7 #include "object-store.h"
8 #include "hash-lookup.h"
12 #include "run-command.h"
13 #include "repository.h"
14 #include "chunk-format.h"
16 #include "pack-bitmap.h"
19 #include "list-objects.h"
21 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
22 #define MIDX_VERSION 1
23 #define MIDX_BYTE_FILE_VERSION 4
24 #define MIDX_BYTE_HASH_VERSION 5
25 #define MIDX_BYTE_NUM_CHUNKS 6
26 #define MIDX_BYTE_NUM_PACKS 8
27 #define MIDX_HEADER_SIZE 12
28 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
30 #define MIDX_CHUNK_ALIGNMENT 4
31 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
32 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
33 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
34 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
35 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
36 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
37 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
38 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
39 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
41 #define PACK_EXPIRED UINT_MAX
43 static uint8_t oid_version(void)
45 switch (hash_algo_by_ptr(the_hash_algo
)) {
51 die(_("invalid hash version"));
55 const unsigned char *get_midx_checksum(struct multi_pack_index
*m
)
57 return m
->data
+ m
->data_len
- the_hash_algo
->rawsz
;
60 char *get_midx_filename(const char *object_dir
)
62 return xstrfmt("%s/pack/multi-pack-index", object_dir
);
65 char *get_midx_rev_filename(struct multi_pack_index
*m
)
67 return xstrfmt("%s/pack/multi-pack-index-%s.rev",
68 m
->object_dir
, hash_to_hex(get_midx_checksum(m
)));
71 static int midx_read_oid_fanout(const unsigned char *chunk_start
,
72 size_t chunk_size
, void *data
)
74 struct multi_pack_index
*m
= data
;
75 m
->chunk_oid_fanout
= (uint32_t *)chunk_start
;
77 if (chunk_size
!= 4 * 256) {
78 error(_("multi-pack-index OID fanout is of the wrong size"));
84 struct multi_pack_index
*load_multi_pack_index(const char *object_dir
, int local
)
86 struct multi_pack_index
*m
= NULL
;
90 void *midx_map
= NULL
;
91 uint32_t hash_version
;
92 char *midx_name
= get_midx_filename(object_dir
);
94 const char *cur_pack_name
;
95 struct chunkfile
*cf
= NULL
;
97 fd
= git_open(midx_name
);
101 if (fstat(fd
, &st
)) {
102 error_errno(_("failed to read %s"), midx_name
);
106 midx_size
= xsize_t(st
.st_size
);
108 if (midx_size
< MIDX_MIN_SIZE
) {
109 error(_("multi-pack-index file %s is too small"), midx_name
);
113 FREE_AND_NULL(midx_name
);
115 midx_map
= xmmap(NULL
, midx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
118 FLEX_ALLOC_STR(m
, object_dir
, object_dir
);
120 m
->data_len
= midx_size
;
123 m
->signature
= get_be32(m
->data
);
124 if (m
->signature
!= MIDX_SIGNATURE
)
125 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
126 m
->signature
, MIDX_SIGNATURE
);
128 m
->version
= m
->data
[MIDX_BYTE_FILE_VERSION
];
129 if (m
->version
!= MIDX_VERSION
)
130 die(_("multi-pack-index version %d not recognized"),
133 hash_version
= m
->data
[MIDX_BYTE_HASH_VERSION
];
134 if (hash_version
!= oid_version()) {
135 error(_("multi-pack-index hash version %u does not match version %u"),
136 hash_version
, oid_version());
139 m
->hash_len
= the_hash_algo
->rawsz
;
141 m
->num_chunks
= m
->data
[MIDX_BYTE_NUM_CHUNKS
];
143 m
->num_packs
= get_be32(m
->data
+ MIDX_BYTE_NUM_PACKS
);
145 cf
= init_chunkfile(NULL
);
147 if (read_table_of_contents(cf
, m
->data
, midx_size
,
148 MIDX_HEADER_SIZE
, m
->num_chunks
))
151 if (pair_chunk(cf
, MIDX_CHUNKID_PACKNAMES
, &m
->chunk_pack_names
) == CHUNK_NOT_FOUND
)
152 die(_("multi-pack-index missing required pack-name chunk"));
153 if (read_chunk(cf
, MIDX_CHUNKID_OIDFANOUT
, midx_read_oid_fanout
, m
) == CHUNK_NOT_FOUND
)
154 die(_("multi-pack-index missing required OID fanout chunk"));
155 if (pair_chunk(cf
, MIDX_CHUNKID_OIDLOOKUP
, &m
->chunk_oid_lookup
) == CHUNK_NOT_FOUND
)
156 die(_("multi-pack-index missing required OID lookup chunk"));
157 if (pair_chunk(cf
, MIDX_CHUNKID_OBJECTOFFSETS
, &m
->chunk_object_offsets
) == CHUNK_NOT_FOUND
)
158 die(_("multi-pack-index missing required object offsets chunk"));
160 pair_chunk(cf
, MIDX_CHUNKID_LARGEOFFSETS
, &m
->chunk_large_offsets
);
162 m
->num_objects
= ntohl(m
->chunk_oid_fanout
[255]);
164 CALLOC_ARRAY(m
->pack_names
, m
->num_packs
);
165 CALLOC_ARRAY(m
->packs
, m
->num_packs
);
167 cur_pack_name
= (const char *)m
->chunk_pack_names
;
168 for (i
= 0; i
< m
->num_packs
; i
++) {
169 m
->pack_names
[i
] = cur_pack_name
;
171 cur_pack_name
+= strlen(cur_pack_name
) + 1;
173 if (i
&& strcmp(m
->pack_names
[i
], m
->pack_names
[i
- 1]) <= 0)
174 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
175 m
->pack_names
[i
- 1],
179 trace2_data_intmax("midx", the_repository
, "load/num_packs", m
->num_packs
);
180 trace2_data_intmax("midx", the_repository
, "load/num_objects", m
->num_objects
);
189 munmap(midx_map
, midx_size
);
195 void close_midx(struct multi_pack_index
*m
)
204 munmap((unsigned char *)m
->data
, m
->data_len
);
206 for (i
= 0; i
< m
->num_packs
; i
++) {
208 m
->packs
[i
]->multi_pack_index
= 0;
210 FREE_AND_NULL(m
->packs
);
211 FREE_AND_NULL(m
->pack_names
);
215 int prepare_midx_pack(struct repository
*r
, struct multi_pack_index
*m
, uint32_t pack_int_id
)
217 struct strbuf pack_name
= STRBUF_INIT
;
218 struct packed_git
*p
;
220 if (pack_int_id
>= m
->num_packs
)
221 die(_("bad pack-int-id: %u (%u total packs)"),
222 pack_int_id
, m
->num_packs
);
224 if (m
->packs
[pack_int_id
])
227 strbuf_addf(&pack_name
, "%s/pack/%s", m
->object_dir
,
228 m
->pack_names
[pack_int_id
]);
230 p
= add_packed_git(pack_name
.buf
, pack_name
.len
, m
->local
);
231 strbuf_release(&pack_name
);
236 p
->multi_pack_index
= 1;
237 m
->packs
[pack_int_id
] = p
;
238 install_packed_git(r
, p
);
239 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
244 int bsearch_midx(const struct object_id
*oid
, struct multi_pack_index
*m
