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 2011 Nexenta Systems, Inc. All rights reserved. 25 * Copyright (c) 2012 by Delphix. All rights reserved. 26 */ 27 28 #include <sys/zfs_context.h> 29 #include <sys/fm/fs/zfs.h> 30 #include <sys/spa.h> 31 #include <sys/spa_impl.h> 32 #include <sys/dmu.h> 33 #include <sys/dmu_tx.h> 34 #include <sys/vdev_impl.h> 35 #include <sys/uberblock_impl.h> 36 #include <sys/metaslab.h> 37 #include <sys/metaslab_impl.h> 38 #include <sys/space_map.h> 39 #include <sys/zio.h> 40 #include <sys/zap.h> 41 #include <sys/fs/zfs.h> 42 #include <sys/arc.h> 43 #include <sys/zil.h> 44 #include <sys/dsl_scan.h> 45 46 /* 47 * Virtual device management. 48 */ 49 50 static vdev_ops_t *vdev_ops_table[] = { 51 &vdev_root_ops, 52 &vdev_raidz_ops, 53 &vdev_mirror_ops, 54 &vdev_replacing_ops, 55 &vdev_spare_ops, 56 &vdev_disk_ops, 57 &vdev_file_ops, 58 &vdev_missing_ops, 59 &vdev_hole_ops, 60 NULL 61 }; 62 63 /* maximum scrub/resilver I/O queue per leaf vdev */ 64 int zfs_scrub_limit = 10; 65 66 /* 67 * Given a vdev type, return the appropriate ops vector. 68 */ 69 static vdev_ops_t * 70 vdev_getops(const char *type) 71 { 72 vdev_ops_t *ops, **opspp; 73 74 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 75 if (strcmp(ops->vdev_op_type, type) == 0) 76 break; 77 78 return (ops); 79 } 80 81 /* 82 * Default asize function: return the MAX of psize with the asize of 83 * all children. This is what's used by anything other than RAID-Z. 84 */ 85 uint64_t 86 vdev_default_asize(vdev_t *vd, uint64_t psize) 87 { 88 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 89 uint64_t csize; 90 91 for (int c = 0; c < vd->vdev_children; c++) { 92 csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 93 asize = MAX(asize, csize); 94 } 95 96 return (asize); 97 } 98 99 /* 100 * Get the minimum allocatable size. We define the allocatable size as 101 * the vdev's asize rounded to the nearest metaslab. This allows us to 102 * replace or attach devices which don't have the same physical size but 103 * can still satisfy the same number of allocations. 104 */ 105 uint64_t 106 vdev_get_min_asize(vdev_t *vd) 107 { 108 vdev_t *pvd = vd->vdev_parent; 109 110 /* 111 * If our parent is NULL (inactive spare or cache) or is the root, 112 * just return our own asize. 113 */ 114 if (pvd == NULL) 115 return (vd->vdev_asize); 116 117 /* 118 * The top-level vdev just returns the allocatable size rounded 119 * to the nearest metaslab. 120 */ 121 if (vd == vd->vdev_top) 122 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 123 124 /* 125 * The allocatable space for a raidz vdev is N * sizeof(smallest child), 126 * so each child must provide at least 1/Nth of its asize. 127 */ 128 if (pvd->vdev_ops == &vdev_raidz_ops) 129 return (pvd->vdev_min_asize / pvd->vdev_children); 130 131 return (pvd->vdev_min_asize); 132 } 133 134 void 135 vdev_set_min_asize(vdev_t *vd) 136 { 137 vd->vdev_min_asize = vdev_get_min_asize(vd); 138 139 for (int c = 0; c < vd->vdev_children; c++) 140 vdev_set_min_asize(vd->vdev_child[c]); 141 } 142 143 vdev_t * 144 vdev_lookup_top(spa_t *spa, uint64_t vdev) 145 { 146 vdev_t *rvd = spa->spa_root_vdev; 147 148 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 149 150 if (vdev < rvd->vdev_children) { 151 ASSERT(rvd->vdev_child[vdev] != NULL); 152 return (rvd->vdev_child[vdev]); 153 } 154 155 return (NULL); 156 } 157 158 vdev_t * 159 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 160 { 161 vdev_t *mvd; 162 163 if (vd->vdev_guid == guid) 164 return (vd); 165 166 for (int c = 0; c < vd->vdev_children; c++) 167 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 168 NULL) 169 return (mvd); 170 171 return (NULL); 172 } 173 174 void 175 vdev_add_child(vdev_t *pvd, vdev_t *cvd) 176 { 177 size_t oldsize, newsize; 178 uint64_t id = cvd->vdev_id; 179 vdev_t **newchild; 180 181 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 182 ASSERT(cvd->vdev_parent == NULL); 183 184 cvd->vdev_parent = pvd; 185 186 if (pvd == NULL) 187 return; 188 189 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 190 191 oldsize = pvd->vdev_children * sizeof (vdev_t *); 192 pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 193 newsize = pvd->vdev_children * sizeof (vdev_t *); 194 195 newchild = kmem_zalloc(newsize, KM_SLEEP); 196 if (pvd->vdev_child != NULL) { 197 bcopy(pvd->vdev_child, newchild, oldsize); 198 kmem_free(pvd->vdev_child, oldsize); 199 } 200 201 pvd->vdev_child = newchild; 202 pvd->vdev_child[id] = cvd; 203 204 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 205 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 206 207 /* 208 * Walk up all ancestors to update guid sum. 209 */ 210 for (; pvd != NULL; pvd = pvd->vdev_parent) 211 pvd->vdev_guid_sum += cvd->vdev_guid_sum; 212 } 213 214 void 215 vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 216 { 217 int c; 218 uint_t id = cvd->vdev_id; 219 220 ASSERT(cvd->vdev_parent == pvd); 221 222 if (pvd == NULL) 223 return; 224 225 ASSERT(id < pvd->vdev_children); 226 ASSERT(pvd->vdev_child[id] == cvd); 227 228 pvd->vdev_child[id] = NULL; 229 cvd->vdev_parent = NULL; 230 231 for (c = 0; c < pvd->vdev_children; c++) 232 if (pvd->vdev_child[c]) 233 break; 234 235 if (c == pvd->vdev_children) { 236 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 237 pvd->vdev_child = NULL; 238 pvd->vdev_children = 0; 239 } 240 241 /* 242 * Walk up all ancestors to update guid sum. 243 */ 244 for (; pvd != NULL; pvd = pvd->vdev_parent) 245 pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 246 } 247 248 /* 249 * Remove any holes in the child array. 250 */ 251 void 252 vdev_compact_children(vdev_t *pvd) 253 { 254 vdev_t **newchild, *cvd; 255 int oldc = pvd->vdev_children; 256 int newc; 257 258 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 259 260 for (int c = newc = 0; c < oldc; c++) 261 if (pvd->vdev_child[c]) 262 newc++; 263 264 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 265 266 for (int c = newc = 0; c < oldc; c++) { 267 if ((cvd = pvd->vdev_child[c]) != NULL) { 268 newchild[newc] = cvd; 269 cvd->vdev_id = newc++; 270 } 271 } 272 273 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 274 pvd->vdev_child = newchild; 275 pvd->vdev_children = newc; 276 } 277 278 /* 279 * Allocate and minimally initialize a vdev_t. 280 */ 281 vdev_t * 282 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 283 { 284 vdev_t *vd; 285 286 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 287 288 if (spa->spa_root_vdev == NULL) { 289 ASSERT(ops == &vdev_root_ops); 290 spa->spa_root_vdev = vd; 291 spa->spa_load_guid = spa_generate_guid(NULL); 292 } 293 294 if (guid == 0 && ops != &vdev_hole_ops) { 295 if (spa->spa_root_vdev == vd) { 296 /* 297 * The root vdev's guid will also be the pool guid, 298 * which must be unique among all pools. 299 */ 300 guid = spa_generate_guid(NULL); 301 } else { 302 /* 303 * Any other vdev's guid must be unique within the pool. 304 */ 305 guid = spa_generate_guid(spa); 306 } 307 ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 308 } 309 310 vd->vdev_spa = spa; 311 vd->vdev_id = id; 312 vd->vdev_guid = guid; 313 vd->vdev_guid_sum = guid; 314 vd->vdev_ops = ops; 315 vd->vdev_state = VDEV_STATE_CLOSED; 316 vd->vdev_ishole = (ops == &vdev_hole_ops); 317 318 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 319 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 320 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 321 for (int t = 0; t < DTL_TYPES; t++) { 322 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 323 &vd->vdev_dtl_lock); 324 } 325 txg_list_create(&vd->vdev_ms_list, 326 offsetof(struct metaslab, ms_txg_node)); 327 txg_list_create(&vd->vdev_dtl_list, 328 offsetof(struct vdev, vdev_dtl_node)); 329 vd->vdev_stat.vs_timestamp = gethrtime(); 330 vdev_queue_init(vd); 331 vdev_cache_init(vd); 332 333 return (vd); 334 } 335 336 /* 337 * Allocate a new vdev. The 'alloctype' is used to control whether we are 338 * creating a new vdev or loading an existing one - the behavior is slightly 339 * different for each case. 340 */ 341 int 342 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 343 int alloctype) 344 { 345 vdev_ops_t *ops; 346 char *type; 347 uint64_t guid = 0, islog, nparity; 348 vdev_t *vd; 349 350 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 351 352 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 353 return (EINVAL); 354 355 if ((ops = vdev_getops(type)) == NULL) 356 return (EINVAL); 357 358 /* 359 * If this is a load, get the vdev guid from the nvlist. 