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 static inline int offset_1st_component(const char *path
)
90 if (has_dos_drive_prefix(path
))
91 return 2 + (path
[2] == '/');
95 int safe_create_leading_directories(char *path
)
97 char *pos
= path
+ offset_1st_component(path
);
101 pos
= strchr(pos
, '/');
105 if (!stat(path
, &st
)) {
107 if (!S_ISDIR(st
.st_mode
)) {
112 else if (mkdir(path
, 0777)) {
116 else if (adjust_shared_perm(path
)) {
125 char * sha1_to_hex(const unsigned char *sha1
)
128 static char hexbuffer
[4][50];
129 static const char hex
[] = "0123456789abcdef";
130 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
133 for (i
= 0; i
< 20; i
++) {
134 unsigned int val
= *sha1
++;
135 *buf
++ = hex
[val
>> 4];
136 *buf
++ = hex
[val
& 0xf];
143 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
146 for (i
= 0; i
< 20; i
++) {
147 static char hex
[] = "0123456789abcdef";
148 unsigned int val
= sha1
[i
];
149 char *pos
= pathbuf
+ i
*2 + (i
> 0);
150 *pos
++ = hex
[val
>> 4];
151 *pos
= hex
[val
& 0xf];
156 * NOTE! This returns a statically allocated buffer, so you have to be
157 * careful about using it. Do an "xstrdup()" if you need to save the
160 * Also note that this returns the location for creating. Reading
161 * SHA1 file can happen from any alternate directory listed in the
162 * DB_ENVIRONMENT environment variable if it is not found in
163 * the primary object database.
165 char *sha1_file_name(const unsigned char *sha1
)
167 static char *name
, *base
;
170 const char *sha1_file_directory
= get_object_directory();
171 int len
= strlen(sha1_file_directory
);
172 base
= xmalloc(len
+ 60);
173 memcpy(base
, sha1_file_directory
, len
);
174 memset(base
+len
, 0, 60);
177 name
= base
+ len
+ 1;
179 fill_sha1_path(name
, sha1
);
183 static char *sha1_get_pack_name(const unsigned char *sha1
,
184 char **name
, char **base
, const char *which
)
186 static const char hex
[] = "0123456789abcdef";
191 const char *sha1_file_directory
= get_object_directory();
192 int len
= strlen(sha1_file_directory
);
193 *base
= xmalloc(len
+ 60);
194 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
195 sha1_file_directory
, which
);
196 *name
= *base
+ len
+ 11;
201 for (i
= 0; i
< 20; i
++) {
202 unsigned int val
= *sha1
++;
203 *buf
++ = hex
[val
>> 4];
204 *buf
++ = hex
[val
& 0xf];
210 char *sha1_pack_name(const unsigned char *sha1
)
212 static char *name
, *base
;
214 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
217 char *sha1_pack_index_name(const unsigned char *sha1
)
219 static char *name
, *base
;
221 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
224 struct alternate_object_database
*alt_odb_list
;
225 static struct alternate_object_database
**alt_odb_tail
;
227 static void read_info_alternates(const char * alternates
, int depth
);
230 * Prepare alternate object database registry.
232 * The variable alt_odb_list points at the list of struct
233 * alternate_object_database. The elements on this list come from
234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
236 * whose contents is similar to that environment variable but can be
237 * LF separated. Its base points at a statically allocated buffer that
238 * contains "/the/directory/corresponding/to/.git/objects/...", while
239 * its name points just after the slash at the end of ".git/objects/"
240 * in the example above, and has enough space to hold 40-byte hex
241 * SHA1, an extra slash for the first level indirection, and the
244 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
247 const char *objdir
= get_object_directory();
248 struct alternate_object_database
*ent
;
249 struct alternate_object_database
*alt
;
250 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int entlen
= pfxlen
+ 43;
255 if (!is_absolute_path(entry
) && relative_base
) {
256 /* Relative alt-odb */
258 base_len
= strlen(relative_base
) + 1;
262 ent
= xmalloc(sizeof(*ent
) + entlen
);
264 if (!is_absolute_path(entry
) && relative_base
) {
265 memcpy(ent
->base
, relative_base
, base_len
- 1);
266 ent
->base
[base_len
- 1] = '/';
267 memcpy(ent
->base
+ base_len
, entry
, len
);
270 memcpy(ent
->base
, entry
, pfxlen
);
272 ent
->name
= ent
->base
+ pfxlen
+ 1;
273 ent
->base
[pfxlen
+ 3] = '/';
274 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
276 /* Detect cases where alternate disappeared */
277 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
278 error("object directory %s does not exist; "
279 "check .git/objects/info/alternates.",
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
288 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
289 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
294 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
299 /* add the alternate entry */
301 alt_odb_tail
= &(ent
->next
);
304 /* recursively add alternates */
305 read_info_alternates(ent
->base
, depth
+ 1);
307 ent
->base
[pfxlen
] = '/';
312 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
313 const char *relative_base
, int depth
)
315 const char *cp
, *last
;
318 error("%s: ignoring alternate object stores, nesting too deep.",
326 if (cp
< ep
&& *cp
== '#') {
327 while (cp
< ep
&& *cp
!= sep
)
332 while (cp
< ep
&& *cp
!= sep
)
335 if (!is_absolute_path(last
) && depth
) {
336 error("%s: ignoring relative alternate object store %s",
337 relative_base
, last
);
339 link_alt_odb_entry(last
, cp
- last
,
340 relative_base
, depth
);
343 while (cp
< ep
&& *cp
== sep
)
349 static void read_info_alternates(const char * relative_base
, int depth
)
354 const char alt_file_name
[] = "info/alternates";
355 /* Given that relative_base is no longer than PATH_MAX,
356 ensure that "path" has enough space to append "/", the
357 file name, "info/alternates", and a trailing NUL. */
358 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
361 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
362 fd
= open(path
, O_RDONLY
);
365 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
369 mapsz
= xsize_t(st
.st_size
);
370 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
373 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
378 void add_to_alternates_file(const char *reference
)
380 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
381 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), 1);
382 char *alt
= mkpath("%s/objects\n", reference
