7 #include "object-store.h"
8 #include "sha1-lookup.h"
11 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
12 #define MIDX_VERSION 1
13 #define MIDX_BYTE_FILE_VERSION 4
14 #define MIDX_BYTE_HASH_VERSION 5
15 #define MIDX_BYTE_NUM_CHUNKS 6
16 #define MIDX_BYTE_NUM_PACKS 8
17 #define MIDX_HASH_VERSION 1
18 #define MIDX_HEADER_SIZE 12
19 #define MIDX_HASH_LEN 20
20 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + MIDX_HASH_LEN)
22 #define MIDX_MAX_CHUNKS 5
23 #define MIDX_CHUNK_ALIGNMENT 4
24 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
25 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
26 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
27 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
28 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
29 #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
30 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
31 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
32 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
33 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
35 static char *get_midx_filename(const char *object_dir
)
37 return xstrfmt("%s/pack/multi-pack-index", object_dir
);
40 struct multi_pack_index
*load_multi_pack_index(const char *object_dir
, int local
)
42 struct multi_pack_index
*m
= NULL
;
46 void *midx_map
= NULL
;
47 uint32_t hash_version
;
48 char *midx_name
= get_midx_filename(object_dir
);
50 const char *cur_pack_name
;
52 fd
= git_open(midx_name
);
57 error_errno(_("failed to read %s"), midx_name
);
61 midx_size
= xsize_t(st
.st_size
);
63 if (midx_size
< MIDX_MIN_SIZE
) {
64 error(_("multi-pack-index file %s is too small"), midx_name
);
68 FREE_AND_NULL(midx_name
);
70 midx_map
= xmmap(NULL
, midx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
72 FLEX_ALLOC_MEM(m
, object_dir
, object_dir
, strlen(object_dir
));
75 m
->data_len
= midx_size
;
78 m
->signature
= get_be32(m
->data
);
79 if (m
->signature
!= MIDX_SIGNATURE
)
80 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
81 m
->signature
, MIDX_SIGNATURE
);
83 m
->version
= m
->data
[MIDX_BYTE_FILE_VERSION
];
84 if (m
->version
!= MIDX_VERSION
)
85 die(_("multi-pack-index version %d not recognized"),
88 hash_version
= m
->data
[MIDX_BYTE_HASH_VERSION
];
89 if (hash_version
!= MIDX_HASH_VERSION
)
90 die(_("hash version %u does not match"), hash_version
);
91 m
->hash_len
= MIDX_HASH_LEN
;
93 m
->num_chunks
= m
->data
[MIDX_BYTE_NUM_CHUNKS
];
95 m
->num_packs
= get_be32(m
->data
+ MIDX_BYTE_NUM_PACKS
);
97 for (i
= 0; i
< m
->num_chunks
; i
++) {
98 uint32_t chunk_id
= get_be32(m
->data
+ MIDX_HEADER_SIZE
+
99 MIDX_CHUNKLOOKUP_WIDTH
* i
);
100 uint64_t chunk_offset
= get_be64(m
->data
+ MIDX_HEADER_SIZE
+ 4 +
101 MIDX_CHUNKLOOKUP_WIDTH
* i
);
103 if (chunk_offset
>= m
->data_len
)
104 die(_("invalid chunk offset (too large)"));
107 case MIDX_CHUNKID_PACKNAMES
:
108 m
->chunk_pack_names
= m
->data
+ chunk_offset
;
111 case MIDX_CHUNKID_OIDFANOUT
:
112 m
->chunk_oid_fanout
= (uint32_t *)(m
->data
+ chunk_offset
);
115 case MIDX_CHUNKID_OIDLOOKUP
:
116 m
->chunk_oid_lookup
= m
->data
+ chunk_offset
;
119 case MIDX_CHUNKID_OBJECTOFFSETS
:
120 m
->chunk_object_offsets
= m
->data
+ chunk_offset
;
123 case MIDX_CHUNKID_LARGEOFFSETS
:
124 m
->chunk_large_offsets
= m
->data
+ chunk_offset
;
128 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
133 * Do nothing on unrecognized chunks, allowing future
134 * extensions to add optional chunks.
