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 unsigned int pack_open_fds
;
421 static unsigned int pack_max_fds
;
422 static size_t peak_pack_mapped
;
423 static size_t pack_mapped
;
424 struct packed_git
*packed_git
;
426 void pack_report(void)
429 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
430 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
431 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
432 sz_fmt(getpagesize()),
433 sz_fmt(packed_git_window_size
),
434 sz_fmt(packed_git_limit
));
436 "pack_report: pack_used_ctr = %10u\n"
437 "pack_report: pack_mmap_calls = %10u\n"
438 "pack_report: pack_open_windows = %10u / %10u\n"
439 "pack_report: pack_mapped = "
440 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
443 pack_open_windows
, peak_pack_open_windows
,
444 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
447 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
450 struct pack_idx_header
*hdr
;
452 uint32_t version
, nr
, i
, *index
;
453 int fd
= git_open_noatime(path
, p
);
458 if (fstat(fd
, &st
)) {
462 idx_size
= xsize_t(st
.st_size
);
463 if (idx_size
< 4 * 256 + 20 + 20) {
465 return error("index file %s is too small", path
);
467 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
471 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
472 version
= ntohl(hdr
->idx_version
);
473 if (version
< 2 || version
> 2) {
474 munmap(idx_map
, idx_size
);
475 return error("index file %s is version %"PRIu32
476 " and is not supported by this binary"
477 " (try upgrading GIT to a newer version)",
486 index
+= 2; /* skip index header */
487 for (i
= 0; i
< 256; i
++) {
488 uint32_t n
= ntohl(index
[i
]);
490 munmap(idx_map
, idx_size
);
491 return error("non-monotonic index %s", path
);
499 * - 256 index entries 4 bytes each
500 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
501 * - 20-byte SHA1 of the packfile
502 * - 20-byte SHA1 file checksum
504 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
505 munmap(idx_map
, idx_size
);
506 return error("wrong index v1 file size in %s", path
);
508 } else if (version
== 2) {
511 * - 8 bytes of header
512 * - 256 index entries 4 bytes each
513 * - 20-byte sha1 entry * nr
514 * - 4-byte crc entry * nr
515 * - 4-byte offset entry * nr
516 * - 20-byte SHA1 of the packfile
517 * - 20-byte SHA1 file checksum
518 * And after the 4-byte offset table might be a
519 * variable sized table containing 8-byte entries
520 * for offsets larger than 2^31.
522 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
523 unsigned long max_size
= min_size
;
525 max_size
+= (nr
- 1)*8;
526 if (idx_size
< min_size
|| idx_size
> max_size
) {
527 munmap(idx_map
, idx_size
);
528 return error("wrong index v2 file size in %s", path
);
530 if (idx_size
!= min_size
&&
532 * make sure we can deal with large pack offsets.
533 * 31-bit signed offset won't be enough, neither
534 * 32-bit unsigned one will be.
536 (sizeof(off_t
) <= 4)) {
537 munmap(idx_map
, idx_size
);
538 return error("pack too large for current definition of off_t in %s", path
);
542 p
->index_version
= version
;
543 p
->index_data
= idx_map
;
544 p
->index_size
= idx_size
;
549 int open_pack_index(struct packed_git
*p
)
557 idx_name
= xstrdup(p
->pack_name
);
558 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
559 ret
= check_packed_git_idx(idx_name
, p
);
564 static void scan_windows(struct packed_git
*p
,
565 struct packed_git
**lru_p
,
566 struct pack_window
**lru_w
,
567 struct pack_window
**lru_l
)
569 struct pack_window
*w
, *w_l
;
571 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
573 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
583 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
585 struct packed_git
*p
, *lru_p
= NULL
;
586 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
589 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
590 for (p
= packed_git
; p
; p
= p
->next
)
591 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
593 munmap(lru_w
->base
, lru_w
->len
);
594 pack_mapped
-= lru_w
->len
;
596 lru_l
->next
= lru_w
->next
;
598 lru_p
->windows
= lru_w
->next
;
599 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
600 close(lru_p
->pack_fd
);
612 void release_pack_memory(size_t need
, int fd
)
614 size_t cur
= pack_mapped
;
615 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
619 void *xmmap(void *start
, size_t length
,
620 int prot
, int flags
, int fd
, off_t offset
)
622 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
623 if (ret
== MAP_FAILED
) {
626 release_pack_memory(length
, fd
);
627 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
628 if (ret
== MAP_FAILED
)
629 die_errno("Out of memory? mmap failed");
634 void close_pack_windows(struct packed_git
*p
)
637 struct pack_window
*w
= p
->windows
;
640 die("pack '%s' still has open windows to it",
642 munmap(w
->base
, w
->len
);
643 pack_mapped
-= w
->len
;
645 p
->windows
= w
->next
;
650 void unuse_pack(struct pack_window
**w_cursor
)
652 struct pack_window
*w
= *w_cursor
;
659 void close_pack_index(struct packed_git
*p
)
662 munmap((void *)p
->index_data
, p
->index_size
);
663 p
->index_data
= NULL
;
668 * This is used by git-repack in case a newly created pack happens to
669 * contain the same set of objects as an existing one. In that case
670 * the resulting file might be different even if its name would be the
671 * same. It is best to close any reference to the old pack before it is
672 * replaced on disk. Of course no index pointers nor windows for given pack
673 * must subsist at this point. If ever objects from this pack are requested
674 * again, the new version of the pack will be reinitialized through
675 * reprepare_packed_git().
677 void free_pack_by_name(const char *pack_name
)
679 struct packed_git
*p
, **pp
= &packed_git
;
683 if (strcmp(pack_name
, p
->pack_name
) == 0) {
684 clear_delta_base_cache();
685 close_pack_windows(p
);
686 if (p
->pack_fd
!= -1) {
691 free(p
->bad_object_sha1
);
701 * Do not call this directly as this leaks p->pack_fd on error return;
702 * call open_packed_git() instead.
