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
28 #define SZ_FMT PRIuMAX
29 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
31 const unsigned char null_sha1
[20];
33 static int git_open_noatime(const char *name
, struct packed_git
*p
);
36 * This is meant to hold a *small* number of objects that you would
37 * want read_sha1_file() to be able to return, but yet you do not want
38 * to write them into the object store (e.g. a browse-only
41 static struct cached_object
{
42 unsigned char sha1
[20];
43 enum object_type type
;
47 static int cached_object_nr
, cached_object_alloc
;
49 static struct cached_object empty_tree
= {
50 EMPTY_TREE_SHA1_BIN_LITERAL
,
56 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
59 struct cached_object
*co
= cached_objects
;
61 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
62 if (!hashcmp(co
->sha1
, sha1
))
65 if (!hashcmp(sha1
, empty_tree
.sha1
))
70 int safe_create_leading_directories(char *path
)
72 char *pos
= path
+ offset_1st_component(path
);
76 pos
= strchr(pos
, '/');
84 if (!stat(path
, &st
)) {
86 if (!S_ISDIR(st
.st_mode
)) {
91 else if (mkdir(path
, 0777)) {
95 else if (adjust_shared_perm(path
)) {
104 int safe_create_leading_directories_const(const char *path
)
106 /* path points to cache entries, so xstrdup before messing with it */
107 char *buf
= xstrdup(path
);
108 int result
= safe_create_leading_directories(buf
);
113 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
116 for (i
= 0; i
< 20; i
++) {
117 static char hex
[] = "0123456789abcdef";
118 unsigned int val
= sha1
[i
];
119 char *pos
= pathbuf
+ i
*2 + (i
> 0);
120 *pos
++ = hex
[val
>> 4];
121 *pos
= hex
[val
& 0xf];
126 * NOTE! This returns a statically allocated buffer, so you have to be
127 * careful about using it. Do an "xstrdup()" if you need to save the
130 * Also note that this returns the location for creating. Reading
131 * SHA1 file can happen from any alternate directory listed in the
132 * DB_ENVIRONMENT environment variable if it is not found in
133 * the primary object database.
135 char *sha1_file_name(const unsigned char *sha1
)
137 static char buf
[PATH_MAX
];
141 objdir
= get_object_directory();
142 len
= strlen(objdir
);
144 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
145 if (len
+ 43 > PATH_MAX
)
146 die("insanely long object directory %s", objdir
);
147 memcpy(buf
, objdir
, len
);
151 fill_sha1_path(buf
+ len
+ 1, sha1
);
155 static char *sha1_get_pack_name(const unsigned char *sha1
,
156 char **name
, char **base
, const char *which
)
158 static const char hex
[] = "0123456789abcdef";
163 const char *sha1_file_directory
= get_object_directory();
164 int len
= strlen(sha1_file_directory
);
165 *base
= xmalloc(len
+ 60);
166 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
167 sha1_file_directory
, which
);
168 *name
= *base
+ len
+ 11;
173 for (i
= 0; i
< 20; i
++) {
174 unsigned int val
= *sha1
++;
175 *buf
++ = hex
[val
>> 4];
176 *buf
++ = hex
[val
& 0xf];
182 char *sha1_pack_name(const unsigned char *sha1
)
184 static char *name
, *base
;
186 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
189 char *sha1_pack_index_name(const unsigned char *sha1
)
191 static char *name
, *base
;
193 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
196 struct alternate_object_database
*alt_odb_list
;
197 static struct alternate_object_database
**alt_odb_tail
;
199 static void read_info_alternates(const char * alternates
, int depth
);
202 * Prepare alternate object database registry.
204 * The variable alt_odb_list points at the list of struct
205 * alternate_object_database. The elements on this list come from
206 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
207 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
208 * whose contents is similar to that environment variable but can be
209 * LF separated. Its base points at a statically allocated buffer that
210 * contains "/the/directory/corresponding/to/.git/objects/...", while
211 * its name points just after the slash at the end of ".git/objects/"
212 * in the example above, and has enough space to hold 40-byte hex
213 * SHA1, an extra slash for the first level indirection, and the
216 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
218 const char *objdir
= get_object_directory();
219 struct alternate_object_database
*ent
;
220 struct alternate_object_database
*alt
;
221 /* 43 = 40-byte + 2 '/' + terminating NUL */
223 int entlen
= pfxlen
+ 43;
226 if (!is_absolute_path(entry
) && relative_base
) {
227 /* Relative alt-odb */
229 base_len
= strlen(relative_base
) + 1;
233 ent
= xmalloc(sizeof(*ent
) + entlen
);
235 if (!is_absolute_path(entry
) && relative_base
) {
236 memcpy(ent
->base
, relative_base
, base_len
- 1);
237 ent
->base
[base_len
- 1] = '/';
238 memcpy(ent
->base
+ base_len
, entry
, len
);
241 memcpy(ent
->base
, entry
, pfxlen
);
243 ent
->name
= ent
->base
+ pfxlen
+ 1;
244 ent
->base
[pfxlen
+ 3] = '/';
245 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
247 /* Detect cases where alternate disappeared */
248 if (!is_directory(ent
->base
)) {
249 error("object directory %s does not exist; "
250 "check .git/objects/info/alternates.",
256 /* Prevent the common mistake of listing the same
257 * thing twice, or object directory itself.
