midx: add progress to expire_midx_packs
[git.git] / midx.c
blobcb2c602603649048526a0a781efd0539aba512f6
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);
76 FLEX_ALLOC_STR(m, object_dir, object_dir);
77 m->fd = fd;
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);
193 close(m->fd);
194 m->fd = -1;
196 for (i = 0; i < m->num_packs; i++) {
197 if (m->packs[i])
198 m->packs[i]->multi_pack_index = 0;
200 FREE_AND_NULL(m->packs);
201 FREE_AND_NULL(m->pack_names);
204 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
206 struct strbuf pack_name = STRBUF_INIT;
207 struct packed_git *p;
209 if (pack_int_id >= m->num_packs)
210 die(_("bad pack-int-id: %u (%u total packs)"),
211 pack_int_id, m->num_packs);
213 if (m->packs[pack_int_id])
214 return 0;
216 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
217 m->pack_names[pack_int_id]);
219 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
220 strbuf_release(&pack_name);
222 if (!p)
223 return 1;
225 p->multi_pack_index = 1;
226 m->packs[pack_int_id] = p;
227 install_packed_git(r, p);
228 list_add_tail(&p->mru, &r->objects->packed_git_mru);
230 return 0;
233 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
235 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
236 the_hash_algo->rawsz, result);
239 struct object_id *nth_midxed_object_oid(struct object_id *oid,
240 struct multi_pack_index *m,
241 uint32_t n)
243 if (n >= m->num_objects)
244 return NULL;
246 hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
247 return oid;
250 static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
252 const unsigned char *offset_data;
253 uint32_t offset32;
255 offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
256 offset32 = get_be32(offset_data + sizeof(uint32_t));
258 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
259 if (sizeof(off_t) < sizeof(uint64_t))
260 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
262 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
263 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
266 return offset32;
269 static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
271 return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
274 static int nth_midxed_pack_entry(struct repository *r,
275 struct multi_pack_index *m,
276 struct pack_entry *e,
277 uint32_t pos)
279 uint32_t pack_int_id;
280 struct packed_git *p;
282 if (pos >= m->num_objects)
283 return 0;
285 pack_int_id = nth_midxed_pack_int_id(m, pos);
287 if (prepare_midx_pack(r, m, pack_int_id))
288 die(_("error preparing packfile from multi-pack-index"));
289 p = m->packs[pack_int_id];
292 * We are about to tell the caller where they can locate the
293 * requested object. We better make sure the packfile is
294 * still here and can be accessed before supplying that
295 * answer, as it may have been deleted since the MIDX was
296 * loaded!
298 if (!is_pack_valid(p))
299 return 0;
301 if (p->num_bad_objects) {
302 uint32_t i;
303 struct object_id oid;
304 nth_midxed_object_oid(&oid, m, pos);
305 for (i = 0; i < p->num_bad_objects; i++)
306 if (hasheq(oid.hash,
307 p->bad_object_sha1 + the_hash_algo->rawsz * i))
308 return 0;
311 e->offset = nth_midxed_offset(m, pos);
312 e->p = p;
314 return 1;
317 int fill_midx_entry(struct repository * r,
318 const struct object_id *oid,
319 struct pack_entry *e,
320 struct multi_pack_index *m)
322 uint32_t pos;
324 if (!bsearch_midx(oid, m, &pos))
325 return 0;
327 return nth_midxed_pack_entry(r, m, e, pos);
330 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
331 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
332 const char *idx_name)
334 /* Skip past any initial matching prefix. */
335 while (*idx_name && *idx_name == *idx_or_pack_name) {
336 idx_name++;
337 idx_or_pack_name++;
341 * If we didn't match completely, we may have matched "pack-1234." and
342 * be left with "idx" and "pack" respectively, which is also OK. We do
343 * not have to check for "idx" and "idx", because that would have been
344 * a complete match (and in that case these strcmps will be false, but
345 * we'll correctly return 0 from the final strcmp() below.
347 * Technically this matches "fooidx" and "foopack", but we'd never have
348 * such names in the first place.