140 if (!m
->chunk_pack_names
)
141 die(_("multi-pack-index missing required pack-name chunk"));
142 if (!m
->chunk_oid_fanout
)
143 die(_("multi-pack-index missing required OID fanout chunk"));
144 if (!m
->chunk_oid_lookup
)
145 die(_("multi-pack-index missing required OID lookup chunk"));
146 if (!m
->chunk_object_offsets
)
147 die(_("multi-pack-index missing required object offsets chunk"));
149 m
->num_objects
= ntohl(m
->chunk_oid_fanout
[255]);
151 m
->pack_names
= xcalloc(m
->num_packs
, sizeof(*m
->pack_names
));
152 m
->packs
= xcalloc(m
->num_packs
, sizeof(*m
->packs
));
154 cur_pack_name
= (const char *)m
->chunk_pack_names
;
155 for (i
= 0; i
< m
->num_packs
; i
++) {
156 m
->pack_names
[i
] = cur_pack_name
;
158 cur_pack_name
+= strlen(cur_pack_name
) + 1;
160 if (i
&& strcmp(m
->pack_names
[i
], m
->pack_names
[i
- 1]) <= 0)
161 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
162 m
->pack_names
[i
- 1],
172 munmap(midx_map
, midx_size
);
178 static void close_midx(struct multi_pack_index
*m
)
181 munmap((unsigned char *)m
->data
, m
->data_len
);
185 for (i
= 0; i
< m
->num_packs
; i
++) {
187 close_pack(m
->packs
[i
]);
191 FREE_AND_NULL(m
->packs
);
192 FREE_AND_NULL(m
->pack_names
);
195 int prepare_midx_pack(struct multi_pack_index
*m
, uint32_t pack_int_id
)
197 struct strbuf pack_name
= STRBUF_INIT
;
199 if (pack_int_id
>= m
->num_packs
)
200 die(_("bad pack-int-id: %u (%u total packs"),
201 pack_int_id
, m
->num_packs
);
203 if (m
->packs
[pack_int_id
])
206 strbuf_addf(&pack_name
, "%s/pack/%s", m
->object_dir
,
207 m
->pack_names
[pack_int_id
]);
209 m
->packs
[pack_int_id
] = add_packed_git(pack_name
.buf
, pack_name
.len
, m
->local
);
210 strbuf_release(&pack_name
);
211 return !m
->packs
[pack_int_id
];
214 int bsearch_midx(const struct object_id
*oid
, struct multi_pack_index
*m
, uint32_t *result
)
216 return bsearch_hash(oid
->hash
, m
->chunk_oid_fanout
, m
->chunk_oid_lookup
,
217 MIDX_HASH_LEN
, result
);
220 struct object_id
*nth_midxed_object_oid(struct object_id
*oid
,
221 struct multi_pack_index
*m
,
224 if (n
>= m
->num_objects
)
227 hashcpy(oid
->hash
, m
->chunk_oid_lookup
+ m
->hash_len
* n
);
231 static off_t
nth_midxed_offset(struct multi_pack_index
*m
, uint32_t pos
)
233 const unsigned char *offset_data
;
236 offset_data
= m
->chunk_object_offsets
+ pos
* MIDX_CHUNK_OFFSET_WIDTH
;
237 offset32
= get_be32(offset_data
+ sizeof(uint32_t));
239 if (m
->chunk_large_offsets
&& offset32
& MIDX_LARGE_OFFSET_NEEDED
) {
240 if (sizeof(off_t
) < sizeof(uint64_t))
241 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
243 offset32
^= MIDX_LARGE_OFFSET_NEEDED
;
244 return get_be64(m
->chunk_large_offsets
+ sizeof(uint64_t) * offset32
);
250 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index
*m
, uint32_t pos
)
252 return get_be32(m
->chunk_object_offsets
+ pos
* MIDX_CHUNK_OFFSET_WIDTH
);
255 static int nth_midxed_pack_entry(struct multi_pack_index
