7 #include "object-store.h"
8 #include "sha1-lookup.h"
12 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
13 #define MIDX_VERSION 1
14 #define MIDX_BYTE_FILE_VERSION 4
15 #define MIDX_BYTE_HASH_VERSION 5
16 #define MIDX_BYTE_NUM_CHUNKS 6
17 #define MIDX_BYTE_NUM_PACKS 8
18 #define MIDX_HASH_VERSION 1
19 #define MIDX_HEADER_SIZE 12
20 #define MIDX_HASH_LEN 20
21 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + MIDX_HASH_LEN)
23 #define MIDX_MAX_CHUNKS 5
24 #define MIDX_CHUNK_ALIGNMENT 4
25 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
26 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
27 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
28 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
29 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
30 #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
31 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
32 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
33 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
34 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
36 static char *get_midx_filename(const char *object_dir
)
38 return xstrfmt("%s/pack/multi-pack-index", object_dir
);
41 struct multi_pack_index
*load_multi_pack_index(const char *object_dir
, int local
)
43 struct multi_pack_index
*m
= NULL
;
47 void *midx_map
= NULL
;
48 uint32_t hash_version
;
49 char *midx_name
= get_midx_filename(object_dir
);
51 const char *cur_pack_name
;
53 fd
= git_open(midx_name
);
58 error_errno(_("failed to read %s"), midx_name
);
62 midx_size
= xsize_t(st
.st_size
);
64 if (midx_size
< MIDX_MIN_SIZE
) {
65 error(_("multi-pack-index file %s is too small"), midx_name
);
69 FREE_AND_NULL(midx_name
);
71 midx_map
= xmmap(NULL
, midx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
73 FLEX_ALLOC_MEM(m
, object_dir
, object_dir
, strlen(object_dir
));
76 m
->data_len
= midx_size
;
79 m
->signature
= get_be32(m
->data
);
80 if (m
->signature
!= MIDX_SIGNATURE
)
81 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
82 m
->signature
, MIDX_SIGNATURE
);
84 m
->version
= m
->data
[MIDX_BYTE_FILE_VERSION
];
85 if (m
->version
!= MIDX_VERSION
)
86 die(_("multi-pack-index version %d not recognized"),
89 hash_version
= m
->data
[MIDX_BYTE_HASH_VERSION
];
90 if (hash_version
!= MIDX_HASH_VERSION
)
91 die(_("hash version %u does not match"), hash_version
);
92 m
->hash_len
= MIDX_HASH_LEN
;
94 m
->num_chunks
= m
->data
[MIDX_BYTE_NUM_CHUNKS
];
96 m
->num_packs
= get_be32(m
->data
+ MIDX_BYTE_NUM_PACKS
);
98 for (i
= 0; i
< m
->num_chunks
; i
++) {
99 uint32_t chunk_id
= get_be32(m
->data
+ MIDX_HEADER_SIZE
+
100 MIDX_CHUNKLOOKUP_WIDTH
* i
);
101 uint64_t chunk_offset
= get_be64(m
->data
+ MIDX_HEADER_SIZE
+ 4 +
102 MIDX_CHUNKLOOKUP_WIDTH
* i
);
104 if (chunk_offset
>= m
->data_len
)
105 die(_("invalid chunk offset (too large)"));
108 case MIDX_CHUNKID_PACKNAMES
:
109 m
->chunk_pack_names
= m
->data
+ chunk_offset
;
112 case MIDX_CHUNKID_OIDFANOUT
:
113 m
->chunk_oid_fanout
= (uint32_t *)(m
->data
+ chunk_offset
);
116 case MIDX_CHUNKID_OIDLOOKUP
:
117 m
->chunk_oid_lookup
= m
->data
+ chunk_offset
;
120 case MIDX_CHUNKID_OBJECTOFFSETS
:
121 m
->chunk_object_offsets
= m
->data
+ chunk_offset
;
124 case MIDX_CHUNKID_LARGEOFFSETS
:
125 m
->chunk_large_offsets
= m
->data
+ chunk_offset
;
129 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
134 * Do nothing on unrecognized chunks, allowing future
135 * extensions to add optional chunks.
