Merge branch 'fd/gitweb-snapshot-conf-doc-fix'
[git.git] / midx.c
blob8a505fd423efccf9986079542b47967cc93df7dc
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"
12 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
13 #define MIDX_VERSION 1
14 #define MIDX_BYTE_FILE_VERSION 4
15 #define MIDX_BYTE_HASH_VERSION 5
16 #define MIDX_BYTE_NUM_CHUNKS 6
17 #define MIDX_BYTE_NUM_PACKS 8
18 #define MIDX_HASH_VERSION 1
19 #define MIDX_HEADER_SIZE 12
20 #define MIDX_HASH_LEN 20
21 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + MIDX_HASH_LEN)
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 static char *get_midx_filename(const char *object_dir)
38 return xstrfmt("%s/pack/multi-pack-index", object_dir);
41 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
43 struct multi_pack_index *m = NULL;
44 int fd;
45 struct stat st;
46 size_t midx_size;
47 void *midx_map = NULL;
48 uint32_t hash_version;
49 char *midx_name = get_midx_filename(object_dir);
50 uint32_t i;
51 const char *cur_pack_name;
53 fd = git_open(midx_name);
55 if (fd < 0)
56 goto cleanup_fail;
57 if (fstat(fd, &st)) {
58 error_errno(_("failed to read %s"), midx_name);
59 goto cleanup_fail;
62 midx_size = xsize_t(st.st_size);
64 if (midx_size < MIDX_MIN_SIZE) {
65 error(_("multi-pack-index file %s is too small"), midx_name);
66 goto cleanup_fail;
69 FREE_AND_NULL(midx_name);
71 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
73 FLEX_ALLOC_MEM(m, object_dir, object_dir, strlen(object_dir));
74 m->fd = fd;
75 m->data = midx_map;
76 m->data_len = midx_size;
77 m->local = local;
79 m->signature = get_be32(m->data);
80 if (m->signature != MIDX_SIGNATURE)
81 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
82 m->signature, MIDX_SIGNATURE);
84 m->version = m->data[MIDX_BYTE_FILE_VERSION];
85 if (m->version != MIDX_VERSION)
86 die(_("multi-pack-index version %d not recognized"),
87 m->version);
89 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
90 if (hash_version != MIDX_HASH_VERSION)
91 die(_("hash version %u does not match"), hash_version);
92 m->hash_len = MIDX_HASH_LEN;
94 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
96 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
98 for (i = 0; i < m->num_chunks; i++) {
99 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
100 MIDX_CHUNKLOOKUP_WIDTH * i);
101 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
102 MIDX_CHUNKLOOKUP_WIDTH * i);
104 if (chunk_offset >= m->data_len)
105 die(_("invalid chunk offset (too large)"));
107 switch (chunk_id) {
108 case MIDX_CHUNKID_PACKNAMES:
109 m->chunk_pack_names = m->data + chunk_offset;
110 break;
112 case MIDX_CHUNKID_OIDFANOUT:
113 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
114 break;
116 case MIDX_CHUNKID_OIDLOOKUP:
117 m->chunk_oid_lookup = m->data + chunk_offset;
118 break;
120 case MIDX_CHUNKID_OBJECTOFFSETS:
121 m->chunk_object_offsets = m->data + chunk_offset;
122 break;
124 case MIDX_CHUNKID_LARGEOFFSETS:
125 m->chunk_large_offsets = m->data + chunk_offset;
126 break;
128 case 0:
129 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
130 break;
132 default:
134 * Do nothing on unrecognized chunks, allowing future
135 * extensions to add optional chunks.
