]>
Commit | Line | Data |
---|---|---|
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 | ||
17 | #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */ | |
18 | #define MIDX_VERSION 1 | |
19 | #define MIDX_BYTE_FILE_VERSION 4 | |
20 | #define MIDX_BYTE_HASH_VERSION 5 | |
21 | #define MIDX_BYTE_NUM_CHUNKS 6 | |
22 | #define MIDX_BYTE_NUM_PACKS 8 | |
23 | #define MIDX_HEADER_SIZE 12 | |
24 | #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz) | |
25 | ||
26 | #define MIDX_CHUNK_ALIGNMENT 4 | |
27 | #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */ | |
28 | #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */ | |
29 | #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */ | |
30 | #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */ | |
31 | #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */ | |
32 | #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256) | |
33 | #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t)) | |
34 | #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t)) | |
35 | #define MIDX_LARGE_OFFSET_NEEDED 0x80000000 | |
36 | ||
37 | #define PACK_EXPIRED UINT_MAX | |
38 | ||
39 | static uint8_t oid_version(void) | |
40 | { | |
41 | switch (hash_algo_by_ptr(the_hash_algo)) { | |
42 | case GIT_HASH_SHA1: | |
43 | return 1; | |
44 | case GIT_HASH_SHA256: | |
45 | return 2; | |
46 | default: | |
47 | die(_("invalid hash version")); | |
48 | } | |
49 | } | |
50 | ||
51 | static const unsigned char *get_midx_checksum(struct multi_pack_index *m) | |
52 | { | |
53 | return m->data + m->data_len - the_hash_algo->rawsz; | |
54 | } | |
55 | ||
56 | static char *get_midx_filename(const char *object_dir) | |
57 | { | |
58 | return xstrfmt("%s/pack/multi-pack-index", object_dir); | |
59 | } | |
60 | ||
61 | char *get_midx_rev_filename(struct multi_pack_index *m) | |
62 | { | |
63 | return xstrfmt("%s/pack/multi-pack-index-%s.rev", | |
64 | m->object_dir, hash_to_hex(get_midx_checksum(m))); | |
65 | } | |
66 | ||
67 | static int midx_read_oid_fanout(const unsigned char *chunk_start, | |
68 | size_t chunk_size, void *data) | |
69 | { | |
70 | struct multi_pack_index *m = data; | |
71 | m->chunk_oid_fanout = (uint32_t *)chunk_start; | |
72 | ||
73 | if (chunk_size != 4 * 256) { | |
74 | error(_("multi-pack-index OID fanout is of the wrong size")); | |
75 | return 1; | |
76 | } | |
77 | return 0; | |
78 | } | |
79 | ||
80 | struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local) | |
81 | { | |
82 | struct multi_pack_index *m = NULL; | |
83 | int fd; | |
84 | struct stat st; | |
85 | size_t midx_size; | |
86 | void *midx_map = NULL; | |
87 | uint32_t hash_version; | |
88 | char *midx_name = get_midx_filename(object_dir); | |
89 | uint32_t i; | |
90 | const char *cur_pack_name; | |
91 | struct chunkfile *cf = NULL; | |
92 | ||
93 | fd = git_open(midx_name); | |
94 | ||
95 | if (fd < 0) | |
96 | goto cleanup_fail; | |
97 | if (fstat(fd, &st)) { | |
98 | error_errno(_("failed to read %s"), midx_name); | |
99 | goto cleanup_fail; | |
100 | } | |
101 | ||
102 | midx_size = xsize_t(st.st_size); | |
103 | ||
104 | if (midx_size < MIDX_MIN_SIZE) { | |
105 | error(_("multi-pack-index file %s is too small"), midx_name); | |
106 | goto cleanup_fail; | |
107 | } | |
108 | ||
109 | FREE_AND_NULL(midx_name); | |
110 | ||
111 | midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
112 | close(fd); | |
113 | ||
114 | FLEX_ALLOC_STR(m, object_dir, object_dir); | |
115 | m->data = midx_map; | |
116 | m->data_len = midx_size; | |
117 | m->local = local; | |
118 | ||
119 | m->signature = get_be32(m->data); | |
120 | if (m->signature != MIDX_SIGNATURE) | |
121 | die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"), | |
122 | m->signature, MIDX_SIGNATURE); | |
123 | ||
124 | m->version = m->data[MIDX_BYTE_FILE_VERSION]; | |
125 | if (m->version != MIDX_VERSION) | |
126 | die(_("multi-pack-index version %d not recognized"), | |
127 | m->version); | |
128 | ||
129 | hash_version = m->data[MIDX_BYTE_HASH_VERSION]; | |
130 | if (hash_version != oid_version()) { | |
131 | error(_("multi-pack-index hash version %u does not match version %u"), | |
132 | hash_version, oid_version()); | |
133 | goto cleanup_fail; | |
134 | } | |
135 | m->hash_len = the_hash_algo->rawsz; | |
136 | ||
137 | m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS]; | |
138 | ||
139 | m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS); | |
140 | ||
141 | cf = init_chunkfile(NULL); | |
142 | ||
143 | if (read_table_of_contents(cf, m->data, midx_size, | |
144 | MIDX_HEADER_SIZE, m->num_chunks)) | |
145 | goto cleanup_fail; | |
146 | ||
147 | if (pair_chunk(cf, MIDX_CHUNKID_PACKNAMES, &m->chunk_pack_names) == CHUNK_NOT_FOUND) | |
148 | die(_("multi-pack-index missing required pack-name chunk")); | |
149 | if (read_chunk(cf, MIDX_CHUNKID_OIDFANOUT, midx_read_oid_fanout, m) == CHUNK_NOT_FOUND) | |
150 | die(_("multi-pack-index missing required OID fanout chunk")); | |
151 | if (pair_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, &m->chunk_oid_lookup) == CHUNK_NOT_FOUND) | |
152 | die(_("multi-pack-index missing required OID lookup chunk")); | |
153 | if (pair_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, &m->chunk_object_offsets) == CHUNK_NOT_FOUND) | |
154 | die(_("multi-pack-index missing required object offsets chunk")); | |
155 | ||
156 | pair_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, &m->chunk_large_offsets); | |
157 | ||
158 | m->num_objects = ntohl(m->chunk_oid_fanout[255]); | |
159 | ||
160 | CALLOC_ARRAY(m->pack_names, m->num_packs); | |
161 | CALLOC_ARRAY(m->packs, m->num_packs); | |
162 | ||
163 | cur_pack_name = (const char *)m->chunk_pack_names; | |
164 | for (i = 0; i < m->num_packs; i++) { | |
165 | m->pack_names[i] = cur_pack_name; | |
166 | ||
167 | cur_pack_name += strlen(cur_pack_name) + 1; | |
168 | ||
169 | if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0) | |
170 | die(_("multi-pack-index pack names out of order: '%s' before '%s'"), | |
171 | m->pack_names[i - 1], | |
172 | m->pack_names[i]); | |
173 | } | |
174 | ||
175 | trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs); | |
176 | trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects); | |
177 | ||
178 | return m; | |
179 | ||
180 | cleanup_fail: | |
181 | free(m); | |
182 | free(midx_name); | |
183 | free(cf); | |
184 | if (midx_map) | |
185 | munmap(midx_map, midx_size); | |
186 | if (0 <= fd) | |
187 | close(fd); | |
188 | return NULL; | |
189 | } | |
190 | ||
191 | void close_midx(struct multi_pack_index *m) | |
192 | { | |
193 | uint32_t i; | |
194 | ||
195 | if (!m) | |
196 | return; | |
197 | ||
