2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s
) { return (unsigned long)s
; }
33 static size_t sz_fmt(size_t s
) { return s
; }
36 const unsigned char null_sha1
[20];
38 const signed char hexval_table
[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
76 for (i
= 0; i
< 20; i
++) {
77 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
86 int safe_create_leading_directories(char *path
)
91 if (is_absolute_path(path
))
95 pos
= strchr(pos
, '/');
99 if (!stat(path
, &st
)) {
101 if (!S_ISDIR(st
.st_mode
)) {
106 else if (mkdir(path
, 0777)) {
110 else if (adjust_shared_perm(path
)) {
119 char *sha1_to_hex(const unsigned char *sha1
)
122 static char hexbuffer
[4][50];
123 static const char hex
[] = "0123456789abcdef";
124 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
127 for (i
= 0; i
< 20; i
++) {
128 unsigned int val
= *sha1
++;
129 *buf
++ = hex
[val
>> 4];
130 *buf
++ = hex
[val
& 0xf];
137 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
140 for (i
= 0; i
< 20; i
++) {
141 static char hex
[] = "0123456789abcdef";
142 unsigned int val
= sha1
[i
];
143 char *pos
= pathbuf
+ i
*2 + (i
> 0);
144 *pos
++ = hex
[val
>> 4];
145 *pos
= hex
[val
& 0xf];
150 * NOTE! This returns a statically allocated buffer, so you have to be
151 * careful about using it. Do an "xstrdup()" if you need to save the
154 * Also note that this returns the location for creating. Reading
155 * SHA1 file can happen from any alternate directory listed in the
156 * DB_ENVIRONMENT environment variable if it is not found in
157 * the primary object database.
159 char *sha1_file_name(const unsigned char *sha1
)
161 static char *name
, *base
;
164 const char *sha1_file_directory
= get_object_directory();
165 int len
= strlen(sha1_file_directory
);
166 base
= xmalloc(len
+ 60);
167 memcpy(base
, sha1_file_directory
, len
);
168 memset(base
+len
, 0, 60);
171 name
= base
+ len
+ 1;
173 fill_sha1_path(name
, sha1
);
177 static char *sha1_get_pack_name(const unsigned char *sha1
,
178 char **name
, char **base
, const char *which
)
180 static const char hex
[] = "0123456789abcdef";
185 const char *sha1_file_directory
= get_object_directory();
186 int len
= strlen(sha1_file_directory
);
187 *base
= xmalloc(len
+ 60);
188 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
189 sha1_file_directory
, which
);
190 *name
= *base
+ len
+ 11;
195 for (i
= 0; i
< 20; i
++) {
196 unsigned int val
= *sha1
++;
197 *buf
++ = hex
[val
>> 4];
198 *buf
++ = hex
[val
& 0xf];
204 char *sha1_pack_name(const unsigned char *sha1
)
206 static char *name
, *base
;
208 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
211 char *sha1_pack_index_name(const unsigned char *sha1
)
213 static char *name
, *base
;
215 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
218 struct alternate_object_database
*alt_odb_list
;
219 static struct alternate_object_database
**alt_odb_tail
;
221 static void read_info_alternates(const char * alternates
, int depth
);
224 * Prepare alternate object database registry.
226 * The variable alt_odb_list points at the list of struct
227 * alternate_object_database. The elements on this list come from
228 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
229 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
230 * whose contents is similar to that environment variable but can be
231 * LF separated. Its base points at a statically allocated buffer that
232 * contains "/the/directory/corresponding/to/.git/objects/...", while
233 * its name points just after the slash at the end of ".git/objects/"
234 * in the example above, and has enough space to hold 40-byte hex
235 * SHA1, an extra slash for the first level indirection, and the
238 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
241 const char *objdir
= get_object_directory();
242 struct alternate_object_database
*ent
;
243 struct alternate_object_database
*alt
;
244 /* 43 = 40-byte + 2 '/' + terminating NUL */
246 int entlen
= pfxlen
+ 43;
249 if (!is_absolute_path(entry
) && relative_base
) {
250 /* Relative alt-odb */
252 base_len
= strlen(relative_base
) + 1;
256 ent
= xmalloc(sizeof(*ent
) + entlen
);
258 if (!is_absolute_path(entry
) && relative_base
) {
259 memcpy(ent
->base
, relative_base
, base_len
- 1);
260 ent
->base
[base_len
- 1] = '/';
261 memcpy(ent
->base
+ base_len
, entry
, len
);
264 memcpy(ent
->base
, entry
, pfxlen
);
266 ent
->name
= ent
->base
+ pfxlen
+ 1;
267 ent
->base
[pfxlen
+ 3] = '/';
268 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
270 /* Detect cases where alternate disappeared */
271 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
272 error("object directory %s does not exist; "
273 "check .git/objects/info/alternates.",
279 /* Prevent the common mistake of listing the same
280 * thing twice, or object directory itself.
