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
32 const unsigned char null_sha1
[20];
34 static unsigned int sha1_file_open_flag
= O_NOATIME
;
36 signed char hexval_table
[256] = {
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
43 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
44 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
45 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
74 for (i
= 0; i
< 20; i
++) {
75 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
84 int safe_create_leading_directories(char *path
)
93 pos
= strchr(pos
, '/');
97 if (!stat(path
, &st
)) {
99 if (!S_ISDIR(st
.st_mode
)) {
104 else if (mkdir(path
, 0777)) {
108 else if (adjust_shared_perm(path
)) {
117 char * sha1_to_hex(const unsigned char *sha1
)
120 static char hexbuffer
[4][50];
121 static const char hex
[] = "0123456789abcdef";
122 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
125 for (i
= 0; i
< 20; i
++) {
126 unsigned int val
= *sha1
++;
127 *buf
++ = hex
[val
>> 4];
128 *buf
++ = hex
[val
& 0xf];
135 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
138 for (i
= 0; i
< 20; i
++) {
139 static char hex
[] = "0123456789abcdef";
140 unsigned int val
= sha1
[i
];
141 char *pos
= pathbuf
+ i
*2 + (i
> 0);
142 *pos
++ = hex
[val
>> 4];
143 *pos
= hex
[val
& 0xf];
148 * NOTE! This returns a statically allocated buffer, so you have to be
149 * careful about using it. Do a "xstrdup()" if you need to save the
152 * Also note that this returns the location for creating. Reading
153 * SHA1 file can happen from any alternate directory listed in the
154 * DB_ENVIRONMENT environment variable if it is not found in
155 * the primary object database.
157 char *sha1_file_name(const unsigned char *sha1
)
159 static char *name
, *base
;
162 const char *sha1_file_directory
= get_object_directory();
163 int len
= strlen(sha1_file_directory
);
164 base
= xmalloc(len
+ 60);
165 memcpy(base
, sha1_file_directory
, len
);
166 memset(base
+len
, 0, 60);
169 name
= base
+ len
+ 1;
171 fill_sha1_path(name
, sha1
);
175 char *sha1_pack_name(const unsigned char *sha1
)
177 static const char hex
[] = "0123456789abcdef";
178 static char *name
, *base
, *buf
;
182 const char *sha1_file_directory
= get_object_directory();
183 int len
= strlen(sha1_file_directory
);
184 base
= xmalloc(len
+ 60);
185 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory
);
186 name
= base
+ len
+ 11;
191 for (i
= 0; i
< 20; i
++) {
192 unsigned int val
= *sha1
++;
193 *buf
++ = hex
[val
>> 4];
194 *buf
++ = hex
[val
& 0xf];
200 char *sha1_pack_index_name(const unsigned char *sha1
)
202 static const char hex
[] = "0123456789abcdef";
203 static char *name
, *base
, *buf
;
207 const char *sha1_file_directory
= get_object_directory();
208 int len
= strlen(sha1_file_directory
);
209 base
= xmalloc(len
+ 60);
210 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory
);
211 name
= base
+ len
+ 11;
216 for (i
= 0; i
< 20; i
++) {
217 unsigned int val
= *sha1
++;
218 *buf
++ = hex
[val
>> 4];
219 *buf
++ = hex
[val
& 0xf];
225 struct alternate_object_database
*alt_odb_list
;
226 static struct alternate_object_database
**alt_odb_tail
;
228 static void read_info_alternates(const char * alternates
, int depth
);
231 * Prepare alternate object database registry.
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
245 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
248 const char *objdir
= get_object_directory();
249 struct alternate_object_database
*ent
;
250 struct alternate_object_database
*alt
;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
253 int entlen
= pfxlen
+ 43;
256 if (*entry
!= '/' && relative_base
) {
257 /* Relative alt-odb */
259 base_len
= strlen(relative_base
) + 1;
263 ent
= xmalloc(sizeof(*ent
) + entlen
);
265 if (*entry
!= '/' && relative_base
) {
266 memcpy(ent
->base
, relative_base
, base_len
- 1);
267 ent
->base
[base_len
- 1] = '/';
268 memcpy(ent
->base
+ base_len
, entry
, len
);
271 memcpy(ent
->base
, entry
, pfxlen
);
273 ent
->name
= ent
->base
+ pfxlen
+ 1;
274 ent
->base
[pfxlen
+ 3] = '/';
275 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
277 /* Detect cases where alternate disappeared */
278 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
289 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
290 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
295 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
300 /* add the alternate entry */
302 alt_odb_tail
= &(ent
->next
);
305 /* recursively add alternates */
306 read_info_alternates(ent
->base
, depth
+ 1);
308 ent
->base
[pfxlen
] = '/';
313 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
314 const char *relative_base
, int depth
)
316 const char *cp
, *last
;
319 error("%s: ignoring alternate object stores, nesting too deep.",
327 if (cp
< ep
&& *cp
== '#') {
328 while (cp
< ep
&& *cp
!= sep
)
333 while (cp
< ep
&& *cp
!= sep
)
336 if ((*last
!= '/') && depth
) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base
, last
);
340 link_alt_odb_entry(last
, cp
- last
,
341 relative_base
, depth
);
344 while (cp
< ep
&& *cp
== sep
)
350 static void read_info_alternates(const char * relative_base
, int depth
)
358 sprintf(path
, "%s/info/alternates", relative_base
);
359 fd
= open(path
, O_RDONLY
);
362 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
366 mapsz
= xsize_t(st
.st_size
);
367 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
370 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
375 void prepare_alt_odb(void)
379 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
384 alt_odb_tail
= &alt_odb_list
;
