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_HEADER_SIZE 12
21 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
23 #define MIDX_MAX_CHUNKS 5
24 #define MIDX_CHUNK_ALIGNMENT 4
25 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
26 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
27 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
28 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
29 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
30 #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
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 char *get_midx_filename(const char *object_dir)
52 return xstrfmt("%s/pack/multi-pack-index", object_dir);
55 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
57 struct multi_pack_index *m = NULL;
61 void *midx_map = NULL;
62 uint32_t hash_version;
63 char *midx_name = get_midx_filename(object_dir);
65 const char *cur_pack_name;
67 fd = git_open(midx_name);
72 error_errno(_("failed to read %s"), midx_name);
76 midx_size = xsize_t(st.st_size);
78 if (midx_size < MIDX_MIN_SIZE) {
79 error(_("multi-pack-index file %s is too small"), midx_name);
83 FREE_AND_NULL(midx_name);
85 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
88 FLEX_ALLOC_STR(m, object_dir, object_dir);
90 m->data_len = midx_size;
93 m->signature = get_be32(m->data);
94 if (m->signature != MIDX_SIGNATURE)
95 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
96 m->signature, MIDX_SIGNATURE);
98 m->version = m->data[MIDX_BYTE_FILE_VERSION];
99 if (m->version != MIDX_VERSION)
100 die(_("multi-pack-index version %d not recognized"),
103 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
104 if (hash_version != oid_version()) {
105 error(_("multi-pack-index hash version %u does not match version %u"),
106 hash_version, oid_version());
109 m->hash_len = the_hash_algo->rawsz;
111 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
113 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
115 for (i = 0; i < m->num_chunks; i++) {
116 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
117 MIDX_CHUNKLOOKUP_WIDTH * i);
118 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
119 MIDX_CHUNKLOOKUP_WIDTH * i);
121 if (chunk_offset >= m->data_len)
122 die(_("invalid chunk offset (too large)"));
125 case MIDX_CHUNKID_PACKNAMES:
126 m->chunk_pack_names = m->data + chunk_offset;
129 case MIDX_CHUNKID_OIDFANOUT:
130 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
133 case MIDX_CHUNKID_OIDLOOKUP:
134 m->chunk_oid_lookup = m->data + chunk_offset;
137 case MIDX_CHUNKID_OBJECTOFFSETS:
138 m->chunk_object_offsets = m->data + chunk_offset;
141 case MIDX_CHUNKID_LARGEOFFSETS:
142 m->chunk_large_offsets = m->data + chunk_offset;
146 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
151 * Do nothing on unrecognized chunks, allowing future
152 * extensions to add optional chunks.
158 if (!m->chunk_pack_names)
159 die(_("multi-pack-index missing required pack-name chunk"));
160 if (!m->chunk_oid_fanout)
161 die(_("multi-pack-index missing required OID fanout chunk"));
162 if (!m->chunk_oid_lookup)
163 die(_("multi-pack-index missing required OID lookup chunk"));
164 if (!m->chunk_object_offsets)
165 die(_("multi-pack-index missing required object offsets chunk"));
167 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
169 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
170 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
172 cur_pack_name = (const char *)m->chunk_pack_names;
173 for (i = 0; i < m->num_packs; i++) {
174 m->pack_names[i] = cur_pack_name;
176 cur_pack_name += strlen(cur_pack_name) + 1;
178 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
179 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
180 m->pack_names[i - 1],
184 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
185 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
193 munmap(midx_map, midx_size);
