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 unsigned int sha1_file_open_flag
= O_NOATIME
;
40 const signed char hexval_table
[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
78 for (i
= 0; i
< 20; i
++) {
79 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
88 int safe_create_leading_directories(char *path
)
93 if (is_absolute_path(path
))
97 pos
= strchr(pos
, '/');
101 if (!stat(path
, &st
)) {
103 if (!S_ISDIR(st
.st_mode
)) {
108 else if (mkdir(path
, 0777)) {
112 else if (adjust_shared_perm(path
)) {
121 char * sha1_to_hex(const unsigned char *sha1
)
124 static char hexbuffer
[4][50];
125 static const char hex
[] = "0123456789abcdef";
126 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
129 for (i
= 0; i
< 20; i
++) {
130 unsigned int val
= *sha1
++;
131 *buf
++ = hex
[val
>> 4];
132 *buf
++ = hex
[val
& 0xf];
139 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
142 for (i
= 0; i
< 20; i
++) {
143 static char hex
[] = "0123456789abcdef";
144 unsigned int val
= sha1
[i
];
145 char *pos
= pathbuf
+ i
*2 + (i
> 0);
146 *pos
++ = hex
[val
>> 4];
147 *pos
= hex
[val
& 0xf];
152 * NOTE! This returns a statically allocated buffer, so you have to be
153 * careful about using it. Do an "xstrdup()" if you need to save the
156 * Also note that this returns the location for creating. Reading
157 * SHA1 file can happen from any alternate directory listed in the
158 * DB_ENVIRONMENT environment variable if it is not found in
159 * the primary object database.
161 char *sha1_file_name(const unsigned char *sha1
)
163 static char *name
, *base
;
166 const char *sha1_file_directory
= get_object_directory();
167 int len
= strlen(sha1_file_directory
);
168 base
= xmalloc(len
+ 60);
169 memcpy(base
, sha1_file_directory
, len
);
170 memset(base
+len
, 0, 60);
173 name
= base
+ len
+ 1;
175 fill_sha1_path(name
, sha1
);
179 static char *sha1_get_pack_name(const unsigned char *sha1
,
180 char **name
, char **base
, const char *which
)
182 static const char hex
[] = "0123456789abcdef";
187 const char *sha1_file_directory
= get_object_directory();
188 int len
= strlen(sha1_file_directory
);
189 *base
= xmalloc(len
+ 60);
190 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
191 sha1_file_directory
, which
);
192 *name
= *base
+ len
+ 11;
197 for (i
= 0; i
< 20; i
++) {
198 unsigned int val
= *sha1
++;
199 *buf
++ = hex
[val
>> 4];
200 *buf
++ = hex
[val
& 0xf];
206 char *sha1_pack_name(const unsigned char *sha1
)
208 static char *name
, *base
;
210 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
213 char *sha1_pack_index_name(const unsigned char *sha1
)
215 static char *name
, *base
;
217 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
220 struct alternate_object_database
*alt_odb_list
;
221 static struct alternate_object_database
**alt_odb_tail
;
223 static void read_info_alternates(const char * alternates
, int depth
);
226 * Prepare alternate object database registry.
228 * The variable alt_odb_list points at the list of struct
229 * alternate_object_database. The elements on this list come from
230 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
231 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
232 * whose contents is similar to that environment variable but can be
233 * LF separated. Its base points at a statically allocated buffer that
234 * contains "/the/directory/corresponding/to/.git/objects/...", while
235 * its name points just after the slash at the end of ".git/objects/"
236 * in the example above, and has enough space to hold 40-byte hex
237 * SHA1, an extra slash for the first level indirection, and the
240 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
243 const char *objdir
= get_object_directory();
244 struct alternate_object_database
*ent
;
245 struct alternate_object_database
*alt
;
246 /* 43 = 40-byte + 2 '/' + terminating NUL */
248 int entlen
= pfxlen
+ 43;
251 if (!is_absolute_path(entry
) && relative_base
) {
252 /* Relative alt-odb */
254 base_len
= strlen(relative_base
) + 1;
258 ent
= xmalloc(sizeof(*ent
) + entlen
);
260 if (!is_absolute_path(entry
) && relative_base
) {
261 memcpy(ent
->base
, relative_base
, base_len
- 1);
262 ent
->base
[base_len
- 1] = '/';
263 memcpy(ent
->base
+ base_len
, entry
, len
);
266 memcpy(ent
->base
, entry
, pfxlen
);
268 ent
->name
= ent
->base
+ pfxlen
+ 1;
269 ent
->base
[pfxlen
+ 3] = '/';
270 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
272 /* Detect cases where alternate disappeared */
273 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
274 error("object directory %s does not exist; "
275 "check .git/objects/info/alternates.",
281 /* Prevent the common mistake of listing the same
282 * thing twice, or object directory itself.
