7 #include "object-store.h"
8 #include "sha1-lookup.h"
12 #include "run-command.h"
14 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
15 #define MIDX_VERSION 1
16 #define MIDX_BYTE_FILE_VERSION 4
17 #define MIDX_BYTE_HASH_VERSION 5
18 #define MIDX_BYTE_NUM_CHUNKS 6
19 #define MIDX_BYTE_NUM_PACKS 8
20 #define MIDX_HASH_VERSION 1
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 char *get_midx_filename(const char *object_dir)
41 return xstrfmt("%s/pack/multi-pack-index", object_dir);
44 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
46 struct multi_pack_index *m = NULL;
50 void *midx_map = NULL;
51 uint32_t hash_version;
52 char *midx_name = get_midx_filename(object_dir);
54 const char *cur_pack_name;
56 fd = git_open(midx_name);
61 error_errno(_("failed to read %s"), midx_name);
65 midx_size = xsize_t(st.st_size);
67 if (midx_size < MIDX_MIN_SIZE) {
68 error(_("multi-pack-index file %s is too small"), midx_name);
72 FREE_AND_NULL(midx_name);
74 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
76 FLEX_ALLOC_STR(m, object_dir, object_dir);
79 m->data_len = midx_size;
82 m->signature = get_be32(m->data);
83 if (m->signature != MIDX_SIGNATURE)
84 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
85 m->signature, MIDX_SIGNATURE);
87 m->version = m->data[MIDX_BYTE_FILE_VERSION];
88 if (m->version != MIDX_VERSION)
89 die(_("multi-pack-index version %d not recognized"),
92 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
93 if (hash_version != MIDX_HASH_VERSION)
94 die(_("hash version %u does not match"), hash_version);
95 m->hash_len = the_hash_algo->rawsz;
97 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
99 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
101 for (i = 0; i < m->num_chunks; i++) {
102 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
103 MIDX_CHUNKLOOKUP_WIDTH * i);
104 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
105 MIDX_CHUNKLOOKUP_WIDTH * i);
107 if (chunk_offset >= m->data_len)
108 die(_("invalid chunk offset (too large)"));
111 case MIDX_CHUNKID_PACKNAMES:
112 m->chunk_pack_names = m->data + chunk_offset;
115 case MIDX_CHUNKID_OIDFANOUT:
116 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
119 case MIDX_CHUNKID_OIDLOOKUP:
120 m->chunk_oid_lookup = m->data + chunk_offset;
123 case MIDX_CHUNKID_OBJECTOFFSETS:
124 m->chunk_object_offsets = m->data + chunk_offset;
127 case MIDX_CHUNKID_LARGEOFFSETS:
128 m->chunk_large_offsets = m->data + chunk_offset;
132 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
137 * Do nothing on unrecognized chunks, allowing future
138 * extensions to add optional chunks.
144 if (!m->chunk_pack_names)
145 die(_("multi-pack-index missing required pack-name chunk"));
146 if (!m->chunk_oid_fanout)
147 die(_("multi-pack-index missing required OID fanout chunk"));
148 if (!m->chunk_oid_lookup)
149 die(_("multi-pack-index missing required OID lookup chunk"));
150 if (!m->chunk_object_offsets)
151 die(_("multi-pack-index missing required object offsets chunk"));
153 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
155 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
156 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
158 cur_pack_name = (const char *)m->chunk_pack_names;
159 for (i = 0; i < m->num_packs; i++) {
160 m->pack_names[i] = cur_pack_name;
162 cur_pack_name += strlen(cur_pack_name) + 1;
164 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
165 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
166 m->pack_names[i - 1],
170 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
171 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
179 munmap(midx_map, midx_size);
185 void close_midx(struct multi_pack_index *m)
192 munmap((unsigned char *)m->data, m->data_len);
196 for (i = 0; i < m->num_packs; i++) {
