FILE
osdmap.c
File preview with syntax highlighting.
linux-6.19.14/net/ceph/osdmap.c
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 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 14static __printf(2, 3) 15void osdmap_info(const struct ceph_osdmap *map, const char *fmt, ...) 16{ 17 struct va_format vaf; 18 va_list args; 19 20 va_start(args, fmt); 21 vaf.fmt = fmt; 22 vaf.va = &args; 23 24 printk(KERN_INFO "%s (%pU e%u): %pV", KBUILD_MODNAME, &map->fsid, 25 map->epoch, &vaf); 26 27 va_end(args); 28} 29 30char *ceph_osdmap_state_str(char *str, int len, u32 state) 31{ 32 if (!len) 33 return str; 34 35 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP)) 36 snprintf(str, len, "exists, up"); 37 else if (state & CEPH_OSD_EXISTS) 38 snprintf(str, len, "exists"); 39 else if (state & CEPH_OSD_UP) 40 snprintf(str, len, "up"); 41 else 42 snprintf(str, len, "doesn't exist"); 43 44 return str; 45} 46 47/* maps */ 48 49static int calc_bits_of(unsigned int t) 50{ 51 int b = 0; 52 while (t) { 53 t = t >> 1; 54 b++; 55 } 56 return b; 57} 58 59/* 60 * the foo_mask is the smallest value 2^n-1 that is >= foo. 61 */ 62static void calc_pg_masks(struct ceph_pg_pool_info *pi) 63{ 64 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1; 65 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1; 66} 67 68/* 69 * decode crush map 70 */ 71static int crush_decode_uniform_bucket(void **p, void *end, 72 struct crush_bucket_uniform *b) 73{ 74 dout("crush_decode_uniform_bucket %p to %p\n", *p, end); 75 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad); 76 b->item_weight = ceph_decode_32(p); 77 return 0; 78bad: 79 return -EINVAL; 80} 81 82static int crush_decode_list_bucket(void **p, void *end, 83 struct crush_bucket_list *b) 84{ 85 int j; 86 dout("crush_decode_list_bucket %p to %p\n", *p, end); 87 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 88 if (b->item_weights == NULL) 89 return -ENOMEM; 90 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 91 if (b->sum_weights == NULL) 92 return -ENOMEM; 93 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); 94 for (j = 0; j < b->h.size; j++) { 95 b->item_weights[j] = ceph_decode_32(p); 96 b->sum_weights[j] = ceph_decode_32(p); 97 } 98 return 0; 99bad: 100 return -EINVAL; 101} 102 103static int crush_decode_tree_bucket(void **p, void *end, 104 struct crush_bucket_tree *b) 105{ 106 int j; 107 dout("crush_decode_tree_bucket %p to %p\n", *p, end); 108 ceph_decode_8_safe(p, end, b->num_nodes, bad); 109 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS); 110 if (b->node_weights == NULL) 111 return -ENOMEM; 112 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad); 113 for (j = 0; j < b->num_nodes; j++) 114 b->node_weights[j] = ceph_decode_32(p); 115 return 0; 116bad: 117 return -EINVAL; 118} 119 120static int crush_decode_straw_bucket(void **p, void *end, 121 struct crush_bucket_straw *b) 122{ 123 int j; 124 dout("crush_decode_straw_bucket %p to %p\n", *p, end); 125 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 126 if (b->item_weights == NULL) 127 return -ENOMEM; 128 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 129 if (b->straws == NULL) 130 return -ENOMEM; 131 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); 132 for (j = 0; j < b->h.size; j++) { 133 b->item_weights[j] = ceph_decode_32(p); 134 b->straws[j] = ceph_decode_32(p); 135 } 136 return 0; 137bad: 138 return -EINVAL; 139} 140 141static int crush_decode_straw2_bucket(void **p, void *end, 142 struct crush_bucket_straw2 *b) 143{ 144 int j; 145 dout("crush_decode_straw2_bucket %p to %p\n", *p, end); 146 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 147 if (b->item_weights == NULL) 148 return -ENOMEM; 149 ceph_decode_need(p, end, b->h.size * sizeof(u32), bad); 150 for (j = 0; j < b->h.size; j++) 151 b->item_weights[j] = ceph_decode_32(p); 152 return 0; 153bad: 154 return -EINVAL; 155} 156 157struct crush_name_node { 158 struct rb_node cn_node; 159 int cn_id; 160 char cn_name[]; 161}; 162 163static struct crush_name_node *alloc_crush_name(size_t name_len) 164{ 165 struct crush_name_node *cn; 166 167 cn = kmalloc(sizeof(*cn) + name_len + 1, GFP_NOIO); 168 if (!cn) 169 return NULL; 170 171 RB_CLEAR_NODE(&cn->cn_node); 172 return cn; 173} 174 175static void free_crush_name(struct crush_name_node *cn) 176{ 177 WARN_ON(!RB_EMPTY_NODE(&cn->cn_node)); 178 179 kfree(cn); 180} 181 182DEFINE_RB_FUNCS(crush_name, struct crush_name_node, cn_id, cn_node) 183 184static int decode_crush_names(void **p, void *end, struct rb_root *root) 185{ 186 u32 n; 187 188 ceph_decode_32_safe(p, end, n, e_inval); 189 while (n--) { 190 struct crush_name_node *cn; 191 int id; 192 u32 name_len; 193 194 ceph_decode_32_safe(p, end, id, e_inval); 195 ceph_decode_32_safe(p, end, name_len, e_inval); 196 ceph_decode_need(p, end, name_len, e_inval); 197 198 cn = alloc_crush_name(name_len); 199 if (!cn) 200 return -ENOMEM; 201 202 cn->cn_id = id; 203 memcpy(cn->cn_name, *p, name_len); 204 cn->cn_name[name_len] = '\0'; 205 *p += name_len; 206 207 if (!__insert_crush_name(root, cn)) { 208 free_crush_name(cn); 209 return -EEXIST; 210 } 211 } 212 213 return 0; 214 215e_inval: 216 return -EINVAL; 217} 218 219void clear_crush_names(struct rb_root *root) 220{ 221 while (!RB_EMPTY_ROOT(root)) { 222 struct crush_name_node *cn = 223 rb_entry(rb_first(root), struct crush_name_node, cn_node); 224 225 erase_crush_name(root, cn); 226 free_crush_name(cn); 227 } 228} 229 230static struct crush_choose_arg_map *alloc_choose_arg_map(void) 231{ 232 struct crush_choose_arg_map *arg_map; 233 234 arg_map = kzalloc(sizeof(*arg_map), GFP_NOIO); 235 if (!arg_map) 236 return NULL; 237 238 RB_CLEAR_NODE(&arg_map->node); 239 return arg_map; 240} 241 242static void free_choose_arg_map(struct crush_choose_arg_map *arg_map) 243{ 244 int i, j; 245 246 if (!arg_map) 247 return; 248 249 WARN_ON(!RB_EMPTY_NODE(&arg_map->node)); 250 251 if (arg_map->args) { 252 for (i = 0; i < arg_map->size; i++) { 253 struct crush_choose_arg *arg = &arg_map->args[i]; 254 if (arg->weight_set) { 255 for (j = 0; j < arg->weight_set_size; j++) 256 kfree(arg->weight_set[j].weights); 257 kfree(arg->weight_set); 258 } 259 kfree(arg->ids); 260 } 261 kfree(arg_map->args); 262 } 263 kfree(arg_map); 264} 265 266DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index, 267 node); 268 269void clear_choose_args(struct crush_map *c) 270{ 271 while (!RB_EMPTY_ROOT(&c->choose_args)) { 272 struct crush_choose_arg_map *arg_map = 273 rb_entry(rb_first(&c->choose_args), 274 struct crush_choose_arg_map, node); 275 276 erase_choose_arg_map(&c->choose_args, arg_map); 277 free_choose_arg_map(arg_map); 278 } 279} 280 281static u32 *decode_array_32_alloc(void **p, void *end, u32 *plen) 282{ 283 u32 *a = NULL; 284 u32 len; 285 int ret; 286 287 ceph_decode_32_safe(p, end, len, e_inval); 288 if (len) { 289 u32 i; 290 291 a = kmalloc_array(len, sizeof(u32), GFP_NOIO); 292 if (!a) { 293 ret = -ENOMEM; 294 goto fail; 295 } 296 297 ceph_decode_need(p, end, len * sizeof(u32), e_inval); 298 for (i = 0; i < len; i++) 299 a[i] = ceph_decode_32(p); 300 } 301 302 *plen = len; 303 return a; 304 305e_inval: 306 ret = -EINVAL; 307fail: 308 kfree(a); 309 return ERR_PTR(ret); 310} 311 312/* 313 * Assumes @arg is zero-initialized. 314 */ 315static int decode_choose_arg(void **p, void *end, struct crush_choose_arg *arg) 316{ 317 int ret; 318 319 ceph_decode_32_safe(p, end, arg->weight_set_size, e_inval); 320 if (arg->weight_set_size) { 321 u32 i; 322 323 arg->weight_set = kmalloc_array(arg->weight_set_size, 324 sizeof(*arg->weight_set), 325 GFP_NOIO); 326 if (!arg->weight_set) 327 return -ENOMEM; 328 329 for (i = 0; i < arg->weight_set_size; i++) { 330 struct crush_weight_set *w = &arg->weight_set[i]; 331 332 w->weights = decode_array_32_alloc(p, end, &w->size); 333 if (IS_ERR(w->weights)) { 334 ret = PTR_ERR(w->weights); 335 w->weights = NULL; 336 return ret; 337 } 338 } 339 } 340 341 arg->ids = decode_array_32_alloc(p, end, &arg->ids_size); 342 if (IS_ERR(arg->ids)) { 343 ret = PTR_ERR(arg->ids); 344 arg->ids = NULL; 345 return ret; 346 } 347 348 return 0; 349 350e_inval: 351 return -EINVAL; 352} 353 354static int decode_choose_args(void **p, void *end, struct crush_map *c) 355{ 356 struct crush_choose_arg_map *arg_map = NULL; 357 u32 num_choose_arg_maps, num_buckets; 358 int ret; 359 360 ceph_decode_32_safe(p, end, num_choose_arg_maps, e_inval); 361 while (num_choose_arg_maps--) { 362 arg_map = alloc_choose_arg_map(); 363 if (!arg_map) { 364 ret = -ENOMEM; 365 goto fail; 366 } 367 368 ceph_decode_64_safe(p, end, arg_map->choose_args_index, 369 e_inval); 370 arg_map->size = c->max_buckets; 371 arg_map->args = kcalloc(arg_map->size, sizeof(*arg_map->args), 372 GFP_NOIO); 373 if (!arg_map->args) { 374 ret = -ENOMEM; 375 goto fail; 376 } 377 378 ceph_decode_32_safe(p, end, num_buckets, e_inval); 379 while (num_buckets--) { 380 struct crush_choose_arg *arg; 381 u32 bucket_index; 382 383 ceph_decode_32_safe(p, end, bucket_index, e_inval); 384 if (bucket_index >= arg_map->size) 385 goto e_inval; 386 387 arg = &arg_map->args[bucket_index]; 388 ret = decode_choose_arg(p, end, arg); 389 if (ret) 390 goto fail; 391 392 if (arg->ids_size && 393 arg->ids_size != c->buckets[bucket_index]->size) 394 goto e_inval; 395 } 396 397 insert_choose_arg_map(&c->choose_args, arg_map); 398 } 399 400 return 0; 401 402e_inval: 403 ret = -EINVAL; 404fail: 405 free_choose_arg_map(arg_map); 406 return ret; 407} 408 409static void crush_finalize(struct crush_map *c) 410{ 411 __s32 b; 412 413 /* Space for the array of pointers to per-bucket workspace */ 414 c->working_size = sizeof(struct crush_work) + 415 c->max_buckets * sizeof(struct crush_work_bucket *); 416 417 for (b = 0; b < c->max_buckets; b++) { 418 if (!c->buckets[b]) 419 continue; 420 421 switch (c->buckets[b]->alg) { 422 default: 423 /* 424 * The base case, permutation variables and 425 * the pointer to the permutation array. 