704 static int open_packed_git_1(struct packed_git
*p
)
707 struct pack_header hdr
;
708 unsigned char sha1
[20];
709 unsigned char *idx_sha1
;
712 if (!p
->index_data
&& open_pack_index(p
))
713 return error("packfile %s index unavailable", p
->pack_name
);
717 unsigned int max_fds
;
719 if (getrlimit(RLIMIT_NOFILE
, &lim
))
720 die_errno("cannot get RLIMIT_NOFILE");
722 max_fds
= lim
.rlim_cur
;
724 /* Save 3 for stdin/stdout/stderr, 22 for work */
726 pack_max_fds
= max_fds
- 25;
731 while (pack_max_fds
<= pack_open_fds
&& unuse_one_window(NULL
, -1))
734 p
->pack_fd
= git_open_noatime(p
->pack_name
, p
);
735 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
739 /* If we created the struct before we had the pack we lack size. */
741 if (!S_ISREG(st
.st_mode
))
742 return error("packfile %s not a regular file", p
->pack_name
);
743 p
->pack_size
= st
.st_size
;
744 } else if (p
->pack_size
!= st
.st_size
)
745 return error("packfile %s size changed", p
->pack_name
);
747 /* We leave these file descriptors open with sliding mmap;
748 * there is no point keeping them open across exec(), though.
750 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
752 return error("cannot determine file descriptor flags");
753 fd_flag
|= FD_CLOEXEC
;
754 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
755 return error("cannot set FD_CLOEXEC");
757 /* Verify we recognize this pack file format. */
758 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
759 return error("file %s is far too short to be a packfile", p
->pack_name
);
760 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
761 return error("file %s is not a GIT packfile", p
->pack_name
);
762 if (!pack_version_ok(hdr
.hdr_version
))
763 return error("packfile %s is version %"PRIu32
" and not"
764 " supported (try upgrading GIT to a newer version)",
765 p
->pack_name
, ntohl(hdr
.hdr_version
));
767 /* Verify the pack matches its index. */
768 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
769 return error("packfile %s claims to have %"PRIu32
" objects"
770 " while index indicates %"PRIu32
" objects",
771 p
->pack_name
, ntohl(hdr
.hdr_entries
),
773 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
774 return error("end of packfile %s is unavailable", p
->pack_name
);
775 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
776 return error("packfile %s signature is unavailable", p
->pack_name
);
777 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
778 if (hashcmp(sha1
, idx_sha1
))
779 return error("packfile %s does not match index", p
->pack_name
);
783 static int open_packed_git(struct packed_git
*p
)
785 if (!open_packed_git_1(p
))
787 if (p
->pack_fd
!= -1) {
795 static int in_window(struct pack_window
*win
, off_t offset
)
797 /* We must promise at least 20 bytes (one hash) after the
798 * offset is available from this window, otherwise the offset
799 * is not actually in this window and a different window (which
800 * has that one hash excess) must be used. This is to support
801 * the object header and delta base parsing routines below.
803 off_t win_off
= win
->offset
;
804 return win_off
<= offset
805 && (offset
+ 20) <= (win_off
+ win
->len
);
808 unsigned char *use_pack(struct packed_git
*p
,
809 struct pack_window
**w_cursor
,
813 struct pack_window
*win
= *w_cursor
;
815 if (p
->pack_fd
== -1 && open_packed_git(p
))
816 die("packfile %s cannot be accessed", p
->pack_name
);
818 /* Since packfiles end in a hash of their content and it's
819 * pointless to ask for an offset into the middle of that
820 * hash, and the in_window function above wouldn't match
821 * don't allow an offset too close to the end of the file.
823 if (offset
> (p
->pack_size
- 20))
824 die("offset beyond end of packfile (truncated pack?)");
826 if (!win
|| !in_window(win
, offset
)) {
829 for (win
= p
->windows
; win
; win
= win
->next
) {
830 if (in_window(win
, offset
))
834 size_t window_align
= packed_git_window_size
/ 2;
836 win
= xcalloc(1, sizeof(*win
));
837 win
->offset
= (offset
/ window_align
) * window_align
;
838 len
= p
->pack_size
- win
->offset
;
839 if (len
> packed_git_window_size
)
840 len
= packed_git_window_size
;
841 win
->len
= (size_t)len
;
842 pack_mapped
+= win
->len
;
843 while (packed_git_limit
< pack_mapped
844 && unuse_one_window(p
, p
->pack_fd
))
846 win
->base
= xmmap(NULL
, win
->len
,
847 PROT_READ
, MAP_PRIVATE
,
848 p
->pack_fd
, win
->offset
);
849 if (win
->base
== MAP_FAILED
)
850 die("packfile %s cannot be mapped: %s",
855 if (pack_mapped
> peak_pack_mapped
)
856 peak_pack_mapped
= pack_mapped
;
857 if (pack_open_windows
> peak_pack_open_windows
)
858 peak_pack_open_windows
= pack_open_windows
;
859 win
->next
= p
->windows
;
863 if (win
!= *w_cursor
) {
864 win
->last_used
= pack_used_ctr
++;
868 offset
-= win
->offset
;
870 *left
= win
->len
- xsize_t(offset
);
871 return win
->base
+ offset
;
874 static struct packed_git
*alloc_packed_git(int extra
)
876 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
877 memset(p
, 0, sizeof(*p
));
882 static void try_to_free_pack_memory(size_t size
)
884 release_pack_memory(size
, -1);
887 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
889 static int have_set_try_to_free_routine
;
891 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
893 if (!have_set_try_to_free_routine
) {
894 have_set_try_to_free_routine
= 1;
895 set_try_to_free_routine(try_to_free_pack_memory
);
899 * Make sure a corresponding .pack file exists and that
900 * the index looks sane.
