4 #include "repository.h"
10 #include "streaming.h"
11 #include "sha1-lookup.h"
15 #include "tree-walk.h"
17 #include "object-store.h"
19 #include "commit-graph.h"
21 char *odb_pack_name(struct strbuf
*buf
,
22 const unsigned char *sha1
,
26 strbuf_addf(buf
, "%s/pack/pack-%s.%s", get_object_directory(),
27 sha1_to_hex(sha1
), ext
);
31 char *sha1_pack_name(const unsigned char *sha1
)
33 static struct strbuf buf
= STRBUF_INIT
;
34 return odb_pack_name(&buf
, sha1
, "pack");
37 char *sha1_pack_index_name(const unsigned char *sha1
)
39 static struct strbuf buf
= STRBUF_INIT
;
40 return odb_pack_name(&buf
, sha1
, "idx");
43 static unsigned int pack_used_ctr
;
44 static unsigned int pack_mmap_calls
;
45 static unsigned int peak_pack_open_windows
;
46 static unsigned int pack_open_windows
;
47 static unsigned int pack_open_fds
;
48 static unsigned int pack_max_fds
;
49 static size_t peak_pack_mapped
;
50 static size_t pack_mapped
;
52 #define SZ_FMT PRIuMAX
53 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
55 void pack_report(void)
58 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
59 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
60 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
61 sz_fmt(getpagesize()),
62 sz_fmt(packed_git_window_size
),
63 sz_fmt(packed_git_limit
));
65 "pack_report: pack_used_ctr = %10u\n"
66 "pack_report: pack_mmap_calls = %10u\n"
67 "pack_report: pack_open_windows = %10u / %10u\n"
68 "pack_report: pack_mapped = "
69 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
72 pack_open_windows
, peak_pack_open_windows
,
73 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
77 * Open and mmap the index file at path, perform a couple of
78 * consistency checks, then record its information to p. Return 0 on
81 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
85 int fd
= git_open(path
), ret
;
87 const unsigned int hashsz
= the_hash_algo
->rawsz
;
95 idx_size
= xsize_t(st
.st_size
);
96 if (idx_size
< 4 * 256 + hashsz
+ hashsz
) {
98 return error("index file %s is too small", path
);
100 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
103 ret
= load_idx(path
, hashsz
, idx_map
, idx_size
, p
);
106 munmap(idx_map
, idx_size
);
111 int load_idx(const char *path
, const unsigned int hashsz
, void *idx_map
,
112 size_t idx_size
, struct packed_git
*p
)
114 struct pack_idx_header
*hdr
= idx_map
;
115 uint32_t version
, nr
, i
, *index
;
117 if (idx_size
< 4 * 256 + hashsz
+ hashsz
)
118 return error("index file %s is too small", path
);
120 return error("empty data");
122 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
123 version
= ntohl(hdr
->idx_version
);
124 if (version
< 2 || version
> 2)
125 return error("index file %s is version %"PRIu32
126 " and is not supported by this binary"
127 " (try upgrading GIT to a newer version)",
135 index
+= 2; /* skip index header */
136 for (i
= 0; i
< 256; i
++) {
137 uint32_t n
= ntohl(index
[i
]);
139 return error("non-monotonic index %s", path
);
146 * - 256 index entries 4 bytes each
147 * - 24-byte entries * nr (object ID + 4-byte offset)
148 * - hash of the packfile
151 if (idx_size
!= 4 * 256 + nr
* (hashsz
+ 4) + hashsz
+ hashsz
)
152 return error("wrong index v1 file size in %s", path
);
153 } else if (version
== 2) {
156 * - 8 bytes of header
157 * - 256 index entries 4 bytes each
158 * - object ID entry * nr
159 * - 4-byte crc entry * nr
160 * - 4-byte offset entry * nr
161 * - hash of the packfile
163 * And after the 4-byte offset table might be a
164 * variable sized table containing 8-byte entries
165 * for offsets larger than 2^31.
167 unsigned long min_size
= 8 + 4*256 + nr
*(hashsz
+ 4 + 4) + hashsz
+ hashsz
;
168 unsigned long max_size
= min_size
;
170 max_size
+= (nr
- 1)*8;
171 if (idx_size
< min_size
|| idx_size
> max_size
)
172 return error("wrong index v2 file size in %s", path
);
173 if (idx_size
!= min_size
&&
175 * make sure we can deal with large pack offsets.
176 * 31-bit signed offset won't be enough, neither
177 * 32-bit unsigned one will be.
179 (sizeof(off_t
) <= 4))
180 return error("pack too large for current definition of off_t in %s", path
);
183 p
->index_version
= version
;
184 p
->index_data
= idx_map
;
185 p
->index_size
= idx_size
;
190 int open_pack_index(struct packed_git
*p
)
199 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
200 BUG("pack_name does not end in .pack");
201 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
202 ret
= check_packed_git_idx(idx_name
, p
);
207 uint32_t get_pack_fanout(struct packed_git
*p
, uint32_t value
)
209 const uint32_t *level1_ofs
= p
->index_data
;
212 if (open_pack_index(p
))
214 level1_ofs
= p
->index_data
;
217 if (p
->index_version
> 1) {
221 return ntohl(level1_ofs
[value
]);
224 static struct packed_git
*alloc_packed_git(int extra
)
226 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
227 memset(p
, 0, sizeof(*p
));
232 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
234 const char *path
= sha1_pack_name(sha1
);
235 size_t alloc
= st_add(strlen(path
), 1);
236 struct packed_git
*p
= alloc_packed_git(alloc
);
238 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
239 hashcpy(p
->sha1
, sha1
);
240 if (check_packed_git_idx(idx_path
, p
)) {
248 static void scan_windows(struct packed_git
*p
,
249 struct packed_git
**lru_p
,
250 struct pack_window
**lru_w
,
251 struct pack_window
**lru_l
)
253 struct pack_window
*w
, *w_l
;
255 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
257 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
267 static int unuse_one_window(struct packed_git
*current
)
269 struct packed_git
*p
, *lru_p
= NULL
;
270 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
273 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
274 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
)
275 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
277 munmap(lru_w
->base
, lru_w
->len
);
278 pack_mapped
-= lru_w
->len
;
280 lru_l
->next
= lru_w
->next
;
282 lru_p
->windows
= lru_w
->next
;
290 void release_pack_memory(size_t need
)
292 size_t cur
= pack_mapped
;
293 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
297 void close_pack_windows(struct packed_git
*p
)
300 struct pack_window
*w
= p
->windows
;
303 die("pack '%s' still has open windows to it",
305 munmap(w
->base
, w
->len
);
306 pack_mapped
-= w
->len
;
308 p
->windows
= w
->next
;
313 static int close_pack_fd(struct packed_git
*p
)
325 void close_pack_index(struct packed_git
*p
)
328 munmap((void *)p
->index_data
, p
->index_size
);
329 p
->index_data
= NULL
;
333 void close_pack(struct packed_git
*p
)
335 close_pack_windows(p
);
340 void close_object_store(struct raw_object_store
*o
)
342 struct packed_git
*p
;
344 for (p
= o
->packed_git
; p
; p
= p
->next
)
346 BUG("want to close pack marked 'do-not-close'");
350 if (o
->multi_pack_index
) {
351 close_midx(o
->multi_pack_index
);
352 o
->multi_pack_index
= NULL
;
355 close_commit_graph(o
);
359 * The LRU pack is the one with the oldest MRU window, preferring packs
360 * with no used windows, or the oldest mtime if it has no windows allocated.
