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
)
83 struct pack_idx_header
*hdr
;
85 uint32_t version
, nr
, i
, *index
;
86 int fd
= git_open(path
);
88 const unsigned int hashsz
= the_hash_algo
->rawsz
;
96 idx_size
= xsize_t(st
.st_size
);
97 if (idx_size
< 4 * 256 + hashsz
+ hashsz
) {
99 return error("index file %s is too small", path
);
101 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
105 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
106 version
= ntohl(hdr
->idx_version
);
107 if (version
< 2 || version
> 2) {
108 munmap(idx_map
, idx_size
);
109 return error("index file %s is version %"PRIu32
110 " and is not supported by this binary"
111 " (try upgrading GIT to a newer version)",
120 index
+= 2; /* skip index header */
121 for (i
= 0; i
< 256; i
++) {
122 uint32_t n
= ntohl(index
[i
]);
124 munmap(idx_map
, idx_size
);
125 return error("non-monotonic index %s", path
);
133 * - 256 index entries 4 bytes each
134 * - 24-byte entries * nr (object ID + 4-byte offset)
135 * - hash of the packfile
138 if (idx_size
!= 4*256 + nr
* (hashsz
+ 4) + hashsz
+ hashsz
) {
139 munmap(idx_map
, idx_size
);
140 return error("wrong index v1 file size in %s", path
);
142 } else if (version
== 2) {
145 * - 8 bytes of header
146 * - 256 index entries 4 bytes each
147 * - object ID entry * nr
148 * - 4-byte crc entry * nr
149 * - 4-byte offset entry * nr
150 * - hash of the packfile
152 * And after the 4-byte offset table might be a
153 * variable sized table containing 8-byte entries
154 * for offsets larger than 2^31.
156 unsigned long min_size
= 8 + 4*256 + nr
*(hashsz
+ 4 + 4) + hashsz
+ hashsz
;
157 unsigned long max_size
= min_size
;
159 max_size
+= (nr
- 1)*8;
160 if (idx_size
< min_size
|| idx_size
> max_size
) {
161 munmap(idx_map
, idx_size
);
162 return error("wrong index v2 file size in %s", path
);
164 if (idx_size
!= min_size
&&
166 * make sure we can deal with large pack offsets.
167 * 31-bit signed offset won't be enough, neither
168 * 32-bit unsigned one will be.
170 (sizeof(off_t
) <= 4)) {
171 munmap(idx_map
, idx_size
);
172 return error("pack too large for current definition of off_t in %s", path
);
176 p
->index_version
= version
;
177 p
->index_data
= idx_map
;
178 p
->index_size
= idx_size
;
183 int open_pack_index(struct packed_git
*p
)
192 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
193 BUG("pack_name does not end in .pack");
194 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
195 ret
= check_packed_git_idx(idx_name
, p
);
200 uint32_t get_pack_fanout(struct packed_git
*p
, uint32_t value
)
202 const uint32_t *level1_ofs
= p
->index_data
;
205 if (open_pack_index(p
))
207 level1_ofs
= p
->index_data
;
210 if (p
->index_version
> 1) {
214 return ntohl(level1_ofs
[value
]);
217 static struct packed_git
*alloc_packed_git(int extra
)
219 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
220 memset(p
, 0, sizeof(*p
));
225 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
227 const char *path
= sha1_pack_name(sha1
);
228 size_t alloc
= st_add(strlen(path
), 1);
229 struct packed_git
*p
= alloc_packed_git(alloc
);
231 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
232 hashcpy(p
->sha1
, sha1
);
233 if (check_packed_git_idx(idx_path
, p
)) {
241 static void scan_windows(struct packed_git
*p
,
242 struct packed_git
**lru_p
,
243 struct pack_window
**lru_w
,
244 struct pack_window
**lru_l
)
246 struct pack_window
*w
, *w_l
;
248 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
250 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
260 static int unuse_one_window(struct packed_git
*current
)
262 struct packed_git
*p
, *lru_p
= NULL
;
263 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
266 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
267 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
)
268 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
270 munmap(lru_w
->base
, lru_w
->len
);
271 pack_mapped
-= lru_w
->len
;
273 lru_l
->next
= lru_w
->next
;
275 lru_p
->windows
= lru_w
->next
;
283 void release_pack_memory(size_t need
)
285 size_t cur
= pack_mapped
;
286 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
290 void close_pack_windows(struct packed_git
*p
)
293 struct pack_window
*w
= p
->windows
;
296 die("pack '%s' still has open windows to it",
298 munmap(w
->base
, w
->len
);
299 pack_mapped
-= w
->len
;
301 p
->windows
= w
->next
;
306 static int close_pack_fd(struct packed_git
*p
)
318 void close_pack_index(struct packed_git
*p
)
321 munmap((void *)p
->index_data
, p
->index_size
);
322 p
->index_data
= NULL
;
326 void close_pack(struct packed_git
*p
)
328 close_pack_windows(p
);
333 void close_all_packs(struct raw_object_store
*o
)
335 struct packed_git
*p
;
337 for (p
= o
->packed_git
; p
; p
= p
->next
)
339 BUG("want to close pack marked 'do-not-close'");
345 * The LRU pack is the one with the oldest MRU window, preferring packs
346 * with no used windows, or the oldest mtime if it has no windows allocated.
348 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
350 struct pack_window
*w
, *this_mru_w
;
351 int has_windows_inuse
= 0;
354 * Reject this pack if it has windows and the previously selected
355 * one does not. If this pack does not have windows, reject
356 * it if the pack file is newer than the previously selected one.
358 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
361 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
363 * Reject this pack if any of its windows are in use,
364 * but the previously selected pack did not have any
365 * inuse windows. Otherwise, record that this pack
366 * has windows in use.
