?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
midx: implement `BTMP` chunk
[git/gitster.git] / midx.c
blobde25612b0c863d8544d04b943d50e55d4ab65f20
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_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
432 {
433 uint32_t first = 0, last = m->num_packs;
434
435 while (first < last) {
436 uint32_t mid = first + (last - first) / 2;
437 const char *current;
438 int cmp;
439
440 current = m->pack_names[mid];
441 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
442 if (!cmp)
443 return 1;
444 if (cmp > 0) {
445 first = mid + 1;
446 continue;
447 }
448 last = mid;
449 }
450
451 return 0;
452 }
453
454 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
455 {
456 struct multi_pack_index *m;
457 struct multi_pack_index *m_search;
458
459 prepare_repo_settings(r);
460 if (!r->settings.core_multi_pack_index)
461 return 0;
462
463 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
464 if (!strcmp(object_dir, m_search->object_dir))
465 return 1;
466
467 m = load_multi_pack_index(object_dir, local);
468
469 if (m) {
470 struct multi_pack_index *mp = r->objects->multi_pack_index;
471 if (mp) {
472 m->next = mp->next;
473 mp->next = m;
474 } else
475 r->objects->multi_pack_index = m;
476 return 1;
477 }
478
479 return 0;
480 }
481
482 static size_t write_midx_header(struct hashfile *f,
483 unsigned char num_chunks,
484 uint32_t num_packs)
485 {
486 hashwrite_be32(f, MIDX_SIGNATURE);
487 hashwrite_u8(f, MIDX_VERSION);
488 hashwrite_u8(f, oid_version(the_hash_algo));
489 hashwrite_u8(f, num_chunks);
490 hashwrite_u8(f, 0); /* unused */
491 hashwrite_be32(f, num_packs);
492
493 return MIDX_HEADER_SIZE;
494 }
495
496 #define BITMAP_POS_UNKNOWN (~((uint32_t)0))
497
498 struct pack_info {
499 uint32_t orig_pack_int_id;
500 char *pack_name;
501 struct packed_git *p;
502
503 uint32_t bitmap_pos;
504 uint32_t bitmap_nr;
505
506 unsigned expired : 1;
507 };
508
509 static void fill_pack_info(struct pack_info *info,
510 struct packed_git *p, const char *pack_name,
511 uint32_t orig_pack_int_id)
512 {
513 memset(info, 0, sizeof(struct pack_info));
514
515 info->orig_pack_int_id = orig_pack_int_id;
516 info->pack_name = xstrdup(pack_name);
517 info->p = p;
518 info->bitmap_pos = BITMAP_POS_UNKNOWN;
519 }
520
521 static int pack_info_compare(const void *_a, const void *_b)
522 {
523 struct pack_info *a = (struct pack_info *)_a;
524 struct pack_info *b = (struct pack_info *)_b;
525 return strcmp(a->pack_name, b->pack_name);
526 }
527
528 static int idx_or_pack_name_cmp(const void *_va, const void *_vb)
529 {
530 const char *pack_name = _va;
531 const struct pack_info *compar = _vb;
532
533 return cmp_idx_or_pack_name(pack_name, compar->pack_name);
534 }
535
536 struct write_midx_context {
537 struct pack_info *info;
538 size_t nr;
539 size_t alloc;
540 struct multi_pack_index *m;
541 struct progress *progress;
542 unsigned pack_paths_checked;
543
544 struct pack_midx_entry *entries;
545 size_t entries_nr;
546
547 uint32_t *pack_perm;
548 uint32_t *pack_order;
549 unsigned large_offsets_needed:1;
550 uint32_t num_large_offsets;
551
552 int preferred_pack_idx;
553
554 struct string_list *to_include;
555 };
556
557 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
558 const char *file_name, void *data)
559 {
560 struct write_midx_context *ctx = data;
561 struct packed_git *p;
562
563 if (ends_with(file_name, ".idx")) {
564 display_progress(ctx->progress, ++ctx->pack_paths_checked);
565 /*
566 * Note that at most one of ctx->m and ctx->to_include are set,
567 * so we are testing midx_contains_pack() and
568 * string_list_has_string() independently (guarded by the
569 * appropriate NULL checks).
570 *
571 * We could support passing to_include while reusing an existing
572 * MIDX, but don't currently since the reuse process drags
573 * forward all packs from an existing MIDX (without checking
574 * whether or not they appear in the to_include list).
575 *
576 * If we added support for that, these next two conditional
577 * should be performed independently (likely checking
578 * to_include before the existing MIDX).
579 */
580 if (ctx->m && midx_contains_pack(ctx->m, file_name))
581 return;
582 else if (ctx->to_include &&
583 !string_list_has_string(ctx->to_include, file_name))
584 return;
585
586 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
587
588 p = add_packed_git(full_path, full_path_len, 0);
589 if (!p) {
590 warning(_("failed to add packfile '%s'"),
591 full_path);
592 return;
593 }
594
595 if (open_pack_index(p)) {
596 warning(_("failed to open pack-index '%s'"),
597 full_path);
598 close_pack(p);
599 free(p);
600 return;
601 }
602
603 fill_pack_info(&ctx->info[ctx->nr], p, file_name, ctx->nr);
604 ctx->nr++;
605 }
606 }
607
608 struct pack_midx_entry {
609 struct object_id oid;
610 uint32_t pack_int_id;
611 time_t pack_mtime;
612 uint64_t offset;
613 unsigned preferred : 1;
614 };
615
616 static int midx_oid_compare(const void *_a, const void *_b)
617 {
618 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
619 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
620 int cmp = oidcmp(&a->oid, &b->oid);
621
622 if (cmp)
623 return cmp;
624
625 /* Sort objects in a preferred pack first when multiple copies exist. */
626 if (a->preferred > b->preferred)
627 return -1;
628 if (a->preferred < b->preferred)
629 return 1;
630
631 if (a->pack_mtime > b->pack_mtime)
632 return -1;
633 else if (a->pack_mtime < b->pack_mtime)
634 return 1;
635
636 return a->pack_int_id - b->pack_int_id;
637 }
638
639 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
640 struct pack_midx_entry *e,
641 uint32_t pos)
642 {
643 if (pos >= m->num_objects)
644 return 1;
645
646 nth_midxed_object_oid(&e->oid, m, pos);
647 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
648 e->offset = nth_midxed_offset(m, pos);
649
650 /* consider objects in midx to be from "old" packs */
651 e->pack_mtime = 0;
652 return 0;
653 }
654
655 static void fill_pack_entry(uint32_t pack_int_id,
656 struct packed_git *p,
657 uint32_t cur_object,
658 struct pack_midx_entry *entry,
659 int preferred)
660 {
661 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
662 die(_("failed to locate object %d in packfile"), cur_object);
663
664 entry->pack_int_id = pack_int_id;
665 entry->pack_mtime = p->mtime;
666
667 entry->offset = nth_packed_object_offset(p, cur_object);
668 entry->preferred = !!preferred;
669 }
670
671 struct midx_fanout {
672 struct pack_midx_entry *entries;
673 size_t nr, alloc;
674 };
675
676 static void midx_fanout_grow(struct midx_fanout *fanout, size_t nr)
677 {
678 if (nr < fanout->nr)
679 BUG("negative growth in midx_fanout_grow() (%"PRIuMAX" < %"PRIuMAX")",
680 (uintmax_t)nr, (uintmax_t)fanout->nr);
681 ALLOC_GROW(fanout->entries, nr, fanout->alloc);
682 }
683
684 static void midx_fanout_sort(struct midx_fanout *fanout)
685 {
686 QSORT(fanout->entries, fanout->nr, midx_oid_compare);
687 }
688
689 static void midx_fanout_add_midx_fanout(struct midx_fanout *fanout,
690 struct multi_pack_index *m,
691 uint32_t cur_fanout,
692 int preferred_pack)
693 {
694 uint32_t start = 0, end;
695 uint32_t cur_object;
696
697 if (cur_fanout)
698 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
699 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
700
701 for (cur_object = start; cur_object < end; cur_object++) {
702 if ((preferred_pack > -1) &&
703 (preferred_pack == nth_midxed_pack_int_id(m, cur_object))) {
704 /*
705 * Objects from preferred packs are added
706 * separately.