350 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
351 return 0;
354 * This not only checks for a complete match, but also orders based on
355 * the first non-identical character, which means our ordering will
356 * match a raw strcmp(). That makes it OK to use this to binary search
357 * a naively-sorted list.
359 return strcmp(idx_or_pack_name, idx_name);
362 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
364 uint32_t first = 0, last = m->num_packs;
366 while (first < last) {
367 uint32_t mid = first + (last - first) / 2;
368 const char *current;
369 int cmp;
371 current = m->pack_names[mid];
372 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
373 if (!cmp)
374 return 1;
375 if (cmp > 0) {
376 first = mid + 1;
377 continue;
379 last = mid;
382 return 0;
385 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
387 struct multi_pack_index *m;
388 struct multi_pack_index *m_search;
389 int config_value;
390 static int env_value = -1;
392 if (env_value < 0)
393 env_value = git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0);
395 if (!env_value &&
396 (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
397 !config_value))
398 return 0;
400 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
401 if (!strcmp(object_dir, m_search->object_dir))
402 return 1;
404 m = load_multi_pack_index(object_dir, local);
406 if (m) {
407 m->next = r->objects->multi_pack_index;
408 r->objects->multi_pack_index = m;
409 return 1;
412 return 0;
415 static size_t write_midx_header(struct hashfile *f,
416 unsigned char num_chunks,
417 uint32_t num_packs)
419 unsigned char byte_values[4];
421 hashwrite_be32(f, MIDX_SIGNATURE);
422 byte_values[0] = MIDX_VERSION;
423 byte_values[1] = MIDX_HASH_VERSION;
424 byte_values[2] = num_chunks;
425 byte_values[3] = 0; /* unused */
426 hashwrite(f, byte_values, sizeof(byte_values));
427 hashwrite_be32(f, num_packs);
429 return MIDX_HEADER_SIZE;
432 struct pack_info {
433 uint32_t orig_pack_int_id;
434 char *pack_name;
435 struct packed_git *p;
436 unsigned expired : 1;
439 static int pack_info_compare(const void *_a, const void *_b)
441 struct pack_info *a = (struct pack_info *)_a;
442 struct pack_info *b = (struct pack_info *)_b;
443 return strcmp(a->pack_name, b->pack_name);
446 struct pack_list {
447 struct pack_info *info;
448 uint32_t nr;
449 uint32_t alloc;
450 struct multi_pack_index *m;
451 struct progress *progress;
452 unsigned pack_paths_checked;
455 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
456 const char *file_name, void *data)
458 struct pack_list *packs = (struct pack_list *)data;
460 if (ends_with(file_name, ".idx")) {
461 display_progress(packs->progress, ++packs->pack_paths_checked);
462 if (packs->m && midx_contains_pack(packs->m, file_name))
463 return;
465 ALLOC_GROW(packs->info, packs->nr + 1, packs->alloc);
467 packs->info[packs->nr].p = add_packed_git(full_path,
468 full_path_len,
471 if (!packs->info[packs->nr].p) {
472 warning(_("failed to add packfile '%s'"),
473 full_path);
474 return;
477 if (open_pack_index(packs->info[packs->nr].p)) {
478 warning(_("failed to open pack-index '%s'"),
479 full_path);
480 close_pack(packs->info[packs->nr].p);
481 FREE_AND_NULL(packs->info[packs->nr].p);
482 return;
485 packs->info[packs->nr].pack_name = xstrdup(file_name);
486 packs->info[packs->nr].orig_pack_int_id = packs->nr;
487 packs->info[packs->nr].expired = 0;
488 packs->nr++;
492 struct pack_midx_entry {
493 struct object_id oid;
494 uint32_t pack_int_id;
495 time_t pack_mtime;
496 uint64_t offset;
499 static int midx_oid_compare(const void *_a, const void *_b)
501 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
502 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
503 int cmp = oidcmp(&a->oid, &b->oid);
505 if (cmp)
506 return cmp;
508 if (a->pack_mtime > b->pack_mtime)
509 return -1;
510 else if (a->pack_mtime < b->pack_mtime)
511 return 1;
513 return a->pack_int_id - b->pack_int_id;
516 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
517 struct pack_midx_entry *e,
518 uint32_t pos)
520 if (pos >= m->num_objects)
521 return 1;
523 nth_midxed_object_oid(&e->oid, m, pos);
524 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
525 e->offset = nth_midxed_offset(m, pos);
527 /* consider objects in midx to be from "old" packs */
528 e->pack_mtime = 0;
529 return 0;
532 static void fill_pack_entry(uint32_t pack_int_id,
533 struct packed_git *p,
534 uint32_t cur_object,
535 struct pack_midx_entry *entry)
537 if (!nth_packed_object_oid(&entry->oid, p, cur_object))
538 die(_("failed to locate object %d in packfile"), cur_object);
540 entry->pack_int_id = pack_int_id;
541 entry->pack_mtime = p->mtime;
543 entry->offset = nth_packed_object_offset(p, cur_object);
547 * It is possible to artificially get into a state where there are many
548 * duplicate copies of objects. That can create high memory pressure if
549 * we are to create a list of all objects before de-duplication. To reduce
550 * this memory pressure without a significant performance drop, automatically
551 * group objects by the first byte of their object id. Use the IDX fanout
552 * tables to group the data, copy to a local array, then sort.
