]>
Commit | Line | Data |
---|---|---|
1 | #include "git-compat-util.h" | |
2 | #include "abspath.h" | |
3 | #include "alloc.h" | |
4 | #include "config.h" | |
5 | #include "csum-file.h" | |
6 | #include "dir.h" | |
7 | #include "gettext.h" | |
8 | #include "hex.h" | |
9 | #include "lockfile.h" | |
10 | #include "packfile.h" | |
11 | #include "object-file.h" | |
12 | #include "object-store.h" | |
13 | #include "hash-lookup.h" | |
14 | #include "midx.h" | |
15 | #include "progress.h" | |
16 | #include "trace2.h" | |
17 | #include "run-command.h" | |
18 | #include "repository.h" | |
19 | #include "chunk-format.h" | |
20 | #include "pack.h" | |
21 | #include "pack-bitmap.h" | |
22 | #include "refs.h" | |
23 | #include "revision.h" | |
24 | #include "list-objects.h" | |
25 | ||
26 | #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */ | |
27 | #define MIDX_VERSION 1 | |
28 | #define MIDX_BYTE_FILE_VERSION 4 | |
29 | #define MIDX_BYTE_HASH_VERSION 5 | |
30 | #define MIDX_BYTE_NUM_CHUNKS 6 | |
31 | #define MIDX_BYTE_NUM_PACKS 8 | |
32 | #define MIDX_HEADER_SIZE 12 | |
33 | #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz) | |
34 | ||
35 | #define MIDX_CHUNK_ALIGNMENT 4 | |
36 | #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */ | |
37 | #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */ | |
38 | #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */ | |
39 | #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */ | |
40 | #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */ | |
41 | #define MIDX_CHUNKID_REVINDEX 0x52494458 /* "RIDX" */ | |
42 | #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256) | |
43 | #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t)) | |
44 | #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t)) | |
45 | #define MIDX_LARGE_OFFSET_NEEDED 0x80000000 | |
46 | ||
47 | #define PACK_EXPIRED UINT_MAX | |
48 | ||
49 | const unsigned char *get_midx_checksum(struct multi_pack_index *m) | |
50 | { | |
51 | return m->data + m->data_len - the_hash_algo->rawsz; | |
52 | } | |
53 | ||
54 | void get_midx_filename(struct strbuf *out, const char *object_dir) | |
55 | { | |
56 | strbuf_addf(out, "%s/pack/multi-pack-index", object_dir); | |
57 | } | |
58 | ||
59 | void get_midx_rev_filename(struct strbuf *out, struct multi_pack_index *m) | |
60 | { | |
61 | get_midx_filename(out, m->object_dir); | |
62 | strbuf_addf(out, "-%s.rev", hash_to_hex(get_midx_checksum(m))); | |
63 | } | |
64 | ||
65 | static int midx_read_oid_fanout(const unsigned char *chunk_start, | |
66 | size_t chunk_size, void *data) | |
67 | { | |
68 | struct multi_pack_index *m = data; | |
69 | m->chunk_oid_fanout = (uint32_t *)chunk_start; | |
70 | ||
71 | if (chunk_size != 4 * 256) { | |
72 | error(_("multi-pack-index OID fanout is of the wrong size")); | |
73 | return 1; | |
74 | } | |
75 | return 0; | |
76 | } | |
77 | ||
78 | struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local) | |
79 | { | |
80 | struct multi_pack_index *m = NULL; | |
81 | int fd; | |
82 | struct stat st; | |
83 | size_t midx_size; | |
84 | void *midx_map = NULL; | |
85 | uint32_t hash_version; | |
86 | struct strbuf midx_name = STRBUF_INIT; | |
87 | uint32_t i; | |
88 | const char *cur_pack_name; | |
89 | struct chunkfile *cf = NULL; | |
90 | ||
91 | get_midx_filename(&midx_name, object_dir); | |
92 | ||
93 | fd = git_open(midx_name.buf); | |
94 | ||
95 | if (fd < 0) | |
96 | goto cleanup_fail; | |
97 | if (fstat(fd, &st)) { | |
98 | error_errno(_("failed to read %s"), midx_name.buf); | |
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.buf); | |
106 | goto cleanup_fail; | |
107 | } | |
108 | ||
109 | strbuf_release(&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(the_hash_algo)) { | |
131 | error(_("multi-pack-index hash version %u does not match version %u"), | |
132 | hash_version, oid_version(the_hash_algo)); | |
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 | if (git_env_bool("GIT_TEST_MIDX_READ_RIDX", 1)) | |
159 | pair_chunk(cf, MIDX_CHUNKID_REVINDEX, &m->chunk_revindex); | |
160 | ||
161 | m->num_objects = ntohl(m->chunk_oid_fanout[255]); | |
162 | ||
163 | CALLOC_ARRAY(m->pack_names, m->num_packs); | |
164 | CALLOC_ARRAY(m->packs, m->num_packs); | |
165 | ||
166 | cur_pack_name = (const char *)m->chunk_pack_names; | |
167 | for (i = 0; i < m->num_packs; i++) { | |
168 | m->pack_names[i] = cur_pack_name; | |
169 | ||
170 | cur_pack_name += strlen(cur_pack_name) + 1; | |
171 | ||
172 | if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0) | |
173 | die(_("multi-pack-index pack names out of order: '%s' before '%s'"), | |
174 | m->pack_names[i - 1], | |
175 | m->pack_names[i]); | |
176 | } | |
177 | ||
178 | trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs); | |
179 | trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects); | |
180 | ||
181 | free_chunkfile(cf); | |
182 | return m; | |
183 | ||
184 | cleanup_fail: | |
185 | free(m); | |
186 | strbuf_release(&midx_name); | |
187 | free_chunkfile(cf); | |
188 | if (midx_map) | |
189 | munmap(midx_map, midx_size); | |
190 | if (0 <= fd) | |
191 | close(fd); | |
192 | return NULL; | |
193 | } | |
194 | ||
195 | void close_midx(struct multi_pack_index *m) | |
196 | { | |
197 | uint32_t i; | |
198 | ||
199 | if (!m) | |
200 | return; | |
201 | ||
202 | close_midx(m->next); | |
203 | ||
204 | munmap((unsigned char *)m->data, m->data_len); | |
205 | ||
206 | for (i = 0; i < m->num_packs; i++) { | |
207 | if (m->packs[i]) | |
208 | m->packs[i]->multi_pack_index = 0; | |
209 | } | |
210 | FREE_AND_NULL(m->packs); | |
211 | FREE_AND_NULL(m->pack_names); | |
212 | free(m); | |
213 | } | |
214 | ||
215 | int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id) | |
216 | { | |
217 | struct strbuf pack_name = STRBUF_INIT; | |
218 | struct packed_git *p; | |
219 | ||
220 | if (pack_int_id >= m->num_packs) | |
221 | die(_("bad pack-int-id: %u (%u total packs)"), | |
222 | pack_int_id, m->num_packs); | |
223 | ||
224 | if (m->packs[pack_int_id]) | |
225 | return 0; | |
226 | ||
227 | strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir, | |
228 | m->pack_names[pack_int_id]); | |
229 | ||
230 | p = add_packed_git(pack_name.buf, pack_name.len, m->local); | |
231 | strbuf_release(&pack_name); | |
232 | ||
233 | if (!p) | |
234 | return 1; | |
235 | ||
236 | p->multi_pack_index = 1; | |
237 | m->packs[pack_int_id] = p; | |
238 | install_packed_git(r, p); | |
239 | list_add_tail(&p->mru, &r->objects->packed_git_mru); | |
240 | ||
241 | return 0; | |
242 | } | |
243 | ||
244 | int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result) | |
245 | { | |
246 | return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup, | |
247 | the_hash_algo->rawsz, result); | |
248 | } | |
249 | ||
250 | struct object_id *nth_midxed_object_oid(struct object_id *oid, | |
251 | struct multi_pack_index *m, | |
252 | uint32_t n) | |
253 | { | |
254 | if (n >= m->num_objects) | |
255 | return NULL; | |
256 | ||
257 | oidread(oid, m->chunk_oid_lookup + st_mult(m->hash_len, n)); | |
258 | return oid; | |
259 | } | |
260 | ||
261 | off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos) | |
262 | { | |
263 | const unsigned char *offset_data; | |
264 | uint32_t offset32; | |
265 | ||
266 | offset_data = m->chunk_object_offsets + (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH; | |
267 | offset32 = get_be32(offset_data + sizeof(uint32_t)); | |
268 | ||
269 | if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) { | |
270 | if (sizeof(off_t) < sizeof(uint64_t)) | |
271 | die(_("multi-pack-index stores a 64-bit offset, but off_t is too small")); | |
272 | ||
273 | offset32 ^= MIDX_LARGE_OFFSET_NEEDED; | |
274 | return get_be64(m->chunk_large_offsets + | |
275 | st_mult(sizeof(uint64_t), offset32)); | |
276 | } | |
277 | ||
278 | return offset32; | |
279 | } | |
280 | ||
281 | uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos) | |
282 | { | |
283 | return get_be32(m->chunk_object_offsets + | |
284 | (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH); | |
285 | } | |
286 | ||
287 | int fill_midx_entry(struct repository *r, | |
288 | const struct object_id *oid, | |
289 | struct pack_entry *e, | |
290 | struct multi_pack_index *m) | |
291 | { | |
292 | uint32_t pos; | |
293 | uint32_t pack_int_id; | |
294 | struct packed_git *p; | |
295 | ||
296 | if (!bsearch_midx(oid, m, &pos)) | |
297 | return 0; | |
298 | ||
299 | if (pos >= m->num_objects) | |
300 | return 0; | |
301 | ||
302 | pack_int_id = nth_midxed_pack_int_id(m, pos); | |
303 | ||
304 | if (prepare_midx_pack(r, m, pack_int_id)) | |
305 | return 0; | |
306 | p = m->packs[pack_int_id]; | |
307 | ||
308 | /* | |
309 | * We are about to tell the caller where they can locate the | |
310 | * requested object. We better make sure the packfile is | |