141 if (!m
->chunk_pack_names
)
142 die(_("multi-pack-index missing required pack-name chunk"));
143 if (!m
->chunk_oid_fanout
)
144 die(_("multi-pack-index missing required OID fanout chunk"));
145 if (!m
->chunk_oid_lookup
)
146 die(_("multi-pack-index missing required OID lookup chunk"));
147 if (!m
->chunk_object_offsets
)
148 die(_("multi-pack-index missing required object offsets chunk"));
150 m
->num_objects
= ntohl(m
->chunk_oid_fanout
[255]);
152 m
->pack_names
= xcalloc(m
->num_packs
, sizeof(*m
->pack_names
));
153 m
->packs
= xcalloc(m
->num_packs
, sizeof(*m
->packs
));
155 cur_pack_name
= (const char *)m
->chunk_pack_names
;
156 for (i
= 0; i
< m
->num_packs
; i
++) {
157 m
->pack_names
[i
] = cur_pack_name
;
159 cur_pack_name
+= strlen(cur_pack_name
) + 1;
161 if (i
&& strcmp(m
->pack_names
[i
], m
->pack_names
[i
- 1]) <= 0)
162 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
163 m
->pack_names
[i
- 1],
173 munmap(midx_map
, midx_size
);
179 static void close_midx(struct multi_pack_index
*m
)
182 munmap((unsigned char *)m
->data
, m
->data_len
);
186 for (i
= 0; i
< m
->num_packs
; i
++) {
188 close_pack(m
->packs
[i
]);
192 FREE_AND_NULL(m
->packs
);
193 FREE_AND_NULL(m
->pack_names
);
196 int prepare_midx_pack(struct multi_pack_index
*m
, uint32_t pack_int_id
)
198 struct strbuf pack_name
= STRBUF_INIT
;
200 if (pack_int_id
>= m
->num_packs
)
201 die(_("bad pack-int-id: %u (%u total packs"),
202 pack_int_id
, m
->num_packs
);
204 if (m
->packs
[pack_int_id
])
207 strbuf_addf(&pack_name
, "%s/pack/%s", m
->object_dir
,
208 m
->pack_names
[pack_int_id
]);
210 m
->packs
[pack_int_id
] = add_packed_git(pack_name
.buf
, pack_name
.len
, m
->local
);
211 strbuf_release(&pack_name
);
212 return !m
->packs
[pack_int_id
];
215 int bsearch_midx(const struct object_id
*oid
, struct multi_pack_index
*m
, uint32_t *result
)
217 return bsearch_hash(oid
->hash
, m
->chunk_oid_fanout
, m
->chunk_oid_lookup
,
218 MIDX_HASH_LEN
, result
);
221 struct object_id
*nth_midxed_object_oid(struct object_id
*oid
,
222 struct multi_pack_index
*m
,
225 if (n
>= m
->num_objects
)
228 hashcpy(oid
->hash
, m
->chunk_oid_lookup
+ m
->hash_len
* n
);
232 static off_t
nth_midxed_offset(struct multi_pack_index
*m
, uint32_t pos
)
234 const unsigned char *offset_data
;
237 offset_data
= m
->chunk_object_offsets
+ pos
* MIDX_CHUNK_OFFSET_WIDTH
;
238 offset32
= get_be32(offset_data
+ sizeof(uint32_t));
240 if (m
->chunk_large_offsets
&& offset32
& MIDX_LARGE_OFFSET_NEEDED
) {
241 if (sizeof(off_t
) < sizeof(uint64_t))
242 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
244 offset32
^= MIDX_LARGE_OFFSET_NEEDED
;
245 return get_be64(m
->chunk_large_offsets
+ sizeof(uint64_t) * offset32
);
251 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index
*m
, uint32_t pos
)
253 return get_be32(m
->chunk_object_offsets
+ pos
* MIDX_CHUNK_OFFSET_WIDTH
);
256 static int nth_midxed_pack_entry(struct multi_pack_index
