http: refactor finish_http_pack_request()
[git.git] / midx.c
blob6d1584ca51d313343bbb65f9baa7c794259c7bf0
1 #include "cache.h"
2 #include "config.h"
3 #include "csum-file.h"
4 #include "dir.h"
5 #include "lockfile.h"
6 #include "packfile.h"
7 #include "object-store.h"
8 #include "sha1-lookup.h"
9 #include "midx.h"
10 #include "progress.h"
11 #include "trace2.h"
12 #include "run-command.h"
14 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
15 #define MIDX_VERSION 1
16 #define MIDX_BYTE_FILE_VERSION 4
17 #define MIDX_BYTE_HASH_VERSION 5
18 #define MIDX_BYTE_NUM_CHUNKS 6
19 #define MIDX_BYTE_NUM_PACKS 8
20 #define MIDX_HASH_VERSION 1
21 #define MIDX_HEADER_SIZE 12
22 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
24 #define MIDX_MAX_CHUNKS 5
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_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
32 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
33 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
34 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
35 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
37 #define PACK_EXPIRED UINT_MAX
39 static char *get_midx_filename(const char *object_dir)
41 return xstrfmt("%s/pack/multi-pack-index", object_dir);
44 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
46 struct multi_pack_index *m = NULL;
47 int fd;
48 struct stat st;
49 size_t midx_size;
50 void *midx_map = NULL;
51 uint32_t hash_version;
52 char *midx_name = get_midx_filename(object_dir);
53 uint32_t i;
54 const char *cur_pack_name;
56 fd = git_open(midx_name);
58 if (fd < 0)
59 goto cleanup_fail;
60 if (fstat(fd, &st)) {
61 error_errno(_("failed to read %s"), midx_name);
62 goto cleanup_fail;
65 midx_size = xsize_t(st.st_size);
67 if (midx_size < MIDX_MIN_SIZE) {
68 error(_("multi-pack-index file %s is too small"), midx_name);
69 goto cleanup_fail;
72 FREE_AND_NULL(midx_name);
74 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
75 close(fd);
77 FLEX_ALLOC_STR(m, object_dir, object_dir);
78 m->data = midx_map;
79 m->data_len = midx_size;
80 m->local = local;
82 m->signature = get_be32(m->data);
83 if (m->signature != MIDX_SIGNATURE)
84 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
85 m->signature, MIDX_SIGNATURE);
87 m->version = m->data[MIDX_BYTE_FILE_VERSION];
88 if (m->version != MIDX_VERSION)
89 die(_("multi-pack-index version %d not recognized"),
90 m->version);
92 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
93 if (hash_version != MIDX_HASH_VERSION)
94 die(_("hash version %u does not match"), hash_version);
95 m->hash_len = the_hash_algo->rawsz;
97 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
99 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
101 for (i = 0; i < m->num_chunks; i++) {
102 uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
103 MIDX_CHUNKLOOKUP_WIDTH * i);
104 uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
105 MIDX_CHUNKLOOKUP_WIDTH * i);
107 if (chunk_offset >= m->data_len)
108 die(_("invalid chunk offset (too large)"));
110 switch (chunk_id) {
111 case MIDX_CHUNKID_PACKNAMES:
112 m->chunk_pack_names = m->data + chunk_offset;
113 break;
115 case MIDX_CHUNKID_OIDFANOUT:
116 m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
117 break;
119 case MIDX_CHUNKID_OIDLOOKUP:
120 m->chunk_oid_lookup = m->data + chunk_offset;
121 break;
123 case MIDX_CHUNKID_OBJECTOFFSETS:
124 m->chunk_object_offsets = m->data + chunk_offset;
125 break;
127 case MIDX_CHUNKID_LARGEOFFSETS:
128 m->chunk_large_offsets = m->data + chunk_offset;
129 break;
131 case 0:
132 die(_("terminating multi-pack-index chunk id appears earlier than expected"));
133 break;
135 default:
137 * Do nothing on unrecognized chunks, allowing future
138 * extensions to add optional chunks.