199 void close_midx(struct multi_pack_index *m)
206 munmap((unsigned char *)m->data, m->data_len);
208 for (i = 0; i < m->num_packs; i++) {
210 m->packs[i]->multi_pack_index = 0;
212 FREE_AND_NULL(m->packs);
213 FREE_AND_NULL(m->pack_names);
216 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
218 struct strbuf pack_name = STRBUF_INIT;
219 struct packed_git *p;
221 if (pack_int_id >= m->num_packs)
222 die(_("bad pack-int-id: %u (%u total packs)"),
223 pack_int_id, m->num_packs);
225 if (m->packs[pack_int_id])
228 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
229 m->pack_names[pack_int_id]);
231 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
232 strbuf_release(&pack_name);
237 p->multi_pack_index = 1;
238 m->packs[pack_int_id] = p;
239 install_packed_git(r, p);
240 list_add_tail(&p->mru, &r->objects->packed_git_mru);
245 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
247 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
248 the_hash_algo->rawsz, result);
251 struct object_id *nth_midxed_object_oid(struct object_id *oid,
252 struct multi_pack_index *m,
255 if (n >= m->num_objects)
258 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
262 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
264 const unsigned char *offset_data;
267 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
268 offset32 = get_be32(offset_data + sizeof(uint32_t));
270 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
271 if (sizeof(off_t) < sizeof(uint64_t))
272 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
274 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
275 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
281 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
283 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
286 static int nth_midxed_pack_entry(struct repository *r,
287 struct multi_pack_index *m,
288 struct pack_entry *e,
291 uint32_t pack_int_id;
292 struct packed_git *p;
294 if (pos >= m->num_objects)
297 pack_int_id = nth_midxed_pack_int_id(m, pos);
299 if (prepare_midx_pack(r, m, pack_int_id))
300 die(_("error preparing packfile from multi-pack-index"));
301 p = m->packs[pack_int_id];
304 * We are about to tell the caller where they can locate the
305 * requested object. We better make sure the packfile is
306 * still here and can be accessed before supplying that
307 * answer, as it may have been deleted since the MIDX was
310 if (!is_pack_valid(p))
313 if (p->num_bad_objects) {
315 struct object_id oid;
316 nth_midxed_object_oid(&oid, m, pos);
317 for (i = 0; i < p->num_bad_objects; i++)
319 p->bad_object_sha1 + the_hash_algo->rawsz * i))
323 e->offset = nth_midxed_offset(m, pos);
329 int fill_midx_entry(struct repository * r,
330 const struct object_id *oid,
331 struct pack_entry *e,
332 struct multi_pack_index *m)
336 if (!bsearch_midx(oid, m, &pos))
339 return nth_midxed_pack_entry(r, m, e, pos);
342 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
343 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
344 const char *idx_name)
346 /* Skip past any initial matching prefix. */
347 while (*idx_name && *idx_name == *idx_or_pack_name) {
353 * If we didn't match completely, we may have matched "pack-1234." and
354 * be left with "idx" and "pack" respectively, which is also OK. We do
355 * not have to check for "idx" and "idx", because that would have been
356 * a complete match (and in that case these strcmps will be false, but
357 * we'll correctly return 0 from the final strcmp() below.
359 * Technically this matches "fooidx" and "foopack", but we'd never have
360 * such names in the first place.
362 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
366 * This not only checks for a complete match, but also orders based on
367 * the first non-identical character, which means our ordering will
368 * match a raw strcmp(). That makes it OK to use this to binary search
369 * a naively-sorted list.