, uint32_t *result
)
246 return bsearch_hash(oid
->hash
, m
->chunk_oid_fanout
, m
->chunk_oid_lookup
,
247 the_hash_algo
->rawsz
, result
);
250 struct object_id
*nth_midxed_object_oid(struct object_id
*oid
,
251 struct multi_pack_index
*m
,
254 if (n
>= m
->num_objects
)
257 oidread(oid
, m
->chunk_oid_lookup
+ m
->hash_len
* n
);
261 off_t
nth_midxed_offset(struct multi_pack_index
*m
, uint32_t pos
)
263 const unsigned char *offset_data
;
266 offset_data
= m
->chunk_object_offsets
+ (off_t
)pos
* MIDX_CHUNK_OFFSET_WIDTH
;
267 offset32
= get_be32(offset_data
+ sizeof(uint32_t));
269 if (m
->chunk_large_offsets
&& offset32
& MIDX_LARGE_OFFSET_NEEDED
) {
270 if (sizeof(off_t
) < sizeof(uint64_t))
271 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
273 offset32
^= MIDX_LARGE_OFFSET_NEEDED
;
274 return get_be64(m
->chunk_large_offsets
+ sizeof(uint64_t) * offset32
);
280 uint32_t nth_midxed_pack_int_id(struct multi_pack_index
*m
, uint32_t pos
)
282 return get_be32(m
->chunk_object_offsets
+
283 (off_t
)pos
* MIDX_CHUNK_OFFSET_WIDTH
);
286 int fill_midx_entry(struct repository
* r
,
287 const struct object_id
*oid
,
288 struct pack_entry
*e
,
289 struct multi_pack_index
*m
)
292 uint32_t pack_int_id
;
293 struct packed_git
*p
;
295 if (!bsearch_midx(oid
, m
, &pos
))
298 if (pos
>= m
->num_objects
)
301 pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
303 if (prepare_midx_pack(r
, m
, pack_int_id
))
305 p
= m
->packs
[pack_int_id
];
308 * We are about to tell the caller where they can locate the
309 * requested object. We better make sure the packfile is
310 * still here and can be accessed before supplying that
311 * answer, as it may have been deleted since the MIDX was
314 if (!is_pack_valid(p
))
317 if (oidset_size(&p
->bad_objects
) &&
318 oidset_contains(&p
->bad_objects
, oid
))
321 e
->offset
= nth_midxed_offset(m
, pos
);
327 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
328 static int cmp_idx_or_pack_name(const char *idx_or_pack_name
,
329 const char *idx_name
)
331 /* Skip past any initial matching prefix. */
332 while (*idx_name
&& *idx_name
== *idx_or_pack_name
) {
338 * If we didn't match completely, we may have matched "pack-1234." and
339 * be left with "idx" and "pack" respectively, which is also OK. We do
340 * not have to check for "idx" and "idx", because that would have been
341 * a complete match (and in that case these strcmps will be false, but
342 * we'll correctly return 0 from the final strcmp() below.
344 * Technically this matches "fooidx" and "foopack", but we'd never have
345 * such names in the first place.
347 if (!strcmp(idx_name
, "idx") && !strcmp(idx_or_pack_name
, "pack"))
351 * This not only checks for a complete match, but also orders based on
352 * the first non-identical character, which means our ordering will
353 * match a raw strcmp(). That makes it OK to use this to binary search
354 * a naively-sorted list.
356 return strcmp(idx_or_pack_name
, idx_name
);
359 int midx_contains_pack(struct multi_pack_index
*m
, const char *idx_or_pack_name
)
361 uint32_t first
= 0, last
= m
->num_packs
;
363 while (first
< last
) {
364 uint32_t mid
= first
+ (last
- first
) / 2;
368 current
= m
->pack_names
[mid
];
369 cmp
= cmp_idx_or_pack_name(idx_or_pack_name
, current
);
382 int prepare_multi_pack_index_one(struct repository
*r
, const char *object_dir
, int local
)
384 struct multi_pack_index
*m
;
385 struct multi_pack_index
*m_search
;
387 prepare_repo_settings(r
);
388 if (!r
->settings
.core_multi_pack_index
)
391 for (m_search
= r
->objects
->multi_pack_index
; m_search
; m_search
= m_search
->next
)
392 if (!strcmp(object_dir
, m_search
->object_dir
))
395 m
= load_multi_pack_index(object_dir
, local
);
398 struct multi_pack_index
*mp
= r
->objects
->multi_pack_index
;
403 r
->objects
->multi_pack_index
= m
;
410 static size_t write_midx_header(struct hashfile
*f
,
411 unsigned char num_chunks
,
414 hashwrite_be32(f
, MIDX_SIGNATURE
);
415 hashwrite_u8(f
, MIDX_VERSION
);
416 hashwrite_u8(f
, oid_version());
417 hashwrite_u8(f
, num_chunks
);
418 hashwrite_u8(f
, 0); /* unused */
419 hashwrite_be32(f
, num_packs
);
421 return MIDX_HEADER_SIZE
;
425 uint32_t orig_pack_int_id
;
427 struct packed_git
*p
;
428 unsigned expired
: 1;
431 static int pack_info_compare(const void *_a
, const void *_b
)
433 struct pack_info
*a
= (struct pack_info
*)_a
;
434 struct pack_info
*b
= (struct pack_info
*)_b
;
435 return strcmp(a
->pack_name
, b
->pack_name
);
438 static int idx_or_pack_name_cmp(const void *_va
, const void *_vb
)
440 const char *pack_name
= _va
;
441 const struct pack_info
*compar
= _vb
;
443 return cmp_idx_or_pack_name(pack_name
, compar
->pack_name
);
446 struct write_midx_context
{
447 struct pack_info
*info
;
450 struct multi_pack_index
*m
;
451 struct progress
*progress
;
452 unsigned pack_paths_checked
;
454 struct pack_midx_entry
*entries
;
458 uint32_t *pack_order
;
459 unsigned large_offsets_needed
:1;
460 uint32_t num_large_offsets
;
462 int preferred_pack_idx
;
464 struct string_list
*to_include
;
467 static void add_pack_to_midx(const char *full_path
, size_t full_path_len
,
468 const char *file_name
, void *data
)
470 struct write_midx_context
*ctx
= data
;
472 if (ends_with(file_name
, ".idx")) {
473 display_progress(ctx
->progress
, ++ctx
->pack_paths_checked
);
475 * Note that at most one of ctx->m and ctx->to_include are set,
476 * so we are testing midx_contains_pack() and
477 * string_list_has_string() independently (guarded by the
478 * appropriate NULL checks).
480 * We could support passing to_include while reusing an existing
481 * MIDX, but don't currently since the reuse process drags
482 * forward all packs from an existing MIDX (without checking
483 * whether or not they appear in the to_include list).
485 * If we added support for that, these next two conditional
486 * should be performed independently (likely checking
487 * to_include before the existing MIDX).
