4 #include "repository.h"
10 #include "hash-lookup.h"
14 #include "tree-walk.h"
16 #include "object-store.h"
18 #include "commit-graph.h"
19 #include "promisor-remote.h"
21 char *odb_pack_name(struct strbuf
*buf
,
22 const unsigned char *hash
,
26 strbuf_addf(buf
, "%s/pack/pack-%s.%s", get_object_directory(),
27 hash_to_hex(hash
), 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
!= st_add(4 * 256 + hashsz
+ hashsz
, st_mult(nr
, hashsz
+ 4)))
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 size_t min_size
= st_add(8 + 4*256 + hashsz
+ hashsz
, st_mult(nr
, hashsz
+ 4 + 4));
168 size_t max_size
= min_size
;
170 max_size
= st_add(max_size
, st_mult(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
);
181 p
->crc_offset
= 8 + 4 * 256 + nr
* hashsz
;
184 p
->index_version
= version
;
185 p
->index_data
= idx_map
;
186 p
->index_size
= idx_size
;
191 int open_pack_index(struct packed_git
*p
)
200 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
201 BUG("pack_name does not end in .pack");
202 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
203 ret
= check_packed_git_idx(idx_name
, p
);
208 uint32_t get_pack_fanout(struct packed_git
*p
, uint32_t value
)
210 const uint32_t *level1_ofs
= p
->index_data
;
213 if (open_pack_index(p
))
215 level1_ofs
= p
->index_data
;
218 if (p
->index_version
> 1) {
222 return ntohl(level1_ofs
[value
]);
225 static struct packed_git
*alloc_packed_git(int extra
)
227 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
228 memset(p
, 0, sizeof(*p
));
233 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
235 const char *path
= sha1_pack_name(sha1
);
236 size_t alloc
= st_add(strlen(path
), 1);
237 struct packed_git
*p
= alloc_packed_git(alloc
);
239 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
240 hashcpy(p
->hash
, sha1
);
241 if (check_packed_git_idx(idx_path
, p
)) {
249 static void scan_windows(struct packed_git
*p
,
250 struct packed_git
**lru_p
,
251 struct pack_window
**lru_w
,
252 struct pack_window
**lru_l
)
254 struct pack_window
*w
, *w_l
;
256 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
258 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
268 static int unuse_one_window(struct packed_git
*current
)
270 struct packed_git
*p
, *lru_p
= NULL
;
271 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
274 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
275 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
)
276 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
278 munmap(lru_w
->base
, lru_w
->len
);
279 pack_mapped
-= lru_w
->len
;
281 lru_l
->next
= lru_w
->next
;
283 lru_p
->windows
= lru_w
->next
;
291 void close_pack_windows(struct packed_git
*p
)
294 struct pack_window
*w
= p
->windows
;
297 die("pack '%s' still has open windows to it",
299 munmap(w
->base
, w
->len
);
300 pack_mapped
-= w
->len
;
302 p
->windows
= w
->next
;
307 int close_pack_fd(struct packed_git
*p
)
319 void close_pack_index(struct packed_git
*p
)
322 munmap((void *)p
->index_data
, p
->index_size
);
323 p
->index_data
= NULL
;
327 static void close_pack_revindex(struct packed_git
*p
)
329 if (!p
->revindex_map
)
332 munmap((void *)p
->revindex_map
, p
->revindex_size
);
333 p
->revindex_map
= NULL
;
334 p
->revindex_data
= NULL
;
337 void close_pack(struct packed_git
*p
)
339 close_pack_windows(p
);
342 close_pack_revindex(p
);
343 oidset_clear(&p
->bad_objects
);
346 void close_object_store(struct raw_object_store
*o
)
348 struct packed_git
*p
;
350 for (p
= o
->packed_git
; p
; p
= p
->next
)
352 BUG("want to close pack marked 'do-not-close'");
356 if (o
->multi_pack_index
) {
357 close_midx(o
->multi_pack_index
);
358 o
->multi_pack_index
= NULL
;
361 close_commit_graph(o
);
364 void unlink_pack_path(const char *pack_name
, int force_delete
)
366 static const char *exts
[] = {".pack", ".idx", ".rev", ".keep", ".bitmap", ".promisor"};
368 struct strbuf buf
= STRBUF_INIT
;
371 strbuf_addstr(&buf
, pack_name
);
372 strip_suffix_mem(buf
.buf
, &buf
.len
, ".pack");
376 strbuf_addstr(&buf
, ".keep");
377 if (!access(buf
.buf
, F_OK
)) {
378 strbuf_release(&buf
);
383 for (i
= 0; i
< ARRAY_SIZE(exts
); i
++) {
384 strbuf_setlen(&buf
, plen
);
385 strbuf_addstr(&buf
, exts
[i
]);
389 strbuf_release(&buf
);
393 * The LRU pack is the one with the oldest MRU window, preferring packs
394 * with no used windows, or the oldest mtime if it has no windows allocated.
396 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
398 struct pack_window
*w
, *this_mru_w
;
399 int has_windows_inuse
= 0;
402 * Reject this pack if it has windows and the previously selected
403 * one does not. If this pack does not have windows, reject
404 * it if the pack file is newer than the previously selected one.
406 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
409 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
411 * Reject this pack if any of its windows are in use,
412 * but the previously selected pack did not have any
413 * inuse windows. Otherwise, record that this pack
414 * has windows in use.
417 if (*accept_windows_inuse
)
418 has_windows_inuse
= 1;
423 if (w
->last_used
> this_mru_w
->last_used
)
427 * Reject this pack if it has windows that have been
428 * used more recently than the previously selected pack.
429 * If the previously selected pack had windows inuse and
430 * we have not encountered a window in this pack that is
431 * inuse, skip this check since we prefer a pack with no
432 * inuse windows to one that has inuse windows.
434 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
435 this_mru_w
->last_used
> (*mru_w
)->last_used
)
444 *accept_windows_inuse
= has_windows_inuse
;
447 static int close_one_pack(void)
449 struct packed_git
*p
, *lru_p
= NULL
;
450 struct pack_window
*mru_w
= NULL
;
451 int accept_windows_inuse
= 1;
453 for (p
= the_repository
->objects
->packed_git
; p
; p
= p
->next
) {
454 if (p
->pack_fd
== -1)
456 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
460 return close_pack_fd(lru_p
);
465 static unsigned int get_max_fd_limit(void)
471 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
478 long open_max
= sysconf(_SC_OPEN_MAX
);
482 * Otherwise, we got -1 for one of the two
485 * (1) sysconf() did not understand _SC_OPEN_MAX
486 * and signaled an error with -1; or
487 * (2) sysconf() said there is no limit.
