7 #include "object-store.h"
8 #include "sha1-lookup.h"
12 #include "run-command.h"
13 #include "repository.h"
15 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
16 #define MIDX_VERSION 1
17 #define MIDX_BYTE_FILE_VERSION 4
18 #define MIDX_BYTE_HASH_VERSION 5
19 #define MIDX_BYTE_NUM_CHUNKS 6
20 #define MIDX_BYTE_NUM_PACKS 8
21 #define MIDX_HEADER_SIZE 12
22 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
24 #define MIDX_MAX_CHUNKS 5
25 #define MIDX_CHUNK_ALIGNMENT 4
26 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
27 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
28 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
29 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
30 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
31 #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
32 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
33 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
34 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
35 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
37 #define PACK_EXPIRED UINT_MAX
39 static uint8_t oid_version(void)
41 switch (hash_algo_by_ptr(the_hash_algo
)) {
47 die(_("invalid hash version"));
51 static char *get_midx_filename(const char *object_dir
)
53 return xstrfmt("%s/pack/multi-pack-index", object_dir
);
56 struct multi_pack_index
*load_multi_pack_index(const char *object_dir
, int local
)
58 struct multi_pack_index
*m
= NULL
;
62 void *midx_map
= NULL
;
63 uint32_t hash_version
;
64 char *midx_name
= get_midx_filename(object_dir
);
66 const char *cur_pack_name
;
68 fd
= git_open(midx_name
);
73 error_errno(_("failed to read %s"), midx_name
);
77 midx_size
= xsize_t(st
.st_size
);
79 if (midx_size
< MIDX_MIN_SIZE
) {
80 error(_("multi-pack-index file %s is too small"), midx_name
);
84 FREE_AND_NULL(midx_name
);
86 midx_map
= xmmap(NULL
, midx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
89 FLEX_ALLOC_STR(m
, object_dir
, object_dir
);
91 m
->data_len
= midx_size
;
94 m
->signature
= get_be32(m
->data
);
95 if (m
->signature
!= MIDX_SIGNATURE
)
96 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
97 m
->signature
, MIDX_SIGNATURE
);
99 m
->version
= m
->data
[MIDX_BYTE_FILE_VERSION
];
100 if (m
->version
!= MIDX_VERSION
)
101 die(_("multi-pack-index version %d not recognized"),
104 hash_version
= m
->data
[MIDX_BYTE_HASH_VERSION
];
105 if (hash_version
!= oid_version()) {
106 error(_("multi-pack-index hash version %u does not match version %u"),
107 hash_version
, oid_version());
110 m
->hash_len
= the_hash_algo
->rawsz
;
112 m
->num_chunks
= m
->data
[MIDX_BYTE_NUM_CHUNKS
];
114 m
->num_packs
= get_be32(m
->data
+ MIDX_BYTE_NUM_PACKS
);
116 for (i
= 0; i
< m
->num_chunks
; i
++) {
117 uint32_t chunk_id
= get_be32(m
->data
+ MIDX_HEADER_SIZE
+
118 MIDX_CHUNKLOOKUP_WIDTH
* i
);
119 uint64_t chunk_offset
= get_be64(m
->data
+ MIDX_HEADER_SIZE
+ 4 +
120 MIDX_CHUNKLOOKUP_WIDTH
* i
);
122 if (chunk_offset
>= m
->data_len
)
123 die(_("invalid chunk offset (too large)"));
126 case MIDX_CHUNKID_PACKNAMES
:
127 m
->chunk_pack_names
= m
->data
+ chunk_offset
;
130 case MIDX_CHUNKID_OIDFANOUT
:
131 m
->chunk_oid_fanout
= (uint32_t *)(m
->data
+ chunk_offset
);
134 case MIDX_CHUNKID_OIDLOOKUP
:
135 m
->chunk_oid_lookup
= m
->data
+ chunk_offset
;
138 case MIDX_CHUNKID_OBJECTOFFSETS
:
139 m
->chunk_object_offsets
= m
->data
+ chunk_offset
;
142 case MIDX_CHUNKID_LARGEOFFSETS
:
143 m
->chunk_large_offsets
= m
->data
+ chunk_offset
;
147 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
152 * Do nothing on unrecognized chunks, allowing future
153 * extensions to add optional chunks.
