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 retry_sync: 1482 for (int c = 0; c < children; c++) 1483 vd->vdev_child[c]->vdev_open_error = 1484 vdev_open(vd->vdev_child[c]); 1485 } else { 1486 tq = taskq_create("vdev_open", children, minclsyspri, 1487 children, children, TASKQ_PREPOPULATE); 1488 if (tq == NULL) 1489 goto retry_sync; 1490 1491 for (int c = 0; c < children; c++) 1492 VERIFY(taskq_dispatch(tq, vdev_open_child, 1493 vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID); 1494 1495 taskq_destroy(tq); 1496 } 1497 1498 vd->vdev_nonrot = B_TRUE; 1499 1500 for (int c = 0; c < children; c++) 1501 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; 1502 } 1503 1504 /* 1505 * Compute the raidz-deflation ratio. Note, we hard-code 1506 * in 128k (1 << 17) because it is the "typical" blocksize. 1507 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 1508 * otherwise it would inconsistently account for existing bp's. 1509 */ 1510 static void 1511 vdev_set_deflate_ratio(vdev_t *vd) 1512 { 1513 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 1514 vd->vdev_deflate_ratio = (1 << 17) / 1515 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 1516 } 1517 } 1518 1519 /* 1520 * Prepare a virtual device for access. 1521 */ 1522 int 1523 vdev_open(vdev_t *vd) 1524 { 1525 spa_t *spa = vd->vdev_spa; 1526 int error; 1527 uint64_t osize = 0; 1528 uint64_t max_osize = 0; 1529 uint64_t asize, max_asize, psize; 1530 uint64_t ashift = 0; 1531 1532 ASSERT(vd->vdev_open_thread == curthread || 1533 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1534 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1535 vd->vdev_state == VDEV_STATE_CANT_OPEN || 1536 vd->vdev_state == VDEV_STATE_OFFLINE); 1537 1538 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1539 vd->vdev_cant_read = B_FALSE; 1540 vd->vdev_cant_write = B_FALSE; 1541 vd->vdev_min_asize = vdev_get_min_asize(vd); 1542 1543 /* 1544 * If this vdev is not removed, check its fault status. If it's 1545 * faulted, bail out of the open. 1546 */ 1547 if (!vd->vdev_removed && vd->vdev_faulted) { 1548 ASSERT(vd->vdev_children == 0); 1549 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1550 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1551 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1552 vd->vdev_label_aux); 1553 return (SET_ERROR(ENXIO)); 1554 } else if (vd->vdev_offline) { 1555 ASSERT(vd->vdev_children == 0); 1556 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1557 return (SET_ERROR(ENXIO)); 1558 } 1559 1560 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1561 1562 /* 1563 * Reset the vdev_reopening flag so that we actually close 1564 * the vdev on error. 1565 */ 1566 vd->vdev_reopening = B_FALSE; 1567 if (zio_injection_enabled && error == 0) 1568 error = zio_handle_device_injection(vd, NULL, ENXIO); 1569 1570 if (error) { 1571 if (vd->vdev_removed && 1572 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1573 vd->vdev_removed = B_FALSE; 1574 1575 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 1576 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 1577 vd->vdev_stat.vs_aux); 1578 } else { 1579 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1580 vd->vdev_stat.vs_aux); 1581 } 1582 return (error); 1583 } 1584 1585 vd->vdev_removed = B_FALSE; 1586 1587 /* 1588 * Recheck the faulted flag now that we have confirmed that 1589 * the vdev is accessible. If we're faulted, bail. 1590 */ 1591 if (vd->vdev_faulted) { 1592 ASSERT(vd->vdev_children == 0); 1593 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1594 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1595 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1596 vd->vdev_label_aux); 1597 return (SET_ERROR(ENXIO)); 1598 } 1599 1600 if (vd->vdev_degraded) { 1601 ASSERT(vd->vdev_children == 0); 1602 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1603 VDEV_AUX_ERR_EXCEEDED); 1604 } else { 1605 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 1606 } 1607 1608 /* 1609 * For hole or missing vdevs we just return success. 1610 */ 1611 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 1612 return (0); 1613 1614 for (int c = 0; c < vd->vdev_children; c++) { 1615 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1616 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1617 VDEV_AUX_NONE); 1618 break; 1619 } 1620 } 1621 1622 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1623 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1624 1625 if (vd->vdev_children == 0) { 1626 if (osize < SPA_MINDEVSIZE) { 1627 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1628 VDEV_AUX_TOO_SMALL); 1629 return (SET_ERROR(EOVERFLOW)); 1630 } 1631 psize = osize; 1632 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1633 max_asize = max_osize - (VDEV_LABEL_START_SIZE + 1634 VDEV_LABEL_END_SIZE); 1635 } else { 1636 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1637 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1638 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1639 VDEV_AUX_TOO_SMALL); 1640 return (SET_ERROR(EOVERFLOW)); 1641 } 1642 psize = 0; 1643 asize = osize; 1644 max_asize = max_osize; 1645 } 1646 1647 vd->vdev_psize = psize; 1648 1649 /* 1650 * Make sure the allocatable size hasn't shrunk too much. 1651 */ 1652 if (asize < vd->vdev_min_asize) { 1653 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1654 VDEV_AUX_BAD_LABEL); 1655 return (SET_ERROR(EINVAL)); 1656 } 1657 1658 if (vd->vdev_asize == 0) { 1659 /* 1660 * This is the first-ever open, so use the computed values. 1661 * For testing purposes, a higher ashift can be requested. 1662 */ 1663 vd->vdev_asize = asize; 1664 vd->vdev_max_asize = max_asize; 1665 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1666 vd->vdev_ashift = MAX(zfs_ashift_min, vd->vdev_ashift); 1667 } else { 1668 /* 1669 * Detect if the alignment requirement has increased. 1670 * We don't want to make the pool unavailable, just 1671 * issue a warning instead. 1672 */ 1673 if (ashift > vd->vdev_top->vdev_ashift && 1674 vd->vdev_ops->vdev_op_leaf) { 1675 cmn_err(CE_WARN, 1676 "Disk, '%s', has a block alignment that is " 1677 "larger than the pool's alignment\n", 1678 vd->vdev_path); 1679 } 1680 vd->vdev_max_asize = max_asize; 1681 } 1682 1683 /* 1684 * If all children are healthy we update asize if either: 1685 * The asize has increased, due to a device expansion caused by dynamic 1686 * LUN growth or vdev replacement, and automatic expansion is enabled; 1687 * making the additional space available. 1688 * 1689 * The asize has decreased, due to a device shrink usually caused by a 1690 * vdev replace with a smaller device. This ensures that calculations 1691 * based of max_asize and asize e.g. esize are always valid. It's safe 1692 * to do this as we've already validated that asize is greater than 1693 * vdev_min_asize. 1694 */ 1695 if (vd->vdev_state == VDEV_STATE_HEALTHY && 1696 ((asize > vd->vdev_asize && 1697 (vd->vdev_expanding || spa->spa_autoexpand)) || 1698 (asize < vd->vdev_asize))) 1699 vd->vdev_asize = asize; 1700 1701 vdev_set_min_asize(vd); 1702 1703 /* 1704 * Ensure we can issue some IO before declaring the 1705 * vdev open for business. 1706 */ 1707 if (vd->vdev_ops->vdev_op_leaf && 1708 (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1709 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1710 VDEV_AUX_ERR_EXCEEDED); 1711 return (error); 1712 } 1713 1714 /* 1715 * Track the min and max ashift values for normal data devices. 1716 * 1717 * DJB - TBD these should perhaps be tracked per allocation class 1718 * (e.g. spa_min_ashift is used to round up post compression buffers) 1719 */ 1720 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1721 vd->vdev_alloc_bias == VDEV_BIAS_NONE && 1722 vd->vdev_aux == NULL) { 1723 if (vd->vdev_ashift > spa->spa_max_ashift) 1724 spa->spa_max_ashift = vd->vdev_ashift; 1725 if (vd->vdev_ashift < spa->spa_min_ashift) 1726 spa->spa_min_ashift = vd->vdev_ashift; 1727 } 1728 1729 /* 1730 * If a leaf vdev has a DTL, and seems healthy, then kick off a 1731 * resilver. But don't do this if we are doing a reopen for a scrub, 1732 * since this would just restart the scrub we are already doing. 1733 */ 1734 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1735 vdev_resilver_needed(vd, NULL, NULL)) 1736 spa_async_request(spa, SPA_ASYNC_RESILVER); 1737 1738 return (0); 1739 } 1740 1741 /* 1742 * Called once the vdevs are all opened, this routine validates the label 1743 * contents. This needs to be done before vdev_load() so that we don't 1744 * inadvertently do repair I/Os to the wrong device. 1745 * 1746 * This function will only return failure if one of the vdevs indicates that it 1747 * has since been destroyed or exported. This is only possible if 1748 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1749 * will be updated but the function will return 0. 1750 */ 1751 int 1752 vdev_validate(vdev_t *vd) 1753 { 1754 spa_t *spa = vd->vdev_spa; 1755 nvlist_t *label; 1756 uint64_t guid = 0, aux_guid = 0, top_guid; 1757 uint64_t state; 1758 nvlist_t *nvl; 1759 uint64_t txg; 1760 1761 if (vdev_validate_skip) 1762 return (0); 1763 1764 for (uint64_t c = 0; c < vd->vdev_children; c++) 1765 if (vdev_validate(vd->vdev_child[c]) != 0) 1766 return (SET_ERROR(EBADF)); 1767 1768 /* 1769 * If the device has already failed, or was marked offline, don't do 1770 * any further validation. Otherwise, label I/O will fail and we will 1771 * overwrite the previous state. 1772 */ 1773 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 1774 return (0); 1775 1776 /* 1777 * If we are performing an extreme rewind, we allow for a label that 1778 * was modified at a point after the current txg. 1779 * If config lock is not held do not check for the txg. spa_sync could 1780 * be updating the vdev's label before updating spa_last_synced_txg. 1781 */ 1782 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 1783 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 1784 txg = UINT64_MAX; 1785 else 1786 txg = spa_last_synced_txg(spa); 1787 1788 if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1789 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1790 VDEV_AUX_BAD_LABEL); 1791 vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 1792 "txg %llu", (u_longlong_t)txg); 1793 return (0); 1794 } 1795 1796 /* 1797 * Determine if this vdev has been split off into another 1798 * pool. If so, then refuse to open it. 1799 */ 1800 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1801 &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1802 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1803 VDEV_AUX_SPLIT_POOL); 1804 nvlist_free(label); 1805 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 1806 return (0); 1807 } 1808 1809 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 1810 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1811 VDEV_AUX_CORRUPT_DATA); 1812 nvlist_free(label); 1813 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1814 ZPOOL_CONFIG_POOL_GUID); 1815 return (0); 1816 } 1817 1818 /* 1819 * If config is not trusted then ignore the spa guid check. This is 1820 * necessary because if the machine crashed during a re-guid the new 1821 * guid might have been written to all of the vdev labels, but not the 1822 * cached config. The check will be performed again once we have the 1823 * trusted config from the MOS. 1824 */ 1825 if (spa->spa_trust_config && guid != spa_guid(spa)) { 1826 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1827 VDEV_AUX_CORRUPT_DATA); 1828 nvlist_free(label); 1829 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 1830 "match config (%llu != %llu)", (u_longlong_t)guid, 1831 (u_longlong_t)spa_guid(spa)); 1832 return (0); 1833 } 1834 1835 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1836 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1837 &aux_guid) != 0) 1838 aux_guid = 0; 1839 1840 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 1841 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1842 VDEV_AUX_CORRUPT_DATA); 1843 nvlist_free(label); 1844 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1845 ZPOOL_CONFIG_GUID); 1846 return (0); 1847 } 1848 1849 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 1850 != 0) { 1851 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1852 VDEV_AUX_CORRUPT_DATA); 1853 nvlist_free(label); 1854 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1855 ZPOOL_CONFIG_TOP_GUID); 1856 return (0); 1857 } 1858 1859 /* 1860 * If this vdev just became a top-level vdev because its sibling was 1861 * detached, it will have adopted the parent's vdev guid -- but the 1862 * label may or may not be on disk yet. Fortunately, either version 1863 * of the label will have the same top guid, so if we're a top-level 1864 * vdev, we can safely compare to that instead. 