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 err = -EINVAL; 321 ceph_decode_32_safe(p, end, yes, bad); 322 if (!yes) { 323 dout("crush_decode NO rule %d off %x %p to %p\n", 324 i, (int)(*p-start), *p, end); 325 c->rules[i] = NULL; 326 continue; 327 } 328 329 dout("crush_decode rule %d off %x %p to %p\n", 330 i, (int)(*p-start), *p, end); 331 332 /* len */ 333 ceph_decode_32_safe(p, end, yes, bad); 334 #if BITS_PER_LONG == 32 335 err = -EINVAL; 336 if (yes > (ULONG_MAX - sizeof(*r)) 337 / sizeof(struct crush_rule_step)) 338 goto bad; 339 #endif 340 r = c->rules[i] = kmalloc(sizeof(*r) + 341 yes*sizeof(struct crush_rule_step), 342 GFP_NOFS); 343 if (r == NULL) 344 goto badmem; 345 dout(" rule %d is at %p\n", i, r); 346 r->len = yes; 347 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */ 348 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad); 349 for (j = 0; j < r->len; j++) { 350 r->steps[j].op = ceph_decode_32(p); 351 r->steps[j].arg1 = ceph_decode_32(p); 352 r->steps[j].arg2 = ceph_decode_32(p); 353 } 354 } 355 356 /* ignore trailing name maps. */ 357 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) { 358 err = skip_name_map(p, end); 359 if (err < 0) 360 goto done; 361 } 362 363 /* tunables */ 364 ceph_decode_need(p, end, 3*sizeof(u32), done); 365 c->choose_local_tries = ceph_decode_32(p); 366 c->choose_local_fallback_tries = ceph_decode_32(p); 367 c->choose_total_tries = ceph_decode_32(p); 368 dout("crush decode tunable choose_local_tries = %d\n", 369 c->choose_local_tries); 370 dout("crush decode tunable choose_local_fallback_tries = %d\n", 371 c->choose_local_fallback_tries); 372 dout("crush decode tunable choose_total_tries = %d\n", 373 c->choose_total_tries); 374 375 ceph_decode_need(p, end, sizeof(u32), done); 376 c->chooseleaf_descend_once = ceph_decode_32(p); 377 dout("crush decode tunable chooseleaf_descend_once = %d\n", 378 c->chooseleaf_descend_once); 379 380 ceph_decode_need(p, end, sizeof(u8), done); 381 c->chooseleaf_vary_r = ceph_decode_8(p); 382 dout("crush decode tunable chooseleaf_vary_r = %d\n", 383 c->chooseleaf_vary_r); 384 385 /* skip straw_calc_version, allowed_bucket_algs */ 386 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done); 387 *p += sizeof(u8) + sizeof(u32); 388 389 ceph_decode_need(p, end, sizeof(u8), done); 390 c->chooseleaf_stable = ceph_decode_8(p); 391 dout("crush decode tunable chooseleaf_stable = %d\n", 392 c->chooseleaf_stable); 393 394 done: 395 crush_finalize(c); 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 /* 686 * last_force_op_resend_preluminous, will be overridden if the 687 * map was encoded with RESEND_ON_SPLIT 688 */ 689 if (ev >= 15) 690 pi->last_force_request_resend = ceph_decode_32(p); 691 else 692 pi->last_force_request_resend = 0; 693 694 if (ev >= 16) 695 *p += 4; /* skip min_read_recency_for_promote */ 696 697 if (ev >= 17) 698 *p += 8; /* skip expected_num_objects */ 699 700 if (ev >= 19) 701 *p += 4; /* skip cache_target_dirty_high_ratio_micro */ 702 703 if (ev >= 20) 704 *p += 4; /* skip min_write_recency_for_promote */ 705 706 if (ev >= 21) 707 *p += 1; /* skip use_gmt_hitset */ 708 709 if (ev >= 22) 710 *p += 1; /* skip fast_read */ 711 712 if (ev >= 23) { 713 *p += 4; /* skip hit_set_grade_decay_rate */ 714 *p += 4; /* skip hit_set_search_last_n */ 715 } 716 717 if (ev >= 24) { 718 /* skip opts */ 719 *p += 1 + 1; /* versions */ 720 len = ceph_decode_32(p); 721 *p += len; 722 } 723 724 if (ev >= 25) 725 pi->last_force_request_resend = ceph_decode_32(p); 726 727 /* ignore the rest */ 728 729 *p = pool_end; 730 calc_pg_masks(pi); 731 return 0; 732 733 bad: 734 return -EINVAL; 735 } 736 737 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map) 738 { 739 struct ceph_pg_pool_info *pi; 740 u32 num, len; 741 u64 pool; 742 743 ceph_decode_32_safe(p, end, num, bad); 744 dout(" %d pool names\n", num); 745 while (num--) { 746 ceph_decode_64_safe(p, end, pool, bad); 747 ceph_decode_32_safe(p, end, len, bad); 748 dout(" pool %llu len %d\n", pool, len); 749 ceph_decode_need(p, end, len, bad); 750 pi = __lookup_pg_pool(&map->pg_pools, pool); 751 if (pi) { 752 char *name = kstrndup(*p, len, GFP_NOFS); 753 754 if (!name) 755 return -ENOMEM; 756 kfree(pi->name); 757 pi->name = name; 758 dout(" name is %s\n", pi->name); 759 } 760 *p += len; 761 } 762 return 0; 763 764 bad: 765 return -EINVAL; 766 } 767 768 /* 769 * osd map 770 */ 771 struct ceph_osdmap *ceph_osdmap_alloc(void) 772 { 773 struct ceph_osdmap *map; 774 775 map = kzalloc(sizeof(*map), GFP_NOIO); 776 if (!map) 777 return NULL; 778 779 map->pg_pools = RB_ROOT; 780 map->pool_max = -1; 781 map->pg_temp = RB_ROOT; 782 map->primary_temp = RB_ROOT; 783 mutex_init(&map->crush_workspace_mutex); 784 785 return map; 786 } 787 788 void ceph_osdmap_destroy(struct ceph_osdmap *map) 789 { 790 dout("osdmap_destroy %p\n", map); 791 if (map->crush) 792 crush_destroy(map->crush); 793 while (!RB_EMPTY_ROOT(&map->pg_temp)) { 794 struct ceph_pg_mapping *pg = 795 rb_entry(rb_first(&map->pg_temp), 796 struct ceph_pg_mapping, node); 797 rb_erase(&pg->node, &map->pg_temp); 798 kfree(pg); 799 } 800 while (!RB_EMPTY_ROOT(&map->primary_temp)) { 801 struct ceph_pg_mapping *pg = 802 rb_entry(rb_first(&map->primary_temp), 803 struct ceph_pg_mapping, node); 804 rb_erase(&pg->node, &map->primary_temp); 805 kfree(pg); 806 } 807 while (!RB_EMPTY_ROOT(&map->pg_pools)) { 808 struct ceph_pg_pool_info *pi = 809 rb_entry(rb_first(&map->pg_pools), 810 struct ceph_pg_pool_info, node); 811 __remove_pg_pool(&map->pg_pools, pi); 812 } 813 kfree(map->osd_state); 814 kfree(map->osd_weight); 815 kfree(map->osd_addr); 816 kfree(map->osd_primary_affinity); 817 kfree(map->crush_workspace); 818 kfree(map); 819 } 820 821 /* 822 * Adjust max_osd value, (re)allocate arrays. 