]> git.ipfire.org Git - thirdparty/git.git/blob - midx.c
Merge branch 'jh/mingw-unlink'
[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
14 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
15 #define MIDX_VERSION 1
16 #define MIDX_BYTE_FILE_VERSION 4
17 #define MIDX_BYTE_HASH_VERSION 5
18 #define MIDX_BYTE_NUM_CHUNKS 6
19 #define MIDX_BYTE_NUM_PACKS 8
20 #define MIDX_HEADER_SIZE 12
21 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
22
23 #define MIDX_MAX_CHUNKS 5
24 #define MIDX_CHUNK_ALIGNMENT 4
25 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
26 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
27 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
28 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
29 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
30 #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
31 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
32 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
33 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
34 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
35
36 #define PACK_EXPIRED UINT_MAX
37
38 static uint8_t oid_version(void)
39 {
40 switch (hash_algo_by_ptr(the_hash_algo)) {
41 case GIT_HASH_SHA1:
42 return 1;
43 case GIT_HASH_SHA256:
44 return 2;
45 default:
46 die(_("invalid hash version"));
47 }
48 }
49
50 static char *get_midx_filename(const char *object_dir)
51 {
52 return xstrfmt("%s/pack/multi-pack-index", object_dir);
53 }
54
55 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
56 {
57 struct multi_pack_index *m = NULL;
58 int fd;
59 struct stat st;
60 size_t midx_size;
61 void *midx_map = NULL;
62 uint32_t hash_version;
63 char *midx_name = get_midx_filename(object_dir);
64 uint32_t i;
65 const char *cur_pack_name;
66
67 fd = git_open(midx_name);
68
69 if (fd < 0)
70 goto cleanup_fail;
71 if (fstat(fd, &st)) {
72 error_errno(_("failed to read %s"), midx_name);
73 goto cleanup_fail;
74 }
75
76 midx_size = xsize_t(st.st_size);
77
78 if (midx_size < MIDX_MIN_SIZE) {
79 error(_("multi-pack-index file %s is too small"), midx_name);
80 goto cleanup_fail;
81 }
82
83 FREE_AND_NULL(midx_name);
84
85 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
86 close(fd);
87
88 FLEX_ALLOC_STR(m, object_dir, object_dir);
89 m->data = midx_map;
90 m->data_len = midx_size;
91 m->local = local;
92
93 m->signature = get_be32(m->data);
94 if (m->signature != MIDX_SIGNATURE)
95 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
96 m->signature, MIDX_SIGNATURE);
97
98 m->version = m->data[MIDX_BYTE_FILE_VERSION];
99 if (m->version != MIDX_VERSION)
100 die(_("multi-pack-index version %d not recognized"),
101 m->version);
102
103 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
104 if (hash_version != oid_version()) {
105 error(_("multi-pack-index hash version %u does not match version %u"),
106 hash_version, oid_version());
107 goto cleanup_fail;
108 }
109 m->hash_len = the_hash_algo->rawsz;
110
111 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
112
113 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
114
115 for (i = 0; i < m->num_chunks; i++) {
116 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
117 MIDX_CHUNKLOOKUP_WIDTH * i);
118 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
119 MIDX_CHUNKLOOKUP_WIDTH * i);
120
121 if (chunk_offset >= m->data_len)
122 die(_("invalid chunk offset (too large)"));
123
124 switch (chunk_id) {
125 case MIDX_CHUNKID_PACKNAMES:
126 m->chunk_pack_names = m->data + chunk_offset;
127 break;
128
129 case MIDX_CHUNKID_OIDFANOUT:
130 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
131 break;
132
133 case MIDX_CHUNKID_OIDLOOKUP:
134 m->chunk_oid_lookup = m->data + chunk_offset;
135 break;
136
137 case MIDX_CHUNKID_OBJECTOFFSETS:
138 m->chunk_object_offsets = m->data + chunk_offset;
139 break;
140
141 case MIDX_CHUNKID_LARGEOFFSETS:
142 m->chunk_large_offsets = m->data + chunk_offset;
143 break;
144
145 case 0:
146 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
147 break;
148
149 default:
150 /*
151 * Do nothing on unrecognized chunks, allowing future
152 * extensions to add optional chunks.