);
383 write_or_die(fd
, alt
, strlen(alt
));
384 if (commit_lock_file(lock
))
385 die("could not close alternates file");
387 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
390 void prepare_alt_odb(void)
397 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
400 alt_odb_tail
= &alt_odb_list
;
401 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
403 read_info_alternates(get_object_directory(), 0);
406 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
408 char *name
= sha1_file_name(sha1
);
409 struct alternate_object_database
*alt
;
414 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
416 fill_sha1_path(name
, sha1
);
417 if (!stat(alt
->base
, st
))
423 static unsigned int pack_used_ctr
;
424 static unsigned int pack_mmap_calls
;
425 static unsigned int peak_pack_open_windows
;
426 static unsigned int pack_open_windows
;
427 static size_t peak_pack_mapped
;
428 static size_t pack_mapped
;
429 struct packed_git
*packed_git
;
431 void pack_report(void)
434 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
435 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
436 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
437 sz_fmt(getpagesize()),
438 sz_fmt(packed_git_window_size
),
439 sz_fmt(packed_git_limit
));
441 "pack_report: pack_used_ctr = %10u\n"
442 "pack_report: pack_mmap_calls = %10u\n"
443 "pack_report: pack_open_windows = %10u / %10u\n"
444 "pack_report: pack_mapped = "
445 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
448 pack_open_windows
, peak_pack_open_windows
,
449 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
452 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
455 struct pack_idx_header
*hdr
;
457 uint32_t version
, nr
, i
, *index
;
458 int fd
= open(path
, O_RDONLY
);
463 if (fstat(fd
, &st
)) {
467 idx_size
= xsize_t(st
.st_size
);
468 if (idx_size
< 4 * 256 + 20 + 20) {
470 return error("index file %s is too small", path
);
472 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
476 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
477 version
= ntohl(hdr
->idx_version
);
478 if (version
< 2 || version
> 2) {
479 munmap(idx_map
, idx_size
);
480 return error("index file %s is version %d"
481 " and is not supported by this binary"
482 " (try upgrading GIT to a newer version)",
491 index
+= 2; /* skip index header */
492 for (i
= 0; i
< 256; i
++) {
493 uint32_t n
= ntohl(index
[i
]);
495 munmap(idx_map
, idx_size
);
496 return error("non-monotonic index %s", path
);
504 * - 256 index entries 4 bytes each
505 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
506 * - 20-byte SHA1 of the packfile
507 * - 20-byte SHA1 file checksum
509 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
510 munmap(idx_map
, idx_size
);
511 return error("wrong index v1 file size in %s", path
);
513 } else if (version
== 2) {
516 * - 8 bytes of header
517 * - 256 index entries 4 bytes each
518 * - 20-byte sha1 entry * nr
519 * - 4-byte crc entry * nr
520 * - 4-byte offset entry * nr
521 * - 20-byte SHA1 of the packfile
522 * - 20-byte SHA1 file checksum
523 * And after the 4-byte offset table might be a
524 * variable sized table containing 8-byte entries
525 * for offsets larger than 2^31.
527 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
528 unsigned long max_size
= min_size
;
530 max_size
+= (nr
- 1)*8;
531 if (idx_size
< min_size
|| idx_size
> max_size
) {
532 munmap(idx_map
, idx_size
);
533 return error("wrong index v2 file size in %s", path
);
535 if (idx_size
!= min_size
&&
537 * make sure we can deal with large pack offsets.
538 * 31-bit signed offset won't be enough, neither
539 * 32-bit unsigned one will be.
541 (sizeof(off_t
) <= 4)) {
542 munmap(idx_map
, idx_size
);
543 return error("pack too large for current definition of off_t in %s", path
);
547 p
->index_version
= version
;
548 p
->index_data
= idx_map
;
549 p
->index_size
= idx_size
;
554 int open_pack_index(struct packed_git
*p
)
562 idx_name
= xstrdup(p
->pack_name
);
563 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
564 ret
= check_packed_git_idx(idx_name
, p
);
569 static void scan_windows(struct packed_git
*p
,
570 struct packed_git
**lru_p
,
571 struct pack_window
**lru_w
,
572 struct pack_window
**lru_l
)
574 struct pack_window
*w
, *w_l
;
576 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
578 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
588 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
590 struct packed_git
*p
, *lru_p
= NULL
;
591 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
594 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
595 for (p
= packed_git
; p
; p
= p
->next
)
596 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
598 munmap(lru_w
->base
, lru_w
->len
);
599 pack_mapped
-= lru_w
->len
;
601 lru_l
->next
= lru_w
->next
;
603 lru_p
->windows
= lru_w
->next
;
604 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
605 close(lru_p
->pack_fd
);
616 void release_pack_memory(size_t need
, int fd
)
618 size_t cur
= pack_mapped
;
619 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
623 void close_pack_windows(struct packed_git
*p
)
626 struct pack_window
*w
= p
->windows
;
629 die("pack '%s' still has open windows to it",
631 munmap(w
->base
, w
->len
);
632 pack_mapped
-= w
->len
;
634 p
->windows
= w
->next
;
639 void unuse_pack(struct pack_window
**w_cursor
)
641 struct pack_window
*w
= *w_cursor
;
649 * Do not call this directly as this leaks p->pack_fd on error return;
650 * call open_packed_git() instead.
652 static int open_packed_git_1(struct packed_git
*p
)
655 struct pack_header hdr
;
656 unsigned char sha1
[20];
657 unsigned char *idx_sha1
;
660 if (!p
->index_data
&& open_pack_index(p
))
661 return error("packfile %s index unavailable", p
->pack_name
);
663 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
664 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
667 /* If we created the struct before we had the pack we lack size. */
669 if (!S_ISREG(st
.st_mode
))
670 return error("packfile %s not a regular file", p
->pack_name
);
671 p
->pack_size
= st
.st_size
;
672 } else if (p
->pack_size
!= st
.st_size
)
673 return error("packfile %s size changed", p
->pack_name
);
675 /* We leave these file descriptors open with sliding mmap;
676 * there is no point keeping them open across exec(), though.
