7 #include "object-store.h"
8 #include "hash-lookup.h"
12 #include "run-command.h"
13 #include "repository.h"
14 #include "chunk-format.h"
17 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
18 #define MIDX_VERSION 1
19 #define MIDX_BYTE_FILE_VERSION 4
20 #define MIDX_BYTE_HASH_VERSION 5
21 #define MIDX_BYTE_NUM_CHUNKS 6
22 #define MIDX_BYTE_NUM_PACKS 8
23 #define MIDX_HEADER_SIZE 12
24 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
26 #define MIDX_CHUNK_ALIGNMENT 4
27 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
28 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
29 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
30 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
31 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
32 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
33 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
34 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
35 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
37 #define PACK_EXPIRED UINT_MAX
39 static uint8_t oid_version(void)
41 switch (hash_algo_by_ptr(the_hash_algo)) {
47 die(_("invalid hash version"));
51 static const unsigned char *get_midx_checksum(struct multi_pack_index *m)
53 return m->data + m->data_len - the_hash_algo->rawsz;
56 static char *get_midx_filename(const char *object_dir)
58 return xstrfmt("%s/pack/multi-pack-index", object_dir);
61 char *get_midx_rev_filename(struct multi_pack_index *m)
63 return xstrfmt("%s/pack/multi-pack-index-%s.rev",
64 m->object_dir, hash_to_hex(get_midx_checksum(m)));
67 static int midx_read_oid_fanout(const unsigned char *chunk_start,
68 size_t chunk_size, void *data)
70 struct multi_pack_index *m = data;
71 m->chunk_oid_fanout = (uint32_t *)chunk_start;
73 if (chunk_size != 4 * 256) {
74 error(_("multi-pack-index OID fanout is of the wrong size"));
80 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
82 struct multi_pack_index *m = NULL;
86 void *midx_map = NULL;
87 uint32_t hash_version;
88 char *midx_name = get_midx_filename(object_dir);
90 const char *cur_pack_name;
91 struct chunkfile *cf = NULL;
93 fd = git_open(midx_name);
98 error_errno(_("failed to read %s"), midx_name);
102 midx_size = xsize_t(st.st_size);
104 if (midx_size < MIDX_MIN_SIZE) {
105 error(_("multi-pack-index file %s is too small"), midx_name);
109 FREE_AND_NULL(midx_name);
111 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
114 FLEX_ALLOC_STR(m, object_dir, object_dir);
116 m->data_len = midx_size;
119 m->signature = get_be32(m->data);
120 if (m->signature != MIDX_SIGNATURE)
121 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
122 m->signature, MIDX_SIGNATURE);
124 m->version = m->data[MIDX_BYTE_FILE_VERSION];
125 if (m->version != MIDX_VERSION)
126 die(_("multi-pack-index version %d not recognized"),
129 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
130 if (hash_version != oid_version()) {
131 error(_("multi-pack-index hash version %u does not match version %u"),
132 hash_version, oid_version());
135 m->hash_len = the_hash_algo->rawsz;
137 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
139 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
141 cf = init_chunkfile(NULL);
143 if (read_table_of_contents(cf, m->data, midx_size,
144 MIDX_HEADER_SIZE, m->num_chunks))
147 if (pair_chunk(cf, MIDX_CHUNKID_PACKNAMES, &m->chunk_pack_names) == CHUNK_NOT_FOUND)
148 die(_("multi-pack-index missing required pack-name chunk"));
149 if (read_chunk(cf, MIDX_CHUNKID_OIDFANOUT, midx_read_oid_fanout, m) == CHUNK_NOT_FOUND)
150 die(_("multi-pack-index missing required OID fanout chunk"));
151 if (pair_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, &m->chunk_oid_lookup) == CHUNK_NOT_FOUND)
152 die(_("multi-pack-index missing required OID lookup chunk"));
153 if (pair_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, &m->chunk_object_offsets) == CHUNK_NOT_FOUND)
154 die(_("multi-pack-index missing required object offsets chunk"));
156 pair_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, &m->chunk_large_offsets);
158 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
160 CALLOC_ARRAY(m->pack_names, m->num_packs);
161 CALLOC_ARRAY(m->packs, m->num_packs);
163 cur_pack_name = (const char *)m->chunk_pack_names;
164 for (i = 0; i < m->num_packs; i++) {
165 m->pack_names[i] = cur_pack_name;