489 * We _could_ clear errno before calling sysconf() to
490 * tell these two cases apart and return a huge number
491 * in the latter case to let the caller cap it to a
492 * value that is not so selfish, but letting the
493 * fallback OPEN_MAX codepath take care of these cases
502 return 1; /* see the caller ;-) */
506 const char *pack_basename(struct packed_git
*p
)
508 const char *ret
= strrchr(p
->pack_name
, '/');
510 ret
= ret
+ 1; /* skip past slash */
512 ret
= p
->pack_name
; /* we only have a base */
517 * Do not call this directly as this leaks p->pack_fd on error return;
518 * call open_packed_git() instead.
520 static int open_packed_git_1(struct packed_git
*p
)
523 struct pack_header hdr
;
524 unsigned char hash
[GIT_MAX_RAWSZ
];
525 unsigned char *idx_hash
;
527 const unsigned hashsz
= the_hash_algo
->rawsz
;
529 if (open_pack_index(p
))
530 return error("packfile %s index unavailable", p
->pack_name
);
533 unsigned int max_fds
= get_max_fd_limit();
535 /* Save 3 for stdin/stdout/stderr, 22 for work */
537 pack_max_fds
= max_fds
- 25;
542 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
545 p
->pack_fd
= git_open(p
->pack_name
);
546 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
550 /* If we created the struct before we had the pack we lack size. */
552 if (!S_ISREG(st
.st_mode
))
553 return error("packfile %s not a regular file", p
->pack_name
);
554 p
->pack_size
= st
.st_size
;
555 } else if (p
->pack_size
!= st
.st_size
)
556 return error("packfile %s size changed", p
->pack_name
);
558 /* Verify we recognize this pack file format. */
559 read_result
= read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
));
561 return error_errno("error reading from %s", p
->pack_name
);
562 if (read_result
!= sizeof(hdr
))
563 return error("file %s is far too short to be a packfile", p
->pack_name
);
564 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
565 return error("file %s is not a GIT packfile", p
->pack_name
);
566 if (!pack_version_ok(hdr
.hdr_version
))
567 return error("packfile %s is version %"PRIu32
" and not"
568 " supported (try upgrading GIT to a newer version)",
569 p
->pack_name
, ntohl(hdr
.hdr_version
));
571 /* Verify the pack matches its index. */
572 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
573 return error("packfile %s claims to have %"PRIu32
" objects"
574 " while index indicates %"PRIu32
" objects",
575 p
->pack_name
, ntohl(hdr
.hdr_entries
),
577 read_result
= pread_in_full(p
->pack_fd
, hash
, hashsz
,
578 p
->pack_size
- hashsz
);
580 return error_errno("error reading from %s", p
->pack_name
);
581 if (read_result
!= hashsz
)
582 return error("packfile %s signature is unavailable", p
->pack_name
);
583 idx_hash
= ((unsigned char *)p
->index_data
) + p
->index_size
- hashsz
* 2;
584 if (!hasheq(hash
, idx_hash
))
585 return error("packfile %s does not match index", p
->pack_name
);
589 static int open_packed_git(struct packed_git
*p
)
591 if (!open_packed_git_1(p
))
597 static int in_window(struct pack_window
*win
, off_t offset
)
599 /* We must promise at least one full hash after the
600 * offset is available from this window, otherwise the offset
601 * is not actually in this window and a different window (which
602 * has that one hash excess) must be used. This is to support
603 * the object header and delta base parsing routines below.
605 off_t win_off
= win
->offset
;
606 return win_off
<= offset
607 && (offset
+ the_hash_algo
->rawsz
) <= (win_off
+ win
->len
);
610 unsigned char *use_pack(struct packed_git
*p
,
611 struct pack_window
**w_cursor
,
615 struct pack_window
*win
= *w_cursor
;
617 /* Since packfiles end in a hash of their content and it's
618 * pointless to ask for an offset into the middle of that
619 * hash, and the in_window function above wouldn't match
620 * don't allow an offset too close to the end of the file.
622 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
623 die("packfile %s cannot be accessed", p
->pack_name
);
624 if (offset
> (p
->pack_size
- the_hash_algo
->rawsz
))
625 die("offset beyond end of packfile (truncated pack?)");
627 die(_("offset before end of packfile (broken .idx?)"));
629 if (!win
|| !in_window(win
, offset
)) {
632 for (win
= p
->windows
; win
; win
= win
->next
) {
633 if (in_window(win
, offset
))
637 size_t window_align
= packed_git_window_size
/ 2;
640 if (p
->pack_fd
== -1 && open_packed_git(p
))
641 die("packfile %s cannot be accessed", p
->pack_name
);
643 CALLOC_ARRAY(win
, 1);
644 win
->offset
= (offset
/ window_align
) * window_align
;
645 len
= p
->pack_size
- win
->offset
;
646 if (len
> packed_git_window_size
)
647 len
= packed_git_window_size
;
648 win
->len
= (size_t)len
;
649 pack_mapped
+= win
->len
;
650 while (packed_git_limit
< pack_mapped
651 && unuse_one_window(p
))
653 win
->base
= xmmap_gently(NULL
, win
->len
,
654 PROT_READ
, MAP_PRIVATE
,
655 p
->pack_fd
, win
->offset
);
656 if (win
->base
== MAP_FAILED
)
657 die_errno(_("packfile %s cannot be mapped%s"),
658 p
->pack_name
, mmap_os_err());
659 if (!win
->offset
&& win
->len
== p
->pack_size
664 if (pack_mapped
> peak_pack_mapped
)
665 peak_pack_mapped
= pack_mapped
;
666 if (pack_open_windows
> peak_pack_open_windows
)
667 peak_pack_open_windows
= pack_open_windows
;
668 win
->next
= p
->windows
;
672 if (win
!= *w_cursor
) {
673 win
->last_used
= pack_used_ctr
++;
677 offset
-= win
->offset
;
679 *left
= win
->len
- xsize_t(offset
);
680 return win
->base
+ offset
;
683 void unuse_pack(struct pack_window
**w_cursor
)
685 struct pack_window
*w
= *w_cursor
;
692 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
696 struct packed_git
*p
;
699 * Make sure a corresponding .pack file exists and that
700 * the index looks sane.