371 return strcmp(idx_or_pack_name, idx_name);
374 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
376 uint32_t first = 0, last = m->num_packs;
378 while (first < last) {
379 uint32_t mid = first + (last - first) / 2;
383 current = m->pack_names[mid];
384 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
397 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
399 struct multi_pack_index *m;
400 struct multi_pack_index *m_search;
402 static int env_value = -1;
405 env_value = git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0);
408 (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
412 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
413 if (!strcmp(object_dir, m_search->object_dir))
416 m = load_multi_pack_index(object_dir, local);
419 struct multi_pack_index *mp = r->objects->multi_pack_index;
424 r->objects->multi_pack_index = m;
431 static size_t write_midx_header(struct hashfile *f,
432 unsigned char num_chunks,
435 unsigned char byte_values[4];
437 hashwrite_be32(f, MIDX_SIGNATURE);
438 byte_values[0] = MIDX_VERSION;
439 byte_values[1] = oid_version();
440 byte_values[2] = num_chunks;
441 byte_values[3] = 0; /* unused */
442 hashwrite(f, byte_values, sizeof(byte_values));
443 hashwrite_be32(f, num_packs);
445 return MIDX_HEADER_SIZE;
449 uint32_t orig_pack_int_id;
451 struct packed_git *p;
452 unsigned expired : 1;
455 static int pack_info_compare(const void *_a, const void *_b)
457 struct pack_info *a = (struct pack_info *)_a;
458 struct pack_info *b = (struct pack_info *)_b;
459 return strcmp(a->pack_name, b->pack_name);
463 struct pack_info *info;
466 struct multi_pack_index *m;
467 struct progress *progress;
468 unsigned pack_paths_checked;
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 pack_list *packs = (struct pack_list *)data;
476 if (ends_with(file_name, ".idx")) {
477 display_progress(packs->progress, ++packs->pack_paths_checked);
478 if (packs->m && midx_contains_pack(packs->m, file_name))
481 ALLOC_GROW(packs->info, packs->nr + 1, packs->alloc);
483 packs->info[packs->nr].p = add_packed_git(full_path,
487 if (!packs->info[packs->nr].p) {
488 warning(_("failed to add packfile '%s'"),
493 if (open_pack_index(packs->info[packs->nr].p)) {
494 warning(_("failed to open pack-index '%s'"),
496 close_pack(packs->info[packs->nr].p);
497 FREE_AND_NULL(packs->info[packs->nr].p);
501 packs->info[packs->nr].pack_name = xstrdup(file_name);
502 packs->info[packs->nr].orig_pack_int_id = packs->nr;
503 packs->info[packs->nr].expired = 0;
508 struct pack_midx_entry {
509 struct object_id oid;
510 uint32_t pack_int_id;
515 static int midx_oid_compare(const void *_a, const void *_b)
517 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
518 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
519 int cmp = oidcmp(&a->oid, &b->oid);
524 if (a->pack_mtime > b->pack_mtime)
526 else if (a->pack_mtime < b->pack_mtime)
529 return a->pack_int_id - b->pack_int_id;
532 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
533 struct pack_midx_entry *e,
536 if (pos >= m->num_objects)
539 nth_midxed_object_oid(&e->oid, m, pos);
540 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
541 e->offset = nth_midxed_offset(m, pos);
543 /* consider objects in midx to be from "old" packs */
548 static void fill_pack_entry(uint32_t pack_int_id,
549 struct packed_git *p,
551 struct pack_midx_entry *entry)
553 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
554 die(_("failed to locate object %d in packfile"), cur_object);
556 entry->pack_int_id = pack_int_id;
557 entry->pack_mtime = p->mtime;
559 entry->offset = nth_packed_object_offset(p, cur_object);
563 * It is possible to artificially get into a state where there are many
564 * duplicate copies of objects. That can create high memory pressure if
565 * we are to create a list of all objects before de-duplication. To reduce
566 * this memory pressure without a significant performance drop, automatically
567 * group objects by the first byte of their object id. Use the IDX fanout
568 * tables to group the data, copy to a local array, then sort.
570 * Copy only the de-duplicated entries (selected by most-recent modified time
571 * of a packfile containing the object).
573 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
574 struct pack_info *info,
576 uint32_t *nr_objects)
578 uint32_t cur_fanout, cur_pack, cur_object;
579 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
580 struct pack_midx_entry *entries_by_fanout = NULL;
581 struct pack_midx_entry *deduplicated_entries = NULL;
582 uint32_t start_pack = m ? m->num_packs : 0;
584 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
585 total_objects += info[cur_pack].p->num_objects;
588 * As we de-duplicate by fanout value, we expect the fanout
589 * slices to be evenly distributed, with some noise. Hence,
590 * allocate slightly more than one 256th.