140 break;
144 if (!m->chunk_pack_names)
145 die(_("multi-pack-index missing required pack-name chunk"));
146 if (!m->chunk_oid_fanout)
147 die(_("multi-pack-index missing required OID fanout chunk"));
148 if (!m->chunk_oid_lookup)
149 die(_("multi-pack-index missing required OID lookup chunk"));
150 if (!m->chunk_object_offsets)
151 die(_("multi-pack-index missing required object offsets chunk"));
153 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
155 m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
156 m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
158 cur_pack_name = (const char *)m->chunk_pack_names;
159 for (i = 0; i < m->num_packs; i++) {
160 m->pack_names[i] = cur_pack_name;
162 cur_pack_name += strlen(cur_pack_name) + 1;
164 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
165 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
166 m->pack_names[i - 1],
167 m->pack_names[i]);
170 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
171 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
173 return m;
175 cleanup_fail:
176 free(m);
177 free(midx_name);
178 if (midx_map)
179 munmap(midx_map, midx_size);
180 if (0 <= fd)
181 close(fd);
182 return NULL;
185 void close_midx(struct multi_pack_index *m)
187 uint32_t i;
189 if (!m)
190 return;
192 munmap((unsigned char *)m->data, m->data_len);
194 for (i = 0; i < m->num_packs; i++) {
195 if (m->packs[i])
196 m->packs[i]->multi_pack_index = 0;
198 FREE_AND_NULL(m->packs);
199 FREE_AND_NULL(m->pack_names);
202 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
204 struct strbuf pack_name = STRBUF_INIT;
205 struct packed_git *p;
207 if (pack_int_id >= m->num_packs)
208 die(_("bad pack-int-id: %u (%u total packs)"),
209 pack_int_id, m->num_packs);
211 if (m->packs[pack_int_id])
212 return 0;
214 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
215 m->pack_names[pack_int_id]);
217 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
218 strbuf_release(&pack_name);
220 if (!p)
221 return 1;
223 p->multi_pack_index = 1;
224 m->packs[pack_int_id] = p;
225 install_packed_git(r, p);
226 list_add_tail(&p->mru, &r->objects->packed_git_mru);
228 return 0;
231 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
233 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
234 the_hash_algo->rawsz, result);
237 struct object_id *nth_midxed_object_oid(struct object_id *oid,
238 struct multi_pack_index *m,
239 uint32_t n)
241 if (n >= m->num_objects)
242 return NULL;
244 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
245 return oid;
248 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
250 const unsigned char *offset_data;
251 uint32_t offset32;
253 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
254 offset32 = get_be32(offset_data + sizeof(uint32_t));
256 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
257 if (sizeof(off_t) < sizeof(uint64_t))
258 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
260 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
261 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
264 return offset32;
267 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
269 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
272 static int nth_midxed_pack_entry(struct repository *r,
273 struct multi_pack_index *m,
274 struct pack_entry *e,
275 uint32_t pos)
277 uint32_t pack_int_id;
278 struct packed_git *p;
280 if (pos >= m->num_objects)
281 return 0;
283 pack_int_id = nth_midxed_pack_int_id(m, pos);
285 if (prepare_midx_pack(r, m, pack_int_id))
286 die(_("error preparing packfile from multi-pack-index"));
287 p = m->packs[pack_int_id];
290 * We are about to tell the caller where they can locate the
291 * requested object. We better make sure the packfile is
292 * still here and can be accessed before supplying that
293 * answer, as it may have been deleted since the MIDX was
294 * loaded!
296 if (!is_pack_valid(p))
297 return 0;
299 if (p->num_bad_objects) {
300 uint32_t i;
301 struct object_id oid;
302 nth_midxed_object_oid(&oid, m, pos);
303 for (i = 0; i < p->num_bad_objects; i++)
304 if (hasheq(oid.hash,
305 p->bad_object_sha1 + the_hash_algo->rawsz * i))
306 return 0;
309 e->offset = nth_midxed_offset(m, pos);
310 e->p = p;
312 return 1;
315 int fill_midx_entry(struct repository * r,
316 const struct object_id *oid,
317 struct pack_entry *e,
318 struct multi_pack_index *m)
320 uint32_t pos;
322 if (!bsearch_midx(oid, m, &pos))
323 return 0;
325 return nth_midxed_pack_entry(r, m, e, pos);
328 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
329 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
330 const char *idx_name)
332 /* Skip past any initial matching prefix. */
333 while (*idx_name && *idx_name == *idx_or_pack_name) {
334 idx_name++;
335 idx_or_pack_name++;
339 * If we didn't match completely, we may have matched "pack-1234." and
340 * be left with "idx" and "pack" respectively, which is also OK. We do
341 * not have to check for "idx" and "idx", because that would have been
342 * a complete match (and in that case these strcmps will be false, but
343 * we'll correctly return 0 from the final strcmp() below.
345 * Technically this matches "fooidx" and "foopack", but we'd never have
346 * such names in the first place.
348 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
349 return 0;
352 * This not only checks for a complete match, but also orders based on
353 * the first non-identical character, which means our ordering will
354 * match a raw strcmp(). That makes it OK to use this to binary search
355 * a naively-sorted list.