*m
, struct pack_entry
*e
, uint32_t pos
)
257 uint32_t pack_int_id
;
258 struct packed_git
*p
;
260 if (pos
>= m
->num_objects
)
263 pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
265 if (prepare_midx_pack(m
, pack_int_id
))
266 die(_("error preparing packfile from multi-pack-index"));
267 p
= m
->packs
[pack_int_id
];
270 * We are about to tell the caller where they can locate the
271 * requested object. We better make sure the packfile is
272 * still here and can be accessed before supplying that
273 * answer, as it may have been deleted since the MIDX was
276 if (!is_pack_valid(p
))
279 if (p
->num_bad_objects
) {
281 struct object_id oid
;
282 nth_midxed_object_oid(&oid
, m
, pos
);
283 for (i
= 0; i
< p
->num_bad_objects
; i
++)
284 if (!hashcmp(oid
.hash
,
285 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
289 e
->offset
= nth_midxed_offset(m
, pos
);
295 int fill_midx_entry(const struct object_id
*oid
, struct pack_entry
*e
, struct multi_pack_index
*m
)
299 if (!bsearch_midx(oid
, m
, &pos
))
302 return nth_midxed_pack_entry(m
, e
, pos
);
305 int midx_contains_pack(struct multi_pack_index
*m
, const char *idx_name
)
307 uint32_t first
= 0, last
= m
->num_packs
;
309 while (first
< last
) {
310 uint32_t mid
= first
+ (last
- first
) / 2;
314 current
= m
->pack_names
[mid
];
315 cmp
= strcmp(idx_name
, current
);
328 int prepare_multi_pack_index_one(struct repository
*r
, const char *object_dir
, int local
)
330 struct multi_pack_index
*m
;
331 struct multi_pack_index
*m_search
;
334 if (repo_config_get_bool(r
, "core.multipackindex", &config_value
) ||
338 for (m_search
= r
->objects
->multi_pack_index
; m_search
; m_search
= m_search
->next
)
339 if (!strcmp(object_dir
, m_search
->object_dir
))
342 m
= load_multi_pack_index(object_dir
, local
);
345 m
->next
= r
->objects
->multi_pack_index
;
346 r
->objects
->multi_pack_index
= m
;
353 static size_t write_midx_header(struct hashfile
*f
,
354 unsigned char num_chunks
,
357 unsigned char byte_values
[4];
359 hashwrite_be32(f
, MIDX_SIGNATURE
);
360 byte_values
[0] = MIDX_VERSION
;
361 byte_values
[1] = MIDX_HASH_VERSION
;
362 byte_values
[2] = num_chunks
;
363 byte_values
[3] = 0; /* unused */
364 hashwrite(f
, byte_values
, sizeof(byte_values
));
365 hashwrite_be32(f
, num_packs
);
367 return MIDX_HEADER_SIZE
;
371 struct packed_git
**list
;
375 uint32_t alloc_names
;
376 size_t pack_name_concat_len
;
377 struct multi_pack_index
*m
;
380 static void add_pack_to_midx(const char *full_path
, size_t full_path_len
,
381 const char *file_name
, void *data
)
383 struct pack_list
*packs
= (struct pack_list
*)data
;
385 if (ends_with(file_name
, ".idx")) {
386 if (packs
->m
&& midx_contains_pack(packs
->m
, file_name
))
389 ALLOC_GROW(packs
->list
, packs
->nr
+ 1, packs
->alloc_list
);
390 ALLOC_GROW(packs
->names
, packs
->nr
+ 1, packs
->alloc_names
);
392 packs
->list
[packs