702 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
706 * ".promisor" is long enough to hold any suffix we're adding (and
707 * the use xsnprintf double-checks that)
709 alloc
= st_add3(path_len
, strlen(".promisor"), 1);
710 p
= alloc_packed_git(alloc
);
711 memcpy(p
->pack_name
, path
, path_len
);
713 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
714 if (!access(p
->pack_name
, F_OK
))
717 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".promisor");
718 if (!access(p
->pack_name
, F_OK
))
719 p
->pack_promisor
= 1;
721 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
722 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
727 /* ok, it looks sane as far as we can check without
728 * actually mapping the pack file.
730 p
->pack_size
= st
.st_size
;
731 p
->pack_local
= local
;
732 p
->mtime
= st
.st_mtime
;
733 if (path_len
< the_hash_algo
->hexsz
||
734 get_sha1_hex(path
+ path_len
- the_hash_algo
->hexsz
, p
->hash
))
739 void install_packed_git(struct repository
*r
, struct packed_git
*pack
)
741 if (pack
->pack_fd
!= -1)
744 pack
->next
= r
->objects
->packed_git
;
745 r
->objects
->packed_git
= pack
;
747 hashmap_entry_init(&pack
->packmap_ent
, strhash(pack
->pack_name
));
748 hashmap_add(&r
->objects
->pack_map
, &pack
->packmap_ent
);
751 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
753 static void report_helper(const struct string_list
*list
,
754 int seen_bits
, int first
, int last
)
756 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
759 for (; first
< last
; first
++)
760 report_garbage(seen_bits
, list
->items
[first
].string
);
763 static void report_pack_garbage(struct string_list
*list
)
765 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
770 string_list_sort(list
);
772 for (i
= 0; i
< list
->nr
; i
++) {
773 const char *path
= list
->items
[i
].string
;
775 strncmp(path
, list
->items
[first
].string
, baselen
)) {
776 report_helper(list
, seen_bits
, first
, i
);
781 const char *dot
= strrchr(path
, '.');
783 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
786 baselen
= dot
- path
+ 1;
789 if (!strcmp(path
+ baselen
, "pack"))
791 else if (!strcmp(path
+ baselen
, "idx"))
794 report_helper(list
, seen_bits
, first
, list
->nr
);
797 void for_each_file_in_pack_dir(const char *objdir
,
798 each_file_in_pack_dir_fn fn
,
801 struct strbuf path
= STRBUF_INIT
;
806 strbuf_addstr(&path
, objdir
);
807 strbuf_addstr(&path
, "/pack");
808 dir
= opendir(path
.buf
);
811 error_errno("unable to open object pack directory: %s",
813 strbuf_release(&path
);
816 strbuf_addch(&path
, '/');
817 dirnamelen
= path
.len
;
818 while ((de
= readdir_skip_dot_and_dotdot(dir
)) != NULL
) {
819 strbuf_setlen(&path
, dirnamelen
);
820 strbuf_addstr(&path
, de
->d_name
);
822 fn(path
.buf
, path
.len
, de
->d_name
, data
);
826 strbuf_release(&path
);
829 struct prepare_pack_data
{
830 struct repository
*r
;
831 struct string_list
*garbage
;
833 struct multi_pack_index
*m
;
836 static void prepare_pack(const char *full_name
, size_t full_name_len
,
837 const char *file_name
, void *_data
)
839 struct prepare_pack_data
*data
= (struct prepare_pack_data
*)_data
;
840 struct packed_git
*p
;
841 size_t base_len
= full_name_len
;
843 if (strip_suffix_mem(full_name
, &base_len
, ".idx") &&
844 !(data
->m
&& midx_contains_pack(data
->m
, file_name
))) {
845 struct hashmap_entry hent
;
846 char *pack_name
= xstrfmt("%.*s.pack", (int)base_len
, full_name
);
847 unsigned int hash
= strhash(pack_name
);
848 hashmap_entry_init(&hent
, hash
);
850 /* Don't reopen a pack we already have. */
851 if (!hashmap_get(&data
->r
->objects
->pack_map
, &hent
, pack_name
)) {
852 p
= add_packed_git(full_name
, full_name_len
, data
->local
);
854 install_packed_git(data
->r
, p
);
862 if (!strcmp(file_name
, "multi-pack-index"))
864 if (starts_with(file_name
, "multi-pack-index") &&
865 (ends_with(file_name
, ".bitmap") || ends_with(file_name
, ".rev")))
867 if (ends_with(file_name
, ".idx") ||
868 ends_with(file_name
, ".rev") ||
869 ends_with(file_name
, ".pack") ||
870 ends_with(file_name
, ".bitmap") ||
871 ends_with(file_name
, ".keep") ||
872 ends_with(file_name
, ".promisor"))
873 string_list_append(data
->garbage
, full_name
);
875 report_garbage(PACKDIR_FILE_GARBAGE
, full_name
);
878 static void prepare_packed_git_one(struct repository
*r
, char *objdir
, int local
)
880 struct prepare_pack_data data
;
881 struct string_list garbage
= STRING_LIST_INIT_DUP
;
883 data
.m
= r
->objects
->multi_pack_index
;
885 /* look for the multi-pack-index for this object directory */
886 while (data
.m
&& strcmp(data
.m
->object_dir
, objdir
))
887 data
.m
= data
.m
->next
;
890 data
.garbage
= &garbage
;
893 for_each_file_in_pack_dir(objdir
, prepare_pack
, &data
);
895 report_pack_garbage(data
.garbage
);
896 string_list_clear(data
.garbage
, 0);
899 static void prepare_packed_git(struct repository
*r
);
901 * Give a fast, rough count of the number of objects in the repository. This
902 * ignores loose objects completely. If you have a lot of them, then either
903 * you should repack because your performance will be awful, or they are
904 * all unreachable objects about to be pruned, in which case they're not really
905 * interesting as a measure of repo size in the first place.
907 unsigned long repo_approximate_object_count(struct repository
*r
)
909 if (!r
->objects
->approximate_object_count_valid
) {
911 struct multi_pack_index
*m
;
912 struct packed_git
*p
;
914 prepare_packed_git(r
);
916 for (m
= get_multi_pack_index(r
); m
; m
= m
->next
)
917 count
+= m
->num_objects
;
918 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
) {
919 if (open_pack_index(p
))
921 count
+= p
->num_objects
;
923 r
->objects
->approximate_object_count
= count
;
924 r
->objects
->approximate_object_count_valid
= 1;
926 return r
->objects
->approximate_object_count
;
929 DEFINE_LIST_SORT(static, sort_packs
, struct packed_git
, next
);
931 static int sort_pack(const struct packed_git
*a
, const struct packed_git
*b
)
936 * Local packs tend to contain objects specific to our
937 * variant of the project than remote ones. In addition,
938 * remote ones could be on a network mounted filesystem.
