1 // SPDX-License-Identifier: GPL-2.0 2 3 #include <linux/ceph/ceph_debug.h> 4 5 #include <linux/module.h> 6 #include <linux/slab.h> 7 #include <asm/div64.h> 8 9 #include <linux/ceph/libceph.h> 10 #include <linux/ceph/osdmap.h> 11 #include <linux/ceph/decode.h> 12 #include <linux/crush/hash.h> 13 #include <linux/crush/mapper.h> 14 15 char *ceph_osdmap_state_str(char *str, int len, u32 state) 16 { 17 if (!len) 18 return str; 19 20 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP)) 21 snprintf(str, len, "exists, up"); 22 else if (state & CEPH_OSD_EXISTS) 23 snprintf(str, len, "exists"); 24 else if (state & CEPH_OSD_UP) 25 snprintf(str, len, "up"); 26 else 27 snprintf(str, len, "doesn't exist"); 28 29 return str; 30 } 31 32 /* maps */ 33 34 static int calc_bits_of(unsigned int t) 35 { 36 int b = 0; 37 while (t) { 38 t = t >> 1; 39 b++; 40 } 41 return b; 42 } 43 44 /* 45 * the foo_mask is the smallest value 2^n-1 that is >= foo. 46 */ 47 static void calc_pg_masks(struct ceph_pg_pool_info *pi) 48 { 49 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1; 50 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1; 51 } 52 53 /* 54 * decode crush map 55 */ 56 static int crush_decode_uniform_bucket(void **p, void *end, 57 struct crush_bucket_uniform *b) 58 { 59 dout("crush_decode_uniform_bucket %p to %p\n", *p, end); 60 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad); 61 b->item_weight = ceph_decode_32(p); 62 return 0; 63 bad: 64 return -EINVAL; 65 } 66 67 static int crush_decode_list_bucket(void **p, void *end, 68 struct crush_bucket_list *b) 69 { 70 int j; 71 dout("crush_decode_list_bucket %p to %p\n", *p, end); 72 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 73 if (b->item_weights == NULL) 74 return -ENOMEM; 75 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 76 if (b->sum_weights == NULL) 77 return -ENOMEM; 78 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); 79 for (j = 0; j < b->h.size; j++) { 80 b->item_weights[j] = ceph_decode_32(p); 81 b->sum_weights[j] = ceph_decode_32(p); 82 } 83 return 0; 84 bad: 85 return -EINVAL; 86 } 87 88 static int crush_decode_tree_bucket(void **p, void *end, 89 struct crush_bucket_tree *b) 90 { 91 int j; 92 dout("crush_decode_tree_bucket %p to %p\n", *p, end); 93 ceph_decode_8_safe(p, end, b->num_nodes, bad); 94 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS); 95 if (b->node_weights == NULL) 96 return -ENOMEM; 97 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad); 98 for (j = 0; j < b->num_nodes; j++) 99 b->node_weights[j] = ceph_decode_32(p); 100 return 0; 101 bad: 102 return -EINVAL; 103 } 104 105 static int crush_decode_straw_bucket(void **p, void *end, 106 struct crush_bucket_straw *b) 107 { 108 int j; 109 dout("crush_decode_straw_bucket %p to %p\n", *p, end); 110 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 111 if (b->item_weights == NULL) 112 return -ENOMEM; 113 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 114 if (b->straws == NULL) 115 return -ENOMEM; 116 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); 117 for (j = 0; j < b->h.size; j++) { 118 b->item_weights[j] = ceph_decode_32(p); 119 b->straws[j] = ceph_decode_32(p); 120 } 121 return 0; 122 bad: 123 return -EINVAL; 124 } 125 126 static int crush_decode_straw2_bucket(void **p, void *end, 127 struct crush_bucket_straw2 *b) 128 { 129 int j; 130 dout("crush_decode_straw2_bucket %p to %p\n", *p, end); 131 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 132 if (b->item_weights == NULL) 133 return -ENOMEM; 134 ceph_decode_need(p, end, b->h.size * sizeof(u32), bad); 135 for (j = 0; j < b->h.size; j++) 136 b->item_weights[j] = ceph_decode_32(p); 137 return 0; 138 bad: 139 return -EINVAL; 140 } 141 142 static struct crush_choose_arg_map *alloc_choose_arg_map(void) 143 { 144 struct crush_choose_arg_map *arg_map; 145 146 arg_map = kzalloc(sizeof(*arg_map), GFP_NOIO); 147 if (!arg_map) 148 return NULL; 149 150 RB_CLEAR_NODE(&arg_map->node); 151 return arg_map; 152 } 153 154 static void free_choose_arg_map(struct crush_choose_arg_map *arg_map) 155 { 156 if (arg_map) { 157 int i, j; 158 159 WARN_ON(!RB_EMPTY_NODE(&arg_map->node)); 160 161 for (i = 0; i < arg_map->size; i++) { 162 struct crush_choose_arg *arg = &arg_map->args[i]; 163 164 for (j = 0; j < arg->weight_set_size; j++) 165 kfree(arg->weight_set[j].weights); 166 kfree(arg->weight_set); 167 kfree(arg->ids); 168 } 169 kfree(arg_map->args); 170 kfree(arg_map); 171 } 172 } 173 174 DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index, 175 node); 176 177 void clear_choose_args(struct crush_map *c) 178 { 179 while (!RB_EMPTY_ROOT(&c->choose_args)) { 180 struct crush_choose_arg_map *arg_map = 181 rb_entry(rb_first(&c->choose_args), 182 struct crush_choose_arg_map, node); 183 184 erase_choose_arg_map(&c->choose_args, arg_map); 185 free_choose_arg_map(arg_map); 186 } 187 } 188 189 static u32 *decode_array_32_alloc(void **p, void *end, u32 *plen) 190 { 191 u32 *a = NULL; 192 u32 len; 193 int ret; 194 195 ceph_decode_32_safe(p, end, len, e_inval); 196 if (len) { 197 u32 i; 198 199 a = kmalloc_array(len, sizeof(u32), GFP_NOIO); 200 if (!a) { 201 ret = -ENOMEM; 202 goto fail; 203 } 204 205 ceph_decode_need(p, end, len * sizeof(u32), e_inval); 206 for (i = 0; i < len; i++) 207 a[i] = ceph_decode_32(p); 208 } 209 210 *plen = len; 211 return a; 212 213 e_inval: 214 ret = -EINVAL; 215 fail: 216 kfree(a); 217 return ERR_PTR(ret); 218 } 219 220 /* 221 * Assumes @arg is zero-initialized. 222 */ 223 static int decode_choose_arg(void **p, void *end, struct crush_choose_arg *arg) 224 { 225 int ret; 226 227 ceph_decode_32_safe(p, end, arg->weight_set_size, e_inval); 228 if (arg->weight_set_size) { 229 u32 i; 230 231 arg->weight_set = kmalloc_array(arg->weight_set_size, 232 sizeof(*arg->weight_set), 233 GFP_NOIO); 234 if (!arg->weight_set) 235 return -ENOMEM; 236 237 for (i = 0; i < arg->weight_set_size; i++) { 238 struct crush_weight_set *w = &arg->weight_set[i]; 239 240 w->weights = decode_array_32_alloc(p, end, &w->size); 241 if (IS_ERR(w->weights)) { 242 ret = PTR_ERR(w->weights); 243 w->weights = NULL; 244 return ret; 245 } 246 } 247 } 248 249 arg->ids = decode_array_32_alloc(p, end, &arg->ids_size); 250 if (IS_ERR(arg->ids)) { 251 ret = PTR_ERR(arg->ids); 252 arg->ids = NULL; 253 return ret; 254 } 255 256 return 0; 257 258 e_inval: 259 return -EINVAL; 260 } 261 262 static int decode_choose_args(void **p, void *end, struct crush_map *c) 263 { 264 struct crush_choose_arg_map *arg_map = NULL; 265 u32 num_choose_arg_maps, num_buckets; 266 int ret; 267 268 ceph_decode_32_safe(p, end, num_choose_arg_maps, e_inval); 269 while (num_choose_arg_maps--) { 270 arg_map = alloc_choose_arg_map(); 271 if (!arg_map) { 272 ret = -ENOMEM; 273 goto fail; 274 } 275 276 ceph_decode_64_safe(p, end, arg_map->choose_args_index, 277 e_inval); 278 arg_map->size = c->max_buckets; 279 arg_map->args = kcalloc(arg_map->size, sizeof(*arg_map->args), 280 GFP_NOIO); 281 if (!arg_map->args) { 282 ret = -ENOMEM; 283 goto fail; 284 } 285 286 ceph_decode_32_safe(p, end, num_buckets, e_inval); 287 while (num_buckets--) { 288 struct crush_choose_arg *arg; 289 u32 bucket_index; 290 291 ceph_decode_32_safe(p, end, bucket_index, e_inval); 292 if (bucket_index >= arg_map->size) 293 goto e_inval; 294 295 arg = &arg_map->args[bucket_index]; 296 ret = decode_choose_arg(p, end, arg); 297 if (ret) 298 goto fail; 299 300 if (arg->ids_size && 301 arg->ids_size != c->buckets[bucket_index]->size) 302 goto e_inval; 303 } 304 305 insert_choose_arg_map(&c->choose_args, arg_map); 306 } 307 308 return 0; 309 310 e_inval: 311 ret = -EINVAL; 312 fail: 313 free_choose_arg_map(arg_map); 314 return ret; 315 } 316 317 static void crush_finalize(struct crush_map *c) 318 { 319 __s32 b; 320 321 /* Space for the array of pointers to per-bucket workspace */ 322 c->working_size = sizeof(struct crush_work) + 323 c->max_buckets * sizeof(struct crush_work_bucket *); 324 325 for (b = 0; b < c->max_buckets; b++) { 326 if (!c->buckets[b]) 327 continue; 328 329 switch (c->buckets[b]->alg) { 330 default: 331 /* 332 * The base case, permutation variables and 333 * the pointer to the permutation array. 