284 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
285 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
290 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
295 /* add the alternate entry */
297 alt_odb_tail
= &(ent
->next
);
300 /* recursively add alternates */
301 read_info_alternates(ent
->base
, depth
+ 1);
303 ent
->base
[pfxlen
] = '/';
308 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
309 const char *relative_base
, int depth
)
311 const char *cp
, *last
;
314 error("%s: ignoring alternate object stores, nesting too deep.",
322 if (cp
< ep
&& *cp
== '#') {
323 while (cp
< ep
&& *cp
!= sep
)
328 while (cp
< ep
&& *cp
!= sep
)
331 if (!is_absolute_path(last
) && depth
) {
332 error("%s: ignoring relative alternate object store %s",
333 relative_base
, last
);
335 link_alt_odb_entry(last
, cp
- last
,
336 relative_base
, depth
);
339 while (cp
< ep
&& *cp
== sep
)
345 static void read_info_alternates(const char * relative_base
, int depth
)
350 const char alt_file_name
[] = "info/alternates";
351 /* Given that relative_base is no longer than PATH_MAX,
352 ensure that "path" has enough space to append "/", the
353 file name, "info/alternates", and a trailing NUL. */
354 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
357 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
358 fd
= open(path
, O_RDONLY
);
361 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
365 mapsz
= xsize_t(st
.st_size
);
366 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
369 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
374 void add_to_alternates_file(const char *reference
)
376 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
377 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), 1);
378 char *alt
= mkpath("%s/objects\n", reference
);
379 write_or_die(fd
, alt
, strlen(alt
));
380 if (commit_lock_file(lock
))
381 die("could not close alternates file");
383 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
386 void prepare_alt_odb(void)
393 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
396 alt_odb_tail
= &alt_odb_list
;
397 link_alt_odb_entries(alt
, alt
+ strlen(alt
), ':', NULL
, 0);
399 read_info_alternates(get_object_directory(), 0);
402 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
404 char *name
= sha1_file_name(sha1
);
405 struct alternate_object_database
*alt
;
410 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
412 fill_sha1_path(name
, sha1
);
413 if (!stat(alt
->base
, st
))
419 static unsigned int pack_used_ctr
;
420 static unsigned int pack_mmap_calls
;
421 static unsigned int peak_pack_open_windows
;
422 static unsigned int pack_open_windows
;
423 static size_t peak_pack_mapped
;
424 static size_t pack_mapped
;
425 struct packed_git
*packed_git
;
427 void pack_report(void)
430 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
431 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
432 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
433 sz_fmt(getpagesize()),
434 sz_fmt(packed_git_window_size
),
435 sz_fmt(packed_git_limit
));
437 "pack_report: pack_used_ctr = %10u\n"
438 "pack_report: pack_mmap_calls = %10u\n"
439 "pack_report: pack_open_windows = %10u / %10u\n"
440 "pack_report: pack_mapped = "
441 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
444 pack_open_windows
, peak_pack_open_windows
,
445 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
448 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
451 struct pack_idx_header
*hdr
;
453 uint32_t version
, nr
, i
, *index
;
454 int fd
= open(path
, O_RDONLY
);
459 if (fstat(fd
, &st
)) {
463 idx_size
= xsize_t(st
.st_size
);
464 if (idx_size
< 4 * 256 + 20 + 20) {
466 return error("index file %s is too small", path
);
468 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
472 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
473 version
= ntohl(hdr
->idx_version
);
474 if (version
< 2 || version
> 2) {
475 munmap(idx_map
, idx_size
);
476 return error("index file %s is version %d"
477 " and is not supported by this binary"
478 " (try upgrading GIT to a newer version)",
487 index
+= 2; /* skip index header */
488 for (i
= 0; i
< 256; i
++) {
489 uint32_t n
= ntohl(index
[i
]);
491 munmap(idx_map
, idx_size
);
492 return error("non-monotonic index %s", path
);
500 * - 256 index entries 4 bytes each
501 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
502 * - 20-byte SHA1 of the packfile
503 * - 20-byte SHA1 file checksum
505 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
506 munmap(idx_map
, idx_size
);
507 return error("wrong index v1 file size in %s", path
);
509 } else if (version
== 2) {
512 * - 8 bytes of header
513 * - 256 index entries 4 bytes each
514 * - 20-byte sha1 entry * nr
515 * - 4-byte crc entry * nr
516 * - 4-byte offset entry * nr
517 * - 20-byte SHA1 of the packfile
518 * - 20-byte SHA1 file checksum
519 * And after the 4-byte offset table might be a
520 * variable sized table containing 8-byte entries
521 * for offsets larger than 2^31.
523 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
524 unsigned long max_size
= min_size
;
526 max_size
+= (nr
- 1)*8;
527 if (idx_size
< min_size
|| idx_size
> max_size
) {
528 munmap(idx_map
, idx_size
);
529 return error("wrong index v2 file size in %s", path
);
531 if (idx_size
!= min_size
&&
533 * make sure we can deal with large pack offsets.
534 * 31-bit signed offset won't be enough, neither
535 * 32-bit unsigned one will be.
537 (sizeof(off_t
) <= 4)) {
538 munmap(idx_map
, idx_size
);
539 return error("pack too large for current definition of off_t in %s", path
);
543 p
->index_version
= version
;
544 p
->index_data
= idx_map
;
545 p
->index_size
= idx_size
;
550 int open_pack_index(struct packed_git
*p
)
558 idx_name
= xstrdup(p
->pack_name
);
559 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
560 ret
= check_packed_git_idx(idx_name
, p
);
565 static void scan_windows(struct packed_git
*p
,
566 struct packed_git
**lru_p
,
567 struct pack_window
**lru_w
,
568 struct pack_window
**lru_l
)
570 struct pack_window
*w
, *w_l
;
572 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
574 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
584 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
586 struct packed_git
*p
, *lru_p
= NULL
;
587 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
590 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
591 for (p
= packed_git
; p
; p
= p
->next
)
592 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
594 munmap(lru_w
->base
, lru_w
->len
);
595 pack_mapped
-= lru_w
->len
;
597 lru_l
->next
= lru_w
->next
;
599 lru_p
->windows
= lru_w
->next
;
600 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
601 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 close_pack_windows(struct packed_git
*p
)
622 struct pack_window
*w
= p
->windows
;
625 die("pack '%s' still has open windows to it",
627 munmap(w
->base
, w
->len
);
628 pack_mapped
-= w
->len
;
630 p
->windows
= w
->next
;
635 void unuse_pack(struct pack_window
**w_cursor
)
637 struct pack_window
*w
= *w_cursor
;
645 * Do not call this directly as this leaks p->pack_fd on error return;
646 * call open_packed_git() instead.