362 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
364 struct pack_window
*w
, *this_mru_w
;
365 int has_windows_inuse
= 0;
368 * Reject this pack if it has windows and the previously selected
369 * one does not. If this pack does not have windows, reject
370 * it if the pack file is newer than the previously selected one.
372 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
375 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
377 * Reject this pack if any of its windows are in use,
378 * but the previously selected pack did not have any
379 * inuse windows. Otherwise, record that this pack
380 * has windows in use.
383 if (*accept_windows_inuse
)
384 has_windows_inuse
= 1;
389 if (w
->last_used
> this_mru_w
->last_used
)
393 * Reject this pack if it has windows that have been
394 * used more recently than the previously selected pack.
395 * If the previously selected pack had windows inuse and
396 * we have not encountered a window in this pack that is
397 * inuse, skip this check since we prefer a pack with no
398 * inuse windows to one that has inuse windows.
400 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
401 this_mru_w
->last_used
> (*mru_w
)->last_used
)
410 *accept_windows_inuse
= has_windows_inuse
;
413 static int close_one_pack(void)
415 struct packed_git
*p
, *lru_p
= NULL
;
416 struct pack_window
*mru_w
= NULL
;
417 int accept_windows_inuse
= 1;
419 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
420 if (p
->pack_fd
== -1)
422 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
426 return close_pack_fd(lru_p
);
431 static unsigned int get_max_fd_limit(void)
437 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
444 long open_max
= sysconf(_SC_OPEN_MAX
);
448 * Otherwise, we got -1 for one of the two
451 * (1) sysconf() did not understand _SC_OPEN_MAX
452 * and signaled an error with -1; or
453 * (2) sysconf() said there is no limit.
455 * We _could_ clear errno before calling sysconf() to
456 * tell these two cases apart and return a huge number
457 * in the latter case to let the caller cap it to a
458 * value that is not so selfish, but letting the
459 * fallback OPEN_MAX codepath take care of these cases
468 return 1; /* see the caller ;-) */
473 * Do not call this directly as this leaks p->pack_fd on error return;
474 * call open_packed_git() instead.
476 static int open_packed_git_1(struct packed_git
*p
)
479 struct pack_header hdr
;
480 unsigned char hash
[GIT_MAX_RAWSZ
];
481 unsigned char *idx_hash
;
484 const unsigned hashsz
= the_hash_algo
->rawsz
;
486 if (!p
->index_data
) {
487 struct multi_pack_index
*m
;
488 const char *pack_name
= strrchr(p
->pack_name
, '/');
490 for (m
= the_repository
->objects
->multi_pack_index
;
492 if (midx_contains_pack(m
, pack_name
))
496 if (!m
&& open_pack_index(p
))
497 return error("packfile %s index unavailable", p
->pack_name
);
501 unsigned int max_fds
= get_max_fd_limit();
503 /* Save 3 for stdin/stdout/stderr, 22 for work */
505 pack_max_fds
= max_fds
- 25;
510 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
513 p
->pack_fd
= git_open(p
->pack_name
);
514 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
518 /* If we created the struct before we had the pack we lack size. */
520 if (!S_ISREG(st
.st_mode
))
521 return error("packfile %s not a regular file", p
->pack_name
);
522 p
->pack_size
= st
.st_size
;
523 } else if (p
->pack_size
!= st
.st_size
)
524 return error("packfile %s size changed", p
->pack_name
);
526 /* We leave these file descriptors open with sliding mmap;
527 * there is no point keeping them open across exec(), though.
529 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
531 return error("cannot determine file descriptor flags");
532 fd_flag
|= FD_CLOEXEC
;
533 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
534 return error("cannot set FD_CLOEXEC");
536 /* Verify we recognize this pack file format. */
537 read_result
= read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
));
539 return error_errno("error reading from %s", p
->pack_name
);
540 if (read_result
!= sizeof(hdr
))
541 return error("file %s is far too short to be a packfile", p
->pack_name
);
542 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
543 return error("file %s is not a GIT packfile", p
->pack_name
);
544 if (!pack_version_ok(hdr
.hdr_version
))
545 return error("packfile %s is version %"PRIu32
" and not"
546 " supported (try upgrading GIT to a newer version)",
547 p
->pack_name
, ntohl(hdr
.hdr_version
));
549 /* Skip index checking if in multi-pack-index */
553 /* Verify the pack matches its index. */
554 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
555 return error("packfile %s claims to have %"PRIu32
" objects"
556 " while index indicates %"PRIu32
" objects",
557 p
->pack_name
, ntohl(hdr
.hdr_entries
),
559 if (lseek(p
->pack_fd
, p
->pack_size
- hashsz
, SEEK_SET
) == -1)
560 return error("end of packfile %s is unavailable", p
->pack_name
);
561 read_result
= read_in_full(p
->pack_fd
, hash
, hashsz
);
563 return error_errno("error reading from %s", p
->pack_name
);
564 if (read_result
!= hashsz
)
565 return error("packfile %s signature is unavailable", p
->pack_name
);
566 idx_hash
= ((unsigned char *)p
->index_data
) + p
->index_size
- hashsz
* 2;
567 if (!hasheq(hash
, idx_hash
))
568 return error("packfile %s does not match index", p
->pack_name
);
572 static int open_packed_git(struct packed_git
*p
)
574 if (!open_packed_git_1(p
))
580 static int in_window(struct pack_window
*win
, off_t offset
)
582 /* We must promise at least one full hash after the
583 * offset is available from this window, otherwise the offset
584 * is not actually in this window and a different window (which
585 * has that one hash excess) must be used. This is to support
586 * the object header and delta base parsing routines below.