489 if (ctx
->m
&& midx_contains_pack(ctx
->m
, file_name
))
491 else if (ctx
->to_include
&&
492 !string_list_has_string(ctx
->to_include
, file_name
))
495 ALLOC_GROW(ctx
->info
, ctx
->nr
+ 1, ctx
->alloc
);
497 ctx
->info
[ctx
->nr
].p
= add_packed_git(full_path
,
501 if (!ctx
->info
[ctx
->nr
].p
) {
502 warning(_("failed to add packfile '%s'"),
507 if (open_pack_index(ctx
->info
[ctx
->nr
].p
)) {
508 warning(_("failed to open pack-index '%s'"),
510 close_pack(ctx
->info
[ctx
->nr
].p
);
511 FREE_AND_NULL(ctx
->info
[ctx
->nr
].p
);
515 ctx
->info
[ctx
->nr
].pack_name
= xstrdup(file_name
);
516 ctx
->info
[ctx
->nr
].orig_pack_int_id
= ctx
->nr
;
517 ctx
->info
[ctx
->nr
].expired
= 0;
522 struct pack_midx_entry
{
523 struct object_id oid
;
524 uint32_t pack_int_id
;
527 unsigned preferred
: 1;
530 static int midx_oid_compare(const void *_a
, const void *_b
)
532 const struct pack_midx_entry
*a
= (const struct pack_midx_entry
*)_a
;
533 const struct pack_midx_entry
*b
= (const struct pack_midx_entry
*)_b
;
534 int cmp
= oidcmp(&a
->oid
, &b
->oid
);
539 /* Sort objects in a preferred pack first when multiple copies exist. */
540 if (a
->preferred
> b
->preferred
)
542 if (a
->preferred
< b
->preferred
)
545 if (a
->pack_mtime
> b
->pack_mtime
)
547 else if (a
->pack_mtime
< b
->pack_mtime
)
550 return a
->pack_int_id
- b
->pack_int_id
;
553 static int nth_midxed_pack_midx_entry(struct multi_pack_index
*m
,
554 struct pack_midx_entry
*e
,
557 if (pos
>= m
->num_objects
)
560 nth_midxed_object_oid(&e
->oid
, m
, pos
);
561 e
->pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
562 e
->offset
= nth_midxed_offset(m
, pos
);
564 /* consider objects in midx to be from "old" packs */
569 static void fill_pack_entry(uint32_t pack_int_id
,
570 struct packed_git
*p
,
572 struct pack_midx_entry
*entry
,
575 if (nth_packed_object_id(&entry
->oid
, p
, cur_object
) < 0)
576 die(_("failed to locate object %d in packfile"), cur_object
);
578 entry
->pack_int_id
= pack_int_id
;
579 entry
->pack_mtime
= p
->mtime
;
581 entry
->offset
= nth_packed_object_offset(p
, cur_object
);
582 entry
->preferred
= !!preferred
;
586 * It is possible to artificially get into a state where there are many
587 * duplicate copies of objects. That can create high memory pressure if
588 * we are to create a list of all objects before de-duplication. To reduce
589 * this memory pressure without a significant performance drop, automatically
590 * group objects by the first byte of their object id. Use the IDX fanout
591 * tables to group the data, copy to a local array, then sort.
593 * Copy only the de-duplicated entries (selected by most-recent modified time
594 * of a packfile containing the object).
596 static struct pack_midx_entry
*get_sorted_entries(struct multi_pack_index
*m
,
597 struct pack_info
*info
,
599 uint32_t *nr_objects
,
602 uint32_t cur_fanout
, cur_pack
, cur_object
;
603 uint32_t alloc_fanout
, alloc_objects
, total_objects
= 0;
604 struct pack_midx_entry
*entries_by_fanout
= NULL
;
605 struct pack_midx_entry
*deduplicated_entries
= NULL
;
606 uint32_t start_pack
= m
? m
->num_packs
: 0;
608 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++)
609 total_objects
+= info
[cur_pack
].p
->num_objects
;
612 * As we de-duplicate by fanout value, we expect the fanout
613 * slices to be evenly distributed, with some noise. Hence,
614 * allocate slightly more than one 256th.
616 alloc_objects
= alloc_fanout
= total_objects
> 3200 ? total_objects
/ 200 : 16;
618 ALLOC_ARRAY(entries_by_fanout
, alloc_fanout
);
619 ALLOC_ARRAY(deduplicated_entries
, alloc_objects
);
622 for (cur_fanout
= 0; cur_fanout
< 256; cur_fanout
++) {
623 uint32_t nr_fanout
= 0;
626 uint32_t start
= 0, end
;
629 start
= ntohl(m
->chunk_oid_fanout
[cur_fanout
- 1]);
630 end
= ntohl(m
->chunk_oid_fanout
[cur_fanout
]);
632 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
633 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
634 nth_midxed_pack_midx_entry(m
,
635 &entries_by_fanout
[nr_fanout
],
637 if (nth_midxed_pack_int_id(m
, cur_object
) == preferred_pack
)
638 entries_by_fanout
[nr_fanout
].preferred
= 1;
640 entries_by_fanout
[nr_fanout
].preferred
= 0;
645 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++) {
646 uint32_t start
= 0, end
;
647 int preferred
= cur_pack
== preferred_pack
;
650 start
= get_pack_fanout(info
[cur_pack
].p
, cur_fanout
- 1);
651 end
= get_pack_fanout(info
[cur_pack
].p
, cur_fanout
);
653 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
654 ALLOC_GROW(entries_by_fanout
, nr_fanout
+ 1, alloc_fanout
);
655 fill_pack_entry(cur_pack
,
658 &entries_by_fanout
[nr_fanout
],
664 QSORT(entries_by_fanout
, nr_fanout
, midx_oid_compare
);
667 * The batch is now sorted by OID and then mtime (descending).
668 * Take only the first duplicate.
670 for (cur_object
= 0; cur_object
< nr_fanout
; cur_object
++) {
671 if (cur_object
&& oideq(&entries_by_fanout
[cur_object
- 1].oid
,
672 &entries_by_fanout
[cur_object
].oid
))
675 ALLOC_GROW(deduplicated_entries
, *nr_objects
+ 1, alloc_objects
);
676 memcpy(&deduplicated_entries
[*nr_objects
],
677 &entries_by_fanout
[cur_object
],
678 sizeof(struct pack_midx_entry
));
683 free(entries_by_fanout
);
684 return deduplicated_entries
;
687 static int write_midx_pack_names(struct hashfile
*f
, void *data
)
689 struct write_midx_context
*ctx
= data
;
691 unsigned char padding
[MIDX_CHUNK_ALIGNMENT
];
694 for (i
= 0; i
< ctx
->nr
; i
++) {
697 if (ctx
->info
[i
].expired
)
700 if (i
&& strcmp(ctx
->info
[i
].pack_name
, ctx
->info
[i
- 1].pack_name
) <= 0)
701 BUG("incorrect pack-file order: %s before %s",
702 ctx
->info
[i
- 1].pack_name
,
703 ctx
->info
[i
].pack_name
);
705 writelen
= strlen(ctx
->info
[i
].pack_name
) + 1;
706 hashwrite(f
, ctx
->info
[i
].pack_name
, writelen
);
710 /* add padding to be aligned */
711 i
= MIDX_CHUNK_ALIGNMENT
- (written
% MIDX_CHUNK_ALIGNMENT
);
712 if (i
< MIDX_CHUNK_ALIGNMENT
) {
713 memset(padding
, 0, sizeof(padding
));
714 hashwrite(f
, padding
, i
);
720 static int write_midx_oid_fanout(struct hashfile
*f
,
723 struct write_midx_context
*ctx
= data
;
724 struct pack_midx_entry
*list
= ctx
->entries
;
725 struct pack_midx_entry
*last
= ctx
->entries
+ ctx
->entries_nr
;
730 * Write the first-level table (the list is sorted,
731 * but we use a 256-entry lookup to be able to avoid
732 * having to do eight extra binary search iterations).
