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
->hash
, 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 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 ;-) */
472 const char *pack_basename(struct packed_git
*p
)
474 const char *ret
= strrchr(p
->pack_name
, '/');
476 ret
= ret
+ 1; /* skip past slash */
478 ret
= p
->pack_name
; /* we only have a base */
483 * Do not call this directly as this leaks p->pack_fd on error return;
484 * call open_packed_git() instead.
486 static int open_packed_git_1(struct packed_git
*p
)
489 struct pack_header hdr
;
490 unsigned char hash
[GIT_MAX_RAWSZ
];
491 unsigned char *idx_hash
;
494 const unsigned hashsz
= the_hash_algo
->rawsz
;
496 if (!p
->index_data
) {
497 struct multi_pack_index
*m
;
498 const char *pack_name
= pack_basename(p
);
500 for (m
= the_repository
->objects
->multi_pack_index
;
502 if (midx_contains_pack(m
, pack_name
))
506 if (!m
&& open_pack_index(p
))
507 return error("packfile %s index unavailable", p
->pack_name
);
511 unsigned int max_fds
= get_max_fd_limit();
513 /* Save 3 for stdin/stdout/stderr, 22 for work */
515 pack_max_fds
= max_fds
- 25;
520 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
523 p
->pack_fd
= git_open(p
->pack_name
);
524 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
528 /* If we created the struct before we had the pack we lack size. */
530 if (!S_ISREG(st
.st_mode
))
531 return error("packfile %s not a regular file", p
->pack_name
);
532 p
->pack_size
= st
.st_size
;
533 } else if (p
->pack_size
!= st
.st_size
)
534 return error("packfile %s size changed", p
->pack_name
);
536 /* We leave these file descriptors open with sliding mmap;
537 * there is no point keeping them open across exec(), though.
539 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
541 return error("cannot determine file descriptor flags");
542 fd_flag
|= FD_CLOEXEC
;
543 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
544 return error("cannot set FD_CLOEXEC");
546 /* Verify we recognize this pack file format. */
547 read_result
= read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
));
549 return error_errno("error reading from %s", p
->pack_name
);
550 if (read_result
!= sizeof(hdr
))
551 return error("file %s is far too short to be a packfile", p
->pack_name
);
552 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
553 return error("file %s is not a GIT packfile", p
->pack_name
);
554 if (!pack_version_ok(hdr
.hdr_version
))
555 return error("packfile %s is version %"PRIu32
" and not"
556 " supported (try upgrading GIT to a newer version)",
557 p
->pack_name
, ntohl(hdr
.hdr_version
));
559 /* Skip index checking if in multi-pack-index */
563 /* Verify the pack matches its index. */
564 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
565 return error("packfile %s claims to have %"PRIu32
" objects"
566 " while index indicates %"PRIu32
" objects",
567 p
->pack_name
, ntohl(hdr
.hdr_entries
),
569 if (lseek(p
->pack_fd
, p
->pack_size
- hashsz
, SEEK_SET
) == -1)
570 return error("end of packfile %s is unavailable", p
->pack_name
);
571 read_result
= read_in_full(p
->pack_fd
, hash
, hashsz
);
573 return error_errno("error reading from %s", p
->pack_name
);
574 if (read_result
!= hashsz
)
575 return error("packfile %s signature is unavailable", p
->pack_name
);
576 idx_hash
= ((unsigned char *)p
->index_data
) + p
->index_size
- hashsz
* 2;
577 if (!hasheq(hash
, idx_hash
))
578 return error("packfile %s does not match index", p
->pack_name
);
582 static int open_packed_git(struct packed_git
*p
)
584 if (!open_packed_git_1(p
))
590 static int in_window(struct pack_window
*win
, off_t offset
)
592 /* We must promise at least one full hash after the
593 * offset is available from this window, otherwise the offset
594 * is not actually in this window and a different window (which
595 * has that one hash excess) must be used. This is to support
596 * the object header and delta base parsing routines below.
598 off_t win_off
= win
->offset
;
599 return win_off
<= offset
600 && (offset
+ the_hash_algo
->rawsz
) <= (win_off
+ win
->len
);
603 unsigned char *use_pack(struct packed_git
*p
,
604 struct pack_window
**w_cursor
,
608 struct pack_window
*win
= *w_cursor
;
610 /* Since packfiles end in a hash of their content and it's
611 * pointless to ask for an offset into the middle of that
612 * hash, and the in_window function above wouldn't match
613 * don't allow an offset too close to the end of the file.
