2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s
) { return (unsigned long)s
; }
33 static size_t sz_fmt(size_t s
) { return s
; }
36 const unsigned char null_sha1
[20];
38 static unsigned int sha1_file_open_flag
= O_NOATIME
;
40 const signed char hexval_table
[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
78 for (i
= 0; i
< 20; i
++) {
79 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
88 static inline int offset_1st_component(const char *path
)
90 if (has_dos_drive_prefix(path
))
91 return 2 + (path
[2] == '/');
95 int safe_create_leading_directories(char *path
)
97 char *pos
= path
+ offset_1st_component(path
);
101 pos
= strchr(pos
, '/');
105 if (!stat(path
, &st
)) {
107 if (!S_ISDIR(st
.st_mode
)) {
112 else if (mkdir(path
, 0777)) {
116 else if (adjust_shared_perm(path
)) {
125 char * sha1_to_hex(const unsigned char *sha1
)
128 static char hexbuffer
[4][50];
129 static const char hex
[] = "0123456789abcdef";
130 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
133 for (i
= 0; i
< 20; i
++) {
134 unsigned int val
= *sha1
++;
135 *buf
++ = hex
[val
>> 4];
136 *buf
++ = hex
[val
& 0xf];
143 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
146 for (i
= 0; i
< 20; i
++) {
147 static char hex
[] = "0123456789abcdef";
148 unsigned int val
= sha1
[i
];
149 char *pos
= pathbuf
+ i
*2 + (i
> 0);
150 *pos
++ = hex
[val
>> 4];
151 *pos
= hex
[val
& 0xf];
156 * NOTE! This returns a statically allocated buffer, so you have to be
157 * careful about using it. Do an "xstrdup()" if you need to save the
160 * Also note that this returns the location for creating. Reading
161 * SHA1 file can happen from any alternate directory listed in the
162 * DB_ENVIRONMENT environment variable if it is not found in
163 * the primary object database.
165 char *sha1_file_name(const unsigned char *sha1
)
167 static char *name
, *base
;
170 const char *sha1_file_directory
= get_object_directory();
171 int len
= strlen(sha1_file_directory
);
172 base
= xmalloc(len
+ 60);
173 memcpy(base
, sha1_file_directory
, len
);
174 memset(base
+len
, 0, 60);
177 name
= base
+ len
+ 1;
179 fill_sha1_path(name
, sha1
);
183 static char *sha1_get_pack_name(const unsigned char *sha1
,
184 char **name
, char **base
)
186 static const char hex
[] = "0123456789abcdef";
191 const char *sha1_file_directory
= get_object_directory();
192 int len
= strlen(sha1_file_directory
);
193 *base
= xmalloc(len
+ 60);
194 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory
);
195 *name
= *base
+ len
+ 11;
200 for (i
= 0; i
< 20; i
++) {
201 unsigned int val
= *sha1
++;
202 *buf
++ = hex
[val
>> 4];
203 *buf
++ = hex
[val
& 0xf];
209 char *sha1_pack_name(const unsigned char *sha1
)
211 static char *name
, *base
;
213 return sha1_get_pack_name(sha1
, &name
, &base
);
216 char *sha1_pack_index_name(const unsigned char *sha1
)
218 static char *name
, *base
;
220 return sha1_get_pack_name(sha1
, &name
, &base
);
223 struct alternate_object_database
*alt_odb_list
;
224 static struct alternate_object_database
**alt_odb_tail
;
226 static void read_info_alternates(const char * alternates
, int depth
);
229 * Prepare alternate object database registry.
231 * The variable alt_odb_list points at the list of struct
232 * alternate_object_database. The elements on this list come from
233 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
234 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
235 * whose contents is similar to that environment variable but can be
236 * LF separated. Its base points at a statically allocated buffer that
237 * contains "/the/directory/corresponding/to/.git/objects/...", while
238 * its name points just after the slash at the end of ".git/objects/"
239 * in the example above, and has enough space to hold 40-byte hex
240 * SHA1, an extra slash for the first level indirection, and the
243 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
246 const char *objdir
= get_object_directory();
247 struct alternate_object_database
*ent
;
248 struct alternate_object_database
*alt
;
249 /* 43 = 40-byte + 2 '/' + terminating NUL */
251 int entlen
= pfxlen
+ 43;
254 if (!is_absolute_path(entry
) && relative_base
) {
255 /* Relative alt-odb */
257 base_len
= strlen(relative_base
) + 1;
261 ent
= xmalloc(sizeof(*ent
) + entlen
);
263 if (!is_absolute_path(entry
) && relative_base
) {
264 memcpy(ent
->base
, relative_base
, base_len
- 1);
265 ent
->base
[base_len
- 1] = '/';
266 memcpy(ent
->base
+ base_len
, entry
, len
);
269 memcpy(ent
->base
, entry
, pfxlen
);
271 ent
->name
= ent
->base
+ pfxlen
+ 1;
272 ent
->base
[pfxlen
+ 3] = '/';
273 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
275 /* Detect cases where alternate disappeared */
276 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
277 error("object directory %s does not exist; "
278 "check .git/objects/info/alternates.",
284 /* Prevent the common mistake of listing the same
285 * thing twice, or object directory itself.
