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 int git_open_noatime(const char *name
, struct packed_git
*p
);
40 int safe_create_leading_directories(char *path
)
42 char *pos
= path
+ offset_1st_component(path
);
46 pos
= strchr(pos
, '/');
54 if (!stat(path
, &st
)) {
56 if (!S_ISDIR(st
.st_mode
)) {
61 else if (mkdir(path
, 0777)) {
65 else if (adjust_shared_perm(path
)) {
74 int safe_create_leading_directories_const(const char *path
)
76 /* path points to cache entries, so xstrdup before messing with it */
77 char *buf
= xstrdup(path
);
78 int result
= safe_create_leading_directories(buf
);
83 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
86 for (i
= 0; i
< 20; i
++) {
87 static char hex
[] = "0123456789abcdef";
88 unsigned int val
= sha1
[i
];
89 char *pos
= pathbuf
+ i
*2 + (i
> 0);
90 *pos
++ = hex
[val
>> 4];
91 *pos
= hex
[val
& 0xf];
96 * NOTE! This returns a statically allocated buffer, so you have to be
97 * careful about using it. Do an "xstrdup()" if you need to save the
100 * Also note that this returns the location for creating. Reading
101 * SHA1 file can happen from any alternate directory listed in the
102 * DB_ENVIRONMENT environment variable if it is not found in
103 * the primary object database.
105 char *sha1_file_name(const unsigned char *sha1
)
107 static char buf
[PATH_MAX
];
111 objdir
= get_object_directory();
112 len
= strlen(objdir
);
114 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
115 if (len
+ 43 > PATH_MAX
)
116 die("insanely long object directory %s", objdir
);
117 memcpy(buf
, objdir
, len
);
121 fill_sha1_path(buf
+ len
+ 1, sha1
);
125 static char *sha1_get_pack_name(const unsigned char *sha1
,
126 char **name
, char **base
, const char *which
)
128 static const char hex
[] = "0123456789abcdef";
133 const char *sha1_file_directory
= get_object_directory();
134 int len
= strlen(sha1_file_directory
);
135 *base
= xmalloc(len
+ 60);
136 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
137 sha1_file_directory
, which
);
138 *name
= *base
+ len
+ 11;
143 for (i
= 0; i
< 20; i
++) {
144 unsigned int val
= *sha1
++;
145 *buf
++ = hex
[val
>> 4];
146 *buf
++ = hex
[val
& 0xf];
152 char *sha1_pack_name(const unsigned char *sha1
)
154 static char *name
, *base
;
156 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
159 char *sha1_pack_index_name(const unsigned char *sha1
)
161 static char *name
, *base
;
163 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
166 struct alternate_object_database
*alt_odb_list
;
167 static struct alternate_object_database
**alt_odb_tail
;
169 static void read_info_alternates(const char * alternates
, int depth
);
172 * Prepare alternate object database registry.
174 * The variable alt_odb_list points at the list of struct
175 * alternate_object_database. The elements on this list come from
176 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
177 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
178 * whose contents is similar to that environment variable but can be
179 * LF separated. Its base points at a statically allocated buffer that
180 * contains "/the/directory/corresponding/to/.git/objects/...", while
181 * its name points just after the slash at the end of ".git/objects/"
182 * in the example above, and has enough space to hold 40-byte hex
183 * SHA1, an extra slash for the first level indirection, and the
186 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
188 const char *objdir
= get_object_directory();
189 struct alternate_object_database
*ent
;
190 struct alternate_object_database
*alt
;
191 /* 43 = 40-byte + 2 '/' + terminating NUL */
193 int entlen
= pfxlen
+ 43;
196 if (!is_absolute_path(entry
) && relative_base
) {
197 /* Relative alt-odb */
199 base_len
= strlen(relative_base
) + 1;
203 ent
= xmalloc(sizeof(*ent
) + entlen
);
205 if (!is_absolute_path(entry
) && relative_base
) {
206 memcpy(ent
->base
, relative_base
, base_len
- 1);
207 ent
->base
[base_len
- 1] = '/';
208 memcpy(ent
->base
+ base_len
, entry
, len
);
211 memcpy(ent
->base
, entry
, pfxlen
);
213 ent
->name
= ent
->base
+ pfxlen
+ 1;
214 ent
->base
[pfxlen
+ 3] = '/';
215 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
217 /* Detect cases where alternate disappeared */
218 if (!is_directory(ent
->base
)) {
219 error("object directory %s does not exist; "
220 "check .git/objects/info/alternates.",
226 /* Prevent the common mistake of listing the same
227 * thing twice, or object directory itself.
229 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
230 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
235 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
240 /* add the alternate entry */
242 alt_odb_tail
= &(ent
->next
);
245 /* recursively add alternates */
246 read_info_alternates(ent
->base
, depth
+ 1);
248 ent
->base
[pfxlen
] = '/';
253 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
254 const char *relative_base
, int depth
)
256 const char *cp
, *last
;
259 error("%s: ignoring alternate object stores, nesting too deep.",
267 if (cp
< ep
&& *cp
== '#') {
268 while (cp
< ep
&& *cp
!= sep
)
273 while (cp
< ep
&& *cp
!= sep
)
276 if (!is_absolute_path(last
) && depth
) {
277 error("%s: ignoring relative alternate object store %s",
278 relative_base
, last
);
280 link_alt_odb_entry(last
, cp
- last
,
281 relative_base
, depth
);
284 while (cp
< ep
&& *cp
== sep
)
290 static void read_info_alternates(const char * relative_base
, int depth
)
295 const char alt_file_name
[] = "info/alternates";
296 /* Given that relative_base is no longer than PATH_MAX,
297 ensure that "path" has enough space to append "/", the
298 file name, "info/alternates", and a trailing NUL. */
299 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
302 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
303 fd
= git_open_noatime(path
, NULL
);
306 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
310 mapsz
= xsize_t(st
.st_size
);
311 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
314 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
319 void add_to_alternates_file(const char *reference
)
321 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
322 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
323 char *alt
= mkpath("%s/objects\n", reference
);
324 write_or_die(fd
, alt
, strlen(alt
));
325 if (commit_lock_file(lock
))
326 die("could not close alternates file");
328 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
331 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
333 struct alternate_object_database
*ent
;
336 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
341 void prepare_alt_odb(void)
348 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
351 alt_odb_tail
= &alt_odb_list
;
352 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
354 read_info_alternates(get_object_directory(), 0);
357 static int has_loose_object_local(const unsigned char *sha1
)
359 char *name
= sha1_file_name(sha1
);
360 return !access(name
, F_OK