->nr
] = add_packed_git(full_path
,
396 if (!packs
->list
[packs
->nr
]) {
397 warning(_("failed to add packfile '%s'"),
402 if (open_pack_index(packs
->list
[packs
->nr
])) {
403 warning(_("failed to open pack-index '%s'"),
405 close_pack(packs
->list
[packs
->nr
]);
406 FREE_AND_NULL(packs
->list
[packs
->nr
]);
410 packs
->names
[packs
->nr
] = xstrdup(file_name
);
411 packs
->pack_name_concat_len
+= strlen(file_name
) + 1;
417 uint32_t pack_int_id
;
421 static int pack_pair_compare(const void *_a
, const void *_b
)
423 struct pack_pair
*a
= (struct pack_pair
*)_a
;
424 struct pack_pair
*b
= (struct pack_pair
*)_b
;
425 return strcmp(a
->pack_name
, b
->pack_name
);
428 static void sort_packs_by_name(char **pack_names
, uint32_t nr_packs
, uint32_t *perm
)
431 struct pack_pair
*pairs
;
433 ALLOC_ARRAY(pairs
, nr_packs
);
435 for (i
= 0; i
< nr_packs
; i
++) {
436 pairs
[i
].pack_int_id
= i
;
437 pairs
[i
].pack_name
= pack_names
[i
];
440 QSORT(pairs
, nr_packs
, pack_pair_compare
);
442 for (i
= 0; i
< nr_packs
; i
++) {
443 pack_names
[i
] = pairs
[i
].pack_name
;
444 perm
[pairs
[i
].pack_int_id
] = i
;
450 struct pack_midx_entry
{
451 struct object_id oid
;
452 uint32_t pack_int_id
;
457 static int midx_oid_compare(const void *_a
, const void *_b
)
459 const struct pack_midx_entry
*a
= (const struct pack_midx_entry
*)_a
;
460 const struct pack_midx_entry
*b
= (const struct pack_midx_entry
*)_b
;
461 int cmp
= oidcmp(&a
->oid
, &b
->oid
);
466 if (a
->pack_mtime
> b
->pack_mtime
)
468 else if (a
->pack_mtime
< b
->pack_mtime
)
471 return a
->pack_int_id
- b
->pack_int_id
;
474 static int nth_midxed_pack_midx_entry(struct multi_pack_index
*m
,
476 struct pack_midx_entry
*e
,
479 if (pos
>= m
->num_objects
)
482 nth_midxed_object_oid(&e
->oid
, m
, pos
);
483 e
->pack_int_id
= pack_perm
[nth_midxed_pack_int_id(m
, pos
)];
484 e
->offset
= nth_midxed_offset(m
, pos
);
486 /* consider objects in midx to be from "old" packs */
491 static void fill_pack_entry(uint32_t pack_int_id
,
492 struct packed_git
*p
,
494 struct pack_midx_entry
*entry
)
496 if (!nth_packed_object_oid(&entry
->oid
, p
, cur_object
))
497 die(_("failed to locate object %d in packfile"), cur_object
);
499 entry
->pack_int_id
= pack_int_id
;
500 entry
->pack_mtime
= p
->mtime
;
502 entry
->offset
= nth_packed_object_offset(p
, cur_object
);
506 * It is possible to artificially get into a state where there are many
507 * duplicate copies of objects. That can create high memory pressure if
508 * we are to create a list of all objects before de-duplication. To reduce
509 * this memory pressure without a significant performance drop, automatically
510 * group objects by the first byte of their object id. Use the IDX fanout
511 * tables to group the data, copy to a local array, then sort.
513 * Copy only the de-duplicated entries (selected by most-recent modified time
514 * of a packfile containing the object).