287 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
288 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
293 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
298 /* add the alternate entry */
300 alt_odb_tail
= &(ent
->next
);
303 /* recursively add alternates */
304 read_info_alternates(ent
->base
, depth
+ 1);
306 ent
->base
[pfxlen
] = '/';
311 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
312 const char *relative_base
, int depth
)
314 const char *cp
, *last
;
317 error("%s: ignoring alternate object stores, nesting too deep.",
325 if (cp
< ep
&& *cp
== '#') {
326 while (cp
< ep
&& *cp
!= sep
)
331 while (cp
< ep
&& *cp
!= sep
)
334 if (!is_absolute_path(last
) && depth
) {
335 error("%s: ignoring relative alternate object store %s",
336 relative_base
, last
);
338 link_alt_odb_entry(last
, cp
- last
,
339 relative_base
, depth
);
342 while (cp
< ep
&& *cp
== sep
)
348 static void read_info_alternates(const char * relative_base
, int depth
)
353 const char alt_file_name
[] = "info/alternates";
354 /* Given that relative_base is no longer than PATH_MAX,
355 ensure that "path" has enough space to append "/", the
356 file name, "info/alternates", and a trailing NUL. */
357 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
360 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
361 fd
= open(path
, O_RDONLY
);
364 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
368 mapsz
= xsize_t(st
.st_size
);
369 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
372 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
377 void add_to_alternates_file(const char *reference
)
379 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
380 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), 1);
381 char *alt
= mkpath("%s/objects\n", reference
);
382 write_or_die(fd
, alt
, strlen(alt
));
383 if (commit_lock_file(lock
))
384 die("could not close alternates file");
386 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
389 void prepare_alt_odb(void)
396 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
399 alt_odb_tail
= &alt_odb_list
;
400 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
402 read_info_alternates(get_object_directory(), 0);
405 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
407 char *name
= sha1_file_name(sha1
);
408 struct alternate_object_database
*alt
;
413 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
415 fill_sha1_path(name
, sha1
);
416 if (!stat(alt
->base
, st
))
422 static unsigned int pack_used_ctr
;
423 static unsigned int pack_mmap_calls
;
424 static unsigned int peak_pack_open_windows
;
425 static unsigned int pack_open_windows
;
426 static size_t peak_pack_mapped
;
427 static size_t pack_mapped
;
428 struct packed_git
*packed_git
;
430 void pack_report(void)
433 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
434 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
435 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
436 sz_fmt(getpagesize()),
437 sz_fmt(packed_git_window_size
),
438 sz_fmt(packed_git_limit
));
440 "pack_report: pack_used_ctr = %10u\n"
441 "pack_report: pack_mmap_calls = %10u\n"
442 "pack_report: pack_open_windows = %10u / %10u\n"
443 "pack_report: pack_mapped = "
444 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
447 pack_open_windows
, peak_pack_open_windows
,
448 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
451 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
454 struct pack_idx_header
*hdr
;
456 uint32_t version
, nr
, i
, *index
;
457 int fd
= open(path
, O_RDONLY
);
462 if (fstat(fd
, &st
)) {
466 idx_size
= xsize_t(st
.st_size
);
467 if (idx_size
< 4 * 256 + 20 + 20) {
469 return error("index file %s is too small", path
);
471 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
475 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
476 version
= ntohl(hdr
->idx_version
);
477 if (version
< 2 || version
> 2) {
478 munmap(idx_map
, idx_size
);
479 return error("index file %s is version %d"
480 " and is not supported by this binary"
481 " (try upgrading GIT to a newer version)",
490 index
+= 2; /* skip index header */
491 for (i
= 0; i
< 256; i
++) {
492 uint32_t n
= ntohl(index
[i
]);
494 munmap(idx_map
, idx_size
);
495 return error("non-monotonic index %s", path
);
503 * - 256 index entries 4 bytes each
504 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
505 * - 20-byte SHA1 of the packfile
506 * - 20-byte SHA1 file checksum
508 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
509 munmap(idx_map
, idx_size
);
510 return error("wrong index v1 file size in %s", path
);
512 } else if (version
== 2) {
515 * - 8 bytes of header
516 * - 256 index entries 4 bytes each
517 * - 20-byte sha1 entry * nr
518 * - 4-byte crc entry * nr
519 * - 4-byte offset entry * nr
520 * - 20-byte SHA1 of the packfile
521 * - 20-byte SHA1 file checksum
522 * And after the 4-byte offset table might be a
523 * variable sized table containing 8-byte entries
524 * for offsets larger than 2^31.
526 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
527 unsigned long max_size
= min_size
;
529 max_size
+= (nr
- 1)*8;
530 if (idx_size
< min_size
|| idx_size
> max_size
) {
531 munmap(idx_map
, idx_size
);
532 return error("wrong index v2 file size in %s", path
);
534 if (idx_size
!= min_size
&&
536 * make sure we can deal with large pack offsets.
537 * 31-bit signed offset won't be enough, neither
538 * 32-bit unsigned one will be.
540 (sizeof(off_t
) <= 4)) {
541 munmap(idx_map
, idx_size
);
542 return error("pack too large for current definition of off_t in %s", path
);
546 p
->index_version
= version
;
547 p
->index_data
= idx_map
;
548 p
->index_size
= idx_size
;
553 int open_pack_index(struct packed_git
*p
)
561 idx_name
= xstrdup(p
->pack_name
);
562 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
563 ret
= check_packed_git_idx(idx_name
, p
);
568 static void scan_windows(struct packed_git
*p
,
569 struct packed_git
**lru_p
,
570 struct pack_window
**lru_w
,
571 struct pack_window
**lru_l
)
573 struct pack_window
*w
, *w_l
;
575 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
577 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
587 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
589 struct packed_git
*p
, *lru_p
= NULL
;
590 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
593 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
594 for (p
= packed_git
; p
; p
= p
->next
)
595 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
597 munmap(lru_w
->base
, lru_w
->len
);
598 pack_mapped
-= lru_w
->len
;
600 lru_l
->next
= lru_w
->next
;
602 lru_p
->windows
= lru_w
->next
;
603 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
604 close(lru_p
->pack_fd
);
615 void release_pack_memory(size_t need
, int fd
)
617 size_t cur
= pack_mapped
;
618 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
622 void close_pack_windows(struct packed_git
*p
)
625 struct pack_window
*w
= p
->windows
;
628 die("pack '%s' still has open windows to it",
630 munmap(w
->base
, w
->len
);
631 pack_mapped
-= w
->len
;
633 p
->windows
= w
->next
;
638 void unuse_pack(struct pack_window
**w_cursor
)
640 struct pack_window
*w
= *w_cursor
;
648 * Do not call this directly as this leaks p->pack_fd on error return;
649 * call open_packed_git() instead.
651 static int open_packed_git_1(struct packed_git
*p
)
654 struct pack_header hdr
;
655 unsigned char sha1
[20];
656 unsigned char *idx_sha1
;
659 if (!p
->index_data
&& open_pack_index(p
))
660 return error("packfile %s index unavailable", p
->pack_name
);
662 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
663 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
666 /* If we created the struct before we had the pack we lack size. */
668 if (!S_ISREG(st
.st_mode
))
669 return error("packfile %s not a regular file", p
->pack_name
);
670 p
->pack_size
= st
.st_size
;
671 } else if (p
->pack_size
!= st
.st_size
)
672 return error("packfile %s size changed", p
->pack_name
);
674 /* We leave these file descriptors open with sliding mmap;
675 * there is no point keeping them open across exec(), though.
