2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s
) { return (unsigned long)s
; }
33 static size_t sz_fmt(size_t s
) { return s
; }
36 const unsigned char null_sha1
[20];
38 const signed char hexval_table
[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
76 for (i
= 0; i
< 20; i
++) {
77 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
86 static inline int offset_1st_component(const char *path
)
88 if (has_dos_drive_prefix(path
))
89 return 2 + (path
[2] == '/');
93 int safe_create_leading_directories(char *path
)
95 char *pos
= path
+ offset_1st_component(path
);
99 pos
= strchr(pos
, '/');
102 while (*++pos
== '/')
107 if (!stat(path
, &st
)) {
109 if (!S_ISDIR(st
.st_mode
)) {
114 else if (mkdir(path
, 0777)) {
118 else if (adjust_shared_perm(path
)) {
127 int safe_create_leading_directories_const(const char *path
)
129 /* path points to cache entries, so xstrdup before messing with it */
130 char *buf
= xstrdup(path
);
131 int result
= safe_create_leading_directories(buf
);
136 char *sha1_to_hex(const unsigned char *sha1
)
139 static char hexbuffer
[4][50];
140 static const char hex
[] = "0123456789abcdef";
141 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
144 for (i
= 0; i
< 20; i
++) {
145 unsigned int val
= *sha1
++;
146 *buf
++ = hex
[val
>> 4];
147 *buf
++ = hex
[val
& 0xf];
154 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
157 for (i
= 0; i
< 20; i
++) {
158 static char hex
[] = "0123456789abcdef";
159 unsigned int val
= sha1
[i
];
160 char *pos
= pathbuf
+ i
*2 + (i
> 0);
161 *pos
++ = hex
[val
>> 4];
162 *pos
= hex
[val
& 0xf];
167 * NOTE! This returns a statically allocated buffer, so you have to be
168 * careful about using it. Do an "xstrdup()" if you need to save the
171 * Also note that this returns the location for creating. Reading
172 * SHA1 file can happen from any alternate directory listed in the
173 * DB_ENVIRONMENT environment variable if it is not found in
174 * the primary object database.
176 char *sha1_file_name(const unsigned char *sha1
)
178 static char *name
, *base
;
181 const char *sha1_file_directory
= get_object_directory();
182 int len
= strlen(sha1_file_directory
);
183 base
= xmalloc(len
+ 60);
184 memcpy(base
, sha1_file_directory
, len
);
185 memset(base
+len
, 0, 60);
188 name
= base
+ len
+ 1;
190 fill_sha1_path(name
, sha1
);
194 static char *sha1_get_pack_name(const unsigned char *sha1
,
195 char **name
, char **base
, const char *which
)
197 static const char hex
[] = "0123456789abcdef";
202 const char *sha1_file_directory
= get_object_directory();
203 int len
= strlen(sha1_file_directory
);
204 *base
= xmalloc(len
+ 60);
205 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
206 sha1_file_directory
, which
);
207 *name
= *base
+ len
+ 11;
212 for (i
= 0; i
< 20; i
++) {
213 unsigned int val
= *sha1
++;
214 *buf
++ = hex
[val
>> 4];
215 *buf
++ = hex
[val
& 0xf];
221 char *sha1_pack_name(const unsigned char *sha1
)
223 static char *name
, *base
;
225 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
228 char *sha1_pack_index_name(const unsigned char *sha1
)
230 static char *name
, *base
;
232 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
235 struct alternate_object_database
*alt_odb_list
;
236 static struct alternate_object_database
**alt_odb_tail
;
238 static void read_info_alternates(const char * alternates
, int depth
);
241 * Prepare alternate object database registry.
243 * The variable alt_odb_list points at the list of struct
244 * alternate_object_database. The elements on this list come from
245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
247 * whose contents is similar to that environment variable but can be
248 * LF separated. Its base points at a statically allocated buffer that
249 * contains "/the/directory/corresponding/to/.git/objects/...", while
250 * its name points just after the slash at the end of ".git/objects/"
251 * in the example above, and has enough space to hold 40-byte hex
252 * SHA1, an extra slash for the first level indirection, and the
255 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
257 const char *objdir
= get_object_directory();
258 struct alternate_object_database
*ent
;
259 struct alternate_object_database
*alt
;
260 /* 43 = 40-byte + 2 '/' + terminating NUL */
262 int entlen
= pfxlen
+ 43;
265 if (!is_absolute_path(entry
) && relative_base
) {
266 /* Relative alt-odb */
268 base_len
= strlen(relative_base
) + 1;
272 ent
= xmalloc(sizeof(*ent
) + entlen
);
274 if (!is_absolute_path(entry
) && relative_base
) {
275 memcpy(ent
->base
, relative_base
, base_len
- 1);
276 ent
->base
[base_len
- 1] = '/';
277 memcpy(ent
->base
+ base_len
, entry
, len
);
280 memcpy(ent
->base
, entry
, pfxlen
);
282 ent
->name
= ent
->base
+ pfxlen
+ 1;
283 ent
->base
[pfxlen
+ 3] = '/';
284 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
286 /* Detect cases where alternate disappeared */
287 if (!is_directory(ent
->base
)) {
288 error("object directory %s does not exist; "
289 "check .git/objects/info/alternates.",
295 /* Prevent the common mistake of listing the same
296 * thing twice, or object directory itself.