137 break;
141 if (!m->chunk_pack_names)
142 die(_("multi-pack-index missing required pack-name chunk"));
143 if (!m->chunk_oid_fanout)
144 die(_("multi-pack-index missing required OID fanout chunk"));
145 if (!m->chunk_oid_lookup)
146 die(_("multi-pack-index missing required OID lookup chunk"));
147 if (!m->chunk_object_offsets)
148 die(_("multi-pack-index missing required object offsets chunk"));
150 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
152 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
153 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
155 cur_pack_name = (const char *)m->chunk_pack_names;
156 for (i = 0; i < m->num_packs; i++) {
157 m->pack_names[i] = cur_pack_name;
159 cur_pack_name += strlen(cur_pack_name) + 1;
161 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
162 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
163 m->pack_names[i - 1],
164 m->pack_names[i]);
167 return m;
169 cleanup_fail:
170 free(m);
171 free(midx_name);
172 if (midx_map)
173 munmap(midx_map, midx_size);
174 if (0 <= fd)
175 close(fd);
176 return NULL;
179 void close_midx(struct multi_pack_index *m)
181 uint32_t i;
183 if (!m)
184 return;
186 munmap((unsigned char *)m->data, m->data_len);
187 close(m->fd);
188 m->fd = -1;
190 for (i = 0; i < m->num_packs; i++) {
191 if (m->packs[i]) {
192 close_pack(m->packs[i]);
193 free(m->packs[i]);
196 FREE_AND_NULL(m->packs);
197 FREE_AND_NULL(m->pack_names);
200 int prepare_midx_pack(struct multi_pack_index *m, uint32_t pack_int_id)
202 struct strbuf pack_name = STRBUF_INIT;
204 if (pack_int_id >= m->num_packs)
205 die(_("bad pack-int-id: %u (%u total packs)"),
206 pack_int_id, m->num_packs);
208 if (m->packs[pack_int_id])
209 return 0;
211 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
212 m->pack_names[pack_int_id]);
214 m->packs[pack_int_id] = add_packed_git(pack_name.buf, pack_name.len, m->local);
215 strbuf_release(&pack_name);
216 return !m->packs[pack_int_id];
219 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
221 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
222 MIDX_HASH_LEN, result);
225 struct object_id *nth_midxed_object_oid(struct object_id *oid,
226 struct multi_pack_index *m,
227 uint32_t n)
229 if (n >= m->num_objects)
230 return NULL;
232 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
233 return oid;
236 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
238 const unsigned char *offset_data;
239 uint32_t offset32;
241 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
242 offset32 = get_be32(offset_data + sizeof(uint32_t));
244 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
245 if (sizeof(off_t) < sizeof(uint64_t))
246 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
248 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
249 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
252 return offset32;
255 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
257 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
260 static int nth_midxed_pack_entry(struct multi_pack_index *m, struct pack_entry *e, uint32_t pos)
262 uint32_t pack_int_id;
263 struct packed_git *p;
265 if (pos >= m->num_objects)
266 return 0;
268 pack_int_id = nth_midxed_pack_int_id(m, pos);
270 if (prepare_midx_pack(m, pack_int_id))
271 die(_("error preparing packfile from multi-pack-index"));
272 p = m->packs[pack_int_id];
275 * We are about to tell the caller where they can locate the
276 * requested object. We better make sure the packfile is
277 * still here and can be accessed before supplying that
278 * answer, as it may have been deleted since the MIDX was
279 * loaded!