311 | * still here and can be accessed before supplying that | |
312 | * answer, as it may have been deleted since the MIDX was | |
313 | * loaded! | |
314 | */ | |
315 | if (!is_pack_valid(p)) | |
316 | return 0; | |
317 | ||
318 | if (oidset_size(&p->bad_objects) && | |
319 | oidset_contains(&p->bad_objects, oid)) | |
320 | return 0; | |
321 | ||
322 | e->offset = nth_midxed_offset(m, pos); | |
323 | e->p = p; | |
324 | ||
325 | return 1; | |
326 | } | |
327 | ||
328 | /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */ | |
329 | static int cmp_idx_or_pack_name(const char *idx_or_pack_name, | |
330 | const char *idx_name) | |
331 | { | |
332 | /* Skip past any initial matching prefix. */ | |
333 | while (*idx_name && *idx_name == *idx_or_pack_name) { | |
334 | idx_name++; | |
335 | idx_or_pack_name++; | |
336 | } | |
337 | ||
338 | /* | |
339 | * If we didn't match completely, we may have matched "pack-1234." and | |
340 | * be left with "idx" and "pack" respectively, which is also OK. We do | |
341 | * not have to check for "idx" and "idx", because that would have been | |
342 | * a complete match (and in that case these strcmps will be false, but | |
343 | * we'll correctly return 0 from the final strcmp() below. | |
344 | * | |
345 | * Technically this matches "fooidx" and "foopack", but we'd never have | |
346 | * such names in the first place. | |
347 | */ | |
348 | if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack")) | |
349 | return 0; | |
350 | ||
351 | /* | |
352 | * This not only checks for a complete match, but also orders based on | |
353 | * the first non-identical character, which means our ordering will | |
354 | * match a raw strcmp(). That makes it OK to use this to binary search | |
355 | * a naively-sorted list. | |
356 | */ | |
357 | return strcmp(idx_or_pack_name, idx_name); | |
358 | } | |
359 | ||
360 | int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name) | |
361 | { | |
362 | uint32_t first = 0, last = m->num_packs; | |
363 | ||
364 | while (first < last) { | |
365 | uint32_t mid = first + (last - first) / 2; | |
366 | const char *current; | |
367 | int cmp; | |
368 | ||
369 | current = m->pack_names[mid]; | |
370 | cmp = cmp_idx_or_pack_name(idx_or_pack_name, current); | |
371 | if (!cmp) | |
372 | return 1; | |
373 | if (cmp > 0) { | |
374 | first = mid + 1; | |
375 | continue; | |
376 | } | |
377 | last = mid; | |
378 | } | |
379 | ||
380 | return 0; | |
381 | } | |
382 | ||
383 | int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local) | |
384 | { | |
385 | struct multi_pack_index *m; | |
386 | struct multi_pack_index *m_search; | |
387 | ||
388 | prepare_repo_settings(r); | |
389 | if (!r->settings.core_multi_pack_index) | |
390 | return 0; | |
391 | ||
392 | for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next) | |
393 | if (!strcmp(object_dir, m_search->object_dir)) | |
394 | return 1; | |
395 | ||
396 | m = load_multi_pack_index(object_dir, local); | |
397 | ||
398 | if (m) { | |
399 | struct multi_pack_index *mp = r->objects->multi_pack_index; | |
400 | if (mp) { | |
401 | m->next = mp->next; | |
402 | mp->next = m; | |
403 | } else | |
404 | r->objects->multi_pack_index = m; | |
405 | return 1; | |
406 | } | |
407 | ||
408 | return 0; | |
409 | } | |
410 | ||
411 | static size_t write_midx_header(struct hashfile *f, | |
412 | unsigned char num_chunks, | |
413 | uint32_t num_packs) | |
414 | { | |
415 | hashwrite_be32(f, MIDX_SIGNATURE); | |
416 | hashwrite_u8(f, MIDX_VERSION); | |
417 | hashwrite_u8(f, oid_version(the_hash_algo)); | |
418 | hashwrite_u8(f, num_chunks); | |
419 | hashwrite_u8(f, 0); /* unused */ | |
420 | hashwrite_be32(f, num_packs); | |
421 | ||
422 | return MIDX_HEADER_SIZE; | |
423 | } | |
424 | ||
425 | struct pack_info { | |
426 | uint32_t orig_pack_int_id; | |
427 | char *pack_name; | |
428 | struct packed_git *p; | |
429 | unsigned expired : 1; | |
430 | }; | |
431 | ||
432 | static int pack_info_compare(const void *_a, const void *_b) | |
433 | { | |
434 | struct pack_info *a = (struct pack_info *)_a; | |
435 | struct pack_info *b = (struct pack_info *)_b; | |
436 | return strcmp(a->pack_name, b->pack_name); | |
437 | } | |
438 | ||
439 | static int idx_or_pack_name_cmp(const void *_va, const void *_vb) | |
440 | { | |
441 | const char *pack_name = _va; | |
442 | const struct pack_info *compar = _vb; | |
443 | ||
444 | return cmp_idx_or_pack_name(pack_name, compar->pack_name); | |
445 | } | |
446 | ||
447 | struct write_midx_context { | |
448 | struct pack_info *info; | |
449 | size_t nr; | |
450 | size_t alloc; | |
451 | struct multi_pack_index *m; | |
452 | struct progress *progress; | |
453 | unsigned pack_paths_checked; | |
454 | ||
455 | struct pack_midx_entry *entries; | |
456 | size_t entries_nr; | |
457 | ||
458 | uint32_t *pack_perm; | |
459 | uint32_t *pack_order; | |
460 | unsigned large_offsets_needed:1; | |
461 | uint32_t num_large_offsets; | |
462 | ||
463 | int preferred_pack_idx; | |
464 | ||
465 | struct string_list *to_include; | |
466 | }; | |
467 | ||
468 | static void add_pack_to_midx(const char *full_path, size_t full_path_len, | |
469 | const char *file_name, void *data) | |
470 | { | |
471 | struct write_midx_context *ctx = data; | |
472 | ||
473 | if (ends_with(file_name, ".idx")) { | |
474 | display_progress(ctx->progress, ++ctx->pack_paths_checked); | |
475 | /* | |
476 | * Note that at most one of ctx->m and ctx->to_include are set, | |
477 | * so we are testing midx_contains_pack() and | |
478 | * string_list_has_string() independently (guarded by the | |
479 | * appropriate NULL checks). | |
480 | * | |
481 | * We could support passing to_include while reusing an existing | |
482 | * MIDX, but don't currently since the reuse process drags | |
483 | * forward all packs from an existing MIDX (without checking | |
484 | * whether or not they appear in the to_include list). | |
485 | * | |
486 | * If we added support for that, these next two conditional | |
487 | * should be performed independently (likely checking | |
488 | * to_include before the existing MIDX). | |
489 | */ | |
490 | if (ctx->m && midx_contains_pack(ctx->m, file_name)) | |
491 | return; | |
492 | else if (ctx->to_include && | |
493 | !string_list_has_string(ctx->to_include, file_name)) | |
494 | return; | |
495 | ||
496 | ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc); | |
497 | ||
498 | ctx->info[ctx->nr].p = add_packed_git(full_path, | |
499 | full_path_len, | |
500 | 0); | |
501 | ||
502 | if (!ctx->info[ctx->nr].p) { | |
503 | warning(_("failed to add packfile '%s'"), | |
504 | full_path); | |
505 | return; | |
506 | } | |
507 | ||
508 | if (open_pack_index(ctx->info[ctx->nr].p)) { | |
509 | warning(_("failed to open pack-index '%s'"), | |
510 | full_path); | |
511 | close_pack(ctx->info[ctx->nr].p); | |
512 | FREE_AND_NULL(ctx->info[ctx->nr].p); | |
513 | return; | |
514 | } | |
515 | ||
516 | ctx->info[ctx->nr].pack_name = xstrdup(file_name); | |
517 | ctx->info[ctx->nr].orig_pack_int_id = ctx->nr; | |
518 | ctx->info[ctx->nr].expired = 0; | |
519 | ctx->nr++; | |
520 | } | |
521 | } | |
522 | ||
523 | struct pack_midx_entry { | |
524 | struct object_id oid; | |
525 | uint32_t pack_int_id; | |
526 | time_t pack_mtime; | |
527 | uint64_t offset; | |
528 | unsigned preferred : 1; | |
529 | }; | |
530 | ||
531 | static int midx_oid_compare(const void *_a, const void *_b) | |
532 | { | |
533 | const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a; | |
534 | const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b; | |
535 | int cmp = oidcmp(&a->oid, &b->oid); | |
536 | ||
537 | if (cmp) | |
538 | return cmp; | |
539 | ||
540 | /* Sort objects in a preferred pack first when multiple copies exist. */ | |
541 | if (a->preferred > b->preferred) | |
542 | return -1; | |
543 | if (a->preferred < b->preferred) | |
544 | return 1; | |
545 | ||
546 | if (a->pack_mtime > b->pack_mtime) | |
547 | return -1; | |
548 | else if (a->pack_mtime < b->pack_mtime) | |
549 | return 1; | |
550 | ||
551 | return a->pack_int_id - b->pack_int_id; | |
552 | } | |
553 | ||
554 | static int nth_midxed_pack_midx_entry(struct multi_pack_index *m, | |
555 | struct pack_midx_entry *e, | |
556 | uint32_t pos) | |
557 | { | |
558 | if (pos >= m->num_objects) | |
559 | return 1; | |
560 | ||
561 | nth_midxed_object_oid(&e->oid, m, pos); | |
562 | e->pack_int_id = nth_midxed_pack_int_id(m, pos); | |
563 | e->offset = nth_midxed_offset(m, pos); | |
564 | ||
565 | /* consider objects in midx to be from "old" packs */ | |
566 | e->pack_mtime = 0; | |
567 | return 0; | |
568 | } | |
569 | ||
570 | static void fill_pack_entry(uint32_t pack_int_id, | |
571 | struct packed_git *p, | |
572 | uint32_t cur_object, | |
573 | struct pack_midx_entry *entry, | |