707 */
708 continue;
709 }
710
711 midx_fanout_grow(fanout, fanout->nr + 1);
712 nth_midxed_pack_midx_entry(m,
713 &fanout->entries[fanout->nr],
714 cur_object);
715 fanout->entries[fanout->nr].preferred = 0;
716 fanout->nr++;
717 }
718 }
719
720 static void midx_fanout_add_pack_fanout(struct midx_fanout *fanout,
721 struct pack_info *info,
722 uint32_t cur_pack,
723 int preferred,
724 uint32_t cur_fanout)
725 {
726 struct packed_git *pack = info[cur_pack].p;
727 uint32_t start = 0, end;
728 uint32_t cur_object;
729
730 if (cur_fanout)
731 start = get_pack_fanout(pack, cur_fanout - 1);
732 end = get_pack_fanout(pack, cur_fanout);
733
734 for (cur_object = start; cur_object < end; cur_object++) {
735 midx_fanout_grow(fanout, fanout->nr + 1);
736 fill_pack_entry(cur_pack,
737 info[cur_pack].p,
738 cur_object,
739 &fanout->entries[fanout->nr],
740 preferred);
741 fanout->nr++;
742 }
743 }
744
745 /*
746 * It is possible to artificially get into a state where there are many
747 * duplicate copies of objects. That can create high memory pressure if
748 * we are to create a list of all objects before de-duplication. To reduce
749 * this memory pressure without a significant performance drop, automatically
750 * group objects by the first byte of their object id. Use the IDX fanout
751 * tables to group the data, copy to a local array, then sort.
752 *
753 * Copy only the de-duplicated entries (selected by most-recent modified time
754 * of a packfile containing the object).
755 */
756 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
757 struct pack_info *info,
758 uint32_t nr_packs,
759 size_t *nr_objects,
760 int preferred_pack)
761 {
762 uint32_t cur_fanout, cur_pack, cur_object;
763 size_t alloc_objects, total_objects = 0;
764 struct midx_fanout fanout = { 0 };
765 struct pack_midx_entry *deduplicated_entries = NULL;
766 uint32_t start_pack = m ? m->num_packs : 0;
767
768 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
769 total_objects = st_add(total_objects,
770 info[cur_pack].p->num_objects);
771
772 /*
773 * As we de-duplicate by fanout value, we expect the fanout
774 * slices to be evenly distributed, with some noise. Hence,
775 * allocate slightly more than one 256th.
776 */
777 alloc_objects = fanout.alloc = total_objects > 3200 ? total_objects / 200 : 16;
778
779 ALLOC_ARRAY(fanout.entries, fanout.alloc);
780 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
781 *nr_objects = 0;
782
783 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
784 fanout.nr = 0;
785
786 if (m)
787 midx_fanout_add_midx_fanout(&fanout, m, cur_fanout,
788 preferred_pack);
789
790 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
791 int preferred = cur_pack == preferred_pack;
792 midx_fanout_add_pack_fanout(&fanout,
793 info, cur_pack,
794 preferred, cur_fanout);
795 }
796
797 if (-1 < preferred_pack && preferred_pack < start_pack)
798 midx_fanout_add_pack_fanout(&fanout, info,
799 preferred_pack, 1,
800 cur_fanout);
801
802 midx_fanout_sort(&fanout);
803
804 /*
805 * The batch is now sorted by OID and then mtime (descending).
806 * Take only the first duplicate.
807 */
808 for (cur_object = 0; cur_object < fanout.nr; cur_object++) {
809 if (cur_object && oideq(&fanout.entries[cur_object - 1].oid,
810 &fanout.entries[cur_object].oid))
811 continue;
812
813 ALLOC_GROW(deduplicated_entries, st_add(*nr_objects, 1),
814 alloc_objects);
815 memcpy(&deduplicated_entries[*nr_objects],
816 &fanout.entries[cur_object],
817 sizeof(struct pack_midx_entry));
818 (*nr_objects)++;
819 }
820 }
821
822 free(fanout.entries);
823 return deduplicated_entries;
824 }
825
826 static int write_midx_pack_names(struct hashfile *f, void *data)
827 {
828 struct write_midx_context *ctx = data;
829 uint32_t i;
830 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
831 size_t written = 0;
832
833 for (i = 0; i < ctx->nr; i++) {
834 size_t writelen;
835
836 if (ctx->info[i].expired)
837 continue;
838
839 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
840 BUG("incorrect pack-file order: %s before %s",
841 ctx->info[i - 1].pack_name,
842 ctx->info[i].pack_name);
843
844 writelen = strlen(ctx->info[i].pack_name) + 1;
845 hashwrite(f, ctx->info[i].pack_name, writelen);
846 written += writelen;
847 }
848
849 /* add padding to be aligned */
850 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
851 if (i < MIDX_CHUNK_ALIGNMENT) {
852 memset(padding, 0, sizeof(padding));
853 hashwrite(f, padding, i);
854 }
855
856 return 0;
857 }
858
859 static int write_midx_bitmapped_packs(struct hashfile *f, void *data)
860 {
861 struct write_midx_context *ctx = data;
862 size_t i;
863
864 for (i = 0; i < ctx->nr; i++) {
865 struct pack_info *pack = &ctx->info[i];
866 if (pack->expired)
867 continue;
868
869 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN && pack->bitmap_nr)
870 BUG("pack '%s' has no bitmap position, but has %d bitmapped object(s)",
871 pack->pack_name, pack->bitmap_nr);
872
873 hashwrite_be32(f, pack->bitmap_pos);
874 hashwrite_be32(f, pack->bitmap_nr);
875 }
876 return 0;
877 }
878
879 static int write_midx_oid_fanout(struct hashfile *f,
880 void *data)
881 {
882 struct write_midx_context *ctx = data;
883 struct pack_midx_entry *list = ctx->entries;
884 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
885 uint32_t count = 0;
886 uint32_t i;
887
888 /*
889 * Write the first-level table (the list is sorted,
890 * but we use a 256-entry lookup to be able to avoid
891 * having to do eight extra binary search iterations).