198 | munmap((unsigned char *)m->data, m->data_len); | |
199 | ||
200 | for (i = 0; i < m->num_packs; i++) { | |
201 | if (m->packs[i]) | |
202 | m->packs[i]->multi_pack_index = 0; | |
203 | } | |
204 | FREE_AND_NULL(m->packs); | |
205 | FREE_AND_NULL(m->pack_names); | |
206 | } | |
207 | ||
208 | int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id) | |
209 | { | |
210 | struct strbuf pack_name = STRBUF_INIT; | |
211 | struct packed_git *p; | |
212 | ||
213 | if (pack_int_id >= m->num_packs) | |
214 | die(_("bad pack-int-id: %u (%u total packs)"), | |
215 | pack_int_id, m->num_packs); | |
216 | ||
217 | if (m->packs[pack_int_id]) | |
218 | return 0; | |
219 | ||
220 | strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir, | |
221 | m->pack_names[pack_int_id]); | |
222 | ||
223 | p = add_packed_git(pack_name.buf, pack_name.len, m->local); | |
224 | strbuf_release(&pack_name); | |
225 | ||
226 | if (!p) | |
227 | return 1; | |
228 | ||
229 | p->multi_pack_index = 1; | |
230 | m->packs[pack_int_id] = p; | |
231 | install_packed_git(r, p); | |
232 | list_add_tail(&p->mru, &r->objects->packed_git_mru); | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
237 | int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result) | |
238 | { | |
239 | return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup, | |
240 | the_hash_algo->rawsz, result); | |
241 | } | |
242 | ||
243 | struct object_id *nth_midxed_object_oid(struct object_id *oid, | |
244 | struct multi_pack_index *m, | |
245 | uint32_t n) | |
246 | { | |
247 | if (n >= m->num_objects) | |
248 | return NULL; | |
249 | ||
250 | oidread(oid, m->chunk_oid_lookup + m->hash_len * n); | |
251 | return oid; | |
252 | } | |
253 | ||
254 | off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos) | |
255 | { | |
256 | const unsigned char *offset_data; | |
257 | uint32_t offset32; | |
258 | ||
259 | offset_data = m->chunk_object_offsets + (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH; | |
260 | offset32 = get_be32(offset_data + sizeof(uint32_t)); | |
261 | ||
262 | if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) { | |
263 | if (sizeof(off_t) < sizeof(uint64_t)) | |
264 | die(_("multi-pack-index stores a 64-bit offset, but off_t is too small")); | |
265 | ||
266 | offset32 ^= MIDX_LARGE_OFFSET_NEEDED; | |
267 | return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32); | |
268 | } | |
269 | ||
270 | return offset32; | |
271 | } | |
272 | ||
273 | uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos) | |
274 | { | |
275 | return get_be32(m->chunk_object_offsets + | |
276 | (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH); | |
277 | } | |
278 | ||
279 | static int nth_midxed_pack_entry(struct repository *r, | |
280 | struct multi_pack_index *m, | |
281 | struct pack_entry *e, | |
282 | uint32_t pos) | |
283 | { | |
284 | uint32_t pack_int_id; | |
285 | struct packed_git *p; | |
286 | ||
287 | if (pos >= m->num_objects) | |
288 | return 0; | |
289 | ||
290 | pack_int_id = nth_midxed_pack_int_id(m, pos); | |
291 | ||
292 | if (prepare_midx_pack(r, m, pack_int_id)) | |
293 | return 0; | |
294 | p = m->packs[pack_int_id]; | |
295 | ||
296 | /* | |
297 | * We are about to tell the caller where they can locate the | |
298 | * requested object. We better make sure the packfile is | |
299 | * still here and can be accessed before supplying that | |
300 | * answer, as it may have been deleted since the MIDX was | |
301 | * loaded! | |
302 | */ | |
303 | if (!is_pack_valid(p)) | |
304 | return 0; | |
305 | ||
306 | if (p->num_bad_objects) { | |
307 | uint32_t i; | |
308 | struct object_id oid; | |
309 | nth_midxed_object_oid(&oid, m, pos); | |
310 | for (i = 0; i < p->num_bad_objects; i++) | |
311 | if (hasheq(oid.hash, | |
312 | p->bad_object_sha1 + the_hash_algo->rawsz * i)) | |
313 | return 0; | |
314 | } | |
315 | ||
316 | e->offset = nth_midxed_offset(m, pos); | |
317 | e->p = p; | |
318 | ||
319 | return 1; | |
320 | } | |
321 | ||
322 | int fill_midx_entry(struct repository * r, | |
323 | const struct object_id *oid, | |
324 | struct pack_entry *e, | |
325 | struct multi_pack_index *m) | |
326 | { | |
327 | uint32_t pos; | |
328 | ||
329 | if (!bsearch_midx(oid, m, &pos)) | |
330 | return 0; | |
331 | ||
332 | return nth_midxed_pack_entry(r, m, e, pos); | |
333 | } | |
334 | ||
335 | /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */ | |
336 | static int cmp_idx_or_pack_name(const char *idx_or_pack_name, | |
337 | const char *idx_name) | |
338 | { | |
339 | /* Skip past any initial matching prefix. */ | |
340 | while (*idx_name && *idx_name == *idx_or_pack_name) { | |
341 | idx_name++; | |
342 | idx_or_pack_name++; | |
343 | } | |
344 | ||
345 | /* | |
346 | * If we didn't match completely, we may have matched "pack-1234." and | |
347 | * be left with "idx" and "pack" respectively, which is also OK. We do | |
348 | * not have to check for "idx" and "idx", because that would have been | |
349 | * a complete match (and in that case these strcmps will be false, but | |
350 | * we'll correctly return 0 from the final strcmp() below. | |
351 | * | |
352 | * Technically this matches "fooidx" and "foopack", but we'd never have | |
353 | * such names in the first place. | |
354 | */ | |
355 | if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack")) | |
356 | return 0; | |
357 | ||
358 | /* | |
359 | * This not only checks for a complete match, but also orders based on | |
360 | * the first non-identical character, which means our ordering will | |
361 | * match a raw strcmp(). That makes it OK to use this to binary search | |
362 | * a naively-sorted list. | |
363 | */ | |
364 | return strcmp(idx_or_pack_name, idx_name); | |
365 | } | |
366 | ||
367 | int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name) | |
368 | { | |
369 | uint32_t first = 0, last = m->num_packs; | |
370 | ||
371 | while (first < last) { | |
372 | uint32_t mid = first + (last - first) / 2; | |
373 | const char *current; | |
374 | int cmp; | |
375 | ||
376 | current = m->pack_names[mid]; | |
377 | cmp = cmp_idx_or_pack_name(idx_or_pack_name, current); | |
378 | if (!cmp) | |
379 | return 1; | |
380 | if (cmp > 0) { | |
381 | first = mid + 1; | |
382 | continue; | |
383 | } | |
384 | last = mid; | |
385 | } | |
386 | ||
387 | return 0; | |
388 | } | |
389 | ||
390 | int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local) | |
391 | { | |
392 | struct multi_pack_index *m; | |
393 | struct multi_pack_index *m_search; | |
394 | ||
395 | prepare_repo_settings(r); | |
396 | if (!r->settings.core_multi_pack_index) | |
397 | return 0; | |
398 | ||
399 | for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next) | |