334 */ 335 c->working_size += sizeof(struct crush_work_bucket); 336 break; 337 } 338 /* Every bucket has a permutation array. */ 339 c->working_size += c->buckets[b]->size * sizeof(__u32); 340 } 341 } 342 343 static struct crush_map *crush_decode(void *pbyval, void *end) 344 { 345 struct crush_map *c; 346 int err; 347 int i, j; 348 void **p = &pbyval; 349 void *start = pbyval; 350 u32 magic; 351 352 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p)); 353 354 c = kzalloc(sizeof(*c), GFP_NOFS); 355 if (c == NULL) 356 return ERR_PTR(-ENOMEM); 357 358 c->choose_args = RB_ROOT; 359 360 /* set tunables to default values */ 361 c->choose_local_tries = 2; 362 c->choose_local_fallback_tries = 5; 363 c->choose_total_tries = 19; 364 c->chooseleaf_descend_once = 0; 365 366 ceph_decode_need(p, end, 4*sizeof(u32), bad); 367 magic = ceph_decode_32(p); 368 if (magic != CRUSH_MAGIC) { 369 pr_err("crush_decode magic %x != current %x\n", 370 (unsigned int)magic, (unsigned int)CRUSH_MAGIC); 371 goto bad; 372 } 373 c->max_buckets = ceph_decode_32(p); 374 c->max_rules = ceph_decode_32(p); 375 c->max_devices = ceph_decode_32(p); 376 377 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS); 378 if (c->buckets == NULL) 379 goto badmem; 380 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS); 381 if (c->rules == NULL) 382 goto badmem; 383 384 /* buckets */ 385 for (i = 0; i < c->max_buckets; i++) { 386 int size = 0; 387 u32 alg; 388 struct crush_bucket *b; 389 390 ceph_decode_32_safe(p, end, alg, bad); 391 if (alg == 0) { 392 c->buckets[i] = NULL; 393 continue; 394 } 395 dout("crush_decode bucket %d off %x %p to %p\n", 396 i, (int)(*p-start), *p, end); 397 398 switch (alg) { 399 case CRUSH_BUCKET_UNIFORM: 400 size = sizeof(struct crush_bucket_uniform); 401 break; 402 case CRUSH_BUCKET_LIST: 403 size = sizeof(struct crush_bucket_list); 404 break; 405 case CRUSH_BUCKET_TREE: 406 size = sizeof(struct crush_bucket_tree); 407 break; 408 case CRUSH_BUCKET_STRAW: 409 size = sizeof(struct crush_bucket_straw); 410 break; 411 case CRUSH_BUCKET_STRAW2: 412 size = sizeof(struct crush_bucket_straw2); 413 break; 414 default: 415 goto bad; 416 } 417 BUG_ON(size == 0); 418 b = c->buckets[i] = kzalloc(size, GFP_NOFS); 419 if (b == NULL) 420 goto badmem; 421 422 ceph_decode_need(p, end, 4*sizeof(u32), bad); 423 b->id = ceph_decode_32(p); 424 b->type = ceph_decode_16(p); 425 b->alg = ceph_decode_8(p); 426 b->hash = ceph_decode_8(p); 427 b->weight = ceph_decode_32(p); 428 b->size = ceph_decode_32(p); 429 430 dout("crush_decode bucket size %d off %x %p to %p\n", 431 b->size, (int)(*p-start), *p, end); 432 433 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS); 434 if (b->items == NULL) 435 goto badmem; 436 437 ceph_decode_need(p, end, b->size*sizeof(u32), bad); 438 for (j = 0; j < b->size; j++) 439 b->items[j] = ceph_decode_32(p); 440 441 switch (b->alg) { 442 case CRUSH_BUCKET_UNIFORM: 443 err = crush_decode_uniform_bucket(p, end, 444 (struct crush_bucket_uniform *)b); 445 if (err < 0) 446 goto fail; 447 break; 448 case CRUSH_BUCKET_LIST: 449 err = crush_decode_list_bucket(p, end, 450 (struct crush_bucket_list *)b); 451 if (err < 0) 452 goto fail; 453 break; 454 case CRUSH_BUCKET_TREE: 455 err = crush_decode_tree_bucket(p, end, 456 (struct crush_bucket_tree *)b); 457 if (err < 0) 458 goto fail; 459 break; 460 case CRUSH_BUCKET_STRAW: 461 err = crush_decode_straw_bucket(p, end, 462 (struct crush_bucket_straw *)b); 463 if (err < 0) 464 goto fail; 465 break; 466 case CRUSH_BUCKET_STRAW2: 467 err = crush_decode_straw2_bucket(p, end, 468 (struct crush_bucket_straw2 *)b); 469 if (err < 0) 470 goto fail; 471 break; 472 } 473 } 474 475 /* rules */ 476 dout("rule vec is %p\n", c->rules); 477 for (i = 0; i < c->max_rules; i++) { 478 u32 yes; 479 struct crush_rule *r; 480 481 ceph_decode_32_safe(p, end, yes, bad); 482 if (!yes) { 483 dout("crush_decode NO rule %d off %x %p to %p\n", 484 i, (int)(*p-start), *p, end); 485 c->rules[i] = NULL; 486 continue; 487 } 488 489 dout("crush_decode rule %d off %x %p to %p\n", 490 i, (int)(*p-start), *p, end); 491 492 /* len */ 493 ceph_decode_32_safe(p, end, yes, bad); 494 #if BITS_PER_LONG == 32 495 if (yes > (ULONG_MAX - sizeof(*r)) 496 / sizeof(struct crush_rule_step)) 497 goto bad; 498 #endif 499 r = c->rules[i] = kmalloc(sizeof(*r) + 500 yes*sizeof(struct crush_rule_step), 501 GFP_NOFS); 502 if (r == NULL) 503 goto badmem; 504 dout(" rule %d is at %p\n", i, r); 505 r->len = yes; 506 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */ 507 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad); 508 for (j = 0; j < r->len; j++) { 509 r->steps[j].op = ceph_decode_32(p); 510 r->steps[j].arg1 = ceph_decode_32(p); 511 r->steps[j].arg2 = ceph_decode_32(p); 512 } 513 } 514 515 ceph_decode_skip_map(p, end, 32, string, bad); /* type_map */ 516 ceph_decode_skip_map(p, end, 32, string, bad); /* name_map */ 517 ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */ 518 519 /* tunables */ 520 ceph_decode_need(p, end, 3*sizeof(u32), done); 521 c->choose_local_tries = ceph_decode_32(p); 522 c->choose_local_fallback_tries = ceph_decode_32(p); 523 c->choose_total_tries = ceph_decode_32(p); 524 dout("crush decode tunable choose_local_tries = %d\n", 525 c->choose_local_tries); 526 dout("crush decode tunable choose_local_fallback_tries = %d\n", 527 c->choose_local_fallback_tries); 528 dout("crush decode tunable choose_total_tries = %d\n", 529 c->choose_total_tries); 530 531 ceph_decode_need(p, end, sizeof(u32), done); 532 c->chooseleaf_descend_once = ceph_decode_32(p); 533 dout("crush decode tunable chooseleaf_descend_once = %d\n", 534 c->chooseleaf_descend_once); 535 536 ceph_decode_need(p, end, sizeof(u8), done); 537 c->chooseleaf_vary_r = ceph_decode_8(p); 538 dout("crush decode tunable chooseleaf_vary_r = %d\n", 539 c->chooseleaf_vary_r); 540 541 /* skip straw_calc_version, allowed_bucket_algs */ 542 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done); 543 *p += sizeof(u8) + sizeof(u32); 544 545 ceph_decode_need(p, end, sizeof(u8), done); 546 c->chooseleaf_stable = ceph_decode_8(p); 547 dout("crush decode tunable chooseleaf_stable = %d\n", 548 c->chooseleaf_stable); 549 550 if (*p != end) { 551 /* class_map */ 552 ceph_decode_skip_map(p, end, 32, 32, bad); 553 /* class_name */ 554 ceph_decode_skip_map(p, end, 32, string, bad); 555 /* class_bucket */ 556 ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad); 557 } 558 559 if (*p != end) { 560 err = decode_choose_args(p, end, c); 561 if (err) 562 goto fail; 563 } 564 565 done: 566 crush_finalize(c); 567 dout("crush_decode success\n"); 568 return c; 569 570 badmem: 571 err = -ENOMEM; 572 fail: 573 dout("crush_decode fail %d\n", err); 574 crush_destroy(c); 575 return ERR_PTR(err); 576 577 bad: 578 err = -EINVAL; 579 goto fail; 580 } 581 582 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs) 583 { 584 if (lhs->pool < rhs->pool) 585 return -1; 586 if (lhs->pool > rhs->pool) 587 return 1; 588 if (lhs->seed < rhs->seed) 589 return -1; 590 if (lhs->seed > rhs->seed) 591 return 1; 592 593 return 0; 594 } 595 596 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs) 597 { 598 int ret; 599 600 ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid); 601 if (ret) 602 return ret; 603 604 if (lhs->shard < rhs->shard) 605 return -1; 606 if (lhs->shard > rhs->shard) 607 return 1; 608 609 return 0; 610 } 611 612 static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len) 613 { 614 struct ceph_pg_mapping *pg; 615 616 pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO); 617 if (!pg) 618 return NULL; 619 620 RB_CLEAR_NODE(&pg->node); 621 return pg; 622 } 623 624 static void free_pg_mapping(struct ceph_pg_mapping *pg) 625 { 626 WARN_ON(!RB_EMPTY_NODE(&pg->node)); 627 628 kfree(pg); 629 } 630 631 /* 632 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid 633 * to a set of osds) and primary_temp (explicit primary setting) 634 */ 635 DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare, 636 RB_BYPTR, const struct ceph_pg *, node) 637 638 /* 639 * rbtree of pg pool info 640 */ 641 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new) 642 { 643 struct rb_node **p = &root->rb_node; 644 struct rb_node *parent = NULL; 645 struct ceph_pg_pool_info *pi = NULL; 646 647 while (*p) { 648 parent = *p; 649 pi = rb_entry(parent, struct ceph_pg_pool_info, node); 650 if (new->id < pi->id) 651 p = &(*p)->rb_left; 652 else if (new->id > pi->id) 653 p = &(*p)->rb_right; 654 else 655 return -EEXIST; 656 } 657 658 rb_link_node(&new->node, parent, p); 659 rb_insert_color(&new->node, root); 660 return 0; 661 } 662 663 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id) 664 { 665 struct ceph_pg_pool_info *pi; 666 struct rb_node *n = root->rb_node; 667 668 while (n) { 669 pi = rb_entry(n, struct ceph_pg_pool_info, node); 670 if (id < pi->id) 671 n = n->rb_left; 672 else if (id > pi->id) 673 n = n->rb_right; 674 else 675 return pi; 676 } 677 return NULL; 678 } 679 680 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id) 681 { 682 return __lookup_pg_pool(&map->pg_pools, id); 683 } 684 685 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id) 686 { 687 struct ceph_pg_pool_info *pi; 688 689 if (id == CEPH_NOPOOL) 690 return NULL; 691 692 if (WARN_ON_ONCE(id > (u64) INT_MAX)) 693 return NULL; 694 695 pi = __lookup_pg_pool(&map->pg_pools, (int) id); 696 697 return pi ? pi->name : NULL; 698 } 699 EXPORT_SYMBOL(ceph_pg_pool_name_by_id); 700 701 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name) 702 { 703 struct rb_node *rbp; 704 705 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) { 706 struct ceph_pg_pool_info *pi = 707 rb_entry(rbp, struct ceph_pg_pool_info, node); 708 if (pi->name && strcmp(pi->name, name) == 0) 709 return pi->id; 710 } 711 return -ENOENT; 712 } 713 EXPORT_SYMBOL(ceph_pg_poolid_by_name); 714 715 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi) 716 { 717 rb_erase(&pi->node, root); 718 kfree(pi->name); 719 kfree(pi); 720 } 721 722 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi) 723 { 724 u8 ev, cv; 725 unsigned len, num; 726 void *pool_end; 727 728 ceph_decode_need(p, end, 2 + 4, bad); 729 ev = ceph_decode_8(p); /* encoding version */ 730 cv = ceph_decode_8(p); /* compat version */ 731 if (ev < 5) { 732 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv); 733 return -EINVAL; 734 } 735 if (cv > 9) { 736 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv); 737 return -EINVAL; 738 } 739 len = ceph_decode_32(p); 740 ceph_decode_need(p, end, len, bad); 741 pool_end = *p + len; 742 743 pi->type = ceph_decode_8(p); 744 pi->size = ceph_decode_8(p); 745 pi->crush_ruleset = ceph_decode_8(p); 746 pi->object_hash = ceph_decode_8(p); 747 748 pi->pg_num = ceph_decode_32(p); 749 pi->pgp_num = ceph_decode_32(p); 750 751 *p += 4 + 4; /* skip lpg* */ 752 *p += 4; /* skip last_change */ 753 *p += 8 + 4; /* skip snap_seq, snap_epoch */ 754 755 /* skip snaps */ 756 num = ceph_decode_32(p); 757 while (num--) { 758 *p += 8; /* snapid key */ 759 *p += 1 + 1; /* versions */ 760 len = ceph_decode_32(p); 761 *p += len; 762 } 763 764 /* skip removed_snaps */ 765 num = ceph_decode_32(p); 766 *p += num * (8 + 8); 767 768 *p += 8; /* skip auid */ 769 pi->flags = ceph_decode_64(p); 770 *p += 4; /* skip crash_replay_interval */ 771 772 if (ev >= 7) 773 pi->min_size = ceph_decode_8(p); 774 else 775 pi->min_size = pi->size - pi->size / 2; 776 777 if (ev >= 8) 778 *p += 8 + 8; /* skip quota_max_* */ 779 780 if (ev >= 9) { 781 /* skip tiers */ 782 num = ceph_decode_32(p); 783 *p += num * 8; 784 785 *p += 8; /* skip tier_of */ 786 *p += 1; /* skip cache_mode */ 787 788 pi->read_tier = ceph_decode_64(p); 789 pi->write_tier = ceph_decode_64(p); 790 } else { 791 pi->read_tier = -1; 792 pi->write_tier = -1; 793 } 794 795 if (ev >= 10) { 796 /* skip properties */ 797 num = ceph_decode_32(p); 798 while (num--) { 799 len = ceph_decode_32(p); 800 *p += len; /* key */ 801 len = ceph_decode_32(p); 802 *p += len; /* val */ 803 } 804 } 805 806 if (ev >= 11) { 807 /* skip hit_set_params */ 808 *p += 1 + 1; /* versions */ 809 len = ceph_decode_32(p); 810 *p += len; 811 812 *p += 4; /* skip hit_set_period */ 813 *p += 4; /* skip hit_set_count */ 814 } 815 816 if (ev >= 12) 817 *p += 4; /* skip stripe_width */ 818 819 if (ev >= 13) { 820 *p += 8; /* skip target_max_bytes */ 821 *p += 8; /* skip target_max_objects */ 822 *p += 4; /* skip cache_target_dirty_ratio_micro */ 823 *p += 4; /* skip cache_target_full_ratio_micro */ 824 *p += 4; /* skip cache_min_flush_age */ 825 *p += 4; /* skip cache_min_evict_age */ 826 } 827 828 if (ev >= 14) { 829 /* skip erasure_code_profile */ 830 len = ceph_decode_32(p); 831 *p += len; 832 } 833 834 /* 835 * last_force_op_resend_preluminous, will be overridden if the 836 * map was encoded with RESEND_ON_SPLIT 837 */ 838 if (ev >= 15) 839 pi->last_force_request_resend = ceph_decode_32(p); 840 else 841 pi->last_force_request_resend = 0; 842 843 if (ev >= 16) 844 *p += 4; /* skip min_read_recency_for_promote */ 845 846 if (ev >= 17) 847 *p += 8; /* skip expected_num_objects */ 848 849 if (ev >= 19) 850 *p += 4; /* skip cache_target_dirty_high_ratio_micro */ 851 852 if (ev >= 20) 853 *p += 4; /* skip min_write_recency_for_promote */ 854 855 if (ev >= 21) 856 *p += 1; /* skip use_gmt_hitset */ 857 858 if (ev >= 22) 859 *p += 1; /* skip fast_read */ 860 861 if (ev >= 23) { 862 *p += 4; /* skip hit_set_grade_decay_rate */ 863 *p += 4; /* skip hit_set_search_last_n */ 864 } 865 866 if (ev >= 24) { 867 /* skip opts */ 868 *p += 1 + 1; /* versions */ 869 len = ceph_decode_32(p); 870 *p += len; 871 } 872 873 if (ev >= 25) 874 pi->last_force_request_resend = ceph_decode_32(p); 875 876 /* ignore the rest */ 877 878 *p = pool_end; 879 calc_pg_masks(pi); 880 return 0; 881 882 bad: 883 return -EINVAL; 884 } 885 886 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map) 887 { 888 struct ceph_pg_pool_info *pi; 889 u32 num, len; 890 u64 pool; 891 892 ceph_decode_32_safe(p, end, num, bad); 893 dout(" %d pool names\n", num); 894 while (num--) { 895 ceph_decode_64_safe(p, end, pool, bad); 896 ceph_decode_32_safe(p, end, len, bad); 897 dout(" pool %llu len %d\n", pool, len); 898 ceph_decode_need(p, end, len, bad); 899 pi = __lookup_pg_pool(&map->pg_pools, pool); 900 if (pi) { 901 char *name = kstrndup(*p, len, GFP_NOFS); 902 903 if (!name) 904 return -ENOMEM; 905 kfree(pi->name); 906 pi->name = name; 907 dout(" name is %s\n", pi->name); 908 } 909 *p += len; 910 } 911 return 0; 912 913 bad: 914 return -EINVAL; 915 } 916 917 /* 918 * osd map 919 */ 920 struct ceph_osdmap *ceph_osdmap_alloc(void) 921 { 922 struct ceph_osdmap *map; 923 924 map = kzalloc(sizeof(*map), GFP_NOIO); 925 if (!map) 926 return NULL; 927 928 map->pg_pools = RB_ROOT; 929 map->pool_max = -1; 930 map->pg_temp = RB_ROOT; 931 map->primary_temp = RB_ROOT; 932 map->pg_upmap = RB_ROOT; 933 map->pg_upmap_items = RB_ROOT; 934 mutex_init(&map->crush_workspace_mutex); 935 936 return map; 937 } 938 939 void ceph_osdmap_destroy(struct ceph_osdmap *map) 940 { 941 dout("osdmap_destroy %p\n", map); 942 if (map->crush) 943 crush_destroy(map->crush); 944 while (!RB_EMPTY_ROOT(&map->pg_temp)) { 945 struct ceph_pg_mapping *pg = 946 rb_entry(rb_first(&map->pg_temp), 947 struct ceph_pg_mapping, node); 948 erase_pg_mapping(&map->pg_temp, pg); 949 free_pg_mapping(pg); 950 } 951 while (!RB_EMPTY_ROOT(&map->primary_temp)) { 952 struct ceph_pg_mapping *pg = 953 rb_entry(rb_first(&map->primary_temp), 954 struct ceph_pg_mapping, node); 955 erase_pg_mapping(&map->primary_temp, pg); 956 free_pg_mapping(pg); 957 } 958 while (!RB_EMPTY_ROOT(&map->pg_upmap)) { 959 struct ceph_pg_mapping *pg = 960 rb_entry(rb_first(&map->pg_upmap), 961 struct ceph_pg_mapping, node); 962 rb_erase(&pg->node, &map->pg_upmap); 963 kfree(pg); 964 } 965 while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) { 966 struct ceph_pg_mapping *pg = 967 rb_entry(rb_first(&map->pg_upmap_items), 968 struct ceph_pg_mapping, node); 969 rb_erase(&pg->node, &map->pg_upmap_items); 970 kfree(pg); 971 } 972 while (!RB_EMPTY_ROOT(&map->pg_pools)) { 973 struct ceph_pg_pool_info *pi = 974 rb_entry(rb_first(&map->pg_pools), 975 struct ceph_pg_pool_info, node); 976 __remove_pg_pool(&map->pg_pools, pi); 977 } 978 kfree(map->osd_state); 979 kfree(map->osd_weight); 980 kfree(map->osd_addr); 981 kfree(map->osd_primary_affinity); 982 kfree(map->crush_workspace); 983 kfree(map); 984 } 985 986 /* 987 * Adjust max_osd value, (re)allocate arrays. 988 * 989 * The new elements are properly initialized. 990 */ 991 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max) 992 { 993 u32 *state; 994 u32 *weight; 995 struct ceph_entity_addr *addr; 996 int i; 997 998 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS); 999 if (!state) 1000 return -ENOMEM; 1001 map->osd_state = state; 1002 1003 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS); 1004 if (!weight) 1005 return -ENOMEM; 1006 map->osd_weight = weight; 1007 1008 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS); 1009 if (!addr) 1010 return -ENOMEM; 1011 map->osd_addr = addr; 1012 1013 for (i = map->max_osd; i < max; i++) { 1014 map->osd_state[i] = 0; 1015 map->osd_weight[i] = CEPH_OSD_OUT; 1016 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr)); 1017 } 1018 1019 if (map->osd_primary_affinity) { 1020 u32 *affinity; 1021 1022 affinity = krealloc(map->osd_primary_affinity, 1023 max*sizeof(*affinity), GFP_NOFS); 1024 if (!affinity) 1025 return -ENOMEM; 1026 map->osd_primary_affinity = affinity; 1027 1028 for (i = map->max_osd; i < max; i++) 1029 map->osd_primary_affinity[i] = 1030 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1031 } 1032 1033 map->max_osd = max; 1034 1035 return 0; 1036 } 1037 1038 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush) 1039 { 1040 void *workspace; 1041 size_t work_size; 1042 1043 if (IS_ERR(crush)) 1044 return PTR_ERR(crush); 1045 1046 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE); 1047 dout("%s work_size %zu bytes\n", __func__, work_size); 1048 workspace = kmalloc(work_size, GFP_NOIO); 1049 if (!workspace) { 1050 crush_destroy(crush); 1051 return -ENOMEM; 1052 } 1053 crush_init_workspace(crush, workspace); 1054 1055 if (map->crush) 1056 crush_destroy(map->crush); 1057 kfree(map->crush_workspace); 1058 map->crush = crush; 1059 map->crush_workspace = workspace; 1060 return 0; 1061 } 1062 1063 #define OSDMAP_WRAPPER_COMPAT_VER 7 1064 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1 1065 1066 /* 1067 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps, 1068 * to struct_v of the client_data section for new (v7 and above) 1069 * osdmaps. 