298 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
299 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
304 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
309 /* add the alternate entry */
311 alt_odb_tail
= &(ent
->next
);
314 /* recursively add alternates */
315 read_info_alternates(ent
->base
, depth
+ 1);
317 ent
->base
[pfxlen
] = '/';
322 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
323 const char *relative_base
, int depth
)
325 const char *cp
, *last
;
328 error("%s: ignoring alternate object stores, nesting too deep.",
336 if (cp
< ep
&& *cp
== '#') {
337 while (cp
< ep
&& *cp
!= sep
)
342 while (cp
< ep
&& *cp
!= sep
)
345 if (!is_absolute_path(last
) && depth
) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base
, last
);
349 link_alt_odb_entry(last
, cp
- last
,
350 relative_base
, depth
);
353 while (cp
< ep
&& *cp
== sep
)
359 static void read_info_alternates(const char * relative_base
, int depth
)
364 const char alt_file_name
[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
371 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
372 fd
= open(path
, O_RDONLY
);
375 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
379 mapsz
= xsize_t(st
.st_size
);
380 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
383 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
388 void add_to_alternates_file(const char *reference
)
390 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
391 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
392 char *alt
= mkpath("%s/objects\n", reference
);
393 write_or_die(fd
, alt
, strlen(alt
));
394 if (commit_lock_file(lock
))
395 die("could not close alternates file");
397 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
400 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
402 struct alternate_object_database
*ent
;
405 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
410 void prepare_alt_odb(void)
417 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
420 alt_odb_tail
= &alt_odb_list
;
421 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
423 read_info_alternates(get_object_directory(), 0);
426 static int has_loose_object_local(const unsigned char *sha1
)
428 char *name
= sha1_file_name(sha1
);
429 return !access(name
, F_OK
);
432 int has_loose_object_nonlocal(const unsigned char *sha1
)
434 struct alternate_object_database
*alt
;
436 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
437 fill_sha1_path(alt
->name
, sha1
);
438 if (!access(alt
->base
, F_OK
))
444 static int has_loose_object(const unsigned char *sha1
)
446 return has_loose_object_local(sha1
) ||
447 has_loose_object_nonlocal(sha1
);
450 static unsigned int pack_used_ctr
;
451 static unsigned int pack_mmap_calls
;
452 static unsigned int peak_pack_open_windows
;
453 static unsigned int pack_open_windows
;
454 static size_t peak_pack_mapped
;
455 static size_t pack_mapped
;
456 struct packed_git
*packed_git
;
458 void pack_report(void)
461 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
462 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
463 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
464 sz_fmt(getpagesize()),
465 sz_fmt(packed_git_window_size
),
466 sz_fmt(packed_git_limit
));
468 "pack_report: pack_used_ctr = %10u\n"
469 "pack_report: pack_mmap_calls = %10u\n"
470 "pack_report: pack_open_windows = %10u / %10u\n"
471 "pack_report: pack_mapped = "
472 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
475 pack_open_windows
, peak_pack_open_windows
,
476 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
479 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
482 struct pack_idx_header
*hdr
;
484 uint32_t version
, nr
, i
, *index
;
485 int fd
= open(path
, O_RDONLY
);
490 if (fstat(fd
, &st
)) {
494 idx_size
= xsize_t(st
.st_size
);
495 if (idx_size
< 4 * 256 + 20 + 20) {
497 return error("index file %s is too small", path
);
499 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
503 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
504 version
= ntohl(hdr
->idx_version
);
505 if (version
< 2 || version
> 2) {
506 munmap(idx_map
, idx_size
);
507 return error("index file %s is version %"PRIu32
508 " and is not supported by this binary"
509 " (try upgrading GIT to a newer version)",
518 index
+= 2; /* skip index header */
519 for (i
= 0; i
< 256; i
++) {
520 uint32_t n
= ntohl(index
[i
]);
522 munmap(idx_map
, idx_size
);
523 return error("non-monotonic index %s", path
);
531 * - 256 index entries 4 bytes each
532 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
533 * - 20-byte SHA1 of the packfile
534 * - 20-byte SHA1 file checksum
536 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
537 munmap(idx_map
, idx_size
);
538 return error("wrong index v1 file size in %s", path
);
540 } else if (version
== 2) {
543 * - 8 bytes of header
544 * - 256 index entries 4 bytes each
545 * - 20-byte sha1 entry * nr
546 * - 4-byte crc entry * nr
547 * - 4-byte offset entry * nr
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 * And after the 4-byte offset table might be a
551 * variable sized table containing 8-byte entries
552 * for offsets larger than 2^31.
554 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
555 unsigned long max_size
= min_size
;
557 max_size
+= (nr
- 1)*8;
558 if (idx_size
< min_size
|| idx_size
> max_size
) {
559 munmap(idx_map
, idx_size
);
560 return error("wrong index v2 file size in %s", path
);
562 if (idx_size
!= min_size
&&
564 * make sure we can deal with large pack offsets.
565 * 31-bit signed offset won't be enough, neither
566 * 32-bit unsigned one will be.
568 (sizeof(off_t
) <= 4)) {
569 munmap(idx_map
, idx_size
);
570 return error("pack too large for current definition of off_t in %s", path
);
574 p
->index_version
= version
;
575 p
->index_data
= idx_map
;
576 p
->index_size
= idx_size
;
581 int open_pack_index(struct packed_git
*p
)
589 idx_name
= xstrdup(p
->pack_name
);
590 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
591 ret
= check_packed_git_idx(idx_name
, p
);
596 static void scan_windows(struct packed_git
*p
,
597 struct packed_git
**lru_p
,
598 struct pack_window
**lru_w
,
599 struct pack_window
**lru_l
)
601 struct pack_window
*w
, *w_l
;
603 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
605 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
615 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
617 struct packed_git
*p
, *lru_p
= NULL
;
618 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
621 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
622 for (p
= packed_git
; p
; p
= p
->next
)
623 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
625 munmap(lru_w
->base
, lru_w
->len
);
626 pack_mapped
-= lru_w
->len
;
628 lru_l
->next
= lru_w
->next
;
630 lru_p
->windows
= lru_w
->next
;
631 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
632 close(lru_p
->pack_fd
);
643 void release_pack_memory(size_t need
, int fd
)
645 size_t cur
= pack_mapped
;
646 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
650 void close_pack_windows(struct packed_git
*p
)
653 struct pack_window
*w
= p
->windows
;
656 die("pack '%s' still has open windows to it",
658 munmap(w
->base
, w
->len
);
659 pack_mapped
-= w
->len
;
661 p
->windows
= w
->next
;
666 void unuse_pack(struct pack_window
**w_cursor
)
668 struct pack_window
*w
= *w_cursor
;
676 * This is used by git-repack in case a newly created pack happens to
677 * contain the same set of objects as an existing one. In that case
678 * the resulting file might be different even if its name would be the
679 * same. It is best to close any reference to the old pack before it is
680 * replaced on disk. Of course no index pointers nor windows for given pack
681 * must subsist at this point. If ever objects from this pack are requested
682 * again, the new version of the pack will be reinitialized through
683 * reprepare_packed_git().
685 void free_pack_by_name(const char *pack_name
)
687 struct packed_git
*p
, **pp
= &packed_git
;
691 if (strcmp(pack_name
, p
->pack_name
) == 0) {
692 clear_delta_base_cache();
693 close_pack_windows(p
);
694 if (p
->pack_fd
!= -1)
697 munmap((void *)p
->index_data
, p
->index_size
);
698 free(p
->bad_object_sha1
);
708 * Do not call this directly as this leaks p->pack_fd on error return;
709 * call open_packed_git() instead.