426 */ 427 c->working_size += sizeof(struct crush_work_bucket); 428 break; 429 } 430 /* Every bucket has a permutation array. */ 431 c->working_size += c->buckets[b]->size * sizeof(__u32); 432 } 433} 434 435static struct crush_map *crush_decode(void *pbyval, void *end) 436{ 437 struct crush_map *c; 438 int err; 439 int i, j; 440 void **p = &pbyval; 441 void *start = pbyval; 442 u32 magic; 443 444 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p)); 445 446 c = kzalloc(sizeof(*c), GFP_NOFS); 447 if (c == NULL) 448 return ERR_PTR(-ENOMEM); 449 450 c->type_names = RB_ROOT; 451 c->names = RB_ROOT; 452 c->choose_args = RB_ROOT; 453 454 /* set tunables to default values */ 455 c->choose_local_tries = 2; 456 c->choose_local_fallback_tries = 5; 457 c->choose_total_tries = 19; 458 c->chooseleaf_descend_once = 0; 459 460 ceph_decode_need(p, end, 4*sizeof(u32), bad); 461 magic = ceph_decode_32(p); 462 if (magic != CRUSH_MAGIC) { 463 pr_err("crush_decode magic %x != current %x\n", 464 (unsigned int)magic, (unsigned int)CRUSH_MAGIC); 465 goto bad; 466 } 467 c->max_buckets = ceph_decode_32(p); 468 c->max_rules = ceph_decode_32(p); 469 c->max_devices = ceph_decode_32(p); 470 471 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS); 472 if (c->buckets == NULL) 473 goto badmem; 474 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS); 475 if (c->rules == NULL) 476 goto badmem; 477 478 /* buckets */ 479 for (i = 0; i < c->max_buckets; i++) { 480 int size = 0; 481 u32 alg; 482 struct crush_bucket *b; 483 484 ceph_decode_32_safe(p, end, alg, bad); 485 if (alg == 0) { 486 c->buckets[i] = NULL; 487 continue; 488 } 489 dout("crush_decode bucket %d off %x %p to %p\n", 490 i, (int)(*p-start), *p, end); 491 492 switch (alg) { 493 case CRUSH_BUCKET_UNIFORM: 494 size = sizeof(struct crush_bucket_uniform); 495 break; 496 case CRUSH_BUCKET_LIST: 497 size = sizeof(struct crush_bucket_list); 498 break; 499 case CRUSH_BUCKET_TREE: 500 size = sizeof(struct crush_bucket_tree); 501 break; 502 case CRUSH_BUCKET_STRAW: 503 size = sizeof(struct crush_bucket_straw); 504 break; 505 case CRUSH_BUCKET_STRAW2: 506 size = sizeof(struct crush_bucket_straw2); 507 break; 508 default: 509 goto bad; 510 } 511 BUG_ON(size == 0); 512 b = c->buckets[i] = kzalloc(size, GFP_NOFS); 513 if (b == NULL) 514 goto badmem; 515 516 ceph_decode_need(p, end, 4*sizeof(u32), bad); 517 b->id = ceph_decode_32(p); 518 b->type = ceph_decode_16(p); 519 b->alg = ceph_decode_8(p); 520 b->hash = ceph_decode_8(p); 521 b->weight = ceph_decode_32(p); 522 b->size = ceph_decode_32(p); 523 524 dout("crush_decode bucket size %d off %x %p to %p\n", 525 b->size, (int)(*p-start), *p, end); 526 527 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS); 528 if (b->items == NULL) 529 goto badmem; 530 531 ceph_decode_need(p, end, b->size*sizeof(u32), bad); 532 for (j = 0; j < b->size; j++) 533 b->items[j] = ceph_decode_32(p); 534 535 switch (b->alg) { 536 case CRUSH_BUCKET_UNIFORM: 537 err = crush_decode_uniform_bucket(p, end, 538 (struct crush_bucket_uniform *)b); 539 if (err < 0) 540 goto fail; 541 break; 542 case CRUSH_BUCKET_LIST: 543 err = crush_decode_list_bucket(p, end, 544 (struct crush_bucket_list *)b); 545 if (err < 0) 546 goto fail; 547 break; 548 case CRUSH_BUCKET_TREE: 549 err = crush_decode_tree_bucket(p, end, 550 (struct crush_bucket_tree *)b); 551 if (err < 0) 552 goto fail; 553 break; 554 case CRUSH_BUCKET_STRAW: 555 err = crush_decode_straw_bucket(p, end, 556 (struct crush_bucket_straw *)b); 557 if (err < 0) 558 goto fail; 559 break; 560 case CRUSH_BUCKET_STRAW2: 561 err = crush_decode_straw2_bucket(p, end, 562 (struct crush_bucket_straw2 *)b); 563 if (err < 0) 564 goto fail; 565 break; 566 } 567 } 568 569 /* rules */ 570 dout("rule vec is %p\n", c->rules); 571 for (i = 0; i < c->max_rules; i++) { 572 u32 yes; 573 struct crush_rule *r; 574 575 ceph_decode_32_safe(p, end, yes, bad); 576 if (!yes) { 577 dout("crush_decode NO rule %d off %x %p to %p\n", 578 i, (int)(*p-start), *p, end); 579 c->rules[i] = NULL; 580 continue; 581 } 582 583 dout("crush_decode rule %d off %x %p to %p\n", 584 i, (int)(*p-start), *p, end); 585 586 /* len */ 587 ceph_decode_32_safe(p, end, yes, bad); 588#if BITS_PER_LONG == 32 589 if (yes > (ULONG_MAX - sizeof(*r)) 590 / sizeof(struct crush_rule_step)) 591 goto bad; 592#endif 593 r = kmalloc(struct_size(r, steps, yes), GFP_NOFS); 594 if (r == NULL) 595 goto badmem; 596 dout(" rule %d is at %p\n", i, r); 597 c->rules[i] = r; 598 r->len = yes; 599 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */ 600 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad); 601 for (j = 0; j < r->len; j++) { 602 r->steps[j].op = ceph_decode_32(p); 603 r->steps[j].arg1 = ceph_decode_32(p); 604 r->steps[j].arg2 = ceph_decode_32(p); 605 } 606 } 607 608 err = decode_crush_names(p, end, &c->type_names); 609 if (err) 610 goto fail; 611 612 err = decode_crush_names(p, end, &c->names); 613 if (err) 614 goto fail; 615 616 ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */ 617 618 /* tunables */ 619 ceph_decode_need(p, end, 3*sizeof(u32), done); 620 c->choose_local_tries = ceph_decode_32(p); 621 c->choose_local_fallback_tries = ceph_decode_32(p); 622 c->choose_total_tries = ceph_decode_32(p); 623 dout("crush decode tunable choose_local_tries = %d\n", 624 c->choose_local_tries); 625 dout("crush decode tunable choose_local_fallback_tries = %d\n", 626 c->choose_local_fallback_tries); 627 dout("crush decode tunable choose_total_tries = %d\n", 628 c->choose_total_tries); 629 630 ceph_decode_need(p, end, sizeof(u32), done); 631 c->chooseleaf_descend_once = ceph_decode_32(p); 632 dout("crush decode tunable chooseleaf_descend_once = %d\n", 633 c->chooseleaf_descend_once); 634 635 ceph_decode_need(p, end, sizeof(u8), done); 636 c->chooseleaf_vary_r = ceph_decode_8(p); 637 dout("crush decode tunable chooseleaf_vary_r = %d\n", 638 c->chooseleaf_vary_r); 639 640 /* skip straw_calc_version, allowed_bucket_algs */ 641 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done); 642 *p += sizeof(u8) + sizeof(u32); 643 644 ceph_decode_need(p, end, sizeof(u8), done); 645 c->chooseleaf_stable = ceph_decode_8(p); 646 dout("crush decode tunable chooseleaf_stable = %d\n", 647 c->chooseleaf_stable); 648 649 if (*p != end) { 650 /* class_map */ 651 ceph_decode_skip_map(p, end, 32, 32, bad); 652 /* class_name */ 653 ceph_decode_skip_map(p, end, 32, string, bad); 654 /* class_bucket */ 655 ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad); 656 } 657 658 if (*p != end) { 659 err = decode_choose_args(p, end, c); 660 if (err) 661 goto fail; 662 } 663 664done: 665 crush_finalize(c); 666 dout("crush_decode success\n"); 667 return c; 668 669badmem: 670 err = -ENOMEM; 671fail: 672 dout("crush_decode fail %d\n", err); 673 crush_destroy(c); 674 return ERR_PTR(err); 675 676bad: 677 err = -EINVAL; 678 goto fail; 679} 680 681int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs) 682{ 683 if (lhs->pool < rhs->pool) 684 return -1; 685 if (lhs->pool > rhs->pool) 686 return 1; 687 if (lhs->seed < rhs->seed) 688 return -1; 689 if (lhs->seed > rhs->seed) 690 return 1; 691 692 return 0; 693} 694 695int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs) 696{ 697 int ret; 698 699 ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid); 700 if (ret) 701 return ret; 702 703 if (lhs->shard < rhs->shard) 704 return -1; 705 if (lhs->shard > rhs->shard) 706 return 1; 707 708 return 0; 709} 710 711static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len) 712{ 713 struct ceph_pg_mapping *pg; 714 715 pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO); 716 if (!pg) 717 return NULL; 718 719 RB_CLEAR_NODE(&pg->node); 720 return pg; 721} 722 723static void free_pg_mapping(struct ceph_pg_mapping *pg) 724{ 725 WARN_ON(!RB_EMPTY_NODE(&pg->node)); 726 727 kfree(pg); 728} 729 730/* 731 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid 732 * to a set of osds) and primary_temp (explicit primary setting) 733 */ 734DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare, 735 RB_BYPTR, const struct ceph_pg *, node) 736 737/* 738 * rbtree of pg pool info 739 */ 740DEFINE_RB_FUNCS(pg_pool, struct ceph_pg_pool_info, id, node) 741 742struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id) 743{ 744 return lookup_pg_pool(&map->pg_pools, id); 745} 746 747const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id) 748{ 749 struct ceph_pg_pool_info *pi; 750 751 if (id == CEPH_NOPOOL) 752 return NULL; 753 754 if (WARN_ON_ONCE(id > (u64) INT_MAX)) 755 return NULL; 756 757 pi = lookup_pg_pool(&map->pg_pools, id); 758 return pi ? pi->name : NULL; 759} 760EXPORT_SYMBOL(ceph_pg_pool_name_by_id); 761 762int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name) 763{ 764 struct rb_node *rbp; 765 766 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) { 767 struct ceph_pg_pool_info *pi = 768 rb_entry(rbp, struct ceph_pg_pool_info, node); 769 if (pi->name && strcmp(pi->name, name) == 0) 770 return pi->id; 771 } 772 return -ENOENT; 773} 774EXPORT_SYMBOL(ceph_pg_poolid_by_name); 775 776u64 ceph_pg_pool_flags(struct ceph_osdmap *map, u64 id) 777{ 778 struct ceph_pg_pool_info *pi; 779 780 pi = lookup_pg_pool(&map->pg_pools, id); 781 return pi ? pi->flags : 0; 782} 783EXPORT_SYMBOL(ceph_pg_pool_flags); 784 785static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi) 786{ 787 erase_pg_pool(root, pi); 788 kfree(pi->name); 789 kfree(pi); 790} 791 792static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi) 793{ 794 u8 ev, cv; 795 unsigned len, num; 796 void *pool_end; 797 798 ceph_decode_need(p, end, 2 + 4, bad); 799 ev = ceph_decode_8(p); /* encoding version */ 800 cv = ceph_decode_8(p); /* compat version */ 801 if (ev < 5) { 802 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv); 803 return -EINVAL; 804 } 805 if (cv > 9) { 806 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv); 807 return -EINVAL; 808 } 809 len = ceph_decode_32(p); 810 ceph_decode_need(p, end, len, bad); 811 pool_end = *p + len; 812 813 ceph_decode_need(p, end, 4 + 4 + 4, bad); 814 pi->type = ceph_decode_8(p); 815 pi->size = ceph_decode_8(p); 816 pi->crush_ruleset = ceph_decode_8(p); 817 pi->object_hash = ceph_decode_8(p); 