167 cur_pack_name += strlen(cur_pack_name) + 1;
169 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
170 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
171 m->pack_names[i - 1],
175 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
176 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
185 munmap(midx_map, midx_size);
191 void close_midx(struct multi_pack_index *m)
198 munmap((unsigned char *)m->data, m->data_len);
200 for (i = 0; i < m->num_packs; i++) {
202 m->packs[i]->multi_pack_index = 0;
204 FREE_AND_NULL(m->packs);
205 FREE_AND_NULL(m->pack_names);
208 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
210 struct strbuf pack_name = STRBUF_INIT;
211 struct packed_git *p;
213 if (pack_int_id >= m->num_packs)
214 die(_("bad pack-int-id: %u (%u total packs)"),
215 pack_int_id, m->num_packs);
217 if (m->packs[pack_int_id])
220 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
221 m->pack_names[pack_int_id]);
223 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
224 strbuf_release(&pack_name);
229 p->multi_pack_index = 1;
230 m->packs[pack_int_id] = p;
231 install_packed_git(r, p);
232 list_add_tail(&p->mru, &r->objects->packed_git_mru);
237 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
239 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
240 the_hash_algo->rawsz, result);
243 struct object_id *nth_midxed_object_oid(struct object_id *oid,
244 struct multi_pack_index *m,
247 if (n >= m->num_objects)
250 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
254 off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
256 const unsigned char *offset_data;
259 offset_data = m->chunk_object_offsets + (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH;
260 offset32 = get_be32(offset_data + sizeof(uint32_t));
262 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
263 if (sizeof(off_t) < sizeof(uint64_t))
264 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
266 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
267 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
273 uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
275 return get_be32(m->chunk_object_offsets +
276 (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH);
279 static int nth_midxed_pack_entry(struct repository *r,
280 struct multi_pack_index *m,
281 struct pack_entry *e,
284 uint32_t pack_int_id;
285 struct packed_git *p;
287 if (pos >= m->num_objects)
290 pack_int_id = nth_midxed_pack_int_id(m, pos);
292 if (prepare_midx_pack(r, m, pack_int_id))
294 p = m->packs[pack_int_id];
297 * We are about to tell the caller where they can locate the
298 * requested object. We better make sure the packfile is
299 * still here and can be accessed before supplying that
300 * answer, as it may have been deleted since the MIDX was
303 if (!is_pack_valid(p))
306 if (p->num_bad_objects) {
308 struct object_id oid;
309 nth_midxed_object_oid(&oid, m, pos);
310 for (i = 0; i < p->num_bad_objects; i++)
312 p->bad_object_sha1 + the_hash_algo->rawsz * i))
316 e->offset = nth_midxed_offset(m, pos);
322 int fill_midx_entry(struct repository * r,
323 const struct object_id *oid,
324 struct pack_entry *e,
325 struct multi_pack_index *m)
329 if (!bsearch_midx(oid, m, &pos))
332 return nth_midxed_pack_entry(r, m, e, pos);
335 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
336 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
337 const char *idx_name)
339 /* Skip past any initial matching prefix. */
340 while (*idx_name && *idx_name == *idx_or_pack_name) {
346 * If we didn't match completely, we may have matched "pack-1234." and
347 * be left with "idx" and "pack" respectively, which is also OK. We do
348 * not have to check for "idx" and "idx", because that would have been
349 * a complete match (and in that case these strcmps will be false, but
350 * we'll correctly return 0 from the final strcmp() below.
352 * Technically this matches "fooidx" and "foopack", but we'd never have
353 * such names in the first place.
355 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
359 * This not only checks for a complete match, but also orders based on
360 * the first non-identical character, which means our ordering will
361 * match a raw strcmp(). That makes it OK to use this to binary search
362 * a naively-sorted list.
