7 #include "object-store.h"
8 #include "sha1-lookup.h"
12 #include "run-command.h"
13 #include "repository.h"
15 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
16 #define MIDX_VERSION 1
17 #define MIDX_BYTE_FILE_VERSION 4
18 #define MIDX_BYTE_HASH_VERSION 5
19 #define MIDX_BYTE_NUM_CHUNKS 6
20 #define MIDX_BYTE_NUM_PACKS 8
21 #define MIDX_HEADER_SIZE 12
22 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
24 #define MIDX_MAX_CHUNKS 5
25 #define MIDX_CHUNK_ALIGNMENT 4
26 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
27 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
28 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
29 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
30 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
31 #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
32 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
33 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
34 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
35 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
37 #define PACK_EXPIRED UINT_MAX
39 static uint8_t oid_version(void)
41 switch (hash_algo_by_ptr(the_hash_algo)) {
47 die(_("invalid hash version"));
51 static char *get_midx_filename(const char *object_dir)
53 return xstrfmt("%s/pack/multi-pack-index", object_dir);
56 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
58 struct multi_pack_index *m = NULL;
62 void *midx_map = NULL;
63 uint32_t hash_version;
64 char *midx_name = get_midx_filename(object_dir);
66 const char *cur_pack_name;
68 fd = git_open(midx_name);
73 error_errno(_("failed to read %s"), midx_name);
77 midx_size = xsize_t(st.st_size);
79 if (midx_size < MIDX_MIN_SIZE) {
80 error(_("multi-pack-index file %s is too small"), midx_name);
84 FREE_AND_NULL(midx_name);
86 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
89 FLEX_ALLOC_STR(m, object_dir, object_dir);
91 m->data_len = midx_size;
94 m->signature = get_be32(m->data);
95 if (m->signature != MIDX_SIGNATURE)
96 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
97 m->signature, MIDX_SIGNATURE);
99 m->version = m->data[MIDX_BYTE_FILE_VERSION];
100 if (m->version != MIDX_VERSION)
101 die(_("multi-pack-index version %d not recognized"),
104 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
105 if (hash_version != oid_version()) {
106 error(_("multi-pack-index hash version %u does not match version %u"),
107 hash_version, oid_version());
110 m->hash_len = the_hash_algo->rawsz;
112 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
114 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
116 for (i = 0; i < m->num_chunks; i++) {
117 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
118 MIDX_CHUNKLOOKUP_WIDTH * i);
119 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
120 MIDX_CHUNKLOOKUP_WIDTH * i);
122 if (chunk_offset >= m->data_len)
123 die(_("invalid chunk offset (too large)"));
126 case MIDX_CHUNKID_PACKNAMES:
127 m->chunk_pack_names = m->data + chunk_offset;
130 case MIDX_CHUNKID_OIDFANOUT:
131 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
134 case MIDX_CHUNKID_OIDLOOKUP:
135 m->chunk_oid_lookup = m->data + chunk_offset;
138 case MIDX_CHUNKID_OBJECTOFFSETS:
139 m->chunk_object_offsets = m->data + chunk_offset;
142 case MIDX_CHUNKID_LARGEOFFSETS:
143 m->chunk_large_offsets = m->data + chunk_offset;
147 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
152 * Do nothing on unrecognized chunks, allowing future
153 * extensions to add optional chunks.