360 * Otherwise, vdev_alloc_common() will generate one for us. 361 */ 362 if (alloctype == VDEV_ALLOC_LOAD) { 363 uint64_t label_id; 364 365 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 366 label_id != id) 367 return (EINVAL); 368 369 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 370 return (EINVAL); 371 } else if (alloctype == VDEV_ALLOC_SPARE) { 372 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 373 return (EINVAL); 374 } else if (alloctype == VDEV_ALLOC_L2CACHE) { 375 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 376 return (EINVAL); 377 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 378 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 379 return (EINVAL); 380 } 381 382 /* 383 * The first allocated vdev must be of type 'root'. 384 */ 385 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 386 return (EINVAL); 387 388 /* 389 * Determine whether we're a log vdev. 390 */ 391 islog = 0; 392 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 393 if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 394 return (ENOTSUP); 395 396 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 397 return (ENOTSUP); 398 399 /* 400 * Set the nparity property for RAID-Z vdevs. 401 */ 402 nparity = -1ULL; 403 if (ops == &vdev_raidz_ops) { 404 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 405 &nparity) == 0) { 406 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 407 return (EINVAL); 408 /* 409 * Previous versions could only support 1 or 2 parity 410 * device. 411 */ 412 if (nparity > 1 && 413 spa_version(spa) < SPA_VERSION_RAIDZ2) 414 return (ENOTSUP); 415 if (nparity > 2 && 416 spa_version(spa) < SPA_VERSION_RAIDZ3) 417 return (ENOTSUP); 418 } else { 419 /* 420 * We require the parity to be specified for SPAs that 421 * support multiple parity levels. 422 */ 423 if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 424 return (EINVAL); 425 /* 426 * Otherwise, we default to 1 parity device for RAID-Z. 427 */ 428 nparity = 1; 429 } 430 } else { 431 nparity = 0; 432 } 433 ASSERT(nparity != -1ULL); 434 435 vd = vdev_alloc_common(spa, id, guid, ops); 436 437 vd->vdev_islog = islog; 438 vd->vdev_nparity = nparity; 439 440 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 441 vd->vdev_path = spa_strdup(vd->vdev_path); 442 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 443 vd->vdev_devid = spa_strdup(vd->vdev_devid); 444 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 445 &vd->vdev_physpath) == 0) 446 vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 448 vd->vdev_fru = spa_strdup(vd->vdev_fru); 449 450 /* 451 * Set the whole_disk property. If it's not specified, leave the value 452 * as -1. 453 */ 454 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 455 &vd->vdev_wholedisk) != 0) 456 vd->vdev_wholedisk = -1ULL; 457 458 /* 459 * Look for the 'not present' flag. This will only be set if the device 460 * was not present at the time of import. 461 */ 462 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 463 &vd->vdev_not_present); 464 465 /* 466 * Get the alignment requirement. 467 */ 468 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 469 470 /* 471 * Retrieve the vdev creation time. 472 */ 473 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 474 &vd->vdev_crtxg); 475 476 /* 477 * If we're a top-level vdev, try to load the allocation parameters. 478 */ 479 if (parent && !parent->vdev_parent && 480 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 481 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 482 &vd->vdev_ms_array); 483 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 484 &vd->vdev_ms_shift); 485 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 486 &vd->vdev_asize); 487 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 488 &vd->vdev_removing); 489 } 490 491 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 492 ASSERT(alloctype == VDEV_ALLOC_LOAD || 493 alloctype == VDEV_ALLOC_ADD || 494 alloctype == VDEV_ALLOC_SPLIT || 495 alloctype == VDEV_ALLOC_ROOTPOOL); 496 vd->vdev_mg = metaslab_group_create(islog ? 497 spa_log_class(spa) : spa_normal_class(spa), vd); 498 } 499 500 /* 501 * If we're a leaf vdev, try to load the DTL object and other state. 502 */ 503 if (vd->vdev_ops->vdev_op_leaf && 504 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 505 alloctype == VDEV_ALLOC_ROOTPOOL)) { 506 if (alloctype == VDEV_ALLOC_LOAD) { 507 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 508 &vd->vdev_dtl_smo.smo_object); 509 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 510 &vd->vdev_unspare); 511 } 512 513 if (alloctype == VDEV_ALLOC_ROOTPOOL) { 514 uint64_t spare = 0; 515 516 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 517 &spare) == 0 && spare) 518 spa_spare_add(vd); 519 } 520 521 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 522 &vd->vdev_offline); 523 524 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING, 525 &vd->vdev_resilvering); 526 527 /* 528 * When importing a pool, we want to ignore the persistent fault 529 * state, as the diagnosis made on another system may not be 530 * valid in the current context. Local vdevs will 531 * remain in the faulted state. 532 */ 533 if (spa_load_state(spa) == SPA_LOAD_OPEN) { 534 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 535 &vd->vdev_faulted); 536 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 537 &vd->vdev_degraded); 538 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 539 &vd->vdev_removed); 540 541 if (vd->vdev_faulted || vd->vdev_degraded) { 542 char *aux; 543 544 vd->vdev_label_aux = 545 VDEV_AUX_ERR_EXCEEDED; 546 if (nvlist_lookup_string(nv, 547 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 548 strcmp(aux, "external") == 0) 549 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 550 } 551 } 552 } 553 554 /* 555 * Add ourselves to the parent's list of children. 556 */ 557 vdev_add_child(parent, vd); 558 559 *vdp = vd; 560 561 return (0); 562 } 563 564 void 565 vdev_free(vdev_t *vd) 566 { 567 spa_t *spa = vd->vdev_spa; 568 569 /* 570 * vdev_free() implies closing the vdev first. This is simpler than 571 * trying to ensure complicated semantics for all callers. 572 */ 573 vdev_close(vd); 574 575 ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 576 ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 577 578 /* 579 * Free all children. 580 */ 581 for (int c = 0; c < vd->vdev_children; c++) 582 vdev_free(vd->vdev_child[c]); 583 584 ASSERT(vd->vdev_child == NULL); 585 ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 586 587 /* 588 * Discard allocation state. 589 */ 590 if (vd->vdev_mg != NULL) { 591 vdev_metaslab_fini(vd); 592 metaslab_group_destroy(vd->vdev_mg); 593 } 594 595 ASSERT3U(vd->vdev_stat.vs_space, ==, 0); 596 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); 597 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 598 599 /* 600 * Remove this vdev from its parent's child list. 601 */ 602 vdev_remove_child(vd->vdev_parent, vd); 603 604 ASSERT(vd->vdev_parent == NULL); 605 606 /* 607 * Clean up vdev structure. 