554 * Copy only the de-duplicated entries (selected by most-recent modified time
555 * of a packfile containing the object).
557 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
558 struct pack_info *info,
559 uint32_t nr_packs,
560 uint32_t *nr_objects)
562 uint32_t cur_fanout, cur_pack, cur_object;
563 uint32_t alloc_fanout, alloc_objects, total_objects = 0;
564 struct pack_midx_entry *entries_by_fanout = NULL;
565 struct pack_midx_entry *deduplicated_entries = NULL;
566 uint32_t start_pack = m ? m->num_packs : 0;
568 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
569 total_objects += info[cur_pack].p->num_objects;
572 * As we de-duplicate by fanout value, we expect the fanout
573 * slices to be evenly distributed, with some noise. Hence,
574 * allocate slightly more than one 256th.
576 alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
578 ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
579 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
580 *nr_objects = 0;
582 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
583 uint32_t nr_fanout = 0;
585 if (m) {
586 uint32_t start = 0, end;
588 if (cur_fanout)
589 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
590 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
592 for (cur_object = start; cur_object < end; cur_object++) {
593 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
594 nth_midxed_pack_midx_entry(m,
595 &entries_by_fanout[nr_fanout],
596 cur_object);
597 nr_fanout++;
601 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
602 uint32_t start = 0, end;
604 if (cur_fanout)
605 start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1);
606 end = get_pack_fanout(info[cur_pack].p, cur_fanout);
608 for (cur_object = start; cur_object < end; cur_object++) {
609 ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
610 fill_pack_entry(cur_pack, info[cur_pack].p, cur_object, &entries_by_fanout[nr_fanout]);
611 nr_fanout++;
615 QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
618 * The batch is now sorted by OID and then mtime (descending).
619 * Take only the first duplicate.
621 for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
622 if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
623 &entries_by_fanout[cur_object].oid))
624 continue;
626 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
627 memcpy(&deduplicated_entries[*nr_objects],
628 &entries_by_fanout[cur_object],
629 sizeof(struct pack_midx_entry));
630 (*nr_objects)++;
634 free(entries_by_fanout);
635 return deduplicated_entries;
638 static size_t write_midx_pack_names(struct hashfile *f,
639 struct pack_info *info,
640 uint32_t num_packs)
642 uint32_t i;
643 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
644 size_t written = 0;
646 for (i = 0; i < num_packs; i++) {
647 size_t writelen;
649 if (info[i].expired)
650 continue;
652 if (i && strcmp(info[i].pack_name, info[i - 1].pack_name) <= 0)
653 BUG("incorrect pack-file order: %s before %s",
654 info[i - 1].pack_name,
655 info[i].pack_name);
657 writelen = strlen(info[i].pack_name) + 1;
658 hashwrite(f, info[i].pack_name, writelen);
659 written += writelen;
662 /* add padding to be aligned */
663 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
664 if (i < MIDX_CHUNK_ALIGNMENT) {
665 memset(padding, 0, sizeof(padding));
666 hashwrite(f, padding, i);
667 written += i;
670 return written;
673 static size_t write_midx_oid_fanout(struct hashfile *f,
674 struct pack_midx_entry *objects,
675 uint32_t nr_objects)
677 struct pack_midx_entry *list = objects;
678 struct pack_midx_entry *last = objects + nr_objects;
679 uint32_t count = 0;
680 uint32_t i;
683 * Write the first-level table (the list is sorted,
684 * but we use a 256-entry lookup to be able to avoid
685 * having to do eight extra binary search iterations).