159 if (!m
->chunk_pack_names
)
160 die(_("multi-pack-index missing required pack-name chunk"));
161 if (!m
->chunk_oid_fanout
)
162 die(_("multi-pack-index missing required OID fanout chunk"));
163 if (!m
->chunk_oid_lookup
)
164 die(_("multi-pack-index missing required OID lookup chunk"));
165 if (!m
->chunk_object_offsets
)
166 die(_("multi-pack-index missing required object offsets chunk"));
168 m
->num_objects
= ntohl(m
->chunk_oid_fanout
[255]);
170 m
->pack_names
= xcalloc(m
->num_packs
, sizeof(*m
->pack_names
));
171 m
->packs
= xcalloc(m
->num_packs
, sizeof(*m
->packs
));
173 cur_pack_name
= (const char *)m
->chunk_pack_names
;
174 for (i
= 0; i
< m
->num_packs
; i
++) {
175 m
->pack_names
[i
] = cur_pack_name
;
177 cur_pack_name
+= strlen(cur_pack_name
) + 1;
179 if (i
&& strcmp(m
->pack_names
[i
], m
->pack_names
[i
- 1]) <= 0)
180 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
181 m
->pack_names
[i
- 1],
185 trace2_data_intmax("midx", the_repository
, "load/num_packs", m
->num_packs
);
186 trace2_data_intmax("midx", the_repository
, "load/num_objects", m
->num_objects
);
194 munmap(midx_map
, midx_size
);
200 void close_midx(struct multi_pack_index
*m
)
207 munmap((unsigned char *)m
->data
, m
->data_len
);
209 for (i
= 0; i
< m
->num_packs
; i
++) {
211 m
->packs
[i
]->multi_pack_index
= 0;
213 FREE_AND_NULL(m
->packs
);
214 FREE_AND_NULL(m
->pack_names
);
217 int prepare_midx_pack(struct repository
*r
, struct multi_pack_index
*m
, uint32_t pack_int_id
)
219 struct strbuf pack_name
= STRBUF_INIT
;
220 struct packed_git
*p
;
222 if (pack_int_id
>= m
->num_packs
)
223 die(_("bad pack-int-id: %u (%u total packs)"),
224 pack_int_id
, m
->num_packs
);
226 if (m
->packs
[pack_int_id
])
229 strbuf_addf(&pack_name
, "%s/pack/%s", m
->object_dir
,
230 m
->pack_names
[pack_int_id
]);
232 p
= add_packed_git(pack_name
.buf
, pack_name
.len
, m
->local
);
233 strbuf_release(&pack_name
);
238 p
->multi_pack_index
= 1;
239 m
->packs
[pack_int_id
] = p
;
240 install_packed_git(r
, p
);
241 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
246 int bsearch_midx(const struct object_id
*oid
, struct multi_pack_index
*m
, uint32_t *result
)
248 return bsearch_hash(oid
->hash
, m
->chunk_oid_fanout
, m
->chunk_oid_lookup
,
249 the_hash_algo
->rawsz
, result
);
252 struct object_id
*nth_midxed_object_oid(struct object_id
*oid
,
253 struct multi_pack_index
*m
,
256 if (n
>= m
->num_objects
)
259 hashcpy(oid
->hash
, m
->chunk_oid_lookup
+ m
->hash_len
* n
);
263 static off_t
nth_midxed_offset(struct multi_pack_index
*m
, uint32_t pos
)
265 const unsigned char *offset_data
;
268 offset_data
= m
->chunk_object_offsets
+ pos
* MIDX_CHUNK_OFFSET_WIDTH
;
269 offset32
= get_be32(offset_data
+ sizeof(uint32_t));
271 if (m
->chunk_large_offsets
&& offset32
& MIDX_LARGE_OFFSET_NEEDED
) {
272 if (sizeof(off_t
) < sizeof(uint64_t))
273 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
275 offset32
^= MIDX_LARGE_OFFSET_NEEDED
;
276 return get_be64(m
->chunk_large_offsets
+ sizeof(uint64_t) * offset32
);
282 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index
*m
, uint32_t pos
)
284 return get_be32(m
->chunk_object_offsets
+ pos
* MIDX_CHUNK_OFFSET_WIDTH
);
287 static int nth_midxed_pack_entry(struct repository
*r
,
288 struct multi_pack_index
*m
,
289 struct pack_entry
*e
,
292 uint32_t pack_int_id
;
293 struct packed_git
*p
;
295 if (pos
>= m
->num_objects
)
298 pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
300 if (prepare_midx_pack(r
, m
, pack_int_id
))
302 p
= m
->packs
[pack_int_id
];
305 * We are about to tell the caller where they can locate the
306 * requested object. We better make sure the packfile is
307 * still here and can be accessed before supplying that
308 * answer, as it may have been deleted since the MIDX was
311 if (!is_pack_valid(p
))
314 if (p
->num_bad_objects
) {
316 struct object_id oid
;
317 nth_midxed_object_oid(&oid
, m
, pos
);
318 for (i
= 0; i
< p
->num_bad_objects
; i
++)
320 p
->bad_object_sha1
+ the_hash_algo
->rawsz
* i
))
324 e
->offset
= nth_midxed_offset(m
, pos
);
330 int fill_midx_entry(struct repository
* r
,
331 const struct object_id
*oid
,
332 struct pack_entry
*e
,
333 struct multi_pack_index
*m
)
337 if (!bsearch_midx(oid
, m
, &pos
))
340 return nth_midxed_pack_entry(r
, m
, e
, pos
);
343 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
344 static int cmp_idx_or_pack_name(const char *idx_or_pack_name
,
345 const char *idx_name
)
347 /* Skip past any initial matching prefix. */
348 while (*idx_name
&& *idx_name
== *idx_or_pack_name
) {
354 * If we didn't match completely, we may have matched "pack-1234." and
355 * be left with "idx" and "pack" respectively, which is also OK. We do
356 * not have to check for "idx" and "idx", because that would have been
357 * a complete match (and in that case these strcmps will be false, but
358 * we'll correctly return 0 from the final strcmp() below.
360 * Technically this matches "fooidx" and "foopack", but we'd never have
361 * such names in the first place.
363 if (!strcmp(idx_name
, "idx") && !strcmp(idx_or_pack_name
, "pack"))
367 * This not only checks for a complete match, but also orders based on
368 * the first non-identical character, which means our ordering will
369 * match a raw strcmp(). That makes it OK to use this to binary search
370 * a naively-sorted list.