608 */ 609 vdev_queue_fini(vd); 610 vdev_cache_fini(vd); 611 612 if (vd->vdev_path) 613 spa_strfree(vd->vdev_path); 614 if (vd->vdev_devid) 615 spa_strfree(vd->vdev_devid); 616 if (vd->vdev_physpath) 617 spa_strfree(vd->vdev_physpath); 618 if (vd->vdev_fru) 619 spa_strfree(vd->vdev_fru); 620 621 if (vd->vdev_isspare) 622 spa_spare_remove(vd); 623 if (vd->vdev_isl2cache) 624 spa_l2cache_remove(vd); 625 626 txg_list_destroy(&vd->vdev_ms_list); 627 txg_list_destroy(&vd->vdev_dtl_list); 628 629 mutex_enter(&vd->vdev_dtl_lock); 630 for (int t = 0; t < DTL_TYPES; t++) { 631 space_map_unload(&vd->vdev_dtl[t]); 632 space_map_destroy(&vd->vdev_dtl[t]); 633 } 634 mutex_exit(&vd->vdev_dtl_lock); 635 636 mutex_destroy(&vd->vdev_dtl_lock); 637 mutex_destroy(&vd->vdev_stat_lock); 638 mutex_destroy(&vd->vdev_probe_lock); 639 640 if (vd == spa->spa_root_vdev) 641 spa->spa_root_vdev = NULL; 642 643 kmem_free(vd, sizeof (vdev_t)); 644 } 645 646 /* 647 * Transfer top-level vdev state from svd to tvd. 648 */ 649 static void 650 vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 651 { 652 spa_t *spa = svd->vdev_spa; 653 metaslab_t *msp; 654 vdev_t *vd; 655 int t; 656 657 ASSERT(tvd == tvd->vdev_top); 658 659 tvd->vdev_ms_array = svd->vdev_ms_array; 660 tvd->vdev_ms_shift = svd->vdev_ms_shift; 661 tvd->vdev_ms_count = svd->vdev_ms_count; 662 663 svd->vdev_ms_array = 0; 664 svd->vdev_ms_shift = 0; 665 svd->vdev_ms_count = 0; 666 667 if (tvd->vdev_mg) 668 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 669 tvd->vdev_mg = svd->vdev_mg; 670 tvd->vdev_ms = svd->vdev_ms; 671 672 svd->vdev_mg = NULL; 673 svd->vdev_ms = NULL; 674 675 if (tvd->vdev_mg != NULL) 676 tvd->vdev_mg->mg_vd = tvd; 677 678 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 679 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 680 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 681 682 svd->vdev_stat.vs_alloc = 0; 683 svd->vdev_stat.vs_space = 0; 684 svd->vdev_stat.vs_dspace = 0; 685 686 for (t = 0; t < TXG_SIZE; t++) { 687 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 688 (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 689 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 690 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 691 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 692 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 693 } 694 695 if (list_link_active(&svd->vdev_config_dirty_node)) { 696 vdev_config_clean(svd); 697 vdev_config_dirty(tvd); 698 } 699 700 if (list_link_active(&svd->vdev_state_dirty_node)) { 701 vdev_state_clean(svd); 702 vdev_state_dirty(tvd); 703 } 704 705 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 706 svd->vdev_deflate_ratio = 0; 707 708 tvd->vdev_islog = svd->vdev_islog; 709 svd->vdev_islog = 0; 710 } 711 712 static void 713 vdev_top_update(vdev_t *tvd, vdev_t *vd) 714 { 715 if (vd == NULL) 716 return; 717 718 vd->vdev_top = tvd; 719 720 for (int c = 0; c < vd->vdev_children; c++) 721 vdev_top_update(tvd, vd->vdev_child[c]); 722 } 723 724 /* 725 * Add a mirror/replacing vdev above an existing vdev. 726 */ 727 vdev_t * 728 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 729 { 730 spa_t *spa = cvd->vdev_spa; 731 vdev_t *pvd = cvd->vdev_parent; 732 vdev_t *mvd; 733 734 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 735 736 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 737 738 mvd->vdev_asize = cvd->vdev_asize; 739 mvd->vdev_min_asize = cvd->vdev_min_asize; 740 mvd->vdev_max_asize = cvd->vdev_max_asize; 741 mvd->vdev_ashift = cvd->vdev_ashift; 742 mvd->vdev_state = cvd->vdev_state; 743 mvd->vdev_crtxg = cvd->vdev_crtxg; 744 745 vdev_remove_child(pvd, cvd); 746 vdev_add_child(pvd, mvd); 747 cvd->vdev_id = mvd->vdev_children; 748 vdev_add_child(mvd, cvd); 749 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 750 751 if (mvd == mvd->vdev_top) 752 vdev_top_transfer(cvd, mvd); 753 754 return (mvd); 755 } 756 757 /* 758 * Remove a 1-way mirror/replacing vdev from the tree. 759 */ 760 void 761 vdev_remove_parent(vdev_t *cvd) 762 { 763 vdev_t *mvd = cvd->vdev_parent; 764 vdev_t *pvd = mvd->vdev_parent; 765 766 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 767 768 ASSERT(mvd->vdev_children == 1); 769 ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 770 mvd->vdev_ops == &vdev_replacing_ops || 771 mvd->vdev_ops == &vdev_spare_ops); 772 cvd->vdev_ashift = mvd->vdev_ashift; 773 774 vdev_remove_child(mvd, cvd); 775 vdev_remove_child(pvd, mvd); 776 777 /* 778 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 779 * Otherwise, we could have detached an offline device, and when we 780 * go to import the pool we'll think we have two top-level vdevs, 781 * instead of a different version of the same top-level vdev. 782 */ 783 if (mvd->vdev_top == mvd) { 784 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 785 cvd->vdev_orig_guid = cvd->vdev_guid; 786 cvd->vdev_guid += guid_delta; 787 cvd->vdev_guid_sum += guid_delta; 788 } 789 cvd->vdev_id = mvd->vdev_id; 790 vdev_add_child(pvd, cvd); 791 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 792 793 if (cvd == cvd->vdev_top) 794 vdev_top_transfer(mvd, cvd); 795 796 ASSERT(mvd->vdev_children == 0); 797 vdev_free(mvd); 798 } 799 800 int 801 vdev_metaslab_init(vdev_t *vd, uint64_t txg) 802 { 803 spa_t *spa = vd->vdev_spa; 804 objset_t *mos = spa->spa_meta_objset; 805 uint64_t m; 806 uint64_t oldc = vd->vdev_ms_count; 807 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 808 metaslab_t **mspp; 809 int error; 810 811 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 812 813 /* 814 * This vdev is not being allocated from yet or is a hole. 815 */ 816 if (vd->vdev_ms_shift == 0) 817 return (0); 818 819 ASSERT(!vd->vdev_ishole); 820 821 /* 822 * Compute the raidz-deflation ratio. Note, we hard-code 823 * in 128k (1 << 17) because it is the current "typical" blocksize. 824 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 825 * or we will inconsistently account for existing bp's. 826 */ 827 vd->vdev_deflate_ratio = (1 << 17) / 828 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 829 830 ASSERT(oldc <= newc); 831 832 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 833 834 if (oldc != 0) { 835 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 836 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 837 } 838 839 vd->vdev_ms = mspp; 840 vd->vdev_ms_count = newc; 841 842 for (m = oldc; m < newc; m++) { 843 space_map_obj_t smo = { 0, 0, 0 }; 844 if (txg == 0) { 845 uint64_t object = 0; 846 error = dmu_read(mos, vd->vdev_ms_array, 847 m * sizeof (uint64_t), sizeof (uint64_t), &object, 848 DMU_READ_PREFETCH); 849 if (error) 850 return (error); 851 if (object != 0) { 852 dmu_buf_t *db; 853 error = dmu_bonus_hold(mos, object, FTAG, &db); 854 if (error) 855 return (error); 856 ASSERT3U(db->db_size, >=, sizeof (smo)); 857 bcopy(db->db_data, &smo, sizeof (smo)); 858 ASSERT3U(smo.smo_object, ==, object); 859 dmu_buf_rele(db, FTAG); 860 } 861 } 862 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 863 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 864 } 865 866 if (txg == 0) 867 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 868 869 /* 870 * If the vdev is being removed we don't activate 871 * the metaslabs since we want to ensure that no new 872 * allocations are performed on this device. 