1 #include "git-compat-util.h"
7 #include "repository.h"
12 #include "streaming.h"
13 #include "hash-lookup.h"
17 #include "tree-walk.h"
19 #include "object-store.h"
21 #include "commit-graph.h"
22 #include "promisor-remote.h"
24 char *odb_pack_name(struct strbuf
*buf
,
25 const unsigned char *hash
,
29 strbuf_addf(buf
, "%s/pack/pack-%s.%s", get_object_directory(),
30 hash_to_hex(hash
), ext
);
34 char *sha1_pack_name(const unsigned char *sha1
)
36 static struct strbuf buf
= STRBUF_INIT
;
37 return odb_pack_name(&buf
, sha1
, "pack");
40 char *sha1_pack_index_name(const unsigned char *sha1
)
42 static struct strbuf buf
= STRBUF_INIT
;
43 return odb_pack_name(&buf
, sha1
, "idx");
46 static unsigned int pack_used_ctr
;
47 static unsigned int pack_mmap_calls
;
48 static unsigned int peak_pack_open_windows
;
49 static unsigned int pack_open_windows
;
50 static unsigned int pack_open_fds
;
51 static unsigned int pack_max_fds
;
52 static size_t peak_pack_mapped
;
53 static size_t pack_mapped
;
55 #define SZ_FMT PRIuMAX
56 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
58 void pack_report(void)
61 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
62 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
63 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
64 sz_fmt(getpagesize()),
65 sz_fmt(packed_git_window_size
),
66 sz_fmt(packed_git_limit
));
68 "pack_report: pack_used_ctr = %10u\n"
69 "pack_report: pack_mmap_calls = %10u\n"
70 "pack_report: pack_open_windows = %10u / %10u\n"
71 "pack_report: pack_mapped = "
72 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
75 pack_open_windows
, peak_pack_open_windows
,
76 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
80 * Open and mmap the index file at path, perform a couple of
81 * consistency checks, then record its information to p. Return 0 on
84 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
88 int fd
= git_open(path
), ret
;
90 const unsigned int hashsz
= the_hash_algo
->rawsz
;
98 idx_size
= xsize_t(st
.st_size
);
99 if (idx_size
< 4 * 256 + hashsz
+ hashsz
) {
101 return error("index file %s is too small", path
);
103 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
106 ret
= load_idx(path
, hashsz
, idx_map
, idx_size
, p
);
109 munmap(idx_map
, idx_size
);
114 int load_idx(const char *path
, const unsigned int hashsz
, void *idx_map
,
115 size_t idx_size
, struct packed_git
*p
)
117 struct pack_idx_header
*hdr
= idx_map
;
118 uint32_t version
, nr
, i
, *index
;
120 if (idx_size
< 4 * 256 + hashsz
+ hashsz
)
121 return error("index file %s is too small", path
);
123 return error("empty data");
125 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
126 version
= ntohl(hdr
->idx_version
);
127 if (version
< 2 || version
> 2)
128 return error("index file %s is version %"PRIu32
129 " and is not supported by this binary"
130 " (try upgrading GIT to a newer version)",
138 index
+= 2; /* skip index header */
139 for (i
= 0; i
< 256; i
++) {
140 uint32_t n
= ntohl(index
[i
]);
142 return error("non-monotonic index %s", path
);
149 * - 256 index entries 4 bytes each
150 * - 24-byte entries * nr (object ID + 4-byte offset)
151 * - hash of the packfile
154 if (idx_size
!= st_add(4 * 256 + hashsz
+ hashsz
, st_mult(nr
, hashsz
+ 4)))
155 return error("wrong index v1 file size in %s", path
);
156 } else if (version
== 2) {
159 * - 8 bytes of header
160 * - 256 index entries 4 bytes each
161 * - object ID entry * nr
162 * - 4-byte crc entry * nr
163 * - 4-byte offset entry * nr
164 * - hash of the packfile
166 * And after the 4-byte offset table might be a
167 * variable sized table containing 8-byte entries
168 * for offsets larger than 2^31.
170 size_t min_size
= st_add(8 + 4*256 + hashsz
+ hashsz
, st_mult(nr
, hashsz
+ 4 + 4));
171 size_t max_size
= min_size
;
173 max_size
= st_add(max_size
, st_mult(nr
- 1, 8));
174 if (idx_size
< min_size
|| idx_size
> max_size
)
175 return error("wrong index v2 file size in %s", path
);
176 if (idx_size
!= min_size
&&
178 * make sure we can deal with large pack offsets.
179 * 31-bit signed offset won't be enough, neither
180 * 32-bit unsigned one will be.
182 (sizeof(off_t
) <= 4))
183 return error("pack too large for current definition of off_t in %s", path
);
184 p
->crc_offset
= 8 + 4 * 256 + nr
* hashsz
;
187 p
->index_version
= version
;
188 p
->index_data
= idx_map
;
189 p
->index_size
= idx_size
;
194 int open_pack_index(struct packed_git
*p
)
203 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
204 BUG("pack_name does not end in .pack");
205 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
206 ret
= check_packed_git_idx(idx_name
, p
);
211 uint32_t get_pack_fanout(struct packed_git
*p
, uint32_t value
)
213 const uint32_t *level1_ofs
= p
->index_data
;
216 if (open_pack_index(p
))
218 level1_ofs
= p
->index_data
;
221 if (p
->index_version
> 1) {
225 return ntohl(level1_ofs
[value
]);
228 static struct packed_git
*alloc_packed_git(int extra
)
230 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
231 memset(p
, 0, sizeof(*p
));
236 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
238 const char *path
= sha1_pack_name(sha1
);
239 size_t alloc
= st_add(strlen(path
), 1);
240 struct packed_git
*p
= alloc_packed_git(alloc
);
242 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
243 hashcpy(p
->hash
, sha1
);
244 if (check_packed_git_idx(idx_path
, p
)) {
252 static void scan_windows(struct packed_git
*p
,
253 struct packed_git
**lru_p
,
254 struct pack_window
**lru_w
,
255 struct pack_window
**lru_l
)
257 struct pack_window
*w
, *w_l
;
259 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
261 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
271 static int unuse_one_window(struct packed_git
*current
)
273 struct packed_git
*p
, *lru_p
= NULL
;
274 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
277 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
278 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
)
279 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
281 munmap(lru_w
->base
, lru_w
->len
);
282 pack_mapped
-= lru_w
->len
;
284 lru_l
->next
= lru_w
->next
;
286 lru_p
->windows
= lru_w
->next
;
294 void close_pack_windows(struct packed_git
*p
)
297 struct pack_window
*w
= p
->windows
;
300 die("pack '%s' still has open windows to it",
302 munmap(w
->base
, w
->len
);
303 pack_mapped
-= w
->len
;
305 p
->windows
= w
->next
;
310 int close_pack_fd(struct packed_git
*p
)
322 void close_pack_index(struct packed_git
*p
)
325 munmap((void *)p
->index_data
, p
->index_size
);
326 p
->index_data
= NULL
;
330 static void close_pack_revindex(struct packed_git
*p
)
332 if (!p
->revindex_map
)
335 munmap((void *)p
->revindex_map
, p
->revindex_size
);
336 p
->revindex_map
= NULL
;
337 p
->revindex_data
= NULL
;
340 static void close_pack_mtimes(struct packed_git
*p
)
345 munmap((void *)p
->mtimes_map
, p
->mtimes_size
);
346 p
->mtimes_map
= NULL
;
349 void close_pack(struct packed_git
*p
)
351 close_pack_windows(p
);
354 close_pack_revindex(p
);
355 close_pack_mtimes(p
);
356 oidset_clear(&p
->bad_objects
);
359 void close_object_store(struct raw_object_store
*o
)
361 struct packed_git
*p
;
363 for (p
= o
->packed_git
; p
; p
= p
->next
)
365 BUG("want to close pack marked 'do-not-close'");
369 if (o
->multi_pack_index
) {
370 close_midx(o
->multi_pack_index
);
371 o
->multi_pack_index
= NULL
;
374 close_commit_graph(o
);
377 void unlink_pack_path(const char *pack_name
, int force_delete
)
379 static const char *exts
[] = {".pack", ".idx", ".rev", ".keep", ".bitmap", ".promisor", ".mtimes"};
381 struct strbuf buf
= STRBUF_INIT
;
384 strbuf_addstr(&buf
, pack_name
);
385 strip_suffix_mem(buf
.buf
, &buf
.len
, ".pack");
389 strbuf_addstr(&buf
, ".keep");
390 if (!access(buf
.buf
, F_OK
)) {
391 strbuf_release(&buf
);
396 for (i
= 0; i
< ARRAY_SIZE(exts
); i
++) {
397 strbuf_setlen(&buf
, plen
);
398 strbuf_addstr(&buf
, exts
[i
]);
402 strbuf_release(&buf
);
406 * The LRU pack is the one with the oldest MRU window, preferring packs
407 * with no used windows, or the oldest mtime if it has no windows allocated.