1865 * However, if the config comes from a cachefile that failed to update 1866 * after the detach, a top-level vdev will appear as a non top-level 1867 * vdev in the config. Also relax the constraints if we perform an 1868 * extreme rewind. 1869 * 1870 * If we split this vdev off instead, then we also check the 1871 * original pool's guid. We don't want to consider the vdev 1872 * corrupt if it is partway through a split operation. 1873 */ 1874 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 1875 boolean_t mismatch = B_FALSE; 1876 if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 1877 if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 1878 mismatch = B_TRUE; 1879 } else { 1880 if (vd->vdev_guid != top_guid && 1881 vd->vdev_top->vdev_guid != guid) 1882 mismatch = B_TRUE; 1883 } 1884 1885 if (mismatch) { 1886 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1887 VDEV_AUX_CORRUPT_DATA); 1888 nvlist_free(label); 1889 vdev_dbgmsg(vd, "vdev_validate: config guid " 1890 "doesn't match label guid"); 1891 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 1892 (u_longlong_t)vd->vdev_guid, 1893 (u_longlong_t)vd->vdev_top->vdev_guid); 1894 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 1895 "aux_guid %llu", (u_longlong_t)guid, 1896 (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1897 return (0); 1898 } 1899 } 1900 1901 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1902 &state) != 0) { 1903 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1904 VDEV_AUX_CORRUPT_DATA); 1905 nvlist_free(label); 1906 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1907 ZPOOL_CONFIG_POOL_STATE); 1908 return (0); 1909 } 1910 1911 nvlist_free(label); 1912 1913 /* 1914 * If this is a verbatim import, no need to check the 1915 * state of the pool. 1916 */ 1917 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1918 spa_load_state(spa) == SPA_LOAD_OPEN && 1919 state != POOL_STATE_ACTIVE) { 1920 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 1921 "for spa %s", (u_longlong_t)state, spa->spa_name); 1922 return (SET_ERROR(EBADF)); 1923 } 1924 1925 /* 1926 * If we were able to open and validate a vdev that was 1927 * previously marked permanently unavailable, clear that state 1928 * now. 1929 */ 1930 if (vd->vdev_not_present) 1931 vd->vdev_not_present = 0; 1932 1933 return (0); 1934 } 1935 1936 static void 1937 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 1938 { 1939 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 1940 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 1941 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 1942 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 1943 dvd->vdev_path, svd->vdev_path); 1944 spa_strfree(dvd->vdev_path); 1945 dvd->vdev_path = spa_strdup(svd->vdev_path); 1946 } 1947 } else if (svd->vdev_path != NULL) { 1948 dvd->vdev_path = spa_strdup(svd->vdev_path); 1949 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 1950 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 1951 } 1952 } 1953 1954 /* 1955 * Recursively copy vdev paths from one vdev to another. Source and destination 1956 * vdev trees must have same geometry otherwise return error. Intended to copy 1957 * paths from userland config into MOS config. 1958 */ 1959 int 1960 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 1961 { 1962 if ((svd->vdev_ops == &vdev_missing_ops) || 1963 (svd->vdev_ishole && dvd->vdev_ishole) || 1964 (dvd->vdev_ops == &vdev_indirect_ops)) 1965 return (0); 1966 1967 if (svd->vdev_ops != dvd->vdev_ops) { 1968 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 1969 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 1970 return (SET_ERROR(EINVAL)); 1971 } 1972 1973 if (svd->vdev_guid != dvd->vdev_guid) { 1974 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 1975 "%llu)", (u_longlong_t)svd->vdev_guid, 1976 (u_longlong_t)dvd->vdev_guid); 1977 return (SET_ERROR(EINVAL)); 1978 } 1979 1980 if (svd->vdev_children != dvd->vdev_children) { 1981 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 1982 "%llu != %llu", (u_longlong_t)svd->vdev_children, 1983 (u_longlong_t)dvd->vdev_children); 1984 return (SET_ERROR(EINVAL)); 1985 } 1986 1987 for (uint64_t i = 0; i < svd->vdev_children; i++) { 1988 int error = vdev_copy_path_strict(svd->vdev_child[i], 1989 dvd->vdev_child[i]); 1990 if (error != 0) 1991 return (error); 1992 } 1993 1994 if (svd->vdev_ops->vdev_op_leaf) 1995 vdev_copy_path_impl(svd, dvd); 1996 1997 return (0); 1998 } 1999 2000 static void 2001 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 2002 { 2003 ASSERT(stvd->vdev_top == stvd); 2004 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 2005 2006 for (uint64_t i = 0; i < dvd->vdev_children; i++) { 2007 vdev_copy_path_search(stvd, dvd->vdev_child[i]); 2008 } 2009 2010 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 2011 return; 2012 2013 /* 2014 * The idea here is that while a vdev can shift positions within 2015 * a top vdev (when replacing, attaching mirror, etc.) it cannot 2016 * step outside of it. 2017 */ 2018 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 2019 2020 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 2021 return; 2022 2023 ASSERT(vd->vdev_ops->vdev_op_leaf); 2024 2025 vdev_copy_path_impl(vd, dvd); 2026 } 2027 2028 /* 2029 * Recursively copy vdev paths from one root vdev to another. Source and 2030 * destination vdev trees may differ in geometry. For each destination leaf 2031 * vdev, search a vdev with the same guid and top vdev id in the source. 2032 * Intended to copy paths from userland config into MOS config. 2033 */ 2034 void 2035 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 2036 { 2037 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 2038 ASSERT(srvd->vdev_ops == &vdev_root_ops); 2039 ASSERT(drvd->vdev_ops == &vdev_root_ops); 2040 2041 for (uint64_t i = 0; i < children; i++) { 2042 vdev_copy_path_search(srvd->vdev_child[i], 2043 drvd->vdev_child[i]); 2044 } 2045 } 2046 2047 /* 2048 * Close a virtual device. 2049 */ 2050 void 2051 vdev_close(vdev_t *vd) 2052 { 2053 spa_t *spa = vd->vdev_spa; 2054 vdev_t *pvd = vd->vdev_parent; 2055 2056 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2057 2058 /* 2059 * If our parent is reopening, then we are as well, unless we are 2060 * going offline. 2061 */ 2062 if (pvd != NULL && pvd->vdev_reopening) 2063 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 2064 2065 vd->vdev_ops->vdev_op_close(vd); 2066 2067 vdev_cache_purge(vd); 2068 2069 /* 2070 * We record the previous state before we close it, so that if we are 2071 * doing a reopen(), we don't generate FMA ereports if we notice that 2072 * it's still faulted. 2073 */ 2074 vd->vdev_prevstate = vd->vdev_state; 2075 2076 if (vd->vdev_offline) 2077 vd->vdev_state = VDEV_STATE_OFFLINE; 2078 else 2079 vd->vdev_state = VDEV_STATE_CLOSED; 2080 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2081 } 2082 2083 void 2084 vdev_hold(vdev_t *vd) 2085 { 2086 spa_t *spa = vd->vdev_spa; 2087 2088 ASSERT(spa_is_root(spa)); 2089 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 2090 return; 2091 2092 for (int c = 0; c < vd->vdev_children; c++) 2093 vdev_hold(vd->vdev_child[c]); 2094 2095 if (vd->vdev_ops->vdev_op_leaf) 2096 vd->vdev_ops->vdev_op_hold(vd); 2097 } 2098 2099 void 2100 vdev_rele(vdev_t *vd) 2101 { 2102 spa_t *spa = vd->vdev_spa; 2103 2104 ASSERT(spa_is_root(spa)); 2105 for (int c = 0; c < vd->vdev_children; c++) 2106 vdev_rele(vd->vdev_child[c]); 2107 2108 if (vd->vdev_ops->vdev_op_leaf) 2109 vd->vdev_ops->vdev_op_rele(vd); 2110 } 2111 2112 /* 2113 * Reopen all interior vdevs and any unopened leaves. We don't actually 2114 * reopen leaf vdevs which had previously been opened as they might deadlock 2115 * on the spa_config_lock. Instead we only obtain the leaf's physical size. 2116 * If the leaf has never been opened then open it, as usual. 2117 */ 2118 void 2119 vdev_reopen(vdev_t *vd) 2120 { 2121 spa_t *spa = vd->vdev_spa; 2122 2123 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2124 2125 /* set the reopening flag unless we're taking the vdev offline */ 2126 vd->vdev_reopening = !vd->vdev_offline; 2127 vdev_close(vd); 2128 (void) vdev_open(vd); 2129 2130 /* 2131 * Call vdev_validate() here to make sure we have the same device. 2132 * Otherwise, a device with an invalid label could be successfully 2133 * opened in response to vdev_reopen(). 2134 */ 2135 if (vd->vdev_aux) { 2136 (void) vdev_validate_aux(vd); 2137 if (vdev_readable(vd) && vdev_writeable(vd) && 2138 vd->vdev_aux == &spa->spa_l2cache && 2139 !l2arc_vdev_present(vd)) 2140 l2arc_add_vdev(spa, vd); 2141 } else { 2142 (void) vdev_validate(vd); 2143 } 2144 2145 /* 2146 * Reassess parent vdev's health. 2147 */ 2148 vdev_propagate_state(vd); 2149 } 2150 2151 int 2152 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2153 { 2154 int error; 2155 2156 /* 2157 * Normally, partial opens (e.g. of a mirror) are allowed. 2158 * For a create, however, we want to fail the request if 2159 * there are any components we can't open. 2160 */ 2161 error = vdev_open(vd); 2162 2163 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2164 vdev_close(vd); 2165 return (error ? error : ENXIO); 2166 } 2167 2168 /* 2169 * Recursively load DTLs and initialize all labels. 2170 */ 2171 if ((error = vdev_dtl_load(vd)) != 0 || 2172 (error = vdev_label_init(vd, txg, isreplacing ? 