372 return strcmp(idx_or_pack_name
, idx_name
);
375 int midx_contains_pack(struct multi_pack_index
*m
, const char *idx_or_pack_name
)
377 uint32_t first
= 0, last
= m
->num_packs
;
379 while (first
< last
) {
380 uint32_t mid
= first
+ (last
- first
) / 2;
384 current
= m
->pack_names
[mid
];
385 cmp
= cmp_idx_or_pack_name(idx_or_pack_name
, current
);
398 int prepare_multi_pack_index_one(struct repository
*r
, const char *object_dir
, int local
)
400 struct multi_pack_index
*m
;
401 struct multi_pack_index
*m_search
;
403 prepare_repo_settings(r
);
404 if (!r
->settings
.core_multi_pack_index
)
407 for (m_search
= r
->objects
->multi_pack_index
; m_search
; m_search
= m_search
->next
)
408 if (!strcmp(object_dir
, m_search
->object_dir
))
411 m
= load_multi_pack_index(object_dir
, local
);
414 struct multi_pack_index
*mp
= r
->objects
->multi_pack_index
;
419 r
->objects
->multi_pack_index
= m
;
426 static size_t write_midx_header(struct hashfile
*f
,
427 unsigned char num_chunks
,
430 hashwrite_be32(f
, MIDX_SIGNATURE
);
431 hashwrite_u8(f
, MIDX_VERSION
);
432 hashwrite_u8(f
, oid_version());
433 hashwrite_u8(f
, num_chunks
);
434 hashwrite_u8(f
, 0); /* unused */
435 hashwrite_be32(f
, num_packs
);
437 return MIDX_HEADER_SIZE
;
441 uint32_t orig_pack_int_id
;
443 struct packed_git
*p
;
444 unsigned expired
: 1;
447 static int pack_info_compare(const void *_a
, const void *_b
)
449 struct pack_info
*a
= (struct pack_info
*)_a
;
450 struct pack_info
*b
= (struct pack_info
*)_b
;
451 return strcmp(a
->pack_name
, b
->pack_name
);
454 struct write_midx_context
{
455 struct pack_info
*info
;
458 struct multi_pack_index
*m
;
459 struct progress
*progress
;
460 unsigned pack_paths_checked
;
463 static void add_pack_to_midx(const char *full_path
, size_t full_path_len
,
464 const char *file_name
, void *data
)
466 struct write_midx_context
*ctx
= data
;
468 if (ends_with(file_name
, ".idx")) {
469 display_progress(ctx
->progress
, ++ctx
->pack_paths_checked
);
470 if (ctx
->m
&& midx_contains_pack(ctx
->m
, file_name
))
473 ALLOC_GROW(ctx
->info
, ctx
->nr
+ 1, ctx
->alloc
);
475 ctx
->info
[ctx
->nr
].p
= add_packed_git(full_path
,
479 if (!ctx
->info
[ctx
->nr
].p
) {
480 warning(_("failed to add packfile '%s'"),
485 if (open_pack_index(ctx
->info
[ctx
->nr
].p
)) {
486 warning(_("failed to open pack-index '%s'"),
488 close_pack(ctx
->info
[ctx
->nr
].p
);
489 FREE_AND_NULL(ctx
->info
[ctx
->nr
].p
);
493 ctx
->info
[ctx
->nr
].pack_name
= xstrdup(file_name
);
494 ctx
->info
[ctx
->nr
].orig_pack_int_id
= ctx
->nr
;
495 ctx
->info
[ctx
->nr
].expired
= 0;
500 struct pack_midx_entry
{
501 struct object_id oid
;
502 uint32_t pack_int_id
;
507 static int midx_oid_compare(const void *_a
, const void *_b
)
509 const struct pack_midx_entry
*a
= (const struct pack_midx_entry
*)_a
;
510 const struct pack_midx_entry
*b
= (const struct pack_midx_entry
*)_b
;
511 int cmp
= oidcmp(&a
->oid
, &b
->oid
);
516 if (a
->pack_mtime
> b
->pack_mtime
)
518 else if (a
->pack_mtime
< b
->pack_mtime
)
521 return a
->pack_int_id
- b
->pack_int_id
;
524 static int nth_midxed_pack_midx_entry(struct multi_pack_index
*m
,
525 struct pack_midx_entry
*e
,
528 if (pos
>= m
->num_objects
)
531 nth_midxed_object_oid(&e
->oid
, m
, pos
);
532 e
->pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
533 e
->offset
= nth_midxed_offset(m
, pos
);
535 /* consider objects in midx to be from "old" packs */
540 static void fill_pack_entry(uint32_t pack_int_id
,
541 struct packed_git
*p
,
543 struct pack_midx_entry
*entry
)
545 if (nth_packed_object_id(&entry
->oid
, p
, cur_object
) < 0)
546 die(_("failed to locate object %d in packfile"), cur_object
);
548 entry
->pack_int_id
= pack_int_id
;
549 entry
->pack_mtime
= p
->mtime
;
551 entry
->offset
= nth_packed_object_offset(p
, cur_object
);
555 * It is possible to artificially get into a state where there are many
556 * duplicate copies of objects. That can create high memory pressure if
557 * we are to create a list of all objects before de-duplication. To reduce
558 * this memory pressure without a significant performance drop, automatically
559 * group objects by the first byte of their object id. Use the IDX fanout
560 * tables to group the data, copy to a local array, then sort.
562 * Copy only the de-duplicated entries (selected by most-recent modified time
563 * of a packfile containing the object).
565 static struct pack_midx_entry
*get_sorted_entries(struct multi_pack_index
*m
,
566 struct pack_info
*info
,
568 uint32_t *nr_objects
)
570 uint32_t cur_fanout
, cur_pack
, cur_object
;
571 uint32_t alloc_fanout
, alloc_objects
, total_objects
= 0;
572 struct pack_midx_entry
*entries_by_fanout
= NULL
;
573 struct pack_midx_entry
*deduplicated_entries
= NULL
;
574 uint32_t start_pack
= m
? m
->num_packs
: 0;
576 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++)
577 total_objects
+= info
[cur_pack
].p
->num_objects
;
580 * As we de-duplicate by fanout value, we expect the fanout
581 * slices to be evenly distributed, with some noise. Hence,
582 * allocate slightly more than one 256th.
