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 *p += 1; /* skip min_size */ 601 602 if (ev >= 8) 603 *p += 8 + 8; /* skip quota_max_* */ 604 605 if (ev >= 9) { 606 /* skip tiers */ 607 num = ceph_decode_32(p); 608 *p += num * 8; 609 610 *p += 8; /* skip tier_of */ 611 *p += 1; /* skip cache_mode */ 612 613 pi->read_tier = ceph_decode_64(p); 614 pi->write_tier = ceph_decode_64(p); 615 } else { 616 pi->read_tier = -1; 617 pi->write_tier = -1; 618 } 619 620 /* ignore the rest */ 621 622 *p = pool_end; 623 calc_pg_masks(pi); 624 return 0; 625 626 bad: 627 return -EINVAL; 628 } 629 630 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map) 631 { 632 struct ceph_pg_pool_info *pi; 633 u32 num, len; 634 u64 pool; 635 636 ceph_decode_32_safe(p, end, num, bad); 637 dout(" %d pool names\n", num); 638 while (num--) { 639 ceph_decode_64_safe(p, end, pool, bad); 640 ceph_decode_32_safe(p, end, len, bad); 641 dout(" pool %llu len %d\n", pool, len); 642 ceph_decode_need(p, end, len, bad); 643 pi = __lookup_pg_pool(&map->pg_pools, pool); 644 if (pi) { 645 char *name = kstrndup(*p, len, GFP_NOFS); 646 647 if (!name) 648 return -ENOMEM; 649 kfree(pi->name); 650 pi->name = name; 651 dout(" name is %s\n", pi->name); 652 } 653 *p += len; 654 } 655 return 0; 656 657 bad: 658 return -EINVAL; 659 } 660 661 /* 662 * osd map 663 */ 664 void ceph_osdmap_destroy(struct ceph_osdmap *map) 665 { 666 dout("osdmap_destroy %p\n", map); 667 if (map->crush) 668 crush_destroy(map->crush); 669 while (!RB_EMPTY_ROOT(&map->pg_temp)) { 670 struct ceph_pg_mapping *pg = 671 rb_entry(rb_first(&map->pg_temp), 672 struct ceph_pg_mapping, node); 673 rb_erase(&pg->node, &map->pg_temp); 674 kfree(pg); 675 } 676 while (!RB_EMPTY_ROOT(&map->primary_temp)) { 677 struct ceph_pg_mapping *pg = 678 rb_entry(rb_first(&map->primary_temp), 679 struct ceph_pg_mapping, node); 680 rb_erase(&pg->node, &map->primary_temp); 681 kfree(pg); 682 } 683 while (!RB_EMPTY_ROOT(&map->pg_pools)) { 684 struct ceph_pg_pool_info *pi = 685 rb_entry(rb_first(&map->pg_pools), 686 struct ceph_pg_pool_info, node); 687 __remove_pg_pool(&map->pg_pools, pi); 688 } 689 kfree(map->osd_state); 690 kfree(map->osd_weight); 691 kfree(map->osd_addr); 692 kfree(map->osd_primary_affinity); 693 kfree(map); 694 } 695 696 /* 697 * Adjust max_osd value, (re)allocate arrays. 698 * 699 * The new elements are properly initialized. 700 */ 701 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max) 702 { 703 u8 *state; 704 u32 *weight; 705 struct ceph_entity_addr *addr; 706 int i; 707 708 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS); 709 if (!state) 710 return -ENOMEM; 711 map->osd_state = state; 712 713 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS); 714 if (!weight) 715 return -ENOMEM; 716 map->osd_weight = weight; 717 718 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS); 719 if (!addr) 720 return -ENOMEM; 721 map->osd_addr = addr; 722 723 for (i = map->max_osd; i < max; i++) { 724 map->osd_state[i] = 0; 725 map->osd_weight[i] = CEPH_OSD_OUT; 726 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr)); 727 } 728 729 if (map->osd_primary_affinity) { 730 u32 *affinity; 731 732 affinity = krealloc(map->osd_primary_affinity, 733 max*sizeof(*affinity), GFP_NOFS); 734 if (!affinity) 735 return -ENOMEM; 736 map->osd_primary_affinity = affinity; 737 738 for (i = map->max_osd; i < max; i++) 739 map->osd_primary_affinity[i] = 740 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 741 } 742 743 map->max_osd = max; 744 745 return 0; 746 } 747 748 #define OSDMAP_WRAPPER_COMPAT_VER 7 749 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1 750 751 /* 752 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps, 753 * to struct_v of the client_data section for new (v7 and above) 754 * osdmaps. 