677 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
679 return error("cannot determine file descriptor flags");
680 fd_flag
|= FD_CLOEXEC
;
681 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
682 return error("cannot set FD_CLOEXEC");
684 /* Verify we recognize this pack file format. */
685 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
686 return error("file %s is far too short to be a packfile", p
->pack_name
);
687 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
688 return error("file %s is not a GIT packfile", p
->pack_name
);
689 if (!pack_version_ok(hdr
.hdr_version
))
690 return error("packfile %s is version %u and not supported"
691 " (try upgrading GIT to a newer version)",
692 p
->pack_name
, ntohl(hdr
.hdr_version
));
694 /* Verify the pack matches its index. */
695 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
696 return error("packfile %s claims to have %u objects"
697 " while index indicates %u objects",
698 p
->pack_name
, ntohl(hdr
.hdr_entries
),
700 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
701 return error("end of packfile %s is unavailable", p
->pack_name
);
702 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
703 return error("packfile %s signature is unavailable", p
->pack_name
);
704 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
705 if (hashcmp(sha1
, idx_sha1
))
706 return error("packfile %s does not match index", p
->pack_name
);
710 static int open_packed_git(struct packed_git
*p
)
712 if (!open_packed_git_1(p
))
714 if (p
->pack_fd
!= -1) {
721 static int in_window(struct pack_window
*win
, off_t offset
)
723 /* We must promise at least 20 bytes (one hash) after the
724 * offset is available from this window, otherwise the offset
725 * is not actually in this window and a different window (which
726 * has that one hash excess) must be used. This is to support
727 * the object header and delta base parsing routines below.
729 off_t win_off
= win
->offset
;
730 return win_off
<= offset
731 && (offset
+ 20) <= (win_off
+ win
->len
);
734 unsigned char* use_pack(struct packed_git
*p
,
735 struct pack_window
**w_cursor
,
739 struct pack_window
*win
= *w_cursor
;
741 if (p
->pack_fd
== -1 && open_packed_git(p
))
742 die("packfile %s cannot be accessed", p
->pack_name
);
744 /* Since packfiles end in a hash of their content and its
745 * pointless to ask for an offset into the middle of that
746 * hash, and the in_window function above wouldn't match
747 * don't allow an offset too close to the end of the file.
749 if (offset
> (p
->pack_size
- 20))
750 die("offset beyond end of packfile (truncated pack?)");
752 if (!win
|| !in_window(win
, offset
)) {
755 for (win
= p
->windows
; win
; win
= win
->next
) {
756 if (in_window(win
, offset
))
760 size_t window_align
= packed_git_window_size
/ 2;
762 win
= xcalloc(1, sizeof(*win
));
763 win
->offset
= (offset
/ window_align
) * window_align
;
764 len
= p
->pack_size
- win
->offset
;
765 if (len
> packed_git_window_size
)
766 len
= packed_git_window_size
;
767 win
->len
= (size_t)len
;
768 pack_mapped
+= win
->len
;
769 while (packed_git_limit
< pack_mapped
770 && unuse_one_window(p
, p
->pack_fd
))
772 win
->base
= xmmap(NULL
, win
->len
,
773 PROT_READ
, MAP_PRIVATE
,
774 p
->pack_fd
, win
->offset
);
775 if (win
->base
== MAP_FAILED
)
776 die("packfile %s cannot be mapped: %s",
781 if (pack_mapped
> peak_pack_mapped
)
782 peak_pack_mapped
= pack_mapped
;
783 if (pack_open_windows
> peak_pack_open_windows
)
784 peak_pack_open_windows
= pack_open_windows
;
785 win
->next
= p
->windows
;
789 if (win
!= *w_cursor
) {
790 win
->last_used
= pack_used_ctr
++;
794 offset
-= win
->offset
;
796 *left
= win
->len
- xsize_t(offset
);
797 return win
->base
+ offset
;
800 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
803 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
806 * Make sure a corresponding .pack file exists and that
807 * the index looks sane.
809 path_len
-= strlen(".idx");
812 memcpy(p
->pack_name
, path
, path_len
);
813 strcpy(p
->pack_name
+ path_len
, ".pack");
814 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
819 /* ok, it looks sane as far as we can check without
820 * actually mapping the pack file.
822 p
->index_version
= 0;
823 p
->index_data
= NULL
;
826 p
->pack_size
= st
.st_size
;
830 p
->pack_local
= local
;
831 p
->mtime
= st
.st_mtime
;
832 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
837 struct packed_git
*parse_pack_index(unsigned char *sha1
)
839 char *path
= sha1_pack_index_name(sha1
);
840 return parse_pack_index_file(sha1
, path
);
843 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
844 const char *idx_path
)
846 const char *path
= sha1_pack_name(sha1
);
847 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
849 if (check_packed_git_idx(idx_path
, p
)) {
854 strcpy(p
->pack_name
, path
);
859 hashcpy(p
->sha1
, sha1
);
863 void install_packed_git(struct packed_git
*pack
)
865 pack
->next
= packed_git
;
869 static void prepare_packed_git_one(char *objdir
, int local
)
871 /* Ensure that this buffer is large enough so that we can
872 append "/pack/" without clobbering the stack even if
873 strlen(objdir) were PATH_MAX. */
874 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
879 sprintf(path
, "%s/pack", objdir
);
884 error("unable to open object pack directory: %s: %s",
885 path
, strerror(errno
));
889 while ((de
= readdir(dir
)) != NULL
) {
890 int namelen
= strlen(de
->d_name
);
891 struct packed_git
*p
;
893 if (!has_extension(de
->d_name
, ".idx"))
896 if (len
+ namelen
+ 1 > sizeof(path
))
899 /* Don't reopen a pack we already have. */
900 strcpy(path
+ len
, de
->d_name
);
901 for (p
= packed_git
; p
; p
= p
->next
) {
902 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
907 /* See if it really is a valid .idx file with corresponding
908 * .pack file that we can map.