282 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
283 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
288 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
293 /* add the alternate entry */
295 alt_odb_tail
= &(ent
->next
);
298 /* recursively add alternates */
299 read_info_alternates(ent
->base
, depth
+ 1);
301 ent
->base
[pfxlen
] = '/';
306 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
307 const char *relative_base
, int depth
)
309 const char *cp
, *last
;
312 error("%s: ignoring alternate object stores, nesting too deep.",
320 if (cp
< ep
&& *cp
== '#') {
321 while (cp
< ep
&& *cp
!= sep
)
326 while (cp
< ep
&& *cp
!= sep
)
329 if (!is_absolute_path(last
) && depth
) {
330 error("%s: ignoring relative alternate object store %s",
331 relative_base
, last
);
333 link_alt_odb_entry(last
, cp
- last
,
334 relative_base
, depth
);
337 while (cp
< ep
&& *cp
== sep
)
343 static void read_info_alternates(const char * relative_base
, int depth
)
348 const char alt_file_name
[] = "info/alternates";
349 /* Given that relative_base is no longer than PATH_MAX,
350 ensure that "path" has enough space to append "/", the
351 file name, "info/alternates", and a trailing NUL. */
352 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
355 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
356 fd
= open(path
, O_RDONLY
);
359 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
363 mapsz
= xsize_t(st
.st_size
);
364 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
367 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
372 void add_to_alternates_file(const char *reference
)
374 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
375 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), 1);
376 char *alt
= mkpath("%s/objects\n", reference
);
377 write_or_die(fd
, alt
, strlen(alt
));
378 if (commit_lock_file(lock
))
379 die("could not close alternates file");
381 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
384 void prepare_alt_odb(void)
391 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
394 alt_odb_tail
= &alt_odb_list
;
395 link_alt_odb_entries(alt
, alt
+ strlen(alt
), ':', NULL
, 0);
397 read_info_alternates(get_object_directory(), 0);
400 static int has_loose_object(const unsigned char *sha1
)
402 char *name
= sha1_file_name(sha1
);
403 struct alternate_object_database
*alt
;
405 if (!access(name
, F_OK
))
408 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
410 fill_sha1_path(name
, sha1
);
411 if (!access(alt
->base
, F_OK
))
417 static unsigned int pack_used_ctr
;
418 static unsigned int pack_mmap_calls
;
419 static unsigned int peak_pack_open_windows
;
420 static unsigned int pack_open_windows
;
421 static size_t peak_pack_mapped
;
422 static size_t pack_mapped
;
423 struct packed_git
*packed_git
;
425 void pack_report(void)
428 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
429 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
430 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
431 sz_fmt(getpagesize()),
432 sz_fmt(packed_git_window_size
),
433 sz_fmt(packed_git_limit
));
435 "pack_report: pack_used_ctr = %10u\n"
436 "pack_report: pack_mmap_calls = %10u\n"
437 "pack_report: pack_open_windows = %10u / %10u\n"
438 "pack_report: pack_mapped = "
439 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
442 pack_open_windows
, peak_pack_open_windows
,
443 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
446 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
449 struct pack_idx_header
*hdr
;
451 uint32_t version
, nr
, i
, *index
;
452 int fd
= open(path
, O_RDONLY
);
457 if (fstat(fd
, &st
)) {
461 idx_size
= xsize_t(st
.st_size
);
462 if (idx_size
< 4 * 256 + 20 + 20) {
464 return error("index file %s is too small", path
);
466 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
470 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
471 version
= ntohl(hdr
->idx_version
);
472 if (version
< 2 || version
> 2) {
473 munmap(idx_map
, idx_size
);
474 return error("index file %s is version %d"
475 " and is not supported by this binary"
476 " (try upgrading GIT to a newer version)",
485 index
+= 2; /* skip index header */
486 for (i
= 0; i
< 256; i
++) {
487 uint32_t n
= ntohl(index
[i
]);
489 munmap(idx_map
, idx_size
);
490 return error("non-monotonic index %s", path
);
498 * - 256 index entries 4 bytes each
499 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
500 * - 20-byte SHA1 of the packfile
501 * - 20-byte SHA1 file checksum
503 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
504 munmap(idx_map
, idx_size
);
505 return error("wrong index v1 file size in %s", path
);
507 } else if (version
== 2) {
510 * - 8 bytes of header
511 * - 256 index entries 4 bytes each
512 * - 20-byte sha1 entry * nr
513 * - 4-byte crc entry * nr
514 * - 4-byte offset entry * nr
515 * - 20-byte SHA1 of the packfile
516 * - 20-byte SHA1 file checksum
517 * And after the 4-byte offset table might be a
518 * variable sized table containing 8-byte entries
519 * for offsets larger than 2^31.
521 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
522 unsigned long max_size
= min_size
;
524 max_size
+= (nr
- 1)*8;
525 if (idx_size
< min_size
|| idx_size
> max_size
) {
526 munmap(idx_map
, idx_size
);
527 return error("wrong index v2 file size in %s", path
);
529 if (idx_size
!= min_size
&&
531 * make sure we can deal with large pack offsets.
532 * 31-bit signed offset won't be enough, neither
533 * 32-bit unsigned one will be.
535 (sizeof(off_t
) <= 4)) {
536 munmap(idx_map
, idx_size
);
537 return error("pack too large for current definition of off_t in %s", path
);
541 p
->index_version
= version
;
542 p
->index_data
= idx_map
;
543 p
->index_size
= idx_size
;
548 int open_pack_index(struct packed_git
*p
)
556 idx_name
= xstrdup(p
->pack_name
);
557 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
558 ret
= check_packed_git_idx(idx_name
, p
);
563 static void scan_windows(struct packed_git
*p
,
564 struct packed_git
**lru_p
,
565 struct pack_window
**lru_w
,
566 struct pack_window
**lru_l
)
568 struct pack_window
*w
, *w_l
;
570 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
572 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
582 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
584 struct packed_git
*p
, *lru_p
= NULL
;
585 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
588 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
589 for (p
= packed_git
; p
; p
= p
->next
)
590 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
592 munmap(lru_w
->base
, lru_w
->len
);
593 pack_mapped
-= lru_w
->len
;
595 lru_l
->next
= lru_w
->next
;
597 lru_p
->windows
= lru_w
->next
;
598 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
599 close(lru_p
->pack_fd
);
610 void release_pack_memory(size_t need
, int fd
)
612 size_t cur
= pack_mapped
;
613 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
617 void close_pack_windows(struct packed_git
*p
)
620 struct pack_window
*w
= p
->windows
;
623 die("pack '%s' still has open windows to it",
625 munmap(w
->base
, w
->len
);
626 pack_mapped
-= w
->len
;
628 p
->windows
= w
->next
;
633 void unuse_pack(struct pack_window
**w_cursor
)
635 struct pack_window
*w
= *w_cursor
;
643 * Do not call this directly as this leaks p->pack_fd on error return;
644 * call open_packed_git() instead.
