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 int safe_create_leading_directories(char *path
)
40 char *pos
= path
+ offset_1st_component(path
);
44 pos
= strchr(pos
, '/');
52 if (!stat(path
, &st
)) {
54 if (!S_ISDIR(st
.st_mode
)) {
59 else if (mkdir(path
, 0777)) {
63 else if (adjust_shared_perm(path
)) {
72 int safe_create_leading_directories_const(const char *path
)
74 /* path points to cache entries, so xstrdup before messing with it */
75 char *buf
= xstrdup(path
);
76 int result
= safe_create_leading_directories(buf
);
81 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
84 for (i
= 0; i
< 20; i
++) {
85 static char hex
[] = "0123456789abcdef";
86 unsigned int val
= sha1
[i
];
87 char *pos
= pathbuf
+ i
*2 + (i
> 0);
88 *pos
++ = hex
[val
>> 4];
89 *pos
= hex
[val
& 0xf];
94 * NOTE! This returns a statically allocated buffer, so you have to be
95 * careful about using it. Do an "xstrdup()" if you need to save the
98 * Also note that this returns the location for creating. Reading
99 * SHA1 file can happen from any alternate directory listed in the
100 * DB_ENVIRONMENT environment variable if it is not found in
101 * the primary object database.
103 char *sha1_file_name(const unsigned char *sha1
)
105 static char buf
[PATH_MAX
];
109 objdir
= get_object_directory();
110 len
= strlen(objdir
);
112 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
113 if (len
+ 43 > PATH_MAX
)
114 die("insanely long object directory %s", objdir
);
115 memcpy(buf
, objdir
, len
);
119 fill_sha1_path(buf
+ len
+ 1, sha1
);
123 static char *sha1_get_pack_name(const unsigned char *sha1
,
124 char **name
, char **base
, const char *which
)
126 static const char hex
[] = "0123456789abcdef";
131 const char *sha1_file_directory
= get_object_directory();
132 int len
= strlen(sha1_file_directory
);
133 *base
= xmalloc(len
+ 60);
134 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
135 sha1_file_directory
, which
);
136 *name
= *base
+ len
+ 11;
141 for (i
= 0; i
< 20; i
++) {
142 unsigned int val
= *sha1
++;
143 *buf
++ = hex
[val
>> 4];
144 *buf
++ = hex
[val
& 0xf];
150 char *sha1_pack_name(const unsigned char *sha1
)
152 static char *name
, *base
;
154 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
157 char *sha1_pack_index_name(const unsigned char *sha1
)
159 static char *name
, *base
;
161 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
164 struct alternate_object_database
*alt_odb_list
;
165 static struct alternate_object_database
**alt_odb_tail
;
167 static void read_info_alternates(const char * alternates
, int depth
);
170 * Prepare alternate object database registry.
172 * The variable alt_odb_list points at the list of struct
173 * alternate_object_database. The elements on this list come from
174 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
175 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
176 * whose contents is similar to that environment variable but can be
177 * LF separated. Its base points at a statically allocated buffer that
178 * contains "/the/directory/corresponding/to/.git/objects/...", while
179 * its name points just after the slash at the end of ".git/objects/"
180 * in the example above, and has enough space to hold 40-byte hex
181 * SHA1, an extra slash for the first level indirection, and the
184 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
186 const char *objdir
= get_object_directory();
187 struct alternate_object_database
*ent
;
188 struct alternate_object_database
*alt
;
189 /* 43 = 40-byte + 2 '/' + terminating NUL */
191 int entlen
= pfxlen
+ 43;
194 if (!is_absolute_path(entry
) && relative_base
) {
195 /* Relative alt-odb */
197 base_len
= strlen(relative_base
) + 1;
201 ent
= xmalloc(sizeof(*ent
) + entlen
);
203 if (!is_absolute_path(entry
) && relative_base
) {
204 memcpy(ent
->base
, relative_base
, base_len
- 1);
205 ent
->base
[base_len
- 1] = '/';
206 memcpy(ent
->base
+ base_len
, entry
, len
);
209 memcpy(ent
->base
, entry
, pfxlen
);
211 ent
->name
= ent
->base
+ pfxlen
+ 1;
212 ent
->base
[pfxlen
+ 3] = '/';
213 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
215 /* Detect cases where alternate disappeared */
216 if (!is_directory(ent
->base
)) {
217 error("object directory %s does not exist; "
218 "check .git/objects/info/alternates.",
224 /* Prevent the common mistake of listing the same
225 * thing twice, or object directory itself.
227 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
228 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
233 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
238 /* add the alternate entry */
240 alt_odb_tail
= &(ent
->next
);
243 /* recursively add alternates */
244 read_info_alternates(ent
->base
, depth
+ 1);
246 ent
->base
[pfxlen
] = '/';
251 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
252 const char *relative_base
, int depth
)
254 const char *cp
, *last
;
257 error("%s: ignoring alternate object stores, nesting too deep.",
265 if (cp
< ep
&& *cp
== '#') {
266 while (cp
< ep
&& *cp
!= sep
)
271 while (cp
< ep
&& *cp
!= sep
)
274 if (!is_absolute_path(last
) && depth
) {
275 error("%s: ignoring relative alternate object store %s",
276 relative_base
, last
);
278 link_alt_odb_entry(last
, cp
- last
,
279 relative_base
, depth
);
282 while (cp
< ep
&& *cp
== sep
)
288 static void read_info_alternates(const char * relative_base
, int depth
)
293 const char alt_file_name
[] = "info/alternates";
294 /* Given that relative_base is no longer than PATH_MAX,
295 ensure that "path" has enough space to append "/", the
296 file name, "info/alternates", and a trailing NUL. */
297 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
300 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
301 fd
= open(path
, O_RDONLY
);
304 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
308 mapsz
= xsize_t(st
.st_size
);
309 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
312 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
317 void add_to_alternates_file(const char *reference
)
319 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
320 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
321 char *alt
= mkpath("%s/objects\n", reference
);
322 write_or_die(fd
, alt
, strlen(alt
));
323 if (commit_lock_file(lock
))
324 die("could not close alternates file");
326 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
329 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
331 struct alternate_object_database
*ent
;
334 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
339 void prepare_alt_odb(void)
346 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
349 alt_odb_tail
= &alt_odb_list
;
350 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
352 read_info_alternates(get_object_directory(), 0);
355 static int has_loose_object_local(const unsigned char *sha1
)
357 char *name
= sha1_file_name(sha1
);
358 return !access(name
, F_OK
);
361 int has_loose_object_nonlocal(const unsigned char *sha1
)
363 struct alternate_object_database
*alt
