1 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/module.h> 5 #include <linux/slab.h> 6 #include <asm/div64.h> 7 8 #include <linux/ceph/libceph.h> 9 #include <linux/ceph/osdmap.h> 10 #include <linux/ceph/decode.h> 11 #include <linux/crush/hash.h> 12 #include <linux/crush/mapper.h> 13 14 char *ceph_osdmap_state_str(char *str, int len, int state) 15 { 16 if (!len) 17 return str; 18 19 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP)) 20 snprintf(str, len, "exists, up"); 21 else if (state & CEPH_OSD_EXISTS) 22 snprintf(str, len, "exists"); 23 else if (state & CEPH_OSD_UP) 24 snprintf(str, len, "up"); 25 else 26 snprintf(str, len, "doesn't exist"); 27 28 return str; 29 } 30 31 /* maps */ 32 33 static int calc_bits_of(unsigned int t) 34 { 35 int b = 0; 36 while (t) { 37 t = t >> 1; 38 b++; 39 } 40 return b; 41 } 42 43 /* 44 * the foo_mask is the smallest value 2^n-1 that is >= foo. 45 */ 46 static void calc_pg_masks(struct ceph_pg_pool_info *pi) 47 { 48 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1; 49 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1; 50 } 51 52 /* 53 * decode crush map 54 */ 55 static int crush_decode_uniform_bucket(void **p, void *end, 56 struct crush_bucket_uniform *b) 57 { 58 dout("crush_decode_uniform_bucket %p to %p\n", *p, end); 59 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad); 60 b->item_weight = ceph_decode_32(p); 61 return 0; 62 bad: 63 return -EINVAL; 64 } 65 66 static int crush_decode_list_bucket(void **p, void *end, 67 struct crush_bucket_list *b) 68 { 69 int j; 70 dout("crush_decode_list_bucket %p to %p\n", *p, end); 71 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 72 if (b->item_weights == NULL) 73 return -ENOMEM; 74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 75 if (b->sum_weights == NULL) 76 return -ENOMEM; 77 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); 78 for (j = 0; j < b->h.size; j++) { 79 b->item_weights[j] = ceph_decode_32(p); 80 b->sum_weights[j] = ceph_decode_32(p); 81 } 82 return 0; 83 bad: 84 return -EINVAL; 85 } 86 87 static int crush_decode_tree_bucket(void **p, void *end, 88 struct crush_bucket_tree *b) 89 { 90 int j; 91 dout("crush_decode_tree_bucket %p to %p\n", *p, end); 92 ceph_decode_8_safe(p, end, b->num_nodes, bad); 93 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS); 94 if (b->node_weights == NULL) 95 return -ENOMEM; 96 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad); 97 for (j = 0; j < b->num_nodes; j++) 98 b->node_weights[j] = ceph_decode_32(p); 99 return 0; 100 bad: 101 return -EINVAL; 102 } 103 104 static int crush_decode_straw_bucket(void **p, void *end, 105 struct crush_bucket_straw *b) 106 { 107 int j; 108 dout("crush_decode_straw_bucket %p to %p\n", *p, end); 109 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 110 if (b->item_weights == NULL) 111 return -ENOMEM; 112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 113 if (b->straws == NULL) 114 return -ENOMEM; 115 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); 116 for (j = 0; j < b->h.size; j++) { 117 b->item_weights[j] = ceph_decode_32(p); 118 b->straws[j] = ceph_decode_32(p); 119 } 120 return 0; 121 bad: 122 return -EINVAL; 123 } 124 125 static int crush_decode_straw2_bucket(void **p, void *end, 126 struct crush_bucket_straw2 *b) 127 { 128 int j; 129 dout("crush_decode_straw2_bucket %p to %p\n", *p, end); 130 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 131 if (b->item_weights == NULL) 132 return -ENOMEM; 133 ceph_decode_need(p, end, b->h.size * sizeof(u32), bad); 134 for (j = 0; j < b->h.size; j++) 135 b->item_weights[j] = ceph_decode_32(p); 136 return 0; 137 bad: 138 return -EINVAL; 139 } 140 141 static int skip_name_map(void **p, void *end) 142 { 143 int len; 144 ceph_decode_32_safe(p, end, len ,bad); 145 while (len--) { 146 int strlen; 147 *p += sizeof(u32); 148 ceph_decode_32_safe(p, end, strlen, bad); 149 *p += strlen; 150 } 151 return 0; 152 bad: 153 return -EINVAL; 154 } 155 156 static void crush_finalize(struct crush_map *c) 157 { 158 __s32 b; 159 160 /* Space for the array of pointers to per-bucket workspace */ 161 c->working_size = sizeof(struct crush_work) + 162 c->max_buckets * sizeof(struct crush_work_bucket *); 163 164 for (b = 0; b < c->max_buckets; b++) { 165 if (!c->buckets[b]) 166 continue; 167 168 switch (c->buckets[b]->alg) { 169 default: 170 /* 171 * The base case, permutation variables and 172 * the pointer to the permutation array. 173 */ 174 c->working_size += sizeof(struct crush_work_bucket); 175 break; 176 } 177 /* Every bucket has a permutation array. */ 178 c->working_size += c->buckets[b]->size * sizeof(__u32); 179 } 180 } 181 182 static struct crush_map *crush_decode(void *pbyval, void *end) 183 { 184 struct crush_map *c; 185 int err = -EINVAL; 186 int i, j; 187 void **p = &pbyval; 188 void *start = pbyval; 189 u32 magic; 190 u32 num_name_maps; 191 192 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p)); 193 194 c = kzalloc(sizeof(*c), GFP_NOFS); 195 if (c == NULL) 196 return ERR_PTR(-ENOMEM); 197 198 /* set tunables to default values */ 199 c->choose_local_tries = 2; 200 c->choose_local_fallback_tries = 5; 201 c->choose_total_tries = 19; 202 c->chooseleaf_descend_once = 0; 203 204 ceph_decode_need(p, end, 4*sizeof(u32), bad); 205 magic = ceph_decode_32(p); 206 if (magic != CRUSH_MAGIC) { 207 pr_err("crush_decode magic %x != current %x\n", 208 (unsigned int)magic, (unsigned int)CRUSH_MAGIC); 209 goto bad; 210 } 211 c->max_buckets = ceph_decode_32(p); 212 c->max_rules = ceph_decode_32(p); 213 c->max_devices = ceph_decode_32(p); 214 215 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS); 216 if (c->buckets == NULL) 217 goto badmem; 218 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS); 219 if (c->rules == NULL) 220 goto badmem; 221 222 /* buckets */ 223 for (i = 0; i < c->max_buckets; i++) { 224 int size = 0; 225 u32 alg; 226 struct crush_bucket *b; 227 228 ceph_decode_32_safe(p, end, alg, bad); 229 if (alg == 0) { 230 c->buckets[i] = NULL; 231 continue; 232 } 233 dout("crush_decode bucket %d off %x %p to %p\n", 234 i, (int)(*p-start), *p, end); 235 236 switch (alg) { 237 case CRUSH_BUCKET_UNIFORM: 238 size = sizeof(struct crush_bucket_uniform); 239 break; 240 case CRUSH_BUCKET_LIST: 241 size = sizeof(struct crush_bucket_list); 242 break; 243 case CRUSH_BUCKET_TREE: 244 size = sizeof(struct crush_bucket_tree); 245 break; 246 case CRUSH_BUCKET_STRAW: 247 size = sizeof(struct crush_bucket_straw); 248 break; 249 case CRUSH_BUCKET_STRAW2: 250 size = sizeof(struct crush_bucket_straw2); 251 break; 252 default: 253 err = -EINVAL; 254 goto bad; 255 } 256 BUG_ON(size == 0); 257 b = c->buckets[i] = kzalloc(size, GFP_NOFS); 258 if (b == NULL) 259 goto badmem; 260 261 ceph_decode_need(p, end, 4*sizeof(u32), bad); 262 b->id = ceph_decode_32(p); 263 b->type = ceph_decode_16(p); 264 b->alg = ceph_decode_8(p); 265 b->hash = ceph_decode_8(p); 266 b->weight = ceph_decode_32(p); 267 b->size = ceph_decode_32(p); 268 269 dout("crush_decode bucket size %d off %x %p to %p\n", 270 b->size, (int)(*p-start), *p, end); 271 272 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS); 