615 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
616 die("packfile %s cannot be accessed", p
->pack_name
);
617 if (offset
> (p
->pack_size
- the_hash_algo
->rawsz
))
618 die("offset beyond end of packfile (truncated pack?)");
620 die(_("offset before end of packfile (broken .idx?)"));
622 if (!win
|| !in_window(win
, offset
)) {
625 for (win
= p
->windows
; win
; win
= win
->next
) {
626 if (in_window(win
, offset
))
630 size_t window_align
= packed_git_window_size
/ 2;
633 if (p
->pack_fd
== -1 && open_packed_git(p
))
634 die("packfile %s cannot be accessed", p
->pack_name
);
636 win
= xcalloc(1, sizeof(*win
));
637 win
->offset
= (offset
/ window_align
) * window_align
;
638 len
= p
->pack_size
- win
->offset
;
639 if (len
> packed_git_window_size
)
640 len
= packed_git_window_size
;
641 win
->len
= (size_t)len
;
642 pack_mapped
+= win
->len
;
643 while (packed_git_limit
< pack_mapped
644 && unuse_one_window(p
))
646 win
->base
= xmmap_gently(NULL
, win
->len
,
647 PROT_READ
, MAP_PRIVATE
,
648 p
->pack_fd
, win
->offset
);
649 if (win
->base
== MAP_FAILED
)
650 die_errno("packfile %s cannot be mapped",
652 if (!win
->offset
&& win
->len
== p
->pack_size
657 if (pack_mapped
> peak_pack_mapped
)
658 peak_pack_mapped
= pack_mapped
;
659 if (pack_open_windows
> peak_pack_open_windows
)
660 peak_pack_open_windows
= pack_open_windows
;
661 win
->next
= p
->windows
;
665 if (win
!= *w_cursor
) {
666 win
->last_used
= pack_used_ctr
++;
670 offset
-= win
->offset
;
672 *left
= win
->len
- xsize_t(offset
);
673 return win
->base
+ offset
;
676 void unuse_pack(struct pack_window
**w_cursor
)
678 struct pack_window
*w
= *w_cursor
;
685 static void try_to_free_pack_memory(size_t size
)
687 release_pack_memory(size
);
690 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
692 static int have_set_try_to_free_routine
;
695 struct packed_git
*p
;
697 if (!have_set_try_to_free_routine
) {
698 have_set_try_to_free_routine
= 1;
699 set_try_to_free_routine(try_to_free_pack_memory
);
703 * Make sure a corresponding .pack file exists and that
704 * the index looks sane.
706 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
710 * ".promisor" is long enough to hold any suffix we're adding (and
711 * the use xsnprintf double-checks that)
713 alloc
= st_add3(path_len
, strlen(".promisor"), 1);
714 p
= alloc_packed_git(alloc
);
715 memcpy(p
->pack_name
, path
, path_len
);
717 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
718 if (!access(p
->pack_name
, F_OK
))
721 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".promisor");
722 if (!access(p
->pack_name
, F_OK
))
723 p
->pack_promisor
= 1;
725 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
726 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
731 /* ok, it looks sane as far as we can check without
732 * actually mapping the pack file.
734 p
->pack_size
= st
.st_size
;
735 p
->pack_local
= local
;
736 p
->mtime
= st
.st_mtime
;
737 if (path_len
< the_hash_algo
->hexsz
||
738 get_sha1_hex(path
+ path_len
- the_hash_algo
->hexsz
, p
->hash
))
743 void install_packed_git(struct repository
*r
, struct packed_git
*pack
)
745 if (pack
->pack_fd
!= -1)
748 pack
->next
= r
->objects
->packed_git
;
749 r
->objects
->packed_git
= pack
;
752 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
754 static void report_helper(const struct string_list
*list
,
755 int seen_bits
, int first
, int last
)
757 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
760 for (; first
< last
; first
++)
761 report_garbage(seen_bits
, list
->items
[first
].string
);
764 static void report_pack_garbage(struct string_list
*list
)
766 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
771 string_list_sort(list
);
773 for (i
= 0; i
< list
->nr
; i
++) {
774 const char *path
= list
->items
[i
].string
;
776 strncmp(path
, list
->items
[first
].string
, baselen
)) {
777 report_helper(list
, seen_bits
, first
, i
);
782 const char *dot
= strrchr(path
, '.');
784 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
787 baselen
= dot
- path
+ 1;