259 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
260 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
265 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
270 /* add the alternate entry */
272 alt_odb_tail
= &(ent
->next
);
275 /* recursively add alternates */
276 read_info_alternates(ent
->base
, depth
+ 1);
278 ent
->base
[pfxlen
] = '/';
283 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
284 const char *relative_base
, int depth
)
286 const char *cp
, *last
;
289 error("%s: ignoring alternate object stores, nesting too deep.",
297 if (cp
< ep
&& *cp
== '#') {
298 while (cp
< ep
&& *cp
!= sep
)
303 while (cp
< ep
&& *cp
!= sep
)
306 if (!is_absolute_path(last
) && depth
) {
307 error("%s: ignoring relative alternate object store %s",
308 relative_base
, last
);
310 link_alt_odb_entry(last
, cp
- last
,
311 relative_base
, depth
);
314 while (cp
< ep
&& *cp
== sep
)
320 static void read_info_alternates(const char * relative_base
, int depth
)
325 const char alt_file_name
[] = "info/alternates";
326 /* Given that relative_base is no longer than PATH_MAX,
327 ensure that "path" has enough space to append "/", the
328 file name, "info/alternates", and a trailing NUL. */
329 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
332 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
333 fd
= git_open_noatime(path
, NULL
);
336 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
340 mapsz
= xsize_t(st
.st_size
);
341 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
344 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
349 void add_to_alternates_file(const char *reference
)
351 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
352 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
353 char *alt
= mkpath("%s/objects\n", reference
);
354 write_or_die(fd
, alt
, strlen(alt
));
355 if (commit_lock_file(lock
))
356 die("could not close alternates file");
358 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
361 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
363 struct alternate_object_database
*ent
;
366 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
371 void prepare_alt_odb(void)
378 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
381 alt_odb_tail
= &alt_odb_list
;
382 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
384 read_info_alternates(get_object_directory(), 0);
387 static int has_loose_object_local(const unsigned char *sha1
)
389 char *name
= sha1_file_name(sha1
);
390 return !access(name
, F_OK
);
393 int has_loose_object_nonlocal(const unsigned char *sha1
)
395 struct alternate_object_database
*alt
;
397 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
398 fill_sha1_path(alt
->name
, sha1
);
399 if (!access(alt
->base
, F_OK
))
405 static int has_loose_object(const unsigned char *sha1
)
407 return has_loose_object_local(sha1
) ||
408 has_loose_object_nonlocal(sha1
);
411 static unsigned int pack_used_ctr
;
412 static unsigned int pack_mmap_calls
;
413 static unsigned int peak_pack_open_windows
;
414 static unsigned int pack_open_windows
;
415 static size_t peak_pack_mapped
;
416 static size_t pack_mapped
;
417 struct packed_git
*packed_git
;
419 void pack_report(void)
422 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
423 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
424 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
425 sz_fmt(getpagesize()),
426 sz_fmt(packed_git_window_size
),
427 sz_fmt(packed_git_limit
));
429 "pack_report: pack_used_ctr = %10u\n"
430 "pack_report: pack_mmap_calls = %10u\n"
431 "pack_report: pack_open_windows = %10u / %10u\n"
432 "pack_report: pack_mapped = "
433 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
436 pack_open_windows
, peak_pack_open_windows
,
437 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
440 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
443 struct pack_idx_header
*hdr
;
445 uint32_t version
, nr
, i
, *index
;
446 int fd
= git_open_noatime(path
, p
);
451 if (fstat(fd
, &st
)) {
455 idx_size
= xsize_t(st
.st_size
);
456 if (idx_size
< 4 * 256 + 20 + 20) {
458 return error("index file %s is too small", path
);
460 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
464 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
465 version
= ntohl(hdr
->idx_version
);
466 if (version
< 2 || version
> 2) {
467 munmap(idx_map
, idx_size
);
468 return error("index file %s is version %"PRIu32
469 " and is not supported by this binary"
470 " (try upgrading GIT to a newer version)",
479 index
+= 2; /* skip index header */
480 for (i
= 0; i
< 256; i
++) {
481 uint32_t n
= ntohl(index
[i
]);
483 munmap(idx_map
, idx_size
);
484 return error("non-monotonic index %s", path
);
492 * - 256 index entries 4 bytes each
493 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
494 * - 20-byte SHA1 of the packfile
495 * - 20-byte SHA1 file checksum
497 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
498 munmap(idx_map
, idx_size
);
499 return error("wrong index v1 file size in %s", path
);
501 } else if (version
== 2) {
504 * - 8 bytes of header
505 * - 256 index entries 4 bytes each
506 * - 20-byte sha1 entry * nr
507 * - 4-byte crc entry * nr
508 * - 4-byte offset entry * nr
509 * - 20-byte SHA1 of the packfile
510 * - 20-byte SHA1 file checksum
511 * And after the 4-byte offset table might be a
512 * variable sized table containing 8-byte entries
513 * for offsets larger than 2^31.
515 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
516 unsigned long max_size
= min_size
;
518 max_size
+= (nr
- 1)*8;
519 if (idx_size
< min_size
|| idx_size
> max_size
) {
520 munmap(idx_map
, idx_size
);
521 return error("wrong index v2 file size in %s", path
);
523 if (idx_size
!= min_size
&&
525 * make sure we can deal with large pack offsets.
526 * 31-bit signed offset won't be enough, neither
527 * 32-bit unsigned one will be.
529 (sizeof(off_t
) <= 4)) {
530 munmap(idx_map
, idx_size
);
531 return error("pack too large for current definition of off_t in %s", path
);
535 p
->index_version
= version
;
536 p
->index_data
= idx_map
;
537 p
->index_size
= idx_size
;
542 int open_pack_index(struct packed_git
*p
)
550 idx_name
= xstrdup(p
->pack_name
);
551 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
552 ret
= check_packed_git_idx(idx_name
, p
);
557 static void scan_windows(struct packed_git
*p
,
558 struct packed_git
**lru_p
,
559 struct pack_window
**lru_w
,
560 struct pack_window
**lru_l
)
562 struct pack_window
*w
, *w_l
;
564 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
566 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
576 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
578 struct packed_git
*p
, *lru_p
= NULL
;
579 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
582 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
583 for (p
= packed_git
; p
; p
= p
->next
)
584 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
586 munmap(lru_w
->base
, lru_w
->len
);
587 pack_mapped
-= lru_w
->len
;
589 lru_l
->next
= lru_w
->next
;
591 lru_p
->windows
= lru_w
->next
;
592 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
593 close(lru_p
->pack_fd
);
604 void release_pack_memory(size_t need
, int fd
)
606 size_t cur
= pack_mapped
;
607 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
611 void *xmmap(void *start
, size_t length
,
612 int prot
, int flags
, int fd
, off_t offset
)
614 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
615 if (ret
== MAP_FAILED
) {
618 release_pack_memory(length
, fd
);
619 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
620 if (ret
== MAP_FAILED
)
621 die_errno("Out of memory? mmap failed");
626 void close_pack_windows(struct packed_git
*p
)
629 struct pack_window
*w
= p
->windows
;
632 die("pack '%s' still has open windows to it",
634 munmap(w
->base
, w
->len
);
635 pack_mapped
-= w
->len
;
637 p
->windows
= w
->next
;
642 void unuse_pack(struct pack_window
**w_cursor
)
644 struct pack_window
*w
= *w_cursor
;
651 void close_pack_index(struct packed_git
*p
)
654 munmap((void *)p
->index_data
, p
->index_size
);
655 p
->index_data
= NULL
;
660 * This is used by git-repack in case a newly created pack happens to
661 * contain the same set of objects as an existing one. In that case
662 * the resulting file might be different even if its name would be the
663 * same. It is best to close any reference to the old pack before it is
664 * replaced on disk. Of course no index pointers nor windows for given pack
665 * must subsist at this point. If ever objects from this pack are requested
666 * again, the new version of the pack will be reinitialized through
667 * reprepare_packed_git().
