multi-pack-index: verify object offsets
[git/git.git] / midx.c
CommitLineData
a3407730 1#include "cache.h"
c4d25228 2#include "config.h"
fc59e748 3#include "csum-file.h"
396f2570 4#include "dir.h"
fc59e748 5#include "lockfile.h"
396f2570 6#include "packfile.h"
4d80560c 7#include "object-store.h"
3715a633 8#include "sha1-lookup.h"
a3407730
DS
9#include "midx.h"
10
fc59e748
DS
11#define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
12#define MIDX_VERSION 1
4d80560c
DS
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
fc59e748
DS
17#define MIDX_HASH_VERSION 1
18#define MIDX_HEADER_SIZE 12
4d80560c
DS
19#define MIDX_HASH_LEN 20
20#define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + MIDX_HASH_LEN)
fc59e748 21
662148c4 22#define MIDX_MAX_CHUNKS 5
32f3c541
DS
23#define MIDX_CHUNK_ALIGNMENT 4
24#define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
d7cacf29 25#define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
0d5b3a5e 26#define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
662148c4
DS
27#define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
28#define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
32f3c541 29#define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
d7cacf29 30#define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
662148c4
DS
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
32f3c541 34
fc59e748
DS
35static char *get_midx_filename(const char *object_dir)
36{
37 return xstrfmt("%s/pack/multi-pack-index", object_dir);
38}
39
2cf489a3 40struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
4d80560c
DS
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);
32f3c541 49 uint32_t i;
3227565c 50 const char *cur_pack_name;
4d80560c
DS
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;
2cf489a3 76 m->local = local;
4d80560c
DS
77
78 m->signature = get_be32(m->data);
53ad0407
DS
79 if (m->signature != MIDX_SIGNATURE)
80 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
4d80560c 81 m->signature, MIDX_SIGNATURE);
4d80560c
DS
82
83 m->version = m->data[MIDX_BYTE_FILE_VERSION];
53ad0407
DS
84 if (m->version != MIDX_VERSION)
85 die(_("multi-pack-index version %d not recognized"),
4d80560c 86 m->version);
4d80560c
DS
87
88 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
53ad0407
DS
89 if (hash_version != MIDX_HASH_VERSION)
90 die(_("hash version %u does not match"), hash_version);
4d80560c
DS
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
32f3c541
DS
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
d3f8e211
DS
103 if (chunk_offset >= m->data_len)
104 die(_("invalid chunk offset (too large)"));
105
32f3c541
DS
106 switch (chunk_id) {
107 case MIDX_CHUNKID_PACKNAMES:
108 m->chunk_pack_names = m->data + chunk_offset;
109 break;
110
d7cacf29
DS
111 case MIDX_CHUNKID_OIDFANOUT:
112 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
113 break;
114
0d5b3a5e
DS
115 case MIDX_CHUNKID_OIDLOOKUP:
116 m->chunk_oid_lookup = m->data + chunk_offset;
117 break;
118
662148c4
DS
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
32f3c541
DS
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"));
d7cacf29
DS
142 if (!m->chunk_oid_fanout)
143 die(_("multi-pack-index missing required OID fanout chunk"));
0d5b3a5e
DS
144 if (!m->chunk_oid_lookup)
145 die(_("multi-pack-index missing required OID lookup chunk"));
662148c4
DS
146 if (!m->chunk_object_offsets)
147 die(_("multi-pack-index missing required object offsets chunk"));
32f3c541 148
d7cacf29
DS
149 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
150
3227565c 151 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
3715a633 152 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
3227565c
DS
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
8e72a3c3
DS
160 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
161 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
3227565c
DS
162 m->pack_names[i - 1],
