Merge branch 'jk/perf-in-worktrees'
[git.git] / midx.c
blobbecfafe65e8029d4381b7302cf9644096f45f293
1 #include "cache.h"
2 #include "config.h"
3 #include "csum-file.h"
4 #include "dir.h"
5 #include "lockfile.h"
6 #include "packfile.h"
7 #include "object-store.h"
8 #include "hash-lookup.h"
9 #include "midx.h"
10 #include "progress.h"
11 #include "trace2.h"
12 #include "run-command.h"
13 #include "repository.h"
14 #include "chunk-format.h"
16 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
17 #define MIDX_VERSION 1
18 #define MIDX_BYTE_FILE_VERSION 4
19 #define MIDX_BYTE_HASH_VERSION 5
20 #define MIDX_BYTE_NUM_CHUNKS 6
21 #define MIDX_BYTE_NUM_PACKS 8
22 #define MIDX_HEADER_SIZE 12
23 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
25 #define MIDX_CHUNK_ALIGNMENT 4
26 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
27 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
28 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
29 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
30 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
31 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
32 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
33 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
34 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
36 #define PACK_EXPIRED UINT_MAX
38 static uint8_t oid_version(void)
40 switch (hash_algo_by_ptr(the_hash_algo)) {
41 case GIT_HASH_SHA1:
42 return 1;
43 case GIT_HASH_SHA256:
44 return 2;
45 default:
46 die(_("invalid hash version"));
50 static char *get_midx_filename(const char *object_dir)
52 return xstrfmt("%s/pack/multi-pack-index", object_dir);
55 static int midx_read_oid_fanout(const unsigned char *chunk_start,
56 size_t chunk_size, void *data)
58 struct multi_pack_index *m = data;
59 m->chunk_oid_fanout = (uint32_t *)chunk_start;
61 if (chunk_size != 4 * 256) {
62 error(_("multi-pack-index OID fanout is of the wrong size"));
63 return 1;
65 return 0;
68 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
70 struct multi_pack_index *m = NULL;
71 int fd;
72 struct stat st;
73 size_t midx_size;
74 void *midx_map = NULL;
75 uint32_t hash_version;
76 char *midx_name = get_midx_filename(object_dir);
77 uint32_t i;
78 const char *cur_pack_name;
79 struct chunkfile *cf = NULL;
81 fd = git_open(midx_name);
83 if (fd < 0)
84 goto cleanup_fail;
85 if (fstat(fd, &st)) {
86 error_errno(_("failed to read %s"), midx_name);
87 goto cleanup_fail;
90 midx_size = xsize_t(st.st_size);
92 if (midx_size < MIDX_MIN_SIZE) {
93 error(_("multi-pack-index file %s is too small"), midx_name);
94 goto cleanup_fail;
97 FREE_AND_NULL(midx_name);
99 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
100 close(fd);
102 FLEX_ALLOC_STR(m, object_dir, object_dir);
103 m->data = midx_map;
104 m->data_len = midx_size;
105 m->local = local;
107 m->signature = get_be32(m->data);
108 if (m->signature != MIDX_SIGNATURE)
109 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
110 m->signature, MIDX_SIGNATURE);
112 m->version = m->data[MIDX_BYTE_FILE_VERSION];
113 if (m->version != MIDX_VERSION)
114 die(_("multi-pack-index version %d not recognized"),
115 m->version);
117 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
118 if (hash_version != oid_version()) {
119 error(_("multi-pack-index hash version %u does not match version %u"),
120 hash_version, oid_version());
121 goto cleanup_fail;
123 m->hash_len = the_hash_algo->rawsz;
125 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
127 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
129 cf = init_chunkfile(NULL);
131 if (read_table_of_contents(cf, m->data, midx_size,
132 MIDX_HEADER_SIZE, m->num_chunks))
133 goto cleanup_fail;
135 if (pair_chunk(cf, MIDX_CHUNKID_PACKNAMES, &m->chunk_pack_names) == CHUNK_NOT_FOUND)
136 die(_("multi-pack-index missing required pack-name chunk"));
137 if (read_chunk(cf, MIDX_CHUNKID_OIDFANOUT, midx_read_oid_fanout, m) == CHUNK_NOT_FOUND)
138 die(_("multi-pack-index missing required OID fanout chunk"));
139 if (pair_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, &m->chunk_oid_lookup) == CHUNK_NOT_FOUND)
140 die(_("multi-pack-index missing required OID lookup chunk"));
141 if (pair_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, &m->chunk_object_offsets) == CHUNK_NOT_FOUND)
142 die(_("multi-pack-index missing required object offsets chunk"));
144 pair_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, &m->chunk_large_offsets);
146 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
148 CALLOC_ARRAY(m->pack_names, m->num_packs);
149 CALLOC_ARRAY(m->packs, m->num_packs);