711 static int open_packed_git_1(struct packed_git
*p
)
714 struct pack_header hdr
;
715 unsigned char sha1
[20];
716 unsigned char *idx_sha1
;
719 if (!p
->index_data
&& open_pack_index(p
))
720 return error("packfile %s index unavailable", p
->pack_name
);
722 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
723 while (p
->pack_fd
< 0 && errno
== EMFILE
&& unuse_one_window(p
, -1))
724 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
725 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
728 /* If we created the struct before we had the pack we lack size. */
730 if (!S_ISREG(st
.st_mode
))
731 return error("packfile %s not a regular file", p
->pack_name
);
732 p
->pack_size
= st
.st_size
;
733 } else if (p
->pack_size
!= st
.st_size
)
734 return error("packfile %s size changed", p
->pack_name
);
736 /* We leave these file descriptors open with sliding mmap;
737 * there is no point keeping them open across exec(), though.
739 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
741 return error("cannot determine file descriptor flags");
742 fd_flag
|= FD_CLOEXEC
;
743 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
744 return error("cannot set FD_CLOEXEC");
746 /* Verify we recognize this pack file format. */
747 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
748 return error("file %s is far too short to be a packfile", p
->pack_name
);
749 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
750 return error("file %s is not a GIT packfile", p
->pack_name
);
751 if (!pack_version_ok(hdr
.hdr_version
))
752 return error("packfile %s is version %"PRIu32
" and not"
753 " supported (try upgrading GIT to a newer version)",
754 p
->pack_name
, ntohl(hdr
.hdr_version
));
756 /* Verify the pack matches its index. */
757 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
758 return error("packfile %s claims to have %"PRIu32
" objects"
759 " while index indicates %"PRIu32
" objects",
760 p
->pack_name
, ntohl(hdr
.hdr_entries
),
762 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
763 return error("end of packfile %s is unavailable", p
->pack_name
);
764 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
765 return error("packfile %s signature is unavailable", p
->pack_name
);
766 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
767 if (hashcmp(sha1
, idx_sha1
))
768 return error("packfile %s does not match index", p
->pack_name
);
772 static int open_packed_git(struct packed_git
*p
)
774 if (!open_packed_git_1(p
))
776 if (p
->pack_fd
!= -1) {
783 static int in_window(struct pack_window
*win
, off_t offset
)
785 /* We must promise at least 20 bytes (one hash) after the
786 * offset is available from this window, otherwise the offset
787 * is not actually in this window and a different window (which
788 * has that one hash excess) must be used. This is to support
789 * the object header and delta base parsing routines below.
791 off_t win_off
= win
->offset
;
792 return win_off
<= offset
793 && (offset
+ 20) <= (win_off
+ win
->len
);
796 unsigned char *use_pack(struct packed_git
*p
,
797 struct pack_window
**w_cursor
,
801 struct pack_window
*win
= *w_cursor
;
803 if (p
->pack_fd
== -1 && open_packed_git(p
))
804 die("packfile %s cannot be accessed", p
->pack_name
);
806 /* Since packfiles end in a hash of their content and it's
807 * pointless to ask for an offset into the middle of that
808 * hash, and the in_window function above wouldn't match
809 * don't allow an offset too close to the end of the file.
811 if (offset
> (p
->pack_size
- 20))
812 die("offset beyond end of packfile (truncated pack?)");
814 if (!win
|| !in_window(win
, offset
)) {
817 for (win
= p
->windows
; win
; win
= win
->next
) {
818 if (in_window(win
, offset
))
822 size_t window_align
= packed_git_window_size
/ 2;
824 win
= xcalloc(1, sizeof(*win
));
825 win
->offset
= (offset
/ window_align
) * window_align
;
826 len
= p
->pack_size
- win
->offset
;
827 if (len
> packed_git_window_size
)
828 len
= packed_git_window_size
;
829 win
->len
= (size_t)len
;
830 pack_mapped
+= win
->len
;
831 while (packed_git_limit
< pack_mapped
832 && unuse_one_window(p
, p
->pack_fd
))
834 win
->base
= xmmap(NULL
, win
->len
,
835 PROT_READ
, MAP_PRIVATE
,
836 p
->pack_fd
, win
->offset
);
837 if (win
->base
== MAP_FAILED
)
838 die("packfile %s cannot be mapped: %s",
843 if (pack_mapped
> peak_pack_mapped
)
844 peak_pack_mapped
= pack_mapped
;
845 if (pack_open_windows
> peak_pack_open_windows
)
846 peak_pack_open_windows
= pack_open_windows
;
847 win
->next
= p
->windows
;
851 if (win
!= *w_cursor
) {
852 win
->last_used
= pack_used_ctr
++;
856 offset
-= win
->offset
;
858 *left
= win
->len
- xsize_t(offset
);
859 return win
->base
+ offset
;
862 static struct packed_git
*alloc_packed_git(int extra
)
864 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
865 memset(p
, 0, sizeof(*p
));
870 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
873 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
876 * Make sure a corresponding .pack file exists and that
877 * the index looks sane.
879 path_len
-= strlen(".idx");
884 memcpy(p
->pack_name
, path
, path_len
);
886 strcpy(p
->pack_name
+ path_len
, ".keep");
887 if (!access(p
->pack_name
, F_OK
))
890 strcpy(p
->pack_name
+ path_len
, ".pack");
891 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
896 /* ok, it looks sane as far as we can check without
897 * actually mapping the pack file.
899 p
->pack_size
= st
.st_size
;
900 p
->pack_local
= local
;
901 p
->mtime
= st
.st_mtime
;
902 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
907 struct packed_git
*parse_pack_index(unsigned char *sha1
)
909 const char *idx_path
= sha1_pack_index_name(sha1
);
910 const char *path
= sha1_pack_name(sha1
);
911 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
913 strcpy(p
->pack_name
, path
);
914 hashcpy(p
->sha1
, sha1
);
915 if (check_packed_git_idx(idx_path
, p
)) {
923 void install_packed_git(struct packed_git
*pack
)
925 pack
->next
= packed_git
;
929 static void prepare_packed_git_one(char *objdir
, int local
)
931 /* Ensure that this buffer is large enough so that we can
932 append "/pack/" without clobbering the stack even if
933 strlen(objdir) were PATH_MAX. */
934 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
939 sprintf(path
, "%s/pack", objdir
);
942 while (!dir
&& errno
== EMFILE
&& unuse_one_window(packed_git
, -1))
946 error("unable to open object pack directory: %s: %s",
947 path
, strerror(errno
));
951 while ((de
= readdir(dir
)) != NULL
) {
952 int namelen
= strlen(de
->d_name
);
953 struct packed_git
*p
;
955 if (!has_extension(de
->d_name
, ".idx"))
958 if (len
+ namelen
+ 1 > sizeof(path
))
961 /* Don't reopen a pack we already have. */
962 strcpy(path
+ len
, de
->d_name
);
963 for (p
= packed_git
; p
; p
= p
->next
) {
964 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
969 /* See if it really is a valid .idx file with corresponding
970 * .pack file that we can map.