);
363 int has_loose_object_nonlocal(const unsigned char *sha1
)
365 struct alternate_object_database
*alt
;
367 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
368 fill_sha1_path(alt
->name
, sha1
);
369 if (!access(alt
->base
, F_OK
))
375 static int has_loose_object(const unsigned char *sha1
)
377 return has_loose_object_local(sha1
) ||
378 has_loose_object_nonlocal(sha1
);
381 static unsigned int pack_used_ctr
;
382 static unsigned int pack_mmap_calls
;
383 static unsigned int peak_pack_open_windows
;
384 static unsigned int pack_open_windows
;
385 static size_t peak_pack_mapped
;
386 static size_t pack_mapped
;
387 struct packed_git
*packed_git
;
389 void pack_report(void)
392 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
393 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
394 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
395 sz_fmt(getpagesize()),
396 sz_fmt(packed_git_window_size
),
397 sz_fmt(packed_git_limit
));
399 "pack_report: pack_used_ctr = %10u\n"
400 "pack_report: pack_mmap_calls = %10u\n"
401 "pack_report: pack_open_windows = %10u / %10u\n"
402 "pack_report: pack_mapped = "
403 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
406 pack_open_windows
, peak_pack_open_windows
,
407 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
410 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
413 struct pack_idx_header
*hdr
;
415 uint32_t version
, nr
, i
, *index
;
416 int fd
= git_open_noatime(path
, p
);
421 if (fstat(fd
, &st
)) {
425 idx_size
= xsize_t(st
.st_size
);
426 if (idx_size
< 4 * 256 + 20 + 20) {
428 return error("index file %s is too small", path
);
430 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
434 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
435 version
= ntohl(hdr
->idx_version
);
436 if (version
< 2 || version
> 2) {
437 munmap(idx_map
, idx_size
);
438 return error("index file %s is version %"PRIu32
439 " and is not supported by this binary"
440 " (try upgrading GIT to a newer version)",
449 index
+= 2; /* skip index header */
450 for (i
= 0; i
< 256; i
++) {
451 uint32_t n
= ntohl(index
[i
]);
453 munmap(idx_map
, idx_size
);
454 return error("non-monotonic index %s", path
);
462 * - 256 index entries 4 bytes each
463 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
464 * - 20-byte SHA1 of the packfile
465 * - 20-byte SHA1 file checksum
467 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
468 munmap(idx_map
, idx_size
);
469 return error("wrong index v1 file size in %s", path
);
471 } else if (version
== 2) {
474 * - 8 bytes of header
475 * - 256 index entries 4 bytes each
476 * - 20-byte sha1 entry * nr
477 * - 4-byte crc entry * nr
478 * - 4-byte offset entry * nr
479 * - 20-byte SHA1 of the packfile
480 * - 20-byte SHA1 file checksum
481 * And after the 4-byte offset table might be a
482 * variable sized table containing 8-byte entries
483 * for offsets larger than 2^31.
485 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
486 unsigned long max_size
= min_size
;
488 max_size
+= (nr
- 1)*8;
489 if (idx_size
< min_size
|| idx_size
> max_size
) {
490 munmap(idx_map
, idx_size
);
491 return error("wrong index v2 file size in %s", path
);
493 if (idx_size
!= min_size
&&
495 * make sure we can deal with large pack offsets.
496 * 31-bit signed offset won't be enough, neither
497 * 32-bit unsigned one will be.
499 (sizeof(off_t
) <= 4)) {
500 munmap(idx_map
, idx_size
);
501 return error("pack too large for current definition of off_t in %s", path
);
505 p
->index_version
= version
;
506 p
->index_data
= idx_map
;
507 p
->index_size
= idx_size
;
512 int open_pack_index(struct packed_git
*p
)
520 idx_name
= xstrdup(p
->pack_name
);
521 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
522 ret
= check_packed_git_idx(idx_name
, p
);
527 static void scan_windows(struct packed_git
*p
,
528 struct packed_git
**lru_p
,
529 struct pack_window
**lru_w
,
530 struct pack_window
**lru_l
)
532 struct pack_window
*w
, *w_l
;
534 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
536 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
546 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
548 struct packed_git
*p
, *lru_p
= NULL
;
549 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
552 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
553 for (p
= packed_git
; p
; p
= p
->next
)
554 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
556 munmap(lru_w
->base
, lru_w
->len
);
557 pack_mapped
-= lru_w
->len
;
559 lru_l
->next
= lru_w
->next
;
561 lru_p
->windows
= lru_w
->next
;
562 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
563 close(lru_p
->pack_fd
);
574 void release_pack_memory(size_t need
, int fd
)
576 size_t cur
= pack_mapped
;
577 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
581 void close_pack_windows(struct packed_git
*p
)
584 struct pack_window
*w
= p
->windows
;
587 die("pack '%s' still has open windows to it",
589 munmap(w
->base
, w
->len
);
590 pack_mapped
-= w
->len
;
592 p
->windows
= w
->next
;
597 void unuse_pack(struct pack_window
**w_cursor
)
599 struct pack_window
*w
= *w_cursor
;
606 void close_pack_index(struct packed_git
*p
)
609 munmap((void *)p
->index_data
, p
->index_size
);
610 p
->index_data
= NULL
;
615 * This is used by git-repack in case a newly created pack happens to
616 * contain the same set of objects as an existing one. In that case
617 * the resulting file might be different even if its name would be the
618 * same. It is best to close any reference to the old pack before it is
619 * replaced on disk. Of course no index pointers nor windows for given pack
620 * must subsist at this point. If ever objects from this pack are requested
621 * again, the new version of the pack will be reinitialized through
622 * reprepare_packed_git().
624 void free_pack_by_name(const char *pack_name
)
626 struct packed_git
*p
, **pp
= &packed_git
;
630 if (strcmp(pack_name
, p
->pack_name
) == 0) {
631 clear_delta_base_cache();
632 close_pack_windows(p
);
633 if (p
->pack_fd
!= -1)
636 free(p
->bad_object_sha1
);
646 * Do not call this directly as this leaks p->pack_fd on error return;
647 * call open_packed_git() instead.
649 static int open_packed_git_1(struct packed_git
*p
)
652 struct pack_header hdr
;
653 unsigned char sha1
[20];
654 unsigned char *idx_sha1
;
657 if (!p
->index_data
&& open_pack_index(p
))
658 return error("packfile %s index unavailable", p
->pack_name
);
660 p
->pack_fd
= git_open_noatime(p
->pack_name
, p
);
661 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
664 /* If we created the struct before we had the pack we lack size. */
666 if (!S_ISREG(st
.st_mode
))
667 return error("packfile %s not a regular file", p
->pack_name
);
668 p
->pack_size
= st
.st_size
;
669 } else if (p
->pack_size
!= st
.st_size
)
670 return error("packfile %s size changed", p
->pack_name
);
672 /* We leave these file descriptors open with sliding mmap;
673 * there is no point keeping them open across exec(), though.