364 return strcmp(idx_or_pack_name, idx_name);
367 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
369 uint32_t first = 0, last = m->num_packs;
371 while (first < last) {
372 uint32_t mid = first + (last - first) / 2;
376 current = m->pack_names[mid];
377 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
390 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
392 struct multi_pack_index *m;
393 struct multi_pack_index *m_search;
395 prepare_repo_settings(r);
396 if (!r->settings.core_multi_pack_index)
399 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
400 if (!strcmp(object_dir, m_search->object_dir))
403 m = load_multi_pack_index(object_dir, local);
406 struct multi_pack_index *mp = r->objects->multi_pack_index;
411 r->objects->multi_pack_index = m;
418 static size_t write_midx_header(struct hashfile *f,
419 unsigned char num_chunks,
422 hashwrite_be32(f, MIDX_SIGNATURE);
423 hashwrite_u8(f, MIDX_VERSION);
424 hashwrite_u8(f, oid_version());
425 hashwrite_u8(f, num_chunks);
426 hashwrite_u8(f, 0); /* unused */
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);
446 static int idx_or_pack_name_cmp(const void *_va, const void *_vb)
448 const char *pack_name = _va;
449 const struct pack_info *compar = _vb;
451 return cmp_idx_or_pack_name(pack_name, compar->pack_name);
454 struct write_midx_context {
455 struct pack_info *info;
458 struct multi_pack_index *m;
459 struct progress *progress;
460 unsigned pack_paths_checked;
462 struct pack_midx_entry *entries;
466 uint32_t *pack_order;
467 unsigned large_offsets_needed:1;
468 uint32_t num_large_offsets;
470 int preferred_pack_idx;
473 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
474 const char *file_name, void *data)
476 struct write_midx_context *ctx = data;
478 if (ends_with(file_name, ".idx")) {
479 display_progress(ctx->progress, ++ctx->pack_paths_checked);
480 if (ctx->m && midx_contains_pack(ctx->m, file_name))
483 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
485 ctx->info[ctx->nr].p = add_packed_git(full_path,
489 if (!ctx->info[ctx->nr].p) {
490 warning(_("failed to add packfile '%s'"),
495 if (open_pack_index(ctx->info[ctx->nr].p)) {
496 warning(_("failed to open pack-index '%s'"),
498 close_pack(ctx->info[ctx->nr].p);
499 FREE_AND_NULL(ctx->info[ctx->nr].p);
503 ctx->info[ctx->nr].pack_name = xstrdup(file_name);
504 ctx->info[ctx->nr].orig_pack_int_id = ctx->nr;
505 ctx->info[ctx->nr].expired = 0;
510 struct pack_midx_entry {
511 struct object_id oid;
512 uint32_t pack_int_id;
515 unsigned preferred : 1;
518 static int midx_oid_compare(const void *_a, const void *_b)
520 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
521 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
522 int cmp = oidcmp(&a->oid, &b->oid);
527 /* Sort objects in a preferred pack first when multiple copies exist. */
528 if (a->preferred > b->preferred)
530 if (a->preferred < b->preferred)
533 if (a->pack_mtime > b->pack_mtime)
535 else if (a->pack_mtime < b->pack_mtime)
538 return a->pack_int_id - b->pack_int_id;
541 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
542 struct pack_midx_entry *e,
545 if (pos >= m->num_objects)
548 nth_midxed_object_oid(&e->oid, m, pos);
549 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
550 e->offset = nth_midxed_offset(m, pos);
552 /* consider objects in midx to be from "old" packs */
557 static void fill_pack_entry(uint32_t pack_int_id,
558 struct packed_git *p,
560 struct pack_midx_entry *entry,
563 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
564 die(_("failed to locate object %d in packfile"), cur_object);
566 entry->pack_int_id = pack_int_id;
567 entry->pack_mtime = p->mtime;
569 entry->offset = nth_packed_object_offset(p, cur_object);
570 entry->preferred = !!preferred;
574 * It is possible to artificially get into a state where there are many
575 * duplicate copies of objects. That can create high memory pressure if
576 * we are to create a list of all objects before de-duplication. To reduce
577 * this memory pressure without a significant performance drop, automatically
578 * group objects by the first byte of their object id. Use the IDX fanout
579 * tables to group the data, copy to a local array, then sort.
581 * Copy only the de-duplicated entries (selected by most-recent modified time
582 * of a packfile containing the object).
584 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
585 struct pack_info *info,
587 uint32_t *nr_objects,
590 uint32_t cur_fanout, cur_pack, cur_object;
591 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
592 struct pack_midx_entry *entries_by_fanout = NULL;
593 struct pack_midx_entry *deduplicated_entries = NULL;
594 uint32_t start_pack = m ? m->num_packs : 0;
596 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
597 total_objects += info[cur_pack].p->num_objects;
600 * As we de-duplicate by fanout value, we expect the fanout
601 * slices to be evenly distributed, with some noise. Hence,
602 * allocate slightly more than one 256th.