588 off_t win_off
= win
->offset
;
589 return win_off
<= offset
590 && (offset
+ the_hash_algo
->rawsz
) <= (win_off
+ win
->len
);
593 unsigned char *use_pack(struct packed_git
*p
,
594 struct pack_window
**w_cursor
,
598 struct pack_window
*win
= *w_cursor
;
600 /* Since packfiles end in a hash of their content and it's
601 * pointless to ask for an offset into the middle of that
602 * hash, and the in_window function above wouldn't match
603 * don't allow an offset too close to the end of the file.
605 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
606 die("packfile %s cannot be accessed", p
->pack_name
);
607 if (offset
> (p
->pack_size
- the_hash_algo
->rawsz
))
608 die("offset beyond end of packfile (truncated pack?)");
610 die(_("offset before end of packfile (broken .idx?)"));
612 if (!win
|| !in_window(win
, offset
)) {
615 for (win
= p
->windows
; win
; win
= win
->next
) {
616 if (in_window(win
, offset
))
620 size_t window_align
= packed_git_window_size
/ 2;
623 if (p
->pack_fd
== -1 && open_packed_git(p
))
624 die("packfile %s cannot be accessed", p
->pack_name
);
626 win
= xcalloc(1, sizeof(*win
));
627 win
->offset
= (offset
/ window_align
) * window_align
;
628 len
= p
->pack_size
- win
->offset
;
629 if (len
> packed_git_window_size
)
630 len
= packed_git_window_size
;
631 win
->len
= (size_t)len
;
632 pack_mapped
+= win
->len
;
633 while (packed_git_limit
< pack_mapped
634 && unuse_one_window(p
))
636 win
->base
= xmmap(NULL
, win
->len
,
637 PROT_READ
, MAP_PRIVATE
,
638 p
->pack_fd
, win
->offset
);
639 if (win
->base
== MAP_FAILED
)
640 die_errno("packfile %s cannot be mapped",
642 if (!win
->offset
&& win
->len
== p
->pack_size
647 if (pack_mapped
> peak_pack_mapped
)
648 peak_pack_mapped
= pack_mapped
;
649 if (pack_open_windows
> peak_pack_open_windows
)
650 peak_pack_open_windows
= pack_open_windows
;
651 win
->next
= p
->windows
;
655 if (win
!= *w_cursor
) {
656 win
->last_used
= pack_used_ctr
++;
660 offset
-= win
->offset
;
662 *left
= win
->len
- xsize_t(offset
);
663 return win
->base
+ offset
;
666 void unuse_pack(struct pack_window
**w_cursor
)
668 struct pack_window
*w
= *w_cursor
;
675 static void try_to_free_pack_memory(size_t size
)
677 release_pack_memory(size
);
680 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
682 static int have_set_try_to_free_routine
;
685 struct packed_git
*p
;
687 if (!have_set_try_to_free_routine
) {
688 have_set_try_to_free_routine
= 1;
689 set_try_to_free_routine(try_to_free_pack_memory
);
693 * Make sure a corresponding .pack file exists and that
694 * the index looks sane.
696 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
700 * ".promisor" is long enough to hold any suffix we're adding (and
701 * the use xsnprintf double-checks that)
703 alloc
= st_add3(path_len
, strlen(".promisor"), 1);
704 p
= alloc_packed_git(alloc
);
705 memcpy(p
->pack_name
, path
, path_len
);
707 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
708 if (!access(p
->pack_name
, F_OK
))
711 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".promisor");
712 if (!access(p
->pack_name
, F_OK
))
713 p
->pack_promisor
= 1;
715 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
716 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
721 /* ok, it looks sane as far as we can check without
722 * actually mapping the pack file.