151 cur_pack_name = (const char *)m->chunk_pack_names;
152 for (i = 0; i < m->num_packs; i++) {
153 m->pack_names[i] = cur_pack_name;
155 cur_pack_name += strlen(cur_pack_name) + 1;
157 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
158 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
159 m->pack_names[i - 1],
160 m->pack_names[i]);
163 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
164 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
166 return m;
168 cleanup_fail:
169 free(m);
170 free(midx_name);
171 free(cf);
172 if (midx_map)
173 munmap(midx_map, midx_size);
174 if (0 <= fd)
175 close(fd);
176 return NULL;
179 void close_midx(struct multi_pack_index *m)
181 uint32_t i;
183 if (!m)
184 return;
186 munmap((unsigned char *)m->data, m->data_len);
188 for (i = 0; i < m->num_packs; i++) {
189 if (m->packs[i])
190 m->packs[i]->multi_pack_index = 0;
192 FREE_AND_NULL(m->packs);
193 FREE_AND_NULL(m->pack_names);
196 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
198 struct strbuf pack_name = STRBUF_INIT;
199 struct packed_git *p;
201 if (pack_int_id >= m->num_packs)
202 die(_("bad pack-int-id: %u (%u total packs)"),
203 pack_int_id, m->num_packs);
205 if (m->packs[pack_int_id])
206 return 0;
208 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
209 m->pack_names[pack_int_id]);
211 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
212 strbuf_release(&pack_name);
214 if (!p)
215 return 1;
217 p->multi_pack_index = 1;
218 m->packs[pack_int_id] = p;
219 install_packed_git(r, p);
220 list_add_tail(&p->mru, &r->objects->packed_git_mru);
222 return 0;
225 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
227 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
228 the_hash_algo->rawsz, result);
231 struct object_id *nth_midxed_object_oid(struct object_id *oid,
232 struct multi_pack_index *m,
233 uint32_t n)
235 if (n >= m->num_objects)
236 return NULL;
238 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
239 return oid;
242 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
244 const unsigned char *offset_data;
245 uint32_t offset32;
247 offset_data = m->chunk_object_offsets + (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH;
248 offset32 = get_be32(offset_data + sizeof(uint32_t));
250 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
251 if (sizeof(off_t) < sizeof(uint64_t))
252 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
254 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
255 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
258 return offset32;
261 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
263 return get_be32(m->chunk_object_offsets +
264 (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH);
267 static int nth_midxed_pack_entry(struct repository *r,
268 struct multi_pack_index *m,
269 struct pack_entry *e,
270 uint32_t pos)
272 uint32_t pack_int_id;
273 struct packed_git *p;
275 if (pos >= m->num_objects)
276 return 0;
278 pack_int_id = nth_midxed_pack_int_id(m, pos);
280 if (prepare_midx_pack(r, m, pack_int_id))
281 return 0;
282 p = m->packs[pack_int_id];
285 * We are about to tell the caller where they can locate the
286 * requested object. We better make sure the packfile is
287 * still here and can be accessed before supplying that
288 * answer, as it may have been deleted since the MIDX was
289 * loaded!
291 if (!is_pack_valid(p))
292 return 0;
294 if (p->num_bad_objects) {
295 uint32_t i;
296 struct object_id oid;
297 nth_midxed_object_oid(&oid, m, pos);
298 for (i = 0; i < p->num_bad_objects; i++)
299 if (hasheq(oid.hash,
300 p->bad_object_sha1 + the_hash_algo->rawsz * i))
301 return 0;
304 e->offset = nth_midxed_offset(m, pos);
305 e->p = p;
307 return 1;
310 int fill_midx_entry(struct repository * r,
311 const struct object_id *oid,
312 struct pack_entry *e,
313 struct multi_pack_index *m)
315 uint32_t pos;
317 if (!bsearch_midx(oid, m, &pos))
318 return 0;
320 return nth_midxed_pack_entry(r, m, e, pos);
323 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
324 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
325 const char *idx_name)
327 /* Skip past any initial matching prefix. */
328 while (*idx_name && *idx_name == *idx_or_pack_name) {
329 idx_name++;
330 idx_or_pack_name++;
334 * If we didn't match completely, we may have matched "pack-1234." and
335 * be left with "idx" and "pack" respectively, which is also OK. We do
336 * not have to check for "idx" and "idx", because that would have been
337 * a complete match (and in that case these strcmps will be false, but
338 * we'll correctly return 0 from the final strcmp() below.