516 static struct pack_midx_entry
*get_sorted_entries(struct multi_pack_index
*m
,
517 struct packed_git
**p
,
520 uint32_t *nr_objects
)
522 uint32_t cur_fanout
, cur_pack
, cur_object
;
523 uint32_t alloc_fanout
, alloc_objects
, total_objects
= 0;
524 struct pack_midx_entry
*entries_by_fanout
= NULL
;
525 struct pack_midx_entry
*deduplicated_entries
= NULL
;
526 uint32_t start_pack
= m
? m
->num_packs
: 0;
528 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++)
529 total_objects
+= p
[cur_pack
]->num_objects
;
532 * As we de-duplicate by fanout value, we expect the fanout
533 * slices to be evenly distributed, with some noise. Hence,
534 * allocate slightly more than one 256th.
536 alloc_objects
= alloc_fanout
= total_objects
> 3200 ? total_objects
/ 200 : 16;
538 ALLOC_ARRAY(entries_by_fanout
, alloc_fanout
);
539 ALLOC_ARRAY(deduplicated_entries
, alloc_objects
);
542 for (cur_fanout
= 0; cur_fanout
< 256; cur_fanout
++) {
543 uint32_t nr_fanout
= 0;
546 uint32_t start
= 0, end
;
549 start
= ntohl(m
->chunk_oid_fanout
[cur_fanout
- 1]);
550 end
= ntohl(m
->chunk_oid_fanout
[cur_fanout
]);
552 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
553 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
554 nth_midxed_pack_midx_entry(m
, perm
,
555 &entries_by_fanout
[nr_fanout
],
561 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++) {
562 uint32_t start
= 0, end
;
565 start
= get_pack_fanout(p
[cur_pack
], cur_fanout
- 1);
566 end
= get_pack_fanout(p
[cur_pack
], cur_fanout
);
568 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
569 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
570 fill_pack_entry(perm
[cur_pack
], p
[cur_pack
], cur_object
, &entries_by_fanout
[nr_fanout
]);
575 QSORT(entries_by_fanout
, nr_fanout
, midx_oid_compare
);
578 * The batch is now sorted by OID and then mtime (descending).
579 * Take only the first duplicate.
581 for (cur_object
= 0; cur_object
< nr_fanout
; cur_object
++) {
582 if (cur_object
&& !oidcmp(&entries_by_fanout
[cur_object
- 1].oid
,
583 &entries_by_fanout
[cur_object
].oid
))
586 ALLOC_GROW(deduplicated_entries
, *nr_objects
+ 1, alloc_objects
);
587 memcpy(&deduplicated_entries
[*nr_objects
],
588 &entries_by_fanout
[cur_object
],
589 sizeof(struct pack_midx_entry
));
594 free(entries_by_fanout
);
595 return deduplicated_entries
;
598 static size_t write_midx_pack_names(struct hashfile
*f
,
603 unsigned char padding
[MIDX_CHUNK_ALIGNMENT
];
606 for (i
= 0; i
< num_packs
; i
++) {
607 size_t writelen
= strlen(pack_names
[i
]) + 1;
609 if (i
&& strcmp(pack_names
[i
], pack_names
[i
- 1]) <= 0)
610 BUG("incorrect pack-file order: %s before %s",
614 hashwrite(f
, pack_names
[i
], writelen
);
618 /* add padding to be aligned */
619 i
= MIDX_CHUNK_ALIGNMENT
- (written
% MIDX_CHUNK_ALIGNMENT
);
620 if (i
< MIDX_CHUNK_ALIGNMENT
) {
621 memset(padding
, 0, sizeof(padding
));
622 hashwrite(f
, padding
, i
);
629 static size_t write_midx_oid_fanout(struct hashfile
*f
,
630 struct pack_midx_entry
*objects
,
633 struct pack_midx_entry
*list
= objects
;
634 struct pack_midx_entry
*last
= objects
+ nr_objects
;
639 * Write the first-level table (the list is sorted,
640 * but we use a 256-entry lookup to be able to avoid
641 * having to do eight extra binary search iterations).