939 * Favor local ones for these reasons.
941 st
= a
->pack_local
- b
->pack_local
;
946 * Younger packs tend to contain more recent objects,
947 * and more recent objects tend to get accessed more
950 if (a
->mtime
< b
->mtime
)
952 else if (a
->mtime
== b
->mtime
)
957 static void rearrange_packed_git(struct repository
*r
)
959 sort_packs(&r
->objects
->packed_git
, sort_pack
);
962 static void prepare_packed_git_mru(struct repository
*r
)
964 struct packed_git
*p
;
966 INIT_LIST_HEAD(&r
->objects
->packed_git_mru
);
968 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
969 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
972 static void prepare_packed_git(struct repository
*r
)
974 struct object_directory
*odb
;
976 if (r
->objects
->packed_git_initialized
)
980 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
) {
981 int local
= (odb
== r
->objects
->odb
);
982 prepare_multi_pack_index_one(r
, odb
->path
, local
);
983 prepare_packed_git_one(r
, odb
->path
, local
);
985 rearrange_packed_git(r
);
987 prepare_packed_git_mru(r
);
988 r
->objects
->packed_git_initialized
= 1;
991 void reprepare_packed_git(struct repository
*r
)
993 struct object_directory
*odb
;
996 for (odb
= r
->objects
->odb
; odb
; odb
= odb
->next
)
997 odb_clear_loose_cache(odb
);
999 r
->objects
->approximate_object_count_valid
= 0;
1000 r
->objects
->packed_git_initialized
= 0;
1001 prepare_packed_git(r
);
1005 struct packed_git
*get_packed_git(struct repository
*r
)
1007 prepare_packed_git(r
);
1008 return r
->objects
->packed_git
;
1011 struct multi_pack_index
*get_multi_pack_index(struct repository
*r
)
1013 prepare_packed_git(r
);
1014 return r
->objects
->multi_pack_index
;
1017 struct multi_pack_index
*get_local_multi_pack_index(struct repository
*r
)
1019 struct multi_pack_index
*m
= get_multi_pack_index(r
);
1021 /* no need to iterate; we always put the local one first (if any) */
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
)
1053 unsigned long used
= 0;
1056 *type
= (c
>> 4) & 7;
1060 if (len
<= used
|| (bitsizeof(long) - 7) < shift
) {
1061 error("bad object header");
1066 size
= st_add(size
, st_left_shift(c
& 0x7f, shift
));
1069 *sizep
= cast_size_t_to_ulong(size
);
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
;
1091 * Note: the window section returned by use_pack() must be
1092 * available throughout git_inflate()'s unlocked execution. To
1093 * ensure no other thread will modify the window in the
1094 * meantime, we rely on the packed_window.inuse_cnt. This
1095 * counter is incremented before window reading and checked
1096 * before window disposal.
1098 * Other worrying sections could be the call to close_pack_fd(),
1099 * which can close packs even with in-use windows, and to
1100 * reprepare_packed_git(). Regarding the former, mmap doc says:
1101 * "closing the file descriptor does not unmap the region". And
1102 * for the latter, it won't re-open already available packs.
1105 st
= git_inflate(&stream
, Z_FINISH
);
1107 curpos
+= stream
.next_in
- in
;
1108 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1109 stream
.total_out
< sizeof(delta_head
));
1110 git_inflate_end(&stream
);
1111 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1112 error("delta data unpack-initial failed");
1116 /* Examine the initial part of the delta to figure out
1121 /* ignore base size */
1122 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1124 /* Read the result size */
1125 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1128 int unpack_object_header(struct packed_git
*p
,
1129 struct pack_window
**w_curs
,
1131 unsigned long *sizep
)
1133 unsigned char *base
;
1136 enum object_type type
;
1138 /* use_pack() assures us we have [base, base + 20) available
1139 * as a range that we can look at. (Its actually the hash
1140 * size that is assured.) With our object header encoding
1141 * the maximum deflated object size is 2^137, which is just
1142 * insane, so we know won't exceed what we have been given.
1144 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1145 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1154 void mark_bad_packed_object(struct packed_git
*p
, const struct object_id
*oid
)
1156 oidset_insert(&p
->bad_objects
, oid
);
1159 const struct packed_git
*has_packed_and_bad(struct repository
*r
,
1160 const struct object_id
*oid
)
1162 struct packed_git
*p
;
1164 for (p
= r
->objects
->packed_git
; p
; p
= p
->next
)
1165 if (oidset_contains(&p
->bad_objects
, oid
))
1170 off_t
get_delta_base(struct packed_git
*p
,
1171 struct pack_window
**w_curs
,
1173 enum object_type type
,
1174 off_t delta_obj_offset
)
1176 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1179 /* use_pack() assured us we have [base_info, base_info + 20)
1180 * as a range that we can look at without walking off the
1181 * end of the mapped window. Its actually the hash size
1182 * that is assured. An OFS_DELTA longer than the hash size
1183 * is stupid, as then a REF_DELTA would be smaller to store.
1185 if (type
== OBJ_OFS_DELTA
) {
1187 unsigned char c
= base_info
[used
++];
1188 base_offset
= c
& 127;
1191 if (!base_offset
|| MSB(base_offset
, 7))
1192 return 0; /* overflow */
1193 c
= base_info
[used
++];
1194 base_offset
= (base_offset
<< 7) + (c
& 127);
1196 base_offset
= delta_obj_offset
- base_offset
;
1197 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1198 return 0; /* out of bound */
1200 } else if (type
== OBJ_REF_DELTA
) {
1201 /* The base entry _must_ be in the same pack */
1202 base_offset
= find_pack_entry_one(base_info
, p
);
1203 *curpos
+= the_hash_algo
->rawsz
;
1205 die("I am totally screwed");
1210 * Like get_delta_base above, but we return the sha1 instead of the pack
1211 * offset. This means it is cheaper for REF deltas (we do not have to do
1212 * the final object lookup), but more expensive for OFS deltas (we
1213 * have to load the revidx to convert the offset back into a sha1).