273 if (b->items == NULL) 274 goto badmem; 275 276 ceph_decode_need(p, end, b->size*sizeof(u32), bad); 277 for (j = 0; j < b->size; j++) 278 b->items[j] = ceph_decode_32(p); 279 280 switch (b->alg) { 281 case CRUSH_BUCKET_UNIFORM: 282 err = crush_decode_uniform_bucket(p, end, 283 (struct crush_bucket_uniform *)b); 284 if (err < 0) 285 goto bad; 286 break; 287 case CRUSH_BUCKET_LIST: 288 err = crush_decode_list_bucket(p, end, 289 (struct crush_bucket_list *)b); 290 if (err < 0) 291 goto bad; 292 break; 293 case CRUSH_BUCKET_TREE: 294 err = crush_decode_tree_bucket(p, end, 295 (struct crush_bucket_tree *)b); 296 if (err < 0) 297 goto bad; 298 break; 299 case CRUSH_BUCKET_STRAW: 300 err = crush_decode_straw_bucket(p, end, 301 (struct crush_bucket_straw *)b); 302 if (err < 0) 303 goto bad; 304 break; 305 case CRUSH_BUCKET_STRAW2: 306 err = crush_decode_straw2_bucket(p, end, 307 (struct crush_bucket_straw2 *)b); 308 if (err < 0) 309 goto bad; 310 break; 311 } 312 } 313 314 /* rules */ 315 dout("rule vec is %p\n", c->rules); 316 for (i = 0; i < c->max_rules; i++) { 317 u32 yes; 318 struct crush_rule *r; 319 320 ceph_decode_32_safe(p, end, yes, bad); 321 if (!yes) { 322 dout("crush_decode NO rule %d off %x %p to %p\n", 323 i, (int)(*p-start), *p, end); 324 c->rules[i] = NULL; 325 continue; 326 } 327 328 dout("crush_decode rule %d off %x %p to %p\n", 329 i, (int)(*p-start), *p, end); 330 331 /* len */ 332 ceph_decode_32_safe(p, end, yes, bad); 333 #if BITS_PER_LONG == 32 334 err = -EINVAL; 335 if (yes > (ULONG_MAX - sizeof(*r)) 336 / sizeof(struct crush_rule_step)) 337 goto bad; 338 #endif 339 r = c->rules[i] = kmalloc(sizeof(*r) + 340 yes*sizeof(struct crush_rule_step), 341 GFP_NOFS); 342 if (r == NULL) 343 goto badmem; 344 dout(" rule %d is at %p\n", i, r); 345 r->len = yes; 346 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */ 347 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad); 348 for (j = 0; j < r->len; j++) { 349 r->steps[j].op = ceph_decode_32(p); 350 r->steps[j].arg1 = ceph_decode_32(p); 351 r->steps[j].arg2 = ceph_decode_32(p); 352 } 353 } 354 355 /* ignore trailing name maps. */ 356 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) { 357 err = skip_name_map(p, end); 358 if (err < 0) 359 goto done; 360 } 361 362 /* tunables */ 363 ceph_decode_need(p, end, 3*sizeof(u32), done); 364 c->choose_local_tries = ceph_decode_32(p); 365 c->choose_local_fallback_tries = ceph_decode_32(p); 366 c->choose_total_tries = ceph_decode_32(p); 367 dout("crush decode tunable choose_local_tries = %d\n", 368 c->choose_local_tries); 369 dout("crush decode tunable choose_local_fallback_tries = %d\n", 370 c->choose_local_fallback_tries); 371 dout("crush decode tunable choose_total_tries = %d\n", 372 c->choose_total_tries); 373 374 ceph_decode_need(p, end, sizeof(u32), done); 375 c->chooseleaf_descend_once = ceph_decode_32(p); 376 dout("crush decode tunable chooseleaf_descend_once = %d\n", 377 c->chooseleaf_descend_once); 378 379 ceph_decode_need(p, end, sizeof(u8), done); 380 c->chooseleaf_vary_r = ceph_decode_8(p); 381 dout("crush decode tunable chooseleaf_vary_r = %d\n", 382 c->chooseleaf_vary_r); 383 384 /* skip straw_calc_version, allowed_bucket_algs */ 385 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done); 386 *p += sizeof(u8) + sizeof(u32); 387 388 ceph_decode_need(p, end, sizeof(u8), done); 389 c->chooseleaf_stable = ceph_decode_8(p); 390 dout("crush decode tunable chooseleaf_stable = %d\n", 391 c->chooseleaf_stable); 392 393 crush_finalize(c); 394 395 done: 396 dout("crush_decode success\n"); 397 return c; 398 399 badmem: 400 err = -ENOMEM; 401 bad: 402 dout("crush_decode fail %d\n", err); 403 crush_destroy(c); 404 return ERR_PTR(err); 405 } 406 407 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs) 408 { 409 if (lhs->pool < rhs->pool) 410 return -1; 411 if (lhs->pool > rhs->pool) 412 return 1; 413 if (lhs->seed < rhs->seed) 414 return -1; 415 if (lhs->seed > rhs->seed) 416 return 1; 417 418 return 0; 419 } 420 421 /* 422 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid 423 * to a set of osds) and primary_temp (explicit primary setting) 424 */ 425 static int __insert_pg_mapping(struct ceph_pg_mapping *new, 426 struct rb_root *root) 427 { 428 struct rb_node **p = &root->rb_node; 429 struct rb_node *parent = NULL; 430 struct ceph_pg_mapping *pg = NULL; 431 int c; 432 433 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new); 434 while (*p) { 435 parent = *p; 436 pg = rb_entry(parent, struct ceph_pg_mapping, node); 437 c = ceph_pg_compare(&new->pgid, &pg->pgid); 438 if (c < 0) 439 p = &(*p)->rb_left; 440 else if (c > 0) 441 p = &(*p)->rb_right; 442 else 443 return -EEXIST; 444 } 445 446 rb_link_node(&new->node, parent, p); 447 rb_insert_color(&new->node, root); 448 return 0; 449 } 450 451 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root, 452 struct ceph_pg pgid) 453 { 454 struct rb_node *n = root->rb_node; 455 struct ceph_pg_mapping *pg; 456 int c; 457 458 while (n) { 459 pg = rb_entry(n, struct ceph_pg_mapping, node); 460 c = ceph_pg_compare(&pgid, &pg->pgid); 461 if (c < 0) { 462 n = n->rb_left; 463 } else if (c > 0) { 464 n = n->rb_right; 465 } else { 466 dout("__lookup_pg_mapping %lld.%x got %p\n", 467 pgid.pool, pgid.seed, pg); 468 return pg; 469 } 470 } 471 return NULL; 472 } 473 474 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid) 475 { 476 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid); 477 478 if (pg) { 479 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed, 480 pg); 481 rb_erase(&pg->node, root); 482 kfree(pg); 483 return 0; 484 } 485 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed); 486 return -ENOENT; 487 } 488 489 /* 490 * rbtree of pg pool info 491 */ 492 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new) 493 { 494 struct rb_node **p = &root->rb_node; 495 struct rb_node *parent = NULL; 496 struct ceph_pg_pool_info *pi = NULL; 497 498 while (*p) { 499 parent = *p; 500 pi = rb_entry(parent, struct ceph_pg_pool_info, node); 501 if (new->id < pi->id) 502 p = &(*p)->rb_left; 503 else if (new->id > pi->id) 504 p = &(*p)->rb_right; 505 else 506 return -EEXIST; 507 } 508 509 rb_link_node(&new->node, parent, p); 510 rb_insert_color(&new->node, root); 511 return 0; 512 } 513 514 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id) 515 { 516 struct ceph_pg_pool_info *pi; 517 struct rb_node *n = root->rb_node; 518 519 while (n) { 520 pi = rb_entry(n, struct ceph_pg_pool_info, node); 521 if (id < pi->id) 522 n = n->rb_left; 523 else if (id > pi->id) 524 n = n->rb_right; 525 else 526 return pi; 527 } 528 return NULL; 529 } 530 531 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id) 532 { 533 return __lookup_pg_pool(&map->pg_pools, id); 534 } 535 536 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id) 537 { 538 struct ceph_pg_pool_info *pi; 539 540 if (id == CEPH_NOPOOL) 541 return NULL; 542 543 if (WARN_ON_ONCE(id > (u64) INT_MAX)) 544 return NULL; 545 546 pi = __lookup_pg_pool(&map->pg_pools, (int) id); 547 548 return pi ? pi->name : NULL; 549 } 550 EXPORT_SYMBOL(ceph_pg_pool_name_by_id); 