198 m->packs[i]->multi_pack_index = 0;
200 FREE_AND_NULL(m->packs);
201 FREE_AND_NULL(m->pack_names);
204 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
206 struct strbuf pack_name = STRBUF_INIT;
207 struct packed_git *p;
209 if (pack_int_id >= m->num_packs)
210 die(_("bad pack-int-id: %u (%u total packs)"),
211 pack_int_id, m->num_packs);
213 if (m->packs[pack_int_id])
216 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
217 m->pack_names[pack_int_id]);
219 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
220 strbuf_release(&pack_name);
225 p->multi_pack_index = 1;
226 m->packs[pack_int_id] = p;
227 install_packed_git(r, p);
228 list_add_tail(&p->mru, &r->objects->packed_git_mru);
233 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
235 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
236 the_hash_algo->rawsz, result);
239 struct object_id *nth_midxed_object_oid(struct object_id *oid,
240 struct multi_pack_index *m,
243 if (n >= m->num_objects)
246 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
250 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
252 const unsigned char *offset_data;
255 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
256 offset32 = get_be32(offset_data + sizeof(uint32_t));
258 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
259 if (sizeof(off_t) < sizeof(uint64_t))
260 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
262 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
263 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
269 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
271 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
274 static int nth_midxed_pack_entry(struct repository *r,
275 struct multi_pack_index *m,
276 struct pack_entry *e,
279 uint32_t pack_int_id;
280 struct packed_git *p;
282 if (pos >= m->num_objects)
285 pack_int_id = nth_midxed_pack_int_id(m, pos);
287 if (prepare_midx_pack(r, m, pack_int_id))
288 die(_("error preparing packfile from multi-pack-index"));
289 p = m->packs[pack_int_id];
292 * We are about to tell the caller where they can locate the
293 * requested object. We better make sure the packfile is
294 * still here and can be accessed before supplying that
295 * answer, as it may have been deleted since the MIDX was
298 if (!is_pack_valid(p))
301 if (p->num_bad_objects) {
303 struct object_id oid;
304 nth_midxed_object_oid(&oid, m, pos);
305 for (i = 0; i < p->num_bad_objects; i++)
307 p->bad_object_sha1 + the_hash_algo->rawsz * i))
311 e->offset = nth_midxed_offset(m, pos);
317 int fill_midx_entry(struct repository * r,
318 const struct object_id *oid,
319 struct pack_entry *e,
320 struct multi_pack_index *m)
324 if (!bsearch_midx(oid, m, &pos))
327 return nth_midxed_pack_entry(r, m, e, pos);
330 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
331 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
332 const char *idx_name)
334 /* Skip past any initial matching prefix. */
335 while (*idx_name && *idx_name == *idx_or_pack_name) {
341 * If we didn't match completely, we may have matched "pack-1234." and
342 * be left with "idx" and "pack" respectively, which is also OK. We do
343 * not have to check for "idx" and "idx", because that would have been
344 * a complete match (and in that case these strcmps will be false, but
345 * we'll correctly return 0 from the final strcmp() below.
347 * Technically this matches "fooidx" and "foopack", but we'd never have
348 * such names in the first place.
350 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
354 * This not only checks for a complete match, but also orders based on
355 * the first non-identical character, which means our ordering will
356 * match a raw strcmp(). That makes it OK to use this to binary search
357 * a naively-sorted list.