910 p
= add_packed_git(path
, len
+ namelen
, local
);
913 install_packed_git(p
);
918 static int sort_pack(const void *a_
, const void *b_
)
920 struct packed_git
*a
= *((struct packed_git
**)a_
);
921 struct packed_git
*b
= *((struct packed_git
**)b_
);
925 * Local packs tend to contain objects specific to our
926 * variant of the project than remote ones. In addition,
927 * remote ones could be on a network mounted filesystem.
928 * Favor local ones for these reasons.
930 st
= a
->pack_local
- b
->pack_local
;
935 * Younger packs tend to contain more recent objects,
936 * and more recent objects tend to get accessed more
939 if (a
->mtime
< b
->mtime
)
941 else if (a
->mtime
== b
->mtime
)
946 static void rearrange_packed_git(void)
948 struct packed_git
**ary
, *p
;
951 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
956 /* prepare an array of packed_git for easier sorting */
957 ary
= xcalloc(n
, sizeof(struct packed_git
*));
958 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
961 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
963 /* link them back again */
964 for (i
= 0; i
< n
- 1; i
++)
965 ary
[i
]->next
= ary
[i
+ 1];
966 ary
[n
- 1]->next
= NULL
;
972 static int prepare_packed_git_run_once
= 0;
973 void prepare_packed_git(void)
975 struct alternate_object_database
*alt
;
977 if (prepare_packed_git_run_once
)
979 prepare_packed_git_one(get_object_directory(), 1);
981 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
983 prepare_packed_git_one(alt
->base
, 0);
986 rearrange_packed_git();
987 prepare_packed_git_run_once
= 1;
990 void reprepare_packed_git(void)
992 prepare_packed_git_run_once
= 0;
993 prepare_packed_git();
996 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
998 unsigned char real_sha1
[20];
999 hash_sha1_file(map
, size
, type
, real_sha1
);
1000 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1003 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1008 char *filename
= find_sha1_file(sha1
, &st
);
1014 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
1016 /* See if it works without O_NOATIME */
1017 switch (sha1_file_open_flag
) {
1019 fd
= open(filename
, O_RDONLY
);
1027 /* If it failed once, it will probably fail again.
1028 * Stop using O_NOATIME
1030 sha1_file_open_flag
= 0;
1032 *size
= xsize_t(st
.st_size
);
1033 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1038 static int legacy_loose_object(unsigned char *map
)
1043 * Is it a zlib-compressed buffer? If so, the first byte
1044 * must be 0x78 (15-bit window size, deflated), and the
1045 * first 16-bit word is evenly divisible by 31
1047 word
= (map
[0] << 8) + map
[1];
1048 if (map
[0] == 0x78 && !(word
% 31))
1054 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1059 unsigned long used
= 0;
1062 *type
= (c
>> 4) & 7;
1068 if (sizeof(long) * 8 <= shift
)
1071 size
+= (c
& 0x7f) << shift
;
1078 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1080 unsigned long size
, used
;
1081 static const char valid_loose_object_type
[8] = {
1083 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1084 0, /* "delta" and others are invalid in a loose object */
1086 enum object_type type
;
1088 /* Get the data stream */
1089 memset(stream
, 0, sizeof(*stream
));
1090 stream
->next_in
= map
;
1091 stream
->avail_in
= mapsize
;
1092 stream
->next_out
= buffer
;
1093 stream
->avail_out
= bufsiz
;
1095 if (legacy_loose_object(map
)) {
1096 inflateInit(stream
);
1097 return inflate(stream
, 0);
1102 * There used to be a second loose object header format which
1103 * was meant to mimic the in-pack format, allowing for direct
1104 * copy of the object data. This format turned up not to be
1105 * really worth it and we don't write it any longer. But we
1106 * can still read it.
1108 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1109 if (!used
|| !valid_loose_object_type
[type
])
1114 /* Set up the stream for the rest.. */
1115 stream
->next_in
= map
;
1116 stream
->avail_in
= mapsize
;
1117 inflateInit(stream
);
1119 /* And generate the fake traditional header */
1120 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1121 typename(type
), size
);
1125 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1127 int bytes
= strlen(buffer
) + 1;
1128 unsigned char *buf
= xmalloc(1+size
);
1132 n
= stream
->total_out
- bytes
;
1135 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1137 if (bytes
<= size
) {
1139 * The above condition must be (bytes <= size), not
1140 * (bytes < size). In other words, even though we
1141 * expect no more output and set avail_out to zer0,
1142 * the input zlib stream may have bytes that express
1143 * "this concludes the stream", and we *do* want to
1146 * Otherwise we would not be able to test that we
1147 * consumed all the input to reach the expected size;
1148 * we also want to check that zlib tells us that all
1149 * went well with status == Z_STREAM_END at the end.
1151 stream
->next_out
= buf
+ bytes
;
1152 stream
->avail_out
= size
- bytes
;
1153 while (status
== Z_OK
)
1154 status
= inflate(stream
, Z_FINISH
);
1157 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1163 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1164 else if (stream
->avail_in
)
1165 error("garbage at end of loose object '%s'",
1172 * We used to just use "sscanf()", but that's actually way
1173 * too permissive for what we want to check. So do an anal
1174 * object header parse by hand.
1176 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1183 * The type can be at most ten bytes (including the
1184 * terminating '\0' that we add), and is followed by
1193 if (i
>= sizeof(type
))
1199 * The length must follow immediately, and be in canonical
1200 * decimal format (ie "010" is not valid).