755 */ 756 static int get_osdmap_client_data_v(void **p, void *end, 757 const char *prefix, u8 *v) 758 { 759 u8 struct_v; 760 761 ceph_decode_8_safe(p, end, struct_v, e_inval); 762 if (struct_v >= 7) { 763 u8 struct_compat; 764 765 ceph_decode_8_safe(p, end, struct_compat, e_inval); 766 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) { 767 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n", 768 struct_v, struct_compat, 769 OSDMAP_WRAPPER_COMPAT_VER, prefix); 770 return -EINVAL; 771 } 772 *p += 4; /* ignore wrapper struct_len */ 773 774 ceph_decode_8_safe(p, end, struct_v, e_inval); 775 ceph_decode_8_safe(p, end, struct_compat, e_inval); 776 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) { 777 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n", 778 struct_v, struct_compat, 779 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix); 780 return -EINVAL; 781 } 782 *p += 4; /* ignore client data struct_len */ 783 } else { 784 u16 version; 785 786 *p -= 1; 787 ceph_decode_16_safe(p, end, version, e_inval); 788 if (version < 6) { 789 pr_warn("got v %d < 6 of %s ceph_osdmap\n", 790 version, prefix); 791 return -EINVAL; 792 } 793 794 /* old osdmap enconding */ 795 struct_v = 0; 796 } 797 798 *v = struct_v; 799 return 0; 800 801 e_inval: 802 return -EINVAL; 803 } 804 805 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map, 806 bool incremental) 807 { 808 u32 n; 809 810 ceph_decode_32_safe(p, end, n, e_inval); 811 while (n--) { 812 struct ceph_pg_pool_info *pi; 813 u64 pool; 814 int ret; 815 816 ceph_decode_64_safe(p, end, pool, e_inval); 817 818 pi = __lookup_pg_pool(&map->pg_pools, pool); 819 if (!incremental || !pi) { 820 pi = kzalloc(sizeof(*pi), GFP_NOFS); 821 if (!pi) 822 return -ENOMEM; 823 824 pi->id = pool; 825 826 ret = __insert_pg_pool(&map->pg_pools, pi); 827 if (ret) { 828 kfree(pi); 829 return ret; 830 } 831 } 832 833 ret = decode_pool(p, end, pi); 834 if (ret) 835 return ret; 836 } 837 838 return 0; 839 840 e_inval: 841 return -EINVAL; 842 } 843 844 static int decode_pools(void **p, void *end, struct ceph_osdmap *map) 845 { 846 return __decode_pools(p, end, map, false); 847 } 848 849 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map) 850 { 851 return __decode_pools(p, end, map, true); 852 } 853 854 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map, 855 bool incremental) 856 { 857 u32 n; 858 859 ceph_decode_32_safe(p, end, n, e_inval); 860 while (n--) { 861 struct ceph_pg pgid; 862 u32 len, i; 863 int ret; 864 865 ret = ceph_decode_pgid(p, end, &pgid); 866 if (ret) 867 return ret; 868 869 ceph_decode_32_safe(p, end, len, e_inval); 870 871 ret = __remove_pg_mapping(&map->pg_temp, pgid); 872 BUG_ON(!incremental && ret != -ENOENT); 873 874 if (!incremental || len > 0) { 875 struct ceph_pg_mapping *pg; 876 877 ceph_decode_need(p, end, len*sizeof(u32), e_inval); 878 879 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32)) 880 return -EINVAL; 881 882 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS); 883 if (!pg) 884 return -ENOMEM; 885 886 pg->pgid = pgid; 887 pg->pg_temp.len = len; 888 for (i = 0; i < len; i++) 889 pg->pg_temp.osds[i] = ceph_decode_32(p); 890 891 ret = __insert_pg_mapping(pg, &map->pg_temp); 892 if (ret) { 893 kfree(pg); 894 return ret; 895 } 896 } 897 } 898 899 return 0; 900 901 e_inval: 902 return -EINVAL; 903 } 904 905 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map) 906 { 907 return __decode_pg_temp(p, end, map, false); 908 } 909 910 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map) 911 { 912 return __decode_pg_temp(p, end, map, true); 913 } 914 915 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map, 916 bool incremental) 917 { 918 u32 n; 919 920 ceph_decode_32_safe(p, end, n, e_inval); 921 while (n--) { 922 struct ceph_pg pgid; 923 u32 osd; 924 int ret; 925 926 ret = ceph_decode_pgid(p, end, &pgid); 927 if (ret) 928 return ret; 929 930 ceph_decode_32_safe(p, end, osd, e_inval); 931 932 ret = __remove_pg_mapping(&map->primary_temp, pgid); 933 BUG_ON(!incremental && ret != -ENOENT); 934 935 if (!incremental || osd != (u32)-1) { 936 struct ceph_pg_mapping *pg; 937 938 pg = kzalloc(sizeof(*pg), GFP_NOFS); 939 if (!pg) 940 return -ENOMEM; 941 942 pg->pgid = pgid; 943 pg->primary_temp.osd = osd; 944 945 ret = __insert_pg_mapping(pg, &map->primary_temp); 946 if (ret) { 947 kfree(pg); 948 return ret; 949 } 950 } 951 } 952 953 return 0; 954 955 e_inval: 956 return -EINVAL; 957 } 958 959 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map) 