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 const 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
)
355 const char alt_file_name
[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
362 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
363 fd
= open(path
, O_RDONLY
);
366 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
370 mapsz
= xsize_t(st
.st_size
);
371 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
374 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
379 void prepare_alt_odb(void)
386 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
389 alt_odb_tail
= &alt_odb_list
;
390 link_alt_odb_entries(alt
, alt
+ strlen(alt
), ':', NULL
, 0);
392 read_info_alternates(get_object_directory(), 0);
395 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
397 char *name
= sha1_file_name(sha1
);
398 struct alternate_object_database
*alt
;
403 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
405 fill_sha1_path(name
, sha1
);
406 if (!stat(alt
->base
, st
))
412 static unsigned int pack_used_ctr
;
413 static unsigned int pack_mmap_calls
;
414 static unsigned int peak_pack_open_windows
;
415 static unsigned int pack_open_windows
;
416 static size_t peak_pack_mapped
;
417 static size_t pack_mapped
;
418 struct packed_git
*packed_git
;
420 void pack_report(void)
423 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
424 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
425 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
426 (size_t) getpagesize(),
427 packed_git_window_size
,
430 "pack_report: pack_used_ctr = %10u\n"
431 "pack_report: pack_mmap_calls = %10u\n"
432 "pack_report: pack_open_windows = %10u / %10u\n"
433 "pack_report: pack_mapped = "
434 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
437 pack_open_windows
, peak_pack_open_windows
,
438 pack_mapped
, peak_pack_mapped
);
441 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
444 struct pack_idx_header
*hdr
;
446 uint32_t version
, nr
, i
, *index
;
447 int fd
= open(path
, O_RDONLY
);
452 if (fstat(fd
, &st
)) {
456 idx_size
= xsize_t(st
.st_size
);
457 if (idx_size
< 4 * 256 + 20 + 20) {
459 return error("index file %s is too small", path
);
461 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
465 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
466 version
= ntohl(hdr
->idx_version
);
467 if (version
< 2 || version
> 2) {
468 munmap(idx_map
, idx_size
);
469 return error("index file %s is version %d"
470 " and is not supported by this binary"
471 " (try upgrading GIT to a newer version)",
480 index
+= 2; /* skip index header */
481 for (i
= 0; i
< 256; i
++) {
482 uint32_t n
= ntohl(index
[i
]);
484 munmap(idx_map
, idx_size
);
485 return error("non-monotonic index %s", path
);
493 * - 256 index entries 4 bytes each
494 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
495 * - 20-byte SHA1 of the packfile
496 * - 20-byte SHA1 file checksum
498 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
499 munmap(idx_map
, idx_size
);
500 return error("wrong index v1 file size in %s", path
);
502 } else if (version
== 2) {
505 * - 8 bytes of header
506 * - 256 index entries 4 bytes each
507 * - 20-byte sha1 entry * nr
508 * - 4-byte crc entry * nr
509 * - 4-byte offset entry * nr
510 * - 20-byte SHA1 of the packfile
511 * - 20-byte SHA1 file checksum
512 * And after the 4-byte offset table might be a
513 * variable sized table containing 8-byte entries
514 * for offsets larger than 2^31.
516 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
517 unsigned long max_size
= min_size
;
519 max_size
+= (nr
- 1)*8;
520 if (idx_size
< min_size
|| idx_size
> max_size
) {
521 munmap(idx_map
, idx_size
);
522 return error("wrong index v2 file size in %s", path
);
524 if (idx_size
!= min_size
&&
526 * make sure we can deal with large pack offsets.
527 * 31-bit signed offset won't be enough, neither
528 * 32-bit unsigned one will be.
530 (sizeof(off_t
) <= 4)) {
531 munmap(idx_map
, idx_size
);
532 return error("pack too large for current definition of off_t in %s", path
);
536 p
->index_version
= version
;
537 p
->index_data
= idx_map
;
538 p
->index_size
= idx_size
;
543 int open_pack_index(struct packed_git
*p
)
551 idx_name
= xstrdup(p
->pack_name
);
552 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
553 ret
= check_packed_git_idx(idx_name
, p
);
558 static void scan_windows(struct packed_git
*p
,
559 struct packed_git
**lru_p
,
560 struct pack_window
**lru_w
,
561 struct pack_window
**lru_l
)
563 struct pack_window
*w
, *w_l
;
565 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
567 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
577 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
579 struct packed_git
*p
, *lru_p
= NULL
;
580 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
583 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
584 for (p
= packed_git
; p
; p
= p
->next
)
585 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
587 munmap(lru_w
->base
, lru_w
->len
);
588 pack_mapped
-= lru_w
->len
;
590 lru_l
->next
= lru_w
->next
;
592 lru_p
->windows
= lru_w
->next
;
593 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
594 close(lru_p
->pack_fd
);
605 void release_pack_memory(size_t need
, int fd
)
607 size_t cur
= pack_mapped
;
608 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
612 void unuse_pack(struct pack_window
**w_cursor
)
614 struct pack_window
*w
= *w_cursor
;
622 * Do not call this directly as this leaks p->pack_fd on error return;
623 * call open_packed_git() instead.