823 * 824 * The new elements are properly initialized. 825 */ 826 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max) 827 { 828 u8 *state; 829 u32 *weight; 830 struct ceph_entity_addr *addr; 831 int i; 832 833 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS); 834 if (!state) 835 return -ENOMEM; 836 map->osd_state = state; 837 838 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS); 839 if (!weight) 840 return -ENOMEM; 841 map->osd_weight = weight; 842 843 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS); 844 if (!addr) 845 return -ENOMEM; 846 map->osd_addr = addr; 847 848 for (i = map->max_osd; i < max; i++) { 849 map->osd_state[i] = 0; 850 map->osd_weight[i] = CEPH_OSD_OUT; 851 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr)); 852 } 853 854 if (map->osd_primary_affinity) { 855 u32 *affinity; 856 857 affinity = krealloc(map->osd_primary_affinity, 858 max*sizeof(*affinity), GFP_NOFS); 859 if (!affinity) 860 return -ENOMEM; 861 map->osd_primary_affinity = affinity; 862 863 for (i = map->max_osd; i < max; i++) 864 map->osd_primary_affinity[i] = 865 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 866 } 867 868 map->max_osd = max; 869 870 return 0; 871 } 872 873 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush) 874 { 875 void *workspace; 876 size_t work_size; 877 878 if (IS_ERR(crush)) 879 return PTR_ERR(crush); 880 881 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE); 882 dout("%s work_size %zu bytes\n", __func__, work_size); 883 workspace = kmalloc(work_size, GFP_NOIO); 884 if (!workspace) { 885 crush_destroy(crush); 886 return -ENOMEM; 887 } 888 crush_init_workspace(crush, workspace); 889 890 if (map->crush) 891 crush_destroy(map->crush); 892 kfree(map->crush_workspace); 893 map->crush = crush; 894 map->crush_workspace = workspace; 895 return 0; 896 } 897 898 #define OSDMAP_WRAPPER_COMPAT_VER 7 899 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1 900 901 /* 902 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps, 903 * to struct_v of the client_data section for new (v7 and above) 904 * osdmaps. 905 */ 906 static int get_osdmap_client_data_v(void **p, void *end, 907 const char *prefix, u8 *v) 908 { 909 u8 struct_v; 910 911 ceph_decode_8_safe(p, end, struct_v, e_inval); 912 if (struct_v >= 7) { 913 u8 struct_compat; 914 915 ceph_decode_8_safe(p, end, struct_compat, e_inval); 916 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) { 917 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n", 918 struct_v, struct_compat, 919 OSDMAP_WRAPPER_COMPAT_VER, prefix); 920 return -EINVAL; 921 } 922 *p += 4; /* ignore wrapper struct_len */ 923 924 ceph_decode_8_safe(p, end, struct_v, e_inval); 925 ceph_decode_8_safe(p, end, struct_compat, e_inval); 926 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) { 927 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n", 928 struct_v, struct_compat, 929 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix); 930 return -EINVAL; 931 } 932 *p += 4; /* ignore client data struct_len */ 933 } else { 934 u16 version; 935 936 *p -= 1; 937 ceph_decode_16_safe(p, end, version, e_inval); 938 if (version < 6) { 939 pr_warn("got v %d < 6 of %s ceph_osdmap\n", 940 version, prefix); 941 return -EINVAL; 942 } 943 944 /* old osdmap enconding */ 945 struct_v = 0; 946 } 947 948 *v = struct_v; 949 return 0; 950 951 e_inval: 952 return -EINVAL; 953 } 954 955 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map, 956 bool incremental) 957 { 958 u32 n; 959 960 ceph_decode_32_safe(p, end, n, e_inval); 961 while (n--) { 962 struct ceph_pg_pool_info *pi; 963 u64 pool; 964 int ret; 965 966 ceph_decode_64_safe(p, end, pool, e_inval); 967 968 pi = __lookup_pg_pool(&map->pg_pools, pool); 969 if (!incremental || !pi) { 970 pi = kzalloc(sizeof(*pi), GFP_NOFS); 971 if (!pi) 972 return -ENOMEM; 973 974 pi->id = pool; 975 976 ret = __insert_pg_pool(&map->pg_pools, pi); 977 if (ret) { 978 kfree(pi); 979 return ret; 980 } 981 } 982 983 ret = decode_pool(p, end, pi); 984 if (ret) 985 return ret; 986 } 987 988 return 0; 989 990 e_inval: 991 return -EINVAL; 992 } 993 994 static int decode_pools(void **p, void *end, struct ceph_osdmap *map) 995 { 996 return __decode_pools(p, end, map, false); 997 } 998 999 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map) 1000 { 1001 return __decode_pools(p, end, map, true); 1002 } 1003 1004 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map, 1005 bool incremental) 1006 { 1007 u32 n; 1008 1009 ceph_decode_32_safe(p, end, n, e_inval); 1010 while (n--) { 1011 struct ceph_pg pgid; 1012 u32 len, i; 1013 int ret; 1014 1015 ret = ceph_decode_pgid(p, end, &pgid); 1016 if (ret) 1017 return ret; 1018 1019 ceph_decode_32_safe(p, end, len, e_inval); 1020 1021 ret = __remove_pg_mapping(&map->pg_temp, pgid); 1022 BUG_ON(!incremental && ret != -ENOENT); 1023 1024 if (!incremental || len > 0) { 1025 struct ceph_pg_mapping *pg; 1026 1027 ceph_decode_need(p, end, len*sizeof(u32), e_inval); 1028 1029 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32)) 1030 return -EINVAL; 1031 1032 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS); 1033 if (!pg) 1034 return -ENOMEM; 1035 1036 pg->pgid = pgid; 1037 pg->pg_temp.len = len; 1038 