734 for (i
= 0; i
< 256; i
++) {
735 struct pack_midx_entry
*next
= list
;
737 while (next
< last
&& next
->oid
.hash
[0] == i
) {
742 hashwrite_be32(f
, count
);
749 static int write_midx_oid_lookup(struct hashfile
*f
,
752 struct write_midx_context
*ctx
= data
;
753 unsigned char hash_len
= the_hash_algo
->rawsz
;
754 struct pack_midx_entry
*list
= ctx
->entries
;
757 for (i
= 0; i
< ctx
->entries_nr
; i
++) {
758 struct pack_midx_entry
*obj
= list
++;
760 if (i
< ctx
->entries_nr
- 1) {
761 struct pack_midx_entry
*next
= list
;
762 if (oidcmp(&obj
->oid
, &next
->oid
) >= 0)
763 BUG("OIDs not in order: %s >= %s",
764 oid_to_hex(&obj
->oid
),
765 oid_to_hex(&next
->oid
));
768 hashwrite(f
, obj
->oid
.hash
, (int)hash_len
);
774 static int write_midx_object_offsets(struct hashfile
*f
,
777 struct write_midx_context
*ctx
= data
;
778 struct pack_midx_entry
*list
= ctx
->entries
;
779 uint32_t i
, nr_large_offset
= 0;
781 for (i
= 0; i
< ctx
->entries_nr
; i
++) {
782 struct pack_midx_entry
*obj
= list
++;
784 if (ctx
->pack_perm
[obj
->pack_int_id
] == PACK_EXPIRED
)
785 BUG("object %s is in an expired pack with int-id %d",
786 oid_to_hex(&obj
->oid
),
789 hashwrite_be32(f
, ctx
->pack_perm
[obj
->pack_int_id
]);
791 if (ctx
->large_offsets_needed
&& obj
->offset
>> 31)
792 hashwrite_be32(f
, MIDX_LARGE_OFFSET_NEEDED
| nr_large_offset
++);
793 else if (!ctx
->large_offsets_needed
&& obj
->offset
>> 32)
794 BUG("object %s requires a large offset (%"PRIx64
") but the MIDX is not writing large offsets!",
795 oid_to_hex(&obj
->oid
),
798 hashwrite_be32(f
, (uint32_t)obj
->offset
);
804 static int write_midx_large_offsets(struct hashfile
*f
,
807 struct write_midx_context
*ctx
= data
;
808 struct pack_midx_entry
*list
= ctx
->entries
;
809 struct pack_midx_entry
*end
= ctx
->entries
+ ctx
->entries_nr
;
810 uint32_t nr_large_offset
= ctx
->num_large_offsets
;
812 while (nr_large_offset
) {
813 struct pack_midx_entry
*obj
;
817 BUG("too many large-offset objects");
820 offset
= obj
->offset
;
825 hashwrite_be64(f
, offset
);
833 struct midx_pack_order_data
{
839 static int midx_pack_order_cmp(const void *va
, const void *vb
)
841 const struct midx_pack_order_data
*a
= va
, *b
= vb
;
842 if (a
->pack
< b
->pack
)
844 else if (a
->pack
> b
->pack
)
846 else if (a
->offset
< b
->offset
)
848 else if (a
->offset
> b
->offset
)
854 static uint32_t *midx_pack_order(struct write_midx_context
*ctx
)
856 struct midx_pack_order_data
*data
;
857 uint32_t *pack_order
;
860 ALLOC_ARRAY(data
, ctx
->entries_nr
);
861 for (i
= 0; i
< ctx
->entries_nr
; i
++) {
862 struct pack_midx_entry
*e
= &ctx
->entries
[i
];
864 data
[i
].pack
= ctx
->pack_perm
[e
->pack_int_id
];
866 data
[i
].pack
|= (1U << 31);
867 data
[i
].offset
= e
->offset
;
870 QSORT(data
, ctx
->entries_nr
, midx_pack_order_cmp
);
872 ALLOC_ARRAY(pack_order
, ctx
->entries_nr
);
873 for (i
= 0; i
< ctx
->entries_nr
; i
++)
874 pack_order
[i
] = data
[i
].nr
;
880 static void write_midx_reverse_index(char *midx_name
, unsigned char *midx_hash
,
881 struct write_midx_context
*ctx
)
883 struct strbuf buf
= STRBUF_INIT
;
884 const char *tmp_file
;
886 strbuf_addf(&buf
, "%s-%s.rev", midx_name
, hash_to_hex(midx_hash
));
888 tmp_file
= write_rev_file_order(NULL
, ctx
->pack_order
, ctx
->entries_nr
,
889 midx_hash
, WRITE_REV
);
891 if (finalize_object_file(tmp_file
, buf
.buf
))
892 die(_("cannot store reverse index file"));
894 strbuf_release(&buf
);
897 static void clear_midx_files_ext(const char *object_dir
, const char *ext
,
898 unsigned char *keep_hash
);
900 static int midx_checksum_valid(struct multi_pack_index
*m
)
902 return hashfile_checksum_valid(m
->data
, m
->data_len
);
905 static void prepare_midx_packing_data(struct packing_data
*pdata
,
906 struct write_midx_context
*ctx
)
910 memset(pdata
, 0, sizeof(struct packing_data
));
911 prepare_packing_data(the_repository
, pdata
);
913 for (i
= 0; i
< ctx
->entries_nr
; i
++) {
914 struct pack_midx_entry
*from
= &ctx
->entries
[ctx
->pack_order
[i
]];
915 struct object_entry
*to
= packlist_alloc(pdata
, &from
->oid
);
917 oe_set_in_pack(pdata
, to
,
918 ctx
->info
[ctx
->pack_perm
[from
->pack_int_id
]].p
);
922 static int add_ref_to_pending(const char *refname
,
923 const struct object_id
*oid
,
924 int flag
, void *cb_data
)
926 struct rev_info
*revs
= (struct rev_info
*)cb_data
;
927 struct object
*object
;
929 if ((flag
& REF_ISSYMREF
) && (flag
& REF_ISBROKEN
)) {
930 warning("symbolic ref is dangling: %s", refname
);
934 object
= parse_object_or_die(oid
, refname
);
935 if (object
->type
!= OBJ_COMMIT
)
938 add_pending_object(revs
, object
, "");
939 if (bitmap_is_preferred_refname(revs
->repo
, refname
))
940 object
->flags
|= NEEDS_BITMAP
;
944 struct bitmap_commit_cb
{
945 struct commit
**commits
;
946 size_t commits_nr
, commits_alloc
;
948 struct write_midx_context
*ctx
;
951 static const struct object_id
*bitmap_oid_access(size_t index
,
952 const void *_entries
)
954 const struct pack_midx_entry
*entries
= _entries
;
955 return &entries
[index
].oid
;
958 static void bitmap_show_commit(struct commit
*commit
, void *_data
)
960 struct bitmap_commit_cb
*data
= _data
;
961 int pos
= oid_pos(&commit
->object
.oid
, data
->ctx
->entries
,
962 data
->ctx
->entries_nr
,
967 ALLOC_GROW(data
->commits
, data
->commits_nr
+ 1, data
->commits_alloc
);
968 data
->commits
[data
->commits_nr
++] = commit
;
971 static int read_refs_snapshot(const char *refs_snapshot
,
972 struct rev_info
*revs
)
974 struct strbuf buf
= STRBUF_INIT
;
975 struct object_id oid
;
976 FILE *f
= xfopen(refs_snapshot
, "r");
978 while (strbuf_getline(&buf
, f
) != EOF
) {
979 struct object
*object
;
982 const char *end
= NULL
;
984 if (buf
.len
&& *buf
.buf
== '+') {
989 if (parse_oid_hex(hex
, &oid
, &end
) < 0)
990 die(_("could not parse line: %s"), buf
.buf
);
992 die(_("malformed line: %s"), buf
.buf
);
994 object
= parse_object_or_die(&oid
, NULL
);
996 object
->flags
|= NEEDS_BITMAP
;
998 add_pending_object(revs
, object
, "");
1002 strbuf_release(&buf
);
1006 static struct commit
**find_commits_for_midx_bitmap(uint32_t *indexed_commits_nr_p
,
1007 const char *refs_snapshot
,
1008 struct write_midx_context
*ctx
)
1010 struct rev_info revs
;
1011 struct bitmap_commit_cb cb
= {0};
1015 repo_init_revisions(the_repository
, &revs
, NULL
);
1016 if (refs_snapshot
) {
1017 read_refs_snapshot(refs_snapshot
, &revs
);
1019 setup_revisions(0, NULL
, &revs
, NULL
);
1020 for_each_ref(add_ref_to_pending
, &revs
);
1024 * Skipping promisor objects here is intentional, since it only excludes
1025 * them from the list of reachable commits that we want to select from
1026 * when computing the selection of MIDX'd commits to receive bitmaps.
