2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
18 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19 #define O_NOATIME 01000000
31 const unsigned char null_sha1
[20];
33 static unsigned int sha1_file_open_flag
= O_NOATIME
;
35 signed char hexval_table
[256] = {
36 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
42 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
43 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
44 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
48 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
70 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
73 for (i
= 0; i
< 20; i
++) {
74 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
83 int safe_create_leading_directories(char *path
)
92 pos
= strchr(pos
, '/');
96 if (!stat(path
, &st
)) {
98 if (!S_ISDIR(st
.st_mode
)) {
103 else if (mkdir(path
, 0777)) {
107 else if (adjust_shared_perm(path
)) {
116 char * sha1_to_hex(const unsigned char *sha1
)
119 static char hexbuffer
[4][50];
120 static const char hex
[] = "0123456789abcdef";
121 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
124 for (i
= 0; i
< 20; i
++) {
125 unsigned int val
= *sha1
++;
126 *buf
++ = hex
[val
>> 4];
127 *buf
++ = hex
[val
& 0xf];
134 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
137 for (i
= 0; i
< 20; i
++) {
138 static char hex
[] = "0123456789abcdef";
139 unsigned int val
= sha1
[i
];
140 char *pos
= pathbuf
+ i
*2 + (i
> 0);
141 *pos
++ = hex
[val
>> 4];
142 *pos
= hex
[val
& 0xf];
147 * NOTE! This returns a statically allocated buffer, so you have to be
148 * careful about using it. Do a "xstrdup()" if you need to save the
151 * Also note that this returns the location for creating. Reading
152 * SHA1 file can happen from any alternate directory listed in the
153 * DB_ENVIRONMENT environment variable if it is not found in
154 * the primary object database.
156 char *sha1_file_name(const unsigned char *sha1
)
158 static char *name
, *base
;
161 const char *sha1_file_directory
= get_object_directory();
162 int len
= strlen(sha1_file_directory
);
163 base
= xmalloc(len
+ 60);
164 memcpy(base
, sha1_file_directory
, len
);
165 memset(base
+len
, 0, 60);
168 name
= base
+ len
+ 1;
170 fill_sha1_path(name
, sha1
);
174 char *sha1_pack_name(const unsigned char *sha1
)
176 static const char hex
[] = "0123456789abcdef";
177 static char *name
, *base
, *buf
;
181 const char *sha1_file_directory
= get_object_directory();
182 int len
= strlen(sha1_file_directory
);
183 base
= xmalloc(len
+ 60);
184 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory
);
185 name
= base
+ len
+ 11;
190 for (i
= 0; i
< 20; i
++) {
191 unsigned int val
= *sha1
++;
192 *buf
++ = hex
[val
>> 4];
193 *buf
++ = hex
[val
& 0xf];
199 char *sha1_pack_index_name(const unsigned char *sha1
)
201 static const char hex
[] = "0123456789abcdef";
202 static char *name
, *base
, *buf
;
206 const char *sha1_file_directory
= get_object_directory();
207 int len
= strlen(sha1_file_directory
);
208 base
= xmalloc(len
+ 60);
209 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory
);
210 name
= base
+ len
+ 11;
215 for (i
= 0; i
< 20; i
++) {
216 unsigned int val
= *sha1
++;
217 *buf
++ = hex
[val
>> 4];
218 *buf
++ = hex
[val
& 0xf];
224 struct alternate_object_database
*alt_odb_list
;
225 static struct alternate_object_database
**alt_odb_tail
;
227 static void read_info_alternates(const char * alternates
, int depth
);
230 * Prepare alternate object database registry.
232 * The variable alt_odb_list points at the list of struct
233 * alternate_object_database. The elements on this list come from
234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
236 * whose contents is similar to that environment variable but can be
237 * LF separated. Its base points at a statically allocated buffer that
238 * contains "/the/directory/corresponding/to/.git/objects/...", while
239 * its name points just after the slash at the end of ".git/objects/"
240 * in the example above, and has enough space to hold 40-byte hex
241 * SHA1, an extra slash for the first level indirection, and the
244 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
247 const char *objdir
= get_object_directory();
248 struct alternate_object_database
*ent
;
249 struct alternate_object_database
*alt
;
250 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int entlen
= pfxlen
+ 43;
255 if (*entry
!= '/' && relative_base
) {
256 /* Relative alt-odb */
258 base_len
= strlen(relative_base
) + 1;
262 ent
= xmalloc(sizeof(*ent
) + entlen
);
264 if (*entry
!= '/' && relative_base
) {
265 memcpy(ent
->base
, relative_base
, base_len
- 1);
266 ent
->base
[base_len
- 1] = '/';
267 memcpy(ent
->base
+ base_len
, entry
, len
);
270 memcpy(ent
->base
, entry
, pfxlen
);
272 ent
->name
= ent
->base
+ pfxlen
+ 1;
273 ent
->base
[pfxlen
+ 3] = '/';
274 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
276 /* Detect cases where alternate disappeared */
277 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
278 error("object directory %s does not exist; "
279 "check .git/objects/info/alternates.",
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
288 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
289 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
294 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
299 /* add the alternate entry */
301 alt_odb_tail
= &(ent
->next
);
304 /* recursively add alternates */
305 read_info_alternates(ent
->base
, depth
+ 1);
307 ent
->base
[pfxlen
] = '/';
312 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
313 const char *relative_base
, int depth
)
315 const char *cp
, *last
;
318 error("%s: ignoring alternate object stores, nesting too deep.",
326 if (cp
< ep
&& *cp
== '#') {
327 while (cp
< ep
&& *cp
!= sep
)
332 while (cp
< ep
&& *cp
!= sep
)