646 static int open_packed_git_1(struct packed_git
*p
)
649 struct pack_header hdr
;
650 unsigned char sha1
[20];
651 unsigned char *idx_sha1
;
654 if (!p
->index_data
&& open_pack_index(p
))
655 return error("packfile %s index unavailable", p
->pack_name
);
657 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
658 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
661 /* If we created the struct before we had the pack we lack size. */
663 if (!S_ISREG(st
.st_mode
))
664 return error("packfile %s not a regular file", p
->pack_name
);
665 p
->pack_size
= st
.st_size
;
666 } else if (p
->pack_size
!= st
.st_size
)
667 return error("packfile %s size changed", p
->pack_name
);
669 /* We leave these file descriptors open with sliding mmap;
670 * there is no point keeping them open across exec(), though.
672 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
674 return error("cannot determine file descriptor flags");
675 fd_flag
|= FD_CLOEXEC
;
676 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
677 return error("cannot set FD_CLOEXEC");
679 /* Verify we recognize this pack file format. */
680 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
681 return error("file %s is far too short to be a packfile", p
->pack_name
);
682 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
683 return error("file %s is not a GIT packfile", p
->pack_name
);
684 if (!pack_version_ok(hdr
.hdr_version
))
685 return error("packfile %s is version %u and not supported"
686 " (try upgrading GIT to a newer version)",
687 p
->pack_name
, ntohl(hdr
.hdr_version
));
689 /* Verify the pack matches its index. */
690 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
691 return error("packfile %s claims to have %u objects"
692 " while index indicates %u objects",
693 p
->pack_name
, ntohl(hdr
.hdr_entries
),
695 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
696 return error("end of packfile %s is unavailable", p
->pack_name
);
697 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
698 return error("packfile %s signature is unavailable", p
->pack_name
);
699 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
700 if (hashcmp(sha1
, idx_sha1
))
701 return error("packfile %s does not match index", p
->pack_name
);
705 static int open_packed_git(struct packed_git
*p
)
707 if (!open_packed_git_1(p
))
709 if (p
->pack_fd
!= -1) {
716 static int in_window(struct pack_window
*win
, off_t offset
)
718 /* We must promise at least 20 bytes (one hash) after the
719 * offset is available from this window, otherwise the offset
720 * is not actually in this window and a different window (which
721 * has that one hash excess) must be used. This is to support
722 * the object header and delta base parsing routines below.
724 off_t win_off
= win
->offset
;
725 return win_off
<= offset
726 && (offset
+ 20) <= (win_off
+ win
->len
);
729 unsigned char* use_pack(struct packed_git
*p
,
730 struct pack_window
**w_cursor
,
734 struct pack_window
*win
= *w_cursor
;
736 if (p
->pack_fd
== -1 && open_packed_git(p
))
737 die("packfile %s cannot be accessed", p
->pack_name
);
739 /* Since packfiles end in a hash of their content and its
740 * pointless to ask for an offset into the middle of that
741 * hash, and the in_window function above wouldn't match
742 * don't allow an offset too close to the end of the file.
744 if (offset
> (p
->pack_size
- 20))
745 die("offset beyond end of packfile (truncated pack?)");
747 if (!win
|| !in_window(win
, offset
)) {
750 for (win
= p
->windows
; win
; win
= win
->next
) {
751 if (in_window(win
, offset
))
755 size_t window_align
= packed_git_window_size
/ 2;
757 win
= xcalloc(1, sizeof(*win
));
758 win
->offset
= (offset
/ window_align
) * window_align
;
759 len
= p
->pack_size
- win
->offset
;
760 if (len
> packed_git_window_size
)
761 len
= packed_git_window_size
;
762 win
->len
= (size_t)len
;
763 pack_mapped
+= win
->len
;
764 while (packed_git_limit
< pack_mapped
765 && unuse_one_window(p
, p
->pack_fd
))
767 win
->base
= xmmap(NULL
, win
->len
,
768 PROT_READ
, MAP_PRIVATE
,
769 p
->pack_fd
, win
->offset
);
770 if (win
->base
== MAP_FAILED
)
771 die("packfile %s cannot be mapped: %s",
776 if (pack_mapped
> peak_pack_mapped
)
777 peak_pack_mapped
= pack_mapped
;
778 if (pack_open_windows
> peak_pack_open_windows
)
779 peak_pack_open_windows
= pack_open_windows
;
780 win
->next
= p
->windows
;
784 if (win
!= *w_cursor
) {
785 win
->last_used
= pack_used_ctr
++;
789 offset
-= win
->offset
;
791 *left
= win
->len
- xsize_t(offset
);
792 return win
->base
+ offset
;
795 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
798 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
801 * Make sure a corresponding .pack file exists and that
802 * the index looks sane.
804 path_len
-= strlen(".idx");
807 memcpy(p
->pack_name
, path
, path_len
);
808 strcpy(p
->pack_name
+ path_len
, ".pack");
809 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
814 /* ok, it looks sane as far as we can check without
815 * actually mapping the pack file.
817 p
->index_version
= 0;
818 p
->index_data
= NULL
;
821 p
->pack_size
= st
.st_size
;
825 p
->pack_local
= local
;
826 p
->mtime
= st
.st_mtime
;
827 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
832 struct packed_git
*parse_pack_index(unsigned char *sha1
)
834 const char *idx_path
= sha1_pack_index_name(sha1
);
835 const char *path
= sha1_pack_name(sha1
);
836 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
838 if (check_packed_git_idx(idx_path
, p
)) {
843 strcpy(p
->pack_name
, path
);
848 hashcpy(p
->sha1
, sha1
);
852 void install_packed_git(struct packed_git
*pack
)
854 pack
->next
= packed_git
;
858 static void prepare_packed_git_one(char *objdir
, int local
)
860 /* Ensure that this buffer is large enough so that we can
861 append "/pack/" without clobbering the stack even if
862 strlen(objdir) were PATH_MAX. */
863 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
868 sprintf(path
, "%s/pack", objdir
);
873 error("unable to open object pack directory: %s: %s",
874 path
, strerror(errno
));
878 while ((de
= readdir(dir
)) != NULL
) {
879 int namelen
= strlen(de
->d_name
);
880 struct packed_git
*p
;
882 if (!has_extension(de
->d_name
, ".idx"))
885 if (len
+ namelen
+ 1 > sizeof(path
))
888 /* Don't reopen a pack we already have. */
889 strcpy(path
+ len
, de
->d_name
);
890 for (p
= packed_git
; p
; p
= p
->next
) {
891 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
896 /* See if it really is a valid .idx file with corresponding
897 * .pack file that we can map.
