midx: enable core.multiPackIndex by default
[git.git] / midx.c
blobef499cf5046fdfa216b2f12efb1f4a88f37a68fc
1 #include "cache.h"
2 #include "config.h"
3 #include "csum-file.h"
4 #include "dir.h"
5 #include "lockfile.h"
6 #include "packfile.h"
7 #include "object-store.h"
8 #include "sha1-lookup.h"
9 #include "midx.h"
10 #include "progress.h"
11 #include "trace2.h"
12 #include "run-command.h"
13 #include "repository.h"
15 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
16 #define MIDX_VERSION 1
17 #define MIDX_BYTE_FILE_VERSION 4
18 #define MIDX_BYTE_HASH_VERSION 5
19 #define MIDX_BYTE_NUM_CHUNKS 6
20 #define MIDX_BYTE_NUM_PACKS 8
21 #define MIDX_HASH_VERSION 1
22 #define MIDX_HEADER_SIZE 12
23 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
25 #define MIDX_MAX_CHUNKS 5
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_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
33 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
34 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
35 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
36 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
38 #define PACK_EXPIRED UINT_MAX
40 static char *get_midx_filename(const char *object_dir)
42 return xstrfmt("%s/pack/multi-pack-index", object_dir);
45 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
47 struct multi_pack_index *m = NULL;
48 int fd;
49 struct stat st;
50 size_t midx_size;
51 void *midx_map = NULL;
52 uint32_t hash_version;
53 char *midx_name = get_midx_filename(object_dir);
54 uint32_t i;
55 const char *cur_pack_name;
57 fd = git_open(midx_name);
59 if (fd < 0)
60 goto cleanup_fail;
61 if (fstat(fd, &st)) {
62 error_errno(_("failed to read %s"), midx_name);
63 goto cleanup_fail;
66 midx_size = xsize_t(st.st_size);
68 if (midx_size < MIDX_MIN_SIZE) {
69 error(_("multi-pack-index file %s is too small"), midx_name);
70 goto cleanup_fail;
73 FREE_AND_NULL(midx_name);
75 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
76 close(fd);
78 FLEX_ALLOC_STR(m, object_dir, object_dir);
79 m->data = midx_map;
80 m->data_len = midx_size;
81 m->local = local;
83 m->signature = get_be32(m->data);
84 if (m->signature != MIDX_SIGNATURE)
85 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
86 m->signature, MIDX_SIGNATURE);
88 m->version = m->data[MIDX_BYTE_FILE_VERSION];
89 if (m->version != MIDX_VERSION)
90 die(_("multi-pack-index version %d not recognized"),
91 m->version);
93 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
94 if (hash_version != MIDX_HASH_VERSION)
95 die(_("hash version %u does not match"), hash_version);
96 m->hash_len = the_hash_algo->rawsz;
98 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
100 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
102 for (i = 0; i < m->num_chunks; i++) {
103 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
104 MIDX_CHUNKLOOKUP_WIDTH * i);
105 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
106 MIDX_CHUNKLOOKUP_WIDTH * i);
108 if (chunk_offset >= m->data_len)
109 die(_("invalid chunk offset (too large)"));
111 switch (chunk_id) {
112 case MIDX_CHUNKID_PACKNAMES:
113 m->chunk_pack_names = m->data + chunk_offset;
114 break;
116 case MIDX_CHUNKID_OIDFANOUT:
117 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
118 break;
120 case MIDX_CHUNKID_OIDLOOKUP:
121 m->chunk_oid_lookup = m->data + chunk_offset;
122 break;
124 case MIDX_CHUNKID_OBJECTOFFSETS:
125 m->chunk_object_offsets = m->data + chunk_offset;
126 break;
128 case MIDX_CHUNKID_LARGEOFFSETS:
129 m->chunk_large_offsets = m->data + chunk_offset;
130 break;
132 case 0:
133 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
134 break;
136 default:
138 * Do nothing on unrecognized chunks, allowing future
139 * extensions to add optional chunks.
