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 error(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
81 m
->signature
, MIDX_SIGNATURE
);
85 m
->version
= m
->data
[MIDX_BYTE_FILE_VERSION
];
86 if (m
->version
!= MIDX_VERSION
) {
87 error(_("multi-pack-index version %d not recognized"),
92 hash_version
= m
->data
[MIDX_BYTE_HASH_VERSION
];
93 if (hash_version
!= MIDX_HASH_VERSION
) {
94 error(_("hash version %u does not match"), hash_version
);
97 m
->hash_len
= MIDX_HASH_LEN
;
99 m
->num_chunks
= m
->data
[MIDX_BYTE_NUM_CHUNKS
];
101 m
->num_packs
= get_be32(m
->data
+ MIDX_BYTE_NUM_PACKS
);
103 for (i
= 0; i
< m
->num_chunks
; i
++) {
104 uint32_t chunk_id
= get_be32(m
->data
+ MIDX_HEADER_SIZE
+
105 MIDX_CHUNKLOOKUP_WIDTH
* i
);
106 uint64_t chunk_offset
= get_be64(m
->data
+ MIDX_HEADER_SIZE
+ 4 +
107 MIDX_CHUNKLOOKUP_WIDTH
* i
);
110 case MIDX_CHUNKID_PACKNAMES
:
111 m
->chunk_pack_names
= m
->data
+ chunk_offset
;
114 case MIDX_CHUNKID_OIDFANOUT
:
115 m
->chunk_oid_fanout
= (uint32_t *)(m
->data
+ chunk_offset
);
118 case MIDX_CHUNKID_OIDLOOKUP
:
119 m
->chunk_oid_lookup
= m
->data
+ chunk_offset
;
122 case MIDX_CHUNKID_OBJECTOFFSETS
:
123 m
->chunk_object_offsets
= m
->data
+ chunk_offset
;
126 case MIDX_CHUNKID_LARGEOFFSETS
:
127 m
->chunk_large_offsets
= m
->data
+ chunk_offset
;
131 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
136 * Do nothing on unrecognized chunks, allowing future
137 * extensions to add optional chunks.
143 if (!m
->chunk_pack_names
)
144 die(_("multi-pack-index missing required pack-name chunk"));
145 if (!m
->chunk_oid_fanout
)
146 die(_("multi-pack-index missing required OID fanout chunk"));
147 if (!m
->chunk_oid_lookup
)
148 die(_("multi-pack-index missing required OID lookup chunk"));
149 if (!m
->chunk_object_offsets
)
150 die(_("multi-pack-index missing required object offsets chunk"));
152 m
->num_objects
= ntohl(m
->chunk_oid_fanout
[255]);
154 m
->pack_names
= xcalloc(m
->num_packs
, sizeof(*m
->pack_names
));
155 m
->packs
= xcalloc(m
->num_packs
, sizeof(*m
->packs
));
157 cur_pack_name
= (const char *)m
->chunk_pack_names
;
158 for (i
= 0; i
< m
->num_packs
; i
++) {
159 m
->pack_names
[i
] = cur_pack_name
;
161 cur_pack_name
+= strlen(cur_pack_name
) + 1;
163 if (i
&& strcmp(m
->pack_names
[i
], m
->pack_names
[i
- 1]) <= 0) {
164 error(_("multi-pack-index pack names out of order: '%s' before '%s'"),
165 m
->pack_names
[i
- 1],
177 munmap(midx_map
, midx_size
);
183 static void close_midx(struct multi_pack_index
*m
)
186 munmap((unsigned char *)m
->data
, m
->data_len
);
190 for (i
= 0; i
< m
->num_packs
; i
++) {
192 close_pack(m
->packs
[i
]);
196 FREE_AND_NULL(m
->packs
);
197 FREE_AND_NULL(m
->pack_names
);
200 int prepare_midx_pack(struct multi_pack_index
*m
, uint32_t pack_int_id
)
202 struct strbuf pack_name
= STRBUF_INIT
;
204 if (pack_int_id
>= m
->num_packs
)
205 BUG("bad pack-int-id");
207 if (m
->packs
[pack_int_id
])
210 strbuf_addf(&pack_name
, "%s/pack/%s", m
->object_dir
,
211 m
->pack_names
[pack_int_id
]);
213 m
->packs
[pack_int_id
] = add_packed_git(pack_name
.buf
, pack_name
.len
, m
->local
);
214 strbuf_release(&pack_name
);
215 return !m
->packs
[pack_int_id
];
218 int bsearch_midx(const struct object_id
*oid
, struct multi_pack_index
*m
, uint32_t *result
)
220 return bsearch_hash(oid
->hash
, m
->chunk_oid_fanout
, m
->chunk_oid_lookup
,
221 MIDX_HASH_LEN