385 link_alt_odb_entries(alt
, alt
+ strlen(alt
), ':', NULL
, 0);
387 read_info_alternates(get_object_directory(), 0);
390 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
392 char *name
= sha1_file_name(sha1
);
393 struct alternate_object_database
*alt
;
398 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
400 fill_sha1_path(name
, sha1
);
401 if (!stat(alt
->base
, st
))
407 static unsigned int pack_used_ctr
;
408 static unsigned int pack_mmap_calls
;
409 static unsigned int peak_pack_open_windows
;
410 static unsigned int pack_open_windows
;
411 static size_t peak_pack_mapped
;
412 static size_t pack_mapped
;
413 struct packed_git
*packed_git
;
418 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
419 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
420 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
421 (size_t) getpagesize(),
422 packed_git_window_size
,
425 "pack_report: pack_used_ctr = %10u\n"
426 "pack_report: pack_mmap_calls = %10u\n"
427 "pack_report: pack_open_windows = %10u / %10u\n"
428 "pack_report: pack_mapped = "
429 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
432 pack_open_windows
, peak_pack_open_windows
,
433 pack_mapped
, peak_pack_mapped
);
436 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
439 struct pack_idx_header
*hdr
;
441 uint32_t version
, nr
, i
, *index
;
442 int fd
= open(path
, O_RDONLY
);
447 if (fstat(fd
, &st
)) {
451 idx_size
= xsize_t(st
.st_size
);
452 if (idx_size
< 4 * 256 + 20 + 20) {
454 return error("index file %s is too small", path
);
456 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
460 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
461 version
= ntohl(hdr
->idx_version
);
462 if (version
< 2 || version
> 2) {
463 munmap(idx_map
, idx_size
);
464 return error("index file %s is version %d"
465 " and is not supported by this binary"
466 " (try upgrading GIT to a newer version)",
475 index
+= 2; /* skip index header */
476 for (i
= 0; i
< 256; i
++) {
477 uint32_t n
= ntohl(index
[i
]);
479 munmap(idx_map
, idx_size
);
480 return error("non-monotonic index %s", path
);
488 * - 256 index entries 4 bytes each
489 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
490 * - 20-byte SHA1 of the packfile
491 * - 20-byte SHA1 file checksum
493 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
494 munmap(idx_map
, idx_size
);
495 return error("wrong index file size in %s", path
);
497 } else if (version
== 2) {
500 * - 8 bytes of header
501 * - 256 index entries 4 bytes each
502 * - 20-byte sha1 entry * nr
503 * - 4-byte crc entry * nr
504 * - 4-byte offset entry * nr
505 * - 20-byte SHA1 of the packfile
506 * - 20-byte SHA1 file checksum
507 * And after the 4-byte offset table might be a
508 * variable sized table containing 8-byte entries
509 * for offsets larger than 2^31.
511 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
512 if (idx_size
< min_size
|| idx_size
> min_size
+ (nr
- 1)*8) {
513 munmap(idx_map
, idx_size
);
514 return error("wrong index file size in %s", path
);
516 if (idx_size
!= min_size
) {
517 /* make sure we can deal with large pack offsets */
518 off_t x
= 0x7fffffffUL
, y
= 0xffffffffUL
;
519 if (x
> (x
+ 1) || y
> (y
+ 1)) {
520 munmap(idx_map
, idx_size
);
521 return error("pack too large for current definition of off_t in %s", path
);
526 p
->index_version
= version
;
527 p
->index_data
= idx_map
;
528 p
->index_size
= idx_size
;
533 int open_pack_index (struct packed_git
*p
)
541 idx_name
= xstrdup(p
->pack_name
);
542 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
543 ret
= check_packed_git_idx(idx_name
, p
);
548 static void scan_windows(struct packed_git
*p
,
549 struct packed_git
**lru_p
,
550 struct pack_window
**lru_w
,
551 struct pack_window
**lru_l
)
553 struct pack_window
*w
, *w_l
;
555 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
557 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
567 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
569 struct packed_git
*p
, *lru_p
= NULL
;
570 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
573 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
574 for (p
= packed_git
; p
; p
= p
->next
)
575 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
577 munmap(lru_w
->base
, lru_w
->len
);
578 pack_mapped
-= lru_w
->len
;
580 lru_l
->next
= lru_w
->next
;
582 lru_p
->windows
= lru_w
->next
;
583 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
584 close(lru_p
->pack_fd
);
595 void release_pack_memory(size_t need
, int fd
)
597 size_t cur
= pack_mapped
;
598 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
602 void unuse_pack(struct pack_window
**w_cursor
)
604 struct pack_window
*w
= *w_cursor
;
612 * Do not call this directly as this leaks p->pack_fd on error return;
613 * call open_packed_git() instead.
615 static int open_packed_git_1(struct packed_git
*p
)
618 struct pack_header hdr
;
619 unsigned char sha1
[20];
620 unsigned char *idx_sha1
;
623 if (!p
->index_data
&& open_pack_index(p
))
624 return error("packfile %s index unavailable", p
->pack_name
);
626 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
627 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
630 /* If we created the struct before we had the pack we lack size. */
632 if (!S_ISREG(st
.st_mode
))
633 return error("packfile %s not a regular file", p
->pack_name
);
634 p
->pack_size
= st
.st_size
;
635 } else if (p
->pack_size
!= st
.st_size
)
636 return error("packfile %s size changed", p
->pack_name
);
638 /* We leave these file descriptors open with sliding mmap;
639 * there is no point keeping them open across exec(), though.
