]> git.ipfire.org Git - thirdparty/git.git/blob - midx.c
Merge branch 'en/merge-ort-api-null-impl'
[thirdparty/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 #include "progress.h"
11 #include "trace2.h"
12 #include "run-command.h"
13 #include "repository.h"
14
15 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
16 #define MIDX_VERSION 1
17 #define MIDX_BYTE_FILE_VERSION 4
18 #define MIDX_BYTE_HASH_VERSION 5
19 #define MIDX_BYTE_NUM_CHUNKS 6
20 #define MIDX_BYTE_NUM_PACKS 8
21 #define MIDX_HEADER_SIZE 12
22 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
23
24 #define MIDX_MAX_CHUNKS 5
25 #define MIDX_CHUNK_ALIGNMENT 4
26 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
27 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
28 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
29 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
30 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
31 #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
32 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
33 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
34 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
35 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
36
37 #define PACK_EXPIRED UINT_MAX
38
39 static uint8_t oid_version(void)
40 {
41 switch (hash_algo_by_ptr(the_hash_algo)) {
42 case GIT_HASH_SHA1:
43 return 1;
44 case GIT_HASH_SHA256:
45 return 2;
46 default:
47 die(_("invalid hash version"));
48 }
49 }
50
51 static char *get_midx_filename(const char *object_dir)
52 {
53 return xstrfmt("%s/pack/multi-pack-index", object_dir);
54 }
55
56 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
57 {
58 struct multi_pack_index *m = NULL;
59 int fd;
60 struct stat st;
61 size_t midx_size;
62 void *midx_map = NULL;
63 uint32_t hash_version;
64 char *midx_name = get_midx_filename(object_dir);
65 uint32_t i;
66 const char *cur_pack_name;
67
68 fd = git_open(midx_name);
69
70 if (fd < 0)
71 goto cleanup_fail;
72 if (fstat(fd, &st)) {
73 error_errno(_("failed to read %s"), midx_name);
74 goto cleanup_fail;
75 }
76
77 midx_size = xsize_t(st.st_size);
78
79 if (midx_size < MIDX_MIN_SIZE) {
80 error(_("multi-pack-index file %s is too small"), midx_name);
81 goto cleanup_fail;
82 }
83
84 FREE_AND_NULL(midx_name);
85
86 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
87 close(fd);
88
89 FLEX_ALLOC_STR(m, object_dir, object_dir);
90 m->data = midx_map;
91 m->data_len = midx_size;
92 m->local = local;
93
94 m->signature = get_be32(m->data);
95 if (m->signature != MIDX_SIGNATURE)
96 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
97 m->signature, MIDX_SIGNATURE);
98
99 m->version = m->data[MIDX_BYTE_FILE_VERSION];
100 if (m->version != MIDX_VERSION)
101 die(_("multi-pack-index version %d not recognized"),
102 m->version);
103
104 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
105 if (hash_version != oid_version()) {
106 error(_("multi-pack-index hash version %u does not match version %u"),
107 hash_version, oid_version());
108 goto cleanup_fail;
109 }
110 m->hash_len = the_hash_algo->rawsz;
111
112 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
113
114 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
115
116 for (i = 0; i < m->num_chunks; i++) {
117 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
118 MIDX_CHUNKLOOKUP_WIDTH * i);
119 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
120 MIDX_CHUNKLOOKUP_WIDTH * i);
121
122 if (chunk_offset >= m->data_len)
123 die(_("invalid chunk offset (too large)"));
124
125 switch (chunk_id) {
126 case MIDX_CHUNKID_PACKNAMES:
127 m->chunk_pack_names = m->data + chunk_offset;
128 break;
129
130 case MIDX_CHUNKID_OIDFANOUT:
131 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
132 break;
133
134 case MIDX_CHUNKID_OIDLOOKUP:
135 m->chunk_oid_lookup = m->data + chunk_offset;
136 break;
137
138 case MIDX_CHUNKID_OBJECTOFFSETS:
139 m->chunk_object_offsets = m->data + chunk_offset;
140 break;
141
142 case MIDX_CHUNKID_LARGEOFFSETS:
143 m->chunk_large_offsets = m->data + chunk_offset;
144 break;
145
146 case 0:
147 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
148 break;
149
150 default:
151 /*
152 * Do nothing on unrecognized chunks, allowing future
153 * extensions to add optional chunks.