648 static int open_packed_git_1(struct packed_git
*p
)
651 struct pack_header hdr
;
652 unsigned char sha1
[20];
653 unsigned char *idx_sha1
;
656 if (!p
->index_data
&& open_pack_index(p
))
657 return error("packfile %s index unavailable", p
->pack_name
);
659 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
660 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
663 /* If we created the struct before we had the pack we lack size. */
665 if (!S_ISREG(st
.st_mode
))
666 return error("packfile %s not a regular file", p
->pack_name
);
667 p
->pack_size
= st
.st_size
;
668 } else if (p
->pack_size
!= st
.st_size
)
669 return error("packfile %s size changed", p
->pack_name
);
671 /* We leave these file descriptors open with sliding mmap;
672 * there is no point keeping them open across exec(), though.
674 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
676 return error("cannot determine file descriptor flags");
677 fd_flag
|= FD_CLOEXEC
;
678 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
679 return error("cannot set FD_CLOEXEC");
681 /* Verify we recognize this pack file format. */
682 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
683 return error("file %s is far too short to be a packfile", p
->pack_name
);
684 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
685 return error("file %s is not a GIT packfile", p
->pack_name
);
686 if (!pack_version_ok(hdr
.hdr_version
))
687 return error("packfile %s is version %u and not supported"
688 " (try upgrading GIT to a newer version)",
689 p
->pack_name
, ntohl(hdr
.hdr_version
));
691 /* Verify the pack matches its index. */
692 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
693 return error("packfile %s claims to have %u objects"
694 " while index indicates %u objects",
695 p
->pack_name
, ntohl(hdr
.hdr_entries
),
697 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
698 return error("end of packfile %s is unavailable", p
->pack_name
);
699 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
700 return error("packfile %s signature is unavailable", p
->pack_name
);
701 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
702 if (hashcmp(sha1
, idx_sha1
))
703 return error("packfile %s does not match index", p
->pack_name
);
707 static int open_packed_git(struct packed_git
*p
)
709 if (!open_packed_git_1(p
))
711 if (p
->pack_fd
!= -1) {
718 static int in_window(struct pack_window
*win
, off_t offset
)
720 /* We must promise at least 20 bytes (one hash) after the
721 * offset is available from this window, otherwise the offset
722 * is not actually in this window and a different window (which
723 * has that one hash excess) must be used. This is to support
724 * the object header and delta base parsing routines below.
726 off_t win_off
= win
->offset
;
727 return win_off
<= offset
728 && (offset
+ 20) <= (win_off
+ win
->len
);
731 unsigned char* use_pack(struct packed_git
*p
,
732 struct pack_window
**w_cursor
,
736 struct pack_window
*win
= *w_cursor
;
738 if (p
->pack_fd
== -1 && open_packed_git(p
))
739 die("packfile %s cannot be accessed", p
->pack_name
);
741 /* Since packfiles end in a hash of their content and its
742 * pointless to ask for an offset into the middle of that
743 * hash, and the in_window function above wouldn't match
744 * don't allow an offset too close to the end of the file.
746 if (offset
> (p
->pack_size
- 20))
747 die("offset beyond end of packfile (truncated pack?)");
749 if (!win
|| !in_window(win
, offset
)) {
752 for (win
= p
->windows
; win
; win
= win
->next
) {
753 if (in_window(win
, offset
))
757 size_t window_align
= packed_git_window_size
/ 2;
759 win
= xcalloc(1, sizeof(*win
));
760 win
->offset
= (offset
/ window_align
) * window_align
;
761 len
= p
->pack_size
- win
->offset
;
762 if (len
> packed_git_window_size
)
763 len
= packed_git_window_size
;
764 win
->len
= (size_t)len
;
765 pack_mapped
+= win
->len
;
766 while (packed_git_limit
< pack_mapped
767 && unuse_one_window(p
, p
->pack_fd
))
769 win
->base
= xmmap(NULL
, win
->len
,
770 PROT_READ
, MAP_PRIVATE
,
771 p
->pack_fd
, win
->offset
);
772 if (win
->base
== MAP_FAILED
)
773 die("packfile %s cannot be mapped: %s",
778 if (pack_mapped
> peak_pack_mapped
)
779 peak_pack_mapped
= pack_mapped
;
780 if (pack_open_windows
> peak_pack_open_windows
)
781 peak_pack_open_windows
= pack_open_windows
;
782 win
->next
= p
->windows
;
786 if (win
!= *w_cursor
) {
787 win
->last_used
= pack_used_ctr
++;
791 offset
-= win
->offset
;
793 *left
= win
->len
- xsize_t(offset
);
794 return win
->base
+ offset
;
797 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
800 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
803 * Make sure a corresponding .pack file exists and that
804 * the index looks sane.
806 path_len
-= strlen(".idx");
809 memcpy(p
->pack_name
, path
, path_len
);
810 strcpy(p
->pack_name
+ path_len
, ".pack");
811 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
816 /* ok, it looks sane as far as we can check without
817 * actually mapping the pack file.
819 p
->index_version
= 0;
820 p
->index_data
= NULL
;
823 p
->pack_size
= st
.st_size
;
827 p
->pack_local
= local
;
828 p
->mtime
= st
.st_mtime
;
829 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
834 struct packed_git
*parse_pack_index(unsigned char *sha1
)
836 char *path
= sha1_pack_index_name(sha1
);
837 return parse_pack_index_file(sha1
, path
);
840 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
841 const char *idx_path
)
843 const char *path
= sha1_pack_name(sha1
);
844 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
846 if (check_packed_git_idx(idx_path
, p
)) {
851 strcpy(p
->pack_name
, path
);
856 hashcpy(p
->sha1
, sha1
);
860 void install_packed_git(struct packed_git
*pack
)
862 pack
->next
= packed_git
;
866 static void prepare_packed_git_one(char *objdir
, int local
)
868 /* Ensure that this buffer is large enough so that we can
869 append "/pack/" without clobbering the stack even if
870 strlen(objdir) were PATH_MAX. */
871 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
876 sprintf(path
, "%s/pack", objdir
);
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)
989 prepare_packed_git_run_once
= 0;
990 prepare_packed_git();
993 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
995 unsigned char real_sha1
[20];
996 hash_sha1_file(map
, size
, type
, real_sha1
);
997 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1000 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1005 char *filename
= find_sha1_file(sha1
, &st
);
1011 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
1013 /* See if it works without O_NOATIME */
1014 switch (sha1_file_open_flag
) {
1016 fd
= open(filename
, O_RDONLY
);
1024 /* If it failed once, it will probably fail again.