*m
, struct pack_entry
*e
, uint32_t pos
)
258 uint32_t pack_int_id
;
259 struct packed_git
*p
;
261 if (pos
>= m
->num_objects
)
264 pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
266 if (prepare_midx_pack(m
, pack_int_id
))
267 die(_("error preparing packfile from multi-pack-index"));
268 p
= m
->packs
[pack_int_id
];
271 * We are about to tell the caller where they can locate the
272 * requested object. We better make sure the packfile is
273 * still here and can be accessed before supplying that
274 * answer, as it may have been deleted since the MIDX was
277 if (!is_pack_valid(p
))
280 if (p
->num_bad_objects
) {
282 struct object_id oid
;
283 nth_midxed_object_oid(&oid
, m
, pos
);
284 for (i
= 0; i
< p
->num_bad_objects
; i
++)
285 if (!hashcmp(oid
.hash
,
286 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
290 e
->offset
= nth_midxed_offset(m
, pos
);
296 int fill_midx_entry(const struct object_id
*oid
, struct pack_entry
*e
, struct multi_pack_index
*m
)
300 if (!bsearch_midx(oid
, m
, &pos
))
303 return nth_midxed_pack_entry(m
, e
, pos
);
306 int midx_contains_pack(struct multi_pack_index
*m
, const char *idx_name
)
308 uint32_t first
= 0, last
= m
->num_packs
;
310 while (first
< last
) {
311 uint32_t mid
= first
+ (last
- first
) / 2;
315 current
= m
->pack_names
[mid
];
316 cmp
= strcmp(idx_name
, current
);
329 int prepare_multi_pack_index_one(struct repository
*r
, const char *object_dir
, int local
)
331 struct multi_pack_index
*m
;
332 struct multi_pack_index
*m_search
;
335 if (repo_config_get_bool(r
, "core.multipackindex", &config_value
) ||
339 for (m_search
= r
->objects
->multi_pack_index
; m_search
; m_search
= m_search
->next
)
340 if (!strcmp(object_dir
, m_search
->object_dir
))
343 m
= load_multi_pack_index(object_dir
, local
);
346 m
->next
= r
->objects
->multi_pack_index
;
347 r
->objects
->multi_pack_index
= m
;
354 static size_t write_midx_header(struct hashfile
*f
,
355 unsigned char num_chunks
,
358 unsigned char byte_values
[4];
360 hashwrite_be32(f
, MIDX_SIGNATURE
);
361 byte_values
[0] = MIDX_VERSION
;
362 byte_values
[1] = MIDX_HASH_VERSION
;
363 byte_values
[2] = num_chunks
;
364 byte_values
[3] = 0; /* unused */
365 hashwrite(f
, byte_values
, sizeof(byte_values
));
366 hashwrite_be32(f
, num_packs
);
368 return MIDX_HEADER_SIZE
;
372 struct packed_git
**list
;
376 uint32_t alloc_names
;
377 size_t pack_name_concat_len
;
378 struct multi_pack_index
*m
;
381 static void add_pack_to_midx(const char *full_path
, size_t full_path_len
,
382 const char *file_name
, void *data
)
384 struct pack_list
*packs
= (struct pack_list
*)data
;
386 if (ends_with(file_name
, ".idx")) {
387 if (packs
->m
&& midx_contains_pack(packs
->m
, file_name
))
390 ALLOC_GROW(packs
->list
, packs
->nr
+ 1, packs
->alloc_list
);
391 ALLOC_GROW(packs
->names
, packs
->nr
+ 1, packs
->alloc_names
);
393 packs
->list
[packs