1028 * Reachability bitmaps do require that their objects be closed under
1029 * reachability, but fetching any objects missing from promisors at this
1030 * point is too late. But, if one of those objects can be reached from
1031 * an another object that is included in the bitmap, then we will
1032 * complain later that we don't have reachability closure (and fail
1035 fetch_if_missing
= 0;
1036 revs
.exclude_promisor_objects
= 1;
1038 if (prepare_revision_walk(&revs
))
1039 die(_("revision walk setup failed"));
1041 traverse_commit_list(&revs
, bitmap_show_commit
, NULL
, &cb
);
1042 if (indexed_commits_nr_p
)
1043 *indexed_commits_nr_p
= cb
.commits_nr
;
1048 static int write_midx_bitmap(char *midx_name
, unsigned char *midx_hash
,
1049 struct write_midx_context
*ctx
,
1050 const char *refs_snapshot
,
1053 struct packing_data pdata
;
1054 struct pack_idx_entry
**index
;
1055 struct commit
**commits
= NULL
;
1056 uint32_t i
, commits_nr
;
1057 uint16_t options
= 0;
1058 char *bitmap_name
= xstrfmt("%s-%s.bitmap", midx_name
, hash_to_hex(midx_hash
));
1061 if (flags
& MIDX_WRITE_BITMAP_HASH_CACHE
)
1062 options
|= BITMAP_OPT_HASH_CACHE
;
1064 prepare_midx_packing_data(&pdata
, ctx
);
1066 commits
= find_commits_for_midx_bitmap(&commits_nr
, refs_snapshot
, ctx
);
1069 * Build the MIDX-order index based on pdata.objects (which is already
1070 * in MIDX order; c.f., 'midx_pack_order_cmp()' for the definition of
1073 ALLOC_ARRAY(index
, pdata
.nr_objects
);
1074 for (i
= 0; i
< pdata
.nr_objects
; i
++)
1075 index
[i
] = &pdata
.objects
[i
].idx
;
1077 bitmap_writer_show_progress(flags
& MIDX_PROGRESS
);
1078 bitmap_writer_build_type_index(&pdata
, index
, pdata
.nr_objects
);
1081 * bitmap_writer_finish expects objects in lex order, but pack_order
1082 * gives us exactly that. use it directly instead of re-sorting the
1085 * This changes the order of objects in 'index' between
1086 * bitmap_writer_build_type_index and bitmap_writer_finish.
1088 * The same re-ordering takes place in the single-pack bitmap code via
1089 * write_idx_file(), which is called by finish_tmp_packfile(), which
1090 * happens between bitmap_writer_build_type_index() and
1091 * bitmap_writer_finish().
1093 for (i
= 0; i
< pdata
.nr_objects
; i
++)
1094 index
[ctx
->pack_order
[i
]] = &pdata
.objects
[i
].idx
;
1096 bitmap_writer_select_commits(commits
, commits_nr
, -1);
1097 ret
= bitmap_writer_build(&pdata
);
1101 bitmap_writer_set_checksum(midx_hash
);
1102 bitmap_writer_finish(index
, pdata
.nr_objects
, bitmap_name
, options
);
1110 static int write_midx_internal(const char *object_dir
,
1111 struct string_list
*packs_to_include
,
1112 struct string_list
*packs_to_drop
,
1113 const char *preferred_pack_name
,
1114 const char *refs_snapshot
,
1118 unsigned char midx_hash
[GIT_MAX_RAWSZ
];
1120 struct hashfile
*f
= NULL
;
1121 struct lock_file lk
;
1122 struct write_midx_context ctx
= { 0 };
1123 struct multi_pack_index
*cur
;
1124 int pack_name_concat_len
= 0;
1125 int dropped_packs
= 0;
1127 struct chunkfile
*cf
;
1129 /* Ensure the given object_dir is local, or a known alternate. */
1130 find_odb(the_repository
, object_dir
);
1132 midx_name
= get_midx_filename(object_dir
);
1133 if (safe_create_leading_directories(midx_name
))
1134 die_errno(_("unable to create leading directories of %s"),
1137 if (!packs_to_include
) {
1139 * Only reference an existing MIDX when not filtering which
1140 * packs to include, since all packs and objects are copied
1141 * blindly from an existing MIDX if one is present.
1143 for (cur
= get_multi_pack_index(the_repository
); cur
; cur
= cur
->next
) {
1144 if (!strcmp(object_dir
, cur
->object_dir
)) {
1151 if (ctx
.m
&& !midx_checksum_valid(ctx
.m
)) {
1152 warning(_("ignoring existing multi-pack-index; checksum mismatch"));
1157 ctx
.alloc
= ctx
.m
? ctx
.m
->num_packs
: 16;
1159 ALLOC_ARRAY(ctx
.info
, ctx
.alloc
);
1162 for (i
= 0; i
< ctx
.m
->num_packs
; i
++) {
1163 ALLOC_GROW(ctx
.info
, ctx
.nr
+ 1, ctx
.alloc
);
1165 ctx
.info
[ctx
.nr
].orig_pack_int_id
= i
;
1166 ctx
.info
[ctx
.nr
].pack_name
= xstrdup(ctx
.m
->pack_names
[i
]);
1167 ctx
.info
[ctx
.nr
].p
= ctx
.m
->packs
[i
];
1168 ctx
.info
[ctx
.nr
].expired
= 0;
1170 if (flags
& MIDX_WRITE_REV_INDEX
) {
1172 * If generating a reverse index, need to have
1173 * packed_git's loaded to compare their
1174 * mtimes and object count.
1176 if (prepare_midx_pack(the_repository
, ctx
.m
, i
)) {
1177 error(_("could not load pack"));
1182 if (open_pack_index(ctx
.m
->packs
[i
]))
1183 die(_("could not open index for %s"),
1184 ctx
.m
->packs
[i
]->pack_name
);
1185 ctx
.info
[ctx
.nr
].p
= ctx
.m
->packs
[i
];
1192 ctx
.pack_paths_checked
= 0;
1193 if (flags
& MIDX_PROGRESS
)
1194 ctx
.progress
= start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
1196 ctx
.progress
= NULL
;
1198 ctx
.to_include
= packs_to_include
;
1200 for_each_file_in_pack_dir(object_dir
, add_pack_to_midx
, &ctx
);
1201 stop_progress(&ctx
.progress
);
1203 if ((ctx
.m
&& ctx
.nr
== ctx
.m
->num_packs
) &&
1204 !(packs_to_include
|| packs_to_drop
)) {
1205 struct bitmap_index
*bitmap_git
;
1207 int want_bitmap
= flags
& MIDX_WRITE_BITMAP
;
1209 bitmap_git
= prepare_midx_bitmap_git(ctx
.m
);
1210 bitmap_exists
= bitmap_git
&& bitmap_is_midx(bitmap_git
);
1211 free_bitmap_index(bitmap_git
);
1213 if (bitmap_exists
|| !want_bitmap
) {
1215 * The correct MIDX already exists, and so does a
1216 * corresponding bitmap (or one wasn't requested).
