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 const signed char hexval_table
[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
76 for (i
= 0; i
< 20; i
++) {
77 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
86 static inline int offset_1st_component(const char *path
)
88 if (has_dos_drive_prefix(path
))
89 return 2 + (path
[2] == '/');
93 int safe_create_leading_directories(char *path
)
95 char *pos
= path
+ offset_1st_component(path
);
99 pos
= strchr(pos
, '/');
102 while (*++pos
== '/')
107 if (!stat(path
, &st
)) {
109 if (!S_ISDIR(st
.st_mode
)) {
114 else if (mkdir(path
, 0777)) {
118 else if (adjust_shared_perm(path
)) {
127 int safe_create_leading_directories_const(const char *path
)
129 /* path points to cache entries, so xstrdup before messing with it */
130 char *buf
= xstrdup(path
);
131 int result
= safe_create_leading_directories(buf
);
136 char *sha1_to_hex(const unsigned char *sha1
)
139 static char hexbuffer
[4][50];
140 static const char hex
[] = "0123456789abcdef";
141 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
144 for (i
= 0; i
< 20; i
++) {
145 unsigned int val
= *sha1
++;
146 *buf
++ = hex
[val
>> 4];
147 *buf
++ = hex
[val
& 0xf];
154 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
157 for (i
= 0; i
< 20; i
++) {
158 static char hex
[] = "0123456789abcdef";
159 unsigned int val
= sha1
[i
];
160 char *pos
= pathbuf
+ i
*2 + (i
> 0);
161 *pos
++ = hex
[val
>> 4];
162 *pos
= hex
[val
& 0xf];
167 * NOTE! This returns a statically allocated buffer, so you have to be
168 * careful about using it. Do an "xstrdup()" if you need to save the
171 * Also note that this returns the location for creating. Reading
172 * SHA1 file can happen from any alternate directory listed in the
173 * DB_ENVIRONMENT environment variable if it is not found in
174 * the primary object database.
176 char *sha1_file_name(const unsigned char *sha1
)
178 static char *name
, *base
;
181 const char *sha1_file_directory
= get_object_directory();
182 int len
= strlen(sha1_file_directory
);
183 base
= xmalloc(len
+ 60);
184 memcpy(base
, sha1_file_directory
, len
);
185 memset(base
+len
, 0, 60);
188 name
= base
+ len
+ 1;
190 fill_sha1_path(name
, sha1
);
194 static char *sha1_get_pack_name(const unsigned char *sha1
,
195 char **name
, char **base
, const char *which
)
197 static const char hex
[] = "0123456789abcdef";
202 const char *sha1_file_directory
= get_object_directory();
203 int len
= strlen(sha1_file_directory
);
204 *base
= xmalloc(len
+ 60);
205 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
206 sha1_file_directory
, which
);
207 *name
= *base
+ len
+ 11;
212 for (i
= 0; i
< 20; i
++) {
213 unsigned int val
= *sha1
++;
214 *buf
++ = hex
[val
>> 4];
215 *buf
++ = hex
[val
& 0xf];
221 char *sha1_pack_name(const unsigned char *sha1
)
223 static char *name
, *base
;
225 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
228 char *sha1_pack_index_name(const unsigned char *sha1
)
230 static char *name
, *base
;
232 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
235 struct alternate_object_database
*alt_odb_list
;
236 static struct alternate_object_database
**alt_odb_tail
;
238 static void read_info_alternates(const char * alternates
, int depth
);
241 * Prepare alternate object database registry.
243 * The variable alt_odb_list points at the list of struct
244 * alternate_object_database. The elements on this list come from
245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
247 * whose contents is similar to that environment variable but can be
248 * LF separated. Its base points at a statically allocated buffer that
249 * contains "/the/directory/corresponding/to/.git/objects/...", while
250 * its name points just after the slash at the end of ".git/objects/"
251 * in the example above, and has enough space to hold 40-byte hex
252 * SHA1, an extra slash for the first level indirection, and the
255 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
257 const char *objdir
= get_object_directory();
258 struct alternate_object_database
*ent
;
259 struct alternate_object_database
*alt
;
260 /* 43 = 40-byte + 2 '/' + terminating NUL */
262 int entlen
= pfxlen
+ 43;
265 if (!is_absolute_path(entry
) && relative_base
) {
266 /* Relative alt-odb */
268 base_len
= strlen(relative_base
) + 1;
272 ent
= xmalloc(sizeof(*ent
) + entlen
);
274 if (!is_absolute_path(entry
) && relative_base
) {
275 memcpy(ent
->base
, relative_base
, base_len
- 1);
276 ent
->base
[base_len
- 1] = '/';
277 memcpy(ent
->base
+ base_len
, entry
, len
);
280 memcpy(ent
->base
, entry
, pfxlen
);
282 ent
->name
= ent
->base
+ pfxlen
+ 1;
283 ent
->base
[pfxlen
+ 3] = '/';
284 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
286 /* Detect cases where alternate disappeared */
287 if (!is_directory(ent
->base
)) {
288 error("object directory %s does not exist; "
289 "check .git/objects/info/alternates.",
295 /* Prevent the common mistake of listing the same
296 * thing twice, or object directory itself.