641 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
643 return error("cannot determine file descriptor flags");
644 fd_flag
|= FD_CLOEXEC
;
645 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
646 return error("cannot set FD_CLOEXEC");
648 /* Verify we recognize this pack file format. */
649 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
650 return error("file %s is far too short to be a packfile", p
->pack_name
);
651 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
652 return error("file %s is not a GIT packfile", p
->pack_name
);
653 if (!pack_version_ok(hdr
.hdr_version
))
654 return error("packfile %s is version %u and not supported"
655 " (try upgrading GIT to a newer version)",
656 p
->pack_name
, ntohl(hdr
.hdr_version
));
658 /* Verify the pack matches its index. */
659 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
660 return error("packfile %s claims to have %u objects"
661 " while index indicates %u objects",
662 p
->pack_name
, ntohl(hdr
.hdr_entries
),
664 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
665 return error("end of packfile %s is unavailable", p
->pack_name
);
666 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
667 return error("packfile %s signature is unavailable", p
->pack_name
);
668 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
669 if (hashcmp(sha1
, idx_sha1
))
670 return error("packfile %s does not match index", p
->pack_name
);
674 static int open_packed_git(struct packed_git
*p
)
676 if (!open_packed_git_1(p
))
678 if (p
->pack_fd
!= -1) {
685 static int in_window(struct pack_window
*win
, off_t offset
)
687 /* We must promise at least 20 bytes (one hash) after the
688 * offset is available from this window, otherwise the offset
689 * is not actually in this window and a different window (which
690 * has that one hash excess) must be used. This is to support
691 * the object header and delta base parsing routines below.
693 off_t win_off
= win
->offset
;
694 return win_off
<= offset
695 && (offset
+ 20) <= (win_off
+ win
->len
);
698 unsigned char* use_pack(struct packed_git
*p
,
699 struct pack_window
**w_cursor
,
703 struct pack_window
*win
= *w_cursor
;
705 if (p
->pack_fd
== -1 && open_packed_git(p
))
706 die("packfile %s cannot be accessed", p
->pack_name
);
708 /* Since packfiles end in a hash of their content and its
709 * pointless to ask for an offset into the middle of that
710 * hash, and the in_window function above wouldn't match
711 * don't allow an offset too close to the end of the file.
713 if (offset
> (p
->pack_size
- 20))
714 die("offset beyond end of packfile (truncated pack?)");
716 if (!win
|| !in_window(win
, offset
)) {
719 for (win
= p
->windows
; win
; win
= win
->next
) {
720 if (in_window(win
, offset
))
724 size_t window_align
= packed_git_window_size
/ 2;
726 win
= xcalloc(1, sizeof(*win
));
727 win
->offset
= (offset
/ window_align
) * window_align
;
728 len
= p
->pack_size
- win
->offset
;
729 if (len
> packed_git_window_size
)
730 len
= packed_git_window_size
;
731 win
->len
= (size_t)len
;
732 pack_mapped
+= win
->len
;
733 while (packed_git_limit
< pack_mapped
734 && unuse_one_window(p
, p
->pack_fd
))
736 win
->base
= xmmap(NULL
, win
->len
,
737 PROT_READ
, MAP_PRIVATE
,
738 p
->pack_fd
, win
->offset
);
739 if (win
->base
== MAP_FAILED
)
740 die("packfile %s cannot be mapped: %s",
745 if (pack_mapped
> peak_pack_mapped
)
746 peak_pack_mapped
= pack_mapped
;
747 if (pack_open_windows
> peak_pack_open_windows
)
748 peak_pack_open_windows
= pack_open_windows
;
749 win
->next
= p
->windows
;
753 if (win
!= *w_cursor
) {
754 win
->last_used
= pack_used_ctr
++;
758 offset
-= win
->offset
;
760 *left
= win
->len
- xsize_t(offset
);
761 return win
->base
+ offset
;
764 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
767 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
770 * Make sure a corresponding .pack file exists and that
771 * the index looks sane.
773 path_len
-= strlen(".idx");
776 memcpy(p
->pack_name
, path
, path_len
);
777 strcpy(p
->pack_name
+ path_len
, ".pack");
778 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
783 /* ok, it looks sane as far as we can check without
784 * actually mapping the pack file.
786 p
->index_version
= 0;
787 p
->index_data
= NULL
;
790 p
->pack_size
= st
.st_size
;
794 p
->pack_local
= local
;
795 p
->mtime
= st
.st_mtime
;
796 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
801 struct packed_git
*parse_pack_index(unsigned char *sha1
)
803 char *path
= sha1_pack_index_name(sha1
);
804 return parse_pack_index_file(sha1
, path
);
807 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
808 const char *idx_path
)
810 const char *path
= sha1_pack_name(sha1
);
811 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
813 if (check_packed_git_idx(idx_path
, p
)) {
818 strcpy(p
->pack_name
, path
);
823 hashcpy(p
->sha1
, sha1
);
827 void install_packed_git(struct packed_git
*pack
)
829 pack
->next
= packed_git
;
833 static void prepare_packed_git_one(char *objdir
, int local
)
840 sprintf(path
, "%s/pack", objdir
);
845 error("unable to open object pack directory: %s: %s",
846 path
, strerror(errno
));
850 while ((de
= readdir(dir
)) != NULL
) {
851 int namelen
= strlen(de
->d_name
);
852 struct packed_git
*p
;
854 if (!has_extension(de
->d_name
, ".idx"))
857 /* Don't reopen a pack we already have. */
858 strcpy(path
+ len
, de
->d_name
);
859 for (p
= packed_git
; p
; p
= p
->next
) {
860 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
865 /* See if it really is a valid .idx file with corresponding
866 * .pack file that we can map.
868 p
= add_packed_git(path
, len
+ namelen
, local
);
871 install_packed_git(p
);
876 static int sort_pack(const void *a_
, const void *b_
)
878 struct packed_git
*a
= *((struct packed_git
**)a_
);
879 struct packed_git
*b
= *((struct packed_git
**)b_
);
883 * Local packs tend to contain objects specific to our
884 * variant of the project than remote ones. In addition,
885 * remote ones could be on a network mounted filesystem.
886 * Favor local ones for these reasons.