687 for (i = 0; i < 256; i++) {
688 struct pack_midx_entry *next = list;
690 while (next < last && next->oid.hash[0] == i) {
691 count++;
692 next++;
695 hashwrite_be32(f, count);
696 list = next;
699 return MIDX_CHUNK_FANOUT_SIZE;
702 static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
703 struct pack_midx_entry *objects,
704 uint32_t nr_objects)
706 struct pack_midx_entry *list = objects;
707 uint32_t i;
708 size_t written = 0;
710 for (i = 0; i < nr_objects; i++) {
711 struct pack_midx_entry *obj = list++;
713 if (i < nr_objects - 1) {
714 struct pack_midx_entry *next = list;
715 if (oidcmp(&obj->oid, &next->oid) >= 0)
716 BUG("OIDs not in order: %s >= %s",
717 oid_to_hex(&obj->oid),
718 oid_to_hex(&next->oid));
721 hashwrite(f, obj->oid.hash, (int)hash_len);
722 written += hash_len;
725 return written;
728 static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
729 uint32_t *perm,
730 struct pack_midx_entry *objects, uint32_t nr_objects)
732 struct pack_midx_entry *list = objects;
733 uint32_t i, nr_large_offset = 0;
734 size_t written = 0;
736 for (i = 0; i < nr_objects; i++) {
737 struct pack_midx_entry *obj = list++;
739 if (perm[obj->pack_int_id] == PACK_EXPIRED)
740 BUG("object %s is in an expired pack with int-id %d",
741 oid_to_hex(&obj->oid),
742 obj->pack_int_id);
744 hashwrite_be32(f, perm[obj->pack_int_id]);
746 if (large_offset_needed && obj->offset >> 31)
747 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
748 else if (!large_offset_needed && obj->offset >> 32)
749 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
750 oid_to_hex(&obj->oid),
751 obj->offset);
752 else
753 hashwrite_be32(f, (uint32_t)obj->offset);
755 written += MIDX_CHUNK_OFFSET_WIDTH;
758 return written;
761 static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
762 struct pack_midx_entry *objects, uint32_t nr_objects)
764 struct pack_midx_entry *list = objects, *end = objects + nr_objects;
765 size_t written = 0;
767 while (nr_large_offset) {
768 struct pack_midx_entry *obj;
769 uint64_t offset;
771 if (list >= end)
772 BUG("too many large-offset objects");
774 obj = list++;
775 offset = obj->offset;
777 if (!(offset >> 31))
778 continue;
780 hashwrite_be32(f, offset >> 32);
781 hashwrite_be32(f, offset & 0xffffffffUL);
782 written += 2 * sizeof(uint32_t);
784 nr_large_offset--;
787 return written;
790 static int write_midx_internal(const char *object_dir, struct multi_pack_index *m,
791 struct string_list *packs_to_drop, unsigned flags)
793 unsigned char cur_chunk, num_chunks = 0;
794 char *midx_name;
795 uint32_t i;
796 struct hashfile *f = NULL;
797 struct lock_file lk;
798 struct pack_list packs;
799 uint32_t *pack_perm = NULL;
800 uint64_t written = 0;
801 uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
802 uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
803 uint32_t nr_entries, num_large_offsets = 0;
804 struct pack_midx_entry *entries = NULL;
805 struct progress *progress = NULL;
806 int large_offsets_needed = 0;
807 int pack_name_concat_len = 0;
808 int dropped_packs = 0;
809 int result = 0;
811 midx_name = get_midx_filename(object_dir);
812 if (safe_create_leading_directories(midx_name)) {
813 UNLEAK(midx_name);
814 die_errno(_("unable to create leading directories of %s"),
815 midx_name);
818 if (m)
819 packs.m = m;
820 else
821 packs.m = load_multi_pack_index(object_dir, 1);
823 packs.nr = 0;
824 packs.alloc = packs.m ? packs.m->num_packs : 16;