159 if (!m->chunk_pack_names)
160 die(_("multi-pack-index missing required pack-name chunk"));
161 if (!m->chunk_oid_fanout)
162 die(_("multi-pack-index missing required OID fanout chunk"));
163 if (!m->chunk_oid_lookup)
164 die(_("multi-pack-index missing required OID lookup chunk"));
165 if (!m->chunk_object_offsets)
166 die(_("multi-pack-index missing required object offsets chunk"));
168 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
170 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
171 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
173 cur_pack_name = (const char *)m->chunk_pack_names;
174 for (i = 0; i < m->num_packs; i++) {
175 m->pack_names[i] = cur_pack_name;
177 cur_pack_name += strlen(cur_pack_name) + 1;
179 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
180 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
181 m->pack_names[i - 1],
185 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
186 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
194 munmap(midx_map, midx_size);
200 void close_midx(struct multi_pack_index *m)
207 munmap((unsigned char *)m->data, m->data_len);
209 for (i = 0; i < m->num_packs; i++) {
211 m->packs[i]->multi_pack_index = 0;
213 FREE_AND_NULL(m->packs);
214 FREE_AND_NULL(m->pack_names);
217 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
219 struct strbuf pack_name = STRBUF_INIT;
220 struct packed_git *p;
222 if (pack_int_id >= m->num_packs)
223 die(_("bad pack-int-id: %u (%u total packs)"),
224 pack_int_id, m->num_packs);
226 if (m->packs[pack_int_id])
229 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
230 m->pack_names[pack_int_id]);
232 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
233 strbuf_release(&pack_name);
238 p->multi_pack_index = 1;
239 m->packs[pack_int_id] = p;
240 install_packed_git(r, p);
241 list_add_tail(&p->mru, &r->objects->packed_git_mru);
246 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
248 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
249 the_hash_algo->rawsz, result);
252 struct object_id *nth_midxed_object_oid(struct object_id *oid,
253 struct multi_pack_index *m,
256 if (n >= m->num_objects)
259 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
263 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
265 const unsigned char *offset_data;
268 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
269 offset32 = get_be32(offset_data + sizeof(uint32_t));
271 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
272 if (sizeof(off_t) < sizeof(uint64_t))
273 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
275 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
276 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
282 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
284 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
287 static int nth_midxed_pack_entry(struct repository *r,
288 struct multi_pack_index *m,
289 struct pack_entry *e,
292 uint32_t pack_int_id;
293 struct packed_git *p;
295 if (pos >= m->num_objects)
298 pack_int_id = nth_midxed_pack_int_id(m, pos);
300 if (prepare_midx_pack(r, m, pack_int_id))
302 p = m->packs[pack_int_id];
305 * We are about to tell the caller where they can locate the
306 * requested object. We better make sure the packfile is
307 * still here and can be accessed before supplying that
308 * answer, as it may have been deleted since the MIDX was
311 if (!is_pack_valid(p))
314 if (p->num_bad_objects) {
316 struct object_id oid;
317 nth_midxed_object_oid(&oid, m, pos);
318 for (i = 0; i < p->num_bad_objects; i++)
320 p->bad_object_sha1 + the_hash_algo->rawsz * i))
324 e->offset = nth_midxed_offset(m, pos);
330 int fill_midx_entry(struct repository * r,
331 const struct object_id *oid,
332 struct pack_entry *e,
333 struct multi_pack_index *m)
337 if (!bsearch_midx(oid, m, &pos))
340 return nth_midxed_pack_entry(r, m, e, pos);
343 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
344 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
345 const char *idx_name)
347 /* Skip past any initial matching prefix. */
348 while (*idx_name && *idx_name == *idx_or_pack_name) {
354 * If we didn't match completely, we may have matched "pack-1234." and
355 * be left with "idx" and "pack" respectively, which is also OK. We do
356 * not have to check for "idx" and "idx", because that would have been
357 * a complete match (and in that case these strcmps will be false, but
358 * we'll correctly return 0 from the final strcmp() below.
360 * Technically this matches "fooidx" and "foopack", but we'd never have
361 * such names in the first place.
363 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
367 * This not only checks for a complete match, but also orders based on
368 * the first non-identical character, which means our ordering will
369 * match a raw strcmp(). That makes it OK to use this to binary search
370 * a naively-sorted list.