625 static int open_packed_git_1(struct packed_git
*p
)
628 struct pack_header hdr
;
629 unsigned char sha1
[20];
630 unsigned char *idx_sha1
;
633 if (!p
->index_data
&& open_pack_index(p
))
634 return error("packfile %s index unavailable", p
->pack_name
);
636 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
637 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
640 /* If we created the struct before we had the pack we lack size. */
642 if (!S_ISREG(st
.st_mode
))
643 return error("packfile %s not a regular file", p
->pack_name
);
644 p
->pack_size
= st
.st_size
;
645 } else if (p
->pack_size
!= st
.st_size
)
646 return error("packfile %s size changed", p
->pack_name
);
648 /* We leave these file descriptors open with sliding mmap;
649 * there is no point keeping them open across exec(), though.
651 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
653 return error("cannot determine file descriptor flags");
654 fd_flag
|= FD_CLOEXEC
;
655 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
656 return error("cannot set FD_CLOEXEC");
658 /* Verify we recognize this pack file format. */
659 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
660 return error("file %s is far too short to be a packfile", p
->pack_name
);
661 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
662 return error("file %s is not a GIT packfile", p
->pack_name
);
663 if (!pack_version_ok(hdr
.hdr_version
))
664 return error("packfile %s is version %u and not supported"
665 " (try upgrading GIT to a newer version)",
666 p
->pack_name
, ntohl(hdr
.hdr_version
));
668 /* Verify the pack matches its index. */
669 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
670 return error("packfile %s claims to have %u objects"
671 " while index indicates %u objects",
672 p
->pack_name
, ntohl(hdr
.hdr_entries
),
674 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
675 return error("end of packfile %s is unavailable", p
->pack_name
);
676 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
677 return error("packfile %s signature is unavailable", p
->pack_name
);
678 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
679 if (hashcmp(sha1
, idx_sha1
))
680 return error("packfile %s does not match index", p
->pack_name
);
684 static int open_packed_git(struct packed_git
*p
)
686 if (!open_packed_git_1(p
))
688 if (p
->pack_fd
!= -1) {
695 static int in_window(struct pack_window
*win
, off_t offset
)
697 /* We must promise at least 20 bytes (one hash) after the
698 * offset is available from this window, otherwise the offset
699 * is not actually in this window and a different window (which
700 * has that one hash excess) must be used. This is to support
701 * the object header and delta base parsing routines below.
703 off_t win_off
= win
->offset
;
704 return win_off
<= offset
705 && (offset
+ 20) <= (win_off
+ win
->len
);
708 unsigned char* use_pack(struct packed_git
*p
,
709 struct pack_window
**w_cursor
,
713 struct pack_window
*win
= *w_cursor
;
715 if (p
->pack_fd
== -1 && open_packed_git(p
))
716 die("packfile %s cannot be accessed", p
->pack_name
);
718 /* Since packfiles end in a hash of their content and its
719 * pointless to ask for an offset into the middle of that
720 * hash, and the in_window function above wouldn't match
721 * don't allow an offset too close to the end of the file.
723 if (offset
> (p
->pack_size
- 20))
724 die("offset beyond end of packfile (truncated pack?)");
726 if (!win
|| !in_window(win
, offset
)) {
729 for (win
= p
->windows
; win
; win
= win
->next
) {
730 if (in_window(win
, offset
))
734 size_t window_align
= packed_git_window_size
/ 2;
736 win
= xcalloc(1, sizeof(*win
));
737 win
->offset
= (offset
/ window_align
) * window_align
;
738 len
= p
->pack_size
- win
->offset
;
739 if (len
> packed_git_window_size
)
740 len
= packed_git_window_size
;
741 win
->len
= (size_t)len
;
742 pack_mapped
+= win
->len
;
743 while (packed_git_limit
< pack_mapped
744 && unuse_one_window(p
, p
->pack_fd
))
746 win
->base
= xmmap(NULL
, win
->len
,
747 PROT_READ
, MAP_PRIVATE
,
748 p
->pack_fd
, win
->offset
);
749 if (win
->base
== MAP_FAILED
)
750 die("packfile %s cannot be mapped: %s",
755 if (pack_mapped
> peak_pack_mapped
)
756 peak_pack_mapped
= pack_mapped
;
757 if (pack_open_windows
> peak_pack_open_windows
)
758 peak_pack_open_windows
= pack_open_windows
;
759 win
->next
= p
->windows
;
763 if (win
!= *w_cursor
) {
764 win
->last_used
= pack_used_ctr
++;
768 offset
-= win
->offset
;
770 *left
= win
->len
- xsize_t(offset
);
771 return win
->base
+ offset
;
774 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
777 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
780 * Make sure a corresponding .pack file exists and that
781 * the index looks sane.
783 path_len
-= strlen(".idx");
786 memcpy(p
->pack_name
, path
, path_len
);
787 strcpy(p
->pack_name
+ path_len
, ".pack");
788 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
793 /* ok, it looks sane as far as we can check without
794 * actually mapping the pack file.
