?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
pack-objects: pass `bitmapped_pack`'s to pack-reuse functions
[alt-git.git] / midx.c
blobbeaf0c0de4472c8c2fcf20902aa0aaf1ed3e4dab
1 #include "git-compat-util.h"
2 #include "abspath.h"
3 #include "config.h"
4 #include "csum-file.h"
5 #include "dir.h"
6 #include "gettext.h"
7 #include "hex.h"
8 #include "lockfile.h"
9 #include "packfile.h"
10 #include "object-file.h"
11 #include "object-store-ll.h"
12 #include "hash-lookup.h"
13 #include "midx.h"
14 #include "progress.h"
15 #include "trace2.h"
16 #include "run-command.h"
17 #include "repository.h"
18 #include "chunk-format.h"
19 #include "pack.h"
20 #include "pack-bitmap.h"
21 #include "refs.h"
22 #include "revision.h"
23 #include "list-objects.h"
24
25 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
26 #define MIDX_VERSION 1
27 #define MIDX_BYTE_FILE_VERSION 4
28 #define MIDX_BYTE_HASH_VERSION 5
29 #define MIDX_BYTE_NUM_CHUNKS 6
30 #define MIDX_BYTE_NUM_PACKS 8
31 #define MIDX_HEADER_SIZE 12
32 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
33
34 #define MIDX_CHUNK_ALIGNMENT 4
35 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
36 #define MIDX_CHUNKID_BITMAPPEDPACKS 0x42544d50 /* "BTMP" */
37 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
38 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
39 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
40 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
41 #define MIDX_CHUNKID_REVINDEX 0x52494458 /* "RIDX" */
42 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
43 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
44 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
45 #define MIDX_CHUNK_BITMAPPED_PACKS_WIDTH (2 * sizeof(uint32_t))
46 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
47
48 #define PACK_EXPIRED UINT_MAX
49
50 const unsigned char *get_midx_checksum(struct multi_pack_index *m)
51 {
52 return m->data + m->data_len - the_hash_algo->rawsz;
53 }
54
55 void get_midx_filename(struct strbuf *out, const char *object_dir)
56 {
57 strbuf_addf(out, "%s/pack/multi-pack-index", object_dir);
58 }
59
60 void get_midx_rev_filename(struct strbuf *out, struct multi_pack_index *m)
61 {
62 get_midx_filename(out, m->object_dir);
63 strbuf_addf(out, "-%s.rev", hash_to_hex(get_midx_checksum(m)));
64 }
65
66 static int midx_read_oid_fanout(const unsigned char *chunk_start,
67 size_t chunk_size, void *data)
68 {
69 int i;
70 struct multi_pack_index *m = data;
71 m->chunk_oid_fanout = (uint32_t *)chunk_start;
72
73 if (chunk_size != 4 * 256) {
74 error(_("multi-pack-index OID fanout is of the wrong size"));
75 return 1;
76 }
77 for (i = 0; i < 255; i++) {
78 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
79 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i+1]);
80
81 if (oid_fanout1 > oid_fanout2) {
82 error(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
83 i, oid_fanout1, oid_fanout2, i + 1);
84 return 1;
85 }
86 }
87 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
88 return 0;
89 }
90
91 static int midx_read_oid_lookup(const unsigned char *chunk_start,
92 size_t chunk_size, void *data)
93 {
94 struct multi_pack_index *m = data;
95 m->chunk_oid_lookup = chunk_start;
96
97 if (chunk_size != st_mult(m->hash_len, m->num_objects)) {
98 error(_("multi-pack-index OID lookup chunk is the wrong size"));
99 return 1;
100 }
101 return 0;
102 }
103
104 static int midx_read_object_offsets(const unsigned char *chunk_start,
105 size_t chunk_size, void *data)
106 {
107 struct multi_pack_index *m = data;
108 m->chunk_object_offsets = chunk_start;
109
110 if (chunk_size != st_mult(m->num_objects, MIDX_CHUNK_OFFSET_WIDTH)) {
111 error(_("multi-pack-index object offset chunk is the wrong size"));
112 return 1;
113 }
114 return 0;
115 }
116
117 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
118 {
119 struct multi_pack_index *m = NULL;
120 int fd;
121 struct stat st;
122 size_t midx_size;
123 void *midx_map = NULL;
124 uint32_t hash_version;
125 struct strbuf midx_name = STRBUF_INIT;
126 uint32_t i;
127 const char *cur_pack_name;
128 struct chunkfile *cf = NULL;
129
130 get_midx_filename(&midx_name, object_dir);
131
132 fd = git_open(midx_name.buf);
133
134 if (fd < 0)
135 goto cleanup_fail;
136 if (fstat(fd, &st)) {
137 error_errno(_("failed to read %s"), midx_name.buf);
138 goto cleanup_fail;
139 }
140
141 midx_size = xsize_t(st.st_size);
142
143 if (midx_size < MIDX_MIN_SIZE) {
144 error(_("multi-pack-index file %s is too small"), midx_name.buf);
145 goto cleanup_fail;
146 }
147
148 strbuf_release(&midx_name);
149
150 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
151 close(fd);
152
153 FLEX_ALLOC_STR(m, object_dir, object_dir);
154 m->data = midx_map;
155 m->data_len = midx_size;
156 m->local = local;
157
158 m->signature = get_be32(m->data);
159 if (m->signature != MIDX_SIGNATURE)
160 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
161 m->signature, MIDX_SIGNATURE);
162
163 m->version = m->data[MIDX_BYTE_FILE_VERSION];
164 if (m->version != MIDX_VERSION)
165 die(_("multi-pack-index version %d not recognized"),
166 m->version);
167
168 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
169 if (hash_version != oid_version(the_hash_algo)) {
170 error(_("multi-pack-index hash version %u does not match version %u"),
171 hash_version, oid_version(the_hash_algo));
172 goto cleanup_fail;
173 }
174 m->hash_len = the_hash_algo->rawsz;
175
176 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
177
178 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
179
180 cf = init_chunkfile(NULL);
181
182 if (read_table_of_contents(cf, m->data, midx_size,
183 MIDX_HEADER_SIZE, m->num_chunks,
184 MIDX_CHUNK_ALIGNMENT))
185 goto cleanup_fail;
186
187 if (pair_chunk(cf, MIDX_CHUNKID_PACKNAMES, &m->chunk_pack_names, &m->chunk_pack_names_len))
188 die(_("multi-pack-index required pack-name chunk missing or corrupted"));
189 if (read_chunk(cf, MIDX_CHUNKID_OIDFANOUT, midx_read_oid_fanout, m))
190 die(_("multi-pack-index required OID fanout chunk missing or corrupted"));
191 if (read_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, midx_read_oid_lookup, m))
192 die(_("multi-pack-index required OID lookup chunk missing or corrupted"));
193 if (read_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, midx_read_object_offsets, m))
194 die(_("multi-pack-index required object offsets chunk missing or corrupted"));
195
196 pair_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, &m->chunk_large_offsets,
197 &m->chunk_large_offsets_len);
198 pair_chunk(cf, MIDX_CHUNKID_BITMAPPEDPACKS,
199 (const unsigned char **)&m->chunk_bitmapped_packs,
200 &m->chunk_bitmapped_packs_len);
201
202 if (git_env_bool("GIT_TEST_MIDX_READ_RIDX", 1))
203 pair_chunk(cf, MIDX_CHUNKID_REVINDEX, &m->chunk_revindex,
204 &m->chunk_revindex_len);
205
206 CALLOC_ARRAY(m->pack_names, m->num_packs);
207 CALLOC_ARRAY(m->packs, m->num_packs);
208
209 cur_pack_name = (const char *)m->chunk_pack_names;
210 for (i = 0; i < m->num_packs; i++) {
211 const char *end;
212 size_t avail = m->chunk_pack_names_len -
213 (cur_pack_name - (const char *)m->chunk_pack_names);
214
215 m->pack_names[i] = cur_pack_name;
216
217 end = memchr(cur_pack_name, '\0', avail);
218 if (!end)
219 die(_("multi-pack-index pack-name chunk is too short"));
220 cur_pack_name = end + 1;
221
222 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
223 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
224 m->pack_names[i - 1],
225 m->pack_names[i]);
226 }
227
228 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
229 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
230
231 free_chunkfile(cf);
232 return m;
233
234 cleanup_fail:
235 free(m);
236 strbuf_release(&midx_name);
237 free_chunkfile(cf);
238 if (midx_map)
239 munmap(midx_map, midx_size);
240 if (0 <= fd)
241 close(fd);
242 return NULL;
243 }
244
245 void close_midx(struct multi_pack_index *m)
246 {
247 uint32_t i;
248
249 if (!m)
250 return;
251
252 close_midx(m->next);
253
254 munmap((unsigned char *)m->data, m->data_len);
255
256 for (i = 0; i < m->num_packs; i++) {
257 if (m->packs[i])
258 m->packs[i]->multi_pack_index = 0;
259 }
260 FREE_AND_NULL(m->packs);
261 FREE_AND_NULL(m->pack_names);
262 free(m);
263 }
264
265 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
266 {
267 struct strbuf pack_name = STRBUF_INIT;
268 struct packed_git *p;
269
270 if (pack_int_id >= m->num_packs)
