2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
18 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19 #define O_NOATIME 01000000
25 const unsigned char null_sha1
[20];
27 static unsigned int sha1_file_open_flag
= O_NOATIME
;
29 static unsigned hexval(char c
)
31 if (c
>= '0' && c
<= '9')
33 if (c
>= 'a' && c
<= 'f')
35 if (c
>= 'A' && c
<= 'F')
40 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
43 for (i
= 0; i
< 20; i
++) {
44 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
53 int safe_create_leading_directories(char *path
)
62 pos
= strchr(pos
, '/');
66 if (!stat(path
, &st
)) {
68 if (!S_ISDIR(st
.st_mode
)) {
73 else if (mkdir(path
, 0777)) {
77 else if (adjust_shared_perm(path
)) {
86 char * sha1_to_hex(const unsigned char *sha1
)
89 static char hexbuffer
[4][50];
90 static const char hex
[] = "0123456789abcdef";
91 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
94 for (i
= 0; i
< 20; i
++) {
95 unsigned int val
= *sha1
++;
96 *buf
++ = hex
[val
>> 4];
97 *buf
++ = hex
[val
& 0xf];
104 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
107 for (i
= 0; i
< 20; i
++) {
108 static char hex
[] = "0123456789abcdef";
109 unsigned int val
= sha1
[i
];
110 char *pos
= pathbuf
+ i
*2 + (i
> 0);
111 *pos
++ = hex
[val
>> 4];
112 *pos
= hex
[val
& 0xf];
117 * NOTE! This returns a statically allocated buffer, so you have to be
118 * careful about using it. Do a "strdup()" if you need to save the
121 * Also note that this returns the location for creating. Reading
122 * SHA1 file can happen from any alternate directory listed in the
123 * DB_ENVIRONMENT environment variable if it is not found in
124 * the primary object database.
126 char *sha1_file_name(const unsigned char *sha1
)
128 static char *name
, *base
;
131 const char *sha1_file_directory
= get_object_directory();
132 int len
= strlen(sha1_file_directory
);
133 base
= xmalloc(len
+ 60);
134 memcpy(base
, sha1_file_directory
, len
);
135 memset(base
+len
, 0, 60);
138 name
= base
+ len
+ 1;
140 fill_sha1_path(name
, sha1
);
144 char *sha1_pack_name(const unsigned char *sha1
)
146 static const char hex
[] = "0123456789abcdef";
147 static char *name
, *base
, *buf
;
151 const char *sha1_file_directory
= get_object_directory();
152 int len
= strlen(sha1_file_directory
);
153 base
= xmalloc(len
+ 60);
154 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory
);
155 name
= base
+ len
+ 11;
160 for (i
= 0; i
< 20; i
++) {
161 unsigned int val
= *sha1
++;
162 *buf
++ = hex
[val
>> 4];
163 *buf
++ = hex
[val
& 0xf];
169 char *sha1_pack_index_name(const unsigned char *sha1
)
171 static const char hex
[] = "0123456789abcdef";
172 static char *name
, *base
, *buf
;
176 const char *sha1_file_directory
= get_object_directory();
177 int len
= strlen(sha1_file_directory
);
178 base
= xmalloc(len
+ 60);
179 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory
);
180 name
= base
+ len
+ 11;
185 for (i
= 0; i
< 20; i
++) {
186 unsigned int val
= *sha1
++;
187 *buf
++ = hex
[val
>> 4];
188 *buf
++ = hex
[val
& 0xf];
194 struct alternate_object_database
*alt_odb_list
;
195 static struct alternate_object_database
**alt_odb_tail
;
197 static void read_info_alternates(const char * alternates
, int depth
);
200 * Prepare alternate object database registry.
202 * The variable alt_odb_list points at the list of struct
203 * alternate_object_database. The elements on this list come from
204 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
205 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
206 * whose contents is similar to that environment variable but can be
207 * LF separated. Its base points at a statically allocated buffer that
208 * contains "/the/directory/corresponding/to/.git/objects/...", while
209 * its name points just after the slash at the end of ".git/objects/"
210 * in the example above, and has enough space to hold 40-byte hex
211 * SHA1, an extra slash for the first level indirection, and the
214 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
217 const char *objdir
= get_object_directory();
218 struct alternate_object_database
*ent
;
219 struct alternate_object_database
*alt
;
220 /* 43 = 40-byte + 2 '/' + terminating NUL */
222 int entlen
= pfxlen
+ 43;
225 if (*entry
!= '/' && relative_base
) {
226 /* Relative alt-odb */
228 base_len
= strlen(relative_base
) + 1;
232 ent
= xmalloc(sizeof(*ent
) + entlen
);
234 if (*entry
!= '/' && relative_base
) {
235 memcpy(ent
->base
, relative_base
, base_len
- 1);
236 ent
->base
[base_len
- 1] = '/';
237 memcpy(ent
->base
+ base_len
, entry
, len
);
240 memcpy(ent
->base
, entry
, pfxlen
);
242 ent
->name
= ent
->base
+ pfxlen
+ 1;
243 ent
->base
[pfxlen
+ 3] = '/';
244 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
246 /* Detect cases where alternate disappeared */
247 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
248 error("object directory %s does not exist; "
249 "check .git/objects/info/alternates.",
255 /* Prevent the common mistake of listing the same
256 * thing twice, or object directory itself.