574 | int preferred) | |
575 | { | |
576 | if (nth_packed_object_id(&entry->oid, p, cur_object) < 0) | |
577 | die(_("failed to locate object %d in packfile"), cur_object); | |
578 | ||
579 | entry->pack_int_id = pack_int_id; | |
580 | entry->pack_mtime = p->mtime; | |
581 | ||
582 | entry->offset = nth_packed_object_offset(p, cur_object); | |
583 | entry->preferred = !!preferred; | |
584 | } | |
585 | ||
586 | struct midx_fanout { | |
587 | struct pack_midx_entry *entries; | |
588 | size_t nr, alloc; | |
589 | }; | |
590 | ||
591 | static void midx_fanout_grow(struct midx_fanout *fanout, size_t nr) | |
592 | { | |
593 | if (nr < fanout->nr) | |
594 | BUG("negative growth in midx_fanout_grow() (%"PRIuMAX" < %"PRIuMAX")", | |
595 | (uintmax_t)nr, (uintmax_t)fanout->nr); | |
596 | ALLOC_GROW(fanout->entries, nr, fanout->alloc); | |
597 | } | |
598 | ||
599 | static void midx_fanout_sort(struct midx_fanout *fanout) | |
600 | { | |
601 | QSORT(fanout->entries, fanout->nr, midx_oid_compare); | |
602 | } | |
603 | ||
604 | static void midx_fanout_add_midx_fanout(struct midx_fanout *fanout, | |
605 | struct multi_pack_index *m, | |
606 | uint32_t cur_fanout, | |
607 | int preferred_pack) | |
608 | { | |
609 | uint32_t start = 0, end; | |
610 | uint32_t cur_object; | |
611 | ||
612 | if (cur_fanout) | |
613 | start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]); | |
614 | end = ntohl(m->chunk_oid_fanout[cur_fanout]); | |
615 | ||
616 | for (cur_object = start; cur_object < end; cur_object++) { | |
617 | if ((preferred_pack > -1) && | |
618 | (preferred_pack == nth_midxed_pack_int_id(m, cur_object))) { | |
619 | /* | |
620 | * Objects from preferred packs are added | |
621 | * separately. | |
622 | */ | |
623 | continue; | |
624 | } | |
625 | ||
626 | midx_fanout_grow(fanout, fanout->nr + 1); | |
627 | nth_midxed_pack_midx_entry(m, | |
628 | &fanout->entries[fanout->nr], | |
629 | cur_object); | |
630 | fanout->entries[fanout->nr].preferred = 0; | |
631 | fanout->nr++; | |
632 | } | |
633 | } | |
634 | ||
635 | static void midx_fanout_add_pack_fanout(struct midx_fanout *fanout, | |
636 | struct pack_info *info, | |
637 | uint32_t cur_pack, | |
638 | int preferred, | |
639 | uint32_t cur_fanout) | |
640 | { | |
641 | struct packed_git *pack = info[cur_pack].p; | |
642 | uint32_t start = 0, end; | |
643 | uint32_t cur_object; | |
644 | ||
645 | if (cur_fanout) | |
646 | start = get_pack_fanout(pack, cur_fanout - 1); | |
647 | end = get_pack_fanout(pack, cur_fanout); | |
648 | ||
649 | for (cur_object = start; cur_object < end; cur_object++) { | |
650 | midx_fanout_grow(fanout, fanout->nr + 1); | |
651 | fill_pack_entry(cur_pack, | |
652 | info[cur_pack].p, | |
653 | cur_object, | |
654 | &fanout->entries[fanout->nr], | |
655 | preferred); | |
656 | fanout->nr++; | |
657 | } | |
658 | } | |
659 | ||
660 | /* | |
661 | * It is possible to artificially get into a state where there are many | |
662 | * duplicate copies of objects. That can create high memory pressure if | |
663 | * we are to create a list of all objects before de-duplication. To reduce | |
664 | * this memory pressure without a significant performance drop, automatically | |
665 | * group objects by the first byte of their object id. Use the IDX fanout | |
666 | * tables to group the data, copy to a local array, then sort. | |
667 | * | |
668 | * Copy only the de-duplicated entries (selected by most-recent modified time | |
669 | * of a packfile containing the object). | |
670 | */ | |
671 | static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m, | |
672 | struct pack_info *info, | |
673 | uint32_t nr_packs, | |
674 | size_t *nr_objects, | |
675 | int preferred_pack) | |
676 | { | |
677 | uint32_t cur_fanout, cur_pack, cur_object; | |
678 | size_t alloc_objects, total_objects = 0; | |
679 | struct midx_fanout fanout = { 0 }; | |
680 | struct pack_midx_entry *deduplicated_entries = NULL; | |
681 | uint32_t start_pack = m ? m->num_packs : 0; | |
682 | ||
683 | for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) | |
684 | total_objects = st_add(total_objects, | |
685 | info[cur_pack].p->num_objects); | |
686 | ||
687 | /* | |
688 | * As we de-duplicate by fanout value, we expect the fanout | |
689 | * slices to be evenly distributed, with some noise. Hence, | |
690 | * allocate slightly more than one 256th. | |
691 | */ | |
692 | alloc_objects = fanout.alloc = total_objects > 3200 ? total_objects / 200 : 16; | |
693 | ||
694 | ALLOC_ARRAY(fanout.entries, fanout.alloc); | |
695 | ALLOC_ARRAY(deduplicated_entries, alloc_objects); | |
696 | *nr_objects = 0; | |
697 | ||
698 | for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) { | |
699 | fanout.nr = 0; | |
700 | ||
701 | if (m) | |
702 | midx_fanout_add_midx_fanout(&fanout, m, cur_fanout, | |
703 | preferred_pack); | |
704 | ||
705 | for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) { | |
706 | int preferred = cur_pack == preferred_pack; | |
707 | midx_fanout_add_pack_fanout(&fanout, | |
708 | info, cur_pack, | |
709 | preferred, cur_fanout); | |
710 | } | |
711 | ||
712 | if (-1 < preferred_pack && preferred_pack < start_pack) | |
713 | midx_fanout_add_pack_fanout(&fanout, info, | |
714 | preferred_pack, 1, | |
715 | cur_fanout); | |
716 | ||
717 | midx_fanout_sort(&fanout); | |
718 | ||
719 | /* | |
720 | * The batch is now sorted by OID and then mtime (descending). | |
721 | * Take only the first duplicate. | |
722 | */ | |
723 | for (cur_object = 0; cur_object < fanout.nr; cur_object++) { | |
724 | if (cur_object && oideq(&fanout.entries[cur_object - 1].oid, | |
725 | &fanout.entries[cur_object].oid)) | |
726 | continue; | |
727 | ||
728 | ALLOC_GROW(deduplicated_entries, st_add(*nr_objects, 1), | |
729 | alloc_objects); | |
730 | memcpy(&deduplicated_entries[*nr_objects], | |
731 | &fanout.entries[cur_object], | |
732 | sizeof(struct pack_midx_entry)); | |
733 | (*nr_objects)++; | |
734 | } | |
735 | } | |
736 | ||
737 | free(fanout.entries); | |
738 | return deduplicated_entries; | |
739 | } | |
740 | ||
741 | static int write_midx_pack_names(struct hashfile *f, void *data) | |
742 | { | |
743 | struct write_midx_context *ctx = data; | |
744 | uint32_t i; | |
745 | unsigned char padding[MIDX_CHUNK_ALIGNMENT]; | |
746 | size_t written = 0; | |
747 | ||
748 | for (i = 0; i < ctx->nr; i++) { | |
749 | size_t writelen; | |
750 | ||
751 | if (ctx->info[i].expired) | |
752 | continue; | |
753 | ||
754 | if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0) | |
755 | BUG("incorrect pack-file order: %s before %s", | |
756 | ctx->info[i - 1].pack_name, | |
757 | ctx->info[i].pack_name); | |
758 | ||
759 | writelen = strlen(ctx->info[i].pack_name) + 1; | |
760 | hashwrite(f, ctx->info[i].pack_name, writelen); | |
761 | written += writelen; | |
762 | } | |
763 | ||
764 | /* add padding to be aligned */ | |
765 | i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT); | |
766 | if (i < MIDX_CHUNK_ALIGNMENT) { | |
767 | memset(padding, 0, sizeof(padding)); | |
768 | hashwrite(f, padding, i); | |
769 | } | |
770 | ||
771 | return 0; | |
772 | } | |
773 | ||
774 | static int write_midx_oid_fanout(struct hashfile *f, | |
775 | void *data) | |
776 | { | |
777 | struct write_midx_context *ctx = data; | |
778 | struct pack_midx_entry *list = ctx->entries; | |
779 | struct pack_midx_entry *last = ctx->entries + ctx->entries_nr; | |
780 | uint32_t count = 0; | |
781 | uint32_t i; | |
782 | ||
783 | /* | |
784 | * Write the first-level table (the list is sorted, | |
785 | * but we use a 256-entry lookup to be able to avoid | |
786 | * having to do eight extra binary search iterations). | |
787 | */ | |
788 | for (i = 0; i < 256; i++) { | |
789 | struct pack_midx_entry *next = list; | |
790 | ||
791 | while (next < last && next->oid.hash[0] == i) { | |
792 | count++; | |
793 | next++; | |
794 | } | |
795 | ||
796 | hashwrite_be32(f, count); | |
797 | list = next; | |
798 | } | |
799 | ||
800 | return 0; | |
801 | } | |
802 | ||
803 | static int write_midx_oid_lookup(struct hashfile *f, | |
804 | void *data) | |
805 | { | |
806 | struct write_midx_context *ctx = data; | |
807 | unsigned char hash_len = the_hash_algo->rawsz; | |
808 | struct pack_midx_entry *list = ctx->entries; | |
809 | uint32_t i; | |
810 | ||
811 | for (i = 0; i < ctx->entries_nr; i++) { | |
812 | struct pack_midx_entry *obj = list++; | |
813 | ||
814 | if (i < ctx->entries_nr - 1) { | |
815 | struct pack_midx_entry *next = list; | |
816 | if (oidcmp(&obj->oid, &next->oid) >= 0) | |
817 | BUG("OIDs not in order: %s >= %s", | |
818 | oid_to_hex(&obj->oid), | |
819 | oid_to_hex(&next->oid)); | |
820 | } | |
821 | ||
822 | hashwrite(f, obj->oid.hash, (int)hash_len); | |
823 | } | |
824 | ||
825 | return 0; | |
826 | } | |
827 | ||
828 | static int write_midx_object_offsets(struct hashfile *f, | |