669 void free_pack_by_name(const char *pack_name
)
671 struct packed_git
*p
, **pp
= &packed_git
;
675 if (strcmp(pack_name
, p
->pack_name
) == 0) {
676 clear_delta_base_cache();
677 close_pack_windows(p
);
678 if (p
->pack_fd
!= -1)
681 free(p
->bad_object_sha1
);
691 * Do not call this directly as this leaks p->pack_fd on error return;
692 * call open_packed_git() instead.
694 static int open_packed_git_1(struct packed_git
*p
)
697 struct pack_header hdr
;
698 unsigned char sha1
[20];
699 unsigned char *idx_sha1
;
702 if (!p
->index_data
&& open_pack_index(p
))
703 return error("packfile %s index unavailable", p
->pack_name
);
705 p
->pack_fd
= git_open_noatime(p
->pack_name
, p
);
706 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
709 /* If we created the struct before we had the pack we lack size. */
711 if (!S_ISREG(st
.st_mode
))
712 return error("packfile %s not a regular file", p
->pack_name
);
713 p
->pack_size
= st
.st_size
;
714 } else if (p
->pack_size
!= st
.st_size
)
715 return error("packfile %s size changed", p
->pack_name
);
717 /* We leave these file descriptors open with sliding mmap;
718 * there is no point keeping them open across exec(), though.
720 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
722 return error("cannot determine file descriptor flags");
723 fd_flag
|= FD_CLOEXEC
;
724 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
725 return error("cannot set FD_CLOEXEC");
727 /* Verify we recognize this pack file format. */
728 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
729 return error("file %s is far too short to be a packfile", p
->pack_name
);
730 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
731 return error("file %s is not a GIT packfile", p
->pack_name
);
732 if (!pack_version_ok(hdr
.hdr_version
))
733 return error("packfile %s is version %"PRIu32
" and not"
734 " supported (try upgrading GIT to a newer version)",
735 p
->pack_name
, ntohl(hdr
.hdr_version
));
737 /* Verify the pack matches its index. */
738 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
739 return error("packfile %s claims to have %"PRIu32
" objects"
740 " while index indicates %"PRIu32
" objects",
741 p
->pack_name
, ntohl(hdr
.hdr_entries
),
743 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
744 return error("end of packfile %s is unavailable", p
->pack_name
);
745 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
746 return error("packfile %s signature is unavailable", p
->pack_name
);
747 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
748 if (hashcmp(sha1
, idx_sha1
))
749 return error("packfile %s does not match index", p
->pack_name
);
753 static int open_packed_git(struct packed_git
*p
)
755 if (!open_packed_git_1(p
))
757 if (p
->pack_fd
!= -1) {
764 static int in_window(struct pack_window
*win
, off_t offset
)
766 /* We must promise at least 20 bytes (one hash) after the
767 * offset is available from this window, otherwise the offset
768 * is not actually in this window and a different window (which
769 * has that one hash excess) must be used. This is to support
770 * the object header and delta base parsing routines below.
772 off_t win_off
= win
->offset
;
773 return win_off
<= offset
774 && (offset
+ 20) <= (win_off
+ win
->len
);
777 unsigned char *use_pack(struct packed_git
*p
,
778 struct pack_window
**w_cursor
,
782 struct pack_window
*win
= *w_cursor
;
784 if (p
->pack_fd
== -1 && open_packed_git(p
))
785 die("packfile %s cannot be accessed", p
->pack_name
);
787 /* Since packfiles end in a hash of their content and it's
788 * pointless to ask for an offset into the middle of that
789 * hash, and the in_window function above wouldn't match
790 * don't allow an offset too close to the end of the file.
792 if (offset
> (p
->pack_size
- 20))
793 die("offset beyond end of packfile (truncated pack?)");
795 if (!win
|| !in_window(win
, offset
)) {
798 for (win
= p
->windows
; win
; win
= win
->next
) {
799 if (in_window(win
, offset
))
803 size_t window_align
= packed_git_window_size
/ 2;
805 win
= xcalloc(1, sizeof(*win
));
806 win
->offset
= (offset
/ window_align
) * window_align
;
807 len
= p
->pack_size
- win
->offset
;
808 if (len
> packed_git_window_size
)
809 len
= packed_git_window_size
;
810 win
->len
= (size_t)len
;
811 pack_mapped
+= win
->len
;
812 while (packed_git_limit
< pack_mapped
813 && unuse_one_window(p
, p
->pack_fd
))
815 win
->base
= xmmap(NULL
, win
->len
,
816 PROT_READ
, MAP_PRIVATE
,
817 p
->pack_fd
, win
->offset
);
818 if (win
->base
== MAP_FAILED
)
819 die("packfile %s cannot be mapped: %s",
824 if (pack_mapped
> peak_pack_mapped
)
825 peak_pack_mapped
= pack_mapped
;
826 if (pack_open_windows
> peak_pack_open_windows
)
827 peak_pack_open_windows
= pack_open_windows
;
828 win
->next
= p
->windows
;
832 if (win
!= *w_cursor
) {
833 win
->last_used
= pack_used_ctr
++;
837 offset
-= win
->offset
;
839 *left
= win
->len
- xsize_t(offset
);
840 return win
->base
+ offset
;
843 static struct packed_git
*alloc_packed_git(int extra
)
845 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
846 memset(p
, 0, sizeof(*p
));
851 static void try_to_free_pack_memory(size_t size
)
853 release_pack_memory(size
, -1);
856 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
858 static int have_set_try_to_free_routine
;
860 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
862 if (!have_set_try_to_free_routine
) {
863 have_set_try_to_free_routine
= 1;
864 set_try_to_free_routine(try_to_free_pack_memory
);
868 * Make sure a corresponding .pack file exists and that
869 * the index looks sane.
871 path_len
-= strlen(".idx");
876 memcpy(p
->pack_name
, path
, path_len
);
878 strcpy(p
->pack_name
+ path_len
, ".keep");
879 if (!access(p
->pack_name
, F_OK
))
882 strcpy(p
->pack_name
+ path_len
, ".pack");
883 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
888 /* ok, it looks sane as far as we can check without
889 * actually mapping the pack file.
