4 #include "repository.h"
10 #include "streaming.h"
11 #include "sha1-lookup.h"
15 #include "tree-walk.h"
17 #include "object-store.h"
20 char *odb_pack_name(struct strbuf
*buf
,
21 const unsigned char *sha1
,
25 strbuf_addf(buf
, "%s/pack/pack-%s.%s", get_object_directory(),
26 sha1_to_hex(sha1
), ext
);
30 char *sha1_pack_name(const unsigned char *sha1
)
32 static struct strbuf buf
= STRBUF_INIT
;
33 return odb_pack_name(&buf
, sha1
, "pack");
36 char *sha1_pack_index_name(const unsigned char *sha1
)
38 static struct strbuf buf
= STRBUF_INIT
;
39 return odb_pack_name(&buf
, sha1
, "idx");
42 static unsigned int pack_used_ctr
;
43 static unsigned int pack_mmap_calls
;
44 static unsigned int peak_pack_open_windows
;
45 static unsigned int pack_open_windows
;
46 static unsigned int pack_open_fds
;
47 static unsigned int pack_max_fds
;
48 static size_t peak_pack_mapped
;
49 static size_t pack_mapped
;
51 #define SZ_FMT PRIuMAX
52 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
54 void pack_report(void)
57 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
58 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
59 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
60 sz_fmt(getpagesize()),
61 sz_fmt(packed_git_window_size
),
62 sz_fmt(packed_git_limit
));
64 "pack_report: pack_used_ctr = %10u\n"
65 "pack_report: pack_mmap_calls = %10u\n"
66 "pack_report: pack_open_windows = %10u / %10u\n"
67 "pack_report: pack_mapped = "
68 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
71 pack_open_windows
, peak_pack_open_windows
,
72 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
76 * Open and mmap the index file at path, perform a couple of
77 * consistency checks, then record its information to p. Return 0 on
80 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
84 int fd
= git_open(path
), ret
;
86 const unsigned int hashsz
= the_hash_algo
->rawsz
;
94 idx_size
= xsize_t(st
.st_size
);
95 if (idx_size
< 4 * 256 + hashsz
+ hashsz
) {
97 return error("index file %s is too small", path
);
99 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
102 ret
= load_idx(path
, hashsz
, idx_map
, idx_size
, p
);
105 munmap(idx_map
, idx_size
);
110 int load_idx(const char *path
, const unsigned int hashsz
, void *idx_map
,
111 size_t idx_size
, struct packed_git
*p
)
113 struct pack_idx_header
*hdr
= idx_map
;
114 uint32_t version
, nr
, i
, *index
;
116 if (idx_size
< 4 * 256 + hashsz
+ hashsz
)
117 return error("index file %s is too small", path
);
119 return error("empty data");
121 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
122 version
= ntohl(hdr
->idx_version
);
123 if (version
< 2 || version
> 2)
124 return error("index file %s is version %"PRIu32
125 " and is not supported by this binary"
126 " (try upgrading GIT to a newer version)",
134 index
+= 2; /* skip index header */
135 for (i
= 0; i
< 256; i
++) {
136 uint32_t n
= ntohl(index
[i
]);
138 return error("non-monotonic index %s", path
);
145 * - 256 index entries 4 bytes each
146 * - 24-byte entries * nr (object ID + 4-byte offset)
147 * - hash of the packfile
150 if (idx_size
!= 4 * 256 + nr
* (hashsz
+ 4) + hashsz
+ hashsz
)
151 return error("wrong index v1 file size in %s", path
);
152 } else if (version
== 2) {
155 * - 8 bytes of header
156 * - 256 index entries 4 bytes each
157 * - object ID entry * nr
158 * - 4-byte crc entry * nr
159 * - 4-byte offset entry * nr
160 * - hash of the packfile
162 * And after the 4-byte offset table might be a
163 * variable sized table containing 8-byte entries
164 * for offsets larger than 2^31.
166 unsigned long min_size
= 8 + 4*256 + nr
*(hashsz
+ 4 + 4) + hashsz
+ hashsz
;
167 unsigned long max_size
= min_size
;
169 max_size
+= (nr
- 1)*8;
170 if (idx_size
< min_size
|| idx_size
> max_size
)
171 return error("wrong index v2 file size in %s", path
);
172 if (idx_size
!= min_size
&&
174 * make sure we can deal with large pack offsets.
175 * 31-bit signed offset won't be enough, neither
176 * 32-bit unsigned one will be.
178 (sizeof(off_t
) <= 4))
179 return error("pack too large for current definition of off_t in %s", path
);
182 p
->index_version
= version
;
183 p
->index_data
= idx_map
;
184 p
->index_size
= idx_size
;
189 int open_pack_index(struct packed_git
*p
)
198 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
199 BUG("pack_name does not end in .pack");
200 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
201 ret
= check_packed_git_idx(idx_name
, p
);
206 uint32_t get_pack_fanout(struct packed_git
*p
, uint32_t value
)
208 const uint32_t *level1_ofs
= p
->index_data
;
211 if (open_pack_index(p
))
213 level1_ofs
= p
->index_data
;
216 if (p
->index_version
> 1) {
220 return ntohl(level1_ofs
[value
]);
223 static struct packed_git
*alloc_packed_git(int extra
)
225 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
226 memset(p
, 0, sizeof(*p
));
231 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
233 const char *path
= sha1_pack_name(sha1
);
234 size_t alloc
= st_add(strlen(path
), 1);
235 struct packed_git
*p
= alloc_packed_git(alloc
);
237 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
238 hashcpy(p
->sha1
, sha1
);
239 if (check_packed_git_idx(idx_path
, p
)) {
247 static void scan_windows(struct packed_git
*p
,
248 struct packed_git
**lru_p
,
249 struct pack_window
**lru_w
,
250 struct pack_window
**lru_l
)
252 struct pack_window
*w
, *w_l
;
254 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
256 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
266 static int unuse_one_window(struct packed_git
*current
)
268 struct packed_git
*p
, *lru_p
= NULL
;
269 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
272 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
273 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
)
274 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
276 munmap(lru_w
->base
, lru_w
->len
);
277 pack_mapped
-= lru_w
->len
;
279 lru_l
->next
= lru_w
->next
;
281 lru_p
->windows
= lru_w
->next
;
289 void release_pack_memory(size_t need
)
291 size_t cur
= pack_mapped
;
292 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
296 void close_pack_windows(struct packed_git
*p
)
299 struct pack_window
*w
= p
->windows
;
302 die("pack '%s' still has open windows to it",
304 munmap(w
->base
, w
->len
);
305 pack_mapped
-= w
->len
;
307 p
->windows
= w
->next
;
312 static int close_pack_fd(struct packed_git
*p
)
324 void close_pack_index(struct packed_git
*p
)
327 munmap((void *)p
->index_data
, p
->index_size
);
328 p
->index_data
= NULL
;
332 void close_pack(struct packed_git
*p
)
334 close_pack_windows(p
);
339 void close_all_packs(struct raw_object_store
*o
)
341 struct packed_git
*p
;
343 for (p
= o
->packed_git
; p
; p
= p
->next
)
345 BUG("want to close pack marked 'do-not-close'");
349 if (o
->multi_pack_index
) {
350 close_midx(o
->multi_pack_index
);
351 o
->multi_pack_index
= NULL
;
356 * The LRU pack is the one with the oldest MRU window, preferring packs
357 * with no used windows, or the oldest mtime if it has no windows allocated.