;
365 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
366 fill_sha1_path(alt
->name
, sha1
);
367 if (!access(alt
->base
, F_OK
))
373 static int has_loose_object(const unsigned char *sha1
)
375 return has_loose_object_local(sha1
) ||
376 has_loose_object_nonlocal(sha1
);
379 static unsigned int pack_used_ctr
;
380 static unsigned int pack_mmap_calls
;
381 static unsigned int peak_pack_open_windows
;
382 static unsigned int pack_open_windows
;
383 static size_t peak_pack_mapped
;
384 static size_t pack_mapped
;
385 struct packed_git
*packed_git
;
387 void pack_report(void)
390 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
391 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
392 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
393 sz_fmt(getpagesize()),
394 sz_fmt(packed_git_window_size
),
395 sz_fmt(packed_git_limit
));
397 "pack_report: pack_used_ctr = %10u\n"
398 "pack_report: pack_mmap_calls = %10u\n"
399 "pack_report: pack_open_windows = %10u / %10u\n"
400 "pack_report: pack_mapped = "
401 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
404 pack_open_windows
, peak_pack_open_windows
,
405 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
408 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
411 struct pack_idx_header
*hdr
;
413 uint32_t version
, nr
, i
, *index
;
414 int fd
= open(path
, O_RDONLY
);
419 if (fstat(fd
, &st
)) {
423 idx_size
= xsize_t(st
.st_size
);
424 if (idx_size
< 4 * 256 + 20 + 20) {
426 return error("index file %s is too small", path
);
428 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
432 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
433 version
= ntohl(hdr
->idx_version
);
434 if (version
< 2 || version
> 2) {
435 munmap(idx_map
, idx_size
);
436 return error("index file %s is version %"PRIu32
437 " and is not supported by this binary"
438 " (try upgrading GIT to a newer version)",
447 index
+= 2; /* skip index header */
448 for (i
= 0; i
< 256; i
++) {
449 uint32_t n
= ntohl(index
[i
]);
451 munmap(idx_map
, idx_size
);
452 return error("non-monotonic index %s", path
);
460 * - 256 index entries 4 bytes each
461 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
462 * - 20-byte SHA1 of the packfile
463 * - 20-byte SHA1 file checksum
465 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
466 munmap(idx_map
, idx_size
);
467 return error("wrong index v1 file size in %s", path
);
469 } else if (version
== 2) {
472 * - 8 bytes of header
473 * - 256 index entries 4 bytes each
474 * - 20-byte sha1 entry * nr
475 * - 4-byte crc entry * nr
476 * - 4-byte offset entry * nr
477 * - 20-byte SHA1 of the packfile
478 * - 20-byte SHA1 file checksum
479 * And after the 4-byte offset table might be a
480 * variable sized table containing 8-byte entries
481 * for offsets larger than 2^31.
483 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
484 unsigned long max_size
= min_size
;
486 max_size
+= (nr
- 1)*8;
487 if (idx_size
< min_size
|| idx_size
> max_size
) {
488 munmap(idx_map
, idx_size
);
489 return error("wrong index v2 file size in %s", path
);
491 if (idx_size
!= min_size
&&
493 * make sure we can deal with large pack offsets.
494 * 31-bit signed offset won't be enough, neither
495 * 32-bit unsigned one will be.
497 (sizeof(off_t
) <= 4)) {
498 munmap(idx_map
, idx_size
);
499 return error("pack too large for current definition of off_t in %s", path
);
503 p
->index_version
= version
;
504 p
->index_data
= idx_map
;
505 p
->index_size
= idx_size
;
510 int open_pack_index(struct packed_git
*p
)
518 idx_name
= xstrdup(p
->pack_name
);
519 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
520 ret
= check_packed_git_idx(idx_name
, p
);
525 static void scan_windows(struct packed_git
*p
,
526 struct packed_git
**lru_p
,
527 struct pack_window
**lru_w
,
528 struct pack_window
**lru_l
)
530 struct pack_window
*w
, *w_l
;
532 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
534 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
544 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
546 struct packed_git
*p
, *lru_p
= NULL
;
547 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
550 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
551 for (p
= packed_git
; p
; p
= p
->next
)
552 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
554 munmap(lru_w
->base
, lru_w
->len
);
555 pack_mapped
-= lru_w
->len
;
557 lru_l
->next
= lru_w
->next
;
559 lru_p
->windows
= lru_w
->next
;
560 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
561 close(lru_p
->pack_fd
);
572 void release_pack_memory(size_t need
, int fd
)
574 size_t cur
= pack_mapped
;
575 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
579 void close_pack_windows(struct packed_git
*p
)
582 struct pack_window
*w
= p
->windows
;
585 die("pack '%s' still has open windows to it",
587 munmap(w
->base
, w
->len
);
588 pack_mapped
-= w
->len
;
590 p
->windows
= w
->next
;
595 void unuse_pack(struct pack_window
**w_cursor
)
597 struct pack_window
*w
= *w_cursor
;
604 void close_pack_index(struct packed_git
*p
)
607 munmap((void *)p
->index_data
, p
->index_size
);
608 p
->index_data
= NULL
;
613 * This is used by git-repack in case a newly created pack happens to
614 * contain the same set of objects as an existing one. In that case
615 * the resulting file might be different even if its name would be the
616 * same. It is best to close any reference to the old pack before it is
617 * replaced on disk. Of course no index pointers nor windows for given pack
618 * must subsist at this point. If ever objects from this pack are requested
619 * again, the new version of the pack will be reinitialized through
620 * reprepare_packed_git().
622 void free_pack_by_name(const char *pack_name
)
624 struct packed_git
*p
, **pp
= &packed_git
;
628 if (strcmp(pack_name
, p
->pack_name
) == 0) {
629 clear_delta_base_cache();
630 close_pack_windows(p
);
631 if (p
->pack_fd
!= -1)
634 free(p
->bad_object_sha1
);
644 * Do not call this directly as this leaks p->pack_fd on error return;
645 * call open_packed_git() instead.
647 static int open_packed_git_1(struct packed_git
*p
)
650 struct pack_header hdr
;
651 unsigned char sha1
[20];
652 unsigned char *idx_sha1
;
655 if (!p
->index_data
&& open_pack_index(p
))
656 return error("packfile %s index unavailable", p
->pack_name
);
658 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
659 while (p
->pack_fd
< 0 && errno
== EMFILE
&& unuse_one_window(p
, -1))
660 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
661 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
664 /* If we created the struct before we had the pack we lack size. */
666 if (!S_ISREG(st
.st_mode
))
667 return error("packfile %s not a regular file", p
->pack_name
);
668 p
->pack_size
= st
.st_size
;
669 } else if (p
->pack_size
!= st
.st_size
)
670 return error("packfile %s size changed", p
->pack_name
);
672 /* We leave these file descriptors open with sliding mmap;
673 * there is no point keeping them open across exec(), though.
