_FAILED
)
627 die_errno("Out of memory? mmap failed");
632 void close_pack_windows(struct packed_git
*p
)
635 struct pack_window
*w
= p
->windows
;
638 die("pack '%s' still has open windows to it",
640 munmap(w
->base
, w
->len
);
641 pack_mapped
-= w
->len
;
643 p
->windows
= w
->next
;
648 void unuse_pack(struct pack_window
**w_cursor
)
650 struct pack_window
*w
= *w_cursor
;
657 void close_pack_index(struct packed_git
*p
)
660 munmap((void *)p
->index_data
, p
->index_size
);
661 p
->index_data
= NULL
;
666 * This is used by git-repack in case a newly created pack happens to
667 * contain the same set of objects as an existing one. In that case
668 * the resulting file might be different even if its name would be the
669 * same. It is best to close any reference to the old pack before it is
670 * replaced on disk. Of course no index pointers nor windows for given pack
671 * must subsist at this point. If ever objects from this pack are requested
672 * again, the new version of the pack will be reinitialized through
673 * reprepare_packed_git().
675 void free_pack_by_name(const char *pack_name
)
677 struct packed_git
*p
, **pp
= &packed_git
;
681 if (strcmp(pack_name
, p
->pack_name
) == 0) {
682 clear_delta_base_cache();
683 close_pack_windows(p
);
684 if (p
->pack_fd
!= -1)
687 free(p
->bad_object_sha1
);
697 * Do not call this directly as this leaks p->pack_fd on error return;
698 * call open_packed_git() instead.
700 static int open_packed_git_1(struct packed_git
*p
)
703 struct pack_header hdr
;
704 unsigned char sha1
[20];
705 unsigned char *idx_sha1
;
708 if (!p
->index_data
&& open_pack_index(p
))
709 return error("packfile %s index unavailable", p
->pack_name
);
711 p
->pack_fd
= git_open_noatime(p
->pack_name
, p
);
712 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
715 /* If we created the struct before we had the pack we lack size. */
717 if (!S_ISREG(st
.st_mode
))
718 return error("packfile %s not a regular file", p
->pack_name
);
719 p
->pack_size
= st
.st_size
;
720 } else if (p
->pack_size
!= st
.st_size
)
721 return error("packfile %s size changed", p
->pack_name
);
723 /* We leave these file descriptors open with sliding mmap;
724 * there is no point keeping them open across exec(), though.
726 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
728 return error("cannot determine file descriptor flags");
729 fd_flag
|= FD_CLOEXEC
;
730 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
731 return error("cannot set FD_CLOEXEC");
733 /* Verify we recognize this pack file format. */
734 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
735 return error("file %s is far too short to be a packfile", p
->pack_name
);
736 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
737 return error("file %s is not a GIT packfile", p
->pack_name
);
738 if (!pack_version_ok(hdr
.hdr_version
))
739 return error("packfile %s is version %"PRIu32
" and not"
740 " supported (try upgrading GIT to a newer version)",
741 p
->pack_name
, ntohl(hdr
.hdr_version
));
743 /* Verify the pack matches its index. */
744 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
745 return error("packfile %s claims to have %"PRIu32
" objects"
746 " while index indicates %"PRIu32
" objects",
747 p
->pack_name
, ntohl(hdr
.hdr_entries
),
749 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
750 return error("end of packfile %s is unavailable", p
->pack_name
);
751 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
752 return error("packfile %s signature is unavailable", p
->pack_name
);
753 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
754 if (hashcmp(sha1
, idx_sha1
))
755 return error("packfile %s does not match index", p
->pack_name
);
759 static int open_packed_git(struct packed_git
*p
)
761 if (!open_packed_git_1(p
))
763 if (p
->pack_fd
!= -1) {
770 static int in_window(struct pack_window
*win
, off_t offset
)
772 /* We must promise at least 20 bytes (one hash) after the
773 * offset is available from this window, otherwise the offset
774 * is not actually in this window and a different window (which
775 * has that one hash excess) must be used. This is to support
776 * the object header and delta base parsing routines below.
778 off_t win_off
= win
->offset
;
779 return win_off
<= offset
780 && (offset
+ 20) <= (win_off
+ win
->len
);
783 unsigned char *use_pack(struct packed_git
*p
,
784 struct pack_window
**w_cursor
,
788 struct pack_window
*win
= *w_cursor
;
790 if (p
->pack_fd
== -1 && open_packed_git(p
))
791 die("packfile %s cannot be accessed", p
->pack_name
);
793 /* Since packfiles end in a hash of their content and it's
794 * pointless to ask for an offset into the middle of that
795 * hash, and the in_window function above wouldn't match
796 * don't allow an offset too close to the end of the file.
