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
)
258 const char *objdir
= get_object_directory();
259 struct alternate_object_database
*ent
;
260 struct alternate_object_database
*alt
;
261 /* 43 = 40-byte + 2 '/' + terminating NUL */
263 int entlen
= pfxlen
+ 43;
266 if (!is_absolute_path(entry
) && relative_base
) {
267 /* Relative alt-odb */
269 base_len
= strlen(relative_base
) + 1;
273 ent
= xmalloc(sizeof(*ent
) + entlen
);
275 if (!is_absolute_path(entry
) && relative_base
) {
276 memcpy(ent
->base
, relative_base
, base_len
- 1);
277 ent
->base
[base_len
- 1] = '/';
278 memcpy(ent
->base
+ base_len
, entry
, len
);
281 memcpy(ent
->base
, entry
, pfxlen
);
283 ent
->name
= ent
->base
+ pfxlen
+ 1;
284 ent
->base
[pfxlen
+ 3] = '/';
285 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
287 /* Detect cases where alternate disappeared */
288 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
289 error("object directory %s does not exist; "
290 "check .git/objects/info/alternates.",
296 /* Prevent the common mistake of listing the same
297 * thing twice, or object directory itself.
299 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
300 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
305 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
310 /* add the alternate entry */
312 alt_odb_tail
= &(ent
->next
);
315 /* recursively add alternates */
316 read_info_alternates(ent
->base
, depth
+ 1);
318 ent
->base
[pfxlen
] = '/';
323 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
324 const char *relative_base
, int depth
)
326 const char *cp
, *last
;
329 error("%s: ignoring alternate object stores, nesting too deep.",
337 if (cp
< ep
&& *cp
== '#') {
338 while (cp
< ep
&& *cp
!= sep
)
343 while (cp
< ep
&& *cp
!= sep
)
346 if (!is_absolute_path(last
) && depth
) {
347 error("%s: ignoring relative alternate object store %s",
348 relative_base
, last
);
350 link_alt_odb_entry(last
, cp
- last
,
351 relative_base
, depth
);
354 while (cp
< ep
&& *cp
== sep
)
360 static void read_info_alternates(const char * relative_base
, int depth
)
365 const char alt_file_name
[] = "info/alternates";
366 /* Given that relative_base is no longer than PATH_MAX,
367 ensure that "path" has enough space to append "/", the
368 file name, "info/alternates", and a trailing NUL. */
369 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
372 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
373 fd
= open(path
, O_RDONLY
);
376 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
380 mapsz
= xsize_t(st
.st_size
);
381 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
384 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
389 void add_to_alternates_file(const char *reference
)
391 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
392 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), 1);
393 char *alt
= mkpath("%s/objects\n", reference
);
394 write_or_die(fd
, alt
, strlen(alt
));
395 if (commit_lock_file(lock
))
396 die("could not close alternates file");
398 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
401 void prepare_alt_odb(void)
408 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
411 alt_odb_tail
= &alt_odb_list
;
412 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
414 read_info_alternates(get_object_directory(), 0);
417 static int has_loose_object(const unsigned char *sha1
)
419 char *name
= sha1_file_name(sha1
);
420 struct alternate_object_database
*alt
;
422 if (!access(name
, F_OK
))
425 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
427 fill_sha1_path(name
, sha1
);
428 if (!access(alt
->base
, F_OK
))
434 static unsigned int pack_used_ctr
;
435 static unsigned int pack_mmap_calls
;
436 static unsigned int peak_pack_open_windows
;
437 static unsigned int pack_open_windows
;
438 static size_t peak_pack_mapped
;
439 static size_t pack_mapped
;
440 struct packed_git
*packed_git
;
442 void pack_report(void)
445 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
446 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
447 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
448 sz_fmt(getpagesize()),
449 sz_fmt(packed_git_window_size
),
450 sz_fmt(packed_git_limit
));
452 "pack_report: pack_used_ctr = %10u\n"
453 "pack_report: pack_mmap_calls = %10u\n"
454 "pack_report: pack_open_windows = %10u / %10u\n"
455 "pack_report: pack_mapped = "
456 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
459 pack_open_windows
, peak_pack_open_windows
,
460 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
463 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
466 struct pack_idx_header
*hdr
;
468 uint32_t version
, nr
, i
, *index
;
469 int fd
= open(path
, O_RDONLY
);
474 if (fstat(fd
, &st
)) {
478 idx_size
= xsize_t(st
.st_size
);
479 if (idx_size
< 4 * 256 + 20 + 20) {
481 return error("index file %s is too small", path
);
483 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
487 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
488 version
= ntohl(hdr
->idx_version
);
489 if (version
< 2 || version
> 2) {
490 munmap(idx_map
, idx_size
);
491 return error("index file %s is version %"PRIu32
492 " and is not supported by this binary"
493 " (try upgrading GIT to a newer version)",
502 index
+= 2; /* skip index header */
503 for (i
= 0; i
< 256; i
++) {
504 uint32_t n
= ntohl(index
[i
]);
506 munmap(idx_map
, idx_size
);
507 return error("non-monotonic index %s", path
);
515 * - 256 index entries 4 bytes each
516 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
517 * - 20-byte SHA1 of the packfile
518 * - 20-byte SHA1 file checksum
520 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
521 munmap(idx_map
, idx_size
);
522 return error("wrong index v1 file size in %s", path
);
524 } else if (version
== 2) {
527 * - 8 bytes of header
528 * - 256 index entries 4 bytes each
529 * - 20-byte sha1 entry * nr
530 * - 4-byte crc entry * nr
531 * - 4-byte offset entry * nr
532 * - 20-byte SHA1 of the packfile
533 * - 20-byte SHA1 file checksum
534 * And after the 4-byte offset table might be a
535 * variable sized table containing 8-byte entries
536 * for offsets larger than 2^31.