790 if (!strcmp(path
+ baselen
, "pack"))
792 else if (!strcmp(path
+ baselen
, "idx"))
795 report_helper(list
, seen_bits
, first
, list
->nr
);
798 void for_each_file_in_pack_dir(const char *objdir
,
799 each_file_in_pack_dir_fn fn
,
802 struct strbuf path
= STRBUF_INIT
;
807 strbuf_addstr(&path
, objdir
);
808 strbuf_addstr(&path
, "/pack");
809 dir
= opendir(path
.buf
);
812 error_errno("unable to open object pack directory: %s",
814 strbuf_release(&path
);
817 strbuf_addch(&path
, '/');
818 dirnamelen
= path
.len
;
819 while ((de
= readdir(dir
)) != NULL
) {
820 if (is_dot_or_dotdot(de
->d_name
))
823 strbuf_setlen(&path
, dirnamelen
);
824 strbuf_addstr(&path
, de
->d_name
);
826 fn(path
.buf
, path
.len
, de
->d_name
, data
);
830 strbuf_release(&path
);
833 struct prepare_pack_data
{
834 struct repository
*r
;
835 struct string_list
*garbage
;
837 struct multi_pack_index
*m
;
840 static void prepare_pack(const char *full_name
, size_t full_name_len
,
841 const char *file_name
, void *_data
)
843 struct prepare_pack_data
*data
= (struct prepare_pack_data
*)_data
;
844 struct packed_git
*p
;
845 size_t base_len
= full_name_len
;
847 if (strip_suffix_mem(full_name
, &base_len
, ".idx") &&
848 !(data
->m
&& midx_contains_pack(data
->m
, file_name
))) {
849 /* Don't reopen a pack we already have. */
850 for (p
= data
->r
->objects
->packed_git
; p
; p
= p
->next
) {
852 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
854 !memcmp(p
->pack_name
, full_name
, len
))
859 p
= add_packed_git(full_name
, full_name_len
, data
->local
);
861 install_packed_git(data
->r
, p
);
868 if (!strcmp(file_name
, "multi-pack-index"))
870 if (ends_with(file_name
, ".idx") ||
871 ends_with(file_name
, ".pack") ||
872 ends_with(file_name
, ".bitmap") ||
873 ends_with(file_name
, ".keep") ||
874 ends_with(file_name
, ".promisor"))
875 string_list_append(data
->garbage
, full_name
);
877 report_garbage(PACKDIR_FILE_GARBAGE
, full_name
);
880 static void prepare_packed_git_one(struct repository
*r
, char *objdir
, int local
)
882 struct prepare_pack_data data
;
883 struct string_list garbage
= STRING_LIST_INIT_DUP
;
885 data
.m
= r
->objects
->multi_pack_index
;
887 /* look for the multi-pack-index for this object directory */
888 while (data
.m
&& strcmp(data
.m
->object_dir
, objdir
))
889 data
.m
= data
.m
->next
;
892 data
.garbage
= &garbage
;
895 for_each_file_in_pack_dir(objdir
, prepare_pack
, &data
);
897 report_pack_garbage(data
.garbage
);
898 string_list_clear(data
.garbage
, 0);
901 static void prepare_packed_git(struct repository
*r
);
903 * Give a fast, rough count of the number of objects in the repository. This
904 * ignores loose objects completely. If you have a lot of them, then either
905 * you should repack because your performance will be awful, or they are
906 * all unreachable objects about to be pruned, in which case they're not really
907 * interesting as a measure of repo size in the first place.
909 unsigned long repo_approximate_object_count(struct repository
*r
)
911 if (!r
->objects
->approximate_object_count_valid
) {
913 struct multi_pack_index
*m
;
914 struct packed_git
*p
;
916 prepare_packed_git(r
);
918 for (m
= get_multi_pack_index(r
); m
; m
= m
->next
)
919 count
+= m
->num_objects
;
920 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
) {
921 if (open_pack_index(p
))
923 count
+= p
->num_objects
;
925 r
->objects
->approximate_object_count
= count
;
927 return r
->objects
->approximate_object_count
;
930 static void *get_next_packed_git(const void *p
)
932 return ((const struct packed_git
*)p
)->next
;
935 static void set_next_packed_git(void *p
, void *next
)
937 ((struct packed_git
*)p
)->next
= next
;
940 static int sort_pack(const void *a_
, const void *b_
)
942 const struct packed_git
*a
= a_
;
943 const struct packed_git
*b
= b_
;
947 * Local packs tend to contain objects specific to our
948 * variant of the project than remote ones. In addition,
949 * remote ones could be on a network mounted filesystem.
950 * Favor local ones for these reasons.