271 die(_("bad pack-int-id: %u (%u total packs)"),
272 pack_int_id, m->num_packs);
273
274 if (m->packs[pack_int_id])
275 return 0;
276
277 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
278 m->pack_names[pack_int_id]);
279
280 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
281 strbuf_release(&pack_name);
282
283 if (!p)
284 return 1;
285
286 p->multi_pack_index = 1;
287 m->packs[pack_int_id] = p;
288 install_packed_git(r, p);
289 list_add_tail(&p->mru, &r->objects->packed_git_mru);
290
291 return 0;
292 }
293
294 int nth_bitmapped_pack(struct repository *r, struct multi_pack_index *m,
295 struct bitmapped_pack *bp, uint32_t pack_int_id)
296 {
297 if (!m->chunk_bitmapped_packs)
298 return error(_("MIDX does not contain the BTMP chunk"));
299
300 if (prepare_midx_pack(r, m, pack_int_id))
301 return error(_("could not load bitmapped pack %"PRIu32), pack_int_id);
302
303 bp->p = m->packs[pack_int_id];
304 bp->bitmap_pos = get_be32((char *)m->chunk_bitmapped_packs +
305 MIDX_CHUNK_BITMAPPED_PACKS_WIDTH * pack_int_id);
306 bp->bitmap_nr = get_be32((char *)m->chunk_bitmapped_packs +
307 MIDX_CHUNK_BITMAPPED_PACKS_WIDTH * pack_int_id +
308 sizeof(uint32_t));
309 bp->pack_int_id = pack_int_id;
310
311 return 0;
312 }
313
314 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
315 {
316 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
317 the_hash_algo->rawsz, result);
318 }
319
320 struct object_id *nth_midxed_object_oid(struct object_id *oid,
321 struct multi_pack_index *m,
322 uint32_t n)
323 {
324 if (n >= m->num_objects)
325 return NULL;
326
327 oidread(oid, m->chunk_oid_lookup + st_mult(m->hash_len, n));
328 return oid;
329 }
330
331 off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
332 {
333 const unsigned char *offset_data;
334 uint32_t offset32;
335
336 offset_data = m->chunk_object_offsets + (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH;
337 offset32 = get_be32(offset_data + sizeof(uint32_t));
338
339 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
340 if (sizeof(off_t) < sizeof(uint64_t))
341 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
342
343 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
344 if (offset32 >= m->chunk_large_offsets_len / sizeof(uint64_t))
345 die(_("multi-pack-index large offset out of bounds"));
346 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
347 }
348
349 return offset32;
350 }
351
352 uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
353 {
354 return get_be32(m->chunk_object_offsets +
355 (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH);
356 }
357
358 int fill_midx_entry(struct repository *r,
359 const struct object_id *oid,
360 struct pack_entry *e,
361 struct multi_pack_index *m)
362 {
363 uint32_t pos;
364 uint32_t pack_int_id;
365 struct packed_git *p;
366
367 if (!bsearch_midx(oid, m, &pos))
368 return 0;
369
370 if (pos >= m->num_objects)
371 return 0;
372
373 pack_int_id = nth_midxed_pack_int_id(m, pos);
374
375 if (prepare_midx_pack(r, m, pack_int_id))
376 return 0;
377 p = m->packs[pack_int_id];
378
379 /*
380 * We are about to tell the caller where they can locate the
381 * requested object. We better make sure the packfile is
382 * still here and can be accessed before supplying that
383 * answer, as it may have been deleted since the MIDX was
384 * loaded!
385 */
386 if (!is_pack_valid(p))
387 return 0;
388
389 if (oidset_size(&p->bad_objects) &&
390 oidset_contains(&p->bad_objects, oid))
391 return 0;
392
393 e->offset = nth_midxed_offset(m, pos);
394 e->p = p;
395
396 return 1;
397 }
398
399 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
400 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
401 const char *idx_name)
402 {
403 /* Skip past any initial matching prefix. */
404 while (*idx_name && *idx_name == *idx_or_pack_name) {
405 idx_name++;
406 idx_or_pack_name++;
407 }
408
409 /*
410 * If we didn't match completely, we may have matched "pack-1234." and
411 * be left with "idx" and "pack" respectively, which is also OK. We do
412 * not have to check for "idx" and "idx", because that would have been
413 * a complete match (and in that case these strcmps will be false, but
414 * we'll correctly return 0 from the final strcmp() below.
415 *
416 * Technically this matches "fooidx" and "foopack", but we'd never have
417 * such names in the first place.
418 */
419 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
420 return 0;
421
422 /*
423 * This not only checks for a complete match, but also orders based on
424 * the first non-identical character, which means our ordering will
425 * match a raw strcmp(). That makes it OK to use this to binary search
426 * a naively-sorted list.
427 */
428 return strcmp(idx_or_pack_name, idx_name);
429 }
430
431 int midx_locate_pack(struct multi_pack_index *m, const char *idx_or_pack_name,
432 uint32_t *pos)
433 {
434 uint32_t first = 0, last = m->num_packs;
435
436 while (first < last) {
437 uint32_t mid = first + (last - first) / 2;
438 const char *current;
439 int cmp;
440
441 current = m->pack_names[mid];
442 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
443 if (!cmp) {
444 if (pos)
445 *pos = mid;
446 return 1;
447 }
448 if (cmp > 0) {
449 first = mid + 1;
450 continue;
451 }
452 last = mid;
453 }
454
455 return 0;
456 }
457
458 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
459 {
460 return midx_locate_pack(m, idx_or_pack_name, NULL);
461 }
462
463 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
464 {
465 struct multi_pack_index *m;
466 struct multi_pack_index *m_search;
467
468 prepare_repo_settings(r);
469 if (!r->settings.core_multi_pack_index)
470 return 0;
471
472 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
473 if (!strcmp(object_dir, m_search->object_dir))
474 return 1;
475
476 m = load_multi_pack_index(object_dir, local);
477
478 if (m) {
479 struct multi_pack_index *mp = r->objects->multi_pack_index;
480 if (mp) {
481 m->next = mp->next;
482 mp->next = m;
483 } else
484 r->objects->multi_pack_index = m;
485 return 1;
486 }
487
488 return 0;
489 }
490
491 static size_t write_midx_header(struct hashfile *f,
492 unsigned char num_chunks,
493 uint32_t num_packs)
494 {
495 hashwrite_be32(f, MIDX_SIGNATURE);
496 hashwrite_u8(f, MIDX_VERSION);
497 hashwrite_u8(f, oid_version(the_hash_algo));
498 hashwrite_u8(f, num_chunks);
499 hashwrite_u8(f, 0); /* unused */
500 hashwrite_be32(f, num_packs);
501
502 return MIDX_HEADER_SIZE;
503 }
504
505 #define BITMAP_POS_UNKNOWN (~((uint32_t)0))
506
507 struct pack_info {
508 uint32_t orig_pack_int_id;
509 char *pack_name;
510 struct packed_git *p;
511
512 uint32_t bitmap_pos;
513 uint32_t bitmap_nr;
514
515 unsigned expired : 1;
516 };
517
518 static void fill_pack_info(struct pack_info *info,
519 struct packed_git *p, const char *pack_name,
520 uint32_t orig_pack_int_id)
521 {
522 memset(info, 0, sizeof(struct pack_info));
523
524 info->orig_pack_int_id = orig_pack_int_id;
525 info->pack_name = xstrdup(pack_name);
526 info->p = p;
527 info->bitmap_pos = BITMAP_POS_UNKNOWN;
528 }
529
530 static int pack_info_compare(const void *_a, const void *_b)
531 {
532 struct pack_info *a = (struct pack_info *)_a;
533 struct pack_info *b = (struct pack_info *)_b;
534 return strcmp(a->pack_name, b->pack_name);
535 }
536
537 static int idx_or_pack_name_cmp(const void *_va, const void *_vb)
538 {
539 const char *pack_name = _va;
540 const struct pack_info *compar = _vb;
541
542 return cmp_idx_or_pack_name(pack_name, compar->pack_name);
543 }
544
545 struct write_midx_context {
546 struct pack_info *info;
547 size_t nr;
548 size_t alloc;
549 struct multi_pack_index *m;
550 struct progress *progress;
551 unsigned pack_paths_checked;
552
553 struct pack_midx_entry *entries;
554 size_t entries_nr;
555
556 uint32_t *pack_perm;
557 uint32_t *pack_order;
558 unsigned large_offsets_needed:1;
559 uint32_t num_large_offsets;
560
561 int preferred_pack_idx;
562
563 struct string_list *to_include;
564 };
565
566 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
567 const char *file_name, void *data)
568 {
569 struct write_midx_context *ctx = data;
570 struct packed_git *p;
571
572 if (ends_with(file_name, ".idx")) {
573 display_progress(ctx->progress, ++ctx->pack_paths_checked);
574 /*
575 * Note that at most one of ctx->m and ctx->to_include are set,
576 * so we are testing midx_contains_pack() and
577 * string_list_has_string() independently (guarded by the
578 * appropriate NULL checks).
