multi-pack-index: verify corrupt chunk lookup table
[git/git.git] / midx.c
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
11 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
12 #define MIDX_VERSION 1
13 #define MIDX_BYTE_FILE_VERSION 4
14 #define MIDX_BYTE_HASH_VERSION 5
15 #define MIDX_BYTE_NUM_CHUNKS 6
16 #define MIDX_BYTE_NUM_PACKS 8
17 #define MIDX_HASH_VERSION 1
18 #define MIDX_HEADER_SIZE 12
19 #define MIDX_HASH_LEN 20
20 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + MIDX_HASH_LEN)
21
22 #define MIDX_MAX_CHUNKS 5
23 #define MIDX_CHUNK_ALIGNMENT 4
24 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
25 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
26 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
27 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
28 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
29 #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
30 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
31 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
32 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
33 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
34
35 static char *get_midx_filename(const char *object_dir)
36 {
37 return xstrfmt("%s/pack/multi-pack-index", object_dir);
38 }
39
40 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
41 {
42 struct multi_pack_index *m = NULL;
43 int fd;
44 struct stat st;
45 size_t midx_size;
46 void *midx_map = NULL;
47 uint32_t hash_version;
48 char *midx_name = get_midx_filename(object_dir);
49 uint32_t i;
50 const char *cur_pack_name;
51
52 fd = git_open(midx_name);
53
54 if (fd < 0)
55 goto cleanup_fail;
56 if (fstat(fd, &st)) {
57 error_errno(_("failed to read %s"), midx_name);
58 goto cleanup_fail;
59 }
60
61 midx_size = xsize_t(st.st_size);
62
63 if (midx_size < MIDX_MIN_SIZE) {
64 error(_("multi-pack-index file %s is too small"), midx_name);
65 goto cleanup_fail;
66 }
67
68 FREE_AND_NULL(midx_name);
69
70 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
71
72 FLEX_ALLOC_MEM(m, object_dir, object_dir, strlen(object_dir));
73 m->fd = fd;
74 m->data = midx_map;
75 m->data_len = midx_size;
76 m->local = local;
77
78 m->signature = get_be32(m->data);
79 if (m->signature != MIDX_SIGNATURE)
80 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
81 m->signature, MIDX_SIGNATURE);
82
83 m->version = m->data[MIDX_BYTE_FILE_VERSION];
84 if (m->version != MIDX_VERSION)
85 die(_("multi-pack-index version %d not recognized"),
86 m->version);
87
88 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
89 if (hash_version != MIDX_HASH_VERSION)
90 die(_("hash version %u does not match"), hash_version);
91 m->hash_len = MIDX_HASH_LEN;
92
93 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
94
95 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
96
97 for (i = 0; i < m->num_chunks; i++) {
98 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
99 MIDX_CHUNKLOOKUP_WIDTH * i);
100 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
101 MIDX_CHUNKLOOKUP_WIDTH * i);
102
103 if (chunk_offset >= m->data_len)
104 die(_("invalid chunk offset (too large)"));
105
106 switch (chunk_id) {
107 case MIDX_CHUNKID_PACKNAMES:
108 m->chunk_pack_names = m->data + chunk_offset;
109 break;
110
111 case MIDX_CHUNKID_OIDFANOUT:
112 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
113 break;
114
115 case MIDX_CHUNKID_OIDLOOKUP:
116 m->chunk_oid_lookup = m->data + chunk_offset;
117 break;
118
119 case MIDX_CHUNKID_OBJECTOFFSETS:
120 m->chunk_object_offsets = m->data + chunk_offset;
121 break;
122
123 case MIDX_CHUNKID_LARGEOFFSETS:
124 m->chunk_large_offsets = m->data + chunk_offset;
125 break;
126
127 case 0:
128 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
129 break;
130
131 default:
132 /*
133 * Do nothing on unrecognized chunks, allowing future
134 * extensions to add optional chunks.
