]>
Commit | Line | Data |
---|---|---|
a3407730 | 1 | #include "cache.h" |
c4d25228 | 2 | #include "config.h" |
fc59e748 | 3 | #include "csum-file.h" |
396f2570 | 4 | #include "dir.h" |
fc59e748 | 5 | #include "lockfile.h" |
396f2570 | 6 | #include "packfile.h" |
4d80560c | 7 | #include "object-store.h" |
3715a633 | 8 | #include "sha1-lookup.h" |
a3407730 DS |
9 | #include "midx.h" |
10 | ||
fc59e748 DS |
11 | #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */ |
12 | #define MIDX_VERSION 1 | |
4d80560c DS |
13 | #define MIDX_BYTE_FILE_VERSION 4 |
14 | #define MIDX_BYTE_HASH_VERSION 5 | |
15 | #define MIDX_BYTE_NUM_CHUNKS 6 | |
16 | #define MIDX_BYTE_NUM_PACKS 8 | |
fc59e748 DS |
17 | #define MIDX_HASH_VERSION 1 |
18 | #define MIDX_HEADER_SIZE 12 | |
4d80560c DS |
19 | #define MIDX_HASH_LEN 20 |
20 | #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + MIDX_HASH_LEN) | |
fc59e748 | 21 | |
662148c4 | 22 | #define MIDX_MAX_CHUNKS 5 |
32f3c541 DS |
23 | #define MIDX_CHUNK_ALIGNMENT 4 |
24 | #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */ | |
d7cacf29 | 25 | #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */ |
0d5b3a5e | 26 | #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */ |
662148c4 DS |
27 | #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */ |
28 | #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */ | |
32f3c541 | 29 | #define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t)) |
d7cacf29 | 30 | #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256) |
662148c4 DS |
31 | #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t)) |
32 | #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t)) | |
33 | #define MIDX_LARGE_OFFSET_NEEDED 0x80000000 | |
32f3c541 | 34 | |
fc59e748 DS |
35 | static char *get_midx_filename(const char *object_dir) |
36 | { | |
37 | return xstrfmt("%s/pack/multi-pack-index", object_dir); | |
38 | } | |
39 | ||
2cf489a3 | 40 | struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local) |
4d80560c DS |
41 | { |
42 | struct multi_pack_index *m = NULL; | |
43 | int fd; | |
44 | struct stat st; | |
45 | size_t midx_size; | |
46 | void *midx_map = NULL; | |
47 | uint32_t hash_version; | |
48 | char *midx_name = get_midx_filename(object_dir); | |
32f3c541 | 49 | uint32_t i; |
3227565c | 50 | const char *cur_pack_name; |
4d80560c DS |
51 | |
52 | fd = git_open(midx_name); | |
53 | ||
54 | if (fd < 0) | |
55 | goto cleanup_fail; | |
56 | if (fstat(fd, &st)) { | |
57 | error_errno(_("failed to read %s"), midx_name); | |
58 | goto cleanup_fail; | |
59 | } | |
60 | ||
61 | midx_size = xsize_t(st.st_size); | |
62 | ||
63 | if (midx_size < MIDX_MIN_SIZE) { | |
64 | error(_("multi-pack-index file %s is too small"), midx_name); | |
65 | goto cleanup_fail; | |
66 | } | |
67 | ||
68 | FREE_AND_NULL(midx_name); | |
69 | ||
70 | midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
71 | ||
72 | FLEX_ALLOC_MEM(m, object_dir, object_dir, strlen(object_dir)); | |
73 | m->fd = fd; | |
74 | m->data = midx_map; | |
75 | m->data_len = midx_size; | |
2cf489a3 | 76 | m->local = local; |
4d80560c DS |
77 | |
78 | m->signature = get_be32(m->data); | |
53ad0407 DS |
79 | if (m->signature != MIDX_SIGNATURE) |
80 | die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"), | |
4d80560c | 81 | m->signature, MIDX_SIGNATURE); |
4d80560c DS |
82 | |
83 | m->version = m->data[MIDX_BYTE_FILE_VERSION]; | |
53ad0407 DS |
84 | if (m->version != MIDX_VERSION) |
85 | die(_("multi-pack-index version %d not recognized"), | |
4d80560c | 86 | m->version); |
4d80560c DS |
87 | |
88 | hash_version = m->data[MIDX_BYTE_HASH_VERSION]; | |
53ad0407 DS |
89 | if (hash_version != MIDX_HASH_VERSION) |
90 | die(_("hash version %u does not match"), hash_version); | |
4d80560c DS |
91 | m->hash_len = MIDX_HASH_LEN; |
92 | ||
93 | m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS]; | |
94 | ||
95 | m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS); | |
96 | ||
32f3c541 DS |
97 | for (i = 0; i < m->num_chunks; i++) { |
98 | uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE + | |
99 | MIDX_CHUNKLOOKUP_WIDTH * i); | |
100 | uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 + | |