359 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
361 struct pack_window
*w
, *this_mru_w
;
362 int has_windows_inuse
= 0;
365 * Reject this pack if it has windows and the previously selected
366 * one does not. If this pack does not have windows, reject
367 * it if the pack file is newer than the previously selected one.
369 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
372 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
374 * Reject this pack if any of its windows are in use,
375 * but the previously selected pack did not have any
376 * inuse windows. Otherwise, record that this pack
377 * has windows in use.
380 if (*accept_windows_inuse
)
381 has_windows_inuse
= 1;
386 if (w
->last_used
> this_mru_w
->last_used
)
390 * Reject this pack if it has windows that have been
391 * used more recently than the previously selected pack.
392 * If the previously selected pack had windows inuse and
393 * we have not encountered a window in this pack that is
394 * inuse, skip this check since we prefer a pack with no
395 * inuse windows to one that has inuse windows.
397 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
398 this_mru_w
->last_used
> (*mru_w
)->last_used
)
407 *accept_windows_inuse
= has_windows_inuse
;
410 static int close_one_pack(void)
412 struct packed_git
*p
, *lru_p
= NULL
;
413 struct pack_window
*mru_w
= NULL
;
414 int accept_windows_inuse
= 1;
416 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
417 if (p
->pack_fd
== -1)
419 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
423 return close_pack_fd(lru_p
);
428 static unsigned int get_max_fd_limit(void)
434 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
441 long open_max
= sysconf(_SC_OPEN_MAX
);
445 * Otherwise, we got -1 for one of the two
448 * (1) sysconf() did not understand _SC_OPEN_MAX
449 * and signaled an error with -1; or
450 * (2) sysconf() said there is no limit.
452 * We _could_ clear errno before calling sysconf() to
453 * tell these two cases apart and return a huge number
454 * in the latter case to let the caller cap it to a
455 * value that is not so selfish, but letting the
456 * fallback OPEN_MAX codepath take care of these cases
465 return 1; /* see the caller ;-) */
470 * Do not call this directly as this leaks p->pack_fd on error return;
471 * call open_packed_git() instead.
473 static int open_packed_git_1(struct packed_git
*p
)
476 struct pack_header hdr
;
477 unsigned char hash
[GIT_MAX_RAWSZ
];
478 unsigned char *idx_hash
;
481 const unsigned hashsz
= the_hash_algo
->rawsz
;
483 if (!p
->index_data
) {
484 struct multi_pack_index
*m
;
485 const char *pack_name
= strrchr(p
->pack_name
, '/');
487 for (m
= the_repository
->objects
->multi_pack_index
;
489 if (midx_contains_pack(m
, pack_name
))
493 if (!m
&& open_pack_index(p
))
494 return error("packfile %s index unavailable", p
->pack_name
);
498 unsigned int max_fds
= get_max_fd_limit();
500 /* Save 3 for stdin/stdout/stderr, 22 for work */
502 pack_max_fds
= max_fds
- 25;
507 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
510 p
->pack_fd
= git_open(p
->pack_name
);
511 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
515 /* If we created the struct before we had the pack we lack size. */
517 if (!S_ISREG(st
.st_mode
))
518 return error("packfile %s not a regular file", p
->pack_name
);
519 p
->pack_size
= st
.st_size
;
520 } else if (p
->pack_size
!= st
.st_size
)
521 return error("packfile %s size changed", p
->pack_name
);
523 /* We leave these file descriptors open with sliding mmap;
524 * there is no point keeping them open across exec(), though.
526 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
528 return error("cannot determine file descriptor flags");
529 fd_flag
|= FD_CLOEXEC
;
530 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
531 return error("cannot set FD_CLOEXEC");
533 /* Verify we recognize this pack file format. */
534 read_result
= read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
));
536 return error_errno("error reading from %s", p
->pack_name
);
537 if (read_result
!= sizeof(hdr
))
538 return error("file %s is far too short to be a packfile", p
->pack_name
);
539 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
540 return error("file %s is not a GIT packfile", p
->pack_name
);
541 if (!pack_version_ok(hdr
.hdr_version
))
542 return error("packfile %s is version %"PRIu32
" and not"
543 " supported (try upgrading GIT to a newer version)",
544 p
->pack_name
, ntohl(hdr
.hdr_version
));
546 /* Skip index checking if in multi-pack-index */
550 /* Verify the pack matches its index. */
551 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
552 return error("packfile %s claims to have %"PRIu32
" objects"
553 " while index indicates %"PRIu32
" objects",
554 p
->pack_name
, ntohl(hdr
.hdr_entries
),
556 if (lseek(p
->pack_fd
, p
->pack_size
- hashsz
, SEEK_SET
) == -1)
557 return error("end of packfile %s is unavailable", p
->pack_name
);
558 read_result
= read_in_full(p
->pack_fd
, hash
, hashsz
);
560 return error_errno("error reading from %s", p
->pack_name
);
561 if (read_result
!= hashsz
)
562 return error("packfile %s signature is unavailable", p
->pack_name
);
563 idx_hash
= ((unsigned char *)p
->index_data
) + p
->index_size
- hashsz
* 2;
564 if (!hasheq(hash
, idx_hash
))
565 return error("packfile %s does not match index", p
->pack_name
);
569 static int open_packed_git(struct packed_git
*p
)
571 if (!open_packed_git_1(p
))
577 static int in_window(struct pack_window
*win
, off_t offset
)
579 /* We must promise at least one full hash after the
580 * offset is available from this window, otherwise the offset
581 * is not actually in this window and a different window (which
582 * has that one hash excess) must be used. This is to support
583 * the object header and delta base parsing routines below.