1025 * Stop using O_NOATIME
1027 sha1_file_open_flag
= 0;
1029 *size
= xsize_t(st
.st_size
);
1030 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1035 static int legacy_loose_object(unsigned char *map
)
1040 * Is it a zlib-compressed buffer? If so, the first byte
1041 * must be 0x78 (15-bit window size, deflated), and the
1042 * first 16-bit word is evenly divisible by 31
1044 word
= (map
[0] << 8) + map
[1];
1045 if (map
[0] == 0x78 && !(word
% 31))
1051 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1056 unsigned long used
= 0;
1059 *type
= (c
>> 4) & 7;
1065 if (sizeof(long) * 8 <= shift
)
1068 size
+= (c
& 0x7f) << shift
;
1075 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1077 unsigned long size
, used
;
1078 static const char valid_loose_object_type
[8] = {
1080 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1081 0, /* "delta" and others are invalid in a loose object */
1083 enum object_type type
;
1085 /* Get the data stream */
1086 memset(stream
, 0, sizeof(*stream
));
1087 stream
->next_in
= map
;
1088 stream
->avail_in
= mapsize
;
1089 stream
->next_out
= buffer
;
1090 stream
->avail_out
= bufsiz
;
1092 if (legacy_loose_object(map
)) {
1093 inflateInit(stream
);
1094 return inflate(stream
, 0);
1099 * There used to be a second loose object header format which
1100 * was meant to mimic the in-pack format, allowing for direct
1101 * copy of the object data. This format turned up not to be
1102 * really worth it and we don't write it any longer. But we
1103 * can still read it.
1105 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1106 if (!used
|| !valid_loose_object_type
[type
])
1111 /* Set up the stream for the rest.. */
1112 stream
->next_in
= map
;
1113 stream
->avail_in
= mapsize
;
1114 inflateInit(stream
);
1116 /* And generate the fake traditional header */
1117 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1118 typename(type
), size
);
1122 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1124 int bytes
= strlen(buffer
) + 1;
1125 unsigned char *buf
= xmalloc(1+size
);
1129 n
= stream
->total_out
- bytes
;
1132 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1134 if (bytes
<= size
) {
1136 * The above condition must be (bytes <= size), not
1137 * (bytes < size). In other words, even though we
1138 * expect no more output and set avail_out to zer0,
1139 * the input zlib stream may have bytes that express
1140 * "this concludes the stream", and we *do* want to
1143 * Otherwise we would not be able to test that we
1144 * consumed all the input to reach the expected size;
1145 * we also want to check that zlib tells us that all
1146 * went well with status == Z_STREAM_END at the end.
1148 stream
->next_out
= buf
+ bytes
;
1149 stream
->avail_out
= size
- bytes
;
1150 while (status
== Z_OK
)
1151 status
= inflate(stream
, Z_FINISH
);
1154 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1160 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1161 else if (stream
->avail_in
)
1162 error("garbage at end of loose object '%s'",
1169 * We used to just use "sscanf()", but that's actually way
1170 * too permissive for what we want to check. So do an anal
1171 * object header parse by hand.
1173 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1180 * The type can be at most ten bytes (including the
1181 * terminating '\0' that we add), and is followed by
1190 if (i
>= sizeof(type
))
1196 * The length must follow immediately, and be in canonical
1197 * decimal format (ie "010" is not valid).
1199 size
= *hdr
++ - '0';
1204 unsigned long c
= *hdr
- '0';
1208 size
= size
* 10 + c
;
1214 * The length must be followed by a zero byte
1216 return *hdr
? -1 : type_from_string(type
);
1219 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1225 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1226 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1229 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1232 unsigned long get_size_from_delta(struct packed_git
*p
,
1233 struct pack_window
**w_curs
,
1236 const unsigned char *data
;
1237 unsigned char delta_head
[20], *in
;
1241 memset(&stream
, 0, sizeof(stream
));
1242 stream
.next_out
= delta_head
;
1243 stream
.avail_out
= sizeof(delta_head
);
1245 inflateInit(&stream
);
1247 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1248 stream
.next_in
= in
;
1249 st
= inflate(&stream
, Z_FINISH
);
1250 curpos
+= stream
.next_in
- in
;
1251 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1252 stream
.total_out
< sizeof(delta_head
));
1253 inflateEnd(&stream
);
1254 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1255 die("delta data unpack-initial failed");
1257 /* Examine the initial part of the delta to figure out
1262 /* ignore base size */
1263 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1265 /* Read the result size */
1266 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1269 static off_t
get_delta_base(struct packed_git
*p
,
1270 struct pack_window
**w_curs
,
1272 enum object_type type
,
1273 off_t delta_obj_offset
)
1275 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1278 /* use_pack() assured us we have [base_info, base_info + 20)
1279 * as a range that we can look at without walking off the
1280 * end of the mapped window. Its actually the hash size
1281 * that is assured. An OFS_DELTA longer than the hash size
1282 * is stupid, as then a REF_DELTA would be smaller to store.