873 */ 874 if (oldc == 0 && !vd->vdev_removing) 875 metaslab_group_activate(vd->vdev_mg); 876 877 if (txg == 0) 878 spa_config_exit(spa, SCL_ALLOC, FTAG); 879 880 return (0); 881 } 882 883 void 884 vdev_metaslab_fini(vdev_t *vd) 885 { 886 uint64_t m; 887 uint64_t count = vd->vdev_ms_count; 888 889 if (vd->vdev_ms != NULL) { 890 metaslab_group_passivate(vd->vdev_mg); 891 for (m = 0; m < count; m++) 892 if (vd->vdev_ms[m] != NULL) 893 metaslab_fini(vd->vdev_ms[m]); 894 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 895 vd->vdev_ms = NULL; 896 } 897 } 898 899 typedef struct vdev_probe_stats { 900 boolean_t vps_readable; 901 boolean_t vps_writeable; 902 int vps_flags; 903 } vdev_probe_stats_t; 904 905 static void 906 vdev_probe_done(zio_t *zio) 907 { 908 spa_t *spa = zio->io_spa; 909 vdev_t *vd = zio->io_vd; 910 vdev_probe_stats_t *vps = zio->io_private; 911 912 ASSERT(vd->vdev_probe_zio != NULL); 913 914 if (zio->io_type == ZIO_TYPE_READ) { 915 if (zio->io_error == 0) 916 vps->vps_readable = 1; 917 if (zio->io_error == 0 && spa_writeable(spa)) { 918 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 919 zio->io_offset, zio->io_size, zio->io_data, 920 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 921 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 922 } else { 923 zio_buf_free(zio->io_data, zio->io_size); 924 } 925 } else if (zio->io_type == ZIO_TYPE_WRITE) { 926 if (zio->io_error == 0) 927 vps->vps_writeable = 1; 928 zio_buf_free(zio->io_data, zio->io_size); 929 } else if (zio->io_type == ZIO_TYPE_NULL) { 930 zio_t *pio; 931 932 vd->vdev_cant_read |= !vps->vps_readable; 933 vd->vdev_cant_write |= !vps->vps_writeable; 934 935 if (vdev_readable(vd) && 936 (vdev_writeable(vd) || !spa_writeable(spa))) { 937 zio->io_error = 0; 938 } else { 939 ASSERT(zio->io_error != 0); 940 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 941 spa, vd, NULL, 0, 0); 942 zio->io_error = ENXIO; 943 } 944 945 mutex_enter(&vd->vdev_probe_lock); 946 ASSERT(vd->vdev_probe_zio == zio); 947 vd->vdev_probe_zio = NULL; 948 mutex_exit(&vd->vdev_probe_lock); 949 950 while ((pio = zio_walk_parents(zio)) != NULL) 951 if (!vdev_accessible(vd, pio)) 952 pio->io_error = ENXIO; 953 954 kmem_free(vps, sizeof (*vps)); 955 } 956 } 957 958 /* 959 * Determine whether this device is accessible by reading and writing 960 * to several known locations: the pad regions of each vdev label 961 * but the first (which we leave alone in case it contains a VTOC). 962 */ 963 zio_t * 964 vdev_probe(vdev_t *vd, zio_t *zio) 965 { 966 spa_t *spa = vd->vdev_spa; 967 vdev_probe_stats_t *vps = NULL; 968 zio_t *pio; 969 970 ASSERT(vd->vdev_ops->vdev_op_leaf); 971 972 /* 973 * Don't probe the probe. 974 */ 975 if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 976 return (NULL); 977 978 /* 979 * To prevent 'probe storms' when a device fails, we create 980 * just one probe i/o at a time. All zios that want to probe 981 * this vdev will become parents of the probe io. 982 */ 983 mutex_enter(&vd->vdev_probe_lock); 984 985 if ((pio = vd->vdev_probe_zio) == NULL) { 986 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 987 988 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 989 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 990 ZIO_FLAG_TRYHARD; 991 992 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 993 /* 994 * vdev_cant_read and vdev_cant_write can only 995 * transition from TRUE to FALSE when we have the 996 * SCL_ZIO lock as writer; otherwise they can only 997 * transition from FALSE to TRUE. This ensures that 998 * any zio looking at these values can assume that 999 * failures persist for the life of the I/O. That's 1000 * important because when a device has intermittent 1001 * connectivity problems, we want to ensure that 1002 * they're ascribed to the device (ENXIO) and not 1003 * the zio (EIO). 1004 * 1005 * Since we hold SCL_ZIO as writer here, clear both 1006 * values so the probe can reevaluate from first 1007 * principles. 1008 */ 1009 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1010 vd->vdev_cant_read = B_FALSE; 1011 vd->vdev_cant_write = B_FALSE; 1012 } 1013 1014 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1015 vdev_probe_done, vps, 1016 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1017 1018 /* 1019 * We can't change the vdev state in this context, so we 1020 * kick off an async task to do it on our behalf. 1021 */ 1022 if (zio != NULL) { 1023 vd->vdev_probe_wanted = B_TRUE; 1024 spa_async_request(spa, SPA_ASYNC_PROBE); 1025 } 1026 } 1027 1028 if (zio != NULL) 1029 zio_add_child(zio, pio); 1030 1031 mutex_exit(&vd->vdev_probe_lock); 1032 1033 if (vps == NULL) { 1034 ASSERT(zio != NULL); 1035 return (NULL); 1036 } 1037 1038 for (int l = 1; l < VDEV_LABELS; l++) { 1039 zio_nowait(zio_read_phys(pio, vd, 1040 vdev_label_offset(vd->vdev_psize, l, 1041 offsetof(vdev_label_t, vl_pad2)), 1042 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1043 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1044 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1045 } 1046 1047 if (zio == NULL) 1048 return (pio); 1049 1050 zio_nowait(pio); 1051 return (NULL); 1052 } 1053 1054 static void 1055 vdev_open_child(void *arg) 1056 { 1057 vdev_t *vd = arg; 1058 1059 vd->vdev_open_thread = curthread; 1060 vd->vdev_open_error = vdev_open(vd); 1061 vd->vdev_open_thread = NULL; 1062 } 1063 1064 boolean_t 1065 vdev_uses_zvols(vdev_t *vd) 1066 { 1067 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1068 strlen(ZVOL_DIR)) == 0) 1069 return (B_TRUE); 1070 for (int c = 0; c < vd->vdev_children; c++) 1071 if (vdev_uses_zvols(vd->vdev_child[c])) 1072 return (B_TRUE); 1073 return (B_FALSE); 1074 } 1075 1076 void 1077 vdev_open_children(vdev_t *vd) 1078 { 1079 taskq_t *tq; 1080 int children = vd->vdev_children; 1081 1082 /* 1083 * in order to handle pools on top of zvols, do the opens 1084 * in a single thread so that the same thread holds the 1085 * spa_namespace_lock 1086 */ 1087 if (vdev_uses_zvols(vd)) { 1088 for (int c = 0; c < children; c++) 1089 vd->vdev_child[c]->vdev_open_error = 1090 vdev_open(vd->vdev_child[c]); 1091 return; 1092 } 1093 tq = taskq_create("vdev_open", children, minclsyspri, 1094 children, children, TASKQ_PREPOPULATE); 1095 1096 for (int c = 0; c < children; c++) 1097 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1098 TQ_SLEEP) != NULL); 1099 1100 taskq_destroy(tq); 1101 } 1102 1103 /* 1104 * Prepare a virtual device for access. 1105 */ 1106 int 1107 vdev_open(vdev_t *vd) 1108 { 1109 spa_t *spa = vd->vdev_spa; 1110 int error; 1111 uint64_t osize = 0; 1112 uint64_t max_osize = 0; 1113 uint64_t asize, max_asize, psize; 1114 uint64_t ashift = 0; 1115 1116 ASSERT(vd->vdev_open_thread == curthread || 1117 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1118 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1119 vd->vdev_state == VDEV_STATE_CANT_OPEN || 1120 vd->vdev_state == VDEV_STATE_OFFLINE); 1121 1122 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1123 vd->vdev_cant_read = B_FALSE; 1124 vd->vdev_cant_write = B_FALSE; 1125 vd->vdev_min_asize = vdev_get_min_asize(vd); 1126 1127 /* 1128 * If this vdev is not removed, check its fault status. If it's 1129 * faulted, bail out of the open. 1130 */ 1131 if (!vd->vdev_removed && vd->vdev_faulted) { 1132 ASSERT(vd->vdev_children == 0); 1133 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1134 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1135 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1136 vd->vdev_label_aux); 1137 return (ENXIO); 1138 } else if (vd->vdev_offline) { 1139 ASSERT(vd->vdev_children == 0); 1140 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1141 return (ENXIO); 1142 } 1143 1144 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1145 1146 /* 1147 * Reset the vdev_reopening flag so that we actually close 1148 * the vdev on error. 