281 if (!is_pack_valid(p))
282 return 0;
284 if (p->num_bad_objects) {
285 uint32_t i;
286 struct object_id oid;
287 nth_midxed_object_oid(&oid, m, pos);
288 for (i = 0; i < p->num_bad_objects; i++)
289 if (hasheq(oid.hash,
290 p->bad_object_sha1 + the_hash_algo->rawsz * i))
291 return 0;
294 e->offset = nth_midxed_offset(m, pos);
295 e->p = p;
297 return 1;
300 int fill_midx_entry(const struct object_id *oid, struct pack_entry *e, struct multi_pack_index *m)
302 uint32_t pos;
304 if (!bsearch_midx(oid, m, &pos))
305 return 0;
307 return nth_midxed_pack_entry(m, e, pos);
310 int midx_contains_pack(struct multi_pack_index *m, const char *idx_name)
312 uint32_t first = 0, last = m->num_packs;
314 while (first < last) {
315 uint32_t mid = first + (last - first) / 2;
316 const char *current;
317 int cmp;
319 current = m->pack_names[mid];
320 cmp = strcmp(idx_name, current);
321 if (!cmp)
322 return 1;
323 if (cmp > 0) {
324 first = mid + 1;
325 continue;
327 last = mid;
330 return 0;
333 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
335 struct multi_pack_index *m;
336 struct multi_pack_index *m_search;
337 int config_value;
338 static int env_value = -1;
340 if (env_value < 0)
341 env_value = git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0);
343 if (!env_value &&
344 (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
345 !config_value))
346 return 0;
348 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
349 if (!strcmp(object_dir, m_search->object_dir))
350 return 1;
352 m = load_multi_pack_index(object_dir, local);
354 if (m) {
355 m->next = r->objects->multi_pack_index;
356 r->objects->multi_pack_index = m;
357 return 1;
360 return 0;
363 static size_t write_midx_header(struct hashfile *f,
364 unsigned char num_chunks,
365 uint32_t num_packs)
367 unsigned char byte_values[4];
369 hashwrite_be32(f, MIDX_SIGNATURE);
370 byte_values[0] = MIDX_VERSION;
371 byte_values[1] = MIDX_HASH_VERSION;
372 byte_values[2] = num_chunks;
373 byte_values[3] = 0; /* unused */
374 hashwrite(f, byte_values, sizeof(byte_values));
375 hashwrite_be32(f, num_packs);
377 return MIDX_HEADER_SIZE;
380 struct pack_list {
381 struct packed_git **list;
382 char **names;
383 uint32_t nr;
384 uint32_t alloc_list;
385 uint32_t alloc_names;
386 size_t pack_name_concat_len;
387 struct multi_pack_index *m;
390 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
391 const char *file_name, void *data)
393 struct pack_list *packs = (struct pack_list *)data;
395 if (ends_with(file_name, ".idx")) {
396 if (packs->m && midx_contains_pack(packs->m, file_name))
397 return;
399 ALLOC_GROW(packs->list, packs->nr + 1, packs->alloc_list);
400 ALLOC_GROW(packs->names, packs->nr + 1, packs->alloc_names);
402 packs->list[packs->nr] = add_packed_git(full_path,
403 full_path_len,
406 if (!packs->list[packs->nr]) {
407 warning(_("failed to add packfile '%s'"),
408 full_path);
409 return;
412 if (open_pack_index(packs->list[packs->nr])) {
413 warning(_("failed to open pack-index '%s'"),
414 full_path);
415 close_pack(packs->list[packs->nr]);
416 FREE_AND_NULL(packs->list[packs->nr]);
417 return;
420 packs->names[packs->nr] = xstrdup(file_name);
421 packs->pack_name_concat_len += strlen(file_name) + 1;
422 packs->nr++;
426 struct pack_pair {
427 uint32_t pack_int_id;
428 char *pack_name;
431 static int pack_pair_compare(const void *_a, const void *_b)
433 struct pack_pair *a = (struct pack_pair *)_a;
434 struct pack_pair *b = (struct pack_pair *)_b;
435 return strcmp(a->pack_name, b->pack_name);
438 static void sort_packs_by_name(char **pack_names, uint32_t nr_packs, uint32_t *perm)
440 uint32_t i;
441 struct pack_pair *pairs;
443 ALLOC_ARRAY(pairs, nr_packs);
445 for (i = 0; i < nr_packs; i++) {
446 pairs[i].pack_int_id = i;
447 pairs[i].pack_name = pack_names[i];
450 QSORT(pairs, nr_packs, pack_pair_compare);
452 for (i = 0; i < nr_packs; i++) {
453 pack_names[i] = pairs[i].pack_name;
454 perm[pairs[i].pack_int_id] = i;