538 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
539 unsigned long max_size
= min_size
;
541 max_size
+= (nr
- 1)*8;
542 if (idx_size
< min_size
|| idx_size
> max_size
) {
543 munmap(idx_map
, idx_size
);
544 return error("wrong index v2 file size in %s", path
);
546 if (idx_size
!= min_size
&&
548 * make sure we can deal with large pack offsets.
549 * 31-bit signed offset won't be enough, neither
550 * 32-bit unsigned one will be.
552 (sizeof(off_t
) <= 4)) {
553 munmap(idx_map
, idx_size
);
554 return error("pack too large for current definition of off_t in %s", path
);
558 p
->index_version
= version
;
559 p
->index_data
= idx_map
;
560 p
->index_size
= idx_size
;
565 int open_pack_index(struct packed_git
*p
)
573 idx_name
= xstrdup(p
->pack_name
);
574 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
575 ret
= check_packed_git_idx(idx_name
, p
);
580 static void scan_windows(struct packed_git
*p
,
581 struct packed_git
**lru_p
,
582 struct pack_window
**lru_w
,
583 struct pack_window
**lru_l
)
585 struct pack_window
*w
, *w_l
;
587 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
589 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
599 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
601 struct packed_git
*p
, *lru_p
= NULL
;
602 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
605 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
606 for (p
= packed_git
; p
; p
= p
->next
)
607 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
609 munmap(lru_w
->base
, lru_w
->len
);
610 pack_mapped
-= lru_w
->len
;
612 lru_l
->next
= lru_w
->next
;
614 lru_p
->windows
= lru_w
->next
;
615 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
616 close(lru_p
->pack_fd
);
627 void release_pack_memory(size_t need
, int fd
)
629 size_t cur
= pack_mapped
;
630 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
634 void close_pack_windows(struct packed_git
*p
)
637 struct pack_window
*w
= p
->windows
;
640 die("pack '%s' still has open windows to it",
642 munmap(w
->base
, w
->len
);
643 pack_mapped
-= w
->len
;
645 p
->windows
= w
->next
;
650 void unuse_pack(struct pack_window
**w_cursor
)
652 struct pack_window
*w
= *w_cursor
;
660 * Do not call this directly as this leaks p->pack_fd on error return;
661 * call open_packed_git() instead.
663 static int open_packed_git_1(struct packed_git
*p
)
666 struct pack_header hdr
;
667 unsigned char sha1
[20];
668 unsigned char *idx_sha1
;
671 if (!p
->index_data
&& open_pack_index(p
))
672 return error("packfile %s index unavailable", p
->pack_name
);
674 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
675 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
678 /* If we created the struct before we had the pack we lack size. */
680 if (!S_ISREG(st
.st_mode
))
681 return error("packfile %s not a regular file", p
->pack_name
);
682 p
->pack_size
= st
.st_size
;
683 } else if (p
->pack_size
!= st
.st_size
)
684 return error("packfile %s size changed", p
->pack_name
);
686 /* We leave these file descriptors open with sliding mmap;
687 * there is no point keeping them open across exec(), though.
689 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
691 return error("cannot determine file descriptor flags");
692 fd_flag
|= FD_CLOEXEC
;
693 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
694 return error("cannot set FD_CLOEXEC");
696 /* Verify we recognize this pack file format. */
697 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
698 return error("file %s is far too short to be a packfile", p
->pack_name
);
699 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
700 return error("file %s is not a GIT packfile", p
->pack_name
);
701 if (!pack_version_ok(hdr
.hdr_version
))
702 return error("packfile %s is version %"PRIu32
" and not"
703 " supported (try upgrading GIT to a newer version)",
704 p
->pack_name
, ntohl(hdr
.hdr_version
));
706 /* Verify the pack matches its index. */
707 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
708 return error("packfile %s claims to have %"PRIu32
" objects"
709 " while index indicates %"PRIu32
" objects",
710 p
->pack_name
, ntohl(hdr
.hdr_entries
),
712 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
713 return error("end of packfile %s is unavailable", p
->pack_name
);
714 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
715 return error("packfile %s signature is unavailable", p
->pack_name
);
716 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
717 if (hashcmp(sha1
, idx_sha1
))
718 return error("packfile %s does not match index", p
->pack_name
);
722 static int open_packed_git(struct packed_git
*p
)
724 if (!open_packed_git_1(p
))
726 if (p
->pack_fd
!= -1) {
733 static int in_window(struct pack_window
*win
, off_t offset
)
735 /* We must promise at least 20 bytes (one hash) after the
736 * offset is available from this window, otherwise the offset
737 * is not actually in this window and a different window (which
738 * has that one hash excess) must be used. This is to support
739 * the object header and delta base parsing routines below.
741 off_t win_off
= win
->offset
;
742 return win_off
<= offset
743 && (offset
+ 20) <= (win_off
+ win
->len
);
746 unsigned char* use_pack(struct packed_git
*p
,
747 struct pack_window
**w_cursor
,
751 struct pack_window
*win
= *w_cursor
;
753 if (p
->pack_fd
== -1 && open_packed_git(p
))
754 die("packfile %s cannot be accessed", p
->pack_name
);
756 /* Since packfiles end in a hash of their content and its
757 * pointless to ask for an offset into the middle of that
758 * hash, and the in_window function above wouldn't match
759 * don't allow an offset too close to the end of the file.