643 for (i
= 0; i
< 256; i
++) {
644 struct pack_midx_entry
*next
= list
;
646 while (next
< last
&& next
->oid
.hash
[0] == i
) {
651 hashwrite_be32(f
, count
);
655 return MIDX_CHUNK_FANOUT_SIZE
;
658 static size_t write_midx_oid_lookup(struct hashfile
*f
, unsigned char hash_len
,
659 struct pack_midx_entry
*objects
,
662 struct pack_midx_entry
*list
= objects
;
666 for (i
= 0; i
< nr_objects
; i
++) {
667 struct pack_midx_entry
*obj
= list
++;
669 if (i
< nr_objects
- 1) {
670 struct pack_midx_entry
*next
= list
;
671 if (oidcmp(&obj
->oid
, &next
->oid
) >= 0)
672 BUG("OIDs not in order: %s >= %s",
673 oid_to_hex(&obj
->oid
),
674 oid_to_hex(&next
->oid
));
677 hashwrite(f
, obj
->oid
.hash
, (int)hash_len
);
684 static size_t write_midx_object_offsets(struct hashfile
*f
, int large_offset_needed
,
685 struct pack_midx_entry
*objects
, uint32_t nr_objects
)
687 struct pack_midx_entry
*list
= objects
;
688 uint32_t i
, nr_large_offset
= 0;
691 for (i
= 0; i
< nr_objects
; i
++) {
692 struct pack_midx_entry
*obj
= list
++;
694 hashwrite_be32(f
, obj
->pack_int_id
);
696 if (large_offset_needed
&& obj
->offset
>> 31)
697 hashwrite_be32(f
, MIDX_LARGE_OFFSET_NEEDED
| nr_large_offset
++);
698 else if (!large_offset_needed
&& obj
->offset
>> 32)
699 BUG("object %s requires a large offset (%"PRIx64
") but the MIDX is not writing large offsets!",
700 oid_to_hex(&obj
->oid
),
703 hashwrite_be32(f
, (uint32_t)obj
->offset
);
705 written
+= MIDX_CHUNK_OFFSET_WIDTH
;
711 static size_t write_midx_large_offsets(struct hashfile
*f
, uint32_t nr_large_offset
,
712 struct pack_midx_entry
*objects
, uint32_t nr_objects
)
714 struct pack_midx_entry
*list
= objects
;
717 while (nr_large_offset
) {
718 struct pack_midx_entry
*obj
= list
++;
719 uint64_t offset
= obj
->offset
;
724 hashwrite_be32(f
, offset
>> 32);
725 hashwrite_be32(f
, offset
& 0xffffffffUL
);
726 written
+= 2 * sizeof(uint32_t);
734 int write_midx_file(const char *object_dir
)
736 unsigned char cur_chunk
, num_chunks
= 0;
739 struct hashfile
*f
= NULL
;
741 struct pack_list packs
;
742 uint32_t *pack_perm
= NULL
;
743 uint64_t written
= 0;
744 uint32_t chunk_ids
[MIDX_MAX_CHUNKS
+ 1];
745 uint64_t chunk_offsets
[MIDX_MAX_CHUNKS
+ 1];
746 uint32_t nr_entries
, num_large_offsets
= 0;
747 struct pack_midx_entry
*entries
= NULL
;
748 int large_offsets_needed
= 0;
750 midx_name
= get_midx_filename(object_dir
);
751 if (safe_create_leading_directories(midx_name
)) {
753 die_errno(_("unable to create leading directories of %s"),
757 packs
.m
= load_multi_pack_index(object_dir
, 1);
760 packs
.alloc_list
= packs
.m
? packs
.m
->num_packs
: 16;
761 packs
.alloc_names
= packs
.alloc_list
;
764 packs
.pack_name_concat_len
= 0;
765 ALLOC_ARRAY(packs
.list
, packs
.alloc_list
);
766 ALLOC_ARRAY(packs
.names
, packs
.alloc_names
);
769 for (i
= 0; i
< packs
.m
->num_packs
; i
++) {
770 ALLOC_GROW(packs