724 p
->pack_size
= st
.st_size
;
725 p
->pack_local
= local
;
726 p
->mtime
= st
.st_mtime
;
727 if (path_len
< the_hash_algo
->hexsz
||
728 get_sha1_hex(path
+ path_len
- the_hash_algo
->hexsz
, p
->sha1
))
733 void install_packed_git(struct repository
*r
, struct packed_git
*pack
)
735 if (pack
->pack_fd
!= -1)
738 pack
->next
= r
->objects
->packed_git
;
739 r
->objects
->packed_git
= pack
;
742 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
744 static void report_helper(const struct string_list
*list
,
745 int seen_bits
, int first
, int last
)
747 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
750 for (; first
< last
; first
++)
751 report_garbage(seen_bits
, list
->items
[first
].string
);
754 static void report_pack_garbage(struct string_list
*list
)
756 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
761 string_list_sort(list
);
763 for (i
= 0; i
< list
->nr
; i
++) {
764 const char *path
= list
->items
[i
].string
;
766 strncmp(path
, list
->items
[first
].string
, baselen
)) {
767 report_helper(list
, seen_bits
, first
, i
);
772 const char *dot
= strrchr(path
, '.');
774 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
777 baselen
= dot
- path
+ 1;
780 if (!strcmp(path
+ baselen
, "pack"))
782 else if (!strcmp(path
+ baselen
, "idx"))
785 report_helper(list
, seen_bits
, first
, list
->nr
);
788 void for_each_file_in_pack_dir(const char *objdir
,
789 each_file_in_pack_dir_fn fn
,
792 struct strbuf path
= STRBUF_INIT
;
797 strbuf_addstr(&path
, objdir
);
798 strbuf_addstr(&path
, "/pack");
799 dir
= opendir(path
.buf
);
802 error_errno("unable to open object pack directory: %s",
804 strbuf_release(&path
);
807 strbuf_addch(&path
, '/');
808 dirnamelen
= path
.len
;
809 while ((de
= readdir(dir
)) != NULL
) {
810 if (is_dot_or_dotdot(de
->d_name
))
813 strbuf_setlen(&path
, dirnamelen
);
814 strbuf_addstr(&path
, de
->d_name
);
816 fn(path
.buf
, path
.len
, de
->d_name
, data
);
820 strbuf_release(&path
);
823 struct prepare_pack_data
{
824 struct repository
*r
;
825 struct string_list
*garbage
;
827 struct multi_pack_index
*m
;
830 static void prepare_pack(const char *full_name
, size_t full_name_len
,
831 const char *file_name
, void *_data
)
833 struct prepare_pack_data
*data
= (struct prepare_pack_data
*)_data
;
834 struct packed_git
*p
;
835 size_t base_len
= full_name_len
;
837 if (strip_suffix_mem(full_name
, &base_len
, ".idx") &&
838 !(data
->m
&& midx_contains_pack(data
->m
, file_name
))) {
839 /* Don't reopen a pack we already have. */
840 for (p
= data
->r
->objects
->packed_git
; p
; p
= p
->next
) {
842 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
844 !memcmp(p
->pack_name
, full_name
, len
))
849 p
= add_packed_git(full_name
, full_name_len
, data
->local
);
851 install_packed_git(data
->r
, p
);
858 if (!strcmp(file_name
, "multi-pack-index"))
860 if (ends_with(file_name
, ".idx") ||
861 ends_with(file_name
, ".pack") ||
862 ends_with(file_name
, ".bitmap") ||
863 ends_with(file_name
, ".keep") ||
864 ends_with(file_name
, ".promisor"))
865 string_list_append(data
->garbage
, full_name
);
867 report_garbage(PACKDIR_FILE_GARBAGE
, full_name
);
870 static void prepare_packed_git_one(struct repository
*r
, char *objdir
, int local
)
872 struct prepare_pack_data data
;
873 struct string_list garbage
= STRING_LIST_INIT_DUP
;
875 data
.m
= r
->objects
->multi_pack_index
;
877 /* look for the multi-pack-index for this object directory */
878 while (data
.m
&& strcmp(data
.m
->object_dir
, objdir
))
879 data
.m
= data
.m
->next
;
882 data
.garbage
= &garbage
;
885 for_each_file_in_pack_dir(objdir
, prepare_pack
, &data
);
887 report_pack_garbage(data
.garbage
);
888 string_list_clear(data
.garbage
, 0);
891 static void prepare_packed_git(struct repository
*r
);
893 * Give a fast, rough count of the number of objects in the repository. This
894 * ignores loose objects completely. If you have a lot of them, then either
895 * you should repack because your performance will be awful, or they are
896 * all unreachable objects about to be pruned, in which case they're not really
897 * interesting as a measure of repo size in the first place.
899 unsigned long approximate_object_count(void)
901 if (!the_repository
->objects
->approximate_object_count_valid
) {
903 struct multi_pack_index
*m
;
904 struct packed_git
*p
;
906 prepare_packed_git(the_repository
);
908 for (m
= get_multi_pack_index(the_repository
); m
; m
= m
->next
)
909 count
+= m
->num_objects
;
910 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
911 if (open_pack_index(p
))
913 count
+= p
->num_objects
;
915 the_repository
->objects
->approximate_object_count
= count
;
917 return the_repository
->objects
->approximate_object_count
;
920 static void *get_next_packed_git(const void *p
)
922 return ((const struct packed_git
*)p
)->next
;
925 static void set_next_packed_git(void *p
, void *next
)
927 ((struct packed_git
*)p
)->next
= next
;
930 static int sort_pack(const void *a_
, const void *b_
)
932 const struct packed_git
*a
= a_
;
933 const struct packed_git
*b
= b_
;
937 * Local packs tend to contain objects specific to our
938 * variant of the project than remote ones. In addition,
939 * remote ones could be on a network mounted filesystem.
940 * Favor local ones for these reasons.
942 st
= a
->pack_local
- b
->pack_local
;
947 * Younger packs tend to contain more recent objects,
948 * and more recent objects tend to get accessed more
951 if (a
->mtime
< b
->mtime
)
953 else if (a
->mtime
== b
->mtime
)
958 static void rearrange_packed_git(struct repository
*r
)
960 r
->objects
->packed_git
= llist_mergesort(
961 r
->objects
->packed_git
, get_next_packed_git
,
962 set_next_packed_git
, sort_pack
);
965 static void prepare_packed_git_mru(struct repository
*r
)
967 struct packed_git
*p
;
969 INIT_LIST_HEAD(&r
->objects
->packed_git_mru
);
971 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