796 p
->index_version
= 0;
797 p
->index_data
= NULL
;
800 p
->pack_size
= st
.st_size
;
804 p
->pack_local
= local
;
805 p
->mtime
= st
.st_mtime
;
806 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
811 struct packed_git
*parse_pack_index(unsigned char *sha1
)
813 char *path
= sha1_pack_index_name(sha1
);
814 return parse_pack_index_file(sha1
, path
);
817 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
818 const char *idx_path
)
820 const char *path
= sha1_pack_name(sha1
);
821 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
823 if (check_packed_git_idx(idx_path
, p
)) {
828 strcpy(p
->pack_name
, path
);
833 hashcpy(p
->sha1
, sha1
);
837 void install_packed_git(struct packed_git
*pack
)
839 pack
->next
= packed_git
;
843 static void prepare_packed_git_one(char *objdir
, int local
)
845 /* Ensure that this buffer is large enough so that we can
846 append "/pack/" without clobbering the stack even if
847 strlen(objdir) were PATH_MAX. */
848 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
853 sprintf(path
, "%s/pack", objdir
);
858 error("unable to open object pack directory: %s: %s",
859 path
, strerror(errno
));
863 while ((de
= readdir(dir
)) != NULL
) {
864 int namelen
= strlen(de
->d_name
);
865 struct packed_git
*p
;
867 if (!has_extension(de
->d_name
, ".idx"))
870 if (len
+ namelen
+ 1 > sizeof(path
))
873 /* Don't reopen a pack we already have. */
874 strcpy(path
+ len
, de
->d_name
);
875 for (p
= packed_git
; p
; p
= p
->next
) {
876 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
881 /* See if it really is a valid .idx file with corresponding
882 * .pack file that we can map.
884 p
= add_packed_git(path
, len
+ namelen
, local
);
887 install_packed_git(p
);
892 static int sort_pack(const void *a_
, const void *b_
)
894 struct packed_git
*a
= *((struct packed_git
**)a_
);
895 struct packed_git
*b
= *((struct packed_git
**)b_
);
899 * Local packs tend to contain objects specific to our
900 * variant of the project than remote ones. In addition,
901 * remote ones could be on a network mounted filesystem.
902 * Favor local ones for these reasons.
904 st
= a
->pack_local
- b
->pack_local
;
909 * Younger packs tend to contain more recent objects,
910 * and more recent objects tend to get accessed more
913 if (a
->mtime
< b
->mtime
)
915 else if (a
->mtime
== b
->mtime
)
920 static void rearrange_packed_git(void)
922 struct packed_git
**ary
, *p
;
925 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
930 /* prepare an array of packed_git for easier sorting */
931 ary
= xcalloc(n
, sizeof(struct packed_git
*));
932 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
935 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
937 /* link them back again */
938 for (i
= 0; i
< n
- 1; i
++)
939 ary
[i
]->next
= ary
[i
+ 1];
940 ary
[n
- 1]->next
= NULL
;
946 static int prepare_packed_git_run_once
= 0;
947 void prepare_packed_git(void)
949 struct alternate_object_database
*alt
;
951 if (prepare_packed_git_run_once
)
953 prepare_packed_git_one(get_object_directory(), 1);
955 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
957 prepare_packed_git_one(alt
->base
, 0);
960 rearrange_packed_git();
961 prepare_packed_git_run_once
= 1;
964 void reprepare_packed_git(void)
966 prepare_packed_git_run_once
= 0;
967 prepare_packed_git();
970 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
972 unsigned char real_sha1
[20];
973 hash_sha1_file(map
, size
, type
, real_sha1
);
974 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
977 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
982 char *filename
= find_sha1_file(sha1
, &st
);
988 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
990 /* See if it works without O_NOATIME */
991 switch (sha1_file_open_flag
) {
993 fd
= open(filename
, O_RDONLY
);
1001 /* If it failed once, it will probably fail again.
1002 * Stop using O_NOATIME
1004 sha1_file_open_flag
= 0;
1006 *size
= xsize_t(st
.st_size
);
1007 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1012 static int legacy_loose_object(unsigned char *map
)
1017 * Is it a zlib-compressed buffer? If so, the first byte
1018 * must be 0x78 (15-bit window size, deflated), and the
1019 * first 16-bit word is evenly divisible by 31
1021 word
= (map
[0] << 8) + map
[1];
1022 if (map
[0] == 0x78 && !(word
% 31))
1028 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1033 unsigned long used
= 0;
1036 *type
= (c
>> 4) & 7;
1042 if (sizeof(long) * 8 <= shift
)
1045 size
+= (c
& 0x7f) << shift
;
1052 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1054 unsigned long size
, used
;
1055 static const char valid_loose_object_type
[8] = {
1057 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1058 0, /* "delta" and others are invalid in a loose object */
1060 enum object_type type
;
1062 /* Get the data stream */
1063 memset(stream
, 0, sizeof(*stream
));
1064 stream
->next_in
= map
;
1065 stream
->avail_in
= mapsize
;
1066 stream
->next_out
= buffer
;
1067 stream
->avail_out
= bufsiz
;
1069 if (legacy_loose_object(map
)) {
1070 inflateInit(stream
);
1071 return inflate(stream
, 0);
1076 * There used to be a second loose object header format which
1077 * was meant to mimic the in-pack format, allowing for direct
1078 * copy of the object data. This format turned up not to be
1079 * really worth it and we don't write it any longer. But we
1080 * can still read it.