761 if (offset
> (p
->pack_size
- 20))
762 die("offset beyond end of packfile (truncated pack?)");
764 if (!win
|| !in_window(win
, offset
)) {
767 for (win
= p
->windows
; win
; win
= win
->next
) {
768 if (in_window(win
, offset
))
772 size_t window_align
= packed_git_window_size
/ 2;
774 win
= xcalloc(1, sizeof(*win
));
775 win
->offset
= (offset
/ window_align
) * window_align
;
776 len
= p
->pack_size
- win
->offset
;
777 if (len
> packed_git_window_size
)
778 len
= packed_git_window_size
;
779 win
->len
= (size_t)len
;
780 pack_mapped
+= win
->len
;
781 while (packed_git_limit
< pack_mapped
782 && unuse_one_window(p
, p
->pack_fd
))
784 win
->base
= xmmap(NULL
, win
->len
,
785 PROT_READ
, MAP_PRIVATE
,
786 p
->pack_fd
, win
->offset
);
787 if (win
->base
== MAP_FAILED
)
788 die("packfile %s cannot be mapped: %s",
793 if (pack_mapped
> peak_pack_mapped
)
794 peak_pack_mapped
= pack_mapped
;
795 if (pack_open_windows
> peak_pack_open_windows
)
796 peak_pack_open_windows
= pack_open_windows
;
797 win
->next
= p
->windows
;
801 if (win
!= *w_cursor
) {
802 win
->last_used
= pack_used_ctr
++;
806 offset
-= win
->offset
;
808 *left
= win
->len
- xsize_t(offset
);
809 return win
->base
+ offset
;
812 static struct packed_git
*alloc_packed_git(int extra
)
814 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
815 memset(p
, 0, sizeof(*p
));
820 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
823 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
826 * Make sure a corresponding .pack file exists and that
827 * the index looks sane.
829 path_len
-= strlen(".idx");
834 memcpy(p
->pack_name
, path
, path_len
);
835 strcpy(p
->pack_name
+ path_len
, ".pack");
836 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
841 /* ok, it looks sane as far as we can check without
842 * actually mapping the pack file.
844 p
->pack_size
= st
.st_size
;
845 p
->pack_local
= local
;
846 p
->mtime
= st
.st_mtime
;
847 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
852 struct packed_git
*parse_pack_index(unsigned char *sha1
)
854 const char *idx_path
= sha1_pack_index_name(sha1
);
855 const char *path
= sha1_pack_name(sha1
);
856 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
858 strcpy(p
->pack_name
, path
);
859 hashcpy(p
->sha1
, sha1
);
860 if (check_packed_git_idx(idx_path
, p
)) {
868 void install_packed_git(struct packed_git
*pack
)
870 pack
->next
= packed_git
;
874 static void prepare_packed_git_one(char *objdir
, int local
)
876 /* Ensure that this buffer is large enough so that we can
877 append "/pack/" without clobbering the stack even if
878 strlen(objdir) were PATH_MAX. */
879 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
884 sprintf(path
, "%s/pack", objdir
);
889 error("unable to open object pack directory: %s: %s",
890 path
, strerror(errno
));
894 while ((de
= readdir(dir
)) != NULL
) {
895 int namelen
= strlen(de
->d_name
);
896 struct packed_git
*p
;
898 if (!has_extension(de
->d_name
, ".idx"))
901 if (len
+ namelen
+ 1 > sizeof(path
))
904 /* Don't reopen a pack we already have. */
905 strcpy(path
+ len
, de
->d_name
);
906 for (p
= packed_git
; p
; p
= p
->next
) {
907 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
912 /* See if it really is a valid .idx file with corresponding
913 * .pack file that we can map.
915 p
= add_packed_git(path
, len
+ namelen
, local
);
918 install_packed_git(p
);
923 static int sort_pack(const void *a_
, const void *b_
)
925 struct packed_git
*a
= *((struct packed_git
**)a_
);
926 struct packed_git
*b
= *((struct packed_git
**)b_
);
930 * Local packs tend to contain objects specific to our
931 * variant of the project than remote ones. In addition,
932 * remote ones could be on a network mounted filesystem.
933 * Favor local ones for these reasons.
935 st
= a
->pack_local
- b
->pack_local
;
940 * Younger packs tend to contain more recent objects,
941 * and more recent objects tend to get accessed more
944 if (a
->mtime
< b
->mtime
)
946 else if (a
->mtime
== b
->mtime
)
951 static void rearrange_packed_git(void)
953 struct packed_git
**ary
, *p
;
956 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
961 /* prepare an array of packed_git for easier sorting */
962 ary
= xcalloc(n
, sizeof(struct packed_git
*));
963 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
966 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
968 /* link them back again */
969 for (i
= 0; i
< n
- 1; i
++)
970 ary
[i
]->next
= ary
[i
+ 1];
971 ary
[n
- 1]->next
= NULL
;
977 static int prepare_packed_git_run_once
= 0;
978 void prepare_packed_git(void)
980 struct alternate_object_database
*alt
;
982 if (prepare_packed_git_run_once
)
984 prepare_packed_git_one(get_object_directory(), 1);
986 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
988 prepare_packed_git_one(alt
->base
, 0);
991 rearrange_packed_git();
992 prepare_packed_git_run_once
= 1;
995 void reprepare_packed_git(void)
998 prepare_packed_git_run_once
= 0;
999 prepare_packed_git();
1002 static void mark_bad_packed_object(struct packed_git
*p
,
1003 const unsigned char *sha1
)
1006 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1007 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1009 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1010 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1011 p
->num_bad_objects
++;
1014 static int has_packed_and_bad(const unsigned char *sha1
)
1016 struct packed_git
*p
;
1019 for (p
= packed_git
; p
; p
= p
->next
)
1020 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1021 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1026 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1028 unsigned char real_sha1
[20];
1029 hash_sha1_file(map
, size
, type
, real_sha1
);
1030 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1033 static int git_open_noatime(const char *name
)
1035 static int sha1_file_open_flag
= O_NOATIME
;
1036 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1038 /* Might the failure be due to O_NOATIME? */