584 alloc_objects
= alloc_fanout
= total_objects
> 3200 ? total_objects
/ 200 : 16;
586 ALLOC_ARRAY(entries_by_fanout
, alloc_fanout
);
587 ALLOC_ARRAY(deduplicated_entries
, alloc_objects
);
590 for (cur_fanout
= 0; cur_fanout
< 256; cur_fanout
++) {
591 uint32_t nr_fanout
= 0;
594 uint32_t start
= 0, end
;
597 start
= ntohl(m
->chunk_oid_fanout
[cur_fanout
- 1]);
598 end
= ntohl(m
->chunk_oid_fanout
[cur_fanout
]);
600 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
601 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
602 nth_midxed_pack_midx_entry(m
,
603 &entries_by_fanout
[nr_fanout
],
609 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++) {
610 uint32_t start
= 0, end
;
613 start
= get_pack_fanout(info
[cur_pack
].p
, cur_fanout
- 1);
614 end
= get_pack_fanout(info
[cur_pack
].p
, cur_fanout
);
616 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
617 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
618 fill_pack_entry(cur_pack
, info
[cur_pack
].p
, cur_object
, &entries_by_fanout
[nr_fanout
]);
623 QSORT(entries_by_fanout
, nr_fanout
, midx_oid_compare
);
626 * The batch is now sorted by OID and then mtime (descending).
627 * Take only the first duplicate.
629 for (cur_object
= 0; cur_object
< nr_fanout
; cur_object
++) {
630 if (cur_object
&& oideq(&entries_by_fanout
[cur_object
- 1].oid
,
631 &entries_by_fanout
[cur_object
].oid
))
634 ALLOC_GROW(deduplicated_entries
, *nr_objects
+ 1, alloc_objects
);
635 memcpy(&deduplicated_entries
[*nr_objects
],
636 &entries_by_fanout
[cur_object
],
637 sizeof(struct pack_midx_entry
));
642 free(entries_by_fanout
);
643 return deduplicated_entries
;
646 static size_t write_midx_pack_names(struct hashfile
*f
, void *data
)
648 struct write_midx_context
*ctx
= data
;
650 unsigned char padding
[MIDX_CHUNK_ALIGNMENT
];
653 for (i
= 0; i
< ctx
->nr
; i
++) {
656 if (ctx
->info
[i
].expired
)
659 if (i
&& strcmp(ctx
->info
[i
].pack_name
, ctx
->info
[i
- 1].pack_name
) <= 0)
660 BUG("incorrect pack-file order: %s before %s",
661 ctx
->info
[i
- 1].pack_name
,
662 ctx
->info
[i
].pack_name
);
664 writelen
= strlen(ctx
->info
[i
].pack_name
) + 1;
665 hashwrite(f
, ctx
->info
[i
].pack_name
, writelen
);
669 /* add padding to be aligned */
670 i
= MIDX_CHUNK_ALIGNMENT
- (written
% MIDX_CHUNK_ALIGNMENT
);
671 if (i
< MIDX_CHUNK_ALIGNMENT
) {
672 memset(padding
, 0, sizeof(padding
));
673 hashwrite(f
, padding
, i
);
680 static size_t write_midx_oid_fanout(struct hashfile
*f
,
681 struct pack_midx_entry
*objects
,
684 struct pack_midx_entry
*list
= objects
;
685 struct pack_midx_entry
*last
= objects
+ nr_objects
;
690 * Write the first-level table (the list is sorted,
691 * but we use a 256-entry lookup to be able to avoid
692 * having to do eight extra binary search iterations).
694 for (i
= 0; i
< 256; i
++) {
695 struct pack_midx_entry
*next
= list
;
697 while (next
< last
&& next
->oid
.hash
[0] == i
) {
702 hashwrite_be32(f
, count
);
706 return MIDX_CHUNK_FANOUT_SIZE
;
709 static size_t write_midx_oid_lookup(struct hashfile
*f
, unsigned char hash_len
,
710 struct pack_midx_entry
*objects
,
713 struct pack_midx_entry
*list
= objects
;
717 for (i
= 0; i
< nr_objects
; i
++) {
718 struct pack_midx_entry
*obj
= list
++;
720 if (i
< nr_objects
- 1) {
721 struct pack_midx_entry
*next
= list
;
722 if (oidcmp(&obj
->oid
, &next
->oid
) >= 0)
723 BUG("OIDs not in order: %s >= %s",
724 oid_to_hex(&obj
->oid
),
725 oid_to_hex(&next
->oid
));
728 hashwrite(f
, obj
->oid
.hash
, (int)hash_len
);
735 static size_t write_midx_object_offsets(struct hashfile
*f
, int large_offset_needed
,
737 struct pack_midx_entry
*objects
, uint32_t nr_objects
)
739 struct pack_midx_entry
*list
= objects
;
740 uint32_t i
, nr_large_offset
= 0;
743 for (i
= 0; i
< nr_objects
; i
++) {
744 struct pack_midx_entry
*obj
= list
++;
746 if (perm
[obj
->pack_int_id
] == PACK_EXPIRED
)
747 BUG("object %s is in an expired pack with int-id %d",
748 oid_to_hex(&obj
->oid
),
751 hashwrite_be32(f
, perm
[obj
->pack_int_id
]);
753 if (large_offset_needed
