2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s
) { return (unsigned long)s
; }
33 static size_t sz_fmt(size_t s
) { return s
; }
36 const unsigned char null_sha1
[20];
38 static unsigned int sha1_file_open_flag
= O_NOATIME
;
40 const signed char hexval_table
[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
78 for (i
= 0; i
< 20; i
++) {
79 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
88 int safe_create_leading_directories(char *path
)
93 if (is_absolute_path(path
))
97 pos
= strchr(pos
, '/');
101 if (!stat(path
, &st
)) {
103 if (!S_ISDIR(st
.st_mode
)) {
108 else if (mkdir(path
, 0777)) {
112 else if (adjust_shared_perm(path
)) {
121 char * sha1_to_hex(const unsigned char *sha1
)
124 static char hexbuffer
[4][50];
125 static const char hex
[] = "0123456789abcdef";
126 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
129 for (i
= 0; i
< 20; i
++) {
130 unsigned int val
= *sha1
++;
131 *buf
++ = hex
[val
>> 4];
132 *buf
++ = hex
[val
& 0xf];
139 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
142 for (i
= 0; i
< 20; i
++) {
143 static char hex
[] = "0123456789abcdef";
144 unsigned int val
= sha1
[i
];
145 char *pos
= pathbuf
+ i
*2 + (i
> 0);
146 *pos
++ = hex
[val
>> 4];
147 *pos
= hex
[val
& 0xf];
152 * NOTE! This returns a statically allocated buffer, so you have to be
153 * careful about using it. Do an "xstrdup()" if you need to save the
156 * Also note that this returns the location for creating. Reading
157 * SHA1 file can happen from any alternate directory listed in the
158 * DB_ENVIRONMENT environment variable if it is not found in
159 * the primary object database.
161 char *sha1_file_name(const unsigned char *sha1
)
163 static char *name
, *base
;
166 const char *sha1_file_directory
= get_object_directory();
167 int len
= strlen(sha1_file_directory
);
168 base
= xmalloc(len
+ 60);
169 memcpy(base
, sha1_file_directory
, len
);
170 memset(base
+len
, 0, 60);
173 name
= base
+ len
+ 1;
175 fill_sha1_path(name
, sha1
);
179 static char *sha1_get_pack_name(const unsigned char *sha1
,
180 char **name
, char **base
)
182 static const char hex
[] = "0123456789abcdef";
187 const char *sha1_file_directory
= get_object_directory();
188 int len
= strlen(sha1_file_directory
);
189 *base
= xmalloc(len
+ 60);
190 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory
);
191 *name
= *base
+ len
+ 11;
196 for (i
= 0; i
< 20; i
++) {
197 unsigned int val
= *sha1
++;
198 *buf
++ = hex
[val
>> 4];
199 *buf
++ = hex
[val
& 0xf];
205 char *sha1_pack_name(const unsigned char *sha1
)
207 static char *name
, *base
;
209 return sha1_get_pack_name(sha1
, &name
, &base
);
212 char *sha1_pack_index_name(const unsigned char *sha1
)
214 static char *name
, *base
;
216 return sha1_get_pack_name(sha1
, &name
, &base
);
219 struct alternate_object_database
*alt_odb_list
;
220 static struct alternate_object_database
**alt_odb_tail
;
222 static void read_info_alternates(const char * alternates
, int depth
);
225 * Prepare alternate object database registry.
227 * The variable alt_odb_list points at the list of struct
228 * alternate_object_database. The elements on this list come from
229 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
230 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
231 * whose contents is similar to that environment variable but can be
232 * LF separated. Its base points at a statically allocated buffer that
233 * contains "/the/directory/corresponding/to/.git/objects/...", while
234 * its name points just after the slash at the end of ".git/objects/"
235 * in the example above, and has enough space to hold 40-byte hex
236 * SHA1, an extra slash for the first level indirection, and the
239 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
242 const char *objdir
= get_object_directory();
243 struct alternate_object_database
*ent
;
244 struct alternate_object_database
*alt
;
245 /* 43 = 40-byte + 2 '/' + terminating NUL */
247 int entlen
= pfxlen
+ 43;
250 if (!is_absolute_path(entry
) && relative_base
) {
251 /* Relative alt-odb */
253 base_len
= strlen(relative_base
) + 1;
257 ent
= xmalloc(sizeof(*ent
) + entlen
);
259 if (!is_absolute_path(entry
) && relative_base
) {
260 memcpy(ent
->base
, relative_base
, base_len
- 1);
261 ent
->base
[base_len
- 1] = '/';
262 memcpy(ent
->base
+ base_len
, entry
, len
);
265 memcpy(ent
->base
, entry
, pfxlen
);
267 ent
->name
= ent
->base
+ pfxlen
+ 1;
268 ent
->base
[pfxlen
+ 3] = '/';
269 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
271 /* Detect cases where alternate disappeared */
272 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
273 error("object directory %s does not exist; "
274 "check .git/objects/info/alternates.",
280 /* Prevent the common mistake of listing the same
281 * thing twice, or object directory itself.