902 path_len
-= strlen(".idx");
907 memcpy(p
->pack_name
, path
, path_len
);
909 strcpy(p
->pack_name
+ path_len
, ".keep");
910 if (!access(p
->pack_name
, F_OK
))
913 strcpy(p
->pack_name
+ path_len
, ".pack");
914 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
919 /* ok, it looks sane as far as we can check without
920 * actually mapping the pack file.
922 p
->pack_size
= st
.st_size
;
923 p
->pack_local
= local
;
924 p
->mtime
= st
.st_mtime
;
925 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
930 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
932 const char *path
= sha1_pack_name(sha1
);
933 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
935 strcpy(p
->pack_name
, path
);
936 hashcpy(p
->sha1
, sha1
);
937 if (check_packed_git_idx(idx_path
, p
)) {
945 void install_packed_git(struct packed_git
*pack
)
947 if (pack
->pack_fd
!= -1)
950 pack
->next
= packed_git
;
954 static void prepare_packed_git_one(char *objdir
, int local
)
956 /* Ensure that this buffer is large enough so that we can
957 append "/pack/" without clobbering the stack even if
958 strlen(objdir) were PATH_MAX. */
959 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
964 sprintf(path
, "%s/pack", objdir
);
969 error("unable to open object pack directory: %s: %s",
970 path
, strerror(errno
));
974 while ((de
= readdir(dir
)) != NULL
) {
975 int namelen
= strlen(de
->d_name
);
976 struct packed_git
*p
;
978 if (!has_extension(de
->d_name
, ".idx"))
981 if (len
+ namelen
+ 1 > sizeof(path
))
984 /* Don't reopen a pack we already have. */
985 strcpy(path
+ len
, de
->d_name
);
986 for (p
= packed_git
; p
; p
= p
->next
) {
987 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
992 /* See if it really is a valid .idx file with corresponding
993 * .pack file that we can map.
995 p
= add_packed_git(path
, len
+ namelen
, local
);
998 install_packed_git(p
);
1003 static int sort_pack(const void *a_
, const void *b_
)
1005 struct packed_git
*a
= *((struct packed_git
**)a_
);
1006 struct packed_git
*b
= *((struct packed_git
**)b_
);
1010 * Local packs tend to contain objects specific to our
1011 * variant of the project than remote ones. In addition,
1012 * remote ones could be on a network mounted filesystem.
1013 * Favor local ones for these reasons.
1015 st
= a
->pack_local
- b
->pack_local
;
1020 * Younger packs tend to contain more recent objects,
1021 * and more recent objects tend to get accessed more
1024 if (a
->mtime
< b
->mtime
)
1026 else if (a
->mtime
== b
->mtime
)
1031 static void rearrange_packed_git(void)
1033 struct packed_git
**ary
, *p
;
1036 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1041 /* prepare an array of packed_git for easier sorting */
1042 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1043 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1046 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1048 /* link them back again */
1049 for (i
= 0; i
< n
- 1; i
++)
1050 ary
[i
]->next
= ary
[i
+ 1];
1051 ary
[n
- 1]->next
= NULL
;
1052 packed_git
= ary
[0];
1057 static int prepare_packed_git_run_once
= 0;
1058 void prepare_packed_git(void)
1060 struct alternate_object_database
*alt
;
1062 if (prepare_packed_git_run_once
)
1064 prepare_packed_git_one(get_object_directory(), 1);
1066 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1068 prepare_packed_git_one(alt
->base
, 0);
1069 alt
->name
[-1] = '/';
1071 rearrange_packed_git();
1072 prepare_packed_git_run_once
= 1;
1075 void reprepare_packed_git(void)
1078 prepare_packed_git_run_once
= 0;
1079 prepare_packed_git();
1082 static void mark_bad_packed_object(struct packed_git
*p
,
1083 const unsigned char *sha1
)
1086 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1087 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1089 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1090 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1091 p
->num_bad_objects
++;
1094 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1096 struct packed_git
*p
;
1099 for (p
= packed_git
; p
; p
= p
->next
)
1100 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1101 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1106 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1108 unsigned char real_sha1
[20];
1109 hash_sha1_file(map
, size
, type
, real_sha1
);
1110 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1113 static int git_open_noatime(const char *name
, struct packed_git
*p
)
1115 static int sha1_file_open_flag
= O_NOATIME
;
1118 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1122 /* Might the failure be due to O_NOATIME? */
1123 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1124 sha1_file_open_flag
= 0;
1132 static int open_sha1_file(const unsigned char *sha1
)
1135 char *name
= sha1_file_name(sha1
);
1136 struct alternate_object_database
*alt
;
1138 fd
= git_open_noatime(name
, NULL
);
1144 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1146 fill_sha1_path(name
, sha1
);
1147 fd
= git_open_noatime(alt
->base
, NULL
);
1154 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1159 fd
= open_sha1_file(sha1
);
1164 if (!fstat(fd
, &st
)) {
1165 *size
= xsize_t(st
.st_size
);
1166 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1173 static int legacy_loose_object(unsigned char *map
)
1178 * Is it a zlib-compressed buffer? If so, the first byte
1179 * must be 0x78 (15-bit window size, deflated), and the
1180 * first 16-bit word is evenly divisible by 31
1182 word
= (map
[0] << 8) + map
[1];
1183 if (map
[0] == 0x78 && !(word
% 31))
1189 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1190 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1193 unsigned long size
, c
;
1194 unsigned long used
= 0;
1197 *type
= (c
>> 4) & 7;
1201 if (len
<= used
|| bitsizeof(long) <= shift
) {
1202 error("bad object header");
1206 size
+= (c
& 0x7f) << shift
;
1213 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1215 unsigned long size
, used
;
1216 static const char valid_loose_object_type
[8] = {
1218 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1219 0, /* "delta" and others are invalid in a loose object */
1221 enum object_type type
;
1223 /* Get the data stream */
1224 memset(stream
, 0, sizeof(*stream
));
1225 stream
->next_in
= map
;
1226 stream
->avail_in
= mapsize
;
1227 stream
->next_out
= buffer
;
1228 stream
->avail_out
= bufsiz
;
1230 if (legacy_loose_object(map
)) {
1231 git_inflate_init(stream
);
1232 return git_inflate(stream
, 0);
1237 * There used to be a second loose object header format which
1238 * was meant to mimic the in-pack format, allowing for direct
1239 * copy of the object data. This format turned up not to be
1240 * really worth it and we don't write it any longer. But we
1241 * can still read it.