1070 */ 1071 static int get_osdmap_client_data_v(void **p, void *end, 1072 const char *prefix, u8 *v) 1073 { 1074 u8 struct_v; 1075 1076 ceph_decode_8_safe(p, end, struct_v, e_inval); 1077 if (struct_v >= 7) { 1078 u8 struct_compat; 1079 1080 ceph_decode_8_safe(p, end, struct_compat, e_inval); 1081 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) { 1082 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n", 1083 struct_v, struct_compat, 1084 OSDMAP_WRAPPER_COMPAT_VER, prefix); 1085 return -EINVAL; 1086 } 1087 *p += 4; /* ignore wrapper struct_len */ 1088 1089 ceph_decode_8_safe(p, end, struct_v, e_inval); 1090 ceph_decode_8_safe(p, end, struct_compat, e_inval); 1091 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) { 1092 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n", 1093 struct_v, struct_compat, 1094 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix); 1095 return -EINVAL; 1096 } 1097 *p += 4; /* ignore client data struct_len */ 1098 } else { 1099 u16 version; 1100 1101 *p -= 1; 1102 ceph_decode_16_safe(p, end, version, e_inval); 1103 if (version < 6) { 1104 pr_warn("got v %d < 6 of %s ceph_osdmap\n", 1105 version, prefix); 1106 return -EINVAL; 1107 } 1108 1109 /* old osdmap enconding */ 1110 struct_v = 0; 1111 } 1112 1113 *v = struct_v; 1114 return 0; 1115 1116 e_inval: 1117 return -EINVAL; 1118 } 1119 1120 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map, 1121 bool incremental) 1122 { 1123 u32 n; 1124 1125 ceph_decode_32_safe(p, end, n, e_inval); 1126 while (n--) { 1127 struct ceph_pg_pool_info *pi; 1128 u64 pool; 1129 int ret; 1130 1131 ceph_decode_64_safe(p, end, pool, e_inval); 1132 1133 pi = __lookup_pg_pool(&map->pg_pools, pool); 1134 if (!incremental || !pi) { 1135 pi = kzalloc(sizeof(*pi), GFP_NOFS); 1136 if (!pi) 1137 return -ENOMEM; 1138 1139 pi->id = pool; 1140 1141 ret = __insert_pg_pool(&map->pg_pools, pi); 1142 if (ret) { 1143 kfree(pi); 1144 return ret; 1145 } 1146 } 1147 1148 ret = decode_pool(p, end, pi); 1149 if (ret) 1150 return ret; 1151 } 1152 1153 return 0; 1154 1155 e_inval: 1156 return -EINVAL; 1157 } 1158 1159 static int decode_pools(void **p, void *end, struct ceph_osdmap *map) 1160 { 1161 return __decode_pools(p, end, map, false); 1162 } 1163 1164 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map) 1165 { 1166 return __decode_pools(p, end, map, true); 1167 } 1168 1169 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool); 1170 1171 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root, 1172 decode_mapping_fn_t fn, bool incremental) 1173 { 1174 u32 n; 1175 1176 WARN_ON(!incremental && !fn); 1177 1178 ceph_decode_32_safe(p, end, n, e_inval); 1179 while (n--) { 1180 struct ceph_pg_mapping *pg; 1181 struct ceph_pg pgid; 1182 int ret; 1183 1184 ret = ceph_decode_pgid(p, end, &pgid); 1185 if (ret) 1186 return ret; 1187 1188 pg = lookup_pg_mapping(mapping_root, &pgid); 1189 if (pg) { 1190 WARN_ON(!incremental); 1191 erase_pg_mapping(mapping_root, pg); 1192 free_pg_mapping(pg); 1193 } 1194 1195 if (fn) { 1196 pg = fn(p, end, incremental); 1197 if (IS_ERR(pg)) 1198 return PTR_ERR(pg); 1199 1200 if (pg) { 1201 pg->pgid = pgid; /* struct */ 1202 insert_pg_mapping(mapping_root, pg); 1203 } 1204 } 1205 } 1206 1207 return 0; 1208 1209 e_inval: 1210 return -EINVAL; 1211 } 1212 1213 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end, 1214 bool incremental) 1215 { 1216 struct ceph_pg_mapping *pg; 1217 u32 len, i; 1218 1219 ceph_decode_32_safe(p, end, len, e_inval); 1220 if (len == 0 && incremental) 1221 return NULL; /* new_pg_temp: [] to remove */ 1222 if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32)) 1223 return ERR_PTR(-EINVAL); 1224 1225 ceph_decode_need(p, end, len * sizeof(u32), e_inval); 1226 pg = alloc_pg_mapping(len * sizeof(u32)); 1227 if (!pg) 1228 return ERR_PTR(-ENOMEM); 1229 1230 pg->pg_temp.len = len; 1231 for (i = 0; i < len; i++) 1232 pg->pg_temp.osds[i] = ceph_decode_32(p); 1233 1234 return pg; 1235 1236 e_inval: 1237 return ERR_PTR(-EINVAL); 1238 } 1239 1240 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map) 1241 { 1242 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp, 1243 false); 1244 } 1245 1246 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map) 1247 { 1248 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp, 1249 true); 1250 } 1251 1252 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end, 1253 bool incremental) 1254 { 1255 struct ceph_pg_mapping *pg; 1256 u32 osd; 1257 1258 ceph_decode_32_safe(p, end, osd, e_inval); 1259 if (osd == (u32)-1 && incremental) 1260 return NULL; /* new_primary_temp: -1 to remove */ 1261 1262 pg = alloc_pg_mapping(0); 1263 if (!pg) 1264 return ERR_PTR(-ENOMEM); 1265 1266 pg->primary_temp.osd = osd; 1267 return pg; 1268 1269 e_inval: 1270 return ERR_PTR(-EINVAL); 1271 } 1272 1273 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map) 1274 { 1275 return decode_pg_mapping(p, end, &map->primary_temp, 1276 __decode_primary_temp, false); 1277 } 1278 1279 static int decode_new_primary_temp(void **p, void *end, 1280 struct ceph_osdmap *map) 1281 { 1282 return decode_pg_mapping(p, end, &map->primary_temp, 1283 __decode_primary_temp, true); 1284 } 1285 1286 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd) 1287 { 1288 BUG_ON(osd >= map->max_osd); 1289 1290 if (!map->osd_primary_affinity) 1291 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1292 1293 return map->osd_primary_affinity[osd]; 1294 } 1295 1296 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff) 1297 { 1298 BUG_ON(osd >= map->max_osd); 1299 1300 if (!map->osd_primary_affinity) { 1301 int i; 1302 1303 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32), 1304 GFP_NOFS); 1305 if (!map->osd_primary_affinity) 1306 return -ENOMEM; 1307 1308 for (i = 0; i < map->max_osd; i++) 1309 map->osd_primary_affinity[i] = 1310 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1311 } 1312 1313 map->osd_primary_affinity[osd] = aff; 1314 1315 return 0; 1316 } 1317 1318 static int decode_primary_affinity(void **p, void *end, 1319 struct ceph_osdmap *map) 1320 { 1321 u32 len, i; 1322 1323 ceph_decode_32_safe(p, end, len, e_inval); 1324 if (len == 0) { 1325 kfree(map->osd_primary_affinity); 1326 map->osd_primary_affinity = NULL; 1327 return 0; 1328 } 1329 if (len != map->max_osd) 1330 goto e_inval; 1331 1332 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval); 1333 1334 for (i = 0; i < map->max_osd; i++) { 1335 int ret; 1336 1337 ret = set_primary_affinity(map, i, ceph_decode_32(p)); 1338 if (ret) 1339 return ret; 1340 } 1341 1342 return 0; 1343 1344 e_inval: 1345 return -EINVAL; 1346 } 1347 1348 static int decode_new_primary_affinity(void **p, void *end, 1349 struct ceph_osdmap *map) 1350 { 1351 u32 n; 1352 1353 ceph_decode_32_safe(p, end, n, e_inval); 1354 while (n--) { 1355 u32 osd, aff; 1356 int ret; 1357 1358 ceph_decode_32_safe(p, end, osd, e_inval); 1359 ceph_decode_32_safe(p, end, aff, e_inval); 1360 1361 ret = set_primary_affinity(map, osd, aff); 1362 if (ret) 1363 return ret; 1364 1365 pr_info("osd%d primary-affinity 0x%x\n", osd, aff); 1366 } 1367 1368 return 0; 1369 1370 e_inval: 1371 return -EINVAL; 1372 } 1373 1374 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end, 1375 bool __unused) 1376 { 1377 return __decode_pg_temp(p, end, false); 1378 } 1379 1380 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map) 1381 { 1382 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap, 1383 false); 1384 } 1385 1386 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map) 1387 { 1388 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap, 1389 true); 1390 } 1391 1392 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map) 1393 { 1394 return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true); 1395 } 1396 1397 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end, 1398 bool __unused) 1399 { 1400 struct