2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
19 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
20 #define O_NOATIME 01000000
28 static unsigned long sz_fmt(size_t s
) { return (unsigned long)s
; }
31 static size_t sz_fmt(size_t s
) { return s
; }
34 const unsigned char null_sha1
[20];
36 static unsigned int sha1_file_open_flag
= O_NOATIME
;
38 const signed char hexval_table
[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
76 for (i
= 0; i
< 20; i
++) {
77 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
86 /* returns the number of chars to skip to first component */
87 static inline int is_path_absolute(const char *path
)
90 if (isalpha(path
[0]) && path
[1] == ':')
91 return 2 + (path
[2] == '/');
92 /* TODO: C:dir/file 'relative' paths are not catered for */
97 int safe_create_leading_directories(char *path
)
99 char *pos
= path
+ is_path_absolute(path
);
103 pos
= strchr(pos
, '/');
107 if (!stat(path
, &st
)) {
109 if (!S_ISDIR(st
.st_mode
)) {
114 else if (mkdir(path
, 0777)) {
118 else if (adjust_shared_perm(path
)) {
127 char * sha1_to_hex(const unsigned char *sha1
)
130 static char hexbuffer
[4][50];
131 static const char hex
[] = "0123456789abcdef";
132 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
135 for (i
= 0; i
< 20; i
++) {
136 unsigned int val
= *sha1
++;
137 *buf
++ = hex
[val
>> 4];
138 *buf
++ = hex
[val
& 0xf];
145 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
148 for (i
= 0; i
< 20; i
++) {
149 static char hex
[] = "0123456789abcdef";
150 unsigned int val
= sha1
[i
];
151 char *pos
= pathbuf
+ i
*2 + (i
> 0);
152 *pos
++ = hex
[val
>> 4];
153 *pos
= hex
[val
& 0xf];
158 * NOTE! This returns a statically allocated buffer, so you have to be
159 * careful about using it. Do a "xstrdup()" if you need to save the
162 * Also note that this returns the location for creating. Reading
163 * SHA1 file can happen from any alternate directory listed in the
164 * DB_ENVIRONMENT environment variable if it is not found in
165 * the primary object database.
167 char *sha1_file_name(const unsigned char *sha1
)
169 static char *name
, *base
;
172 const char *sha1_file_directory
= get_object_directory();
173 int len
= strlen(sha1_file_directory
);
174 base
= xmalloc(len
+ 60);
175 memcpy(base
, sha1_file_directory
, len
);
176 memset(base
+len
, 0, 60);
179 name
= base
+ len
+ 1;
181 fill_sha1_path(name
, sha1
);
185 char *sha1_pack_name(const unsigned char *sha1
)
187 static const char hex
[] = "0123456789abcdef";
188 static char *name
, *base
, *buf
;
192 const char *sha1_file_directory
= get_object_directory();
193 int len
= strlen(sha1_file_directory
);
194 base
= xmalloc(len
+ 60);
195 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory
);
196 name
= base
+ len
+ 11;
201 for (i
= 0; i
< 20; i
++) {
202 unsigned int val
= *sha1
++;
203 *buf
++ = hex
[val
>> 4];
204 *buf
++ = hex
[val
& 0xf];
210 char *sha1_pack_index_name(const unsigned char *sha1
)
212 static const char hex
[] = "0123456789abcdef";
213 static char *name
, *base
, *buf
;
217 const char *sha1_file_directory
= get_object_directory();
218 int len
= strlen(sha1_file_directory
);
219 base
= xmalloc(len
+ 60);
220 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory
);
221 name
= base
+ len
+ 11;
226 for (i
= 0; i
< 20; i
++) {
227 unsigned int val
= *sha1
++;
228 *buf
++ = hex
[val
>> 4];
229 *buf
++ = hex
[val
& 0xf];
235 struct alternate_object_database
*alt_odb_list
;
236 static struct alternate_object_database
**alt_odb_tail
;
238 static void read_info_alternates(const char * alternates
, int depth
);
241 * Prepare alternate object database registry.
243 * The variable alt_odb_list points at the list of struct
244 * alternate_object_database. The elements on this list come from
245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
247 * whose contents is similar to that environment variable but can be
248 * LF separated. Its base points at a statically allocated buffer that
249 * contains "/the/directory/corresponding/to/.git/objects/...", while
250 * its name points just after the slash at the end of ".git/objects/"
251 * in the example above, and has enough space to hold 40-byte hex
252 * SHA1, an extra slash for the first level indirection, and the
255 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
258 const char *objdir
= get_object_directory();
259 struct alternate_object_database
*ent
;
260 struct alternate_object_database
*alt
;
261 /* 43 = 40-byte + 2 '/' + terminating NUL */
263 int entlen
= pfxlen
+ 43;
266 if (!is_absolute_path(entry
) && relative_base
) {
267 /* Relative alt-odb */
269 base_len
= strlen(relative_base
) + 1;
273 ent
= xmalloc(sizeof(*ent
) + entlen
);
275 if (!is_absolute_path(entry
) && relative_base
) {
276 memcpy(ent
->base
, relative_base
, base_len
- 1);
277 ent
->base
[base_len
- 1] = '/';
278 memcpy(ent
->base
+ base_len
, entry
, len
);
281 memcpy(ent
->base
, entry
, pfxlen
);
283 ent
->name
= ent
->base
+ pfxlen
+ 1;
284 ent
->base
[pfxlen
+ 3] = '/';
285 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
287 /* Detect cases where alternate disappeared */
288 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
289 error("object directory %s does not exist; "
290 "check .git/objects/info/alternates.",
296 /* Prevent the common mistake of listing the same
297 * thing twice, or object directory itself.