372 return strcmp(idx_or_pack_name, idx_name);
375 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
377 uint32_t first = 0, last = m->num_packs;
379 while (first < last) {
380 uint32_t mid = first + (last - first) / 2;
384 current = m->pack_names[mid];
385 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
398 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
400 struct multi_pack_index *m;
401 struct multi_pack_index *m_search;
403 prepare_repo_settings(r);
404 if (!r->settings.core_multi_pack_index)
407 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
408 if (!strcmp(object_dir, m_search->object_dir))
411 m = load_multi_pack_index(object_dir, local);
414 struct multi_pack_index *mp = r->objects->multi_pack_index;
419 r->objects->multi_pack_index = m;
426 static size_t write_midx_header(struct hashfile *f,
427 unsigned char num_chunks,
430 hashwrite_be32(f, MIDX_SIGNATURE);
431 hashwrite_u8(f, MIDX_VERSION);
432 hashwrite_u8(f, oid_version());
433 hashwrite_u8(f, num_chunks);
434 hashwrite_u8(f, 0); /* unused */
435 hashwrite_be32(f, num_packs);
437 return MIDX_HEADER_SIZE;
441 uint32_t orig_pack_int_id;
443 struct packed_git *p;
444 unsigned expired : 1;
447 static int pack_info_compare(const void *_a, const void *_b)
449 struct pack_info *a = (struct pack_info *)_a;
450 struct pack_info *b = (struct pack_info *)_b;
451 return strcmp(a->pack_name, b->pack_name);
454 struct write_midx_context {
455 struct pack_info *info;
458 struct multi_pack_index *m;
459 struct progress *progress;
460 unsigned pack_paths_checked;
462 struct pack_midx_entry *entries;
466 unsigned large_offsets_needed:1;
469 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
470 const char *file_name, void *data)
472 struct write_midx_context *ctx = data;
474 if (ends_with(file_name, ".idx")) {
475 display_progress(ctx->progress, ++ctx->pack_paths_checked);
476 if (ctx->m && midx_contains_pack(ctx->m, file_name))
479 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
481 ctx->info[ctx->nr].p = add_packed_git(full_path,
485 if (!ctx->info[ctx->nr].p) {
486 warning(_("failed to add packfile '%s'"),
491 if (open_pack_index(ctx->info[ctx->nr].p)) {
492 warning(_("failed to open pack-index '%s'"),
494 close_pack(ctx->info[ctx->nr].p);
495 FREE_AND_NULL(ctx->info[ctx->nr].p);
499 ctx->info[ctx->nr].pack_name = xstrdup(file_name);
500 ctx->info[ctx->nr].orig_pack_int_id = ctx->nr;
501 ctx->info[ctx->nr].expired = 0;
506 struct pack_midx_entry {
507 struct object_id oid;
508 uint32_t pack_int_id;
513 static int midx_oid_compare(const void *_a, const void *_b)
515 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
516 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
517 int cmp = oidcmp(&a->oid, &b->oid);
522 if (a->pack_mtime > b->pack_mtime)
524 else if (a->pack_mtime < b->pack_mtime)
527 return a->pack_int_id - b->pack_int_id;
530 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
531 struct pack_midx_entry *e,
534 if (pos >= m->num_objects)
537 nth_midxed_object_oid(&e->oid, m, pos);
538 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
539 e->offset = nth_midxed_offset(m, pos);
541 /* consider objects in midx to be from "old" packs */
546 static void fill_pack_entry(uint32_t pack_int_id,
547 struct packed_git *p,
549 struct pack_midx_entry *entry)
551 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
552 die(_("failed to locate object %d in packfile"), cur_object);
554 entry->pack_int_id = pack_int_id;
555 entry->pack_mtime = p->mtime;
557 entry->offset = nth_packed_object_offset(p, cur_object);
561 * It is possible to artificially get into a state where there are many
562 * duplicate copies of objects. That can create high memory pressure if
563 * we are to create a list of all objects before de-duplication. To reduce
564 * this memory pressure without a significant performance drop, automatically
565 * group objects by the first byte of their object id. Use the IDX fanout
566 * tables to group the data, copy to a local array, then sort.