2173 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2174 vdev_close(vd); 2175 return (error); 2176 } 2177 2178 return (0); 2179 } 2180 2181 void 2182 vdev_metaslab_set_size(vdev_t *vd) 2183 { 2184 uint64_t asize = vd->vdev_asize; 2185 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2186 uint64_t ms_shift; 2187 2188 /* BEGIN CSTYLED */ 2189 /* 2190 * There are two dimensions to the metaslab sizing calculation: 2191 * the size of the metaslab and the count of metaslabs per vdev. 2192 * 2193 * The default values used below are a good balance between memory 2194 * usage (larger metaslab size means more memory needed for loaded 2195 * metaslabs; more metaslabs means more memory needed for the 2196 * metaslab_t structs), metaslab load time (larger metaslabs take 2197 * longer to load), and metaslab sync time (more metaslabs means 2198 * more time spent syncing all of them). 2199 * 2200 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2201 * The range of the dimensions are as follows: 2202 * 2203 * 2^29 <= ms_size <= 2^34 2204 * 16 <= ms_count <= 131,072 2205 * 2206 * On the lower end of vdev sizes, we aim for metaslabs sizes of 2207 * at least 512MB (2^29) to minimize fragmentation effects when 2208 * testing with smaller devices. However, the count constraint 2209 * of at least 16 metaslabs will override this minimum size goal. 2210 * 2211 * On the upper end of vdev sizes, we aim for a maximum metaslab 2212 * size of 16GB. However, we will cap the total count to 2^17 2213 * metaslabs to keep our memory footprint in check and let the 2214 * metaslab size grow from there if that limit is hit. 2215 * 2216 * The net effect of applying above constrains is summarized below. 2217 * 2218 * vdev size metaslab count 2219 * --------------|----------------- 2220 * < 8GB ~16 2221 * 8GB - 100GB one per 512MB 2222 * 100GB - 3TB ~200 2223 * 3TB - 2PB one per 16GB 2224 * > 2PB ~131,072 2225 * -------------------------------- 2226 * 2227 * Finally, note that all of the above calculate the initial 2228 * number of metaslabs. Expanding a top-level vdev will result 2229 * in additional metaslabs being allocated making it possible 2230 * to exceed the zfs_vdev_ms_count_limit. 2231 */ 2232 /* END CSTYLED */ 2233 2234 if (ms_count < zfs_vdev_min_ms_count) 2235 ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2236 else if (ms_count > zfs_vdev_default_ms_count) 2237 ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2238 else 2239 ms_shift = zfs_vdev_default_ms_shift; 2240 2241 if (ms_shift < SPA_MAXBLOCKSHIFT) { 2242 ms_shift = SPA_MAXBLOCKSHIFT; 2243 } else if (ms_shift > zfs_vdev_max_ms_shift) { 2244 ms_shift = zfs_vdev_max_ms_shift; 2245 /* cap the total count to constrain memory footprint */ 2246 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2247 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 2248 } 2249 2250 vd->vdev_ms_shift = ms_shift; 2251 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2252 } 2253 2254 void 2255 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2256 { 2257 ASSERT(vd == vd->vdev_top); 2258 /* indirect vdevs don't have metaslabs or dtls */ 2259 ASSERT(vdev_is_concrete(vd) || flags == 0); 2260 ASSERT(ISP2(flags)); 2261 ASSERT(spa_writeable(vd->vdev_spa)); 2262 2263 if (flags & VDD_METASLAB) 2264 (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2265 2266 if (flags & VDD_DTL) 2267 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2268 2269 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2270 } 2271 2272 void 2273 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 2274 { 2275 for (int c = 0; c < vd->vdev_children; c++) 2276 vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 2277 2278 if (vd->vdev_ops->vdev_op_leaf) 2279 vdev_dirty(vd->vdev_top, flags, vd, txg); 2280 } 2281 2282 /* 2283 * DTLs. 2284 * 2285 * A vdev's DTL (dirty time log) is the set of transaction groups for which 2286 * the vdev has less than perfect replication. There are four kinds of DTL: 2287 * 2288 * DTL_MISSING: txgs for which the vdev has no valid copies of the data 2289 * 2290 * DTL_PARTIAL: txgs for which data is available, but not fully replicated 2291 * 2292 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 2293 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 2294 * txgs that was scrubbed. 2295 * 2296 * DTL_OUTAGE: txgs which cannot currently be read, whether due to 2297 * persistent errors or just some device being offline. 2298 * Unlike the other three, the DTL_OUTAGE map is not generally 2299 * maintained; it's only computed when needed, typically to 2300 * determine whether a device can be detached. 2301 * 2302 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 2303 * either has the data or it doesn't. 2304 * 2305 * For interior vdevs such as mirror and RAID-Z the picture is more complex. 2306 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 2307 * if any child is less than fully replicated, then so is its parent. 2308 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 2309 * comprising only those txgs which appear in 'maxfaults' or more children; 2310 * those are the txgs we don't have enough replication to read. For example, 2311 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 2312 * thus, its DTL_MISSING consists of the set of txgs that appear in more than 2313 * two child DTL_MISSING maps. 2314 * 2315 * It should be clear from the above that to compute the DTLs and outage maps 2316 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 2317 * Therefore, that is all we keep on disk. When loading the pool, or after 2318 * a configuration change, we generate all other DTLs from first principles. 2319 */ 2320 void 2321 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2322 { 2323 range_tree_t *rt = vd->vdev_dtl[t]; 2324 2325 ASSERT(t < DTL_TYPES); 2326 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2327 ASSERT(spa_writeable(vd->vdev_spa)); 2328 2329 mutex_enter(&vd->vdev_dtl_lock); 2330 if (!range_tree_contains(rt, txg, size)) 2331 range_tree_add(rt, txg, size); 2332 mutex_exit(&vd->vdev_dtl_lock); 2333 } 2334 2335 boolean_t 2336 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2337 { 2338 range_tree_t *rt = vd->vdev_dtl[t]; 2339 boolean_t dirty = B_FALSE; 2340 2341 ASSERT(t < DTL_TYPES); 2342 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2343 2344 /* 2345 * While we are loading the pool, the DTLs have not been loaded yet. 2346 * Ignore the DTLs and try all devices. This avoids a recursive 2347 * mutex enter on the vdev_dtl_lock, and also makes us try hard 2348 * when loading the pool (relying on the checksum to ensure that 2349 * we get the right data -- note that we while loading, we are 2350 * only reading the MOS, which is always checksummed). 2351 */ 2352 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 2353 return (B_FALSE); 2354 2355 mutex_enter(&vd->vdev_dtl_lock); 2356 if (!range_tree_is_empty(rt)) 2357 dirty = range_tree_contains(rt, txg, size); 2358 mutex_exit(&vd->vdev_dtl_lock); 2359 2360 return (dirty); 2361 } 2362 2363 boolean_t 2364 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 2365 { 2366 range_tree_t *rt = vd->vdev_dtl[t]; 2367 boolean_t empty; 2368 2369 mutex_enter(&vd->vdev_dtl_lock); 2370 empty = range_tree_is_empty(rt); 2371 mutex_exit(&vd->vdev_dtl_lock); 2372 2373 return (empty); 2374 } 2375 2376 /* 2377 * Returns B_TRUE if vdev determines offset needs to be resilvered. 2378 */ 2379 boolean_t 2380 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize) 2381 { 2382 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2383 2384 if (vd->vdev_ops->vdev_op_need_resilver == NULL || 2385 vd->vdev_ops->vdev_op_leaf) 2386 return (B_TRUE); 2387 2388 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize)); 2389 } 2390 2391 /* 2392 * Returns the lowest txg in the DTL range. 2393 */ 2394 static uint64_t 2395 vdev_dtl_min(vdev_t *vd) 2396 { 2397 range_seg_t *rs; 2398 2399 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2400 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2401 ASSERT0(vd->vdev_children); 2402 2403 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 2404 return (rs->rs_start - 1); 2405 } 2406 2407 /* 2408 * Returns the highest txg in the DTL. 2409 */ 2410 static uint64_t 2411 vdev_dtl_max(vdev_t *vd) 2412 { 2413 range_seg_t *rs; 2414 2415 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2416 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2417 ASSERT0(vd->vdev_children); 2418 2419 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 2420 return (rs->rs_end); 2421 } 2422 2423 /* 2424 * Determine if a resilvering vdev should remove any DTL entries from 2425 * its range. If the vdev was resilvering for the entire duration of the 2426 * scan then it should excise that range from its DTLs. Otherwise, this 2427 * vdev is considered partially resilvered and should leave its DTL 2428 * entries intact. The comment in vdev_dtl_reassess() describes how we 2429 * excise the DTLs. 2430 */ 2431 static boolean_t 2432 vdev_dtl_should_excise(vdev_t *vd) 2433 { 2434 spa_t *spa = vd->vdev_spa; 2435 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2436 2437 ASSERT0(scn->scn_phys.scn_errors); 2438 ASSERT0(vd->vdev_children); 2439 2440 if (vd->vdev_state < VDEV_STATE_DEGRADED) 2441 return (B_FALSE); 2442 2443 if (vd->vdev_resilver_txg == 0 || 2444 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2445 return (B_TRUE); 2446 2447 /* 2448 * When a resilver is initiated the scan will assign the scn_max_txg 2449 * value to the highest txg value that exists in all DTLs. If this 2450 * device's max DTL is not part of this scan (i.e. it is not in 2451 * the range (scn_min_txg, scn_max_txg] then it is not eligible 2452 * for excision. 2453 */ 2454 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2455 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2456 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2457 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2458 return (B_TRUE); 2459 } 2460 return (B_FALSE); 2461 } 2462 2463 /* 2464 * Reassess DTLs after a config change or scrub completion. 2465 */ 2466 void 2467 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2468 { 2469 spa_t *spa = vd->vdev_spa; 2470 avl_tree_t reftree; 2471 int minref; 2472 2473 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2474 2475 for (int c = 0; c < vd->vdev_children; c++) 2476 vdev_dtl_reassess(vd->vdev_child[c], txg, 2477 scrub_txg, scrub_done); 2478 2479 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 2480 return; 2481 2482 if (vd->vdev_ops->vdev_op_leaf) { 2483 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2484 2485 mutex_enter(&vd->vdev_dtl_lock); 2486 2487 /* 2488 * If we've completed a scan cleanly then determine 2489 * if this vdev should remove any DTLs. We only want to 2490 * excise regions on vdevs that were available during 2491 * the entire duration of this scan. 2492 */ 2493 if (scrub_txg != 0 && 2494 (spa->spa_scrub_started || 2495 (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2496 vdev_dtl_should_excise(vd)) { 2497 /* 2498 * We completed a scrub up to scrub_txg. If we 2499 * did it without rebooting, then the scrub dtl 2500 * will be valid, so excise the old region and 2501 * fold in the scrub dtl. Otherwise, leave the 2502 * dtl as-is if there was an error. 2503 * 2504 * There's little trick here: to excise the beginning 2505 * of the DTL_MISSING map, we put it into a reference 2506 * tree and then add a segment with refcnt -1 that 2507 * covers the range [0, scrub_txg). This means 2508 * that each txg in that range has refcnt -1 or 0. 