163 m->pack_names[i]);
3227565c
DS
164 }
165
4d80560c
DS
166 return m;
167
168cleanup_fail:
169 free(m);
170 free(midx_name);
171 if (midx_map)
172 munmap(midx_map, midx_size);
173 if (0 <= fd)
174 close(fd);
175 return NULL;
176}
177
a40498a1
DS
178static void close_midx(struct multi_pack_index *m)
179{
180 uint32_t i;
181 munmap((unsigned char *)m->data, m->data_len);
182 close(m->fd);
183 m->fd = -1;
184
185 for (i = 0; i < m->num_packs; i++) {
186 if (m->packs[i]) {
187 close_pack(m->packs[i]);
188 free(m->packs);
189 }
190 }
191 FREE_AND_NULL(m->packs);
192 FREE_AND_NULL(m->pack_names);
193}
194
0bff5269 195int prepare_midx_pack(struct multi_pack_index *m, uint32_t pack_int_id)
3715a633
DS
196{
197 struct strbuf pack_name = STRBUF_INIT;
198
199 if (pack_int_id >= m->num_packs)
cc6af73c
DS
200 die(_("bad pack-int-id: %u (%u total packs"),
201 pack_int_id, m->num_packs);
3715a633
DS
202
203 if (m->packs[pack_int_id])
204 return 0;
205
206 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
207 m->pack_names[pack_int_id]);
208
2cf489a3 209 m->packs[pack_int_id] = add_packed_git(pack_name.buf, pack_name.len, m->local);
3715a633
DS
210 strbuf_release(&pack_name);
211 return !m->packs[pack_int_id];
212}
213
214int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
215{
216 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
217 MIDX_HASH_LEN, result);
218}
219
8aac67a1
DS
220struct object_id *nth_midxed_object_oid(struct object_id *oid,
221 struct multi_pack_index *m,
222 uint32_t n)
223{
224 if (n >= m->num_objects)
225 return NULL;
226
227 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
228 return oid;
229}
230
3715a633
DS
231static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
232{
233 const unsigned char *offset_data;
234 uint32_t offset32;
235
236 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
237 offset32 = get_be32(offset_data + sizeof(uint32_t));
238
239 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
d8ac9ee1 240 if (sizeof(off_t) < sizeof(uint64_t))
3715a633
DS
241 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
242
243 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
244 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
245 }
246
247 return offset32;
248}
249
250static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
251{
252 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
253}
254
255static int nth_midxed_pack_entry(struct multi_pack_index *m, struct pack_entry *e, uint32_t pos)
256{
257 uint32_t pack_int_id;
258 struct packed_git *p;
259
260 if (pos >= m->num_objects)
261 return 0;
262
263 pack_int_id = nth_midxed_pack_int_id(m, pos);
264
265 if (prepare_midx_pack(m, pack_int_id))
266 die(_("error preparing packfile from multi-pack-index"));
267 p = m->packs[pack_int_id];
268
269 /*
270 * We are about to tell the caller where they can locate the
271 * requested object. We better make sure the packfile is
272 * still here and can be accessed before supplying that
273 * answer, as it may have been deleted since the MIDX was
274 * loaded!
275 */
276 if (!is_pack_valid(p))
277 return 0;
278
c39b02ae
DS
279 if (p->num_bad_objects) {
280 uint32_t i;
281 struct object_id oid;
282 nth_midxed_object_oid(&oid, m, pos);
283 for (i = 0; i < p->num_bad_objects; i++)
284 if (!hashcmp(oid.hash,
285 p->bad_object_sha1 + the_hash_algo->rawsz * i))
286 return 0;
287 }
288
3715a633
DS
289 e->offset = nth_midxed_offset(m, pos);
290 e->p = p;
291
292 return 1;
293}
294
295int fill_midx_entry(const struct object_id *oid, struct pack_entry *e, struct multi_pack_index *m)
296{
297 uint32_t pos;
298
299 if (!bsearch_midx(oid, m, &pos))
300 return 0;
301
302 return nth_midxed_pack_entry(m, e, pos);
303}
304
a40498a1
DS
305int midx_contains_pack(struct multi_pack_index *m, const char *idx_name)
306{
307 uint32_t first = 0, last = m->num_packs;
308
309 while (first < last) {
310 uint32_t mid = first + (last - first) / 2;
