1 #include "git-compat-util.h"
10 #include "object-file.h"
11 #include "object-store-ll.h"
12 #include "hash-lookup.h"
16 #include "run-command.h"
17 #include "repository.h"
18 #include "chunk-format.h"
20 #include "pack-bitmap.h"
23 #include "list-objects.h"
25 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
26 #define MIDX_VERSION 1
27 #define MIDX_BYTE_FILE_VERSION 4
28 #define MIDX_BYTE_HASH_VERSION 5
29 #define MIDX_BYTE_NUM_CHUNKS 6
30 #define MIDX_BYTE_NUM_PACKS 8
31 #define MIDX_HEADER_SIZE 12
32 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
34 #define MIDX_CHUNK_ALIGNMENT 4
35 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
36 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
37 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
38 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
39 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
40 #define MIDX_CHUNKID_REVINDEX 0x52494458 /* "RIDX" */
41 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
42 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
43 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
44 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
46 #define PACK_EXPIRED UINT_MAX
48 const unsigned char *get_midx_checksum(struct multi_pack_index
*m
)
50 return m
->data
+ m
->data_len
- the_hash_algo
->rawsz
;
53 void get_midx_filename(struct strbuf
*out
, const char *object_dir
)
55 strbuf_addf(out
, "%s/pack/multi-pack-index", object_dir
);
58 void get_midx_rev_filename(struct strbuf
*out
, struct multi_pack_index
*m
)
60 get_midx_filename(out
, m
->object_dir
);
61 strbuf_addf(out
, "-%s.rev", hash_to_hex(get_midx_checksum(m
)));
64 static int midx_read_oid_fanout(const unsigned char *chunk_start
,
65 size_t chunk_size
, void *data
)
68 struct multi_pack_index
*m
= data
;
69 m
->chunk_oid_fanout
= (uint32_t *)chunk_start
;
71 if (chunk_size
!= 4 * 256) {
72 error(_("multi-pack-index OID fanout is of the wrong size"));
75 for (i
= 0; i
< 255; i
++) {
76 uint32_t oid_fanout1
= ntohl(m
->chunk_oid_fanout
[i
]);
77 uint32_t oid_fanout2
= ntohl(m
->chunk_oid_fanout
[i
+1]);
79 if (oid_fanout1
> oid_fanout2
) {
80 error(_("oid fanout out of order: fanout[%d] = %"PRIx32
" > %"PRIx32
" = fanout[%d]"),
81 i
, oid_fanout1
, oid_fanout2
, i
+ 1);
85 m
->num_objects
= ntohl(m
->chunk_oid_fanout
[255]);
89 static int midx_read_oid_lookup(const unsigned char *chunk_start
,
90 size_t chunk_size
, void *data
)
92 struct multi_pack_index
*m
= data
;
93 m
->chunk_oid_lookup
= chunk_start
;
95 if (chunk_size
!= st_mult(m
->hash_len
, m
->num_objects
)) {
96 error(_("multi-pack-index OID lookup chunk is the wrong size"));
102 static int midx_read_object_offsets(const unsigned char *chunk_start
,
103 size_t chunk_size
, void *data
)
105 struct multi_pack_index
*m
= data
;
106 m
->chunk_object_offsets
= chunk_start
;
108 if (chunk_size
!= st_mult(m
->num_objects
, MIDX_CHUNK_OFFSET_WIDTH
)) {
109 error(_("multi-pack-index object offset chunk is the wrong size"));
115 struct multi_pack_index
*load_multi_pack_index(const char *object_dir
, int local
)
117 struct multi_pack_index
*m
= NULL
;
121 void *midx_map
= NULL
;
122 uint32_t hash_version
;
123 struct strbuf midx_name
= STRBUF_INIT
;
125 const char *cur_pack_name
;
126 struct chunkfile
*cf
= NULL
;
128 get_midx_filename(&midx_name
, object_dir
);
130 fd
= git_open(midx_name
.buf
);
134 if (fstat(fd
, &st
)) {
135 error_errno(_("failed to read %s"), midx_name
.buf
);
139 midx_size
= xsize_t(st
.st_size
);
141 if (midx_size
< MIDX_MIN_SIZE
) {
142 error(_("multi-pack-index file %s is too small"), midx_name
.buf
);
146 strbuf_release(&midx_name
);
148 midx_map
= xmmap(NULL
, midx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
151 FLEX_ALLOC_STR(m
, object_dir
, object_dir
);
153 m
->data_len
= midx_size
;
156 m
->signature
= get_be32(m
->data
);
157 if (m
->signature
!= MIDX_SIGNATURE
)
158 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
159 m
->signature
, MIDX_SIGNATURE
);
161 m
->version
= m
->data
[MIDX_BYTE_FILE_VERSION
];
162 if (m
->version
!= MIDX_VERSION
)
163 die(_("multi-pack-index version %d not recognized"),
166 hash_version
= m
->data
[MIDX_BYTE_HASH_VERSION
];
167 if (hash_version
!= oid_version(the_hash_algo
)) {
168 error(_("multi-pack-index hash version %u does not match version %u"),
169 hash_version
, oid_version(the_hash_algo
));
172 m
->hash_len
= the_hash_algo
->rawsz
;
174 m
->num_chunks
= m
->data
[MIDX_BYTE_NUM_CHUNKS
];
176 m
->num_packs
= get_be32(m
->data
+ MIDX_BYTE_NUM_PACKS
);
178 cf
= init_chunkfile(NULL
);
180 if (read_table_of_contents(cf
, m
->data
, midx_size
,
181 MIDX_HEADER_SIZE
, m
->num_chunks
,
182 MIDX_CHUNK_ALIGNMENT
))
185 if (pair_chunk(cf
, MIDX_CHUNKID_PACKNAMES
, &m
->chunk_pack_names
, &m
->chunk_pack_names_len
))
186 die(_("multi-pack-index required pack-name chunk missing or corrupted"));
187 if (read_chunk(cf
, MIDX_CHUNKID_OIDFANOUT
, midx_read_oid_fanout
, m
))
188 die(_("multi-pack-index required OID fanout chunk missing or corrupted"));
189 if (read_chunk(cf
, MIDX_CHUNKID_OIDLOOKUP
, midx_read_oid_lookup
, m
))
190 die(_("multi-pack-index required OID lookup chunk missing or corrupted"));
191 if (read_chunk(cf
, MIDX_CHUNKID_OBJECTOFFSETS
, midx_read_object_offsets
, m
))
192 die(_("multi-pack-index required object offsets chunk missing or corrupted"));
194 pair_chunk(cf
, MIDX_CHUNKID_LARGEOFFSETS
, &m
->chunk_large_offsets
,
195 &m
->chunk_large_offsets_len
);
197 if (git_env_bool("GIT_TEST_MIDX_READ_RIDX", 1))
198 pair_chunk(cf
, MIDX_CHUNKID_REVINDEX
, &m
->chunk_revindex
,
199 &m
->chunk_revindex_len
);
201 CALLOC_ARRAY(m
->pack_names
, m
->num_packs
);
202 CALLOC_ARRAY(m
->packs
, m
->num_packs
);
204 cur_pack_name
= (const char *)m
->chunk_pack_names
;
205 for (i
= 0; i
< m
->num_packs
; i
++) {
207 size_t avail
= m
->chunk_pack_names_len
-
208 (cur_pack_name
- (const char *)m
->chunk_pack_names
);
210 m
->pack_names
[i
] = cur_pack_name
;
212 end
= memchr(cur_pack_name
, '\0', avail
);
214 die(_("multi-pack-index pack-name chunk is too short"));
215 cur_pack_name
= end
+ 1;
217 if (i
&& strcmp(m
->pack_names
[i
], m
->pack_names
[i
- 1]) <= 0)
218 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
219 m
->pack_names
[i
- 1],
223 trace2_data_intmax("midx", the_repository
, "load/num_packs", m
->num_packs
);
224 trace2_data_intmax("midx", the_repository
, "load/num_objects", m
->num_objects
);
231 strbuf_release(&midx_name
);
234 munmap(midx_map
, midx_size
);
240 void close_midx(struct multi_pack_index
*m
)
249 munmap((unsigned char *)m
->data
, m
->data_len
);
251 for (i
= 0; i
< m
->num_packs
; i
++) {
253 m
->packs
[i
]->multi_pack_index
= 0;
255 FREE_AND_NULL(m
->packs
);
256 FREE_AND_NULL(m
->pack_names
);
260 int prepare_midx_pack(struct repository
*r