340 * Technically this matches "fooidx" and "foopack", but we'd never have
341 * such names in the first place.
343 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
344 return 0;
347 * This not only checks for a complete match, but also orders based on
348 * the first non-identical character, which means our ordering will
349 * match a raw strcmp(). That makes it OK to use this to binary search
350 * a naively-sorted list.
352 return strcmp(idx_or_pack_name, idx_name);
355 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
357 uint32_t first = 0, last = m->num_packs;
359 while (first < last) {
360 uint32_t mid = first + (last - first) / 2;
361 const char *current;
362 int cmp;
364 current = m->pack_names[mid];
365 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
366 if (!cmp)
367 return 1;
368 if (cmp > 0) {
369 first = mid + 1;
370 continue;
372 last = mid;
375 return 0;
378 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
380 struct multi_pack_index *m;
381 struct multi_pack_index *m_search;
383 prepare_repo_settings(r);
384 if (!r->settings.core_multi_pack_index)
385 return 0;
387 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
388 if (!strcmp(object_dir, m_search->object_dir))
389 return 1;
391 m = load_multi_pack_index(object_dir, local);
393 if (m) {
394 struct multi_pack_index *mp = r->objects->multi_pack_index;
395 if (mp) {
396 m->next = mp->next;
397 mp->next = m;
398 } else
399 r->objects->multi_pack_index = m;
400 return 1;
403 return 0;
406 static size_t write_midx_header(struct hashfile *f,
407 unsigned char num_chunks,
408 uint32_t num_packs)
410 hashwrite_be32(f, MIDX_SIGNATURE);
411 hashwrite_u8(f, MIDX_VERSION);
412 hashwrite_u8(f, oid_version());
413 hashwrite_u8(f, num_chunks);
414 hashwrite_u8(f, 0); /* unused */
415 hashwrite_be32(f, num_packs);
417 return MIDX_HEADER_SIZE;
420 struct pack_info {
421 uint32_t orig_pack_int_id;
422 char *pack_name;
423 struct packed_git *p;
424 unsigned expired : 1;
427 static int pack_info_compare(const void *_a, const void *_b)
429 struct pack_info *a = (struct pack_info *)_a;
430 struct pack_info *b = (struct pack_info *)_b;
431 return strcmp(a->pack_name, b->pack_name);
434 struct write_midx_context {
435 struct pack_info *info;
436 uint32_t nr;
437 uint32_t alloc;
438 struct multi_pack_index *m;
439 struct progress *progress;
440 unsigned pack_paths_checked;
442 struct pack_midx_entry *entries;
443 uint32_t entries_nr;
445 uint32_t *pack_perm;
446 unsigned large_offsets_needed:1;
447 uint32_t num_large_offsets;
450 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
451 const char *file_name, void *data)
453 struct write_midx_context *ctx = data;
455 if (ends_with(file_name, ".idx")) {
456 display_progress(ctx->progress, ++ctx->pack_paths_checked);
457 if (ctx->m && midx_contains_pack(ctx->m, file_name))
458 return;
460 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
462 ctx->info[ctx->nr].p = add_packed_git(full_path,
463 full_path_len,
466 if (!ctx->info[ctx->nr].p) {
467 warning(_("failed to add packfile '%s'"),
468 full_path);
469 return;
472 if (open_pack_index(ctx->info[ctx->nr].p)) {
473 warning(_("failed to open pack-index '%s'"),
474 full_path);
475 close_pack(ctx->info[ctx->nr].p);
476 FREE_AND_NULL(ctx->info[ctx->nr].p);
477 return;
480 ctx->info[ctx->nr].pack_name = xstrdup(file_name);
481 ctx->info[ctx->nr].orig_pack_int_id = ctx->nr;
482 ctx->info[ctx->nr].expired = 0;
483 ctx->nr++;
487 struct pack_midx_entry {
488 struct object_id oid;
489 uint32_t pack_int_id;
490 time_t pack_mtime;
491 uint64_t offset;
494 static int midx_oid_compare(const void *_a, const void *_b)
496 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
497 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
498 int cmp = oidcmp(&a->oid, &b->oid);
500 if (cmp)
501 return cmp;
503 if (a->pack_mtime > b->pack_mtime)
504 return -1;
505 else if (a->pack_mtime < b->pack_mtime)
506 return 1;
508 return a->pack_int_id - b->pack_int_id;
511 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
512 struct pack_midx_entry *e,
513 uint32_t pos)
515 if (pos >= m->num_objects)
516 return 1;
518 nth_midxed_object_oid(&e->oid, m, pos);
519 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
520 e->offset = nth_midxed_offset(m, pos);
522 /* consider objects in midx to be from "old" packs */
523 e->pack_mtime = 0;
524 return 0;
527 static void fill_pack_entry(uint32_t pack_int_id,
528 struct packed_git *p,
529 uint32_t cur_object,
530 struct pack_midx_entry *entry)
532 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
533 die(_("failed to locate object %d in packfile"), cur_object);
535 entry->pack_int_id = pack_int_id;
536 entry->pack_mtime = p->mtime;
538 entry->offset = nth_packed_object_offset(p, cur_object);
542 * It is possible to artificially get into a state where there are many
543 * duplicate copies of objects. That can create high memory pressure if
544 * we are to create a list of all objects before de-duplication. To reduce
545 * this memory pressure without a significant performance drop, automatically
546 * group objects by the first byte of their object id. Use the IDX fanout
547 * tables to group the data, copy to a local array, then sort.