141 break;
145 if (!m->chunk_pack_names)
146 die(_("multi-pack-index missing required pack-name chunk"));
147 if (!m->chunk_oid_fanout)
148 die(_("multi-pack-index missing required OID fanout chunk"));
149 if (!m->chunk_oid_lookup)
150 die(_("multi-pack-index missing required OID lookup chunk"));
151 if (!m->chunk_object_offsets)
152 die(_("multi-pack-index missing required object offsets chunk"));
154 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
156 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
157 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
159 cur_pack_name = (const char *)m->chunk_pack_names;
160 for (i = 0; i < m->num_packs; i++) {
161 m->pack_names[i] = cur_pack_name;
163 cur_pack_name += strlen(cur_pack_name) + 1;
165 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
166 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
167 m->pack_names[i - 1],
168 m->pack_names[i]);
171 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
172 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
174 return m;
176 cleanup_fail:
177 free(m);
178 free(midx_name);
179 if (midx_map)
180 munmap(midx_map, midx_size);
181 if (0 <= fd)
182 close(fd);
183 return NULL;
186 void close_midx(struct multi_pack_index *m)
188 uint32_t i;
190 if (!m)
191 return;
193 munmap((unsigned char *)m->data, m->data_len);
195 for (i = 0; i < m->num_packs; i++) {
196 if (m->packs[i])
197 m->packs[i]->multi_pack_index = 0;
199 FREE_AND_NULL(m->packs);
200 FREE_AND_NULL(m->pack_names);
203 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
205 struct strbuf pack_name = STRBUF_INIT;
206 struct packed_git *p;
208 if (pack_int_id >= m->num_packs)
209 die(_("bad pack-int-id: %u (%u total packs)"),
210 pack_int_id, m->num_packs);
212 if (m->packs[pack_int_id])
213 return 0;
215 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
216 m->pack_names[pack_int_id]);
218 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
219 strbuf_release(&pack_name);
221 if (!p)
222 return 1;
224 p->multi_pack_index = 1;
225 m->packs[pack_int_id] = p;
226 install_packed_git(r, p);
227 list_add_tail(&p->mru, &r->objects->packed_git_mru);
229 return 0;
232 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
234 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
235 the_hash_algo->rawsz, result);
238 struct object_id *nth_midxed_object_oid(struct object_id *oid,
239 struct multi_pack_index *m,
240 uint32_t n)
242 if (n >= m->num_objects)
243 return NULL;
245 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
246 return oid;
249 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
251 const unsigned char *offset_data;
252 uint32_t offset32;
254 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
255 offset32 = get_be32(offset_data + sizeof(uint32_t));
257 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
258 if (sizeof(off_t) < sizeof(uint64_t))
259 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
261 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
262 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
265 return offset32;
268 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
270 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
273 static int nth_midxed_pack_entry(struct repository *r,
274 struct multi_pack_index *m,
275 struct pack_entry *e,
276 uint32_t pos)
278 uint32_t pack_int_id;
279 struct packed_git *p;
281 if (pos >= m->num_objects)
282 return 0;
284 pack_int_id = nth_midxed_pack_int_id(m, pos);
286 if (prepare_midx_pack(r, m, pack_int_id))
287 die(_("error preparing packfile from multi-pack-index"));
288 p = m->packs[pack_int_id];
291 * We are about to tell the caller where they can locate the
292 * requested object. We better make sure the packfile is
293 * still here and can be accessed before supplying that
294 * answer, as it may have been deleted since the MIDX was
295 * loaded!
297 if (!is_pack_valid(p))
298 return 0;
300 if (p->num_bad_objects) {
301 uint32_t i;
302 struct object_id oid;
303 nth_midxed_object_oid(&oid, m, pos);
304 for (i = 0; i < p->num_bad_objects; i++)
305 if (hasheq(oid.hash,
306 p->bad_object_sha1 + the_hash_algo->rawsz * i))
307 return 0;
310 e->offset = nth_midxed_offset(m, pos);
311 e->p = p;
313 return 1;
316 int fill_midx_entry(struct repository * r,
317 const struct object_id *oid,
318 struct pack_entry *e,
319 struct multi_pack_index *m)
321 uint32_t pos;
323 if (!bsearch_midx(oid, m, &pos))
324 return 0;
326 return nth_midxed_pack_entry(r, m, e, pos);
329 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
330 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
331 const char *idx_name)
333 /* Skip past any initial matching prefix. */
334 while (*idx_name && *idx_name == *idx_or_pack_name) {
335 idx_name++;
336 idx_or_pack_name++;
340 * If we didn't match completely, we may have matched "pack-1234." and
341 * be left with "idx" and "pack" respectively, which is also OK. We do
342 * not have to check for "idx" and "idx", because that would have been
343 * a complete match (and in that case these strcmps will be false, but
344 * we'll correctly return 0 from the final strcmp() below.
346 * Technically this matches "fooidx" and "foopack", but we'd never have
347 * such names in the first place.
349 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
350 return 0;
353 * This not only checks for a complete match, but also orders based on
354 * the first non-identical character, which means our ordering will
355 * match a raw strcmp(). That makes it OK to use this to binary search
356 * a naively-sorted list.