1082 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1083 if (!used
|| !valid_loose_object_type
[type
])
1088 /* Set up the stream for the rest.. */
1089 stream
->next_in
= map
;
1090 stream
->avail_in
= mapsize
;
1091 inflateInit(stream
);
1093 /* And generate the fake traditional header */
1094 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1095 typename(type
), size
);
1099 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1101 int bytes
= strlen(buffer
) + 1;
1102 unsigned char *buf
= xmalloc(1+size
);
1106 n
= stream
->total_out
- bytes
;
1109 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1111 if (bytes
<= size
) {
1113 * The above condition must be (bytes <= size), not
1114 * (bytes < size). In other words, even though we
1115 * expect no more output and set avail_out to zer0,
1116 * the input zlib stream may have bytes that express
1117 * "this concludes the stream", and we *do* want to
1120 * Otherwise we would not be able to test that we
1121 * consumed all the input to reach the expected size;
1122 * we also want to check that zlib tells us that all
1123 * went well with status == Z_STREAM_END at the end.
1125 stream
->next_out
= buf
+ bytes
;
1126 stream
->avail_out
= size
- bytes
;
1127 while (status
== Z_OK
)
1128 status
= inflate(stream
, Z_FINISH
);
1131 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1137 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1138 else if (stream
->avail_in
)
1139 error("garbage at end of loose object '%s'",
1146 * We used to just use "sscanf()", but that's actually way
1147 * too permissive for what we want to check. So do an anal
1148 * object header parse by hand.
1150 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1157 * The type can be at most ten bytes (including the
1158 * terminating '\0' that we add), and is followed by
1167 if (i
>= sizeof(type
))
1173 * The length must follow immediately, and be in canonical
1174 * decimal format (ie "010" is not valid).
1176 size
= *hdr
++ - '0';
1181 unsigned long c
= *hdr
- '0';
1185 size
= size
* 10 + c
;
1191 * The length must be followed by a zero byte
1193 return *hdr
? -1 : type_from_string(type
);
1196 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1202 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1203 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1206 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1209 unsigned long get_size_from_delta(struct packed_git
*p
,
1210 struct pack_window
**w_curs
,
1213 const unsigned char *data
;
1214 unsigned char delta_head
[20], *in
;
1218 memset(&stream
, 0, sizeof(stream
));
1219 stream
.next_out
= delta_head
;
1220 stream
.avail_out
= sizeof(delta_head
);
1222 inflateInit(&stream
);
1224 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1225 stream
.next_in
= in
;
1226 st
= inflate(&stream
, Z_FINISH
);
1227 curpos
+= stream
.next_in
- in
;
1228 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1229 stream
.total_out
< sizeof(delta_head
));
1230 inflateEnd(&stream
);
1231 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1232 die("delta data unpack-initial failed");
1234 /* Examine the initial part of the delta to figure out
1239 /* ignore base size */
1240 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1242 /* Read the result size */
1243 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1246 static off_t
get_delta_base(struct packed_git
*p
,
1247 struct pack_window
**w_curs
,
1249 enum object_type type
,
1250 off_t delta_obj_offset
)
1252 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1255 /* use_pack() assured us we have [base_info, base_info + 20)
1256 * as a range that we can look at without walking off the
1257 * end of the mapped window. Its actually the hash size
1258 * that is assured. An OFS_DELTA longer than the hash size
1259 * is stupid, as then a REF_DELTA would be smaller to store.
1261 if (type
== OBJ_OFS_DELTA
) {
1263 unsigned char c
= base_info
[used
++];
1264 base_offset
= c
& 127;
1267 if (!base_offset
|| MSB(base_offset
, 7))
1268 die("offset value overflow for delta base object");
1269 c
= base_info
[used
++];
1270 base_offset
= (base_offset
<< 7) + (c
& 127);
1272 base_offset
= delta_obj_offset
- base_offset
;
1273 if (base_offset
>= delta_obj_offset
)
1274 die("delta base offset out of bound");
1276 } else if (type
== OBJ_REF_DELTA
) {
1277 /* The base entry _must_ be in the same pack */
1278 base_offset
= find_pack_entry_one(base_info
, p
);
1280 die("failed to find delta-pack base object %s",
1281 sha1_to_hex(base_info
));
1284 die("I am totally screwed");
1288 /* forward declaration for a mutually recursive function */
1289 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1290 unsigned long *sizep
);
1292 static int packed_delta_info(struct packed_git
*p
,
1293 struct pack_window
**w_curs
,
1295 enum object_type type
,
1297 unsigned long *sizep
)
1301 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1302 type
= packed_object_info(p
, base_offset
, NULL
);
1304 /* We choose to only get the type of the base object and
1305 * ignore potentially corrupt pack file that expects the delta
1306 * based on a base with a wrong size. This saves tons of
1310 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1315 static int unpack_object_header(struct packed_git
*p
,
1316 struct pack_window
**w_curs
,
1318 unsigned long *sizep
)
1320 unsigned char *base
;
1323 enum object_type type
;
1325 /* use_pack() assures us we have [base, base + 20) available
1326 * as a range that we can look at at. (Its actually the hash
1327 * size that is assured.) With our object header encoding
1328 * the maximum deflated object size is 2^137, which is just
1329 * insane, so we know won't exceed what we have been given.