154 */
155 break;
156 }
157 }
158
159 if (!m->chunk_pack_names)
160 die(_("multi-pack-index missing required pack-name chunk"));
161 if (!m->chunk_oid_fanout)
162 die(_("multi-pack-index missing required OID fanout chunk"));
163 if (!m->chunk_oid_lookup)
164 die(_("multi-pack-index missing required OID lookup chunk"));
165 if (!m->chunk_object_offsets)
166 die(_("multi-pack-index missing required object offsets chunk"));
167
168 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
169
170 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
171 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
172
173 cur_pack_name = (const char *)m->chunk_pack_names;
174 for (i = 0; i < m->num_packs; i++) {
175 m->pack_names[i] = cur_pack_name;
176
177 cur_pack_name += strlen(cur_pack_name) + 1;
178
179 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
180 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
181 m->pack_names[i - 1],
182 m->pack_names[i]);
183 }
184
185 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
186 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
187
188 return m;
189
190 cleanup_fail:
191 free(m);
192 free(midx_name);
193 if (midx_map)
194 munmap(midx_map, midx_size);
195 if (0 <= fd)
196 close(fd);
197 return NULL;
198 }
199
200 void close_midx(struct multi_pack_index *m)
201 {
202 uint32_t i;
203
204 if (!m)
205 return;
206
207 munmap((unsigned char *)m->data, m->data_len);
208
209 for (i = 0; i < m->num_packs; i++) {
210 if (m->packs[i])
211 m->packs[i]->multi_pack_index = 0;
212 }
213 FREE_AND_NULL(m->packs);
214 FREE_AND_NULL(m->pack_names);
215 }
216
217 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
218 {
219 struct strbuf pack_name = STRBUF_INIT;
220 struct packed_git *p;
221
222 if (pack_int_id >= m->num_packs)
223 die(_("bad pack-int-id: %u (%u total packs)"),
224 pack_int_id, m->num_packs);
225
226 if (m->packs[pack_int_id])
227 return 0;
228
229 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
230 m->pack_names[pack_int_id]);
231
232 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
233 strbuf_release(&pack_name);
234
235 if (!p)
236 return 1;
237
238 p->multi_pack_index = 1;
239 m->packs[pack_int_id] = p;
240 install_packed_git(r, p);
241 list_add_tail(&p->mru, &r->objects->packed_git_mru);
242
243 return 0;
244 }
245
246 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
247 {
248 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
249 the_hash_algo->rawsz, result);
250 }
251
252 struct object_id *nth_midxed_object_oid(struct object_id *oid,
253 struct multi_pack_index *m,
254 uint32_t n)
255 {
256 if (n >= m->num_objects)
257 return NULL;
258
259 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
260 return oid;
261 }
262
263 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
264 {
265 const unsigned char *offset_data;
266 uint32_t offset32;
267
268 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
269 offset32 = get_be32(offset_data + sizeof(uint32_t));
270
271 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
272 if (sizeof(off_t) < sizeof(uint64_t))
273 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
274
275 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
276 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
277 }
278
279 return offset32;
280 }
281
282 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
283 {
284 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
285 }
286
287 static int nth_midxed_pack_entry(struct repository *r,
288 struct multi_pack_index *m,
289 struct pack_entry *e,
290 uint32_t pos)
291 {
292 uint32_t pack_int_id;
293 struct packed_git *p;
294
295 if (pos >= m->num_objects)
296 return 0;
297
298 pack_int_id = nth_midxed_pack_int_id(m, pos);
299
300 if (prepare_midx_pack(r, m, pack_int_id))
301 die(_("error preparing packfile from multi-pack-index"));
302 p = m->packs[pack_int_id];
303
304 /*
305 * We are about to tell the caller where they can locate the
306 * requested object. We better make sure the packfile is
307 * still here and can be accessed before supplying that
308 * answer, as it may have been deleted since the MIDX was
309 * loaded!
310 */
311 if (!is_pack_valid(p))
312 return 0;
313
314 if (p->num_bad_objects) {
315 uint32_t i;
316 struct object_id oid;
317 nth_midxed_object_oid(&oid, m, pos);
318 for (i = 0; i < p->num_bad_objects; i++)
319 if (hasheq(oid.hash,
320 p->bad_object_sha1 + the_hash_algo->rawsz * i))
321 return 0;
322 }
323
324 e->offset = nth_midxed_offset(m, pos);
325 e->p = p;
326
327 return 1;
328 }
329
330 int fill_midx_entry(struct repository * r,
331 const struct object_id *oid,
332 struct pack_entry *e,
333 struct multi_pack_index *m)
334 {
335 uint32_t pos;
336
337 if (!bsearch_midx(oid, m, &pos))
338 return 0;
339
340 return nth_midxed_pack_entry(r, m, e, pos);
341 }
342
343 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
344 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
345 const char *idx_name)
346 {
347 /* Skip past any initial matching prefix. */
348 while (*idx_name && *idx_name == *idx_or_pack_name) {
349 idx_name++;
350 idx_or_pack_name++;
351 }
352
353 /*
354 * If we didn't match completely, we may have matched "pack-1234." and
355 * be left with "idx" and "pack" respectively, which is also OK. We do
356 * not have to check for "idx" and "idx", because that would have been
357 * a complete match (and in that case these strcmps will be false, but
358 * we'll correctly return 0 from the final strcmp() below.
359 *
360 * Technically this matches "fooidx" and "foopack", but we'd never have
361 * such names in the first place.
362 */
363 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
364 return 0;
365
366 /*
367 * This not only checks for a complete match, but also orders based on
368 * the first non-identical character, which means our ordering will
369 * match a raw strcmp(). That makes it OK to use this to binary search
370 * a naively-sorted list.