604 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
606 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
607 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
610 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
611 uint32_t nr_fanout = 0;
614 uint32_t start = 0, end;
617 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
618 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
620 for (cur_object = start; cur_object < end; cur_object++) {
621 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
622 nth_midxed_pack_midx_entry(m,
623 &entries_by_fanout[nr_fanout],
625 if (nth_midxed_pack_int_id(m, cur_object) == preferred_pack)
626 entries_by_fanout[nr_fanout].preferred = 1;
628 entries_by_fanout[nr_fanout].preferred = 0;
633 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
634 uint32_t start = 0, end;
635 int preferred = cur_pack == preferred_pack;
638 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
639 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
641 for (cur_object = start; cur_object < end; cur_object++) {
642 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
643 fill_pack_entry(cur_pack,
646 &entries_by_fanout[nr_fanout],
652 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
655 * The batch is now sorted by OID and then mtime (descending).
656 * Take only the first duplicate.
658 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
659 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
660 &entries_by_fanout[cur_object].oid))
663 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
664 memcpy(&deduplicated_entries[*nr_objects],
665 &entries_by_fanout[cur_object],
666 sizeof(struct pack_midx_entry));
671 free(entries_by_fanout);
672 return deduplicated_entries;
675 static int write_midx_pack_names(struct hashfile *f, void *data)
677 struct write_midx_context *ctx = data;
679 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
682 for (i = 0; i < ctx->nr; i++) {
685 if (ctx->info[i].expired)
688 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
689 BUG("incorrect pack-file order: %s before %s",
690 ctx->info[i - 1].pack_name,
691 ctx->info[i].pack_name);
693 writelen = strlen(ctx->info[i].pack_name) + 1;
694 hashwrite(f, ctx->info[i].pack_name, writelen);
698 /* add padding to be aligned */
699 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
700 if (i < MIDX_CHUNK_ALIGNMENT) {
701 memset(padding, 0, sizeof(padding));
702 hashwrite(f, padding, i);
708 static int write_midx_oid_fanout(struct hashfile *f,
711 struct write_midx_context *ctx = data;
712 struct pack_midx_entry *list = ctx->entries;
713 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
718 * Write the first-level table (the list is sorted,
719 * but we use a 256-entry lookup to be able to avoid
720 * having to do eight extra binary search iterations).
722 for (i = 0; i < 256; i++) {
723 struct pack_midx_entry *next = list;
725 while (next < last && next->oid.hash[0] == i) {
730 hashwrite_be32(f, count);
737 static int write_midx_oid_lookup(struct hashfile *f,
740 struct write_midx_context *ctx = data;
741 unsigned char hash_len = the_hash_algo->rawsz;
742 struct pack_midx_entry *list = ctx->entries;
745 for (i = 0; i < ctx->entries_nr; i++) {
746 struct pack_midx_entry *obj = list++;
748 if (i < ctx->entries_nr - 1) {
749 struct pack_midx_entry *next = list;
750 if (oidcmp(&obj->oid, &next->oid) >= 0)
751 BUG("OIDs not in order: %s >= %s",
752 oid_to_hex(&obj->oid),
753 oid_to_hex(&next->oid));
756 hashwrite(f, obj->oid.hash, (int)hash_len);
762 static int write_midx_object_offsets(struct hashfile *f,
765 struct write_midx_context *ctx = data;
766 struct pack_midx_entry *list = ctx->entries;
767 uint32_t i, nr_large_offset = 0;
769 for (i = 0; i < ctx->entries_nr; i++) {
770 struct pack_midx_entry *obj = list++;
772 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
773 BUG("object %s is in an expired pack with int-id %d",
774 oid_to_hex(&obj->oid),
777 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
779 if (ctx->large_offsets_needed && obj->offset >> 31)
780 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
781 else if (!ctx->large_offsets_needed && obj->offset >> 32)
