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 int safe_create_leading_directories(char *path
)
88 char *pos
= path
+ offset_1st_component(path
);
92 pos
= strchr(pos
, '/');
100 if (!stat(path
, &st
)) {
102 if (!S_ISDIR(st
.st_mode
)) {
107 else if (mkdir(path
, 0777)) {
111 else if (adjust_shared_perm(path
)) {
120 int safe_create_leading_directories_const(const char *path
)
122 /* path points to cache entries, so xstrdup before messing with it */
123 char *buf
= xstrdup(path
);
124 int result
= safe_create_leading_directories(buf
);
129 char *sha1_to_hex(const unsigned char *sha1
)
132 static char hexbuffer
[4][50];
133 static const char hex
[] = "0123456789abcdef";
134 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
137 for (i
= 0; i
< 20; i
++) {
138 unsigned int val
= *sha1
++;
139 *buf
++ = hex
[val
>> 4];
140 *buf
++ = hex
[val
& 0xf];
147 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
150 for (i
= 0; i
< 20; i
++) {
151 static char hex
[] = "0123456789abcdef";
152 unsigned int val
= sha1
[i
];
153 char *pos
= pathbuf
+ i
*2 + (i
> 0);
154 *pos
++ = hex
[val
>> 4];
155 *pos
= hex
[val
& 0xf];
160 * NOTE! This returns a statically allocated buffer, so you have to be
161 * careful about using it. Do an "xstrdup()" if you need to save the
164 * Also note that this returns the location for creating. Reading
165 * SHA1 file can happen from any alternate directory listed in the
166 * DB_ENVIRONMENT environment variable if it is not found in
167 * the primary object database.
169 char *sha1_file_name(const unsigned char *sha1
)
171 static char *name
, *base
;
174 const char *sha1_file_directory
= get_object_directory();
175 int len
= strlen(sha1_file_directory
);
176 base
= xmalloc(len
+ 60);
177 memcpy(base
, sha1_file_directory
, len
);
178 memset(base
+len
, 0, 60);
181 name
= base
+ len
+ 1;
183 fill_sha1_path(name
, sha1
);
187 static char *sha1_get_pack_name(const unsigned char *sha1
,
188 char **name
, char **base
, const char *which
)
190 static const char hex
[] = "0123456789abcdef";
195 const char *sha1_file_directory
= get_object_directory();
196 int len
= strlen(sha1_file_directory
);
197 *base
= xmalloc(len
+ 60);
198 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
199 sha1_file_directory
, which
);
200 *name
= *base
+ len
+ 11;
205 for (i
= 0; i
< 20; i
++) {
206 unsigned int val
= *sha1
++;
207 *buf
++ = hex
[val
>> 4];
208 *buf
++ = hex
[val
& 0xf];
214 char *sha1_pack_name(const unsigned char *sha1
)
216 static char *name
, *base
;
218 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
221 char *sha1_pack_index_name(const unsigned char *sha1
)
223 static char *name
, *base
;
225 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
228 struct alternate_object_database
*alt_odb_list
;
229 static struct alternate_object_database
**alt_odb_tail
;
231 static void read_info_alternates(const char * alternates
, int depth
);
234 * Prepare alternate object database registry.
236 * The variable alt_odb_list points at the list of struct
237 * alternate_object_database. The elements on this list come from
238 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
239 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
240 * whose contents is similar to that environment variable but can be
241 * LF separated. Its base points at a statically allocated buffer that
242 * contains "/the/directory/corresponding/to/.git/objects/...", while
243 * its name points just after the slash at the end of ".git/objects/"
244 * in the example above, and has enough space to hold 40-byte hex
245 * SHA1, an extra slash for the first level indirection, and the
248 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
250 const char *objdir
= get_object_directory();
251 struct alternate_object_database
*ent
;
252 struct alternate_object_database
*alt
;
253 /* 43 = 40-byte + 2 '/' + terminating NUL */
255 int entlen
= pfxlen
+ 43;
258 if (!is_absolute_path(entry
) && relative_base
) {
259 /* Relative alt-odb */
261 base_len
= strlen(relative_base
) + 1;
265 ent
= xmalloc(sizeof(*ent
) + entlen
);
267 if (!is_absolute_path(entry
) && relative_base
) {
268 memcpy(ent
->base
, relative_base
, base_len
- 1);
269 ent
->base
[base_len
- 1] = '/';
270 memcpy(ent
->base
+ base_len
, entry
, len
);
273 memcpy(ent
->base
, entry
, pfxlen
);
275 ent
->name
= ent
->base
+ pfxlen
+ 1;
276 ent
->base
[pfxlen
+ 3] = '/';
277 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
279 /* Detect cases where alternate disappeared */
280 if (!is_directory(ent
->base
)) {
281 error("object directory %s does not exist; "
282 "check .git/objects/info/alternates.",
288 /* Prevent the common mistake of listing the same
289 * thing twice, or object directory itself.
