parse-options: decouple "--end-of-options" and "--"
[git/gitster.git] / midx.c
blob7cfad04a24027c1f0703dafdeeb31989414a13c5
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"
15 #include "pack.h"
16 #include "pack-bitmap.h"
17 #include "refs.h"
18 #include "revision.h"
19 #include "list-objects.h"
21 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
22 #define MIDX_VERSION 1
23 #define MIDX_BYTE_FILE_VERSION 4
24 #define MIDX_BYTE_HASH_VERSION 5
25 #define MIDX_BYTE_NUM_CHUNKS 6
26 #define MIDX_BYTE_NUM_PACKS 8
27 #define MIDX_HEADER_SIZE 12
28 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
30 #define MIDX_CHUNK_ALIGNMENT 4
31 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
32 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
33 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
34 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
35 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
36 #define MIDX_CHUNKID_REVINDEX 0x52494458 /* "RIDX" */
37 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
38 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
39 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
40 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
42 #define PACK_EXPIRED UINT_MAX
44 const unsigned char *get_midx_checksum(struct multi_pack_index *m)
46 return m->data + m->data_len - the_hash_algo->rawsz;
49 void get_midx_filename(struct strbuf *out, const char *object_dir)
51 strbuf_addf(out, "%s/pack/multi-pack-index", object_dir);
54 void get_midx_rev_filename(struct strbuf *out, struct multi_pack_index *m)
56 get_midx_filename(out, m->object_dir);
57 strbuf_addf(out, "-%s.rev", hash_to_hex(get_midx_checksum(m)));
60 static int midx_read_oid_fanout(const unsigned char *chunk_start,
61 size_t chunk_size, void *data)
63 struct multi_pack_index *m = data;
64 m->chunk_oid_fanout = (uint32_t *)chunk_start;
66 if (chunk_size != 4 * 256) {
67 error(_("multi-pack-index OID fanout is of the wrong size"));
68 return 1;
70 return 0;
73 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
75 struct multi_pack_index *m = NULL;
76 int fd;
77 struct stat st;
78 size_t midx_size;
79 void *midx_map = NULL;
80 uint32_t hash_version;
81 struct strbuf midx_name = STRBUF_INIT;
82 uint32_t i;
83 const char *cur_pack_name;
84 struct chunkfile *cf = NULL;
86 get_midx_filename(&midx_name, object_dir);
88 fd = git_open(midx_name.buf);
90 if (fd < 0)
91 goto cleanup_fail;
92 if (fstat(fd, &st)) {
93 error_errno(_("failed to read %s"), midx_name.buf);
94 goto cleanup_fail;
97 midx_size = xsize_t(st.st_size);
99 if (midx_size < MIDX_MIN_SIZE) {
100 error(_("multi-pack-index file %s is too small"), midx_name.buf);
101 goto cleanup_fail;
104 strbuf_release(&midx_name);
106 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
107 close(fd);
109 FLEX_ALLOC_STR(m, object_dir, object_dir);
110 m->data = midx_map;
111 m->data_len = midx_size;
112 m->local = local;
114 m->signature = get_be32(m->data);
115 if (m->signature != MIDX_SIGNATURE)
116 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
117 m->signature, MIDX_SIGNATURE);
119 m->version = m->data[MIDX_BYTE_FILE_VERSION];
120 if (m->version != MIDX_VERSION)
121 die(_("multi-pack-index version %d not recognized"),
122 m->version);
124 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
125 if (hash_version != oid_version(the_hash_algo)) {
126 error(_("multi-pack-index hash version %u does not match version %u"),
127 hash_version, oid_version(the_hash_algo));
128 goto cleanup_fail;
130 m->hash_len = the_hash_algo->rawsz;
132 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
134 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
136 cf = init_chunkfile(NULL);
138 if (read_table_of_contents(cf, m->data, midx_size,
139 MIDX_HEADER_SIZE, m->num_chunks))
140 goto cleanup_fail;
142 if (pair_chunk(cf, MIDX_CHUNKID_PACKNAMES, &m->chunk_pack_names) == CHUNK_NOT_FOUND)
143 die(_("multi-pack-index missing required pack-name chunk"));
144 if (read_chunk(cf, MIDX_CHUNKID_OIDFANOUT, midx_read_oid_fanout, m) == CHUNK_NOT_FOUND)
145 die(_("multi-pack-index missing required OID fanout chunk"));
146 if (pair_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, &m->chunk_oid_lookup) == CHUNK_NOT_FOUND)
147 die(_("multi-pack-index missing required OID lookup chunk"));
148 if (pair_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, &m->chunk_object_offsets) == CHUNK_NOT_FOUND)
149 die(_("multi-pack-index missing required object offsets chunk"));
151 pair_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, &m->chunk_large_offsets);
153 if (git_env_bool("GIT_TEST_MIDX_READ_RIDX", 1))
154 pair_chunk(cf, MIDX_CHUNKID_REVINDEX, &m->chunk_revindex);
156 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
158 CALLOC_ARRAY(m->pack_names, m->num_packs);
159 CALLOC_ARRAY(m->packs, m->num_packs);
161 cur_pack_name = (const char *)m->chunk_pack_names;
162 for (i = 0; i < m->num_packs; i++) {
163 m->pack_names[i] = cur_pack_name;
165 cur_pack_name += strlen(cur_pack_name) + 1;
167 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
168 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
169 m->pack_names[i - 1],
170 m->pack_names[i]);
173 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
174 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
176 free_chunkfile(cf);
177 return m;
179 cleanup_fail:
180 free(m);
181 strbuf_release(&midx_name);
182 free_chunkfile(cf);
183 if (midx_map)
184 munmap(midx_map, midx_size);
185 if (0 <= fd)
186 close(fd);
187 return NULL;
190 void close_midx(struct multi_pack_index *m)
192 uint32_t i;
194 if (!m)
195 return;
197 close_midx(m->next);
199 munmap((unsigned char *)m->data, m->data_len);
201 for (i = 0; i < m->num_packs; i++) {
202 if (m->packs[i])
203 m->packs[i]->multi_pack_index = 0;
205 FREE_AND_NULL(m->packs);
206 FREE_AND_NULL(m->pack_names);
207 free(m);
210 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
212 struct strbuf pack_name = STRBUF_INIT;
213 struct packed_git *p;
215 if (pack_int_id >= m->num_packs)
216 die(_("bad pack-int-id: %u (%u total packs)"),
217 pack_int_id, m->num_packs);
219 if (m->packs[pack_int_id])
220 return 0;
222 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
223 m->pack_names[pack_int_id]);
225 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
226 strbuf_release(&pack_name);
228 if (!p)
229 return 1;
231 p->multi_pack_index = 1;
232 m->packs[pack_int_id] = p;
233 install_packed_git(r, p);
234 list_add_tail(&p->mru, &r->objects->packed_git_mru);
236 return 0;
239 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
241 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
242 the_hash_algo->rawsz, result);
245 struct object_id *nth_midxed_object_oid(struct object_id *oid,
246 struct multi_pack_index *m,
247 uint32_t n)
249 if (n >= m->num_objects)
250 return NULL;
252 oidread(oid, m->chunk_oid_lookup + m->hash_len * n);
253 return oid;
256 off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
258 const unsigned char *offset_data;
259 uint32_t offset32;
261 offset_data = m->chunk_object_offsets + (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH;
262 offset32 = get_be32(offset_data + sizeof(uint32_t));
264 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
265 if (sizeof(off_t) < sizeof(uint64_t))
266 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
268 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
269 return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
272 return offset32;
275 uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
277 return get_be32(m->chunk_object_offsets +
278 (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH);
281 int fill_midx_entry(struct repository *r,
282 const struct object_id *oid,
283 struct pack_entry *e,
284 struct multi_pack_index *m)
286 uint32_t pos;
287 uint32_t pack_int_id;
288 struct packed_git *p;
290 if (!bsearch_midx(oid, m, &pos))
291 return 0;
293 if (pos >= m->num_objects)
294 return 0;
296 pack_int_id = nth_midxed_pack_int_id(m, pos);
298 if (prepare_midx_pack(r, m, pack_int_id))
299 return 0;
300 p = m->packs[pack_int_id];
303 * We are about to tell the caller where they can locate the
304 * requested object. We better make sure the packfile is
305 * still here and can be accessed before supplying that
306 * answer, as it may have been deleted since the MIDX was
307 * loaded!