1215 static int get_delta_base_oid(struct packed_git
*p
,
1216 struct pack_window
**w_curs
,
1218 struct object_id
*oid
,
1219 enum object_type type
,
1220 off_t delta_obj_offset
)
1222 if (type
== OBJ_REF_DELTA
) {
1223 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1226 } else if (type
== OBJ_OFS_DELTA
) {
1228 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1229 type
, delta_obj_offset
);
1234 if (offset_to_pack_pos(p
, base_offset
, &base_pos
) < 0)
1237 return nth_packed_object_id(oid
, p
,
1238 pack_pos_to_index(p
, base_pos
));
1243 static int retry_bad_packed_offset(struct repository
*r
,
1244 struct packed_git
*p
,
1249 struct object_id oid
;
1250 if (offset_to_pack_pos(p
, obj_offset
, &pos
) < 0)
1252 nth_packed_object_id(&oid
, p
, pack_pos_to_index(p
, pos
));
1253 mark_bad_packed_object(p
, &oid
);
1254 type
= oid_object_info(r
, &oid
, NULL
);
1255 if (type
<= OBJ_NONE
)
1260 #define POI_STACK_PREALLOC 64
1262 static enum object_type
packed_to_object_type(struct repository
*r
,
1263 struct packed_git
*p
,
1265 enum object_type type
,
1266 struct pack_window
**w_curs
,
1269 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1270 off_t
*poi_stack
= small_poi_stack
;
1271 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1273 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1276 /* Push the object we're going to leave behind */
1277 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1278 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1279 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
1280 COPY_ARRAY(poi_stack
, small_poi_stack
, poi_stack_nr
);
1282 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1284 poi_stack
[poi_stack_nr
++] = obj_offset
;
1285 /* If parsing the base offset fails, just unwind */
1286 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1289 curpos
= obj_offset
= base_offset
;
1290 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1291 if (type
<= OBJ_NONE
) {
1292 /* If getting the base itself fails, we first
1293 * retry the base, otherwise unwind */
1294 type
= retry_bad_packed_offset(r
, p
, base_offset
);
1295 if (type
> OBJ_NONE
)
1309 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1310 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1315 if (poi_stack
!= small_poi_stack
)
1320 while (poi_stack_nr
) {
1321 obj_offset
= poi_stack
[--poi_stack_nr
];
1322 type
= retry_bad_packed_offset(r
, p
, obj_offset
);
1323 if (type
> OBJ_NONE
)
1330 static struct hashmap delta_base_cache
;
1331 static size_t delta_base_cached
;
1333 static LIST_HEAD(delta_base_cache_lru
);
1335 struct delta_base_cache_key
{
1336 struct packed_git
*p
;
1340 struct delta_base_cache_entry
{
1341 struct hashmap_entry ent
;
1342 struct delta_base_cache_key key
;
1343 struct list_head lru
;
1346 enum object_type type
;
1349 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1353 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
1354 hash
+= (hash
>> 8) + (hash
>> 16);
1358 static struct delta_base_cache_entry
*
1359 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1361 struct hashmap_entry entry
, *e
;
1362 struct delta_base_cache_key key
;
1364 if (!delta_base_cache
.cmpfn
)
1367 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
1369 key
.base_offset
= base_offset
;
1370 e
= hashmap_get(&delta_base_cache
, &entry
, &key
);
1371 return e
? container_of(e
, struct delta_base_cache_entry
, ent
) : NULL
;
1374 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
1375 const struct delta_base_cache_key
*b
)
1377 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
1380 static int delta_base_cache_hash_cmp(const void *unused_cmp_data
,
1381 const struct hashmap_entry
*va
,
1382 const struct hashmap_entry
*vb
,
1385 const struct delta_base_cache_entry
*a
, *b
;
1386 const struct delta_base_cache_key
*key
= vkey
;
1388 a
= container_of(va
, const struct delta_base_cache_entry
, ent
);
1389 b
= container_of(vb
, const struct delta_base_cache_entry
, ent
);
1392 return !delta_base_cache_key_eq(&a
->key
, key
);
1394 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
1397 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1399 return !!get_delta_base_cache_entry(p
, base_offset
);
1403 * Remove the entry from the cache, but do _not_ free the associated
1404 * entry data. The caller takes ownership of the "data" buffer, and
1405 * should copy out any fields it wants before detaching.
1407 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1409 hashmap_remove(&delta_base_cache
, &ent
->ent
, &ent
->key
);
1410 list_del(&ent
->lru
);
1411 delta_base_cached
-= ent
->size
;
1415 static void *cache_or_unpack_entry(struct repository
*r
, struct packed_git
*p
,
1416 off_t base_offset
, unsigned long *base_size
,
1417 enum object_type
*type
)
1419 struct delta_base_cache_entry
*ent
;
1421 ent
= get_delta_base_cache_entry(p
, base_offset
);
1423 return unpack_entry(r
, p
, base_offset
, type
, base_size
);
1428 *base_size
= ent
->size
;
1429 return xmemdupz(ent
->data
, ent
->size
);
1432 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1435 detach_delta_base_cache_entry(ent
);
1438 void clear_delta_base_cache(void)
1440 struct list_head
*lru
, *tmp
;
1441 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1442 struct delta_base_cache_entry
*entry
=
1443 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1444 release_delta_base_cache(entry
);
1448 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1449 void *base
, unsigned long base_size
, enum object_type type
)
1451 struct delta_base_cache_entry
*ent
;
1452 struct list_head
*lru
, *tmp
;
1455 * Check required to avoid redundant entries when more than one thread
1456 * is unpacking the same object, in unpack_entry() (since its phases I
1457 * and III might run concurrently across multiple threads).
1459 if (in_delta_base_cache(p
, base_offset
)) {
1464 delta_base_cached
+= base_size
;
1466 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
1467 struct delta_base_cache_entry
*f
=
1468 list_entry(lru
, struct delta_base_cache_entry
, lru
);
1469 if (delta_base_cached
<= delta_base_cache_limit
)
1471 release_delta_base_cache(f
);
1474 ent
= xmalloc(sizeof(*ent
));
1476 ent
->key
.base_offset
= base_offset
;
1479 ent
->size
= base_size
;
1480 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
1482 if (!delta_base_cache
.cmpfn
)
1483 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, NULL
, 0);
1484 hashmap_entry_init(&ent
->ent
, pack_entry_hash(p
, base_offset
));
1485 hashmap_add(&delta_base_cache
, &ent
->ent
);
1488 int packed_object_info(struct repository
*r
, struct packed_git
*p
,
1489 off_t obj_offset
, struct object_info
*oi
)
1491 struct pack_window
*w_curs
= NULL
;
1493 off_t curpos
= obj_offset
;
1494 enum object_type type
;
1497 * We always get the representation type, but only convert it to
1498 * a "real" type later if the caller is interested.
