freshlybakedca.ke
Git fork
1#include "git-compat-util.h"
2#include "abspath.h"
3#include "config.h"
4#include "hex.h"
5#include "lockfile.h"
6#include "packfile.h"
7#include "object-file.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 "chunk-format.h"
14#include "pack-bitmap.h"
15#include "refs.h"
16#include "revision.h"
17#include "list-objects.h"
18#include "path.h"
19#include "pack-revindex.h"
20
21#define PACK_EXPIRED UINT_MAX
22#define BITMAP_POS_UNKNOWN (~((uint32_t)0))
23#define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
24#define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
25#define NO_PREFERRED_PACK (~((uint32_t)0))
26
27extern int midx_checksum_valid(struct multi_pack_index *m);
28extern void clear_midx_files_ext(struct odb_source *source, const char *ext,
29 const char *keep_hash);
30extern void clear_incremental_midx_files_ext(struct odb_source *source,
31 const char *ext,
32 const char **keep_hashes,
33 uint32_t hashes_nr);
34extern int cmp_idx_or_pack_name(const char *idx_or_pack_name,
35 const char *idx_name);
36
37static size_t write_midx_header(const struct git_hash_algo *hash_algo,
38 struct hashfile *f, unsigned char num_chunks,
39 uint32_t num_packs)
40{
41 hashwrite_be32(f, MIDX_SIGNATURE);
42 hashwrite_u8(f, MIDX_VERSION);
43 hashwrite_u8(f, oid_version(hash_algo));
44 hashwrite_u8(f, num_chunks);
45 hashwrite_u8(f, 0); /* unused */
46 hashwrite_be32(f, num_packs);
47
48 return MIDX_HEADER_SIZE;
49}
50
51struct pack_info {
52 uint32_t orig_pack_int_id;
53 char *pack_name;
54 struct packed_git *p;
55
56 uint32_t bitmap_pos;
57 uint32_t bitmap_nr;
58
59 unsigned expired : 1;
60};
61
62static void fill_pack_info(struct pack_info *info,
63 struct packed_git *p, const char *pack_name,
64 uint32_t orig_pack_int_id)
65{
66 memset(info, 0, sizeof(struct pack_info));
67
68 info->orig_pack_int_id = orig_pack_int_id;
69 info->pack_name = xstrdup(pack_name);
70 info->p = p;
71 info->bitmap_pos = BITMAP_POS_UNKNOWN;
72}
73
74static int pack_info_compare(const void *_a, const void *_b)
75{
76 struct pack_info *a = (struct pack_info *)_a;
77 struct pack_info *b = (struct pack_info *)_b;
78 return strcmp(a->pack_name, b->pack_name);
79}
80
81static int idx_or_pack_name_cmp(const void *_va, const void *_vb)
82{
83 const char *pack_name = _va;
84 const struct pack_info *compar = _vb;
85
86 return cmp_idx_or_pack_name(pack_name, compar->pack_name);
87}
88
89struct write_midx_context {
90 struct pack_info *info;
91 size_t nr;
92 size_t alloc;
93 struct multi_pack_index *m;
94 struct multi_pack_index *base_midx;
95 struct progress *progress;
96 unsigned pack_paths_checked;
97
98 struct pack_midx_entry *entries;
99 size_t entries_nr;
100
101 uint32_t *pack_perm;
102 uint32_t *pack_order;
103 unsigned large_offsets_needed:1;
104 uint32_t num_large_offsets;
105
106 uint32_t preferred_pack_idx;
107
108 int incremental;
109 uint32_t num_multi_pack_indexes_before;
110
111 struct string_list *to_include;
112
113 struct repository *repo;
114 struct odb_source *source;
115};
116
117static int should_include_pack(const struct write_midx_context *ctx,
118 const char *file_name)
119{
120 /*
121 * Note that at most one of ctx->m and ctx->to_include are set,
122 * so we are testing midx_contains_pack() and
123 * string_list_has_string() independently (guarded by the
124 * appropriate NULL checks).
125 *
126 * We could support passing to_include while reusing an existing
127 * MIDX, but don't currently since the reuse process drags
128 * forward all packs from an existing MIDX (without checking
129 * whether or not they appear in the to_include list).
130 *
131 * If we added support for that, these next two conditional
132 * should be performed independently (likely checking
133 * to_include before the existing MIDX).
134 */
135 if (ctx->m && midx_contains_pack(ctx->m, file_name))
136 return 0;
137 else if (ctx->base_midx && midx_contains_pack(ctx->base_midx,
138 file_name))
139 return 0;
140 else if (ctx->to_include &&
141 !string_list_has_string(ctx->to_include, file_name))
142 return 0;
143 return 1;
144}
145
146static void add_pack_to_midx(const char *full_path, size_t full_path_len,
147 const char *file_name, void *data)
148{
149 struct write_midx_context *ctx = data;
150 struct packed_git *p;
151
152 if (ends_with(file_name, ".idx")) {
153 display_progress(ctx->progress, ++ctx->pack_paths_checked);
154
155 if (!should_include_pack(ctx, file_name))
156 return;
157
158 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
159 p = add_packed_git(ctx->repo, full_path, full_path_len, 0);
160 if (!p) {
161 warning(_("failed to add packfile '%s'"),
162 full_path);
163 return;
164 }
165
166 if (open_pack_index(p)) {
167 warning(_("failed to open pack-index '%s'"),
168 full_path);
169 close_pack(p);
170 free(p);
171 return;
172 }
173
174 fill_pack_info(&ctx->info[ctx->nr], p, file_name, ctx->nr);
175 ctx->nr++;
176 }
177}
178
179struct pack_midx_entry {
180 struct object_id oid;
181 uint32_t pack_int_id;
182 time_t pack_mtime;
183 uint64_t offset;
184 unsigned preferred : 1;
185};
186
187static int midx_oid_compare(const void *_a, const void *_b)
188{
189 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
190 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
191 int cmp = oidcmp(&a->oid, &b->oid);
192
193 if (cmp)
194 return cmp;
195
196 /* Sort objects in a preferred pack first when multiple copies exist. */
197 if (a->preferred > b->preferred)
198 return -1;
199 if (a->preferred < b->preferred)
200 return 1;
201
202 if (a->pack_mtime > b->pack_mtime)
203 return -1;
204 else if (a->pack_mtime < b->pack_mtime)
205 return 1;
206
207 return a->pack_int_id - b->pack_int_id;
208}
209
210static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
211 struct pack_midx_entry *e,
212 uint32_t pos)
213{
214 if (pos >= m->num_objects + m->num_objects_in_base)
215 return 1;
216
217 nth_midxed_object_oid(&e->oid, m, pos);
218 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
219 e->offset = nth_midxed_offset(m, pos);
220
221 /* consider objects in midx to be from "old" packs */
222 e->pack_mtime = 0;
223 return 0;
224}
225
226static void fill_pack_entry(uint32_t pack_int_id,
227 struct packed_git *p,
228 uint32_t cur_object,
229 struct pack_midx_entry *entry,
230 int preferred)
231{
232 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
233 die(_("failed to locate object %d in packfile"), cur_object);
234
235 entry->pack_int_id = pack_int_id;
236 entry->pack_mtime = p->mtime;
237
238 entry->offset = nth_packed_object_offset(p, cur_object);
239 entry->preferred = !!preferred;
240}
241
242struct midx_fanout {
243 struct pack_midx_entry *entries;
244 size_t nr, alloc;
245};
246
247static void midx_fanout_grow(struct midx_fanout *fanout, size_t nr)
248{
249 if (nr < fanout->nr)
250 BUG("negative growth in midx_fanout_grow() (%"PRIuMAX" < %"PRIuMAX")",
251 (uintmax_t)nr, (uintmax_t)fanout->nr);
252 ALLOC_GROW(fanout->entries, nr, fanout->alloc);
253}
254
255static void midx_fanout_sort(struct midx_fanout *fanout)
256{
257 QSORT(fanout->entries, fanout->nr, midx_oid_compare);
258}
259
260static void midx_fanout_add_midx_fanout(struct midx_fanout *fanout,
261 struct multi_pack_index *m,
262 uint32_t cur_fanout,
263 uint32_t preferred_pack)
264{
265 uint32_t start = m->num_objects_in_base, end;
266 uint32_t cur_object;
267
268 if (m->base_midx)
269 midx_fanout_add_midx_fanout(fanout, m->base_midx, cur_fanout,
270 preferred_pack);
271
272 if (cur_fanout)
273 start += ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
274 end = m->num_objects_in_base + ntohl(m->chunk_oid_fanout[cur_fanout]);
275
276 for (cur_object = start; cur_object < end; cur_object++) {
277 if ((preferred_pack != NO_PREFERRED_PACK) &&
278 (preferred_pack == nth_midxed_pack_int_id(m, cur_object))) {
279 /*
280 * Objects from preferred packs are added
281 * separately.