357 return strcmp(idx_or_pack_name, idx_name);
360 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
362 uint32_t first = 0, last = m->num_packs;
364 while (first < last) {
365 uint32_t mid = first + (last - first) / 2;
366 const char *current;
367 int cmp;
369 current = m->pack_names[mid];
370 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
371 if (!cmp)
372 return 1;
373 if (cmp > 0) {
374 first = mid + 1;
375 continue;
377 last = mid;
380 return 0;
383 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
385 struct multi_pack_index *m;
386 struct multi_pack_index *m_search;
387 int config_value;
388 static int env_value = -1;
390 if (env_value < 0)
391 env_value = git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0);
393 if (!env_value &&
394 (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
395 !config_value))
396 return 0;
398 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
399 if (!strcmp(object_dir, m_search->object_dir))
400 return 1;
402 m = load_multi_pack_index(object_dir, local);
404 if (m) {
405 m->next = r->objects->multi_pack_index;
406 r->objects->multi_pack_index = m;
407 return 1;
410 return 0;
413 static size_t write_midx_header(struct hashfile *f,
414 unsigned char num_chunks,
415 uint32_t num_packs)
417 unsigned char byte_values[4];
419 hashwrite_be32(f, MIDX_SIGNATURE);
420 byte_values[0] = MIDX_VERSION;
421 byte_values[1] = MIDX_HASH_VERSION;
422 byte_values[2] = num_chunks;
423 byte_values[3] = 0; /* unused */
424 hashwrite(f, byte_values, sizeof(byte_values));
425 hashwrite_be32(f, num_packs);
427 return MIDX_HEADER_SIZE;
430 struct pack_info {
431 uint32_t orig_pack_int_id;
432 char *pack_name;
433 struct packed_git *p;
434 unsigned expired : 1;
437 static int pack_info_compare(const void *_a, const void *_b)
439 struct pack_info *a = (struct pack_info *)_a;
440 struct pack_info *b = (struct pack_info *)_b;
441 return strcmp(a->pack_name, b->pack_name);
444 struct pack_list {
445 struct pack_info *info;
446 uint32_t nr;
447 uint32_t alloc;
448 struct multi_pack_index *m;
449 struct progress *progress;
450 unsigned pack_paths_checked;
453 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
454 const char *file_name, void *data)
456 struct pack_list *packs = (struct pack_list *)data;
458 if (ends_with(file_name, ".idx")) {
459 display_progress(packs->progress, ++packs->pack_paths_checked);
460 if (packs->m && midx_contains_pack(packs->m, file_name))
461 return;
463 ALLOC_GROW(packs->info, packs->nr + 1, packs->alloc);
465 packs->info[packs->nr].p = add_packed_git(full_path,
466 full_path_len,
469 if (!packs->info[packs->nr].p) {
470 warning(_("failed to add packfile '%s'"),
471 full_path);
472 return;
475 if (open_pack_index(packs->info[packs->nr].p)) {
476 warning(_("failed to open pack-index '%s'"),
477 full_path);
478 close_pack(packs->info[packs->nr].p);
479 FREE_AND_NULL(packs->info[packs->nr].p);
480 return;
483 packs->info[packs->nr].pack_name = xstrdup(file_name);
484 packs->info[packs->nr].orig_pack_int_id = packs->nr;
485 packs->info[packs->nr].expired = 0;
486 packs->nr++;
490 struct pack_midx_entry {
491 struct object_id oid;
492 uint32_t pack_int_id;
493 time_t pack_mtime;
494 uint64_t offset;
497 static int midx_oid_compare(const void *_a, const void *_b)
499 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
500 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
501 int cmp = oidcmp(&a->oid, &b->oid);
503 if (cmp)
504 return cmp;
506 if (a->pack_mtime > b->pack_mtime)
507 return -1;
508 else if (a->pack_mtime < b->pack_mtime)
509 return 1;
511 return a->pack_int_id - b->pack_int_id;
514 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
515 struct pack_midx_entry *e,
516 uint32_t pos)
518 if (pos >= m->num_objects)
519 return 1;
521 nth_midxed_object_oid(&e->oid, m, pos);
522 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
523 e->offset = nth_midxed_offset(m, pos);
525 /* consider objects in midx to be from "old" packs */
526 e->pack_mtime = 0;
527 return 0;
530 static void fill_pack_entry(uint32_t pack_int_id,
531 struct packed_git *p,
532 uint32_t cur_object,
533 struct pack_midx_entry *entry)
535 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
536 die(_("failed to locate object %d in packfile"), cur_object);
538 entry->pack_int_id = pack_int_id;
539 entry->pack_mtime = p->mtime;
541 entry->offset = nth_packed_object_offset(p, cur_object);
545 * It is possible to artificially get into a state where there are many
546 * duplicate copies of objects. That can create high memory pressure if
547 * we are to create a list of all objects before de-duplication. To reduce
548 * this memory pressure without a significant performance drop, automatically
549 * group objects by the first byte of their object id. Use the IDX fanout
550 * tables to group the data, copy to a local array, then sort.
552 * Copy only the de-duplicated entries (selected by most-recent modified time
553 * of a packfile containing the object).