952 st
= a
->pack_local
- b
->pack_local
;
957 * Younger packs tend to contain more recent objects,
958 * and more recent objects tend to get accessed more
961 if (a
->mtime
< b
->mtime
)
963 else if (a
->mtime
== b
->mtime
)
968 static void rearrange_packed_git(struct repository
*r
)
970 r
->objects
->packed_git
= llist_mergesort(
971 r
->objects
->packed_git
, get_next_packed_git
,
972 set_next_packed_git
, sort_pack
);
975 static void prepare_packed_git_mru(struct repository
*r
)
977 struct packed_git
*p
;
979 INIT_LIST_HEAD(&r
->objects
->packed_git_mru
);
981 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
982 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
985 static void prepare_packed_git(struct repository
*r
)
987 struct object_directory
*odb
;
989 if (r
->objects
->packed_git_initialized
)
993 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
994 int local
= (odb
== r
->objects
->odb
);
995 prepare_multi_pack_index_one(r
, odb
->path
, local
);
996 prepare_packed_git_one(r
, odb
->path
, local
);
998 rearrange_packed_git(r
);
1000 prepare_packed_git_mru(r
);
1001 r
->objects
->packed_git_initialized
= 1;
1004 void reprepare_packed_git(struct repository
*r
)
1006 struct object_directory
*odb
;
1008 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
)
1009 odb_clear_loose_cache(odb
);
1011 r
->objects
->approximate_object_count_valid
= 0;
1012 r
->objects
->packed_git_initialized
= 0;
1013 prepare_packed_git(r
);
1016 struct packed_git
*get_packed_git(struct repository
*r
)
1018 prepare_packed_git(r
);
1019 return r
->objects
->packed_git
;
1022 struct multi_pack_index
*get_multi_pack_index(struct repository
*r
)
1024 prepare_packed_git(r
);
1025 return r
->objects
->multi_pack_index
;
1028 struct packed_git
*get_all_packs(struct repository
*r
)
1030 struct multi_pack_index
*m
;
1032 prepare_packed_git(r
);
1033 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
1035 for (i
= 0; i
< m
->num_packs
; i
++)
1036 prepare_midx_pack(r
, m
, i
);
1039 return r
->objects
->packed_git
;
1042 struct list_head
*get_packed_git_mru(struct repository
*r
)
1044 prepare_packed_git(r
);
1045 return &r
->objects
->packed_git_mru
;
1048 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1049 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1052 unsigned long size
, c
;
1053 unsigned long used
= 0;
1056 *type
= (c
>> 4) & 7;
1060 if (len
<= used
|| bitsizeof(long) <= shift
) {
1061 error("bad object header");
1066 size
+= (c
& 0x7f) << shift
;
1073 unsigned long get_size_from_delta(struct packed_git
*p
,
1074 struct pack_window
**w_curs
,
1077 const unsigned char *data
;
1078 unsigned char delta_head
[20], *in
;
1082 memset(&stream
, 0, sizeof(stream
));
1083 stream
.next_out
= delta_head
;
1084 stream
.avail_out
= sizeof(delta_head
);
1086 git_inflate_init(&stream
);
1088 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1089 stream
.next_in
= in
;
1090 st
= git_inflate(&stream
, Z_FINISH
);
1091 curpos
+= stream
.next_in
- in
;
1092 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1093 stream
.total_out
< sizeof(delta_head
));
1094 git_inflate_end(&stream
);
1095 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1096 error("delta data unpack-initial failed");
1100 /* Examine the initial part of the delta to figure out
1105 /* ignore base size */
1106 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1108 /* Read the result size */
1109 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1112 int unpack_object_header(struct packed_git
*p
,
1113 struct pack_window
**w_curs
,
1115 unsigned long *sizep
)
1117 unsigned char *base
;
1120 enum object_type type
;
1122 /* use_pack() assures us we have [base, base + 20) available
1123 * as a range that we can look at. (Its actually the hash
1124 * size that is assured.) With our object header encoding
1125 * the maximum deflated object size is 2^137, which is just
1126 * insane, so we know won't exceed what we have been given.
1128 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1129 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1138 void mark_bad_packed_object(struct packed_git
*p
, const unsigned char *sha1
)
1141 const unsigned hashsz
= the_hash_algo
->rawsz
;
1142 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1143 if (hasheq(sha1
, p
->bad_object_sha1
+ hashsz
* i
))
1145 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1146 st_mult(GIT_MAX_RAWSZ
,
1147 st_add(p
->num_bad_objects
, 1)));
1148 hashcpy(p
->bad_object_sha1
+ hashsz
* p
->num_bad_objects
, sha1
);
1149 p
->num_bad_objects
++;
1152 const struct packed_git
*has_packed_and_bad(struct repository
*r
,
1153 const unsigned char *sha1
)
1155 struct packed_git
*p
;
1158 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
1159 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1161 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
1166 static off_t
get_delta_base(struct packed_git
*p
,
1167 struct pack_window
**w_curs
,
1169 enum object_type type
,
1170 off_t delta_obj_offset
)
1172 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1175 /* use_pack() assured us we have [base_info, base_info + 20)
1176 * as a range that we can look at without walking off the
1177 * end of the mapped window. Its actually the hash size
1178 * that is assured. An OFS_DELTA longer than the hash size
1179 * is stupid, as then a REF_DELTA would be smaller to store.