400 | if (!strcmp(object_dir, m_search->object_dir)) | |
401 | return 1; | |
402 | ||
403 | m = load_multi_pack_index(object_dir, local); | |
404 | ||
405 | if (m) { | |
406 | struct multi_pack_index *mp = r->objects->multi_pack_index; | |
407 | if (mp) { | |
408 | m->next = mp->next; | |
409 | mp->next = m; | |
410 | } else | |
411 | r->objects->multi_pack_index = m; | |
412 | return 1; | |
413 | } | |
414 | ||
415 | return 0; | |
416 | } | |
417 | ||
418 | static size_t write_midx_header(struct hashfile *f, | |
419 | unsigned char num_chunks, | |
420 | uint32_t num_packs) | |
421 | { | |
422 | hashwrite_be32(f, MIDX_SIGNATURE); | |
423 | hashwrite_u8(f, MIDX_VERSION); | |
424 | hashwrite_u8(f, oid_version()); | |
425 | hashwrite_u8(f, num_chunks); | |
426 | hashwrite_u8(f, 0); /* unused */ | |
427 | hashwrite_be32(f, num_packs); | |
428 | ||
429 | return MIDX_HEADER_SIZE; | |
430 | } | |
431 | ||
432 | struct pack_info { | |
433 | uint32_t orig_pack_int_id; | |
434 | char *pack_name; | |
435 | struct packed_git *p; | |
436 | unsigned expired : 1; | |
437 | }; | |
438 | ||
439 | static int pack_info_compare(const void *_a, const void *_b) | |
440 | { | |
441 | struct pack_info *a = (struct pack_info *)_a; | |
442 | struct pack_info *b = (struct pack_info *)_b; | |
443 | return strcmp(a->pack_name, b->pack_name); | |
444 | } | |
445 | ||
446 | static int idx_or_pack_name_cmp(const void *_va, const void *_vb) | |
447 | { | |
448 | const char *pack_name = _va; | |
449 | const struct pack_info *compar = _vb; | |
450 | ||
451 | return cmp_idx_or_pack_name(pack_name, compar->pack_name); | |
452 | } | |
453 | ||
454 | struct write_midx_context { | |
455 | struct pack_info *info; | |
456 | uint32_t nr; | |
457 | uint32_t alloc; | |
458 | struct multi_pack_index *m; | |
459 | struct progress *progress; | |
460 | unsigned pack_paths_checked; | |
461 | ||
462 | struct pack_midx_entry *entries; | |
463 | uint32_t entries_nr; | |
464 | ||
465 | uint32_t *pack_perm; | |
466 | uint32_t *pack_order; | |
467 | unsigned large_offsets_needed:1; | |
468 | uint32_t num_large_offsets; | |
469 | ||
470 | int preferred_pack_idx; | |
471 | }; | |
472 | ||
473 | static void add_pack_to_midx(const char *full_path, size_t full_path_len, | |
474 | const char *file_name, void *data) | |
475 | { | |
476 | struct write_midx_context *ctx = data; | |
477 | ||
478 | if (ends_with(file_name, ".idx")) { | |
479 | display_progress(ctx->progress, ++ctx->pack_paths_checked); | |
480 | if (ctx->m && midx_contains_pack(ctx->m, file_name)) | |
481 | return; | |
482 | ||
483 | ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc); | |
484 | ||
485 | ctx->info[ctx->nr].p = add_packed_git(full_path, | |
486 | full_path_len, | |
487 | 0); | |
488 | ||
489 | if (!ctx->info[ctx->nr].p) { | |
490 | warning(_("failed to add packfile '%s'"), | |
491 | full_path); | |
492 | return; | |
493 | } | |
494 | ||
495 | if (open_pack_index(ctx->info[ctx->nr].p)) { | |
496 | warning(_("failed to open pack-index '%s'"), | |
497 | full_path); | |
498 | close_pack(ctx->info[ctx->nr].p); | |
499 | FREE_AND_NULL(ctx->info[ctx->nr].p); | |
500 | return; | |
501 | } | |
502 | ||
503 | ctx->info[ctx->nr].pack_name = xstrdup(file_name); | |
504 | ctx->info[ctx->nr].orig_pack_int_id = ctx->nr; | |
505 | ctx->info[ctx->nr].expired = 0; | |
506 | ctx->nr++; | |
507 | } | |
508 | } | |
509 | ||
510 | struct pack_midx_entry { | |
511 | struct object_id oid; | |
512 | uint32_t pack_int_id; | |
513 | time_t pack_mtime; | |
514 | uint64_t offset; | |
515 | unsigned preferred : 1; | |
516 | }; | |
517 | ||
518 | static int midx_oid_compare(const void *_a, const void *_b) | |
519 | { | |
520 | const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a; | |
521 | const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b; | |
522 | int cmp = oidcmp(&a->oid, &b->oid); | |
523 | ||
524 | if (cmp) | |
525 | return cmp; | |
526 | ||
527 | /* Sort objects in a preferred pack first when multiple copies exist. */ | |
528 | if (a->preferred > b->preferred) | |
529 | return -1; | |
530 | if (a->preferred < b->preferred) | |
531 | return 1; | |
532 | ||
533 | if (a->pack_mtime > b->pack_mtime) | |
534 | return -1; | |
535 | else if (a->pack_mtime < b->pack_mtime) | |
536 | return 1; | |
537 | ||
538 | return a->pack_int_id - b->pack_int_id; | |
539 | } | |
540 | ||
541 | static int nth_midxed_pack_midx_entry(struct multi_pack_index *m, | |
542 | struct pack_midx_entry *e, | |
543 | uint32_t pos) | |
544 | { | |
545 | if (pos >= m->num_objects) | |
546 | return 1; | |
547 | ||
548 | nth_midxed_object_oid(&e->oid, m, pos); | |
549 | e->pack_int_id = nth_midxed_pack_int_id(m, pos); | |
550 | e->offset = nth_midxed_offset(m, pos); | |
551 | ||
552 | /* consider objects in midx to be from "old" packs */ | |
553 | e->pack_mtime = 0; | |
554 | return 0; | |
555 | } | |
556 | ||
557 | static void fill_pack_entry(uint32_t pack_int_id, | |
558 | struct packed_git *p, | |
559 | uint32_t cur_object, | |
560 | struct pack_midx_entry *entry, | |
561 | int preferred) | |
562 | { | |
563 | if (nth_packed_object_id(&entry->oid, p, cur_object) < 0) | |
564 | die(_("failed to locate object %d in packfile"), cur_object); | |
565 | ||
566 | entry->pack_int_id = pack_int_id; | |
567 | entry->pack_mtime = p->mtime; | |
568 | ||
569 | entry->offset = nth_packed_object_offset(p, cur_object); | |
570 | entry->preferred = !!preferred; | |
571 | } | |
572 | ||
573 | /* | |
574 | * It is possible to artificially get into a state where there are many | |
575 | * duplicate copies of objects. That can create high memory pressure if | |
576 | * we are to create a list of all objects before de-duplication. To reduce | |
577 | * this memory pressure without a significant performance drop, automatically | |
578 | * group objects by the first byte of their object id. Use the IDX fanout | |
579 | * tables to group the data, copy to a local array, then sort. | |
580 | * | |
581 | * Copy only the de-duplicated entries (selected by most-recent modified time | |
582 | * of a packfile containing the object). | |
583 | */ | |
584 | static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m, | |
585 | struct pack_info *info, | |
586 | uint32_t nr_packs, | |
587 | uint32_t *nr_objects, | |
588 | int preferred_pack) | |
589 | { | |
590 | uint32_t cur_fanout, cur_pack, cur_object; | |
591 | uint32_t alloc_fanout, alloc_objects, total_objects = 0; | |
592 | struct pack_midx_entry *entries_by_fanout = NULL; | |
593 | struct pack_midx_entry *deduplicated_entries = NULL; | |
594 | uint32_t start_pack = m ? m->num_packs : 0; | |
595 | ||