153 */
154 break;
155 }
156 }
157
158 if (!m->chunk_pack_names)
159 die(_("multi-pack-index missing required pack-name chunk"));
160 if (!m->chunk_oid_fanout)
161 die(_("multi-pack-index missing required OID fanout chunk"));
162 if (!m->chunk_oid_lookup)
163 die(_("multi-pack-index missing required OID lookup chunk"));
164 if (!m->chunk_object_offsets)
165 die(_("multi-pack-index missing required object offsets chunk"));
166
167 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
168
169 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
170 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
171
172 cur_pack_name = (const char *)m->chunk_pack_names;
173 for (i = 0; i < m->num_packs; i++) {
174 m->pack_names[i] = cur_pack_name;
175
176 cur_pack_name += strlen(cur_pack_name) + 1;
177
178 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
179 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
180 m->pack_names[i - 1],
181 m->pack_names[i]);
182 }
183
184 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
185 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
186
187 return m;
188
189 cleanup_fail:
190 free(m);
191 free(midx_name);
192 if (midx_map)
193 munmap(midx_map, midx_size);
194 if (0 <= fd)
195 close(fd);
196 return NULL;
197 }
198
199 void close_midx(struct multi_pack_index *m)
200 {
201 uint32_t i;
202
203 if (!m)
204 return;
205
206 munmap((unsigned char *)m->data, m->data_len);
207
208 for (i = 0; i < m->num_packs; i++) {
209 if (m->packs[i])
210 m->packs[i]->multi_pack_index = 0;
211 }
212 FREE_AND_NULL(m->packs);
213 FREE_AND_NULL(m->pack_names);
214 }
215
216 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
217 {
218 struct strbuf pack_name = STRBUF_INIT;
219 struct packed_git *p;
220
221 if (pack_int_id >= m->num_packs)
222 die(_("bad pack-int-id: %u (%u total packs)"),
223 pack_int_id, m->num_packs);
224
225 if (m->packs[pack_int_id])
226 return 0;
227
228 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
229 m->pack_names[pack_int_id]);
230
231 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
232 strbuf_release(&pack_name);
233
234 if (!p)
235 return 1;
236
237 p->multi_pack_index = 1;
238 m->packs[pack_int_id] = p;
239 install_packed_git(r, p);
240 list_add_tail(&p->mru, &r->objects->packed_git_mru);
241
242 return 0;
243 }
244
245 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
246 {
247 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
248 the_hash_algo->rawsz, result);
249 }
250
251 struct object_id *nth_midxed_object_oid(struct object_id *oid,
252 struct multi_pack_index *m,
253 uint32_t n)
254 {
255 if (n >= m->num_objects)
256 return NULL;
257
258 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
259 return oid;
260 }
261
262 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
263 {
264 const unsigned char *offset_data;
265 uint32_t offset32;
266
267 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
268 offset32 = get_be32(offset_data + sizeof(uint32_t));
269
270 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
271 if (sizeof(off_t) < sizeof(uint64_t))
272 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
273
274 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
275 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
276 }
277
278 return offset32;
279 }
280
281 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
282 {
283 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
284 }
285
286 static int nth_midxed_pack_entry(struct repository *r,
287 struct multi_pack_index *m,
288 struct pack_entry *e,
289 uint32_t pos)
290 {
291 uint32_t pack_int_id;
292 struct packed_git *p;
293
294 if (pos >= m->num_objects)
295 return 0;
296
297 pack_int_id = nth_midxed_pack_int_id(m, pos);
298
299 if (prepare_midx_pack(r, m, pack_int_id))
300 die(_("error preparing packfile from multi-pack-index"));
301 p = m->packs[pack_int_id];
302
303 /*
304 * We are about to tell the caller where they can locate the
305 * requested object. We better make sure the packfile is
306 * still here and can be accessed before supplying that
307 * answer, as it may have been deleted since the MIDX was
308 * loaded!
309 */
310 if (!is_pack_valid(p))
311 return 0;
312
313 if (p->num_bad_objects) {
314 uint32_t i;
315 struct object_id oid;
316 nth_midxed_object_oid(&oid, m, pos);
317 for (i = 0; i < p->num_bad_objects; i++)
318 if (hasheq(oid.hash,
319 p->bad_object_sha1 + the_hash_algo->rawsz * i))
320 return 0;
321 }
322
323 e->offset = nth_midxed_offset(m, pos);
324 e->p = p;
325
326 return 1;
327 }
328
329 int fill_midx_entry(struct repository * r,
330 const struct object_id *oid,
331 struct pack_entry *e,
332 struct multi_pack_index *m)
333 {
334 uint32_t pos;
335
336 if (!bsearch_midx(oid, m, &pos))
337 return 0;
338
339 return nth_midxed_pack_entry(r, m, e, pos);
340 }
341
342 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
343 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
344 const char *idx_name)
345 {
346 /* Skip past any initial matching prefix. */
347 while (*idx_name && *idx_name == *idx_or_pack_name) {
348 idx_name++;
349 idx_or_pack_name++;
350 }
351
352 /*
353 * If we didn't match completely, we may have matched "pack-1234." and
354 * be left with "idx" and "pack" respectively, which is also OK. We do
355 * not have to check for "idx" and "idx", because that would have been
356 * a complete match (and in that case these strcmps will be false, but
357 * we'll correctly return 0 from the final strcmp() below.
358 *
359 * Technically this matches "fooidx" and "foopack", but we'd never have
360 * such names in the first place.
361 */
362 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
363 return 0;
364
365 /*
366 * This not only checks for a complete match, but also orders based on
367 * the first non-identical character, which means our ordering will
368 * match a raw strcmp(). That makes it OK to use this to binary search
369 * a naively-sorted list.