258 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
259 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
264 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
269 /* add the alternate entry */
271 alt_odb_tail
= &(ent
->next
);
274 /* recursively add alternates */
275 read_info_alternates(ent
->base
, depth
+ 1);
277 ent
->base
[pfxlen
] = '/';
282 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
283 const char *relative_base
, int depth
)
285 const char *cp
, *last
;
288 error("%s: ignoring alternate object stores, nesting too deep.",
296 if (cp
< ep
&& *cp
== '#') {
297 while (cp
< ep
&& *cp
!= sep
)
302 while (cp
< ep
&& *cp
!= sep
)
305 if ((*last
!= '/') && depth
) {
306 error("%s: ignoring relative alternate object store %s",
307 relative_base
, last
);
309 link_alt_odb_entry(last
, cp
- last
,
310 relative_base
, depth
);
313 while (cp
< ep
&& *cp
== sep
)
319 static void read_info_alternates(const char * relative_base
, int depth
)
326 sprintf(path
, "%s/info/alternates", relative_base
);
327 fd
= open(path
, O_RDONLY
);
330 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
334 map
= mmap(NULL
, st
.st_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
336 if (map
== MAP_FAILED
)
339 link_alt_odb_entries(map
, map
+ st
.st_size
, '\n', relative_base
, depth
);
341 munmap(map
, st
.st_size
);
344 void prepare_alt_odb(void)
348 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
353 alt_odb_tail
= &alt_odb_list
;
354 link_alt_odb_entries(alt
, alt
+ strlen(alt
), ':', NULL
, 0);
356 read_info_alternates(get_object_directory(), 0);
359 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
361 char *name
= sha1_file_name(sha1
);
362 struct alternate_object_database
*alt
;
367 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
369 fill_sha1_path(name
, sha1
);
370 if (!stat(alt
->base
, st
))
376 #define PACK_MAX_SZ (1<<26)
377 static int pack_used_ctr
;
378 static unsigned long pack_mapped
;
379 struct packed_git
*packed_git
;
381 static int check_packed_git_idx(const char *path
, unsigned long *idx_size_
,
386 unsigned long idx_size
;
388 int fd
= open(path
, O_RDONLY
);
392 if (fstat(fd
, &st
)) {
396 idx_size
= st
.st_size
;
397 idx_map
= mmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
399 if (idx_map
== MAP_FAILED
)
404 *idx_size_
= idx_size
;
406 /* check index map */
407 if (idx_size
< 4*256 + 20 + 20)
408 return error("index file too small");
410 for (i
= 0; i
< 256; i
++) {
411 unsigned int n
= ntohl(index
[i
]);
413 return error("non-monotonic index");
419 * - 256 index entries 4 bytes each
420 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
421 * - 20-byte SHA1 of the packfile
422 * - 20-byte SHA1 file checksum
424 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20)
425 return error("wrong index file size");
430 static int unuse_one_packed_git(void)
432 struct packed_git
*p
, *lru
= NULL
;
434 for (p
= packed_git
; p
; p
= p
->next
) {
435 if (p
->pack_use_cnt
|| !p
->pack_base
)
437 if (!lru
|| p
->pack_last_used
< lru
->pack_last_used
)
442 munmap(lru
->pack_base
, lru
->pack_size
);
443 lru
->pack_base
= NULL
;
447 void unuse_packed_git(struct packed_git
*p
)
452 int use_packed_git(struct packed_git
*p
)
456 /* We created the struct before we had the pack */
457 stat(p
->pack_name
, &st
);
458 if (!S_ISREG(st
.st_mode
))
459 die("packfile %s not a regular file", p
->pack_name
);
460 p
->pack_size
= st
.st_size
;
467 pack_mapped
+= p
->pack_size
;
468 while (PACK_MAX_SZ
< pack_mapped
&& unuse_one_packed_git())
470 fd
= open(p
->pack_name
, O_RDONLY
);
472 die("packfile %s cannot be opened", p
->pack_name
);
473 if (fstat(fd
, &st
)) {
475 die("packfile %s cannot be opened", p
->pack_name
);
477 if (st
.st_size
!= p
->pack_size
)
478 die("packfile %s size mismatch.", p
->pack_name
);
479 map
= mmap(NULL
, p
->pack_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
481 if (map
== MAP_FAILED
)
482 die("packfile %s cannot be mapped.", p
->pack_name
);
485 /* Check if the pack file matches with the index file.