782 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
783 oid_to_hex(&obj->oid),
786 hashwrite_be32(f, (uint32_t)obj->offset);
792 static int write_midx_large_offsets(struct hashfile *f,
795 struct write_midx_context *ctx = data;
796 struct pack_midx_entry *list = ctx->entries;
797 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
798 uint32_t nr_large_offset = ctx->num_large_offsets;
800 while (nr_large_offset) {
801 struct pack_midx_entry *obj;
805 BUG("too many large-offset objects");
808 offset = obj->offset;
813 hashwrite_be64(f, offset);
821 struct midx_pack_order_data {
827 static int midx_pack_order_cmp(const void *va, const void *vb)
829 const struct midx_pack_order_data *a = va, *b = vb;
830 if (a->pack < b->pack)
832 else if (a->pack > b->pack)
834 else if (a->offset < b->offset)
836 else if (a->offset > b->offset)
842 static uint32_t *midx_pack_order(struct write_midx_context *ctx)
844 struct midx_pack_order_data *data;
845 uint32_t *pack_order;
848 ALLOC_ARRAY(data, ctx->entries_nr);
849 for (i = 0; i < ctx->entries_nr; i++) {
850 struct pack_midx_entry *e = &ctx->entries[i];
852 data[i].pack = ctx->pack_perm[e->pack_int_id];
854 data[i].pack |= (1U << 31);
855 data[i].offset = e->offset;
858 QSORT(data, ctx->entries_nr, midx_pack_order_cmp);
860 ALLOC_ARRAY(pack_order, ctx->entries_nr);
861 for (i = 0; i < ctx->entries_nr; i++)
862 pack_order[i] = data[i].nr;
868 static void write_midx_reverse_index(char *midx_name, unsigned char *midx_hash,
869 struct write_midx_context *ctx)
871 struct strbuf buf = STRBUF_INIT;
872 const char *tmp_file;
874 strbuf_addf(&buf, "%s-%s.rev", midx_name, hash_to_hex(midx_hash));
876 tmp_file = write_rev_file_order(NULL, ctx->pack_order, ctx->entries_nr,
877 midx_hash, WRITE_REV);
879 if (finalize_object_file(tmp_file, buf.buf))
880 die(_("cannot store reverse index file"));
882 strbuf_release(&buf);
885 static void clear_midx_files_ext(struct repository *r, const char *ext,
886 unsigned char *keep_hash);
888 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
889 struct string_list *packs_to_drop,
890 const char *preferred_pack_name,
894 unsigned char midx_hash[GIT_MAX_RAWSZ];
896 struct hashfile *f = NULL;
898 struct write_midx_context ctx = { 0 };
899 int pack_name_concat_len = 0;
900 int dropped_packs = 0;
902 struct chunkfile *cf;
904 midx_name = get_midx_filename(object_dir);
905 if (safe_create_leading_directories(midx_name))
906 die_errno(_("unable to create leading directories of %s"),
912 ctx.m = load_multi_pack_index(object_dir, 1);
915 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
917 ALLOC_ARRAY(ctx.info, ctx.alloc);
920 for (i = 0; i < ctx.m->num_packs; i++) {
921 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
923 ctx.info[ctx.nr].orig_pack_int_id = i;
924 ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]);
925 ctx.info[ctx.nr].p = NULL;
926 ctx.info[ctx.nr].expired = 0;
931 ctx.pack_paths_checked = 0;
932 if (flags & MIDX_PROGRESS)
933 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
937 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
938 stop_progress(&ctx.progress);
940 if (ctx.m && ctx.nr == ctx.m->num_packs && !packs_to_drop)
943 ctx.preferred_pack_idx = -1;
944 if (preferred_pack_name) {
945 for (i = 0; i < ctx.nr; i++) {
946 if (!cmp_idx_or_pack_name(preferred_pack_name,
947 ctx.info[i].pack_name)) {
948 ctx.preferred_pack_idx = i;
954 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr,
955 ctx.preferred_pack_idx);
957 ctx.large_offsets_needed = 0;
958 for (i = 0; i < ctx.entries_nr; i++) {
959 if (ctx.entries[i].offset > 0x7fffffff)
960 ctx.num_large_offsets++;
961 if (ctx.entries[i].offset > 0xffffffff)
962 ctx.large_offsets_needed = 1;
965 QSORT(ctx.info, ctx.nr, pack_info_compare);
967 if (packs_to_drop && packs_to_drop->nr) {
969 int missing_drops = 0;
971 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
972 int cmp = strcmp(ctx.info[i].pack_name,
973 packs_to_drop->items[drop_index].string);
977 ctx.info[i].expired = 1;
978 } else if (cmp > 0) {
979 error(_("did not see pack-file %s to drop"),
980 packs_to_drop->items[drop_index].string);