972 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
975 static void prepare_packed_git(struct repository
*r
)
977 struct object_directory
*odb
;
979 if (r
->objects
->packed_git_initialized
)
983 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
984 int local
= (odb
== r
->objects
->odb
);
985 prepare_multi_pack_index_one(r
, odb
->path
, local
);
986 prepare_packed_git_one(r
, odb
->path
, local
);
988 rearrange_packed_git(r
);
990 r
->objects
->all_packs
= NULL
;
992 prepare_packed_git_mru(r
);
993 r
->objects
->packed_git_initialized
= 1;
996 void reprepare_packed_git(struct repository
*r
)
998 struct object_directory
*odb
;
1000 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
)
1001 odb_clear_loose_cache(odb
);
1003 r
->objects
->approximate_object_count_valid
= 0;
1004 r
->objects
->packed_git_initialized
= 0;
1005 prepare_packed_git(r
);
1008 struct packed_git
*get_packed_git(struct repository
*r
)
1010 prepare_packed_git(r
);
1011 return r
->objects
->packed_git
;
1014 struct multi_pack_index
*get_multi_pack_index(struct repository
*r
)
1016 prepare_packed_git(r
);
1017 return r
->objects
->multi_pack_index
;
1020 struct packed_git
*get_all_packs(struct repository
*r
)
1022 prepare_packed_git(r
);
1024 if (!r
->objects
->all_packs
) {
1025 struct packed_git
*p
= r
->objects
->packed_git
;
1026 struct multi_pack_index
*m
;
1028 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
1030 for (i
= 0; i
< m
->num_packs
; i
++) {
1031 if (!prepare_midx_pack(m
, i
)) {
1032 m
->packs
[i
]->next
= p
;
1038 r
->objects
->all_packs
= p
;
1041 return r
->objects
->all_packs
;
1044 struct list_head
*get_packed_git_mru(struct repository
*r
)
1046 prepare_packed_git(r
);
1047 return &r
->objects
->packed_git_mru
;
1050 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1051 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1054 unsigned long size
, c
;
1055 unsigned long used
= 0;
1058 *type
= (c
>> 4) & 7;
1062 if (len
<= used
|| bitsizeof(long) <= shift
) {
1063 error("bad object header");
1068 size
+= (c
& 0x7f) << shift
;
1075 unsigned long get_size_from_delta(struct packed_git
*p
,
1076 struct pack_window
**w_curs
,
1079 const unsigned char *data
;
1080 unsigned char delta_head
[20], *in
;
1084 memset(&stream
, 0, sizeof(stream
));
1085 stream
.next_out
= delta_head
;
1086 stream
.avail_out
= sizeof(delta_head
);
1088 git_inflate_init(&stream
);
1090 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1091 stream
.next_in
= in
;
1092 st
= git_inflate(&stream
, Z_FINISH
);
1093 curpos
+= stream
.next_in
- in
;
1094 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1095 stream
.total_out
< sizeof(delta_head
));
1096 git_inflate_end(&stream
);
1097 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1098 error("delta data unpack-initial failed");
1102 /* Examine the initial part of the delta to figure out
1107 /* ignore base size */
1108 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1110 /* Read the result size */
1111 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1114 int unpack_object_header(struct packed_git
*p
,
1115 struct pack_window
**w_curs
,
1117 unsigned long *sizep
)
1119 unsigned char *base
;
1122 enum object_type type
;
1124 /* use_pack() assures us we have [base, base + 20) available
1125 * as a range that we can look at. (Its actually the hash
1126 * size that is assured.) With our object header encoding
1127 * the maximum deflated object size is 2^137, which is just
1128 * insane, so we know won't exceed what we have been given.
1130 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1131 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1140 void mark_bad_packed_object(struct packed_git
*p
, const unsigned char *sha1
)
1143 const unsigned hashsz
= the_hash_algo
->rawsz
;
1144 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1145 if (hasheq(sha1
, p
->bad_object_sha1
+ hashsz
* i
))
1147 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1148 st_mult(GIT_MAX_RAWSZ
,
1149 st_add(p
->num_bad_objects
, 1)));
1150 hashcpy(p
->bad_object_sha1
+ hashsz
* p
->num_bad_objects
, sha1
);
1151 p
->num_bad_objects
++;
1154 const struct packed_git
*has_packed_and_bad(struct repository
*r
,
1155 const unsigned char *sha1
)
1157 struct packed_git
*p
;
1160 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
1161 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1163 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
1168 static off_t
get_delta_base(struct packed_git
*p
,
1169 struct pack_window
**w_curs
,
1171 enum object_type type
,
1172 off_t delta_obj_offset
)
1174 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1177 /* use_pack() assured us we have [base_info, base_info + 20)
1178 * as a range that we can look at without walking off the
1179 * end of the mapped window. Its actually the hash size
1180 * that is assured. An OFS_DELTA longer than the hash size
1181 * is stupid, as then a REF_DELTA would be smaller to store.
1183 if (type
== OBJ_OFS_DELTA
) {
1185 unsigned char c
= base_info
[used
++];
1186 base_offset
= c
& 127;
1189 if (!base_offset
|| MSB(base_offset
, 7))
1190 return 0; /* overflow */
1191 c
= base_info
[used
++];
1192 base_offset
= (base_offset
<< 7) + (c
& 127);
1194 base_offset
= delta_obj_offset
- base_offset
;
1195 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1196 return 0; /* out of bound */
1198 } else if (type
== OBJ_REF_DELTA
) {
1199 /* The base entry _must_ be in the same pack */
1200 base_offset
= find_pack_entry_one(base_info
, p
);
1201 *curpos
+= the_hash_algo
->rawsz
;
1203 die("I am totally screwed");
1208 * Like get_delta_base above, but we return the sha1 instead of the pack
1209 * offset. This means it is cheaper for REF deltas (we do not have to do
1210 * the final object lookup), but more expensive for OFS deltas (we
1211 * have to load the revidx to convert the offset back into a sha1).