488 if (hashcmp((unsigned char *)(p
->index_base
) +
490 (unsigned char *)p
->pack_base
+
491 p
->pack_size
- 20)) {
492 die("packfile %s does not match index.", p
->pack_name
);
495 p
->pack_last_used
= pack_used_ctr
++;
500 struct packed_git
*add_packed_git(char *path
, int path_len
, int local
)
503 struct packed_git
*p
;
504 unsigned long idx_size
;
506 unsigned char sha1
[20];
508 if (check_packed_git_idx(path
, &idx_size
, &idx_map
))
511 /* do we have a corresponding .pack file? */
512 strcpy(path
+ path_len
- 4, ".pack");
513 if (stat(path
, &st
) || !S_ISREG(st
.st_mode
)) {
514 munmap(idx_map
, idx_size
);
517 /* ok, it looks sane as far as we can check without
518 * actually mapping the pack file.
520 p
= xmalloc(sizeof(*p
) + path_len
+ 2);
521 strcpy(p
->pack_name
, path
);
522 p
->index_size
= idx_size
;
523 p
->pack_size
= st
.st_size
;
524 p
->index_base
= idx_map
;
527 p
->pack_last_used
= 0;
529 p
->pack_local
= local
;
530 if ((path_len
> 44) && !get_sha1_hex(path
+ path_len
- 44, sha1
))
531 memcpy(p
->sha1
, sha1
, 20);
535 struct packed_git
*parse_pack_index(unsigned char *sha1
)
537 char *path
= sha1_pack_index_name(sha1
);
538 return parse_pack_index_file(sha1
, path
);
541 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
, char *idx_path
)
543 struct packed_git
*p
;
544 unsigned long idx_size
;
548 if (check_packed_git_idx(idx_path
, &idx_size
, &idx_map
))
551 path
= sha1_pack_name(sha1
);
553 p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
554 strcpy(p
->pack_name
, path
);
555 p
->index_size
= idx_size
;
557 p
->index_base
= idx_map
;
560 p
->pack_last_used
= 0;
562 memcpy(p
->sha1
, sha1
, 20);
566 void install_packed_git(struct packed_git
*pack
)
568 pack
->next
= packed_git
;
572 static void prepare_packed_git_one(char *objdir
, int local
)
579 sprintf(path
, "%s/pack", objdir
);
584 error("unable to open object pack directory: %s: %s",
585 path
, strerror(errno
));
589 while ((de
= readdir(dir
)) != NULL
) {
590 int namelen
= strlen(de
->d_name
);
591 struct packed_git
*p
;
593 if (!has_extension(de
->d_name
, ".idx"))
596 /* we have .idx. Is it a file we can map? */
597 strcpy(path
+ len
, de
->d_name
);
598 for (p
= packed_git
; p
; p
= p
->next
) {
599 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
604 p
= add_packed_git(path
, len
+ namelen
, local
);
607 p
->next
= packed_git
;
613 static int prepare_packed_git_run_once
= 0;
614 void prepare_packed_git(void)
616 struct alternate_object_database
*alt
;
618 if (prepare_packed_git_run_once
)
620 prepare_packed_git_one(get_object_directory(), 1);
622 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
624 prepare_packed_git_one(alt
->base
, 0);
627 prepare_packed_git_run_once
= 1;
630 static void reprepare_packed_git(void)
632 prepare_packed_git_run_once
= 0;
633 prepare_packed_git();
636 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
639 unsigned char real_sha1
[20];
643 SHA1_Update(&c
, header
, 1+sprintf(header
, "%s %lu", type
, size
));
644 SHA1_Update(&c
, map
, size
);
645 SHA1_Final(real_sha1
, &c
);
646 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
649 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
654 char *filename
= find_sha1_file(sha1
, &st
);
660 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
662 /* See if it works without O_NOATIME */
663 switch (sha1_file_open_flag
) {
665 fd
= open(filename
, O_RDONLY
);
673 /* If it failed once, it will probably fail again.
674 * Stop using O_NOATIME
676 sha1_file_open_flag
= 0;
678 map
= mmap(NULL
, st
.st_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
680 if (map
== MAP_FAILED
)
686 int legacy_loose_object(unsigned char *map
)
691 * Is it a zlib-compressed buffer? If so, the first byte
692 * must be 0x78 (15-bit window size, deflated), and the
693 * first 16-bit word is evenly divisible by 31
695 word
= (map
[0] << 8) + map
[1];
696 if (map
[0] == 0x78 && !(word
% 31))
702 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
707 static const char *typename
[8] = {
709 "commit", "tree", "blob", "tag",
714 /* Get the data stream */
715 memset(stream
, 0, sizeof(*stream
));
716 stream
->next_in
= map
;
717 stream
->avail_in
= mapsize
;
718 stream
->next_out
= buffer
;
719 stream
->avail_out
= bufsiz
;
721 if (legacy_loose_object(map
)) {
723 return inflate(stream
, 0);
728 type
= typename
[(c
>> 4) & 7];
735 if (bits
>= 8*sizeof(long))
738 size
+= (c
& 0x7f) << bits
;
743 /* Set up the stream for the rest.. */
744 stream
->next_in
= map
;
745 stream
->avail_in
= mapsize
;
748 /* And generate the fake traditional header */
749 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu", type
, size
);
753 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
)
755 int bytes
= strlen(buffer
) + 1;
756 unsigned char *buf
= xmalloc(1+size
);
759 n
= stream
->total_out
- bytes
;
762 memcpy(buf
, (char *) buffer
+ bytes
, n
);
765 stream
->next_out
= buf
+ bytes
;
766 stream
->avail_out
= size
- bytes
;
767 while (inflate(stream
, Z_FINISH
) == Z_OK
)
776 * We used to just use "sscanf()", but that's actually way
777 * too permissive for what we want to check. So do an anal
778 * object header parse by hand.