825 packs.info = NULL;
826 ALLOC_ARRAY(packs.info, packs.alloc);
828 if (packs.m) {
829 for (i = 0; i < packs.m->num_packs; i++) {
830 ALLOC_GROW(packs.info, packs.nr + 1, packs.alloc);
832 packs.info[packs.nr].orig_pack_int_id = i;
833 packs.info[packs.nr].pack_name = xstrdup(packs.m->pack_names[i]);
834 packs.info[packs.nr].p = NULL;
835 packs.info[packs.nr].expired = 0;
836 packs.nr++;
840 packs.pack_paths_checked = 0;
841 if (flags & MIDX_PROGRESS)
842 packs.progress = start_progress(_("Adding packfiles to multi-pack-index"), 0);
843 else
844 packs.progress = NULL;
846 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
847 stop_progress(&packs.progress);
849 if (packs.m && packs.nr == packs.m->num_packs && !packs_to_drop)
850 goto cleanup;
852 entries = get_sorted_entries(packs.m, packs.info, packs.nr, &nr_entries);
854 for (i = 0; i < nr_entries; i++) {
855 if (entries[i].offset > 0x7fffffff)
856 num_large_offsets++;
857 if (entries[i].offset > 0xffffffff)
858 large_offsets_needed = 1;
861 QSORT(packs.info, packs.nr, pack_info_compare);
863 if (packs_to_drop && packs_to_drop->nr) {
864 int drop_index = 0;
865 int missing_drops = 0;
867 for (i = 0; i < packs.nr && drop_index < packs_to_drop->nr; i++) {
868 int cmp = strcmp(packs.info[i].pack_name,
869 packs_to_drop->items[drop_index].string);
871 if (!cmp) {
872 drop_index++;
873 packs.info[i].expired = 1;
874 } else if (cmp > 0) {
875 error(_("did not see pack-file %s to drop"),
876 packs_to_drop->items[drop_index].string);
877 drop_index++;
878 missing_drops++;
879 i--;
880 } else {
881 packs.info[i].expired = 0;
885 if (missing_drops) {
886 result = 1;
887 goto cleanup;
892 * pack_perm stores a permutation between pack-int-ids from the
893 * previous multi-pack-index to the new one we are writing:
895 * pack_perm[old_id] = new_id
897 ALLOC_ARRAY(pack_perm, packs.nr);
898 for (i = 0; i < packs.nr; i++) {
899 if (packs.info[i].expired) {
900 dropped_packs++;
901 pack_perm[packs.info[i].orig_pack_int_id] = PACK_EXPIRED;
902 } else {
903 pack_perm[packs.info[i].orig_pack_int_id] = i - dropped_packs;
907 for (i = 0; i < packs.nr; i++) {
908 if (!packs.info[i].expired)
909 pack_name_concat_len += strlen(packs.info[i].pack_name) + 1;
912 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
913 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
914 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
916 hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
917 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
918 FREE_AND_NULL(midx_name);
920 if (packs.m)
921 close_midx(packs.m);
923 cur_chunk = 0;
924 num_chunks = large_offsets_needed ? 5 : 4;
926 written = write_midx_header(f, num_chunks, packs.nr - dropped_packs);
928 chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
929 chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
931 cur_chunk++;
932 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
933 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + pack_name_concat_len;
935 cur_chunk++;
936 chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
937 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
939 cur_chunk++;
940 chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
941 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * the_hash_algo->rawsz;
943 cur_chunk++;
944 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
945 if (large_offsets_needed) {
946 chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