283 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
284 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
289 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
294 /* add the alternate entry */
296 alt_odb_tail
= &(ent
->next
);
299 /* recursively add alternates */
300 read_info_alternates(ent
->base
, depth
+ 1);
302 ent
->base
[pfxlen
] = '/';
307 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
308 const char *relative_base
, int depth
)
310 const char *cp
, *last
;
313 error("%s: ignoring alternate object stores, nesting too deep.",
321 if (cp
< ep
&& *cp
== '#') {
322 while (cp
< ep
&& *cp
!= sep
)
327 while (cp
< ep
&& *cp
!= sep
)
330 if (!is_absolute_path(last
) && depth
) {
331 error("%s: ignoring relative alternate object store %s",
332 relative_base
, last
);
334 link_alt_odb_entry(last
, cp
- last
,
335 relative_base
, depth
);
338 while (cp
< ep
&& *cp
== sep
)
344 static void read_info_alternates(const char * relative_base
, int depth
)
349 const char alt_file_name
[] = "info/alternates";
350 /* Given that relative_base is no longer than PATH_MAX,
351 ensure that "path" has enough space to append "/", the
352 file name, "info/alternates", and a trailing NUL. */
353 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
356 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
357 fd
= open(path
, O_RDONLY
);
360 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
364 mapsz
= xsize_t(st
.st_size
);
365 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
368 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
373 void add_to_alternates_file(const char *reference
)
375 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
376 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), 1);
377 char *alt
= mkpath("%s/objects\n", reference
);
378 write_or_die(fd
, alt
, strlen(alt
));
379 if (commit_lock_file(lock
))
380 die("could not close alternates file");
382 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
385 void prepare_alt_odb(void)
392 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
395 alt_odb_tail
= &alt_odb_list
;
396 link_alt_odb_entries(alt
, alt
+ strlen(alt
), ':', NULL
, 0);
398 read_info_alternates(get_object_directory(), 0);
401 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
403 char *name
= sha1_file_name(sha1
);
404 struct alternate_object_database
*alt
;
409 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
411 fill_sha1_path(name
, sha1
);
412 if (!stat(alt
->base
, st
))
418 static unsigned int pack_used_ctr
;
419 static unsigned int pack_mmap_calls
;
420 static unsigned int peak_pack_open_windows
;
421 static unsigned int pack_open_windows
;
422 static size_t peak_pack_mapped
;
423 static size_t pack_mapped
;
424 struct packed_git
*packed_git
;
426 void pack_report(void)
429 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
430 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
431 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
432 sz_fmt(getpagesize()),
433 sz_fmt(packed_git_window_size
),
434 sz_fmt(packed_git_limit
));
436 "pack_report: pack_used_ctr = %10u\n"
437 "pack_report: pack_mmap_calls = %10u\n"
438 "pack_report: pack_open_windows = %10u / %10u\n"
439 "pack_report: pack_mapped = "
440 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
443 pack_open_windows
, peak_pack_open_windows
,
444 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
447 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
450 struct pack_idx_header
*hdr
;
452 uint32_t version
, nr
, i
, *index
;
453 int fd
= open(path
, O_RDONLY
);
458 if (fstat(fd
, &st
)) {
462 idx_size
= xsize_t(st
.st_size
);
463 if (idx_size
< 4 * 256 + 20 + 20) {
465 return error("index file %s is too small", path
);
467 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
471 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
472 version
= ntohl(hdr
->idx_version
);
473 if (version
< 2 || version
> 2) {
474 munmap(idx_map
, idx_size
);
475 return error("index file %s is version %d"
476 " and is not supported by this binary"
477 " (try upgrading GIT to a newer version)",
486 index
+= 2; /* skip index header */
487 for (i
= 0; i
< 256; i
++) {
488 uint32_t n
= ntohl(index
[i
]);
490 munmap(idx_map
, idx_size
);
491 return error("non-monotonic index %s", path
);
499 * - 256 index entries 4 bytes each
500 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
501 * - 20-byte SHA1 of the packfile
502 * - 20-byte SHA1 file checksum
504 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
505 munmap(idx_map
, idx_size
);
506 return error("wrong index v1 file size in %s", path
);
508 } else if (version
== 2) {
511 * - 8 bytes of header
512 * - 256 index entries 4 bytes each
513 * - 20-byte sha1 entry * nr
514 * - 4-byte crc entry * nr
515 * - 4-byte offset entry * nr
516 * - 20-byte SHA1 of the packfile
517 * - 20-byte SHA1 file checksum
518 * And after the 4-byte offset table might be a
519 * variable sized table containing 8-byte entries
520 * for offsets larger than 2^31.
522 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
523 unsigned long max_size
= min_size
;
525 max_size
+= (nr
- 1)*8;
526 if (idx_size
< min_size
|| idx_size
> max_size
) {
527 munmap(idx_map
, idx_size
);
528 return error("wrong index v2 file size in %s", path
);
530 if (idx_size
!= min_size
&&
532 * make sure we can deal with large pack offsets.
533 * 31-bit signed offset won't be enough, neither
534 * 32-bit unsigned one will be.
536 (sizeof(off_t
) <= 4)) {
537 munmap(idx_map
, idx_size
);
538 return error("pack too large for current definition of off_t in %s", path
);
542 p
->index_version
= version
;
543 p
->index_data
= idx_map
;
544 p
->index_size
= idx_size
;
549 int open_pack_index(struct packed_git
*p
)
557 idx_name
= xstrdup(p
->pack_name
);
558 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
559 ret
= check_packed_git_idx(idx_name
, p
);
564 static void scan_windows(struct packed_git
*p
,
565 struct packed_git
**lru_p
,
566 struct pack_window
**lru_w
,
567 struct pack_window
**lru_l
)
569 struct pack_window
*w
, *w_l
;
571 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
573 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
583 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
585 struct packed_git
*p
, *lru_p
= NULL
;
586 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
589 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
590 for (p
= packed_git
; p
; p
= p
->next
)
591 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
593 munmap(lru_w
->base
, lru_w
->len
);
594 pack_mapped
-= lru_w
->len
;
596 lru_l
->next
= lru_w
->next
;
598 lru_p
->windows
= lru_w
->next
;
599 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
600 close(lru_p
->pack_fd
);
611 void release_pack_memory(size_t need
, int fd
)
613 size_t cur
= pack_mapped
;
614 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
618 void close_pack_windows(struct packed_git
*p
)
621 struct pack_window
*w
= p
->windows
;
624 die("pack '%s' still has open windows to it",
626 munmap(w
->base
, w
->len
);
627 pack_mapped
-= w
->len
;
629 p
->windows
= w
->next
;
634 void unuse_pack(struct pack_window
**w_cursor
)
636 struct pack_window
*w
= *w_cursor
;
644 * Do not call this directly as this leaks p->pack_fd on error return;
645 * call open_packed_git() instead.
647 static int open_packed_git_1(struct packed_git
*p
)
650 struct pack_header hdr
;
651 unsigned char sha1
[20];
652 unsigned char *idx_sha1
;
655 if (!p
->index_data
&& open_pack_index(p
))
656 return error("packfile %s index unavailable", p
->pack_name
);
658 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
659 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
662 /* If we created the struct before we had the pack we lack size. */
664 if (!S_ISREG(st
.st_mode
))
665 return error("packfile %s not a regular file", p
->pack_name
);
666 p
->pack_size
= st
.st_size
;
667 } else if (p
->pack_size
!= st
.st_size
)
668 return error("packfile %s size changed", p
->pack_name
);
670 /* We leave these file descriptors open with sliding mmap;
671 * there is no point keeping them open across exec(), though.