675 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
677 return error("cannot determine file descriptor flags");
678 fd_flag
|= FD_CLOEXEC
;
679 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
680 return error("cannot set FD_CLOEXEC");
682 /* Verify we recognize this pack file format. */
683 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
684 return error("file %s is far too short to be a packfile", p
->pack_name
);
685 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
686 return error("file %s is not a GIT packfile", p
->pack_name
);
687 if (!pack_version_ok(hdr
.hdr_version
))
688 return error("packfile %s is version %"PRIu32
" and not"
689 " supported (try upgrading GIT to a newer version)",
690 p
->pack_name
, ntohl(hdr
.hdr_version
));
692 /* Verify the pack matches its index. */
693 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
694 return error("packfile %s claims to have %"PRIu32
" objects"
695 " while index indicates %"PRIu32
" objects",
696 p
->pack_name
, ntohl(hdr
.hdr_entries
),
698 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
699 return error("end of packfile %s is unavailable", p
->pack_name
);
700 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
701 return error("packfile %s signature is unavailable", p
->pack_name
);
702 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
703 if (hashcmp(sha1
, idx_sha1
))
704 return error("packfile %s does not match index", p
->pack_name
);
708 static int open_packed_git(struct packed_git
*p
)
710 if (!open_packed_git_1(p
))
712 if (p
->pack_fd
!= -1) {
719 static int in_window(struct pack_window
*win
, off_t offset
)
721 /* We must promise at least 20 bytes (one hash) after the
722 * offset is available from this window, otherwise the offset
723 * is not actually in this window and a different window (which
724 * has that one hash excess) must be used. This is to support
725 * the object header and delta base parsing routines below.
727 off_t win_off
= win
->offset
;
728 return win_off
<= offset
729 && (offset
+ 20) <= (win_off
+ win
->len
);
732 unsigned char *use_pack(struct packed_git
*p
,
733 struct pack_window
**w_cursor
,
737 struct pack_window
*win
= *w_cursor
;
739 if (p
->pack_fd
== -1 && open_packed_git(p
))
740 die("packfile %s cannot be accessed", p
->pack_name
);
742 /* Since packfiles end in a hash of their content and it's
743 * pointless to ask for an offset into the middle of that
744 * hash, and the in_window function above wouldn't match
745 * don't allow an offset too close to the end of the file.
747 if (offset
> (p
->pack_size
- 20))
748 die("offset beyond end of packfile (truncated pack?)");
750 if (!win
|| !in_window(win
, offset
)) {
753 for (win
= p
->windows
; win
; win
= win
->next
) {
754 if (in_window(win
, offset
))
758 size_t window_align
= packed_git_window_size
/ 2;
760 win
= xcalloc(1, sizeof(*win
));
761 win
->offset
= (offset
/ window_align
) * window_align
;
762 len
= p
->pack_size
- win
->offset
;
763 if (len
> packed_git_window_size
)
764 len
= packed_git_window_size
;
765 win
->len
= (size_t)len
;
766 pack_mapped
+= win
->len
;
767 while (packed_git_limit
< pack_mapped
768 && unuse_one_window(p
, p
->pack_fd
))
770 win
->base
= xmmap(NULL
, win
->len
,
771 PROT_READ
, MAP_PRIVATE
,
772 p
->pack_fd
, win
->offset
);
773 if (win
->base
== MAP_FAILED
)
774 die("packfile %s cannot be mapped: %s",
779 if (pack_mapped
> peak_pack_mapped
)
780 peak_pack_mapped
= pack_mapped
;
781 if (pack_open_windows
> peak_pack_open_windows
)
782 peak_pack_open_windows
= pack_open_windows
;
783 win
->next
= p
->windows
;
787 if (win
!= *w_cursor
) {
788 win
->last_used
= pack_used_ctr
++;
792 offset
-= win
->offset
;
794 *left
= win
->len
- xsize_t(offset
);
795 return win
->base
+ offset
;
798 static struct packed_git
*alloc_packed_git(int extra
)
800 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
801 memset(p
, 0, sizeof(*p
));
806 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
809 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
812 * Make sure a corresponding .pack file exists and that
813 * the index looks sane.
815 path_len
-= strlen(".idx");
820 memcpy(p
->pack_name
, path
, path_len
);
822 strcpy(p
->pack_name
+ path_len
, ".keep");
823 if (!access(p
->pack_name
, F_OK
))
826 strcpy(p
->pack_name
+ path_len
, ".pack");
827 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
832 /* ok, it looks sane as far as we can check without
833 * actually mapping the pack file.
835 p
->pack_size
= st
.st_size
;
836 p
->pack_local
= local
;
837 p
->mtime
= st
.st_mtime
;
838 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
843 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
845 const char *path
= sha1_pack_name(sha1
);
846 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
848 strcpy(p
->pack_name
, path
);
849 hashcpy(p
->sha1
, sha1
);
850 if (check_packed_git_idx(idx_path
, p
)) {
858 void install_packed_git(struct packed_git
*pack
)
860 pack
->next
= packed_git
;
864 static void prepare_packed_git_one(char *objdir
, int local
)
866 /* Ensure that this buffer is large enough so that we can
867 append "/pack/" without clobbering the stack even if
868 strlen(objdir) were PATH_MAX. */
869 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
874 sprintf(path
, "%s/pack", objdir
);
877 while (!dir
&& errno
== EMFILE
&& unuse_one_window(packed_git
, -1))
881 error("unable to open object pack directory: %s: %s",
882 path
, strerror(errno
));
886 while ((de
= readdir(dir
)) != NULL
) {
887 int namelen
= strlen(de
->d_name
);
888 struct packed_git
*p
;
890 if (!has_extension(de
->d_name
, ".idx"))
893 if (len
+ namelen
+ 1 > sizeof(path
))
896 /* Don't reopen a pack we already have. */
897 strcpy(path
+ len
, de
->d_name
);
898 for (p
= packed_git
; p
; p
= p
->next
) {
899 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
904 /* See if it really is a valid .idx file with corresponding
905 * .pack file that we can map.
907 p
= add_packed_git(path
, len
+ namelen
, local
);
910 install_packed_git(p
);
915 static int sort_pack(const void *a_
, const void *b_
)
917 struct packed_git
*a
= *((struct packed_git
**)a_
);
918 struct packed_git
*b
= *((struct packed_git
**)b_
);
922 * Local packs tend to contain objects specific to our
923 * variant of the project than remote ones. In addition,
924 * remote ones could be on a network mounted filesystem.