409 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
411 struct pack_window
*w
, *this_mru_w
;
412 int has_windows_inuse
= 0;
415 * Reject this pack if it has windows and the previously selected
416 * one does not. If this pack does not have windows, reject
417 * it if the pack file is newer than the previously selected one.
419 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
422 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
424 * Reject this pack if any of its windows are in use,
425 * but the previously selected pack did not have any
426 * inuse windows. Otherwise, record that this pack
427 * has windows in use.
430 if (*accept_windows_inuse
)
431 has_windows_inuse
= 1;
436 if (w
->last_used
> this_mru_w
->last_used
)
440 * Reject this pack if it has windows that have been
441 * used more recently than the previously selected pack.
442 * If the previously selected pack had windows inuse and
443 * we have not encountered a window in this pack that is
444 * inuse, skip this check since we prefer a pack with no
445 * inuse windows to one that has inuse windows.
447 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
448 this_mru_w
->last_used
> (*mru_w
)->last_used
)
457 *accept_windows_inuse
= has_windows_inuse
;
460 static int close_one_pack(void)
462 struct packed_git
*p
, *lru_p
= NULL
;
463 struct pack_window
*mru_w
= NULL
;
464 int accept_windows_inuse
= 1;
466 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
467 if (p
->pack_fd
== -1)
469 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
473 return close_pack_fd(lru_p
);
478 static unsigned int get_max_fd_limit(void)
484 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
491 long open_max
= sysconf(_SC_OPEN_MAX
);
495 * Otherwise, we got -1 for one of the two
498 * (1) sysconf() did not understand _SC_OPEN_MAX
499 * and signaled an error with -1; or
500 * (2) sysconf() said there is no limit.
502 * We _could_ clear errno before calling sysconf() to
503 * tell these two cases apart and return a huge number
504 * in the latter case to let the caller cap it to a
505 * value that is not so selfish, but letting the
506 * fallback OPEN_MAX codepath take care of these cases
515 return 1; /* see the caller ;-) */
519 const char *pack_basename(struct packed_git
*p
)
521 const char *ret
= strrchr(p
->pack_name
, '/');
523 ret
= ret
+ 1; /* skip past slash */
525 ret
= p
->pack_name
; /* we only have a base */
530 * Do not call this directly as this leaks p->pack_fd on error return;
531 * call open_packed_git() instead.
533 static int open_packed_git_1(struct packed_git
*p
)
536 struct pack_header hdr
;
537 unsigned char hash
[GIT_MAX_RAWSZ
];
538 unsigned char *idx_hash
;
540 const unsigned hashsz
= the_hash_algo
->rawsz
;
542 if (open_pack_index(p
))
543 return error("packfile %s index unavailable", p
->pack_name
);
546 unsigned int max_fds
= get_max_fd_limit();
548 /* Save 3 for stdin/stdout/stderr, 22 for work */
550 pack_max_fds
= max_fds
- 25;
555 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
558 p
->pack_fd
= git_open(p
->pack_name
);
559 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
563 /* If we created the struct before we had the pack we lack size. */
565 if (!S_ISREG(st
.st_mode
))
566 return error("packfile %s not a regular file", p
->pack_name
);
567 p
->pack_size
= st
.st_size
;
568 } else if (p
->pack_size
!= st
.st_size
)
569 return error("packfile %s size changed", p
->pack_name
);
571 /* Verify we recognize this pack file format. */
572 read_result
= read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
));
574 return error_errno("error reading from %s", p
->pack_name
);
575 if (read_result
!= sizeof(hdr
))
576 return error("file %s is far too short to be a packfile", p
->pack_name
);
577 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
578 return error("file %s is not a GIT packfile", p
->pack_name
);
579 if (!pack_version_ok(hdr
.hdr_version
))
580 return error("packfile %s is version %"PRIu32
" and not"
581 " supported (try upgrading GIT to a newer version)",
582 p
->pack_name
, ntohl(hdr
.hdr_version
));
584 /* Verify the pack matches its index. */
585 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
586 return error("packfile %s claims to have %"PRIu32
" objects"
587 " while index indicates %"PRIu32
" objects",
588 p
->pack_name
, ntohl(hdr
.hdr_entries
),
590 read_result
= pread_in_full(p
->pack_fd
, hash
, hashsz
,
591 p
->pack_size
- hashsz
);
593 return error_errno("error reading from %s", p
->pack_name
);
594 if (read_result
!= hashsz
)
595 return error("packfile %s signature is unavailable", p
->pack_name
);
596 idx_hash
= ((unsigned char *)p
->index_data
) + p
->index_size
- hashsz
* 2;
597 if (!hasheq(hash
, idx_hash
))
598 return error("packfile %s does not match index", p
->pack_name
);
602 static int open_packed_git(struct packed_git
*p
)
604 if (!open_packed_git_1(p
))
610 static int in_window(struct pack_window
*win
, off_t offset
)
612 /* We must promise at least one full hash after the
613 * offset is available from this window, otherwise the offset
614 * is not actually in this window and a different window (which
615 * has that one hash excess) must be used. This is to support
616 * the object header and delta base parsing routines below.