298 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
299 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
304 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
309 /* add the alternate entry */
311 alt_odb_tail
= &(ent
->next
);
314 /* recursively add alternates */
315 read_info_alternates(ent
->base
, depth
+ 1);
317 ent
->base
[pfxlen
] = '/';
322 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
323 const char *relative_base
, int depth
)
325 const char *cp
, *last
;
328 error("%s: ignoring alternate object stores, nesting too deep.",
336 if (cp
< ep
&& *cp
== '#') {
337 while (cp
< ep
&& *cp
!= sep
)
342 while (cp
< ep
&& *cp
!= sep
)
345 if (!is_absolute_path(last
) && depth
) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base
, last
);
349 link_alt_odb_entry(last
, cp
- last
,
350 relative_base
, depth
);
353 while (cp
< ep
&& *cp
== sep
)
359 static void read_info_alternates(const char * relative_base
, int depth
)
364 const char alt_file_name
[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
371 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
372 fd
= open(path
, O_RDONLY
);
375 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
379 mapsz
= xsize_t(st
.st_size
);
380 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
383 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
388 void add_to_alternates_file(const char *reference
)
390 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
391 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
392 char *alt
= mkpath("%s/objects\n", reference
);
393 write_or_die(fd
, alt
, strlen(alt
));
394 if (commit_lock_file(lock
))
395 die("could not close alternates file");
397 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
400 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
402 struct alternate_object_database
*ent
;
405 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
410 void prepare_alt_odb(void)
417 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
420 alt_odb_tail
= &alt_odb_list
;
421 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
423 read_info_alternates(get_object_directory(), 0);
426 static int has_loose_object_local(const unsigned char *sha1
)
428 char *name
= sha1_file_name(sha1
);
429 return !access(name
, F_OK
);
432 int has_loose_object_nonlocal(const unsigned char *sha1
)
434 struct alternate_object_database
*alt
;
436 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
437 fill_sha1_path(alt
->name
, sha1
);
438 if (!access(alt
->base
, F_OK
))
444 static int has_loose_object(const unsigned char *sha1
)
446 return has_loose_object_local(sha1
) ||
447 has_loose_object_nonlocal(sha1
);
450 static unsigned int pack_used_ctr
;
451 static unsigned int pack_mmap_calls
;
452 static unsigned int peak_pack_open_windows
;
453 static unsigned int pack_open_windows
;
454 static size_t peak_pack_mapped
;
455 static size_t pack_mapped
;
456 struct packed_git
*packed_git
;
458 void pack_report(void)
461 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
462 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
463 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
464 sz_fmt(getpagesize()),
465 sz_fmt(packed_git_window_size
),
466 sz_fmt(packed_git_limit
));
468 "pack_report: pack_used_ctr = %10u\n"
469 "pack_report: pack_mmap_calls = %10u\n"
470 "pack_report: pack_open_windows = %10u / %10u\n"
471 "pack_report: pack_mapped = "
472 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
475 pack_open_windows
, peak_pack_open_windows
,
476 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
479 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
482 struct pack_idx_header
*hdr
;
484 uint32_t version
, nr
, i
, *index
;
485 int fd
= open(path
, O_RDONLY
);
490 if (fstat(fd
, &st
)) {
494 idx_size
= xsize_t(st
.st_size
);
495 if (idx_size
< 4 * 256 + 20 + 20) {
497 return error("index file %s is too small", path
);
499 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
503 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
504 version
= ntohl(hdr
->idx_version
);
505 if (version
< 2 || version
> 2) {
506 munmap(idx_map
, idx_size
);
507 return error("index file %s is version %"PRIu32
508 " and is not supported by this binary"
509 " (try upgrading GIT to a newer version)",
518 index
+= 2; /* skip index header */
519 for (i
= 0; i
< 256; i
++) {
520 uint32_t n
= ntohl(index
[i
]);
522 munmap(idx_map
, idx_size
);
523 return error("non-monotonic index %s", path
);
531 * - 256 index entries 4 bytes each
532 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
533 * - 20-byte SHA1 of the packfile
534 * - 20-byte SHA1 file checksum
536 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
537 munmap(idx_map
, idx_size
);
538 return error("wrong index v1 file size in %s", path
);
540 } else if (version
== 2) {
543 * - 8 bytes of header
544 * - 256 index entries 4 bytes each
545 * - 20-byte sha1 entry * nr
546 * - 4-byte crc entry * nr
547 * - 4-byte offset entry * nr
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 * And after the 4-byte offset table might be a
551 * variable sized table containing 8-byte entries
552 * for offsets larger than 2^31.
554 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
555 unsigned long max_size
= min_size
;
557 max_size
+= (nr
- 1)*8;
558 if (idx_size
< min_size
|| idx_size
> max_size
) {
559 munmap(idx_map
, idx_size
);
560 return error("wrong index v2 file size in %s", path
);
562 if (idx_size
!= min_size
&&
564 * make sure we can deal with large pack offsets.
565 * 31-bit signed offset won't be enough, neither
566 * 32-bit unsigned one will be.
568 (sizeof(off_t
) <= 4)) {
569 munmap(idx_map
, idx_size
);
570 return error("pack too large for current definition of off_t in %s", path
);
574 p
->index_version
= version
;
575 p
->index_data
= idx_map
;
576 p
->index_size
= idx_size
;
581 int open_pack_index(struct packed_git
*p
)
589 idx_name
= xstrdup(p
->pack_name
);
590 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
591 ret
= check_packed_git_idx(idx_name
, p
);
596 static void scan_windows(struct packed_git
*p
,
597 struct packed_git
**lru_p
,
598 struct pack_window
**lru_w
,
599 struct pack_window
**lru_l
)
601 struct pack_window
*w
, *w_l
;
603 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
605 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
615 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
617 struct packed_git
*p
, *lru_p
= NULL
;
618 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
621 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
622 for (p
= packed_git
; p
; p
= p
->next
)
623 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
625 munmap(lru_w
->base
, lru_w
->len
);
626 pack_mapped
-= lru_w
->len
;
628 lru_l
->next
= lru_w
->next
;
630 lru_p
->windows
= lru_w
->next
;
631 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
632 close(lru_p
->pack_fd
);
643 void release_pack_memory(size_t need
, int fd
)
645 size_t cur
= pack_mapped
;
646 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
650 void close_pack_windows(struct packed_git
*p
)
653 struct pack_window
*w
= p
->windows
;
656 die("pack '%s' still has open windows to it",
658 munmap(w
->base
, w
->len
);
659 pack_mapped
-= w
->len
;
661 p
->windows
= w
->next
;
666 void unuse_pack(struct pack_window
**w_cursor
)
668 struct pack_window
*w
= *w_cursor
;
676 * This is used by git-repack in case a newly created pack happens to
677 * contain the same set of objects as an existing one. In that case
678 * the resulting file might be different even if its name would be the
679 * same. It is best to close any reference to the old pack before it is
680 * replaced on disk. Of course no index pointers nor windows for given pack
681 * must subsist at this point. If ever objects from this pack are requested
682 * again, the new version of the pack will be reinitialized through
683 * reprepare_packed_git().
685 void free_pack_by_name(const char *pack_name
)
687 struct packed_git
*p
, **pp
= &packed_git
;
691 if (strcmp(pack_name
, p
->pack_name
) == 0) {
692 close_pack_windows(p
);
693 if (p
->pack_fd
!= -1)
696 munmap((void *)p
->index_data
, p
->index_size
);
697 free(p
->bad_object_sha1
);
707 * Do not call this directly as this leaks p->pack_fd on error return;
708 * call open_packed_git() instead.