, struct multi_pack_index
*m
, uint32_t pack_int_id
)
262 struct strbuf pack_name
= STRBUF_INIT
;
263 struct packed_git
*p
;
265 if (pack_int_id
>= m
->num_packs
)
266 die(_("bad pack-int-id: %u (%u total packs)"),
267 pack_int_id
, m
->num_packs
);
269 if (m
->packs
[pack_int_id
])
272 strbuf_addf(&pack_name
, "%s/pack/%s", m
->object_dir
,
273 m
->pack_names
[pack_int_id
]);
275 p
= add_packed_git(pack_name
.buf
, pack_name
.len
, m
->local
);
276 strbuf_release(&pack_name
);
281 p
->multi_pack_index
= 1;
282 m
->packs
[pack_int_id
] = p
;
283 install_packed_git(r
, p
);
284 list_add_tail(&p
->mru
, &r
->objects
->packed_git_mru
);
289 int bsearch_midx(const struct object_id
*oid
, struct multi_pack_index
*m
, uint32_t *result
)
291 return bsearch_hash(oid
->hash
, m
->chunk_oid_fanout
, m
->chunk_oid_lookup
,
292 the_hash_algo
->rawsz
, result
);
295 struct object_id
*nth_midxed_object_oid(struct object_id
*oid
,
296 struct multi_pack_index
*m
,
299 if (n
>= m
->num_objects
)
302 oidread(oid
, m
->chunk_oid_lookup
+ st_mult(m
->hash_len
, n
));
306 off_t
nth_midxed_offset(struct multi_pack_index
*m
, uint32_t pos
)
308 const unsigned char *offset_data
;
311 offset_data
= m
->chunk_object_offsets
+ (off_t
)pos
* MIDX_CHUNK_OFFSET_WIDTH
;
312 offset32
= get_be32(offset_data
+ sizeof(uint32_t));
314 if (m
->chunk_large_offsets
&& offset32
& MIDX_LARGE_OFFSET_NEEDED
) {
315 if (sizeof(off_t
) < sizeof(uint64_t))
316 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
318 offset32
^= MIDX_LARGE_OFFSET_NEEDED
;
319 if (offset32
>= m
->chunk_large_offsets_len
/ sizeof(uint64_t))
320 die(_("multi-pack-index large offset out of bounds"));
321 return get_be64(m
->chunk_large_offsets
+ sizeof(uint64_t) * offset32
);
327 uint32_t nth_midxed_pack_int_id(struct multi_pack_index
*m
, uint32_t pos
)
329 return get_be32(m
->chunk_object_offsets
+
330 (off_t
)pos
* MIDX_CHUNK_OFFSET_WIDTH
);
333 int fill_midx_entry(struct repository
*r
,
334 const struct object_id
*oid
,
335 struct pack_entry
*e
,
336 struct multi_pack_index
*m
)
339 uint32_t pack_int_id
;
340 struct packed_git
*p
;
342 if (!bsearch_midx(oid
, m
, &pos
))
345 if (pos
>= m
->num_objects
)
348 pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
350 if (prepare_midx_pack(r
, m
, pack_int_id
))
352 p
= m
->packs
[pack_int_id
];
355 * We are about to tell the caller where they can locate the
356 * requested object. We better make sure the packfile is
357 * still here and can be accessed before supplying that
358 * answer, as it may have been deleted since the MIDX was
361 if (!is_pack_valid(p
))
364 if (oidset_size(&p
->bad_objects
) &&
365 oidset_contains(&p
->bad_objects
, oid
))
368 e
->offset
= nth_midxed_offset(m
, pos
);
374 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
375 static int cmp_idx_or_pack_name(const char *idx_or_pack_name
,
376 const char *idx_name
)
378 /* Skip past any initial matching prefix. */
379 while (*idx_name
&& *idx_name
== *idx_or_pack_name
) {
385 * If we didn't match completely, we may have matched "pack-1234." and
386 * be left with "idx" and "pack" respectively, which is also OK. We do
387 * not have to check for "idx" and "idx", because that would have been
388 * a complete match (and in that case these strcmps will be false, but
389 * we'll correctly return 0 from the final strcmp() below.
391 * Technically this matches "fooidx" and "foopack", but we'd never have
392 * such names in the first place.
394 if (!strcmp(idx_name
, "idx") && !strcmp(idx_or_pack_name
, "pack"))
398 * This not only checks for a complete match, but also orders based on
399 * the first non-identical character, which means our ordering will
400 * match a raw strcmp(). That makes it OK to use this to binary search
401 * a naively-sorted list.
403 return strcmp(idx_or_pack_name
, idx_name
);
406 int midx_contains_pack(struct multi_pack_index
*m
, const char *idx_or_pack_name
)
408 uint32_t first
= 0, last
= m
->num_packs
;
410 while (first
< last
) {
411 uint32_t mid
= first
+ (last
- first
) / 2;
415 current
= m
->pack_names
[mid
];
416 cmp
= cmp_idx_or_pack_name(idx_or_pack_name
, current
);
429 int prepare_multi_pack_index_one(struct repository
*r
, const char *object_dir
, int local
)
431 struct multi_pack_index
*m
;
432 struct multi_pack_index
*m_search
;
434 prepare_repo_settings(r
);
435 if (!r
->settings
.core_multi_pack_index
)
438 for (m_search
= r
->objects
->multi_pack_index
; m_search
; m_search
= m_search
->next
)
439 if (!strcmp(object_dir
, m_search
->object_dir
))
442 m
= load_multi_pack_index(object_dir
, local
);
445 struct multi_pack_index
*mp
= r
->objects
->multi_pack_index
;
450 r
->objects
->multi_pack_index
= m
;
457 static size_t write_midx_header(struct hashfile
*f
,
458 unsigned char num_chunks
,
461 hashwrite_be32(f
, MIDX_SIGNATURE
);
462 hashwrite_u8(f
, MIDX_VERSION
);
463 hashwrite_u8(f
, oid_version(the_hash_algo
));
464 hashwrite_u8(f
, num_chunks
);
465 hashwrite_u8(f
, 0); /* unused */
466 hashwrite_be32(f
, num_packs
);
468 return MIDX_HEADER_SIZE
;
472 uint32_t orig_pack_int_id
;
474 struct packed_git
*p
;
475 unsigned expired
: 1;
478 static int pack_info_compare(const void *_a
, const void *_b
)
480 struct pack_info
*a
= (struct pack_info
*)_a
;
481 struct pack_info
*b
= (struct pack_info
*)_b
;
482 return strcmp(a
->pack_name
, b
->pack_name
);
485 static int idx_or_pack_name_cmp(const void *_va
, const void *_vb
)
487 const char *pack_name
= _va
;
488 const struct pack_info
*compar
= _vb
;
490 return cmp_idx_or_pack_name(pack_name
, compar
->pack_name
);
493 struct write_midx_context
{
494 struct pack_info
*info
;
497 struct multi_pack_index
*m
;
498 struct progress
*progress
;
499 unsigned pack_paths_checked
;
501 struct pack_midx_entry
*entries
;
505 uint32_t *pack_order
;
506 unsigned large_offsets_needed
:1;
507 uint32_t num_large_offsets
;
509 int preferred_pack_idx
;
511 struct string_list
*to_include
;
514 static void add_pack_to_midx(const char *full_path
, size_t full_path_len
,
515 const char *file_name
, void *data
)
517 struct write_midx_context
*ctx
= data
;
519 if (ends_with(file_name
, ".idx")) {
520 display_progress(ctx
->progress
, ++ctx
->pack_paths_checked
);
522 * Note that at most one of ctx->m and ctx->to_include are set,
523 * so we are testing midx_contains_pack() and
524 * string_list_has_string() independently (guarded by the
525 * appropriate NULL checks).
527 * We could support passing to_include while reusing an existing
528 * MIDX, but don't currently since the reuse process drags
529 * forward all packs from an existing MIDX (without checking
530 * whether or not they appear in the to_include list).
532 * If we added support for that, these next two conditional
533 * should be performed independently (likely checking
534 * to_include before the existing MIDX).