1039 if (fd
< 0 && errno
!= ENOENT
&& sha1_file_open_flag
) {
1040 fd
= open(name
, O_RDONLY
);
1042 sha1_file_open_flag
= 0;
1047 static int open_sha1_file(const unsigned char *sha1
)
1050 char *name
= sha1_file_name(sha1
);
1051 struct alternate_object_database
*alt
;
1053 fd
= git_open_noatime(name
);
1059 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1061 fill_sha1_path(name
, sha1
);
1062 fd
= git_open_noatime(alt
->base
);
1069 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1074 fd
= open_sha1_file(sha1
);
1079 if (!fstat(fd
, &st
)) {
1080 *size
= xsize_t(st
.st_size
);
1081 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1088 static int legacy_loose_object(unsigned char *map
)
1093 * Is it a zlib-compressed buffer? If so, the first byte
1094 * must be 0x78 (15-bit window size, deflated), and the
1095 * first 16-bit word is evenly divisible by 31
1097 word
= (map
[0] << 8) + map
[1];
1098 if (map
[0] == 0x78 && !(word
% 31))
1104 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1109 unsigned long used
= 0;
1112 *type
= (c
>> 4) & 7;
1118 if (sizeof(long) * 8 <= shift
)
1121 size
+= (c
& 0x7f) << shift
;
1128 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1130 unsigned long size
, used
;
1131 static const char valid_loose_object_type
[8] = {
1133 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1134 0, /* "delta" and others are invalid in a loose object */
1136 enum object_type type
;
1138 /* Get the data stream */
1139 memset(stream
, 0, sizeof(*stream
));
1140 stream
->next_in
= map
;
1141 stream
->avail_in
= mapsize
;
1142 stream
->next_out
= buffer
;
1143 stream
->avail_out
= bufsiz
;
1145 if (legacy_loose_object(map
)) {
1146 inflateInit(stream
);
1147 return inflate(stream
, 0);
1152 * There used to be a second loose object header format which
1153 * was meant to mimic the in-pack format, allowing for direct
1154 * copy of the object data. This format turned up not to be
1155 * really worth it and we don't write it any longer. But we
1156 * can still read it.
1158 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1159 if (!used
|| !valid_loose_object_type
[type
])
1164 /* Set up the stream for the rest.. */
1165 stream
->next_in
= map
;
1166 stream
->avail_in
= mapsize
;
1167 inflateInit(stream
);
1169 /* And generate the fake traditional header */
1170 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1171 typename(type
), size
);
1175 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1177 int bytes
= strlen(buffer
) + 1;
1178 unsigned char *buf
= xmalloc(1+size
);
1182 n
= stream
->total_out
- bytes
;
1185 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1187 if (bytes
<= size
) {
1189 * The above condition must be (bytes <= size), not
1190 * (bytes < size). In other words, even though we
1191 * expect no more output and set avail_out to zer0,
1192 * the input zlib stream may have bytes that express
1193 * "this concludes the stream", and we *do* want to
1196 * Otherwise we would not be able to test that we
1197 * consumed all the input to reach the expected size;
1198 * we also want to check that zlib tells us that all
1199 * went well with status == Z_STREAM_END at the end.
1201 stream
->next_out
= buf
+ bytes
;
1202 stream
->avail_out
= size
- bytes
;
1203 while (status
== Z_OK
)
1204 status
= inflate(stream
, Z_FINISH
);
1207 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1213 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1214 else if (stream
->avail_in
)
1215 error("garbage at end of loose object '%s'",
1222 * We used to just use "sscanf()", but that's actually way
1223 * too permissive for what we want to check. So do an anal
1224 * object header parse by hand.
1226 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1233 * The type can be at most ten bytes (including the
1234 * terminating '\0' that we add), and is followed by
1243 if (i
>= sizeof(type
))
1249 * The length must follow immediately, and be in canonical
1250 * decimal format (ie "010" is not valid).
1252 size
= *hdr
++ - '0';
1257 unsigned long c
= *hdr
- '0';
1261 size
= size
* 10 + c
;
1267 * The length must be followed by a zero byte
1269 return *hdr
? -1 : type_from_string(type
);
1272 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1278 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1279 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1282 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1285 unsigned long get_size_from_delta(struct packed_git
*p
,
1286 struct pack_window
**w_curs
,
1289 const unsigned char *data
;
1290 unsigned char delta_head
[20], *in
;
1294 memset(&stream
, 0, sizeof(stream
));
1295 stream
.next_out
= delta_head
;
1296 stream
.avail_out
= sizeof(delta_head
);
1298 inflateInit(&stream
);
1300 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1301 stream
.next_in
= in
;
1302 st
= inflate(&stream
, Z_FINISH
);
1303 curpos
+= stream
.next_in
- in
;
1304 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1305 stream
.total_out
< sizeof(delta_head
));
1306 inflateEnd(&stream
);
1307 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1308 die("delta data unpack-initial failed");
1310 /* Examine the initial part of the delta to figure out
1315 /* ignore base size */
1316 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1318 /* Read the result size */
1319 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1322 static off_t
get_delta_base(struct packed_git
*p
,
1323 struct pack_window
**w_curs
,
1325 enum object_type type
,
1326 off_t delta_obj_offset
)
1328 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1331 /* use_pack() assured us we have [base_info, base_info + 20)
1332 * as a range that we can look at without walking off the
1333 * end of the mapped window. Its actually the hash size
1334 * that is assured. An OFS_DELTA longer than the hash size
1335 * is stupid, as then a REF_DELTA would be smaller to store.
