1 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/module.h> 5 #include <linux/slab.h> 6 #include <asm/div64.h> 7 8 #include <linux/ceph/libceph.h> 9 #include <linux/ceph/osdmap.h> 10 #include <linux/ceph/decode.h> 11 #include <linux/crush/hash.h> 12 #include <linux/crush/mapper.h> 13 14 char *ceph_osdmap_state_str(char *str, int len, int state) 15 { 16 if (!len) 17 return str; 18 19 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP)) 20 snprintf(str, len, "exists, up"); 21 else if (state & CEPH_OSD_EXISTS) 22 snprintf(str, len, "exists"); 23 else if (state & CEPH_OSD_UP) 24 snprintf(str, len, "up"); 25 else 26 snprintf(str, len, "doesn't exist"); 27 28 return str; 29 } 30 31 /* maps */ 32 33 static int calc_bits_of(unsigned int t) 34 { 35 int b = 0; 36 while (t) { 37 t = t >> 1; 38 b++; 39 } 40 return b; 41 } 42 43 /* 44 * the foo_mask is the smallest value 2^n-1 that is >= foo. 45 */ 46 static void calc_pg_masks(struct ceph_pg_pool_info *pi) 47 { 48 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1; 49 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1; 50 } 51 52 /* 53 * decode crush map 54 */ 55 static int crush_decode_uniform_bucket(void **p, void *end, 56 struct crush_bucket_uniform *b) 57 { 58 dout("crush_decode_uniform_bucket %p to %p\n", *p, end); 59 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad); 60 b->item_weight = ceph_decode_32(p); 61 return 0; 62 bad: 63 return -EINVAL; 64 } 65 66 static int crush_decode_list_bucket(void **p, void *end, 67 struct crush_bucket_list *b) 68 { 69 int j; 70 dout("crush_decode_list_bucket %p to %p\n", *p, end); 71 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 72 if (b->item_weights == NULL) 73 return -ENOMEM; 74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 75 if (b->sum_weights == NULL) 76 return -ENOMEM; 77 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); 78 for (j = 0; j < b->h.size; j++) { 79 b->item_weights[j] = ceph_decode_32(p); 80 b->sum_weights[j] = ceph_decode_32(p); 81 } 82 return 0; 83 bad: 84 return -EINVAL; 85 } 86 87 static int crush_decode_tree_bucket(void **p, void *end, 88 struct crush_bucket_tree *b) 89 { 90 int j; 91 dout("crush_decode_tree_bucket %p to %p\n", *p, end); 92 ceph_decode_8_safe(p, end, b->num_nodes, bad); 93 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS); 94 if (b->node_weights == NULL) 95 return -ENOMEM; 96 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad); 97 for (j = 0; j < b->num_nodes; j++) 98 b->node_weights[j] = ceph_decode_32(p); 99 return 0; 100 bad: 101 return -EINVAL; 102 } 103 104 static int crush_decode_straw_bucket(void **p, void *end, 105 struct crush_bucket_straw *b) 106 { 107 int j; 108 dout("crush_decode_straw_bucket %p to %p\n", *p, end); 109 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 110 if (b->item_weights == NULL) 111 return -ENOMEM; 112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 113 if (b->straws == NULL) 114 return -ENOMEM; 115 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); 116 for (j = 0; j < b->h.size; j++) { 117 b->item_weights[j] = ceph_decode_32(p); 118 b->straws[j] = ceph_decode_32(p); 119 } 120 return 0; 121 bad: 122 return -EINVAL; 123 } 124 125 static int crush_decode_straw2_bucket(void **p, void *end, 126 struct crush_bucket_straw2 *b) 127 { 128 int j; 129 dout("crush_decode_straw2_bucket %p to %p\n", *p, end); 130 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); 131 if (b->item_weights == NULL) 132 return -ENOMEM; 133 ceph_decode_need(p, end, b->h.size * sizeof(u32), bad); 134 for (j = 0; j < b->h.size; j++) 135 b->item_weights[j] = ceph_decode_32(p); 136 return 0; 137 bad: 138 return -EINVAL; 139 } 140 141 static int skip_name_map(void **p, void *end) 142 { 143 int len; 144 ceph_decode_32_safe(p, end, len ,bad); 145 while (len--) { 146 int strlen; 147 *p += sizeof(u32); 148 ceph_decode_32_safe(p, end, strlen, bad); 149 *p += strlen; 150 } 151 return 0; 152 bad: 153 return -EINVAL; 154 } 155 156 static struct crush_map *crush_decode(void *pbyval, void *end) 157 { 158 struct crush_map *c; 159 int err = -EINVAL; 160 int i, j; 161 void **p = &pbyval; 162 void *start = pbyval; 163 u32 magic; 164 u32 num_name_maps; 165 166 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p)); 167 168 c = kzalloc(sizeof(*c), GFP_NOFS); 169 if (c == NULL) 170 return ERR_PTR(-ENOMEM); 171 172 /* set tunables to default values */ 173 c->choose_local_tries = 2; 174 c->choose_local_fallback_tries = 5; 175 c->choose_total_tries = 19; 176 c->chooseleaf_descend_once = 0; 177 178 ceph_decode_need(p, end, 4*sizeof(u32), bad); 179 magic = ceph_decode_32(p); 180 if (magic != CRUSH_MAGIC) { 181 pr_err("crush_decode magic %x != current %x\n", 182 (unsigned int)magic, (unsigned int)CRUSH_MAGIC); 183 goto bad; 184 } 185 c->max_buckets = ceph_decode_32(p); 186 c->max_rules = ceph_decode_32(p); 187 c->max_devices = ceph_decode_32(p); 188 189 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS); 190 if (c->buckets == NULL) 191 goto badmem; 192 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS); 193 if (c->rules == NULL) 194 goto badmem; 195 196 /* buckets */ 197 for (i = 0; i < c->max_buckets; i++) { 198 int size = 0; 199 u32 alg; 200 struct crush_bucket *b; 201 202 ceph_decode_32_safe(p, end, alg, bad); 203 if (alg == 0) { 204 c->buckets[i] = NULL; 205 continue; 206 } 207 dout("crush_decode bucket %d off %x %p to %p\n", 208 i, (int)(*p-start), *p, end); 209 210 switch (alg) { 211 case CRUSH_BUCKET_UNIFORM: 212 size = sizeof(struct crush_bucket_uniform); 213 break; 214 case CRUSH_BUCKET_LIST: 215 size = sizeof(struct crush_bucket_list); 216 break; 217 case CRUSH_BUCKET_TREE: 218 size = sizeof(struct crush_bucket_tree); 219 break; 220 case CRUSH_BUCKET_STRAW: 221 size = sizeof(struct crush_bucket_straw); 222 break; 223 case CRUSH_BUCKET_STRAW2: 224 size = sizeof(struct crush_bucket_straw2); 225 break; 226 default: 227 err = -EINVAL; 228 goto bad; 229 } 230 BUG_ON(size == 0); 231 b = c->buckets[i] = kzalloc(size, GFP_NOFS); 232 if (b == NULL) 233 goto badmem; 234 235 ceph_decode_need(p, end, 4*sizeof(u32), bad); 236 b->id = ceph_decode_32(p); 237 b->type = ceph_decode_16(p); 238 b->alg = ceph_decode_8(p); 239 b->hash = ceph_decode_8(p); 