678 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
680 return error("cannot determine file descriptor flags");
681 fd_flag
|= FD_CLOEXEC
;
682 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
683 return error("cannot set FD_CLOEXEC");
685 /* Verify we recognize this pack file format. */
686 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
687 return error("file %s is far too short to be a packfile", p
->pack_name
);
688 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
689 return error("file %s is not a GIT packfile", p
->pack_name
);
690 if (!pack_version_ok(hdr
.hdr_version
))
691 return error("packfile %s is version %u and not supported"
692 " (try upgrading GIT to a newer version)",
693 p
->pack_name
, ntohl(hdr
.hdr_version
));
695 /* Verify the pack matches its index. */
696 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
697 return error("packfile %s claims to have %u objects"
698 " while index indicates %u objects",
699 p
->pack_name
, ntohl(hdr
.hdr_entries
),
701 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
702 return error("end of packfile %s is unavailable", p
->pack_name
);
703 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
704 return error("packfile %s signature is unavailable", p
->pack_name
);
705 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
706 if (hashcmp(sha1
, idx_sha1
))
707 return error("packfile %s does not match index", p
->pack_name
);
711 static int open_packed_git(struct packed_git
*p
)
713 if (!open_packed_git_1(p
))
715 if (p
->pack_fd
!= -1) {
722 static int in_window(struct pack_window
*win
, off_t offset
)
724 /* We must promise at least 20 bytes (one hash) after the
725 * offset is available from this window, otherwise the offset
726 * is not actually in this window and a different window (which
727 * has that one hash excess) must be used. This is to support
728 * the object header and delta base parsing routines below.
730 off_t win_off
= win
->offset
;
731 return win_off
<= offset
732 && (offset
+ 20) <= (win_off
+ win
->len
);
735 unsigned char* use_pack(struct packed_git
*p
,
736 struct pack_window
**w_cursor
,
740 struct pack_window
*win
= *w_cursor
;
742 if (p
->pack_fd
== -1 && open_packed_git(p
))
743 die("packfile %s cannot be accessed", p
->pack_name
);
745 /* Since packfiles end in a hash of their content and its
746 * pointless to ask for an offset into the middle of that
747 * hash, and the in_window function above wouldn't match
748 * don't allow an offset too close to the end of the file.
750 if (offset
> (p
->pack_size
- 20))
751 die("offset beyond end of packfile (truncated pack?)");
753 if (!win
|| !in_window(win
, offset
)) {
756 for (win
= p
->windows
; win
; win
= win
->next
) {
757 if (in_window(win
, offset
))
761 size_t window_align
= packed_git_window_size
/ 2;
763 win
= xcalloc(1, sizeof(*win
));
764 win
->offset
= (offset
/ window_align
) * window_align
;
765 len
= p
->pack_size
- win
->offset
;
766 if (len
> packed_git_window_size
)
767 len
= packed_git_window_size
;
768 win
->len
= (size_t)len
;
769 pack_mapped
+= win
->len
;
770 while (packed_git_limit
< pack_mapped
771 && unuse_one_window(p
, p
->pack_fd
))
773 win
->base
= xmmap(NULL
, win
->len
,
774 PROT_READ
, MAP_PRIVATE
,
775 p
->pack_fd
, win
->offset
);
776 if (win
->base
== MAP_FAILED
)
777 die("packfile %s cannot be mapped: %s",
782 if (pack_mapped
> peak_pack_mapped
)
783 peak_pack_mapped
= pack_mapped
;
784 if (pack_open_windows
> peak_pack_open_windows
)
785 peak_pack_open_windows
= pack_open_windows
;
786 win
->next
= p
->windows
;
790 if (win
!= *w_cursor
) {
791 win
->last_used
= pack_used_ctr
++;
795 offset
-= win
->offset
;
797 *left
= win
->len
- xsize_t(offset
);
798 return win
->base
+ offset
;
801 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
804 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
807 * Make sure a corresponding .pack file exists and that
808 * the index looks sane.
810 path_len
-= strlen(".idx");
813 memcpy(p
->pack_name
, path
, path_len
);
814 strcpy(p
->pack_name
+ path_len
, ".pack");
815 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
820 /* ok, it looks sane as far as we can check without
821 * actually mapping the pack file.
823 p
->index_version
= 0;
824 p
->index_data
= NULL
;
827 p
->pack_size
= st
.st_size
;
831 p
->pack_local
= local
;
832 p
->mtime
= st
.st_mtime
;
833 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
838 struct packed_git
*parse_pack_index(unsigned char *sha1
)
840 char *path
= sha1_pack_index_name(sha1
);
841 return parse_pack_index_file(sha1
, path
);
844 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
845 const char *idx_path
)
847 const char *path
= sha1_pack_name(sha1
);
848 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
850 if (check_packed_git_idx(idx_path
, p
)) {
855 strcpy(p
->pack_name
, path
);
860 hashcpy(p
->sha1
, sha1
);
864 void install_packed_git(struct packed_git
*pack
)
866 pack
->next
= packed_git
;
870 static void prepare_packed_git_one(char *objdir
, int local
)
872 /* Ensure that this buffer is large enough so that we can
873 append "/pack/" without clobbering the stack even if
874 strlen(objdir) were PATH_MAX. */
875 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
880 sprintf(path
, "%s/pack", objdir
);
885 error("unable to open object pack directory: %s: %s",
886 path
, strerror(errno
));
890 while ((de
= readdir(dir
)) != NULL
) {
891 int namelen
= strlen(de
->d_name
);
892 struct packed_git
*p
;
894 if (!has_extension(de
->d_name
, ".idx"))
897 if (len
+ namelen
+ 1 > sizeof(path
))
900 /* Don't reopen a pack we already have. */
901 strcpy(path
+ len
, de
->d_name
);
902 for (p
= packed_git
; p
; p
= p
->next
) {
903 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
908 /* See if it really is a valid .idx file with corresponding
909 * .pack file that we can map.
911 p
= add_packed_git(path
, len
+ namelen
, local
);
914 install_packed_git(p
);
919 static int sort_pack(const void *a_
, const void *b_
)
921 struct packed_git
*a
= *((struct packed_git
**)a_
);
922 struct packed_git
*b
= *((struct packed_git
**)b_
);
926 * Local packs tend to contain objects specific to our
927 * variant of the project than remote ones. In addition,
928 * remote ones could be on a network mounted filesystem.
