1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved. 25 * Copyright 2017 Nexenta Systems, Inc. 26 * Copyright (c) 2014 Integros [integros.com] 27 * Copyright 2016 Toomas Soome <tsoome@me.com> 28 * Copyright 2017 Joyent, Inc. 29 * Copyright (c) 2017, Intel Corporation. 30 */ 31 32 #include <sys/zfs_context.h> 33 #include <sys/fm/fs/zfs.h> 34 #include <sys/spa.h> 35 #include <sys/spa_impl.h> 36 #include <sys/bpobj.h> 37 #include <sys/dmu.h> 38 #include <sys/dmu_tx.h> 39 #include <sys/dsl_dir.h> 40 #include <sys/vdev_impl.h> 41 #include <sys/uberblock_impl.h> 42 #include <sys/metaslab.h> 43 #include <sys/metaslab_impl.h> 44 #include <sys/space_map.h> 45 #include <sys/space_reftree.h> 46 #include <sys/zio.h> 47 #include <sys/zap.h> 48 #include <sys/fs/zfs.h> 49 #include <sys/arc.h> 50 #include <sys/zil.h> 51 #include <sys/dsl_scan.h> 52 #include <sys/abd.h> 53 #include <sys/vdev_initialize.h> 54 55 /* 56 * Virtual device management. 57 */ 58 59 static vdev_ops_t *vdev_ops_table[] = { 60 &vdev_root_ops, 61 &vdev_raidz_ops, 62 &vdev_mirror_ops, 63 &vdev_replacing_ops, 64 &vdev_spare_ops, 65 &vdev_disk_ops, 66 &vdev_file_ops, 67 &vdev_missing_ops, 68 &vdev_hole_ops, 69 &vdev_indirect_ops, 70 NULL 71 }; 72 73 /* maximum scrub/resilver I/O queue per leaf vdev */ 74 int zfs_scrub_limit = 10; 75 76 /* default target for number of metaslabs per top-level vdev */ 77 int zfs_vdev_default_ms_count = 200; 78 79 /* minimum number of metaslabs per top-level vdev */ 80 int zfs_vdev_min_ms_count = 16; 81 82 /* practical upper limit of total metaslabs per top-level vdev */ 83 int zfs_vdev_ms_count_limit = 1ULL << 17; 84 85 /* lower limit for metaslab size (512M) */ 86 int zfs_vdev_default_ms_shift = 29; 87 88 /* upper limit for metaslab size (16G) */ 89 int zfs_vdev_max_ms_shift = 34; 90 91 boolean_t vdev_validate_skip = B_FALSE; 92 93 /* 94 * Since the DTL space map of a vdev is not expected to have a lot of 95 * entries, we default its block size to 4K. 96 */ 97 int vdev_dtl_sm_blksz = (1 << 12); 98 99 /* 100 * vdev-wide space maps that have lots of entries written to them at 101 * the end of each transaction can benefit from a higher I/O bandwidth 102 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 103 */ 104 int vdev_standard_sm_blksz = (1 << 17); 105 106 int zfs_ashift_min; 107 108 /*PRINTFLIKE2*/ 109 void 110 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 111 { 112 va_list adx; 113 char buf[256]; 114 115 va_start(adx, fmt); 116 (void) vsnprintf(buf, sizeof (buf), fmt, adx); 117 va_end(adx); 118 119 if (vd->vdev_path != NULL) { 120 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 121 vd->vdev_path, buf); 122 } else { 123 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 124 vd->vdev_ops->vdev_op_type, 125 (u_longlong_t)vd->vdev_id, 126 (u_longlong_t)vd->vdev_guid, buf); 127 } 128 } 129 130 void 131 vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 132 { 133 char state[20]; 134 135 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 136 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 137 vd->vdev_ops->vdev_op_type); 138 return; 139 } 140 141 switch (vd->vdev_state) { 142 case VDEV_STATE_UNKNOWN: 143 (void) snprintf(state, sizeof (state), "unknown"); 144 break; 145 case VDEV_STATE_CLOSED: 146 (void) snprintf(state, sizeof (state), "closed"); 147 break; 148 case VDEV_STATE_OFFLINE: 149 (void) snprintf(state, sizeof (state), "offline"); 150 break; 151 case VDEV_STATE_REMOVED: 152 (void) snprintf(state, sizeof (state), "removed"); 153 break; 154 case VDEV_STATE_CANT_OPEN: 155 (void) snprintf(state, sizeof (state), "can't open"); 156 break; 157 case VDEV_STATE_FAULTED: 158 (void) snprintf(state, sizeof (state), "faulted"); 159 break; 160 case VDEV_STATE_DEGRADED: 161 (void) snprintf(state, sizeof (state), "degraded"); 162 break; 163 case VDEV_STATE_HEALTHY: 164 (void) snprintf(state, sizeof (state), "healthy"); 165 break; 166 default: 167 (void) snprintf(state, sizeof (state), "<state %u>", 168 (uint_t)vd->vdev_state); 169 } 170 171 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 172 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, 173 vd->vdev_islog ? " (log)" : "", 174 (u_longlong_t)vd->vdev_guid, 175 vd->vdev_path ? vd->vdev_path : "N/A", state); 176 177 for (uint64_t i = 0; i < vd->vdev_children; i++) 178 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 179 } 180 181 /* 182 * Given a vdev type, return the appropriate ops vector. 183 */ 184 static vdev_ops_t * 185 vdev_getops(const char *type) 186 { 187 vdev_ops_t *ops, **opspp; 188 189 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 190 if (strcmp(ops->vdev_op_type, type) == 0) 191 break; 192 193 return (ops); 194 } 195 196 /* 197 * Derive the enumerated alloction bias from string input. 198 * String origin is either the per-vdev zap or zpool(1M). 199 */ 200 static vdev_alloc_bias_t 201 vdev_derive_alloc_bias(const char *bias) 202 { 203 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 204 205 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0) 206 alloc_bias = VDEV_BIAS_LOG; 207 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) 208 alloc_bias = VDEV_BIAS_SPECIAL; 209 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0) 210 alloc_bias = VDEV_BIAS_DEDUP; 211 212 return (alloc_bias); 213 } 214 215 /* ARGSUSED */ 216 void 217 vdev_default_xlate(vdev_t *vd, const range_seg_t *in, range_seg_t *res) 218 { 219 res->rs_start = in->rs_start; 220 res->rs_end = in->rs_end; 221 } 222 223 /* 224 * Default asize function: return the MAX of psize with the asize of 225 * all children. This is what's used by anything other than RAID-Z. 226 */ 227 uint64_t 228 vdev_default_asize(vdev_t *vd, uint64_t psize) 229 { 230 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 231 uint64_t csize; 232 233 for (int c = 0; c < vd->vdev_children; c++) { 234 csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 235 asize = MAX(asize, csize); 236 } 237 238 return (asize); 239 } 240 241 /* 242 * Get the minimum allocatable size. We define the allocatable size as 243 * the vdev's asize rounded to the nearest metaslab. This allows us to 244 * replace or attach devices which don't have the same physical size but 245 * can still satisfy the same number of allocations. 246 */ 247 uint64_t 248 vdev_get_min_asize(vdev_t *vd) 249 { 250 vdev_t *pvd = vd->vdev_parent; 251 252 /* 253 * If our parent is NULL (inactive spare or cache) or is the root, 254 * just return our own asize. 255 */ 256 if (pvd == NULL) 257 return (vd->vdev_asize); 258 259 /* 260 * The top-level vdev just returns the allocatable size rounded 261 * to the nearest metaslab. 262 */ 263 if (vd == vd->vdev_top) 264 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 265 266 /* 267 * The allocatable space for a raidz vdev is N * sizeof(smallest child), 268 * so each child must provide at least 1/Nth of its asize. 269 */ 270 if (pvd->vdev_ops == &vdev_raidz_ops) 271 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 272 pvd->vdev_children); 273 274 return (pvd->vdev_min_asize); 275 } 276 277 void 278 vdev_set_min_asize(vdev_t *vd) 279 { 280 vd->vdev_min_asize = vdev_get_min_asize(vd); 281 282 for (int c = 0; c < vd->vdev_children; c++) 283 vdev_set_min_asize(vd->vdev_child[c]); 284 } 285 286 vdev_t * 287 vdev_lookup_top(spa_t *spa, uint64_t vdev) 288 { 289 vdev_t *rvd = spa->spa_root_vdev; 290 291 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 292 293 if (vdev < rvd->vdev_children) { 294 ASSERT(rvd->vdev_child[vdev] != NULL); 295 return (rvd->vdev_child[vdev]); 296 } 297 298 return (NULL); 299 } 300 301 vdev_t * 302 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 303 { 304 vdev_t *mvd; 305 306 if (vd->vdev_guid == guid) 307 return (vd); 308 309 for (int c = 0; c < vd->vdev_children; c++) 310 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 311 NULL) 312 return (mvd); 313 314 return (NULL); 315 } 316 317 static int 318 vdev_count_leaves_impl(vdev_t *vd) 319 { 320 int n = 0; 321 322 if (vd->vdev_ops->vdev_op_leaf) 323 return (1); 324 325 for (int c = 0; c < vd->vdev_children; c++) 326 n += vdev_count_leaves_impl(vd->vdev_child[c]); 327 328 return (n); 329 } 330 331 int 332 vdev_count_leaves(spa_t *spa) 333 { 334 return (vdev_count_leaves_impl(spa->spa_root_vdev)); 335 } 336 337 void 338 vdev_add_child(vdev_t *pvd, vdev_t *cvd) 339 { 340 size_t oldsize, newsize; 341 uint64_t id = cvd->vdev_id; 342 vdev_t **newchild; 343 spa_t *spa = cvd->vdev_spa; 344 345 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 346 ASSERT(cvd->vdev_parent == NULL); 347 348 cvd->vdev_parent = pvd; 349 350 if (pvd == NULL) 351 return; 352 353 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 354 355 oldsize = pvd->vdev_children * sizeof (vdev_t *); 356 pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 357 newsize = pvd->vdev_children * sizeof (vdev_t *); 358 359 newchild = kmem_zalloc(newsize, KM_SLEEP); 360 if (pvd->vdev_child != NULL) { 361 bcopy(pvd->vdev_child, newchild, oldsize); 362 kmem_free(pvd->vdev_child, oldsize); 363 } 364 365 pvd->vdev_child = newchild; 366 pvd->vdev_child[id] = cvd; 367 368 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 369 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 370 371 /* 372 * Walk up all ancestors to update guid sum. 373 */ 374 for (; pvd != NULL; pvd = pvd->vdev_parent) 375 pvd->vdev_guid_sum += cvd->vdev_guid_sum; 376 377 if (cvd->vdev_ops->vdev_op_leaf) { 378 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); 379 cvd->vdev_spa->spa_leaf_list_gen++; 380 } 381 } 382 383 void 384 vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 385 { 386 int c; 387 uint_t id = cvd->vdev_id; 388 389 ASSERT(cvd->vdev_parent == pvd); 390 391 if (pvd == NULL) 392 return; 393 394 ASSERT(id < pvd->vdev_children); 395 ASSERT(pvd->vdev_child[id] == cvd); 396 397 pvd->vdev_child[id] = NULL; 398 cvd->vdev_parent = NULL; 399 400 for (c = 0; c < pvd->vdev_children; c++) 401 if (pvd->vdev_child[c]) 402 break; 403 404 if (c == pvd->vdev_children) { 405 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 406 pvd->vdev_child = NULL; 407 pvd->vdev_children = 0; 408 } 409 410 if (cvd->vdev_ops->vdev_op_leaf) { 411 spa_t *spa = cvd->vdev_spa; 412 list_remove(&spa->spa_leaf_list, cvd); 413 spa->spa_leaf_list_gen++; 414 } 415 416 /* 417 * Walk up all ancestors to update guid sum. 418 */ 419 for (; pvd != NULL; pvd = pvd->vdev_parent) 420 pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 421 } 422 423 /* 424 * Remove any holes in the child array. 425 */ 426 void 427 vdev_compact_children(vdev_t *pvd) 428 { 429 vdev_t **newchild, *cvd; 430 int oldc = pvd->vdev_children; 431 int newc; 432 433 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 434 435 for (int c = newc = 0; c < oldc; c++) 436 if (pvd->vdev_child[c]) 437 newc++; 438 439 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 440 441 for (int c = newc = 0; c < oldc; c++) { 442 if ((cvd = pvd->vdev_child[c]) != NULL) { 443 newchild[newc] = cvd; 444 cvd->vdev_id = newc++; 445 } 446 } 447 448 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 449 pvd->vdev_child = newchild; 450 pvd->vdev_children = newc; 451 } 452 453 /* 454 * Allocate and minimally initialize a vdev_t. 455 */ 456 vdev_t * 457 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 458 { 459 vdev_t *vd; 460 vdev_indirect_config_t *vic; 461 462 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 463 vic = &vd->vdev_indirect_config; 464 465 if (spa->spa_root_vdev == NULL) { 466 ASSERT(ops == &vdev_root_ops); 467 spa->spa_root_vdev = vd; 468 spa->spa_load_guid = spa_generate_guid(NULL); 469 } 470 471 if (guid == 0 && ops != &vdev_hole_ops) { 472 if (spa->spa_root_vdev == vd) { 473 /* 474 * The root vdev's guid will also be the pool guid, 475 * which must be unique among all pools. 476 */ 477 guid = spa_generate_guid(NULL); 478 } else { 479 /* 480 * Any other vdev's guid must be unique within the pool. 481 */ 482 guid = spa_generate_guid(spa); 483 } 484 ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 485 } 486 487 vd->vdev_spa = spa; 488 vd->vdev_id = id; 489 vd->vdev_guid = guid; 490 vd->vdev_guid_sum = guid; 491 vd->vdev_ops = ops; 492 vd->vdev_state = VDEV_STATE_CLOSED; 493 vd->vdev_ishole = (ops == &vdev_hole_ops); 494 vic->vic_prev_indirect_vdev = UINT64_MAX; 495 496 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 497 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 498 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL); 499 500 list_link_init(&vd->vdev_leaf_node); 501 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 502 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 503 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 504 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); 505 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); 506 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); 507 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); 508 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); 509 510 for (int t = 0; t < DTL_TYPES; t++) { 511 vd->vdev_dtl[t] = range_tree_create(NULL, NULL); 512 } 513 txg_list_create(&vd->vdev_ms_list, spa, 514 offsetof(struct metaslab, ms_txg_node)); 515 txg_list_create(&vd->vdev_dtl_list, spa, 516 offsetof(struct vdev, vdev_dtl_node)); 517 vd->vdev_stat.vs_timestamp = gethrtime(); 518 vdev_queue_init(vd); 519 vdev_cache_init(vd); 520 521 return (vd); 522 } 523 524 /* 525 * Allocate a new vdev. The 'alloctype' is used to control whether we are 526 * creating a new vdev or loading an existing one - the behavior is slightly 527 * different for each case. 528 */ 529 int 530 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 531 int alloctype) 532 { 533 vdev_ops_t *ops; 534 char *type; 535 uint64_t guid = 0, islog, nparity; 536 vdev_t *vd; 537 vdev_indirect_config_t *vic; 538 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 539 boolean_t top_level = (parent && !parent->vdev_parent); 540 541 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 542 543 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 544 return (SET_ERROR(EINVAL)); 545 546 if ((ops = vdev_getops(type)) == NULL) 547 return (SET_ERROR(EINVAL)); 548 549 /* 550 * If this is a load, get the vdev guid from the nvlist. 551 * Otherwise, vdev_alloc_common() will generate one for us. 552 */ 553 if (alloctype == VDEV_ALLOC_LOAD) { 554 uint64_t label_id; 555 556 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 557 label_id != id) 558 return (SET_ERROR(EINVAL)); 559 560 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 561 return (SET_ERROR(EINVAL)); 562 } else if (alloctype == VDEV_ALLOC_SPARE) { 563 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 564 return (SET_ERROR(EINVAL)); 565 } else if (alloctype == VDEV_ALLOC_L2CACHE) { 566 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 567 return (SET_ERROR(EINVAL)); 568 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 569 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 570 return (SET_ERROR(EINVAL)); 571 } 572 573 /* 574 * The first allocated vdev must be of type 'root'. 575 */ 576 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 577 return (SET_ERROR(EINVAL)); 578 579 /* 580 * Determine whether we're a log vdev. 581 */ 582 islog = 0; 583 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 584 if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 585 return (SET_ERROR(ENOTSUP)); 586 587 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 588 return (SET_ERROR(ENOTSUP)); 589 590 /* 591 * Set the nparity property for RAID-Z vdevs. 592 */ 593 nparity = -1ULL; 594 if (ops == &vdev_raidz_ops) { 595 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 596 &nparity) == 0) { 597 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 598 return (SET_ERROR(EINVAL)); 599 /* 600 * Previous versions could only support 1 or 2 parity 601 * device. 602 */ 603 if (nparity > 1 && 604 spa_version(spa) < SPA_VERSION_RAIDZ2) 605 return (SET_ERROR(ENOTSUP)); 606 if (nparity > 2 && 607 spa_version(spa) < SPA_VERSION_RAIDZ3) 608 return (SET_ERROR(ENOTSUP)); 609 } else { 610 /* 611 * We require the parity to be specified for SPAs that 612 * support multiple parity levels. 