925 * Favor local ones for these reasons.
927 st
= a
->pack_local
- b
->pack_local
;
932 * Younger packs tend to contain more recent objects,
933 * and more recent objects tend to get accessed more
936 if (a
->mtime
< b
->mtime
)
938 else if (a
->mtime
== b
->mtime
)
943 static void rearrange_packed_git(void)
945 struct packed_git
**ary
, *p
;
948 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
953 /* prepare an array of packed_git for easier sorting */
954 ary
= xcalloc(n
, sizeof(struct packed_git
*));
955 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
958 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
960 /* link them back again */
961 for (i
= 0; i
< n
- 1; i
++)
962 ary
[i
]->next
= ary
[i
+ 1];
963 ary
[n
- 1]->next
= NULL
;
969 static int prepare_packed_git_run_once
= 0;
970 void prepare_packed_git(void)
972 struct alternate_object_database
*alt
;
974 if (prepare_packed_git_run_once
)
976 prepare_packed_git_one(get_object_directory(), 1);
978 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
980 prepare_packed_git_one(alt
->base
, 0);
983 rearrange_packed_git();
984 prepare_packed_git_run_once
= 1;
987 void reprepare_packed_git(void)
990 prepare_packed_git_run_once
= 0;
991 prepare_packed_git();
994 static void mark_bad_packed_object(struct packed_git
*p
,
995 const unsigned char *sha1
)
998 for (i
= 0; i
< p
->num_bad_objects
; i
++)
999 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1001 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1002 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1003 p
->num_bad_objects
++;
1006 static int has_packed_and_bad(const unsigned char *sha1
)
1008 struct packed_git
*p
;
1011 for (p
= packed_git
; p
; p
= p
->next
)
1012 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1013 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1018 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1020 unsigned char real_sha1
[20];
1021 hash_sha1_file(map
, size
, type
, real_sha1
);
1022 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1025 static int git_open_noatime(const char *name
)
1027 static int sha1_file_open_flag
= O_NOATIME
;
1028 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1030 /* Might the failure be due to O_NOATIME? */
1031 if (fd
< 0 && errno
!= ENOENT
&& sha1_file_open_flag
) {
1032 fd
= open(name
, O_RDONLY
);
1034 sha1_file_open_flag
= 0;
1039 static int open_sha1_file(const unsigned char *sha1
)
1042 char *name
= sha1_file_name(sha1
);
1043 struct alternate_object_database
*alt
;
1045 fd
= git_open_noatime(name
);
1051 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1053 fill_sha1_path(name
, sha1
);
1054 fd
= git_open_noatime(alt
->base
);
1061 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1066 fd
= open_sha1_file(sha1
);
1071 if (!fstat(fd
, &st
)) {
1072 *size
= xsize_t(st
.st_size
);
1073 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1081 * There used to be a second loose object header format which
1082 * was meant to mimic the in-pack format, allowing for direct
1083 * copy of the object data. This format turned up not to be
1084 * really worth it and we no longer write loose objects in that
1087 static int experimental_loose_object(unsigned char *map
)
1092 * Is it a zlib-compressed buffer? If so, the first byte
1093 * must be 0x78 (15-bit window size, deflated), and the
1094 * first 16-bit word is evenly divisible by 31. If so,
1095 * we are looking at the official format, not the experimental
1098 word
= (map
[0] << 8) + map
[1];
1099 if (map
[0] == 0x78 && !(word
% 31))
1105 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1106 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1109 unsigned long size
, c
;
1110 unsigned long used
= 0;
1113 *type
= (c
>> 4) & 7;
1117 if (len
<= used
|| bitsizeof(long) <= shift
) {
1118 error("bad object header");
1122 size
+= (c
& 0x7f) << shift
;
1129 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1131 unsigned long size
, used
;
1132 static const char valid_loose_object_type
[8] = {
1134 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1135 0, /* "delta" and others are invalid in a loose object */
1137 enum object_type type
;
1139 /* Get the data stream */
1140 memset(stream
, 0, sizeof(*stream
));
1141 stream
->next_in
= map
;
1142 stream
->avail_in
= mapsize
;
1143 stream
->next_out
= buffer
;
1144 stream
->avail_out
= bufsiz
;
1146 if (experimental_loose_object(map
)) {
1148 * The old experimental format we no longer produce;
1149 * we can still read it.
1151 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1152 if (!used
|| !valid_loose_object_type
[type
])
1157 /* Set up the stream for the rest.. */
1158 stream
->next_in
= map
;
1159 stream
->avail_in
= mapsize
;
1160 git_inflate_init(stream
);
1162 /* And generate the fake traditional header */
1163 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1164 typename(type
), size
);
1167 git_inflate_init(stream
);
1168 return git_inflate(stream
, 0);
1171 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1173 int bytes
= strlen(buffer
) + 1;
1174 unsigned char *buf
= xmallocz(size
);
1178 n
= stream
->total_out
- bytes
;
1181 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1183 if (bytes
<= size
) {
1185 * The above condition must be (bytes <= size), not
1186 * (bytes < size). In other words, even though we
1187 * expect no more output and set avail_out to zer0,
1188 * the input zlib stream may have bytes that express
1189 * "this concludes the stream", and we *do* want to
1192 * Otherwise we would not be able to test that we
1193 * consumed all the input to reach the expected size;
1194 * we also want to check that zlib tells us that all
1195 * went well with status == Z_STREAM_END at the end.
1197 stream
->next_out
= buf
+ bytes
;
1198 stream
->avail_out
= size
- bytes
;
1199 while (status
== Z_OK
)
1200 status
= git_inflate(stream
, Z_FINISH
);
1202 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1203 git_inflate_end(stream
);
1208 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1209 else if (stream
->avail_in
)
1210 error("garbage at end of loose object '%s'",
1217 * We used to just use "sscanf()", but that's actually way
1218 * too permissive for what we want to check. So do an anal
1219 * object header parse by hand.
1221 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1228 * The type can be at most ten bytes (including the
1229 * terminating '\0' that we add), and is followed by
1238 if (i
>= sizeof(type
))
1244 * The length must follow immediately, and be in canonical
1245 * decimal format (ie "010" is not valid).