, result
);
224 struct object_id
*nth_midxed_object_oid(struct object_id
*oid
,
225 struct multi_pack_index
*m
,
228 if (n
>= m
->num_objects
)
231 hashcpy(oid
->hash
, m
->chunk_oid_lookup
+ m
->hash_len
* n
);
235 static off_t
nth_midxed_offset(struct multi_pack_index
*m
, uint32_t pos
)
237 const unsigned char *offset_data
;
240 offset_data
= m
->chunk_object_offsets
+ pos
* MIDX_CHUNK_OFFSET_WIDTH
;
241 offset32
= get_be32(offset_data
+ sizeof(uint32_t));
243 if (m
->chunk_large_offsets
&& offset32
& MIDX_LARGE_OFFSET_NEEDED
) {
244 if (sizeof(offset32
) < sizeof(uint64_t))
245 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
247 offset32
^= MIDX_LARGE_OFFSET_NEEDED
;
248 return get_be64(m
->chunk_large_offsets
+ sizeof(uint64_t) * offset32
);
254 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index
*m
, uint32_t pos
)
256 return get_be32(m
->chunk_object_offsets
+ pos
* MIDX_CHUNK_OFFSET_WIDTH
);
259 static int nth_midxed_pack_entry(struct multi_pack_index
*m
, struct pack_entry
*e
, uint32_t pos
)
261 uint32_t pack_int_id
;
262 struct packed_git
*p
;
264 if (pos
>= m
->num_objects
)
267 pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
269 if (prepare_midx_pack(m
, pack_int_id
))
270 die(_("error preparing packfile from multi-pack-index"));
271 p
= m
->packs
[pack_int_id
];
274 * We are about to tell the caller where they can locate the
275 * requested object. We better make sure the packfile is
276 * still here and can be accessed before supplying that
277 * answer, as it may have been deleted since the MIDX was
280 if (!is_pack_valid(p
))
283 if (p
->num_bad_objects
) {
285 struct object_id oid
;
286 nth_midxed_object_oid(&oid
, m
, pos
);
287 for (i
= 0; i
< p
->num_bad_objects
; i
++)
288 if (!hashcmp(oid
.hash
,
289 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
293 e
->offset
= nth_midxed_offset(m
, pos
);
299 int fill_midx_entry(const struct object_id
*oid
, struct pack_entry
*e
, struct multi_pack_index
*m
)
303 if (!bsearch_midx(oid
, m
, &pos
))
306 return nth_midxed_pack_entry(m
, e
, pos
);
309 int midx_contains_pack(struct multi_pack_index
*m
, const char *idx_name
)
311 uint32_t first
= 0, last
= m
->num_packs
;
313 while (first
< last
) {
314 uint32_t mid
= first
+ (last
- first
) / 2;
318 current
= m
->pack_names
[mid
];
319 cmp
= strcmp(idx_name
, current
);
332 int prepare_multi_pack_index_one(struct repository
*r
, const char *object_dir
, int local
)
334 struct multi_pack_index
*m
;
335 struct multi_pack_index
*m_search
;
338 if (repo_config_get_bool(r
, "core.multipackindex", &config_value
) ||
342 for (m_search
= r
->objects
->multi_pack_index
; m_search
; m_search
= m_search
->next
)
343 if (!strcmp(object_dir
, m_search
->object_dir
))
346 m
= load_multi_pack_index(object_dir
, local
);
349 m
->next
= r
->objects
->multi_pack_index
;
350 r
->objects
->multi_pack_index
= m
;
357 static size_t write_midx_header(struct hashfile
*f
,
358 unsigned char num_chunks
,
361 unsigned char byte_values
[4];
363 hashwrite_be32(f
, MIDX_SIGNATURE
);
364 byte_values
[0] = MIDX_VERSION
;
365 byte_values
[1] = MIDX_HASH_VERSION
;
366 byte_values
[2] = num_chunks
;
367 byte_values
[3] = 0; /* unused */
368 hashwrite(f
, byte_values
, sizeof(byte_values
));
369 hashwrite_be32(f
, num_packs
);
371 return MIDX_HEADER_SIZE
;
375 struct packed_git
**list
;
379 uint32_t alloc_names
;
380 size_t pack_name_concat_len
;