309 if (!is_pack_valid(p))
310 return 0;
312 if (oidset_size(&p->bad_objects) &&
313 oidset_contains(&p->bad_objects, oid))
314 return 0;
316 e->offset = nth_midxed_offset(m, pos);
317 e->p = p;
319 return 1;
322 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
323 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
324 const char *idx_name)
326 /* Skip past any initial matching prefix. */
327 while (*idx_name && *idx_name == *idx_or_pack_name) {
328 idx_name++;
329 idx_or_pack_name++;
333 * If we didn't match completely, we may have matched "pack-1234." and
334 * be left with "idx" and "pack" respectively, which is also OK. We do
335 * not have to check for "idx" and "idx", because that would have been
336 * a complete match (and in that case these strcmps will be false, but
337 * we'll correctly return 0 from the final strcmp() below.
339 * Technically this matches "fooidx" and "foopack", but we'd never have
340 * such names in the first place.
342 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
343 return 0;
346 * This not only checks for a complete match, but also orders based on
347 * the first non-identical character, which means our ordering will
348 * match a raw strcmp(). That makes it OK to use this to binary search
349 * a naively-sorted list.
351 return strcmp(idx_or_pack_name, idx_name);
354 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
356 uint32_t first = 0, last = m->num_packs;
358 while (first < last) {
359 uint32_t mid = first + (last - first) / 2;
360 const char *current;
361 int cmp;
363 current = m->pack_names[mid];
364 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
365 if (!cmp)
366 return 1;
367 if (cmp > 0) {
368 first = mid + 1;
369 continue;
371 last = mid;
374 return 0;
377 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
379 struct multi_pack_index *m;
380 struct multi_pack_index *m_search;
382 prepare_repo_settings(r);
383 if (!r->settings.core_multi_pack_index)
384 return 0;
386 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
387 if (!strcmp(object_dir, m_search->object_dir))
388 return 1;
390 m = load_multi_pack_index(object_dir, local);
392 if (m) {
393 struct multi_pack_index *mp = r->objects->multi_pack_index;
394 if (mp) {
395 m->next = mp->next;
396 mp->next = m;
397 } else
398 r->objects->multi_pack_index = m;
399 return 1;
402 return 0;
405 static size_t write_midx_header(struct hashfile *f,
406 unsigned char num_chunks,
407 uint32_t num_packs)
409 hashwrite_be32(f, MIDX_SIGNATURE);
410 hashwrite_u8(f, MIDX_VERSION);
411 hashwrite_u8(f, oid_version(the_hash_algo));
412 hashwrite_u8(f, num_chunks);
413 hashwrite_u8(f, 0); /* unused */
414 hashwrite_be32(f, num_packs);
416 return MIDX_HEADER_SIZE;
419 struct pack_info {
420 uint32_t orig_pack_int_id;
421 char *pack_name;
422 struct packed_git *p;
423 unsigned expired : 1;
426 static int pack_info_compare(const void *_a, const void *_b)
428 struct pack_info *a = (struct pack_info *)_a;
429 struct pack_info *b = (struct pack_info *)_b;
430 return strcmp(a->pack_name, b->pack_name);
433 static int idx_or_pack_name_cmp(const void *_va, const void *_vb)
435 const char *pack_name = _va;
436 const struct pack_info *compar = _vb;
438 return cmp_idx_or_pack_name(pack_name, compar->pack_name);
441 struct write_midx_context {
442 struct pack_info *info;
443 uint32_t nr;
444 uint32_t alloc;
445 struct multi_pack_index *m;
446 struct progress *progress;
447 unsigned pack_paths_checked;
449 struct pack_midx_entry *entries;
450 uint32_t entries_nr;
452 uint32_t *pack_perm;
453 uint32_t *pack_order;
454 unsigned large_offsets_needed:1;
455 uint32_t num_large_offsets;
457 int preferred_pack_idx;
459 struct string_list *to_include;
462 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
463 const char *file_name, void *data)
465 struct write_midx_context *ctx = data;
467 if (ends_with(file_name, ".idx")) {
468 display_progress(ctx->progress, ++ctx->pack_paths_checked);
470 * Note that at most one of ctx->m and ctx->to_include are set,
471 * so we are testing midx_contains_pack() and
472 * string_list_has_string() independently (guarded by the
473 * appropriate NULL checks).
475 * We could support passing to_include while reusing an existing
476 * MIDX, but don't currently since the reuse process drags
477 * forward all packs from an existing MIDX (without checking
478 * whether or not they appear in the to_include list).
480 * If we added support for that, these next two conditional
481 * should be performed independently (likely checking
482 * to_include before the existing MIDX).
484 if (ctx->m && midx_contains_pack(ctx->m, file_name))
485 return;
486 else if (ctx->to_include &&
487 !string_list_has_string(ctx->to_include, file_name))
488 return;
490 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
492 ctx->info[ctx->nr].p = add_packed_git(full_path,
493 full_path_len,
496 if (!ctx->info[ctx->nr].p) {
497 warning(_("failed to add packfile '%s'"),
498 full_path);
499 return;
502 if (open_pack_index(ctx->info[ctx->nr].p)) {
503 warning(_("failed to open pack-index '%s'"),
504 full_path);
505 close_pack(ctx->info[ctx->nr].p);
506 FREE_AND_NULL(ctx->info[ctx->nr].p);
507 return;
510 ctx->info[ctx->nr].pack_name = xstrdup(file_name);
511 ctx->info[ctx->nr].orig_pack_int_id = ctx->nr;
512 ctx->info[ctx->nr].expired = 0;
513 ctx->nr++;
517 struct pack_midx_entry {
518 struct object_id oid;
519 uint32_t pack_int_id;
520 time_t pack_mtime;
521 uint64_t offset;
522 unsigned preferred : 1;
525 static int midx_oid_compare(const void *_a, const void *_b)
527 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
528 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
529 int cmp = oidcmp(&a->oid, &b->oid);
531 if (cmp)
532 return cmp;
534 /* Sort objects in a preferred pack first when multiple copies exist. */
535 if (a->preferred > b->preferred)
536 return -1;
537 if (a->preferred < b->preferred)
538 return 1;
540 if (a->pack_mtime > b->pack_mtime)
541 return -1;
542 else if (a->pack_mtime < b->pack_mtime)
543 return 1;
545 return a->pack_int_id - b->pack_int_id;
548 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
549 struct pack_midx_entry *e,
550 uint32_t pos)
552 if (pos >= m->num_objects)
553 return 1;
555 nth_midxed_object_oid(&e->oid, m, pos);
556 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
557 e->offset = nth_midxed_offset(m, pos);
559 /* consider objects in midx to be from "old" packs */
560 e->pack_mtime = 0;
561 return 0;
564 static void fill_pack_entry(uint32_t pack_int_id,
565 struct packed_git *p,
566 uint32_t cur_object,
567 struct pack_midx_entry *entry,
568 int preferred)
570 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
571 die(_("failed to locate object %d in packfile"), cur_object);
573 entry->pack_int_id = pack_int_id;
574 entry->pack_mtime = p->mtime;
576 entry->offset = nth_packed_object_offset(p, cur_object);
577 entry->preferred = !!preferred;
580 struct midx_fanout {
581 struct pack_midx_entry *entries;
582 uint32_t nr;
583 uint32_t alloc;
586 static void midx_fanout_grow(struct midx_fanout *fanout, uint32_t nr)
588 ALLOC_GROW(fanout->entries, nr, fanout->alloc);
591 static void midx_fanout_sort(struct midx_fanout *fanout)
593 QSORT(fanout->entries, fanout->nr, midx_oid_compare);
596 static void midx_fanout_add_midx_fanout(struct midx_fanout *fanout,
597 struct multi_pack_index *m,
598 uint32_t cur_fanout,
599 int preferred_pack)
601 uint32_t start = 0, end;
602 uint32_t cur_object;
604 if (cur_fanout)
605 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
606 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
608 for (cur_object = start; cur_object < end; cur_object++) {
609 if ((preferred_pack > -1) &&
610 (preferred_pack == nth_midxed_pack_int_id(m, cur_object))) {
612 * Objects from preferred packs are added
613 * separately.