101 | MIDX_CHUNKLOOKUP_WIDTH * i); | |
102 | ||
103 | switch (chunk_id) { | |
104 | case MIDX_CHUNKID_PACKNAMES: | |
105 | m->chunk_pack_names = m->data + chunk_offset; | |
106 | break; | |
107 | ||
d7cacf29 DS |
108 | case MIDX_CHUNKID_OIDFANOUT: |
109 | m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset); | |
110 | break; | |
111 | ||
0d5b3a5e DS |
112 | case MIDX_CHUNKID_OIDLOOKUP: |
113 | m->chunk_oid_lookup = m->data + chunk_offset; | |
114 | break; | |
115 | ||
662148c4 DS |
116 | case MIDX_CHUNKID_OBJECTOFFSETS: |
117 | m->chunk_object_offsets = m->data + chunk_offset; | |
118 | break; | |
119 | ||
120 | case MIDX_CHUNKID_LARGEOFFSETS: | |
121 | m->chunk_large_offsets = m->data + chunk_offset; | |
122 | break; | |
123 | ||
32f3c541 DS |
124 | case 0: |
125 | die(_("terminating multi-pack-index chunk id appears earlier than expected")); | |
126 | break; | |
127 | ||
128 | default: | |
129 | /* | |
130 | * Do nothing on unrecognized chunks, allowing future | |
131 | * extensions to add optional chunks. | |
132 | */ | |
133 | break; | |
134 | } | |
135 | } | |
136 | ||
137 | if (!m->chunk_pack_names) | |
138 | die(_("multi-pack-index missing required pack-name chunk")); | |
d7cacf29 DS |
139 | if (!m->chunk_oid_fanout) |
140 | die(_("multi-pack-index missing required OID fanout chunk")); | |
0d5b3a5e DS |
141 | if (!m->chunk_oid_lookup) |
142 | die(_("multi-pack-index missing required OID lookup chunk")); | |
662148c4 DS |
143 | if (!m->chunk_object_offsets) |
144 | die(_("multi-pack-index missing required object offsets chunk")); | |
32f3c541 | 145 | |
d7cacf29 DS |
146 | m->num_objects = ntohl(m->chunk_oid_fanout[255]); |
147 | ||
3227565c | 148 | m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names)); |
3715a633 | 149 | m->packs = xcalloc(m->num_packs, sizeof(*m->packs)); |
3227565c DS |
150 | |
151 | cur_pack_name = (const char *)m->chunk_pack_names; | |
152 | for (i = 0; i < m->num_packs; i++) { | |
153 | m->pack_names[i] = cur_pack_name; | |
154 | ||
155 | cur_pack_name += strlen(cur_pack_name) + 1; | |
156 | ||
157 | if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0) { | |
158 | error(_("multi-pack-index pack names out of order: '%s' before '%s'"), | |
159 | m->pack_names[i - 1], | |
160 | m->pack_names[i]); | |
161 | goto cleanup_fail; | |
162 | } | |
163 | } | |
164 | ||
4d80560c DS |
165 | return m; |
166 | ||
167 | cleanup_fail: | |
168 | free(m); | |
169 | free(midx_name); | |
170 | if (midx_map) | |
171 | munmap(midx_map, midx_size); | |
172 | if (0 <= fd) | |
173 | close(fd); | |
174 | return NULL; | |
175 | } | |
176 | ||
a40498a1 DS |
177 | static void close_midx(struct multi_pack_index *m) |
178 | { | |
179 | uint32_t i; | |
180 | munmap((unsigned char *)m->data, m->data_len); | |
181 | close(m->fd); | |
182 | m->fd = -1; | |
183 | ||
184 | for (i = 0; i < m->num_packs; i++) { | |
185 | if (m->packs[i]) { | |
186 | close_pack(m->packs[i]); | |
187 | free(m->packs); | |
188 | } | |
189 | } | |
190 | FREE_AND_NULL(m->packs); | |
191 | FREE_AND_NULL(m->pack_names); | |
192 | } | |
193 | ||
0bff5269 | 194 | int prepare_midx_pack(struct multi_pack_index *m, uint32_t pack_int_id) |
3715a633 DS |
195 | { |
196 | struct strbuf pack_name = STRBUF_INIT; | |
197 | ||
198 | if (pack_int_id >= m->num_packs) | |
199 | BUG("bad pack-int-id"); | |
200 | ||
201 | if (m->packs[pack_int_id]) | |
202 | return 0; | |
203 | ||
204 | strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir, | |
205 | m->pack_names[pack_int_id]); | |
206 | ||
2cf489a3 | 207 | m->packs[pack_int_id] = add_packed_git(pack_name.buf, pack_name.len, m->local); |
3715a633 DS |
208 | strbuf_release(&pack_name); |
209 | return !m->packs[pack_int_id]; | |
210 | } | |
211 | ||
212 | int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result) | |
213 | { | |
214 | return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup, | |
215 | MIDX_HASH_LEN, result); | |
216 | } | |
217 | ||
8aac67a1 DS |
218 | struct object_id *nth_midxed_object_oid(struct object_id *oid, |
219 | struct multi_pack_index *m, | |
220 | uint32_t n) | |
221 | { | |
222 | if (n >= m->num_objects) | |
223 | return NULL; | |
224 | ||