972 p
= add_packed_git(path
, len
+ namelen
, local
);
975 install_packed_git(p
);
980 static int sort_pack(const void *a_
, const void *b_
)
982 struct packed_git
*a
= *((struct packed_git
**)a_
);
983 struct packed_git
*b
= *((struct packed_git
**)b_
);
987 * Local packs tend to contain objects specific to our
988 * variant of the project than remote ones. In addition,
989 * remote ones could be on a network mounted filesystem.
990 * Favor local ones for these reasons.
992 st
= a
->pack_local
- b
->pack_local
;
997 * Younger packs tend to contain more recent objects,
998 * and more recent objects tend to get accessed more
1001 if (a
->mtime
< b
->mtime
)
1003 else if (a
->mtime
== b
->mtime
)
1008 static void rearrange_packed_git(void)
1010 struct packed_git
**ary
, *p
;
1013 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1018 /* prepare an array of packed_git for easier sorting */
1019 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1020 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1023 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1025 /* link them back again */
1026 for (i
= 0; i
< n
- 1; i
++)
1027 ary
[i
]->next
= ary
[i
+ 1];
1028 ary
[n
- 1]->next
= NULL
;
1029 packed_git
= ary
[0];
1034 static int prepare_packed_git_run_once
= 0;
1035 void prepare_packed_git(void)
1037 struct alternate_object_database
*alt
;
1039 if (prepare_packed_git_run_once
)
1041 prepare_packed_git_one(get_object_directory(), 1);
1043 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1045 prepare_packed_git_one(alt
->base
, 0);
1046 alt
->name
[-1] = '/';
1048 rearrange_packed_git();
1049 prepare_packed_git_run_once
= 1;
1052 void reprepare_packed_git(void)
1055 prepare_packed_git_run_once
= 0;
1056 prepare_packed_git();
1059 static void mark_bad_packed_object(struct packed_git
*p
,
1060 const unsigned char *sha1
)
1063 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1064 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1066 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1067 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1068 p
->num_bad_objects
++;
1071 static int has_packed_and_bad(const unsigned char *sha1
)
1073 struct packed_git
*p
;
1076 for (p
= packed_git
; p
; p
= p
->next
)
1077 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1078 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1083 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1085 unsigned char real_sha1
[20];
1086 hash_sha1_file(map
, size
, type
, real_sha1
);
1087 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1090 static int git_open_noatime(const char *name
)
1092 static int sha1_file_open_flag
= O_NOATIME
;
1093 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1095 /* Might the failure be due to O_NOATIME? */
1096 if (fd
< 0 && errno
!= ENOENT
&& sha1_file_open_flag
) {
1097 fd
= open(name
, O_RDONLY
);
1099 sha1_file_open_flag
= 0;
1104 static int open_sha1_file(const unsigned char *sha1
)
1107 char *name
= sha1_file_name(sha1
);
1108 struct alternate_object_database
*alt
;
1110 fd
= git_open_noatime(name
);
1116 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1118 fill_sha1_path(name
, sha1
);
1119 fd
= git_open_noatime(alt
->base
);
1126 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1131 fd
= open_sha1_file(sha1
);
1136 if (!fstat(fd
, &st
)) {
1137 *size
= xsize_t(st
.st_size
);
1138 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1145 static int legacy_loose_object(unsigned char *map
)
1150 * Is it a zlib-compressed buffer? If so, the first byte
1151 * must be 0x78 (15-bit window size, deflated), and the
1152 * first 16-bit word is evenly divisible by 31
1154 word
= (map
[0] << 8) + map
[1];
1155 if (map
[0] == 0x78 && !(word
% 31))
1161 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1162 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1167 unsigned long used
= 0;
1170 *type
= (c
>> 4) & 7;
1174 if (len
<= used
|| sizeof(long) * 8 <= shift
) {
1175 error("bad object header");
1179 size
+= (c
& 0x7f) << shift
;
1186 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1188 unsigned long size
, used
;
1189 static const char valid_loose_object_type
[8] = {
1191 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1192 0, /* "delta" and others are invalid in a loose object */
1194 enum object_type type
;
1196 /* Get the data stream */
1197 memset(stream
, 0, sizeof(*stream
));
1198 stream
->next_in
= map
;
1199 stream
->avail_in
= mapsize
;
1200 stream
->next_out
= buffer
;
1201 stream
->avail_out
= bufsiz
;
1203 if (legacy_loose_object(map
)) {
1204 git_inflate_init(stream
);
1205 return git_inflate(stream
, 0);
1210 * There used to be a second loose object header format which
1211 * was meant to mimic the in-pack format, allowing for direct
1212 * copy of the object data. This format turned up not to be
1213 * really worth it and we don't write it any longer. But we
1214 * can still read it.
1216 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1217 if (!used
|| !valid_loose_object_type
[type
])
1222 /* Set up the stream for the rest.. */
1223 stream
->next_in
= map
;
1224 stream
->avail_in
= mapsize
;
1225 git_inflate_init(stream
);
1227 /* And generate the fake traditional header */
1228 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1229 typename(type
), size
);
1233 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1235 int bytes
= strlen(buffer
) + 1;
1236 unsigned char *buf
= xmalloc(1+size
);
1240 n
= stream
->total_out
- bytes
;
1243 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1245 if (bytes
<= size
) {
1247 * The above condition must be (bytes <= size), not
1248 * (bytes < size). In other words, even though we
1249 * expect no more output and set avail_out to zer0,
1250 * the input zlib stream may have bytes that express
1251 * "this concludes the stream", and we *do* want to
1254 * Otherwise we would not be able to test that we
1255 * consumed all the input to reach the expected size;
1256 * we also want to check that zlib tells us that all
1257 * went well with status == Z_STREAM_END at the end.