135 */
136 break;
137 }
138 }
139
140 if (!m->chunk_pack_names)
141 die(_("multi-pack-index missing required pack-name chunk"));
142 if (!m->chunk_oid_fanout)
143 die(_("multi-pack-index missing required OID fanout chunk"));
144 if (!m->chunk_oid_lookup)
145 die(_("multi-pack-index missing required OID lookup chunk"));
146 if (!m->chunk_object_offsets)
147 die(_("multi-pack-index missing required object offsets chunk"));
148
149 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
150
151 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
152 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
153
154 cur_pack_name = (const char *)m->chunk_pack_names;
155 for (i = 0; i < m->num_packs; i++) {
156 m->pack_names[i] = cur_pack_name;
157
158 cur_pack_name += strlen(cur_pack_name) + 1;
159
160 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0) {
161 error(_("multi-pack-index pack names out of order: '%s' before '%s'"),
162 m->pack_names[i - 1],
163 m->pack_names[i]);
164 goto cleanup_fail;
165 }
166 }
167
168 return m;
169
170 cleanup_fail:
171 free(m);
172 free(midx_name);
173 if (midx_map)
174 munmap(midx_map, midx_size);
175 if (0 <= fd)
176 close(fd);
177 return NULL;
178 }
179
180 static void close_midx(struct multi_pack_index *m)
181 {
182 uint32_t i;
183 munmap((unsigned char *)m->data, m->data_len);
184 close(m->fd);
185 m->fd = -1;
186
187 for (i = 0; i < m->num_packs; i++) {
188 if (m->packs[i]) {
189 close_pack(m->packs[i]);
190 free(m->packs);
191 }
192 }
193 FREE_AND_NULL(m->packs);
194 FREE_AND_NULL(m->pack_names);
195 }
196
197 int prepare_midx_pack(struct multi_pack_index *m, uint32_t pack_int_id)
198 {
199 struct strbuf pack_name = STRBUF_INIT;
200
201 if (pack_int_id >= m->num_packs)
202 BUG("bad pack-int-id");
203
204 if (m->packs[pack_int_id])
205 return 0;
206
207 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
208 m->pack_names[pack_int_id]);
209
210 m->packs[pack_int_id] = add_packed_git(pack_name.buf, pack_name.len, m->local);
211 strbuf_release(&pack_name);
212 return !m->packs[pack_int_id];
213 }
214
215 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
216 {
217 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
218 MIDX_HASH_LEN, result);
219 }
220
221 struct object_id *nth_midxed_object_oid(struct object_id *oid,
222 struct multi_pack_index *m,
223 uint32_t n)
224 {
225 if (n >= m->num_objects)
226 return NULL;
227
228 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
229 return oid;
230 }
231
232 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
233 {
234 const unsigned char *offset_data;
235 uint32_t offset32;
236
237 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
238 offset32 = get_be32(offset_data + sizeof(uint32_t));
239
240 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
241 if (sizeof(offset32) < sizeof(uint64_t))
242 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
243
244 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
245 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
246 }
247
248 return offset32;
249 }
250
251 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
252 {
253 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
254 }
255
256 static int nth_midxed_pack_entry(struct multi_pack_index *m, struct pack_entry *e, uint32_t pos)
257 {
258 uint32_t pack_int_id;
259 struct packed_git *p;
260
261 if (pos >= m->num_objects)
262 return 0;
263
264 pack_int_id = nth_midxed_pack_int_id(m, pos);
265
266 if (prepare_midx_pack(m, pack_int_id))
267 die(_("error preparing packfile from multi-pack-index"));
268 p = m->packs[pack_int_id];
269
270 /*
271 * We are about to tell the caller where they can locate the
272 * requested object. We better make sure the packfile is
273 * still here and can be accessed before supplying that
274 * answer, as it may have been deleted since the MIDX was
275 * loaded!