618 off_t win_off
= win
->offset
;
619 return win_off
<= offset
620 && (offset
+ the_hash_algo
->rawsz
) <= (win_off
+ win
->len
);
623 unsigned char *use_pack(struct packed_git
*p
,
624 struct pack_window
**w_cursor
,
628 struct pack_window
*win
= *w_cursor
;
630 /* Since packfiles end in a hash of their content and it's
631 * pointless to ask for an offset into the middle of that
632 * hash, and the in_window function above wouldn't match
633 * don't allow an offset too close to the end of the file.
635 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
636 die("packfile %s cannot be accessed", p
->pack_name
);
637 if (offset
> (p
->pack_size
- the_hash_algo
->rawsz
))
638 die("offset beyond end of packfile (truncated pack?)");
640 die(_("offset before end of packfile (broken .idx?)"));
642 if (!win
|| !in_window(win
, offset
)) {
645 for (win
= p
->windows
; win
; win
= win
->next
) {
646 if (in_window(win
, offset
))
650 size_t window_align
= packed_git_window_size
/ 2;
653 if (p
->pack_fd
== -1 && open_packed_git(p
))
654 die("packfile %s cannot be accessed", p
->pack_name
);
656 CALLOC_ARRAY(win
, 1);
657 win
->offset
= (offset
/ window_align
) * window_align
;
658 len
= p
->pack_size
- win
->offset
;
659 if (len
> packed_git_window_size
)
660 len
= packed_git_window_size
;
661 win
->len
= (size_t)len
;
662 pack_mapped
+= win
->len
;
663 while (packed_git_limit
< pack_mapped
664 && unuse_one_window(p
))
666 win
->base
= xmmap_gently(NULL
, win
->len
,
667 PROT_READ
, MAP_PRIVATE
,
668 p
->pack_fd
, win
->offset
);
669 if (win
->base
== MAP_FAILED
)
670 die_errno(_("packfile %s cannot be mapped%s"),
671 p
->pack_name
, mmap_os_err());
672 if (!win
->offset
&& win
->len
== p
->pack_size
677 if (pack_mapped
> peak_pack_mapped
)
678 peak_pack_mapped
= pack_mapped
;
679 if (pack_open_windows
> peak_pack_open_windows
)
680 peak_pack_open_windows
= pack_open_windows
;
681 win
->next
= p
->windows
;
685 if (win
!= *w_cursor
) {
686 win
->last_used
= pack_used_ctr
++;
690 offset
-= win
->offset
;
692 *left
= win
->len
- xsize_t(offset
);
693 return win
->base
+ offset
;
696 void unuse_pack(struct pack_window
**w_cursor
)
698 struct pack_window
*w
= *w_cursor
;
705 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
709 struct packed_git
*p
;
712 * Make sure a corresponding .pack file exists and that
713 * the index looks sane.
715 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
719 * ".promisor" is long enough to hold any suffix we're adding (and
720 * the use xsnprintf double-checks that)
722 alloc
= st_add3(path_len
, strlen(".promisor"), 1);
723 p
= alloc_packed_git(alloc
);
724 memcpy(p
->pack_name
, path
, path_len
);
726 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
727 if (!access(p
->pack_name
, F_OK
))
730 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".promisor");
731 if (!access(p
->pack_name
, F_OK
))
732 p
->pack_promisor
= 1;
734 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".mtimes");
735 if (!access(p
->pack_name
, F_OK
))
738 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
739 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
744 /* ok, it looks sane as far as we can check without
745 * actually mapping the pack file.
747 p
->pack_size
= st
.st_size
;
748 p
->pack_local
= local
;
749 p
->mtime
= st
.st_mtime
;
750 if (path_len
< the_hash_algo
->hexsz
||
751 get_sha1_hex(path
+ path_len
- the_hash_algo
->hexsz
, p
->hash
))
756 void install_packed_git(struct repository
*r
, struct packed_git
*pack
)
758 if (pack
->pack_fd
!= -1)
761 pack
->next
= r
->objects
->packed_git
;
762 r
->objects
->packed_git
= pack
;
764 hashmap_entry_init(&pack
->packmap_ent
, strhash(pack
->pack_name
));
765 hashmap_add(&r
->objects
->pack_map
, &pack
->packmap_ent
);
768 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
770 static void report_helper(const struct string_list
*list
,
771 int seen_bits
, int first
, int last
)
773 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
776 for (; first
< last
; first
++)
777 report_garbage(seen_bits
, list
->items
[first
].string
);
780 static void report_pack_garbage(struct string_list
*list
)
782 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
787 string_list_sort(list
);
789 for (i
= 0; i
< list
->nr
; i
++) {
790 const char *path
= list
->items
[i
].string
;
792 strncmp(path
, list
->items
[first
].string
, baselen
)) {
793 report_helper(list
, seen_bits
, first
, i
);
798 const char *dot
= strrchr(path
, '.');
800 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
803 baselen
= dot
- path
+ 1;
806 if (!strcmp(path
+ baselen
, "pack"))
808 else if (!strcmp(path
+ baselen
, "idx"))
811 report_helper(list
, seen_bits
, first
, list
->nr
);
814 void for_each_file_in_pack_dir(const char *objdir
,
815 each_file_in_pack_dir_fn fn
,
818 struct strbuf path
= STRBUF_INIT
;
823 strbuf_addstr(&path
, objdir
);
824 strbuf_addstr(&path
, "/pack");
825 dir
= opendir(path
.buf
);
828 error_errno("unable to open object pack directory: %s",
830 strbuf_release(&path
);
833 strbuf_addch(&path
, '/');
834 dirnamelen
= path
.len
;
835 while ((de
= readdir_skip_dot_and_dotdot(dir
)) != NULL
) {
836 strbuf_setlen(&path
, dirnamelen
);
837 strbuf_addstr(&path
, de
->d_name
);
839 fn(path
.buf
, path
.len
, de
->d_name
, data
);
843 strbuf_release(&path
);
846 struct prepare_pack_data
{
847 struct repository
*r
;
848 struct string_list
*garbage
;
850 struct multi_pack_index
*m
;
853 static void prepare_pack(const char *full_name
, size_t full_name_len
,
854 const char *file_name
, void *_data
)
856 struct prepare_pack_data
*data
= (struct prepare_pack_data
*)_data
;
857 struct packed_git
*p
;
858 size_t base_len
= full_name_len
;
860 if (strip_suffix_mem(full_name
, &base_len
, ".idx") &&
861 !(data
->m
&& midx_contains_pack(data
->m
, file_name
))) {
862 struct hashmap_entry hent
;
863 char *pack_name
= xstrfmt("%.*s.pack", (int)base_len
, full_name
);
864 unsigned int hash
= strhash(pack_name
);
865 hashmap_entry_init(&hent
, hash
);
867 /* Don't reopen a pack we already have. */
868 if (!hashmap_get(&data
->r
->objects
->pack_map
, &hent
, pack_name
)) {
869 p
= add_packed_git(full_name
, full_name_len
, data
->local
);
871 install_packed_git(data
->r
, p
);
879 if (!strcmp(file_name
, "multi-pack-index"))
881 if (starts_with(file_name
, "multi-pack-index") &&
882 (ends_with(file_name
, ".bitmap") || ends_with(file_name
, ".rev")))
884 if (ends_with(file_name
, ".idx") ||
885 ends_with(file_name
, ".rev") ||
886 ends_with(file_name
, ".pack") ||
887 ends_with(file_name
, ".bitmap") ||
888 ends_with(file_name
, ".keep") ||
889 ends_with(file_name
, ".promisor") ||
890 ends_with(file_name
, ".mtimes"))
891 string_list_append(data
->garbage
, full_name
);
893 report_garbage(PACKDIR_FILE_GARBAGE
, full_name
);
896 static void prepare_packed_git_one(struct repository
*r
, char *objdir
, int local
)
898 struct prepare_pack_data data
;
899 struct string_list garbage
= STRING_LIST_INIT_DUP
;
901 data
.m
= r
->objects
->multi_pack_index
;
903 /* look for the multi-pack-index for this object directory */
904 while (data
.m
&& strcmp(data
.m
->object_dir
, objdir
))
905 data
.m
= data
.m
->next
;
908 data
.garbage
= &garbage
;
911 for_each_file_in_pack_dir(objdir
, prepare_pack
, &data
);
913 report_pack_garbage(data
.garbage
);
914 string_list_clear(data
.garbage
, 0);
917 static void prepare_packed_git(struct repository
*r
);
919 * Give a fast, rough count of the number of objects in the repository. This
920 * ignores loose objects completely. If you have a lot of them, then either
921 * you should repack because your performance will be awful, or they are
922 * all unreachable objects about to be pruned, in which case they're not really
923 * interesting as a measure of repo size in the first place.