960 { 961 return __decode_primary_temp(p, end, map, false); 962 } 963 964 static int decode_new_primary_temp(void **p, void *end, 965 struct ceph_osdmap *map) 966 { 967 return __decode_primary_temp(p, end, map, true); 968 } 969 970 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd) 971 { 972 BUG_ON(osd >= map->max_osd); 973 974 if (!map->osd_primary_affinity) 975 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 976 977 return map->osd_primary_affinity[osd]; 978 } 979 980 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff) 981 { 982 BUG_ON(osd >= map->max_osd); 983 984 if (!map->osd_primary_affinity) { 985 int i; 986 987 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32), 988 GFP_NOFS); 989 if (!map->osd_primary_affinity) 990 return -ENOMEM; 991 992 for (i = 0; i < map->max_osd; i++) 993 map->osd_primary_affinity[i] = 994 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; 995 } 996 997 map->osd_primary_affinity[osd] = aff; 998 999 return 0; 1000 } 1001 1002 static int decode_primary_affinity(void **p, void *end, 1003 struct ceph_osdmap *map) 1004 { 1005 u32 len, i; 1006 1007 ceph_decode_32_safe(p, end, len, e_inval); 1008 if (len == 0) { 1009 kfree(map->osd_primary_affinity); 1010 map->osd_primary_affinity = NULL; 1011 return 0; 1012 } 1013 if (len != map->max_osd) 1014 goto e_inval; 1015 1016 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval); 1017 1018 for (i = 0; i < map->max_osd; i++) { 1019 int ret; 1020 1021 ret = set_primary_affinity(map, i, ceph_decode_32(p)); 1022 if (ret) 1023 return ret; 1024 } 1025 1026 return 0; 1027 1028 e_inval: 1029 return -EINVAL; 1030 } 1031 1032 static int decode_new_primary_affinity(void **p, void *end, 1033 struct ceph_osdmap *map) 1034 { 1035 u32 n; 1036 1037 ceph_decode_32_safe(p, end, n, e_inval); 1038 while (n--) { 1039 u32 osd, aff; 1040 int ret; 1041 1042 ceph_decode_32_safe(p, end, osd, e_inval); 1043 ceph_decode_32_safe(p, end, aff, e_inval); 1044 1045 ret = set_primary_affinity(map, osd, aff); 1046 if (ret) 1047 return ret; 1048 1049 pr_info("osd%d primary-affinity 0x%x\n", osd, aff); 1050 } 1051 1052 return 0; 1053 1054 e_inval: 1055 return -EINVAL; 1056 } 1057 1058 /* 1059 * decode a full map. 1060 */ 1061 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map) 1062 { 1063 u8 struct_v; 1064 u32 epoch = 0; 1065 void *start = *p; 1066 u32 max; 1067 u32 len, i; 1068 int err; 1069 1070 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1071 1072 err = get_osdmap_client_data_v(p, end, "full", &struct_v); 1073 if (err) 1074 goto bad; 1075 1076 /* fsid, epoch, created, modified */ 1077 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) + 1078 sizeof(map->created) + sizeof(map->modified), e_inval); 1079 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid)); 1080 epoch = map->epoch = ceph_decode_32(p); 1081 ceph_decode_copy(p, &map->created, sizeof(map->created)); 1082 ceph_decode_copy(p, &map->modified, sizeof(map->modified)); 1083 1084 /* pools */ 1085 err = decode_pools(p, end, map); 1086 if (err) 1087 goto bad; 1088 1089 /* pool_name */ 1090 err = decode_pool_names(p, end, map); 1091 if (err) 1092 goto bad; 1093 1094 ceph_decode_32_safe(p, end, map->pool_max, e_inval); 1095 1096 ceph_decode_32_safe(p, end, map->flags, e_inval); 1097 1098 /* max_osd */ 1099 ceph_decode_32_safe(p, end, max, e_inval); 1100 1101 /* (re)alloc osd arrays */ 1102 err = osdmap_set_max_osd(map, max); 1103 if (err) 1104 goto bad; 1105 1106 /* osd_state, osd_weight, osd_addrs->client_addr */ 1107 ceph_decode_need(p, end, 3*sizeof(u32) + 1108 map->max_osd*(1 + sizeof(*map->osd_weight) + 1109 sizeof(*map->osd_addr)), e_inval); 1110 1111 if (ceph_decode_32(p) != map->max_osd) 1112 goto e_inval; 1113 1114 ceph_decode_copy(p, map->osd_state, map->max_osd); 1115 1116 if (ceph_decode_32(p) != map->max_osd) 1117 goto e_inval; 1118 1119 for (i = 0; i < map->max_osd; i++) 1120 map->osd_weight[i] = ceph_decode_32(p); 1121 1122 if (ceph_decode_32(p) != map->max_osd) 1123 goto e_inval; 1124 1125 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr)); 1126 for (i = 0; i < map->max_osd; i++) 1127 ceph_decode_addr(&map->osd_addr[i]); 