371 */
372 return strcmp(idx_or_pack_name, idx_name);
373 }
374
375 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
376 {
377 uint32_t first = 0, last = m->num_packs;
378
379 while (first < last) {
380 uint32_t mid = first + (last - first) / 2;
381 const char *current;
382 int cmp;
383
384 current = m->pack_names[mid];
385 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
386 if (!cmp)
387 return 1;
388 if (cmp > 0) {
389 first = mid + 1;
390 continue;
391 }
392 last = mid;
393 }
394
395 return 0;
396 }
397
398 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
399 {
400 struct multi_pack_index *m;
401 struct multi_pack_index *m_search;
402
403 prepare_repo_settings(r);
404 if (!r->settings.core_multi_pack_index)
405 return 0;
406
407 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
408 if (!strcmp(object_dir, m_search->object_dir))
409 return 1;
410
411 m = load_multi_pack_index(object_dir, local);
412
413 if (m) {
414 struct multi_pack_index *mp = r->objects->multi_pack_index;
415 if (mp) {
416 m->next = mp->next;
417 mp->next = m;
418 } else
419 r->objects->multi_pack_index = m;
420 return 1;
421 }
422
423 return 0;
424 }
425
426 static size_t write_midx_header(struct hashfile *f,
427 unsigned char num_chunks,
428 uint32_t num_packs)
429 {
430 hashwrite_be32(f, MIDX_SIGNATURE);
431 hashwrite_u8(f, MIDX_VERSION);
432 hashwrite_u8(f, oid_version());
433 hashwrite_u8(f, num_chunks);
434 hashwrite_u8(f, 0); /* unused */
435 hashwrite_be32(f, num_packs);
436
437 return MIDX_HEADER_SIZE;
438 }
439
440 struct pack_info {
441 uint32_t orig_pack_int_id;
442 char *pack_name;
443 struct packed_git *p;
444 unsigned expired : 1;
445 };
446
447 static int pack_info_compare(const void *_a, const void *_b)
448 {
449 struct pack_info *a = (struct pack_info *)_a;
450 struct pack_info *b = (struct pack_info *)_b;
451 return strcmp(a->pack_name, b->pack_name);
452 }
453
454 struct pack_list {
455 struct pack_info *info;
456 uint32_t nr;
457 uint32_t alloc;
458 struct multi_pack_index *m;
459 struct progress *progress;
460 unsigned pack_paths_checked;
461 };
462
463 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
464 const char *file_name, void *data)
465 {
466 struct pack_list *packs = (struct pack_list *)data;
467
468 if (ends_with(file_name, ".idx")) {
469 display_progress(packs->progress, ++packs->pack_paths_checked);
470 if (packs->m && midx_contains_pack(packs->m, file_name))
471 return;
472
473 ALLOC_GROW(packs->info, packs->nr + 1, packs->alloc);
474
475 packs->info[packs->nr].p = add_packed_git(full_path,
476 full_path_len,
477 0);
478
479 if (!packs->info[packs->nr].p) {
480 warning(_("failed to add packfile '%s'"),
481 full_path);
482 return;
483 }
484
485 if (open_pack_index(packs->info[packs->nr].p)) {
486 warning(_("failed to open pack-index '%s'"),
487 full_path);
488 close_pack(packs->info[packs->nr].p);
489 FREE_AND_NULL(packs->info[packs->nr].p);
490 return;
491 }
492
493 packs->info[packs->nr].pack_name = xstrdup(file_name);
494 packs->info[packs->nr].orig_pack_int_id = packs->nr;
495 packs->info[packs->nr].expired = 0;
496 packs->nr++;
497 }
498 }
499
500 struct pack_midx_entry {
501 struct object_id oid;
502 uint32_t pack_int_id;
503 time_t pack_mtime;
504 uint64_t offset;
505 };
506
507 static int midx_oid_compare(const void *_a, const void *_b)
508 {
509 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
510 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
511 int cmp = oidcmp(&a->oid, &b->oid);
512
513 if (cmp)
514 return cmp;
515
516 if (a->pack_mtime > b->pack_mtime)
517 return -1;
518 else if (a->pack_mtime < b->pack_mtime)
519 return 1;
520
521 return a->pack_int_id - b->pack_int_id;
522 }
523
524 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
525 struct pack_midx_entry *e,
526 uint32_t pos)
527 {
528 if (pos >= m->num_objects)
529 return 1;
530
531 nth_midxed_object_oid(&e->oid, m, pos);
532 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
533 e->offset = nth_midxed_offset(m, pos);
534
535 /* consider objects in midx to be from "old" packs */
536 e->pack_mtime = 0;
537 return 0;
538 }
539
540 static void fill_pack_entry(uint32_t pack_int_id,
541 struct packed_git *p,
542 uint32_t cur_object,
543 struct pack_midx_entry *entry)
544 {
545 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
546 die(_("failed to locate object %d in packfile"), cur_object);
547
548 entry->pack_int_id = pack_int_id;
549 entry->pack_mtime = p->mtime;
550
551 entry->offset = nth_packed_object_offset(p, cur_object);
552 }
553
554 /*
555 * It is possible to artificially get into a state where there are many
556 * duplicate copies of objects. That can create high memory pressure if
557 * we are to create a list of all objects before de-duplication. To reduce
558 * this memory pressure without a significant performance drop, automatically
559 * group objects by the first byte of their object id. Use the IDX fanout
560 * tables to group the data, copy to a local array, then sort.
561 *
562 * Copy only the de-duplicated entries (selected by most-recent modified time
563 * of a packfile containing the object).