1337 if (type
== OBJ_OFS_DELTA
) {
1339 unsigned char c
= base_info
[used
++];
1340 base_offset
= c
& 127;
1343 if (!base_offset
|| MSB(base_offset
, 7))
1344 return 0; /* overflow */
1345 c
= base_info
[used
++];
1346 base_offset
= (base_offset
<< 7) + (c
& 127);
1348 base_offset
= delta_obj_offset
- base_offset
;
1349 if (base_offset
>= delta_obj_offset
)
1350 return 0; /* out of bound */
1352 } else if (type
== OBJ_REF_DELTA
) {
1353 /* The base entry _must_ be in the same pack */
1354 base_offset
= find_pack_entry_one(base_info
, p
);
1357 die("I am totally screwed");
1361 /* forward declaration for a mutually recursive function */
1362 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1363 unsigned long *sizep
);
1365 static int packed_delta_info(struct packed_git
*p
,
1366 struct pack_window
**w_curs
,
1368 enum object_type type
,
1370 unsigned long *sizep
)
1374 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1375 type
= packed_object_info(p
, base_offset
, NULL
);
1377 /* We choose to only get the type of the base object and
1378 * ignore potentially corrupt pack file that expects the delta
1379 * based on a base with a wrong size. This saves tons of
1383 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1388 static int unpack_object_header(struct packed_git
*p
,
1389 struct pack_window
**w_curs
,
1391 unsigned long *sizep
)
1393 unsigned char *base
;
1396 enum object_type type
;
1398 /* use_pack() assures us we have [base, base + 20) available
1399 * as a range that we can look at at. (Its actually the hash
1400 * size that is assured.) With our object header encoding
1401 * the maximum deflated object size is 2^137, which is just
1402 * insane, so we know won't exceed what we have been given.
1404 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1405 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1407 die("object offset outside of pack file");
1413 const char *packed_object_info_detail(struct packed_git
*p
,
1415 unsigned long *size
,
1416 unsigned long *store_size
,
1417 unsigned int *delta_chain_length
,
1418 unsigned char *base_sha1
)
1420 struct pack_window
*w_curs
= NULL
;
1422 unsigned long dummy
;
1423 unsigned char *next_sha1
;
1424 enum object_type type
;
1425 struct revindex_entry
*revidx
;
1427 *delta_chain_length
= 0;
1428 curpos
= obj_offset
;
1429 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1431 revidx
= find_pack_revindex(p
, obj_offset
);
1432 *store_size
= revidx
[1].offset
- obj_offset
;
1437 die("pack %s contains unknown object type %d",
1438 p
->pack_name
, type
);
1443 unuse_pack(&w_curs
);
1444 return typename(type
);
1446 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1448 die("pack %s contains bad delta base reference of type %s",
1449 p
->pack_name
, typename(type
));
1450 if (*delta_chain_length
== 0) {
1451 revidx
= find_pack_revindex(p
, obj_offset
);
1452 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1456 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1457 if (*delta_chain_length
== 0)
1458 hashcpy(base_sha1
, next_sha1
);
1459 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1462 (*delta_chain_length
)++;
1463 curpos
= obj_offset
;
1464 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1468 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1469 unsigned long *sizep
)
1471 struct pack_window
*w_curs
= NULL
;
1473 off_t curpos
= obj_offset
;
1474 enum object_type type
;
1476 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1481 type
= packed_delta_info(p
, &w_curs
, curpos
,
1482 type
, obj_offset
, sizep
);
1492 die("pack %s contains unknown object type %d",
1493 p
->pack_name
, type
);
1495 unuse_pack(&w_curs
);
1499 static void *unpack_compressed_entry(struct packed_git
*p
,
1500 struct pack_window
**w_curs
,
1506 unsigned char *buffer
, *in
;
1508 buffer
= xmalloc(size
+ 1);
1510 memset(&stream
, 0, sizeof(stream
));
1511 stream
.next_out
= buffer
;
1512 stream
.avail_out
= size
;
1514 inflateInit(&stream
);
1516 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1517 stream
.next_in
= in
;
1518 st
= inflate(&stream
, Z_FINISH
);
1519 curpos
+= stream
.next_in
- in
;
1520 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1521 inflateEnd(&stream
);
1522 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1530 #define MAX_DELTA_CACHE (256)
1532 static size_t delta_base_cached
;
1534 static struct delta_base_cache_lru_list
{
1535 struct delta_base_cache_lru_list
*prev
;
1536 struct delta_base_cache_lru_list
*next
;
1537 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1539 static struct delta_base_cache_entry
{
1540 struct delta_base_cache_lru_list lru
;
1542 struct packed_git
*p
;
1545 enum object_type type
;
1546 } delta_base_cache
[MAX_DELTA_CACHE
];
1548 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1552 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1553 hash
+= (hash
>> 8) + (hash
>> 16);
1554 return hash
% MAX_DELTA_CACHE
;
1557 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1558 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1561 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1562 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1565 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1566 goto found_cache_entry
;
1567 return unpack_entry(p
, base_offset
, type
, base_size
);
1572 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1573 ent
->lru
.prev
->next
= ent
->lru
.next
;
1574 delta_base_cached
-= ent
->size
;
1576 ret
= xmemdupz(ent
->data
, ent
->size
);
1579 *base_size
= ent
->size
;
1583 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1588 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1589 ent
->lru
.prev
->next
= ent
->lru
.next
;
1590 delta_base_cached
-= ent
->size
;
1594 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1595 void *base
, unsigned long base_size
, enum object_type type
)
1597 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1598 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1599 struct delta_base_cache_lru_list
*lru
;
1601 release_delta_base_cache(ent
);
1602 delta_base_cached
+= base_size
;
1604 for (lru
= delta_base_cache_lru
.next
;
1605 delta_base_cached
> delta_base_cache_limit
1606 && lru
!= &delta_base_cache_lru
;
1608 struct delta_base_cache_entry
*f
= (void *)lru
;
1609 if (f
->type
== OBJ_BLOB
)
1610 release_delta_base_cache(f
);
1612 for (lru
= delta_base_cache_lru
.next
;
1613 delta_base_cached
> delta_base_cache_limit
1614 && lru
!= &delta_base_cache_lru
;
1616 struct delta_base_cache_entry
*f
= (void *)lru
;
1617 release_delta_base_cache(f
);
1621 ent
->base_offset
= base_offset
;
1624 ent
->size
= base_size
;
1625 ent
->lru
.next
= &delta_base_cache_lru
;
1626 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1627 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1628 delta_base_cache_lru
.prev
= &ent
->lru
;
1631 static void *unpack_delta_entry(struct packed_git
*p
,
1632 struct pack_window
**w_curs
,
1634 unsigned long delta_size
,
1636 enum object_type
*type
,
1637 unsigned long *sizep
)
1639 void *delta_data
, *result
, *base
;
1640 unsigned long base_size
;
1643 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1645 error("failed to validate delta base reference "
1646 "at offset %"PRIuMAX
" from %s",
1647 (uintmax_t)curpos
, p
->pack_name
);
1651 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1654 * We're probably in deep shit, but let's try to fetch
1655 * the required base anyway from another pack or loose.