780 static int parse_sha1_header(char *hdr
, char *type
, unsigned long *sizep
)
786 * The type can be at most ten bytes (including the
787 * terminating '\0' that we add), and is followed by
802 * The length must follow immediately, and be in canonical
803 * decimal format (ie "010" is not valid).
810 unsigned long c
= *hdr
- '0';
814 size
= size
* 10 + c
;
820 * The length must be followed by a zero byte
822 return *hdr
? -1 : 0;
825 void * unpack_sha1_file(void *map
, unsigned long mapsize
, char *type
, unsigned long *size
)
831 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
832 if (ret
< Z_OK
|| parse_sha1_header(hdr
, type
, size
) < 0)
835 return unpack_sha1_rest(&stream
, hdr
, *size
);
838 /* forward declaration for a mutually recursive function */
839 static int packed_object_info(struct pack_entry
*entry
,
840 char *type
, unsigned long *sizep
);
842 static int packed_delta_info(unsigned char *base_sha1
,
843 unsigned long delta_size
,
846 unsigned long *sizep
,
847 struct packed_git
*p
)
849 struct pack_entry base_ent
;
852 die("truncated pack file");
854 /* The base entry _must_ be in the same pack */
855 if (!find_pack_entry_one(base_sha1
, &base_ent
, p
))
856 die("failed to find delta-pack base object %s",
857 sha1_to_hex(base_sha1
));
859 /* We choose to only get the type of the base object and
860 * ignore potentially corrupt pack file that expects the delta
861 * based on a base with a wrong size. This saves tons of
865 if (packed_object_info(&base_ent
, type
, NULL
))
866 die("cannot get info for delta-pack base");
869 const unsigned char *data
;
870 unsigned char delta_head
[64];
871 unsigned long result_size
;
875 memset(&stream
, 0, sizeof(stream
));
877 data
= stream
.next_in
= base_sha1
+ 20;
878 stream
.avail_in
= left
- 20;
879 stream
.next_out
= delta_head
;
880 stream
.avail_out
= sizeof(delta_head
);
882 inflateInit(&stream
);
883 st
= inflate(&stream
, Z_FINISH
);
885 if ((st
!= Z_STREAM_END
) &&
886 stream
.total_out
!= sizeof(delta_head
))
887 die("delta data unpack-initial failed");
889 /* Examine the initial part of the delta to figure out
894 /* ignore base size */
895 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
897 /* Read the result size */
898 result_size
= get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
899 *sizep
= result_size
;
904 static unsigned long unpack_object_header(struct packed_git
*p
, unsigned long offset
,
905 enum object_type
*type
, unsigned long *sizep
)
908 unsigned char *pack
, c
;
911 if (offset
>= p
->pack_size
)
912 die("object offset outside of pack file");
914 pack
= (unsigned char *) p
->pack_base
+ offset
;
917 *type
= (c
>> 4) & 7;
921 if (offset
>= p
->pack_size
)
922 die("object offset outside of pack file");
925 size
+= (c
& 0x7f) << shift
;
932 int check_reuse_pack_delta(struct packed_git
*p
, unsigned long offset
,
933 unsigned char *base
, unsigned long *sizep
,
934 enum object_type
*kindp
)
941 ptr
= unpack_object_header(p
, ptr
, kindp
, sizep
);
942 if (*kindp
!= OBJ_DELTA
)
944 memcpy(base
, (unsigned char *) p
->pack_base
+ ptr
, 20);
951 void packed_object_info_detail(struct pack_entry
*e
,
954 unsigned long *store_size
,
955 unsigned int *delta_chain_length
,
956 unsigned char *base_sha1
)
958 struct packed_git
*p
= e
->p
;
959 unsigned long offset
;
961 enum object_type kind
;
963 offset
= unpack_object_header(p
, e
->offset
, &kind
, size
);
964 pack
= (unsigned char *) p
->pack_base
+ offset
;
965 if (kind
!= OBJ_DELTA
)
966 *delta_chain_length
= 0;
968 unsigned int chain_length
= 0;
969 if (p
->pack_size
<= offset
+ 20)
970 die("pack file %s records an incomplete delta base",
972 memcpy(base_sha1
, pack
, 20);
974 struct pack_entry base_ent
;
977 find_pack_entry_one(pack
, &base_ent
, p
);
978 offset
= unpack_object_header(p
, base_ent
.offset
,
980 pack
= (unsigned char *) p
->pack_base
+ offset
;
982 } while (kind
== OBJ_DELTA
);
983 *delta_chain_length
= chain_length
;
987 strcpy(type
, commit_type
);
990 strcpy(type
, tree_type
);
993 strcpy(type
, blob_type
);
996 strcpy(type
, tag_type
);
999 die("corrupted pack file %s containing object of kind %d",