579 *
580 * We could support passing to_include while reusing an existing
581 * MIDX, but don't currently since the reuse process drags
582 * forward all packs from an existing MIDX (without checking
583 * whether or not they appear in the to_include list).
584 *
585 * If we added support for that, these next two conditional
586 * should be performed independently (likely checking
587 * to_include before the existing MIDX).
588 */
589 if (ctx->m && midx_contains_pack(ctx->m, file_name))
590 return;
591 else if (ctx->to_include &&
592 !string_list_has_string(ctx->to_include, file_name))
593 return;
594
595 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
596
597 p = add_packed_git(full_path, full_path_len, 0);
598 if (!p) {
599 warning(_("failed to add packfile '%s'"),
600 full_path);
601 return;
602 }
603
604 if (open_pack_index(p)) {
605 warning(_("failed to open pack-index '%s'"),
606 full_path);
607 close_pack(p);
608 free(p);
609 return;
610 }
611
612 fill_pack_info(&ctx->info[ctx->nr], p, file_name, ctx->nr);
613 ctx->nr++;
614 }
615 }
616
617 struct pack_midx_entry {
618 struct object_id oid;
619 uint32_t pack_int_id;
620 time_t pack_mtime;
621 uint64_t offset;
622 unsigned preferred : 1;
623 };
624
625 static int midx_oid_compare(const void *_a, const void *_b)
626 {
627 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
628 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
629 int cmp = oidcmp(&a->oid, &b->oid);
630
631 if (cmp)
632 return cmp;
633
634 /* Sort objects in a preferred pack first when multiple copies exist. */
635 if (a->preferred > b->preferred)
636 return -1;
637 if (a->preferred < b->preferred)
638 return 1;
639
640 if (a->pack_mtime > b->pack_mtime)
641 return -1;
642 else if (a->pack_mtime < b->pack_mtime)
643 return 1;
644
645 return a->pack_int_id - b->pack_int_id;
646 }
647
648 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
649 struct pack_midx_entry *e,
650 uint32_t pos)
651 {
652 if (pos >= m->num_objects)
653 return 1;
654
655 nth_midxed_object_oid(&e->oid, m, pos);
656 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
657 e->offset = nth_midxed_offset(m, pos);
658
659 /* consider objects in midx to be from "old" packs */
660 e->pack_mtime = 0;
661 return 0;
662 }
663
664 static void fill_pack_entry(uint32_t pack_int_id,
665 struct packed_git *p,
666 uint32_t cur_object,
667 struct pack_midx_entry *entry,
668 int preferred)
669 {
670 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
671 die(_("failed to locate object %d in packfile"), cur_object);
672
673 entry->pack_int_id = pack_int_id;
674 entry->pack_mtime = p->mtime;
675
676 entry->offset = nth_packed_object_offset(p, cur_object);
677 entry->preferred = !!preferred;
678 }
679
680 struct midx_fanout {
681 struct pack_midx_entry *entries;
682 size_t nr, alloc;
683 };
684
685 static void midx_fanout_grow(struct midx_fanout *fanout, size_t nr)
686 {
687 if (nr < fanout->nr)
688 BUG("negative growth in midx_fanout_grow() (%"PRIuMAX" < %"PRIuMAX")",
689 (uintmax_t)nr, (uintmax_t)fanout->nr);
690 ALLOC_GROW(fanout->entries, nr, fanout->alloc);
691 }
692
693 static void midx_fanout_sort(struct midx_fanout *fanout)
694 {
695 QSORT(fanout->entries, fanout->nr, midx_oid_compare);
696 }
697
698 static void midx_fanout_add_midx_fanout(struct midx_fanout *fanout,
699 struct multi_pack_index *m,
700 uint32_t cur_fanout,
701 int preferred_pack)
702 {
703 uint32_t start = 0, end;
704 uint32_t cur_object;
705
706 if (cur_fanout)
707 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
708 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
709
710 for (cur_object = start; cur_object < end; cur_object++) {
711 if ((preferred_pack > -1) &&
712 (preferred_pack == nth_midxed_pack_int_id(m, cur_object))) {
713 /*
714 * Objects from preferred packs are added
715 * separately.
716 */
717 continue;
718 }
719
720 midx_fanout_grow(fanout, fanout->nr + 1);
721 nth_midxed_pack_midx_entry(m,
722 &fanout->entries[fanout->nr],
723 cur_object);
724 fanout->entries[fanout->nr].preferred = 0;
725 fanout->nr++;
726 }
727 }
728
729 static void midx_fanout_add_pack_fanout(struct midx_fanout *fanout,
730 struct pack_info *info,
731 uint32_t cur_pack,
732 int preferred,
733 uint32_t cur_fanout)
734 {
735 struct packed_git *pack = info[cur_pack].p;
736 uint32_t start = 0, end;
737 uint32_t cur_object;
738
739 if (cur_fanout)
740 start = get_pack_fanout(pack, cur_fanout - 1);
741 end = get_pack_fanout(pack, cur_fanout);
742
743 for (cur_object = start; cur_object < end; cur_object++) {
744 midx_fanout_grow(fanout, fanout->nr + 1);
745 fill_pack_entry(cur_pack,
746 info[cur_pack].p,
747 cur_object,
748 &fanout->entries[fanout->nr],
749 preferred);
750 fanout->nr++;
751 }
752 }
753
754 /*
755 * It is possible to artificially get into a state where there are many
756 * duplicate copies of objects. That can create high memory pressure if
757 * we are to create a list of all objects before de-duplication. To reduce
758 * this memory pressure without a significant performance drop, automatically
759 * group objects by the first byte of their object id. Use the IDX fanout
760 * tables to group the data, copy to a local array, then sort.
761 *
762 * Copy only the de-duplicated entries (selected by most-recent modified time
763 * of a packfile containing the object).
764 */
765 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
766 struct pack_info *info,
767 uint32_t nr_packs,
768 size_t *nr_objects,
769 int preferred_pack)
770 {
771 uint32_t cur_fanout, cur_pack, cur_object;
772 size_t alloc_objects, total_objects = 0;
773 struct midx_fanout fanout = { 0 };
774 struct pack_midx_entry *deduplicated_entries = NULL;
775 uint32_t start_pack = m ? m->num_packs : 0;
776
777 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
778 total_objects = st_add(total_objects,
779 info[cur_pack].p->num_objects);
780
781 /*
782 * As we de-duplicate by fanout value, we expect the fanout
783 * slices to be evenly distributed, with some noise. Hence,
784 * allocate slightly more than one 256th.
785 */
786 alloc_objects = fanout.alloc = total_objects > 3200 ? total_objects / 200 : 16;
787
788 ALLOC_ARRAY(fanout.entries, fanout.alloc);
789 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
790 *nr_objects = 0;
791
792 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
793 fanout.nr = 0;
794
795 if (m)
796 midx_fanout_add_midx_fanout(&fanout, m, cur_fanout,
797 preferred_pack);
798
799 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
800 int preferred = cur_pack == preferred_pack;
801 midx_fanout_add_pack_fanout(&fanout,
802 info, cur_pack,
803 preferred, cur_fanout);
804 }
805
806 if (-1 < preferred_pack && preferred_pack < start_pack)
807 midx_fanout_add_pack_fanout(&fanout, info,
808 preferred_pack, 1,
809 cur_fanout);
810
811 midx_fanout_sort(&fanout);
812
813 /*
814 * The batch is now sorted by OID and then mtime (descending).
815 * Take only the first duplicate.
816 */
817 for (cur_object = 0; cur_object < fanout.nr; cur_object++) {
818 if (cur_object && oideq(&fanout.entries[cur_object - 1].oid,
819 &fanout.entries[cur_object].oid))
820 continue;
821
822 ALLOC_GROW(deduplicated_entries, st_add(*nr_objects, 1),
823 alloc_objects);
824 memcpy(&deduplicated_entries[*nr_objects],
825 &fanout.entries[cur_object],
826 sizeof(struct pack_midx_entry));
827 (*nr_objects)++;
828 }
829 }
830
831 free(fanout.entries);
832 return deduplicated_entries;
833 }
834
835 static int write_midx_pack_names(struct hashfile *f, void *data)
836 {
837 struct write_midx_context *ctx = data;
838 uint32_t i;
839 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
840 size_t written = 0;
841
842 for (i = 0; i < ctx->nr; i++) {
843 size_t writelen;
844
845 if (ctx->info[i].expired)
846 continue;
847
848 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
849 BUG("incorrect pack-file order: %s before %s",
850 ctx->info[i - 1].pack_name,
851 ctx->info[i].pack_name);
852
853 writelen = strlen(ctx->info[i].pack_name) + 1;
854 hashwrite(f, ctx->info[i].pack_name, writelen);
855 written += writelen;
856 }
857
858 /* add padding to be aligned */
859 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
860 if (i < MIDX_CHUNK_ALIGNMENT) {
861 memset(padding, 0, sizeof(padding));
862 hashwrite(f, padding, i);
863 }
864
865 return 0;
866 }
867
868 static int write_midx_bitmapped_packs(struct hashfile *f, void *data)
869 {
870 struct write_midx_context *ctx = data;
871 size_t i;
872
873 for (i = 0; i < ctx->nr; i++) {
874 struct pack_info *pack = &ctx->info[i];
875 if (pack->expired)
876 continue;
877
878 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN && pack->bitmap_nr)
879 BUG("pack '%s' has no bitmap position, but has %d bitmapped object(s)",
880 pack->pack_name, pack->bitmap_nr);
881
882 hashwrite_be32(f, pack->bitmap_pos);
883 hashwrite_be32(f, pack->bitmap_nr);
884 }
885 return 0;
886 }
887
888 static int write_midx_oid_fanout(struct hashfile *f,
889 void *data)
890 {
891 struct write_midx_context *ctx = data;
892 struct pack_midx_entry *list = ctx->entries;
893 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
894 uint32_t count = 0;
895 uint32_t i;
896
897 /*
898 * Write the first-level table (the list is sorted,
899 * but we use a 256-entry lookup to be able to avoid
900 * having to do eight extra binary search iterations).