596 | for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) | |
597 | total_objects += info[cur_pack].p->num_objects; | |
598 | ||
599 | /* | |
600 | * As we de-duplicate by fanout value, we expect the fanout | |
601 | * slices to be evenly distributed, with some noise. Hence, | |
602 | * allocate slightly more than one 256th. | |
603 | */ | |
604 | alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16; | |
605 | ||
606 | ALLOC_ARRAY(entries_by_fanout, alloc_fanout); | |
607 | ALLOC_ARRAY(deduplicated_entries, alloc_objects); | |
608 | *nr_objects = 0; | |
609 | ||
610 | for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) { | |
611 | uint32_t nr_fanout = 0; | |
612 | ||
613 | if (m) { | |
614 | uint32_t start = 0, end; | |
615 | ||
616 | if (cur_fanout) | |
617 | start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]); | |
618 | end = ntohl(m->chunk_oid_fanout[cur_fanout]); | |
619 | ||
620 | for (cur_object = start; cur_object < end; cur_object++) { | |
621 | ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout); | |
622 | nth_midxed_pack_midx_entry(m, | |
623 | &entries_by_fanout[nr_fanout], | |
624 | cur_object); | |
625 | if (nth_midxed_pack_int_id(m, cur_object) == preferred_pack) | |
626 | entries_by_fanout[nr_fanout].preferred = 1; | |
627 | else | |
628 | entries_by_fanout[nr_fanout].preferred = 0; | |
629 | nr_fanout++; | |
630 | } | |
631 | } | |
632 | ||
633 | for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) { | |
634 | uint32_t start = 0, end; | |
635 | int preferred = cur_pack == preferred_pack; | |
636 | ||
637 | if (cur_fanout) | |
638 | start = get_pack_fanout(info[cur_pack].p, cur_fanout - 1); | |
639 | end = get_pack_fanout(info[cur_pack].p, cur_fanout); | |
640 | ||
641 | for (cur_object = start; cur_object < end; cur_object++) { | |
642 | ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout); | |
643 | fill_pack_entry(cur_pack, | |
644 | info[cur_pack].p, | |
645 | cur_object, | |
646 | &entries_by_fanout[nr_fanout], | |
647 | preferred); | |
648 | nr_fanout++; | |
649 | } | |
650 | } | |
651 | ||
652 | QSORT(entries_by_fanout, nr_fanout, midx_oid_compare); | |
653 | ||
654 | /* | |
655 | * The batch is now sorted by OID and then mtime (descending). | |
656 | * Take only the first duplicate. | |
657 | */ | |
658 | for (cur_object = 0; cur_object < nr_fanout; cur_object++) { | |
659 | if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid, | |
660 | &entries_by_fanout[cur_object].oid)) | |
661 | continue; | |
662 | ||
663 | ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects); | |
664 | memcpy(&deduplicated_entries[*nr_objects], | |
665 | &entries_by_fanout[cur_object], | |
666 | sizeof(struct pack_midx_entry)); | |
667 | (*nr_objects)++; | |
668 | } | |
669 | } | |
670 | ||
671 | free(entries_by_fanout); | |
672 | return deduplicated_entries; | |
673 | } | |
674 | ||
675 | static int write_midx_pack_names(struct hashfile *f, void *data) | |
676 | { | |
677 | struct write_midx_context *ctx = data; | |
678 | uint32_t i; | |
679 | unsigned char padding[MIDX_CHUNK_ALIGNMENT]; | |
680 | size_t written = 0; | |
681 | ||
682 | for (i = 0; i < ctx->nr; i++) { | |
683 | size_t writelen; | |
684 | ||
685 | if (ctx->info[i].expired) | |
686 | continue; | |
687 | ||
688 | if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0) | |
689 | BUG("incorrect pack-file order: %s before %s", | |
690 | ctx->info[i - 1].pack_name, | |
691 | ctx->info[i].pack_name); | |
692 | ||
693 | writelen = strlen(ctx->info[i].pack_name) + 1; | |
694 | hashwrite(f, ctx->info[i].pack_name, writelen); | |
695 | written += writelen; | |
696 | } | |
697 | ||
698 | /* add padding to be aligned */ | |
699 | i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT); | |
700 | if (i < MIDX_CHUNK_ALIGNMENT) { | |
701 | memset(padding, 0, sizeof(padding)); | |
702 | hashwrite(f, padding, i); | |
703 | } | |
704 | ||
705 | return 0; | |
706 | } | |
707 | ||
708 | static int write_midx_oid_fanout(struct hashfile *f, | |
709 | void *data) | |
710 | { | |
711 | struct write_midx_context *ctx = data; | |
712 | struct pack_midx_entry *list = ctx->entries; | |
713 | struct pack_midx_entry *last = ctx->entries + ctx->entries_nr; | |
714 | uint32_t count = 0; | |
715 | uint32_t i; | |
716 | ||
717 | /* | |
718 | * Write the first-level table (the list is sorted, | |
719 | * but we use a 256-entry lookup to be able to avoid | |
720 | * having to do eight extra binary search iterations). | |
721 | */ | |
722 | for (i = 0; i < 256; i++) { | |
723 | struct pack_midx_entry *next = list; | |
724 | ||
725 | while (next < last && next->oid.hash[0] == i) { | |
726 | count++; | |
727 | next++; | |
728 | } | |
729 | ||
730 | hashwrite_be32(f, count); | |
731 | list = next; | |
732 | } | |
733 | ||
734 | return 0; | |
735 | } | |
736 | ||
737 | static int write_midx_oid_lookup(struct hashfile *f, | |
738 | void *data) | |
739 | { | |
740 | struct write_midx_context *ctx = data; | |
741 | unsigned char hash_len = the_hash_algo->rawsz; | |
742 | struct pack_midx_entry *list = ctx->entries; | |
743 | uint32_t i; | |
744 | ||
745 | for (i = 0; i < ctx->entries_nr; i++) { | |
746 | struct pack_midx_entry *obj = list++; | |
747 | ||
748 | if (i < ctx->entries_nr - 1) { | |
749 | struct pack_midx_entry *next = list; | |
750 | if (oidcmp(&obj->oid, &next->oid) >= 0) | |
751 | BUG("OIDs not in order: %s >= %s", | |
752 | oid_to_hex(&obj->oid), | |
753 | oid_to_hex(&next->oid)); | |
754 | } | |
755 | ||
756 | hashwrite(f, obj->oid.hash, (int)hash_len); | |
757 | } | |
758 | ||
759 | return 0; | |
760 | } | |
761 | ||
762 | static int write_midx_object_offsets(struct hashfile *f, | |
763 | void *data) | |
764 | { | |
765 | struct write_midx_context *ctx = data; | |
766 | struct pack_midx_entry *list = ctx->entries; | |
767 | uint32_t i, nr_large_offset = 0; | |
768 | ||
769 | for (i = 0; i < ctx->entries_nr; i++) { | |
770 | struct pack_midx_entry *obj = list++; | |
771 | ||
772 | if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED) | |
773 | BUG("object %s is in an expired pack with int-id %d", | |
774 | oid_to_hex(&obj->oid), | |
775 | obj->pack_int_id); | |
776 | ||
777 | hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]); | |
778 | ||
779 | if (ctx->large_offsets_needed && obj->offset >> 31) | |
780 | hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++); | |
781 | else if (!ctx->large_offsets_needed && obj->offset >> 32) | |
782 | BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!", | |
783 | oid_to_hex(&obj->oid), | |
784 | obj->offset); | |
785 | else | |
786 | hashwrite_be32(f, (uint32_t)obj->offset); | |
787 | } | |
788 | ||
789 | return 0; | |
790 | } | |
