multi-pack-index: store local property
[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);
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
32f3c541
DS
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
d7cacf29
DS
114 case MIDX_CHUNKID_OIDFANOUT:
115 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
116 break;
117
0d5b3a5e
DS
118 case MIDX_CHUNKID_OIDLOOKUP:
119 m->chunk_oid_lookup = m->data + chunk_offset;
120 break;
121
662148c4
DS
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
32f3c541
DS
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"));
d7cacf29
DS
145 if (!m->chunk_oid_fanout)
146 die(_("multi-pack-index missing required OID fanout chunk"));
0d5b3a5e
DS
147 if (!m->chunk_oid_lookup)
148 die(_("multi-pack-index missing required OID lookup chunk"));
662148c4
DS
149 if (!m->chunk_object_offsets)
150 die(_("multi-pack-index missing required object offsets chunk"));
32f3c541 151
d7cacf29
DS
152 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
153
3227565c 154 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
3715a633 155 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
3227565c
DS
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
4d80560c
DS
171 return m;
172
173cleanup_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
a40498a1
DS
183static void close_midx(struct multi_pack_index *m)
184{
185 uint32_t i;
186 munmap((unsigned char *)m->data, m->data_len);
187 close(m->fd);
188 m->fd = -1;
189
190 for (i = 0; i < m->num_packs; i++) {
191 if (m->packs[i]) {
192 close_pack(m->packs[i]);
193 free(m->packs);
194 }
195 }
196 FREE_AND_NULL(m->packs);
197 FREE_AND_NULL(m->pack_names);
198}
199
3715a633
DS
200static int prepare_midx_pack(struct multi_pack_index *m, uint32_t pack_int_id)
201{
202 struct strbuf pack_name = STRBUF_INIT;
203
204 if (pack_int_id >= m->num_packs)
205 BUG("bad pack-int-id");
206
207 if (m->packs[pack_int_id])
208 return 0;
209
210 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
211 m->pack_names[pack_int_id]);
212
2cf489a3 213 m->packs[pack_int_id] = add_packed_git(pack_name.buf, pack_name.len, m->local);
3715a633
DS
214 strbuf_release(&pack_name);
215 return !m->packs[pack_int_id];
216}
217
218int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
219{
220 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
221 MIDX_HASH_LEN, result);
222}
223
8aac67a1
DS
224struct object_id *nth_midxed_object_oid(struct object_id *oid,
225 struct multi_pack_index *m,
226 uint32_t n)
227{
228 if (n >= m->num_objects)
229 return NULL;
230
231 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
232 return oid;
233}
234
3715a633
DS
235static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
236{
237 const unsigned char *offset_data;
238 uint32_t offset32;
239
240 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
241 offset32 = get_be32(offset_data + sizeof(uint32_t));
242
243 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
244 if (sizeof(offset32) < sizeof(uint64_t))
245 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
246
247 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
248 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
249 }
250
251 return offset32;
252}
253
254static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
255{
256 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
257}
258
259static int nth_midxed_pack_entry(struct multi_pack_index *m, struct pack_entry *e, uint32_t pos)
260{
261 uint32_t pack_int_id;
262 struct packed_git *p;
263
264 if (pos >= m->num_objects)
265 return 0;
266
267 pack_int_id = nth_midxed_pack_int_id(m, pos);
268
269 if (prepare_midx_pack(m, pack_int_id))
270 die(_("error preparing packfile from multi-pack-index"));
271 p = m->packs[pack_int_id];
272
273 /*
274 * We are about to tell the caller where they can locate the
275 * requested object. We better make sure the packfile is
276 * still here and can be accessed before supplying that
277 * answer, as it may have been deleted since the MIDX was
278 * loaded!