1243 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1244 if (!used
|| !valid_loose_object_type
[type
])
1249 /* Set up the stream for the rest.. */
1250 stream
->next_in
= map
;
1251 stream
->avail_in
= mapsize
;
1252 git_inflate_init(stream
);
1254 /* And generate the fake traditional header */
1255 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1256 typename(type
), size
);
1260 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1262 int bytes
= strlen(buffer
) + 1;
1263 unsigned char *buf
= xmallocz(size
);
1267 n
= stream
->total_out
- bytes
;
1270 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1272 if (bytes
<= size
) {
1274 * The above condition must be (bytes <= size), not
1275 * (bytes < size). In other words, even though we
1276 * expect no more output and set avail_out to zer0,
1277 * the input zlib stream may have bytes that express
1278 * "this concludes the stream", and we *do* want to
1281 * Otherwise we would not be able to test that we
1282 * consumed all the input to reach the expected size;
1283 * we also want to check that zlib tells us that all
1284 * went well with status == Z_STREAM_END at the end.
1286 stream
->next_out
= buf
+ bytes
;
1287 stream
->avail_out
= size
- bytes
;
1288 while (status
== Z_OK
)
1289 status
= git_inflate(stream
, Z_FINISH
);
1291 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1292 git_inflate_end(stream
);
1297 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1298 else if (stream
->avail_in
)
1299 error("garbage at end of loose object '%s'",
1306 * We used to just use "sscanf()", but that's actually way
1307 * too permissive for what we want to check. So do an anal
1308 * object header parse by hand.
1310 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1317 * The type can be at most ten bytes (including the
1318 * terminating '\0' that we add), and is followed by
1327 if (i
>= sizeof(type
))
1333 * The length must follow immediately, and be in canonical
1334 * decimal format (ie "010" is not valid).
1336 size
= *hdr
++ - '0';
1341 unsigned long c
= *hdr
- '0';
1345 size
= size
* 10 + c
;
1351 * The length must be followed by a zero byte
1353 return *hdr
? -1 : type_from_string(type
);
1356 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1362 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1363 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1366 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1369 unsigned long get_size_from_delta(struct packed_git
*p
,
1370 struct pack_window
**w_curs
,
1373 const unsigned char *data
;
1374 unsigned char delta_head
[20], *in
;
1378 memset(&stream
, 0, sizeof(stream
));
1379 stream
.next_out
= delta_head
;
1380 stream
.avail_out
= sizeof(delta_head
);
1382 git_inflate_init(&stream
);
1384 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1385 stream
.next_in
= in
;
1386 st
= git_inflate(&stream
, Z_FINISH
);
1387 curpos
+= stream
.next_in
- in
;
1388 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1389 stream
.total_out
< sizeof(delta_head
));
1390 git_inflate_end(&stream
);
1391 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1392 error("delta data unpack-initial failed");
1396 /* Examine the initial part of the delta to figure out
1401 /* ignore base size */
1402 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1404 /* Read the result size */
1405 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1408 static off_t
get_delta_base(struct packed_git
*p
,
1409 struct pack_window
**w_curs
,
1411 enum object_type type
,
1412 off_t delta_obj_offset
)
1414 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1417 /* use_pack() assured us we have [base_info, base_info + 20)
1418 * as a range that we can look at without walking off the
1419 * end of the mapped window. Its actually the hash size
1420 * that is assured. An OFS_DELTA longer than the hash size
1421 * is stupid, as then a REF_DELTA would be smaller to store.
1423 if (type
== OBJ_OFS_DELTA
) {
1425 unsigned char c
= base_info
[used
++];
1426 base_offset
= c
& 127;
1429 if (!base_offset
|| MSB(base_offset
, 7))
1430 return 0; /* overflow */
1431 c
= base_info
[used
++];
1432 base_offset
= (base_offset
<< 7) + (c
& 127);
1434 base_offset
= delta_obj_offset
- base_offset
;
1435 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1436 return 0; /* out of bound */
1438 } else if (type
== OBJ_REF_DELTA
) {
1439 /* The base entry _must_ be in the same pack */
1440 base_offset
= find_pack_entry_one(base_info
, p
);
1443 die("I am totally screwed");
1447 /* forward declaration for a mutually recursive function */
1448 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1449 unsigned long *sizep
);
1451 static int packed_delta_info(struct packed_git
*p
,
1452 struct pack_window
**w_curs
,
1454 enum object_type type
,
1456 unsigned long *sizep
)
1460 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1463 type
= packed_object_info(p
, base_offset
, NULL
);
1464 if (type
<= OBJ_NONE
) {
1465 struct revindex_entry
*revidx
;
1466 const unsigned char *base_sha1
;
1467 revidx
= find_pack_revindex(p
, base_offset
);
1470 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1471 mark_bad_packed_object(p
, base_sha1
);
1472 type
= sha1_object_info(base_sha1
, NULL
);
1473 if (type
<= OBJ_NONE
)
1477 /* We choose to only get the type of the base object and
1478 * ignore potentially corrupt pack file that expects the delta
1479 * based on a base with a wrong size. This saves tons of
1483 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1491 static int unpack_object_header(struct packed_git
*p
,
1492 struct pack_window
**w_curs
,
1494 unsigned long *sizep
)
1496 unsigned char *base
;
1499 enum object_type type
;
1501 /* use_pack() assures us we have [base, base + 20) available
1502 * as a range that we can look at at. (Its actually the hash
1503 * size that is assured.) With our object header encoding
1504 * the maximum deflated object size is 2^137, which is just
1505 * insane, so we know won't exceed what we have been given.