592 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
594 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
595 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
598 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
599 uint32_t nr_fanout = 0;
602 uint32_t start = 0, end;
605 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
606 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
608 for (cur_object = start; cur_object < end; cur_object++) {
609 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
610 nth_midxed_pack_midx_entry(m,
611 &entries_by_fanout[nr_fanout],
617 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
618 uint32_t start = 0, end;
621 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
622 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
624 for (cur_object = start; cur_object < end; cur_object++) {
625 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
626 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
631 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
634 * The batch is now sorted by OID and then mtime (descending).
635 * Take only the first duplicate.
637 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
638 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
639 &entries_by_fanout[cur_object].oid))
642 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
643 memcpy(&deduplicated_entries[*nr_objects],
644 &entries_by_fanout[cur_object],
645 sizeof(struct pack_midx_entry));
650 free(entries_by_fanout);
651 return deduplicated_entries;
654 static size_t write_midx_pack_names(struct hashfile *f,
655 struct pack_info *info,
659 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
662 for (i = 0; i < num_packs; i++) {
668 if (i && strcmp(info[i].pack_name, info[i - 1].pack_name) <= 0)
669 BUG("incorrect pack-file order: %s before %s",
670 info[i - 1].pack_name,
673 writelen = strlen(info[i].pack_name) + 1;
674 hashwrite(f, info[i].pack_name, writelen);
678 /* add padding to be aligned */
679 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
680 if (i < MIDX_CHUNK_ALIGNMENT) {
681 memset(padding, 0, sizeof(padding));
682 hashwrite(f, padding, i);
689 static size_t write_midx_oid_fanout(struct hashfile *f,
690 struct pack_midx_entry *objects,
693 struct pack_midx_entry *list = objects;
694 struct pack_midx_entry *last = objects + nr_objects;
699 * Write the first-level table (the list is sorted,
700 * but we use a 256-entry lookup to be able to avoid
701 * having to do eight extra binary search iterations).
703 for (i = 0; i < 256; i++) {
704 struct pack_midx_entry *next = list;
706 while (next < last && next->oid.hash[0] == i) {
711 hashwrite_be32(f, count);
715 return MIDX_CHUNK_FANOUT_SIZE;
718 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
719 struct pack_midx_entry *objects,
722 struct pack_midx_entry *list = objects;
726 for (i = 0; i < nr_objects; i++) {
727 struct pack_midx_entry *obj = list++;
729 if (i < nr_objects - 1) {
730 struct pack_midx_entry *next = list;
731 if (oidcmp(&obj->oid, &next->oid) >= 0)
732 BUG("OIDs not in order: %s >= %s",
733 oid_to_hex(&obj->oid),
734 oid_to_hex(&next->oid));
737 hashwrite(f, obj->oid.hash, (int)hash_len);
744 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
746 struct pack_midx_entry *objects, uint32_t nr_objects)
748 struct pack_midx_entry *list = objects;
749 uint32_t i, nr_large_offset = 0;
752 for (i = 0; i < nr_objects; i++) {
753 struct pack_midx_entry *obj = list++;
755 if (perm[obj->pack_int_id] == PACK_EXPIRED)
756 BUG("object %s is in an expired pack with int-id %d",
757 oid_to_hex(&obj->oid),
760 hashwrite_be32(f, perm[obj->pack_int_id]);
762 if (large_offset_needed && obj->offset >> 31)
763 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