276 */
277 if (!is_pack_valid(p))
278 return 0;
279
280 if (p->num_bad_objects) {
281 uint32_t i;
282 struct object_id oid;
283 nth_midxed_object_oid(&oid, m, pos);
284 for (i = 0; i < p->num_bad_objects; i++)
285 if (!hashcmp(oid.hash,
286 p->bad_object_sha1 + the_hash_algo->rawsz * i))
287 return 0;
288 }
289
290 e->offset = nth_midxed_offset(m, pos);
291 e->p = p;
292
293 return 1;
294 }
295
296 int fill_midx_entry(const struct object_id *oid, struct pack_entry *e, struct multi_pack_index *m)
297 {
298 uint32_t pos;
299
300 if (!bsearch_midx(oid, m, &pos))
301 return 0;
302
303 return nth_midxed_pack_entry(m, e, pos);
304 }
305
306 int midx_contains_pack(struct multi_pack_index *m, const char *idx_name)
307 {
308 uint32_t first = 0, last = m->num_packs;
309
310 while (first < last) {
311 uint32_t mid = first + (last - first) / 2;
312 const char *current;
313 int cmp;
314
315 current = m->pack_names[mid];
316 cmp = strcmp(idx_name, current);
317 if (!cmp)
318 return 1;
319 if (cmp > 0) {
320 first = mid + 1;
321 continue;
322 }
323 last = mid;
324 }
325
326 return 0;
327 }
328
329 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
330 {
331 struct multi_pack_index *m;
332 struct multi_pack_index *m_search;
333 int config_value;
334
335 if (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
336 !config_value)
337 return 0;
338
339 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
340 if (!strcmp(object_dir, m_search->object_dir))
341 return 1;
342
343 m = load_multi_pack_index(object_dir, local);
344
345 if (m) {
346 m->next = r->objects->multi_pack_index;
347 r->objects->multi_pack_index = m;
348 return 1;
349 }
350
351 return 0;
352 }
353
354 static size_t write_midx_header(struct hashfile *f,
355 unsigned char num_chunks,
356 uint32_t num_packs)
357 {
358 unsigned char byte_values[4];
359
360 hashwrite_be32(f, MIDX_SIGNATURE);
361 byte_values[0] = MIDX_VERSION;
362 byte_values[1] = MIDX_HASH_VERSION;
363 byte_values[2] = num_chunks;
364 byte_values[3] = 0; /* unused */
365 hashwrite(f, byte_values, sizeof(byte_values));
366 hashwrite_be32(f, num_packs);
367
368 return MIDX_HEADER_SIZE;
369 }
370
371 struct pack_list {
372 struct packed_git **list;
373 char **names;
374 uint32_t nr;
375 uint32_t alloc_list;
376 uint32_t alloc_names;
377 size_t pack_name_concat_len;
378 struct multi_pack_index *m;
379 };
380
381 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
382 const char *file_name, void *data)
383 {
384 struct pack_list *packs = (struct pack_list *)data;
385
386 if (ends_with(file_name, ".idx")) {
387 if (packs->m && midx_contains_pack(packs->m, file_name))
388 return;
389
390 ALLOC_GROW(packs->list, packs->nr + 1, packs->alloc_list);
391 ALLOC_GROW(packs->names, packs->nr + 1, packs->alloc_names);
392
393 packs->list[packs->nr] = add_packed_git(full_path,
394 full_path_len,
395 0);
396
397 if (!packs->list[packs->nr]) {
398 warning(_("failed to add packfile '%s'"),
399 full_path);
400 return;
401 }
402
403 if (open_pack_index(packs->list[packs->nr])) {
404 warning(_("failed to open pack-index '%s'"),
405 full_path);
406 close_pack(packs->list[packs->nr]);
407 FREE_AND_NULL(packs->list[packs->nr]);
408 return;
409 }
410
411 packs->names[packs->nr] = xstrdup(file_name);
412 packs->pack_name_concat_len += strlen(file_name) + 1;
413 packs->nr++;
414 }
415 }
416
417 struct pack_pair {
418 uint32_t pack_int_id;
419 char *pack_name;
420 };
421