568 * Copy only the de-duplicated entries (selected by most-recent modified time
569 * of a packfile containing the object).
571 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
572 struct pack_info *info,
574 uint32_t *nr_objects)
576 uint32_t cur_fanout, cur_pack, cur_object;
577 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
578 struct pack_midx_entry *entries_by_fanout = NULL;
579 struct pack_midx_entry *deduplicated_entries = NULL;
580 uint32_t start_pack = m ? m->num_packs : 0;
582 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
583 total_objects += info[cur_pack].p->num_objects;
586 * As we de-duplicate by fanout value, we expect the fanout
587 * slices to be evenly distributed, with some noise. Hence,
588 * allocate slightly more than one 256th.
590 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
592 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
593 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
596 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
597 uint32_t nr_fanout = 0;
600 uint32_t start = 0, end;
603 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
604 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
606 for (cur_object = start; cur_object < end; cur_object++) {
607 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
608 nth_midxed_pack_midx_entry(m,
609 &entries_by_fanout[nr_fanout],
615 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
616 uint32_t start = 0, end;
619 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
620 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
622 for (cur_object = start; cur_object < end; cur_object++) {
623 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
624 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
629 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
632 * The batch is now sorted by OID and then mtime (descending).
633 * Take only the first duplicate.
635 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
636 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
637 &entries_by_fanout[cur_object].oid))
640 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
641 memcpy(&deduplicated_entries[*nr_objects],
642 &entries_by_fanout[cur_object],
643 sizeof(struct pack_midx_entry));
648 free(entries_by_fanout);
649 return deduplicated_entries;
652 static size_t write_midx_pack_names(struct hashfile *f, void *data)
654 struct write_midx_context *ctx = data;
656 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
659 for (i = 0; i < ctx->nr; i++) {
662 if (ctx->info[i].expired)
665 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
666 BUG("incorrect pack-file order: %s before %s",
667 ctx->info[i - 1].pack_name,
668 ctx->info[i].pack_name);
670 writelen = strlen(ctx->info[i].pack_name) + 1;
671 hashwrite(f, ctx->info[i].pack_name, writelen);
675 /* add padding to be aligned */
676 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
677 if (i < MIDX_CHUNK_ALIGNMENT) {
678 memset(padding, 0, sizeof(padding));
679 hashwrite(f, padding, i);
686 static size_t write_midx_oid_fanout(struct hashfile *f,
689 struct write_midx_context *ctx = data;
690 struct pack_midx_entry *list = ctx->entries;
691 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
696 * Write the first-level table (the list is sorted,
697 * but we use a 256-entry lookup to be able to avoid
698 * having to do eight extra binary search iterations).