359 return strcmp(idx_or_pack_name, idx_name);
362 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
364 uint32_t first = 0, last = m->num_packs;
366 while (first < last) {
367 uint32_t mid = first + (last - first) / 2;
371 current = m->pack_names[mid];
372 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
385 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
387 struct multi_pack_index *m;
388 struct multi_pack_index *m_search;
390 static int env_value = -1;
393 env_value = git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0);
396 (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
400 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
401 if (!strcmp(object_dir, m_search->object_dir))
404 m = load_multi_pack_index(object_dir, local);
407 m->next = r->objects->multi_pack_index;
408 r->objects->multi_pack_index = m;
415 static size_t write_midx_header(struct hashfile *f,
416 unsigned char num_chunks,
419 unsigned char byte_values[4];
421 hashwrite_be32(f, MIDX_SIGNATURE);
422 byte_values[0] = MIDX_VERSION;
423 byte_values[1] = MIDX_HASH_VERSION;
424 byte_values[2] = num_chunks;
425 byte_values[3] = 0; /* unused */
426 hashwrite(f, byte_values, sizeof(byte_values));
427 hashwrite_be32(f, num_packs);
429 return MIDX_HEADER_SIZE;
433 uint32_t orig_pack_int_id;
435 struct packed_git *p;
436 unsigned expired : 1;
439 static int pack_info_compare(const void *_a, const void *_b)
441 struct pack_info *a = (struct pack_info *)_a;
442 struct pack_info *b = (struct pack_info *)_b;
443 return strcmp(a->pack_name, b->pack_name);
447 struct pack_info *info;
450 struct multi_pack_index *m;
453 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
454 const char *file_name, void *data)
456 struct pack_list *packs = (struct pack_list *)data;
458 if (ends_with(file_name, ".idx")) {
459 if (packs->m && midx_contains_pack(packs->m, file_name))
462 ALLOC_GROW(packs->info, packs->nr + 1, packs->alloc);
464 packs->info[packs->nr].p = add_packed_git(full_path,
468 if (!packs->info[packs->nr].p) {
469 warning(_("failed to add packfile '%s'"),
474 if (open_pack_index(packs->info[packs->nr].p)) {
475 warning(_("failed to open pack-index '%s'"),
477 close_pack(packs->info[packs->nr].p);
478 FREE_AND_NULL(packs->info[packs->nr].p);
482 packs->info[packs->nr].pack_name = xstrdup(file_name);
483 packs->info[packs->nr].orig_pack_int_id = packs->nr;
484 packs->info[packs->nr].expired = 0;
489 struct pack_midx_entry {
490 struct object_id oid;
491 uint32_t pack_int_id;
496 static int midx_oid_compare(const void *_a, const void *_b)
498 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
499 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
500 int cmp = oidcmp(&a->oid, &b->oid);
505 if (a->pack_mtime > b->pack_mtime)
507 else if (a->pack_mtime < b->pack_mtime)
510 return a->pack_int_id - b->pack_int_id;
513 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
514 struct pack_midx_entry *e,
517 if (pos >= m->num_objects)
520 nth_midxed_object_oid(&e->oid, m, pos);
521 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
522 e->offset = nth_midxed_offset(m, pos);
524 /* consider objects in midx to be from "old" packs */
529 static void fill_pack_entry(uint32_t pack_int_id,
530 struct packed_git *p,
532 struct pack_midx_entry *entry)
534 if (!nth_packed_object_oid(&entry->oid, p, cur_object))
535 die(_("failed to locate object %d in packfile"), cur_object);
537 entry->pack_int_id = pack_int_id;
538 entry->pack_mtime = p->mtime;
540 entry->offset = nth_packed_object_offset(p, cur_object);
544 * It is possible to artificially get into a state where there are many
545 * duplicate copies of objects. That can create high memory pressure if
546 * we are to create a list of all objects before de-duplication. To reduce
547 * this memory pressure without a significant performance drop, automatically
548 * group objects by the first byte of their object id. Use the IDX fanout
549 * tables to group the data, copy to a local array, then sort.
551 * Copy only the de-duplicated entries (selected by most-recent modified time
552 * of a packfile containing the object).
554 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
555 struct pack_info *info,
557 uint32_t *nr_objects)
559 uint32_t cur_fanout, cur_pack, cur_object;
560 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
561 struct pack_midx_entry *entries_by_fanout = NULL;
562 struct pack_midx_entry *deduplicated_entries = NULL;
563 uint32_t start_pack = m ? m->num_packs : 0;
565 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
566 total_objects += info[cur_pack].p->num_objects;
569 * As we de-duplicate by fanout value, we expect the fanout
570 * slices to be evenly distributed, with some noise. Hence,
571 * allocate slightly more than one 256th.
573 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
575 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
576 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
579 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
580 uint32_t nr_fanout = 0;
583 uint32_t start = 0, end;
586 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
587 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
589 for (cur_object = start; cur_object < end; cur_object++) {
590 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
591 nth_midxed_pack_midx_entry(m,
592 &entries_by_fanout[nr_fanout],
598 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
599 uint32_t start = 0, end;
602 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
603 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
605 for (cur_object = start; cur_object < end; cur_object++) {
606 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
607 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
612 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
615 * The batch is now sorted by OID and then mtime (descending).