613 */ 614 if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 615 return (SET_ERROR(EINVAL)); 616 /* 617 * Otherwise, we default to 1 parity device for RAID-Z. 618 */ 619 nparity = 1; 620 } 621 } else { 622 nparity = 0; 623 } 624 ASSERT(nparity != -1ULL); 625 626 /* 627 * If creating a top-level vdev, check for allocation classes input 628 */ 629 if (top_level && alloctype == VDEV_ALLOC_ADD) { 630 char *bias; 631 632 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 633 &bias) == 0) { 634 alloc_bias = vdev_derive_alloc_bias(bias); 635 636 /* spa_vdev_add() expects feature to be enabled */ 637 if (spa->spa_load_state != SPA_LOAD_CREATE && 638 !spa_feature_is_enabled(spa, 639 SPA_FEATURE_ALLOCATION_CLASSES)) { 640 return (SET_ERROR(ENOTSUP)); 641 } 642 } 643 } 644 645 vd = vdev_alloc_common(spa, id, guid, ops); 646 vic = &vd->vdev_indirect_config; 647 648 vd->vdev_islog = islog; 649 vd->vdev_nparity = nparity; 650 if (top_level && alloc_bias != VDEV_BIAS_NONE) 651 vd->vdev_alloc_bias = alloc_bias; 652 653 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 654 vd->vdev_path = spa_strdup(vd->vdev_path); 655 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 656 vd->vdev_devid = spa_strdup(vd->vdev_devid); 657 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 658 &vd->vdev_physpath) == 0) 659 vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 660 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 661 vd->vdev_fru = spa_strdup(vd->vdev_fru); 662 663 /* 664 * Set the whole_disk property. If it's not specified, leave the value 665 * as -1. 666 */ 667 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 668 &vd->vdev_wholedisk) != 0) 669 vd->vdev_wholedisk = -1ULL; 670 671 ASSERT0(vic->vic_mapping_object); 672 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 673 &vic->vic_mapping_object); 674 ASSERT0(vic->vic_births_object); 675 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 676 &vic->vic_births_object); 677 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 678 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 679 &vic->vic_prev_indirect_vdev); 680 681 /* 682 * Look for the 'not present' flag. This will only be set if the device 683 * was not present at the time of import. 684 */ 685 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 686 &vd->vdev_not_present); 687 688 /* 689 * Get the alignment requirement. 690 */ 691 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 692 693 /* 694 * Retrieve the vdev creation time. 695 */ 696 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 697 &vd->vdev_crtxg); 698 699 /* 700 * If we're a top-level vdev, try to load the allocation parameters. 701 */ 702 if (top_level && 703 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 704 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 705 &vd->vdev_ms_array); 706 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 707 &vd->vdev_ms_shift); 708 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 709 &vd->vdev_asize); 710 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 711 &vd->vdev_removing); 712 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 713 &vd->vdev_top_zap); 714 } else { 715 ASSERT0(vd->vdev_top_zap); 716 } 717 718 if (top_level && alloctype != VDEV_ALLOC_ATTACH) { 719 ASSERT(alloctype == VDEV_ALLOC_LOAD || 720 alloctype == VDEV_ALLOC_ADD || 721 alloctype == VDEV_ALLOC_SPLIT || 722 alloctype == VDEV_ALLOC_ROOTPOOL); 723 /* Note: metaslab_group_create() is now deferred */ 724 } 725 726 if (vd->vdev_ops->vdev_op_leaf && 727 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 728 (void) nvlist_lookup_uint64(nv, 729 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 730 } else { 731 ASSERT0(vd->vdev_leaf_zap); 732 } 733 734 /* 735 * If we're a leaf vdev, try to load the DTL object and other state. 736 */ 737 738 if (vd->vdev_ops->vdev_op_leaf && 739 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 740 alloctype == VDEV_ALLOC_ROOTPOOL)) { 741 if (alloctype == VDEV_ALLOC_LOAD) { 742 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 743 &vd->vdev_dtl_object); 744 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 745 &vd->vdev_unspare); 746 } 747 748 if (alloctype == VDEV_ALLOC_ROOTPOOL) { 749 uint64_t spare = 0; 750 751 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 752 &spare) == 0 && spare) 753 spa_spare_add(vd); 754 } 755 756 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 757 &vd->vdev_offline); 758 759 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 760 &vd->vdev_resilver_txg); 761 762 /* 763 * When importing a pool, we want to ignore the persistent fault 764 * state, as the diagnosis made on another system may not be 765 * valid in the current context. Local vdevs will 766 * remain in the faulted state. 767 */ 768 if (spa_load_state(spa) == SPA_LOAD_OPEN) { 769 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 770 &vd->vdev_faulted); 771 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 772 &vd->vdev_degraded); 773 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 774 &vd->vdev_removed); 775 776 if (vd->vdev_faulted || vd->vdev_degraded) { 777 char *aux; 778 779 vd->vdev_label_aux = 780 VDEV_AUX_ERR_EXCEEDED; 781 if (nvlist_lookup_string(nv, 782 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 783 strcmp(aux, "external") == 0) 784 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 785 } 786 } 787 } 788 789 /* 790 * Add ourselves to the parent's list of children. 791 */ 792 vdev_add_child(parent, vd); 793 794 *vdp = vd; 795 796 return (0); 797 } 798 799 void 800 vdev_free(vdev_t *vd) 801 { 802 spa_t *spa = vd->vdev_spa; 803 ASSERT3P(vd->vdev_initialize_thread, ==, NULL); 804 805 /* 806 * Scan queues are normally destroyed at the end of a scan. If the 807 * queue exists here, that implies the vdev is being removed while 808 * the scan is still running. 809 */ 810 if (vd->vdev_scan_io_queue != NULL) { 811 mutex_enter(&vd->vdev_scan_io_queue_lock); 812 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); 813 vd->vdev_scan_io_queue = NULL; 814 mutex_exit(&vd->vdev_scan_io_queue_lock); 815 } 816 817 /* 818 * vdev_free() implies closing the vdev first. This is simpler than 819 * trying to ensure complicated semantics for all callers. 820 */ 821 vdev_close(vd); 822 823 ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 824 ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 825 826 /* 827 * Free all children. 828 */ 829 for (int c = 0; c < vd->vdev_children; c++) 830 vdev_free(vd->vdev_child[c]); 831 832 ASSERT(vd->vdev_child == NULL); 833 ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 834 ASSERT(vd->vdev_initialize_thread == NULL); 835 836 /* 837 * Discard allocation state. 838 */ 839 if (vd->vdev_mg != NULL) { 840 vdev_metaslab_fini(vd); 841 metaslab_group_destroy(vd->vdev_mg); 842 } 843 844 ASSERT0(vd->vdev_stat.vs_space); 845 ASSERT0(vd->vdev_stat.vs_dspace); 846 ASSERT0(vd->vdev_stat.vs_alloc); 847 848 /* 849 * Remove this vdev from its parent's child list. 850 */ 851 vdev_remove_child(vd->vdev_parent, vd); 852 853 ASSERT(vd->vdev_parent == NULL); 854 ASSERT(!list_link_active(&vd->vdev_leaf_node)); 855 856 /* 857 * Clean up vdev structure. 858 */ 859 vdev_queue_fini(vd); 860 vdev_cache_fini(vd); 861 862 if (vd->vdev_path) 863 spa_strfree(vd->vdev_path); 864 if (vd->vdev_devid) 865 spa_strfree(vd->vdev_devid); 866 if (vd->vdev_physpath) 867 spa_strfree(vd->vdev_physpath); 868 if (vd->vdev_fru) 869 spa_strfree(vd->vdev_fru); 870 871 if (vd->vdev_isspare) 872 spa_spare_remove(vd); 873 if (vd->vdev_isl2cache) 874 spa_l2cache_remove(vd); 875 876 txg_list_destroy(&vd->vdev_ms_list); 877 txg_list_destroy(&vd->vdev_dtl_list); 878 879 mutex_enter(&vd->vdev_dtl_lock); 880 space_map_close(vd->vdev_dtl_sm); 881 for (int t = 0; t < DTL_TYPES; t++) { 882 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 883 range_tree_destroy(vd->vdev_dtl[t]); 884 } 885 mutex_exit(&vd->vdev_dtl_lock); 886 887 EQUIV(vd->vdev_indirect_births != NULL, 888 vd->vdev_indirect_mapping != NULL); 889 if (vd->vdev_indirect_births != NULL) { 890 vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 891 vdev_indirect_births_close(vd->vdev_indirect_births); 892 } 893 894 if (vd->vdev_obsolete_sm != NULL) { 895 ASSERT(vd->vdev_removing || 896 vd->vdev_ops == &vdev_indirect_ops); 897 space_map_close(vd->vdev_obsolete_sm); 898 vd->vdev_obsolete_sm = NULL; 899 } 900 range_tree_destroy(vd->vdev_obsolete_segments); 901 rw_destroy(&vd->vdev_indirect_rwlock); 902 mutex_destroy(&vd->vdev_obsolete_lock); 903 904 mutex_destroy(&vd->vdev_dtl_lock); 905 mutex_destroy(&vd->vdev_stat_lock); 906 mutex_destroy(&vd->vdev_probe_lock); 907 mutex_destroy(&vd->vdev_scan_io_queue_lock); 908 mutex_destroy(&vd->vdev_initialize_lock); 909 mutex_destroy(&vd->vdev_initialize_io_lock); 910 cv_destroy(&vd->vdev_initialize_io_cv); 911 cv_destroy(&vd->vdev_initialize_cv); 912 913 if (vd == spa->spa_root_vdev) 914 spa->spa_root_vdev = NULL; 915 916 kmem_free(vd, sizeof (vdev_t)); 917 } 918 919 /* 920 * Transfer top-level vdev state from svd to tvd. 921 */ 922 static void 923 vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 924 { 925 spa_t *spa = svd->vdev_spa; 926 metaslab_t *msp; 927 vdev_t *vd; 928 int t; 929 930 ASSERT(tvd == tvd->vdev_top); 931 932 tvd->vdev_ms_array = svd->vdev_ms_array; 933 tvd->vdev_ms_shift = svd->vdev_ms_shift; 934 tvd->vdev_ms_count = svd->vdev_ms_count; 935 tvd->vdev_top_zap = svd->vdev_top_zap; 936 937 svd->vdev_ms_array = 0; 938 svd->vdev_ms_shift = 0; 939 svd->vdev_ms_count = 0; 940 svd->vdev_top_zap = 0; 941 942 if (tvd->vdev_mg) 943 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 944 tvd->vdev_mg = svd->vdev_mg; 945 tvd->vdev_ms = svd->vdev_ms; 946 947 svd->vdev_mg = NULL; 948 svd->vdev_ms = NULL; 949 950 if (tvd->vdev_mg != NULL) 951 tvd->vdev_mg->mg_vd = tvd; 952 953 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 954 svd->vdev_checkpoint_sm = NULL; 955 956 tvd->vdev_alloc_bias = svd->vdev_alloc_bias; 957 svd->vdev_alloc_bias = VDEV_BIAS_NONE; 958 959 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 960 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 961 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 962 963 svd->vdev_stat.vs_alloc = 0; 964 svd->vdev_stat.vs_space = 0; 965 svd->vdev_stat.vs_dspace = 0; 966 967 /* 968 * State which may be set on a top-level vdev that's in the 969 * process of being removed. 970 */ 971 ASSERT0(tvd->vdev_indirect_config.vic_births_object); 972 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); 973 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); 974 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); 975 ASSERT3P(tvd->vdev_indirect_births, ==, NULL); 976 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); 977 ASSERT0(tvd->vdev_removing); 978 tvd->vdev_removing = svd->vdev_removing; 979 tvd->vdev_indirect_config = svd->vdev_indirect_config; 980 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; 981 tvd->vdev_indirect_births = svd->vdev_indirect_births; 982 range_tree_swap(&svd->vdev_obsolete_segments, 983 &tvd->vdev_obsolete_segments); 984 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; 985 svd->vdev_indirect_config.vic_mapping_object = 0; 986 svd->vdev_indirect_config.vic_births_object = 0; 987 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; 988 svd->vdev_indirect_mapping = NULL; 989 svd->vdev_indirect_births = NULL; 990 svd->vdev_obsolete_sm = NULL; 991 svd->vdev_removing = 0; 992 993 for (t = 0; t < TXG_SIZE; t++) { 994 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 995 (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 996 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 997 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 998 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 999 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 1000 } 1001 1002 if (list_link_active(&svd->vdev_config_dirty_node)) { 1003 vdev_config_clean(svd); 1004 vdev_config_dirty(tvd); 1005 } 1006 1007 if (list_link_active(&svd->vdev_state_dirty_node)) { 1008 vdev_state_clean(svd); 1009 vdev_state_dirty(tvd); 1010 } 1011 1012 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 1013 svd->vdev_deflate_ratio = 0; 1014 1015 tvd->vdev_islog = svd->vdev_islog; 1016 svd->vdev_islog = 0; 1017 1018 dsl_scan_io_queue_vdev_xfer(svd, tvd); 1019 } 1020 1021 static void 1022 vdev_top_update(vdev_t *tvd, vdev_t *vd) 1023 { 1024 if (vd == NULL) 1025 return; 1026 1027 vd->vdev_top = tvd; 1028 1029 for (int c = 0; c < vd->vdev_children; c++) 1030 vdev_top_update(tvd, vd->vdev_child[c]); 1031 } 1032 1033 /* 1034 * Add a mirror/replacing vdev above an existing vdev. 1035 */ 1036 vdev_t * 1037 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 1038 { 1039 spa_t *spa = cvd->vdev_spa; 1040 vdev_t *pvd = cvd->vdev_parent; 1041 vdev_t *mvd; 1042 1043 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1044 1045 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 1046 1047 mvd->vdev_asize = cvd->vdev_asize; 1048 mvd->vdev_min_asize = cvd->vdev_min_asize; 1049 mvd->vdev_max_asize = cvd->vdev_max_asize; 1050 mvd->vdev_psize = cvd->vdev_psize; 1051 mvd->vdev_ashift = cvd->vdev_ashift; 1052 mvd->vdev_state = cvd->vdev_state; 1053 mvd->vdev_crtxg = cvd->vdev_crtxg; 1054 1055 vdev_remove_child(pvd, cvd); 1056 vdev_add_child(pvd, mvd); 1057 cvd->vdev_id = mvd->vdev_children; 1058 vdev_add_child(mvd, cvd); 1059 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1060 1061 if (mvd == mvd->vdev_top) 1062 vdev_top_transfer(cvd, mvd); 1063 1064 return (mvd); 1065 } 1066 1067 /* 1068 * Remove a 1-way mirror/replacing vdev from the tree. 1069 */ 1070 void 1071 vdev_remove_parent(vdev_t *cvd) 1072 { 1073 vdev_t *mvd = cvd->vdev_parent; 1074 vdev_t *pvd = mvd->vdev_parent; 1075 1076 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1077 1078 ASSERT(mvd->vdev_children == 1); 1079 ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 1080 mvd->vdev_ops == &vdev_replacing_ops || 1081 mvd->vdev_ops == &vdev_spare_ops); 1082 cvd->vdev_ashift = mvd->vdev_ashift; 1083 1084 vdev_remove_child(mvd, cvd); 1085 vdev_remove_child(pvd, mvd); 1086 1087 /* 1088 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 1089 * Otherwise, we could have detached an offline device, and when we 1090 * go to import the pool we'll think we have two top-level vdevs, 1091 * instead of a different version of the same top-level vdev. 1092 */ 1093 if (mvd->vdev_top == mvd) { 1094 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 1095 cvd->vdev_orig_guid = cvd->vdev_guid; 1096 cvd->vdev_guid += guid_delta; 1097 cvd->vdev_guid_sum += guid_delta; 1098 } 1099 cvd->vdev_id = mvd->vdev_id; 1100 vdev_add_child(pvd, cvd); 1101 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1102 1103 if (cvd == cvd->vdev_top) 1104 vdev_top_transfer(mvd, cvd); 1105 1106 ASSERT(mvd->vdev_children == 0); 1107 vdev_free(mvd); 1108 } 1109 1110 static void 1111 vdev_metaslab_group_create(vdev_t *vd) 1112 { 1113 spa_t *spa = vd->vdev_spa; 1114 1115 /* 1116 * metaslab_group_create was delayed until allocation bias was available 1117 */ 1118 if (vd->vdev_mg == NULL) { 1119 metaslab_class_t *mc; 1120 1121 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE) 1122 vd->vdev_alloc_bias = VDEV_BIAS_LOG; 1123 1124 ASSERT3U(vd->vdev_islog, ==, 1125 (vd->vdev_alloc_bias == VDEV_BIAS_LOG)); 1126 1127 switch (vd->vdev_alloc_bias) { 1128 case VDEV_BIAS_LOG: 1129 mc = spa_log_class(spa); 1130 break; 1131 case VDEV_BIAS_SPECIAL: 1132 mc = spa_special_class(spa); 1133 break; 1134 case VDEV_BIAS_DEDUP: 1135 mc = spa_dedup_class(spa); 1136 break; 1137 default: 1138 mc = spa_normal_class(spa); 1139 } 1140 1141 vd->vdev_mg = metaslab_group_create(mc, vd, 1142 spa->spa_alloc_count); 1143 1144 /* 1145 * The spa ashift values currently only reflect the 1146 * general vdev classes. Class destination is late 1147 * binding so ashift checking had to wait until now 1148 */ 1149 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1150 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) { 1151 if (vd->vdev_ashift > spa->spa_max_ashift) 1152 spa->spa_max_ashift = vd->vdev_ashift; 1153 if (vd->vdev_ashift < spa->spa_min_ashift) 1154 spa->spa_min_ashift = vd->vdev_ashift; 1155 } 1156 } 1157 } 1158 1159 int 1160 vdev_metaslab_init(vdev_t *vd, uint64_t txg) 1161 { 1162 spa_t *spa = vd->vdev_spa; 1163 objset_t *mos = spa->spa_meta_objset; 1164 uint64_t m; 1165 uint64_t oldc = vd->vdev_ms_count; 1166 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 1167 metaslab_t **mspp; 1168 int error; 1169 boolean_t expanding = (oldc != 0); 1170 1171 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1172 1173 /* 1174 * This vdev is not being allocated from yet or is a hole. 1175 */ 1176 if (vd->vdev_ms_shift == 0) 1177 return (0); 1178 1179 ASSERT(!vd->vdev_ishole); 1180 1181 ASSERT(oldc <= newc); 1182 1183 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1184 1185 if (expanding) { 1186 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 1187 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1188 } 1189 1190 vd->vdev_ms = mspp; 1191 vd->vdev_ms_count = newc; 1192 for (m = oldc; m < newc; m++) { 1193 uint64_t object = 0; 1194 1195 /* 1196 * vdev_ms_array may be 0 if we are creating the "fake" 1197 * metaslabs for an indirect vdev for zdb's leak detection. 