1656 * This is costly but should happen only in the presence
1657 * of a corrupted pack, and is better than failing outright.
1659 struct revindex_entry
*revidx
= find_pack_revindex(p
, base_offset
);
1660 const unsigned char *base_sha1
=
1661 nth_packed_object_sha1(p
, revidx
->nr
);
1662 error("failed to read delta base object %s"
1663 " at offset %"PRIuMAX
" from %s",
1664 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1666 mark_bad_packed_object(p
, base_sha1
);
1667 base
= read_object(base_sha1
, type
, &base_size
);
1672 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1674 error("failed to unpack compressed delta "
1675 "at offset %"PRIuMAX
" from %s",
1676 (uintmax_t)curpos
, p
->pack_name
);
1680 result
= patch_delta(base
, base_size
,
1681 delta_data
, delta_size
,
1684 die("failed to apply delta");
1686 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1690 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1691 enum object_type
*type
, unsigned long *sizep
)
1693 struct pack_window
*w_curs
= NULL
;
1694 off_t curpos
= obj_offset
;
1697 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1701 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1702 obj_offset
, type
, sizep
);
1708 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1712 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1713 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1715 unuse_pack(&w_curs
);
1719 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1722 const unsigned char *index
= p
->index_data
;
1724 if (open_pack_index(p
))
1726 index
= p
->index_data
;
1728 if (n
>= p
->num_objects
)
1731 if (p
->index_version
== 1) {
1732 return index
+ 24 * n
+ 4;
1735 return index
+ 20 * n
;
1739 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1741 const unsigned char *index
= p
->index_data
;
1743 if (p
->index_version
== 1) {
1744 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1747 index
+= 8 + p
->num_objects
* (20 + 4);
1748 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1749 if (!(off
& 0x80000000))
1751 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1752 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1753 ntohl(*((uint32_t *)(index
+ 4)));
1757 off_t
find_pack_entry_one(const unsigned char *sha1
,
1758 struct packed_git
*p
)
1760 const uint32_t *level1_ofs
= p
->index_data
;
1761 const unsigned char *index
= p
->index_data
;
1762 unsigned hi
, lo
, stride
;
1763 static int use_lookup
= -1;
1764 static int debug_lookup
= -1;
1766 if (debug_lookup
< 0)
1767 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1770 if (open_pack_index(p
))
1772 level1_ofs
= p
->index_data
;
1773 index
= p
->index_data
;
1775 if (p
->index_version
> 1) {
1780 hi
= ntohl(level1_ofs
[*sha1
]);
1781 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1782 if (p
->index_version
> 1) {
1790 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1791 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1794 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1796 int pos
= sha1_entry_pos(index
, stride
, 0,
1797 lo
, hi
, p
->num_objects
, sha1
);
1800 return nth_packed_object_offset(p
, pos
);
1804 unsigned mi
= (lo
+ hi
) / 2;
1805 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1808 printf("lo %u hi %u rg %u mi %u\n",
1809 lo
, hi
, hi
- lo
, mi
);
1811 return nth_packed_object_offset(p
, mi
);
1820 int matches_pack_name(struct packed_git
*p
, const char *name
)
1822 const char *last_c
, *c
;
1824 if (!strcmp(p
->pack_name
, name
))
1827 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1832 if (!strcmp(last_c
, name
))
1838 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1840 static struct packed_git
*last_found
= (void *)1;
1841 struct packed_git
*p
;
1844 prepare_packed_git();
1847 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1850 if (ignore_packed
) {
1852 for (ig
= ignore_packed
; *ig
; ig
++)
1853 if (matches_pack_name(p
, *ig
))
1859 if (p
->num_bad_objects
) {
1861 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1862 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1866 offset
= find_pack_entry_one(sha1
, p
);
1869 * We are about to tell the caller where they can
1870 * locate the requested object. We better make
1871 * sure the packfile is still here and can be
1872 * accessed before supplying that answer, as
1873 * it may have been deleted since the index
1876 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1877 error("packfile %s cannot be accessed", p
->pack_name
);
1882 hashcpy(e
->sha1
, sha1
);
1888 if (p
== last_found
)
1892 if (p
== last_found
)
1898 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1899 struct packed_git
*packs
)
1901 struct packed_git
*p
;
1903 for (p
= packs
; p
; p
= p
->next
) {
1904 if (find_pack_entry_one(sha1
, p
))
1911 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1914 unsigned long mapsize
, size
;
1919 map
= map_sha1_file(sha1
, &mapsize
);
1921 return error("unable to find %s", sha1_to_hex(sha1
));
1922 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1923 status
= error("unable to unpack %s header",
1925 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1926 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1929 inflateEnd(&stream
);
1930 munmap(map
, mapsize
);
1934 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1936 struct pack_entry e
;
1939 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1940 /* Most likely it's a loose object. */
1941 status
= sha1_loose_object_info(sha1
, sizep
);
1945 /* Not a loose object; someone else may have just packed it. */
1946 reprepare_packed_git();
1947 if (!find_pack_entry(sha1
, &e
, NULL
))
1950 return packed_object_info(e
.p
, e
.offset
, sizep
);
1953 static void *read_packed_sha1(const unsigned char *sha1
,
1954 enum object_type
*type
, unsigned long *size
)
1956 struct pack_entry e
;
1959 if (!find_pack_entry(sha1
, &e
, NULL
))
1961 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1964 * We're probably in deep shit, but let's try to fetch
1965 * the required object anyway from another pack or loose.