1284 if (type
== OBJ_OFS_DELTA
) {
1286 unsigned char c
= base_info
[used
++];
1287 base_offset
= c
& 127;
1290 if (!base_offset
|| MSB(base_offset
, 7))
1291 die("offset value overflow for delta base object");
1292 c
= base_info
[used
++];
1293 base_offset
= (base_offset
<< 7) + (c
& 127);
1295 base_offset
= delta_obj_offset
- base_offset
;
1296 if (base_offset
>= delta_obj_offset
)
1297 die("delta base offset out of bound");
1299 } else if (type
== OBJ_REF_DELTA
) {
1300 /* The base entry _must_ be in the same pack */
1301 base_offset
= find_pack_entry_one(base_info
, p
);
1303 die("failed to find delta-pack base object %s",
1304 sha1_to_hex(base_info
));
1307 die("I am totally screwed");
1311 /* forward declaration for a mutually recursive function */
1312 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1313 unsigned long *sizep
);
1315 static int packed_delta_info(struct packed_git
*p
,
1316 struct pack_window
**w_curs
,
1318 enum object_type type
,
1320 unsigned long *sizep
)
1324 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1325 type
= packed_object_info(p
, base_offset
, NULL
);
1327 /* We choose to only get the type of the base object and
1328 * ignore potentially corrupt pack file that expects the delta
1329 * based on a base with a wrong size. This saves tons of
1333 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1338 static int unpack_object_header(struct packed_git
*p
,
1339 struct pack_window
**w_curs
,
1341 unsigned long *sizep
)
1343 unsigned char *base
;
1346 enum object_type type
;
1348 /* use_pack() assures us we have [base, base + 20) available
1349 * as a range that we can look at at. (Its actually the hash
1350 * size that is assured.) With our object header encoding
1351 * the maximum deflated object size is 2^137, which is just
1352 * insane, so we know won't exceed what we have been given.
1354 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1355 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1357 die("object offset outside of pack file");
1363 const char *packed_object_info_detail(struct packed_git
*p
,
1365 unsigned long *size
,
1366 unsigned long *store_size
,
1367 unsigned int *delta_chain_length
,
1368 unsigned char *base_sha1
)
1370 struct pack_window
*w_curs
= NULL
;
1372 unsigned long dummy
;
1373 unsigned char *next_sha1
;
1374 enum object_type type
;
1375 struct revindex_entry
*revidx
;
1377 *delta_chain_length
= 0;
1378 curpos
= obj_offset
;
1379 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1381 revidx
= find_pack_revindex(p
, obj_offset
);
1382 *store_size
= revidx
[1].offset
- obj_offset
;
1387 die("pack %s contains unknown object type %d",
1388 p
->pack_name
, type
);
1393 unuse_pack(&w_curs
);
1394 return typename(type
);
1396 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1397 if (*delta_chain_length
== 0) {
1398 revidx
= find_pack_revindex(p
, obj_offset
);
1399 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1403 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1404 if (*delta_chain_length
== 0)
1405 hashcpy(base_sha1
, next_sha1
);
1406 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1409 (*delta_chain_length
)++;
1410 curpos
= obj_offset
;
1411 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1415 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1416 unsigned long *sizep
)
1418 struct pack_window
*w_curs
= NULL
;
1420 off_t curpos
= obj_offset
;
1421 enum object_type type
;
1423 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1428 type
= packed_delta_info(p
, &w_curs
, curpos
,
1429 type
, obj_offset
, sizep
);
1439 die("pack %s contains unknown object type %d",
1440 p
->pack_name
, type
);
1442 unuse_pack(&w_curs
);
1446 static void *unpack_compressed_entry(struct packed_git
*p
,
1447 struct pack_window
**w_curs
,
1453 unsigned char *buffer
, *in
;
1455 buffer
= xmalloc(size
+ 1);
1457 memset(&stream
, 0, sizeof(stream
));
1458 stream
.next_out
= buffer
;
1459 stream
.avail_out
= size
;
1461 inflateInit(&stream
);
1463 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1464 stream
.next_in
= in
;
1465 st
= inflate(&stream
, Z_FINISH
);
1466 curpos
+= stream
.next_in
- in
;
1467 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1468 inflateEnd(&stream
);
1469 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1477 #define MAX_DELTA_CACHE (256)
1479 static size_t delta_base_cached
;
1481 static struct delta_base_cache_lru_list
{
1482 struct delta_base_cache_lru_list
*prev
;
1483 struct delta_base_cache_lru_list
*next
;
1484 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1486 static struct delta_base_cache_entry
{
1487 struct delta_base_cache_lru_list lru
;
1489 struct packed_git
*p
;
1492 enum object_type type
;
1493 } delta_base_cache
[MAX_DELTA_CACHE
];
1495 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1499 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1500 hash
+= (hash
>> 8) + (hash
>> 16);
1501 return hash
% MAX_DELTA_CACHE
;
1504 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1505 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1508 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1509 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1512 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1513 goto found_cache_entry
;
1514 return unpack_entry(p
, base_offset
, type
, base_size
);
1519 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1520 ent
->lru
.prev
->next
= ent
->lru
.next
;
1521 delta_base_cached
-= ent
->size
;
1523 ret
= xmemdupz(ent
->data
, ent
->size
);
1526 *base_size
= ent
->size
;
1530 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1535 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1536 ent
->lru
.prev
->next
= ent
->lru
.next
;
1537 delta_base_cached
-= ent
->size
;
1541 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1542 void *base
, unsigned long base_size
, enum object_type type
)
1544 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1545 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1546 struct delta_base_cache_lru_list
*lru
;
1548 release_delta_base_cache(ent
);
1549 delta_base_cached
+= base_size
;
1551 for (lru
= delta_base_cache_lru
.next
;
1552 delta_base_cached
> delta_base_cache_limit
1553 && lru
!= &delta_base_cache_lru
;
1555 struct delta_base_cache_entry
*f
= (void *)lru
;
1556 if (f
->type
== OBJ_BLOB
)
1557 release_delta_base_cache(f
);
1559 for (lru
= delta_base_cache_lru
.next
;
1560 delta_base_cached
> delta_base_cache_limit