673 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
675 return error("cannot determine file descriptor flags");
676 fd_flag
|= FD_CLOEXEC
;
677 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
678 return error("cannot set FD_CLOEXEC");
680 /* Verify we recognize this pack file format. */
681 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
682 return error("file %s is far too short to be a packfile", p
->pack_name
);
683 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
684 return error("file %s is not a GIT packfile", p
->pack_name
);
685 if (!pack_version_ok(hdr
.hdr_version
))
686 return error("packfile %s is version %u and not supported"
687 " (try upgrading GIT to a newer version)",
688 p
->pack_name
, ntohl(hdr
.hdr_version
));
690 /* Verify the pack matches its index. */
691 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
692 return error("packfile %s claims to have %u objects"
693 " while index indicates %u objects",
694 p
->pack_name
, ntohl(hdr
.hdr_entries
),
696 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
697 return error("end of packfile %s is unavailable", p
->pack_name
);
698 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
699 return error("packfile %s signature is unavailable", p
->pack_name
);
700 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
701 if (hashcmp(sha1
, idx_sha1
))
702 return error("packfile %s does not match index", p
->pack_name
);
706 static int open_packed_git(struct packed_git
*p
)
708 if (!open_packed_git_1(p
))
710 if (p
->pack_fd
!= -1) {
717 static int in_window(struct pack_window
*win
, off_t offset
)
719 /* We must promise at least 20 bytes (one hash) after the
720 * offset is available from this window, otherwise the offset
721 * is not actually in this window and a different window (which
722 * has that one hash excess) must be used. This is to support
723 * the object header and delta base parsing routines below.
725 off_t win_off
= win
->offset
;
726 return win_off
<= offset
727 && (offset
+ 20) <= (win_off
+ win
->len
);
730 unsigned char* use_pack(struct packed_git
*p
,
731 struct pack_window
**w_cursor
,
735 struct pack_window
*win
= *w_cursor
;
737 if (p
->pack_fd
== -1 && open_packed_git(p
))
738 die("packfile %s cannot be accessed", p
->pack_name
);
740 /* Since packfiles end in a hash of their content and its
741 * pointless to ask for an offset into the middle of that
742 * hash, and the in_window function above wouldn't match
743 * don't allow an offset too close to the end of the file.
745 if (offset
> (p
->pack_size
- 20))
746 die("offset beyond end of packfile (truncated pack?)");
748 if (!win
|| !in_window(win
, offset
)) {
751 for (win
= p
->windows
; win
; win
= win
->next
) {
752 if (in_window(win
, offset
))
756 size_t window_align
= packed_git_window_size
/ 2;
758 win
= xcalloc(1, sizeof(*win
));
759 win
->offset
= (offset
/ window_align
) * window_align
;
760 len
= p
->pack_size
- win
->offset
;
761 if (len
> packed_git_window_size
)
762 len
= packed_git_window_size
;
763 win
->len
= (size_t)len
;
764 pack_mapped
+= win
->len
;
765 while (packed_git_limit
< pack_mapped
766 && unuse_one_window(p
, p
->pack_fd
))
768 win
->base
= xmmap(NULL
, win
->len
,
769 PROT_READ
, MAP_PRIVATE
,
770 p
->pack_fd
, win
->offset
);
771 if (win
->base
== MAP_FAILED
)
772 die("packfile %s cannot be mapped: %s",
777 if (pack_mapped
> peak_pack_mapped
)
778 peak_pack_mapped
= pack_mapped
;
779 if (pack_open_windows
> peak_pack_open_windows
)
780 peak_pack_open_windows
= pack_open_windows
;
781 win
->next
= p
->windows
;
785 if (win
!= *w_cursor
) {
786 win
->last_used
= pack_used_ctr
++;
790 offset
-= win
->offset
;
792 *left
= win
->len
- xsize_t(offset
);
793 return win
->base
+ offset
;
796 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
799 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
802 * Make sure a corresponding .pack file exists and that
803 * the index looks sane.
805 path_len
-= strlen(".idx");
808 memcpy(p
->pack_name
, path
, path_len
);
809 strcpy(p
->pack_name
+ path_len
, ".pack");
810 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
815 /* ok, it looks sane as far as we can check without
816 * actually mapping the pack file.
818 p
->index_version
= 0;
819 p
->index_data
= NULL
;
822 p
->pack_size
= st
.st_size
;
826 p
->pack_local
= local
;
827 p
->mtime
= st
.st_mtime
;
828 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
833 struct packed_git
*parse_pack_index(unsigned char *sha1
)
835 char *path
= sha1_pack_index_name(sha1
);
836 return parse_pack_index_file(sha1
, path
);
839 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
840 const char *idx_path
)
842 const char *path
= sha1_pack_name(sha1
);
843 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
845 if (check_packed_git_idx(idx_path
, p
)) {
850 strcpy(p
->pack_name
, path
);
855 hashcpy(p
->sha1
, sha1
);
859 void install_packed_git(struct packed_git
*pack
)
861 pack
->next
= packed_git
;
865 static void prepare_packed_git_one(char *objdir
, int local
)
867 /* Ensure that this buffer is large enough so that we can
868 append "/pack/" without clobbering the stack even if
869 strlen(objdir) were PATH_MAX. */
870 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
875 sprintf(path
, "%s/pack", objdir
);
880 error("unable to open object pack directory: %s: %s",
881 path
, strerror(errno
));
885 while ((de
= readdir(dir
)) != NULL
) {
886 int namelen
= strlen(de
->d_name
);
887 struct packed_git
*p
;
889 if (!has_extension(de
->d_name
, ".idx"))
892 if (len
+ namelen
+ 1 > sizeof(path
))
895 /* Don't reopen a pack we already have. */
896 strcpy(path
+ len
, de
->d_name
);
897 for (p
= packed_git
; p
; p
= p
->next
) {
898 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
903 /* See if it really is a valid .idx file with corresponding
904 * .pack file that we can map.