369 if (*accept_windows_inuse
)
370 has_windows_inuse
= 1;
375 if (w
->last_used
> this_mru_w
->last_used
)
379 * Reject this pack if it has windows that have been
380 * used more recently than the previously selected pack.
381 * If the previously selected pack had windows inuse and
382 * we have not encountered a window in this pack that is
383 * inuse, skip this check since we prefer a pack with no
384 * inuse windows to one that has inuse windows.
386 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
387 this_mru_w
->last_used
> (*mru_w
)->last_used
)
396 *accept_windows_inuse
= has_windows_inuse
;
399 static int close_one_pack(void)
401 struct packed_git
*p
, *lru_p
= NULL
;
402 struct pack_window
*mru_w
= NULL
;
403 int accept_windows_inuse
= 1;
405 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
406 if (p
->pack_fd
== -1)
408 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
412 return close_pack_fd(lru_p
);
417 static unsigned int get_max_fd_limit(void)
423 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
430 long open_max
= sysconf(_SC_OPEN_MAX
);
434 * Otherwise, we got -1 for one of the two
437 * (1) sysconf() did not understand _SC_OPEN_MAX
438 * and signaled an error with -1; or
439 * (2) sysconf() said there is no limit.
441 * We _could_ clear errno before calling sysconf() to
442 * tell these two cases apart and return a huge number
443 * in the latter case to let the caller cap it to a
444 * value that is not so selfish, but letting the
445 * fallback OPEN_MAX codepath take care of these cases
454 return 1; /* see the caller ;-) */
459 * Do not call this directly as this leaks p->pack_fd on error return;
460 * call open_packed_git() instead.
462 static int open_packed_git_1(struct packed_git
*p
)
465 struct pack_header hdr
;
466 unsigned char hash
[GIT_MAX_RAWSZ
];
467 unsigned char *idx_hash
;
470 const unsigned hashsz
= the_hash_algo
->rawsz
;
472 if (!p
->index_data
) {
473 struct multi_pack_index
*m
;
474 const char *pack_name
= strrchr(p
->pack_name
, '/');
476 for (m
= the_repository
->objects
->multi_pack_index
;
478 if (midx_contains_pack(m
, pack_name
))
482 if (!m
&& open_pack_index(p
))
483 return error("packfile %s index unavailable", p
->pack_name
);
487 unsigned int max_fds
= get_max_fd_limit();
489 /* Save 3 for stdin/stdout/stderr, 22 for work */
491 pack_max_fds
= max_fds
- 25;
496 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
499 p
->pack_fd
= git_open(p
->pack_name
);
500 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
504 /* If we created the struct before we had the pack we lack size. */
506 if (!S_ISREG(st
.st_mode
))
507 return error("packfile %s not a regular file", p
->pack_name
);
508 p
->pack_size
= st
.st_size
;
509 } else if (p
->pack_size
!= st
.st_size
)
510 return error("packfile %s size changed", p
->pack_name
);
512 /* We leave these file descriptors open with sliding mmap;
513 * there is no point keeping them open across exec(), though.
515 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
517 return error("cannot determine file descriptor flags");
518 fd_flag
|= FD_CLOEXEC
;
519 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
520 return error("cannot set FD_CLOEXEC");
522 /* Verify we recognize this pack file format. */
523 read_result
= read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
));
525 return error_errno("error reading from %s", p
->pack_name
);
526 if (read_result
!= sizeof(hdr
))
527 return error("file %s is far too short to be a packfile", p
->pack_name
);
528 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
529 return error("file %s is not a GIT packfile", p
->pack_name
);
530 if (!pack_version_ok(hdr
.hdr_version
))
531 return error("packfile %s is version %"PRIu32
" and not"
532 " supported (try upgrading GIT to a newer version)",
533 p
->pack_name
, ntohl(hdr
.hdr_version
));
535 /* Skip index checking if in multi-pack-index */
539 /* Verify the pack matches its index. */
540 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
541 return error("packfile %s claims to have %"PRIu32
" objects"
542 " while index indicates %"PRIu32
" objects",
543 p
->pack_name
, ntohl(hdr
.hdr_entries
),
545 if (lseek(p
->pack_fd
, p
->pack_size
- hashsz
, SEEK_SET
) == -1)
546 return error("end of packfile %s is unavailable", p
->pack_name
);
547 read_result
= read_in_full(p
->pack_fd
, hash
, hashsz
);
549 return error_errno("error reading from %s", p
->pack_name
);
550 if (read_result
!= hashsz
)
551 return error("packfile %s signature is unavailable", p
->pack_name
);
552 idx_hash
= ((unsigned char *)p
->index_data
) + p
->index_size
- hashsz
* 2;
553 if (hashcmp(hash
, idx_hash
))
554 return error("packfile %s does not match index", p
->pack_name
);
558 static int open_packed_git(struct packed_git
*p
)
560 if (!open_packed_git_1(p
))
566 static int in_window(struct pack_window
*win
, off_t offset
)
568 /* We must promise at least one full hash after the
569 * offset is available from this window, otherwise the offset
570 * is not actually in this window and a different window (which
571 * has that one hash excess) must be used. This is to support
572 * the object header and delta base parsing routines below.
574 off_t win_off
= win
->offset
;
575 return win_off
<= offset
576 && (offset
+ the_hash_algo
->rawsz
) <= (win_off
+ win
->len
);
579 unsigned char *use_pack(struct packed_git
*p
,
580 struct pack_window
**w_cursor
,
584 struct pack_window
*win
= *w_cursor
;
586 /* Since packfiles end in a hash of their content and it's
587 * pointless to ask for an offset into the middle of that
588 * hash, and the in_window function above wouldn't match
589 * don't allow an offset too close to the end of the file.