901 */
902 for (i = 0; i < 256; i++) {
903 struct pack_midx_entry *next = list;
904
905 while (next < last && next->oid.hash[0] == i) {
906 count++;
907 next++;
908 }
909
910 hashwrite_be32(f, count);
911 list = next;
912 }
913
914 return 0;
915 }
916
917 static int write_midx_oid_lookup(struct hashfile *f,
918 void *data)
919 {
920 struct write_midx_context *ctx = data;
921 unsigned char hash_len = the_hash_algo->rawsz;
922 struct pack_midx_entry *list = ctx->entries;
923 uint32_t i;
924
925 for (i = 0; i < ctx->entries_nr; i++) {
926 struct pack_midx_entry *obj = list++;
927
928 if (i < ctx->entries_nr - 1) {
929 struct pack_midx_entry *next = list;
930 if (oidcmp(&obj->oid, &next->oid) >= 0)
931 BUG("OIDs not in order: %s >= %s",
932 oid_to_hex(&obj->oid),
933 oid_to_hex(&next->oid));
934 }
935
936 hashwrite(f, obj->oid.hash, (int)hash_len);
937 }
938
939 return 0;
940 }
941
942 static int write_midx_object_offsets(struct hashfile *f,
943 void *data)
944 {
945 struct write_midx_context *ctx = data;
946 struct pack_midx_entry *list = ctx->entries;
947 uint32_t i, nr_large_offset = 0;
948
949 for (i = 0; i < ctx->entries_nr; i++) {
950 struct pack_midx_entry *obj = list++;
951
952 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
953 BUG("object %s is in an expired pack with int-id %d",
954 oid_to_hex(&obj->oid),
955 obj->pack_int_id);
956
957 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
958
959 if (ctx->large_offsets_needed && obj->offset >> 31)
960 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
961 else if (!ctx->large_offsets_needed && obj->offset >> 32)
962 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
963 oid_to_hex(&obj->oid),
964 obj->offset);
965 else
966 hashwrite_be32(f, (uint32_t)obj->offset);
967 }
968
969 return 0;
970 }
971
972 static int write_midx_large_offsets(struct hashfile *f,
973 void *data)
974 {
975 struct write_midx_context *ctx = data;
976 struct pack_midx_entry *list = ctx->entries;
977 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
978 uint32_t nr_large_offset = ctx->num_large_offsets;
979
980 while (nr_large_offset) {
981 struct pack_midx_entry *obj;
982 uint64_t offset;
983
984 if (list >= end)
985 BUG("too many large-offset objects");
986
987 obj = list++;
988 offset = obj->offset;
989
990 if (!(offset >> 31))
991 continue;
992
993 hashwrite_be64(f, offset);
994
995 nr_large_offset--;
996 }
997
998 return 0;
999 }
1000
1001 static int write_midx_revindex(struct hashfile *f,
1002 void *data)
1003 {
1004 struct write_midx_context *ctx = data;
1005 uint32_t i;
1006
1007 for (i = 0; i < ctx->entries_nr; i++)
1008 hashwrite_be32(f, ctx->pack_order[i]);
1009
1010 return 0;
1011 }
1012
1013 struct midx_pack_order_data {
1014 uint32_t nr;
1015 uint32_t pack;
1016 off_t offset;
1017 };
1018
1019 static int midx_pack_order_cmp(const void *va, const void *vb)
1020 {
1021 const struct midx_pack_order_data *a = va, *b = vb;
1022 if (a->pack < b->pack)
1023 return -1;
1024 else if (a->pack > b->pack)
1025 return 1;
1026 else if (a->offset < b->offset)
1027 return -1;
1028 else if (a->offset > b->offset)
1029 return 1;
1030 else
1031 return 0;
1032 }
1033
1034 static uint32_t *midx_pack_order(struct write_midx_context *ctx)
1035 {
1036 struct midx_pack_order_data *data;
1037 uint32_t *pack_order;
1038 uint32_t i;
1039
1040 trace2_region_enter("midx", "midx_pack_order", the_repository);
1041
1042 ALLOC_ARRAY(data, ctx->entries_nr);
1043 for (i = 0; i < ctx->entries_nr; i++) {
1044 struct pack_midx_entry *e = &ctx->entries[i];
1045 data[i].nr = i;
1046 data[i].pack = ctx->pack_perm[e->pack_int_id];
1047 if (!e->preferred)
1048 data[i].pack |= (1U << 31);
1049 data[i].offset = e->offset;
1050 }
1051
1052 QSORT(data, ctx->entries_nr, midx_pack_order_cmp);
1053
1054 ALLOC_ARRAY(pack_order, ctx->entries_nr);
1055 for (i = 0; i < ctx->entries_nr; i++) {
1056 struct pack_midx_entry *e = &ctx->entries[data[i].nr];
1057 struct pack_info *pack = &ctx->info[ctx->pack_perm[e->pack_int_id]];
1058 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN)
1059 pack->bitmap_pos = i;
1060 pack->bitmap_nr++;
1061 pack_order[i] = data[i].nr;
1062 }
1063 for (i = 0; i < ctx->nr; i++) {
1064 struct pack_info *pack = &ctx->info[ctx->pack_perm[i]];
1065 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN)
1066 pack->bitmap_pos = 0;
1067 }
1068 free(data);
1069
1070 trace2_region_leave("midx", "midx_pack_order", the_repository);
1071
1072 return pack_order;
1073 }
1074
1075 static void write_midx_reverse_index(char *midx_name, unsigned char *midx_hash,
1076 struct write_midx_context *ctx)
1077 {
1078 struct strbuf buf = STRBUF_INIT;
1079 const char *tmp_file;
1080
1081 trace2_region_enter("midx", "write_midx_reverse_index", the_repository);
1082
1083 strbuf_addf(&buf, "%s-%s.rev", midx_name, hash_to_hex(midx_hash));
1084
1085 tmp_file = write_rev_file_order(NULL, ctx->pack_order, ctx->entries_nr,
1086 midx_hash, WRITE_REV);
1087
1088 if (finalize_object_file(tmp_file, buf.buf))
1089 die(_("cannot store reverse index file"));
1090
1091 strbuf_release(&buf);
1092
1093 trace2_region_leave("midx", "write_midx_reverse_index", the_repository);
1094 }
1095
1096 static void clear_midx_files_ext(const char *object_dir, const char *ext,
1097 unsigned char *keep_hash);
1098
1099 static int midx_checksum_valid(struct multi_pack_index *m)
1100 {
1101 return hashfile_checksum_valid(m->data, m->data_len);
1102 }
1103
1104 static void prepare_midx_packing_data(struct packing_data *pdata,
1105 struct write_midx_context *ctx)
1106 {
1107 uint32_t i;
1108
1109 trace2_region_enter("midx", "prepare_midx_packing_data", the_repository);
1110
1111 memset(pdata, 0, sizeof(struct packing_data));
1112 prepare_packing_data(the_repository, pdata);
1113
1114 for (i = 0; i < ctx->entries_nr; i++) {
1115 struct pack_midx_entry *from = &ctx->entries[ctx->pack_order[i]];
1116 struct object_entry *to = packlist_alloc(pdata, &from->oid);
1117
1118 oe_set_in_pack(pdata, to,
1119 ctx->info[ctx->pack_perm[from->pack_int_id]].p);
1120 }
1121
1122 trace2_region_leave("midx", "prepare_midx_packing_data", the_repository);
1123 }
1124
1125 static int add_ref_to_pending(const char *refname,
1126 const struct object_id *oid,
1127 int flag, void *cb_data)
1128 {
1129 struct rev_info *revs = (struct rev_info*)cb_data;
1130 struct object_id peeled;
1131 struct object *object;
1132
1133 if ((flag & REF_ISSYMREF) && (flag & REF_ISBROKEN)) {
1134 warning("symbolic ref is dangling: %s", refname);
1135 return 0;
1136 }
1137
1138 if (!peel_iterated_oid(oid, &peeled))
1139 oid = &peeled;
1140
1141 object = parse_object_or_die(oid, refname);
1142 if (object->type != OBJ_COMMIT)
1143 return 0;
1144
1145 add_pending_object(revs, object, "");
1146 if (bitmap_is_preferred_refname(revs->repo, refname))
1147 object->flags |= NEEDS_BITMAP;
1148 return 0;
1149 }
1150
1151 struct bitmap_commit_cb {
1152 struct commit **commits;
1153 size_t commits_nr, commits_alloc;
1154
1155 struct write_midx_context *ctx;
1156 };
1157
1158 static const struct object_id *bitmap_oid_access(size_t index,
1159 const void *_entries)
1160 {