536 if (ctx
->m
&& midx_contains_pack(ctx
->m
, file_name
))
538 else if (ctx
->to_include
&&
539 !string_list_has_string(ctx
->to_include
, file_name
))
542 ALLOC_GROW(ctx
->info
, ctx
->nr
+ 1, ctx
->alloc
);
544 ctx
->info
[ctx
->nr
].p
= add_packed_git(full_path
,
548 if (!ctx
->info
[ctx
->nr
].p
) {
549 warning(_("failed to add packfile '%s'"),
554 if (open_pack_index(ctx
->info
[ctx
->nr
].p
)) {
555 warning(_("failed to open pack-index '%s'"),
557 close_pack(ctx
->info
[ctx
->nr
].p
);
558 FREE_AND_NULL(ctx
->info
[ctx
->nr
].p
);
562 ctx
->info
[ctx
->nr
].pack_name
= xstrdup(file_name
);
563 ctx
->info
[ctx
->nr
].orig_pack_int_id
= ctx
->nr
;
564 ctx
->info
[ctx
->nr
].expired
= 0;
569 struct pack_midx_entry
{
570 struct object_id oid
;
571 uint32_t pack_int_id
;
574 unsigned preferred
: 1;
577 static int midx_oid_compare(const void *_a
, const void *_b
)
579 const struct pack_midx_entry
*a
= (const struct pack_midx_entry
*)_a
;
580 const struct pack_midx_entry
*b
= (const struct pack_midx_entry
*)_b
;
581 int cmp
= oidcmp(&a
->oid
, &b
->oid
);
586 /* Sort objects in a preferred pack first when multiple copies exist. */
587 if (a
->preferred
> b
->preferred
)
589 if (a
->preferred
< b
->preferred
)
592 if (a
->pack_mtime
> b
->pack_mtime
)
594 else if (a
->pack_mtime
< b
->pack_mtime
)
597 return a
->pack_int_id
- b
->pack_int_id
;
600 static int nth_midxed_pack_midx_entry(struct multi_pack_index
*m
,
601 struct pack_midx_entry
*e
,
604 if (pos
>= m
->num_objects
)
607 nth_midxed_object_oid(&e
->oid
, m
, pos
);
608 e
->pack_int_id
= nth_midxed_pack_int_id(m
, pos
);
609 e
->offset
= nth_midxed_offset(m
, pos
);
611 /* consider objects in midx to be from "old" packs */
616 static void fill_pack_entry(uint32_t pack_int_id
,
617 struct packed_git
*p
,
619 struct pack_midx_entry
*entry
,
622 if (nth_packed_object_id(&entry
->oid
, p
, cur_object
) < 0)
623 die(_("failed to locate object %d in packfile"), cur_object
);
625 entry
->pack_int_id
= pack_int_id
;
626 entry
->pack_mtime
= p
->mtime
;
628 entry
->offset
= nth_packed_object_offset(p
, cur_object
);
629 entry
->preferred
= !!preferred
;
633 struct pack_midx_entry
*entries
;
637 static void midx_fanout_grow(struct midx_fanout
*fanout
, size_t nr
)
640 BUG("negative growth in midx_fanout_grow() (%"PRIuMAX
" < %"PRIuMAX
")",
641 (uintmax_t)nr
, (uintmax_t)fanout
->nr
);
642 ALLOC_GROW(fanout
->entries
, nr
, fanout
->alloc
);
645 static void midx_fanout_sort(struct midx_fanout
*fanout
)
647 QSORT(fanout
->entries
, fanout
->nr
, midx_oid_compare
);
650 static void midx_fanout_add_midx_fanout(struct midx_fanout
*fanout
,
651 struct multi_pack_index
*m
,
655 uint32_t start
= 0, end
;
659 start
= ntohl(m
->chunk_oid_fanout
[cur_fanout
- 1]);
660 end
= ntohl(m
->chunk_oid_fanout
[cur_fanout
]);
662 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
663 if ((preferred_pack
> -1) &&
664 (preferred_pack
== nth_midxed_pack_int_id(m
, cur_object
))) {
666 * Objects from preferred packs are added
672 midx_fanout_grow(fanout
, fanout
->nr
+ 1);
673 nth_midxed_pack_midx_entry(m
,
674 &fanout
->entries
[fanout
->nr
],
676 fanout
->entries
[fanout
->nr
].preferred
= 0;
681 static void midx_fanout_add_pack_fanout(struct midx_fanout
*fanout
,
682 struct pack_info
*info
,
687 struct packed_git
*pack
= info
[cur_pack
].p
;
688 uint32_t start
= 0, end
;
692 start
= get_pack_fanout(pack
, cur_fanout
- 1);
693 end
= get_pack_fanout(pack
, cur_fanout
);
695 for (cur_object
= start
; cur_object
< end
; cur_object
++) {
696 midx_fanout_grow(fanout
, fanout
->nr
+ 1);
697 fill_pack_entry(cur_pack
,
700 &fanout
->entries
[fanout
->nr
],
707 * It is possible to artificially get into a state where there are many
708 * duplicate copies of objects. That can create high memory pressure if
709 * we are to create a list of all objects before de-duplication. To reduce
710 * this memory pressure without a significant performance drop, automatically
711 * group objects by the first byte of their object id. Use the IDX fanout
712 * tables to group the data, copy to a local array, then sort.
714 * Copy only the de-duplicated entries (selected by most-recent modified time
715 * of a packfile containing the object).
717 static struct pack_midx_entry
*get_sorted_entries(struct multi_pack_index
*m
,
718 struct pack_info
*info
,
723 uint32_t cur_fanout
, cur_pack
, cur_object
;
724 size_t alloc_objects
, total_objects
= 0;
725 struct midx_fanout fanout
= { 0 };
726 struct pack_midx_entry
*deduplicated_entries
= NULL
;
727 uint32_t start_pack
= m
? m
->num_packs
: 0;
729 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++)
730 total_objects
= st_add(total_objects
,
731 info
[cur_pack
].p
->num_objects
);
734 * As we de-duplicate by fanout value, we expect the fanout
735 * slices to be evenly distributed, with some noise. Hence,
736 * allocate slightly more than one 256th.
738 alloc_objects
= fanout
.alloc
= total_objects
> 3200 ? total_objects
/ 200 : 16;
740 ALLOC_ARRAY(fanout
.entries
, fanout
.alloc
);
741 ALLOC_ARRAY(deduplicated_entries
, alloc_objects
);
744 for (cur_fanout
= 0; cur_fanout
< 256; cur_fanout
++) {
748 midx_fanout_add_midx_fanout(&fanout
, m
, cur_fanout
,
751 for (cur_pack
= start_pack
; cur_pack
< nr_packs
; cur_pack
++) {
752 int preferred
= cur_pack
== preferred_pack
;
753 midx_fanout_add_pack_fanout(&fanout
,
755 preferred
, cur_fanout
);
758 if (-1 < preferred_pack
&& preferred_pack
< start_pack
)
759 midx_fanout_add_pack_fanout(&fanout
, info
,
763 midx_fanout_sort(&fanout
);
766 * The batch is now sorted by OID and then mtime (descending).
767 * Take only the first duplicate.
769 for (cur_object
= 0; cur_object
< fanout
.nr
; cur_object
++) {
770 if (cur_object
&& oideq(&fanout
.entries
[cur_object
- 1].oid
,
771 &fanout
.entries
[cur_object
].oid
))
774 ALLOC_GROW(deduplicated_entries
, st_add(*nr_objects
, 1),
776 memcpy(&deduplicated_entries
[*nr_objects
],
777 &fanout
.entries
[cur_object
],
778 sizeof(struct pack_midx_entry
));
783 free(fanout
.entries
);
784 return deduplicated_entries
;
787 static int write_midx_pack_names(struct hashfile
*f
, void *data
)
789 struct write_midx_context
*ctx
= data
;
791 unsigned char padding
[MIDX_CHUNK_ALIGNMENT
];
794 for (i
= 0; i
< ctx
->nr
; i
++) {
797 if (ctx
->info
[i
].expired
)
800 if (i
&& strcmp(ctx
->info
[i
].pack_name
, ctx
->info
[i
- 1].pack_name
) <= 0)
801 BUG("incorrect pack-file order: %s before %s",
802 ctx
->info
[i
- 1].pack_name
,
803 ctx
->info
[i
].pack_name
);
805 writelen
= strlen(ctx
->info
[i
].pack_name
) + 1;
806 hashwrite(f
, ctx
->info
[i
].pack_name
, writelen
);
810 /* add padding to be aligned */
811 i
= MIDX_CHUNK_ALIGNMENT
- (written
% MIDX_CHUNK_ALIGNMENT
);
812 if (i
< MIDX_CHUNK_ALIGNMENT
) {
813 memset(padding
, 0, sizeof(padding
));
814 hashwrite(f
, padding
, i
);
820 static int write_midx_oid_fanout(struct hashfile
*f
,
823 struct write_midx_context
*ctx
= data
;
824 struct pack_midx_entry
*list
= ctx
->entries
;
825 struct pack_midx_entry
*last
= ctx
->entries
+ ctx
->entries_nr
;
830 * Write the first-level table (the list is sorted,
831 * but we use a 256-entry lookup to be able to avoid
832 * having to do eight extra binary search iterations).