457 free(pairs);
460 struct pack_midx_entry {
461 struct object_id oid;
462 uint32_t pack_int_id;
463 time_t pack_mtime;
464 uint64_t offset;
467 static int midx_oid_compare(const void *_a, const void *_b)
469 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
470 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
471 int cmp = oidcmp(&a->oid, &b->oid);
473 if (cmp)
474 return cmp;
476 if (a->pack_mtime > b->pack_mtime)
477 return -1;
478 else if (a->pack_mtime < b->pack_mtime)
479 return 1;
481 return a->pack_int_id - b->pack_int_id;
484 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
485 uint32_t *pack_perm,
486 struct pack_midx_entry *e,
487 uint32_t pos)
489 if (pos >= m->num_objects)
490 return 1;
492 nth_midxed_object_oid(&e->oid, m, pos);
493 e->pack_int_id = pack_perm[nth_midxed_pack_int_id(m, pos)];
494 e->offset = nth_midxed_offset(m, pos);
496 /* consider objects in midx to be from "old" packs */
497 e->pack_mtime = 0;
498 return 0;
501 static void fill_pack_entry(uint32_t pack_int_id,
502 struct packed_git *p,
503 uint32_t cur_object,
504 struct pack_midx_entry *entry)
506 if (!nth_packed_object_oid(&entry->oid, p, cur_object))
507 die(_("failed to locate object %d in packfile"), cur_object);
509 entry->pack_int_id = pack_int_id;
510 entry->pack_mtime = p->mtime;
512 entry->offset = nth_packed_object_offset(p, cur_object);
516 * It is possible to artificially get into a state where there are many
517 * duplicate copies of objects. That can create high memory pressure if
518 * we are to create a list of all objects before de-duplication. To reduce
519 * this memory pressure without a significant performance drop, automatically
520 * group objects by the first byte of their object id. Use the IDX fanout
521 * tables to group the data, copy to a local array, then sort.
523 * Copy only the de-duplicated entries (selected by most-recent modified time
524 * of a packfile containing the object).
526 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
527 struct packed_git **p,
528 uint32_t *perm,
529 uint32_t nr_packs,
530 uint32_t *nr_objects)
532 uint32_t cur_fanout, cur_pack, cur_object;
533 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
534 struct pack_midx_entry *entries_by_fanout = NULL;
535 struct pack_midx_entry *deduplicated_entries = NULL;
536 uint32_t start_pack = m ? m->num_packs : 0;
538 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
539 total_objects += p[cur_pack]->num_objects;
542 * As we de-duplicate by fanout value, we expect the fanout
543 * slices to be evenly distributed, with some noise. Hence,
544 * allocate slightly more than one 256th.
546 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
548 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
549 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
550 *nr_objects = 0;
552 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
553 uint32_t nr_fanout = 0;
555 if (m) {
556 uint32_t start = 0, end;
558 if (cur_fanout)
559 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
560 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
562 for (cur_object = start; cur_object < end; cur_object++) {
563 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
564 nth_midxed_pack_midx_entry(m, perm,
565 &entries_by_fanout[nr_fanout],
566 cur_object);
567 nr_fanout++;
571 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
572 uint32_t start = 0, end;
574 if (cur_fanout)
575 start = get_pack_fanout(p[cur_pack], cur_fanout - 1);
576 end = get_pack_fanout(p[cur_pack], cur_fanout);
578 for (cur_object = start; cur_object < end; cur_object++) {
579 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
580 fill_pack_entry(perm[cur_pack], p[cur_pack], cur_object, &entries_by_fanout[nr_fanout]);
581 nr_fanout++;
585 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
588 * The batch is now sorted by OID and then mtime (descending).
589 * Take only the first duplicate.