370 */
371 return strcmp(idx_or_pack_name, idx_name);
372 }
373
374 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
375 {
376 uint32_t first = 0, last = m->num_packs;
377
378 while (first < last) {
379 uint32_t mid = first + (last - first) / 2;
380 const char *current;
381 int cmp;
382
383 current = m->pack_names[mid];
384 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
385 if (!cmp)
386 return 1;
387 if (cmp > 0) {
388 first = mid + 1;
389 continue;
390 }
391 last = mid;
392 }
393
394 return 0;
395 }
396
397 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
398 {
399 struct multi_pack_index *m;
400 struct multi_pack_index *m_search;
401 int config_value;
402 static int env_value = -1;
403
404 if (env_value < 0)
405 env_value = git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0);
406
407 if (!env_value &&
408 (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
409 !config_value))
410 return 0;
411
412 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
413 if (!strcmp(object_dir, m_search->object_dir))
414 return 1;
415
416 m = load_multi_pack_index(object_dir, local);
417
418 if (m) {
419 m->next = r->objects->multi_pack_index;
420 r->objects->multi_pack_index = m;
421 return 1;
422 }
423
424 return 0;
425 }
426
427 static size_t write_midx_header(struct hashfile *f,
428 unsigned char num_chunks,
429 uint32_t num_packs)
430 {
431 unsigned char byte_values[4];
432
433 hashwrite_be32(f, MIDX_SIGNATURE);
434 byte_values[0] = MIDX_VERSION;
435 byte_values[1] = oid_version();
436 byte_values[2] = num_chunks;
437 byte_values[3] = 0; /* unused */
438 hashwrite(f, byte_values, sizeof(byte_values));
439 hashwrite_be32(f, num_packs);
440
441 return MIDX_HEADER_SIZE;
442 }
443
444 struct pack_info {
445 uint32_t orig_pack_int_id;
446 char *pack_name;
447 struct packed_git *p;
448 unsigned expired : 1;
449 };
450
451 static int pack_info_compare(const void *_a, const void *_b)
452 {
453 struct pack_info *a = (struct pack_info *)_a;
454 struct pack_info *b = (struct pack_info *)_b;
455 return strcmp(a->pack_name, b->pack_name);
456 }
457
458 struct pack_list {
459 struct pack_info *info;
460 uint32_t nr;
461 uint32_t alloc;
462 struct multi_pack_index *m;
463 struct progress *progress;
464 unsigned pack_paths_checked;
465 };
466
467 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
468 const char *file_name, void *data)
469 {
470 struct pack_list *packs = (struct pack_list *)data;
471
472 if (ends_with(file_name, ".idx")) {
473 display_progress(packs->progress, ++packs->pack_paths_checked);
474 if (packs->m && midx_contains_pack(packs->m, file_name))
475 return;
476
477 ALLOC_GROW(packs->info, packs->nr + 1, packs->alloc);
478
479 packs->info[packs->nr].p = add_packed_git(full_path,
480 full_path_len,
481 0);
482
483 if (!packs->info[packs->nr].p) {
484 warning(_("failed to add packfile '%s'"),
485 full_path);
486 return;
487 }
488
489 if (open_pack_index(packs->info[packs->nr].p)) {
490 warning(_("failed to open pack-index '%s'"),
491 full_path);
492 close_pack(packs->info[packs->nr].p);
493 FREE_AND_NULL(packs->info[packs->nr].p);
494 return;
495 }
496
497 packs->info[packs->nr].pack_name = xstrdup(file_name);
498 packs->info[packs->nr].orig_pack_int_id = packs->nr;
499 packs->info[packs->nr].expired = 0;
500 packs->nr++;
501 }
502 }
503
504 struct pack_midx_entry {
505 struct object_id oid;
506 uint32_t pack_int_id;
507 time_t pack_mtime;
508 uint64_t offset;
509 };
510
511 static int midx_oid_compare(const void *_a, const void *_b)
512 {
513 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
514 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
515 int cmp = oidcmp(&a->oid, &b->oid);
516
517 if (cmp)
518 return cmp;
519
520 if (a->pack_mtime > b->pack_mtime)
521 return -1;
522 else if (a->pack_mtime < b->pack_mtime)
523 return 1;
524
525 return a->pack_int_id - b->pack_int_id;
526 }
527
528 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
529 struct pack_midx_entry *e,
530 uint32_t pos)
531 {
532 if (pos >= m->num_objects)
533 return 1;
534
535 nth_midxed_object_oid(&e->oid, m, pos);
536 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
537 e->offset = nth_midxed_offset(m, pos);
538
539 /* consider objects in midx to be from "old" packs */
540 e->pack_mtime = 0;
541 return 0;
542 }
543
544 static void fill_pack_entry(uint32_t pack_int_id,
545 struct packed_git *p,
546 uint32_t cur_object,
547 struct pack_midx_entry *entry)
548 {
549 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
550 die(_("failed to locate object %d in packfile"), cur_object);
551
552 entry->pack_int_id = pack_int_id;
553 entry->pack_mtime = p->mtime;
554
555 entry->offset = nth_packed_object_offset(p, cur_object);
556 }
557
558 /*
559 * It is possible to artificially get into a state where there are many
560 * duplicate copies of objects. That can create high memory pressure if
561 * we are to create a list of all objects before de-duplication. To reduce
562 * this memory pressure without a significant performance drop, automatically
563 * group objects by the first byte of their object id. Use the IDX fanout
564 * tables to group the data, copy to a local array, then sort.
565 *
566 * Copy only the de-duplicated entries (selected by most-recent modified time
567 * of a packfile containing the object).
568 */
569 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
570 struct pack_info *info,
571 uint32_t nr_packs,
572 uint32_t *nr_objects)
573 {
574 uint32_t cur_fanout, cur_pack, cur_object;
575 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
576 struct pack_midx_entry *entries_by_fanout = NULL;
577 struct pack_midx_entry *deduplicated_entries = NULL;
578 uint32_t start_pack = m ? m->num_packs : 0;
579
580 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
581 total_objects += info[cur_pack].p->num_objects;
582
583 /*
584 * As we de-duplicate by fanout value, we expect the fanout
585 * slices to be evenly distributed, with some noise. Hence,
586 * allocate slightly more than one 256th.