829 | void *data) | |
830 | { | |
831 | struct write_midx_context *ctx = data; | |
832 | struct pack_midx_entry *list = ctx->entries; | |
833 | uint32_t i, nr_large_offset = 0; | |
834 | ||
835 | for (i = 0; i < ctx->entries_nr; i++) { | |
836 | struct pack_midx_entry *obj = list++; | |
837 | ||
838 | if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED) | |
839 | BUG("object %s is in an expired pack with int-id %d", | |
840 | oid_to_hex(&obj->oid), | |
841 | obj->pack_int_id); | |
842 | ||
843 | hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]); | |
844 | ||
845 | if (ctx->large_offsets_needed && obj->offset >> 31) | |
846 | hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++); | |
847 | else if (!ctx->large_offsets_needed && obj->offset >> 32) | |
848 | BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!", | |
849 | oid_to_hex(&obj->oid), | |
850 | obj->offset); | |
851 | else | |
852 | hashwrite_be32(f, (uint32_t)obj->offset); | |
853 | } | |
854 | ||
855 | return 0; | |
856 | } | |
857 | ||
858 | static int write_midx_large_offsets(struct hashfile *f, | |
859 | void *data) | |
860 | { | |
861 | struct write_midx_context *ctx = data; | |
862 | struct pack_midx_entry *list = ctx->entries; | |
863 | struct pack_midx_entry *end = ctx->entries + ctx->entries_nr; | |
864 | uint32_t nr_large_offset = ctx->num_large_offsets; | |
865 | ||
866 | while (nr_large_offset) { | |
867 | struct pack_midx_entry *obj; | |
868 | uint64_t offset; | |
869 | ||
870 | if (list >= end) | |
871 | BUG("too many large-offset objects"); | |
872 | ||
873 | obj = list++; | |
874 | offset = obj->offset; | |
875 | ||
876 | if (!(offset >> 31)) | |
877 | continue; | |
878 | ||
879 | hashwrite_be64(f, offset); | |
880 | ||
881 | nr_large_offset--; | |
882 | } | |
883 | ||
884 | return 0; | |
885 | } | |
886 | ||
887 | static int write_midx_revindex(struct hashfile *f, | |
888 | void *data) | |
889 | { | |
890 | struct write_midx_context *ctx = data; | |
891 | uint32_t i; | |
892 | ||
893 | for (i = 0; i < ctx->entries_nr; i++) | |
894 | hashwrite_be32(f, ctx->pack_order[i]); | |
895 | ||
896 | return 0; | |
897 | } | |
898 | ||
899 | struct midx_pack_order_data { | |
900 | uint32_t nr; | |
901 | uint32_t pack; | |
902 | off_t offset; | |
903 | }; | |
904 | ||
905 | static int midx_pack_order_cmp(const void *va, const void *vb) | |
906 | { | |
907 | const struct midx_pack_order_data *a = va, *b = vb; | |
908 | if (a->pack < b->pack) | |
909 | return -1; | |
910 | else if (a->pack > b->pack) | |
911 | return 1; | |
912 | else if (a->offset < b->offset) | |
913 | return -1; | |
914 | else if (a->offset > b->offset) | |
915 | return 1; | |
916 | else | |
917 | return 0; | |
918 | } | |
919 | ||
920 | static uint32_t *midx_pack_order(struct write_midx_context *ctx) | |
921 | { | |
922 | struct midx_pack_order_data *data; | |
923 | uint32_t *pack_order; | |
924 | uint32_t i; | |
925 | ||
926 | trace2_region_enter("midx", "midx_pack_order", the_repository); | |
927 | ||
928 | ALLOC_ARRAY(data, ctx->entries_nr); | |
929 | for (i = 0; i < ctx->entries_nr; i++) { | |
930 | struct pack_midx_entry *e = &ctx->entries[i]; | |
931 | data[i].nr = i; | |
932 | data[i].pack = ctx->pack_perm[e->pack_int_id]; | |
933 | if (!e->preferred) | |
934 | data[i].pack |= (1U << 31); | |
935 | data[i].offset = e->offset; | |
936 | } | |
937 | ||
938 | QSORT(data, ctx->entries_nr, midx_pack_order_cmp); | |
939 | ||
940 | ALLOC_ARRAY(pack_order, ctx->entries_nr); | |
941 | for (i = 0; i < ctx->entries_nr; i++) | |
942 | pack_order[i] = data[i].nr; | |
943 | free(data); | |
944 | ||
945 | trace2_region_leave("midx", "midx_pack_order", the_repository); | |
946 | ||
947 | return pack_order; | |
948 | } | |
949 | ||
950 | static void write_midx_reverse_index(char *midx_name, unsigned char *midx_hash, | |
951 | struct write_midx_context *ctx) | |
952 | { | |
953 | struct strbuf buf = STRBUF_INIT; | |
954 | const char *tmp_file; | |
955 | ||
956 | trace2_region_enter("midx", "write_midx_reverse_index", the_repository); | |
957 | ||
958 | strbuf_addf(&buf, "%s-%s.rev", midx_name, hash_to_hex(midx_hash)); | |
959 | ||
960 | tmp_file = write_rev_file_order(NULL, ctx->pack_order, ctx->entries_nr, | |
961 | midx_hash, WRITE_REV); | |
962 | ||
963 | if (finalize_object_file(tmp_file, buf.buf)) | |
964 | die(_("cannot store reverse index file")); | |
965 | ||
966 | strbuf_release(&buf); | |
967 | ||
968 | trace2_region_leave("midx", "write_midx_reverse_index", the_repository); | |
969 | } | |
970 | ||
971 | static void clear_midx_files_ext(const char *object_dir, const char *ext, | |
972 | unsigned char *keep_hash); | |
973 | ||
974 | static int midx_checksum_valid(struct multi_pack_index *m) | |
975 | { | |
976 | return hashfile_checksum_valid(m->data, m->data_len); | |
977 | } | |
978 | ||
979 | static void prepare_midx_packing_data(struct packing_data *pdata, | |
980 | struct write_midx_context *ctx) | |
981 | { | |
982 | uint32_t i; | |
983 | ||
984 | trace2_region_enter("midx", "prepare_midx_packing_data", the_repository); | |
985 | ||
986 | memset(pdata, 0, sizeof(struct packing_data)); | |
987 | prepare_packing_data(the_repository, pdata); | |
988 | ||
989 | for (i = 0; i < ctx->entries_nr; i++) { | |
990 | struct pack_midx_entry *from = &ctx->entries[ctx->pack_order[i]]; | |
991 | struct object_entry *to = packlist_alloc(pdata, &from->oid); | |
992 | ||
993 | oe_set_in_pack(pdata, to, | |
994 | ctx->info[ctx->pack_perm[from->pack_int_id]].p); | |
995 | } | |
996 | ||
997 | trace2_region_leave("midx", "prepare_midx_packing_data", the_repository); | |
998 | } | |
999 | ||
1000 | static int add_ref_to_pending(const char *refname, | |
1001 | const struct object_id *oid, | |
1002 | int flag, void *cb_data) | |
1003 | { | |
1004 | struct rev_info *revs = (struct rev_info*)cb_data; | |
1005 | struct object_id peeled; | |
1006 | struct object *object; | |
1007 | ||
1008 | if ((flag & REF_ISSYMREF) && (flag & REF_ISBROKEN)) { | |
1009 | warning("symbolic ref is dangling: %s", refname); | |
1010 | return 0; | |
1011 | } | |
1012 | ||
1013 | if (!peel_iterated_oid(oid, &peeled)) | |
1014 | oid = &peeled; | |
1015 | ||
1016 | object = parse_object_or_die(oid, refname); | |
1017 | if (object->type != OBJ_COMMIT) | |
1018 | return 0; | |
1019 | ||
1020 | add_pending_object(revs, object, ""); | |
1021 | if (bitmap_is_preferred_refname(revs->repo, refname)) | |
1022 | object->flags |= NEEDS_BITMAP; | |
1023 | return 0; | |
1024 | } | |
1025 | ||
1026 | struct bitmap_commit_cb { | |
1027 | struct commit **commits; | |
1028 | size_t commits_nr, commits_alloc; | |
1029 | ||
1030 | struct write_midx_context *ctx; | |
1031 | }; | |
1032 | ||
1033 | static const struct object_id *bitmap_oid_access(size_t index, | |
1034 | const void *_entries) | |
1035 | { | |
1036 | const struct pack_midx_entry *entries = _entries; | |
1037 | return &entries[index].oid; | |
1038 | } | |
1039 | ||
1040 | static void bitmap_show_commit(struct commit *commit, void *_data) | |
1041 | { | |
1042 | struct bitmap_commit_cb *data = _data; | |
1043 | int pos = oid_pos(&commit->object.oid, data->ctx->entries, | |
1044 | data->ctx->entries_nr, | |
1045 | bitmap_oid_access); | |
1046 | if (pos < 0) | |
1047 | return; | |
1048 | ||
1049 | ALLOC_GROW(data->commits, data->commits_nr + 1, data->commits_alloc); | |
1050 | data->commits[data->commits_nr++] = commit; | |
1051 | } | |
1052 | ||
1053 | static int read_refs_snapshot(const char *refs_snapshot, | |
1054 | struct rev_info *revs) | |
1055 | { | |
1056 | struct strbuf buf = STRBUF_INIT; | |
1057 | struct object_id oid; | |
1058 | FILE *f = xfopen(refs_snapshot, "r"); | |
1059 | ||
1060 | while (strbuf_getline(&buf, f) != EOF) { | |
1061 | struct object *object; | |
1062 | int preferred = 0; | |
1063 | char *hex = buf.buf; | |
1064 | const char *end = NULL; | |
1065 | ||
1066 | if (buf.len && *buf.buf == '+') { | |
1067 | preferred = 1; | |
1068 | hex = &buf.buf[1]; | |
1069 | } | |
1070 | ||
1071 | if (parse_oid_hex(hex, &oid, &end) < 0) | |
1072 | die(_("could not parse line: %s"), buf.buf); | |
1073 | if (*end) | |
1074 | die(_("malformed line: %s"), buf.buf); | |
1075 | ||
1076 | object = parse_object_or_die(&oid, NULL); | |
1077 | if (preferred) | |
1078 | object->flags |= NEEDS_BITMAP; | |
1079 | ||
1080 | add_pending_object(revs, object, ""); | |
1081 | } | |
1082 | ||
1083 | fclose(f); | |
1084 | strbuf_release(&buf); | |
1085 | return 0; | |
1086 | } | |
1087 | ||
1088 | static struct commit **find_commits_for_midx_bitmap(uint32_t *indexed_commits_nr_p, | |
1089 | const char *refs_snapshot, | |
1090 | struct write_midx_context *ctx) | |
1091 | { | |
1092 | struct rev_info revs; | |
1093 | struct bitmap_commit_cb cb = {0}; | |
1094 | ||
1095 | trace2_region_enter("midx", "find_commits_for_midx_bitmap", | |