555 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
556 struct pack_info *info,
557 uint32_t nr_packs,
558 uint32_t *nr_objects)
560 uint32_t cur_fanout, cur_pack, cur_object;
561 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
562 struct pack_midx_entry *entries_by_fanout = NULL;
563 struct pack_midx_entry *deduplicated_entries = NULL;
564 uint32_t start_pack = m ? m->num_packs : 0;
566 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
567 total_objects += info[cur_pack].p->num_objects;
570 * As we de-duplicate by fanout value, we expect the fanout
571 * slices to be evenly distributed, with some noise. Hence,
572 * allocate slightly more than one 256th.
574 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
576 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
577 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
578 *nr_objects = 0;
580 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
581 uint32_t nr_fanout = 0;
583 if (m) {
584 uint32_t start = 0, end;
586 if (cur_fanout)
587 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
588 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
590 for (cur_object = start; cur_object < end; cur_object++) {
591 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
592 nth_midxed_pack_midx_entry(m,
593 &entries_by_fanout[nr_fanout],
594 cur_object);
595 nr_fanout++;
599 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
600 uint32_t start = 0, end;
602 if (cur_fanout)
603 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
604 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
606 for (cur_object = start; cur_object < end; cur_object++) {
607 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
608 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
609 nr_fanout++;
613 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
616 * The batch is now sorted by OID and then mtime (descending).
617 * Take only the first duplicate.
619 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
620 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
621 &entries_by_fanout[cur_object].oid))
622 continue;
624 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
625 memcpy(&deduplicated_entries[*nr_objects],
626 &entries_by_fanout[cur_object],
627 sizeof(struct pack_midx_entry));
628 (*nr_objects)++;
632 free(entries_by_fanout);
633 return deduplicated_entries;
636 static size_t write_midx_pack_names(struct hashfile *f,
637 struct pack_info *info,
638 uint32_t num_packs)
640 uint32_t i;
641 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
642 size_t written = 0;
644 for (i = 0; i < num_packs; i++) {
645 size_t writelen;
647 if (info[i].expired)
648 continue;
650 if (i && strcmp(info[i].pack_name, info[i - 1].pack_name) <= 0)
651 BUG("incorrect pack-file order: %s before %s",
652 info[i - 1].pack_name,
653 info[i].pack_name);
655 writelen = strlen(info[i].pack_name) + 1;
656 hashwrite(f, info[i].pack_name, writelen);
657 written += writelen;
660 /* add padding to be aligned */
661 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
662 if (i < MIDX_CHUNK_ALIGNMENT) {
663 memset(padding, 0, sizeof(padding));
664 hashwrite(f, padding, i);
665 written += i;
668 return written;
671 static size_t write_midx_oid_fanout(struct hashfile *f,
672 struct pack_midx_entry *objects,
673 uint32_t nr_objects)
675 struct pack_midx_entry *list = objects;
676 struct pack_midx_entry *last = objects + nr_objects;
677 uint32_t count = 0;
678 uint32_t i;
681 * Write the first-level table (the list is sorted,
682 * but we use a 256-entry lookup to be able to avoid
683 * having to do eight extra binary search iterations).
685 for (i = 0; i < 256; i++) {
686 struct pack_midx_entry *next = list;
688 while (next < last && next->oid.hash[0] == i) {
689 count++;
690 next++;
693 hashwrite_be32(f, count);
694 list = next;
697 return MIDX_CHUNK_FANOUT_SIZE;
700 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
701 struct pack_midx_entry *objects,
702 uint32_t nr_objects)
704 struct pack_midx_entry *list = objects;
705 uint32_t i;
706 size_t written = 0;
708 for (i = 0; i < nr_objects; i++) {
709 struct pack_midx_entry *obj = list++;
711 if (i < nr_objects - 1) {
712 struct pack_midx_entry *next = list;
713 if (oidcmp(&obj->oid, &next->oid) >= 0)
714 BUG("OIDs not in order: %s >= %s",
715 oid_to_hex(&obj->oid),
716 oid_to_hex(&next->oid));
719 hashwrite(f, obj->oid.hash, (int)hash_len);
720 written += hash_len;
723 return written;
726 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
727 uint32_t *perm,
728 struct pack_midx_entry *objects, uint32_t nr_objects)
730 struct pack_midx_entry *list = objects;
731 uint32_t i, nr_large_offset = 0;