948 cur_chunk++;
949 chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
950 num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
953 chunk_ids[cur_chunk] = 0;
955 for (i = 0; i <= num_chunks; i++) {
956 if (i && chunk_offsets[i] < chunk_offsets[i - 1])
957 BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
958 chunk_offsets[i - 1],
959 chunk_offsets[i]);
961 if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
962 BUG("chunk offset %"PRIu64" is not properly aligned",
963 chunk_offsets[i]);
965 hashwrite_be32(f, chunk_ids[i]);
966 hashwrite_be32(f, chunk_offsets[i] >> 32);
967 hashwrite_be32(f, chunk_offsets[i]);
969 written += MIDX_CHUNKLOOKUP_WIDTH;
972 if (flags & MIDX_PROGRESS)
973 progress = start_progress(_("Writing chunks to multi-pack-index"),
974 num_chunks);
975 for (i = 0; i < num_chunks; i++) {
976 if (written != chunk_offsets[i])
977 BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
978 chunk_offsets[i],
979 written,
980 chunk_ids[i]);
982 switch (chunk_ids[i]) {
983 case MIDX_CHUNKID_PACKNAMES:
984 written += write_midx_pack_names(f, packs.info, packs.nr);
985 break;
987 case MIDX_CHUNKID_OIDFANOUT:
988 written += write_midx_oid_fanout(f, entries, nr_entries);
989 break;
991 case MIDX_CHUNKID_OIDLOOKUP:
992 written += write_midx_oid_lookup(f, the_hash_algo->rawsz, entries, nr_entries);
993 break;
995 case MIDX_CHUNKID_OBJECTOFFSETS:
996 written += write_midx_object_offsets(f, large_offsets_needed, pack_perm, entries, nr_entries);
997 break;
999 case MIDX_CHUNKID_LARGEOFFSETS:
1000 written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
1001 break;
1003 default:
1004 BUG("trying to write unknown chunk id %"PRIx32,
1005 chunk_ids[i]);
1008 display_progress(progress, i + 1);
1010 stop_progress(&progress);
1012 if (written != chunk_offsets[num_chunks])
1013 BUG("incorrect final offset %"PRIu64" != %"PRIu64,
1014 written,
1015 chunk_offsets[num_chunks]);
1017 finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1018 commit_lock_file(&lk);
1020 cleanup:
1021 for (i = 0; i < packs.nr; i++) {
1022 if (packs.info[i].p) {
1023 close_pack(packs.info[i].p);
1024 free(packs.info[i].p);
1026 free(packs.info[i].pack_name);
1029 free(packs.info);
1030 free(entries);
1031 free(pack_perm);
1032 free(midx_name);
1033 return result;
1036 int write_midx_file(const char *object_dir, unsigned flags)
1038 return write_midx_internal(object_dir, NULL, NULL, flags);
1041 void clear_midx_file(struct repository *r)
1043 char *midx = get_midx_filename(r->objects->odb->path);
1045 if (r->objects && r->objects->multi_pack_index) {
1046 close_midx(r->objects->multi_pack_index);
1047 r->objects->multi_pack_index = NULL;
1050 if (remove_path(midx)) {
1051 UNLEAK(midx);
1052 die(_("failed to clear multi-pack-index at %s"), midx);
1055 free(midx);
1058 static int verify_midx_error;
1060 static void midx_report(const char *fmt, ...)
1062 va_list ap;
1063 verify_midx_error = 1;
1064 va_start(ap, fmt);
1065 vfprintf(stderr, fmt, ap);
1066 fprintf(stderr, "\n");
1067 va_end(ap);
1070 struct pair_pos_vs_id
1072 uint32_t pos;
1073 uint32_t pack_int_id;
1076 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1078 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1079 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1081 return b->pack_int_id - a->pack_int_id;
1085 * Limit calls to display_progress() for performance reasons.
1086 * The interval here was arbitrarily chosen.