888 st
= a
->pack_local
- b
->pack_local
;
893 * Younger packs tend to contain more recent objects,
894 * and more recent objects tend to get accessed more
897 if (a
->mtime
< b
->mtime
)
899 else if (a
->mtime
== b
->mtime
)
904 static void rearrange_packed_git(void)
906 struct packed_git
**ary
, *p
;
909 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
914 /* prepare an array of packed_git for easier sorting */
915 ary
= xcalloc(n
, sizeof(struct packed_git
*));
916 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
919 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
921 /* link them back again */
922 for (i
= 0; i
< n
- 1; i
++)
923 ary
[i
]->next
= ary
[i
+ 1];
924 ary
[n
- 1]->next
= NULL
;
930 static int prepare_packed_git_run_once
= 0;
931 void prepare_packed_git(void)
933 struct alternate_object_database
*alt
;
935 if (prepare_packed_git_run_once
)
937 prepare_packed_git_one(get_object_directory(), 1);
939 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
941 prepare_packed_git_one(alt
->base
, 0);
944 rearrange_packed_git();
945 prepare_packed_git_run_once
= 1;
948 void reprepare_packed_git(void)
950 prepare_packed_git_run_once
= 0;
951 prepare_packed_git();
954 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
956 unsigned char real_sha1
[20];
957 hash_sha1_file(map
, size
, type
, real_sha1
);
958 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
961 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
966 char *filename
= find_sha1_file(sha1
, &st
);
972 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
974 /* See if it works without O_NOATIME */
975 switch (sha1_file_open_flag
) {
977 fd
= open(filename
, O_RDONLY
);
985 /* If it failed once, it will probably fail again.
986 * Stop using O_NOATIME
988 sha1_file_open_flag
= 0;
990 *size
= xsize_t(st
.st_size
);
991 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
996 static int legacy_loose_object(unsigned char *map
)
1001 * Is it a zlib-compressed buffer? If so, the first byte
1002 * must be 0x78 (15-bit window size, deflated), and the
1003 * first 16-bit word is evenly divisible by 31
1005 word
= (map
[0] << 8) + map
[1];
1006 if (map
[0] == 0x78 && !(word
% 31))
1012 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1017 unsigned long used
= 0;
1020 *type
= (c
>> 4) & 7;
1026 if (sizeof(long) * 8 <= shift
)
1029 size
+= (c
& 0x7f) << shift
;
1036 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1038 unsigned long size
, used
;
1039 static const char valid_loose_object_type
[8] = {
1041 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1042 0, /* "delta" and others are invalid in a loose object */
1044 enum object_type type
;
1046 /* Get the data stream */
1047 memset(stream
, 0, sizeof(*stream
));
1048 stream
->next_in
= map
;
1049 stream
->avail_in
= mapsize
;
1050 stream
->next_out
= buffer
;
1051 stream
->avail_out
= bufsiz
;
1053 if (legacy_loose_object(map
)) {
1054 inflateInit(stream
);
1055 return inflate(stream
, 0);
1060 * There used to be a second loose object header format which
1061 * was meant to mimic the in-pack format, allowing for direct
1062 * copy of the object data. This format turned up not to be
1063 * really worth it and we don't write it any longer. But we
1064 * can still read it.
1066 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1067 if (!used
|| !valid_loose_object_type
[type
])
1072 /* Set up the stream for the rest.. */
1073 stream
->next_in
= map
;
1074 stream
->avail_in
= mapsize
;
1075 inflateInit(stream
);
1077 /* And generate the fake traditional header */
1078 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1079 typename(type
), size
);
1083 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1085 int bytes
= strlen(buffer
) + 1;
1086 unsigned char *buf
= xmalloc(1+size
);
1090 n
= stream
->total_out
- bytes
;
1093 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1095 if (bytes
<= size
) {
1097 * The above condition must be (bytes <= size), not
1098 * (bytes < size). In other words, even though we
1099 * expect no more output and set avail_out to zer0,
1100 * the input zlib stream may have bytes that express
1101 * "this concludes the stream", and we *do* want to
1104 * Otherwise we would not be able to test that we
1105 * consumed all the input to reach the expected size;
1106 * we also want to check that zlib tells us that all
1107 * went well with status == Z_STREAM_END at the end.
1109 stream
->next_out
= buf
+ bytes
;
1110 stream
->avail_out
= size
- bytes
;
1111 while (status
== Z_OK
)
1112 status
= inflate(stream
, Z_FINISH
);
1115 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1121 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1122 else if (stream
->avail_in
)
1123 error("garbage at end of loose object '%s'",
1130 * We used to just use "sscanf()", but that's actually way
1131 * too permissive for what we want to check. So do an anal
1132 * object header parse by hand.
1134 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1141 * The type can be at most ten bytes (including the
1142 * terminating '\0' that we add), and is followed by
1151 if (i
>= sizeof(type
))
1157 * The length must follow immediately, and be in canonical
1158 * decimal format (ie "010" is not valid).
1160 size
= *hdr
++ - '0';
1165 unsigned long c
= *hdr
- '0';
1169 size
= size
* 10 + c
;
1175 * The length must be followed by a zero byte
1177 return *hdr
? -1 : type_from_string(type
);
1180 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1186 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1187 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1190 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1193 unsigned long get_size_from_delta(struct packed_git
*p
,
1194 struct pack_window
**w_curs
,
1197 const unsigned char *data
;
1198 unsigned char delta_head
[20], *in
;
1202 memset(&stream
, 0, sizeof(stream
));
1203 stream
.next_out
= delta_head
;
1204 stream
.avail_out
= sizeof(delta_head
);
1206 inflateInit(&stream
);
1208 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1209 stream
.next_in
= in
;
1210 st
= inflate(&stream
, Z_FINISH
);
1211 curpos
+= stream
.next_in
- in
;
1212 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1213 stream
.total_out
< sizeof(delta_head
));
1214 inflateEnd(&stream
);
1215 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1216 die("delta data unpack-initial failed");
1218 /* Examine the initial part of the delta to figure out
1223 /* ignore base size */
1224 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1226 /* Read the result size */
1227 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1230 static off_t
get_delta_base(struct packed_git
*p
,
1231 struct pack_window
**w_curs
,
1233 enum object_type type
,
1234 off_t delta_obj_offset
)
1236 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1239 /* use_pack() assured us we have [base_info, base_info + 20)
1240 * as a range that we can look at without walking off the
1241 * end of the mapped window. Its actually the hash size
1242 * that is assured. An OFS_DELTA longer than the hash size
1243 * is stupid, as then a REF_DELTA would be smaller to store.