710 static int open_packed_git_1(struct packed_git
*p
)
713 struct pack_header hdr
;
714 unsigned char sha1
[20];
715 unsigned char *idx_sha1
;
718 if (!p
->index_data
&& open_pack_index(p
))
719 return error("packfile %s index unavailable", p
->pack_name
);
721 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
722 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
725 /* If we created the struct before we had the pack we lack size. */
727 if (!S_ISREG(st
.st_mode
))
728 return error("packfile %s not a regular file", p
->pack_name
);
729 p
->pack_size
= st
.st_size
;
730 } else if (p
->pack_size
!= st
.st_size
)
731 return error("packfile %s size changed", p
->pack_name
);
733 /* We leave these file descriptors open with sliding mmap;
734 * there is no point keeping them open across exec(), though.
736 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
738 return error("cannot determine file descriptor flags");
739 fd_flag
|= FD_CLOEXEC
;
740 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
741 return error("cannot set FD_CLOEXEC");
743 /* Verify we recognize this pack file format. */
744 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
745 return error("file %s is far too short to be a packfile", p
->pack_name
);
746 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
747 return error("file %s is not a GIT packfile", p
->pack_name
);
748 if (!pack_version_ok(hdr
.hdr_version
))
749 return error("packfile %s is version %"PRIu32
" and not"
750 " supported (try upgrading GIT to a newer version)",
751 p
->pack_name
, ntohl(hdr
.hdr_version
));
753 /* Verify the pack matches its index. */
754 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
755 return error("packfile %s claims to have %"PRIu32
" objects"
756 " while index indicates %"PRIu32
" objects",
757 p
->pack_name
, ntohl(hdr
.hdr_entries
),
759 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
760 return error("end of packfile %s is unavailable", p
->pack_name
);
761 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
762 return error("packfile %s signature is unavailable", p
->pack_name
);
763 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
764 if (hashcmp(sha1
, idx_sha1
))
765 return error("packfile %s does not match index", p
->pack_name
);
769 static int open_packed_git(struct packed_git
*p
)
771 if (!open_packed_git_1(p
))
773 if (p
->pack_fd
!= -1) {
780 static int in_window(struct pack_window
*win
, off_t offset
)
782 /* We must promise at least 20 bytes (one hash) after the
783 * offset is available from this window, otherwise the offset
784 * is not actually in this window and a different window (which
785 * has that one hash excess) must be used. This is to support
786 * the object header and delta base parsing routines below.
788 off_t win_off
= win
->offset
;
789 return win_off
<= offset
790 && (offset
+ 20) <= (win_off
+ win
->len
);
793 unsigned char* use_pack(struct packed_git
*p
,
794 struct pack_window
**w_cursor
,
798 struct pack_window
*win
= *w_cursor
;
800 if (p
->pack_fd
== -1 && open_packed_git(p
))
801 die("packfile %s cannot be accessed", p
->pack_name
);
803 /* Since packfiles end in a hash of their content and its
804 * pointless to ask for an offset into the middle of that
805 * hash, and the in_window function above wouldn't match
806 * don't allow an offset too close to the end of the file.
808 if (offset
> (p
->pack_size
- 20))
809 die("offset beyond end of packfile (truncated pack?)");
811 if (!win
|| !in_window(win
, offset
)) {
814 for (win
= p
->windows
; win
; win
= win
->next
) {
815 if (in_window(win
, offset
))
819 size_t window_align
= packed_git_window_size
/ 2;
821 win
= xcalloc(1, sizeof(*win
));
822 win
->offset
= (offset
/ window_align
) * window_align
;
823 len
= p
->pack_size
- win
->offset
;
824 if (len
> packed_git_window_size
)
825 len
= packed_git_window_size
;
826 win
->len
= (size_t)len
;
827 pack_mapped
+= win
->len
;
828 while (packed_git_limit
< pack_mapped
829 && unuse_one_window(p
, p
->pack_fd
))
831 win
->base
= xmmap(NULL
, win
->len
,
832 PROT_READ
, MAP_PRIVATE
,
833 p
->pack_fd
, win
->offset
);
834 if (win
->base
== MAP_FAILED
)
835 die("packfile %s cannot be mapped: %s",
840 if (pack_mapped
> peak_pack_mapped
)
841 peak_pack_mapped
= pack_mapped
;
842 if (pack_open_windows
> peak_pack_open_windows
)
843 peak_pack_open_windows
= pack_open_windows
;
844 win
->next
= p
->windows
;
848 if (win
!= *w_cursor
) {
849 win
->last_used
= pack_used_ctr
++;
853 offset
-= win
->offset
;
855 *left
= win
->len
- xsize_t(offset
);
856 return win
->base
+ offset
;
859 static struct packed_git
*alloc_packed_git(int extra
)
861 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
862 memset(p
, 0, sizeof(*p
));
867 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
870 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
873 * Make sure a corresponding .pack file exists and that
874 * the index looks sane.
876 path_len
-= strlen(".idx");
881 memcpy(p
->pack_name
, path
, path_len
);
883 strcpy(p
->pack_name
+ path_len
, ".keep");
884 if (!access(p
->pack_name
, F_OK
))
887 strcpy(p
->pack_name
+ path_len
, ".pack");
888 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
893 /* ok, it looks sane as far as we can check without
894 * actually mapping the pack file.
896 p
->pack_size
= st
.st_size
;
897 p
->pack_local
= local
;
898 p
->mtime
= st
.st_mtime
;
899 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
904 struct packed_git
*parse_pack_index(unsigned char *sha1
)
906 const char *idx_path
= sha1_pack_index_name(sha1
);
907 const char *path
= sha1_pack_name(sha1
);
908 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
910 strcpy(p
->pack_name
, path
);
911 hashcpy(p
->sha1
, sha1
);
912 if (check_packed_git_idx(idx_path
, p
)) {
920 void install_packed_git(struct packed_git
*pack
)
922 pack
->next
= packed_git
;
926 static void prepare_packed_git_one(char *objdir
, int local
)
928 /* Ensure that this buffer is large enough so that we can
929 append "/pack/" without clobbering the stack even if
930 strlen(objdir) were PATH_MAX. */
931 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
936 sprintf(path
, "%s/pack", objdir
);
941 error("unable to open object pack directory: %s: %s",
942 path
, strerror(errno
));
946 while ((de
= readdir(dir
)) != NULL
) {
947 int namelen
= strlen(de
->d_name
);
948 struct packed_git
*p
;
950 if (!has_extension(de
->d_name
, ".idx"))
953 if (len
+ namelen
+ 1 > sizeof(path
))
956 /* Don't reopen a pack we already have. */
957 strcpy(path
+ len
, de
->d_name
);
958 for (p
= packed_git
; p
; p
= p
->next
) {
959 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
964 /* See if it really is a valid .idx file with corresponding
965 * .pack file that we can map.