564 */
565 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
566 struct pack_info *info,
567 uint32_t nr_packs,
568 uint32_t *nr_objects)
569 {
570 uint32_t cur_fanout, cur_pack, cur_object;
571 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
572 struct pack_midx_entry *entries_by_fanout = NULL;
573 struct pack_midx_entry *deduplicated_entries = NULL;
574 uint32_t start_pack = m ? m->num_packs : 0;
575
576 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
577 total_objects += info[cur_pack].p->num_objects;
578
579 /*
580 * As we de-duplicate by fanout value, we expect the fanout
581 * slices to be evenly distributed, with some noise. Hence,
582 * allocate slightly more than one 256th.
583 */
584 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
585
586 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
587 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
588 *nr_objects = 0;
589
590 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
591 uint32_t nr_fanout = 0;
592
593 if (m) {
594 uint32_t start = 0, end;
595
596 if (cur_fanout)
597 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
598 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
599
600 for (cur_object = start; cur_object < end; cur_object++) {
601 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
602 nth_midxed_pack_midx_entry(m,
603 &entries_by_fanout[nr_fanout],
604 cur_object);
605 nr_fanout++;
606 }
607 }
608
609 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
610 uint32_t start = 0, end;
611
612 if (cur_fanout)
613 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
614 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
615
616 for (cur_object = start; cur_object < end; cur_object++) {
617 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
618 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
619 nr_fanout++;
620 }
621 }
622
623 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
624
625 /*
626 * The batch is now sorted by OID and then mtime (descending).
627 * Take only the first duplicate.
628 */
629 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
630 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
631 &entries_by_fanout[cur_object].oid))
632 continue;
633
634 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
635 memcpy(&deduplicated_entries[*nr_objects],
636 &entries_by_fanout[cur_object],
637 sizeof(struct pack_midx_entry));
638 (*nr_objects)++;
639 }
640 }
641
642 free(entries_by_fanout);
643 return deduplicated_entries;
644 }
645
646 static size_t write_midx_pack_names(struct hashfile *f,
647 struct pack_info *info,
648 uint32_t num_packs)
649 {
650 uint32_t i;
651 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
652 size_t written = 0;
653
654 for (i = 0; i < num_packs; i++) {
655 size_t writelen;
656
657 if (info[i].expired)
658 continue;
659
660 if (i && strcmp(info[i].pack_name, info[i - 1].pack_name) <= 0)
661 BUG("incorrect pack-file order: %s before %s",
662 info[i - 1].pack_name,
663 info[i].pack_name);
664
665 writelen = strlen(info[i].pack_name) + 1;
666 hashwrite(f, info[i].pack_name, writelen);
667 written += writelen;
668 }
669
670 /* add padding to be aligned */
671 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
672 if (i < MIDX_CHUNK_ALIGNMENT) {
673 memset(padding, 0, sizeof(padding));
674 hashwrite(f, padding, i);
675 written += i;
676 }
677
678 return written;
679 }
680
681 static size_t write_midx_oid_fanout(struct hashfile *f,
682 struct pack_midx_entry *objects,
683 uint32_t nr_objects)
684 {
685 struct pack_midx_entry *list = objects;
686 struct pack_midx_entry *last = objects + nr_objects;
687 uint32_t count = 0;
688 uint32_t i;
689
690 /*
691 * Write the first-level table (the list is sorted,
692 * but we use a 256-entry lookup to be able to avoid
693 * having to do eight extra binary search iterations).
694 */
695 for (i = 0; i < 256; i++) {
696 struct pack_midx_entry *next = list;
697
698 while (next < last && next->oid.hash[0] == i) {
699 count++;
700 next++;
701 }
702
703 hashwrite_be32(f, count);
704 list = next;
705 }
706
707 return MIDX_CHUNK_FANOUT_SIZE;
708 }
709
710 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
711 struct pack_midx_entry *objects,
712 uint32_t nr_objects)
713 {
714 struct pack_midx_entry *list = objects;
715 uint32_t i;
716 size_t written = 0;
717
718 for (i = 0; i < nr_objects; i++) {
719 struct pack_midx_entry *obj = list++;
720
721 if (i < nr_objects - 1) {
722 struct pack_midx_entry *next = list;
723 if (oidcmp(&obj->oid, &next->oid) >= 0)
724 BUG("OIDs not in order: %s >= %s",
725 oid_to_hex(&obj->oid),
726 oid_to_hex(&next->oid));
727 }
728
729 hashwrite(f, obj->oid.hash, (int)hash_len);
730 written += hash_len;
731 }
732
733 return written;
734 }
735
736 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
737 uint32_t *perm,
738 struct pack_midx_entry *objects, uint32_t nr_objects)
739 {
740 struct pack_midx_entry *list = objects;