279 */
280 if (!is_pack_valid(p))
281 return 0;
282
283 e->offset = nth_midxed_offset(m, pos);
284 e->p = p;
285
286 return 1;
287}
288
289int fill_midx_entry(const struct object_id *oid, struct pack_entry *e, struct multi_pack_index *m)
290{
291 uint32_t pos;
292
293 if (!bsearch_midx(oid, m, &pos))
294 return 0;
295
296 return nth_midxed_pack_entry(m, e, pos);
297}
298
a40498a1
DS
299int midx_contains_pack(struct multi_pack_index *m, const char *idx_name)
300{
301 uint32_t first = 0, last = m->num_packs;
302
303 while (first < last) {
304 uint32_t mid = first + (last - first) / 2;
305 const char *current;
306 int cmp;
307
308 current = m->pack_names[mid];
309 cmp = strcmp(idx_name, current);
310 if (!cmp)
311 return 1;
312 if (cmp > 0) {
313 first = mid + 1;
314 continue;
315 }
316 last = mid;
317 }
318
319 return 0;
320}
321
2cf489a3 322int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
c4d25228
DS
323{
324 struct multi_pack_index *m = r->objects->multi_pack_index;
325 struct multi_pack_index *m_search;
326 int config_value;
327
328 if (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
329 !config_value)
330 return 0;
331
332 for (m_search = m; m_search; m_search = m_search->next)
333 if (!strcmp(object_dir, m_search->object_dir))
334 return 1;
335
2cf489a3 336 r->objects->multi_pack_index = load_multi_pack_index(object_dir, local);
c4d25228
DS
337
338 if (r->objects->multi_pack_index) {
339 r->objects->multi_pack_index->next = m;
340 return 1;
341 }
342
343 return 0;
344}
345
fc59e748
DS
346static size_t write_midx_header(struct hashfile *f,
347 unsigned char num_chunks,
348 uint32_t num_packs)
349{
350 unsigned char byte_values[4];
351
352 hashwrite_be32(f, MIDX_SIGNATURE);
353 byte_values[0] = MIDX_VERSION;
354 byte_values[1] = MIDX_HASH_VERSION;
355 byte_values[2] = num_chunks;
356 byte_values[3] = 0; /* unused */
357 hashwrite(f, byte_values, sizeof(byte_values));
358 hashwrite_be32(f, num_packs);
359
360 return MIDX_HEADER_SIZE;
361}
362
396f2570
DS
363struct pack_list {
364 struct packed_git **list;
32f3c541 365 char **names;
396f2570
DS
366 uint32_t nr;
367 uint32_t alloc_list;
32f3c541
DS
368 uint32_t alloc_names;
369 size_t pack_name_concat_len;
a40498a1 370 struct multi_pack_index *m;
396f2570
DS
371};
372
373static void add_pack_to_midx(const char *full_path, size_t full_path_len,
374 const char *file_name, void *data)
375{
376 struct pack_list *packs = (struct pack_list *)data;
377
378 if (ends_with(file_name, ".idx")) {
a40498a1
DS
379 if (packs->m && midx_contains_pack(packs->m, file_name))
380 return;
381
396f2570 382 ALLOC_GROW(packs->list, packs->nr + 1, packs->alloc_list);
32f3c541 383 ALLOC_GROW(packs->names, packs->nr + 1, packs->alloc_names);
396f2570
DS
384
385 packs->list[packs->nr] = add_packed_git(full_path,
386 full_path_len,
387 0);
fe1ed56f 388
396f2570
DS
389 if (!packs->list[packs->nr]) {
390 warning(_("failed to add packfile '%s'"),
391 full_path);
392 return;
393 }
394
fe1ed56f
DS
395 if (open_pack_index(packs->list[packs->nr])) {
396 warning(_("failed to open pack-index '%s'"),
397 full_path);
398 close_pack(packs->list[packs->nr]);
399 FREE_AND_NULL(packs->list[packs->nr]);
400 return;
401 }
402
32f3c541
DS
403 packs->names[packs->nr] = xstrdup(file_name);
404 packs->pack_name_concat_len += strlen(file_name) + 1;
396f2570
DS
405 packs->nr++;
406 }
407}
408
32f3c541
DS
409struct pack_pair {
410 uint32_t pack_int_id;
411 char *pack_name;
412};
413
414static int pack_pair_compare(const void *_a, const void *_b)