335 if ((*last
!= '/') && depth
) {
336 error("%s: ignoring relative alternate object store %s",
337 relative_base
, last
);
339 link_alt_odb_entry(last
, cp
- last
,
340 relative_base
, depth
);
343 while (cp
< ep
&& *cp
== sep
)
349 static void read_info_alternates(const char * relative_base
, int depth
)
357 sprintf(path
, "%s/info/alternates", relative_base
);
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 prepare_alt_odb(void)
378 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
383 alt_odb_tail
= &alt_odb_list
;
384 link_alt_odb_entries(alt
, alt
+ strlen(alt
), ':', NULL
, 0);
386 read_info_alternates(get_object_directory(), 0);
389 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
391 char *name
= sha1_file_name(sha1
);
392 struct alternate_object_database
*alt
;
397 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
399 fill_sha1_path(name
, sha1
);
400 if (!stat(alt
->base
, st
))
406 static unsigned int pack_used_ctr
;
407 static unsigned int pack_mmap_calls
;
408 static unsigned int peak_pack_open_windows
;
409 static unsigned int pack_open_windows
;
410 static size_t peak_pack_mapped
;
411 static size_t pack_mapped
;
412 struct packed_git
*packed_git
;
417 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
418 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
419 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
420 (size_t) getpagesize(),
421 packed_git_window_size
,
424 "pack_report: pack_used_ctr = %10u\n"
425 "pack_report: pack_mmap_calls = %10u\n"
426 "pack_report: pack_open_windows = %10u / %10u\n"
427 "pack_report: pack_mapped = "
428 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
431 pack_open_windows
, peak_pack_open_windows
,
432 pack_mapped
, peak_pack_mapped
);
435 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
438 struct pack_idx_header
*hdr
;
440 uint32_t version
, nr
, i
, *index
;
441 int fd
= open(path
, O_RDONLY
);
446 if (fstat(fd
, &st
)) {
450 idx_size
= xsize_t(st
.st_size
);
451 if (idx_size
< 4 * 256 + 20 + 20) {
453 return error("index file %s is too small", path
);
455 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
459 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
460 version
= ntohl(hdr
->idx_version
);
461 if (version
< 2 || version
> 2) {
462 munmap(idx_map
, idx_size
);
463 return error("index file %s is version %d"
464 " and is not supported by this binary"
465 " (try upgrading GIT to a newer version)",
474 index
+= 2; /* skip index header */
475 for (i
= 0; i
< 256; i
++) {
476 uint32_t n
= ntohl(index
[i
]);
478 munmap(idx_map
, idx_size
);
479 return error("non-monotonic index %s", path
);
487 * - 256 index entries 4 bytes each
488 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
489 * - 20-byte SHA1 of the packfile
490 * - 20-byte SHA1 file checksum
492 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
493 munmap(idx_map
, idx_size
);
494 return error("wrong index file size in %s", path
);
496 } else if (version
== 2) {
499 * - 8 bytes of header
500 * - 256 index entries 4 bytes each
501 * - 20-byte sha1 entry * nr
502 * - 4-byte crc entry * nr
503 * - 4-byte offset entry * nr
504 * - 20-byte SHA1 of the packfile
505 * - 20-byte SHA1 file checksum
506 * And after the 4-byte offset table might be a
507 * variable sized table containing 8-byte entries
508 * for offsets larger than 2^31.
510 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
511 if (idx_size
< min_size
|| idx_size
> min_size
+ (nr
- 1)*8) {
512 munmap(idx_map
, idx_size
);
513 return error("wrong index file size in %s", path
);
515 if (idx_size
!= min_size
) {
516 /* make sure we can deal with large pack offsets */
517 off_t x
= 0x7fffffffUL
, y
= 0xffffffffUL
;
518 if (x
> (x
+ 1) || y
> (y
+ 1)) {
519 munmap(idx_map
, idx_size
);
520 return error("pack too large for current definition of off_t in %s", path
);
525 p
->index_version
= version
;
526 p
->index_data
= idx_map
;
527 p
->index_size
= idx_size
;
532 static void scan_windows(struct packed_git
*p
,
533 struct packed_git
**lru_p
,
534 struct pack_window
**lru_w
,
535 struct pack_window
**lru_l
)
537 struct pack_window
*w
, *w_l
;
539 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
541 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
551 static int unuse_one_window(struct packed_git
*current
)
553 struct packed_git
*p
, *lru_p
= NULL
;
554 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
557 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
558 for (p
= packed_git
; p
; p
= p
->next
)
559 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
561 munmap(lru_w
->base
, lru_w
->len
);
562 pack_mapped
-= lru_w
->len
;
564 lru_l
->next
= lru_w
->next
;
566 lru_p
->windows
= lru_w
->next
;
567 if (!lru_p
->windows
&& lru_p
!= current
) {
568 close(lru_p
->pack_fd
);
579 void release_pack_memory(size_t need
)
581 size_t cur
= pack_mapped
;
582 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
586 void unuse_pack(struct pack_window
**w_cursor
)
588 struct pack_window
*w
= *w_cursor
;
596 * Do not call this directly as this leaks p->pack_fd on error return;
597 * call open_packed_git() instead.
599 static int open_packed_git_1(struct packed_git
*p
)
602 struct pack_header hdr
;
603 unsigned char sha1
[20];
604 unsigned char *idx_sha1
;
607 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
608 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
611 /* If we created the struct before we had the pack we lack size. */
613 if (!S_ISREG(st
.st_mode
))
614 return error("packfile %s not a regular file", p
->pack_name
);
615 p
->pack_size
= st
.st_size
;
616 } else if (p
->pack_size
!= st
.st_size
)
617 return error("packfile %s size changed", p
->pack_name
);
619 /* We leave these file descriptors open with sliding mmap;
620 * there is no point keeping them open across exec(), though.