906 p
= add_packed_git(path
, len
+ namelen
, local
);
909 install_packed_git(p
);
914 static int sort_pack(const void *a_
, const void *b_
)
916 struct packed_git
*a
= *((struct packed_git
**)a_
);
917 struct packed_git
*b
= *((struct packed_git
**)b_
);
921 * Local packs tend to contain objects specific to our
922 * variant of the project than remote ones. In addition,
923 * remote ones could be on a network mounted filesystem.
924 * Favor local ones for these reasons.
926 st
= a
->pack_local
- b
->pack_local
;
931 * Younger packs tend to contain more recent objects,
932 * and more recent objects tend to get accessed more
935 if (a
->mtime
< b
->mtime
)
937 else if (a
->mtime
== b
->mtime
)
942 static void rearrange_packed_git(void)
944 struct packed_git
**ary
, *p
;
947 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
952 /* prepare an array of packed_git for easier sorting */
953 ary
= xcalloc(n
, sizeof(struct packed_git
*));
954 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
957 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
959 /* link them back again */
960 for (i
= 0; i
< n
- 1; i
++)
961 ary
[i
]->next
= ary
[i
+ 1];
962 ary
[n
- 1]->next
= NULL
;
968 static int prepare_packed_git_run_once
= 0;
969 void prepare_packed_git(void)
971 struct alternate_object_database
*alt
;
973 if (prepare_packed_git_run_once
)
975 prepare_packed_git_one(get_object_directory(), 1);
977 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
979 prepare_packed_git_one(alt
->base
, 0);
982 rearrange_packed_git();
983 prepare_packed_git_run_once
= 1;
986 void reprepare_packed_git(void)
988 prepare_packed_git_run_once
= 0;
989 prepare_packed_git();
992 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
994 unsigned char real_sha1
[20];
995 hash_sha1_file(map
, size
, type
, real_sha1
);
996 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
999 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1004 char *filename
= find_sha1_file(sha1
, &st
);
1010 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
1012 /* See if it works without O_NOATIME */
1013 switch (sha1_file_open_flag
) {
1015 fd
= open(filename
, O_RDONLY
);
1023 /* If it failed once, it will probably fail again.
1024 * Stop using O_NOATIME
1026 sha1_file_open_flag
= 0;
1028 *size
= xsize_t(st
.st_size
);
1029 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1034 static int legacy_loose_object(unsigned char *map
)
1039 * Is it a zlib-compressed buffer? If so, the first byte
1040 * must be 0x78 (15-bit window size, deflated), and the
1041 * first 16-bit word is evenly divisible by 31
1043 word
= (map
[0] << 8) + map
[1];
1044 if (map
[0] == 0x78 && !(word
% 31))
1050 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1055 unsigned long used
= 0;
1058 *type
= (c
>> 4) & 7;
1064 if (sizeof(long) * 8 <= shift
)
1067 size
+= (c
& 0x7f) << shift
;
1074 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1076 unsigned long size
, used
;
1077 static const char valid_loose_object_type
[8] = {
1079 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1080 0, /* "delta" and others are invalid in a loose object */
1082 enum object_type type
;
1084 /* Get the data stream */
1085 memset(stream
, 0, sizeof(*stream
));
1086 stream
->next_in
= map
;
1087 stream
->avail_in
= mapsize
;
1088 stream
->next_out
= buffer
;
1089 stream
->avail_out
= bufsiz
;
1091 if (legacy_loose_object(map
)) {
1092 inflateInit(stream
);
1093 return inflate(stream
, 0);
1098 * There used to be a second loose object header format which
1099 * was meant to mimic the in-pack format, allowing for direct
1100 * copy of the object data. This format turned up not to be
1101 * really worth it and we don't write it any longer. But we
1102 * can still read it.
1104 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1105 if (!used
|| !valid_loose_object_type
[type
])
1110 /* Set up the stream for the rest.. */
1111 stream
->next_in
= map
;
1112 stream
->avail_in
= mapsize
;
1113 inflateInit(stream
);
1115 /* And generate the fake traditional header */
1116 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1117 typename(type
), size
);
1121 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1123 int bytes
= strlen(buffer
) + 1;
1124 unsigned char *buf
= xmalloc(1+size
);
1128 n
= stream
->total_out
- bytes
;
1131 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1133 if (bytes
<= size
) {
1135 * The above condition must be (bytes <= size), not
1136 * (bytes < size). In other words, even though we
1137 * expect no more output and set avail_out to zer0,
1138 * the input zlib stream may have bytes that express
1139 * "this concludes the stream", and we *do* want to
1142 * Otherwise we would not be able to test that we
1143 * consumed all the input to reach the expected size;
1144 * we also want to check that zlib tells us that all
1145 * went well with status == Z_STREAM_END at the end.
1147 stream
->next_out
= buf
+ bytes
;
1148 stream
->avail_out
= size
- bytes
;
1149 while (status
== Z_OK
)
1150 status
= inflate(stream
, Z_FINISH
);
1153 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1159 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1160 else if (stream
->avail_in
)
1161 error("garbage at end of loose object '%s'",
1168 * We used to just use "sscanf()", but that's actually way
1169 * too permissive for what we want to check. So do an anal
1170 * object header parse by hand.
1172 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1179 * The type can be at most ten bytes (including the
1180 * terminating '\0' that we add), and is followed by
1189 if (i
>= sizeof(type
))
1195 * The length must follow immediately, and be in canonical
1196 * decimal format (ie "010" is not valid).