700 for (i = 0; i < 256; i++) {
701 struct pack_midx_entry *next = list;
703 while (next < last && next->oid.hash[0] == i) {
708 hashwrite_be32(f, count);
712 return MIDX_CHUNK_FANOUT_SIZE;
715 static size_t write_midx_oid_lookup(struct hashfile *f,
718 struct write_midx_context *ctx = data;
719 unsigned char hash_len = the_hash_algo->rawsz;
720 struct pack_midx_entry *list = ctx->entries;
724 for (i = 0; i < ctx->entries_nr; i++) {
725 struct pack_midx_entry *obj = list++;
727 if (i < ctx->entries_nr - 1) {
728 struct pack_midx_entry *next = list;
729 if (oidcmp(&obj->oid, &next->oid) >= 0)
730 BUG("OIDs not in order: %s >= %s",
731 oid_to_hex(&obj->oid),
732 oid_to_hex(&next->oid));
735 hashwrite(f, obj->oid.hash, (int)hash_len);
742 static size_t write_midx_object_offsets(struct hashfile *f,
745 struct write_midx_context *ctx = data;
746 struct pack_midx_entry *list = ctx->entries;
747 uint32_t i, nr_large_offset = 0;
750 for (i = 0; i < ctx->entries_nr; i++) {
751 struct pack_midx_entry *obj = list++;
753 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
754 BUG("object %s is in an expired pack with int-id %d",
755 oid_to_hex(&obj->oid),
758 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
760 if (ctx->large_offsets_needed && obj->offset >> 31)
761 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
762 else if (!ctx->large_offsets_needed && obj->offset >> 32)
763 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
764 oid_to_hex(&obj->oid),
767 hashwrite_be32(f, (uint32_t)obj->offset);
769 written += MIDX_CHUNK_OFFSET_WIDTH;
775 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
776 struct pack_midx_entry *objects, uint32_t nr_objects)
778 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
781 while (nr_large_offset) {
782 struct pack_midx_entry *obj;
786 BUG("too many large-offset objects");
789 offset = obj->offset;
794 written += hashwrite_be64(f, offset);
802 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
803 struct string_list *packs_to_drop, unsigned flags)
805 unsigned char cur_chunk, num_chunks = 0;
808 struct hashfile *f = NULL;
810 struct write_midx_context ctx = { 0 };
811 uint64_t written = 0;
812 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
813 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
814 uint32_t num_large_offsets = 0;
815 struct progress *progress = NULL;
816 int pack_name_concat_len = 0;
817 int dropped_packs = 0;
820 midx_name = get_midx_filename(object_dir);
821 if (safe_create_leading_directories(midx_name))
822 die_errno(_("unable to create leading directories of %s"),
828 ctx.m = load_multi_pack_index(object_dir, 1);
831 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
833 ALLOC_ARRAY(ctx.info, ctx.alloc);
836 for (i = 0; i < ctx.m->num_packs; i++) {
837 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
839 ctx.info[ctx.nr].orig_pack_int_id = i;
840 ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]);
841 ctx.info[ctx.nr].p = NULL;
842 ctx.info[ctx.nr].expired = 0;
847 ctx.pack_paths_checked = 0;
848 if (flags & MIDX_PROGRESS)
849 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
853 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
854 stop_progress(&ctx.progress);
856 if (ctx.m && ctx.nr == ctx.m->num_packs && !packs_to_drop)
859 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr);
861 ctx.large_offsets_needed = 0;
862 for (i = 0; i < ctx.entries_nr; i++) {
863 if (ctx.entries[i].offset > 0x7fffffff)
865 if (ctx.entries[i].offset > 0xffffffff)
866 ctx.large_offsets_needed = 1;
869 QSORT(ctx.info, ctx.nr, pack_info_compare);
871 if (packs_to_drop && packs_to_drop->nr) {
873 int missing_drops = 0;
875 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
876 int cmp = strcmp(ctx.info[i].pack_name,
877 packs_to_drop->items[drop_index].string);
881 ctx.info[i].expired = 1;
882 } else if (cmp > 0) {
883 error(_("did not see pack-file %s to drop"),
884 packs_to_drop->items[drop_index].string);
889 ctx.info[i].expired = 0;
900 * pack_perm stores a permutation between pack-int-ids from the
901 * previous multi-pack-index to the new one we are writing:
903 * pack_perm[old_id] = new_id
905 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
906 for (i = 0; i < ctx.nr; i++) {
907 if (ctx.info[i].expired) {
909 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