622 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
624 return error("cannot determine file descriptor flags");
625 fd_flag
|= FD_CLOEXEC
;
626 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
627 return error("cannot set FD_CLOEXEC");
629 /* Verify we recognize this pack file format. */
630 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
631 return error("file %s is far too short to be a packfile", p
->pack_name
);
632 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
633 return error("file %s is not a GIT packfile", p
->pack_name
);
634 if (!pack_version_ok(hdr
.hdr_version
))
635 return error("packfile %s is version %u and not supported"
636 " (try upgrading GIT to a newer version)",
637 p
->pack_name
, ntohl(hdr
.hdr_version
));
639 /* Verify the pack matches its index. */
640 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
641 return error("packfile %s claims to have %u objects"
642 " while index indicates %u objects",
643 p
->pack_name
, ntohl(hdr
.hdr_entries
),
645 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
646 return error("end of packfile %s is unavailable", p
->pack_name
);
647 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
648 return error("packfile %s signature is unavailable", p
->pack_name
);
649 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
650 if (hashcmp(sha1
, idx_sha1
))
651 return error("packfile %s does not match index", p
->pack_name
);
655 static int open_packed_git(struct packed_git
*p
)
657 if (!open_packed_git_1(p
))
659 if (p
->pack_fd
!= -1) {
666 static int in_window(struct pack_window
*win
, off_t offset
)
668 /* We must promise at least 20 bytes (one hash) after the
669 * offset is available from this window, otherwise the offset
670 * is not actually in this window and a different window (which
671 * has that one hash excess) must be used. This is to support
672 * the object header and delta base parsing routines below.
674 off_t win_off
= win
->offset
;
675 return win_off
<= offset
676 && (offset
+ 20) <= (win_off
+ win
->len
);
679 unsigned char* use_pack(struct packed_git
*p
,
680 struct pack_window
**w_cursor
,
684 struct pack_window
*win
= *w_cursor
;
686 if (p
->pack_fd
== -1 && open_packed_git(p
))
687 die("packfile %s cannot be accessed", p
->pack_name
);
689 /* Since packfiles end in a hash of their content and its
690 * pointless to ask for an offset into the middle of that
691 * hash, and the in_window function above wouldn't match
692 * don't allow an offset too close to the end of the file.
694 if (offset
> (p
->pack_size
- 20))
695 die("offset beyond end of packfile (truncated pack?)");
697 if (!win
|| !in_window(win
, offset
)) {
700 for (win
= p
->windows
; win
; win
= win
->next
) {
701 if (in_window(win
, offset
))
705 size_t window_align
= packed_git_window_size
/ 2;
707 win
= xcalloc(1, sizeof(*win
));
708 win
->offset
= (offset
/ window_align
) * window_align
;
709 len
= p
->pack_size
- win
->offset
;
710 if (len
> packed_git_window_size
)
711 len
= packed_git_window_size
;
712 win
->len
= (size_t)len
;
713 pack_mapped
+= win
->len
;
714 while (packed_git_limit
< pack_mapped
715 && unuse_one_window(p
))
717 win
->base
= xmmap(NULL
, win
->len
,
718 PROT_READ
, MAP_PRIVATE
,
719 p
->pack_fd
, win
->offset
);
720 if (win
->base
== MAP_FAILED
)
721 die("packfile %s cannot be mapped: %s",
726 if (pack_mapped
> peak_pack_mapped
)
727 peak_pack_mapped
= pack_mapped
;
728 if (pack_open_windows
> peak_pack_open_windows
)
729 peak_pack_open_windows
= pack_open_windows
;
730 win
->next
= p
->windows
;
734 if (win
!= *w_cursor
) {
735 win
->last_used
= pack_used_ctr
++;
739 offset
-= win
->offset
;
741 *left
= win
->len
- xsize_t(offset
);
742 return win
->base
+ offset
;
745 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
748 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
751 * Make sure a corresponding .pack file exists and that
752 * the index looks sane.
754 path_len
-= strlen(".idx");
757 memcpy(p
->pack_name
, path
, path_len
);
758 strcpy(p
->pack_name
+ path_len
, ".pack");
759 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
) ||
760 check_packed_git_idx(path
, p
)) {
765 /* ok, it looks sane as far as we can check without
766 * actually mapping the pack file.
768 p
->pack_size
= st
.st_size
;
772 p
->pack_local
= local
;
773 p
->mtime
= st
.st_mtime
;
774 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
779 struct packed_git
*parse_pack_index(unsigned char *sha1
)
781 char *path
= sha1_pack_index_name(sha1
);
782 return parse_pack_index_file(sha1
, path
);
785 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
786 const char *idx_path
)
788 const char *path
= sha1_pack_name(sha1
);
789 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
791 if (check_packed_git_idx(idx_path
, p
)) {
796 strcpy(p
->pack_name
, path
);
801 hashcpy(p
->sha1
, sha1
);
805 void install_packed_git(struct packed_git
*pack
)
807 pack
->next
= packed_git
;
811 static void prepare_packed_git_one(char *objdir
, int local
)
818 sprintf(path
, "%s/pack", objdir
);
823 error("unable to open object pack directory: %s: %s",
824 path
, strerror(errno
));
828 while ((de
= readdir(dir
)) != NULL
) {
829 int namelen
= strlen(de
->d_name
);
830 struct packed_git
*p
;
832 if (!has_extension(de
->d_name
, ".idx"))
835 /* Don't reopen a pack we already have. */
836 strcpy(path
+ len
, de
->d_name
);
837 for (p
= packed_git
; p
; p
= p
->next
) {
838 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
843 /* See if it really is a valid .idx file with corresponding
844 * .pack file that we can map.
846 p
= add_packed_git(path
, len
+ namelen
, local
);
849 install_packed_git(p
);
854 static int sort_pack(const void *a_
, const void *b_
)
856 struct packed_git
*a
= *((struct packed_git
**)a_
);
857 struct packed_git
*b
= *((struct packed_git
**)b_
);
861 * Local packs tend to contain objects specific to our
862 * variant of the project than remote ones. In addition,
863 * remote ones could be on a network mounted filesystem.
864 * Favor local ones for these reasons.