551 552 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name) 553 { 554 struct rb_node *rbp; 555 556 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) { 557 struct ceph_pg_pool_info *pi = 558 rb_entry(rbp, struct ceph_pg_pool_info, node); 559 if (pi->name && strcmp(pi->name, name) == 0) 560 return pi->id; 561 } 562 return -ENOENT; 563 } 564 EXPORT_SYMBOL(ceph_pg_poolid_by_name); 565 566 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi) 567 { 568 rb_erase(&pi->node, root); 569 kfree(pi->name); 570 kfree(pi); 571 } 572 573 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi) 574 { 575 u8 ev, cv; 576 unsigned len, num; 577 void *pool_end; 578 579 ceph_decode_need(p, end, 2 + 4, bad); 580 ev = ceph_decode_8(p); /* encoding version */ 581 cv = ceph_decode_8(p); /* compat version */ 582 if (ev < 5) { 583 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv); 584 return -EINVAL; 585 } 586 if (cv > 9) { 587 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv); 588 return -EINVAL; 589 } 590 len = ceph_decode_32(p); 591 ceph_decode_need(p, end, len, bad); 592 pool_end = *p + len; 593 594 pi->type = ceph_decode_8(p); 595 pi->size = ceph_decode_8(p); 596 pi->crush_ruleset = ceph_decode_8(p); 597 pi->object_hash = ceph_decode_8(p); 598 599 pi->pg_num = ceph_decode_32(p); 600 pi->pgp_num = ceph_decode_32(p); 601 602 *p += 4 + 4; /* skip lpg* */ 603 *p += 4; /* skip last_change */ 604 *p += 8 + 4; /* skip snap_seq, snap_epoch */ 605 606 /* skip snaps */ 607 num = ceph_decode_32(p); 608 while (num--) { 609 *p += 8; /* snapid key */ 610 *p += 1 + 1; /* versions */ 611 len = ceph_decode_32(p); 612 *p += len; 613 } 614 615 /* skip removed_snaps */ 616 num = ceph_decode_32(p); 617 *p += num * (8 + 8); 618 619 *p += 8; /* skip auid */ 620 pi->flags = ceph_decode_64(p); 621 *p += 4; /* skip crash_replay_interval */ 622 623 if (ev >= 7) 624 pi->min_size = ceph_decode_8(p); 625 else 626 pi->min_size = pi->size - pi->size / 2; 627 628 if (ev >= 8) 629 *p += 8 + 8; /* skip quota_max_* */ 630 631 if (ev >= 9) { 632 /* skip tiers */ 633 num = ceph_decode_32(p); 634 *p += num * 8; 635 636 *p += 8; /* skip tier_of */ 637 *p += 1; /* skip cache_mode */ 638 639 pi->read_tier = ceph_decode_64(p); 640 pi->write_tier = ceph_decode_64(p); 641 } else { 642 pi->read_tier = -1; 643 pi->write_tier = -1; 644 } 645 646 if (ev >= 10) { 647 /* skip properties */ 648 num = ceph_decode_32(p); 649 while (num--) { 650 len = ceph_decode_32(p); 651 *p += len; /* key */ 652 len = ceph_decode_32(p); 653 *p += len; /* val */ 654 } 655 } 656 657 if (ev >= 11) { 658 /* skip hit_set_params */ 659 *p += 1 + 1; /* versions */ 660 len = ceph_decode_32(p); 661 *p += len; 662 663 *p += 4; /* skip hit_set_period */ 664 *p += 4; /* skip hit_set_count */ 665 } 666 667 if (ev >= 12) 668 *p += 4; /* skip stripe_width */ 669 670 if (ev >= 13) { 671 *p += 8; /* skip target_max_bytes */ 672 *p += 8; /* skip target_max_objects */ 673 *p += 4; /* skip cache_target_dirty_ratio_micro */ 674 *p += 4; /* skip cache_target_full_ratio_micro */ 675 *p += 4; /* skip cache_min_flush_age */ 676 *p += 4; /* skip cache_min_evict_age */ 677 } 678 679 if (ev >= 14) { 680 /* skip erasure_code_profile */ 681 len = ceph_decode_32(p); 682 *p += len; 683 } 684 685 if (ev >= 15) 686 pi->last_force_request_resend = ceph_decode_32(p); 687 else 688 pi->last_force_request_resend = 0; 689 690 /* ignore the rest */ 691 692 *p = pool_end; 693 calc_pg_masks(pi); 694 return 0; 695 696 bad: 697 return -EINVAL; 698 } 699 700 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map) 701 { 702 struct ceph_pg_pool_info *pi; 703 u32 num, len; 704 u64 pool; 705 706 ceph_decode_32_safe(p, end, num, bad); 707 dout(" %d pool names\n", num); 708 while (num--) { 709 ceph_decode_64_safe(p, end, pool, bad); 710 ceph_decode_32_safe(p, end, len, bad); 711 dout(" pool %llu len %d\n", pool, len); 712 ceph_decode_need(p, end, len, bad); 713 pi = __lookup_pg_pool(&map->pg_pools, pool); 714 if (pi) { 715 char *name = kstrndup(*p, len, GFP_NOFS); 716 717 if (!name) 718 return -ENOMEM; 719 kfree(pi->name); 720 pi->name = name; 721 dout(" name is %s\n", pi->name); 722 } 723 *p += len; 724 } 725 return 0; 726 727 bad: 728 return -EINVAL; 729 } 730 731 /* 732 * osd map 733 */ 734 struct ceph_osdmap *ceph_osdmap_alloc(void) 735 { 736 struct ceph_osdmap *map; 737 738 map = kzalloc(sizeof(*map), GFP_NOIO); 739 if (!map) 740 return NULL; 741 742 map->pg_pools = RB_ROOT; 743 map->pool_max = -1; 744 map->pg_temp = RB_ROOT; 745 map->primary_temp = RB_ROOT; 746 mutex_init(&map->crush_scratch_mutex); 747 748 return map; 749 } 750 751 void ceph_osdmap_destroy(struct ceph_osdmap *map) 752 { 753 dout("osdmap_destroy %p\n", map); 754 if (map->crush) 755 crush_destroy(map->crush); 756 while (!RB_EMPTY_ROOT(&map->pg_temp)) { 757 struct ceph_pg_mapping *pg = 758 rb_entry(rb_first(&map->pg_temp), 759 struct ceph_pg_mapping, node); 760 rb_erase(&pg->node, &map->pg_temp); 761 kfree(pg); 762 } 763 while (!RB_EMPTY_ROOT(&map->primary_temp)) { 764 struct ceph_pg_mapping *pg = 765 rb_entry(rb_first(&map->primary_temp), 766 struct ceph_pg_mapping, node); 767 rb_erase(&pg->node, &map->primary_temp); 768 kfree(pg); 769 } 770 while (!RB_EMPTY_ROOT(&map->pg_pools)) { 771 struct ceph_pg_pool_info *pi = 772 rb_entry(rb_first(&map->pg_pools), 773 struct ceph_pg_pool_info, node); 774 __remove_pg_pool(&map->pg_pools, pi); 775 } 776 kfree(map->osd_state); 777 kfree(map->osd_weight); 778 kfree(map->osd_addr); 779 kfree(map->osd_primary_affinity); 780 kfree(map->crush_workspace); 781 kfree(map); 782 } 783 784 /* 785 * Adjust max_osd value, (re)allocate arrays. 786 * 787 * The new elements are properly initialized. 788 */ 789 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max) 790 { 791 u8 *state; 792 u32 *weight; 793 struct ceph_entity_addr *addr; 794 int i; 795 796 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS); 797 if (!state) 798 return -ENOMEM; 799 map->osd_state = state; 800 801 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS); 802 if (!weight) 803 return -ENOMEM; 804 map->osd_weight = weight; 805 806 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS); 807 if (!addr) 808 return -ENOMEM; 809 map->osd_addr = addr; 810 811 for (i = map->max_osd; i < max; i++) { 812 map->osd_state[i] = 0; 813 map->osd_weight[i] = CEPH_OSD_OUT; 814 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr)); 815 } 816 817 if (map->osd_primary_affinity) { 818 u32 *affinity; 819 820 affinity = krealloc(map->osd_primary_affinity, 821 max*sizeof(*affinity), GFP_NOFS); 822 if (!affinity) 823 return -ENOMEM; 824 map->osd_primary_affinity = affinity; 825 826 for (i = map->max_osd; i < max; i++) 827 map->osd_primary_affinity[i] = 828 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 829 } 830 831 map->max_osd = max; 832 833 return 0; 834 } 835 836 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush) 837 { 838 void *workspace; 839 840 if (IS_ERR(crush)) 841 return PTR_ERR(crush); 842 843 workspace = kmalloc(crush->working_size, GFP_NOIO); 844 if (!workspace) { 845 crush_destroy(crush); 846 return -ENOMEM; 847 } 848 crush_init_workspace(crush, workspace); 849 850 if (map->crush) 851 crush_destroy(map->crush); 852 kfree(map->crush_workspace); 853 map->crush = crush; 854 map->crush_workspace = workspace; 855 return 0; 856 } 857 858 #define OSDMAP_WRAPPER_COMPAT_VER 7 859 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1 860 861 /* 862 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps, 863 * to struct_v of the client_data section for new (v7 and above) 864 * osdmaps. 