for (i = 0; i < len; i++) 1039 pg->pg_temp.osds[i] = ceph_decode_32(p); 1040 1041 ret = __insert_pg_mapping(pg, &map->pg_temp); 1042 if (ret) { 1043 kfree(pg); 1044 return ret; 1045 } 1046 } 1047 } 1048 1049 return 0; 1050 1051 e_inval: 1052 return -EINVAL; 1053 } 1054 1055 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map) 1056 { 1057 return __decode_pg_temp(p, end, map, false); 1058 } 1059 1060 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map) 1061 { 1062 return __decode_pg_temp(p, end, map, true); 1063 } 1064 1065 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map, 1066 bool incremental) 1067 { 1068 u32 n; 1069 1070 ceph_decode_32_safe(p, end, n, e_inval); 1071 while (n--) { 1072 struct ceph_pg pgid; 1073 u32 osd; 1074 int ret; 1075 1076 ret = ceph_decode_pgid(p, end, &pgid); 1077 if (ret) 1078 return ret; 1079 1080 ceph_decode_32_safe(p, end, osd, e_inval); 1081 1082 ret = __remove_pg_mapping(&map->primary_temp, pgid); 1083 BUG_ON(!incremental && ret != -ENOENT); 1084 1085 if (!incremental || osd != (u32)-1) { 1086 struct ceph_pg_mapping *pg; 1087 1088 pg = kzalloc(sizeof(*pg), GFP_NOFS); 1089 if (!pg) 1090 return -ENOMEM; 1091 1092 pg->pgid = pgid; 1093 pg->primary_temp.osd = osd; 1094 1095 ret = __insert_pg_mapping(pg, &map->primary_temp); 1096 if (ret) { 1097 kfree(pg); 1098 return ret; 1099 } 1100 } 1101 } 1102 1103 return 0; 1104 1105 e_inval: 1106 return -EINVAL; 1107 } 1108 1109 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map) 1110 { 1111 return __decode_primary_temp(p, end, map, false); 1112 } 1113 1114 static int decode_new_primary_temp(void **p, void *end, 1115 struct ceph_osdmap *map) 1116 { 1117 return __decode_primary_temp(p, end, map, true); 1118 } 1119 1120 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd) 1121 { 1122 BUG_ON(osd >= map->max_osd); 1123 1124 if (!map->osd_primary_affinity) 1125 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1126 1127 return map->osd_primary_affinity[osd]; 1128 } 1129 1130 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff) 1131 { 1132 BUG_ON(osd >= map->max_osd); 1133 1134 if (!map->osd_primary_affinity) { 1135 int i; 1136 1137 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32), 1138 GFP_NOFS); 1139 if (!map->osd_primary_affinity) 1140 return -ENOMEM; 1141 1142 for (i = 0; i < map->max_osd; i++) 1143 map->osd_primary_affinity[i] = 1144 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1145 } 1146 1147 map->osd_primary_affinity[osd] = aff; 1148 1149 return 0; 1150 } 1151 1152 static int decode_primary_affinity(void **p, void *end, 1153 struct ceph_osdmap *map) 1154 { 1155 u32 len, i; 1156 1157 ceph_decode_32_safe(p, end, len, e_inval); 1158 if (len == 0) { 1159 kfree(map->osd_primary_affinity); 1160 map->osd_primary_affinity = NULL; 1161 return 0; 1162 } 1163 if (len != map->max_osd) 1164 goto e_inval; 1165 1166 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval); 1167 1168 for (i = 0; i < map->max_osd; i++) { 1169 int ret; 1170 1171 ret = set_primary_affinity(map, i, ceph_decode_32(p)); 1172 if (ret) 1173 return ret; 1174 } 1175 1176 return 0; 1177 1178 e_inval: 1179 return -EINVAL; 1180 } 1181 1182 static int decode_new_primary_affinity(void **p, void *end, 1183 struct ceph_osdmap *map) 1184 { 1185 u32 n; 1186 1187 ceph_decode_32_safe(p, end, n, e_inval); 1188 while (n--) { 1189 u32 osd, aff; 1190 int ret; 1191 1192 ceph_decode_32_safe(p, end, osd, e_inval); 1193 ceph_decode_32_safe(p, end, aff, e_inval); 1194 1195 ret = set_primary_affinity(map, osd, aff); 1196 if (ret) 1197 return ret; 1198 1199 pr_info("osd%d primary-affinity 0x%x\n", osd, aff); 1200 } 1201 1202 return 0; 1203 1204 e_inval: 1205 return -EINVAL; 1206 } 1207 1208 /* 1209 * decode a full map. 1210 */ 1211 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map) 1212 { 1213 u8 struct_v; 1214 u32 epoch = 0; 1215 void *start = *p; 1216 u32 max; 1217 u32 len, i; 1218 int err; 1219 1220 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1221 1222 err = get_osdmap_client_data_v(p, end, "full", &struct_v); 1223 if (err) 1224 goto bad; 1225 1226 /* fsid, epoch, created, modified */ 1227 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) + 1228 sizeof(map->created) + sizeof(map->modified), e_inval); 1229 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid)); 1230 epoch = map->epoch = ceph_decode_32(p); 1231 ceph_decode_copy(p, &map->created, sizeof(map->created)); 1232 ceph_decode_copy(p, &map->modified, sizeof(map->modified)); 1233 1234 /* pools */ 1235 err = decode_pools(p, end, map); 1236 if (err) 1237 goto bad; 1238 1239 /* pool_name */ 1240 err = decode_pool_names(p, end, map); 1241 if (err) 1242 goto bad; 1243 1244 ceph_decode_32_safe(p, end, map->pool_max, e_inval); 1245 1246 ceph_decode_32_safe(p, end, map->flags, e_inval); 1247 1248 /* max_osd */ 1249 ceph_decode_32_safe(p, end, max, e_inval); 1250 1251 /* (re)alloc osd arrays */ 1252 err = osdmap_set_max_osd(map, max); 1253 if (err) 1254 goto bad; 1255 1256 /* osd_state, osd_weight, osd_addrs->client_addr */ 1257 ceph_decode_need(p, end, 3*sizeof(u32) + 1258 map->max_osd*(1 + sizeof(*map->osd_weight) + 1259 sizeof(*map->osd_addr)), e_inval); 1260 1261 if (ceph_decode_32(p) != map->max_osd) 1262 goto e_inval; 1263 1264 ceph_decode_copy(p, map->osd_state, map->max_osd); 1265 1266 if (ceph_decode_32(p) != map->max_osd) 1267 goto e_inval; 1268 1269 for (i = 0; i < map->max_osd; i++) 1270 map->osd_weight[i] = ceph_decode_32(p); 1271 1272 if (ceph_decode_32(p) != map->max_osd) 