2509 * We then add DTL_SCRUB with a refcnt of 2, so that 2510 * entries in the range [0, scrub_txg) will have a 2511 * positive refcnt -- either 1 or 2. We then convert 2512 * the reference tree into the new DTL_MISSING map. 2513 */ 2514 space_reftree_create(&reftree); 2515 space_reftree_add_map(&reftree, 2516 vd->vdev_dtl[DTL_MISSING], 1); 2517 space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 2518 space_reftree_add_map(&reftree, 2519 vd->vdev_dtl[DTL_SCRUB], 2); 2520 space_reftree_generate_map(&reftree, 2521 vd->vdev_dtl[DTL_MISSING], 1); 2522 space_reftree_destroy(&reftree); 2523 } 2524 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 2525 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 2526 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2527 if (scrub_done) 2528 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 2529 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 2530 if (!vdev_readable(vd)) 2531 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 2532 else 2533 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 2534 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2535 2536 /* 2537 * If the vdev was resilvering and no longer has any 2538 * DTLs then reset its resilvering flag. 2539 */ 2540 if (vd->vdev_resilver_txg != 0 && 2541 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 2542 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) 2543 vd->vdev_resilver_txg = 0; 2544 2545 mutex_exit(&vd->vdev_dtl_lock); 2546 2547 if (txg != 0) 2548 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2549 return; 2550 } 2551 2552 mutex_enter(&vd->vdev_dtl_lock); 2553 for (int t = 0; t < DTL_TYPES; t++) { 2554 /* account for child's outage in parent's missing map */ 2555 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 2556 if (t == DTL_SCRUB) 2557 continue; /* leaf vdevs only */ 2558 if (t == DTL_PARTIAL) 2559 minref = 1; /* i.e. non-zero */ 2560 else if (vd->vdev_nparity != 0) 2561 minref = vd->vdev_nparity + 1; /* RAID-Z */ 2562 else 2563 minref = vd->vdev_children; /* any kind of mirror */ 2564 space_reftree_create(&reftree); 2565 for (int c = 0; c < vd->vdev_children; c++) { 2566 vdev_t *cvd = vd->vdev_child[c]; 2567 mutex_enter(&cvd->vdev_dtl_lock); 2568 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 2569 mutex_exit(&cvd->vdev_dtl_lock); 2570 } 2571 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 2572 space_reftree_destroy(&reftree); 2573 } 2574 mutex_exit(&vd->vdev_dtl_lock); 2575 } 2576 2577 int 2578 vdev_dtl_load(vdev_t *vd) 2579 { 2580 spa_t *spa = vd->vdev_spa; 2581 objset_t *mos = spa->spa_meta_objset; 2582 int error = 0; 2583 2584 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 2585 ASSERT(vdev_is_concrete(vd)); 2586 2587 error = space_map_open(&vd->vdev_dtl_sm, mos, 2588 vd->vdev_dtl_object, 0, -1ULL, 0); 2589 if (error) 2590 return (error); 2591 ASSERT(vd->vdev_dtl_sm != NULL); 2592 2593 mutex_enter(&vd->vdev_dtl_lock); 2594 error = space_map_load(vd->vdev_dtl_sm, 2595 vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 2596 mutex_exit(&vd->vdev_dtl_lock); 2597 2598 return (error); 2599 } 2600 2601 for (int c = 0; c < vd->vdev_children; c++) { 2602 error = vdev_dtl_load(vd->vdev_child[c]); 2603 if (error != 0) 2604 break; 2605 } 2606 2607 return (error); 2608 } 2609 2610 static void 2611 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) 2612 { 2613 spa_t *spa = vd->vdev_spa; 2614 objset_t *mos = spa->spa_meta_objset; 2615 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 2616 const char *string; 2617 2618 ASSERT(alloc_bias != VDEV_BIAS_NONE); 2619 2620 string = 2621 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : 2622 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : 2623 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; 2624 2625 ASSERT(string != NULL); 2626 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 2627 1, strlen(string) + 1, string, tx)); 2628 2629 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { 2630 spa_activate_allocation_classes(spa, tx); 2631 } 2632 } 2633 2634 void 2635 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2636 { 2637 spa_t *spa = vd->vdev_spa; 2638 2639 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2640 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2641 zapobj, tx)); 2642 } 2643 2644 uint64_t 2645 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2646 { 2647 spa_t *spa = vd->vdev_spa; 2648 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2649 DMU_OT_NONE, 0, tx); 2650 2651 ASSERT(zap != 0); 2652 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2653 zap, tx)); 2654 2655 return (zap); 2656 } 2657 2658 void 2659 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2660 { 2661 if (vd->vdev_ops != &vdev_hole_ops && 2662 vd->vdev_ops != &vdev_missing_ops && 2663 vd->vdev_ops != &vdev_root_ops && 2664 !vd->vdev_top->vdev_removing) { 2665 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2666 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2667 } 2668 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2669 vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2670 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) 2671 vdev_zap_allocation_data(vd, tx); 2672 } 2673 } 2674 2675 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2676 vdev_construct_zaps(vd->vdev_child[i], tx); 2677 } 2678 } 2679 2680 void 2681 vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2682 { 2683 spa_t *spa = vd->vdev_spa; 2684 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2685 objset_t *mos = spa->spa_meta_objset; 2686 range_tree_t *rtsync; 2687 dmu_tx_t *tx; 2688 uint64_t object = space_map_object(vd->vdev_dtl_sm); 2689 2690 ASSERT(vdev_is_concrete(vd)); 2691 ASSERT(vd->vdev_ops->vdev_op_leaf); 2692 2693 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2694 2695 if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 2696 mutex_enter(&vd->vdev_dtl_lock); 2697 space_map_free(vd->vdev_dtl_sm, tx); 2698 space_map_close(vd->vdev_dtl_sm); 2699 vd->vdev_dtl_sm = NULL; 2700 mutex_exit(&vd->vdev_dtl_lock); 2701 2702 /* 2703 * We only destroy the leaf ZAP for detached leaves or for 2704 * removed log devices. Removed data devices handle leaf ZAP 2705 * cleanup later, once cancellation is no longer possible. 2706 */ 2707 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2708 vd->vdev_top->vdev_islog)) { 2709 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2710 vd->vdev_leaf_zap = 0; 2711 } 2712 2713 dmu_tx_commit(tx); 2714 return; 2715 } 2716 2717 if (vd->vdev_dtl_sm == NULL) { 2718 uint64_t new_object; 2719 2720 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx); 2721 VERIFY3U(new_object, !=, 0); 2722 2723 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 2724 0, -1ULL, 0)); 2725 ASSERT(vd->vdev_dtl_sm != NULL); 2726 } 2727 2728 rtsync = range_tree_create(NULL, NULL); 2729 2730 mutex_enter(&vd->vdev_dtl_lock); 2731 range_tree_walk(rt, range_tree_add, rtsync); 2732 mutex_exit(&vd->vdev_dtl_lock); 2733 2734 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx); 2735 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 2736 range_tree_vacate(rtsync, NULL, NULL); 2737 2738 range_tree_destroy(rtsync); 2739 2740 /* 2741 * If the object for the space map has changed then dirty 2742 * the top level so that we update the config. 2743 */ 2744 if (object != space_map_object(vd->vdev_dtl_sm)) { 2745 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 2746 "new object %llu", (u_longlong_t)txg, spa_name(spa), 2747 (u_longlong_t)object, 2748 (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 2749 vdev_config_dirty(vd->vdev_top); 2750 } 2751 2752 dmu_tx_commit(tx); 2753 } 2754 2755 /* 2756 * Determine whether the specified vdev can be offlined/detached/removed 2757 * without losing data. 2758 */ 2759 boolean_t 2760 vdev_dtl_required(vdev_t *vd) 2761 { 2762 spa_t *spa = vd->vdev_spa; 2763 vdev_t *tvd = vd->vdev_top; 2764 uint8_t cant_read = vd->vdev_cant_read; 2765 boolean_t required; 2766 2767 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2768 2769 if (vd == spa->spa_root_vdev || vd == tvd) 2770 return (B_TRUE); 2771 2772 /* 2773 * Temporarily mark the device as unreadable, and then determine 2774 * whether this results in any DTL outages in the top-level vdev. 2775 * If not, we can safely offline/detach/remove the device. 2776 */ 2777 vd->vdev_cant_read = B_TRUE; 2778 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 2779 required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 2780 vd->vdev_cant_read = cant_read; 2781 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 2782 2783 if (!required && zio_injection_enabled) 2784 required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2785 2786 return (required); 2787 } 2788 2789 /* 2790 * Determine if resilver is needed, and if so the txg range. 2791 */ 2792 boolean_t 2793 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2794 { 2795 boolean_t needed = B_FALSE; 2796 uint64_t thismin = UINT64_MAX; 2797 uint64_t thismax = 0; 2798 2799 if (vd->vdev_children == 0) { 2800 mutex_enter(&vd->vdev_dtl_lock); 2801 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 2802 vdev_writeable(vd)) { 2803 2804 thismin = vdev_dtl_min(vd); 2805 thismax = vdev_dtl_max(vd); 2806 needed = B_TRUE; 2807 } 2808 mutex_exit(&vd->vdev_dtl_lock); 2809 } else { 2810 for (int c = 0; c < vd->vdev_children; c++) { 2811 vdev_t *cvd = vd->vdev_child[c]; 2812 uint64_t cmin, cmax; 2813 2814 if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2815 thismin = MIN(thismin, cmin); 2816 thismax = MAX(thismax, cmax); 2817 needed = B_TRUE; 2818 } 2819 } 2820 } 2821 2822 if (needed && minp) { 2823 *minp = thismin; 2824 *maxp = thismax; 2825 } 2826 return (needed); 2827 } 2828 2829 /* 2830 * Gets the checkpoint space map object from the vdev's ZAP. 2831 * Returns the spacemap object, or 0 if it wasn't in the ZAP 2832 * or the ZAP doesn't exist yet. 2833 */ 2834 int 2835 vdev_checkpoint_sm_object(vdev_t *vd) 2836 { 2837 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 2838 if (vd->vdev_top_zap == 0) { 2839 return (0); 2840 } 2841 2842 uint64_t sm_obj = 0; 2843 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 2844 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); 2845 2846 ASSERT(err == 0 || err == ENOENT); 2847 2848 return (sm_obj); 2849 } 2850 2851 int 2852 vdev_load(vdev_t *vd) 2853 { 2854 int error = 0; 2855 /* 2856 * Recursively load all children. 2857 */ 2858 for (int c = 0; c < vd->vdev_children; c++) { 2859 error = vdev_load(vd->vdev_child[c]); 2860 if (error != 0) { 2861 return (error); 2862 } 2863 } 2864 2865 vdev_set_deflate_ratio(vd); 2866 2867 /* 2868 * On spa_load path, grab the allocation bias from our zap 2869 */ 2870 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 2871 spa_t *spa = vd->vdev_spa; 2872 char bias_str[64]; 2873 2874 if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, 2875 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), 2876 bias_str) == 0) { 2877 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); 2878 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); 2879 } 2880 } 2881 2882 /* 2883 * If this is a top-level vdev, initialize its metaslabs. 