591 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
592 die("packfile %s cannot be accessed", p
->pack_name
);
593 if (offset
> (p
->pack_size
- the_hash_algo
->rawsz
))
594 die("offset beyond end of packfile (truncated pack?)");
596 die(_("offset before end of packfile (broken .idx?)"));
598 if (!win
|| !in_window(win
, offset
)) {
601 for (win
= p
->windows
; win
; win
= win
->next
) {
602 if (in_window(win
, offset
))
606 size_t window_align
= packed_git_window_size
/ 2;
609 if (p
->pack_fd
== -1 && open_packed_git(p
))
610 die("packfile %s cannot be accessed", p
->pack_name
);
612 win
= xcalloc(1, sizeof(*win
));
613 win
->offset
= (offset
/ window_align
) * window_align
;
614 len
= p
->pack_size
- win
->offset
;
615 if (len
> packed_git_window_size
)
616 len
= packed_git_window_size
;
617 win
->len
= (size_t)len
;
618 pack_mapped
+= win
->len
;
619 while (packed_git_limit
< pack_mapped
620 && unuse_one_window(p
))
622 win
->base
= xmmap(NULL
, win
->len
,
623 PROT_READ
, MAP_PRIVATE
,
624 p
->pack_fd
, win
->offset
);
625 if (win
->base
== MAP_FAILED
)
626 die_errno("packfile %s cannot be mapped",
628 if (!win
->offset
&& win
->len
== p
->pack_size
633 if (pack_mapped
> peak_pack_mapped
)
634 peak_pack_mapped
= pack_mapped
;
635 if (pack_open_windows
> peak_pack_open_windows
)
636 peak_pack_open_windows
= pack_open_windows
;
637 win
->next
= p
->windows
;
641 if (win
!= *w_cursor
) {
642 win
->last_used
= pack_used_ctr
++;
646 offset
-= win
->offset
;
648 *left
= win
->len
- xsize_t(offset
);
649 return win
->base
+ offset
;
652 void unuse_pack(struct pack_window
**w_cursor
)
654 struct pack_window
*w
= *w_cursor
;
661 static void try_to_free_pack_memory(size_t size
)
663 release_pack_memory(size
);
666 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
668 static int have_set_try_to_free_routine
;
671 struct packed_git
*p
;
673 if (!have_set_try_to_free_routine
) {
674 have_set_try_to_free_routine
= 1;
675 set_try_to_free_routine(try_to_free_pack_memory
);
679 * Make sure a corresponding .pack file exists and that
680 * the index looks sane.
682 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
686 * ".promisor" is long enough to hold any suffix we're adding (and
687 * the use xsnprintf double-checks that)
689 alloc
= st_add3(path_len
, strlen(".promisor"), 1);
690 p
= alloc_packed_git(alloc
);
691 memcpy(p
->pack_name
, path
, path_len
);
693 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
694 if (!access(p
->pack_name
, F_OK
))
697 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".promisor");
698 if (!access(p
->pack_name
, F_OK
))
699 p
->pack_promisor
= 1;
701 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
702 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
707 /* ok, it looks sane as far as we can check without
708 * actually mapping the pack file.
710 p
->pack_size
= st
.st_size
;
711 p
->pack_local
= local
;
712 p
->mtime
= st
.st_mtime
;
713 if (path_len
< the_hash_algo
->hexsz
||
714 get_sha1_hex(path
+ path_len
- the_hash_algo
->hexsz
, p
->sha1
))
719 void install_packed_git(struct repository
*r
, struct packed_git
*pack
)
721 if (pack
->pack_fd
!= -1)
724 pack
->next
= r
->objects
->packed_git
;
725 r
->objects
->packed_git
= pack
;
728 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
730 static void report_helper(const struct string_list
*list
,
731 int seen_bits
, int first
, int last
)
733 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
736 for (; first
< last
; first
++)
737 report_garbage(seen_bits
, list
->items
[first
].string
);
740 static void report_pack_garbage(struct string_list
*list
)
742 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
747 string_list_sort(list
);
749 for (i
= 0; i
< list
->nr
; i
++) {
750 const char *path
= list
->items
[i
].string
;
752 strncmp(path
, list
->items
[first
].string
, baselen
)) {
753 report_helper(list
, seen_bits
, first
, i
);
758 const char *dot
= strrchr(path
, '.');
760 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
763 baselen
= dot
- path
+ 1;
766 if (!strcmp(path
+ baselen
, "pack"))
768 else if (!strcmp(path
+ baselen
, "idx"))
771 report_helper(list
, seen_bits
, first
, list
->nr
);
774 void for_each_file_in_pack_dir(const char *objdir
,
775 each_file_in_pack_dir_fn fn
,
778 struct strbuf path
= STRBUF_INIT
;
783 strbuf_addstr(&path
, objdir
);
784 strbuf_addstr(&path
, "/pack");
785 dir
= opendir(path
.buf
);
788 error_errno("unable to open object pack directory: %s",
790 strbuf_release(&path
);
793 strbuf_addch(&path
, '/');
794 dirnamelen
= path
.len
;
795 while ((de
= readdir(dir
)) != NULL
) {
796 if (is_dot_or_dotdot(de
->d_name
))
799 strbuf_setlen(&path
, dirnamelen
);
800 strbuf_addstr(&path
, de
->d_name
);
802 fn(path
.buf
, path
.len
, de
->d_name
, data
);
806 strbuf_release(&path
);
809 struct prepare_pack_data
{
810 struct repository
*r
;
811 struct string_list
*garbage
;
813 struct multi_pack_index
*m
;
816 static void prepare_pack(const char *full_name
, size_t full_name_len
,
817 const char *file_name
, void *_data
)
819 struct prepare_pack_data
*data
= (struct prepare_pack_data
*)_data
;
820 struct packed_git
*p
;
821 size_t base_len
= full_name_len
;
823 if (strip_suffix_mem(full_name
, &base_len
, ".idx")) {
824 if (data
->m
&& midx_contains_pack(data
->m
, file_name
))
826 /* Don't reopen a pack we already have. */
827 for (p
= data
->r
->objects
->packed_git
; p
; p
= p
->next
) {
829 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
831 !memcmp(p
->pack_name
, full_name
, len
))
836 p
= add_packed_git(full_name
, full_name_len
, data
->local
);
838 install_packed_git(data
->r
, p
);
845 if (ends_with(file_name
, ".idx") ||
846 ends_with(file_name
, ".pack") ||
847 ends_with(file_name
, ".bitmap") ||
848 ends_with(file_name
, ".keep") ||
849 ends_with(file_name
, ".promisor"))
850 string_list_append(data
->garbage
, full_name
);
852 report_garbage(PACKDIR_FILE_GARBAGE
, full_name
);
855 static void prepare_packed_git_one(struct repository
*r
, char *objdir
, int local
)
857 struct prepare_pack_data data
;
858 struct string_list garbage
= STRING_LIST_INIT_DUP
;
860 data
.m
= r
->objects
->multi_pack_index
;
862 /* look for the multi-pack-index for this object directory */
863 while (data
.m
&& strcmp(data
.m
->object_dir
, objdir
))
864 data
.m
= data
.m
->next
;
867 data
.garbage
= &garbage
;
870 for_each_file_in_pack_dir(objdir
, prepare_pack
, &data
);
872 report_pack_garbage(data
.garbage
);
873 string_list_clear(data
.garbage
, 0);
876 static void prepare_packed_git(struct repository
*r
);
878 * Give a fast, rough count of the number of objects in the repository. This
879 * ignores loose objects completely. If you have a lot of them, then either
880 * you should repack because your performance will be awful, or they are
881 * all unreachable objects about to be pruned, in which case they're not really
882 * interesting as a measure of repo size in the first place.