1181 if (type
== OBJ_OFS_DELTA
) {
1183 unsigned char c
= base_info
[used
++];
1184 base_offset
= c
& 127;
1187 if (!base_offset
|| MSB(base_offset
, 7))
1188 return 0; /* overflow */
1189 c
= base_info
[used
++];
1190 base_offset
= (base_offset
<< 7) + (c
& 127);
1192 base_offset
= delta_obj_offset
- base_offset
;
1193 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1194 return 0; /* out of bound */
1196 } else if (type
== OBJ_REF_DELTA
) {
1197 /* The base entry _must_ be in the same pack */
1198 base_offset
= find_pack_entry_one(base_info
, p
);
1199 *curpos
+= the_hash_algo
->rawsz
;
1201 die("I am totally screwed");
1206 * Like get_delta_base above, but we return the sha1 instead of the pack
1207 * offset. This means it is cheaper for REF deltas (we do not have to do
1208 * the final object lookup), but more expensive for OFS deltas (we
1209 * have to load the revidx to convert the offset back into a sha1).
1211 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1212 struct pack_window
**w_curs
,
1214 enum object_type type
,
1215 off_t delta_obj_offset
)
1217 if (type
== OBJ_REF_DELTA
) {
1218 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1220 } else if (type
== OBJ_OFS_DELTA
) {
1221 struct revindex_entry
*revidx
;
1222 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1223 type
, delta_obj_offset
);
1228 revidx
= find_pack_revindex(p
, base_offset
);
1232 return nth_packed_object_sha1(p
, revidx
->nr
);
1237 static int retry_bad_packed_offset(struct repository
*r
,
1238 struct packed_git
*p
,
1242 struct revindex_entry
*revidx
;
1243 struct object_id oid
;
1244 revidx
= find_pack_revindex(p
, obj_offset
);
1247 nth_packed_object_oid(&oid
, p
, revidx
->nr
);
1248 mark_bad_packed_object(p
, oid
.hash
);
1249 type
= oid_object_info(r
, &oid
, NULL
);
1250 if (type
<= OBJ_NONE
)
1255 #define POI_STACK_PREALLOC 64
1257 static enum object_type
packed_to_object_type(struct repository
*r
,
1258 struct packed_git
*p
,
1260 enum object_type type
,
1261 struct pack_window
**w_curs
,
1264 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1265 off_t
*poi_stack
= small_poi_stack
;
1266 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1268 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1271 /* Push the object we're going to leave behind */
1272 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1273 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1274 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
1275 COPY_ARRAY(poi_stack
, small_poi_stack
, poi_stack_nr
);
1277 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1279 poi_stack
[poi_stack_nr
++] = obj_offset
;
1280 /* If parsing the base offset fails, just unwind */
1281 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1284 curpos
= obj_offset
= base_offset
;
1285 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1286 if (type
<= OBJ_NONE
) {
1287 /* If getting the base itself fails, we first
1288 * retry the base, otherwise unwind */
1289 type
= retry_bad_packed_offset(r
, p
, base_offset
);
1290 if (type
> OBJ_NONE
)
1304 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1305 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1310 if (poi_stack
!= small_poi_stack
)
1315 while (poi_stack_nr
) {
1316 obj_offset
= poi_stack
[--poi_stack_nr
];
1317 type
= retry_bad_packed_offset(r
, p
, obj_offset
);
1318 if (type
> OBJ_NONE
)
1325 static struct hashmap delta_base_cache
;
1326 static size_t delta_base_cached
;
1328 static LIST_HEAD(delta_base_cache_lru
);
1330 struct delta_base_cache_key
{
1331 struct packed_git
*p
;
1335 struct delta_base_cache_entry
{
1336 struct hashmap hash
;
1337 struct delta_base_cache_key key
;
1338 struct list_head lru
;
1341 enum object_type type
;
1344 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1348 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
1349 hash
+= (hash
>> 8) + (hash
>> 16);
1353 static struct delta_base_cache_entry
*
1354 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1356 struct hashmap_entry entry
;
1357 struct delta_base_cache_key key
;
1359 if (!delta_base_cache
.cmpfn
)
1362 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
1364 key
.base_offset
= base_offset
;
1365 return hashmap_get(&delta_base_cache
, &entry
, &key
);
1368 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
1369 const struct delta_base_cache_key
*b
)
1371 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
1374 static int delta_base_cache_hash_cmp(const void *unused_cmp_data
,
1375 const void *va
, const void *vb
,
1378 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
1379 const struct delta_base_cache_key
*key
= vkey
;
1381 return !delta_base_cache_key_eq(&a
->key
, key
);
1383 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
1386 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1388 return !!get_delta_base_cache_entry(p
, base_offset
);
1392 * Remove the entry from the cache, but do _not_ free the associated
1393 * entry data. The caller takes ownership of the "data" buffer, and
1394 * should copy out any fields it wants before detaching.