1202 size
= *hdr
++ - '0';
1207 unsigned long c
= *hdr
- '0';
1211 size
= size
* 10 + c
;
1217 * The length must be followed by a zero byte
1219 return *hdr
? -1 : type_from_string(type
);
1222 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1228 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1229 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1232 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1235 unsigned long get_size_from_delta(struct packed_git
*p
,
1236 struct pack_window
**w_curs
,
1239 const unsigned char *data
;
1240 unsigned char delta_head
[20], *in
;
1244 memset(&stream
, 0, sizeof(stream
));
1245 stream
.next_out
= delta_head
;
1246 stream
.avail_out
= sizeof(delta_head
);
1248 inflateInit(&stream
);
1250 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1251 stream
.next_in
= in
;
1252 st
= inflate(&stream
, Z_FINISH
);
1253 curpos
+= stream
.next_in
- in
;
1254 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1255 stream
.total_out
< sizeof(delta_head
));
1256 inflateEnd(&stream
);
1257 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1258 die("delta data unpack-initial failed");
1260 /* Examine the initial part of the delta to figure out
1265 /* ignore base size */
1266 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1268 /* Read the result size */
1269 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1272 static off_t
get_delta_base(struct packed_git
*p
,
1273 struct pack_window
**w_curs
,
1275 enum object_type type
,
1276 off_t delta_obj_offset
)
1278 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1281 /* use_pack() assured us we have [base_info, base_info + 20)
1282 * as a range that we can look at without walking off the
1283 * end of the mapped window. Its actually the hash size
1284 * that is assured. An OFS_DELTA longer than the hash size
1285 * is stupid, as then a REF_DELTA would be smaller to store.
1287 if (type
== OBJ_OFS_DELTA
) {
1289 unsigned char c
= base_info
[used
++];
1290 base_offset
= c
& 127;
1293 if (!base_offset
|| MSB(base_offset
, 7))
1294 die("offset value overflow for delta base object");
1295 c
= base_info
[used
++];
1296 base_offset
= (base_offset
<< 7) + (c
& 127);
1298 base_offset
= delta_obj_offset
- base_offset
;
1299 if (base_offset
>= delta_obj_offset
)
1300 die("delta base offset out of bound");
1302 } else if (type
== OBJ_REF_DELTA
) {
1303 /* The base entry _must_ be in the same pack */
1304 base_offset
= find_pack_entry_one(base_info
, p
);
1306 die("failed to find delta-pack base object %s",
1307 sha1_to_hex(base_info
));
1310 die("I am totally screwed");
1314 /* forward declaration for a mutually recursive function */
1315 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1316 unsigned long *sizep
);
1318 static int packed_delta_info(struct packed_git
*p
,
1319 struct pack_window
**w_curs
,
1321 enum object_type type
,
1323 unsigned long *sizep
)
1327 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1328 type
= packed_object_info(p
, base_offset
, NULL
);
1330 /* We choose to only get the type of the base object and
1331 * ignore potentially corrupt pack file that expects the delta
1332 * based on a base with a wrong size. This saves tons of
1336 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1341 static int unpack_object_header(struct packed_git
*p
,
1342 struct pack_window
**w_curs
,
1344 unsigned long *sizep
)
1346 unsigned char *base
;
1349 enum object_type type
;
1351 /* use_pack() assures us we have [base, base + 20) available
1352 * as a range that we can look at at. (Its actually the hash
1353 * size that is assured.) With our object header encoding
1354 * the maximum deflated object size is 2^137, which is just
1355 * insane, so we know won't exceed what we have been given.
1357 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1358 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1360 die("object offset outside of pack file");
1366 const char *packed_object_info_detail(struct packed_git
*p
,
1368 unsigned long *size
,
1369 unsigned long *store_size
,
1370 unsigned int *delta_chain_length
,
1371 unsigned char *base_sha1
)
1373 struct pack_window
*w_curs
= NULL
;
1375 unsigned long dummy
;
1376 unsigned char *next_sha1
;
1377 enum object_type type
;
1378 struct revindex_entry
*revidx
;
1380 *delta_chain_length
= 0;
1381 curpos
= obj_offset
;
1382 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1384 revidx
= find_pack_revindex(p
, obj_offset
);
1385 *store_size
= revidx
[1].offset
- obj_offset
;
1390 die("pack %s contains unknown object type %d",
1391 p
->pack_name
, type
);
1396 unuse_pack(&w_curs
);
1397 return typename(type
);
1399 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1400 if (*delta_chain_length
== 0) {
1401 revidx
= find_pack_revindex(p
, obj_offset
);
1402 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1406 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1407 if (*delta_chain_length
== 0)
1408 hashcpy(base_sha1
, next_sha1
);
1409 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1412 (*delta_chain_length
)++;
1413 curpos
= obj_offset
;
1414 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1418 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1419 unsigned long *sizep
)
1421 struct pack_window
*w_curs
= NULL
;
1423 off_t curpos
= obj_offset
;
1424 enum object_type type
;
1426 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1431 type
= packed_delta_info(p
, &w_curs
, curpos
,
1432 type
, obj_offset
, sizep
);
1442 die("pack %s contains unknown object type %d",
1443 p
->pack_name
, type
);
1445 unuse_pack(&w_curs
);
1449 static void *unpack_compressed_entry(struct packed_git
*p
,
1450 struct pack_window
**w_curs
,
1456 unsigned char *buffer
, *in
;
1458 buffer
= xmalloc(size
+ 1);
1460 memset(&stream
, 0, sizeof(stream
));
1461 stream
.next_out
= buffer
;
1462 stream
.avail_out
= size
;
1464 inflateInit(&stream
);
1466 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1467 stream
.next_in
= in
;
1468 st
= inflate(&stream
, Z_FINISH
);
1469 curpos
+= stream
.next_in
- in
;
1470 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1471 inflateEnd(&stream
);
1472 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1480 #define MAX_DELTA_CACHE (256)
1482 static size_t delta_base_cached
;
1484 static struct delta_base_cache_lru_list
{
1485 struct delta_base_cache_lru_list
*prev
;
1486 struct delta_base_cache_lru_list
*next
;
1487 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1489 static struct delta_base_cache_entry
{
1490 struct delta_base_cache_lru_list lru
;
1492 struct packed_git
*p
;
1495 enum object_type type
;
1496 } delta_base_cache
[MAX_DELTA_CACHE
];
1498 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1502 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1503 hash
+= (hash
>> 8) + (hash
>> 16);
1504 return hash
% MAX_DELTA_CACHE
;
1507 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1508 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1511 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1512 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1515 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1516 goto found_cache_entry