422 static int pack_pair_compare(const void *_a, const void *_b)
423 {
424 struct pack_pair *a = (struct pack_pair *)_a;
425 struct pack_pair *b = (struct pack_pair *)_b;
426 return strcmp(a->pack_name, b->pack_name);
427 }
428
429 static void sort_packs_by_name(char **pack_names, uint32_t nr_packs, uint32_t *perm)
430 {
431 uint32_t i;
432 struct pack_pair *pairs;
433
434 ALLOC_ARRAY(pairs, nr_packs);
435
436 for (i = 0; i < nr_packs; i++) {
437 pairs[i].pack_int_id = i;
438 pairs[i].pack_name = pack_names[i];
439 }
440
441 QSORT(pairs, nr_packs, pack_pair_compare);
442
443 for (i = 0; i < nr_packs; i++) {
444 pack_names[i] = pairs[i].pack_name;
445 perm[pairs[i].pack_int_id] = i;
446 }
447
448 free(pairs);
449 }
450
451 struct pack_midx_entry {
452 struct object_id oid;
453 uint32_t pack_int_id;
454 time_t pack_mtime;
455 uint64_t offset;
456 };
457
458 static int midx_oid_compare(const void *_a, const void *_b)
459 {
460 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
461 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
462 int cmp = oidcmp(&a->oid, &b->oid);
463
464 if (cmp)
465 return cmp;
466
467 if (a->pack_mtime > b->pack_mtime)
468 return -1;
469 else if (a->pack_mtime < b->pack_mtime)
470 return 1;
471
472 return a->pack_int_id - b->pack_int_id;
473 }
474
475 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
476 uint32_t *pack_perm,
477 struct pack_midx_entry *e,
478 uint32_t pos)
479 {
480 if (pos >= m->num_objects)
481 return 1;
482
483 nth_midxed_object_oid(&e->oid, m, pos);
484 e->pack_int_id = pack_perm[nth_midxed_pack_int_id(m, pos)];
485 e->offset = nth_midxed_offset(m, pos);
486
487 /* consider objects in midx to be from "old" packs */
488 e->pack_mtime = 0;
489 return 0;
490 }
491
492 static void fill_pack_entry(uint32_t pack_int_id,
493 struct packed_git *p,
494 uint32_t cur_object,
495 struct pack_midx_entry *entry)
496 {
497 if (!nth_packed_object_oid(&entry->oid, p, cur_object))
498 die(_("failed to locate object %d in packfile"), cur_object);
499
500 entry->pack_int_id = pack_int_id;
501 entry->pack_mtime = p->mtime;
502
503 entry->offset = nth_packed_object_offset(p, cur_object);
504 }
505
506 /*
507 * It is possible to artificially get into a state where there are many
508 * duplicate copies of objects. That can create high memory pressure if
509 * we are to create a list of all objects before de-duplication. To reduce
510 * this memory pressure without a significant performance drop, automatically
511 * group objects by the first byte of their object id. Use the IDX fanout
512 * tables to group the data, copy to a local array, then sort.
513 *
514 * Copy only the de-duplicated entries (selected by most-recent modified time
515 * of a packfile containing the object).
516 */
517 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
518 struct packed_git **p,
519 uint32_t *perm,
520 uint32_t nr_packs,
521 uint32_t *nr_objects)
522 {
523 uint32_t cur_fanout, cur_pack, cur_object;
524 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
525 struct pack_midx_entry *entries_by_fanout = NULL;
526 struct pack_midx_entry *deduplicated_entries = NULL;
527 uint32_t start_pack = m ? m->num_packs : 0;
528
529 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
530 total_objects += p[cur_pack]->num_objects;
531
532 /*
533 * As we de-duplicate by fanout value, we expect the fanout
534 * slices to be evenly distributed, with some noise. Hence,
535 * allocate slightly more than one 256th.