791 | ||
792 | static int write_midx_large_offsets(struct hashfile *f, | |
793 | void *data) | |
794 | { | |
795 | struct write_midx_context *ctx = data; | |
796 | struct pack_midx_entry *list = ctx->entries; | |
797 | struct pack_midx_entry *end = ctx->entries + ctx->entries_nr; | |
798 | uint32_t nr_large_offset = ctx->num_large_offsets; | |
799 | ||
800 | while (nr_large_offset) { | |
801 | struct pack_midx_entry *obj; | |
802 | uint64_t offset; | |
803 | ||
804 | if (list >= end) | |
805 | BUG("too many large-offset objects"); | |
806 | ||
807 | obj = list++; | |
808 | offset = obj->offset; | |
809 | ||
810 | if (!(offset >> 31)) | |
811 | continue; | |
812 | ||
813 | hashwrite_be64(f, offset); | |
814 | ||
815 | nr_large_offset--; | |
816 | } | |
817 | ||
818 | return 0; | |
819 | } | |
820 | ||
821 | struct midx_pack_order_data { | |
822 | uint32_t nr; | |
823 | uint32_t pack; | |
824 | off_t offset; | |
825 | }; | |
826 | ||
827 | static int midx_pack_order_cmp(const void *va, const void *vb) | |
828 | { | |
829 | const struct midx_pack_order_data *a = va, *b = vb; | |
830 | if (a->pack < b->pack) | |
831 | return -1; | |
832 | else if (a->pack > b->pack) | |
833 | return 1; | |
834 | else if (a->offset < b->offset) | |
835 | return -1; | |
836 | else if (a->offset > b->offset) | |
837 | return 1; | |
838 | else | |
839 | return 0; | |
840 | } | |
841 | ||
842 | static uint32_t *midx_pack_order(struct write_midx_context *ctx) | |
843 | { | |
844 | struct midx_pack_order_data *data; | |
845 | uint32_t *pack_order; | |
846 | uint32_t i; | |
847 | ||
848 | ALLOC_ARRAY(data, ctx->entries_nr); | |
849 | for (i = 0; i < ctx->entries_nr; i++) { | |
850 | struct pack_midx_entry *e = &ctx->entries[i]; | |
851 | data[i].nr = i; | |
852 | data[i].pack = ctx->pack_perm[e->pack_int_id]; | |
853 | if (!e->preferred) | |
854 | data[i].pack |= (1U << 31); | |
855 | data[i].offset = e->offset; | |
856 | } | |
857 | ||
858 | QSORT(data, ctx->entries_nr, midx_pack_order_cmp); | |
859 | ||
860 | ALLOC_ARRAY(pack_order, ctx->entries_nr); | |
861 | for (i = 0; i < ctx->entries_nr; i++) | |
862 | pack_order[i] = data[i].nr; | |
863 | free(data); | |
864 | ||
865 | return pack_order; | |
866 | } | |
867 | ||
868 | static void write_midx_reverse_index(char *midx_name, unsigned char *midx_hash, | |
869 | struct write_midx_context *ctx) | |
870 | { | |
871 | struct strbuf buf = STRBUF_INIT; | |
872 | const char *tmp_file; | |
873 | ||
874 | strbuf_addf(&buf, "%s-%s.rev", midx_name, hash_to_hex(midx_hash)); | |
875 | ||
876 | tmp_file = write_rev_file_order(NULL, ctx->pack_order, ctx->entries_nr, | |
877 | midx_hash, WRITE_REV); | |
878 | ||
879 | if (finalize_object_file(tmp_file, buf.buf)) | |
880 | die(_("cannot store reverse index file")); | |
881 | ||
882 | strbuf_release(&buf); | |
883 | } | |
884 | ||
885 | static void clear_midx_files_ext(const char *object_dir, const char *ext, | |
886 | unsigned char *keep_hash); | |
887 | ||
888 | static int midx_checksum_valid(struct multi_pack_index *m) | |
889 | { | |
890 | return hashfile_checksum_valid(m->data, m->data_len); | |
891 | } | |
892 | ||
893 | static int write_midx_internal(const char *object_dir, struct multi_pack_index *m, | |
894 | struct string_list *packs_to_drop, | |
895 | const char *preferred_pack_name, | |
896 | unsigned flags) | |
897 | { | |
898 | char *midx_name; | |
899 | unsigned char midx_hash[GIT_MAX_RAWSZ]; | |
900 | uint32_t i; | |
901 | struct hashfile *f = NULL; | |
902 | struct lock_file lk; | |
903 | struct write_midx_context ctx = { 0 }; | |
904 | int pack_name_concat_len = 0; | |
905 | int dropped_packs = 0; | |
906 | int result = 0; | |
907 | struct chunkfile *cf; | |
908 | ||
909 | midx_name = get_midx_filename(object_dir); | |
910 | if (safe_create_leading_directories(midx_name)) | |
911 | die_errno(_("unable to create leading directories of %s"), | |
912 | midx_name); | |
913 | ||
914 | if (m) | |
915 | ctx.m = m; | |
916 | else | |
917 | ctx.m = load_multi_pack_index(object_dir, 1); | |
918 | ||
919 | if (ctx.m && !midx_checksum_valid(ctx.m)) { | |
920 | warning(_("ignoring existing multi-pack-index; checksum mismatch")); | |
921 | ctx.m = NULL; | |
922 | } | |
923 | ||
924 | ctx.nr = 0; | |
925 | ctx.alloc = ctx.m ? ctx.m->num_packs : 16; | |
926 | ctx.info = NULL; | |
927 | ALLOC_ARRAY(ctx.info, ctx.alloc); | |
928 | ||
929 | if (ctx.m) { | |
930 | for (i = 0; i < ctx.m->num_packs; i++) { | |
931 | ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc); | |
932 | ||
933 | ctx.info[ctx.nr].orig_pack_int_id = i; | |
934 | ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]); | |
935 | ctx.info[ctx.nr].p = NULL; | |
936 | ctx.info[ctx.nr].expired = 0; | |
937 | ctx.nr++; | |
938 | } | |
939 | } | |
940 | ||
941 | ctx.pack_paths_checked = 0; | |
942 | if (flags & MIDX_PROGRESS) | |
943 | ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0); | |
944 | else | |
945 | ctx.progress = NULL; | |
946 | ||
947 | for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx); | |
948 | stop_progress(&ctx.progress); | |
949 | ||
950 | if (ctx.m && ctx.nr == ctx.m->num_packs && !packs_to_drop) | |
951 | goto cleanup; | |
952 | ||
953 | ctx.preferred_pack_idx = -1; | |
954 | if (preferred_pack_name) { | |
955 | for (i = 0; i < ctx.nr; i++) { | |
956 | if (!cmp_idx_or_pack_name(preferred_pack_name, | |
957 | ctx.info[i].pack_name)) { | |
958 | ctx.preferred_pack_idx = i; | |
959 | break; | |
960 | } | |
961 | } | |
962 | } | |
963 | ||
964 | ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr, | |
965 | ctx.preferred_pack_idx); | |
966 | ||
967 | ctx.large_offsets_needed = 0; | |
968 | for (i = 0; i < ctx.entries_nr; i++) { | |
969 | if (ctx.entries[i].offset > 0x7fffffff) | |
970 | ctx.num_large_offsets++; | |
971 | if (ctx.entries[i].offset > 0xffffffff) | |
972 | ctx.large_offsets_needed = 1; | |
973 | } | |
974 | ||
975 | QSORT(ctx.info, ctx.nr, pack_info_compare); | |
976 | ||
977 | if (packs_to_drop && packs_to_drop->nr) { | |
978 | int drop_index = 0; | |
979 | int missing_drops = 0; | |
980 | ||
981 | for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) { | |
982 | int cmp = strcmp(ctx.info[i].pack_name, | |
983 | packs_to_drop->items[drop_index].string); | |
984 | ||
985 | if (!cmp) { | |
986 | drop_index++; | |
987 | ctx.info[i].expired = 1; | |
988 | } else if (cmp > 0) { | |
989 | error(_("did not see pack-file %s to drop"), | |
990 | packs_to_drop->items[drop_index].string); | |
991 | drop_index++; | |
992 | missing_drops++; | |
993 | i--; | |
994 | } else { | |
995 | ctx.info[i].expired = 0; | |
996 | } | |
997 | } | |
998 | ||
999 | if (missing_drops) { | |