1198 size
= *hdr
++ - '0';
1203 unsigned long c
= *hdr
- '0';
1207 size
= size
* 10 + c
;
1213 * The length must be followed by a zero byte
1215 return *hdr
? -1 : type_from_string(type
);
1218 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1224 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1225 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1228 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1231 unsigned long get_size_from_delta(struct packed_git
*p
,
1232 struct pack_window
**w_curs
,
1235 const unsigned char *data
;
1236 unsigned char delta_head
[20], *in
;
1240 memset(&stream
, 0, sizeof(stream
));
1241 stream
.next_out
= delta_head
;
1242 stream
.avail_out
= sizeof(delta_head
);
1244 inflateInit(&stream
);
1246 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1247 stream
.next_in
= in
;
1248 st
= inflate(&stream
, Z_FINISH
);
1249 curpos
+= stream
.next_in
- in
;
1250 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1251 stream
.total_out
< sizeof(delta_head
));
1252 inflateEnd(&stream
);
1253 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1254 die("delta data unpack-initial failed");
1256 /* Examine the initial part of the delta to figure out
1261 /* ignore base size */
1262 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1264 /* Read the result size */
1265 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1268 static off_t
get_delta_base(struct packed_git
*p
,
1269 struct pack_window
**w_curs
,
1271 enum object_type type
,
1272 off_t delta_obj_offset
)
1274 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1277 /* use_pack() assured us we have [base_info, base_info + 20)
1278 * as a range that we can look at without walking off the
1279 * end of the mapped window. Its actually the hash size
1280 * that is assured. An OFS_DELTA longer than the hash size
1281 * is stupid, as then a REF_DELTA would be smaller to store.
1283 if (type
== OBJ_OFS_DELTA
) {
1285 unsigned char c
= base_info
[used
++];
1286 base_offset
= c
& 127;
1289 if (!base_offset
|| MSB(base_offset
, 7))
1290 die("offset value overflow for delta base object");
1291 c
= base_info
[used
++];
1292 base_offset
= (base_offset
<< 7) + (c
& 127);
1294 base_offset
= delta_obj_offset
- base_offset
;
1295 if (base_offset
>= delta_obj_offset
)
1296 die("delta base offset out of bound");
1298 } else if (type
== OBJ_REF_DELTA
) {
1299 /* The base entry _must_ be in the same pack */
1300 base_offset
= find_pack_entry_one(base_info
, p
);
1302 die("failed to find delta-pack base object %s",
1303 sha1_to_hex(base_info
));
1306 die("I am totally screwed");
1310 /* forward declaration for a mutually recursive function */
1311 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1312 unsigned long *sizep
);
1314 static int packed_delta_info(struct packed_git
*p
,
1315 struct pack_window
**w_curs
,
1317 enum object_type type
,
1319 unsigned long *sizep
)
1323 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1324 type
= packed_object_info(p
, base_offset
, NULL
);
1326 /* We choose to only get the type of the base object and
1327 * ignore potentially corrupt pack file that expects the delta
1328 * based on a base with a wrong size. This saves tons of
1332 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1337 static int unpack_object_header(struct packed_git
*p
,
1338 struct pack_window
**w_curs
,
1340 unsigned long *sizep
)
1342 unsigned char *base
;
1345 enum object_type type
;
1347 /* use_pack() assures us we have [base, base + 20) available
1348 * as a range that we can look at at. (Its actually the hash
1349 * size that is assured.) With our object header encoding
1350 * the maximum deflated object size is 2^137, which is just
1351 * insane, so we know won't exceed what we have been given.
1353 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1354 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1356 die("object offset outside of pack file");
1362 const char *packed_object_info_detail(struct packed_git
*p
,
1364 unsigned long *size
,
1365 unsigned long *store_size
,
1366 unsigned int *delta_chain_length
,
1367 unsigned char *base_sha1
)
1369 struct pack_window
*w_curs
= NULL
;
1371 unsigned long dummy
;
1372 unsigned char *next_sha1
;
1373 enum object_type type
;
1374 struct revindex_entry
*revidx
;
1376 *delta_chain_length
= 0;
1377 curpos
= obj_offset
;
1378 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1380 revidx
= find_pack_revindex(p
, obj_offset
);
1381 *store_size
= revidx
[1].offset
- obj_offset
;
1386 die("pack %s contains unknown object type %d",
1387 p
->pack_name
, type
);
1392 unuse_pack(&w_curs
);
1393 return typename(type
);
1395 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1396 if (*delta_chain_length
== 0) {
1397 revidx
= find_pack_revindex(p
, obj_offset
);
1398 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1402 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1403 if (*delta_chain_length
== 0)
1404 hashcpy(base_sha1
, next_sha1
);
1405 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1408 (*delta_chain_length
)++;
1409 curpos
= obj_offset
;
1410 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1414 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1415 unsigned long *sizep
)
1417 struct pack_window
*w_curs
= NULL
;
1419 off_t curpos
= obj_offset
;
1420 enum object_type type
;
1422 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1427 type
= packed_delta_info(p
, &w_curs
, curpos
,
1428 type
, obj_offset
, sizep
);
1438 die("pack %s contains unknown object type %d",
1439 p
->pack_name
, type
);
1441 unuse_pack(&w_curs
);
1445 static void *unpack_compressed_entry(struct packed_git
*p
,
1446 struct pack_window
**w_curs
,
1452 unsigned char *buffer
, *in
;
1454 buffer
= xmalloc(size
+ 1);
1456 memset(&stream
, 0, sizeof(stream
));
1457 stream
.next_out
= buffer
;
1458 stream
.avail_out
= size
;
1460 inflateInit(&stream
);
1462 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1463 stream
.next_in
= in
;
1464 st
= inflate(&stream
, Z_FINISH
);
1465 curpos
+= stream
.next_in
- in
;
1466 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1467 inflateEnd(&stream
);
1468 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1476 #define MAX_DELTA_CACHE (256)
1478 static size_t delta_base_cached
;
1480 static struct delta_base_cache_lru_list
{
1481 struct delta_base_cache_lru_list
*prev
;
1482 struct delta_base_cache_lru_list
*next
;
1483 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1485 static struct delta_base_cache_entry
{
1486 struct delta_base_cache_lru_list lru
;
1488 struct packed_git
*p
;
1491 enum object_type type
;
1492 } delta_base_cache
[MAX_DELTA_CACHE
];
1494 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1498 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1499 hash
+= (hash
>> 8) + (hash
>> 16);
1500 return hash
% MAX_DELTA_CACHE
;
1503 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1504 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1507 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1508 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1511 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1512 goto found_cache_entry