1128 1129 /* pg_temp */ 1130 err = decode_pg_temp(p, end, map); 1131 if (err) 1132 goto bad; 1133 1134 /* primary_temp */ 1135 if (struct_v >= 1) { 1136 err = decode_primary_temp(p, end, map); 1137 if (err) 1138 goto bad; 1139 } 1140 1141 /* primary_affinity */ 1142 if (struct_v >= 2) { 1143 err = decode_primary_affinity(p, end, map); 1144 if (err) 1145 goto bad; 1146 } else { 1147 /* XXX can this happen? */ 1148 kfree(map->osd_primary_affinity); 1149 map->osd_primary_affinity = NULL; 1150 } 1151 1152 /* crush */ 1153 ceph_decode_32_safe(p, end, len, e_inval); 1154 map->crush = crush_decode(*p, min(*p + len, end)); 1155 if (IS_ERR(map->crush)) { 1156 err = PTR_ERR(map->crush); 1157 map->crush = NULL; 1158 goto bad; 1159 } 1160 *p += len; 1161 1162 /* ignore the rest */ 1163 *p = end; 1164 1165 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1166 return 0; 1167 1168 e_inval: 1169 err = -EINVAL; 1170 bad: 1171 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1172 err, epoch, (int)(*p - start), *p, start, end); 1173 print_hex_dump(KERN_DEBUG, "osdmap: ", 1174 DUMP_PREFIX_OFFSET, 16, 1, 1175 start, end - start, true); 1176 return err; 1177 } 1178 1179 /* 1180 * Allocate and decode a full map. 1181 */ 1182 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end) 1183 { 1184 struct ceph_osdmap *map; 1185 int ret; 1186 1187 map = kzalloc(sizeof(*map), GFP_NOFS); 1188 if (!map) 1189 return ERR_PTR(-ENOMEM); 1190 1191 map->pg_temp = RB_ROOT; 1192 map->primary_temp = RB_ROOT; 1193 mutex_init(&map->crush_scratch_mutex); 1194 1195 ret = osdmap_decode(p, end, map); 1196 if (ret) { 1197 ceph_osdmap_destroy(map); 1198 return ERR_PTR(ret); 1199 } 1200 1201 return map; 1202 } 1203 1204 /* 1205 * decode and apply an incremental map update. 1206 */ 1207 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 1208 struct ceph_osdmap *map) 1209 { 1210 struct crush_map *newcrush = NULL; 1211 struct ceph_fsid fsid; 1212 u32 epoch = 0; 1213 struct ceph_timespec modified; 1214 s32 len; 1215 u64 pool; 1216 __s64 new_pool_max; 1217 __s32 new_flags, max; 1218 void *start = *p; 1219 int err; 1220 u8 struct_v; 1221 1222 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p)); 1223 1224 err = get_osdmap_client_data_v(p, end, "inc", &struct_v); 1225 if (err) 1226 goto bad; 1227 1228 /* fsid, epoch, modified, new_pool_max, new_flags */ 1229 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) + 1230 sizeof(u64) + sizeof(u32), e_inval); 1231 ceph_decode_copy(p, &fsid, sizeof(fsid)); 1232 epoch = ceph_decode_32(p); 1233 BUG_ON(epoch != map->epoch+1); 1234 ceph_decode_copy(p, &modified, sizeof(modified)); 1235 new_pool_max = ceph_decode_64(p); 1236 new_flags = ceph_decode_32(p); 1237 1238 /* full map? */ 1239 ceph_decode_32_safe(p, end, len, e_inval); 1240 if (len > 0) { 1241 dout("apply_incremental full map len %d, %p to %p\n", 1242 len, *p, end); 1243 return ceph_osdmap_decode(p, min(*p+len, end)); 1244 } 1245 1246 /* new crush? */ 1247 ceph_decode_32_safe(p, end, len, e_inval); 1248 if (len > 0) { 1249 newcrush = crush_decode(*p, min(*p+len, end)); 1250 if (IS_ERR(newcrush)) { 1251 err = PTR_ERR(newcrush); 1252 newcrush = NULL; 1253 goto bad; 1254 } 1255 *p += len; 1256 } 1257 1258 /* new flags? */ 1259 if (new_flags >= 0) 1260 map->flags = new_flags; 1261 if (new_pool_max >= 0) 1262 map->pool_max = new_pool_max; 1263 1264 /* new max? */ 1265 ceph_decode_32_safe(p, end, max, e_inval); 1266 if (max >= 0) { 1267 err = osdmap_set_max_osd(map, max); 1268 if (err) 1269 goto bad; 1270 } 1271 1272 map->epoch++; 1273 map->modified = modified; 1274 if (newcrush) { 1275 if (map->crush) 1276 crush_destroy(map->crush); 1277 map->crush = newcrush; 1278 newcrush = NULL; 1279 } 1280 1281 /* new_pools */ 1282 err = decode_new_pools(p, end, map); 1283 if (err) 1284 goto bad; 1285 1286 /* new_pool_names */ 1287 err = decode_pool_names(p, end, map); 1288 if (err) 1289 goto bad; 1290 1291 /* old_pool */ 1292 ceph_decode_32_safe(p, end, len, e_inval); 1293 while (len--) { 1294 struct ceph_pg_pool_info *pi; 1295 1296 ceph_decode_64_safe(p, end, pool, e_inval); 1297 pi = __lookup_pg_pool(&map->pg_pools, pool); 1298 if (pi) 1299 __remove_pg_pool(&map->pg_pools, pi); 1300 } 1301 1302 /* new_up */ 1303 ceph_decode_32_safe(p, end, len, e_inval); 1304 while (len--) { 1305 u32 osd; 1306 struct ceph_entity_addr addr; 1307 ceph_decode_32_safe(p, end, osd, e_inval); 1308 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval); 1309 ceph_decode_addr(&addr); 1310 pr_info("osd%d up\n", osd); 1311 BUG_ON(osd >= map->max_osd); 1312 map->osd_state[osd] |= CEPH_OSD_UP | CEPH_OSD_EXISTS; 1313 map->osd_addr[osd] = addr; 1314 } 1315 1316 /* new_state */ 1317 ceph_decode_32_safe(p, end, len, e_inval); 1318 while (len--) { 1319 u32 osd; 1320 u8 xorstate; 1321 ceph_decode_32_safe(p, end, osd, e_inval); 1322 xorstate = **(u8 **)p; 1323 (*p)++; /* clean flag */ 1324 if (xorstate == 0) 1325 xorstate = CEPH_OSD_UP; 1326 if (xorstate & CEPH_OSD_UP) 1327 pr_info("osd%d down\n", osd); 1328 if (osd < map->max_osd) 1329 map->osd_state[osd] ^= xorstate; 1330 } 1331 1332 /* new_weight */ 1333 ceph_decode_32_safe(p, end, len, e_inval); 1334 while (len--) { 1335 u32 osd, off; 1336 ceph_decode_need(p, end, sizeof(u32)*2, e_inval); 1337 osd = ceph_decode_32(p); 1338 off = ceph_decode_32(p); 1339 pr_info("osd%d weight 0x%x %s\n", osd, off, 1340 off == CEPH_OSD_IN ? "(in)" : 1341 (off == CEPH_OSD_OUT ? "(out)" : "")); 1342 if (osd < map->max_osd) 1343 map->osd_weight[osd] = off; 1344 } 1345 1346 /* new_pg_temp */ 1347 err = decode_new_pg_temp(p, end, map); 1348 if (err) 1349 goto bad; 1350 1351 /* new_primary_temp */ 1352 if (struct_v >= 1) { 1353 err = decode_new_primary_temp(p, end, map); 1354 if (err) 1355 goto bad; 1356 } 1357 1358 /* new_primary_affinity */ 1359 if (struct_v >= 2) { 1360 err = decode_new_primary_affinity(p, end, map); 1361 if (err) 1362 goto bad; 1363 } 1364 1365 /* ignore the rest */ 1366 *p = end; 1367 1368 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd); 1369 return map; 1370 1371 e_inval: 1372 err = -EINVAL; 1373 bad: 1374 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n", 1375 err, epoch, (int)(*p - start), *p, start, end); 1376 print_hex_dump(KERN_DEBUG, "osdmap: ", 1377 DUMP_PREFIX_OFFSET, 16, 1, 1378 start, end - start, true); 1379 if (newcrush) 1380 crush_destroy(newcrush); 1381 return ERR_PTR(err); 1382 } 1383 1384 void ceph_oid_copy(struct ceph_object_id *dest, 1385 const struct ceph_object_id *src) 1386 { 1387 WARN_ON(!ceph_oid_empty(dest)); 1388 1389 if (src->name != src->inline_name) { 1390 /* very rare, see ceph_object_id definition */ 1391 dest->name = kmalloc(src->name_len + 1, 1392 GFP_NOIO | __GFP_NOFAIL); 1393 } 1394 1395 memcpy(dest->name, src->name, src->name_len + 1); 1396 dest->name_len = src->name_len; 1397 } 1398 EXPORT_SYMBOL(ceph_oid_copy); 1399 1400 static __printf(2, 0) 1401 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap) 1402 { 1403 int len; 1404 1405 WARN_ON(!ceph_oid_empty(oid)); 1406 1407 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap); 1408 if (len >= sizeof(oid->inline_name)) 1409 return len; 1410 1411 oid->name_len = len; 1412 return 0; 1413 } 1414 1415 /* 1416 * If oid doesn't fit into inline buffer, BUG. 1417 */ 1418 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...) 1419 { 1420 va_list ap; 1421 1422 va_start(ap, fmt); 1423 BUG_ON(oid_printf_vargs(oid, fmt, ap)); 1424 va_end(ap); 1425 } 1426 EXPORT_SYMBOL(ceph_oid_printf); 1427 1428 static __printf(3, 0) 1429 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp, 1430 const char *fmt, va_list ap) 1431 { 1432 va_list aq; 1433 int len; 1434 1435 va_copy(aq, ap); 1436 len = oid_printf_vargs(oid, fmt, aq); 1437 va_end(aq); 1438 1439 if (len) { 1440 char *external_name; 1441 1442 external_name = kmalloc(len + 1, gfp); 1443 if (!external_name) 1444 return -ENOMEM; 1445 1446 oid->name = external_name; 1447 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len); 1448 oid->name_len = len; 1449 } 1450 1451 return 0; 1452 } 1453 1454 /* 1455 * If oid doesn't fit into inline buffer, allocate. 1456 */ 1457 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp, 1458 const char *fmt, ...) 