2884 */ 2885 if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 2886 vdev_metaslab_group_create(vd); 2887 2888 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 2889 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2890 VDEV_AUX_CORRUPT_DATA); 2891 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 2892 "asize=%llu", (u_longlong_t)vd->vdev_ashift, 2893 (u_longlong_t)vd->vdev_asize); 2894 return (SET_ERROR(ENXIO)); 2895 } 2896 2897 error = vdev_metaslab_init(vd, 0); 2898 if (error != 0) { 2899 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 2900 "[error=%d]", error); 2901 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2902 VDEV_AUX_CORRUPT_DATA); 2903 return (error); 2904 } 2905 2906 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); 2907 if (checkpoint_sm_obj != 0) { 2908 objset_t *mos = spa_meta_objset(vd->vdev_spa); 2909 ASSERT(vd->vdev_asize != 0); 2910 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 2911 2912 error = space_map_open(&vd->vdev_checkpoint_sm, 2913 mos, checkpoint_sm_obj, 0, vd->vdev_asize, 2914 vd->vdev_ashift); 2915 if (error != 0) { 2916 vdev_dbgmsg(vd, "vdev_load: space_map_open " 2917 "failed for checkpoint spacemap (obj %llu) " 2918 "[error=%d]", 2919 (u_longlong_t)checkpoint_sm_obj, error); 2920 return (error); 2921 } 2922 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 2923 2924 /* 2925 * Since the checkpoint_sm contains free entries 2926 * exclusively we can use space_map_allocated() to 2927 * indicate the cumulative checkpointed space that 2928 * has been freed. 2929 */ 2930 vd->vdev_stat.vs_checkpoint_space = 2931 -space_map_allocated(vd->vdev_checkpoint_sm); 2932 vd->vdev_spa->spa_checkpoint_info.sci_dspace += 2933 vd->vdev_stat.vs_checkpoint_space; 2934 } 2935 } 2936 2937 /* 2938 * If this is a leaf vdev, load its DTL. 2939 */ 2940 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 2941 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2942 VDEV_AUX_CORRUPT_DATA); 2943 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 2944 "[error=%d]", error); 2945 return (error); 2946 } 2947 2948 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 2949 if (obsolete_sm_object != 0) { 2950 objset_t *mos = vd->vdev_spa->spa_meta_objset; 2951 ASSERT(vd->vdev_asize != 0); 2952 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 2953 2954 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 2955 obsolete_sm_object, 0, vd->vdev_asize, 0))) { 2956 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2957 VDEV_AUX_CORRUPT_DATA); 2958 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 2959 "obsolete spacemap (obj %llu) [error=%d]", 2960 (u_longlong_t)obsolete_sm_object, error); 2961 return (error); 2962 } 2963 } 2964 2965 return (0); 2966 } 2967 2968 /* 2969 * The special vdev case is used for hot spares and l2cache devices. Its 2970 * sole purpose it to set the vdev state for the associated vdev. To do this, 2971 * we make sure that we can open the underlying device, then try to read the 2972 * label, and make sure that the label is sane and that it hasn't been 2973 * repurposed to another pool. 2974 */ 2975 int 2976 vdev_validate_aux(vdev_t *vd) 2977 { 2978 nvlist_t *label; 2979 uint64_t guid, version; 2980 uint64_t state; 2981 2982 if (!vdev_readable(vd)) 2983 return (0); 2984 2985 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 2986 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2987 VDEV_AUX_CORRUPT_DATA); 2988 return (-1); 2989 } 2990 2991 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2992 !SPA_VERSION_IS_SUPPORTED(version) || 2993 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 2994 guid != vd->vdev_guid || 2995 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 2996 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2997 VDEV_AUX_CORRUPT_DATA); 2998 nvlist_free(label); 2999 return (-1); 3000 } 3001 3002 /* 3003 * We don't actually check the pool state here. If it's in fact in 3004 * use by another pool, we update this fact on the fly when requested. 3005 */ 3006 nvlist_free(label); 3007 return (0); 3008 } 3009 3010 /* 3011 * Free the objects used to store this vdev's spacemaps, and the array 3012 * that points to them. 3013 */ 3014 void 3015 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 3016 { 3017 if (vd->vdev_ms_array == 0) 3018 return; 3019 3020 objset_t *mos = vd->vdev_spa->spa_meta_objset; 3021 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 3022 size_t array_bytes = array_count * sizeof (uint64_t); 3023 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 3024 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 3025 array_bytes, smobj_array, 0)); 3026 3027 for (uint64_t i = 0; i < array_count; i++) { 3028 uint64_t smobj = smobj_array[i]; 3029 if (smobj == 0) 3030 continue; 3031 3032 space_map_free_obj(mos, smobj, tx); 3033 } 3034 3035 kmem_free(smobj_array, array_bytes); 3036 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 3037 vd->vdev_ms_array = 0; 3038 } 3039 3040 static void 3041 vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 3042 { 3043 spa_t *spa = vd->vdev_spa; 3044 3045 ASSERT(vd->vdev_islog); 3046 ASSERT(vd == vd->vdev_top); 3047 ASSERT3U(txg, ==, spa_syncing_txg(spa)); 3048 3049 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 3050 3051 vdev_destroy_spacemaps(vd, tx); 3052 if (vd->vdev_top_zap != 0) { 3053 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 3054 vd->vdev_top_zap = 0; 3055 } 3056 3057 dmu_tx_commit(tx); 3058 } 3059 3060 void 3061 vdev_sync_done(vdev_t *vd, uint64_t txg) 3062 { 3063 metaslab_t *msp; 3064 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 3065 3066 ASSERT(vdev_is_concrete(vd)); 3067 3068 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 3069 != NULL) 3070 metaslab_sync_done(msp, txg); 3071 3072 if (reassess) 3073 metaslab_sync_reassess(vd->vdev_mg); 3074 } 3075 3076 void 3077 vdev_sync(vdev_t *vd, uint64_t txg) 3078 { 3079 spa_t *spa = vd->vdev_spa; 3080 vdev_t *lvd; 3081 metaslab_t *msp; 3082 3083 ASSERT3U(txg, ==, spa->spa_syncing_txg); 3084 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3085 if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 3086 ASSERT(vd->vdev_removing || 3087 vd->vdev_ops == &vdev_indirect_ops); 3088 3089 vdev_indirect_sync_obsolete(vd, tx); 3090 3091 /* 3092 * If the vdev is indirect, it can't have dirty 3093 * metaslabs or DTLs. 3094 */ 3095 if (vd->vdev_ops == &vdev_indirect_ops) { 3096 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 3097 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 3098 dmu_tx_commit(tx); 3099 return; 3100 } 3101 } 3102 3103 ASSERT(vdev_is_concrete(vd)); 3104 3105 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 3106 !vd->vdev_removing) { 3107 ASSERT(vd == vd->vdev_top); 3108 ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 3109 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 3110 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 3111 ASSERT(vd->vdev_ms_array != 0); 3112 vdev_config_dirty(vd); 3113 } 3114 3115 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 3116 metaslab_sync(msp, txg); 3117 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 3118 } 3119 3120 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 3121 vdev_dtl_sync(lvd, txg); 3122 3123 /* 3124 * If this is an empty log device being removed, destroy the 3125 * metadata associated with it. 3126 */ 3127 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 3128 vdev_remove_empty_log(vd, txg); 3129 3130 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 3131 dmu_tx_commit(tx); 3132 } 3133 3134 uint64_t 3135 vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 3136 { 3137 return (vd->vdev_ops->vdev_op_asize(vd, psize)); 3138 } 3139 3140 /* 3141 * Mark the given vdev faulted. A faulted vdev behaves as if the device could 3142 * not be opened, and no I/O is attempted. 3143 */ 3144 int 3145 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3146 { 3147 vdev_t *vd, *tvd; 3148 3149 spa_vdev_state_enter(spa, SCL_NONE); 3150 3151 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3152 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3153 3154 if (!vd->vdev_ops->vdev_op_leaf) 3155 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3156 3157 tvd = vd->vdev_top; 3158 3159 /* 3160 * We don't directly use the aux state here, but if we do a 3161 * vdev_reopen(), we need this value to be present to remember why we 3162 * were faulted. 3163 */ 3164 vd->vdev_label_aux = aux; 3165 3166 /* 3167 * Faulted state takes precedence over degraded. 3168 */ 3169 vd->vdev_delayed_close = B_FALSE; 3170 vd->vdev_faulted = 1ULL; 3171 vd->vdev_degraded = 0ULL; 3172 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 3173 3174 /* 3175 * If this device has the only valid copy of the data, then 3176 * back off and simply mark the vdev as degraded instead. 3177 */ 3178 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 3179 vd->vdev_degraded = 1ULL; 3180 vd->vdev_faulted = 0ULL; 3181 3182 /* 3183 * If we reopen the device and it's not dead, only then do we 3184 * mark it degraded. 3185 */ 3186 vdev_reopen(tvd); 3187 3188 if (vdev_readable(vd)) 3189 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 3190 } 3191 3192 return (spa_vdev_state_exit(spa, vd, 0)); 3193 } 3194 3195 /* 3196 * Mark the given vdev degraded. A degraded vdev is purely an indication to the 3197 * user that something is wrong. The vdev continues to operate as normal as far 3198 * as I/O is concerned. 3199 */ 3200 int 3201 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3202 { 3203 vdev_t *vd; 3204 3205 spa_vdev_state_enter(spa, SCL_NONE); 3206 3207 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3208 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3209 3210 if (!vd->vdev_ops->vdev_op_leaf) 3211 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3212 3213 /* 3214 * If the vdev is already faulted, then don't do anything. 3215 */ 3216 if (vd->vdev_faulted || vd->vdev_degraded) 3217 return (spa_vdev_state_exit(spa, NULL, 0)); 3218 3219 vd->vdev_degraded = 1ULL; 3220 if (!vdev_is_dead(vd)) 3221 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3222 aux); 3223 3224 return (spa_vdev_state_exit(spa, vd, 0)); 3225 } 3226 3227 /* 3228 * Online the given vdev. 3229 * 3230 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 3231 * spare device should be detached when the device finishes resilvering. 