865 */ 866 static int get_osdmap_client_data_v(void **p, void *end, 867 const char *prefix, u8 *v) 868 { 869 u8 struct_v; 870 871 ceph_decode_8_safe(p, end, struct_v, e_inval); 872 if (struct_v >= 7) { 873 u8 struct_compat; 874 875 ceph_decode_8_safe(p, end, struct_compat, e_inval); 876 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) { 877 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n", 878 struct_v, struct_compat, 879 OSDMAP_WRAPPER_COMPAT_VER, prefix); 880 return -EINVAL; 881 } 882 *p += 4; /* ignore wrapper struct_len */ 883 884 ceph_decode_8_safe(p, end, struct_v, e_inval); 885 ceph_decode_8_safe(p, end, struct_compat, e_inval); 886 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) { 887 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n", 888 struct_v, struct_compat, 889 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix); 890 return -EINVAL; 891 } 892 *p += 4; /* ignore client data struct_len */ 893 } else { 894 u16 version; 895 896 *p -= 1; 897 ceph_decode_16_safe(p, end, version, e_inval); 898 if (version < 6) { 899 pr_warn("got v %d < 6 of %s ceph_osdmap\n", 900 version, prefix); 901 return -EINVAL; 902 } 903 904 /* old osdmap enconding */ 905 struct_v = 0; 906 } 907 908 *v = struct_v; 909 return 0; 910 911 e_inval: 912 return -EINVAL; 913 } 914 915 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map, 916 bool incremental) 917 { 918 u32 n; 919 920 ceph_decode_32_safe(p, end, n, e_inval); 921 while (n--) { 922 struct ceph_pg_pool_info *pi; 923 u64 pool; 924 int ret; 925 926 ceph_decode_64_safe(p, end, pool, e_inval); 927 928 pi = __lookup_pg_pool(&map->pg_pools, pool); 929 if (!incremental || !pi) { 930 pi = kzalloc(sizeof(*pi), GFP_NOFS); 931 if (!pi) 932 return -ENOMEM; 933 934 pi->id = pool; 935 936 ret = __insert_pg_pool(&map->pg_pools, pi); 937 if (ret) { 938 kfree(pi); 939 return ret; 940 } 941 } 942 943 ret = decode_pool(p, end, pi); 944 if (ret) 945 return ret; 946 } 947 948 return 0; 949 950 e_inval: 951 return -EINVAL; 952 } 953 954 static int decode_pools(void **p, void *end, struct ceph_osdmap *map) 955 { 956 return __decode_pools(p, end, map, false); 957 } 958 959 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map) 960 { 961 return __decode_pools(p, end, map, true); 962 } 963 964 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map, 965 bool incremental) 966 { 967 u32 n; 968 969 ceph_decode_32_safe(p, end, n, e_inval); 970 while (n--) { 971 struct ceph_pg pgid; 972 u32 len, i; 973 int ret; 974 975 ret = ceph_decode_pgid(p, end, &pgid); 976 if (ret) 977 return ret; 978 979 ceph_decode_32_safe(p, end, len, e_inval); 980 981 ret = __remove_pg_mapping(&map->pg_temp, pgid); 982 BUG_ON(!incremental && ret != -ENOENT); 983 984 if (!incremental || len > 0) { 985 struct ceph_pg_mapping *pg; 986 987 ceph_decode_need(p, end, len*sizeof(u32), e_inval); 988 989 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32)) 990 return -EINVAL; 991 992 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS); 993 if (!pg) 994 return -ENOMEM; 995 996 pg->pgid = pgid; 997 pg->pg_temp.len = len; 998 for (i = 0; i < len; i++) 999 pg->pg_temp.osds[i] = ceph_decode_32(p); 1000 1001 ret = __insert_pg_mapping(pg, &map->pg_temp); 1002 if (ret) { 1003 kfree(pg); 1004 return ret; 1005 } 1006 } 1007 } 1008 1009 return 0; 1010 1011 e_inval: 1012 return -EINVAL; 1013 } 1014 1015 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map) 1016 { 1017 return __decode_pg_temp(p, end, map, false); 1018 } 1019 1020 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map) 1021 { 1022 return __decode_pg_temp(p, end, map, true); 1023 } 1024 1025 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map, 1026 bool incremental) 1027 { 1028 u32 n; 1029 1030 ceph_decode_32_safe(p, end, n, e_inval); 1031 while (n--) { 1032 struct ceph_pg pgid; 1033 u32 osd; 1034 int ret; 1035 1036 ret = ceph_decode_pgid(p, end, &pgid); 1037 if (ret) 1038 return ret; 1039 1040 ceph_decode_32_safe(p, end, osd, e_inval); 1041 1042 ret = __remove_pg_mapping(&map->primary_temp, pgid); 1043 BUG_ON(!incremental && ret != -ENOENT); 1044 1045 if (!incremental || osd != (u32)-1) { 1046 struct ceph_pg_mapping *pg; 1047 1048 pg = kzalloc(sizeof(*pg), GFP_NOFS); 1049 if (!pg) 1050 return -ENOMEM; 1051 1052 pg->pgid = pgid; 1053 pg->primary_temp.osd = osd; 1054 1055 ret = __insert_pg_mapping(pg, &map->primary_temp); 1056 if (ret) { 1057 kfree(pg); 1058 return ret; 1059 } 1060 } 1061 } 1062 1063 return 0; 1064 1065 e_inval: 1066 return -EINVAL; 1067 } 1068 1069 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map) 1070 { 1071 return __decode_primary_temp(p, end, map, false); 1072 } 1073 1074 static int decode_new_primary_temp(void **p, void *end, 1075 struct ceph_osdmap *map) 1076 { 1077 return __decode_primary_temp(p, end, map, true); 1078 } 1079 1080 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd) 1081 { 1082 BUG_ON(osd >= map->max_osd); 1083 1084 if (!map->osd_primary_affinity) 1085 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1086 1087 return map->osd_primary_affinity[osd]; 1088 } 1089 1090 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff) 1091 { 1092 BUG_ON(osd >= map->max_osd); 1093 1094 if (!map->osd_primary_affinity) { 1095 int i; 1096 1097 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32), 1098 GFP_NOFS); 1099 if (!map->osd_primary_affinity) 1100 return -ENOMEM; 1101 1102 for (i = 0; i < map->max_osd; i++) 1103 map->osd_primary_affinity[i] = 1104 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1105 } 1106 1107 map->osd_primary_affinity[osd] = aff; 1108 1109 return 0; 1110 } 1111 1112 static int decode_primary_affinity(void **p, void *end, 1113 struct ceph_osdmap *map) 1114 { 1115 u32 len, i; 1116 1117 ceph_decode_32_safe(p, end, len, e_inval); 1118 if (len == 0) { 1119 kfree(map->osd_primary_affinity); 1120 map->osd_primary_affinity = NULL; 1121 return 0; 1122 } 1123 if (len != map->max_osd) 1124 goto e_inval; 1125 1126 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval); 1127 1128 for (i = 0; i < map->max_osd; i++) { 1129 int ret; 1130 1131 ret = set_primary_affinity(map, i, ceph_decode_32(p)); 1132 if (ret) 1133 return ret; 1134 } 1135 1136 return 0; 1137 1138 e_inval: 1139 return -EINVAL; 1140 } 1141 1142 static int decode_new_primary_affinity(void **p, void *end, 1143 struct ceph_osdmap *map) 1144 { 1145 u32 n; 1146 1147 ceph_decode_32_safe(p, end, n, e_inval); 1148 while (n--) { 1149 u32 osd, aff; 1150 int ret; 1151 1152 ceph_decode_32_safe(p, end, osd, e_inval); 1153 ceph_decode_32_safe(p, end, aff, e_inval); 1154 1155 ret = set_primary_affinity(map, osd, aff); 1156 if (ret) 1157 return ret; 