884 unsigned long approximate_object_count(void)
886 if (!the_repository
->objects
->approximate_object_count_valid
) {
888 struct multi_pack_index
*m
;
889 struct packed_git
*p
;
891 prepare_packed_git(the_repository
);
893 for (m
= get_multi_pack_index(the_repository
); m
; m
= m
->next
)
894 count
+= m
->num_objects
;
895 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
896 if (open_pack_index(p
))
898 count
+= p
->num_objects
;
900 the_repository
->objects
->approximate_object_count
= count
;
902 return the_repository
->objects
->approximate_object_count
;
905 static void *get_next_packed_git(const void *p
)
907 return ((const struct packed_git
*)p
)->next
;
910 static void set_next_packed_git(void *p
, void *next
)
912 ((struct packed_git
*)p
)->next
= next
;
915 static int sort_pack(const void *a_
, const void *b_
)
917 const struct packed_git
*a
= a_
;
918 const struct packed_git
*b
= b_
;
922 * Local packs tend to contain objects specific to our
923 * variant of the project than remote ones. In addition,
924 * remote ones could be on a network mounted filesystem.
925 * Favor local ones for these reasons.
927 st
= a
->pack_local
- b
->pack_local
;
932 * Younger packs tend to contain more recent objects,
933 * and more recent objects tend to get accessed more
936 if (a
->mtime
< b
->mtime
)
938 else if (a
->mtime
== b
->mtime
)
943 static void rearrange_packed_git(struct repository
*r
)
945 r
->objects
->packed_git
= llist_mergesort(
946 r
->objects
->packed_git
, get_next_packed_git
,
947 set_next_packed_git
, sort_pack
);
950 static void prepare_packed_git_mru(struct repository
*r
)
952 struct packed_git
*p
;
954 INIT_LIST_HEAD(&r
->objects
->packed_git_mru
);
956 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
957 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
960 static void prepare_packed_git(struct repository
*r
)
962 struct alternate_object_database
*alt
;
964 if (r
->objects
->packed_git_initialized
)
966 prepare_multi_pack_index_one(r
, r
->objects
->objectdir
, 1);
967 prepare_packed_git_one(r
, r
->objects
->objectdir
, 1);
969 for (alt
= r
->objects
->alt_odb_list
; alt
; alt
= alt
->next
) {
970 prepare_multi_pack_index_one(r
, alt
->path
, 0);
971 prepare_packed_git_one(r
, alt
->path
, 0);
973 rearrange_packed_git(r
);
974 prepare_packed_git_mru(r
);
975 r
->objects
->packed_git_initialized
= 1;
978 void reprepare_packed_git(struct repository
*r
)
980 r
->objects
->approximate_object_count_valid
= 0;
981 r
->objects
->packed_git_initialized
= 0;
982 prepare_packed_git(r
);
985 struct packed_git
*get_packed_git(struct repository
*r
)
987 prepare_packed_git(r
);
988 return r
->objects
->packed_git
;
991 struct multi_pack_index
*get_multi_pack_index(struct repository
*r
)
993 prepare_packed_git(r
);
994 return r
->objects
->multi_pack_index
;
997 struct list_head
*get_packed_git_mru(struct repository
*r
)
999 prepare_packed_git(r
);
1000 return &r
->objects
->packed_git_mru
;
1003 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1004 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1007 unsigned long size
, c
;
1008 unsigned long used
= 0;
1011 *type
= (c
>> 4) & 7;
1015 if (len
<= used
|| bitsizeof(long) <= shift
) {
1016 error("bad object header");
1021 size
+= (c
& 0x7f) << shift
;
1028 unsigned long get_size_from_delta(struct packed_git
*p
,
1029 struct pack_window
**w_curs
,
1032 const unsigned char *data
;
1033 unsigned char delta_head
[20], *in
;
1037 memset(&stream
, 0, sizeof(stream
));
1038 stream
.next_out
= delta_head
;
1039 stream
.avail_out
= sizeof(delta_head
);
1041 git_inflate_init(&stream
);
1043 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1044 stream
.next_in
= in
;
1045 st
= git_inflate(&stream
, Z_FINISH
);
1046 curpos
+= stream
.next_in
- in
;
1047 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1048 stream
.total_out
< sizeof(delta_head
));
1049 git_inflate_end(&stream
);
1050 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1051 error("delta data unpack-initial failed");
1055 /* Examine the initial part of the delta to figure out
1060 /* ignore base size */
1061 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1063 /* Read the result size */
1064 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1067 int unpack_object_header(struct packed_git
*p
,
1068 struct pack_window
**w_curs
,
1070 unsigned long *sizep
)
1072 unsigned char *base
;
1075 enum object_type type
;
1077 /* use_pack() assures us we have [base, base + 20) available
1078 * as a range that we can look at. (Its actually the hash
1079 * size that is assured.) With our object header encoding
1080 * the maximum deflated object size is 2^137, which is just
1081 * insane, so we know won't exceed what we have been given.