549 * Copy only the de-duplicated entries (selected by most-recent modified time
550 * of a packfile containing the object).
552 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
553 struct pack_info *info,
554 uint32_t nr_packs,
555 uint32_t *nr_objects)
557 uint32_t cur_fanout, cur_pack, cur_object;
558 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
559 struct pack_midx_entry *entries_by_fanout = NULL;
560 struct pack_midx_entry *deduplicated_entries = NULL;
561 uint32_t start_pack = m ? m->num_packs : 0;
563 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
564 total_objects += info[cur_pack].p->num_objects;
567 * As we de-duplicate by fanout value, we expect the fanout
568 * slices to be evenly distributed, with some noise. Hence,
569 * allocate slightly more than one 256th.
571 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
573 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
574 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
575 *nr_objects = 0;
577 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
578 uint32_t nr_fanout = 0;
580 if (m) {
581 uint32_t start = 0, end;
583 if (cur_fanout)
584 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
585 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
587 for (cur_object = start; cur_object < end; cur_object++) {
588 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
589 nth_midxed_pack_midx_entry(m,
590 &entries_by_fanout[nr_fanout],
591 cur_object);
592 nr_fanout++;
596 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
597 uint32_t start = 0, end;
599 if (cur_fanout)
600 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
601 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
603 for (cur_object = start; cur_object < end; cur_object++) {
604 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
605 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
606 nr_fanout++;
610 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
613 * The batch is now sorted by OID and then mtime (descending).
614 * Take only the first duplicate.
616 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
617 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
618 &entries_by_fanout[cur_object].oid))
619 continue;
621 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
622 memcpy(&deduplicated_entries[*nr_objects],
623 &entries_by_fanout[cur_object],
624 sizeof(struct pack_midx_entry));
625 (*nr_objects)++;
629 free(entries_by_fanout);
630 return deduplicated_entries;
633 static int write_midx_pack_names(struct hashfile *f, void *data)
635 struct write_midx_context *ctx = data;
636 uint32_t i;
637 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
638 size_t written = 0;
640 for (i = 0; i < ctx->nr; i++) {
641 size_t writelen;
643 if (ctx->info[i].expired)
644 continue;
646 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
647 BUG("incorrect pack-file order: %s before %s",
648 ctx->info[i - 1].pack_name,
649 ctx->info[i].pack_name);
651 writelen = strlen(ctx->info[i].pack_name) + 1;
652 hashwrite(f, ctx->info[i].pack_name, writelen);
653 written += writelen;
656 /* add padding to be aligned */
657 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
658 if (i < MIDX_CHUNK_ALIGNMENT) {
659 memset(padding, 0, sizeof(padding));
660 hashwrite(f, padding, i);
663 return 0;
666 static int write_midx_oid_fanout(struct hashfile *f,
667 void *data)
669 struct write_midx_context *ctx = data;
670 struct pack_midx_entry *list = ctx->entries;
671 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
672 uint32_t count = 0;
673 uint32_t i;
676 * Write the first-level table (the list is sorted,
677 * but we use a 256-entry lookup to be able to avoid
678 * having to do eight extra binary search iterations).