1966 * This should happen only in the presence of a corrupted
1967 * pack, and is better than failing outright.
1969 error("failed to read object %s at offset %"PRIuMAX
" from %s",
1970 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
1971 mark_bad_packed_object(e
.p
, sha1
);
1972 data
= read_object(sha1
, type
, size
);
1978 * This is meant to hold a *small* number of objects that you would
1979 * want read_sha1_file() to be able to return, but yet you do not want
1980 * to write them into the object store (e.g. a browse-only
1983 static struct cached_object
{
1984 unsigned char sha1
[20];
1985 enum object_type type
;
1989 static int cached_object_nr
, cached_object_alloc
;
1991 static struct cached_object empty_tree
= {
1992 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1993 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1994 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2000 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
2003 struct cached_object
*co
= cached_objects
;
2005 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
2006 if (!hashcmp(co
->sha1
, sha1
))
2009 if (!hashcmp(sha1
, empty_tree
.sha1
))
2014 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2015 unsigned char *sha1
)
2017 struct cached_object
*co
;
2019 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2020 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2022 if (cached_object_alloc
<= cached_object_nr
) {
2023 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2024 cached_objects
= xrealloc(cached_objects
,
2025 sizeof(*cached_objects
) *
2026 cached_object_alloc
);
2028 co
= &cached_objects
[cached_object_nr
++];
2031 co
->buf
= xmalloc(len
);
2032 memcpy(co
->buf
, buf
, len
);
2033 hashcpy(co
->sha1
, sha1
);
2037 void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2038 unsigned long *size
)
2040 unsigned long mapsize
;
2042 struct cached_object
*co
;
2044 co
= find_cached_object(sha1
);
2048 return xmemdupz(co
->buf
, co
->size
);
2051 buf
= read_packed_sha1(sha1
, type
, size
);
2054 map
= map_sha1_file(sha1
, &mapsize
);
2056 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2057 munmap(map
, mapsize
);
2060 reprepare_packed_git();
2061 return read_packed_sha1(sha1
, type
, size
);
2064 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
2065 unsigned long *size
)
2067 void *data
= read_object(sha1
, type
, size
);
2068 /* legacy behavior is to die on corrupted objects */
2069 if (!data
&& (has_loose_object(sha1
) || has_packed_and_bad(sha1
)))
2070 die("object %s is corrupted", sha1_to_hex(sha1
));
2074 void *read_object_with_reference(const unsigned char *sha1
,
2075 const char *required_type_name
,
2076 unsigned long *size
,
2077 unsigned char *actual_sha1_return
)
2079 enum object_type type
, required_type
;
2081 unsigned long isize
;
2082 unsigned char actual_sha1
[20];
2084 required_type
= type_from_string(required_type_name
);
2085 hashcpy(actual_sha1
, sha1
);
2087 int ref_length
= -1;
2088 const char *ref_type
= NULL
;
2090 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2093 if (type
== required_type
) {
2095 if (actual_sha1_return
)
2096 hashcpy(actual_sha1_return
, actual_sha1
);
2099 /* Handle references */
2100 else if (type
== OBJ_COMMIT
)
2102 else if (type
== OBJ_TAG
)
2103 ref_type
= "object ";
2108 ref_length
= strlen(ref_type
);
2110 if (ref_length
+ 40 > isize
||
2111 memcmp(buffer
, ref_type
, ref_length
) ||
2112 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2117 /* Now we have the ID of the referred-to object in
2118 * actual_sha1. Check again. */
2122 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2123 const char *type
, unsigned char *sha1
,
2124 char *hdr
, int *hdrlen
)
2128 /* Generate the header */
2129 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2133 SHA1_Update(&c
, hdr
, *hdrlen
);
2134 SHA1_Update(&c
, buf
, len
);
2135 SHA1_Final(sha1
, &c
);
2139 * Move the just written object into its final resting place
2141 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2144 if (link(tmpfile
, filename
))
2148 * Coda hack - coda doesn't like cross-directory links,
2149 * so we fall back to a rename, which will mean that it
2150 * won't be able to check collisions, but that's not a
2153 * The same holds for FAT formatted media.