892 */
893 for (i = 0; i < 256; i++) {
894 struct pack_midx_entry *next = list;
895
896 while (next < last && next->oid.hash[0] == i) {
897 count++;
898 next++;
899 }
900
901 hashwrite_be32(f, count);
902 list = next;
903 }
904
905 return 0;
906 }
907
908 static int write_midx_oid_lookup(struct hashfile *f,
909 void *data)
910 {
911 struct write_midx_context *ctx = data;
912 unsigned char hash_len = the_hash_algo->rawsz;
913 struct pack_midx_entry *list = ctx->entries;
914 uint32_t i;
915
916 for (i = 0; i < ctx->entries_nr; i++) {
917 struct pack_midx_entry *obj = list++;
918
919 if (i < ctx->entries_nr - 1) {
920 struct pack_midx_entry *next = list;
921 if (oidcmp(&obj->oid, &next->oid) >= 0)
922 BUG("OIDs not in order: %s >= %s",
923 oid_to_hex(&obj->oid),
924 oid_to_hex(&next->oid));
925 }
926
927 hashwrite(f, obj->oid.hash, (int)hash_len);
928 }
929
930 return 0;
931 }
932
933 static int write_midx_object_offsets(struct hashfile *f,
934 void *data)
935 {
936 struct write_midx_context *ctx = data;
937 struct pack_midx_entry *list = ctx->entries;
938 uint32_t i, nr_large_offset = 0;
939
940 for (i = 0; i < ctx->entries_nr; i++) {
941 struct pack_midx_entry *obj = list++;
942
943 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
944 BUG("object %s is in an expired pack with int-id %d",
945 oid_to_hex(&obj->oid),
946 obj->pack_int_id);
947
948 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
949
950 if (ctx->large_offsets_needed && obj->offset >> 31)
951 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
952 else if (!ctx->large_offsets_needed && obj->offset >> 32)
953 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
954 oid_to_hex(&obj->oid),
955 obj->offset);
956 else
957 hashwrite_be32(f, (uint32_t)obj->offset);
958 }
959
960 return 0;
961 }
962
963 static int write_midx_large_offsets(struct hashfile *f,
964 void *data)
965 {
966 struct write_midx_context *ctx = data;
967 struct pack_midx_entry *list = ctx->entries;
968 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
969 uint32_t nr_large_offset = ctx->num_large_offsets;
970
971 while (nr_large_offset) {
972 struct pack_midx_entry *obj;
973 uint64_t offset;
974
975 if (list >= end)
976 BUG("too many large-offset objects");
977
978 obj = list++;
979 offset = obj->offset;
980
981 if (!(offset >> 31))
982 continue;
983
984 hashwrite_be64(f, offset);
985
986 nr_large_offset--;
987 }
988
989 return 0;
990 }
991
992 static int write_midx_revindex(struct hashfile *f,
993 void *data)
994 {
995 struct write_midx_context *ctx = data;
996 uint32_t i;
997
998 for (i = 0; i < ctx->entries_nr; i++)
999 hashwrite_be32(f, ctx->pack_order[i]);
1000
1001 return 0;
1002 }
1003
1004 struct midx_pack_order_data {
1005 uint32_t nr;
1006 uint32_t pack;
1007 off_t offset;
1008 };
1009
1010 static int midx_pack_order_cmp(const void *va, const void *vb)
1011 {
1012 const struct midx_pack_order_data *a = va, *b = vb;
1013 if (a->pack < b->pack)
1014 return -1;
1015 else if (a->pack > b->pack)
1016 return 1;
1017 else if (a->offset < b->offset)
1018 return -1;
1019 else if (a->offset > b->offset)
1020 return 1;
1021 else
1022 return 0;
1023 }
1024
1025 static uint32_t *midx_pack_order(struct write_midx_context *ctx)
1026 {
1027 struct midx_pack_order_data *data;
1028 uint32_t *pack_order;
1029 uint32_t i;
1030
1031 trace2_region_enter("midx", "midx_pack_order", the_repository);
1032
1033 ALLOC_ARRAY(data, ctx->entries_nr);
1034 for (i = 0; i < ctx->entries_nr; i++) {
1035 struct pack_midx_entry *e = &ctx->entries[i];
1036 data[i].nr = i;
1037 data[i].pack = ctx->pack_perm[e->pack_int_id];
1038 if (!e->preferred)
1039 data[i].pack |= (1U << 31);
1040 data[i].offset = e->offset;
1041 }
1042
1043 QSORT(data, ctx->entries_nr, midx_pack_order_cmp);
1044
1045 ALLOC_ARRAY(pack_order, ctx->entries_nr);
1046 for (i = 0; i < ctx->entries_nr; i++) {
1047 struct pack_midx_entry *e = &ctx->entries[data[i].nr];
1048 struct pack_info *pack = &ctx->info[ctx->pack_perm[e->pack_int_id]];
1049 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN)
1050 pack->bitmap_pos = i;
1051 pack->bitmap_nr++;
1052 pack_order[i] = data[i].nr;
1053 }
1054 for (i = 0; i < ctx->nr; i++) {
1055 struct pack_info *pack = &ctx->info[ctx->pack_perm[i]];
1056 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN)
1057 pack->bitmap_pos = 0;
1058 }
1059 free(data);
1060
1061 trace2_region_leave("midx", "midx_pack_order", the_repository);
1062
1063 return pack_order;
1064 }
1065
1066 static void write_midx_reverse_index(char *midx_name, unsigned char *midx_hash,
1067 struct write_midx_context *ctx)
1068 {
1069 struct strbuf buf = STRBUF_INIT;
1070 const char *tmp_file;
1071
1072 trace2_region_enter("midx", "write_midx_reverse_index", the_repository);
1073
1074 strbuf_addf(&buf, "%s-%s.rev", midx_name, hash_to_hex(midx_hash));
1075
1076 tmp_file = write_rev_file_order(NULL, ctx->pack_order, ctx->entries_nr,
1077 midx_hash, WRITE_REV);
1078
1079 if (finalize_object_file(tmp_file, buf.buf))
1080 die(_("cannot store reverse index file"));
1081
1082 strbuf_release(&buf);
1083
1084 trace2_region_leave("midx", "write_midx_reverse_index", the_repository);
1085 }
1086
1087 static void clear_midx_files_ext(const char *object_dir, const char *ext,
1088 unsigned char *keep_hash);
1089
1090 static int midx_checksum_valid(struct multi_pack_index *m)
1091 {
1092 return hashfile_checksum_valid(m->data, m->data_len);
1093 }
1094
1095 static void prepare_midx_packing_data(struct packing_data *pdata,
1096 struct write_midx_context *ctx)
1097 {
1098 uint32_t i;
1099
1100 trace2_region_enter("midx", "prepare_midx_packing_data", the_repository);
1101
1102 memset(pdata, 0, sizeof(struct packing_data));
1103 prepare_packing_data(the_repository, pdata);
1104
1105 for (i = 0; i < ctx->entries_nr; i++) {
1106 struct pack_midx_entry *from = &ctx->entries[ctx->pack_order[i]];
1107 struct object_entry *to = packlist_alloc(pdata, &from->oid);