818 pi->pg_num = ceph_decode_32(p); 819 pi->pgp_num = ceph_decode_32(p); 820 821 /* lpg*, last_change, snap_seq, snap_epoch */ 822 ceph_decode_skip_n(p, end, 8 + 4 + 8 + 4, bad); 823 824 /* skip snaps */ 825 ceph_decode_32_safe(p, end, num, bad); 826 while (num--) { 827 /* snapid key, pool snap (with versions) */ 828 ceph_decode_skip_n(p, end, 8 + 2, bad); 829 ceph_decode_skip_string(p, end, bad); 830 } 831 832 /* removed_snaps */ 833 ceph_decode_skip_map(p, end, 64, 64, bad); 834 835 ceph_decode_need(p, end, 8 + 8 + 4, bad); 836 *p += 8; /* skip auid */ 837 pi->flags = ceph_decode_64(p); 838 *p += 4; /* skip crash_replay_interval */ 839 840 if (ev >= 7) 841 ceph_decode_8_safe(p, end, pi->min_size, bad); 842 else 843 pi->min_size = pi->size - pi->size / 2; 844 845 if (ev >= 8) 846 /* quota_max_* */ 847 ceph_decode_skip_n(p, end, 8 + 8, bad); 848 849 if (ev >= 9) { 850 /* tiers */ 851 ceph_decode_skip_set(p, end, 64, bad); 852 853 ceph_decode_need(p, end, 8 + 1 + 8 + 8, bad); 854 *p += 8; /* skip tier_of */ 855 *p += 1; /* skip cache_mode */ 856 pi->read_tier = ceph_decode_64(p); 857 pi->write_tier = ceph_decode_64(p); 858 } else { 859 pi->read_tier = -1; 860 pi->write_tier = -1; 861 } 862 863 if (ev >= 10) 864 /* properties */ 865 ceph_decode_skip_map(p, end, string, string, bad); 866 867 if (ev >= 11) { 868 /* hit_set_params (with versions) */ 869 ceph_decode_skip_n(p, end, 2, bad); 870 ceph_decode_skip_string(p, end, bad); 871 872 /* hit_set_period, hit_set_count */ 873 ceph_decode_skip_n(p, end, 4 + 4, bad); 874 } 875 876 if (ev >= 12) 877 /* stripe_width */ 878 ceph_decode_skip_32(p, end, bad); 879 880 if (ev >= 13) 881 /* target_max_*, cache_target_*, cache_min_* */ 882 ceph_decode_skip_n(p, end, 16 + 8 + 8, bad); 883 884 if (ev >= 14) 885 /* erasure_code_profile */ 886 ceph_decode_skip_string(p, end, bad); 887 888 /* 889 * last_force_op_resend_preluminous, will be overridden if the 890 * map was encoded with RESEND_ON_SPLIT 891 */ 892 if (ev >= 15) 893 ceph_decode_32_safe(p, end, pi->last_force_request_resend, bad); 894 else 895 pi->last_force_request_resend = 0; 896 897 if (ev >= 16) 898 /* min_read_recency_for_promote */ 899 ceph_decode_skip_32(p, end, bad); 900 901 if (ev >= 17) 902 /* expected_num_objects */ 903 ceph_decode_skip_64(p, end, bad); 904 905 if (ev >= 19) 906 /* cache_target_dirty_high_ratio_micro */ 907 ceph_decode_skip_32(p, end, bad); 908 909 if (ev >= 20) 910 /* min_write_recency_for_promote */ 911 ceph_decode_skip_32(p, end, bad); 912 913 if (ev >= 21) 914 /* use_gmt_hitset */ 915 ceph_decode_skip_8(p, end, bad); 916 917 if (ev >= 22) 918 /* fast_read */ 919 ceph_decode_skip_8(p, end, bad); 920 921 if (ev >= 23) 922 /* hit_set_grade_decay_rate, hit_set_search_last_n */ 923 ceph_decode_skip_n(p, end, 4 + 4, bad); 924 925 if (ev >= 24) { 926 /* opts (with versions) */ 927 ceph_decode_skip_n(p, end, 2, bad); 928 ceph_decode_skip_string(p, end, bad); 929 } 930 931 if (ev >= 25) 932 ceph_decode_32_safe(p, end, pi->last_force_request_resend, bad); 933 934 /* ignore the rest */ 935 936 *p = pool_end; 937 calc_pg_masks(pi); 938 return 0; 939 940bad: 941 return -EINVAL; 942} 943 944static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map) 945{ 946 struct ceph_pg_pool_info *pi; 947 u32 num, len; 948 u64 pool; 949 950 ceph_decode_32_safe(p, end, num, bad); 951 dout(" %d pool names\n", num); 952 while (num--) { 953 ceph_decode_64_safe(p, end, pool, bad); 954 ceph_decode_32_safe(p, end, len, bad); 955 dout(" pool %llu len %d\n", pool, len); 956 ceph_decode_need(p, end, len, bad); 957 pi = lookup_pg_pool(&map->pg_pools, pool); 958 if (pi) { 959 char *name = kstrndup(*p, len, GFP_NOFS); 960 961 if (!name) 962 return -ENOMEM; 963 kfree(pi->name); 964 pi->name = name; 965 dout(" name is %s\n", pi->name); 966 } 967 *p += len; 968 } 969 return 0; 970 971bad: 972 return -EINVAL; 973} 974 975/* 976 * CRUSH workspaces 977 * 978 * workspace_manager framework borrowed from fs/btrfs/compression.c. 979 * Two simplifications: there is only one type of workspace and there 980 * is always at least one workspace. 981 */ 982static struct crush_work *alloc_workspace(const struct crush_map *c) 983{ 984 struct crush_work *work; 985 size_t work_size; 986 987 WARN_ON(!c->working_size); 988 work_size = crush_work_size(c, CEPH_PG_MAX_SIZE); 989 dout("%s work_size %zu bytes\n", __func__, work_size); 990 991 work = kvmalloc(work_size, GFP_NOIO); 992 if (!work) 993 return NULL; 994 995 INIT_LIST_HEAD(&work->item); 996 crush_init_workspace(c, work); 997 return work; 998} 999 1000static void free_workspace(struct crush_work *work) 1001{ 1002 WARN_ON(!list_empty(&work->item)); 1003 kvfree(work); 1004} 1005 1006static void init_workspace_manager(struct workspace_manager *wsm) 1007{ 1008 INIT_LIST_HEAD(&wsm->idle_ws); 1009 spin_lock_init(&wsm->ws_lock); 1010 atomic_set(&wsm->total_ws, 0); 1011 wsm->free_ws = 0; 1012 init_waitqueue_head(&wsm->ws_wait); 1013} 1014 1015static void add_initial_workspace(struct workspace_manager *wsm, 1016 struct crush_work *work) 1017{ 1018 WARN_ON(!list_empty(&wsm->idle_ws)); 1019 1020 list_add(&work->item, &wsm->idle_ws); 1021 atomic_set(&wsm->total_ws, 1); 1022 wsm->free_ws = 1; 1023} 1024 1025static void cleanup_workspace_manager(struct workspace_manager *wsm) 1026{ 1027 struct crush_work *work; 1028 1029 while (!list_empty(&wsm->idle_ws)) { 1030 work = list_first_entry(&wsm->idle_ws, struct crush_work, 1031 item); 1032 list_del_init(&work->item); 1033 free_workspace(work); 1034 } 1035 atomic_set(&wsm->total_ws, 0); 1036 wsm->free_ws = 0; 1037} 1038 1039/* 1040 * Finds an available workspace or allocates a new one. If it's not 1041 * possible to allocate a new one, waits until there is one. 1042 */ 1043static struct crush_work *get_workspace(struct workspace_manager *wsm, 1044 const struct crush_map *c) 1045{ 1046 struct crush_work *work; 1047 int cpus = num_online_cpus(); 1048 1049again: 1050 spin_lock(&wsm->ws_lock); 1051 if (!list_empty(&wsm->idle_ws)) { 1052 work = list_first_entry(&wsm->idle_ws, struct crush_work, 1053 item); 1054 list_del_init(&work->item); 1055 wsm->free_ws--; 1056 spin_unlock(&wsm->ws_lock); 1057 return work; 1058 1059 } 1060 if (atomic_read(&wsm->total_ws) > cpus) { 1061 DEFINE_WAIT(wait); 1062 1063 spin_unlock(&wsm->ws_lock); 1064 prepare_to_wait(&wsm->ws_wait, &wait, TASK_UNINTERRUPTIBLE); 1065 if (atomic_read(&wsm->total_ws) > cpus && !wsm->free_ws) 1066 schedule(); 1067 finish_wait(&wsm->ws_wait, &wait); 1068 goto again; 1069 } 1070 atomic_inc(&wsm->total_ws); 1071 spin_unlock(&wsm->ws_lock); 1072 1073 work = alloc_workspace(c); 1074 if (!work) { 1075 atomic_dec(&wsm->total_ws); 1076 wake_up(&wsm->ws_wait); 1077 1078 /* 1079 * Do not return the error but go back to waiting. We 1080 * have the initial workspace and the CRUSH computation 1081 * time is bounded so we will get it eventually. 1082 */ 1083 WARN_ON(atomic_read(&wsm->total_ws) < 1); 1084 goto again; 1085 } 1086 return work; 1087} 1088 1089/* 1090 * Puts a workspace back on the list or frees it if we have enough 1091 * idle ones sitting around. 1092 */ 1093static void put_workspace(struct workspace_manager *wsm, 1094 struct crush_work *work) 1095{ 1096 spin_lock(&wsm->ws_lock); 1097 if (wsm->free_ws <= num_online_cpus()) { 1098 list_add(&work->item, &wsm->idle_ws); 1099 wsm->free_ws++; 1100 spin_unlock(&wsm->ws_lock); 1101 goto wake; 1102 } 1103 spin_unlock(&wsm->ws_lock); 1104 1105 free_workspace(work); 1106 atomic_dec(&wsm->total_ws); 1107wake: 1108 if (wq_has_sleeper(&wsm->ws_wait)) 1109 wake_up(&wsm->ws_wait); 1110} 1111 1112/* 1113 * osd map 1114 */ 1115struct ceph_osdmap *ceph_osdmap_alloc(void) 1116{ 1117 struct ceph_osdmap *map; 1118 1119 map = kzalloc(sizeof(*map), GFP_NOIO); 1120 if (!map) 1121 return NULL; 1122 1123 map->pg_pools = RB_ROOT; 1124 map->pool_max = -1; 1125 map->pg_temp = RB_ROOT; 1126 map->primary_temp = RB_ROOT; 1127 map->pg_upmap = RB_ROOT; 1128 map->pg_upmap_items = RB_ROOT; 1129 1130 init_workspace_manager(&map->crush_wsm); 1131 1132 return map; 1133} 1134 1135void ceph_osdmap_destroy(struct ceph_osdmap *map) 1136{ 1137 dout("osdmap_destroy %p\n", map); 1138 1139 if (map->crush) 1140 crush_destroy(map->crush); 1141 cleanup_workspace_manager(&map->crush_wsm); 1142 1143 while (!RB_EMPTY_ROOT(&map->pg_temp)) { 1144 struct ceph_pg_mapping *pg = 1145 rb_entry(rb_first(&map->pg_temp), 1146 struct ceph_pg_mapping, node); 1147 erase_pg_mapping(&map->pg_temp, pg); 1148 free_pg_mapping(pg); 1149 } 1150 while (!RB_EMPTY_ROOT(&map->primary_temp)) { 1151 struct ceph_pg_mapping *pg = 1152 rb_entry(rb_first(&map->primary_temp), 1153 struct ceph_pg_mapping, node); 1154 erase_pg_mapping(&map->primary_temp, pg); 1155 free_pg_mapping(pg); 1156 } 1157 while (!RB_EMPTY_ROOT(&map->pg_upmap)) { 1158 struct ceph_pg_mapping *pg = 1159 rb_entry(rb_first(&map->pg_upmap), 1160 struct ceph_pg_mapping, node); 1161 rb_erase(&pg->node, &map->pg_upmap); 1162 kfree(pg); 1163 } 1164 while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) { 1165 struct ceph_pg_mapping *pg = 1166 rb_entry(rb_first(&map->pg_upmap_items), 1167 struct ceph_pg_mapping, node); 1168 rb_erase(&pg->node, &map->pg_upmap_items); 1169 kfree(pg); 1170 } 1171 while (!RB_EMPTY_ROOT(&map->pg_pools)) { 1172 struct ceph_pg_pool_info *pi = 1173 rb_entry(rb_first(&map->pg_pools), 1174 struct ceph_pg_pool_info, node); 1175 __remove_pg_pool(&map->pg_pools, pi); 1176 } 1177 kvfree(map->osd_state); 1178 kvfree(map->osd_weight); 1179 kvfree(map->osd_addr); 1180 kvfree(map->osd_primary_affinity); 1181 kfree(map); 1182} 1183 1184/* 1185 * Adjust max_osd value, (re)allocate arrays. 1186 * 1187 * The new elements are properly initialized. 