;
1517 return unpack_entry(p
, base_offset
, type
, base_size
);
1522 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1523 ent
->lru
.prev
->next
= ent
->lru
.next
;
1524 delta_base_cached
-= ent
->size
;
1526 ret
= xmemdupz(ent
->data
, ent
->size
);
1529 *base_size
= ent
->size
;
1533 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1538 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1539 ent
->lru
.prev
->next
= ent
->lru
.next
;
1540 delta_base_cached
-= ent
->size
;
1544 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1545 void *base
, unsigned long base_size
, enum object_type type
)
1547 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1548 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1549 struct delta_base_cache_lru_list
*lru
;
1551 release_delta_base_cache(ent
);
1552 delta_base_cached
+= base_size
;
1554 for (lru
= delta_base_cache_lru
.next
;
1555 delta_base_cached
> delta_base_cache_limit
1556 && lru
!= &delta_base_cache_lru
;
1558 struct delta_base_cache_entry
*f
= (void *)lru
;
1559 if (f
->type
== OBJ_BLOB
)
1560 release_delta_base_cache(f
);
1562 for (lru
= delta_base_cache_lru
.next
;
1563 delta_base_cached
> delta_base_cache_limit
1564 && lru
!= &delta_base_cache_lru
;
1566 struct delta_base_cache_entry
*f
= (void *)lru
;
1567 release_delta_base_cache(f
);
1571 ent
->base_offset
= base_offset
;
1574 ent
->size
= base_size
;
1575 ent
->lru
.next
= &delta_base_cache_lru
;
1576 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1577 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1578 delta_base_cache_lru
.prev
= &ent
->lru
;
1581 static void *unpack_delta_entry(struct packed_git
*p
,
1582 struct pack_window
**w_curs
,
1584 unsigned long delta_size
,
1586 enum object_type
*type
,
1587 unsigned long *sizep
)
1589 void *delta_data
, *result
, *base
;
1590 unsigned long base_size
;
1593 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1594 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1596 die("failed to read delta base object"
1597 " at %"PRIuMAX
" from %s",
1598 (uintmax_t)base_offset
, p
->pack_name
);
1600 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1602 die("failed to unpack compressed delta"
1603 " at %"PRIuMAX
" from %s",
1604 (uintmax_t)curpos
, p
->pack_name
);
1605 result
= patch_delta(base
, base_size
,
1606 delta_data
, delta_size
,
1609 die("failed to apply delta");
1611 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1615 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1616 enum object_type
*type
, unsigned long *sizep
)
1618 struct pack_window
*w_curs
= NULL
;
1619 off_t curpos
= obj_offset
;
1622 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1626 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1627 obj_offset
, type
, sizep
);
1633 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1636 die("unknown object type %i in %s", *type
, p
->pack_name
);
1638 unuse_pack(&w_curs
);
1642 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1645 const unsigned char *index
= p
->index_data
;
1647 if (open_pack_index(p
))
1649 index
= p
->index_data
;
1651 if (n
>= p
->num_objects
)
1654 if (p
->index_version
== 1) {
1655 return index
+ 24 * n
+ 4;
1658 return index
+ 20 * n
;
1662 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1664 const unsigned char *index
= p
->index_data
;
1666 if (p
->index_version
== 1) {
1667 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1670 index
+= 8 + p
->num_objects
* (20 + 4);
1671 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1672 if (!(off
& 0x80000000))
1674 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1675 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1676 ntohl(*((uint32_t *)(index
+ 4)));
1680 off_t
find_pack_entry_one(const unsigned char *sha1
,
1681 struct packed_git
*p
)
1683 const uint32_t *level1_ofs
= p
->index_data
;
1684 const unsigned char *index
= p
->index_data
;
1685 unsigned hi
, lo
, stride
;
1686 static int use_lookup
= -1;
1687 static int debug_lookup
= -1;
1689 if (debug_lookup
< 0)
1690 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1693 if (open_pack_index(p
))
1695 level1_ofs
= p
->index_data
;
1696 index
= p
->index_data
;
1698 if (p
->index_version
> 1) {
1703 hi
= ntohl(level1_ofs
[*sha1
]);
1704 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1705 if (p
->index_version
> 1) {
1713 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1714 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1717 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1719 int pos
= sha1_entry_pos(index
, stride
, 0,
1720 lo
, hi
, p
->num_objects
, sha1
);
1723 return nth_packed_object_offset(p
, pos
);
1727 unsigned mi
= (lo
+ hi
) / 2;
1728 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1731 printf("lo %u hi %u rg %u mi %u\n",
1732 lo
, hi
, hi
- lo
, mi
);
1734 return nth_packed_object_offset(p
, mi
);
1743 int matches_pack_name(struct packed_git
*p
, const char *name
)
1745 const char *last_c
, *c
;
1747 if (!strcmp(p
->pack_name
, name
))
1750 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1755 if (!strcmp(last_c
, name
))
1761 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1763 static struct packed_git
*last_found
= (void *)1;
1764 struct packed_git
*p
;
1767 prepare_packed_git();
1770 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1773 if (ignore_packed
) {
1775 for (ig
= ignore_packed
; *ig
; ig
++)
1776 if (matches_pack_name(p
, *ig
))
1782 offset
= find_pack_entry_one(sha1
, p
);
1785 * We are about to tell the caller where they can
1786 * locate the requested object. We better make
1787 * sure the packfile is still here and can be
1788 * accessed before supplying that answer, as
1789 * it may have been deleted since the index
1792 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1793 error("packfile %s cannot be accessed", p
->pack_name
);
1798 hashcpy(e
->sha1
, sha1
);
1804 if (p
== last_found
)
1808 if (p
== last_found
)
1814 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1815 struct packed_git
*packs
)
1817 struct packed_git
*p
;
1819 for (p
= packs
; p
; p
= p
->next
) {
1820 if (find_pack_entry_one(sha1
, p
))
1827 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1830 unsigned long mapsize
, size
;
1835 map
= map_sha1_file(sha1
, &mapsize
);
1837 return error("unable to find %s", sha1_to_hex(sha1
));
1838 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1839 status
= error("unable to unpack %s header",
1841 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1842 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1845 inflateEnd(&stream
);
1846 munmap(map
, mapsize
);
1850 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1852 struct pack_entry e
;
1854 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1855 reprepare_packed_git();
1856 if (!find_pack_entry(sha1
, &e
, NULL
))
1857 return sha1_loose_object_info(sha1
, sizep
);