1259 stream
->next_out
= buf
+ bytes
;
1260 stream
->avail_out
= size
- bytes
;
1261 while (status
== Z_OK
)
1262 status
= git_inflate(stream
, Z_FINISH
);
1265 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1266 git_inflate_end(stream
);
1271 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1272 else if (stream
->avail_in
)
1273 error("garbage at end of loose object '%s'",
1280 * We used to just use "sscanf()", but that's actually way
1281 * too permissive for what we want to check. So do an anal
1282 * object header parse by hand.
1284 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1291 * The type can be at most ten bytes (including the
1292 * terminating '\0' that we add), and is followed by
1301 if (i
>= sizeof(type
))
1307 * The length must follow immediately, and be in canonical
1308 * decimal format (ie "010" is not valid).
1310 size
= *hdr
++ - '0';
1315 unsigned long c
= *hdr
- '0';
1319 size
= size
* 10 + c
;
1325 * The length must be followed by a zero byte
1327 return *hdr
? -1 : type_from_string(type
);
1330 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1336 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1337 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1340 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1343 unsigned long get_size_from_delta(struct packed_git
*p
,
1344 struct pack_window
**w_curs
,
1347 const unsigned char *data
;
1348 unsigned char delta_head
[20], *in
;
1352 memset(&stream
, 0, sizeof(stream
));
1353 stream
.next_out
= delta_head
;
1354 stream
.avail_out
= sizeof(delta_head
);
1356 git_inflate_init(&stream
);
1358 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1359 stream
.next_in
= in
;
1360 st
= git_inflate(&stream
, Z_FINISH
);
1361 curpos
+= stream
.next_in
- in
;
1362 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1363 stream
.total_out
< sizeof(delta_head
));
1364 git_inflate_end(&stream
);
1365 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1366 error("delta data unpack-initial failed");
1370 /* Examine the initial part of the delta to figure out
1375 /* ignore base size */
1376 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1378 /* Read the result size */
1379 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1382 static off_t
get_delta_base(struct packed_git
*p
,
1383 struct pack_window
**w_curs
,
1385 enum object_type type
,
1386 off_t delta_obj_offset
)
1388 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1391 /* use_pack() assured us we have [base_info, base_info + 20)
1392 * as a range that we can look at without walking off the
1393 * end of the mapped window. Its actually the hash size
1394 * that is assured. An OFS_DELTA longer than the hash size
1395 * is stupid, as then a REF_DELTA would be smaller to store.
1397 if (type
== OBJ_OFS_DELTA
) {
1399 unsigned char c
= base_info
[used
++];
1400 base_offset
= c
& 127;
1403 if (!base_offset
|| MSB(base_offset
, 7))
1404 return 0; /* overflow */
1405 c
= base_info
[used
++];
1406 base_offset
= (base_offset
<< 7) + (c
& 127);
1408 base_offset
= delta_obj_offset
- base_offset
;
1409 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1410 return 0; /* out of bound */
1412 } else if (type
== OBJ_REF_DELTA
) {
1413 /* The base entry _must_ be in the same pack */
1414 base_offset
= find_pack_entry_one(base_info
, p
);
1417 die("I am totally screwed");
1421 /* forward declaration for a mutually recursive function */
1422 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1423 unsigned long *sizep
);
1425 static int packed_delta_info(struct packed_git
*p
,
1426 struct pack_window
**w_curs
,
1428 enum object_type type
,
1430 unsigned long *sizep
)
1434 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1437 type
= packed_object_info(p
, base_offset
, NULL
);
1438 if (type
<= OBJ_NONE
) {
1439 struct revindex_entry
*revidx
;
1440 const unsigned char *base_sha1
;
1441 revidx
= find_pack_revindex(p
, base_offset
);
1444 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1445 mark_bad_packed_object(p
, base_sha1
);
1446 type
= sha1_object_info(base_sha1
, NULL
);
1447 if (type
<= OBJ_NONE
)
1451 /* We choose to only get the type of the base object and
1452 * ignore potentially corrupt pack file that expects the delta
1453 * based on a base with a wrong size. This saves tons of
1457 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1465 static int unpack_object_header(struct packed_git
*p
,
1466 struct pack_window
**w_curs
,
1468 unsigned long *sizep
)
1470 unsigned char *base
;
1473 enum object_type type
;
1475 /* use_pack() assures us we have [base, base + 20) available
1476 * as a range that we can look at at. (Its actually the hash
1477 * size that is assured.) With our object header encoding
1478 * the maximum deflated object size is 2^137, which is just
1479 * insane, so we know won't exceed what we have been given.
1481 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1482 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1491 const char *packed_object_info_detail(struct packed_git
*p
,
1493 unsigned long *size
,
1494 unsigned long *store_size
,
1495 unsigned int *delta_chain_length
,
1496 unsigned char *base_sha1
)
1498 struct pack_window
*w_curs
= NULL
;
1500 unsigned long dummy
;
1501 unsigned char *next_sha1
;
1502 enum object_type type
;
1503 struct revindex_entry
*revidx
;
1505 *delta_chain_length
= 0;
1506 curpos
= obj_offset
;
1507 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1509 revidx
= find_pack_revindex(p
, obj_offset
);
1510 *store_size
= revidx
[1].offset
- obj_offset
;
1515 die("pack %s contains unknown object type %d",
1516 p
->pack_name
, type
);
1521 unuse_pack(&w_curs
);
1522 return typename(type
);
1524 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1526 die("pack %s contains bad delta base reference of type %s",
1527 p
->pack_name
, typename(type
));
1528 if (*delta_chain_length
== 0) {
1529 revidx
= find_pack_revindex(p
, obj_offset
);
1530 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1534 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1535 if (*delta_chain_length
== 0)
1536 hashcpy(base_sha1
, next_sha1
);
1537 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1540 (*delta_chain_length
)++;
1541 curpos
= obj_offset
;
1542 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1546 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1547 unsigned long *sizep
)
1549 struct pack_window
*w_curs
= NULL
;
1551 off_t curpos
= obj_offset
;
1552 enum object_type type
;
1554 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1559 type
= packed_delta_info(p
, &w_curs
, curpos
,
1560 type
, obj_offset
, sizep
);
1570 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1571 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1574 unuse_pack(&w_curs
);
1578 static void *unpack_compressed_entry(struct packed_git
*p
,
1579 struct pack_window
**w_curs
,
1585 unsigned char *buffer
, *in
;
1587 buffer
= xmalloc(size
+ 1);
1589 memset(&stream
, 0, sizeof(stream
));
1590 stream
.next_out
= buffer
;
1591 stream
.avail_out
= size
;
1593 git_inflate_init(&stream
);
1595 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1596 stream
.next_in
= in
;
1597 st
= git_inflate(&stream