899 p
= add_packed_git(path
, len
+ namelen
, local
);
902 install_packed_git(p
);
907 static int sort_pack(const void *a_
, const void *b_
)
909 struct packed_git
*a
= *((struct packed_git
**)a_
);
910 struct packed_git
*b
= *((struct packed_git
**)b_
);
914 * Local packs tend to contain objects specific to our
915 * variant of the project than remote ones. In addition,
916 * remote ones could be on a network mounted filesystem.
917 * Favor local ones for these reasons.
919 st
= a
->pack_local
- b
->pack_local
;
924 * Younger packs tend to contain more recent objects,
925 * and more recent objects tend to get accessed more
928 if (a
->mtime
< b
->mtime
)
930 else if (a
->mtime
== b
->mtime
)
935 static void rearrange_packed_git(void)
937 struct packed_git
**ary
, *p
;
940 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
945 /* prepare an array of packed_git for easier sorting */
946 ary
= xcalloc(n
, sizeof(struct packed_git
*));
947 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
950 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
952 /* link them back again */
953 for (i
= 0; i
< n
- 1; i
++)
954 ary
[i
]->next
= ary
[i
+ 1];
955 ary
[n
- 1]->next
= NULL
;
961 static int prepare_packed_git_run_once
= 0;
962 void prepare_packed_git(void)
964 struct alternate_object_database
*alt
;
966 if (prepare_packed_git_run_once
)
968 prepare_packed_git_one(get_object_directory(), 1);
970 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
972 prepare_packed_git_one(alt
->base
, 0);
975 rearrange_packed_git();
976 prepare_packed_git_run_once
= 1;
979 void reprepare_packed_git(void)
981 prepare_packed_git_run_once
= 0;
982 prepare_packed_git();
985 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
987 unsigned char real_sha1
[20];
988 hash_sha1_file(map
, size
, type
, real_sha1
);
989 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
992 static int git_open_noatime(const char *name
)
994 static int sha1_file_open_flag
= O_NOATIME
;
995 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
997 /* Might the failure be due to O_NOATIME? */
998 if (fd
< 0 && errno
!= ENOENT
&& sha1_file_open_flag
) {
999 fd
= open(name
, O_RDONLY
);
1001 sha1_file_open_flag
= 0;
1006 static int open_sha1_file(const unsigned char *sha1
)
1009 char *name
= sha1_file_name(sha1
);
1010 struct alternate_object_database
*alt
;
1012 fd
= git_open_noatime(name
);
1018 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1020 fill_sha1_path(name
, sha1
);
1021 fd
= git_open_noatime(alt
->base
);
1028 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1033 fd
= open_sha1_file(sha1
);
1038 if (!fstat(fd
, &st
)) {
1039 *size
= xsize_t(st
.st_size
);
1040 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1047 static int legacy_loose_object(unsigned char *map
)
1052 * Is it a zlib-compressed buffer? If so, the first byte
1053 * must be 0x78 (15-bit window size, deflated), and the
1054 * first 16-bit word is evenly divisible by 31
1056 word
= (map
[0] << 8) + map
[1];
1057 if (map
[0] == 0x78 && !(word
% 31))
1063 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1068 unsigned long used
= 0;
1071 *type
= (c
>> 4) & 7;
1077 if (sizeof(long) * 8 <= shift
)
1080 size
+= (c
& 0x7f) << shift
;
1087 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1089 unsigned long size
, used
;
1090 static const char valid_loose_object_type
[8] = {
1092 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1093 0, /* "delta" and others are invalid in a loose object */
1095 enum object_type type
;
1097 /* Get the data stream */
1098 memset(stream
, 0, sizeof(*stream
));
1099 stream
->next_in
= map
;
1100 stream
->avail_in
= mapsize
;
1101 stream
->next_out
= buffer
;
1102 stream
->avail_out
= bufsiz
;
1104 if (legacy_loose_object(map
)) {
1105 inflateInit(stream
);
1106 return inflate(stream
, 0);
1111 * There used to be a second loose object header format which
1112 * was meant to mimic the in-pack format, allowing for direct
1113 * copy of the object data. This format turned up not to be
1114 * really worth it and we don't write it any longer. But we
1115 * can still read it.
1117 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1118 if (!used
|| !valid_loose_object_type
[type
])
1123 /* Set up the stream for the rest.. */
1124 stream
->next_in
= map
;
1125 stream
->avail_in
= mapsize
;
1126 inflateInit(stream
);
1128 /* And generate the fake traditional header */
1129 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1130 typename(type
), size
);
1134 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1136 int bytes
= strlen(buffer
) + 1;
1137 unsigned char *buf
= xmalloc(1+size
);
1141 n
= stream
->total_out
- bytes
;
1144 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1146 if (bytes
<= size
) {
1148 * The above condition must be (bytes <= size), not
1149 * (bytes < size). In other words, even though we
1150 * expect no more output and set avail_out to zer0,
1151 * the input zlib stream may have bytes that express
1152 * "this concludes the stream", and we *do* want to
1155 * Otherwise we would not be able to test that we
1156 * consumed all the input to reach the expected size;
1157 * we also want to check that zlib tells us that all
1158 * went well with status == Z_STREAM_END at the end.