680 for (i = 0; i < 256; i++) {
681 struct pack_midx_entry *next = list;
683 while (next < last && next->oid.hash[0] == i) {
684 count++;
685 next++;
688 hashwrite_be32(f, count);
689 list = next;
692 return 0;
695 static int write_midx_oid_lookup(struct hashfile *f,
696 void *data)
698 struct write_midx_context *ctx = data;
699 unsigned char hash_len = the_hash_algo->rawsz;
700 struct pack_midx_entry *list = ctx->entries;
701 uint32_t i;
703 for (i = 0; i < ctx->entries_nr; i++) {
704 struct pack_midx_entry *obj = list++;
706 if (i < ctx->entries_nr - 1) {
707 struct pack_midx_entry *next = list;
708 if (oidcmp(&obj->oid, &next->oid) >= 0)
709 BUG("OIDs not in order: %s >= %s",
710 oid_to_hex(&obj->oid),
711 oid_to_hex(&next->oid));
714 hashwrite(f, obj->oid.hash, (int)hash_len);
717 return 0;
720 static int write_midx_object_offsets(struct hashfile *f,
721 void *data)
723 struct write_midx_context *ctx = data;
724 struct pack_midx_entry *list = ctx->entries;
725 uint32_t i, nr_large_offset = 0;
727 for (i = 0; i < ctx->entries_nr; i++) {
728 struct pack_midx_entry *obj = list++;
730 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
731 BUG("object %s is in an expired pack with int-id %d",
732 oid_to_hex(&obj->oid),
733 obj->pack_int_id);
735 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
737 if (ctx->large_offsets_needed && obj->offset >> 31)
738 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
739 else if (!ctx->large_offsets_needed && obj->offset >> 32)
740 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
741 oid_to_hex(&obj->oid),
742 obj->offset);
743 else
744 hashwrite_be32(f, (uint32_t)obj->offset);
747 return 0;
750 static int write_midx_large_offsets(struct hashfile *f,
751 void *data)
753 struct write_midx_context *ctx = data;
754 struct pack_midx_entry *list = ctx->entries;
755 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
756 uint32_t nr_large_offset = ctx->num_large_offsets;
758 while (nr_large_offset) {
759 struct pack_midx_entry *obj;
760 uint64_t offset;
762 if (list >= end)
763 BUG("too many large-offset objects");
765 obj = list++;
766 offset = obj->offset;
768 if (!(offset >> 31))
769 continue;
771 hashwrite_be64(f, offset);
773 nr_large_offset--;
776 return 0;
779 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
780 struct string_list *packs_to_drop, unsigned flags)
782 char *midx_name;
783 uint32_t i;
784 struct hashfile *f = NULL;
785 struct lock_file lk;
786 struct write_midx_context ctx = { 0 };
787 int pack_name_concat_len = 0;
788 int dropped_packs = 0;
789 int result = 0;
790 struct chunkfile *cf;
792 midx_name = get_midx_filename(object_dir);
793 if (safe_create_leading_directories(midx_name))
794 die_errno(_("unable to create leading directories of %s"),
795 midx_name);
797 if (m)
798 ctx.m = m;
799 else
800 ctx.m = load_multi_pack_index(object_dir, 1);
802 ctx.nr = 0;
803 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
804 ctx.info = NULL;
805 ALLOC_ARRAY(ctx.info, ctx.alloc);
807 if (ctx.m) {
808 for (i = 0; i < ctx.m->num_packs; i++) {
809 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
811 ctx.info[ctx.nr].orig_pack_int_id = i;
812 ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]);
813 ctx.info[ctx.nr].p = NULL;
814 ctx.info[ctx.nr].expired = 0;
815 ctx.nr++;
819 ctx.pack_paths_checked = 0;
820 if (flags & MIDX_PROGRESS)
821 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
822 else
823 ctx.progress = NULL;
825 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
826 stop_progress(&ctx.progress);
828 if (ctx.m && ctx.nr == ctx.m->num_packs && !packs_to_drop)
829 goto cleanup;
831 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr);
833 ctx.large_offsets_needed = 0;
834 for (i = 0; i < ctx.entries_nr; i++) {
835 if (ctx.entries[i].offset > 0x7fffffff)
836 ctx.num_large_offsets++;
837 if (ctx.entries[i].offset > 0xffffffff)
838 ctx.large_offsets_needed = 1;
841 QSORT(ctx.info, ctx.nr, pack_info_compare);
843 if (packs_to_drop && packs_to_drop->nr) {
844 int drop_index = 0;
845 int missing_drops = 0;
847 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
848 int cmp = strcmp(ctx.info[i].pack_name,
849 packs_to_drop->items[drop_index].string);
851 if (!cmp) {
852 drop_index++;
853 ctx.info[i].expired = 1;
854 } else if (cmp > 0) {
855 error(_("did not see pack-file %s to drop"),
856 packs_to_drop->items[drop_index].string);
857 drop_index++;
858 missing_drops++;
859 i--;
860 } else {
861 ctx.info[i].expired = 0;
865 if (missing_drops) {
866 result = 1;
867 goto cleanup;
872 * pack_perm stores a permutation between pack-int-ids from the
873 * previous multi-pack-index to the new one we are writing:
875 * pack_perm[old_id] = new_id
877 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
878 for (i = 0; i < ctx.nr; i++) {
879 if (ctx.info[i].expired) {
880 dropped_packs++;
881 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
882 } else {
883 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
887 for (i = 0; i < ctx.nr; i++) {
888 if (!ctx.info[i].expired)
889 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
892 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
893 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
894 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
896 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
897 f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk));
898 FREE_AND_NULL(midx_name);
900 if (ctx.m)
901 close_midx(ctx.m);
903 if (ctx.nr - dropped_packs == 0) {
904 error(_("no pack files to index."));
905 result = 1;
906 goto cleanup;
909 cf = init_chunkfile(f);
911 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
912 write_midx_pack_names);
913 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
914 write_midx_oid_fanout);
915 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
916 (size_t)ctx.entries_nr * the_hash_algo->rawsz,
917 write_midx_oid_lookup);
918 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
919 (size_t)ctx.entries_nr * MIDX_CHUNK_OFFSET_WIDTH,
920 write_midx_object_offsets);
922 if (ctx.large_offsets_needed)
923 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
924 (size_t)ctx.num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH,
925 write_midx_large_offsets);
927 write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs);
928 write_chunkfile(cf, &ctx);
930 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
931 free_chunkfile(cf);
932 commit_lock_file(&lk);
934 cleanup:
935 for (i = 0; i < ctx.nr; i++) {
936 if (ctx.info[i].p) {
937 close_pack(ctx.info[i].p);
938 free(ctx.info[i].p);
940 free(ctx.info[i].pack_name);
943 free(ctx.info);
944 free(ctx.entries);
945 free(ctx.pack_perm);
946 free(midx_name);
947 return result;
950 int write_midx_file(const char *object_dir, unsigned flags)
952 return write_midx_internal(object_dir, NULL, NULL, flags);
955 void clear_midx_file(struct repository *r)
957 char *midx = get_midx_filename(r->objects->odb->path);
959 if (r->objects && r->objects->multi_pack_index) {
960 close_midx(r->objects->multi_pack_index);
961 r->objects->multi_pack_index = NULL;
964 if (remove_path(midx))
965 die(_("failed to clear multi-pack-index at %s"), midx);
967 free(midx);
970 static int verify_midx_error;
972 static void midx_report(const char *fmt, ...)
974 va_list ap;
975 verify_midx_error = 1;
976 va_start(ap, fmt);
977 vfprintf(stderr, fmt, ap);
978 fprintf(stderr, "\n");
979 va_end(ap);
982 struct pair_pos_vs_id
984 uint32_t pos;
985 uint32_t pack_int_id;
988 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
990 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
991 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
993 return b->pack_int_id - a->pack_int_id;
997 * Limit calls to display_progress() for performance reasons.
998 * The interval here was arbitrarily chosen.
1000 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1001 #define midx_display_sparse_progress(progress, n) \
1002 do { \
1003 uint64_t _n = (n); \
1004 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1005 display_progress(progress, _n); \
1006 } while (0)
1008 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1010 struct pair_pos_vs_id *pairs = NULL;
1011 uint32_t i;
1012 struct progress *progress = NULL;
1013 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1014 verify_midx_error = 0;
1016 if (!m) {
1017 int result = 0;
1018 struct stat sb;
1019 char *filename = get_midx_filename(object_dir);
1020 if (!stat(filename, &sb)) {
1021 error(_("multi-pack-index file exists, but failed to parse"));
1022 result = 1;
1024 free(filename);
1025 return result;
1028 if (flags & MIDX_PROGRESS)
1029 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1030 m->num_packs);
1031 for (i = 0; i < m->num_packs; i++) {
1032 if (prepare_midx_pack(r, m, i))
1033 midx_report("failed to load pack in position %d", i);
1035 display_progress(progress, i + 1);
1037 stop_progress(&progress);
1039 for (i = 0; i < 255; i++) {
1040 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1041 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1043 if (oid_fanout1 > oid_fanout2)
1044 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1045 i, oid_fanout1, oid_fanout2, i + 1);
1048 if (m->num_objects == 0) {
1049 midx_report(_("the midx contains no oid"));
1051 * Remaining tests assume that we have objects, so we can
1052 * return here.
