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 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
17 #define MIDX_VERSION 1
18 #define MIDX_BYTE_FILE_VERSION 4
19 #define MIDX_BYTE_HASH_VERSION 5
20 #define MIDX_BYTE_NUM_CHUNKS 6
21 #define MIDX_BYTE_NUM_PACKS 8
22 #define MIDX_HEADER_SIZE 12
23 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
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_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
32 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
33 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
34 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
36 #define PACK_EXPIRED UINT_MAX
38 static uint8_t oid_version(void)
40 switch (hash_algo_by_ptr(the_hash_algo)) {
46 die(_("invalid hash version"));
50 static const unsigned char *get_midx_checksum(struct multi_pack_index *m)
52 return m->data + m->data_len - the_hash_algo->rawsz;
55 static char *get_midx_filename(const char *object_dir)
57 return xstrfmt("%s/pack/multi-pack-index", object_dir);
60 char *get_midx_rev_filename(struct multi_pack_index *m)
62 return xstrfmt("%s/pack/multi-pack-index-%s.rev",
63 m->object_dir, hash_to_hex(get_midx_checksum(m)));
66 static int midx_read_oid_fanout(const unsigned char *chunk_start,
67 size_t chunk_size, void *data)
69 struct multi_pack_index *m = data;
70 m->chunk_oid_fanout = (uint32_t *)chunk_start;
72 if (chunk_size != 4 * 256) {
73 error(_("multi-pack-index OID fanout is of the wrong size"));
79 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
81 struct multi_pack_index *m = NULL;
85 void *midx_map = NULL;
86 uint32_t hash_version;
87 char *midx_name = get_midx_filename(object_dir);
89 const char *cur_pack_name;
90 struct chunkfile *cf = NULL;
92 fd = git_open(midx_name);
97 error_errno(_("failed to read %s"), midx_name);
101 midx_size = xsize_t(st.st_size);
103 if (midx_size < MIDX_MIN_SIZE) {
104 error(_("multi-pack-index file %s is too small"), midx_name);
108 FREE_AND_NULL(midx_name);
110 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
113 FLEX_ALLOC_STR(m, object_dir, object_dir);
115 m->data_len = midx_size;
118 m->signature = get_be32(m->data);
119 if (m->signature != MIDX_SIGNATURE)
120 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
121 m->signature, MIDX_SIGNATURE);
123 m->version = m->data[MIDX_BYTE_FILE_VERSION];
124 if (m->version != MIDX_VERSION)
125 die(_("multi-pack-index version %d not recognized"),
128 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
129 if (hash_version != oid_version()) {
130 error(_("multi-pack-index hash version %u does not match version %u"),
131 hash_version, oid_version());
134 m->hash_len = the_hash_algo->rawsz;
136 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
138 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
140 cf = init_chunkfile(NULL);
142 if (read_table_of_contents(cf, m->data, midx_size,
143 MIDX_HEADER_SIZE, m->num_chunks))
146 if (pair_chunk(cf, MIDX_CHUNKID_PACKNAMES, &m->chunk_pack_names) == CHUNK_NOT_FOUND)
147 die(_("multi-pack-index missing required pack-name chunk"));
148 if (read_chunk(cf, MIDX_CHUNKID_OIDFANOUT, midx_read_oid_fanout, m) == CHUNK_NOT_FOUND)
149 die(_("multi-pack-index missing required OID fanout chunk"));
150 if (pair_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, &m->chunk_oid_lookup) == CHUNK_NOT_FOUND)
151 die(_("multi-pack-index missing required OID lookup chunk"));
152 if (pair_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, &m->chunk_object_offsets) == CHUNK_NOT_FOUND)
153 die(_("multi-pack-index missing required object offsets chunk"));
155 pair_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, &m->chunk_large_offsets);
157 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
159 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
160 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
162 cur_pack_name = (const char *)m->chunk_pack_names;
163 for (i = 0; i < m->num_packs; i++) {
164 m->pack_names[i] = cur_pack_name;
166 cur_pack_name += strlen(cur_pack_name) + 1;
168 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
169 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