1108
1109 oe_set_in_pack(pdata, to,
1110 ctx->info[ctx->pack_perm[from->pack_int_id]].p);
1111 }
1112
1113 trace2_region_leave("midx", "prepare_midx_packing_data", the_repository);
1114 }
1115
1116 static int add_ref_to_pending(const char *refname,
1117 const struct object_id *oid,
1118 int flag, void *cb_data)
1119 {
1120 struct rev_info *revs = (struct rev_info*)cb_data;
1121 struct object_id peeled;
1122 struct object *object;
1123
1124 if ((flag & REF_ISSYMREF) && (flag & REF_ISBROKEN)) {
1125 warning("symbolic ref is dangling: %s", refname);
1126 return 0;
1127 }
1128
1129 if (!peel_iterated_oid(oid, &peeled))
1130 oid = &peeled;
1131
1132 object = parse_object_or_die(oid, refname);
1133 if (object->type != OBJ_COMMIT)
1134 return 0;
1135
1136 add_pending_object(revs, object, "");
1137 if (bitmap_is_preferred_refname(revs->repo, refname))
1138 object->flags |= NEEDS_BITMAP;
1139 return 0;
1140 }
1141
1142 struct bitmap_commit_cb {
1143 struct commit **commits;
1144 size_t commits_nr, commits_alloc;
1145
1146 struct write_midx_context *ctx;
1147 };
1148
1149 static const struct object_id *bitmap_oid_access(size_t index,
1150 const void *_entries)
1151 {
1152 const struct pack_midx_entry *entries = _entries;
1153 return &entries[index].oid;
1154 }
1155
1156 static void bitmap_show_commit(struct commit *commit, void *_data)
1157 {
1158 struct bitmap_commit_cb *data = _data;
1159 int pos = oid_pos(&commit->object.oid, data->ctx->entries,
1160 data->ctx->entries_nr,
1161 bitmap_oid_access);
1162 if (pos < 0)
1163 return;
1164
1165 ALLOC_GROW(data->commits, data->commits_nr + 1, data->commits_alloc);
1166 data->commits[data->commits_nr++] = commit;
1167 }
1168
1169 static int read_refs_snapshot(const char *refs_snapshot,
1170 struct rev_info *revs)
1171 {
1172 struct strbuf buf = STRBUF_INIT;
1173 struct object_id oid;
1174 FILE *f = xfopen(refs_snapshot, "r");
1175
1176 while (strbuf_getline(&buf, f) != EOF) {
1177 struct object *object;
1178 int preferred = 0;
1179 char *hex = buf.buf;
1180 const char *end = NULL;
1181
1182 if (buf.len && *buf.buf == '+') {
1183 preferred = 1;
1184 hex = &buf.buf[1];
1185 }
1186
1187 if (parse_oid_hex(hex, &oid, &end) < 0)
1188 die(_("could not parse line: %s"), buf.buf);
1189 if (*end)
1190 die(_("malformed line: %s"), buf.buf);
1191
1192 object = parse_object_or_die(&oid, NULL);
1193 if (preferred)
1194 object->flags |= NEEDS_BITMAP;
1195
1196 add_pending_object(revs, object, "");
1197 }
1198
1199 fclose(f);
1200 strbuf_release(&buf);
1201 return 0;
1202 }
1203
1204 static struct commit **find_commits_for_midx_bitmap(uint32_t *indexed_commits_nr_p,
1205 const char *refs_snapshot,
1206 struct write_midx_context *ctx)
1207 {
1208 struct rev_info revs;
1209 struct bitmap_commit_cb cb = {0};
1210
1211 trace2_region_enter("midx", "find_commits_for_midx_bitmap",
1212 the_repository);
1213
1214 cb.ctx = ctx;
1215
1216 repo_init_revisions(the_repository, &revs, NULL);
1217 if (refs_snapshot) {
1218 read_refs_snapshot(refs_snapshot, &revs);
1219 } else {
1220 setup_revisions(0, NULL, &revs, NULL);
1221 for_each_ref(add_ref_to_pending, &revs);
1222 }
1223
1224 /*
1225 * Skipping promisor objects here is intentional, since it only excludes
1226 * them from the list of reachable commits that we want to select from
1227 * when computing the selection of MIDX'd commits to receive bitmaps.
1228 *
1229 * Reachability bitmaps do require that their objects be closed under
1230 * reachability, but fetching any objects missing from promisors at this
1231 * point is too late. But, if one of those objects can be reached from
1232 * an another object that is included in the bitmap, then we will
1233 * complain later that we don't have reachability closure (and fail
1234 * appropriately).
1235 */
1236 fetch_if_missing = 0;
1237 revs.exclude_promisor_objects = 1;
1238
1239 if (prepare_revision_walk(&revs))
1240 die(_("revision walk setup failed"));
1241
1242 traverse_commit_list(&revs, bitmap_show_commit, NULL, &cb);
1243 if (indexed_commits_nr_p)
1244 *indexed_commits_nr_p = cb.commits_nr;
1245
1246 release_revisions(&revs);
1247
1248 trace2_region_leave("midx", "find_commits_for_midx_bitmap",
1249 the_repository);
1250
1251 return cb.commits;
1252 }
1253
1254 static int write_midx_bitmap(const char *midx_name,
1255 const unsigned char *midx_hash,
1256 struct packing_data *pdata,
1257 struct commit **commits,
1258 uint32_t commits_nr,
1259 uint32_t *pack_order,
1260 unsigned flags)
1261 {
1262 int ret, i;
1263 uint16_t options = 0;
1264 struct pack_idx_entry **index;
1265 char *bitmap_name = xstrfmt("%s-%s.bitmap", midx_name,
1266 hash_to_hex(midx_hash));
1267
1268 trace2_region_enter("midx", "write_midx_bitmap", the_repository);
1269
1270 if (flags & MIDX_WRITE_BITMAP_HASH_CACHE)
1271 options |= BITMAP_OPT_HASH_CACHE;
1272
1273 if (flags & MIDX_WRITE_BITMAP_LOOKUP_TABLE)
1274 options |= BITMAP_OPT_LOOKUP_TABLE;
1275
1276 /*
1277 * Build the MIDX-order index based on pdata.objects (which is already
1278 * in MIDX order; c.f., 'midx_pack_order_cmp()' for the definition of
1279 * this order).
1280 */
1281 ALLOC_ARRAY(index, pdata->nr_objects);
1282 for (i = 0; i < pdata->nr_objects; i++)
1283 index[i] = &pdata->objects[i].idx;
1284
1285 bitmap_writer_show_progress(flags & MIDX_PROGRESS);
1286 bitmap_writer_build_type_index(pdata, index, pdata->nr_objects);
1287
1288 /*
1289 * bitmap_writer_finish expects objects in lex order, but pack_order
1290 * gives us exactly that. use it directly instead of re-sorting the
1291 * array.
1292 *
1293 * This changes the order of objects in 'index' between
1294 * bitmap_writer_build_type_index and bitmap_writer_finish.
1295 *
1296 * The same re-ordering takes place in the single-pack bitmap code via
1297 * write_idx_file(), which is called by finish_tmp_packfile(), which
1298 * happens between bitmap_writer_build_type_index() and
1299 * bitmap_writer_finish().