1859 return packed_object_info(e
.p
, e
.offset
, sizep
);
1862 static void *read_packed_sha1(const unsigned char *sha1
,
1863 enum object_type
*type
, unsigned long *size
)
1865 struct pack_entry e
;
1867 if (!find_pack_entry(sha1
, &e
, NULL
))
1870 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1874 * This is meant to hold a *small* number of objects that you would
1875 * want read_sha1_file() to be able to return, but yet you do not want
1876 * to write them into the object store (e.g. a browse-only
1879 static struct cached_object
{
1880 unsigned char sha1
[20];
1881 enum object_type type
;
1885 static int cached_object_nr
, cached_object_alloc
;
1887 static struct cached_object empty_tree
= {
1888 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1889 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1890 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1896 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1899 struct cached_object
*co
= cached_objects
;
1901 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1902 if (!hashcmp(co
->sha1
, sha1
))
1905 if (!hashcmp(sha1
, empty_tree
.sha1
))
1910 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1911 unsigned char *sha1
)
1913 struct cached_object
*co
;
1915 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1916 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1918 if (cached_object_alloc
<= cached_object_nr
) {
1919 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1920 cached_objects
= xrealloc(cached_objects
,
1921 sizeof(*cached_objects
) *
1922 cached_object_alloc
);
1924 co
= &cached_objects
[cached_object_nr
++];
1927 co
->buf
= xmalloc(len
);
1928 memcpy(co
->buf
, buf
, len
);
1929 hashcpy(co
->sha1
, sha1
);
1933 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1934 unsigned long *size
)
1936 unsigned long mapsize
;
1938 struct cached_object
*co
;
1940 co
= find_cached_object(sha1
);
1944 return xmemdupz(co
->buf
, co
->size
);
1947 buf
= read_packed_sha1(sha1
, type
, size
);
1950 map
= map_sha1_file(sha1
, &mapsize
);
1952 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1953 munmap(map
, mapsize
);
1956 reprepare_packed_git();
1957 return read_packed_sha1(sha1
, type
, size
);
1960 void *read_object_with_reference(const unsigned char *sha1
,
1961 const char *required_type_name
,
1962 unsigned long *size
,
1963 unsigned char *actual_sha1_return
)
1965 enum object_type type
, required_type
;
1967 unsigned long isize
;
1968 unsigned char actual_sha1
[20];
1970 required_type
= type_from_string(required_type_name
);
1971 hashcpy(actual_sha1
, sha1
);
1973 int ref_length
= -1;
1974 const char *ref_type
= NULL
;
1976 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1979 if (type
== required_type
) {
1981 if (actual_sha1_return
)
1982 hashcpy(actual_sha1_return
, actual_sha1
);
1985 /* Handle references */
1986 else if (type
== OBJ_COMMIT
)
1988 else if (type
== OBJ_TAG
)
1989 ref_type
= "object ";
1994 ref_length
= strlen(ref_type
);
1996 if (ref_length
+ 40 > isize
||
1997 memcmp(buffer
, ref_type
, ref_length
) ||
1998 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2003 /* Now we have the ID of the referred-to object in
2004 * actual_sha1. Check again. */
2008 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2009 const char *type
, unsigned char *sha1
,
2010 char *hdr
, int *hdrlen
)
2014 /* Generate the header */
2015 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2019 SHA1_Update(&c
, hdr
, *hdrlen
);
2020 SHA1_Update(&c
, buf
, len
);
2021 SHA1_Final(sha1
, &c
);
2025 * Link the tempfile to the final place, possibly creating the
2026 * last directory level as you do so.
2028 * Returns the errno on failure, 0 on success.
2030 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
2035 if (!link(tmpfile
, filename
))
2039 * Try to mkdir the last path component if that failed.
2041 * Re-try the "link()" regardless of whether the mkdir
2042 * succeeds, since a race might mean that somebody
2046 dir
= strrchr(filename
, '/');
2049 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
2054 if (!link(tmpfile
, filename
))
2062 * Move the just written object into its final resting place
2064 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2066 int ret
= link_temp_to_file(tmpfile
, filename
);
2069 * Coda hack - coda doesn't like cross-directory links,
2070 * so we fall back to a rename, which will mean that it
2071 * won't be able to check collisions, but that's not a
2074 * The same holds for FAT formatted media.
2076 * When this succeeds, we just return 0. We have nothing
2079 if (ret
&& ret
!= EEXIST
) {
2080 if (!rename(tmpfile
, filename
))
2086 if (ret
!= EEXIST
) {
2087 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2089 /* FIXME!!! Collision check here ? */
2095 static int write_buffer(int fd
, const void *buf
, size_t len
)
2097 if (write_in_full(fd
, buf
, len
) < 0)
2098 return error("file write error (%s)", strerror(errno
));
2102 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2103 unsigned char *sha1
)
2107 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2111 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2112 void *buf
, unsigned long len
, time_t mtime
)
2115 unsigned char *compressed
;
2118 static char tmpfile
[PATH_MAX
];
2120 filename
= sha1_file_name(sha1
);
2121 fd
= open(filename
, O_RDONLY
);
2124 * FIXME!!! We might do collision checking here, but we'd
2125 * need to uncompress the old file and check it. Later.
2131 if (errno
!= ENOENT
) {
2132 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2135 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2137 fd
= mkstemp(tmpfile
);
2140 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2142 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2146 memset(&stream
, 0, sizeof(stream
));
2147 deflateInit(&stream
, zlib_compression_level
);
2148 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2149 compressed
= xmalloc(size
);
2152 stream
.next_out
= compressed
;
2153 stream
.avail_out
= size
;
2155 /* First header.. */
2156 stream
.next_in
= (unsigned char *)hdr
;
2157 stream
.avail_in
= hdrlen
;
2158 while (deflate(&stream
, 0) == Z_OK
)
2161 /* Then the data itself.. */
2162 stream
.next_in
= buf
;
2163 stream
.avail_in
= len
;
2164 ret
= deflate(&stream
, Z_FINISH
);
2165 if (ret
!= Z_STREAM_END
)
2166 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2168 ret
= deflateEnd(&stream
);
2170 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2172 size
= stream
.total_out
;
2174 if (write_buffer(fd
, compressed
, size
) < 0)
2175 die("unable to write sha1 file");
2178 die("unable to write sha1 file");
2184 utb
.modtime
= mtime
;
2185 if (utime(tmpfile
, &utb
) < 0)
2186 warning("failed utime() on %s: %s",
2187 tmpfile
, strerror(errno
));
2190 return move_temp_to_file(tmpfile
, filename
);
2193 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2195 unsigned char sha1
[20];
2199 /* Normally if we have it in the pack then we do not bother writing
2200 * it out into .git/objects/??/?{38} file.