, Z_FINISH
);
1598 curpos
+= stream
.next_in
- in
;
1599 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1600 git_inflate_end(&stream
);
1601 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1609 #define MAX_DELTA_CACHE (256)
1611 static size_t delta_base_cached
;
1613 static struct delta_base_cache_lru_list
{
1614 struct delta_base_cache_lru_list
*prev
;
1615 struct delta_base_cache_lru_list
*next
;
1616 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1618 static struct delta_base_cache_entry
{
1619 struct delta_base_cache_lru_list lru
;
1621 struct packed_git
*p
;
1624 enum object_type type
;
1625 } delta_base_cache
[MAX_DELTA_CACHE
];
1627 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1631 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1632 hash
+= (hash
>> 8) + (hash
>> 16);
1633 return hash
% MAX_DELTA_CACHE
;
1636 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1637 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1640 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1641 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1644 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1645 return unpack_entry(p
, base_offset
, type
, base_size
);
1649 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1650 ent
->lru
.prev
->next
= ent
->lru
.next
;
1651 delta_base_cached
-= ent
->size
;
1653 ret
= xmemdupz(ent
->data
, ent
->size
);
1656 *base_size
= ent
->size
;
1660 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1665 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1666 ent
->lru
.prev
->next
= ent
->lru
.next
;
1667 delta_base_cached
-= ent
->size
;
1671 void clear_delta_base_cache(void)
1674 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1675 release_delta_base_cache(&delta_base_cache
[p
]);
1678 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1679 void *base
, unsigned long base_size
, enum object_type type
)
1681 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1682 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1683 struct delta_base_cache_lru_list
*lru
;
1685 release_delta_base_cache(ent
);
1686 delta_base_cached
+= base_size
;
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 if (f
->type
== OBJ_BLOB
)
1694 release_delta_base_cache(f
);
1696 for (lru
= delta_base_cache_lru
.next
;
1697 delta_base_cached
> delta_base_cache_limit
1698 && lru
!= &delta_base_cache_lru
;
1700 struct delta_base_cache_entry
*f
= (void *)lru
;
1701 release_delta_base_cache(f
);
1705 ent
->base_offset
= base_offset
;
1708 ent
->size
= base_size
;
1709 ent
->lru
.next
= &delta_base_cache_lru
;
1710 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1711 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1712 delta_base_cache_lru
.prev
= &ent
->lru
;
1715 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1716 unsigned long *size
);
1718 static void *unpack_delta_entry(struct packed_git
*p
,
1719 struct pack_window
**w_curs
,
1721 unsigned long delta_size
,
1723 enum object_type
*type
,
1724 unsigned long *sizep
)
1726 void *delta_data
, *result
, *base
;
1727 unsigned long base_size
;
1730 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1732 error("failed to validate delta base reference "
1733 "at offset %"PRIuMAX
" from %s",
1734 (uintmax_t)curpos
, p
->pack_name
);
1738 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1741 * We're probably in deep shit, but let's try to fetch
1742 * the required base anyway from another pack or loose.
1743 * This is costly but should happen only in the presence
1744 * of a corrupted pack, and is better than failing outright.
1746 struct revindex_entry
*revidx
;
1747 const unsigned char *base_sha1
;
1748 revidx
= find_pack_revindex(p
, base_offset
);
1751 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1752 error("failed to read delta base object %s"
1753 " at offset %"PRIuMAX
" from %s",
1754 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1756 mark_bad_packed_object(p
, base_sha1
);
1757 base
= read_object(base_sha1
, type
, &base_size
);
1762 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1764 error("failed to unpack compressed delta "
1765 "at offset %"PRIuMAX
" from %s",
1766 (uintmax_t)curpos
, p
->pack_name
);
1770 result
= patch_delta(base
, base_size
,
1771 delta_data
, delta_size
,
1774 die("failed to apply delta");
1776 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1780 int do_check_packed_object_crc
;
1782 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1783 enum object_type
*type
, unsigned long *sizep
)
1785 struct pack_window
*w_curs
= NULL
;
1786 off_t curpos
= obj_offset
;
1789 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1790 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1791 unsigned long len
= revidx
[1].offset
- obj_offset
;
1792 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1793 const unsigned char *sha1
=
1794 nth_packed_object_sha1(p
, revidx
->nr
);
1795 error("bad packed object CRC for %s",
1797 mark_bad_packed_object(p
, sha1
);
1798 unuse_pack(&w_curs
);
1803 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1807 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1808 obj_offset
, type
, sizep
);
1814 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1818 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1819 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1821 unuse_pack(&w_curs
);
1825 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1828 const unsigned char *index
= p
->index_data
;
1830 if (open_pack_index(p
))
1832 index
= p
->index_data
;
1834 if (n
>= p
->num_objects
)
1837 if (p
->index_version
== 1) {
1838 return index
+ 24 * n
+ 4;
1841 return index
+ 20 * n
;
1845 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1847 const unsigned char *index
= p
->index_data
;
1849 if (p
->index_version
== 1) {
1850 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1853 index
+= 8 + p
->num_objects
* (20 + 4);
1854 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1855 if (!(off
& 0x80000000))
1857 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1858 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1859 ntohl(*((uint32_t *)(index
+ 4)));
1863 off_t
find_pack_entry_one(const unsigned char *sha1
,
1864 struct packed_git
*p
)
1866 const uint32_t *level1_ofs
= p
->index_data
;
1867 const unsigned char *index
= p
->index_data
;
1868 unsigned hi
, lo
, stride
;
1869 static int use_lookup
= -1;
1870 static int debug_lookup
= -1;
1872 if (debug_lookup
< 0)
1873 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1876 if (open_pack_index(p
))
1878 level1_ofs
= p
->index_data
;
1879 index
= p
->index_data
;
1881 if (p
->index_version
> 1) {
1886 hi
= ntohl(level1_ofs
[*sha1
]);
1887 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1888 if (p
->index_version
> 1) {
1896 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1897 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1900 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1902 int pos
= sha1_entry_pos(index
, stride
, 0,
1903 lo
, hi
, p
->num_objects
, sha1
);
1906 return nth_packed_object_offset(p
, pos
);
1910 unsigned mi
= (lo
+ hi
) / 2;
1911 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1914 printf("lo %u hi %u rg %u mi %u\n",