282 */
283 continue;
284 }
285
286 midx_fanout_grow(fanout, fanout->nr + 1);
287 nth_midxed_pack_midx_entry(m,
288 &fanout->entries[fanout->nr],
289 cur_object);
290 fanout->entries[fanout->nr].preferred = 0;
291 fanout->nr++;
292 }
293}
294
295static void midx_fanout_add_pack_fanout(struct midx_fanout *fanout,
296 struct pack_info *info,
297 uint32_t cur_pack,
298 int preferred,
299 uint32_t cur_fanout)
300{
301 struct packed_git *pack = info[cur_pack].p;
302 uint32_t start = 0, end;
303 uint32_t cur_object;
304
305 if (cur_fanout)
306 start = get_pack_fanout(pack, cur_fanout - 1);
307 end = get_pack_fanout(pack, cur_fanout);
308
309 for (cur_object = start; cur_object < end; cur_object++) {
310 midx_fanout_grow(fanout, fanout->nr + 1);
311 fill_pack_entry(cur_pack,
312 info[cur_pack].p,
313 cur_object,
314 &fanout->entries[fanout->nr],
315 preferred);
316 fanout->nr++;
317 }
318}
319
320/*
321 * It is possible to artificially get into a state where there are many
322 * duplicate copies of objects. That can create high memory pressure if
323 * we are to create a list of all objects before de-duplication. To reduce
324 * this memory pressure without a significant performance drop, automatically
325 * group objects by the first byte of their object id. Use the IDX fanout
326 * tables to group the data, copy to a local array, then sort.
327 *
328 * Copy only the de-duplicated entries (selected by most-recent modified time
329 * of a packfile containing the object).
330 */
331static void compute_sorted_entries(struct write_midx_context *ctx,
332 uint32_t start_pack)
333{
334 uint32_t cur_fanout, cur_pack, cur_object;
335 size_t alloc_objects, total_objects = 0;
336 struct midx_fanout fanout = { 0 };
337
338 for (cur_pack = start_pack; cur_pack < ctx->nr; cur_pack++)
339 total_objects = st_add(total_objects,
340 ctx->info[cur_pack].p->num_objects);
341
342 /*
343 * As we de-duplicate by fanout value, we expect the fanout
344 * slices to be evenly distributed, with some noise. Hence,
345 * allocate slightly more than one 256th.
346 */
347 alloc_objects = fanout.alloc = total_objects > 3200 ? total_objects / 200 : 16;
348
349 ALLOC_ARRAY(fanout.entries, fanout.alloc);
350 ALLOC_ARRAY(ctx->entries, alloc_objects);
351 ctx->entries_nr = 0;
352
353 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
354 fanout.nr = 0;
355
356 if (ctx->m && !ctx->incremental)
357 midx_fanout_add_midx_fanout(&fanout, ctx->m, cur_fanout,
358 ctx->preferred_pack_idx);
359
360 for (cur_pack = start_pack; cur_pack < ctx->nr; cur_pack++) {
361 int preferred = cur_pack == ctx->preferred_pack_idx;
362 midx_fanout_add_pack_fanout(&fanout,
363 ctx->info, cur_pack,
364 preferred, cur_fanout);
365 }
366
367 if (ctx->preferred_pack_idx != NO_PREFERRED_PACK &&
368 ctx->preferred_pack_idx < start_pack)
369 midx_fanout_add_pack_fanout(&fanout, ctx->info,
370 ctx->preferred_pack_idx, 1,
371 cur_fanout);
372
373 midx_fanout_sort(&fanout);
374
375 /*
376 * The batch is now sorted by OID and then mtime (descending).
377 * Take only the first duplicate.
378 */
379 for (cur_object = 0; cur_object < fanout.nr; cur_object++) {
380 if (cur_object && oideq(&fanout.entries[cur_object - 1].oid,
381 &fanout.entries[cur_object].oid))
382 continue;
383 if (ctx->incremental && ctx->base_midx &&
384 midx_has_oid(ctx->base_midx,
385 &fanout.entries[cur_object].oid))
386 continue;
387
388 ALLOC_GROW(ctx->entries, st_add(ctx->entries_nr, 1),
389 alloc_objects);
390 memcpy(&ctx->entries[ctx->entries_nr],
391 &fanout.entries[cur_object],
392 sizeof(struct pack_midx_entry));
393 ctx->entries_nr++;
394 }
395 }
396
397 free(fanout.entries);
398}
399
400static int write_midx_pack_names(struct hashfile *f, void *data)
401{
402 struct write_midx_context *ctx = data;
403 uint32_t i;
404 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
405 size_t written = 0;
406
407 for (i = 0; i < ctx->nr; i++) {
408 size_t writelen;
409
410 if (ctx->info[i].expired)
411 continue;
412
413 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
414 BUG("incorrect pack-file order: %s before %s",
415 ctx->info[i - 1].pack_name,
416 ctx->info[i].pack_name);
417
418 writelen = strlen(ctx->info[i].pack_name) + 1;
419 hashwrite(f, ctx->info[i].pack_name, writelen);
420 written += writelen;
421 }
422
423 /* add padding to be aligned */
424 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
425 if (i < MIDX_CHUNK_ALIGNMENT) {
426 memset(padding, 0, sizeof(padding));
427 hashwrite(f, padding, i);
428 }
429
430 return 0;
431}
432
433static int write_midx_bitmapped_packs(struct hashfile *f, void *data)
434{
435 struct write_midx_context *ctx = data;
436 size_t i;
437
438 for (i = 0; i < ctx->nr; i++) {
439 struct pack_info *pack = &ctx->info[i];
440 if (pack->expired)
441 continue;
442
443 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN && pack->bitmap_nr)
444 BUG("pack '%s' has no bitmap position, but has %d bitmapped object(s)",
445 pack->pack_name, pack->bitmap_nr);
446
447 hashwrite_be32(f, pack->bitmap_pos);
448 hashwrite_be32(f, pack->bitmap_nr);
449 }
450 return 0;
451}
452
453static int write_midx_oid_fanout(struct hashfile *f,
454 void *data)
455{
456 struct write_midx_context *ctx = data;
457 struct pack_midx_entry *list = ctx->entries;
458 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
459 uint32_t count = 0;
460 uint32_t i;
461
462 /*
463 * Write the first-level table (the list is sorted,
464 * but we use a 256-entry lookup to be able to avoid
465 * having to do eight extra binary search iterations).