1507 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1508 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1517 const char *packed_object_info_detail(struct packed_git
*p
,
1519 unsigned long *size
,
1520 unsigned long *store_size
,
1521 unsigned int *delta_chain_length
,
1522 unsigned char *base_sha1
)
1524 struct pack_window
*w_curs
= NULL
;
1526 unsigned long dummy
;
1527 unsigned char *next_sha1
;
1528 enum object_type type
;
1529 struct revindex_entry
*revidx
;
1531 *delta_chain_length
= 0;
1532 curpos
= obj_offset
;
1533 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1535 revidx
= find_pack_revindex(p
, obj_offset
);
1536 *store_size
= revidx
[1].offset
- obj_offset
;
1541 die("pack %s contains unknown object type %d",
1542 p
->pack_name
, type
);
1547 unuse_pack(&w_curs
);
1548 return typename(type
);
1550 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1552 die("pack %s contains bad delta base reference of type %s",
1553 p
->pack_name
, typename(type
));
1554 if (*delta_chain_length
== 0) {
1555 revidx
= find_pack_revindex(p
, obj_offset
);
1556 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1560 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1561 if (*delta_chain_length
== 0)
1562 hashcpy(base_sha1
, next_sha1
);
1563 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1566 (*delta_chain_length
)++;
1567 curpos
= obj_offset
;
1568 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1572 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1573 unsigned long *sizep
)
1575 struct pack_window
*w_curs
= NULL
;
1577 off_t curpos
= obj_offset
;
1578 enum object_type type
;
1580 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1585 type
= packed_delta_info(p
, &w_curs
, curpos
,
1586 type
, obj_offset
, sizep
);
1596 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1597 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1600 unuse_pack(&w_curs
);
1604 static void *unpack_compressed_entry(struct packed_git
*p
,
1605 struct pack_window
**w_curs
,
1611 unsigned char *buffer
, *in
;
1613 buffer
= xmallocz(size
);
1614 memset(&stream
, 0, sizeof(stream
));
1615 stream
.next_out
= buffer
;
1616 stream
.avail_out
= size
+ 1;
1618 git_inflate_init(&stream
);
1620 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1621 stream
.next_in
= in
;
1622 st
= git_inflate(&stream
, Z_FINISH
);
1623 if (!stream
.avail_out
)
1624 break; /* the payload is larger than it should be */
1625 curpos
+= stream
.next_in
- in
;
1626 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1627 git_inflate_end(&stream
);
1628 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1636 #define MAX_DELTA_CACHE (256)
1638 static size_t delta_base_cached
;
1640 static struct delta_base_cache_lru_list
{
1641 struct delta_base_cache_lru_list
*prev
;
1642 struct delta_base_cache_lru_list
*next
;
1643 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1645 static struct delta_base_cache_entry
{
1646 struct delta_base_cache_lru_list lru
;
1648 struct packed_git
*p
;
1651 enum object_type type
;
1652 } delta_base_cache
[MAX_DELTA_CACHE
];
1654 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1658 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1659 hash
+= (hash
>> 8) + (hash
>> 16);
1660 return hash
% MAX_DELTA_CACHE
;
1663 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1664 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1667 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1668 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1671 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1672 return unpack_entry(p
, base_offset
, type
, base_size
);
1676 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1677 ent
->lru
.prev
->next
= ent
->lru
.next
;
1678 delta_base_cached
-= ent
->size
;
1680 ret
= xmemdupz(ent
->data
, ent
->size
);
1683 *base_size
= ent
->size
;
1687 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1692 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1693 ent
->lru
.prev
->next
= ent
->lru
.next
;
1694 delta_base_cached
-= ent
->size
;
1698 void clear_delta_base_cache(void)
1701 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1702 release_delta_base_cache(&delta_base_cache
[p
]);
1705 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1706 void *base
, unsigned long base_size
, enum object_type type
)
1708 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1709 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1710 struct delta_base_cache_lru_list
*lru
;
1712 release_delta_base_cache(ent
);
1713 delta_base_cached
+= base_size
;
1715 for (lru
= delta_base_cache_lru
.next
;
1716 delta_base_cached
> delta_base_cache_limit
1717 && lru
!= &delta_base_cache_lru
;
1719 struct delta_base_cache_entry
*f
= (void *)lru
;
1720 if (f
->type
== OBJ_BLOB
)
1721 release_delta_base_cache(f
);
1723 for (lru
= delta_base_cache_lru
.next
;
1724 delta_base_cached
> delta_base_cache_limit
1725 && lru
!= &delta_base_cache_lru
;
1727 struct delta_base_cache_entry
*f
= (void *)lru
;
1728 release_delta_base_cache(f
);
1732 ent
->base_offset
= base_offset
;
1735 ent
->size
= base_size
;
1736 ent
->lru
.next
= &delta_base_cache_lru
;
1737 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1738 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1739 delta_base_cache_lru
.prev
= &ent
->lru
;
1742 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1743 unsigned long *size
);
1745 static void *unpack_delta_entry(struct packed_git
*p
,
1746 struct pack_window
**w_curs
,
1748 unsigned long delta_size
,
1750 enum object_type
*type
,
1751 unsigned long *sizep
)
1753 void *delta_data
, *result
, *base
;
1754 unsigned long base_size
;
1757 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1759 error("failed to validate delta base reference "
1760 "at offset %"PRIuMAX
" from %s",
1761 (uintmax_t)curpos
, p
->pack_name
);
1765 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1768 * We're probably in deep shit, but let's try to fetch
1769 * the required base anyway from another pack or loose.