834 for (i
= 0; i
< 256; i
++) {
835 struct pack_midx_entry
*next
= list
;
837 while (next
< last
&& next
->oid
.hash
[0] == i
) {
842 hashwrite_be32(f
, count
);
849 static int write_midx_oid_lookup(struct hashfile
*f
,
852 struct write_midx_context
*ctx
= data
;
853 unsigned char hash_len
= the_hash_algo
->rawsz
;
854 struct pack_midx_entry
*list
= ctx
->entries
;
857 for (i
= 0; i
< ctx
->entries_nr
; i
++) {
858 struct pack_midx_entry
*obj
= list
++;
860 if (i
< ctx
->entries_nr
- 1) {
861 struct pack_midx_entry
*next
= list
;
862 if (oidcmp(&obj
->oid
, &next
->oid
) >= 0)
863 BUG("OIDs not in order: %s >= %s",
864 oid_to_hex(&obj
->oid
),
865 oid_to_hex(&next
->oid
));
868 hashwrite(f
, obj
->oid
.hash
, (int)hash_len
);
874 static int write_midx_object_offsets(struct hashfile
*f
,
877 struct write_midx_context
*ctx
= data
;
878 struct pack_midx_entry
*list
= ctx
->entries
;
879 uint32_t i
, nr_large_offset
= 0;
881 for (i
= 0; i
< ctx
->entries_nr
; i
++) {
882 struct pack_midx_entry
*obj
= list
++;
884 if (ctx
->pack_perm
[obj
->pack_int_id
] == PACK_EXPIRED
)
885 BUG("object %s is in an expired pack with int-id %d",
886 oid_to_hex(&obj
->oid
),
889 hashwrite_be32(f
, ctx
->pack_perm
[obj
->pack_int_id
]);
891 if (ctx
->large_offsets_needed
&& obj
->offset
>> 31)
892 hashwrite_be32(f
, MIDX_LARGE_OFFSET_NEEDED
| nr_large_offset
++);
893 else if (!ctx
->large_offsets_needed
&& obj
->offset
>> 32)
894 BUG("object %s requires a large offset (%"PRIx64
") but the MIDX is not writing large offsets!",
895 oid_to_hex(&obj
->oid
),
898 hashwrite_be32(f
, (uint32_t)obj
->offset
);
904 static int write_midx_large_offsets(struct hashfile
*f
,
907 struct write_midx_context
*ctx
= data
;
908 struct pack_midx_entry
*list
= ctx
->entries
;
909 struct pack_midx_entry
*end
= ctx
->entries
+ ctx
->entries_nr
;
910 uint32_t nr_large_offset
= ctx
->num_large_offsets
;
912 while (nr_large_offset
) {
913 struct pack_midx_entry
*obj
;
917 BUG("too many large-offset objects");
920 offset
= obj
->offset
;
925 hashwrite_be64(f
, offset
);
933 static int write_midx_revindex(struct hashfile
*f
,
936 struct write_midx_context
*ctx
= data
;
939 for (i
= 0; i
< ctx
->entries_nr
; i
++)
940 hashwrite_be32(f
, ctx
->pack_order
[i
]);
945 struct midx_pack_order_data
{
951 static int midx_pack_order_cmp(const void *va
, const void *vb
)
953 const struct midx_pack_order_data
*a
= va
, *b
= vb
;
954 if (a
->pack
< b
->pack
)
956 else if (a
->pack
> b
->pack
)
958 else if (a
->offset
< b
->offset
)
960 else if (a
->offset
> b
->offset
)
966 static uint32_t *midx_pack_order(struct write_midx_context
*ctx
)
968 struct midx_pack_order_data
*data
;
969 uint32_t *pack_order
;
972 trace2_region_enter("midx", "midx_pack_order", the_repository
);
974 ALLOC_ARRAY(data
, ctx
->entries_nr
);
975 for (i
= 0; i
< ctx
->entries_nr
; i
++) {
976 struct pack_midx_entry
*e
= &ctx
->entries
[i
];
978 data
[i
].pack
= ctx
->pack_perm
[e
->pack_int_id
];
980 data
[i
].pack
|= (1U << 31);
981 data
[i
].offset
= e
->offset
;
984 QSORT(data
, ctx
->entries_nr
, midx_pack_order_cmp
);
986 ALLOC_ARRAY(pack_order
, ctx
->entries_nr
);
987 for (i
= 0; i
< ctx
->entries_nr
; i
++)
988 pack_order
[i
] = data
[i
].nr
;
991 trace2_region_leave("midx", "midx_pack_order", the_repository
);
996 static void write_midx_reverse_index(char *midx_name
, unsigned char *midx_hash
,
997 struct write_midx_context
*ctx
)
999 struct strbuf buf
= STRBUF_INIT
;
1000 const char *tmp_file
;
1002 trace2_region_enter("midx", "write_midx_reverse_index", the_repository
);
1004 strbuf_addf(&buf
, "%s-%s.rev", midx_name
, hash_to_hex(midx_hash
));
1006 tmp_file
= write_rev_file_order(NULL
, ctx
->pack_order
, ctx
->entries_nr
,
1007 midx_hash
, WRITE_REV
);
1009 if (finalize_object_file(tmp_file
, buf
.buf
))
1010 die(_("cannot store reverse index file"));
1012 strbuf_release(&buf
);
1014 trace2_region_leave("midx", "write_midx_reverse_index", the_repository
);
1017 static void clear_midx_files_ext(const char *object_dir
, const char *ext
,
1018 unsigned char *keep_hash
);
1020 static int midx_checksum_valid(struct multi_pack_index
*m
)
1022 return hashfile_checksum_valid(m
->data
, m
->data_len
);
1025 static void prepare_midx_packing_data(struct packing_data
*pdata
,
1026 struct write_midx_context
*ctx
)
1030 trace2_region_enter("midx", "prepare_midx_packing_data", the_repository
);
1032 memset(pdata
, 0, sizeof(struct packing_data
));
1033 prepare_packing_data(the_repository
, pdata
);
1035 for (i
= 0; i
< ctx
->entries_nr
; i
++) {
1036 struct pack_midx_entry
*from
= &ctx
->entries
[ctx
->pack_order
[i
]];
1037 struct object_entry
*to
= packlist_alloc(pdata
, &from
->oid
);
1039 oe_set_in_pack(pdata
, to
,
1040 ctx
->info
[ctx
->pack_perm
[from
->pack_int_id
]].p
);
1043 trace2_region_leave("midx", "prepare_midx_packing_data", the_repository
);
1046 static int add_ref_to_pending(const char *refname
,
1047 const struct object_id
*oid
,
1048 int flag
, void *cb_data
)
1050 struct rev_info
*revs
= (struct rev_info
*)cb_data
;
1051 struct object_id peeled
;
1052 struct object
*object
;
1054 if ((flag
& REF_ISSYMREF
) && (flag
& REF_ISBROKEN
)) {
1055 warning("symbolic ref is dangling: %s", refname
);
1059 if (!peel_iterated_oid(oid
, &peeled
))
1062 object
= parse_object_or_die(oid
, refname
);
1063 if (object
->type
!= OBJ_COMMIT
)
1066 add_pending_object(revs
, object
, "");
1067 if (bitmap_is_preferred_refname(revs
->repo
, refname
))
1068 object
->flags
|= NEEDS_BITMAP
;
1072 struct bitmap_commit_cb
{
1073 struct commit
**commits
;
1074 size_t commits_nr
, commits_alloc
;
1076 struct write_midx_context
*ctx
;
1079 static const struct object_id
*bitmap_oid_access(size_t index
,
1080 const void *_entries
)
1082 const struct pack_midx_entry
*entries
= _entries
;
1083 return &entries
[index
].oid
;
1086 static void bitmap_show_commit(struct commit
*commit
, void *_data
)
1088 struct bitmap_commit_cb
*data
= _data
;
1089 int pos
= oid_pos(&commit
->object
.oid
, data
->ctx
->entries
,
1090 data
->ctx
->entries_nr
,
1095 ALLOC_GROW(data
->commits
, data
->commits_nr
+ 1, data
->commits_alloc
);
1096 data
->commits
[data
->commits_nr
++] = commit
;
1099 static int read_refs_snapshot(const char *refs_snapshot
,
1100 struct rev_info
*revs
)
1102 struct strbuf buf
= STRBUF_INIT
;
1103 struct object_id oid
;
1104 FILE *f
= xfopen(refs_snapshot
, "r");
1106 while (strbuf_getline(&buf
, f
) != EOF
) {
1107 struct object
*object
;
1109 char *hex
= buf
.buf
;
1110 const char *end
= NULL
;
1112 if (buf
.len
&& *buf
.buf
== '+') {
1117 if (parse_oid_hex(hex
, &oid
, &end
) < 0)
1118 die(_("could not parse line: %s"), buf
.buf
);
1120 die(_("malformed line: %s"), buf
.buf
);
1122 object
= parse_object_or_die(&oid
, NULL
);
1124 object
->flags
|= NEEDS_BITMAP
;
1126 add_pending_object(revs
, object
, "");
1130 strbuf_release(&buf
);
1134 static struct commit
**find_commits_for_midx_bitmap(uint32_t *indexed_commits_nr_p
,
1135 const char *refs_snapshot
,
1136 struct write_midx_context
*ctx
)
1138 struct rev_info revs
;
1139 struct bitmap_commit_cb cb
= {0};
1141 trace2_region_enter("midx", "find_commits_for_midx_bitmap",
1146 repo_init_revisions(the_repository
, &revs
, NULL
);
1147 if (refs_snapshot
) {
1148 read_refs_snapshot(refs_snapshot
, &revs
);
1150 setup_revisions(0, NULL
, &revs
, NULL
);
1151 for_each_ref(add_ref_to_pending
, &revs
);
1155 * Skipping promisor objects here is intentional, since it only excludes
1156 * them from the list of reachable commits that we want to select from
1157 * when computing the selection of MIDX'd commits to receive bitmaps.