1219 clear_midx_files_ext(object_dir
, ".bitmap",
1225 if (preferred_pack_name
) {
1227 for (i
= 0; i
< ctx
.nr
; i
++) {
1228 if (!cmp_idx_or_pack_name(preferred_pack_name
,
1229 ctx
.info
[i
].pack_name
)) {
1230 ctx
.preferred_pack_idx
= i
;
1237 warning(_("unknown preferred pack: '%s'"),
1238 preferred_pack_name
);
1239 } else if (ctx
.nr
&&
1240 (flags
& (MIDX_WRITE_REV_INDEX
| MIDX_WRITE_BITMAP
))) {
1241 struct packed_git
*oldest
= ctx
.info
[ctx
.preferred_pack_idx
].p
;
1242 ctx
.preferred_pack_idx
= 0;
1244 if (packs_to_drop
&& packs_to_drop
->nr
)
1245 BUG("cannot write a MIDX bitmap during expiration");
1248 * set a preferred pack when writing a bitmap to ensure that
1249 * the pack from which the first object is selected in pseudo
1250 * pack-order has all of its objects selected from that pack
1251 * (and not another pack containing a duplicate)
1253 for (i
= 1; i
< ctx
.nr
; i
++) {
1254 struct packed_git
*p
= ctx
.info
[i
].p
;
1256 if (!oldest
->num_objects
|| p
->mtime
< oldest
->mtime
) {
1258 ctx
.preferred_pack_idx
= i
;
1262 if (!oldest
->num_objects
) {
1264 * If all packs are empty; unset the preferred index.
1265 * This is acceptable since there will be no duplicate
1266 * objects to resolve, so the preferred value doesn't
1269 ctx
.preferred_pack_idx
= -1;
1273 * otherwise don't mark any pack as preferred to avoid
1274 * interfering with expiration logic below
1276 ctx
.preferred_pack_idx
= -1;
1279 if (ctx
.preferred_pack_idx
> -1) {
1280 struct packed_git
*preferred
= ctx
.info
[ctx
.preferred_pack_idx
].p
;
1281 if (!preferred
->num_objects
) {
1282 error(_("cannot select preferred pack %s with no objects"),
1283 preferred
->pack_name
);
1289 ctx
.entries
= get_sorted_entries(ctx
.m
, ctx
.info
, ctx
.nr
, &ctx
.entries_nr
,
1290 ctx
.preferred_pack_idx
);
1292 ctx
.large_offsets_needed
= 0;
1293 for (i
= 0; i
< ctx
.entries_nr
; i
++) {
1294 if (ctx
.entries
[i
].offset
> 0x7fffffff)
1295 ctx
.num_large_offsets
++;
1296 if (ctx
.entries
[i
].offset
> 0xffffffff)
1297 ctx
.large_offsets_needed
= 1;
1300 QSORT(ctx
.info
, ctx
.nr
, pack_info_compare
);
1302 if (packs_to_drop
&& packs_to_drop
->nr
) {
1304 int missing_drops
= 0;
1306 for (i
= 0; i
< ctx
.nr
&& drop_index
< packs_to_drop
->nr
; i
++) {
1307 int cmp
= strcmp(ctx
.info
[i
].pack_name
,
1308 packs_to_drop
->items
[drop_index
].string
);
1312 ctx
.info
[i
].expired
= 1;
1313 } else if (cmp
> 0) {
1314 error(_("did not see pack-file %s to drop"),
1315 packs_to_drop
->items
[drop_index
].string
);
1320 ctx
.info
[i
].expired
= 0;
1324 if (missing_drops
) {
1331 * pack_perm stores a permutation between pack-int-ids from the
1332 * previous multi-pack-index to the new one we are writing:
1334 * pack_perm[old_id] = new_id
1336 ALLOC_ARRAY(ctx
.pack_perm
, ctx
.nr
);
1337 for (i
= 0; i
< ctx
.nr
; i
++) {
1338 if (ctx
.info
[i
].expired
) {
1340 ctx
.pack_perm
[ctx
.info
[i
].orig_pack_int_id
] = PACK_EXPIRED
;
1342 ctx
.pack_perm
[ctx
.info
[i
].orig_pack_int_id
] = i
- dropped_packs
;
1346 for (i
= 0; i
< ctx
.nr
; i
++) {
1347 if (!ctx
.info
[i
].expired
)
1348 pack_name_concat_len
+= strlen(ctx
.info
[i
].pack_name
) + 1;
1351 /* Check that the preferred pack wasn't expired (if given). */
1352 if (preferred_pack_name
) {
1353 struct pack_info
*preferred
= bsearch(preferred_pack_name
,
1356 idx_or_pack_name_cmp
);
1358 uint32_t perm
= ctx
.pack_perm
[preferred
->orig_pack_int_id
];
1359 if (perm
== PACK_EXPIRED
)
1360 warning(_("preferred pack '%s' is expired"),
1361 preferred_pack_name
);
1365 if (pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
)
1366 pack_name_concat_len
+= MIDX_CHUNK_ALIGNMENT
-
1367 (pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
);
1369 hold_lock_file_for_update(&lk
, midx_name
, LOCK_DIE_ON_ERROR
);
1370 f
= hashfd(get_lock_file_fd(&lk
), get_lock_file_path(&lk
));
1372 if (ctx
.nr
- dropped_packs
== 0) {
1373 error(_("no pack files to index."));
1378 cf
= init_chunkfile(f
);
1380 add_chunk(cf
, MIDX_CHUNKID_PACKNAMES
, pack_name_concat_len
,
1381 write_midx_pack_names
);
1382 add_chunk(cf
, MIDX_CHUNKID_OIDFANOUT
, MIDX_CHUNK_FANOUT_SIZE
,
1383 write_midx_oid_fanout
);
1384 add_chunk(cf
, MIDX_CHUNKID_OIDLOOKUP
,
1385 (size_t)ctx
.entries_nr
* the_hash_algo
->rawsz
,
1386 write_midx_oid_lookup
);
1387 add_chunk(cf
, MIDX_CHUNKID_OBJECTOFFSETS
,
1388 (size_t)ctx
.entries_nr
* MIDX_CHUNK_OFFSET_WIDTH
,
1389 write_midx_object_offsets
);
1391 if (ctx
.large_offsets_needed
)
1392 add_chunk(cf
, MIDX_CHUNKID_LARGEOFFSETS
,
1393 (size_t)ctx
.num_large_offsets
* MIDX_CHUNK_LARGE_OFFSET_WIDTH
,
1394 write_midx_large_offsets
);
1396 write_midx_header(f
, get_num_chunks(cf
), ctx
.nr
- dropped_packs
);
1397 write_chunkfile(cf
, &ctx
);
1399 finalize_hashfile(f
, midx_hash
, CSUM_FSYNC
| CSUM_HASH_IN_STREAM
);
1402 if (flags
& (MIDX_WRITE_REV_INDEX
| MIDX_WRITE_BITMAP
))
1403 ctx
.pack_order
= midx_pack_order(&ctx
);
1405 if (flags
& MIDX_WRITE_REV_INDEX
)
1406 write_midx_reverse_index(midx_name
, midx_hash
, &ctx
);
1407 if (flags
& MIDX_WRITE_BITMAP
) {
1408 if (write_midx_bitmap(midx_name
, midx_hash
, &ctx
,
1409 refs_snapshot
, flags
) < 0) {
1410 error(_("could not write multi-pack bitmap"));
1417 close_object_store(the_repository
->objects
);
1419 commit_lock_file(&lk
);
1421 clear_midx_files_ext(object_dir
, ".bitmap", midx_hash
);
1422 clear_midx_files_ext(object_dir
, ".rev", midx_hash
);
1425 for (i
= 0; i
< ctx
.nr
; i
++) {
1426 if (ctx
.info
[i
].p
) {
1427 close_pack(ctx
.info
[i
].p
);
1428 free(ctx
.info
[i
].p
);
1430 free(ctx
.info
[i
].pack_name
);
1435 free(ctx
.pack_perm
);
1436 free(ctx
.pack_order
);
1442 int write_midx_file(const char *object_dir
,
1443 const char *preferred_pack_name
,
1444 const char *refs_snapshot
,
1447 return write_midx_internal(object_dir
, NULL
, NULL