615 continue;
618 midx_fanout_grow(fanout, fanout->nr + 1);
619 nth_midxed_pack_midx_entry(m,
620 &fanout->entries[fanout->nr],
621 cur_object);
622 fanout->entries[fanout->nr].preferred = 0;
623 fanout->nr++;
627 static void midx_fanout_add_pack_fanout(struct midx_fanout *fanout,
628 struct pack_info *info,
629 uint32_t cur_pack,
630 int preferred,
631 uint32_t cur_fanout)
633 struct packed_git *pack = info[cur_pack].p;
634 uint32_t start = 0, end;
635 uint32_t cur_object;
637 if (cur_fanout)
638 start = get_pack_fanout(pack, cur_fanout - 1);
639 end = get_pack_fanout(pack, cur_fanout);
641 for (cur_object = start; cur_object < end; cur_object++) {
642 midx_fanout_grow(fanout, fanout->nr + 1);
643 fill_pack_entry(cur_pack,
644 info[cur_pack].p,
645 cur_object,
646 &fanout->entries[fanout->nr],
647 preferred);
648 fanout->nr++;
653 * It is possible to artificially get into a state where there are many
654 * duplicate copies of objects. That can create high memory pressure if
655 * we are to create a list of all objects before de-duplication. To reduce
656 * this memory pressure without a significant performance drop, automatically
657 * group objects by the first byte of their object id. Use the IDX fanout
658 * tables to group the data, copy to a local array, then sort.
660 * Copy only the de-duplicated entries (selected by most-recent modified time
661 * of a packfile containing the object).
663 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
664 struct pack_info *info,
665 uint32_t nr_packs,
666 uint32_t *nr_objects,
667 int preferred_pack)
669 uint32_t cur_fanout, cur_pack, cur_object;
670 uint32_t alloc_objects, total_objects = 0;
671 struct midx_fanout fanout = { 0 };
672 struct pack_midx_entry *deduplicated_entries = NULL;
673 uint32_t start_pack = m ? m->num_packs : 0;
675 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
676 total_objects += info[cur_pack].p->num_objects;
679 * As we de-duplicate by fanout value, we expect the fanout
680 * slices to be evenly distributed, with some noise. Hence,
681 * allocate slightly more than one 256th.
683 alloc_objects = fanout.alloc = total_objects > 3200 ? total_objects / 200 : 16;
685 ALLOC_ARRAY(fanout.entries, fanout.alloc);
686 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
687 *nr_objects = 0;
689 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
690 fanout.nr = 0;
692 if (m)
693 midx_fanout_add_midx_fanout(&fanout, m, cur_fanout,
694 preferred_pack);
696 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
697 int preferred = cur_pack == preferred_pack;
698 midx_fanout_add_pack_fanout(&fanout,
699 info, cur_pack,
700 preferred, cur_fanout);
703 if (-1 < preferred_pack && preferred_pack < start_pack)
704 midx_fanout_add_pack_fanout(&fanout, info,
705 preferred_pack, 1,
706 cur_fanout);
708 midx_fanout_sort(&fanout);
711 * The batch is now sorted by OID and then mtime (descending).
712 * Take only the first duplicate.
714 for (cur_object = 0; cur_object < fanout.nr; cur_object++) {
715 if (cur_object && oideq(&fanout.entries[cur_object - 1].oid,
716 &fanout.entries[cur_object].oid))
717 continue;
719 ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
720 memcpy(&deduplicated_entries[*nr_objects],
721 &fanout.entries[cur_object],
722 sizeof(struct pack_midx_entry));
723 (*nr_objects)++;
727 free(fanout.entries);
728 return deduplicated_entries;
731 static int write_midx_pack_names(struct hashfile *f, void *data)
733 struct write_midx_context *ctx = data;
734 uint32_t i;
735 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
736 size_t written = 0;
738 for (i = 0; i < ctx->nr; i++) {
739 size_t writelen;
741 if (ctx->info[i].expired)
742 continue;
744 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
745 BUG("incorrect pack-file order: %s before %s",
746 ctx->info[i - 1].pack_name,
747 ctx->info[i].pack_name);
749 writelen = strlen(ctx->info[i].pack_name) + 1;
750 hashwrite(f, ctx->info[i].pack_name, writelen);
751 written += writelen;
754 /* add padding to be aligned */
755 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
756 if (i < MIDX_CHUNK_ALIGNMENT) {
757 memset(padding, 0, sizeof(padding));
758 hashwrite(f, padding, i);
761 return 0;
764 static int write_midx_oid_fanout(struct hashfile *f,
765 void *data)
767 struct write_midx_context *ctx = data;
768 struct pack_midx_entry *list = ctx->entries;
769 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
770 uint32_t count = 0;
771 uint32_t i;
774 * Write the first-level table (the list is sorted,
775 * but we use a 256-entry lookup to be able to avoid
776 * having to do eight extra binary search iterations).