1159 * Reachability bitmaps do require that their objects be closed under
1160 * reachability, but fetching any objects missing from promisors at this
1161 * point is too late. But, if one of those objects can be reached from
1162 * an another object that is included in the bitmap, then we will
1163 * complain later that we don't have reachability closure (and fail
1166 fetch_if_missing
= 0;
1167 revs
.exclude_promisor_objects
= 1;
1169 if (prepare_revision_walk(&revs
))
1170 die(_("revision walk setup failed"));
1172 traverse_commit_list(&revs
, bitmap_show_commit
, NULL
, &cb
);
1173 if (indexed_commits_nr_p
)
1174 *indexed_commits_nr_p
= cb
.commits_nr
;
1176 release_revisions(&revs
);
1178 trace2_region_leave("midx", "find_commits_for_midx_bitmap",
1184 static int write_midx_bitmap(const char *midx_name
,
1185 const unsigned char *midx_hash
,
1186 struct packing_data
*pdata
,
1187 struct commit
**commits
,
1188 uint32_t commits_nr
,
1189 uint32_t *pack_order
,
1193 uint16_t options
= 0;
1194 struct pack_idx_entry
**index
;
1195 char *bitmap_name
= xstrfmt("%s-%s.bitmap", midx_name
,
1196 hash_to_hex(midx_hash
));
1198 trace2_region_enter("midx", "write_midx_bitmap", the_repository
);
1200 if (flags
& MIDX_WRITE_BITMAP_HASH_CACHE
)
1201 options
|= BITMAP_OPT_HASH_CACHE
;
1203 if (flags
& MIDX_WRITE_BITMAP_LOOKUP_TABLE
)
1204 options
|= BITMAP_OPT_LOOKUP_TABLE
;
1207 * Build the MIDX-order index based on pdata.objects (which is already
1208 * in MIDX order; c.f., 'midx_pack_order_cmp()' for the definition of
1211 ALLOC_ARRAY(index
, pdata
->nr_objects
);
1212 for (i
= 0; i
< pdata
->nr_objects
; i
++)
1213 index
[i
] = &pdata
->objects
[i
].idx
;
1215 bitmap_writer_show_progress(flags
& MIDX_PROGRESS
);
1216 bitmap_writer_build_type_index(pdata
, index
, pdata
->nr_objects
);
1219 * bitmap_writer_finish expects objects in lex order, but pack_order
1220 * gives us exactly that. use it directly instead of re-sorting the
1223 * This changes the order of objects in 'index' between
1224 * bitmap_writer_build_type_index and bitmap_writer_finish.
1226 * The same re-ordering takes place in the single-pack bitmap code via
1227 * write_idx_file(), which is called by finish_tmp_packfile(), which
1228 * happens between bitmap_writer_build_type_index() and
1229 * bitmap_writer_finish().
1231 for (i
= 0; i
< pdata
->nr_objects
; i
++)
1232 index
[pack_order
[i
]] = &pdata
->objects
[i
].idx
;
1234 bitmap_writer_select_commits(commits
, commits_nr
, -1);
1235 ret
= bitmap_writer_build(pdata
);
1239 bitmap_writer_set_checksum(midx_hash
);
1240 bitmap_writer_finish(index
, pdata
->nr_objects
, bitmap_name
, options
);
1246 trace2_region_leave("midx", "write_midx_bitmap", the_repository
);
1251 static struct multi_pack_index
*lookup_multi_pack_index(struct repository
*r
,
1252 const char *object_dir
)
1254 struct multi_pack_index
*result
= NULL
;
1255 struct multi_pack_index
*cur
;
1256 char *obj_dir_real
= real_pathdup(object_dir
, 1);
1257 struct strbuf cur_path_real
= STRBUF_INIT
;
1259 /* Ensure the given object_dir is local, or a known alternate. */
1260 find_odb(r
, obj_dir_real
);
1262 for (cur
= get_multi_pack_index(r
); cur
; cur
= cur
->next
) {
1263 strbuf_realpath(&cur_path_real
, cur
->object_dir
, 1);
1264 if (!strcmp(obj_dir_real
, cur_path_real
.buf
)) {
1272 strbuf_release(&cur_path_real
);
1276 static int write_midx_internal(const char *object_dir
,
1277 struct string_list
*packs_to_include
,
1278 struct string_list
*packs_to_drop
,
1279 const char *preferred_pack_name
,
1280 const char *refs_snapshot
,
1283 struct strbuf midx_name
= STRBUF_INIT
;
1284 unsigned char midx_hash
[GIT_MAX_RAWSZ
];
1286 struct hashfile
*f
= NULL
;
1287 struct lock_file lk
;
1288 struct write_midx_context ctx
= { 0 };
1289 int pack_name_concat_len
= 0;
1290 int dropped_packs
= 0;
1292 struct chunkfile
*cf
;
1294 trace2_region_enter("midx", "write_midx_internal", the_repository
);
1296 get_midx_filename(&midx_name
, object_dir
);
1297 if (safe_create_leading_directories(midx_name
.buf
))
1298 die_errno(_("unable to create leading directories of %s"),
1301 if (!packs_to_include
) {
1303 * Only reference an existing MIDX when not filtering which
1304 * packs to include, since all packs and objects are copied
1305 * blindly from an existing MIDX if one is present.
1307 ctx
.m
= lookup_multi_pack_index(the_repository
, object_dir
);
1310 if (ctx
.m
&& !midx_checksum_valid(ctx
.m
)) {
1311 warning(_("ignoring existing multi-pack-index; checksum mismatch"));
1316 ctx
.alloc
= ctx
.m
? ctx
.m
->num_packs
: 16;
1318 ALLOC_ARRAY(ctx
.info
, ctx
.alloc
);
1321 for (i
= 0; i
< ctx
.m
->num_packs
; i
++) {
1322 ALLOC_GROW(ctx
.info
, ctx
.nr
+ 1, ctx
.alloc
);
1324 ctx
.info
[ctx
.nr
].orig_pack_int_id
= i
;
1325 ctx
.info
[ctx
.nr
].pack_name
= xstrdup(ctx
.m
->pack_names
[i
]);
1326 ctx
.info
[ctx
.nr
].p
= ctx
.m
->packs
[i
];
1327 ctx
.info
[ctx
.nr
].expired
= 0;
1329 if (flags
& MIDX_WRITE_REV_INDEX
) {
1331 * If generating a reverse index, need to have
1332 * packed_git's loaded to compare their
1333 * mtimes and object count.