ceph_pg_mapping *pg; 1401 u32 len, i; 1402 1403 ceph_decode_32_safe(p, end, len, e_inval); 1404 if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32))) 1405 return ERR_PTR(-EINVAL); 1406 1407 ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval); 1408 pg = alloc_pg_mapping(2 * len * sizeof(u32)); 1409 if (!pg) 1410 return ERR_PTR(-ENOMEM); 1411 1412 pg->pg_upmap_items.len = len; 1413 for (i = 0; i < len; i++) { 1414 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p); 1415 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p); 1416 } 1417 1418 return pg; 1419 1420 e_inval: 1421 return ERR_PTR(-EINVAL); 1422 } 1423 1424 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map) 1425 { 1426 return decode_pg_mapping(p, end, &map->pg_upmap_items, 1427 __decode_pg_upmap_items, false); 1428 } 1429 1430 static int decode_new_pg_upmap_items(void **p, void *end, 1431 struct ceph_osdmap *map) 1432 { 1433 return decode_pg_mapping(p, end, &map->pg_upmap_items, 1434 __decode_pg_upmap_items, true); 1435 } 1436 1437 static int decode_old_pg_upmap_items(void **p, void *end, 1438 struct ceph_osdmap *map) 1439 { 1440 return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true); 1441 } 1442 1443 /* 1444 * decode a full map. 1445 */ 1446 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map) 1447 { 1448 u8 struct_v; 1449 u32 epoch = 0; 1450 void *start = *p; 1451 u32 max; 1452 u32 len, i; 1453 int err; 1454 1455 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1456 1457 err = get_osdmap_client_data_v(p, end, "full", &struct_v); 1458 if (err) 1459 goto bad; 1460 1461 /* fsid, epoch, created, modified */ 1462 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) + 1463 sizeof(map->created) + sizeof(map->modified), e_inval); 1464 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid)); 1465 epoch = map->epoch = ceph_decode_32(p); 1466 ceph_decode_copy(p, &map->created, sizeof(map->created)); 1467 ceph_decode_copy(p, &map->modified, sizeof(map->modified)); 1468 1469 /* pools */ 1470 err = decode_pools(p, end, map); 1471 if (err) 1472 goto bad; 1473 1474 /* pool_name */ 1475 err = decode_pool_names(p, end, map); 1476 if (err) 1477 goto bad; 1478 1479 ceph_decode_32_safe(p, end, map->pool_max, e_inval); 1480 1481 ceph_decode_32_safe(p, end, map->flags, e_inval); 1482 1483 /* max_osd */ 1484 ceph_decode_32_safe(p, end, max, e_inval); 1485 1486 /* (re)alloc osd arrays */ 1487 err = osdmap_set_max_osd(map, max); 1488 if (err) 1489 goto bad; 1490 1491 /* osd_state, osd_weight, osd_addrs->client_addr */ 1492 ceph_decode_need(p, end, 3*sizeof(u32) + 1493 map->max_osd*((struct_v >= 5 ? sizeof(u32) : 1494 sizeof(u8)) + 1495 sizeof(*map->osd_weight) + 1496 sizeof(*map->osd_addr)), e_inval); 1497 1498 if (ceph_decode_32(p) != map->max_osd) 1499 goto e_inval; 1500 1501 if (struct_v >= 5) { 1502 for (i = 0; i < map->max_osd; i++) 1503 map->osd_state[i] = ceph_decode_32(p); 1504 } else { 1505 for (i = 0; i < map->max_osd; i++) 1506 map->osd_state[i] = ceph_decode_8(p); 1507 } 1508 1509 if (ceph_decode_32(p) != map->max_osd) 1510 goto e_inval; 1511 1512 for (i = 0; i < map->max_osd; i++) 1513 map->osd_weight[i] = ceph_decode_32(p); 1514 1515 if (ceph_decode_32(p) != map->max_osd) 1516 goto e_inval; 1517 1518 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr)); 1519 for (i = 0; i < map->max_osd; i++) 1520 ceph_decode_addr(&map->osd_addr[i]); 1521 1522 /* pg_temp */ 1523 err = decode_pg_temp(p, end, map); 1524 if (err) 1525 goto bad; 1526 1527 /* primary_temp */ 1528 if (struct_v >= 1) { 1529 err = decode_primary_temp(p, end, map); 1530 if (err) 1531 goto bad; 1532 } 1533 1534 /* primary_affinity */ 1535 if (struct_v >= 2) { 1536 err = decode_primary_affinity(p, end, map); 1537 if (err) 1538 goto bad; 1539 } else { 1540 WARN_ON(map->osd_primary_affinity); 1541 } 1542 1543 /* crush */ 1544 ceph_decode_32_safe(p, end, len, e_inval); 1545 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end))); 1546 if (err) 1547 goto bad; 1548 1549 *p += len; 1550 if (struct_v >= 3) { 1551 /* erasure_code_profiles */ 1552 ceph_decode_skip_map_of_map(p, end, string, string, string, 1553 e_inval); 1554 } 1555 1556 if (struct_v >= 4) { 1557 err = decode_pg_upmap(p, end, map); 1558 if (err) 1559 goto bad; 1560 1561 err = decode_pg_upmap_items(p, end, map); 1562 if (err) 1563 goto bad; 1564 } else { 1565 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap)); 1566 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items)); 1567 } 1568 1569 /* ignore the rest */ 1570 *p = end; 1571 1572 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1573 return 0; 1574 1575 e_inval: 1576 err = -EINVAL; 1577 bad: 1578 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1579 err, epoch, (int)(*p - start), *p, start, end); 1580 print_hex_dump(KERN_DEBUG, "osdmap: ", 1581 DUMP_PREFIX_OFFSET, 16, 1, 1582 start, end - start, true); 1583 return err; 1584 } 1585 1586 /* 1587 * Allocate and decode a full map. 1588 */ 1589 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end) 1590 { 1591 struct ceph_osdmap *map; 1592 int ret; 1593 1594 map = ceph_osdmap_alloc(); 1595 if (!map) 1596 return ERR_PTR(-ENOMEM); 1597 1598 ret = osdmap_decode(p, end, map); 1599 if (ret) { 1600 ceph_osdmap_destroy(map); 1601 return ERR_PTR(ret); 1602 } 1603 1604 return map; 1605 } 1606 1607 /* 1608 * Encoding order is (new_up_client, new_state, new_weight). Need to 1609 * apply in the (new_weight, new_state, new_up_client) order, because 1610 * an incremental map may look like e.g. 1611 * 1612 * new_up_client: { osd=6, addr=... } # set osd_state and addr 1613 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state 1614 */ 1615 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v, 1616 struct ceph_osdmap *map) 1617 { 1618 void *new_up_client; 1619 void *new_state; 1620 void *new_weight_end; 1621 u32 len; 1622 1623 new_up_client = *p; 1624 ceph_decode_32_safe(p, end, len, e_inval); 1625 len *= sizeof(u32) + sizeof(struct ceph_entity_addr); 1626 ceph_decode_need(p, end, len, e_inval); 1627 *p += len; 1628 1629 new_state = *p; 1630 ceph_decode_32_safe(p, end, len, e_inval); 1631 len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8)); 1632 ceph_decode_need(p, end, len, e_inval); 1633 *p += len; 1634 1635 /* new_weight */ 1636 ceph_decode_32_safe(p, end, len, e_inval); 1637 while (len--) { 1638 s32 osd; 1639 u32 w; 1640 1641 ceph_decode_need(p, end, 2*sizeof(u32), e_inval); 1642 osd = ceph_decode_32(p); 1643 w = ceph_decode_32(p); 1644 BUG_ON(osd >= map->max_osd); 1645 pr_info("osd%d weight 0x%x %s\n", osd, w, 1646 w == CEPH_OSD_IN ? "(in)" : 1647 (w == CEPH_OSD_OUT ? "(out)" : "")); 1648 map->osd_weight[osd] = w; 1649 1650 /* 1651 * If we are marking in, set the EXISTS, and clear the 1652 * AUTOOUT and NEW bits. 1653 */ 1654 if (w) { 1655 map->osd_state[osd] |= CEPH_OSD_EXISTS; 1656 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT | 1657 CEPH_OSD_NEW); 1658 } 1659 } 1660 new_weight_end = *p; 1661 1662 /* new_state (up/down) */ 1663 *p = new_state; 1664 len = ceph_decode_32(p); 1665 while (len--) { 1666 s32 osd; 1667 u32 xorstate; 1668 int ret; 1669 1670 osd = ceph_decode_32(p); 1671 if (struct_v >= 5) 1672 xorstate = ceph_decode_32(p); 1673 else 1674 xorstate = ceph_decode_8(p); 1675 if (xorstate == 0) 1676 xorstate = CEPH_OSD_UP; 1677 BUG_ON(osd >= map->max_osd); 1678 if ((map->osd_state[osd] & CEPH_OSD_UP) && 1679 (xorstate & CEPH_OSD_UP)) 1680 pr_info("osd%d down\n", osd); 1681 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) && 1682 (xorstate & CEPH_OSD_EXISTS)) { 1683 pr_info("osd%d does not exist\n", osd); 1684 ret = set_primary_affinity(map, osd, 1685 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY); 1686 if (ret) 1687 return ret; 1688 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr)); 1689 map->osd_state[osd] = 0; 1690 } else { 1691 map->osd_state[osd] ^= xorstate; 1692 } 1693 } 1694 1695 /* new_up_client */ 1696 *p = new_up_client; 1697 len = ceph_decode_32(p); 1698 while (len--) { 1699 s32 osd; 1700 struct ceph_entity_addr addr; 1701 1702 osd = ceph_decode_32(p); 1703 ceph_decode_copy(p, &addr, sizeof(addr)); 1704 ceph_decode_addr(&addr); 1705 BUG_ON(osd >= map->max_osd); 1706 pr_info("osd%d up\n", osd); 1707 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP; 1708 map->osd_addr[osd] = addr; 1709 } 1710 1711 *p = new_weight_end; 1712 return 0; 1713 1714 e_inval: 1715 return -EINVAL; 1716 } 1717 1718 /* 1719 * decode and apply an incremental map update. 