732 size_t written = 0;
734 for (i = 0; i < nr_objects; i++) {
735 struct pack_midx_entry *obj = list++;
737 if (perm[obj->pack_int_id] == PACK_EXPIRED)
738 BUG("object %s is in an expired pack with int-id %d",
739 oid_to_hex(&obj->oid),
740 obj->pack_int_id);
742 hashwrite_be32(f, perm[obj->pack_int_id]);
744 if (large_offset_needed && obj->offset >> 31)
745 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
746 else if (!large_offset_needed && obj->offset >> 32)
747 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
748 oid_to_hex(&obj->oid),
749 obj->offset);
750 else
751 hashwrite_be32(f, (uint32_t)obj->offset);
753 written += MIDX_CHUNK_OFFSET_WIDTH;
756 return written;
759 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
760 struct pack_midx_entry *objects, uint32_t nr_objects)
762 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
763 size_t written = 0;
765 while (nr_large_offset) {
766 struct pack_midx_entry *obj;
767 uint64_t offset;
769 if (list >= end)
770 BUG("too many large-offset objects");
772 obj = list++;
773 offset = obj->offset;
775 if (!(offset >> 31))
776 continue;
778 hashwrite_be32(f, offset >> 32);
779 hashwrite_be32(f, offset & 0xffffffffUL);
780 written += 2 * sizeof(uint32_t);
782 nr_large_offset--;
785 return written;
788 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
789 struct string_list *packs_to_drop, unsigned flags)
791 unsigned char cur_chunk, num_chunks = 0;
792 char *midx_name;
793 uint32_t i;
794 struct hashfile *f = NULL;
795 struct lock_file lk;
796 struct pack_list packs;
797 uint32_t *pack_perm = NULL;
798 uint64_t written = 0;
799 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
800 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
801 uint32_t nr_entries, num_large_offsets = 0;
802 struct pack_midx_entry *entries = NULL;
803 struct progress *progress = NULL;
804 int large_offsets_needed = 0;
805 int pack_name_concat_len = 0;
806 int dropped_packs = 0;
807 int result = 0;
809 midx_name = get_midx_filename(object_dir);
810 if (safe_create_leading_directories(midx_name)) {
811 UNLEAK(midx_name);
812 die_errno(_("unable to create leading directories of %s"),
813 midx_name);
816 if (m)
817 packs.m = m;
818 else
819 packs.m = load_multi_pack_index(object_dir, 1);
821 packs.nr = 0;
822 packs.alloc = packs.m ? packs.m->num_packs : 16;
823 packs.info = NULL;
824 ALLOC_ARRAY(packs.info, packs.alloc);
826 if (packs.m) {
827 for (i = 0; i < packs.m->num_packs; i++) {
828 ALLOC_GROW(packs.info, packs.nr + 1, packs.alloc);
830 packs.info[packs.nr].orig_pack_int_id = i;
831 packs.info[packs.nr].pack_name = xstrdup(packs.m->pack_names[i]);
832 packs.info[packs.nr].p = NULL;
833 packs.info[packs.nr].expired = 0;
834 packs.nr++;
838 packs.pack_paths_checked = 0;
839 if (flags & MIDX_PROGRESS)
840 packs.progress = start_progress(_("Adding packfiles to multi-pack-index"), 0);
841 else
842 packs.progress = NULL;
844 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
845 stop_progress(&packs.progress);
847 if (packs.m && packs.nr == packs.m->num_packs && !packs_to_drop)
848 goto cleanup;
850 entries = get_sorted_entries(packs.m, packs.info, packs.nr, &nr_entries);
852 for (i = 0; i < nr_entries; i++) {
853 if (entries[i].offset > 0x7fffffff)
854 num_large_offsets++;
855 if (entries[i].offset > 0xffffffff)
856 large_offsets_needed = 1;
859 QSORT(packs.info, packs.nr, pack_info_compare);
861 if (packs_to_drop && packs_to_drop->nr) {
862 int drop_index = 0;
863 int missing_drops = 0;
865 for (i = 0; i < packs.nr && drop_index < packs_to_drop->nr; i++) {
866 int cmp = strcmp(packs.info[i].pack_name,
867 packs_to_drop->items[drop_index].string);
869 if (!cmp) {
870 drop_index++;
871 packs.info[i].expired = 1;
872 } else if (cmp > 0) {
873 error(_("did not see pack-file %s to drop"),
874 packs_to_drop->items[drop_index].string);
875 drop_index++;
876 missing_drops++;
877 i--;
878 } else {
879 packs.info[i].expired = 0;
883 if (missing_drops) {
884 result = 1;
885 goto cleanup;
890 * pack_perm stores a permutation between pack-int-ids from the
891 * previous multi-pack-index to the new one we are writing:
893 * pack_perm[old_id] = new_id
895 ALLOC_ARRAY(pack_perm, packs.nr);
896 for (i = 0; i < packs.nr; i++) {
897 if (packs.info[i].expired) {
898 dropped_packs++;
899 pack_perm[packs.info[i].orig_pack_int_id] = PACK_EXPIRED;
900 } else {
901 pack_perm[packs.info[i].orig_pack_int_id] = i - dropped_packs;
905 for (i = 0; i < packs.nr; i++) {
906 if (!packs.info[i].expired)
907 pack_name_concat_len += strlen(packs.info[i].pack_name) + 1;
910 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
911 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