466 */
467 for (i = 0; i < 256; i++) {
468 struct pack_midx_entry *next = list;
469
470 while (next < last && next->oid.hash[0] == i) {
471 count++;
472 next++;
473 }
474
475 hashwrite_be32(f, count);
476 list = next;
477 }
478
479 return 0;
480}
481
482static int write_midx_oid_lookup(struct hashfile *f,
483 void *data)
484{
485 struct write_midx_context *ctx = data;
486 unsigned char hash_len = ctx->repo->hash_algo->rawsz;
487 struct pack_midx_entry *list = ctx->entries;
488 uint32_t i;
489
490 for (i = 0; i < ctx->entries_nr; i++) {
491 struct pack_midx_entry *obj = list++;
492
493 if (i < ctx->entries_nr - 1) {
494 struct pack_midx_entry *next = list;
495 if (oidcmp(&obj->oid, &next->oid) >= 0)
496 BUG("OIDs not in order: %s >= %s",
497 oid_to_hex(&obj->oid),
498 oid_to_hex(&next->oid));
499 }
500
501 hashwrite(f, obj->oid.hash, (int)hash_len);
502 }
503
504 return 0;
505}
506
507static int write_midx_object_offsets(struct hashfile *f,
508 void *data)
509{
510 struct write_midx_context *ctx = data;
511 struct pack_midx_entry *list = ctx->entries;
512 uint32_t i, nr_large_offset = 0;
513
514 for (i = 0; i < ctx->entries_nr; i++) {
515 struct pack_midx_entry *obj = list++;
516
517 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
518 BUG("object %s is in an expired pack with int-id %d",
519 oid_to_hex(&obj->oid),
520 obj->pack_int_id);
521
522 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
523
524 if (ctx->large_offsets_needed && obj->offset >> 31)
525 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
526 else if (!ctx->large_offsets_needed && obj->offset >> 32)
527 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
528 oid_to_hex(&obj->oid),
529 obj->offset);
530 else
531 hashwrite_be32(f, (uint32_t)obj->offset);
532 }
533
534 return 0;
535}
536
537static int write_midx_large_offsets(struct hashfile *f,
538 void *data)
539{
540 struct write_midx_context *ctx = data;
541 struct pack_midx_entry *list = ctx->entries;
542 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
543 uint32_t nr_large_offset = ctx->num_large_offsets;
544
545 while (nr_large_offset) {
546 struct pack_midx_entry *obj;
547 uint64_t offset;
548
549 if (list >= end)
550 BUG("too many large-offset objects");
551
552 obj = list++;
553 offset = obj->offset;
554
555 if (!(offset >> 31))
556 continue;
557
558 hashwrite_be64(f, offset);
559
560 nr_large_offset--;
561 }
562
563 return 0;
564}
565
566static int write_midx_revindex(struct hashfile *f,
567 void *data)
568{
569 struct write_midx_context *ctx = data;
570 uint32_t i, nr_base;
571
572 if (ctx->incremental && ctx->base_midx)
573 nr_base = ctx->base_midx->num_objects +
574 ctx->base_midx->num_objects_in_base;
575 else
576 nr_base = 0;
577
578 for (i = 0; i < ctx->entries_nr; i++)
579 hashwrite_be32(f, ctx->pack_order[i] + nr_base);
580
581 return 0;
582}
583
584struct midx_pack_order_data {
585 uint32_t nr;
586 uint32_t pack;
587 off_t offset;
588};
589
590static int midx_pack_order_cmp(const void *va, const void *vb)
591{
592 const struct midx_pack_order_data *a = va, *b = vb;
593 if (a->pack < b->pack)
594 return -1;
595 else if (a->pack > b->pack)
596 return 1;
597 else if (a->offset < b->offset)
598 return -1;
599 else if (a->offset > b->offset)
600 return 1;
601 else
602 return 0;
603}
604
605static uint32_t *midx_pack_order(struct write_midx_context *ctx)
606{
607 struct midx_pack_order_data *data;
608 uint32_t *pack_order, base_objects = 0;
609 uint32_t i;
610
611 trace2_region_enter("midx", "midx_pack_order", ctx->repo);
612
613 if (ctx->incremental && ctx->base_midx)
614 base_objects = ctx->base_midx->num_objects +
615 ctx->base_midx->num_objects_in_base;
616
617 ALLOC_ARRAY(pack_order, ctx->entries_nr);
618 ALLOC_ARRAY(data, ctx->entries_nr);
619
620 for (i = 0; i < ctx->entries_nr; i++) {
621 struct pack_midx_entry *e = &ctx->entries[i];
622 data[i].nr = i;
623 data[i].pack = ctx->pack_perm[e->pack_int_id];
624 if (!e->preferred)
625 data[i].pack |= (1U << 31);
626 data[i].offset = e->offset;
627 }
628
629 QSORT(data, ctx->entries_nr, midx_pack_order_cmp);
630
631 for (i = 0; i < ctx->entries_nr; i++) {
632 struct pack_midx_entry *e = &ctx->entries[data[i].nr];
633 struct pack_info *pack = &ctx->info[ctx->pack_perm[e->pack_int_id]];
634 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN)
635 pack->bitmap_pos = i + base_objects;
636 pack->bitmap_nr++;
637 pack_order[i] = data[i].nr;
638 }
639 for (i = 0; i < ctx->nr; i++) {
640 struct pack_info *pack = &ctx->info[ctx->pack_perm[i]];
641 if (pack->bitmap_pos == BITMAP_POS_UNKNOWN)
642 pack->bitmap_pos = 0;
643 }
644 free(data);
645
646 trace2_region_leave("midx", "midx_pack_order", ctx->repo);
647
648 return pack_order;
649}
650
651static void write_midx_reverse_index(struct write_midx_context *ctx,
652 unsigned char *midx_hash)
653{
654 struct strbuf buf = STRBUF_INIT;
655 char *tmp_file;
656
657 trace2_region_enter("midx", "write_midx_reverse_index", ctx->repo);
658
659 if (ctx->incremental)
660 get_split_midx_filename_ext(ctx->source, &buf,
661 midx_hash, MIDX_EXT_REV);
662 else
663 get_midx_filename_ext(ctx->source, &buf,
664 midx_hash, MIDX_EXT_REV);
665
666 tmp_file = write_rev_file_order(ctx->repo, NULL, ctx->pack_order,
667 ctx->entries_nr, midx_hash, WRITE_REV);
668
669 if (finalize_object_file(ctx->repo, tmp_file, buf.buf))
670 die(_("cannot store reverse index file"));
671
672 strbuf_release(&buf);
673 free(tmp_file);
674
675 trace2_region_leave("midx", "write_midx_reverse_index", ctx->repo);
676}
677
678static void prepare_midx_packing_data(struct packing_data *pdata,
679 struct write_midx_context *ctx)
680{
681 uint32_t i;
682
683 trace2_region_enter("midx", "prepare_midx_packing_data", ctx->repo);
684
685 memset(pdata, 0, sizeof(struct packing_data));
686 prepare_packing_data(ctx->repo, pdata);
687
688 for (i = 0; i < ctx->entries_nr; i++) {
689 uint32_t pos = ctx->pack_order[i];
690 struct pack_midx_entry *from = &ctx->entries[pos];
691 struct object_entry *to = packlist_alloc(pdata, &from->oid);
692
693 oe_set_in_pack(pdata, to,