778 for (i = 0; i < 256; i++) {
779 struct pack_midx_entry *next = list;
781 while (next < last && next->oid.hash[0] == i) {
782 count++;
783 next++;
786 hashwrite_be32(f, count);
787 list = next;
790 return 0;
793 static int write_midx_oid_lookup(struct hashfile *f,
794 void *data)
796 struct write_midx_context *ctx = data;
797 unsigned char hash_len = the_hash_algo->rawsz;
798 struct pack_midx_entry *list = ctx->entries;
799 uint32_t i;
801 for (i = 0; i < ctx->entries_nr; i++) {
802 struct pack_midx_entry *obj = list++;
804 if (i < ctx->entries_nr - 1) {
805 struct pack_midx_entry *next = list;
806 if (oidcmp(&obj->oid, &next->oid) >= 0)
807 BUG("OIDs not in order: %s >= %s",
808 oid_to_hex(&obj->oid),
809 oid_to_hex(&next->oid));
812 hashwrite(f, obj->oid.hash, (int)hash_len);
815 return 0;
818 static int write_midx_object_offsets(struct hashfile *f,
819 void *data)
821 struct write_midx_context *ctx = data;
822 struct pack_midx_entry *list = ctx->entries;
823 uint32_t i, nr_large_offset = 0;
825 for (i = 0; i < ctx->entries_nr; i++) {
826 struct pack_midx_entry *obj = list++;
828 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
829 BUG("object %s is in an expired pack with int-id %d",
830 oid_to_hex(&obj->oid),
831 obj->pack_int_id);
833 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
835 if (ctx->large_offsets_needed && obj->offset >> 31)
836 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
837 else if (!ctx->large_offsets_needed && obj->offset >> 32)
838 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
839 oid_to_hex(&obj->oid),
840 obj->offset);
841 else
842 hashwrite_be32(f, (uint32_t)obj->offset);
845 return 0;
848 static int write_midx_large_offsets(struct hashfile *f,
849 void *data)
851 struct write_midx_context *ctx = data;
852 struct pack_midx_entry *list = ctx->entries;
853 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
854 uint32_t nr_large_offset = ctx->num_large_offsets;
856 while (nr_large_offset) {
857 struct pack_midx_entry *obj;
858 uint64_t offset;
860 if (list >= end)
861 BUG("too many large-offset objects");
863 obj = list++;
864 offset = obj->offset;
866 if (!(offset >> 31))
867 continue;
869 hashwrite_be64(f, offset);
871 nr_large_offset--;
874 return 0;
877 static int write_midx_revindex(struct hashfile *f,
878 void *data)
880 struct write_midx_context *ctx = data;
881 uint32_t i;
883 for (i = 0; i < ctx->entries_nr; i++)
884 hashwrite_be32(f, ctx->pack_order[i]);
886 return 0;
889 struct midx_pack_order_data {
890 uint32_t nr;
891 uint32_t pack;
892 off_t offset;
895 static int midx_pack_order_cmp(const void *va, const void *vb)
897 const struct midx_pack_order_data *a = va, *b = vb;
898 if (a->pack < b->pack)
899 return -1;
900 else if (a->pack > b->pack)
901 return 1;
902 else if (a->offset < b->offset)
903 return -1;
904 else if (a->offset > b->offset)
905 return 1;
906 else
907 return 0;
910 static uint32_t *midx_pack_order(struct write_midx_context *ctx)
912 struct midx_pack_order_data *data;
913 uint32_t *pack_order;
914 uint32_t i;
916 trace2_region_enter("midx", "midx_pack_order", the_repository);
918 ALLOC_ARRAY(data, ctx->entries_nr);
919 for (i = 0; i < ctx->entries_nr; i++) {
920 struct pack_midx_entry *e = &ctx->entries[i];
921 data[i].nr = i;
922 data[i].pack = ctx->pack_perm[e->pack_int_id];
923 if (!e->preferred)
924 data[i].pack |= (1U << 31);
925 data[i].offset = e->offset;
928 QSORT(data, ctx->entries_nr, midx_pack_order_cmp);
930 ALLOC_ARRAY(pack_order, ctx->entries_nr);
931 for (i = 0; i < ctx->entries_nr; i++)
932 pack_order[i] = data[i].nr;
933 free(data);
935 trace2_region_leave("midx", "midx_pack_order", the_repository);
937 return pack_order;
940 static void write_midx_reverse_index(char *midx_name, unsigned char *midx_hash,
941 struct write_midx_context *ctx)
943 struct strbuf buf = STRBUF_INIT;
944 const char *tmp_file;
946 trace2_region_enter("midx", "write_midx_reverse_index", the_repository);
948 strbuf_addf(&buf, "%s-%s.rev", midx_name, hash_to_hex(midx_hash));
950 tmp_file = write_rev_file_order(NULL, ctx->pack_order, ctx->entries_nr,
951 midx_hash, WRITE_REV);
953 if (finalize_object_file(tmp_file, buf.buf))
954 die(_("cannot store reverse index file"));
956 strbuf_release(&buf);
958 trace2_region_leave("midx", "write_midx_reverse_index", the_repository);
961 static void clear_midx_files_ext(const char *object_dir, const char *ext,
962 unsigned char *keep_hash);
964 static int midx_checksum_valid(struct multi_pack_index *m)
966 return hashfile_checksum_valid(m->data, m->data_len);
969 static void prepare_midx_packing_data(struct packing_data *pdata,
970 struct write_midx_context *ctx)
972 uint32_t i;
974 trace2_region_enter("midx", "prepare_midx_packing_data", the_repository);
976 memset(pdata, 0, sizeof(struct packing_data));
977 prepare_packing_data(the_repository, pdata);
979 for (i = 0; i < ctx->entries_nr; i++) {
980 struct pack_midx_entry *from = &ctx->entries[ctx->pack_order[i]];
981 struct object_entry *to = packlist_alloc(pdata, &from->oid);
983 oe_set_in_pack(pdata, to,
984 ctx->info[ctx->pack_perm[from->pack_int_id]].p);
987 trace2_region_leave("midx", "prepare_midx_packing_data", the_repository);
990 static int add_ref_to_pending(const char *refname,
991 const struct object_id *oid,
992 int flag, void *cb_data)
994 struct rev_info *revs = (struct rev_info*)cb_data;
995 struct object_id peeled;
996 struct object *object;
998 if ((flag & REF_ISSYMREF) && (flag & REF_ISBROKEN)) {
999 warning("symbolic ref is dangling: %s", refname);
1000 return 0;
1003 if (!peel_iterated_oid(oid, &peeled))
1004 oid = &peeled;
1006 object = parse_object_or_die(oid, refname);
1007 if (object->type != OBJ_COMMIT)
1008 return 0;
1010 add_pending_object(revs, object, "");
1011 if (bitmap_is_preferred_refname(revs->repo, refname))
1012 object->flags |= NEEDS_BITMAP;
1013 return 0;
1016 struct bitmap_commit_cb {
1017 struct commit **commits;
1018 size_t commits_nr, commits_alloc;
1020 struct write_midx_context *ctx;
1023 static const struct object_id *bitmap_oid_access(size_t index,
1024 const void *_entries)
1026 const struct pack_midx_entry *entries = _entries;
1027 return &entries[index].oid;
1030 static void bitmap_show_commit(struct commit *commit, void *_data)
1032 struct bitmap_commit_cb *data = _data;
1033 int pos = oid_pos(&commit->object.oid, data->ctx->entries,
1034 data->ctx->entries_nr,
1035 bitmap_oid_access);
1036 if (pos < 0)
1037 return;
1039 ALLOC_GROW(data->commits, data->commits_nr + 1, data->commits_alloc);
1040 data->commits[data->commits_nr++] = commit;
1043 static int read_refs_snapshot(const char *refs_snapshot,
1044 struct rev_info *revs)
1046 struct strbuf buf = STRBUF_INIT;
1047 struct object_id oid;
1048 FILE *f = xfopen(refs_snapshot, "r");
1050 while (strbuf_getline(&buf, f) != EOF) {
1051 struct object *object;
1052 int preferred = 0;
1053 char *hex = buf.buf;
1054 const char *end = NULL;
1056 if (buf.len && *buf.buf == '+') {
1057 preferred = 1;
1058 hex = &buf.buf[1];
1061 if (parse_oid_hex(hex, &oid, &end) < 0)
1062 die(_("could not parse line: %s"), buf.buf);
1063 if (*end)
1064 die(_("malformed line: %s"), buf.buf);
1066 object = parse_object_or_die(&oid, NULL);
1067 if (preferred)
1068 object->flags |= NEEDS_BITMAP;
1070 add_pending_object(revs, object, "");
1073 fclose(f);
1074 strbuf_release(&buf);
1075 return 0;
1078 static struct commit **find_commits_for_midx_bitmap(uint32_t *indexed_commits_nr_p,
1079 const char *refs_snapshot,
1080 struct write_midx_context *ctx)
1082 struct rev_info revs;
1083 struct bitmap_commit_cb cb = {0};
1085 trace2_region_enter("midx", "find_commits_for_midx_bitmap",
1086 the_repository);
1088 cb.ctx = ctx;
1090 repo_init_revisions(the_repository, &revs, NULL);
1091 if (refs_snapshot) {
1092 read_refs_snapshot(refs_snapshot, &revs);
1093 } else {
1094 setup_revisions(0, NULL, &revs, NULL);
1095 for_each_ref(add_ref_to_pending, &revs);
1099 * Skipping promisor objects here is intentional, since it only excludes
1100 * them from the list of reachable commits that we want to select from
1101 * when computing the selection of MIDX'd commits to receive bitmaps.