1501 *oi
->contentp
= cache_or_unpack_entry(r
, p
, obj_offset
, oi
->sizep
,
1506 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1509 if (!oi
->contentp
&& oi
->sizep
) {
1510 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1511 off_t tmp_pos
= curpos
;
1512 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1518 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1519 if (*oi
->sizep
== 0) {
1528 if (oi
->disk_sizep
) {
1530 if (offset_to_pack_pos(p
, obj_offset
, &pos
) < 0) {
1531 error("could not find object at offset %"PRIuMAX
" "
1532 "in pack %s", (uintmax_t)obj_offset
, p
->pack_name
);
1537 *oi
->disk_sizep
= pack_pos_to_offset(p
, pos
+ 1) - obj_offset
;
1540 if (oi
->typep
|| oi
->type_name
) {
1541 enum object_type ptot
;
1542 ptot
= packed_to_object_type(r
, p
, obj_offset
,
1543 type
, &w_curs
, curpos
);
1546 if (oi
->type_name
) {
1547 const char *tn
= type_name(ptot
);
1549 strbuf_addstr(oi
->type_name
, tn
);
1557 if (oi
->delta_base_oid
) {
1558 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1559 if (get_delta_base_oid(p
, &w_curs
, curpos
,
1561 type
, obj_offset
) < 0) {
1566 oidclr(oi
->delta_base_oid
);
1569 oi
->whence
= in_delta_base_cache(p
, obj_offset
) ? OI_DBCACHED
:
1573 unuse_pack(&w_curs
);
1577 static void *unpack_compressed_entry(struct packed_git
*p
,
1578 struct pack_window
**w_curs
,
1584 unsigned char *buffer
, *in
;
1586 buffer
= xmallocz_gently(size
);
1589 memset(&stream
, 0, sizeof(stream
));
1590 stream
.next_out
= buffer
;
1591 stream
.avail_out
= size
+ 1;
1593 git_inflate_init(&stream
);
1595 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1596 stream
.next_in
= in
;
1598 * Note: we must ensure the window section returned by
1599 * use_pack() will be available throughout git_inflate()'s
1600 * unlocked execution. Please refer to the comment at
1601 * get_size_from_delta() to see how this is done.
1604 st
= git_inflate(&stream
, Z_FINISH
);
1606 if (!stream
.avail_out
)
1607 break; /* the payload is larger than it should be */
1608 curpos
+= stream
.next_in
- in
;
1609 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1610 git_inflate_end(&stream
);
1611 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1616 /* versions of zlib can clobber unconsumed portion of outbuf */
1617 buffer
[size
] = '\0';
1622 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1624 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
1625 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
1626 p
->pack_name
, (uintmax_t)obj_offset
);
1629 int do_check_packed_object_crc
;
1631 #define UNPACK_ENTRY_STACK_PREALLOC 64
1632 struct unpack_entry_stack_ent
{
1638 static void *read_object(struct repository
*r
,
1639 const struct object_id
*oid
,
1640 enum object_type
*type
,
1641 unsigned long *size
)
1643 struct object_info oi
= OBJECT_INFO_INIT
;
1647 oi
.contentp
= &content
;
1649 if (oid_object_info_extended(r
, oid
, &oi
, 0) < 0)
1654 void *unpack_entry(struct repository
*r
, struct packed_git
*p
, off_t obj_offset
,
1655 enum object_type
*final_type
, unsigned long *final_size
)
1657 struct pack_window
*w_curs
= NULL
;
1658 off_t curpos
= obj_offset
;
1661 enum object_type type
;
1662 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
1663 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
1664 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
1665 int base_from_cache
= 0;
1667 write_pack_access_log(p
, obj_offset
);
1669 /* PHASE 1: drill down to the innermost base object */
1673 struct delta_base_cache_entry
*ent
;
1675 ent
= get_delta_base_cache_entry(p
, curpos
);
1680 detach_delta_base_cache_entry(ent
);
1681 base_from_cache
= 1;
1685 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1686 uint32_t pack_pos
, index_pos
;
1689 if (offset_to_pack_pos(p
, obj_offset
, &pack_pos
) < 0) {
1690 error("could not find object at offset %"PRIuMAX
" in pack %s",
1691 (uintmax_t)obj_offset
, p
->pack_name
);
1696 len
= pack_pos_to_offset(p
, pack_pos
+ 1) - obj_offset
;
1697 index_pos
= pack_pos_to_index(p
, pack_pos
);
1698 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, index_pos
)) {
1699 struct object_id oid
;
1700 nth_packed_object_id(&oid
, p
, index_pos
);
1701 error("bad packed object CRC for %s",
1703 mark_bad_packed_object(p
, &oid
);
1709 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1710 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
1713 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1715 error("failed to validate delta base reference "
1716 "at offset %"PRIuMAX
" from %s",
1717 (uintmax_t)curpos
, p
->pack_name
);
1718 /* bail to phase 2, in hopes of recovery */
1723 /* push object, proceed to base */
1724 if (delta_stack_nr
>= delta_stack_alloc
1725 && delta_stack
== small_delta_stack
) {
1726 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
1727 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
1728 COPY_ARRAY(delta_stack
, small_delta_stack
,
1731 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
1733 i
= delta_stack_nr
++;
1734 delta_stack
[i
].obj_offset
= obj_offset
;
1735 delta_stack
[i
].curpos
= curpos
;
1736 delta_stack
[i
].size
= size
;
1738 curpos
= obj_offset
= base_offset
;
1741 /* PHASE 2: handle the base */
1746 BUG("unpack_entry: left loop at a valid delta");
1752 if (!base_from_cache
)
1753 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
1757 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1758 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1761 /* PHASE 3: apply deltas in order */
1764 * 'data' holds the base data, or NULL if there was corruption
1766 while (delta_stack_nr
) {
1769 void *external_base
= NULL
;
1770 unsigned long delta_size
, base_size
= size
;
1772 off_t base_obj_offset
= obj_offset
;
1778 * We're probably in deep shit, but let's try to fetch
1779 * the required base anyway from another pack or loose.
1780 * This is costly but should happen only in the presence
1781 * of a corrupted pack, and is better than failing outright.