985 ctx.info[i].expired = 0;
996 * pack_perm stores a permutation between pack-int-ids from the
997 * previous multi-pack-index to the new one we are writing:
999 * pack_perm[old_id] = new_id
1001 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
1002 for (i = 0; i < ctx.nr; i++) {
1003 if (ctx.info[i].expired) {
1005 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
1007 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
1011 for (i = 0; i < ctx.nr; i++) {
1012 if (!ctx.info[i].expired)
1013 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
1016 /* Check that the preferred pack wasn't expired (if given). */
1017 if (preferred_pack_name) {
1018 struct pack_info *preferred = bsearch(preferred_pack_name,
1021 idx_or_pack_name_cmp);
1024 warning(_("unknown preferred pack: '%s'"),
1025 preferred_pack_name);
1027 uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id];
1028 if (perm == PACK_EXPIRED)
1029 warning(_("preferred pack '%s' is expired"),
1030 preferred_pack_name);
1034 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
1035 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
1036 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
1038 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
1039 f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk));
1044 if (ctx.nr - dropped_packs == 0) {
1045 error(_("no pack files to index."));
1050 cf = init_chunkfile(f);
1052 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
1053 write_midx_pack_names);
1054 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
1055 write_midx_oid_fanout);
1056 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
1057 (size_t)ctx.entries_nr * the_hash_algo->rawsz,
1058 write_midx_oid_lookup);
1059 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
1060 (size_t)ctx.entries_nr * MIDX_CHUNK_OFFSET_WIDTH,
1061 write_midx_object_offsets);
1063 if (ctx.large_offsets_needed)
1064 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
1065 (size_t)ctx.num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH,
1066 write_midx_large_offsets);
1068 write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs);
1069 write_chunkfile(cf, &ctx);
1071 finalize_hashfile(f, midx_hash, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1074 if (flags & MIDX_WRITE_REV_INDEX)
1075 ctx.pack_order = midx_pack_order(&ctx);
1077 if (flags & MIDX_WRITE_REV_INDEX)
1078 write_midx_reverse_index(midx_name, midx_hash, &ctx);
1079 clear_midx_files_ext(the_repository, ".rev", midx_hash);
1081 commit_lock_file(&lk);
1084 for (i = 0; i < ctx.nr; i++) {
1085 if (ctx.info[i].p) {
1086 close_pack(ctx.info[i].p);
1087 free(ctx.info[i].p);
1089 free(ctx.info[i].pack_name);
1094 free(ctx.pack_perm);
1095 free(ctx.pack_order);
1100 int write_midx_file(const char *object_dir,
1101 const char *preferred_pack_name,
1104 return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name,
1108 struct clear_midx_data {
1113 static void clear_midx_file_ext(const char *full_path, size_t full_path_len,
1114 const char *file_name, void *_data)
1116 struct clear_midx_data *data = _data;
1118 if (!(starts_with(file_name, "multi-pack-index-") &&
1119 ends_with(file_name, data->ext)))
1121 if (data->keep && !strcmp(data->keep, file_name))
1124 if (unlink(full_path))
1125 die_errno(_("failed to remove %s"), full_path);
1128 static void clear_midx_files_ext(struct repository *r, const char *ext,
1129 unsigned char *keep_hash)
1131 struct clear_midx_data data;
1132 memset(&data, 0, sizeof(struct clear_midx_data));
1135 data.keep = xstrfmt("multi-pack-index-%s%s",
1136 hash_to_hex(keep_hash), ext);
1139 for_each_file_in_pack_dir(r->objects->odb->path,
1140 clear_midx_file_ext,
1146 void clear_midx_file(struct repository *r)
1148 char *midx = get_midx_filename(r->objects->odb->path);
1150 if (r->objects && r->objects->multi_pack_index) {
1151 close_midx(r->objects->multi_pack_index);
1152 r->objects->multi_pack_index = NULL;
1155 if (remove_path(midx))
1156 die(_("failed to clear multi-pack-index at %s"), midx);
1158 clear_midx_files_ext(r, ".rev", NULL);
1163 static int verify_midx_error;
1165 static void midx_report(const char *fmt, ...)