381 struct multi_pack_index
*m
;
384 static void add_pack_to_midx(const char *full_path
, size_t full_path_len
,
385 const char *file_name
, void *data
)
387 struct pack_list
*packs
= (struct pack_list
*)data
;
389 if (ends_with(file_name
, ".idx")) {
390 if (packs
->m
&& midx_contains_pack(packs
->m
, file_name
))
393 ALLOC_GROW(packs
->list
, packs
->nr
+ 1, packs
->alloc_list
);
394 ALLOC_GROW(packs
->names
, packs
->nr
+ 1, packs
->alloc_names
);
396 packs
->list
[packs
->nr
] = add_packed_git(full_path
,
400 if (!packs
->list
[packs
->nr
]) {
401 warning(_("failed to add packfile '%s'"),
406 if (open_pack_index(packs
->list
[packs
->nr
])) {
407 warning(_("failed to open pack-index '%s'"),
409 close_pack(packs
->list
[packs
->nr
]);
410 FREE_AND_NULL(packs
->list
[packs
->nr
]);
414 packs
->names
[packs
->nr
] = xstrdup(file_name
);
415 packs
->pack_name_concat_len
+= strlen(file_name
) + 1;
421 uint32_t pack_int_id
;
425 static int pack_pair_compare(const void *_a
, const void *_b
)
427 struct pack_pair
*a
= (struct pack_pair
*)_a
;
428 struct pack_pair
*b
= (struct pack_pair
*)_b
;
429 return strcmp(a
->pack_name
, b
->pack_name
);
432 static void sort_packs_by_name(char **pack_names
, uint32_t nr_packs
, uint32_t *perm
)
435 struct pack_pair
*pairs
;
437 ALLOC_ARRAY(pairs
, nr_packs
);
439 for (i
= 0; i
< nr_packs
; i
++) {
440 pairs
[i
].pack_int_id
= i
;
441 pairs
[i
].pack_name
= pack_names
[i
];
444 QSORT(pairs
, nr_packs
, pack_pair_compare
);
446 for (i
= 0; i
< nr_packs
; i
++) {
447 pack_names
[i
] = pairs
[i
].pack_name
;
448 perm
[pairs
[i
].pack_int_id
] = i
;
454 struct pack_midx_entry
{
455 struct object_id oid
;
456 uint32_t pack_int_id
;
461 static int midx_oid_compare(const void *_a
, const void *_b
)
463 const struct pack_midx_entry
*a
= (const struct pack_midx_entry
*)_a
;
464 const struct pack_midx_entry
*b
= (const struct pack_midx_entry
*)_b
;
465 int cmp
= oidcmp(&a
->oid
, &b
->oid
);
470 if (a
->pack_mtime
> b
->pack_mtime
)
472 else if (a
->pack_mtime
< b
->pack_mtime
)
475 return a
->pack_int_id
- b
->pack_int_id
;
478 static int nth_midxed_pack_midx_entry(struct multi_pack_index
*m
,
480 struct pack_midx_entry
*e
,
483 if (pos
>= m
->num_objects
)
486 nth_midxed_object_oid(&e
->oid
, m
, pos
);
487 e
->pack_int_id
= pack_perm
[nth_midxed_pack_int_id(m
, pos
)];
488 e
->offset
= nth_midxed_offset(m
, pos
);
490 /* consider objects in midx to be from "old" packs */
495 static void fill_pack_entry(uint32_t pack_int_id
,
496 struct packed_git
*p
,
498 struct pack_midx_entry
*entry
)
500 if (!nth_packed_object_oid(&entry
->oid
, p
, cur_object
))
501 die(_("failed to locate object %d in packfile"), cur_object
);
503 entry
->pack_int_id
= pack_int_id
;
504 entry
->pack_mtime
= p
->mtime
;
506 entry
->offset
= nth_packed_object_offset(p
, cur_object
);
510 * It is possible to artificially get into a state where there are many
511 * duplicate copies of objects. That can create high memory pressure if
512 * we are to create a list of all objects before de-duplication. To reduce
513 * this memory pressure without a significant performance drop, automatically
514 * group objects by the first byte of their object id. Use the IDX fanout
515 * tables to group the data, copy to a local array, then sort.
517 * Copy only the de-duplicated entries (selected by most-recent modified time
518 * of a packfile containing the object).