1245 if (type
== OBJ_OFS_DELTA
) {
1247 unsigned char c
= base_info
[used
++];
1248 base_offset
= c
& 127;
1251 if (!base_offset
|| MSB(base_offset
, 7))
1252 die("offset value overflow for delta base object");
1253 c
= base_info
[used
++];
1254 base_offset
= (base_offset
<< 7) + (c
& 127);
1256 base_offset
= delta_obj_offset
- base_offset
;
1257 if (base_offset
>= delta_obj_offset
)
1258 die("delta base offset out of bound");
1260 } else if (type
== OBJ_REF_DELTA
) {
1261 /* The base entry _must_ be in the same pack */
1262 base_offset
= find_pack_entry_one(base_info
, p
);
1264 die("failed to find delta-pack base object %s",
1265 sha1_to_hex(base_info
));
1268 die("I am totally screwed");
1272 /* forward declaration for a mutually recursive function */
1273 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1274 unsigned long *sizep
);
1276 static int packed_delta_info(struct packed_git
*p
,
1277 struct pack_window
**w_curs
,
1279 enum object_type type
,
1281 unsigned long *sizep
)
1285 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1286 type
= packed_object_info(p
, base_offset
, NULL
);
1288 /* We choose to only get the type of the base object and
1289 * ignore potentially corrupt pack file that expects the delta
1290 * based on a base with a wrong size. This saves tons of
1294 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1299 static int unpack_object_header(struct packed_git
*p
,
1300 struct pack_window
**w_curs
,
1302 unsigned long *sizep
)
1304 unsigned char *base
;
1307 enum object_type type
;
1309 /* use_pack() assures us we have [base, base + 20) available
1310 * as a range that we can look at at. (Its actually the hash
1311 * size that is assured.) With our object header encoding
1312 * the maximum deflated object size is 2^137, which is just
1313 * insane, so we know won't exceed what we have been given.
1315 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1316 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1318 die("object offset outside of pack file");
1324 const char *packed_object_info_detail(struct packed_git
*p
,
1326 unsigned long *size
,
1327 unsigned long *store_size
,
1328 unsigned int *delta_chain_length
,
1329 unsigned char *base_sha1
)
1331 struct pack_window
*w_curs
= NULL
;
1333 unsigned long dummy
;
1334 unsigned char *next_sha1
;
1335 enum object_type type
;
1337 *delta_chain_length
= 0;
1338 curpos
= obj_offset
;
1339 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1344 die("pack %s contains unknown object type %d",
1345 p
->pack_name
, type
);
1350 *store_size
= 0; /* notyet */
1351 unuse_pack(&w_curs
);
1352 return typename(type
);
1354 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1355 if (*delta_chain_length
== 0) {
1356 /* TODO: find base_sha1 as pointed by curpos */
1361 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1362 if (*delta_chain_length
== 0)
1363 hashcpy(base_sha1
, next_sha1
);
1364 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1367 (*delta_chain_length
)++;
1368 curpos
= obj_offset
;
1369 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1373 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1374 unsigned long *sizep
)
1376 struct pack_window
*w_curs
= NULL
;
1378 off_t curpos
= obj_offset
;
1379 enum object_type type
;
1381 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1386 type
= packed_delta_info(p
, &w_curs
, curpos
,
1387 type
, obj_offset
, sizep
);
1397 die("pack %s contains unknown object type %d",
1398 p
->pack_name
, type
);
1400 unuse_pack(&w_curs
);
1404 static void *unpack_compressed_entry(struct packed_git
*p
,
1405 struct pack_window
**w_curs
,
1411 unsigned char *buffer
, *in
;
1413 buffer
= xmalloc(size
+ 1);
1415 memset(&stream
, 0, sizeof(stream
));
1416 stream
.next_out
= buffer
;
1417 stream
.avail_out
= size
;
1419 inflateInit(&stream
);
1421 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1422 stream
.next_in
= in
;
1423 st
= inflate(&stream
, Z_FINISH
);
1424 curpos
+= stream
.next_in
- in
;
1425 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1426 inflateEnd(&stream
);
1427 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1435 #define MAX_DELTA_CACHE (256)
1437 static size_t delta_base_cached
;
1439 static struct delta_base_cache_lru_list
{
1440 struct delta_base_cache_lru_list
*prev
;
1441 struct delta_base_cache_lru_list
*next
;
1442 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1444 static struct delta_base_cache_entry
{
1445 struct delta_base_cache_lru_list lru
;
1447 struct packed_git
*p
;
1450 enum object_type type
;
1451 } delta_base_cache
[MAX_DELTA_CACHE
];
1453 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1457 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1458 hash
+= (hash
>> 8) + (hash
>> 16);
1459 return hash
% MAX_DELTA_CACHE
;
1462 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1463 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1466 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1467 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1470 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1471 goto found_cache_entry
;
1472 return unpack_entry(p
, base_offset
, type
, base_size
);
1477 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1478 ent
->lru
.prev
->next
= ent
->lru
.next
;
1479 delta_base_cached
-= ent
->size
;
1482 ret
= xmalloc(ent
->size
+ 1);
1483 memcpy(ret
, ent
->data
, ent
->size
);
1484 ((char *)ret
)[ent
->size
] = 0;
1487 *base_size
= ent
->size
;
1491 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1496 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1497 ent
->lru
.prev
->next
= ent
->lru
.next
;
1498 delta_base_cached
-= ent
->size
;
1502 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1503 void *base
, unsigned long base_size
, enum object_type type
)
1505 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1506 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1507 struct delta_base_cache_lru_list
*lru
;
1509 release_delta_base_cache(ent
);
1510 delta_base_cached
+= base_size