616 * Take only the first duplicate.
618 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
619 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
620 &entries_by_fanout[cur_object].oid))
623 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
624 memcpy(&deduplicated_entries[*nr_objects],
625 &entries_by_fanout[cur_object],
626 sizeof(struct pack_midx_entry));
631 free(entries_by_fanout);
632 return deduplicated_entries;
635 static size_t write_midx_pack_names(struct hashfile *f,
636 struct pack_info *info,
640 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
643 for (i = 0; i < num_packs; i++) {
649 if (i && strcmp(info[i].pack_name, info[i - 1].pack_name) <= 0)
650 BUG("incorrect pack-file order: %s before %s",
651 info[i - 1].pack_name,
654 writelen = strlen(info[i].pack_name) + 1;
655 hashwrite(f, info[i].pack_name, writelen);
659 /* add padding to be aligned */
660 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
661 if (i < MIDX_CHUNK_ALIGNMENT) {
662 memset(padding, 0, sizeof(padding));
663 hashwrite(f, padding, i);
670 static size_t write_midx_oid_fanout(struct hashfile *f,
671 struct pack_midx_entry *objects,
674 struct pack_midx_entry *list = objects;
675 struct pack_midx_entry *last = objects + nr_objects;
680 * Write the first-level table (the list is sorted,
681 * but we use a 256-entry lookup to be able to avoid
682 * having to do eight extra binary search iterations).
684 for (i = 0; i < 256; i++) {
685 struct pack_midx_entry *next = list;
687 while (next < last && next->oid.hash[0] == i) {
692 hashwrite_be32(f, count);
696 return MIDX_CHUNK_FANOUT_SIZE;
699 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
700 struct pack_midx_entry *objects,
703 struct pack_midx_entry *list = objects;
707 for (i = 0; i < nr_objects; i++) {
708 struct pack_midx_entry *obj = list++;
710 if (i < nr_objects - 1) {
711 struct pack_midx_entry *next = list;
712 if (oidcmp(&obj->oid, &next->oid) >= 0)
713 BUG("OIDs not in order: %s >= %s",
714 oid_to_hex(&obj->oid),
715 oid_to_hex(&next->oid));
718 hashwrite(f, obj->oid.hash, (int)hash_len);
725 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
727 struct pack_midx_entry *objects, uint32_t nr_objects)
729 struct pack_midx_entry *list = objects;
730 uint32_t i, nr_large_offset = 0;
733 for (i = 0; i < nr_objects; i++) {
734 struct pack_midx_entry *obj = list++;
736 if (perm[obj->pack_int_id] == PACK_EXPIRED)
737 BUG("object %s is in an expired pack with int-id %d",
738 oid_to_hex(&obj->oid),
741 hashwrite_be32(f, perm[obj->pack_int_id]);
743 if (large_offset_needed && obj->offset >> 31)
744 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
745 else if (!large_offset_needed && obj->offset >> 32)
746 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
747 oid_to_hex(&obj->oid),
750 hashwrite_be32(f, (uint32_t)obj->offset);
752 written += MIDX_CHUNK_OFFSET_WIDTH;
758 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
759 struct pack_midx_entry *objects, uint32_t nr_objects)
761 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
764 while (nr_large_offset) {
765 struct pack_midx_entry *obj;
769 BUG("too many large-offset objects");
772 offset = obj->offset;
777 hashwrite_be32(f, offset >> 32);
778 hashwrite_be32(f, offset & 0xffffffffUL);
779 written += 2 * sizeof(uint32_t);
787 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
788 struct string_list *packs_to_drop)
790 unsigned char cur_chunk, num_chunks = 0;
793 struct hashfile *f = NULL;
795 struct pack_list packs;
796 uint32_t *pack_perm = NULL;
797 uint64_t written = 0;
798 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
799 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
800 uint32_t nr_entries, num_large_offsets = 0;
801 struct pack_midx_entry *entries = NULL;
802 int large_offsets_needed = 0;
803 int pack_name_concat_len = 0;
804 int dropped_packs = 0;
807 midx_name = get_midx_filename(object_dir);
808 if (safe_create_leading_directories(midx_name)) {
810 die_errno(_("unable to create leading directories of %s"),
817 packs.m = load_multi_pack_index(object_dir, 1);
820 packs.alloc = packs.m ? packs.m->num_packs : 16;
822 ALLOC_ARRAY(packs.info, packs.alloc);
825 for (i = 0; i < packs.m->num_packs; i++) {
826 ALLOC_GROW(packs.info, packs.nr + 1, packs.alloc);
828 packs.info[packs.nr].orig_pack_int_id = i;
829 packs.info[packs.nr].pack_name = xstrdup(packs.m->pack_names[i]);
830 packs.info[packs.nr].p = NULL;
831 packs.info[packs.nr].expired = 0;
836 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
838 if (packs.m && packs.nr == packs.m->num_packs && !packs_to_drop)