1915 lo
, hi
, hi
- lo
, mi
);
1917 return nth_packed_object_offset(p
, mi
);
1926 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1928 static struct packed_git
*last_found
= (void *)1;
1929 struct packed_git
*p
;
1932 prepare_packed_git();
1935 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1938 if (p
->num_bad_objects
) {
1940 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1941 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1945 offset
= find_pack_entry_one(sha1
, p
);
1948 * We are about to tell the caller where they can
1949 * locate the requested object. We better make
1950 * sure the packfile is still here and can be
1951 * accessed before supplying that answer, as
1952 * it may have been deleted since the index
1955 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1956 error("packfile %s cannot be accessed", p
->pack_name
);
1961 hashcpy(e
->sha1
, sha1
);
1967 if (p
== last_found
)
1971 if (p
== last_found
)
1977 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1978 struct packed_git
*packs
)
1980 struct packed_git
*p
;
1982 for (p
= packs
; p
; p
= p
->next
) {
1983 if (find_pack_entry_one(sha1
, p
))
1990 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1993 unsigned long mapsize
, size
;
1998 map
= map_sha1_file(sha1
, &mapsize
);
2000 return error("unable to find %s", sha1_to_hex(sha1
));
2001 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2002 status
= error("unable to unpack %s header",
2004 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2005 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2008 git_inflate_end(&stream
);
2009 munmap(map
, mapsize
);
2013 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2015 struct pack_entry e
;
2018 if (!find_pack_entry(sha1
, &e
)) {
2019 /* Most likely it's a loose object. */
2020 status
= sha1_loose_object_info(sha1
, sizep
);
2024 /* Not a loose object; someone else may have just packed it. */
2025 reprepare_packed_git();
2026 if (!find_pack_entry(sha1
, &e
))
2030 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
2032 mark_bad_packed_object(e
.p
, sha1
);
2033 status
= sha1_object_info(sha1
, sizep
);
2039 static void *read_packed_sha1(const unsigned char *sha1
,
2040 enum object_type
*type
, unsigned long *size
)
2042 struct pack_entry e
;
2045 if (!find_pack_entry(sha1
, &e
))
2047 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2050 * We're probably in deep shit, but let's try to fetch
2051 * the required object anyway from another pack or loose.
2052 * This should happen only in the presence of a corrupted
2053 * pack, and is better than failing outright.
2055 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2056 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2057 mark_bad_packed_object(e
.p
, sha1
);
2058 data
= read_object(sha1
, type
, size
);
2064 * This is meant to hold a *small* number of objects that you would
2065 * want read_sha1_file() to be able to return, but yet you do not want
2066 * to write them into the object store (e.g. a browse-only
2069 static struct cached_object
{
2070 unsigned char sha1
[20];
2071 enum object_type type
;
2075 static int cached_object_nr
, cached_object_alloc
;
2077 static struct cached_object empty_tree
= {
2078 EMPTY_TREE_SHA1_BIN
,
2084 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
2087 struct cached_object
*co
= cached_objects
;
2089 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
2090 if (!hashcmp(co
->sha1
, sha1
))
2093 if (!hashcmp(sha1
, empty_tree
.sha1
))
2098 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2099 unsigned char *sha1
)
2101 struct cached_object
*co
;
2103 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2104 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2106 if (cached_object_alloc
<= cached_object_nr
) {
2107 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2108 cached_objects
= xrealloc(cached_objects
,
2109 sizeof(*cached_objects
) *
2110 cached_object_alloc
);
2112 co
= &cached_objects
[cached_object_nr
++];
2115 co
->buf
= xmalloc(len
);
2116 memcpy(co
->buf
, buf
, len
);
2117 hashcpy(co
->sha1
, sha1
);
2121 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2122 unsigned long *size
)
2124 unsigned long mapsize
;
2126 struct cached_object
*co
;
2128 co
= find_cached_object(sha1
);
2132 return xmemdupz(co
->buf
, co
->size
);
2135 buf
= read_packed_sha1(sha1
, type
, size
);
2138 map
= map_sha1_file(sha1
, &mapsize
);
2140 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2141 munmap(map
, mapsize
);
2144 reprepare_packed_git();
2145 return read_packed_sha1(sha1
, type
, size
);
2148 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
2149 unsigned long *size
)
2151 const unsigned char *repl
= lookup_replace_object(sha1
);
2152 void *data
= read_object(repl
, type
, size
);
2154 /* die if we replaced an object with one that does not exist */
2155 if (!data
&& repl
!= sha1
)
2156 die("replacement %s not found for %s",
2157 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2159 /* legacy behavior is to die on corrupted objects */
2160 if (!data
&& (has_loose_object(repl
) || has_packed_and_bad(repl
)))
2161 die("object %s is corrupted", sha1_to_hex(repl
));
2166 void *read_object_with_reference(const unsigned char *sha1
,
2167 const char *required_type_name
,
2168 unsigned long *size
,
2169 unsigned char *actual_sha1_return
)
2171 enum object_type type
, required_type
;
2173 unsigned long isize
;
2174 unsigned char actual_sha1
[20];
2176 required_type
= type_from_string(required_type_name
);
2177 hashcpy(actual_sha1
, sha1
);
2179 int ref_length
= -1;
2180 const char *ref_type
= NULL
;
2182 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2185 if (type
== required_type
) {
2187 if (actual_sha1_return
)
2188 hashcpy(actual_sha1_return
, actual_sha1
);
2191 /* Handle references */
2192 else if (type
== OBJ_COMMIT
)
2194 else if (type
== OBJ_TAG
)
2195 ref_type
= "object ";
2200 ref_length
= strlen(ref_type
);
2202 if (ref_length
+ 40 > isize
||
2203 memcmp(buffer
, ref_type
, ref_length
) ||
2204 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2209 /* Now we have the ID of the referred-to object in
2210 * actual_sha1. Check again. */
2214 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2215 const char *type
, unsigned char *sha1
,
2216 char *hdr
, int *hdrlen
)
2220 /* Generate the header */
2221 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2225 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2226 git_SHA1_Update(&c
, buf
, len
);
2227 git_SHA1_Final(sha1
, &c
);
2231 * Move the just written object into its final resting place.
2232 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2233 * "moving" is only a part of what it does, when no patch between
2234 * master to pu changes the call sites of this function.
2236 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2240 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2242 else if (link(tmpfile
, filename
))
2246 * Coda hack - coda doesn't like cross-directory links,
2247 * so we fall back to a rename, which will mean that it
2248 * won't be able to check collisions, but that's not a
2251 * The same holds for FAT formatted media.