967 p
= add_packed_git(path
, len
+ namelen
, local
);
970 install_packed_git(p
);
975 static int sort_pack(const void *a_
, const void *b_
)
977 struct packed_git
*a
= *((struct packed_git
**)a_
);
978 struct packed_git
*b
= *((struct packed_git
**)b_
);
982 * Local packs tend to contain objects specific to our
983 * variant of the project than remote ones. In addition,
984 * remote ones could be on a network mounted filesystem.
985 * Favor local ones for these reasons.
987 st
= a
->pack_local
- b
->pack_local
;
992 * Younger packs tend to contain more recent objects,
993 * and more recent objects tend to get accessed more
996 if (a
->mtime
< b
->mtime
)
998 else if (a
->mtime
== b
->mtime
)
1003 static void rearrange_packed_git(void)
1005 struct packed_git
**ary
, *p
;
1008 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1013 /* prepare an array of packed_git for easier sorting */
1014 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1015 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1018 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1020 /* link them back again */
1021 for (i
= 0; i
< n
- 1; i
++)
1022 ary
[i
]->next
= ary
[i
+ 1];
1023 ary
[n
- 1]->next
= NULL
;
1024 packed_git
= ary
[0];
1029 static int prepare_packed_git_run_once
= 0;
1030 void prepare_packed_git(void)
1032 struct alternate_object_database
*alt
;
1034 if (prepare_packed_git_run_once
)
1036 prepare_packed_git_one(get_object_directory(), 1);
1038 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1040 prepare_packed_git_one(alt
->base
, 0);
1041 alt
->name
[-1] = '/';
1043 rearrange_packed_git();
1044 prepare_packed_git_run_once
= 1;
1047 void reprepare_packed_git(void)
1050 prepare_packed_git_run_once
= 0;
1051 prepare_packed_git();
1054 static void mark_bad_packed_object(struct packed_git
*p
,
1055 const unsigned char *sha1
)
1058 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1059 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1061 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1062 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1063 p
->num_bad_objects
++;
1066 static int has_packed_and_bad(const unsigned char *sha1
)
1068 struct packed_git
*p
;
1071 for (p
= packed_git
; p
; p
= p
->next
)
1072 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1073 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1078 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1080 unsigned char real_sha1
[20];
1081 hash_sha1_file(map
, size
, type
, real_sha1
);
1082 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1085 static int git_open_noatime(const char *name
)
1087 static int sha1_file_open_flag
= O_NOATIME
;
1088 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1090 /* Might the failure be due to O_NOATIME? */
1091 if (fd
< 0 && errno
!= ENOENT
&& sha1_file_open_flag
) {
1092 fd
= open(name
, O_RDONLY
);
1094 sha1_file_open_flag
= 0;
1099 static int open_sha1_file(const unsigned char *sha1
)
1102 char *name
= sha1_file_name(sha1
);
1103 struct alternate_object_database
*alt
;
1105 fd
= git_open_noatime(name
);
1111 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1113 fill_sha1_path(name
, sha1
);
1114 fd
= git_open_noatime(alt
->base
);
1121 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1126 fd
= open_sha1_file(sha1
);
1131 if (!fstat(fd
, &st
)) {
1132 *size
= xsize_t(st
.st_size
);
1133 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1140 static int legacy_loose_object(unsigned char *map
)
1145 * Is it a zlib-compressed buffer? If so, the first byte
1146 * must be 0x78 (15-bit window size, deflated), and the
1147 * first 16-bit word is evenly divisible by 31
1149 word
= (map
[0] << 8) + map
[1];
1150 if (map
[0] == 0x78 && !(word
% 31))
1156 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1157 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1162 unsigned long used
= 0;
1165 *type
= (c
>> 4) & 7;
1169 if (len
<= used
|| sizeof(long) * 8 <= shift
) {
1170 error("bad object header");
1174 size
+= (c
& 0x7f) << shift
;
1181 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1183 unsigned long size
, used
;
1184 static const char valid_loose_object_type
[8] = {
1186 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1187 0, /* "delta" and others are invalid in a loose object */
1189 enum object_type type
;
1191 /* Get the data stream */
1192 memset(stream
, 0, sizeof(*stream
));
1193 stream
->next_in
= map
;
1194 stream
->avail_in
= mapsize
;
1195 stream
->next_out
= buffer
;
1196 stream
->avail_out
= bufsiz
;
1198 if (legacy_loose_object(map
)) {
1199 git_inflate_init(stream
);
1200 return git_inflate(stream
, 0);
1205 * There used to be a second loose object header format which
1206 * was meant to mimic the in-pack format, allowing for direct
1207 * copy of the object data. This format turned up not to be
1208 * really worth it and we don't write it any longer. But we
1209 * can still read it.
1211 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1212 if (!used
|| !valid_loose_object_type
[type
])
1217 /* Set up the stream for the rest.. */
1218 stream
->next_in
= map
;
1219 stream
->avail_in
= mapsize
;
1220 git_inflate_init(stream
);
1222 /* And generate the fake traditional header */
1223 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1224 typename(type
), size
);
1228 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1230 int bytes
= strlen(buffer
) + 1;
1231 unsigned char *buf
= xmalloc(1+size
);
1235 n
= stream
->total_out
- bytes
;
1238 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1240 if (bytes
<= size
) {
1242 * The above condition must be (bytes <= size), not
1243 * (bytes < size). In other words, even though we
1244 * expect no more output and set avail_out to zer0,
1245 * the input zlib stream may have bytes that express
1246 * "this concludes the stream", and we *do* want to
1249 * Otherwise we would not be able to test that we
1250 * consumed all the input to reach the expected size;
1251 * we also want to check that zlib tells us that all
1252 * went well with status == Z_STREAM_END at the end.
1254 stream
->next_out
= buf
+ bytes
;
1255 stream
->avail_out
= size
- bytes
;
1256 while (status
== Z_OK
)
1257 status
= git_inflate(stream
, Z_FINISH
);
1260 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1261 git_inflate_end(stream
);
1266 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1267 else if (stream
->avail_in
)
1268 error("garbage at end of loose object '%s'",
1275 * We used to just use "sscanf()", but that's actually way
1276 * too permissive for what we want to check. So do an anal
1277 * object header parse by hand.
