midx: allow marking a pack as preferred
[git/debian.git] / midx.c
blobb8f42bf31e48cfd84b411241446cd6ce40c2fc8a
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 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
149 m->packs = xcalloc(m->num_packs, sizeof(*m->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 static int idx_or_pack_name_cmp(const void *_va, const void *_vb)
436 const char *pack_name = _va;
437 const struct pack_info *compar = _vb;
439 return cmp_idx_or_pack_name(pack_name, compar->pack_name);
442 struct write_midx_context {
443 struct pack_info *info;
444 uint32_t nr;
445 uint32_t alloc;
446 struct multi_pack_index *m;
447 struct progress *progress;
448 unsigned pack_paths_checked;
450 struct pack_midx_entry *entries;
451 uint32_t entries_nr;
453 uint32_t *pack_perm;
454 unsigned large_offsets_needed:1;
455 uint32_t num_large_offsets;
457 int preferred_pack_idx;
460 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
461 const char *file_name, void *data)
463 struct write_midx_context *ctx = data;
465 if (ends_with(file_name, ".idx")) {
466 display_progress(ctx->progress, ++ctx->pack_paths_checked);
467 if (ctx->m && midx_contains_pack(ctx->m, file_name))
468 return;
470 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
472 ctx->info[ctx->nr].p = add_packed_git(full_path,
473 full_path_len,
476 if (!ctx->info[ctx->nr].p) {
477 warning(_("failed to add packfile '%s'"),
478 full_path);
479 return;
482 if (open_pack_index(ctx->info[ctx->nr].p)) {
483 warning(_("failed to open pack-index '%s'"),
484 full_path);
485 close_pack(ctx->info[ctx->nr].p);
486 FREE_AND_NULL(ctx->info[ctx->nr].p);
487 return;
490 ctx->info[ctx->nr].pack_name = xstrdup(file_name);
491 ctx->info[ctx->nr].orig_pack_int_id = ctx->nr;
492 ctx->info[ctx->nr].expired = 0;
493 ctx->nr++;
497 struct pack_midx_entry {
498 struct object_id oid;
499 uint32_t pack_int_id;
500 time_t pack_mtime;
501 uint64_t offset;
502 unsigned preferred : 1;
505 static int midx_oid_compare(const void *_a, const void *_b)
507 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
508 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
509 int cmp = oidcmp(&a->oid, &b->oid);
511 if (cmp)
512 return cmp;
514 /* Sort objects in a preferred pack first when multiple copies exist. */
515 if (a->preferred > b->preferred)
516 return -1;
517 if (a->preferred < b->preferred)
518 return 1;
520 if (a->pack_mtime > b->pack_mtime)
521 return -1;
522 else if (a->pack_mtime < b->pack_mtime)
523 return 1;
525 return a->pack_int_id - b->pack_int_id;
528 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
529 struct pack_midx_entry *e,
530 uint32_t pos)
532 if (pos >= m->num_objects)
533 return 1;
535 nth_midxed_object_oid(&e->oid, m, pos);
536 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
537 e->offset = nth_midxed_offset(m, pos);
539 /* consider objects in midx to be from "old" packs */
540 e->pack_mtime = 0;
541 return 0;
544 static void fill_pack_entry(uint32_t pack_int_id,
545 struct packed_git *p,
546 uint32_t cur_object,
547 struct pack_midx_entry *entry,
548 int preferred)
550 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
551 die(_("failed to locate object %d in packfile"), cur_object);
553 entry->pack_int_id = pack_int_id;
554 entry->pack_mtime = p->mtime;
556 entry->offset = nth_packed_object_offset(p, cur_object);
557 entry->preferred = !!preferred;
561 * It is possible to artificially get into a state where there are many
562 * duplicate copies of objects. That can create high memory pressure if
563 * we are to create a list of all objects before de-duplication. To reduce
564 * this memory pressure without a significant performance drop, automatically
565 * group objects by the first byte of their object id. Use the IDX fanout
566 * tables to group the data, copy to a local array, then sort.
568 * Copy only the de-duplicated entries (selected by most-recent modified time
569 * of a packfile containing the object).
571 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
572 struct pack_info *info,
573 uint32_t nr_packs,
574 uint32_t *nr_objects,
575 int preferred_pack)
577 uint32_t cur_fanout, cur_pack, cur_object;
578 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
579 struct pack_midx_entry *entries_by_fanout = NULL;
580 struct pack_midx_entry *deduplicated_entries = NULL;
581 uint32_t start_pack = m ? m->num_packs : 0;
583 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
584 total_objects += info[cur_pack].p->num_objects;
587 * As we de-duplicate by fanout value, we expect the fanout
588 * slices to be evenly distributed, with some noise. Hence,
589 * allocate slightly more than one 256th.