170 m->pack_names[i - 1],
174 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
175 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
184 munmap(midx_map, midx_size);
190 void close_midx(struct multi_pack_index *m)
197 munmap((unsigned char *)m->data, m->data_len);
199 for (i = 0; i < m->num_packs; i++) {
201 m->packs[i]->multi_pack_index = 0;
203 FREE_AND_NULL(m->packs);
204 FREE_AND_NULL(m->pack_names);
207 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
209 struct strbuf pack_name = STRBUF_INIT;
210 struct packed_git *p;
212 if (pack_int_id >= m->num_packs)
213 die(_("bad pack-int-id: %u (%u total packs)"),
214 pack_int_id, m->num_packs);
216 if (m->packs[pack_int_id])
219 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
220 m->pack_names[pack_int_id]);
222 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
223 strbuf_release(&pack_name);
228 p->multi_pack_index = 1;
229 m->packs[pack_int_id] = p;
230 install_packed_git(r, p);
231 list_add_tail(&p->mru, &r->objects->packed_git_mru);
236 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
238 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
239 the_hash_algo->rawsz, result);
242 struct object_id *nth_midxed_object_oid(struct object_id *oid,
243 struct multi_pack_index *m,
246 if (n >= m->num_objects)
249 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
253 off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
255 const unsigned char *offset_data;
258 offset_data = m->chunk_object_offsets + (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH;
259 offset32 = get_be32(offset_data + sizeof(uint32_t));
261 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
262 if (sizeof(off_t) < sizeof(uint64_t))
263 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
265 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
266 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
272 uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
274 return get_be32(m->chunk_object_offsets +
275 (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH);
278 static int nth_midxed_pack_entry(struct repository *r,
279 struct multi_pack_index *m,
280 struct pack_entry *e,
283 uint32_t pack_int_id;
284 struct packed_git *p;
286 if (pos >= m->num_objects)
289 pack_int_id = nth_midxed_pack_int_id(m, pos);
291 if (prepare_midx_pack(r, m, pack_int_id))
293 p = m->packs[pack_int_id];
296 * We are about to tell the caller where they can locate the
297 * requested object. We better make sure the packfile is
298 * still here and can be accessed before supplying that
299 * answer, as it may have been deleted since the MIDX was
302 if (!is_pack_valid(p))
305 if (p->num_bad_objects) {
307 struct object_id oid;
308 nth_midxed_object_oid(&oid, m, pos);
309 for (i = 0; i < p->num_bad_objects; i++)
311 p->bad_object_sha1 + the_hash_algo->rawsz * i))
315 e->offset = nth_midxed_offset(m, pos);
321 int fill_midx_entry(struct repository * r,
322 const struct object_id *oid,
323 struct pack_entry *e,
324 struct multi_pack_index *m)
328 if (!bsearch_midx(oid, m, &pos))
331 return nth_midxed_pack_entry(r, m, e, pos);
334 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
335 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
336 const char *idx_name)
338 /* Skip past any initial matching prefix. */
339 while (*idx_name && *idx_name == *idx_or_pack_name) {
345 * If we didn't match completely, we may have matched "pack-1234." and
346 * be left with "idx" and "pack" respectively, which is also OK. We do
347 * not have to check for "idx" and "idx", because that would have been
348 * a complete match (and in that case these strcmps will be false, but
349 * we'll correctly return 0 from the final strcmp() below.
351 * Technically this matches "fooidx" and "foopack", but we'd never have
352 * such names in the first place.
354 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
358 * This not only checks for a complete match, but also orders based on
359 * the first non-identical character, which means our ordering will
360 * match a raw strcmp(). That makes it OK to use this to binary search
361 * a naively-sorted list.