1000 | result = 1; | |
1001 | goto cleanup; | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | /* | |
1006 | * pack_perm stores a permutation between pack-int-ids from the | |
1007 | * previous multi-pack-index to the new one we are writing: | |
1008 | * | |
1009 | * pack_perm[old_id] = new_id | |
1010 | */ | |
1011 | ALLOC_ARRAY(ctx.pack_perm, ctx.nr); | |
1012 | for (i = 0; i < ctx.nr; i++) { | |
1013 | if (ctx.info[i].expired) { | |
1014 | dropped_packs++; | |
1015 | ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED; | |
1016 | } else { | |
1017 | ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs; | |
1018 | } | |
1019 | } | |
1020 | ||
1021 | for (i = 0; i < ctx.nr; i++) { | |
1022 | if (!ctx.info[i].expired) | |
1023 | pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1; | |
1024 | } | |
1025 | ||
1026 | /* Check that the preferred pack wasn't expired (if given). */ | |
1027 | if (preferred_pack_name) { | |
1028 | struct pack_info *preferred = bsearch(preferred_pack_name, | |
1029 | ctx.info, ctx.nr, | |
1030 | sizeof(*ctx.info), | |
1031 | idx_or_pack_name_cmp); | |
1032 | ||
1033 | if (!preferred) | |
1034 | warning(_("unknown preferred pack: '%s'"), | |
1035 | preferred_pack_name); | |
1036 | else { | |
1037 | uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id]; | |
1038 | if (perm == PACK_EXPIRED) | |
1039 | warning(_("preferred pack '%s' is expired"), | |
1040 | preferred_pack_name); | |
1041 | } | |
1042 | } | |
1043 | ||
1044 | if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT) | |
1045 | pack_name_concat_len += MIDX_CHUNK_ALIGNMENT - | |
1046 | (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT); | |
1047 | ||
1048 | hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR); | |
1049 | f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk)); | |
1050 | ||
1051 | if (ctx.m) | |
1052 | close_midx(ctx.m); | |
1053 | ||
1054 | if (ctx.nr - dropped_packs == 0) { | |
1055 | error(_("no pack files to index.")); | |
1056 | result = 1; | |
1057 | goto cleanup; | |
1058 | } | |
1059 | ||
1060 | cf = init_chunkfile(f); | |
1061 | ||
1062 | add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len, | |
1063 | write_midx_pack_names); | |
1064 | add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE, | |
1065 | write_midx_oid_fanout); | |
1066 | add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, | |
1067 | (size_t)ctx.entries_nr * the_hash_algo->rawsz, | |
1068 | write_midx_oid_lookup); | |
1069 | add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, | |
1070 | (size_t)ctx.entries_nr * MIDX_CHUNK_OFFSET_WIDTH, | |
1071 | write_midx_object_offsets); | |
1072 | ||
1073 | if (ctx.large_offsets_needed) | |
1074 | add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, | |
1075 | (size_t)ctx.num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH, | |
1076 | write_midx_large_offsets); | |
1077 | ||
1078 | write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs); | |
1079 | write_chunkfile(cf, &ctx); | |
1080 | ||
1081 | finalize_hashfile(f, midx_hash, CSUM_FSYNC | CSUM_HASH_IN_STREAM); | |
1082 | free_chunkfile(cf); | |
1083 | ||
1084 | if (flags & MIDX_WRITE_REV_INDEX) | |
1085 | ctx.pack_order = midx_pack_order(&ctx); | |
1086 | ||
1087 | if (flags & MIDX_WRITE_REV_INDEX) | |
1088 | write_midx_reverse_index(midx_name, midx_hash, &ctx); | |
1089 | clear_midx_files_ext(object_dir, ".rev", midx_hash); | |
1090 | ||
1091 | commit_lock_file(&lk); | |
1092 | ||
1093 | cleanup: | |
1094 | for (i = 0; i < ctx.nr; i++) { | |
1095 | if (ctx.info[i].p) { | |
1096 | close_pack(ctx.info[i].p); | |
1097 | free(ctx.info[i].p); | |
1098 | } | |
1099 | free(ctx.info[i].pack_name); | |
1100 | } | |
1101 | ||
1102 | free(ctx.info); | |
1103 | free(ctx.entries); | |
1104 | free(ctx.pack_perm); | |
1105 | free(ctx.pack_order); | |
1106 | free(midx_name); | |
1107 | return result; | |
1108 | } | |
1109 | ||
1110 | int write_midx_file(const char *object_dir, | |
1111 | const char *preferred_pack_name, | |
1112 | unsigned flags) | |
1113 | { | |
1114 | return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name, | |
1115 | flags); | |
1116 | } | |
1117 | ||
1118 | struct clear_midx_data { | |
1119 | char *keep; | |
1120 | const char *ext; | |
1121 | }; | |
1122 | ||
1123 | static void clear_midx_file_ext(const char *full_path, size_t full_path_len, | |
1124 | const char *file_name, void *_data) | |
1125 | { | |
1126 | struct clear_midx_data *data = _data; | |
1127 | ||
1128 | if (!(starts_with(file_name, "multi-pack-index-") && | |
1129 | ends_with(file_name, data->ext))) | |
1130 | return; | |
1131 | if (data->keep && !strcmp(data->keep, file_name)) | |
1132 | return; | |
1133 | ||
1134 | if (unlink(full_path)) | |
1135 | die_errno(_("failed to remove %s"), full_path); | |
1136 | } | |
1137 | ||
1138 | static void clear_midx_files_ext(const char *object_dir, const char *ext, | |
1139 | unsigned char *keep_hash) | |
1140 | { | |
1141 | struct clear_midx_data data; | |
1142 | memset(&data, 0, sizeof(struct clear_midx_data)); | |
1143 | ||
1144 | if (keep_hash) | |
1145 | data.keep = xstrfmt("multi-pack-index-%s%s", | |
1146 | hash_to_hex(keep_hash), ext); | |
1147 | data.ext = ext; | |
1148 | ||
1149 | for_each_file_in_pack_dir(object_dir, | |
1150 | clear_midx_file_ext, | |
1151 | &data); | |
1152 | ||
1153 | free(data.keep); | |
1154 | } | |
1155 | ||
1156 | void clear_midx_file(struct repository *r) | |
1157 | { | |
1158 | char *midx = get_midx_filename(r->objects->odb->path); | |
1159 | ||
1160 | if (r->objects && r->objects->multi_pack_index) { | |
1161 | close_midx(r->objects->multi_pack_index); | |
1162 | r->objects->multi_pack_index = NULL; | |
1163 | } | |
1164 | ||
1165 | if (remove_path(midx)) | |
1166 | die(_("failed to clear multi-pack-index at %s"), midx); | |
1167 | ||
1168 | clear_midx_files_ext(r->objects->odb->path, ".rev", NULL); | |
1169 | ||
1170 | free(midx); | |
1171 | } | |
1172 | ||
1173 | static int verify_midx_error; | |
1174 | ||
1175 | __attribute__((format (printf, 1, 2))) | |
1176 | static void midx_report(const char *fmt, ...) | |
1177 | { | |
1178 | va_list ap; | |
1179 | verify_midx_error = 1; | |
1180 | va_start(ap, fmt); | |
1181 | vfprintf(stderr, fmt, ap); | |
1182 | fprintf(stderr, "\n"); | |
1183 | va_end(ap); | |
1184 | } | |
1185 | ||
1186 | struct pair_pos_vs_id | |
1187 | { | |
1188 | uint32_t pos; | |
1189 | uint32_t pack_int_id; | |
1190 | }; | |
1191 | ||
1192 | static int compare_pair_pos_vs_id(const void *_a, const void *_b) | |
1193 | { | |
1194 | struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a; | |
1195 | struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b; | |