1088 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1089 #define midx_display_sparse_progress(progress, n) \
1090 do { \
1091 uint64_t _n = (n); \
1092 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1093 display_progress(progress, _n); \
1094 } while (0)
1096 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1098 struct pair_pos_vs_id *pairs = NULL;
1099 uint32_t i;
1100 struct progress *progress;
1101 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1102 verify_midx_error = 0;
1104 if (!m)
1105 return 0;
1107 progress = start_progress(_("Looking for referenced packfiles"),
1108 m->num_packs);
1109 for (i = 0; i < m->num_packs; i++) {
1110 if (prepare_midx_pack(r, m, i))
1111 midx_report("failed to load pack in position %d", i);
1113 display_progress(progress, i + 1);
1115 stop_progress(&progress);
1117 for (i = 0; i < 255; i++) {
1118 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1119 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1121 if (oid_fanout1 > oid_fanout2)
1122 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1123 i, oid_fanout1, oid_fanout2, i + 1);
1126 progress = start_sparse_progress(_("Verifying OID order in MIDX"),
1127 m->num_objects - 1);
1128 for (i = 0; i < m->num_objects - 1; i++) {
1129 struct object_id oid1, oid2;
1131 nth_midxed_object_oid(&oid1, m, i);
1132 nth_midxed_object_oid(&oid2, m, i + 1);
1134 if (oidcmp(&oid1, &oid2) >= 0)
1135 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1136 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1138 midx_display_sparse_progress(progress, i + 1);
1140 stop_progress(&progress);
1143 * Create an array mapping each object to its packfile id. Sort it
1144 * to group the objects by packfile. Use this permutation to visit
1145 * each of the objects and only require 1 packfile to be open at a
1146 * time.
1148 ALLOC_ARRAY(pairs, m->num_objects);
1149 for (i = 0; i < m->num_objects; i++) {
1150 pairs[i].pos = i;
1151 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1154 progress = start_sparse_progress(_("Sorting objects by packfile"),
1155 m->num_objects);
1156 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1157 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1158 stop_progress(&progress);
1160 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1161 for (i = 0; i < m->num_objects; i++) {
1162 struct object_id oid;
1163 struct pack_entry e;
1164 off_t m_offset, p_offset;
1166 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1167 m->packs[pairs[i-1].pack_int_id])
1169 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1170 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1173 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1175 if (!fill_midx_entry(r, &oid, &e, m)) {
1176 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1177 pairs[i].pos, oid_to_hex(&oid));
1178 continue;
1181 if (open_pack_index(e.p)) {
1182 midx_report(_("failed to load pack-index for packfile %s"),
1183 e.p->pack_name);
1184 break;
1187 m_offset = e.offset;
1188 p_offset = find_pack_entry_one(oid.hash, e.p);
1190 if (m_offset != p_offset)
1191 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1192 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1194 midx_display_sparse_progress(progress, i + 1);
1196 stop_progress(&progress);
1198 free(pairs);
1200 return verify_midx_error;
1203 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1205 uint32_t i, *count, result = 0;
1206 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1207 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1208 struct progress *progress = NULL;
1210 if (!m)
1211 return 0;
1213 count = xcalloc(m->num_packs, sizeof(uint32_t));
1215 if (flags & MIDX_PROGRESS)
1216 progress = start_progress(_("Counting referenced objects"),
1217 m->num_objects);
1218 for (i = 0; i < m->num_objects; i++) {
1219 int pack_int_id = nth_midxed_pack_int_id(m, i);
1220 count[pack_int_id]++;
1221 display_progress(progress, i + 1);
1223 stop_progress(&progress);
1225 if (flags & MIDX_PROGRESS)
1226 progress = start_progress(_("Finding and deleting unreferenced packfiles"),
1227 m->num_packs);
1228 for (i = 0; i < m->num_packs; i++) {