911 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
915 for (i = 0; i < ctx.nr; i++) {
916 if (!ctx.info[i].expired)
917 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
920 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
921 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
922 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
924 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
925 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
926 FREE_AND_NULL(midx_name);
932 num_chunks = ctx.large_offsets_needed ? 5 : 4;
934 if (ctx.nr - dropped_packs == 0) {
935 error(_("no pack files to index."));
940 written = write_midx_header(f, num_chunks, ctx.nr - dropped_packs);
942 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
943 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
946 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
947 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + pack_name_concat_len;
950 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
951 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
954 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
955 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + ctx.entries_nr * the_hash_algo->rawsz;
958 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + ctx.entries_nr * MIDX_CHUNK_OFFSET_WIDTH;
959 if (ctx.large_offsets_needed) {
960 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
963 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
964 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
967 chunk_ids[cur_chunk] = 0;
969 for (i = 0; i <= num_chunks; i++) {
970 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
971 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
972 chunk_offsets[i - 1],
975 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
976 BUG("chunk offset %"PRIu64" is not properly aligned",
979 hashwrite_be32(f, chunk_ids[i]);
980 hashwrite_be64(f, chunk_offsets[i]);
982 written += MIDX_CHUNKLOOKUP_WIDTH;
985 if (flags & MIDX_PROGRESS)
986 progress = start_delayed_progress(_("Writing chunks to multi-pack-index"),
988 for (i = 0; i < num_chunks; i++) {
989 if (written != chunk_offsets[i])
990 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
995 switch (chunk_ids[i]) {
996 case MIDX_CHUNKID_PACKNAMES:
997 written += write_midx_pack_names(f, &ctx);
1000 case MIDX_CHUNKID_OIDFANOUT:
1001 written += write_midx_oid_fanout(f, &ctx);
1004 case MIDX_CHUNKID_OIDLOOKUP:
1005 written += write_midx_oid_lookup(f, &ctx);
1008 case MIDX_CHUNKID_OBJECTOFFSETS:
1009 written += write_midx_object_offsets(f, &ctx);
1012 case MIDX_CHUNKID_LARGEOFFSETS:
1013 written += write_midx_large_offsets(f, num_large_offsets, ctx.entries, ctx.entries_nr);
1017 BUG("trying to write unknown chunk id %"PRIx32,
1021 display_progress(progress, i + 1);
1023 stop_progress(&progress);
1025 if (written != chunk_offsets[num_chunks])
1026 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
1028 chunk_offsets[num_chunks]);
1030 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1031 commit_lock_file(&lk);
1034 for (i = 0; i < ctx.nr; i++) {
1035 if (ctx.info[i].p) {
1036 close_pack(ctx.info[i].p);
1037 free(ctx.info[i].p);
1039 free(ctx.info[i].pack_name);
1044 free(ctx.pack_perm);
1049 int write_midx_file(const char *object_dir, unsigned flags)
1051 return write_midx_internal(object_dir, NULL, NULL, flags);
1054 void clear_midx_file(struct repository *r)
1056 char *midx = get_midx_filename(r->objects->odb->path);
1058 if (r->objects && r->objects->multi_pack_index) {
1059 close_midx(r->objects->multi_pack_index);
1060 r->objects->multi_pack_index = NULL;
1063 if (remove_path(midx))
1064 die(_("failed to clear multi-pack-index at %s"), midx);
1069 static int verify_midx_error;
1071 static void midx_report(const char *fmt, ...)
1074 verify_midx_error = 1;
1076 vfprintf(stderr, fmt, ap);
1077 fprintf(stderr, "\n");
1081 struct pair_pos_vs_id
1084 uint32_t pack_int_id;
1087 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1089 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1090 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1092 return b->pack_int_id - a->pack_int_id;
1096 * Limit calls to display_progress() for performance reasons.
1097 * The interval here was arbitrarily chosen.