240 b->weight = ceph_decode_32(p); 241 b->size = ceph_decode_32(p); 242 243 dout("crush_decode bucket size %d off %x %p to %p\n", 244 b->size, (int)(*p-start), *p, end); 245 246 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS); 247 if (b->items == NULL) 248 goto badmem; 249 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS); 250 if (b->perm == NULL) 251 goto badmem; 252 b->perm_n = 0; 253 254 ceph_decode_need(p, end, b->size*sizeof(u32), bad); 255 for (j = 0; j < b->size; j++) 256 b->items[j] = ceph_decode_32(p); 257 258 switch (b->alg) { 259 case CRUSH_BUCKET_UNIFORM: 260 err = crush_decode_uniform_bucket(p, end, 261 (struct crush_bucket_uniform *)b); 262 if (err < 0) 263 goto bad; 264 break; 265 case CRUSH_BUCKET_LIST: 266 err = crush_decode_list_bucket(p, end, 267 (struct crush_bucket_list *)b); 268 if (err < 0) 269 goto bad; 270 break; 271 case CRUSH_BUCKET_TREE: 272 err = crush_decode_tree_bucket(p, end, 273 (struct crush_bucket_tree *)b); 274 if (err < 0) 275 goto bad; 276 break; 277 case CRUSH_BUCKET_STRAW: 278 err = crush_decode_straw_bucket(p, end, 279 (struct crush_bucket_straw *)b); 280 if (err < 0) 281 goto bad; 282 break; 283 case CRUSH_BUCKET_STRAW2: 284 err = crush_decode_straw2_bucket(p, end, 285 (struct crush_bucket_straw2 *)b); 286 if (err < 0) 287 goto bad; 288 break; 289 } 290 } 291 292 /* rules */ 293 dout("rule vec is %p\n", c->rules); 294 for (i = 0; i < c->max_rules; i++) { 295 u32 yes; 296 struct crush_rule *r; 297 298 ceph_decode_32_safe(p, end, yes, bad); 299 if (!yes) { 300 dout("crush_decode NO rule %d off %x %p to %p\n", 301 i, (int)(*p-start), *p, end); 302 c->rules[i] = NULL; 303 continue; 304 } 305 306 dout("crush_decode rule %d off %x %p to %p\n", 307 i, (int)(*p-start), *p, end); 308 309 /* len */ 310 ceph_decode_32_safe(p, end, yes, bad); 311 #if BITS_PER_LONG == 32 312 err = -EINVAL; 313 if (yes > (ULONG_MAX - sizeof(*r)) 314 / sizeof(struct crush_rule_step)) 315 goto bad; 316 #endif 317 r = c->rules[i] = kmalloc(sizeof(*r) + 318 yes*sizeof(struct crush_rule_step), 319 GFP_NOFS); 320 if (r == NULL) 321 goto badmem; 322 dout(" rule %d is at %p\n", i, r); 323 r->len = yes; 324 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */ 325 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad); 326 for (j = 0; j < r->len; j++) { 327 r->steps[j].op = ceph_decode_32(p); 328 r->steps[j].arg1 = ceph_decode_32(p); 329 r->steps[j].arg2 = ceph_decode_32(p); 330 } 331 } 332 333 /* ignore trailing name maps. */ 334 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) { 335 err = skip_name_map(p, end); 336 if (err < 0) 337 goto done; 338 } 339 340 /* tunables */ 341 ceph_decode_need(p, end, 3*sizeof(u32), done); 342 c->choose_local_tries = ceph_decode_32(p); 343 c->choose_local_fallback_tries = ceph_decode_32(p); 344 c->choose_total_tries = ceph_decode_32(p); 345 dout("crush decode tunable choose_local_tries = %d\n", 346 c->choose_local_tries); 347 dout("crush decode tunable choose_local_fallback_tries = %d\n", 348 c->choose_local_fallback_tries); 349 dout("crush decode tunable choose_total_tries = %d\n", 350 c->choose_total_tries); 351 352 ceph_decode_need(p, end, sizeof(u32), done); 353 c->chooseleaf_descend_once = ceph_decode_32(p); 354 dout("crush decode tunable chooseleaf_descend_once = %d\n", 355 c->chooseleaf_descend_once); 356 357 ceph_decode_need(p, end, sizeof(u8), done); 358 c->chooseleaf_vary_r = ceph_decode_8(p); 359 dout("crush decode tunable chooseleaf_vary_r = %d\n", 360 c->chooseleaf_vary_r); 361 362 /* skip straw_calc_version, allowed_bucket_algs */ 363 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done); 364 *p += sizeof(u8) + sizeof(u32); 365 366 ceph_decode_need(p, end, sizeof(u8), done); 367 c->chooseleaf_stable = ceph_decode_8(p); 368 dout("crush decode tunable chooseleaf_stable = %d\n", 369 c->chooseleaf_stable); 370 371 done: 372 dout("crush_decode success\n"); 373 return c; 374 375 badmem: 376 err = -ENOMEM; 377 bad: 378 dout("crush_decode fail %d\n", err); 379 crush_destroy(c); 380 return ERR_PTR(err); 381 } 382 383 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs) 384 { 385 if (lhs->pool < rhs->pool) 386 return -1; 387 if (lhs->pool > rhs->pool) 388 return 1; 389 if (lhs->seed < rhs->seed) 390 return -1; 391 if (lhs->seed > rhs->seed) 392 return 1; 393 394 return 0; 395 } 396 397 /* 398 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid 399 * to a set of osds) and primary_temp (explicit primary setting) 400 */ 401 static int __insert_pg_mapping(struct ceph_pg_mapping *new, 402 struct rb_root *root) 403 { 404 struct rb_node **p = &root->rb_node; 405 struct rb_node *parent = NULL; 406 struct ceph_pg_mapping *pg = NULL; 407 int c; 408 409 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new); 410 while (*p) { 411 parent = *p; 412 pg = rb_entry(parent, struct ceph_pg_mapping, node); 413 c = ceph_pg_compare(&new->pgid, &pg->pgid); 414 if (c < 0) 415 p = &(*p)->rb_left; 416 else if (c > 0) 417 p = &(*p)->rb_right; 418 else 419 return -EEXIST; 420 } 421 422 rb_link_node(&new->node, parent, p); 423 rb_insert_color(&new->node, root); 424 return 0; 425 } 426 427 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root, 428 struct ceph_pg pgid) 429 { 430 struct rb_node *n = root->rb_node; 431 struct ceph_pg_mapping *pg; 432 int c; 433 434 while (n) { 435 pg = rb_entry(n, struct ceph_pg_mapping, node); 436 c = ceph_pg_compare(&pgid, &pg->pgid); 437 if (c < 0) { 438 n = n->rb_left; 439 } else if (c > 0) { 440 n = n->rb_right; 441 } else { 442 dout("__lookup_pg_mapping %lld.