1103 * Reachability bitmaps do require that their objects be closed under
1104 * reachability, but fetching any objects missing from promisors at this
1105 * point is too late. But, if one of those objects can be reached from
1106 * an another object that is included in the bitmap, then we will
1107 * complain later that we don't have reachability closure (and fail
1108 * appropriately).
1110 fetch_if_missing = 0;
1111 revs.exclude_promisor_objects = 1;
1113 if (prepare_revision_walk(&revs))
1114 die(_("revision walk setup failed"));
1116 traverse_commit_list(&revs, bitmap_show_commit, NULL, &cb);
1117 if (indexed_commits_nr_p)
1118 *indexed_commits_nr_p = cb.commits_nr;
1120 release_revisions(&revs);
1122 trace2_region_leave("midx", "find_commits_for_midx_bitmap",
1123 the_repository);
1125 return cb.commits;
1128 static int write_midx_bitmap(const char *midx_name,
1129 const unsigned char *midx_hash,
1130 struct packing_data *pdata,
1131 struct commit **commits,
1132 uint32_t commits_nr,
1133 uint32_t *pack_order,
1134 unsigned flags)
1136 int ret, i;
1137 uint16_t options = 0;
1138 struct pack_idx_entry **index;
1139 char *bitmap_name = xstrfmt("%s-%s.bitmap", midx_name,
1140 hash_to_hex(midx_hash));
1142 trace2_region_enter("midx", "write_midx_bitmap", the_repository);
1144 if (flags & MIDX_WRITE_BITMAP_HASH_CACHE)
1145 options |= BITMAP_OPT_HASH_CACHE;
1147 if (flags & MIDX_WRITE_BITMAP_LOOKUP_TABLE)
1148 options |= BITMAP_OPT_LOOKUP_TABLE;
1151 * Build the MIDX-order index based on pdata.objects (which is already
1152 * in MIDX order; c.f., 'midx_pack_order_cmp()' for the definition of
1153 * this order).
1155 ALLOC_ARRAY(index, pdata->nr_objects);
1156 for (i = 0; i < pdata->nr_objects; i++)
1157 index[i] = &pdata->objects[i].idx;
1159 bitmap_writer_show_progress(flags & MIDX_PROGRESS);
1160 bitmap_writer_build_type_index(pdata, index, pdata->nr_objects);
1163 * bitmap_writer_finish expects objects in lex order, but pack_order
1164 * gives us exactly that. use it directly instead of re-sorting the
1165 * array.
1167 * This changes the order of objects in 'index' between
1168 * bitmap_writer_build_type_index and bitmap_writer_finish.
1170 * The same re-ordering takes place in the single-pack bitmap code via
1171 * write_idx_file(), which is called by finish_tmp_packfile(), which
1172 * happens between bitmap_writer_build_type_index() and
1173 * bitmap_writer_finish().
1175 for (i = 0; i < pdata->nr_objects; i++)
1176 index[pack_order[i]] = &pdata->objects[i].idx;
1178 bitmap_writer_select_commits(commits, commits_nr, -1);
1179 ret = bitmap_writer_build(pdata);
1180 if (ret < 0)
1181 goto cleanup;
1183 bitmap_writer_set_checksum(midx_hash);
1184 bitmap_writer_finish(index, pdata->nr_objects, bitmap_name, options);
1186 cleanup:
1187 free(index);
1188 free(bitmap_name);
1190 trace2_region_leave("midx", "write_midx_bitmap", the_repository);
1192 return ret;
1195 static struct multi_pack_index *lookup_multi_pack_index(struct repository *r,
1196 const char *object_dir)
1198 struct multi_pack_index *result = NULL;
1199 struct multi_pack_index *cur;
1200 char *obj_dir_real = real_pathdup(object_dir, 1);
1201 struct strbuf cur_path_real = STRBUF_INIT;
1203 /* Ensure the given object_dir is local, or a known alternate. */
1204 find_odb(r, obj_dir_real);
1206 for (cur = get_multi_pack_index(r); cur; cur = cur->next) {
1207 strbuf_realpath(&cur_path_real, cur->object_dir, 1);
1208 if (!strcmp(obj_dir_real, cur_path_real.buf)) {
1209 result = cur;
1210 goto cleanup;
1214 cleanup:
1215 free(obj_dir_real);
1216 strbuf_release(&cur_path_real);
1217 return result;
1220 static int write_midx_internal(const char *object_dir,
1221 struct string_list *packs_to_include,
1222 struct string_list *packs_to_drop,
1223 const char *preferred_pack_name,
1224 const char *refs_snapshot,
1225 unsigned flags)
1227 struct strbuf midx_name = STRBUF_INIT;
1228 unsigned char midx_hash[GIT_MAX_RAWSZ];
1229 uint32_t i;
1230 struct hashfile *f = NULL;
1231 struct lock_file lk;
1232 struct write_midx_context ctx = { 0 };
1233 int pack_name_concat_len = 0;
1234 int dropped_packs = 0;
1235 int result = 0;
1236 struct chunkfile *cf;
1238 trace2_region_enter("midx", "write_midx_internal", the_repository);
1240 get_midx_filename(&midx_name, object_dir);
1241 if (safe_create_leading_directories(midx_name.buf))
1242 die_errno(_("unable to create leading directories of %s"),
1243 midx_name.buf);
1245 if (!packs_to_include) {
1247 * Only reference an existing MIDX when not filtering which
1248 * packs to include, since all packs and objects are copied
1249 * blindly from an existing MIDX if one is present.
1251 ctx.m = lookup_multi_pack_index(the_repository, object_dir);
1254 if (ctx.m && !midx_checksum_valid(ctx.m)) {
1255 warning(_("ignoring existing multi-pack-index; checksum mismatch"));
1256 ctx.m = NULL;
1259 ctx.nr = 0;
1260 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
1261 ctx.info = NULL;
1262 ALLOC_ARRAY(ctx.info, ctx.alloc);
1264 if (ctx.m) {
1265 for (i = 0; i < ctx.m->num_packs; i++) {
1266 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
1268 ctx.info[ctx.nr].orig_pack_int_id = i;
1269 ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]);
1270 ctx.info[ctx.nr].p = ctx.m->packs[i];
1271 ctx.info[ctx.nr].expired = 0;
1273 if (flags & MIDX_WRITE_REV_INDEX) {
1275 * If generating a reverse index, need to have
1276 * packed_git's loaded to compare their
1277 * mtimes and object count.