1188 */ 1189static int osdmap_set_max_osd(struct ceph_osdmap *map, u32 max) 1190{ 1191 u32 *state; 1192 u32 *weight; 1193 struct ceph_entity_addr *addr; 1194 u32 to_copy; 1195 int i; 1196 1197 dout("%s old %u new %u\n", __func__, map->max_osd, max); 1198 if (max == map->max_osd) 1199 return 0; 1200 1201 state = kvmalloc(array_size(max, sizeof(*state)), GFP_NOFS); 1202 weight = kvmalloc(array_size(max, sizeof(*weight)), GFP_NOFS); 1203 addr = kvmalloc(array_size(max, sizeof(*addr)), GFP_NOFS); 1204 if (!state || !weight || !addr) { 1205 kvfree(state); 1206 kvfree(weight); 1207 kvfree(addr); 1208 return -ENOMEM; 1209 } 1210 1211 to_copy = min(map->max_osd, max); 1212 if (map->osd_state) { 1213 memcpy(state, map->osd_state, to_copy * sizeof(*state)); 1214 memcpy(weight, map->osd_weight, to_copy * sizeof(*weight)); 1215 memcpy(addr, map->osd_addr, to_copy * sizeof(*addr)); 1216 kvfree(map->osd_state); 1217 kvfree(map->osd_weight); 1218 kvfree(map->osd_addr); 1219 } 1220 1221 map->osd_state = state; 1222 map->osd_weight = weight; 1223 map->osd_addr = addr; 1224 for (i = map->max_osd; i < max; i++) { 1225 map->osd_state[i] = 0; 1226 map->osd_weight[i] = CEPH_OSD_OUT; 1227 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr)); 1228 } 1229 1230 if (map->osd_primary_affinity) { 1231 u32 *affinity; 1232 1233 affinity = kvmalloc(array_size(max, sizeof(*affinity)), 1234 GFP_NOFS); 1235 if (!affinity) 1236 return -ENOMEM; 1237 1238 memcpy(affinity, map->osd_primary_affinity, 1239 to_copy * sizeof(*affinity)); 1240 kvfree(map->osd_primary_affinity); 1241 1242 map->osd_primary_affinity = affinity; 1243 for (i = map->max_osd; i < max; i++) 1244 map->osd_primary_affinity[i] = 1245 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1246 } 1247 1248 map->max_osd = max; 1249 1250 return 0; 1251} 1252 1253static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush) 1254{ 1255 struct crush_work *work; 1256 1257 if (IS_ERR(crush)) 1258 return PTR_ERR(crush); 1259 1260 work = alloc_workspace(crush); 1261 if (!work) { 1262 crush_destroy(crush); 1263 return -ENOMEM; 1264 } 1265 1266 if (map->crush) 1267 crush_destroy(map->crush); 1268 cleanup_workspace_manager(&map->crush_wsm); 1269 map->crush = crush; 1270 add_initial_workspace(&map->crush_wsm, work); 1271 return 0; 1272} 1273 1274#define OSDMAP_WRAPPER_COMPAT_VER 7 1275#define OSDMAP_CLIENT_DATA_COMPAT_VER 1 1276 1277/* 1278 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps, 1279 * to struct_v of the client_data section for new (v7 and above) 1280 * osdmaps. 1281 */ 1282static int get_osdmap_client_data_v(void **p, void *end, 1283 const char *prefix, u8 *v) 1284{ 1285 u8 struct_v; 1286 1287 ceph_decode_8_safe(p, end, struct_v, e_inval); 1288 if (struct_v >= 7) { 1289 u8 struct_compat; 1290 1291 ceph_decode_8_safe(p, end, struct_compat, e_inval); 1292 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) { 1293 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n", 1294 struct_v, struct_compat, 1295 OSDMAP_WRAPPER_COMPAT_VER, prefix); 1296 return -EINVAL; 1297 } 1298 *p += 4; /* ignore wrapper struct_len */ 1299 1300 ceph_decode_8_safe(p, end, struct_v, e_inval); 1301 ceph_decode_8_safe(p, end, struct_compat, e_inval); 1302 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) { 1303 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n", 1304 struct_v, struct_compat, 1305 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix); 1306 return -EINVAL; 1307 } 1308 *p += 4; /* ignore client data struct_len */ 1309 } else { 1310 u16 version; 1311 1312 *p -= 1; 1313 ceph_decode_16_safe(p, end, version, e_inval); 1314 if (version < 6) { 1315 pr_warn("got v %d < 6 of %s ceph_osdmap\n", 1316 version, prefix); 1317 return -EINVAL; 1318 } 1319 1320 /* old osdmap encoding */ 1321 struct_v = 0; 1322 } 1323 1324 *v = struct_v; 1325 return 0; 1326 1327e_inval: 1328 return -EINVAL; 1329} 1330 1331static int __decode_pools(void **p, void *end, struct ceph_osdmap *map, 1332 bool incremental) 1333{ 1334 u32 n; 1335 1336 ceph_decode_32_safe(p, end, n, e_inval); 1337 while (n--) { 1338 struct ceph_pg_pool_info *pi; 1339 u64 pool; 1340 int ret; 1341 1342 ceph_decode_64_safe(p, end, pool, e_inval); 1343 1344 pi = lookup_pg_pool(&map->pg_pools, pool); 1345 if (!incremental || !pi) { 1346 pi = kzalloc(sizeof(*pi), GFP_NOFS); 1347 if (!pi) 1348 return -ENOMEM; 1349 1350 RB_CLEAR_NODE(&pi->node); 1351 pi->id = pool; 1352 1353 if (!__insert_pg_pool(&map->pg_pools, pi)) { 1354 kfree(pi); 1355 return -EEXIST; 1356 } 1357 } 1358 1359 ret = decode_pool(p, end, pi); 1360 if (ret) 1361 return ret; 1362 } 1363 1364 return 0; 1365 1366e_inval: 1367 return -EINVAL; 1368} 1369 1370static int decode_pools(void **p, void *end, struct ceph_osdmap *map) 1371{ 1372 return __decode_pools(p, end, map, false); 1373} 1374 1375static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map) 1376{ 1377 return __decode_pools(p, end, map, true); 1378} 1379 1380typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool); 1381 1382static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root, 1383 decode_mapping_fn_t fn, bool incremental) 1384{ 1385 u32 n; 1386 1387 WARN_ON(!incremental && !fn); 1388 1389 ceph_decode_32_safe(p, end, n, e_inval); 1390 while (n--) { 1391 struct ceph_pg_mapping *pg; 1392 struct ceph_pg pgid; 1393 int ret; 1394 1395 ret = ceph_decode_pgid(p, end, &pgid); 1396 if (ret) 1397 return ret; 1398 1399 pg = lookup_pg_mapping(mapping_root, &pgid); 1400 if (pg) { 1401 WARN_ON(!incremental); 1402 erase_pg_mapping(mapping_root, pg); 1403 free_pg_mapping(pg); 1404 } 1405 1406 if (fn) { 1407 pg = fn(p, end, incremental); 1408 if (IS_ERR(pg)) 1409 return PTR_ERR(pg); 1410 1411 if (pg) { 1412 pg->pgid = pgid; /* struct */ 1413 insert_pg_mapping(mapping_root, pg); 1414 } 1415 } 1416 } 1417 1418 return 0; 1419 1420e_inval: 1421 return -EINVAL; 1422} 1423 1424static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end, 1425 bool incremental) 1426{ 1427 struct ceph_pg_mapping *pg; 1428 u32 len, i; 1429 1430 ceph_decode_32_safe(p, end, len, e_inval); 1431 if (len == 0 && incremental) 1432 return NULL; /* new_pg_temp: [] to remove */ 1433 if ((size_t)len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32)) 1434 return ERR_PTR(-EINVAL); 1435 1436 ceph_decode_need(p, end, len * sizeof(u32), e_inval); 1437 pg = alloc_pg_mapping(len * sizeof(u32)); 1438 if (!pg) 1439 return ERR_PTR(-ENOMEM); 1440 1441 pg->pg_temp.len = len; 1442 for (i = 0; i < len; i++) 1443 pg->pg_temp.osds[i] = ceph_decode_32(p); 1444 1445 return pg; 1446 1447e_inval: 1448 return ERR_PTR(-EINVAL); 1449} 1450 1451static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map) 1452{ 1453 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp, 1454 false); 1455} 1456 1457static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map) 1458{ 1459 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp, 1460 true); 1461} 1462 1463static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end, 1464 bool incremental) 1465{ 1466 struct ceph_pg_mapping *pg; 1467 u32 osd; 1468 1469 ceph_decode_32_safe(p, end, osd, e_inval); 1470 if (osd == (u32)-1 && incremental) 1471 return NULL; /* new_primary_temp: -1 to remove */ 1472 1473 pg = alloc_pg_mapping(0); 1474 if (!pg) 1475 return ERR_PTR(-ENOMEM); 1476 1477 pg->primary_temp.osd = osd; 1478 return pg; 1479 1480e_inval: 1481 return ERR_PTR(-EINVAL); 1482} 1483 1484static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map) 1485{ 1486 return decode_pg_mapping(p, end, &map->primary_temp, 1487 __decode_primary_temp, false); 1488} 1489 1490static int decode_new_primary_temp(void **p, void *end, 1491 struct ceph_osdmap *map) 1492{ 1493 return decode_pg_mapping(p, end, &map->primary_temp, 1494 __decode_primary_temp, true); 1495} 1496 1497u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd) 1498{ 1499 if (!map->osd_primary_affinity) 1500 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1501 1502 return map->osd_primary_affinity[osd]; 1503} 1504 1505static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff) 1506{ 1507 if (!map->osd_primary_affinity) { 1508 int i; 1509 1510 map->osd_primary_affinity = kvmalloc( 1511 array_size(map->max_osd, sizeof(*map->osd_primary_affinity)), 1512 GFP_NOFS); 1513 if (!map->osd_primary_affinity) 1514 return -ENOMEM; 1515 1516 for (i = 0; i < map->max_osd; i++) 1517 map->osd_primary_affinity[i] = 1518 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1519 } 1520 1521 map->osd_primary_affinity[osd] = aff; 1522 1523 return 0; 1524} 1525 1526static int decode_primary_affinity(void **p, void *end, 1527 struct ceph_osdmap *map) 1528{ 1529 u32 len, i; 1530 1531 ceph_decode_32_safe(p, end, len, e_inval); 1532 if (len == 0) { 1533 kvfree(map->osd_primary_affinity); 1534 map->osd_primary_affinity = NULL; 1535 return 0; 1536 } 1537 if (len != map->max_osd) 1538 goto e_inval; 1539 1540 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval); 1541 1542 for (i = 0; i < map->max_osd; i++) { 1543 int ret; 1544 1545 ret = set_primary_affinity(map, i, ceph_decode_32(p)); 1546 if (ret) 1547 return ret; 1548 } 1549 1550 return 0; 1551 1552e_inval: 1553 return -EINVAL; 1554} 1555 1556static int decode_new_primary_affinity(void **p, void *end, 1557 struct ceph_osdmap *map) 1558{ 1559 u32 n; 1560 1561 ceph_decode_32_safe(p, end, n, e_inval); 1562 while (n--) { 1563 u32 osd, aff; 1564 int ret; 1565 1566 ceph_decode_32_safe(p, end, osd, e_inval); 1567 ceph_decode_32_safe(p, end, aff, e_inval); 1568 if (osd >= map->max_osd) 1569 goto e_inval; 1570 1571 ret = set_primary_affinity(map, osd, aff); 1572 if (ret) 1573 return ret; 1574 1575 osdmap_info(map, "osd%d primary-affinity 0x%x\n", osd, aff); 1576 } 1577 1578 return 0; 1579 1580e_inval: 1581 return -EINVAL; 1582} 1583 1584static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end, 1585 bool __unused) 1586{ 1587 return __decode_pg_temp(p, end, false); 1588} 1589 1590static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map) 1591{ 1592 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap, 1593 false); 1594} 1595 1596static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map) 1597{ 1598 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap, 1599 true); 1600} 1601 1602static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map) 1603{ 1604 return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true); 1605} 1606 1607static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end, 