741 uint32_t i, nr_large_offset = 0;
742 size_t written = 0;
743
744 for (i = 0; i < nr_objects; i++) {
745 struct pack_midx_entry *obj = list++;
746
747 if (perm[obj->pack_int_id] == PACK_EXPIRED)
748 BUG("object %s is in an expired pack with int-id %d",
749 oid_to_hex(&obj->oid),
750 obj->pack_int_id);
751
752 hashwrite_be32(f, perm[obj->pack_int_id]);
753
754 if (large_offset_needed && obj->offset >> 31)
755 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
756 else if (!large_offset_needed && obj->offset >> 32)
757 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
758 oid_to_hex(&obj->oid),
759 obj->offset);
760 else
761 hashwrite_be32(f, (uint32_t)obj->offset);
762
763 written += MIDX_CHUNK_OFFSET_WIDTH;
764 }
765
766 return written;
767 }
768
769 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
770 struct pack_midx_entry *objects, uint32_t nr_objects)
771 {
772 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
773 size_t written = 0;
774
775 while (nr_large_offset) {
776 struct pack_midx_entry *obj;
777 uint64_t offset;
778
779 if (list >= end)
780 BUG("too many large-offset objects");
781
782 obj = list++;
783 offset = obj->offset;
784
785 if (!(offset >> 31))
786 continue;
787
788 hashwrite_be32(f, offset >> 32);
789 hashwrite_be32(f, offset & 0xffffffffUL);
790 written += 2 * sizeof(uint32_t);
791
792 nr_large_offset--;
793 }
794
795 return written;
796 }
797
798 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
799 struct string_list *packs_to_drop, unsigned flags)
800 {
801 unsigned char cur_chunk, num_chunks = 0;
802 char *midx_name;
803 uint32_t i;
804 struct hashfile *f = NULL;
805 struct lock_file lk;
806 struct pack_list packs;
807 uint32_t *pack_perm = NULL;
808 uint64_t written = 0;
809 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
810 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
811 uint32_t nr_entries, num_large_offsets = 0;
812 struct pack_midx_entry *entries = NULL;
813 struct progress *progress = NULL;
814 int large_offsets_needed = 0;
815 int pack_name_concat_len = 0;
816 int dropped_packs = 0;
817 int result = 0;
818
819 midx_name = get_midx_filename(object_dir);
820 if (safe_create_leading_directories(midx_name))
821 die_errno(_("unable to create leading directories of %s"),
822 midx_name);
823
824 if (m)
825 packs.m = m;
826 else
827 packs.m = load_multi_pack_index(object_dir, 1);
828
829 packs.nr = 0;
830 packs.alloc = packs.m ? packs.m->num_packs : 16;
831 packs.info = NULL;
832 ALLOC_ARRAY(packs.info, packs.alloc);
833
834 if (packs.m) {
835 for (i = 0; i < packs.m->num_packs; i++) {
836 ALLOC_GROW(packs.info, packs.nr + 1, packs.alloc);
837
838 packs.info[packs.nr].orig_pack_int_id = i;
839 packs.info[packs.nr].pack_name = xstrdup(packs.m->pack_names[i]);
840 packs.info[packs.nr].p = NULL;
841 packs.info[packs.nr].expired = 0;
842 packs.nr++;
843 }
844 }
845
846 packs.pack_paths_checked = 0;
847 if (flags & MIDX_PROGRESS)
848 packs.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
849 else
850 packs.progress = NULL;
851
852 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
853 stop_progress(&packs.progress);
854
855 if (packs.m && packs.nr == packs.m->num_packs && !packs_to_drop)
856 goto cleanup;
857
858 entries = get_sorted_entries(packs.m, packs.info, packs.nr, &nr_entries);
859
860 for (i = 0; i < nr_entries; i++) {
861 if (entries[i].offset > 0x7fffffff)
862 num_large_offsets++;
863 if (entries[i].offset > 0xffffffff)
864 large_offsets_needed = 1;
865 }
866
867 QSORT(packs.info, packs.nr, pack_info_compare);
868
869 if (packs_to_drop && packs_to_drop->nr) {
870 int drop_index = 0;
871 int missing_drops = 0;
872
873 for (i = 0; i < packs.nr && drop_index < packs_to_drop->nr; i++) {
874 int cmp = strcmp(packs.info[i].pack_name,
875 packs_to_drop->items[drop_index].string);
876
877 if (!cmp) {
878 drop_index++;
879 packs.info[i].expired = 1;
880 } else if (cmp > 0) {
881 error(_("did not see pack-file %s to drop"),
882 packs_to_drop->items[drop_index].string);
883 drop_index++;
884 missing_drops++;
885 i--;
886 } else {
887 packs.info[i].expired = 0;
888 }
889 }
890
891 if (missing_drops) {
892 result = 1;
893 goto cleanup;
894 }
895 }
896
897 /*
898 * pack_perm stores a permutation between pack-int-ids from the
899 * previous multi-pack-index to the new one we are writing:
900 *
901 * pack_perm[old_id] = new_id
902 */
903 ALLOC_ARRAY(pack_perm, packs.nr);
904 for (i = 0; i < packs.nr; i++) {
905 if (packs.info[i].expired) {
906 dropped_packs++;
907 pack_perm[packs.info[i].orig_pack_int_id] = PACK_EXPIRED;
908 } else {
909 pack_perm[packs.info[i].orig_pack_int_id] = i - dropped_packs;
910 }
911 }
912
913 for (i = 0; i < packs.nr; i++) {
914 if (!packs.info[i].expired)
915 pack_name_concat_len += strlen(packs.info[i].pack_name) + 1;
916 }
917
918 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
919 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