299 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
300 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
305 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
310 /* add the alternate entry */
312 alt_odb_tail
= &(ent
->next
);
315 /* recursively add alternates */
316 read_info_alternates(ent
->base
, depth
+ 1);
318 ent
->base
[pfxlen
] = '/';
323 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
324 const char *relative_base
, int depth
)
326 const char *cp
, *last
;
329 error("%s: ignoring alternate object stores, nesting too deep.",
337 if (cp
< ep
&& *cp
== '#') {
338 while (cp
< ep
&& *cp
!= sep
)
343 while (cp
< ep
&& *cp
!= sep
)
346 if (!is_absolute_path(last
) && depth
) {
347 error("%s: ignoring relative alternate object store %s",
348 relative_base
, last
);
350 link_alt_odb_entry(last
, cp
- last
,
351 relative_base
, depth
);
354 while (cp
< ep
&& *cp
== sep
)
360 static void read_info_alternates(const char * relative_base
, int depth
)
365 const char alt_file_name
[] = "info/alternates";
366 /* Given that relative_base is no longer than PATH_MAX,
367 ensure that "path" has enough space to append "/", the
368 file name, "info/alternates", and a trailing NUL. */
369 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
372 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
373 fd
= open(path
, O_RDONLY
);
376 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
380 mapsz
= xsize_t(st
.st_size
);
381 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
384 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
389 void prepare_alt_odb(void)
396 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
399 alt_odb_tail
= &alt_odb_list
;
400 link_alt_odb_entries(alt
, alt
+ strlen(alt
), ':', NULL
, 0);
402 read_info_alternates(get_object_directory(), 0);
405 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
407 char *name
= sha1_file_name(sha1
);
408 struct alternate_object_database
*alt
;
413 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
415 fill_sha1_path(name
, sha1
);
416 if (!stat(alt
->base
, st
))
422 static unsigned int pack_used_ctr
;
423 static unsigned int pack_mmap_calls
;
424 static unsigned int peak_pack_open_windows
;
425 static unsigned int pack_open_windows
;
426 static size_t peak_pack_mapped
;
427 static size_t pack_mapped
;
428 struct packed_git
*packed_git
;
430 void pack_report(void)
433 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
434 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
435 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
436 sz_fmt(getpagesize()),
437 sz_fmt(packed_git_window_size
),
438 sz_fmt(packed_git_limit
));
440 "pack_report: pack_used_ctr = %10u\n"
441 "pack_report: pack_mmap_calls = %10u\n"
442 "pack_report: pack_open_windows = %10u / %10u\n"
443 "pack_report: pack_mapped = "
444 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
447 pack_open_windows
, peak_pack_open_windows
,
448 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
451 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
454 struct pack_idx_header
*hdr
;
456 uint32_t version
, nr
, i
, *index
;
457 int fd
= open(path
, O_RDONLY
);
462 if (fstat(fd
, &st
)) {
466 idx_size
= xsize_t(st
.st_size
);
467 if (idx_size
< 4 * 256 + 20 + 20) {
469 return error("index file %s is too small", path
);
471 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
475 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
476 version
= ntohl(hdr
->idx_version
);
477 if (version
< 2 || version
> 2) {
478 munmap(idx_map
, idx_size
);
479 return error("index file %s is version %d"
480 " and is not supported by this binary"
481 " (try upgrading GIT to a newer version)",
490 index
+= 2; /* skip index header */
491 for (i
= 0; i
< 256; i
++) {
492 uint32_t n
= ntohl(index
[i
]);
494 munmap(idx_map
, idx_size
);
495 return error("non-monotonic index %s", path
);
503 * - 256 index entries 4 bytes each
504 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
505 * - 20-byte SHA1 of the packfile
506 * - 20-byte SHA1 file checksum
508 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
509 munmap(idx_map
, idx_size
);
510 return error("wrong index v1 file size in %s", path
);
512 } else if (version
== 2) {
515 * - 8 bytes of header
516 * - 256 index entries 4 bytes each
517 * - 20-byte sha1 entry * nr
518 * - 4-byte crc entry * nr
519 * - 4-byte offset entry * nr
520 * - 20-byte SHA1 of the packfile
521 * - 20-byte SHA1 file checksum
522 * And after the 4-byte offset table might be a
523 * variable sized table containing 8-byte entries
524 * for offsets larger than 2^31.
526 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
527 unsigned long max_size
= min_size
;
529 max_size
+= (nr
- 1)*8;
530 if (idx_size
< min_size
|| idx_size
> max_size
) {
531 munmap(idx_map
, idx_size
);
532 return error("wrong index v2 file size in %s", path
);
534 if (idx_size
!= min_size
&&
536 * make sure we can deal with large pack offsets.
537 * 31-bit signed offset won't be enough, neither
538 * 32-bit unsigned one will be.
540 (sizeof(off_t
) <= 4)) {
541 munmap(idx_map
, idx_size
);
542 return error("pack too large for current definition of off_t in %s", path
);
546 p
->index_version
= version
;
547 p
->index_data
= idx_map
;
548 p
->index_size
= idx_size
;
553 int open_pack_index(struct packed_git
*p
)
561 idx_name
= xstrdup(p
->pack_name
);
562 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
563 ret
= check_packed_git_idx(idx_name
, p
);
568 static void scan_windows(struct packed_git
*p
,
569 struct packed_git
**lru_p
,
570 struct pack_window
**lru_w
,
571 struct pack_window
**lru_l
)
573 struct pack_window
*w
, *w_l
;
575 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
577 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
587 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
589 struct packed_git
*p
, *lru_p
= NULL
;
590 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
593 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
594 for (p
= packed_git
; p
; p
= p
->next
)
595 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
597 munmap(lru_w
->base
, lru_w
->len
);
598 pack_mapped
-= lru_w
->len
;
600 lru_l
->next
= lru_w
->next
;
602 lru_p
->windows
= lru_w
->next
;
603 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
604 close(lru_p
->pack_fd
);
615 void release_pack_memory(size_t need
, int fd
)
617 size_t cur
= pack_mapped
;
618 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
622 void unuse_pack(struct pack_window
**w_cursor
)
624 struct pack_window
*w
= *w_cursor
;
632 * Do not call this directly as this leaks p->pack_fd on error return;
633 * call open_packed_git() instead.