1608 bool __unused) 1609{ 1610 struct ceph_pg_mapping *pg; 1611 u32 len, i; 1612 1613 ceph_decode_32_safe(p, end, len, e_inval); 1614 if ((size_t)len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32))) 1615 return ERR_PTR(-EINVAL); 1616 1617 ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval); 1618 pg = alloc_pg_mapping(2 * len * sizeof(u32)); 1619 if (!pg) 1620 return ERR_PTR(-ENOMEM); 1621 1622 pg->pg_upmap_items.len = len; 1623 for (i = 0; i < len; i++) { 1624 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p); 1625 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p); 1626 } 1627 1628 return pg; 1629 1630e_inval: 1631 return ERR_PTR(-EINVAL); 1632} 1633 1634static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map) 1635{ 1636 return decode_pg_mapping(p, end, &map->pg_upmap_items, 1637 __decode_pg_upmap_items, false); 1638} 1639 1640static int decode_new_pg_upmap_items(void **p, void *end, 1641 struct ceph_osdmap *map) 1642{ 1643 return decode_pg_mapping(p, end, &map->pg_upmap_items, 1644 __decode_pg_upmap_items, true); 1645} 1646 1647static int decode_old_pg_upmap_items(void **p, void *end, 1648 struct ceph_osdmap *map) 1649{ 1650 return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true); 1651} 1652 1653/* 1654 * decode a full map. 1655 */ 1656static int osdmap_decode(void **p, void *end, bool msgr2, 1657 struct ceph_osdmap *map) 1658{ 1659 u8 struct_v; 1660 u32 epoch = 0; 1661 void *start = *p; 1662 u32 max; 1663 u32 len, i; 1664 int err; 1665 1666 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1667 1668 err = get_osdmap_client_data_v(p, end, "full", &struct_v); 1669 if (err) 1670 goto bad; 1671 1672 /* fsid, epoch, created, modified */ 1673 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) + 1674 sizeof(map->created) + sizeof(map->modified), e_inval); 1675 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid)); 1676 epoch = map->epoch = ceph_decode_32(p); 1677 ceph_decode_copy(p, &map->created, sizeof(map->created)); 1678 ceph_decode_copy(p, &map->modified, sizeof(map->modified)); 1679 1680 /* pools */ 1681 err = decode_pools(p, end, map); 1682 if (err) 1683 goto bad; 1684 1685 /* pool_name */ 1686 err = decode_pool_names(p, end, map); 1687 if (err) 1688 goto bad; 1689 1690 ceph_decode_32_safe(p, end, map->pool_max, e_inval); 1691 1692 ceph_decode_32_safe(p, end, map->flags, e_inval); 1693 1694 /* max_osd */ 1695 ceph_decode_32_safe(p, end, max, e_inval); 1696 1697 /* (re)alloc osd arrays */ 1698 err = osdmap_set_max_osd(map, max); 1699 if (err) 1700 goto bad; 1701 1702 /* osd_state, osd_weight, osd_addrs->client_addr */ 1703 ceph_decode_need(p, end, 3*sizeof(u32) + 1704 map->max_osd*(struct_v >= 5 ? sizeof(u32) : 1705 sizeof(u8)) + 1706 sizeof(*map->osd_weight), e_inval); 1707 if (ceph_decode_32(p) != map->max_osd) 1708 goto e_inval; 1709 1710 if (struct_v >= 5) { 1711 for (i = 0; i < map->max_osd; i++) 1712 map->osd_state[i] = ceph_decode_32(p); 1713 } else { 1714 for (i = 0; i < map->max_osd; i++) 1715 map->osd_state[i] = ceph_decode_8(p); 1716 } 1717 1718 if (ceph_decode_32(p) != map->max_osd) 1719 goto e_inval; 1720 1721 for (i = 0; i < map->max_osd; i++) 1722 map->osd_weight[i] = ceph_decode_32(p); 1723 1724 if (ceph_decode_32(p) != map->max_osd) 1725 goto e_inval; 1726 1727 for (i = 0; i < map->max_osd; i++) { 1728 struct ceph_entity_addr *addr = &map->osd_addr[i]; 1729 1730 if (struct_v >= 8) 1731 err = ceph_decode_entity_addrvec(p, end, msgr2, addr); 1732 else 1733 err = ceph_decode_entity_addr(p, end, addr); 1734 if (err) 1735 goto bad; 1736 1737 dout("%s osd%d addr %s\n", __func__, i, ceph_pr_addr(addr)); 1738 } 1739 1740 /* pg_temp */ 1741 err = decode_pg_temp(p, end, map); 1742 if (err) 1743 goto bad; 1744 1745 /* primary_temp */ 1746 if (struct_v >= 1) { 1747 err = decode_primary_temp(p, end, map); 1748 if (err) 1749 goto bad; 1750 } 1751 1752 /* primary_affinity */ 1753 if (struct_v >= 2) { 1754 err = decode_primary_affinity(p, end, map); 1755 if (err) 1756 goto bad; 1757 } else { 1758 WARN_ON(map->osd_primary_affinity); 1759 } 1760 1761 /* crush */ 1762 ceph_decode_32_safe(p, end, len, e_inval); 1763 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end))); 1764 if (err) 1765 goto bad; 1766 1767 *p += len; 1768 if (struct_v >= 3) { 1769 /* erasure_code_profiles */ 1770 ceph_decode_skip_map_of_map(p, end, string, string, string, 1771 e_inval); 1772 } 1773 1774 if (struct_v >= 4) { 1775 err = decode_pg_upmap(p, end, map); 1776 if (err) 1777 goto bad; 1778 1779 err = decode_pg_upmap_items(p, end, map); 1780 if (err) 1781 goto bad; 1782 } else { 1783 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap)); 1784 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items)); 1785 } 1786 1787 /* ignore the rest */ 1788 *p = end; 1789 1790 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1791 return 0; 1792 1793e_inval: 1794 err = -EINVAL; 1795bad: 1796 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1797 err, epoch, (int)(*p - start), *p, start, end); 1798 print_hex_dump(KERN_DEBUG, "osdmap: ", 1799 DUMP_PREFIX_OFFSET, 16, 1, 1800 start, end - start, true); 1801 return err; 1802} 1803 1804/* 1805 * Allocate and decode a full map. 1806 */ 1807struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end, bool msgr2) 1808{ 1809 struct ceph_osdmap *map; 1810 int ret; 1811 1812 map = ceph_osdmap_alloc(); 1813 if (!map) 1814 return ERR_PTR(-ENOMEM); 1815 1816 ret = osdmap_decode(p, end, msgr2, map); 1817 if (ret) { 1818 ceph_osdmap_destroy(map); 1819 return ERR_PTR(ret); 1820 } 1821 1822 return map; 1823} 1824 1825/* 1826 * Encoding order is (new_up_client, new_state, new_weight). Need to 1827 * apply in the (new_weight, new_state, new_up_client) order, because 1828 * an incremental map may look like e.g. 1829 * 1830 * new_up_client: { osd=6, addr=... } # set osd_state and addr 1831 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state 1832 */ 1833static int decode_new_up_state_weight(void **p, void *end, u8 struct_v, 1834 bool msgr2, struct ceph_osdmap *map) 1835{ 1836 void *new_up_client; 1837 void *new_state; 1838 void *new_weight_end; 1839 u32 len; 1840 int ret; 1841 int i; 1842 1843 new_up_client = *p; 1844 ceph_decode_32_safe(p, end, len, e_inval); 1845 for (i = 0; i < len; ++i) { 1846 struct ceph_entity_addr addr; 1847 1848 ceph_decode_skip_32(p, end, e_inval); 1849 if (struct_v >= 7) 1850 ret = ceph_decode_entity_addrvec(p, end, msgr2, &addr); 1851 else 1852 ret = ceph_decode_entity_addr(p, end, &addr); 1853 if (ret) 1854 return ret; 1855 } 1856 1857 new_state = *p; 1858 ceph_decode_32_safe(p, end, len, e_inval); 1859 len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8)); 1860 ceph_decode_need(p, end, len, e_inval); 1861 *p += len; 1862 1863 /* new_weight */ 1864 ceph_decode_32_safe(p, end, len, e_inval); 1865 while (len--) { 1866 s32 osd; 1867 u32 w; 1868 1869 ceph_decode_need(p, end, 2*sizeof(u32), e_inval); 1870 osd = ceph_decode_32(p); 1871 w = ceph_decode_32(p); 1872 if (osd >= map->max_osd) 1873 goto e_inval; 1874 1875 osdmap_info(map, "osd%d weight 0x%x %s\n", osd, w, 1876 w == CEPH_OSD_IN ? "(in)" : 1877 (w == CEPH_OSD_OUT ? "(out)" : "")); 1878 map->osd_weight[osd] = w; 1879 1880 /* 1881 * If we are marking in, set the EXISTS, and clear the 1882 * AUTOOUT and NEW bits. 1883 */ 1884 if (w) { 1885 map->osd_state[osd] |= CEPH_OSD_EXISTS; 1886 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT | 1887 CEPH_OSD_NEW); 1888 } 1889 } 1890 new_weight_end = *p; 1891 1892 /* new_state (up/down) */ 1893 *p = new_state; 1894 len = ceph_decode_32(p); 1895 while (len--) { 1896 s32 osd; 1897 u32 xorstate; 1898 1899 osd = ceph_decode_32(p); 1900 if (osd >= map->max_osd) 1901 goto e_inval; 1902 1903 if (struct_v >= 5) 1904 xorstate = ceph_decode_32(p); 1905 else 1906 xorstate = ceph_decode_8(p); 1907 if (xorstate == 0) 1908 xorstate = CEPH_OSD_UP; 1909 if ((map->osd_state[osd] & CEPH_OSD_UP) && 1910 (xorstate & CEPH_OSD_UP)) 1911 osdmap_info(map, "osd%d down\n", osd); 1912 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) && 1913 (xorstate & CEPH_OSD_EXISTS)) { 1914 osdmap_info(map, "osd%d does not exist\n", osd); 1915 ret = set_primary_affinity(map, osd, 1916 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY); 1917 if (ret) 1918 return ret; 1919 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr)); 1920 map->osd_state[osd] = 0; 1921 } else { 1922 map->osd_state[osd] ^= xorstate; 1923 } 1924 } 1925 1926 /* new_up_client */ 1927 *p = new_up_client; 1928 len = ceph_decode_32(p); 1929 while (len--) { 1930 s32 osd; 1931 struct ceph_entity_addr addr; 1932 1933 osd = ceph_decode_32(p); 1934 if (osd >= map->max_osd) 1935 goto e_inval; 1936 1937 if (struct_v >= 7) 1938 ret = ceph_decode_entity_addrvec(p, end, msgr2, &addr); 1939 else 1940 ret = ceph_decode_entity_addr(p, end, &addr); 1941 if (ret) 1942 return ret; 1943 1944 dout("%s osd%d addr %s\n", __func__, osd, ceph_pr_addr(&addr)); 1945 1946 osdmap_info(map, "osd%d up\n", osd); 1947 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP; 1948 map->osd_addr[osd] = addr; 1949 } 1950 1951 *p = new_weight_end; 1952 return 0; 1953 1954e_inval: 1955 return -EINVAL; 1956} 1957 1958/* 1959 * decode and apply an incremental map update. 