1300 */
1301 for (i = 0; i < pdata->nr_objects; i++)
1302 index[pack_order[i]] = &pdata->objects[i].idx;
1303
1304 bitmap_writer_select_commits(commits, commits_nr, -1);
1305 ret = bitmap_writer_build(pdata);
1306 if (ret < 0)
1307 goto cleanup;
1308
1309 bitmap_writer_set_checksum(midx_hash);
1310 bitmap_writer_finish(index, pdata->nr_objects, bitmap_name, options);
1311
1312 cleanup:
1313 free(index);
1314 free(bitmap_name);
1315
1316 trace2_region_leave("midx", "write_midx_bitmap", the_repository);
1317
1318 return ret;
1319 }
1320
1321 static struct multi_pack_index *lookup_multi_pack_index(struct repository *r,
1322 const char *object_dir)
1323 {
1324 struct multi_pack_index *result = NULL;
1325 struct multi_pack_index *cur;
1326 char *obj_dir_real = real_pathdup(object_dir, 1);
1327 struct strbuf cur_path_real = STRBUF_INIT;
1328
1329 /* Ensure the given object_dir is local, or a known alternate. */
1330 find_odb(r, obj_dir_real);
1331
1332 for (cur = get_multi_pack_index(r); cur; cur = cur->next) {
1333 strbuf_realpath(&cur_path_real, cur->object_dir, 1);
1334 if (!strcmp(obj_dir_real, cur_path_real.buf)) {
1335 result = cur;
1336 goto cleanup;
1337 }
1338 }
1339
1340 cleanup:
1341 free(obj_dir_real);
1342 strbuf_release(&cur_path_real);
1343 return result;
1344 }
1345
1346 static int write_midx_internal(const char *object_dir,
1347 struct string_list *packs_to_include,
1348 struct string_list *packs_to_drop,
1349 const char *preferred_pack_name,
1350 const char *refs_snapshot,
1351 unsigned flags)
1352 {
1353 struct strbuf midx_name = STRBUF_INIT;
1354 unsigned char midx_hash[GIT_MAX_RAWSZ];
1355 uint32_t i;
1356 struct hashfile *f = NULL;
1357 struct lock_file lk;
1358 struct write_midx_context ctx = { 0 };
1359 int bitmapped_packs_concat_len = 0;
1360 int pack_name_concat_len = 0;
1361 int dropped_packs = 0;
1362 int result = 0;
1363 struct chunkfile *cf;
1364
1365 trace2_region_enter("midx", "write_midx_internal", the_repository);
1366
1367 get_midx_filename(&midx_name, object_dir);
1368 if (safe_create_leading_directories(midx_name.buf))
1369 die_errno(_("unable to create leading directories of %s"),
1370 midx_name.buf);
1371
1372 if (!packs_to_include) {
1373 /*
1374 * Only reference an existing MIDX when not filtering which
1375 * packs to include, since all packs and objects are copied
1376 * blindly from an existing MIDX if one is present.
1377 */
1378 ctx.m = lookup_multi_pack_index(the_repository, object_dir);
1379 }
1380
1381 if (ctx.m && !midx_checksum_valid(ctx.m)) {
1382 warning(_("ignoring existing multi-pack-index; checksum mismatch"));
1383 ctx.m = NULL;
1384 }
1385
1386 ctx.nr = 0;
1387 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
1388 ctx.info = NULL;
1389 ALLOC_ARRAY(ctx.info, ctx.alloc);
1390
1391 if (ctx.m) {
1392 for (i = 0; i < ctx.m->num_packs; i++) {
1393 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
1394
1395 if (flags & MIDX_WRITE_REV_INDEX) {
1396 /*
1397 * If generating a reverse index, need to have
1398 * packed_git's loaded to compare their
1399 * mtimes and object count.
1400 */
1401 if (prepare_midx_pack(the_repository, ctx.m, i)) {
1402 error(_("could not load pack"));
1403 result = 1;
1404 goto cleanup;
1405 }
1406
1407 if (open_pack_index(ctx.m->packs[i]))
1408 die(_("could not open index for %s"),
1409 ctx.m->packs[i]->pack_name);
1410 }
1411
1412 fill_pack_info(&ctx.info[ctx.nr++], ctx.m->packs[i],
1413 ctx.m->pack_names[i], i);
1414 }
1415 }
1416
1417 ctx.pack_paths_checked = 0;
1418 if (flags & MIDX_PROGRESS)
1419 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
1420 else
1421 ctx.progress = NULL;
1422
1423 ctx.to_include = packs_to_include;
1424
1425 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
1426 stop_progress(&ctx.progress);
1427
1428 if ((ctx.m && ctx.nr == ctx.m->num_packs) &&
1429 !(packs_to_include || packs_to_drop)) {
1430 struct bitmap_index *bitmap_git;
1431 int bitmap_exists;
1432 int want_bitmap = flags & MIDX_WRITE_BITMAP;
1433
1434 bitmap_git = prepare_midx_bitmap_git(ctx.m);
1435 bitmap_exists = bitmap_git && bitmap_is_midx(bitmap_git);
1436 free_bitmap_index(bitmap_git);
1437
1438 if (bitmap_exists || !want_bitmap) {
1439 /*
1440 * The correct MIDX already exists, and so does a
1441 * corresponding bitmap (or one wasn't requested).
1442 */
1443 if (!want_bitmap)
1444 clear_midx_files_ext(object_dir, ".bitmap",
1445 NULL);
1446 goto cleanup;
1447 }
1448 }
1449
1450 if (preferred_pack_name) {
1451 ctx.preferred_pack_idx = -1;
1452
1453 for (i = 0; i < ctx.nr; i++) {
1454 if (!cmp_idx_or_pack_name(preferred_pack_name,
1455 ctx.info[i].pack_name)) {
1456 ctx.preferred_pack_idx = i;
1457 break;
1458 }
1459 }
1460
1461 if (ctx.preferred_pack_idx == -1)
1462 warning(_("unknown preferred pack: '%s'"),
1463 preferred_pack_name);
1464 } else if (ctx.nr &&
1465 (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP))) {
1466 struct packed_git *oldest = ctx.info[ctx.preferred_pack_idx].p;
1467 ctx.preferred_pack_idx = 0;
1468
1469 if (packs_to_drop && packs_to_drop->nr)
1470 BUG("cannot write a MIDX bitmap during expiration");
1471
1472 /*
1473 * set a preferred pack when writing a bitmap to ensure that
1474 * the pack from which the first object is selected in pseudo
1475 * pack-order has all of its objects selected from that pack
1476 * (and not another pack containing a duplicate)
1477 */
1478 for (i = 1; i < ctx.nr; i++) {
1479 struct packed_git *p = ctx.info[i].p;
1480
1481 if (!oldest->num_objects || p->mtime < oldest->mtime) {
1482 oldest = p;
1483 ctx.preferred_pack_idx = i;
1484 }
1485 }
1486
1487 if (!oldest->num_objects) {
1488 /*
1489 * If all packs are empty; unset the preferred index.
1490 * This is acceptable since there will be no duplicate
1491 * objects to resolve, so the preferred value doesn't
1492 * matter.