891 p
->pack_size
= st
.st_size
;
892 p
->pack_local
= local
;
893 p
->mtime
= st
.st_mtime
;
894 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
899 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
901 const char *path
= sha1_pack_name(sha1
);
902 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
904 strcpy(p
->pack_name
, path
);
905 hashcpy(p
->sha1
, sha1
);
906 if (check_packed_git_idx(idx_path
, p
)) {
914 void install_packed_git(struct packed_git
*pack
)
916 pack
->next
= packed_git
;
920 static void prepare_packed_git_one(char *objdir
, int local
)
922 /* Ensure that this buffer is large enough so that we can
923 append "/pack/" without clobbering the stack even if
924 strlen(objdir) were PATH_MAX. */
925 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
930 sprintf(path
, "%s/pack", objdir
);
933 while (!dir
&& errno
== EMFILE
&& unuse_one_window(NULL
, -1))
937 error("unable to open object pack directory: %s: %s",
938 path
, strerror(errno
));
942 while ((de
= readdir(dir
)) != NULL
) {
943 int namelen
= strlen(de
->d_name
);
944 struct packed_git
*p
;
946 if (!has_extension(de
->d_name
, ".idx"))
949 if (len
+ namelen
+ 1 > sizeof(path
))
952 /* Don't reopen a pack we already have. */
953 strcpy(path
+ len
, de
->d_name
);
954 for (p
= packed_git
; p
; p
= p
->next
) {
955 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
960 /* See if it really is a valid .idx file with corresponding
961 * .pack file that we can map.
963 p
= add_packed_git(path
, len
+ namelen
, local
);
966 install_packed_git(p
);
971 static int sort_pack(const void *a_
, const void *b_
)
973 struct packed_git
*a
= *((struct packed_git
**)a_
);
974 struct packed_git
*b
= *((struct packed_git
**)b_
);
978 * Local packs tend to contain objects specific to our
979 * variant of the project than remote ones. In addition,
980 * remote ones could be on a network mounted filesystem.
981 * Favor local ones for these reasons.
983 st
= a
->pack_local
- b
->pack_local
;
988 * Younger packs tend to contain more recent objects,
989 * and more recent objects tend to get accessed more
992 if (a
->mtime
< b
->mtime
)
994 else if (a
->mtime
== b
->mtime
)
999 static void rearrange_packed_git(void)
1001 struct packed_git
**ary
, *p
;
1004 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1009 /* prepare an array of packed_git for easier sorting */
1010 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1011 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1014 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1016 /* link them back again */
1017 for (i
= 0; i
< n
- 1; i
++)
1018 ary
[i
]->next
= ary
[i
+ 1];
1019 ary
[n
- 1]->next
= NULL
;
1020 packed_git
= ary
[0];
1025 static int prepare_packed_git_run_once
= 0;
1026 void prepare_packed_git(void)
1028 struct alternate_object_database
*alt
;
1030 if (prepare_packed_git_run_once
)
1032 prepare_packed_git_one(get_object_directory(), 1);
1034 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1036 prepare_packed_git_one(alt
->base
, 0);
1037 alt
->name
[-1] = '/';
1039 rearrange_packed_git();
1040 prepare_packed_git_run_once
= 1;
1043 void reprepare_packed_git(void)
1046 prepare_packed_git_run_once
= 0;
1047 prepare_packed_git();
1050 static void mark_bad_packed_object(struct packed_git
*p
,
1051 const unsigned char *sha1
)
1054 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1055 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1057 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1058 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1059 p
->num_bad_objects
++;
1062 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1064 struct packed_git
*p
;
1067 for (p
= packed_git
; p
; p
= p
->next
)
1068 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1069 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1074 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1076 unsigned char real_sha1
[20];
1077 hash_sha1_file(map
, size
, type
, real_sha1
);
1078 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1081 static int git_open_noatime(const char *name
, struct packed_git
*p
)
1083 static int sha1_file_open_flag
= O_NOATIME
;
1086 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1090 /* Might the failure be insufficient file descriptors? */
1091 if (errno
== EMFILE
) {
1092 if (unuse_one_window(p
, -1))
1098 /* Might the failure be due to O_NOATIME? */
1099 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1100 sha1_file_open_flag
= 0;
1108 static int open_sha1_file(const unsigned char *sha1
)
1111 char *name
= sha1_file_name(sha1
);
1112 struct alternate_object_database
*alt
;
1114 fd
= git_open_noatime(name
, NULL
);
1120 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1122 fill_sha1_path(name
, sha1
);
1123 fd
= git_open_noatime(alt
->base
, NULL
);
1130 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1135 fd
= open_sha1_file(sha1
);
1140 if (!fstat(fd
, &st
)) {
1141 *size
= xsize_t(st
.st_size
);
1142 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1149 static int legacy_loose_object(unsigned char *map
)
1154 * Is it a zlib-compressed buffer? If so, the first byte
1155 * must be 0x78 (15-bit window size, deflated), and the
1156 * first 16-bit word is evenly divisible by 31
1158 word
= (map
[0] << 8) + map
[1];
1159 if (map
[0] == 0x78 && !(word
% 31))
1165 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1166 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1169 unsigned long size
, c
;
1170 unsigned long used
= 0;
1173 *type
= (c
>> 4) & 7;
1177 if (len
<= used
|| bitsizeof(long) <= shift
) {
1178 error("bad object header");
1182 size
+= (c
& 0x7f) << shift
;
1189 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1191 unsigned long size
, used
;
1192 static const char valid_loose_object_type
[8] = {
1194 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1195 0, /* "delta" and others are invalid in a loose object */
1197 enum object_type type
;
1199 /* Get the data stream */
1200 memset(stream
, 0, sizeof(*stream
));
1201 stream
->next_in
= map
;
1202 stream
->avail_in
= mapsize
;
1203 stream
->next_out
= buffer
;
1204 stream
->avail_out
= bufsiz
;
1206 if (legacy_loose_object(map
)) {
1207 git_inflate_init(stream
);
1208 return git_inflate(stream
, 0);
1213 * There used to be a second loose object header format which
1214 * was meant to mimic the in-pack format, allowing for direct
1215 * copy of the object data. This format turned up not to be
1216 * really worth it and we don't write it any longer. But we
1217 * can still read it.
1219 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1220 if (!used
|| !valid_loose_object_type
[type
])
1225 /* Set up the stream for the rest.. */
1226 stream
->next_in
= map
;
1227 stream
->avail_in
= mapsize
;
1228 git_inflate_init(stream
);
1230 /* And generate the fake traditional header */
1231 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1232 typename(type
), size
);
1236 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1238 int bytes
= strlen(buffer
) + 1;
1239 unsigned char *buf
= xmallocz(size
);
1243 n
= stream
->total_out
- bytes
;
1246 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1248 if (bytes
<= size
) {
1250 * The above condition must be (bytes <= size), not
1251 * (bytes < size). In other words, even though we
1252 * expect no more output and set avail_out to zer0,
1253 * the input zlib stream may have bytes that express
1254 * "this concludes the stream", and we *do* want to
1257 * Otherwise we would not be able to test that we
1258 * consumed all the input to reach the expected size;
1259 * we also want to check that zlib tells us that all
1260 * went well with status == Z_STREAM_END at the end.