585 off_t win_off
= win
->offset
;
586 return win_off
<= offset
587 && (offset
+ the_hash_algo
->rawsz
) <= (win_off
+ win
->len
);
590 unsigned char *use_pack(struct packed_git
*p
,
591 struct pack_window
**w_cursor
,
595 struct pack_window
*win
= *w_cursor
;
597 /* Since packfiles end in a hash of their content and it's
598 * pointless to ask for an offset into the middle of that
599 * hash, and the in_window function above wouldn't match
600 * don't allow an offset too close to the end of the file.
602 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
603 die("packfile %s cannot be accessed", p
->pack_name
);
604 if (offset
> (p
->pack_size
- the_hash_algo
->rawsz
))
605 die("offset beyond end of packfile (truncated pack?)");
607 die(_("offset before end of packfile (broken .idx?)"));
609 if (!win
|| !in_window(win
, offset
)) {
612 for (win
= p
->windows
; win
; win
= win
->next
) {
613 if (in_window(win
, offset
))
617 size_t window_align
= packed_git_window_size
/ 2;
620 if (p
->pack_fd
== -1 && open_packed_git(p
))
621 die("packfile %s cannot be accessed", p
->pack_name
);
623 win
= xcalloc(1, sizeof(*win
));
624 win
->offset
= (offset
/ window_align
) * window_align
;
625 len
= p
->pack_size
- win
->offset
;
626 if (len
> packed_git_window_size
)
627 len
= packed_git_window_size
;
628 win
->len
= (size_t)len
;
629 pack_mapped
+= win
->len
;
630 while (packed_git_limit
< pack_mapped
631 && unuse_one_window(p
))
633 win
->base
= xmmap(NULL
, win
->len
,
634 PROT_READ
, MAP_PRIVATE
,
635 p
->pack_fd
, win
->offset
);
636 if (win
->base
== MAP_FAILED
)
637 die_errno("packfile %s cannot be mapped",
639 if (!win
->offset
&& win
->len
== p
->pack_size
644 if (pack_mapped
> peak_pack_mapped
)
645 peak_pack_mapped
= pack_mapped
;
646 if (pack_open_windows
> peak_pack_open_windows
)
647 peak_pack_open_windows
= pack_open_windows
;
648 win
->next
= p
->windows
;
652 if (win
!= *w_cursor
) {
653 win
->last_used
= pack_used_ctr
++;
657 offset
-= win
->offset
;
659 *left
= win
->len
- xsize_t(offset
);
660 return win
->base
+ offset
;
663 void unuse_pack(struct pack_window
**w_cursor
)
665 struct pack_window
*w
= *w_cursor
;
672 static void try_to_free_pack_memory(size_t size
)
674 release_pack_memory(size
);
677 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
679 static int have_set_try_to_free_routine
;
682 struct packed_git
*p
;
684 if (!have_set_try_to_free_routine
) {
685 have_set_try_to_free_routine
= 1;
686 set_try_to_free_routine(try_to_free_pack_memory
);
690 * Make sure a corresponding .pack file exists and that
691 * the index looks sane.
693 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
697 * ".promisor" is long enough to hold any suffix we're adding (and
698 * the use xsnprintf double-checks that)
700 alloc
= st_add3(path_len
, strlen(".promisor"), 1);
701 p
= alloc_packed_git(alloc
);
702 memcpy(p
->pack_name
, path
, path_len
);
704 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
705 if (!access(p
->pack_name
, F_OK
))
708 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".promisor");
709 if (!access(p
->pack_name
, F_OK
))
710 p
->pack_promisor
= 1;
712 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
713 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
718 /* ok, it looks sane as far as we can check without
719 * actually mapping the pack file.
721 p
->pack_size
= st
.st_size
;
722 p
->pack_local
= local
;
723 p
->mtime
= st
.st_mtime
;
724 if (path_len
< the_hash_algo
->hexsz
||
725 get_sha1_hex(path
+ path_len
- the_hash_algo
->hexsz
, p
->sha1
))
730 void install_packed_git(struct repository
*r
, struct packed_git
*pack
)
732 if (pack
->pack_fd
!= -1)
735 pack
->next
= r
->objects
->packed_git
;
736 r
->objects
->packed_git
= pack
;
739 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
741 static void report_helper(const struct string_list
*list
,
742 int seen_bits
, int first
, int last
)
744 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
747 for (; first
< last
; first
++)
748 report_garbage(seen_bits
, list
->items
[first
].string
);
751 static void report_pack_garbage(struct string_list
*list
)
753 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
758 string_list_sort(list
);
760 for (i
= 0; i
< list
->nr
; i
++) {
761 const char *path
= list
->items
[i
].string
;
763 strncmp(path
, list
->items
[first
].string
, baselen
)) {
764 report_helper(list
, seen_bits
, first
, i
);
769 const char *dot
= strrchr(path
, '.');
771 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
774 baselen
= dot
- path
+ 1;
777 if (!strcmp(path
+ baselen
, "pack"))
779 else if (!strcmp(path
+ baselen
, "idx"))
782 report_helper(list
, seen_bits
, first
, list
->nr
);
785 void for_each_file_in_pack_dir(const char *objdir
,
786 each_file_in_pack_dir_fn fn
,
789 struct strbuf path
= STRBUF_INIT
;
794 strbuf_addstr(&path
, objdir
);
795 strbuf_addstr(&path
, "/pack");
796 dir
= opendir(path
.buf
);
799 error_errno("unable to open object pack directory: %s",
801 strbuf_release(&path
);
804 strbuf_addch(&path