635 static int open_packed_git_1(struct packed_git
*p
)
638 struct pack_header hdr
;
639 unsigned char sha1
[20];
640 unsigned char *idx_sha1
;
643 if (!p
->index_data
&& open_pack_index(p
))
644 return error("packfile %s index unavailable", p
->pack_name
);
646 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
647 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
650 /* If we created the struct before we had the pack we lack size. */
652 if (!S_ISREG(st
.st_mode
))
653 return error("packfile %s not a regular file", p
->pack_name
);
654 p
->pack_size
= st
.st_size
;
655 } else if (p
->pack_size
!= st
.st_size
)
656 return error("packfile %s size changed", p
->pack_name
);
658 /* We leave these file descriptors open with sliding mmap;
659 * there is no point keeping them open across exec(), though.
661 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
663 return error("cannot determine file descriptor flags");
664 fd_flag
|= FD_CLOEXEC
;
665 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
666 return error("cannot set FD_CLOEXEC");
668 /* Verify we recognize this pack file format. */
669 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
670 return error("file %s is far too short to be a packfile", p
->pack_name
);
671 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
672 return error("file %s is not a GIT packfile", p
->pack_name
);
673 if (!pack_version_ok(hdr
.hdr_version
))
674 return error("packfile %s is version %u and not supported"
675 " (try upgrading GIT to a newer version)",
676 p
->pack_name
, ntohl(hdr
.hdr_version
));
678 /* Verify the pack matches its index. */
679 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
680 return error("packfile %s claims to have %u objects"
681 " while index indicates %u objects",
682 p
->pack_name
, ntohl(hdr
.hdr_entries
),
684 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
685 return error("end of packfile %s is unavailable", p
->pack_name
);
686 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
687 return error("packfile %s signature is unavailable", p
->pack_name
);
688 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
689 if (hashcmp(sha1
, idx_sha1
))
690 return error("packfile %s does not match index", p
->pack_name
);
694 static int open_packed_git(struct packed_git
*p
)
696 if (!open_packed_git_1(p
))
698 if (p
->pack_fd
!= -1) {
705 static int in_window(struct pack_window
*win
, off_t offset
)
707 /* We must promise at least 20 bytes (one hash) after the
708 * offset is available from this window, otherwise the offset
709 * is not actually in this window and a different window (which
710 * has that one hash excess) must be used. This is to support
711 * the object header and delta base parsing routines below.
713 off_t win_off
= win
->offset
;
714 return win_off
<= offset
715 && (offset
+ 20) <= (win_off
+ win
->len
);
718 unsigned char* use_pack(struct packed_git
*p
,
719 struct pack_window
**w_cursor
,
723 struct pack_window
*win
= *w_cursor
;
725 if (p
->pack_fd
== -1 && open_packed_git(p
))
726 die("packfile %s cannot be accessed", p
->pack_name
);
728 /* Since packfiles end in a hash of their content and its
729 * pointless to ask for an offset into the middle of that
730 * hash, and the in_window function above wouldn't match
731 * don't allow an offset too close to the end of the file.
733 if (offset
> (p
->pack_size
- 20))
734 die("offset beyond end of packfile (truncated pack?)");
736 if (!win
|| !in_window(win
, offset
)) {
739 for (win
= p
->windows
; win
; win
= win
->next
) {
740 if (in_window(win
, offset
))
744 size_t window_align
= packed_git_window_size
/ 2;
746 win
= xcalloc(1, sizeof(*win
));
747 win
->offset
= (offset
/ window_align
) * window_align
;
748 len
= p
->pack_size
- win
->offset
;
749 if (len
> packed_git_window_size
)
750 len
= packed_git_window_size
;
751 win
->len
= (size_t)len
;
752 pack_mapped
+= win
->len
;
753 while (packed_git_limit
< pack_mapped
754 && unuse_one_window(p
, p
->pack_fd
))
756 win
->base
= xmmap(NULL
, win
->len
,
757 PROT_READ
, MAP_PRIVATE
,
758 p
->pack_fd
, win
->offset
);
759 if (win
->base
== MAP_FAILED
)
760 die("packfile %s cannot be mapped: %s",
765 if (pack_mapped
> peak_pack_mapped
)
766 peak_pack_mapped
= pack_mapped
;
767 if (pack_open_windows
> peak_pack_open_windows
)
768 peak_pack_open_windows
= pack_open_windows
;
769 win
->next
= p
->windows
;
773 if (win
!= *w_cursor
) {
774 win
->last_used
= pack_used_ctr
++;
778 offset
-= win
->offset
;
780 *left
= win
->len
- xsize_t(offset
);
781 return win
->base
+ offset
;
784 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
787 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
790 * Make sure a corresponding .pack file exists and that
791 * the index looks sane.
793 path_len
-= strlen(".idx");
796 memcpy(p
->pack_name
, path
, path_len
);
797 strcpy(p
->pack_name
+ path_len
, ".pack");
798 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
803 /* ok, it looks sane as far as we can check without
804 * actually mapping the pack file.
806 p
->index_version
= 0;
807 p
->index_data
= NULL
;
810 p
->pack_size
= st
.st_size
;
814 p
->pack_local
= local
;
815 p
->mtime
= st
.st_mtime
;
816 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
821 struct packed_git
*parse_pack_index(unsigned char *sha1
)
823 char *path
= sha1_pack_index_name(sha1
);
824 return parse_pack_index_file(sha1
, path
);
827 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
828 const char *idx_path
)
830 const char *path
= sha1_pack_name(sha1
);
831 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
833 if (check_packed_git_idx(idx_path
, p
)) {
838 strcpy(p
->pack_name
, path
);
843 hashcpy(p
->sha1
, sha1
);
847 void install_packed_git(struct packed_git
*pack
)
849 pack
->next
= packed_git
;
853 static void prepare_packed_git_one(char *objdir
, int local
)
855 /* Ensure that this buffer is large enough so that we can
856 append "/pack/" without clobbering the stack even if
857 strlen(objdir) were PATH_MAX. */
858 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
863 sprintf(path
, "%s/pack", objdir
);
868 error("unable to open object pack directory: %s: %s",
869 path
, strerror(errno
));
873 while ((de
= readdir(dir
)) != NULL
) {
874 int namelen
= strlen(de
->d_name
);
875 struct packed_git
*p
;
877 if (!has_extension(de
->d_name
, ".idx"))
880 if (len
+ namelen
+ 1 > sizeof(path
))
883 /* Don't reopen a pack we already have. */
884 strcpy(path
+ len
, de
->d_name
);
885 for (p
= packed_git
; p
; p
= p
->next
) {
886 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
891 /* See if it really is a valid .idx file with corresponding
892 * .pack file that we can map.