358 return strcmp(idx_or_pack_name, idx_name);
361 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
363 uint32_t first = 0, last = m->num_packs;
365 while (first < last) {
366 uint32_t mid = first + (last - first) / 2;
367 const char *current;
368 int cmp;
370 current = m->pack_names[mid];
371 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
372 if (!cmp)
373 return 1;
374 if (cmp > 0) {
375 first = mid + 1;
376 continue;
378 last = mid;
381 return 0;
384 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
386 struct multi_pack_index *m;
387 struct multi_pack_index *m_search;
389 prepare_repo_settings(r);
390 if (!r->settings.core_multi_pack_index)
391 return 0;
393 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
394 if (!strcmp(object_dir, m_search->object_dir))
395 return 1;
397 m = load_multi_pack_index(object_dir, local);
399 if (m) {
400 m->next = r->objects->multi_pack_index;
401 r->objects->multi_pack_index = m;
402 return 1;
405 return 0;
408 static size_t write_midx_header(struct hashfile *f,
409 unsigned char num_chunks,
410 uint32_t num_packs)
412 unsigned char byte_values[4];
414 hashwrite_be32(f, MIDX_SIGNATURE);
415 byte_values[0] = MIDX_VERSION;
416 byte_values[1] = MIDX_HASH_VERSION;
417 byte_values[2] = num_chunks;
418 byte_values[3] = 0; /* unused */
419 hashwrite(f, byte_values, sizeof(byte_values));
420 hashwrite_be32(f, num_packs);
422 return MIDX_HEADER_SIZE;
425 struct pack_info {
426 uint32_t orig_pack_int_id;
427 char *pack_name;
428 struct packed_git *p;
429 unsigned expired : 1;
432 static int pack_info_compare(const void *_a, const void *_b)
434 struct pack_info *a = (struct pack_info *)_a;
435 struct pack_info *b = (struct pack_info *)_b;
436 return strcmp(a->pack_name, b->pack_name);
439 struct pack_list {
440 struct pack_info *info;
441 uint32_t nr;
442 uint32_t alloc;
443 struct multi_pack_index *m;
444 struct progress *progress;
445 unsigned pack_paths_checked;
448 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
449 const char *file_name, void *data)
451 struct pack_list *packs = (struct pack_list *)data;
453 if (ends_with(file_name, ".idx")) {
454 display_progress(packs->progress, ++packs->pack_paths_checked);
455 if (packs->m && midx_contains_pack(packs->m, file_name))
456 return;
458 ALLOC_GROW(packs->info, packs->nr + 1, packs->alloc);
460 packs->info[packs->nr].p = add_packed_git(full_path,
461 full_path_len,
464 if (!packs->info[packs->nr].p) {
465 warning(_("failed to add packfile '%s'"),
466 full_path);
467 return;
470 if (open_pack_index(packs->info[packs->nr].p)) {
471 warning(_("failed to open pack-index '%s'"),
472 full_path);
473 close_pack(packs->info[packs->nr].p);
474 FREE_AND_NULL(packs->info[packs->nr].p);
475 return;
478 packs->info[packs->nr].pack_name = xstrdup(file_name);
479 packs->info[packs->nr].orig_pack_int_id = packs->nr;
480 packs->info[packs->nr].expired = 0;
481 packs->nr++;
485 struct pack_midx_entry {
486 struct object_id oid;
487 uint32_t pack_int_id;
488 time_t pack_mtime;
489 uint64_t offset;
492 static int midx_oid_compare(const void *_a, const void *_b)
494 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
495 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
496 int cmp = oidcmp(&a->oid, &b->oid);
498 if (cmp)
499 return cmp;
501 if (a->pack_mtime > b->pack_mtime)
502 return -1;
503 else if (a->pack_mtime < b->pack_mtime)
504 return 1;
506 return a->pack_int_id - b->pack_int_id;
509 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
510 struct pack_midx_entry *e,
511 uint32_t pos)
513 if (pos >= m->num_objects)
514 return 1;
516 nth_midxed_object_oid(&e->oid, m, pos);
517 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
518 e->offset = nth_midxed_offset(m, pos);
520 /* consider objects in midx to be from "old" packs */
521 e->pack_mtime = 0;
522 return 0;
525 static void fill_pack_entry(uint32_t pack_int_id,
526 struct packed_git *p,
527 uint32_t cur_object,
528 struct pack_midx_entry *entry)
530 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
531 die(_("failed to locate object %d in packfile"), cur_object);
533 entry->pack_int_id = pack_int_id;
534 entry->pack_mtime = p->mtime;
536 entry->offset = nth_packed_object_offset(p, cur_object);
540 * It is possible to artificially get into a state where there are many
541 * duplicate copies of objects. That can create high memory pressure if
542 * we are to create a list of all objects before de-duplication. To reduce
543 * this memory pressure without a significant performance drop, automatically
544 * group objects by the first byte of their object id. Use the IDX fanout
545 * tables to group the data, copy to a local array, then sort.
547 * Copy only the de-duplicated entries (selected by most-recent modified time
548 * of a packfile containing the object).