1262 stream
->next_out
= buf
+ bytes
;
1263 stream
->avail_out
= size
- bytes
;
1264 while (status
== Z_OK
)
1265 status
= git_inflate(stream
, Z_FINISH
);
1267 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1268 git_inflate_end(stream
);
1273 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1274 else if (stream
->avail_in
)
1275 error("garbage at end of loose object '%s'",
1282 * We used to just use "sscanf()", but that's actually way
1283 * too permissive for what we want to check. So do an anal
1284 * object header parse by hand.
1286 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1293 * The type can be at most ten bytes (including the
1294 * terminating '\0' that we add), and is followed by
1303 if (i
>= sizeof(type
))
1309 * The length must follow immediately, and be in canonical
1310 * decimal format (ie "010" is not valid).
1312 size
= *hdr
++ - '0';
1317 unsigned long c
= *hdr
- '0';
1321 size
= size
* 10 + c
;
1327 * The length must be followed by a zero byte
1329 return *hdr
? -1 : type_from_string(type
);
1332 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1338 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1339 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1342 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1345 unsigned long get_size_from_delta(struct packed_git
*p
,
1346 struct pack_window
**w_curs
,
1349 const unsigned char *data
;
1350 unsigned char delta_head
[20], *in
;
1354 memset(&stream
, 0, sizeof(stream
));
1355 stream
.next_out
= delta_head
;
1356 stream
.avail_out
= sizeof(delta_head
);
1358 git_inflate_init(&stream
);
1360 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1361 stream
.next_in
= in
;
1362 st
= git_inflate(&stream
, Z_FINISH
);
1363 curpos
+= stream
.next_in
- in
;
1364 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1365 stream
.total_out
< sizeof(delta_head
));
1366 git_inflate_end(&stream
);
1367 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1368 error("delta data unpack-initial failed");
1372 /* Examine the initial part of the delta to figure out
1377 /* ignore base size */
1378 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1380 /* Read the result size */
1381 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1384 static off_t
get_delta_base(struct packed_git
*p
,
1385 struct pack_window
**w_curs
,
1387 enum object_type type
,
1388 off_t delta_obj_offset
)
1390 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1393 /* use_pack() assured us we have [base_info, base_info + 20)
1394 * as a range that we can look at without walking off the
1395 * end of the mapped window. Its actually the hash size
1396 * that is assured. An OFS_DELTA longer than the hash size
1397 * is stupid, as then a REF_DELTA would be smaller to store.
1399 if (type
== OBJ_OFS_DELTA
) {
1401 unsigned char c
= base_info
[used
++];
1402 base_offset
= c
& 127;
1405 if (!base_offset
|| MSB(base_offset
, 7))
1406 return 0; /* overflow */
1407 c
= base_info
[used
++];
1408 base_offset
= (base_offset
<< 7) + (c
& 127);
1410 base_offset
= delta_obj_offset
- base_offset
;
1411 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1412 return 0; /* out of bound */
1414 } else if (type
== OBJ_REF_DELTA
) {
1415 /* The base entry _must_ be in the same pack */
1416 base_offset
= find_pack_entry_one(base_info
, p
);
1419 die("I am totally screwed");
1423 /* forward declaration for a mutually recursive function */
1424 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1425 unsigned long *sizep
);
1427 static int packed_delta_info(struct packed_git
*p
,
1428 struct pack_window
**w_curs
,
1430 enum object_type type
,
1432 unsigned long *sizep
)
1436 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1439 type
= packed_object_info(p
, base_offset
, NULL
);
1440 if (type
<= OBJ_NONE
) {
1441 struct revindex_entry
*revidx
;
1442 const unsigned char *base_sha1
;
1443 revidx
= find_pack_revindex(p
, base_offset
);
1446 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1447 mark_bad_packed_object(p
, base_sha1
);
1448 type
= sha1_object_info(base_sha1
, NULL
);
1449 if (type
<= OBJ_NONE
)
1453 /* We choose to only get the type of the base object and
1454 * ignore potentially corrupt pack file that expects the delta
1455 * based on a base with a wrong size. This saves tons of
1459 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1467 static int unpack_object_header(struct packed_git
*p
,
1468 struct pack_window
**w_curs
,
1470 unsigned long *sizep
)
1472 unsigned char *base
;
1475 enum object_type type
;
1477 /* use_pack() assures us we have [base, base + 20) available
1478 * as a range that we can look at at. (Its actually the hash
1479 * size that is assured.) With our object header encoding
1480 * the maximum deflated object size is 2^137, which is just
1481 * insane, so we know won't exceed what we have been given.
1483 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1484 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1493 const char *packed_object_info_detail(struct packed_git
*p
,
1495 unsigned long *size
,
1496 unsigned long *store_size
,
1497 unsigned int *delta_chain_length
,
1498 unsigned char *base_sha1
)
1500 struct pack_window
*w_curs
= NULL
;
1502 unsigned long dummy
;
1503 unsigned char *next_sha1
;
1504 enum object_type type
;
1505 struct revindex_entry
*revidx
;
1507 *delta_chain_length
= 0;
1508 curpos
= obj_offset
;
1509 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1511 revidx
= find_pack_revindex(p
, obj_offset
);
1512 *store_size
= revidx
[1].offset
- obj_offset
;
1517 die("pack %s contains unknown object type %d",
1518 p
->pack_name
, type
);
1523 unuse_pack(&w_curs
);
1524 return typename(type
);
1526 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1528 die("pack %s contains bad delta base reference of type %s",
1529 p
->pack_name
, typename(type
));
1530 if (*delta_chain_length
== 0) {
1531 revidx
= find_pack_revindex(p
, obj_offset
);
1532 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1536 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1537 if (*delta_chain_length
== 0)
1538 hashcpy(base_sha1
, next_sha1
);
1539 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1542 (*delta_chain_length
)++;
1543 curpos
= obj_offset
;
1544 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1548 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1549 unsigned long *sizep
)
1551 struct pack_window
*w_curs
= NULL
;
1553 off_t curpos
= obj_offset
;
1554 enum object_type type
;
1556 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1561 type
= packed_delta_info(p
, &w_curs
, curpos
,
1562 type
, obj_offset
, sizep
);
1572 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1573 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1576 unuse_pack(&w_curs
);
1580 static void *unpack_compressed_entry(struct packed_git
*p
,
1581 struct pack_window
**w_curs
,
1587 unsigned char *buffer
, *in
;
1589 buffer
= xmallocz(size
);
1590 memset(&stream
, 0, sizeof(stream
));
1591 stream
.next_out
= buffer
;
1592 stream
.avail_out
= size
+ 1;
1594 git_inflate_init(&stream
);
1596 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1597 stream
.next_in
= in
;
1598 st
= git_inflate(&stream
, Z_FINISH
);
1599 if (!stream
.avail_out
)
1600 break; /* the payload is larger than it should be */
1601 curpos
+= stream
.next_in
- in
;
1602 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1603 git_inflate_end(&stream
);
1604 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1612 #define MAX_DELTA_CACHE (256)
1614 static size_t delta_base_cached
;
1616 static struct delta_base_cache_lru_list
{
1617 struct delta_base_cache_lru_list
*prev
;
1618 struct delta_base_cache_lru_list
*next
;
1619 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1621 static struct delta_base_cache_entry
{
1622 struct delta_base_cache_lru_list lru
;
1624 struct packed_git
*p
;
1627 enum object_type type
;
1628 } delta_base_cache
[MAX_DELTA_CACHE
];
1630 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1634 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1635 hash
+= (hash
>> 8) + (hash
>> 16);
1636 return hash
% MAX_DELTA_CACHE
;
1639 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1640 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1643 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1644 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1647 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1648 return unpack_entry(p
, base_offset
, type
, base_size
);
1652 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1653 ent
->lru
.prev
->next
= ent
->lru
.next
;
1654 delta_base_cached