1331 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1332 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1334 die("object offset outside of pack file");
1340 const char *packed_object_info_detail(struct packed_git
*p
,
1342 unsigned long *size
,
1343 unsigned long *store_size
,
1344 unsigned int *delta_chain_length
,
1345 unsigned char *base_sha1
)
1347 struct pack_window
*w_curs
= NULL
;
1349 unsigned long dummy
;
1350 unsigned char *next_sha1
;
1351 enum object_type type
;
1353 *delta_chain_length
= 0;
1354 curpos
= obj_offset
;
1355 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1360 die("pack %s contains unknown object type %d",
1361 p
->pack_name
, type
);
1366 *store_size
= 0; /* notyet */
1367 unuse_pack(&w_curs
);
1368 return typename(type
);
1370 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1371 if (*delta_chain_length
== 0) {
1372 /* TODO: find base_sha1 as pointed by curpos */
1377 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1378 if (*delta_chain_length
== 0)
1379 hashcpy(base_sha1
, next_sha1
);
1380 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1383 (*delta_chain_length
)++;
1384 curpos
= obj_offset
;
1385 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1389 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1390 unsigned long *sizep
)
1392 struct pack_window
*w_curs
= NULL
;
1394 off_t curpos
= obj_offset
;
1395 enum object_type type
;
1397 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1402 type
= packed_delta_info(p
, &w_curs
, curpos
,
1403 type
, obj_offset
, sizep
);
1413 die("pack %s contains unknown object type %d",
1414 p
->pack_name
, type
);
1416 unuse_pack(&w_curs
);
1420 static void *unpack_compressed_entry(struct packed_git
*p
,
1421 struct pack_window
**w_curs
,
1427 unsigned char *buffer
, *in
;
1429 buffer
= xmalloc(size
+ 1);
1431 memset(&stream
, 0, sizeof(stream
));
1432 stream
.next_out
= buffer
;
1433 stream
.avail_out
= size
;
1435 inflateInit(&stream
);
1437 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1438 stream
.next_in
= in
;
1439 st
= inflate(&stream
, Z_FINISH
);
1440 curpos
+= stream
.next_in
- in
;
1441 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1442 inflateEnd(&stream
);
1443 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1451 #define MAX_DELTA_CACHE (256)
1453 static size_t delta_base_cached
;
1455 static struct delta_base_cache_lru_list
{
1456 struct delta_base_cache_lru_list
*prev
;
1457 struct delta_base_cache_lru_list
*next
;
1458 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1460 static struct delta_base_cache_entry
{
1461 struct delta_base_cache_lru_list lru
;
1463 struct packed_git
*p
;
1466 enum object_type type
;
1467 } delta_base_cache
[MAX_DELTA_CACHE
];
1469 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1473 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1474 hash
+= (hash
>> 8) + (hash
>> 16);
1475 return hash
% MAX_DELTA_CACHE
;
1478 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1479 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1482 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1483 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1486 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1487 goto found_cache_entry
;
1488 return unpack_entry(p
, base_offset
, type
, base_size
);
1493 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1494 ent
->lru
.prev
->next
= ent
->lru
.next
;
1495 delta_base_cached
-= ent
->size
;
1497 ret
= xmemdupz(ent
->data
, ent
->size
);
1500 *base_size
= ent
->size
;
1504 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1509 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1510 ent
->lru
.prev
->next
= ent
->lru
.next
;
1511 delta_base_cached
-= ent
->size
;
1515 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1516 void *base
, unsigned long base_size
, enum object_type type
)
1518 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1519 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1520 struct delta_base_cache_lru_list
*lru
;
1522 release_delta_base_cache(ent
);
1523 delta_base_cached
+= base_size
;
1525 for (lru
= delta_base_cache_lru
.next
;
1526 delta_base_cached
> delta_base_cache_limit
1527 && lru
!= &delta_base_cache_lru
;
1529 struct delta_base_cache_entry
*f
= (void *)lru
;
1530 if (f
->type
== OBJ_BLOB
)
1531 release_delta_base_cache(f
);
1533 for (lru
= delta_base_cache_lru
.next
;
1534 delta_base_cached
> delta_base_cache_limit
1535 && lru
!= &delta_base_cache_lru
;
1537 struct delta_base_cache_entry
*f
= (void *)lru
;
1538 release_delta_base_cache(f
);
1542 ent
->base_offset
= base_offset
;
1545 ent
->size
= base_size
;
1546 ent
->lru
.next
= &delta_base_cache_lru
;
1547 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1548 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1549 delta_base_cache_lru
.prev
= &ent
->lru
;
1552 static void *unpack_delta_entry(struct packed_git
*p
,
1553 struct pack_window
**w_curs
,
1555 unsigned long delta_size
,
1557 enum object_type
*type
,
1558 unsigned long *sizep
)
1560 void *delta_data
, *result
, *base
;
1561 unsigned long base_size
;
1564 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1565 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1567 die("failed to read delta base object"
1568 " at %"PRIuMAX
" from %s",
1569 (uintmax_t)base_offset
, p
->pack_name
);
1571 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1573 die("failed to unpack compressed delta"
