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