1000 p
->pack_name
, kind
);
1002 *store_size
= 0; /* notyet */
1005 static int packed_object_info(struct pack_entry
*entry
,
1006 char *type
, unsigned long *sizep
)
1008 struct packed_git
*p
= entry
->p
;
1009 unsigned long offset
, size
, left
;
1010 unsigned char *pack
;
1011 enum object_type kind
;
1014 if (use_packed_git(p
))
1015 die("cannot map packed file");
1017 offset
= unpack_object_header(p
, entry
->offset
, &kind
, &size
);
1018 pack
= (unsigned char *) p
->pack_base
+ offset
;
1019 left
= p
->pack_size
- offset
;
1023 retval
= packed_delta_info(pack
, size
, left
, type
, sizep
, p
);
1024 unuse_packed_git(p
);
1027 strcpy(type
, commit_type
);
1030 strcpy(type
, tree_type
);
1033 strcpy(type
, blob_type
);
1036 strcpy(type
, tag_type
);
1039 die("corrupted pack file %s containing object of kind %d",
1040 p
->pack_name
, kind
);
1044 unuse_packed_git(p
);
1048 /* forward declaration for a mutually recursive function */
1049 static void *unpack_entry(struct pack_entry
*, char *, unsigned long *);
1051 static void *unpack_delta_entry(unsigned char *base_sha1
,
1052 unsigned long delta_size
,
1055 unsigned long *sizep
,
1056 struct packed_git
*p
)
1058 struct pack_entry base_ent
;
1059 void *data
, *delta_data
, *result
, *base
;
1060 unsigned long data_size
, result_size
, base_size
;
1065 die("truncated pack file");
1067 /* The base entry _must_ be in the same pack */
1068 if (!find_pack_entry_one(base_sha1
, &base_ent
, p
))
1069 die("failed to find delta-pack base object %s",
1070 sha1_to_hex(base_sha1
));
1071 base
= unpack_entry_gently(&base_ent
, type
, &base_size
);
1073 die("failed to read delta-pack base object %s",
1074 sha1_to_hex(base_sha1
));
1076 data
= base_sha1
+ 20;
1077 data_size
= left
- 20;
1078 delta_data
= xmalloc(delta_size
);
1080 memset(&stream
, 0, sizeof(stream
));
1082 stream
.next_in
= data
;
1083 stream
.avail_in
= data_size
;
1084 stream
.next_out
= delta_data
;
1085 stream
.avail_out
= delta_size
;
1087 inflateInit(&stream
);
1088 st
= inflate(&stream
, Z_FINISH
);
1089 inflateEnd(&stream
);
1090 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= delta_size
)
1091 die("delta data unpack failed");
1093 result
= patch_delta(base
, base_size
,
1094 delta_data
, delta_size
,
1097 die("failed to apply delta");
1100 *sizep
= result_size
;
1104 static void *unpack_non_delta_entry(unsigned char *data
,
1110 unsigned char *buffer
;
1112 buffer
= xmalloc(size
+ 1);
1114 memset(&stream
, 0, sizeof(stream
));
1115 stream
.next_in
= data
;
1116 stream
.avail_in
= left
;
1117 stream
.next_out
= buffer
;
1118 stream
.avail_out
= size
;
1120 inflateInit(&stream
);
1121 st
= inflate(&stream
, Z_FINISH
);
1122 inflateEnd(&stream
);
1123 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1131 static void *unpack_entry(struct pack_entry
*entry
,
1132 char *type
, unsigned long *sizep
)
1134 struct packed_git
*p
= entry
->p
;
1137 if (use_packed_git(p
))
1138 die("cannot map packed file");
1139 retval
= unpack_entry_gently(entry
, type
, sizep
);
1140 unuse_packed_git(p
);
1142 die("corrupted pack file %s", p
->pack_name
);
1146 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
1147 void *unpack_entry_gently(struct pack_entry
*entry
,
1148 char *type
, unsigned long *sizep
)
1150 struct packed_git
*p
= entry
->p
;
1151 unsigned long offset
, size
, left
;
1152 unsigned char *pack
;
1153 enum object_type kind
;
1156 offset
= unpack_object_header(p
, entry
->offset
, &kind
, &size
);
1157 pack
= (unsigned char *) p
->pack_base
+ offset
;
1158 left
= p
->pack_size
- offset
;
1161 retval
= unpack_delta_entry(pack
, size
, left
, type
, sizep
, p
);
1164 strcpy(type
, commit_type
);
1167 strcpy(type
, tree_type
);
1170 strcpy(type
, blob_type
);
1173 strcpy(type
, tag_type
);
1179 retval
= unpack_non_delta_entry(pack
, size
, left
);
1183 int num_packed_objects(const struct packed_git
*p
)
1185 /* See check_packed_git_idx() */
1186 return (p
->index_size
- 20 - 20 - 4*256) / 24;
1189 int nth_packed_object_sha1(const struct packed_git
*p
, int n
,
1190 unsigned char* sha1
)
1192 void *index