550 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
551 struct pack_info *info,
552 uint32_t nr_packs,
553 uint32_t *nr_objects)
555 uint32_t cur_fanout, cur_pack, cur_object;
556 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
557 struct pack_midx_entry *entries_by_fanout = NULL;
558 struct pack_midx_entry *deduplicated_entries = NULL;
559 uint32_t start_pack = m ? m->num_packs : 0;
561 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
562 total_objects += info[cur_pack].p->num_objects;
565 * As we de-duplicate by fanout value, we expect the fanout
566 * slices to be evenly distributed, with some noise. Hence,
567 * allocate slightly more than one 256th.
569 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
571 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
572 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
573 *nr_objects = 0;
575 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
576 uint32_t nr_fanout = 0;
578 if (m) {
579 uint32_t start = 0, end;
581 if (cur_fanout)
582 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
583 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
585 for (cur_object = start; cur_object < end; cur_object++) {
586 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
587 nth_midxed_pack_midx_entry(m,
588 &entries_by_fanout[nr_fanout],
589 cur_object);
590 nr_fanout++;
594 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
595 uint32_t start = 0, end;
597 if (cur_fanout)
598 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
599 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
601 for (cur_object = start; cur_object < end; cur_object++) {
602 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
603 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
604 nr_fanout++;
608 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
611 * The batch is now sorted by OID and then mtime (descending).
612 * Take only the first duplicate.
614 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
615 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
616 &entries_by_fanout[cur_object].oid))
617 continue;
619 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
620 memcpy(&deduplicated_entries[*nr_objects],
621 &entries_by_fanout[cur_object],
622 sizeof(struct pack_midx_entry));
623 (*nr_objects)++;
627 free(entries_by_fanout);
628 return deduplicated_entries;
631 static size_t write_midx_pack_names(struct hashfile *f,
632 struct pack_info *info,
633 uint32_t num_packs)
635 uint32_t i;
636 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
637 size_t written = 0;
639 for (i = 0; i < num_packs; i++) {
640 size_t writelen;
642 if (info[i].expired)
643 continue;
645 if (i && strcmp(info[i].pack_name, info[i - 1].pack_name) <= 0)
646 BUG("incorrect pack-file order: %s before %s",
647 info[i - 1].pack_name,
648 info[i].pack_name);
650 writelen = strlen(info[i].pack_name) + 1;
651 hashwrite(f, info[i].pack_name, writelen);
652 written += writelen;
655 /* add padding to be aligned */
656 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
657 if (i < MIDX_CHUNK_ALIGNMENT) {
658 memset(padding, 0, sizeof(padding));
659 hashwrite(f, padding, i);
660 written += i;
663 return written;
666 static size_t write_midx_oid_fanout(struct hashfile *f,
667 struct pack_midx_entry *objects,
668 uint32_t nr_objects)
670 struct pack_midx_entry *list = objects;
671 struct pack_midx_entry *last = objects + nr_objects;
672 uint32_t count = 0;
673 uint32_t i;
676 * Write the first-level table (the list is sorted,
677 * but we use a 256-entry lookup to be able to avoid
678 * having to do eight extra binary search iterations).
680 for (i = 0; i < 256; i++) {
681 struct pack_midx_entry *next = list;
683 while (next < last && next->oid.hash[0] == i) {
684 count++;
685 next++;
688 hashwrite_be32(f, count);
689 list = next;
692 return MIDX_CHUNK_FANOUT_SIZE;
695 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
696 struct pack_midx_entry *objects,
697 uint32_t nr_objects)
699 struct pack_midx_entry *list = objects;
700 uint32_t i;
701 size_t written = 0;
703 for (i = 0; i < nr_objects; i++) {
704 struct pack_midx_entry *obj = list++;
706 if (i < nr_objects - 1) {
707 struct pack_midx_entry *next = list;
708 if (oidcmp(&obj->oid, &next->oid) >= 0)
709 BUG("OIDs not in order: %s >= %s",
710 oid_to_hex(&obj->oid),
711 oid_to_hex(&next->oid));
714 hashwrite(f, obj->oid.hash, (int)hash_len);
715 written += hash_len;
718 return written;
721 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
722 uint32_t *perm,
723 struct pack_midx_entry *objects, uint32_t nr_objects)
725 struct pack_midx_entry *list = objects;
726 uint32_t i, nr_large_offset = 0;