1083 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1084 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1093 void mark_bad_packed_object(struct packed_git
*p
, const unsigned char *sha1
)
1096 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1097 if (!hashcmp(sha1
, p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* i
))
1099 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1100 st_mult(GIT_MAX_RAWSZ
,
1101 st_add(p
->num_bad_objects
, 1)));
1102 hashcpy(p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* p
->num_bad_objects
, sha1
);
1103 p
->num_bad_objects
++;
1106 const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1108 struct packed_git
*p
;
1111 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
)
1112 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1114 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
1119 static off_t
get_delta_base(struct packed_git
*p
,
1120 struct pack_window
**w_curs
,
1122 enum object_type type
,
1123 off_t delta_obj_offset
)
1125 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1128 /* use_pack() assured us we have [base_info, base_info + 20)
1129 * as a range that we can look at without walking off the
1130 * end of the mapped window. Its actually the hash size
1131 * that is assured. An OFS_DELTA longer than the hash size
1132 * is stupid, as then a REF_DELTA would be smaller to store.
1134 if (type
== OBJ_OFS_DELTA
) {
1136 unsigned char c
= base_info
[used
++];
1137 base_offset
= c
& 127;
1140 if (!base_offset
|| MSB(base_offset
, 7))
1141 return 0; /* overflow */
1142 c
= base_info
[used
++];
1143 base_offset
= (base_offset
<< 7) + (c
& 127);
1145 base_offset
= delta_obj_offset
- base_offset
;
1146 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1147 return 0; /* out of bound */
1149 } else if (type
== OBJ_REF_DELTA
) {
1150 /* The base entry _must_ be in the same pack */
1151 base_offset
= find_pack_entry_one(base_info
, p
);
1152 *curpos
+= the_hash_algo
->rawsz
;
1154 die("I am totally screwed");
1159 * Like get_delta_base above, but we return the sha1 instead of the pack
1160 * offset. This means it is cheaper for REF deltas (we do not have to do
1161 * the final object lookup), but more expensive for OFS deltas (we
1162 * have to load the revidx to convert the offset back into a sha1).
1164 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1165 struct pack_window
**w_curs
,
1167 enum object_type type
,
1168 off_t delta_obj_offset
)
1170 if (type
== OBJ_REF_DELTA
) {
1171 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1173 } else if (type
== OBJ_OFS_DELTA
) {
1174 struct revindex_entry
*revidx
;
1175 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1176 type
, delta_obj_offset
);
1181 revidx
= find_pack_revindex(p
, base_offset
);
1185 return nth_packed_object_sha1(p
, revidx
->nr
);
1190 static int retry_bad_packed_offset(struct repository
*r
,
1191 struct packed_git
*p
,
1195 struct revindex_entry
*revidx
;
1196 struct object_id oid
;
1197 revidx
= find_pack_revindex(p
, obj_offset
);
1200 nth_packed_object_oid(&oid
, p
, revidx
->nr
);
1201 mark_bad_packed_object(p
, oid
.hash
);
1202 type
= oid_object_info(r
, &oid
, NULL
);
1203 if (type
<= OBJ_NONE
)
1208 #define POI_STACK_PREALLOC 64
1210 static enum object_type
packed_to_object_type(struct repository
*r
,
1211 struct packed_git
*p
,
1213 enum object_type type
,
1214 struct pack_window
**w_curs
,
1217 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1218 off_t
*poi_stack
= small_poi_stack
;
1219 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1221 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1224 /* Push the object we're going to leave behind */
1225 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1226 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1227 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
1228 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1230 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1232 poi_stack
[poi_stack_nr
++] = obj_offset
;
1233 /* If parsing the base offset fails, just unwind */
1234 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1237 curpos
= obj_offset
= base_offset
;
1238 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1239 if (type
<= OBJ_NONE
) {
1240 /* If getting the base itself fails, we first
1241 * retry the base, otherwise unwind */
1242 type
= retry_bad_packed_offset(r
, p
, base_offset
);
1243 if (type
> OBJ_NONE
)
1257 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1258 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1263 if (poi_stack
!= small_poi_stack
)
1268 while (poi_stack_nr
) {
1269 obj_offset
= poi_stack
[--poi_stack_nr
];
1270 type
= retry_bad_packed_offset(r
, p
, obj_offset
);
1271 if (type
> OBJ_NONE
)
1278 static struct hashmap delta_base_cache
;
1279 static size_t delta_base_cached
;
1281 static LIST_HEAD(delta_base_cache_lru
);
1283 struct delta_base_cache_key
{
1284 struct packed_git
*p
;
1288 struct delta_base_cache_entry
{
1289 struct hashmap hash
;
1290 struct delta_base_cache_key key
;
1291 struct list_head lru
;
1294 enum object_type type
;
1297 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1301 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
1302 hash
+= (hash
>> 8) + (hash
>> 16);
1306 static struct delta_base_cache_entry
*
1307 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1309 struct hashmap_entry entry
;
1310 struct delta_base_cache_key key
;
1312 if (!delta_base_cache
.cmpfn
)
1315 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
1317 key
.base_offset
= base_offset
;
1318 return hashmap_get(&delta_base_cache
, &entry
, &key
);
1321 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
1322 const struct delta_base_cache_key
*b
)
1324 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
1327 static int delta_base_cache_hash_cmp(const void *unused_cmp_data
,
1328 const void *va
, const void *vb
,
1331 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
1332 const struct delta_base_cache_key
*key
= vkey
;
1334 return !delta_base_cache_key_eq(&a
->key
, key
);
1336 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
1339 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1341 return !!get_delta_base_cache_entry(p
, base_offset
);
1345 * Remove the entry from the cache, but do _not_ free the associated
1346 * entry data. The caller takes ownership of the "data" buffer, and
1347 * should copy out any fields it wants before detaching.