1158 1159 pr_info("osd%d primary-affinity 0x%x\n", osd, aff); 1160 } 1161 1162 return 0; 1163 1164 e_inval: 1165 return -EINVAL; 1166 } 1167 1168 /* 1169 * decode a full map. 1170 */ 1171 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map) 1172 { 1173 u8 struct_v; 1174 u32 epoch = 0; 1175 void *start = *p; 1176 u32 max; 1177 u32 len, i; 1178 int err; 1179 1180 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1181 1182 err = get_osdmap_client_data_v(p, end, "full", &struct_v); 1183 if (err) 1184 goto bad; 1185 1186 /* fsid, epoch, created, modified */ 1187 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) + 1188 sizeof(map->created) + sizeof(map->modified), e_inval); 1189 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid)); 1190 epoch = map->epoch = ceph_decode_32(p); 1191 ceph_decode_copy(p, &map->created, sizeof(map->created)); 1192 ceph_decode_copy(p, &map->modified, sizeof(map->modified)); 1193 1194 /* pools */ 1195 err = decode_pools(p, end, map); 1196 if (err) 1197 goto bad; 1198 1199 /* pool_name */ 1200 err = decode_pool_names(p, end, map); 1201 if (err) 1202 goto bad; 1203 1204 ceph_decode_32_safe(p, end, map->pool_max, e_inval); 1205 1206 ceph_decode_32_safe(p, end, map->flags, e_inval); 1207 1208 /* max_osd */ 1209 ceph_decode_32_safe(p, end, max, e_inval); 1210 1211 /* (re)alloc osd arrays */ 1212 err = osdmap_set_max_osd(map, max); 1213 if (err) 1214 goto bad; 1215 1216 /* osd_state, osd_weight, osd_addrs->client_addr */ 1217 ceph_decode_need(p, end, 3*sizeof(u32) + 1218 map->max_osd*(1 + sizeof(*map->osd_weight) + 1219 sizeof(*map->osd_addr)), e_inval); 1220 1221 if (ceph_decode_32(p) != map->max_osd) 1222 goto e_inval; 1223 1224 ceph_decode_copy(p, map->osd_state, map->max_osd); 1225 1226 if (ceph_decode_32(p) != map->max_osd) 1227 goto e_inval; 1228 1229 for (i = 0; i < map->max_osd; i++) 1230 map->osd_weight[i] = ceph_decode_32(p); 1231 1232 if (ceph_decode_32(p) != map->max_osd) 1233 goto e_inval; 1234 1235 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr)); 1236 for (i = 0; i < map->max_osd; i++) 1237 ceph_decode_addr(&map->osd_addr[i]); 1238 1239 /* pg_temp */ 1240 err = decode_pg_temp(p, end, map); 1241 if (err) 1242 goto bad; 1243 1244 /* primary_temp */ 1245 if (struct_v >= 1) { 1246 err = decode_primary_temp(p, end, map); 1247 if (err) 1248 goto bad; 1249 } 1250 1251 /* primary_affinity */ 1252 if (struct_v >= 2) { 1253 err = decode_primary_affinity(p, end, map); 1254 if (err) 1255 goto bad; 1256 } else { 1257 /* XXX can this happen? */ 1258 kfree(map->osd_primary_affinity); 1259 map->osd_primary_affinity = NULL; 1260 } 1261 1262 /* crush */ 1263 ceph_decode_32_safe(p, end, len, e_inval); 1264 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end))); 1265 if (err) 1266 goto bad; 1267 1268 /* ignore the rest */ 1269 *p = end; 1270 1271 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1272 return 0; 1273 1274 e_inval: 1275 err = -EINVAL; 1276 bad: 1277 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1278 err, epoch, (int)(*p - start), *p, start, end); 1279 print_hex_dump(KERN_DEBUG, "osdmap: ", 1280 DUMP_PREFIX_OFFSET, 16, 1, 1281 start, end - start, true); 1282 return err; 1283 } 1284 1285 /* 1286 * Allocate and decode a full map. 1287 */ 1288 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end) 1289 { 1290 struct ceph_osdmap *map; 1291 int ret; 1292 1293 map = ceph_osdmap_alloc(); 1294 if (!map) 1295 return ERR_PTR(-ENOMEM); 1296 1297 ret = osdmap_decode(p, end, map); 1298 if (ret) { 1299 ceph_osdmap_destroy(map); 1300 return ERR_PTR(ret); 1301 } 1302 1303 return map; 1304 } 1305 1306 /* 1307 * Encoding order is (new_up_client, new_state, new_weight). Need to 1308 * apply in the (new_weight, new_state, new_up_client) order, because 1309 * an incremental map may look like e.g. 1310 * 1311 * new_up_client: { osd=6, addr=... } # set osd_state and addr 1312 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state 1313 */ 1314 static int decode_new_up_state_weight(void **p, void *end, 1315 struct ceph_osdmap *map) 1316 { 1317 void *new_up_client; 1318 void *new_state; 1319 void *new_weight_end; 1320 u32 len; 1321 1322 new_up_client = *p; 1323 ceph_decode_32_safe(p, end, len, e_inval); 1324 len *= sizeof(u32) + sizeof(struct ceph_entity_addr); 1325 ceph_decode_need(p, end, len, e_inval); 1326 *p += len; 1327 1328 new_state = *p; 1329 ceph_decode_32_safe(p, end, len, e_inval); 1330 len *= sizeof(u32) + sizeof(u8); 1331 ceph_decode_need(p, end, len, e_inval); 1332 *p += len; 1333 1334 /* new_weight */ 1335 ceph_decode_32_safe(p, end, len, e_inval); 1336 while (len--) { 1337 s32 osd; 1338 u32 w; 1339 1340 ceph_decode_need(p, end, 2*sizeof(u32), e_inval); 1341 osd = ceph_decode_32(p); 1342 w = ceph_decode_32(p); 1343 BUG_ON(osd >= map->max_osd); 1344 pr_info("osd%d weight 0x%x %s\n", osd, w, 1345 w == CEPH_OSD_IN ? "(in)" : 1346 (w == CEPH_OSD_OUT ? "(out)" : "")); 1347 map->osd_weight[osd] = w; 1348 1349 /* 1350 * If we are marking in, set the EXISTS, and clear the 1351 * AUTOOUT and NEW bits. 1352 */ 1353 if (w) { 1354 map->osd_state[osd] |= CEPH_OSD_EXISTS; 1355 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT | 1356 CEPH_OSD_NEW); 1357 } 1358 } 1359 new_weight_end = *p; 1360 1361 /* new_state (up/down) */ 1362 *p = new_state; 1363 len = ceph_decode_32(p); 1364 while (len--) { 1365 s32 osd; 1366 u8 xorstate; 1367 int ret; 1368 1369 osd = ceph_decode_32(p); 1370 xorstate = ceph_decode_8(p); 1371 if (xorstate == 0) 1372 xorstate = CEPH_OSD_UP; 1373 BUG_ON(osd >= map->max_osd); 1374 if ((map->osd_state[osd] & CEPH_OSD_UP) && 1375 (xorstate & CEPH_OSD_UP)) 1376 pr_info("osd%d down\n", osd); 1377 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) && 1378 (xorstate & CEPH_OSD_EXISTS)) { 1379 pr_info("osd%d does not exist\n", osd); 1380 map->osd_weight[osd] = CEPH_OSD_IN; 1381 ret = set_primary_affinity(map, osd, 1382 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY); 1383 if (ret) 1384 return ret; 1385 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr)); 1386 map->osd_state[osd] = 0; 1387 } else { 1388 map->osd_state[osd] ^= xorstate; 1389 } 1390 } 1391 1392 /* new_up_client */ 1393 *p = new_up_client; 1394 len = ceph_decode_32(p); 1395 while (len--) { 1396 s32 osd; 1397 struct ceph_entity_addr addr; 1398 1399 osd = ceph_decode_32(p); 1400 ceph_decode_copy(p, &addr, sizeof(addr)); 1401 ceph_decode_addr(&addr); 1402 BUG_ON(osd >= map->max_osd); 1403 pr_info("osd%d up\n", osd); 1404 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP; 1405 map->osd_addr[osd] = addr; 1406 } 1407 1408 *p = new_weight_end; 1409 return 0; 1410 1411 e_inval: 1412 return -EINVAL; 1413 } 1414 1415 /* 1416 * decode and apply an incremental map update. 