1279 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1286 * The type can be at most ten bytes (including the
1287 * terminating '\0' that we add), and is followed by
1296 if (i
>= sizeof(type
))
1302 * The length must follow immediately, and be in canonical
1303 * decimal format (ie "010" is not valid).
1305 size
= *hdr
++ - '0';
1310 unsigned long c
= *hdr
- '0';
1314 size
= size
* 10 + c
;
1320 * The length must be followed by a zero byte
1322 return *hdr
? -1 : type_from_string(type
);
1325 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1331 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1332 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1335 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1338 unsigned long get_size_from_delta(struct packed_git
*p
,
1339 struct pack_window
**w_curs
,
1342 const unsigned char *data
;
1343 unsigned char delta_head
[20], *in
;
1347 memset(&stream
, 0, sizeof(stream
));
1348 stream
.next_out
= delta_head
;
1349 stream
.avail_out
= sizeof(delta_head
);
1351 git_inflate_init(&stream
);
1353 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1354 stream
.next_in
= in
;
1355 st
= git_inflate(&stream
, Z_FINISH
);
1356 curpos
+= stream
.next_in
- in
;
1357 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1358 stream
.total_out
< sizeof(delta_head
));
1359 git_inflate_end(&stream
);
1360 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1361 error("delta data unpack-initial failed");
1365 /* Examine the initial part of the delta to figure out
1370 /* ignore base size */
1371 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1373 /* Read the result size */
1374 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1377 static off_t
get_delta_base(struct packed_git
*p
,
1378 struct pack_window
**w_curs
,
1380 enum object_type type
,
1381 off_t delta_obj_offset
)
1383 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1386 /* use_pack() assured us we have [base_info, base_info + 20)
1387 * as a range that we can look at without walking off the
1388 * end of the mapped window. Its actually the hash size
1389 * that is assured. An OFS_DELTA longer than the hash size
1390 * is stupid, as then a REF_DELTA would be smaller to store.
1392 if (type
== OBJ_OFS_DELTA
) {
1394 unsigned char c
= base_info
[used
++];
1395 base_offset
= c
& 127;
1398 if (!base_offset
|| MSB(base_offset
, 7))
1399 return 0; /* overflow */
1400 c
= base_info
[used
++];
1401 base_offset
= (base_offset
<< 7) + (c
& 127);
1403 base_offset
= delta_obj_offset
- base_offset
;
1404 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1405 return 0; /* out of bound */
1407 } else if (type
== OBJ_REF_DELTA
) {
1408 /* The base entry _must_ be in the same pack */
1409 base_offset
= find_pack_entry_one(base_info
, p
);
1412 die("I am totally screwed");
1416 /* forward declaration for a mutually recursive function */
1417 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1418 unsigned long *sizep
);
1420 static int packed_delta_info(struct packed_git
*p
,
1421 struct pack_window
**w_curs
,
1423 enum object_type type
,
1425 unsigned long *sizep
)
1429 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1432 type
= packed_object_info(p
, base_offset
, NULL
);
1433 if (type
<= OBJ_NONE
) {
1434 struct revindex_entry
*revidx
;
1435 const unsigned char *base_sha1
;
1436 revidx
= find_pack_revindex(p
, base_offset
);
1439 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1440 mark_bad_packed_object(p
, base_sha1
);
1441 type
= sha1_object_info(base_sha1
, NULL
);
1442 if (type
<= OBJ_NONE
)
1446 /* We choose to only get the type of the base object and
1447 * ignore potentially corrupt pack file that expects the delta
1448 * based on a base with a wrong size. This saves tons of
1452 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1460 static int unpack_object_header(struct packed_git
*p
,
1461 struct pack_window
**w_curs
,
1463 unsigned long *sizep
)
1465 unsigned char *base
;
1468 enum object_type type
;
1470 /* use_pack() assures us we have [base, base + 20) available
1471 * as a range that we can look at at. (Its actually the hash
1472 * size that is assured.) With our object header encoding
1473 * the maximum deflated object size is 2^137, which is just
1474 * insane, so we know won't exceed what we have been given.
1476 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1477 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1486 const char *packed_object_info_detail(struct packed_git
*p
,
1488 unsigned long *size
,
1489 unsigned long *store_size
,
1490 unsigned int *delta_chain_length
,
1491 unsigned char *base_sha1
)
1493 struct pack_window
*w_curs
= NULL
;
1495 unsigned long dummy
;
1496 unsigned char *next_sha1
;
1497 enum object_type type
;
1498 struct revindex_entry
*revidx
;
1500 *delta_chain_length
= 0;
1501 curpos
= obj_offset
;
1502 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1504 revidx
= find_pack_revindex(p
, obj_offset
);
1505 *store_size
= revidx
[1].offset
- obj_offset
;
1510 die("pack %s contains unknown object type %d",
1511 p
->pack_name
, type
);
1516 unuse_pack(&w_curs
);
1517 return typename(type
);
1519 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1521 die("pack %s contains bad delta base reference of type %s",
1522 p
->pack_name
, typename(type
));
1523 if (*delta_chain_length
== 0) {
1524 revidx
= find_pack_revindex(p
, obj_offset
);
1525 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1529 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1530 if (*delta_chain_length
== 0)
1531 hashcpy(base_sha1
, next_sha1
);
1532 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1535 (*delta_chain_length
)++;
1536 curpos
= obj_offset
;
1537 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1541 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1542 unsigned long *sizep
)
1544 struct pack_window
*w_curs
= NULL
;
1546 off_t curpos
= obj_offset
;
1547 enum object_type type
;
1549 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1554 type
= packed_delta_info(p
, &w_curs
, curpos
,
1555 type
, obj_offset
, sizep
);
1565 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1566 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1569 unuse_pack(&w_curs
);
1573 static void *unpack_compressed_entry(struct packed_git
*p
,
1574 struct pack_window
**w_curs
,
1580 unsigned char *buffer
, *in
;
1582 buffer
= xmalloc(size
+ 1);
1584 memset(&stream
, 0, sizeof(stream
));
1585 stream
.next_out
= buffer
;
1586 stream
.avail_out
= size
;
1588 git_inflate_init(&stream
);
1590 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1591 stream
.next_in
= in
;
1592 st
= git_inflate(&stream
, Z_FINISH
);
1593 curpos
+= stream
.next_in
- in
;
1594 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1595 git_inflate_end(&stream
);
1596 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1604 #define MAX_DELTA_CACHE (256)
1606 static size_t delta_base_cached
;