798 if (offset
> (p
->pack_size
- 20))
799 die("offset beyond end of packfile (truncated pack?)");
801 if (!win
|| !in_window(win
, offset
)) {
804 for (win
= p
->windows
; win
; win
= win
->next
) {
805 if (in_window(win
, offset
))
809 size_t window_align
= packed_git_window_size
/ 2;
811 win
= xcalloc(1, sizeof(*win
));
812 win
->offset
= (offset
/ window_align
) * window_align
;
813 len
= p
->pack_size
- win
->offset
;
814 if (len
> packed_git_window_size
)
815 len
= packed_git_window_size
;
816 win
->len
= (size_t)len
;
817 pack_mapped
+= win
->len
;
818 while (packed_git_limit
< pack_mapped
819 && unuse_one_window(p
, p
->pack_fd
))
821 win
->base
= xmmap(NULL
, win
->len
,
822 PROT_READ
, MAP_PRIVATE
,
823 p
->pack_fd
, win
->offset
);
824 if (win
->base
== MAP_FAILED
)
825 die("packfile %s cannot be mapped: %s",
830 if (pack_mapped
> peak_pack_mapped
)
831 peak_pack_mapped
= pack_mapped
;
832 if (pack_open_windows
> peak_pack_open_windows
)
833 peak_pack_open_windows
= pack_open_windows
;
834 win
->next
= p
->windows
;
838 if (win
!= *w_cursor
) {
839 win
->last_used
= pack_used_ctr
++;
843 offset
-= win
->offset
;
845 *left
= win
->len
- xsize_t(offset
);
846 return win
->base
+ offset
;
849 static struct packed_git
*alloc_packed_git(int extra
)
851 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
852 memset(p
, 0, sizeof(*p
));
857 static void try_to_free_pack_memory(size_t size
)
859 release_pack_memory(size
, -1);
862 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
864 static int have_set_try_to_free_routine
;
866 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
868 if (!have_set_try_to_free_routine
) {
869 have_set_try_to_free_routine
= 1;
870 set_try_to_free_routine(try_to_free_pack_memory
);
874 * Make sure a corresponding .pack file exists and that
875 * the index looks sane.
877 path_len
-= strlen(".idx");
882 memcpy(p
->pack_name
, path
, path_len
);
884 strcpy(p
->pack_name
+ path_len
, ".keep");
885 if (!access(p
->pack_name
, F_OK
))
888 strcpy(p
->pack_name
+ path_len
, ".pack");
889 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
894 /* ok, it looks sane as far as we can check without
895 * actually mapping the pack file.
897 p
->pack_size
= st
.st_size
;
898 p
->pack_local
= local
;
899 p
->mtime
= st
.st_mtime
;
900 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
905 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
907 const char *path
= sha1_pack_name(sha1
);
908 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
910 strcpy(p
->pack_name
, path
);
911 hashcpy(p
->sha1
, sha1
);
912 if (check_packed_git_idx(idx_path
, p
)) {
920 void install_packed_git(struct packed_git
*pack
)
922 pack
->next
= packed_git
;
926 static void prepare_packed_git_one(char *objdir
, int local
)
928 /* Ensure that this buffer is large enough so that we can
929 append "/pack/" without clobbering the stack even if
930 strlen(objdir) were PATH_MAX. */
931 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
936 sprintf(path
, "%s/pack", objdir
);
939 while (!dir
&& errno
== EMFILE
&& unuse_one_window(NULL
, -1))
943 error("unable to open object pack directory: %s: %s",
944 path
, strerror(errno
));
948 while ((de
= readdir(dir
)) != NULL
) {
949 int namelen
= strlen(de
->d_name
);
950 struct packed_git
*p
;
952 if (!has_extension(de
->d_name
, ".idx"))
955 if (len
+ namelen
+ 1 > sizeof(path
))
958 /* Don't reopen a pack we already have. */
959 strcpy(path
+ len
, de
->d_name
);
960 for (p
= packed_git
; p
; p
= p
->next
) {
961 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
966 /* See if it really is a valid .idx file with corresponding
967 * .pack file that we can map.
969 p
= add_packed_git(path
, len
+ namelen
, local
);
972 install_packed_git(p
);
977 static int sort_pack(const void *a_
, const void *b_
)
979 struct packed_git
*a
= *((struct packed_git
**)a_
);
980 struct packed_git
*b
= *((struct packed_git
**)b_
);
984 * Local packs tend to contain objects specific to our
985 * variant of the project than remote ones. In addition,
986 * remote ones could be on a network mounted filesystem.
987 * Favor local ones for these reasons.