727 size_t written = 0;
729 for (i = 0; i < nr_objects; i++) {
730 struct pack_midx_entry *obj = list++;
732 if (perm[obj->pack_int_id] == PACK_EXPIRED)
733 BUG("object %s is in an expired pack with int-id %d",
734 oid_to_hex(&obj->oid),
735 obj->pack_int_id);
737 hashwrite_be32(f, perm[obj->pack_int_id]);
739 if (large_offset_needed && obj->offset >> 31)
740 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
741 else if (!large_offset_needed && obj->offset >> 32)
742 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
743 oid_to_hex(&obj->oid),
744 obj->offset);
745 else
746 hashwrite_be32(f, (uint32_t)obj->offset);
748 written += MIDX_CHUNK_OFFSET_WIDTH;
751 return written;
754 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
755 struct pack_midx_entry *objects, uint32_t nr_objects)
757 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
758 size_t written = 0;
760 while (nr_large_offset) {
761 struct pack_midx_entry *obj;
762 uint64_t offset;
764 if (list >= end)
765 BUG("too many large-offset objects");
767 obj = list++;
768 offset = obj->offset;
770 if (!(offset >> 31))
771 continue;
773 hashwrite_be32(f, offset >> 32);
774 hashwrite_be32(f, offset & 0xffffffffUL);
775 written += 2 * sizeof(uint32_t);
777 nr_large_offset--;
780 return written;
783 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
784 struct string_list *packs_to_drop, unsigned flags)
786 unsigned char cur_chunk, num_chunks = 0;
787 char *midx_name;
788 uint32_t i;
789 struct hashfile *f = NULL;
790 struct lock_file lk;
791 struct pack_list packs;
792 uint32_t *pack_perm = NULL;
793 uint64_t written = 0;
794 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
795 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
796 uint32_t nr_entries, num_large_offsets = 0;
797 struct pack_midx_entry *entries = NULL;
798 struct progress *progress = NULL;
799 int large_offsets_needed = 0;
800 int pack_name_concat_len = 0;
801 int dropped_packs = 0;
802 int result = 0;
804 midx_name = get_midx_filename(object_dir);
805 if (safe_create_leading_directories(midx_name)) {
806 UNLEAK(midx_name);
807 die_errno(_("unable to create leading directories of %s"),
808 midx_name);
811 if (m)
812 packs.m = m;
813 else
814 packs.m = load_multi_pack_index(object_dir, 1);
816 packs.nr = 0;
817 packs.alloc = packs.m ? packs.m->num_packs : 16;
818 packs.info = NULL;
819 ALLOC_ARRAY(packs.info, packs.alloc);
821 if (packs.m) {
822 for (i = 0; i < packs.m->num_packs; i++) {
823 ALLOC_GROW(packs.info, packs.nr + 1, packs.alloc);
825 packs.info[packs.nr].orig_pack_int_id = i;
826 packs.info[packs.nr].pack_name = xstrdup(packs.m->pack_names[i]);
827 packs.info[packs.nr].p = NULL;
828 packs.info[packs.nr].expired = 0;
829 packs.nr++;
833 packs.pack_paths_checked = 0;
834 if (flags & MIDX_PROGRESS)
835 packs.progress = start_progress(_("Adding packfiles to multi-pack-index"), 0);
836 else
837 packs.progress = NULL;
839 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
840 stop_progress(&packs.progress);
842 if (packs.m && packs.nr == packs.m->num_packs && !packs_to_drop)
843 goto cleanup;
845 entries = get_sorted_entries(packs.m, packs.info, packs.nr, &nr_entries);
847 for (i = 0; i < nr_entries; i++) {
848 if (entries[i].offset > 0x7fffffff)
849 num_large_offsets++;
850 if (entries[i].offset > 0xffffffff)
851 large_offsets_needed = 1;
854 QSORT(packs.info, packs.nr, pack_info_compare);
856 if (packs_to_drop && packs_to_drop->nr) {
857 int drop_index = 0;
858 int missing_drops = 0;
860 for (i = 0; i < packs.nr && drop_index < packs_to_drop->nr; i++) {
861 int cmp = strcmp(packs.info[i].pack_name,
862 packs_to_drop->items[drop_index].string);
864 if (!cmp) {
865 drop_index++;
866 packs.info[i].expired = 1;
867 } else if (cmp > 0) {
868 error(_("did not see pack-file %s to drop"),
869 packs_to_drop->items[drop_index].string);
870 drop_index++;
871 missing_drops++;
872 i--;
873 } else {
874 packs.info[i].expired = 0;
878 if (missing_drops) {
879 result = 1;
880 goto cleanup;
885 * pack_perm stores a permutation between pack-int-ids from the
886 * previous multi-pack-index to the new one we are writing:
888 * pack_perm[old_id] = new_id
890 ALLOC_ARRAY(pack_perm, packs.nr);
891 for (i = 0; i < packs.nr; i++) {
892 if (packs.info[i].expired) {
893 dropped_packs++;
894 pack_perm[packs.info[i].orig_pack_int_id] = PACK_EXPIRED;
895 } else {
896 pack_perm[packs.info[i].orig_pack_int_id] = i - dropped_packs;
900 for (i = 0; i < packs.nr; i++) {
901 if (!packs.info[i].expired)
902 pack_name_concat_len += strlen(packs.info[i].pack_name) + 1;
905 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
906 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