1720 */ 1721 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 1722 struct ceph_osdmap *map) 1723 { 1724 struct ceph_fsid fsid; 1725 u32 epoch = 0; 1726 struct ceph_timespec modified; 1727 s32 len; 1728 u64 pool; 1729 __s64 new_pool_max; 1730 __s32 new_flags, max; 1731 void *start = *p; 1732 int err; 1733 u8 struct_v; 1734 1735 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1736 1737 err = get_osdmap_client_data_v(p, end, "inc", &struct_v); 1738 if (err) 1739 goto bad; 1740 1741 /* fsid, epoch, modified, new_pool_max, new_flags */ 1742 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) + 1743 sizeof(u64) + sizeof(u32), e_inval); 1744 ceph_decode_copy(p, &fsid, sizeof(fsid)); 1745 epoch = ceph_decode_32(p); 1746 BUG_ON(epoch != map->epoch+1); 1747 ceph_decode_copy(p, &modified, sizeof(modified)); 1748 new_pool_max = ceph_decode_64(p); 1749 new_flags = ceph_decode_32(p); 1750 1751 /* full map? */ 1752 ceph_decode_32_safe(p, end, len, e_inval); 1753 if (len > 0) { 1754 dout("apply_incremental full map len %d, %p to %p\n", 1755 len, *p, end); 1756 return ceph_osdmap_decode(p, min(*p+len, end)); 1757 } 1758 1759 /* new crush? */ 1760 ceph_decode_32_safe(p, end, len, e_inval); 1761 if (len > 0) { 1762 err = osdmap_set_crush(map, 1763 crush_decode(*p, min(*p + len, end))); 1764 if (err) 1765 goto bad; 1766 *p += len; 1767 } 1768 1769 /* new flags? */ 1770 if (new_flags >= 0) 1771 map->flags = new_flags; 1772 if (new_pool_max >= 0) 1773 map->pool_max = new_pool_max; 1774 1775 /* new max? */ 1776 ceph_decode_32_safe(p, end, max, e_inval); 1777 if (max >= 0) { 1778 err = osdmap_set_max_osd(map, max); 1779 if (err) 1780 goto bad; 1781 } 1782 1783 map->epoch++; 1784 map->modified = modified; 1785 1786 /* new_pools */ 1787 err = decode_new_pools(p, end, map); 1788 if (err) 1789 goto bad; 1790 1791 /* new_pool_names */ 1792 err = decode_pool_names(p, end, map); 1793 if (err) 1794 goto bad; 1795 1796 /* old_pool */ 1797 ceph_decode_32_safe(p, end, len, e_inval); 1798 while (len--) { 1799 struct ceph_pg_pool_info *pi; 1800 1801 ceph_decode_64_safe(p, end, pool, e_inval); 1802 pi = __lookup_pg_pool(&map->pg_pools, pool); 1803 if (pi) 1804 __remove_pg_pool(&map->pg_pools, pi); 1805 } 1806 1807 /* new_up_client, new_state, new_weight */ 1808 err = decode_new_up_state_weight(p, end, struct_v, map); 1809 if (err) 1810 goto bad; 1811 1812 /* new_pg_temp */ 1813 err = decode_new_pg_temp(p, end, map); 1814 if (err) 1815 goto bad; 1816 1817 /* new_primary_temp */ 1818 if (struct_v >= 1) { 1819 err = decode_new_primary_temp(p, end, map); 1820 if (err) 1821 goto bad; 1822 } 1823 1824 /* new_primary_affinity */ 1825 if (struct_v >= 2) { 1826 err = decode_new_primary_affinity(p, end, map); 1827 if (err) 1828 goto bad; 1829 } 1830 1831 if (struct_v >= 3) { 1832 /* new_erasure_code_profiles */ 1833 ceph_decode_skip_map_of_map(p, end, string, string, string, 1834 e_inval); 1835 /* old_erasure_code_profiles */ 1836 ceph_decode_skip_set(p, end, string, e_inval); 1837 } 1838 1839 if (struct_v >= 4) { 1840 err = decode_new_pg_upmap(p, end, map); 1841 if (err) 1842 goto bad; 1843 1844 err = decode_old_pg_upmap(p, end, map); 1845 if (err) 1846 goto bad; 1847 1848 err = decode_new_pg_upmap_items(p, end, map); 1849 if (err) 1850 goto bad; 1851 1852 err = decode_old_pg_upmap_items(p, end, map); 1853 if (err) 1854 goto bad; 1855 } 1856 1857 /* ignore the rest */ 1858 *p = end; 1859 1860 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1861 return map; 1862 1863 e_inval: 1864 err = -EINVAL; 1865 bad: 1866 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1867 err, epoch, (int)(*p - start), *p, start, end); 1868 print_hex_dump(KERN_DEBUG, "osdmap: ", 1869 DUMP_PREFIX_OFFSET, 16, 1, 1870 start, end - start, true); 1871 return ERR_PTR(err); 1872 } 1873 1874 void ceph_oloc_copy(struct ceph_object_locator *dest, 1875 const struct ceph_object_locator *src) 1876 { 1877 ceph_oloc_destroy(dest); 1878 1879 dest->pool = src->pool; 1880 if (src->pool_ns) 1881 dest->pool_ns = ceph_get_string(src->pool_ns); 1882 else 1883 dest->pool_ns = NULL; 1884 } 1885 EXPORT_SYMBOL(ceph_oloc_copy); 1886 1887 void ceph_oloc_destroy(struct ceph_object_locator *oloc) 1888 { 1889 ceph_put_string(oloc->pool_ns); 1890 } 1891 EXPORT_SYMBOL(ceph_oloc_destroy); 1892 1893 void ceph_oid_copy(struct ceph_object_id *dest, 1894 const struct ceph_object_id *src) 1895 { 1896 ceph_oid_destroy(dest); 1897 1898 if (src->name != src->inline_name) { 1899 /* very rare, see ceph_object_id definition */ 1900 dest->name = kmalloc(src->name_len + 1, 1901 GFP_NOIO | __GFP_NOFAIL); 1902 } else { 1903 dest->name = dest->inline_name; 1904 } 1905 memcpy(dest->name, src->name, src->name_len + 1); 1906 dest->name_len = src->name_len; 1907 } 1908 EXPORT_SYMBOL(ceph_oid_copy); 1909 1910 static __printf(2, 0) 1911 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap) 1912 { 1913 int len; 1914 1915 WARN_ON(!ceph_oid_empty(oid)); 1916 1917 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap); 1918 if (len >= sizeof(oid->inline_name)) 1919 return len; 1920 1921 oid->name_len = len; 1922 return 0; 1923 } 1924 1925 /* 1926 * If oid doesn't fit into inline buffer, BUG. 1927 */ 1928 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...) 1929 { 1930 va_list ap; 1931 1932 va_start(ap, fmt); 1933 BUG_ON(oid_printf_vargs(oid, fmt, ap)); 1934 va_end(ap); 1935 } 1936 EXPORT_SYMBOL(ceph_oid_printf); 1937 1938 static __printf(3, 0) 1939 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp, 1940 const char *fmt, va_list ap) 1941 { 1942 va_list aq; 1943 int len; 1944 1945 va_copy(aq, ap); 1946 len = oid_printf_vargs(oid, fmt, aq); 1947 va_end(aq); 1948 1949 if (len) { 1950 char *external_name; 1951 1952 external_name = kmalloc(len + 1, gfp); 1953 if (!external_name) 1954 return -ENOMEM; 1955 1956 oid->name = external_name; 1957 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len); 1958 oid->name_len = len; 1959 } 1960 1961 return 0; 1962 } 1963 1964 /* 1965 * If oid doesn't fit into inline buffer, allocate. 1966 */ 1967 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp, 1968 const char *fmt, ...) 1969 { 1970 va_list ap; 1971 int ret; 1972 1973 va_start(ap, fmt); 1974 ret = oid_aprintf_vargs(oid, gfp, fmt, ap); 1975 va_end(ap); 1976 1977 return ret; 1978 } 1979 EXPORT_SYMBOL(ceph_oid_aprintf); 1980 1981 void ceph_oid_destroy(struct ceph_object_id *oid) 1982 { 1983 if (oid->name != oid->inline_name) 1984 kfree(oid->name); 1985 } 1986 EXPORT_SYMBOL(ceph_oid_destroy); 1987 1988 /* 1989 * osds only 1990 */ 1991 static bool __osds_equal(const struct ceph_osds *lhs, 1992 const struct ceph_osds *rhs) 1993 { 1994 if (lhs->size == rhs->size && 1995 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0]))) 1996 return true; 1997 1998 return false; 1999 } 2000 2001 /* 2002 * osds + primary 2003 */ 2004 static bool osds_equal(const struct ceph_osds *lhs, 2005 const struct ceph_osds *rhs) 2006 { 2007 if (__osds_equal(lhs, rhs) && 2008 lhs->primary == rhs->primary) 2009 return true; 2010 2011 return false; 2012 } 2013 2014 static bool osds_valid(const struct ceph_osds *set) 2015 { 2016 /* non-empty set */ 2017 if (set->size > 0 && set->primary >= 0) 2018 return true; 2019 2020 /* empty can_shift_osds set */ 2021 if (!set->size && set->primary == -1) 2022 return true; 2023 2024 /* empty !can_shift_osds set - all NONE */ 2025 if (set->size > 0 && set->primary == -1) { 2026 int i; 2027 2028 for (i = 0; i < set->size; i++) { 2029 if (set->osds[i] != CRUSH_ITEM_NONE) 2030 break; 2031 } 2032 if (i == set->size) 2033 return true; 2034 } 2035 2036 return false; 2037 } 2038 2039 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src) 2040 { 2041 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0])); 2042 dest->size = src->size; 2043 dest->primary = src->primary; 2044 } 2045 2046 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num, 2047 u32 new_pg_num) 2048 { 2049 int old_bits = calc_bits_of(old_pg_num); 2050 int old_mask = (1 << old_bits) - 1; 2051 int n; 2052 2053 WARN_ON(pgid->seed >= old_pg_num); 2054 if (new_pg_num <= old_pg_num) 2055 return false; 2056 2057 for (n = 1; ; n++) { 2058 int next_bit = n << (old_bits - 1); 2059 u32 s = next_bit | pgid->seed; 2060 2061 if (s < old_pg_num || s == pgid->seed) 2062 continue; 2063 if (s >= new_pg_num) 2064 break; 2065 2066 s = ceph_stable_mod(s, old_pg_num, old_mask); 2067 if (s == pgid->seed) 2068 return true; 2069 } 2070 2071 return false; 2072 } 2073 2074 bool ceph_is_new_interval(const struct ceph_osds *old_acting, 2075 const struct ceph_osds *new_acting, 2076 const struct ceph_osds *old_up, 2077 const struct ceph_osds *new_up, 2078 int old_size, 2079 int new_size, 2080 int old_min_size, 2081 int new_min_size, 2082 u32 old_pg_num, 2083 u32 new_pg_num, 2084 bool old_sort_bitwise, 2085 bool new_sort_bitwise, 2086 bool old_recovery_deletes, 2087 bool