1335 if (prepare_midx_pack(the_repository
, ctx
.m
, i
)) {
1336 error(_("could not load pack"));
1341 if (open_pack_index(ctx
.m
->packs
[i
]))
1342 die(_("could not open index for %s"),
1343 ctx
.m
->packs
[i
]->pack_name
);
1344 ctx
.info
[ctx
.nr
].p
= ctx
.m
->packs
[i
];
1351 ctx
.pack_paths_checked
= 0;
1352 if (flags
& MIDX_PROGRESS
)
1353 ctx
.progress
= start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
1355 ctx
.progress
= NULL
;
1357 ctx
.to_include
= packs_to_include
;
1359 for_each_file_in_pack_dir(object_dir
, add_pack_to_midx
, &ctx
);
1360 stop_progress(&ctx
.progress
);
1362 if ((ctx
.m
&& ctx
.nr
== ctx
.m
->num_packs
) &&
1363 !(packs_to_include
|| packs_to_drop
)) {
1364 struct bitmap_index
*bitmap_git
;
1366 int want_bitmap
= flags
& MIDX_WRITE_BITMAP
;
1368 bitmap_git
= prepare_midx_bitmap_git(ctx
.m
);
1369 bitmap_exists
= bitmap_git
&& bitmap_is_midx(bitmap_git
);
1370 free_bitmap_index(bitmap_git
);
1372 if (bitmap_exists
|| !want_bitmap
) {
1374 * The correct MIDX already exists, and so does a
1375 * corresponding bitmap (or one wasn't requested).
1378 clear_midx_files_ext(object_dir
, ".bitmap",
1384 if (preferred_pack_name
) {
1385 ctx
.preferred_pack_idx
= -1;
1387 for (i
= 0; i
< ctx
.nr
; i
++) {
1388 if (!cmp_idx_or_pack_name(preferred_pack_name
,
1389 ctx
.info
[i
].pack_name
)) {
1390 ctx
.preferred_pack_idx
= i
;
1395 if (ctx
.preferred_pack_idx
== -1)
1396 warning(_("unknown preferred pack: '%s'"),
1397 preferred_pack_name
);
1398 } else if (ctx
.nr
&&
1399 (flags
& (MIDX_WRITE_REV_INDEX
| MIDX_WRITE_BITMAP
))) {
1400 struct packed_git
*oldest
= ctx
.info
[ctx
.preferred_pack_idx
].p
;
1401 ctx
.preferred_pack_idx
= 0;
1403 if (packs_to_drop
&& packs_to_drop
->nr
)
1404 BUG("cannot write a MIDX bitmap during expiration");
1407 * set a preferred pack when writing a bitmap to ensure that
1408 * the pack from which the first object is selected in pseudo
1409 * pack-order has all of its objects selected from that pack
1410 * (and not another pack containing a duplicate)
1412 for (i
= 1; i
< ctx
.nr
; i
++) {
1413 struct packed_git
*p
= ctx
.info
[i
].p
;
1415 if (!oldest
->num_objects
|| p
->mtime
< oldest
->mtime
) {
1417 ctx
.preferred_pack_idx
= i
;
1421 if (!oldest
->num_objects
) {
1423 * If all packs are empty; unset the preferred index.
1424 * This is acceptable since there will be no duplicate
1425 * objects to resolve, so the preferred value doesn't
1428 ctx
.preferred_pack_idx
= -1;
1432 * otherwise don't mark any pack as preferred to avoid
1433 * interfering with expiration logic below
1435 ctx
.preferred_pack_idx
= -1;
1438 if (ctx
.preferred_pack_idx
> -1) {
1439 struct packed_git
*preferred
= ctx
.info
[ctx
.preferred_pack_idx
].p
;
1440 if (!preferred
->num_objects
) {
1441 error(_("cannot select preferred pack %s with no objects"),
1442 preferred
->pack_name
);
1448 ctx
.entries
= get_sorted_entries(ctx
.m
, ctx
.info
, ctx
.nr
, &ctx
.entries_nr
,
1449 ctx
.preferred_pack_idx
);
1451 ctx
.large_offsets_needed
= 0;
1452 for (i
= 0; i
< ctx
.entries_nr
; i
++) {
1453 if (ctx
.entries
[i
].offset
> 0x7fffffff)
1454 ctx
.num_large_offsets
++;
1455 if (ctx
.entries
[i
].offset
> 0xffffffff)
1456 ctx
.large_offsets_needed
= 1;
1459 QSORT(ctx
.info
, ctx
.nr
, pack_info_compare
);
1461 if (packs_to_drop
&& packs_to_drop
->nr
) {
1463 int missing_drops
= 0;
1465 for (i
= 0; i
< ctx
.nr
&& drop_index
< packs_to_drop
->nr
; i
++) {
1466 int cmp
= strcmp(ctx
.info
[i
].pack_name
,
1467 packs_to_drop
->items
[drop_index
].string
);
1471 ctx
.info
[i
].expired
= 1;
1472 } else if (cmp
> 0) {
1473 error(_("did not see pack-file %s to drop"),
1474 packs_to_drop
->items
[drop_index
].string
);
1479 ctx
.info
[i
].expired
= 0;
1483 if (missing_drops
) {
1490 * pack_perm stores a permutation between pack-int-ids from the
1491 * previous multi-pack-index to the new one we are writing:
1493 * pack_perm[old_id] = new_id
1495 ALLOC_ARRAY(ctx
.pack_perm
, ctx
.nr
);
1496 for (i
= 0; i
< ctx
.nr
; i
++) {
1497 if (ctx
.info
[i
].expired
) {
1499 ctx
.pack_perm
[ctx
.info
[i
].orig_pack_int_id
] = PACK_EXPIRED
;
1501 ctx
.pack_perm
[ctx
.info
[i
].orig_pack_int_id
] = i
- dropped_packs
;
1505 for (i
= 0; i
< ctx
.nr
; i
++) {
1506 if (!ctx
.info
[i
].expired
)
1507 pack_name_concat_len
+= strlen(ctx
.info
[i
].pack_name
) + 1;
1510 /* Check that the preferred pack wasn't expired (if given). */
1511 if (preferred_pack_name
) {
1512 struct pack_info
*preferred
= bsearch(preferred_pack_name
,
1515 idx_or_pack_name_cmp
);
1517 uint32_t perm
= ctx
.pack_perm
[preferred
->orig_pack_int_id
];
1518 if (perm
== PACK_EXPIRED
)
1519 warning(_("preferred pack '%s' is expired"),
1520 preferred_pack_name
);
1524 if (pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
)
1525 pack_name_concat_len
+= MIDX_CHUNK_ALIGNMENT
-
1526 (pack_name_concat_len
% MIDX_CHUNK_ALIGNMENT
);
1528 hold_lock_file_for_update(&lk
, midx_name
.buf
, LOCK_DIE_ON_ERROR
);
1529 f
= hashfd(get_lock_file_fd(&lk
), get_lock_file_path(&lk
));
1531 if (ctx
.nr
- dropped_packs
== 0) {
1532 error(_("no pack files to index."));
1537 if (!ctx
.entries_nr
) {
1538 if (flags
& MIDX_WRITE_BITMAP
)
1539 warning(_("refusing to write multi-pack .bitmap without any objects"));
1540 flags
&= ~(MIDX_WRITE_REV_INDEX
| MIDX_WRITE_BITMAP
);
1543 cf
= init_chunkfile(f
);
1545 add_chunk(cf
, MIDX_CHUNKID_PACKNAMES
, pack_name_concat_len
,
1546 write_midx_pack_names
);
1547 add_chunk(cf
, MIDX_CHUNKID_OIDFANOUT
, MIDX_CHUNK_FANOUT_SIZE
,
1548 write_midx_oid_fanout
);
1549 add_chunk(cf
, MIDX_CHUNKID_OIDLOOKUP
,
1550 st_mult(ctx
.entries_nr
, the_hash_algo
->rawsz
),
1551 write_midx_oid_lookup
);
1552 add_chunk(cf
, MIDX_CHUNKID_OBJECTOFFSETS
,
1553 st_mult(ctx
.entries_nr
, MIDX_CHUNK_OFFSET_WIDTH
),
1554 write_midx_object_offsets
);
1556 if (ctx
.large_offsets_needed
)
1557 add_chunk(cf
, MIDX_CHUNKID_LARGEOFFSETS
,
1558 st_mult(ctx
.num_large_offsets
,
1559 MIDX_CHUNK_LARGE_OFFSET_WIDTH
),
1560 write_midx_large_offsets
);
1562 if (flags
& (MIDX_WRITE_REV_INDEX
| MIDX_WRITE_BITMAP
)) {
1563 ctx
.pack_order
= midx_pack_order(&ctx
);
1564 add_chunk(cf
, MIDX_CHUNKID_REVINDEX
,
1565 st_mult(ctx
.entries_nr
, sizeof(uint32_t)),
1566 write_midx_revindex
);
1569 write_midx_header(f
, get_num_chunks(cf
), ctx
.nr
- dropped_packs
);
1570 write_chunkfile(cf
, &ctx
);
1572 finalize_hashfile(f
, midx_hash
, FSYNC_COMPONENT_PACK_METADATA
,
1573 CSUM_FSYNC
| CSUM_HASH_IN_STREAM
);
1576 if (flags
& MIDX_WRITE_REV_INDEX
&&
1577 git_env_bool("GIT_TEST_MIDX_WRITE_REV", 0))
1578 write_midx_reverse_index(midx_name
.buf
, midx_hash
, &ctx
);
1580 if (flags
& MIDX_WRITE_BITMAP
) {
1581 struct packing_data pdata
;
1582 struct commit
**commits
;
1583 uint32_t commits_nr
;
1585 if (!ctx
.entries_nr
)
1586 BUG("cannot write a bitmap without any objects");
1588 prepare_midx_packing_data(&pdata
, &ctx
);
1590 commits
= find_commits_for_midx_bitmap(&commits_nr
, refs_snapshot
, &ctx
);
1593 * The previous steps translated the information from
1594 * 'entries' into information suitable for constructing
1595 * bitmaps. We no longer need that array, so clear it to
1596 * reduce memory pressure.