= p
->index_base
+ 256;
1193 if (n
< 0 || num_packed_objects(p
) <= n
)
1195 memcpy(sha1
, (char *) index
+ (24 * n
) + 4, 20);
1199 int find_pack_entry_one(const unsigned char *sha1
,
1200 struct pack_entry
*e
, struct packed_git
*p
)
1202 unsigned int *level1_ofs
= p
->index_base
;
1203 int hi
= ntohl(level1_ofs
[*sha1
]);
1204 int lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1205 void *index
= p
->index_base
+ 256;
1208 int mi
= (lo
+ hi
) / 2;
1209 int cmp
= hashcmp((unsigned char *)index
+ (24 * mi
) + 4, sha1
);
1211 e
->offset
= ntohl(*((unsigned int *) ((char *) index
+ (24 * mi
))));
1212 memcpy(e
->sha1
, sha1
, 20);
1224 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1226 struct packed_git
*p
;
1227 prepare_packed_git();
1229 for (p
= packed_git
; p
; p
= p
->next
) {
1230 if (find_pack_entry_one(sha1
, e
, p
))
1236 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1237 struct packed_git
*packs
)
1239 struct packed_git
*p
;
1240 struct pack_entry e
;
1242 for (p
= packs
; p
; p
= p
->next
) {
1243 if (find_pack_entry_one(sha1
, &e
, p
))
1250 int sha1_object_info(const unsigned char *sha1
, char *type
, unsigned long *sizep
)
1253 unsigned long mapsize
, size
;
1258 map
= map_sha1_file(sha1
, &mapsize
);
1260 struct pack_entry e
;
1262 if (find_pack_entry(sha1
, &e
))
1263 return packed_object_info(&e
, type
, sizep
);
1264 reprepare_packed_git();
1265 if (find_pack_entry(sha1
, &e
))
1266 return packed_object_info(&e
, type
, sizep
);
1267 return error("unable to find %s", sha1_to_hex(sha1
));
1269 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1270 status
= error("unable to unpack %s header",
1272 if (parse_sha1_header(hdr
, type
, &size
) < 0)
1273 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1279 inflateEnd(&stream
);
1280 munmap(map
, mapsize
);
1284 static void *read_packed_sha1(const unsigned char *sha1
, char *type
, unsigned long *size
)
1286 struct pack_entry e
;
1288 if (!find_pack_entry(sha1
, &e
)) {
1289 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
1292 return unpack_entry(&e
, type
, size
);
1295 void * read_sha1_file(const unsigned char *sha1
, char *type
, unsigned long *size
)
1297 unsigned long mapsize
;
1299 struct pack_entry e
;
1301 if (find_pack_entry(sha1
, &e
))
1302 return read_packed_sha1(sha1
, type
, size
);
1303 map
= map_sha1_file(sha1
, &mapsize
);
1305 buf
= unpack_sha1_file(map
, mapsize
, type
, size
);
1306 munmap(map
, mapsize
);
1309 reprepare_packed_git();
1310 if (find_pack_entry(sha1
, &e
))
1311 return read_packed_sha1(sha1
, type
, size
);
1315 void *read_object_with_reference(const unsigned char *sha1
,
1316 const char *required_type
,
1317 unsigned long *size
,
1318 unsigned char *actual_sha1_return
)
1322 unsigned long isize
;
1323 unsigned char actual_sha1
[20];
1325 memcpy(actual_sha1
, sha1
, 20);
1327 int ref_length
= -1;
1328 const char *ref_type
= NULL
;
1330 buffer
= read_sha1_file(actual_sha1
, type
, &isize
);
1333 if (!strcmp(type
, required_type
)) {
1335 if (actual_sha1_return
)
1336 memcpy(actual_sha1_return
, actual_sha1
, 20);
1339 /* Handle references */
1340 else if (!strcmp(type
, commit_type
))
1342 else if (!strcmp(type
, tag_type
))
1343 ref_type
= "object ";
1348 ref_length
= strlen(ref_type
);
1350 if (memcmp(buffer
, ref_type
, ref_length
) ||
1351 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
1356 /* Now we have the ID of the referred-to object in
1357 * actual_sha1. Check again. */
1361 char *write_sha1_file_prepare(void *buf
,
1364 unsigned char *sha1
,
1370 /* Generate the header */
1371 *hdrlen
= sprintf((char *)hdr
, "%s %lu", type
, len
)+1;
1375 SHA1_Update(&c
, hdr
, *hdrlen
);
1376 SHA1_Update(&c
, buf
, len
);
1377 SHA1_Final(sha1
, &c
);
1379 return sha1_file_name(sha1
);
1383 * Link the tempfile to the final place, possibly creating the
1384 * last directory level as you do so.
1386 * Returns the errno on failure, 0 on success.
1388 static int link_temp_to_file(const char *tmpfile
, char *filename
)
1393 if (!link(tmpfile
, filename
))
1397 * Try to mkdir the last path component if that failed.