675 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
677 return error("cannot determine file descriptor flags");
678 fd_flag
|= FD_CLOEXEC
;
679 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
680 return error("cannot set FD_CLOEXEC");
682 /* Verify we recognize this pack file format. */
683 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
684 return error("file %s is far too short to be a packfile", p
->pack_name
);
685 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
686 return error("file %s is not a GIT packfile", p
->pack_name
);
687 if (!pack_version_ok(hdr
.hdr_version
))
688 return error("packfile %s is version %"PRIu32
" and not"
689 " supported (try upgrading GIT to a newer version)",
690 p
->pack_name
, ntohl(hdr
.hdr_version
));
692 /* Verify the pack matches its index. */
693 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
694 return error("packfile %s claims to have %"PRIu32
" objects"
695 " while index indicates %"PRIu32
" objects",
696 p
->pack_name
, ntohl(hdr
.hdr_entries
),
698 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
699 return error("end of packfile %s is unavailable", p
->pack_name
);
700 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
701 return error("packfile %s signature is unavailable", p
->pack_name
);
702 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
703 if (hashcmp(sha1
, idx_sha1
))
704 return error("packfile %s does not match index", p
->pack_name
);
708 static int open_packed_git(struct packed_git
*p
)
710 if (!open_packed_git_1(p
))
712 if (p
->pack_fd
!= -1) {
719 static int in_window(struct pack_window
*win
, off_t offset
)
721 /* We must promise at least 20 bytes (one hash) after the
722 * offset is available from this window, otherwise the offset
723 * is not actually in this window and a different window (which
724 * has that one hash excess) must be used. This is to support
725 * the object header and delta base parsing routines below.
727 off_t win_off
= win
->offset
;
728 return win_off
<= offset
729 && (offset
+ 20) <= (win_off
+ win
->len
);
732 unsigned char *use_pack(struct packed_git
*p
,
733 struct pack_window
**w_cursor
,
737 struct pack_window
*win
= *w_cursor
;
739 if (p
->pack_fd
== -1 && open_packed_git(p
))
740 die("packfile %s cannot be accessed", p
->pack_name
);
742 /* Since packfiles end in a hash of their content and it's
743 * pointless to ask for an offset into the middle of that
744 * hash, and the in_window function above wouldn't match
745 * don't allow an offset too close to the end of the file.
747 if (offset
> (p
->pack_size
- 20))
748 die("offset beyond end of packfile (truncated pack?)");
750 if (!win
|| !in_window(win
, offset
)) {
753 for (win
= p
->windows
; win
; win
= win
->next
) {
754 if (in_window(win
, offset
))
758 size_t window_align
= packed_git_window_size
/ 2;
760 win
= xcalloc(1, sizeof(*win
));
761 win
->offset
= (offset
/ window_align
) * window_align
;
762 len
= p
->pack_size
- win
->offset
;
763 if (len
> packed_git_window_size
)
764 len
= packed_git_window_size
;
765 win
->len
= (size_t)len
;
766 pack_mapped
+= win
->len
;
767 while (packed_git_limit
< pack_mapped
768 && unuse_one_window(p
, p
->pack_fd
))
770 win
->base
= xmmap(NULL
, win
->len
,
771 PROT_READ
, MAP_PRIVATE
,
772 p
->pack_fd
, win
->offset
);
773 if (win
->base
== MAP_FAILED
)
774 die("packfile %s cannot be mapped: %s",
779 if (pack_mapped
> peak_pack_mapped
)
780 peak_pack_mapped
= pack_mapped
;
781 if (pack_open_windows
> peak_pack_open_windows
)
782 peak_pack_open_windows
= pack_open_windows
;
783 win
->next
= p
->windows
;
787 if (win
!= *w_cursor
) {
788 win
->last_used
= pack_used_ctr
++;
792 offset
-= win
->offset
;
794 *left
= win
->len
- xsize_t(offset
);
795 return win
->base
+ offset
;
798 static struct packed_git
*alloc_packed_git(int extra
)
800 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
801 memset(p
, 0, sizeof(*p
));
806 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
809 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
812 * Make sure a corresponding .pack file exists and that
813 * the index looks sane.
815 path_len
-= strlen(".idx");
820 memcpy(p
->pack_name
, path
, path_len
);
822 strcpy(p
->pack_name
+ path_len
, ".keep");
823 if (!access(p
->pack_name
, F_OK
))
826 strcpy(p
->pack_name
+ path_len
, ".pack");
827 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
832 /* ok, it looks sane as far as we can check without
833 * actually mapping the pack file.
835 p
->pack_size
= st
.st_size
;
836 p
->pack_local
= local
;
837 p
->mtime
= st
.st_mtime
;
838 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
843 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
845 const char *path
= sha1_pack_name(sha1
);
846 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
848 strcpy(p
->pack_name
, path
);
849 hashcpy(p
->sha1
, sha1
);
850 if (check_packed_git_idx(idx_path
, p
)) {
858 void install_packed_git(struct packed_git
*pack
)
860 pack
->next
= packed_git
;
864 static void prepare_packed_git_one(char *objdir
, int local
)
866 /* Ensure that this buffer is large enough so that we can
867 append "/pack/" without clobbering the stack even if
868 strlen(objdir) were PATH_MAX. */
869 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
874 sprintf(path
, "%s/pack", objdir
);
877 while (!dir
&& errno
== EMFILE
&& unuse_one_window(NULL
, -1))
881 error("unable to open object pack directory: %s: %s",
882 path
, strerror(errno
));
886 while ((de
= readdir(dir
)) != NULL
) {
887 int namelen
= strlen(de
->d_name
);
888 struct packed_git
*p
;
890 if (!has_extension(de
->d_name
, ".idx"))
893 if (len
+ namelen
+ 1 > sizeof(path
))
896 /* Don't reopen a pack we already have. */
897 strcpy(path
+ len
, de
->d_name
);
898 for (p
= packed_git
; p
; p
= p
->next
) {
899 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
904 /* See if it really is a valid .idx file with corresponding
905 * .pack file that we can map.
907 p
= add_packed_git(path
, len
+ namelen
, local
);
910 install_packed_git(p
);
915 static int sort_pack(const void *a_
, const void *b_
)
917 struct packed_git
*a
= *((struct packed_git
**)a_
);
918 struct packed_git
*b
= *((struct packed_git
**)b_
);
922 * Local packs tend to contain objects specific to our
923 * variant of the project than remote ones. In addition,
924 * remote ones could be on a network mounted filesystem.