311 const char *current;
312 int cmp;
313
314 current = m->pack_names[mid];
315 cmp = strcmp(idx_name, current);
316 if (!cmp)
317 return 1;
318 if (cmp > 0) {
319 first = mid + 1;
320 continue;
321 }
322 last = mid;
323 }
324
325 return 0;
326}
327
2cf489a3 328int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
c4d25228 329{
29e2016b 330 struct multi_pack_index *m;
c4d25228
DS
331 struct multi_pack_index *m_search;
332 int config_value;
333
334 if (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
335 !config_value)
336 return 0;
337
29e2016b 338 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
c4d25228
DS
339 if (!strcmp(object_dir, m_search->object_dir))
340 return 1;
341
29e2016b 342 m = load_multi_pack_index(object_dir, local);
c4d25228 343
29e2016b
DS
344 if (m) {
345 m->next = r->objects->multi_pack_index;
346 r->objects->multi_pack_index = m;
c4d25228
DS
347 return 1;
348 }
349
350 return 0;
351}
352
fc59e748
DS
353static size_t write_midx_header(struct hashfile *f,
354 unsigned char num_chunks,
355 uint32_t num_packs)
356{
357 unsigned char byte_values[4];
358
359 hashwrite_be32(f, MIDX_SIGNATURE);
360 byte_values[0] = MIDX_VERSION;
361 byte_values[1] = MIDX_HASH_VERSION;
362 byte_values[2] = num_chunks;
363 byte_values[3] = 0; /* unused */
364 hashwrite(f, byte_values, sizeof(byte_values));
365 hashwrite_be32(f, num_packs);
366
367 return MIDX_HEADER_SIZE;
368}
369
396f2570
DS
370struct pack_list {
371 struct packed_git **list;
32f3c541 372 char **names;
396f2570
DS
373 uint32_t nr;
374 uint32_t alloc_list;
32f3c541
DS
375 uint32_t alloc_names;
376 size_t pack_name_concat_len;
a40498a1 377 struct multi_pack_index *m;
396f2570
DS
378};
379
380static void add_pack_to_midx(const char *full_path, size_t full_path_len,
381 const char *file_name, void *data)
382{
383 struct pack_list *packs = (struct pack_list *)data;
384
385 if (ends_with(file_name, ".idx")) {
a40498a1
DS
386 if (packs->m && midx_contains_pack(packs->m, file_name))
387 return;
388
396f2570 389 ALLOC_GROW(packs->list, packs->nr + 1, packs->alloc_list);
32f3c541 390 ALLOC_GROW(packs->names, packs->nr + 1, packs->alloc_names);
396f2570
DS
391
392 packs->list[packs->nr] = add_packed_git(full_path,
393 full_path_len,
394 0);
fe1ed56f 395
396f2570
DS
396 if (!packs->list[packs->nr]) {
397 warning(_("failed to add packfile '%s'"),
398 full_path);
399 return;
400 }
401
fe1ed56f
DS
402 if (open_pack_index(packs->list[packs->nr])) {
403 warning(_("failed to open pack-index '%s'"),
404 full_path);
405 close_pack(packs->list[packs->nr]);
406 FREE_AND_NULL(packs->list[packs->nr]);
407 return;
408 }
409
32f3c541
DS
410 packs->names[packs->nr] = xstrdup(file_name);
411 packs->pack_name_concat_len += strlen(file_name) + 1;
396f2570
DS
412 packs->nr++;
413 }
414}
415
32f3c541
DS
416struct pack_pair {
417 uint32_t pack_int_id;
418 char *pack_name;
419};
420
421static int pack_pair_compare(const void *_a, const void *_b)
422{
423 struct pack_pair *a = (struct pack_pair *)_a;
424 struct pack_pair *b = (struct pack_pair *)_b;
425 return strcmp(a->pack_name, b->pack_name);
426}
427
428static void sort_packs_by_name(char **pack_names, uint32_t nr_packs, uint32_t *perm)
429{
430 uint32_t i;
431 struct pack_pair *pairs;
432
433 ALLOC_ARRAY(pairs, nr_packs);
434
435 for (i = 0; i < nr_packs; i++) {
436 pairs[i].pack_int_id = i;
437 pairs[i].pack_name = pack_names[i];
438 }
439
440 QSORT(pairs, nr_packs, pack_pair_compare);
441
442 for (i = 0; i < nr_packs; i++) {
443 pack_names[i] = pairs[i].pack_name;
444 perm[pairs[i].pack_int_id] = i;
445 }
446
447 free(pairs);
448}
449
fe1ed56f
DS
450struct pack_midx_entry {
451 struct object_id oid;
452 uint32_t pack_int_id;
453 time_t pack_mtime;
454 uint64_t offset;
455};
456
457static int midx_oid_compare(const void *_a, const void *_b)
458{
459 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
460 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