989 st
= a
->pack_local
- b
->pack_local
;
994 * Younger packs tend to contain more recent objects,
995 * and more recent objects tend to get accessed more
998 if (a
->mtime
< b
->mtime
)
1000 else if (a
->mtime
== b
->mtime
)
1005 static void rearrange_packed_git(void)
1007 struct packed_git
**ary
, *p
;
1010 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1015 /* prepare an array of packed_git for easier sorting */
1016 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1017 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1020 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1022 /* link them back again */
1023 for (i
= 0; i
< n
- 1; i
++)
1024 ary
[i
]->next
= ary
[i
+ 1];
1025 ary
[n
- 1]->next
= NULL
;
1026 packed_git
= ary
[0];
1031 static int prepare_packed_git_run_once
= 0;
1032 void prepare_packed_git(void)
1034 struct alternate_object_database
*alt
;
1036 if (prepare_packed_git_run_once
)
1038 prepare_packed_git_one(get_object_directory(), 1);
1040 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1042 prepare_packed_git_one(alt
->base
, 0);
1043 alt
->name
[-1] = '/';
1045 rearrange_packed_git();
1046 prepare_packed_git_run_once
= 1;
1049 void reprepare_packed_git(void)
1052 prepare_packed_git_run_once
= 0;
1053 prepare_packed_git();
1056 static void mark_bad_packed_object(struct packed_git
*p
,
1057 const unsigned char *sha1
)
1060 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1061 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1063 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1064 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1065 p
->num_bad_objects
++;
1068 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1070 struct packed_git
*p
;
1073 for (p
= packed_git
; p
; p
= p
->next
)
1074 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1075 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1080 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1082 unsigned char real_sha1
[20];
1083 hash_sha1_file(map
, size
, type
, real_sha1
);
1084 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1087 static int git_open_noatime(const char *name
, struct packed_git
*p
)
1089 static int sha1_file_open_flag
= O_NOATIME
;
1092 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1096 /* Might the failure be insufficient file descriptors? */
1097 if (errno
== EMFILE
) {
1098 if (unuse_one_window(p
, -1))
1104 /* Might the failure be due to O_NOATIME? */
1105 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1106 sha1_file_open_flag
= 0;
1114 static int open_sha1_file(const unsigned char *sha1
)
1117 char *name
= sha1_file_name(sha1
);
1118 struct alternate_object_database
*alt
;
1120 fd
= git_open_noatime(name
, NULL
);
1126 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1128 fill_sha1_path(name
, sha1
);
1129 fd
= git_open_noatime(alt
->base
, NULL
);
1136 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1141 fd
= open_sha1_file(sha1
);
1146 if (!fstat(fd
, &st
)) {
1147 *size
= xsize_t(st
.st_size
);
1148 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1155 static int legacy_loose_object(unsigned char *map
)
1160 * Is it a zlib-compressed buffer? If so, the first byte
1161 * must be 0x78 (15-bit window size, deflated), and the
1162 * first 16-bit word is evenly divisible by 31
1164 word
= (map
[0] << 8) + map
[1];
1165 if (map
[0] == 0x78 && !(word
% 31))
1171 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1172 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1175 unsigned long size
, c
;
1176 unsigned long used
= 0;
1179 *type
= (c
>> 4) & 7;
1183 if (len
<= used
|| bitsizeof(long) <= shift
) {
1184 error("bad object header");
1188 size
+= (c
& 0x7f) << shift
;
1195 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1197 unsigned long size
, used
;
1198 static const char valid_loose_object_type
[8] = {
1200 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1201 0, /* "delta" and others are invalid in a loose object */
1203 enum object_type type
;
1205 /* Get the data stream */
1206 memset(stream
, 0, sizeof(*stream
));
1207 stream
->next_in
= map
;
1208 stream
->avail_in
= mapsize
;
1209 stream
->next_out
= buffer
;
1210 stream
->avail_out
= bufsiz
;
1212 if (legacy_loose_object(map
)) {
1213 git_inflate_init(stream
);
1214 return git_inflate(stream
, 0);
1219 * There used to be a second loose object header format which
1220 * was meant to mimic the in-pack format, allowing for direct
1221 * copy of the object data. This format turned up not to be
1222 * really worth it and we don't write it any longer. But we
1223 * can still read it.
1225 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1226 if (!used
|| !valid_loose_object_type
[type
])
1231 /* Set up the stream for the rest.. */
1232 stream
->next_in
= map
;
1233 stream
->avail_in
= mapsize
;
1234 git_inflate_init(stream
);
1236 /* And generate the fake traditional header */
1237 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1238 typename(type
), size
);
1242 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1244 int bytes
= strlen(buffer
) + 1;
1245 unsigned char *buf
= xmallocz(size
);
1249 n
= stream
->total_out
- bytes
;
1252 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1254 if (bytes
<= size
) {
1256 * The above condition must be (bytes <= size), not
1257 * (bytes < size). In other words, even though we
1258 * expect no more output and set avail_out to zer0,
1259 * the input zlib stream may have bytes that express
1260 * "this concludes the stream", and we *do* want to
1263 * Otherwise we would not be able to test that we
1264 * consumed all the input to reach the expected size;
1265 * we also want to check that zlib tells us that all
1266 * went well with status == Z_STREAM_END at the end.
1268 stream
->next_out
= buf
+ bytes
;
1269 stream
->avail_out
= size
- bytes
;
1270 while (status
== Z_OK
)
1271 status
= git_inflate(stream
, Z_FINISH
);
1273 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1274 git_inflate_end(stream
);
1279 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1280 else if (stream
->avail_in
)
1281 error("garbage at end of loose object '%s'",
1288 * We used to just use "sscanf()", but that's actually way
1289 * too permissive for what we want to check. So do an anal
1290 * object header parse by hand.
1292 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1299 * The type can be at most ten bytes (including the
1300 * terminating '\0' that we add), and is followed by
1309 if (i
>= sizeof(type
))
1315 * The length must follow immediately, and be in canonical
1316 * decimal format (ie "010" is not valid).