1273 goto e_inval; 1274 1275 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr)); 1276 for (i = 0; i < map->max_osd; i++) 1277 ceph_decode_addr(&map->osd_addr[i]); 1278 1279 /* pg_temp */ 1280 err = decode_pg_temp(p, end, map); 1281 if (err) 1282 goto bad; 1283 1284 /* primary_temp */ 1285 if (struct_v >= 1) { 1286 err = decode_primary_temp(p, end, map); 1287 if (err) 1288 goto bad; 1289 } 1290 1291 /* primary_affinity */ 1292 if (struct_v >= 2) { 1293 err = decode_primary_affinity(p, end, map); 1294 if (err) 1295 goto bad; 1296 } else { 1297 /* XXX can this happen? */ 1298 kfree(map->osd_primary_affinity); 1299 map->osd_primary_affinity = NULL; 1300 } 1301 1302 /* crush */ 1303 ceph_decode_32_safe(p, end, len, e_inval); 1304 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end))); 1305 if (err) 1306 goto bad; 1307 1308 /* ignore the rest */ 1309 *p = end; 1310 1311 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1312 return 0; 1313 1314 e_inval: 1315 err = -EINVAL; 1316 bad: 1317 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1318 err, epoch, (int)(*p - start), *p, start, end); 1319 print_hex_dump(KERN_DEBUG, "osdmap: ", 1320 DUMP_PREFIX_OFFSET, 16, 1, 1321 start, end - start, true); 1322 return err; 1323 } 1324 1325 /* 1326 * Allocate and decode a full map. 1327 */ 1328 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end) 1329 { 1330 struct ceph_osdmap *map; 1331 int ret; 1332 1333 map = ceph_osdmap_alloc(); 1334 if (!map) 1335 return ERR_PTR(-ENOMEM); 1336 1337 ret = osdmap_decode(p, end, map); 1338 if (ret) { 1339 ceph_osdmap_destroy(map); 1340 return ERR_PTR(ret); 1341 } 1342 1343 return map; 1344 } 1345 1346 /* 1347 * Encoding order is (new_up_client, new_state, new_weight). Need to 1348 * apply in the (new_weight, new_state, new_up_client) order, because 1349 * an incremental map may look like e.g. 1350 * 1351 * new_up_client: { osd=6, addr=... } # set osd_state and addr 1352 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state 1353 */ 1354 static int decode_new_up_state_weight(void **p, void *end, 1355 struct ceph_osdmap *map) 1356 { 1357 void *new_up_client; 1358 void *new_state; 1359 void *new_weight_end; 1360 u32 len; 1361 1362 new_up_client = *p; 1363 ceph_decode_32_safe(p, end, len, e_inval); 1364 len *= sizeof(u32) + sizeof(struct ceph_entity_addr); 1365 ceph_decode_need(p, end, len, e_inval); 1366 *p += len; 1367 1368 new_state = *p; 1369 ceph_decode_32_safe(p, end, len, e_inval); 1370 len *= sizeof(u32) + sizeof(u8); 1371 ceph_decode_need(p, end, len, e_inval); 1372 *p += len; 1373 1374 /* new_weight */ 1375 ceph_decode_32_safe(p, end, len, e_inval); 1376 while (len--) { 1377 s32 osd; 1378 u32 w; 1379 1380 ceph_decode_need(p, end, 2*sizeof(u32), e_inval); 1381 osd = ceph_decode_32(p); 1382 w = ceph_decode_32(p); 1383 BUG_ON(osd >= map->max_osd); 1384 pr_info("osd%d weight 0x%x %s\n", osd, w, 1385 w == CEPH_OSD_IN ? "(in)" : 1386 (w == CEPH_OSD_OUT ? "(out)" : "")); 1387 map->osd_weight[osd] = w; 1388 1389 /* 1390 * If we are marking in, set the EXISTS, and clear the 1391 * AUTOOUT and NEW bits. 1392 */ 1393 if (w) { 1394 map->osd_state[osd] |= CEPH_OSD_EXISTS; 1395 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT | 1396 CEPH_OSD_NEW); 1397 } 1398 } 1399 new_weight_end = *p; 1400 1401 /* new_state (up/down) */ 1402 *p = new_state; 1403 len = ceph_decode_32(p); 1404 while (len--) { 1405 s32 osd; 1406 u8 xorstate; 1407 int ret; 1408 1409 osd = ceph_decode_32(p); 1410 xorstate = ceph_decode_8(p); 1411 if (xorstate == 0) 1412 xorstate = CEPH_OSD_UP; 1413 BUG_ON(osd >= map->max_osd); 1414 if ((map->osd_state[osd] & CEPH_OSD_UP) && 1415 (xorstate & CEPH_OSD_UP)) 1416 pr_info("osd%d down\n", osd); 1417 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) && 1418 (xorstate & CEPH_OSD_EXISTS)) { 1419 pr_info("osd%d does not exist\n", osd); 1420 ret = set_primary_affinity(map, osd, 1421 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY); 1422 if (ret) 1423 return ret; 1424 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr)); 1425 map->osd_state[osd] = 0; 1426 } else { 1427 map->osd_state[osd] ^= xorstate; 1428 } 1429 } 1430 1431 /* new_up_client */ 1432 *p = new_up_client; 1433 len = ceph_decode_32(p); 1434 while (len--) { 1435 s32 osd; 1436 struct ceph_entity_addr addr; 1437 1438 osd = ceph_decode_32(p); 1439 ceph_decode_copy(p, &addr, sizeof(addr)); 1440 ceph_decode_addr(&addr); 1441 BUG_ON(osd >= map->max_osd); 1442 pr_info("osd%d up\n", osd); 1443 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP; 1444 map->osd_addr[osd] = addr; 1445 } 1446 1447 *p = new_weight_end; 1448 return 0; 1449 1450 e_inval: 1451 return -EINVAL; 1452 } 1453 1454 /* 1455 * decode and apply an incremental map update. 