291 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
292 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
297 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
302 /* add the alternate entry */
304 alt_odb_tail
= &(ent
->next
);
307 /* recursively add alternates */
308 read_info_alternates(ent
->base
, depth
+ 1);
310 ent
->base
[pfxlen
] = '/';
315 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
316 const char *relative_base
, int depth
)
318 const char *cp
, *last
;
321 error("%s: ignoring alternate object stores, nesting too deep.",
329 if (cp
< ep
&& *cp
== '#') {
330 while (cp
< ep
&& *cp
!= sep
)
335 while (cp
< ep
&& *cp
!= sep
)
338 if (!is_absolute_path(last
) && depth
) {
339 error("%s: ignoring relative alternate object store %s",
340 relative_base
, last
);
342 link_alt_odb_entry(last
, cp
- last
,
343 relative_base
, depth
);
346 while (cp
< ep
&& *cp
== sep
)
352 static void read_info_alternates(const char * relative_base
, int depth
)
357 const char alt_file_name
[] = "info/alternates";
358 /* Given that relative_base is no longer than PATH_MAX,
359 ensure that "path" has enough space to append "/", the
360 file name, "info/alternates", and a trailing NUL. */
361 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
364 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
365 fd
= open(path
, O_RDONLY
);
368 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
372 mapsz
= xsize_t(st
.st_size
);
373 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
376 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
381 void add_to_alternates_file(const char *reference
)
383 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
384 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
385 char *alt
= mkpath("%s/objects\n", reference
);
386 write_or_die(fd
, alt
, strlen(alt
));
387 if (commit_lock_file(lock
))
388 die("could not close alternates file");
390 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
393 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
395 struct alternate_object_database
*ent
;
398 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
403 void prepare_alt_odb(void)
410 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
413 alt_odb_tail
= &alt_odb_list
;
414 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
416 read_info_alternates(get_object_directory(), 0);
419 static int has_loose_object_local(const unsigned char *sha1
)
421 char *name
= sha1_file_name(sha1
);
422 return !access(name
, F_OK
);
425 int has_loose_object_nonlocal(const unsigned char *sha1
)
427 struct alternate_object_database
*alt
;
429 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
430 fill_sha1_path(alt
->name
, sha1
);
431 if (!access(alt
->base
, F_OK
))
437 static int has_loose_object(const unsigned char *sha1
)
439 return has_loose_object_local(sha1
) ||
440 has_loose_object_nonlocal(sha1
);
443 static unsigned int pack_used_ctr
;
444 static unsigned int pack_mmap_calls
;
445 static unsigned int peak_pack_open_windows
;
446 static unsigned int pack_open_windows
;
447 static size_t peak_pack_mapped
;
448 static size_t pack_mapped
;
449 struct packed_git
*packed_git
;
451 void pack_report(void)
454 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size
),
459 sz_fmt(packed_git_limit
));
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
468 pack_open_windows
, peak_pack_open_windows
,
469 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
472 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
475 struct pack_idx_header
*hdr
;
477 uint32_t version
, nr
, i
, *index
;
478 int fd
= open(path
, O_RDONLY
);
483 if (fstat(fd
, &st
)) {
487 idx_size
= xsize_t(st
.st_size
);
488 if (idx_size
< 4 * 256 + 20 + 20) {
490 return error("index file %s is too small", path
);
492 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
496 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
497 version
= ntohl(hdr
->idx_version
);
498 if (version
< 2 || version
> 2) {
499 munmap(idx_map
, idx_size
);
500 return error("index file %s is version %"PRIu32
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
511 index
+= 2; /* skip index header */
512 for (i
= 0; i
< 256; i
++) {
513 uint32_t n
= ntohl(index
[i
]);
515 munmap(idx_map
, idx_size
);
516 return error("non-monotonic index %s", path
);
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
529 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
530 munmap(idx_map
, idx_size
);
531 return error("wrong index v1 file size in %s", path
);
533 } else if (version
== 2) {
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
547 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
548 unsigned long max_size
= min_size
;
550 max_size
+= (nr
- 1)*8;
551 if (idx_size
< min_size
|| idx_size
> max_size
) {
552 munmap(idx_map
, idx_size
);
553 return error("wrong index v2 file size in %s", path
);
555 if (idx_size
!= min_size
&&
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
561 (sizeof(off_t
) <= 4)) {
562 munmap(idx_map
, idx_size
);
563 return error("pack too large for current definition of off_t in %s", path
);
567 p
->index_version
= version
;
568 p
->index_data
= idx_map
;
569 p
->index_size
= idx_size
;
574 int open_pack_index(struct packed_git
*p
)
582 idx_name
= xstrdup(p
->pack_name
);
583 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
584 ret
= check_packed_git_idx(idx_name
, p
);
589 static void scan_windows(struct packed_git
*p
,
590 struct packed_git
**lru_p
,
591 struct pack_window
**lru_w
,
592 struct pack_window
**lru_l
)
594 struct pack_window
*w
, *w_l
;
596 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
598 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
608 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
610 struct packed_git
*p
, *lru_p
= NULL
;
611 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
614 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
615 for (p
= packed_git
; p
; p
= p
->next
)
616 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
618 munmap(lru_w
->base
, lru_w
->len
);
619 pack_mapped
-= lru_w
->len
;
621 lru_l
->next
= lru_w
->next
;
623 lru_p
->windows
= lru_w
->next
;
624 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
625 close(lru_p
->pack_fd
);
636 void release_pack_memory(size_t need
, int fd
)
638 size_t cur
= pack_mapped
;
639 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
643 void close_pack_windows(struct packed_git
*p
)
646 struct pack_window
*w
= p
->windows
;
649 die("pack '%s' still has open windows to it",
651 munmap(w
->base
, w
->len
);
652 pack_mapped
-= w
->len
;
654 p
->windows
= w
->next
;
659 void unuse_pack(struct pack_window
**w_cursor
)
661 struct pack_window
*w
= *w_cursor
;
669 * This is used by git-repack in case a newly created pack happens to
670 * contain the same set of objects as an existing one. In that case
671 * the resulting file might be different even if its name would be the
672 * same. It is best to close any reference to the old pack before it is
673 * replaced on disk. Of course no index pointers nor windows for given pack
674 * must subsist at this point. If ever objects from this pack are requested
675 * again, the new version of the pack will be reinitialized through
676 * reprepare_packed_git().
678 void free_pack_by_name(const char *pack_name
)
680 struct packed_git
*p
, **pp
= &packed_git
;
684 if (strcmp(pack_name
, p
->pack_name
) == 0) {
685 clear_delta_base_cache();
686 close_pack_windows(p
);
687 if (p
->pack_fd
!= -1)
690 munmap((void *)p
->index_data
, p
->index_size
);
691 free(p
->bad_object_sha1
);
701 * Do not call this directly as this leaks p->pack_fd on error return;
702 * call open_packed_git() instead.
704 static int open_packed_git_1(struct packed_git
*p
)
707 struct pack_header hdr
;
708 unsigned char sha1
[20];
709 unsigned char *idx_sha1
;
712 if (!p
->index_data
&& open_pack_index(p
))
713 return error("packfile %s index unavailable", p
->pack_name
);
715 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
716 while (p
->pack_fd
< 0 && errno
== EMFILE
&& unuse_one_window(p
, -1))
717 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
718 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
721 /* If we created the struct before we had the pack we lack size. */
723 if (!S_ISREG(st
.st_mode
))
724 return error("packfile %s not a regular file", p
->pack_name
);
725 p
->pack_size
= st
.st_size
;
726 } else if (p
->pack_size
!= st
.st_size
)
727 return error("packfile %s size changed", p
->pack_name
);
729 /* We leave these file descriptors open with sliding mmap;
730 * there is no point keeping them open across exec(), though.