1096 | the_repository); | |
1097 | ||
1098 | cb.ctx = ctx; | |
1099 | ||
1100 | repo_init_revisions(the_repository, &revs, NULL); | |
1101 | if (refs_snapshot) { | |
1102 | read_refs_snapshot(refs_snapshot, &revs); | |
1103 | } else { | |
1104 | setup_revisions(0, NULL, &revs, NULL); | |
1105 | for_each_ref(add_ref_to_pending, &revs); | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * Skipping promisor objects here is intentional, since it only excludes | |
1110 | * them from the list of reachable commits that we want to select from | |
1111 | * when computing the selection of MIDX'd commits to receive bitmaps. | |
1112 | * | |
1113 | * Reachability bitmaps do require that their objects be closed under | |
1114 | * reachability, but fetching any objects missing from promisors at this | |
1115 | * point is too late. But, if one of those objects can be reached from | |
1116 | * an another object that is included in the bitmap, then we will | |
1117 | * complain later that we don't have reachability closure (and fail | |
1118 | * appropriately). | |
1119 | */ | |
1120 | fetch_if_missing = 0; | |
1121 | revs.exclude_promisor_objects = 1; | |
1122 | ||
1123 | if (prepare_revision_walk(&revs)) | |
1124 | die(_("revision walk setup failed")); | |
1125 | ||
1126 | traverse_commit_list(&revs, bitmap_show_commit, NULL, &cb); | |
1127 | if (indexed_commits_nr_p) | |
1128 | *indexed_commits_nr_p = cb.commits_nr; | |
1129 | ||
1130 | release_revisions(&revs); | |
1131 | ||
1132 | trace2_region_leave("midx", "find_commits_for_midx_bitmap", | |
1133 | the_repository); | |
1134 | ||
1135 | return cb.commits; | |
1136 | } | |
1137 | ||
1138 | static int write_midx_bitmap(const char *midx_name, | |
1139 | const unsigned char *midx_hash, | |
1140 | struct packing_data *pdata, | |
1141 | struct commit **commits, | |
1142 | uint32_t commits_nr, | |
1143 | uint32_t *pack_order, | |
1144 | unsigned flags) | |
1145 | { | |
1146 | int ret, i; | |
1147 | uint16_t options = 0; | |
1148 | struct pack_idx_entry **index; | |
1149 | char *bitmap_name = xstrfmt("%s-%s.bitmap", midx_name, | |
1150 | hash_to_hex(midx_hash)); | |
1151 | ||
1152 | trace2_region_enter("midx", "write_midx_bitmap", the_repository); | |
1153 | ||
1154 | if (flags & MIDX_WRITE_BITMAP_HASH_CACHE) | |
1155 | options |= BITMAP_OPT_HASH_CACHE; | |
1156 | ||
1157 | if (flags & MIDX_WRITE_BITMAP_LOOKUP_TABLE) | |
1158 | options |= BITMAP_OPT_LOOKUP_TABLE; | |
1159 | ||
1160 | /* | |
1161 | * Build the MIDX-order index based on pdata.objects (which is already | |
1162 | * in MIDX order; c.f., 'midx_pack_order_cmp()' for the definition of | |
1163 | * this order). | |
1164 | */ | |
1165 | ALLOC_ARRAY(index, pdata->nr_objects); | |
1166 | for (i = 0; i < pdata->nr_objects; i++) | |
1167 | index[i] = &pdata->objects[i].idx; | |
1168 | ||
1169 | bitmap_writer_show_progress(flags & MIDX_PROGRESS); | |
1170 | bitmap_writer_build_type_index(pdata, index, pdata->nr_objects); | |
1171 | ||
1172 | /* | |
1173 | * bitmap_writer_finish expects objects in lex order, but pack_order | |
1174 | * gives us exactly that. use it directly instead of re-sorting the | |
1175 | * array. | |
1176 | * | |
1177 | * This changes the order of objects in 'index' between | |
1178 | * bitmap_writer_build_type_index and bitmap_writer_finish. | |
1179 | * | |
1180 | * The same re-ordering takes place in the single-pack bitmap code via | |
1181 | * write_idx_file(), which is called by finish_tmp_packfile(), which | |
1182 | * happens between bitmap_writer_build_type_index() and | |
1183 | * bitmap_writer_finish(). | |
1184 | */ | |
1185 | for (i = 0; i < pdata->nr_objects; i++) | |
1186 | index[pack_order[i]] = &pdata->objects[i].idx; | |
1187 | ||
1188 | bitmap_writer_select_commits(commits, commits_nr, -1); | |
1189 | ret = bitmap_writer_build(pdata); | |
1190 | if (ret < 0) | |
1191 | goto cleanup; | |
1192 | ||
1193 | bitmap_writer_set_checksum(midx_hash); | |
1194 | bitmap_writer_finish(index, pdata->nr_objects, bitmap_name, options); | |
1195 | ||
1196 | cleanup: | |
1197 | free(index); | |
1198 | free(bitmap_name); | |
1199 | ||
1200 | trace2_region_leave("midx", "write_midx_bitmap", the_repository); | |
1201 | ||
1202 | return ret; | |
1203 | } | |
1204 | ||
1205 | static struct multi_pack_index *lookup_multi_pack_index(struct repository *r, | |
1206 | const char *object_dir) | |
1207 | { | |
1208 | struct multi_pack_index *result = NULL; | |
1209 | struct multi_pack_index *cur; | |
1210 | char *obj_dir_real = real_pathdup(object_dir, 1); | |
1211 | struct strbuf cur_path_real = STRBUF_INIT; | |
1212 | ||
1213 | /* Ensure the given object_dir is local, or a known alternate. */ | |
1214 | find_odb(r, obj_dir_real); | |
1215 | ||
1216 | for (cur = get_multi_pack_index(r); cur; cur = cur->next) { | |
1217 | strbuf_realpath(&cur_path_real, cur->object_dir, 1); | |
1218 | if (!strcmp(obj_dir_real, cur_path_real.buf)) { | |
1219 | result = cur; | |
1220 | goto cleanup; | |
1221 | } | |
1222 | } | |
1223 | ||
1224 | cleanup: | |
1225 | free(obj_dir_real); | |
1226 | strbuf_release(&cur_path_real); | |
1227 | return result; | |
1228 | } | |
1229 | ||
1230 | static int write_midx_internal(const char *object_dir, | |
1231 | struct string_list *packs_to_include, | |
1232 | struct string_list *packs_to_drop, | |
1233 | const char *preferred_pack_name, | |
1234 | const char *refs_snapshot, | |
1235 | unsigned flags) | |
1236 | { | |
1237 | struct strbuf midx_name = STRBUF_INIT; | |
1238 | unsigned char midx_hash[GIT_MAX_RAWSZ]; | |
1239 | uint32_t i; | |
1240 | struct hashfile *f = NULL; | |
1241 | struct lock_file lk; | |
1242 | struct write_midx_context ctx = { 0 }; | |
1243 | int pack_name_concat_len = 0; | |
1244 | int dropped_packs = 0; | |
1245 | int result = 0; | |
1246 | struct chunkfile *cf; | |
1247 | ||
1248 | trace2_region_enter("midx", "write_midx_internal", the_repository); | |
1249 | ||
1250 | get_midx_filename(&midx_name, object_dir); | |
1251 | if (safe_create_leading_directories(midx_name.buf)) | |
1252 | die_errno(_("unable to create leading directories of %s"), | |
1253 | midx_name.buf); | |
1254 | ||
1255 | if (!packs_to_include) { | |
1256 | /* | |
1257 | * Only reference an existing MIDX when not filtering which | |
1258 | * packs to include, since all packs and objects are copied | |
1259 | * blindly from an existing MIDX if one is present. | |
1260 | */ | |
1261 | ctx.m = lookup_multi_pack_index(the_repository, object_dir); | |
1262 | } | |
1263 | ||
1264 | if (ctx.m && !midx_checksum_valid(ctx.m)) { | |
1265 | warning(_("ignoring existing multi-pack-index; checksum mismatch")); | |
1266 | ctx.m = NULL; | |
1267 | } | |
1268 | ||
1269 | ctx.nr = 0; | |
1270 | ctx.alloc = ctx.m ? ctx.m->num_packs : 16; | |
1271 | ctx.info = NULL; | |
1272 | ALLOC_ARRAY(ctx.info, ctx.alloc); | |
1273 | ||
1274 | if (ctx.m) { | |
1275 | for (i = 0; i < ctx.m->num_packs; i++) { | |
1276 | ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc); | |
1277 | ||
1278 | ctx.info[ctx.nr].orig_pack_int_id = i; | |
1279 | ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]); | |
1280 | ctx.info[ctx.nr].p = ctx.m->packs[i]; | |
1281 | ctx.info[ctx.nr].expired = 0; | |
1282 | ||
1283 | if (flags & MIDX_WRITE_REV_INDEX) { | |
1284 | /* | |
1285 | * If generating a reverse index, need to have | |
1286 | * packed_git's loaded to compare their | |
1287 | * mtimes and object count. | |
1288 | */ | |
1289 | if (prepare_midx_pack(the_repository, ctx.m, i)) { | |
1290 | error(_("could not load pack")); | |
1291 | result = 1; | |
1292 | goto cleanup; | |
1293 | } | |
1294 | ||
1295 | if (open_pack_index(ctx.m->packs[i])) | |
1296 | die(_("could not open index for %s"), | |
1297 | ctx.m->packs[i]->pack_name); | |
1298 | ctx.info[ctx.nr].p = ctx.m->packs[i]; | |
1299 | } | |
1300 | ||
1301 | ctx.nr++; | |
1302 | } | |
1303 | } | |
1304 | ||
1305 | ctx.pack_paths_checked = 0; | |
1306 | if (flags & MIDX_PROGRESS) | |
1307 | ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0); | |
1308 | else | |
1309 | ctx.progress = NULL; | |
1310 | ||
1311 | ctx.to_include = packs_to_include; | |
1312 | ||
1313 | for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx); | |
1314 | stop_progress(&ctx.progress); | |
1315 | ||
1316 | if ((ctx.m && ctx.nr == ctx.m->num_packs) && | |
1317 | !(packs_to_include || packs_to_drop)) { | |
1318 | struct bitmap_index *bitmap_git; | |
1319 | int bitmap_exists; | |
1320 | int want_bitmap = flags & MIDX_WRITE_BITMAP; | |
1321 | ||
1322 | bitmap_git = prepare_midx_bitmap_git(ctx.m); | |
1323 | bitmap_exists = bitmap_git && bitmap_is_midx(bitmap_git); | |
1324 | free_bitmap_index(bitmap_git); | |
1325 | ||
1326 | if (bitmap_exists || !want_bitmap) { | |
1327 | /* | |
1328 | * The correct MIDX already exists, and so does a | |