1493 */
1494 ctx.preferred_pack_idx = -1;
1495 }
1496 } else {
1497 /*
1498 * otherwise don't mark any pack as preferred to avoid
1499 * interfering with expiration logic below
1500 */
1501 ctx.preferred_pack_idx = -1;
1502 }
1503
1504 if (ctx.preferred_pack_idx > -1) {
1505 struct packed_git *preferred = ctx.info[ctx.preferred_pack_idx].p;
1506 if (!preferred->num_objects) {
1507 error(_("cannot select preferred pack %s with no objects"),
1508 preferred->pack_name);
1509 result = 1;
1510 goto cleanup;
1511 }
1512 }
1513
1514 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr,
1515 ctx.preferred_pack_idx);
1516
1517 ctx.large_offsets_needed = 0;
1518 for (i = 0; i < ctx.entries_nr; i++) {
1519 if (ctx.entries[i].offset > 0x7fffffff)
1520 ctx.num_large_offsets++;
1521 if (ctx.entries[i].offset > 0xffffffff)
1522 ctx.large_offsets_needed = 1;
1523 }
1524
1525 QSORT(ctx.info, ctx.nr, pack_info_compare);
1526
1527 if (packs_to_drop && packs_to_drop->nr) {
1528 int drop_index = 0;
1529 int missing_drops = 0;
1530
1531 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
1532 int cmp = strcmp(ctx.info[i].pack_name,
1533 packs_to_drop->items[drop_index].string);
1534
1535 if (!cmp) {
1536 drop_index++;
1537 ctx.info[i].expired = 1;
1538 } else if (cmp > 0) {
1539 error(_("did not see pack-file %s to drop"),
1540 packs_to_drop->items[drop_index].string);
1541 drop_index++;
1542 missing_drops++;
1543 i--;
1544 } else {
1545 ctx.info[i].expired = 0;
1546 }
1547 }
1548
1549 if (missing_drops) {
1550 result = 1;
1551 goto cleanup;
1552 }
1553 }
1554
1555 /*
1556 * pack_perm stores a permutation between pack-int-ids from the
1557 * previous multi-pack-index to the new one we are writing:
1558 *
1559 * pack_perm[old_id] = new_id
1560 */
1561 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
1562 for (i = 0; i < ctx.nr; i++) {
1563 if (ctx.info[i].expired) {
1564 dropped_packs++;
1565 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
1566 } else {
1567 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
1568 }
1569 }
1570
1571 for (i = 0; i < ctx.nr; i++) {
1572 if (ctx.info[i].expired)
1573 continue;
1574 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
1575 bitmapped_packs_concat_len += 2 * sizeof(uint32_t);
1576 }
1577
1578 /* Check that the preferred pack wasn't expired (if given). */
1579 if (preferred_pack_name) {
1580 struct pack_info *preferred = bsearch(preferred_pack_name,
1581 ctx.info, ctx.nr,
1582 sizeof(*ctx.info),
1583 idx_or_pack_name_cmp);
1584 if (preferred) {
1585 uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id];
1586 if (perm == PACK_EXPIRED)
1587 warning(_("preferred pack '%s' is expired"),
1588 preferred_pack_name);
1589 }
1590 }
1591
1592 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
1593 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
1594 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
1595
1596 hold_lock_file_for_update(&lk, midx_name.buf, LOCK_DIE_ON_ERROR);
1597 f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk));
1598
1599 if (ctx.nr - dropped_packs == 0) {
1600 error(_("no pack files to index."));
1601 result = 1;
1602 goto cleanup;
1603 }
1604
1605 if (!ctx.entries_nr) {
1606 if (flags & MIDX_WRITE_BITMAP)
1607 warning(_("refusing to write multi-pack .bitmap without any objects"));
1608 flags &= ~(MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP);
1609 }
1610
1611 cf = init_chunkfile(f);
1612
1613 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
1614 write_midx_pack_names);
1615 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
1616 write_midx_oid_fanout);
1617 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
1618 st_mult(ctx.entries_nr, the_hash_algo->rawsz),
1619 write_midx_oid_lookup);
1620 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
1621 st_mult(ctx.entries_nr, MIDX_CHUNK_OFFSET_WIDTH),
1622 write_midx_object_offsets);
1623
1624 if (ctx.large_offsets_needed)
1625 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
1626 st_mult(ctx.num_large_offsets,
1627 MIDX_CHUNK_LARGE_OFFSET_WIDTH),
1628 write_midx_large_offsets);
1629
1630 if (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP)) {
1631 ctx.pack_order = midx_pack_order(&ctx);
1632 add_chunk(cf, MIDX_CHUNKID_REVINDEX,
1633 st_mult(ctx.entries_nr, sizeof(uint32_t)),
1634 write_midx_revindex);
1635 add_chunk(cf, MIDX_CHUNKID_BITMAPPEDPACKS,
1636 bitmapped_packs_concat_len,
1637 write_midx_bitmapped_packs);
1638 }
1639
1640 write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs);
1641 write_chunkfile(cf, &ctx);
1642
1643 finalize_hashfile(f, midx_hash, FSYNC_COMPONENT_PACK_METADATA,
1644 CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1645 free_chunkfile(cf);
1646
1647 if (flags & MIDX_WRITE_REV_INDEX &&
1648 git_env_bool("GIT_TEST_MIDX_WRITE_REV", 0))
1649 write_midx_reverse_index(midx_name.buf, midx_hash, &ctx);
1650
1651 if (flags & MIDX_WRITE_BITMAP) {
1652 struct packing_data pdata;
1653 struct commit **commits;
1654 uint32_t commits_nr;
1655
1656 if (!ctx.entries_nr)
1657 BUG("cannot write a bitmap without any objects");
1658
1659 prepare_midx_packing_data(&pdata, &ctx);
1660
1661 commits = find_commits_for_midx_bitmap(&commits_nr, refs_snapshot, &ctx);
1662
1663 /*
1664 * The previous steps translated the information from
1665 * 'entries' into information suitable for constructing
1666 * bitmaps. We no longer need that array, so clear it to
1667 * reduce memory pressure.
1668 */
1669 FREE_AND_NULL(ctx.entries);
1670 ctx.entries_nr = 0;
1671
1672 if (write_midx_bitmap(midx_name.buf, midx_hash, &pdata,
1673 commits, commits_nr, ctx.pack_order,
1674 flags) < 0) {
1675 error(_("could not write multi-pack bitmap"));
1676 result = 1;
1677 clear_packing_data(&pdata);
1678 free(commits);
1679 goto cleanup;
1680 }
1681
1682 clear_packing_data(&pdata);
1683 free(commits);
1684 }
1685 /*
1686 * NOTE: Do not use ctx.entries beyond this point, since it might
1687 * have been freed in the previous if block.