1456 */ 1457 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 1458 struct ceph_osdmap *map) 1459 { 1460 struct ceph_fsid fsid; 1461 u32 epoch = 0; 1462 struct ceph_timespec modified; 1463 s32 len; 1464 u64 pool; 1465 __s64 new_pool_max; 1466 __s32 new_flags, max; 1467 void *start = *p; 1468 int err; 1469 u8 struct_v; 1470 1471 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1472 1473 err = get_osdmap_client_data_v(p, end, "inc", &struct_v); 1474 if (err) 1475 goto bad; 1476 1477 /* fsid, epoch, modified, new_pool_max, new_flags */ 1478 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) + 1479 sizeof(u64) + sizeof(u32), e_inval); 1480 ceph_decode_copy(p, &fsid, sizeof(fsid)); 1481 epoch = ceph_decode_32(p); 1482 BUG_ON(epoch != map->epoch+1); 1483 ceph_decode_copy(p, &modified, sizeof(modified)); 1484 new_pool_max = ceph_decode_64(p); 1485 new_flags = ceph_decode_32(p); 1486 1487 /* full map? */ 1488 ceph_decode_32_safe(p, end, len, e_inval); 1489 if (len > 0) { 1490 dout("apply_incremental full map len %d, %p to %p\n", 1491 len, *p, end); 1492 return ceph_osdmap_decode(p, min(*p+len, end)); 1493 } 1494 1495 /* new crush? */ 1496 ceph_decode_32_safe(p, end, len, e_inval); 1497 if (len > 0) { 1498 err = osdmap_set_crush(map, 1499 crush_decode(*p, min(*p + len, end))); 1500 if (err) 1501 goto bad; 1502 *p += len; 1503 } 1504 1505 /* new flags? */ 1506 if (new_flags >= 0) 1507 map->flags = new_flags; 1508 if (new_pool_max >= 0) 1509 map->pool_max = new_pool_max; 1510 1511 /* new max? */ 1512 ceph_decode_32_safe(p, end, max, e_inval); 1513 if (max >= 0) { 1514 err = osdmap_set_max_osd(map, max); 1515 if (err) 1516 goto bad; 1517 } 1518 1519 map->epoch++; 1520 map->modified = modified; 1521 1522 /* new_pools */ 1523 err = decode_new_pools(p, end, map); 1524 if (err) 1525 goto bad; 1526 1527 /* new_pool_names */ 1528 err = decode_pool_names(p, end, map); 1529 if (err) 1530 goto bad; 1531 1532 /* old_pool */ 1533 ceph_decode_32_safe(p, end, len, e_inval); 1534 while (len--) { 1535 struct ceph_pg_pool_info *pi; 1536 1537 ceph_decode_64_safe(p, end, pool, e_inval); 1538 pi = __lookup_pg_pool(&map->pg_pools, pool); 1539 if (pi) 1540 __remove_pg_pool(&map->pg_pools, pi); 1541 } 1542 1543 /* new_up_client, new_state, new_weight */ 1544 err = decode_new_up_state_weight(p, end, map); 1545 if (err) 1546 goto bad; 1547 1548 /* new_pg_temp */ 1549 err = decode_new_pg_temp(p, end, map); 1550 if (err) 1551 goto bad; 1552 1553 /* new_primary_temp */ 1554 if (struct_v >= 1) { 1555 err = decode_new_primary_temp(p, end, map); 1556 if (err) 1557 goto bad; 1558 } 1559 1560 /* new_primary_affinity */ 1561 if (struct_v >= 2) { 1562 err = decode_new_primary_affinity(p, end, map); 1563 if (err) 1564 goto bad; 1565 } 1566 1567 /* ignore the rest */ 1568 *p = end; 1569 1570 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1571 return map; 1572 1573 e_inval: 1574 err = -EINVAL; 1575 bad: 1576 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1577 err, epoch, (int)(*p - start), *p, start, end); 1578 print_hex_dump(KERN_DEBUG, "osdmap: ", 1579 DUMP_PREFIX_OFFSET, 16, 1, 1580 start, end - start, true); 1581 return ERR_PTR(err); 1582 } 1583 1584 void ceph_oloc_copy(struct ceph_object_locator *dest, 1585 const struct ceph_object_locator *src) 1586 { 1587 ceph_oloc_destroy(dest); 1588 1589 dest->pool = src->pool; 1590 if (src->pool_ns) 1591 dest->pool_ns = ceph_get_string(src->pool_ns); 1592 else 1593 dest->pool_ns = NULL; 1594 } 1595 EXPORT_SYMBOL(ceph_oloc_copy); 1596 1597 void ceph_oloc_destroy(struct ceph_object_locator *oloc) 1598 { 1599 ceph_put_string(oloc->pool_ns); 1600 } 1601 EXPORT_SYMBOL(ceph_oloc_destroy); 1602 1603 void ceph_oid_copy(struct ceph_object_id *dest, 1604 const struct ceph_object_id *src) 1605 { 1606 ceph_oid_destroy(dest); 1607 1608 if (src->name != src->inline_name) { 1609 /* very rare, see ceph_object_id definition */ 1610 dest->name = kmalloc(src->name_len + 1, 1611 GFP_NOIO | __GFP_NOFAIL); 1612 } else { 1613 dest->name = dest->inline_name; 1614 } 1615 memcpy(dest->name, src->name, src->name_len + 1); 1616 dest->name_len = src->name_len; 1617 } 1618 EXPORT_SYMBOL(ceph_oid_copy); 1619 1620 static __printf(2, 0) 1621 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap) 1622 { 1623 int len; 1624 1625 WARN_ON(!ceph_oid_empty(oid)); 1626 1627 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap); 1628 if (len >= sizeof(oid->inline_name)) 1629 return len; 1630 1631 oid->name_len = len; 1632 return 0; 1633 } 1634 1635 /* 1636 * If oid doesn't fit into inline buffer, BUG. 1637 */ 1638 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...) 1639 { 1640 va_list ap; 1641 1642 va_start(ap, fmt); 1643 BUG_ON(oid_printf_vargs(oid, fmt, ap)); 1644 va_end(ap); 1645 } 1646 EXPORT_SYMBOL(ceph_oid_printf); 1647 1648 static __printf(3, 0) 1649 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp, 1650 const char *fmt, va_list ap) 1651 { 1652 va_list aq; 1653 int len; 1654 1655 va_copy(aq, ap); 1656 len = oid_printf_vargs(oid, fmt, aq); 1657 va_end(aq); 1658 1659 if (len) { 1660 char *external_name; 1661 1662 external_name = kmalloc(len + 1, gfp); 1663 if (!external_name) 1664 return -ENOMEM; 1665 1666 oid->name = external_name; 1667 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len); 1668 oid->name_len = len; 1669 } 1670 1671 return 0; 1672 } 1673 1674 /* 1675 * If oid doesn't fit into inline buffer, allocate. 1676 */ 1677 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp, 1678 const char *fmt, ...) 