, '/');
805 dirnamelen
= path
.len
;
806 while ((de
= readdir(dir
)) != NULL
) {
807 if (is_dot_or_dotdot(de
->d_name
))
810 strbuf_setlen(&path
, dirnamelen
);
811 strbuf_addstr(&path
, de
->d_name
);
813 fn(path
.buf
, path
.len
, de
->d_name
, data
);
817 strbuf_release(&path
);
820 struct prepare_pack_data
{
821 struct repository
*r
;
822 struct string_list
*garbage
;
824 struct multi_pack_index
*m
;
827 static void prepare_pack(const char *full_name
, size_t full_name_len
,
828 const char *file_name
, void *_data
)
830 struct prepare_pack_data
*data
= (struct prepare_pack_data
*)_data
;
831 struct packed_git
*p
;
832 size_t base_len
= full_name_len
;
834 if (strip_suffix_mem(full_name
, &base_len
, ".idx") &&
835 !(data
->m
&& midx_contains_pack(data
->m
, file_name
))) {
836 /* Don't reopen a pack we already have. */
837 for (p
= data
->r
->objects
->packed_git
; p
; p
= p
->next
) {
839 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
841 !memcmp(p
->pack_name
, full_name
, len
))
846 p
= add_packed_git(full_name
, full_name_len
, data
->local
);
848 install_packed_git(data
->r
, p
);
855 if (!strcmp(file_name
, "multi-pack-index"))
857 if (ends_with(file_name
, ".idx") ||
858 ends_with(file_name
, ".pack") ||
859 ends_with(file_name
, ".bitmap") ||
860 ends_with(file_name
, ".keep") ||
861 ends_with(file_name
, ".promisor"))
862 string_list_append(data
->garbage
, full_name
);
864 report_garbage(PACKDIR_FILE_GARBAGE
, full_name
);
867 static void prepare_packed_git_one(struct repository
*r
, char *objdir
, int local
)
869 struct prepare_pack_data data
;
870 struct string_list garbage
= STRING_LIST_INIT_DUP
;
872 data
.m
= r
->objects
->multi_pack_index
;
874 /* look for the multi-pack-index for this object directory */
875 while (data
.m
&& strcmp(data
.m
->object_dir
, objdir
))
876 data
.m
= data
.m
->next
;
879 data
.garbage
= &garbage
;
882 for_each_file_in_pack_dir(objdir
, prepare_pack
, &data
);
884 report_pack_garbage(data
.garbage
);
885 string_list_clear(data
.garbage
, 0);
888 static void prepare_packed_git(struct repository
*r
);
890 * Give a fast, rough count of the number of objects in the repository. This
891 * ignores loose objects completely. If you have a lot of them, then either
892 * you should repack because your performance will be awful, or they are
893 * all unreachable objects about to be pruned, in which case they're not really
894 * interesting as a measure of repo size in the first place.
896 unsigned long approximate_object_count(void)
898 if (!the_repository
->objects
->approximate_object_count_valid
) {
900 struct multi_pack_index
*m
;
901 struct packed_git
*p
;
903 prepare_packed_git(the_repository
);
905 for (m
= get_multi_pack_index(the_repository
); m
; m
= m
->next
)
906 count
+= m
->num_objects
;
907 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
908 if (open_pack_index(p
))
910 count
+= p
->num_objects
;
912 the_repository
->objects
->approximate_object_count
= count
;
914 return the_repository
->objects
->approximate_object_count
;
917 static void *get_next_packed_git(const void *p
)
919 return ((const struct packed_git
*)p
)->next
;
922 static void set_next_packed_git(void *p
, void *next
)
924 ((struct packed_git
*)p
)->next
= next
;
927 static int sort_pack(const void *a_
, const void *b_
)
929 const struct packed_git
*a
= a_
;
930 const struct packed_git
*b
= b_
;
934 * Local packs tend to contain objects specific to our
935 * variant of the project than remote ones. In addition,
936 * remote ones could be on a network mounted filesystem.
937 * Favor local ones for these reasons.
939 st
= a
->pack_local
- b
->pack_local
;
944 * Younger packs tend to contain more recent objects,
945 * and more recent objects tend to get accessed more
948 if (a
->mtime
< b
->mtime
)
950 else if (a
->mtime
== b
->mtime
)
955 static void rearrange_packed_git(struct repository
*r
)
957 r
->objects
->packed_git
= llist_mergesort(
958 r
->objects
->packed_git
, get_next_packed_git
,
959 set_next_packed_git
, sort_pack
);
962 static void prepare_packed_git_mru(struct repository
*r
)
964 struct packed_git
*p
;
966 INIT_LIST_HEAD(&r
->objects
->packed_git_mru
);
968 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
969 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
972 static void prepare_packed_git(struct repository
*r
)
974 struct alternate_object_database
*alt
;
976 if (r
->objects
->packed_git_initialized
)
978 prepare_multi_pack_index_one(r
, r
->objects
->objectdir
, 1);
979 prepare_packed_git_one(r
, r
->objects
->objectdir
, 1);
981 for (alt
= r
->objects
->alt_odb_list
; alt
; alt
= alt
->next
) {
982 prepare_multi_pack_index_one(r
, alt
->path
, 0);
983 prepare_packed_git_one(r
, alt
->path
, 0);
985 rearrange_packed_git(r
);
987 r
->objects
->all_packs
= NULL
;
989 prepare_packed_git_mru(r
);
990 r
->objects
->packed_git_initialized
= 1;
993 void reprepare_packed_git(struct repository
*r
)
995 r
->objects
->approximate_object_count_valid
= 0;
996 r
->objects
->packed_git_initialized
= 0;
997 prepare_packed_git(r