2202 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2204 hashcpy(returnsha1
, sha1
);
2205 if (has_sha1_file(sha1
))
2207 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2210 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2215 enum object_type type
;
2219 if (find_sha1_file(sha1
, &st
))
2221 buf
= read_packed_sha1(sha1
, &type
, &len
);
2223 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2224 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2225 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2229 * We need to unpack and recompress the object for writing
2230 * it out to a different file.
2232 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2236 unsigned char *unpacked
;
2238 enum object_type type
;
2243 /* need to unpack and recompress it by itself */
2244 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2246 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2248 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2251 memset(&stream
, 0, sizeof(stream
));
2252 deflateInit(&stream
, zlib_compression_level
);
2253 size
= deflateBound(&stream
, len
+ hdrlen
);
2254 buf
= xmalloc(size
);
2257 stream
.next_out
= buf
;
2258 stream
.avail_out
= size
;
2260 /* First header.. */
2261 stream
.next_in
= (void *)hdr
;
2262 stream
.avail_in
= hdrlen
;
2263 while (deflate(&stream
, 0) == Z_OK
)
2266 /* Then the data itself.. */
2267 stream
.next_in
= unpacked
;
2268 stream
.avail_in
= len
;
2269 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2271 deflateEnd(&stream
);
2274 *objsize
= stream
.total_out
;
2278 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2281 unsigned long objsize
;
2282 void *buf
= map_sha1_file(sha1
, &objsize
);
2285 retval
= write_buffer(fd
, buf
, objsize
);
2286 munmap(buf
, objsize
);
2290 buf
= repack_object(sha1
, &objsize
);
2291 retval
= write_buffer(fd
, buf
, objsize
);
2296 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2297 size_t bufsize
, size_t *bufposn
)
2299 char tmpfile
[PATH_MAX
];
2302 unsigned char real_sha1
[20];
2303 unsigned char discard
[4096];
2307 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2309 local
= mkstemp(tmpfile
);
2312 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2314 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2317 memset(&stream
, 0, sizeof(stream
));
2319 inflateInit(&stream
);
2326 stream
.avail_in
= *bufposn
;
2327 stream
.next_in
= (unsigned char *) buffer
;
2329 stream
.next_out
= discard
;
2330 stream
.avail_out
= sizeof(discard
);
2331 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2332 SHA1_Update(&c
, discard
, sizeof(discard
) -
2334 } while (stream
.avail_in
&& ret
== Z_OK
);
2335 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2336 die("unable to write sha1 file");
2337 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2339 *bufposn
= stream
.avail_in
;
2343 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2348 return error("Connection closed?");
2349 perror("Reading from connection");
2354 inflateEnd(&stream
);
2356 fchmod(local
, 0444);
2357 if (close(local
) != 0)
2358 die("unable to write sha1 file");
2359 SHA1_Final(real_sha1
, &c
);
2360 if (ret
!= Z_STREAM_END
) {
2362 return error("File %s corrupted", sha1_to_hex(sha1
));
2364 if (hashcmp(sha1
, real_sha1
)) {
2366 return error("File %s has bad hash", sha1_to_hex(sha1
));
2369 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2372 int has_pack_index(const unsigned char *sha1
)
2375 if (stat(sha1_pack_index_name(sha1
), &st
))
2380 int has_pack_file(const unsigned char *sha1
)
2383 if (stat(sha1_pack_name(sha1
), &st
))
2388 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2390 struct pack_entry e
;
2391 return find_pack_entry(sha1
, &e
, ignore_packed
);
2394 int has_sha1_file(const unsigned char *sha1
)
2397 struct pack_entry e
;
2399 if (find_pack_entry(sha1
, &e
, NULL
))
2401 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2404 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2409 strbuf_init(&buf
, 0);
2410 if (strbuf_read(&buf
, fd
, 4096) < 0) {
2411 strbuf_release(&buf
);
2418 ret
= write_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2420 ret
= hash_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2421 strbuf_release(&buf
);
2426 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2427 enum object_type type
, const char *path
)
2429 size_t size
= xsize_t(st
->st_size
);
2431 int ret
, re_allocated
= 0;
2434 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2441 * Convert blobs to git internal format
2443 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2445 strbuf_init(&nbuf
, 0);
2446 if (convert_to_git(path
, buf
, size
, &nbuf
,
2447 write_object
? safe_crlf
: 0)) {
2449 buf
= strbuf_detach(&nbuf
, &size
);
2455 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2457 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2467 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2473 switch (st
->st_mode
& S_IFMT
) {
2475 fd
= open(path
, O_RDONLY
);
2477 return error("open(\"%s\"): %s", path
,
2479 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2480 return error("%s: failed to insert into database",
2484 len
= xsize_t(st
->st_size
);
2485 target
= xmalloc(len
+ 1);
2486 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2487 char *errstr
= strerror(errno
);
2489 return error("readlink(\"%s\"): %s", path
,
2493 hash_sha1_file(target
, len
, blob_type
, sha1
);
2494 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2495 return error("%s: failed to insert into database",
2500 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2502 return error("%s: unsupported file type", path
);
2507 int read_pack_header(int fd
, struct pack_header
*header
)
2509 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2510 /* "eof before pack header was fully read" */
2511 return PH_ERROR_EOF
;
2513 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2514 /* "protocol error (pack signature mismatch detected)" */
2515 return PH_ERROR_PACK_SIGNATURE
;
2516 if (!pack_version_ok(header
->hdr_version
))
2517 /* "protocol error (pack version unsupported)" */
2518 return PH_ERROR_PROTOCOL
;