1318 size
= *hdr
++ - '0';
1323 unsigned long c
= *hdr
- '0';
1327 size
= size
* 10 + c
;
1333 * The length must be followed by a zero byte
1335 return *hdr
? -1 : type_from_string(type
);
1338 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1344 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1345 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1348 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1351 unsigned long get_size_from_delta(struct packed_git
*p
,
1352 struct pack_window
**w_curs
,
1355 const unsigned char *data
;
1356 unsigned char delta_head
[20], *in
;
1360 memset(&stream
, 0, sizeof(stream
));
1361 stream
.next_out
= delta_head
;
1362 stream
.avail_out
= sizeof(delta_head
);
1364 git_inflate_init(&stream
);
1366 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1367 stream
.next_in
= in
;
1368 st
= git_inflate(&stream
, Z_FINISH
);
1369 curpos
+= stream
.next_in
- in
;
1370 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1371 stream
.total_out
< sizeof(delta_head
));
1372 git_inflate_end(&stream
);
1373 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1374 error("delta data unpack-initial failed");
1378 /* Examine the initial part of the delta to figure out
1383 /* ignore base size */
1384 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1386 /* Read the result size */
1387 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1390 static off_t
get_delta_base(struct packed_git
*p
,
1391 struct pack_window
**w_curs
,
1393 enum object_type type
,
1394 off_t delta_obj_offset
)
1396 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1399 /* use_pack() assured us we have [base_info, base_info + 20)
1400 * as a range that we can look at without walking off the
1401 * end of the mapped window. Its actually the hash size
1402 * that is assured. An OFS_DELTA longer than the hash size
1403 * is stupid, as then a REF_DELTA would be smaller to store.
1405 if (type
== OBJ_OFS_DELTA
) {
1407 unsigned char c
= base_info
[used
++];
1408 base_offset
= c
& 127;
1411 if (!base_offset
|| MSB(base_offset
, 7))
1412 return 0; /* overflow */
1413 c
= base_info
[used
++];
1414 base_offset
= (base_offset
<< 7) + (c
& 127);
1416 base_offset
= delta_obj_offset
- base_offset
;
1417 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1418 return 0; /* out of bound */
1420 } else if (type
== OBJ_REF_DELTA
) {
1421 /* The base entry _must_ be in the same pack */
1422 base_offset
= find_pack_entry_one(base_info
, p
);
1425 die("I am totally screwed");
1429 /* forward declaration for a mutually recursive function */
1430 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1431 unsigned long *sizep
);
1433 static int packed_delta_info(struct packed_git
*p
,
1434 struct pack_window
**w_curs
,
1436 enum object_type type
,
1438 unsigned long *sizep
)
1442 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1445 type
= packed_object_info(p
, base_offset
, NULL
);
1446 if (type
<= OBJ_NONE
) {
1447 struct revindex_entry
*revidx
;
1448 const unsigned char *base_sha1
;
1449 revidx
= find_pack_revindex(p
, base_offset
);
1452 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1453 mark_bad_packed_object(p
, base_sha1
);
1454 type
= sha1_object_info(base_sha1
, NULL
);
1455 if (type
<= OBJ_NONE
)
1459 /* We choose to only get the type of the base object and
1460 * ignore potentially corrupt pack file that expects the delta
1461 * based on a base with a wrong size. This saves tons of
1465 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1473 static int unpack_object_header(struct packed_git
*p
,
1474 struct pack_window
**w_curs
,
1476 unsigned long *sizep
)
1478 unsigned char *base
;
1481 enum object_type type
;
1483 /* use_pack() assures us we have [base, base + 20) available
1484 * as a range that we can look at at. (Its actually the hash
1485 * size that is assured.) With our object header encoding
1486 * the maximum deflated object size is 2^137, which is just
1487 * insane, so we know won't exceed what we have been given.
1489 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1490 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1499 const char *packed_object_info_detail(struct packed_git
*p
,
1501 unsigned long *size
,
1502 unsigned long *store_size
,
1503 unsigned int *delta_chain_length
,
1504 unsigned char *base_sha1
)
1506 struct pack_window
*w_curs
= NULL
;
1508 unsigned long dummy
;
1509 unsigned char *next_sha1
;
1510 enum object_type type
;
1511 struct revindex_entry
*revidx
;
1513 *delta_chain_length
= 0;
1514 curpos
= obj_offset
;
1515 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1517 revidx
= find_pack_revindex(p
, obj_offset
);
1518 *store_size
= revidx
[1].offset
- obj_offset
;
1523 die("pack %s contains unknown object type %d",
1524 p
->pack_name
, type
);
1529 unuse_pack(&w_curs
);
1530 return typename(type
);
1532 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1534 die("pack %s contains bad delta base reference of type %s",
1535 p
->pack_name
, typename(type
));
1536 if (*delta_chain_length
== 0) {
1537 revidx
= find_pack_revindex(p
, obj_offset
);
1538 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1542 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1543 if (*delta_chain_length
== 0)
1544 hashcpy(base_sha1
, next_sha1
);
1545 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1548 (*delta_chain_length
)++;
1549 curpos
= obj_offset
;
1550 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1554 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1555 unsigned long *sizep
)
1557 struct pack_window
*w_curs
= NULL
;
1559 off_t curpos
= obj_offset
;
1560 enum object_type type
;
1562 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1567 type
= packed_delta_info(p
, &w_curs
, curpos
,
1568 type
, obj_offset
, sizep
);
1578 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1579 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1582 unuse_pack(&w_curs
);
1586 static void *unpack_compressed_entry(struct packed_git