587 */
588 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
589
590 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
591 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
592 *nr_objects = 0;
593
594 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
595 uint32_t nr_fanout = 0;
596
597 if (m) {
598 uint32_t start = 0, end;
599
600 if (cur_fanout)
601 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
602 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
603
604 for (cur_object = start; cur_object < end; cur_object++) {
605 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
606 nth_midxed_pack_midx_entry(m,
607 &entries_by_fanout[nr_fanout],
608 cur_object);
609 nr_fanout++;
610 }
611 }
612
613 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
614 uint32_t start = 0, end;
615
616 if (cur_fanout)
617 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
618 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
619
620 for (cur_object = start; cur_object < end; cur_object++) {
621 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
622 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
623 nr_fanout++;
624 }
625 }
626
627 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
628
629 /*
630 * The batch is now sorted by OID and then mtime (descending).
631 * Take only the first duplicate.
632 */
633 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
634 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
635 &entries_by_fanout[cur_object].oid))
636 continue;
637
638 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
639 memcpy(&deduplicated_entries[*nr_objects],
640 &entries_by_fanout[cur_object],
641 sizeof(struct pack_midx_entry));
642 (*nr_objects)++;
643 }
644 }
645
646 free(entries_by_fanout);
647 return deduplicated_entries;
648 }
649
650 static size_t write_midx_pack_names(struct hashfile *f,
651 struct pack_info *info,
652 uint32_t num_packs)
653 {
654 uint32_t i;
655 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
656 size_t written = 0;
657
658 for (i = 0; i < num_packs; i++) {
659 size_t writelen;
660
661 if (info[i].expired)
662 continue;
663
664 if (i && strcmp(info[i].pack_name, info[i - 1].pack_name) <= 0)
665 BUG("incorrect pack-file order: %s before %s",
666 info[i - 1].pack_name,
667 info[i].pack_name);
668
669 writelen = strlen(info[i].pack_name) + 1;
670 hashwrite(f, info[i].pack_name, writelen);
671 written += writelen;
672 }
673
674 /* add padding to be aligned */
675 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
676 if (i < MIDX_CHUNK_ALIGNMENT) {
677 memset(padding, 0, sizeof(padding));
678 hashwrite(f, padding, i);
679 written += i;
680 }
681
682 return written;
683 }
684
685 static size_t write_midx_oid_fanout(struct hashfile *f,
686 struct pack_midx_entry *objects,
687 uint32_t nr_objects)
688 {
689 struct pack_midx_entry *list = objects;
690 struct pack_midx_entry *last = objects + nr_objects;
691 uint32_t count = 0;
692 uint32_t i;
693
694 /*
695 * Write the first-level table (the list is sorted,
696 * but we use a 256-entry lookup to be able to avoid
697 * having to do eight extra binary search iterations).
698 */
699 for (i = 0; i < 256; i++) {
700 struct pack_midx_entry *next = list;
701
702 while (next < last && next->oid.hash[0] == i) {
703 count++;
704 next++;
705 }
706
707 hashwrite_be32(f, count);
708 list = next;
709 }
710
711 return MIDX_CHUNK_FANOUT_SIZE;
712 }
713
714 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
715 struct pack_midx_entry *objects,
716 uint32_t nr_objects)
717 {
718 struct pack_midx_entry *list = objects;
719 uint32_t i;
720 size_t written = 0;
721
722 for (i = 0; i < nr_objects; i++) {
723 struct pack_midx_entry *obj = list++;
724
725 if (i < nr_objects - 1) {
726 struct pack_midx_entry *next = list;
727 if (oidcmp(&obj->oid, &next->oid) >= 0)
728 BUG("OIDs not in order: %s >= %s",
729 oid_to_hex(&obj->oid),
730 oid_to_hex(&next->oid));
731 }
732
733 hashwrite(f, obj->oid.hash, (int)hash_len);
734 written += hash_len;
735 }
736
737 return written;
738 }
739
740 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
741 uint32_t *perm,
742 struct pack_midx_entry *objects, uint32_t nr_objects)
743 {
744 struct pack_midx_entry *list = objects;
745 uint32_t i, nr_large_offset = 0;
746 size_t written = 0;
747
748 for (i = 0; i < nr_objects; i++) {
749 struct pack_midx_entry *obj = list++;
750
751 if (perm[obj->pack_int_id] == PACK_EXPIRED)
752 BUG("object %s is in an expired pack with int-id %d",
753 oid_to_hex(&obj->oid),
754 obj->pack_int_id);
755
756 hashwrite_be32(f, perm[obj->pack_int_id]);
757
758 if (large_offset_needed && obj->offset >> 31)
759 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
760 else if (!large_offset_needed && obj->offset >> 32)