536 */
537 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
538
539 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
540 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
541 *nr_objects = 0;
542
543 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
544 uint32_t nr_fanout = 0;
545
546 if (m) {
547 uint32_t start = 0, end;
548
549 if (cur_fanout)
550 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
551 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
552
553 for (cur_object = start; cur_object < end; cur_object++) {
554 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
555 nth_midxed_pack_midx_entry(m, perm,
556 &entries_by_fanout[nr_fanout],
557 cur_object);
558 nr_fanout++;
559 }
560 }
561
562 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
563 uint32_t start = 0, end;
564
565 if (cur_fanout)
566 start = get_pack_fanout(p[cur_pack], cur_fanout - 1);
567 end = get_pack_fanout(p[cur_pack], cur_fanout);
568
569 for (cur_object = start; cur_object < end; cur_object++) {
570 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
571 fill_pack_entry(perm[cur_pack], p[cur_pack], cur_object, &entries_by_fanout[nr_fanout]);
572 nr_fanout++;
573 }
574 }
575
576 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
577
578 /*
579 * The batch is now sorted by OID and then mtime (descending).
580 * Take only the first duplicate.
581 */
582 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
583 if (cur_object && !oidcmp(&entries_by_fanout[cur_object - 1].oid,
584 &entries_by_fanout[cur_object].oid))
585 continue;
586
587 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
588 memcpy(&deduplicated_entries[*nr_objects],
589 &entries_by_fanout[cur_object],
590 sizeof(struct pack_midx_entry));
591 (*nr_objects)++;
592 }
593 }
594
595 free(entries_by_fanout);
596 return deduplicated_entries;
597 }
598
599 static size_t write_midx_pack_names(struct hashfile *f,
600 char **pack_names,
601 uint32_t num_packs)
602 {
603 uint32_t i;
604 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
605 size_t written = 0;
606
607 for (i = 0; i < num_packs; i++) {
608 size_t writelen = strlen(pack_names[i]) + 1;
609
610 if (i && strcmp(pack_names[i], pack_names[i - 1]) <= 0)
611 BUG("incorrect pack-file order: %s before %s",
612 pack_names[i - 1],
613 pack_names[i]);
614
615 hashwrite(f, pack_names[i], writelen);
616 written += writelen;
617 }
618
619 /* add padding to be aligned */
620 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
621 if (i < MIDX_CHUNK_ALIGNMENT) {
622 memset(padding, 0, sizeof(padding));
623 hashwrite(f, padding, i);
624 written += i;
625 }
626
627 return written;
628 }
629
630 static size_t write_midx_oid_fanout(struct hashfile *f,
631 struct pack_midx_entry *objects,
632 uint32_t nr_objects)
633 {
634 struct pack_midx_entry *list = objects;
635 struct pack_midx_entry *last = objects + nr_objects;
636 uint32_t count = 0;
637 uint32_t i;
638
639 /*
640 * Write the first-level table (the list is sorted,
641 * but we use a 256-entry lookup to be able to avoid
642 * having to do eight extra binary search iterations).