866 st
= a
->pack_local
- b
->pack_local
;
871 * Younger packs tend to contain more recent objects,
872 * and more recent objects tend to get accessed more
875 if (a
->mtime
< b
->mtime
)
877 else if (a
->mtime
== b
->mtime
)
882 static void rearrange_packed_git(void)
884 struct packed_git
**ary
, *p
;
887 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
892 /* prepare an array of packed_git for easier sorting */
893 ary
= xcalloc(n
, sizeof(struct packed_git
*));
894 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
897 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
899 /* link them back again */
900 for (i
= 0; i
< n
- 1; i
++)
901 ary
[i
]->next
= ary
[i
+ 1];
902 ary
[n
- 1]->next
= NULL
;
908 static int prepare_packed_git_run_once
= 0;
909 void prepare_packed_git(void)
911 struct alternate_object_database
*alt
;
913 if (prepare_packed_git_run_once
)
915 prepare_packed_git_one(get_object_directory(), 1);
917 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
919 prepare_packed_git_one(alt
->base
, 0);
922 rearrange_packed_git();
923 prepare_packed_git_run_once
= 1;
926 void reprepare_packed_git(void)
928 prepare_packed_git_run_once
= 0;
929 prepare_packed_git();
932 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
934 unsigned char real_sha1
[20];
935 hash_sha1_file(map
, size
, type
, real_sha1
);
936 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
939 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
944 char *filename
= find_sha1_file(sha1
, &st
);
950 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
952 /* See if it works without O_NOATIME */
953 switch (sha1_file_open_flag
) {
955 fd
= open(filename
, O_RDONLY
);
963 /* If it failed once, it will probably fail again.
964 * Stop using O_NOATIME
966 sha1_file_open_flag
= 0;
968 *size
= xsize_t(st
.st_size
);
969 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
974 int legacy_loose_object(unsigned char *map
)
979 * Is it a zlib-compressed buffer? If so, the first byte
980 * must be 0x78 (15-bit window size, deflated), and the
981 * first 16-bit word is evenly divisible by 31
983 word
= (map
[0] << 8) + map
[1];
984 if (map
[0] == 0x78 && !(word
% 31))
990 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
995 unsigned long used
= 0;
998 *type
= (c
>> 4) & 7;
1004 if (sizeof(long) * 8 <= shift
)
1007 size
+= (c
& 0x7f) << shift
;
1014 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1016 unsigned long size
, used
;
1017 static const char valid_loose_object_type
[8] = {
1019 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1020 0, /* "delta" and others are invalid in a loose object */
1022 enum object_type type
;
1024 /* Get the data stream */
1025 memset(stream
, 0, sizeof(*stream
));
1026 stream
->next_in
= map
;
1027 stream
->avail_in
= mapsize
;
1028 stream
->next_out
= buffer
;
1029 stream
->avail_out
= bufsiz
;
1031 if (legacy_loose_object(map
)) {
1032 inflateInit(stream
);
1033 return inflate(stream
, 0);
1036 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1037 if (!used
|| !valid_loose_object_type
[type
])
1042 /* Set up the stream for the rest.. */
1043 stream
->next_in
= map
;
1044 stream
->avail_in
= mapsize
;
1045 inflateInit(stream
);
1047 /* And generate the fake traditional header */
1048 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1049 typename(type
), size
);
1053 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1055 int bytes
= strlen(buffer
) + 1;
1056 unsigned char *buf
= xmalloc(1+size
);
1060 n
= stream
->total_out
- bytes
;
1063 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1065 if (bytes
<= size
) {
1067 * The above condition must be (bytes <= size), not
1068 * (bytes < size). In other words, even though we
1069 * expect no more output and set avail_out to zer0,
1070 * the input zlib stream may have bytes that express
1071 * "this concludes the stream", and we *do* want to
1074 * Otherwise we would not be able to test that we
1075 * consumed all the input to reach the expected size;
1076 * we also want to check that zlib tells us that all
1077 * went well with status == Z_STREAM_END at the end.
1079 stream
->next_out
= buf
+ bytes
;
1080 stream
->avail_out
= size
- bytes
;
1081 while (status
== Z_OK
)
1082 status
= inflate(stream
, Z_FINISH
);
1085 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1091 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1092 else if (stream
->avail_in
)
1093 error("garbage at end of loose object '%s'",
1100 * We used to just use "sscanf()", but that's actually way
1101 * too permissive for what we want to check. So do an anal
1102 * object header parse by hand.
1104 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1111 * The type can be at most ten bytes (including the
1112 * terminating '\0' that we add), and is followed by
1121 if (i
>= sizeof(type
))
1127 * The length must follow immediately, and be in canonical
1128 * decimal format (ie "010" is not valid).
1130 size
= *hdr
++ - '0';
1135 unsigned long c
= *hdr
- '0';
1139 size
= size
* 10 + c
;
1145 * The length must be followed by a zero byte
1147 return *hdr
? -1 : type_from_string(type
);
1150 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1156 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1157 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1160 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1163 static off_t
get_delta_base(struct packed_git
*p
,
1164 struct pack_window
**w_curs
,
1166 enum object_type type
,
1167 off_t delta_obj_offset
)
1169 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1172 /* use_pack() assured us we have [base_info, base_info + 20)
1173 * as a range that we can look at without walking off the
1174 * end of the mapped window. Its actually the hash size
1175 * that is assured. An OFS_DELTA longer than the hash size
1176 * is stupid, as then a REF_DELTA would be smaller to store.