1168 verify_midx_error = 1;
1170 vfprintf(stderr, fmt, ap);
1171 fprintf(stderr, "\n");
1175 struct pair_pos_vs_id
1178 uint32_t pack_int_id;
1181 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1183 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1184 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1186 return b->pack_int_id - a->pack_int_id;
1190 * Limit calls to display_progress() for performance reasons.
1191 * The interval here was arbitrarily chosen.
1193 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1194 #define midx_display_sparse_progress(progress, n) \
1196 uint64_t _n = (n); \
1197 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1198 display_progress(progress, _n); \
1201 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1203 struct pair_pos_vs_id *pairs = NULL;
1205 struct progress *progress = NULL;
1206 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1207 verify_midx_error = 0;
1212 char *filename = get_midx_filename(object_dir);
1213 if (!stat(filename, &sb)) {
1214 error(_("multi-pack-index file exists, but failed to parse"));
1221 if (flags & MIDX_PROGRESS)
1222 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1224 for (i = 0; i < m->num_packs; i++) {
1225 if (prepare_midx_pack(r, m, i))
1226 midx_report("failed to load pack in position %d", i);
1228 display_progress(progress, i + 1);
1230 stop_progress(&progress);
1232 for (i = 0; i < 255; i++) {
1233 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1234 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1236 if (oid_fanout1 > oid_fanout2)
1237 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1238 i, oid_fanout1, oid_fanout2, i + 1);
1241 if (m->num_objects == 0) {
1242 midx_report(_("the midx contains no oid"));
1244 * Remaining tests assume that we have objects, so we can
1247 return verify_midx_error;
1250 if (flags & MIDX_PROGRESS)
1251 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1252 m->num_objects - 1);
1253 for (i = 0; i < m->num_objects - 1; i++) {
1254 struct object_id oid1, oid2;
1256 nth_midxed_object_oid(&oid1, m, i);
1257 nth_midxed_object_oid(&oid2, m, i + 1);
1259 if (oidcmp(&oid1, &oid2) >= 0)
1260 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1261 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1263 midx_display_sparse_progress(progress, i + 1);
1265 stop_progress(&progress);
1268 * Create an array mapping each object to its packfile id. Sort it
1269 * to group the objects by packfile. Use this permutation to visit
1270 * each of the objects and only require 1 packfile to be open at a
1273 ALLOC_ARRAY(pairs, m->num_objects);
1274 for (i = 0; i < m->num_objects; i++) {
1276 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1279 if (flags & MIDX_PROGRESS)
1280 progress = start_sparse_progress(_("Sorting objects by packfile"),
1282 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1283 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1284 stop_progress(&progress);
1286 if (flags & MIDX_PROGRESS)
1287 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1288 for (i = 0; i < m->num_objects; i++) {
1289 struct object_id oid;
1290 struct pack_entry e;
1291 off_t m_offset, p_offset;
1293 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1294 m->packs[pairs[i-1].pack_int_id])
1296 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1297 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1300 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1302 if (!fill_midx_entry(r, &oid, &e, m)) {
1303 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1304 pairs[i].pos, oid_to_hex(&oid));
1308 if (open_pack_index(e.p)) {
1309 midx_report(_("failed to load pack-index for packfile %s"),
1314 m_offset = e.offset;
1315 p_offset = find_pack_entry_one(oid.hash, e.p);
1317 if (m_offset != p_offset)
1318 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1319 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1321 midx_display_sparse_progress(progress, i + 1);
1323 stop_progress(&progress);
1327 return verify_midx_error;
1330 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1332 uint32_t i, *count, result = 0;
1333 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1334 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1335 struct progress *progress = NULL;
1340 CALLOC_ARRAY(count, m->num_packs);
1342 if (flags & MIDX_PROGRESS)
1343 progress = start_delayed_progress(_("Counting referenced objects"),
1345 for (i = 0; i < m->num_objects; i++) {