929 * Favor local ones for these reasons.
931 st
= a
->pack_local
- b
->pack_local
;
936 * Younger packs tend to contain more recent objects,
937 * and more recent objects tend to get accessed more
940 if (a
->mtime
< b
->mtime
)
942 else if (a
->mtime
== b
->mtime
)
947 static void rearrange_packed_git(void)
949 struct packed_git
**ary
, *p
;
952 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
957 /* prepare an array of packed_git for easier sorting */
958 ary
= xcalloc(n
, sizeof(struct packed_git
*));
959 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
962 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
964 /* link them back again */
965 for (i
= 0; i
< n
- 1; i
++)
966 ary
[i
]->next
= ary
[i
+ 1];
967 ary
[n
- 1]->next
= NULL
;
973 static int prepare_packed_git_run_once
= 0;
974 void prepare_packed_git(void)
976 struct alternate_object_database
*alt
;
978 if (prepare_packed_git_run_once
)
980 prepare_packed_git_one(get_object_directory(), 1);
982 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
984 prepare_packed_git_one(alt
->base
, 0);
987 rearrange_packed_git();
988 prepare_packed_git_run_once
= 1;
991 void reprepare_packed_git(void)
993 prepare_packed_git_run_once
= 0;
994 prepare_packed_git();
997 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
999 unsigned char real_sha1
[20];
1000 hash_sha1_file(map
, size
, type
, real_sha1
);
1001 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1004 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1009 char *filename
= find_sha1_file(sha1
, &st
);
1015 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
1017 /* See if it works without O_NOATIME */
1018 switch (sha1_file_open_flag
) {
1020 fd
= open(filename
, O_RDONLY
);
1028 /* If it failed once, it will probably fail again.
1029 * Stop using O_NOATIME
1031 sha1_file_open_flag
= 0;
1033 *size
= xsize_t(st
.st_size
);
1034 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1039 static int legacy_loose_object(unsigned char *map
)
1044 * Is it a zlib-compressed buffer? If so, the first byte
1045 * must be 0x78 (15-bit window size, deflated), and the
1046 * first 16-bit word is evenly divisible by 31
1048 word
= (map
[0] << 8) + map
[1];
1049 if (map
[0] == 0x78 && !(word
% 31))
1055 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1060 unsigned long used
= 0;
1063 *type
= (c
>> 4) & 7;
1069 if (sizeof(long) * 8 <= shift
)
1072 size
+= (c
& 0x7f) << shift
;
1079 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1081 unsigned long size
, used
;
1082 static const char valid_loose_object_type
[8] = {
1084 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1085 0, /* "delta" and others are invalid in a loose object */
1087 enum object_type type
;
1089 /* Get the data stream */
1090 memset(stream
, 0, sizeof(*stream
));
1091 stream
->next_in
= map
;
1092 stream
->avail_in
= mapsize
;
1093 stream
->next_out
= buffer
;
1094 stream
->avail_out
= bufsiz
;
1096 if (legacy_loose_object(map
)) {
1097 inflateInit(stream
);
1098 return inflate(stream
, 0);
1103 * There used to be a second loose object header format which
1104 * was meant to mimic the in-pack format, allowing for direct
1105 * copy of the object data. This format turned up not to be
1106 * really worth it and we don't write it any longer. But we
1107 * can still read it.
1109 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1110 if (!used
|| !valid_loose_object_type
[type
])
1115 /* Set up the stream for the rest.. */
1116 stream
->next_in
= map
;
1117 stream
->avail_in
= mapsize
;
1118 inflateInit(stream
);
1120 /* And generate the fake traditional header */
1121 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1122 typename(type
), size
);
1126 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1128 int bytes
= strlen(buffer
) + 1;
1129 unsigned char *buf
= xmalloc(1+size
);
1133 n
= stream
->total_out
- bytes
;
1136 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1138 if (bytes
<= size
) {
1140 * The above condition must be (bytes <= size), not
1141 * (bytes < size). In other words, even though we
1142 * expect no more output and set avail_out to zer0,
1143 * the input zlib stream may have bytes that express
1144 * "this concludes the stream", and we *do* want to
1147 * Otherwise we would not be able to test that we
1148 * consumed all the input to reach the expected size;
1149 * we also want to check that zlib tells us that all
1150 * went well with status == Z_STREAM_END at the end.
1152 stream
->next_out
= buf
+ bytes
;
1153 stream
->avail_out
= size
- bytes
;
1154 while (status
== Z_OK
)
1155 status
= inflate(stream
, Z_FINISH
);
1158 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1164 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1165 else if (stream
->avail_in
)
1166 error("garbage at end of loose object '%s'",
1173 * We used to just use "sscanf()", but that's actually way
1174 * too permissive for what we want to check. So do an anal
1175 * object header parse by hand.
1177 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1184 * The type can be at most ten bytes (including the
1185 * terminating '\0' that we add), and is followed by
1194 if (i
>= sizeof(type
))
1200 * The length must follow immediately, and be in canonical
1201 * decimal format (ie "010" is not valid).
1203 size
= *hdr
++ - '0';
1208 unsigned long c
= *hdr
- '0';
1212 size
= size
* 10 + c
;
1218 * The length must be followed by a zero byte
1220 return *hdr
? -1 : type_from_string(type
);
1223 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1229 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1230 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1233 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1236 unsigned long get_size_from_delta(struct packed_git
*p
,
1237 struct pack_window
**w_curs
,
1240 const unsigned char *data
;
1241 unsigned char delta_head
[20], *in
;
1245 memset(&stream
, 0, sizeof(stream
));
1246 stream
.next_out
= delta_head
;
1247 stream
.avail_out
= sizeof(delta_head
);
1249 inflateInit(&stream
);
1251 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1252 stream
.next_in
= in
;
1253 st
= inflate(&stream
, Z_FINISH
);
1254 curpos
+= stream
.next_in
- in
;
1255 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1256 stream
.total_out
< sizeof(delta_head
));
1257 inflateEnd(&stream
);
1258 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1259 die("delta data unpack-initial failed");
1261 /* Examine the initial part of the delta to figure out
1266 /* ignore base size */
1267 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1269 /* Read the result size */
1270 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1273 static off_t
get_delta_base(struct packed_git
*p
,
1274 struct pack_window
**w_curs
,
1276 enum object_type type
,
1277 off_t delta_obj_offset
)
1279 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1282 /* use_pack() assured us we have [base_info, base_info + 20)
1283 * as a range that we can look at without walking off the
1284 * end of the mapped window. Its actually the hash size
1285 * that is assured. An OFS_DELTA longer than the hash size
1286 * is stupid, as then a REF_DELTA would be smaller to store.