1329 | * corresponding bitmap (or one wasn't requested). | |
1330 | */ | |
1331 | if (!want_bitmap) | |
1332 | clear_midx_files_ext(object_dir, ".bitmap", | |
1333 | NULL); | |
1334 | goto cleanup; | |
1335 | } | |
1336 | } | |
1337 | ||
1338 | if (preferred_pack_name) { | |
1339 | ctx.preferred_pack_idx = -1; | |
1340 | ||
1341 | for (i = 0; i < ctx.nr; i++) { | |
1342 | if (!cmp_idx_or_pack_name(preferred_pack_name, | |
1343 | ctx.info[i].pack_name)) { | |
1344 | ctx.preferred_pack_idx = i; | |
1345 | break; | |
1346 | } | |
1347 | } | |
1348 | ||
1349 | if (ctx.preferred_pack_idx == -1) | |
1350 | warning(_("unknown preferred pack: '%s'"), | |
1351 | preferred_pack_name); | |
1352 | } else if (ctx.nr && | |
1353 | (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP))) { | |
1354 | struct packed_git *oldest = ctx.info[ctx.preferred_pack_idx].p; | |
1355 | ctx.preferred_pack_idx = 0; | |
1356 | ||
1357 | if (packs_to_drop && packs_to_drop->nr) | |
1358 | BUG("cannot write a MIDX bitmap during expiration"); | |
1359 | ||
1360 | /* | |
1361 | * set a preferred pack when writing a bitmap to ensure that | |
1362 | * the pack from which the first object is selected in pseudo | |
1363 | * pack-order has all of its objects selected from that pack | |
1364 | * (and not another pack containing a duplicate) | |
1365 | */ | |
1366 | for (i = 1; i < ctx.nr; i++) { | |
1367 | struct packed_git *p = ctx.info[i].p; | |
1368 | ||
1369 | if (!oldest->num_objects || p->mtime < oldest->mtime) { | |
1370 | oldest = p; | |
1371 | ctx.preferred_pack_idx = i; | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | if (!oldest->num_objects) { | |
1376 | /* | |
1377 | * If all packs are empty; unset the preferred index. | |
1378 | * This is acceptable since there will be no duplicate | |
1379 | * objects to resolve, so the preferred value doesn't | |
1380 | * matter. | |
1381 | */ | |
1382 | ctx.preferred_pack_idx = -1; | |
1383 | } | |
1384 | } else { | |
1385 | /* | |
1386 | * otherwise don't mark any pack as preferred to avoid | |
1387 | * interfering with expiration logic below | |
1388 | */ | |
1389 | ctx.preferred_pack_idx = -1; | |
1390 | } | |
1391 | ||
1392 | if (ctx.preferred_pack_idx > -1) { | |
1393 | struct packed_git *preferred = ctx.info[ctx.preferred_pack_idx].p; | |
1394 | if (!preferred->num_objects) { | |
1395 | error(_("cannot select preferred pack %s with no objects"), | |
1396 | preferred->pack_name); | |
1397 | result = 1; | |
1398 | goto cleanup; | |
1399 | } | |
1400 | } | |
1401 | ||
1402 | ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr, | |
1403 | ctx.preferred_pack_idx); | |
1404 | ||
1405 | ctx.large_offsets_needed = 0; | |
1406 | for (i = 0; i < ctx.entries_nr; i++) { | |
1407 | if (ctx.entries[i].offset > 0x7fffffff) | |
1408 | ctx.num_large_offsets++; | |
1409 | if (ctx.entries[i].offset > 0xffffffff) | |
1410 | ctx.large_offsets_needed = 1; | |
1411 | } | |
1412 | ||
1413 | QSORT(ctx.info, ctx.nr, pack_info_compare); | |
1414 | ||
1415 | if (packs_to_drop && packs_to_drop->nr) { | |
1416 | int drop_index = 0; | |
1417 | int missing_drops = 0; | |
1418 | ||
1419 | for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) { | |
1420 | int cmp = strcmp(ctx.info[i].pack_name, | |
1421 | packs_to_drop->items[drop_index].string); | |
1422 | ||
1423 | if (!cmp) { | |
1424 | drop_index++; | |
1425 | ctx.info[i].expired = 1; | |
1426 | } else if (cmp > 0) { | |
1427 | error(_("did not see pack-file %s to drop"), | |
1428 | packs_to_drop->items[drop_index].string); | |
1429 | drop_index++; | |
1430 | missing_drops++; | |
1431 | i--; | |
1432 | } else { | |
1433 | ctx.info[i].expired = 0; | |
1434 | } | |
1435 | } | |
1436 | ||
1437 | if (missing_drops) { | |
1438 | result = 1; | |
1439 | goto cleanup; | |
1440 | } | |
1441 | } | |
1442 | ||
1443 | /* | |
1444 | * pack_perm stores a permutation between pack-int-ids from the | |
1445 | * previous multi-pack-index to the new one we are writing: | |
1446 | * | |
1447 | * pack_perm[old_id] = new_id | |
1448 | */ | |
1449 | ALLOC_ARRAY(ctx.pack_perm, ctx.nr); | |
1450 | for (i = 0; i < ctx.nr; i++) { | |
1451 | if (ctx.info[i].expired) { | |
1452 | dropped_packs++; | |
1453 | ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED; | |
1454 | } else { | |
1455 | ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs; | |
1456 | } | |
1457 | } | |
1458 | ||
1459 | for (i = 0; i < ctx.nr; i++) { | |
1460 | if (!ctx.info[i].expired) | |
1461 | pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1; | |
1462 | } | |
1463 | ||
1464 | /* Check that the preferred pack wasn't expired (if given). */ | |
1465 | if (preferred_pack_name) { | |
1466 | struct pack_info *preferred = bsearch(preferred_pack_name, | |
1467 | ctx.info, ctx.nr, | |
1468 | sizeof(*ctx.info), | |
1469 | idx_or_pack_name_cmp); | |
1470 | if (preferred) { | |
1471 | uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id]; | |
1472 | if (perm == PACK_EXPIRED) | |
1473 | warning(_("preferred pack '%s' is expired"), | |
1474 | preferred_pack_name); | |
1475 | } | |
1476 | } | |
1477 | ||
1478 | if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT) | |
1479 | pack_name_concat_len += MIDX_CHUNK_ALIGNMENT - | |
1480 | (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT); | |
1481 | ||
1482 | hold_lock_file_for_update(&lk, midx_name.buf, LOCK_DIE_ON_ERROR); | |
1483 | f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk)); | |
1484 | ||
1485 | if (ctx.nr - dropped_packs == 0) { | |
1486 | error(_("no pack files to index.")); | |
1487 | result = 1; | |
1488 | goto cleanup; | |
1489 | } | |
1490 | ||
1491 | if (!ctx.entries_nr) { | |
1492 | if (flags & MIDX_WRITE_BITMAP) | |
1493 | warning(_("refusing to write multi-pack .bitmap without any objects")); | |
1494 | flags &= ~(MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP); | |
1495 | } | |
1496 | ||
1497 | cf = init_chunkfile(f); | |
1498 | ||
1499 | add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len, | |
1500 | write_midx_pack_names); | |
1501 | add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE, | |
1502 | write_midx_oid_fanout); | |
1503 | add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, | |
1504 | (size_t)ctx.entries_nr * the_hash_algo->rawsz, | |
1505 | write_midx_oid_lookup); | |
1506 | add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, | |
1507 | (size_t)ctx.entries_nr * MIDX_CHUNK_OFFSET_WIDTH, | |
1508 | write_midx_object_offsets); | |
1509 | ||
1510 | if (ctx.large_offsets_needed) | |
1511 | add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, | |
1512 | (size_t)ctx.num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH, | |
1513 | write_midx_large_offsets); | |
1514 | ||
1515 | if (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP)) { | |
1516 | ctx.pack_order = midx_pack_order(&ctx); | |
1517 | add_chunk(cf, MIDX_CHUNKID_REVINDEX, | |
1518 | ctx.entries_nr * sizeof(uint32_t), | |
1519 | write_midx_revindex); | |
1520 | } | |
1521 | ||
1522 | write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs); | |
1523 | write_chunkfile(cf, &ctx); | |
1524 | ||
1525 | finalize_hashfile(f, midx_hash, FSYNC_COMPONENT_PACK_METADATA, | |
1526 | CSUM_FSYNC | CSUM_HASH_IN_STREAM); | |
1527 | free_chunkfile(cf); | |
1528 | ||
1529 | if (flags & MIDX_WRITE_REV_INDEX && | |
1530 | git_env_bool("GIT_TEST_MIDX_WRITE_REV", 0)) | |
1531 | write_midx_reverse_index(midx_name.buf, midx_hash, &ctx); | |
1532 | ||
1533 | if (flags & MIDX_WRITE_BITMAP) { | |
1534 | struct packing_data pdata; | |
1535 | struct commit **commits; | |
1536 | uint32_t commits_nr; | |
1537 | ||
1538 | if (!ctx.entries_nr) | |
1539 | BUG("cannot write a bitmap without any objects"); | |
1540 | ||
1541 | prepare_midx_packing_data(&pdata, &ctx); | |
1542 | ||
1543 | commits = find_commits_for_midx_bitmap(&commits_nr, refs_snapshot, &ctx); | |
1544 | ||
1545 | /* | |
1546 | * The previous steps translated the information from | |
1547 | * 'entries' into information suitable for constructing | |
1548 | * bitmaps. We no longer need that array, so clear it to | |
1549 | * reduce memory pressure. | |
1550 | */ | |
1551 | FREE_AND_NULL(ctx.entries); | |
1552 | ctx.entries_nr = 0; | |
1553 | ||
1554 | if (write_midx_bitmap(midx_name.buf, midx_hash, &pdata, | |
1555 | commits, commits_nr, ctx.pack_order, | |
1556 | flags) < 0) { | |
1557 | error(_("could not write multi-pack bitmap")); | |
1558 | result = 1; | |
1559 | goto cleanup; | |
1560 | } | |
1561 | } | |
1562 | /* | |
1563 | * NOTE: Do not use ctx.entries beyond this point, since it might | |
1564 | * have been freed in the previous if block. | |
1565 | */ | |
1566 | ||
1567 | if (ctx.m) | |
1568 | close_object_store(the_repository->objects); | |
1569 | ||
1570 | if (commit_lock_file(&lk) < 0) | |