;
1512 for (lru
= delta_base_cache_lru
.next
;
1513 delta_base_cached
> delta_base_cache_limit
1514 && lru
!= &delta_base_cache_lru
;
1516 struct delta_base_cache_entry
*f
= (void *)lru
;
1517 if (f
->type
== OBJ_BLOB
)
1518 release_delta_base_cache(f
);
1520 for (lru
= delta_base_cache_lru
.next
;
1521 delta_base_cached
> delta_base_cache_limit
1522 && lru
!= &delta_base_cache_lru
;
1524 struct delta_base_cache_entry
*f
= (void *)lru
;
1525 release_delta_base_cache(f
);
1529 ent
->base_offset
= base_offset
;
1532 ent
->size
= base_size
;
1533 ent
->lru
.next
= &delta_base_cache_lru
;
1534 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1535 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1536 delta_base_cache_lru
.prev
= &ent
->lru
;
1539 static void *unpack_delta_entry(struct packed_git
*p
,
1540 struct pack_window
**w_curs
,
1542 unsigned long delta_size
,
1544 enum object_type
*type
,
1545 unsigned long *sizep
)
1547 void *delta_data
, *result
, *base
;
1548 unsigned long base_size
;
1551 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1552 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1554 die("failed to read delta base object"
1555 " at %"PRIuMAX
" from %s",
1556 (uintmax_t)base_offset
, p
->pack_name
);
1558 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1559 result
= patch_delta(base
, base_size
,
1560 delta_data
, delta_size
,
1563 die("failed to apply delta");
1565 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1569 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1570 enum object_type
*type
, unsigned long *sizep
)
1572 struct pack_window
*w_curs
= NULL
;
1573 off_t curpos
= obj_offset
;
1576 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1580 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1581 obj_offset
, type
, sizep
);
1587 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1590 die("unknown object type %i in %s", *type
, p
->pack_name
);
1592 unuse_pack(&w_curs
);
1596 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1599 const unsigned char *index
= p
->index_data
;
1601 if (open_pack_index(p
))
1603 index
= p
->index_data
;
1605 if (n
>= p
->num_objects
)
1608 if (p
->index_version
== 1) {
1609 return index
+ 24 * n
+ 4;
1612 return index
+ 20 * n
;
1616 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1618 const unsigned char *index
= p
->index_data
;
1620 if (p
->index_version
== 1) {
1621 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1624 index
+= 8 + p
->num_objects
* (20 + 4);
1625 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1626 if (!(off
& 0x80000000))
1628 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1629 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1630 ntohl(*((uint32_t *)(index
+ 4)));
1634 off_t
find_pack_entry_one(const unsigned char *sha1
,
1635 struct packed_git
*p
)
1637 const uint32_t *level1_ofs
= p
->index_data
;
1638 const unsigned char *index
= p
->index_data
;
1642 if (open_pack_index(p
))
1644 level1_ofs
= p
->index_data
;
1645 index
= p
->index_data
;
1647 if (p
->index_version
> 1) {
1652 hi
= ntohl(level1_ofs
[*sha1
]);
1653 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1656 unsigned mi
= (lo
+ hi
) / 2;
1657 unsigned x
= (p
->index_version
> 1) ? (mi
* 20) : (mi
* 24 + 4);
1658 int cmp
= hashcmp(index
+ x
, sha1
);
1660 return nth_packed_object_offset(p
, mi
);
1669 static int matches_pack_name(struct packed_git
*p
, const char *ig
)
1671 const char *last_c
, *c
;
1673 if (!strcmp(p
->pack_name
, ig
))
1676 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1681 if (!strcmp(last_c
, ig
))
1687 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1689 struct packed_git
*p
;
1692 prepare_packed_git();
1694 for (p
= packed_git
; p
; p
= p
->next
) {
1695 if (ignore_packed
) {
1697 for (ig
= ignore_packed
; *ig
; ig
++)
1698 if (!matches_pack_name(p
, *ig
))
1703 offset
= find_pack_entry_one(sha1
, p
);
1706 * We are about to tell the caller where they can
1707 * locate the requested object. We better make
1708 * sure the packfile is still here and can be
1709 * accessed before supplying that answer, as
1710 * it may have been deleted since the index
1713 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1714 error("packfile %s cannot be accessed", p
->pack_name
);
1719 hashcpy(e
->sha1
, sha1
);
1726 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1727 struct packed_git
*packs
)
1729 struct packed_git
*p
;
1731 for (p
= packs
; p
; p
= p
->next
) {
1732 if (find_pack_entry_one(sha1
, p
))
1739 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1742 unsigned long mapsize
, size
;
1747 map
= map_sha1_file(sha1
, &mapsize
);
1749 return error("unable to find %s", sha1_to_hex(sha1
));
1750 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1751 status
= error("unable to unpack %s header",
1753 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1754 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1757 inflateEnd(&stream
);
1758 munmap(map
, mapsize
);
1762 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1764 struct pack_entry e
;
1766 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1767 reprepare_packed_git();
1768 if (!find_pack_entry(sha1
, &e
, NULL
))
1769 return sha1_loose_object_info(sha1
, sizep
);
1771 return packed_object_info(e
.p
, e
.offset
, sizep
);
1774 static void *read_packed_sha1(const unsigned char *sha1
,
1775 enum object_type
*type
, unsigned long *size
)
1777 struct pack_entry e
;
1779 if (!find_pack_entry(sha1
, &e
, NULL
))
1782 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1786 * This is meant to hold a *small* number of objects that you would
1787 * want read_sha1_file() to be able to return, but yet you do not want
1788 * to write them into the object store (e.g. a browse-only
1791 static struct cached_object
{
1792 unsigned char sha1
[20];
1793 enum object_type type
;
1797 static int cached_object_nr
, cached_object_alloc
;
1799 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1802 struct cached_object
*co
= cached_objects