920 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
921
922 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
923 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
924 FREE_AND_NULL(midx_name);
925
926 if (packs.m)
927 close_midx(packs.m);
928
929 cur_chunk = 0;
930 num_chunks = large_offsets_needed ? 5 : 4;
931
932 if (packs.nr - dropped_packs == 0) {
933 error(_("no pack files to index."));
934 result = 1;
935 goto cleanup;
936 }
937
938 written = write_midx_header(f, num_chunks, packs.nr - dropped_packs);
939
940 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
941 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
942
943 cur_chunk++;
944 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
945 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + pack_name_concat_len;
946
947 cur_chunk++;
948 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
949 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
950
951 cur_chunk++;
952 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
953 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * the_hash_algo->rawsz;
954
955 cur_chunk++;
956 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
957 if (large_offsets_needed) {
958 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
959
960 cur_chunk++;
961 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
962 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
963 }
964
965 chunk_ids[cur_chunk] = 0;
966
967 for (i = 0; i <= num_chunks; i++) {
968 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
969 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
970 chunk_offsets[i - 1],
971 chunk_offsets[i]);
972
973 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
974 BUG("chunk offset %"PRIu64" is not properly aligned",
975 chunk_offsets[i]);
976
977 hashwrite_be32(f, chunk_ids[i]);
978 hashwrite_be32(f, chunk_offsets[i] >> 32);
979 hashwrite_be32(f, chunk_offsets[i]);
980
981 written += MIDX_CHUNKLOOKUP_WIDTH;
982 }
983
984 if (flags & MIDX_PROGRESS)
985 progress = start_delayed_progress(_("Writing chunks to multi-pack-index"),
986 num_chunks);
987 for (i = 0; i < num_chunks; i++) {
988 if (written != chunk_offsets[i])
989 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
990 chunk_offsets[i],
991 written,
992 chunk_ids[i]);
993
994 switch (chunk_ids[i]) {
995 case MIDX_CHUNKID_PACKNAMES:
996 written += write_midx_pack_names(f, packs.info, packs.nr);
997 break;
998
999 case MIDX_CHUNKID_OIDFANOUT:
1000 written += write_midx_oid_fanout(f, entries, nr_entries);
1001 break;
1002
1003 case MIDX_CHUNKID_OIDLOOKUP:
1004 written += write_midx_oid_lookup(f, the_hash_algo->rawsz, entries, nr_entries);
1005 break;
1006
1007 case MIDX_CHUNKID_OBJECTOFFSETS:
1008 written += write_midx_object_offsets(f, large_offsets_needed, pack_perm, entries, nr_entries);
1009 break;
1010
1011 case MIDX_CHUNKID_LARGEOFFSETS:
1012 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
1013 break;
1014
1015 default:
1016 BUG("trying to write unknown chunk id %"PRIx32,
1017 chunk_ids[i]);
1018 }
1019
1020 display_progress(progress, i + 1);
1021 }
1022 stop_progress(&progress);
1023
1024 if (written != chunk_offsets[num_chunks])
1025 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
1026 written,
1027 chunk_offsets[num_chunks]);
1028
1029 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1030 commit_lock_file(&lk);
1031
1032 cleanup:
1033 for (i = 0; i < packs.nr; i++) {
1034 if (packs.info[i].p) {
1035 close_pack(packs.info[i].p);
1036 free(packs.info[i].p);
1037 }
1038 free(packs.info[i].pack_name);
1039 }
1040
1041 free(packs.info);
1042 free(entries);
1043 free(pack_perm);
1044 free(midx_name);
1045 return result;
1046 }
1047
1048 int write_midx_file(const char *object_dir, unsigned flags)
1049 {
1050 return write_midx_internal(object_dir, NULL, NULL, flags);
1051 }
1052
1053 void clear_midx_file(struct repository *r)
1054 {
1055 char *midx = get_midx_filename(r->objects->odb->path);
1056
1057 if (r->objects && r->objects->multi_pack_index) {
1058 close_midx(r->objects->multi_pack_index);
1059 r->objects->multi_pack_index = NULL;
1060 }
1061
1062 if (remove_path(midx))
1063 die(_("failed to clear multi-pack-index at %s"), midx);
1064
1065 free(midx);
1066 }
1067
1068 static int verify_midx_error;
1069
1070 static void midx_report(const char *fmt, ...)
1071 {
1072 va_list ap;
1073 verify_midx_error = 1;
1074 va_start(ap, fmt);
1075 vfprintf(stderr, fmt, ap);
1076 fprintf(stderr, "\n");
1077 va_end(ap);
1078 }
1079
1080 struct pair_pos_vs_id
1081 {
1082 uint32_t pos;
1083 uint32_t pack_int_id;
1084 };
1085
1086 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1087 {
1088 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1089 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1090
1091 return b->pack_int_id - a->pack_int_id;
1092 }
1093
1094 /*
1095 * Limit calls to display_progress() for performance reasons.
1096 * The interval here was arbitrarily chosen.