415{
416 struct pack_pair *a = (struct pack_pair *)_a;
417 struct pack_pair *b = (struct pack_pair *)_b;
418 return strcmp(a->pack_name, b->pack_name);
419}
420
421static void sort_packs_by_name(char **pack_names, uint32_t nr_packs, uint32_t *perm)
422{
423 uint32_t i;
424 struct pack_pair *pairs;
425
426 ALLOC_ARRAY(pairs, nr_packs);
427
428 for (i = 0; i < nr_packs; i++) {
429 pairs[i].pack_int_id = i;
430 pairs[i].pack_name = pack_names[i];
431 }
432
433 QSORT(pairs, nr_packs, pack_pair_compare);
434
435 for (i = 0; i < nr_packs; i++) {
436 pack_names[i] = pairs[i].pack_name;
437 perm[pairs[i].pack_int_id] = i;
438 }
439
440 free(pairs);
441}
442
fe1ed56f
DS
443struct pack_midx_entry {
444 struct object_id oid;
445 uint32_t pack_int_id;
446 time_t pack_mtime;
447 uint64_t offset;
448};
449
450static int midx_oid_compare(const void *_a, const void *_b)
451{
452 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
453 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
454 int cmp = oidcmp(&a->oid, &b->oid);
455
456 if (cmp)
457 return cmp;
458
459 if (a->pack_mtime > b->pack_mtime)
460 return -1;
461 else if (a->pack_mtime < b->pack_mtime)
462 return 1;
463
464 return a->pack_int_id - b->pack_int_id;
465}
466
a40498a1
DS
467static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
468 uint32_t *pack_perm,
469 struct pack_midx_entry *e,
470 uint32_t pos)
471{
472 if (pos >= m->num_objects)
473 return 1;
474
475 nth_midxed_object_oid(&e->oid, m, pos);
476 e->pack_int_id = pack_perm[nth_midxed_pack_int_id(m, pos)];
477 e->offset = nth_midxed_offset(m, pos);
478
479 /* consider objects in midx to be from "old" packs */
480 e->pack_mtime = 0;
481 return 0;
482}
483
fe1ed56f
DS
484static void fill_pack_entry(uint32_t pack_int_id,
485 struct packed_git *p,
486 uint32_t cur_object,
487 struct pack_midx_entry *entry)
488{
489 if (!nth_packed_object_oid(&entry->oid, p, cur_object))
490 die(_("failed to locate object %d in packfile"), cur_object);
491
492 entry->pack_int_id = pack_int_id;
493 entry->pack_mtime = p->mtime;
494
495 entry->offset = nth_packed_object_offset(p, cur_object);
496}
497
498/*
499 * It is possible to artificially get into a state where there are many
500 * duplicate copies of objects. That can create high memory pressure if
501 * we are to create a list of all objects before de-duplication. To reduce
502 * this memory pressure without a significant performance drop, automatically
503 * group objects by the first byte of their object id. Use the IDX fanout
504 * tables to group the data, copy to a local array, then sort.
505 *
506 * Copy only the de-duplicated entries (selected by most-recent modified time
507 * of a packfile containing the object).
508 */
a40498a1
DS
509static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
510 struct packed_git **p,
fe1ed56f
DS
511 uint32_t *perm,
512 uint32_t nr_packs,
513 uint32_t *nr_objects)
514{
515 uint32_t cur_fanout, cur_pack, cur_object;
516 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
517 struct pack_midx_entry *entries_by_fanout = NULL;
518 struct pack_midx_entry *deduplicated_entries = NULL;
a40498a1 519 uint32_t start_pack = m ? m->num_packs : 0;
fe1ed56f 520
a40498a1 521 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
fe1ed56f
DS
522 total_objects += p[cur_pack]->num_objects;
523
524 /*
525 * As we de-duplicate by fanout value, we expect the fanout
526 * slices to be evenly distributed, with some noise. Hence,
527 * allocate slightly more than one 256th.