1196 | ||
1197 | return b->pack_int_id - a->pack_int_id; | |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * Limit calls to display_progress() for performance reasons. | |
1202 | * The interval here was arbitrarily chosen. | |
1203 | */ | |
1204 | #define SPARSE_PROGRESS_INTERVAL (1 << 12) | |
1205 | #define midx_display_sparse_progress(progress, n) \ | |
1206 | do { \ | |
1207 | uint64_t _n = (n); \ | |
1208 | if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \ | |
1209 | display_progress(progress, _n); \ | |
1210 | } while (0) | |
1211 | ||
1212 | int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags) | |
1213 | { | |
1214 | struct pair_pos_vs_id *pairs = NULL; | |
1215 | uint32_t i; | |
1216 | struct progress *progress = NULL; | |
1217 | struct multi_pack_index *m = load_multi_pack_index(object_dir, 1); | |
1218 | verify_midx_error = 0; | |
1219 | ||
1220 | if (!m) { | |
1221 | int result = 0; | |
1222 | struct stat sb; | |
1223 | char *filename = get_midx_filename(object_dir); | |
1224 | if (!stat(filename, &sb)) { | |
1225 | error(_("multi-pack-index file exists, but failed to parse")); | |
1226 | result = 1; | |
1227 | } | |
1228 | free(filename); | |
1229 | return result; | |
1230 | } | |
1231 | ||
1232 | if (!midx_checksum_valid(m)) | |
1233 | midx_report(_("incorrect checksum")); | |
1234 | ||
1235 | if (flags & MIDX_PROGRESS) | |
1236 | progress = start_delayed_progress(_("Looking for referenced packfiles"), | |
1237 | m->num_packs); | |
1238 | for (i = 0; i < m->num_packs; i++) { | |
1239 | if (prepare_midx_pack(r, m, i)) | |
1240 | midx_report("failed to load pack in position %d", i); | |
1241 | ||
1242 | display_progress(progress, i + 1); | |
1243 | } | |
1244 | stop_progress(&progress); | |
1245 | ||
1246 | for (i = 0; i < 255; i++) { | |
1247 | uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]); | |
1248 | uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]); | |
1249 | ||
1250 | if (oid_fanout1 > oid_fanout2) | |
1251 | midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"), | |
1252 | i, oid_fanout1, oid_fanout2, i + 1); | |
1253 | } | |
1254 | ||
1255 | if (m->num_objects == 0) { | |
1256 | midx_report(_("the midx contains no oid")); | |
1257 | /* | |
1258 | * Remaining tests assume that we have objects, so we can | |
1259 | * return here. | |
1260 | */ | |
1261 | return verify_midx_error; | |
1262 | } | |
1263 | ||
1264 | if (flags & MIDX_PROGRESS) | |
1265 | progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"), | |
1266 | m->num_objects - 1); | |
1267 | for (i = 0; i < m->num_objects - 1; i++) { | |
1268 | struct object_id oid1, oid2; | |
1269 | ||
1270 | nth_midxed_object_oid(&oid1, m, i); | |
1271 | nth_midxed_object_oid(&oid2, m, i + 1); | |
1272 | ||
1273 | if (oidcmp(&oid1, &oid2) >= 0) | |
1274 | midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"), | |
1275 | i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1); | |
1276 | ||
1277 | midx_display_sparse_progress(progress, i + 1); | |
1278 | } | |
1279 | stop_progress(&progress); | |
1280 | ||
1281 | /* | |
1282 | * Create an array mapping each object to its packfile id. Sort it | |
1283 | * to group the objects by packfile. Use this permutation to visit | |
1284 | * each of the objects and only require 1 packfile to be open at a | |
1285 | * time. | |
1286 | */ | |
1287 | ALLOC_ARRAY(pairs, m->num_objects); | |
1288 | for (i = 0; i < m->num_objects; i++) { | |
1289 | pairs[i].pos = i; | |
1290 | pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i); | |
1291 | } | |
1292 | ||
1293 | if (flags & MIDX_PROGRESS) | |
1294 | progress = start_sparse_progress(_("Sorting objects by packfile"), | |
1295 | m->num_objects); | |
1296 | display_progress(progress, 0); /* TODO: Measure QSORT() progress */ | |
1297 | QSORT(pairs, m->num_objects, compare_pair_pos_vs_id); | |
1298 | stop_progress(&progress); | |
1299 | ||
1300 | if (flags & MIDX_PROGRESS) | |
1301 | progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects); | |
1302 | for (i = 0; i < m->num_objects; i++) { | |
1303 | struct object_id oid; | |
1304 | struct pack_entry e; | |
1305 | off_t m_offset, p_offset; | |
1306 | ||
1307 | if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id && | |
1308 | m->packs[pairs[i-1].pack_int_id]) | |
1309 | { | |
1310 | close_pack_fd(m->packs[pairs[i-1].pack_int_id]); | |
1311 | close_pack_index(m->packs[pairs[i-1].pack_int_id]); | |
1312 | } | |
1313 | ||
1314 | nth_midxed_object_oid(&oid, m, pairs[i].pos); | |
1315 | ||
1316 | if (!fill_midx_entry(r, &oid, &e, m)) { | |
1317 | midx_report(_("failed to load pack entry for oid[%d] = %s"), | |
1318 | pairs[i].pos, oid_to_hex(&oid)); | |
1319 | continue; | |
1320 | } | |
1321 | ||
1322 | if (open_pack_index(e.p)) { | |
1323 | midx_report(_("failed to load pack-index for packfile %s"), | |
1324 | e.p->pack_name); | |
1325 | break; | |
1326 | } | |
1327 | ||
1328 | m_offset = e.offset; | |
1329 | p_offset = find_pack_entry_one(oid.hash, e.p); | |
1330 | ||
1331 | if (m_offset != p_offset) | |
1332 | midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64), | |
1333 | pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset); | |
1334 | ||
1335 | midx_display_sparse_progress(progress, i + 1); | |
1336 | } | |
1337 | stop_progress(&progress); | |
1338 | ||
1339 | free(pairs); | |
1340 | ||
1341 | return verify_midx_error; | |
1342 | } | |
1343 | ||
1344 | int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags) | |
1345 | { | |
1346 | uint32_t i, *count, result = 0; | |
1347 | struct string_list packs_to_drop = STRING_LIST_INIT_DUP; | |
1348 | struct multi_pack_index *m = load_multi_pack_index(object_dir, 1); | |
1349 | struct progress *progress = NULL; | |
1350 | ||
1351 | if (!m) | |
1352 | return 0; | |
1353 | ||
1354 | CALLOC_ARRAY(count, m->num_packs); | |
1355 | ||
1356 | if (flags & MIDX_PROGRESS) | |
1357 | progress = start_delayed_progress(_("Counting referenced objects"), | |
1358 | m->num_objects); | |
1359 | for (i = 0; i < m->num_objects; i++) { | |
1360 | int pack_int_id = nth_midxed_pack_int_id(m, i); | |
1361 | count[pack_int_id]++; | |
1362 | display_progress(progress, i + 1); | |
1363 | } | |
1364 | stop_progress(&progress); | |
1365 | ||
1366 | if (flags & MIDX_PROGRESS) | |
1367 | progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"), | |
1368 | m->num_packs); | |
1369 | for (i = 0; i < m->num_packs; i++) { | |
1370 | char *pack_name; | |
1371 | display_progress(progress, i + 1); | |
1372 | ||
1373 | if (count[i]) | |
1374 | continue; | |
1375 | ||
1376 | if (prepare_midx_pack(r, m, i)) | |
1377 | continue; | |
1378 | ||
1379 | if (m->packs[i]->pack_keep) | |