%x got %p\n", 443 pgid.pool, pgid.seed, pg); 444 return pg; 445 } 446 } 447 return NULL; 448 } 449 450 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid) 451 { 452 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid); 453 454 if (pg) { 455 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed, 456 pg); 457 rb_erase(&pg->node, root); 458 kfree(pg); 459 return 0; 460 } 461 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed); 462 return -ENOENT; 463 } 464 465 /* 466 * rbtree of pg pool info 467 */ 468 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new) 469 { 470 struct rb_node **p = &root->rb_node; 471 struct rb_node *parent = NULL; 472 struct ceph_pg_pool_info *pi = NULL; 473 474 while (*p) { 475 parent = *p; 476 pi = rb_entry(parent, struct ceph_pg_pool_info, node); 477 if (new->id < pi->id) 478 p = &(*p)->rb_left; 479 else if (new->id > pi->id) 480 p = &(*p)->rb_right; 481 else 482 return -EEXIST; 483 } 484 485 rb_link_node(&new->node, parent, p); 486 rb_insert_color(&new->node, root); 487 return 0; 488 } 489 490 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id) 491 { 492 struct ceph_pg_pool_info *pi; 493 struct rb_node *n = root->rb_node; 494 495 while (n) { 496 pi = rb_entry(n, struct ceph_pg_pool_info, node); 497 if (id < pi->id) 498 n = n->rb_left; 499 else if (id > pi->id) 500 n = n->rb_right; 501 else 502 return pi; 503 } 504 return NULL; 505 } 506 507 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id) 508 { 509 return __lookup_pg_pool(&map->pg_pools, id); 510 } 511 512 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id) 513 { 514 struct ceph_pg_pool_info *pi; 515 516 if (id == CEPH_NOPOOL) 517 return NULL; 518 519 if (WARN_ON_ONCE(id > (u64) INT_MAX)) 520 return NULL; 521 522 pi = __lookup_pg_pool(&map->pg_pools, (int) id); 523 524 return pi ? pi->name : NULL; 525 } 526 EXPORT_SYMBOL(ceph_pg_pool_name_by_id); 527 528 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name) 529 { 530 struct rb_node *rbp; 531 532 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) { 533 struct ceph_pg_pool_info *pi = 534 rb_entry(rbp, struct ceph_pg_pool_info, node); 535 if (pi->name && strcmp(pi->name, name) == 0) 536 return pi->id; 537 } 538 return -ENOENT; 539 } 540 EXPORT_SYMBOL(ceph_pg_poolid_by_name); 541 542 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi) 543 { 544 rb_erase(&pi->node, root); 545 kfree(pi->name); 546 kfree(pi); 547 } 548 549 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi) 550 { 551 u8 ev, cv; 552 unsigned len, num; 553 void *pool_end; 554 555 ceph_decode_need(p, end, 2 + 4, bad); 556 ev = ceph_decode_8(p); /* encoding version */ 557 cv = ceph_decode_8(p); /* compat version */ 558 if (ev < 5) { 559 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv); 560 return -EINVAL; 561 } 562 if (cv > 9) { 563 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv); 564 return -EINVAL; 565 } 566 len = ceph_decode_32(p); 567 ceph_decode_need(p, end, len, bad); 568 pool_end = *p + len; 569 570 pi->type = ceph_decode_8(p); 571 pi->size = ceph_decode_8(p); 572 pi->crush_ruleset = ceph_decode_8(p); 573 pi->object_hash = ceph_decode_8(p); 574 575 pi->pg_num = ceph_decode_32(p); 576 pi->pgp_num = ceph_decode_32(p); 577 578 *p += 4 + 4; /* skip lpg* */ 579 *p += 4; /* skip last_change */ 580 *p += 8 + 4; /* skip snap_seq, snap_epoch */ 581 582 /* skip snaps */ 583 num = ceph_decode_32(p); 584 while (num--) { 585 *p += 8; /* snapid key */ 586 *p += 1 + 1; /* versions */ 587 len = ceph_decode_32(p); 588 *p += len; 589 } 590 591 /* skip removed_snaps */ 592 num = ceph_decode_32(p); 593 *p += num * (8 + 8); 594 595 *p += 8; /* skip auid */ 596 pi->flags = ceph_decode_64(p); 597 *p += 4; /* skip crash_replay_interval */ 598 599 if (ev >= 7) 600 pi->min_size = ceph_decode_8(p); 601 else 602 pi->min_size = pi->size - pi->size / 2; 603 604 if (ev >= 8) 605 *p += 8 + 8; /* skip quota_max_* */ 606 607 if (ev >= 9) { 608 /* skip tiers */ 609 num = ceph_decode_32(p); 610 *p += num * 8; 611 612 *p += 8; /* skip tier_of */ 613 *p += 1; /* skip cache_mode */ 614 615 pi->read_tier = ceph_decode_64(p); 616 pi->write_tier = ceph_decode_64(p); 617 } else { 618 pi->read_tier = -1; 619 pi->write_tier = -1; 620 } 621 622 if (ev >= 10) { 623 /* skip properties */ 624 num = ceph_decode_32(p); 625 while (num--) { 626 len = ceph_decode_32(p); 627 *p += len; /* key */ 628 len = ceph_decode_32(p); 629 *p += len; /* val */ 630 } 631 } 632 633 if (ev >= 11) { 634 /* skip hit_set_params */ 635 *p += 1 + 1; /* versions */ 636 len = ceph_decode_32(p); 637 *p += len; 638 639 *p += 4; /* skip hit_set_period */ 640 *p += 4; /* skip hit_set_count */ 641 } 642 643 if (ev >= 12) 644 *p += 4; /* skip stripe_width */ 645 646 if (ev >= 13) { 647 *p += 8; /* skip target_max_bytes */ 648 *p += 8; /* skip target_max_objects */ 649 *p += 4; /* skip cache_target_dirty_ratio_micro */ 650 *p += 4; /* skip cache_target_full_ratio_micro */ 651 *p += 4; /* skip cache_min_flush_age */ 652 *p += 4; /* skip cache_min_evict_age */ 653 } 654 655 if (ev >= 14) { 656 /* skip erasure_code_profile */ 657 len = ceph_decode_32(p); 658 *p += len; 659 } 660 661 if (ev >= 15) 662 pi->last_force_request_resend = ceph_decode_32(p); 663 else 664 pi->last_force_request_resend = 0; 665 666 /* ignore the rest */ 667 668 *p = pool_end; 669 calc_pg_masks(pi); 670 return 0; 671 672 bad: 673 return -EINVAL; 674 } 675 676 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map) 677 { 678 struct ceph_pg_pool_info *pi; 679 u32 num, len; 680 u64 pool; 681 682 ceph_decode_32_safe(p, end, num, bad); 683 dout(" %d pool names\n", num); 684 while (num--) { 685 ceph_decode_64_safe(p, end, pool, bad); 686 ceph_decode_32_safe(p, end, len, bad); 687 dout(" pool %llu len %d\n", pool, len); 688 ceph_decode_need(p, end, len, bad); 689 pi = __lookup_pg_pool(&map->pg_pools, pool); 690 if (pi) { 691 char *name = kstrndup(*p, len, GFP_NOFS); 692 693 if (!name) 694 return -ENOMEM; 695 kfree(pi->name); 696 pi->name = name; 697 dout(" name is %s\n", pi->name); 698 } 699 *p += len; 700 } 701 return 0; 702 703 bad: 704 return -EINVAL; 705 } 706 707 /* 708 * osd map 709 */ 710 void ceph_osdmap_destroy(struct ceph_osdmap *map) 711 { 712 dout("osdmap_destroy %p\n", map); 713 if (map->crush) 714 crush_destroy(map->crush); 715 while (!RB_EMPTY_ROOT(&map->pg_temp)) { 716 struct ceph_pg_mapping *pg = 717 rb_entry(rb_first(&map->pg_temp), 718 struct ceph_pg_mapping, node); 719 rb_erase(&pg->node, &map->pg_temp); 720 kfree(pg); 721 } 722 while (!RB_EMPTY_ROOT(&map->primary_temp)) { 723 struct ceph_pg_mapping *pg = 724 rb_entry(rb_first(&map->primary_temp), 725 struct ceph_pg_mapping, node); 726 rb_erase(&pg->node, &map->primary_temp); 727 kfree(pg); 728 } 729 while (!RB_EMPTY_ROOT(&map->pg_pools)) { 730 struct ceph_pg_pool_info *pi = 731 rb_entry(rb_first(&map->pg_pools), 732 struct ceph_pg_pool_info, node); 733 __remove_pg_pool(&map->pg_pools, pi); 734 } 735 kfree(map->osd_state); 736 kfree(map->osd_weight); 737 kfree(map->osd_addr); 738 kfree(map->osd_primary_affinity); 739 kfree(map); 740 } 741 742 /* 743 * Adjust max_osd value, (re)allocate arrays. 