1770 * This is costly but should happen only in the presence
1771 * of a corrupted pack, and is better than failing outright.
1773 struct revindex_entry
*revidx
;
1774 const unsigned char *base_sha1
;
1775 revidx
= find_pack_revindex(p
, base_offset
);
1778 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1779 error("failed to read delta base object %s"
1780 " at offset %"PRIuMAX
" from %s",
1781 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1783 mark_bad_packed_object(p
, base_sha1
);
1784 base
= read_object(base_sha1
, type
, &base_size
);
1789 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1791 error("failed to unpack compressed delta "
1792 "at offset %"PRIuMAX
" from %s",
1793 (uintmax_t)curpos
, p
->pack_name
);
1797 result
= patch_delta(base
, base_size
,
1798 delta_data
, delta_size
,
1801 die("failed to apply delta");
1803 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1807 int do_check_packed_object_crc
;
1809 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1810 enum object_type
*type
, unsigned long *sizep
)
1812 struct pack_window
*w_curs
= NULL
;
1813 off_t curpos
= obj_offset
;
1816 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1817 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1818 unsigned long len
= revidx
[1].offset
- obj_offset
;
1819 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1820 const unsigned char *sha1
=
1821 nth_packed_object_sha1(p
, revidx
->nr
);
1822 error("bad packed object CRC for %s",
1824 mark_bad_packed_object(p
, sha1
);
1825 unuse_pack(&w_curs
);
1830 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1834 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1835 obj_offset
, type
, sizep
);
1841 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1845 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1846 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1848 unuse_pack(&w_curs
);
1852 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1855 const unsigned char *index
= p
->index_data
;
1857 if (open_pack_index(p
))
1859 index
= p
->index_data
;
1861 if (n
>= p
->num_objects
)
1864 if (p
->index_version
== 1) {
1865 return index
+ 24 * n
+ 4;
1868 return index
+ 20 * n
;
1872 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1874 const unsigned char *index
= p
->index_data
;
1876 if (p
->index_version
== 1) {
1877 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1880 index
+= 8 + p
->num_objects
* (20 + 4);
1881 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1882 if (!(off
& 0x80000000))
1884 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1885 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1886 ntohl(*((uint32_t *)(index
+ 4)));
1890 off_t
find_pack_entry_one(const unsigned char *sha1
,
1891 struct packed_git
*p
)
1893 const uint32_t *level1_ofs
= p
->index_data
;
1894 const unsigned char *index
= p
->index_data
;
1895 unsigned hi
, lo
, stride
;
1896 static int use_lookup
= -1;
1897 static int debug_lookup
= -1;
1899 if (debug_lookup
< 0)
1900 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1903 if (open_pack_index(p
))
1905 level1_ofs
= p
->index_data
;
1906 index
= p
->index_data
;
1908 if (p
->index_version
> 1) {
1913 hi
= ntohl(level1_ofs
[*sha1
]);
1914 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1915 if (p
->index_version
> 1) {
1923 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1924 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1927 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1929 int pos
= sha1_entry_pos(index
, stride
, 0,
1930 lo
, hi
, p
->num_objects
, sha1
);
1933 return nth_packed_object_offset(p
, pos
);
1937 unsigned mi
= (lo
+ hi
) / 2;
1938 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1941 printf("lo %u hi %u rg %u mi %u\n",
1942 lo
, hi
, hi
- lo
, mi
);
1944 return nth_packed_object_offset(p
, mi
);
1953 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1955 static struct packed_git
*last_found
= (void *)1;
1956 struct packed_git
*p
;
1959 prepare_packed_git();
1962 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1965 if (p
->num_bad_objects
) {
1967 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1968 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1972 offset
= find_pack_entry_one(sha1
, p
);
1975 * We are about to tell the caller where they can
1976 * locate the requested object. We better make
1977 * sure the packfile is still here and can be
1978 * accessed before supplying that answer, as
1979 * it may have been deleted since the index
1982 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1983 error("packfile %s cannot be accessed", p
->pack_name
);
1988 hashcpy(e
->sha1
, sha1
);
1994 if (p
== last_found
)
1998 if (p
== last_found
)
2004 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2005 struct packed_git
*packs
)
2007 struct packed_git
*p
;
2009 for (p
= packs
; p
; p
= p
->next
) {
2010 if (find_pack_entry_one(sha1
, p
))
2017 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2020 unsigned long mapsize
, size
;
2025 map
= map_sha1_file(sha1
, &mapsize
);
2027 return error("unable to find %s", sha1_to_hex(sha1
));
2028 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2029 status
= error("unable to unpack %s header",
2031 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2032 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2035 git_inflate_end(&stream
);
2036 munmap(map
, mapsize
);
2040 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2042 struct cached_object
*co
;
2043 struct pack_entry e
;
2046 co
= find_cached_object(sha1
);
2053 if (!find_pack_entry(sha1
, &e
)) {
2054 /* Most likely it's a loose object. */
2055 status
= sha1_loose_object_info(sha1
, sizep
);
2059 /* Not a loose object; someone else may have just packed it. */
2060 reprepare_packed_git();
2061 if (!find_pack_entry(sha1
, &e
))
2065 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
2067 mark_bad_packed_object(e
.p
, sha1
);
2068 status
= sha1_object_info(sha1
, sizep
);
2074 static void *read_packed_sha1(const unsigned char *sha1
,
2075 enum object_type
*type
, unsigned long *size
)
2077 struct pack_entry e
;
2080 if (!find_pack_entry(sha1
, &e
))
2082 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2085 * We're probably in deep shit, but let's try to fetch
2086 * the required object anyway from another pack or loose.