225 | hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n); | |
226 | return oid; | |
227 | } | |
228 | ||
3715a633 DS |
229 | static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos) |
230 | { | |
231 | const unsigned char *offset_data; | |
232 | uint32_t offset32; | |
233 | ||
234 | offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH; | |
235 | offset32 = get_be32(offset_data + sizeof(uint32_t)); | |
236 | ||
237 | if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) { | |
238 | if (sizeof(offset32) < sizeof(uint64_t)) | |
239 | die(_("multi-pack-index stores a 64-bit offset, but off_t is too small")); | |
240 | ||
241 | offset32 ^= MIDX_LARGE_OFFSET_NEEDED; | |
242 | return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32); | |
243 | } | |
244 | ||
245 | return offset32; | |
246 | } | |
247 | ||
248 | static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos) | |
249 | { | |
250 | return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH); | |
251 | } | |
252 | ||
253 | static int nth_midxed_pack_entry(struct multi_pack_index *m, struct pack_entry *e, uint32_t pos) | |
254 | { | |
255 | uint32_t pack_int_id; | |
256 | struct packed_git *p; | |
257 | ||
258 | if (pos >= m->num_objects) | |
259 | return 0; | |
260 | ||
261 | pack_int_id = nth_midxed_pack_int_id(m, pos); | |
262 | ||
263 | if (prepare_midx_pack(m, pack_int_id)) | |
264 | die(_("error preparing packfile from multi-pack-index")); | |
265 | p = m->packs[pack_int_id]; | |
266 | ||
267 | /* | |
268 | * We are about to tell the caller where they can locate the | |
269 | * requested object. We better make sure the packfile is | |
270 | * still here and can be accessed before supplying that | |
271 | * answer, as it may have been deleted since the MIDX was | |
272 | * loaded! | |
273 | */ | |
274 | if (!is_pack_valid(p)) | |
275 | return 0; | |
276 | ||
c39b02ae DS |
277 | if (p->num_bad_objects) { |
278 | uint32_t i; | |
279 | struct object_id oid; | |
280 | nth_midxed_object_oid(&oid, m, pos); | |
281 | for (i = 0; i < p->num_bad_objects; i++) | |
282 | if (!hashcmp(oid.hash, | |
283 | p->bad_object_sha1 + the_hash_algo->rawsz * i)) | |
284 | return 0; | |
285 | } | |
286 | ||
3715a633 DS |
287 | e->offset = nth_midxed_offset(m, pos); |
288 | e->p = p; | |
289 | ||
290 | return 1; | |
291 | } | |
292 | ||
293 | int fill_midx_entry(const struct object_id *oid, struct pack_entry *e, struct multi_pack_index *m) | |
294 | { | |
295 | uint32_t pos; | |
296 | ||
297 | if (!bsearch_midx(oid, m, &pos)) | |
298 | return 0; | |
299 | ||
300 | return nth_midxed_pack_entry(m, e, pos); | |
301 | } | |
302 | ||
a40498a1 DS |
303 | int midx_contains_pack(struct multi_pack_index *m, const char *idx_name) |
304 | { | |
305 | uint32_t first = 0, last = m->num_packs; | |
306 | ||
307 | while (first < last) { | |
308 | uint32_t mid = first + (last - first) / 2; | |
309 | const char *current; | |
310 | int cmp; | |
311 | ||
312 | current = m->pack_names[mid]; | |
313 | cmp = strcmp(idx_name, current); | |
314 | if (!cmp) | |
315 | return 1; | |
316 | if (cmp > 0) { | |
317 | first = mid + 1; | |
318 | continue; | |
319 | } | |
320 | last = mid; | |
321 | } | |
322 | ||
323 | return 0; | |
324 | } | |
325 | ||
2cf489a3 | 326 | int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local) |
c4d25228 | 327 | { |
29e2016b | 328 | struct multi_pack_index *m; |
c4d25228 DS |
329 | struct multi_pack_index *m_search; |
330 | int config_value; | |
331 | ||
332 | if (repo_config_get_bool(r, "core.multipackindex", &config_value) || | |
333 | !config_value) | |
334 | return 0; | |
335 | ||
29e2016b | 336 | for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next) |
c4d25228 DS |
337 | if (!strcmp(object_dir, m_search->object_dir)) |
338 | return 1; | |
339 | ||
29e2016b | 340 | m = load_multi_pack_index(object_dir, local); |
c4d25228 | 341 | |
29e2016b DS |
342 | if (m) { |
343 | m->next = r->objects->multi_pack_index; | |
344 | r->objects->multi_pack_index = m; | |
c4d25228 DS |
345 | return 1; |
346 | } | |
347 | ||
348 | return 0; | |
349 | } | |
350 | ||
fc59e748 DS |
351 | static size_t write_midx_header(struct hashfile *f, |
352 | unsigned char num_chunks, | |