1417 */ 1418 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 1419 struct ceph_osdmap *map) 1420 { 1421 struct ceph_fsid fsid; 1422 u32 epoch = 0; 1423 struct ceph_timespec modified; 1424 s32 len; 1425 u64 pool; 1426 __s64 new_pool_max; 1427 __s32 new_flags, max; 1428 void *start = *p; 1429 int err; 1430 u8 struct_v; 1431 1432 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1433 1434 err = get_osdmap_client_data_v(p, end, "inc", &struct_v); 1435 if (err) 1436 goto bad; 1437 1438 /* fsid, epoch, modified, new_pool_max, new_flags */ 1439 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) + 1440 sizeof(u64) + sizeof(u32), e_inval); 1441 ceph_decode_copy(p, &fsid, sizeof(fsid)); 1442 epoch = ceph_decode_32(p); 1443 BUG_ON(epoch != map->epoch+1); 1444 ceph_decode_copy(p, &modified, sizeof(modified)); 1445 new_pool_max = ceph_decode_64(p); 1446 new_flags = ceph_decode_32(p); 1447 1448 /* full map? */ 1449 ceph_decode_32_safe(p, end, len, e_inval); 1450 if (len > 0) { 1451 dout("apply_incremental full map len %d, %p to %p\n", 1452 len, *p, end); 1453 return ceph_osdmap_decode(p, min(*p+len, end)); 1454 } 1455 1456 /* new crush? */ 1457 ceph_decode_32_safe(p, end, len, e_inval); 1458 if (len > 0) { 1459 err = osdmap_set_crush(map, 1460 crush_decode(*p, min(*p + len, end))); 1461 if (err) 1462 goto bad; 1463 *p += len; 1464 } 1465 1466 /* new flags? */ 1467 if (new_flags >= 0) 1468 map->flags = new_flags; 1469 if (new_pool_max >= 0) 1470 map->pool_max = new_pool_max; 1471 1472 /* new max? */ 1473 ceph_decode_32_safe(p, end, max, e_inval); 1474 if (max >= 0) { 1475 err = osdmap_set_max_osd(map, max); 1476 if (err) 1477 goto bad; 1478 } 1479 1480 map->epoch++; 1481 map->modified = modified; 1482 1483 /* new_pools */ 1484 err = decode_new_pools(p, end, map); 1485 if (err) 1486 goto bad; 1487 1488 /* new_pool_names */ 1489 err = decode_pool_names(p, end, map); 1490 if (err) 1491 goto bad; 1492 1493 /* old_pool */ 1494 ceph_decode_32_safe(p, end, len, e_inval); 1495 while (len--) { 1496 struct ceph_pg_pool_info *pi; 1497 1498 ceph_decode_64_safe(p, end, pool, e_inval); 1499 pi = __lookup_pg_pool(&map->pg_pools, pool); 1500 if (pi) 1501 __remove_pg_pool(&map->pg_pools, pi); 1502 } 1503 1504 /* new_up_client, new_state, new_weight */ 1505 err = decode_new_up_state_weight(p, end, map); 1506 if (err) 1507 goto bad; 1508 1509 /* new_pg_temp */ 1510 err = decode_new_pg_temp(p, end, map); 1511 if (err) 1512 goto bad; 1513 1514 /* new_primary_temp */ 1515 if (struct_v >= 1) { 1516 err = decode_new_primary_temp(p, end, map); 1517 if (err) 1518 goto bad; 1519 } 1520 1521 /* new_primary_affinity */ 1522 if (struct_v >= 2) { 1523 err = decode_new_primary_affinity(p, end, map); 1524 if (err) 1525 goto bad; 1526 } 1527 1528 /* ignore the rest */ 1529 *p = end; 1530 1531 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1532 return map; 1533 1534 e_inval: 1535 err = -EINVAL; 1536 bad: 1537 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1538 err, epoch, (int)(*p - start), *p, start, end); 1539 print_hex_dump(KERN_DEBUG, "osdmap: ", 1540 DUMP_PREFIX_OFFSET, 16, 1, 1541 start, end - start, true); 1542 return ERR_PTR(err); 1543 } 1544 1545 void ceph_oloc_copy(struct ceph_object_locator *dest, 1546 const struct ceph_object_locator *src) 1547 { 1548 WARN_ON(!ceph_oloc_empty(dest)); 1549 WARN_ON(dest->pool_ns); /* empty() only covers ->pool */ 1550 1551 dest->pool = src->pool; 1552 if (src->pool_ns) 1553 dest->pool_ns = ceph_get_string(src->pool_ns); 1554 } 1555 EXPORT_SYMBOL(ceph_oloc_copy); 1556 1557 void ceph_oloc_destroy(struct ceph_object_locator *oloc) 1558 { 1559 ceph_put_string(oloc->pool_ns); 1560 } 1561 EXPORT_SYMBOL(ceph_oloc_destroy); 1562 1563 void ceph_oid_copy(struct ceph_object_id *dest, 1564 const struct ceph_object_id *src) 1565 { 1566 WARN_ON(!ceph_oid_empty(dest)); 1567 1568 if (src->name != src->inline_name) { 1569 /* very rare, see ceph_object_id definition */ 1570 dest->name = kmalloc(src->name_len + 1, 1571 GFP_NOIO | __GFP_NOFAIL); 1572 } 1573 1574 memcpy(dest->name, src->name, src->name_len + 1); 1575 dest->name_len = src->name_len; 1576 } 1577 EXPORT_SYMBOL(ceph_oid_copy); 1578 1579 static __printf(2, 0) 1580 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap) 1581 { 1582 int len; 1583 1584 WARN_ON(!ceph_oid_empty(oid)); 1585 1586 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap); 1587 if (len >= sizeof(oid->inline_name)) 1588 return len; 1589 1590 oid->name_len = len; 1591 return 0; 1592 } 1593 1594 /* 1595 * If oid doesn't fit into inline buffer, BUG. 1596 */ 1597 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...) 1598 { 1599 va_list ap; 1600 1601 va_start(ap, fmt); 1602 BUG_ON(oid_printf_vargs(oid, fmt, ap)); 1603 va_end(ap); 1604 } 1605 EXPORT_SYMBOL(ceph_oid_printf); 1606 1607 static __printf(3, 0) 1608 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp, 1609 const char *fmt, va_list ap) 1610 { 1611 va_list aq; 1612 int len; 1613 1614 va_copy(aq, ap); 1615 len = oid_printf_vargs(oid, fmt, aq); 1616 va_end(aq); 1617 1618 if (len) { 1619 char *external_name; 1620 1621 external_name = kmalloc(len + 1, gfp); 1622 if (!external_name) 1623 return -ENOMEM; 1624 1625 oid->name = external_name; 1626 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len); 1627 oid->name_len = len; 1628 } 1629 1630 return 0; 1631 } 1632 1633 /* 1634 * If oid doesn't fit into inline buffer, allocate. 1635 */ 1636 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp, 1637 const char *fmt, ...) 1638 { 1639 va_list ap; 1640 int ret; 1641 1642 va_start(ap, fmt); 1643 ret = oid_aprintf_vargs(oid, gfp, fmt, ap); 1644 va_end(ap); 1645 1646 return ret; 1647 } 1648 EXPORT_SYMBOL(ceph_oid_aprintf); 1649 1650 void ceph_oid_destroy(struct ceph_object_id *oid) 1651 { 1652 if (oid->name != oid->inline_name) 1653 kfree(oid->name); 1654 } 1655 EXPORT_SYMBOL(ceph_oid_destroy); 1656 1657 /* 1658 * osds only 1659 */ 1660 static bool __osds_equal(const struct ceph_osds *lhs, 1661 const struct ceph_osds *rhs) 1662 { 1663 if (lhs->size == rhs->size && 1664 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0]))) 1665 return true; 1666 1667 return false; 1668 } 1669 1670 /* 1671 * osds + primary 1672 */ 1673 static bool osds_equal(const struct ceph_osds *lhs, 1674 const struct ceph_osds *rhs) 1675 { 1676 if (__osds_equal(lhs, rhs) && 1677 lhs->primary == rhs->primary) 1678 return true; 1679 1680 return false; 1681 } 1682 1683 static bool osds_valid(const struct ceph_osds *set) 1684 { 1685 /* non-empty set */ 1686 if (set->size > 0 && set->primary >= 0) 1687 return true; 1688 1689 /* empty can_shift_osds set */ 1690 if (!set->size && set->primary == -1) 1691 return true; 1692 1693 /* empty !can_shift_osds set - all NONE */ 1694 if (set->size > 0 && set->primary == -1) { 1695 int i; 1696 1697 for (i = 0; i < set->size; i++) { 1698 if (set->osds[i] != CRUSH_ITEM_NONE) 1699 break; 1700 } 1701 if (i == set->size) 1702 return true; 1703 } 1704 1705 return false; 1706 } 1707 1708 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src) 1709 { 1710 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0])); 1711 dest->size = src->size; 1712 dest->primary = src->primary; 1713 } 1714 1715 static bool is_split(const struct ceph_pg *pgid, 1716 u32 old_pg_num, 1717 u32 new_pg_num) 1718 { 1719 int old_bits = calc_bits_of(old_pg_num); 1720 int old_mask = (1 << old_bits) - 1; 1721 int n; 1722 1723 WARN_ON(pgid->seed >= old_pg_num); 1724 if (new_pg_num <= old_pg_num) 1725 return false; 1726 1727 for (n = 