1247 size
= *hdr
++ - '0';
1252 unsigned long c
= *hdr
- '0';
1256 size
= size
* 10 + c
;
1262 * The length must be followed by a zero byte
1264 return *hdr
? -1 : type_from_string(type
);
1267 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1273 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1274 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1277 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1280 unsigned long get_size_from_delta(struct packed_git
*p
,
1281 struct pack_window
**w_curs
,
1284 const unsigned char *data
;
1285 unsigned char delta_head
[20], *in
;
1289 memset(&stream
, 0, sizeof(stream
));
1290 stream
.next_out
= delta_head
;
1291 stream
.avail_out
= sizeof(delta_head
);
1293 git_inflate_init(&stream
);
1295 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1296 stream
.next_in
= in
;
1297 st
= git_inflate(&stream
, Z_FINISH
);
1298 curpos
+= stream
.next_in
- in
;
1299 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1300 stream
.total_out
< sizeof(delta_head
));
1301 git_inflate_end(&stream
);
1302 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1303 error("delta data unpack-initial failed");
1307 /* Examine the initial part of the delta to figure out
1312 /* ignore base size */
1313 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1315 /* Read the result size */
1316 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1319 static off_t
get_delta_base(struct packed_git
*p
,
1320 struct pack_window
**w_curs
,
1322 enum object_type type
,
1323 off_t delta_obj_offset
)
1325 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1328 /* use_pack() assured us we have [base_info, base_info + 20)
1329 * as a range that we can look at without walking off the
1330 * end of the mapped window. Its actually the hash size
1331 * that is assured. An OFS_DELTA longer than the hash size
1332 * is stupid, as then a REF_DELTA would be smaller to store.
1334 if (type
== OBJ_OFS_DELTA
) {
1336 unsigned char c
= base_info
[used
++];
1337 base_offset
= c
& 127;
1340 if (!base_offset
|| MSB(base_offset
, 7))
1341 return 0; /* overflow */
1342 c
= base_info
[used
++];
1343 base_offset
= (base_offset
<< 7) + (c
& 127);
1345 base_offset
= delta_obj_offset
- base_offset
;
1346 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1347 return 0; /* out of bound */
1349 } else if (type
== OBJ_REF_DELTA
) {
1350 /* The base entry _must_ be in the same pack */
1351 base_offset
= find_pack_entry_one(base_info
, p
);
1354 die("I am totally screwed");
1358 /* forward declaration for a mutually recursive function */
1359 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1360 unsigned long *sizep
);
1362 static int packed_delta_info(struct packed_git
*p
,
1363 struct pack_window
**w_curs
,
1365 enum object_type type
,
1367 unsigned long *sizep
)
1371 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1374 type
= packed_object_info(p
, base_offset
, NULL
);
1375 if (type
<= OBJ_NONE
) {
1376 struct revindex_entry
*revidx
;
1377 const unsigned char *base_sha1
;
1378 revidx
= find_pack_revindex(p
, base_offset
);
1381 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1382 mark_bad_packed_object(p
, base_sha1
);
1383 type
= sha1_object_info(base_sha1
, NULL
);
1384 if (type
<= OBJ_NONE
)
1388 /* We choose to only get the type of the base object and
1389 * ignore potentially corrupt pack file that expects the delta
1390 * based on a base with a wrong size. This saves tons of
1394 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1402 static int unpack_object_header(struct packed_git
*p
,
1403 struct pack_window
**w_curs
,
1405 unsigned long *sizep
)
1407 unsigned char *base
;
1410 enum object_type type
;
1412 /* use_pack() assures us we have [base, base + 20) available
1413 * as a range that we can look at at. (Its actually the hash
1414 * size that is assured.) With our object header encoding
1415 * the maximum deflated object size is 2^137, which is just
1416 * insane, so we know won't exceed what we have been given.
1418 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1419 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1428 const char *packed_object_info_detail(struct packed_git
*p
,
1430 unsigned long *size
,
1431 unsigned long *store_size
,
1432 unsigned int *delta_chain_length
,
1433 unsigned char *base_sha1
)
1435 struct pack_window
*w_curs
= NULL
;
1437 unsigned long dummy
;
1438 unsigned char *next_sha1
;
1439 enum object_type type
;
1440 struct revindex_entry
*revidx
;
1442 *delta_chain_length
= 0;
1443 curpos
= obj_offset
;
1444 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1446 revidx
= find_pack_revindex(p
, obj_offset
);
1447 *store_size
= revidx
[1].offset
- obj_offset
;
1452 die("pack %s contains unknown object type %d",
1453 p
->pack_name
, type
);
1458 unuse_pack(&w_curs
);
1459 return typename(type
);
1461 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1463 die("pack %s contains bad delta base reference of type %s",
1464 p
->pack_name
, typename(type
));
1465 if (*delta_chain_length
== 0) {
1466 revidx
= find_pack_revindex(p
, obj_offset
);
1467 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1471 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1472 if (*delta_chain_length
== 0)
1473 hashcpy(base_sha1
, next_sha1
);
1474 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1477 (*delta_chain_length
)++;
1478 curpos
= obj_offset
;
1479 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1483 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1484 unsigned long *sizep
)
1486 struct pack_window
*w_curs
= NULL
;
1488 off_t curpos
= obj_offset
;
1489 enum object_type type
;
1491 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1496 type
= packed_delta_info(p
, &w_curs
, curpos
,
1497 type
, obj_offset
, sizep
);
1507 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1508 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1511 unuse_pack(&w_curs
);
1515 static void *unpack_compressed_entry(struct packed_git
*p
,
1516 struct pack_window
**w_curs
,
1522 unsigned char *buffer
, *in
;
1524 buffer
= xmallocz(size
);
1525 memset(&stream
, 0, sizeof(stream
));
1526 stream
.next_out
= buffer
;
1527 stream
.avail_out
= size
+ 1;
1529 git_inflate_init(&stream
);
1531 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1532 stream
.next_in
= in
;
1533 st
= git_inflate(&stream
, Z_FINISH
);
1534 if (!stream
.avail_out
)
1535 break; /* the payload is larger than it should be */
1536 curpos
+= stream
.next_in
- in
;
1537 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1538 git_inflate_end(&stream
);
1539 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1547 #define MAX_DELTA_CACHE (256)
1549 static size_t delta_base_cached
;
1551 static struct delta_base_cache_lru_list
{
1552 struct delta_base_cache_lru_list
*prev
;
1553 struct delta_base_cache_lru_list
*next
;
1554 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1556 static struct delta_base_cache_entry
{
1557 struct delta_base_cache_lru_list lru
;
1559 struct packed_git
*p
;
1562 enum object_type type
;
1563 } delta_base_cache
[MAX_DELTA_CACHE
];
1565 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1569 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1570 hash
+= (hash
>> 8) + (hash
>> 16);
1571 return hash
% MAX_DELTA_CACHE
;
1574 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1575 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1578 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1579 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1582 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1583 return unpack_entry(p
, base_offset
, type
, base_size
);
1587 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1588 ent
->lru
.prev
->next
= ent
->lru
.next
;
1589 delta_base_cached
-= ent
->size
;
1591 ret
= xmemdupz(ent
->data
, ent
->size
);
1594 *base_size
= ent
->size
;
1598 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1603 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1604 ent
->lru
.prev
->next
= ent
->lru
.next
;
1605 delta_base_cached
-= ent
->size
;
1609 void clear_delta_base_cache(void)