*p
,
1587 struct pack_window
**w_curs
,
1593 unsigned char *buffer
, *in
;
1595 buffer
= xmallocz(size
);
1596 memset(&stream
, 0, sizeof(stream
));
1597 stream
.next_out
= buffer
;
1598 stream
.avail_out
= size
+ 1;
1600 git_inflate_init(&stream
);
1602 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1603 stream
.next_in
= in
;
1604 st
= git_inflate(&stream
, Z_FINISH
);
1605 if (!stream
.avail_out
)
1606 break; /* the payload is larger than it should be */
1607 curpos
+= stream
.next_in
- in
;
1608 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1609 git_inflate_end(&stream
);
1610 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1618 #define MAX_DELTA_CACHE (256)
1620 static size_t delta_base_cached
;
1622 static struct delta_base_cache_lru_list
{
1623 struct delta_base_cache_lru_list
*prev
;
1624 struct delta_base_cache_lru_list
*next
;
1625 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1627 static struct delta_base_cache_entry
{
1628 struct delta_base_cache_lru_list lru
;
1630 struct packed_git
*p
;
1633 enum object_type type
;
1634 } delta_base_cache
[MAX_DELTA_CACHE
];
1636 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1640 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1641 hash
+= (hash
>> 8) + (hash
>> 16);
1642 return hash
% MAX_DELTA_CACHE
;
1645 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1646 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1649 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1650 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1653 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1654 return unpack_entry(p
, base_offset
, type
, base_size
);
1658 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1659 ent
->lru
.prev
->next
= ent
->lru
.next
;
1660 delta_base_cached
-= ent
->size
;
1662 ret
= xmemdupz(ent
->data
, ent
->size
);
1665 *base_size
= ent
->size
;
1669 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1674 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1675 ent
->lru
.prev
->next
= ent
->lru
.next
;
1676 delta_base_cached
-= ent
->size
;
1680 void clear_delta_base_cache(void)
1683 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1684 release_delta_base_cache(&delta_base_cache
[p
]);
1687 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1688 void *base
, unsigned long base_size
, enum object_type type
)
1690 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1691 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1692 struct delta_base_cache_lru_list
*lru
;
1694 release_delta_base_cache(ent
);
1695 delta_base_cached
+= base_size
;
1697 for (lru
= delta_base_cache_lru
.next
;
1698 delta_base_cached
> delta_base_cache_limit
1699 && lru
!= &delta_base_cache_lru
;
1701 struct delta_base_cache_entry
*f
= (void *)lru
;
1702 if (f
->type
== OBJ_BLOB
)
1703 release_delta_base_cache(f
);
1705 for (lru
= delta_base_cache_lru
.next
;
1706 delta_base_cached
> delta_base_cache_limit
1707 && lru
!= &delta_base_cache_lru
;
1709 struct delta_base_cache_entry
*f
= (void *)lru
;
1710 release_delta_base_cache(f
);
1714 ent
->base_offset
= base_offset
;
1717 ent
->size
= base_size
;
1718 ent
->lru
.next
= &delta_base_cache_lru
;
1719 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1720 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1721 delta_base_cache_lru
.prev
= &ent
->lru
;
1724 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1725 unsigned long *size
);
1727 static void *unpack_delta_entry(struct packed_git
*p
,
1728 struct pack_window
**w_curs
,
1730 unsigned long delta_size
,
1732 enum object_type
*type
,
1733 unsigned long *sizep
)
1735 void *delta_data
, *result
, *base
;
1736 unsigned long base_size
;
1739 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1741 error("failed to validate delta base reference "
1742 "at offset %"PRIuMAX
" from %s",
1743 (uintmax_t)curpos
, p
->pack_name
);
1747 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1750 * We're probably in deep shit, but let's try to fetch
1751 * the required base anyway from another pack or loose.
1752 * This is costly but should happen only in the presence
1753 * of a corrupted pack, and is better than failing outright.
1755 struct revindex_entry
*revidx
;
1756 const unsigned char *base_sha1
;
1757 revidx
= find_pack_revindex(p
, base_offset
);
1760 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1761 error("failed to read delta base object %s"
1762 " at offset %"PRIuMAX
" from %s",
1763 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1765 mark_bad_packed_object(p
, base_sha1
);
1766 base
= read_object(base_sha1
, type
, &base_size
);
1771 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1773 error("failed to unpack compressed delta "
1774 "at offset %"PRIuMAX
" from %s",
1775 (uintmax_t)curpos
, p
->pack_name
);
1779 result
= patch_delta(base
, base_size
,
1780 delta_data
, delta_size
,
1783 die("failed to apply delta");
1785 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1789 int do_check_packed_object_crc
;
1791 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1792 enum object_type
*type
, unsigned long *sizep
)
1794 struct pack_window
*w_curs
= NULL
;
1795 off_t curpos
= obj_offset
;
1798 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1799 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1800 unsigned long len
= revidx
[1].offset
- obj_offset
;
1801 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1802 const unsigned char *sha1
=
1803 nth_packed_object_sha1(p
, revidx
->nr
);
1804 error("bad packed object CRC for %s",
1806 mark_bad_packed_object(p
, sha1
);
1807 unuse_pack(&w_curs
);
1812 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1816 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1817 obj_offset
, type
, sizep
);
1823 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1827 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1828 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1830 unuse_pack(&w_curs
);