732 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
734 return error("cannot determine file descriptor flags");
735 fd_flag
|= FD_CLOEXEC
;
736 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
737 return error("cannot set FD_CLOEXEC");
739 /* Verify we recognize this pack file format. */
740 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
741 return error("file %s is far too short to be a packfile", p
->pack_name
);
742 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
743 return error("file %s is not a GIT packfile", p
->pack_name
);
744 if (!pack_version_ok(hdr
.hdr_version
))
745 return error("packfile %s is version %"PRIu32
" and not"
746 " supported (try upgrading GIT to a newer version)",
747 p
->pack_name
, ntohl(hdr
.hdr_version
));
749 /* Verify the pack matches its index. */
750 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
751 return error("packfile %s claims to have %"PRIu32
" objects"
752 " while index indicates %"PRIu32
" objects",
753 p
->pack_name
, ntohl(hdr
.hdr_entries
),
755 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
756 return error("end of packfile %s is unavailable", p
->pack_name
);
757 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
758 return error("packfile %s signature is unavailable", p
->pack_name
);
759 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
760 if (hashcmp(sha1
, idx_sha1
))
761 return error("packfile %s does not match index", p
->pack_name
);
765 static int open_packed_git(struct packed_git
*p
)
767 if (!open_packed_git_1(p
))
769 if (p
->pack_fd
!= -1) {
776 static int in_window(struct pack_window
*win
, off_t offset
)
778 /* We must promise at least 20 bytes (one hash) after the
779 * offset is available from this window, otherwise the offset
780 * is not actually in this window and a different window (which
781 * has that one hash excess) must be used. This is to support
782 * the object header and delta base parsing routines below.
784 off_t win_off
= win
->offset
;
785 return win_off
<= offset
786 && (offset
+ 20) <= (win_off
+ win
->len
);
789 unsigned char *use_pack(struct packed_git
*p
,
790 struct pack_window
**w_cursor
,
794 struct pack_window
*win
= *w_cursor
;
796 if (p
->pack_fd
== -1 && open_packed_git(p
))
797 die("packfile %s cannot be accessed", p
->pack_name
);
799 /* Since packfiles end in a hash of their content and it's
800 * pointless to ask for an offset into the middle of that
801 * hash, and the in_window function above wouldn't match
802 * don't allow an offset too close to the end of the file.
804 if (offset
> (p
->pack_size
- 20))
805 die("offset beyond end of packfile (truncated pack?)");
807 if (!win
|| !in_window(win
, offset
)) {
810 for (win
= p
->windows
; win
; win
= win
->next
) {
811 if (in_window(win
, offset
))
815 size_t window_align
= packed_git_window_size
/ 2;
817 win
= xcalloc(1, sizeof(*win
));
818 win
->offset
= (offset
/ window_align
) * window_align
;
819 len
= p
->pack_size
- win
->offset
;
820 if (len
> packed_git_window_size
)
821 len
= packed_git_window_size
;
822 win
->len
= (size_t)len
;
823 pack_mapped
+= win
->len
;
824 while (packed_git_limit
< pack_mapped
825 && unuse_one_window(p
, p
->pack_fd
))
827 win
->base
= xmmap(NULL
, win
->len
,
828 PROT_READ
, MAP_PRIVATE
,
829 p
->pack_fd
, win
->offset
);
830 if (win
->base
== MAP_FAILED
)
831 die("packfile %s cannot be mapped: %s",
836 if (pack_mapped
> peak_pack_mapped
)
837 peak_pack_mapped
= pack_mapped
;
838 if (pack_open_windows
> peak_pack_open_windows
)
839 peak_pack_open_windows
= pack_open_windows
;
840 win
->next
= p
->windows
;
844 if (win
!= *w_cursor
) {
845 win
->last_used
= pack_used_ctr
++;
849 offset
-= win
->offset
;
851 *left
= win
->len
- xsize_t(offset
);
852 return win
->base
+ offset
;
855 static struct packed_git
*alloc_packed_git(int extra
)
857 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
858 memset(p
, 0, sizeof(*p
));
863 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
866 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
869 * Make sure a corresponding .pack file exists and that
870 * the index looks sane.
872 path_len
-= strlen(".idx");
877 memcpy(p
->pack_name
, path
, path_len
);
879 strcpy(p
->pack_name
+ path_len
, ".keep");
880 if (!access(p
->pack_name
, F_OK
))
883 strcpy(p
->pack_name
+ path_len
, ".pack");
884 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
889 /* ok, it looks sane as far as we can check without
890 * actually mapping the pack file.
892 p
->pack_size
= st
.st_size
;
893 p
->pack_local
= local
;
894 p
->mtime
= st
.st_mtime
;
895 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
900 struct packed_git
*parse_pack_index(unsigned char *sha1
)
902 const char *idx_path
= sha1_pack_index_name(sha1
);
903 const char *path
= sha1_pack_name(sha1
);
904 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
906 strcpy(p
->pack_name
, path
);
907 hashcpy(p
->sha1
, sha1
);
908 if (check_packed_git_idx(idx_path
, p
)) {
916 void install_packed_git(struct packed_git
*pack
)
918 pack
->next
= packed_git
;
922 static void prepare_packed_git_one(char *objdir
, int local
)
924 /* Ensure that this buffer is large enough so that we can
925 append "/pack/" without clobbering the stack even if
926 strlen(objdir) were PATH_MAX. */
927 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
932 sprintf(path
, "%s/pack", objdir
);
935 while (!dir
&& errno
== EMFILE
&& unuse_one_window(packed_git
, -1))
939 error("unable to open object pack directory: %s: %s",
940 path
, strerror(errno
));
944 while ((de
= readdir(dir
)) != NULL
) {
945 int namelen
= strlen(de
->d_name
);
946 struct packed_git
*p
;
948 if (!has_extension(de
->d_name
, ".idx"))
951 if (len
+ namelen
+ 1 > sizeof(path
))
954 /* Don't reopen a pack we already have. */
955 strcpy(path
+ len
, de
->d_name
);
956 for (p
= packed_git
; p
; p
= p
->next
) {
957 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
962 /* See if it really is a valid .idx file with corresponding
963 * .pack file that we can map.
965 p
= add_packed_git(path
, len
+ namelen
, local
);
968 install_packed_git(p
);
973 static int sort_pack(const void *a_
, const void *b_
)
975 struct packed_git
*a
= *((struct packed_git
**)a_
);
976 struct packed_git
*b
= *((struct packed_git
**)b_
);
980 * Local packs tend to contain objects specific to our
981 * variant of the project than remote ones. In addition,
982 * remote ones could be on a network mounted filesystem.