1459 { 1460 va_list ap; 1461 int ret; 1462 1463 va_start(ap, fmt); 1464 ret = oid_aprintf_vargs(oid, gfp, fmt, ap); 1465 va_end(ap); 1466 1467 return ret; 1468 } 1469 EXPORT_SYMBOL(ceph_oid_aprintf); 1470 1471 void ceph_oid_destroy(struct ceph_object_id *oid) 1472 { 1473 if (oid->name != oid->inline_name) 1474 kfree(oid->name); 1475 } 1476 EXPORT_SYMBOL(ceph_oid_destroy); 1477 1478 static bool osds_valid(const struct ceph_osds *set) 1479 { 1480 /* non-empty set */ 1481 if (set->size > 0 && set->primary >= 0) 1482 return true; 1483 1484 /* empty can_shift_osds set */ 1485 if (!set->size && set->primary == -1) 1486 return true; 1487 1488 /* empty !can_shift_osds set - all NONE */ 1489 if (set->size > 0 && set->primary == -1) { 1490 int i; 1491 1492 for (i = 0; i < set->size; i++) { 1493 if (set->osds[i] != CRUSH_ITEM_NONE) 1494 break; 1495 } 1496 if (i == set->size) 1497 return true; 1498 } 1499 1500 return false; 1501 } 1502 1503 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src) 1504 { 1505 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0])); 1506 dest->size = src->size; 1507 dest->primary = src->primary; 1508 } 1509 1510 /* 1511 * calculate file layout from given offset, length. 1512 * fill in correct oid, logical length, and object extent 1513 * offset, length. 1514 * 1515 * for now, we write only a single su, until we can 1516 * pass a stride back to the caller. 1517 */ 1518 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout, 1519 u64 off, u64 len, 1520 u64 *ono, 1521 u64 *oxoff, u64 *oxlen) 1522 { 1523 u32 osize = le32_to_cpu(layout->fl_object_size); 1524 u32 su = le32_to_cpu(layout->fl_stripe_unit); 1525 u32 sc = le32_to_cpu(layout->fl_stripe_count); 1526 u32 bl, stripeno, stripepos, objsetno; 1527 u32 su_per_object; 1528 u64 t, su_offset; 1529 1530 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len, 1531 osize, su); 1532 if (su == 0 || sc == 0) 1533 goto invalid; 1534 su_per_object = osize / su; 1535 if (su_per_object == 0) 1536 goto invalid; 1537 dout("osize %u / su %u = su_per_object %u\n", osize, su, 1538 su_per_object); 1539 1540 if ((su & ~PAGE_MASK) != 0) 1541 goto invalid; 1542 1543 /* bl = *off / su; */ 1544 t = off; 1545 do_div(t, su); 1546 bl = t; 1547 dout("off %llu / su %u = bl %u\n", off, su, bl); 1548 1549 stripeno = bl / sc; 1550 stripepos = bl % sc; 1551 objsetno = stripeno / su_per_object; 1552 1553 *ono = objsetno * sc + stripepos; 1554 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono); 1555 1556 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */ 1557 t = off; 1558 su_offset = do_div(t, su); 1559 *oxoff = su_offset + (stripeno % su_per_object) * su; 1560 1561 /* 1562 * Calculate the length of the extent being written to the selected 1563 * object. This is the minimum of the full length requested (len) or 1564 * the remainder of the current stripe being written to. 1565 */ 1566 *oxlen = min_t(u64, len, su - su_offset); 1567 1568 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen); 1569 return 0; 1570 1571 invalid: 1572 dout(" invalid layout\n"); 1573 *ono = 0; 1574 *oxoff = 0; 1575 *oxlen = 0; 1576 return -EINVAL; 1577 } 1578 EXPORT_SYMBOL(ceph_calc_file_object_mapping); 1579 1580 /* 1581 * Map an object into a PG. 1582 * 1583 * Should only be called with target_oid and target_oloc (as opposed to 1584 * base_oid and base_oloc), since tiering isn't taken into account. 1585 */ 1586 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 1587 struct ceph_object_id *oid, 1588 struct ceph_object_locator *oloc, 1589 struct ceph_pg *raw_pgid) 1590 { 1591 struct ceph_pg_pool_info *pi; 1592 1593 pi = ceph_pg_pool_by_id(osdmap, oloc->pool); 1594 if (!pi) 1595 return -ENOENT; 1596 1597 raw_pgid->pool = oloc->pool; 1598 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name, 1599 oid->name_len); 1600 1601 dout("%s %*pE -> raw_pgid %llu.%x\n", __func__, oid->name_len, 1602 oid->name, raw_pgid->pool, raw_pgid->seed); 1603 return 0; 1604 } 1605 EXPORT_SYMBOL(ceph_object_locator_to_pg); 1606 1607 /* 1608 * Map a raw PG (full precision ps) into an actual PG. 1609 */ 1610 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi, 1611 const struct ceph_pg *raw_pgid, 1612 struct ceph_pg *pgid) 1613 { 1614 pgid->pool = raw_pgid->pool; 1615 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num, 1616 pi->pg_num_mask); 1617 } 1618 1619 /* 1620 * Map a raw PG (full precision ps) into a placement ps (placement 1621 * seed). Include pool id in that value so that different pools don't 1622 * use the same seeds. 