1679 { 1680 va_list ap; 1681 int ret; 1682 1683 va_start(ap, fmt); 1684 ret = oid_aprintf_vargs(oid, gfp, fmt, ap); 1685 va_end(ap); 1686 1687 return ret; 1688 } 1689 EXPORT_SYMBOL(ceph_oid_aprintf); 1690 1691 void ceph_oid_destroy(struct ceph_object_id *oid) 1692 { 1693 if (oid->name != oid->inline_name) 1694 kfree(oid->name); 1695 } 1696 EXPORT_SYMBOL(ceph_oid_destroy); 1697 1698 /* 1699 * osds only 1700 */ 1701 static bool __osds_equal(const struct ceph_osds *lhs, 1702 const struct ceph_osds *rhs) 1703 { 1704 if (lhs->size == rhs->size && 1705 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0]))) 1706 return true; 1707 1708 return false; 1709 } 1710 1711 /* 1712 * osds + primary 1713 */ 1714 static bool osds_equal(const struct ceph_osds *lhs, 1715 const struct ceph_osds *rhs) 1716 { 1717 if (__osds_equal(lhs, rhs) && 1718 lhs->primary == rhs->primary) 1719 return true; 1720 1721 return false; 1722 } 1723 1724 static bool osds_valid(const struct ceph_osds *set) 1725 { 1726 /* non-empty set */ 1727 if (set->size > 0 && set->primary >= 0) 1728 return true; 1729 1730 /* empty can_shift_osds set */ 1731 if (!set->size && set->primary == -1) 1732 return true; 1733 1734 /* empty !can_shift_osds set - all NONE */ 1735 if (set->size > 0 && set->primary == -1) { 1736 int i; 1737 1738 for (i = 0; i < set->size; i++) { 1739 if (set->osds[i] != CRUSH_ITEM_NONE) 1740 break; 1741 } 1742 if (i == set->size) 1743 return true; 1744 } 1745 1746 return false; 1747 } 1748 1749 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src) 1750 { 1751 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0])); 1752 dest->size = src->size; 1753 dest->primary = src->primary; 1754 } 1755 1756 static bool is_split(const struct ceph_pg *pgid, 1757 u32 old_pg_num, 1758 u32 new_pg_num) 1759 { 1760 int old_bits = calc_bits_of(old_pg_num); 1761 int old_mask = (1 << old_bits) - 1; 1762 int n; 1763 1764 WARN_ON(pgid->seed >= old_pg_num); 1765 if (new_pg_num <= old_pg_num) 1766 return false; 1767 1768 for (n = 1; ; n++) { 1769 int next_bit = n << (old_bits - 1); 1770 u32 s = next_bit | pgid->seed; 1771 1772 if (s < old_pg_num || s == pgid->seed) 1773 continue; 1774 if (s >= new_pg_num) 1775 break; 1776 1777 s = ceph_stable_mod(s, old_pg_num, old_mask); 1778 if (s == pgid->seed) 1779 return true; 1780 } 1781 1782 return false; 1783 } 1784 1785 bool ceph_is_new_interval(const struct ceph_osds *old_acting, 1786 const struct ceph_osds *new_acting, 1787 const struct ceph_osds *old_up, 1788 const struct ceph_osds *new_up, 1789 int old_size, 1790 int new_size, 1791 int old_min_size, 1792 int new_min_size, 1793 u32 old_pg_num, 1794 u32 new_pg_num, 1795 bool old_sort_bitwise, 1796 bool new_sort_bitwise, 1797 const struct ceph_pg *pgid) 1798 { 1799 return !osds_equal(old_acting, new_acting) || 1800 !osds_equal(old_up, new_up) || 1801 old_size != new_size || 1802 old_min_size != new_min_size || 1803 is_split(pgid, old_pg_num, new_pg_num) || 1804 old_sort_bitwise != new_sort_bitwise; 1805 } 1806 1807 static int calc_pg_rank(int osd, const struct ceph_osds *acting) 1808 { 1809 int i; 1810 1811 for (i = 0; i < acting->size; i++) { 1812 if (acting->osds[i] == osd) 1813 return i; 1814 } 1815 1816 return -1; 1817 } 1818 1819 static bool primary_changed(const struct ceph_osds *old_acting, 1820 const struct ceph_osds *new_acting) 1821 { 1822 if (!old_acting->size && !new_acting->size) 1823 return false; /* both still empty */ 1824 1825 if (!old_acting->size ^ !new_acting->size) 1826 return true; /* was empty, now not, or vice versa */ 1827 1828 if (old_acting->primary != new_acting->primary) 1829 return true; /* primary changed */ 1830 1831 if (calc_pg_rank(old_acting->primary, old_acting) != 1832 calc_pg_rank(new_acting->primary, new_acting)) 1833 return true; 1834 1835 return false; /* same primary (tho replicas may have changed) */ 1836 } 1837 1838 bool ceph_osds_changed(const struct ceph_osds *old_acting, 1839 const struct ceph_osds *new_acting, 1840 bool any_change) 1841 { 1842 if (primary_changed(old_acting, new_acting)) 1843 return true; 1844 1845 if (any_change && !__osds_equal(old_acting, new_acting)) 1846 return true; 1847 1848 return false; 1849 } 1850 1851 /* 1852 * calculate file layout from given offset, length. 1853 * fill in correct oid, logical length, and object extent 1854 * offset, length. 1855 * 1856 * for now, we write only a single su, until we can 1857 * pass a stride back to the caller. 1858 */ 1859 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout, 1860 u64 off, u64 len, 1861 u64 *ono, 1862 u64 *oxoff, u64 *oxlen) 1863 { 1864 u32 osize = layout->object_size; 1865 u32 su = layout->stripe_unit; 1866 u32 sc = layout->stripe_count; 1867 u32 bl, stripeno, stripepos, objsetno; 1868 u32 su_per_object; 1869 u64 t, su_offset; 1870 1871 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len, 1872 osize, su); 1873 if (su == 0 || sc == 0) 1874 goto invalid; 1875 su_per_object = osize / su; 1876 if (su_per_object == 0) 1877 goto invalid; 1878 dout("osize %u / su %u = su_per_object %u\n", osize, su, 1879 su_per_object); 1880 1881 if ((su & ~PAGE_MASK) != 0) 1882 goto invalid; 1883 1884 /* bl = *off / su; */ 1885 t = off; 1886 do_div(t, su); 1887 bl = t; 1888 dout("off %llu / su %u = bl %u\n", off, su, bl); 1889 1890 stripeno = bl / sc; 1891 stripepos = bl % sc; 1892 objsetno = stripeno / su_per_object; 1893 1894 *ono = objsetno * sc + stripepos; 1895 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono); 1896 1897 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */ 1898 t = off; 1899 su_offset = do_div(t, su); 1900 *oxoff = su_offset + (stripeno % su_per_object) * su; 1901 1902 /* 1903 * Calculate the length of the extent being written to the selected 1904 * object. This is the minimum of the full length requested (len) or 1905 * the remainder of the current stripe being written to. 1906 */ 1907 *oxlen = min_t(u64, len, su - su_offset); 1908 1909 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen); 1910 return 0; 1911 1912 invalid: 1913 dout(" invalid layout\n"); 1914 *ono = 0; 1915 *oxoff = 0; 1916 *oxlen = 0; 1917 return -EINVAL; 1918 } 1919 EXPORT_SYMBOL(ceph_calc_file_object_mapping); 1920 1921 /* 1922 * Map an object into a PG. 1923 * 1924 * Should only be called with target_oid and target_oloc (as opposed to 1925 * base_oid and base_oloc), since tiering isn't taken into account. 