694 ctx->info[ctx->pack_perm[from->pack_int_id]].p);
695 }
696
697 trace2_region_leave("midx", "prepare_midx_packing_data", ctx->repo);
698}
699
700static int add_ref_to_pending(const char *refname, const char *referent UNUSED,
701 const struct object_id *oid,
702 int flag, void *cb_data)
703{
704 struct rev_info *revs = (struct rev_info*)cb_data;
705 struct object_id peeled;
706 struct object *object;
707
708 if ((flag & REF_ISSYMREF) && (flag & REF_ISBROKEN)) {
709 warning("symbolic ref is dangling: %s", refname);
710 return 0;
711 }
712
713 if (!peel_iterated_oid(revs->repo, oid, &peeled))
714 oid = &peeled;
715
716 object = parse_object_or_die(revs->repo, oid, refname);
717 if (object->type != OBJ_COMMIT)
718 return 0;
719
720 add_pending_object(revs, object, "");
721 if (bitmap_is_preferred_refname(revs->repo, refname))
722 object->flags |= NEEDS_BITMAP;
723 return 0;
724}
725
726struct bitmap_commit_cb {
727 struct commit **commits;
728 size_t commits_nr, commits_alloc;
729
730 struct write_midx_context *ctx;
731};
732
733static const struct object_id *bitmap_oid_access(size_t index,
734 const void *_entries)
735{
736 const struct pack_midx_entry *entries = _entries;
737 return &entries[index].oid;
738}
739
740static void bitmap_show_commit(struct commit *commit, void *_data)
741{
742 struct bitmap_commit_cb *data = _data;
743 int pos = oid_pos(&commit->object.oid, data->ctx->entries,
744 data->ctx->entries_nr,
745 bitmap_oid_access);
746 if (pos < 0)
747 return;
748
749 ALLOC_GROW(data->commits, data->commits_nr + 1, data->commits_alloc);
750 data->commits[data->commits_nr++] = commit;
751}
752
753static int read_refs_snapshot(const char *refs_snapshot,
754 struct rev_info *revs)
755{
756 struct strbuf buf = STRBUF_INIT;
757 struct object_id oid;
758 FILE *f = xfopen(refs_snapshot, "r");
759
760 while (strbuf_getline(&buf, f) != EOF) {
761 struct object *object;
762 int preferred = 0;
763 char *hex = buf.buf;
764 const char *end = NULL;
765
766 if (buf.len && *buf.buf == '+') {
767 preferred = 1;
768 hex = &buf.buf[1];
769 }
770
771 if (parse_oid_hex_algop(hex, &oid, &end, revs->repo->hash_algo) < 0)
772 die(_("could not parse line: %s"), buf.buf);
773 if (*end)
774 die(_("malformed line: %s"), buf.buf);
775
776 object = parse_object_or_die(revs->repo, &oid, NULL);
777 if (preferred)
778 object->flags |= NEEDS_BITMAP;
779
780 add_pending_object(revs, object, "");
781 }
782
783 fclose(f);
784 strbuf_release(&buf);
785 return 0;
786}
787
788static struct commit **find_commits_for_midx_bitmap(uint32_t *indexed_commits_nr_p,
789 const char *refs_snapshot,
790 struct write_midx_context *ctx)
791{
792 struct rev_info revs;
793 struct bitmap_commit_cb cb = {0};
794
795 trace2_region_enter("midx", "find_commits_for_midx_bitmap", ctx->repo);
796
797 cb.ctx = ctx;
798
799 repo_init_revisions(ctx->repo, &revs, NULL);
800 if (refs_snapshot) {
801 read_refs_snapshot(refs_snapshot, &revs);
802 } else {
803 setup_revisions(0, NULL, &revs, NULL);
804 refs_for_each_ref(get_main_ref_store(ctx->repo),
805 add_ref_to_pending, &revs);
806 }
807
808 /*
809 * Skipping promisor objects here is intentional, since it only excludes
810 * them from the list of reachable commits that we want to select from
811 * when computing the selection of MIDX'd commits to receive bitmaps.
812 *
813 * Reachability bitmaps do require that their objects be closed under
814 * reachability, but fetching any objects missing from promisors at this
815 * point is too late. But, if one of those objects can be reached from
816 * an another object that is included in the bitmap, then we will
817 * complain later that we don't have reachability closure (and fail
818 * appropriately).
819 */
820 fetch_if_missing = 0;
821 revs.exclude_promisor_objects = 1;
822
823 if (prepare_revision_walk(&revs))
824 die(_("revision walk setup failed"));
825
826 traverse_commit_list(&revs, bitmap_show_commit, NULL, &cb);
827 if (indexed_commits_nr_p)
828 *indexed_commits_nr_p = cb.commits_nr;
829
830 release_revisions(&revs);
831
832 trace2_region_leave("midx", "find_commits_for_midx_bitmap", ctx->repo);
833
834 return cb.commits;
835}
836
837static int write_midx_bitmap(struct write_midx_context *ctx,
838 const unsigned char *midx_hash,
839 struct packing_data *pdata,
840 struct commit **commits,
841 uint32_t commits_nr,
842 unsigned flags)
843{
844 int ret;
845 uint16_t options = 0;
846 struct bitmap_writer writer;
847 struct pack_idx_entry **index;
848 struct strbuf bitmap_name = STRBUF_INIT;
849
850 trace2_region_enter("midx", "write_midx_bitmap", ctx->repo);
851
852 if (ctx->incremental)
853 get_split_midx_filename_ext(ctx->source, &bitmap_name,
854 midx_hash, MIDX_EXT_BITMAP);
855 else
856 get_midx_filename_ext(ctx->source, &bitmap_name,
857 midx_hash, MIDX_EXT_BITMAP);
858
859 if (flags & MIDX_WRITE_BITMAP_HASH_CACHE)
860 options |= BITMAP_OPT_HASH_CACHE;
861
862 if (flags & MIDX_WRITE_BITMAP_LOOKUP_TABLE)
863 options |= BITMAP_OPT_LOOKUP_TABLE;
864
865 /*
866 * Build the MIDX-order index based on pdata.objects (which is already
867 * in MIDX order; c.f., 'midx_pack_order_cmp()' for the definition of
868 * this order).
869 */
870 ALLOC_ARRAY(index, pdata->nr_objects);
871 for (uint32_t i = 0; i < pdata->nr_objects; i++)
872 index[i] = &pdata->objects[i].idx;
873
874 bitmap_writer_init(&writer, ctx->repo, pdata,
875 ctx->incremental ? ctx->base_midx : NULL);
876 bitmap_writer_show_progress(&writer, flags & MIDX_PROGRESS);
877 bitmap_writer_build_type_index(&writer, index);
878
879 /*
880 * bitmap_writer_finish expects objects in lex order, but pack_order
881 * gives us exactly that. use it directly instead of re-sorting the
882 * array.
883 *
884 * This changes the order of objects in 'index' between
885 * bitmap_writer_build_type_index and bitmap_writer_finish.
886 *
887 * The same re-ordering takes place in the single-pack bitmap code via
888 * write_idx_file(), which is called by finish_tmp_packfile(), which
889 * happens between bitmap_writer_build_type_index() and
890 * bitmap_writer_finish().