764 else if (!large_offset_needed && obj->offset >> 32)
765 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
766 oid_to_hex(&obj->oid),
769 hashwrite_be32(f, (uint32_t)obj->offset);
771 written += MIDX_CHUNK_OFFSET_WIDTH;
777 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
778 struct pack_midx_entry *objects, uint32_t nr_objects)
780 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
783 while (nr_large_offset) {
784 struct pack_midx_entry *obj;
788 BUG("too many large-offset objects");
791 offset = obj->offset;
796 hashwrite_be32(f, offset >> 32);
797 hashwrite_be32(f, offset & 0xffffffffUL);
798 written += 2 * sizeof(uint32_t);
806 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
807 struct string_list *packs_to_drop, unsigned flags)
809 unsigned char cur_chunk, num_chunks = 0;
812 struct hashfile *f = NULL;
814 struct pack_list packs;
815 uint32_t *pack_perm = NULL;
816 uint64_t written = 0;
817 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
818 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
819 uint32_t nr_entries, num_large_offsets = 0;
820 struct pack_midx_entry *entries = NULL;
821 struct progress *progress = NULL;
822 int large_offsets_needed = 0;
823 int pack_name_concat_len = 0;
824 int dropped_packs = 0;
827 midx_name = get_midx_filename(object_dir);
828 if (safe_create_leading_directories(midx_name))
829 die_errno(_("unable to create leading directories of %s"),
835 packs.m = load_multi_pack_index(object_dir, 1);
838 packs.alloc = packs.m ? packs.m->num_packs : 16;
840 ALLOC_ARRAY(packs.info, packs.alloc);
843 for (i = 0; i < packs.m->num_packs; i++) {
844 ALLOC_GROW(packs.info, packs.nr + 1, packs.alloc);
846 packs.info[packs.nr].orig_pack_int_id = i;
847 packs.info[packs.nr].pack_name = xstrdup(packs.m->pack_names[i]);
848 packs.info[packs.nr].p = NULL;
849 packs.info[packs.nr].expired = 0;
854 packs.pack_paths_checked = 0;
855 if (flags & MIDX_PROGRESS)
856 packs.progress = start_progress(_("Adding packfiles to multi-pack-index"), 0);
858 packs.progress = NULL;
860 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
861 stop_progress(&packs.progress);
863 if (packs.m && packs.nr == packs.m->num_packs && !packs_to_drop)
866 entries = get_sorted_entries(packs.m, packs.info, packs.nr, &nr_entries);
868 for (i = 0; i < nr_entries; i++) {
869 if (entries[i].offset > 0x7fffffff)
871 if (entries[i].offset > 0xffffffff)
872 large_offsets_needed = 1;
875 QSORT(packs.info, packs.nr, pack_info_compare);
877 if (packs_to_drop && packs_to_drop->nr) {
879 int missing_drops = 0;
881 for (i = 0; i < packs.nr && drop_index < packs_to_drop->nr; i++) {
882 int cmp = strcmp(packs.info[i].pack_name,
883 packs_to_drop->items[drop_index].string);
887 packs.info[i].expired = 1;
888 } else if (cmp > 0) {
889 error(_("did not see pack-file %s to drop"),
890 packs_to_drop->items[drop_index].string);
895 packs.info[i].expired = 0;
906 * pack_perm stores a permutation between pack-int-ids from the
907 * previous multi-pack-index to the new one we are writing:
909 * pack_perm[old_id] = new_id
911 ALLOC_ARRAY(pack_perm, packs.nr);
912 for (i = 0; i < packs.nr; i++) {
913 if (packs.info[i].expired) {
915 pack_perm[packs.info[i].orig_pack_int_id] = PACK_EXPIRED;
917 pack_perm[packs.info[i].orig_pack_int_id] = i - dropped_packs;
921 for (i = 0; i < packs.nr; i++) {
922 if (!packs.info[i].expired)
923 pack_name_concat_len += strlen(packs.info[i].pack_name) + 1;
926 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
927 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
928 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
930 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
931 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
932 FREE_AND_NULL(midx_name);
938 num_chunks = large_offsets_needed ? 5 : 4;