1612 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1613 release_delta_base_cache(&delta_base_cache
[p
]);
1616 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1617 void *base
, unsigned long base_size
, enum object_type type
)
1619 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1620 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1621 struct delta_base_cache_lru_list
*lru
;
1623 release_delta_base_cache(ent
);
1624 delta_base_cached
+= base_size
;
1626 for (lru
= delta_base_cache_lru
.next
;
1627 delta_base_cached
> delta_base_cache_limit
1628 && lru
!= &delta_base_cache_lru
;
1630 struct delta_base_cache_entry
*f
= (void *)lru
;
1631 if (f
->type
== OBJ_BLOB
)
1632 release_delta_base_cache(f
);
1634 for (lru
= delta_base_cache_lru
.next
;
1635 delta_base_cached
> delta_base_cache_limit
1636 && lru
!= &delta_base_cache_lru
;
1638 struct delta_base_cache_entry
*f
= (void *)lru
;
1639 release_delta_base_cache(f
);
1643 ent
->base_offset
= base_offset
;
1646 ent
->size
= base_size
;
1647 ent
->lru
.next
= &delta_base_cache_lru
;
1648 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1649 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1650 delta_base_cache_lru
.prev
= &ent
->lru
;
1653 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1654 unsigned long *size
);
1656 static void *unpack_delta_entry(struct packed_git
*p
,
1657 struct pack_window
**w_curs
,
1659 unsigned long delta_size
,
1661 enum object_type
*type
,
1662 unsigned long *sizep
)
1664 void *delta_data
, *result
, *base
;
1665 unsigned long base_size
;
1668 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1670 error("failed to validate delta base reference "
1671 "at offset %"PRIuMAX
" from %s",
1672 (uintmax_t)curpos
, p
->pack_name
);
1676 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1679 * We're probably in deep shit, but let's try to fetch
1680 * the required base anyway from another pack or loose.
1681 * This is costly but should happen only in the presence
1682 * of a corrupted pack, and is better than failing outright.
1684 struct revindex_entry
*revidx
;
1685 const unsigned char *base_sha1
;
1686 revidx
= find_pack_revindex(p
, base_offset
);
1689 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1690 error("failed to read delta base object %s"
1691 " at offset %"PRIuMAX
" from %s",
1692 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1694 mark_bad_packed_object(p
, base_sha1
);
1695 base
= read_object(base_sha1
, type
, &base_size
);
1700 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1702 error("failed to unpack compressed delta "
1703 "at offset %"PRIuMAX
" from %s",
1704 (uintmax_t)curpos
, p
->pack_name
);
1708 result
= patch_delta(base
, base_size
,
1709 delta_data
, delta_size
,
1712 die("failed to apply delta");
1714 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1718 int do_check_packed_object_crc
;
1720 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1721 enum object_type
*type
, unsigned long *sizep
)
1723 struct pack_window
*w_curs
= NULL
;
1724 off_t curpos
= obj_offset
;
1727 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1728 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1729 unsigned long len
= revidx
[1].offset
- obj_offset
;
1730 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1731 const unsigned char *sha1
=
1732 nth_packed_object_sha1(p
, revidx
->nr
);
1733 error("bad packed object CRC for %s",
1735 mark_bad_packed_object(p
, sha1
);
1736 unuse_pack(&w_curs
);
1741 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1745 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1746 obj_offset
, type
, sizep
);
1752 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1756 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1757 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1759 unuse_pack(&w_curs
);
1763 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1766 const unsigned char *index
= p
->index_data
;
1768 if (open_pack_index(p
))
1770 index
= p
->index_data
;
1772 if (n
>= p
->num_objects
)
1775 if (p
->index_version
== 1) {
1776 return index
+ 24 * n
+ 4;
1779 return index
+ 20 * n
;
1783 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1785 const unsigned char *index
= p
->index_data
;
1787 if (p
->index_version
== 1) {
1788 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1791 index
+= 8 + p
->num_objects
* (20 + 4);
1792 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1793 if (!(off
& 0x80000000))
1795 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1796 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1797 ntohl(*((uint32_t *)(index
+ 4)));
1801 off_t
find_pack_entry_one(const unsigned char *sha1
,
1802 struct packed_git
*p
)
1804 const uint32_t *level1_ofs
= p
->index_data
;
1805 const unsigned char *index
= p
->index_data
;
1806 unsigned hi
, lo
, stride
;
1807 static int use_lookup
= -1;
1808 static int debug_lookup
= -1;
1810 if (debug_lookup
< 0)
1811 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1814 if (open_pack_index(p
))
1816 level1_ofs
= p
->index_data
;
1817 index
= p
->index_data
;
1819 if (p
->index_version
> 1) {
1824 hi
= ntohl(level1_ofs
[*sha1
]);
1825 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1826 if (p
->index_version
> 1) {
1834 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1835 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1838 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1840 int pos
= sha1_entry_pos(index
, stride
, 0,
1841 lo
, hi
, p
->num_objects
, sha1
);
1844 return nth_packed_object_offset(p
, pos
);
1848 unsigned mi
= (lo
+ hi
) / 2;
1849 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1852 printf("lo %u hi %u rg %u mi %u\n",
1853 lo
, hi
, hi
- lo
, mi
);
1855 return nth_packed_object_offset(p
, mi
);
1864 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1866 static struct packed_git
*last_found
= (void *)1;
1867 struct packed_git
*p
;
1870 prepare_packed_git();
1873 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1876 if (p
->num_bad_objects
) {
1878 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1879 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1883 offset
= find_pack_entry_one(sha1
, p
);
1886 * We are about to tell the caller where they can
1887 * locate the requested object. We better make
1888 * sure the packfile is still here and can be
1889 * accessed before supplying that answer, as
1890 * it may have been deleted since the index
1893 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1894 error("packfile %s cannot be accessed", p
->pack_name
);
1899 hashcpy(e
->sha1
, sha1
);
1905 if (p
== last_found
)
1909 if (p
== last_found
)
1915 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1916 struct packed_git
*packs
)
1918 struct packed_git
*p
;
1920 for (p
= packs
; p
; p
= p
->next
) {
1921 if (find_pack_entry_one(sha1
, p
))
1928 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1931 unsigned long mapsize
, size
;
1936 map
= map_sha1_file(sha1
, &mapsize
);
1938 return error("unable to find %s", sha1_to_hex(sha1
));
1939 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1940 status
= error("unable to unpack %s header",
1942 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1943 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1946 git_inflate_end(&stream
);
1947 munmap(map
, mapsize
);
1951 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1953 struct pack_entry e
;
1956 if (!find_pack_entry(sha1
, &e
)) {
1957 /* Most likely it's a loose object. */
1958 status
= sha1_loose_object_info(sha1
, sizep
);
1962 /* Not a loose object; someone else may have just packed it. */
1963 reprepare_packed_git();
1964 if (!find_pack_entry(sha1
, &e
))
1968 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
1970 mark_bad_packed_object(e
.p
, sha1
);
1971 status
= sha1_object_info(sha1
, sizep
);
1977 static void *read_packed_sha1(const unsigned char *sha1
,
1978 enum object_type
*type
, unsigned long *size
)
1980 struct pack_entry e
;
1983 if (!find_pack_entry(sha1
, &e
))
1985 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1988 * We're probably in deep shit, but let's try to fetch
1989 * the required object anyway from another pack or loose.