925 * Favor local ones for these reasons.
927 st
= a
->pack_local
- b
->pack_local
;
932 * Younger packs tend to contain more recent objects,
933 * and more recent objects tend to get accessed more
936 if (a
->mtime
< b
->mtime
)
938 else if (a
->mtime
== b
->mtime
)
943 static void rearrange_packed_git(void)
945 struct packed_git
**ary
, *p
;
948 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
953 /* prepare an array of packed_git for easier sorting */
954 ary
= xcalloc(n
, sizeof(struct packed_git
*));
955 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
958 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
960 /* link them back again */
961 for (i
= 0; i
< n
- 1; i
++)
962 ary
[i
]->next
= ary
[i
+ 1];
963 ary
[n
- 1]->next
= NULL
;
969 static int prepare_packed_git_run_once
= 0;
970 void prepare_packed_git(void)
972 struct alternate_object_database
*alt
;
974 if (prepare_packed_git_run_once
)
976 prepare_packed_git_one(get_object_directory(), 1);
978 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
980 prepare_packed_git_one(alt
->base
, 0);
983 rearrange_packed_git();
984 prepare_packed_git_run_once
= 1;
987 void reprepare_packed_git(void)
990 prepare_packed_git_run_once
= 0;
991 prepare_packed_git();
994 static void mark_bad_packed_object(struct packed_git
*p
,
995 const unsigned char *sha1
)
998 for (i
= 0; i
< p
->num_bad_objects
; i
++)
999 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1001 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1002 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1003 p
->num_bad_objects
++;
1006 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1008 struct packed_git
*p
;
1011 for (p
= packed_git
; p
; p
= p
->next
)
1012 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1013 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1018 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1020 unsigned char real_sha1
[20];
1021 hash_sha1_file(map
, size
, type
, real_sha1
);
1022 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1025 static int git_open_noatime(const char *name
, struct packed_git
*p
)
1027 static int sha1_file_open_flag
= O_NOATIME
;
1030 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1034 /* Might the failure be insufficient file descriptors? */
1035 if (errno
== EMFILE
) {
1036 if (unuse_one_window(p
, -1))
1042 /* Might the failure be due to O_NOATIME? */
1043 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1044 sha1_file_open_flag
= 0;
1052 static int open_sha1_file(const unsigned char *sha1
)
1055 char *name
= sha1_file_name(sha1
);
1056 struct alternate_object_database
*alt
;
1058 fd
= git_open_noatime(name
, NULL
);
1064 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1066 fill_sha1_path(name
, sha1
);
1067 fd
= git_open_noatime(alt
->base
, NULL
);
1074 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1079 fd
= open_sha1_file(sha1
);
1084 if (!fstat(fd
, &st
)) {
1085 *size
= xsize_t(st
.st_size
);
1086 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1093 static int legacy_loose_object(unsigned char *map
)
1098 * Is it a zlib-compressed buffer? If so, the first byte
1099 * must be 0x78 (15-bit window size, deflated), and the
1100 * first 16-bit word is evenly divisible by 31
1102 word
= (map
[0] << 8) + map
[1];
1103 if (map
[0] == 0x78 && !(word
% 31))
1109 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1110 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1113 unsigned long size
, c
;
1114 unsigned long used
= 0;
1117 *type
= (c
>> 4) & 7;
1121 if (len
<= used
|| bitsizeof(long) <= shift
) {
1122 error("bad object header");
1126 size
+= (c
& 0x7f) << shift
;
1133 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1135 unsigned long size
, used
;
1136 static const char valid_loose_object_type
[8] = {
1138 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1139 0, /* "delta" and others are invalid in a loose object */
1141 enum object_type type
;
1143 /* Get the data stream */
1144 memset(stream
, 0, sizeof(*stream
));
1145 stream
->next_in
= map
;
1146 stream
->avail_in
= mapsize
;
1147 stream
->next_out
= buffer
;
1148 stream
->avail_out
= bufsiz
;
1150 if (legacy_loose_object(map
)) {
1151 git_inflate_init(stream
);
1152 return git_inflate(stream
, 0);
1157 * There used to be a second loose object header format which
1158 * was meant to mimic the in-pack format, allowing for direct
1159 * copy of the object data. This format turned up not to be
1160 * really worth it and we don't write it any longer. But we
1161 * can still read it.
1163 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1164 if (!used
|| !valid_loose_object_type
[type
])
1169 /* Set up the stream for the rest.. */
1170 stream
->next_in
= map
;
1171 stream
->avail_in
= mapsize
;
1172 git_inflate_init(stream
);
1174 /* And generate the fake traditional header */
1175 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1176 typename(type
), size
);
1180 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1182 int bytes
= strlen(buffer
) + 1;
1183 unsigned char *buf
= xmallocz(size
);
1187 n
= stream
->total_out
- bytes
;
1190 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1192 if (bytes
<= size
) {
1194 * The above condition must be (bytes <= size), not
1195 * (bytes < size). In other words, even though we
1196 * expect no more output and set avail_out to zer0,
1197 * the input zlib stream may have bytes that express
1198 * "this concludes the stream", and we *do* want to
1201 * Otherwise we would not be able to test that we
1202 * consumed all the input to reach the expected size;
1203 * we also want to check that zlib tells us that all
1204 * went well with status == Z_STREAM_END at the end.
1206 stream
->next_out
= buf
+ bytes
;
1207 stream
->avail_out
= size
- bytes
;
1208 while (status
== Z_OK
)
1209 status
= git_inflate(stream
, Z_FINISH
);
1211 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1212 git_inflate_end(stream
);
1217 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1218 else if (stream
->avail_in
)
1219 error("garbage at end of loose object '%s'",
1226 * We used to just use "sscanf()", but that's actually way
1227 * too permissive for what we want to check. So do an anal
1228 * object header parse by hand.
1230 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1237 * The type can be at most ten bytes (including the
1238 * terminating '\0' that we add), and is followed by
1247 if (i
>= sizeof(type
))
1253 * The length must follow immediately, and be in canonical
1254 * decimal format (ie "010" is not valid).