->nr
] = add_packed_git(full_path
,
397 if (!packs
->list
[packs
->nr
]) {
398 warning(_("failed to add packfile '%s'"),
403 if (open_pack_index(packs
->list
[packs
->nr
])) {
404 warning(_("failed to open pack-index '%s'"),
406 close_pack(packs
->list
[packs
->nr
]);
407 FREE_AND_NULL(packs
->list
[packs
->nr
]);
411 packs
->names
[packs
->nr
] = xstrdup(file_name
);
412 packs
->pack_name_concat_len
+= strlen(file_name
) + 1;
418 uint32_t pack_int_id
;
422 static int pack_pair_compare(const void *_a
, const void *_b
)
424 struct pack_pair
*a
= (struct pack_pair
*)_a
;
425 struct pack_pair
*b
= (struct pack_pair
*)_b
;
426 return strcmp(a
->pack_name
, b
->pack_name
);
429 static void sort_packs_by_name(char **pack_names
, uint32_t nr_packs
, uint32_t *perm
)
432 struct pack_pair
*pairs
;
434 ALLOC_ARRAY(pairs
, nr_packs
);
436 for (i
= 0; i
< nr_packs
; i
++) {
437 pairs
[i
].pack_int_id
= i
;
438 pairs
[i
].pack_name
= pack_names
[i
];
441 QSORT(pairs
, nr_packs
, pack_pair_compare
);
443 for (i
= 0; i
< nr_packs
; i
++) {
444 pack_names
[i
] = pairs
[i
].pack_name
;
445 perm
[pairs
[i
].pack_int_id
] = i
;
451 struct pack_midx_entry
{
452 struct object_id oid
;
453 uint32_t pack_int_id
;
458 static int midx_oid_compare(const void *_a
, const void *_b
)
460 const struct pack_midx_entry
*a
= (const struct pack_midx_entry
*)_a
;
461 const struct pack_midx_entry
*b
= (const struct pack_midx_entry
*)_b
;
462 int cmp
= oidcmp(&a
->oid
, &b
->oid
);
467 if (a
->pack_mtime
> b
->pack_mtime
)
469 else if (a
->pack_mtime
< b
->pack_mtime
)
472 return a
->pack_int_id
- b
->pack_int_id
;
475 static int nth_midxed_pack_midx_entry(struct multi_pack_index
*m
,
477 struct pack_midx_entry
*e
,
480 if (pos
>= m
->num_objects
)
483 nth_midxed_object_oid(&e
->oid
, m
, pos
);
484 e
->pack_int_id
= pack_perm
[nth_midxed_pack_int_id(m
, pos
)];
485 e
->offset
= nth_midxed_offset(m
, pos
);
487 /* consider objects in midx to be from "old" packs */
492 static void fill_pack_entry(uint32_t pack_int_id
,
493 struct packed_git
*p
,
495 struct pack_midx_entry
*entry
)
497 if (!nth_packed_object_oid(&entry
->oid
, p
, cur_object
))
498 die(_("failed to locate object %d in packfile"), cur_object
);
500 entry
->pack_int_id
= pack_int_id
;
501 entry
->pack_mtime
= p
->mtime
;
503 entry
->offset
= nth_packed_object_offset(p
, cur_object
);
507 * It is possible to artificially get into a state where there are many
508 * duplicate copies of objects. That can create high memory pressure if
509 * we are to create a list of all objects before de-duplication. To reduce
510 * this memory pressure without a significant performance drop, automatically
511 * group objects by the first byte of their object id. Use the IDX fanout
512 * tables to group the data, copy to a local array, then sort.
514 * Copy only the de-duplicated entries (selected by most-recent modified time
515 * of a packfile containing the object).