2155 * When this succeeds, we just return 0. We have nothing
2158 if (ret
&& ret
!= EEXIST
) {
2159 if (!rename(tmpfile
, filename
))
2165 if (ret
!= EEXIST
) {
2166 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2168 /* FIXME!!! Collision check here ? */
2174 static int write_buffer(int fd
, const void *buf
, size_t len
)
2176 if (write_in_full(fd
, buf
, len
) < 0)
2177 return error("file write error (%s)", strerror(errno
));
2181 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2182 unsigned char *sha1
)
2186 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2190 /* Finalize a file on disk, and close it. */
2191 static void close_sha1_file(int fd
)
2193 if (fsync_object_files
)
2194 fsync_or_die(fd
, "sha1 file");
2197 die("unable to write sha1 file");
2200 /* Size of directory component, including the ending '/' */
2201 static inline int directory_size(const char *filename
)
2203 const char *s
= strrchr(filename
, '/');
2206 return s
- filename
+ 1;
2210 * This creates a temporary file in the same directory as the final
2213 * We want to avoid cross-directory filename renames, because those
2214 * can have problems on various filesystems (FAT, NFS, Coda).
2216 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2218 int fd
, dirlen
= directory_size(filename
);
2220 if (dirlen
+ 20 > bufsiz
) {
2221 errno
= ENAMETOOLONG
;
2224 memcpy(buffer
, filename
, dirlen
);
2225 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2226 fd
= mkstemp(buffer
);
2227 if (fd
< 0 && dirlen
) {
2228 /* Make sure the directory exists */
2229 memcpy(buffer
, filename
, dirlen
);
2230 buffer
[dirlen
-1] = 0;
2231 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2235 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2236 fd
= mkstemp(buffer
);
2241 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2242 void *buf
, unsigned long len
, time_t mtime
)
2245 unsigned char *compressed
;
2248 static char tmpfile
[PATH_MAX
];
2250 filename
= sha1_file_name(sha1
);
2251 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2254 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2256 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2260 memset(&stream
, 0, sizeof(stream
));
2261 deflateInit(&stream
, zlib_compression_level
);
2262 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2263 compressed
= xmalloc(size
);
2266 stream
.next_out
= compressed
;
2267 stream
.avail_out
= size
;
2269 /* First header.. */
2270 stream
.next_in
= (unsigned char *)hdr
;
2271 stream
.avail_in
= hdrlen
;
2272 while (deflate(&stream
, 0) == Z_OK
)
2275 /* Then the data itself.. */
2276 stream
.next_in
= buf
;
2277 stream
.avail_in
= len
;
2278 ret
= deflate(&stream
, Z_FINISH
);
2279 if (ret
!= Z_STREAM_END
)
2280 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2282 ret
= deflateEnd(&stream
);
2284 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2286 size
= stream
.total_out
;
2288 if (write_buffer(fd
, compressed
, size
) < 0)
2289 die("unable to write sha1 file");
2290 close_sha1_file(fd
);
2296 utb
.modtime
= mtime
;
2297 if (utime(tmpfile
, &utb
) < 0)
2298 warning("failed utime() on %s: %s",
2299 tmpfile
, strerror(errno
));
2302 return move_temp_to_file(tmpfile
, filename
);
2305 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2307 unsigned char sha1
[20];
2311 /* Normally if we have it in the pack then we do not bother writing
2312 * it out into .git/objects/??/?{38} file.
2314 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2316 hashcpy(returnsha1
, sha1
);
2317 if (has_sha1_file(sha1
))
2319 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2322 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2326 enum object_type type
;
2330 if (has_loose_object(sha1
))
2332 buf
= read_packed_sha1(sha1
, &type
, &len
);
2334 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2335 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2336 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2339 int has_pack_index(const unsigned char *sha1
)
2342 if (stat(sha1_pack_index_name(sha1
), &st
))
2347 int has_pack_file(const unsigned char *sha1
)
2350 if (stat(sha1_pack_name(sha1
), &st
))
2355 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2357 struct pack_entry e
;
2358 return find_pack_entry(sha1
, &e
, ignore_packed
);
2361 int has_sha1_file(const unsigned char *sha1
)
2363 struct pack_entry e
;
2365 if (find_pack_entry(sha1
, &e
, NULL
))
2367 return has_loose_object(sha1
);
2370 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2371 int write_object
, enum object_type type
, const char *path
)
2373 int ret
, re_allocated
= 0;
2379 * Convert blobs to git internal format
2381 if ((type
== OBJ_BLOB
) && path
) {
2383 strbuf_init(&nbuf
, 0);
2384 if (convert_to_git(path
, buf
, size
, &nbuf
,
2385 write_object
? safe_crlf
: 0)) {
2386 buf
= strbuf_detach(&nbuf
, &size
);
2392 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2394 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2400 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2401 enum object_type type
, const char *path
)
2404 size_t size
= xsize_t(st
->st_size
);
2406 if (!S_ISREG(st
->st_mode
)) {
2408 strbuf_init(&sbuf
, 0);
2409 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2410 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2414 strbuf_release(&sbuf
);
2416 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2417 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2420 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2425 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2431 switch (st
->st_mode
& S_IFMT
) {
2433 fd
= open(path
, O_RDONLY
);
2435 return error("open(\"%s\"): %s", path
,
2437 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2438 return error("%s: failed to insert into database",
2442 len
= xsize_t(st
->st_size
);
2443 target
= xmalloc(len
+ 1);
2444 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2445 char *errstr
= strerror(errno
);
2447 return error("readlink(\"%s\"): %s", path
,
2451 hash_sha1_file(target
, len
, blob_type
, sha1
);
2452 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2453 return error("%s: failed to insert into database",
2458 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2460 return error("%s: unsupported file type", path
);
2465 int read_pack_header(int fd
, struct pack_header
*header
)
2467 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2468 /* "eof before pack header was fully read" */
2469 return PH_ERROR_EOF
;
2471 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2472 /* "protocol error (pack signature mismatch detected)" */
2473 return PH_ERROR_PACK_SIGNATURE
;
2474 if (!pack_version_ok(header
->hdr_version
))
2475 /* "protocol error (pack version unsupported)" */
2476 return PH_ERROR_PROTOCOL
;