891 */
892 for (uint32_t i = 0; i < pdata->nr_objects; i++)
893 index[ctx->pack_order[i]] = &pdata->objects[i].idx;
894
895 bitmap_writer_select_commits(&writer, commits, commits_nr);
896 ret = bitmap_writer_build(&writer);
897 if (ret < 0)
898 goto cleanup;
899
900 bitmap_writer_set_checksum(&writer, midx_hash);
901 bitmap_writer_finish(&writer, index, bitmap_name.buf, options);
902
903cleanup:
904 free(index);
905 strbuf_release(&bitmap_name);
906 bitmap_writer_free(&writer);
907
908 trace2_region_leave("midx", "write_midx_bitmap", ctx->repo);
909
910 return ret;
911}
912
913static int fill_packs_from_midx(struct write_midx_context *ctx)
914{
915 struct multi_pack_index *m;
916
917 for (m = ctx->m; m; m = m->base_midx) {
918 uint32_t i;
919
920 for (i = 0; i < m->num_packs; i++) {
921 if (prepare_midx_pack(m, m->num_packs_in_base + i))
922 return error(_("could not load pack"));
923
924 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
925 fill_pack_info(&ctx->info[ctx->nr++], m->packs[i],
926 m->pack_names[i],
927 m->num_packs_in_base + i);
928 }
929 }
930 return 0;
931}
932
933static struct {
934 const char *non_split;
935 const char *split;
936} midx_exts[] = {
937 {NULL, MIDX_EXT_MIDX},
938 {MIDX_EXT_BITMAP, MIDX_EXT_BITMAP},
939 {MIDX_EXT_REV, MIDX_EXT_REV},
940};
941
942static int link_midx_to_chain(struct multi_pack_index *m)
943{
944 struct strbuf from = STRBUF_INIT;
945 struct strbuf to = STRBUF_INIT;
946 int ret = 0;
947 size_t i;
948
949 if (!m || m->has_chain) {
950 /*
951 * Either no MIDX previously existed, or it was already
952 * part of a MIDX chain. In both cases, we have nothing
953 * to link, so return early.
954 */
955 goto done;
956 }
957
958 for (i = 0; i < ARRAY_SIZE(midx_exts); i++) {
959 const unsigned char *hash = get_midx_checksum(m);
960
961 get_midx_filename_ext(m->source, &from,
962 hash, midx_exts[i].non_split);
963 get_split_midx_filename_ext(m->source, &to, hash,
964 midx_exts[i].split);
965
966 if (link(from.buf, to.buf) < 0 && errno != ENOENT) {
967 ret = error_errno(_("unable to link '%s' to '%s'"),
968 from.buf, to.buf);
969 goto done;
970 }
971
972 strbuf_reset(&from);
973 strbuf_reset(&to);
974 }
975
976done:
977 strbuf_release(&from);
978 strbuf_release(&to);
979 return ret;
980}
981
982static void clear_midx_files(struct odb_source *source,
983 const char **hashes, uint32_t hashes_nr,
984 unsigned incremental)
985{
986 /*
987 * if incremental:
988 * - remove all non-incremental MIDX files
989 * - remove any incremental MIDX files not in the current one
990 *
991 * if non-incremental:
992 * - remove all incremental MIDX files
993 * - remove any non-incremental MIDX files not matching the current
994 * hash
995 */
996 struct strbuf buf = STRBUF_INIT;
997 const char *exts[] = { MIDX_EXT_BITMAP, MIDX_EXT_REV, MIDX_EXT_MIDX };
998 uint32_t i, j;
999
1000 for (i = 0; i < ARRAY_SIZE(exts); i++) {
1001 clear_incremental_midx_files_ext(source, exts[i],
1002 hashes, hashes_nr);
1003 for (j = 0; j < hashes_nr; j++)
1004 clear_midx_files_ext(source, exts[i], hashes[j]);
1005 }
1006
1007 if (incremental)
1008 get_midx_filename(source, &buf);
1009 else
1010 get_midx_chain_filename(source, &buf);
1011
1012 if (unlink(buf.buf) && errno != ENOENT)
1013 die_errno(_("failed to clear multi-pack-index at %s"), buf.buf);
1014
1015 strbuf_release(&buf);
1016}
1017
1018static int write_midx_internal(struct odb_source *source,
1019 struct string_list *packs_to_include,
1020 struct string_list *packs_to_drop,
1021 const char *preferred_pack_name,
1022 const char *refs_snapshot,
1023 unsigned flags)
1024{
1025 struct repository *r = source->odb->repo;
1026 struct strbuf midx_name = STRBUF_INIT;
1027 unsigned char midx_hash[GIT_MAX_RAWSZ];
1028 uint32_t start_pack;
1029 struct hashfile *f = NULL;
1030 struct lock_file lk;
1031 struct tempfile *incr;
1032 struct write_midx_context ctx = {
1033 .preferred_pack_idx = NO_PREFERRED_PACK,
1034 };
1035 int bitmapped_packs_concat_len = 0;
1036 int pack_name_concat_len = 0;
1037 int dropped_packs = 0;
1038 int result = -1;
1039 const char **keep_hashes = NULL;
1040 struct chunkfile *cf;
1041
1042 trace2_region_enter("midx", "write_midx_internal", r);
1043
1044 ctx.repo = r;
1045 ctx.source = source;
1046
1047 ctx.incremental = !!(flags & MIDX_WRITE_INCREMENTAL);
1048
1049 if (ctx.incremental)
1050 strbuf_addf(&midx_name,
1051 "%s/pack/multi-pack-index.d/tmp_midx_XXXXXX",
1052 source->path);
1053 else
1054 get_midx_filename(source, &midx_name);
1055 if (safe_create_leading_directories(r, midx_name.buf))
1056 die_errno(_("unable to create leading directories of %s"),
1057 midx_name.buf);
1058
1059 if (!packs_to_include || ctx.incremental) {
1060 struct multi_pack_index *m = get_multi_pack_index(source);
1061 if (m && !midx_checksum_valid(m)) {
1062 warning(_("ignoring existing multi-pack-index; checksum mismatch"));
1063 m = NULL;
1064 }
1065
1066 if (m) {
1067 /*
1068 * Only reference an existing MIDX when not filtering
1069 * which packs to include, since all packs and objects
1070 * are copied blindly from an existing MIDX if one is
1071 * present.
1072 */
1073 if (ctx.incremental)
1074 ctx.base_midx = m;
1075 else if (!packs_to_include)
1076 ctx.m = m;
1077 }
1078 }
1079
1080 ctx.nr = 0;
1081 ctx.alloc = ctx.m ? ctx.m->num_packs + ctx.m->num_packs_in_base : 16;
1082 ctx.info = NULL;
1083 ALLOC_ARRAY(ctx.info, ctx.alloc);
1084
1085 if (ctx.incremental) {
1086 struct multi_pack_index *m = ctx.base_midx;
1087 while (m) {
1088 if (flags & MIDX_WRITE_BITMAP && load_midx_revindex(m)) {
1089 error(_("could not load reverse index for MIDX %s"),
1090 hash_to_hex_algop(get_midx_checksum(m),
1091 m->source->odb->repo->hash_algo));
1092 goto cleanup;
1093 }
1094 ctx.num_multi_pack_indexes_before++;
1095 m = m->base_midx;
1096 }
1097 } else if (ctx.m && fill_packs_from_midx(&ctx)) {
1098 goto cleanup;
1099 }
1100
1101 start_pack = ctx.nr;
1102
1103 ctx.pack_paths_checked = 0;
1104 if (flags & MIDX_PROGRESS)
1105 ctx.progress = start_delayed_progress(r,
1106 _("Adding packfiles to multi-pack-index"), 0);
1107 else
1108 ctx.progress = NULL;
1109
1110 ctx.to_include = packs_to_include;
1111
1112 for_each_file_in_pack_dir(source->path, add_pack_to_midx, &ctx);
1113 stop_progress(&ctx.progress);
1114
1115 if ((ctx.m && ctx.nr == ctx.m->num_packs + ctx.m->num_packs_in_base) &&
1116 !ctx.incremental &&
1117 !(packs_to_include || packs_to_drop)) {
1118 struct bitmap_index *bitmap_git;
1119 int bitmap_exists;
1120 int want_bitmap = flags & MIDX_WRITE_BITMAP;
1121
1122 bitmap_git = prepare_midx_bitmap_git(ctx.m);
1123 bitmap_exists = bitmap_git && bitmap_is_midx(bitmap_git);
1124 free_bitmap_index(bitmap_git);
1125
1126 if (bitmap_exists || !want_bitmap) {
1127 /*
1128 * The correct MIDX already exists, and so does a
1129 * corresponding bitmap (or one wasn't requested).