1198 * See zdb_leak_init(). 1199 */ 1200 if (txg == 0 && vd->vdev_ms_array != 0) { 1201 error = dmu_read(mos, vd->vdev_ms_array, 1202 m * sizeof (uint64_t), sizeof (uint64_t), &object, 1203 DMU_READ_PREFETCH); 1204 if (error != 0) { 1205 vdev_dbgmsg(vd, "unable to read the metaslab " 1206 "array [error=%d]", error); 1207 return (error); 1208 } 1209 } 1210 1211 #ifndef _KERNEL 1212 /* 1213 * To accomodate zdb_leak_init() fake indirect 1214 * metaslabs, we allocate a metaslab group for 1215 * indirect vdevs which normally don't have one. 1216 */ 1217 if (vd->vdev_mg == NULL) { 1218 ASSERT0(vdev_is_concrete(vd)); 1219 vdev_metaslab_group_create(vd); 1220 } 1221 #endif 1222 error = metaslab_init(vd->vdev_mg, m, object, txg, 1223 &(vd->vdev_ms[m])); 1224 if (error != 0) { 1225 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 1226 error); 1227 return (error); 1228 } 1229 } 1230 1231 if (txg == 0) 1232 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1233 1234 /* 1235 * If the vdev is being removed we don't activate 1236 * the metaslabs since we want to ensure that no new 1237 * allocations are performed on this device. 1238 */ 1239 if (!expanding && !vd->vdev_removing) { 1240 metaslab_group_activate(vd->vdev_mg); 1241 } 1242 1243 if (txg == 0) 1244 spa_config_exit(spa, SCL_ALLOC, FTAG); 1245 1246 return (0); 1247 } 1248 1249 void 1250 vdev_metaslab_fini(vdev_t *vd) 1251 { 1252 if (vd->vdev_checkpoint_sm != NULL) { 1253 ASSERT(spa_feature_is_active(vd->vdev_spa, 1254 SPA_FEATURE_POOL_CHECKPOINT)); 1255 space_map_close(vd->vdev_checkpoint_sm); 1256 /* 1257 * Even though we close the space map, we need to set its 1258 * pointer to NULL. The reason is that vdev_metaslab_fini() 1259 * may be called multiple times for certain operations 1260 * (i.e. when destroying a pool) so we need to ensure that 1261 * this clause never executes twice. This logic is similar 1262 * to the one used for the vdev_ms clause below. 1263 */ 1264 vd->vdev_checkpoint_sm = NULL; 1265 } 1266 1267 if (vd->vdev_ms != NULL) { 1268 metaslab_group_t *mg = vd->vdev_mg; 1269 metaslab_group_passivate(mg); 1270 1271 uint64_t count = vd->vdev_ms_count; 1272 for (uint64_t m = 0; m < count; m++) { 1273 metaslab_t *msp = vd->vdev_ms[m]; 1274 if (msp != NULL) 1275 metaslab_fini(msp); 1276 } 1277 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1278 vd->vdev_ms = NULL; 1279 1280 vd->vdev_ms_count = 0; 1281 1282 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 1283 ASSERT0(mg->mg_histogram[i]); 1284 } 1285 ASSERT0(vd->vdev_ms_count); 1286 } 1287 1288 typedef struct vdev_probe_stats { 1289 boolean_t vps_readable; 1290 boolean_t vps_writeable; 1291 int vps_flags; 1292 } vdev_probe_stats_t; 1293 1294 static void 1295 vdev_probe_done(zio_t *zio) 1296 { 1297 spa_t *spa = zio->io_spa; 1298 vdev_t *vd = zio->io_vd; 1299 vdev_probe_stats_t *vps = zio->io_private; 1300 1301 ASSERT(vd->vdev_probe_zio != NULL); 1302 1303 if (zio->io_type == ZIO_TYPE_READ) { 1304 if (zio->io_error == 0) 1305 vps->vps_readable = 1; 1306 if (zio->io_error == 0 && spa_writeable(spa)) { 1307 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1308 zio->io_offset, zio->io_size, zio->io_abd, 1309 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1310 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1311 } else { 1312 abd_free(zio->io_abd); 1313 } 1314 } else if (zio->io_type == ZIO_TYPE_WRITE) { 1315 if (zio->io_error == 0) 1316 vps->vps_writeable = 1; 1317 abd_free(zio->io_abd); 1318 } else if (zio->io_type == ZIO_TYPE_NULL) { 1319 zio_t *pio; 1320 1321 vd->vdev_cant_read |= !vps->vps_readable; 1322 vd->vdev_cant_write |= !vps->vps_writeable; 1323 1324 if (vdev_readable(vd) && 1325 (vdev_writeable(vd) || !spa_writeable(spa))) { 1326 zio->io_error = 0; 1327 } else { 1328 ASSERT(zio->io_error != 0); 1329 vdev_dbgmsg(vd, "failed probe"); 1330 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 1331 spa, vd, NULL, 0, 0); 1332 zio->io_error = SET_ERROR(ENXIO); 1333 } 1334 1335 mutex_enter(&vd->vdev_probe_lock); 1336 ASSERT(vd->vdev_probe_zio == zio); 1337 vd->vdev_probe_zio = NULL; 1338 mutex_exit(&vd->vdev_probe_lock); 1339 1340 zio_link_t *zl = NULL; 1341 while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1342 if (!vdev_accessible(vd, pio)) 1343 pio->io_error = SET_ERROR(ENXIO); 1344 1345 kmem_free(vps, sizeof (*vps)); 1346 } 1347 } 1348 1349 /* 1350 * Determine whether this device is accessible. 1351 * 1352 * Read and write to several known locations: the pad regions of each 1353 * vdev label but the first, which we leave alone in case it contains 1354 * a VTOC. 1355 */ 1356 zio_t * 1357 vdev_probe(vdev_t *vd, zio_t *zio) 1358 { 1359 spa_t *spa = vd->vdev_spa; 1360 vdev_probe_stats_t *vps = NULL; 1361 zio_t *pio; 1362 1363 ASSERT(vd->vdev_ops->vdev_op_leaf); 1364 1365 /* 1366 * Don't probe the probe. 1367 */ 1368 if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1369 return (NULL); 1370 1371 /* 1372 * To prevent 'probe storms' when a device fails, we create 1373 * just one probe i/o at a time. All zios that want to probe 1374 * this vdev will become parents of the probe io. 1375 */ 1376 mutex_enter(&vd->vdev_probe_lock); 1377 1378 if ((pio = vd->vdev_probe_zio) == NULL) { 1379 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1380 1381 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1382 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 1383 ZIO_FLAG_TRYHARD; 1384 1385 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1386 /* 1387 * vdev_cant_read and vdev_cant_write can only 1388 * transition from TRUE to FALSE when we have the 1389 * SCL_ZIO lock as writer; otherwise they can only 1390 * transition from FALSE to TRUE. This ensures that 1391 * any zio looking at these values can assume that 1392 * failures persist for the life of the I/O. That's 1393 * important because when a device has intermittent 1394 * connectivity problems, we want to ensure that 1395 * they're ascribed to the device (ENXIO) and not 1396 * the zio (EIO). 1397 * 1398 * Since we hold SCL_ZIO as writer here, clear both 1399 * values so the probe can reevaluate from first 1400 * principles. 1401 */ 1402 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1403 vd->vdev_cant_read = B_FALSE; 1404 vd->vdev_cant_write = B_FALSE; 1405 } 1406 1407 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1408 vdev_probe_done, vps, 1409 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1410 1411 /* 1412 * We can't change the vdev state in this context, so we 1413 * kick off an async task to do it on our behalf. 1414 */ 1415 if (zio != NULL) { 1416 vd->vdev_probe_wanted = B_TRUE; 1417 spa_async_request(spa, SPA_ASYNC_PROBE); 1418 } 1419 } 1420 1421 if (zio != NULL) 1422 zio_add_child(zio, pio); 1423 1424 mutex_exit(&vd->vdev_probe_lock); 1425 1426 if (vps == NULL) { 1427 ASSERT(zio != NULL); 1428 return (NULL); 1429 } 1430 1431 for (int l = 1; l < VDEV_LABELS; l++) { 1432 zio_nowait(zio_read_phys(pio, vd, 1433 vdev_label_offset(vd->vdev_psize, l, 1434 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 1435 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1436 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1437 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1438 } 1439 1440 if (zio == NULL) 1441 return (pio); 1442 1443 zio_nowait(pio); 1444 return (NULL); 1445 } 1446 1447 static void 1448 vdev_open_child(void *arg) 1449 { 1450 vdev_t *vd = arg; 1451 1452 vd->vdev_open_thread = curthread; 1453 vd->vdev_open_error = vdev_open(vd); 1454 vd->vdev_open_thread = NULL; 1455 } 1456 1457 boolean_t 1458 vdev_uses_zvols(vdev_t *vd) 1459 { 1460 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1461 strlen(ZVOL_DIR)) == 0) 1462 return (B_TRUE); 1463 for (int c = 0; c < vd->vdev_children; c++) 1464 if (vdev_uses_zvols(vd->vdev_child[c])) 1465 return (B_TRUE); 1466 return (B_FALSE); 1467 } 1468 1469 void 1470 vdev_open_children(vdev_t *vd) 1471 { 1472 taskq_t *tq; 1473 int children = vd->vdev_children; 1474 1475 /* 1476 * in order to handle pools on top of zvols, do the opens 1477 * in a single thread so that the same thread holds the 1478 * spa_namespace_lock 1479 */ 1480 if (vdev_uses_zvols(vd)) { 1481 for (int c = 0; c < children; c++) 1482 vd->vdev_child[c]->vdev_open_error = 1483 vdev_open(vd->vdev_child[c]); 1484 return; 1485 } 1486 tq = taskq_create("vdev_open", children, minclsyspri, 1487 children, children, TASKQ_PREPOPULATE); 1488 1489 for (int c = 0; c < children; c++) 1490 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1491 TQ_SLEEP) != TASKQID_INVALID); 1492 1493 taskq_destroy(tq); 1494 } 1495 1496 /* 1497 * Compute the raidz-deflation ratio. Note, we hard-code 1498 * in 128k (1 << 17) because it is the "typical" blocksize. 1499 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 1500 * otherwise it would inconsistently account for existing bp's. 1501 */ 1502 static void 1503 vdev_set_deflate_ratio(vdev_t *vd) 1504 { 1505 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 1506 vd->vdev_deflate_ratio = (1 << 17) / 1507 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 1508 } 1509 } 1510 1511 /* 1512 * Prepare a virtual device for access. 1513 */ 1514 int 1515 vdev_open(vdev_t *vd) 1516 { 1517 spa_t *spa = vd->vdev_spa; 1518 int error; 1519 uint64_t osize = 0; 1520 uint64_t max_osize = 0; 1521 uint64_t asize, max_asize, psize; 1522 uint64_t ashift = 0; 1523 1524 ASSERT(vd->vdev_open_thread == curthread || 1525 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1526 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1527 vd->vdev_state == VDEV_STATE_CANT_OPEN || 1528 vd->vdev_state == VDEV_STATE_OFFLINE); 1529 1530 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1531 vd->vdev_cant_read = B_FALSE; 1532 vd->vdev_cant_write = B_FALSE; 1533 vd->vdev_min_asize = vdev_get_min_asize(vd); 1534 1535 /* 1536 * If this vdev is not removed, check its fault status. If it's 1537 * faulted, bail out of the open. 1538 */ 1539 if (!vd->vdev_removed && vd->vdev_faulted) { 1540 ASSERT(vd->vdev_children == 0); 1541 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1542 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1543 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1544 vd->vdev_label_aux); 1545 return (SET_ERROR(ENXIO)); 1546 } else if (vd->vdev_offline) { 1547 ASSERT(vd->vdev_children == 0); 1548 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1549 return (SET_ERROR(ENXIO)); 1550 } 1551 1552 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1553 1554 /* 1555 * Reset the vdev_reopening flag so that we actually close 1556 * the vdev on error. 1557 */ 1558 vd->vdev_reopening = B_FALSE; 1559 if (zio_injection_enabled && error == 0) 1560 error = zio_handle_device_injection(vd, NULL, ENXIO); 1561 1562 if (error) { 1563 if (vd->vdev_removed && 1564 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1565 vd->vdev_removed = B_FALSE; 1566 1567 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 1568 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 1569 vd->vdev_stat.vs_aux); 1570 } else { 1571 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1572 vd->vdev_stat.vs_aux); 1573 } 1574 return (error); 1575 } 1576 1577 vd->vdev_removed = B_FALSE; 1578 1579 /* 1580 * Recheck the faulted flag now that we have confirmed that 1581 * the vdev is accessible. If we're faulted, bail. 1582 */ 1583 if (vd->vdev_faulted) { 1584 ASSERT(vd->vdev_children == 0); 1585 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1586 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1587 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1588 vd->vdev_label_aux); 1589 return (SET_ERROR(ENXIO)); 1590 } 1591 1592 if (vd->vdev_degraded) { 1593 ASSERT(vd->vdev_children == 0); 1594 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1595 VDEV_AUX_ERR_EXCEEDED); 1596 } else { 1597 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 1598 } 1599 1600 /* 1601 * For hole or missing vdevs we just return success. 1602 */ 1603 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 1604 return (0); 1605 1606 for (int c = 0; c < vd->vdev_children; c++) { 1607 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1608 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1609 VDEV_AUX_NONE); 1610 break; 1611 } 1612 } 1613 1614 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1615 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1616 1617 if (vd->vdev_children == 0) { 1618 if (osize < SPA_MINDEVSIZE) { 1619 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1620 VDEV_AUX_TOO_SMALL); 1621 return (SET_ERROR(EOVERFLOW)); 1622 } 1623 psize = osize; 1624 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1625 max_asize = max_osize - (VDEV_LABEL_START_SIZE + 1626 VDEV_LABEL_END_SIZE); 1627 } else { 1628 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1629 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1630 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1631 VDEV_AUX_TOO_SMALL); 1632 return (SET_ERROR(EOVERFLOW)); 1633 } 1634 psize = 0; 1635 asize = osize; 1636 max_asize = max_osize; 1637 } 1638 1639 vd->vdev_psize = psize; 1640 1641 /* 1642 * Make sure the allocatable size hasn't shrunk too much. 1643 */ 1644 if (asize < vd->vdev_min_asize) { 1645 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1646 VDEV_AUX_BAD_LABEL); 1647 return (SET_ERROR(EINVAL)); 1648 } 1649 1650 if (vd->vdev_asize == 0) { 1651 /* 1652 * This is the first-ever open, so use the computed values. 1653 * For testing purposes, a higher ashift can be requested. 1654 */ 1655 vd->vdev_asize = asize; 1656 vd->vdev_max_asize = max_asize; 1657 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1658 vd->vdev_ashift = MAX(zfs_ashift_min, vd->vdev_ashift); 1659 } else { 1660 /* 1661 * Detect if the alignment requirement has increased. 1662 * We don't want to make the pool unavailable, just 1663 * issue a warning instead. 1664 */ 1665 if (ashift > vd->vdev_top->vdev_ashift && 1666 vd->vdev_ops->vdev_op_leaf) { 1667 cmn_err(CE_WARN, 1668 "Disk, '%s', has a block alignment that is " 1669 "larger than the pool's alignment\n", 1670 vd->vdev_path); 1671 } 1672 vd->vdev_max_asize = max_asize; 1673 } 1674 1675 /* 1676 * If all children are healthy we update asize if either: 1677 * The asize has increased, due to a device expansion caused by dynamic 1678 * LUN growth or vdev replacement, and automatic expansion is enabled; 1679 * making the additional space available. 1680 * 1681 * The asize has decreased, due to a device shrink usually caused by a 1682 * vdev replace with a smaller device. This ensures that calculations 1683 * based of max_asize and asize e.g. esize are always valid. It's safe 1684 * to do this as we've already validated that asize is greater than 1685 * vdev_min_asize. 1686 */ 1687 if (vd->vdev_state == VDEV_STATE_HEALTHY && 1688 ((asize > vd->vdev_asize && 1689 (vd->vdev_expanding || spa->spa_autoexpand)) || 1690 (asize < vd->vdev_asize))) 1691 vd->vdev_asize = asize; 1692 1693 vdev_set_min_asize(vd); 1694 1695 /* 1696 * Ensure we can issue some IO before declaring the 1697 * vdev open for business. 1698 */ 1699 if (vd->vdev_ops->vdev_op_leaf && 1700 (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1701 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1702 VDEV_AUX_ERR_EXCEEDED); 1703 return (error); 1704 } 1705 1706 /* 1707 * Track the min and max ashift values for normal data devices. 