894 p
= add_packed_git(path
, len
+ namelen
, local
);
897 install_packed_git(p
);
902 static int sort_pack(const void *a_
, const void *b_
)
904 struct packed_git
*a
= *((struct packed_git
**)a_
);
905 struct packed_git
*b
= *((struct packed_git
**)b_
);
909 * Local packs tend to contain objects specific to our
910 * variant of the project than remote ones. In addition,
911 * remote ones could be on a network mounted filesystem.
912 * Favor local ones for these reasons.
914 st
= a
->pack_local
- b
->pack_local
;
919 * Younger packs tend to contain more recent objects,
920 * and more recent objects tend to get accessed more
923 if (a
->mtime
< b
->mtime
)
925 else if (a
->mtime
== b
->mtime
)
930 static void rearrange_packed_git(void)
932 struct packed_git
**ary
, *p
;
935 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
940 /* prepare an array of packed_git for easier sorting */
941 ary
= xcalloc(n
, sizeof(struct packed_git
*));
942 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
945 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
947 /* link them back again */
948 for (i
= 0; i
< n
- 1; i
++)
949 ary
[i
]->next
= ary
[i
+ 1];
950 ary
[n
- 1]->next
= NULL
;
956 static int prepare_packed_git_run_once
= 0;
957 void prepare_packed_git(void)
959 struct alternate_object_database
*alt
;
961 if (prepare_packed_git_run_once
)
963 prepare_packed_git_one(get_object_directory(), 1);
965 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
967 prepare_packed_git_one(alt
->base
, 0);
970 rearrange_packed_git();
971 prepare_packed_git_run_once
= 1;
974 void reprepare_packed_git(void)
976 prepare_packed_git_run_once
= 0;
977 prepare_packed_git();
980 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
982 unsigned char real_sha1
[20];
983 hash_sha1_file(map
, size
, type
, real_sha1
);
984 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
987 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
992 char *filename
= find_sha1_file(sha1
, &st
);
998 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
1000 /* See if it works without O_NOATIME */
1001 switch (sha1_file_open_flag
) {
1003 fd
= open(filename
, O_RDONLY
);
1011 /* If it failed once, it will probably fail again.
1012 * Stop using O_NOATIME
1014 sha1_file_open_flag
= 0;
1016 *size
= xsize_t(st
.st_size
);
1017 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1022 static int legacy_loose_object(unsigned char *map
)
1027 * Is it a zlib-compressed buffer? If so, the first byte
1028 * must be 0x78 (15-bit window size, deflated), and the
1029 * first 16-bit word is evenly divisible by 31
1031 word
= (map
[0] << 8) + map
[1];
1032 if (map
[0] == 0x78 && !(word
% 31))
1038 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1043 unsigned long used
= 0;
1046 *type
= (c
>> 4) & 7;
1052 if (sizeof(long) * 8 <= shift
)
1055 size
+= (c
& 0x7f) << shift
;
1062 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1064 unsigned long size
, used
;
1065 static const char valid_loose_object_type
[8] = {
1067 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1068 0, /* "delta" and others are invalid in a loose object */
1070 enum object_type type
;
1072 /* Get the data stream */
1073 memset(stream
, 0, sizeof(*stream
));
1074 stream
->next_in
= map
;
1075 stream
->avail_in
= mapsize
;
1076 stream
->next_out
= buffer
;
1077 stream
->avail_out
= bufsiz
;
1079 if (legacy_loose_object(map
)) {
1080 inflateInit(stream
);
1081 return inflate(stream
, 0);
1086 * There used to be a second loose object header format which
1087 * was meant to mimic the in-pack format, allowing for direct
1088 * copy of the object data. This format turned up not to be
1089 * really worth it and we don't write it any longer. But we
1090 * can still read it.
1092 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1093 if (!used
|| !valid_loose_object_type
[type
])
1098 /* Set up the stream for the rest.. */
1099 stream
->next_in
= map
;
1100 stream
->avail_in
= mapsize
;
1101 inflateInit(stream
);
1103 /* And generate the fake traditional header */
1104 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1105 typename(type
), size
);
1109 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1111 int bytes
= strlen(buffer
) + 1;
1112 unsigned char *buf
= xmalloc(1+size
);
1116 n
= stream
->total_out
- bytes
;
1119 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1121 if (bytes
<= size
) {
1123 * The above condition must be (bytes <= size), not
1124 * (bytes < size). In other words, even though we
1125 * expect no more output and set avail_out to zer0,
1126 * the input zlib stream may have bytes that express
1127 * "this concludes the stream", and we *do* want to
1130 * Otherwise we would not be able to test that we
1131 * consumed all the input to reach the expected size;
1132 * we also want to check that zlib tells us that all
1133 * went well with status == Z_STREAM_END at the end.
1135 stream
->next_out
= buf
+ bytes
;
1136 stream
->avail_out
= size
- bytes
;
1137 while (status
== Z_OK
)
1138 status
= inflate(stream
, Z_FINISH
);
1141 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1147 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1148 else if (stream
->avail_in
)
1149 error("garbage at end of loose object '%s'",
1156 * We used to just use "sscanf()", but that's actually way
1157 * too permissive for what we want to check. So do an anal
1158 * object header parse by hand.
1160 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1167 * The type can be at most ten bytes (including the
1168 * terminating '\0' that we add), and is followed by
1177 if (i
>= sizeof(type
))
1183 * The length must follow immediately, and be in canonical
1184 * decimal format (ie "010" is not valid).