1960 */ 1961struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, bool msgr2, 1962 struct ceph_osdmap *map) 1963{ 1964 struct ceph_fsid fsid; 1965 u32 epoch = 0; 1966 struct ceph_timespec modified; 1967 s32 len; 1968 u64 pool; 1969 __s64 new_pool_max; 1970 __s32 new_flags, max; 1971 void *start = *p; 1972 int err; 1973 u8 struct_v; 1974 1975 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1976 1977 err = get_osdmap_client_data_v(p, end, "inc", &struct_v); 1978 if (err) 1979 goto bad; 1980 1981 /* fsid, epoch, modified, new_pool_max, new_flags */ 1982 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) + 1983 sizeof(u64) + sizeof(u32), e_inval); 1984 ceph_decode_copy(p, &fsid, sizeof(fsid)); 1985 epoch = ceph_decode_32(p); 1986 ceph_decode_copy(p, &modified, sizeof(modified)); 1987 new_pool_max = ceph_decode_64(p); 1988 new_flags = ceph_decode_32(p); 1989 1990 if (epoch != map->epoch + 1) 1991 goto e_inval; 1992 1993 /* full map? */ 1994 ceph_decode_32_safe(p, end, len, e_inval); 1995 if (len > 0) { 1996 dout("apply_incremental full map len %d, %p to %p\n", 1997 len, *p, end); 1998 return ceph_osdmap_decode(p, min(*p+len, end), msgr2); 1999 } 2000 2001 /* new crush? */ 2002 ceph_decode_32_safe(p, end, len, e_inval); 2003 if (len > 0) { 2004 err = osdmap_set_crush(map, 2005 crush_decode(*p, min(*p + len, end))); 2006 if (err) 2007 goto bad; 2008 *p += len; 2009 } 2010 2011 /* new flags? */ 2012 if (new_flags >= 0) 2013 map->flags = new_flags; 2014 if (new_pool_max >= 0) 2015 map->pool_max = new_pool_max; 2016 2017 /* new max? */ 2018 ceph_decode_32_safe(p, end, max, e_inval); 2019 if (max >= 0) { 2020 err = osdmap_set_max_osd(map, max); 2021 if (err) 2022 goto bad; 2023 } 2024 2025 map->epoch++; 2026 map->modified = modified; 2027 2028 /* new_pools */ 2029 err = decode_new_pools(p, end, map); 2030 if (err) 2031 goto bad; 2032 2033 /* new_pool_names */ 2034 err = decode_pool_names(p, end, map); 2035 if (err) 2036 goto bad; 2037 2038 /* old_pool */ 2039 ceph_decode_32_safe(p, end, len, e_inval); 2040 while (len--) { 2041 struct ceph_pg_pool_info *pi; 2042 2043 ceph_decode_64_safe(p, end, pool, e_inval); 2044 pi = lookup_pg_pool(&map->pg_pools, pool); 2045 if (pi) 2046 __remove_pg_pool(&map->pg_pools, pi); 2047 } 2048 2049 /* new_up_client, new_state, new_weight */ 2050 err = decode_new_up_state_weight(p, end, struct_v, msgr2, map); 2051 if (err) 2052 goto bad; 2053 2054 /* new_pg_temp */ 2055 err = decode_new_pg_temp(p, end, map); 2056 if (err) 2057 goto bad; 2058 2059 /* new_primary_temp */ 2060 if (struct_v >= 1) { 2061 err = decode_new_primary_temp(p, end, map); 2062 if (err) 2063 goto bad; 2064 } 2065 2066 /* new_primary_affinity */ 2067 if (struct_v >= 2) { 2068 err = decode_new_primary_affinity(p, end, map); 2069 if (err) 2070 goto bad; 2071 } 2072 2073 if (struct_v >= 3) { 2074 /* new_erasure_code_profiles */ 2075 ceph_decode_skip_map_of_map(p, end, string, string, string, 2076 e_inval); 2077 /* old_erasure_code_profiles */ 2078 ceph_decode_skip_set(p, end, string, e_inval); 2079 } 2080 2081 if (struct_v >= 4) { 2082 err = decode_new_pg_upmap(p, end, map); 2083 if (err) 2084 goto bad; 2085 2086 err = decode_old_pg_upmap(p, end, map); 2087 if (err) 2088 goto bad; 2089 2090 err = decode_new_pg_upmap_items(p, end, map); 2091 if (err) 2092 goto bad; 2093 2094 err = decode_old_pg_upmap_items(p, end, map); 2095 if (err) 2096 goto bad; 2097 } 2098 2099 /* ignore the rest */ 2100 *p = end; 2101 2102 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 2103 return map; 2104 2105e_inval: 2106 err = -EINVAL; 2107bad: 2108 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 2109 err, epoch, (int)(*p - start), *p, start, end); 2110 print_hex_dump(KERN_DEBUG, "osdmap: ", 2111 DUMP_PREFIX_OFFSET, 16, 1, 2112 start, end - start, true); 2113 return ERR_PTR(err); 2114} 2115 2116void ceph_oloc_copy(struct ceph_object_locator *dest, 2117 const struct ceph_object_locator *src) 2118{ 2119 ceph_oloc_destroy(dest); 2120 2121 dest->pool = src->pool; 2122 if (src->pool_ns) 2123 dest->pool_ns = ceph_get_string(src->pool_ns); 2124 else 2125 dest->pool_ns = NULL; 2126} 2127EXPORT_SYMBOL(ceph_oloc_copy); 2128 2129void ceph_oloc_destroy(struct ceph_object_locator *oloc) 2130{ 2131 ceph_put_string(oloc->pool_ns); 2132} 2133EXPORT_SYMBOL(ceph_oloc_destroy); 2134 2135void ceph_oid_copy(struct ceph_object_id *dest, 2136 const struct ceph_object_id *src) 2137{ 2138 ceph_oid_destroy(dest); 2139 2140 if (src->name != src->inline_name) { 2141 /* very rare, see ceph_object_id definition */ 2142 dest->name = kmalloc(src->name_len + 1, 2143 GFP_NOIO | __GFP_NOFAIL); 2144 } else { 2145 dest->name = dest->inline_name; 2146 } 2147 memcpy(dest->name, src->name, src->name_len + 1); 2148 dest->name_len = src->name_len; 2149} 2150EXPORT_SYMBOL(ceph_oid_copy); 2151 2152static __printf(2, 0) 2153int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap) 2154{ 2155 int len; 2156 2157 WARN_ON(!ceph_oid_empty(oid)); 2158 2159 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap); 2160 if (len >= sizeof(oid->inline_name)) 2161 return len; 2162 2163 oid->name_len = len; 2164 return 0; 2165} 2166 2167/* 2168 * If oid doesn't fit into inline buffer, BUG. 2169 */ 2170void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...) 2171{ 2172 va_list ap; 2173 2174 va_start(ap, fmt); 2175 BUG_ON(oid_printf_vargs(oid, fmt, ap)); 2176 va_end(ap); 2177} 2178EXPORT_SYMBOL(ceph_oid_printf); 2179 2180static __printf(3, 0) 2181int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp, 2182 const char *fmt, va_list ap) 2183{ 2184 va_list aq; 2185 int len; 2186 2187 va_copy(aq, ap); 2188 len = oid_printf_vargs(oid, fmt, aq); 2189 va_end(aq); 2190 2191 if (len) { 2192 char *external_name; 2193 2194 external_name = kmalloc(len + 1, gfp); 2195 if (!external_name) 2196 return -ENOMEM; 2197 2198 oid->name = external_name; 2199 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len); 2200 oid->name_len = len; 2201 } 2202 2203 return 0; 2204} 2205 2206/* 2207 * If oid doesn't fit into inline buffer, allocate. 2208 */ 2209int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp, 2210 const char *fmt, ...) 2211{ 2212 va_list ap; 2213 int ret; 2214 2215 va_start(ap, fmt); 2216 ret = oid_aprintf_vargs(oid, gfp, fmt, ap); 2217 va_end(ap); 2218 2219 return ret; 2220} 2221EXPORT_SYMBOL(ceph_oid_aprintf); 2222 2223void ceph_oid_destroy(struct ceph_object_id *oid) 2224{ 2225 if (oid->name != oid->inline_name) 2226 kfree(oid->name); 2227} 2228EXPORT_SYMBOL(ceph_oid_destroy); 2229 2230/* 2231 * osds only 2232 */ 2233static bool __osds_equal(const struct ceph_osds *lhs, 2234 const struct ceph_osds *rhs) 2235{ 2236 if (lhs->size == rhs->size && 2237 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0]))) 2238 return true; 2239 2240 return false; 2241} 2242 2243/* 2244 * osds + primary 2245 */ 2246static bool osds_equal(const struct ceph_osds *lhs, 2247 const struct ceph_osds *rhs) 2248{ 2249 if (__osds_equal(lhs, rhs) && 2250 lhs->primary == rhs->primary) 2251 return true; 2252 2253 return false; 2254} 2255 2256static bool osds_valid(const struct ceph_osds *set) 2257{ 2258 /* non-empty set */ 2259 if (set->size > 0 && set->primary >= 0) 2260 return true; 2261 2262 /* empty can_shift_osds set */ 2263 if (!set->size && set->primary == -1) 2264 return true; 2265 2266 /* empty !can_shift_osds set - all NONE */ 2267 if (set->size > 0 && set->primary == -1) { 2268 int i; 2269 2270 for (i = 0; i < set->size; i++) { 2271 if (set->osds[i] != CRUSH_ITEM_NONE) 2272 break; 2273 } 2274 if (i == set->size) 2275 return true; 2276 } 2277 2278 return false; 2279} 2280 2281void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src) 2282{ 2283 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0])); 2284 dest->size = src->size; 2285 dest->primary = src->primary; 2286} 2287 2288bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num, 2289 u32 new_pg_num) 2290{ 2291 int old_bits = calc_bits_of(old_pg_num); 2292 int old_mask = (1 << old_bits) - 1; 2293 int n; 2294 2295 WARN_ON(pgid->seed >= old_pg_num); 2296 if (new_pg_num <= old_pg_num) 2297 return false; 2298 2299 for (n = 1; ; n++) { 2300 int next_bit = n << (old_bits - 1); 2301 u32 s = next_bit | pgid->seed; 2302 2303 if (s < old_pg_num || s == pgid->seed) 2304 continue; 2305 if (s >= new_pg_num) 2306 break; 2307 2308 s = ceph_stable_mod(s, old_pg_num, old_mask); 2309 if (s == pgid->seed) 2310 return true; 2311 } 2312 2313 return false; 2314} 2315 2316bool ceph_is_new_interval(const struct ceph_osds *old_acting, 2317 const struct ceph_osds *new_acting, 2318 const struct ceph_osds *old_up, 2319 const struct ceph_osds *new_up, 2320 int old_size, 2321 int new_size, 2322 int old_min_size, 2323 int new_min_size, 2324 u32 old_pg_num, 2325 u32 new_pg_num, 2326 bool old_sort_bitwise, 2327 bool new_sort_bitwise, 2328 bool old_recovery_deletes, 2329 bool new_recovery_deletes, 2330 const struct ceph_pg *pgid) 2331{ 2332 return !osds_equal(old_acting, new_acting) || 2333 !osds_equal(old_up, new_up) || 2334 old_size != new_size || 2335 old_min_size != new_min_size || 2336 ceph_pg_is_split(pgid, old_pg_num, new_pg_num) || 2337 old_sort_bitwise != new_sort_bitwise || 2338 old_recovery_deletes != new_recovery_deletes; 2339} 2340 2341static int calc_pg_rank(int osd, const struct ceph_osds *acting) 2342{ 2343 int i; 2344 2345 for (i = 0; i < acting->size; i++) { 2346 if (acting->osds[i] == osd) 2347 return i; 2348 } 2349 2350 return -1; 2351} 2352 2353static bool primary_changed(const struct ceph_osds *old_acting, 2354 const struct ceph_osds *new_acting) 2355{ 2356 if (!old_acting->size && !new_acting->size) 2357 return false; /* both still empty */ 2358 2359 if (!old_acting->size ^ !new_acting->size) 2360 return true; /* was empty, now not, or vice versa */ 2361 2362 if (old_acting->primary != new_acting->primary) 2363 return true; /* primary changed */ 2364 2365 if (calc_pg_rank(old_acting->primary, old_acting) != 2366 calc_pg_rank(new_acting->primary, new_acting)) 2367 return true; 2368 2369 return false; /* same primary (tho replicas may have changed) */ 2370} 2371 2372bool ceph_osds_changed(const struct ceph_osds *old_acting, 2373 const struct ceph_osds *new_acting, 2374 bool any_change) 2375{ 2376 if (primary_changed(old_acting, new_acting)) 2377 return true; 2378 2379 if (any_change && !__osds_equal(old_acting, new_acting)) 2380 return true; 2381 2382 return false; 2383} 2384 2385/* 2386 * Map an object into a PG. 2387 * 2388 * Should only be called with target_oid and target_oloc (as opposed to 2389 * base_oid and base_oloc), since tiering isn't taken into account. 