1708 * 1709 * DJB - TBD these should perhaps be tracked per allocation class 1710 * (e.g. spa_min_ashift is used to round up post compression buffers) 1711 */ 1712 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1713 vd->vdev_alloc_bias == VDEV_BIAS_NONE && 1714 vd->vdev_aux == NULL) { 1715 if (vd->vdev_ashift > spa->spa_max_ashift) 1716 spa->spa_max_ashift = vd->vdev_ashift; 1717 if (vd->vdev_ashift < spa->spa_min_ashift) 1718 spa->spa_min_ashift = vd->vdev_ashift; 1719 } 1720 1721 /* 1722 * If a leaf vdev has a DTL, and seems healthy, then kick off a 1723 * resilver. But don't do this if we are doing a reopen for a scrub, 1724 * since this would just restart the scrub we are already doing. 1725 */ 1726 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1727 vdev_resilver_needed(vd, NULL, NULL)) 1728 spa_async_request(spa, SPA_ASYNC_RESILVER); 1729 1730 return (0); 1731 } 1732 1733 /* 1734 * Called once the vdevs are all opened, this routine validates the label 1735 * contents. This needs to be done before vdev_load() so that we don't 1736 * inadvertently do repair I/Os to the wrong device. 1737 * 1738 * This function will only return failure if one of the vdevs indicates that it 1739 * has since been destroyed or exported. This is only possible if 1740 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1741 * will be updated but the function will return 0. 1742 */ 1743 int 1744 vdev_validate(vdev_t *vd) 1745 { 1746 spa_t *spa = vd->vdev_spa; 1747 nvlist_t *label; 1748 uint64_t guid = 0, aux_guid = 0, top_guid; 1749 uint64_t state; 1750 nvlist_t *nvl; 1751 uint64_t txg; 1752 1753 if (vdev_validate_skip) 1754 return (0); 1755 1756 for (uint64_t c = 0; c < vd->vdev_children; c++) 1757 if (vdev_validate(vd->vdev_child[c]) != 0) 1758 return (SET_ERROR(EBADF)); 1759 1760 /* 1761 * If the device has already failed, or was marked offline, don't do 1762 * any further validation. Otherwise, label I/O will fail and we will 1763 * overwrite the previous state. 1764 */ 1765 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 1766 return (0); 1767 1768 /* 1769 * If we are performing an extreme rewind, we allow for a label that 1770 * was modified at a point after the current txg. 1771 * If config lock is not held do not check for the txg. spa_sync could 1772 * be updating the vdev's label before updating spa_last_synced_txg. 1773 */ 1774 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 1775 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 1776 txg = UINT64_MAX; 1777 else 1778 txg = spa_last_synced_txg(spa); 1779 1780 if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1781 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1782 VDEV_AUX_BAD_LABEL); 1783 vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 1784 "txg %llu", (u_longlong_t)txg); 1785 return (0); 1786 } 1787 1788 /* 1789 * Determine if this vdev has been split off into another 1790 * pool. If so, then refuse to open it. 1791 */ 1792 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1793 &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1794 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1795 VDEV_AUX_SPLIT_POOL); 1796 nvlist_free(label); 1797 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 1798 return (0); 1799 } 1800 1801 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 1802 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1803 VDEV_AUX_CORRUPT_DATA); 1804 nvlist_free(label); 1805 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1806 ZPOOL_CONFIG_POOL_GUID); 1807 return (0); 1808 } 1809 1810 /* 1811 * If config is not trusted then ignore the spa guid check. This is 1812 * necessary because if the machine crashed during a re-guid the new 1813 * guid might have been written to all of the vdev labels, but not the 1814 * cached config. The check will be performed again once we have the 1815 * trusted config from the MOS. 1816 */ 1817 if (spa->spa_trust_config && guid != spa_guid(spa)) { 1818 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1819 VDEV_AUX_CORRUPT_DATA); 1820 nvlist_free(label); 1821 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 1822 "match config (%llu != %llu)", (u_longlong_t)guid, 1823 (u_longlong_t)spa_guid(spa)); 1824 return (0); 1825 } 1826 1827 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1828 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1829 &aux_guid) != 0) 1830 aux_guid = 0; 1831 1832 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 1833 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1834 VDEV_AUX_CORRUPT_DATA); 1835 nvlist_free(label); 1836 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1837 ZPOOL_CONFIG_GUID); 1838 return (0); 1839 } 1840 1841 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 1842 != 0) { 1843 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1844 VDEV_AUX_CORRUPT_DATA); 1845 nvlist_free(label); 1846 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1847 ZPOOL_CONFIG_TOP_GUID); 1848 return (0); 1849 } 1850 1851 /* 1852 * If this vdev just became a top-level vdev because its sibling was 1853 * detached, it will have adopted the parent's vdev guid -- but the 1854 * label may or may not be on disk yet. Fortunately, either version 1855 * of the label will have the same top guid, so if we're a top-level 1856 * vdev, we can safely compare to that instead. 1857 * However, if the config comes from a cachefile that failed to update 1858 * after the detach, a top-level vdev will appear as a non top-level 1859 * vdev in the config. Also relax the constraints if we perform an 1860 * extreme rewind. 1861 * 1862 * If we split this vdev off instead, then we also check the 1863 * original pool's guid. We don't want to consider the vdev 1864 * corrupt if it is partway through a split operation. 1865 */ 1866 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 1867 boolean_t mismatch = B_FALSE; 1868 if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 1869 if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 1870 mismatch = B_TRUE; 1871 } else { 1872 if (vd->vdev_guid != top_guid && 1873 vd->vdev_top->vdev_guid != guid) 1874 mismatch = B_TRUE; 1875 } 1876 1877 if (mismatch) { 1878 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1879 VDEV_AUX_CORRUPT_DATA); 1880 nvlist_free(label); 1881 vdev_dbgmsg(vd, "vdev_validate: config guid " 1882 "doesn't match label guid"); 1883 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 1884 (u_longlong_t)vd->vdev_guid, 1885 (u_longlong_t)vd->vdev_top->vdev_guid); 1886 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 1887 "aux_guid %llu", (u_longlong_t)guid, 1888 (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1889 return (0); 1890 } 1891 } 1892 1893 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1894 &state) != 0) { 1895 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1896 VDEV_AUX_CORRUPT_DATA); 1897 nvlist_free(label); 1898 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1899 ZPOOL_CONFIG_POOL_STATE); 1900 return (0); 1901 } 1902 1903 nvlist_free(label); 1904 1905 /* 1906 * If this is a verbatim import, no need to check the 1907 * state of the pool. 1908 */ 1909 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1910 spa_load_state(spa) == SPA_LOAD_OPEN && 1911 state != POOL_STATE_ACTIVE) { 1912 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 1913 "for spa %s", (u_longlong_t)state, spa->spa_name); 1914 return (SET_ERROR(EBADF)); 1915 } 1916 1917 /* 1918 * If we were able to open and validate a vdev that was 1919 * previously marked permanently unavailable, clear that state 1920 * now. 1921 */ 1922 if (vd->vdev_not_present) 1923 vd->vdev_not_present = 0; 1924 1925 return (0); 1926 } 1927 1928 static void 1929 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 1930 { 1931 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 1932 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 1933 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 1934 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 1935 dvd->vdev_path, svd->vdev_path); 1936 spa_strfree(dvd->vdev_path); 1937 dvd->vdev_path = spa_strdup(svd->vdev_path); 1938 } 1939 } else if (svd->vdev_path != NULL) { 1940 dvd->vdev_path = spa_strdup(svd->vdev_path); 1941 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 1942 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 1943 } 1944 } 1945 1946 /* 1947 * Recursively copy vdev paths from one vdev to another. Source and destination 1948 * vdev trees must have same geometry otherwise return error. Intended to copy 1949 * paths from userland config into MOS config. 1950 */ 1951 int 1952 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 1953 { 1954 if ((svd->vdev_ops == &vdev_missing_ops) || 1955 (svd->vdev_ishole && dvd->vdev_ishole) || 1956 (dvd->vdev_ops == &vdev_indirect_ops)) 1957 return (0); 1958 1959 if (svd->vdev_ops != dvd->vdev_ops) { 1960 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 1961 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 1962 return (SET_ERROR(EINVAL)); 1963 } 1964 1965 if (svd->vdev_guid != dvd->vdev_guid) { 1966 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 1967 "%llu)", (u_longlong_t)svd->vdev_guid, 1968 (u_longlong_t)dvd->vdev_guid); 1969 return (SET_ERROR(EINVAL)); 1970 } 1971 1972 if (svd->vdev_children != dvd->vdev_children) { 1973 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 1974 "%llu != %llu", (u_longlong_t)svd->vdev_children, 1975 (u_longlong_t)dvd->vdev_children); 1976 return (SET_ERROR(EINVAL)); 1977 } 1978 1979 for (uint64_t i = 0; i < svd->vdev_children; i++) { 1980 int error = vdev_copy_path_strict(svd->vdev_child[i], 1981 dvd->vdev_child[i]); 1982 if (error != 0) 1983 return (error); 1984 } 1985 1986 if (svd->vdev_ops->vdev_op_leaf) 1987 vdev_copy_path_impl(svd, dvd); 1988 1989 return (0); 1990 } 1991 1992 static void 1993 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 1994 { 1995 ASSERT(stvd->vdev_top == stvd); 1996 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 1997 1998 for (uint64_t i = 0; i < dvd->vdev_children; i++) { 1999 vdev_copy_path_search(stvd, dvd->vdev_child[i]); 2000 } 2001 2002 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 2003 return; 2004 2005 /* 2006 * The idea here is that while a vdev can shift positions within 2007 * a top vdev (when replacing, attaching mirror, etc.) it cannot 2008 * step outside of it. 2009 */ 2010 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 2011 2012 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 2013 return; 2014 2015 ASSERT(vd->vdev_ops->vdev_op_leaf); 2016 2017 vdev_copy_path_impl(vd, dvd); 2018 } 2019 2020 /* 2021 * Recursively copy vdev paths from one root vdev to another. Source and 2022 * destination vdev trees may differ in geometry. For each destination leaf 2023 * vdev, search a vdev with the same guid and top vdev id in the source. 2024 * Intended to copy paths from userland config into MOS config. 2025 */ 2026 void 2027 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 2028 { 2029 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 2030 ASSERT(srvd->vdev_ops == &vdev_root_ops); 2031 ASSERT(drvd->vdev_ops == &vdev_root_ops); 2032 2033 for (uint64_t i = 0; i < children; i++) { 2034 vdev_copy_path_search(srvd->vdev_child[i], 2035 drvd->vdev_child[i]); 2036 } 2037 } 2038 2039 /* 2040 * Close a virtual device. 2041 */ 2042 void 2043 vdev_close(vdev_t *vd) 2044 { 2045 spa_t *spa = vd->vdev_spa; 2046 vdev_t *pvd = vd->vdev_parent; 2047 2048 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2049 2050 /* 2051 * If our parent is reopening, then we are as well, unless we are 2052 * going offline. 2053 */ 2054 if (pvd != NULL && pvd->vdev_reopening) 2055 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 2056 2057 vd->vdev_ops->vdev_op_close(vd); 2058 2059 vdev_cache_purge(vd); 2060 2061 /* 2062 * We record the previous state before we close it, so that if we are 2063 * doing a reopen(), we don't generate FMA ereports if we notice that 2064 * it's still faulted. 2065 */ 2066 vd->vdev_prevstate = vd->vdev_state; 2067 2068 if (vd->vdev_offline) 2069 vd->vdev_state = VDEV_STATE_OFFLINE; 2070 else 2071 vd->vdev_state = VDEV_STATE_CLOSED; 2072 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2073 } 2074 2075 void 2076 vdev_hold(vdev_t *vd) 2077 { 2078 spa_t *spa = vd->vdev_spa; 2079 2080 ASSERT(spa_is_root(spa)); 2081 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 2082 return; 2083 2084 for (int c = 0; c < vd->vdev_children; c++) 2085 vdev_hold(vd->vdev_child[c]); 2086 2087 if (vd->vdev_ops->vdev_op_leaf) 2088 vd->vdev_ops->vdev_op_hold(vd); 2089 } 2090 2091 void 2092 vdev_rele(vdev_t *vd) 2093 { 2094 spa_t *spa = vd->vdev_spa; 2095 2096 ASSERT(spa_is_root(spa)); 2097 for (int c = 0; c < vd->vdev_children; c++) 2098 vdev_rele(vd->vdev_child[c]); 2099 2100 if (vd->vdev_ops->vdev_op_leaf) 2101 vd->vdev_ops->vdev_op_rele(vd); 2102 } 2103 2104 /* 2105 * Reopen all interior vdevs and any unopened leaves. We don't actually 2106 * reopen leaf vdevs which had previously been opened as they might deadlock 2107 * on the spa_config_lock. Instead we only obtain the leaf's physical size. 2108 * If the leaf has never been opened then open it, as usual. 2109 */ 2110 void 2111 vdev_reopen(vdev_t *vd) 2112 { 2113 spa_t *spa = vd->vdev_spa; 2114 2115 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2116 2117 /* set the reopening flag unless we're taking the vdev offline */ 2118 vd->vdev_reopening = !vd->vdev_offline; 2119 vdev_close(vd); 2120 (void) vdev_open(vd); 2121 2122 /* 2123 * Call vdev_validate() here to make sure we have the same device. 2124 * Otherwise, a device with an invalid label could be successfully 2125 * opened in response to vdev_reopen(). 2126 */ 2127 if (vd->vdev_aux) { 2128 (void) vdev_validate_aux(vd); 2129 if (vdev_readable(vd) && vdev_writeable(vd) && 2130 vd->vdev_aux == &spa->spa_l2cache && 2131 !l2arc_vdev_present(vd)) 2132 l2arc_add_vdev(spa, vd); 2133 } else { 2134 (void) vdev_validate(vd); 2135 } 2136 2137 /* 2138 * Reassess parent vdev's health. 2139 */ 2140 vdev_propagate_state(vd); 2141 } 2142 2143 int 2144 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2145 { 2146 int error; 2147 2148 /* 2149 * Normally, partial opens (e.g. of a mirror) are allowed. 2150 * For a create, however, we want to fail the request if 2151 * there are any components we can't open. 2152 */ 2153 error = vdev_open(vd); 2154 2155 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2156 vdev_close(vd); 2157 return (error ? error : ENXIO); 2158 } 2159 2160 /* 2161 * Recursively load DTLs and initialize all labels. 2162 */ 2163 if ((error = vdev_dtl_load(vd)) != 0 || 2164 (error = vdev_label_init(vd, txg, isreplacing ? 2165 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2166 vdev_close(vd); 2167 return (error); 2168 } 2169 2170 return (0); 2171 } 2172 2173 void 2174 vdev_metaslab_set_size(vdev_t *vd) 2175 { 2176 uint64_t asize = vd->vdev_asize; 2177 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2178 uint64_t ms_shift; 2179 2180 /* BEGIN CSTYLED */ 2181 /* 2182 * There are two dimensions to the metaslab sizing calculation: 2183 * the size of the metaslab and the count of metaslabs per vdev. 2184 * 2185 * The default values used below are a good balance between memory 2186 * usage (larger metaslab size means more memory needed for loaded 2187 * metaslabs; more metaslabs means more memory needed for the 2188 * metaslab_t structs), metaslab load time (larger metaslabs take 2189 * longer to load), and metaslab sync time (more metaslabs means 2190 * more time spent syncing all of them). 2191 * 2192 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2193 * The range of the dimensions are as follows: 2194 * 2195 * 2^29 <= ms_size <= 2^34 2196 * 16 <= ms_count <= 131,072 2197 * 2198 * On the lower end of vdev sizes, we aim for metaslabs sizes of 2199 * at least 512MB (2^29) to minimize fragmentation effects when 2200 * testing with smaller devices. However, the count constraint 2201 * of at least 16 metaslabs will override this minimum size goal. 2202 * 2203 * On the upper end of vdev sizes, we aim for a maximum metaslab 2204 * size of 16GB. However, we will cap the total count to 2^17 2205 * metaslabs to keep our memory footprint in check and let the 2206 * metaslab size grow from there if that limit is hit. 2207 * 2208 * The net effect of applying above constrains is summarized below. 2209 * 2210 * vdev size metaslab count 2211 * --------------|----------------- 2212 * < 8GB ~16 2213 * 8GB - 100GB one per 512MB 2214 * 100GB - 3TB ~200 2215 * 3TB - 2PB one per 16GB 2216 * > 2PB ~131,072 2217 * -------------------------------- 2218 * 2219 * Finally, note that all of the above calculate the initial 2220 * number of metaslabs. Expanding a top-level vdev will result 2221 * in additional metaslabs being allocated making it possible 2222 * to exceed the zfs_vdev_ms_count_limit. 