1926 */ 1927 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 1928 struct ceph_object_id *oid, 1929 struct ceph_object_locator *oloc, 1930 struct ceph_pg *raw_pgid) 1931 { 1932 struct ceph_pg_pool_info *pi; 1933 1934 pi = ceph_pg_pool_by_id(osdmap, oloc->pool); 1935 if (!pi) 1936 return -ENOENT; 1937 1938 if (!oloc->pool_ns) { 1939 raw_pgid->pool = oloc->pool; 1940 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name, 1941 oid->name_len); 1942 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name, 1943 raw_pgid->pool, raw_pgid->seed); 1944 } else { 1945 char stack_buf[256]; 1946 char *buf = stack_buf; 1947 int nsl = oloc->pool_ns->len; 1948 size_t total = nsl + 1 + oid->name_len; 1949 1950 if (total > sizeof(stack_buf)) { 1951 buf = kmalloc(total, GFP_NOIO); 1952 if (!buf) 1953 return -ENOMEM; 1954 } 1955 memcpy(buf, oloc->pool_ns->str, nsl); 1956 buf[nsl] = '\037'; 1957 memcpy(buf + nsl + 1, oid->name, oid->name_len); 1958 raw_pgid->pool = oloc->pool; 1959 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total); 1960 if (buf != stack_buf) 1961 kfree(buf); 1962 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__, 1963 oid->name, nsl, oloc->pool_ns->str, 1964 raw_pgid->pool, raw_pgid->seed); 1965 } 1966 return 0; 1967 } 1968 EXPORT_SYMBOL(ceph_object_locator_to_pg); 1969 1970 /* 1971 * Map a raw PG (full precision ps) into an actual PG. 1972 */ 1973 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi, 1974 const struct ceph_pg *raw_pgid, 1975 struct ceph_pg *pgid) 1976 { 1977 pgid->pool = raw_pgid->pool; 1978 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num, 1979 pi->pg_num_mask); 1980 } 1981 1982 /* 1983 * Map a raw PG (full precision ps) into a placement ps (placement 1984 * seed). Include pool id in that value so that different pools don't 1985 * use the same seeds. 1986 */ 1987 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi, 1988 const struct ceph_pg *raw_pgid) 1989 { 1990 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) { 1991 /* hash pool id and seed so that pool PGs do not overlap */ 1992 return crush_hash32_2(CRUSH_HASH_RJENKINS1, 1993 ceph_stable_mod(raw_pgid->seed, 1994 pi->pgp_num, 1995 pi->pgp_num_mask), 1996 raw_pgid->pool); 1997 } else { 1998 /* 1999 * legacy behavior: add ps and pool together. this is 2000 * not a great approach because the PGs from each pool 2001 * will overlap on top of each other: 0.5 == 1.4 == 2002 * 2.3 == ... 2003 */ 2004 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num, 2005 pi->pgp_num_mask) + 2006 (unsigned)raw_pgid->pool; 2007 } 2008 } 2009 2010 static int do_crush(struct ceph_osdmap *map, int ruleno, int x, 2011 int *result, int result_max, 2012 const __u32 *weight, int weight_max) 2013 { 2014 int r; 2015 2016 BUG_ON(result_max > CEPH_PG_MAX_SIZE); 2017 2018 mutex_lock(&map->crush_workspace_mutex); 2019 r = crush_do_rule(map->crush, ruleno, x, result, result_max, 2020 weight, weight_max, map->crush_workspace); 2021 mutex_unlock(&map->crush_workspace_mutex); 2022 2023 return r; 2024 } 2025 2026 /* 2027 * Calculate raw set (CRUSH output) for given PG. The result may 2028 * contain nonexistent OSDs. ->primary is undefined for a raw set. 2029 * 2030 * Placement seed (CRUSH input) is returned through @ppps. 2031 */ 2032 static void pg_to_raw_osds(struct ceph_osdmap *osdmap, 2033 struct ceph_pg_pool_info *pi, 2034 const struct ceph_pg *raw_pgid, 2035 struct ceph_osds *raw, 2036 u32 *ppps) 2037 { 2038 u32 pps = raw_pg_to_pps(pi, raw_pgid); 2039 int ruleno; 2040 int len; 2041 2042 ceph_osds_init(raw); 2043 if (ppps) 2044 *ppps = pps; 2045 2046 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type, 2047 pi->size); 2048 if (ruleno < 0) { 2049 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n", 2050 pi->id, pi->crush_ruleset, pi->type, pi->size); 2051 return; 2052 } 2053 2054 if (pi->size > ARRAY_SIZE(raw->osds)) { 2055 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n", 2056 pi->id, pi->crush_ruleset, pi->type, pi->size, 2057 ARRAY_SIZE(raw->osds)); 2058 return; 2059 } 2060 2061 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size, 2062 osdmap->osd_weight, osdmap->max_osd); 2063 if (len < 0) { 2064 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n", 2065 len, ruleno, pi->id, pi->crush_ruleset, pi->type, 2066 pi->size); 2067 return; 2068 } 2069 2070 raw->size = len; 2071 } 2072 2073 /* 2074 * Given raw set, calculate up set and up primary. By definition of an 2075 * up set, the result won't contain nonexistent or down OSDs. 2076 * 2077 * This is done in-place - on return @set is the up set. If it's 2078 * empty, ->primary will remain undefined. 