1186 size
= *hdr
++ - '0';
1191 unsigned long c
= *hdr
- '0';
1195 size
= size
* 10 + c
;
1201 * The length must be followed by a zero byte
1203 return *hdr
? -1 : type_from_string(type
);
1206 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1212 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1213 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1216 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1219 unsigned long get_size_from_delta(struct packed_git
*p
,
1220 struct pack_window
**w_curs
,
1223 const unsigned char *data
;
1224 unsigned char delta_head
[20], *in
;
1228 memset(&stream
, 0, sizeof(stream
));
1229 stream
.next_out
= delta_head
;
1230 stream
.avail_out
= sizeof(delta_head
);
1232 inflateInit(&stream
);
1234 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1235 stream
.next_in
= in
;
1236 st
= inflate(&stream
, Z_FINISH
);
1237 curpos
+= stream
.next_in
- in
;
1238 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1239 stream
.total_out
< sizeof(delta_head
));
1240 inflateEnd(&stream
);
1241 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1242 die("delta data unpack-initial failed");
1244 /* Examine the initial part of the delta to figure out
1249 /* ignore base size */
1250 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1252 /* Read the result size */
1253 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1256 static off_t
get_delta_base(struct packed_git
*p
,
1257 struct pack_window
**w_curs
,
1259 enum object_type type
,
1260 off_t delta_obj_offset
)
1262 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1265 /* use_pack() assured us we have [base_info, base_info + 20)
1266 * as a range that we can look at without walking off the
1267 * end of the mapped window. Its actually the hash size
1268 * that is assured. An OFS_DELTA longer than the hash size
1269 * is stupid, as then a REF_DELTA would be smaller to store.
1271 if (type
== OBJ_OFS_DELTA
) {
1273 unsigned char c
= base_info
[used
++];
1274 base_offset
= c
& 127;
1277 if (!base_offset
|| MSB(base_offset
, 7))
1278 die("offset value overflow for delta base object");
1279 c
= base_info
[used
++];
1280 base_offset
= (base_offset
<< 7) + (c
& 127);
1282 base_offset
= delta_obj_offset
- base_offset
;
1283 if (base_offset
>= delta_obj_offset
)
1284 die("delta base offset out of bound");
1286 } else if (type
== OBJ_REF_DELTA
) {
1287 /* The base entry _must_ be in the same pack */
1288 base_offset
= find_pack_entry_one(base_info
, p
);
1290 die("failed to find delta-pack base object %s",
1291 sha1_to_hex(base_info
));
1294 die("I am totally screwed");
1298 /* forward declaration for a mutually recursive function */
1299 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1300 unsigned long *sizep
);
1302 static int packed_delta_info(struct packed_git
*p
,
1303 struct pack_window
**w_curs
,
1305 enum object_type type
,
1307 unsigned long *sizep
)
1311 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1312 type
= packed_object_info(p
, base_offset
, NULL
);
1314 /* We choose to only get the type of the base object and
1315 * ignore potentially corrupt pack file that expects the delta
1316 * based on a base with a wrong size. This saves tons of
1320 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1325 static int unpack_object_header(struct packed_git
*p
,
1326 struct pack_window
**w_curs
,
1328 unsigned long *sizep
)
1330 unsigned char *base
;
1333 enum object_type type
;
1335 /* use_pack() assures us we have [base, base + 20) available
1336 * as a range that we can look at at. (Its actually the hash
1337 * size that is assured.) With our object header encoding
1338 * the maximum deflated object size is 2^137, which is just
1339 * insane, so we know won't exceed what we have been given.
1341 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1342 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1344 die("object offset outside of pack file");
1350 const char *packed_object_info_detail(struct packed_git
*p
,
1352 unsigned long *size
,
1353 unsigned long *store_size
,
1354 unsigned int *delta_chain_length
,
1355 unsigned char *base_sha1
)
1357 struct pack_window
*w_curs
= NULL
;
1359 unsigned long dummy
;
1360 unsigned char *next_sha1
;
1361 enum object_type type
;
1363 *delta_chain_length
= 0;
1364 curpos
= obj_offset
;
1365 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1370 die("pack %s contains unknown object type %d",
1371 p
->pack_name
, type
);
1376 *store_size
= 0; /* notyet */
1377 unuse_pack(&w_curs
);
1378 return typename(type
);
1380 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1381 if (*delta_chain_length
== 0) {
1382 /* TODO: find base_sha1 as pointed by curpos */
1387 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1388 if (*delta_chain_length
== 0)
1389 hashcpy(base_sha1
, next_sha1
);
1390 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1393 (*delta_chain_length
)++;
1394 curpos
= obj_offset
;
1395 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1399 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1400 unsigned long *sizep
)
1402 struct pack_window
*w_curs
= NULL
;
1404 off_t curpos
= obj_offset
;
1405 enum object_type type
;
1407 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1412 type
= packed_delta_info(p
, &w_curs
, curpos
,
1413 type
, obj_offset
, sizep
);
1423 die("pack %s contains unknown object type %d",
1424 p
->pack_name
, type
);
1426 unuse_pack(&w_curs
);
1430 static void *unpack_compressed_entry(struct packed_git
*p
,
1431 struct pack_window
**w_curs
,
1437 unsigned char *buffer
, *in
;
1439 buffer
= xmalloc(size
+ 1);
1441 memset(&stream
, 0, sizeof(stream
));
1442 stream
.next_out
= buffer
;
1443 stream
.avail_out
= size
;
1445 inflateInit(&stream
);
1447 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1448 stream
.next_in
= in
;
1449 st
= inflate(&stream
, Z_FINISH
);
1450 curpos
+= stream
.next_in
- in
;
1451 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1452 inflateEnd(&stream
);
1453 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1461 #define MAX_DELTA_CACHE (256)
1463 static size_t delta_base_cached
;
1465 static struct delta_base_cache_lru_list
{
1466 struct delta_base_cache_lru_list
*prev
;
1467 struct delta_base_cache_lru_list