1346 int pack_int_id = nth_midxed_pack_int_id(m, i);
1347 count[pack_int_id]++;
1348 display_progress(progress, i + 1);
1350 stop_progress(&progress);
1352 if (flags & MIDX_PROGRESS)
1353 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1355 for (i = 0; i < m->num_packs; i++) {
1357 display_progress(progress, i + 1);
1362 if (prepare_midx_pack(r, m, i))
1365 if (m->packs[i]->pack_keep)
1368 pack_name = xstrdup(m->packs[i]->pack_name);
1369 close_pack(m->packs[i]);
1371 string_list_insert(&packs_to_drop, m->pack_names[i]);
1372 unlink_pack_path(pack_name, 0);
1375 stop_progress(&progress);
1379 if (packs_to_drop.nr)
1380 result = write_midx_internal(object_dir, m, &packs_to_drop, NULL, flags);
1382 string_list_clear(&packs_to_drop, 0);
1386 struct repack_info {
1388 uint32_t referenced_objects;
1389 uint32_t pack_int_id;
1392 static int compare_by_mtime(const void *a_, const void *b_)
1394 const struct repack_info *a, *b;
1396 a = (const struct repack_info *)a_;
1397 b = (const struct repack_info *)b_;
1399 if (a->mtime < b->mtime)
1401 if (a->mtime > b->mtime)
1406 static int fill_included_packs_all(struct repository *r,
1407 struct multi_pack_index *m,
1408 unsigned char *include_pack)
1410 uint32_t i, count = 0;
1411 int pack_kept_objects = 0;
1413 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1415 for (i = 0; i < m->num_packs; i++) {
1416 if (prepare_midx_pack(r, m, i))
1418 if (!pack_kept_objects && m->packs[i]->pack_keep)
1421 include_pack[i] = 1;
1428 static int fill_included_packs_batch(struct repository *r,
1429 struct multi_pack_index *m,
1430 unsigned char *include_pack,
1433 uint32_t i, packs_to_repack;
1435 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1436 int pack_kept_objects = 0;
1438 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1440 for (i = 0; i < m->num_packs; i++) {
1441 pack_info[i].pack_int_id = i;
1443 if (prepare_midx_pack(r, m, i))
1446 pack_info[i].mtime = m->packs[i]->mtime;
1449 for (i = 0; batch_size && i < m->num_objects; i++) {
1450 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1451 pack_info[pack_int_id].referenced_objects++;
1454 QSORT(pack_info, m->num_packs, compare_by_mtime);
1457 packs_to_repack = 0;
1458 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1459 int pack_int_id = pack_info[i].pack_int_id;
1460 struct packed_git *p = m->packs[pack_int_id];
1461 size_t expected_size;
1465 if (!pack_kept_objects && p->pack_keep)
1467 if (open_pack_index(p) || !p->num_objects)
1470 expected_size = (size_t)(p->pack_size
1471 * pack_info[i].referenced_objects);
1472 expected_size /= p->num_objects;
1474 if (expected_size >= batch_size)
1478 total_size += expected_size;
1479 include_pack[pack_int_id] = 1;
1484 if (packs_to_repack < 2)
1490 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1494 unsigned char *include_pack;
1495 struct child_process cmd = CHILD_PROCESS_INIT;
1497 struct strbuf base_name = STRBUF_INIT;
1498 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1501 * When updating the default for these configuration
1502 * variables in builtin/repack.c, these must be adjusted
1505 int delta_base_offset = 1;
1506 int use_delta_islands = 0;
1511 CALLOC_ARRAY(include_pack, m->num_packs);
1514 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1516 } else if (fill_included_packs_all(r, m, include_pack))
1519 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1520 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1522 strvec_push(&cmd.args, "pack-objects");
1524 strbuf_addstr(&base_name, object_dir);
1525 strbuf_addstr(&base_name, "/pack/pack");
1526 strvec_push(&cmd.args, base_name.buf);
1528 if (delta_base_offset)
1529 strvec_push(&cmd.args, "--delta-base-offset");
1530 if (use_delta_islands)
1531 strvec_push(&cmd.args, "--delta-islands");
1533 if (flags & MIDX_PROGRESS)
1534 strvec_push(&cmd.args, "--progress");
1536 strvec_push(&cmd.args, "-q");
1538 strbuf_release(&base_name);
1541 cmd.in = cmd.out = -1;
1543 if (start_command(&cmd)) {
1544 error(_("could not start pack-objects"));
1549 cmd_in = xfdopen(cmd.in, "w");
1551 for (i = 0; i < m->num_objects; i++) {
1552 struct object_id oid;
1553 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1555 if (!include_pack[pack_int_id])
1558 nth_midxed_object_oid(&oid, m, i);
1559 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
1563 if (finish_command(&cmd)) {
1564 error(_("could not finish pack-objects"));
1569 result = write_midx_internal(object_dir, m, NULL, NULL, flags);