925 unsigned long repo_approximate_object_count(struct repository
*r
)
927 if (!r
->objects
->approximate_object_count_valid
) {
929 struct multi_pack_index
*m
;
930 struct packed_git
*p
;
932 prepare_packed_git(r
);
934 for (m
= get_multi_pack_index(r
); m
; m
= m
->next
)
935 count
+= m
->num_objects
;
936 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
) {
937 if (open_pack_index(p
))
939 count
+= p
->num_objects
;
941 r
->objects
->approximate_object_count
= count
;
942 r
->objects
->approximate_object_count_valid
= 1;
944 return r
->objects
->approximate_object_count
;
947 DEFINE_LIST_SORT(static, sort_packs
, struct packed_git
, next
);
949 static int sort_pack(const struct packed_git
*a
, const struct packed_git
*b
)
954 * Local packs tend to contain objects specific to our
955 * variant of the project than remote ones. In addition,
956 * remote ones could be on a network mounted filesystem.
957 * Favor local ones for these reasons.
959 st
= a
->pack_local
- b
->pack_local
;
964 * Younger packs tend to contain more recent objects,
965 * and more recent objects tend to get accessed more
968 if (a
->mtime
< b
->mtime
)
970 else if (a
->mtime
== b
->mtime
)
975 static void rearrange_packed_git(struct repository
*r
)
977 sort_packs(&r
->objects
->packed_git
, sort_pack
);
980 static void prepare_packed_git_mru(struct repository
*r
)
982 struct packed_git
*p
;
984 INIT_LIST_HEAD(&r
->objects
->packed_git_mru
);
986 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
987 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
990 static void prepare_packed_git(struct repository
*r
)
992 struct object_directory
*odb
;
994 if (r
->objects
->packed_git_initialized
)
998 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
999 int local
= (odb
== r
->objects
->odb
);
1000 prepare_multi_pack_index_one(r
, odb
->path
, local
);
1001 prepare_packed_git_one(r
, odb
->path
, local
);
1003 rearrange_packed_git(r
);
1005 prepare_packed_git_mru(r
);
1006 r
->objects
->packed_git_initialized
= 1;
1009 void reprepare_packed_git(struct repository
*r
)
1011 struct object_directory
*odb
;
1016 * Reprepare alt odbs, in case the alternates file was modified
1017 * during the course of this process. This only _adds_ odbs to
1018 * the linked list, so existing odbs will continue to exist for
1019 * the lifetime of the process.
1021 r
->objects
->loaded_alternates
= 0;
1024 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
)
1025 odb_clear_loose_cache(odb
);
1027 r
->objects
->approximate_object_count_valid
= 0;
1028 r
->objects
->packed_git_initialized
= 0;
1029 prepare_packed_git(r
);
1033 struct packed_git
*get_packed_git(struct repository
*r
)
1035 prepare_packed_git(r
);
1036 return r
->objects
->packed_git
;
1039 struct multi_pack_index
*get_multi_pack_index(struct repository
*r
)
1041 prepare_packed_git(r
);
1042 return r
->objects
->multi_pack_index
;
1045 struct multi_pack_index
*get_local_multi_pack_index(struct repository
*r
)
1047 struct multi_pack_index
*m
= get_multi_pack_index(r
);
1049 /* no need to iterate; we always put the local one first (if any) */
1056 struct packed_git
*get_all_packs(struct repository
*r
)
1058 struct multi_pack_index
*m
;
1060 prepare_packed_git(r
);
1061 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
1063 for (i
= 0; i
< m
->num_packs
; i
++)
1064 prepare_midx_pack(r
, m
, i
);
1067 return r
->objects
->packed_git
;
1070 struct list_head
*get_packed_git_mru(struct repository
*r
)
1072 prepare_packed_git(r
);
1073 return &r
->objects
->packed_git_mru
;
1076 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1077 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1081 unsigned long used
= 0;
1084 *type
= (c
>> 4) & 7;
1088 if (len
<= used
|| (bitsizeof(long) - 7) < shift
) {
1089 error("bad object header");
1094 size
= st_add(size
, st_left_shift(c
& 0x7f, shift
));
1097 *sizep
= cast_size_t_to_ulong(size
);
1101 unsigned long get_size_from_delta(struct packed_git
*p
,
1102 struct pack_window
**w_curs
,
1105 const unsigned char *data
;
1106 unsigned char delta_head
[20], *in
;
1110 memset(&stream
, 0, sizeof(stream
));
1111 stream
.next_out
= delta_head
;
1112 stream
.avail_out
= sizeof(delta_head
);
1114 git_inflate_init(&stream
);
1116 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1117 stream
.next_in
= in
;
1119 * Note: the window section returned by use_pack() must be
1120 * available throughout git_inflate()'s unlocked execution. To
1121 * ensure no other thread will modify the window in the
1122 * meantime, we rely on the packed_window.inuse_cnt. This
1123 * counter is incremented before window reading and checked
1124 * before window disposal.
1126 * Other worrying sections could be the call to close_pack_fd(),
1127 * which can close packs even with in-use windows, and to
1128 * reprepare_packed_git(). Regarding the former, mmap doc says:
1129 * "closing the file descriptor does not unmap the region". And
1130 * for the latter, it won't re-open already available packs.