1608 static struct delta_base_cache_lru_list
{
1609 struct delta_base_cache_lru_list
*prev
;
1610 struct delta_base_cache_lru_list
*next
;
1611 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1613 static struct delta_base_cache_entry
{
1614 struct delta_base_cache_lru_list lru
;
1616 struct packed_git
*p
;
1619 enum object_type type
;
1620 } delta_base_cache
[MAX_DELTA_CACHE
];
1622 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1626 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1627 hash
+= (hash
>> 8) + (hash
>> 16);
1628 return hash
% MAX_DELTA_CACHE
;
1631 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1632 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1635 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1636 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1639 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1640 return unpack_entry(p
, base_offset
, type
, base_size
);
1644 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1645 ent
->lru
.prev
->next
= ent
->lru
.next
;
1646 delta_base_cached
-= ent
->size
;
1648 ret
= xmemdupz(ent
->data
, ent
->size
);
1651 *base_size
= ent
->size
;
1655 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1660 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1661 ent
->lru
.prev
->next
= ent
->lru
.next
;
1662 delta_base_cached
-= ent
->size
;
1666 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1667 void *base
, unsigned long base_size
, enum object_type type
)
1669 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1670 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1671 struct delta_base_cache_lru_list
*lru
;
1673 release_delta_base_cache(ent
);
1674 delta_base_cached
+= base_size
;
1676 for (lru
= delta_base_cache_lru
.next
;
1677 delta_base_cached
> delta_base_cache_limit
1678 && lru
!= &delta_base_cache_lru
;
1680 struct delta_base_cache_entry
*f
= (void *)lru
;
1681 if (f
->type
== OBJ_BLOB
)
1682 release_delta_base_cache(f
);
1684 for (lru
= delta_base_cache_lru
.next
;
1685 delta_base_cached
> delta_base_cache_limit
1686 && lru
!= &delta_base_cache_lru
;
1688 struct delta_base_cache_entry
*f
= (void *)lru
;
1689 release_delta_base_cache(f
);
1693 ent
->base_offset
= base_offset
;
1696 ent
->size
= base_size
;
1697 ent
->lru
.next
= &delta_base_cache_lru
;
1698 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1699 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1700 delta_base_cache_lru
.prev
= &ent
->lru
;
1703 static void *unpack_delta_entry(struct packed_git
*p
,
1704 struct pack_window
**w_curs
,
1706 unsigned long delta_size
,
1708 enum object_type
*type
,
1709 unsigned long *sizep
)
1711 void *delta_data
, *result
, *base
;
1712 unsigned long base_size
;
1715 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1717 error("failed to validate delta base reference "
1718 "at offset %"PRIuMAX
" from %s",
1719 (uintmax_t)curpos
, p
->pack_name
);
1723 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1726 * We're probably in deep shit, but let's try to fetch
1727 * the required base anyway from another pack or loose.
1728 * This is costly but should happen only in the presence
1729 * of a corrupted pack, and is better than failing outright.
1731 struct revindex_entry
*revidx
;
1732 const unsigned char *base_sha1
;
1733 revidx
= find_pack_revindex(p
, base_offset
);
1736 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1737 error("failed to read delta base object %s"
1738 " at offset %"PRIuMAX
" from %s",
1739 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1741 mark_bad_packed_object(p
, base_sha1
);
1742 base
= read_object(base_sha1
, type
, &base_size
);
1747 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1749 error("failed to unpack compressed delta "
1750 "at offset %"PRIuMAX
" from %s",
1751 (uintmax_t)curpos
, p
->pack_name
);
1755 result
= patch_delta(base
, base_size
,
1756 delta_data
, delta_size
,
1759 die("failed to apply delta");
1761 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1765 int do_check_packed_object_crc
;
1767 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1768 enum object_type
*type
, unsigned long *sizep
)
1770 struct pack_window
*w_curs
= NULL
;
1771 off_t curpos
= obj_offset
;
1774 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1775 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1776 unsigned long len
= revidx
[1].offset
- obj_offset
;
1777 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1778 const unsigned char *sha1
=
1779 nth_packed_object_sha1(p
, revidx
->nr
);
1780 error("bad packed object CRC for %s",
1782 mark_bad_packed_object(p
, sha1
);
1783 unuse_pack(&w_curs
);
1788 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1792 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1793 obj_offset
, type
, sizep
);
1799 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1803 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1804 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1806 unuse_pack(&w_curs
);
1810 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1813 const unsigned char *index
= p
->index_data
;
1815 if (open_pack_index(p
))
1817 index
= p
->index_data
;
1819 if (n
>= p
->num_objects
)
1822 if (p
->index_version
== 1) {
1823 return index
+ 24 * n
+ 4;
1826 return index
+ 20 * n
;
1830 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1832 const unsigned char *index
= p
->index_data
;
1834 if (p
->index_version
== 1) {
1835 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1838 index
+= 8 + p
->num_objects
* (20 + 4);
1839 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1840 if (!(off
& 0x80000000))
1842 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1843 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1844 ntohl(*((uint32_t *)(index
+ 4)));
1848 off_t
find_pack_entry_one(const unsigned char *sha1
,
1849 struct packed_git
*p
)
1851 const uint32_t *level1_ofs
= p
->index_data
;
1852 const unsigned char *index
= p
->index_data
;
1853 unsigned hi
, lo
, stride
;
1854 static int use_lookup
= -1;
1855 static int debug_lookup
= -1;
1857 if (debug_lookup
< 0)
1858 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1861 if (open_pack_index(p
))
1863 level1_ofs
= p
->index_data
;
1864 index
= p
->index_data
;
1866 if (p
->index_version
> 1) {
1871 hi
= ntohl(level1_ofs
[*sha1
]);
1872 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1873 if (p
->index_version
> 1) {
1881 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1882 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1885 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1887 int pos
= sha1_entry_pos(index
, stride
, 0,
1888 lo
, hi
, p
->num_objects
, sha1
);
1891 return nth_packed_object_offset(p
, pos
);
1895 unsigned mi
= (lo
+ hi
) / 2;
1896 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1899 printf("lo %u hi %u rg %u mi %u\n",
1900 lo
, hi
, hi
- lo
, mi
);
1902 return nth_packed_object_offset(p
, mi
);