517 static struct pack_midx_entry
*get_sorted_entries(struct multi_pack_index
*m
,
518 struct packed_git
**p
,
521 uint32_t *nr_objects
)
523 uint32_t cur_fanout
, cur_pack
, cur_object
;
524 uint32_t alloc_fanout
, alloc_objects
, total_objects
= 0;
525 struct pack_midx_entry
*entries_by_fanout
= NULL
;
526 struct pack_midx_entry
*deduplicated_entries
= NULL
;
527 uint32_t start_pack
= m
? m
->num_packs
: 0;
529 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++)
530 total_objects
+= p
[cur_pack
]->num_objects
;
533 * As we de-duplicate by fanout value, we expect the fanout
534 * slices to be evenly distributed, with some noise. Hence,
535 * allocate slightly more than one 256th.
537 alloc_objects
= alloc_fanout
= total_objects
> 3200 ? total_objects
/ 200 : 16;
539 ALLOC_ARRAY(entries_by_fanout
, alloc_fanout
);
540 ALLOC_ARRAY(deduplicated_entries
, alloc_objects
);
543 for (cur_fanout
= 0; cur_fanout
< 256; cur_fanout
++) {
544 uint32_t nr_fanout
= 0;
547 uint32_t start
= 0, end
;
550 start
= ntohl(m
->chunk_oid_fanout
[cur_fanout
- 1]);
551 end
= ntohl(m
->chunk_oid_fanout
[cur_fanout
]);
553 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
554 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
555 nth_midxed_pack_midx_entry(m
, perm
,
556 &entries_by_fanout
[nr_fanout
],
562 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++) {
563 uint32_t start
= 0, end
;
566 start
= get_pack_fanout(p
[cur_pack
], cur_fanout
- 1);
567 end
= get_pack_fanout(p
[cur_pack
], cur_fanout
);
569 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
570 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
571 fill_pack_entry(perm
[cur_pack
], p
[cur_pack
], cur_object
, &entries_by_fanout
[nr_fanout
]);
576 QSORT(entries_by_fanout
, nr_fanout
, midx_oid_compare
);
579 * The batch is now sorted by OID and then mtime (descending).
580 * Take only the first duplicate.
582 for (cur_object
= 0; cur_object
< nr_fanout
; cur_object
++) {
583 if (cur_object
&& !oidcmp(&entries_by_fanout
[cur_object
- 1].oid
,
584 &entries_by_fanout
[cur_object
].oid
))
587 ALLOC_GROW(deduplicated_entries
, *nr_objects
+ 1, alloc_objects
);
588 memcpy(&deduplicated_entries
[*nr_objects
],
589 &entries_by_fanout
[cur_object
],
590 sizeof(struct pack_midx_entry
));
595 free(entries_by_fanout
);
596 return deduplicated_entries
;
599 static size_t write_midx_pack_names(struct hashfile
*f
,
604 unsigned char padding
[MIDX_CHUNK_ALIGNMENT
];
607 for (i
= 0; i
< num_packs
; i
++) {
608 size_t writelen
= strlen(pack_names
[i
]) + 1;
610 if (i
&& strcmp(pack_names
[i
], pack_names
[i
- 1]) <= 0)
611 BUG("incorrect pack-file order: %s before %s",
615 hashwrite(f
, pack_names
[i
], writelen
);
619 /* add padding to be aligned */
620 i
= MIDX_CHUNK_ALIGNMENT
- (written
% MIDX_CHUNK_ALIGNMENT
);
621 if (i
< MIDX_CHUNK_ALIGNMENT
) {
622 memset(padding
, 0, sizeof(padding
));
623 hashwrite(f
, padding
, i
);
630 static size_t write_midx_oid_fanout(struct hashfile
*f
,
631 struct pack_midx_entry
*objects
,
634 struct pack_midx_entry
*list
= objects
;
635 struct pack_midx_entry
*last
= objects
+ nr_objects
;
640 * Write the first-level table (the list is sorted,
641 * but we use a 256-entry lookup to be able to avoid
642 * having to do eight extra binary search iterations).