new_recovery_deletes, 2088 const struct ceph_pg *pgid) 2089 { 2090 return !osds_equal(old_acting, new_acting) || 2091 !osds_equal(old_up, new_up) || 2092 old_size != new_size || 2093 old_min_size != new_min_size || 2094 ceph_pg_is_split(pgid, old_pg_num, new_pg_num) || 2095 old_sort_bitwise != new_sort_bitwise || 2096 old_recovery_deletes != new_recovery_deletes; 2097 } 2098 2099 static int calc_pg_rank(int osd, const struct ceph_osds *acting) 2100 { 2101 int i; 2102 2103 for (i = 0; i < acting->size; i++) { 2104 if (acting->osds[i] == osd) 2105 return i; 2106 } 2107 2108 return -1; 2109 } 2110 2111 static bool primary_changed(const struct ceph_osds *old_acting, 2112 const struct ceph_osds *new_acting) 2113 { 2114 if (!old_acting->size && !new_acting->size) 2115 return false; /* both still empty */ 2116 2117 if (!old_acting->size ^ !new_acting->size) 2118 return true; /* was empty, now not, or vice versa */ 2119 2120 if (old_acting->primary != new_acting->primary) 2121 return true; /* primary changed */ 2122 2123 if (calc_pg_rank(old_acting->primary, old_acting) != 2124 calc_pg_rank(new_acting->primary, new_acting)) 2125 return true; 2126 2127 return false; /* same primary (tho replicas may have changed) */ 2128 } 2129 2130 bool ceph_osds_changed(const struct ceph_osds *old_acting, 2131 const struct ceph_osds *new_acting, 2132 bool any_change) 2133 { 2134 if (primary_changed(old_acting, new_acting)) 2135 return true; 2136 2137 if (any_change && !__osds_equal(old_acting, new_acting)) 2138 return true; 2139 2140 return false; 2141 } 2142 2143 /* 2144 * Map a file extent to a stripe unit within an object. 2145 * Fill in objno, offset into object, and object extent length (i.e. the 2146 * number of bytes mapped, less than or equal to @l->stripe_unit). 2147 * 2148 * Example for stripe_count = 3, stripes_per_object = 4: 2149 * 2150 * blockno | 0 3 6 9 | 1 4 7 10 | 2 5 8 11 | 12 15 18 21 | 13 16 19 2151 * stripeno | 0 1 2 3 | 0 1 2 3 | 0 1 2 3 | 4 5 6 7 | 4 5 6 2152 * stripepos | 0 | 1 | 2 | 0 | 1 2153 * objno | 0 | 1 | 2 | 3 | 4 2154 * objsetno | 0 | 1 2155 */ 2156 void ceph_calc_file_object_mapping(struct ceph_file_layout *l, 2157 u64 off, u64 len, 2158 u64 *objno, u64 *objoff, u32 *xlen) 2159 { 2160 u32 stripes_per_object = l->object_size / l->stripe_unit; 2161 u64 blockno; /* which su in the file (i.e. globally) */ 2162 u32 blockoff; /* offset into su */ 2163 u64 stripeno; /* which stripe */ 2164 u32 stripepos; /* which su in the stripe, 2165 which object in the object set */ 2166 u64 objsetno; /* which object set */ 2167 u32 objsetpos; /* which stripe in the object set */ 2168 2169 blockno = div_u64_rem(off, l->stripe_unit, &blockoff); 2170 stripeno = div_u64_rem(blockno, l->stripe_count, &stripepos); 2171 objsetno = div_u64_rem(stripeno, stripes_per_object, &objsetpos); 2172 2173 *objno = objsetno * l->stripe_count + stripepos; 2174 *objoff = objsetpos * l->stripe_unit + blockoff; 2175 *xlen = min_t(u64, len, l->stripe_unit - blockoff); 2176 } 2177 EXPORT_SYMBOL(ceph_calc_file_object_mapping); 2178 2179 /* 2180 * Map an object into a PG. 2181 * 2182 * Should only be called with target_oid and target_oloc (as opposed to 2183 * base_oid and base_oloc), since tiering isn't taken into account. 2184 */ 2185 int __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi, 2186 const struct ceph_object_id *oid, 2187 const struct ceph_object_locator *oloc, 2188 struct ceph_pg *raw_pgid) 2189 { 2190 WARN_ON(pi->id != oloc->pool); 2191 2192 if (!oloc->pool_ns) { 2193 raw_pgid->pool = oloc->pool; 2194 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name, 2195 oid->name_len); 2196 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name, 2197 raw_pgid->pool, raw_pgid->seed); 2198 } else { 2199 char stack_buf[256]; 2200 char *buf = stack_buf; 2201 int nsl = oloc->pool_ns->len; 2202 size_t total = nsl + 1 + oid->name_len; 2203 2204 if (total > sizeof(stack_buf)) { 2205 buf = kmalloc(total, GFP_NOIO); 2206 if (!buf) 2207 return -ENOMEM; 2208 } 2209 memcpy(buf, oloc->pool_ns->str, nsl); 2210 buf[nsl] = '\037'; 2211 memcpy(buf + nsl + 1, oid->name, oid->name_len); 2212 raw_pgid->pool = oloc->pool; 2213 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total); 2214 if (buf != stack_buf) 2215 kfree(buf); 2216 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__, 2217 oid->name, nsl, oloc->pool_ns->str, 2218 raw_pgid->pool, raw_pgid->seed); 2219 } 2220 return 0; 2221 } 2222 2223 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 2224 const struct ceph_object_id *oid, 2225 const struct ceph_object_locator *oloc, 2226 struct ceph_pg *raw_pgid) 2227 { 2228 struct ceph_pg_pool_info *pi; 2229 2230 pi = ceph_pg_pool_by_id(osdmap, oloc->pool); 2231 if (!pi) 2232 return -ENOENT; 2233 2234 return __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid); 2235 } 2236 EXPORT_SYMBOL(ceph_object_locator_to_pg); 2237 2238 /* 2239 * Map a raw PG (full precision ps) into an actual PG. 2240 */ 2241 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi, 2242 const struct ceph_pg *raw_pgid, 2243 struct ceph_pg *pgid) 2244 { 2245 pgid->pool = raw_pgid->pool; 2246 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num, 2247 pi->pg_num_mask); 2248 } 2249 2250 /* 2251 * Map a raw PG (full precision ps) into a placement ps (placement 2252 * seed). Include pool id in that value so that different pools don't 2253 * use the same seeds. 2254 */ 2255 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi, 2256 const struct ceph_pg *raw_pgid) 2257 { 2258 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) { 2259 /* hash pool id and seed so that pool PGs do not overlap */ 2260 return crush_hash32_2(CRUSH_HASH_RJENKINS1, 2261 ceph_stable_mod(raw_pgid->seed, 2262 pi->pgp_num, 2263 pi->pgp_num_mask), 2264 raw_pgid->pool); 2265 } else { 2266 /* 2267 * legacy behavior: add ps and pool together. this is 2268 * not a great approach because the PGs from each pool 2269 * will overlap on top of each other: 0.5 == 1.4 == 2270 * 2.3 == ... 2271 */ 2272 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num, 2273 pi->pgp_num_mask) + 2274 (unsigned)raw_pgid->pool; 2275 } 2276 } 2277 2278 /* 2279 * Magic value used for a "default" fallback choose_args, used if the 2280 * crush_choose_arg_map passed to do_crush() does not exist. If this 2281 * also doesn't exist, fall back to canonical weights. 2282 */ 2283 #define CEPH_DEFAULT_CHOOSE_ARGS -1 2284 2285 static int do_crush(struct ceph_osdmap *map, int ruleno, int x, 2286 int *result, int result_max, 2287 const __u32 *weight, int weight_max, 2288 s64 choose_args_index) 2289 { 2290 struct crush_choose_arg_map *arg_map; 2291 int r; 2292 2293 BUG_ON(result_max > CEPH_PG_MAX_SIZE); 2294 2295 arg_map = lookup_choose_arg_map(&map->crush->choose_args, 2296 choose_args_index); 2297 if (!arg_map) 2298 arg_map = lookup_choose_arg_map(&map->crush->choose_args, 2299 CEPH_DEFAULT_CHOOSE_ARGS); 2300 2301 mutex_lock(&map->crush_workspace_mutex); 2302 r = crush_do_rule(map->crush, ruleno, x, result, result_max, 2303 weight, weight_max, map->crush_workspace, 2304 arg_map ? arg_map->args : NULL); 2305 mutex_unlock(&map->crush_workspace_mutex); 2306 2307 return r; 2308 } 2309 2310 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap, 2311 struct ceph_pg_pool_info *pi, 2312 struct ceph_osds *set) 2313 { 2314 int i; 2315 2316 if (ceph_can_shift_osds(pi)) { 2317 int removed = 0; 2318 2319 /* shift left */ 2320 for (i = 0; i < set->size; i++) { 2321 if (!ceph_osd_exists(osdmap, set->osds[i])) { 2322 removed++; 2323 continue; 2324 } 2325 if (removed) 2326 set->osds[i - removed] = set->osds[i]; 2327 } 2328 set->size -= removed; 2329 } else { 2330 /* set dne devices to NONE */ 2331 for (i = 0; i < set->size; i++) { 2332 if (!ceph_osd_exists(osdmap, set->osds[i])) 2333 set->osds[i] = CRUSH_ITEM_NONE; 2334 } 2335 } 2336 } 2337 2338 /* 2339 * Calculate raw set (CRUSH output) for given PG and filter out 2340 * nonexistent OSDs. ->primary is undefined for a raw set. 2341 * 2342 * Placement seed (CRUSH input) is returned through @ppps. 2343 */ 2344 static void pg_to_raw_osds(struct ceph_osdmap *osdmap, 2345 struct ceph_pg_pool_info *pi, 2346 const struct ceph_pg *raw_pgid, 2347 struct ceph_osds *raw, 2348 u32 *ppps) 2349 { 2350 u32 pps = raw_pg_to_pps(pi, raw_pgid); 2351 int ruleno; 2352 int len; 2353 2354 ceph_osds_init(raw); 2355 if (ppps) 2356 *ppps = pps; 2357 2358 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type, 2359 pi->size); 2360 if (ruleno < 0) { 2361 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n", 2362 pi->id, pi->crush_ruleset, pi->type, pi->size); 2363 return; 2364 } 2365 2366 if (pi->size > ARRAY_SIZE(raw->osds)) { 