841 entries = get_sorted_entries(packs.m, packs.info, packs.nr, &nr_entries);
843 for (i = 0; i < nr_entries; i++) {
844 if (entries[i].offset > 0x7fffffff)
846 if (entries[i].offset > 0xffffffff)
847 large_offsets_needed = 1;
850 QSORT(packs.info, packs.nr, pack_info_compare);
852 if (packs_to_drop && packs_to_drop->nr) {
854 int missing_drops = 0;
856 for (i = 0; i < packs.nr && drop_index < packs_to_drop->nr; i++) {
857 int cmp = strcmp(packs.info[i].pack_name,
858 packs_to_drop->items[drop_index].string);
862 packs.info[i].expired = 1;
863 } else if (cmp > 0) {
864 error(_("did not see pack-file %s to drop"),
865 packs_to_drop->items[drop_index].string);
870 packs.info[i].expired = 0;
881 * pack_perm stores a permutation between pack-int-ids from the
882 * previous multi-pack-index to the new one we are writing:
884 * pack_perm[old_id] = new_id
886 ALLOC_ARRAY(pack_perm, packs.nr);
887 for (i = 0; i < packs.nr; i++) {
888 if (packs.info[i].expired) {
890 pack_perm[packs.info[i].orig_pack_int_id] = PACK_EXPIRED;
892 pack_perm[packs.info[i].orig_pack_int_id] = i - dropped_packs;
896 for (i = 0; i < packs.nr; i++) {
897 if (!packs.info[i].expired)
898 pack_name_concat_len += strlen(packs.info[i].pack_name) + 1;
901 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
902 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
903 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
905 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
906 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
907 FREE_AND_NULL(midx_name);
913 num_chunks = large_offsets_needed ? 5 : 4;
915 written = write_midx_header(f, num_chunks, packs.nr - dropped_packs);
917 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
918 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
921 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
922 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + pack_name_concat_len;
925 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
926 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
929 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
930 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * the_hash_algo->rawsz;
933 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
934 if (large_offsets_needed) {
935 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
938 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
939 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
942 chunk_ids[cur_chunk] = 0;
944 for (i = 0; i <= num_chunks; i++) {
945 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
946 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
947 chunk_offsets[i - 1],
950 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
951 BUG("chunk offset %"PRIu64" is not properly aligned",
954 hashwrite_be32(f, chunk_ids[i]);
955 hashwrite_be32(f, chunk_offsets[i] >> 32);
956 hashwrite_be32(f, chunk_offsets[i]);
958 written += MIDX_CHUNKLOOKUP_WIDTH;
961 for (i = 0; i < num_chunks; i++) {
962 if (written != chunk_offsets[i])
963 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
968 switch (chunk_ids[i]) {
969 case MIDX_CHUNKID_PACKNAMES:
970 written += write_midx_pack_names(f, packs.info, packs.nr);
973 case MIDX_CHUNKID_OIDFANOUT:
974 written += write_midx_oid_fanout(f, entries, nr_entries);
977 case MIDX_CHUNKID_OIDLOOKUP:
978 written += write_midx_oid_lookup(f, the_hash_algo->rawsz, entries, nr_entries);
981 case MIDX_CHUNKID_OBJECTOFFSETS:
982 written += write_midx_object_offsets(f, large_offsets_needed, pack_perm, entries, nr_entries);
985 case MIDX_CHUNKID_LARGEOFFSETS:
986 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
990 BUG("trying to write unknown chunk id %"PRIx32,
995 if (written != chunk_offsets[num_chunks])
996 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
998 chunk_offsets[num_chunks]);
1000 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1001 commit_lock_file(&lk);
1004 for (i = 0; i < packs.nr; i++) {
1005 if (packs.info[i].p) {
1006 close_pack(packs.info[i].p);
1007 free(packs.info[i].p);
1009 free(packs.info[i].pack_name);
1019 int write_midx_file(const char *object_dir)
1021 return write_midx_internal(object_dir, NULL, NULL);
1024 void clear_midx_file(struct repository *r)
1026 char *midx = get_midx_filename(r->objects->odb->path);
1028 if (r->objects && r->objects->multi_pack_index) {
1029 close_midx(r->objects->multi_pack_index);
1030 r->objects->multi_pack_index = NULL;
1033 if (remove_path(midx)) {
1035 die(_("failed to clear multi-pack-index at %s"), midx);
1041 static int verify_midx_error;
1043 static void midx_report(const char *fmt, ...)