591 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
593 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
594 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
595 *nr_objects = 0;
597 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
598 uint32_t nr_fanout = 0;
600 if (m) {
601 uint32_t start = 0, end;
603 if (cur_fanout)
604 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
605 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
607 for (cur_object = start; cur_object < end; cur_object++) {
608 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
609 nth_midxed_pack_midx_entry(m,
610 &entries_by_fanout[nr_fanout],
611 cur_object);
612 if (nth_midxed_pack_int_id(m, cur_object) == preferred_pack)
613 entries_by_fanout[nr_fanout].preferred = 1;
614 else
615 entries_by_fanout[nr_fanout].preferred = 0;
616 nr_fanout++;
620 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
621 uint32_t start = 0, end;
622 int preferred = cur_pack == preferred_pack;
624 if (cur_fanout)
625 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
626 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
628 for (cur_object = start; cur_object < end; cur_object++) {
629 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
630 fill_pack_entry(cur_pack,
631 info[cur_pack].p,
632 cur_object,
633 &entries_by_fanout[nr_fanout],
634 preferred);
635 nr_fanout++;
639 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
642 * The batch is now sorted by OID and then mtime (descending).
643 * Take only the first duplicate.
645 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
646 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
647 &entries_by_fanout[cur_object].oid))
648 continue;
650 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
651 memcpy(&deduplicated_entries[*nr_objects],
652 &entries_by_fanout[cur_object],
653 sizeof(struct pack_midx_entry));
654 (*nr_objects)++;
658 free(entries_by_fanout);
659 return deduplicated_entries;
662 static int write_midx_pack_names(struct hashfile *f, void *data)
664 struct write_midx_context *ctx = data;
665 uint32_t i;
666 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
667 size_t written = 0;
669 for (i = 0; i < ctx->nr; i++) {
670 size_t writelen;
672 if (ctx->info[i].expired)
673 continue;
675 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
676 BUG("incorrect pack-file order: %s before %s",
677 ctx->info[i - 1].pack_name,
678 ctx->info[i].pack_name);
680 writelen = strlen(ctx->info[i].pack_name) + 1;
681 hashwrite(f, ctx->info[i].pack_name, writelen);
682 written += writelen;
685 /* add padding to be aligned */
686 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
687 if (i < MIDX_CHUNK_ALIGNMENT) {
688 memset(padding, 0, sizeof(padding));
689 hashwrite(f, padding, i);
692 return 0;
695 static int write_midx_oid_fanout(struct hashfile *f,
696 void *data)
698 struct write_midx_context *ctx = data;
699 struct pack_midx_entry *list = ctx->entries;
700 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
701 uint32_t count = 0;
702 uint32_t i;
705 * Write the first-level table (the list is sorted,
706 * but we use a 256-entry lookup to be able to avoid
707 * having to do eight extra binary search iterations).
709 for (i = 0; i < 256; i++) {
710 struct pack_midx_entry *next = list;
712 while (next < last && next->oid.hash[0] == i) {
713 count++;
714 next++;
717 hashwrite_be32(f, count);
718 list = next;
721 return 0;
724 static int write_midx_oid_lookup(struct hashfile *f,
725 void *data)
727 struct write_midx_context *ctx = data;
728 unsigned char hash_len = the_hash_algo->rawsz;
729 struct pack_midx_entry *list = ctx->entries;
730 uint32_t i;
732 for (i = 0; i < ctx->entries_nr; i++) {
733 struct pack_midx_entry *obj = list++;
735 if (i < ctx->entries_nr - 1) {
736 struct pack_midx_entry *next = list;
737 if (oidcmp(&obj->oid, &next->oid) >= 0)
738 BUG("OIDs not in order: %s >= %s",
739 oid_to_hex(&obj->oid),
740 oid_to_hex(&next->oid));
743 hashwrite(f, obj->oid.hash, (int)hash_len);
746 return 0;
749 static int write_midx_object_offsets(struct hashfile *f,
750 void *data)
752 struct write_midx_context *ctx = data;
753 struct pack_midx_entry *list = ctx->entries;
754 uint32_t i, nr_large_offset = 0;
756 for (i = 0; i < ctx->entries_nr; i++) {
757 struct pack_midx_entry *obj = list++;
759 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
760 BUG("object %s is in an expired pack with int-id %d",
761 oid_to_hex(&obj->oid),
762 obj->pack_int_id);