1990 * This should happen only in the presence of a corrupted
1991 * pack, and is better than failing outright.
1993 error("failed to read object %s at offset %"PRIuMAX
" from %s",
1994 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
1995 mark_bad_packed_object(e
.p
, sha1
);
1996 data
= read_object(sha1
, type
, size
);
2002 * This is meant to hold a *small* number of objects that you would
2003 * want read_sha1_file() to be able to return, but yet you do not want
2004 * to write them into the object store (e.g. a browse-only
2007 static struct cached_object
{
2008 unsigned char sha1
[20];
2009 enum object_type type
;
2013 static int cached_object_nr
, cached_object_alloc
;
2015 static struct cached_object empty_tree
= {
2016 EMPTY_TREE_SHA1_BIN
,
2022 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
2025 struct cached_object
*co
= cached_objects
;
2027 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
2028 if (!hashcmp(co
->sha1
, sha1
))
2031 if (!hashcmp(sha1
, empty_tree
.sha1
))
2036 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2037 unsigned char *sha1
)
2039 struct cached_object
*co
;
2041 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2042 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2044 if (cached_object_alloc
<= cached_object_nr
) {
2045 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2046 cached_objects
= xrealloc(cached_objects
,
2047 sizeof(*cached_objects
) *
2048 cached_object_alloc
);
2050 co
= &cached_objects
[cached_object_nr
++];
2053 co
->buf
= xmalloc(len
);
2054 memcpy(co
->buf
, buf
, len
);
2055 hashcpy(co
->sha1
, sha1
);
2059 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2060 unsigned long *size
)
2062 unsigned long mapsize
;
2064 struct cached_object
*co
;
2066 co
= find_cached_object(sha1
);
2070 return xmemdupz(co
->buf
, co
->size
);
2073 buf
= read_packed_sha1(sha1
, type
, size
);
2076 map
= map_sha1_file(sha1
, &mapsize
);
2078 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2079 munmap(map
, mapsize
);
2082 reprepare_packed_git();
2083 return read_packed_sha1(sha1
, type
, size
);
2086 void *read_sha1_file_repl(const unsigned char *sha1
,
2087 enum object_type
*type
,
2088 unsigned long *size
,
2089 const unsigned char **replacement
)
2091 const unsigned char *repl
= lookup_replace_object(sha1
);
2092 void *data
= read_object(repl
, type
, size
);
2094 /* die if we replaced an object with one that does not exist */
2095 if (!data
&& repl
!= sha1
)
2096 die("replacement %s not found for %s",
2097 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2099 /* legacy behavior is to die on corrupted objects */
2100 if (!data
&& (has_loose_object(repl
) || has_packed_and_bad(repl
)))
2101 die("object %s is corrupted", sha1_to_hex(repl
));
2104 *replacement
= repl
;
2109 void *read_object_with_reference(const unsigned char *sha1
,
2110 const char *required_type_name
,
2111 unsigned long *size
,
2112 unsigned char *actual_sha1_return
)
2114 enum object_type type
, required_type
;
2116 unsigned long isize
;
2117 unsigned char actual_sha1
[20];
2119 required_type
= type_from_string(required_type_name
);
2120 hashcpy(actual_sha1
, sha1
);
2122 int ref_length
= -1;
2123 const char *ref_type
= NULL
;
2125 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2128 if (type
== required_type
) {
2130 if (actual_sha1_return
)
2131 hashcpy(actual_sha1_return
, actual_sha1
);
2134 /* Handle references */
2135 else if (type
== OBJ_COMMIT
)
2137 else if (type
== OBJ_TAG
)
2138 ref_type
= "object ";
2143 ref_length
= strlen(ref_type
);
2145 if (ref_length
+ 40 > isize
||
2146 memcmp(buffer
, ref_type
, ref_length
) ||
2147 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2152 /* Now we have the ID of the referred-to object in
2153 * actual_sha1. Check again. */
2157 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2158 const char *type
, unsigned char *sha1
,
2159 char *hdr
, int *hdrlen
)
2163 /* Generate the header */
2164 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2168 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2169 git_SHA1_Update(&c
, buf
, len
);
2170 git_SHA1_Final(sha1
, &c
);
2174 * Move the just written object into its final resting place.
2175 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2176 * "moving" is only a part of what it does, when no patch between
2177 * master to pu changes the call sites of this function.
2179 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2183 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2185 else if (link(tmpfile
, filename
))
2189 * Coda hack - coda doesn't like cross-directory links,
2190 * so we fall back to a rename, which will mean that it
2191 * won't be able to check collisions, but that's not a
2194 * The same holds for FAT formatted media.
2196 * When this succeeds, we just return. We have nothing
2199 if (ret
&& ret
!= EEXIST
) {
2201 if (!rename(tmpfile
, filename
))
2205 unlink_or_warn(tmpfile
);
2207 if (ret
!= EEXIST
) {
2208 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2210 /* FIXME!!! Collision check here ? */
2214 if (adjust_shared_perm(filename
))
2215 return error("unable to set permission to '%s'", filename
);
2219 static int write_buffer(int fd
, const void *buf
, size_t len
)
2221 if (write_in_full(fd
, buf
, len
) < 0)
2222 return error("file write error (%s)", strerror(errno
));
2226 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2227 unsigned char *sha1
)
2231 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2235 /* Finalize a file on disk, and close it. */
2236 static void close_sha1_file(int fd
)
2238 if (fsync_object_files
)
2239 fsync_or_die(fd
, "sha1 file");
2241 die_errno("error when closing sha1 file");
2244 /* Size of directory component, including the ending '/' */
2245 static inline int directory_size(const char *filename
)
2247 const char *s
= strrchr(filename
, '/');
2250 return s
- filename
+ 1;
2254 * This creates a temporary file in the same directory as the final
2257 * We want to avoid cross-directory filename renames, because those
2258 * can have problems on various filesystems (FAT, NFS, Coda).