1161 const struct pack_midx_entry *entries = _entries;
1162 return &entries[index].oid;
1163 }
1164
1165 static void bitmap_show_commit(struct commit *commit, void *_data)
1166 {
1167 struct bitmap_commit_cb *data = _data;
1168 int pos = oid_pos(&commit->object.oid, data->ctx->entries,
1169 data->ctx->entries_nr,
1170 bitmap_oid_access);
1171 if (pos < 0)
1172 return;
1173
1174 ALLOC_GROW(data->commits, data->commits_nr + 1, data->commits_alloc);
1175 data->commits[data->commits_nr++] = commit;
1176 }
1177
1178 static int read_refs_snapshot(const char *refs_snapshot,
1179 struct rev_info *revs)
1180 {
1181 struct strbuf buf = STRBUF_INIT;
1182 struct object_id oid;
1183 FILE *f = xfopen(refs_snapshot, "r");
1184
1185 while (strbuf_getline(&buf, f) != EOF) {
1186 struct object *object;
1187 int preferred = 0;
1188 char *hex = buf.buf;
1189 const char *end = NULL;
1190
1191 if (buf.len && *buf.buf == '+') {
1192 preferred = 1;
1193 hex = &buf.buf[1];
1194 }
1195
1196 if (parse_oid_hex(hex, &oid, &end) < 0)
1197 die(_("could not parse line: %s"), buf.buf);
1198 if (*end)
1199 die(_("malformed line: %s"), buf.buf);
1200
1201 object = parse_object_or_die(&oid, NULL);
1202 if (preferred)
1203 object->flags |= NEEDS_BITMAP;
1204
1205 add_pending_object(revs, object, "");
1206 }
1207
1208 fclose(f);
1209 strbuf_release(&buf);
1210 return 0;
1211 }
1212
1213 static struct commit **find_commits_for_midx_bitmap(uint32_t *indexed_commits_nr_p,
1214 const char *refs_snapshot,
1215 struct write_midx_context *ctx)
1216 {
1217 struct rev_info revs;
1218 struct bitmap_commit_cb cb = {0};
1219
1220 trace2_region_enter("midx", "find_commits_for_midx_bitmap",
1221 the_repository);
1222
1223 cb.ctx = ctx;
1224
1225 repo_init_revisions(the_repository, &revs, NULL);
1226 if (refs_snapshot) {
1227 read_refs_snapshot(refs_snapshot, &revs);
1228 } else {
1229 setup_revisions(0, NULL, &revs, NULL);
1230 for_each_ref(add_ref_to_pending, &revs);
1231 }
1232
1233 /*
1234 * Skipping promisor objects here is intentional, since it only excludes
1235 * them from the list of reachable commits that we want to select from
1236 * when computing the selection of MIDX'd commits to receive bitmaps.
1237 *
1238 * Reachability bitmaps do require that their objects be closed under
1239 * reachability, but fetching any objects missing from promisors at this
1240 * point is too late. But, if one of those objects can be reached from
1241 * an another object that is included in the bitmap, then we will
1242 * complain later that we don't have reachability closure (and fail
1243 * appropriately).
1244 */
1245 fetch_if_missing = 0;
1246 revs.exclude_promisor_objects = 1;
1247
1248 if (prepare_revision_walk(&revs))
1249 die(_("revision walk setup failed"));
1250
1251 traverse_commit_list(&revs, bitmap_show_commit, NULL, &cb);
1252 if (indexed_commits_nr_p)
1253 *indexed_commits_nr_p = cb.commits_nr;
1254
1255 release_revisions(&revs);
1256
1257 trace2_region_leave("midx", "find_commits_for_midx_bitmap",
1258 the_repository);
1259
1260 return cb.commits;
1261 }
1262
1263 static int write_midx_bitmap(const char *midx_name,
1264 const unsigned char *midx_hash,
1265 struct packing_data *pdata,
1266 struct commit **commits,
1267 uint32_t commits_nr,
1268 uint32_t *pack_order,
1269 unsigned flags)
1270 {
1271 int ret, i;
1272 uint16_t options = 0;
1273 struct pack_idx_entry **index;
1274 char *bitmap_name = xstrfmt("%s-%s.bitmap", midx_name,
1275 hash_to_hex(midx_hash));
1276
1277 trace2_region_enter("midx", "write_midx_bitmap", the_repository);
1278
1279 if (flags & MIDX_WRITE_BITMAP_HASH_CACHE)
1280 options |= BITMAP_OPT_HASH_CACHE;
1281
1282 if (flags & MIDX_WRITE_BITMAP_LOOKUP_TABLE)
1283 options |= BITMAP_OPT_LOOKUP_TABLE;
1284
1285 /*
1286 * Build the MIDX-order index based on pdata.objects (which is already
1287 * in MIDX order; c.f., 'midx_pack_order_cmp()' for the definition of
1288 * this order).
1289 */
1290 ALLOC_ARRAY(index, pdata->nr_objects);
1291 for (i = 0; i < pdata->nr_objects; i++)
1292 index[i] = &pdata->objects[i].idx;
1293
1294 bitmap_writer_show_progress(flags & MIDX_PROGRESS);
1295 bitmap_writer_build_type_index(pdata, index, pdata->nr_objects);
1296
1297 /*
1298 * bitmap_writer_finish expects objects in lex order, but pack_order
1299 * gives us exactly that. use it directly instead of re-sorting the
1300 * array.
1301 *
1302 * This changes the order of objects in 'index' between
1303 * bitmap_writer_build_type_index and bitmap_writer_finish.
1304 *
1305 * The same re-ordering takes place in the single-pack bitmap code via
1306 * write_idx_file(), which is called by finish_tmp_packfile(), which
1307 * happens between bitmap_writer_build_type_index() and
1308 * bitmap_writer_finish().
1309 */
1310 for (i = 0; i < pdata->nr_objects; i++)
1311 index[pack_order[i]] = &pdata->objects[i].idx;
1312
1313 bitmap_writer_select_commits(commits, commits_nr, -1);
1314 ret = bitmap_writer_build(pdata);
1315 if (ret < 0)
1316 goto cleanup;
1317
1318 bitmap_writer_set_checksum(midx_hash);
1319 bitmap_writer_finish(index, pdata->nr_objects, bitmap_name, options);
1320
1321 cleanup:
1322 free(index);
1323 free(bitmap_name);
1324
1325 trace2_region_leave("midx", "write_midx_bitmap", the_repository);
1326
1327 return ret;
1328 }
1329
1330 static struct multi_pack_index *lookup_multi_pack_index(struct repository *r,
1331 const char *object_dir)
1332 {
1333 struct multi_pack_index *result = NULL;
1334 struct multi_pack_index *cur;
1335 char *obj_dir_real = real_pathdup(object_dir, 1);
1336 struct strbuf cur_path_real = STRBUF_INIT;
1337
1338 /* Ensure the given object_dir is local, or a known alternate. */
1339 find_odb(r, obj_dir_real);
1340
1341 for (cur = get_multi_pack_index(r); cur; cur = cur->next) {
1342 strbuf_realpath(&cur_path_real, cur->object_dir, 1);
1343 if (!strcmp(obj_dir_real, cur_path_real.buf)) {
1344 result = cur;
1345 goto cleanup;
1346 }
1347 }
1348
1349 cleanup:
1350 free(obj_dir_real);
1351 strbuf_release(&cur_path_real);
1352 return result;
1353 }
1354
1355 static int write_midx_internal(const char *object_dir,
1356 struct string_list *packs_to_include,
1357 struct string_list *packs_to_drop,
1358 const char *preferred_pack_name,
1359 const char *refs_snapshot,
1360 unsigned flags)
1361 {
1362 struct strbuf midx_name = STRBUF_INIT;
1363 unsigned char midx_hash[GIT_MAX_RAWSZ];
1364 uint32_t i;
1365 struct hashfile *f = NULL;
1366 struct lock_file lk;
1367 struct write_midx_context ctx = { 0 };
1368 int bitmapped_packs_concat_len = 0;
1369 int pack_name_concat_len = 0;
1370 int dropped_packs = 0;
1371 int result = 0;
1372 struct chunkfile *cf;
1373
1374 trace2_region_enter("midx", "write_midx_internal", the_repository);
1375
1376 get_midx_filename(&midx_name, object_dir);
1377 if (safe_create_leading_directories(midx_name.buf))
1378 die_errno(_("unable to create leading directories of %s"),
1379 midx_name.buf);
1380
1381 if (!packs_to_include) {
1382 /*
1383 * Only reference an existing MIDX when not filtering which
1384 * packs to include, since all packs and objects are copied
1385 * blindly from an existing MIDX if one is present.