1133 st
= git_inflate(&stream
, Z_FINISH
);
1135 curpos
+= stream
.next_in
- in
;
1136 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1137 stream
.total_out
< sizeof(delta_head
));
1138 git_inflate_end(&stream
);
1139 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1140 error("delta data unpack-initial failed");
1144 /* Examine the initial part of the delta to figure out
1149 /* ignore base size */
1150 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1152 /* Read the result size */
1153 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1156 int unpack_object_header(struct packed_git
*p
,
1157 struct pack_window
**w_curs
,
1159 unsigned long *sizep
)
1161 unsigned char *base
;
1164 enum object_type type
;
1166 /* use_pack() assures us we have [base, base + 20) available
1167 * as a range that we can look at. (Its actually the hash
1168 * size that is assured.) With our object header encoding
1169 * the maximum deflated object size is 2^137, which is just
1170 * insane, so we know won't exceed what we have been given.
1172 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1173 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1182 void mark_bad_packed_object(struct packed_git
*p
, const struct object_id
*oid
)
1184 oidset_insert(&p
->bad_objects
, oid
);
1187 const struct packed_git
*has_packed_and_bad(struct repository
*r
,
1188 const struct object_id
*oid
)
1190 struct packed_git
*p
;
1192 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
1193 if (oidset_contains(&p
->bad_objects
, oid
))
1198 off_t
get_delta_base(struct packed_git
*p
,
1199 struct pack_window
**w_curs
,
1201 enum object_type type
,
1202 off_t delta_obj_offset
)
1204 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1207 /* use_pack() assured us we have [base_info, base_info + 20)
1208 * as a range that we can look at without walking off the
1209 * end of the mapped window. Its actually the hash size
1210 * that is assured. An OFS_DELTA longer than the hash size
1211 * is stupid, as then a REF_DELTA would be smaller to store.
1213 if (type
== OBJ_OFS_DELTA
) {
1215 unsigned char c
= base_info
[used
++];
1216 base_offset
= c
& 127;
1219 if (!base_offset
|| MSB(base_offset
, 7))
1220 return 0; /* overflow */
1221 c
= base_info
[used
++];
1222 base_offset
= (base_offset
<< 7) + (c
& 127);
1224 base_offset
= delta_obj_offset
- base_offset
;
1225 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1226 return 0; /* out of bound */
1228 } else if (type
== OBJ_REF_DELTA
) {
1229 /* The base entry _must_ be in the same pack */
1230 base_offset
= find_pack_entry_one(base_info
, p
);
1231 *curpos
+= the_hash_algo
->rawsz
;
1233 die("I am totally screwed");
1238 * Like get_delta_base above, but we return the sha1 instead of the pack
1239 * offset. This means it is cheaper for REF deltas (we do not have to do
1240 * the final object lookup), but more expensive for OFS deltas (we
1241 * have to load the revidx to convert the offset back into a sha1).
1243 static int get_delta_base_oid(struct packed_git
*p
,
1244 struct pack_window
**w_curs
,
1246 struct object_id
*oid
,
1247 enum object_type type
,
1248 off_t delta_obj_offset
)
1250 if (type
== OBJ_REF_DELTA
) {
1251 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1254 } else if (type
== OBJ_OFS_DELTA
) {
1256 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1257 type
, delta_obj_offset
);
1262 if (offset_to_pack_pos(p
, base_offset
, &base_pos
) < 0)
1265 return nth_packed_object_id(oid
, p
,
1266 pack_pos_to_index(p
, base_pos
));
1271 static int retry_bad_packed_offset(struct repository
*r
,
1272 struct packed_git
*p
,
1277 struct object_id oid
;
1278 if (offset_to_pack_pos(p
, obj_offset
, &pos
) < 0)
1280 nth_packed_object_id(&oid
, p
, pack_pos_to_index(p
, pos
));
1281 mark_bad_packed_object(p
, &oid
);
1282 type
= oid_object_info(r
, &oid
, NULL
);
1283 if (type
<= OBJ_NONE
)
1288 #define POI_STACK_PREALLOC 64
1290 static enum object_type
packed_to_object_type(struct repository
*r
,
1291 struct packed_git
*p
,
1293 enum object_type type
,
1294 struct pack_window
**w_curs
,
1297 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1298 off_t
*poi_stack
= small_poi_stack
;
1299 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1301 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1304 /* Push the object we're going to leave behind */
1305 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1306 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1307 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
1308 COPY_ARRAY(poi_stack
, small_poi_stack
, poi_stack_nr
);
1310 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1312 poi_stack
[poi_stack_nr
++] = obj_offset
;
1313 /* If parsing the base offset fails, just unwind */
1314 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1317 curpos
= obj_offset
= base_offset
;
1318 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1319 if (type
<= OBJ_NONE
) {
1320 /* If getting the base itself fails, we first
1321 * retry the base, otherwise unwind */
1322 type
= retry_bad_packed_offset(r
, p
, base_offset
);
1323 if (type
> OBJ_NONE
)
1337 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1338 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1343 if (poi_stack
!= small_poi_stack
)
1348 while (poi_stack_nr
) {
1349 obj_offset
= poi_stack
[--poi_stack_nr
];
1350 type
= retry_bad_packed_offset(r
, p
, obj_offset
);
1351 if (type
> OBJ_NONE
)
1358 static struct hashmap delta_base_cache
;
1359 static size_t delta_base_cached
;
1361 static LIST_HEAD(delta_base_cache_lru
);
1363 struct delta_base_cache_key
{
1364 struct packed_git
*p
;
1368 struct delta_base_cache_entry
{
1369 struct hashmap_entry ent
;
1370 struct delta_base_cache_key key
;
1371 struct list_head lru
;
1374 enum object_type type
;
1377 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1381 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
1382 hash
+= (hash
>> 8) + (hash
>> 16);
1386 static struct delta_base_cache_entry
*
1387 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1389 struct hashmap_entry entry
, *e
;
1390 struct delta_base_cache_key key
;
1392 if (!delta_base_cache
.cmpfn
)
1395 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
1397 key
.base_offset
= base_offset
;
1398 e
= hashmap_get(&delta_base_cache
, &entry
, &key
);
1399 return e
? container_of(e
, struct delta_base_cache_entry
, ent
) : NULL
;
1402 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
1403 const struct delta_base_cache_key
*b
)
1405 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
1408 static int delta_base_cache_hash_cmp(const void *cmp_data UNUSED
,
1409 const struct hashmap_entry
*va
,
1410 const struct hashmap_entry
*vb
,
1413 const struct delta_base_cache_entry
*a
, *b
;
1414 const struct delta_base_cache_key
*key
= vkey
;
1416 a
= container_of(va
, const struct delta_base_cache_entry
, ent
);
1417 b
= container_of(vb
, const struct delta_base_cache_entry
, ent
);
1420 return !delta_base_cache_key_eq(&a
->key
, key
);
1422 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
1425 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1427 return !!get_delta_base_cache_entry(p
, base_offset
);
1431 * Remove the entry from the cache, but do _not_ free the associated
1432 * entry data. The caller takes ownership of the "data" buffer, and
1433 * should copy out any fields it wants before detaching.