363 return strcmp(idx_or_pack_name, idx_name);
366 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
368 uint32_t first = 0, last = m->num_packs;
370 while (first < last) {
371 uint32_t mid = first + (last - first) / 2;
375 current = m->pack_names[mid];
376 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
389 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
391 struct multi_pack_index *m;
392 struct multi_pack_index *m_search;
394 prepare_repo_settings(r);
395 if (!r->settings.core_multi_pack_index)
398 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
399 if (!strcmp(object_dir, m_search->object_dir))
402 m = load_multi_pack_index(object_dir, local);
405 struct multi_pack_index *mp = r->objects->multi_pack_index;
410 r->objects->multi_pack_index = m;
417 static size_t write_midx_header(struct hashfile *f,
418 unsigned char num_chunks,
421 hashwrite_be32(f, MIDX_SIGNATURE);
422 hashwrite_u8(f, MIDX_VERSION);
423 hashwrite_u8(f, oid_version());
424 hashwrite_u8(f, num_chunks);
425 hashwrite_u8(f, 0); /* unused */
426 hashwrite_be32(f, num_packs);
428 return MIDX_HEADER_SIZE;
432 uint32_t orig_pack_int_id;
434 struct packed_git *p;
435 unsigned expired : 1;
438 static int pack_info_compare(const void *_a, const void *_b)
440 struct pack_info *a = (struct pack_info *)_a;
441 struct pack_info *b = (struct pack_info *)_b;
442 return strcmp(a->pack_name, b->pack_name);
445 static int idx_or_pack_name_cmp(const void *_va, const void *_vb)
447 const char *pack_name = _va;
448 const struct pack_info *compar = _vb;
450 return cmp_idx_or_pack_name(pack_name, compar->pack_name);
453 struct write_midx_context {
454 struct pack_info *info;
457 struct multi_pack_index *m;
458 struct progress *progress;
459 unsigned pack_paths_checked;
461 struct pack_midx_entry *entries;
465 unsigned large_offsets_needed:1;
466 uint32_t num_large_offsets;
468 int preferred_pack_idx;
471 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
472 const char *file_name, void *data)
474 struct write_midx_context *ctx = data;
476 if (ends_with(file_name, ".idx")) {
477 display_progress(ctx->progress, ++ctx->pack_paths_checked);
478 if (ctx->m && midx_contains_pack(ctx->m, file_name))
481 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
483 ctx->info[ctx->nr].p = add_packed_git(full_path,
487 if (!ctx->info[ctx->nr].p) {
488 warning(_("failed to add packfile '%s'"),
493 if (open_pack_index(ctx->info[ctx->nr].p)) {
494 warning(_("failed to open pack-index '%s'"),
496 close_pack(ctx->info[ctx->nr].p);
497 FREE_AND_NULL(ctx->info[ctx->nr].p);
501 ctx->info[ctx->nr].pack_name = xstrdup(file_name);
502 ctx->info[ctx->nr].orig_pack_int_id = ctx->nr;
503 ctx->info[ctx->nr].expired = 0;
508 struct pack_midx_entry {
509 struct object_id oid;
510 uint32_t pack_int_id;
513 unsigned preferred : 1;
516 static int midx_oid_compare(const void *_a, const void *_b)
518 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
519 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
520 int cmp = oidcmp(&a->oid, &b->oid);
525 /* Sort objects in a preferred pack first when multiple copies exist. */
526 if (a->preferred > b->preferred)
528 if (a->preferred < b->preferred)
531 if (a->pack_mtime > b->pack_mtime)
533 else if (a->pack_mtime < b->pack_mtime)
536 return a->pack_int_id - b->pack_int_id;
539 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
540 struct pack_midx_entry *e,
543 if (pos >= m->num_objects)
546 nth_midxed_object_oid(&e->oid, m, pos);
547 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
548 e->offset = nth_midxed_offset(m, pos);
550 /* consider objects in midx to be from "old" packs */
555 static void fill_pack_entry(uint32_t pack_int_id,
556 struct packed_git *p,
558 struct pack_midx_entry *entry,
561 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
562 die(_("failed to locate object %d in packfile"), cur_object);
564 entry->pack_int_id = pack_int_id;
565 entry->pack_mtime = p->mtime;
567 entry->offset = nth_packed_object_offset(p, cur_object);
568 entry->preferred = !!preferred;
572 * It is possible to artificially get into a state where there are many
573 * duplicate copies of objects. That can create high memory pressure if
574 * we are to create a list of all objects before de-duplication. To reduce
575 * this memory pressure without a significant performance drop, automatically
576 * group objects by the first byte of their object id. Use the IDX fanout
577 * tables to group the data, copy to a local array, then sort.