983 * Favor local ones for these reasons.
985 st
= a
->pack_local
- b
->pack_local
;
990 * Younger packs tend to contain more recent objects,
991 * and more recent objects tend to get accessed more
994 if (a
->mtime
< b
->mtime
)
996 else if (a
->mtime
== b
->mtime
)
1001 static void rearrange_packed_git(void)
1003 struct packed_git
**ary
, *p
;
1006 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1011 /* prepare an array of packed_git for easier sorting */
1012 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1013 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1016 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1018 /* link them back again */
1019 for (i
= 0; i
< n
- 1; i
++)
1020 ary
[i
]->next
= ary
[i
+ 1];
1021 ary
[n
- 1]->next
= NULL
;
1022 packed_git
= ary
[0];
1027 static int prepare_packed_git_run_once
= 0;
1028 void prepare_packed_git(void)
1030 struct alternate_object_database
*alt
;
1032 if (prepare_packed_git_run_once
)
1034 prepare_packed_git_one(get_object_directory(), 1);
1036 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1038 prepare_packed_git_one(alt
->base
, 0);
1039 alt
->name
[-1] = '/';
1041 rearrange_packed_git();
1042 prepare_packed_git_run_once
= 1;
1045 void reprepare_packed_git(void)
1048 prepare_packed_git_run_once
= 0;
1049 prepare_packed_git();
1052 static void mark_bad_packed_object(struct packed_git
*p
,
1053 const unsigned char *sha1
)
1056 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1057 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1059 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1060 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1061 p
->num_bad_objects
++;
1064 static int has_packed_and_bad(const unsigned char *sha1
)
1066 struct packed_git
*p
;
1069 for (p
= packed_git
; p
; p
= p
->next
)
1070 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1071 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1076 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1078 unsigned char real_sha1
[20];
1079 hash_sha1_file(map
, size
, type
, real_sha1
);
1080 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1083 static int git_open_noatime(const char *name
)
1085 static int sha1_file_open_flag
= O_NOATIME
;
1086 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1088 /* Might the failure be due to O_NOATIME? */
1089 if (fd
< 0 && errno
!= ENOENT
&& sha1_file_open_flag
) {
1090 fd
= open(name
, O_RDONLY
);
1092 sha1_file_open_flag
= 0;
1097 static int open_sha1_file(const unsigned char *sha1
)
1100 char *name
= sha1_file_name(sha1
);
1101 struct alternate_object_database
*alt
;
1103 fd
= git_open_noatime(name
);
1109 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1111 fill_sha1_path(name
, sha1
);
1112 fd
= git_open_noatime(alt
->base
);
1119 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1124 fd
= open_sha1_file(sha1
);
1129 if (!fstat(fd
, &st
)) {
1130 *size
= xsize_t(st
.st_size
);
1131 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1138 static int legacy_loose_object(unsigned char *map
)
1143 * Is it a zlib-compressed buffer? If so, the first byte
1144 * must be 0x78 (15-bit window size, deflated), and the
1145 * first 16-bit word is evenly divisible by 31
1147 word
= (map
[0] << 8) + map
[1];
1148 if (map
[0] == 0x78 && !(word
% 31))
1154 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1155 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1158 unsigned long size
, c
;
1159 unsigned long used
= 0;
1162 *type
= (c
>> 4) & 7;
1166 if (len
<= used
|| bitsizeof(long) <= shift
) {
1167 error("bad object header");
1171 size
+= (c
& 0x7f) << shift
;
1178 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1180 unsigned long size
, used
;
1181 static const char valid_loose_object_type
[8] = {
1183 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1184 0, /* "delta" and others are invalid in a loose object */
1186 enum object_type type
;
1188 /* Get the data stream */
1189 memset(stream
, 0, sizeof(*stream
));
1190 stream
->next_in
= map
;
1191 stream
->avail_in
= mapsize
;
1192 stream
->next_out
= buffer
;
1193 stream
->avail_out
= bufsiz
;
1195 if (legacy_loose_object(map
)) {
1196 git_inflate_init(stream
);
1197 return git_inflate(stream
, 0);
1202 * There used to be a second loose object header format which
1203 * was meant to mimic the in-pack format, allowing for direct
1204 * copy of the object data. This format turned up not to be
1205 * really worth it and we don't write it any longer. But we
1206 * can still read it.
1208 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1209 if (!used
|| !valid_loose_object_type
[type
])
1214 /* Set up the stream for the rest.. */
1215 stream
->next_in
= map
;
1216 stream
->avail_in
= mapsize
;
1217 git_inflate_init(stream
);
1219 /* And generate the fake traditional header */
1220 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1221 typename(type
), size
);
1225 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1227 int bytes
= strlen(buffer
) + 1;
1228 unsigned char *buf
= xmallocz(size
);
1232 n
= stream
->total_out
- bytes
;
1235 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1237 if (bytes
<= size
) {
1239 * The above condition must be (bytes <= size), not
1240 * (bytes < size). In other words, even though we
1241 * expect no more output and set avail_out to zer0,
1242 * the input zlib stream may have bytes that express
1243 * "this concludes the stream", and we *do* want to
1246 * Otherwise we would not be able to test that we
1247 * consumed all the input to reach the expected size;
1248 * we also want to check that zlib tells us that all
1249 * went well with status == Z_STREAM_END at the end.
1251 stream
->next_out
= buf
+ bytes
;
1252 stream
->avail_out
= size
- bytes
;
1253 while (status
== Z_OK
)
1254 status
= git_inflate(stream
, Z_FINISH
);
1256 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1257 git_inflate_end(stream
);
1262 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1263 else if (stream
->avail_in
)
1264 error("garbage at end of loose object '%s'",
1271 * We used to just use "sscanf()", but that's actually way
1272 * too permissive for what we want to check. So do an anal
1273 * object header parse by hand.
1275 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1282 * The type can be at most ten bytes (including the
1283 * terminating '\0' that we add), and is followed by
1292 if (i
>= sizeof(type
))
1298 * The length must follow immediately, and be in canonical
1299 * decimal format (ie "010" is not valid).