1688 */
1689
1690 if (ctx.m)
1691 close_object_store(the_repository->objects);
1692
1693 if (commit_lock_file(&lk) < 0)
1694 die_errno(_("could not write multi-pack-index"));
1695
1696 clear_midx_files_ext(object_dir, ".bitmap", midx_hash);
1697 clear_midx_files_ext(object_dir, ".rev", midx_hash);
1698
1699 cleanup:
1700 for (i = 0; i < ctx.nr; i++) {
1701 if (ctx.info[i].p) {
1702 close_pack(ctx.info[i].p);
1703 free(ctx.info[i].p);
1704 }
1705 free(ctx.info[i].pack_name);
1706 }
1707
1708 free(ctx.info);
1709 free(ctx.entries);
1710 free(ctx.pack_perm);
1711 free(ctx.pack_order);
1712 strbuf_release(&midx_name);
1713
1714 trace2_region_leave("midx", "write_midx_internal", the_repository);
1715
1716 return result;
1717 }
1718
1719 int write_midx_file(const char *object_dir,
1720 const char *preferred_pack_name,
1721 const char *refs_snapshot,
1722 unsigned flags)
1723 {
1724 return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name,
1725 refs_snapshot, flags);
1726 }
1727
1728 int write_midx_file_only(const char *object_dir,
1729 struct string_list *packs_to_include,
1730 const char *preferred_pack_name,
1731 const char *refs_snapshot,
1732 unsigned flags)
1733 {
1734 return write_midx_internal(object_dir, packs_to_include, NULL,
1735 preferred_pack_name, refs_snapshot, flags);
1736 }
1737
1738 struct clear_midx_data {
1739 char *keep;
1740 const char *ext;
1741 };
1742
1743 static void clear_midx_file_ext(const char *full_path, size_t full_path_len UNUSED,
1744 const char *file_name, void *_data)
1745 {
1746 struct clear_midx_data *data = _data;
1747
1748 if (!(starts_with(file_name, "multi-pack-index-") &&
1749 ends_with(file_name, data->ext)))
1750 return;
1751 if (data->keep && !strcmp(data->keep, file_name))
1752 return;
1753
1754 if (unlink(full_path))
1755 die_errno(_("failed to remove %s"), full_path);
1756 }
1757
1758 static void clear_midx_files_ext(const char *object_dir, const char *ext,
1759 unsigned char *keep_hash)
1760 {
1761 struct clear_midx_data data;
1762 memset(&data, 0, sizeof(struct clear_midx_data));
1763
1764 if (keep_hash)
1765 data.keep = xstrfmt("multi-pack-index-%s%s",
1766 hash_to_hex(keep_hash), ext);
1767 data.ext = ext;
1768
1769 for_each_file_in_pack_dir(object_dir,
1770 clear_midx_file_ext,
1771 &data);
1772
1773 free(data.keep);
1774 }
1775
1776 void clear_midx_file(struct repository *r)
1777 {
1778 struct strbuf midx = STRBUF_INIT;
1779
1780 get_midx_filename(&midx, r->objects->odb->path);
1781
1782 if (r->objects && r->objects->multi_pack_index) {
1783 close_midx(r->objects->multi_pack_index);
1784 r->objects->multi_pack_index = NULL;
1785 }
1786
1787 if (remove_path(midx.buf))
1788 die(_("failed to clear multi-pack-index at %s"), midx.buf);
1789
1790 clear_midx_files_ext(r->objects->odb->path, ".bitmap", NULL);
1791 clear_midx_files_ext(r->objects->odb->path, ".rev", NULL);
1792
1793 strbuf_release(&midx);
1794 }
1795
1796 static int verify_midx_error;
1797
1798 __attribute__((format (printf, 1, 2)))
1799 static void midx_report(const char *fmt, ...)
1800 {
1801 va_list ap;
1802 verify_midx_error = 1;
1803 va_start(ap, fmt);
1804 vfprintf(stderr, fmt, ap);
1805 fprintf(stderr, "\n");
1806 va_end(ap);
1807 }
1808
1809 struct pair_pos_vs_id
1810 {
1811 uint32_t pos;
1812 uint32_t pack_int_id;
1813 };
1814
1815 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1816 {
1817 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1818 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1819
1820 return b->pack_int_id - a->pack_int_id;
1821 }
1822
1823 /*
1824 * Limit calls to display_progress() for performance reasons.
1825 * The interval here was arbitrarily chosen.
1826 */
1827 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1828 #define midx_display_sparse_progress(progress, n) \
1829 do { \
1830 uint64_t _n = (n); \
1831 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1832 display_progress(progress, _n); \
1833 } while (0)
1834
1835 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1836 {
1837 struct pair_pos_vs_id *pairs = NULL;
1838 uint32_t i;
1839 struct progress *progress = NULL;
1840 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1841 verify_midx_error = 0;
1842
1843 if (!m) {
1844 int result = 0;
1845 struct stat sb;
1846 struct strbuf filename = STRBUF_INIT;
1847
1848 get_midx_filename(&filename, object_dir);
1849
1850 if (!stat(filename.buf, &sb)) {
1851 error(_("multi-pack-index file exists, but failed to parse"));
1852 result = 1;
1853 }
1854 strbuf_release(&filename);
1855 return result;
1856 }
1857
1858 if (!midx_checksum_valid(m))
1859 midx_report(_("incorrect checksum"));
1860
1861 if (flags & MIDX_PROGRESS)
1862 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1863 m->num_packs);
1864 for (i = 0; i < m->num_packs; i++) {
1865 if (prepare_midx_pack(r, m, i))
1866 midx_report("failed to load pack in position %d", i);
1867
1868 display_progress(progress, i + 1);
1869 }
1870 stop_progress(&progress);
1871
1872 if (m->num_objects == 0) {
1873 midx_report(_("the midx contains no oid"));
1874 /*
1875 * Remaining tests assume that we have objects, so we can
1876 * return here.
1877 */
1878 goto cleanup;
1879 }
1880
1881 if (flags & MIDX_PROGRESS)
1882 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1883 m->num_objects - 1);
1884 for (i = 0; i < m->num_objects - 1; i++) {
1885 struct object_id oid1, oid2;
1886
1887 nth_midxed_object_oid(&oid1, m, i);
1888 nth_midxed_object_oid(&oid2, m, i + 1);
1889
1890 if (oidcmp(&oid1, &oid2) >= 0)
1891 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1892 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1893
1894 midx_display_sparse_progress(progress, i + 1);
1895 }
1896 stop_progress(&progress);
1897
1898 /*
1899 * Create an array mapping each object to its packfile id. Sort it
1900 * to group the objects by packfile. Use this permutation to visit
1901 * each of the objects and only require 1 packfile to be open at a
1902 * time.