912 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
914 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
915 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
916 FREE_AND_NULL(midx_name);
918 if (packs.m)
919 close_midx(packs.m);
921 cur_chunk = 0;
922 num_chunks = large_offsets_needed ? 5 : 4;
924 if (packs.nr - dropped_packs == 0) {
925 error(_("no pack files to index."));
926 result = 1;
927 goto cleanup;
930 written = write_midx_header(f, num_chunks, packs.nr - dropped_packs);
932 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
933 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
935 cur_chunk++;
936 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
937 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + pack_name_concat_len;
939 cur_chunk++;
940 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
941 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
943 cur_chunk++;
944 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
945 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * the_hash_algo->rawsz;
947 cur_chunk++;
948 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
949 if (large_offsets_needed) {
950 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
952 cur_chunk++;
953 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
954 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
957 chunk_ids[cur_chunk] = 0;
959 for (i = 0; i <= num_chunks; i++) {
960 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
961 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
962 chunk_offsets[i - 1],
963 chunk_offsets[i]);
965 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
966 BUG("chunk offset %"PRIu64" is not properly aligned",
967 chunk_offsets[i]);
969 hashwrite_be32(f, chunk_ids[i]);
970 hashwrite_be32(f, chunk_offsets[i] >> 32);
971 hashwrite_be32(f, chunk_offsets[i]);
973 written += MIDX_CHUNKLOOKUP_WIDTH;
976 if (flags & MIDX_PROGRESS)
977 progress = start_progress(_("Writing chunks to multi-pack-index"),
978 num_chunks);
979 for (i = 0; i < num_chunks; i++) {
980 if (written != chunk_offsets[i])
981 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
982 chunk_offsets[i],
983 written,
984 chunk_ids[i]);
986 switch (chunk_ids[i]) {
987 case MIDX_CHUNKID_PACKNAMES:
988 written += write_midx_pack_names(f, packs.info, packs.nr);
989 break;
991 case MIDX_CHUNKID_OIDFANOUT:
992 written += write_midx_oid_fanout(f, entries, nr_entries);
993 break;
995 case MIDX_CHUNKID_OIDLOOKUP:
996 written += write_midx_oid_lookup(f, the_hash_algo->rawsz, entries, nr_entries);
997 break;
999 case MIDX_CHUNKID_OBJECTOFFSETS:
1000 written += write_midx_object_offsets(f, large_offsets_needed, pack_perm, entries, nr_entries);
1001 break;
1003 case MIDX_CHUNKID_LARGEOFFSETS:
1004 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
1005 break;
1007 default:
1008 BUG("trying to write unknown chunk id %"PRIx32,
1009 chunk_ids[i]);
1012 display_progress(progress, i + 1);
1014 stop_progress(&progress);
1016 if (written != chunk_offsets[num_chunks])
1017 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
1018 written,
1019 chunk_offsets[num_chunks]);
1021 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1022 commit_lock_file(&lk);
1024 cleanup:
1025 for (i = 0; i < packs.nr; i++) {
1026 if (packs.info[i].p) {
1027 close_pack(packs.info[i].p);
1028 free(packs.info[i].p);
1030 free(packs.info[i].pack_name);
1033 free(packs.info);
1034 free(entries);
1035 free(pack_perm);
1036 free(midx_name);
1037 return result;
1040 int write_midx_file(const char *object_dir, unsigned flags)
1042 return write_midx_internal(object_dir, NULL, NULL, flags);
1045 void clear_midx_file(struct repository *r)
1047 char *midx = get_midx_filename(r->objects->odb->path);
1049 if (r->objects && r->objects->multi_pack_index) {
1050 close_midx(r->objects->multi_pack_index);
1051 r->objects->multi_pack_index = NULL;
1054 if (remove_path(midx)) {
1055 UNLEAK(midx);
1056 die(_("failed to clear multi-pack-index at %s"), midx);
1059 free(midx);
1062 static int verify_midx_error;
1064 static void midx_report(const char *fmt, ...)
1066 va_list ap;
1067 verify_midx_error = 1;
1068 va_start(ap, fmt);
1069 vfprintf(stderr, fmt, ap);
1070 fprintf(stderr, "\n");
1071 va_end(ap);
1074 struct pair_pos_vs_id
1076 uint32_t pos;
1077 uint32_t pack_int_id;
1080 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1082 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1083 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1085 return b->pack_int_id - a->pack_int_id;
1089 * Limit calls to display_progress() for performance reasons.
1090 * The interval here was arbitrarily chosen.