1301 size
= *hdr
++ - '0';
1306 unsigned long c
= *hdr
- '0';
1310 size
= size
* 10 + c
;
1316 * The length must be followed by a zero byte
1318 return *hdr
? -1 : type_from_string(type
);
1321 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1327 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1328 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1331 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1334 unsigned long get_size_from_delta(struct packed_git
*p
,
1335 struct pack_window
**w_curs
,
1338 const unsigned char *data
;
1339 unsigned char delta_head
[20], *in
;
1343 memset(&stream
, 0, sizeof(stream
));
1344 stream
.next_out
= delta_head
;
1345 stream
.avail_out
= sizeof(delta_head
);
1347 git_inflate_init(&stream
);
1349 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1350 stream
.next_in
= in
;
1351 st
= git_inflate(&stream
, Z_FINISH
);
1352 curpos
+= stream
.next_in
- in
;
1353 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1354 stream
.total_out
< sizeof(delta_head
));
1355 git_inflate_end(&stream
);
1356 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1357 error("delta data unpack-initial failed");
1361 /* Examine the initial part of the delta to figure out
1366 /* ignore base size */
1367 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1369 /* Read the result size */
1370 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1373 static off_t
get_delta_base(struct packed_git
*p
,
1374 struct pack_window
**w_curs
,
1376 enum object_type type
,
1377 off_t delta_obj_offset
)
1379 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1382 /* use_pack() assured us we have [base_info, base_info + 20)
1383 * as a range that we can look at without walking off the
1384 * end of the mapped window. Its actually the hash size
1385 * that is assured. An OFS_DELTA longer than the hash size
1386 * is stupid, as then a REF_DELTA would be smaller to store.
1388 if (type
== OBJ_OFS_DELTA
) {
1390 unsigned char c
= base_info
[used
++];
1391 base_offset
= c
& 127;
1394 if (!base_offset
|| MSB(base_offset
, 7))
1395 return 0; /* overflow */
1396 c
= base_info
[used
++];
1397 base_offset
= (base_offset
<< 7) + (c
& 127);
1399 base_offset
= delta_obj_offset
- base_offset
;
1400 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1401 return 0; /* out of bound */
1403 } else if (type
== OBJ_REF_DELTA
) {
1404 /* The base entry _must_ be in the same pack */
1405 base_offset
= find_pack_entry_one(base_info
, p
);
1408 die("I am totally screwed");
1412 /* forward declaration for a mutually recursive function */
1413 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1414 unsigned long *sizep
);
1416 static int packed_delta_info(struct packed_git
*p
,
1417 struct pack_window
**w_curs
,
1419 enum object_type type
,
1421 unsigned long *sizep
)
1425 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1428 type
= packed_object_info(p
, base_offset
, NULL
);
1429 if (type
<= OBJ_NONE
) {
1430 struct revindex_entry
*revidx
;
1431 const unsigned char *base_sha1
;
1432 revidx
= find_pack_revindex(p
, base_offset
);
1435 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1436 mark_bad_packed_object(p
, base_sha1
);
1437 type
= sha1_object_info(base_sha1
, NULL
);
1438 if (type
<= OBJ_NONE
)
1442 /* We choose to only get the type of the base object and
1443 * ignore potentially corrupt pack file that expects the delta
1444 * based on a base with a wrong size. This saves tons of
1448 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1456 static int unpack_object_header(struct packed_git
*p
,
1457 struct pack_window
**w_curs
,
1459 unsigned long *sizep
)
1461 unsigned char *base
;
1464 enum object_type type
;
1466 /* use_pack() assures us we have [base, base + 20) available
1467 * as a range that we can look at at. (Its actually the hash
1468 * size that is assured.) With our object header encoding
1469 * the maximum deflated object size is 2^137, which is just
1470 * insane, so we know won't exceed what we have been given.
1472 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1473 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1482 const char *packed_object_info_detail(struct packed_git
*p
,
1484 unsigned long *size
,
1485 unsigned long *store_size
,
1486 unsigned int *delta_chain_length
,
1487 unsigned char *base_sha1
)
1489 struct pack_window
*w_curs
= NULL
;
1491 unsigned long dummy
;
1492 unsigned char *next_sha1
;
1493 enum object_type type
;
1494 struct revindex_entry
*revidx
;
1496 *delta_chain_length
= 0;
1497 curpos
= obj_offset
;
1498 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1500 revidx
= find_pack_revindex(p
, obj_offset
);
1501 *store_size
= revidx
[1].offset
- obj_offset
;
1506 die("pack %s contains unknown object type %d",
1507 p
->pack_name
, type
);
1512 unuse_pack(&w_curs
);
1513 return typename(type
);
1515 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1517 die("pack %s contains bad delta base reference of type %s",
1518 p
->pack_name
, typename(type
));
1519 if (*delta_chain_length
== 0) {
1520 revidx
= find_pack_revindex(p
, obj_offset
);
1521 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1525 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1526 if (*delta_chain_length
== 0)
1527 hashcpy(base_sha1
, next_sha1
);
1528 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1531 (*delta_chain_length
)++;
1532 curpos
= obj_offset
;
1533 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1537 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1538 unsigned long *sizep
)
1540 struct pack_window
*w_curs
= NULL
;
1542 off_t curpos
= obj_offset
;
1543 enum object_type type
;
1545 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1550 type
= packed_delta_info(p
, &w_curs
, curpos
,
1551 type
, obj_offset
, sizep
);
1561 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1562 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1565 unuse_pack(&w_curs
);
1569 static void *unpack_compressed_entry(struct packed_git
*p
,
1570 struct pack_window
**w_curs
,
1576 unsigned char *buffer
, *in
;
1578 buffer
= xmallocz(size
);
1579 memset(&stream
, 0, sizeof(stream
));
1580 stream
.next_out
= buffer
;
1581 stream
.avail_out
= size
+ 1;
1583 git_inflate_init(&stream
);
1585 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1586 stream
.next_in
= in
;
1587 st
= git_inflate(&stream
, Z_FINISH
);
1588 if (!stream
.avail_out
)
1589 break; /* the payload is larger than it should be */
1590 curpos
+= stream
.next_in
- in
;
1591 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1592 git_inflate_end(&stream
);
1593 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1601 #define MAX_DELTA_CACHE (256)
1603 static size_t delta_base_cached
;
1605 static struct delta_base_cache_lru_list
{
1606 struct delta_base_cache_lru_list
*prev
;
1607 struct delta_base_cache_lru_list
*next
;
1608 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1610 static struct delta_base_cache_entry
{
1611 struct delta_base_cache_lru_list lru
;
1613 struct packed_git
*p
;
1616 enum object_type type
;
1617 } delta_base_cache
[MAX_DELTA_CACHE
];
1619 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1623 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1624 hash
+= (hash
>> 8) + (hash
>> 16);
1625 return hash
% MAX_DELTA_CACHE
;
1628 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1629 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1632 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1633 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1636 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1637 return unpack_entry(p
, base_offset
, type
, base_size
);
1641 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1642 ent
->lru
.prev
->next
= ent
->lru
.next
;
1643 delta_base_cached
-= ent
->size
;
1645 ret
= xmemdupz(ent
->data
, ent
->size
);
1648 *base_size
= ent
->size
;
1652 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1657 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1658 ent
->lru
.prev
->next
= ent
->lru
.next
;
1659 delta_base_cached
-= ent
->size
;
1663 void clear_delta_base_cache(void)