;
1513 return unpack_entry(p
, base_offset
, type
, base_size
);
1518 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1519 ent
->lru
.prev
->next
= ent
->lru
.next
;
1520 delta_base_cached
-= ent
->size
;
1522 ret
= xmemdupz(ent
->data
, ent
->size
);
1525 *base_size
= ent
->size
;
1529 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1534 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1535 ent
->lru
.prev
->next
= ent
->lru
.next
;
1536 delta_base_cached
-= ent
->size
;
1540 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1541 void *base
, unsigned long base_size
, enum object_type type
)
1543 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1544 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1545 struct delta_base_cache_lru_list
*lru
;
1547 release_delta_base_cache(ent
);
1548 delta_base_cached
+= base_size
;
1550 for (lru
= delta_base_cache_lru
.next
;
1551 delta_base_cached
> delta_base_cache_limit
1552 && lru
!= &delta_base_cache_lru
;
1554 struct delta_base_cache_entry
*f
= (void *)lru
;
1555 if (f
->type
== OBJ_BLOB
)
1556 release_delta_base_cache(f
);
1558 for (lru
= delta_base_cache_lru
.next
;
1559 delta_base_cached
> delta_base_cache_limit
1560 && lru
!= &delta_base_cache_lru
;
1562 struct delta_base_cache_entry
*f
= (void *)lru
;
1563 release_delta_base_cache(f
);
1567 ent
->base_offset
= base_offset
;
1570 ent
->size
= base_size
;
1571 ent
->lru
.next
= &delta_base_cache_lru
;
1572 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1573 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1574 delta_base_cache_lru
.prev
= &ent
->lru
;
1577 static void *unpack_delta_entry(struct packed_git
*p
,
1578 struct pack_window
**w_curs
,
1580 unsigned long delta_size
,
1582 enum object_type
*type
,
1583 unsigned long *sizep
)
1585 void *delta_data
, *result
, *base
;
1586 unsigned long base_size
;
1589 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1590 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1592 die("failed to read delta base object"
1593 " at %"PRIuMAX
" from %s",
1594 (uintmax_t)base_offset
, p
->pack_name
);
1596 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1598 die("failed to unpack compressed delta"
1599 " at %"PRIuMAX
" from %s",
1600 (uintmax_t)curpos
, p
->pack_name
);
1601 result
= patch_delta(base
, base_size
,
1602 delta_data
, delta_size
,
1605 die("failed to apply delta");
1607 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1611 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1612 enum object_type
*type
, unsigned long *sizep
)
1614 struct pack_window
*w_curs
= NULL
;
1615 off_t curpos
= obj_offset
;
1618 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1622 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1623 obj_offset
, type
, sizep
);
1629 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1632 die("unknown object type %i in %s", *type
, p
->pack_name
);
1634 unuse_pack(&w_curs
);
1638 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1641 const unsigned char *index
= p
->index_data
;
1643 if (open_pack_index(p
))
1645 index
= p
->index_data
;
1647 if (n
>= p
->num_objects
)
1650 if (p
->index_version
== 1) {
1651 return index
+ 24 * n
+ 4;
1654 return index
+ 20 * n
;
1658 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1660 const unsigned char *index
= p
->index_data
;
1662 if (p
->index_version
== 1) {
1663 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1666 index
+= 8 + p
->num_objects
* (20 + 4);
1667 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1668 if (!(off
& 0x80000000))
1670 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1671 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1672 ntohl(*((uint32_t *)(index
+ 4)));
1676 off_t
find_pack_entry_one(const unsigned char *sha1
,
1677 struct packed_git
*p
)
1679 const uint32_t *level1_ofs
= p
->index_data
;
1680 const unsigned char *index
= p
->index_data
;
1681 unsigned hi
, lo
, stride
;
1682 static int use_lookup
= -1;
1683 static int debug_lookup
= -1;
1685 if (debug_lookup
< 0)
1686 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1689 if (open_pack_index(p
))
1691 level1_ofs
= p
->index_data
;
1692 index
= p
->index_data
;
1694 if (p
->index_version
> 1) {
1699 hi
= ntohl(level1_ofs
[*sha1
]);
1700 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1701 if (p
->index_version
> 1) {
1709 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1710 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1713 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1715 int pos
= sha1_entry_pos(index
, stride
, 0,
1716 lo
, hi
, p
->num_objects
, sha1
);
1719 return nth_packed_object_offset(p
, pos
);
1723 unsigned mi
= (lo
+ hi
) / 2;
1724 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1727 printf("lo %u hi %u rg %u mi %u\n",
1728 lo
, hi
, hi
- lo
, mi
);
1730 return nth_packed_object_offset(p
, mi
);
1739 int matches_pack_name(struct packed_git
*p
, const char *name
)
1741 const char *last_c
, *c
;
1743 if (!strcmp(p
->pack_name
, name
))
1746 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1751 if (!strcmp(last_c
, name
))
1757 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1759 static struct packed_git
*last_found
= (void *)1;
1760 struct packed_git
*p
;
1763 prepare_packed_git();
1766 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1769 if (ignore_packed
) {
1771 for (ig
= ignore_packed
; *ig
; ig
++)
1772 if (matches_pack_name(p
, *ig
))
1778 offset
= find_pack_entry_one(sha1
, p
);
1781 * We are about to tell the caller where they can
1782 * locate the requested object. We better make
1783 * sure the packfile is still here and can be
1784 * accessed before supplying that answer, as
1785 * it may have been deleted since the index
1788 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1789 error("packfile %s cannot be accessed", p
->pack_name
);
1794 hashcpy(e
->sha1
, sha1
);
1800 if (p
== last_found
)
1804 if (p
== last_found
)
1810 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1811 struct packed_git
*packs
)
1813 struct packed_git
*p
;
1815 for (p
= packs
; p
; p
= p
->next
) {
1816 if (find_pack_entry_one(sha1
, p
))
1823 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1826 unsigned long mapsize
, size
;
1831 map
= map_sha1_file(sha1
, &mapsize
);
1833 return error("unable to find %s", sha1_to_hex(sha1
));
1834 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1835 status
= error("unable to unpack %s header",
1837 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1838 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1841 inflateEnd(&stream
);
1842 munmap(map
, mapsize
);
1846 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1848 struct pack_entry e
;
1850 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1851 reprepare_packed_git();
1852 if (!find_pack_entry(sha1
, &e
, NULL
))
1853 return sha1_loose_object_info(sha1
, sizep
);