940 if (packs.nr - dropped_packs == 0) {
941 error(_("no pack files to index."));
946 written = write_midx_header(f, num_chunks, packs.nr - dropped_packs);
948 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
949 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
952 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
953 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + pack_name_concat_len;
956 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
957 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
960 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
961 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * the_hash_algo->rawsz;
964 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
965 if (large_offsets_needed) {
966 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
969 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
970 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
973 chunk_ids[cur_chunk] = 0;
975 for (i = 0; i <= num_chunks; i++) {
976 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
977 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
978 chunk_offsets[i - 1],
981 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
982 BUG("chunk offset %"PRIu64" is not properly aligned",
985 hashwrite_be32(f, chunk_ids[i]);
986 hashwrite_be32(f, chunk_offsets[i] >> 32);
987 hashwrite_be32(f, chunk_offsets[i]);
989 written += MIDX_CHUNKLOOKUP_WIDTH;
992 if (flags & MIDX_PROGRESS)
993 progress = start_progress(_("Writing chunks to multi-pack-index"),
995 for (i = 0; i < num_chunks; i++) {
996 if (written != chunk_offsets[i])
997 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
1002 switch (chunk_ids[i]) {
1003 case MIDX_CHUNKID_PACKNAMES:
1004 written += write_midx_pack_names(f, packs.info, packs.nr);
1007 case MIDX_CHUNKID_OIDFANOUT:
1008 written += write_midx_oid_fanout(f, entries, nr_entries);
1011 case MIDX_CHUNKID_OIDLOOKUP:
1012 written += write_midx_oid_lookup(f, the_hash_algo->rawsz, entries, nr_entries);
1015 case MIDX_CHUNKID_OBJECTOFFSETS:
1016 written += write_midx_object_offsets(f, large_offsets_needed, pack_perm, entries, nr_entries);
1019 case MIDX_CHUNKID_LARGEOFFSETS:
1020 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
1024 BUG("trying to write unknown chunk id %"PRIx32,
1028 display_progress(progress, i + 1);
1030 stop_progress(&progress);
1032 if (written != chunk_offsets[num_chunks])
1033 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
1035 chunk_offsets[num_chunks]);
1037 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1038 commit_lock_file(&lk);
1041 for (i = 0; i < packs.nr; i++) {
1042 if (packs.info[i].p) {
1043 close_pack(packs.info[i].p);
1044 free(packs.info[i].p);
1046 free(packs.info[i].pack_name);
1056 int write_midx_file(const char *object_dir, unsigned flags)
1058 return write_midx_internal(object_dir, NULL, NULL, flags);
1061 void clear_midx_file(struct repository *r)
1063 char *midx = get_midx_filename(r->objects->odb->path);
1065 if (r->objects && r->objects->multi_pack_index) {
1066 close_midx(r->objects->multi_pack_index);
1067 r->objects->multi_pack_index = NULL;
1070 if (remove_path(midx))
1071 die(_("failed to clear multi-pack-index at %s"), midx);
1076 static int verify_midx_error;
1078 static void midx_report(const char *fmt, ...)
1081 verify_midx_error = 1;
1083 vfprintf(stderr, fmt, ap);
1084 fprintf(stderr, "\n");
1088 struct pair_pos_vs_id
1091 uint32_t pack_int_id;
1094 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1096 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1097 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1099 return b->pack_int_id - a->pack_int_id;
1103 * Limit calls to display_progress() for performance reasons.
1104 * The interval here was arbitrarily chosen.