579 * Copy only the de-duplicated entries (selected by most-recent modified time
580 * of a packfile containing the object).
582 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
583 struct pack_info *info,
585 uint32_t *nr_objects,
588 uint32_t cur_fanout, cur_pack, cur_object;
589 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
590 struct pack_midx_entry *entries_by_fanout = NULL;
591 struct pack_midx_entry *deduplicated_entries = NULL;
592 uint32_t start_pack = m ? m->num_packs : 0;
594 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
595 total_objects += info[cur_pack].p->num_objects;
598 * As we de-duplicate by fanout value, we expect the fanout
599 * slices to be evenly distributed, with some noise. Hence,
600 * allocate slightly more than one 256th.
602 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
604 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
605 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
608 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
609 uint32_t nr_fanout = 0;
612 uint32_t start = 0, end;
615 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
616 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
618 for (cur_object = start; cur_object < end; cur_object++) {
619 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
620 nth_midxed_pack_midx_entry(m,
621 &entries_by_fanout[nr_fanout],
623 if (nth_midxed_pack_int_id(m, cur_object) == preferred_pack)
624 entries_by_fanout[nr_fanout].preferred = 1;
626 entries_by_fanout[nr_fanout].preferred = 0;
631 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
632 uint32_t start = 0, end;
633 int preferred = cur_pack == preferred_pack;
636 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
637 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
639 for (cur_object = start; cur_object < end; cur_object++) {
640 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
641 fill_pack_entry(cur_pack,
644 &entries_by_fanout[nr_fanout],
650 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
653 * The batch is now sorted by OID and then mtime (descending).
654 * Take only the first duplicate.
656 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
657 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
658 &entries_by_fanout[cur_object].oid))
661 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
662 memcpy(&deduplicated_entries[*nr_objects],
663 &entries_by_fanout[cur_object],
664 sizeof(struct pack_midx_entry));
669 free(entries_by_fanout);
670 return deduplicated_entries;
673 static int write_midx_pack_names(struct hashfile *f, void *data)
675 struct write_midx_context *ctx = data;
677 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
680 for (i = 0; i < ctx->nr; i++) {
683 if (ctx->info[i].expired)
686 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
687 BUG("incorrect pack-file order: %s before %s",
688 ctx->info[i - 1].pack_name,
689 ctx->info[i].pack_name);
691 writelen = strlen(ctx->info[i].pack_name) + 1;
692 hashwrite(f, ctx->info[i].pack_name, writelen);
696 /* add padding to be aligned */
697 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
698 if (i < MIDX_CHUNK_ALIGNMENT) {
699 memset(padding, 0, sizeof(padding));
700 hashwrite(f, padding, i);
706 static int write_midx_oid_fanout(struct hashfile *f,
709 struct write_midx_context *ctx = data;
710 struct pack_midx_entry *list = ctx->entries;
711 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
716 * Write the first-level table (the list is sorted,
717 * but we use a 256-entry lookup to be able to avoid
718 * having to do eight extra binary search iterations).