2367 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n", 2368 pi->id, pi->crush_ruleset, pi->type, pi->size, 2369 ARRAY_SIZE(raw->osds)); 2370 return; 2371 } 2372 2373 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size, 2374 osdmap->osd_weight, osdmap->max_osd, pi->id); 2375 if (len < 0) { 2376 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n", 2377 len, ruleno, pi->id, pi->crush_ruleset, pi->type, 2378 pi->size); 2379 return; 2380 } 2381 2382 raw->size = len; 2383 remove_nonexistent_osds(osdmap, pi, raw); 2384 } 2385 2386 /* apply pg_upmap[_items] mappings */ 2387 static void apply_upmap(struct ceph_osdmap *osdmap, 2388 const struct ceph_pg *pgid, 2389 struct ceph_osds *raw) 2390 { 2391 struct ceph_pg_mapping *pg; 2392 int i, j; 2393 2394 pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid); 2395 if (pg) { 2396 /* make sure targets aren't marked out */ 2397 for (i = 0; i < pg->pg_upmap.len; i++) { 2398 int osd = pg->pg_upmap.osds[i]; 2399 2400 if (osd != CRUSH_ITEM_NONE && 2401 osd < osdmap->max_osd && 2402 osdmap->osd_weight[osd] == 0) { 2403 /* reject/ignore explicit mapping */ 2404 return; 2405 } 2406 } 2407 for (i = 0; i < pg->pg_upmap.len; i++) 2408 raw->osds[i] = pg->pg_upmap.osds[i]; 2409 raw->size = pg->pg_upmap.len; 2410 /* check and apply pg_upmap_items, if any */ 2411 } 2412 2413 pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid); 2414 if (pg) { 2415 /* 2416 * Note: this approach does not allow a bidirectional swap, 2417 * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1]. 2418 */ 2419 for (i = 0; i < pg->pg_upmap_items.len; i++) { 2420 int from = pg->pg_upmap_items.from_to[i][0]; 2421 int to = pg->pg_upmap_items.from_to[i][1]; 2422 int pos = -1; 2423 bool exists = false; 2424 2425 /* make sure replacement doesn't already appear */ 2426 for (j = 0; j < raw->size; j++) { 2427 int osd = raw->osds[j]; 2428 2429 if (osd == to) { 2430 exists = true; 2431 break; 2432 } 2433 /* ignore mapping if target is marked out */ 2434 if (osd == from && pos < 0 && 2435 !(to != CRUSH_ITEM_NONE && 2436 to < osdmap->max_osd && 2437 osdmap->osd_weight[to] == 0)) { 2438 pos = j; 2439 } 2440 } 2441 if (!exists && pos >= 0) 2442 raw->osds[pos] = to; 2443 } 2444 } 2445 } 2446 2447 /* 2448 * Given raw set, calculate up set and up primary. By definition of an 2449 * up set, the result won't contain nonexistent or down OSDs. 2450 * 2451 * This is done in-place - on return @set is the up set. If it's 2452 * empty, ->primary will remain undefined. 2453 */ 2454 static void raw_to_up_osds(struct ceph_osdmap *osdmap, 2455 struct ceph_pg_pool_info *pi, 2456 struct ceph_osds *set) 2457 { 2458 int i; 2459 2460 /* ->primary is undefined for a raw set */ 2461 BUG_ON(set->primary != -1); 2462 2463 if (ceph_can_shift_osds(pi)) { 2464 int removed = 0; 2465 2466 /* shift left */ 2467 for (i = 0; i < set->size; i++) { 2468 if (ceph_osd_is_down(osdmap, set->osds[i])) { 2469 removed++; 2470 continue; 2471 } 2472 if (removed) 2473 set->osds[i - removed] = set->osds[i]; 2474 } 2475 set->size -= removed; 2476 if (set->size > 0) 2477 set->primary = set->osds[0]; 2478 } else { 2479 /* set down/dne devices to NONE */ 2480 for (i = set->size - 1; i >= 0; i--) { 2481 if (ceph_osd_is_down(osdmap, set->osds[i])) 2482 set->osds[i] = CRUSH_ITEM_NONE; 2483 else 2484 set->primary = set->osds[i]; 2485 } 2486 } 2487 } 2488 2489 static void apply_primary_affinity(struct ceph_osdmap *osdmap, 2490 struct ceph_pg_pool_info *pi, 2491 u32 pps, 2492 struct ceph_osds *up) 2493 { 2494 int i; 2495 int pos = -1; 2496 2497 /* 2498 * Do we have any non-default primary_affinity values for these 2499 * osds? 2500 */ 2501 if (!osdmap->osd_primary_affinity) 2502 return; 2503 2504 for (i = 0; i < up->size; i++) { 2505 int osd = up->osds[i]; 2506 2507 if (osd != CRUSH_ITEM_NONE && 2508 osdmap->osd_primary_affinity[osd] != 2509 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) { 2510 break; 2511 } 2512 } 2513 if (i == up->size) 2514 return; 2515 2516 /* 2517 * Pick the primary. Feed both the seed (for the pg) and the 2518 * osd into the hash/rng so that a proportional fraction of an 2519 * osd's pgs get rejected as primary. 2520 */ 2521 for (i = 0; i < up->size; i++) { 2522 int osd = up->osds[i]; 2523 u32 aff; 2524 2525 if (osd == CRUSH_ITEM_NONE) 2526 continue; 2527 2528 aff = osdmap->osd_primary_affinity[osd]; 2529 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY && 2530 (crush_hash32_2(CRUSH_HASH_RJENKINS1, 2531 pps, osd) >> 16) >= aff) { 2532 /* 2533 * We chose not to use this primary. Note it 2534 * anyway as a fallback in case we don't pick 2535 * anyone else, but keep looking. 2536 */ 2537 if (pos < 0) 2538 pos = i; 2539 } else { 2540 pos = i; 2541 break; 2542 } 2543 } 2544 if (pos < 0) 2545 return; 2546 2547 up->primary = up->osds[pos]; 2548 2549 if (ceph_can_shift_osds(pi) && pos > 0) { 2550 /* move the new primary to the front */ 2551 for (i = pos; i > 0; i--) 2552 up->osds[i] = up->osds[i - 1]; 2553 up->osds[0] = up->primary; 2554 } 2555 } 2556 2557 /* 2558 * Get pg_temp and primary_temp mappings for given PG. 2559 * 2560 * Note that a PG may have none, only pg_temp, only primary_temp or 2561 * both pg_temp and primary_temp mappings. This means @temp isn't 2562 * always a valid OSD set on return: in the "only primary_temp" case, 2563 * @temp will have its ->primary >= 0 but ->size == 0. 2564 */ 2565 static void get_temp_osds(struct ceph_osdmap *osdmap, 2566 struct ceph_pg_pool_info *pi, 2567 const struct ceph_pg *pgid, 2568 struct ceph_osds *temp) 2569 { 2570 struct ceph_pg_mapping *pg; 2571 int i; 2572 2573 ceph_osds_init(temp); 2574 2575 /* pg_temp? */ 2576 pg = lookup_pg_mapping(&osdmap->pg_temp, pgid); 2577 if (pg) { 2578 for (i = 0; i < pg->pg_temp.len; i++) { 2579 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) { 2580 if (ceph_can_shift_osds(pi)) 2581 continue; 2582 2583 temp->osds[temp->size++] = CRUSH_ITEM_NONE; 2584 } else { 2585 temp->osds[temp->size++] = pg->pg_temp.osds[i]; 2586 } 2587 } 2588 2589 /* apply pg_temp's primary */ 2590 for (i = 0; i < temp->size; i++) { 2591 if (temp->osds[i] != CRUSH_ITEM_NONE) { 2592 temp->primary = temp->osds[i]; 2593 break; 2594 } 2595 } 2596 } 2597 2598 /* primary_temp? */ 2599 pg = lookup_pg_mapping(&osdmap->primary_temp, pgid); 2600 if (pg) 2601 temp->primary = pg->primary_temp.osd; 2602 } 2603 2604 /* 2605 * Map a PG to its acting set as well as its up set. 2606 * 2607 * Acting set is used for data mapping purposes, while up set can be 2608 * recorded for detecting interval changes and deciding whether to 2609 * resend a request. 2610 */ 2611 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 2612 struct ceph_pg_pool_info *pi, 2613 const struct ceph_pg *raw_pgid, 2614 struct ceph_osds *up, 2615 struct ceph_osds *acting) 2616 { 2617 struct ceph_pg pgid; 2618 u32 pps; 2619 2620 WARN_ON(pi->id != raw_pgid->pool); 2621 raw_pg_to_pg(pi, raw_pgid, &pgid); 2622 2623 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps); 2624 apply_upmap(osdmap, &pgid, up); 2625 raw_to_up_osds(osdmap, pi, up); 2626 apply_primary_affinity(osdmap, pi, pps, up); 2627 get_temp_osds(osdmap, pi, &pgid, acting); 2628 if (!acting->size) { 2629 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0])); 2630 acting->size = up->size; 2631 if (acting->primary == -1) 2632 acting->primary = up->primary; 2633 } 2634 WARN_ON(!osds_valid(up) || !osds_valid(acting)); 2635 } 2636 2637 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap, 2638 struct ceph_pg_pool_info *pi, 2639 const struct ceph_pg *raw_pgid, 2640 struct ceph_spg *spgid) 2641 { 2642 struct ceph_pg pgid; 2643 struct ceph_osds up, acting; 2644 int i; 2645 2646 WARN_ON(pi->id != raw_pgid->pool); 2647 raw_pg_to_pg(pi, raw_pgid, &pgid); 2648 2649 if (ceph_can_shift_osds(pi)) { 2650 spgid->pgid = pgid; /* struct */ 2651 spgid->shard = CEPH_SPG_NOSHARD; 2652 return true; 2653 } 2654 2655 ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting); 2656 for (i = 0; i < acting.size; i++) { 2657 if (acting.osds[i] == acting.primary) { 2658 spgid->pgid = pgid; /* struct */ 2659 spgid->shard = i; 2660 return true; 2661 } 2662 } 2663 2664 return false; 2665 } 2666 2667 /* 2668 * Return acting primary for given PG, or -1 if none. 2669 */ 2670 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 2671 const struct ceph_pg *raw_pgid) 2672 { 2673 struct ceph_pg_pool_info *pi; 2674 struct ceph_osds up, acting; 2675 2676 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool); 2677 if (!pi) 2678 return -1; 2679 2680 ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting); 2681 return acting.primary; 2682 } 2683 EXPORT_SYMBOL(ceph_pg_to_acting_primary); 2684