1149 */ 1150 vd->vdev_reopening = B_FALSE; 1151 if (zio_injection_enabled && error == 0) 1152 error = zio_handle_device_injection(vd, NULL, ENXIO); 1153 1154 if (error) { 1155 if (vd->vdev_removed && 1156 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1157 vd->vdev_removed = B_FALSE; 1158 1159 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1160 vd->vdev_stat.vs_aux); 1161 return (error); 1162 } 1163 1164 vd->vdev_removed = B_FALSE; 1165 1166 /* 1167 * Recheck the faulted flag now that we have confirmed that 1168 * the vdev is accessible. If we're faulted, bail. 1169 */ 1170 if (vd->vdev_faulted) { 1171 ASSERT(vd->vdev_children == 0); 1172 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1173 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1174 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1175 vd->vdev_label_aux); 1176 return (ENXIO); 1177 } 1178 1179 if (vd->vdev_degraded) { 1180 ASSERT(vd->vdev_children == 0); 1181 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1182 VDEV_AUX_ERR_EXCEEDED); 1183 } else { 1184 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 1185 } 1186 1187 /* 1188 * For hole or missing vdevs we just return success. 1189 */ 1190 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 1191 return (0); 1192 1193 for (int c = 0; c < vd->vdev_children; c++) { 1194 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1195 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1196 VDEV_AUX_NONE); 1197 break; 1198 } 1199 } 1200 1201 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1202 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1203 1204 if (vd->vdev_children == 0) { 1205 if (osize < SPA_MINDEVSIZE) { 1206 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1207 VDEV_AUX_TOO_SMALL); 1208 return (EOVERFLOW); 1209 } 1210 psize = osize; 1211 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1212 max_asize = max_osize - (VDEV_LABEL_START_SIZE + 1213 VDEV_LABEL_END_SIZE); 1214 } else { 1215 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1216 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1217 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1218 VDEV_AUX_TOO_SMALL); 1219 return (EOVERFLOW); 1220 } 1221 psize = 0; 1222 asize = osize; 1223 max_asize = max_osize; 1224 } 1225 1226 vd->vdev_psize = psize; 1227 1228 /* 1229 * Make sure the allocatable size hasn't shrunk. 1230 */ 1231 if (asize < vd->vdev_min_asize) { 1232 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1233 VDEV_AUX_BAD_LABEL); 1234 return (EINVAL); 1235 } 1236 1237 if (vd->vdev_asize == 0) { 1238 /* 1239 * This is the first-ever open, so use the computed values. 1240 * For testing purposes, a higher ashift can be requested. 1241 */ 1242 vd->vdev_asize = asize; 1243 vd->vdev_max_asize = max_asize; 1244 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1245 } else { 1246 /* 1247 * Make sure the alignment requirement hasn't increased. 1248 */ 1249 if (ashift > vd->vdev_top->vdev_ashift) { 1250 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1251 VDEV_AUX_BAD_LABEL); 1252 return (EINVAL); 1253 } 1254 vd->vdev_max_asize = max_asize; 1255 } 1256 1257 /* 1258 * If all children are healthy and the asize has increased, 1259 * then we've experienced dynamic LUN growth. If automatic 1260 * expansion is enabled then use the additional space. 1261 */ 1262 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1263 (vd->vdev_expanding || spa->spa_autoexpand)) 1264 vd->vdev_asize = asize; 1265 1266 vdev_set_min_asize(vd); 1267 1268 /* 1269 * Ensure we can issue some IO before declaring the 1270 * vdev open for business. 1271 */ 1272 if (vd->vdev_ops->vdev_op_leaf && 1273 (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1274 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1275 VDEV_AUX_ERR_EXCEEDED); 1276 return (error); 1277 } 1278 1279 /* 1280 * If a leaf vdev has a DTL, and seems healthy, then kick off a 1281 * resilver. But don't do this if we are doing a reopen for a scrub, 1282 * since this would just restart the scrub we are already doing. 1283 */ 1284 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1285 vdev_resilver_needed(vd, NULL, NULL)) 1286 spa_async_request(spa, SPA_ASYNC_RESILVER); 1287 1288 return (0); 1289 } 1290 1291 /* 1292 * Called once the vdevs are all opened, this routine validates the label 1293 * contents. This needs to be done before vdev_load() so that we don't 1294 * inadvertently do repair I/Os to the wrong device. 1295 * 1296 * If 'strict' is false ignore the spa guid check. This is necessary because 1297 * if the machine crashed during a re-guid the new guid might have been written 1298 * to all of the vdev labels, but not the cached config. The strict check 1299 * will be performed when the pool is opened again using the mos config. 1300 * 1301 * This function will only return failure if one of the vdevs indicates that it 1302 * has since been destroyed or exported. This is only possible if 1303 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1304 * will be updated but the function will return 0. 1305 */ 1306 int 1307 vdev_validate(vdev_t *vd, boolean_t strict) 1308 { 1309 spa_t *spa = vd->vdev_spa; 1310 nvlist_t *label; 1311 uint64_t guid = 0, top_guid; 1312 uint64_t state; 1313 1314 for (int c = 0; c < vd->vdev_children; c++) 1315 if (vdev_validate(vd->vdev_child[c], strict) != 0) 1316 return (EBADF); 1317 1318 /* 1319 * If the device has already failed, or was marked offline, don't do 1320 * any further validation. Otherwise, label I/O will fail and we will 1321 * overwrite the previous state. 1322 */ 1323 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1324 uint64_t aux_guid = 0; 1325 nvlist_t *nvl; 1326 1327 if ((label = vdev_label_read_config(vd)) == NULL) { 1328 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1329 VDEV_AUX_BAD_LABEL); 1330 return (0); 1331 } 1332 1333 /* 1334 * Determine if this vdev has been split off into another 1335 * pool. If so, then refuse to open it. 1336 */ 1337 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1338 &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1339 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1340 VDEV_AUX_SPLIT_POOL); 1341 nvlist_free(label); 1342 return (0); 1343 } 1344 1345 if (strict && (nvlist_lookup_uint64(label, 1346 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1347 guid != spa_guid(spa))) { 1348 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1349 VDEV_AUX_CORRUPT_DATA); 1350 nvlist_free(label); 1351 return (0); 1352 } 1353 1354 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1355 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1356 &aux_guid) != 0) 1357 aux_guid = 0; 1358 1359 /* 1360 * If this vdev just became a top-level vdev because its 1361 * sibling was detached, it will have adopted the parent's 1362 * vdev guid -- but the label may or may not be on disk yet. 1363 * Fortunately, either version of the label will have the 1364 * same top guid, so if we're a top-level vdev, we can 1365 * safely compare to that instead. 1366 * 1367 * If we split this vdev off instead, then we also check the 1368 * original pool's guid. We don't want to consider the vdev 1369 * corrupt if it is partway through a split operation. 