528 */
529 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
530
531 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
532 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
533 *nr_objects = 0;
534
535 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
536 uint32_t nr_fanout = 0;
537
a40498a1
DS
538 if (m) {
539 uint32_t start = 0, end;
540
541 if (cur_fanout)
542 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
543 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
544
545 for (cur_object = start; cur_object < end; cur_object++) {
546 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
547 nth_midxed_pack_midx_entry(m, perm,
548 &entries_by_fanout[nr_fanout],
549 cur_object);
550 nr_fanout++;
551 }
552 }
553
554 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
fe1ed56f
DS
555 uint32_t start = 0, end;
556
557 if (cur_fanout)
558 start = get_pack_fanout(p[cur_pack], cur_fanout - 1);
559 end = get_pack_fanout(p[cur_pack], cur_fanout);
560
561 for (cur_object = start; cur_object < end; cur_object++) {
562 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
563 fill_pack_entry(perm[cur_pack], p[cur_pack], cur_object, &entries_by_fanout[nr_fanout]);
564 nr_fanout++;
565 }
566 }
567
568 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
569
570 /*
571 * The batch is now sorted by OID and then mtime (descending).
572 * Take only the first duplicate.
573 */
574 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
575 if (cur_object && !oidcmp(&entries_by_fanout[cur_object - 1].oid,
576 &entries_by_fanout[cur_object].oid))
577 continue;
578
579 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
580 memcpy(&deduplicated_entries[*nr_objects],
581 &entries_by_fanout[cur_object],
582 sizeof(struct pack_midx_entry));
583 (*nr_objects)++;
584 }
585 }
586
587 free(entries_by_fanout);
588 return deduplicated_entries;
589}
590
32f3c541
DS
591static size_t write_midx_pack_names(struct hashfile *f,
592 char **pack_names,
593 uint32_t num_packs)
594{
595 uint32_t i;
596 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
597 size_t written = 0;
598
599 for (i = 0; i < num_packs; i++) {
600 size_t writelen = strlen(pack_names[i]) + 1;
601
602 if (i && strcmp(pack_names[i], pack_names[i - 1]) <= 0)
603 BUG("incorrect pack-file order: %s before %s",
604 pack_names[i - 1],
605 pack_names[i]);
606
607 hashwrite(f, pack_names[i], writelen);
608 written += writelen;
609 }
610
611 /* add padding to be aligned */
612 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
613 if (i < MIDX_CHUNK_ALIGNMENT) {
614 memset(padding, 0, sizeof(padding));
615 hashwrite(f, padding, i);
616 written += i;
617 }
618
619 return written;
620}
621
d7cacf29
DS
622static size_t write_midx_oid_fanout(struct hashfile *f,
623 struct pack_midx_entry *objects,
624 uint32_t nr_objects)
625{
626 struct pack_midx_entry *list = objects;
627 struct pack_midx_entry *last = objects + nr_objects;
628 uint32_t count = 0;
629 uint32_t i;
630
631 /*
632 * Write the first-level table (the list is sorted,
633 * but we use a 256-entry lookup to be able to avoid
634 * having to do eight extra binary search iterations).
635 */
636 for (i = 0; i < 256; i++) {
637 struct pack_midx_entry *next = list;
638
639 while (next < last && next->oid.hash[0] == i) {
640 count++;
641 next++;
642 }
643
644 hashwrite_be32(f, count);
645 list = next;
646 }
647
648 return MIDX_CHUNK_FANOUT_SIZE;
649}
650
0d5b3a5e
DS
651static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
652 struct pack_midx_entry *objects,
653 uint32_t nr_objects)
654{
655 struct pack_midx_entry *list = objects;
656 uint32_t i;
657 size_t written = 0;
658
659 for (i = 0; i < nr_objects; i++) {
660 struct pack_midx_entry *obj = list++;
661
662 if (i < nr_objects - 1) {
663 struct pack_midx_entry *next = list;
664 if (oidcmp(&obj->oid, &next->oid) >= 0)
665 BUG("OIDs not in order: %s >= %s",