1623 */ 1624 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi, 1625 const struct ceph_pg *raw_pgid) 1626 { 1627 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) { 1628 /* hash pool id and seed so that pool PGs do not overlap */ 1629 return crush_hash32_2(CRUSH_HASH_RJENKINS1, 1630 ceph_stable_mod(raw_pgid->seed, 1631 pi->pgp_num, 1632 pi->pgp_num_mask), 1633 raw_pgid->pool); 1634 } else { 1635 /* 1636 * legacy behavior: add ps and pool together. this is 1637 * not a great approach because the PGs from each pool 1638 * will overlap on top of each other: 0.5 == 1.4 == 1639 * 2.3 == ... 1640 */ 1641 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num, 1642 pi->pgp_num_mask) + 1643 (unsigned)raw_pgid->pool; 1644 } 1645 } 1646 1647 static int do_crush(struct ceph_osdmap *map, int ruleno, int x, 1648 int *result, int result_max, 1649 const __u32 *weight, int weight_max) 1650 { 1651 int r; 1652 1653 BUG_ON(result_max > CEPH_PG_MAX_SIZE); 1654 1655 mutex_lock(&map->crush_scratch_mutex); 1656 r = crush_do_rule(map->crush, ruleno, x, result, result_max, 1657 weight, weight_max, map->crush_scratch_ary); 1658 mutex_unlock(&map->crush_scratch_mutex); 1659 1660 return r; 1661 } 1662 1663 /* 1664 * Calculate raw set (CRUSH output) for given PG. The result may 1665 * contain nonexistent OSDs. ->primary is undefined for a raw set. 1666 * 1667 * Placement seed (CRUSH input) is returned through @ppps. 1668 */ 1669 static void pg_to_raw_osds(struct ceph_osdmap *osdmap, 1670 struct ceph_pg_pool_info *pi, 1671 const struct ceph_pg *raw_pgid, 1672 struct ceph_osds *raw, 1673 u32 *ppps) 1674 { 1675 u32 pps = raw_pg_to_pps(pi, raw_pgid); 1676 int ruleno; 1677 int len; 1678 1679 ceph_osds_init(raw); 1680 if (ppps) 1681 *ppps = pps; 1682 1683 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type, 1684 pi->size); 1685 if (ruleno < 0) { 1686 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n", 1687 pi->id, pi->crush_ruleset, pi->type, pi->size); 1688 return; 1689 } 1690 1691 len = do_crush(osdmap, ruleno, pps, raw->osds, 1692 min_t(int, pi->size, ARRAY_SIZE(raw->osds)), 1693 osdmap->osd_weight, osdmap->max_osd); 1694 if (len < 0) { 1695 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n", 1696 len, ruleno, pi->id, pi->crush_ruleset, pi->type, 1697 pi->size); 1698 return; 1699 } 1700 1701 raw->size = len; 1702 } 1703 1704 /* 1705 * Given raw set, calculate up set and up primary. By definition of an 1706 * up set, the result won't contain nonexistent or down OSDs. 1707 * 1708 * This is done in-place - on return @set is the up set. If it's 1709 * empty, ->primary will remain undefined. 1710 */ 1711 static void raw_to_up_osds(struct ceph_osdmap *osdmap, 1712 struct ceph_pg_pool_info *pi, 1713 struct ceph_osds *set) 1714 { 1715 int i; 1716 1717 /* ->primary is undefined for a raw set */ 1718 BUG_ON(set->primary != -1); 1719 1720 if (ceph_can_shift_osds(pi)) { 1721 int removed = 0; 1722 1723 /* shift left */ 1724 for (i = 0; i < set->size; i++) { 1725 if (ceph_osd_is_down(osdmap, set->osds[i])) { 1726 removed++; 1727 continue; 1728 } 1729 if (removed) 1730 set->osds[i - removed] = set->osds[i]; 1731 } 1732 set->size -= removed; 1733 if (set->size > 0) 1734 set->primary = set->osds[0]; 1735 } else { 1736 /* set down/dne devices to NONE */ 1737 for (i = set->size - 1; i >= 0; i--) { 1738 if (ceph_osd_is_down(osdmap, set->osds[i])) 1739 set->osds[i] = CRUSH_ITEM_NONE; 1740 else 1741 set->primary = set->osds[i]; 1742 } 1743 } 1744 } 1745 1746 static void apply_primary_affinity(struct ceph_osdmap *osdmap, 1747 struct ceph_pg_pool_info *pi, 1748 u32 pps, 1749 struct ceph_osds *up) 1750 { 1751 int i; 1752 int pos = -1; 1753 1754 /* 1755 * Do we have any non-default primary_affinity values for these 1756 * osds? 1757 */ 1758 if (!osdmap->osd_primary_affinity) 1759 return; 1760 1761 for (i = 0; i < up->size; i++) { 1762 int osd = up->osds[i]; 1763 1764 if (osd != CRUSH_ITEM_NONE && 1765 osdmap->osd_primary_affinity[osd] != 1766 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) { 1767 break; 1768 } 1769 } 1770 if (i == up->size) 1771 return; 1772 1773 /* 1774 * Pick the primary. Feed both the seed (for the pg) and the 1775 * osd into the hash/rng so that a proportional fraction of an 1776 * osd's pgs get rejected as primary. 