907 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
909 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
910 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
911 FREE_AND_NULL(midx_name);
913 if (packs.m)
914 close_midx(packs.m);
916 cur_chunk = 0;
917 num_chunks = large_offsets_needed ? 5 : 4;
919 if (packs.nr - dropped_packs == 0) {
920 error(_("no pack files to index."));
921 result = 1;
922 goto cleanup;
925 written = write_midx_header(f, num_chunks, packs.nr - dropped_packs);
927 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
928 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
930 cur_chunk++;
931 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
932 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + pack_name_concat_len;
934 cur_chunk++;
935 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
936 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
938 cur_chunk++;
939 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
940 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * the_hash_algo->rawsz;
942 cur_chunk++;
943 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
944 if (large_offsets_needed) {
945 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
947 cur_chunk++;
948 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
949 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
952 chunk_ids[cur_chunk] = 0;
954 for (i = 0; i <= num_chunks; i++) {
955 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
956 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
957 chunk_offsets[i - 1],
958 chunk_offsets[i]);
960 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
961 BUG("chunk offset %"PRIu64" is not properly aligned",
962 chunk_offsets[i]);
964 hashwrite_be32(f, chunk_ids[i]);
965 hashwrite_be32(f, chunk_offsets[i] >> 32);
966 hashwrite_be32(f, chunk_offsets[i]);
968 written += MIDX_CHUNKLOOKUP_WIDTH;
971 if (flags & MIDX_PROGRESS)
972 progress = start_progress(_("Writing chunks to multi-pack-index"),
973 num_chunks);
974 for (i = 0; i < num_chunks; i++) {
975 if (written != chunk_offsets[i])
976 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
977 chunk_offsets[i],
978 written,
979 chunk_ids[i]);
981 switch (chunk_ids[i]) {
982 case MIDX_CHUNKID_PACKNAMES:
983 written += write_midx_pack_names(f, packs.info, packs.nr);
984 break;
986 case MIDX_CHUNKID_OIDFANOUT:
987 written += write_midx_oid_fanout(f, entries, nr_entries);
988 break;
990 case MIDX_CHUNKID_OIDLOOKUP:
991 written += write_midx_oid_lookup(f, the_hash_algo->rawsz, entries, nr_entries);
992 break;
994 case MIDX_CHUNKID_OBJECTOFFSETS:
995 written += write_midx_object_offsets(f, large_offsets_needed, pack_perm, entries, nr_entries);
996 break;
998 case MIDX_CHUNKID_LARGEOFFSETS:
999 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
1000 break;
1002 default:
1003 BUG("trying to write unknown chunk id %"PRIx32,
1004 chunk_ids[i]);
1007 display_progress(progress, i + 1);
1009 stop_progress(&progress);
1011 if (written != chunk_offsets[num_chunks])
1012 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
1013 written,
1014 chunk_offsets[num_chunks]);
1016 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1017 commit_lock_file(&lk);
1019 cleanup:
1020 for (i = 0; i < packs.nr; i++) {
1021 if (packs.info[i].p) {
1022 close_pack(packs.info[i].p);
1023 free(packs.info[i].p);
1025 free(packs.info[i].pack_name);
1028 free(packs.info);
1029 free(entries);
1030 free(pack_perm);
1031 free(midx_name);
1032 return result;
1035 int write_midx_file(const char *object_dir, unsigned flags)
1037 return write_midx_internal(object_dir, NULL, NULL, flags);
1040 void clear_midx_file(struct repository *r)
1042 char *midx = get_midx_filename(r->objects->odb->path);
1044 if (r->objects && r->objects->multi_pack_index) {
1045 close_midx(r->objects->multi_pack_index);
1046 r->objects->multi_pack_index = NULL;
1049 if (remove_path(midx)) {
1050 UNLEAK(midx);
1051 die(_("failed to clear multi-pack-index at %s"), midx);
1054 free(midx);
1057 static int verify_midx_error;
1059 static void midx_report(const char *fmt, ...)
1061 va_list ap;
1062 verify_midx_error = 1;
1063 va_start(ap, fmt);
1064 vfprintf(stderr, fmt, ap);
1065 fprintf(stderr, "\n");
1066 va_end(ap);
1069 struct pair_pos_vs_id
1071 uint32_t pos;
1072 uint32_t pack_int_id;
1075 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1077 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1078 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1080 return b->pack_int_id - a->pack_int_id;
1084 * Limit calls to display_progress() for performance reasons.
1085 * The interval here was arbitrarily chosen.