353 | uint32_t num_packs) | |
354 | { | |
355 | unsigned char byte_values[4]; | |
356 | ||
357 | hashwrite_be32(f, MIDX_SIGNATURE); | |
358 | byte_values[0] = MIDX_VERSION; | |
359 | byte_values[1] = MIDX_HASH_VERSION; | |
360 | byte_values[2] = num_chunks; | |
361 | byte_values[3] = 0; /* unused */ | |
362 | hashwrite(f, byte_values, sizeof(byte_values)); | |
363 | hashwrite_be32(f, num_packs); | |
364 | ||
365 | return MIDX_HEADER_SIZE; | |
366 | } | |
367 | ||
396f2570 DS |
368 | struct pack_list { |
369 | struct packed_git **list; | |
32f3c541 | 370 | char **names; |
396f2570 DS |
371 | uint32_t nr; |
372 | uint32_t alloc_list; | |
32f3c541 DS |
373 | uint32_t alloc_names; |
374 | size_t pack_name_concat_len; | |
a40498a1 | 375 | struct multi_pack_index *m; |
396f2570 DS |
376 | }; |
377 | ||
378 | static void add_pack_to_midx(const char *full_path, size_t full_path_len, | |
379 | const char *file_name, void *data) | |
380 | { | |
381 | struct pack_list *packs = (struct pack_list *)data; | |
382 | ||
383 | if (ends_with(file_name, ".idx")) { | |
a40498a1 DS |
384 | if (packs->m && midx_contains_pack(packs->m, file_name)) |
385 | return; | |
386 | ||
396f2570 | 387 | ALLOC_GROW(packs->list, packs->nr + 1, packs->alloc_list); |
32f3c541 | 388 | ALLOC_GROW(packs->names, packs->nr + 1, packs->alloc_names); |
396f2570 DS |
389 | |
390 | packs->list[packs->nr] = add_packed_git(full_path, | |
391 | full_path_len, | |
392 | 0); | |
fe1ed56f | 393 | |
396f2570 DS |
394 | if (!packs->list[packs->nr]) { |
395 | warning(_("failed to add packfile '%s'"), | |
396 | full_path); | |
397 | return; | |
398 | } | |
399 | ||
fe1ed56f DS |
400 | if (open_pack_index(packs->list[packs->nr])) { |
401 | warning(_("failed to open pack-index '%s'"), | |
402 | full_path); | |
403 | close_pack(packs->list[packs->nr]); | |
404 | FREE_AND_NULL(packs->list[packs->nr]); | |
405 | return; | |
406 | } | |
407 | ||
32f3c541 DS |
408 | packs->names[packs->nr] = xstrdup(file_name); |
409 | packs->pack_name_concat_len += strlen(file_name) + 1; | |
396f2570 DS |
410 | packs->nr++; |
411 | } | |
412 | } | |
413 | ||
32f3c541 DS |
414 | struct pack_pair { |
415 | uint32_t pack_int_id; | |
416 | char *pack_name; | |
417 | }; | |
418 | ||
419 | static int pack_pair_compare(const void *_a, const void *_b) | |
420 | { | |
421 | struct pack_pair *a = (struct pack_pair *)_a; | |
422 | struct pack_pair *b = (struct pack_pair *)_b; | |
423 | return strcmp(a->pack_name, b->pack_name); | |
424 | } | |
425 | ||
426 | static void sort_packs_by_name(char **pack_names, uint32_t nr_packs, uint32_t *perm) | |
427 | { | |
428 | uint32_t i; | |
429 | struct pack_pair *pairs; | |
430 | ||
431 | ALLOC_ARRAY(pairs, nr_packs); | |
432 | ||
433 | for (i = 0; i < nr_packs; i++) { | |
434 | pairs[i].pack_int_id = i; | |
435 | pairs[i].pack_name = pack_names[i]; | |
436 | } | |
437 | ||
438 | QSORT(pairs, nr_packs, pack_pair_compare); | |
439 | ||
440 | for (i = 0; i < nr_packs; i++) { | |
441 | pack_names[i] = pairs[i].pack_name; | |
442 | perm[pairs[i].pack_int_id] = i; | |
443 | } | |
444 | ||
445 | free(pairs); | |
446 | } | |
447 | ||
fe1ed56f DS |
448 | struct pack_midx_entry { |
449 | struct object_id oid; | |
450 | uint32_t pack_int_id; | |
451 | time_t pack_mtime; | |
452 | uint64_t offset; | |
453 | }; | |
454 | ||
455 | static int midx_oid_compare(const void *_a, const void *_b) | |
456 | { | |
457 | const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a; | |
458 | const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b; | |
459 | int cmp = oidcmp(&a->oid, &b->oid); | |
460 | ||
461 | if (cmp) | |
462 | return cmp; | |
463 | ||
464 | if (a->pack_mtime > b->pack_mtime) | |
465 | return -1; | |
466 | else if (a->pack_mtime < b->pack_mtime) | |
467 | return 1; | |
468 | ||
469 | return a->pack_int_id - b->pack_int_id; | |
470 | } | |
471 | ||
a40498a1 DS |
472 | static int nth_midxed_pack_midx_entry(struct multi_pack_index *m, |
473 | uint32_t *pack_perm, | |
474 | struct pack_midx_entry *e, | |
475 | uint32_t pos) | |
476 | { | |
477 | if (pos >= m->num_objects) | |
478 | return 1; | |
479 | ||
480 | nth_midxed_object_oid(&e->oid, m, pos); | |