1130 */
1131 if (!want_bitmap)
1132 clear_midx_files_ext(source, "bitmap", NULL);
1133 result = 0;
1134 goto cleanup;
1135 }
1136 }
1137
1138 if (ctx.incremental && !ctx.nr) {
1139 result = 0;
1140 goto cleanup; /* nothing to do */
1141 }
1142
1143 if (preferred_pack_name) {
1144 ctx.preferred_pack_idx = NO_PREFERRED_PACK;
1145
1146 for (size_t i = 0; i < ctx.nr; i++) {
1147 if (!cmp_idx_or_pack_name(preferred_pack_name,
1148 ctx.info[i].pack_name)) {
1149 ctx.preferred_pack_idx = i;
1150 break;
1151 }
1152 }
1153
1154 if (ctx.preferred_pack_idx == NO_PREFERRED_PACK)
1155 warning(_("unknown preferred pack: '%s'"),
1156 preferred_pack_name);
1157 } else if (ctx.nr &&
1158 (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP))) {
1159 struct packed_git *oldest = ctx.info[0].p;
1160 ctx.preferred_pack_idx = 0;
1161
1162 /*
1163 * Attempt opening the pack index to populate num_objects.
1164 * Ignore failiures as they can be expected and are not
1165 * fatal during this selection time.
1166 */
1167 open_pack_index(oldest);
1168
1169 if (packs_to_drop && packs_to_drop->nr)
1170 BUG("cannot write a MIDX bitmap during expiration");
1171
1172 /*
1173 * set a preferred pack when writing a bitmap to ensure that
1174 * the pack from which the first object is selected in pseudo
1175 * pack-order has all of its objects selected from that pack
1176 * (and not another pack containing a duplicate)
1177 */
1178 for (size_t i = 1; i < ctx.nr; i++) {
1179 struct packed_git *p = ctx.info[i].p;
1180
1181 if (!oldest->num_objects || p->mtime < oldest->mtime) {
1182 oldest = p;
1183 open_pack_index(oldest);
1184 ctx.preferred_pack_idx = i;
1185 }
1186 }
1187
1188 if (!oldest->num_objects) {
1189 /*
1190 * If all packs are empty; unset the preferred index.
1191 * This is acceptable since there will be no duplicate
1192 * objects to resolve, so the preferred value doesn't
1193 * matter.
1194 */
1195 ctx.preferred_pack_idx = NO_PREFERRED_PACK;
1196 }
1197 } else {
1198 /*
1199 * otherwise don't mark any pack as preferred to avoid
1200 * interfering with expiration logic below
1201 */
1202 ctx.preferred_pack_idx = NO_PREFERRED_PACK;
1203 }
1204
1205 if (ctx.preferred_pack_idx != NO_PREFERRED_PACK) {
1206 struct packed_git *preferred = ctx.info[ctx.preferred_pack_idx].p;
1207
1208 if (open_pack_index(preferred))
1209 die(_("failed to open preferred pack %s"),
1210 ctx.info[ctx.preferred_pack_idx].pack_name);
1211
1212 if (!preferred->num_objects) {
1213 error(_("cannot select preferred pack %s with no objects"),
1214 preferred->pack_name);
1215 goto cleanup;
1216 }
1217 }
1218
1219 compute_sorted_entries(&ctx, start_pack);
1220
1221 ctx.large_offsets_needed = 0;
1222 for (size_t i = 0; i < ctx.entries_nr; i++) {
1223 if (ctx.entries[i].offset > 0x7fffffff)
1224 ctx.num_large_offsets++;
1225 if (ctx.entries[i].offset > 0xffffffff)
1226 ctx.large_offsets_needed = 1;
1227 }
1228
1229 QSORT(ctx.info, ctx.nr, pack_info_compare);
1230
1231 if (packs_to_drop && packs_to_drop->nr) {
1232 size_t drop_index = 0;
1233 int missing_drops = 0;
1234
1235 for (size_t i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
1236 int cmp = strcmp(ctx.info[i].pack_name,
1237 packs_to_drop->items[drop_index].string);
1238
1239 if (!cmp) {
1240 drop_index++;
1241 ctx.info[i].expired = 1;
1242 } else if (cmp > 0) {
1243 error(_("did not see pack-file %s to drop"),
1244 packs_to_drop->items[drop_index].string);
1245 drop_index++;
1246 missing_drops++;
1247 i--;
1248 } else {
1249 ctx.info[i].expired = 0;
1250 }
1251 }
1252
1253 if (missing_drops)
1254 goto cleanup;
1255 }
1256
1257 /*
1258 * pack_perm stores a permutation between pack-int-ids from the
1259 * previous multi-pack-index to the new one we are writing:
1260 *
1261 * pack_perm[old_id] = new_id
1262 */
1263 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
1264 for (size_t i = 0; i < ctx.nr; i++) {
1265 if (ctx.info[i].expired) {
1266 dropped_packs++;
1267 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
1268 } else {
1269 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
1270 }
1271 }
1272
1273 for (size_t i = 0; i < ctx.nr; i++) {
1274 if (ctx.info[i].expired)
1275 continue;
1276 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
1277 bitmapped_packs_concat_len += 2 * sizeof(uint32_t);
1278 }
1279
1280 /* Check that the preferred pack wasn't expired (if given). */
1281 if (preferred_pack_name) {
1282 struct pack_info *preferred = bsearch(preferred_pack_name,
1283 ctx.info, ctx.nr,
1284 sizeof(*ctx.info),
1285 idx_or_pack_name_cmp);
1286 if (preferred) {
1287 uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id];
1288 if (perm == PACK_EXPIRED)
1289 warning(_("preferred pack '%s' is expired"),
1290 preferred_pack_name);
1291 }
1292 }
1293
1294 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
1295 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
1296 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
1297
1298 if (ctx.nr - dropped_packs == 0) {
1299 error(_("no pack files to index."));
1300 goto cleanup;
1301 }
1302
1303 if (!ctx.entries_nr) {
1304 if (flags & MIDX_WRITE_BITMAP)
1305 warning(_("refusing to write multi-pack .bitmap without any objects"));
1306 flags &= ~(MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP);
1307 }
1308
1309 if (ctx.incremental) {
1310 struct strbuf lock_name = STRBUF_INIT;
1311
1312 get_midx_chain_filename(source, &lock_name);
1313 hold_lock_file_for_update(&lk, lock_name.buf, LOCK_DIE_ON_ERROR);
1314 strbuf_release(&lock_name);
1315
1316 incr = mks_tempfile_m(midx_name.buf, 0444);
1317 if (!incr) {
1318 error(_("unable to create temporary MIDX layer"));
1319 goto cleanup;
1320 }
1321
1322 if (adjust_shared_perm(r, get_tempfile_path(incr))) {
1323 error(_("unable to adjust shared permissions for '%s'"),
1324 get_tempfile_path(incr));
1325 goto cleanup;
1326 }
1327
1328 f = hashfd(r->hash_algo, get_tempfile_fd(incr),
1329 get_tempfile_path(incr));
1330 } else {
1331 hold_lock_file_for_update(&lk, midx_name.buf, LOCK_DIE_ON_ERROR);
1332 f = hashfd(r->hash_algo, get_lock_file_fd(&lk),
1333 get_lock_file_path(&lk));
1334 }
1335
1336 cf = init_chunkfile(f);
1337
1338 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
1339 write_midx_pack_names);
1340 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
1341 write_midx_oid_fanout);
1342 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
1343 st_mult(ctx.entries_nr, r->hash_algo->rawsz),
1344 write_midx_oid_lookup);
1345 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
1346 st_mult(ctx.entries_nr, MIDX_CHUNK_OFFSET_WIDTH),
1347 write_midx_object_offsets);
1348
1349 if (ctx.large_offsets_needed)
1350 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
1351 st_mult(ctx.num_large_offsets,
1352 MIDX_CHUNK_LARGE_OFFSET_WIDTH),
1353 write_midx_large_offsets);
1354
1355 if (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP)) {
1356 ctx.pack_order = midx_pack_order(&ctx);
1357 add_chunk(cf, MIDX_CHUNKID_REVINDEX,
1358 st_mult(ctx.entries_nr, sizeof(uint32_t)),
1359 write_midx_revindex);
1360 add_chunk(cf, MIDX_CHUNKID_BITMAPPEDPACKS,
1361 bitmapped_packs_concat_len,
1362 write_midx_bitmapped_packs);
1363 }
1364
1365 write_midx_header(r->hash_algo, f, get_num_chunks(cf),
1366 ctx.nr - dropped_packs);
1367 write_chunkfile(cf, &ctx);
1368
1369 finalize_hashfile(f, midx_hash, FSYNC_COMPONENT_PACK_METADATA,
1370 CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1371 free_chunkfile(cf);
1372
1373 if (flags & MIDX_WRITE_REV_INDEX &&
1374 git_env_bool("GIT_TEST_MIDX_WRITE_REV", 0))
1375 write_midx_reverse_index(&ctx, midx_hash);
1376
1377 if (flags & MIDX_WRITE_BITMAP) {
1378 struct packing_data pdata;
1379 struct commit **commits;
1380 uint32_t commits_nr;
1381
1382 if (!ctx.entries_nr)
1383 BUG("cannot write a bitmap without any objects");
1384
1385 prepare_midx_packing_data(&pdata, &ctx);
1386
1387 commits = find_commits_for_midx_bitmap(&commits_nr, refs_snapshot, &ctx);
1388
1389 /*
1390 * The previous steps translated the information from
1391 * 'entries' into information suitable for constructing
1392 * bitmaps. We no longer need that array, so clear it to
1393 * reduce memory pressure.