1046 verify_midx_error = 1;
1048 vfprintf(stderr, fmt, ap);
1049 fprintf(stderr, "\n");
1053 struct pair_pos_vs_id
1056 uint32_t pack_int_id;
1059 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1061 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1062 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1064 return b->pack_int_id - a->pack_int_id;
1068 * Limit calls to display_progress() for performance reasons.
1069 * The interval here was arbitrarily chosen.
1071 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1072 #define midx_display_sparse_progress(progress, n) \
1074 uint64_t _n = (n); \
1075 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1076 display_progress(progress, _n); \
1079 int verify_midx_file(struct repository *r, const char *object_dir)
1081 struct pair_pos_vs_id *pairs = NULL;
1083 struct progress *progress;
1084 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1085 verify_midx_error = 0;
1090 progress = start_progress(_("Looking for referenced packfiles"),
1092 for (i = 0; i < m->num_packs; i++) {
1093 if (prepare_midx_pack(r, m, i))
1094 midx_report("failed to load pack in position %d", i);
1096 display_progress(progress, i + 1);
1098 stop_progress(&progress);
1100 for (i = 0; i < 255; i++) {
1101 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1102 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1104 if (oid_fanout1 > oid_fanout2)
1105 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1106 i, oid_fanout1, oid_fanout2, i + 1);
1109 progress = start_sparse_progress(_("Verifying OID order in MIDX"),
1110 m->num_objects - 1);
1111 for (i = 0; i < m->num_objects - 1; i++) {
1112 struct object_id oid1, oid2;
1114 nth_midxed_object_oid(&oid1, m, i);
1115 nth_midxed_object_oid(&oid2, m, i + 1);
1117 if (oidcmp(&oid1, &oid2) >= 0)
1118 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1119 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1121 midx_display_sparse_progress(progress, i + 1);
1123 stop_progress(&progress);
1126 * Create an array mapping each object to its packfile id. Sort it
1127 * to group the objects by packfile. Use this permutation to visit
1128 * each of the objects and only require 1 packfile to be open at a
1131 ALLOC_ARRAY(pairs, m->num_objects);
1132 for (i = 0; i < m->num_objects; i++) {
1134 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1137 progress = start_sparse_progress(_("Sorting objects by packfile"),
1139 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1140 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1141 stop_progress(&progress);
1143 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1144 for (i = 0; i < m->num_objects; i++) {
1145 struct object_id oid;
1146 struct pack_entry e;
1147 off_t m_offset, p_offset;
1149 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1150 m->packs[pairs[i-1].pack_int_id])
1152 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1153 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1156 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1158 if (!fill_midx_entry(r, &oid, &e, m)) {
1159 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1160 pairs[i].pos, oid_to_hex(&oid));
1164 if (open_pack_index(e.p)) {
1165 midx_report(_("failed to load pack-index for packfile %s"),
1170 m_offset = e.offset;
1171 p_offset = find_pack_entry_one(oid.hash, e.p);
1173 if (m_offset != p_offset)
1174 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1175 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1177 midx_display_sparse_progress(progress, i + 1);
1179 stop_progress(&progress);
1183 return verify_midx_error;