1386 */
1387 ctx.m = lookup_multi_pack_index(the_repository, object_dir);
1388 }
1389
1390 if (ctx.m && !midx_checksum_valid(ctx.m)) {
1391 warning(_("ignoring existing multi-pack-index; checksum mismatch"));
1392 ctx.m = NULL;
1393 }
1394
1395 ctx.nr = 0;
1396 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
1397 ctx.info = NULL;
1398 ALLOC_ARRAY(ctx.info, ctx.alloc);
1399
1400 if (ctx.m) {
1401 for (i = 0; i < ctx.m->num_packs; i++) {
1402 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
1403
1404 if (flags & MIDX_WRITE_REV_INDEX) {
1405 /*
1406 * If generating a reverse index, need to have
1407 * packed_git's loaded to compare their
1408 * mtimes and object count.
1409 */
1410 if (prepare_midx_pack(the_repository, ctx.m, i)) {
1411 error(_("could not load pack"));
1412 result = 1;
1413 goto cleanup;
1414 }
1415
1416 if (open_pack_index(ctx.m->packs[i]))
1417 die(_("could not open index for %s"),
1418 ctx.m->packs[i]->pack_name);
1419 }
1420
1421 fill_pack_info(&ctx.info[ctx.nr++], ctx.m->packs[i],
1422 ctx.m->pack_names[i], i);
1423 }
1424 }
1425
1426 ctx.pack_paths_checked = 0;
1427 if (flags & MIDX_PROGRESS)
1428 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
1429 else
1430 ctx.progress = NULL;
1431
1432 ctx.to_include = packs_to_include;
1433
1434 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
1435 stop_progress(&ctx.progress);
1436
1437 if ((ctx.m && ctx.nr == ctx.m->num_packs) &&
1438 !(packs_to_include || packs_to_drop)) {
1439 struct bitmap_index *bitmap_git;
1440 int bitmap_exists;
1441 int want_bitmap = flags & MIDX_WRITE_BITMAP;
1442
1443 bitmap_git = prepare_midx_bitmap_git(ctx.m);
1444 bitmap_exists = bitmap_git && bitmap_is_midx(bitmap_git);
1445 free_bitmap_index(bitmap_git);
1446
1447 if (bitmap_exists || !want_bitmap) {
1448 /*
1449 * The correct MIDX already exists, and so does a
1450 * corresponding bitmap (or one wasn't requested).
1451 */
1452 if (!want_bitmap)
1453 clear_midx_files_ext(object_dir, ".bitmap",
1454 NULL);
1455 goto cleanup;
1456 }
1457 }
1458
1459 if (preferred_pack_name) {
1460 ctx.preferred_pack_idx = -1;
1461
1462 for (i = 0; i < ctx.nr; i++) {
1463 if (!cmp_idx_or_pack_name(preferred_pack_name,
1464 ctx.info[i].pack_name)) {
1465 ctx.preferred_pack_idx = i;
1466 break;
1467 }
1468 }
1469
1470 if (ctx.preferred_pack_idx == -1)
1471 warning(_("unknown preferred pack: '%s'"),
1472 preferred_pack_name);
1473 } else if (ctx.nr &&
1474 (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP))) {
1475 struct packed_git *oldest = ctx.info[ctx.preferred_pack_idx].p;
1476 ctx.preferred_pack_idx = 0;
1477
1478 if (packs_to_drop && packs_to_drop->nr)
1479 BUG("cannot write a MIDX bitmap during expiration");
1480
1481 /*
1482 * set a preferred pack when writing a bitmap to ensure that
1483 * the pack from which the first object is selected in pseudo
1484 * pack-order has all of its objects selected from that pack
1485 * (and not another pack containing a duplicate)
1486 */
1487 for (i = 1; i < ctx.nr; i++) {
1488 struct packed_git *p = ctx.info[i].p;
1489
1490 if (!oldest->num_objects || p->mtime < oldest->mtime) {
1491 oldest = p;
1492 ctx.preferred_pack_idx = i;
1493 }
1494 }
1495
1496 if (!oldest->num_objects) {
1497 /*
1498 * If all packs are empty; unset the preferred index.
1499 * This is acceptable since there will be no duplicate
1500 * objects to resolve, so the preferred value doesn't
1501 * matter.
1502 */
1503 ctx.preferred_pack_idx = -1;
1504 }
1505 } else {
1506 /*
1507 * otherwise don't mark any pack as preferred to avoid
1508 * interfering with expiration logic below
1509 */
1510 ctx.preferred_pack_idx = -1;
1511 }
1512
1513 if (ctx.preferred_pack_idx > -1) {
1514 struct packed_git *preferred = ctx.info[ctx.preferred_pack_idx].p;
1515 if (!preferred->num_objects) {
1516 error(_("cannot select preferred pack %s with no objects"),
1517 preferred->pack_name);
1518 result = 1;
1519 goto cleanup;
1520 }
1521 }
1522
1523 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr,
1524 ctx.preferred_pack_idx);
1525
1526 ctx.large_offsets_needed = 0;
1527 for (i = 0; i < ctx.entries_nr; i++) {
1528 if (ctx.entries[i].offset > 0x7fffffff)
1529 ctx.num_large_offsets++;
1530 if (ctx.entries[i].offset > 0xffffffff)
1531 ctx.large_offsets_needed = 1;
1532 }
1533
1534 QSORT(ctx.info, ctx.nr, pack_info_compare);
1535
1536 if (packs_to_drop && packs_to_drop->nr) {
1537 int drop_index = 0;
1538 int missing_drops = 0;
1539
1540 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
1541 int cmp = strcmp(ctx.info[i].pack_name,
1542 packs_to_drop->items[drop_index].string);
1543
1544 if (!cmp) {
1545 drop_index++;
1546 ctx.info[i].expired = 1;
1547 } else if (cmp > 0) {
1548 error(_("did not see pack-file %s to drop"),
1549 packs_to_drop->items[drop_index].string);
1550 drop_index++;
1551 missing_drops++;
1552 i--;
1553 } else {
1554 ctx.info[i].expired = 0;
1555 }
1556 }
1557
1558 if (missing_drops) {
1559 result = 1;
1560 goto cleanup;
1561 }
1562 }
1563
1564 /*
1565 * pack_perm stores a permutation between pack-int-ids from the
1566 * previous multi-pack-index to the new one we are writing:
1567 *
1568 * pack_perm[old_id] = new_id
1569 */
1570 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
1571 for (i = 0; i < ctx.nr; i++) {
1572 if (ctx.info[i].expired) {
1573 dropped_packs++;
1574 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
1575 } else {
1576 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
1577 }
1578 }
1579
1580 for (i = 0; i < ctx.nr; i++) {
1581 if (ctx.info[i].expired)
1582 continue;
1583 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
1584 bitmapped_packs_concat_len += 2 * sizeof(uint32_t);
1585 }
1586
1587 /* Check that the preferred pack wasn't expired (if given). */
1588 if (preferred_pack_name) {
1589 struct pack_info *preferred = bsearch(preferred_pack_name,
1590 ctx.info, ctx.nr,
1591 sizeof(*ctx.info),
1592 idx_or_pack_name_cmp);
1593 if (preferred) {
1594 uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id];
1595 if (perm == PACK_EXPIRED)
1596 warning(_("preferred pack '%s' is expired"),
1597 preferred_pack_name);
1598 }
1599 }
1600
1601 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
1602 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
1603 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
1604
1605 hold_lock_file_for_update(&lk, midx_name.buf, LOCK_DIE_ON_ERROR);
1606 f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk));
1607
1608 if (ctx.nr - dropped_packs == 0) {
1609 error(_("no pack files to index."));
1610 result = 1;
1611 goto cleanup;
1612 }
1613
1614 if (!ctx.entries_nr) {
1615 if (flags & MIDX_WRITE_BITMAP)
1616 warning(_("refusing to write multi-pack .bitmap without any objects"));
1617 flags &= ~(MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP);
1618 }
1619
1620 cf = init_chunkfile(f);
1621
1622 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
1623 write_midx_pack_names);
1624 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
1625 write_midx_oid_fanout);
1626 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
1627 st_mult(ctx.entries_nr, the_hash_algo->rawsz),
1628 write_midx_oid_lookup);
1629 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
1630 st_mult(ctx.entries_nr, MIDX_CHUNK_OFFSET_WIDTH),
1631 write_midx_object_offsets);
1632
1633 if (ctx.large_offsets_needed)
1634 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
1635 st_mult(ctx.num_large_offsets,
1636 MIDX_CHUNK_LARGE_OFFSET_WIDTH),
1637 write_midx_large_offsets);
1638
1639 if (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP)) {
1640 ctx.pack_order = midx_pack_order(&ctx);
1641 add_chunk(cf, MIDX_CHUNKID_REVINDEX,
1642 st_mult(ctx.entries_nr, sizeof(uint32_t)),
1643 write_midx_revindex);
1644 add_chunk(cf, MIDX_CHUNKID_BITMAPPEDPACKS,
1645 bitmapped_packs_concat_len,
1646 write_midx_bitmapped_packs);
1647 }
1648
1649 write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs);
1650 write_chunkfile(cf, &ctx);
1651
1652 finalize_hashfile(f, midx_hash, FSYNC_COMPONENT_PACK_METADATA,
1653 CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1654 free_chunkfile(cf);
1655
1656 if (flags & MIDX_WRITE_REV_INDEX &&
1657 git_env_bool("GIT_TEST_MIDX_WRITE_REV", 0))
1658 write_midx_reverse_index(midx_name.buf, midx_hash, &ctx);
1659
1660 if (flags & MIDX_WRITE_BITMAP) {
1661 struct packing_data pdata;
1662 struct commit **commits;
1663 uint32_t commits_nr;
1664
1665 if (!ctx.entries_nr)
1666 BUG("cannot write a bitmap without any objects");
1667
1668 prepare_midx_packing_data(&pdata, &ctx);
1669
1670 commits = find_commits_for_midx_bitmap(&commits_nr, refs_snapshot, &ctx);
1671
1672 /*
1673 * The previous steps translated the information from
1674 * 'entries' into information suitable for constructing
1675 * bitmaps. We no longer need that array, so clear it to
1676 * reduce memory pressure.