1178 if (type
== OBJ_OFS_DELTA
) {
1180 unsigned char c
= base_info
[used
++];
1181 base_offset
= c
& 127;
1184 if (!base_offset
|| MSB(base_offset
, 7))
1185 die("offset value overflow for delta base object");
1186 c
= base_info
[used
++];
1187 base_offset
= (base_offset
<< 7) + (c
& 127);
1189 base_offset
= delta_obj_offset
- base_offset
;
1190 if (base_offset
>= delta_obj_offset
)
1191 die("delta base offset out of bound");
1193 } else if (type
== OBJ_REF_DELTA
) {
1194 /* The base entry _must_ be in the same pack */
1195 base_offset
= find_pack_entry_one(base_info
, p
);
1197 die("failed to find delta-pack base object %s",
1198 sha1_to_hex(base_info
));
1201 die("I am totally screwed");
1205 /* forward declaration for a mutually recursive function */
1206 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1207 unsigned long *sizep
);
1209 static int packed_delta_info(struct packed_git
*p
,
1210 struct pack_window
**w_curs
,
1212 enum object_type type
,
1214 unsigned long *sizep
)
1218 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1219 type
= packed_object_info(p
, base_offset
, NULL
);
1221 /* We choose to only get the type of the base object and
1222 * ignore potentially corrupt pack file that expects the delta
1223 * based on a base with a wrong size. This saves tons of
1227 const unsigned char *data
;
1228 unsigned char delta_head
[20], *in
;
1232 memset(&stream
, 0, sizeof(stream
));
1233 stream
.next_out
= delta_head
;
1234 stream
.avail_out
= sizeof(delta_head
);
1236 inflateInit(&stream
);
1238 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1239 stream
.next_in
= in
;
1240 st
= inflate(&stream
, Z_FINISH
);
1241 curpos
+= stream
.next_in
- in
;
1242 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
)
1243 && stream
.total_out
< sizeof(delta_head
));
1244 inflateEnd(&stream
);
1245 if ((st
!= Z_STREAM_END
) &&
1246 stream
.total_out
!= sizeof(delta_head
))
1247 die("delta data unpack-initial failed");
1249 /* Examine the initial part of the delta to figure out
1254 /* ignore base size */
1255 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1257 /* Read the result size */
1258 *sizep
= get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1264 static int unpack_object_header(struct packed_git
*p
,
1265 struct pack_window
**w_curs
,
1267 unsigned long *sizep
)
1269 unsigned char *base
;
1272 enum object_type type
;
1274 /* use_pack() assures us we have [base, base + 20) available
1275 * as a range that we can look at at. (Its actually the hash
1276 * size that is assured.) With our object header encoding
1277 * the maximum deflated object size is 2^137, which is just
1278 * insane, so we know won't exceed what we have been given.
1280 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1281 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1283 die("object offset outside of pack file");
1289 const char *packed_object_info_detail(struct packed_git
*p
,
1291 unsigned long *size
,
1292 unsigned long *store_size
,
1293 unsigned int *delta_chain_length
,
1294 unsigned char *base_sha1
)
1296 struct pack_window
*w_curs
= NULL
;
1298 unsigned long dummy
;
1299 unsigned char *next_sha1
;
1300 enum object_type type
;
1302 *delta_chain_length
= 0;
1303 curpos
= obj_offset
;
1304 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1309 die("pack %s contains unknown object type %d",
1310 p
->pack_name
, type
);
1315 *store_size
= 0; /* notyet */
1316 unuse_pack(&w_curs
);
1317 return typename(type
);
1319 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1320 if (*delta_chain_length
== 0) {
1321 /* TODO: find base_sha1 as pointed by curpos */
1326 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1327 if (*delta_chain_length
== 0)
1328 hashcpy(base_sha1
, next_sha1
);
1329 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1332 (*delta_chain_length
)++;
1333 curpos
= obj_offset
;
1334 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1338 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1339 unsigned long *sizep
)
1341 struct pack_window
*w_curs
= NULL
;
1343 off_t curpos
= obj_offset
;
1344 enum object_type type
;
1346 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1351 type
= packed_delta_info(p
, &w_curs
, curpos
,
1352 type
, obj_offset
, sizep
);
1362 die("pack %s contains unknown object type %d",
1363 p
->pack_name
, type
);
1365 unuse_pack(&w_curs
);
1369 static void *unpack_compressed_entry(struct packed_git
*p
,
1370 struct pack_window
**w_curs
,
1376 unsigned char *buffer
, *in
;
1378 buffer
= xmalloc(size
+ 1);
1380 memset(&stream
, 0, sizeof(stream
));
1381 stream
.next_out
= buffer
;
1382 stream
.avail_out
= size
;
1384 inflateInit(&stream
);
1386 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1387 stream
.next_in
= in
;
1388 st
= inflate(&stream
, Z_FINISH
);
1389 curpos
+= stream
.next_in
- in
;
1390 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1391 inflateEnd(&stream
);
1392 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1400 #define MAX_DELTA_CACHE (256)
1402 static size_t delta_base_cached
;
1404 static struct delta_base_cache_lru_list
{
1405 struct delta_base_cache_lru_list
*prev
;
1406 struct delta_base_cache_lru_list
*next
;
1407 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1409 static struct delta_base_cache_entry
{
1410 struct delta_base_cache_lru_list lru
;
1412 struct packed_git
*p
;
1415 enum object_type type
;
1416 } delta_base_cache
[MAX_DELTA_CACHE
];
1418 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1422 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1423 hash
+= (hash
>> 8) + (hash
>> 16);
1424 return hash
% MAX_DELTA_CACHE
;
1427 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1428 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1431 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1432 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1435 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1436 goto found_cache_entry
;
1437 return unpack_entry(p
, base_offset
, type
, base_size
);
1442 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1443 ent
->lru
.prev
->next
= ent
->lru
.next
;
1444 delta_base_cached
-= ent
->size
;
1447 ret
= xmalloc(ent
->size
+ 1);
1448 memcpy(ret
, ent
->data
, ent
->size
);
1449 ((char *)ret
)[ent
->size
] = 0;
1452 *base_size
= ent
->size
;
1456 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1461 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1462 ent
->lru
.prev