1435 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1437 hashmap_remove(&delta_base_cache
, &ent
->ent
, &ent
->key
);
1438 list_del(&ent
->lru
);
1439 delta_base_cached
-= ent
->size
;
1443 static void *cache_or_unpack_entry(struct repository
*r
, struct packed_git
*p
,
1444 off_t base_offset
, unsigned long *base_size
,
1445 enum object_type
*type
)
1447 struct delta_base_cache_entry
*ent
;
1449 ent
= get_delta_base_cache_entry(p
, base_offset
);
1451 return unpack_entry(r
, p
, base_offset
, type
, base_size
);
1456 *base_size
= ent
->size
;
1457 return xmemdupz(ent
->data
, ent
->size
);
1460 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1463 detach_delta_base_cache_entry(ent
);
1466 void clear_delta_base_cache(void)
1468 struct list_head
*lru
, *tmp
;
1469 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1470 struct delta_base_cache_entry
*entry
=
1471 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1472 release_delta_base_cache(entry
);
1476 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1477 void *base
, unsigned long base_size
, enum object_type type
)
1479 struct delta_base_cache_entry
*ent
;
1480 struct list_head
*lru
, *tmp
;
1483 * Check required to avoid redundant entries when more than one thread
1484 * is unpacking the same object, in unpack_entry() (since its phases I
1485 * and III might run concurrently across multiple threads).
1487 if (in_delta_base_cache(p
, base_offset
)) {
1492 delta_base_cached
+= base_size
;
1494 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1495 struct delta_base_cache_entry
*f
=
1496 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1497 if (delta_base_cached
<= delta_base_cache_limit
)
1499 release_delta_base_cache(f
);
1502 ent
= xmalloc(sizeof(*ent
));
1504 ent
->key
.base_offset
= base_offset
;
1507 ent
->size
= base_size
;
1508 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
1510 if (!delta_base_cache
.cmpfn
)
1511 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, NULL
, 0);
1512 hashmap_entry_init(&ent
->ent
, pack_entry_hash(p
, base_offset
));
1513 hashmap_add(&delta_base_cache
, &ent
->ent
);
1516 int packed_object_info(struct repository
*r
, struct packed_git
*p
,
1517 off_t obj_offset
, struct object_info
*oi
)
1519 struct pack_window
*w_curs
= NULL
;
1521 off_t curpos
= obj_offset
;
1522 enum object_type type
;
1525 * We always get the representation type, but only convert it to
1526 * a "real" type later if the caller is interested.
1529 *oi
->contentp
= cache_or_unpack_entry(r
, p
, obj_offset
, oi
->sizep
,
1534 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1537 if (!oi
->contentp
&& oi
->sizep
) {
1538 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1539 off_t tmp_pos
= curpos
;
1540 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1546 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1547 if (*oi
->sizep
== 0) {
1556 if (oi
->disk_sizep
) {
1558 if (offset_to_pack_pos(p
, obj_offset
, &pos
) < 0) {
1559 error("could not find object at offset %"PRIuMAX
" "
1560 "in pack %s", (uintmax_t)obj_offset
, p
->pack_name
);
1565 *oi
->disk_sizep
= pack_pos_to_offset(p
, pos
+ 1) - obj_offset
;
1568 if (oi
->typep
|| oi
->type_name
) {
1569 enum object_type ptot
;
1570 ptot
= packed_to_object_type(r
, p
, obj_offset
,
1571 type
, &w_curs
, curpos
);
1574 if (oi
->type_name
) {
1575 const char *tn
= type_name(ptot
);
1577 strbuf_addstr(oi
->type_name
, tn
);
1585 if (oi
->delta_base_oid
) {
1586 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1587 if (get_delta_base_oid(p
, &w_curs
, curpos
,
1589 type
, obj_offset
) < 0) {
1594 oidclr(oi
->delta_base_oid
);
1597 oi
->whence
= in_delta_base_cache(p
, obj_offset
) ? OI_DBCACHED
:
1601 unuse_pack(&w_curs
);
1605 static void *unpack_compressed_entry(struct packed_git
*p
,
1606 struct pack_window
**w_curs
,
1612 unsigned char *buffer
, *in
;
1614 buffer
= xmallocz_gently(size
);
1617 memset(&stream
, 0, sizeof(stream
));
1618 stream
.next_out
= buffer
;
1619 stream
.avail_out
= size
+ 1;
1621 git_inflate_init(&stream
);
1623 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1624 stream
.next_in
= in
;
1626 * Note: we must ensure the window section returned by
1627 * use_pack() will be available throughout git_inflate()'s
1628 * unlocked execution. Please refer to the comment at
1629 * get_size_from_delta() to see how this is done.
1632 st
= git_inflate(&stream
, Z_FINISH
);
1634 if (!stream
.avail_out
)
1635 break; /* the payload is larger than it should be */
1636 curpos
+= stream
.next_in
- in
;
1637 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1638 git_inflate_end(&stream
);
1639 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1644 /* versions of zlib can clobber unconsumed portion of outbuf */
1645 buffer
[size
] = '\0';
1650 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1652 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
1653 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
1654 p
->pack_name
, (uintmax_t)obj_offset
);
1657 int do_check_packed_object_crc
;
1659 #define UNPACK_ENTRY_STACK_PREALLOC 64
1660 struct unpack_entry_stack_ent
{
1666 void *unpack_entry(struct repository
*r
, struct packed_git
*p
, off_t obj_offset
,
1667 enum object_type
*final_type
, unsigned long *final_size
)
1669 struct pack_window
*w_curs
= NULL
;
1670 off_t curpos
= obj_offset
;
1673 enum object_type type
;
1674 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
1675 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
1676 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
1677 int base_from_cache
= 0;
1679 write_pack_access_log(p
, obj_offset
);
1681 /* PHASE 1: drill down to the innermost base object */
1685 struct delta_base_cache_entry
*ent
;
1687 ent
= get_delta_base_cache_entry(p
, curpos
);
1692 detach_delta_base_cache_entry(ent
);
1693 base_from_cache
= 1;
1697 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1698 uint32_t pack_pos
, index_pos
;
1701 if (offset_to_pack_pos(p
, obj_offset
, &pack_pos
) < 0) {
1702 error("could not find object at offset %"PRIuMAX
" in pack %s",
1703 (uintmax_t)obj_offset
, p
->pack_name
);
1708 len
= pack_pos_to_offset(p
, pack_pos
+ 1) - obj_offset
;
1709 index_pos
= pack_pos_to_index(p
, pack_pos
);
1710 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, index_pos
)) {
1711 struct object_id oid
;
1712 nth_packed_object_id(&oid
, p
, index_pos
);
1713 error("bad packed object CRC for %s",
1715 mark_bad_packed_object(p
, &oid
);
1721 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1722 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
1725 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1727 error("failed to validate delta base reference "
1728 "at offset %"PRIuMAX
" from %s",
1729 (uintmax_t)curpos
, p
->pack_name
);
1730 /* bail to phase 2, in hopes of recovery */
1735 /* push object, proceed to base */
1736 if (delta_stack_nr
>= delta_stack_alloc
1737 && delta_stack
== small_delta_stack
) {
1738 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
1739 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
1740 COPY_ARRAY(delta_stack
, small_delta_stack
,
1743 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
1745 i
= delta_stack_nr
++;
1746 delta_stack
[i
].obj_offset
= obj_offset
;
1747 delta_stack
[i
].curpos
= curpos
;
1748 delta_stack
[i
].size
= size
;
1750 curpos
= obj_offset
= base_offset
;
1753 /* PHASE 2: handle the base */
1758 BUG("unpack_entry: left loop at a valid delta");
1764 if (!base_from_cache
)
1765 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
1769 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1770 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1773 /* PHASE 3: apply deltas in order */
1776 * 'data' holds the base data, or NULL if there was corruption
1778 while (delta_stack_nr
) {
1781 void *external_base
= NULL
;
1782 unsigned long delta_size
, base_size
= size
;
1784 off_t base_obj_offset
= obj_offset
;
1790 * We're probably in deep shit, but let's try to fetch
1791 * the required base anyway from another pack or loose.