1677 */
1678 FREE_AND_NULL(ctx.entries);
1679 ctx.entries_nr = 0;
1680
1681 if (write_midx_bitmap(midx_name.buf, midx_hash, &pdata,
1682 commits, commits_nr, ctx.pack_order,
1683 flags) < 0) {
1684 error(_("could not write multi-pack bitmap"));
1685 result = 1;
1686 clear_packing_data(&pdata);
1687 free(commits);
1688 goto cleanup;
1689 }
1690
1691 clear_packing_data(&pdata);
1692 free(commits);
1693 }
1694 /*
1695 * NOTE: Do not use ctx.entries beyond this point, since it might
1696 * have been freed in the previous if block.
1697 */
1698
1699 if (ctx.m)
1700 close_object_store(the_repository->objects);
1701
1702 if (commit_lock_file(&lk) < 0)
1703 die_errno(_("could not write multi-pack-index"));
1704
1705 clear_midx_files_ext(object_dir, ".bitmap", midx_hash);
1706 clear_midx_files_ext(object_dir, ".rev", midx_hash);
1707
1708 cleanup:
1709 for (i = 0; i < ctx.nr; i++) {
1710 if (ctx.info[i].p) {
1711 close_pack(ctx.info[i].p);
1712 free(ctx.info[i].p);
1713 }
1714 free(ctx.info[i].pack_name);
1715 }
1716
1717 free(ctx.info);
1718 free(ctx.entries);
1719 free(ctx.pack_perm);
1720 free(ctx.pack_order);
1721 strbuf_release(&midx_name);
1722
1723 trace2_region_leave("midx", "write_midx_internal", the_repository);
1724
1725 return result;
1726 }
1727
1728 int write_midx_file(const char *object_dir,
1729 const char *preferred_pack_name,
1730 const char *refs_snapshot,
1731 unsigned flags)
1732 {
1733 return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name,
1734 refs_snapshot, flags);
1735 }
1736
1737 int write_midx_file_only(const char *object_dir,
1738 struct string_list *packs_to_include,
1739 const char *preferred_pack_name,
1740 const char *refs_snapshot,
1741 unsigned flags)
1742 {
1743 return write_midx_internal(object_dir, packs_to_include, NULL,
1744 preferred_pack_name, refs_snapshot, flags);
1745 }
1746
1747 struct clear_midx_data {
1748 char *keep;
1749 const char *ext;
1750 };
1751
1752 static void clear_midx_file_ext(const char *full_path, size_t full_path_len UNUSED,
1753 const char *file_name, void *_data)
1754 {
1755 struct clear_midx_data *data = _data;
1756
1757 if (!(starts_with(file_name, "multi-pack-index-") &&
1758 ends_with(file_name, data->ext)))
1759 return;
1760 if (data->keep && !strcmp(data->keep, file_name))
1761 return;
1762
1763 if (unlink(full_path))
1764 die_errno(_("failed to remove %s"), full_path);
1765 }
1766
1767 static void clear_midx_files_ext(const char *object_dir, const char *ext,
1768 unsigned char *keep_hash)
1769 {
1770 struct clear_midx_data data;
1771 memset(&data, 0, sizeof(struct clear_midx_data));
1772
1773 if (keep_hash)
1774 data.keep = xstrfmt("multi-pack-index-%s%s",
1775 hash_to_hex(keep_hash), ext);
1776 data.ext = ext;
1777
1778 for_each_file_in_pack_dir(object_dir,
1779 clear_midx_file_ext,
1780 &data);
1781
1782 free(data.keep);
1783 }
1784
1785 void clear_midx_file(struct repository *r)
1786 {
1787 struct strbuf midx = STRBUF_INIT;
1788
1789 get_midx_filename(&midx, r->objects->odb->path);
1790
1791 if (r->objects && r->objects->multi_pack_index) {
1792 close_midx(r->objects->multi_pack_index);
1793 r->objects->multi_pack_index = NULL;
1794 }
1795
1796 if (remove_path(midx.buf))
1797 die(_("failed to clear multi-pack-index at %s"), midx.buf);
1798
1799 clear_midx_files_ext(r->objects->odb->path, ".bitmap", NULL);
1800 clear_midx_files_ext(r->objects->odb->path, ".rev", NULL);
1801
1802 strbuf_release(&midx);
1803 }
1804
1805 static int verify_midx_error;
1806
1807 __attribute__((format (printf, 1, 2)))
1808 static void midx_report(const char *fmt, ...)
1809 {
1810 va_list ap;
1811 verify_midx_error = 1;
1812 va_start(ap, fmt);
1813 vfprintf(stderr, fmt, ap);
1814 fprintf(stderr, "\n");
1815 va_end(ap);
1816 }
1817
1818 struct pair_pos_vs_id
1819 {
1820 uint32_t pos;
1821 uint32_t pack_int_id;
1822 };
1823
1824 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1825 {
1826 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1827 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1828
1829 return b->pack_int_id - a->pack_int_id;
1830 }
1831
1832 /*
1833 * Limit calls to display_progress() for performance reasons.
1834 * The interval here was arbitrarily chosen.
1835 */
1836 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1837 #define midx_display_sparse_progress(progress, n) \
1838 do { \
1839 uint64_t _n = (n); \
1840 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1841 display_progress(progress, _n); \
1842 } while (0)
1843
1844 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1845 {
1846 struct pair_pos_vs_id *pairs = NULL;
1847 uint32_t i;
1848 struct progress *progress = NULL;
1849 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1850 verify_midx_error = 0;
1851
1852 if (!m) {
1853 int result = 0;
1854 struct stat sb;
1855 struct strbuf filename = STRBUF_INIT;
1856
1857 get_midx_filename(&filename, object_dir);
1858
1859 if (!stat(filename.buf, &sb)) {
1860 error(_("multi-pack-index file exists, but failed to parse"));
1861 result = 1;
1862 }
1863 strbuf_release(&filename);
1864 return result;
1865 }
1866
1867 if (!midx_checksum_valid(m))
1868 midx_report(_("incorrect checksum"));
1869
1870 if (flags & MIDX_PROGRESS)
1871 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1872 m->num_packs);
1873 for (i = 0; i < m->num_packs; i++) {
1874 if (prepare_midx_pack(r, m, i))
1875 midx_report("failed to load pack in position %d", i);
1876
1877 display_progress(progress, i + 1);
1878 }
1879 stop_progress(&progress);
1880
1881 if (m->num_objects == 0) {
1882 midx_report(_("the midx contains no oid"));
1883 /*
1884 * Remaining tests assume that we have objects, so we can
1885 * return here.
1886 */
1887 goto cleanup;
1888 }
1889
1890 if (flags & MIDX_PROGRESS)
1891 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1892 m->num_objects - 1);
1893 for (i = 0; i < m->num_objects - 1; i++) {
1894 struct object_id oid1, oid2;
1895
1896 nth_midxed_object_oid(&oid1, m, i);
1897 nth_midxed_object_oid(&oid2, m, i + 1);
1898
1899 if (oidcmp(&oid1, &oid2) >= 0)
1900 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1901 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1902
1903 midx_display_sparse_progress(progress, i + 1);
1904 }
1905 stop_progress(&progress);
1906
1907 /*
1908 * Create an array mapping each object to its packfile id. Sort it
1909 * to group the objects by packfile. Use this permutation to visit
1910 * each of the objects and only require 1 packfile to be open at a
1911 * time.