->next
= ent
->lru
.next
;
1463 delta_base_cached
-= ent
->size
;
1467 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1468 void *base
, unsigned long base_size
, enum object_type type
)
1470 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1471 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1472 struct delta_base_cache_lru_list
*lru
;
1474 release_delta_base_cache(ent
);
1475 delta_base_cached
+= base_size
;
1477 for (lru
= delta_base_cache_lru
.next
;
1478 delta_base_cached
> delta_base_cache_limit
1479 && lru
!= &delta_base_cache_lru
;
1481 struct delta_base_cache_entry
*f
= (void *)lru
;
1482 if (f
->type
== OBJ_BLOB
)
1483 release_delta_base_cache(f
);
1485 for (lru
= delta_base_cache_lru
.next
;
1486 delta_base_cached
> delta_base_cache_limit
1487 && lru
!= &delta_base_cache_lru
;
1489 struct delta_base_cache_entry
*f
= (void *)lru
;
1490 release_delta_base_cache(f
);
1494 ent
->base_offset
= base_offset
;
1497 ent
->size
= base_size
;
1498 ent
->lru
.next
= &delta_base_cache_lru
;
1499 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1500 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1501 delta_base_cache_lru
.prev
= &ent
->lru
;
1504 static void *unpack_delta_entry(struct packed_git
*p
,
1505 struct pack_window
**w_curs
,
1507 unsigned long delta_size
,
1509 enum object_type
*type
,
1510 unsigned long *sizep
)
1512 void *delta_data
, *result
, *base
;
1513 unsigned long base_size
;
1516 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1517 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1519 die("failed to read delta base object"
1520 " at %"PRIuMAX
" from %s",
1521 (uintmax_t)base_offset
, p
->pack_name
);
1523 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1524 result
= patch_delta(base
, base_size
,
1525 delta_data
, delta_size
,
1528 die("failed to apply delta");
1530 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1534 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1535 enum object_type
*type
, unsigned long *sizep
)
1537 struct pack_window
*w_curs
= NULL
;
1538 off_t curpos
= obj_offset
;
1541 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1545 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1546 obj_offset
, type
, sizep
);
1552 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1555 die("unknown object type %i in %s", *type
, p
->pack_name
);
1557 unuse_pack(&w_curs
);
1561 const unsigned char *nth_packed_object_sha1(const struct packed_git
*p
,
1564 const unsigned char *index
= p
->index_data
;
1565 if (n
>= p
->num_objects
)
1568 if (p
->index_version
== 1) {
1569 return index
+ 24 * n
+ 4;
1572 return index
+ 20 * n
;
1576 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1578 const unsigned char *index
= p
->index_data
;
1580 if (p
->index_version
== 1) {
1581 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1584 index
+= 8 + p
->num_objects
* (20 + 4);
1585 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1586 if (!(off
& 0x80000000))
1588 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1589 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1590 ntohl(*((uint32_t *)(index
+ 4)));
1594 off_t
find_pack_entry_one(const unsigned char *sha1
,
1595 struct packed_git
*p
)
1597 const uint32_t *level1_ofs
= p
->index_data
;
1598 const unsigned char *index
= p
->index_data
;
1601 if (p
->index_version
> 1) {
1606 hi
= ntohl(level1_ofs
[*sha1
]);
1607 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1610 unsigned mi
= (lo
+ hi
) / 2;
1611 unsigned x
= (p
->index_version
> 1) ? (mi
* 20) : (mi
* 24 + 4);
1612 int cmp
= hashcmp(index
+ x
, sha1
);
1614 return nth_packed_object_offset(p
, mi
);
1623 static int matches_pack_name(struct packed_git
*p
, const char *ig
)
1625 const char *last_c
, *c
;
1627 if (!strcmp(p
->pack_name
, ig
))
1630 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1635 if (!strcmp(last_c
, ig
))
1641 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1643 struct packed_git
*p
;
1646 prepare_packed_git();
1648 for (p
= packed_git
; p
; p
= p
->next
) {
1649 if (ignore_packed
) {
1651 for (ig
= ignore_packed
; *ig
; ig
++)
1652 if (!matches_pack_name(p
, *ig
))
1657 offset
= find_pack_entry_one(sha1
, p
);
1660 * We are about to tell the caller where they can
1661 * locate the requested object. We better make
1662 * sure the packfile is still here and can be
1663 * accessed before supplying that answer, as
1664 * it may have been deleted since the index
1667 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1668 error("packfile %s cannot be accessed", p
->pack_name
);
1673 hashcpy(e
->sha1
, sha1
);
1680 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1681 struct packed_git
*packs
)
1683 struct packed_git
*p
;
1685 for (p
= packs
; p
; p
= p
->next
) {
1686 if (find_pack_entry_one(sha1
, p
))
1693 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1696 unsigned long mapsize
, size
;
1701 map
= map_sha1_file(sha1
, &mapsize
);
1703 return error("unable to find %s", sha1_to_hex(sha1
));
1704 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1705 status
= error("unable to unpack %s header",
1707 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1708 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1711 inflateEnd(&stream
);
1712 munmap(map
, mapsize
);
1716 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1718 struct pack_entry e
;
1720 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1721 reprepare_packed_git();
1722 if (!find_pack_entry(sha1
, &e
, NULL
))
1723 return sha1_loose_object_info(sha1
, sizep
);
1725 return packed_object_info(e
.p
, e
.offset
, sizep
);
1728 static void *read_packed_sha1(const unsigned char *sha1
,
1729 enum object_type
*type
, unsigned long *size
)
1731 struct pack_entry e
;
1733 if (!find_pack_entry(sha1
, &e
, NULL
))
1736 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1740 * This is meant to hold a *small* number of objects that you would
1741 * want read_sha1_file() to be able to return, but yet you do not want
1742 * to write them into the object store (e.g. a browse-only
1745 static struct cached_object
{
1746 unsigned char sha1
[20];
1747 enum object_type type
;
1751 static int cached_object_nr
, cached_object_alloc
;
1753 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1756 struct cached_object
*co
= cached_objects
;
1758 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1759 if (!hashcmp(co
->sha1
, sha1
))