1288 if (type
== OBJ_OFS_DELTA
) {
1290 unsigned char c
= base_info
[used
++];
1291 base_offset
= c
& 127;
1294 if (!base_offset
|| MSB(base_offset
, 7))
1295 die("offset value overflow for delta base object");
1296 c
= base_info
[used
++];
1297 base_offset
= (base_offset
<< 7) + (c
& 127);
1299 base_offset
= delta_obj_offset
- base_offset
;
1300 if (base_offset
>= delta_obj_offset
)
1301 die("delta base offset out of bound");
1303 } else if (type
== OBJ_REF_DELTA
) {
1304 /* The base entry _must_ be in the same pack */
1305 base_offset
= find_pack_entry_one(base_info
, p
);
1307 die("failed to find delta-pack base object %s",
1308 sha1_to_hex(base_info
));
1311 die("I am totally screwed");
1315 /* forward declaration for a mutually recursive function */
1316 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1317 unsigned long *sizep
);
1319 static int packed_delta_info(struct packed_git
*p
,
1320 struct pack_window
**w_curs
,
1322 enum object_type type
,
1324 unsigned long *sizep
)
1328 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1329 type
= packed_object_info(p
, base_offset
, NULL
);
1331 /* We choose to only get the type of the base object and
1332 * ignore potentially corrupt pack file that expects the delta
1333 * based on a base with a wrong size. This saves tons of
1337 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1342 static int unpack_object_header(struct packed_git
*p
,
1343 struct pack_window
**w_curs
,
1345 unsigned long *sizep
)
1347 unsigned char *base
;
1350 enum object_type type
;
1352 /* use_pack() assures us we have [base, base + 20) available
1353 * as a range that we can look at at. (Its actually the hash
1354 * size that is assured.) With our object header encoding
1355 * the maximum deflated object size is 2^137, which is just
1356 * insane, so we know won't exceed what we have been given.
1358 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1359 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1361 die("object offset outside of pack file");
1367 const char *packed_object_info_detail(struct packed_git
*p
,
1369 unsigned long *size
,
1370 unsigned long *store_size
,
1371 unsigned int *delta_chain_length
,
1372 unsigned char *base_sha1
)
1374 struct pack_window
*w_curs
= NULL
;
1376 unsigned long dummy
;
1377 unsigned char *next_sha1
;
1378 enum object_type type
;
1379 struct revindex_entry
*revidx
;
1381 *delta_chain_length
= 0;
1382 curpos
= obj_offset
;
1383 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1385 revidx
= find_pack_revindex(p
, obj_offset
);
1386 *store_size
= revidx
[1].offset
- obj_offset
;
1391 die("pack %s contains unknown object type %d",
1392 p
->pack_name
, type
);
1397 unuse_pack(&w_curs
);
1398 return typename(type
);
1400 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1401 if (*delta_chain_length
== 0) {
1402 revidx
= find_pack_revindex(p
, obj_offset
);
1403 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1407 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1408 if (*delta_chain_length
== 0)
1409 hashcpy(base_sha1
, next_sha1
);
1410 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1413 (*delta_chain_length
)++;
1414 curpos
= obj_offset
;
1415 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1419 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1420 unsigned long *sizep
)
1422 struct pack_window
*w_curs
= NULL
;
1424 off_t curpos
= obj_offset
;
1425 enum object_type type
;
1427 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1432 type
= packed_delta_info(p
, &w_curs
, curpos
,
1433 type
, obj_offset
, sizep
);
1443 die("pack %s contains unknown object type %d",
1444 p
->pack_name
, type
);
1446 unuse_pack(&w_curs
);
1450 static void *unpack_compressed_entry(struct packed_git
*p
,
1451 struct pack_window
**w_curs
,
1457 unsigned char *buffer
, *in
;
1459 buffer
= xmalloc(size
+ 1);
1461 memset(&stream
, 0, sizeof(stream
));
1462 stream
.next_out
= buffer
;
1463 stream
.avail_out
= size
;
1465 inflateInit(&stream
);
1467 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1468 stream
.next_in
= in
;
1469 st
= inflate(&stream
, Z_FINISH
);
1470 curpos
+= stream
.next_in
- in
;
1471 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1472 inflateEnd(&stream
);
1473 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1481 #define MAX_DELTA_CACHE (256)
1483 static size_t delta_base_cached
;
1485 static struct delta_base_cache_lru_list
{
1486 struct delta_base_cache_lru_list
*prev
;
1487 struct delta_base_cache_lru_list
*next
;
1488 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1490 static struct delta_base_cache_entry
{
1491 struct delta_base_cache_lru_list lru
;
1493 struct packed_git
*p
;
1496 enum object_type type
;
1497 } delta_base_cache
[MAX_DELTA_CACHE
];
1499 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1503 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1504 hash
+= (hash
>> 8) + (hash
>> 16);
1505 return hash
% MAX_DELTA_CACHE
;
1508 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1509 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1512 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1513 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1516 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1517 goto found_cache_entry
;
1518 return unpack_entry(p
, base_offset
, type
, base_size
);
1523 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1524 ent
->lru
.prev
->next
= ent
->lru
.next
;
1525 delta_base_cached
-= ent
->size
;
1527 ret
= xmemdupz(ent
->data
, ent
->size
);
1530 *base_size
= ent
->size
;
1534 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1539 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1540 ent
->lru
.prev
->next
= ent
->lru
.next
;
1541 delta_base_cached
-= ent
->size
;
1545 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1546 void *base
, unsigned long base_size
, enum object_type type
)
1548 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1549 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1550 struct delta_base_cache_lru_list
*lru
;
1552 release_delta_base_cache(ent
);
1553 delta_base_cached
+= base_size
;
1555 for (lru
= delta_base_cache_lru
.next
;
1556 delta_base_cached
> delta_base_cache_limit
1557 && lru
!= &delta_base_cache_lru
;
1559 struct delta_base_cache_entry
*f
= (void *)lru
;
1560 if (f
->type
== OBJ_BLOB
)
1561 release_delta_base_cache(f
);
1563 for (lru
= delta_base_cache_lru