481 | e->pack_int_id = pack_perm[nth_midxed_pack_int_id(m, pos)]; | |
482 | e->offset = nth_midxed_offset(m, pos); | |
483 | ||
484 | /* consider objects in midx to be from "old" packs */ | |
485 | e->pack_mtime = 0; | |
486 | return 0; | |
487 | } | |
488 | ||
fe1ed56f DS |
489 | static void fill_pack_entry(uint32_t pack_int_id, |
490 | struct packed_git *p, | |
491 | uint32_t cur_object, | |
492 | struct pack_midx_entry *entry) | |
493 | { | |
494 | if (!nth_packed_object_oid(&entry->oid, p, cur_object)) | |
495 | die(_("failed to locate object %d in packfile"), cur_object); | |
496 | ||
497 | entry->pack_int_id = pack_int_id; | |
498 | entry->pack_mtime = p->mtime; | |
499 | ||
500 | entry->offset = nth_packed_object_offset(p, cur_object); | |
501 | } | |
502 | ||
503 | /* | |
504 | * It is possible to artificially get into a state where there are many | |
505 | * duplicate copies of objects. That can create high memory pressure if | |
506 | * we are to create a list of all objects before de-duplication. To reduce | |
507 | * this memory pressure without a significant performance drop, automatically | |
508 | * group objects by the first byte of their object id. Use the IDX fanout | |
509 | * tables to group the data, copy to a local array, then sort. | |
510 | * | |
511 | * Copy only the de-duplicated entries (selected by most-recent modified time | |
512 | * of a packfile containing the object). | |
513 | */ | |
a40498a1 DS |
514 | static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m, |
515 | struct packed_git **p, | |
fe1ed56f DS |
516 | uint32_t *perm, |
517 | uint32_t nr_packs, | |
518 | uint32_t *nr_objects) | |
519 | { | |
520 | uint32_t cur_fanout, cur_pack, cur_object; | |
521 | uint32_t alloc_fanout, alloc_objects, total_objects = 0; | |
522 | struct pack_midx_entry *entries_by_fanout = NULL; | |
523 | struct pack_midx_entry *deduplicated_entries = NULL; | |
a40498a1 | 524 | uint32_t start_pack = m ? m->num_packs : 0; |
fe1ed56f | 525 | |
a40498a1 | 526 | for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) |
fe1ed56f DS |
527 | total_objects += p[cur_pack]->num_objects; |
528 | ||
529 | /* | |
530 | * As we de-duplicate by fanout value, we expect the fanout | |
531 | * slices to be evenly distributed, with some noise. Hence, | |
532 | * allocate slightly more than one 256th. | |
533 | */ | |
534 | alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16; | |
535 | ||
536 | ALLOC_ARRAY(entries_by_fanout, alloc_fanout); | |
537 | ALLOC_ARRAY(deduplicated_entries, alloc_objects); | |
538 | *nr_objects = 0; | |
539 | ||
540 | for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) { | |
541 | uint32_t nr_fanout = 0; | |
542 | ||
a40498a1 DS |
543 | if (m) { |
544 | uint32_t start = 0, end; | |
545 | ||
546 | if (cur_fanout) | |
547 | start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]); | |
548 | end = ntohl(m->chunk_oid_fanout[cur_fanout]); | |
549 | ||
550 | for (cur_object = start; cur_object < end; cur_object++) { | |
551 | ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout); | |
552 | nth_midxed_pack_midx_entry(m, perm, | |
553 | &entries_by_fanout[nr_fanout], | |
554 | cur_object); | |
555 | nr_fanout++; | |
556 | } | |
557 | } | |
558 | ||
559 | for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) { | |
fe1ed56f DS |
560 | uint32_t start = 0, end; |
561 | ||
562 | if (cur_fanout) | |
563 | start = get_pack_fanout(p[cur_pack], cur_fanout - 1); | |
564 | end = get_pack_fanout(p[cur_pack], cur_fanout); | |
565 | ||
566 | for (cur_object = start; cur_object < end; cur_object++) { | |
567 | ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout); | |
568 | fill_pack_entry(perm[cur_pack], p[cur_pack], cur_object, &entries_by_fanout[nr_fanout]); | |
569 | nr_fanout++; | |
570 | } | |
571 | } | |
572 | ||
573 | QSORT(entries_by_fanout, nr_fanout, midx_oid_compare); | |
574 | ||
575 | /* | |
576 | * The batch is now sorted by OID and then mtime (descending). | |
577 | * Take only the first duplicate. | |
578 | */ | |
579 | for (cur_object = 0; cur_object < nr_fanout; cur_object++) { | |
580 | if (cur_object && !oidcmp(&entries_by_fanout[cur_object - 1].oid, | |