&& obj
->offset
>> 31)
754 hashwrite_be32(f
, MIDX_LARGE_OFFSET_NEEDED
| nr_large_offset
++);
755 else if (!large_offset_needed
&& obj
->offset
>> 32)
756 BUG("object %s requires a large offset (%"PRIx64
") but the MIDX is not writing large offsets!",
757 oid_to_hex(&obj
->oid
),
760 hashwrite_be32(f
, (uint32_t)obj
->offset
);
762 written
+= MIDX_CHUNK_OFFSET_WIDTH
;
768 static size_t write_midx_large_offsets(struct hashfile
*f
, uint32_t nr_large_offset
,
769 struct pack_midx_entry
*objects
, uint32_t nr_objects
)
771 struct pack_midx_entry
*list
= objects
, *end
= objects
+ nr_objects
;
774 while (nr_large_offset
) {
775 struct pack_midx_entry
*obj
;
779 BUG("too many large-offset objects");
782 offset
= obj
->offset
;
787 written
+= hashwrite_be64(f
, offset
);
795 static int write_midx_internal(const char *object_dir
, struct multi_pack_index
*m
,
796 struct string_list
*packs_to_drop
, unsigned flags
)
798 unsigned char cur_chunk
, num_chunks
= 0;
801 struct hashfile
*f
= NULL
;
803 struct write_midx_context ctx
= { 0 };
804 uint32_t *pack_perm
= NULL
;
805 uint64_t written
= 0;
806 uint32_t chunk_ids
[MIDX_MAX_CHUNKS
+ 1];
807 uint64_t chunk_offsets
[MIDX_MAX_CHUNKS
+ 1];
808 uint32_t nr_entries
, num_large_offsets
= 0;
809 struct pack_midx_entry
*entries
= NULL
;
810 struct progress
*progress
= NULL
;
811 int large_offsets_needed
= 0;
812 int pack_name_concat_len
= 0;
813 int dropped_packs
= 0;
816 midx_name
= get_midx_filename(object_dir
);
817 if (safe_create_leading_directories(midx_name
))
818 die_errno(_("unable to create leading directories of %s"),
824 ctx
.m
= load_multi_pack_index(object_dir
, 1);
827 ctx
.alloc
= ctx
.m
? ctx
.m
->num_packs
: 16;
829 ALLOC_ARRAY(ctx
.info
, ctx
.alloc
);
832 for (i
= 0; i
< ctx
.m
->num_packs
; i
++) {
833 ALLOC_GROW(ctx
.info
, ctx
.nr
+ 1, ctx
.alloc
);
835 ctx
.info
[ctx
.nr
].orig_pack_int_id
= i
;
836 ctx
.info
[ctx
.nr
].pack_name
= xstrdup(ctx
.m
->pack_names
[i
]);
837 ctx
.info
[ctx
.nr
].p
= NULL
;
838 ctx
.info
[ctx
.nr
].expired
= 0;
843 ctx
.pack_paths_checked
= 0;
844 if (flags
& MIDX_PROGRESS
)
845 ctx
.progress
= start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
849 for_each_file_in_pack_dir(object_dir
, add_pack_to_midx
, &ctx
);
850 stop_progress(&ctx
.progress
);
852 if (ctx
.m
&& ctx
.nr
== ctx
.m
->num_packs
&& !packs_to_drop
)
855 entries
= get_sorted_entries(ctx
.m
, ctx
.info
, ctx
.nr
, &nr_entries
);
857 for (i
= 0; i
< nr_entries
; i
++) {
858 if (entries
[i
].offset
> 0x7fffffff)
860 if (entries
[i
].offset
> 0xffffffff)
861 large_offsets_needed
= 1;
864 QSORT(ctx
.info
, ctx
.nr
, pack_info_compare
);
866 if (packs_to_drop
&& packs_to_drop
->nr
) {
868 int missing_drops
= 0;
870 for (i
= 0; i
< ctx
.nr
&& drop_index
< packs_to_drop
->nr
; i
++) {
871 int cmp
= strcmp(ctx
.info
[i
].pack_name
,
872 packs_to_drop
->items
[drop_index
].string
);
876 ctx
.info
[i
].expired
= 1;
877 } else if (cmp
> 0) {
878 error(_("did not see pack-file %s to drop"),
879 packs_to_drop
->items
[drop_index
].string
);
884 ctx
.info
[i
].expired
= 0;
895 * pack_perm stores a permutation between pack-int-ids from the
896 * previous multi-pack-index to the new one we are writing:
898 * pack_perm[old_id] = new_id
900 ALLOC_ARRAY(pack_perm
, ctx
.nr
);
901 for (i
= 0; i
< ctx
.nr
; i
++) {
902 if (ctx
.info
[i
].expired
) {
904 pack_perm
[ctx
.info
[i
].orig_pack_int_id
] = PACK_EXPIRED
;
906 pack_perm
[ctx
.info
[i
].orig_pack_int_id
] = i
- dropped_packs
;
910 for (i
= 0; i
< ctx
.nr
; i
++) {
911 if (!ctx
.info
[i
].expired
)
912 pack_name_concat_len
+= strlen(ctx
.info
[i
].pack_name
) + 1;
915 if (pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
)
916 pack_name_concat_len
+= MIDX_CHUNK_ALIGNMENT
-
917 (pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
);
919 hold_lock_file_for_update(&lk
, midx_name
, LOCK_DIE_ON_ERROR
);
920 f
= hashfd(lk
.tempfile
->fd
, lk
.tempfile
->filename
.buf
);
921 FREE_AND_NULL(midx_name
);
927 num_chunks
= large_offsets_needed
? 5 : 4;
929 if (ctx
.nr