591 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
592 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
593 &entries_by_fanout[cur_object].oid))
594 continue;
596 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
597 memcpy(&deduplicated_entries[*nr_objects],
598 &entries_by_fanout[cur_object],
599 sizeof(struct pack_midx_entry));
600 (*nr_objects)++;
604 free(entries_by_fanout);
605 return deduplicated_entries;
608 static size_t write_midx_pack_names(struct hashfile *f,
609 char **pack_names,
610 uint32_t num_packs)
612 uint32_t i;
613 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
614 size_t written = 0;
616 for (i = 0; i < num_packs; i++) {
617 size_t writelen = strlen(pack_names[i]) + 1;
619 if (i && strcmp(pack_names[i], pack_names[i - 1]) <= 0)
620 BUG("incorrect pack-file order: %s before %s",
621 pack_names[i - 1],
622 pack_names[i]);
624 hashwrite(f, pack_names[i], writelen);
625 written += writelen;
628 /* add padding to be aligned */
629 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
630 if (i < MIDX_CHUNK_ALIGNMENT) {
631 memset(padding, 0, sizeof(padding));
632 hashwrite(f, padding, i);
633 written += i;
636 return written;
639 static size_t write_midx_oid_fanout(struct hashfile *f,
640 struct pack_midx_entry *objects,
641 uint32_t nr_objects)
643 struct pack_midx_entry *list = objects;
644 struct pack_midx_entry *last = objects + nr_objects;
645 uint32_t count = 0;
646 uint32_t i;
649 * Write the first-level table (the list is sorted,
650 * but we use a 256-entry lookup to be able to avoid
651 * having to do eight extra binary search iterations).
653 for (i = 0; i < 256; i++) {
654 struct pack_midx_entry *next = list;
656 while (next < last && next->oid.hash[0] == i) {
657 count++;
658 next++;
661 hashwrite_be32(f, count);
662 list = next;
665 return MIDX_CHUNK_FANOUT_SIZE;
668 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
669 struct pack_midx_entry *objects,
670 uint32_t nr_objects)
672 struct pack_midx_entry *list = objects;
673 uint32_t i;
674 size_t written = 0;
676 for (i = 0; i < nr_objects; i++) {
677 struct pack_midx_entry *obj = list++;
679 if (i < nr_objects - 1) {
680 struct pack_midx_entry *next = list;
681 if (oidcmp(&obj->oid, &next->oid) >= 0)
682 BUG("OIDs not in order: %s >= %s",
683 oid_to_hex(&obj->oid),
684 oid_to_hex(&next->oid));
687 hashwrite(f, obj->oid.hash, (int)hash_len);
688 written += hash_len;
691 return written;
694 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
695 struct pack_midx_entry *objects, uint32_t nr_objects)
697 struct pack_midx_entry *list = objects;
698 uint32_t i, nr_large_offset = 0;
699 size_t written = 0;
701 for (i = 0; i < nr_objects; i++) {
702 struct pack_midx_entry *obj = list++;
704 hashwrite_be32(f, obj->pack_int_id);
706 if (large_offset_needed && obj->offset >> 31)
707 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
708 else if (!large_offset_needed && obj->offset >> 32)
709 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
710 oid_to_hex(&obj->oid),
711 obj->offset);
712 else
713 hashwrite_be32(f, (uint32_t)obj->offset);
715 written += MIDX_CHUNK_OFFSET_WIDTH;
718 return written;
721 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
722 struct pack_midx_entry *objects, uint32_t nr_objects)
724 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
725 size_t written = 0;
727 while (nr_large_offset) {
728 struct pack_midx_entry *obj;
729 uint64_t offset;
731 if (list >= end)
732 BUG("too many large-offset objects");
734 obj = list++;
735 offset = obj->offset;
737 if (!(offset >> 31))
738 continue;
740 hashwrite_be32(f, offset >> 32);
741 hashwrite_be32(f, offset & 0xffffffffUL);
742 written += 2 * sizeof(uint32_t);
744 nr_large_offset--;
747 return written;