1574 " at %"PRIuMAX
" from %s",
1575 (uintmax_t)curpos
, p
->pack_name
);
1576 result
= patch_delta(base
, base_size
,
1577 delta_data
, delta_size
,
1580 die("failed to apply delta");
1582 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1586 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1587 enum object_type
*type
, unsigned long *sizep
)
1589 struct pack_window
*w_curs
= NULL
;
1590 off_t curpos
= obj_offset
;
1593 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1597 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1598 obj_offset
, type
, sizep
);
1604 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1607 die("unknown object type %i in %s", *type
, p
->pack_name
);
1609 unuse_pack(&w_curs
);
1613 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1616 const unsigned char *index
= p
->index_data
;
1618 if (open_pack_index(p
))
1620 index
= p
->index_data
;
1622 if (n
>= p
->num_objects
)
1625 if (p
->index_version
== 1) {
1626 return index
+ 24 * n
+ 4;
1629 return index
+ 20 * n
;
1633 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1635 const unsigned char *index
= p
->index_data
;
1637 if (p
->index_version
== 1) {
1638 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1641 index
+= 8 + p
->num_objects
* (20 + 4);
1642 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1643 if (!(off
& 0x80000000))
1645 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1646 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1647 ntohl(*((uint32_t *)(index
+ 4)));
1651 off_t
find_pack_entry_one(const unsigned char *sha1
,
1652 struct packed_git
*p
)
1654 const uint32_t *level1_ofs
= p
->index_data
;
1655 const unsigned char *index
= p
->index_data
;
1659 if (open_pack_index(p
))
1661 level1_ofs
= p
->index_data
;
1662 index
= p
->index_data
;
1664 if (p
->index_version
> 1) {
1669 hi
= ntohl(level1_ofs
[*sha1
]);
1670 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1673 unsigned mi
= (lo
+ hi
) / 2;
1674 unsigned x
= (p
->index_version
> 1) ? (mi
* 20) : (mi
* 24 + 4);
1675 int cmp
= hashcmp(index
+ x
, sha1
);
1677 return nth_packed_object_offset(p
, mi
);
1686 int matches_pack_name(struct packed_git
*p
, const char *name
)
1688 const char *last_c
, *c
;
1690 if (!strcmp(p
->pack_name
, name
))
1693 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1698 if (!strcmp(last_c
, name
))
1704 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1706 static struct packed_git
*last_found
= (void *)1;
1707 struct packed_git
*p
;
1710 prepare_packed_git();
1713 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1716 if (ignore_packed
) {
1718 for (ig
= ignore_packed
; *ig
; ig
++)
1719 if (matches_pack_name(p
, *ig
))
1725 offset
= find_pack_entry_one(sha1
, p
);
1728 * We are about to tell the caller where they can
1729 * locate the requested object. We better make
1730 * sure the packfile is still here and can be
1731 * accessed before supplying that answer, as
1732 * it may have been deleted since the index
1735 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1736 error("packfile %s cannot be accessed", p
->pack_name
);
1741 hashcpy(e
->sha1
, sha1
);
1747 if (p
== last_found
)
1751 if (p
== last_found
)
1757 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1758 struct packed_git
*packs
)
1760 struct packed_git
*p
;
1762 for (p
= packs
; p
; p
= p
->next
) {
1763 if (find_pack_entry_one(sha1
, p
))
1770 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1773 unsigned long mapsize
, size
;
1778 map
= map_sha1_file(sha1
, &mapsize
);
1780 return error("unable to find %s", sha1_to_hex(sha1
));
1781 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1782 status
= error("unable to unpack %s header",
1784 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1785 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1788 inflateEnd(&stream
);
1789 munmap(map
, mapsize
);
1793 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1795 struct pack_entry e
;
1797 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1798 reprepare_packed_git();
1799 if (!find_pack_entry(sha1
, &e
, NULL
))
1800 return sha1_loose_object_info(sha1
, sizep
);
1802 return packed_object_info(e
.p
, e
.offset
, sizep
);
1805 static void *read_packed_sha1(const unsigned char *sha1
,
1806 enum object_type
*type
, unsigned long *size
)
1808 struct pack_entry e
;
1810 if (!find_pack_entry(sha1
, &e
, NULL
))
1813 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1817 * This is meant to hold a *small* number of objects that you would
1818 * want read_sha1_file() to be able to return, but yet you do not want
1819 * to write them into the object store (e.g. a browse-only
1822 static struct cached_object
{
1823 unsigned char sha1
[20];
1824 enum object_type type
;
1828 static int cached_object_nr
, cached_object_alloc
;
1830 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1833 struct cached_object
*co
= cached_objects
;
1835 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1836 if (!hashcmp(co
->sha1
, sha1
))
1842 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1843 unsigned char *sha1
)
1845 struct cached_object
*co
;
1847 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1848 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1850 if (cached_object_alloc
<= cached_object_nr
) {
1851 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1852 cached_objects
= xrealloc(cached_objects
,
1853 sizeof(*cached_objects