);
1000 struct packed_git
*get_packed_git(struct repository
*r
)
1002 prepare_packed_git(r
);
1003 return r
->objects
->packed_git
;
1006 struct multi_pack_index
*get_multi_pack_index(struct repository
*r
)
1008 prepare_packed_git(r
);
1009 return r
->objects
->multi_pack_index
;
1012 struct packed_git
*get_all_packs(struct repository
*r
)
1014 prepare_packed_git(r
);
1016 if (!r
->objects
->all_packs
) {
1017 struct packed_git
*p
= r
->objects
->packed_git
;
1018 struct multi_pack_index
*m
;
1020 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
1022 for (i
= 0; i
< m
->num_packs
; i
++) {
1023 if (!prepare_midx_pack(m
, i
)) {
1024 m
->packs
[i
]->next
= p
;
1030 r
->objects
->all_packs
= p
;
1033 return r
->objects
->all_packs
;
1036 struct list_head
*get_packed_git_mru(struct repository
*r
)
1038 prepare_packed_git(r
);
1039 return &r
->objects
->packed_git_mru
;
1042 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1043 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1046 unsigned long size
, c
;
1047 unsigned long used
= 0;
1050 *type
= (c
>> 4) & 7;
1054 if (len
<= used
|| bitsizeof(long) <= shift
) {
1055 error("bad object header");
1060 size
+= (c
& 0x7f) << shift
;
1067 unsigned long get_size_from_delta(struct packed_git
*p
,
1068 struct pack_window
**w_curs
,
1071 const unsigned char *data
;
1072 unsigned char delta_head
[20], *in
;
1076 memset(&stream
, 0, sizeof(stream
));
1077 stream
.next_out
= delta_head
;
1078 stream
.avail_out
= sizeof(delta_head
);
1080 git_inflate_init(&stream
);
1082 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1083 stream
.next_in
= in
;
1084 st
= git_inflate(&stream
, Z_FINISH
);
1085 curpos
+= stream
.next_in
- in
;
1086 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1087 stream
.total_out
< sizeof(delta_head
));
1088 git_inflate_end(&stream
);
1089 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1090 error("delta data unpack-initial failed");
1094 /* Examine the initial part of the delta to figure out
1099 /* ignore base size */
1100 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1102 /* Read the result size */
1103 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1106 int unpack_object_header(struct packed_git
*p
,
1107 struct pack_window
**w_curs
,
1109 unsigned long *sizep
)
1111 unsigned char *base
;
1114 enum object_type type
;
1116 /* use_pack() assures us we have [base, base + 20) available
1117 * as a range that we can look at. (Its actually the hash
1118 * size that is assured.) With our object header encoding
1119 * the maximum deflated object size is 2^137, which is just
1120 * insane, so we know won't exceed what we have been given.
1122 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1123 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1132 void mark_bad_packed_object(struct packed_git
*p
, const unsigned char *sha1
)
1135 const unsigned hashsz
= the_hash_algo
->rawsz
;
1136 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1137 if (hasheq(sha1
, p
->bad_object_sha1
+ hashsz
* i
))
1139 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1140 st_mult(GIT_MAX_RAWSZ
,
1141 st_add(p
->num_bad_objects
, 1)));
1142 hashcpy(p
->bad_object_sha1
+ hashsz
* p
->num_bad_objects
, sha1
);
1143 p
->num_bad_objects
++;
1146 const struct packed_git
*has_packed_and_bad(struct repository
*r
,
1147 const unsigned char *sha1
)
1149 struct packed_git
*p
;
1152 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
1153 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1155 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
1160 static off_t
get_delta_base(struct packed_git
*p
,
1161 struct pack_window
**w_curs
,
1163 enum object_type type
,
1164 off_t delta_obj_offset
)
1166 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1169 /* use_pack() assured us we have [base_info, base_info + 20)
1170 * as a range that we can look at without walking off the
1171 * end of the mapped window. Its actually the hash size
1172 * that is assured. An OFS_DELTA longer than the hash size
1173 * is stupid, as then a REF_DELTA would be smaller to store.
1175 if (type
== OBJ_OFS_DELTA
) {
1177 unsigned char c
= base_info
[used
++];
1178 base_offset
= c
& 127;
1181 if (!base_offset
|| MSB(base_offset
, 7))
1182 return 0; /* overflow */
1183 c
= base_info
[used
++];
1184 base_offset
= (base_offset
<< 7) + (c
& 127);
1186 base_offset
= delta_obj_offset
- base_offset
;
1187 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1188 return 0; /* out of bound */
1190 } else if (type
== OBJ_REF_DELTA
) {
1191 /* The base entry _must_ be in the same pack */
1192 base_offset
= find_pack_entry_one(base_info
, p
);
1193 *curpos
+= the_hash_algo
->rawsz
;
1195 die("I am totally screwed");
1200 * Like get_delta_base above, but we return the sha1 instead of the pack
1201 * offset. This means it is cheaper for REF deltas (we do not have to do
1202 * the final object lookup), but more expensive for OFS deltas (we
1203 * have to load the revidx to convert the offset back into a sha1).