1099 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1100 #define midx_display_sparse_progress(progress, n) \
1102 uint64_t _n = (n); \
1103 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1104 display_progress(progress, _n); \
1107 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1109 struct pair_pos_vs_id *pairs = NULL;
1111 struct progress *progress = NULL;
1112 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1113 verify_midx_error = 0;
1118 char *filename = get_midx_filename(object_dir);
1119 if (!stat(filename, &sb)) {
1120 error(_("multi-pack-index file exists, but failed to parse"));
1127 if (flags & MIDX_PROGRESS)
1128 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1130 for (i = 0; i < m->num_packs; i++) {
1131 if (prepare_midx_pack(r, m, i))
1132 midx_report("failed to load pack in position %d", i);
1134 display_progress(progress, i + 1);
1136 stop_progress(&progress);
1138 for (i = 0; i < 255; i++) {
1139 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1140 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1142 if (oid_fanout1 > oid_fanout2)
1143 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1144 i, oid_fanout1, oid_fanout2, i + 1);
1147 if (m->num_objects == 0) {
1148 midx_report(_("the midx contains no oid"));
1150 * Remaining tests assume that we have objects, so we can
1153 return verify_midx_error;
1156 if (flags & MIDX_PROGRESS)
1157 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1158 m->num_objects - 1);
1159 for (i = 0; i < m->num_objects - 1; i++) {
1160 struct object_id oid1, oid2;
1162 nth_midxed_object_oid(&oid1, m, i);
1163 nth_midxed_object_oid(&oid2, m, i + 1);
1165 if (oidcmp(&oid1, &oid2) >= 0)
1166 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1167 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1169 midx_display_sparse_progress(progress, i + 1);
1171 stop_progress(&progress);
1174 * Create an array mapping each object to its packfile id. Sort it
1175 * to group the objects by packfile. Use this permutation to visit
1176 * each of the objects and only require 1 packfile to be open at a
1179 ALLOC_ARRAY(pairs, m->num_objects);
1180 for (i = 0; i < m->num_objects; i++) {
1182 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1185 if (flags & MIDX_PROGRESS)
1186 progress = start_sparse_progress(_("Sorting objects by packfile"),
1188 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1189 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1190 stop_progress(&progress);
1192 if (flags & MIDX_PROGRESS)
1193 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1194 for (i = 0; i < m->num_objects; i++) {
1195 struct object_id oid;
1196 struct pack_entry e;
1197 off_t m_offset, p_offset;
1199 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1200 m->packs[pairs[i-1].pack_int_id])
1202 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1203 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1206 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1208 if (!fill_midx_entry(r, &oid, &e, m)) {
1209 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1210 pairs[i].pos, oid_to_hex(&oid));
1214 if (open_pack_index(e.p)) {
1215 midx_report(_("failed to load pack-index for packfile %s"),
1220 m_offset = e.offset;
1221 p_offset = find_pack_entry_one(oid.hash, e.p);
1223 if (m_offset != p_offset)
1224 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1225 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1227 midx_display_sparse_progress(progress, i + 1);
1229 stop_progress(&progress);
1233 return verify_midx_error;
1236 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1238 uint32_t i, *count, result = 0;
1239 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1240 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1241 struct progress *progress = NULL;
1246 count = xcalloc(m->num_packs, sizeof(uint32_t));
1248 if (flags & MIDX_PROGRESS)
1249 progress = start_delayed_progress(_("Counting referenced objects"),
1251 for (i = 0; i < m->num_objects; i++) {
1252 int pack_int_id = nth_midxed_pack_int_id(m, i);
1253 count[pack_int_id]++;
1254 display_progress(progress, i + 1);
1256 stop_progress(&progress);
1258 if (flags & MIDX_PROGRESS)
1259 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1261 for (i = 0; i < m->num_packs; i++) {