1792 * This is costly but should happen only in the presence
1793 * of a corrupted pack, and is better than failing outright.
1796 struct object_id base_oid
;
1797 if (!(offset_to_pack_pos(p
, obj_offset
, &pos
))) {
1798 struct object_info oi
= OBJECT_INFO_INIT
;
1800 nth_packed_object_id(&base_oid
, p
,
1801 pack_pos_to_index(p
, pos
));
1802 error("failed to read delta base object %s"
1803 " at offset %"PRIuMAX
" from %s",
1804 oid_to_hex(&base_oid
), (uintmax_t)obj_offset
,
1806 mark_bad_packed_object(p
, &base_oid
);
1809 oi
.sizep
= &base_size
;
1810 oi
.contentp
= &base
;
1811 if (oid_object_info_extended(r
, &base_oid
, &oi
, 0) < 0)
1814 external_base
= base
;
1818 i
= --delta_stack_nr
;
1819 obj_offset
= delta_stack
[i
].obj_offset
;
1820 curpos
= delta_stack
[i
].curpos
;
1821 delta_size
= delta_stack
[i
].size
;
1826 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
1829 error("failed to unpack compressed delta "
1830 "at offset %"PRIuMAX
" from %s",
1831 (uintmax_t)curpos
, p
->pack_name
);
1834 data
= patch_delta(base
, base_size
, delta_data
,
1838 * We could not apply the delta; warn the user, but
1839 * keep going. Our failure will be noticed either in
1840 * the next iteration of the loop, or if this is the
1841 * final delta, in the caller when we return NULL.
1842 * Those code paths will take care of making a more
1843 * explicit warning and retrying with another copy of
1847 error("failed to apply delta");
1851 * We delay adding `base` to the cache until the end of the loop
1852 * because unpack_compressed_entry() momentarily releases the
1853 * obj_read_mutex, giving another thread the chance to access
1854 * the cache. Therefore, if `base` was already there, this other
1855 * thread could free() it (e.g. to make space for another entry)
1856 * before we are done using it.
1859 add_delta_base_cache(p
, base_obj_offset
, base
, base_size
, type
);
1862 free(external_base
);
1871 unuse_pack(&w_curs
);
1873 if (delta_stack
!= small_delta_stack
)
1879 int bsearch_pack(const struct object_id
*oid
, const struct packed_git
*p
, uint32_t *result
)
1881 const unsigned char *index_fanout
= p
->index_data
;
1882 const unsigned char *index_lookup
;
1883 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1884 int index_lookup_width
;
1887 BUG("bsearch_pack called without a valid pack-index");
1889 index_lookup
= index_fanout
+ 4 * 256;
1890 if (p
->index_version
== 1) {
1891 index_lookup_width
= hashsz
+ 4;
1894 index_lookup_width
= hashsz
;
1899 return bsearch_hash(oid
->hash
, (const uint32_t*)index_fanout
,
1900 index_lookup
, index_lookup_width
, result
);
1903 int nth_packed_object_id(struct object_id
*oid
,
1904 struct packed_git
*p
,
1907 const unsigned char *index
= p
->index_data
;
1908 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1910 if (open_pack_index(p
))
1912 index
= p
->index_data
;
1914 if (n
>= p
->num_objects
)
1917 if (p
->index_version
== 1) {
1918 oidread(oid
, index
+ (hashsz
+ 4) * n
+ 4);
1921 oidread(oid
, index
+ hashsz
* n
);
1926 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
1928 const unsigned char *ptr
= vptr
;
1929 const unsigned char *start
= p
->index_data
;
1930 const unsigned char *end
= start
+ p
->index_size
;
1932 die(_("offset before start of pack index for %s (corrupt index?)"),
1934 /* No need to check for underflow; .idx files must be at least 8 bytes */
1936 die(_("offset beyond end of pack index for %s (truncated index?)"),
1940 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1942 const unsigned char *index
= p
->index_data
;
1943 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1945 if (p
->index_version
== 1) {
1946 return ntohl(*((uint32_t *)(index
+ (hashsz
+ 4) * (size_t)n
)));
1949 index
+= 8 + (size_t)p
->num_objects
* (hashsz
+ 4);
1950 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1951 if (!(off
& 0x80000000))
1953 index
+= (size_t)p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1954 check_pack_index_ptr(p
, index
);
1955 return get_be64(index
);
1959 off_t
find_pack_entry_one(const unsigned char *sha1
,
1960 struct packed_git
*p
)
1962 const unsigned char *index
= p
->index_data
;
1963 struct object_id oid
;
1967 if (open_pack_index(p
))
1971 hashcpy(oid
.hash
, sha1
);
1972 if (bsearch_pack(&oid
, p
, &result
))
1973 return nth_packed_object_offset(p
, result
);
1977 int is_pack_valid(struct packed_git
*p
)
1979 /* An already open pack is known to be valid. */
1980 if (p
->pack_fd
!= -1)
1983 /* If the pack has one window completely covering the
1984 * file size, the pack is known to be valid even if
1985 * the descriptor is not currently open.