643 */
644 for (i = 0; i < 256; i++) {
645 struct pack_midx_entry *next = list;
646
647 while (next < last && next->oid.hash[0] == i) {
648 count++;
649 next++;
650 }
651
652 hashwrite_be32(f, count);
653 list = next;
654 }
655
656 return MIDX_CHUNK_FANOUT_SIZE;
657 }
658
659 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
660 struct pack_midx_entry *objects,
661 uint32_t nr_objects)
662 {
663 struct pack_midx_entry *list = objects;
664 uint32_t i;
665 size_t written = 0;
666
667 for (i = 0; i < nr_objects; i++) {
668 struct pack_midx_entry *obj = list++;
669
670 if (i < nr_objects - 1) {
671 struct pack_midx_entry *next = list;
672 if (oidcmp(&obj->oid, &next->oid) >= 0)
673 BUG("OIDs not in order: %s >= %s",
674 oid_to_hex(&obj->oid),
675 oid_to_hex(&next->oid));
676 }
677
678 hashwrite(f, obj->oid.hash, (int)hash_len);
679 written += hash_len;
680 }
681
682 return written;
683 }
684
685 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
686 struct pack_midx_entry *objects, uint32_t nr_objects)
687 {
688 struct pack_midx_entry *list = objects;
689 uint32_t i, nr_large_offset = 0;
690 size_t written = 0;
691
692 for (i = 0; i < nr_objects; i++) {
693 struct pack_midx_entry *obj = list++;
694
695 hashwrite_be32(f, obj->pack_int_id);
696
697 if (large_offset_needed && obj->offset >> 31)
698 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
699 else if (!large_offset_needed && obj->offset >> 32)
700 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
701 oid_to_hex(&obj->oid),
702 obj->offset);
703 else
704 hashwrite_be32(f, (uint32_t)obj->offset);
705
706 written += MIDX_CHUNK_OFFSET_WIDTH;
707 }
708
709 return written;
710 }
711
712 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
713 struct pack_midx_entry *objects, uint32_t nr_objects)
714 {
715 struct pack_midx_entry *list = objects;
716 size_t written = 0;
717
718 while (nr_large_offset) {
719 struct pack_midx_entry *obj = list++;
720 uint64_t offset = obj->offset;
721
722 if (!(offset >> 31))
723 continue;
724
725 hashwrite_be32(f, offset >> 32);
726 hashwrite_be32(f, offset & 0xffffffffUL);
727 written += 2 * sizeof(uint32_t);
728
729 nr_large_offset--;
730 }
731
732 return written;
733 }
734
735 int write_midx_file(const char *object_dir)
736 {
737 unsigned char cur_chunk, num_chunks = 0;
738 char *midx_name;
739 uint32_t i;
740 struct hashfile *f = NULL;
741 struct lock_file lk;
742 struct pack_list packs;
743 uint32_t *pack_perm = NULL;
744 uint64_t written = 0;
745 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
746 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
747 uint32_t nr_entries, num_large_offsets = 0;
748 struct pack_midx_entry *entries = NULL;
749 int large_offsets_needed = 0;
750
751 midx_name = get_midx_filename(object_dir);
752 if (safe_create_leading_directories(midx_name)) {
753 UNLEAK(midx_name);
754 die_errno(_("unable to create leading directories of %s"),
755 midx_name);
756 }
757
758 packs.m = load_multi_pack_index(object_dir, 1);
759
760 packs.nr = 0;
761 packs.alloc_list = packs.m ? packs.m->num_packs : 16;
762 packs.alloc_names = packs.alloc_list;
763 packs.list = NULL;
764 packs.names = NULL;
765 packs.pack_name_concat_len = 0;
766 ALLOC_ARRAY(packs.list, packs.alloc_list);
767 ALLOC_ARRAY(packs.names, packs.alloc_names);
768
769 if (packs.m) {
770 for (i = 0; i < packs.m->num_packs; i++) {
771 ALLOC_GROW(packs.list, packs.nr + 1, packs.alloc_list);
772 ALLOC_GROW(packs.names, packs.nr + 1, packs.alloc_names);
773
774 packs.list[packs.nr] = NULL;
775 packs.names[packs.nr] = xstrdup(packs.m->pack_names[i]);
776 packs.pack_name_concat_len += strlen(packs.names[packs.nr]) + 1;
777 packs.nr++;
778 }
779 }
780
781 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
782
783 if (packs.m && packs.nr == packs.m->num_packs)
784 goto cleanup;
785
786 if (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
787 packs.pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
788 (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
789
790 ALLOC_ARRAY(pack_perm, packs.nr);