761 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
762 oid_to_hex(&obj->oid),
763 obj->offset);
764 else
765 hashwrite_be32(f, (uint32_t)obj->offset);
766
767 written += MIDX_CHUNK_OFFSET_WIDTH;
768 }
769
770 return written;
771 }
772
773 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
774 struct pack_midx_entry *objects, uint32_t nr_objects)
775 {
776 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
777 size_t written = 0;
778
779 while (nr_large_offset) {
780 struct pack_midx_entry *obj;
781 uint64_t offset;
782
783 if (list >= end)
784 BUG("too many large-offset objects");
785
786 obj = list++;
787 offset = obj->offset;
788
789 if (!(offset >> 31))
790 continue;
791
792 hashwrite_be32(f, offset >> 32);
793 hashwrite_be32(f, offset & 0xffffffffUL);
794 written += 2 * sizeof(uint32_t);
795
796 nr_large_offset--;
797 }
798
799 return written;
800 }
801
802 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
803 struct string_list *packs_to_drop, unsigned flags)
804 {
805 unsigned char cur_chunk, num_chunks = 0;
806 char *midx_name;
807 uint32_t i;
808 struct hashfile *f = NULL;
809 struct lock_file lk;
810 struct pack_list packs;
811 uint32_t *pack_perm = NULL;
812 uint64_t written = 0;
813 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
814 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
815 uint32_t nr_entries, num_large_offsets = 0;
816 struct pack_midx_entry *entries = NULL;
817 struct progress *progress = NULL;
818 int large_offsets_needed = 0;
819 int pack_name_concat_len = 0;
820 int dropped_packs = 0;
821 int result = 0;
822
823 midx_name = get_midx_filename(object_dir);
824 if (safe_create_leading_directories(midx_name)) {
825 UNLEAK(midx_name);
826 die_errno(_("unable to create leading directories of %s"),
827 midx_name);
828 }
829
830 if (m)
831 packs.m = m;
832 else
833 packs.m = load_multi_pack_index(object_dir, 1);
834
835 packs.nr = 0;
836 packs.alloc = packs.m ? packs.m->num_packs : 16;
837 packs.info = NULL;
838 ALLOC_ARRAY(packs.info, packs.alloc);
839
840 if (packs.m) {
841 for (i = 0; i < packs.m->num_packs; i++) {
842 ALLOC_GROW(packs.info, packs.nr + 1, packs.alloc);
843
844 packs.info[packs.nr].orig_pack_int_id = i;
845 packs.info[packs.nr].pack_name = xstrdup(packs.m->pack_names[i]);
846 packs.info[packs.nr].p = NULL;
847 packs.info[packs.nr].expired = 0;
848 packs.nr++;
849 }
850 }
851
852 packs.pack_paths_checked = 0;
853 if (flags & MIDX_PROGRESS)
854 packs.progress = start_progress(_("Adding packfiles to multi-pack-index"), 0);
855 else
856 packs.progress = NULL;
857
858 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
859 stop_progress(&packs.progress);
860
861 if (packs.m && packs.nr == packs.m->num_packs && !packs_to_drop)
862 goto cleanup;
863
864 entries = get_sorted_entries(packs.m, packs.info, packs.nr, &nr_entries);
865
866 for (i = 0; i < nr_entries; i++) {
867 if (entries[i].offset > 0x7fffffff)
868 num_large_offsets++;
869 if (entries[i].offset > 0xffffffff)
870 large_offsets_needed = 1;
871 }
872
873 QSORT(packs.info, packs.nr, pack_info_compare);
874
875 if (packs_to_drop && packs_to_drop->nr) {
876 int drop_index = 0;
877 int missing_drops = 0;
878
879 for (i = 0; i < packs.nr && drop_index < packs_to_drop->nr; i++) {
880 int cmp = strcmp(packs.info[i].pack_name,
881 packs_to_drop->items[drop_index].string);
882
883 if (!cmp) {
884 drop_index++;
885 packs.info[i].expired = 1;
886 } else if (cmp > 0) {
887 error(_("did not see pack-file %s to drop"),
888 packs_to_drop->items[drop_index].string);
889 drop_index++;
890 missing_drops++;
891 i--;
892 } else {
893 packs.info[i].expired = 0;
894 }
895 }
896
897 if (missing_drops) {
898 result = 1;
899 goto cleanup;
900 }
901 }
902
903 /*
904 * pack_perm stores a permutation between pack-int-ids from the
905 * previous multi-pack-index to the new one we are writing:
906 *
907 * pack_perm[old_id] = new_id
908 */
909 ALLOC_ARRAY(pack_perm, packs.nr);
910 for (i = 0; i < packs.nr; i++) {
911 if (packs.info[i].expired) {
912 dropped_packs++;
913 pack_perm[packs.info[i].orig_pack_int_id] = PACK_EXPIRED;
914 } else {
915 pack_perm[packs.info[i].orig_pack_int_id] = i - dropped_packs;
916 }
917 }
918
919 for (i = 0; i < packs.nr; i++) {
920 if (!packs.info[i].expired)
921 pack_name_concat_len += strlen(packs.info[i].pack_name) + 1;
922 }
923
924 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
925 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
926 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
927
928 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
929 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
930 FREE_AND_NULL(midx_name);
931
932 if (packs.m)
933 close_midx(packs.m);
934