1561 && lru
!= &delta_base_cache_lru
;
1563 struct delta_base_cache_entry
*f
= (void *)lru
;
1564 release_delta_base_cache(f
);
1568 ent
->base_offset
= base_offset
;
1571 ent
->size
= base_size
;
1572 ent
->lru
.next
= &delta_base_cache_lru
;
1573 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1574 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1575 delta_base_cache_lru
.prev
= &ent
->lru
;
1578 static void *unpack_delta_entry(struct packed_git
*p
,
1579 struct pack_window
**w_curs
,
1581 unsigned long delta_size
,
1583 enum object_type
*type
,
1584 unsigned long *sizep
)
1586 void *delta_data
, *result
, *base
;
1587 unsigned long base_size
;
1590 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1591 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1593 die("failed to read delta base object"
1594 " at %"PRIuMAX
" from %s",
1595 (uintmax_t)base_offset
, p
->pack_name
);
1597 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1599 die("failed to unpack compressed delta"
1600 " at %"PRIuMAX
" from %s",
1601 (uintmax_t)curpos
, p
->pack_name
);
1602 result
= patch_delta(base
, base_size
,
1603 delta_data
, delta_size
,
1606 die("failed to apply delta");
1608 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1612 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1613 enum object_type
*type
, unsigned long *sizep
)
1615 struct pack_window
*w_curs
= NULL
;
1616 off_t curpos
= obj_offset
;
1619 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1623 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1624 obj_offset
, type
, sizep
);
1630 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1633 die("unknown object type %i in %s", *type
, p
->pack_name
);
1635 unuse_pack(&w_curs
);
1639 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1642 const unsigned char *index
= p
->index_data
;
1644 if (open_pack_index(p
))
1646 index
= p
->index_data
;
1648 if (n
>= p
->num_objects
)
1651 if (p
->index_version
== 1) {
1652 return index
+ 24 * n
+ 4;
1655 return index
+ 20 * n
;
1659 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1661 const unsigned char *index
= p
->index_data
;
1663 if (p
->index_version
== 1) {
1664 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1667 index
+= 8 + p
->num_objects
* (20 + 4);
1668 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1669 if (!(off
& 0x80000000))
1671 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1672 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1673 ntohl(*((uint32_t *)(index
+ 4)));
1677 off_t
find_pack_entry_one(const unsigned char *sha1
,
1678 struct packed_git
*p
)
1680 const uint32_t *level1_ofs
= p
->index_data
;
1681 const unsigned char *index
= p
->index_data
;
1682 unsigned hi
, lo
, stride
;
1683 static int use_lookup
= -1;
1684 static int debug_lookup
= -1;
1686 if (debug_lookup
< 0)
1687 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1690 if (open_pack_index(p
))
1692 level1_ofs
= p
->index_data
;
1693 index
= p
->index_data
;
1695 if (p
->index_version
> 1) {
1700 hi
= ntohl(level1_ofs
[*sha1
]);
1701 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1702 if (p
->index_version
> 1) {
1710 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1711 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1714 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1716 int pos
= sha1_entry_pos(index
, stride
, 0,
1717 lo
, hi
, p
->num_objects
, sha1
);
1720 return nth_packed_object_offset(p
, pos
);
1724 unsigned mi
= (lo
+ hi
) / 2;
1725 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1728 printf("lo %u hi %u rg %u mi %u\n",
1729 lo
, hi
, hi
- lo
, mi
);
1731 return nth_packed_object_offset(p
, mi
);
1740 int matches_pack_name(struct packed_git
*p
, const char *name
)
1742 const char *last_c
, *c
;
1744 if (!strcmp(p
->pack_name
, name
))
1747 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1752 if (!strcmp(last_c
, name
))
1758 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1760 static struct packed_git
*last_found
= (void *)1;
1761 struct packed_git
*p
;
1764 prepare_packed_git();
1767 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1770 if (ignore_packed
) {
1772 for (ig
= ignore_packed
; *ig
; ig
++)
1773 if (matches_pack_name(p
, *ig
))
1779 offset
= find_pack_entry_one(sha1
, p
);
1782 * We are about to tell the caller where they can
1783 * locate the requested object. We better make
1784 * sure the packfile is still here and can be
1785 * accessed before supplying that answer, as
1786 * it may have been deleted since the index
1789 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1790 error("packfile %s cannot be accessed", p
->pack_name
);
1795 hashcpy(e
->sha1
, sha1
);
1801 if (p
== last_found
)
1805 if (p
== last_found
)
1811 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1812 struct packed_git
*packs
)
1814 struct packed_git
*p
;
1816 for (p
= packs
; p
; p
= p
->next
) {
1817 if (find_pack_entry_one(sha1
, p
))
1824 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1827 unsigned long mapsize
, size
;
1832 map
= map_sha1_file(sha1
, &mapsize
);
1834 return error("unable to find %s", sha1_to_hex(sha1
));
1835 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1836 status
= error("unable to unpack %s header",
1838 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1839 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1842 inflateEnd(&stream
);
1843 munmap(map
, mapsize
);
1847 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1849 struct pack_entry e
;
1851 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1852 reprepare_packed_git();
1853 if (!find_pack_entry(sha1
, &e
, NULL
))
1854 return sha1_loose_object_info(sha1
, sizep
);
1856 return packed_object_info(e
.p
, e
.offset
, sizep
);
1859 static void *read_packed_sha1(const unsigned char *sha1
,
1860 enum object_type
*type
, unsigned long *size
)
1862 struct pack_entry e
;
1864 if (!find_pack_entry(sha1
, &e
, NULL
))
1867 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1871 * This is meant to hold a *small* number of objects that you would
1872 * want read_sha1_file() to be able to return, but yet you do not want
1873 * to write them into the object store (e.g. a browse-only
1876 static struct cached_object
{
1877 unsigned char sha1
[20];
1878 enum object_type type
;
1882 static int cached_object_nr
, cached_object_alloc
;
1884 static struct cached_object empty_tree
= {