2260 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2262 int fd
, dirlen
= directory_size(filename
);
2264 if (dirlen
+ 20 > bufsiz
) {
2265 errno
= ENAMETOOLONG
;
2268 memcpy(buffer
, filename
, dirlen
);
2269 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2270 fd
= git_mkstemp_mode(buffer
, 0444);
2271 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2272 /* Make sure the directory exists */
2273 memcpy(buffer
, filename
, dirlen
);
2274 buffer
[dirlen
-1] = 0;
2275 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2279 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2280 fd
= git_mkstemp_mode(buffer
, 0444);
2285 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2286 const void *buf
, unsigned long len
, time_t mtime
)
2289 unsigned char compressed
[4096];
2292 unsigned char parano_sha1
[20];
2294 static char tmpfile
[PATH_MAX
];
2296 filename
= sha1_file_name(sha1
);
2297 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2298 while (fd
< 0 && errno
== EMFILE
&& unuse_one_window(packed_git
, -1))
2299 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2301 if (errno
== EACCES
)
2302 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2304 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2308 memset(&stream
, 0, sizeof(stream
));
2309 deflateInit(&stream
, zlib_compression_level
);
2310 stream
.next_out
= compressed
;
2311 stream
.avail_out
= sizeof(compressed
);
2314 /* First header.. */
2315 stream
.next_in
= (unsigned char *)hdr
;
2316 stream
.avail_in
= hdrlen
;
2317 while (deflate(&stream
, 0) == Z_OK
)
2319 git_SHA1_Update(&c
, hdr
, hdrlen
);
2321 /* Then the data itself.. */
2322 stream
.next_in
= (void *)buf
;
2323 stream
.avail_in
= len
;
2325 unsigned char *in0
= stream
.next_in
;
2326 ret
= deflate(&stream
, Z_FINISH
);
2327 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2328 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2329 die("unable to write sha1 file");
2330 stream
.next_out
= compressed
;
2331 stream
.avail_out
= sizeof(compressed
);
2332 } while (ret
== Z_OK
);
2334 if (ret
!= Z_STREAM_END
)
2335 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2336 ret
= deflateEnd(&stream
);
2338 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2339 git_SHA1_Final(parano_sha1
, &c
);
2340 if (hashcmp(sha1
, parano_sha1
) != 0)
2341 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2343 close_sha1_file(fd
);
2348 utb
.modtime
= mtime
;
2349 if (utime(tmpfile
, &utb
) < 0)
2350 warning("failed utime() on %s: %s",
2351 tmpfile
, strerror(errno
));
2354 return move_temp_to_file(tmpfile
, filename
);
2357 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2359 unsigned char sha1
[20];
2363 /* Normally if we have it in the pack then we do not bother writing
2364 * it out into .git/objects/??/?{38} file.
2366 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2368 hashcpy(returnsha1
, sha1
);
2369 if (has_sha1_file(sha1
))
2371 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2374 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2378 enum object_type type
;
2383 if (has_loose_object(sha1
))
2385 buf
= read_packed_sha1(sha1
, &type
, &len
);
2387 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2388 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2389 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2395 int has_pack_index(const unsigned char *sha1
)
2398 if (stat(sha1_pack_index_name(sha1
), &st
))
2403 int has_sha1_pack(const unsigned char *sha1
)
2405 struct pack_entry e
;
2406 return find_pack_entry(sha1
, &e
);
2409 int has_sha1_file(const unsigned char *sha1
)
2411 struct pack_entry e
;
2413 if (find_pack_entry(sha1
, &e
))
2415 return has_loose_object(sha1
);
2418 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2419 int write_object
, enum object_type type
, const char *path
)
2421 int ret
, re_allocated
= 0;
2427 * Convert blobs to git internal format
2429 if ((type
== OBJ_BLOB
) && path
) {
2430 struct strbuf nbuf
= STRBUF_INIT
;
2431 if (convert_to_git(path
, buf
, size
, &nbuf
,
2432 write_object
? safe_crlf
: 0)) {
2433 buf
= strbuf_detach(&nbuf
, &size
);
2439 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2441 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2447 #define SMALL_FILE_SIZE (32*1024)
2449 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2450 enum object_type type
, const char *path
)
2453 size_t size
= xsize_t(st
->st_size
);
2455 if (!S_ISREG(st
->st_mode
)) {
2456 struct strbuf sbuf
= STRBUF_INIT
;
2457 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2458 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2462 strbuf_release(&sbuf
);
2464 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2465 } else if (size
<= SMALL_FILE_SIZE
) {
2466 char *buf
= xmalloc(size
);
2467 if (size
== read_in_full(fd
, buf
, size
))
2468 ret
= index_mem(sha1
, buf
, size
, write_object
, type
,
2471 ret
= error("short read %s", strerror(errno
));
2474 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2475 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2482 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2485 struct strbuf sb
= STRBUF_INIT
;
2487 switch (st
->st_mode
& S_IFMT
) {
2489 fd
= open(path
, O_RDONLY
);
2491 return error("open(\"%s\"): %s", path
,
2493 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2494 return error("%s: failed to insert into database",
2498 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2499 char *errstr
= strerror(errno
);
2500 return error("readlink(\"%s\"): %s", path
,
2504 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2505 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2506 return error("%s: failed to insert into database",
2508 strbuf_release(&sb
);
2511 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2513 return error("%s: unsupported file type", path
);
2518 int read_pack_header(int fd
, struct pack_header
*header
)
2520 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2521 /* "eof before pack header was fully read" */
2522 return PH_ERROR_EOF
;
2524 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2525 /* "protocol error (pack signature mismatch detected)" */
2526 return PH_ERROR_PACK_SIGNATURE
;
2527 if (!pack_version_ok(header
->hdr_version
))
2528 /* "protocol error (pack version unsupported)" */
2529 return PH_ERROR_PROTOCOL
;
2533 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2535 enum object_type type
= sha1_object_info(sha1
, NULL
);
2537 die("%s is not a valid object", sha1_to_hex(sha1
));
2539 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),