644 for (i
= 0; i
< 256; i
++) {
645 struct pack_midx_entry
*next
= list
;
647 while (next
< last
&& next
->oid
.hash
[0] == i
) {
652 hashwrite_be32(f
, count
);
656 return MIDX_CHUNK_FANOUT_SIZE
;
659 static size_t write_midx_oid_lookup(struct hashfile
*f
, unsigned char hash_len
,
660 struct pack_midx_entry
*objects
,
663 struct pack_midx_entry
*list
= objects
;
667 for (i
= 0; i
< nr_objects
; i
++) {
668 struct pack_midx_entry
*obj
= list
++;
670 if (i
< nr_objects
- 1) {
671 struct pack_midx_entry
*next
= list
;
672 if (oidcmp(&obj
->oid
, &next
->oid
) >= 0)
673 BUG("OIDs not in order: %s >= %s",
674 oid_to_hex(&obj
->oid
),
675 oid_to_hex(&next
->oid
));
678 hashwrite(f
, obj
->oid
.hash
, (int)hash_len
);
685 static size_t write_midx_object_offsets(struct hashfile
*f
, int large_offset_needed
,
686 struct pack_midx_entry
*objects
, uint32_t nr_objects
)
688 struct pack_midx_entry
*list
= objects
;
689 uint32_t i
, nr_large_offset
= 0;
692 for (i
= 0; i
< nr_objects
; i
++) {
693 struct pack_midx_entry
*obj
= list
++;
695 hashwrite_be32(f
, obj
->pack_int_id
);
697 if (large_offset_needed
&& obj
->offset
>> 31)
698 hashwrite_be32(f
, MIDX_LARGE_OFFSET_NEEDED
| nr_large_offset
++);
699 else if (!large_offset_needed
&& obj
->offset
>> 32)
700 BUG("object %s requires a large offset (%"PRIx64
") but the MIDX is not writing large offsets!",
701 oid_to_hex(&obj
->oid
),
704 hashwrite_be32(f
, (uint32_t)obj
->offset
);
706 written
+= MIDX_CHUNK_OFFSET_WIDTH
;
712 static size_t write_midx_large_offsets(struct hashfile
*f
, uint32_t nr_large_offset
,
713 struct pack_midx_entry
*objects
, uint32_t nr_objects
)
715 struct pack_midx_entry
*list
= objects
;
718 while (nr_large_offset
) {
719 struct pack_midx_entry
*obj
= list
++;
720 uint64_t offset
= obj
->offset
;
725 hashwrite_be32(f
, offset
>> 32);
726 hashwrite_be32(f
, offset
& 0xffffffffUL
);
727 written
+= 2 * sizeof(uint32_t);
735 int write_midx_file(const char *object_dir
)
737 unsigned char cur_chunk
, num_chunks
= 0;
740 struct hashfile
*f
= NULL
;
742 struct pack_list packs
;
743 uint32_t *pack_perm
= NULL
;
744 uint64_t written
= 0;
745 uint32_t chunk_ids
[MIDX_MAX_CHUNKS
+ 1];
746 uint64_t chunk_offsets
[MIDX_MAX_CHUNKS
+ 1];
747 uint32_t nr_entries
, num_large_offsets
= 0;
748 struct pack_midx_entry
*entries
= NULL
;
749 int large_offsets_needed
= 0;
751 midx_name
= get_midx_filename(object_dir
);
752 if (safe_create_leading_directories(midx_name
)) {
754 die_errno(_("unable to create leading directories of %s"),
758 packs
.m
= load_multi_pack_index(object_dir
, 1);
761 packs
.alloc_list
= packs
.m
? packs
.m
->num_packs
: 16;
762 packs
.alloc_names
= packs
.alloc_list
;
765 packs
.pack_name_concat_len
= 0;
766 ALLOC_ARRAY(packs
.list
, packs
.alloc_list
);
767 ALLOC_ARRAY(packs
.names
, packs
.alloc_names
);
770 for (i
= 0; i
< packs
.m
->num_packs
; i
++) {
771 ALLOC_GROW(packs
.list
, packs
.nr
+ 1, packs
.alloc_list
);
772 ALLOC_GROW(packs