1370 */ 1371 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1372 &guid) != 0 || 1373 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1374 &top_guid) != 0 || 1375 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1376 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1377 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1378 VDEV_AUX_CORRUPT_DATA); 1379 nvlist_free(label); 1380 return (0); 1381 } 1382 1383 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1384 &state) != 0) { 1385 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1386 VDEV_AUX_CORRUPT_DATA); 1387 nvlist_free(label); 1388 return (0); 1389 } 1390 1391 nvlist_free(label); 1392 1393 /* 1394 * If this is a verbatim import, no need to check the 1395 * state of the pool. 1396 */ 1397 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1398 spa_load_state(spa) == SPA_LOAD_OPEN && 1399 state != POOL_STATE_ACTIVE) 1400 return (EBADF); 1401 1402 /* 1403 * If we were able to open and validate a vdev that was 1404 * previously marked permanently unavailable, clear that state 1405 * now. 1406 */ 1407 if (vd->vdev_not_present) 1408 vd->vdev_not_present = 0; 1409 } 1410 1411 return (0); 1412 } 1413 1414 /* 1415 * Close a virtual device. 1416 */ 1417 void 1418 vdev_close(vdev_t *vd) 1419 { 1420 spa_t *spa = vd->vdev_spa; 1421 vdev_t *pvd = vd->vdev_parent; 1422 1423 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1424 1425 /* 1426 * If our parent is reopening, then we are as well, unless we are 1427 * going offline. 1428 */ 1429 if (pvd != NULL && pvd->vdev_reopening) 1430 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1431 1432 vd->vdev_ops->vdev_op_close(vd); 1433 1434 vdev_cache_purge(vd); 1435 1436 /* 1437 * We record the previous state before we close it, so that if we are 1438 * doing a reopen(), we don't generate FMA ereports if we notice that 1439 * it's still faulted. 1440 */ 1441 vd->vdev_prevstate = vd->vdev_state; 1442 1443 if (vd->vdev_offline) 1444 vd->vdev_state = VDEV_STATE_OFFLINE; 1445 else 1446 vd->vdev_state = VDEV_STATE_CLOSED; 1447 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1448 } 1449 1450 void 1451 vdev_hold(vdev_t *vd) 1452 { 1453 spa_t *spa = vd->vdev_spa; 1454 1455 ASSERT(spa_is_root(spa)); 1456 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1457 return; 1458 1459 for (int c = 0; c < vd->vdev_children; c++) 1460 vdev_hold(vd->vdev_child[c]); 1461 1462 if (vd->vdev_ops->vdev_op_leaf) 1463 vd->vdev_ops->vdev_op_hold(vd); 1464 } 1465 1466 void 1467 vdev_rele(vdev_t *vd) 1468 { 1469 spa_t *spa = vd->vdev_spa; 1470 1471 ASSERT(spa_is_root(spa)); 1472 for (int c = 0; c < vd->vdev_children; c++) 1473 vdev_rele(vd->vdev_child[c]); 1474 1475 if (vd->vdev_ops->vdev_op_leaf) 1476 vd->vdev_ops->vdev_op_rele(vd); 1477 } 1478 1479 /* 1480 * Reopen all interior vdevs and any unopened leaves. We don't actually 1481 * reopen leaf vdevs which had previously been opened as they might deadlock 1482 * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1483 * If the leaf has never been opened then open it, as usual. 1484 */ 1485 void 1486 vdev_reopen(vdev_t *vd) 1487 { 1488 spa_t *spa = vd->vdev_spa; 1489 1490 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1491 1492 /* set the reopening flag unless we're taking the vdev offline */ 1493 vd->vdev_reopening = !vd->vdev_offline; 1494 vdev_close(vd); 1495 (void) vdev_open(vd); 1496 1497 /* 1498 * Call vdev_validate() here to make sure we have the same device. 1499 * Otherwise, a device with an invalid label could be successfully 1500 * opened in response to vdev_reopen(). 1501 */ 1502 if (vd->vdev_aux) { 1503 (void) vdev_validate_aux(vd); 1504 if (vdev_readable(vd) && vdev_writeable(vd) && 1505 vd->vdev_aux == &spa->spa_l2cache && 1506 !l2arc_vdev_present(vd)) 1507 l2arc_add_vdev(spa, vd); 1508 } else { 1509 (void) vdev_validate(vd, B_TRUE); 1510 } 1511 1512 /* 1513 * Reassess parent vdev's health. 1514 */ 1515 vdev_propagate_state(vd); 1516 } 1517 1518 int 1519 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1520 { 1521 int error; 1522 1523 /* 1524 * Normally, partial opens (e.g. of a mirror) are allowed. 1525 * For a create, however, we want to fail the request if 1526 * there are any components we can't open. 1527 */ 1528 error = vdev_open(vd); 1529 1530 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1531 vdev_close(vd); 1532 return (error ? error : ENXIO); 1533 } 1534 1535 /* 1536 * Recursively initialize all labels. 1537 */ 1538 if ((error = vdev_label_init(vd, txg, isreplacing ? 1539 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1540 vdev_close(vd); 1541 return (error); 1542 } 1543 1544 return (0); 1545 } 1546 1547 void 1548 vdev_metaslab_set_size(vdev_t *vd) 1549 { 1550 /* 1551 * Aim for roughly 200 metaslabs per vdev. 1552 */ 1553 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1554 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1555 } 1556 1557 void 1558 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1559 { 1560 ASSERT(vd == vd->vdev_top); 1561 ASSERT(!vd->vdev_ishole); 1562 ASSERT(ISP2(flags)); 1563 ASSERT(spa_writeable(vd->vdev_spa)); 1564 1565 if (flags & VDD_METASLAB) 1566 (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1567 1568 if (flags & VDD_DTL) 1569 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1570 1571 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1572 } 1573 1574 /* 1575 * DTLs. 1576 * 1577 * A vdev's DTL (dirty time log) is the set of transaction groups for which 1578 * the vdev has less than perfect replication. There are four kinds of DTL: 1579 * 1580 * DTL_MISSING: txgs for which the vdev has no valid copies of the data 1581 * 1582 * DTL_PARTIAL: txgs for which data is available, but not fully replicated 1583 * 1584 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 1585 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 1586 * txgs that was scrubbed. 1587 * 1588 * DTL_OUTAGE: txgs which cannot currently be read, whether due to 1589 * persistent errors or just some device being offline. 1590 * Unlike the other three, the DTL_OUTAGE map is not generally 1591 * maintained; it's only computed when needed, typically to 1592 * determine whether a device can be detached. 1593 * 1594 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 1595 * either has the data or it doesn't. 1596 * 1597 * For interior vdevs such as mirror and RAID-Z the picture is more complex. 1598 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 1599 * if any child is less than fully replicated, then so is its parent. 1600 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 1601 * comprising only those txgs which appear in 'maxfaults' or more children; 1602 * those are the txgs we don't have enough replication to read. For example, 1603 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 1604 * thus, its DTL_MISSING consists of the set of txgs that appear in more than 1605 * two child DTL_MISSING maps. 1606 * 1607 * It should be clear from the above that to compute the DTLs and outage maps 1608 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 1609 * Therefore, that is all we keep on disk. When loading the pool, or after 1610 * a configuration change, we generate all other DTLs from first principles. 1611 */ 1612 void 1613 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1614 { 1615 space_map_t *sm = &vd->vdev_dtl[t]; 1616 1617 ASSERT(t < DTL_TYPES); 1618 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1619 ASSERT(spa_writeable(vd->vdev_spa)); 1620 1621 mutex_enter(sm->sm_lock); 1622 if (!space_map_contains(sm, txg, size)) 1623 space_map_add(sm, txg, size); 1624 mutex_exit(sm->sm_lock); 1625 } 1626 1627 boolean_t 1628 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1629 { 1630 space_map_t *sm = &vd->vdev_dtl[t]; 1631 boolean_t dirty = B_FALSE; 1632 1633 ASSERT(t < DTL_TYPES); 1634 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1635 1636 mutex_enter(sm->sm_lock); 1637 if (sm->sm_space != 0) 1638 dirty = space_map_contains(sm, txg, size); 1639 mutex_exit(sm->sm_lock); 1640 1641 return (dirty); 1642 } 1643 1644 boolean_t 1645 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 1646 { 1647 space_map_t *sm = &vd->vdev_dtl[t]; 1648 boolean_t empty; 1649 1650 mutex_enter(sm->sm_lock); 1651 empty = (sm->sm_space == 0); 1652 mutex_exit(sm->sm_lock); 1653 1654 return (empty); 1655 } 1656 1657 /* 1658 * Reassess DTLs after a config change or scrub completion. 