1229 char *pack_name;
1230 display_progress(progress, i + 1);
1232 if (count[i])
1233 continue;
1235 if (prepare_midx_pack(r, m, i))
1236 continue;
1238 if (m->packs[i]->pack_keep)
1239 continue;
1241 pack_name = xstrdup(m->packs[i]->pack_name);
1242 close_pack(m->packs[i]);
1244 string_list_insert(&packs_to_drop, m->pack_names[i]);
1245 unlink_pack_path(pack_name, 0);
1246 free(pack_name);
1248 stop_progress(&progress);
1250 free(count);
1252 if (packs_to_drop.nr)
1253 result = write_midx_internal(object_dir, m, &packs_to_drop, flags);
1255 string_list_clear(&packs_to_drop, 0);
1256 return result;
1259 struct repack_info {
1260 timestamp_t mtime;
1261 uint32_t referenced_objects;
1262 uint32_t pack_int_id;
1265 static int compare_by_mtime(const void *a_, const void *b_)
1267 const struct repack_info *a, *b;
1269 a = (const struct repack_info *)a_;
1270 b = (const struct repack_info *)b_;
1272 if (a->mtime < b->mtime)
1273 return -1;
1274 if (a->mtime > b->mtime)
1275 return 1;
1276 return 0;
1279 static int fill_included_packs_all(struct multi_pack_index *m,
1280 unsigned char *include_pack)
1282 uint32_t i;
1284 for (i = 0; i < m->num_packs; i++)
1285 include_pack[i] = 1;
1287 return m->num_packs < 2;
1290 static int fill_included_packs_batch(struct repository *r,
1291 struct multi_pack_index *m,
1292 unsigned char *include_pack,
1293 size_t batch_size)
1295 uint32_t i, packs_to_repack;
1296 size_t total_size;
1297 struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info));
1299 for (i = 0; i < m->num_packs; i++) {
1300 pack_info[i].pack_int_id = i;
1302 if (prepare_midx_pack(r, m, i))
1303 continue;
1305 pack_info[i].mtime = m->packs[i]->mtime;
1308 for (i = 0; batch_size && i < m->num_objects; i++) {
1309 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1310 pack_info[pack_int_id].referenced_objects++;
1313 QSORT(pack_info, m->num_packs, compare_by_mtime);
1315 total_size = 0;
1316 packs_to_repack = 0;
1317 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1318 int pack_int_id = pack_info[i].pack_int_id;
1319 struct packed_git *p = m->packs[pack_int_id];
1320 size_t expected_size;
1322 if (!p)
1323 continue;
1324 if (open_pack_index(p) || !p->num_objects)
1325 continue;
1327 expected_size = (size_t)(p->pack_size
1328 * pack_info[i].referenced_objects);
1329 expected_size /= p->num_objects;
1331 if (expected_size >= batch_size)
1332 continue;
1334 packs_to_repack++;
1335 total_size += expected_size;
1336 include_pack[pack_int_id] = 1;
1339 free(pack_info);
1341 if (total_size < batch_size || packs_to_repack < 2)
1342 return 1;
1344 return 0;
1347 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
1349 int result = 0;
1350 uint32_t i;
1351 unsigned char *include_pack;
1352 struct child_process cmd = CHILD_PROCESS_INIT;
1353 struct strbuf base_name = STRBUF_INIT;
1354 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1356 if (!m)
1357 return 0;
1359 include_pack = xcalloc(m->num_packs, sizeof(unsigned char));
1361 if (batch_size) {
1362 if (fill_included_packs_batch(r, m, include_pack, batch_size))
1363 goto cleanup;
1364 } else if (fill_included_packs_all(m, include_pack))
1365 goto cleanup;
1367 argv_array_push(&cmd.args, "pack-objects");
1369 strbuf_addstr(&base_name, object_dir);
1370 strbuf_addstr(&base_name, "/pack/pack");
1371 argv_array_push(&cmd.args, base_name.buf);
1372 strbuf_release(&base_name);
1374 cmd.git_cmd = 1;
1375 cmd.in = cmd.out = -1;
1377 if (start_command(&cmd)) {
1378 error(_("could not start pack-objects"));
1379 result = 1;
1380 goto cleanup;
1383 for (i = 0; i < m->num_objects; i++) {
1384 struct object_id oid;
1385 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1387 if (!include_pack[pack_int_id])
1388 continue;
1390 nth_midxed_object_oid(&oid, m, i);
1391 xwrite(cmd.in, oid_to_hex(&oid), the_hash_algo->hexsz);
1392 xwrite(cmd.in, "\n", 1);
1394 close(cmd.in);
1396 if (finish_command(&cmd)) {
1397 error(_("could not finish pack-objects"));
1398 result = 1;
1399 goto cleanup;
1402 result = write_midx_internal(object_dir, m, NULL, flags);
1403 m = NULL;
1405 cleanup:
1406 if (m)
1407 close_midx(m);
1408 free(include_pack);
1409 return result;