764 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
766 if (ctx->large_offsets_needed && obj->offset >> 31)
767 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
768 else if (!ctx->large_offsets_needed && obj->offset >> 32)
769 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
770 oid_to_hex(&obj->oid),
771 obj->offset);
772 else
773 hashwrite_be32(f, (uint32_t)obj->offset);
776 return 0;
779 static int write_midx_large_offsets(struct hashfile *f,
780 void *data)
782 struct write_midx_context *ctx = data;
783 struct pack_midx_entry *list = ctx->entries;
784 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
785 uint32_t nr_large_offset = ctx->num_large_offsets;
787 while (nr_large_offset) {
788 struct pack_midx_entry *obj;
789 uint64_t offset;
791 if (list >= end)
792 BUG("too many large-offset objects");
794 obj = list++;
795 offset = obj->offset;
797 if (!(offset >> 31))
798 continue;
800 hashwrite_be64(f, offset);
802 nr_large_offset--;
805 return 0;
808 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
809 struct string_list *packs_to_drop,
810 const char *preferred_pack_name,
811 unsigned flags)
813 char *midx_name;
814 uint32_t i;
815 struct hashfile *f = NULL;
816 struct lock_file lk;
817 struct write_midx_context ctx = { 0 };
818 int pack_name_concat_len = 0;
819 int dropped_packs = 0;
820 int result = 0;
821 struct chunkfile *cf;
823 midx_name = get_midx_filename(object_dir);
824 if (safe_create_leading_directories(midx_name))
825 die_errno(_("unable to create leading directories of %s"),
826 midx_name);
828 if (m)
829 ctx.m = m;
830 else
831 ctx.m = load_multi_pack_index(object_dir, 1);
833 ctx.nr = 0;
834 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
835 ctx.info = NULL;
836 ALLOC_ARRAY(ctx.info, ctx.alloc);
838 if (ctx.m) {
839 for (i = 0; i < ctx.m->num_packs; i++) {
840 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
842 ctx.info[ctx.nr].orig_pack_int_id = i;
843 ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]);
844 ctx.info[ctx.nr].p = NULL;
845 ctx.info[ctx.nr].expired = 0;
846 ctx.nr++;
850 ctx.pack_paths_checked = 0;
851 if (flags & MIDX_PROGRESS)
852 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
853 else
854 ctx.progress = NULL;
856 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
857 stop_progress(&ctx.progress);
859 if (ctx.m && ctx.nr == ctx.m->num_packs && !packs_to_drop)
860 goto cleanup;
862 ctx.preferred_pack_idx = -1;
863 if (preferred_pack_name) {
864 for (i = 0; i < ctx.nr; i++) {
865 if (!cmp_idx_or_pack_name(preferred_pack_name,
866 ctx.info[i].pack_name)) {
867 ctx.preferred_pack_idx = i;
868 break;
873 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr,
874 ctx.preferred_pack_idx);
876 ctx.large_offsets_needed = 0;
877 for (i = 0; i < ctx.entries_nr; i++) {
878 if (ctx.entries[i].offset > 0x7fffffff)
879 ctx.num_large_offsets++;
880 if (ctx.entries[i].offset > 0xffffffff)
881 ctx.large_offsets_needed = 1;
884 QSORT(ctx.info, ctx.nr, pack_info_compare);
886 if (packs_to_drop && packs_to_drop->nr) {
887 int drop_index = 0;
888 int missing_drops = 0;
890 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
891 int cmp = strcmp(ctx.info[i].pack_name,
892 packs_to_drop->items[drop_index].string);
894 if (!cmp) {
895 drop_index++;
896 ctx.info[i].expired = 1;
897 } else if (cmp > 0) {
898 error(_("did not see pack-file %s to drop"),
899 packs_to_drop->items[drop_index].string);
900 drop_index++;
901 missing_drops++;
902 i--;
903 } else {
904 ctx.info[i].expired = 0;
908 if (missing_drops) {
909 result = 1;
910 goto cleanup;
915 * pack_perm stores a permutation between pack-int-ids from the
916 * previous multi-pack-index to the new one we are writing:
918 * pack_perm[old_id] = new_id
920 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
921 for (i = 0; i < ctx.nr; i++) {
922 if (ctx.info[i].expired) {
923 dropped_packs++;
924 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
925 } else {
926 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
930 for (i = 0; i < ctx.nr; i++) {
931 if (!ctx.info[i].expired)
932 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
935 /* Check that the preferred pack wasn't expired (if given). */
936 if (preferred_pack_name) {
937 struct pack_info *preferred = bsearch(preferred_pack_name,
938 ctx.info, ctx.nr,
939 sizeof(*ctx.info),
940 idx_or_pack_name_cmp);
942 if (!preferred)
943 warning(_("unknown preferred pack: '%s'"),
944 preferred_pack_name);
945 else {
946 uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id];