935 cur_chunk = 0;
936 num_chunks = large_offsets_needed ? 5 : 4;
937
938 if (packs.nr - dropped_packs == 0) {
939 error(_("no pack files to index."));
940 result = 1;
941 goto cleanup;
942 }
943
944 written = write_midx_header(f, num_chunks, packs.nr - dropped_packs);
945
946 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
947 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
948
949 cur_chunk++;
950 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
951 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + pack_name_concat_len;
952
953 cur_chunk++;
954 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
955 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
956
957 cur_chunk++;
958 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
959 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * the_hash_algo->rawsz;
960
961 cur_chunk++;
962 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
963 if (large_offsets_needed) {
964 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
965
966 cur_chunk++;
967 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
968 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
969 }
970
971 chunk_ids[cur_chunk] = 0;
972
973 for (i = 0; i <= num_chunks; i++) {
974 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
975 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
976 chunk_offsets[i - 1],
977 chunk_offsets[i]);
978
979 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
980 BUG("chunk offset %"PRIu64" is not properly aligned",
981 chunk_offsets[i]);
982
983 hashwrite_be32(f, chunk_ids[i]);
984 hashwrite_be32(f, chunk_offsets[i] >> 32);
985 hashwrite_be32(f, chunk_offsets[i]);
986
987 written += MIDX_CHUNKLOOKUP_WIDTH;
988 }
989
990 if (flags & MIDX_PROGRESS)
991 progress = start_progress(_("Writing chunks to multi-pack-index"),
992 num_chunks);
993 for (i = 0; i < num_chunks; i++) {
994 if (written != chunk_offsets[i])
995 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
996 chunk_offsets[i],
997 written,
998 chunk_ids[i]);
999
1000 switch (chunk_ids[i]) {
1001 case MIDX_CHUNKID_PACKNAMES:
1002 written += write_midx_pack_names(f, packs.info, packs.nr);
1003 break;
1004
1005 case MIDX_CHUNKID_OIDFANOUT:
1006 written += write_midx_oid_fanout(f, entries, nr_entries);
1007 break;
1008
1009 case MIDX_CHUNKID_OIDLOOKUP:
1010 written += write_midx_oid_lookup(f, the_hash_algo->rawsz, entries, nr_entries);
1011 break;
1012
1013 case MIDX_CHUNKID_OBJECTOFFSETS:
1014 written += write_midx_object_offsets(f, large_offsets_needed, pack_perm, entries, nr_entries);
1015 break;
1016
1017 case MIDX_CHUNKID_LARGEOFFSETS:
1018 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
1019 break;
1020
1021 default:
1022 BUG("trying to write unknown chunk id %"PRIx32,
1023 chunk_ids[i]);
1024 }
1025
1026 display_progress(progress, i + 1);
1027 }
1028 stop_progress(&progress);
1029
1030 if (written != chunk_offsets[num_chunks])
1031 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
1032 written,
1033 chunk_offsets[num_chunks]);
1034
1035 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1036 commit_lock_file(&lk);
1037
1038 cleanup:
1039 for (i = 0; i < packs.nr; i++) {
1040 if (packs.info[i].p) {
1041 close_pack(packs.info[i].p);
1042 free(packs.info[i].p);
1043 }
1044 free(packs.info[i].pack_name);
1045 }
1046
1047 free(packs.info);
1048 free(entries);
1049 free(pack_perm);
1050 free(midx_name);
1051 return result;
1052 }
1053
1054 int write_midx_file(const char *object_dir, unsigned flags)
1055 {
1056 return write_midx_internal(object_dir, NULL, NULL, flags);
1057 }
1058
1059 void clear_midx_file(struct repository *r)
1060 {
1061 char *midx = get_midx_filename(r->objects->odb->path);
1062
1063 if (r->objects && r->objects->multi_pack_index) {
1064 close_midx(r->objects->multi_pack_index);
1065 r->objects->multi_pack_index = NULL;
1066 }
1067
1068 if (remove_path(midx)) {
1069 UNLEAK(midx);
1070 die(_("failed to clear multi-pack-index at %s"), midx);
1071 }
1072
1073 free(midx);
1074 }
1075
1076 static int verify_midx_error;
1077
1078 static void midx_report(const char *fmt, ...)
1079 {
1080 va_list ap;
1081 verify_midx_error = 1;
1082 va_start(ap, fmt);
1083 vfprintf(stderr, fmt, ap);
1084 fprintf(stderr, "\n");
1085 va_end(ap);
1086 }
1087
1088 struct pair_pos_vs_id
1089 {
1090 uint32_t pos;
1091 uint32_t pack_int_id;
1092 };
1093
1094 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1095 {
1096 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1097 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1098
1099 return b->pack_int_id - a->pack_int_id;
1100 }
1101
1102 /*
1103 * Limit calls to display_progress() for performance reasons.
1104 * The interval here was arbitrarily chosen.