1160 stream
->next_out
= buf
+ bytes
;
1161 stream
->avail_out
= size
- bytes
;
1162 while (status
== Z_OK
)
1163 status
= inflate(stream
, Z_FINISH
);
1166 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1172 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1173 else if (stream
->avail_in
)
1174 error("garbage at end of loose object '%s'",
1181 * We used to just use "sscanf()", but that's actually way
1182 * too permissive for what we want to check. So do an anal
1183 * object header parse by hand.
1185 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1192 * The type can be at most ten bytes (including the
1193 * terminating '\0' that we add), and is followed by
1202 if (i
>= sizeof(type
))
1208 * The length must follow immediately, and be in canonical
1209 * decimal format (ie "010" is not valid).
1211 size
= *hdr
++ - '0';
1216 unsigned long c
= *hdr
- '0';
1220 size
= size
* 10 + c
;
1226 * The length must be followed by a zero byte
1228 return *hdr
? -1 : type_from_string(type
);
1231 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1237 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1238 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1241 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1244 unsigned long get_size_from_delta(struct packed_git
*p
,
1245 struct pack_window
**w_curs
,
1248 const unsigned char *data
;
1249 unsigned char delta_head
[20], *in
;
1253 memset(&stream
, 0, sizeof(stream
));
1254 stream
.next_out
= delta_head
;
1255 stream
.avail_out
= sizeof(delta_head
);
1257 inflateInit(&stream
);
1259 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1260 stream
.next_in
= in
;
1261 st
= inflate(&stream
, Z_FINISH
);
1262 curpos
+= stream
.next_in
- in
;
1263 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1264 stream
.total_out
< sizeof(delta_head
));
1265 inflateEnd(&stream
);
1266 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1267 die("delta data unpack-initial failed");
1269 /* Examine the initial part of the delta to figure out
1274 /* ignore base size */
1275 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1277 /* Read the result size */
1278 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1281 static off_t
get_delta_base(struct packed_git
*p
,
1282 struct pack_window
**w_curs
,
1284 enum object_type type
,
1285 off_t delta_obj_offset
)
1287 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1290 /* use_pack() assured us we have [base_info, base_info + 20)
1291 * as a range that we can look at without walking off the
1292 * end of the mapped window. Its actually the hash size
1293 * that is assured. An OFS_DELTA longer than the hash size
1294 * is stupid, as then a REF_DELTA would be smaller to store.
1296 if (type
== OBJ_OFS_DELTA
) {
1298 unsigned char c
= base_info
[used
++];
1299 base_offset
= c
& 127;
1302 if (!base_offset
|| MSB(base_offset
, 7))
1303 die("offset value overflow for delta base object");
1304 c
= base_info
[used
++];
1305 base_offset
= (base_offset
<< 7) + (c
& 127);
1307 base_offset
= delta_obj_offset
- base_offset
;
1308 if (base_offset
>= delta_obj_offset
)
1309 die("delta base offset out of bound");
1311 } else if (type
== OBJ_REF_DELTA
) {
1312 /* The base entry _must_ be in the same pack */
1313 base_offset
= find_pack_entry_one(base_info
, p
);
1315 die("failed to find delta-pack base object %s",
1316 sha1_to_hex(base_info
));
1319 die("I am totally screwed");
1323 /* forward declaration for a mutually recursive function */
1324 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1325 unsigned long *sizep
);
1327 static int packed_delta_info(struct packed_git
*p
,
1328 struct pack_window
**w_curs
,
1330 enum object_type type
,
1332 unsigned long *sizep
)
1336 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1337 type
= packed_object_info(p
, base_offset
, NULL
);
1339 /* We choose to only get the type of the base object and
1340 * ignore potentially corrupt pack file that expects the delta
1341 * based on a base with a wrong size. This saves tons of
1345 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1350 static int unpack_object_header(struct packed_git
*p
,
1351 struct pack_window
**w_curs
,
1353 unsigned long *sizep
)
1355 unsigned char *base
;
1358 enum object_type type
;
1360 /* use_pack() assures us we have [base, base + 20) available
1361 * as a range that we can look at at. (Its actually the hash
1362 * size that is assured.) With our object header encoding
1363 * the maximum deflated object size is 2^137, which is just
1364 * insane, so we know won't exceed what we have been given.