1; ; n++) { 1728 int next_bit = n << (old_bits - 1); 1729 u32 s = next_bit | pgid->seed; 1730 1731 if (s < old_pg_num || s == pgid->seed) 1732 continue; 1733 if (s >= new_pg_num) 1734 break; 1735 1736 s = ceph_stable_mod(s, old_pg_num, old_mask); 1737 if (s == pgid->seed) 1738 return true; 1739 } 1740 1741 return false; 1742 } 1743 1744 bool ceph_is_new_interval(const struct ceph_osds *old_acting, 1745 const struct ceph_osds *new_acting, 1746 const struct ceph_osds *old_up, 1747 const struct ceph_osds *new_up, 1748 int old_size, 1749 int new_size, 1750 int old_min_size, 1751 int new_min_size, 1752 u32 old_pg_num, 1753 u32 new_pg_num, 1754 bool old_sort_bitwise, 1755 bool new_sort_bitwise, 1756 const struct ceph_pg *pgid) 1757 { 1758 return !osds_equal(old_acting, new_acting) || 1759 !osds_equal(old_up, new_up) || 1760 old_size != new_size || 1761 old_min_size != new_min_size || 1762 is_split(pgid, old_pg_num, new_pg_num) || 1763 old_sort_bitwise != new_sort_bitwise; 1764 } 1765 1766 static int calc_pg_rank(int osd, const struct ceph_osds *acting) 1767 { 1768 int i; 1769 1770 for (i = 0; i < acting->size; i++) { 1771 if (acting->osds[i] == osd) 1772 return i; 1773 } 1774 1775 return -1; 1776 } 1777 1778 static bool primary_changed(const struct ceph_osds *old_acting, 1779 const struct ceph_osds *new_acting) 1780 { 1781 if (!old_acting->size && !new_acting->size) 1782 return false; /* both still empty */ 1783 1784 if (!old_acting->size ^ !new_acting->size) 1785 return true; /* was empty, now not, or vice versa */ 1786 1787 if (old_acting->primary != new_acting->primary) 1788 return true; /* primary changed */ 1789 1790 if (calc_pg_rank(old_acting->primary, old_acting) != 1791 calc_pg_rank(new_acting->primary, new_acting)) 1792 return true; 1793 1794 return false; /* same primary (tho replicas may have changed) */ 1795 } 1796 1797 bool ceph_osds_changed(const struct ceph_osds *old_acting, 1798 const struct ceph_osds *new_acting, 1799 bool any_change) 1800 { 1801 if (primary_changed(old_acting, new_acting)) 1802 return true; 1803 1804 if (any_change && !__osds_equal(old_acting, new_acting)) 1805 return true; 1806 1807 return false; 1808 } 1809 1810 /* 1811 * calculate file layout from given offset, length. 1812 * fill in correct oid, logical length, and object extent 1813 * offset, length. 1814 * 1815 * for now, we write only a single su, until we can 1816 * pass a stride back to the caller. 1817 */ 1818 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout, 1819 u64 off, u64 len, 1820 u64 *ono, 1821 u64 *oxoff, u64 *oxlen) 1822 { 1823 u32 osize = layout->object_size; 1824 u32 su = layout->stripe_unit; 1825 u32 sc = layout->stripe_count; 1826 u32 bl, stripeno, stripepos, objsetno; 1827 u32 su_per_object; 1828 u64 t, su_offset; 1829 1830 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len, 1831 osize, su); 1832 if (su == 0 || sc == 0) 1833 goto invalid; 1834 su_per_object = osize / su; 1835 if (su_per_object == 0) 1836 goto invalid; 1837 dout("osize %u / su %u = su_per_object %u\n", osize, su, 1838 su_per_object); 1839 1840 if ((su & ~PAGE_MASK) != 0) 1841 goto invalid; 1842 1843 /* bl = *off / su; */ 1844 t = off; 1845 do_div(t, su); 1846 bl = t; 1847 dout("off %llu / su %u = bl %u\n", off, su, bl); 1848 1849 stripeno = bl / sc; 1850 stripepos = bl % sc; 1851 objsetno = stripeno / su_per_object; 1852 1853 *ono = objsetno * sc + stripepos; 1854 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono); 1855 1856 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */ 1857 t = off; 1858 su_offset = do_div(t, su); 1859 *oxoff = su_offset + (stripeno % su_per_object) * su; 1860 1861 /* 1862 * Calculate the length of the extent being written to the selected 1863 * object. This is the minimum of the full length requested (len) or 1864 * the remainder of the current stripe being written to. 1865 */ 1866 *oxlen = min_t(u64, len, su - su_offset); 1867 1868 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen); 1869 return 0; 1870 1871 invalid: 1872 dout(" invalid layout\n"); 1873 *ono = 0; 1874 *oxoff = 0; 1875 *oxlen = 0; 1876 return -EINVAL; 1877 } 1878 EXPORT_SYMBOL(ceph_calc_file_object_mapping); 1879 1880 /* 1881 * Map an object into a PG. 1882 * 1883 * Should only be called with target_oid and target_oloc (as opposed to 1884 * base_oid and base_oloc), since tiering isn't taken into account. 1885 */ 1886 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 1887 struct ceph_object_id *oid, 1888 struct ceph_object_locator *oloc, 1889 struct ceph_pg *raw_pgid) 1890 { 1891 struct ceph_pg_pool_info *pi; 1892 1893 pi = ceph_pg_pool_by_id(osdmap, oloc->pool); 1894 if (!pi) 1895 return -ENOENT; 1896 1897 if (!oloc->pool_ns) { 1898 raw_pgid->pool = oloc->pool; 1899 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name, 1900 oid->name_len); 1901 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name, 1902 raw_pgid->pool, raw_pgid->seed); 1903 } else { 1904 char stack_buf[256]; 1905 char *buf = stack_buf; 1906 int nsl = oloc->pool_ns->len; 1907 size_t total = nsl + 1 + oid->name_len; 1908 1909 if (total > sizeof(stack_buf)) { 1910 buf = kmalloc(total, GFP_NOIO); 1911 if (!buf) 1912 return -ENOMEM; 1913 } 1914 memcpy(buf, oloc->pool_ns->str, nsl); 1915 buf[nsl] = '\037'; 1916 memcpy(buf + nsl + 1, oid->name, oid->name_len); 1917 raw_pgid->pool = oloc->pool; 1918 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total); 1919 if (buf != stack_buf) 1920 kfree(buf); 1921 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__, 1922 oid->name, nsl, oloc->pool_ns->str, 1923 raw_pgid->pool, raw_pgid->seed); 1924 } 1925 return 0; 1926 } 1927 EXPORT_SYMBOL(ceph_object_locator_to_pg); 1928 1929 /* 1930 * Map a raw PG (full precision ps) into an actual PG. 1931 */ 1932 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi, 1933 const struct ceph_pg *raw_pgid, 1934 struct ceph_pg *pgid) 1935 { 1936 pgid->pool = raw_pgid->pool; 1937 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num, 1938 pi->pg_num_mask); 1939 } 1940 1941 /* 1942 * Map a raw PG (full precision ps) into a placement ps (placement 1943 * seed). Include pool id in that value so that different pools don't 1944 * use the same seeds. 1945 */ 1946 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi, 1947 const struct ceph_pg *raw_pgid) 1948 { 1949 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) { 1950 /* hash pool id and seed so that pool PGs do not overlap */ 1951 return crush_hash32_2(CRUSH_HASH_RJENKINS1, 1952 ceph_stable_mod(raw_pgid->seed, 1953 pi->pgp_num, 1954 pi->pgp_num_mask), 1955 raw_pgid->pool); 1956 } else { 1957 /* 1958 * legacy behavior: add ps and pool together. this is 1959 * not a great approach because the PGs from each pool 1960 * will overlap on top of each other: 0.5 == 1.4 == 1961 * 2.3 == ... 