1213 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1214 struct pack_window
**w_curs
,
1216 enum object_type type
,
1217 off_t delta_obj_offset
)
1219 if (type
== OBJ_REF_DELTA
) {
1220 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1222 } else if (type
== OBJ_OFS_DELTA
) {
1223 struct revindex_entry
*revidx
;
1224 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1225 type
, delta_obj_offset
);
1230 revidx
= find_pack_revindex(p
, base_offset
);
1234 return nth_packed_object_sha1(p
, revidx
->nr
);
1239 static int retry_bad_packed_offset(struct repository
*r
,
1240 struct packed_git
*p
,
1244 struct revindex_entry
*revidx
;
1245 struct object_id oid
;
1246 revidx
= find_pack_revindex(p
, obj_offset
);
1249 nth_packed_object_oid(&oid
, p
, revidx
->nr
);
1250 mark_bad_packed_object(p
, oid
.hash
);
1251 type
= oid_object_info(r
, &oid
, NULL
);
1252 if (type
<= OBJ_NONE
)
1257 #define POI_STACK_PREALLOC 64
1259 static enum object_type
packed_to_object_type(struct repository
*r
,
1260 struct packed_git
*p
,
1262 enum object_type type
,
1263 struct pack_window
**w_curs
,
1266 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1267 off_t
*poi_stack
= small_poi_stack
;
1268 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1270 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1273 /* Push the object we're going to leave behind */
1274 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1275 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1276 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
1277 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1279 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1281 poi_stack
[poi_stack_nr
++] = obj_offset
;
1282 /* If parsing the base offset fails, just unwind */
1283 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1286 curpos
= obj_offset
= base_offset
;
1287 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1288 if (type
<= OBJ_NONE
) {
1289 /* If getting the base itself fails, we first
1290 * retry the base, otherwise unwind */
1291 type
= retry_bad_packed_offset(r
, p
, base_offset
);
1292 if (type
> OBJ_NONE
)
1306 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1307 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1312 if (poi_stack
!= small_poi_stack
)
1317 while (poi_stack_nr
) {
1318 obj_offset
= poi_stack
[--poi_stack_nr
];
1319 type
= retry_bad_packed_offset(r
, p
, obj_offset
);
1320 if (type
> OBJ_NONE
)
1327 static struct hashmap delta_base_cache
;
1328 static size_t delta_base_cached
;
1330 static LIST_HEAD(delta_base_cache_lru
);
1332 struct delta_base_cache_key
{
1333 struct packed_git
*p
;
1337 struct delta_base_cache_entry
{
1338 struct hashmap hash
;
1339 struct delta_base_cache_key key
;
1340 struct list_head lru
;
1343 enum object_type type
;
1346 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1350 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
1351 hash
+= (hash
>> 8) + (hash
>> 16);
1355 static struct delta_base_cache_entry
*
1356 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1358 struct hashmap_entry entry
;
1359 struct delta_base_cache_key key
;
1361 if (!delta_base_cache
.cmpfn
)
1364 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
1366 key
.base_offset
= base_offset
;
1367 return hashmap_get(&delta_base_cache
, &entry
, &key
);
1370 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
1371 const struct delta_base_cache_key
*b
)
1373 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
1376 static int delta_base_cache_hash_cmp(const void *unused_cmp_data
,
1377 const void *va
, const void *vb
,
1380 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
1381 const struct delta_base_cache_key
*key
= vkey
;
1383 return !delta_base_cache_key_eq(&a
->key
, key
);
1385 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
1388 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1390 return !!get_delta_base_cache_entry(p
, base_offset
);
1394 * Remove the entry from the cache, but do _not_ free the associated
1395 * entry data. The caller takes ownership of the "data" buffer, and
1396 * should copy out any fields it wants before detaching.
1398 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1400 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
1401 list_del(&ent
->lru
);
1402 delta_base_cached
-= ent
->size
;
1406 static void *cache_or_unpack_entry(struct repository
*r
, struct packed_git
*p
,
1407 off_t base_offset
, unsigned long *base_size
,
1408 enum object_type
*type
)
1410 struct delta_base_cache_entry
*ent
;
1412 ent
= get_delta_base_cache_entry(p
, base_offset
);
1414 return unpack_entry(r
, p
, base_offset
, type
, base_size
);
1419 *base_size
= ent
->size
;
1420 return xmemdupz(ent
->data
, ent
->size
);
1423 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1426 detach_delta_base_cache_entry(ent
);
1429 void clear_delta_base_cache(void)
1431 struct list_head
*lru
, *tmp
;
1432 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1433 struct delta_base_cache_entry
*entry
=
1434 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1435 release_delta_base_cache(entry
);
1439 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1440 void *base
, unsigned long base_size
, enum object_type type
)
1442 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
1443 struct list_head
*lru
, *tmp
;
1445 delta_base_cached
+= base_size
;
1447 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1448 struct delta_base_cache_entry
*f
=
1449 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1450 if (delta_base_cached
<= delta_base_cache_limit
)
1452 release_delta_base_cache(f
);
1456 ent
->key
.base_offset
= base_offset
;
1459 ent
->size
= base_size
;
1460 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
1462 if (!delta_base_cache
.cmpfn
)
1463 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, NULL
, 0);
1464 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
1465 hashmap_add(&delta_base_cache
, ent
);
1468 int packed_object_info(struct repository
*r
, struct packed_git
*p
,
1469 off_t obj_offset
, struct object_info
*oi
)
1471 struct pack_window
*w_curs
= NULL
;
1473 off_t curpos
= obj_offset
;
1474 enum object_type type
;
1477 * We always get the representation type, but only convert it to
1478 * a "real" type later if the caller is interested.
1481 *oi
->contentp
= cache_or_unpack_entry(r
, p
, obj_offset
, oi
->sizep
,
1486 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1489 if (!oi
->contentp
&& oi
->sizep
) {
1490 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1491 off_t tmp_pos
= curpos
;
1492 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1498 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1499 if (*oi
->sizep
== 0) {
1508 if (oi
->disk_sizep
) {
1509 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1510 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1513 if (oi
->typep
|| oi
->type_name
) {
1514 enum object_type ptot
;
1515 ptot
= packed_to_object_type(r
, p
, obj_offset
,
1516 type
, &w_curs
, curpos
);
1519 if (oi
->type_name
) {
1520 const char *tn
= type_name(ptot
);
1522 strbuf_addstr(oi
->type_name
, tn
);
1530 if (oi
->delta_base_sha1
) {
1531 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1532 const unsigned char *base
;
1534 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1541 hashcpy(oi
->delta_base_sha1
, base
);
1543 hashclr(oi
->delta_base_sha1
);
1546 oi
->whence
= in_delta_base_cache(p
, obj_offset
) ? OI_DBCACHED
:
1550 unuse_pack(&w_curs
);
1554 static void *unpack_compressed_entry(struct packed_git
*p
,
1555 struct pack_window
**w_curs
,
1561 unsigned char *buffer
, *in
;
1563 buffer
= xmallocz_gently(size
);
1566 memset(&stream
, 0, sizeof(stream
));
1567 stream
.next_out
= buffer
;
1568 stream
.avail_out
= size
+ 1;
1570 git_inflate_init(&stream
);
1572 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1573 stream
.next_in
= in
;
1574 st
= git_inflate(&stream
, Z_FINISH
);
1575 if (!stream
.avail_out
)