1087 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1088 #define midx_display_sparse_progress(progress, n) \
1089 do { \
1090 uint64_t _n = (n); \
1091 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1092 display_progress(progress, _n); \
1093 } while (0)
1095 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1097 struct pair_pos_vs_id *pairs = NULL;
1098 uint32_t i;
1099 struct progress *progress = NULL;
1100 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1101 verify_midx_error = 0;
1103 if (!m)
1104 return 0;
1106 if (flags & MIDX_PROGRESS)
1107 progress = start_progress(_("Looking for referenced packfiles"),
1108 m->num_packs);
1109 for (i = 0; i < m->num_packs; i++) {
1110 if (prepare_midx_pack(r, m, i))
1111 midx_report("failed to load pack in position %d", i);
1113 display_progress(progress, i + 1);
1115 stop_progress(&progress);
1117 for (i = 0; i < 255; i++) {
1118 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1119 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1121 if (oid_fanout1 > oid_fanout2)
1122 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1123 i, oid_fanout1, oid_fanout2, i + 1);
1126 if (m->num_objects == 0) {
1127 midx_report(_("the midx contains no oid"));
1129 * Remaining tests assume that we have objects, so we can
1130 * return here.
1132 return verify_midx_error;
1135 if (flags & MIDX_PROGRESS)
1136 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1137 m->num_objects - 1);
1138 for (i = 0; i < m->num_objects - 1; i++) {
1139 struct object_id oid1, oid2;
1141 nth_midxed_object_oid(&oid1, m, i);
1142 nth_midxed_object_oid(&oid2, m, i + 1);
1144 if (oidcmp(&oid1, &oid2) >= 0)
1145 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1146 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1148 midx_display_sparse_progress(progress, i + 1);
1150 stop_progress(&progress);
1153 * Create an array mapping each object to its packfile id. Sort it
1154 * to group the objects by packfile. Use this permutation to visit
1155 * each of the objects and only require 1 packfile to be open at a
1156 * time.
1158 ALLOC_ARRAY(pairs, m->num_objects);
1159 for (i = 0; i < m->num_objects; i++) {
1160 pairs[i].pos = i;
1161 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1164 if (flags & MIDX_PROGRESS)
1165 progress = start_sparse_progress(_("Sorting objects by packfile"),
1166 m->num_objects);
1167 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1168 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1169 stop_progress(&progress);
1171 if (flags & MIDX_PROGRESS)
1172 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1173 for (i = 0; i < m->num_objects; i++) {
1174 struct object_id oid;
1175 struct pack_entry e;
1176 off_t m_offset, p_offset;
1178 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1179 m->packs[pairs[i-1].pack_int_id])
1181 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1182 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1185 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1187 if (!fill_midx_entry(r, &oid, &e, m)) {
1188 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1189 pairs[i].pos, oid_to_hex(&oid));
1190 continue;
1193 if (open_pack_index(e.p)) {
1194 midx_report(_("failed to load pack-index for packfile %s"),
1195 e.p->pack_name);
1196 break;
1199 m_offset = e.offset;
1200 p_offset = find_pack_entry_one(oid.hash, e.p);
1202 if (m_offset != p_offset)
1203 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1204 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1206 midx_display_sparse_progress(progress, i + 1);
1208 stop_progress(&progress);
1210 free(pairs);
1212 return verify_midx_error;
1215 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1217 uint32_t i, *count, result = 0;
1218 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1219 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1220 struct progress *progress = NULL;
1222 if (!m)
1223 return 0;
1225 count = xcalloc(m->num_packs, sizeof(uint32_t));
1227 if (flags & MIDX_PROGRESS)
1228 progress = start_progress(_("Counting referenced objects"),
1229 m->num_objects);
1230 for (i = 0; i < m->num_objects; i++) {
1231 int pack_int_id = nth_midxed_pack_int_id(m, i);
1232 count[pack_int_id]++;
1233 display_progress(progress, i + 1);
1235 stop_progress(&progress);
1237 if (flags & MIDX_PROGRESS)
1238 progress = start_progress(_("Finding and deleting unreferenced packfiles"),
1239 m->num_packs);
1240 for (i = 0; i < m->num_packs; i++) {
1241 char *pack_name;
1242 display_progress(progress, i + 1);
1244 if (count[i])
1245 continue;
1247 if (prepare_midx_pack(r, m, i))
1248 continue;
1250 if (m->packs[i]->pack_keep)
1251 continue;
1253 pack_name = xstrdup(m->packs[i]->pack_name);
1254 close_pack(m->packs[i]);
1256 string_list_insert(&packs_to_drop, m->pack_names[i]);