.list
, packs
.nr
+ 1, packs
.alloc_list
);
771 ALLOC_GROW(packs
.names
, packs
.nr
+ 1, packs
.alloc_names
);
773 packs
.list
[packs
.nr
] = NULL
;
774 packs
.names
[packs
.nr
] = xstrdup(packs
.m
->pack_names
[i
]);
775 packs
.pack_name_concat_len
+= strlen(packs
.names
[packs
.nr
]) + 1;
780 for_each_file_in_pack_dir(object_dir
, add_pack_to_midx
, &packs
);
782 if (packs
.m
&& packs
.nr
== packs
.m
->num_packs
)
785 if (packs
.pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
)
786 packs
.pack_name_concat_len
+= MIDX_CHUNK_ALIGNMENT
-
787 (packs
.pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
);
789 ALLOC_ARRAY(pack_perm
, packs
.nr
);
790 sort_packs_by_name(packs
.names
, packs
.nr
, pack_perm
);
792 entries
= get_sorted_entries(packs
.m
, packs
.list
, pack_perm
, packs
.nr
, &nr_entries
);
794 for (i
= 0; i
< nr_entries
; i
++) {
795 if (entries
[i
].offset
> 0x7fffffff)
797 if (entries
[i
].offset
> 0xffffffff)
798 large_offsets_needed
= 1;
801 hold_lock_file_for_update(&lk
, midx_name
, LOCK_DIE_ON_ERROR
);
802 f
= hashfd(lk
.tempfile
->fd
, lk
.tempfile
->filename
.buf
);
803 FREE_AND_NULL(midx_name
);
809 num_chunks
= large_offsets_needed
? 5 : 4;
811 written
= write_midx_header(f
, num_chunks
, packs
.nr
);
813 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_PACKNAMES
;
814 chunk_offsets
[cur_chunk
] = written
+ (num_chunks
+ 1) * MIDX_CHUNKLOOKUP_WIDTH
;
817 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OIDFANOUT
;
818 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + packs
.pack_name_concat_len
;
821 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OIDLOOKUP
;
822 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + MIDX_CHUNK_FANOUT_SIZE
;
825 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OBJECTOFFSETS
;
826 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + nr_entries
* MIDX_HASH_LEN
;
829 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + nr_entries
* MIDX_CHUNK_OFFSET_WIDTH
;
830 if (large_offsets_needed
) {
831 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_LARGEOFFSETS
;
834 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] +
835 num_large_offsets
* MIDX_CHUNK_LARGE_OFFSET_WIDTH
;
838 chunk_ids
[cur_chunk
] = 0;
840 for (i
= 0; i
<= num_chunks
; i
++) {
841 if (i
&& chunk_offsets
[i
] < chunk_offsets
[i
- 1])
842 BUG("incorrect chunk offsets: %"PRIu64
" before %"PRIu64
,
843 chunk_offsets
[i
- 1],
846 if (chunk_offsets
[i
] % MIDX_CHUNK_ALIGNMENT
)
847 BUG("chunk offset %"PRIu64
" is not properly aligned",
850 hashwrite_be32(f
, chunk_ids
[i
]);
851 hashwrite_be32(f
, chunk_offsets
[i
] >> 32);
852 hashwrite_be32(f
, chunk_offsets
[i
]);
854 written
+= MIDX_CHUNKLOOKUP_WIDTH
;
857 for (i
= 0; i
< num_chunks
; i
++) {
858 if (written
!= chunk_offsets
[i
])
859 BUG("incorrect chunk offset (%"PRIu64
" != %"PRIu64
") for chunk id %"PRIx32
,
864 switch (chunk_ids