744 * 745 * The new elements are properly initialized. 746 */ 747 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max) 748 { 749 u8 *state; 750 u32 *weight; 751 struct ceph_entity_addr *addr; 752 int i; 753 754 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS); 755 if (!state) 756 return -ENOMEM; 757 map->osd_state = state; 758 759 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS); 760 if (!weight) 761 return -ENOMEM; 762 map->osd_weight = weight; 763 764 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS); 765 if (!addr) 766 return -ENOMEM; 767 map->osd_addr = addr; 768 769 for (i = map->max_osd; i < max; i++) { 770 map->osd_state[i] = 0; 771 map->osd_weight[i] = CEPH_OSD_OUT; 772 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr)); 773 } 774 775 if (map->osd_primary_affinity) { 776 u32 *affinity; 777 778 affinity = krealloc(map->osd_primary_affinity, 779 max*sizeof(*affinity), GFP_NOFS); 780 if (!affinity) 781 return -ENOMEM; 782 map->osd_primary_affinity = affinity; 783 784 for (i = map->max_osd; i < max; i++) 785 map->osd_primary_affinity[i] = 786 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 787 } 788 789 map->max_osd = max; 790 791 return 0; 792 } 793 794 #define OSDMAP_WRAPPER_COMPAT_VER 7 795 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1 796 797 /* 798 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps, 799 * to struct_v of the client_data section for new (v7 and above) 800 * osdmaps. 801 */ 802 static int get_osdmap_client_data_v(void **p, void *end, 803 const char *prefix, u8 *v) 804 { 805 u8 struct_v; 806 807 ceph_decode_8_safe(p, end, struct_v, e_inval); 808 if (struct_v >= 7) { 809 u8 struct_compat; 810 811 ceph_decode_8_safe(p, end, struct_compat, e_inval); 812 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) { 813 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n", 814 struct_v, struct_compat, 815 OSDMAP_WRAPPER_COMPAT_VER, prefix); 816 return -EINVAL; 817 } 818 *p += 4; /* ignore wrapper struct_len */ 819 820 ceph_decode_8_safe(p, end, struct_v, e_inval); 821 ceph_decode_8_safe(p, end, struct_compat, e_inval); 822 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) { 823 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n", 824 struct_v, struct_compat, 825 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix); 826 return -EINVAL; 827 } 828 *p += 4; /* ignore client data struct_len */ 829 } else { 830 u16 version; 831 832 *p -= 1; 833 ceph_decode_16_safe(p, end, version, e_inval); 834 if (version < 6) { 835 pr_warn("got v %d < 6 of %s ceph_osdmap\n", 836 version, prefix); 837 return -EINVAL; 838 } 839 840 /* old osdmap enconding */ 841 struct_v = 0; 842 } 843 844 *v = struct_v; 845 return 0; 846 847 e_inval: 848 return -EINVAL; 849 } 850 851 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map, 852 bool incremental) 853 { 854 u32 n; 855 856 ceph_decode_32_safe(p, end, n, e_inval); 857 while (n--) { 858 struct ceph_pg_pool_info *pi; 859 u64 pool; 860 int ret; 861 862 ceph_decode_64_safe(p, end, pool, e_inval); 863 864 pi = __lookup_pg_pool(&map->pg_pools, pool); 865 if (!incremental || !pi) { 866 pi = kzalloc(sizeof(*pi), GFP_NOFS); 867 if (!pi) 868 return -ENOMEM; 869 870 pi->id = pool; 871 872 ret = __insert_pg_pool(&map->pg_pools, pi); 873 if (ret) { 874 kfree(pi); 875 return ret; 876 } 877 } 878 879 ret = decode_pool(p, end, pi); 880 if (ret) 881 return ret; 882 } 883 884 return 0; 885 886 e_inval: 887 return -EINVAL; 888 } 889 890 static int decode_pools(void **p, void *end, struct ceph_osdmap *map) 891 { 892 return __decode_pools(p, end, map, false); 893 } 894 895 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map) 896 { 897 return __decode_pools(p, end, map, true); 898 } 899 900 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map, 901 bool incremental) 902 { 903 u32 n; 904 905 ceph_decode_32_safe(p, end, n, e_inval); 906 while (n--) { 907 struct ceph_pg pgid; 908 u32 len, i; 909 int ret; 910 911 ret = ceph_decode_pgid(p, end, &pgid); 912 if (ret) 913 return ret; 914 915 ceph_decode_32_safe(p, end, len, e_inval); 916 917 ret = __remove_pg_mapping(&map->pg_temp, pgid); 918 BUG_ON(!incremental && ret != -ENOENT); 919 920 if (!incremental || len > 0) { 921 struct ceph_pg_mapping *pg; 922 923 ceph_decode_need(p, end, len*sizeof(u32), e_inval); 924 925 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32)) 926 return -EINVAL; 927 928 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS); 929 if (!pg) 930 return -ENOMEM; 931 932 pg->pgid = pgid; 933 pg->pg_temp.len = len; 934 for (i = 0; i < len; i++) 935 pg->pg_temp.osds[i] = ceph_decode_32(p); 936 937 ret = __insert_pg_mapping(pg, &map->pg_temp); 938 if (ret) { 939 kfree(pg); 940 return ret; 941 } 942 } 943 } 944 945 return 0; 946 947 e_inval: 948 return -EINVAL; 949 } 950 951 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map) 952 { 953 return __decode_pg_temp(p, end, map, false); 954 } 955 956 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map) 957 { 958 return __decode_pg_temp(p, end, map, true); 959 } 960 961 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map, 962 bool incremental) 963 { 964 u32 n; 965 966 ceph_decode_32_safe(p, end, n, e_inval); 967 while (n--) { 968 struct ceph_pg pgid; 969 u32 osd; 970 int ret; 971 972 ret = ceph_decode_pgid(p, end, &pgid); 973 if (ret) 974 return ret; 975 976 ceph_decode_32_safe(p, end, osd, e_inval); 977 978 ret = __remove_pg_mapping(&map->primary_temp, pgid); 979 BUG_ON(!incremental && ret != -ENOENT); 980 981 if (!incremental || osd != (u32)-1) { 982 struct ceph_pg_mapping *pg; 983 984 pg = kzalloc(sizeof(*pg), GFP_NOFS); 985 if (!pg) 986 return -ENOMEM; 987 988 pg->pgid = pgid; 989 pg->primary_temp.osd = osd; 990 991 ret = __insert_pg_mapping(pg, &map->primary_temp); 992 if (ret) { 993 kfree(pg); 994 return ret; 995 } 996 } 997 } 998 999 return 0; 1000 1001 e_inval: 1002 return -EINVAL; 1003 } 1004 1005 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map) 1006 { 1007 return __decode_primary_temp(p, end, map, false); 1008 } 1009 1010 static int decode_new_primary_temp(void **p, void *end, 1011 struct ceph_osdmap *map) 1012 { 1013 return __decode_primary_temp(p, end, map, true); 1014 } 1015 1016 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd) 1017 { 1018 BUG_ON(osd >= map->max_osd); 1019 1020 if (!map->osd_primary_affinity) 1021 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1022 1023 return map->osd_primary_affinity[osd]; 1024 } 1025 1026 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff) 1027 { 1028 BUG_ON(osd >= map->max_osd); 1029 1030 if (!map->osd_primary_affinity) { 1031 int i; 1032 1033 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32), 1034 GFP_NOFS); 1035 if (!map->osd_primary_affinity) 1036 return -ENOMEM; 1037 1038 for (i = 0; i < map->max_osd; i++) 1039 map->osd_primary_affinity[i] = 1040 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 1041 } 1042 1043 map->osd_primary_affinity[osd] = aff; 1044 1045 return 0; 1046 } 1047 1048 static int decode_primary_affinity(void **p, void *end, 1049 struct ceph_osdmap *map) 1050 { 1051 u32 len, i; 1052 1053 ceph_decode_32_safe(p, end, len, e_inval); 1054 if (len == 0) { 1055 kfree(map->osd_primary_affinity); 1056 map->osd_primary_affinity = NULL; 1057 return 0; 1058 } 1059 if (len != map->max_osd) 1060 goto e_inval; 1061 1062 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval); 1063 1064 for (i = 0; i < map->max_osd; i++) { 1065 int ret; 1066 1067 ret = set_primary_affinity(map, i, ceph_decode_32(p)); 1068 if (ret) 1069 return ret; 1070 } 1071 1072 return 0; 1073 1074 e_inval: 1075 return -EINVAL; 1076 } 1077 1078 static int decode_new_primary_affinity(void **p, void *end, 1079 struct ceph_osdmap *map) 1080 { 1081 u32 n; 1082 1083 ceph_decode_32_safe(p, end, n, e_inval); 1084 while (n--) { 1085 u32 osd, aff; 1086 int ret; 1087 1088 ceph_decode_32_safe(p, end, osd, e_inval); 1089 ceph_decode_32_safe(p, end, aff, e_inval); 1090 1091 ret = set_primary_affinity(map, osd, aff); 1092 if (ret) 1093 return ret; 1094 1095 pr_info("osd%d primary-affinity 0x%x\n", osd, aff); 1096 } 1097 1098 return 0; 1099 1100 e_inval: 1101 return -EINVAL; 1102 } 1103 1104 /* 1105 * decode a full map. 1106 */ 1107 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map) 1108 { 1109 u8 struct_v; 1110 u32 epoch = 0; 1111 void *start = *p; 1112 u32 max; 1113 u32 len, i; 1114 int err; 1115 1116 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1117 1118 err = get_osdmap_client_data_v(p, end, "full", &struct_v); 1119 if (err) 1120 goto bad; 1121 1122 /* fsid, epoch, created, modified */ 1123 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) + 1124 sizeof(map->created) + sizeof(map->modified), e_inval); 1125 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid)); 1126 epoch = map->epoch = ceph_decode_32(p); 1127 ceph_decode_copy(p, &map->created, sizeof(map->created)); 1128 ceph_decode_copy(p, &map->modified, sizeof(map->modified)); 1129 1130 /* pools */ 1131 err = decode_pools(p, end, map); 1132 if (err) 1133 goto bad; 1134 1135 /* pool_name */ 1136 err = decode_pool_names(p, end, map); 1137 if (err) 1138 goto bad; 1139 1140 ceph_decode_32_safe(p, end, map->pool_max, e_inval); 1141 1142 ceph_decode_32_safe(p, end, map->flags, e_inval); 1143 1144 /* max_osd */ 1145 ceph_decode_32_safe(p, end, max, e_inval); 1146 1147 /* (re)alloc osd arrays */ 1148 err = osdmap_set_max_osd(map, max); 1149 if (err) 1150 goto bad; 1151 1152 /* osd_state, osd_weight, osd_addrs->client_addr */ 1153 ceph_decode_need(p, end, 3*sizeof(u32) + 1154 map->max_osd*(1 + sizeof(*map->osd_weight) + 1155 sizeof(*map->osd_addr)), e_inval); 1156 1157 if (ceph_decode_32(p) != map->max_osd) 1158 goto e_inval; 1159 1160 ceph_decode_copy(p, map->osd_state, map->max_osd); 1161 1162 if (ceph_decode_32(p) != map->max_osd) 1163 goto e_inval; 1164 1165 for (i = 0; i < map->max_osd; i++) 1166 map->osd_weight[i] = ceph_decode_32(p); 1167 1168 if (ceph_decode_32(p) != map->max_osd) 1169 goto e_inval; 1170 1171 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr)); 1172 for (i = 0; i < map->max_osd; i++) 1173 ceph_decode_addr(&map->osd_addr[i]); 1174 1175 /* pg_temp */ 1176 err = decode_pg_temp(p, end, map); 1177 if (err) 1178 goto bad; 1179 1180 /* primary_temp */ 1181 if (struct_v >= 1) { 1182 err = decode_primary_temp(p, end, map); 1183 if (err) 1184 goto bad; 1185 } 1186 1187 /* primary_affinity */ 1188 if (struct_v >= 2) { 1189 err = decode_primary_affinity(p, end, map); 1190 if (err) 1191 goto bad; 1192 } else { 1193 /* XXX can this happen? */ 1194 kfree(map->osd_primary_affinity); 1195 map->osd_primary_affinity = NULL; 1196 } 1197 1198 /* crush */ 1199 ceph_decode_32_safe(p, end, len, e_inval); 1200 map->crush = crush_decode(*p, min(*p + len, end)); 1201 if (IS_ERR(map->crush)) { 1202 err = PTR_ERR(map->crush); 1203 map->crush = NULL; 1204 goto bad; 1205 } 1206 *p += len; 1207 1208 /* ignore the rest */ 1209 *p = end; 1210 1211 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1212 return 0; 1213 1214 e_inval: 1215 err = -EINVAL; 1216 bad: 1217 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1218 err, epoch, (int)(*p - start), *p, start, end); 1219 print_hex_dump(KERN_DEBUG, "osdmap: ", 1220 DUMP_PREFIX_OFFSET, 16, 1, 1221 start, end - start, true); 1222 return err; 1223 } 1224 1225 /* 1226 * Allocate and decode a full map. 1227 */ 1228 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end) 1229 { 1230 struct ceph_osdmap *map; 1231 int ret; 1232 1233 map = kzalloc(sizeof(*map), GFP_NOFS); 1234 if (!map) 1235 return ERR_PTR(-ENOMEM); 1236 1237 map->pg_temp = RB_ROOT; 1238 map->primary_temp = RB_ROOT; 1239 mutex_init(&map->crush_scratch_mutex); 1240 1241 ret = osdmap_decode(p, end, map); 1242 if (ret) { 1243 ceph_osdmap_destroy(map); 1244 return ERR_PTR(ret); 1245 } 1246 1247 return map; 1248 } 1249 1250 /* 1251 * decode and apply an incremental map update. 