2087 * This should happen only in the presence of a corrupted
2088 * pack, and is better than failing outright.
2090 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2091 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2092 mark_bad_packed_object(e
.p
, sha1
);
2093 data
= read_object(sha1
, type
, size
);
2098 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2099 unsigned char *sha1
)
2101 struct cached_object
*co
;
2103 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2104 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2106 if (cached_object_alloc
<= cached_object_nr
) {
2107 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2108 cached_objects
= xrealloc(cached_objects
,
2109 sizeof(*cached_objects
) *
2110 cached_object_alloc
);
2112 co
= &cached_objects
[cached_object_nr
++];
2115 co
->buf
= xmalloc(len
);
2116 memcpy(co
->buf
, buf
, len
);
2117 hashcpy(co
->sha1
, sha1
);
2121 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2122 unsigned long *size
)
2124 unsigned long mapsize
;
2126 struct cached_object
*co
;
2128 co
= find_cached_object(sha1
);
2132 return xmemdupz(co
->buf
, co
->size
);
2135 buf
= read_packed_sha1(sha1
, type
, size
);
2138 map
= map_sha1_file(sha1
, &mapsize
);
2140 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2141 munmap(map
, mapsize
);
2144 reprepare_packed_git();
2145 return read_packed_sha1(sha1
, type
, size
);
2149 * This function dies on corrupt objects; the callers who want to
2150 * deal with them should arrange to call read_object() and give error
2151 * messages themselves.
2153 void *read_sha1_file_repl(const unsigned char *sha1
,
2154 enum object_type
*type
,
2155 unsigned long *size
,
2156 const unsigned char **replacement
)
2158 const unsigned char *repl
= lookup_replace_object(sha1
);
2161 const struct packed_git
*p
;
2164 data
= read_object(repl
, type
, size
);
2167 *replacement
= repl
;
2171 if (errno
&& errno
!= ENOENT
)
2172 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2174 /* die if we replaced an object with one that does not exist */
2176 die("replacement %s not found for %s",
2177 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2179 if (has_loose_object(repl
)) {
2180 path
= sha1_file_name(sha1
);
2181 die("loose object %s (stored in %s) is corrupt",
2182 sha1_to_hex(repl
), path
);
2185 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2186 die("packed object %s (stored in %s) is corrupt",
2187 sha1_to_hex(repl
), p
->pack_name
);
2192 void *read_object_with_reference(const unsigned char *sha1
,
2193 const char *required_type_name
,
2194 unsigned long *size
,
2195 unsigned char *actual_sha1_return
)
2197 enum object_type type
, required_type
;
2199 unsigned long isize
;
2200 unsigned char actual_sha1
[20];
2202 required_type
= type_from_string(required_type_name
);
2203 hashcpy(actual_sha1
, sha1
);
2205 int ref_length
= -1;
2206 const char *ref_type
= NULL
;
2208 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2211 if (type
== required_type
) {
2213 if (actual_sha1_return
)
2214 hashcpy(actual_sha1_return
, actual_sha1
);
2217 /* Handle references */
2218 else if (type
== OBJ_COMMIT
)
2220 else if (type
== OBJ_TAG
)
2221 ref_type
= "object ";
2226 ref_length
= strlen(ref_type
);
2228 if (ref_length
+ 40 > isize
||
2229 memcmp(buffer
, ref_type
, ref_length
) ||
2230 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2235 /* Now we have the ID of the referred-to object in
2236 * actual_sha1. Check again. */
2240 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2241 const char *type
, unsigned char *sha1
,
2242 char *hdr
, int *hdrlen
)
2246 /* Generate the header */
2247 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2251 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2252 git_SHA1_Update(&c
, buf
, len
);
2253 git_SHA1_Final(sha1
, &c
);
2257 * Move the just written object into its final resting place.
2258 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2259 * "moving" is only a part of what it does, when no patch between
2260 * master to pu changes the call sites of this function.
2262 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2266 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2268 else if (link(tmpfile
, filename
))
2272 * Coda hack - coda doesn't like cross-directory links,
2273 * so we fall back to a rename, which will mean that it
2274 * won't be able to check collisions, but that's not a
2277 * The same holds for FAT formatted media.
2279 * When this succeeds, we just return. We have nothing
2282 if (ret
&& ret
!= EEXIST
) {
2284 if (!rename(tmpfile
, filename
))
2288 unlink_or_warn(tmpfile
);
2290 if (ret
!= EEXIST
) {
2291 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2293 /* FIXME!!! Collision check here ? */
2297 if (adjust_shared_perm(filename
))
2298 return error("unable to set permission to '%s'", filename
);
2302 static int write_buffer(int fd
, const void *buf
, size_t len
)
2304 if (write_in_full(fd
, buf
, len
) < 0)
2305 return error("file write error (%s)", strerror(errno
));
2309 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2310 unsigned char *sha1
)
2314 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2318 /* Finalize a file on disk, and close it. */
2319 static void close_sha1_file(int fd
)
2321 if (fsync_object_files
)
2322 fsync_or_die(fd
, "sha1 file");
2324 die_errno("error when closing sha1 file");
2327 /* Size of directory component, including the ending '/' */
2328 static inline int directory_size(const char *filename
)
2330 const char *s
= strrchr(filename
, '/');
2333 return s
- filename
+ 1;
2337 * This creates a temporary file in the same directory as the final
2340 * We want to avoid cross-directory filename renames, because those
2341 * can have problems on various filesystems (FAT, NFS, Coda).