1349 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1351 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
1352 list_del(&ent
->lru
);
1353 delta_base_cached
-= ent
->size
;
1357 static void *cache_or_unpack_entry(struct repository
*r
, struct packed_git
*p
,
1358 off_t base_offset
, unsigned long *base_size
,
1359 enum object_type
*type
)
1361 struct delta_base_cache_entry
*ent
;
1363 ent
= get_delta_base_cache_entry(p
, base_offset
);
1365 return unpack_entry(r
, p
, base_offset
, type
, base_size
);
1370 *base_size
= ent
->size
;
1371 return xmemdupz(ent
->data
, ent
->size
);
1374 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1377 detach_delta_base_cache_entry(ent
);
1380 void clear_delta_base_cache(void)
1382 struct list_head
*lru
, *tmp
;
1383 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1384 struct delta_base_cache_entry
*entry
=
1385 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1386 release_delta_base_cache(entry
);
1390 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1391 void *base
, unsigned long base_size
, enum object_type type
)
1393 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
1394 struct list_head
*lru
, *tmp
;
1396 delta_base_cached
+= base_size
;
1398 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1399 struct delta_base_cache_entry
*f
=
1400 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1401 if (delta_base_cached
<= delta_base_cache_limit
)
1403 release_delta_base_cache(f
);
1407 ent
->key
.base_offset
= base_offset
;
1410 ent
->size
= base_size
;
1411 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
1413 if (!delta_base_cache
.cmpfn
)
1414 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, NULL
, 0);
1415 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
1416 hashmap_add(&delta_base_cache
, ent
);
1419 int packed_object_info(struct repository
*r
, struct packed_git
*p
,
1420 off_t obj_offset
, struct object_info
*oi
)
1422 struct pack_window
*w_curs
= NULL
;
1424 off_t curpos
= obj_offset
;
1425 enum object_type type
;
1428 * We always get the representation type, but only convert it to
1429 * a "real" type later if the caller is interested.
1432 *oi
->contentp
= cache_or_unpack_entry(r
, p
, obj_offset
, oi
->sizep
,
1437 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1440 if (!oi
->contentp
&& oi
->sizep
) {
1441 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1442 off_t tmp_pos
= curpos
;
1443 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1449 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1450 if (*oi
->sizep
== 0) {
1459 if (oi
->disk_sizep
) {
1460 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1461 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1464 if (oi
->typep
|| oi
->type_name
) {
1465 enum object_type ptot
;
1466 ptot
= packed_to_object_type(r
, p
, obj_offset
,
1467 type
, &w_curs
, curpos
);
1470 if (oi
->type_name
) {
1471 const char *tn
= type_name(ptot
);
1473 strbuf_addstr(oi
->type_name
, tn
);
1481 if (oi
->delta_base_sha1
) {
1482 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1483 const unsigned char *base
;
1485 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1492 hashcpy(oi
->delta_base_sha1
, base
);
1494 hashclr(oi
->delta_base_sha1
);
1497 oi
->whence
= in_delta_base_cache(p
, obj_offset
) ? OI_DBCACHED
:
1501 unuse_pack(&w_curs
);
1505 static void *unpack_compressed_entry(struct packed_git
*p
,
1506 struct pack_window
**w_curs
,
1512 unsigned char *buffer
, *in
;
1514 buffer
= xmallocz_gently(size
);
1517 memset(&stream
, 0, sizeof(stream
));
1518 stream
.next_out
= buffer
;
1519 stream
.avail_out
= size
+ 1;
1521 git_inflate_init(&stream
);
1523 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1524 stream
.next_in
= in
;
1525 st
= git_inflate(&stream
, Z_FINISH
);
1526 if (!stream
.avail_out
)
1527 break; /* the payload is larger than it should be */
1528 curpos
+= stream
.next_in
- in
;
1529 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1530 git_inflate_end(&stream
);
1531 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1536 /* versions of zlib can clobber unconsumed portion of outbuf */
1537 buffer
[size
] = '\0';
1542 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1544 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
1545 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
1546 p
->pack_name
, (uintmax_t)obj_offset
);
1549 int do_check_packed_object_crc
;
1551 #define UNPACK_ENTRY_STACK_PREALLOC 64
1552 struct unpack_entry_stack_ent
{
1558 static void *read_object(struct repository
*r
,
1559 const struct object_id
*oid
,
1560 enum object_type
*type
,
1561 unsigned long *size
)
1563 struct object_info oi
= OBJECT_INFO_INIT
;
1567 oi
.contentp
= &content
;
1569 if (oid_object_info_extended(r
, oid
, &oi
, 0) < 0)
1574 void *unpack_entry(struct repository
*r
, struct packed_git
*p
, off_t obj_offset
,
1575 enum object_type
*final_type
, unsigned long *final_size
)
1577 struct pack_window
*w_curs
= NULL
;
1578 off_t curpos
= obj_offset
;
1581 enum object_type type
;
1582 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
1583 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
1584 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
1585 int base_from_cache
= 0;
1587 write_pack_access_log(p
, obj_offset
);
1589 /* PHASE 1: drill down to the innermost base object */
1593 struct delta_base_cache_entry
*ent
;
1595 ent
= get_delta_base_cache_entry(p
, curpos
);
1600 detach_delta_base_cache_entry(ent
);
1601 base_from_cache
= 1;
1605 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1606 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1607 off_t len
= revidx
[1].offset
- obj_offset
;
1608 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1609 struct object_id oid
;
1610 nth_packed_object_oid(&oid
, p
, revidx
->nr
);
1611 error("bad packed object CRC for %s",
1613 mark_bad_packed_object(p
, oid