1784 struct object_id base_oid
;
1785 if (!(offset_to_pack_pos(p
, obj_offset
, &pos
))) {
1786 nth_packed_object_id(&base_oid
, p
,
1787 pack_pos_to_index(p
, pos
));
1788 error("failed to read delta base object %s"
1789 " at offset %"PRIuMAX
" from %s",
1790 oid_to_hex(&base_oid
), (uintmax_t)obj_offset
,
1792 mark_bad_packed_object(p
, &base_oid
);
1793 base
= read_object(r
, &base_oid
, &type
, &base_size
);
1794 external_base
= base
;
1798 i
= --delta_stack_nr
;
1799 obj_offset
= delta_stack
[i
].obj_offset
;
1800 curpos
= delta_stack
[i
].curpos
;
1801 delta_size
= delta_stack
[i
].size
;
1806 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
1809 error("failed to unpack compressed delta "
1810 "at offset %"PRIuMAX
" from %s",
1811 (uintmax_t)curpos
, p
->pack_name
);
1814 data
= patch_delta(base
, base_size
, delta_data
,
1818 * We could not apply the delta; warn the user, but
1819 * keep going. Our failure will be noticed either in
1820 * the next iteration of the loop, or if this is the
1821 * final delta, in the caller when we return NULL.
1822 * Those code paths will take care of making a more
1823 * explicit warning and retrying with another copy of
1827 error("failed to apply delta");
1831 * We delay adding `base` to the cache until the end of the loop
1832 * because unpack_compressed_entry() momentarily releases the
1833 * obj_read_mutex, giving another thread the chance to access
1834 * the cache. Therefore, if `base` was already there, this other
1835 * thread could free() it (e.g. to make space for another entry)
1836 * before we are done using it.
1839 add_delta_base_cache(p
, base_obj_offset
, base
, base_size
, type
);
1842 free(external_base
);
1851 unuse_pack(&w_curs
);
1853 if (delta_stack
!= small_delta_stack
)
1859 int bsearch_pack(const struct object_id
*oid
, const struct packed_git
*p
, uint32_t *result
)
1861 const unsigned char *index_fanout
= p
->index_data
;
1862 const unsigned char *index_lookup
;
1863 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1864 int index_lookup_width
;
1867 BUG("bsearch_pack called without a valid pack-index");
1869 index_lookup
= index_fanout
+ 4 * 256;
1870 if (p
->index_version
== 1) {
1871 index_lookup_width
= hashsz
+ 4;
1874 index_lookup_width
= hashsz
;
1879 return bsearch_hash(oid
->hash
, (const uint32_t*)index_fanout
,
1880 index_lookup
, index_lookup_width
, result
);
1883 int nth_packed_object_id(struct object_id
*oid
,
1884 struct packed_git
*p
,
1887 const unsigned char *index
= p
->index_data
;
1888 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1890 if (open_pack_index(p
))
1892 index
= p
->index_data
;
1894 if (n
>= p
->num_objects
)
1897 if (p
->index_version
== 1) {
1898 oidread(oid
, index
+ (hashsz
+ 4) * n
+ 4);
1901 oidread(oid
, index
+ hashsz
* n
);
1906 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
1908 const unsigned char *ptr
= vptr
;
1909 const unsigned char *start
= p
->index_data
;
1910 const unsigned char *end
= start
+ p
->index_size
;
1912 die(_("offset before start of pack index for %s (corrupt index?)"),
1914 /* No need to check for underflow; .idx files must be at least 8 bytes */
1916 die(_("offset beyond end of pack index for %s (truncated index?)"),
1920 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1922 const unsigned char *index
= p
->index_data
;
1923 const unsigned int hashsz
= the_hash_algo
->rawsz
;
1925 if (p
->index_version
== 1) {
1926 return ntohl(*((uint32_t *)(index
+ (hashsz
+ 4) * (size_t)n
)));
1929 index
+= 8 + (size_t)p
->num_objects
* (hashsz
+ 4);
1930 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1931 if (!(off
& 0x80000000))
1933 index
+= (size_t)p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1934 check_pack_index_ptr(p
, index
);
1935 return get_be64(index
);
1939 off_t
find_pack_entry_one(const unsigned char *sha1
,
1940 struct packed_git
*p
)
1942 const unsigned char *index
= p
->index_data
;
1943 struct object_id oid
;
1947 if (open_pack_index(p
))
1951 hashcpy(oid
.hash
, sha1
);
1952 if (bsearch_pack(&oid
, p
, &result
))
1953 return nth_packed_object_offset(p
, result
);
1957 int is_pack_valid(struct packed_git
*p
)
1959 /* An already open pack is known to be valid. */
1960 if (p
->pack_fd
!= -1)
1963 /* If the pack has one window completely covering the
1964 * file size, the pack is known to be valid even if
1965 * the descriptor is not currently open.
1968 struct pack_window
*w
= p
->windows
;
1970 if (!w
->offset
&& w
->len
== p
->pack_size
)
1974 /* Force the pack to open to prove its valid. */
1975 return !open_packed_git(p
);
1978 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1979 struct packed_git
*packs
)
1981 struct packed_git
*p
;
1983 for (p
= packs
; p
; p
= p
->next
) {
1984 if (find_pack_entry_one(sha1
, p
))
1991 static int fill_pack_entry(const struct object_id
*oid
,
1992 struct pack_entry
*e
,
1993 struct packed_git
*p
)
1997 if (oidset_size(&p
->bad_objects
) &&
1998 oidset_contains(&p
->bad_objects
, oid
))
2001 offset
= find_pack_entry_one(oid
->hash
, p
);
2006 * We are about to tell the caller where they can locate the
2007 * requested object. We better make sure the packfile is
2008 * still here and can be accessed before supplying that
2009 * answer, as it may have been deleted since the index was
2012 if (!is_pack_valid(p
))
2019 int find_pack_entry(struct repository
*r
, const struct object_id
*oid
, struct pack_entry
*e
)
2021 struct list_head
*pos
;
2022 struct multi_pack_index
*m
;
2024 prepare_packed_git(r
);
2025 if (!r
->objects
->packed_git
&& !r
->objects
->multi_pack_index
)
2028 for (m
= r
->objects
->multi_pack_index
; m
; m
= m
->next
) {
2029 if (fill_midx_entry(r
, oid
, e
, m
))
2033 list_for_each(pos
, &r
->objects
->packed_git_mru
) {
2034 struct packed_git
*p
= list_entry(pos
, struct packed_git
, mru
);
2035 if (!p
->multi_pack_index
&& fill_pack_entry(oid
, e
, p
)) {
2036 list_move(&p
->mru
, &r
->objects
->packed_git_mru
);
2043 static void maybe_invalidate_kept_pack_cache(struct repository
*r
,
2046 if (!r
->objects
->kept_pack_cache
.packs
)
2048 if (r
->objects
->kept_pack_cache
.flags
== flags
)
2050 FREE_AND_NULL(r
->objects
->kept_pack_cache
.packs
);
2051 r
->objects
->kept_pack_cache
.flags
= 0;
2054 static struct packed_git
**kept_pack_cache(struct repository
*r
, unsigned flags
)
2056 maybe_invalidate_kept_pack_cache(r
, flags
);
2058 if (!r
->objects
->kept_pack_cache
.packs
) {
2059 struct packed_git
**packs
= NULL
;
2060 size_t nr
= 0, alloc
= 0;
2061 struct packed_git
*p
;
2064 * We want "all" packs here, because we need to cover ones that
2065 * are used by a midx, as well. We need to look in every one of
2066 * them (instead of the midx itself) to cover duplicates. It's
2067 * possible that an object is found in two packs that the midx
2068 * covers, one kept and one not kept, but the midx returns only
2069 * the non-kept version.