581 | &entries_by_fanout[cur_object].oid)) | |
582 | continue; | |
583 | ||
584 | ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects); | |
585 | memcpy(&deduplicated_entries[*nr_objects], | |
586 | &entries_by_fanout[cur_object], | |
587 | sizeof(struct pack_midx_entry)); | |
588 | (*nr_objects)++; | |
589 | } | |
590 | } | |
591 | ||
592 | free(entries_by_fanout); | |
593 | return deduplicated_entries; | |
594 | } | |
595 | ||
32f3c541 DS |
596 | static size_t write_midx_pack_names(struct hashfile *f, |
597 | char **pack_names, | |
598 | uint32_t num_packs) | |
599 | { | |
600 | uint32_t i; | |
601 | unsigned char padding[MIDX_CHUNK_ALIGNMENT]; | |
602 | size_t written = 0; | |
603 | ||
604 | for (i = 0; i < num_packs; i++) { | |
605 | size_t writelen = strlen(pack_names[i]) + 1; | |
606 | ||
607 | if (i && strcmp(pack_names[i], pack_names[i - 1]) <= 0) | |
608 | BUG("incorrect pack-file order: %s before %s", | |
609 | pack_names[i - 1], | |
610 | pack_names[i]); | |
611 | ||
612 | hashwrite(f, pack_names[i], writelen); | |
613 | written += writelen; | |
614 | } | |
615 | ||
616 | /* add padding to be aligned */ | |
617 | i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT); | |
618 | if (i < MIDX_CHUNK_ALIGNMENT) { | |
619 | memset(padding, 0, sizeof(padding)); | |
620 | hashwrite(f, padding, i); | |
621 | written += i; | |
622 | } | |
623 | ||
624 | return written; | |
625 | } | |
626 | ||
d7cacf29 DS |
627 | static size_t write_midx_oid_fanout(struct hashfile *f, |
628 | struct pack_midx_entry *objects, | |
629 | uint32_t nr_objects) | |
630 | { | |
631 | struct pack_midx_entry *list = objects; | |
632 | struct pack_midx_entry *last = objects + nr_objects; | |
633 | uint32_t count = 0; | |
634 | uint32_t i; | |
635 | ||
636 | /* | |
637 | * Write the first-level table (the list is sorted, | |
638 | * but we use a 256-entry lookup to be able to avoid | |
639 | * having to do eight extra binary search iterations). | |
640 | */ | |
641 | for (i = 0; i < 256; i++) { | |
642 | struct pack_midx_entry *next = list; | |
643 | ||
644 | while (next < last && next->oid.hash[0] == i) { | |
645 | count++; | |
646 | next++; | |
647 | } | |
648 | ||
649 | hashwrite_be32(f, count); | |
650 | list = next; | |
651 | } | |
652 | ||
653 | return MIDX_CHUNK_FANOUT_SIZE; | |
654 | } | |
655 | ||
0d5b3a5e DS |
656 | static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len, |
657 | struct pack_midx_entry *objects, | |
658 | uint32_t nr_objects) | |
659 | { | |
660 | struct pack_midx_entry *list = objects; | |
661 | uint32_t i; | |
662 | size_t written = 0; | |
663 | ||
664 | for (i = 0; i < nr_objects; i++) { | |
665 | struct pack_midx_entry *obj = list++; | |
666 | ||
667 | if (i < nr_objects - 1) { | |
668 | struct pack_midx_entry *next = list; | |
669 | if (oidcmp(&obj->oid, &next->oid) >= 0) | |
670 | BUG("OIDs not in order: %s >= %s", | |
671 | oid_to_hex(&obj->oid), | |
672 | oid_to_hex(&next->oid)); | |
673 | } | |
674 | ||
675 | hashwrite(f, obj->oid.hash, (int)hash_len); | |
676 | written += hash_len; | |
677 | } | |
678 | ||
679 | return written; | |
680 | } | |
681 | ||
662148c4 DS |
682 | static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed, |
683 | struct pack_midx_entry *objects, uint32_t nr_objects) | |
684 | { | |
685 | struct pack_midx_entry *list = objects; | |
686 | uint32_t i, nr_large_offset = 0; | |
687 | size_t written = 0; | |
688 | ||
689 | for (i = 0; i < nr_objects; i++) { | |
690 | struct pack_midx_entry *obj = list++; | |
691 | ||
692 | hashwrite_be32(f, obj->pack_int_id); | |
693 | ||
694 | if (large_offset_needed && obj->offset >> 31) | |
695 | hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++); | |
696 | else if (!large_offset_needed && obj->offset >> 32) | |
697 | BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!", | |
698 | oid_to_hex(&obj->oid), | |
699 | obj->offset); | |
700 | else | |
701 | hashwrite_be32(f, (uint32_t)obj->offset); | |
702 | ||
703 | written += MIDX_CHUNK_OFFSET_WIDTH; | |
704 | } | |
705 | ||
706 | return written; | |
707 | } | |
708 | ||
709 | static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset, | |
710 | struct pack_midx_entry *objects, uint32_t nr_objects) | |