.names
, packs
.nr
+ 1, packs
.alloc_names
);
774 packs
.list
[packs
.nr
] = NULL
;
775 packs
.names
[packs
.nr
] = xstrdup(packs
.m
->pack_names
[i
]);
776 packs
.pack_name_concat_len
+= strlen(packs
.names
[packs
.nr
]) + 1;
781 for_each_file_in_pack_dir(object_dir
, add_pack_to_midx
, &packs
);
783 if (packs
.m
&& packs
.nr
== packs
.m
->num_packs
)
786 if (packs
.pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
)
787 packs
.pack_name_concat_len
+= MIDX_CHUNK_ALIGNMENT
-
788 (packs
.pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
);
790 ALLOC_ARRAY(pack_perm
, packs
.nr
);
791 sort_packs_by_name(packs
.names
, packs
.nr
, pack_perm
);
793 entries
= get_sorted_entries(packs
.m
, packs
.list
, pack_perm
, packs
.nr
, &nr_entries
);
795 for (i
= 0; i
< nr_entries
; i
++) {
796 if (entries
[i
].offset
> 0x7fffffff)
798 if (entries
[i
].offset
> 0xffffffff)
799 large_offsets_needed
= 1;
802 hold_lock_file_for_update(&lk
, midx_name
, LOCK_DIE_ON_ERROR
);
803 f
= hashfd(lk
.tempfile
->fd
, lk
.tempfile
->filename
.buf
);
804 FREE_AND_NULL(midx_name
);
810 num_chunks
= large_offsets_needed
? 5 : 4;
812 written
= write_midx_header(f
, num_chunks
, packs
.nr
);
814 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_PACKNAMES
;
815 chunk_offsets
[cur_chunk
] = written
+ (num_chunks
+ 1) * MIDX_CHUNKLOOKUP_WIDTH
;
818 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OIDFANOUT
;
819 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + packs
.pack_name_concat_len
;
822 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OIDLOOKUP
;
823 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + MIDX_CHUNK_FANOUT_SIZE
;
826 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OBJECTOFFSETS
;
827 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + nr_entries
* MIDX_HASH_LEN
;
830 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + nr_entries
* MIDX_CHUNK_OFFSET_WIDTH
;
831 if (large_offsets_needed
) {
832 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_LARGEOFFSETS
;
835 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] +
836 num_large_offsets
* MIDX_CHUNK_LARGE_OFFSET_WIDTH
;
839 chunk_ids
[cur_chunk
] = 0;
841 for (i
= 0; i
<= num_chunks
; i
++) {
842 if (i
&& chunk_offsets
[i
] < chunk_offsets
[i
- 1])
843 BUG("incorrect chunk offsets: %"PRIu64
" before %"PRIu64
,
844 chunk_offsets
[i
- 1],
847 if (chunk_offsets
[i
] % MIDX_CHUNK_ALIGNMENT
)
848 BUG("chunk offset %"PRIu64
" is not properly aligned",
851 hashwrite_be32(f
, chunk_ids
[i
]);
852 hashwrite_be32(f
, chunk_offsets
[i
] >> 32);
853 hashwrite_be32(f
, chunk_offsets
[i
]);
855 written
+= MIDX_CHUNKLOOKUP_WIDTH
;
858 for (i
= 0; i
< num_chunks
; i
++) {
859 if (written
!= chunk_offsets
[i
])
860 BUG("incorrect chunk offset (%"PRIu64
" != %"PRIu64
") for chunk id %"PRIx32
,
865 switch (chunk_ids
[i
]) {
866 case MIDX_CHUNKID_PACKNAMES
:
867 written
+= write_midx_pack_names(f
, packs
.names
, packs
.nr
);