*next
;
1468 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1470 static struct delta_base_cache_entry
{
1471 struct delta_base_cache_lru_list lru
;
1473 struct packed_git
*p
;
1476 enum object_type type
;
1477 } delta_base_cache
[MAX_DELTA_CACHE
];
1479 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1483 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1484 hash
+= (hash
>> 8) + (hash
>> 16);
1485 return hash
% MAX_DELTA_CACHE
;
1488 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1489 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1492 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1493 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1496 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1497 goto found_cache_entry
;
1498 return unpack_entry(p
, base_offset
, type
, base_size
);
1503 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1504 ent
->lru
.prev
->next
= ent
->lru
.next
;
1505 delta_base_cached
-= ent
->size
;
1507 ret
= xmemdupz(ent
->data
, ent
->size
);
1510 *base_size
= ent
->size
;
1514 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1519 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1520 ent
->lru
.prev
->next
= ent
->lru
.next
;
1521 delta_base_cached
-= ent
->size
;
1525 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1526 void *base
, unsigned long base_size
, enum object_type type
)
1528 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1529 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1530 struct delta_base_cache_lru_list
*lru
;
1532 release_delta_base_cache(ent
);
1533 delta_base_cached
+= base_size
;
1535 for (lru
= delta_base_cache_lru
.next
;
1536 delta_base_cached
> delta_base_cache_limit
1537 && lru
!= &delta_base_cache_lru
;
1539 struct delta_base_cache_entry
*f
= (void *)lru
;
1540 if (f
->type
== OBJ_BLOB
)
1541 release_delta_base_cache(f
);
1543 for (lru
= delta_base_cache_lru
.next
;
1544 delta_base_cached
> delta_base_cache_limit
1545 && lru
!= &delta_base_cache_lru
;
1547 struct delta_base_cache_entry
*f
= (void *)lru
;
1548 release_delta_base_cache(f
);
1552 ent
->base_offset
= base_offset
;
1555 ent
->size
= base_size
;
1556 ent
->lru
.next
= &delta_base_cache_lru
;
1557 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1558 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1559 delta_base_cache_lru
.prev
= &ent
->lru
;
1562 static void *unpack_delta_entry(struct packed_git
*p
,
1563 struct pack_window
**w_curs
,
1565 unsigned long delta_size
,
1567 enum object_type
*type
,
1568 unsigned long *sizep
)
1570 void *delta_data
, *result
, *base
;
1571 unsigned long base_size
;
1574 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1575 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1577 die("failed to read delta base object"
1578 " at %"PRIuMAX
" from %s",
1579 (uintmax_t)base_offset
, p
->pack_name
);
1581 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1583 die("failed to unpack compressed delta"
1584 " at %"PRIuMAX
" from %s",
1585 (uintmax_t)curpos
, p
->pack_name
);
1586 result
= patch_delta(base
, base_size
,
1587 delta_data
, delta_size
,
1590 die("failed to apply delta");
1592 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1596 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1597 enum object_type
*type
, unsigned long *sizep
)
1599 struct pack_window
*w_curs
= NULL
;
1600 off_t curpos
= obj_offset
;
1603 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1607 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1608 obj_offset
, type
, sizep
);
1614 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1617 die("unknown object type %i in %s", *type
, p
->pack_name
);
1619 unuse_pack(&w_curs
);
1623 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1626 const unsigned char *index
= p
->index_data
;
1628 if (open_pack_index(p
))
1630 index
= p
->index_data
;
1632 if (n
>= p
->num_objects
)
1635 if (p
->index_version
== 1) {
1636 return index
+ 24 * n
+ 4;
1639 return index
+ 20 * n
;
1643 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1645 const unsigned char *index
= p
->index_data
;
1647 if (p
->index_version
== 1) {
1648 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1651 index
+= 8 + p
->num_objects
* (20 + 4);
1652 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1653 if (!(off
& 0x80000000))
1655 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1656 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1657 ntohl(*((uint32_t *)(index
+ 4)));
1661 off_t
find_pack_entry_one(const unsigned char *sha1
,
1662 struct packed_git
*p
)
1664 const uint32_t *level1_ofs
= p
->index_data
;
1665 const unsigned char *index
= p
->index_data
;
1669 if (open_pack_index(p
))
1671 level1_ofs
= p
->index_data
;
1672 index
= p
->index_data
;
1674 if (p
->index_version
> 1) {
1679 hi
= ntohl(level1_ofs
[*sha1
]);
1680 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1683 unsigned mi
= (lo
+ hi
) / 2;
1684 unsigned x
= (p
->index_version
> 1) ? (mi
* 20) : (mi
* 24 + 4);
1685 int cmp
= hashcmp(index
+ x
, sha1
);
1687 return nth_packed_object_offset(p
, mi
);
1696 int matches_pack_name(struct packed_git
*p
, const char *name
)
1698 const char *last_c
, *c
;
1700 if (!strcmp(p
->pack_name
, name
))
1703 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1708 if (!strcmp(last_c
, name
))
1714 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1716 static struct packed_git
*last_found
= (void *)1;
1717 struct packed_git
*p
;
1720 prepare_packed_git();
1723 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1726 if (ignore_packed
) {
1728 for (ig
= ignore_packed
; *ig
; ig
++)
1729 if (matches_pack_name(p
, *ig
))
1735 offset
= find_pack_entry_one(sha1
, p
);
1738 * We are about to tell the caller where they can
1739 * locate the requested object. We better make
1740 * sure the packfile is still here and can be
1741 * accessed before supplying that answer, as
1742 * it may have been deleted since the index
1745 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1746 error("packfile %s cannot be accessed", p
->pack_name
);
1751 hashcpy(e
->sha1
, sha1
);
1757 if (p
== last_found
)
1761 if (p
== last_found
)
1767 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1768 struct packed_git
*packs
)