947 if (perm == PACK_EXPIRED)
948 warning(_("preferred pack '%s' is expired"),
949 preferred_pack_name);
953 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
954 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
955 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
957 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
958 f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk));
959 FREE_AND_NULL(midx_name);
961 if (ctx.m)
962 close_midx(ctx.m);
964 if (ctx.nr - dropped_packs == 0) {
965 error(_("no pack files to index."));
966 result = 1;
967 goto cleanup;
970 cf = init_chunkfile(f);
972 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
973 write_midx_pack_names);
974 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
975 write_midx_oid_fanout);
976 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
977 (size_t)ctx.entries_nr * the_hash_algo->rawsz,
978 write_midx_oid_lookup);
979 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
980 (size_t)ctx.entries_nr * MIDX_CHUNK_OFFSET_WIDTH,
981 write_midx_object_offsets);
983 if (ctx.large_offsets_needed)
984 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
985 (size_t)ctx.num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH,
986 write_midx_large_offsets);
988 write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs);
989 write_chunkfile(cf, &ctx);
991 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
992 free_chunkfile(cf);
993 commit_lock_file(&lk);
995 cleanup:
996 for (i = 0; i < ctx.nr; i++) {
997 if (ctx.info[i].p) {
998 close_pack(ctx.info[i].p);
999 free(ctx.info[i].p);
1001 free(ctx.info[i].pack_name);
1004 free(ctx.info);
1005 free(ctx.entries);
1006 free(ctx.pack_perm);
1007 free(midx_name);
1008 return result;
1011 int write_midx_file(const char *object_dir,
1012 const char *preferred_pack_name,
1013 unsigned flags)
1015 return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name,
1016 flags);
1019 void clear_midx_file(struct repository *r)
1021 char *midx = get_midx_filename(r->objects->odb->path);
1023 if (r->objects && r->objects->multi_pack_index) {
1024 close_midx(r->objects->multi_pack_index);
1025 r->objects->multi_pack_index = NULL;
1028 if (remove_path(midx))
1029 die(_("failed to clear multi-pack-index at %s"), midx);
1031 free(midx);
1034 static int verify_midx_error;
1036 static void midx_report(const char *fmt, ...)
1038 va_list ap;
1039 verify_midx_error = 1;
1040 va_start(ap, fmt);
1041 vfprintf(stderr, fmt, ap);
1042 fprintf(stderr, "\n");
1043 va_end(ap);
1046 struct pair_pos_vs_id
1048 uint32_t pos;
1049 uint32_t pack_int_id;
1052 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1054 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1055 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1057 return b->pack_int_id - a->pack_int_id;
1061 * Limit calls to display_progress() for performance reasons.
1062 * The interval here was arbitrarily chosen.
1064 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1065 #define midx_display_sparse_progress(progress, n) \
1066 do { \
1067 uint64_t _n = (n); \
1068 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1069 display_progress(progress, _n); \
1070 } while (0)
1072 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1074 struct pair_pos_vs_id *pairs = NULL;
1075 uint32_t i;
1076 struct progress *progress = NULL;
1077 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1078 verify_midx_error = 0;
1080 if (!m) {
1081 int result = 0;
1082 struct stat sb;
1083 char *filename = get_midx_filename(object_dir);
1084 if (!stat(filename, &sb)) {
1085 error(_("multi-pack-index file exists, but failed to parse"));
1086 result = 1;
1088 free(filename);
1089 return result;
1092 if (flags & MIDX_PROGRESS)
1093 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1094 m->num_packs);
1095 for (i = 0; i < m->num_packs; i++) {
1096 if (prepare_midx_pack(r, m, i))
1097 midx_report("failed to load pack in position %d", i);
1099 display_progress(progress, i + 1);
1101 stop_progress(&progress);
1103 for (i = 0; i < 255; i++) {
1104 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1105 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1107 if (oid_fanout1 > oid_fanout2)
1108 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1109 i, oid_fanout1, oid_fanout2, i + 1);
1112 if (m->num_objects == 0) {
1113 midx_report(_("the midx contains no oid"));
1115 * Remaining tests assume that we have objects, so we can
1116 * return here.