1855 return packed_object_info(e
.p
, e
.offset
, sizep
);
1858 static void *read_packed_sha1(const unsigned char *sha1
,
1859 enum object_type
*type
, unsigned long *size
)
1861 struct pack_entry e
;
1863 if (!find_pack_entry(sha1
, &e
, NULL
))
1866 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1870 * This is meant to hold a *small* number of objects that you would
1871 * want read_sha1_file() to be able to return, but yet you do not want
1872 * to write them into the object store (e.g. a browse-only
1875 static struct cached_object
{
1876 unsigned char sha1
[20];
1877 enum object_type type
;
1881 static int cached_object_nr
, cached_object_alloc
;
1883 static struct cached_object empty_tree
= {
1884 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1885 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1886 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1892 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1895 struct cached_object
*co
= cached_objects
;
1897 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1898 if (!hashcmp(co
->sha1
, sha1
))
1901 if (!hashcmp(sha1
, empty_tree
.sha1
))
1906 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1907 unsigned char *sha1
)
1909 struct cached_object
*co
;
1911 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1912 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1914 if (cached_object_alloc
<= cached_object_nr
) {
1915 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1916 cached_objects
= xrealloc(cached_objects
,
1917 sizeof(*cached_objects
) *
1918 cached_object_alloc
);
1920 co
= &cached_objects
[cached_object_nr
++];
1923 co
->buf
= xmalloc(len
);
1924 memcpy(co
->buf
, buf
, len
);
1925 hashcpy(co
->sha1
, sha1
);
1929 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1930 unsigned long *size
)
1932 unsigned long mapsize
;
1934 struct cached_object
*co
;
1936 co
= find_cached_object(sha1
);
1940 return xmemdupz(co
->buf
, co
->size
);
1943 buf
= read_packed_sha1(sha1
, type
, size
);
1946 map
= map_sha1_file(sha1
, &mapsize
);
1948 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1949 munmap(map
, mapsize
);
1952 reprepare_packed_git();
1953 return read_packed_sha1(sha1
, type
, size
);
1956 void *read_object_with_reference(const unsigned char *sha1
,
1957 const char *required_type_name
,
1958 unsigned long *size
,
1959 unsigned char *actual_sha1_return
)
1961 enum object_type type
, required_type
;
1963 unsigned long isize
;
1964 unsigned char actual_sha1
[20];
1966 required_type
= type_from_string(required_type_name
);
1967 hashcpy(actual_sha1
, sha1
);
1969 int ref_length
= -1;
1970 const char *ref_type
= NULL
;
1972 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1975 if (type
== required_type
) {
1977 if (actual_sha1_return
)
1978 hashcpy(actual_sha1_return
, actual_sha1
);
1981 /* Handle references */
1982 else if (type
== OBJ_COMMIT
)
1984 else if (type
== OBJ_TAG
)
1985 ref_type
= "object ";
1990 ref_length
= strlen(ref_type
);
1992 if (ref_length
+ 40 > isize
||
1993 memcmp(buffer
, ref_type
, ref_length
) ||
1994 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1999 /* Now we have the ID of the referred-to object in
2000 * actual_sha1. Check again. */
2004 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2005 const char *type
, unsigned char *sha1
,
2006 char *hdr
, int *hdrlen
)
2010 /* Generate the header */
2011 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2015 SHA1_Update(&c
, hdr
, *hdrlen
);
2016 SHA1_Update(&c
, buf
, len
);
2017 SHA1_Final(sha1
, &c
);
2021 * Link the tempfile to the final place, possibly creating the
2022 * last directory level as you do so.
2024 * Returns the errno on failure, 0 on success.
2026 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
2031 if (!link(tmpfile
, filename
))
2035 * Try to mkdir the last path component if that failed.
2037 * Re-try the "link()" regardless of whether the mkdir
2038 * succeeds, since a race might mean that somebody
2042 dir
= strrchr(filename
, '/');
2045 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
2050 if (!link(tmpfile
, filename
))
2058 * Move the just written object into its final resting place
2060 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2062 int ret
= link_temp_to_file(tmpfile
, filename
);
2065 * Coda hack - coda doesn't like cross-directory links,
2066 * so we fall back to a rename, which will mean that it
2067 * won't be able to check collisions, but that's not a
2070 * The same holds for FAT formatted media.
2072 * When this succeeds, we just return 0. We have nothing
2075 if (ret
&& ret
!= EEXIST
) {
2076 if (!rename(tmpfile
, filename
))
2082 if (ret
!= EEXIST
) {
2083 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2085 /* FIXME!!! Collision check here ? */
2091 static int write_buffer(int fd
, const void *buf
, size_t len
)
2093 if (write_in_full(fd
, buf
, len
) < 0)
2094 return error("file write error (%s)", strerror(errno
));
2098 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2099 unsigned char *sha1
)
2103 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2107 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2108 void *buf
, unsigned long len
, time_t mtime
)
2111 unsigned char *compressed
;
2114 static char tmpfile
[PATH_MAX
];
2116 filename
= sha1_file_name(sha1
);
2117 fd
= open(filename
, O_RDONLY
);
2120 * FIXME!!! We might do collision checking here, but we'd
2121 * need to uncompress the old file and check it. Later.
2127 if (errno
!= ENOENT
) {
2128 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2131 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2133 fd
= mkstemp(tmpfile
);
2136 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2138 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2142 memset(&stream
, 0, sizeof(stream
));
2143 deflateInit(&stream
, zlib_compression_level
);
2144 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2145 compressed
= xmalloc(size
);
2148 stream
.next_out
= compressed
;
2149 stream
.avail_out
= size
;
2151 /* First header.. */
2152 stream
.next_in
= (unsigned char *)hdr
;
2153 stream
.avail_in
= hdrlen
;
2154 while (deflate(&stream
, 0) == Z_OK
)
2157 /* Then the data itself.. */
2158 stream
.next_in
= buf
;
2159 stream
.avail_in
= len
;
2160 ret
= deflate(&stream
, Z_FINISH
);
2161 if (ret
!= Z_STREAM_END
)
2162 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2164 ret
= deflateEnd(&stream
);
2166 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2168 size
= stream
.total_out
;
2170 if (write_buffer(fd
, compressed
, size
) < 0)
2171 die("unable to write sha1 file");
2174 die("unable to write sha1 file");
2180 utb
.modtime
= mtime
;
2181 if (utime(tmpfile
, &utb
) < 0)
2182 warning("failed utime() on %s: %s",
2183 tmpfile
, strerror(errno
));
2186 return move_temp_to_file(tmpfile
, filename
);
2189 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2191 unsigned char sha1
[20];
2195 /* Normally if we have it in the pack then we do not bother writing
2196 * it out into .git/objects/??/?{38} file.