461 int cmp = oidcmp(&a->oid, &b->oid);
462
463 if (cmp)
464 return cmp;
465
466 if (a->pack_mtime > b->pack_mtime)
467 return -1;
468 else if (a->pack_mtime < b->pack_mtime)
469 return 1;
470
471 return a->pack_int_id - b->pack_int_id;
472}
473
a40498a1
DS
474static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
475 uint32_t *pack_perm,
476 struct pack_midx_entry *e,
477 uint32_t pos)
478{
479 if (pos >= m->num_objects)
480 return 1;
481
482 nth_midxed_object_oid(&e->oid, m, pos);
483 e->pack_int_id = pack_perm[nth_midxed_pack_int_id(m, pos)];
484 e->offset = nth_midxed_offset(m, pos);
485
486 /* consider objects in midx to be from "old" packs */
487 e->pack_mtime = 0;
488 return 0;
489}
490
fe1ed56f
DS
491static void fill_pack_entry(uint32_t pack_int_id,
492 struct packed_git *p,
493 uint32_t cur_object,
494 struct pack_midx_entry *entry)
495{
496 if (!nth_packed_object_oid(&entry->oid, p, cur_object))
497 die(_("failed to locate object %d in packfile"), cur_object);
498
499 entry->pack_int_id = pack_int_id;
500 entry->pack_mtime = p->mtime;
501
502 entry->offset = nth_packed_object_offset(p, cur_object);
503}
504
505/*
506 * It is possible to artificially get into a state where there are many
507 * duplicate copies of objects. That can create high memory pressure if
508 * we are to create a list of all objects before de-duplication. To reduce
509 * this memory pressure without a significant performance drop, automatically
510 * group objects by the first byte of their object id. Use the IDX fanout
511 * tables to group the data, copy to a local array, then sort.
512 *
513 * Copy only the de-duplicated entries (selected by most-recent modified time
514 * of a packfile containing the object).
515 */
a40498a1
DS
516static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
517 struct packed_git **p,
fe1ed56f
DS
518 uint32_t *perm,
519 uint32_t nr_packs,
520 uint32_t *nr_objects)
521{
522 uint32_t cur_fanout, cur_pack, cur_object;
523 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
524 struct pack_midx_entry *entries_by_fanout = NULL;
525 struct pack_midx_entry *deduplicated_entries = NULL;
a40498a1 526 uint32_t start_pack = m ? m->num_packs : 0;
fe1ed56f 527
a40498a1 528 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
fe1ed56f
DS
529 total_objects += p[cur_pack]->num_objects;
530
531 /*
532 * As we de-duplicate by fanout value, we expect the fanout
533 * slices to be evenly distributed, with some noise. Hence,
534 * allocate slightly more than one 256th.
535 */
536 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
537
538 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
539 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
540 *nr_objects = 0;
541
542 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
543 uint32_t nr_fanout = 0;
544
a40498a1
DS
545 if (m) {
546 uint32_t start = 0, end;
547
548 if (cur_fanout)
549 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
550 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
551
552 for (cur_object = start; cur_object < end; cur_object++) {
553 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
554 nth_midxed_pack_midx_entry(m, perm,
555 &entries_by_fanout[nr_fanout],
556 cur_object);
557 nr_fanout++;
558 }
559 }
560
561 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
fe1ed56f
DS
562 uint32_t start = 0, end;
563
564 if (cur_fanout)
565 start = get_pack_fanout(p[cur_pack], cur_fanout - 1);
566 end = get_pack_fanout(p[cur_pack], cur_fanout);
567
568 for (cur_object = start; cur_object < end; cur_object++) {
569 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
570 fill_pack_entry(perm[cur_pack], p[cur_pack], cur_object, &entries_by_fanout[nr_fanout]);
571 nr_fanout++;
572 }
573 }
574
575 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
576
577 /*
578 * The batch is now sorted by OID and then mtime (descending).
579 * Take only the first duplicate.