1911 int matches_pack_name(struct packed_git
*p
, const char *name
)
1913 const char *last_c
, *c
;
1915 if (!strcmp(p
->pack_name
, name
))
1918 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1923 if (!strcmp(last_c
, name
))
1929 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1931 static struct packed_git
*last_found
= (void *)1;
1932 struct packed_git
*p
;
1935 prepare_packed_git();
1938 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1941 if (ignore_packed
) {
1943 for (ig
= ignore_packed
; *ig
; ig
++)
1944 if (matches_pack_name(p
, *ig
))
1950 if (p
->num_bad_objects
) {
1952 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1953 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1957 offset
= find_pack_entry_one(sha1
, p
);
1960 * We are about to tell the caller where they can
1961 * locate the requested object. We better make
1962 * sure the packfile is still here and can be
1963 * accessed before supplying that answer, as
1964 * it may have been deleted since the index
1967 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1968 error("packfile %s cannot be accessed", p
->pack_name
);
1973 hashcpy(e
->sha1
, sha1
);
1979 if (p
== last_found
)
1983 if (p
== last_found
)
1989 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1990 struct packed_git
*packs
)
1992 struct packed_git
*p
;
1994 for (p
= packs
; p
; p
= p
->next
) {
1995 if (find_pack_entry_one(sha1
, p
))
2002 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2005 unsigned long mapsize
, size
;
2010 map
= map_sha1_file(sha1
, &mapsize
);
2012 return error("unable to find %s", sha1_to_hex(sha1
));
2013 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2014 status
= error("unable to unpack %s header",
2016 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2017 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2020 git_inflate_end(&stream
);
2021 munmap(map
, mapsize
);
2025 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2027 struct pack_entry e
;
2030 if (!find_pack_entry(sha1
, &e
, NULL
)) {
2031 /* Most likely it's a loose object. */
2032 status
= sha1_loose_object_info(sha1
, sizep
);
2036 /* Not a loose object; someone else may have just packed it. */
2037 reprepare_packed_git();
2038 if (!find_pack_entry(sha1
, &e
, NULL
))
2042 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
2044 mark_bad_packed_object(e
.p
, sha1
);
2045 status
= sha1_object_info(sha1
, sizep
);
2051 static void *read_packed_sha1(const unsigned char *sha1
,
2052 enum object_type
*type
, unsigned long *size
)
2054 struct pack_entry e
;
2057 if (!find_pack_entry(sha1
, &e
, NULL
))
2059 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2062 * We're probably in deep shit, but let's try to fetch
2063 * the required object anyway from another pack or loose.
2064 * This should happen only in the presence of a corrupted
2065 * pack, and is better than failing outright.
2067 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2068 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2069 mark_bad_packed_object(e
.p
, sha1
);
2070 data
= read_object(sha1
, type
, size
);
2076 * This is meant to hold a *small* number of objects that you would
2077 * want read_sha1_file() to be able to return, but yet you do not want
2078 * to write them into the object store (e.g. a browse-only
2081 static struct cached_object
{
2082 unsigned char sha1
[20];
2083 enum object_type type
;
2087 static int cached_object_nr
, cached_object_alloc
;
2089 static struct cached_object empty_tree
= {
2090 EMPTY_TREE_SHA1_BIN
,
2096 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
2099 struct cached_object
*co
= cached_objects
;
2101 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
2102 if (!hashcmp(co
->sha1
, sha1
))
2105 if (!hashcmp(sha1
, empty_tree
.sha1
))
2110 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2111 unsigned char *sha1
)
2113 struct cached_object
*co
;
2115 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2116 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2118 if (cached_object_alloc
<= cached_object_nr
) {
2119 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2120 cached_objects
= xrealloc(cached_objects
,
2121 sizeof(*cached_objects
) *
2122 cached_object_alloc
);
2124 co
= &cached_objects
[cached_object_nr
++];
2127 co
->buf
= xmalloc(len
);
2128 memcpy(co
->buf
, buf
, len
);
2129 hashcpy(co
->sha1
, sha1
);
2133 void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2134 unsigned long *size
)
2136 unsigned long mapsize
;
2138 struct cached_object
*co
;
2140 co
= find_cached_object(sha1
);
2144 return xmemdupz(co
->buf
, co
->size
);
2147 buf
= read_packed_sha1(sha1
, type
, size
);
2150 map
= map_sha1_file(sha1
, &mapsize
);
2152 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2153 munmap(map
, mapsize
);
2156 reprepare_packed_git();
2157 return read_packed_sha1(sha1
, type
, size
);
2160 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
2161 unsigned long *size
)
2163 void *data
= read_object(sha1
, type
, size
);
2164 /* legacy behavior is to die on corrupted objects */
2165 if (!data
&& (has_loose_object(sha1
) || has_packed_and_bad(sha1
)))
2166 die("object %s is corrupted", sha1_to_hex(sha1
));
2170 void *read_object_with_reference(const unsigned char *sha1
,
2171 const char *required_type_name
,
2172 unsigned long *size
,
2173 unsigned char *actual_sha1_return
)
2175 enum object_type type
, required_type
;
2177 unsigned long isize
;
2178 unsigned char actual_sha1
[20];
2180 required_type
= type_from_string(required_type_name
);
2181 hashcpy(actual_sha1
, sha1
);
2183 int ref_length
= -1;
2184 const char *ref_type
= NULL
;
2186 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2189 if (type
== required_type
) {
2191 if (actual_sha1_return
)
2192 hashcpy(actual_sha1_return
, actual_sha1
);
2195 /* Handle references */
2196 else if (type
== OBJ_COMMIT
)
2198 else if (type
== OBJ_TAG
)
2199 ref_type
= "object ";
2204 ref_length
= strlen(ref_type
);
2206 if (ref_length
+ 40 > isize
||
2207 memcmp(buffer
, ref_type
, ref_length
) ||
2208 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2213 /* Now we have the ID of the referred-to object in
2214 * actual_sha1. Check again. */
2218 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2219 const char *type
, unsigned char *sha1
,
2220 char *hdr
, int *hdrlen
)
2224 /* Generate the header */
2225 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2229 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2230 git_SHA1_Update(&c
, buf
, len
);
2231 git_SHA1_Final(sha1
, &c
);
2235 * Move the just written object into its final resting place
2237 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2240 if (link(tmpfile
, filename
))
2244 * Coda hack - coda doesn't like cross-directory links,
2245 * so we fall back to a rename, which will mean that it
2246 * won't be able to check collisions, but that's not a
2249 * The same holds for FAT formatted media.