1765 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1766 unsigned char *sha1
)
1768 struct cached_object
*co
;
1770 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1771 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1773 if (cached_object_alloc
<= cached_object_nr
) {
1774 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1775 cached_objects
= xrealloc(cached_objects
,
1776 sizeof(*cached_objects
) *
1777 cached_object_alloc
);
1779 co
= &cached_objects
[cached_object_nr
++];
1782 co
->buf
= xmalloc(len
);
1783 memcpy(co
->buf
, buf
, len
);
1784 hashcpy(co
->sha1
, sha1
);
1788 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1789 unsigned long *size
)
1791 unsigned long mapsize
;
1793 struct cached_object
*co
;
1795 co
= find_cached_object(sha1
);
1797 buf
= xmalloc(co
->size
+ 1);
1798 memcpy(buf
, co
->buf
, co
->size
);
1799 ((char*)buf
)[co
->size
] = 0;
1805 buf
= read_packed_sha1(sha1
, type
, size
);
1808 map
= map_sha1_file(sha1
, &mapsize
);
1810 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1811 munmap(map
, mapsize
);
1814 reprepare_packed_git();
1815 return read_packed_sha1(sha1
, type
, size
);
1818 void *read_object_with_reference(const unsigned char *sha1
,
1819 const char *required_type_name
,
1820 unsigned long *size
,
1821 unsigned char *actual_sha1_return
)
1823 enum object_type type
, required_type
;
1825 unsigned long isize
;
1826 unsigned char actual_sha1
[20];
1828 required_type
= type_from_string(required_type_name
);
1829 hashcpy(actual_sha1
, sha1
);
1831 int ref_length
= -1;
1832 const char *ref_type
= NULL
;
1834 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1837 if (type
== required_type
) {
1839 if (actual_sha1_return
)
1840 hashcpy(actual_sha1_return
, actual_sha1
);
1843 /* Handle references */
1844 else if (type
== OBJ_COMMIT
)
1846 else if (type
== OBJ_TAG
)
1847 ref_type
= "object ";
1852 ref_length
= strlen(ref_type
);
1854 if (memcmp(buffer
, ref_type
, ref_length
) ||
1855 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1860 /* Now we have the ID of the referred-to object in
1861 * actual_sha1. Check again. */
1865 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
1866 const char *type
, unsigned char *sha1
,
1867 char *hdr
, int *hdrlen
)
1871 /* Generate the header */
1872 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
1876 SHA1_Update(&c
, hdr
, *hdrlen
);
1877 SHA1_Update(&c
, buf
, len
);
1878 SHA1_Final(sha1
, &c
);
1882 * Link the tempfile to the final place, possibly creating the
1883 * last directory level as you do so.
1885 * Returns the errno on failure, 0 on success.
1887 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
1892 if (!link(tmpfile
, filename
))
1896 * Try to mkdir the last path component if that failed.
1898 * Re-try the "link()" regardless of whether the mkdir
1899 * succeeds, since a race might mean that somebody
1903 dir
= strrchr(filename
, '/');
1906 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
1911 if (!link(tmpfile
, filename
))
1919 * Move the just written object into its final resting place
1921 int move_temp_to_file(const char *tmpfile
, const char *filename
)
1923 int ret
= link_temp_to_file(tmpfile
, filename
);
1926 * Coda hack - coda doesn't like cross-directory links,
1927 * so we fall back to a rename, which will mean that it
1928 * won't be able to check collisions, but that's not a
1931 * The same holds for FAT formatted media.
1933 * When this succeeds, we just return 0. We have nothing
1936 if (ret
&& ret
!= EEXIST
) {
1937 if (!rename(tmpfile
, filename
))
1943 if (ret
!= EEXIST
) {
1944 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
1946 /* FIXME!!! Collision check here ? */
1952 static int write_buffer(int fd
, const void *buf
, size_t len
)
1954 if (write_in_full(fd
, buf
, len
) < 0)
1955 return error("file write error (%s)", strerror(errno
));
1959 static int write_binary_header(unsigned char *hdr
, enum object_type type
, unsigned long len
)
1964 c
= (type
<< 4) | (len
& 15);
1977 static void setup_object_header(z_stream
*stream
, const char *type
, unsigned long len
)
1979 int obj_type
, hdrlen
;
1981 if (use_legacy_headers
) {
1982 while (deflate(stream
, 0) == Z_OK
)
1986 obj_type
= type_from_string(type
);
1987 hdrlen
= write_binary_header(stream
->next_out
, obj_type
, len
);
1988 stream
->total_out
= hdrlen
;
1989 stream
->next_out
+= hdrlen
;
1990 stream
->avail_out
-= hdrlen
;
1993 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
1994 unsigned char *sha1
)
1998 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2002 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2005 unsigned char *compressed
;
2007 unsigned char sha1
[20];
2009 static char tmpfile
[PATH_MAX
];
2013 /* Normally if we have it in the pack then we do not bother writing
2014 * it out into .git/objects/??/?{38} file.
2016 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2017 filename
= sha1_file_name(sha1
);
2019 hashcpy(returnsha1
, sha1
);
2020 if (has_sha1_file(sha1
))
2022 fd
= open(filename
, O_RDONLY
);
2025 * FIXME!!! We might do collision checking here, but we'd
2026 * need to uncompress the old file and check it. Later.
2032 if (errno
!= ENOENT
) {
2033 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2036 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2038 fd
= mkstemp(tmpfile
);
2041 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2043 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2047 memset(&stream
, 0, sizeof(stream
));
2048 deflateInit(&stream
, zlib_compression_level
);
2049 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2050 compressed
= xmalloc(size
);
2053 stream
.next_out
= compressed
;
2054 stream
.avail_out
= size
;
2056 /* First header.. */
2057 stream
.next_in
= (unsigned char *)hdr
;
2058 stream
.avail_in
= hdrlen
;
2059 setup_object_header(&stream
, type
, len
);
2061 /* Then the data itself.. */
2062 stream
.next_in
= buf
;
2063 stream
.avail_in
= len
;
2064 ret
= deflate(&stream
, Z_FINISH
);
2065 if (ret
!= Z_STREAM_END
)
2066 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2068 ret
= deflateEnd(&stream
);
2070 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2072 size
= stream
.total_out
;
2074 if (write_buffer(fd
, compressed
, size
) < 0)
2075 die("unable to write sha1 file");
2078 die("unable to write sha1 file");
2081 return move_temp_to_file(tmpfile
, filename
);
2085 * We need to unpack and recompress the object for writing
2086 * it out to a different file.