580 */
581 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
582 if (cur_object && !oidcmp(&entries_by_fanout[cur_object - 1].oid,
583 &entries_by_fanout[cur_object].oid))
584 continue;
585
586 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
587 memcpy(&deduplicated_entries[*nr_objects],
588 &entries_by_fanout[cur_object],
589 sizeof(struct pack_midx_entry));
590 (*nr_objects)++;
591 }
592 }
593
594 free(entries_by_fanout);
595 return deduplicated_entries;
596}
597
32f3c541
DS
598static size_t write_midx_pack_names(struct hashfile *f,
599 char **pack_names,
600 uint32_t num_packs)
601{
602 uint32_t i;
603 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
604 size_t written = 0;
605
606 for (i = 0; i < num_packs; i++) {
607 size_t writelen = strlen(pack_names[i]) + 1;
608
609 if (i && strcmp(pack_names[i], pack_names[i - 1]) <= 0)
610 BUG("incorrect pack-file order: %s before %s",
611 pack_names[i - 1],
612 pack_names[i]);
613
614 hashwrite(f, pack_names[i], writelen);
615 written += writelen;
616 }
617
618 /* add padding to be aligned */
619 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
620 if (i < MIDX_CHUNK_ALIGNMENT) {
621 memset(padding, 0, sizeof(padding));
622 hashwrite(f, padding, i);
623 written += i;
624 }
625
626 return written;
627}
628
d7cacf29
DS
629static size_t write_midx_oid_fanout(struct hashfile *f,
630 struct pack_midx_entry *objects,
631 uint32_t nr_objects)
632{
633 struct pack_midx_entry *list = objects;
634 struct pack_midx_entry *last = objects + nr_objects;
635 uint32_t count = 0;
636 uint32_t i;
637
638 /*
639 * Write the first-level table (the list is sorted,
640 * but we use a 256-entry lookup to be able to avoid
641 * having to do eight extra binary search iterations).
642 */
643 for (i = 0; i < 256; i++) {
644 struct pack_midx_entry *next = list;
645
646 while (next < last && next->oid.hash[0] == i) {
647 count++;
648 next++;
649 }
650
651 hashwrite_be32(f, count);
652 list = next;
653 }
654
655 return MIDX_CHUNK_FANOUT_SIZE;
656}
657
0d5b3a5e
DS
658static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
659 struct pack_midx_entry *objects,
660 uint32_t nr_objects)
661{
662 struct pack_midx_entry *list = objects;
663 uint32_t i;
664 size_t written = 0;
665
666 for (i = 0; i < nr_objects; i++) {
667 struct pack_midx_entry *obj = list++;
668
669 if (i < nr_objects - 1) {
670 struct pack_midx_entry *next = list;
671 if (oidcmp(&obj->oid, &next->oid) >= 0)
672 BUG("OIDs not in order: %s >= %s",
673 oid_to_hex(&obj->oid),
674 oid_to_hex(&next->oid));
675 }
676
677 hashwrite(f, obj->oid.hash, (int)hash_len);
678 written += hash_len;
679 }
680
681 return written;
682}
683
662148c4
DS
684static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
685 struct pack_midx_entry *objects, uint32_t nr_objects)
686{
687 struct pack_midx_entry *list = objects;
688 uint32_t i, nr_large_offset = 0;
689 size_t written = 0;
690
691 for (i = 0; i < nr_objects; i++) {
692 struct pack_midx_entry *obj = list++;
693
694 hashwrite_be32(f, obj->pack_int_id);
695
696 if (large_offset_needed && obj->offset >> 31)
697 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
698 else if (!large_offset_needed && obj->offset >> 32)
699 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
700 oid_to_hex(&obj->oid),
701 obj->offset);
702 else
703 hashwrite_be32(f, (uint32_t)obj->offset);
704
705 written += MIDX_CHUNK_OFFSET_WIDTH;
706 }
707
708 return written;
709}
710
711static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
712 struct pack_midx_entry *objects, uint32_t nr_objects)
713{
714 struct pack_midx_entry *list = objects;
715 size_t written = 0;
716
717 while (nr_large_offset) {
718 struct pack_midx_entry *obj = list++;
719 uint64_t offset = obj->offset;
720
721 if (!(offset >> 31))
722 continue;
723
724 hashwrite_be32(f, offset >> 32);
725 hashwrite_be32(f, offset & 0xffffffffUL);
726 written += 2 * sizeof(uint32_t);
727
728 nr_large_offset--;
729 }
730
731 return written;
732}
733
a3407730
DS
734int write_midx_file(const char *object_dir)
735{
32f3c541 736 unsigned char cur_chunk, num_chunks = 0;