1279 if (prepare_midx_pack(the_repository, ctx.m, i)) {
1280 error(_("could not load pack"));
1281 result = 1;
1282 goto cleanup;
1285 if (open_pack_index(ctx.m->packs[i]))
1286 die(_("could not open index for %s"),
1287 ctx.m->packs[i]->pack_name);
1288 ctx.info[ctx.nr].p = ctx.m->packs[i];
1291 ctx.nr++;
1295 ctx.pack_paths_checked = 0;
1296 if (flags & MIDX_PROGRESS)
1297 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
1298 else
1299 ctx.progress = NULL;
1301 ctx.to_include = packs_to_include;
1303 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
1304 stop_progress(&ctx.progress);
1306 if ((ctx.m && ctx.nr == ctx.m->num_packs) &&
1307 !(packs_to_include || packs_to_drop)) {
1308 struct bitmap_index *bitmap_git;
1309 int bitmap_exists;
1310 int want_bitmap = flags & MIDX_WRITE_BITMAP;
1312 bitmap_git = prepare_midx_bitmap_git(ctx.m);
1313 bitmap_exists = bitmap_git && bitmap_is_midx(bitmap_git);
1314 free_bitmap_index(bitmap_git);
1316 if (bitmap_exists || !want_bitmap) {
1318 * The correct MIDX already exists, and so does a
1319 * corresponding bitmap (or one wasn't requested).
1321 if (!want_bitmap)
1322 clear_midx_files_ext(object_dir, ".bitmap",
1323 NULL);
1324 goto cleanup;
1328 if (preferred_pack_name) {
1329 int found = 0;
1330 for (i = 0; i < ctx.nr; i++) {
1331 if (!cmp_idx_or_pack_name(preferred_pack_name,
1332 ctx.info[i].pack_name)) {
1333 ctx.preferred_pack_idx = i;
1334 found = 1;
1335 break;
1339 if (!found)
1340 warning(_("unknown preferred pack: '%s'"),
1341 preferred_pack_name);
1342 } else if (ctx.nr &&
1343 (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP))) {
1344 struct packed_git *oldest = ctx.info[ctx.preferred_pack_idx].p;
1345 ctx.preferred_pack_idx = 0;
1347 if (packs_to_drop && packs_to_drop->nr)
1348 BUG("cannot write a MIDX bitmap during expiration");
1351 * set a preferred pack when writing a bitmap to ensure that
1352 * the pack from which the first object is selected in pseudo
1353 * pack-order has all of its objects selected from that pack
1354 * (and not another pack containing a duplicate)
1356 for (i = 1; i < ctx.nr; i++) {
1357 struct packed_git *p = ctx.info[i].p;
1359 if (!oldest->num_objects || p->mtime < oldest->mtime) {
1360 oldest = p;
1361 ctx.preferred_pack_idx = i;
1365 if (!oldest->num_objects) {
1367 * If all packs are empty; unset the preferred index.
1368 * This is acceptable since there will be no duplicate
1369 * objects to resolve, so the preferred value doesn't
1370 * matter.
1372 ctx.preferred_pack_idx = -1;
1374 } else {
1376 * otherwise don't mark any pack as preferred to avoid
1377 * interfering with expiration logic below
1379 ctx.preferred_pack_idx = -1;
1382 if (ctx.preferred_pack_idx > -1) {
1383 struct packed_git *preferred = ctx.info[ctx.preferred_pack_idx].p;
1384 if (!preferred->num_objects) {
1385 error(_("cannot select preferred pack %s with no objects"),
1386 preferred->pack_name);
1387 result = 1;
1388 goto cleanup;
1392 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr,
1393 ctx.preferred_pack_idx);
1395 ctx.large_offsets_needed = 0;
1396 for (i = 0; i < ctx.entries_nr; i++) {
1397 if (ctx.entries[i].offset > 0x7fffffff)
1398 ctx.num_large_offsets++;
1399 if (ctx.entries[i].offset > 0xffffffff)
1400 ctx.large_offsets_needed = 1;
1403 QSORT(ctx.info, ctx.nr, pack_info_compare);
1405 if (packs_to_drop && packs_to_drop->nr) {
1406 int drop_index = 0;
1407 int missing_drops = 0;
1409 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
1410 int cmp = strcmp(ctx.info[i].pack_name,
1411 packs_to_drop->items[drop_index].string);
1413 if (!cmp) {
1414 drop_index++;
1415 ctx.info[i].expired = 1;
1416 } else if (cmp > 0) {
1417 error(_("did not see pack-file %s to drop"),
1418 packs_to_drop->items[drop_index].string);
1419 drop_index++;
1420 missing_drops++;
1421 i--;
1422 } else {
1423 ctx.info[i].expired = 0;
1427 if (missing_drops) {
1428 result = 1;
1429 goto cleanup;
1434 * pack_perm stores a permutation between pack-int-ids from the
1435 * previous multi-pack-index to the new one we are writing:
1437 * pack_perm[old_id] = new_id
1439 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
1440 for (i = 0; i < ctx.nr; i++) {
1441 if (ctx.info[i].expired) {
1442 dropped_packs++;
1443 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
1444 } else {
1445 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
1449 for (i = 0; i < ctx.nr; i++) {
1450 if (!ctx.info[i].expired)
1451 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
1454 /* Check that the preferred pack wasn't expired (if given). */
1455 if (preferred_pack_name) {
1456 struct pack_info *preferred = bsearch(preferred_pack_name,
1457 ctx.info, ctx.nr,
1458 sizeof(*ctx.info),
1459 idx_or_pack_name_cmp);
1460 if (preferred) {
1461 uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id];
1462 if (perm == PACK_EXPIRED)
1463 warning(_("preferred pack '%s' is expired"),
1464 preferred_pack_name);
1468 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
1469 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
1470 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
1472 hold_lock_file_for_update(&lk, midx_name.buf, LOCK_DIE_ON_ERROR);
1473 f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk));
1475 if (ctx.nr - dropped_packs == 0) {
1476 error(_("no pack files to index."));
1477 result = 1;
1478 goto cleanup;
1481 if (!ctx.entries_nr) {
1482 if (flags & MIDX_WRITE_BITMAP)
1483 warning(_("refusing to write multi-pack .bitmap without any objects"));
1484 flags &= ~(MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP);
1487 cf = init_chunkfile(f);
1489 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
1490 write_midx_pack_names);
1491 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
1492 write_midx_oid_fanout);
1493 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
1494 (size_t)ctx.entries_nr * the_hash_algo->rawsz,
1495 write_midx_oid_lookup);
1496 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
1497 (size_t)ctx.entries_nr * MIDX_CHUNK_OFFSET_WIDTH,
1498 write_midx_object_offsets);
1500 if (ctx.large_offsets_needed)
1501 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
1502 (size_t)ctx.num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH,
1503 write_midx_large_offsets);
1505 if (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP)) {
1506 ctx.pack_order = midx_pack_order(&ctx);
1507 add_chunk(cf, MIDX_CHUNKID_REVINDEX,
1508 ctx.entries_nr * sizeof(uint32_t),
1509 write_midx_revindex);
1512 write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs);
1513 write_chunkfile(cf, &ctx);
1515 finalize_hashfile(f, midx_hash, FSYNC_COMPONENT_PACK_METADATA,
1516 CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1517 free_chunkfile(cf);
1519 if (flags & MIDX_WRITE_REV_INDEX &&
1520 git_env_bool("GIT_TEST_MIDX_WRITE_REV", 0))
1521 write_midx_reverse_index(midx_name.buf, midx_hash, &ctx);
1523 if (flags & MIDX_WRITE_BITMAP) {
1524 struct packing_data pdata;
1525 struct commit **commits;
1526 uint32_t commits_nr;
1528 if (!ctx.entries_nr)
1529 BUG("cannot write a bitmap without any objects");
1531 prepare_midx_packing_data(&pdata, &ctx);
1533 commits = find_commits_for_midx_bitmap(&commits_nr, refs_snapshot, &ctx);
1536 * The previous steps translated the information from
1537 * 'entries' into information suitable for constructing
1538 * bitmaps. We no longer need that array, so clear it to
1539 * reduce memory pressure.