791 sort_packs_by_name(packs.names, packs.nr, pack_perm);
792
793 entries = get_sorted_entries(packs.m, packs.list, pack_perm, packs.nr, &nr_entries);
794
795 for (i = 0; i < nr_entries; i++) {
796 if (entries[i].offset > 0x7fffffff)
797 num_large_offsets++;
798 if (entries[i].offset > 0xffffffff)
799 large_offsets_needed = 1;
800 }
801
802 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
803 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
804 FREE_AND_NULL(midx_name);
805
806 if (packs.m)
807 close_midx(packs.m);
808
809 cur_chunk = 0;
810 num_chunks = large_offsets_needed ? 5 : 4;
811
812 written = write_midx_header(f, num_chunks, packs.nr);
813
814 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
815 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
816
817 cur_chunk++;
818 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
819 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + packs.pack_name_concat_len;
820
821 cur_chunk++;
822 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
823 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
824
825 cur_chunk++;
826 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
827 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_HASH_LEN;
828
829 cur_chunk++;
830 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
831 if (large_offsets_needed) {
832 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
833
834 cur_chunk++;
835 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
836 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
837 }
838
839 chunk_ids[cur_chunk] = 0;
840
841 for (i = 0; i <= num_chunks; i++) {
842 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
843 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
844 chunk_offsets[i - 1],
845 chunk_offsets[i]);
846
847 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
848 BUG("chunk offset %"PRIu64" is not properly aligned",
849 chunk_offsets[i]);
850
851 hashwrite_be32(f, chunk_ids[i]);
852 hashwrite_be32(f, chunk_offsets[i] >> 32);
853 hashwrite_be32(f, chunk_offsets[i]);
854
855 written += MIDX_CHUNKLOOKUP_WIDTH;
856 }
857
858 for (i = 0; i < num_chunks; i++) {
859 if (written != chunk_offsets[i])
860 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
861 chunk_offsets[i],
862 written,
863 chunk_ids[i]);
864
865 switch (chunk_ids[i]) {
866 case MIDX_CHUNKID_PACKNAMES:
867 written += write_midx_pack_names(f, packs.names, packs.nr);
868 break;
869
870 case MIDX_CHUNKID_OIDFANOUT:
871 written += write_midx_oid_fanout(f, entries, nr_entries);
872 break;
873
874 case MIDX_CHUNKID_OIDLOOKUP:
875 written += write_midx_oid_lookup(f, MIDX_HASH_LEN, entries, nr_entries);
876 break;
877
878 case MIDX_CHUNKID_OBJECTOFFSETS:
879 written += write_midx_object_offsets(f, large_offsets_needed, entries, nr_entries);
880 break;
881
882 case MIDX_CHUNKID_LARGEOFFSETS:
883 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
884 break;
885
886 default:
887 BUG("trying to write unknown chunk id %"PRIx32,
888 chunk_ids[i]);
889 }
890 }
891
892 if (written != chunk_offsets[num_chunks])
893 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
894 written,
895 chunk_offsets[num_chunks]);
896
897 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
898 commit_lock_file(&lk);
899
900 cleanup:
901 for (i = 0; i < packs.nr; i++) {
902 if (packs.list[i]) {
903 close_pack(packs.list[i]);
904 free(packs.list[i]);
905 }
906 free(packs.names[i]);
907 }
908
909 free(packs.list);
910 free(packs.names);
911 free(entries);
912 free(pack_perm);
913 free(midx_name);
914 return 0;
915 }
916
917 void clear_midx_file(const char *object_dir)
918 {
919 char *midx = get_midx_filename(object_dir);
920
921 if (remove_path(midx)) {
922 UNLEAK(midx);
923 die(_("failed to clear multi-pack-index at %s"), midx);
924 }
925
926 free(midx);
927 }
928
929 static int verify_midx_error;
930
931 int verify_midx_file(const char *object_dir)
932 {
933 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
934 verify_midx_error = 0;
935
936 if (!m)
937 return 0;
938
939 return verify_midx_error;
940 }