2088 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2092 unsigned char *unpacked
;
2094 enum object_type type
;
2099 /* need to unpack and recompress it by itself */
2100 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2102 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2104 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2107 memset(&stream
, 0, sizeof(stream
));
2108 deflateInit(&stream
, zlib_compression_level
);
2109 size
= deflateBound(&stream
, len
+ hdrlen
);
2110 buf
= xmalloc(size
);
2113 stream
.next_out
= buf
;
2114 stream
.avail_out
= size
;
2116 /* First header.. */
2117 stream
.next_in
= (void *)hdr
;
2118 stream
.avail_in
= hdrlen
;
2119 while (deflate(&stream
, 0) == Z_OK
)
2122 /* Then the data itself.. */
2123 stream
.next_in
= unpacked
;
2124 stream
.avail_in
= len
;
2125 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2127 deflateEnd(&stream
);
2130 *objsize
= stream
.total_out
;
2134 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2137 unsigned long objsize
;
2138 void *buf
= map_sha1_file(sha1
, &objsize
);
2141 retval
= write_buffer(fd
, buf
, objsize
);
2142 munmap(buf
, objsize
);
2146 buf
= repack_object(sha1
, &objsize
);
2147 retval
= write_buffer(fd
, buf
, objsize
);
2152 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2153 size_t bufsize
, size_t *bufposn
)
2155 char tmpfile
[PATH_MAX
];
2158 unsigned char real_sha1
[20];
2159 unsigned char discard
[4096];
2163 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2165 local
= mkstemp(tmpfile
);
2168 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2170 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2173 memset(&stream
, 0, sizeof(stream
));
2175 inflateInit(&stream
);
2182 stream
.avail_in
= *bufposn
;
2183 stream
.next_in
= (unsigned char *) buffer
;
2185 stream
.next_out
= discard
;
2186 stream
.avail_out
= sizeof(discard
);
2187 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2188 SHA1_Update(&c
, discard
, sizeof(discard
) -
2190 } while (stream
.avail_in
&& ret
== Z_OK
);
2191 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2192 die("unable to write sha1 file");
2193 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2195 *bufposn
= stream
.avail_in
;
2199 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2204 return error("Connection closed?");
2205 perror("Reading from connection");
2210 inflateEnd(&stream
);
2212 fchmod(local
, 0444);
2213 if (close(local
) != 0)
2214 die("unable to write sha1 file");
2215 SHA1_Final(real_sha1
, &c
);
2216 if (ret
!= Z_STREAM_END
) {
2218 return error("File %s corrupted", sha1_to_hex(sha1
));
2220 if (hashcmp(sha1
, real_sha1
)) {
2222 return error("File %s has bad hash", sha1_to_hex(sha1
));
2225 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2228 int has_pack_index(const unsigned char *sha1
)
2231 if (stat(sha1_pack_index_name(sha1
), &st
))
2236 int has_pack_file(const unsigned char *sha1
)
2239 if (stat(sha1_pack_name(sha1
), &st
))
2244 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2246 struct pack_entry e
;
2247 return find_pack_entry(sha1
, &e
, ignore_packed
);
2250 int has_sha1_file(const unsigned char *sha1
)
2253 struct pack_entry e
;
2255 if (find_pack_entry(sha1
, &e
, NULL
))
2257 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2261 * reads from fd as long as possible into a supplied buffer of size bytes.
2262 * If necessary the buffer's size is increased using realloc()
2264 * returns 0 if anything went fine and -1 otherwise
2266 * NOTE: both buf and size may change, but even when -1 is returned
2267 * you still have to free() it yourself.
2269 int read_pipe(int fd
, char** return_buf
, unsigned long* return_size
)
2271 char* buf
= *return_buf
;
2272 unsigned long size
= *return_size
;
2274 unsigned long off
= 0;
2277 iret
= xread(fd
, buf
+ off
, size
- off
);
2282 buf
= xrealloc(buf
, size
);
2295 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2297 unsigned long size
= 4096;
2298 char *buf
= xmalloc(size
);
2301 if (read_pipe(fd
, &buf
, &size
)) {
2309 ret
= write_sha1_file(buf
, size
, type
, sha1
);
2311 ret
= hash_sha1_file(buf
, size
, type
, sha1
);
2316 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2317 enum object_type type
, const char *path
)
2319 size_t size
= xsize_t(st
->st_size
);
2321 int ret
, re_allocated
= 0;
2324 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2331 * Convert blobs to git internal format
2333 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2334 unsigned long nsize
= size
;
2336 if (convert_to_git(path
, &nbuf
, &nsize
)) {
2346 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2348 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2358 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2364 switch (st
->st_mode
& S_IFMT
) {
2366 fd
= open(path
, O_RDONLY
);
2368 return error("open(\"%s\"): %s", path
,
2370 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2371 return error("%s: failed to insert into database",
2375 len
= xsize_t(st
->st_size
);
2376 target
= xmalloc(len
+ 1);
2377 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2378 char *errstr
= strerror(errno
);
2380 return error("readlink(\"%s\"): %s", path
,
2384 hash_sha1_file(target
, len
, blob_type
, sha1
);
2385 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2386 return error("%s: failed to insert into database",
2391 return error("%s: unsupported file type", path
);
2396 int read_pack_header(int fd
, struct pack_header
*header
)
2398 char *c
= (char*)header
;
2399 ssize_t remaining
= sizeof(struct pack_header
);
2401 ssize_t r
= xread(fd
, c
, remaining
);
2403 /* "eof before pack header was fully read" */
2404 return PH_ERROR_EOF
;
2407 } while (remaining
> 0);
2408 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2409 /* "protocol error (pack signature mismatch detected)" */
2410 return PH_ERROR_PACK_SIGNATURE
;
2411 if (!pack_version_ok(header
->hdr_version
))
2412 /* "protocol error (pack version unsupported)" */
2413 return PH_ERROR_PROTOCOL
;