fc59e748 737 char *midx_name;
396f2570 738 uint32_t i;
fc59e748
DS
739 struct hashfile *f = NULL;
740 struct lock_file lk;
396f2570 741 struct pack_list packs;
32f3c541
DS
742 uint32_t *pack_perm = NULL;
743 uint64_t written = 0;
744 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
745 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
662148c4 746 uint32_t nr_entries, num_large_offsets = 0;
fe1ed56f 747 struct pack_midx_entry *entries = NULL;
662148c4 748 int large_offsets_needed = 0;
fc59e748
DS
749
750 midx_name = get_midx_filename(object_dir);
751 if (safe_create_leading_directories(midx_name)) {
752 UNLEAK(midx_name);
753 die_errno(_("unable to create leading directories of %s"),
754 midx_name);
755 }
756
2cf489a3 757 packs.m = load_multi_pack_index(object_dir, 1);
a40498a1 758
396f2570 759 packs.nr = 0;
a40498a1
DS
760 packs.alloc_list = packs.m ? packs.m->num_packs : 16;
761 packs.alloc_names = packs.alloc_list;
396f2570 762 packs.list = NULL;
a40498a1 763 packs.names = NULL;
32f3c541 764 packs.pack_name_concat_len = 0;
396f2570 765 ALLOC_ARRAY(packs.list, packs.alloc_list);
32f3c541 766 ALLOC_ARRAY(packs.names, packs.alloc_names);
396f2570 767
a40498a1
DS
768 if (packs.m) {
769 for (i = 0; i < packs.m->num_packs; i++) {
770 ALLOC_GROW(packs.list, packs.nr + 1, packs.alloc_list);
771 ALLOC_GROW(packs.names, packs.nr + 1, packs.alloc_names);
772
773 packs.list[packs.nr] = NULL;
774 packs.names[packs.nr] = xstrdup(packs.m->pack_names[i]);
775 packs.pack_name_concat_len += strlen(packs.names[packs.nr]) + 1;
776 packs.nr++;
777 }
778 }
779
396f2570
DS
780 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
781
a40498a1
DS
782 if (packs.m && packs.nr == packs.m->num_packs)
783 goto cleanup;
784
32f3c541
DS
785 if (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
786 packs.pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
787 (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
788
789 ALLOC_ARRAY(pack_perm, packs.nr);
790 sort_packs_by_name(packs.names, packs.nr, pack_perm);
791
a40498a1
DS
792 entries = get_sorted_entries(packs.m, packs.list, pack_perm, packs.nr, &nr_entries);
793
662148c4
DS
794 for (i = 0; i < nr_entries; i++) {
795 if (entries[i].offset > 0x7fffffff)
796 num_large_offsets++;
797 if (entries[i].offset > 0xffffffff)
798 large_offsets_needed = 1;
799 }
fe1ed56f 800
fc59e748
DS
801 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
802 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
803 FREE_AND_NULL(midx_name);
804
a40498a1
DS
805 if (packs.m)
806 close_midx(packs.m);
807
32f3c541 808 cur_chunk = 0;
662148c4 809 num_chunks = large_offsets_needed ? 5 : 4;
32f3c541
DS
810
811 written = write_midx_header(f, num_chunks, packs.nr);
812
813 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
814 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
815
816 cur_chunk++;
d7cacf29 817 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
32f3c541
DS
818 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + packs.pack_name_concat_len;
819
d7cacf29
DS
820 cur_chunk++;
821 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
822 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
823
0d5b3a5e 824 cur_chunk++;
662148c4 825 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
0d5b3a5e
DS
826 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_HASH_LEN;
827
662148c4
DS
828 cur_chunk++;
829 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
830 if (large_offsets_needed) {
831 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
832
833 cur_chunk++;
834 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
835 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
836 }
837
838 chunk_ids[cur_chunk] = 0;
839
32f3c541
DS
840 for (i = 0; i <= num_chunks; i++) {
841 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
842 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
843 chunk_offsets[i - 1],
844 chunk_offsets[i]);
845
846 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
847 BUG("chunk offset %"PRIu64" is not properly aligned",
848 chunk_offsets[i]);