1659 */ 1660 void 1661 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1662 { 1663 spa_t *spa = vd->vdev_spa; 1664 avl_tree_t reftree; 1665 int minref; 1666 1667 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1668 1669 for (int c = 0; c < vd->vdev_children; c++) 1670 vdev_dtl_reassess(vd->vdev_child[c], txg, 1671 scrub_txg, scrub_done); 1672 1673 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 1674 return; 1675 1676 if (vd->vdev_ops->vdev_op_leaf) { 1677 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1678 1679 mutex_enter(&vd->vdev_dtl_lock); 1680 if (scrub_txg != 0 && 1681 (spa->spa_scrub_started || 1682 (scn && scn->scn_phys.scn_errors == 0))) { 1683 /* 1684 * We completed a scrub up to scrub_txg. If we 1685 * did it without rebooting, then the scrub dtl 1686 * will be valid, so excise the old region and 1687 * fold in the scrub dtl. Otherwise, leave the 1688 * dtl as-is if there was an error. 1689 * 1690 * There's little trick here: to excise the beginning 1691 * of the DTL_MISSING map, we put it into a reference 1692 * tree and then add a segment with refcnt -1 that 1693 * covers the range [0, scrub_txg). This means 1694 * that each txg in that range has refcnt -1 or 0. 1695 * We then add DTL_SCRUB with a refcnt of 2, so that 1696 * entries in the range [0, scrub_txg) will have a 1697 * positive refcnt -- either 1 or 2. We then convert 1698 * the reference tree into the new DTL_MISSING map. 1699 */ 1700 space_map_ref_create(&reftree); 1701 space_map_ref_add_map(&reftree, 1702 &vd->vdev_dtl[DTL_MISSING], 1); 1703 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 1704 space_map_ref_add_map(&reftree, 1705 &vd->vdev_dtl[DTL_SCRUB], 2); 1706 space_map_ref_generate_map(&reftree, 1707 &vd->vdev_dtl[DTL_MISSING], 1); 1708 space_map_ref_destroy(&reftree); 1709 } 1710 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 1711 space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1712 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1713 if (scrub_done) 1714 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 1715 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 1716 if (!vdev_readable(vd)) 1717 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 1718 else 1719 space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1720 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1721 mutex_exit(&vd->vdev_dtl_lock); 1722 1723 if (txg != 0) 1724 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1725 return; 1726 } 1727 1728 mutex_enter(&vd->vdev_dtl_lock); 1729 for (int t = 0; t < DTL_TYPES; t++) { 1730 /* account for child's outage in parent's missing map */ 1731 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 1732 if (t == DTL_SCRUB) 1733 continue; /* leaf vdevs only */ 1734 if (t == DTL_PARTIAL) 1735 minref = 1; /* i.e. non-zero */ 1736 else if (vd->vdev_nparity != 0) 1737 minref = vd->vdev_nparity + 1; /* RAID-Z */ 1738 else 1739 minref = vd->vdev_children; /* any kind of mirror */ 1740 space_map_ref_create(&reftree); 1741 for (int c = 0; c < vd->vdev_children; c++) { 1742 vdev_t *cvd = vd->vdev_child[c]; 1743 mutex_enter(&cvd->vdev_dtl_lock); 1744 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); 1745 mutex_exit(&cvd->vdev_dtl_lock); 1746 } 1747 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 1748 space_map_ref_destroy(&reftree); 1749 } 1750 mutex_exit(&vd->vdev_dtl_lock); 1751 } 1752 1753 static int 1754 vdev_dtl_load(vdev_t *vd) 1755 { 1756 spa_t *spa = vd->vdev_spa; 1757 space_map_obj_t *smo = &vd->vdev_dtl_smo; 1758 objset_t *mos = spa->spa_meta_objset; 1759 dmu_buf_t *db; 1760 int error; 1761 1762 ASSERT(vd->vdev_children == 0); 1763 1764 if (smo->smo_object == 0) 1765 return (0); 1766 1767 ASSERT(!vd->vdev_ishole); 1768 1769 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1770 return (error); 1771 1772 ASSERT3U(db->db_size, >=, sizeof (*smo)); 1773 bcopy(db->db_data, smo, sizeof (*smo)); 1774 dmu_buf_rele(db, FTAG); 1775 1776 mutex_enter(&vd->vdev_dtl_lock); 1777 error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 1778 NULL, SM_ALLOC, smo, mos); 1779 mutex_exit(&vd->vdev_dtl_lock); 1780 1781 return (error); 1782 } 1783 1784 void 1785 vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1786 { 1787 spa_t *spa = vd->vdev_spa; 1788 space_map_obj_t *smo = &vd->vdev_dtl_smo; 1789 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1790 objset_t *mos = spa->spa_meta_objset; 1791 space_map_t smsync; 1792 kmutex_t smlock; 1793 dmu_buf_t *db; 1794 dmu_tx_t *tx; 1795 1796 ASSERT(!vd->vdev_ishole); 1797 1798 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1799 1800 if (vd->vdev_detached) { 1801 if (smo->smo_object != 0) { 1802 int err = dmu_object_free(mos, smo->smo_object, tx); 1803 ASSERT3U(err, ==, 0); 1804 smo->smo_object = 0; 1805 } 1806 dmu_tx_commit(tx); 1807 return; 1808 } 1809 1810 if (smo->smo_object == 0) { 1811 ASSERT(smo->smo_objsize == 0); 1812 ASSERT(smo->smo_alloc == 0); 1813 smo->smo_object = dmu_object_alloc(mos, 1814 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1815 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1816 ASSERT(smo->smo_object != 0); 1817 vdev_config_dirty(vd->vdev_top); 1818 } 1819 1820 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1821 1822 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1823 &smlock); 1824 1825 mutex_enter(&smlock); 1826 1827 mutex_enter(&vd->vdev_dtl_lock); 1828 space_map_walk(sm, space_map_add, &smsync); 1829 mutex_exit(&vd->vdev_dtl_lock); 1830 1831 space_map_truncate(smo, mos, tx); 1832 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1833 1834 space_map_destroy(&smsync); 1835 1836 mutex_exit(&smlock); 1837 mutex_destroy(&smlock); 1838 1839 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1840 dmu_buf_will_dirty(db, tx); 1841 ASSERT3U(db->db_size, >=, sizeof (*smo)); 1842 bcopy(smo, db->db_data, sizeof (*smo)); 1843 dmu_buf_rele(db, FTAG); 1844 1845 dmu_tx_commit(tx); 1846 } 1847 1848 /* 1849 * Determine whether the specified vdev can be offlined/detached/removed 1850 * without losing data. 1851 */ 1852 boolean_t 1853 vdev_dtl_required(vdev_t *vd) 1854 { 1855 spa_t *spa = vd->vdev_spa; 1856 vdev_t *tvd = vd->vdev_top; 1857 uint8_t cant_read = vd->vdev_cant_read; 1858 boolean_t required; 1859 1860 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1861 1862 if (vd == spa->spa_root_vdev || vd == tvd) 1863 return (B_TRUE); 1864 1865 /* 1866 * Temporarily mark the device as unreadable, and then determine 1867 * whether this results in any DTL outages in the top-level vdev. 1868 * If not, we can safely offline/detach/remove the device. 1869 */ 1870 vd->vdev_cant_read = B_TRUE; 1871 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1872 required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 1873 vd->vdev_cant_read = cant_read; 1874 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1875 1876 if (!required && zio_injection_enabled) 1877 required = !!zio_handle_device_injection(vd, NULL, ECHILD); 1878 1879 return (required); 1880 } 1881 1882 /* 1883 * Determine if resilver is needed, and if so the txg range. 