1576 break; /* the payload is larger than it should be */
1577 curpos
+= stream
.next_in
- in
;
1578 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1579 git_inflate_end(&stream
);
1580 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1585 /* versions of zlib can clobber unconsumed portion of outbuf */
1586 buffer
[size
] = '\0';
1591 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1593 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
1594 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
1595 p
->pack_name
, (uintmax_t)obj_offset
);
1598 int do_check_packed_object_crc
;
1600 #define UNPACK_ENTRY_STACK_PREALLOC 64
1601 struct unpack_entry_stack_ent
{
1607 static void *read_object(struct repository
*r
,
1608 const struct object_id
*oid
,
1609 enum object_type
*type
,
1610 unsigned long *size
)
1612 struct object_info oi
= OBJECT_INFO_INIT
;
1616 oi
.contentp
= &content
;
1618 if (oid_object_info_extended(r
, oid
, &oi
, 0) < 0)
1623 void *unpack_entry(struct repository
*r
, struct packed_git
*p
, off_t obj_offset
,
1624 enum object_type
*final_type
, unsigned long *final_size
)
1626 struct pack_window
*w_curs
= NULL
;
1627 off_t curpos
= obj_offset
;
1630 enum object_type type
;
1631 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
1632 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
1633 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
1634 int base_from_cache
= 0;
1636 write_pack_access_log(p
, obj_offset
);
1638 /* PHASE 1: drill down to the innermost base object */
1642 struct delta_base_cache_entry
*ent
;
1644 ent
= get_delta_base_cache_entry(p
, curpos
);
1649 detach_delta_base_cache_entry(ent
);
1650 base_from_cache
= 1;
1654 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1655 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1656 off_t len
= revidx
[1].offset
- obj_offset
;
1657 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1658 struct object_id oid
;
1659 nth_packed_object_oid(&oid
, p
, revidx
->nr
);
1660 error("bad packed object CRC for %s",
1662 mark_bad_packed_object(p
, oid
.hash
);
1668 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1669 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
1672 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1674 error("failed to validate delta base reference "
1675 "at offset %"PRIuMAX
" from %s",
1676 (uintmax_t)curpos
, p
->pack_name
);
1677 /* bail to phase 2, in hopes of recovery */
1682 /* push object, proceed to base */
1683 if (delta_stack_nr
>= delta_stack_alloc
1684 && delta_stack
== small_delta_stack
) {
1685 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
1686 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
1687 memcpy(delta_stack
, small_delta_stack
,
1688 sizeof(*delta_stack
)*delta_stack_nr
);
1690 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
1692 i
= delta_stack_nr
++;
1693 delta_stack
[i
].obj_offset
= obj_offset
;
1694 delta_stack
[i
].curpos
= curpos
;
1695 delta_stack
[i
].size
= size
;
1697 curpos
= obj_offset
= base_offset
;
1700 /* PHASE 2: handle the base */
1705 BUG("unpack_entry: left loop at a valid delta");
1711 if (!base_from_cache
)
1712 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
1716 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1717 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1720 /* PHASE 3: apply deltas in order */
1723 * 'data' holds the base data, or NULL if there was corruption
1725 while (delta_stack_nr
) {
1728 void *external_base
= NULL
;
1729 unsigned long delta_size
, base_size
= size
;
1735 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
1739 * We're probably in deep shit, but let's try to fetch
1740 * the required base anyway from another pack or loose.
1741 * This is costly but should happen only in the presence
1742 * of a corrupted pack, and is better than failing outright.
1744 struct revindex_entry
*revidx
;
1745 struct object_id base_oid
;
1746 revidx
= find_pack_revindex(p
, obj_offset
);
1748 nth_packed_object_oid(&base_oid
, p
, revidx
->nr
);
1749 error("failed to read delta base object %s"
1750 " at offset %"PRIuMAX
" from %s",
1751 oid_to_hex(&base_oid
), (uintmax_t)obj_offset
,
1753 mark_bad_packed_object(p
, base_oid
.hash
);
1754 base
= read_object(r
, &base_oid
, &type
, &base_size
);
1755 external_base
= base
;
1759 i
= --delta_stack_nr
;
1760 obj_offset
= delta_stack
[i
].obj_offset
;
1761 curpos
= delta_stack
[i
].curpos
;
1762 delta_size
= delta_stack
[i
].size
;
1767 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
1770 error("failed to unpack compressed delta "
1771 "at offset %"PRIuMAX
" from %s",
1772 (uintmax_t)curpos
, p
->pack_name
);
1774 free(external_base
);
1778 data
= patch_delta(base
, base_size
,
1779 delta_data
, delta_size
,
1783 * We could not apply the delta; warn the user, but keep going.
1784 * Our failure will be noticed either in the next iteration of
1785 * the loop, or if this is the final delta, in the caller when
1786 * we return NULL. Those code paths will take care of making
1787 * a more explicit warning and retrying with another copy of
1791 error("failed to apply delta");
1794 free(external_base
);
1803 unuse_pack(&w_curs
);
1805 if (delta_stack
!= small_delta_stack
)
1811 int bsearch_pack(const struct object_id
*oid
, const struct packed_git
*p
, uint32_t *result
)
1813 const unsigned char *index_fanout
= p
->index_data
;
1814 const unsigned char *index_lookup
;
1815 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1816 int index_lookup_width
;
1819 BUG("bsearch_pack called without a valid pack-index");
1821 index_lookup
= index_fanout
+ 4 * 256;
1822 if (p
->index_version
== 1) {
1823 index_lookup_width
= hashsz
+ 4;
1826 index_lookup_width
= hashsz
;
1831 return bsearch_hash(oid
->hash
, (const uint32_t*)index_fanout
,
1832 index_lookup
, index_lookup_width
, result
);
1835 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1838 const unsigned char *index
= p
->index_data
;
1839 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1841 if (open_pack_index(p
))
1843 index
= p
->index_data
;
1845 if (n
>= p
->num_objects
)
1848 if (p
->index_version
== 1) {
1849 return index
+ (hashsz
+ 4) * n
+ 4;
1852 return index
+ hashsz
* n
;
1856 const struct object_id
*nth_packed_object_oid(struct object_id
*oid
,
1857 struct packed_git
*p
,
1860 const unsigned char *hash
= nth_packed_object_sha1(p
, n
);
1863 hashcpy(oid
->hash
, hash
);
1867 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
1869 const unsigned char *ptr
= vptr
;
1870 const unsigned char *start
= p
->index_data
;
1871 const unsigned char *end
= start
+ p
->index_size
;
1873 die(_("offset before start of pack index for %s (corrupt index?)"),
1875 /* No need to check for underflow; .idx files must be at least 8 bytes */
1877 die(_("offset beyond end of pack index for %s (truncated index?)"),
1881 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1883 const unsigned char *index
= p
->index_data
;
1884 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1886 if (p
->index_version
== 1) {
1887 return ntohl(*((uint32_t *)(index
+ (hashsz
+ 4) * n
)));
1890 index
+= 8 + p
->num_objects
* (hashsz
+ 4);
1891 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1892 if (!(off
& 0x80000000))
1894 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1895 check_pack_index_ptr(p
, index
);
1896 return get_be64(index
);
1900 off_t
find_pack_entry_one(const unsigned char *sha1
,
1901 struct packed_git
*p
)
1903 const unsigned char *index
= p
->index_data
;
1904 struct object_id oid
;
1908 if (open_pack_index(p
))
1912 hashcpy(oid
.hash
, sha1
);
1913 if (bsearch_pack(&oid
, p
, &result
))
1914 return nth_packed_object_offset(p
, result
);
1918 int is_pack_valid(struct packed_git
*p
)
1920 /* An already open pack is known to be valid. */
1921 if (p
->pack_fd
!= -1)
1924 /* If the pack has one window completely covering the
1925 * file size, the pack is known to be valid even if
1926 * the descriptor is not currently open.