1571 | die_errno(_("could not write multi-pack-index")); | |
1572 | ||
1573 | clear_midx_files_ext(object_dir, ".bitmap", midx_hash); | |
1574 | clear_midx_files_ext(object_dir, ".rev", midx_hash); | |
1575 | ||
1576 | cleanup: | |
1577 | for (i = 0; i < ctx.nr; i++) { | |
1578 | if (ctx.info[i].p) { | |
1579 | close_pack(ctx.info[i].p); | |
1580 | free(ctx.info[i].p); | |
1581 | } | |
1582 | free(ctx.info[i].pack_name); | |
1583 | } | |
1584 | ||
1585 | free(ctx.info); | |
1586 | free(ctx.entries); | |
1587 | free(ctx.pack_perm); | |
1588 | free(ctx.pack_order); | |
1589 | strbuf_release(&midx_name); | |
1590 | ||
1591 | trace2_region_leave("midx", "write_midx_internal", the_repository); | |
1592 | ||
1593 | return result; | |
1594 | } | |
1595 | ||
1596 | int write_midx_file(const char *object_dir, | |
1597 | const char *preferred_pack_name, | |
1598 | const char *refs_snapshot, | |
1599 | unsigned flags) | |
1600 | { | |
1601 | return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name, | |
1602 | refs_snapshot, flags); | |
1603 | } | |
1604 | ||
1605 | int write_midx_file_only(const char *object_dir, | |
1606 | struct string_list *packs_to_include, | |
1607 | const char *preferred_pack_name, | |
1608 | const char *refs_snapshot, | |
1609 | unsigned flags) | |
1610 | { | |
1611 | return write_midx_internal(object_dir, packs_to_include, NULL, | |
1612 | preferred_pack_name, refs_snapshot, flags); | |
1613 | } | |
1614 | ||
1615 | struct clear_midx_data { | |
1616 | char *keep; | |
1617 | const char *ext; | |
1618 | }; | |
1619 | ||
1620 | static void clear_midx_file_ext(const char *full_path, size_t full_path_len UNUSED, | |
1621 | const char *file_name, void *_data) | |
1622 | { | |
1623 | struct clear_midx_data *data = _data; | |
1624 | ||
1625 | if (!(starts_with(file_name, "multi-pack-index-") && | |
1626 | ends_with(file_name, data->ext))) | |
1627 | return; | |
1628 | if (data->keep && !strcmp(data->keep, file_name)) | |
1629 | return; | |
1630 | ||
1631 | if (unlink(full_path)) | |
1632 | die_errno(_("failed to remove %s"), full_path); | |
1633 | } | |
1634 | ||
1635 | static void clear_midx_files_ext(const char *object_dir, const char *ext, | |
1636 | unsigned char *keep_hash) | |
1637 | { | |
1638 | struct clear_midx_data data; | |
1639 | memset(&data, 0, sizeof(struct clear_midx_data)); | |
1640 | ||
1641 | if (keep_hash) | |
1642 | data.keep = xstrfmt("multi-pack-index-%s%s", | |
1643 | hash_to_hex(keep_hash), ext); | |
1644 | data.ext = ext; | |
1645 | ||
1646 | for_each_file_in_pack_dir(object_dir, | |
1647 | clear_midx_file_ext, | |
1648 | &data); | |
1649 | ||
1650 | free(data.keep); | |
1651 | } | |
1652 | ||
1653 | void clear_midx_file(struct repository *r) | |
1654 | { | |
1655 | struct strbuf midx = STRBUF_INIT; | |
1656 | ||
1657 | get_midx_filename(&midx, r->objects->odb->path); | |
1658 | ||
1659 | if (r->objects && r->objects->multi_pack_index) { | |
1660 | close_midx(r->objects->multi_pack_index); | |
1661 | r->objects->multi_pack_index = NULL; | |
1662 | } | |
1663 | ||
1664 | if (remove_path(midx.buf)) | |
1665 | die(_("failed to clear multi-pack-index at %s"), midx.buf); | |
1666 | ||
1667 | clear_midx_files_ext(r->objects->odb->path, ".bitmap", NULL); | |
1668 | clear_midx_files_ext(r->objects->odb->path, ".rev", NULL); | |
1669 | ||
1670 | strbuf_release(&midx); | |
1671 | } | |
1672 | ||
1673 | static int verify_midx_error; | |
1674 | ||
1675 | __attribute__((format (printf, 1, 2))) | |
1676 | static void midx_report(const char *fmt, ...) | |
1677 | { | |
1678 | va_list ap; | |
1679 | verify_midx_error = 1; | |
1680 | va_start(ap, fmt); | |
1681 | vfprintf(stderr, fmt, ap); | |
1682 | fprintf(stderr, "\n"); | |
1683 | va_end(ap); | |
1684 | } | |
1685 | ||
1686 | struct pair_pos_vs_id | |
1687 | { | |
1688 | uint32_t pos; | |
1689 | uint32_t pack_int_id; | |
1690 | }; | |
1691 | ||
1692 | static int compare_pair_pos_vs_id(const void *_a, const void *_b) | |
1693 | { | |
1694 | struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a; | |
1695 | struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b; | |
1696 | ||
1697 | return b->pack_int_id - a->pack_int_id; | |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * Limit calls to display_progress() for performance reasons. | |
1702 | * The interval here was arbitrarily chosen. | |
1703 | */ | |
1704 | #define SPARSE_PROGRESS_INTERVAL (1 << 12) | |
1705 | #define midx_display_sparse_progress(progress, n) \ | |
1706 | do { \ | |
1707 | uint64_t _n = (n); \ | |
1708 | if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \ | |
1709 | display_progress(progress, _n); \ | |
1710 | } while (0) | |
1711 | ||
1712 | int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags) | |
1713 | { | |
1714 | struct pair_pos_vs_id *pairs = NULL; | |
1715 | uint32_t i; | |
1716 | struct progress *progress = NULL; | |
1717 | struct multi_pack_index *m = load_multi_pack_index(object_dir, 1); | |
1718 | verify_midx_error = 0; | |
1719 | ||
1720 | if (!m) { | |
1721 | int result = 0; | |
1722 | struct stat sb; | |
1723 | struct strbuf filename = STRBUF_INIT; | |
1724 | ||
1725 | get_midx_filename(&filename, object_dir); | |
1726 | ||
1727 | if (!stat(filename.buf, &sb)) { | |
1728 | error(_("multi-pack-index file exists, but failed to parse")); | |
1729 | result = 1; | |
1730 | } | |
1731 | strbuf_release(&filename); | |
1732 | return result; | |
1733 | } | |
1734 | ||
1735 | if (!midx_checksum_valid(m)) | |
1736 | midx_report(_("incorrect checksum")); | |
1737 | ||
1738 | if (flags & MIDX_PROGRESS) | |
1739 | progress = start_delayed_progress(_("Looking for referenced packfiles"), | |
1740 | m->num_packs); | |
1741 | for (i = 0; i < m->num_packs; i++) { | |
1742 | if (prepare_midx_pack(r, m, i)) | |
1743 | midx_report("failed to load pack in position %d", i); | |
1744 | ||
1745 | display_progress(progress, i + 1); | |
1746 | } | |
1747 | stop_progress(&progress); | |
1748 | ||
1749 | for (i = 0; i < 255; i++) { | |
1750 | uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]); | |
1751 | uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]); | |
1752 | ||
1753 | if (oid_fanout1 > oid_fanout2) | |
1754 | midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"), | |
1755 | i, oid_fanout1, oid_fanout2, i + 1); | |
1756 | } | |
1757 | ||
1758 | if (m->num_objects == 0) { | |
1759 | midx_report(_("the midx contains no oid")); | |
1760 | /* | |
1761 | * Remaining tests assume that we have objects, so we can | |
1762 | * return here. | |
1763 | */ | |
1764 | goto cleanup; | |
1765 | } | |
1766 | ||
1767 | if (flags & MIDX_PROGRESS) | |
1768 | progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"), | |
1769 | m->num_objects - 1); | |
1770 | for (i = 0; i < m->num_objects - 1; i++) { | |
1771 | struct object_id oid1, oid2; | |
1772 | ||
1773 | nth_midxed_object_oid(&oid1, m, i); | |
1774 | nth_midxed_object_oid(&oid2, m, i + 1); | |
1775 | ||
1776 | if (oidcmp(&oid1, &oid2) >= 0) | |
1777 | midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"), | |
1778 | i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1); | |
1779 | ||
1780 | midx_display_sparse_progress(progress, i + 1); | |
1781 | } | |
1782 | stop_progress(&progress); | |
1783 | ||
1784 | /* | |
1785 | * Create an array mapping each object to its packfile id. Sort it | |
1786 | * to group the objects by packfile. Use this permutation to visit | |
1787 | * each of the objects and only require 1 packfile to be open at a | |
1788 | * time. | |
1789 | */ | |
1790 | ALLOC_ARRAY(pairs, m->num_objects); | |
1791 | for (i = 0; i < m->num_objects; i++) { | |
1792 | pairs[i].pos = i; | |
1793 | pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i); | |
1794 | } | |
1795 | ||
1796 | if (flags & MIDX_PROGRESS) | |
1797 | progress = start_sparse_progress(_("Sorting objects by packfile"), | |
1798 | m->num_objects); | |
1799 | display_progress(progress, 0); /* TODO: Measure QSORT() progress */ | |
1800 | QSORT(pairs, m->num_objects, compare_pair_pos_vs_id); | |
1801 | stop_progress(&progress); | |
1802 | ||
1803 | if (flags & MIDX_PROGRESS) | |
1804 | progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects); | |
1805 | for (i = 0; i < m->num_objects; i++) { | |
1806 | struct object_id oid; | |
1807 | struct pack_entry e; | |
1808 | off_t m_offset, p_offset; | |
1809 | ||
1810 | if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id && | |
1811 | m->packs[pairs[i-1].pack_int_id]) | |
1812 | { | |
1813 | close_pack_fd(m->packs[pairs[i-1].pack_int_id]); | |
1814 | close_pack_index(m->packs[pairs[i-1].pack_int_id]); | |
1815 | } | |
1816 | ||
1817 | nth_midxed_object_oid(&oid, m, pairs[i].pos); | |
1818 | ||
1819 | if (!fill_midx_entry(r, &oid, &e, m)) { | |
1820 | midx_report(_("failed to load pack entry for oid[%d] = %s"), | |