1256 size
= *hdr
++ - '0';
1261 unsigned long c
= *hdr
- '0';
1265 size
= size
* 10 + c
;
1271 * The length must be followed by a zero byte
1273 return *hdr
? -1 : type_from_string(type
);
1276 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1282 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1283 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1286 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1289 unsigned long get_size_from_delta(struct packed_git
*p
,
1290 struct pack_window
**w_curs
,
1293 const unsigned char *data
;
1294 unsigned char delta_head
[20], *in
;
1298 memset(&stream
, 0, sizeof(stream
));
1299 stream
.next_out
= delta_head
;
1300 stream
.avail_out
= sizeof(delta_head
);
1302 git_inflate_init(&stream
);
1304 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1305 stream
.next_in
= in
;
1306 st
= git_inflate(&stream
, Z_FINISH
);
1307 curpos
+= stream
.next_in
- in
;
1308 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1309 stream
.total_out
< sizeof(delta_head
));
1310 git_inflate_end(&stream
);
1311 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1312 error("delta data unpack-initial failed");
1316 /* Examine the initial part of the delta to figure out
1321 /* ignore base size */
1322 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1324 /* Read the result size */
1325 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1328 static off_t
get_delta_base(struct packed_git
*p
,
1329 struct pack_window
**w_curs
,
1331 enum object_type type
,
1332 off_t delta_obj_offset
)
1334 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1337 /* use_pack() assured us we have [base_info, base_info + 20)
1338 * as a range that we can look at without walking off the
1339 * end of the mapped window. Its actually the hash size
1340 * that is assured. An OFS_DELTA longer than the hash size
1341 * is stupid, as then a REF_DELTA would be smaller to store.
1343 if (type
== OBJ_OFS_DELTA
) {
1345 unsigned char c
= base_info
[used
++];
1346 base_offset
= c
& 127;
1349 if (!base_offset
|| MSB(base_offset
, 7))
1350 return 0; /* overflow */
1351 c
= base_info
[used
++];
1352 base_offset
= (base_offset
<< 7) + (c
& 127);
1354 base_offset
= delta_obj_offset
- base_offset
;
1355 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1356 return 0; /* out of bound */
1358 } else if (type
== OBJ_REF_DELTA
) {
1359 /* The base entry _must_ be in the same pack */
1360 base_offset
= find_pack_entry_one(base_info
, p
);
1363 die("I am totally screwed");
1367 /* forward declaration for a mutually recursive function */
1368 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1369 unsigned long *sizep
);
1371 static int packed_delta_info(struct packed_git
*p
,
1372 struct pack_window
**w_curs
,
1374 enum object_type type
,
1376 unsigned long *sizep
)
1380 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1383 type
= packed_object_info(p
, base_offset
, NULL
);
1384 if (type
<= OBJ_NONE
) {
1385 struct revindex_entry
*revidx
;
1386 const unsigned char *base_sha1
;
1387 revidx
= find_pack_revindex(p
, base_offset
);
1390 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1391 mark_bad_packed_object(p
, base_sha1
);
1392 type
= sha1_object_info(base_sha1
, NULL
);
1393 if (type
<= OBJ_NONE
)
1397 /* We choose to only get the type of the base object and
1398 * ignore potentially corrupt pack file that expects the delta
1399 * based on a base with a wrong size. This saves tons of
1403 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1411 static int unpack_object_header(struct packed_git
*p
,
1412 struct pack_window
**w_curs
,
1414 unsigned long *sizep
)
1416 unsigned char *base
;
1419 enum object_type type
;
1421 /* use_pack() assures us we have [base, base + 20) available
1422 * as a range that we can look at at. (Its actually the hash
1423 * size that is assured.) With our object header encoding
1424 * the maximum deflated object size is 2^137, which is just
1425 * insane, so we know won't exceed what we have been given.
1427 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1428 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1437 const char *packed_object_info_detail(struct packed_git
*p
,
1439 unsigned long *size
,
1440 unsigned long *store_size
,
1441 unsigned int *delta_chain_length
,
1442 unsigned char *base_sha1
)
1444 struct pack_window
*w_curs
= NULL
;
1446 unsigned long dummy
;
1447 unsigned char *next_sha1
;
1448 enum object_type type
;
1449 struct revindex_entry
*revidx
;
1451 *delta_chain_length
= 0;
1452 curpos
= obj_offset
;
1453 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1455 revidx
= find_pack_revindex(p
, obj_offset
);
1456 *store_size
= revidx
[1].offset
- obj_offset
;
1461 die("pack %s contains unknown object type %d",
1462 p
->pack_name
, type
);
1467 unuse_pack(&w_curs
);
1468 return typename(type
);
1470 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1472 die("pack %s contains bad delta base reference of type %s",
1473 p
->pack_name
, typename(type
));
1474 if (*delta_chain_length
== 0) {
1475 revidx
= find_pack_revindex(p
, obj_offset
);
1476 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1480 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1481 if (*delta_chain_length
== 0)
1482 hashcpy(base_sha1
, next_sha1
);
1483 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1486 (*delta_chain_length
)++;
1487 curpos
= obj_offset
;
1488 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1492 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1493 unsigned long *sizep
)
1495 struct pack_window
*w_curs
= NULL
;
1497 off_t curpos
= obj_offset
;
1498 enum object_type type
;
1500 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1505 type
= packed_delta_info(p
, &w_curs
, curpos
,
1506 type
, obj_offset
, sizep
);
1516 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1517 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1520 unuse_pack(&w_curs
);
1524 static void *unpack_compressed_entry(struct packed_git
*p
,
1525 struct pack_window
**w_curs
,
1531 unsigned char *buffer
, *in
;
1533 buffer
= xmallocz(size
);
1534 memset(&stream
, 0, sizeof(stream
));
1535 stream
.next_out
= buffer
;
1536 stream
.avail_out
= size
+ 1;
1538 git_inflate_init(&stream
);
1540 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1541 stream
.next_in
= in
;
1542 st
= git_inflate(&stream
, Z_FINISH
);
1543 if (!stream
.avail_out
)
1544 break; /* the payload is larger than it should be */
1545 curpos
+= stream
.next_in
- in
;
1546 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1547 git_inflate_end(&stream
);
1548 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1556 #define MAX_DELTA_CACHE (256)
1558 static size_t delta_base_cached
;
1560 static struct delta_base_cache_lru_list
{
1561 struct delta_base_cache_lru_list
*prev
;
1562 struct delta_base_cache_lru_list
*next
;
1563 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1565 static struct delta_base_cache_entry
{
1566 struct delta_base_cache_lru_list lru
;
1568 struct packed_git
*p
;
1571 enum object_type type
;
1572 } delta_base_cache
[MAX_DELTA_CACHE
];
1574 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1578 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1579 hash
+= (hash
>> 8) + (hash
>> 16);
1580 return hash
% MAX_DELTA_CACHE
;
1583 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1584 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1587 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1588 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1591 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1592 return unpack_entry(p
, base_offset
, type
, base_size
);
1596 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1597 ent
->lru
.prev
->next
= ent
->lru
.next
;
1598 delta_base_cached
-= ent
->size
;
1600 ret
= xmemdupz(ent
->data
, ent
->size
);
1603 *base_size
= ent
->size
;
1607 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1612 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1613 ent
->lru
.prev
->next
= ent
->lru
.next
;
1614 delta_base_cached
-= ent
->size
;
1618 void clear_delta_base_cache(void)