1962 */ 1963 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num, 1964 pi->pgp_num_mask) + 1965 (unsigned)raw_pgid->pool; 1966 } 1967 } 1968 1969 static int do_crush(struct ceph_osdmap *map, int ruleno, int x, 1970 int *result, int result_max, 1971 const __u32 *weight, int weight_max) 1972 { 1973 int r; 1974 1975 BUG_ON(result_max > CEPH_PG_MAX_SIZE); 1976 1977 mutex_lock(&map->crush_scratch_mutex); 1978 r = crush_do_rule(map->crush, ruleno, x, result, result_max, 1979 weight, weight_max, map->crush_workspace, 1980 map->crush_scratch_ary); 1981 mutex_unlock(&map->crush_scratch_mutex); 1982 1983 return r; 1984 } 1985 1986 /* 1987 * Calculate raw set (CRUSH output) for given PG. The result may 1988 * contain nonexistent OSDs. ->primary is undefined for a raw set. 1989 * 1990 * Placement seed (CRUSH input) is returned through @ppps. 1991 */ 1992 static void pg_to_raw_osds(struct ceph_osdmap *osdmap, 1993 struct ceph_pg_pool_info *pi, 1994 const struct ceph_pg *raw_pgid, 1995 struct ceph_osds *raw, 1996 u32 *ppps) 1997 { 1998 u32 pps = raw_pg_to_pps(pi, raw_pgid); 1999 int ruleno; 2000 int len; 2001 2002 ceph_osds_init(raw); 2003 if (ppps) 2004 *ppps = pps; 2005 2006 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type, 2007 pi->size); 2008 if (ruleno < 0) { 2009 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n", 2010 pi->id, pi->crush_ruleset, pi->type, pi->size); 2011 return; 2012 } 2013 2014 len = do_crush(osdmap, ruleno, pps, raw->osds, 2015 min_t(int, pi->size, ARRAY_SIZE(raw->osds)), 2016 osdmap->osd_weight, osdmap->max_osd); 2017 if (len < 0) { 2018 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n", 2019 len, ruleno, pi->id, pi->crush_ruleset, pi->type, 2020 pi->size); 2021 return; 2022 } 2023 2024 raw->size = len; 2025 } 2026 2027 /* 2028 * Given raw set, calculate up set and up primary. By definition of an 2029 * up set, the result won't contain nonexistent or down OSDs. 2030 * 2031 * This is done in-place - on return @set is the up set. If it's 2032 * empty, ->primary will remain undefined. 2033 */ 2034 static void raw_to_up_osds(struct ceph_osdmap *osdmap, 2035 struct ceph_pg_pool_info *pi, 2036 struct ceph_osds *set) 2037 { 2038 int i; 2039 2040 /* ->primary is undefined for a raw set */ 2041 BUG_ON(set->primary != -1); 2042 2043 if (ceph_can_shift_osds(pi)) { 2044 int removed = 0; 2045 2046 /* shift left */ 2047 for (i = 0; i < set->size; i++) { 2048 if (ceph_osd_is_down(osdmap, set->osds[i])) { 2049 removed++; 2050 continue; 2051 } 2052 if (removed) 2053 set->osds[i - removed] = set->osds[i]; 2054 } 2055 set->size -= removed; 2056 if (set->size > 0) 2057 set->primary = set->osds[0]; 2058 } else { 2059 /* set down/dne devices to NONE */ 2060 for (i = set->size - 1; i >= 0; i--) { 2061 if (ceph_osd_is_down(osdmap, set->osds[i])) 2062 set->osds[i] = CRUSH_ITEM_NONE; 2063 else 2064 set->primary = set->osds[i]; 2065 } 2066 } 2067 } 2068 2069 static void apply_primary_affinity(struct ceph_osdmap *osdmap, 2070 struct ceph_pg_pool_info *pi, 2071 u32 pps, 2072 struct ceph_osds *up) 2073 { 2074 int i; 2075 int pos = -1; 2076 2077 /* 2078 * Do we have any non-default primary_affinity values for these 2079 * osds? 2080 */ 2081 if (!osdmap->osd_primary_affinity) 2082 return; 2083 2084 for (i = 0; i < up->size; i++) { 2085 int osd = up->osds[i]; 2086 2087 if (osd != CRUSH_ITEM_NONE && 2088 osdmap->osd_primary_affinity[osd] != 2089 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) { 2090 break; 2091 } 2092 } 2093 if (i == up->size) 2094 return; 2095 2096 /* 2097 * Pick the primary. Feed both the seed (for the pg) and the 2098 * osd into the hash/rng so that a proportional fraction of an 2099 * osd's pgs get rejected as primary. 2100 */ 2101 for (i = 0; i < up->size; i++) { 2102 int osd = up->osds[i]; 2103 u32 aff; 2104 2105 if (osd == CRUSH_ITEM_NONE) 2106 continue; 2107 2108 aff = osdmap->osd_primary_affinity[osd]; 2109 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY && 2110 (crush_hash32_2(CRUSH_HASH_RJENKINS1, 2111 pps, osd) >> 16) >= aff) { 2112 /* 2113 * We chose not to use this primary. Note it 2114 * anyway as a fallback in case we don't pick 2115 * anyone else, but keep looking. 2116 */ 2117 if (pos < 0) 2118 pos = i; 2119 } else { 2120 pos = i; 2121 break; 2122 } 2123 } 2124 if (pos < 0) 2125 return; 2126 2127 up->primary = up->osds[pos]; 2128 2129 if (ceph_can_shift_osds(pi) && pos > 0) { 2130 /* move the new primary to the front */ 2131 for (i = pos; i > 0; i--) 2132 up->osds[i] = up->osds[i - 1]; 2133 up->osds[0] = up->primary; 2134 } 2135 } 2136 2137 /* 2138 * Get pg_temp and primary_temp mappings for given PG. 2139 * 2140 * Note that a PG may have none, only pg_temp, only primary_temp or 2141 * both pg_temp and primary_temp mappings. This means @temp isn't 2142 * always a valid OSD set on return: in the "only primary_temp" case, 2143 * @temp will have its ->primary >= 0 but ->size == 0. 2144 */ 2145 static void get_temp_osds(struct ceph_osdmap *osdmap, 2146 struct ceph_pg_pool_info *pi, 2147 const struct ceph_pg *raw_pgid, 2148 struct ceph_osds *temp) 2149 { 2150 struct ceph_pg pgid; 2151 struct ceph_pg_mapping *pg; 2152 int i; 2153 2154 raw_pg_to_pg(pi, raw_pgid, &pgid); 2155 ceph_osds_init(temp); 2156 2157 /* pg_temp? */ 2158 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid); 2159 if (pg) { 2160 for (i = 0; i < pg->pg_temp.len; i++) { 2161 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) { 2162 if (ceph_can_shift_osds(pi)) 2163 continue; 2164 2165 temp->osds[temp->size++] = CRUSH_ITEM_NONE; 2166 } else { 2167 temp->osds[temp->size++] = pg->pg_temp.osds[i]; 2168 } 2169 } 2170 2171 /* apply pg_temp's primary */ 2172 for (i = 0; i < temp->size; i++) { 2173 if (temp->osds[i] != CRUSH_ITEM_NONE) { 2174 temp->primary = temp->osds[i]; 2175 break; 2176 } 2177 } 2178 } 2179 2180 /* primary_temp? */ 2181 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid); 2182 if (pg) 2183 temp->primary = pg->primary_temp.osd; 2184 } 2185 2186 /* 2187 * Map a PG to its acting set as well as its up set. 2188 * 2189 * Acting set is used for data mapping purposes, while up set can be 2190 * recorded for detecting interval changes and deciding whether to 2191 * resend a request. 2192 */ 2193 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 2194 const struct ceph_pg *raw_pgid, 2195 struct ceph_osds *up, 2196 struct ceph_osds *acting) 2197 { 2198 struct ceph_pg_pool_info *pi; 2199 u32 pps; 2200 2201 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool); 2202 if (!pi) { 2203 ceph_osds_init(up); 2204 ceph_osds_init(acting); 2205 goto out; 2206 } 2207 2208 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps); 2209 raw_to_up_osds(osdmap, pi, up); 2210 apply_primary_affinity(osdmap, pi, pps, up); 2211 get_temp_osds(osdmap, pi, raw_pgid, acting); 2212 if (!acting->size) { 2213 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0])); 2214 acting->size = up->size; 2215 if (acting->primary == -1) 2216 acting->primary = up->primary; 2217 } 2218 out: 2219 WARN_ON(!osds_valid(up) || !osds_valid(acting)); 2220 } 2221 2222 /* 2223 * Return acting primary for given PG, or -1 if none. 2224 */ 2225 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 2226 const struct ceph_pg *raw_pgid) 2227 { 2228 struct ceph_osds up, acting; 2229 2230 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting); 2231 return acting.primary; 2232 } 2233 EXPORT_SYMBOL(ceph_pg_to_acting_primary); 2234