1912 */
1913 ALLOC_ARRAY(pairs, m->num_objects);
1914 for (i = 0; i < m->num_objects; i++) {
1915 pairs[i].pos = i;
1916 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1917 }
1918
1919 if (flags & MIDX_PROGRESS)
1920 progress = start_sparse_progress(_("Sorting objects by packfile"),
1921 m->num_objects);
1922 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1923 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1924 stop_progress(&progress);
1925
1926 if (flags & MIDX_PROGRESS)
1927 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1928 for (i = 0; i < m->num_objects; i++) {
1929 struct object_id oid;
1930 struct pack_entry e;
1931 off_t m_offset, p_offset;
1932
1933 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1934 m->packs[pairs[i-1].pack_int_id])
1935 {
1936 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1937 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1938 }
1939
1940 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1941
1942 if (!fill_midx_entry(r, &oid, &e, m)) {
1943 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1944 pairs[i].pos, oid_to_hex(&oid));
1945 continue;
1946 }
1947
1948 if (open_pack_index(e.p)) {
1949 midx_report(_("failed to load pack-index for packfile %s"),
1950 e.p->pack_name);
1951 break;
1952 }
1953
1954 m_offset = e.offset;
1955 p_offset = find_pack_entry_one(oid.hash, e.p);
1956
1957 if (m_offset != p_offset)
1958 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1959 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1960
1961 midx_display_sparse_progress(progress, i + 1);
1962 }
1963 stop_progress(&progress);
1964
1965 cleanup:
1966 free(pairs);
1967 close_midx(m);
1968
1969 return verify_midx_error;
1970 }
1971
1972 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1973 {
1974 uint32_t i, *count, result = 0;
1975 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1976 struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir);
1977 struct progress *progress = NULL;
1978
1979 if (!m)
1980 return 0;
1981
1982 CALLOC_ARRAY(count, m->num_packs);
1983
1984 if (flags & MIDX_PROGRESS)
1985 progress = start_delayed_progress(_("Counting referenced objects"),
1986 m->num_objects);
1987 for (i = 0; i < m->num_objects; i++) {
1988 int pack_int_id = nth_midxed_pack_int_id(m, i);
1989 count[pack_int_id]++;
1990 display_progress(progress, i + 1);
1991 }
1992 stop_progress(&progress);
1993
1994 if (flags & MIDX_PROGRESS)
1995 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1996 m->num_packs);
1997 for (i = 0; i < m->num_packs; i++) {
1998 char *pack_name;
1999 display_progress(progress, i + 1);
2000
2001 if (count[i])
2002 continue;
2003
2004 if (prepare_midx_pack(r, m, i))
2005 continue;
2006
2007 if (m->packs[i]->pack_keep || m->packs[i]->is_cruft)
2008 continue;
2009
2010 pack_name = xstrdup(m->packs[i]->pack_name);
2011 close_pack(m->packs[i]);
2012
2013 string_list_insert(&packs_to_drop, m->pack_names[i]);
2014 unlink_pack_path(pack_name, 0);
2015 free(pack_name);
2016 }
2017 stop_progress(&progress);
2018
2019 free(count);
2020
2021 if (packs_to_drop.nr)
2022 result = write_midx_internal(object_dir, NULL, &packs_to_drop, NULL, NULL, flags);
2023
2024 string_list_clear(&packs_to_drop, 0);
2025
2026 return result;
2027 }
2028
2029 struct repack_info {
2030 timestamp_t mtime;
2031 uint32_t referenced_objects;
2032 uint32_t pack_int_id;
2033 };
2034
2035 static int compare_by_mtime(const void *a_, const void *b_)
2036 {
2037 const struct repack_info *a, *b;
2038
2039 a = (const struct repack_info *)a_;
2040 b = (const struct repack_info *)b_;
2041
2042 if (a->mtime < b->mtime)
2043 return -1;
2044 if (a->mtime > b->mtime)
2045 return 1;
2046 return 0;
2047 }
2048
2049 static int fill_included_packs_all(struct repository *r,
2050 struct multi_pack_index *m,
2051 unsigned char *include_pack)
2052 {
2053 uint32_t i, count = 0;
2054 int pack_kept_objects = 0;
2055
2056 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
2057
2058 for (i = 0; i < m->num_packs; i++) {
2059 if (prepare_midx_pack(r, m, i))
2060 continue;
2061 if (!pack_kept_objects && m->packs[i]->pack_keep)
2062 continue;
2063 if (m->packs[i]->is_cruft)
2064 continue;
2065
2066 include_pack[i] = 1;
2067 count++;
2068 }
2069
2070 return count < 2;
2071 }
2072
2073 static int fill_included_packs_batch(struct repository *r,
2074 struct multi_pack_index *m,
2075 unsigned char *include_pack,
2076 size_t batch_size)
2077 {
2078 uint32_t i, packs_to_repack;
2079 size_t total_size;
2080 struct repack_info *pack_info;
2081 int pack_kept_objects = 0;
2082
2083 CALLOC_ARRAY(pack_info, m->num_packs);
2084
2085 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
2086
2087 for (i = 0; i < m->num_packs; i++) {
2088 pack_info[i].pack_int_id = i;
2089
2090 if (prepare_midx_pack(r, m, i))
2091 continue;
2092
2093 pack_info[i].mtime = m->packs[i]->mtime;
2094 }
2095
2096 for (i = 0; i < m->num_objects; i++) {
2097 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
2098 pack_info[pack_int_id].referenced_objects++;
2099 }
2100
2101 QSORT(pack_info, m->num_packs, compare_by_mtime);
2102
2103 total_size = 0;
2104 packs_to_repack = 0;
2105 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
2106 int pack_int_id = pack_info[i].pack_int_id;
2107 struct packed_git *p = m->packs[pack_int_id];
2108 size_t expected_size;
2109
2110 if (!p)
2111 continue;
2112 if (!pack_kept_objects && p->pack_keep)
2113 continue;
2114 if (p->is_cruft)
2115 continue;
2116 if (open_pack_index(p) || !p->num_objects)
2117 continue;
2118
2119 expected_size = st_mult(p->pack_size,
2120 pack_info[i].referenced_objects);
2121 expected_size /= p->num_objects;
2122
2123 if (expected_size >= batch_size)
2124 continue;
2125
2126 packs_to_repack++;
2127 total_size += expected_size;
2128 include_pack[pack_int_id] = 1;
2129 }
2130
2131 free(pack_info);
2132
2133 if (packs_to_repack < 2)
2134 return 1;
2135
2136 return 0;
2137 }
2138
2139 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
2140 {
2141 int result = 0;
2142 uint32_t i;
2143 unsigned char *include_pack;
2144 struct child_process cmd = CHILD_PROCESS_INIT;
2145 FILE *cmd_in;
2146 struct strbuf base_name = STRBUF_INIT;
2147 struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir);
2148
2149 /*
2150 * When updating the default for these configuration
2151 * variables in builtin/repack.c, these must be adjusted
2152 * to match.
2153 */
2154 int delta_base_offset = 1;
2155 int use_delta_islands = 0;
2156
2157 if (!m)
2158 return 0;
2159
2160 CALLOC_ARRAY(include_pack, m->num_packs);
2161
2162 if (batch_size) {
2163 if (fill_included_packs_batch(r, m, include_pack, batch_size))
2164 goto cleanup;
2165 } else if (fill_included_packs_all(r, m, include_pack))
2166 goto cleanup;
2167
2168 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
2169 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
2170
2171 strvec_push(&cmd.args, "pack-objects");
2172
2173 strbuf_addstr(&base_name, object_dir);
2174 strbuf_addstr(&base_name, "/pack/pack");
2175 strvec_push(&cmd.args, base_name.buf);
2176
2177 if (delta_base_offset)
2178 strvec_push(&cmd.args, "--delta-base-offset");
2179 if (use_delta_islands)
2180 strvec_push(&cmd.args, "--delta-islands");
2181
2182 if (flags & MIDX_PROGRESS)
2183 strvec_push(&cmd.args, "--progress");
2184 else
2185 strvec_push(&cmd.args, "-q");
2186
2187 strbuf_release(&base_name);
2188
2189 cmd.git_cmd = 1;
2190 cmd.in = cmd.out = -1;
2191
2192 if (start_command(&cmd)) {
2193 error(_("could not start pack-objects"));
2194 result = 1;
2195 goto cleanup;
2196 }
2197
2198 cmd_in = xfdopen(cmd.in, "w");
2199
2200 for (i = 0; i < m->num_objects; i++) {
2201 struct object_id oid;
2202 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
2203
2204 if (!include_pack[pack_int_id])
2205 continue;
2206
2207 nth_midxed_object_oid(&oid, m, i);
2208 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
2209 }
2210 fclose(cmd_in);
2211
2212 if (finish_command(&cmd)) {
2213 error(_("could not finish pack-objects"));
2214 result = 1;
2215 goto cleanup;
2216 }
2217
2218 result = write_midx_internal(object_dir, NULL, NULL, NULL, NULL, flags);
2219
2220 cleanup:
2221 free(include_pack);
2222 return result;
2223 }
Alternate source for your git.git needs