1621 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1622 release_delta_base_cache(&delta_base_cache
[p
]);
1625 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1626 void *base
, unsigned long base_size
, enum object_type type
)
1628 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1629 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1630 struct delta_base_cache_lru_list
*lru
;
1632 release_delta_base_cache(ent
);
1633 delta_base_cached
+= base_size
;
1635 for (lru
= delta_base_cache_lru
.next
;
1636 delta_base_cached
> delta_base_cache_limit
1637 && lru
!= &delta_base_cache_lru
;
1639 struct delta_base_cache_entry
*f
= (void *)lru
;
1640 if (f
->type
== OBJ_BLOB
)
1641 release_delta_base_cache(f
);
1643 for (lru
= delta_base_cache_lru
.next
;
1644 delta_base_cached
> delta_base_cache_limit
1645 && lru
!= &delta_base_cache_lru
;
1647 struct delta_base_cache_entry
*f
= (void *)lru
;
1648 release_delta_base_cache(f
);
1652 ent
->base_offset
= base_offset
;
1655 ent
->size
= base_size
;
1656 ent
->lru
.next
= &delta_base_cache_lru
;
1657 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1658 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1659 delta_base_cache_lru
.prev
= &ent
->lru
;
1662 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1663 unsigned long *size
);
1665 static void *unpack_delta_entry(struct packed_git
*p
,
1666 struct pack_window
**w_curs
,
1668 unsigned long delta_size
,
1670 enum object_type
*type
,
1671 unsigned long *sizep
)
1673 void *delta_data
, *result
, *base
;
1674 unsigned long base_size
;
1677 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1679 error("failed to validate delta base reference "
1680 "at offset %"PRIuMAX
" from %s",
1681 (uintmax_t)curpos
, p
->pack_name
);
1685 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1688 * We're probably in deep shit, but let's try to fetch
1689 * the required base anyway from another pack or loose.
1690 * This is costly but should happen only in the presence
1691 * of a corrupted pack, and is better than failing outright.
1693 struct revindex_entry
*revidx
;
1694 const unsigned char *base_sha1
;
1695 revidx
= find_pack_revindex(p
, base_offset
);
1698 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1699 error("failed to read delta base object %s"
1700 " at offset %"PRIuMAX
" from %s",
1701 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1703 mark_bad_packed_object(p
, base_sha1
);
1704 base
= read_object(base_sha1
, type
, &base_size
);
1709 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1711 error("failed to unpack compressed delta "
1712 "at offset %"PRIuMAX
" from %s",
1713 (uintmax_t)curpos
, p
->pack_name
);
1717 result
= patch_delta(base
, base_size
,
1718 delta_data
, delta_size
,
1721 die("failed to apply delta");
1723 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1727 int do_check_packed_object_crc
;
1729 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1730 enum object_type
*type
, unsigned long *sizep
)
1732 struct pack_window
*w_curs
= NULL
;
1733 off_t curpos
= obj_offset
;
1736 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1737 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1738 unsigned long len
= revidx
[1].offset
- obj_offset
;
1739 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1740 const unsigned char *sha1
=
1741 nth_packed_object_sha1(p
, revidx
->nr
);
1742 error("bad packed object CRC for %s",
1744 mark_bad_packed_object(p
, sha1
);
1745 unuse_pack(&w_curs
);
1750 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1754 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1755 obj_offset
, type
, sizep
);
1761 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1765 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1766 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1768 unuse_pack(&w_curs
);
1772 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1775 const unsigned char *index
= p
->index_data
;
1777 if (open_pack_index(p
))
1779 index
= p
->index_data
;
1781 if (n
>= p
->num_objects
)
1784 if (p
->index_version
== 1) {
1785 return index
+ 24 * n
+ 4;
1788 return index
+ 20 * n
;
1792 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1794 const unsigned char *index
= p
->index_data
;
1796 if (p
->index_version
== 1) {
1797 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1800 index
+= 8 + p
->num_objects
* (20 + 4);
1801 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1802 if (!(off
& 0x80000000))
1804 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1805 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1806 ntohl(*((uint32_t *)(index
+ 4)));
1810 off_t
find_pack_entry_one(const unsigned char *sha1
,
1811 struct packed_git
*p
)
1813 const uint32_t *level1_ofs
= p
->index_data
;
1814 const unsigned char *index
= p
->index_data
;
1815 unsigned hi
, lo
, stride
;
1816 static int use_lookup
= -1;
1817 static int debug_lookup
= -1;
1819 if (debug_lookup
< 0)
1820 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1823 if (open_pack_index(p
))
1825 level1_ofs
= p
->index_data
;
1826 index
= p
->index_data
;
1828 if (p
->index_version
> 1) {
1833 hi
= ntohl(level1_ofs
[*sha1
]);
1834 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1835 if (p
->index_version
> 1) {
1843 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1844 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1847 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1849 int pos
= sha1_entry_pos(index
, stride
, 0,
1850 lo
, hi
, p
->num_objects
, sha1
);
1853 return nth_packed_object_offset(p
, pos
);
1857 unsigned mi
= (lo
+ hi
) / 2;
1858 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1861 printf("lo %u hi %u rg %u mi %u\n",
1862 lo
, hi
, hi
- lo
, mi
);
1864 return nth_packed_object_offset(p
, mi
);
1873 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1875 static struct packed_git
*last_found
= (void *)1;
1876 struct packed_git
*p
;
1879 prepare_packed_git();
1882 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1885 if (p
->num_bad_objects
) {
1887 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1888 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1892 offset
= find_pack_entry_one(sha1
, p
);
1895 * We are about to tell the caller where they can
1896 * locate the requested object. We better make
1897 * sure the packfile is still here and can be
1898 * accessed before supplying that answer, as
1899 * it may have been deleted since the index
1902 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1903 error("packfile %s cannot be accessed", p
->pack_name
);
1908 hashcpy(e
->sha1
, sha1
);
1914 if (p
== last_found
)
1918 if (p
== last_found
)
1924 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1925 struct packed_git
*packs
)
1927 struct packed_git
*p
;
1929 for (p
= packs
; p
; p
= p
->next
) {
1930 if (find_pack_entry_one(sha1
, p
))
1937 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1940 unsigned long mapsize
, size
;
1945 map
= map_sha1_file(sha1
, &mapsize
);
1947 return error("unable to find %s", sha1_to_hex(sha1
));
1948 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1949 status
= error("unable to unpack %s header",
1951 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1952 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1955 git_inflate_end(&stream
);
1956 munmap(map
, mapsize
);
1960 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1962 struct pack_entry e
;
1965 if (!find_pack_entry(sha1
, &e
)) {
1966 /* Most likely it's a loose object. */
1967 status
= sha1_loose_object_info(sha1
, sizep
);
1971 /* Not a loose object; someone else may have just packed it. */
1972 reprepare_packed_git();
1973 if (!find_pack_entry(sha1
, &e
))
1977 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
1979 mark_bad_packed_object(e
.p
, sha1
);
1980 status
= sha1_object_info(sha1
, sizep
);
1986 static void *read_packed_sha1(const unsigned char *sha1
,
1987 enum object_type
*type
, unsigned long *size
)
1989 struct pack_entry e
;
1992 if (!find_pack_entry(sha1
, &e
))
1994 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1997 * We're probably in deep shit, but let's try to fetch
1998 * the required object anyway from another pack or loose.