1054 return verify_midx_error;
1057 if (flags & MIDX_PROGRESS)
1058 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1059 m->num_objects - 1);
1060 for (i = 0; i < m->num_objects - 1; i++) {
1061 struct object_id oid1, oid2;
1063 nth_midxed_object_oid(&oid1, m, i);
1064 nth_midxed_object_oid(&oid2, m, i + 1);
1066 if (oidcmp(&oid1, &oid2) >= 0)
1067 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1068 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1070 midx_display_sparse_progress(progress, i + 1);
1072 stop_progress(&progress);
1075 * Create an array mapping each object to its packfile id. Sort it
1076 * to group the objects by packfile. Use this permutation to visit
1077 * each of the objects and only require 1 packfile to be open at a
1078 * time.
1080 ALLOC_ARRAY(pairs, m->num_objects);
1081 for (i = 0; i < m->num_objects; i++) {
1082 pairs[i].pos = i;
1083 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1086 if (flags & MIDX_PROGRESS)
1087 progress = start_sparse_progress(_("Sorting objects by packfile"),
1088 m->num_objects);
1089 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1090 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1091 stop_progress(&progress);
1093 if (flags & MIDX_PROGRESS)
1094 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1095 for (i = 0; i < m->num_objects; i++) {
1096 struct object_id oid;
1097 struct pack_entry e;
1098 off_t m_offset, p_offset;
1100 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1101 m->packs[pairs[i-1].pack_int_id])
1103 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1104 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1107 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1109 if (!fill_midx_entry(r, &oid, &e, m)) {
1110 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1111 pairs[i].pos, oid_to_hex(&oid));
1112 continue;
1115 if (open_pack_index(e.p)) {
1116 midx_report(_("failed to load pack-index for packfile %s"),
1117 e.p->pack_name);
1118 break;
1121 m_offset = e.offset;
1122 p_offset = find_pack_entry_one(oid.hash, e.p);
1124 if (m_offset != p_offset)
1125 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1126 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1128 midx_display_sparse_progress(progress, i + 1);
1130 stop_progress(&progress);
1132 free(pairs);
1134 return verify_midx_error;
1137 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1139 uint32_t i, *count, result = 0;
1140 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1141 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1142 struct progress *progress = NULL;
1144 if (!m)
1145 return 0;
1147 CALLOC_ARRAY(count, m->num_packs);
1149 if (flags & MIDX_PROGRESS)
1150 progress = start_delayed_progress(_("Counting referenced objects"),
1151 m->num_objects);
1152 for (i = 0; i < m->num_objects; i++) {
1153 int pack_int_id = nth_midxed_pack_int_id(m, i);
1154 count[pack_int_id]++;
1155 display_progress(progress, i + 1);
1157 stop_progress(&progress);
1159 if (flags & MIDX_PROGRESS)
1160 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1161 m->num_packs);
1162 for (i = 0; i < m->num_packs; i++) {
1163 char *pack_name;
1164 display_progress(progress, i + 1);
1166 if (count[i])
1167 continue;
1169 if (prepare_midx_pack(r, m, i))
1170 continue;
1172 if (m->packs[i]->pack_keep)
1173 continue;
1175 pack_name = xstrdup(m->packs[i]->pack_name);
1176 close_pack(m->packs[i]);
1178 string_list_insert(&packs_to_drop, m->pack_names[i]);
1179 unlink_pack_path(pack_name, 0);
1180 free(pack_name);
1182 stop_progress(&progress);
1184 free(count);
1186 if (packs_to_drop.nr)
1187 result = write_midx_internal(object_dir, m, &packs_to_drop, flags);
1189 string_list_clear(&packs_to_drop, 0);
1190 return result;
1193 struct repack_info {
1194 timestamp_t mtime;
1195 uint32_t referenced_objects;
1196 uint32_t pack_int_id;
1199 static int compare_by_mtime(const void *a_, const void *b_)
1201 const struct repack_info *a, *b;
1203 a = (const struct repack_info *)a_;
1204 b = (const struct repack_info *)b_;
1206 if (a->mtime < b->mtime)
1207 return -1;
1208 if (a->mtime > b->mtime)
1209 return 1;