666 oid_to_hex(&obj->oid),
667 oid_to_hex(&next->oid));
668 }
669
670 hashwrite(f, obj->oid.hash, (int)hash_len);
671 written += hash_len;
672 }
673
674 return written;
675}
676
662148c4
DS
677static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
678 struct pack_midx_entry *objects, uint32_t nr_objects)
679{
680 struct pack_midx_entry *list = objects;
681 uint32_t i, nr_large_offset = 0;
682 size_t written = 0;
683
684 for (i = 0; i < nr_objects; i++) {
685 struct pack_midx_entry *obj = list++;
686
687 hashwrite_be32(f, obj->pack_int_id);
688
689 if (large_offset_needed && obj->offset >> 31)
690 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
691 else if (!large_offset_needed && obj->offset >> 32)
692 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
693 oid_to_hex(&obj->oid),
694 obj->offset);
695 else
696 hashwrite_be32(f, (uint32_t)obj->offset);
697
698 written += MIDX_CHUNK_OFFSET_WIDTH;
699 }
700
701 return written;
702}
703
704static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
705 struct pack_midx_entry *objects, uint32_t nr_objects)
706{
707 struct pack_midx_entry *list = objects;
708 size_t written = 0;
709
710 while (nr_large_offset) {
711 struct pack_midx_entry *obj = list++;
712 uint64_t offset = obj->offset;
713
714 if (!(offset >> 31))
715 continue;
716
717 hashwrite_be32(f, offset >> 32);
718 hashwrite_be32(f, offset & 0xffffffffUL);
719 written += 2 * sizeof(uint32_t);
720
721 nr_large_offset--;
722 }
723
724 return written;
725}
726
a3407730
DS
727int write_midx_file(const char *object_dir)
728{
32f3c541 729 unsigned char cur_chunk, num_chunks = 0;
fc59e748 730 char *midx_name;
396f2570 731 uint32_t i;
fc59e748
DS
732 struct hashfile *f = NULL;
733 struct lock_file lk;
396f2570 734 struct pack_list packs;
32f3c541
DS
735 uint32_t *pack_perm = NULL;
736 uint64_t written = 0;
737 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
738 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
662148c4 739 uint32_t nr_entries, num_large_offsets = 0;
fe1ed56f 740 struct pack_midx_entry *entries = NULL;
662148c4 741 int large_offsets_needed = 0;
fc59e748
DS
742
743 midx_name = get_midx_filename(object_dir);
744 if (safe_create_leading_directories(midx_name)) {
745 UNLEAK(midx_name);
746 die_errno(_("unable to create leading directories of %s"),
747 midx_name);
748 }
749
2cf489a3 750 packs.m = load_multi_pack_index(object_dir, 1);
a40498a1 751
396f2570 752 packs.nr = 0;
a40498a1
DS
753 packs.alloc_list = packs.m ? packs.m->num_packs : 16;
754 packs.alloc_names = packs.alloc_list;
396f2570 755 packs.list = NULL;
a40498a1 756 packs.names = NULL;
32f3c541 757 packs.pack_name_concat_len = 0;
396f2570 758 ALLOC_ARRAY(packs.list, packs.alloc_list);
32f3c541 759 ALLOC_ARRAY(packs.names, packs.alloc_names);
396f2570 760
a40498a1
DS
761 if (packs.m) {
762 for (i = 0; i < packs.m->num_packs; i++) {
763 ALLOC_GROW(packs.list, packs.nr + 1, packs.alloc_list);
764 ALLOC_GROW(packs.names, packs.nr + 1, packs.alloc_names);
765
766 packs.list[packs.nr] = NULL;
767 packs.names[packs.nr] = xstrdup(packs.m->pack_names[i]);
768 packs.pack_name_concat_len += strlen(packs.names[packs.nr]) + 1;
769 packs.nr++;
770 }
771 }
772
396f2570
DS
773 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
774
a40498a1
DS
775 if (packs.m && packs.nr == packs.m->num_packs)
776 goto cleanup;
777
32f3c541
DS
778 if (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
779 packs.pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
780 (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
781
782 ALLOC_ARRAY(pack_perm, packs.nr);
783 sort_packs_by_name(packs.names, packs.nr, pack_perm);
784
a40498a1
DS
785 entries = get_sorted_entries(packs.m, packs.list, pack_perm, packs.nr, &nr_entries);
786
662148c4
DS
787 for (i = 0; i < nr_entries; i++) {
788 if (entries[i].offset > 0x7fffffff)