1399 * Re-try the "link()" regardless of whether the mkdir
1400 * succeeds, since a race might mean that somebody
1404 dir
= strrchr(filename
, '/');
1407 mkdir(filename
, 0777);
1408 if (adjust_shared_perm(filename
))
1411 if (!link(tmpfile
, filename
))
1419 * Move the just written object into its final resting place
1421 int move_temp_to_file(const char *tmpfile
, char *filename
)
1423 int ret
= link_temp_to_file(tmpfile
, filename
);
1426 * Coda hack - coda doesn't like cross-directory links,
1427 * so we fall back to a rename, which will mean that it
1428 * won't be able to check collisions, but that's not a
1431 * The same holds for FAT formatted media.
1433 * When this succeeds, we just return 0. We have nothing
1436 if (ret
&& ret
!= EEXIST
) {
1437 if (!rename(tmpfile
, filename
))
1443 if (ret
!= EEXIST
) {
1444 fprintf(stderr
, "unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
1447 /* FIXME!!! Collision check here ? */
1453 static int write_buffer(int fd
, const void *buf
, size_t len
)
1458 size
= write(fd
, buf
, len
);
1460 return error("file write: disk full");
1462 if (errno
== EINTR
|| errno
== EAGAIN
)
1464 return error("file write error (%s)", strerror(errno
));
1467 buf
= (char *) buf
+ size
;
1472 static int write_binary_header(unsigned char *hdr
, enum object_type type
, unsigned long len
)
1477 c
= (type
<< 4) | (len
& 15);
1490 static void setup_object_header(z_stream
*stream
, const char *type
, unsigned long len
)
1494 if (use_legacy_headers
) {
1495 while (deflate(stream
, 0) == Z_OK
)
1499 if (!strcmp(type
, blob_type
))
1500 obj_type
= OBJ_BLOB
;
1501 else if (!strcmp(type
, tree_type
))
1502 obj_type
= OBJ_TREE
;
1503 else if (!strcmp(type
, commit_type
))
1504 obj_type
= OBJ_COMMIT
;
1505 else if (!strcmp(type
, tag_type
))
1508 die("trying to generate bogus object of type '%s'", type
);
1509 hdr
= write_binary_header(stream
->next_out
, obj_type
, len
);
1510 stream
->total_out
= hdr
;
1511 stream
->next_out
+= hdr
;
1512 stream
->avail_out
-= hdr
;
1515 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
1518 unsigned char *compressed
;
1520 unsigned char sha1
[20];
1522 static char tmpfile
[PATH_MAX
];
1523 unsigned char hdr
[50];
1526 /* Normally if we have it in the pack then we do not bother writing
1527 * it out into .git/objects/??/?{38} file.
1529 filename
= write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
1531 memcpy(returnsha1
, sha1
, 20);
1532 if (has_sha1_file(sha1
))
1534 fd
= open(filename
, O_RDONLY
);
1537 * FIXME!!! We might do collision checking here, but we'd
1538 * need to uncompress the old file and check it. Later.
1544 if (errno
!= ENOENT
) {
1545 fprintf(stderr
, "sha1 file %s: %s\n", filename
, strerror(errno
));
1549 snprintf(tmpfile
, sizeof(tmpfile
), "%s/obj_XXXXXX", get_object_directory());
1551 fd
= mkstemp(tmpfile
);
1553 fprintf(stderr
, "unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
1558 memset(&stream
, 0, sizeof(stream
));
1559 deflateInit(&stream
, zlib_compression_level
);
1560 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
1561 compressed
= xmalloc(size
);
1564 stream
.next_out
= compressed
;
1565 stream
.avail_out
= size
;
1567 /* First header.. */
1568 stream
.next_in
= hdr
;
1569 stream
.avail_in
= hdrlen
;
1570 setup_object_header(&stream
, type
, len
);
1572 /* Then the data itself.. */
1573 stream
.next_in
= buf
;
1574 stream
.avail_in
= len
;
1575 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
1577 deflateEnd(&stream
);
1578 size
= stream
.total_out
;
1580 if (write_buffer(fd
, compressed
, size
) < 0)
1581 die("unable to write sha1 file");
1586 return move_temp_to_file(tmpfile
, filename
);
1590 * We need to unpack and recompress the object for writing
1591 * it out to a different file.