, preferred_pack_name
,
1448 refs_snapshot
, flags
);
1451 int write_midx_file_only(const char *object_dir
,
1452 struct string_list
*packs_to_include
,
1453 const char *preferred_pack_name
,
1454 const char *refs_snapshot
,
1457 return write_midx_internal(object_dir
, packs_to_include
, NULL
,
1458 preferred_pack_name
, refs_snapshot
, flags
);
1461 struct clear_midx_data
{
1466 static void clear_midx_file_ext(const char *full_path
, size_t full_path_len
,
1467 const char *file_name
, void *_data
)
1469 struct clear_midx_data
*data
= _data
;
1471 if (!(starts_with(file_name
, "multi-pack-index-") &&
1472 ends_with(file_name
, data
->ext
)))
1474 if (data
->keep
&& !strcmp(data
->keep
, file_name
))
1477 if (unlink(full_path
))
1478 die_errno(_("failed to remove %s"), full_path
);
1481 static void clear_midx_files_ext(const char *object_dir
, const char *ext
,
1482 unsigned char *keep_hash
)
1484 struct clear_midx_data data
;
1485 memset(&data
, 0, sizeof(struct clear_midx_data
));
1488 data
.keep
= xstrfmt("multi-pack-index-%s%s",
1489 hash_to_hex(keep_hash
), ext
);
1492 for_each_file_in_pack_dir(object_dir
,
1493 clear_midx_file_ext
,
1499 void clear_midx_file(struct repository
*r
)
1501 char *midx
= get_midx_filename(r
->objects
->odb
->path
);
1503 if (r
->objects
&& r
->objects
->multi_pack_index
) {
1504 close_midx(r
->objects
->multi_pack_index
);
1505 r
->objects
->multi_pack_index
= NULL
;
1508 if (remove_path(midx
))
1509 die(_("failed to clear multi-pack-index at %s"), midx
);
1511 clear_midx_files_ext(r
->objects
->odb
->path
, ".bitmap", NULL
);
1512 clear_midx_files_ext(r
->objects
->odb
->path
, ".rev", NULL
);
1517 static int verify_midx_error
;
1519 __attribute__((format (printf
, 1, 2)))
1520 static void midx_report(const char *fmt
, ...)
1523 verify_midx_error
= 1;
1525 vfprintf(stderr
, fmt
, ap
);
1526 fprintf(stderr
, "\n");
1530 struct pair_pos_vs_id
1533 uint32_t pack_int_id
;
1536 static int compare_pair_pos_vs_id(const void *_a
, const void *_b
)
1538 struct pair_pos_vs_id
*a
= (struct pair_pos_vs_id
*)_a
;
1539 struct pair_pos_vs_id
*b
= (struct pair_pos_vs_id
*)_b
;
1541 return b
->pack_int_id
- a
->pack_int_id
;
1545 * Limit calls to display_progress() for performance reasons.
1546 * The interval here was arbitrarily chosen.
1548 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1549 #define midx_display_sparse_progress(progress, n) \
1551 uint64_t _n = (n); \
1552 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1553 display_progress(progress, _n); \
1556 int verify_midx_file(struct repository
*r
, const char *object_dir
, unsigned flags
)
1558 struct pair_pos_vs_id
*pairs
= NULL
;
1560 struct progress
*progress
= NULL
;
1561 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
1562 verify_midx_error
= 0;
1567 char *filename
= get_midx_filename(object_dir
);
1568 if (!stat(filename
, &sb
)) {
1569 error(_("multi-pack-index file exists, but failed to parse"));
1576 if (!midx_checksum_valid(m
))
1577 midx_report(_("incorrect checksum"));
1579 if (flags
& MIDX_PROGRESS
)
1580 progress
= start_delayed_progress(_("Looking for referenced packfiles"),
1582 for (i
= 0; i
< m
->num_packs
; i
++) {
1583 if (prepare_midx_pack(r
, m
, i
))
1584 midx_report("failed to load pack in position %d", i
);
1586 display_progress(progress
, i
+ 1);
1588 stop_progress(&progress
);
1590 for (i
= 0; i
< 255; i
++) {
1591 uint32_t oid_fanout1
= ntohl(m
->chunk_oid_fanout
[i
]);
1592 uint32_t oid_fanout2
= ntohl(m
->chunk_oid_fanout
[i
+ 1]);
1594 if (oid_fanout1
> oid_fanout2
)
1595 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32
" > %"PRIx32
" = fanout[%d]"),
1596 i
, oid_fanout1
, oid_fanout2
, i
+ 1);
1599 if (m
->num_objects
== 0) {
1600 midx_report(_("the midx contains no oid"));
1602 * Remaining tests assume that we have objects, so we can
1605 return verify_midx_error
;
1608 if (flags
& MIDX_PROGRESS
)
1609 progress
= start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1610 m
->num_objects
- 1);
1611 for (i
= 0; i
< m
->num_objects
- 1; i
++) {
1612 struct object_id oid1
, oid2
;
1614 nth_midxed_object_oid(&oid1
, m
, i
);
1615 nth_midxed_object_oid(&oid2
, m
, i
+ 1);
1617 if (oidcmp(&oid1
, &oid2
) >= 0)
1618 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1619 i
, oid_to_hex(&oid1
), oid_to_hex(&oid2
), i
+ 1);
1621 midx_display_sparse_progress(progress
, i
+ 1);
1623 stop_progress(&progress
);
1626 * Create an array mapping each object to its packfile id. Sort it
1627 * to group the objects by packfile. Use this permutation to visit
1628 * each of the objects and only require 1 packfile to be open at a
1631 ALLOC_ARRAY(pairs
, m
->num_objects
);
1632 for (i
= 0; i
< m
->num_objects
; i
++) {
1634 pairs
[i
].pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1637 if (flags
& MIDX_PROGRESS
)
1638 progress
= start_sparse_progress(_("Sorting objects by packfile"),
1640 display_progress(progress
, 0); /* TODO: Measure QSORT() progress */
1641 QSORT(pairs
, m
->num_objects
, compare_pair_pos_vs_id
);
1642 stop_progress(&progress
);
1644 if (flags
& MIDX_PROGRESS
)
1645 progress
= start_sparse_progress(_("Verifying object offsets"), m
->num_objects
);
1646 for (i
= 0; i
< m
->num_objects
; i
++) {
1647 struct object_id oid
;
1648 struct pack_entry e
;
1649 off_t m_offset
, p_offset
;
1651 if (i
> 0 && pairs
[i
-1].pack_int_id
!= pairs
[i
].pack_int_id
&&
1652 m
->packs
[pairs
[i
-1].pack_int_id
])
1654 close_pack_fd(m
->packs
[pairs
[i
-1].pack_int_id
]);
1655 close_pack_index(m
->packs
[pairs
[i
-1].pack_int_id
]);
1658 nth_midxed_object_oid(&oid
, m
, pairs
[i
].pos
);
1660 if (!fill_midx_entry(r
, &oid
, &e
, m
)) {
1661 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1662 pairs
[i
].pos
, oid_to_hex(&oid
));
1666 if (open_pack_index(e
.p
)) {