1666 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1667 release_delta_base_cache(&delta_base_cache
[p
]);
1670 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1671 void *base
, unsigned long base_size
, enum object_type type
)
1673 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1674 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1675 struct delta_base_cache_lru_list
*lru
;
1677 release_delta_base_cache(ent
);
1678 delta_base_cached
+= base_size
;
1680 for (lru
= delta_base_cache_lru
.next
;
1681 delta_base_cached
> delta_base_cache_limit
1682 && lru
!= &delta_base_cache_lru
;
1684 struct delta_base_cache_entry
*f
= (void *)lru
;
1685 if (f
->type
== OBJ_BLOB
)
1686 release_delta_base_cache(f
);
1688 for (lru
= delta_base_cache_lru
.next
;
1689 delta_base_cached
> delta_base_cache_limit
1690 && lru
!= &delta_base_cache_lru
;
1692 struct delta_base_cache_entry
*f
= (void *)lru
;
1693 release_delta_base_cache(f
);
1697 ent
->base_offset
= base_offset
;
1700 ent
->size
= base_size
;
1701 ent
->lru
.next
= &delta_base_cache_lru
;
1702 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1703 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1704 delta_base_cache_lru
.prev
= &ent
->lru
;
1707 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1708 unsigned long *size
);
1710 static void *unpack_delta_entry(struct packed_git
*p
,
1711 struct pack_window
**w_curs
,
1713 unsigned long delta_size
,
1715 enum object_type
*type
,
1716 unsigned long *sizep
)
1718 void *delta_data
, *result
, *base
;
1719 unsigned long base_size
;
1722 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1724 error("failed to validate delta base reference "
1725 "at offset %"PRIuMAX
" from %s",
1726 (uintmax_t)curpos
, p
->pack_name
);
1730 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1733 * We're probably in deep shit, but let's try to fetch
1734 * the required base anyway from another pack or loose.
1735 * This is costly but should happen only in the presence
1736 * of a corrupted pack, and is better than failing outright.
1738 struct revindex_entry
*revidx
;
1739 const unsigned char *base_sha1
;
1740 revidx
= find_pack_revindex(p
, base_offset
);
1743 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1744 error("failed to read delta base object %s"
1745 " at offset %"PRIuMAX
" from %s",
1746 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1748 mark_bad_packed_object(p
, base_sha1
);
1749 base
= read_object(base_sha1
, type
, &base_size
);
1754 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1756 error("failed to unpack compressed delta "
1757 "at offset %"PRIuMAX
" from %s",
1758 (uintmax_t)curpos
, p
->pack_name
);
1762 result
= patch_delta(base
, base_size
,
1763 delta_data
, delta_size
,
1766 die("failed to apply delta");
1768 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1772 int do_check_packed_object_crc
;
1774 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1775 enum object_type
*type
, unsigned long *sizep
)
1777 struct pack_window
*w_curs
= NULL
;
1778 off_t curpos
= obj_offset
;
1781 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1782 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1783 unsigned long len
= revidx
[1].offset
- obj_offset
;
1784 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1785 const unsigned char *sha1
=
1786 nth_packed_object_sha1(p
, revidx
->nr
);
1787 error("bad packed object CRC for %s",
1789 mark_bad_packed_object(p
, sha1
);
1790 unuse_pack(&w_curs
);
1795 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1799 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1800 obj_offset
, type
, sizep
);
1806 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1810 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1811 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1813 unuse_pack(&w_curs
);
1817 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1820 const unsigned char *index
= p
->index_data
;
1822 if (open_pack_index(p
))
1824 index
= p
->index_data
;
1826 if (n
>= p
->num_objects
)
1829 if (p
->index_version
== 1) {
1830 return index
+ 24 * n
+ 4;
1833 return index
+ 20 * n
;
1837 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1839 const unsigned char *index
= p
->index_data
;
1841 if (p
->index_version
== 1) {
1842 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1845 index
+= 8 + p
->num_objects
* (20 + 4);
1846 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1847 if (!(off
& 0x80000000))
1849 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1850 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1851 ntohl(*((uint32_t *)(index
+ 4)));
1855 off_t
find_pack_entry_one(const unsigned char *sha1
,
1856 struct packed_git
*p
)
1858 const uint32_t *level1_ofs
= p
->index_data
;
1859 const unsigned char *index
= p
->index_data
;
1860 unsigned hi
, lo
, stride
;
1861 static int use_lookup
= -1;
1862 static int debug_lookup
= -1;
1864 if (debug_lookup
< 0)
1865 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1868 if (open_pack_index(p
))
1870 level1_ofs
= p
->index_data
;
1871 index
= p
->index_data
;
1873 if (p
->index_version
> 1) {
1878 hi
= ntohl(level1_ofs
[*sha1
]);
1879 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1880 if (p
->index_version
> 1) {
1888 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1889 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1892 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1894 int pos
= sha1_entry_pos(index
, stride
, 0,
1895 lo
, hi
, p
->num_objects
, sha1
);
1898 return nth_packed_object_offset(p
, pos
);
1902 unsigned mi
= (lo
+ hi
) / 2;
1903 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1906 printf("lo %u hi %u rg %u mi %u\n",
1907 lo
, hi
, hi
- lo
, mi
);
1909 return nth_packed_object_offset(p
, mi
);
1918 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1920 static struct packed_git
*last_found
= (void *)1;
1921 struct packed_git
*p
;
1924 prepare_packed_git();
1927 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1930 if (p
->num_bad_objects
) {
1932 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1933 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1937 offset
= find_pack_entry_one(sha1
, p
);
1940 * We are about to tell the caller where they can
1941 * locate the requested object. We better make
1942 * sure the packfile is still here and can be
1943 * accessed before supplying that answer, as
1944 * it may have been deleted since the index
1947 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1948 error("packfile %s cannot be accessed", p
->pack_name
);
1953 hashcpy(e
->sha1
, sha1
);
1959 if (p
== last_found
)
1963 if (p
== last_found
)
1969 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1970 struct packed_git
*packs
)
1972 struct packed_git
*p
;
1974 for (p
= packs
; p
; p
= p
->next
) {
1975 if (find_pack_entry_one(sha1
, p
))
1982 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1985 unsigned long mapsize
, size
;
1990 map
= map_sha1_file(sha1
, &mapsize
);
1992 return error("unable to find %s", sha1_to_hex(sha1
));
1993 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1994 status
= error("unable to unpack %s header",
1996 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1997 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2000 git_inflate_end(&stream
);
2001 munmap(map
, mapsize
);
2005 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2007 struct pack_entry e
;
2010 if (!find_pack_entry(sha1
, &e
)) {
2011 /* Most likely it's a loose object. */
2012 status
= sha1_loose_object_info(sha1
, sizep
);
2016 /* Not a loose object; someone else may have just packed it. */
2017 reprepare_packed_git();
2018 if (!find_pack_entry(sha1
, &e
))
2022 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
2024 mark_bad_packed_object(e
.p
, sha1
);
2025 status
= sha1_object_info(sha1
, sizep
);
2031 static void *read_packed_sha1(const unsigned char *sha1
,
2032 enum object_type
*type
, unsigned long *size
)
2034 struct pack_entry e
;
2037 if (!find_pack_entry(sha1
, &e
))
2039 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2042 * We're probably in deep shit, but let's try to fetch
2043 * the required object anyway from another pack or loose.