720 for (i = 0; i < 256; i++) {
721 struct pack_midx_entry *next = list;
723 while (next < last && next->oid.hash[0] == i) {
728 hashwrite_be32(f, count);
735 static int write_midx_oid_lookup(struct hashfile *f,
738 struct write_midx_context *ctx = data;
739 unsigned char hash_len = the_hash_algo->rawsz;
740 struct pack_midx_entry *list = ctx->entries;
743 for (i = 0; i < ctx->entries_nr; i++) {
744 struct pack_midx_entry *obj = list++;
746 if (i < ctx->entries_nr - 1) {
747 struct pack_midx_entry *next = list;
748 if (oidcmp(&obj->oid, &next->oid) >= 0)
749 BUG("OIDs not in order: %s >= %s",
750 oid_to_hex(&obj->oid),
751 oid_to_hex(&next->oid));
754 hashwrite(f, obj->oid.hash, (int)hash_len);
760 static int write_midx_object_offsets(struct hashfile *f,
763 struct write_midx_context *ctx = data;
764 struct pack_midx_entry *list = ctx->entries;
765 uint32_t i, nr_large_offset = 0;
767 for (i = 0; i < ctx->entries_nr; i++) {
768 struct pack_midx_entry *obj = list++;
770 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
771 BUG("object %s is in an expired pack with int-id %d",
772 oid_to_hex(&obj->oid),
775 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
777 if (ctx->large_offsets_needed && obj->offset >> 31)
778 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
779 else if (!ctx->large_offsets_needed && obj->offset >> 32)
780 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
781 oid_to_hex(&obj->oid),
784 hashwrite_be32(f, (uint32_t)obj->offset);
790 static int write_midx_large_offsets(struct hashfile *f,
793 struct write_midx_context *ctx = data;
794 struct pack_midx_entry *list = ctx->entries;
795 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
796 uint32_t nr_large_offset = ctx->num_large_offsets;
798 while (nr_large_offset) {
799 struct pack_midx_entry *obj;
803 BUG("too many large-offset objects");
806 offset = obj->offset;
811 hashwrite_be64(f, offset);
819 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
820 struct string_list *packs_to_drop,
821 const char *preferred_pack_name,
825 unsigned char midx_hash[GIT_MAX_RAWSZ];
827 struct hashfile *f = NULL;
829 struct write_midx_context ctx = { 0 };
830 int pack_name_concat_len = 0;
831 int dropped_packs = 0;
833 struct chunkfile *cf;
835 midx_name = get_midx_filename(object_dir);
836 if (safe_create_leading_directories(midx_name))
837 die_errno(_("unable to create leading directories of %s"),
843 ctx.m = load_multi_pack_index(object_dir, 1);
846 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
848 ALLOC_ARRAY(ctx.info, ctx.alloc);
851 for (i = 0; i < ctx.m->num_packs; i++) {
852 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
854 ctx.info[ctx.nr].orig_pack_int_id = i;
855 ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]);
856 ctx.info[ctx.nr].p = NULL;
857 ctx.info[ctx.nr].expired = 0;
862 ctx.pack_paths_checked = 0;
863 if (flags & MIDX_PROGRESS)
864 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
868 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
869 stop_progress(&ctx.progress);
871 if (ctx.m && ctx.nr == ctx.m->num_packs && !packs_to_drop)
874 ctx.preferred_pack_idx = -1;
875 if (preferred_pack_name) {
876 for (i = 0; i < ctx.nr; i++) {
877 if (!cmp_idx_or_pack_name(preferred_pack_name,
878 ctx.info[i].pack_name)) {
879 ctx.preferred_pack_idx = i;
885 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr,
886 ctx.preferred_pack_idx);
888 ctx.large_offsets_needed = 0;
889 for (i = 0; i < ctx.entries_nr; i++) {
890 if (ctx.entries[i].offset > 0x7fffffff)
891 ctx.num_large_offsets++;
892 if (ctx.entries[i].offset > 0xffffffff)
893 ctx.large_offsets_needed = 1;
896 QSORT(ctx.info, ctx.nr, pack_info_compare);
898 if (packs_to_drop && packs_to_drop->nr) {
900 int missing_drops = 0;