1210 return 0;
1213 static int fill_included_packs_all(struct repository *r,
1214 struct multi_pack_index *m,
1215 unsigned char *include_pack)
1217 uint32_t i, count = 0;
1218 int pack_kept_objects = 0;
1220 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1222 for (i = 0; i < m->num_packs; i++) {
1223 if (prepare_midx_pack(r, m, i))
1224 continue;
1225 if (!pack_kept_objects && m->packs[i]->pack_keep)
1226 continue;
1228 include_pack[i] = 1;
1229 count++;
1232 return count < 2;
1235 static int fill_included_packs_batch(struct repository *r,
1236 struct multi_pack_index *m,
1237 unsigned char *include_pack,
1238 size_t batch_size)
1240 uint32_t i, packs_to_repack;
1241 size_t total_size;
1242 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1243 int pack_kept_objects = 0;
1245 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1247 for (i = 0; i < m->num_packs; i++) {
1248 pack_info[i].pack_int_id = i;
1250 if (prepare_midx_pack(r, m, i))
1251 continue;
1253 pack_info[i].mtime = m->packs[i]->mtime;
1256 for (i = 0; batch_size && i < m->num_objects; i++) {
1257 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1258 pack_info[pack_int_id].referenced_objects++;
1261 QSORT(pack_info, m->num_packs, compare_by_mtime);
1263 total_size = 0;
1264 packs_to_repack = 0;
1265 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1266 int pack_int_id = pack_info[i].pack_int_id;
1267 struct packed_git *p = m->packs[pack_int_id];
1268 size_t expected_size;
1270 if (!p)
1271 continue;
1272 if (!pack_kept_objects && p->pack_keep)
1273 continue;
1274 if (open_pack_index(p) || !p->num_objects)
1275 continue;
1277 expected_size = (size_t)(p->pack_size
1278 * pack_info[i].referenced_objects);
1279 expected_size /= p->num_objects;
1281 if (expected_size >= batch_size)
1282 continue;
1284 packs_to_repack++;
1285 total_size += expected_size;
1286 include_pack[pack_int_id] = 1;
1289 free(pack_info);
1291 if (packs_to_repack < 2)
1292 return 1;
1294 return 0;
1297 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1299 int result = 0;
1300 uint32_t i;
1301 unsigned char *include_pack;
1302 struct child_process cmd = CHILD_PROCESS_INIT;
1303 FILE *cmd_in;
1304 struct strbuf base_name = STRBUF_INIT;
1305 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1308 * When updating the default for these configuration
1309 * variables in builtin/repack.c, these must be adjusted
1310 * to match.
1312 int delta_base_offset = 1;
1313 int use_delta_islands = 0;
1315 if (!m)
1316 return 0;
1318 CALLOC_ARRAY(include_pack, m->num_packs);
1320 if (batch_size) {
1321 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1322 goto cleanup;
1323 } else if (fill_included_packs_all(r, m, include_pack))
1324 goto cleanup;
1326 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1327 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1329 strvec_push(&cmd.args, "pack-objects");
1331 strbuf_addstr(&base_name, object_dir);
1332 strbuf_addstr(&base_name, "/pack/pack");
1333 strvec_push(&cmd.args, base_name.buf);
1335 if (delta_base_offset)
1336 strvec_push(&cmd.args, "--delta-base-offset");
1337 if (use_delta_islands)
1338 strvec_push(&cmd.args, "--delta-islands");
1340 if (flags & MIDX_PROGRESS)
1341 strvec_push(&cmd.args, "--progress");
1342 else
1343 strvec_push(&cmd.args, "-q");
1345 strbuf_release(&base_name);
1347 cmd.git_cmd = 1;
1348 cmd.in = cmd.out = -1;
1350 if (start_command(&cmd)) {
1351 error(_("could not start pack-objects"));
1352 result = 1;
1353 goto cleanup;
1356 cmd_in = xfdopen(cmd.in, "w");
1358 for (i = 0; i < m->num_objects; i++) {
1359 struct object_id oid;
1360 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1362 if (!include_pack[pack_int_id])
1363 continue;
1365 nth_midxed_object_oid(&oid, m, i);
1366 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
1368 fclose(cmd_in);
1370 if (finish_command(&cmd)) {
1371 error(_("could not finish pack-objects"));
1372 result = 1;
1373 goto cleanup;
1376 result = write_midx_internal(object_dir, m, NULL, flags);
1377 m = NULL;
1379 cleanup:
1380 if (m)
1381 close_midx(m);
1382 free(include_pack);
1383 return result;