;
1804 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1805 if (!hashcmp(co
->sha1
, sha1
))
1811 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1812 unsigned char *sha1
)
1814 struct cached_object
*co
;
1816 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1817 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1819 if (cached_object_alloc
<= cached_object_nr
) {
1820 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1821 cached_objects
= xrealloc(cached_objects
,
1822 sizeof(*cached_objects
) *
1823 cached_object_alloc
);
1825 co
= &cached_objects
[cached_object_nr
++];
1828 co
->buf
= xmalloc(len
);
1829 memcpy(co
->buf
, buf
, len
);
1830 hashcpy(co
->sha1
, sha1
);
1834 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1835 unsigned long *size
)
1837 unsigned long mapsize
;
1839 struct cached_object
*co
;
1841 co
= find_cached_object(sha1
);
1843 buf
= xmalloc(co
->size
+ 1);
1844 memcpy(buf
, co
->buf
, co
->size
);
1845 ((char*)buf
)[co
->size
] = 0;
1851 buf
= read_packed_sha1(sha1
, type
, size
);
1854 map
= map_sha1_file(sha1
, &mapsize
);
1856 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1857 munmap(map
, mapsize
);
1860 reprepare_packed_git();
1861 return read_packed_sha1(sha1
, type
, size
);
1864 void *read_object_with_reference(const unsigned char *sha1
,
1865 const char *required_type_name
,
1866 unsigned long *size
,
1867 unsigned char *actual_sha1_return
)
1869 enum object_type type
, required_type
;
1871 unsigned long isize
;
1872 unsigned char actual_sha1
[20];
1874 required_type
= type_from_string(required_type_name
);
1875 hashcpy(actual_sha1
, sha1
);
1877 int ref_length
= -1;
1878 const char *ref_type
= NULL
;
1880 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1883 if (type
== required_type
) {
1885 if (actual_sha1_return
)
1886 hashcpy(actual_sha1_return
, actual_sha1
);
1889 /* Handle references */
1890 else if (type
== OBJ_COMMIT
)
1892 else if (type
== OBJ_TAG
)
1893 ref_type
= "object ";
1898 ref_length
= strlen(ref_type
);
1900 if (memcmp(buffer
, ref_type
, ref_length
) ||
1901 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1906 /* Now we have the ID of the referred-to object in
1907 * actual_sha1. Check again. */
1911 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
1912 const char *type
, unsigned char *sha1
,
1913 char *hdr
, int *hdrlen
)
1917 /* Generate the header */
1918 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
1922 SHA1_Update(&c
, hdr
, *hdrlen
);
1923 SHA1_Update(&c
, buf
, len
);
1924 SHA1_Final(sha1
, &c
);
1928 * Link the tempfile to the final place, possibly creating the
1929 * last directory level as you do so.
1931 * Returns the errno on failure, 0 on success.
1933 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
1938 if (!link(tmpfile
, filename
))
1942 * Try to mkdir the last path component if that failed.
1944 * Re-try the "link()" regardless of whether the mkdir
1945 * succeeds, since a race might mean that somebody
1949 dir
= strrchr(filename
, '/');
1952 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
1957 if (!link(tmpfile
, filename
))
1965 * Move the just written object into its final resting place
1967 int move_temp_to_file(const char *tmpfile
, const char *filename
)
1969 int ret
= link_temp_to_file(tmpfile
, filename
);
1972 * Coda hack - coda doesn't like cross-directory links,
1973 * so we fall back to a rename, which will mean that it
1974 * won't be able to check collisions, but that's not a
1977 * The same holds for FAT formatted media.
1979 * When this succeeds, we just return 0. We have nothing
1982 if (ret
&& ret
!= EEXIST
) {
1983 if (!rename(tmpfile
, filename
))
1989 if (ret
!= EEXIST
) {
1990 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
1992 /* FIXME!!! Collision check here ? */
1998 static int write_buffer(int fd
, const void *buf
, size_t len
)
2000 if (write_in_full(fd
, buf
, len
) < 0)
2001 return error("file write error (%s)", strerror(errno
));
2005 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2006 unsigned char *sha1
)
2010 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2014 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2017 unsigned char *compressed
;
2019 unsigned char sha1
[20];
2021 static char tmpfile
[PATH_MAX
];
2025 /* Normally if we have it in the pack then we do not bother writing
2026 * it out into .git/objects/??/?{38} file.
2028 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2029 filename
= sha1_file_name(sha1
);
2031 hashcpy(returnsha1
, sha1
);
2032 if (has_sha1_file(sha1
))
2034 fd
= open(filename
, O_RDONLY
);
2037 * FIXME!!! We might do collision checking here, but we'd
2038 * need to uncompress the old file and check it. Later.
2044 if (errno
!= ENOENT
) {
2045 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2048 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2050 fd
= mkstemp(tmpfile
);
2053 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2055 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2059 memset(&stream
, 0, sizeof(stream
));
2060 deflateInit(&stream
, zlib_compression_level
);
2061 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2062 compressed
= xmalloc(size
);
2065 stream
.next_out
= compressed
;
2066 stream
.avail_out
= size
;
2068 /* First header.. */
2069 stream
.next_in
= (unsigned char *)hdr
;
2070 stream
.avail_in
= hdrlen
;
2071 while (deflate(&stream
, 0) == Z_OK
)
2074 /* Then the data itself.. */
2075 stream
.next_in
= buf
;
2076 stream
.avail_in
= len
;
2077 ret
= deflate(&stream
, Z_FINISH
);
2078 if (ret
!= Z_STREAM_END
)
2079 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2081 ret
= deflateEnd(&stream
);
2083 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2085 size
= stream
.total_out
;
2087 if (write_buffer(fd
, compressed
, size
) < 0)
2088 die("unable to write sha1 file");
2091 die("unable to write sha1 file");
2094 return move_temp_to_file(tmpfile
, filename
);
2098 * We need to unpack and recompress the object for writing
2099 * it out to a different file.