2253 * When this succeeds, we just return. We have nothing
2256 if (ret
&& ret
!= EEXIST
) {
2258 if (!rename(tmpfile
, filename
))
2262 unlink_or_warn(tmpfile
);
2264 if (ret
!= EEXIST
) {
2265 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2267 /* FIXME!!! Collision check here ? */
2271 if (set_shared_perm(filename
, (S_IFREG
|0444)))
2272 return error("unable to set permission to '%s'", filename
);
2276 static int write_buffer(int fd
, const void *buf
, size_t len
)
2278 if (write_in_full(fd
, buf
, len
) < 0)
2279 return error("file write error (%s)", strerror(errno
));
2283 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2284 unsigned char *sha1
)
2288 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2292 /* Finalize a file on disk, and close it. */
2293 static void close_sha1_file(int fd
)
2295 if (fsync_object_files
)
2296 fsync_or_die(fd
, "sha1 file");
2298 die("error when closing sha1 file (%s)", strerror(errno
));
2301 /* Size of directory component, including the ending '/' */
2302 static inline int directory_size(const char *filename
)
2304 const char *s
= strrchr(filename
, '/');
2307 return s
- filename
+ 1;
2311 * This creates a temporary file in the same directory as the final
2314 * We want to avoid cross-directory filename renames, because those
2315 * can have problems on various filesystems (FAT, NFS, Coda).
2317 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2319 int fd
, dirlen
= directory_size(filename
);
2321 if (dirlen
+ 20 > bufsiz
) {
2322 errno
= ENAMETOOLONG
;
2325 memcpy(buffer
, filename
, dirlen
);
2326 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2327 fd
= mkstemp(buffer
);
2328 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2329 /* Make sure the directory exists */
2330 memcpy(buffer
, filename
, dirlen
);
2331 buffer
[dirlen
-1] = 0;
2332 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2336 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2337 fd
= mkstemp(buffer
);
2342 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2343 void *buf
, unsigned long len
, time_t mtime
)
2347 unsigned char *compressed
;
2350 static char tmpfile
[PATH_MAX
];
2352 filename
= sha1_file_name(sha1
);
2353 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2354 while (fd
< 0 && errno
== EMFILE
&& unuse_one_window(packed_git
, -1))
2355 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2357 if (errno
== EACCES
)
2358 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2360 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2364 memset(&stream
, 0, sizeof(stream
));
2365 deflateInit(&stream
, zlib_compression_level
);
2366 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2367 compressed
= xmalloc(size
);
2370 stream
.next_out
= compressed
;
2371 stream
.avail_out
= size
;
2373 /* First header.. */
2374 stream
.next_in
= (unsigned char *)hdr
;
2375 stream
.avail_in
= hdrlen
;
2376 while (deflate(&stream
, 0) == Z_OK
)
2379 /* Then the data itself.. */
2380 stream
.next_in
= buf
;
2381 stream
.avail_in
= len
;
2382 ret
= deflate(&stream
, Z_FINISH
);
2383 if (ret
!= Z_STREAM_END
)
2384 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2386 ret
= deflateEnd(&stream
);
2388 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2390 size
= stream
.total_out
;
2392 if (write_buffer(fd
, compressed
, size
) < 0)
2393 die("unable to write sha1 file");
2394 close_sha1_file(fd
);
2400 utb
.modtime
= mtime
;
2401 if (utime(tmpfile
, &utb
) < 0)
2402 warning("failed utime() on %s: %s",
2403 tmpfile
, strerror(errno
));
2406 return move_temp_to_file(tmpfile
, filename
);
2409 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2411 unsigned char sha1
[20];
2415 /* Normally if we have it in the pack then we do not bother writing
2416 * it out into .git/objects/??/?{38} file.
2418 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2420 hashcpy(returnsha1
, sha1
);
2421 if (has_sha1_file(sha1
))
2423 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2426 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2430 enum object_type type
;
2435 if (has_loose_object(sha1
))
2437 buf
= read_packed_sha1(sha1
, &type
, &len
);
2439 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2440 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2441 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2447 int has_pack_index(const unsigned char *sha1
)
2450 if (stat(sha1_pack_index_name(sha1
), &st
))
2455 int has_pack_file(const unsigned char *sha1
)
2458 if (stat(sha1_pack_name(sha1
), &st
))
2463 int has_sha1_pack(const unsigned char *sha1
)
2465 struct pack_entry e
;
2466 return find_pack_entry(sha1
, &e
);
2469 int has_sha1_file(const unsigned char *sha1
)
2471 struct pack_entry e
;
2473 if (find_pack_entry(sha1
, &e
))
2475 return has_loose_object(sha1
);
2478 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2479 int write_object
, enum object_type type
, const char *path
)
2481 int ret
, re_allocated
= 0;
2487 * Convert blobs to git internal format
2489 if ((type
== OBJ_BLOB
) && path
) {
2490 struct strbuf nbuf
= STRBUF_INIT
;
2491 if (convert_to_git(path
, buf
, size
, &nbuf
,
2492 write_object
? safe_crlf
: 0)) {
2493 buf
= strbuf_detach(&nbuf
, &size
);
2499 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2501 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2507 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2508 enum object_type type
, const char *path
)
2511 size_t size
= xsize_t(st
->st_size
);
2513 if (!S_ISREG(st
->st_mode
)) {
2514 struct strbuf sbuf
= STRBUF_INIT
;
2515 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2516 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2520 strbuf_release(&sbuf
);
2522 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2523 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2526 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2531 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2534 struct strbuf sb
= STRBUF_INIT
;
2536 switch (st
->st_mode
& S_IFMT
) {
2538 fd
= open(path
, O_RDONLY
);
2540 return error("open(\"%s\"): %s", path
,
2542 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2543 return error("%s: failed to insert into database",
2547 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2548 char *errstr
= strerror(errno
);
2549 return error("readlink(\"%s\"): %s", path
,
2553 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2554 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2555 return error("%s: failed to insert into database",
2557 strbuf_release(&sb
);
2560 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2562 return error("%s: unsupported file type", path
);
2567 int read_pack_header(int fd
, struct pack_header
*header
)
2569 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2570 /* "eof before pack header was fully read" */
2571 return PH_ERROR_EOF
;
2573 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2574 /* "protocol error (pack signature mismatch detected)" */
2575 return PH_ERROR_PACK_SIGNATURE
;
2576 if (!pack_version_ok(header
->hdr_version
))
2577 /* "protocol error (pack version unsupported)" */
2578 return PH_ERROR_PROTOCOL
;