1885 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1886 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1887 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1893 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1896 struct cached_object
*co
= cached_objects
;
1898 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1899 if (!hashcmp(co
->sha1
, sha1
))
1902 if (!hashcmp(sha1
, empty_tree
.sha1
))
1907 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1908 unsigned char *sha1
)
1910 struct cached_object
*co
;
1912 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1913 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1915 if (cached_object_alloc
<= cached_object_nr
) {
1916 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1917 cached_objects
= xrealloc(cached_objects
,
1918 sizeof(*cached_objects
) *
1919 cached_object_alloc
);
1921 co
= &cached_objects
[cached_object_nr
++];
1924 co
->buf
= xmalloc(len
);
1925 memcpy(co
->buf
, buf
, len
);
1926 hashcpy(co
->sha1
, sha1
);
1930 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1931 unsigned long *size
)
1933 unsigned long mapsize
;
1935 struct cached_object
*co
;
1937 co
= find_cached_object(sha1
);
1941 return xmemdupz(co
->buf
, co
->size
);
1944 buf
= read_packed_sha1(sha1
, type
, size
);
1947 map
= map_sha1_file(sha1
, &mapsize
);
1949 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1950 munmap(map
, mapsize
);
1953 reprepare_packed_git();
1954 return read_packed_sha1(sha1
, type
, size
);
1957 void *read_object_with_reference(const unsigned char *sha1
,
1958 const char *required_type_name
,
1959 unsigned long *size
,
1960 unsigned char *actual_sha1_return
)
1962 enum object_type type
, required_type
;
1964 unsigned long isize
;
1965 unsigned char actual_sha1
[20];
1967 required_type
= type_from_string(required_type_name
);
1968 hashcpy(actual_sha1
, sha1
);
1970 int ref_length
= -1;
1971 const char *ref_type
= NULL
;
1973 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1976 if (type
== required_type
) {
1978 if (actual_sha1_return
)
1979 hashcpy(actual_sha1_return
, actual_sha1
);
1982 /* Handle references */
1983 else if (type
== OBJ_COMMIT
)
1985 else if (type
== OBJ_TAG
)
1986 ref_type
= "object ";
1991 ref_length
= strlen(ref_type
);
1993 if (ref_length
+ 40 > isize
||
1994 memcmp(buffer
, ref_type
, ref_length
) ||
1995 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2000 /* Now we have the ID of the referred-to object in
2001 * actual_sha1. Check again. */
2005 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2006 const char *type
, unsigned char *sha1
,
2007 char *hdr
, int *hdrlen
)
2011 /* Generate the header */
2012 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2016 SHA1_Update(&c
, hdr
, *hdrlen
);
2017 SHA1_Update(&c
, buf
, len
);
2018 SHA1_Final(sha1
, &c
);
2022 * Link the tempfile to the final place, possibly creating the
2023 * last directory level as you do so.
2025 * Returns the errno on failure, 0 on success.
2027 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
2032 if (!link(tmpfile
, filename
))
2036 * Try to mkdir the last path component if that failed.
2038 * Re-try the "link()" regardless of whether the mkdir
2039 * succeeds, since a race might mean that somebody
2043 dir
= strrchr(filename
, '/');
2046 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
2051 if (!link(tmpfile
, filename
))
2059 * Move the just written object into its final resting place
2061 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2063 int ret
= link_temp_to_file(tmpfile
, filename
);
2066 * Coda hack - coda doesn't like cross-directory links,
2067 * so we fall back to a rename, which will mean that it
2068 * won't be able to check collisions, but that's not a
2071 * The same holds for FAT formatted media.
2073 * When this succeeds, we just return 0. We have nothing
2076 if (ret
&& ret
!= EEXIST
) {
2077 if (!rename(tmpfile
, filename
))
2083 if (ret
!= EEXIST
) {
2084 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2086 /* FIXME!!! Collision check here ? */
2092 static int write_buffer(int fd
, const void *buf
, size_t len
)
2094 if (write_in_full(fd
, buf
, len
) < 0)
2095 return error("file write error (%s)", strerror(errno
));
2099 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2100 unsigned char *sha1
)
2104 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2108 /* Finalize a file on disk, and close it. */
2109 static void close_sha1_file(int fd
)
2111 /* For safe-mode, we could fsync_or_die(fd, "sha1 file") here */
2114 die("unable to write sha1 file");
2117 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2118 void *buf
, unsigned long len
, time_t mtime
)
2121 unsigned char *compressed
;
2124 static char tmpfile
[PATH_MAX
];
2126 filename
= sha1_file_name(sha1
);
2127 fd
= open(filename
, O_RDONLY
);
2130 * FIXME!!! We might do collision checking here, but we'd
2131 * need to uncompress the old file and check it. Later.
2137 if (errno
!= ENOENT
) {
2138 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2141 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2143 fd
= mkstemp(tmpfile
);
2146 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2148 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2152 memset(&stream
, 0, sizeof(stream
));
2153 deflateInit(&stream
, zlib_compression_level
);
2154 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2155 compressed
= xmalloc(size
);
2158 stream
.next_out
= compressed
;
2159 stream
.avail_out
= size
;
2161 /* First header.. */
2162 stream
.next_in
= (unsigned char *)hdr
;
2163 stream
.avail_in
= hdrlen
;
2164 while (deflate(&stream
, 0) == Z_OK
)
2167 /* Then the data itself.. */
2168 stream
.next_in
= buf
;
2169 stream
.avail_in
= len
;
2170 ret
= deflate(&stream
, Z_FINISH
);
2171 if (ret
!= Z_STREAM_END
)
2172 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2174 ret
= deflateEnd(&stream
);
2176 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2178 size
= stream
.total_out
;
2180 if (write_buffer(fd
, compressed
, size
) < 0)
2181 die("unable to write sha1 file");
2182 close_sha1_file(fd
);
2188 utb
.modtime
= mtime
;
2189 if (utime(tmpfile
, &utb
) < 0)
2190 warning("failed utime() on %s: %s",
2191 tmpfile
, strerror(errno
));
2194 return move_temp_to_file(tmpfile
, filename
);
2197 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2199 unsigned char sha1
[20];
2203 /* Normally if we have it in the pack then we do not bother writing
2204 * it out into .git/objects/??/?{38} file.