2223 */ 2224 /* END CSTYLED */ 2225 2226 if (ms_count < zfs_vdev_min_ms_count) 2227 ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2228 else if (ms_count > zfs_vdev_default_ms_count) 2229 ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2230 else 2231 ms_shift = zfs_vdev_default_ms_shift; 2232 2233 if (ms_shift < SPA_MAXBLOCKSHIFT) { 2234 ms_shift = SPA_MAXBLOCKSHIFT; 2235 } else if (ms_shift > zfs_vdev_max_ms_shift) { 2236 ms_shift = zfs_vdev_max_ms_shift; 2237 /* cap the total count to constrain memory footprint */ 2238 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2239 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 2240 } 2241 2242 vd->vdev_ms_shift = ms_shift; 2243 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2244 } 2245 2246 void 2247 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2248 { 2249 ASSERT(vd == vd->vdev_top); 2250 /* indirect vdevs don't have metaslabs or dtls */ 2251 ASSERT(vdev_is_concrete(vd) || flags == 0); 2252 ASSERT(ISP2(flags)); 2253 ASSERT(spa_writeable(vd->vdev_spa)); 2254 2255 if (flags & VDD_METASLAB) 2256 (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2257 2258 if (flags & VDD_DTL) 2259 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2260 2261 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2262 } 2263 2264 void 2265 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 2266 { 2267 for (int c = 0; c < vd->vdev_children; c++) 2268 vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 2269 2270 if (vd->vdev_ops->vdev_op_leaf) 2271 vdev_dirty(vd->vdev_top, flags, vd, txg); 2272 } 2273 2274 /* 2275 * DTLs. 2276 * 2277 * A vdev's DTL (dirty time log) is the set of transaction groups for which 2278 * the vdev has less than perfect replication. There are four kinds of DTL: 2279 * 2280 * DTL_MISSING: txgs for which the vdev has no valid copies of the data 2281 * 2282 * DTL_PARTIAL: txgs for which data is available, but not fully replicated 2283 * 2284 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 2285 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 2286 * txgs that was scrubbed. 2287 * 2288 * DTL_OUTAGE: txgs which cannot currently be read, whether due to 2289 * persistent errors or just some device being offline. 2290 * Unlike the other three, the DTL_OUTAGE map is not generally 2291 * maintained; it's only computed when needed, typically to 2292 * determine whether a device can be detached. 2293 * 2294 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 2295 * either has the data or it doesn't. 2296 * 2297 * For interior vdevs such as mirror and RAID-Z the picture is more complex. 2298 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 2299 * if any child is less than fully replicated, then so is its parent. 2300 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 2301 * comprising only those txgs which appear in 'maxfaults' or more children; 2302 * those are the txgs we don't have enough replication to read. For example, 2303 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 2304 * thus, its DTL_MISSING consists of the set of txgs that appear in more than 2305 * two child DTL_MISSING maps. 2306 * 2307 * It should be clear from the above that to compute the DTLs and outage maps 2308 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 2309 * Therefore, that is all we keep on disk. When loading the pool, or after 2310 * a configuration change, we generate all other DTLs from first principles. 2311 */ 2312 void 2313 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2314 { 2315 range_tree_t *rt = vd->vdev_dtl[t]; 2316 2317 ASSERT(t < DTL_TYPES); 2318 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2319 ASSERT(spa_writeable(vd->vdev_spa)); 2320 2321 mutex_enter(&vd->vdev_dtl_lock); 2322 if (!range_tree_contains(rt, txg, size)) 2323 range_tree_add(rt, txg, size); 2324 mutex_exit(&vd->vdev_dtl_lock); 2325 } 2326 2327 boolean_t 2328 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2329 { 2330 range_tree_t *rt = vd->vdev_dtl[t]; 2331 boolean_t dirty = B_FALSE; 2332 2333 ASSERT(t < DTL_TYPES); 2334 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2335 2336 /* 2337 * While we are loading the pool, the DTLs have not been loaded yet. 2338 * Ignore the DTLs and try all devices. This avoids a recursive 2339 * mutex enter on the vdev_dtl_lock, and also makes us try hard 2340 * when loading the pool (relying on the checksum to ensure that 2341 * we get the right data -- note that we while loading, we are 2342 * only reading the MOS, which is always checksummed). 2343 */ 2344 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 2345 return (B_FALSE); 2346 2347 mutex_enter(&vd->vdev_dtl_lock); 2348 if (!range_tree_is_empty(rt)) 2349 dirty = range_tree_contains(rt, txg, size); 2350 mutex_exit(&vd->vdev_dtl_lock); 2351 2352 return (dirty); 2353 } 2354 2355 boolean_t 2356 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 2357 { 2358 range_tree_t *rt = vd->vdev_dtl[t]; 2359 boolean_t empty; 2360 2361 mutex_enter(&vd->vdev_dtl_lock); 2362 empty = range_tree_is_empty(rt); 2363 mutex_exit(&vd->vdev_dtl_lock); 2364 2365 return (empty); 2366 } 2367 2368 /* 2369 * Returns B_TRUE if vdev determines offset needs to be resilvered. 2370 */ 2371 boolean_t 2372 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize) 2373 { 2374 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2375 2376 if (vd->vdev_ops->vdev_op_need_resilver == NULL || 2377 vd->vdev_ops->vdev_op_leaf) 2378 return (B_TRUE); 2379 2380 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize)); 2381 } 2382 2383 /* 2384 * Returns the lowest txg in the DTL range. 2385 */ 2386 static uint64_t 2387 vdev_dtl_min(vdev_t *vd) 2388 { 2389 range_seg_t *rs; 2390 2391 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2392 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2393 ASSERT0(vd->vdev_children); 2394 2395 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 2396 return (rs->rs_start - 1); 2397 } 2398 2399 /* 2400 * Returns the highest txg in the DTL. 2401 */ 2402 static uint64_t 2403 vdev_dtl_max(vdev_t *vd) 2404 { 2405 range_seg_t *rs; 2406 2407 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2408 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2409 ASSERT0(vd->vdev_children); 2410 2411 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 2412 return (rs->rs_end); 2413 } 2414 2415 /* 2416 * Determine if a resilvering vdev should remove any DTL entries from 2417 * its range. If the vdev was resilvering for the entire duration of the 2418 * scan then it should excise that range from its DTLs. Otherwise, this 2419 * vdev is considered partially resilvered and should leave its DTL 2420 * entries intact. The comment in vdev_dtl_reassess() describes how we 2421 * excise the DTLs. 2422 */ 2423 static boolean_t 2424 vdev_dtl_should_excise(vdev_t *vd) 2425 { 2426 spa_t *spa = vd->vdev_spa; 2427 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2428 2429 ASSERT0(scn->scn_phys.scn_errors); 2430 ASSERT0(vd->vdev_children); 2431 2432 if (vd->vdev_state < VDEV_STATE_DEGRADED) 2433 return (B_FALSE); 2434 2435 if (vd->vdev_resilver_txg == 0 || 2436 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2437 return (B_TRUE); 2438 2439 /* 2440 * When a resilver is initiated the scan will assign the scn_max_txg 2441 * value to the highest txg value that exists in all DTLs. If this 2442 * device's max DTL is not part of this scan (i.e. it is not in 2443 * the range (scn_min_txg, scn_max_txg] then it is not eligible 2444 * for excision. 2445 */ 2446 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2447 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2448 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2449 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2450 return (B_TRUE); 2451 } 2452 return (B_FALSE); 2453 } 2454 2455 /* 2456 * Reassess DTLs after a config change or scrub completion. 2457 */ 2458 void 2459 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2460 { 2461 spa_t *spa = vd->vdev_spa; 2462 avl_tree_t reftree; 2463 int minref; 2464 2465 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2466 2467 for (int c = 0; c < vd->vdev_children; c++) 2468 vdev_dtl_reassess(vd->vdev_child[c], txg, 2469 scrub_txg, scrub_done); 2470 2471 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 2472 return; 2473 2474 if (vd->vdev_ops->vdev_op_leaf) { 2475 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2476 2477 mutex_enter(&vd->vdev_dtl_lock); 2478 2479 /* 2480 * If we've completed a scan cleanly then determine 2481 * if this vdev should remove any DTLs. We only want to 2482 * excise regions on vdevs that were available during 2483 * the entire duration of this scan. 2484 */ 2485 if (scrub_txg != 0 && 2486 (spa->spa_scrub_started || 2487 (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2488 vdev_dtl_should_excise(vd)) { 2489 /* 2490 * We completed a scrub up to scrub_txg. If we 2491 * did it without rebooting, then the scrub dtl 2492 * will be valid, so excise the old region and 2493 * fold in the scrub dtl. Otherwise, leave the 2494 * dtl as-is if there was an error. 2495 * 2496 * There's little trick here: to excise the beginning 2497 * of the DTL_MISSING map, we put it into a reference 2498 * tree and then add a segment with refcnt -1 that 2499 * covers the range [0, scrub_txg). This means 2500 * that each txg in that range has refcnt -1 or 0. 2501 * We then add DTL_SCRUB with a refcnt of 2, so that 2502 * entries in the range [0, scrub_txg) will have a 2503 * positive refcnt -- either 1 or 2. We then convert 2504 * the reference tree into the new DTL_MISSING map. 2505 */ 2506 space_reftree_create(&reftree); 2507 space_reftree_add_map(&reftree, 2508 vd->vdev_dtl[DTL_MISSING], 1); 2509 space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 2510 space_reftree_add_map(&reftree, 2511 vd->vdev_dtl[DTL_SCRUB], 2); 2512 space_reftree_generate_map(&reftree, 2513 vd->vdev_dtl[DTL_MISSING], 1); 2514 space_reftree_destroy(&reftree); 2515 } 2516 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 2517 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 2518 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2519 if (scrub_done) 2520 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 2521 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 2522 if (!vdev_readable(vd)) 2523 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 2524 else 2525 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 2526 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2527 2528 /* 2529 * If the vdev was resilvering and no longer has any 2530 * DTLs then reset its resilvering flag. 2531 */ 2532 if (vd->vdev_resilver_txg != 0 && 2533 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 2534 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) 2535 vd->vdev_resilver_txg = 0; 2536 2537 mutex_exit(&vd->vdev_dtl_lock); 2538 2539 if (txg != 0) 2540 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2541 return; 2542 } 2543 2544 mutex_enter(&vd->vdev_dtl_lock); 2545 for (int t = 0; t < DTL_TYPES; t++) { 2546 /* account for child's outage in parent's missing map */ 2547 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 2548 if (t == DTL_SCRUB) 2549 continue; /* leaf vdevs only */ 2550 if (t == DTL_PARTIAL) 2551 minref = 1; /* i.e. non-zero */ 2552 else if (vd->vdev_nparity != 0) 2553 minref = vd->vdev_nparity + 1; /* RAID-Z */ 2554 else 2555 minref = vd->vdev_children; /* any kind of mirror */ 2556 space_reftree_create(&reftree); 2557 for (int c = 0; c < vd->vdev_children; c++) { 2558 vdev_t *cvd = vd->vdev_child[c]; 2559 mutex_enter(&cvd->vdev_dtl_lock); 2560 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 2561 mutex_exit(&cvd->vdev_dtl_lock); 2562 } 2563 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 2564 space_reftree_destroy(&reftree); 2565 } 2566 mutex_exit(&vd->vdev_dtl_lock); 2567 } 2568 2569 int 2570 vdev_dtl_load(vdev_t *vd) 2571 { 2572 spa_t *spa = vd->vdev_spa; 2573 objset_t *mos = spa->spa_meta_objset; 2574 int error = 0; 2575 2576 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 2577 ASSERT(vdev_is_concrete(vd)); 2578 2579 error = space_map_open(&vd->vdev_dtl_sm, mos, 2580 vd->vdev_dtl_object, 0, -1ULL, 0); 2581 if (error) 2582 return (error); 2583 ASSERT(vd->vdev_dtl_sm != NULL); 2584 2585 mutex_enter(&vd->vdev_dtl_lock); 2586 error = space_map_load(vd->vdev_dtl_sm, 2587 vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 2588 mutex_exit(&vd->vdev_dtl_lock); 2589 2590 return (error); 2591 } 2592 2593 for (int c = 0; c < vd->vdev_children; c++) { 2594 error = vdev_dtl_load(vd->vdev_child[c]); 2595 if (error != 0) 2596 break; 2597 } 2598 2599 return (error); 2600 } 2601 2602 static void 2603 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) 2604 { 2605 spa_t *spa = vd->vdev_spa; 2606 objset_t *mos = spa->spa_meta_objset; 2607 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 2608 const char *string; 2609 2610 ASSERT(alloc_bias != VDEV_BIAS_NONE); 2611 2612 string = 2613 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : 2614 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : 2615 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; 2616 2617 ASSERT(string != NULL); 2618 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 2619 1, strlen(string) + 1, string, tx)); 2620 2621 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { 2622 spa_activate_allocation_classes(spa, tx); 2623 } 2624 } 2625 2626 void 2627 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2628 { 2629 spa_t *spa = vd->vdev_spa; 2630 2631 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2632 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2633 zapobj, tx)); 2634 } 2635 2636 uint64_t 2637 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2638 { 2639 spa_t *spa = vd->vdev_spa; 2640 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2641 DMU_OT_NONE, 0, tx); 2642 2643 ASSERT(zap != 0); 2644 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2645 zap, tx)); 2646 2647 return (zap); 2648 } 2649 2650 void 2651 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2652 { 2653 if (vd->vdev_ops != &vdev_hole_ops && 2654 vd->vdev_ops != &vdev_missing_ops && 2655 vd->vdev_ops != &vdev_root_ops && 2656 !vd->vdev_top->vdev_removing) { 2657 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2658 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2659 } 2660 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2661 vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2662 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) 2663 vdev_zap_allocation_data(vd, tx); 2664 } 2665 } 2666 2667 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2668 vdev_construct_zaps(vd->vdev_child[i], tx); 2669 } 2670 } 2671 2672 void 2673 vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2674 { 2675 spa_t *spa = vd->vdev_spa; 2676 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2677 objset_t *mos = spa->spa_meta_objset; 2678 range_tree_t *rtsync; 2679 dmu_tx_t *tx; 2680 uint64_t object = space_map_object(vd->vdev_dtl_sm); 2681 2682 ASSERT(vdev_is_concrete(vd)); 2683 ASSERT(vd->vdev_ops->vdev_op_leaf); 2684 2685 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2686 2687 if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 2688 mutex_enter(&vd->vdev_dtl_lock); 2689 space_map_free(vd->vdev_dtl_sm, tx); 2690 space_map_close(vd->vdev_dtl_sm); 2691 vd->vdev_dtl_sm = NULL; 2692 mutex_exit(&vd->vdev_dtl_lock); 2693 2694 /* 2695 * We only destroy the leaf ZAP for detached leaves or for 2696 * removed log devices. Removed data devices handle leaf ZAP 2697 * cleanup later, once cancellation is no longer possible. 2698 */ 2699 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2700 vd->vdev_top->vdev_islog)) { 2701 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2702 vd->vdev_leaf_zap = 0; 2703 } 2704 2705 dmu_tx_commit(tx); 2706 return; 2707 } 2708 2709 if (vd->vdev_dtl_sm == NULL) { 2710 uint64_t new_object; 2711 2712 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx); 2713 VERIFY3U(new_object, !