1598 FREE_AND_NULL(ctx
.entries
);
1601 if (write_midx_bitmap(midx_name
.buf
, midx_hash
, &pdata
,
1602 commits
, commits_nr
, ctx
.pack_order
,
1604 error(_("could not write multi-pack bitmap"));
1610 * NOTE: Do not use ctx.entries beyond this point, since it might
1611 * have been freed in the previous if block.
1615 close_object_store(the_repository
->objects
);
1617 if (commit_lock_file(&lk
) < 0)
1618 die_errno(_("could not write multi-pack-index"));
1620 clear_midx_files_ext(object_dir
, ".bitmap", midx_hash
);
1621 clear_midx_files_ext(object_dir
, ".rev", midx_hash
);
1624 for (i
= 0; i
< ctx
.nr
; i
++) {
1625 if (ctx
.info
[i
].p
) {
1626 close_pack(ctx
.info
[i
].p
);
1627 free(ctx
.info
[i
].p
);
1629 free(ctx
.info
[i
].pack_name
);
1634 free(ctx
.pack_perm
);
1635 free(ctx
.pack_order
);
1636 strbuf_release(&midx_name
);
1638 trace2_region_leave("midx", "write_midx_internal", the_repository
);
1643 int write_midx_file(const char *object_dir
,
1644 const char *preferred_pack_name
,
1645 const char *refs_snapshot
,
1648 return write_midx_internal(object_dir
, NULL
, NULL
, preferred_pack_name
,
1649 refs_snapshot
, flags
);
1652 int write_midx_file_only(const char *object_dir
,
1653 struct string_list
*packs_to_include
,
1654 const char *preferred_pack_name
,
1655 const char *refs_snapshot
,
1658 return write_midx_internal(object_dir
, packs_to_include
, NULL
,
1659 preferred_pack_name
, refs_snapshot
, flags
);
1662 struct clear_midx_data
{
1667 static void clear_midx_file_ext(const char *full_path
, size_t full_path_len UNUSED
,
1668 const char *file_name
, void *_data
)
1670 struct clear_midx_data
*data
= _data
;
1672 if (!(starts_with(file_name
, "multi-pack-index-") &&
1673 ends_with(file_name
, data
->ext
)))
1675 if (data
->keep
&& !strcmp(data
->keep
, file_name
))
1678 if (unlink(full_path
))
1679 die_errno(_("failed to remove %s"), full_path
);
1682 static void clear_midx_files_ext(const char *object_dir
, const char *ext
,
1683 unsigned char *keep_hash
)
1685 struct clear_midx_data data
;
1686 memset(&data
, 0, sizeof(struct clear_midx_data
));
1689 data
.keep
= xstrfmt("multi-pack-index-%s%s",
1690 hash_to_hex(keep_hash
), ext
);
1693 for_each_file_in_pack_dir(object_dir
,
1694 clear_midx_file_ext
,
1700 void clear_midx_file(struct repository
*r
)
1702 struct strbuf midx
= STRBUF_INIT
;
1704 get_midx_filename(&midx
, r
->objects
->odb
->path
);
1706 if (r
->objects
&& r
->objects
->multi_pack_index
) {
1707 close_midx(r
->objects
->multi_pack_index
);
1708 r
->objects
->multi_pack_index
= NULL
;
1711 if (remove_path(midx
.buf
))
1712 die(_("failed to clear multi-pack-index at %s"), midx
.buf
);
1714 clear_midx_files_ext(r
->objects
->odb
->path
, ".bitmap", NULL
);
1715 clear_midx_files_ext(r
->objects
->odb
->path
, ".rev", NULL
);
1717 strbuf_release(&midx
);
1720 static int verify_midx_error
;
1722 __attribute__((format (printf
, 1, 2)))
1723 static void midx_report(const char *fmt
, ...)
1726 verify_midx_error
= 1;
1728 vfprintf(stderr
, fmt
, ap
);
1729 fprintf(stderr
, "\n");
1733 struct pair_pos_vs_id
1736 uint32_t pack_int_id
;
1739 static int compare_pair_pos_vs_id(const void *_a
, const void *_b
)
1741 struct pair_pos_vs_id
*a
= (struct pair_pos_vs_id
*)_a
;
1742 struct pair_pos_vs_id
*b
= (struct pair_pos_vs_id
*)_b
;
1744 return b
->pack_int_id
- a
->pack_int_id
;
1748 * Limit calls to display_progress() for performance reasons.
1749 * The interval here was arbitrarily chosen.
1751 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1752 #define midx_display_sparse_progress(progress, n) \
1754 uint64_t _n = (n); \
1755 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1756 display_progress(progress, _n); \
1759 int verify_midx_file(struct repository
*r
, const char *object_dir
, unsigned flags
)
1761 struct pair_pos_vs_id
*pairs
= NULL
;
1763 struct progress
*progress
= NULL
;
1764 struct multi_pack_index
*m
= load_multi_pack_index(object_dir
, 1);
1765 verify_midx_error
= 0;
1770 struct strbuf filename
= STRBUF_INIT
;
1772 get_midx_filename(&filename
, object_dir
);
1774 if (!stat(filename
.buf
, &sb
)) {
1775 error(_("multi-pack-index file exists, but failed to parse"));
1778 strbuf_release(&filename
);
1782 if (!midx_checksum_valid(m
))
1783 midx_report(_("incorrect checksum"));
1785 if (flags
& MIDX_PROGRESS
)
1786 progress
= start_delayed_progress(_("Looking for referenced packfiles"),
1788 for (i
= 0; i
< m
->num_packs
; i
++) {
1789 if (prepare_midx_pack(r
, m
, i
))
1790 midx_report("failed to load pack in position %d", i
);
1792 display_progress(progress
, i
+ 1);
1794 stop_progress(&progress
);
1796 if (m
->num_objects
== 0) {
1797 midx_report(_("the midx contains no oid"));
1799 * Remaining tests assume that we have objects, so we can
1805 if (flags
& MIDX_PROGRESS
)
1806 progress
= start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1807 m
->num_objects
- 1);
1808 for (i
= 0; i
< m
->num_objects
- 1; i
++) {
1809 struct object_id oid1
, oid2
;
1811 nth_midxed_object_oid(&oid1
, m
, i
);
1812 nth_midxed_object_oid(&oid2
, m
, i
+ 1);
1814 if (oidcmp(&oid1
, &oid2
) >= 0)
1815 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1816 i
, oid_to_hex(&oid1
), oid_to_hex(&oid2
), i
+ 1);
1818 midx_display_sparse_progress(progress
, i
+ 1);
1820 stop_progress(&progress
);
1823 * Create an array mapping each object to its packfile id. Sort it
1824 * to group the objects by packfile. Use this permutation to visit
1825 * each of the objects and only require 1 packfile to be open at a
1828 ALLOC_ARRAY(pairs
, m
->num_objects
);
1829 for (i
= 0; i
< m
->num_objects
; i
++) {
1831 pairs
[i
].pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1834 if (flags
& MIDX_PROGRESS
)
1835 progress
= start_sparse_progress(_("Sorting objects by packfile"),
1837 display_progress(progress
, 0); /* TODO: Measure QSORT() progress */
1838 QSORT(pairs
, m
->num_objects
, compare_pair_pos_vs_id
);
1839 stop_progress(&progress
);
1841 if (flags
& MIDX_PROGRESS
)
1842 progress
= start_sparse_progress(_("Verifying object offsets"), m
->num_objects
);
1843 for (i
= 0; i
< m
->num_objects
; i
++) {
1844 struct object_id oid
;
1845 struct pack_entry e
;
1846 off_t m_offset
, p_offset
;
1848 if (i
> 0 && pairs
[i
-1].pack_int_id
!= pairs
[i
].pack_int_id