1396 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1398 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
1399 list_del(&ent
->lru
);
1400 delta_base_cached
-= ent
->size
;
1404 static void *cache_or_unpack_entry(struct repository
*r
, struct packed_git
*p
,
1405 off_t base_offset
, unsigned long *base_size
,
1406 enum object_type
*type
)
1408 struct delta_base_cache_entry
*ent
;
1410 ent
= get_delta_base_cache_entry(p
, base_offset
);
1412 return unpack_entry(r
, p
, base_offset
, type
, base_size
);
1417 *base_size
= ent
->size
;
1418 return xmemdupz(ent
->data
, ent
->size
);
1421 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1424 detach_delta_base_cache_entry(ent
);
1427 void clear_delta_base_cache(void)
1429 struct list_head
*lru
, *tmp
;
1430 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1431 struct delta_base_cache_entry
*entry
=
1432 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1433 release_delta_base_cache(entry
);
1437 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1438 void *base
, unsigned long base_size
, enum object_type type
)
1440 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
1441 struct list_head
*lru
, *tmp
;
1443 delta_base_cached
+= base_size
;
1445 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1446 struct delta_base_cache_entry
*f
=
1447 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1448 if (delta_base_cached
<= delta_base_cache_limit
)
1450 release_delta_base_cache(f
);
1454 ent
->key
.base_offset
= base_offset
;
1457 ent
->size
= base_size
;
1458 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
1460 if (!delta_base_cache
.cmpfn
)
1461 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, NULL
, 0);
1462 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
1463 hashmap_add(&delta_base_cache
, ent
);
1466 int packed_object_info(struct repository
*r
, struct packed_git
*p
,
1467 off_t obj_offset
, struct object_info
*oi
)
1469 struct pack_window
*w_curs
= NULL
;
1471 off_t curpos
= obj_offset
;
1472 enum object_type type
;
1475 * We always get the representation type, but only convert it to
1476 * a "real" type later if the caller is interested.
1479 *oi
->contentp
= cache_or_unpack_entry(r
, p
, obj_offset
, oi
->sizep
,
1484 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1487 if (!oi
->contentp
&& oi
->sizep
) {
1488 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1489 off_t tmp_pos
= curpos
;
1490 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1496 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1497 if (*oi
->sizep
== 0) {
1506 if (oi
->disk_sizep
) {
1507 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1508 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1511 if (oi
->typep
|| oi
->type_name
) {
1512 enum object_type ptot
;
1513 ptot
= packed_to_object_type(r
, p
, obj_offset
,
1514 type
, &w_curs
, curpos
);
1517 if (oi
->type_name
) {
1518 const char *tn
= type_name(ptot
);
1520 strbuf_addstr(oi
->type_name
, tn
);
1528 if (oi
->delta_base_sha1
) {
1529 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1530 const unsigned char *base
;
1532 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1539 hashcpy(oi
->delta_base_sha1
, base
);
1541 hashclr(oi
->delta_base_sha1
);
1544 oi
->whence
= in_delta_base_cache(p
, obj_offset
) ? OI_DBCACHED
:
1548 unuse_pack(&w_curs
);
1552 static void *unpack_compressed_entry(struct packed_git
*p
,
1553 struct pack_window
**w_curs
,
1559 unsigned char *buffer
, *in
;
1561 buffer
= xmallocz_gently(size
);
1564 memset(&stream
, 0, sizeof(stream
));
1565 stream
.next_out
= buffer
;
1566 stream
.avail_out
= size
+ 1;
1568 git_inflate_init(&stream
);
1570 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1571 stream
.next_in
= in
;
1572 st
= git_inflate(&stream
, Z_FINISH
);
1573 if (!stream
.avail_out
)
1574 break; /* the payload is larger than it should be */
1575 curpos
+= stream
.next_in
- in
;
1576 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1577 git_inflate_end(&stream
);
1578 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1583 /* versions of zlib can clobber unconsumed portion of outbuf */
1584 buffer
[size
] = '\0';
1589 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1591 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