3232 * Second, the online should be treated like a 'test' online case, so no FMA 3233 * events are generated if the device fails to open. 3234 */ 3235 int 3236 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 3237 { 3238 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 3239 boolean_t wasoffline; 3240 vdev_state_t oldstate; 3241 3242 spa_vdev_state_enter(spa, SCL_NONE); 3243 3244 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3245 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3246 3247 if (!vd->vdev_ops->vdev_op_leaf) 3248 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3249 3250 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 3251 oldstate = vd->vdev_state; 3252 3253 tvd = vd->vdev_top; 3254 vd->vdev_offline = B_FALSE; 3255 vd->vdev_tmpoffline = B_FALSE; 3256 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3257 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3258 3259 /* XXX - L2ARC 1.0 does not support expansion */ 3260 if (!vd->vdev_aux) { 3261 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3262 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 3263 } 3264 3265 vdev_reopen(tvd); 3266 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 3267 3268 if (!vd->vdev_aux) { 3269 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3270 pvd->vdev_expanding = B_FALSE; 3271 } 3272 3273 if (newstate) 3274 *newstate = vd->vdev_state; 3275 if ((flags & ZFS_ONLINE_UNSPARE) && 3276 !vdev_is_dead(vd) && vd->vdev_parent && 3277 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3278 vd->vdev_parent->vdev_child[0] == vd) 3279 vd->vdev_unspare = B_TRUE; 3280 3281 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3282 3283 /* XXX - L2ARC 1.0 does not support expansion */ 3284 if (vd->vdev_aux) 3285 return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3286 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3287 } 3288 3289 /* Restart initializing if necessary */ 3290 mutex_enter(&vd->vdev_initialize_lock); 3291 if (vdev_writeable(vd) && 3292 vd->vdev_initialize_thread == NULL && 3293 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 3294 (void) vdev_initialize(vd); 3295 } 3296 mutex_exit(&vd->vdev_initialize_lock); 3297 3298 if (wasoffline || 3299 (oldstate < VDEV_STATE_DEGRADED && 3300 vd->vdev_state >= VDEV_STATE_DEGRADED)) 3301 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 3302 3303 return (spa_vdev_state_exit(spa, vd, 0)); 3304 } 3305 3306 static int 3307 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3308 { 3309 vdev_t *vd, *tvd; 3310 int error = 0; 3311 uint64_t generation; 3312 metaslab_group_t *mg; 3313 3314 top: 3315 spa_vdev_state_enter(spa, SCL_ALLOC); 3316 3317 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3318 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3319 3320 if (!vd->vdev_ops->vdev_op_leaf) 3321 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3322 3323 tvd = vd->vdev_top; 3324 mg = tvd->vdev_mg; 3325 generation = spa->spa_config_generation + 1; 3326 3327 /* 3328 * If the device isn't already offline, try to offline it. 3329 */ 3330 if (!vd->vdev_offline) { 3331 /* 3332 * If this device has the only valid copy of some data, 3333 * don't allow it to be offlined. Log devices are always 3334 * expendable. 3335 */ 3336 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3337 vdev_dtl_required(vd)) 3338 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3339 3340 /* 3341 * If the top-level is a slog and it has had allocations 3342 * then proceed. We check that the vdev's metaslab group 3343 * is not NULL since it's possible that we may have just 3344 * added this vdev but not yet initialized its metaslabs. 3345 */ 3346 if (tvd->vdev_islog && mg != NULL) { 3347 /* 3348 * Prevent any future allocations. 3349 */ 3350 metaslab_group_passivate(mg); 3351 (void) spa_vdev_state_exit(spa, vd, 0); 3352 3353 error = spa_reset_logs(spa); 3354 3355 /* 3356 * If the log device was successfully reset but has 3357 * checkpointed data, do not offline it. 3358 */ 3359 if (error == 0 && 3360 tvd->vdev_checkpoint_sm != NULL) { 3361 error = ZFS_ERR_CHECKPOINT_EXISTS; 3362 } 3363 3364 spa_vdev_state_enter(spa, SCL_ALLOC); 3365 3366 /* 3367 * Check to see if the config has changed. 3368 */ 3369 if (error || generation != spa->spa_config_generation) { 3370 metaslab_group_activate(mg); 3371 if (error) 3372 return (spa_vdev_state_exit(spa, 3373 vd, error)); 3374 (void) spa_vdev_state_exit(spa, vd, 0); 3375 goto top; 3376 } 3377 ASSERT0(tvd->vdev_stat.vs_alloc); 3378 } 3379 3380 /* 3381 * Offline this device and reopen its top-level vdev. 3382 * If the top-level vdev is a log device then just offline 3383 * it. Otherwise, if this action results in the top-level 3384 * vdev becoming unusable, undo it and fail the request. 3385 */ 3386 vd->vdev_offline = B_TRUE; 3387 vdev_reopen(tvd); 3388 3389 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3390 vdev_is_dead(tvd)) { 3391 vd->vdev_offline = B_FALSE; 3392 vdev_reopen(tvd); 3393 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3394 } 3395 3396 /* 3397 * Add the device back into the metaslab rotor so that 3398 * once we online the device it's open for business. 3399 */ 3400 if (tvd->vdev_islog && mg != NULL) 3401 metaslab_group_activate(mg); 3402 } 3403 3404 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3405 3406 return (spa_vdev_state_exit(spa, vd, 0)); 3407 } 3408 3409 int 3410 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3411 { 3412 int error; 3413 3414 mutex_enter(&spa->spa_vdev_top_lock); 3415 error = vdev_offline_locked(spa, guid, flags); 3416 mutex_exit(&spa->spa_vdev_top_lock); 3417 3418 return (error); 3419 } 3420 3421 /* 3422 * Clear the error counts associated with this vdev. Unlike vdev_online() and 3423 * vdev_offline(), we assume the spa config is locked. We also clear all 3424 * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3425 */ 3426 void 3427 vdev_clear(spa_t *spa, vdev_t *vd) 3428 { 3429 vdev_t *rvd = spa->spa_root_vdev; 3430 3431 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3432 3433 if (vd == NULL) 3434 vd = rvd; 3435 3436 vd->vdev_stat.vs_read_errors = 0; 3437 vd->vdev_stat.vs_write_errors = 0; 3438 vd->vdev_stat.vs_checksum_errors = 0; 3439 3440 for (int c = 0; c < vd->vdev_children; c++) 3441 vdev_clear(spa, vd->vdev_child[c]); 3442 3443 /* 3444 * It makes no sense to "clear" an indirect vdev. 3445 */ 3446 if (!vdev_is_concrete(vd)) 3447 return; 3448 3449 /* 3450 * If we're in the FAULTED state or have experienced failed I/O, then 3451 * clear the persistent state and attempt to reopen the device. We 3452 * also mark the vdev config dirty, so that the new faulted state is 3453 * written out to disk. 3454 */ 3455 if (vd->vdev_faulted || vd->vdev_degraded || 3456 !vdev_readable(vd) || !vdev_writeable(vd)) { 3457 3458 /* 3459 * When reopening in reponse to a clear event, it may be due to 3460 * a fmadm repair request. In this case, if the device is 3461 * still broken, we want to still post the ereport again. 3462 */ 3463 vd->vdev_forcefault = B_TRUE; 3464 3465 vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3466 vd->vdev_cant_read = B_FALSE; 3467 vd->vdev_cant_write = B_FALSE; 3468 3469 vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 3470 3471 vd->vdev_forcefault = B_FALSE; 3472 3473 if (vd != rvd && vdev_writeable(vd->vdev_top)) 3474 vdev_state_dirty(vd->vdev_top); 3475 3476 if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 3477 spa_async_request(spa, SPA_ASYNC_RESILVER); 3478 3479 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 3480 } 3481 3482 /* 3483 * When clearing a FMA-diagnosed fault, we always want to 3484 * unspare the device, as we assume that the original spare was 3485 * done in response to the FMA fault. 3486 */ 3487 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3488 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3489 vd->vdev_parent->vdev_child[0] == vd) 3490 vd->vdev_unspare = B_TRUE; 3491 } 3492 3493 boolean_t 3494 vdev_is_dead(vdev_t *vd) 3495 { 3496 /* 3497 * Holes and missing devices are always considered "dead". 3498 * This simplifies the code since we don't have to check for 3499 * these types of devices in the various code paths. 3500 * Instead we rely on the fact that we skip over dead devices 3501 * before issuing I/O to them. 3502 */ 3503 return (vd->vdev_state < VDEV_STATE_DEGRADED || 3504 vd->vdev_ops == &vdev_hole_ops || 3505 vd->vdev_ops == &vdev_missing_ops); 3506 } 3507 3508 boolean_t 3509 vdev_readable(vdev_t *vd) 3510 { 3511 return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 3512 } 3513 3514 boolean_t 3515 vdev_writeable(vdev_t *vd) 3516 { 3517 return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 3518 vdev_is_concrete(vd)); 3519 } 3520 3521 boolean_t 3522 vdev_allocatable(vdev_t *vd) 3523 { 3524 uint64_t state = vd->vdev_state; 3525 3526 /* 3527 * We currently allow allocations from vdevs which may be in the 3528 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3529 * fails to reopen then we'll catch it later when we're holding 3530 * the proper locks. Note that we have to get the vdev state 3531 * in a local variable because although it changes atomically, 3532 * we're asking two separate questions about it. 3533 */ 3534 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 3535 !vd->vdev_cant_write && vdev_is_concrete(vd) && 3536 vd->vdev_mg->mg_initialized); 3537 } 3538 3539 boolean_t 3540 vdev_accessible(vdev_t *vd, zio_t *zio) 3541 { 3542 ASSERT(zio->io_vd == vd); 3543 3544 if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3545 return (B_FALSE); 3546 3547 if (zio->io_type == ZIO_TYPE_READ) 3548 return (!vd->vdev_cant_read); 3549 3550 if (zio->io_type == ZIO_TYPE_WRITE) 3551 return (!vd->vdev_cant_write); 3552 3553 return (B_TRUE); 3554 } 3555 3556 boolean_t 3557 vdev_is_spacemap_addressable(vdev_t *vd) 3558 { 3559 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 3560 return (B_TRUE); 3561 3562 /* 3563 * If double-word space map entries are not enabled we assume 3564 * 47 bits of the space map entry are dedicated to the entry's 3565 * offset (see SM_OFFSET_BITS in space_map.h). We then use that 3566 * to calculate the maximum address that can be described by a 3567 * space map entry for the given device. 3568 */ 3569 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 3570 3571 if (shift >= 63) /* detect potential overflow */ 3572 return (B_TRUE); 3573 3574 return (vd->vdev_asize < (1ULL << shift)); 3575 } 3576 3577 /* 3578 * Get statistics for the given vdev. 3579 */ 3580 void 3581 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3582 { 3583 spa_t *spa = vd->vdev_spa; 3584 vdev_t *rvd = spa->spa_root_vdev; 3585 vdev_t *tvd = vd->vdev_top; 3586 3587 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 3588 3589 mutex_enter(&vd->vdev_stat_lock); 3590 bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3591 vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3592 vs->vs_state = vd->vdev_state; 3593 vs->vs_rsize = vdev_get_min_asize(vd); 3594 if (vd->vdev_ops->vdev_op_leaf) { 3595 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 3596 /* 3597 * Report intializing progress. Since we don't have the 3598 * initializing locks held, this is only an estimate (although a 3599 * fairly accurate one). 3600 */ 3601 vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done; 3602 vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est; 3603 vs->vs_initialize_state = vd->vdev_initialize_state; 3604 vs->vs_initialize_action_time = vd->vdev_initialize_action_time; 3605 } 3606 /* 3607 * Report expandable space on top-level, non-auxillary devices only. 3608 * The expandable space is reported in terms of metaslab sized units 3609 * since that determines how much space the pool can expand. 3610 */ 3611 if (vd->vdev_aux == NULL && tvd != NULL) { 3612 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize - 3613 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift); 3614 } 3615 if (vd->vdev_aux == NULL && vd == vd->vdev_top && 3616 vdev_is_concrete(vd)) { 3617 vs->vs_fragmentation = (vd->vdev_mg != NULL) ? 