1667 midx_report(_("failed to load pack-index for packfile %s"),
1672 m_offset
= e
.offset
;
1673 p_offset
= find_pack_entry_one(oid
.hash
, e
.p
);
1675 if (m_offset
!= p_offset
)
1676 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64
" != %"PRIx64
),
1677 pairs
[i
].pos
, oid_to_hex(&oid
), m_offset
, p_offset
);
1679 midx_display_sparse_progress(progress
, i
+ 1);
1681 stop_progress(&progress
);
1685 return verify_midx_error
;
1688 int expire_midx_packs(struct repository
*r
, const char *object_dir
, unsigned flags
)
1690 uint32_t i
, *count
, result
= 0;
1691 struct string_list packs_to_drop
= STRING_LIST_INIT_DUP
;
1692 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
1693 struct progress
*progress
= NULL
;
1698 CALLOC_ARRAY(count
, m
->num_packs
);
1700 if (flags
& MIDX_PROGRESS
)
1701 progress
= start_delayed_progress(_("Counting referenced objects"),
1703 for (i
= 0; i
< m
->num_objects
; i
++) {
1704 int pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1705 count
[pack_int_id
]++;
1706 display_progress(progress
, i
+ 1);
1708 stop_progress(&progress
);
1710 if (flags
& MIDX_PROGRESS
)
1711 progress
= start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1713 for (i
= 0; i
< m
->num_packs
; i
++) {
1715 display_progress(progress
, i
+ 1);
1720 if (prepare_midx_pack(r
, m
, i
))
1723 if (m
->packs
[i
]->pack_keep
)
1726 pack_name
= xstrdup(m
->packs
[i
]->pack_name
);
1727 close_pack(m
->packs
[i
]);
1729 string_list_insert(&packs_to_drop
, m
->pack_names
[i
]);
1730 unlink_pack_path(pack_name
, 0);
1733 stop_progress(&progress
);
1737 if (packs_to_drop
.nr
) {
1738 result
= write_midx_internal(object_dir
, NULL
, &packs_to_drop
, NULL
, NULL
, flags
);
1742 string_list_clear(&packs_to_drop
, 0);
1746 struct repack_info
{
1748 uint32_t referenced_objects
;
1749 uint32_t pack_int_id
;
1752 static int compare_by_mtime(const void *a_
, const void *b_
)
1754 const struct repack_info
*a
, *b
;
1756 a
= (const struct repack_info
*)a_
;
1757 b
= (const struct repack_info
*)b_
;
1759 if (a
->mtime
< b
->mtime
)
1761 if (a
->mtime
> b
->mtime
)
1766 static int fill_included_packs_all(struct repository
*r
,
1767 struct multi_pack_index
*m
,
1768 unsigned char *include_pack
)
1770 uint32_t i
, count
= 0;
1771 int pack_kept_objects
= 0;
1773 repo_config_get_bool(r
, "repack.packkeptobjects", &pack_kept_objects
);
1775 for (i
= 0; i
< m
->num_packs
; i
++) {
1776 if (prepare_midx_pack(r
, m
, i
))
1778 if (!pack_kept_objects
&& m
->packs
[i
]->pack_keep
)
1781 include_pack
[i
] = 1;
1788 static int fill_included_packs_batch(struct repository
*r
,
1789 struct multi_pack_index
*m
,
1790 unsigned char *include_pack
,
1793 uint32_t i
, packs_to_repack
;
1795 struct repack_info
*pack_info
= xcalloc(m
->num_packs
, sizeof(struct repack_info
));
1796 int pack_kept_objects
= 0;
1798 repo_config_get_bool(r
, "repack.packkeptobjects", &pack_kept_objects
);
1800 for (i
= 0; i
< m
->num_packs
; i
++) {
1801 pack_info
[i
].pack_int_id
= i
;
1803 if (prepare_midx_pack(r
, m
, i
))
1806 pack_info
[i
].mtime
= m
->packs
[i
]->mtime
;
1809 for (i
= 0; batch_size
&& i
< m
->num_objects
; i
++) {
1810 uint32_t pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1811 pack_info
[pack_int_id
].referenced_objects
++;
1814 QSORT(pack_info
, m
->num_packs
, compare_by_mtime
);
1817 packs_to_repack
= 0;
1818 for (i
= 0; total_size
< batch_size
&& i
< m
->num_packs
; i
++) {
1819 int pack_int_id
= pack_info
[i
].pack_int_id
;
1820 struct packed_git
*p
= m
->packs
[pack_int_id
];
1821 size_t expected_size
;
1825 if (!pack_kept_objects
&& p
->pack_keep
)
1827 if (open_pack_index(p
) || !p
->num_objects
)
1830 expected_size
= (size_t)(p
->pack_size
1831 * pack_info
[i
].referenced_objects
);
1832 expected_size
/= p
->num_objects
;
1834 if (expected_size
>= batch_size
)
1838 total_size
+= expected_size
;
1839 include_pack
[pack_int_id
] = 1;
1844 if (packs_to_repack
< 2)
1850 int midx_repack(struct repository
*r
, const char *object_dir
, size_t batch_size
, unsigned flags
)
1854 unsigned char *include_pack
;
1855 struct child_process cmd
= CHILD_PROCESS_INIT
;
1857 struct strbuf base_name
= STRBUF_INIT
;
1858 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
1861 * When updating the default for these configuration
1862 * variables in builtin/repack.c, these must be adjusted
1865 int delta_base_offset
= 1;
1866 int use_delta_islands
= 0;
1871 CALLOC_ARRAY(include_pack
, m
->num_packs
);
1874 if (fill_included_packs_batch(r
, m
, include_pack
, batch_size
))
1876 } else if (fill_included_packs_all(r
, m
, include_pack
))
1879 repo_config_get_bool(r
, "repack.usedeltabaseoffset", &delta_base_offset
);
1880 repo_config_get_bool(r
, "repack.usedeltaislands", &use_delta_islands
);
1882 strvec_push(&cmd
.args
, "pack-objects");
1884 strbuf_addstr(&base_name
, object_dir
);
1885 strbuf_addstr(&base_name
, "/pack/pack");
1886 strvec_push(&cmd
.args
, base_name
.buf
);
1888 if (delta_base_offset
)
1889 strvec_push(&cmd
.args
, "--delta-base-offset");
1890 if (use_delta_islands
)
1891 strvec_push(&cmd
.args
, "--delta-islands");
1893 if (flags
& MIDX_PROGRESS
)
1894 strvec_push(&cmd
.args
, "--progress");
1896 strvec_push(&cmd
.args
, "-q");
1898 strbuf_release(&base_name
);
1901 cmd
.in
= cmd
.out
= -1;
1903 if (start_command(&cmd
)) {
1904 error(_("could not start pack-objects"));
1909 cmd_in
= xfdopen(cmd
.in
, "w");
1911 for (i
= 0; i
< m
->num_objects
; i
++) {
1912 struct object_id oid
;
1913 uint32_t pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1915 if (!include_pack
[pack_int_id
])
1918 nth_midxed_object_oid(&oid
, m
, i
);
1919 fprintf(cmd_in
, "%s\n", oid_to_hex(&oid
));
1923 if (finish_command(&cmd
)) {
1924 error(_("could not finish pack-objects"));
1929 result
= write_midx_internal(object_dir
, NULL
, NULL
, NULL
, NULL
, flags
);