1105 */
1106 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1107 #define midx_display_sparse_progress(progress, n) \
1108 do { \
1109 uint64_t _n = (n); \
1110 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1111 display_progress(progress, _n); \
1112 } while (0)
1113
1114 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1115 {
1116 struct pair_pos_vs_id *pairs = NULL;
1117 uint32_t i;
1118 struct progress *progress = NULL;
1119 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1120 verify_midx_error = 0;
1121
1122 if (!m) {
1123 int result = 0;
1124 struct stat sb;
1125 char *filename = get_midx_filename(object_dir);
1126 if (!stat(filename, &sb)) {
1127 error(_("multi-pack-index file exists, but failed to parse"));
1128 result = 1;
1129 }
1130 free(filename);
1131 return result;
1132 }
1133
1134 if (flags & MIDX_PROGRESS)
1135 progress = start_progress(_("Looking for referenced packfiles"),
1136 m->num_packs);
1137 for (i = 0; i < m->num_packs; i++) {
1138 if (prepare_midx_pack(r, m, i))
1139 midx_report("failed to load pack in position %d", i);
1140
1141 display_progress(progress, i + 1);
1142 }
1143 stop_progress(&progress);
1144
1145 for (i = 0; i < 255; i++) {
1146 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1147 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1148
1149 if (oid_fanout1 > oid_fanout2)
1150 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1151 i, oid_fanout1, oid_fanout2, i + 1);
1152 }
1153
1154 if (m->num_objects == 0) {
1155 midx_report(_("the midx contains no oid"));
1156 /*
1157 * Remaining tests assume that we have objects, so we can
1158 * return here.
1159 */
1160 return verify_midx_error;
1161 }
1162
1163 if (flags & MIDX_PROGRESS)
1164 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1165 m->num_objects - 1);
1166 for (i = 0; i < m->num_objects - 1; i++) {
1167 struct object_id oid1, oid2;
1168
1169 nth_midxed_object_oid(&oid1, m, i);
1170 nth_midxed_object_oid(&oid2, m, i + 1);
1171
1172 if (oidcmp(&oid1, &oid2) >= 0)
1173 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1174 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1175
1176 midx_display_sparse_progress(progress, i + 1);
1177 }
1178 stop_progress(&progress);
1179
1180 /*
1181 * Create an array mapping each object to its packfile id. Sort it
1182 * to group the objects by packfile. Use this permutation to visit
1183 * each of the objects and only require 1 packfile to be open at a
1184 * time.
1185 */
1186 ALLOC_ARRAY(pairs, m->num_objects);
1187 for (i = 0; i < m->num_objects; i++) {
1188 pairs[i].pos = i;
1189 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1190 }
1191
1192 if (flags & MIDX_PROGRESS)
1193 progress = start_sparse_progress(_("Sorting objects by packfile"),
1194 m->num_objects);
1195 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1196 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1197 stop_progress(&progress);
1198
1199 if (flags & MIDX_PROGRESS)
1200 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1201 for (i = 0; i < m->num_objects; i++) {
1202 struct object_id oid;
1203 struct pack_entry e;
1204 off_t m_offset, p_offset;
1205
1206 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1207 m->packs[pairs[i-1].pack_int_id])
1208 {
1209 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1210 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1211 }
1212
1213 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1214
1215 if (!fill_midx_entry(r, &oid, &e, m)) {
1216 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1217 pairs[i].pos, oid_to_hex(&oid));
1218 continue;
1219 }
1220
1221 if (open_pack_index(e.p)) {
1222 midx_report(_("failed to load pack-index for packfile %s"),
1223 e.p->pack_name);
1224 break;
1225 }
1226
1227 m_offset = e.offset;
1228 p_offset = find_pack_entry_one(oid.hash, e.p);
1229
1230 if (m_offset != p_offset)
1231 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1232 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1233
1234 midx_display_sparse_progress(progress, i + 1);
1235 }
1236 stop_progress(&progress);
1237
1238 free(pairs);
1239
1240 return verify_midx_error;
1241 }
1242
1243 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1244 {
1245 uint32_t i, *count, result = 0;
1246 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1247 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1248 struct progress *progress = NULL;
1249
1250 if (!m)
1251 return 0;
1252
1253 count = xcalloc(m->num_packs, sizeof(uint32_t));
1254
1255 if (flags & MIDX_PROGRESS)
1256 progress = start_progress(_("Counting referenced objects"),
1257 m->num_objects);
1258 for (i = 0; i < m->num_objects; i++) {
1259 int pack_int_id = nth_midxed_pack_int_id(m, i);
1260 count[pack_int_id]++;
1261 display_progress(progress, i + 1);
1262 }
1263 stop_progress(&progress);
1264
1265 if (flags & MIDX_PROGRESS)
1266 progress = start_progress(_("Finding and deleting unreferenced packfiles"),
1267 m->num_packs);
1268 for (i = 0; i < m->num_packs; i++) {
1269 char *pack_name;
1270 display_progress(progress, i + 1);
1271
1272 if (count[i])
1273 continue;
1274
1275 if (prepare_midx_pack(r, m, i))
1276 continue;
1277
1278 if (m->packs[i]->pack_keep)
1279 continue;
1280
1281 pack_name = xstrdup(m->packs[i]->pack_name);