1097 */
1098 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1099 #define midx_display_sparse_progress(progress, n) \
1100 do { \
1101 uint64_t _n = (n); \
1102 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1103 display_progress(progress, _n); \
1104 } while (0)
1105
1106 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1107 {
1108 struct pair_pos_vs_id *pairs = NULL;
1109 uint32_t i;
1110 struct progress *progress = NULL;
1111 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1112 verify_midx_error = 0;
1113
1114 if (!m) {
1115 int result = 0;
1116 struct stat sb;
1117 char *filename = get_midx_filename(object_dir);
1118 if (!stat(filename, &sb)) {
1119 error(_("multi-pack-index file exists, but failed to parse"));
1120 result = 1;
1121 }
1122 free(filename);
1123 return result;
1124 }
1125
1126 if (flags & MIDX_PROGRESS)
1127 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1128 m->num_packs);
1129 for (i = 0; i < m->num_packs; i++) {
1130 if (prepare_midx_pack(r, m, i))
1131 midx_report("failed to load pack in position %d", i);
1132
1133 display_progress(progress, i + 1);
1134 }
1135 stop_progress(&progress);
1136
1137 for (i = 0; i < 255; i++) {
1138 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1139 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1140
1141 if (oid_fanout1 > oid_fanout2)
1142 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1143 i, oid_fanout1, oid_fanout2, i + 1);
1144 }
1145
1146 if (m->num_objects == 0) {
1147 midx_report(_("the midx contains no oid"));
1148 /*
1149 * Remaining tests assume that we have objects, so we can
1150 * return here.
1151 */
1152 return verify_midx_error;
1153 }
1154
1155 if (flags & MIDX_PROGRESS)
1156 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1157 m->num_objects - 1);
1158 for (i = 0; i < m->num_objects - 1; i++) {
1159 struct object_id oid1, oid2;
1160
1161 nth_midxed_object_oid(&oid1, m, i);
1162 nth_midxed_object_oid(&oid2, m, i + 1);
1163
1164 if (oidcmp(&oid1, &oid2) >= 0)
1165 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1166 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1167
1168 midx_display_sparse_progress(progress, i + 1);
1169 }
1170 stop_progress(&progress);
1171
1172 /*
1173 * Create an array mapping each object to its packfile id. Sort it
1174 * to group the objects by packfile. Use this permutation to visit
1175 * each of the objects and only require 1 packfile to be open at a
1176 * time.
1177 */
1178 ALLOC_ARRAY(pairs, m->num_objects);
1179 for (i = 0; i < m->num_objects; i++) {
1180 pairs[i].pos = i;
1181 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1182 }
1183
1184 if (flags & MIDX_PROGRESS)
1185 progress = start_sparse_progress(_("Sorting objects by packfile"),
1186 m->num_objects);
1187 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1188 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1189 stop_progress(&progress);
1190
1191 if (flags & MIDX_PROGRESS)
1192 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1193 for (i = 0; i < m->num_objects; i++) {
1194 struct object_id oid;
1195 struct pack_entry e;
1196 off_t m_offset, p_offset;
1197
1198 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1199 m->packs[pairs[i-1].pack_int_id])
1200 {
1201 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1202 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1203 }
1204
1205 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1206
1207 if (!fill_midx_entry(r, &oid, &e, m)) {
1208 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1209 pairs[i].pos, oid_to_hex(&oid));
1210 continue;
1211 }
1212
1213 if (open_pack_index(e.p)) {
1214 midx_report(_("failed to load pack-index for packfile %s"),
1215 e.p->pack_name);
1216 break;
1217 }
1218
1219 m_offset = e.offset;
1220 p_offset = find_pack_entry_one(oid.hash, e.p);
1221
1222 if (m_offset != p_offset)
1223 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1224 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1225
1226 midx_display_sparse_progress(progress, i + 1);
1227 }
1228 stop_progress(&progress);
1229
1230 free(pairs);
1231
1232 return verify_midx_error;
1233 }
1234
1235 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1236 {
1237 uint32_t i, *count, result = 0;
1238 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1239 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1240 struct progress *progress = NULL;
1241
1242 if (!m)
1243 return 0;
1244
1245 count = xcalloc(m->num_packs, sizeof(uint32_t));
1246
1247 if (flags & MIDX_PROGRESS)
1248 progress = start_delayed_progress(_("Counting referenced objects"),
1249 m->num_objects);
1250 for (i = 0; i < m->num_objects; i++) {
1251 int pack_int_id = nth_midxed_pack_int_id(m, i);
1252 count[pack_int_id]++;
1253 display_progress(progress, i + 1);
1254 }
1255 stop_progress(&progress);
1256
1257 if (flags & MIDX_PROGRESS)
1258 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1259 m->num_packs);
1260 for (i = 0; i < m->num_packs; i++) {
1261 char *pack_name;
1262 display_progress(progress, i + 1);
1263
1264 if (count[i])
1265 continue;
1266
1267 if (prepare_midx_pack(r, m, i))
1268 continue;
1269
1270 if (m->packs[i]->pack_keep)
1271 continue;
1272
1273 pack_name = xstrdup(m->packs[i]->pack_name);