1821 | pairs[i].pos, oid_to_hex(&oid)); | |
1822 | continue; | |
1823 | } | |
1824 | ||
1825 | if (open_pack_index(e.p)) { | |
1826 | midx_report(_("failed to load pack-index for packfile %s"), | |
1827 | e.p->pack_name); | |
1828 | break; | |
1829 | } | |
1830 | ||
1831 | m_offset = e.offset; | |
1832 | p_offset = find_pack_entry_one(oid.hash, e.p); | |
1833 | ||
1834 | if (m_offset != p_offset) | |
1835 | midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64), | |
1836 | pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset); | |
1837 | ||
1838 | midx_display_sparse_progress(progress, i + 1); | |
1839 | } | |
1840 | stop_progress(&progress); | |
1841 | ||
1842 | cleanup: | |
1843 | free(pairs); | |
1844 | close_midx(m); | |
1845 | ||
1846 | return verify_midx_error; | |
1847 | } | |
1848 | ||
1849 | int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags) | |
1850 | { | |
1851 | uint32_t i, *count, result = 0; | |
1852 | struct string_list packs_to_drop = STRING_LIST_INIT_DUP; | |
1853 | struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir); | |
1854 | struct progress *progress = NULL; | |
1855 | ||
1856 | if (!m) | |
1857 | return 0; | |
1858 | ||
1859 | CALLOC_ARRAY(count, m->num_packs); | |
1860 | ||
1861 | if (flags & MIDX_PROGRESS) | |
1862 | progress = start_delayed_progress(_("Counting referenced objects"), | |
1863 | m->num_objects); | |
1864 | for (i = 0; i < m->num_objects; i++) { | |
1865 | int pack_int_id = nth_midxed_pack_int_id(m, i); | |
1866 | count[pack_int_id]++; | |
1867 | display_progress(progress, i + 1); | |
1868 | } | |
1869 | stop_progress(&progress); | |
1870 | ||
1871 | if (flags & MIDX_PROGRESS) | |
1872 | progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"), | |
1873 | m->num_packs); | |
1874 | for (i = 0; i < m->num_packs; i++) { | |
1875 | char *pack_name; | |
1876 | display_progress(progress, i + 1); | |
1877 | ||
1878 | if (count[i]) | |
1879 | continue; | |
1880 | ||
1881 | if (prepare_midx_pack(r, m, i)) | |
1882 | continue; | |
1883 | ||
1884 | if (m->packs[i]->pack_keep || m->packs[i]->is_cruft) | |
1885 | continue; | |
1886 | ||
1887 | pack_name = xstrdup(m->packs[i]->pack_name); | |
1888 | close_pack(m->packs[i]); | |
1889 | ||
1890 | string_list_insert(&packs_to_drop, m->pack_names[i]); | |
1891 | unlink_pack_path(pack_name, 0); | |
1892 | free(pack_name); | |
1893 | } | |
1894 | stop_progress(&progress); | |
1895 | ||
1896 | free(count); | |
1897 | ||
1898 | if (packs_to_drop.nr) | |
1899 | result = write_midx_internal(object_dir, NULL, &packs_to_drop, NULL, NULL, flags); | |
1900 | ||
1901 | string_list_clear(&packs_to_drop, 0); | |
1902 | ||
1903 | return result; | |
1904 | } | |
1905 | ||
1906 | struct repack_info { | |
1907 | timestamp_t mtime; | |
1908 | uint32_t referenced_objects; | |
1909 | uint32_t pack_int_id; | |
1910 | }; | |
1911 | ||
1912 | static int compare_by_mtime(const void *a_, const void *b_) | |
1913 | { | |
1914 | const struct repack_info *a, *b; | |
1915 | ||
1916 | a = (const struct repack_info *)a_; | |
1917 | b = (const struct repack_info *)b_; | |
1918 | ||
1919 | if (a->mtime < b->mtime) | |
1920 | return -1; | |
1921 | if (a->mtime > b->mtime) | |
1922 | return 1; | |
1923 | return 0; | |
1924 | } | |
1925 | ||
1926 | static int fill_included_packs_all(struct repository *r, | |
1927 | struct multi_pack_index *m, | |
1928 | unsigned char *include_pack) | |
1929 | { | |
1930 | uint32_t i, count = 0; | |
1931 | int pack_kept_objects = 0; | |
1932 | ||
1933 | repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects); | |
1934 | ||
1935 | for (i = 0; i < m->num_packs; i++) { | |
1936 | if (prepare_midx_pack(r, m, i)) | |
1937 | continue; | |
1938 | if (!pack_kept_objects && m->packs[i]->pack_keep) | |
1939 | continue; | |
1940 | if (m->packs[i]->is_cruft) | |
1941 | continue; | |
1942 | ||
1943 | include_pack[i] = 1; | |
1944 | count++; | |
1945 | } | |
1946 | ||
1947 | return count < 2; | |
1948 | } | |
1949 | ||
1950 | static int fill_included_packs_batch(struct repository *r, | |
1951 | struct multi_pack_index *m, | |
1952 | unsigned char *include_pack, | |
1953 | size_t batch_size) | |
1954 | { | |
1955 | uint32_t i, packs_to_repack; | |
1956 | size_t total_size; | |
1957 | struct repack_info *pack_info; | |
1958 | int pack_kept_objects = 0; | |
1959 | ||
1960 | CALLOC_ARRAY(pack_info, m->num_packs); | |
1961 | ||
1962 | repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects); | |
1963 | ||
1964 | for (i = 0; i < m->num_packs; i++) { | |
1965 | pack_info[i].pack_int_id = i; | |
1966 | ||
1967 | if (prepare_midx_pack(r, m, i)) | |
1968 | continue; | |
1969 | ||
1970 | pack_info[i].mtime = m->packs[i]->mtime; | |
1971 | } | |
1972 | ||
1973 | for (i = 0; i < m->num_objects; i++) { | |
1974 | uint32_t pack_int_id = nth_midxed_pack_int_id(m, i); | |
1975 | pack_info[pack_int_id].referenced_objects++; | |
1976 | } | |
1977 | ||
1978 | QSORT(pack_info, m->num_packs, compare_by_mtime); | |
1979 | ||
1980 | total_size = 0; | |
1981 | packs_to_repack = 0; | |
1982 | for (i = 0; total_size < batch_size && i < m->num_packs; i++) { | |
1983 | int pack_int_id = pack_info[i].pack_int_id; | |
1984 | struct packed_git *p = m->packs[pack_int_id]; | |
1985 | size_t expected_size; | |
1986 | ||
1987 | if (!p) | |
1988 | continue; | |
1989 | if (!pack_kept_objects && p->pack_keep) | |
1990 | continue; | |
1991 | if (p->is_cruft) | |
1992 | continue; | |
1993 | if (open_pack_index(p) || !p->num_objects) | |
1994 | continue; | |
1995 | ||
1996 | expected_size = (size_t)(p->pack_size | |
1997 | * pack_info[i].referenced_objects); | |
1998 | expected_size /= p->num_objects; | |
1999 | ||
2000 | if (expected_size >= batch_size) | |
2001 | continue; | |
2002 | ||
2003 | packs_to_repack++; | |
2004 | total_size += expected_size; | |
2005 | include_pack[pack_int_id] = 1; | |
2006 | } | |
2007 | ||
2008 | free(pack_info); | |
2009 | ||
2010 | if (packs_to_repack < 2) | |
2011 | return 1; | |
2012 | ||
2013 | return 0; | |
2014 | } | |
2015 | ||
2016 | int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags) | |
2017 | { | |
2018 | int result = 0; | |
2019 | uint32_t i; | |
2020 | unsigned char *include_pack; | |
2021 | struct child_process cmd = CHILD_PROCESS_INIT; | |
2022 | FILE *cmd_in; | |
2023 | struct strbuf base_name = STRBUF_INIT; | |
2024 | struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir); | |
2025 | ||
2026 | /* | |
2027 | * When updating the default for these configuration | |
2028 | * variables in builtin/repack.c, these must be adjusted | |
2029 | * to match. | |
2030 | */ | |
2031 | int delta_base_offset = 1; | |
2032 | int use_delta_islands = 0; | |
2033 | ||
2034 | if (!m) | |
2035 | return 0; | |
2036 | ||
2037 | CALLOC_ARRAY(include_pack, m->num_packs); | |
2038 | ||
2039 | if (batch_size) { | |
2040 | if (fill_included_packs_batch(r, m, include_pack, batch_size)) | |
2041 | goto cleanup; | |
2042 | } else if (fill_included_packs_all(r, m, include_pack)) | |
2043 | goto cleanup; | |
2044 | ||
2045 | repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset); | |
2046 | repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands); | |
2047 | ||
2048 | strvec_push(&cmd.args, "pack-objects"); | |
2049 | ||
2050 | strbuf_addstr(&base_name, object_dir); | |
2051 | strbuf_addstr(&base_name, "/pack/pack"); | |
2052 | strvec_push(&cmd.args, base_name.buf); | |
2053 | ||
2054 | if (delta_base_offset) | |
2055 | strvec_push(&cmd.args, "--delta-base-offset"); | |
2056 | if (use_delta_islands) | |
2057 | strvec_push(&cmd.args, "--delta-islands"); | |
2058 | ||
2059 | if (flags & MIDX_PROGRESS) | |
2060 | strvec_push(&cmd.args, "--progress"); | |
2061 | else | |
2062 | strvec_push(&cmd.args, "-q"); | |
2063 | ||
2064 | strbuf_release(&base_name); | |
2065 | ||
2066 | cmd.git_cmd = 1; | |
2067 | cmd.in = cmd.out = -1; | |
2068 | ||
2069 | if (start_command(&cmd)) { | |
2070 | error(_("could not start pack-objects")); | |
2071 | result = 1; | |
2072 | goto cleanup; | |
2073 | } | |
2074 | ||
2075 | cmd_in = xfdopen(cmd.in, "w"); | |
2076 | ||
2077 | for (i = 0; i < m->num_objects; i++) { | |
2078 | struct object_id oid; | |
2079 | uint32_t pack_int_id = nth_midxed_pack_int_id(m, i); | |
2080 | ||
2081 | if (!include_pack[pack_int_id]) | |
2082 | continue; | |
2083 | ||
2084 | nth_midxed_object_oid(&oid, m, i); | |
2085 | fprintf(cmd_in, "%s\n", oid_to_hex(&oid)); | |
2086 | } | |
2087 | fclose(cmd_in); | |
2088 | ||
2089 | if (finish_command(&cmd)) { | |
2090 | error(_("could not finish pack-objects")); | |
2091 | result = 1; | |
2092 | goto cleanup; | |
2093 | } | |
2094 | ||
2095 | result = write_midx_internal(object_dir, NULL, NULL, NULL, NULL, flags); | |
2096 | ||
2097 | cleanup: | |
2098 | free(include_pack); | |
2099 | return result; | |
2100 | } |