1770 struct packed_git
*p
;
1772 for (p
= packs
; p
; p
= p
->next
) {
1773 if (find_pack_entry_one(sha1
, p
))
1780 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1783 unsigned long mapsize
, size
;
1788 map
= map_sha1_file(sha1
, &mapsize
);
1790 return error("unable to find %s", sha1_to_hex(sha1
));
1791 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1792 status
= error("unable to unpack %s header",
1794 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1795 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1798 inflateEnd(&stream
);
1799 munmap(map
, mapsize
);
1803 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1805 struct pack_entry e
;
1807 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1808 reprepare_packed_git();
1809 if (!find_pack_entry(sha1
, &e
, NULL
))
1810 return sha1_loose_object_info(sha1
, sizep
);
1812 return packed_object_info(e
.p
, e
.offset
, sizep
);
1815 static void *read_packed_sha1(const unsigned char *sha1
,
1816 enum object_type
*type
, unsigned long *size
)
1818 struct pack_entry e
;
1820 if (!find_pack_entry(sha1
, &e
, NULL
))
1823 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1827 * This is meant to hold a *small* number of objects that you would
1828 * want read_sha1_file() to be able to return, but yet you do not want
1829 * to write them into the object store (e.g. a browse-only
1832 static struct cached_object
{
1833 unsigned char sha1
[20];
1834 enum object_type type
;
1838 static int cached_object_nr
, cached_object_alloc
;
1840 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1843 struct cached_object
*co
= cached_objects
;
1845 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1846 if (!hashcmp(co
->sha1
, sha1
))
1852 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1853 unsigned char *sha1
)
1855 struct cached_object
*co
;
1857 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1858 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1860 if (cached_object_alloc
<= cached_object_nr
) {
1861 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1862 cached_objects
= xrealloc(cached_objects
,
1863 sizeof(*cached_objects
) *
1864 cached_object_alloc
);
1866 co
= &cached_objects
[cached_object_nr
++];
1869 co
->buf
= xmalloc(len
);
1870 memcpy(co
->buf
, buf
, len
);
1871 hashcpy(co
->sha1
, sha1
);
1875 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1876 unsigned long *size
)
1878 unsigned long mapsize
;
1880 struct cached_object
*co
;
1882 co
= find_cached_object(sha1
);
1886 return xmemdupz(co
->buf
, co
->size
);
1889 buf
= read_packed_sha1(sha1
, type
, size
);
1892 map
= map_sha1_file(sha1
, &mapsize
);
1894 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1895 munmap(map
, mapsize
);
1898 reprepare_packed_git();
1899 return read_packed_sha1(sha1
, type
, size
);
1902 void *read_object_with_reference(const unsigned char *sha1
,
1903 const char *required_type_name
,
1904 unsigned long *size
,
1905 unsigned char *actual_sha1_return
)
1907 enum object_type type
, required_type
;
1909 unsigned long isize
;
1910 unsigned char actual_sha1
[20];
1912 required_type
= type_from_string(required_type_name
);
1913 hashcpy(actual_sha1
, sha1
);
1915 int ref_length
= -1;
1916 const char *ref_type
= NULL
;
1918 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1921 if (type
== required_type
) {
1923 if (actual_sha1_return
)
1924 hashcpy(actual_sha1_return
, actual_sha1
);
1927 /* Handle references */
1928 else if (type
== OBJ_COMMIT
)
1930 else if (type
== OBJ_TAG
)
1931 ref_type
= "object ";
1936 ref_length
= strlen(ref_type
);
1938 if (memcmp(buffer
, ref_type
, ref_length
) ||
1939 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1944 /* Now we have the ID of the referred-to object in
1945 * actual_sha1. Check again. */
1949 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
1950 const char *type
, unsigned char *sha1
,
1951 char *hdr
, int *hdrlen
)
1955 /* Generate the header */
1956 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
1960 SHA1_Update(&c
, hdr
, *hdrlen
);
1961 SHA1_Update(&c
, buf
, len
);
1962 SHA1_Final(sha1
, &c
);
1966 * Link the tempfile to the final place, possibly creating the
1967 * last directory level as you do so.
1969 * Returns the errno on failure, 0 on success.
1971 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
1976 if (!link(tmpfile
, filename
))
1980 * Try to mkdir the last path component if that failed.
1982 * Re-try the "link()" regardless of whether the mkdir
1983 * succeeds, since a race might mean that somebody
1987 dir
= strrchr(filename
, '/');
1990 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
1995 if (!link(tmpfile
, filename
))
2003 * Move the just written object into its final resting place
2005 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2007 int ret
= link_temp_to_file(tmpfile
, filename
);
2010 * Coda hack - coda doesn't like cross-directory links,
2011 * so we fall back to a rename, which will mean that it
2012 * won't be able to check collisions, but that's not a
2015 * The same holds for FAT formatted media.
2017 * When this succeeds, we just return 0. We have nothing
2020 if (ret
&& ret
!= EEXIST
) {
2021 if (!rename(tmpfile
, filename
))
2027 if (ret
!= EEXIST
) {
2028 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2030 /* FIXME!!! Collision check here ? */
2036 static int write_buffer(int fd
, const void *buf
, size_t len
)
2038 if (write_in_full(fd
, buf
, len
) < 0)
2039 return error("file write error (%s)", strerror(errno
));
2043 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2044 unsigned char *sha1
)
2048 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2052 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2055 unsigned char *compressed
;
2057 unsigned char sha1
[20];
2059 static char tmpfile
[PATH_MAX
];
2063 /* Normally if we have it in the pack then we do not bother writing
2064 * it out into .git/objects/??/?{38} file.
2066 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2067 filename
= sha1_file_name(sha1
);
2069 hashcpy(returnsha1
, sha1
);
2070 if (has_sha1_file(sha1
))
2072 fd
= open(filename
, O_RDONLY
);
2075 * FIXME!!! We might do collision checking here, but we'd
2076 * need to uncompress the old file and check it. Later.