520 static struct pack_midx_entry
*get_sorted_entries(struct multi_pack_index
*m
,
521 struct packed_git
**p
,
524 uint32_t *nr_objects
)
526 uint32_t cur_fanout
, cur_pack
, cur_object
;
527 uint32_t alloc_fanout
, alloc_objects
, total_objects
= 0;
528 struct pack_midx_entry
*entries_by_fanout
= NULL
;
529 struct pack_midx_entry
*deduplicated_entries
= NULL
;
530 uint32_t start_pack
= m
? m
->num_packs
: 0;
532 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++)
533 total_objects
+= p
[cur_pack
]->num_objects
;
536 * As we de-duplicate by fanout value, we expect the fanout
537 * slices to be evenly distributed, with some noise. Hence,
538 * allocate slightly more than one 256th.
540 alloc_objects
= alloc_fanout
= total_objects
> 3200 ? total_objects
/ 200 : 16;
542 ALLOC_ARRAY(entries_by_fanout
, alloc_fanout
);
543 ALLOC_ARRAY(deduplicated_entries
, alloc_objects
);
546 for (cur_fanout
= 0; cur_fanout
< 256; cur_fanout
++) {
547 uint32_t nr_fanout
= 0;
550 uint32_t start
= 0, end
;
553 start
= ntohl(m
->chunk_oid_fanout
[cur_fanout
- 1]);
554 end
= ntohl(m
->chunk_oid_fanout
[cur_fanout
]);
556 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
557 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
558 nth_midxed_pack_midx_entry(m
, perm
,
559 &entries_by_fanout
[nr_fanout
],
565 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++) {
566 uint32_t start
= 0, end
;
569 start
= get_pack_fanout(p
[cur_pack
], cur_fanout
- 1);
570 end
= get_pack_fanout(p
[cur_pack
], cur_fanout
);
572 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
573 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
574 fill_pack_entry(perm
[cur_pack
], p
[cur_pack
], cur_object
, &entries_by_fanout
[nr_fanout
]);
579 QSORT(entries_by_fanout
, nr_fanout
, midx_oid_compare
);
582 * The batch is now sorted by OID and then mtime (descending).
583 * Take only the first duplicate.
585 for (cur_object
= 0; cur_object
< nr_fanout
; cur_object
++) {
586 if (cur_object
&& !oidcmp(&entries_by_fanout
[cur_object
- 1].oid
,
587 &entries_by_fanout
[cur_object
].oid
))
590 ALLOC_GROW(deduplicated_entries
, *nr_objects
+ 1, alloc_objects
);
591 memcpy(&deduplicated_entries
[*nr_objects
],
592 &entries_by_fanout
[cur_object
],
593 sizeof(struct pack_midx_entry
));
598 free(entries_by_fanout
);
599 return deduplicated_entries
;
602 static size_t write_midx_pack_names(struct hashfile
*f
,
607 unsigned char padding
[MIDX_CHUNK_ALIGNMENT
];
610 for (i
= 0; i
< num_packs
; i
++) {
611 size_t writelen
= strlen(pack_names
[i
]) + 1;
613 if (i
&& strcmp(pack_names
[i
], pack_names
[i
- 1]) <= 0)
614 BUG("incorrect pack-file order: %s before %s",
618 hashwrite(f
, pack_names
[i
], writelen
);
622 /* add padding to be aligned */
623 i
= MIDX_CHUNK_ALIGNMENT
- (written
% MIDX_CHUNK_ALIGNMENT
);
624 if (i
< MIDX_CHUNK_ALIGNMENT
) {
625 memset(padding
, 0, sizeof(padding
));
626 hashwrite(f
, padding
, i
);
633 static size_t write_midx_oid_fanout(struct hashfile
*f
,
634 struct pack_midx_entry
*objects
,
637 struct pack_midx_entry
*list
= objects
;
638 struct pack_midx_entry
*last
= objects
+ nr_objects
;
643 * Write the first-level table (the list is sorted,
644 * but we use a 256-entry lookup to be able to avoid
645 * having to do eight extra binary search iterations).