1274 close_pack(m->packs[i]);
1275
1276 string_list_insert(&packs_to_drop, m->pack_names[i]);
1277 unlink_pack_path(pack_name, 0);
1278 free(pack_name);
1279 }
1280 stop_progress(&progress);
1281
1282 free(count);
1283
1284 if (packs_to_drop.nr)
1285 result = write_midx_internal(object_dir, m, &packs_to_drop, flags);
1286
1287 string_list_clear(&packs_to_drop, 0);
1288 return result;
1289 }
1290
1291 struct repack_info {
1292 timestamp_t mtime;
1293 uint32_t referenced_objects;
1294 uint32_t pack_int_id;
1295 };
1296
1297 static int compare_by_mtime(const void *a_, const void *b_)
1298 {
1299 const struct repack_info *a, *b;
1300
1301 a = (const struct repack_info *)a_;
1302 b = (const struct repack_info *)b_;
1303
1304 if (a->mtime < b->mtime)
1305 return -1;
1306 if (a->mtime > b->mtime)
1307 return 1;
1308 return 0;
1309 }
1310
1311 static int fill_included_packs_all(struct repository *r,
1312 struct multi_pack_index *m,
1313 unsigned char *include_pack)
1314 {
1315 uint32_t i, count = 0;
1316 int pack_kept_objects = 0;
1317
1318 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1319
1320 for (i = 0; i < m->num_packs; i++) {
1321 if (prepare_midx_pack(r, m, i))
1322 continue;
1323 if (!pack_kept_objects && m->packs[i]->pack_keep)
1324 continue;
1325
1326 include_pack[i] = 1;
1327 count++;
1328 }
1329
1330 return count < 2;
1331 }
1332
1333 static int fill_included_packs_batch(struct repository *r,
1334 struct multi_pack_index *m,
1335 unsigned char *include_pack,
1336 size_t batch_size)
1337 {
1338 uint32_t i, packs_to_repack;
1339 size_t total_size;
1340 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1341 int pack_kept_objects = 0;
1342
1343 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1344
1345 for (i = 0; i < m->num_packs; i++) {
1346 pack_info[i].pack_int_id = i;
1347
1348 if (prepare_midx_pack(r, m, i))
1349 continue;
1350
1351 pack_info[i].mtime = m->packs[i]->mtime;
1352 }
1353
1354 for (i = 0; batch_size && i < m->num_objects; i++) {
1355 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1356 pack_info[pack_int_id].referenced_objects++;
1357 }
1358
1359 QSORT(pack_info, m->num_packs, compare_by_mtime);
1360
1361 total_size = 0;
1362 packs_to_repack = 0;
1363 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1364 int pack_int_id = pack_info[i].pack_int_id;
1365 struct packed_git *p = m->packs[pack_int_id];
1366 size_t expected_size;
1367
1368 if (!p)
1369 continue;
1370 if (!pack_kept_objects && p->pack_keep)
1371 continue;
1372 if (open_pack_index(p) || !p->num_objects)
1373 continue;
1374
1375 expected_size = (size_t)(p->pack_size
1376 * pack_info[i].referenced_objects);
1377 expected_size /= p->num_objects;
1378
1379 if (expected_size >= batch_size)
1380 continue;
1381
1382 packs_to_repack++;
1383 total_size += expected_size;
1384 include_pack[pack_int_id] = 1;
1385 }
1386
1387 free(pack_info);
1388
1389 if (packs_to_repack < 2)
1390 return 1;
1391
1392 return 0;
1393 }
1394
1395 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1396 {
1397 int result = 0;
1398 uint32_t i;
1399 unsigned char *include_pack;
1400 struct child_process cmd = CHILD_PROCESS_INIT;
1401 FILE *cmd_in;
1402 struct strbuf base_name = STRBUF_INIT;
1403 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1404
1405 /*
1406 * When updating the default for these configuration
1407 * variables in builtin/repack.c, these must be adjusted
1408 * to match.
1409 */
1410 int delta_base_offset = 1;
1411 int use_delta_islands = 0;
1412
1413 if (!m)
1414 return 0;
1415
1416 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1417
1418 if (batch_size) {
1419 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1420 goto cleanup;
1421 } else if (fill_included_packs_all(r, m, include_pack))
1422 goto cleanup;
1423
1424 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1425 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1426
1427 strvec_push(&cmd.args, "pack-objects");
1428
1429 strbuf_addstr(&base_name, object_dir);
1430 strbuf_addstr(&base_name, "/pack/pack");
1431 strvec_push(&cmd.args, base_name.buf);
1432
1433 if (delta_base_offset)
1434 strvec_push(&cmd.args, "--delta-base-offset");
1435 if (use_delta_islands)
1436 strvec_push(&cmd.args, "--delta-islands");
1437
1438 if (flags & MIDX_PROGRESS)
1439 strvec_push(&cmd.args, "--progress");
1440 else
1441 strvec_push(&cmd.args, "-q");
1442
1443 strbuf_release(&base_name);
1444
1445 cmd.git_cmd = 1;
1446 cmd.in = cmd.out = -1;
1447
1448 if (start_command(&cmd)) {
1449 error(_("could not start pack-objects"));
1450 result = 1;
1451 goto cleanup;
1452 }
1453
1454 cmd_in = xfdopen(cmd.in, "w");
1455
1456 for (i = 0; i < m->num_objects; i++) {
1457 struct object_id oid;
1458 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1459
1460 if (!include_pack[pack_int_id])
1461 continue;
1462
1463 nth_midxed_object_oid(&oid, m, i);
1464 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
1465 }
1466 fclose(cmd_in);
1467
1468 if (finish_command(&cmd)) {
1469 error(_("could not finish pack-objects"));
1470 result = 1;
1471 goto cleanup;
1472 }
1473
1474 result = write_midx_internal(object_dir, m, NULL, flags);
1475 m = NULL;
1476
1477 cleanup:
1478 if (m)
1479 close_midx(m);
1480 free(include_pack);
1481 return result;
1482 }