) *
1854 cached_object_alloc
);
1856 co
= &cached_objects
[cached_object_nr
++];
1859 co
->buf
= xmalloc(len
);
1860 memcpy(co
->buf
, buf
, len
);
1861 hashcpy(co
->sha1
, sha1
);
1865 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1866 unsigned long *size
)
1868 unsigned long mapsize
;
1870 struct cached_object
*co
;
1872 co
= find_cached_object(sha1
);
1876 return xmemdupz(co
->buf
, co
->size
);
1879 buf
= read_packed_sha1(sha1
, type
, size
);
1882 map
= map_sha1_file(sha1
, &mapsize
);
1884 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1885 munmap(map
, mapsize
);
1888 reprepare_packed_git();
1889 return read_packed_sha1(sha1
, type
, size
);
1892 void *read_object_with_reference(const unsigned char *sha1
,
1893 const char *required_type_name
,
1894 unsigned long *size
,
1895 unsigned char *actual_sha1_return
)
1897 enum object_type type
, required_type
;
1899 unsigned long isize
;
1900 unsigned char actual_sha1
[20];
1902 required_type
= type_from_string(required_type_name
);
1903 hashcpy(actual_sha1
, sha1
);
1905 int ref_length
= -1;
1906 const char *ref_type
= NULL
;
1908 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1911 if (type
== required_type
) {
1913 if (actual_sha1_return
)
1914 hashcpy(actual_sha1_return
, actual_sha1
);
1917 /* Handle references */
1918 else if (type
== OBJ_COMMIT
)
1920 else if (type
== OBJ_TAG
)
1921 ref_type
= "object ";
1926 ref_length
= strlen(ref_type
);
1928 if (memcmp(buffer
, ref_type
, ref_length
) ||
1929 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1934 /* Now we have the ID of the referred-to object in
1935 * actual_sha1. Check again. */
1939 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
1940 const char *type
, unsigned char *sha1
,
1941 char *hdr
, int *hdrlen
)
1945 /* Generate the header */
1946 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
1950 SHA1_Update(&c
, hdr
, *hdrlen
);
1951 SHA1_Update(&c
, buf
, len
);
1952 SHA1_Final(sha1
, &c
);
1956 * Link the tempfile to the final place, possibly creating the
1957 * last directory level as you do so.
1959 * Returns the errno on failure, 0 on success.
1961 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
1966 if (!link(tmpfile
, filename
))
1970 * Try to mkdir the last path component if that failed.
1972 * Re-try the "link()" regardless of whether the mkdir
1973 * succeeds, since a race might mean that somebody
1977 dir
= strrchr(filename
, '/');
1980 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
1985 if (!link(tmpfile
, filename
))
1993 * Move the just written object into its final resting place
1995 int move_temp_to_file(const char *tmpfile
, const char *filename
)
1997 int ret
= link_temp_to_file(tmpfile
, filename
);
2000 * Coda hack - coda doesn't like cross-directory links,
2001 * so we fall back to a rename, which will mean that it
2002 * won't be able to check collisions, but that's not a
2005 * The same holds for FAT formatted media.
2007 * When this succeeds, we just return 0. We have nothing
2010 if (ret
&& ret
!= EEXIST
) {
2011 if (!rename(tmpfile
, filename
))
2017 if (ret
!= EEXIST
) {
2018 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2020 /* FIXME!!! Collision check here ? */
2026 static int write_buffer(int fd
, const void *buf
, size_t len
)
2028 if (write_in_full(fd
, buf
, len
) < 0)
2029 return error("file write error (%s)", strerror(errno
));
2033 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2034 unsigned char *sha1
)
2038 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2042 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2045 unsigned char *compressed
;
2047 unsigned char sha1
[20];
2049 static char tmpfile
[PATH_MAX
];
2053 /* Normally if we have it in the pack then we do not bother writing
2054 * it out into .git/objects/??/?{38} file.
2056 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2057 filename
= sha1_file_name(sha1
);
2059 hashcpy(returnsha1
, sha1
);
2060 if (has_sha1_file(sha1
))
2062 fd
= open(filename
, O_RDONLY
);
2065 * FIXME!!! We might do collision checking here, but we'd
2066 * need to uncompress the old file and check it. Later.
2072 if (errno
!= ENOENT
) {
2073 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2076 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2078 fd
= mkstemp(tmpfile
);
2081 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2083 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2087 memset(&stream
, 0, sizeof(stream
));
2088 deflateInit(&stream
, zlib_compression_level
);
2089 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2090 compressed
= xmalloc(size
);
2093 stream
.next_out
= compressed
;
2094 stream
.avail_out
= size
;
2096 /* First header.. */
2097 stream
.next_in
= (unsigned char *)hdr
;
2098 stream
.avail_in
= hdrlen
;
2099 while (deflate(&stream
, 0) == Z_OK
)
2102 /* Then the data itself.. */
2103 stream
.next_in
= buf
;
2104 stream
.avail_in
= len
;
2105 ret
= deflate(&stream
, Z_FINISH
);
2106 if (ret
!= Z_STREAM_END
)
2107 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2109 ret
= deflateEnd(&stream
);
2111 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2113 size
= stream
.total_out
;
2115 if (write_buffer(fd
, compressed
, size
) < 0)
2116 die("unable to write sha1 file");
2119 die("unable to write sha1 file");
2122 return move_temp_to_file(tmpfile
, filename
);
2126 * We need to unpack and recompress the object for writing
2127 * it out to a different file.