647 for (i
= 0; i
< 256; i
++) {
648 struct pack_midx_entry
*next
= list
;
650 while (next
< last
&& next
->oid
.hash
[0] == i
) {
655 hashwrite_be32(f
, count
);
659 return MIDX_CHUNK_FANOUT_SIZE
;
662 static size_t write_midx_oid_lookup(struct hashfile
*f
, unsigned char hash_len
,
663 struct pack_midx_entry
*objects
,
666 struct pack_midx_entry
*list
= objects
;
670 for (i
= 0; i
< nr_objects
; i
++) {
671 struct pack_midx_entry
*obj
= list
++;
673 if (i
< nr_objects
- 1) {
674 struct pack_midx_entry
*next
= list
;
675 if (oidcmp(&obj
->oid
, &next
->oid
) >= 0)
676 BUG("OIDs not in order: %s >= %s",
677 oid_to_hex(&obj
->oid
),
678 oid_to_hex(&next
->oid
));
681 hashwrite(f
, obj
->oid
.hash
, (int)hash_len
);
688 static size_t write_midx_object_offsets(struct hashfile
*f
, int large_offset_needed
,
689 struct pack_midx_entry
*objects
, uint32_t nr_objects
)
691 struct pack_midx_entry
*list
= objects
;
692 uint32_t i
, nr_large_offset
= 0;
695 for (i
= 0; i
< nr_objects
; i
++) {
696 struct pack_midx_entry
*obj
= list
++;
698 hashwrite_be32(f
, obj
->pack_int_id
);
700 if (large_offset_needed
&& obj
->offset
>> 31)
701 hashwrite_be32(f
, MIDX_LARGE_OFFSET_NEEDED
| nr_large_offset
++);
702 else if (!large_offset_needed
&& obj
->offset
>> 32)
703 BUG("object %s requires a large offset (%"PRIx64
") but the MIDX is not writing large offsets!",
704 oid_to_hex(&obj
->oid
),
707 hashwrite_be32(f
, (uint32_t)obj
->offset
);
709 written
+= MIDX_CHUNK_OFFSET_WIDTH
;
715 static size_t write_midx_large_offsets(struct hashfile
*f
, uint32_t nr_large_offset
,
716 struct pack_midx_entry
*objects
, uint32_t nr_objects
)
718 struct pack_midx_entry
*list
= objects
;
721 while (nr_large_offset
) {
722 struct pack_midx_entry
*obj
= list
++;
723 uint64_t offset
= obj
->offset
;
728 hashwrite_be32(f
, offset
>> 32);
729 hashwrite_be32(f
, offset
& 0xffffffffUL
);
730 written
+= 2 * sizeof(uint32_t);
738 int write_midx_file(const char *object_dir
)
740 unsigned char cur_chunk
, num_chunks
= 0;
743 struct hashfile
*f
= NULL
;
745 struct pack_list packs
;
746 uint32_t *pack_perm
= NULL
;
747 uint64_t written
= 0;
748 uint32_t chunk_ids
[MIDX_MAX_CHUNKS
+ 1];
749 uint64_t chunk_offsets
[MIDX_MAX_CHUNKS
+ 1];
750 uint32_t nr_entries
, num_large_offsets
= 0;
751 struct pack_midx_entry
*entries
= NULL
;
752 int large_offsets_needed
= 0;
754 midx_name
= get_midx_filename(object_dir
);
755 if (safe_create_leading_directories(midx_name
)) {
757 die_errno(_("unable to create leading directories of %s"),
761 packs
.m
= load_multi_pack_index(object_dir
, 1);
764 packs
.alloc_list
= packs
.m
? packs
.m
->num_packs
: 16;
765 packs
.alloc_names
= packs
.alloc_list
;
768 packs
.pack_name_concat_len
= 0;
769 ALLOC_ARRAY(packs
.list
, packs
.alloc_list
);
770 ALLOC_ARRAY(packs
.names
, packs
.alloc_names
);
773 for (i
= 0; i
< packs
.m
->num_packs
; i
++) {
774 ALLOC_GROW(packs
.list
, packs
.nr
+ 1, packs
.alloc_list
);
775 ALLOC_GROW(packs
.names
, packs
.nr
+ 1, packs
.alloc_names
);
777 packs
.list
[packs
.nr
] = NULL
;
778 packs
.names
[packs
.nr
] = xstrdup(packs
.m
->pack_names
[i
]);
779 packs
.pack_name_concat_len
+= strlen(packs
.names
[packs
.nr
]) + 1;
784 for_each_file_in_pack_dir(object_dir
, add_pack_to_midx
, &packs
);
786 if (packs
.m
&& packs
.nr
== packs
.m
->num_packs
)
789 if (packs
.pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
)
790 packs
.pack_name_concat_len
+= MIDX_CHUNK_ALIGNMENT
-
791 (packs
.pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
);
793 ALLOC_ARRAY(pack_perm
, packs
.nr
);
794 sort_packs_by_name(packs
.names
, packs
.nr
, pack_perm
);
796 entries
= get_sorted_entries(packs
.m
, packs
.list
, pack_perm
, packs
.nr
, &nr_entries
);
798 for (i
= 0; i
< nr_entries
; i
++) {
799 if (entries
[i
].offset
> 0x7fffffff)
801 if (entries
[i
].offset
> 0xffffffff)
802 large_offsets_needed
= 1;
805 hold_lock_file_for_update(&lk
, midx_name
, LOCK_DIE_ON_ERROR
);
806 f
= hashfd(lk
.tempfile
->fd
, lk
.tempfile
->filename
.buf
);
807 FREE_AND_NULL(midx_name
);
813 num_chunks
= large_offsets_needed
? 5 : 4;
815 written
= write_midx_header(f
, num_chunks
, packs
.nr
);
817 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_PACKNAMES
;
818 chunk_offsets
[cur_chunk
] = written
+ (num_chunks
+ 1) * MIDX_CHUNKLOOKUP_WIDTH
;
821 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OIDFANOUT
;
822 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + packs
.pack_name_concat_len
;
825 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OIDLOOKUP
;
826 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + MIDX_CHUNK_FANOUT_SIZE
;
829 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OBJECTOFFSETS
;
830 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + nr_entries
* MIDX_HASH_LEN
;
833 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + nr_entries
* MIDX_CHUNK_OFFSET_WIDTH
;
834 if (large_offsets_needed
) {
835 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_LARGEOFFSETS
;
838 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] +
839 num_large_offsets
* MIDX_CHUNK_LARGE_OFFSET_WIDTH
;
842 chunk_ids
[cur_chunk
] = 0;
844 for (i
= 0; i
<= num_chunks
; i
++) {
845 if (i
&& chunk_offsets
[i
] < chunk_offsets
[i
- 1])
846 BUG("incorrect chunk offsets: %"PRIu64
" before %"PRIu64
,
847 chunk_offsets
[i
- 1],
850 if (chunk_offsets
[i
] % MIDX_CHUNK_ALIGNMENT
)
851 BUG("chunk offset %"PRIu64
" is not properly aligned",
854 hashwrite_be32(f
, chunk_ids
[i
]);
855 hashwrite_be32(f
, chunk_offsets
[i
] >> 32);
856 hashwrite_be32(f
, chunk_offsets
[i
]);
858 written
+= MIDX_CHUNKLOOKUP_WIDTH
;
861 for (i
= 0; i
< num_chunks
; i
++) {
862 if (written
!= chunk_offsets
[i
])
863 BUG("incorrect chunk offset (%"PRIu64
" != %"PRIu64
") for chunk id %"PRIx32
,
868 switch (chunk_ids
[i
]) {
869 case MIDX_CHUNKID_PACKNAMES
:
870 written
+= write_midx_pack_names(f
, packs
.names
, packs
.nr
);
873 case MIDX_CHUNKID_OIDFANOUT
:
874 written
+= write_midx_oid_fanout(f
, entries
, nr_entries
);
877 case MIDX_CHUNKID_OIDLOOKUP
:
878 written
+= write_midx_oid_lookup(f
, MIDX_HASH_LEN
, entries
, nr_entries
);
881 case MIDX_CHUNKID_OBJECTOFFSETS
:
882 written
+= write_midx_object_offsets(f
, large_offsets_needed
, entries
, nr_entries
);
885 case MIDX_CHUNKID_LARGEOFFSETS
:
886 written
+= write_midx_large_offsets(f
, num_large_offsets
, entries
, nr_entries
);
890 BUG("trying to write unknown chunk id %"PRIx32
,
895 if (written
!= chunk_offsets
[num_chunks
])
896 BUG("incorrect final offset %"PRIu64
" != %"PRIu64
,
898 chunk_offsets
[num_chunks
]);
900 finalize_hashfile(f
, NULL
, CSUM_FSYNC
| CSUM_HASH_IN_STREAM
);
901 commit_lock_file(&lk
);
904 for (i
= 0; i
< packs
.nr
; i
++) {
906 close_pack(packs
.list
[i
]);
909 free(packs
.names
[i
]);
920 void clear_midx_file(const char *object_dir
)
922 char *midx
= get_midx_filename(object_dir
);
924 if (remove_path(midx
)) {
926 die(_("failed to clear multi-pack-index at %s"), midx
);