- dropped_packs
== 0) {
930 error(_("no pack files to index."));
935 written
= write_midx_header(f
, num_chunks
, ctx
.nr
- dropped_packs
);
937 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_PACKNAMES
;
938 chunk_offsets
[cur_chunk
] = written
+ (num_chunks
+ 1) * MIDX_CHUNKLOOKUP_WIDTH
;
941 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OIDFANOUT
;
942 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + pack_name_concat_len
;
945 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OIDLOOKUP
;
946 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + MIDX_CHUNK_FANOUT_SIZE
;
949 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_OBJECTOFFSETS
;
950 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + nr_entries
* the_hash_algo
->rawsz
;
953 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] + nr_entries
* MIDX_CHUNK_OFFSET_WIDTH
;
954 if (large_offsets_needed
) {
955 chunk_ids
[cur_chunk
] = MIDX_CHUNKID_LARGEOFFSETS
;
958 chunk_offsets
[cur_chunk
] = chunk_offsets
[cur_chunk
- 1] +
959 num_large_offsets
* MIDX_CHUNK_LARGE_OFFSET_WIDTH
;
962 chunk_ids
[cur_chunk
] = 0;
964 for (i
= 0; i
<= num_chunks
; i
++) {
965 if (i
&& chunk_offsets
[i
] < chunk_offsets
[i
- 1])
966 BUG("incorrect chunk offsets: %"PRIu64
" before %"PRIu64
,
967 chunk_offsets
[i
- 1],
970 if (chunk_offsets
[i
] % MIDX_CHUNK_ALIGNMENT
)
971 BUG("chunk offset %"PRIu64
" is not properly aligned",
974 hashwrite_be32(f
, chunk_ids
[i
]);
975 hashwrite_be64(f
, chunk_offsets
[i
]);
977 written
+= MIDX_CHUNKLOOKUP_WIDTH
;
980 if (flags
& MIDX_PROGRESS
)
981 progress
= start_delayed_progress(_("Writing chunks to multi-pack-index"),
983 for (i
= 0; i
< num_chunks
; i
++) {
984 if (written
!= chunk_offsets
[i
])
985 BUG("incorrect chunk offset (%"PRIu64
" != %"PRIu64
") for chunk id %"PRIx32
,
990 switch (chunk_ids
[i
]) {
991 case MIDX_CHUNKID_PACKNAMES
:
992 written
+= write_midx_pack_names(f
, &ctx
);
995 case MIDX_CHUNKID_OIDFANOUT
:
996 written
+= write_midx_oid_fanout(f
, entries
, nr_entries
);
999 case MIDX_CHUNKID_OIDLOOKUP
:
1000 written
+= write_midx_oid_lookup(f
, the_hash_algo
->rawsz
, entries
, nr_entries
);
1003 case MIDX_CHUNKID_OBJECTOFFSETS
:
1004 written
+= write_midx_object_offsets(f
, large_offsets_needed
, pack_perm
, entries
, nr_entries
);
1007 case MIDX_CHUNKID_LARGEOFFSETS
:
1008 written
+= write_midx_large_offsets(f
, num_large_offsets
, entries
, nr_entries
);
1012 BUG("trying to write unknown chunk id %"PRIx32
,
1016 display_progress(progress
, i
+ 1);
1018 stop_progress(&progress
);
1020 if (written
!= chunk_offsets
[num_chunks
])
1021 BUG("incorrect final offset %"PRIu64
" != %"PRIu64
,
1023 chunk_offsets
[num_chunks
]);
1025 finalize_hashfile(f
, NULL
, CSUM_FSYNC
| CSUM_HASH_IN_STREAM
);
1026 commit_lock_file(&lk
);
1029 for (i
= 0; i
< ctx
.nr
; i
++) {
1030 if (ctx
.info
[i
].p
) {
1031 close_pack(ctx
.info
[i
].p
);
1032 free(ctx
.info
[i
].p
);
1034 free(ctx
.info
[i
].pack_name
);
1044 int write_midx_file(const char *object_dir
, unsigned flags
)
1046 return write_midx_internal(object_dir
, NULL
, NULL
, flags
);
1049 void clear_midx_file(struct repository
*r
)
1051 char *midx
= get_midx_filename(r
->objects
->odb
->path
);
1053 if (r
->objects
&& r
->objects
->multi_pack_index
) {
1054 close_midx(r
->objects
->multi_pack_index
);
1055 r
->objects
->multi_pack_index
= NULL
;
1058 if (remove_path(midx
))
1059 die(_("failed to clear multi-pack-index at %s"), midx
);
1064 static int verify_midx_error
;
1066 static void midx_report(const char *fmt
, ...)
1069 verify_midx_error
= 1;
1071 vfprintf(stderr
, fmt
, ap
);
1072 fprintf(stderr
, "\n");
1076 struct pair_pos_vs_id
1079 uint32_t pack_int_id
;
1082 static int compare_pair_pos_vs_id(const void *_a
, const void *_b
)
1084 struct pair_pos_vs_id
*a
= (struct pair_pos_vs_id
*)_a
;
1085 struct pair_pos_vs_id
*b
= (struct pair_pos_vs_id
*)_b
;
1087 return b
->pack_int_id
- a
->pack_int_id
;
1091 * Limit calls to display_progress() for performance reasons.
1092 * The interval here was arbitrarily chosen.