1118 return verify_midx_error;
1121 if (flags & MIDX_PROGRESS)
1122 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1123 m->num_objects - 1);
1124 for (i = 0; i < m->num_objects - 1; i++) {
1125 struct object_id oid1, oid2;
1127 nth_midxed_object_oid(&oid1, m, i);
1128 nth_midxed_object_oid(&oid2, m, i + 1);
1130 if (oidcmp(&oid1, &oid2) >= 0)
1131 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1132 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1134 midx_display_sparse_progress(progress, i + 1);
1136 stop_progress(&progress);
1139 * Create an array mapping each object to its packfile id. Sort it
1140 * to group the objects by packfile. Use this permutation to visit
1141 * each of the objects and only require 1 packfile to be open at a
1142 * time.
1144 ALLOC_ARRAY(pairs, m->num_objects);
1145 for (i = 0; i < m->num_objects; i++) {
1146 pairs[i].pos = i;
1147 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1150 if (flags & MIDX_PROGRESS)
1151 progress = start_sparse_progress(_("Sorting objects by packfile"),
1152 m->num_objects);
1153 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1154 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1155 stop_progress(&progress);
1157 if (flags & MIDX_PROGRESS)
1158 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1159 for (i = 0; i < m->num_objects; i++) {
1160 struct object_id oid;
1161 struct pack_entry e;
1162 off_t m_offset, p_offset;
1164 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1165 m->packs[pairs[i-1].pack_int_id])
1167 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1168 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1171 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1173 if (!fill_midx_entry(r, &oid, &e, m)) {
1174 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1175 pairs[i].pos, oid_to_hex(&oid));
1176 continue;
1179 if (open_pack_index(e.p)) {
1180 midx_report(_("failed to load pack-index for packfile %s"),
1181 e.p->pack_name);
1182 break;
1185 m_offset = e.offset;
1186 p_offset = find_pack_entry_one(oid.hash, e.p);
1188 if (m_offset != p_offset)
1189 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1190 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1192 midx_display_sparse_progress(progress, i + 1);
1194 stop_progress(&progress);
1196 free(pairs);
1198 return verify_midx_error;
1201 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1203 uint32_t i, *count, result = 0;
1204 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1205 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1206 struct progress *progress = NULL;
1208 if (!m)
1209 return 0;
1211 count = xcalloc(m->num_packs, sizeof(uint32_t));
1213 if (flags & MIDX_PROGRESS)
1214 progress = start_delayed_progress(_("Counting referenced objects"),
1215 m->num_objects);
1216 for (i = 0; i < m->num_objects; i++) {
1217 int pack_int_id = nth_midxed_pack_int_id(m, i);
1218 count[pack_int_id]++;
1219 display_progress(progress, i + 1);
1221 stop_progress(&progress);
1223 if (flags & MIDX_PROGRESS)
1224 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1225 m->num_packs);
1226 for (i = 0; i < m->num_packs; i++) {
1227 char *pack_name;
1228 display_progress(progress, i + 1);
1230 if (count[i])
1231 continue;
1233 if (prepare_midx_pack(r, m, i))
1234 continue;
1236 if (m->packs[i]->pack_keep)
1237 continue;
1239 pack_name = xstrdup(m->packs[i]->pack_name);
1240 close_pack(m->packs[i]);
1242 string_list_insert(&packs_to_drop, m->pack_names[i]);
1243 unlink_pack_path(pack_name, 0);
1244 free(pack_name);
1246 stop_progress(&progress);
1248 free(count);
1250 if (packs_to_drop.nr)
1251 result = write_midx_internal(object_dir, m, &packs_to_drop, NULL, flags);
1253 string_list_clear(&packs_to_drop, 0);
1254 return result;
1257 struct repack_info {
1258 timestamp_t mtime;
1259 uint32_t referenced_objects;
1260 uint32_t pack_int_id;
1263 static int compare_by_mtime(const void *a_, const void *b_)
1265 const struct repack_info *a, *b;
1267 a = (const struct repack_info *)a_;
1268 b = (const struct repack_info *)b_;
1270 if (a->mtime < b->mtime)
1271 return -1;
1272 if (a->mtime > b->mtime)
1273 return 1;