1394 */
1395 FREE_AND_NULL(ctx.entries);
1396 ctx.entries_nr = 0;
1397
1398 if (write_midx_bitmap(&ctx,
1399 midx_hash, &pdata, commits, commits_nr,
1400 flags) < 0) {
1401 error(_("could not write multi-pack bitmap"));
1402 clear_packing_data(&pdata);
1403 free(commits);
1404 goto cleanup;
1405 }
1406
1407 clear_packing_data(&pdata);
1408 free(commits);
1409 }
1410 /*
1411 * NOTE: Do not use ctx.entries beyond this point, since it might
1412 * have been freed in the previous if block.
1413 */
1414
1415 if (ctx.num_multi_pack_indexes_before == UINT32_MAX)
1416 die(_("too many multi-pack-indexes"));
1417
1418 CALLOC_ARRAY(keep_hashes, ctx.num_multi_pack_indexes_before + 1);
1419
1420 if (ctx.incremental) {
1421 FILE *chainf = fdopen_lock_file(&lk, "w");
1422 struct strbuf final_midx_name = STRBUF_INIT;
1423 struct multi_pack_index *m = ctx.base_midx;
1424
1425 if (!chainf) {
1426 error_errno(_("unable to open multi-pack-index chain file"));
1427 goto cleanup;
1428 }
1429
1430 if (link_midx_to_chain(ctx.base_midx) < 0)
1431 goto cleanup;
1432
1433 get_split_midx_filename_ext(source, &final_midx_name,
1434 midx_hash, MIDX_EXT_MIDX);
1435
1436 if (rename_tempfile(&incr, final_midx_name.buf) < 0) {
1437 error_errno(_("unable to rename new multi-pack-index layer"));
1438 goto cleanup;
1439 }
1440
1441 strbuf_release(&final_midx_name);
1442
1443 keep_hashes[ctx.num_multi_pack_indexes_before] =
1444 xstrdup(hash_to_hex_algop(midx_hash, r->hash_algo));
1445
1446 for (uint32_t i = 0; i < ctx.num_multi_pack_indexes_before; i++) {
1447 uint32_t j = ctx.num_multi_pack_indexes_before - i - 1;
1448
1449 keep_hashes[j] = xstrdup(hash_to_hex_algop(get_midx_checksum(m),
1450 r->hash_algo));
1451 m = m->base_midx;
1452 }
1453
1454 for (uint32_t i = 0; i <= ctx.num_multi_pack_indexes_before; i++)
1455 fprintf(get_lock_file_fp(&lk), "%s\n", keep_hashes[i]);
1456 } else {
1457 keep_hashes[ctx.num_multi_pack_indexes_before] =
1458 xstrdup(hash_to_hex_algop(midx_hash, r->hash_algo));
1459 }
1460
1461 if (ctx.m || ctx.base_midx)
1462 close_object_store(ctx.repo->objects);
1463
1464 if (commit_lock_file(&lk) < 0)
1465 die_errno(_("could not write multi-pack-index"));
1466
1467 clear_midx_files(source, keep_hashes,
1468 ctx.num_multi_pack_indexes_before + 1,
1469 ctx.incremental);
1470 result = 0;
1471
1472cleanup:
1473 for (size_t i = 0; i < ctx.nr; i++) {
1474 if (ctx.info[i].p) {
1475 close_pack(ctx.info[i].p);
1476 free(ctx.info[i].p);
1477 }
1478 free(ctx.info[i].pack_name);
1479 }
1480
1481 free(ctx.info);
1482 free(ctx.entries);
1483 free(ctx.pack_perm);
1484 free(ctx.pack_order);
1485 if (keep_hashes) {
1486 for (uint32_t i = 0; i <= ctx.num_multi_pack_indexes_before; i++)
1487 free((char *)keep_hashes[i]);
1488 free(keep_hashes);
1489 }
1490 strbuf_release(&midx_name);
1491
1492 trace2_region_leave("midx", "write_midx_internal", r);
1493
1494 return result;
1495}
1496
1497int write_midx_file(struct odb_source *source,
1498 const char *preferred_pack_name,
1499 const char *refs_snapshot, unsigned flags)
1500{
1501 return write_midx_internal(source, NULL, NULL,
1502 preferred_pack_name, refs_snapshot,
1503 flags);
1504}
1505
1506int write_midx_file_only(struct odb_source *source,
1507 struct string_list *packs_to_include,
1508 const char *preferred_pack_name,
1509 const char *refs_snapshot, unsigned flags)
1510{
1511 return write_midx_internal(source, packs_to_include, NULL,
1512 preferred_pack_name, refs_snapshot, flags);
1513}
1514
1515int expire_midx_packs(struct odb_source *source, unsigned flags)
1516{
1517 uint32_t i, *count, result = 0;
1518 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1519 struct multi_pack_index *m = get_multi_pack_index(source);
1520 struct progress *progress = NULL;
1521
1522 if (!m)
1523 return 0;
1524
1525 if (m->base_midx)
1526 die(_("cannot expire packs from an incremental multi-pack-index"));
1527
1528 CALLOC_ARRAY(count, m->num_packs);
1529
1530 if (flags & MIDX_PROGRESS)
1531 progress = start_delayed_progress(
1532 source->odb->repo,
1533 _("Counting referenced objects"),
1534 m->num_objects);
1535 for (i = 0; i < m->num_objects; i++) {
1536 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1537 count[pack_int_id]++;
1538 display_progress(progress, i + 1);
1539 }
1540 stop_progress(&progress);
1541
1542 if (flags & MIDX_PROGRESS)
1543 progress = start_delayed_progress(
1544 source->odb->repo,
1545 _("Finding and deleting unreferenced packfiles"),
1546 m->num_packs);
1547 for (i = 0; i < m->num_packs; i++) {
1548 char *pack_name;
1549 display_progress(progress, i + 1);
1550
1551 if (count[i])
1552 continue;
1553
1554 if (prepare_midx_pack(m, i))
1555 continue;
1556
1557 if (m->packs[i]->pack_keep || m->packs[i]->is_cruft)
1558 continue;
1559
1560 pack_name = xstrdup(m->packs[i]->pack_name);
1561 close_pack(m->packs[i]);
1562
1563 string_list_insert(&packs_to_drop, m->pack_names[i]);
1564 unlink_pack_path(pack_name, 0);
1565 free(pack_name);
1566 }
1567 stop_progress(&progress);
1568
1569 free(count);
1570
1571 if (packs_to_drop.nr)
1572 result = write_midx_internal(source, NULL,
1573 &packs_to_drop, NULL, NULL, flags);
1574
1575 string_list_clear(&packs_to_drop, 0);
1576
1577 return result;
1578}
1579
1580struct repack_info {
1581 timestamp_t mtime;
1582 uint32_t referenced_objects;
1583 uint32_t pack_int_id;
1584};
1585