-= ent
->size
;
1656 ret
= xmemdupz(ent
->data
, ent
->size
);
1659 *base_size
= ent
->size
;
1663 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1668 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1669 ent
->lru
.prev
->next
= ent
->lru
.next
;
1670 delta_base_cached
-= ent
->size
;
1674 void clear_delta_base_cache(void)
1677 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1678 release_delta_base_cache(&delta_base_cache
[p
]);
1681 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1682 void *base
, unsigned long base_size
, enum object_type type
)
1684 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1685 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1686 struct delta_base_cache_lru_list
*lru
;
1688 release_delta_base_cache(ent
);
1689 delta_base_cached
+= base_size
;
1691 for (lru
= delta_base_cache_lru
.next
;
1692 delta_base_cached
> delta_base_cache_limit
1693 && lru
!= &delta_base_cache_lru
;
1695 struct delta_base_cache_entry
*f
= (void *)lru
;
1696 if (f
->type
== OBJ_BLOB
)
1697 release_delta_base_cache(f
);
1699 for (lru
= delta_base_cache_lru
.next
;
1700 delta_base_cached
> delta_base_cache_limit
1701 && lru
!= &delta_base_cache_lru
;
1703 struct delta_base_cache_entry
*f
= (void *)lru
;
1704 release_delta_base_cache(f
);
1708 ent
->base_offset
= base_offset
;
1711 ent
->size
= base_size
;
1712 ent
->lru
.next
= &delta_base_cache_lru
;
1713 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1714 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1715 delta_base_cache_lru
.prev
= &ent
->lru
;
1718 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1719 unsigned long *size
);
1721 static void *unpack_delta_entry(struct packed_git
*p
,
1722 struct pack_window
**w_curs
,
1724 unsigned long delta_size
,
1726 enum object_type
*type
,
1727 unsigned long *sizep
)
1729 void *delta_data
, *result
, *base
;
1730 unsigned long base_size
;
1733 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1735 error("failed to validate delta base reference "
1736 "at offset %"PRIuMAX
" from %s",
1737 (uintmax_t)curpos
, p
->pack_name
);
1741 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1744 * We're probably in deep shit, but let's try to fetch
1745 * the required base anyway from another pack or loose.
1746 * This is costly but should happen only in the presence
1747 * of a corrupted pack, and is better than failing outright.
1749 struct revindex_entry
*revidx
;
1750 const unsigned char *base_sha1
;
1751 revidx
= find_pack_revindex(p
, base_offset
);
1754 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1755 error("failed to read delta base object %s"
1756 " at offset %"PRIuMAX
" from %s",
1757 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1759 mark_bad_packed_object(p
, base_sha1
);
1760 base
= read_object(base_sha1
, type
, &base_size
);
1765 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1767 error("failed to unpack compressed delta "
1768 "at offset %"PRIuMAX
" from %s",
1769 (uintmax_t)curpos
, p
->pack_name
);
1773 result
= patch_delta(base
, base_size
,
1774 delta_data
, delta_size
,
1777 die("failed to apply delta");
1779 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1783 int do_check_packed_object_crc
;
1785 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1786 enum object_type
*type
, unsigned long *sizep
)
1788 struct pack_window
*w_curs
= NULL
;
1789 off_t curpos
= obj_offset
;
1792 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1793 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1794 unsigned long len
= revidx
[1].offset
- obj_offset
;
1795 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1796 const unsigned char *sha1
=
1797 nth_packed_object_sha1(p
, revidx
->nr
);
1798 error("bad packed object CRC for %s",
1800 mark_bad_packed_object(p
, sha1
);
1801 unuse_pack(&w_curs
);
1806 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1810 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1811 obj_offset
, type
, sizep
);
1817 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1821 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1822 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1824 unuse_pack(&w_curs
);
1828 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1831 const unsigned char *index
= p
->index_data
;
1833 if (open_pack_index(p
))
1835 index
= p
->index_data
;
1837 if (n
>= p
->num_objects
)
1840 if (p
->index_version
== 1) {
1841 return index
+ 24 * n
+ 4;
1844 return index
+ 20 * n
;
1848 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1850 const unsigned char *index
= p
->index_data
;
1852 if (p
->index_version
== 1) {
1853 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1856 index
+= 8 + p
->num_objects
* (20 + 4);
1857 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1858 if (!(off
& 0x80000000))
1860 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1861 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1862 ntohl(*((uint32_t *)(index
+ 4)));
1866 off_t
find_pack_entry_one(const unsigned char *sha1
,
1867 struct packed_git
*p
)
1869 const uint32_t *level1_ofs
= p
->index_data
;
1870 const unsigned char *index
= p
->index_data
;
1871 unsigned hi
, lo
, stride
;
1872 static int use_lookup
= -1;
1873 static int debug_lookup
= -1;
1875 if (debug_lookup
< 0)
1876 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1879 if (open_pack_index(p
))
1881 level1_ofs
= p
->index_data
;
1882 index
= p
->index_data
;
1884 if (p
->index_version
> 1) {
1889 hi
= ntohl(level1_ofs
[*sha1
]);
1890 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1891 if (p
->index_version
> 1) {
1899 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
1900 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1903 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1905 int pos
= sha1_entry_pos(index
, stride
, 0,
1906 lo
, hi
, p
->num_objects
, sha1
);
1909 return nth_packed_object_offset(p
, pos
);
1913 unsigned mi
= (lo
+ hi
) / 2;
1914 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1917 printf("lo %u hi %u rg %u mi %u\n",
1918 lo
, hi
, hi
- lo
, mi
);
1920 return nth_packed_object_offset(p
, mi
);
1929 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
1931 static struct packed_git
*last_found
= (void *)1;
1932 struct packed_git
*p
;
1935 prepare_packed_git();
1938 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1941 if (p
->num_bad_objects
) {
1943 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1944 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1948 offset
= find_pack_entry_one(sha1
, p
);
1951 * We are about to tell the caller where they can
1952 * locate the requested object. We better make
1953 * sure the packfile is still here and can be
1954 * accessed before supplying that answer, as
1955 * it may have been deleted since the index
1958 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1959 error("packfile %s cannot be accessed", p
->pack_name
);
1964 hashcpy(e
->sha1
, sha1
);
1970 if (p
== last_found
)
1974 if (p
== last_found
)
1980 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1981 struct packed_git
*packs
)
1983 struct packed_git
*p
;
1985 for (p
= packs
; p
; p
= p
->next
) {
1986 if (find_pack_entry_one(sha1
, p
))
1993 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1996 unsigned long mapsize
, size
;
2001 map
= map_sha1_file(sha1
, &mapsize
);
2003 return error("unable to find %s", sha1_to_hex(sha1
));
2004 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2005 status
= error("unable to unpack %s header",
2007 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2008 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2011 git_inflate_end(&stream
);
2012 munmap(map
, mapsize
);
2016 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2018 struct cached_object
*co
;
2019 struct pack_entry e
;
2022 co
= find_cached_object(sha1
);
2029 if (!find_pack_entry(sha1
, &e
)) {
2030 /* Most likely it's a loose object. */
2031 status
= sha1_loose_object_info(sha1
, sizep
);
2035 /* Not a loose object; someone else may have just packed it. */
2036 reprepare_packed_git();
2037 if (!find_pack_entry(sha1
, &e
))
2041 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
2043 mark_bad_packed_object(e
.p
, sha1
);
2044 status
= sha1_object_info(sha1
, sizep
);
2050 static void *read_packed_sha1(const unsigned char *sha1
,
2051 enum object_type
*type
, unsigned long *size
)
2053 struct pack_entry e
;
2056 if (!find_pack_entry(sha1
, &e
))
2058 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2061 * We're probably in deep shit, but let's try to fetch
2062 * the required object anyway from another pack or loose.