1592 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
1593 p
->pack_name
, (uintmax_t)obj_offset
);
1596 int do_check_packed_object_crc
;
1598 #define UNPACK_ENTRY_STACK_PREALLOC 64
1599 struct unpack_entry_stack_ent
{
1605 static void *read_object(struct repository
*r
,
1606 const struct object_id
*oid
,
1607 enum object_type
*type
,
1608 unsigned long *size
)
1610 struct object_info oi
= OBJECT_INFO_INIT
;
1614 oi
.contentp
= &content
;
1616 if (oid_object_info_extended(r
, oid
, &oi
, 0) < 0)
1621 void *unpack_entry(struct repository
*r
, struct packed_git
*p
, off_t obj_offset
,
1622 enum object_type
*final_type
, unsigned long *final_size
)
1624 struct pack_window
*w_curs
= NULL
;
1625 off_t curpos
= obj_offset
;
1628 enum object_type type
;
1629 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
1630 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
1631 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
1632 int base_from_cache
= 0;
1634 write_pack_access_log(p
, obj_offset
);
1636 /* PHASE 1: drill down to the innermost base object */
1640 struct delta_base_cache_entry
*ent
;
1642 ent
= get_delta_base_cache_entry(p
, curpos
);
1647 detach_delta_base_cache_entry(ent
);
1648 base_from_cache
= 1;
1652 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1653 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1654 off_t len
= revidx
[1].offset
- obj_offset
;
1655 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1656 struct object_id oid
;
1657 nth_packed_object_oid(&oid
, p
, revidx
->nr
);
1658 error("bad packed object CRC for %s",
1660 mark_bad_packed_object(p
, oid
.hash
);
1666 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1667 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
1670 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1672 error("failed to validate delta base reference "
1673 "at offset %"PRIuMAX
" from %s",
1674 (uintmax_t)curpos
, p
->pack_name
);
1675 /* bail to phase 2, in hopes of recovery */
1680 /* push object, proceed to base */
1681 if (delta_stack_nr
>= delta_stack_alloc
1682 && delta_stack
== small_delta_stack
) {
1683 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
1684 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
1685 COPY_ARRAY(delta_stack
, small_delta_stack
,
1688 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
1690 i
= delta_stack_nr
++;
1691 delta_stack
[i
].obj_offset
= obj_offset
;
1692 delta_stack
[i
].curpos
= curpos
;
1693 delta_stack
[i
].size
= size
;
1695 curpos
= obj_offset
= base_offset
;
1698 /* PHASE 2: handle the base */
1703 BUG("unpack_entry: left loop at a valid delta");
1709 if (!base_from_cache
)
1710 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
1714 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1715 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1718 /* PHASE 3: apply deltas in order */
1721 * 'data' holds the base data, or NULL if there was corruption
1723 while (delta_stack_nr
) {
1726 void *external_base
= NULL
;
1727 unsigned long delta_size
, base_size
= size
;
1733 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
1737 * We're probably in deep shit, but let's try to fetch
1738 * the required base anyway from another pack or loose.
1739 * This is costly but should happen only in the presence
1740 * of a corrupted pack, and is better than failing outright.
1742 struct revindex_entry
*revidx
;
1743 struct object_id base_oid
;
1744 revidx
= find_pack_revindex(p
, obj_offset
);
1746 nth_packed_object_oid(&base_oid
, p
, revidx
->nr
);
1747 error("failed to read delta base object %s"
1748 " at offset %"PRIuMAX
" from %s",
1749 oid_to_hex(&base_oid
), (uintmax_t)obj_offset
,
1751 mark_bad_packed_object(p
, base_oid
.hash
);
1752 base
= read_object(r
, &base_oid
, &type
, &base_size
);
1753 external_base
= base
;
1757 i
= --delta_stack_nr
;
1758 obj_offset
= delta_stack
[i
].obj_offset
;
1759 curpos
= delta_stack
[i
].curpos
;
1760 delta_size
= delta_stack
[i
].size
;
1765 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
1768 error("failed to unpack compressed delta "
1769 "at offset %"PRIuMAX
" from %s",
1770 (uintmax_t)curpos
, p
->pack_name
);
1772 free(external_base
);
1776 data
= patch_delta(base
, base_size
,
1777 delta_data
, delta_size
,
1781 * We could not apply the delta; warn the user, but keep going.