2390 */ 2391void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi, 2392 const struct ceph_object_id *oid, 2393 const struct ceph_object_locator *oloc, 2394 struct ceph_pg *raw_pgid) 2395{ 2396 WARN_ON(pi->id != oloc->pool); 2397 2398 if (!oloc->pool_ns) { 2399 raw_pgid->pool = oloc->pool; 2400 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name, 2401 oid->name_len); 2402 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name, 2403 raw_pgid->pool, raw_pgid->seed); 2404 } else { 2405 char stack_buf[256]; 2406 char *buf = stack_buf; 2407 int nsl = oloc->pool_ns->len; 2408 size_t total = nsl + 1 + oid->name_len; 2409 2410 if (total > sizeof(stack_buf)) 2411 buf = kmalloc(total, GFP_NOIO | __GFP_NOFAIL); 2412 memcpy(buf, oloc->pool_ns->str, nsl); 2413 buf[nsl] = '\037'; 2414 memcpy(buf + nsl + 1, oid->name, oid->name_len); 2415 raw_pgid->pool = oloc->pool; 2416 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total); 2417 if (buf != stack_buf) 2418 kfree(buf); 2419 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__, 2420 oid->name, nsl, oloc->pool_ns->str, 2421 raw_pgid->pool, raw_pgid->seed); 2422 } 2423} 2424 2425int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 2426 const struct ceph_object_id *oid, 2427 const struct ceph_object_locator *oloc, 2428 struct ceph_pg *raw_pgid) 2429{ 2430 struct ceph_pg_pool_info *pi; 2431 2432 pi = ceph_pg_pool_by_id(osdmap, oloc->pool); 2433 if (!pi) 2434 return -ENOENT; 2435 2436 __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid); 2437 return 0; 2438} 2439EXPORT_SYMBOL(ceph_object_locator_to_pg); 2440 2441/* 2442 * Map a raw PG (full precision ps) into an actual PG. 2443 */ 2444static void raw_pg_to_pg(struct ceph_pg_pool_info *pi, 2445 const struct ceph_pg *raw_pgid, 2446 struct ceph_pg *pgid) 2447{ 2448 pgid->pool = raw_pgid->pool; 2449 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num, 2450 pi->pg_num_mask); 2451} 2452 2453/* 2454 * Map a raw PG (full precision ps) into a placement ps (placement 2455 * seed). Include pool id in that value so that different pools don't 2456 * use the same seeds. 2457 */ 2458static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi, 2459 const struct ceph_pg *raw_pgid) 2460{ 2461 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) { 2462 /* hash pool id and seed so that pool PGs do not overlap */ 2463 return crush_hash32_2(CRUSH_HASH_RJENKINS1, 2464 ceph_stable_mod(raw_pgid->seed, 2465 pi->pgp_num, 2466 pi->pgp_num_mask), 2467 raw_pgid->pool); 2468 } else { 2469 /* 2470 * legacy behavior: add ps and pool together. this is 2471 * not a great approach because the PGs from each pool 2472 * will overlap on top of each other: 0.5 == 1.4 == 2473 * 2.3 == ... 2474 */ 2475 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num, 2476 pi->pgp_num_mask) + 2477 (unsigned)raw_pgid->pool; 2478 } 2479} 2480 2481/* 2482 * Magic value used for a "default" fallback choose_args, used if the 2483 * crush_choose_arg_map passed to do_crush() does not exist. If this 2484 * also doesn't exist, fall back to canonical weights. 2485 */ 2486#define CEPH_DEFAULT_CHOOSE_ARGS -1 2487 2488static int do_crush(struct ceph_osdmap *map, int ruleno, int x, 2489 int *result, int result_max, 2490 const __u32 *weight, int weight_max, 2491 s64 choose_args_index) 2492{ 2493 struct crush_choose_arg_map *arg_map; 2494 struct crush_work *work; 2495 int r; 2496 2497 BUG_ON(result_max > CEPH_PG_MAX_SIZE); 2498 2499 arg_map = lookup_choose_arg_map(&map->crush->choose_args, 2500 choose_args_index); 2501 if (!arg_map) 2502 arg_map = lookup_choose_arg_map(&map->crush->choose_args, 2503 CEPH_DEFAULT_CHOOSE_ARGS); 2504 2505 work = get_workspace(&map->crush_wsm, map->crush); 2506 r = crush_do_rule(map->crush, ruleno, x, result, result_max, 2507 weight, weight_max, work, 2508 arg_map ? arg_map->args : NULL); 2509 put_workspace(&map->crush_wsm, work); 2510 return r; 2511} 2512 2513static void remove_nonexistent_osds(struct ceph_osdmap *osdmap, 2514 struct ceph_pg_pool_info *pi, 2515 struct ceph_osds *set) 2516{ 2517 int i; 2518 2519 if (ceph_can_shift_osds(pi)) { 2520 int removed = 0; 2521 2522 /* shift left */ 2523 for (i = 0; i < set->size; i++) { 2524 if (!ceph_osd_exists(osdmap, set->osds[i])) { 2525 removed++; 2526 continue; 2527 } 2528 if (removed) 2529 set->osds[i - removed] = set->osds[i]; 2530 } 2531 set->size -= removed; 2532 } else { 2533 /* set dne devices to NONE */ 2534 for (i = 0; i < set->size; i++) { 2535 if (!ceph_osd_exists(osdmap, set->osds[i])) 2536 set->osds[i] = CRUSH_ITEM_NONE; 2537 } 2538 } 2539} 2540 2541/* 2542 * Calculate raw set (CRUSH output) for given PG and filter out 2543 * nonexistent OSDs. ->primary is undefined for a raw set. 2544 * 2545 * Placement seed (CRUSH input) is returned through @ppps. 2546 */ 2547static void pg_to_raw_osds(struct ceph_osdmap *osdmap, 2548 struct ceph_pg_pool_info *pi, 2549 const struct ceph_pg *raw_pgid, 2550 struct ceph_osds *raw, 2551 u32 *ppps) 2552{ 2553 u32 pps = raw_pg_to_pps(pi, raw_pgid); 2554 int ruleno; 2555 int len; 2556 2557 ceph_osds_init(raw); 2558 if (ppps) 2559 *ppps = pps; 2560 2561 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type, 2562 pi->size); 2563 if (ruleno < 0) { 2564 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n", 2565 pi->id, pi->crush_ruleset, pi->type, pi->size); 2566 return; 2567 } 2568 2569 if (pi->size > ARRAY_SIZE(raw->osds)) { 2570 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n", 2571 pi->id, pi->crush_ruleset, pi->type, pi->size, 2572 ARRAY_SIZE(raw->osds)); 2573 return; 2574 } 2575 2576 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size, 2577 osdmap->osd_weight, osdmap->max_osd, pi->id); 2578 if (len < 0) { 2579 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n", 2580 len, ruleno, pi->id, pi->crush_ruleset, pi->type, 2581 pi->size); 2582 return; 2583 } 2584 2585 raw->size = len; 2586 remove_nonexistent_osds(osdmap, pi, raw); 2587} 2588 2589/* apply pg_upmap[_items] mappings */ 2590static void apply_upmap(struct ceph_osdmap *osdmap, 2591 const struct ceph_pg *pgid, 2592 struct ceph_osds *raw) 2593{ 2594 struct ceph_pg_mapping *pg; 2595 int i, j; 2596 2597 pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid); 2598 if (pg) { 2599 /* make sure targets aren't marked out */ 2600 for (i = 0; i < pg->pg_upmap.len; i++) { 2601 int osd = pg->pg_upmap.osds[i]; 2602 2603 if (osd != CRUSH_ITEM_NONE && 2604 osd < osdmap->max_osd && 2605 osdmap->osd_weight[osd] == 0) { 2606 /* reject/ignore explicit mapping */ 2607 return; 2608 } 2609 } 2610 for (i = 0; i < pg->pg_upmap.len; i++) 2611 raw->osds[i] = pg->pg_upmap.osds[i]; 2612 raw->size = pg->pg_upmap.len; 2613 /* check and apply pg_upmap_items, if any */ 2614 } 2615 2616 pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid); 2617 if (pg) { 2618 /* 2619 * Note: this approach does not allow a bidirectional swap, 2620 * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1]. 2621 */ 2622 for (i = 0; i < pg->pg_upmap_items.len; i++) { 2623 int from = pg->pg_upmap_items.from_to[i][0]; 2624 int to = pg->pg_upmap_items.from_to[i][1]; 2625 int pos = -1; 2626 bool exists = false; 2627 2628 /* make sure replacement doesn't already appear */ 2629 for (j = 0; j < raw->size; j++) { 2630 int osd = raw->osds[j]; 2631 2632 if (osd == to) { 2633 exists = true; 2634 break; 2635 } 2636 /* ignore mapping if target is marked out */ 2637 if (osd == from && pos < 0 && 2638 !(to != CRUSH_ITEM_NONE && 2639 to < osdmap->max_osd && 2640 osdmap->osd_weight[to] == 0)) { 2641 pos = j; 2642 } 2643 } 2644 if (!exists && pos >= 0) 2645 raw->osds[pos] = to; 2646 } 2647 } 2648} 2649 2650/* 2651 * Given raw set, calculate up set and up primary. By definition of an 2652 * up set, the result won't contain nonexistent or down OSDs. 2653 * 2654 * This is done in-place - on return @set is the up set. If it's 2655 * empty, ->primary will remain undefined. 2656 */ 2657static void raw_to_up_osds(struct ceph_osdmap *osdmap, 2658 struct ceph_pg_pool_info *pi, 2659 struct ceph_osds *set) 2660{ 2661 int i; 2662 2663 /* ->primary is undefined for a raw set */ 2664 BUG_ON(set->primary != -1); 2665 2666 if (ceph_can_shift_osds(pi)) { 2667 int removed = 0; 2668 2669 /* shift left */ 2670 for (i = 0; i < set->size; i++) { 2671 if (ceph_osd_is_down(osdmap, set->osds[i])) { 2672 removed++; 2673 continue; 2674 } 2675 if (removed) 2676 set->osds[i - removed] = set->osds[i]; 2677 } 2678 set->size -= removed; 2679 if (set->size > 0) 2680 set->primary = set->osds[0]; 2681 } else { 2682 /* set down/dne devices to NONE */ 2683 for (i = set->size - 1; i >= 0; i--) { 2684 if (ceph_osd_is_down(osdmap, set->osds[i])) 2685 set->osds[i] = CRUSH_ITEM_NONE; 2686 else 2687 set->primary = set->osds[i]; 2688 } 2689 } 2690} 2691 2692static void apply_primary_affinity(struct ceph_osdmap *osdmap, 2693 struct ceph_pg_pool_info *pi, 2694 u32 pps, 2695 struct ceph_osds *up) 2696{ 2697 int i; 2698 int pos = -1; 2699 2700 /* 2701 * Do we have any non-default primary_affinity values for these 2702 * osds? 2703 */ 2704 if (!osdmap->osd_primary_affinity) 2705 return; 2706 2707 for (i = 0; i < up->size; i++) { 2708 int osd = up->osds[i]; 2709 2710 if (osd != CRUSH_ITEM_NONE && 2711 osdmap->osd_primary_affinity[osd] != 2712 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) { 2713 break; 2714 } 2715 } 2716 if (i == up->size) 2717 return; 2718 2719 /* 2720 * Pick the primary. Feed both the seed (for the pg) and the 2721 * osd into the hash/rng so that a proportional fraction of an 2722 * osd's pgs get rejected as primary. 2723 */ 2724 for (i = 0; i < up->size; i++) { 2725 int osd = up->osds[i]; 2726 u32 aff; 2727 2728 if (osd == CRUSH_ITEM_NONE) 2729 continue; 2730 2731 aff = osdmap->osd_primary_affinity[osd]; 2732 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY && 2733 (crush_hash32_2(CRUSH_HASH_RJENKINS1, 2734 pps, osd) >> 16) >= aff) { 2735 /* 2736 * We chose not to use this primary. Note it 2737 * anyway as a fallback in case we don't pick 2738 * anyone else, but keep looking. 