2079 */ 2080 static void raw_to_up_osds(struct ceph_osdmap *osdmap, 2081 struct ceph_pg_pool_info *pi, 2082 struct ceph_osds *set) 2083 { 2084 int i; 2085 2086 /* ->primary is undefined for a raw set */ 2087 BUG_ON(set->primary != -1); 2088 2089 if (ceph_can_shift_osds(pi)) { 2090 int removed = 0; 2091 2092 /* shift left */ 2093 for (i = 0; i < set->size; i++) { 2094 if (ceph_osd_is_down(osdmap, set->osds[i])) { 2095 removed++; 2096 continue; 2097 } 2098 if (removed) 2099 set->osds[i - removed] = set->osds[i]; 2100 } 2101 set->size -= removed; 2102 if (set->size > 0) 2103 set->primary = set->osds[0]; 2104 } else { 2105 /* set down/dne devices to NONE */ 2106 for (i = set->size - 1; i >= 0; i--) { 2107 if (ceph_osd_is_down(osdmap, set->osds[i])) 2108 set->osds[i] = CRUSH_ITEM_NONE; 2109 else 2110 set->primary = set->osds[i]; 2111 } 2112 } 2113 } 2114 2115 static void apply_primary_affinity(struct ceph_osdmap *osdmap, 2116 struct ceph_pg_pool_info *pi, 2117 u32 pps, 2118 struct ceph_osds *up) 2119 { 2120 int i; 2121 int pos = -1; 2122 2123 /* 2124 * Do we have any non-default primary_affinity values for these 2125 * osds? 2126 */ 2127 if (!osdmap->osd_primary_affinity) 2128 return; 2129 2130 for (i = 0; i < up->size; i++) { 2131 int osd = up->osds[i]; 2132 2133 if (osd != CRUSH_ITEM_NONE && 2134 osdmap->osd_primary_affinity[osd] != 2135 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) { 2136 break; 2137 } 2138 } 2139 if (i == up->size) 2140 return; 2141 2142 /* 2143 * Pick the primary. Feed both the seed (for the pg) and the 2144 * osd into the hash/rng so that a proportional fraction of an 2145 * osd's pgs get rejected as primary. 2146 */ 2147 for (i = 0; i < up->size; i++) { 2148 int osd = up->osds[i]; 2149 u32 aff; 2150 2151 if (osd == CRUSH_ITEM_NONE) 2152 continue; 2153 2154 aff = osdmap->osd_primary_affinity[osd]; 2155 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY && 2156 (crush_hash32_2(CRUSH_HASH_RJENKINS1, 2157 pps, osd) >> 16) >= aff) { 2158 /* 2159 * We chose not to use this primary. Note it 2160 * anyway as a fallback in case we don't pick 2161 * anyone else, but keep looking. 2162 */ 2163 if (pos < 0) 2164 pos = i; 2165 } else { 2166 pos = i; 2167 break; 2168 } 2169 } 2170 if (pos < 0) 2171 return; 2172 2173 up->primary = up->osds[pos]; 2174 2175 if (ceph_can_shift_osds(pi) && pos > 0) { 2176 /* move the new primary to the front */ 2177 for (i = pos; i > 0; i--) 2178 up->osds[i] = up->osds[i - 1]; 2179 up->osds[0] = up->primary; 2180 } 2181 } 2182 2183 /* 2184 * Get pg_temp and primary_temp mappings for given PG. 2185 * 2186 * Note that a PG may have none, only pg_temp, only primary_temp or 2187 * both pg_temp and primary_temp mappings. This means @temp isn't 2188 * always a valid OSD set on return: in the "only primary_temp" case, 2189 * @temp will have its ->primary >= 0 but ->size == 0. 2190 */ 2191 static void get_temp_osds(struct ceph_osdmap *osdmap, 2192 struct ceph_pg_pool_info *pi, 2193 const struct ceph_pg *raw_pgid, 2194 struct ceph_osds *temp) 2195 { 2196 struct ceph_pg pgid; 2197 struct ceph_pg_mapping *pg; 2198 int i; 2199 2200 raw_pg_to_pg(pi, raw_pgid, &pgid); 2201 ceph_osds_init(temp); 2202 2203 /* pg_temp? */ 2204 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid); 2205 if (pg) { 2206 for (i = 0; i < pg->pg_temp.len; i++) { 2207 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) { 2208 if (ceph_can_shift_osds(pi)) 2209 continue; 2210 2211 temp->osds[temp->size++] = CRUSH_ITEM_NONE; 2212 } else { 2213 temp->osds[temp->size++] = pg->pg_temp.osds[i]; 2214 } 2215 } 2216 2217 /* apply pg_temp's primary */ 2218 for (i = 0; i < temp->size; i++) { 2219 if (temp->osds[i] != CRUSH_ITEM_NONE) { 2220 temp->primary = temp->osds[i]; 2221 break; 2222 } 2223 } 2224 } 2225 2226 /* primary_temp? */ 2227 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid); 2228 if (pg) 2229 temp->primary = pg->primary_temp.osd; 2230 } 2231 2232 /* 2233 * Map a PG to its acting set as well as its up set. 2234 * 2235 * Acting set is used for data mapping purposes, while up set can be 2236 * recorded for detecting interval changes and deciding whether to 2237 * resend a request. 2238 */ 2239 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 2240 const struct ceph_pg *raw_pgid, 2241 struct ceph_osds *up, 2242 struct ceph_osds *acting) 2243 { 2244 struct ceph_pg_pool_info *pi; 2245 u32 pps; 2246 2247 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool); 2248 if (!pi) { 2249 ceph_osds_init(up); 2250 ceph_osds_init(acting); 2251 goto out; 2252 } 2253 2254 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps); 2255 raw_to_up_osds(osdmap, pi, up); 2256 apply_primary_affinity(osdmap, pi, pps, up); 2257 get_temp_osds(osdmap, pi, raw_pgid, acting); 2258 if (!acting->size) { 2259 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0])); 2260 acting->size = up->size; 2261 if (acting->primary == -1) 2262 acting->primary = up->primary; 2263 } 2264 out: 2265 WARN_ON(!osds_valid(up) || !osds_valid(acting)); 2266 } 2267 2268 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap, 2269 const struct ceph_pg *raw_pgid, 2270 struct ceph_spg *spgid) 2271 { 2272 struct ceph_pg_pool_info *pi; 2273 struct ceph_pg pgid; 2274 struct ceph_osds up, acting; 2275 int i; 2276 2277 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool); 2278 if (!pi) 2279 return false; 2280 2281 raw_pg_to_pg(pi, raw_pgid, &pgid); 2282 2283 if (ceph_can_shift_osds(pi)) { 2284 spgid->pgid = pgid; /* struct */ 2285 spgid->shard = CEPH_SPG_NOSHARD; 2286 return true; 2287 } 2288 2289 ceph_pg_to_up_acting_osds(osdmap, &pgid, &up, &acting); 2290 for (i = 0; i < acting.size; i++) { 2291 if (acting.osds[i] == acting.primary) { 2292 spgid->pgid = pgid; /* struct */ 2293 spgid->shard = i; 2294 return true; 2295 } 2296 } 2297 2298 return false; 2299 } 2300 2301 /* 2302 * Return acting primary for given PG, or -1 if none. 2303 */ 2304 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 2305 const struct ceph_pg *raw_pgid) 2306 { 2307 struct ceph_osds up, acting; 2308 2309 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting); 2310 return acting.primary; 2311 } 2312 EXPORT_SYMBOL(ceph_pg_to_acting_primary); 2313