1903 */
1904 ALLOC_ARRAY(pairs, m->num_objects);
1905 for (i = 0; i < m->num_objects; i++) {
1906 pairs[i].pos = i;
1907 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1908 }
1909
1910 if (flags & MIDX_PROGRESS)
1911 progress = start_sparse_progress(_("Sorting objects by packfile"),
1912 m->num_objects);
1913 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1914 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1915 stop_progress(&progress);
1916
1917 if (flags & MIDX_PROGRESS)
1918 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1919 for (i = 0; i < m->num_objects; i++) {
1920 struct object_id oid;
1921 struct pack_entry e;
1922 off_t m_offset, p_offset;
1923
1924 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1925 m->packs[pairs[i-1].pack_int_id])
1926 {
1927 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1928 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1929 }
1930
1931 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1932
1933 if (!fill_midx_entry(r, &oid, &e, m)) {
1934 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1935 pairs[i].pos, oid_to_hex(&oid));
1936 continue;
1937 }
1938
1939 if (open_pack_index(e.p)) {
1940 midx_report(_("failed to load pack-index for packfile %s"),
1941 e.p->pack_name);
1942 break;
1943 }
1944
1945 m_offset = e.offset;
1946 p_offset = find_pack_entry_one(oid.hash, e.p);
1947
1948 if (m_offset != p_offset)
1949 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1950 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1951
1952 midx_display_sparse_progress(progress, i + 1);
1953 }
1954 stop_progress(&progress);
1955
1956 cleanup:
1957 free(pairs);
1958 close_midx(m);
1959
1960 return verify_midx_error;
1961 }
1962
1963 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1964 {
1965 uint32_t i, *count, result = 0;
1966 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1967 struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir);
1968 struct progress *progress = NULL;
1969
1970 if (!m)
1971 return 0;
1972
1973 CALLOC_ARRAY(count, m->num_packs);
1974
1975 if (flags & MIDX_PROGRESS)
1976 progress = start_delayed_progress(_("Counting referenced objects"),
1977 m->num_objects);
1978 for (i = 0; i < m->num_objects; i++) {
1979 int pack_int_id = nth_midxed_pack_int_id(m, i);
1980 count[pack_int_id]++;
1981 display_progress(progress, i + 1);
1982 }
1983 stop_progress(&progress);
1984
1985 if (flags & MIDX_PROGRESS)
1986 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1987 m->num_packs);
1988 for (i = 0; i < m->num_packs; i++) {
1989 char *pack_name;
1990 display_progress(progress, i + 1);
1991
1992 if (count[i])
1993 continue;
1994
1995 if (prepare_midx_pack(r, m, i))
1996 continue;
1997
1998 if (m->packs[i]->pack_keep || m->packs[i]->is_cruft)
1999 continue;
2000
2001 pack_name = xstrdup(m->packs[i]->pack_name);
2002 close_pack(m->packs[i]);
2003
2004 string_list_insert(&packs_to_drop, m->pack_names[i]);
2005 unlink_pack_path(pack_name, 0);
2006 free(pack_name);
2007 }
2008 stop_progress(&progress);
2009
2010 free(count);
2011
2012 if (packs_to_drop.nr)
2013 result = write_midx_internal(object_dir, NULL, &packs_to_drop, NULL, NULL, flags);
2014
2015 string_list_clear(&packs_to_drop, 0);
2016
2017 return result;
2018 }
2019
2020 struct repack_info {
2021 timestamp_t mtime;
2022 uint32_t referenced_objects;
2023 uint32_t pack_int_id;
2024 };
2025
2026 static int compare_by_mtime(const void *a_, const void *b_)
2027 {
2028 const struct repack_info *a, *b;
2029
2030 a = (const struct repack_info *)a_;
2031 b = (const struct repack_info *)b_;
2032
2033 if (a->mtime < b->mtime)
2034 return -1;
2035 if (a->mtime > b->mtime)
2036 return 1;
2037 return 0;
2038 }
2039
2040 static int fill_included_packs_all(struct repository *r,
2041 struct multi_pack_index *m,
2042 unsigned char *include_pack)
2043 {
2044 uint32_t i, count = 0;
2045 int pack_kept_objects = 0;
2046
2047 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
2048
2049 for (i = 0; i < m->num_packs; i++) {
2050 if (prepare_midx_pack(r, m, i))
2051 continue;
2052 if (!pack_kept_objects && m->packs[i]->pack_keep)
2053 continue;
2054 if (m->packs[i]->is_cruft)
2055 continue;
2056
2057 include_pack[i] = 1;
2058 count++;
2059 }
2060
2061 return count < 2;
2062 }
2063
2064 static int fill_included_packs_batch(struct repository *r,
2065 struct multi_pack_index *m,
2066 unsigned char *include_pack,
2067 size_t batch_size)
2068 {
2069 uint32_t i, packs_to_repack;
2070 size_t total_size;
2071 struct repack_info *pack_info;
2072 int pack_kept_objects = 0;
2073
2074 CALLOC_ARRAY(pack_info, m->num_packs);
2075
2076 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
2077
2078 for (i = 0; i < m->num_packs; i++) {
2079 pack_info[i].pack_int_id = i;
2080
2081 if (prepare_midx_pack(r, m, i))
2082 continue;
2083
2084 pack_info[i].mtime = m->packs[i]->mtime;
2085 }
2086
2087 for (i = 0; i < m->num_objects; i++) {
2088 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
2089 pack_info[pack_int_id].referenced_objects++;
2090 }
2091
2092 QSORT(pack_info, m->num_packs, compare_by_mtime);
2093
2094 total_size = 0;
2095 packs_to_repack = 0;
2096 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
2097 int pack_int_id = pack_info[i].pack_int_id;
2098 struct packed_git *p = m->packs[pack_int_id];
2099 size_t expected_size;
2100
2101 if (!p)
2102 continue;
2103 if (!pack_kept_objects && p->pack_keep)
2104 continue;
2105 if (p->is_cruft)
2106 continue;
2107 if (open_pack_index(p) || !p->num_objects)
2108 continue;
2109
2110 expected_size = st_mult(p->pack_size,
2111 pack_info[i].referenced_objects);
2112 expected_size /= p->num_objects;
2113
2114 if (expected_size >= batch_size)
2115 continue;
2116
2117 packs_to_repack++;
2118 total_size += expected_size;
2119 include_pack[pack_int_id] = 1;
2120 }
2121
2122 free(pack_info);
2123
2124 if (packs_to_repack < 2)
2125 return 1;
2126
2127 return 0;
2128 }
2129
2130 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
2131 {
2132 int result = 0;
2133 uint32_t i;
2134 unsigned char *include_pack;
2135 struct child_process cmd = CHILD_PROCESS_INIT;
2136 FILE *cmd_in;
2137 struct strbuf base_name = STRBUF_INIT;
2138 struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir);
2139
2140 /*
2141 * When updating the default for these configuration
2142 * variables in builtin/repack.c, these must be adjusted
2143 * to match.
2144 */
2145 int delta_base_offset = 1;
2146 int use_delta_islands = 0;
2147
2148 if (!m)
2149 return 0;
2150
2151 CALLOC_ARRAY(include_pack, m->num_packs);
2152
2153 if (batch_size) {
2154 if (fill_included_packs_batch(r, m, include_pack, batch_size))
2155 goto cleanup;
2156 } else if (fill_included_packs_all(r, m, include_pack))
2157 goto cleanup;
2158
2159 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
2160 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
2161
2162 strvec_push(&cmd.args, "pack-objects");
2163
2164 strbuf_addstr(&base_name, object_dir);
2165 strbuf_addstr(&base_name, "/pack/pack");
2166 strvec_push(&cmd.args, base_name.buf);
2167
2168 if (delta_base_offset)
2169 strvec_push(&cmd.args, "--delta-base-offset");
2170 if (use_delta_islands)
2171 strvec_push(&cmd.args, "--delta-islands");
2172
2173 if (flags & MIDX_PROGRESS)
2174 strvec_push(&cmd.args, "--progress");
2175 else
2176 strvec_push(&cmd.args, "-q");
2177
2178 strbuf_release(&base_name);
2179
2180 cmd.git_cmd = 1;
2181 cmd.in = cmd.out = -1;
2182
2183 if (start_command(&cmd)) {
2184 error(_("could not start pack-objects"));
2185 result = 1;
2186 goto cleanup;
2187 }
2188
2189 cmd_in = xfdopen(cmd.in, "w");
2190
2191 for (i = 0; i < m->num_objects; i++) {
2192 struct object_id oid;
2193 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
2194
2195 if (!include_pack[pack_int_id])
2196 continue;
2197
2198 nth_midxed_object_oid(&oid, m, i);
2199 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
2200 }
2201 fclose(cmd_in);
2202
2203 if (finish_command(&cmd)) {
2204 error(_("could not finish pack-objects"));
2205 result = 1;
2206 goto cleanup;
2207 }
2208
2209 result = write_midx_internal(object_dir, NULL, NULL, NULL, NULL, flags);
2210
2211 cleanup:
2212 free(include_pack);
2213 return result;
2214 }
Mirror of Junio's Git repository with all the branches