789 num_large_offsets++;
790 if (entries[i].offset > 0xffffffff)
791 large_offsets_needed = 1;
792 }
fe1ed56f 793
fc59e748
DS
794 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
795 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
796 FREE_AND_NULL(midx_name);
797
a40498a1
DS
798 if (packs.m)
799 close_midx(packs.m);
800
32f3c541 801 cur_chunk = 0;
662148c4 802 num_chunks = large_offsets_needed ? 5 : 4;
32f3c541
DS
803
804 written = write_midx_header(f, num_chunks, packs.nr);
805
806 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
807 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
808
809 cur_chunk++;
d7cacf29 810 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
32f3c541
DS
811 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + packs.pack_name_concat_len;
812
d7cacf29
DS
813 cur_chunk++;
814 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
815 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
816
0d5b3a5e 817 cur_chunk++;
662148c4 818 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
0d5b3a5e
DS
819 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_HASH_LEN;
820
662148c4
DS
821 cur_chunk++;
822 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
823 if (large_offsets_needed) {
824 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
825
826 cur_chunk++;
827 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
828 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
829 }
830
831 chunk_ids[cur_chunk] = 0;
832
32f3c541
DS
833 for (i = 0; i <= num_chunks; i++) {
834 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
835 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
836 chunk_offsets[i - 1],
837 chunk_offsets[i]);
838
839 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
840 BUG("chunk offset %"PRIu64" is not properly aligned",
841 chunk_offsets[i]);
842
843 hashwrite_be32(f, chunk_ids[i]);
844 hashwrite_be32(f, chunk_offsets[i] >> 32);
845 hashwrite_be32(f, chunk_offsets[i]);
846
847 written += MIDX_CHUNKLOOKUP_WIDTH;
848 }
849
850 for (i = 0; i < num_chunks; i++) {
851 if (written != chunk_offsets[i])
852 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
853 chunk_offsets[i],
854 written,
855 chunk_ids[i]);
856
857 switch (chunk_ids[i]) {
858 case MIDX_CHUNKID_PACKNAMES:
859 written += write_midx_pack_names(f, packs.names, packs.nr);
860 break;
861
d7cacf29
DS
862 case MIDX_CHUNKID_OIDFANOUT:
863 written += write_midx_oid_fanout(f, entries, nr_entries);
864 break;
865
0d5b3a5e
DS
866 case MIDX_CHUNKID_OIDLOOKUP:
867 written += write_midx_oid_lookup(f, MIDX_HASH_LEN, entries, nr_entries);
868 break;
869
662148c4
DS
870 case MIDX_CHUNKID_OBJECTOFFSETS:
871 written += write_midx_object_offsets(f, large_offsets_needed, entries, nr_entries);
872 break;
873
874 case MIDX_CHUNKID_LARGEOFFSETS:
875 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
876 break;
877
32f3c541
DS
878 default:
879 BUG("trying to write unknown chunk id %"PRIx32,
880 chunk_ids[i]);
881 }
882 }
883
884 if (written != chunk_offsets[num_chunks])
885 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
886 written,
887 chunk_offsets[num_chunks]);
fc59e748
DS
888
889 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
890 commit_lock_file(&lk);
891
a40498a1 892cleanup:
396f2570
DS
893 for (i = 0; i < packs.nr; i++) {
894 if (packs.list[i]) {
895 close_pack(packs.list[i]);
896 free(packs.list[i]);
897 }
32f3c541 898 free(packs.names[i]);
396f2570
DS
899 }
900
901 free(packs.list);
32f3c541 902 free(packs.names);
fe1ed56f 903 free(entries);
a40498a1
DS
904 free(pack_perm);
905 free(midx_name);
a3407730
DS
906 return 0;
907}
525e18c0
DS
908
909void clear_midx_file(const char *object_dir)
910{
911 char *midx = get_midx_filename(object_dir);
912
913 if (remove_path(midx)) {
914 UNLEAK(midx);
915 die(_("failed to clear multi-pack-index at %s"), midx);
916 }
917
918 free(midx);
919}