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