2071 for (p
= get_all_packs(r
); p
; p
= p
->next
) {
2072 if ((p
->pack_keep
&& (flags
& ON_DISK_KEEP_PACKS
)) ||
2073 (p
->pack_keep_in_core
&& (flags
& IN_CORE_KEEP_PACKS
))) {
2074 ALLOC_GROW(packs
, nr
+ 1, alloc
);
2078 ALLOC_GROW(packs
, nr
+ 1, alloc
);
2081 r
->objects
->kept_pack_cache
.packs
= packs
;
2082 r
->objects
->kept_pack_cache
.flags
= flags
;
2085 return r
->objects
->kept_pack_cache
.packs
;
2088 int find_kept_pack_entry(struct repository
*r
,
2089 const struct object_id
*oid
,
2091 struct pack_entry
*e
)
2093 struct packed_git
**cache
;
2095 for (cache
= kept_pack_cache(r
, flags
); *cache
; cache
++) {
2096 struct packed_git
*p
= *cache
;
2097 if (fill_pack_entry(oid
, e
, p
))
2104 int has_object_pack(const struct object_id
*oid
)
2106 struct pack_entry e
;
2107 return find_pack_entry(the_repository
, oid
, &e
);
2110 int has_object_kept_pack(const struct object_id
*oid
, unsigned flags
)
2112 struct pack_entry e
;
2113 return find_kept_pack_entry(the_repository
, oid
, flags
, &e
);
2116 int has_pack_index(const unsigned char *sha1
)
2119 if (stat(sha1_pack_index_name(sha1
), &st
))
2124 int for_each_object_in_pack(struct packed_git
*p
,
2125 each_packed_object_fn cb
, void *data
,
2126 enum for_each_object_flags flags
)
2131 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
) {
2132 if (load_pack_revindex(p
))
2136 for (i
= 0; i
< p
->num_objects
; i
++) {
2138 struct object_id oid
;
2141 * We are iterating "i" from 0 up to num_objects, but its
2142 * meaning may be different, depending on the requested output
2145 * - in object-name order, it is the same as the index order
2146 * used by nth_packed_object_id(), so we can pass it
2149 * - in pack-order, it is pack position, which we must
2150 * convert to an index position in order to get the oid.
2152 if (flags
& FOR_EACH_OBJECT_PACK_ORDER
)
2153 index_pos
= pack_pos_to_index(p
, i
);
2157 if (nth_packed_object_id(&oid
, p
, index_pos
) < 0)
2158 return error("unable to get sha1 of object %u in %s",
2159 index_pos
, p
->pack_name
);
2161 r
= cb(&oid
, p
, index_pos
, data
);
2168 int for_each_packed_object(each_packed_object_fn cb
, void *data
,
2169 enum for_each_object_flags flags
)
2171 struct packed_git
*p
;
2173 int pack_errors
= 0;
2175 prepare_packed_git(the_repository
);
2176 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
2177 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
2179 if ((flags
& FOR_EACH_OBJECT_PROMISOR_ONLY
) &&
2182 if ((flags
& FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS
) &&
2183 p
->pack_keep_in_core
)
2185 if ((flags
& FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS
) &&
2188 if (open_pack_index(p
)) {
2192 r
= for_each_object_in_pack(p
, cb
, data
, flags
);
2196 return r
? r
: pack_errors
;
2199 static int add_promisor_object(const struct object_id
*oid
,
2200 struct packed_git
*pack
,
2204 struct oidset
*set
= set_
;
2205 struct object
*obj
= parse_object(the_repository
, oid
);
2209 oidset_insert(set
, oid
);
2212 * If this is a tree, commit, or tag, the objects it refers
2213 * to are also promisor objects. (Blobs refer to no objects->)
2215 if (obj
->type
== OBJ_TREE
) {
2216 struct tree
*tree
= (struct tree
*)obj
;
2217 struct tree_desc desc
;
2218 struct name_entry entry
;
2219 if (init_tree_desc_gently(&desc
, tree
->buffer
, tree
->size
))
2221 * Error messages are given when packs are
2222 * verified, so do not print any here.
2225 while (tree_entry_gently(&desc
, &entry
))
2226 oidset_insert(set
, &entry
.oid
);
2227 free_tree_buffer(tree
);
2228 } else if (obj
->type
== OBJ_COMMIT
) {
2229 struct commit
*commit
= (struct commit
*) obj
;
2230 struct commit_list
*parents
= commit
->parents
;
2232 oidset_insert(set
, get_commit_tree_oid(commit
));
2233 for (; parents
; parents
= parents
->next
)
2234 oidset_insert(set
, &parents
->item
->object
.oid
);
2235 } else if (obj
->type
== OBJ_TAG
) {
2236 struct tag
*tag
= (struct tag
*) obj
;
2237 oidset_insert(set
, get_tagged_oid(tag
));
2242 int is_promisor_object(const struct object_id
*oid
)
2244 static struct oidset promisor_objects
;
2245 static int promisor_objects_prepared
;
2247 if (!promisor_objects_prepared
) {
2248 if (has_promisor_remote()) {
2249 for_each_packed_object(add_promisor_object
,
2251 FOR_EACH_OBJECT_PROMISOR_ONLY
);
2253 promisor_objects_prepared
= 1;
2255 return oidset_contains(&promisor_objects
, oid
);