2206 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2208 hashcpy(returnsha1
, sha1
);
2209 if (has_sha1_file(sha1
))
2211 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2214 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2219 enum object_type type
;
2223 if (find_sha1_file(sha1
, &st
))
2225 buf
= read_packed_sha1(sha1
, &type
, &len
);
2227 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2228 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2229 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2233 * We need to unpack and recompress the object for writing
2234 * it out to a different file.
2236 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2240 unsigned char *unpacked
;
2242 enum object_type type
;
2247 /* need to unpack and recompress it by itself */
2248 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2250 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2252 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2255 memset(&stream
, 0, sizeof(stream
));
2256 deflateInit(&stream
, zlib_compression_level
);
2257 size
= deflateBound(&stream
, len
+ hdrlen
);
2258 buf
= xmalloc(size
);
2261 stream
.next_out
= buf
;
2262 stream
.avail_out
= size
;
2264 /* First header.. */
2265 stream
.next_in
= (void *)hdr
;
2266 stream
.avail_in
= hdrlen
;
2267 while (deflate(&stream
, 0) == Z_OK
)
2270 /* Then the data itself.. */
2271 stream
.next_in
= unpacked
;
2272 stream
.avail_in
= len
;
2273 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2275 deflateEnd(&stream
);
2278 *objsize
= stream
.total_out
;
2282 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2285 unsigned long objsize
;
2286 void *buf
= map_sha1_file(sha1
, &objsize
);
2289 retval
= write_buffer(fd
, buf
, objsize
);
2290 munmap(buf
, objsize
);
2294 buf
= repack_object(sha1
, &objsize
);
2295 retval
= write_buffer(fd
, buf
, objsize
);
2300 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2301 size_t bufsize
, size_t *bufposn
)
2303 char tmpfile
[PATH_MAX
];
2306 unsigned char real_sha1
[20];
2307 unsigned char discard
[4096];
2311 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2313 local
= mkstemp(tmpfile
);
2316 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2318 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2321 memset(&stream
, 0, sizeof(stream
));
2323 inflateInit(&stream
);
2330 stream
.avail_in
= *bufposn
;
2331 stream
.next_in
= (unsigned char *) buffer
;
2333 stream
.next_out
= discard
;
2334 stream
.avail_out
= sizeof(discard
);
2335 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2336 SHA1_Update(&c
, discard
, sizeof(discard
) -
2338 } while (stream
.avail_in
&& ret
== Z_OK
);
2339 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2340 die("unable to write sha1 file");
2341 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2343 *bufposn
= stream
.avail_in
;
2347 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2352 return error("Connection closed?");
2353 perror("Reading from connection");
2358 inflateEnd(&stream
);
2360 close_sha1_file(local
);
2361 SHA1_Final(real_sha1
, &c
);
2362 if (ret
!= Z_STREAM_END
) {
2364 return error("File %s corrupted", sha1_to_hex(sha1
));
2366 if (hashcmp(sha1
, real_sha1
)) {
2368 return error("File %s has bad hash", sha1_to_hex(sha1
));
2371 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2374 int has_pack_index(const unsigned char *sha1
)
2377 if (stat(sha1_pack_index_name(sha1
), &st
))
2382 int has_pack_file(const unsigned char *sha1
)
2385 if (stat(sha1_pack_name(sha1
), &st
))
2390 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2392 struct pack_entry e
;
2393 return find_pack_entry(sha1
, &e
, ignore_packed
);
2396 int has_sha1_file(const unsigned char *sha1
)
2399 struct pack_entry e
;
2401 if (find_pack_entry(sha1
, &e
, NULL
))
2403 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2406 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2411 strbuf_init(&buf
, 0);
2412 if (strbuf_read(&buf
, fd
, 4096) < 0) {
2413 strbuf_release(&buf
);
2420 ret
= write_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2422 ret
= hash_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2423 strbuf_release(&buf
);
2428 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2429 enum object_type type
, const char *path
)
2431 size_t size
= xsize_t(st
->st_size
);
2433 int ret
, re_allocated
= 0;
2436 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2443 * Convert blobs to git internal format
2445 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2447 strbuf_init(&nbuf
, 0);
2448 if (convert_to_git(path
, buf
, size
, &nbuf
,
2449 write_object
? safe_crlf
: 0)) {
2451 buf
= strbuf_detach(&nbuf
, &size
);
2457 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2459 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2469 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2475 switch (st
->st_mode
& S_IFMT
) {
2477 fd
= open(path
, O_RDONLY
);
2479 return error("open(\"%s\"): %s", path
,
2481 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2482 return error("%s: failed to insert into database",
2486 len
= xsize_t(st
->st_size
);
2487 target
= xmalloc(len
+ 1);
2488 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2489 char *errstr
= strerror(errno
);
2491 return error("readlink(\"%s\"): %s", path
,
2495 hash_sha1_file(target
, len
, blob_type
, sha1
);
2496 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2497 return error("%s: failed to insert into database",
2502 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2504 return error("%s: unsupported file type", path
);
2509 int read_pack_header(int fd
, struct pack_header
*header
)
2511 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2512 /* "eof before pack header was fully read" */
2513 return PH_ERROR_EOF
;
2515 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2516 /* "protocol error (pack signature mismatch detected)" */
2517 return PH_ERROR_PACK_SIGNATURE
;
2518 if (!pack_version_ok(header
->hdr_version
))
2519 /* "protocol error (pack version unsupported)" */
2520 return PH_ERROR_PROTOCOL
;