.next
;
1564 delta_base_cached
> delta_base_cache_limit
1565 && lru
!= &delta_base_cache_lru
;
1567 struct delta_base_cache_entry
*f
= (void *)lru
;
1568 release_delta_base_cache(f
);
1572 ent
->base_offset
= base_offset
;
1575 ent
->size
= base_size
;
1576 ent
->lru
.next
= &delta_base_cache_lru
;
1577 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1578 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1579 delta_base_cache_lru
.prev
= &ent
->lru
;
1582 static void *unpack_delta_entry(struct packed_git
*p
,
1583 struct pack_window
**w_curs
,
1585 unsigned long delta_size
,
1587 enum object_type
*type
,
1588 unsigned long *sizep
)
1590 void *delta_data
, *result
, *base
;
1591 unsigned long base_size
;
1594 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1595 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1597 die("failed to read delta base object"
1598 " at %"PRIuMAX
" from %s",
1599 (uintmax_t)base_offset
, p
->pack_name
);
1601 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1603 die("failed to unpack compressed delta"
1604 " at %"PRIuMAX
" from %s",
1605 (uintmax_t)curpos
, p
->pack_name
);
1606 result
= patch_delta(base
, base_size
,
1607 delta_data
, delta_size
,
1610 die("failed to apply delta");
1612 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1616 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1617 enum object_type
*type
, unsigned long *sizep
)
1619 struct pack_window
*w_curs
= NULL
;
1620 off_t curpos
= obj_offset
;
1623 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1627 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1628 obj_offset
, type
, sizep
);
1634 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1637 die("unknown object type %i in %s", *type
, p
->pack_name
);
1639 unuse_pack(&w_curs
);
1643 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1646 const unsigned char *index
= p
->index_data
;
1648 if (open_pack_index(p
))
1650 index
= p
->index_data
;
1652 if (n
>= p
->num_objects
)
1655 if (p
->index_version
== 1) {
1656 return index
+ 24 * n
+ 4;
1659 return index
+ 20 * n
;
1663 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1665 const unsigned char *index
= p
->index_data
;
1667 if (p
->index_version
== 1) {
1668 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1671 index
+= 8 + p
->num_objects
* (20 + 4);
1672 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1673 if (!(off
& 0x80000000))
1675 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1676 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1677 ntohl(*((uint32_t *)(index
+ 4)));
1681 off_t
find_pack_entry_one(const unsigned char *sha1
,
1682 struct packed_git
*p
)
1684 const uint32_t *level1_ofs
= p
->index_data
;
1685 const unsigned char *index
= p
->index_data
;
1686 unsigned hi
, lo
, stride
;
1687 static int use_lookup
= -1;
1688 static int debug_lookup
= -1;
1690 if (debug_lookup
< 0)
1691 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1694 if (open_pack_index(p
))
1696 level1_ofs
= p
->index_data
;
1697 index
= p
->index_data
;
1699 if (p
->index_version
> 1) {
1704 hi
= ntohl(level1_ofs
[*sha1
]);
1705 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1706 if (p
->index_version
> 1) {
1714 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1715 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1718 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1720 int pos
= sha1_entry_pos(index
, stride
, 0,
1721 lo
, hi
, p
->num_objects
, sha1
);
1724 return nth_packed_object_offset(p
, pos
);
1728 unsigned mi
= (lo
+ hi
) / 2;
1729 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1732 printf("lo %u hi %u rg %u mi %u\n",
1733 lo
, hi
, hi
- lo
, mi
);
1735 return nth_packed_object_offset(p
, mi
);
1744 int matches_pack_name(struct packed_git
*p
, const char *name
)
1746 const char *last_c
, *c
;
1748 if (!strcmp(p
->pack_name
, name
))
1751 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1756 if (!strcmp(last_c
, name
))
1762 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1764 static struct packed_git
*last_found
= (void *)1;
1765 struct packed_git
*p
;
1768 prepare_packed_git();
1771 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1774 if (ignore_packed
) {
1776 for (ig
= ignore_packed
; *ig
; ig
++)
1777 if (matches_pack_name(p
, *ig
))
1783 offset
= find_pack_entry_one(sha1
, p
);
1786 * We are about to tell the caller where they can
1787 * locate the requested object. We better make
1788 * sure the packfile is still here and can be
1789 * accessed before supplying that answer, as
1790 * it may have been deleted since the index
1793 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1794 error("packfile %s cannot be accessed", p
->pack_name
);
1799 hashcpy(e
->sha1
, sha1
);
1805 if (p
== last_found
)
1809 if (p
== last_found
)
1815 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1816 struct packed_git
*packs
)
1818 struct packed_git
*p
;
1820 for (p
= packs
; p
; p
= p
->next
) {
1821 if (find_pack_entry_one(sha1
, p
))
1828 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1831 unsigned long mapsize
, size
;
1836 map
= map_sha1_file(sha1
, &mapsize
);
1838 return error("unable to find %s", sha1_to_hex(sha1
));
1839 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1840 status
= error("unable to unpack %s header",
1842 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1843 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1846 inflateEnd(&stream
);
1847 munmap(map
, mapsize
);
1851 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1853 struct pack_entry e
;
1855 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1856 reprepare_packed_git();
1857 if (!find_pack_entry(sha1
, &e
, NULL
))
1858 return sha1_loose_object_info(sha1
, sizep
);
1860 return packed_object_info(e
.p
, e
.offset
, sizep
);
1863 static void *read_packed_sha1(const unsigned char *sha1
,
1864 enum object_type
*type
, unsigned long *size
)
1866 struct pack_entry e
;
1868 if (!find_pack_entry(sha1
, &e
, NULL
))
1871 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1875 * This is meant to hold a *small* number of objects that you would
1876 * want read_sha1_file() to be able to return, but yet you do not want
1877 * to write them into the object store (e.g. a browse-only
1880 static struct cached_object
{
1881 unsigned char sha1
[20];
1882 enum object_type type
;
1886 static int cached_object_nr