1366 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1367 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1369 die("object offset outside of pack file");
1375 const char *packed_object_info_detail(struct packed_git
*p
,
1377 unsigned long *size
,
1378 unsigned long *store_size
,
1379 unsigned int *delta_chain_length
,
1380 unsigned char *base_sha1
)
1382 struct pack_window
*w_curs
= NULL
;
1384 unsigned long dummy
;
1385 unsigned char *next_sha1
;
1386 enum object_type type
;
1387 struct revindex_entry
*revidx
;
1389 *delta_chain_length
= 0;
1390 curpos
= obj_offset
;
1391 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1393 revidx
= find_pack_revindex(p
, obj_offset
);
1394 *store_size
= revidx
[1].offset
- obj_offset
;
1399 die("pack %s contains unknown object type %d",
1400 p
->pack_name
, type
);
1405 unuse_pack(&w_curs
);
1406 return typename(type
);
1408 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1409 if (*delta_chain_length
== 0) {
1410 revidx
= find_pack_revindex(p
, obj_offset
);
1411 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1415 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1416 if (*delta_chain_length
== 0)
1417 hashcpy(base_sha1
, next_sha1
);
1418 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1421 (*delta_chain_length
)++;
1422 curpos
= obj_offset
;
1423 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1427 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1428 unsigned long *sizep
)
1430 struct pack_window
*w_curs
= NULL
;
1432 off_t curpos
= obj_offset
;
1433 enum object_type type
;
1435 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1440 type
= packed_delta_info(p
, &w_curs
, curpos
,
1441 type
, obj_offset
, sizep
);
1451 die("pack %s contains unknown object type %d",
1452 p
->pack_name
, type
);
1454 unuse_pack(&w_curs
);
1458 static void *unpack_compressed_entry(struct packed_git
*p
,
1459 struct pack_window
**w_curs
,
1465 unsigned char *buffer
, *in
;
1467 buffer
= xmalloc(size
+ 1);
1469 memset(&stream
, 0, sizeof(stream
));
1470 stream
.next_out
= buffer
;
1471 stream
.avail_out
= size
;
1473 inflateInit(&stream
);
1475 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1476 stream
.next_in
= in
;
1477 st
= inflate(&stream
, Z_FINISH
);
1478 curpos
+= stream
.next_in
- in
;
1479 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1480 inflateEnd(&stream
);
1481 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1489 #define MAX_DELTA_CACHE (256)
1491 static size_t delta_base_cached
;
1493 static struct delta_base_cache_lru_list
{
1494 struct delta_base_cache_lru_list
*prev
;
1495 struct delta_base_cache_lru_list
*next
;
1496 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1498 static struct delta_base_cache_entry
{
1499 struct delta_base_cache_lru_list lru
;
1501 struct packed_git
*p
;
1504 enum object_type type
;
1505 } delta_base_cache
[MAX_DELTA_CACHE
];
1507 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1511 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1512 hash
+= (hash
>> 8) + (hash
>> 16);
1513 return hash
% MAX_DELTA_CACHE
;
1516 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1517 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1520 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1521 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1524 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1525 goto found_cache_entry
;
1526 return unpack_entry(p
, base_offset
, type
, base_size
);
1531 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1532 ent
->lru
.prev
->next
= ent
->lru
.next
;
1533 delta_base_cached
-= ent
->size
;
1535 ret
= xmemdupz(ent
->data
, ent
->size
);
1538 *base_size
= ent
->size
;
1542 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1547 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1548 ent
->lru
.prev
->next
= ent
->lru
.next
;
1549 delta_base_cached
-= ent
->size
;
1553 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1554 void *base
, unsigned long base_size
, enum object_type type
)
1556 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1557 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1558 struct delta_base_cache_lru_list
*lru
;
1560 release_delta_base_cache(ent
);
1561 delta_base_cached
+= base_size
;
1563 for (lru
= delta_base_cache_lru
.next
;
1564 delta_base_cached
> delta_base_cache_limit
1565 && lru
!= &delta_base_cache_lru
;
1567 struct delta_base_cache_entry
*f
= (void *)lru
;
1568 if (f
->type
== OBJ_BLOB
)
1569 release_delta_base_cache(f
);
1571 for (lru
= delta_base_cache_lru
.next
;
1572 delta_base_cached
> delta_base_cache_limit
1573 && lru
!= &delta_base_cache_lru
;
1575 struct delta_base_cache_entry
*f
= (void *)lru
;
1576 release_delta_base_cache(f
);
1580 ent
->base_offset
= base_offset
;
1583 ent
->size
= base_size
;
1584 ent
->lru
.next
= &delta_base_cache_lru
;
1585 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1586 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1587 delta_base_cache_lru
.prev
= &ent
->lru
;
1590 static void *unpack_delta_entry(struct packed_git
*p
,
1591 struct pack_window
**w_curs
,
1593 unsigned long delta_size
,
1595 enum object_type
*type
,
1596 unsigned long *sizep
)
1598 void *delta_data
, *result
, *base
;
1599 unsigned long base_size
;
1602 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1603 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1605 die("failed to read delta base object"
1606 " at %"PRIuMAX
" from %s",
1607 (uintmax_t)base_offset
, p
->pack_name
);
1609 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1611 die("failed to unpack compressed delta"
1612 " at %"PRIuMAX
" from %s",
1613 (uintmax_t)curpos
, p
->pack_name
);
1614 result
= patch_delta(base
, base_size
,
1615 delta_data
, delta_size
,
1618 die("failed to apply delta");
1620 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1624 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1625 enum object_type
*type
, unsigned long *sizep
)
1627 struct pack_window
*w_curs
= NULL
;
1628 off_t curpos
= obj_offset
;
1631 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1635 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1636 obj_offset
, type
, sizep
);
1642 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1645 die("unknown object type %i in %s", *type
, p
->pack_name
);
1647 unuse_pack(&w_curs
);
1651 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1654 const unsigned char *index
= p
->index_data
;
1656 if (open_pack_index(p
))
1658 index
= p
->index_data
;
1660 if (n
>= p
->num_objects
)
1663 if (p
->index_version
== 1) {
1664 return index
+ 24 * n
+ 4;
1667 return index
+ 20 * n
;
1671 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1673 const unsigned char *index
= p
->index_data
;
1675 if (p
->index_version
== 1) {
1676 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1679 index
+= 8 + p
->num_objects
* (20 + 4);
1680 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1681 if (!(off
& 0x80000000))
1683 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1684 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1685 ntohl(*((uint32_t *)(index