2251 * When this succeeds, we just return 0. We have nothing
2254 if (ret
&& ret
!= EEXIST
) {
2255 if (!rename(tmpfile
, filename
))
2261 if (ret
!= EEXIST
) {
2262 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2264 /* FIXME!!! Collision check here ? */
2270 static int write_buffer(int fd
, const void *buf
, size_t len
)
2272 if (write_in_full(fd
, buf
, len
) < 0)
2273 return error("file write error (%s)", strerror(errno
));
2277 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2278 unsigned char *sha1
)
2282 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2286 /* Finalize a file on disk, and close it. */
2287 static void close_sha1_file(int fd
)
2289 if (fsync_object_files
)
2290 fsync_or_die(fd
, "sha1 file");
2293 die("unable to write sha1 file");
2296 /* Size of directory component, including the ending '/' */
2297 static inline int directory_size(const char *filename
)
2299 const char *s
= strrchr(filename
, '/');
2302 return s
- filename
+ 1;
2306 * This creates a temporary file in the same directory as the final
2309 * We want to avoid cross-directory filename renames, because those
2310 * can have problems on various filesystems (FAT, NFS, Coda).
2312 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2314 int fd
, dirlen
= directory_size(filename
);
2316 if (dirlen
+ 20 > bufsiz
) {
2317 errno
= ENAMETOOLONG
;
2320 memcpy(buffer
, filename
, dirlen
);
2321 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2322 fd
= mkstemp(buffer
);
2323 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2324 /* Make sure the directory exists */
2325 memcpy(buffer
, filename
, dirlen
);
2326 buffer
[dirlen
-1] = 0;
2327 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2331 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2332 fd
= mkstemp(buffer
);
2337 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2338 void *buf
, unsigned long len
, time_t mtime
)
2342 unsigned char *compressed
;
2345 static char tmpfile
[PATH_MAX
];
2347 filename
= sha1_file_name(sha1
);
2348 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2350 if (errno
== EACCES
)
2351 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2353 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2357 memset(&stream
, 0, sizeof(stream
));
2358 deflateInit(&stream
, zlib_compression_level
);
2359 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2360 compressed
= xmalloc(size
);
2363 stream
.next_out
= compressed
;
2364 stream
.avail_out
= size
;
2366 /* First header.. */
2367 stream
.next_in
= (unsigned char *)hdr
;
2368 stream
.avail_in
= hdrlen
;
2369 while (deflate(&stream
, 0) == Z_OK
)
2372 /* Then the data itself.. */
2373 stream
.next_in
= buf
;
2374 stream
.avail_in
= len
;
2375 ret
= deflate(&stream
, Z_FINISH
);
2376 if (ret
!= Z_STREAM_END
)
2377 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2379 ret
= deflateEnd(&stream
);
2381 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2383 size
= stream
.total_out
;
2385 if (write_buffer(fd
, compressed
, size
) < 0)
2386 die("unable to write sha1 file");
2387 close_sha1_file(fd
);
2393 utb
.modtime
= mtime
;
2394 if (utime(tmpfile
, &utb
) < 0)
2395 warning("failed utime() on %s: %s",
2396 tmpfile
, strerror(errno
));
2399 return move_temp_to_file(tmpfile
, filename
);
2402 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2404 unsigned char sha1
[20];
2408 /* Normally if we have it in the pack then we do not bother writing
2409 * it out into .git/objects/??/?{38} file.
2411 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2413 hashcpy(returnsha1
, sha1
);
2414 if (has_sha1_file(sha1
))
2416 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2419 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2423 enum object_type type
;
2428 if (has_loose_object(sha1
))
2430 buf
= read_packed_sha1(sha1
, &type
, &len
);
2432 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2433 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2434 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2440 int has_pack_index(const unsigned char *sha1
)
2443 if (stat(sha1_pack_index_name(sha1
), &st
))
2448 int has_pack_file(const unsigned char *sha1
)
2451 if (stat(sha1_pack_name(sha1
), &st
))
2456 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2458 struct pack_entry e
;
2459 return find_pack_entry(sha1
, &e
, ignore_packed
);
2462 int has_sha1_file(const unsigned char *sha1
)
2464 struct pack_entry e
;
2466 if (find_pack_entry(sha1
, &e
, NULL
))
2468 return has_loose_object(sha1
);
2471 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2472 int write_object
, enum object_type type
, const char *path
)
2474 int ret
, re_allocated
= 0;
2480 * Convert blobs to git internal format
2482 if ((type
== OBJ_BLOB
) && path
) {
2483 struct strbuf nbuf
= STRBUF_INIT
;
2484 if (convert_to_git(path
, buf
, size
, &nbuf
,
2485 write_object
? safe_crlf
: 0)) {
2486 buf
= strbuf_detach(&nbuf
, &size
);
2492 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2494 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2500 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2501 enum object_type type
, const char *path
)
2504 size_t size
= xsize_t(st
->st_size
);
2506 if (!S_ISREG(st
->st_mode
)) {
2507 struct strbuf sbuf
= STRBUF_INIT
;
2508 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2509 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2513 strbuf_release(&sbuf
);
2515 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2516 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2519 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2524 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2527 struct strbuf sb
= STRBUF_INIT
;
2529 switch (st
->st_mode
& S_IFMT
) {
2531 fd
= open(path
, O_RDONLY
);
2533 return error("open(\"%s\"): %s", path
,
2535 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2536 return error("%s: failed to insert into database",
2540 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2541 char *errstr
= strerror(errno
);
2542 return error("readlink(\"%s\"): %s", path
,
2546 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2547 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2548 return error("%s: failed to insert into database",
2550 strbuf_release(&sb
);
2553 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2555 return error("%s: unsupported file type", path
);
2560 int read_pack_header(int fd
, struct pack_header
*header
)
2562 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2563 /* "eof before pack header was fully read" */
2564 return PH_ERROR_EOF
;
2566 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2567 /* "protocol error (pack signature mismatch detected)" */
2568 return PH_ERROR_PACK_SIGNATURE
;
2569 if (!pack_version_ok(header
->hdr_version
))
2570 /* "protocol error (pack version unsupported)" */
2571 return PH_ERROR_PROTOCOL
;