[i
]) {
865 case MIDX_CHUNKID_PACKNAMES
:
866 written
+= write_midx_pack_names(f
, packs
.names
, packs
.nr
);
869 case MIDX_CHUNKID_OIDFANOUT
:
870 written
+= write_midx_oid_fanout(f
, entries
, nr_entries
);
873 case MIDX_CHUNKID_OIDLOOKUP
:
874 written
+= write_midx_oid_lookup(f
, MIDX_HASH_LEN
, entries
, nr_entries
);
877 case MIDX_CHUNKID_OBJECTOFFSETS
:
878 written
+= write_midx_object_offsets(f
, large_offsets_needed
, entries
, nr_entries
);
881 case MIDX_CHUNKID_LARGEOFFSETS
:
882 written
+= write_midx_large_offsets(f
, num_large_offsets
, entries
, nr_entries
);
886 BUG("trying to write unknown chunk id %"PRIx32
,
891 if (written
!= chunk_offsets
[num_chunks
])
892 BUG("incorrect final offset %"PRIu64
" != %"PRIu64
,
894 chunk_offsets
[num_chunks
]);
896 finalize_hashfile(f
, NULL
, CSUM_FSYNC
| CSUM_HASH_IN_STREAM
);
897 commit_lock_file(&lk
);
900 for (i
= 0; i
< packs
.nr
; i
++) {
902 close_pack(packs
.list
[i
]);
905 free(packs
.names
[i
]);
916 void clear_midx_file(const char *object_dir
)
918 char *midx
= get_midx_filename(object_dir
);
920 if (remove_path(midx
)) {
922 die(_("failed to clear multi-pack-index at %s"), midx
);
928 static int verify_midx_error
;
930 static void midx_report(const char *fmt
, ...)
933 verify_midx_error
= 1;
935 vfprintf(stderr
, fmt
, ap
);
936 fprintf(stderr
, "\n");
940 int verify_midx_file(const char *object_dir
)
943 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
944 verify_midx_error
= 0;
949 for (i
= 0; i
< m
->num_packs
; i
++) {
950 if (prepare_midx_pack(m
, i
))
951 midx_report("failed to load pack in position %d", i
);
954 for (i
= 0; i
< 255; i
++) {
955 uint32_t oid_fanout1
= ntohl(m
->chunk_oid_fanout
[i
]);
956 uint32_t oid_fanout2
= ntohl(m
->chunk_oid_fanout
[i
+ 1]);
958 if (oid_fanout1
> oid_fanout2
)
959 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32
" > %"PRIx32
" = fanout[%d]"),
960 i
, oid_fanout1
, oid_fanout2
, i
+ 1);
963 for (i
= 0; i
< m
->num_objects
- 1; i
++) {
964 struct object_id oid1
, oid2
;
966 nth_midxed_object_oid(&oid1
, m
, i
);
967 nth_midxed_object_oid(&oid2
, m
, i
+ 1);
969 if (oidcmp(&oid1
, &oid2
) >= 0)
970 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
971 i
, oid_to_hex(&oid1
), oid_to_hex(&oid2
), i
+ 1);
974 for (i
= 0; i
< m
->num_objects
; i
++) {
975 struct object_id oid
;
977 off_t m_offset
, p_offset
;
979 nth_midxed_object_oid(&oid
, m
, i
);
980 if (!fill_midx_entry(&oid
, &e
, m
)) {
981 midx_report(_("failed to load pack entry for oid[%d] = %s"),
982 i
, oid_to_hex(&oid
));
986 if (open_pack_index(e
.p
)) {
987 midx_report(_("failed to load pack-index for packfile %s"),
993 p_offset
= find_pack_entry_one(oid
.hash
, e
.p
);
995 if (m_offset
!= p_offset
)
996 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64
" != %"PRIx64
),
997 i
, oid_to_hex(&oid
), m_offset
, p_offset
);
1000 return verify_midx_error
;