1106 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1107 #define midx_display_sparse_progress(progress, n) \
1109 uint64_t _n = (n); \
1110 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1111 display_progress(progress, _n); \
1114 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1116 struct pair_pos_vs_id *pairs = NULL;
1118 struct progress *progress = NULL;
1119 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1120 verify_midx_error = 0;
1125 char *filename = get_midx_filename(object_dir);
1126 if (!stat(filename, &sb)) {
1127 error(_("multi-pack-index file exists, but failed to parse"));
1134 if (flags & MIDX_PROGRESS)
1135 progress = start_progress(_("Looking for referenced packfiles"),
1137 for (i = 0; i < m->num_packs; i++) {
1138 if (prepare_midx_pack(r, m, i))
1139 midx_report("failed to load pack in position %d", i);
1141 display_progress(progress, i + 1);
1143 stop_progress(&progress);
1145 for (i = 0; i < 255; i++) {
1146 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1147 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1149 if (oid_fanout1 > oid_fanout2)
1150 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1151 i, oid_fanout1, oid_fanout2, i + 1);
1154 if (m->num_objects == 0) {
1155 midx_report(_("the midx contains no oid"));
1157 * Remaining tests assume that we have objects, so we can
1160 return verify_midx_error;
1163 if (flags & MIDX_PROGRESS)
1164 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1165 m->num_objects - 1);
1166 for (i = 0; i < m->num_objects - 1; i++) {
1167 struct object_id oid1, oid2;
1169 nth_midxed_object_oid(&oid1, m, i);
1170 nth_midxed_object_oid(&oid2, m, i + 1);
1172 if (oidcmp(&oid1, &oid2) >= 0)
1173 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1174 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1176 midx_display_sparse_progress(progress, i + 1);
1178 stop_progress(&progress);
1181 * Create an array mapping each object to its packfile id. Sort it
1182 * to group the objects by packfile. Use this permutation to visit
1183 * each of the objects and only require 1 packfile to be open at a
1186 ALLOC_ARRAY(pairs, m->num_objects);
1187 for (i = 0; i < m->num_objects; i++) {
1189 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1192 if (flags & MIDX_PROGRESS)
1193 progress = start_sparse_progress(_("Sorting objects by packfile"),
1195 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1196 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1197 stop_progress(&progress);
1199 if (flags & MIDX_PROGRESS)
1200 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1201 for (i = 0; i < m->num_objects; i++) {
1202 struct object_id oid;
1203 struct pack_entry e;
1204 off_t m_offset, p_offset;
1206 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1207 m->packs[pairs[i-1].pack_int_id])
1209 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1210 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1213 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1215 if (!fill_midx_entry(r, &oid, &e, m)) {
1216 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1217 pairs[i].pos, oid_to_hex(&oid));
1221 if (open_pack_index(e.p)) {
1222 midx_report(_("failed to load pack-index for packfile %s"),
1227 m_offset = e.offset;
1228 p_offset = find_pack_entry_one(oid.hash, e.p);
1230 if (m_offset != p_offset)
1231 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1232 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1234 midx_display_sparse_progress(progress, i + 1);
1236 stop_progress(&progress);
1240 return verify_midx_error;
1243 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1245 uint32_t i, *count, result = 0;
1246 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1247 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1248 struct progress *progress = NULL;
1253 count = xcalloc(m->num_packs, sizeof(uint32_t));
1255 if (flags & MIDX_PROGRESS)
1256 progress = start_progress(_("Counting referenced objects"),
1258 for (i = 0; i < m->num_objects; i++) {
1259 int pack_int_id = nth_midxed_pack_int_id(m, i);
1260 count[pack_int_id]++;
1261 display_progress(progress, i + 1);
1263 stop_progress(&progress);
1265 if (flags & MIDX_PROGRESS)
1266 progress = start_progress(_("Finding and deleting unreferenced packfiles"),