750 int write_midx_file(const char *object_dir)
752 unsigned char cur_chunk, num_chunks = 0;
753 char *midx_name;
754 uint32_t i;
755 struct hashfile *f = NULL;
756 struct lock_file lk;
757 struct pack_list packs;
758 uint32_t *pack_perm = NULL;
759 uint64_t written = 0;
760 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
761 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
762 uint32_t nr_entries, num_large_offsets = 0;
763 struct pack_midx_entry *entries = NULL;
764 int large_offsets_needed = 0;
766 midx_name = get_midx_filename(object_dir);
767 if (safe_create_leading_directories(midx_name)) {
768 UNLEAK(midx_name);
769 die_errno(_("unable to create leading directories of %s"),
770 midx_name);
773 packs.m = load_multi_pack_index(object_dir, 1);
775 packs.nr = 0;
776 packs.alloc_list = packs.m ? packs.m->num_packs : 16;
777 packs.alloc_names = packs.alloc_list;
778 packs.list = NULL;
779 packs.names = NULL;
780 packs.pack_name_concat_len = 0;
781 ALLOC_ARRAY(packs.list, packs.alloc_list);
782 ALLOC_ARRAY(packs.names, packs.alloc_names);
784 if (packs.m) {
785 for (i = 0; i < packs.m->num_packs; i++) {
786 ALLOC_GROW(packs.list, packs.nr + 1, packs.alloc_list);
787 ALLOC_GROW(packs.names, packs.nr + 1, packs.alloc_names);
789 packs.list[packs.nr] = NULL;
790 packs.names[packs.nr] = xstrdup(packs.m->pack_names[i]);
791 packs.pack_name_concat_len += strlen(packs.names[packs.nr]) + 1;
792 packs.nr++;
796 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
798 if (packs.m && packs.nr == packs.m->num_packs)
799 goto cleanup;
801 if (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
802 packs.pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
803 (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
805 ALLOC_ARRAY(pack_perm, packs.nr);
806 sort_packs_by_name(packs.names, packs.nr, pack_perm);
808 entries = get_sorted_entries(packs.m, packs.list, pack_perm, packs.nr, &nr_entries);
810 for (i = 0; i < nr_entries; i++) {
811 if (entries[i].offset > 0x7fffffff)
812 num_large_offsets++;
813 if (entries[i].offset > 0xffffffff)
814 large_offsets_needed = 1;
817 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
818 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
819 FREE_AND_NULL(midx_name);
821 if (packs.m)
822 close_midx(packs.m);
824 cur_chunk = 0;
825 num_chunks = large_offsets_needed ? 5 : 4;
827 written = write_midx_header(f, num_chunks, packs.nr);
829 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
830 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
832 cur_chunk++;
833 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
834 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + packs.pack_name_concat_len;
836 cur_chunk++;
837 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
838 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
840 cur_chunk++;
841 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
842 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_HASH_LEN;
844 cur_chunk++;
845 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
846 if (large_offsets_needed) {
847 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
849 cur_chunk++;
850 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
851 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
854 chunk_ids[cur_chunk] = 0;
856 for (i = 0; i <= num_chunks; i++) {
857 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
858 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
859 chunk_offsets[i - 1],
860 chunk_offsets[i]);
862 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
863 BUG("chunk offset %"PRIu64" is not properly aligned",
864 chunk_offsets[i]);
866 hashwrite_be32(f, chunk_ids[i]);
867 hashwrite_be32(f, chunk_offsets[i] >> 32);
868 hashwrite_be32(f, chunk_offsets[i]);