1092 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1093 #define midx_display_sparse_progress(progress, n) \
1094 do { \
1095 uint64_t _n = (n); \
1096 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1097 display_progress(progress, _n); \
1098 } while (0)
1100 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1102 struct pair_pos_vs_id *pairs = NULL;
1103 uint32_t i;
1104 struct progress *progress = NULL;
1105 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1106 verify_midx_error = 0;
1108 if (!m)
1109 return 0;
1111 if (flags & MIDX_PROGRESS)
1112 progress = start_progress(_("Looking for referenced packfiles"),
1113 m->num_packs);
1114 for (i = 0; i < m->num_packs; i++) {
1115 if (prepare_midx_pack(r, m, i))
1116 midx_report("failed to load pack in position %d", i);
1118 display_progress(progress, i + 1);
1120 stop_progress(&progress);
1122 for (i = 0; i < 255; i++) {
1123 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1124 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1126 if (oid_fanout1 > oid_fanout2)
1127 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1128 i, oid_fanout1, oid_fanout2, i + 1);
1131 if (m->num_objects == 0) {
1132 midx_report(_("the midx contains no oid"));
1134 * Remaining tests assume that we have objects, so we can
1135 * return here.
1137 return verify_midx_error;
1140 if (flags & MIDX_PROGRESS)
1141 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1142 m->num_objects - 1);
1143 for (i = 0; i < m->num_objects - 1; i++) {
1144 struct object_id oid1, oid2;
1146 nth_midxed_object_oid(&oid1, m, i);
1147 nth_midxed_object_oid(&oid2, m, i + 1);
1149 if (oidcmp(&oid1, &oid2) >= 0)
1150 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1151 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1153 midx_display_sparse_progress(progress, i + 1);
1155 stop_progress(&progress);
1158 * Create an array mapping each object to its packfile id. Sort it
1159 * to group the objects by packfile. Use this permutation to visit
1160 * each of the objects and only require 1 packfile to be open at a
1161 * time.
1163 ALLOC_ARRAY(pairs, m->num_objects);
1164 for (i = 0; i < m->num_objects; i++) {
1165 pairs[i].pos = i;
1166 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1169 if (flags & MIDX_PROGRESS)
1170 progress = start_sparse_progress(_("Sorting objects by packfile"),
1171 m->num_objects);
1172 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1173 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1174 stop_progress(&progress);
1176 if (flags & MIDX_PROGRESS)
1177 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1178 for (i = 0; i < m->num_objects; i++) {
1179 struct object_id oid;
1180 struct pack_entry e;
1181 off_t m_offset, p_offset;
1183 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1184 m->packs[pairs[i-1].pack_int_id])
1186 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1187 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1190 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1192 if (!fill_midx_entry(r, &oid, &e, m)) {
1193 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1194 pairs[i].pos, oid_to_hex(&oid));
1195 continue;
1198 if (open_pack_index(e.p)) {
1199 midx_report(_("failed to load pack-index for packfile %s"),
1200 e.p->pack_name);
1201 break;
1204 m_offset = e.offset;
1205 p_offset = find_pack_entry_one(oid.hash, e.p);
1207 if (m_offset != p_offset)
1208 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1209 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1211 midx_display_sparse_progress(progress, i + 1);
1213 stop_progress(&progress);
1215 free(pairs);
1217 return verify_midx_error;
1220 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1222 uint32_t i, *count, result = 0;
1223 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1224 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1225 struct progress *progress = NULL;
1227 if (!m)
1228 return 0;
1230 count = xcalloc(m->num_packs, sizeof(uint32_t));
1232 if (flags & MIDX_PROGRESS)
1233 progress = start_progress(_("Counting referenced objects"),
1234 m->num_objects);
1235 for (i = 0; i < m->num_objects; i++) {
1236 int pack_int_id = nth_midxed_pack_int_id(m, i);
1237 count[pack_int_id]++;
1238 display_progress(progress, i + 1);
1240 stop_progress(&progress);
1242 if (flags & MIDX_PROGRESS)
1243 progress = start_progress(_("Finding and deleting unreferenced packfiles"),
1244 m->num_packs);
1245 for (i = 0; i < m->num_packs; i++) {
1246 char *pack_name;
1247 display_progress(progress, i + 1);
1249 if (count[i])
1250 continue;
1252 if (prepare_midx_pack(r, m, i))
1253 continue;
1255 if (m->packs[i]->pack_keep)
1256 continue;
1258 pack_name = xstrdup(m->packs[i]->pack_name);
1259 close_pack(m->packs[i]);
1261 string_list_insert(&packs_to_drop, m->pack_names[i]);