1274 return 0;
1277 static int fill_included_packs_all(struct repository *r,
1278 struct multi_pack_index *m,
1279 unsigned char *include_pack)
1281 uint32_t i, count = 0;
1282 int pack_kept_objects = 0;
1284 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1286 for (i = 0; i < m->num_packs; i++) {
1287 if (prepare_midx_pack(r, m, i))
1288 continue;
1289 if (!pack_kept_objects && m->packs[i]->pack_keep)
1290 continue;
1292 include_pack[i] = 1;
1293 count++;
1296 return count < 2;
1299 static int fill_included_packs_batch(struct repository *r,
1300 struct multi_pack_index *m,
1301 unsigned char *include_pack,
1302 size_t batch_size)
1304 uint32_t i, packs_to_repack;
1305 size_t total_size;
1306 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1307 int pack_kept_objects = 0;
1309 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1311 for (i = 0; i < m->num_packs; i++) {
1312 pack_info[i].pack_int_id = i;
1314 if (prepare_midx_pack(r, m, i))
1315 continue;
1317 pack_info[i].mtime = m->packs[i]->mtime;
1320 for (i = 0; batch_size && i < m->num_objects; i++) {
1321 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1322 pack_info[pack_int_id].referenced_objects++;
1325 QSORT(pack_info, m->num_packs, compare_by_mtime);
1327 total_size = 0;
1328 packs_to_repack = 0;
1329 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1330 int pack_int_id = pack_info[i].pack_int_id;
1331 struct packed_git *p = m->packs[pack_int_id];
1332 size_t expected_size;
1334 if (!p)
1335 continue;
1336 if (!pack_kept_objects && p->pack_keep)
1337 continue;
1338 if (open_pack_index(p) || !p->num_objects)
1339 continue;
1341 expected_size = (size_t)(p->pack_size
1342 * pack_info[i].referenced_objects);
1343 expected_size /= p->num_objects;
1345 if (expected_size >= batch_size)
1346 continue;
1348 packs_to_repack++;
1349 total_size += expected_size;
1350 include_pack[pack_int_id] = 1;
1353 free(pack_info);
1355 if (packs_to_repack < 2)
1356 return 1;
1358 return 0;
1361 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1363 int result = 0;
1364 uint32_t i;
1365 unsigned char *include_pack;
1366 struct child_process cmd = CHILD_PROCESS_INIT;
1367 FILE *cmd_in;
1368 struct strbuf base_name = STRBUF_INIT;
1369 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1372 * When updating the default for these configuration
1373 * variables in builtin/repack.c, these must be adjusted
1374 * to match.
1376 int delta_base_offset = 1;
1377 int use_delta_islands = 0;
1379 if (!m)
1380 return 0;
1382 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1384 if (batch_size) {
1385 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1386 goto cleanup;
1387 } else if (fill_included_packs_all(r, m, include_pack))
1388 goto cleanup;
1390 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1391 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1393 strvec_push(&cmd.args, "pack-objects");
1395 strbuf_addstr(&base_name, object_dir);
1396 strbuf_addstr(&base_name, "/pack/pack");
1397 strvec_push(&cmd.args, base_name.buf);
1399 if (delta_base_offset)
1400 strvec_push(&cmd.args, "--delta-base-offset");
1401 if (use_delta_islands)
1402 strvec_push(&cmd.args, "--delta-islands");
1404 if (flags & MIDX_PROGRESS)
1405 strvec_push(&cmd.args, "--progress");
1406 else
1407 strvec_push(&cmd.args, "-q");
1409 strbuf_release(&base_name);
1411 cmd.git_cmd = 1;
1412 cmd.in = cmd.out = -1;
1414 if (start_command(&cmd)) {
1415 error(_("could not start pack-objects"));
1416 result = 1;
1417 goto cleanup;
1420 cmd_in = xfdopen(cmd.in, "w");
1422 for (i = 0; i < m->num_objects; i++) {
1423 struct object_id oid;
1424 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1426 if (!include_pack[pack_int_id])
1427 continue;
1429 nth_midxed_object_oid(&oid, m, i);
1430 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
1432 fclose(cmd_in);
1434 if (finish_command(&cmd)) {
1435 error(_("could not finish pack-objects"));
1436 result = 1;
1437 goto cleanup;
1440 result = write_midx_internal(object_dir, m, NULL, NULL, flags);
1441 m = NULL;
1443 cleanup:
1444 if (m)
1445 close_midx(m);
1446 free(include_pack);
1447 return result;