1263 display_progress(progress, i + 1);
1268 if (prepare_midx_pack(r, m, i))
1271 if (m->packs[i]->pack_keep)
1274 pack_name = xstrdup(m->packs[i]->pack_name);
1275 close_pack(m->packs[i]);
1277 string_list_insert(&packs_to_drop, m->pack_names[i]);
1278 unlink_pack_path(pack_name, 0);
1281 stop_progress(&progress);
1285 if (packs_to_drop.nr)
1286 result = write_midx_internal(object_dir, m, &packs_to_drop, flags);
1288 string_list_clear(&packs_to_drop, 0);
1292 struct repack_info {
1294 uint32_t referenced_objects;
1295 uint32_t pack_int_id;
1298 static int compare_by_mtime(const void *a_, const void *b_)
1300 const struct repack_info *a, *b;
1302 a = (const struct repack_info *)a_;
1303 b = (const struct repack_info *)b_;
1305 if (a->mtime < b->mtime)
1307 if (a->mtime > b->mtime)
1312 static int fill_included_packs_all(struct repository *r,
1313 struct multi_pack_index *m,
1314 unsigned char *include_pack)
1316 uint32_t i, count = 0;
1317 int pack_kept_objects = 0;
1319 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1321 for (i = 0; i < m->num_packs; i++) {
1322 if (prepare_midx_pack(r, m, i))
1324 if (!pack_kept_objects && m->packs[i]->pack_keep)
1327 include_pack[i] = 1;
1334 static int fill_included_packs_batch(struct repository *r,
1335 struct multi_pack_index *m,
1336 unsigned char *include_pack,
1339 uint32_t i, packs_to_repack;
1341 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1342 int pack_kept_objects = 0;
1344 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1346 for (i = 0; i < m->num_packs; i++) {
1347 pack_info[i].pack_int_id = i;
1349 if (prepare_midx_pack(r, m, i))
1352 pack_info[i].mtime = m->packs[i]->mtime;
1355 for (i = 0; batch_size && i < m->num_objects; i++) {
1356 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1357 pack_info[pack_int_id].referenced_objects++;
1360 QSORT(pack_info, m->num_packs, compare_by_mtime);
1363 packs_to_repack = 0;
1364 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1365 int pack_int_id = pack_info[i].pack_int_id;
1366 struct packed_git *p = m->packs[pack_int_id];
1367 size_t expected_size;
1371 if (!pack_kept_objects && p->pack_keep)
1373 if (open_pack_index(p) || !p->num_objects)
1376 expected_size = (size_t)(p->pack_size
1377 * pack_info[i].referenced_objects);
1378 expected_size /= p->num_objects;
1380 if (expected_size >= batch_size)
1384 total_size += expected_size;
1385 include_pack[pack_int_id] = 1;
1390 if (packs_to_repack < 2)
1396 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1400 unsigned char *include_pack;
1401 struct child_process cmd = CHILD_PROCESS_INIT;
1403 struct strbuf base_name = STRBUF_INIT;
1404 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1407 * When updating the default for these configuration
1408 * variables in builtin/repack.c, these must be adjusted
1411 int delta_base_offset = 1;
1412 int use_delta_islands = 0;
1417 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1420 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1422 } else if (fill_included_packs_all(r, m, include_pack))
1425 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1426 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1428 strvec_push(&cmd.args, "pack-objects");
1430 strbuf_addstr(&base_name, object_dir);
1431 strbuf_addstr(&base_name, "/pack/pack");
1432 strvec_push(&cmd.args, base_name.buf);
1434 if (delta_base_offset)
1435 strvec_push(&cmd.args, "--delta-base-offset");
1436 if (use_delta_islands)
1437 strvec_push(&cmd.args, "--delta-islands");
1439 if (flags & MIDX_PROGRESS)
1440 strvec_push(&cmd.args, "--progress");
1442 strvec_push(&cmd.args, "-q");
1444 strbuf_release(&base_name);
1447 cmd.in = cmd.out = -1;
1449 if (start_command(&cmd)) {
1450 error(_("could not start pack-objects"));
1455 cmd_in = xfdopen(cmd.in, "w");
1457 for (i = 0; i < m->num_objects; i++) {
1458 struct object_id oid;
1459 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1461 if (!include_pack[pack_int_id])
1464 nth_midxed_object_oid(&oid, m, i);
1465 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
1469 if (finish_command(&cmd)) {
1470 error(_("could not finish pack-objects"));
1475 result = write_midx_internal(object_dir, m, NULL, flags);