1988 struct pack_window
*w
= p
->windows
;
1990 if (!w
->offset
&& w
->len
== p
->pack_size
)
1994 /* Force the pack to open to prove its valid. */
1995 return !open_packed_git(p
);
1998 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1999 struct packed_git
*packs
)
2001 struct packed_git
*p
;
2003 for (p
= packs
; p
; p
= p
->next
) {
2004 if (find_pack_entry_one(sha1
, p
))
2011 static int fill_pack_entry(const struct object_id
*oid
,
2012 struct pack_entry
*e
,
2013 struct packed_git
*p
)
2017 if (oidset_size(&p
->bad_objects
) &&
2018 oidset_contains(&p
->bad_objects
, oid
))
2021 offset
= find_pack_entry_one(oid
->hash
, p
);
2026 * We are about to tell the caller where they can locate the
2027 * requested object. We better make sure the packfile is
2028 * still here and can be accessed before supplying that
2029 * answer, as it may have been deleted since the index was
2032 if (!is_pack_valid(p
))
2039 int find_pack_entry(struct repository
*r
, const struct object_id
*oid
, struct pack_entry
*e
)
2041 struct list_head
*pos
;
2042 struct multi_pack_index
*m
;
2044 prepare_packed_git(r
);
2045 if (!r
->objects
->packed_git
&& !r
->objects
->multi_pack_index
)
2048 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
2049 if (fill_midx_entry(r
, oid
, e
, m
))
2053 list_for_each(pos
, &r
->objects
->packed_git_mru
) {
2054 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
2055 if (!p
->multi_pack_index
&& fill_pack_entry(oid
, e
, p
)) {
2056 list_move(&p
->mru
, &r
->objects
->packed_git_mru
);
2063 static void maybe_invalidate_kept_pack_cache(struct repository
*r
,
2066 if (!r
->objects
->kept_pack_cache
.packs
)
2068 if (r
->objects
->kept_pack_cache
.flags
== flags
)
2070 FREE_AND_NULL(r
->objects
->kept_pack_cache
.packs
);
2071 r
->objects
->kept_pack_cache
.flags
= 0;
2074 static struct packed_git
**kept_pack_cache(struct repository
*r
, unsigned flags
)
2076 maybe_invalidate_kept_pack_cache(r
, flags
);
2078 if (!r
->objects
->kept_pack_cache
.packs
) {
2079 struct packed_git
**packs
= NULL
;
2080 size_t nr
= 0, alloc
= 0;
2081 struct packed_git
*p
;
2084 * We want "all" packs here, because we need to cover ones that
2085 * are used by a midx, as well. We need to look in every one of
2086 * them (instead of the midx itself) to cover duplicates. It's
2087 * possible that an object is found in two packs that the midx
2088 * covers, one kept and one not kept, but the midx returns only
2089 * the non-kept version.
2091 for (p
= get_all_packs(r
); p
; p
= p
->next
) {
2092 if ((p
->pack_keep
&& (flags
& ON_DISK_KEEP_PACKS
)) ||
2093 (p
->pack_keep_in_core
&& (flags
& IN_CORE_KEEP_PACKS
))) {
2094 ALLOC_GROW(packs
, nr
+ 1, alloc
);
2098 ALLOC_GROW(packs
, nr
+ 1, alloc
);
2101 r
->objects
->kept_pack_cache
.packs
= packs
;
2102 r
->objects
->kept_pack_cache
.flags
= flags
;
2105 return r
->objects
->kept_pack_cache
.packs
;
2108 int find_kept_pack_entry(struct repository
*r
,
2109 const struct object_id
*oid
,
2111 struct pack_entry
*e
)
2113 struct packed_git
**cache
;
2115 for (cache
= kept_pack_cache(r
, flags
); *cache
; cache
++) {
2116 struct packed_git
*p
= *cache
;
2117 if (fill_pack_entry(oid
, e
, p
))
2124 int has_object_pack(const struct object_id
*oid
)
2126 struct pack_entry e
;
2127 return find_pack_entry(the_repository
, oid
, &e
);
2130 int has_object_kept_pack(const struct object_id
*oid
, unsigned flags
)
2132 struct pack_entry e
;
2133 return find_kept_pack_entry(the_repository
, oid
, flags
, &e
);
2136 int has_pack_index(const unsigned char *sha1
)
2139 if (stat(sha1_pack_index_name(sha1
), &st
))
2144 int for_each_object_in_pack(struct packed_git
*p
,
2145 each_packed_object_fn cb
, void *data
,
2146 enum for_each_object_flags flags
)
2151 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
) {
2152 if (load_pack_revindex(p
))
2156 for (i
= 0; i
< p
->num_objects
; i
++) {
2158 struct object_id oid
;
2161 * We are iterating "i" from 0 up to num_objects, but its
2162 * meaning may be different, depending on the requested output
2165 * - in object-name order, it is the same as the index order
2166 * used by nth_packed_object_id(), so we can pass it
2169 * - in pack-order, it is pack position, which we must
2170 * convert to an index position in order to get the oid.
2172 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
)
2173 index_pos
= pack_pos_to_index(p
, i
);
2177 if (nth_packed_object_id(&oid
, p
, index_pos
) < 0)
2178 return error("unable to get sha1 of object %u in %s",
2179 index_pos
, p
->pack_name
);
2181 r
= cb(&oid
, p
, index_pos
, data
);
2188 int for_each_packed_object(each_packed_object_fn cb
, void *data
,
2189 enum for_each_object_flags flags
)
2191 struct packed_git
*p
;
2193 int pack_errors
= 0;
2195 prepare_packed_git(the_repository
);
2196 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
2197 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
2199 if ((flags
& FOR_EACH_OBJECT_PROMISOR_ONLY
) &&
2202 if ((flags
& FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS
) &&
2203 p
->pack_keep_in_core
)
2205 if ((flags
& FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS
) &&
2208 if (open_pack_index(p
)) {
2212 r
= for_each_object_in_pack(p
, cb
, data
, flags
);
2216 return r
? r
: pack_errors
;
2219 static int add_promisor_object(const struct object_id
*oid
,
2220 struct packed_git
*pack UNUSED
,
2221 uint32_t pos UNUSED
,
2224 struct oidset
*set
= set_
;
2226 int we_parsed_object
;
2228 obj
= lookup_object(the_repository
, oid
);
2229 if (obj
&& obj
->parsed
) {
2230 we_parsed_object
= 0;
2232 we_parsed_object
= 1;
2233 obj
= parse_object(the_repository
, oid
);
2239 oidset_insert(set
, oid
);
2242 * If this is a tree, commit, or tag, the objects it refers
2243 * to are also promisor objects. (Blobs refer to no objects->)
2245 if (obj
->type
== OBJ_TREE
) {
2246 struct tree
*tree
= (struct tree
*)obj
;
2247 struct tree_desc desc
;
2248 struct name_entry entry
;
2249 if (init_tree_desc_gently(&desc
, tree
->buffer
, tree
->size
, 0))
2251 * Error messages are given when packs are
2252 * verified, so do not print any here.
2255 while (tree_entry_gently(&desc
, &entry
))
2256 oidset_insert(set
, &entry
.oid
);
2257 if (we_parsed_object
)
2258 free_tree_buffer(tree
);
2259 } else if (obj
->type
== OBJ_COMMIT
) {
2260 struct commit
*commit
= (struct commit
*) obj
;
2261 struct commit_list
*parents
= commit
->parents
;
2263 oidset_insert(set
, get_commit_tree_oid(commit
));
2264 for (; parents
; parents
= parents
->next
)
2265 oidset_insert(set
, &parents
->item
->object
.oid
);
2266 } else if (obj
->type
== OBJ_TAG
) {
2267 struct tag
*tag
= (struct tag
*) obj
;
2268 oidset_insert(set
, get_tagged_oid(tag
));
2273 int is_promisor_object(const struct object_id
*oid
)
2275 static struct oidset promisor_objects
;
2276 static int promisor_objects_prepared
;
2278 if (!promisor_objects_prepared
) {
2279 if (has_promisor_remote()) {
2280 for_each_packed_object(add_promisor_object
,
2282 FOR_EACH_OBJECT_PROMISOR_ONLY
|
2283 FOR_EACH_OBJECT_PACK_ORDER
);
2285 promisor_objects_prepared
= 1;
2287 return oidset_contains(&promisor_objects
, oid
);