1834 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1837 const unsigned char *index
= p
->index_data
;
1839 if (open_pack_index(p
))
1841 index
= p
->index_data
;
1843 if (n
>= p
->num_objects
)
1846 if (p
->index_version
== 1) {
1847 return index
+ 24 * n
+ 4;
1850 return index
+ 20 * n
;
1854 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1856 const unsigned char *index
= p
->index_data
;
1858 if (p
->index_version
== 1) {
1859 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1862 index
+= 8 + p
->num_objects
* (20 + 4);
1863 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1864 if (!(off
& 0x80000000))
1866 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1867 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1868 ntohl(*((uint32_t *)(index
+ 4)));
1872 off_t
find_pack_entry_one(const unsigned char *sha1
,
1873 struct packed_git
*p
)
1875 const uint32_t *level1_ofs
= p
->index_data
;
1876 const unsigned char *index
= p
->index_data
;
1877 unsigned hi
, lo
, stride
;
1878 static int use_lookup
= -1;
1879 static int debug_lookup
= -1;
1881 if (debug_lookup
< 0)
1882 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1885 if (open_pack_index(p
))
1887 level1_ofs
= p
->index_data
;
1888 index
= p
->index_data
;
1890 if (p
->index_version
> 1) {
1895 hi
= ntohl(level1_ofs
[*sha1
]);
1896 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1897 if (p
->index_version
> 1) {
1905 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1906 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1909 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1911 int pos
= sha1_entry_pos(index
, stride
, 0,
1912 lo
, hi
, p
->num_objects
, sha1
);
1915 return nth_packed_object_offset(p
, pos
);
1919 unsigned mi
= (lo
+ hi
) / 2;
1920 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1923 printf("lo %u hi %u rg %u mi %u\n",
1924 lo
, hi
, hi
- lo
, mi
);
1926 return nth_packed_object_offset(p
, mi
);
1935 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1937 static struct packed_git
*last_found
= (void *)1;
1938 struct packed_git
*p
;
1941 prepare_packed_git();
1944 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1947 if (p
->num_bad_objects
) {
1949 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1950 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1954 offset
= find_pack_entry_one(sha1
, p
);
1957 * We are about to tell the caller where they can
1958 * locate the requested object. We better make
1959 * sure the packfile is still here and can be
1960 * accessed before supplying that answer, as
1961 * it may have been deleted since the index
1964 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1965 error("packfile %s cannot be accessed", p
->pack_name
);
1970 hashcpy(e
->sha1
, sha1
);
1976 if (p
== last_found
)
1980 if (p
== last_found
)
1986 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1987 struct packed_git
*packs
)
1989 struct packed_git
*p
;
1991 for (p
= packs
; p
; p
= p
->next
) {
1992 if (find_pack_entry_one(sha1
, p
))
1999 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2002 unsigned long mapsize
, size
;
2007 map
= map_sha1_file(sha1
, &mapsize
);
2009 return error("unable to find %s", sha1_to_hex(sha1
));
2010 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2011 status
= error("unable to unpack %s header",
2013 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2014 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2017 git_inflate_end(&stream
);
2018 munmap(map
, mapsize
);
2022 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2024 struct cached_object
*co
;
2025 struct pack_entry e
;
2028 co
= find_cached_object(sha1
);
2035 if (!find_pack_entry(sha1
, &e
)) {
2036 /* Most likely it's a loose object. */
2037 status
= sha1_loose_object_info(sha1
, sizep
);
2041 /* Not a loose object; someone else may have just packed it. */
2042 reprepare_packed_git();
2043 if (!find_pack_entry(sha1
, &e
))
2047 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
2049 mark_bad_packed_object(e
.p
, sha1
);
2050 status
= sha1_object_info(sha1
, sizep
);
2056 static void *read_packed_sha1(const unsigned char *sha1
,
2057 enum object_type
*type
, unsigned long *size
)
2059 struct pack_entry e
;
2062 if (!find_pack_entry(sha1
, &e
))
2064 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2067 * We're probably in deep shit, but let's try to fetch
2068 * the required object anyway from another pack or loose.
2069 * This should happen only in the presence of a corrupted
2070 * pack, and is better than failing outright.
2072 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2073 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2074 mark_bad_packed_object(e
.p
, sha1
);
2075 data
= read_object(sha1
, type
, size
);
2080 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2081 unsigned char *sha1
)
2083 struct cached_object
*co
;
2085 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2086 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2088 if (cached_object_alloc
<= cached_object_nr
) {
2089 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2090 cached_objects
= xrealloc(cached_objects
,
2091 sizeof(*cached_objects
) *
2092 cached_object_alloc
);
2094 co
= &cached_objects
[cached_object_nr
++];
2097 co
->buf
= xmalloc(len
);
2098 memcpy(co
->buf
, buf
, len
);
2099 hashcpy(co
->sha1
, sha1
);
2103 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2104 unsigned long *size
)
2106 unsigned long mapsize
;
2108 struct cached_object
*co
;
2110 co
= find_cached_object(sha1
);
2114 return xmemdupz(co
->buf
, co
->size
);
2117 buf
= read_packed_sha1(sha1
, type
, size
);
2120 map
= map_sha1_file(sha1
, &mapsize
);
2122 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2123 munmap(map
, mapsize
);
2126 reprepare_packed_git();
2127 return read_packed_sha1(sha1
, type
, size
);
2131 * This function dies on corrupt objects; the callers who want to
2132 * deal with them should arrange to call read_object() and give error
2133 * messages themselves.