1380 | continue; | |
1381 | ||
1382 | pack_name = xstrdup(m->packs[i]->pack_name); | |
1383 | close_pack(m->packs[i]); | |
1384 | ||
1385 | string_list_insert(&packs_to_drop, m->pack_names[i]); | |
1386 | unlink_pack_path(pack_name, 0); | |
1387 | free(pack_name); | |
1388 | } | |
1389 | stop_progress(&progress); | |
1390 | ||
1391 | free(count); | |
1392 | ||
1393 | if (packs_to_drop.nr) | |
1394 | result = write_midx_internal(object_dir, m, &packs_to_drop, NULL, flags); | |
1395 | ||
1396 | string_list_clear(&packs_to_drop, 0); | |
1397 | return result; | |
1398 | } | |
1399 | ||
1400 | struct repack_info { | |
1401 | timestamp_t mtime; | |
1402 | uint32_t referenced_objects; | |
1403 | uint32_t pack_int_id; | |
1404 | }; | |
1405 | ||
1406 | static int compare_by_mtime(const void *a_, const void *b_) | |
1407 | { | |
1408 | const struct repack_info *a, *b; | |
1409 | ||
1410 | a = (const struct repack_info *)a_; | |
1411 | b = (const struct repack_info *)b_; | |
1412 | ||
1413 | if (a->mtime < b->mtime) | |
1414 | return -1; | |
1415 | if (a->mtime > b->mtime) | |
1416 | return 1; | |
1417 | return 0; | |
1418 | } | |
1419 | ||
1420 | static int fill_included_packs_all(struct repository *r, | |
1421 | struct multi_pack_index *m, | |
1422 | unsigned char *include_pack) | |
1423 | { | |
1424 | uint32_t i, count = 0; | |
1425 | int pack_kept_objects = 0; | |
1426 | ||
1427 | repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects); | |
1428 | ||
1429 | for (i = 0; i < m->num_packs; i++) { | |
1430 | if (prepare_midx_pack(r, m, i)) | |
1431 | continue; | |
1432 | if (!pack_kept_objects && m->packs[i]->pack_keep) | |
1433 | continue; | |
1434 | ||
1435 | include_pack[i] = 1; | |
1436 | count++; | |
1437 | } | |
1438 | ||
1439 | return count < 2; | |
1440 | } | |
1441 | ||
1442 | static int fill_included_packs_batch(struct repository *r, | |
1443 | struct multi_pack_index *m, | |
1444 | unsigned char *include_pack, | |
1445 | size_t batch_size) | |
1446 | { | |
1447 | uint32_t i, packs_to_repack; | |
1448 | size_t total_size; | |
1449 | struct repack_info *pack_info = xcalloc(m->num_packs, sizeof(struct repack_info)); | |
1450 | int pack_kept_objects = 0; | |
1451 | ||
1452 | repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects); | |
1453 | ||
1454 | for (i = 0; i < m->num_packs; i++) { | |
1455 | pack_info[i].pack_int_id = i; | |
1456 | ||
1457 | if (prepare_midx_pack(r, m, i)) | |
1458 | continue; | |
1459 | ||
1460 | pack_info[i].mtime = m->packs[i]->mtime; | |
1461 | } | |
1462 | ||
1463 | for (i = 0; batch_size && i < m->num_objects; i++) { | |
1464 | uint32_t pack_int_id = nth_midxed_pack_int_id(m, i); | |
1465 | pack_info[pack_int_id].referenced_objects++; | |
1466 | } | |
1467 | ||
1468 | QSORT(pack_info, m->num_packs, compare_by_mtime); | |
1469 | ||
1470 | total_size = 0; | |
1471 | packs_to_repack = 0; | |
1472 | for (i = 0; total_size < batch_size && i < m->num_packs; i++) { | |
1473 | int pack_int_id = pack_info[i].pack_int_id; | |
1474 | struct packed_git *p = m->packs[pack_int_id]; | |
1475 | size_t expected_size; | |
1476 | ||
1477 | if (!p) | |
1478 | continue; | |
1479 | if (!pack_kept_objects && p->pack_keep) | |
1480 | continue; | |
1481 | if (open_pack_index(p) || !p->num_objects) | |
1482 | continue; | |
1483 | ||
1484 | expected_size = (size_t)(p->pack_size | |
1485 | * pack_info[i].referenced_objects); | |
1486 | expected_size /= p->num_objects; | |
1487 | ||
1488 | if (expected_size >= batch_size) | |
1489 | continue; | |
1490 | ||
1491 | packs_to_repack++; | |
1492 | total_size += expected_size; | |
1493 | include_pack[pack_int_id] = 1; | |
1494 | } | |
1495 | ||
1496 | free(pack_info); | |
1497 | ||
1498 | if (packs_to_repack < 2) | |
1499 | return 1; | |
1500 | ||
1501 | return 0; | |
1502 | } | |
1503 | ||
1504 | int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags) | |
1505 | { | |
1506 | int result = 0; | |
1507 | uint32_t i; | |
1508 | unsigned char *include_pack; | |
1509 | struct child_process cmd = CHILD_PROCESS_INIT; | |
1510 | FILE *cmd_in; | |
1511 | struct strbuf base_name = STRBUF_INIT; | |
1512 | struct multi_pack_index *m = load_multi_pack_index(object_dir, 1); | |
1513 | ||
1514 | /* | |
1515 | * When updating the default for these configuration | |
1516 | * variables in builtin/repack.c, these must be adjusted | |
1517 | * to match. | |
1518 | */ | |
1519 | int delta_base_offset = 1; | |
1520 | int use_delta_islands = 0; | |
1521 | ||
1522 | if (!m) | |
1523 | return 0; | |
1524 | ||
1525 | CALLOC_ARRAY(include_pack, m->num_packs); | |
1526 | ||
1527 | if (batch_size) { | |
1528 | if (fill_included_packs_batch(r, m, include_pack, batch_size)) | |
1529 | goto cleanup; | |
1530 | } else if (fill_included_packs_all(r, m, include_pack)) | |
1531 | goto cleanup; | |
1532 | ||
1533 | repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset); | |
1534 | repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands); | |
1535 | ||
1536 | strvec_push(&cmd.args, "pack-objects"); | |
1537 | ||
1538 | strbuf_addstr(&base_name, object_dir); | |
1539 | strbuf_addstr(&base_name, "/pack/pack"); | |
1540 | strvec_push(&cmd.args, base_name.buf); | |
1541 | ||
1542 | if (delta_base_offset) | |
1543 | strvec_push(&cmd.args, "--delta-base-offset"); | |
1544 | if (use_delta_islands) | |
1545 | strvec_push(&cmd.args, "--delta-islands"); | |
1546 | ||
1547 | if (flags & MIDX_PROGRESS) | |
1548 | strvec_push(&cmd.args, "--progress"); | |
1549 | else | |
1550 | strvec_push(&cmd.args, "-q"); | |
1551 | ||
1552 | strbuf_release(&base_name); | |
1553 | ||
1554 | cmd.git_cmd = 1; | |
1555 | cmd.in = cmd.out = -1; | |
1556 | ||
1557 | if (start_command(&cmd)) { | |
1558 | error(_("could not start pack-objects")); | |
1559 | result = 1; | |
1560 | goto cleanup; | |
1561 | } | |
1562 | ||
1563 | cmd_in = xfdopen(cmd.in, "w"); | |
1564 | ||
1565 | for (i = 0; i < m->num_objects; i++) { | |
1566 | struct object_id oid; | |
1567 | uint32_t pack_int_id = nth_midxed_pack_int_id(m, i); | |
1568 | ||
1569 | if (!include_pack[pack_int_id]) | |
1570 | continue; | |
1571 | ||
1572 | nth_midxed_object_oid(&oid, m, i); | |
1573 | fprintf(cmd_in, "%s\n", oid_to_hex(&oid)); | |
1574 | } | |
1575 | fclose(cmd_in); | |
1576 | ||
1577 | if (finish_command(&cmd)) { | |
1578 | error(_("could not finish pack-objects")); | |
1579 | result = 1; | |
1580 | goto cleanup; | |
1581 | } | |
1582 | ||
1583 | result = write_midx_internal(object_dir, m, NULL, NULL, flags); | |
1584 | m = NULL; | |
1585 | ||
1586 | cleanup: | |
1587 | if (m) | |
1588 | close_midx(m); | |
1589 | free(include_pack); | |
1590 | return result; | |
1591 | } |