711 | { | |
712 | struct pack_midx_entry *list = objects; | |
713 | size_t written = 0; | |
714 | ||
715 | while (nr_large_offset) { | |
716 | struct pack_midx_entry *obj = list++; | |
717 | uint64_t offset = obj->offset; | |
718 | ||
719 | if (!(offset >> 31)) | |
720 | continue; | |
721 | ||
722 | hashwrite_be32(f, offset >> 32); | |
723 | hashwrite_be32(f, offset & 0xffffffffUL); | |
724 | written += 2 * sizeof(uint32_t); | |
725 | ||
726 | nr_large_offset--; | |
727 | } | |
728 | ||
729 | return written; | |
730 | } | |
731 | ||
a3407730 DS |
732 | int write_midx_file(const char *object_dir) |
733 | { | |
32f3c541 | 734 | unsigned char cur_chunk, num_chunks = 0; |
fc59e748 | 735 | char *midx_name; |
396f2570 | 736 | uint32_t i; |
fc59e748 DS |
737 | struct hashfile *f = NULL; |
738 | struct lock_file lk; | |
396f2570 | 739 | struct pack_list packs; |
32f3c541 DS |
740 | uint32_t *pack_perm = NULL; |
741 | uint64_t written = 0; | |
742 | uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1]; | |
743 | uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1]; | |
662148c4 | 744 | uint32_t nr_entries, num_large_offsets = 0; |
fe1ed56f | 745 | struct pack_midx_entry *entries = NULL; |
662148c4 | 746 | int large_offsets_needed = 0; |
fc59e748 DS |
747 | |
748 | midx_name = get_midx_filename(object_dir); | |
749 | if (safe_create_leading_directories(midx_name)) { | |
750 | UNLEAK(midx_name); | |
751 | die_errno(_("unable to create leading directories of %s"), | |
752 | midx_name); | |
753 | } | |
754 | ||
2cf489a3 | 755 | packs.m = load_multi_pack_index(object_dir, 1); |
a40498a1 | 756 | |
396f2570 | 757 | packs.nr = 0; |
a40498a1 DS |
758 | packs.alloc_list = packs.m ? packs.m->num_packs : 16; |
759 | packs.alloc_names = packs.alloc_list; | |
396f2570 | 760 | packs.list = NULL; |
a40498a1 | 761 | packs.names = NULL; |
32f3c541 | 762 | packs.pack_name_concat_len = 0; |
396f2570 | 763 | ALLOC_ARRAY(packs.list, packs.alloc_list); |
32f3c541 | 764 | ALLOC_ARRAY(packs.names, packs.alloc_names); |
396f2570 | 765 | |
a40498a1 DS |
766 | if (packs.m) { |
767 | for (i = 0; i < packs.m->num_packs; i++) { | |
768 | ALLOC_GROW(packs.list, packs.nr + 1, packs.alloc_list); | |
769 | ALLOC_GROW(packs.names, packs.nr + 1, packs.alloc_names); | |
770 | ||
771 | packs.list[packs.nr] = NULL; | |
772 | packs.names[packs.nr] = xstrdup(packs.m->pack_names[i]); | |
773 | packs.pack_name_concat_len += strlen(packs.names[packs.nr]) + 1; | |
774 | packs.nr++; | |
775 | } | |
776 | } | |
777 | ||
396f2570 DS |
778 | for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs); |
779 | ||
a40498a1 DS |
780 | if (packs.m && packs.nr == packs.m->num_packs) |
781 | goto cleanup; | |
782 | ||
32f3c541 DS |
783 | if (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT) |
784 | packs.pack_name_concat_len += MIDX_CHUNK_ALIGNMENT - | |
785 | (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT); | |
786 | ||
787 | ALLOC_ARRAY(pack_perm, packs.nr); | |
788 | sort_packs_by_name(packs.names, packs.nr, pack_perm); | |
789 | ||
a40498a1 DS |
790 | entries = get_sorted_entries(packs.m, packs.list, pack_perm, packs.nr, &nr_entries); |
791 | ||
662148c4 DS |
792 | for (i = 0; i < nr_entries; i++) { |
793 | if (entries[i].offset > 0x7fffffff) | |
794 | num_large_offsets++; | |
795 | if (entries[i].offset > 0xffffffff) | |
796 | large_offsets_needed = 1; | |
797 | } | |
fe1ed56f | 798 | |
fc59e748 DS |
799 | hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR); |
800 | f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf); | |
801 | FREE_AND_NULL(midx_name); | |
802 | ||
a40498a1 DS |
803 | if (packs.m) |
804 | close_midx(packs.m); | |
805 | ||
32f3c541 | 806 | cur_chunk = 0; |
662148c4 | 807 | num_chunks = large_offsets_needed ? 5 : 4; |
32f3c541 DS |
808 | |
809 | written = write_midx_header(f, num_chunks, packs.nr); | |
810 | ||
811 | chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES; | |
812 | chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH; | |
813 | ||
814 | cur_chunk++; | |
d7cacf29 | 815 | chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT; |
32f3c541 DS |