1186 int expire_midx_packs(struct repository *r, const char *object_dir)
1188 uint32_t i, *count, result = 0;
1189 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1190 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1195 count = xcalloc(m->num_packs, sizeof(uint32_t));
1196 for (i = 0; i < m->num_objects; i++) {
1197 int pack_int_id = nth_midxed_pack_int_id(m, i);
1198 count[pack_int_id]++;
1201 for (i = 0; i < m->num_packs; i++) {
1207 if (prepare_midx_pack(r, m, i))
1210 if (m->packs[i]->pack_keep)
1213 pack_name = xstrdup(m->packs[i]->pack_name);
1214 close_pack(m->packs[i]);
1216 string_list_insert(&packs_to_drop, m->pack_names[i]);
1217 unlink_pack_path(pack_name, 0);
1223 if (packs_to_drop.nr)
1224 result = write_midx_internal(object_dir, m, &packs_to_drop);
1226 string_list_clear(&packs_to_drop, 0);
1230 struct repack_info {
1232 uint32_t referenced_objects;
1233 uint32_t pack_int_id;
1236 static int compare_by_mtime(const void *a_, const void *b_)
1238 const struct repack_info *a, *b;
1240 a = (const struct repack_info *)a_;
1241 b = (const struct repack_info *)b_;
1243 if (a->mtime < b->mtime)
1245 if (a->mtime > b->mtime)
1250 static int fill_included_packs_all(struct multi_pack_index *m,
1251 unsigned char *include_pack)
1255 for (i = 0; i < m->num_packs; i++)
1256 include_pack[i] = 1;
1258 return m->num_packs < 2;
1261 static int fill_included_packs_batch(struct repository *r,
1262 struct multi_pack_index *m,
1263 unsigned char *include_pack,
1266 uint32_t i, packs_to_repack;
1268 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1270 for (i = 0; i < m->num_packs; i++) {
1271 pack_info[i].pack_int_id = i;
1273 if (prepare_midx_pack(r, m, i))
1276 pack_info[i].mtime = m->packs[i]->mtime;
1279 for (i = 0; batch_size && i < m->num_objects; i++) {
1280 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1281 pack_info[pack_int_id].referenced_objects++;
1284 QSORT(pack_info, m->num_packs, compare_by_mtime);
1287 packs_to_repack = 0;
1288 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1289 int pack_int_id = pack_info[i].pack_int_id;
1290 struct packed_git *p = m->packs[pack_int_id];
1291 size_t expected_size;
1295 if (open_pack_index(p) || !p->num_objects)
1298 expected_size = (size_t)(p->pack_size
1299 * pack_info[i].referenced_objects);
1300 expected_size /= p->num_objects;
1302 if (expected_size >= batch_size)
1306 total_size += expected_size;
1307 include_pack[pack_int_id] = 1;
1312 if (total_size < batch_size || packs_to_repack < 2)
1318 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size)
1322 unsigned char *include_pack;
1323 struct child_process cmd = CHILD_PROCESS_INIT;
1324 struct strbuf base_name = STRBUF_INIT;
1325 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1330 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1333 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1335 } else if (fill_included_packs_all(m, include_pack))
1338 argv_array_push(&cmd.args, "pack-objects");
1340 strbuf_addstr(&base_name, object_dir);
1341 strbuf_addstr(&base_name, "/pack/pack");
1342 argv_array_push(&cmd.args, base_name.buf);
1343 strbuf_release(&base_name);
1346 cmd.in = cmd.out = -1;
1348 if (start_command(&cmd)) {
1349 error(_("could not start pack-objects"));
1354 for (i = 0; i < m->num_objects; i++) {
1355 struct object_id oid;
1356 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1358 if (!include_pack[pack_int_id])
1361 nth_midxed_object_oid(&oid, m, i);
1362 xwrite(cmd.in, oid_to_hex(&oid), the_hash_algo->hexsz);
1363 xwrite(cmd.in, "\n", 1);
1367 if (finish_command(&cmd)) {
1368 error(_("could not finish pack-objects"));
1373 result = write_midx_internal(object_dir, m, NULL);