2101 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2105 unsigned char *unpacked
;
2107 enum object_type type
;
2112 /* need to unpack and recompress it by itself */
2113 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2115 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2117 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2120 memset(&stream
, 0, sizeof(stream
));
2121 deflateInit(&stream
, zlib_compression_level
);
2122 size
= deflateBound(&stream
, len
+ hdrlen
);
2123 buf
= xmalloc(size
);
2126 stream
.next_out
= buf
;
2127 stream
.avail_out
= size
;
2129 /* First header.. */
2130 stream
.next_in
= (void *)hdr
;
2131 stream
.avail_in
= hdrlen
;
2132 while (deflate(&stream
, 0) == Z_OK
)
2135 /* Then the data itself.. */
2136 stream
.next_in
= unpacked
;
2137 stream
.avail_in
= len
;
2138 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2140 deflateEnd(&stream
);
2143 *objsize
= stream
.total_out
;
2147 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2150 unsigned long objsize
;
2151 void *buf
= map_sha1_file(sha1
, &objsize
);
2154 retval
= write_buffer(fd
, buf
, objsize
);
2155 munmap(buf
, objsize
);
2159 buf
= repack_object(sha1
, &objsize
);
2160 retval
= write_buffer(fd
, buf
, objsize
);
2165 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2166 size_t bufsize
, size_t *bufposn
)
2168 char tmpfile
[PATH_MAX
];
2171 unsigned char real_sha1
[20];
2172 unsigned char discard
[4096];
2176 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2178 local
= mkstemp(tmpfile
);
2181 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2183 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2186 memset(&stream
, 0, sizeof(stream
));
2188 inflateInit(&stream
);
2195 stream
.avail_in
= *bufposn
;
2196 stream
.next_in
= (unsigned char *) buffer
;
2198 stream
.next_out
= discard
;
2199 stream
.avail_out
= sizeof(discard
);
2200 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2201 SHA1_Update(&c
, discard
, sizeof(discard
) -
2203 } while (stream
.avail_in
&& ret
== Z_OK
);
2204 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2205 die("unable to write sha1 file");
2206 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2208 *bufposn
= stream
.avail_in
;
2212 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2217 return error("Connection closed?");
2218 perror("Reading from connection");
2223 inflateEnd(&stream
);
2225 fchmod(local
, 0444);
2226 if (close(local
) != 0)
2227 die("unable to write sha1 file");
2228 SHA1_Final(real_sha1
, &c
);
2229 if (ret
!= Z_STREAM_END
) {
2231 return error("File %s corrupted", sha1_to_hex(sha1
));
2233 if (hashcmp(sha1
, real_sha1
)) {
2235 return error("File %s has bad hash", sha1_to_hex(sha1
));
2238 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2241 int has_pack_index(const unsigned char *sha1
)
2244 if (stat(sha1_pack_index_name(sha1
), &st
))
2249 int has_pack_file(const unsigned char *sha1
)
2252 if (stat(sha1_pack_name(sha1
), &st
))
2257 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2259 struct pack_entry e
;
2260 return find_pack_entry(sha1
, &e
, ignore_packed
);
2263 int has_sha1_file(const unsigned char *sha1
)
2266 struct pack_entry e
;
2268 if (find_pack_entry(sha1
, &e
, NULL
))
2270 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2274 * reads from fd as long as possible into a supplied buffer of size bytes.
2275 * If necessary the buffer's size is increased using realloc()
2277 * returns 0 if anything went fine and -1 otherwise
2279 * NOTE: both buf and size may change, but even when -1 is returned
2280 * you still have to free() it yourself.
2282 int read_pipe(int fd
, char** return_buf
, unsigned long* return_size
)
2284 char* buf
= *return_buf
;
2285 unsigned long size
= *return_size
;
2287 unsigned long off
= 0;
2290 iret
= xread(fd
, buf
+ off
, size
- off
);
2295 buf
= xrealloc(buf
, size
);
2308 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2310 unsigned long size
= 4096;
2311 char *buf
= xmalloc(size
);
2314 if (read_pipe(fd
, &buf
, &size
)) {
2322 ret
= write_sha1_file(buf
, size
, type
, sha1
);
2324 ret
= hash_sha1_file(buf
, size
, type
, sha1
);
2329 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2330 enum object_type type
, const char *path
)
2332 size_t size
= xsize_t(st
->st_size
);
2334 int ret
, re_allocated
= 0;
2337 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2344 * Convert blobs to git internal format
2346 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2347 unsigned long nsize
= size
;
2348 char *nbuf
= convert_to_git(path
, buf
, &nsize
);
2358 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2360 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2370 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2376 switch (st
->st_mode
& S_IFMT
) {
2378 fd
= open(path
, O_RDONLY
);
2380 return error("open(\"%s\"): %s", path
,
2382 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2383 return error("%s: failed to insert into database",
2387 len
= xsize_t(st
->st_size
);
2388 target
= xmalloc(len
+ 1);
2389 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2390 char *errstr
= strerror(errno
);
2392 return error("readlink(\"%s\"): %s", path
,
2396 hash_sha1_file(target
, len
, blob_type
, sha1
);
2397 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2398 return error("%s: failed to insert into database",
2403 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2405 return error("%s: unsupported file type", path
);
2410 int read_pack_header(int fd
, struct pack_header
*header
)
2412 char *c
= (char*)header
;
2413 ssize_t remaining
= sizeof(struct pack_header
);
2415 ssize_t r
= xread(fd
, c
, remaining
);
2417 /* "eof before pack header was fully read" */
2418 return PH_ERROR_EOF
;
2421 } while (remaining
> 0);
2422 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2423 /* "protocol error (pack signature mismatch detected)" */
2424 return PH_ERROR_PACK_SIGNATURE
;
2425 if (!pack_version_ok(header
->hdr_version
))
2426 /* "protocol error (pack version unsupported)" */
2427 return PH_ERROR_PROTOCOL
;