816 | chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + packs.pack_name_concat_len; |
817 | ||
d7cacf29 DS |
818 | cur_chunk++; |
819 | chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP; | |
820 | chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE; | |
821 | ||
0d5b3a5e | 822 | cur_chunk++; |
662148c4 | 823 | chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS; |
0d5b3a5e DS |
824 | chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_HASH_LEN; |
825 | ||
662148c4 DS |
826 | cur_chunk++; |
827 | chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH; | |
828 | if (large_offsets_needed) { | |
829 | chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS; | |
830 | ||
831 | cur_chunk++; | |
832 | chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + | |
833 | num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH; | |
834 | } | |
835 | ||
836 | chunk_ids[cur_chunk] = 0; | |
837 | ||
32f3c541 DS |
838 | for (i = 0; i <= num_chunks; i++) { |
839 | if (i && chunk_offsets[i] < chunk_offsets[i - 1]) | |
840 | BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64, | |
841 | chunk_offsets[i - 1], | |
842 | chunk_offsets[i]); | |
843 | ||
844 | if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT) | |
845 | BUG("chunk offset %"PRIu64" is not properly aligned", | |
846 | chunk_offsets[i]); | |
847 | ||
848 | hashwrite_be32(f, chunk_ids[i]); | |
849 | hashwrite_be32(f, chunk_offsets[i] >> 32); | |
850 | hashwrite_be32(f, chunk_offsets[i]); | |
851 | ||
852 | written += MIDX_CHUNKLOOKUP_WIDTH; | |
853 | } | |
854 | ||
855 | for (i = 0; i < num_chunks; i++) { | |
856 | if (written != chunk_offsets[i]) | |
857 | BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32, | |
858 | chunk_offsets[i], | |
859 | written, | |
860 | chunk_ids[i]); | |
861 | ||
862 | switch (chunk_ids[i]) { | |
863 | case MIDX_CHUNKID_PACKNAMES: | |
864 | written += write_midx_pack_names(f, packs.names, packs.nr); | |
865 | break; | |
866 | ||
d7cacf29 DS |
867 | case MIDX_CHUNKID_OIDFANOUT: |
868 | written += write_midx_oid_fanout(f, entries, nr_entries); | |
869 | break; | |
870 | ||
0d5b3a5e DS |
871 | case MIDX_CHUNKID_OIDLOOKUP: |
872 | written += write_midx_oid_lookup(f, MIDX_HASH_LEN, entries, nr_entries); | |
873 | break; | |
874 | ||
662148c4 DS |
875 | case MIDX_CHUNKID_OBJECTOFFSETS: |
876 | written += write_midx_object_offsets(f, large_offsets_needed, entries, nr_entries); | |
877 | break; | |
878 | ||
879 | case MIDX_CHUNKID_LARGEOFFSETS: | |
880 | written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries); | |
881 | break; | |
882 | ||
32f3c541 DS |
883 | default: |
884 | BUG("trying to write unknown chunk id %"PRIx32, | |
885 | chunk_ids[i]); | |
886 | } | |
887 | } | |
888 | ||
889 | if (written != chunk_offsets[num_chunks]) | |
890 | BUG("incorrect final offset %"PRIu64" != %"PRIu64, | |
891 | written, | |
892 | chunk_offsets[num_chunks]); | |
fc59e748 DS |
893 | |
894 | finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM); | |
895 | commit_lock_file(&lk); | |
896 | ||
a40498a1 | 897 | cleanup: |
396f2570 DS |
898 | for (i = 0; i < packs.nr; i++) { |
899 | if (packs.list[i]) { | |
900 | close_pack(packs.list[i]); | |
901 | free(packs.list[i]); | |
902 | } | |
32f3c541 | 903 | free(packs.names[i]); |
396f2570 DS |
904 | } |
905 | ||
906 | free(packs.list); | |
32f3c541 | 907 | free(packs.names); |
fe1ed56f | 908 | free(entries); |
a40498a1 DS |
909 | free(pack_perm); |
910 | free(midx_name); | |
a3407730 DS |
911 | return 0; |
912 | } | |
525e18c0 DS |
913 | |
914 | void clear_midx_file(const char *object_dir) | |
915 | { | |
916 | char *midx = get_midx_filename(object_dir); | |
917 | ||
918 | if (remove_path(midx)) { | |
919 | UNLEAK(midx); | |
920 | die(_("failed to clear multi-pack-index at %s"), midx); | |
921 | } | |
922 | ||
923 | free(midx); | |
924 | } | |
56ee7ff1 DS |
925 | |
926 | static int verify_midx_error; | |
927 | ||
928 | int verify_midx_file(const char *object_dir) | |
929 | { | |
930 | struct multi_pack_index *m = load_multi_pack_index(object_dir, 1); | |
931 | verify_midx_error = 0; | |
932 | ||
933 | if (!m) | |
934 | return 0; | |
935 | ||
936 | return verify_midx_error; | |
937 | } |