1268 for (i = 0; i < m->num_packs; i++) {
1270 display_progress(progress, i + 1);
1275 if (prepare_midx_pack(r, m, i))
1278 if (m->packs[i]->pack_keep)
1281 pack_name = xstrdup(m->packs[i]->pack_name);
1282 close_pack(m->packs[i]);
1284 string_list_insert(&packs_to_drop, m->pack_names[i]);
1285 unlink_pack_path(pack_name, 0);
1288 stop_progress(&progress);
1292 if (packs_to_drop.nr)
1293 result = write_midx_internal(object_dir, m, &packs_to_drop, flags);
1295 string_list_clear(&packs_to_drop, 0);
1299 struct repack_info {
1301 uint32_t referenced_objects;
1302 uint32_t pack_int_id;
1305 static int compare_by_mtime(const void *a_, const void *b_)
1307 const struct repack_info *a, *b;
1309 a = (const struct repack_info *)a_;
1310 b = (const struct repack_info *)b_;
1312 if (a->mtime < b->mtime)
1314 if (a->mtime > b->mtime)
1319 static int fill_included_packs_all(struct repository *r,
1320 struct multi_pack_index *m,
1321 unsigned char *include_pack)
1323 uint32_t i, count = 0;
1324 int pack_kept_objects = 0;
1326 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1328 for (i = 0; i < m->num_packs; i++) {
1329 if (prepare_midx_pack(r, m, i))
1331 if (!pack_kept_objects && m->packs[i]->pack_keep)
1334 include_pack[i] = 1;
1341 static int fill_included_packs_batch(struct repository *r,
1342 struct multi_pack_index *m,
1343 unsigned char *include_pack,
1346 uint32_t i, packs_to_repack;
1348 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1349 int pack_kept_objects = 0;
1351 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1353 for (i = 0; i < m->num_packs; i++) {
1354 pack_info[i].pack_int_id = i;
1356 if (prepare_midx_pack(r, m, i))
1359 pack_info[i].mtime = m->packs[i]->mtime;
1362 for (i = 0; batch_size && i < m->num_objects; i++) {
1363 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1364 pack_info[pack_int_id].referenced_objects++;
1367 QSORT(pack_info, m->num_packs, compare_by_mtime);
1370 packs_to_repack = 0;
1371 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1372 int pack_int_id = pack_info[i].pack_int_id;
1373 struct packed_git *p = m->packs[pack_int_id];
1374 size_t expected_size;
1378 if (!pack_kept_objects && p->pack_keep)
1380 if (open_pack_index(p) || !p->num_objects)
1383 expected_size = (size_t)(p->pack_size
1384 * pack_info[i].referenced_objects);
1385 expected_size /= p->num_objects;
1387 if (expected_size >= batch_size)
1391 total_size += expected_size;
1392 include_pack[pack_int_id] = 1;
1397 if (packs_to_repack < 2)
1403 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1407 unsigned char *include_pack;
1408 struct child_process cmd = CHILD_PROCESS_INIT;
1410 struct strbuf base_name = STRBUF_INIT;
1411 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1414 * When updating the default for these configuration
1415 * variables in builtin/repack.c, these must be adjusted
1418 int delta_base_offset = 1;
1419 int use_delta_islands = 0;
1424 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1427 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1429 } else if (fill_included_packs_all(r, m, include_pack))
1432 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1433 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1435 strvec_push(&cmd.args, "pack-objects");
1437 strbuf_addstr(&base_name, object_dir);
1438 strbuf_addstr(&base_name, "/pack/pack");
1439 strvec_push(&cmd.args, base_name.buf);
1441 if (delta_base_offset)
1442 strvec_push(&cmd.args, "--delta-base-offset");
1443 if (use_delta_islands)
1444 strvec_push(&cmd.args, "--delta-islands");
1446 if (flags & MIDX_PROGRESS)
1447 strvec_push(&cmd.args, "--progress");
1449 strvec_push(&cmd.args, "-q");
1451 strbuf_release(&base_name);
1454 cmd.in = cmd.out = -1;
1456 if (start_command(&cmd)) {
1457 error(_("could not start pack-objects"));
1462 cmd_in = xfdopen(cmd.in, "w");
1464 for (i = 0; i < m->num_objects; i++) {
1465 struct object_id oid;
1466 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1468 if (!include_pack[pack_int_id])
1471 nth_midxed_object_oid(&oid, m, i);
1472 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
1476 if (finish_command(&cmd)) {
1477 error(_("could not finish pack-objects"));
1482 result = write_midx_internal(object_dir, m, NULL, flags);