902 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
903 int cmp = strcmp(ctx.info[i].pack_name,
904 packs_to_drop->items[drop_index].string);
908 ctx.info[i].expired = 1;
909 } else if (cmp > 0) {
910 error(_("did not see pack-file %s to drop"),
911 packs_to_drop->items[drop_index].string);
916 ctx.info[i].expired = 0;
927 * pack_perm stores a permutation between pack-int-ids from the
928 * previous multi-pack-index to the new one we are writing:
930 * pack_perm[old_id] = new_id
932 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
933 for (i = 0; i < ctx.nr; i++) {
934 if (ctx.info[i].expired) {
936 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
938 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
942 for (i = 0; i < ctx.nr; i++) {
943 if (!ctx.info[i].expired)
944 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
947 /* Check that the preferred pack wasn't expired (if given). */
948 if (preferred_pack_name) {
949 struct pack_info *preferred = bsearch(preferred_pack_name,
952 idx_or_pack_name_cmp);
955 warning(_("unknown preferred pack: '%s'"),
956 preferred_pack_name);
958 uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id];
959 if (perm == PACK_EXPIRED)
960 warning(_("preferred pack '%s' is expired"),
961 preferred_pack_name);
965 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
966 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
967 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
969 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
970 f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk));
975 if (ctx.nr - dropped_packs == 0) {
976 error(_("no pack files to index."));
981 cf = init_chunkfile(f);
983 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
984 write_midx_pack_names);
985 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
986 write_midx_oid_fanout);
987 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
988 (size_t)ctx.entries_nr * the_hash_algo->rawsz,
989 write_midx_oid_lookup);
990 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
991 (size_t)ctx.entries_nr * MIDX_CHUNK_OFFSET_WIDTH,
992 write_midx_object_offsets);
994 if (ctx.large_offsets_needed)
995 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
996 (size_t)ctx.num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH,
997 write_midx_large_offsets);
999 write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs);
1000 write_chunkfile(cf, &ctx);
1002 finalize_hashfile(f, midx_hash, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1004 commit_lock_file(&lk);
1007 for (i = 0; i < ctx.nr; i++) {
1008 if (ctx.info[i].p) {
1009 close_pack(ctx.info[i].p);
1010 free(ctx.info[i].p);
1012 free(ctx.info[i].pack_name);
1017 free(ctx.pack_perm);
1022 int write_midx_file(const char *object_dir,
1023 const char *preferred_pack_name,
1026 return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name,
1030 void clear_midx_file(struct repository *r)
1032 char *midx = get_midx_filename(r->objects->odb->path);
1034 if (r->objects && r->objects->multi_pack_index) {
1035 close_midx(r->objects->multi_pack_index);
1036 r->objects->multi_pack_index = NULL;
1039 if (remove_path(midx))
1040 die(_("failed to clear multi-pack-index at %s"), midx);
1045 static int verify_midx_error;
1047 static void midx_report(const char *fmt, ...)
1050 verify_midx_error = 1;
1052 vfprintf(stderr, fmt, ap);
1053 fprintf(stderr, "\n");
1057 struct pair_pos_vs_id
1060 uint32_t pack_int_id;
1063 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1065 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1066 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1068 return b->pack_int_id - a->pack_int_id;
1072 * Limit calls to display_progress() for performance reasons.
1073 * The interval here was arbitrarily chosen.