1094 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1095 #define midx_display_sparse_progress(progress, n) \
1097 uint64_t _n = (n); \
1098 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1099 display_progress(progress, _n); \
1102 int verify_midx_file(struct repository
*r
, const char *object_dir
, unsigned flags
)
1104 struct pair_pos_vs_id
*pairs
= NULL
;
1106 struct progress
*progress
= NULL
;
1107 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
1108 verify_midx_error
= 0;
1113 char *filename
= get_midx_filename(object_dir
);
1114 if (!stat(filename
, &sb
)) {
1115 error(_("multi-pack-index file exists, but failed to parse"));
1122 if (flags
& MIDX_PROGRESS
)
1123 progress
= start_delayed_progress(_("Looking for referenced packfiles"),
1125 for (i
= 0; i
< m
->num_packs
; i
++) {
1126 if (prepare_midx_pack(r
, m
, i
))
1127 midx_report("failed to load pack in position %d", i
);
1129 display_progress(progress
, i
+ 1);
1131 stop_progress(&progress
);
1133 for (i
= 0; i
< 255; i
++) {
1134 uint32_t oid_fanout1
= ntohl(m
->chunk_oid_fanout
[i
]);
1135 uint32_t oid_fanout2
= ntohl(m
->chunk_oid_fanout
[i
+ 1]);
1137 if (oid_fanout1
> oid_fanout2
)
1138 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32
" > %"PRIx32
" = fanout[%d]"),
1139 i
, oid_fanout1
, oid_fanout2
, i
+ 1);
1142 if (m
->num_objects
== 0) {
1143 midx_report(_("the midx contains no oid"));
1145 * Remaining tests assume that we have objects, so we can
1148 return verify_midx_error
;
1151 if (flags
& MIDX_PROGRESS
)
1152 progress
= start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1153 m
->num_objects
- 1);
1154 for (i
= 0; i
< m
->num_objects
- 1; i
++) {
1155 struct object_id oid1
, oid2
;
1157 nth_midxed_object_oid(&oid1
, m
, i
);
1158 nth_midxed_object_oid(&oid2
, m
, i
+ 1);
1160 if (oidcmp(&oid1
, &oid2
) >= 0)
1161 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1162 i
, oid_to_hex(&oid1
), oid_to_hex(&oid2
), i
+ 1);
1164 midx_display_sparse_progress(progress
, i
+ 1);
1166 stop_progress(&progress
);
1169 * Create an array mapping each object to its packfile id. Sort it
1170 * to group the objects by packfile. Use this permutation to visit
1171 * each of the objects and only require 1 packfile to be open at a
1174 ALLOC_ARRAY(pairs
, m
->num_objects
);
1175 for (i
= 0; i
< m
->num_objects
; i
++) {
1177 pairs
[i
].pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1180 if (flags
& MIDX_PROGRESS
)
1181 progress
= start_sparse_progress(_("Sorting objects by packfile"),
1183 display_progress(progress
, 0); /* TODO: Measure QSORT() progress */
1184 QSORT(pairs
, m
->num_objects
, compare_pair_pos_vs_id
);
1185 stop_progress(&progress
);
1187 if (flags
& MIDX_PROGRESS
)
1188 progress
= start_sparse_progress(_("Verifying object offsets"), m
->num_objects
);
1189 for (i
= 0; i
< m
->num_objects
; i
++) {
1190 struct object_id oid
;
1191 struct pack_entry e
;
1192 off_t m_offset
, p_offset
;
1194 if (i
> 0 && pairs
[i
-1].pack_int_id
!= pairs
[i
].pack_int_id
&&
1195 m
->packs
[pairs
[i
-1].pack_int_id
])
1197 close_pack_fd(m
->packs
[pairs
[i
-1].pack_int_id
]);
1198 close_pack_index(m
->packs
[pairs
[i
-1].pack_int_id
]);
1201 nth_midxed_object_oid(&oid
, m
, pairs
[i
].pos
);
1203 if (!fill_midx_entry(r
, &oid
, &e
, m
)) {
1204 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1205 pairs
[i
].pos
, oid_to_hex(&oid
));
1209 if (open_pack_index(e
.p
)) {
1210 midx_report(_("failed to load pack-index for packfile %s"),
1215 m_offset
= e
.offset
;
1216 p_offset
= find_pack_entry_one(oid
.hash
, e
.p
);
1218 if (m_offset
!= p_offset
)
1219 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64
" != %"PRIx64
),
1220 pairs
[i
].pos
, oid_to_hex(&oid
), m_offset
, p_offset
);
1222 midx_display_sparse_progress(progress
, i
+ 1);
1224 stop_progress(&progress
);
1228 return verify_midx_error
;
1231 int expire_midx_packs(struct repository
*r
, const char *object_dir
, unsigned flags
)
1233 uint32_t i
, *count
, result
= 0;
1234 struct string_list packs_to_drop
= STRING_LIST_INIT_DUP
;
1235 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
1236 struct progress
*progress
= NULL
;
1241 count
= xcalloc(m
->num_packs
, sizeof(uint32_t));
1243 if (flags
& MIDX_PROGRESS
)
1244 progress
= start_delayed_progress(_("Counting referenced objects"),
1246 for (i
= 0; i
< m
->num_objects
; i
++) {
1247 int pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1248 count
[pack_int_id
]++;
1249 display_progress(progress
, i
+ 1);
1251 stop_progress(&progress
);
1253 if (flags
& MIDX_PROGRESS
)
1254 progress
= start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1256 for (i