2063 * This should happen only in the presence of a corrupted
2064 * pack, and is better than failing outright.
2066 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2067 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2068 mark_bad_packed_object(e
.p
, sha1
);
2069 data
= read_object(sha1
, type
, size
);
2074 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2075 unsigned char *sha1
)
2077 struct cached_object
*co
;
2079 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2080 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2082 if (cached_object_alloc
<= cached_object_nr
) {
2083 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2084 cached_objects
= xrealloc(cached_objects
,
2085 sizeof(*cached_objects
) *
2086 cached_object_alloc
);
2088 co
= &cached_objects
[cached_object_nr
++];
2091 co
->buf
= xmalloc(len
);
2092 memcpy(co
->buf
, buf
, len
);
2093 hashcpy(co
->sha1
, sha1
);
2097 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2098 unsigned long *size
)
2100 unsigned long mapsize
;
2102 struct cached_object
*co
;
2104 co
= find_cached_object(sha1
);
2108 return xmemdupz(co
->buf
, co
->size
);
2111 buf
= read_packed_sha1(sha1
, type
, size
);
2114 map
= map_sha1_file(sha1
, &mapsize
);
2116 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2117 munmap(map
, mapsize
);
2120 reprepare_packed_git();
2121 return read_packed_sha1(sha1
, type
, size
);
2125 * This function dies on corrupt objects; the callers who want to
2126 * deal with them should arrange to call read_object() and give error
2127 * messages themselves.
2129 void *read_sha1_file_repl(const unsigned char *sha1
,
2130 enum object_type
*type
,
2131 unsigned long *size
,
2132 const unsigned char **replacement
)
2134 const unsigned char *repl
= lookup_replace_object(sha1
);
2137 const struct packed_git
*p
;
2140 data
= read_object(repl
, type
, size
);
2143 *replacement
= repl
;
2147 if (errno
&& errno
!= ENOENT
)
2148 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2150 /* die if we replaced an object with one that does not exist */
2152 die("replacement %s not found for %s",
2153 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2155 if (has_loose_object(repl
)) {
2156 path
= sha1_file_name(sha1
);
2157 die("loose object %s (stored in %s) is corrupt",
2158 sha1_to_hex(repl
), path
);
2161 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2162 die("packed object %s (stored in %s) is corrupt",
2163 sha1_to_hex(repl
), p
->pack_name
);
2168 void *read_object_with_reference(const unsigned char *sha1
,
2169 const char *required_type_name
,
2170 unsigned long *size
,
2171 unsigned char *actual_sha1_return
)
2173 enum object_type type
, required_type
;
2175 unsigned long isize
;
2176 unsigned char actual_sha1
[20];
2178 required_type
= type_from_string(required_type_name
);
2179 hashcpy(actual_sha1
, sha1
);
2181 int ref_length
= -1;
2182 const char *ref_type
= NULL
;
2184 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2187 if (type
== required_type
) {
2189 if (actual_sha1_return
)
2190 hashcpy(actual_sha1_return
, actual_sha1
);
2193 /* Handle references */
2194 else if (type
== OBJ_COMMIT
)
2196 else if (type
== OBJ_TAG
)
2197 ref_type
= "object ";
2202 ref_length
= strlen(ref_type
);
2204 if (ref_length
+ 40 > isize
||
2205 memcmp(buffer
, ref_type
, ref_length
) ||
2206 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2211 /* Now we have the ID of the referred-to object in
2212 * actual_sha1. Check again. */
2216 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2217 const char *type
, unsigned char *sha1
,
2218 char *hdr
, int *hdrlen
)
2222 /* Generate the header */
2223 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2227 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2228 git_SHA1_Update(&c
, buf
, len
);
2229 git_SHA1_Final(sha1
, &c
);
2233 * Move the just written object into its final resting place.
2234 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2235 * "moving" is only a part of what it does, when no patch between
2236 * master to pu changes the call sites of this function.
2238 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2242 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2244 else if (link(tmpfile
, filename
))
2248 * Coda hack - coda doesn't like cross-directory links,
2249 * so we fall back to a rename, which will mean that it
2250 * won't be able to check collisions, but that's not a
2253 * The same holds for FAT formatted media.