=, 0); 2714 2715 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 2716 0, -1ULL, 0)); 2717 ASSERT(vd->vdev_dtl_sm != NULL); 2718 } 2719 2720 rtsync = range_tree_create(NULL, NULL); 2721 2722 mutex_enter(&vd->vdev_dtl_lock); 2723 range_tree_walk(rt, range_tree_add, rtsync); 2724 mutex_exit(&vd->vdev_dtl_lock); 2725 2726 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx); 2727 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 2728 range_tree_vacate(rtsync, NULL, NULL); 2729 2730 range_tree_destroy(rtsync); 2731 2732 /* 2733 * If the object for the space map has changed then dirty 2734 * the top level so that we update the config. 2735 */ 2736 if (object != space_map_object(vd->vdev_dtl_sm)) { 2737 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 2738 "new object %llu", (u_longlong_t)txg, spa_name(spa), 2739 (u_longlong_t)object, 2740 (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 2741 vdev_config_dirty(vd->vdev_top); 2742 } 2743 2744 dmu_tx_commit(tx); 2745 } 2746 2747 /* 2748 * Determine whether the specified vdev can be offlined/detached/removed 2749 * without losing data. 2750 */ 2751 boolean_t 2752 vdev_dtl_required(vdev_t *vd) 2753 { 2754 spa_t *spa = vd->vdev_spa; 2755 vdev_t *tvd = vd->vdev_top; 2756 uint8_t cant_read = vd->vdev_cant_read; 2757 boolean_t required; 2758 2759 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2760 2761 if (vd == spa->spa_root_vdev || vd == tvd) 2762 return (B_TRUE); 2763 2764 /* 2765 * Temporarily mark the device as unreadable, and then determine 2766 * whether this results in any DTL outages in the top-level vdev. 2767 * If not, we can safely offline/detach/remove the device. 2768 */ 2769 vd->vdev_cant_read = B_TRUE; 2770 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 2771 required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 2772 vd->vdev_cant_read = cant_read; 2773 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 2774 2775 if (!required && zio_injection_enabled) 2776 required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2777 2778 return (required); 2779 } 2780 2781 /* 2782 * Determine if resilver is needed, and if so the txg range. 2783 */ 2784 boolean_t 2785 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2786 { 2787 boolean_t needed = B_FALSE; 2788 uint64_t thismin = UINT64_MAX; 2789 uint64_t thismax = 0; 2790 2791 if (vd->vdev_children == 0) { 2792 mutex_enter(&vd->vdev_dtl_lock); 2793 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 2794 vdev_writeable(vd)) { 2795 2796 thismin = vdev_dtl_min(vd); 2797 thismax = vdev_dtl_max(vd); 2798 needed = B_TRUE; 2799 } 2800 mutex_exit(&vd->vdev_dtl_lock); 2801 } else { 2802 for (int c = 0; c < vd->vdev_children; c++) { 2803 vdev_t *cvd = vd->vdev_child[c]; 2804 uint64_t cmin, cmax; 2805 2806 if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2807 thismin = MIN(thismin, cmin); 2808 thismax = MAX(thismax, cmax); 2809 needed = B_TRUE; 2810 } 2811 } 2812 } 2813 2814 if (needed && minp) { 2815 *minp = thismin; 2816 *maxp = thismax; 2817 } 2818 return (needed); 2819 } 2820 2821 /* 2822 * Gets the checkpoint space map object from the vdev's ZAP. 2823 * Returns the spacemap object, or 0 if it wasn't in the ZAP 2824 * or the ZAP doesn't exist yet. 2825 */ 2826 int 2827 vdev_checkpoint_sm_object(vdev_t *vd) 2828 { 2829 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 2830 if (vd->vdev_top_zap == 0) { 2831 return (0); 2832 } 2833 2834 uint64_t sm_obj = 0; 2835 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 2836 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); 2837 2838 ASSERT(err == 0 || err == ENOENT); 2839 2840 return (sm_obj); 2841 } 2842 2843 int 2844 vdev_load(vdev_t *vd) 2845 { 2846 int error = 0; 2847 /* 2848 * Recursively load all children. 2849 */ 2850 for (int c = 0; c < vd->vdev_children; c++) { 2851 error = vdev_load(vd->vdev_child[c]); 2852 if (error != 0) { 2853 return (error); 2854 } 2855 } 2856 2857 vdev_set_deflate_ratio(vd); 2858 2859 /* 2860 * On spa_load path, grab the allocation bias from our zap 2861 */ 2862 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 2863 spa_t *spa = vd->vdev_spa; 2864 char bias_str[64]; 2865 2866 if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, 2867 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), 2868 bias_str) == 0) { 2869 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); 2870 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); 2871 } 2872 } 2873 2874 /* 2875 * If this is a top-level vdev, initialize its metaslabs. 2876 */ 2877 if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 2878 vdev_metaslab_group_create(vd); 2879 2880 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 2881 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2882 VDEV_AUX_CORRUPT_DATA); 2883 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 2884 "asize=%llu", (u_longlong_t)vd->vdev_ashift, 2885 (u_longlong_t)vd->vdev_asize); 2886 return (SET_ERROR(ENXIO)); 2887 } 2888 2889 error = vdev_metaslab_init(vd, 0); 2890 if (error != 0) { 2891 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 2892 "[error=%d]", error); 2893 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2894 VDEV_AUX_CORRUPT_DATA); 2895 return (error); 2896 } 2897 2898 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); 2899 if (checkpoint_sm_obj != 0) { 2900 objset_t *mos = spa_meta_objset(vd->vdev_spa); 2901 ASSERT(vd->vdev_asize != 0); 2902 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 2903 2904 error = space_map_open(&vd->vdev_checkpoint_sm, 2905 mos, checkpoint_sm_obj, 0, vd->vdev_asize, 2906 vd->vdev_ashift); 2907 if (error != 0) { 2908 vdev_dbgmsg(vd, "vdev_load: space_map_open " 2909 "failed for checkpoint spacemap (obj %llu) " 2910 "[error=%d]", 2911 (u_longlong_t)checkpoint_sm_obj, error); 2912 return (error); 2913 } 2914 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 2915 2916 /* 2917 * Since the checkpoint_sm contains free entries 2918 * exclusively we can use space_map_allocated() to 2919 * indicate the cumulative checkpointed space that 2920 * has been freed. 2921 */ 2922 vd->vdev_stat.vs_checkpoint_space = 2923 -space_map_allocated(vd->vdev_checkpoint_sm); 2924 vd->vdev_spa->spa_checkpoint_info.sci_dspace += 2925 vd->vdev_stat.vs_checkpoint_space; 2926 } 2927 } 2928 2929 /* 2930 * If this is a leaf vdev, load its DTL. 2931 */ 2932 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 2933 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2934 VDEV_AUX_CORRUPT_DATA); 2935 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 2936 "[error=%d]", error); 2937 return (error); 2938 } 2939 2940 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 2941 if (obsolete_sm_object != 0) { 2942 objset_t *mos = vd->vdev_spa->spa_meta_objset; 2943 ASSERT(vd->vdev_asize != 0); 2944 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 2945 2946 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 2947 obsolete_sm_object, 0, vd->vdev_asize, 0))) { 2948 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2949 VDEV_AUX_CORRUPT_DATA); 2950 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 2951 "obsolete spacemap (obj %llu) [error=%d]", 2952 (u_longlong_t)obsolete_sm_object, error); 2953 return (error); 2954 } 2955 } 2956 2957 return (0); 2958 } 2959 2960 /* 2961 * The special vdev case is used for hot spares and l2cache devices. Its 2962 * sole purpose it to set the vdev state for the associated vdev. To do this, 2963 * we make sure that we can open the underlying device, then try to read the 2964 * label, and make sure that the label is sane and that it hasn't been 2965 * repurposed to another pool. 2966 */ 2967 int 2968 vdev_validate_aux(vdev_t *vd) 2969 { 2970 nvlist_t *label; 2971 uint64_t guid, version; 2972 uint64_t state; 2973 2974 if (!vdev_readable(vd)) 2975 return (0); 2976 2977 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 2978 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2979 VDEV_AUX_CORRUPT_DATA); 2980 return (-1); 2981 } 2982 2983 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2984 !SPA_VERSION_IS_SUPPORTED(version) || 2985 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 2986 guid != vd->vdev_guid || 2987 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 2988 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2989 VDEV_AUX_CORRUPT_DATA); 2990 nvlist_free(label); 2991 return (-1); 2992 } 2993 2994 /* 2995 * We don't actually check the pool state here. If it's in fact in 2996 * use by another pool, we update this fact on the fly when requested. 2997 */ 2998 nvlist_free(label); 2999 return (0); 3000 } 3001 3002 /* 3003 * Free the objects used to store this vdev's spacemaps, and the array 3004 * that points to them. 3005 */ 3006 void 3007 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 3008 { 3009 if (vd->vdev_ms_array == 0) 3010 return; 3011 3012 objset_t *mos = vd->vdev_spa->spa_meta_objset; 3013 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 3014 size_t array_bytes = array_count * sizeof (uint64_t); 3015 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 3016 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 3017 array_bytes, smobj_array, 0)); 3018 3019 for (uint64_t i = 0; i < array_count; i++) { 3020 uint64_t smobj = smobj_array[i]; 3021 if (smobj == 0) 3022 continue; 3023 3024 space_map_free_obj(mos, smobj, tx); 3025 } 3026 3027 kmem_free(smobj_array, array_bytes); 3028 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 3029 vd->vdev_ms_array = 0; 3030 } 3031 3032 static void 3033 vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 3034 { 3035 spa_t *spa = vd->vdev_spa; 3036 3037 ASSERT(vd->vdev_islog); 3038 ASSERT(vd == vd->vdev_top); 3039 ASSERT3U(txg, ==, spa_syncing_txg(spa)); 3040 3041 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 3042 3043 vdev_destroy_spacemaps(vd, tx); 3044 if (vd->vdev_top_zap != 0) { 3045 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 3046 vd->vdev_top_zap = 0; 3047 } 3048 3049 dmu_tx_commit(tx); 3050 } 3051 3052 void 3053 vdev_sync_done(vdev_t *vd, uint64_t txg) 3054 { 3055 metaslab_t *msp; 3056 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 3057 3058 ASSERT(vdev_is_concrete(vd)); 3059 3060 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 3061 != NULL) 3062 metaslab_sync_done(msp, txg); 3063 3064 if (reassess) 3065 metaslab_sync_reassess(vd->vdev_mg); 3066 } 3067 3068 void 3069 vdev_sync(vdev_t *vd, uint64_t txg) 3070 { 3071 spa_t *spa = vd->vdev_spa; 3072 vdev_t *lvd; 3073 metaslab_t *msp; 3074 3075 ASSERT3U(txg, ==, spa->spa_syncing_txg); 3076 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3077 if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 3078 ASSERT(vd->vdev_removing || 3079 vd->vdev_ops == &vdev_indirect_ops); 3080 3081 vdev_indirect_sync_obsolete(vd, tx); 3082 3083 /* 3084 * If the vdev is indirect, it can't have dirty 3085 * metaslabs or DTLs. 3086 */ 3087 if (vd->vdev_ops == &vdev_indirect_ops) { 3088 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 3089 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 3090 dmu_tx_commit(tx); 3091 return; 3092 } 3093 } 3094 3095 ASSERT(vdev_is_concrete(vd)); 3096 3097 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 3098 !vd->vdev_removing) { 3099 ASSERT(vd == vd->vdev_top); 3100 ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 3101 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 3102 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 3103 ASSERT(vd->vdev_ms_array != 0); 3104 vdev_config_dirty(vd); 3105 } 3106 3107 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 3108 metaslab_sync(msp, txg); 3109 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 3110 } 3111 3112 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 3113 vdev_dtl_sync(lvd, txg); 3114 3115 /* 3116 * If this is an empty log device being removed, destroy the 3117 * metadata associated with it. 3118 */ 3119 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 3120 vdev_remove_empty_log(vd, txg); 3121 3122 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 3123 dmu_tx_commit(tx); 3124 } 3125 3126 uint64_t 3127 vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 3128 { 3129 return (vd->vdev_ops->vdev_op_asize(vd, psize)); 3130 } 3131 3132 /* 3133 * Mark the given vdev faulted. A faulted vdev behaves as if the device could 3134 * not be opened, and no I/O is attempted. 3135 */ 3136 int 3137 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3138 { 3139 vdev_t *vd, *tvd; 3140 3141 spa_vdev_state_enter(spa, SCL_NONE); 3142 3143 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3144 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3145 3146 if (!vd->vdev_ops->vdev_op_leaf) 3147 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3148 3149 tvd = vd->vdev_top; 3150 3151 /* 3152 * We don't directly use the aux state here, but if we do a 3153 * vdev_reopen(), we need this value to be present to remember why we 3154 * were faulted. 3155 */ 3156 vd->vdev_label_aux = aux; 3157 3158 /* 3159 * Faulted state takes precedence over degraded. 3160 */ 3161 vd->vdev_delayed_close = B_FALSE; 3162 vd->vdev_faulted = 1ULL; 3163 vd->vdev_degraded = 0ULL; 3164 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 3165 3166 /* 3167 * If this device has the only valid copy of the data, then 3168 * back off and simply mark the vdev as degraded instead. 3169 */ 3170 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 3171 vd->vdev_degraded = 1ULL; 3172 vd->vdev_faulted = 0ULL; 3173 3174 /* 3175 * If we reopen the device and it's not dead, only then do we 3176 * mark it degraded. 3177 */ 3178 vdev_reopen(tvd); 3179 3180 if (vdev_readable(vd)) 3181 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 3182 } 3183 3184 return (spa_vdev_state_exit(spa, vd, 0)); 3185 } 3186 3187 /* 3188 * Mark the given vdev degraded. A degraded vdev is purely an indication to the 3189 * user that something is wrong. The vdev continues to operate as normal as far 3190 * as I/O is concerned. 3191 */ 3192 int 3193 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3194 { 3195 vdev_t *vd; 3196 3197 spa_vdev_state_enter(spa, SCL_NONE); 3198 3199 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3200 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3201 3202 if (!vd->vdev_ops->vdev_op_leaf) 3203 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3204 3205 /* 3206 * If the vdev is already faulted, then don't do anything. 3207 */ 3208 if (vd->vdev_faulted || vd->vdev_degraded) 3209 return (spa_vdev_state_exit(spa, NULL, 0)); 3210 3211 vd->vdev_degraded = 1ULL; 3212 if (!vdev_is_dead(vd)) 3213 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3214 aux); 3215 3216 return (spa_vdev_state_exit(spa, vd, 0)); 3217 } 3218 3219 /* 3220 * Online the given vdev. 3221 * 3222 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 3223 * spare device should be detached when the device finishes resilvering. 