3618 vd->vdev_mg->mg_fragmentation : 0; 3619 } 3620 3621 /* 3622 * If we're getting stats on the root vdev, aggregate the I/O counts 3623 * over all top-level vdevs (i.e. the direct children of the root). 3624 */ 3625 if (vd == rvd) { 3626 for (int c = 0; c < rvd->vdev_children; c++) { 3627 vdev_t *cvd = rvd->vdev_child[c]; 3628 vdev_stat_t *cvs = &cvd->vdev_stat; 3629 3630 for (int t = 0; t < ZIO_TYPES; t++) { 3631 vs->vs_ops[t] += cvs->vs_ops[t]; 3632 vs->vs_bytes[t] += cvs->vs_bytes[t]; 3633 } 3634 cvs->vs_scan_removing = cvd->vdev_removing; 3635 } 3636 } 3637 mutex_exit(&vd->vdev_stat_lock); 3638 } 3639 3640 void 3641 vdev_clear_stats(vdev_t *vd) 3642 { 3643 mutex_enter(&vd->vdev_stat_lock); 3644 vd->vdev_stat.vs_space = 0; 3645 vd->vdev_stat.vs_dspace = 0; 3646 vd->vdev_stat.vs_alloc = 0; 3647 mutex_exit(&vd->vdev_stat_lock); 3648 } 3649 3650 void 3651 vdev_scan_stat_init(vdev_t *vd) 3652 { 3653 vdev_stat_t *vs = &vd->vdev_stat; 3654 3655 for (int c = 0; c < vd->vdev_children; c++) 3656 vdev_scan_stat_init(vd->vdev_child[c]); 3657 3658 mutex_enter(&vd->vdev_stat_lock); 3659 vs->vs_scan_processed = 0; 3660 mutex_exit(&vd->vdev_stat_lock); 3661 } 3662 3663 void 3664 vdev_stat_update(zio_t *zio, uint64_t psize) 3665 { 3666 spa_t *spa = zio->io_spa; 3667 vdev_t *rvd = spa->spa_root_vdev; 3668 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3669 vdev_t *pvd; 3670 uint64_t txg = zio->io_txg; 3671 vdev_stat_t *vs = &vd->vdev_stat; 3672 zio_type_t type = zio->io_type; 3673 int flags = zio->io_flags; 3674 3675 /* 3676 * If this i/o is a gang leader, it didn't do any actual work. 3677 */ 3678 if (zio->io_gang_tree) 3679 return; 3680 3681 if (zio->io_error == 0) { 3682 /* 3683 * If this is a root i/o, don't count it -- we've already 3684 * counted the top-level vdevs, and vdev_get_stats() will 3685 * aggregate them when asked. This reduces contention on 3686 * the root vdev_stat_lock and implicitly handles blocks 3687 * that compress away to holes, for which there is no i/o. 3688 * (Holes never create vdev children, so all the counters 3689 * remain zero, which is what we want.) 3690 * 3691 * Note: this only applies to successful i/o (io_error == 0) 3692 * because unlike i/o counts, errors are not additive. 3693 * When reading a ditto block, for example, failure of 3694 * one top-level vdev does not imply a root-level error. 3695 */ 3696 if (vd == rvd) 3697 return; 3698 3699 ASSERT(vd == zio->io_vd); 3700 3701 if (flags & ZIO_FLAG_IO_BYPASS) 3702 return; 3703 3704 mutex_enter(&vd->vdev_stat_lock); 3705 3706 if (flags & ZIO_FLAG_IO_REPAIR) { 3707 if (flags & ZIO_FLAG_SCAN_THREAD) { 3708 dsl_scan_phys_t *scn_phys = 3709 &spa->spa_dsl_pool->dp_scan->scn_phys; 3710 uint64_t *processed = &scn_phys->scn_processed; 3711 3712 /* XXX cleanup? */ 3713 if (vd->vdev_ops->vdev_op_leaf) 3714 atomic_add_64(processed, psize); 3715 vs->vs_scan_processed += psize; 3716 } 3717 3718 if (flags & ZIO_FLAG_SELF_HEAL) 3719 vs->vs_self_healed += psize; 3720 } 3721 3722 vs->vs_ops[type]++; 3723 vs->vs_bytes[type] += psize; 3724 3725 mutex_exit(&vd->vdev_stat_lock); 3726 return; 3727 } 3728 3729 if (flags & ZIO_FLAG_SPECULATIVE) 3730 return; 3731 3732 /* 3733 * If this is an I/O error that is going to be retried, then ignore the 3734 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 3735 * hard errors, when in reality they can happen for any number of 3736 * innocuous reasons (bus resets, MPxIO link failure, etc). 3737 */ 3738 if (zio->io_error == EIO && 3739 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 3740 return; 3741 3742 /* 3743 * Intent logs writes won't propagate their error to the root 3744 * I/O so don't mark these types of failures as pool-level 3745 * errors. 3746 */ 3747 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 3748 return; 3749 3750 mutex_enter(&vd->vdev_stat_lock); 3751 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3752 if (zio->io_error == ECKSUM) 3753 vs->vs_checksum_errors++; 3754 else 3755 vs->vs_read_errors++; 3756 } 3757 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3758 vs->vs_write_errors++; 3759 mutex_exit(&vd->vdev_stat_lock); 3760 3761 if (spa->spa_load_state == SPA_LOAD_NONE && 3762 type == ZIO_TYPE_WRITE && txg != 0 && 3763 (!(flags & ZIO_FLAG_IO_REPAIR) || 3764 (flags & ZIO_FLAG_SCAN_THREAD) || 3765 spa->spa_claiming)) { 3766 /* 3767 * This is either a normal write (not a repair), or it's 3768 * a repair induced by the scrub thread, or it's a repair 3769 * made by zil_claim() during spa_load() in the first txg. 3770 * In the normal case, we commit the DTL change in the same 3771 * txg as the block was born. In the scrub-induced repair 3772 * case, we know that scrubs run in first-pass syncing context, 3773 * so we commit the DTL change in spa_syncing_txg(spa). 3774 * In the zil_claim() case, we commit in spa_first_txg(spa). 3775 * 3776 * We currently do not make DTL entries for failed spontaneous 3777 * self-healing writes triggered by normal (non-scrubbing) 3778 * reads, because we have no transactional context in which to 3779 * do so -- and it's not clear that it'd be desirable anyway. 3780 */ 3781 if (vd->vdev_ops->vdev_op_leaf) { 3782 uint64_t commit_txg = txg; 3783 if (flags & ZIO_FLAG_SCAN_THREAD) { 3784 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3785 ASSERT(spa_sync_pass(spa) == 1); 3786 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 3787 commit_txg = spa_syncing_txg(spa); 3788 } else if (spa->spa_claiming) { 3789 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3790 commit_txg = spa_first_txg(spa); 3791 } 3792 ASSERT(commit_txg >= spa_syncing_txg(spa)); 3793 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 3794 return; 3795 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3796 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 3797 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 3798 } 3799 if (vd != rvd) 3800 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 3801 } 3802 } 3803 3804 int64_t 3805 vdev_deflated_space(vdev_t *vd, int64_t space) 3806 { 3807 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); 3808 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 3809 3810 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); 3811 } 3812 3813 /* 3814 * Update the in-core space usage stats for this vdev and the root vdev. 3815 */ 3816 void 3817 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 3818 int64_t space_delta) 3819 { 3820 int64_t dspace_delta; 3821 spa_t *spa = vd->vdev_spa; 3822 vdev_t *rvd = spa->spa_root_vdev; 3823 3824 ASSERT(vd == vd->vdev_top); 3825 3826 /* 3827 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 3828 * factor. We must calculate this here and not at the root vdev 3829 * because the root vdev's psize-to-asize is simply the max of its 3830 * childrens', thus not accurate enough for us. 3831 */ 3832 dspace_delta = vdev_deflated_space(vd, space_delta); 3833 3834 mutex_enter(&vd->vdev_stat_lock); 3835 vd->vdev_stat.vs_alloc += alloc_delta; 3836 vd->vdev_stat.vs_space += space_delta; 3837 vd->vdev_stat.vs_dspace += dspace_delta; 3838 mutex_exit(&vd->vdev_stat_lock); 3839 3840 /* every class but log contributes to root space stats */ 3841 if (vd->vdev_mg != NULL && !vd->vdev_islog) { 3842 mutex_enter(&rvd->vdev_stat_lock); 3843 rvd->vdev_stat.vs_alloc += alloc_delta; 3844 rvd->vdev_stat.vs_space += space_delta; 3845 rvd->vdev_stat.vs_dspace += dspace_delta; 3846 mutex_exit(&rvd->vdev_stat_lock); 3847 } 3848 /* Note: metaslab_class_space_update moved to metaslab_space_update */ 3849 } 3850 3851 /* 3852 * Mark a top-level vdev's config as dirty, placing it on the dirty list 3853 * so that it will be written out next time the vdev configuration is synced. 3854 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 3855 */ 3856 void 3857 vdev_config_dirty(vdev_t *vd) 3858 { 3859 spa_t *spa = vd->vdev_spa; 3860 vdev_t *rvd = spa->spa_root_vdev; 3861 int c; 3862 3863 ASSERT(spa_writeable(spa)); 3864 3865 /* 3866 * If this is an aux vdev (as with l2cache and spare devices), then we 3867 * update the vdev config manually and set the sync flag. 3868 */ 3869 if (vd->vdev_aux != NULL) { 3870 spa_aux_vdev_t *sav = vd->vdev_aux; 3871 nvlist_t **aux; 3872 uint_t naux; 3873 3874 for (c = 0; c < sav->sav_count; c++) { 3875 if (sav->sav_vdevs[c] == vd) 3876 break; 3877 } 3878 3879 if (c == sav->sav_count) { 3880 /* 3881 * We're being removed. There's nothing more to do. 3882 */ 3883 ASSERT(sav->sav_sync == B_TRUE); 3884 return; 3885 } 3886 3887 sav->sav_sync = B_TRUE; 3888 3889 if (nvlist_lookup_nvlist_array(sav->sav_config, 3890 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 3891 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 3892 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 3893 } 3894 3895 ASSERT(c < naux); 3896 3897 /* 3898 * Setting the nvlist in the middle if the array is a little 3899 * sketchy, but it will work. 3900 */ 3901 nvlist_free(aux[c]); 3902 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3903 3904 return; 3905 } 3906 3907 /* 3908 * The dirty list is protected by the SCL_CONFIG lock. The caller 3909 * must either hold SCL_CONFIG as writer, or must be the sync thread 3910 * (which holds SCL_CONFIG as reader). There's only one sync thread, 3911 * so this is sufficient to ensure mutual exclusion. 3912 */ 3913 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3914 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3915 spa_config_held(spa, SCL_CONFIG, RW_READER))); 3916 3917 if (vd == rvd) { 3918 for (c = 0; c < rvd->vdev_children; c++) 3919 vdev_config_dirty(rvd->vdev_child[c]); 3920 } else { 3921 ASSERT(vd == vd->vdev_top); 3922 3923 if (!list_link_active(&vd->vdev_config_dirty_node) && 3924 vdev_is_concrete(vd)) { 3925 list_insert_head(&spa->spa_config_dirty_list, vd); 3926 } 3927 } 3928 } 3929 3930 void 3931 vdev_config_clean(vdev_t *vd) 3932 { 3933 spa_t *spa = vd->vdev_spa; 3934 3935 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3936 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3937 spa_config_held(spa, SCL_CONFIG, RW_READER))); 3938 3939 ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 3940 list_remove(&spa->spa_config_dirty_list, vd); 3941 } 3942 3943 /* 3944 * Mark a top-level vdev's state as dirty, so that the next pass of 3945 * spa_sync() can convert this into vdev_config_dirty(). We distinguish 3946 * the state changes from larger config changes because they require 3947 * much less locking, and are often needed for administrative actions. 3948 */ 3949 void 3950 vdev_state_dirty(vdev_t *vd) 3951 { 3952 spa_t *spa = vd->vdev_spa; 3953 3954 ASSERT(spa_writeable(spa)); 3955 ASSERT(vd == vd->vdev_top); 3956 3957 /* 3958 * The state list is protected by the SCL_STATE lock. The caller 3959 * must either hold SCL_STATE as writer, or must be the sync thread 3960 * (which holds SCL_STATE as reader). There's only one sync thread, 3961 * so this is sufficient to ensure mutual exclusion. 