+ 4)));
1689 off_t
find_pack_entry_one(const unsigned char *sha1
,
1690 struct packed_git
*p
)
1692 const uint32_t *level1_ofs
= p
->index_data
;
1693 const unsigned char *index
= p
->index_data
;
1694 unsigned hi
, lo
, stride
;
1695 static int use_lookup
= -1;
1696 static int debug_lookup
= -1;
1698 if (debug_lookup
< 0)
1699 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1702 if (open_pack_index(p
))
1704 level1_ofs
= p
->index_data
;
1705 index
= p
->index_data
;
1707 if (p
->index_version
> 1) {
1712 hi
= ntohl(level1_ofs
[*sha1
]);
1713 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1714 if (p
->index_version
> 1) {
1722 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1723 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1726 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1728 int pos
= sha1_entry_pos(index
, stride
, 0,
1729 lo
, hi
, p
->num_objects
, sha1
);
1732 return nth_packed_object_offset(p
, pos
);
1736 unsigned mi
= (lo
+ hi
) / 2;
1737 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1740 printf("lo %u hi %u rg %u mi %u\n",
1741 lo
, hi
, hi
- lo
, mi
);
1743 return nth_packed_object_offset(p
, mi
);
1752 int matches_pack_name(struct packed_git
*p
, const char *name
)
1754 const char *last_c
, *c
;
1756 if (!strcmp(p
->pack_name
, name
))
1759 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1764 if (!strcmp(last_c
, name
))
1770 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1772 static struct packed_git
*last_found
= (void *)1;
1773 struct packed_git
*p
;
1776 prepare_packed_git();
1779 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1782 if (ignore_packed
) {
1784 for (ig
= ignore_packed
; *ig
; ig
++)
1785 if (matches_pack_name(p
, *ig
))
1791 offset
= find_pack_entry_one(sha1
, p
);
1794 * We are about to tell the caller where they can
1795 * locate the requested object. We better make
1796 * sure the packfile is still here and can be
1797 * accessed before supplying that answer, as
1798 * it may have been deleted since the index
1801 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1802 error("packfile %s cannot be accessed", p
->pack_name
);
1807 hashcpy(e
->sha1
, sha1
);
1813 if (p
== last_found
)
1817 if (p
== last_found
)
1823 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1824 struct packed_git
*packs
)
1826 struct packed_git
*p
;
1828 for (p
= packs
; p
; p
= p
->next
) {
1829 if (find_pack_entry_one(sha1
, p
))
1836 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1839 unsigned long mapsize
, size
;
1844 map
= map_sha1_file(sha1
, &mapsize
);
1846 return error("unable to find %s", sha1_to_hex(sha1
));
1847 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1848 status
= error("unable to unpack %s header",
1850 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1851 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1854 inflateEnd(&stream
);
1855 munmap(map
, mapsize
);
1859 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1861 struct pack_entry e
;
1863 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1864 reprepare_packed_git();
1865 if (!find_pack_entry(sha1
, &e
, NULL
))
1866 return sha1_loose_object_info(sha1
, sizep
);
1868 return packed_object_info(e
.p
, e
.offset
, sizep
);
1871 static void *read_packed_sha1(const unsigned char *sha1
,
1872 enum object_type
*type
, unsigned long *size
)
1874 struct pack_entry e
;
1876 if (!find_pack_entry(sha1
, &e
, NULL
))
1879 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1883 * This is meant to hold a *small* number of objects that you would
1884 * want read_sha1_file() to be able to return, but yet you do not want
1885 * to write them into the object store (e.g. a browse-only
1888 static struct cached_object
{
1889 unsigned char sha1
[20];
1890 enum object_type type
;
1894 static int cached_object_nr
, cached_object_alloc
;
1896 static struct cached_object empty_tree
= {
1897 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1898 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1899 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1905 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1908 struct cached_object
*co
= cached_objects
;
1910 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1911 if (!hashcmp(co
->sha1
, sha1
))
1914 if (!hashcmp(sha1
, empty_tree
.sha1
))
1919 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1920 unsigned char *sha1
)
1922 struct cached_object
*co
;
1924 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1925 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1927 if (cached_object_alloc
<= cached_object_nr
) {
1928 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1929 cached_objects
= xrealloc(cached_objects
,
1930 sizeof(*cached_objects
) *
1931 cached_object_alloc
);
1933 co
= &cached_objects
[cached_object_nr
++];
1936 co
->buf
= xmalloc(len
);
1937 memcpy(co
->buf
, buf
, len
);
1938 hashcpy(co
->sha1
, sha1
);
1942 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1943 unsigned long *size
)
1945 unsigned long mapsize
;
1947 struct cached_object
*co
;
1949 co
= find_cached_object(sha1
);
1953 return xmemdupz(co
->buf
, co
->size
);
1956 buf
= read_packed_sha1(sha1
, type
, size
);
1959 map
= map_sha1_file(sha1
, &mapsize
);
1961 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1962 munmap(map
, mapsize
);
1965 reprepare_packed_git();
1966 return read_packed_sha1(sha1
, type
, size
);
1969 void *read_object_with_reference(const unsigned char *sha1
,
1970 const char *required_type_name
,
1971 unsigned long *size
,
1972 unsigned char *actual_sha1_return
)
1974 enum object_type type
, required_type
;
1976 unsigned long isize
;
1977 unsigned char actual_sha1
[20];
1979 required_type
= type_from_string(required_type_name
);
1980 hashcpy(actual_sha1
, sha1
);
1982 int ref_length
= -1;
1983 const char *ref_type
= NULL
;
1985 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1988 if (type
== required_type
) {
1990 if (actual_sha1_return
)
1991 hashcpy(actual_sha1_return
, actual_sha1
);
1994 /* Handle references */
1995 else if (type
== OBJ_COMMIT
)
1997 else if (type
== OBJ_TAG
)
1998 ref_type
= "object ";
2003 ref_length
= strlen(ref_type
);
2005 if (ref_length
+ 40 > isize
||
2006 memcmp(buffer
, ref_type
, ref_length
) ||
2007 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2012 /* Now we have the ID of the referred-to object in
2013 * actual_sha1. Check again. */
2017 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2018 const char *type
, unsigned char *sha1
,
2019 char *hdr
, int *hdrlen
)
2023 /* Generate the header */
2024 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2028 SHA1_Update(&c
, hdr
, *hdrlen
);
2029 SHA1_Update(&c
, buf
, len
);
2030 SHA1_Final(sha1
, &c
);
2034 * Move the just written object into its final resting place
2036 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2038 int ret
= link(tmpfile
, filename
);
2041 * Coda hack - coda doesn't like cross-directory links,
2042 * so we fall back to a rename, which will mean that it
2043 * won't be able to check collisions, but that's not a
2046 * The same holds for FAT formatted media.