2129 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2133 unsigned char *unpacked
;
2135 enum object_type type
;
2140 /* need to unpack and recompress it by itself */
2141 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2143 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2145 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2148 memset(&stream
, 0, sizeof(stream
));
2149 deflateInit(&stream
, zlib_compression_level
);
2150 size
= deflateBound(&stream
, len
+ hdrlen
);
2151 buf
= xmalloc(size
);
2154 stream
.next_out
= buf
;
2155 stream
.avail_out
= size
;
2157 /* First header.. */
2158 stream
.next_in
= (void *)hdr
;
2159 stream
.avail_in
= hdrlen
;
2160 while (deflate(&stream
, 0) == Z_OK
)
2163 /* Then the data itself.. */
2164 stream
.next_in
= unpacked
;
2165 stream
.avail_in
= len
;
2166 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2168 deflateEnd(&stream
);
2171 *objsize
= stream
.total_out
;
2175 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2178 unsigned long objsize
;
2179 void *buf
= map_sha1_file(sha1
, &objsize
);
2182 retval
= write_buffer(fd
, buf
, objsize
);
2183 munmap(buf
, objsize
);
2187 buf
= repack_object(sha1
, &objsize
);
2188 retval
= write_buffer(fd
, buf
, objsize
);
2193 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2194 size_t bufsize
, size_t *bufposn
)
2196 char tmpfile
[PATH_MAX
];
2199 unsigned char real_sha1
[20];
2200 unsigned char discard
[4096];
2204 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2206 local
= mkstemp(tmpfile
);
2209 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2211 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2214 memset(&stream
, 0, sizeof(stream
));
2216 inflateInit(&stream
);
2223 stream
.avail_in
= *bufposn
;
2224 stream
.next_in
= (unsigned char *) buffer
;
2226 stream
.next_out
= discard
;
2227 stream
.avail_out
= sizeof(discard
);
2228 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2229 SHA1_Update(&c
, discard
, sizeof(discard
) -
2231 } while (stream
.avail_in
&& ret
== Z_OK
);
2232 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2233 die("unable to write sha1 file");
2234 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2236 *bufposn
= stream
.avail_in
;
2240 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2245 return error("Connection closed?");
2246 perror("Reading from connection");
2251 inflateEnd(&stream
);
2253 fchmod(local
, 0444);
2254 if (close(local
) != 0)
2255 die("unable to write sha1 file");
2256 SHA1_Final(real_sha1
, &c
);
2257 if (ret
!= Z_STREAM_END
) {
2259 return error("File %s corrupted", sha1_to_hex(sha1
));
2261 if (hashcmp(sha1
, real_sha1
)) {
2263 return error("File %s has bad hash", sha1_to_hex(sha1
));
2266 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2269 int has_pack_index(const unsigned char *sha1
)
2272 if (stat(sha1_pack_index_name(sha1
), &st
))
2277 int has_pack_file(const unsigned char *sha1
)
2280 if (stat(sha1_pack_name(sha1
), &st
))
2285 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2287 struct pack_entry e
;
2288 return find_pack_entry(sha1
, &e
, ignore_packed
);
2291 int has_sha1_file(const unsigned char *sha1
)
2294 struct pack_entry e
;
2296 if (find_pack_entry(sha1
, &e
, NULL
))
2298 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2301 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2306 strbuf_init(&buf
, 0);
2307 if (strbuf_read(&buf
, fd
, 4096) < 0) {
2308 strbuf_release(&buf
);
2315 ret
= write_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2317 ret
= hash_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2318 strbuf_release(&buf
);
2323 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2324 enum object_type type
, const char *path
)
2326 size_t size
= xsize_t(st
->st_size
);
2328 int ret
, re_allocated
= 0;
2331 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2338 * Convert blobs to git internal format
2340 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2342 strbuf_init(&nbuf
, 0);
2343 if (convert_to_git(path
, buf
, size
, &nbuf
)) {
2345 buf
= strbuf_detach(&nbuf
, &size
);
2351 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2353 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2363 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2369 switch (st
->st_mode
& S_IFMT
) {
2371 fd
= open(path
, O_RDONLY
);
2373 return error("open(\"%s\"): %s", path
,
2375 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2376 return error("%s: failed to insert into database",
2380 len
= xsize_t(st
->st_size
);
2381 target
= xmalloc(len
+ 1);
2382 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2383 char *errstr
= strerror(errno
);
2385 return error("readlink(\"%s\"): %s", path
,
2389 hash_sha1_file(target
, len
, blob_type
, sha1
);
2390 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2391 return error("%s: failed to insert into database",
2396 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2398 return error("%s: unsupported file type", path
);
2403 int read_pack_header(int fd
, struct pack_header
*header
)
2405 char *c
= (char*)header
;
2406 ssize_t remaining
= sizeof(struct pack_header
);
2408 ssize_t r
= xread(fd
, c
, remaining
);
2410 /* "eof before pack header was fully read" */
2411 return PH_ERROR_EOF
;
2414 } while (remaining
> 0);
2415 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2416 /* "protocol error (pack signature mismatch detected)" */
2417 return PH_ERROR_PACK_SIGNATURE
;
2418 if (!pack_version_ok(header
->hdr_version
))
2419 /* "protocol error (pack version unsupported)" */
2420 return PH_ERROR_PROTOCOL
;