2343 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2345 int fd
, dirlen
= directory_size(filename
);
2347 if (dirlen
+ 20 > bufsiz
) {
2348 errno
= ENAMETOOLONG
;
2351 memcpy(buffer
, filename
, dirlen
);
2352 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2353 fd
= git_mkstemp_mode(buffer
, 0444);
2354 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2355 /* Make sure the directory exists */
2356 memcpy(buffer
, filename
, dirlen
);
2357 buffer
[dirlen
-1] = 0;
2358 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2362 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2363 fd
= git_mkstemp_mode(buffer
, 0444);
2368 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2369 const void *buf
, unsigned long len
, time_t mtime
)
2372 unsigned char compressed
[4096];
2375 unsigned char parano_sha1
[20];
2377 static char tmpfile
[PATH_MAX
];
2379 filename
= sha1_file_name(sha1
);
2380 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2382 if (errno
== EACCES
)
2383 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2385 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2389 memset(&stream
, 0, sizeof(stream
));
2390 deflateInit(&stream
, zlib_compression_level
);
2391 stream
.next_out
= compressed
;
2392 stream
.avail_out
= sizeof(compressed
);
2395 /* First header.. */
2396 stream
.next_in
= (unsigned char *)hdr
;
2397 stream
.avail_in
= hdrlen
;
2398 while (deflate(&stream
, 0) == Z_OK
)
2400 git_SHA1_Update(&c
, hdr
, hdrlen
);
2402 /* Then the data itself.. */
2403 stream
.next_in
= (void *)buf
;
2404 stream
.avail_in
= len
;
2406 unsigned char *in0
= stream
.next_in
;
2407 ret
= deflate(&stream
, Z_FINISH
);
2408 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2409 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2410 die("unable to write sha1 file");
2411 stream
.next_out
= compressed
;
2412 stream
.avail_out
= sizeof(compressed
);
2413 } while (ret
== Z_OK
);
2415 if (ret
!= Z_STREAM_END
)
2416 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2417 ret
= deflateEnd(&stream
);
2419 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2420 git_SHA1_Final(parano_sha1
, &c
);
2421 if (hashcmp(sha1
, parano_sha1
) != 0)
2422 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2424 close_sha1_file(fd
);
2429 utb
.modtime
= mtime
;
2430 if (utime(tmpfile
, &utb
) < 0)
2431 warning("failed utime() on %s: %s",
2432 tmpfile
, strerror(errno
));
2435 return move_temp_to_file(tmpfile
, filename
);
2438 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2440 unsigned char sha1
[20];
2444 /* Normally if we have it in the pack then we do not bother writing
2445 * it out into .git/objects/??/?{38} file.
2447 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2449 hashcpy(returnsha1
, sha1
);
2450 if (has_sha1_file(sha1
))
2452 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2455 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2459 enum object_type type
;
2464 if (has_loose_object(sha1
))
2466 buf
= read_packed_sha1(sha1
, &type
, &len
);
2468 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2469 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2470 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2476 int has_pack_index(const unsigned char *sha1
)
2479 if (stat(sha1_pack_index_name(sha1
), &st
))
2484 int has_sha1_pack(const unsigned char *sha1
)
2486 struct pack_entry e
;
2487 return find_pack_entry(sha1
, &e
);
2490 int has_sha1_file(const unsigned char *sha1
)
2492 struct pack_entry e
;
2494 if (find_pack_entry(sha1
, &e
))
2496 return has_loose_object(sha1
);
2499 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2500 int write_object
, enum object_type type
, const char *path
)
2502 int ret
, re_allocated
= 0;
2508 * Convert blobs to git internal format
2510 if ((type
== OBJ_BLOB
) && path
) {
2511 struct strbuf nbuf
= STRBUF_INIT
;
2512 if (convert_to_git(path
, buf
, size
, &nbuf
,
2513 write_object
? safe_crlf
: 0)) {
2514 buf
= strbuf_detach(&nbuf
, &size
);
2520 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2522 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2528 #define SMALL_FILE_SIZE (32*1024)
2530 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2531 enum object_type type
, const char *path
)
2534 size_t size
= xsize_t(st
->st_size
);
2536 if (!S_ISREG(st
->st_mode
)) {
2537 struct strbuf sbuf
= STRBUF_INIT
;
2538 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2539 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2543 strbuf_release(&sbuf
);
2545 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2546 } else if (size
<= SMALL_FILE_SIZE
) {
2547 char *buf
= xmalloc(size
);
2548 if (size
== read_in_full(fd
, buf
, size
))
2549 ret
= index_mem(sha1
, buf
, size
, write_object
, type
,
2552 ret
= error("short read %s", strerror(errno
));
2555 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2556 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2563 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2566 struct strbuf sb
= STRBUF_INIT
;
2568 switch (st
->st_mode
& S_IFMT
) {
2570 fd
= open(path
, O_RDONLY
);
2572 return error("open(\"%s\"): %s", path
,
2574 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2575 return error("%s: failed to insert into database",
2579 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2580 char *errstr
= strerror(errno
);
2581 return error("readlink(\"%s\"): %s", path
,
2585 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2586 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2587 return error("%s: failed to insert into database",
2589 strbuf_release(&sb
);
2592 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2594 return error("%s: unsupported file type", path
);
2599 int read_pack_header(int fd
, struct pack_header
*header
)
2601 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2602 /* "eof before pack header was fully read" */
2603 return PH_ERROR_EOF
;
2605 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2606 /* "protocol error (pack signature mismatch detected)" */
2607 return PH_ERROR_PACK_SIGNATURE
;
2608 if (!pack_version_ok(header
->hdr_version
))
2609 /* "protocol error (pack version unsupported)" */
2610 return PH_ERROR_PROTOCOL
;
2614 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2616 enum object_type type
= sha1_object_info(sha1
, NULL
);
2618 die("%s is not a valid object", sha1_to_hex(sha1
));
2620 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),