2198 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2200 hashcpy(returnsha1
, sha1
);
2201 if (has_sha1_file(sha1
))
2203 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2206 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2211 enum object_type type
;
2215 if (find_sha1_file(sha1
, &st
))
2217 buf
= read_packed_sha1(sha1
, &type
, &len
);
2219 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2220 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2221 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2225 * We need to unpack and recompress the object for writing
2226 * it out to a different file.
2228 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2232 unsigned char *unpacked
;
2234 enum object_type type
;
2239 /* need to unpack and recompress it by itself */
2240 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2242 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2244 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2247 memset(&stream
, 0, sizeof(stream
));
2248 deflateInit(&stream
, zlib_compression_level
);
2249 size
= deflateBound(&stream
, len
+ hdrlen
);
2250 buf
= xmalloc(size
);
2253 stream
.next_out
= buf
;
2254 stream
.avail_out
= size
;
2256 /* First header.. */
2257 stream
.next_in
= (void *)hdr
;
2258 stream
.avail_in
= hdrlen
;
2259 while (deflate(&stream
, 0) == Z_OK
)
2262 /* Then the data itself.. */
2263 stream
.next_in
= unpacked
;
2264 stream
.avail_in
= len
;
2265 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2267 deflateEnd(&stream
);
2270 *objsize
= stream
.total_out
;
2274 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2277 unsigned long objsize
;
2278 void *buf
= map_sha1_file(sha1
, &objsize
);
2281 retval
= write_buffer(fd
, buf
, objsize
);
2282 munmap(buf
, objsize
);
2286 buf
= repack_object(sha1
, &objsize
);
2287 retval
= write_buffer(fd
, buf
, objsize
);
2292 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2293 size_t bufsize
, size_t *bufposn
)
2295 char tmpfile
[PATH_MAX
];
2298 unsigned char real_sha1
[20];
2299 unsigned char discard
[4096];
2303 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2305 local
= mkstemp(tmpfile
);
2308 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2310 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2313 memset(&stream
, 0, sizeof(stream
));
2315 inflateInit(&stream
);
2322 stream
.avail_in
= *bufposn
;
2323 stream
.next_in
= (unsigned char *) buffer
;
2325 stream
.next_out
= discard
;
2326 stream
.avail_out
= sizeof(discard
);
2327 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2328 SHA1_Update(&c
, discard
, sizeof(discard
) -
2330 } while (stream
.avail_in
&& ret
== Z_OK
);
2331 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2332 die("unable to write sha1 file");
2333 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2335 *bufposn
= stream
.avail_in
;
2339 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2344 return error("Connection closed?");
2345 perror("Reading from connection");
2350 inflateEnd(&stream
);
2352 fchmod(local
, 0444);
2353 if (close(local
) != 0)
2354 die("unable to write sha1 file");
2355 SHA1_Final(real_sha1
, &c
);
2356 if (ret
!= Z_STREAM_END
) {
2358 return error("File %s corrupted", sha1_to_hex(sha1
));
2360 if (hashcmp(sha1
, real_sha1
)) {
2362 return error("File %s has bad hash", sha1_to_hex(sha1
));
2365 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2368 int has_pack_index(const unsigned char *sha1
)
2371 if (stat(sha1_pack_index_name(sha1
), &st
))
2376 int has_pack_file(const unsigned char *sha1
)
2379 if (stat(sha1_pack_name(sha1
), &st
))
2384 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2386 struct pack_entry e
;
2387 return find_pack_entry(sha1
, &e
, ignore_packed
);
2390 int has_sha1_file(const unsigned char *sha1
)
2393 struct pack_entry e
;
2395 if (find_pack_entry(sha1
, &e
, NULL
))
2397 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2400 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2405 strbuf_init(&buf
, 0);
2406 if (strbuf_read(&buf
, fd
, 4096) < 0) {
2407 strbuf_release(&buf
);
2414 ret
= write_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2416 ret
= hash_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2417 strbuf_release(&buf
);
2422 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2423 enum object_type type
, const char *path
)
2425 size_t size
= xsize_t(st
->st_size
);
2427 int ret
, re_allocated
= 0;
2430 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2437 * Convert blobs to git internal format
2439 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2441 strbuf_init(&nbuf
, 0);
2442 if (convert_to_git(path
, buf
, size
, &nbuf
,
2443 write_object
? safe_crlf
: 0)) {
2445 buf
= strbuf_detach(&nbuf
, &size
);
2451 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2453 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2463 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2469 switch (st
->st_mode
& S_IFMT
) {
2471 fd
= open(path
, O_RDONLY
);
2473 return error("open(\"%s\"): %s", path
,
2475 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2476 return error("%s: failed to insert into database",
2480 len
= xsize_t(st
->st_size
);
2481 target
= xmalloc(len
+ 1);
2482 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2483 char *errstr
= strerror(errno
);
2485 return error("readlink(\"%s\"): %s", path
,
2489 hash_sha1_file(target
, len
, blob_type
, sha1
);
2490 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2491 return error("%s: failed to insert into database",
2496 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2498 return error("%s: unsupported file type", path
);
2503 int read_pack_header(int fd
, struct pack_header
*header
)
2505 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2506 /* "eof before pack header was fully read" */
2507 return PH_ERROR_EOF
;
2509 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2510 /* "protocol error (pack signature mismatch detected)" */
2511 return PH_ERROR_PACK_SIGNATURE
;
2512 if (!pack_version_ok(header
->hdr_version
))
2513 /* "protocol error (pack version unsupported)" */
2514 return PH_ERROR_PROTOCOL
;