1282 close_pack(m->packs[i]);
1283
1284 string_list_insert(&packs_to_drop, m->pack_names[i]);
1285 unlink_pack_path(pack_name, 0);
1286 free(pack_name);
1287 }
1288 stop_progress(&progress);
1289
1290 free(count);
1291
1292 if (packs_to_drop.nr)
1293 result = write_midx_internal(object_dir, m, &packs_to_drop, flags);
1294
1295 string_list_clear(&packs_to_drop, 0);
1296 return result;
1297 }
1298
1299 struct repack_info {
1300 timestamp_t mtime;
1301 uint32_t referenced_objects;
1302 uint32_t pack_int_id;
1303 };
1304
1305 static int compare_by_mtime(const void *a_, const void *b_)
1306 {
1307 const struct repack_info *a, *b;
1308
1309 a = (const struct repack_info *)a_;
1310 b = (const struct repack_info *)b_;
1311
1312 if (a->mtime < b->mtime)
1313 return -1;
1314 if (a->mtime > b->mtime)
1315 return 1;
1316 return 0;
1317 }
1318
1319 static int fill_included_packs_all(struct repository *r,
1320 struct multi_pack_index *m,
1321 unsigned char *include_pack)
1322 {
1323 uint32_t i, count = 0;
1324 int pack_kept_objects = 0;
1325
1326 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1327
1328 for (i = 0; i < m->num_packs; i++) {
1329 if (prepare_midx_pack(r, m, i))
1330 continue;
1331 if (!pack_kept_objects && m->packs[i]->pack_keep)
1332 continue;
1333
1334 include_pack[i] = 1;
1335 count++;
1336 }
1337
1338 return count < 2;
1339 }
1340
1341 static int fill_included_packs_batch(struct repository *r,
1342 struct multi_pack_index *m,
1343 unsigned char *include_pack,
1344 size_t batch_size)
1345 {
1346 uint32_t i, packs_to_repack;
1347 size_t total_size;
1348 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1349 int pack_kept_objects = 0;
1350
1351 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1352
1353 for (i = 0; i < m->num_packs; i++) {
1354 pack_info[i].pack_int_id = i;
1355
1356 if (prepare_midx_pack(r, m, i))
1357 continue;
1358
1359 pack_info[i].mtime = m->packs[i]->mtime;
1360 }
1361
1362 for (i = 0; batch_size && i < m->num_objects; i++) {
1363 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1364 pack_info[pack_int_id].referenced_objects++;
1365 }
1366
1367 QSORT(pack_info, m->num_packs, compare_by_mtime);
1368
1369 total_size = 0;
1370 packs_to_repack = 0;
1371 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1372 int pack_int_id = pack_info[i].pack_int_id;
1373 struct packed_git *p = m->packs[pack_int_id];
1374 size_t expected_size;
1375
1376 if (!p)
1377 continue;
1378 if (!pack_kept_objects && p->pack_keep)
1379 continue;
1380 if (open_pack_index(p) || !p->num_objects)
1381 continue;
1382
1383 expected_size = (size_t)(p->pack_size
1384 * pack_info[i].referenced_objects);
1385 expected_size /= p->num_objects;
1386
1387 if (expected_size >= batch_size)
1388 continue;
1389
1390 packs_to_repack++;
1391 total_size += expected_size;
1392 include_pack[pack_int_id] = 1;
1393 }
1394
1395 free(pack_info);
1396
1397 if (total_size < batch_size || packs_to_repack < 2)
1398 return 1;
1399
1400 return 0;
1401 }
1402
1403 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1404 {
1405 int result = 0;
1406 uint32_t i;
1407 unsigned char *include_pack;
1408 struct child_process cmd = CHILD_PROCESS_INIT;
1409 struct strbuf base_name = STRBUF_INIT;
1410 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1411
1412 /*
1413 * When updating the default for these configuration
1414 * variables in builtin/repack.c, these must be adjusted
1415 * to match.
1416 */
1417 int delta_base_offset = 1;
1418 int use_delta_islands = 0;
1419
1420 if (!m)
1421 return 0;
1422
1423 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1424
1425 if (batch_size) {
1426 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1427 goto cleanup;
1428 } else if (fill_included_packs_all(r, m, include_pack))
1429 goto cleanup;
1430
1431 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1432 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1433
1434 strvec_push(&cmd.args, "pack-objects");
1435
1436 strbuf_addstr(&base_name, object_dir);
1437 strbuf_addstr(&base_name, "/pack/pack");
1438 strvec_push(&cmd.args, base_name.buf);
1439
1440 if (delta_base_offset)
1441 strvec_push(&cmd.args, "--delta-base-offset");
1442 if (use_delta_islands)
1443 strvec_push(&cmd.args, "--delta-islands");
1444
1445 if (flags & MIDX_PROGRESS)
1446 strvec_push(&cmd.args, "--progress");
1447 else
1448 strvec_push(&cmd.args, "-q");
1449
1450 strbuf_release(&base_name);
1451
1452 cmd.git_cmd = 1;
1453 cmd.in = cmd.out = -1;
1454
1455 if (start_command(&cmd)) {
1456 error(_("could not start pack-objects"));
1457 result = 1;
1458 goto cleanup;
1459 }
1460
1461 for (i = 0; i < m->num_objects; i++) {
1462 struct object_id oid;
1463 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1464
1465 if (!include_pack[pack_int_id])
1466 continue;
1467
1468 nth_midxed_object_oid(&oid, m, i);
1469 xwrite(cmd.in, oid_to_hex(&oid), the_hash_algo->hexsz);
1470 xwrite(cmd.in, "\n", 1);
1471 }
1472 close(cmd.in);
1473
1474 if (finish_command(&cmd)) {
1475 error(_("could not finish pack-objects"));
1476 result = 1;
1477 goto cleanup;
1478 }
1479
1480 result = write_midx_internal(object_dir, m, NULL, flags);
1481 m = NULL;
1482
1483 cleanup:
1484 if (m)
1485 close_midx(m);
1486 free(include_pack);
1487 return result;
1488 }