1257 unlink_pack_path(pack_name, 0);
1258 free(pack_name);
1260 stop_progress(&progress);
1262 free(count);
1264 if (packs_to_drop.nr)
1265 result = write_midx_internal(object_dir, m, &packs_to_drop, flags);
1267 string_list_clear(&packs_to_drop, 0);
1268 return result;
1271 struct repack_info {
1272 timestamp_t mtime;
1273 uint32_t referenced_objects;
1274 uint32_t pack_int_id;
1277 static int compare_by_mtime(const void *a_, const void *b_)
1279 const struct repack_info *a, *b;
1281 a = (const struct repack_info *)a_;
1282 b = (const struct repack_info *)b_;
1284 if (a->mtime < b->mtime)
1285 return -1;
1286 if (a->mtime > b->mtime)
1287 return 1;
1288 return 0;
1291 static int fill_included_packs_all(struct repository *r,
1292 struct multi_pack_index *m,
1293 unsigned char *include_pack)
1295 uint32_t i, count = 0;
1296 int pack_kept_objects = 0;
1298 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1300 for (i = 0; i < m->num_packs; i++) {
1301 if (prepare_midx_pack(r, m, i))
1302 continue;
1303 if (!pack_kept_objects && m->packs[i]->pack_keep)
1304 continue;
1306 include_pack[i] = 1;
1307 count++;
1310 return count < 2;
1313 static int fill_included_packs_batch(struct repository *r,
1314 struct multi_pack_index *m,
1315 unsigned char *include_pack,
1316 size_t batch_size)
1318 uint32_t i, packs_to_repack;
1319 size_t total_size;
1320 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1321 int pack_kept_objects = 0;
1323 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1325 for (i = 0; i < m->num_packs; i++) {
1326 pack_info[i].pack_int_id = i;
1328 if (prepare_midx_pack(r, m, i))
1329 continue;
1331 pack_info[i].mtime = m->packs[i]->mtime;
1334 for (i = 0; batch_size && i < m->num_objects; i++) {
1335 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1336 pack_info[pack_int_id].referenced_objects++;
1339 QSORT(pack_info, m->num_packs, compare_by_mtime);
1341 total_size = 0;
1342 packs_to_repack = 0;
1343 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1344 int pack_int_id = pack_info[i].pack_int_id;
1345 struct packed_git *p = m->packs[pack_int_id];
1346 size_t expected_size;
1348 if (!p)
1349 continue;
1350 if (!pack_kept_objects && p->pack_keep)
1351 continue;
1352 if (open_pack_index(p) || !p->num_objects)
1353 continue;
1355 expected_size = (size_t)(p->pack_size
1356 * pack_info[i].referenced_objects);
1357 expected_size /= p->num_objects;
1359 if (expected_size >= batch_size)
1360 continue;
1362 packs_to_repack++;
1363 total_size += expected_size;
1364 include_pack[pack_int_id] = 1;
1367 free(pack_info);
1369 if (total_size < batch_size || packs_to_repack < 2)
1370 return 1;
1372 return 0;
1375 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1377 int result = 0;
1378 uint32_t i;
1379 unsigned char *include_pack;
1380 struct child_process cmd = CHILD_PROCESS_INIT;
1381 struct strbuf base_name = STRBUF_INIT;
1382 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1385 * When updating the default for these configuration
1386 * variables in builtin/repack.c, these must be adjusted
1387 * to match.
1389 int delta_base_offset = 1;
1390 int use_delta_islands = 0;
1392 if (!m)
1393 return 0;
1395 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1397 if (batch_size) {
1398 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1399 goto cleanup;
1400 } else if (fill_included_packs_all(r, m, include_pack))
1401 goto cleanup;
1403 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1404 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1406 strvec_push(&cmd.args, "pack-objects");
1408 strbuf_addstr(&base_name, object_dir);
1409 strbuf_addstr(&base_name, "/pack/pack");
1410 strvec_push(&cmd.args, base_name.buf);
1412 if (delta_base_offset)
1413 strvec_push(&cmd.args, "--delta-base-offset");
1414 if (use_delta_islands)
1415 strvec_push(&cmd.args, "--delta-islands");
1417 if (flags & MIDX_PROGRESS)
1418 strvec_push(&cmd.args, "--progress");
1419 else
1420 strvec_push(&cmd.args, "-q");
1422 strbuf_release(&base_name);
1424 cmd.git_cmd = 1;
1425 cmd.in = cmd.out = -1;
1427 if (start_command(&cmd)) {
1428 error(_("could not start pack-objects"));
1429 result = 1;
1430 goto cleanup;
1433 for (i = 0; i < m->num_objects; i++) {
1434 struct object_id oid;
1435 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1437 if (!include_pack[pack_int_id])
1438 continue;
1440 nth_midxed_object_oid(&oid, m, i);
1441 xwrite(cmd.in, oid_to_hex(&oid), the_hash_algo->hexsz);
1442 xwrite(cmd.in, "\n", 1);
1444 close(cmd.in);
1446 if (finish_command(&cmd)) {
1447 error(_("could not finish pack-objects"));
1448 result = 1;
1449 goto cleanup;
1452 result = write_midx_internal(object_dir, m, NULL, flags);
1453 m = NULL;
1455 cleanup:
1456 if (m)
1457 close_midx(m);
1458 free(include_pack);
1459 return result;