1252 */ 1253 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 1254 struct ceph_osdmap *map) 1255 { 1256 struct crush_map *newcrush = NULL; 1257 struct ceph_fsid fsid; 1258 u32 epoch = 0; 1259 struct ceph_timespec modified; 1260 s32 len; 1261 u64 pool; 1262 __s64 new_pool_max; 1263 __s32 new_flags, max; 1264 void *start = *p; 1265 int err; 1266 u8 struct_v; 1267 1268 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1269 1270 err = get_osdmap_client_data_v(p, end, "inc", &struct_v); 1271 if (err) 1272 goto bad; 1273 1274 /* fsid, epoch, modified, new_pool_max, new_flags */ 1275 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) + 1276 sizeof(u64) + sizeof(u32), e_inval); 1277 ceph_decode_copy(p, &fsid, sizeof(fsid)); 1278 epoch = ceph_decode_32(p); 1279 BUG_ON(epoch != map->epoch+1); 1280 ceph_decode_copy(p, &modified, sizeof(modified)); 1281 new_pool_max = ceph_decode_64(p); 1282 new_flags = ceph_decode_32(p); 1283 1284 /* full map? */ 1285 ceph_decode_32_safe(p, end, len, e_inval); 1286 if (len > 0) { 1287 dout("apply_incremental full map len %d, %p to %p\n", 1288 len, *p, end); 1289 return ceph_osdmap_decode(p, min(*p+len, end)); 1290 } 1291 1292 /* new crush? */ 1293 ceph_decode_32_safe(p, end, len, e_inval); 1294 if (len > 0) { 1295 newcrush = crush_decode(*p, min(*p+len, end)); 1296 if (IS_ERR(newcrush)) { 1297 err = PTR_ERR(newcrush); 1298 newcrush = NULL; 1299 goto bad; 1300 } 1301 *p += len; 1302 } 1303 1304 /* new flags? */ 1305 if (new_flags >= 0) 1306 map->flags = new_flags; 1307 if (new_pool_max >= 0) 1308 map->pool_max = new_pool_max; 1309 1310 /* new max? */ 1311 ceph_decode_32_safe(p, end, max, e_inval); 1312 if (max >= 0) { 1313 err = osdmap_set_max_osd(map, max); 1314 if (err) 1315 goto bad; 1316 } 1317 1318 map->epoch++; 1319 map->modified = modified; 1320 if (newcrush) { 1321 if (map->crush) 1322 crush_destroy(map->crush); 1323 map->crush = newcrush; 1324 newcrush = NULL; 1325 } 1326 1327 /* new_pools */ 1328 err = decode_new_pools(p, end, map); 1329 if (err) 1330 goto bad; 1331 1332 /* new_pool_names */ 1333 err = decode_pool_names(p, end, map); 1334 if (err) 1335 goto bad; 1336 1337 /* old_pool */ 1338 ceph_decode_32_safe(p, end, len, e_inval); 1339 while (len--) { 1340 struct ceph_pg_pool_info *pi; 1341 1342 ceph_decode_64_safe(p, end, pool, e_inval); 1343 pi = __lookup_pg_pool(&map->pg_pools, pool); 1344 if (pi) 1345 __remove_pg_pool(&map->pg_pools, pi); 1346 } 1347 1348 /* new_up */ 1349 ceph_decode_32_safe(p, end, len, e_inval); 1350 while (len--) { 1351 u32 osd; 1352 struct ceph_entity_addr addr; 1353 ceph_decode_32_safe(p, end, osd, e_inval); 1354 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval); 1355 ceph_decode_addr(&addr); 1356 pr_info("osd%d up\n", osd); 1357 BUG_ON(osd >= map->max_osd); 1358 map->osd_state[osd] |= CEPH_OSD_UP | CEPH_OSD_EXISTS; 1359 map->osd_addr[osd] = addr; 1360 } 1361 1362 /* new_state */ 1363 ceph_decode_32_safe(p, end, len, e_inval); 1364 while (len--) { 1365 u32 osd; 1366 u8 xorstate; 1367 ceph_decode_32_safe(p, end, osd, e_inval); 1368 xorstate = **(u8 **)p; 1369 (*p)++; /* clean flag */ 1370 if (xorstate == 0) 1371 xorstate = CEPH_OSD_UP; 1372 if (xorstate & CEPH_OSD_UP) 1373 pr_info("osd%d down\n", osd); 1374 if (osd < map->max_osd) 1375 map->osd_state[osd] ^= xorstate; 1376 } 1377 1378 /* new_weight */ 1379 ceph_decode_32_safe(p, end, len, e_inval); 1380 while (len--) { 1381 u32 osd, off; 1382 ceph_decode_need(p, end, sizeof(u32)*2, e_inval); 1383 osd = ceph_decode_32(p); 1384 off = ceph_decode_32(p); 1385 pr_info("osd%d weight 0x%x %s\n", osd, off, 1386 off == CEPH_OSD_IN ? "(in)" : 1387 (off == CEPH_OSD_OUT ? "(out)" : "")); 1388 if (osd < map->max_osd) 1389 map->osd_weight[osd] = off; 1390 } 1391 1392 /* new_pg_temp */ 1393 err = decode_new_pg_temp(p, end, map); 1394 if (err) 1395 goto bad; 1396 1397 /* new_primary_temp */ 1398 if (struct_v >= 1) { 1399 err = decode_new_primary_temp(p, end, map); 1400 if (err) 1401 goto bad; 1402 } 1403 1404 /* new_primary_affinity */ 1405 if (struct_v >= 2) { 1406 err = decode_new_primary_affinity(p, end, map); 1407 if (err) 1408 goto bad; 1409 } 1410 1411 /* ignore the rest */ 1412 *p = end; 1413 1414 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1415 return map; 1416 1417 e_inval: 1418 err = -EINVAL; 1419 bad: 1420 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1421 err, epoch, (int)(*p - start), *p, start, end); 1422 print_hex_dump(KERN_DEBUG, "osdmap: ", 1423 DUMP_PREFIX_OFFSET, 16, 1, 1424 start, end - start, true); 1425 if (newcrush) 1426 crush_destroy(newcrush); 1427 return ERR_PTR(err); 1428 } 1429 1430 void ceph_oid_copy(struct ceph_object_id *dest, 1431 const struct ceph_object_id *src) 1432 { 1433 WARN_ON(!ceph_oid_empty(dest)); 1434 1435 if (src->name != src->inline_name) { 1436 /* very rare, see ceph_object_id definition */ 1437 dest->name = kmalloc(src->name_len + 1, 1438 GFP_NOIO | __GFP_NOFAIL); 1439 } 1440 1441 memcpy(dest->name, src->name, src->name_len + 1); 1442 dest->name_len = src->name_len; 1443 } 1444 EXPORT_SYMBOL(ceph_oid_copy); 1445 1446 static __printf(2, 0) 1447 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap) 1448 { 1449 int len; 1450 1451 WARN_ON(!ceph_oid_empty(oid)); 1452 1453 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap); 1454 if (len >= sizeof(oid->inline_name)) 1455 return len; 1456 1457 oid->name_len = len; 1458 return 0; 1459 } 1460 1461 /* 1462 * If oid doesn't fit into inline buffer, BUG. 1463 */ 1464 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...) 1465 { 1466 va_list ap; 1467 1468 va_start(ap, fmt); 1469 BUG_ON(oid_printf_vargs(oid, fmt, ap)); 1470 va_end(ap); 1471 } 1472 EXPORT_SYMBOL(ceph_oid_printf); 1473 1474 static __printf(3, 0) 1475 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp, 1476 const char *fmt, va_list ap) 1477 { 1478 va_list aq; 1479 int len; 1480 1481 va_copy(aq, ap); 1482 len = oid_printf_vargs(oid, fmt, aq); 1483 va_end(aq); 1484 1485 if (len) { 1486 char *external_name; 1487 1488 external_name = kmalloc(len + 1, gfp); 1489 if (!external_name) 1490 return -ENOMEM; 1491 1492 oid->name = external_name; 1493 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len); 1494 oid->name_len = len; 1495 } 1496 1497 return 0; 1498 } 1499 1500 /* 1501 * If oid doesn't fit into inline buffer, allocate. 1502 */ 1503 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp, 1504 const char *fmt, ...) 