2048 * When this succeeds, we just return 0. We have nothing
2051 if (ret
&& ret
!= EEXIST
) {
2052 if (!rename(tmpfile
, filename
))
2058 if (ret
!= EEXIST
) {
2059 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2061 /* FIXME!!! Collision check here ? */
2067 static int write_buffer(int fd
, const void *buf
, size_t len
)
2069 if (write_in_full(fd
, buf
, len
) < 0)
2070 return error("file write error (%s)", strerror(errno
));
2074 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2075 unsigned char *sha1
)
2079 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2083 /* Finalize a file on disk, and close it. */
2084 static void close_sha1_file(int fd
)
2086 if (fsync_object_files
)
2087 fsync_or_die(fd
, "sha1 file");
2090 die("unable to write sha1 file");
2093 /* Size of directory component, including the ending '/' */
2094 static inline int directory_size(const char *filename
)
2096 const char *s
= strrchr(filename
, '/');
2099 return s
- filename
+ 1;
2103 * This creates a temporary file in the same directory as the final
2106 * We want to avoid cross-directory filename renames, because those
2107 * can have problems on various filesystems (FAT, NFS, Coda).
2109 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2111 int fd
, dirlen
= directory_size(filename
);
2113 if (dirlen
+ 20 > bufsiz
) {
2114 errno
= ENAMETOOLONG
;
2117 memcpy(buffer
, filename
, dirlen
);
2118 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2119 fd
= mkstemp(buffer
);
2120 if (fd
< 0 && dirlen
) {
2121 /* Make sure the directory exists */
2122 buffer
[dirlen
-1] = 0;
2123 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2127 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2128 fd
= mkstemp(buffer
);
2133 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2134 void *buf
, unsigned long len
, time_t mtime
)
2137 unsigned char *compressed
;
2140 static char tmpfile
[PATH_MAX
];
2142 filename
= sha1_file_name(sha1
);
2143 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2146 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2148 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2152 memset(&stream
, 0, sizeof(stream
));
2153 deflateInit(&stream
, zlib_compression_level
);
2154 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2155 compressed
= xmalloc(size
);
2158 stream
.next_out
= compressed
;
2159 stream
.avail_out
= size
;
2161 /* First header.. */
2162 stream
.next_in
= (unsigned char *)hdr
;
2163 stream
.avail_in
= hdrlen
;
2164 while (deflate(&stream
, 0) == Z_OK
)
2167 /* Then the data itself.. */
2168 stream
.next_in
= buf
;
2169 stream
.avail_in
= len
;
2170 ret
= deflate(&stream
, Z_FINISH
);
2171 if (ret
!= Z_STREAM_END
)
2172 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2174 ret
= deflateEnd(&stream
);
2176 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2178 size
= stream
.total_out
;
2180 if (write_buffer(fd
, compressed
, size
) < 0)
2181 die("unable to write sha1 file");
2182 close_sha1_file(fd
);
2188 utb
.modtime
= mtime
;
2189 if (utime(tmpfile
, &utb
) < 0)
2190 warning("failed utime() on %s: %s",
2191 tmpfile
, strerror(errno
));
2194 return move_temp_to_file(tmpfile
, filename
);
2197 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2199 unsigned char sha1
[20];
2203 /* Normally if we have it in the pack then we do not bother writing
2204 * it out into .git/objects/??/?{38} file.
2206 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2208 hashcpy(returnsha1
, sha1
);
2209 if (has_sha1_file(sha1
))
2211 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2214 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2218 enum object_type type
;
2222 if (has_loose_object(sha1
))
2224 buf
= read_packed_sha1(sha1
, &type
, &len
);
2226 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2227 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2228 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2231 int has_pack_index(const unsigned char *sha1
)
2234 if (stat(sha1_pack_index_name(sha1
), &st
))
2239 int has_pack_file(const unsigned char *sha1
)
2242 if (stat(sha1_pack_name(sha1
), &st
))
2247 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2249 struct pack_entry e
;
2250 return find_pack_entry(sha1
, &e
, ignore_packed
);
2253 int has_sha1_file(const unsigned char *sha1
)
2255 struct pack_entry e
;
2257 if (find_pack_entry(sha1
, &e
, NULL
))
2259 return has_loose_object(sha1
);
2262 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2267 strbuf_init(&buf
, 0);
2268 if (strbuf_read(&buf
, fd
, 4096) < 0) {
2269 strbuf_release(&buf
);
2276 ret
= write_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2278 ret
= hash_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2279 strbuf_release(&buf
);
2284 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2285 enum object_type type
, const char *path
)
2287 size_t size
= xsize_t(st
->st_size
);
2289 int ret
, re_allocated
= 0;
2292 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2299 * Convert blobs to git internal format
2301 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2303 strbuf_init(&nbuf
, 0);
2304 if (convert_to_git(path
, buf
, size
, &nbuf
,
2305 write_object
? safe_crlf
: 0)) {
2307 buf
= strbuf_detach(&nbuf
, &size
);
2313 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2315 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2325 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2331 switch (st
->st_mode
& S_IFMT
) {
2333 fd
= open(path
, O_RDONLY
);
2335 return error("open(\"%s\"): %s", path
,
2337 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2338 return error("%s: failed to insert into database",
2342 len
= xsize_t(st
->st_size
);
2343 target
= xmalloc(len
+ 1);
2344 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2345 char *errstr
= strerror(errno
);
2347 return error("readlink(\"%s\"): %s", path
,
2351 hash_sha1_file(target
, len
, blob_type
, sha1
);
2352 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2353 return error("%s: failed to insert into database",
2358 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2360 return error("%s: unsupported file type", path
);
2365 int read_pack_header(int fd
, struct pack_header
*header
)
2367 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2368 /* "eof before pack header was fully read" */
2369 return PH_ERROR_EOF
;
2371 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2372 /* "protocol error (pack signature mismatch detected)" */
2373 return PH_ERROR_PACK_SIGNATURE
;
2374 if (!pack_version_ok(header
->hdr_version
))
2375 /* "protocol error (pack version unsupported)" */
2376 return PH_ERROR_PROTOCOL
;