849
850 hashwrite_be32(f, chunk_ids[i]);
851 hashwrite_be32(f, chunk_offsets[i] >> 32);
852 hashwrite_be32(f, chunk_offsets[i]);
853
854 written += MIDX_CHUNKLOOKUP_WIDTH;
855 }
856
857 for (i = 0; i < num_chunks; i++) {
858 if (written != chunk_offsets[i])
859 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
860 chunk_offsets[i],
861 written,
862 chunk_ids[i]);
863
864 switch (chunk_ids[i]) {
865 case MIDX_CHUNKID_PACKNAMES:
866 written += write_midx_pack_names(f, packs.names, packs.nr);
867 break;
868
d7cacf29
DS
869 case MIDX_CHUNKID_OIDFANOUT:
870 written += write_midx_oid_fanout(f, entries, nr_entries);
871 break;
872
0d5b3a5e
DS
873 case MIDX_CHUNKID_OIDLOOKUP:
874 written += write_midx_oid_lookup(f, MIDX_HASH_LEN, entries, nr_entries);
875 break;
876
662148c4
DS
877 case MIDX_CHUNKID_OBJECTOFFSETS:
878 written += write_midx_object_offsets(f, large_offsets_needed, entries, nr_entries);
879 break;
880
881 case MIDX_CHUNKID_LARGEOFFSETS:
882 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
883 break;
884
32f3c541
DS
885 default:
886 BUG("trying to write unknown chunk id %"PRIx32,
887 chunk_ids[i]);
888 }
889 }
890
891 if (written != chunk_offsets[num_chunks])
892 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
893 written,
894 chunk_offsets[num_chunks]);
fc59e748
DS
895
896 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
897 commit_lock_file(&lk);
898
a40498a1 899cleanup:
396f2570
DS
900 for (i = 0; i < packs.nr; i++) {
901 if (packs.list[i]) {
902 close_pack(packs.list[i]);
903 free(packs.list[i]);
904 }
32f3c541 905 free(packs.names[i]);
396f2570
DS
906 }
907
908 free(packs.list);
32f3c541 909 free(packs.names);
fe1ed56f 910 free(entries);
a40498a1
DS
911 free(pack_perm);
912 free(midx_name);
a3407730
DS
913 return 0;
914}
525e18c0
DS
915
916void clear_midx_file(const char *object_dir)
917{
918 char *midx = get_midx_filename(object_dir);
919
920 if (remove_path(midx)) {
921 UNLEAK(midx);
922 die(_("failed to clear multi-pack-index at %s"), midx);
923 }
924
925 free(midx);
926}
56ee7ff1
DS
927
928static int verify_midx_error;
929
d4bf1d88
DS
930static void midx_report(const char *fmt, ...)
931{
932 va_list ap;
933 verify_midx_error = 1;
934 va_start(ap, fmt);
935 vfprintf(stderr, fmt, ap);
936 fprintf(stderr, "\n");
937 va_end(ap);
938}
939
56ee7ff1
DS
940int verify_midx_file(const char *object_dir)
941{
d4bf1d88 942 uint32_t i;
56ee7ff1
DS
943 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
944 verify_midx_error = 0;
945
946 if (!m)
947 return 0;
948
d4bf1d88
DS
949 for (i = 0; i < m->num_packs; i++) {
950 if (prepare_midx_pack(m, i))
951 midx_report("failed to load pack in position %d", i);
952 }
953
2f23d3f3
DS
954 for (i = 0; i < 255; i++) {
955 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
956 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
957
958 if (oid_fanout1 > oid_fanout2)
959 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
960 i, oid_fanout1, oid_fanout2, i + 1);
961 }
962
55c5648d
DS
963 for (i = 0; i < m->num_objects - 1; i++) {
964 struct object_id oid1, oid2;
965
966 nth_midxed_object_oid(&oid1, m, i);
967 nth_midxed_object_oid(&oid2, m, i + 1);
968
969 if (oidcmp(&oid1, &oid2) >= 0)
970 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
971 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
972 }
973
cc6af73c
DS
974 for (i = 0; i < m->num_objects; i++) {
975 struct object_id oid;
976 struct pack_entry e;
977 off_t m_offset, p_offset;
978
979 nth_midxed_object_oid(&oid, m, i);
980 if (!fill_midx_entry(&oid, &e, m)) {
981 midx_report(_("failed to load pack entry for oid[%d] = %s"),
982 i, oid_to_hex(&oid));
983 continue;
984 }
985
986 if (open_pack_index(e.p)) {
987 midx_report(_("failed to load pack-index for packfile %s"),
988 e.p->pack_name);
989 break;
990 }
991
992 m_offset = e.offset;
993 p_offset = find_pack_entry_one(oid.hash, e.p);
994
995 if (m_offset != p_offset)
996 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
997 i, oid_to_hex(&oid), m_offset, p_offset);
998 }
999
56ee7ff1
DS
1000 return verify_midx_error;
1001}