1541 FREE_AND_NULL(ctx.entries);
1542 ctx.entries_nr = 0;
1544 if (write_midx_bitmap(midx_name.buf, midx_hash, &pdata,
1545 commits, commits_nr, ctx.pack_order,
1546 flags) < 0) {
1547 error(_("could not write multi-pack bitmap"));
1548 result = 1;
1549 goto cleanup;
1553 * NOTE: Do not use ctx.entries beyond this point, since it might
1554 * have been freed in the previous if block.
1557 if (ctx.m)
1558 close_object_store(the_repository->objects);
1560 if (commit_lock_file(&lk) < 0)
1561 die_errno(_("could not write multi-pack-index"));
1563 clear_midx_files_ext(object_dir, ".bitmap", midx_hash);
1564 clear_midx_files_ext(object_dir, ".rev", midx_hash);
1566 cleanup:
1567 for (i = 0; i < ctx.nr; i++) {
1568 if (ctx.info[i].p) {
1569 close_pack(ctx.info[i].p);
1570 free(ctx.info[i].p);
1572 free(ctx.info[i].pack_name);
1575 free(ctx.info);
1576 free(ctx.entries);
1577 free(ctx.pack_perm);
1578 free(ctx.pack_order);
1579 strbuf_release(&midx_name);
1581 trace2_region_leave("midx", "write_midx_internal", the_repository);
1583 return result;
1586 int write_midx_file(const char *object_dir,
1587 const char *preferred_pack_name,
1588 const char *refs_snapshot,
1589 unsigned flags)
1591 return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name,
1592 refs_snapshot, flags);
1595 int write_midx_file_only(const char *object_dir,
1596 struct string_list *packs_to_include,
1597 const char *preferred_pack_name,
1598 const char *refs_snapshot,
1599 unsigned flags)
1601 return write_midx_internal(object_dir, packs_to_include, NULL,
1602 preferred_pack_name, refs_snapshot, flags);
1605 struct clear_midx_data {
1606 char *keep;
1607 const char *ext;
1610 static void clear_midx_file_ext(const char *full_path, size_t full_path_len,
1611 const char *file_name, void *_data)
1613 struct clear_midx_data *data = _data;
1615 if (!(starts_with(file_name, "multi-pack-index-") &&
1616 ends_with(file_name, data->ext)))
1617 return;
1618 if (data->keep && !strcmp(data->keep, file_name))
1619 return;
1621 if (unlink(full_path))
1622 die_errno(_("failed to remove %s"), full_path);
1625 static void clear_midx_files_ext(const char *object_dir, const char *ext,
1626 unsigned char *keep_hash)
1628 struct clear_midx_data data;
1629 memset(&data, 0, sizeof(struct clear_midx_data));
1631 if (keep_hash)
1632 data.keep = xstrfmt("multi-pack-index-%s%s",
1633 hash_to_hex(keep_hash), ext);
1634 data.ext = ext;
1636 for_each_file_in_pack_dir(object_dir,
1637 clear_midx_file_ext,
1638 &data);
1640 free(data.keep);
1643 void clear_midx_file(struct repository *r)
1645 struct strbuf midx = STRBUF_INIT;
1647 get_midx_filename(&midx, r->objects->odb->path);
1649 if (r->objects && r->objects->multi_pack_index) {
1650 close_midx(r->objects->multi_pack_index);
1651 r->objects->multi_pack_index = NULL;
1654 if (remove_path(midx.buf))
1655 die(_("failed to clear multi-pack-index at %s"), midx.buf);
1657 clear_midx_files_ext(r->objects->odb->path, ".bitmap", NULL);
1658 clear_midx_files_ext(r->objects->odb->path, ".rev", NULL);
1660 strbuf_release(&midx);
1663 static int verify_midx_error;
1665 __attribute__((format (printf, 1, 2)))
1666 static void midx_report(const char *fmt, ...)
1668 va_list ap;
1669 verify_midx_error = 1;
1670 va_start(ap, fmt);
1671 vfprintf(stderr, fmt, ap);
1672 fprintf(stderr, "\n");
1673 va_end(ap);
1676 struct pair_pos_vs_id
1678 uint32_t pos;
1679 uint32_t pack_int_id;
1682 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1684 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1685 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1687 return b->pack_int_id - a->pack_int_id;
1691 * Limit calls to display_progress() for performance reasons.
1692 * The interval here was arbitrarily chosen.
1694 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1695 #define midx_display_sparse_progress(progress, n) \
1696 do { \
1697 uint64_t _n = (n); \
1698 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1699 display_progress(progress, _n); \
1700 } while (0)
1702 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1704 struct pair_pos_vs_id *pairs = NULL;
1705 uint32_t i;
1706 struct progress *progress = NULL;
1707 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1708 verify_midx_error = 0;
1710 if (!m) {
1711 int result = 0;
1712 struct stat sb;
1713 struct strbuf filename = STRBUF_INIT;
1715 get_midx_filename(&filename, object_dir);
1717 if (!stat(filename.buf, &sb)) {
1718 error(_("multi-pack-index file exists, but failed to parse"));
1719 result = 1;
1721 strbuf_release(&filename);
1722 return result;
1725 if (!midx_checksum_valid(m))
1726 midx_report(_("incorrect checksum"));
1728 if (flags & MIDX_PROGRESS)
1729 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1730 m->num_packs);
1731 for (i = 0; i < m->num_packs; i++) {
1732 if (prepare_midx_pack(r, m, i))
1733 midx_report("failed to load pack in position %d", i);
1735 display_progress(progress, i + 1);
1737 stop_progress(&progress);
1739 for (i = 0; i < 255; i++) {
1740 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1741 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1743 if (oid_fanout1 > oid_fanout2)
1744 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1745 i, oid_fanout1, oid_fanout2, i + 1);
1748 if (m->num_objects == 0) {
1749 midx_report(_("the midx contains no oid"));
1751 * Remaining tests assume that we have objects, so we can
1752 * return here.
1754 goto cleanup;
1757 if (flags & MIDX_PROGRESS)
1758 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1759 m->num_objects - 1);
1760 for (i = 0; i < m->num_objects - 1; i++) {
1761 struct object_id oid1, oid2;
1763 nth_midxed_object_oid(&oid1, m, i);
1764 nth_midxed_object_oid(&oid2, m, i + 1);
1766 if (oidcmp(&oid1, &oid2) >= 0)
1767 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1768 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1770 midx_display_sparse_progress(progress, i + 1);
1772 stop_progress(&progress);
1775 * Create an array mapping each object to its packfile id. Sort it
1776 * to group the objects by packfile. Use this permutation to visit
1777 * each of the objects and only require 1 packfile to be open at a
1778 * time.