1075 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1076 #define midx_display_sparse_progress(progress, n) \
1078 uint64_t _n = (n); \
1079 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1080 display_progress(progress, _n); \
1083 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1085 struct pair_pos_vs_id *pairs = NULL;
1087 struct progress *progress = NULL;
1088 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1089 verify_midx_error = 0;
1094 char *filename = get_midx_filename(object_dir);
1095 if (!stat(filename, &sb)) {
1096 error(_("multi-pack-index file exists, but failed to parse"));
1103 if (flags & MIDX_PROGRESS)
1104 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1106 for (i = 0; i < m->num_packs; i++) {
1107 if (prepare_midx_pack(r, m, i))
1108 midx_report("failed to load pack in position %d", i);
1110 display_progress(progress, i + 1);
1112 stop_progress(&progress);
1114 for (i = 0; i < 255; i++) {
1115 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1116 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1118 if (oid_fanout1 > oid_fanout2)
1119 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1120 i, oid_fanout1, oid_fanout2, i + 1);
1123 if (m->num_objects == 0) {
1124 midx_report(_("the midx contains no oid"));
1126 * Remaining tests assume that we have objects, so we can
1129 return verify_midx_error;
1132 if (flags & MIDX_PROGRESS)
1133 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1134 m->num_objects - 1);
1135 for (i = 0; i < m->num_objects - 1; i++) {
1136 struct object_id oid1, oid2;
1138 nth_midxed_object_oid(&oid1, m, i);
1139 nth_midxed_object_oid(&oid2, m, i + 1);
1141 if (oidcmp(&oid1, &oid2) >= 0)
1142 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1143 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1145 midx_display_sparse_progress(progress, i + 1);
1147 stop_progress(&progress);
1150 * Create an array mapping each object to its packfile id. Sort it
1151 * to group the objects by packfile. Use this permutation to visit
1152 * each of the objects and only require 1 packfile to be open at a
1155 ALLOC_ARRAY(pairs, m->num_objects);
1156 for (i = 0; i < m->num_objects; i++) {
1158 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1161 if (flags & MIDX_PROGRESS)
1162 progress = start_sparse_progress(_("Sorting objects by packfile"),
1164 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1165 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1166 stop_progress(&progress);
1168 if (flags & MIDX_PROGRESS)
1169 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1170 for (i = 0; i < m->num_objects; i++) {
1171 struct object_id oid;
1172 struct pack_entry e;
1173 off_t m_offset, p_offset;
1175 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1176 m->packs[pairs[i-1].pack_int_id])
1178 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1179 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1182 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1184 if (!fill_midx_entry(r, &oid, &e, m)) {
1185 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1186 pairs[i].pos, oid_to_hex(&oid));
1190 if (open_pack_index(e.p)) {
1191 midx_report(_("failed to load pack-index for packfile %s"),
1196 m_offset = e.offset;
1197 p_offset = find_pack_entry_one(oid.hash, e.p);
1199 if (m_offset != p_offset)
1200 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1201 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1203 midx_display_sparse_progress(progress, i + 1);
1205 stop_progress(&progress);
1209 return verify_midx_error;
1212 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1214 uint32_t i, *count, result = 0;
1215 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1216 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1217 struct progress *progress = NULL;
1222 count = xcalloc(m->num_packs, sizeof(uint32_t));
1224 if (flags & MIDX_PROGRESS)
1225 progress = start_delayed_progress(_("Counting referenced objects"),
1227 for (i = 0; i < m->num_objects; i++) {
1228 int pack_int_id = nth_midxed_pack_int_id(m, i);
1229 count[pack_int_id]++;
1230 display_progress(progress, i + 1);
1232 stop_progress(&progress);
1234 if (flags & MIDX_PROGRESS)
1235 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1237 for (i = 0; i < m->num_packs; i++) {