1586static int compare_by_mtime(const void *a_, const void *b_)
1587{
1588 const struct repack_info *a, *b;
1589
1590 a = (const struct repack_info *)a_;
1591 b = (const struct repack_info *)b_;
1592
1593 if (a->mtime < b->mtime)
1594 return -1;
1595 if (a->mtime > b->mtime)
1596 return 1;
1597 return 0;
1598}
1599
1600static int want_included_pack(struct multi_pack_index *m,
1601 int pack_kept_objects,
1602 uint32_t pack_int_id)
1603{
1604 struct packed_git *p;
1605 if (prepare_midx_pack(m, pack_int_id))
1606 return 0;
1607 p = m->packs[pack_int_id];
1608 if (!pack_kept_objects && p->pack_keep)
1609 return 0;
1610 if (p->is_cruft)
1611 return 0;
1612 if (open_pack_index(p) || !p->num_objects)
1613 return 0;
1614 return 1;
1615}
1616
1617static void fill_included_packs_all(struct repository *r,
1618 struct multi_pack_index *m,
1619 unsigned char *include_pack)
1620{
1621 uint32_t i;
1622 int pack_kept_objects = 0;
1623
1624 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1625
1626 for (i = 0; i < m->num_packs; i++) {
1627 if (!want_included_pack(m, pack_kept_objects, i))
1628 continue;
1629
1630 include_pack[i] = 1;
1631 }
1632}
1633
1634static void fill_included_packs_batch(struct repository *r,
1635 struct multi_pack_index *m,
1636 unsigned char *include_pack,
1637 size_t batch_size)
1638{
1639 uint32_t i;
1640 size_t total_size;
1641 struct repack_info *pack_info;
1642 int pack_kept_objects = 0;
1643
1644 CALLOC_ARRAY(pack_info, m->num_packs);
1645
1646 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1647
1648 for (i = 0; i < m->num_packs; i++) {
1649 pack_info[i].pack_int_id = i;
1650
1651 if (prepare_midx_pack(m, i))
1652 continue;
1653
1654 pack_info[i].mtime = m->packs[i]->mtime;
1655 }
1656
1657 for (i = 0; i < m->num_objects; i++) {
1658 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1659 pack_info[pack_int_id].referenced_objects++;
1660 }
1661
1662 QSORT(pack_info, m->num_packs, compare_by_mtime);
1663
1664 total_size = 0;
1665 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1666 uint32_t pack_int_id = pack_info[i].pack_int_id;
1667 struct packed_git *p = m->packs[pack_int_id];
1668 uint64_t expected_size;
1669
1670 if (!want_included_pack(m, pack_kept_objects, pack_int_id))
1671 continue;
1672
1673 /*
1674 * Use shifted integer arithmetic to calculate the
1675 * expected pack size to ~4 significant digits without
1676 * overflow for packsizes less that 1PB.
1677 */
1678 expected_size = (uint64_t)pack_info[i].referenced_objects << 14;
1679 expected_size /= p->num_objects;
1680 expected_size = u64_mult(expected_size, p->pack_size);
1681 expected_size = u64_add(expected_size, 1u << 13) >> 14;
1682
1683 if (expected_size >= batch_size)
1684 continue;
1685
1686 if (unsigned_add_overflows(total_size, (size_t)expected_size))
1687 total_size = SIZE_MAX;
1688 else
1689 total_size += expected_size;
1690
1691 include_pack[pack_int_id] = 1;
1692 }
1693
1694 free(pack_info);
1695}
1696
1697int midx_repack(struct odb_source *source, size_t batch_size, unsigned flags)
1698{
1699 struct repository *r = source->odb->repo;
1700 int result = 0;
1701 uint32_t i, packs_to_repack = 0;
1702 unsigned char *include_pack;
1703 struct child_process cmd = CHILD_PROCESS_INIT;
1704 FILE *cmd_in;
1705 struct multi_pack_index *m = get_multi_pack_index(source);
1706
1707 /*
1708 * When updating the default for these configuration
1709 * variables in builtin/repack.c, these must be adjusted
1710 * to match.
1711 */
1712 int delta_base_offset = 1;
1713 int use_delta_islands = 0;
1714
1715 if (!m)
1716 return 0;
1717 if (m->base_midx)
1718 die(_("cannot repack an incremental multi-pack-index"));
1719
1720 CALLOC_ARRAY(include_pack, m->num_packs);
1721
1722 if (batch_size)
1723 fill_included_packs_batch(r, m, include_pack, batch_size);
1724 else
1725 fill_included_packs_all(r, m, include_pack);
1726
1727 for (i = 0; i < m->num_packs; i++) {
1728 if (include_pack[i])
1729 packs_to_repack++;
1730 }
1731 if (packs_to_repack <= 1)
1732 goto cleanup;
1733
1734 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
1735 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
1736
1737 strvec_push(&cmd.args, "pack-objects");
1738
1739 strvec_pushf(&cmd.args, "%s/pack/pack", source->path);
1740
1741 if (delta_base_offset)
1742 strvec_push(&cmd.args, "--delta-base-offset");
1743 if (use_delta_islands)
1744 strvec_push(&cmd.args, "--delta-islands");
1745
1746 if (flags & MIDX_PROGRESS)
1747 strvec_push(&cmd.args, "--progress");
1748 else
1749 strvec_push(&cmd.args, "-q");
1750
1751 cmd.git_cmd = 1;
1752 cmd.in = cmd.out = -1;
1753
1754 if (start_command(&cmd)) {
1755 error(_("could not start pack-objects"));
1756 result = 1;
1757 goto cleanup;
1758 }
1759
1760 cmd_in = xfdopen(cmd.in, "w");
1761
1762 for (i = 0; i < m->num_objects; i++) {
1763 struct object_id oid;
1764 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1765
1766 if (!include_pack[pack_int_id])
1767 continue;
1768
1769 nth_midxed_object_oid(&oid, m, i);
1770 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
1771 }
1772 fclose(cmd_in);
1773
1774 if (finish_command(&cmd)) {
1775 error(_("could not finish pack-objects"));
1776 result = 1;
1777 goto cleanup;
1778 }
1779
1780 result = write_midx_internal(source, NULL, NULL, NULL, NULL,
1781 flags);
1782
1783cleanup:
1784 free(include_pack);
1785 return result;
1786}