2135 void *read_sha1_file_repl(const unsigned char *sha1
,
2136 enum object_type
*type
,
2137 unsigned long *size
,
2138 const unsigned char **replacement
)
2140 const unsigned char *repl
= lookup_replace_object(sha1
);
2143 const struct packed_git
*p
;
2146 data
= read_object(repl
, type
, size
);
2149 *replacement
= repl
;
2153 if (errno
&& errno
!= ENOENT
)
2154 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2156 /* die if we replaced an object with one that does not exist */
2158 die("replacement %s not found for %s",
2159 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2161 if (has_loose_object(repl
)) {
2162 path
= sha1_file_name(sha1
);
2163 die("loose object %s (stored in %s) is corrupt",
2164 sha1_to_hex(repl
), path
);
2167 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2168 die("packed object %s (stored in %s) is corrupt",
2169 sha1_to_hex(repl
), p
->pack_name
);
2174 void *read_object_with_reference(const unsigned char *sha1
,
2175 const char *required_type_name
,
2176 unsigned long *size
,
2177 unsigned char *actual_sha1_return
)
2179 enum object_type type
, required_type
;
2181 unsigned long isize
;
2182 unsigned char actual_sha1
[20];
2184 required_type
= type_from_string(required_type_name
);
2185 hashcpy(actual_sha1
, sha1
);
2187 int ref_length
= -1;
2188 const char *ref_type
= NULL
;
2190 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2193 if (type
== required_type
) {
2195 if (actual_sha1_return
)
2196 hashcpy(actual_sha1_return
, actual_sha1
);
2199 /* Handle references */
2200 else if (type
== OBJ_COMMIT
)
2202 else if (type
== OBJ_TAG
)
2203 ref_type
= "object ";
2208 ref_length
= strlen(ref_type
);
2210 if (ref_length
+ 40 > isize
||
2211 memcmp(buffer
, ref_type
, ref_length
) ||
2212 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2217 /* Now we have the ID of the referred-to object in
2218 * actual_sha1. Check again. */
2222 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2223 const char *type
, unsigned char *sha1
,
2224 char *hdr
, int *hdrlen
)
2228 /* Generate the header */
2229 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2233 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2234 git_SHA1_Update(&c
, buf
, len
);
2235 git_SHA1_Final(sha1
, &c
);
2239 * Move the just written object into its final resting place.
2240 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2241 * "moving" is only a part of what it does, when no patch between
2242 * master to pu changes the call sites of this function.
2244 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2248 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2250 else if (link(tmpfile
, filename
))
2254 * Coda hack - coda doesn't like cross-directory links,
2255 * so we fall back to a rename, which will mean that it
2256 * won't be able to check collisions, but that's not a
2259 * The same holds for FAT formatted media.
2261 * When this succeeds, we just return. We have nothing
2264 if (ret
&& ret
!= EEXIST
) {
2266 if (!rename(tmpfile
, filename
))
2270 unlink_or_warn(tmpfile
);
2272 if (ret
!= EEXIST
) {
2273 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2275 /* FIXME!!! Collision check here ? */
2279 if (adjust_shared_perm(filename
))
2280 return error("unable to set permission to '%s'", filename
);
2284 static int write_buffer(int fd
, const void *buf
, size_t len
)
2286 if (write_in_full(fd
, buf
, len
) < 0)
2287 return error("file write error (%s)", strerror(errno
));
2291 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2292 unsigned char *sha1
)
2296 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2300 /* Finalize a file on disk, and close it. */
2301 static void close_sha1_file(int fd
)
2303 if (fsync_object_files
)
2304 fsync_or_die(fd
, "sha1 file");
2306 die_errno("error when closing sha1 file");
2309 /* Size of directory component, including the ending '/' */
2310 static inline int directory_size(const char *filename
)
2312 const char *s
= strrchr(filename
, '/');
2315 return s
- filename
+ 1;
2319 * This creates a temporary file in the same directory as the final
2322 * We want to avoid cross-directory filename renames, because those
2323 * can have problems on various filesystems (FAT, NFS, Coda).
2325 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2327 int fd
, dirlen
= directory_size(filename
);
2329 if (dirlen
+ 20 > bufsiz
) {
2330 errno
= ENAMETOOLONG
;
2333 memcpy(buffer
, filename
, dirlen
);
2334 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2335 fd
= git_mkstemp_mode(buffer
, 0444);
2336 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2337 /* Make sure the directory exists */
2338 memcpy(buffer
, filename
, dirlen
);
2339 buffer
[dirlen
-1] = 0;
2340 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2344 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2345 fd
= git_mkstemp_mode(buffer
, 0444);
2350 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2351 const void *buf
, unsigned long len
, time_t mtime
)
2354 unsigned char compressed
[4096];
2357 unsigned char parano_sha1
[20];
2359 static char tmpfile
[PATH_MAX
];
2361 filename
= sha1_file_name(sha1
);
2362 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2363 while (fd
< 0 && errno
== EMFILE
&& unuse_one_window(NULL
, -1))
2364 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2366 if (errno
== EACCES
)
2367 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2369 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2373 memset(&stream
, 0, sizeof(stream
));
2374 deflateInit(&stream
, zlib_compression_level
);
2375 stream
.next_out
= compressed
;
2376 stream
.avail_out
= sizeof(compressed
);
2379 /* First header.. */
2380 stream
.next_in
= (unsigned char *)hdr
;
2381 stream
.avail_in
= hdrlen
;
2382 while (deflate(&stream
, 0) == Z_OK
)
2384 git_SHA1_Update(&c
, hdr
, hdrlen
);
2386 /* Then the data itself.. */
2387 stream
.next_in
= (void *)buf
;
2388 stream
.avail_in
= len
;
2390 unsigned char *in0
= stream
.next_in
;
2391 ret
= deflate(&stream
, Z_FINISH
);
2392 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2393 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2394 die("unable to write sha1 file");
2395 stream
.next_out
= compressed
;
2396 stream
.avail_out
= sizeof(compressed
);
2397 } while (ret
== Z_OK
);
2399 if (ret
!= Z_STREAM_END
)
2400 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2401 ret
= deflateEnd(&stream
);
2403 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2404 git_SHA1_Final(parano_sha1
, &c
);
2405 if (hashcmp(sha1
, parano_sha1
) != 0)
2406 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2408 close_sha1_file(fd
);
2413 utb
.modtime
= mtime
;
2414 if (utime(tmpfile
, &utb
) < 0)
2415 warning("failed utime() on %s: %s",
2416 tmpfile
, strerror(errno
));
2419 return move_temp_to_file(tmpfile
, filename
);
2422 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2424 unsigned char sha1
[20];
2428 /* Normally if we have it in the pack then we do not bother writing
2429 * it out into .git/objects/??/?{38} file.