1239 display_progress(progress, i + 1);
1244 if (prepare_midx_pack(r, m, i))
1247 if (m->packs[i]->pack_keep)
1250 pack_name = xstrdup(m->packs[i]->pack_name);
1251 close_pack(m->packs[i]);
1253 string_list_insert(&packs_to_drop, m->pack_names[i]);
1254 unlink_pack_path(pack_name, 0);
1257 stop_progress(&progress);
1261 if (packs_to_drop.nr)
1262 result = write_midx_internal(object_dir, m, &packs_to_drop, NULL, flags);
1264 string_list_clear(&packs_to_drop, 0);
1268 struct repack_info {
1270 uint32_t referenced_objects;
1271 uint32_t pack_int_id;
1274 static int compare_by_mtime(const void *a_, const void *b_)
1276 const struct repack_info *a, *b;
1278 a = (const struct repack_info *)a_;
1279 b = (const struct repack_info *)b_;
1281 if (a->mtime < b->mtime)
1283 if (a->mtime > b->mtime)
1288 static int fill_included_packs_all(struct repository *r,
1289 struct multi_pack_index *m,
1290 unsigned char *include_pack)
1292 uint32_t i, count = 0;
1293 int pack_kept_objects = 0;
1295 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1297 for (i = 0; i < m->num_packs; i++) {
1298 if (prepare_midx_pack(r, m, i))
1300 if (!pack_kept_objects && m->packs[i]->pack_keep)
1303 include_pack[i] = 1;
1310 static int fill_included_packs_batch(struct repository *r,
1311 struct multi_pack_index *m,
1312 unsigned char *include_pack,
1315 uint32_t i, packs_to_repack;
1317 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1318 int pack_kept_objects = 0;
1320 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1322 for (i = 0; i < m->num_packs; i++) {
1323 pack_info[i].pack_int_id = i;
1325 if (prepare_midx_pack(r, m, i))
1328 pack_info[i].mtime = m->packs[i]->mtime;
1331 for (i = 0; batch_size && i < m->num_objects; i++) {
1332 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1333 pack_info[pack_int_id].referenced_objects++;
1336 QSORT(pack_info, m->num_packs, compare_by_mtime);
1339 packs_to_repack = 0;
1340 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1341 int pack_int_id = pack_info[i].pack_int_id;
1342 struct packed_git *p = m->packs[pack_int_id];
1343 size_t expected_size;
1347 if (!pack_kept_objects && p->pack_keep)
1349 if (open_pack_index(p) || !p->num_objects)
1352 expected_size = (size_t)(p->pack_size
1353 * pack_info[i].referenced_objects);
1354 expected_size /= p->num_objects;
1356 if (expected_size >= batch_size)
1360 total_size += expected_size;
1361 include_pack[pack_int_id] = 1;
1366 if (packs_to_repack < 2)
1372 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1376 unsigned char *include_pack;
1377 struct child_process cmd = CHILD_PROCESS_INIT;
1379 struct strbuf base_name = STRBUF_INIT;
1380 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1383 * When updating the default for these configuration
1384 * variables in builtin/repack.c, these must be adjusted
1387 int delta_base_offset = 1;
1388 int use_delta_islands = 0;
1393 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1396 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1398 } else if (fill_included_packs_all(r, m, include_pack))
1401 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1402 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1404 strvec_push(&cmd.args, "pack-objects");
1406 strbuf_addstr(&base_name, object_dir);
1407 strbuf_addstr(&base_name, "/pack/pack");
1408 strvec_push(&cmd.args, base_name.buf);
1410 if (delta_base_offset)
1411 strvec_push(&cmd.args, "--delta-base-offset");
1412 if (use_delta_islands)
1413 strvec_push(&cmd.args, "--delta-islands");
1415 if (flags & MIDX_PROGRESS)
1416 strvec_push(&cmd.args, "--progress");
1418 strvec_push(&cmd.args, "-q");
1420 strbuf_release(&base_name);
1423 cmd.in = cmd.out = -1;
1425 if (start_command(&cmd)) {
1426 error(_("could not start pack-objects"));
1431 cmd_in = xfdopen(cmd.in, "w");
1433 for (i = 0; i < m->num_objects; i++) {
1434 struct object_id oid;
1435 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1437 if (!include_pack[pack_int_id])
1440 nth_midxed_object_oid(&oid, m, i);
1441 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
1445 if (finish_command(&cmd)) {
1446 error(_("could not finish pack-objects"));
1451 result = write_midx_internal(object_dir, m, NULL, NULL, flags);