2082 if (errno
!= ENOENT
) {
2083 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2086 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2088 fd
= mkstemp(tmpfile
);
2091 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2093 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2097 memset(&stream
, 0, sizeof(stream
));
2098 deflateInit(&stream
, zlib_compression_level
);
2099 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2100 compressed
= xmalloc(size
);
2103 stream
.next_out
= compressed
;
2104 stream
.avail_out
= size
;
2106 /* First header.. */
2107 stream
.next_in
= (unsigned char *)hdr
;
2108 stream
.avail_in
= hdrlen
;
2109 while (deflate(&stream
, 0) == Z_OK
)
2112 /* Then the data itself.. */
2113 stream
.next_in
= buf
;
2114 stream
.avail_in
= len
;
2115 ret
= deflate(&stream
, Z_FINISH
);
2116 if (ret
!= Z_STREAM_END
)
2117 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2119 ret
= deflateEnd(&stream
);
2121 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2123 size
= stream
.total_out
;
2125 if (write_buffer(fd
, compressed
, size
) < 0)
2126 die("unable to write sha1 file");
2129 die("unable to write sha1 file");
2132 return move_temp_to_file(tmpfile
, filename
);
2136 * We need to unpack and recompress the object for writing
2137 * it out to a different file.
2139 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2143 unsigned char *unpacked
;
2145 enum object_type type
;
2150 /* need to unpack and recompress it by itself */
2151 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2153 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2155 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2158 memset(&stream
, 0, sizeof(stream
));
2159 deflateInit(&stream
, zlib_compression_level
);
2160 size
= deflateBound(&stream
, len
+ hdrlen
);
2161 buf
= xmalloc(size
);
2164 stream
.next_out
= buf
;
2165 stream
.avail_out
= size
;
2167 /* First header.. */
2168 stream
.next_in
= (void *)hdr
;
2169 stream
.avail_in
= hdrlen
;
2170 while (deflate(&stream
, 0) == Z_OK
)
2173 /* Then the data itself.. */
2174 stream
.next_in
= unpacked
;
2175 stream
.avail_in
= len
;
2176 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2178 deflateEnd(&stream
);
2181 *objsize
= stream
.total_out
;
2185 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2188 unsigned long objsize
;
2189 void *buf
= map_sha1_file(sha1
, &objsize
);
2192 retval
= write_buffer(fd
, buf
, objsize
);
2193 munmap(buf
, objsize
);
2197 buf
= repack_object(sha1
, &objsize
);
2198 retval
= write_buffer(fd
, buf
, objsize
);
2203 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2204 size_t bufsize
, size_t *bufposn
)
2206 char tmpfile
[PATH_MAX
];
2209 unsigned char real_sha1
[20];
2210 unsigned char discard
[4096];
2214 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2216 local
= mkstemp(tmpfile
);
2219 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2221 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2224 memset(&stream
, 0, sizeof(stream
));
2226 inflateInit(&stream
);
2233 stream
.avail_in
= *bufposn
;
2234 stream
.next_in
= (unsigned char *) buffer
;
2236 stream
.next_out
= discard
;
2237 stream
.avail_out
= sizeof(discard
);
2238 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2239 SHA1_Update(&c
, discard
, sizeof(discard
) -
2241 } while (stream
.avail_in
&& ret
== Z_OK
);
2242 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2243 die("unable to write sha1 file");
2244 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2246 *bufposn
= stream
.avail_in
;
2250 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2255 return error("Connection closed?");
2256 perror("Reading from connection");
2261 inflateEnd(&stream
);
2263 fchmod(local
, 0444);
2264 if (close(local
) != 0)
2265 die("unable to write sha1 file");
2266 SHA1_Final(real_sha1
, &c
);
2267 if (ret
!= Z_STREAM_END
) {
2269 return error("File %s corrupted", sha1_to_hex(sha1
));
2271 if (hashcmp(sha1
, real_sha1
)) {
2273 return error("File %s has bad hash", sha1_to_hex(sha1
));
2276 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2279 int has_pack_index(const unsigned char *sha1
)
2282 if (stat(sha1_pack_index_name(sha1
), &st
))
2287 int has_pack_file(const unsigned char *sha1
)
2290 if (stat(sha1_pack_name(sha1
), &st
))
2295 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2297 struct pack_entry e
;
2298 return find_pack_entry(sha1
, &e
, ignore_packed
);
2301 int has_sha1_file(const unsigned char *sha1
)
2304 struct pack_entry e
;
2306 if (find_pack_entry(sha1
, &e
, NULL
))
2308 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2311 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2316 strbuf_init(&buf
, 0);
2317 if (strbuf_read(&buf
, fd
, 4096) < 0) {
2318 strbuf_release(&buf
);
2325 ret
= write_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2327 ret
= hash_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2328 strbuf_release(&buf
);
2333 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2334 enum object_type type
, const char *path
)
2336 size_t size
= xsize_t(st
->st_size
);
2338 int ret
, re_allocated
= 0;
2341 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2348 * Convert blobs to git internal format
2350 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2352 strbuf_init(&nbuf
, 0);
2353 if (convert_to_git(path
, buf
, size
, &nbuf
)) {
2355 buf
= strbuf_detach(&nbuf
, &size
);
2361 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2363 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2373 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2379 switch (st
->st_mode
& S_IFMT
) {
2381 fd
= open(path
, O_RDONLY
);
2383 return error("open(\"%s\"): %s", path
,
2385 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2386 return error("%s: failed to insert into database",
2390 len
= xsize_t(st
->st_size
);
2391 target
= xmalloc(len
+ 1);
2392 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2393 char *errstr
= strerror(errno
);
2395 return error("readlink(\"%s\"): %s", path
,
2399 hash_sha1_file(target
, len
, blob_type
, sha1
);
2400 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2401 return error("%s: failed to insert into database",
2406 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2408 return error("%s: unsupported file type", path
);
2413 int read_pack_header(int fd
, struct pack_header
*header
)
2415 char *c
= (char*)header
;
2416 ssize_t remaining
= sizeof(struct pack_header
);
2418 ssize_t r
= xread(fd
, c
, remaining
);
2420 /* "eof before pack header was fully read" */
2421 return PH_ERROR_EOF
;
2424 } while (remaining
> 0);
2425 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2426 /* "protocol error (pack signature mismatch detected)" */
2427 return PH_ERROR_PACK_SIGNATURE
;
2428 if (!pack_version_ok(header
->hdr_version
))
2429 /* "protocol error (pack version unsupported)" */
2430 return PH_ERROR_PROTOCOL
;