1205 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1206 struct pack_window
**w_curs
,
1208 enum object_type type
,
1209 off_t delta_obj_offset
)
1211 if (type
== OBJ_REF_DELTA
) {
1212 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1214 } else if (type
== OBJ_OFS_DELTA
) {
1215 struct revindex_entry
*revidx
;
1216 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1217 type
, delta_obj_offset
);
1222 revidx
= find_pack_revindex(p
, base_offset
);
1226 return nth_packed_object_sha1(p
, revidx
->nr
);
1231 static int retry_bad_packed_offset(struct repository
*r
,
1232 struct packed_git
*p
,
1236 struct revindex_entry
*revidx
;
1237 struct object_id oid
;
1238 revidx
= find_pack_revindex(p
, obj_offset
);
1241 nth_packed_object_oid(&oid
, p
, revidx
->nr
);
1242 mark_bad_packed_object(p
, oid
.hash
);
1243 type
= oid_object_info(r
, &oid
, NULL
);
1244 if (type
<= OBJ_NONE
)
1249 #define POI_STACK_PREALLOC 64
1251 static enum object_type
packed_to_object_type(struct repository
*r
,
1252 struct packed_git
*p
,
1254 enum object_type type
,
1255 struct pack_window
**w_curs
,
1258 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1259 off_t
*poi_stack
= small_poi_stack
;
1260 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1262 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1265 /* Push the object we're going to leave behind */
1266 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1267 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1268 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
1269 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1271 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1273 poi_stack
[poi_stack_nr
++] = obj_offset
;
1274 /* If parsing the base offset fails, just unwind */
1275 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1278 curpos
= obj_offset
= base_offset
;
1279 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1280 if (type
<= OBJ_NONE
) {
1281 /* If getting the base itself fails, we first
1282 * retry the base, otherwise unwind */
1283 type
= retry_bad_packed_offset(r
, p
, base_offset
);
1284 if (type
> OBJ_NONE
)
1298 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1299 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1304 if (poi_stack
!= small_poi_stack
)
1309 while (poi_stack_nr
) {
1310 obj_offset
= poi_stack
[--poi_stack_nr
];
1311 type
= retry_bad_packed_offset(r
, p
, obj_offset
);
1312 if (type
> OBJ_NONE
)
1319 static struct hashmap delta_base_cache
;
1320 static size_t delta_base_cached
;
1322 static LIST_HEAD(delta_base_cache_lru
);
1324 struct delta_base_cache_key
{
1325 struct packed_git
*p
;
1329 struct delta_base_cache_entry
{
1330 struct hashmap hash
;
1331 struct delta_base_cache_key key
;
1332 struct list_head lru
;
1335 enum object_type type
;
1338 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1342 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
1343 hash
+= (hash
>> 8) + (hash
>> 16);
1347 static struct delta_base_cache_entry
*
1348 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1350 struct hashmap_entry entry
;
1351 struct delta_base_cache_key key
;
1353 if (!delta_base_cache
.cmpfn
)
1356 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
1358 key
.base_offset
= base_offset
;
1359 return hashmap_get(&delta_base_cache
, &entry
, &key
);
1362 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
1363 const struct delta_base_cache_key
*b
)
1365 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
1368 static int delta_base_cache_hash_cmp(const void *unused_cmp_data
,
1369 const void *va
, const void *vb
,
1372 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
1373 const struct delta_base_cache_key
*key
= vkey
;
1375 return !delta_base_cache_key_eq(&a
->key
, key
);
1377 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
1380 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1382 return !!get_delta_base_cache_entry(p
, base_offset
);
1386 * Remove the entry from the cache, but do _not_ free the associated
1387 * entry data. The caller takes ownership of the "data" buffer, and
1388 * should copy out any fields it wants before detaching.
1390 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1392 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
1393 list_del(&ent
->lru
);
1394 delta_base_cached
-= ent
->size
;
1398 static void *cache_or_unpack_entry(struct repository
*r
, struct packed_git
*p
,
1399 off_t base_offset
, unsigned long *base_size
,
1400 enum object_type
*type
)
1402 struct delta_base_cache_entry
*ent
;
1404 ent
= get_delta_base_cache_entry(p
, base_offset
);
1406 return unpack_entry(r
, p
, base_offset
, type
, base_size
);
1411 *base_size
= ent
->size
;
1412 return xmemdupz(ent
->data
, ent
->size
);
1415 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1418 detach_delta_base_cache_entry(ent
);
1421 void clear_delta_base_cache(void)
1423 struct list_head
*lru
, *tmp
;
1424 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1425 struct delta_base_cache_entry
*entry
=
1426 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1427 release_delta_base_cache(entry
);
1431 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1432 void *base
, unsigned long base_size
, enum object_type type
)
1434 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
1435 struct list_head
*lru
, *tmp
;
1437 delta_base_cached
+= base_size
;
1439 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1440 struct delta_base_cache_entry
*f
=
1441 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1442 if (delta_base_cached
<= delta_base_cache_limit
)
1444 release_delta_base_cache(f
);
1448 ent
->key
.base_offset
= base_offset
;
1451 ent
->size
= base_size
;
1452 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
1454 if (!delta_base_cache
.cmpfn
)
1455 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, NULL
, 0);
1456 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
1457 hashmap_add(&delta_base_cache
, ent
);
1460 int packed_object_info(struct repository
*r
, struct packed_git
*p
,
1461 off_t obj_offset
, struct object_info
*oi
)
1463 struct pack_window
*w_curs
= NULL
;
1465 off_t curpos
= obj_offset
;
1466 enum object_type type
;
1469 * We always get the representation type, but only convert it to
1470 * a "real" type later if the caller is interested.
1473 *oi
->contentp
= cache_or_unpack_entry(r
, p
, obj_offset
, oi
->sizep
,
1478 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1481 if (!oi
->contentp
&& oi
->sizep
) {
1482 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1483 off_t tmp_pos
= curpos
;
1484 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1490 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1491 if (*oi
->sizep
== 0) {
1500 if (oi
->disk_sizep
) {
1501 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1502 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1505 if (oi
->typep
|| oi
->type_name
) {
1506 enum object_type ptot
;
1507 ptot
= packed_to_object_type(r
, p
, obj_offset
,
1508 type
, &w_curs
, curpos
);
1511 if (oi
->type_name
) {
1512 const char *tn
= type_name(ptot
);
1514 strbuf_addstr(oi
->type_name
, tn
);
1522 if (oi
->delta_base_sha1
) {
1523 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1524 const unsigned char *base
;
1526 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1533 hashcpy(oi
->delta_base_sha1
, base
);
1535 hashclr(oi
->delta_base_sha1
);
1538 oi
->whence
= in_delta_base_cache(p
, obj_offset
) ? OI_DBCACHED
:
1542 unuse_pack(&w_curs
);
1546 static void *unpack_compressed_entry(struct packed_git
*p
,
1547 struct pack_window
**w_curs
,
1553 unsigned char *buffer
, *in
;
1555 buffer
= xmallocz_gently(size
);
1558 memset(&stream
, 0, sizeof(stream
));
1559 stream
.next_out
= buffer
;
1560 stream
.avail_out
= size
+ 1;
1562 git_inflate_init(&stream
);
1564 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1565 stream
.next_in
= in
;
1566 st
= git_inflate(&stream