2431 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2433 hashcpy(returnsha1
, sha1
);
2434 if (has_sha1_file(sha1
))
2436 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2439 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2443 enum object_type type
;
2448 if (has_loose_object(sha1
))
2450 buf
= read_packed_sha1(sha1
, &type
, &len
);
2452 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2453 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2454 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2460 int has_pack_index(const unsigned char *sha1
)
2463 if (stat(sha1_pack_index_name(sha1
), &st
))
2468 int has_sha1_pack(const unsigned char *sha1
)
2470 struct pack_entry e
;
2471 return find_pack_entry(sha1
, &e
);
2474 int has_sha1_file(const unsigned char *sha1
)
2476 struct pack_entry e
;
2478 if (find_pack_entry(sha1
, &e
))
2480 return has_loose_object(sha1
);
2483 static void check_tree(const void *buf
, size_t size
)
2485 struct tree_desc desc
;
2486 struct name_entry entry
;
2488 init_tree_desc(&desc
, buf
, size
);
2489 while (tree_entry(&desc
, &entry
))
2491 * tree_entry() will die() on malformed entries */
2495 static void check_commit(const void *buf
, size_t size
)
2498 memset(&c
, 0, sizeof(c
));
2499 if (parse_commit_buffer(&c
, buf
, size
))
2500 die("corrupt commit");
2503 static void check_tag(const void *buf
, size_t size
)
2506 memset(&t
, 0, sizeof(t
));
2507 if (parse_tag_buffer(&t
, buf
, size
))
2511 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2512 int write_object
, enum object_type type
,
2513 const char *path
, int format_check
)
2515 int ret
, re_allocated
= 0;
2521 * Convert blobs to git internal format
2523 if ((type
== OBJ_BLOB
) && path
) {
2524 struct strbuf nbuf
= STRBUF_INIT
;
2525 if (convert_to_git(path
, buf
, size
, &nbuf
,
2526 write_object
? safe_crlf
: 0)) {
2527 buf
= strbuf_detach(&nbuf
, &size
);
2532 if (type
== OBJ_TREE
)
2533 check_tree(buf
, size
);
2534 if (type
== OBJ_COMMIT
)
2535 check_commit(buf
, size
);
2536 if (type
== OBJ_TAG
)
2537 check_tag(buf
, size
);
2541 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2543 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2549 #define SMALL_FILE_SIZE (32*1024)
2551 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2552 enum object_type type
, const char *path
, int format_check
)
2555 size_t size
= xsize_t(st
->st_size
);
2557 if (!S_ISREG(st
->st_mode
)) {
2558 struct strbuf sbuf
= STRBUF_INIT
;
2559 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2560 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2561 type
, path
, format_check
);
2564 strbuf_release(&sbuf
);
2566 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
,
2568 } else if (size
<= SMALL_FILE_SIZE
) {
2569 char *buf
= xmalloc(size
);
2570 if (size
== read_in_full(fd
, buf
, size
))
2571 ret
= index_mem(sha1
, buf
, size
, write_object
, type
,
2572 path
, format_check
);
2574 ret
= error("short read %s", strerror(errno
));
2577 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2578 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
,
2586 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2589 struct strbuf sb
= STRBUF_INIT
;
2591 switch (st
->st_mode
& S_IFMT
) {
2593 fd
= open(path
, O_RDONLY
);
2595 return error("open(\"%s\"): %s", path
,
2597 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
, 0) < 0)
2598 return error("%s: failed to insert into database",
2602 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2603 char *errstr
= strerror(errno
);
2604 return error("readlink(\"%s\"): %s", path
,
2608 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2609 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2610 return error("%s: failed to insert into database",
2612 strbuf_release(&sb
);
2615 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2617 return error("%s: unsupported file type", path
);
2622 int read_pack_header(int fd
, struct pack_header
*header
)
2624 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2625 /* "eof before pack header was fully read" */
2626 return PH_ERROR_EOF
;
2628 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2629 /* "protocol error (pack signature mismatch detected)" */
2630 return PH_ERROR_PACK_SIGNATURE
;
2631 if (!pack_version_ok(header
->hdr_version
))
2632 /* "protocol error (pack version unsupported)" */
2633 return PH_ERROR_PROTOCOL
;
2637 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2639 enum object_type type
= sha1_object_info(sha1
, NULL
);
2641 die("%s is not a valid object", sha1_to_hex(sha1
));
2643 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),