1593 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
1597 unsigned char *unpacked
;
1604 /* need to unpack and recompress it by itself */
1605 unpacked
= read_packed_sha1(sha1
, type
, &len
);
1607 hdrlen
= sprintf(hdr
, "%s %lu", type
, len
) + 1;
1610 memset(&stream
, 0, sizeof(stream
));
1611 deflateInit(&stream
, zlib_compression_level
);
1612 size
= deflateBound(&stream
, len
+ hdrlen
);
1613 buf
= xmalloc(size
);
1616 stream
.next_out
= buf
;
1617 stream
.avail_out
= size
;
1619 /* First header.. */
1620 stream
.next_in
= (void *)hdr
;
1621 stream
.avail_in
= hdrlen
;
1622 while (deflate(&stream
, 0) == Z_OK
)
1625 /* Then the data itself.. */
1626 stream
.next_in
= unpacked
;
1627 stream
.avail_in
= len
;
1628 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
1630 deflateEnd(&stream
);
1633 *objsize
= stream
.total_out
;
1637 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
1640 unsigned long objsize
;
1641 void *buf
= map_sha1_file(sha1
, &objsize
);
1644 retval
= write_buffer(fd
, buf
, objsize
);
1645 munmap(buf
, objsize
);
1649 buf
= repack_object(sha1
, &objsize
);
1650 retval
= write_buffer(fd
, buf
, objsize
);
1655 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
1656 size_t bufsize
, size_t *bufposn
)
1658 char tmpfile
[PATH_MAX
];
1661 unsigned char real_sha1
[20];
1662 unsigned char discard
[4096];
1666 snprintf(tmpfile
, sizeof(tmpfile
), "%s/obj_XXXXXX", get_object_directory());
1668 local
= mkstemp(tmpfile
);
1670 return error("Couldn't open %s for %s",
1671 tmpfile
, sha1_to_hex(sha1
));
1673 memset(&stream
, 0, sizeof(stream
));
1675 inflateInit(&stream
);
1682 stream
.avail_in
= *bufposn
;
1683 stream
.next_in
= (unsigned char *) buffer
;
1685 stream
.next_out
= discard
;
1686 stream
.avail_out
= sizeof(discard
);
1687 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
1688 SHA1_Update(&c
, discard
, sizeof(discard
) -
1690 } while (stream
.avail_in
&& ret
== Z_OK
);
1691 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
1692 die("unable to write sha1 file");
1693 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
1695 *bufposn
= stream
.avail_in
;
1699 size
= read(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
1704 return error("Connection closed?");
1705 perror("Reading from connection");
1710 inflateEnd(&stream
);
1713 SHA1_Final(real_sha1
, &c
);
1714 if (ret
!= Z_STREAM_END
) {
1716 return error("File %s corrupted", sha1_to_hex(sha1
));
1718 if (hashcmp(sha1
, real_sha1
)) {
1720 return error("File %s has bad hash", sha1_to_hex(sha1
));
1723 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
1726 int has_pack_index(const unsigned char *sha1
)
1729 if (stat(sha1_pack_index_name(sha1
), &st
))
1734 int has_pack_file(const unsigned char *sha1
)
1737 if (stat(sha1_pack_name(sha1
), &st
))
1742 int has_sha1_pack(const unsigned char *sha1
)
1744 struct pack_entry e
;
1745 return find_pack_entry(sha1
, &e
);
1748 int has_sha1_file(const unsigned char *sha1
)
1751 struct pack_entry e
;
1753 if (find_pack_entry(sha1
, &e
))
1755 return find_sha1_file(sha1
, &st
) ? 1 : 0;
1759 * reads from fd as long as possible into a supplied buffer of size bytes.
1760 * If necessary the buffer's size is increased using realloc()
1762 * returns 0 if anything went fine and -1 otherwise
1764 * NOTE: both buf and size may change, but even when -1 is returned
1765 * you still have to free() it yourself.
1767 int read_pipe(int fd
, char** return_buf
, unsigned long* return_size
)
1769 char* buf
= *return_buf
;
1770 unsigned long size
= *return_size
;
1772 unsigned long off
= 0;
1775 iret
= xread(fd
, buf
+ off
, size
- off
);
1780 buf
= realloc(buf
, size
);
1793 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
1795 unsigned long size
= 4096;
1796 char *buf
= malloc(size
);
1798 unsigned char hdr
[50];
1801 if (read_pipe(fd
, &buf
, &size
)) {
1809 ret
= write_sha1_file(buf
, size
, type
, sha1
);
1811 write_sha1_file_prepare(buf
, size
, type
, sha1
, hdr
, &hdrlen
);
1818 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
, const char *type
)
1820 unsigned long size
= st
->st_size
;
1823 unsigned char hdr
[50];
1828 buf
= mmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1830 if (buf
== MAP_FAILED
)
1836 ret
= write_sha1_file(buf
, size
, type
, sha1
);
1838 write_sha1_file_prepare(buf
, size
, type
, sha1
, hdr
, &hdrlen
);
1846 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
1851 switch (st
->st_mode
& S_IFMT
) {
1853 fd
= open(path
, O_RDONLY
);
1855 return error("open(\"%s\"): %s", path
,
1857 if (index_fd(sha1
, fd
, st
, write_object
, NULL
) < 0)
1858 return error("%s: failed to insert into database",
1862 target
= xmalloc(st
->st_size
+1);
1863 if (readlink(path
, target
, st
->st_size
+1) != st
->st_size
) {
1864 char *errstr
= strerror(errno
);
1866 return error("readlink(\"%s\"): %s", path
,
1869 if (!write_object
) {
1870 unsigned char hdr
[50];
1872 write_sha1_file_prepare(target
, st
->st_size
, blob_type
,
1873 sha1
, hdr
, &hdrlen
);
1874 } else if (write_sha1_file(target
, st
->st_size
, blob_type
, sha1
))
1875 return error("%s: failed to insert into database",
1880 return error("%s: unsupported file type", path
);