, Z_FINISH
);
1567 if (!stream
.avail_out
)
1568 break; /* the payload is larger than it should be */
1569 curpos
+= stream
.next_in
- in
;
1570 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1571 git_inflate_end(&stream
);
1572 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1577 /* versions of zlib can clobber unconsumed portion of outbuf */
1578 buffer
[size
] = '\0';
1583 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1585 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
1586 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
1587 p
->pack_name
, (uintmax_t)obj_offset
);
1590 int do_check_packed_object_crc
;
1592 #define UNPACK_ENTRY_STACK_PREALLOC 64
1593 struct unpack_entry_stack_ent
{
1599 static void *read_object(struct repository
*r
,
1600 const struct object_id
*oid
,
1601 enum object_type
*type
,
1602 unsigned long *size
)
1604 struct object_info oi
= OBJECT_INFO_INIT
;
1608 oi
.contentp
= &content
;
1610 if (oid_object_info_extended(r
, oid
, &oi
, 0) < 0)
1615 void *unpack_entry(struct repository
*r
, struct packed_git
*p
, off_t obj_offset
,
1616 enum object_type
*final_type
, unsigned long *final_size
)
1618 struct pack_window
*w_curs
= NULL
;
1619 off_t curpos
= obj_offset
;
1622 enum object_type type
;
1623 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
1624 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
1625 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
1626 int base_from_cache
= 0;
1628 write_pack_access_log(p
, obj_offset
);
1630 /* PHASE 1: drill down to the innermost base object */
1634 struct delta_base_cache_entry
*ent
;
1636 ent
= get_delta_base_cache_entry(p
, curpos
);
1641 detach_delta_base_cache_entry(ent
);
1642 base_from_cache
= 1;
1646 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1647 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1648 off_t len
= revidx
[1].offset
- obj_offset
;
1649 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1650 struct object_id oid
;
1651 nth_packed_object_oid(&oid
, p
, revidx
->nr
);
1652 error("bad packed object CRC for %s",
1654 mark_bad_packed_object(p
, oid
.hash
);
1660 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1661 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
1664 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1666 error("failed to validate delta base reference "
1667 "at offset %"PRIuMAX
" from %s",
1668 (uintmax_t)curpos
, p
->pack_name
);
1669 /* bail to phase 2, in hopes of recovery */
1674 /* push object, proceed to base */
1675 if (delta_stack_nr
>= delta_stack_alloc
1676 && delta_stack
== small_delta_stack
) {
1677 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
1678 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
1679 memcpy(delta_stack
, small_delta_stack
,
1680 sizeof(*delta_stack
)*delta_stack_nr
);
1682 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
1684 i
= delta_stack_nr
++;
1685 delta_stack
[i
].obj_offset
= obj_offset
;
1686 delta_stack
[i
].curpos
= curpos
;
1687 delta_stack
[i
].size
= size
;
1689 curpos
= obj_offset
= base_offset
;
1692 /* PHASE 2: handle the base */
1697 BUG("unpack_entry: left loop at a valid delta");
1703 if (!base_from_cache
)
1704 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
1708 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1709 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1712 /* PHASE 3: apply deltas in order */
1715 * 'data' holds the base data, or NULL if there was corruption
1717 while (delta_stack_nr
) {
1720 void *external_base
= NULL
;
1721 unsigned long delta_size
, base_size
= size
;
1727 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
1731 * We're probably in deep shit, but let's try to fetch
1732 * the required base anyway from another pack or loose.
1733 * This is costly but should happen only in the presence
1734 * of a corrupted pack, and is better than failing outright.
1736 struct revindex_entry
*revidx
;
1737 struct object_id base_oid
;
1738 revidx
= find_pack_revindex(p
, obj_offset
);
1740 nth_packed_object_oid(&base_oid
, p
, revidx
->nr
);
1741 error("failed to read delta base object %s"
1742 " at offset %"PRIuMAX
" from %s",
1743 oid_to_hex(&base_oid
), (uintmax_t)obj_offset
,
1745 mark_bad_packed_object(p
, base_oid
.hash
);
1746 base
= read_object(r
, &base_oid
, &type
, &base_size
);
1747 external_base
= base
;
1751 i
= --delta_stack_nr
;
1752 obj_offset
= delta_stack
[i
].obj_offset
;
1753 curpos
= delta_stack
[i
].curpos
;
1754 delta_size
= delta_stack
[i
].size
;
1759 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
1762 error("failed to unpack compressed delta "
1763 "at offset %"PRIuMAX
" from %s",
1764 (uintmax_t)curpos
, p
->pack_name
);
1766 free(external_base
);
1770 data
= patch_delta(base
, base_size
,
1771 delta_data
, delta_size
,
1775 * We could not apply the delta; warn the user, but keep going.
1776 * Our failure will be noticed either in the next iteration of
1777 * the loop, or if this is the final delta, in the caller when
1778 * we return NULL. Those code paths will take care of making
1779 * a more explicit warning and retrying with another copy of
1783 error("failed to apply delta");
1786 free(external_base
);
1795 unuse_pack(&w_curs
);
1797 if (delta_stack
!= small_delta_stack
)
1803 int bsearch_pack(const struct object_id
*oid
, const struct packed_git
*p
, uint32_t *result
)
1805 const unsigned char *index_fanout
= p
->index_data
;
1806 const unsigned char *index_lookup
;
1807 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1808 int index_lookup_width
;
1811 BUG("bsearch_pack called without a valid pack-index");
1813 index_lookup
= index_fanout
+ 4 * 256;
1814 if (p
->index_version
== 1) {
1815 index_lookup_width
= hashsz
+ 4;
1818 index_lookup_width
= hashsz
;
1823 return bsearch_hash(oid
->hash
, (const uint32_t*)index_fanout
,
1824 index_lookup
, index_lookup_width
, result
);
1827 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1830 const unsigned char *index
= p
->index_data
;
1831 const unsigned int hashsz
= the_hash_algo
->rawsz
;
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
+ (hashsz
+ 4) * n
+ 4;
1844 return index
+ hashsz
* n
;
1848 const struct object_id
*nth_packed_object_oid(struct object_id
*oid
,
1849 struct packed_git
*p
,
1852 const unsigned char *hash
= nth_packed_object_sha1(p
, n
);
1855 hashcpy(oid
->hash
, hash
);
1859 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
1861 const unsigned char *ptr
= vptr
;
1862 const unsigned char *start
= p
->index_data
;
1863 const unsigned char *end
= start
+ p
->index_size
;
1865 die(_("offset before start of pack index for %s (corrupt index?)"),
1867 /* No need to check for underflow; .idx files must be at least 8 bytes */
1869 die(_("offset beyond end of pack index for %s (truncated index?)"),
1873 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1875 const unsigned char *index
= p
->index_data
;
1876 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1878 if (p
->index_version
== 1) {
1879 return ntohl(*((uint32_t *)(index
+ (hashsz
+ 4) * n
)));
1882 index
+= 8 + p
->num_objects
* (hashsz
+ 4);
1883 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1884 if (!(off
& 0x80000000))
1886 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1887 check_pack_index_ptr(p
, index
);
1888 return get_be64(index
);
1892 off_t
find_pack_entry_one(const unsigned char *sha1
,
1893 struct packed_git
*p
)
1895 const unsigned char *index
= p
->index_data
;
1896 struct object_id oid
;
1900 if (open_pack_index(p
))
1904 hashcpy(oid
.hash
, sha1
);
1905 if (bsearch_pack(&oid
, p
, &result
))
1906 return nth_packed_object_offset(p
, result
);
1910 int is_pack_valid(struct packed_git
*p
)
1912 /* An already open pack is known to be valid. */
1913 if (p
->pack_fd
!= -1)
1916 /* If the pack has one window completely covering the
1917 * file size, the pack is known to be valid even if
1918 * the descriptor is not currently open.