1929 struct pack_window
*w
= p
->windows
;
1931 if (!w
->offset
&& w
->len
== p
->pack_size
)
1935 /* Force the pack to open to prove its valid. */
1936 return !open_packed_git(p
);
1939 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1940 struct packed_git
*packs
)
1942 struct packed_git
*p
;
1944 for (p
= packs
; p
; p
= p
->next
) {
1945 if (find_pack_entry_one(sha1
, p
))
1952 static int fill_pack_entry(const struct object_id
*oid
,
1953 struct pack_entry
*e
,
1954 struct packed_git
*p
)
1958 if (p
->num_bad_objects
) {
1960 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1961 if (hasheq(oid
->hash
,
1962 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
1966 offset
= find_pack_entry_one(oid
->hash
, p
);
1971 * We are about to tell the caller where they can locate the
1972 * requested object. We better make sure the packfile is
1973 * still here and can be accessed before supplying that
1974 * answer, as it may have been deleted since the index was
1977 if (!is_pack_valid(p
))
1984 int find_pack_entry(struct repository
*r
, const struct object_id
*oid
, struct pack_entry
*e
)
1986 struct list_head
*pos
;
1987 struct multi_pack_index
*m
;
1989 prepare_packed_git(r
);
1990 if (!r
->objects
->packed_git
&& !r
->objects
->multi_pack_index
)
1993 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
1994 if (fill_midx_entry(oid
, e
, m
))
1998 list_for_each(pos
, &r
->objects
->packed_git_mru
) {
1999 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
2000 if (fill_pack_entry(oid
, e
, p
)) {
2001 list_move(&p
->mru
, &r
->objects
->packed_git_mru
);
2008 int has_object_pack(const struct object_id
*oid
)
2010 struct pack_entry e
;
2011 return find_pack_entry(the_repository
, oid
, &e
);
2014 int has_pack_index(const unsigned char *sha1
)
2017 if (stat(sha1_pack_index_name(sha1
), &st
))
2022 int for_each_object_in_pack(struct packed_git
*p
,
2023 each_packed_object_fn cb
, void *data
,
2024 enum for_each_object_flags flags
)
2029 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
)
2030 load_pack_revindex(p
);
2032 for (i
= 0; i
< p
->num_objects
; i
++) {
2034 struct object_id oid
;
2036 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
)
2037 pos
= p
->revindex
[i
].nr
;
2041 if (!nth_packed_object_oid(&oid
, p
, pos
))
2042 return error("unable to get sha1 of object %u in %s",
2045 r
= cb(&oid
, p
, pos
, data
);
2052 int for_each_packed_object(each_packed_object_fn cb
, void *data
,
2053 enum for_each_object_flags flags
)
2055 struct packed_git
*p
;
2057 int pack_errors
= 0;
2059 prepare_packed_git(the_repository
);
2060 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
2061 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
2063 if ((flags
& FOR_EACH_OBJECT_PROMISOR_ONLY
) &&
2066 if (open_pack_index(p
)) {
2070 r
= for_each_object_in_pack(p
, cb
, data
, flags
);
2074 return r
? r
: pack_errors
;
2077 static int add_promisor_object(const struct object_id
*oid
,
2078 struct packed_git
*pack
,
2082 struct oidset
*set
= set_
;
2083 struct object
*obj
= parse_object(the_repository
, oid
);
2087 oidset_insert(set
, oid
);
2090 * If this is a tree, commit, or tag, the objects it refers
2091 * to are also promisor objects. (Blobs refer to no objects->)
2093 if (obj
->type
== OBJ_TREE
) {
2094 struct tree
*tree
= (struct tree
*)obj
;
2095 struct tree_desc desc
;
2096 struct name_entry entry
;
2097 if (init_tree_desc_gently(&desc
, tree
->buffer
, tree
->size
))
2099 * Error messages are given when packs are
2100 * verified, so do not print any here.
2103 while (tree_entry_gently(&desc
, &entry
))
2104 oidset_insert(set
, &entry
.oid
);
2105 } else if (obj
->type
== OBJ_COMMIT
) {
2106 struct commit
*commit
= (struct commit
*) obj
;
2107 struct commit_list
*parents
= commit
->parents
;
2109 oidset_insert(set
, get_commit_tree_oid(commit
));
2110 for (; parents
; parents
= parents
->next
)
2111 oidset_insert(set
, &parents
->item
->object
.oid
);
2112 } else if (obj
->type
== OBJ_TAG
) {
2113 struct tag
*tag
= (struct tag
*) obj
;
2114 oidset_insert(set
, &tag
->tagged
->oid
);
2119 int is_promisor_object(const struct object_id
*oid
)
2121 static struct oidset promisor_objects
;
2122 static int promisor_objects_prepared
;
2124 if (!promisor_objects_prepared
) {
2125 if (repository_format_partial_clone
) {
2126 for_each_packed_object(add_promisor_object
,
2128 FOR_EACH_OBJECT_PROMISOR_ONLY
);
2130 promisor_objects_prepared
= 1;
2132 return oidset_contains(&promisor_objects
, oid
);