= 0; i
< m
->num_packs
; i
++) {
1258 display_progress(progress
, i
+ 1);
1263 if (prepare_midx_pack(r
, m
, i
))
1266 if (m
->packs
[i
]->pack_keep
)
1269 pack_name
= xstrdup(m
->packs
[i
]->pack_name
);
1270 close_pack(m
->packs
[i
]);
1272 string_list_insert(&packs_to_drop
, m
->pack_names
[i
]);
1273 unlink_pack_path(pack_name
, 0);
1276 stop_progress(&progress
);
1280 if (packs_to_drop
.nr
)
1281 result
= write_midx_internal(object_dir
, m
, &packs_to_drop
, flags
);
1283 string_list_clear(&packs_to_drop
, 0);
1287 struct repack_info
{
1289 uint32_t referenced_objects
;
1290 uint32_t pack_int_id
;
1293 static int compare_by_mtime(const void *a_
, const void *b_
)
1295 const struct repack_info
*a
, *b
;
1297 a
= (const struct repack_info
*)a_
;
1298 b
= (const struct repack_info
*)b_
;
1300 if (a
->mtime
< b
->mtime
)
1302 if (a
->mtime
> b
->mtime
)
1307 static int fill_included_packs_all(struct repository
*r
,
1308 struct multi_pack_index
*m
,
1309 unsigned char *include_pack
)
1311 uint32_t i
, count
= 0;
1312 int pack_kept_objects
= 0;
1314 repo_config_get_bool(r
, "repack.packkeptobjects", &pack_kept_objects
);
1316 for (i
= 0; i
< m
->num_packs
; i
++) {
1317 if (prepare_midx_pack(r
, m
, i
))
1319 if (!pack_kept_objects
&& m
->packs
[i
]->pack_keep
)
1322 include_pack
[i
] = 1;
1329 static int fill_included_packs_batch(struct repository
*r
,
1330 struct multi_pack_index
*m
,
1331 unsigned char *include_pack
,
1334 uint32_t i
, packs_to_repack
;
1336 struct repack_info
*pack_info
= xcalloc(m
->num_packs
, sizeof(struct repack_info
));
1337 int pack_kept_objects
= 0;
1339 repo_config_get_bool(r
, "repack.packkeptobjects", &pack_kept_objects
);
1341 for (i
= 0; i
< m
->num_packs
; i
++) {
1342 pack_info
[i
].pack_int_id
= i
;
1344 if (prepare_midx_pack(r
, m
, i
))
1347 pack_info
[i
].mtime
= m
->packs
[i
]->mtime
;
1350 for (i
= 0; batch_size
&& i
< m
->num_objects
; i
++) {
1351 uint32_t pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1352 pack_info
[pack_int_id
].referenced_objects
++;
1355 QSORT(pack_info
, m
->num_packs
, compare_by_mtime
);
1358 packs_to_repack
= 0;
1359 for (i
= 0; total_size
< batch_size
&& i
< m
->num_packs
; i
++) {
1360 int pack_int_id
= pack_info
[i
].pack_int_id
;
1361 struct packed_git
*p
= m
->packs
[pack_int_id
];
1362 size_t expected_size
;
1366 if (!pack_kept_objects
&& p
->pack_keep
)
1368 if (open_pack_index(p
) || !p
->num_objects
)
1371 expected_size
= (size_t)(p
->pack_size
1372 * pack_info
[i
].referenced_objects
);
1373 expected_size
/= p
->num_objects
;
1375 if (expected_size
>= batch_size
)
1379 total_size
+= expected_size
;
1380 include_pack
[pack_int_id
] = 1;
1385 if (packs_to_repack
< 2)
1391 int midx_repack(struct repository
*r
, const char *object_dir
, size_t batch_size
, unsigned flags
)
1395 unsigned char *include_pack
;
1396 struct child_process cmd
= CHILD_PROCESS_INIT
;
1398 struct strbuf base_name
= STRBUF_INIT
;
1399 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
1402 * When updating the default for these configuration
1403 * variables in builtin/repack.c, these must be adjusted
1406 int delta_base_offset
= 1;
1407 int use_delta_islands
= 0;
1412 include_pack
= xcalloc(m
->num_packs
, sizeof(unsigned char));
1415 if (fill_included_packs_batch(r
, m
, include_pack
, batch_size
))
1417 } else if (fill_included_packs_all(r
, m
, include_pack
))
1420 repo_config_get_bool(r
, "repack.usedeltabaseoffset", &delta_base_offset
);
1421 repo_config_get_bool(r
, "repack.usedeltaislands", &use_delta_islands
);
1423 strvec_push(&cmd
.args
, "pack-objects");
1425 strbuf_addstr(&base_name
, object_dir
);
1426 strbuf_addstr(&base_name
, "/pack/pack");
1427 strvec_push(&cmd
.args
, base_name
.buf
);
1429 if (delta_base_offset
)
1430 strvec_push(&cmd
.args
, "--delta-base-offset");
1431 if (use_delta_islands
)
1432 strvec_push(&cmd
.args
, "--delta-islands");
1434 if (flags
& MIDX_PROGRESS
)
1435 strvec_push(&cmd
.args
, "--progress");
1437 strvec_push(&cmd
.args
, "-q");
1439 strbuf_release(&base_name
);
1442 cmd
.in
= cmd
.out
= -1;
1444 if (start_command(&cmd
)) {
1445 error(_("could not start pack-objects"));
1450 cmd_in
= xfdopen(cmd
.in
, "w");
1452 for (i
= 0; i
< m
->num_objects
; i
++) {
1453 struct object_id oid
;
1454 uint32_t pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1456 if (!include_pack
[pack_int_id
])
1459 nth_midxed_object_oid(&oid
, m
, i
);
1460 fprintf(cmd_in
, "%s\n", oid_to_hex(&oid
));
1464 if (finish_command(&cmd
)) {
1465 error(_("could not finish pack-objects"));
1470 result
= write_midx_internal(object_dir
, m
, NULL
, flags
);