1777 */ 1778 for (i = 0; i < up->size; i++) { 1779 int osd = up->osds[i]; 1780 u32 aff; 1781 1782 if (osd == CRUSH_ITEM_NONE) 1783 continue; 1784 1785 aff = osdmap->osd_primary_affinity[osd]; 1786 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY && 1787 (crush_hash32_2(CRUSH_HASH_RJENKINS1, 1788 pps, osd) >> 16) >= aff) { 1789 /* 1790 * We chose not to use this primary. Note it 1791 * anyway as a fallback in case we don't pick 1792 * anyone else, but keep looking. 1793 */ 1794 if (pos < 0) 1795 pos = i; 1796 } else { 1797 pos = i; 1798 break; 1799 } 1800 } 1801 if (pos < 0) 1802 return; 1803 1804 up->primary = up->osds[pos]; 1805 1806 if (ceph_can_shift_osds(pi) && pos > 0) { 1807 /* move the new primary to the front */ 1808 for (i = pos; i > 0; i--) 1809 up->osds[i] = up->osds[i - 1]; 1810 up->osds[0] = up->primary; 1811 } 1812 } 1813 1814 /* 1815 * Get pg_temp and primary_temp mappings for given PG. 1816 * 1817 * Note that a PG may have none, only pg_temp, only primary_temp or 1818 * both pg_temp and primary_temp mappings. This means @temp isn't 1819 * always a valid OSD set on return: in the "only primary_temp" case, 1820 * @temp will have its ->primary >= 0 but ->size == 0. 1821 */ 1822 static void get_temp_osds(struct ceph_osdmap *osdmap, 1823 struct ceph_pg_pool_info *pi, 1824 const struct ceph_pg *raw_pgid, 1825 struct ceph_osds *temp) 1826 { 1827 struct ceph_pg pgid; 1828 struct ceph_pg_mapping *pg; 1829 int i; 1830 1831 raw_pg_to_pg(pi, raw_pgid, &pgid); 1832 ceph_osds_init(temp); 1833 1834 /* pg_temp? */ 1835 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid); 1836 if (pg) { 1837 for (i = 0; i < pg->pg_temp.len; i++) { 1838 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) { 1839 if (ceph_can_shift_osds(pi)) 1840 continue; 1841 1842 temp->osds[temp->size++] = CRUSH_ITEM_NONE; 1843 } else { 1844 temp->osds[temp->size++] = pg->pg_temp.osds[i]; 1845 } 1846 } 1847 1848 /* apply pg_temp's primary */ 1849 for (i = 0; i < temp->size; i++) { 1850 if (temp->osds[i] != CRUSH_ITEM_NONE) { 1851 temp->primary = temp->osds[i]; 1852 break; 1853 } 1854 } 1855 } 1856 1857 /* primary_temp? */ 1858 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid); 1859 if (pg) 1860 temp->primary = pg->primary_temp.osd; 1861 } 1862 1863 /* 1864 * Map a PG to its acting set as well as its up set. 1865 * 1866 * Acting set is used for data mapping purposes, while up set can be 1867 * recorded for detecting interval changes and deciding whether to 1868 * resend a request. 1869 */ 1870 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 1871 const struct ceph_pg *raw_pgid, 1872 struct ceph_osds *up, 1873 struct ceph_osds *acting) 1874 { 1875 struct ceph_pg_pool_info *pi; 1876 u32 pps; 1877 1878 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool); 1879 if (!pi) { 1880 ceph_osds_init(up); 1881 ceph_osds_init(acting); 1882 goto out; 1883 } 1884 1885 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps); 1886 raw_to_up_osds(osdmap, pi, up); 1887 apply_primary_affinity(osdmap, pi, pps, up); 1888 get_temp_osds(osdmap, pi, raw_pgid, acting); 1889 if (!acting->size) { 1890 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0])); 1891 acting->size = up->size; 1892 if (acting->primary == -1) 1893 acting->primary = up->primary; 1894 } 1895 out: 1896 WARN_ON(!osds_valid(up) || !osds_valid(acting)); 1897 } 1898 1899 /* 1900 * Return acting primary for given PG, or -1 if none. 1901 */ 1902 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 1903 const struct ceph_pg *raw_pgid) 1904 { 1905 struct ceph_osds up, acting; 1906 1907 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting); 1908 return acting.primary; 1909 } 1910 EXPORT_SYMBOL(ceph_pg_to_acting_primary); 1911