.hash
);
1619 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1620 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
1623 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1625 error("failed to validate delta base reference "
1626 "at offset %"PRIuMAX
" from %s",
1627 (uintmax_t)curpos
, p
->pack_name
);
1628 /* bail to phase 2, in hopes of recovery */
1633 /* push object, proceed to base */
1634 if (delta_stack_nr
>= delta_stack_alloc
1635 && delta_stack
== small_delta_stack
) {
1636 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
1637 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
1638 memcpy(delta_stack
, small_delta_stack
,
1639 sizeof(*delta_stack
)*delta_stack_nr
);
1641 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
1643 i
= delta_stack_nr
++;
1644 delta_stack
[i
].obj_offset
= obj_offset
;
1645 delta_stack
[i
].curpos
= curpos
;
1646 delta_stack
[i
].size
= size
;
1648 curpos
= obj_offset
= base_offset
;
1651 /* PHASE 2: handle the base */
1656 BUG("unpack_entry: left loop at a valid delta");
1662 if (!base_from_cache
)
1663 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
1667 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1668 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1671 /* PHASE 3: apply deltas in order */
1674 * 'data' holds the base data, or NULL if there was corruption
1676 while (delta_stack_nr
) {
1679 void *external_base
= NULL
;
1680 unsigned long delta_size
, base_size
= size
;
1686 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
1690 * We're probably in deep shit, but let's try to fetch
1691 * the required base anyway from another pack or loose.
1692 * This is costly but should happen only in the presence
1693 * of a corrupted pack, and is better than failing outright.
1695 struct revindex_entry
*revidx
;
1696 struct object_id base_oid
;
1697 revidx
= find_pack_revindex(p
, obj_offset
);
1699 nth_packed_object_oid(&base_oid
, p
, revidx
->nr
);
1700 error("failed to read delta base object %s"
1701 " at offset %"PRIuMAX
" from %s",
1702 oid_to_hex(&base_oid
), (uintmax_t)obj_offset
,
1704 mark_bad_packed_object(p
, base_oid
.hash
);
1705 base
= read_object(r
, &base_oid
, &type
, &base_size
);
1706 external_base
= base
;
1710 i
= --delta_stack_nr
;
1711 obj_offset
= delta_stack
[i
].obj_offset
;
1712 curpos
= delta_stack
[i
].curpos
;
1713 delta_size
= delta_stack
[i
].size
;
1718 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
1721 error("failed to unpack compressed delta "
1722 "at offset %"PRIuMAX
" from %s",
1723 (uintmax_t)curpos
, p
->pack_name
);
1725 free(external_base
);
1729 data
= patch_delta(base
, base_size
,
1730 delta_data
, delta_size
,
1734 * We could not apply the delta; warn the user, but keep going.
1735 * Our failure will be noticed either in the next iteration of
1736 * the loop, or if this is the final delta, in the caller when
1737 * we return NULL. Those code paths will take care of making
1738 * a more explicit warning and retrying with another copy of
1742 error("failed to apply delta");
1745 free(external_base
);
1754 unuse_pack(&w_curs
);
1756 if (delta_stack
!= small_delta_stack
)
1762 int bsearch_pack(const struct object_id
*oid
, const struct packed_git
*p
, uint32_t *result
)
1764 const unsigned char *index_fanout
= p
->index_data
;
1765 const unsigned char *index_lookup
;
1766 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1767 int index_lookup_width
;
1770 BUG("bsearch_pack called without a valid pack-index");
1772 index_lookup
= index_fanout
+ 4 * 256;
1773 if (p
->index_version
== 1) {
1774 index_lookup_width
= hashsz
+ 4;
1777 index_lookup_width
= hashsz
;
1782 return bsearch_hash(oid
->hash
, (const uint32_t*)index_fanout
,
1783 index_lookup
, index_lookup_width
, result
);
1786 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1789 const unsigned char *index
= p
->index_data
;
1790 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1792 if (open_pack_index(p
))
1794 index
= p
->index_data
;
1796 if (n
>= p
->num_objects
)
1799 if (p
->index_version
== 1) {
1800 return index
+ (hashsz
+ 4) * n
+ 4;
1803 return index
+ hashsz
* n
;
1807 const struct object_id
*nth_packed_object_oid(struct object_id
*oid
,
1808 struct packed_git
*p
,
1811 const unsigned char *hash
= nth_packed_object_sha1(p
, n
);
1814 hashcpy(oid
->hash
, hash
);
1818 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
1820 const unsigned char *ptr
= vptr
;
1821 const unsigned char *start
= p
->index_data
;
1822 const unsigned char *end
= start
+ p
->index_size
;
1824 die(_("offset before start of pack index for %s (corrupt index?)"),
1826 /* No need to check for underflow; .idx files must be at least 8 bytes */
1828 die(_("offset beyond end of pack index for %s (truncated index?)"),
1832 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1834 const unsigned char *index
= p
->index_data
;
1835 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1837 if (p
->index_version
== 1) {
1838 return ntohl(*((uint32_t *)(index
+ (hashsz
+ 4) * n
)));
1841 index
+= 8 + p
->num_objects
* (hashsz
+ 4);
1842 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1843 if (!(off
& 0x80000000))
1845 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1846 check_pack_index_ptr(p
, index
);
1847 return get_be64(index
);
1851 off_t
find_pack_entry_one(const unsigned char *sha1
,
1852 struct packed_git
*p
)
1854 const unsigned char *index
= p
->index_data
;
1855 struct object_id oid
;
1859 if (open_pack_index(p
))
1863 hashcpy(oid
.hash
, sha1
);
1864 if (bsearch_pack(&oid
, p
, &result
))
1865 return nth_packed_object_offset(p
, result
);
1869 int is_pack_valid(struct packed_git
*p
)
1871 /* An already open pack is known to be valid. */
1872 if (p
->pack_fd
!= -1)
1875 /* If the pack has one window completely covering the
1876 * file size, the pack is known to be valid even if
1877 * the descriptor is not currently open.