2044 * This should happen only in the presence of a corrupted
2045 * pack, and is better than failing outright.
2047 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2048 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2049 mark_bad_packed_object(e
.p
, sha1
);
2050 data
= read_object(sha1
, type
, size
);
2056 * This is meant to hold a *small* number of objects that you would
2057 * want read_sha1_file() to be able to return, but yet you do not want
2058 * to write them into the object store (e.g. a browse-only
2061 static struct cached_object
{
2062 unsigned char sha1
[20];
2063 enum object_type type
;
2067 static int cached_object_nr
, cached_object_alloc
;
2069 static struct cached_object empty_tree
= {
2070 EMPTY_TREE_SHA1_BIN
,
2076 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
2079 struct cached_object
*co
= cached_objects
;
2081 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
2082 if (!hashcmp(co
->sha1
, sha1
))
2085 if (!hashcmp(sha1
, empty_tree
.sha1
))
2090 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2091 unsigned char *sha1
)
2093 struct cached_object
*co
;
2095 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2096 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2098 if (cached_object_alloc
<= cached_object_nr
) {
2099 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2100 cached_objects
= xrealloc(cached_objects
,
2101 sizeof(*cached_objects
) *
2102 cached_object_alloc
);
2104 co
= &cached_objects
[cached_object_nr
++];
2107 co
->buf
= xmalloc(len
);
2108 memcpy(co
->buf
, buf
, len
);
2109 hashcpy(co
->sha1
, sha1
);
2113 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2114 unsigned long *size
)
2116 unsigned long mapsize
;
2118 struct cached_object
*co
;
2120 co
= find_cached_object(sha1
);
2124 return xmemdupz(co
->buf
, co
->size
);
2127 buf
= read_packed_sha1(sha1
, type
, size
);
2130 map
= map_sha1_file(sha1
, &mapsize
);
2132 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2133 munmap(map
, mapsize
);
2136 reprepare_packed_git();
2137 return read_packed_sha1(sha1
, type
, size
);
2140 void *read_sha1_file_repl(const unsigned char *sha1
,
2141 enum object_type
*type
,
2142 unsigned long *size
,
2143 const unsigned char **replacement
)
2145 const unsigned char *repl
= lookup_replace_object(sha1
);
2146 void *data
= read_object(repl
, type
, size
);
2148 /* die if we replaced an object with one that does not exist */
2149 if (!data
&& repl
!= sha1
)
2150 die("replacement %s not found for %s",
2151 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2153 /* legacy behavior is to die on corrupted objects */
2154 if (!data
&& (has_loose_object(repl
) || has_packed_and_bad(repl
)))
2155 die("object %s is corrupted", sha1_to_hex(repl
));
2158 *replacement
= repl
;
2163 void *read_object_with_reference(const unsigned char *sha1
,
2164 const char *required_type_name
,
2165 unsigned long *size
,
2166 unsigned char *actual_sha1_return
)
2168 enum object_type type
, required_type
;
2170 unsigned long isize
;
2171 unsigned char actual_sha1
[20];
2173 required_type
= type_from_string(required_type_name
);
2174 hashcpy(actual_sha1
, sha1
);
2176 int ref_length
= -1;
2177 const char *ref_type
= NULL
;
2179 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2182 if (type
== required_type
) {
2184 if (actual_sha1_return
)
2185 hashcpy(actual_sha1_return
, actual_sha1
);
2188 /* Handle references */
2189 else if (type
== OBJ_COMMIT
)
2191 else if (type
== OBJ_TAG
)
2192 ref_type
= "object ";
2197 ref_length
= strlen(ref_type
);
2199 if (ref_length
+ 40 > isize
||
2200 memcmp(buffer
, ref_type
, ref_length
) ||
2201 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2206 /* Now we have the ID of the referred-to object in
2207 * actual_sha1. Check again. */
2211 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2212 const char *type
, unsigned char *sha1
,
2213 char *hdr
, int *hdrlen
)
2217 /* Generate the header */
2218 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2222 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2223 git_SHA1_Update(&c
, buf
, len
);
2224 git_SHA1_Final(sha1
, &c
);
2228 * Move the just written object into its final resting place.
2229 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2230 * "moving" is only a part of what it does, when no patch between
2231 * master to pu changes the call sites of this function.
2233 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2237 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2239 else if (link(tmpfile
, filename
))
2243 * Coda hack - coda doesn't like cross-directory links,
2244 * so we fall back to a rename, which will mean that it
2245 * won't be able to check collisions, but that's not a
2248 * The same holds for FAT formatted media.