1505 { 1506 va_list ap; 1507 int ret; 1508 1509 va_start(ap, fmt); 1510 ret = oid_aprintf_vargs(oid, gfp, fmt, ap); 1511 va_end(ap); 1512 1513 return ret; 1514 } 1515 EXPORT_SYMBOL(ceph_oid_aprintf); 1516 1517 void ceph_oid_destroy(struct ceph_object_id *oid) 1518 { 1519 if (oid->name != oid->inline_name) 1520 kfree(oid->name); 1521 } 1522 EXPORT_SYMBOL(ceph_oid_destroy); 1523 1524 static bool osds_valid(const struct ceph_osds *set) 1525 { 1526 /* non-empty set */ 1527 if (set->size > 0 && set->primary >= 0) 1528 return true; 1529 1530 /* empty can_shift_osds set */ 1531 if (!set->size && set->primary == -1) 1532 return true; 1533 1534 /* empty !can_shift_osds set - all NONE */ 1535 if (set->size > 0 && set->primary == -1) { 1536 int i; 1537 1538 for (i = 0; i < set->size; i++) { 1539 if (set->osds[i] != CRUSH_ITEM_NONE) 1540 break; 1541 } 1542 if (i == set->size) 1543 return true; 1544 } 1545 1546 return false; 1547 } 1548 1549 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src) 1550 { 1551 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0])); 1552 dest->size = src->size; 1553 dest->primary = src->primary; 1554 } 1555 1556 /* 1557 * calculate file layout from given offset, length. 1558 * fill in correct oid, logical length, and object extent 1559 * offset, length. 1560 * 1561 * for now, we write only a single su, until we can 1562 * pass a stride back to the caller. 1563 */ 1564 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout, 1565 u64 off, u64 len, 1566 u64 *ono, 1567 u64 *oxoff, u64 *oxlen) 1568 { 1569 u32 osize = le32_to_cpu(layout->fl_object_size); 1570 u32 su = le32_to_cpu(layout->fl_stripe_unit); 1571 u32 sc = le32_to_cpu(layout->fl_stripe_count); 1572 u32 bl, stripeno, stripepos, objsetno; 1573 u32 su_per_object; 1574 u64 t, su_offset; 1575 1576 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len, 1577 osize, su); 1578 if (su == 0 || sc == 0) 1579 goto invalid; 1580 su_per_object = osize / su; 1581 if (su_per_object == 0) 1582 goto invalid; 1583 dout("osize %u / su %u = su_per_object %u\n", osize, su, 1584 su_per_object); 1585 1586 if ((su & ~PAGE_MASK) != 0) 1587 goto invalid; 1588 1589 /* bl = *off / su; */ 1590 t = off; 1591 do_div(t, su); 1592 bl = t; 1593 dout("off %llu / su %u = bl %u\n", off, su, bl); 1594 1595 stripeno = bl / sc; 1596 stripepos = bl % sc; 1597 objsetno = stripeno / su_per_object; 1598 1599 *ono = objsetno * sc + stripepos; 1600 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono); 1601 1602 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */ 1603 t = off; 1604 su_offset = do_div(t, su); 1605 *oxoff = su_offset + (stripeno % su_per_object) * su; 1606 1607 /* 1608 * Calculate the length of the extent being written to the selected 1609 * object. This is the minimum of the full length requested (len) or 1610 * the remainder of the current stripe being written to. 1611 */ 1612 *oxlen = min_t(u64, len, su - su_offset); 1613 1614 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen); 1615 return 0; 1616 1617 invalid: 1618 dout(" invalid layout\n"); 1619 *ono = 0; 1620 *oxoff = 0; 1621 *oxlen = 0; 1622 return -EINVAL; 1623 } 1624 EXPORT_SYMBOL(ceph_calc_file_object_mapping); 1625 1626 /* 1627 * Map an object into a PG. 1628 * 1629 * Should only be called with target_oid and target_oloc (as opposed to 1630 * base_oid and base_oloc), since tiering isn't taken into account. 1631 */ 1632 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 1633 struct ceph_object_id *oid, 1634 struct ceph_object_locator *oloc, 1635 struct ceph_pg *raw_pgid) 1636 { 1637 struct ceph_pg_pool_info *pi; 1638 1639 pi = ceph_pg_pool_by_id(osdmap, oloc->pool); 1640 if (!pi) 1641 return -ENOENT; 1642 1643 raw_pgid->pool = oloc->pool; 1644 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name, 1645 oid->name_len); 1646 1647 dout("%s %*pE -> raw_pgid %llu.%x\n", __func__, oid->name_len, 1648 oid->name, raw_pgid->pool, raw_pgid->seed); 1649 return 0; 1650 } 1651 EXPORT_SYMBOL(ceph_object_locator_to_pg); 1652 1653 /* 1654 * Map a raw PG (full precision ps) into an actual PG. 1655 */ 1656 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi, 1657 const struct ceph_pg *raw_pgid, 1658 struct ceph_pg *pgid) 1659 { 1660 pgid->pool = raw_pgid->pool; 1661 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num, 1662 pi->pg_num_mask); 1663 } 1664 1665 /* 1666 * Map a raw PG (full precision ps) into a placement ps (placement 1667 * seed). Include pool id in that value so that different pools don't 1668 * use the same seeds. 1669 */ 1670 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi, 1671 const struct ceph_pg *raw_pgid) 1672 { 1673 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) { 1674 /* hash pool id and seed so that pool PGs do not overlap */ 1675 return crush_hash32_2(CRUSH_HASH_RJENKINS1, 1676 ceph_stable_mod(raw_pgid->seed, 1677 pi->pgp_num, 1678 pi->pgp_num_mask), 1679 raw_pgid->pool); 1680 } else { 1681 /* 1682 * legacy behavior: add ps and pool together. this is 1683 * not a great approach because the PGs from each pool 1684 * will overlap on top of each other: 0.5 == 1.4 == 1685 * 2.3 == ... 1686 */ 1687 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num, 1688 pi->pgp_num_mask) + 1689 (unsigned)raw_pgid->pool; 1690 } 1691 } 1692 1693 static int do_crush(struct ceph_osdmap *map, int ruleno, int x, 1694 int *result, int result_max, 1695 const __u32 *weight, int weight_max) 1696 { 1697 int r; 1698 1699 BUG_ON(result_max > CEPH_PG_MAX_SIZE); 1700 1701 mutex_lock(&map->crush_scratch_mutex); 1702 r = crush_do_rule(map->crush, ruleno, x, result, result_max, 1703 weight, weight_max, map->crush_scratch_ary); 1704 mutex_unlock(&map->crush_scratch_mutex); 1705 1706 return r; 1707 } 1708 1709 /* 1710 * Calculate raw set (CRUSH output) for given PG. The result may 1711 * contain nonexistent OSDs. ->primary is undefined for a raw set. 1712 * 1713 * Placement seed (CRUSH input) is returned through @ppps. 1714 */ 1715 static void pg_to_raw_osds(struct ceph_osdmap *osdmap, 1716 struct ceph_pg_pool_info *pi, 1717 const struct ceph_pg *raw_pgid, 1718 struct ceph_osds *raw, 1719 u32 *ppps) 1720 { 1721 u32 pps = raw_pg_to_pps(pi, raw_pgid); 1722 int ruleno; 1723 int len; 1724 1725 ceph_osds_init(raw); 1726 if (ppps) 1727 *ppps = pps; 1728 1729 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type, 1730 pi->size); 1731 if (ruleno < 0) { 1732 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n", 1733 pi->id, pi->crush_ruleset, pi->type, pi->size); 1734 return; 1735 } 1736 1737 len = do_crush(osdmap, ruleno, pps, raw->osds, 1738 min_t(int, pi->size, ARRAY_SIZE(raw->osds)), 1739 osdmap->osd_weight, osdmap->max_osd); 1740 if (len < 0) { 1741 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n", 1742 len, ruleno, pi->id, pi->crush_ruleset, pi->type, 1743 pi->size); 1744 return; 1745 } 1746 1747 raw->size = len; 1748 } 1749 1750 /* 1751 * Given raw set, calculate up set and up primary. By definition of an 1752 * up set, the result won't contain nonexistent or down OSDs. 1753 * 1754 * This is done in-place - on return @set is the up set. If it's 1755 * empty, ->primary will remain undefined. 