1999 * This should happen only in the presence of a corrupted
2000 * pack, and is better than failing outright.
2002 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2003 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2004 mark_bad_packed_object(e
.p
, sha1
);
2005 data
= read_object(sha1
, type
, size
);
2011 * This is meant to hold a *small* number of objects that you would
2012 * want read_sha1_file() to be able to return, but yet you do not want
2013 * to write them into the object store (e.g. a browse-only
2016 static struct cached_object
{
2017 unsigned char sha1
[20];
2018 enum object_type type
;
2022 static int cached_object_nr
, cached_object_alloc
;
2024 static struct cached_object empty_tree
= {
2025 EMPTY_TREE_SHA1_BIN
,
2031 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
2034 struct cached_object
*co
= cached_objects
;
2036 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
2037 if (!hashcmp(co
->sha1
, sha1
))
2040 if (!hashcmp(sha1
, empty_tree
.sha1
))
2045 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2046 unsigned char *sha1
)
2048 struct cached_object
*co
;
2050 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2051 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2053 if (cached_object_alloc
<= cached_object_nr
) {
2054 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2055 cached_objects
= xrealloc(cached_objects
,
2056 sizeof(*cached_objects
) *
2057 cached_object_alloc
);
2059 co
= &cached_objects
[cached_object_nr
++];
2062 co
->buf
= xmalloc(len
);
2063 memcpy(co
->buf
, buf
, len
);
2064 hashcpy(co
->sha1
, sha1
);
2068 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2069 unsigned long *size
)
2071 unsigned long mapsize
;
2073 struct cached_object
*co
;
2075 co
= find_cached_object(sha1
);
2079 return xmemdupz(co
->buf
, co
->size
);
2082 buf
= read_packed_sha1(sha1
, type
, size
);
2085 map
= map_sha1_file(sha1
, &mapsize
);
2087 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2088 munmap(map
, mapsize
);
2091 reprepare_packed_git();
2092 return read_packed_sha1(sha1
, type
, size
);
2096 * This function dies on corrupt objects; the callers who want to
2097 * deal with them should arrange to call read_object() and give error
2098 * messages themselves.
2100 void *read_sha1_file_repl(const unsigned char *sha1
,
2101 enum object_type
*type
,
2102 unsigned long *size
,
2103 const unsigned char **replacement
)
2105 const unsigned char *repl
= lookup_replace_object(sha1
);
2108 const struct packed_git
*p
;
2111 data
= read_object(repl
, type
, size
);
2114 *replacement
= repl
;
2118 if (errno
!= ENOENT
)
2119 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2121 /* die if we replaced an object with one that does not exist */
2123 die("replacement %s not found for %s",
2124 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2126 if (has_loose_object(repl
)) {
2127 path
= sha1_file_name(sha1
);
2128 die("loose object %s (stored in %s) is corrupt",
2129 sha1_to_hex(repl
), path
);
2132 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2133 die("packed object %s (stored in %s) is corrupt",
2134 sha1_to_hex(repl
), p
->pack_name
);
2139 void *read_object_with_reference(const unsigned char *sha1
,
2140 const char *required_type_name
,
2141 unsigned long *size
,
2142 unsigned char *actual_sha1_return
)
2144 enum object_type type
, required_type
;
2146 unsigned long isize
;
2147 unsigned char actual_sha1
[20];
2149 required_type
= type_from_string(required_type_name
);
2150 hashcpy(actual_sha1
, sha1
);
2152 int ref_length
= -1;
2153 const char *ref_type
= NULL
;
2155 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2158 if (type
== required_type
) {
2160 if (actual_sha1_return
)
2161 hashcpy(actual_sha1_return
, actual_sha1
);
2164 /* Handle references */
2165 else if (type
== OBJ_COMMIT
)
2167 else if (type
== OBJ_TAG
)
2168 ref_type
= "object ";
2173 ref_length
= strlen(ref_type
);
2175 if (ref_length
+ 40 > isize
||
2176 memcmp(buffer
, ref_type
, ref_length
) ||
2177 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2182 /* Now we have the ID of the referred-to object in
2183 * actual_sha1. Check again. */
2187 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2188 const char *type
, unsigned char *sha1
,
2189 char *hdr
, int *hdrlen
)
2193 /* Generate the header */
2194 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2198 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2199 git_SHA1_Update(&c
, buf
, len
);
2200 git_SHA1_Final(sha1
, &c
);
2204 * Move the just written object into its final resting place.
2205 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2206 * "moving" is only a part of what it does, when no patch between
2207 * master to pu changes the call sites of this function.
2209 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2213 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2215 else if (link(tmpfile
, filename
))
2219 * Coda hack - coda doesn't like cross-directory links,
2220 * so we fall back to a rename, which will mean that it
2221 * won't be able to check collisions, but that's not a
2224 * The same holds for FAT formatted media.
2226 * When this succeeds, we just return. We have nothing
2229 if (ret
&& ret
!= EEXIST
) {
2231 if (!rename(tmpfile
, filename
))
2235 unlink_or_warn(tmpfile
);
2237 if (ret
!= EEXIST
) {
2238 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2240 /* FIXME!!! Collision check here ? */
2244 if (adjust_shared_perm(filename
))
2245 return error("unable to set permission to '%s'", filename
);
2249 static int write_buffer(int fd
, const void *buf
, size_t len
)
2251 if (write_in_full(fd
, buf
, len
) < 0)
2252 return error("file write error (%s)", strerror(errno
));
2256 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2257 unsigned char *sha1
)
2261 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2265 /* Finalize a file on disk, and close it. */
2266 static void close_sha1_file(int fd
)
2268 if (fsync_object_files
)
2269 fsync_or_die(fd
, "sha1 file");
2271 die_errno("error when closing sha1 file");
2274 /* Size of directory component, including the ending '/' */
2275 static inline int directory_size(const char *filename
)
2277 const char *s
= strrchr(filename
, '/');
2280 return s
- filename
+ 1;
2284 * This creates a temporary file in the same directory as the final
2287 * We want to avoid cross-directory filename renames, because those
2288 * can have problems on various filesystems (FAT, NFS, Coda).