2250 * When this succeeds, we just return. We have nothing
2253 if (ret
&& ret
!= EEXIST
) {
2255 if (!rename(tmpfile
, filename
))
2259 unlink_or_warn(tmpfile
);
2261 if (ret
!= EEXIST
) {
2262 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2264 /* FIXME!!! Collision check here ? */
2268 if (set_shared_perm(filename
, (S_IFREG
|0444)))
2269 return error("unable to set permission to '%s'", filename
);
2273 static int write_buffer(int fd
, const void *buf
, size_t len
)
2275 if (write_in_full(fd
, buf
, len
) < 0)
2276 return error("file write error (%s)", strerror(errno
));
2280 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2281 unsigned char *sha1
)
2285 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2289 /* Finalize a file on disk, and close it. */
2290 static void close_sha1_file(int fd
)
2292 if (fsync_object_files
)
2293 fsync_or_die(fd
, "sha1 file");
2295 die_errno("error when closing sha1 file");
2298 /* Size of directory component, including the ending '/' */
2299 static inline int directory_size(const char *filename
)
2301 const char *s
= strrchr(filename
, '/');
2304 return s
- filename
+ 1;
2308 * This creates a temporary file in the same directory as the final
2311 * We want to avoid cross-directory filename renames, because those
2312 * can have problems on various filesystems (FAT, NFS, Coda).
2314 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2316 int fd
, dirlen
= directory_size(filename
);
2318 if (dirlen
+ 20 > bufsiz
) {
2319 errno
= ENAMETOOLONG
;
2322 memcpy(buffer
, filename
, dirlen
);
2323 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2324 fd
= mkstemp(buffer
);
2325 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2326 /* Make sure the directory exists */
2327 memcpy(buffer
, filename
, dirlen
);
2328 buffer
[dirlen
-1] = 0;
2329 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2333 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2334 fd
= mkstemp(buffer
);
2339 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2340 void *buf
, unsigned long len
, time_t mtime
)
2344 unsigned char *compressed
;
2347 static char tmpfile
[PATH_MAX
];
2349 filename
= sha1_file_name(sha1
);
2350 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2351 while (fd
< 0 && errno
== EMFILE
&& unuse_one_window(packed_git
, -1))
2352 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2354 if (errno
== EACCES
)
2355 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2357 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2361 memset(&stream
, 0, sizeof(stream
));
2362 deflateInit(&stream
, zlib_compression_level
);
2363 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2364 compressed
= xmalloc(size
);
2367 stream
.next_out
= compressed
;
2368 stream
.avail_out
= size
;
2370 /* First header.. */
2371 stream
.next_in
= (unsigned char *)hdr
;
2372 stream
.avail_in
= hdrlen
;
2373 while (deflate(&stream
, 0) == Z_OK
)
2376 /* Then the data itself.. */
2377 stream
.next_in
= buf
;
2378 stream
.avail_in
= len
;
2379 ret
= deflate(&stream
, Z_FINISH
);
2380 if (ret
!= Z_STREAM_END
)
2381 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2383 ret
= deflateEnd(&stream
);
2385 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2387 size
= stream
.total_out
;
2389 if (write_buffer(fd
, compressed
, size
) < 0)
2390 die("unable to write sha1 file");
2391 close_sha1_file(fd
);
2397 utb
.modtime
= mtime
;
2398 if (utime(tmpfile
, &utb
) < 0)
2399 warning("failed utime() on %s: %s",
2400 tmpfile
, strerror(errno
));
2403 return move_temp_to_file(tmpfile
, filename
);
2406 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2408 unsigned char sha1
[20];
2412 /* Normally if we have it in the pack then we do not bother writing
2413 * it out into .git/objects/??/?{38} file.
2415 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2417 hashcpy(returnsha1
, sha1
);
2418 if (has_sha1_file(sha1
))
2420 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2423 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2427 enum object_type type
;
2432 if (has_loose_object(sha1
))
2434 buf
= read_packed_sha1(sha1
, &type
, &len
);
2436 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2437 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2438 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2444 int has_pack_index(const unsigned char *sha1
)
2447 if (stat(sha1_pack_index_name(sha1
), &st
))
2452 int has_pack_file(const unsigned char *sha1
)
2455 if (stat(sha1_pack_name(sha1
), &st
))
2460 int has_sha1_pack(const unsigned char *sha1
)
2462 struct pack_entry e
;
2463 return find_pack_entry(sha1
, &e
);
2466 int has_sha1_file(const unsigned char *sha1
)
2468 struct pack_entry e
;
2470 if (find_pack_entry(sha1
, &e
))
2472 return has_loose_object(sha1
);
2475 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2476 int write_object
, enum object_type type
, const char *path
)
2478 int ret
, re_allocated
= 0;
2484 * Convert blobs to git internal format
2486 if ((type
== OBJ_BLOB
) && path
) {
2487 struct strbuf nbuf
= STRBUF_INIT
;
2488 if (convert_to_git(path
, buf
, size
, &nbuf
,
2489 write_object
? safe_crlf
: 0)) {
2490 buf
= strbuf_detach(&nbuf
, &size
);
2496 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2498 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2504 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2505 enum object_type type
, const char *path
)
2508 size_t size
= xsize_t(st
->st_size
);
2510 if (!S_ISREG(st
->st_mode
)) {
2511 struct strbuf sbuf
= STRBUF_INIT
;
2512 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2513 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2517 strbuf_release(&sbuf
);
2519 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2520 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2523 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2528 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2531 struct strbuf sb
= STRBUF_INIT
;
2533 switch (st
->st_mode
& S_IFMT
) {
2535 fd
= open(path
, O_RDONLY
);
2537 return error("open(\"%s\"): %s", path
,
2539 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2540 return error("%s: failed to insert into database",
2544 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2545 char *errstr
= strerror(errno
);
2546 return error("readlink(\"%s\"): %s", path
,
2550 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2551 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2552 return error("%s: failed to insert into database",
2554 strbuf_release(&sb
);
2557 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2559 return error("%s: unsupported file type", path
);
2564 int read_pack_header(int fd
, struct pack_header
*header
)
2566 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2567 /* "eof before pack header was fully read" */
2568 return PH_ERROR_EOF
;
2570 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2571 /* "protocol error (pack signature mismatch detected)" */
2572 return PH_ERROR_PACK_SIGNATURE
;
2573 if (!pack_version_ok(header
->hdr_version
))
2574 /* "protocol error (pack version unsupported)" */
2575 return PH_ERROR_PROTOCOL
;