&&
1849 m
->packs
[pairs
[i
-1].pack_int_id
])
1851 close_pack_fd(m
->packs
[pairs
[i
-1].pack_int_id
]);
1852 close_pack_index(m
->packs
[pairs
[i
-1].pack_int_id
]);
1855 nth_midxed_object_oid(&oid
, m
, pairs
[i
].pos
);
1857 if (!fill_midx_entry(r
, &oid
, &e
, m
)) {
1858 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1859 pairs
[i
].pos
, oid_to_hex(&oid
));
1863 if (open_pack_index(e
.p
)) {
1864 midx_report(_("failed to load pack-index for packfile %s"),
1869 m_offset
= e
.offset
;
1870 p_offset
= find_pack_entry_one(oid
.hash
, e
.p
);
1872 if (m_offset
!= p_offset
)
1873 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64
" != %"PRIx64
),
1874 pairs
[i
].pos
, oid_to_hex(&oid
), m_offset
, p_offset
);
1876 midx_display_sparse_progress(progress
, i
+ 1);
1878 stop_progress(&progress
);
1884 return verify_midx_error
;
1887 int expire_midx_packs(struct repository
*r
, const char *object_dir
, unsigned flags
)
1889 uint32_t i
, *count
, result
= 0;
1890 struct string_list packs_to_drop
= STRING_LIST_INIT_DUP
;
1891 struct multi_pack_index
*m
= lookup_multi_pack_index(r
, object_dir
);
1892 struct progress
*progress
= NULL
;
1897 CALLOC_ARRAY(count
, m
->num_packs
);
1899 if (flags
& MIDX_PROGRESS
)
1900 progress
= start_delayed_progress(_("Counting referenced objects"),
1902 for (i
= 0; i
< m
->num_objects
; i
++) {
1903 int pack_int_id
= nth_midxed_pack_int_id(m
, i
);
1904 count
[pack_int_id
]++;
1905 display_progress(progress
, i
+ 1);
1907 stop_progress(&progress
);
1909 if (flags
& MIDX_PROGRESS
)
1910 progress
= start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1912 for (i
= 0; i
< m
->num_packs
; i
++) {
1914 display_progress(progress
, i
+ 1);
1919 if (prepare_midx_pack(r
, m
, i
))
1922 if (m
->packs
[i
]->pack_keep
|| m
->packs
[i
]->is_cruft
)
1925 pack_name
= xstrdup(m
->packs
[i
]->pack_name
);
1926 close_pack(m
->packs
[i
]);
1928 string_list_insert(&packs_to_drop
, m
->pack_names
[i
]);
1929 unlink_pack_path(pack_name
, 0);
1932 stop_progress(&progress
);
1936 if (packs_to_drop
.nr
)
1937 result
= write_midx_internal(object_dir
, NULL
, &packs_to_drop
, NULL
, NULL
, flags
);
1939 string_list_clear(&packs_to_drop
, 0);
1944 struct repack_info
{
1946 uint32_t referenced_objects
;
1947 uint32_t pack_int_id
;
1950 static int compare_by_mtime(const void *a_
, const void *b_
)
1952 const struct repack_info
*a
, *b
;
1954 a
= (const struct repack_info
*)a_
;
1955 b
= (const struct repack_info
*)b_
;
1957 if (a
->mtime
< b
->mtime
)
1959 if (a
->mtime
> b
->mtime
)
1964 static int fill_included_packs_all(struct repository
*r
,
1965 struct multi_pack_index
*m
,
1966 unsigned char *include_pack
)
1968 uint32_t i
, count
= 0;
1969 int pack_kept_objects
= 0;
1971 repo_config_get_bool(r
, "repack.packkeptobjects", &pack_kept_objects
);
1973 for (i
= 0; i
< m
->num_packs
; i
++) {
1974 if (prepare_midx_pack(r
, m
, i
))
1976 if (!pack_kept_objects
&& m
->packs
[i
]->pack_keep
)
1978 if (m
->packs
[i
]->is_cruft
)
1981 include_pack
[i
] = 1;
1988 static int fill_included_packs_batch(struct repository
*r
,
1989 struct multi_pack_index
*m
,
1990 unsigned char *include_pack
,
1993 uint32_t i
, packs_to_repack
;
1995 struct repack_info
*pack_info
;
1996 int pack_kept_objects
= 0;
1998 CALLOC_ARRAY(pack_info
, m
->num_packs
);
2000 repo_config_get_bool(r
, "repack.packkeptobjects", &pack_kept_objects
);
2002 for (i
= 0; i
< m
->num_packs
; i
++) {
2003 pack_info
[i
].pack_int_id
= i
;
2005 if (prepare_midx_pack(r
, m
, i
))
2008 pack_info
[i
].mtime
= m
->packs
[i
]->mtime
;
2011 for (i
= 0; i
< m
->num_objects
; i
++) {
2012 uint32_t pack_int_id
= nth_midxed_pack_int_id(m
, i
);
2013 pack_info
[pack_int_id
].referenced_objects
++;
2016 QSORT(pack_info
, m
->num_packs
, compare_by_mtime
);
2019 packs_to_repack
= 0;
2020 for (i
= 0; total_size
< batch_size
&& i
< m
->num_packs
; i
++) {
2021 int pack_int_id
= pack_info
[i
].pack_int_id
;
2022 struct packed_git
*p
= m
->packs
[pack_int_id
];
2023 size_t expected_size
;
2027 if (!pack_kept_objects
&& p
->pack_keep
)
2031 if (open_pack_index(p
) || !p
->num_objects
)
2034 expected_size
= st_mult(p
->pack_size
,
2035 pack_info
[i
].referenced_objects
);
2036 expected_size
/= p
->num_objects
;
2038 if (expected_size
>= batch_size
)
2042 total_size
+= expected_size
;
2043 include_pack
[pack_int_id
] = 1;
2048 if (packs_to_repack
< 2)
2054 int midx_repack(struct repository
*r
, const char *object_dir
, size_t batch_size
, unsigned flags
)
2058 unsigned char *include_pack
;
2059 struct child_process cmd
= CHILD_PROCESS_INIT
;
2061 struct strbuf base_name
= STRBUF_INIT
;
2062 struct multi_pack_index
*m
= lookup_multi_pack_index(r
, object_dir
);
2065 * When updating the default for these configuration
2066 * variables in builtin/repack.c, these must be adjusted
2069 int delta_base_offset
= 1;
2070 int use_delta_islands
= 0;
2075 CALLOC_ARRAY(include_pack
, m
->num_packs
);
2078 if (fill_included_packs_batch(r
, m
, include_pack
, batch_size
))
2080 } else if (fill_included_packs_all(r
, m
, include_pack
))
2083 repo_config_get_bool(r
, "repack.usedeltabaseoffset", &delta_base_offset
);
2084 repo_config_get_bool(r
, "repack.usedeltaislands", &use_delta_islands
);
2086 strvec_push(&cmd
.args
, "pack-objects");
2088 strbuf_addstr(&base_name
, object_dir
);
2089 strbuf_addstr(&base_name
, "/pack/pack");
2090 strvec_push(&cmd
.args
, base_name
.buf
);
2092 if (delta_base_offset
)
2093 strvec_push(&cmd
.args
, "--delta-base-offset");
2094 if (use_delta_islands
)
2095 strvec_push(&cmd
.args
, "--delta-islands");
2097 if (flags
& MIDX_PROGRESS
)
2098 strvec_push(&cmd
.args
, "--progress");
2100 strvec_push(&cmd
.args
, "-q");
2102 strbuf_release(&base_name
);
2105 cmd
.in
= cmd
.out
= -1;
2107 if (start_command(&cmd
)) {
2108 error(_("could not start pack-objects"));
2113 cmd_in
= xfdopen(cmd
.in
, "w");
2115 for (i
= 0; i
< m
->num_objects
; i
++) {
2116 struct object_id oid
;
2117 uint32_t pack_int_id
= nth_midxed_pack_int_id(m
, i
);
2119 if (!include_pack
[pack_int_id
])
2122 nth_midxed_object_oid(&oid
, m
, i
);
2123 fprintf(cmd_in
, "%s\n", oid_to_hex(&oid
));
2127 if (finish_command(&cmd
)) {
2128 error(_("could not finish pack-objects"));
2133 result
= write_midx_internal(object_dir
, NULL
, NULL
, NULL
, NULL
, flags
);