3962 */ 3963 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3964 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3965 spa_config_held(spa, SCL_STATE, RW_READER))); 3966 3967 if (!list_link_active(&vd->vdev_state_dirty_node) && 3968 vdev_is_concrete(vd)) 3969 list_insert_head(&spa->spa_state_dirty_list, vd); 3970 } 3971 3972 void 3973 vdev_state_clean(vdev_t *vd) 3974 { 3975 spa_t *spa = vd->vdev_spa; 3976 3977 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3978 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3979 spa_config_held(spa, SCL_STATE, RW_READER))); 3980 3981 ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3982 list_remove(&spa->spa_state_dirty_list, vd); 3983 } 3984 3985 /* 3986 * Propagate vdev state up from children to parent. 3987 */ 3988 void 3989 vdev_propagate_state(vdev_t *vd) 3990 { 3991 spa_t *spa = vd->vdev_spa; 3992 vdev_t *rvd = spa->spa_root_vdev; 3993 int degraded = 0, faulted = 0; 3994 int corrupted = 0; 3995 vdev_t *child; 3996 3997 if (vd->vdev_children > 0) { 3998 for (int c = 0; c < vd->vdev_children; c++) { 3999 child = vd->vdev_child[c]; 4000 4001 /* 4002 * Don't factor holes or indirect vdevs into the 4003 * decision. 4004 */ 4005 if (!vdev_is_concrete(child)) 4006 continue; 4007 4008 if (!vdev_readable(child) || 4009 (!vdev_writeable(child) && spa_writeable(spa))) { 4010 /* 4011 * Root special: if there is a top-level log 4012 * device, treat the root vdev as if it were 4013 * degraded. 4014 */ 4015 if (child->vdev_islog && vd == rvd) 4016 degraded++; 4017 else 4018 faulted++; 4019 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 4020 degraded++; 4021 } 4022 4023 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 4024 corrupted++; 4025 } 4026 4027 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 4028 4029 /* 4030 * Root special: if there is a top-level vdev that cannot be 4031 * opened due to corrupted metadata, then propagate the root 4032 * vdev's aux state as 'corrupt' rather than 'insufficient 4033 * replicas'. 4034 */ 4035 if (corrupted && vd == rvd && 4036 rvd->vdev_state == VDEV_STATE_CANT_OPEN) 4037 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 4038 VDEV_AUX_CORRUPT_DATA); 4039 } 4040 4041 if (vd->vdev_parent) 4042 vdev_propagate_state(vd->vdev_parent); 4043 } 4044 4045 /* 4046 * Set a vdev's state. If this is during an open, we don't update the parent 4047 * state, because we're in the process of opening children depth-first. 4048 * Otherwise, we propagate the change to the parent. 4049 * 4050 * If this routine places a device in a faulted state, an appropriate ereport is 4051 * generated. 4052 */ 4053 void 4054 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 4055 { 4056 uint64_t save_state; 4057 spa_t *spa = vd->vdev_spa; 4058 4059 if (state == vd->vdev_state) { 4060 vd->vdev_stat.vs_aux = aux; 4061 return; 4062 } 4063 4064 save_state = vd->vdev_state; 4065 4066 vd->vdev_state = state; 4067 vd->vdev_stat.vs_aux = aux; 4068 4069 /* 4070 * If we are setting the vdev state to anything but an open state, then 4071 * always close the underlying device unless the device has requested 4072 * a delayed close (i.e. we're about to remove or fault the device). 4073 * Otherwise, we keep accessible but invalid devices open forever. 4074 * We don't call vdev_close() itself, because that implies some extra 4075 * checks (offline, etc) that we don't want here. This is limited to 4076 * leaf devices, because otherwise closing the device will affect other 4077 * children. 4078 */ 4079 if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 4080 vd->vdev_ops->vdev_op_leaf) 4081 vd->vdev_ops->vdev_op_close(vd); 4082 4083 /* 4084 * If we have brought this vdev back into service, we need 4085 * to notify fmd so that it can gracefully repair any outstanding 4086 * cases due to a missing device. We do this in all cases, even those 4087 * that probably don't correlate to a repaired fault. This is sure to 4088 * catch all cases, and we let the zfs-retire agent sort it out. If 4089 * this is a transient state it's OK, as the retire agent will 4090 * double-check the state of the vdev before repairing it. 4091 */ 4092 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 4093 vd->vdev_prevstate != state) 4094 zfs_post_state_change(spa, vd); 4095 4096 if (vd->vdev_removed && 4097 state == VDEV_STATE_CANT_OPEN && 4098 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 4099 /* 4100 * If the previous state is set to VDEV_STATE_REMOVED, then this 4101 * device was previously marked removed and someone attempted to 4102 * reopen it. If this failed due to a nonexistent device, then 4103 * keep the device in the REMOVED state. We also let this be if 4104 * it is one of our special test online cases, which is only 4105 * attempting to online the device and shouldn't generate an FMA 4106 * fault. 4107 */ 4108 vd->vdev_state = VDEV_STATE_REMOVED; 4109 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 4110 } else if (state == VDEV_STATE_REMOVED) { 4111 vd->vdev_removed = B_TRUE; 4112 } else if (state == VDEV_STATE_CANT_OPEN) { 4113 /* 4114 * If we fail to open a vdev during an import or recovery, we 4115 * mark it as "not available", which signifies that it was 4116 * never there to begin with. Failure to open such a device 4117 * is not considered an error. 4118 */ 4119 if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 4120 spa_load_state(spa) == SPA_LOAD_RECOVER) && 4121 vd->vdev_ops->vdev_op_leaf) 4122 vd->vdev_not_present = 1; 4123 4124 /* 4125 * Post the appropriate ereport. If the 'prevstate' field is 4126 * set to something other than VDEV_STATE_UNKNOWN, it indicates 4127 * that this is part of a vdev_reopen(). In this case, we don't 4128 * want to post the ereport if the device was already in the 4129 * CANT_OPEN state beforehand. 4130 * 4131 * If the 'checkremove' flag is set, then this is an attempt to 4132 * online the device in response to an insertion event. If we 4133 * hit this case, then we have detected an insertion event for a 4134 * faulted or offline device that wasn't in the removed state. 4135 * In this scenario, we don't post an ereport because we are 4136 * about to replace the device, or attempt an online with 4137 * vdev_forcefault, which will generate the fault for us. 4138 */ 4139 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 4140 !vd->vdev_not_present && !vd->vdev_checkremove && 4141 vd != spa->spa_root_vdev) { 4142 const char *class; 4143 4144 switch (aux) { 4145 case VDEV_AUX_OPEN_FAILED: 4146 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 4147 break; 4148 case VDEV_AUX_CORRUPT_DATA: 4149 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 4150 break; 4151 case VDEV_AUX_NO_REPLICAS: 4152 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 4153 break; 4154 case VDEV_AUX_BAD_GUID_SUM: 4155 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 4156 break; 4157 case VDEV_AUX_TOO_SMALL: 4158 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 4159 break; 4160 case VDEV_AUX_BAD_LABEL: 4161 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 4162 break; 4163 default: 4164 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 4165 } 4166 4167 zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 4168 } 4169 4170 /* Erase any notion of persistent removed state */ 4171 vd->vdev_removed = B_FALSE; 4172 } else { 4173 vd->vdev_removed = B_FALSE; 4174 } 4175 4176 if (!isopen && vd->vdev_parent) 4177 vdev_propagate_state(vd->vdev_parent); 4178 } 4179 4180 boolean_t 4181 vdev_children_are_offline(vdev_t *vd) 4182 { 4183 ASSERT(!vd->vdev_ops->vdev_op_leaf); 4184 4185 for (uint64_t i = 0; i < vd->vdev_children; i++) { 4186 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 4187 return (B_FALSE); 4188 } 4189 4190 return (B_TRUE); 4191 } 4192 4193 /* 4194 * Check the vdev configuration to ensure that it's capable of supporting 4195 * a root pool. We do not support partial configuration. 4196 * In addition, only a single top-level vdev is allowed. 4197 */ 4198 boolean_t 4199 vdev_is_bootable(vdev_t *vd) 4200 { 4201 if (!vd->vdev_ops->vdev_op_leaf) { 4202 char *vdev_type = vd->vdev_ops->vdev_op_type; 4203 4204 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 4205 vd->vdev_children > 1) { 4206 return (B_FALSE); 4207 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 || 4208 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) { 4209 return (B_FALSE); 4210 } 4211 } 4212 4213 for (int c = 0; c < vd->vdev_children; c++) { 4214 if (!vdev_is_bootable(vd->vdev_child[c])) 4215 return (B_FALSE); 4216 } 4217 return (B_TRUE); 4218 } 4219 4220 boolean_t 4221 vdev_is_concrete(vdev_t *vd) 4222 { 4223 vdev_ops_t *ops = vd->vdev_ops; 4224 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 4225 ops == &vdev_missing_ops || ops == &vdev_root_ops) { 4226 return (B_FALSE); 4227 } else { 4228 return (B_TRUE); 4229 } 4230 } 4231 4232 /* 4233 * Determine if a log device has valid content. If the vdev was 4234 * removed or faulted in the MOS config then we know that 4235 * the content on the log device has already been written to the pool. 4236 */ 4237 boolean_t 4238 vdev_log_state_valid(vdev_t *vd) 4239 { 4240 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 4241 !vd->vdev_removed) 4242 return (B_TRUE); 4243 4244 for (int c = 0; c < vd->vdev_children; c++) 4245 if (vdev_log_state_valid(vd->vdev_child[c])) 4246 return (B_TRUE); 4247 4248 return (B_FALSE); 4249 } 4250 4251 /* 4252 * Expand a vdev if possible. 4253 */ 4254 void 4255 vdev_expand(vdev_t *vd, uint64_t txg) 4256 { 4257 ASSERT(vd->vdev_top == vd); 4258 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4259 ASSERT(vdev_is_concrete(vd)); 4260 4261 vdev_set_deflate_ratio(vd); 4262 4263 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 4264 vdev_is_concrete(vd)) { 4265 vdev_metaslab_group_create(vd); 4266 VERIFY(vdev_metaslab_init(vd, txg) == 0); 4267 vdev_config_dirty(vd); 4268 } 4269 } 4270 4271 /* 4272 * Split a vdev. 4273 */ 4274 void 4275 vdev_split(vdev_t *vd) 4276 { 4277 vdev_t *cvd, *pvd = vd->vdev_parent; 4278 4279 vdev_remove_child(pvd, vd); 4280 vdev_compact_children(pvd); 4281 4282 cvd = pvd->vdev_child[0]; 4283 if (pvd->vdev_children == 1) { 4284 vdev_remove_parent(cvd); 4285 cvd->vdev_splitting = B_TRUE; 4286 } 4287 vdev_propagate_state(cvd); 4288 } 4289 4290 void 4291 vdev_deadman(vdev_t *vd) 4292 { 4293 for (int c = 0; c < vd->vdev_children; c++) { 4294 vdev_t *cvd = vd->vdev_child[c]; 4295 4296 vdev_deadman(cvd); 4297 } 4298 4299 if (vd->vdev_ops->vdev_op_leaf) { 4300 vdev_queue_t *vq = &vd->vdev_queue; 4301 4302 mutex_enter(&vq->vq_lock); 4303 if (avl_numnodes(&vq->vq_active_tree) > 0) { 4304 spa_t *spa = vd->vdev_spa; 4305 zio_t *fio; 4306 uint64_t delta; 4307 4308 /* 4309 * Look at the head of all the pending queues, 4310 * if any I/O has been outstanding for longer than 4311 * the spa_deadman_synctime we panic the system. 4312 */ 4313 fio = avl_first(&vq->vq_active_tree); 4314 delta = gethrtime() - fio->io_timestamp; 4315 if (delta > spa_deadman_synctime(spa)) { 4316 vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 4317 "%lluns, delta %lluns, last io %lluns", 4318 fio->io_timestamp, (u_longlong_t)delta, 4319 vq->vq_io_complete_ts); 4320 fm_panic("I/O to pool '%s' appears to be " 4321 "hung.", spa_name(spa)); 4322 } 4323 } 4324 mutex_exit(&vq->vq_lock); 4325 } 4326 } 4327