870 written += MIDX_CHUNKLOOKUP_WIDTH;
873 for (i = 0; i < num_chunks; i++) {
874 if (written != chunk_offsets[i])
875 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
876 chunk_offsets[i],
877 written,
878 chunk_ids[i]);
880 switch (chunk_ids[i]) {
881 case MIDX_CHUNKID_PACKNAMES:
882 written += write_midx_pack_names(f, packs.names, packs.nr);
883 break;
885 case MIDX_CHUNKID_OIDFANOUT:
886 written += write_midx_oid_fanout(f, entries, nr_entries);
887 break;
889 case MIDX_CHUNKID_OIDLOOKUP:
890 written += write_midx_oid_lookup(f, MIDX_HASH_LEN, entries, nr_entries);
891 break;
893 case MIDX_CHUNKID_OBJECTOFFSETS:
894 written += write_midx_object_offsets(f, large_offsets_needed, entries, nr_entries);
895 break;
897 case MIDX_CHUNKID_LARGEOFFSETS:
898 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
899 break;
901 default:
902 BUG("trying to write unknown chunk id %"PRIx32,
903 chunk_ids[i]);
907 if (written != chunk_offsets[num_chunks])
908 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
909 written,
910 chunk_offsets[num_chunks]);
912 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
913 commit_lock_file(&lk);
915 cleanup:
916 for (i = 0; i < packs.nr; i++) {
917 if (packs.list[i]) {
918 close_pack(packs.list[i]);
919 free(packs.list[i]);
921 free(packs.names[i]);
924 free(packs.list);
925 free(packs.names);
926 free(entries);
927 free(pack_perm);
928 free(midx_name);
929 return 0;
932 void clear_midx_file(struct repository *r)
934 char *midx = get_midx_filename(r->objects->odb->path);
936 if (r->objects && r->objects->multi_pack_index) {
937 close_midx(r->objects->multi_pack_index);
938 r->objects->multi_pack_index = NULL;
941 if (remove_path(midx)) {
942 UNLEAK(midx);
943 die(_("failed to clear multi-pack-index at %s"), midx);
946 free(midx);
949 static int verify_midx_error;
951 static void midx_report(const char *fmt, ...)
953 va_list ap;
954 verify_midx_error = 1;
955 va_start(ap, fmt);
956 vfprintf(stderr, fmt, ap);
957 fprintf(stderr, "\n");
958 va_end(ap);
961 int verify_midx_file(const char *object_dir)
963 uint32_t i;
964 struct progress *progress;
965 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
966 verify_midx_error = 0;
968 if (!m)
969 return 0;
971 for (i = 0; i < m->num_packs; i++) {
972 if (prepare_midx_pack(m, i))
973 midx_report("failed to load pack in position %d", i);
976 for (i = 0; i < 255; i++) {
977 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
978 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
980 if (oid_fanout1 > oid_fanout2)
981 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
982 i, oid_fanout1, oid_fanout2, i + 1);
985 for (i = 0; i < m->num_objects - 1; i++) {
986 struct object_id oid1, oid2;
988 nth_midxed_object_oid(&oid1, m, i);
989 nth_midxed_object_oid(&oid2, m, i + 1);
991 if (oidcmp(&oid1, &oid2) >= 0)
992 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
993 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
996 progress = start_progress(_("Verifying object offsets"), m->num_objects);
997 for (i = 0; i < m->num_objects; i++) {
998 struct object_id oid;
999 struct pack_entry e;
1000 off_t m_offset, p_offset;
1002 nth_midxed_object_oid(&oid, m, i);
1003 if (!fill_midx_entry(&oid, &e, m)) {
1004 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1005 i, oid_to_hex(&oid));
1006 continue;
1009 if (open_pack_index(e.p)) {
1010 midx_report(_("failed to load pack-index for packfile %s"),
1011 e.p->pack_name);
1012 break;
1015 m_offset = e.offset;
1016 p_offset = find_pack_entry_one(oid.hash, e.p);
1018 if (m_offset != p_offset)
1019 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1020 i, oid_to_hex(&oid), m_offset, p_offset);
1022 display_progress(progress, i + 1);
1024 stop_progress(&progress);
1026 return verify_midx_error;