1921 struct pack_window
*w
= p
->windows
;
1923 if (!w
->offset
&& w
->len
== p
->pack_size
)
1927 /* Force the pack to open to prove its valid. */
1928 return !open_packed_git(p
);
1931 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1932 struct packed_git
*packs
)
1934 struct packed_git
*p
;
1936 for (p
= packs
; p
; p
= p
->next
) {
1937 if (find_pack_entry_one(sha1
, p
))
1944 static int fill_pack_entry(const struct object_id
*oid
,
1945 struct pack_entry
*e
,
1946 struct packed_git
*p
)
1950 if (p
->num_bad_objects
) {
1952 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1953 if (hasheq(oid
->hash
,
1954 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
1958 offset
= find_pack_entry_one(oid
->hash
, p
);
1963 * We are about to tell the caller where they can locate the
1964 * requested object. We better make sure the packfile is
1965 * still here and can be accessed before supplying that
1966 * answer, as it may have been deleted since the index was
1969 if (!is_pack_valid(p
))
1976 int find_pack_entry(struct repository
*r
, const struct object_id
*oid
, struct pack_entry
*e
)
1978 struct list_head
*pos
;
1979 struct multi_pack_index
*m
;
1981 prepare_packed_git(r
);
1982 if (!r
->objects
->packed_git
&& !r
->objects
->multi_pack_index
)
1985 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
1986 if (fill_midx_entry(oid
, e
, m
))
1990 list_for_each(pos
, &r
->objects
->packed_git_mru
) {
1991 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1992 if (fill_pack_entry(oid
, e
, p
)) {
1993 list_move(&p
->mru
, &r
->objects
->packed_git_mru
);
2000 int has_object_pack(const struct object_id
*oid
)
2002 struct pack_entry e
;
2003 return find_pack_entry(the_repository
, oid
, &e
);
2006 int has_pack_index(const unsigned char *sha1
)
2009 if (stat(sha1_pack_index_name(sha1
), &st
))
2014 int for_each_object_in_pack(struct packed_git
*p
,
2015 each_packed_object_fn cb
, void *data
,
2016 enum for_each_object_flags flags
)
2021 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
)
2022 load_pack_revindex(p
);
2024 for (i
= 0; i
< p
->num_objects
; i
++) {
2026 struct object_id oid
;
2028 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
)
2029 pos
= p
->revindex
[i
].nr
;
2033 if (!nth_packed_object_oid(&oid
, p
, pos
))
2034 return error("unable to get sha1 of object %u in %s",
2037 r
= cb(&oid
, p
, pos
, data
);
2044 int for_each_packed_object(each_packed_object_fn cb
, void *data
,
2045 enum for_each_object_flags flags
)
2047 struct packed_git
*p
;
2049 int pack_errors
= 0;
2051 prepare_packed_git(the_repository
);
2052 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
2053 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
2055 if ((flags
& FOR_EACH_OBJECT_PROMISOR_ONLY
) &&
2058 if (open_pack_index(p
)) {
2062 r
= for_each_object_in_pack(p
, cb
, data
, flags
);
2066 return r
? r
: pack_errors
;
2069 static int add_promisor_object(const struct object_id
*oid
,
2070 struct packed_git
*pack
,
2074 struct oidset
*set
= set_
;
2075 struct object
*obj
= parse_object(the_repository
, oid
);
2079 oidset_insert(set
, oid
);
2082 * If this is a tree, commit, or tag, the objects it refers
2083 * to are also promisor objects. (Blobs refer to no objects->)
2085 if (obj
->type
== OBJ_TREE
) {
2086 struct tree
*tree
= (struct tree
*)obj
;
2087 struct tree_desc desc
;
2088 struct name_entry entry
;
2089 if (init_tree_desc_gently(&desc
, tree
->buffer
, tree
->size
))
2091 * Error messages are given when packs are
2092 * verified, so do not print any here.
2095 while (tree_entry_gently(&desc
, &entry
))
2096 oidset_insert(set
, entry
.oid
);
2097 } else if (obj
->type
== OBJ_COMMIT
) {
2098 struct commit
*commit
= (struct commit
*) obj
;
2099 struct commit_list
*parents
= commit
->parents
;
2101 oidset_insert(set
, get_commit_tree_oid(commit
));
2102 for (; parents
; parents
= parents
->next
)
2103 oidset_insert(set
, &parents
->item
->object
.oid
);
2104 } else if (obj
->type
== OBJ_TAG
) {
2105 struct tag
*tag
= (struct tag
*) obj
;
2106 oidset_insert(set
, &tag
->tagged
->oid
);
2111 int is_promisor_object(const struct object_id
*oid
)
2113 static struct oidset promisor_objects
;
2114 static int promisor_objects_prepared
;
2116 if (!promisor_objects_prepared
) {
2117 if (repository_format_partial_clone
) {
2118 for_each_packed_object(add_promisor_object
,
2120 FOR_EACH_OBJECT_PROMISOR_ONLY
);
2122 promisor_objects_prepared
= 1;
2124 return oidset_contains(&promisor_objects
, oid
);