2290 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2292 int fd
, dirlen
= directory_size(filename
);
2294 if (dirlen
+ 20 > bufsiz
) {
2295 errno
= ENAMETOOLONG
;
2298 memcpy(buffer
, filename
, dirlen
);
2299 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2300 fd
= git_mkstemp_mode(buffer
, 0444);
2301 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2302 /* Make sure the directory exists */
2303 memcpy(buffer
, filename
, dirlen
);
2304 buffer
[dirlen
-1] = 0;
2305 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2309 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2310 fd
= git_mkstemp_mode(buffer
, 0444);
2315 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2316 const void *buf
, unsigned long len
, time_t mtime
)
2319 unsigned char compressed
[4096];
2322 unsigned char parano_sha1
[20];
2324 static char tmpfile
[PATH_MAX
];
2326 filename
= sha1_file_name(sha1
);
2327 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2328 while (fd
< 0 && errno
== EMFILE
&& unuse_one_window(NULL
, -1))
2329 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2331 if (errno
== EACCES
)
2332 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2334 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2338 memset(&stream
, 0, sizeof(stream
));
2339 deflateInit(&stream
, zlib_compression_level
);
2340 stream
.next_out
= compressed
;
2341 stream
.avail_out
= sizeof(compressed
);
2344 /* First header.. */
2345 stream
.next_in
= (unsigned char *)hdr
;
2346 stream
.avail_in
= hdrlen
;
2347 while (deflate(&stream
, 0) == Z_OK
)
2349 git_SHA1_Update(&c
, hdr
, hdrlen
);
2351 /* Then the data itself.. */
2352 stream
.next_in
= (void *)buf
;
2353 stream
.avail_in
= len
;
2355 unsigned char *in0
= stream
.next_in
;
2356 ret
= deflate(&stream
, Z_FINISH
);
2357 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2358 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2359 die("unable to write sha1 file");
2360 stream
.next_out
= compressed
;
2361 stream
.avail_out
= sizeof(compressed
);
2362 } while (ret
== Z_OK
);
2364 if (ret
!= Z_STREAM_END
)
2365 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2366 ret
= deflateEnd(&stream
);
2368 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2369 git_SHA1_Final(parano_sha1
, &c
);
2370 if (hashcmp(sha1
, parano_sha1
) != 0)
2371 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2373 close_sha1_file(fd
);
2378 utb
.modtime
= mtime
;
2379 if (utime(tmpfile
, &utb
) < 0)
2380 warning("failed utime() on %s: %s",
2381 tmpfile
, strerror(errno
));
2384 return move_temp_to_file(tmpfile
, filename
);
2387 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2389 unsigned char sha1
[20];
2393 /* Normally if we have it in the pack then we do not bother writing
2394 * it out into .git/objects/??/?{38} file.
2396 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2398 hashcpy(returnsha1
, sha1
);
2399 if (has_sha1_file(sha1
))
2401 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2404 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2408 enum object_type type
;
2413 if (has_loose_object(sha1
))
2415 buf
= read_packed_sha1(sha1
, &type
, &len
);
2417 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2418 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2419 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2425 int has_pack_index(const unsigned char *sha1
)
2428 if (stat(sha1_pack_index_name(sha1
), &st
))
2433 int has_sha1_pack(const unsigned char *sha1
)
2435 struct pack_entry e
;
2436 return find_pack_entry(sha1
, &e
);
2439 int has_sha1_file(const unsigned char *sha1
)
2441 struct pack_entry e
;
2443 if (find_pack_entry(sha1
, &e
))
2445 return has_loose_object(sha1
);
2448 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2449 int write_object
, enum object_type type
, const char *path
)
2451 int ret
, re_allocated
= 0;
2457 * Convert blobs to git internal format
2459 if ((type
== OBJ_BLOB
) && path
) {
2460 struct strbuf nbuf
= STRBUF_INIT
;
2461 if (convert_to_git(path
, buf
, size
, &nbuf
,
2462 write_object
? safe_crlf
: 0)) {
2463 buf
= strbuf_detach(&nbuf
, &size
);
2469 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2471 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2477 #define SMALL_FILE_SIZE (32*1024)
2479 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2480 enum object_type type
, const char *path
)
2483 size_t size
= xsize_t(st
->st_size
);
2485 if (!S_ISREG(st
->st_mode
)) {
2486 struct strbuf sbuf
= STRBUF_INIT
;
2487 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2488 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2492 strbuf_release(&sbuf
);
2494 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2495 } else if (size
<= SMALL_FILE_SIZE
) {
2496 char *buf
= xmalloc(size
);
2497 if (size
== read_in_full(fd
, buf
, size
))
2498 ret
= index_mem(sha1
, buf
, size
, write_object
, type
,
2501 ret
= error("short read %s", strerror(errno
));
2504 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2505 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2512 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2515 struct strbuf sb
= STRBUF_INIT
;
2517 switch (st
->st_mode
& S_IFMT
) {
2519 fd
= open(path
, O_RDONLY
);
2521 return error("open(\"%s\"): %s", path
,
2523 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2524 return error("%s: failed to insert into database",
2528 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2529 char *errstr
= strerror(errno
);
2530 return error("readlink(\"%s\"): %s", path
,
2534 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2535 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2536 return error("%s: failed to insert into database",
2538 strbuf_release(&sb
);
2541 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2543 return error("%s: unsupported file type", path
);
2548 int read_pack_header(int fd
, struct pack_header
*header
)
2550 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2551 /* "eof before pack header was fully read" */
2552 return PH_ERROR_EOF
;
2554 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2555 /* "protocol error (pack signature mismatch detected)" */
2556 return PH_ERROR_PACK_SIGNATURE
;
2557 if (!pack_version_ok(header
->hdr_version
))
2558 /* "protocol error (pack version unsupported)" */
2559 return PH_ERROR_PROTOCOL
;
2563 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2565 enum object_type type
= sha1_object_info(sha1
, NULL
);
2567 die("%s is not a valid object", sha1_to_hex(sha1
));
2569 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),