2739 */ 2740 if (pos < 0) 2741 pos = i; 2742 } else { 2743 pos = i; 2744 break; 2745 } 2746 } 2747 if (pos < 0) 2748 return; 2749 2750 up->primary = up->osds[pos]; 2751 2752 if (ceph_can_shift_osds(pi) && pos > 0) { 2753 /* move the new primary to the front */ 2754 for (i = pos; i > 0; i--) 2755 up->osds[i] = up->osds[i - 1]; 2756 up->osds[0] = up->primary; 2757 } 2758} 2759 2760/* 2761 * Get pg_temp and primary_temp mappings for given PG. 2762 * 2763 * Note that a PG may have none, only pg_temp, only primary_temp or 2764 * both pg_temp and primary_temp mappings. This means @temp isn't 2765 * always a valid OSD set on return: in the "only primary_temp" case, 2766 * @temp will have its ->primary >= 0 but ->size == 0. 2767 */ 2768static void get_temp_osds(struct ceph_osdmap *osdmap, 2769 struct ceph_pg_pool_info *pi, 2770 const struct ceph_pg *pgid, 2771 struct ceph_osds *temp) 2772{ 2773 struct ceph_pg_mapping *pg; 2774 int i; 2775 2776 ceph_osds_init(temp); 2777 2778 /* pg_temp? */ 2779 pg = lookup_pg_mapping(&osdmap->pg_temp, pgid); 2780 if (pg) { 2781 for (i = 0; i < pg->pg_temp.len; i++) { 2782 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) { 2783 if (ceph_can_shift_osds(pi)) 2784 continue; 2785 2786 temp->osds[temp->size++] = CRUSH_ITEM_NONE; 2787 } else { 2788 temp->osds[temp->size++] = pg->pg_temp.osds[i]; 2789 } 2790 } 2791 2792 /* apply pg_temp's primary */ 2793 for (i = 0; i < temp->size; i++) { 2794 if (temp->osds[i] != CRUSH_ITEM_NONE) { 2795 temp->primary = temp->osds[i]; 2796 break; 2797 } 2798 } 2799 } 2800 2801 /* primary_temp? */ 2802 pg = lookup_pg_mapping(&osdmap->primary_temp, pgid); 2803 if (pg) 2804 temp->primary = pg->primary_temp.osd; 2805} 2806 2807/* 2808 * Map a PG to its acting set as well as its up set. 2809 * 2810 * Acting set is used for data mapping purposes, while up set can be 2811 * recorded for detecting interval changes and deciding whether to 2812 * resend a request. 2813 */ 2814void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 2815 struct ceph_pg_pool_info *pi, 2816 const struct ceph_pg *raw_pgid, 2817 struct ceph_osds *up, 2818 struct ceph_osds *acting) 2819{ 2820 struct ceph_pg pgid; 2821 u32 pps; 2822 2823 WARN_ON(pi->id != raw_pgid->pool); 2824 raw_pg_to_pg(pi, raw_pgid, &pgid); 2825 2826 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps); 2827 apply_upmap(osdmap, &pgid, up); 2828 raw_to_up_osds(osdmap, pi, up); 2829 apply_primary_affinity(osdmap, pi, pps, up); 2830 get_temp_osds(osdmap, pi, &pgid, acting); 2831 if (!acting->size) { 2832 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0])); 2833 acting->size = up->size; 2834 if (acting->primary == -1) 2835 acting->primary = up->primary; 2836 } 2837 WARN_ON(!osds_valid(up) || !osds_valid(acting)); 2838} 2839 2840bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap, 2841 struct ceph_pg_pool_info *pi, 2842 const struct ceph_pg *raw_pgid, 2843 struct ceph_spg *spgid) 2844{ 2845 struct ceph_pg pgid; 2846 struct ceph_osds up, acting; 2847 int i; 2848 2849 WARN_ON(pi->id != raw_pgid->pool); 2850 raw_pg_to_pg(pi, raw_pgid, &pgid); 2851 2852 if (ceph_can_shift_osds(pi)) { 2853 spgid->pgid = pgid; /* struct */ 2854 spgid->shard = CEPH_SPG_NOSHARD; 2855 return true; 2856 } 2857 2858 ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting); 2859 for (i = 0; i < acting.size; i++) { 2860 if (acting.osds[i] == acting.primary) { 2861 spgid->pgid = pgid; /* struct */ 2862 spgid->shard = i; 2863 return true; 2864 } 2865 } 2866 2867 return false; 2868} 2869 2870/* 2871 * Return acting primary for given PG, or -1 if none. 2872 */ 2873int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 2874 const struct ceph_pg *raw_pgid) 2875{ 2876 struct ceph_pg_pool_info *pi; 2877 struct ceph_osds up, acting; 2878 2879 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool); 2880 if (!pi) 2881 return -1; 2882 2883 ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting); 2884 return acting.primary; 2885} 2886EXPORT_SYMBOL(ceph_pg_to_acting_primary); 2887 2888static struct crush_loc_node *alloc_crush_loc(size_t type_name_len, 2889 size_t name_len) 2890{ 2891 struct crush_loc_node *loc; 2892 2893 loc = kmalloc(sizeof(*loc) + type_name_len + name_len + 2, GFP_NOIO); 2894 if (!loc) 2895 return NULL; 2896 2897 RB_CLEAR_NODE(&loc->cl_node); 2898 return loc; 2899} 2900 2901static void free_crush_loc(struct crush_loc_node *loc) 2902{ 2903 WARN_ON(!RB_EMPTY_NODE(&loc->cl_node)); 2904 2905 kfree(loc); 2906} 2907 2908static int crush_loc_compare(const struct crush_loc *loc1, 2909 const struct crush_loc *loc2) 2910{ 2911 return strcmp(loc1->cl_type_name, loc2->cl_type_name) ?: 2912 strcmp(loc1->cl_name, loc2->cl_name); 2913} 2914 2915DEFINE_RB_FUNCS2(crush_loc, struct crush_loc_node, cl_loc, crush_loc_compare, 2916 RB_BYPTR, const struct crush_loc *, cl_node) 2917 2918/* 2919 * Parses a set of <bucket type name>':'<bucket name> pairs separated 2920 * by '|', e.g. "rack:foo1|rack:foo2|datacenter:bar". 2921 * 2922 * Note that @crush_location is modified by strsep(). 2923 */ 2924int ceph_parse_crush_location(char *crush_location, struct rb_root *locs) 2925{ 2926 struct crush_loc_node *loc; 2927 const char *type_name, *name, *colon; 2928 size_t type_name_len, name_len; 2929 2930 dout("%s '%s'\n", __func__, crush_location); 2931 while ((type_name = strsep(&crush_location, "|"))) { 2932 colon = strchr(type_name, ':'); 2933 if (!colon) 2934 return -EINVAL; 2935 2936 type_name_len = colon - type_name; 2937 if (type_name_len == 0) 2938 return -EINVAL; 2939 2940 name = colon + 1; 2941 name_len = strlen(name); 2942 if (name_len == 0) 2943 return -EINVAL; 2944 2945 loc = alloc_crush_loc(type_name_len, name_len); 2946 if (!loc) 2947 return -ENOMEM; 2948 2949 loc->cl_loc.cl_type_name = loc->cl_data; 2950 memcpy(loc->cl_loc.cl_type_name, type_name, type_name_len); 2951 loc->cl_loc.cl_type_name[type_name_len] = '\0'; 2952 2953 loc->cl_loc.cl_name = loc->cl_data + type_name_len + 1; 2954 memcpy(loc->cl_loc.cl_name, name, name_len); 2955 loc->cl_loc.cl_name[name_len] = '\0'; 2956 2957 if (!__insert_crush_loc(locs, loc)) { 2958 free_crush_loc(loc); 2959 return -EEXIST; 2960 } 2961 2962 dout("%s type_name '%s' name '%s'\n", __func__, 2963 loc->cl_loc.cl_type_name, loc->cl_loc.cl_name); 2964 } 2965 2966 return 0; 2967} 2968 2969int ceph_compare_crush_locs(struct rb_root *locs1, struct rb_root *locs2) 2970{ 2971 struct rb_node *n1 = rb_first(locs1); 2972 struct rb_node *n2 = rb_first(locs2); 2973 int ret; 2974 2975 for ( ; n1 && n2; n1 = rb_next(n1), n2 = rb_next(n2)) { 2976 struct crush_loc_node *loc1 = 2977 rb_entry(n1, struct crush_loc_node, cl_node); 2978 struct crush_loc_node *loc2 = 2979 rb_entry(n2, struct crush_loc_node, cl_node); 2980 2981 ret = crush_loc_compare(&loc1->cl_loc, &loc2->cl_loc); 2982 if (ret) 2983 return ret; 2984 } 2985 2986 if (!n1 && n2) 2987 return -1; 2988 if (n1 && !n2) 2989 return 1; 2990 return 0; 2991} 2992 2993void ceph_clear_crush_locs(struct rb_root *locs) 2994{ 2995 while (!RB_EMPTY_ROOT(locs)) { 2996 struct crush_loc_node *loc = 2997 rb_entry(rb_first(locs), struct crush_loc_node, cl_node); 2998 2999 erase_crush_loc(locs, loc); 3000 free_crush_loc(loc); 3001 } 3002} 3003 3004/* 3005 * [a-zA-Z0-9-_.]+ 3006 */ 3007static bool is_valid_crush_name(const char *name) 3008{ 3009 do { 3010 if (!('a' <= *name && *name <= 'z') && 3011 !('A' <= *name && *name <= 'Z') && 3012 !('0' <= *name && *name <= '9') && 3013 *name != '-' && *name != '_' && *name != '.') 3014 return false; 3015 } while (*++name != '\0'); 3016 3017 return true; 3018} 3019 3020/* 3021 * Gets the parent of an item. Returns its id (<0 because the 3022 * parent is always a bucket), type id (>0 for the same reason, 3023 * via @parent_type_id) and location (via @parent_loc). If no 3024 * parent, returns 0. 3025 * 3026 * Does a linear search, as there are no parent pointers of any 3027 * kind. Note that the result is ambiguous for items that occur 3028 * multiple times in the map. 3029 */ 3030static int get_immediate_parent(struct crush_map *c, int id, 3031 u16 *parent_type_id, 3032 struct crush_loc *parent_loc) 3033{ 3034 struct crush_bucket *b; 3035 struct crush_name_node *type_cn, *cn; 3036 int i, j; 3037 3038 for (i = 0; i < c->max_buckets; i++) { 3039 b = c->buckets[i]; 3040 if (!b) 3041 continue; 3042 3043 /* ignore per-class shadow hierarchy */ 3044 cn = lookup_crush_name(&c->names, b->id); 3045 if (!cn || !is_valid_crush_name(cn->cn_name)) 3046 continue; 3047 3048 for (j = 0; j < b->size; j++) { 3049 if (b->items[j] != id) 3050 continue; 3051 3052 *parent_type_id = b->type; 3053 type_cn = lookup_crush_name(&c->type_names, b->type); 3054 parent_loc->cl_type_name = type_cn->cn_name; 3055 parent_loc->cl_name = cn->cn_name; 3056 return b->id; 3057 } 3058 } 3059 3060 return 0; /* no parent */ 3061} 3062 3063/* 3064 * Calculates the locality/distance from an item to a client 3065 * location expressed in terms of CRUSH hierarchy as a set of 3066 * (bucket type name, bucket name) pairs. Specifically, looks 3067 * for the lowest-valued bucket type for which the location of 3068 * @id matches one of the locations in @locs, so for standard 3069 * bucket types (host = 1, rack = 3, datacenter = 8, zone = 9) 3070 * a matching host is closer than a matching rack and a matching 3071 * data center is closer than a matching zone. 3072 * 3073 * Specifying multiple locations (a "multipath" location) such 3074 * as "rack=foo1 rack=foo2 datacenter=bar" is allowed -- @locs 3075 * is a multimap. The locality will be: 3076 * 3077 * - 3 for OSDs in racks foo1 and foo2 3078 * - 8 for OSDs in data center bar 3079 * - -1 for all other OSDs 3080 * 3081 * The lowest possible bucket type is 1, so the best locality 3082 * for an OSD is 1 (i.e. a matching host). Locality 0 would be 3083 * the OSD itself. 3084 */ 3085int ceph_get_crush_locality(struct ceph_osdmap *osdmap, int id, 3086 struct rb_root *locs) 3087{ 3088 struct crush_loc loc; 3089 u16 type_id; 3090 3091 /* 3092 * Instead of repeated get_immediate_parent() calls, 3093 * the location of @id could be obtained with a single 3094 * depth-first traversal. 3095 */ 3096 for (;;) { 3097 id = get_immediate_parent(osdmap->crush, id, &type_id, &loc); 3098 if (id >= 0) 3099 return -1; /* not local */ 3100 3101 if (lookup_crush_loc(locs, &loc)) 3102 return type_id; 3103 } 3104}
~/
Read-only file preview.