1756 */ 1757 static void raw_to_up_osds(struct ceph_osdmap *osdmap, 1758 struct ceph_pg_pool_info *pi, 1759 struct ceph_osds *set) 1760 { 1761 int i; 1762 1763 /* ->primary is undefined for a raw set */ 1764 BUG_ON(set->primary != -1); 1765 1766 if (ceph_can_shift_osds(pi)) { 1767 int removed = 0; 1768 1769 /* shift left */ 1770 for (i = 0; i < set->size; i++) { 1771 if (ceph_osd_is_down(osdmap, set->osds[i])) { 1772 removed++; 1773 continue; 1774 } 1775 if (removed) 1776 set->osds[i - removed] = set->osds[i]; 1777 } 1778 set->size -= removed; 1779 if (set->size > 0) 1780 set->primary = set->osds[0]; 1781 } else { 1782 /* set down/dne devices to NONE */ 1783 for (i = set->size - 1; i >= 0; i--) { 1784 if (ceph_osd_is_down(osdmap, set->osds[i])) 1785 set->osds[i] = CRUSH_ITEM_NONE; 1786 else 1787 set->primary = set->osds[i]; 1788 } 1789 } 1790 } 1791 1792 static void apply_primary_affinity(struct ceph_osdmap *osdmap, 1793 struct ceph_pg_pool_info *pi, 1794 u32 pps, 1795 struct ceph_osds *up) 1796 { 1797 int i; 1798 int pos = -1; 1799 1800 /* 1801 * Do we have any non-default primary_affinity values for these 1802 * osds? 1803 */ 1804 if (!osdmap->osd_primary_affinity) 1805 return; 1806 1807 for (i = 0; i < up->size; i++) { 1808 int osd = up->osds[i]; 1809 1810 if (osd != CRUSH_ITEM_NONE && 1811 osdmap->osd_primary_affinity[osd] != 1812 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) { 1813 break; 1814 } 1815 } 1816 if (i == up->size) 1817 return; 1818 1819 /* 1820 * Pick the primary. Feed both the seed (for the pg) and the 1821 * osd into the hash/rng so that a proportional fraction of an 1822 * osd's pgs get rejected as primary. 1823 */ 1824 for (i = 0; i < up->size; i++) { 1825 int osd = up->osds[i]; 1826 u32 aff; 1827 1828 if (osd == CRUSH_ITEM_NONE) 1829 continue; 1830 1831 aff = osdmap->osd_primary_affinity[osd]; 1832 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY && 1833 (crush_hash32_2(CRUSH_HASH_RJENKINS1, 1834 pps, osd) >> 16) >= aff) { 1835 /* 1836 * We chose not to use this primary. Note it 1837 * anyway as a fallback in case we don't pick 1838 * anyone else, but keep looking. 1839 */ 1840 if (pos < 0) 1841 pos = i; 1842 } else { 1843 pos = i; 1844 break; 1845 } 1846 } 1847 if (pos < 0) 1848 return; 1849 1850 up->primary = up->osds[pos]; 1851 1852 if (ceph_can_shift_osds(pi) && pos > 0) { 1853 /* move the new primary to the front */ 1854 for (i = pos; i > 0; i--) 1855 up->osds[i] = up->osds[i - 1]; 1856 up->osds[0] = up->primary; 1857 } 1858 } 1859 1860 /* 1861 * Get pg_temp and primary_temp mappings for given PG. 1862 * 1863 * Note that a PG may have none, only pg_temp, only primary_temp or 1864 * both pg_temp and primary_temp mappings. This means @temp isn't 1865 * always a valid OSD set on return: in the "only primary_temp" case, 1866 * @temp will have its ->primary >= 0 but ->size == 0. 1867 */ 1868 static void get_temp_osds(struct ceph_osdmap *osdmap, 1869 struct ceph_pg_pool_info *pi, 1870 const struct ceph_pg *raw_pgid, 1871 struct ceph_osds *temp) 1872 { 1873 struct ceph_pg pgid; 1874 struct ceph_pg_mapping *pg; 1875 int i; 1876 1877 raw_pg_to_pg(pi, raw_pgid, &pgid); 1878 ceph_osds_init(temp); 1879 1880 /* pg_temp? */ 1881 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid); 1882 if (pg) { 1883 for (i = 0; i < pg->pg_temp.len; i++) { 1884 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) { 1885 if (ceph_can_shift_osds(pi)) 1886 continue; 1887 1888 temp->osds[temp->size++] = CRUSH_ITEM_NONE; 1889 } else { 1890 temp->osds[temp->size++] = pg->pg_temp.osds[i]; 1891 } 1892 } 1893 1894 /* apply pg_temp's primary */ 1895 for (i = 0; i < temp->size; i++) { 1896 if (temp->osds[i] != CRUSH_ITEM_NONE) { 1897 temp->primary = temp->osds[i]; 1898 break; 1899 } 1900 } 1901 } 1902 1903 /* primary_temp? */ 1904 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid); 1905 if (pg) 1906 temp->primary = pg->primary_temp.osd; 1907 } 1908 1909 /* 1910 * Map a PG to its acting set as well as its up set. 1911 * 1912 * Acting set is used for data mapping purposes, while up set can be 1913 * recorded for detecting interval changes and deciding whether to 1914 * resend a request. 1915 */ 1916 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 1917 const struct ceph_pg *raw_pgid, 1918 struct ceph_osds *up, 1919 struct ceph_osds *acting) 1920 { 1921 struct ceph_pg_pool_info *pi; 1922 u32 pps; 1923 1924 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool); 1925 if (!pi) { 1926 ceph_osds_init(up); 1927 ceph_osds_init(acting); 1928 goto out; 1929 } 1930 1931 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps); 1932 raw_to_up_osds(osdmap, pi, up); 1933 apply_primary_affinity(osdmap, pi, pps, up); 1934 get_temp_osds(osdmap, pi, raw_pgid, acting); 1935 if (!acting->size) { 1936 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0])); 1937 acting->size = up->size; 1938 if (acting->primary == -1) 1939 acting->primary = up->primary; 1940 } 1941 out: 1942 WARN_ON(!osds_valid(up) || !osds_valid(acting)); 1943 } 1944 1945 /* 1946 * Return acting primary for given PG, or -1 if none. 1947 */ 1948 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 1949 const struct ceph_pg *raw_pgid) 1950 { 1951 struct ceph_osds up, acting; 1952 1953 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting); 1954 return acting.primary; 1955 } 1956 EXPORT_SYMBOL(ceph_pg_to_acting_primary); 1957