, cached_object_alloc
;
1888 static struct cached_object empty_tree
= {
1889 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1890 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1891 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1897 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1900 struct cached_object
*co
= cached_objects
;
1902 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1903 if (!hashcmp(co
->sha1
, sha1
))
1906 if (!hashcmp(sha1
, empty_tree
.sha1
))
1911 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1912 unsigned char *sha1
)
1914 struct cached_object
*co
;
1916 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1917 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1919 if (cached_object_alloc
<= cached_object_nr
) {
1920 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1921 cached_objects
= xrealloc(cached_objects
,
1922 sizeof(*cached_objects
) *
1923 cached_object_alloc
);
1925 co
= &cached_objects
[cached_object_nr
++];
1928 co
->buf
= xmalloc(len
);
1929 memcpy(co
->buf
, buf
, len
);
1930 hashcpy(co
->sha1
, sha1
);
1934 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1935 unsigned long *size
)
1937 unsigned long mapsize
;
1939 struct cached_object
*co
;
1941 co
= find_cached_object(sha1
);
1945 return xmemdupz(co
->buf
, co
->size
);
1948 buf
= read_packed_sha1(sha1
, type
, size
);
1951 map
= map_sha1_file(sha1
, &mapsize
);
1953 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1954 munmap(map
, mapsize
);
1957 reprepare_packed_git();
1958 return read_packed_sha1(sha1
, type
, size
);
1961 void *read_object_with_reference(const unsigned char *sha1
,
1962 const char *required_type_name
,
1963 unsigned long *size
,
1964 unsigned char *actual_sha1_return
)
1966 enum object_type type
, required_type
;
1968 unsigned long isize
;
1969 unsigned char actual_sha1
[20];
1971 required_type
= type_from_string(required_type_name
);
1972 hashcpy(actual_sha1
, sha1
);
1974 int ref_length
= -1;
1975 const char *ref_type
= NULL
;
1977 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1980 if (type
== required_type
) {
1982 if (actual_sha1_return
)
1983 hashcpy(actual_sha1_return
, actual_sha1
);
1986 /* Handle references */
1987 else if (type
== OBJ_COMMIT
)
1989 else if (type
== OBJ_TAG
)
1990 ref_type
= "object ";
1995 ref_length
= strlen(ref_type
);
1997 if (ref_length
+ 40 > isize
||
1998 memcmp(buffer
, ref_type
, ref_length
) ||
1999 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2004 /* Now we have the ID of the referred-to object in
2005 * actual_sha1. Check again. */
2009 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2010 const char *type
, unsigned char *sha1
,
2011 char *hdr
, int *hdrlen
)
2015 /* Generate the header */
2016 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2020 SHA1_Update(&c
, hdr
, *hdrlen
);
2021 SHA1_Update(&c
, buf
, len
);
2022 SHA1_Final(sha1
, &c
);
2026 * Link the tempfile to the final place, possibly creating the
2027 * last directory level as you do so.
2029 * Returns the errno on failure, 0 on success.
2031 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
2036 if (!link(tmpfile
, filename
))
2040 * Try to mkdir the last path component if that failed.
2042 * Re-try the "link()" regardless of whether the mkdir
2043 * succeeds, since a race might mean that somebody
2047 dir
= strrchr(filename
, '/');
2050 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
2055 if (!link(tmpfile
, filename
))
2063 * Move the just written object into its final resting place
2065 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2067 int ret
= link_temp_to_file(tmpfile
, filename
);
2070 * Coda hack - coda doesn't like cross-directory links,
2071 * so we fall back to a rename, which will mean that it
2072 * won't be able to check collisions, but that's not a
2075 * The same holds for FAT formatted media.
2077 * When this succeeds, we just return 0. We have nothing
2080 if (ret
&& ret
!= EEXIST
) {
2081 if (!rename(tmpfile
, filename
))
2087 if (ret
!= EEXIST
) {
2088 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2090 /* FIXME!!! Collision check here ? */
2096 static int write_buffer(int fd
, const void *buf
, size_t len
)
2098 if (write_in_full(fd
, buf
, len
) < 0)
2099 return error("file write error (%s)", strerror(errno
));
2103 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2104 unsigned char *sha1
)
2108 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2112 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2113 void *buf
, unsigned long len
, time_t mtime
)
2116 unsigned char *compressed
;
2119 static char tmpfile
[PATH_MAX
];
2121 filename
= sha1_file_name(sha1
);
2122 fd
= open(filename
, O_RDONLY
);
2125 * FIXME!!! We might do collision checking here, but we'd
2126 * need to uncompress the old file and check it. Later.
2132 if (errno
!= ENOENT
) {
2133 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2136 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2138 fd
= mkstemp(tmpfile
);
2141 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2143 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2147 memset(&stream
, 0, sizeof(stream
));
2148 deflateInit(&stream
, zlib_compression_level
);
2149 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2150 compressed
= xmalloc(size
);
2153 stream
.next_out
= compressed
;
2154 stream
.avail_out
= size
;
2156 /* First header.. */
2157 stream
.next_in
= (unsigned char *)hdr
;
2158 stream
.avail_in
= hdrlen
;
2159 while (deflate(&stream
, 0) == Z_OK
)
2162 /* Then the data itself.. */
2163 stream
.next_in
= buf
;
2164 stream
.avail_in
= len
;
2165 ret
= deflate(&stream
, Z_FINISH
);
2166 if (ret
!= Z_STREAM_END
)
2167 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2169 ret
= deflateEnd(&stream
);
2171 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2173 size
= stream
.total_out
;
2175 if (write_buffer(fd
, compressed
, size
) < 0)
2176 die("unable to write sha1 file");
2179 die("unable to write sha1 file");
2185 utb
.modtime
= mtime
;
2186 if (utime(tmpfile
, &utb
) < 0)
2187 warning("failed utime() on %s: %s",
2188 tmpfile
, strerror(errno
));
2191 return move_temp_to_file(tmpfile
, filename
);
2194 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2196 unsigned char sha1
[20];
2200 /* Normally if we have it in the pack then we do not bother writing
2201 * it out into .git/objects/??/?{38} file.