1880 struct pack_window
*w
= p
->windows
;
1882 if (!w
->offset
&& w
->len
== p
->pack_size
)
1886 /* Force the pack to open to prove its valid. */
1887 return !open_packed_git(p
);
1890 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1891 struct packed_git
*packs
)
1893 struct packed_git
*p
;
1895 for (p
= packs
; p
; p
= p
->next
) {
1896 if (find_pack_entry_one(sha1
, p
))
1903 static int fill_pack_entry(const struct object_id
*oid
,
1904 struct pack_entry
*e
,
1905 struct packed_git
*p
)
1909 if (p
->num_bad_objects
) {
1911 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1912 if (!hashcmp(oid
->hash
,
1913 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
1917 offset
= find_pack_entry_one(oid
->hash
, p
);
1922 * We are about to tell the caller where they can locate the
1923 * requested object. We better make sure the packfile is
1924 * still here and can be accessed before supplying that
1925 * answer, as it may have been deleted since the index was
1928 if (!is_pack_valid(p
))
1935 int find_pack_entry(struct repository
*r
, const struct object_id
*oid
, struct pack_entry
*e
)
1937 struct list_head
*pos
;
1938 struct multi_pack_index
*m
;
1940 prepare_packed_git(r
);
1941 if (!r
->objects
->packed_git
&& !r
->objects
->multi_pack_index
)
1944 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
1945 if (fill_midx_entry(oid
, e
, m
))
1949 list_for_each(pos
, &r
->objects
->packed_git_mru
) {
1950 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1951 if (fill_pack_entry(oid
, e
, p
)) {
1952 list_move(&p
->mru
, &r
->objects
->packed_git_mru
);
1959 int has_object_pack(const struct object_id
*oid
)
1961 struct pack_entry e
;
1962 return find_pack_entry(the_repository
, oid
, &e
);
1965 int has_pack_index(const unsigned char *sha1
)
1968 if (stat(sha1_pack_index_name(sha1
), &st
))
1973 int for_each_object_in_pack(struct packed_git
*p
,
1974 each_packed_object_fn cb
, void *data
,
1975 enum for_each_object_flags flags
)
1980 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
)
1981 load_pack_revindex(p
);
1983 for (i
= 0; i
< p
->num_objects
; i
++) {
1985 struct object_id oid
;
1987 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
)
1988 pos
= p
->revindex
[i
].nr
;
1992 if (!nth_packed_object_oid(&oid
, p
, pos
))
1993 return error("unable to get sha1 of object %u in %s",
1996 r
= cb(&oid
, p
, pos
, data
);
2003 int for_each_packed_object(each_packed_object_fn cb
, void *data
,
2004 enum for_each_object_flags flags
)
2006 struct packed_git
*p
;
2008 int pack_errors
= 0;
2010 prepare_packed_git(the_repository
);
2011 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
2012 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
2014 if ((flags
& FOR_EACH_OBJECT_PROMISOR_ONLY
) &&
2017 if (open_pack_index(p
)) {
2021 r
= for_each_object_in_pack(p
, cb
, data
, flags
);
2025 return r
? r
: pack_errors
;
2028 static int add_promisor_object(const struct object_id
*oid
,
2029 struct packed_git
*pack
,
2033 struct oidset
*set
= set_
;
2034 struct object
*obj
= parse_object(the_repository
, oid
);
2038 oidset_insert(set
, oid
);
2041 * If this is a tree, commit, or tag, the objects it refers
2042 * to are also promisor objects. (Blobs refer to no objects->)
2044 if (obj
->type
== OBJ_TREE
) {
2045 struct tree
*tree
= (struct tree
*)obj
;
2046 struct tree_desc desc
;
2047 struct name_entry entry
;
2048 if (init_tree_desc_gently(&desc
, tree
->buffer
, tree
->size
))
2050 * Error messages are given when packs are
2051 * verified, so do not print any here.
2054 while (tree_entry_gently(&desc
, &entry
))
2055 oidset_insert(set
, entry
.oid
);
2056 } else if (obj
->type
== OBJ_COMMIT
) {
2057 struct commit
*commit
= (struct commit
*) obj
;
2058 struct commit_list
*parents
= commit
->parents
;
2060 oidset_insert(set
, get_commit_tree_oid(commit
));
2061 for (; parents
; parents
= parents
->next
)
2062 oidset_insert(set
, &parents
->item
->object
.oid
);
2063 } else if (obj
->type
== OBJ_TAG
) {
2064 struct tag
*tag
= (struct tag
*) obj
;
2065 oidset_insert(set
, &tag
->tagged
->oid
);
2070 int is_promisor_object(const struct object_id
*oid
)
2072 static struct oidset promisor_objects
;
2073 static int promisor_objects_prepared
;
2075 if (!promisor_objects_prepared
) {
2076 if (repository_format_partial_clone
) {
2077 for_each_packed_object(add_promisor_object
,
2079 FOR_EACH_OBJECT_PROMISOR_ONLY
);
2081 promisor_objects_prepared
= 1;
2083 return oidset_contains(&promisor_objects
, oid
);