3224 * Second, the online should be treated like a 'test' online case, so no FMA 3225 * events are generated if the device fails to open. 3226 */ 3227 int 3228 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 3229 { 3230 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 3231 boolean_t wasoffline; 3232 vdev_state_t oldstate; 3233 3234 spa_vdev_state_enter(spa, SCL_NONE); 3235 3236 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3237 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3238 3239 if (!vd->vdev_ops->vdev_op_leaf) 3240 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3241 3242 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 3243 oldstate = vd->vdev_state; 3244 3245 tvd = vd->vdev_top; 3246 vd->vdev_offline = B_FALSE; 3247 vd->vdev_tmpoffline = B_FALSE; 3248 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3249 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3250 3251 /* XXX - L2ARC 1.0 does not support expansion */ 3252 if (!vd->vdev_aux) { 3253 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3254 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 3255 } 3256 3257 vdev_reopen(tvd); 3258 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 3259 3260 if (!vd->vdev_aux) { 3261 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3262 pvd->vdev_expanding = B_FALSE; 3263 } 3264 3265 if (newstate) 3266 *newstate = vd->vdev_state; 3267 if ((flags & ZFS_ONLINE_UNSPARE) && 3268 !vdev_is_dead(vd) && vd->vdev_parent && 3269 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3270 vd->vdev_parent->vdev_child[0] == vd) 3271 vd->vdev_unspare = B_TRUE; 3272 3273 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3274 3275 /* XXX - L2ARC 1.0 does not support expansion */ 3276 if (vd->vdev_aux) 3277 return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3278 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3279 } 3280 3281 /* Restart initializing if necessary */ 3282 mutex_enter(&vd->vdev_initialize_lock); 3283 if (vdev_writeable(vd) && 3284 vd->vdev_initialize_thread == NULL && 3285 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 3286 (void) vdev_initialize(vd); 3287 } 3288 mutex_exit(&vd->vdev_initialize_lock); 3289 3290 if (wasoffline || 3291 (oldstate < VDEV_STATE_DEGRADED && 3292 vd->vdev_state >= VDEV_STATE_DEGRADED)) 3293 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 3294 3295 return (spa_vdev_state_exit(spa, vd, 0)); 3296 } 3297 3298 static int 3299 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3300 { 3301 vdev_t *vd, *tvd; 3302 int error = 0; 3303 uint64_t generation; 3304 metaslab_group_t *mg; 3305 3306 top: 3307 spa_vdev_state_enter(spa, SCL_ALLOC); 3308 3309 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3310 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3311 3312 if (!vd->vdev_ops->vdev_op_leaf) 3313 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3314 3315 tvd = vd->vdev_top; 3316 mg = tvd->vdev_mg; 3317 generation = spa->spa_config_generation + 1; 3318 3319 /* 3320 * If the device isn't already offline, try to offline it. 3321 */ 3322 if (!vd->vdev_offline) { 3323 /* 3324 * If this device has the only valid copy of some data, 3325 * don't allow it to be offlined. Log devices are always 3326 * expendable. 3327 */ 3328 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3329 vdev_dtl_required(vd)) 3330 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3331 3332 /* 3333 * If the top-level is a slog and it has had allocations 3334 * then proceed. We check that the vdev's metaslab group 3335 * is not NULL since it's possible that we may have just 3336 * added this vdev but not yet initialized its metaslabs. 3337 */ 3338 if (tvd->vdev_islog && mg != NULL) { 3339 /* 3340 * Prevent any future allocations. 3341 */ 3342 metaslab_group_passivate(mg); 3343 (void) spa_vdev_state_exit(spa, vd, 0); 3344 3345 error = spa_reset_logs(spa); 3346 3347 /* 3348 * If the log device was successfully reset but has 3349 * checkpointed data, do not offline it. 3350 */ 3351 if (error == 0 && 3352 tvd->vdev_checkpoint_sm != NULL) { 3353 error = ZFS_ERR_CHECKPOINT_EXISTS; 3354 } 3355 3356 spa_vdev_state_enter(spa, SCL_ALLOC); 3357 3358 /* 3359 * Check to see if the config has changed. 3360 */ 3361 if (error || generation != spa->spa_config_generation) { 3362 metaslab_group_activate(mg); 3363 if (error) 3364 return (spa_vdev_state_exit(spa, 3365 vd, error)); 3366 (void) spa_vdev_state_exit(spa, vd, 0); 3367 goto top; 3368 } 3369 ASSERT0(tvd->vdev_stat.vs_alloc); 3370 } 3371 3372 /* 3373 * Offline this device and reopen its top-level vdev. 3374 * If the top-level vdev is a log device then just offline 3375 * it. Otherwise, if this action results in the top-level 3376 * vdev becoming unusable, undo it and fail the request. 3377 */ 3378 vd->vdev_offline = B_TRUE; 3379 vdev_reopen(tvd); 3380 3381 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3382 vdev_is_dead(tvd)) { 3383 vd->vdev_offline = B_FALSE; 3384 vdev_reopen(tvd); 3385 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3386 } 3387 3388 /* 3389 * Add the device back into the metaslab rotor so that 3390 * once we online the device it's open for business. 3391 */ 3392 if (tvd->vdev_islog && mg != NULL) 3393 metaslab_group_activate(mg); 3394 } 3395 3396 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3397 3398 return (spa_vdev_state_exit(spa, vd, 0)); 3399 } 3400 3401 int 3402 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3403 { 3404 int error; 3405 3406 mutex_enter(&spa->spa_vdev_top_lock); 3407 error = vdev_offline_locked(spa, guid, flags); 3408 mutex_exit(&spa->spa_vdev_top_lock); 3409 3410 return (error); 3411 } 3412 3413 /* 3414 * Clear the error counts associated with this vdev. Unlike vdev_online() and 3415 * vdev_offline(), we assume the spa config is locked. We also clear all 3416 * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3417 */ 3418 void 3419 vdev_clear(spa_t *spa, vdev_t *vd) 3420 { 3421 vdev_t *rvd = spa->spa_root_vdev; 3422 3423 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3424 3425 if (vd == NULL) 3426 vd = rvd; 3427 3428 vd->vdev_stat.vs_read_errors = 0; 3429 vd->vdev_stat.vs_write_errors = 0; 3430 vd->vdev_stat.vs_checksum_errors = 0; 3431 3432 for (int c = 0; c < vd->vdev_children; c++) 3433 vdev_clear(spa, vd->vdev_child[c]); 3434 3435 /* 3436 * It makes no sense to "clear" an indirect vdev. 3437 */ 3438 if (!vdev_is_concrete(vd)) 3439 return; 3440 3441 /* 3442 * If we're in the FAULTED state or have experienced failed I/O, then 3443 * clear the persistent state and attempt to reopen the device. We 3444 * also mark the vdev config dirty, so that the new faulted state is 3445 * written out to disk. 3446 */ 3447 if (vd->vdev_faulted || vd->vdev_degraded || 3448 !vdev_readable(vd) || !vdev_writeable(vd)) { 3449 3450 /* 3451 * When reopening in reponse to a clear event, it may be due to 3452 * a fmadm repair request. In this case, if the device is 3453 * still broken, we want to still post the ereport again. 3454 */ 3455 vd->vdev_forcefault = B_TRUE; 3456 3457 vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3458 vd->vdev_cant_read = B_FALSE; 3459 vd->vdev_cant_write = B_FALSE; 3460 3461 vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 3462 3463 vd->vdev_forcefault = B_FALSE; 3464 3465 if (vd != rvd && vdev_writeable(vd->vdev_top)) 3466 vdev_state_dirty(vd->vdev_top); 3467 3468 if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 3469 spa_async_request(spa, SPA_ASYNC_RESILVER); 3470 3471 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 3472 } 3473 3474 /* 3475 * When clearing a FMA-diagnosed fault, we always want to 3476 * unspare the device, as we assume that the original spare was 3477 * done in response to the FMA fault. 3478 */ 3479 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3480 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3481 vd->vdev_parent->vdev_child[0] == vd) 3482 vd->vdev_unspare = B_TRUE; 3483 } 3484 3485 boolean_t 3486 vdev_is_dead(vdev_t *vd) 3487 { 3488 /* 3489 * Holes and missing devices are always considered "dead". 3490 * This simplifies the code since we don't have to check for 3491 * these types of devices in the various code paths. 3492 * Instead we rely on the fact that we skip over dead devices 3493 * before issuing I/O to them. 3494 */ 3495 return (vd->vdev_state < VDEV_STATE_DEGRADED || 3496 vd->vdev_ops == &vdev_hole_ops || 3497 vd->vdev_ops == &vdev_missing_ops); 3498 } 3499 3500 boolean_t 3501 vdev_readable(vdev_t *vd) 3502 { 3503 return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 3504 } 3505 3506 boolean_t 3507 vdev_writeable(vdev_t *vd) 3508 { 3509 return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 3510 vdev_is_concrete(vd)); 3511 } 3512 3513 boolean_t 3514 vdev_allocatable(vdev_t *vd) 3515 { 3516 uint64_t state = vd->vdev_state; 3517 3518 /* 3519 * We currently allow allocations from vdevs which may be in the 3520 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3521 * fails to reopen then we'll catch it later when we're holding 3522 * the proper locks. Note that we have to get the vdev state 3523 * in a local variable because although it changes atomically, 3524 * we're asking two separate questions about it. 3525 */ 3526 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 3527 !vd->vdev_cant_write && vdev_is_concrete(vd) && 3528 vd->vdev_mg->mg_initialized); 3529 } 3530 3531 boolean_t 3532 vdev_accessible(vdev_t *vd, zio_t *zio) 3533 { 3534 ASSERT(zio->io_vd == vd); 3535 3536 if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3537 return (B_FALSE); 3538 3539 if (zio->io_type == ZIO_TYPE_READ) 3540 return (!vd->vdev_cant_read); 3541 3542 if (zio->io_type == ZIO_TYPE_WRITE) 3543 return (!vd->vdev_cant_write); 3544 3545 return (B_TRUE); 3546 } 3547 3548 boolean_t 3549 vdev_is_spacemap_addressable(vdev_t *vd) 3550 { 3551 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 3552 return (B_TRUE); 3553 3554 /* 3555 * If double-word space map entries are not enabled we assume 3556 * 47 bits of the space map entry are dedicated to the entry's 3557 * offset (see SM_OFFSET_BITS in space_map.h). We then use that 3558 * to calculate the maximum address that can be described by a 3559 * space map entry for the given device. 3560 */ 3561 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 3562 3563 if (shift >= 63) /* detect potential overflow */ 3564 return (B_TRUE); 3565 3566 return (vd->vdev_asize < (1ULL << shift)); 3567 } 3568 3569 /* 3570 * Get statistics for the given vdev. 3571 */ 3572 void 3573 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3574 { 3575 spa_t *spa = vd->vdev_spa; 3576 vdev_t *rvd = spa->spa_root_vdev; 3577 vdev_t *tvd = vd->vdev_top; 3578 3579 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 3580 3581 mutex_enter(&vd->vdev_stat_lock); 3582 bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3583 vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3584 vs->vs_state = vd->vdev_state; 3585 vs->vs_rsize = vdev_get_min_asize(vd); 3586 if (vd->vdev_ops->vdev_op_leaf) { 3587 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 3588 /* 3589 * Report intializing progress. Since we don't have the 3590 * initializing locks held, this is only an estimate (although a 3591 * fairly accurate one). 3592 */ 3593 vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done; 3594 vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est; 3595 vs->vs_initialize_state = vd->vdev_initialize_state; 3596 vs->vs_initialize_action_time = vd->vdev_initialize_action_time; 3597 } 3598 /* 3599 * Report expandable space on top-level, non-auxillary devices only. 3600 * The expandable space is reported in terms of metaslab sized units 3601 * since that determines how much space the pool can expand. 3602 */ 3603 if (vd->vdev_aux == NULL && tvd != NULL) { 3604 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize - 3605 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift); 3606 } 3607 if (vd->vdev_aux == NULL && vd == vd->vdev_top && 3608 vdev_is_concrete(vd)) { 3609 vs->vs_fragmentation = (vd->vdev_mg != NULL) ? 3610 vd->vdev_mg->mg_fragmentation : 0; 3611 } 3612 3613 /* 3614 * If we're getting stats on the root vdev, aggregate the I/O counts 3615 * over all top-level vdevs (i.e. the direct children of the root). 3616 */ 3617 if (vd == rvd) { 3618 for (int c = 0; c < rvd->vdev_children; c++) { 3619 vdev_t *cvd = rvd->vdev_child[c]; 3620 vdev_stat_t *cvs = &cvd->vdev_stat; 3621 3622 for (int t = 0; t < ZIO_TYPES; t++) { 3623 vs->vs_ops[t] += cvs->vs_ops[t]; 3624 vs->vs_bytes[t] += cvs->vs_bytes[t]; 3625 } 3626 cvs->vs_scan_removing = cvd->vdev_removing; 3627 } 3628 } 3629 mutex_exit(&vd->vdev_stat_lock); 3630 } 3631 3632 void 3633 vdev_clear_stats(vdev_t *vd) 3634 { 3635 mutex_enter(&vd->vdev_stat_lock); 3636 vd->vdev_stat.vs_space = 0; 3637 vd->vdev_stat.vs_dspace = 0; 3638 vd->vdev_stat.vs_alloc = 0; 3639 mutex_exit(&vd->vdev_stat_lock); 3640 } 3641 3642 void 3643 vdev_scan_stat_init(vdev_t *vd) 3644 { 3645 vdev_stat_t *vs = &vd->vdev_stat; 3646 3647 for (int c = 0; c < vd->vdev_children; c++) 3648 vdev_scan_stat_init(vd->vdev_child[c]); 3649 3650 mutex_enter(&vd->vdev_stat_lock); 3651 vs->vs_scan_processed = 0; 3652 mutex_exit(&vd->vdev_stat_lock); 3653 } 3654 3655 void 3656 vdev_stat_update(zio_t *zio, uint64_t psize) 3657 { 3658 spa_t *spa = zio->io_spa; 3659 vdev_t *rvd = spa->spa_root_vdev; 3660 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3661 vdev_t *pvd; 3662 uint64_t txg = zio->io_txg; 3663 vdev_stat_t *vs = &vd->vdev_stat; 3664 zio_type_t type = zio->io_type; 3665 int flags = zio->io_flags; 3666 3667 /* 3668 * If this i/o is a gang leader, it didn't do any actual work. 3669 */ 3670 if (zio->io_gang_tree) 3671 return; 3672 3673 if (zio->io_error == 0) { 3674 /* 3675 * If this is a root i/o, don't count it -- we've already 3676 * counted the top-level vdevs, and vdev_get_stats() will 3677 * aggregate them when asked. This reduces contention on 3678 * the root vdev_stat_lock and implicitly handles blocks 3679 * that compress away to holes, for which there is no i/o. 3680 * (Holes never create vdev children, so all the counters 3681 * remain zero, which is what we want.) 3682 * 3683 * Note: this only applies to successful i/o (io_error == 0) 3684 * because unlike i/o counts, errors are not additive. 3685 * When reading a ditto block, for example, failure of 3686 * one top-level vdev does not imply a root-level error. 3687 */ 3688 if (vd == rvd) 3689 return; 3690 3691 ASSERT(vd == zio->io_vd); 3692 3693 if (flags & ZIO_FLAG_IO_BYPASS) 3694 return; 3695 3696 mutex_enter(&vd->vdev_stat_lock); 3697 3698 if (flags & ZIO_FLAG_IO_REPAIR) { 3699 if (flags & ZIO_FLAG_SCAN_THREAD) { 3700 dsl_scan_phys_t *scn_phys = 3701 &spa->spa_dsl_pool->dp_scan->scn_phys; 3702 uint64_t *processed = &scn_phys->scn_processed; 3703 3704 /* XXX cleanup? */ 3705 if (vd->vdev_ops->vdev_op_leaf) 3706 atomic_add_64(processed, psize); 3707 vs->vs_scan_processed += psize; 3708 } 3709 3710 if (flags & ZIO_FLAG_SELF_HEAL) 3711 vs->vs_self_healed += psize; 3712 } 3713 3714 vs->vs_ops[type]++; 3715 vs->vs_bytes[type] += psize; 3716 3717 mutex_exit(&vd->vdev_stat_lock); 3718 return; 3719 } 3720 3721 if (flags & ZIO_FLAG_SPECULATIVE) 3722 return; 3723 3724 /* 3725 * If this is an I/O error that is going to be retried, then ignore the 3726 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 3727 * hard errors, when in reality they can happen for any number of 3728 * innocuous reasons (bus resets, MPxIO link failure, etc). 3729 */ 3730 if (zio->io_error == EIO && 3731 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 3732 return; 3733 3734 /* 3735 * Intent logs writes won't propagate their error to the root 3736 * I/O so don't mark these types of failures as pool-level 3737 * errors. 3738 */ 3739 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 3740 return; 3741 3742 mutex_enter(&vd->vdev_stat_lock); 3743 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3744 if (zio->io_error == ECKSUM) 3745 vs->vs_checksum_errors++; 3746 else 3747 vs->vs_read_errors++; 3748 } 3749 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3750 vs->vs_write_errors++; 3751 mutex_exit(&vd->vdev_stat_lock); 3752 3753 if (spa->spa_load_state == SPA_LOAD_NONE && 3754 type == ZIO_TYPE_WRITE && txg != 0 && 3755 (!(flags & ZIO_FLAG_IO_REPAIR) || 3756 (flags & ZIO_FLAG_SCAN_THREAD) || 3757 spa->spa_claiming)) { 3758 /* 3759 * This is either a normal write (not a repair), or it's 3760 * a repair induced by the scrub thread, or it's a repair 3761 * made by zil_claim() during spa_load() in the first txg. 3762 * In the normal case, we commit the DTL change in the same 3763 * txg as the block was born. In the scrub-induced repair 3764 * case, we know that scrubs run in first-pass syncing context, 3765 * so we commit the DTL change in spa_syncing_txg(spa). 3766 * In the zil_claim() case, we commit in spa_first_txg(spa). 3767 * 3768 * We currently do not make DTL entries for failed spontaneous 3769 * self-healing writes triggered by normal (non-scrubbing) 3770 * reads, because we have no transactional context in which to 3771 * do so -- and it's not clear that it'd be desirable anyway. 