1262 unlink_pack_path(pack_name, 0);
1263 free(pack_name);
1265 stop_progress(&progress);
1267 free(count);
1269 if (packs_to_drop.nr)
1270 result = write_midx_internal(object_dir, m, &packs_to_drop, flags);
1272 string_list_clear(&packs_to_drop, 0);
1273 return result;
1276 struct repack_info {
1277 timestamp_t mtime;
1278 uint32_t referenced_objects;
1279 uint32_t pack_int_id;
1282 static int compare_by_mtime(const void *a_, const void *b_)
1284 const struct repack_info *a, *b;
1286 a = (const struct repack_info *)a_;
1287 b = (const struct repack_info *)b_;
1289 if (a->mtime < b->mtime)
1290 return -1;
1291 if (a->mtime > b->mtime)
1292 return 1;
1293 return 0;
1296 static int fill_included_packs_all(struct repository *r,
1297 struct multi_pack_index *m,
1298 unsigned char *include_pack)
1300 uint32_t i, count = 0;
1301 int pack_kept_objects = 0;
1303 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1305 for (i = 0; i < m->num_packs; i++) {
1306 if (prepare_midx_pack(r, m, i))
1307 continue;
1308 if (!pack_kept_objects && m->packs[i]->pack_keep)
1309 continue;
1311 include_pack[i] = 1;
1312 count++;
1315 return count < 2;
1318 static int fill_included_packs_batch(struct repository *r,
1319 struct multi_pack_index *m,
1320 unsigned char *include_pack,
1321 size_t batch_size)
1323 uint32_t i, packs_to_repack;
1324 size_t total_size;
1325 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1326 int pack_kept_objects = 0;
1328 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1330 for (i = 0; i < m->num_packs; i++) {
1331 pack_info[i].pack_int_id = i;
1333 if (prepare_midx_pack(r, m, i))
1334 continue;
1336 pack_info[i].mtime = m->packs[i]->mtime;
1339 for (i = 0; batch_size && i < m->num_objects; i++) {
1340 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1341 pack_info[pack_int_id].referenced_objects++;
1344 QSORT(pack_info, m->num_packs, compare_by_mtime);
1346 total_size = 0;
1347 packs_to_repack = 0;
1348 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1349 int pack_int_id = pack_info[i].pack_int_id;
1350 struct packed_git *p = m->packs[pack_int_id];
1351 size_t expected_size;
1353 if (!p)
1354 continue;
1355 if (!pack_kept_objects && p->pack_keep)
1356 continue;
1357 if (open_pack_index(p) || !p->num_objects)
1358 continue;
1360 expected_size = (size_t)(p->pack_size
1361 * pack_info[i].referenced_objects);
1362 expected_size /= p->num_objects;
1364 if (expected_size >= batch_size)
1365 continue;
1367 packs_to_repack++;
1368 total_size += expected_size;
1369 include_pack[pack_int_id] = 1;
1372 free(pack_info);
1374 if (total_size < batch_size || packs_to_repack < 2)
1375 return 1;
1377 return 0;
1380 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1382 int result = 0;
1383 uint32_t i;
1384 unsigned char *include_pack;
1385 struct child_process cmd = CHILD_PROCESS_INIT;
1386 struct strbuf base_name = STRBUF_INIT;
1387 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1390 * When updating the default for these configuration
1391 * variables in builtin/repack.c, these must be adjusted
1392 * to match.
1394 int delta_base_offset = 1;
1395 int use_delta_islands = 0;
1397 if (!m)
1398 return 0;
1400 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1402 if (batch_size) {
1403 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1404 goto cleanup;
1405 } else if (fill_included_packs_all(r, m, include_pack))
1406 goto cleanup;
1408 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1409 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1411 argv_array_push(&cmd.args, "pack-objects");
1413 strbuf_addstr(&base_name, object_dir);
1414 strbuf_addstr(&base_name, "/pack/pack");
1415 argv_array_push(&cmd.args, base_name.buf);
1417 if (delta_base_offset)
1418 argv_array_push(&cmd.args, "--delta-base-offset");
1419 if (use_delta_islands)
1420 argv_array_push(&cmd.args, "--delta-islands");
1422 if (flags & MIDX_PROGRESS)
1423 argv_array_push(&cmd.args, "--progress");
1424 else
1425 argv_array_push(&cmd.args, "-q");
1427 strbuf_release(&base_name);
1429 cmd.git_cmd = 1;
1430 cmd.in = cmd.out = -1;
1432 if (start_command(&cmd)) {
1433 error(_("could not start pack-objects"));
1434 result = 1;
1435 goto cleanup;
1438 for (i = 0; i < m->num_objects; i++) {
1439 struct object_id oid;
1440 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1442 if (!include_pack[pack_int_id])
1443 continue;
1445 nth_midxed_object_oid(&oid, m, i);
1446 xwrite(cmd.in, oid_to_hex(&oid), the_hash_algo->hexsz);
1447 xwrite(cmd.in, "\n", 1);
1449 close(cmd.in);
1451 if (finish_command(&cmd)) {
1452 error(_("could not finish pack-objects"));
1453 result = 1;
1454 goto cleanup;
1457 result = write_midx_internal(object_dir, m, NULL, flags);
1458 m = NULL;
1460 cleanup:
1461 if (m)
1462 close_midx(m);
1463 free(include_pack);
1464 return result;