1780 ALLOC_ARRAY(pairs, m->num_objects);
1781 for (i = 0; i < m->num_objects; i++) {
1782 pairs[i].pos = i;
1783 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1786 if (flags & MIDX_PROGRESS)
1787 progress = start_sparse_progress(_("Sorting objects by packfile"),
1788 m->num_objects);
1789 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1790 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1791 stop_progress(&progress);
1793 if (flags & MIDX_PROGRESS)
1794 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1795 for (i = 0; i < m->num_objects; i++) {
1796 struct object_id oid;
1797 struct pack_entry e;
1798 off_t m_offset, p_offset;
1800 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1801 m->packs[pairs[i-1].pack_int_id])
1803 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1804 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1807 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1809 if (!fill_midx_entry(r, &oid, &e, m)) {
1810 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1811 pairs[i].pos, oid_to_hex(&oid));
1812 continue;
1815 if (open_pack_index(e.p)) {
1816 midx_report(_("failed to load pack-index for packfile %s"),
1817 e.p->pack_name);
1818 break;
1821 m_offset = e.offset;
1822 p_offset = find_pack_entry_one(oid.hash, e.p);
1824 if (m_offset != p_offset)
1825 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1826 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1828 midx_display_sparse_progress(progress, i + 1);
1830 stop_progress(&progress);
1832 cleanup:
1833 free(pairs);
1834 close_midx(m);
1836 return verify_midx_error;
1839 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1841 uint32_t i, *count, result = 0;
1842 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1843 struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir);
1844 struct progress *progress = NULL;
1846 if (!m)
1847 return 0;
1849 CALLOC_ARRAY(count, m->num_packs);
1851 if (flags & MIDX_PROGRESS)
1852 progress = start_delayed_progress(_("Counting referenced objects"),
1853 m->num_objects);
1854 for (i = 0; i < m->num_objects; i++) {
1855 int pack_int_id = nth_midxed_pack_int_id(m, i);
1856 count[pack_int_id]++;
1857 display_progress(progress, i + 1);
1859 stop_progress(&progress);
1861 if (flags & MIDX_PROGRESS)
1862 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1863 m->num_packs);
1864 for (i = 0; i < m->num_packs; i++) {
1865 char *pack_name;
1866 display_progress(progress, i + 1);
1868 if (count[i])
1869 continue;
1871 if (prepare_midx_pack(r, m, i))
1872 continue;
1874 if (m->packs[i]->pack_keep || m->packs[i]->is_cruft)
1875 continue;
1877 pack_name = xstrdup(m->packs[i]->pack_name);
1878 close_pack(m->packs[i]);
1880 string_list_insert(&packs_to_drop, m->pack_names[i]);
1881 unlink_pack_path(pack_name, 0);
1882 free(pack_name);
1884 stop_progress(&progress);
1886 free(count);
1888 if (packs_to_drop.nr)
1889 result = write_midx_internal(object_dir, NULL, &packs_to_drop, NULL, NULL, flags);
1891 string_list_clear(&packs_to_drop, 0);
1893 return result;
1896 struct repack_info {
1897 timestamp_t mtime;
1898 uint32_t referenced_objects;
1899 uint32_t pack_int_id;
1902 static int compare_by_mtime(const void *a_, const void *b_)
1904 const struct repack_info *a, *b;
1906 a = (const struct repack_info *)a_;
1907 b = (const struct repack_info *)b_;
1909 if (a->mtime < b->mtime)
1910 return -1;
1911 if (a->mtime > b->mtime)
1912 return 1;
1913 return 0;
1916 static int fill_included_packs_all(struct repository *r,
1917 struct multi_pack_index *m,
1918 unsigned char *include_pack)
1920 uint32_t i, count = 0;
1921 int pack_kept_objects = 0;
1923 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1925 for (i = 0; i < m->num_packs; i++) {
1926 if (prepare_midx_pack(r, m, i))
1927 continue;
1928 if (!pack_kept_objects && m->packs[i]->pack_keep)
1929 continue;
1930 if (m->packs[i]->is_cruft)
1931 continue;
1933 include_pack[i] = 1;
1934 count++;
1937 return count < 2;
1940 static int fill_included_packs_batch(struct repository *r,
1941 struct multi_pack_index *m,
1942 unsigned char *include_pack,
1943 size_t batch_size)
1945 uint32_t i, packs_to_repack;
1946 size_t total_size;
1947 struct repack_info *pack_info;
1948 int pack_kept_objects = 0;
1950 CALLOC_ARRAY(pack_info, m->num_packs);
1952 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1954 for (i = 0; i < m->num_packs; i++) {
1955 pack_info[i].pack_int_id = i;
1957 if (prepare_midx_pack(r, m, i))
1958 continue;
1960 pack_info[i].mtime = m->packs[i]->mtime;
1963 for (i = 0; i < m->num_objects; i++) {
1964 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1965 pack_info[pack_int_id].referenced_objects++;
1968 QSORT(pack_info, m->num_packs, compare_by_mtime);
1970 total_size = 0;
1971 packs_to_repack = 0;
1972 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1973 int pack_int_id = pack_info[i].pack_int_id;
1974 struct packed_git *p = m->packs[pack_int_id];
1975 size_t expected_size;
1977 if (!p)
1978 continue;
1979 if (!pack_kept_objects && p->pack_keep)
1980 continue;
1981 if (p->is_cruft)
1982 continue;
1983 if (open_pack_index(p) || !p->num_objects)
1984 continue;
1986 expected_size = (size_t)(p->pack_size
1987 * pack_info[i].referenced_objects);
1988 expected_size /= p->num_objects;
1990 if (expected_size >= batch_size)
1991 continue;
1993 packs_to_repack++;
1994 total_size += expected_size;
1995 include_pack[pack_int_id] = 1;
1998 free(pack_info);
2000 if (packs_to_repack < 2)
2001 return 1;
2003 return 0;
2006 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
2008 int result = 0;
2009 uint32_t i;
2010 unsigned char *include_pack;
2011 struct child_process cmd = CHILD_PROCESS_INIT;
2012 FILE *cmd_in;
2013 struct strbuf base_name = STRBUF_INIT;
2014 struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir);
2017 * When updating the default for these configuration
2018 * variables in builtin/repack.c, these must be adjusted
2019 * to match.
2021 int delta_base_offset = 1;
2022 int use_delta_islands = 0;
2024 if (!m)
2025 return 0;
2027 CALLOC_ARRAY(include_pack, m->num_packs);
2029 if (batch_size) {
2030 if (fill_included_packs_batch(r, m, include_pack, batch_size))
2031 goto cleanup;
2032 } else if (fill_included_packs_all(r, m, include_pack))
2033 goto cleanup;
2035 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
2036 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
2038 strvec_push(&cmd.args, "pack-objects");
2040 strbuf_addstr(&base_name, object_dir);
2041 strbuf_addstr(&base_name, "/pack/pack");
2042 strvec_push(&cmd.args, base_name.buf);
2044 if (delta_base_offset)
2045 strvec_push(&cmd.args, "--delta-base-offset");
2046 if (use_delta_islands)
2047 strvec_push(&cmd.args, "--delta-islands");
2049 if (flags & MIDX_PROGRESS)
2050 strvec_push(&cmd.args, "--progress");
2051 else
2052 strvec_push(&cmd.args, "-q");
2054 strbuf_release(&base_name);
2056 cmd.git_cmd = 1;
2057 cmd.in = cmd.out = -1;
2059 if (start_command(&cmd)) {
2060 error(_("could not start pack-objects"));
2061 result = 1;
2062 goto cleanup;
2065 cmd_in = xfdopen(cmd.in, "w");
2067 for (i = 0; i < m->num_objects; i++) {
2068 struct object_id oid;
2069 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
2071 if (!include_pack[pack_int_id])
2072 continue;
2074 nth_midxed_object_oid(&oid, m, i);
2075 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
2077 fclose(cmd_in);
2079 if (finish_command(&cmd)) {
2080 error(_("could not finish pack-objects"));
2081 result = 1;
2082 goto cleanup;
2085 result = write_midx_internal(object_dir, NULL, NULL, NULL, NULL, flags);
2087 cleanup:
2088 free(include_pack);
2089 return result;