2203 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2205 hashcpy(returnsha1
, sha1
);
2206 if (has_sha1_file(sha1
))
2208 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2211 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2216 enum object_type type
;
2220 if (find_sha1_file(sha1
, &st
))
2222 buf
= read_packed_sha1(sha1
, &type
, &len
);
2224 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2225 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2226 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2230 * We need to unpack and recompress the object for writing
2231 * it out to a different file.
2233 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2237 unsigned char *unpacked
;
2239 enum object_type type
;
2244 /* need to unpack and recompress it by itself */
2245 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2247 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2249 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2252 memset(&stream
, 0, sizeof(stream
));
2253 deflateInit(&stream
, zlib_compression_level
);
2254 size
= deflateBound(&stream
, len
+ hdrlen
);
2255 buf
= xmalloc(size
);
2258 stream
.next_out
= buf
;
2259 stream
.avail_out
= size
;
2261 /* First header.. */
2262 stream
.next_in
= (void *)hdr
;
2263 stream
.avail_in
= hdrlen
;
2264 while (deflate(&stream
, 0) == Z_OK
)
2267 /* Then the data itself.. */
2268 stream
.next_in
= unpacked
;
2269 stream
.avail_in
= len
;
2270 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2272 deflateEnd(&stream
);
2275 *objsize
= stream
.total_out
;
2279 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2282 unsigned long objsize
;
2283 void *buf
= map_sha1_file(sha1
, &objsize
);
2286 retval
= write_buffer(fd
, buf
, objsize
);
2287 munmap(buf
, objsize
);
2291 buf
= repack_object(sha1
, &objsize
);
2292 retval
= write_buffer(fd
, buf
, objsize
);
2297 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2298 size_t bufsize
, size_t *bufposn
)
2300 char tmpfile
[PATH_MAX
];
2303 unsigned char real_sha1
[20];
2304 unsigned char discard
[4096];
2308 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2310 local
= mkstemp(tmpfile
);
2313 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2315 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2318 memset(&stream
, 0, sizeof(stream
));
2320 inflateInit(&stream
);
2327 stream
.avail_in
= *bufposn
;
2328 stream
.next_in
= (unsigned char *) buffer
;
2330 stream
.next_out
= discard
;
2331 stream
.avail_out
= sizeof(discard
);
2332 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2333 SHA1_Update(&c
, discard
, sizeof(discard
) -
2335 } while (stream
.avail_in
&& ret
== Z_OK
);
2336 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2337 die("unable to write sha1 file");
2338 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2340 *bufposn
= stream
.avail_in
;
2344 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2349 return error("Connection closed?");
2350 perror("Reading from connection");
2355 inflateEnd(&stream
);
2357 fchmod(local
, 0444);
2358 if (close(local
) != 0)
2359 die("unable to write sha1 file");
2360 SHA1_Final(real_sha1
, &c
);
2361 if (ret
!= Z_STREAM_END
) {
2363 return error("File %s corrupted", sha1_to_hex(sha1
));
2365 if (hashcmp(sha1
, real_sha1
)) {
2367 return error("File %s has bad hash", sha1_to_hex(sha1
));
2370 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2373 int has_pack_index(const unsigned char *sha1
)
2376 if (stat(sha1_pack_index_name(sha1
), &st
))
2381 int has_pack_file(const unsigned char *sha1
)
2384 if (stat(sha1_pack_name(sha1
), &st
))
2389 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2391 struct pack_entry e
;
2392 return find_pack_entry(sha1
, &e
, ignore_packed
);
2395 int has_sha1_file(const unsigned char *sha1
)
2398 struct pack_entry e
;
2400 if (find_pack_entry(sha1
, &e
, NULL
))
2402 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2405 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2410 strbuf_init(&buf
, 0);
2411 if (strbuf_read(&buf
, fd
, 4096) < 0) {
2412 strbuf_release(&buf
);
2419 ret
= write_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2421 ret
= hash_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2422 strbuf_release(&buf
);
2427 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2428 enum object_type type
, const char *path
)
2430 size_t size
= xsize_t(st
->st_size
);
2432 int ret
, re_allocated
= 0;
2435 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2442 * Convert blobs to git internal format
2444 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2446 strbuf_init(&nbuf
, 0);
2447 if (convert_to_git(path
, buf
, size
, &nbuf
,
2448 write_object
? safe_crlf
: 0)) {
2450 buf
= strbuf_detach(&nbuf
, &size
);
2456 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2458 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2468 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2474 switch (st
->st_mode
& S_IFMT
) {
2476 fd
= open(path
, O_RDONLY
);
2478 return error("open(\"%s\"): %s", path
,
2480 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2481 return error("%s: failed to insert into database",
2485 len
= xsize_t(st
->st_size
);
2486 target
= xmalloc(len
+ 1);
2487 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2488 char *errstr
= strerror(errno
);
2490 return error("readlink(\"%s\"): %s", path
,
2494 hash_sha1_file(target
, len
, blob_type
, sha1
);
2495 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2496 return error("%s: failed to insert into database",
2501 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2503 return error("%s: unsupported file type", path
);
2508 int read_pack_header(int fd
, struct pack_header
*header
)
2510 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2511 /* "eof before pack header was fully read" */
2512 return PH_ERROR_EOF
;
2514 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2515 /* "protocol error (pack signature mismatch detected)" */
2516 return PH_ERROR_PACK_SIGNATURE
;
2517 if (!pack_version_ok(header
->hdr_version
))
2518 /* "protocol error (pack version unsupported)" */
2519 return PH_ERROR_PROTOCOL
;