870 case MIDX_CHUNKID_OIDFANOUT
:
871 written
+= write_midx_oid_fanout(f
, entries
, nr_entries
);
874 case MIDX_CHUNKID_OIDLOOKUP
:
875 written
+= write_midx_oid_lookup(f
, MIDX_HASH_LEN
, entries
, nr_entries
);
878 case MIDX_CHUNKID_OBJECTOFFSETS
:
879 written
+= write_midx_object_offsets(f
, large_offsets_needed
, entries
, nr_entries
);
882 case MIDX_CHUNKID_LARGEOFFSETS
:
883 written
+= write_midx_large_offsets(f
, num_large_offsets
, entries
, nr_entries
);
887 BUG("trying to write unknown chunk id %"PRIx32
,
892 if (written
!= chunk_offsets
[num_chunks
])
893 BUG("incorrect final offset %"PRIu64
" != %"PRIu64
,
895 chunk_offsets
[num_chunks
]);
897 finalize_hashfile(f
, NULL
, CSUM_FSYNC
| CSUM_HASH_IN_STREAM
);
898 commit_lock_file(&lk
);
901 for (i
= 0; i
< packs
.nr
; i
++) {
903 close_pack(packs
.list
[i
]);
906 free(packs
.names
[i
]);
917 void clear_midx_file(const char *object_dir
)
919 char *midx
= get_midx_filename(object_dir
);
921 if (remove_path(midx
)) {
923 die(_("failed to clear multi-pack-index at %s"), midx
);
929 static int verify_midx_error
;
931 static void midx_report(const char *fmt
, ...)
934 verify_midx_error
= 1;
936 vfprintf(stderr
, fmt
, ap
);
937 fprintf(stderr
, "\n");
941 int verify_midx_file(const char *object_dir
)
944 struct progress
*progress
= NULL
;
945 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
946 verify_midx_error
= 0;
951 for (i
= 0; i
< m
->num_packs
; i
++) {
952 if (prepare_midx_pack(m
, i
))
953 midx_report("failed to load pack in position %d", i
);
956 for (i
= 0; i
< 255; i
++) {
957 uint32_t oid_fanout1
= ntohl(m
->chunk_oid_fanout
[i
]);
958 uint32_t oid_fanout2
= ntohl(m
->chunk_oid_fanout
[i
+ 1]);
960 if (oid_fanout1
> oid_fanout2
)
961 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32
" > %"PRIx32
" = fanout[%d]"),
962 i
, oid_fanout1
, oid_fanout2
, i
+ 1);
965 for (i
= 0; i
< m
->num_objects
- 1; i
++) {
966 struct object_id oid1
, oid2
;
968 nth_midxed_object_oid(&oid1
, m
, i
);
969 nth_midxed_object_oid(&oid2
, m
, i
+ 1);
971 if (oidcmp(&oid1
, &oid2
) >= 0)
972 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
973 i
, oid_to_hex(&oid1
), oid_to_hex(&oid2
), i
+ 1);
976 progress
= start_progress(_("Verifying object offsets"), m
->num_objects
);
977 for (i
= 0; i
< m
->num_objects
; i
++) {
978 struct object_id oid
;
980 off_t m_offset
, p_offset
;
982 nth_midxed_object_oid(&oid
, m
, i
);
983 if (!fill_midx_entry(&oid
, &e
, m
)) {
984 midx_report(_("failed to load pack entry for oid[%d] = %s"),
985 i
, oid_to_hex(&oid
));
989 if (open_pack_index(e
.p
)) {
990 midx_report(_("failed to load pack-index for packfile %s"),
996 p_offset
= find_pack_entry_one(oid
.hash
, e
.p
);
998 if (m_offset
!= p_offset
)
999 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64
" != %"PRIx64
),
1000 i
, oid_to_hex(&oid
), m_offset
, p_offset
);
1002 display_progress(progress
, i
+ 1);
1004 stop_progress(&progress
);
1006 return verify_midx_error
;