1782 * Our failure will be noticed either in the next iteration of
1783 * the loop, or if this is the final delta, in the caller when
1784 * we return NULL. Those code paths will take care of making
1785 * a more explicit warning and retrying with another copy of
1789 error("failed to apply delta");
1792 free(external_base
);
1801 unuse_pack(&w_curs
);
1803 if (delta_stack
!= small_delta_stack
)
1809 int bsearch_pack(const struct object_id
*oid
, const struct packed_git
*p
, uint32_t *result
)
1811 const unsigned char *index_fanout
= p
->index_data
;
1812 const unsigned char *index_lookup
;
1813 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1814 int index_lookup_width
;
1817 BUG("bsearch_pack called without a valid pack-index");
1819 index_lookup
= index_fanout
+ 4 * 256;
1820 if (p
->index_version
== 1) {
1821 index_lookup_width
= hashsz
+ 4;
1824 index_lookup_width
= hashsz
;
1829 return bsearch_hash(oid
->hash
, (const uint32_t*)index_fanout
,
1830 index_lookup
, index_lookup_width
, result
);
1833 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1836 const unsigned char *index
= p
->index_data
;
1837 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1839 if (open_pack_index(p
))
1841 index
= p
->index_data
;
1843 if (n
>= p
->num_objects
)
1846 if (p
->index_version
== 1) {
1847 return index
+ (hashsz
+ 4) * n
+ 4;
1850 return index
+ hashsz
* n
;
1854 const struct object_id
*nth_packed_object_oid(struct object_id
*oid
,
1855 struct packed_git
*p
,
1858 const unsigned char *hash
= nth_packed_object_sha1(p
, n
);
1861 hashcpy(oid
->hash
, hash
);
1865 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
1867 const unsigned char *ptr
= vptr
;
1868 const unsigned char *start
= p
->index_data
;
1869 const unsigned char *end
= start
+ p
->index_size
;
1871 die(_("offset before start of pack index for %s (corrupt index?)"),
1873 /* No need to check for underflow; .idx files must be at least 8 bytes */
1875 die(_("offset beyond end of pack index for %s (truncated index?)"),
1879 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1881 const unsigned char *index
= p
->index_data
;
1882 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1884 if (p
->index_version
== 1) {
1885 return ntohl(*((uint32_t *)(index
+ (hashsz
+ 4) * n
)));
1888 index
+= 8 + p
->num_objects
* (hashsz
+ 4);
1889 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1890 if (!(off
& 0x80000000))
1892 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1893 check_pack_index_ptr(p
, index
);
1894 return get_be64(index
);
1898 off_t
find_pack_entry_one(const unsigned char *sha1
,
1899 struct packed_git
*p
)
1901 const unsigned char *index
= p
->index_data
;
1902 struct object_id oid
;
1906 if (open_pack_index(p
))
1910 hashcpy(oid
.hash
, sha1
);
1911 if (bsearch_pack(&oid
, p
, &result
))
1912 return nth_packed_object_offset(p
, result
);
1916 int is_pack_valid(struct packed_git
*p
)
1918 /* An already open pack is known to be valid. */
1919 if (p
->pack_fd
!= -1)
1922 /* If the pack has one window completely covering the
1923 * file size, the pack is known to be valid even if
1924 * the descriptor is not currently open.
1927 struct pack_window
*w
= p
->windows
;
1929 if (!w
->offset
&& w
->len
== p
->pack_size
)
1933 /* Force the pack to open to prove its valid. */
1934 return !open_packed_git(p
);
1937 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1938 struct packed_git
*packs
)
1940 struct packed_git
*p
;
1942 for (p
= packs
; p
; p
= p
->next
) {
1943 if (find_pack_entry_one(sha1
, p
))
1950 static int fill_pack_entry(const struct object_id
*oid
,
1951 struct pack_entry
*e
,
1952 struct packed_git
*p
)
1956 if (p
->num_bad_objects
) {
1958 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1959 if (hasheq(oid
->hash
,
1960 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
1964 offset
= find_pack_entry_one(oid
->hash
, p
);
1969 * We are about to tell the caller where they can locate the
1970 * requested object. We better make sure the packfile is
1971 * still here and can be accessed before supplying that
1972 * answer, as it may have been deleted since the index was
1975 if (!is_pack_valid(p
))
1982 int find_pack_entry(struct repository
*r
, const struct object_id
*oid
, struct pack_entry
*e
)
1984 struct list_head
*pos
;
1985 struct multi_pack_index
*m
;
1987 prepare_packed_git(r
);
1988 if (!r
->objects
->packed_git
&& !r
->objects
->multi_pack_index
)
1991 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
1992 if (fill_midx_entry(r
, oid
, e
, m
))
1996 list_for_each(pos
, &r
->objects
->packed_git_mru
) {
1997 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
1998 if (!p
->multi_pack_index
&& fill_pack_entry(oid
, e
, p
)) {
1999 list_move(&p
->mru
, &r
->objects
->packed_git_mru
);
2006 int has_object_pack(const struct object_id
*oid
)
2008 struct pack_entry e
;
2009 return find_pack_entry(the_repository
, oid
, &e
);
2012 int has_pack_index(const unsigned char *sha1
)
2015 if (stat(sha1_pack_index_name(sha1
), &st
))
2020 int for_each_object_in_pack(struct packed_git
*p
,
2021 each_packed_object_fn cb
, void *data
,
2022 enum for_each_object_flags flags
)
2027 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
) {
2028 if (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
);