3772 */ 3773 if (vd->vdev_ops->vdev_op_leaf) { 3774 uint64_t commit_txg = txg; 3775 if (flags & ZIO_FLAG_SCAN_THREAD) { 3776 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3777 ASSERT(spa_sync_pass(spa) == 1); 3778 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 3779 commit_txg = spa_syncing_txg(spa); 3780 } else if (spa->spa_claiming) { 3781 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3782 commit_txg = spa_first_txg(spa); 3783 } 3784 ASSERT(commit_txg >= spa_syncing_txg(spa)); 3785 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 3786 return; 3787 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3788 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 3789 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 3790 } 3791 if (vd != rvd) 3792 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 3793 } 3794 } 3795 3796 int64_t 3797 vdev_deflated_space(vdev_t *vd, int64_t space) 3798 { 3799 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); 3800 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 3801 3802 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); 3803 } 3804 3805 /* 3806 * Update the in-core space usage stats for this vdev and the root vdev. 3807 */ 3808 void 3809 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 3810 int64_t space_delta) 3811 { 3812 int64_t dspace_delta; 3813 spa_t *spa = vd->vdev_spa; 3814 vdev_t *rvd = spa->spa_root_vdev; 3815 3816 ASSERT(vd == vd->vdev_top); 3817 3818 /* 3819 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 3820 * factor. We must calculate this here and not at the root vdev 3821 * because the root vdev's psize-to-asize is simply the max of its 3822 * childrens', thus not accurate enough for us. 3823 */ 3824 dspace_delta = vdev_deflated_space(vd, space_delta); 3825 3826 mutex_enter(&vd->vdev_stat_lock); 3827 vd->vdev_stat.vs_alloc += alloc_delta; 3828 vd->vdev_stat.vs_space += space_delta; 3829 vd->vdev_stat.vs_dspace += dspace_delta; 3830 mutex_exit(&vd->vdev_stat_lock); 3831 3832 /* every class but log contributes to root space stats */ 3833 if (vd->vdev_mg != NULL && !vd->vdev_islog) { 3834 mutex_enter(&rvd->vdev_stat_lock); 3835 rvd->vdev_stat.vs_alloc += alloc_delta; 3836 rvd->vdev_stat.vs_space += space_delta; 3837 rvd->vdev_stat.vs_dspace += dspace_delta; 3838 mutex_exit(&rvd->vdev_stat_lock); 3839 } 3840 /* Note: metaslab_class_space_update moved to metaslab_space_update */ 3841 } 3842 3843 /* 3844 * Mark a top-level vdev's config as dirty, placing it on the dirty list 3845 * so that it will be written out next time the vdev configuration is synced. 3846 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 3847 */ 3848 void 3849 vdev_config_dirty(vdev_t *vd) 3850 { 3851 spa_t *spa = vd->vdev_spa; 3852 vdev_t *rvd = spa->spa_root_vdev; 3853 int c; 3854 3855 ASSERT(spa_writeable(spa)); 3856 3857 /* 3858 * If this is an aux vdev (as with l2cache and spare devices), then we 3859 * update the vdev config manually and set the sync flag. 3860 */ 3861 if (vd->vdev_aux != NULL) { 3862 spa_aux_vdev_t *sav = vd->vdev_aux; 3863 nvlist_t **aux; 3864 uint_t naux; 3865 3866 for (c = 0; c < sav->sav_count; c++) { 3867 if (sav->sav_vdevs[c] == vd) 3868 break; 3869 } 3870 3871 if (c == sav->sav_count) { 3872 /* 3873 * We're being removed. There's nothing more to do. 3874 */ 3875 ASSERT(sav->sav_sync == B_TRUE); 3876 return; 3877 } 3878 3879 sav->sav_sync = B_TRUE; 3880 3881 if (nvlist_lookup_nvlist_array(sav->sav_config, 3882 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 3883 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 3884 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 3885 } 3886 3887 ASSERT(c < naux); 3888 3889 /* 3890 * Setting the nvlist in the middle if the array is a little 3891 * sketchy, but it will work. 3892 */ 3893 nvlist_free(aux[c]); 3894 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3895 3896 return; 3897 } 3898 3899 /* 3900 * The dirty list is protected by the SCL_CONFIG lock. The caller 3901 * must either hold SCL_CONFIG as writer, or must be the sync thread 3902 * (which holds SCL_CONFIG as reader). There's only one sync thread, 3903 * so this is sufficient to ensure mutual exclusion. 3904 */ 3905 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3906 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3907 spa_config_held(spa, SCL_CONFIG, RW_READER))); 3908 3909 if (vd == rvd) { 3910 for (c = 0; c < rvd->vdev_children; c++) 3911 vdev_config_dirty(rvd->vdev_child[c]); 3912 } else { 3913 ASSERT(vd == vd->vdev_top); 3914 3915 if (!list_link_active(&vd->vdev_config_dirty_node) && 3916 vdev_is_concrete(vd)) { 3917 list_insert_head(&spa->spa_config_dirty_list, vd); 3918 } 3919 } 3920 } 3921 3922 void 3923 vdev_config_clean(vdev_t *vd) 3924 { 3925 spa_t *spa = vd->vdev_spa; 3926 3927 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3928 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3929 spa_config_held(spa, SCL_CONFIG, RW_READER))); 3930 3931 ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 3932 list_remove(&spa->spa_config_dirty_list, vd); 3933 } 3934 3935 /* 3936 * Mark a top-level vdev's state as dirty, so that the next pass of 3937 * spa_sync() can convert this into vdev_config_dirty(). We distinguish 3938 * the state changes from larger config changes because they require 3939 * much less locking, and are often needed for administrative actions. 3940 */ 3941 void 3942 vdev_state_dirty(vdev_t *vd) 3943 { 3944 spa_t *spa = vd->vdev_spa; 3945 3946 ASSERT(spa_writeable(spa)); 3947 ASSERT(vd == vd->vdev_top); 3948 3949 /* 3950 * The state list is protected by the SCL_STATE lock. The caller 3951 * must either hold SCL_STATE as writer, or must be the sync thread 3952 * (which holds SCL_STATE as reader). There's only one sync thread, 3953 * so this is sufficient to ensure mutual exclusion. 3954 */ 3955 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3956 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3957 spa_config_held(spa, SCL_STATE, RW_READER))); 3958 3959 if (!list_link_active(&vd->vdev_state_dirty_node) && 3960 vdev_is_concrete(vd)) 3961 list_insert_head(&spa->spa_state_dirty_list, vd); 3962 } 3963 3964 void 3965 vdev_state_clean(vdev_t *vd) 3966 { 3967 spa_t *spa = vd->vdev_spa; 3968 3969 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3970 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3971 spa_config_held(spa, SCL_STATE, RW_READER))); 3972 3973 ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3974 list_remove(&spa->spa_state_dirty_list, vd); 3975 } 3976 3977 /* 3978 * Propagate vdev state up from children to parent. 3979 */ 3980 void 3981 vdev_propagate_state(vdev_t *vd) 3982 { 3983 spa_t *spa = vd->vdev_spa; 3984 vdev_t *rvd = spa->spa_root_vdev; 3985 int degraded = 0, faulted = 0; 3986 int corrupted = 0; 3987 vdev_t *child; 3988 3989 if (vd->vdev_children > 0) { 3990 for (int c = 0; c < vd->vdev_children; c++) { 3991 child = vd->vdev_child[c]; 3992 3993 /* 3994 * Don't factor holes or indirect vdevs into the 3995 * decision. 3996 */ 3997 if (!vdev_is_concrete(child)) 3998 continue; 3999 4000 if (!vdev_readable(child) || 4001 (!vdev_writeable(child) && spa_writeable(spa))) { 4002 /* 4003 * Root special: if there is a top-level log 4004 * device, treat the root vdev as if it were 4005 * degraded. 4006 */ 4007 if (child->vdev_islog && vd == rvd) 4008 degraded++; 4009 else 4010 faulted++; 4011 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 4012 degraded++; 4013 } 4014 4015 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 4016 corrupted++; 4017 } 4018 4019 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 4020 4021 /* 4022 * Root special: if there is a top-level vdev that cannot be 4023 * opened due to corrupted metadata, then propagate the root 4024 * vdev's aux state as 'corrupt' rather than 'insufficient 4025 * replicas'. 4026 */ 4027 if (corrupted && vd == rvd && 4028 rvd->vdev_state == VDEV_STATE_CANT_OPEN) 4029 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 4030 VDEV_AUX_CORRUPT_DATA); 4031 } 4032 4033 if (vd->vdev_parent) 4034 vdev_propagate_state(vd->vdev_parent); 4035 } 4036 4037 /* 4038 * Set a vdev's state. If this is during an open, we don't update the parent 4039 * state, because we're in the process of opening children depth-first. 4040 * Otherwise, we propagate the change to the parent. 4041 * 4042 * If this routine places a device in a faulted state, an appropriate ereport is 4043 * generated. 4044 */ 4045 void 4046 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 4047 { 4048 uint64_t save_state; 4049 spa_t *spa = vd->vdev_spa; 4050 4051 if (state == vd->vdev_state) { 4052 vd->vdev_stat.vs_aux = aux; 4053 return; 4054 } 4055 4056 save_state = vd->vdev_state; 4057 4058 vd->vdev_state = state; 4059 vd->vdev_stat.vs_aux = aux; 4060 4061 /* 4062 * If we are setting the vdev state to anything but an open state, then 4063 * always close the underlying device unless the device has requested 4064 * a delayed close (i.e. we're about to remove or fault the device). 4065 * Otherwise, we keep accessible but invalid devices open forever. 4066 * We don't call vdev_close() itself, because that implies some extra 4067 * checks (offline, etc) that we don't want here. This is limited to 4068 * leaf devices, because otherwise closing the device will affect other 4069 * children. 4070 */ 4071 if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 4072 vd->vdev_ops->vdev_op_leaf) 4073 vd->vdev_ops->vdev_op_close(vd); 4074 4075 /* 4076 * If we have brought this vdev back into service, we need 4077 * to notify fmd so that it can gracefully repair any outstanding 4078 * cases due to a missing device. We do this in all cases, even those 4079 * that probably don't correlate to a repaired fault. This is sure to 4080 * catch all cases, and we let the zfs-retire agent sort it out. If 4081 * this is a transient state it's OK, as the retire agent will 4082 * double-check the state of the vdev before repairing it. 4083 */ 4084 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 4085 vd->vdev_prevstate != state) 4086 zfs_post_state_change(spa, vd); 4087 4088 if (vd->vdev_removed && 4089 state == VDEV_STATE_CANT_OPEN && 4090 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 4091 /* 4092 * If the previous state is set to VDEV_STATE_REMOVED, then this 4093 * device was previously marked removed and someone attempted to 4094 * reopen it. If this failed due to a nonexistent device, then 4095 * keep the device in the REMOVED state. We also let this be if 4096 * it is one of our special test online cases, which is only 4097 * attempting to online the device and shouldn't generate an FMA 4098 * fault. 4099 */ 4100 vd->vdev_state = VDEV_STATE_REMOVED; 4101 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 4102 } else if (state == VDEV_STATE_REMOVED) { 4103 vd->vdev_removed = B_TRUE; 4104 } else if (state == VDEV_STATE_CANT_OPEN) { 4105 /* 4106 * If we fail to open a vdev during an import or recovery, we 4107 * mark it as "not available", which signifies that it was 4108 * never there to begin with. Failure to open such a device 4109 * is not considered an error. 4110 */ 4111 if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 4112 spa_load_state(spa) == SPA_LOAD_RECOVER) && 4113 vd->vdev_ops->vdev_op_leaf) 4114 vd->vdev_not_present = 1; 4115 4116 /* 4117 * Post the appropriate ereport. If the 'prevstate' field is 4118 * set to something other than VDEV_STATE_UNKNOWN, it indicates 4119 * that this is part of a vdev_reopen(). In this case, we don't 4120 * want to post the ereport if the device was already in the 4121 * CANT_OPEN state beforehand. 4122 * 4123 * If the 'checkremove' flag is set, then this is an attempt to 4124 * online the device in response to an insertion event. If we 4125 * hit this case, then we have detected an insertion event for a 4126 * faulted or offline device that wasn't in the removed state. 4127 * In this scenario, we don't post an ereport because we are 4128 * about to replace the device, or attempt an online with 4129 * vdev_forcefault, which will generate the fault for us. 4130 */ 4131 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 4132 !vd->vdev_not_present && !vd->vdev_checkremove && 4133 vd != spa->spa_root_vdev) { 4134 const char *class; 4135 4136 switch (aux) { 4137 case VDEV_AUX_OPEN_FAILED: 4138 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 4139 break; 4140 case VDEV_AUX_CORRUPT_DATA: 4141 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 4142 break; 4143 case VDEV_AUX_NO_REPLICAS: 4144 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 4145 break; 4146 case VDEV_AUX_BAD_GUID_SUM: 4147 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 4148 break; 4149 case VDEV_AUX_TOO_SMALL: 4150 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 4151 break; 4152 case VDEV_AUX_BAD_LABEL: 4153 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 4154 break; 4155 default: 4156 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 4157 } 4158 4159 zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 4160 } 4161 4162 /* Erase any notion of persistent removed state */ 4163 vd->vdev_removed = B_FALSE; 4164 } else { 4165 vd->vdev_removed = B_FALSE; 4166 } 4167 4168 if (!isopen && vd->vdev_parent) 4169 vdev_propagate_state(vd->vdev_parent); 4170 } 4171 4172 boolean_t 4173 vdev_children_are_offline(vdev_t *vd) 4174 { 4175 ASSERT(!vd->vdev_ops->vdev_op_leaf); 4176 4177 for (uint64_t i = 0; i < vd->vdev_children; i++) { 4178 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 4179 return (B_FALSE); 4180 } 4181 4182 return (B_TRUE); 4183 } 4184 4185 /* 4186 * Check the vdev configuration to ensure that it's capable of supporting 4187 * a root pool. We do not support partial configuration. 4188 * In addition, only a single top-level vdev is allowed. 4189 */ 4190 boolean_t 4191 vdev_is_bootable(vdev_t *vd) 4192 { 4193 if (!vd->vdev_ops->vdev_op_leaf) { 4194 char *vdev_type = vd->vdev_ops->vdev_op_type; 4195 4196 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 4197 vd->vdev_children > 1) { 4198 return (B_FALSE); 4199 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 || 4200 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) { 4201 return (B_FALSE); 4202 } 4203 } 4204 4205 for (int c = 0; c < vd->vdev_children; c++) { 4206 if (!vdev_is_bootable(vd->vdev_child[c])) 4207 return (B_FALSE); 4208 } 4209 return (B_TRUE); 4210 } 4211 4212 boolean_t 4213 vdev_is_concrete(vdev_t *vd) 4214 { 4215 vdev_ops_t *ops = vd->vdev_ops; 4216 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 4217 ops == &vdev_missing_ops || ops == &vdev_root_ops) { 4218 return (B_FALSE); 4219 } else { 4220 return (B_TRUE); 4221 } 4222 } 4223 4224 /* 4225 * Determine if a log device has valid content. If the vdev was 4226 * removed or faulted in the MOS config then we know that 4227 * the content on the log device has already been written to the pool. 4228 */ 4229 boolean_t 4230 vdev_log_state_valid(vdev_t *vd) 4231 { 4232 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 4233 !vd->vdev_removed) 4234 return (B_TRUE); 4235 4236 for (int c = 0; c < vd->vdev_children; c++) 4237 if (vdev_log_state_valid(vd->vdev_child[c])) 4238 return (B_TRUE); 4239 4240 return (B_FALSE); 4241 } 4242 4243 /* 4244 * Expand a vdev if possible. 4245 */ 4246 void 4247 vdev_expand(vdev_t *vd, uint64_t txg) 4248 { 4249 ASSERT(vd->vdev_top == vd); 4250 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4251 ASSERT(vdev_is_concrete(vd)); 4252 4253 vdev_set_deflate_ratio(vd); 4254 4255 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 4256 vdev_is_concrete(vd)) { 4257 vdev_metaslab_group_create(vd); 4258 VERIFY(vdev_metaslab_init(vd, txg) == 0); 4259 vdev_config_dirty(vd); 4260 } 4261 } 4262 4263 /* 4264 * Split a vdev. 4265 */ 4266 void 4267 vdev_split(vdev_t *vd) 4268 { 4269 vdev_t *cvd, *pvd = vd->vdev_parent; 4270 4271 vdev_remove_child(pvd, vd); 4272 vdev_compact_children(pvd); 4273 4274 cvd = pvd->vdev_child[0]; 4275 if (pvd->vdev_children == 1) { 4276 vdev_remove_parent(cvd); 4277 cvd->vdev_splitting = B_TRUE; 4278 } 4279 vdev_propagate_state(cvd); 4280 } 4281 4282 void 4283 vdev_deadman(vdev_t *vd) 4284 { 4285 for (int c = 0; c < vd->vdev_children; c++) { 4286 vdev_t *cvd = vd->vdev_child[c]; 4287 4288 vdev_deadman(cvd); 4289 } 4290 4291 if (vd->vdev_ops->vdev_op_leaf) { 4292 vdev_queue_t *vq = &vd->vdev_queue; 4293 4294 mutex_enter(&vq->vq_lock); 4295 if (avl_numnodes(&vq->vq_active_tree) > 0) { 4296 spa_t *spa = vd->vdev_spa; 4297 zio_t *fio; 4298 uint64_t delta; 4299 4300 /* 4301 * Look at the head of all the pending queues, 4302 * if any I/O has been outstanding for longer than 4303 * the spa_deadman_synctime we panic the system. 4304 */ 4305 fio = avl_first(&vq->vq_active_tree); 4306 delta = gethrtime() - fio->io_timestamp; 4307 if (delta > spa_deadman_synctime(spa)) { 4308 vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 4309 "%lluns, delta %lluns, last io %lluns", 4310 fio->io_timestamp, (u_longlong_t)delta, 4311 vq->vq_io_complete_ts); 4312 fm_panic("I/O to pool '%s' appears to be " 4313 "hung.", spa_name(spa)); 4314 } 4315 } 4316 mutex_exit(&vq->vq_lock); 4317 } 4318 } 4319