2255 * When this succeeds, we just return. We have nothing
2258 if (ret
&& ret
!= EEXIST
) {
2260 if (!rename(tmpfile
, filename
))
2264 unlink_or_warn(tmpfile
);
2266 if (ret
!= EEXIST
) {
2267 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2269 /* FIXME!!! Collision check here ? */
2273 if (adjust_shared_perm(filename
))
2274 return error("unable to set permission to '%s'", filename
);
2278 static int write_buffer(int fd
, const void *buf
, size_t len
)
2280 if (write_in_full(fd
, buf
, len
) < 0)
2281 return error("file write error (%s)", strerror(errno
));
2285 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2286 unsigned char *sha1
)
2290 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2294 /* Finalize a file on disk, and close it. */
2295 static void close_sha1_file(int fd
)
2297 if (fsync_object_files
)
2298 fsync_or_die(fd
, "sha1 file");
2300 die_errno("error when closing sha1 file");
2303 /* Size of directory component, including the ending '/' */
2304 static inline int directory_size(const char *filename
)
2306 const char *s
= strrchr(filename
, '/');
2309 return s
- filename
+ 1;
2313 * This creates a temporary file in the same directory as the final
2316 * We want to avoid cross-directory filename renames, because those
2317 * can have problems on various filesystems (FAT, NFS, Coda).
2319 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2321 int fd
, dirlen
= directory_size(filename
);
2323 if (dirlen
+ 20 > bufsiz
) {
2324 errno
= ENAMETOOLONG
;
2327 memcpy(buffer
, filename
, dirlen
);
2328 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2329 fd
= git_mkstemp_mode(buffer
, 0444);
2330 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2331 /* Make sure the directory exists */
2332 memcpy(buffer
, filename
, dirlen
);
2333 buffer
[dirlen
-1] = 0;
2334 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2338 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2339 fd
= git_mkstemp_mode(buffer
, 0444);
2344 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2345 const void *buf
, unsigned long len
, time_t mtime
)
2348 unsigned char compressed
[4096];
2351 unsigned char parano_sha1
[20];
2353 static char tmpfile
[PATH_MAX
];
2355 filename
= sha1_file_name(sha1
);
2356 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2357 while (fd
< 0 && errno
== EMFILE
&& unuse_one_window(NULL
, -1))
2358 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2360 if (errno
== EACCES
)
2361 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2363 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2367 memset(&stream
, 0, sizeof(stream
));
2368 deflateInit(&stream
, zlib_compression_level
);
2369 stream
.next_out
= compressed
;
2370 stream
.avail_out
= sizeof(compressed
);
2373 /* First header.. */
2374 stream
.next_in
= (unsigned char *)hdr
;
2375 stream
.avail_in
= hdrlen
;
2376 while (deflate(&stream
, 0) == Z_OK
)
2378 git_SHA1_Update(&c
, hdr
, hdrlen
);
2380 /* Then the data itself.. */
2381 stream
.next_in
= (void *)buf
;
2382 stream
.avail_in
= len
;
2384 unsigned char *in0
= stream
.next_in
;
2385 ret
= deflate(&stream
, Z_FINISH
);
2386 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2387 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2388 die("unable to write sha1 file");
2389 stream
.next_out
= compressed
;
2390 stream
.avail_out
= sizeof(compressed
);
2391 } while (ret
== Z_OK
);
2393 if (ret
!= Z_STREAM_END
)
2394 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2395 ret
= deflateEnd(&stream
);
2397 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2398 git_SHA1_Final(parano_sha1
, &c
);
2399 if (hashcmp(sha1
, parano_sha1
) != 0)
2400 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2402 close_sha1_file(fd
);
2407 utb
.modtime
= mtime
;
2408 if (utime(tmpfile
, &utb
) < 0)
2409 warning("failed utime() on %s: %s",
2410 tmpfile
, strerror(errno
));
2413 return move_temp_to_file(tmpfile
, filename
);
2416 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2418 unsigned char sha1
[20];
2422 /* Normally if we have it in the pack then we do not bother writing
2423 * it out into .git/objects/??/?{38} file.
2425 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2427 hashcpy(returnsha1
, sha1
);
2428 if (has_sha1_file(sha1
))
2430 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2433 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2437 enum object_type type
;
2442 if (has_loose_object(sha1
))
2444 buf
= read_packed_sha1(sha1
, &type
, &len
);
2446 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2447 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2448 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2454 int has_pack_index(const unsigned char *sha1
)
2457 if (stat(sha1_pack_index_name(sha1
), &st
))
2462 int has_sha1_pack(const unsigned char *sha1
)
2464 struct pack_entry e
;
2465 return find_pack_entry(sha1
, &e
);
2468 int has_sha1_file(const unsigned char *sha1
)
2470 struct pack_entry e
;
2472 if (find_pack_entry(sha1
, &e
))
2474 return has_loose_object(sha1
);
2477 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2478 int write_object
, enum object_type type
, const char *path
)
2480 int ret
, re_allocated
= 0;
2486 * Convert blobs to git internal format
2488 if ((type
== OBJ_BLOB
) && path
) {
2489 struct strbuf nbuf
= STRBUF_INIT
;
2490 if (convert_to_git(path
, buf
, size
, &nbuf
,
2491 write_object
? safe_crlf
: 0)) {
2492 buf
= strbuf_detach(&nbuf
, &size
);
2498 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2500 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2506 #define SMALL_FILE_SIZE (32*1024)
2508 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2509 enum object_type type
, const char *path
)
2512 size_t size
= xsize_t(st
->st_size
);
2514 if (!S_ISREG(st
->st_mode
)) {
2515 struct strbuf sbuf
= STRBUF_INIT
;
2516 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2517 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2521 strbuf_release(&sbuf
);
2523 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
);
2524 } else if (size
<= SMALL_FILE_SIZE
) {
2525 char *buf
= xmalloc(size
);
2526 if (size
== read_in_full(fd
, buf
, size
))
2527 ret
= index_mem(sha1
, buf
, size
, write_object
, type
,
2530 ret
= error("short read %s", strerror(errno
));
2533 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2534 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
);
2541 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2544 struct strbuf sb
= STRBUF_INIT
;
2546 switch (st
->st_mode
& S_IFMT
) {
2548 fd
= open(path
, O_RDONLY
);
2550 return error("open(\"%s\"): %s", path
,
2552 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2553 return error("%s: failed to insert into database",
2557 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2558 char *errstr
= strerror(errno
);
2559 return error("readlink(\"%s\"): %s", path
,
2563 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2564 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2565 return error("%s: failed to insert into database",
2567 strbuf_release(&sb
);
2570 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2572 return error("%s: unsupported file type", path
);
2577 int read_pack_header(int fd
, struct pack_header
*header
)
2579 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2580 /* "eof before pack header was fully read" */
2581 return PH_ERROR_EOF
;
2583 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2584 /* "protocol error (pack signature mismatch detected)" */
2585 return PH_ERROR_PACK_SIGNATURE
;
2586 if (!pack_version_ok(header
->hdr_version
))
2587 /* "protocol error (pack version unsupported)" */
2588 return PH_ERROR_PROTOCOL
;
2592 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2594 enum object_type type
= sha1_object_info(sha1
, NULL
);
2596 die("%s is not a valid object", sha1_to_hex(sha1
));
2598 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),