1884 */ 1885 boolean_t 1886 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1887 { 1888 boolean_t needed = B_FALSE; 1889 uint64_t thismin = UINT64_MAX; 1890 uint64_t thismax = 0; 1891 1892 if (vd->vdev_children == 0) { 1893 mutex_enter(&vd->vdev_dtl_lock); 1894 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 1895 vdev_writeable(vd)) { 1896 space_seg_t *ss; 1897 1898 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1899 thismin = ss->ss_start - 1; 1900 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1901 thismax = ss->ss_end; 1902 needed = B_TRUE; 1903 } 1904 mutex_exit(&vd->vdev_dtl_lock); 1905 } else { 1906 for (int c = 0; c < vd->vdev_children; c++) { 1907 vdev_t *cvd = vd->vdev_child[c]; 1908 uint64_t cmin, cmax; 1909 1910 if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1911 thismin = MIN(thismin, cmin); 1912 thismax = MAX(thismax, cmax); 1913 needed = B_TRUE; 1914 } 1915 } 1916 } 1917 1918 if (needed && minp) { 1919 *minp = thismin; 1920 *maxp = thismax; 1921 } 1922 return (needed); 1923 } 1924 1925 void 1926 vdev_load(vdev_t *vd) 1927 { 1928 /* 1929 * Recursively load all children. 1930 */ 1931 for (int c = 0; c < vd->vdev_children; c++) 1932 vdev_load(vd->vdev_child[c]); 1933 1934 /* 1935 * If this is a top-level vdev, initialize its metaslabs. 1936 */ 1937 if (vd == vd->vdev_top && !vd->vdev_ishole && 1938 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1939 vdev_metaslab_init(vd, 0) != 0)) 1940 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1941 VDEV_AUX_CORRUPT_DATA); 1942 1943 /* 1944 * If this is a leaf vdev, load its DTL. 1945 */ 1946 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1947 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1948 VDEV_AUX_CORRUPT_DATA); 1949 } 1950 1951 /* 1952 * The special vdev case is used for hot spares and l2cache devices. Its 1953 * sole purpose it to set the vdev state for the associated vdev. To do this, 1954 * we make sure that we can open the underlying device, then try to read the 1955 * label, and make sure that the label is sane and that it hasn't been 1956 * repurposed to another pool. 1957 */ 1958 int 1959 vdev_validate_aux(vdev_t *vd) 1960 { 1961 nvlist_t *label; 1962 uint64_t guid, version; 1963 uint64_t state; 1964 1965 if (!vdev_readable(vd)) 1966 return (0); 1967 1968 if ((label = vdev_label_read_config(vd)) == NULL) { 1969 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1970 VDEV_AUX_CORRUPT_DATA); 1971 return (-1); 1972 } 1973 1974 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1975 version > SPA_VERSION || 1976 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 1977 guid != vd->vdev_guid || 1978 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 1979 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1980 VDEV_AUX_CORRUPT_DATA); 1981 nvlist_free(label); 1982 return (-1); 1983 } 1984 1985 /* 1986 * We don't actually check the pool state here. If it's in fact in 1987 * use by another pool, we update this fact on the fly when requested. 1988 */ 1989 nvlist_free(label); 1990 return (0); 1991 } 1992 1993 void 1994 vdev_remove(vdev_t *vd, uint64_t txg) 1995 { 1996 spa_t *spa = vd->vdev_spa; 1997 objset_t *mos = spa->spa_meta_objset; 1998 dmu_tx_t *tx; 1999 2000 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 2001 2002 if (vd->vdev_dtl_smo.smo_object) { 2003 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0); 2004 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); 2005 vd->vdev_dtl_smo.smo_object = 0; 2006 } 2007 2008 if (vd->vdev_ms != NULL) { 2009 for (int m = 0; m < vd->vdev_ms_count; m++) { 2010 metaslab_t *msp = vd->vdev_ms[m]; 2011 2012 if (msp == NULL || msp->ms_smo.smo_object == 0) 2013 continue; 2014 2015 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0); 2016 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); 2017 msp->ms_smo.smo_object = 0; 2018 } 2019 } 2020 2021 if (vd->vdev_ms_array) { 2022 (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 2023 vd->vdev_ms_array = 0; 2024 vd->vdev_ms_shift = 0; 2025 } 2026 dmu_tx_commit(tx); 2027 } 2028 2029 void 2030 vdev_sync_done(vdev_t *vd, uint64_t txg) 2031 { 2032 metaslab_t *msp; 2033 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2034 2035 ASSERT(!vd->vdev_ishole); 2036 2037 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2038 metaslab_sync_done(msp, txg); 2039 2040 if (reassess) 2041 metaslab_sync_reassess(vd->vdev_mg); 2042 } 2043 2044 void 2045 vdev_sync(vdev_t *vd, uint64_t txg) 2046 { 2047 spa_t *spa = vd->vdev_spa; 2048 vdev_t *lvd; 2049 metaslab_t *msp; 2050 dmu_tx_t *tx; 2051 2052 ASSERT(!vd->vdev_ishole); 2053 2054 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2055 ASSERT(vd == vd->vdev_top); 2056 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2057 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2058 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2059 ASSERT(vd->vdev_ms_array != 0); 2060 vdev_config_dirty(vd); 2061 dmu_tx_commit(tx); 2062 } 2063 2064 /* 2065 * Remove the metadata associated with this vdev once it's empty. 2066 */ 2067 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 2068 vdev_remove(vd, txg); 2069 2070 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2071 metaslab_sync(msp, txg); 2072 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2073 } 2074 2075 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2076 vdev_dtl_sync(lvd, txg); 2077 2078 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2079 } 2080 2081 uint64_t 2082 vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2083 { 2084 return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2085 } 2086 2087 /* 2088 * Mark the given vdev faulted. A faulted vdev behaves as if the device could 2089 * not be opened, and no I/O is attempted. 2090 */ 2091 int 2092 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2093 { 2094 vdev_t *vd, *tvd; 2095 2096 spa_vdev_state_enter(spa, SCL_NONE); 2097 2098 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2099 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2100 2101 if (!vd->vdev_ops->vdev_op_leaf) 2102 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2103 2104 tvd = vd->vdev_top; 2105 2106 /* 2107 * We don't directly use the aux state here, but if we do a 2108 * vdev_reopen(), we need this value to be present to remember why we 2109 * were faulted. 2110 */ 2111 vd->vdev_label_aux = aux; 2112 2113 /* 2114 * Faulted state takes precedence over degraded. 2115 */ 2116 vd->vdev_delayed_close = B_FALSE; 2117 vd->vdev_faulted = 1ULL; 2118 vd->vdev_degraded = 0ULL; 2119 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 2120 2121 /* 2122 * If this device has the only valid copy of the data, then 2123 * back off and simply mark the vdev as degraded instead. 2124 */ 2125 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 2126 vd->vdev_degraded = 1ULL; 2127 vd->vdev_faulted = 0ULL; 2128 2129 /* 2130 * If we reopen the device and it's not dead, only then do we 2131 * mark it degraded. 2132 */ 2133 vdev_reopen(tvd); 2134 2135 if (vdev_readable(vd)) 2136 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 2137 } 2138 2139 return (spa_vdev_state_exit(spa, vd, 0)); 2140 } 2141 2142 /* 2143 * Mark the given vdev degraded. A degraded vdev is purely an indication to the 2144 * user that something is wrong. The vdev continues to operate as normal as far 2145 * as I/O is concerned. 2146 */ 2147 int 2148 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2149 { 2150 vdev_t *vd; 2151 2152 spa_vdev_state_enter(spa, SCL_NONE); 2153 2154 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2155 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2156 2157 if (!vd->vdev_ops->vdev_op_leaf) 2158 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2159 2160 /* 2161 * If the vdev is already faulted, then don't do anything. 2162 */ 2163 if (vd->vdev_faulted || vd->vdev_degraded) 2164 return (spa_vdev_state_exit(spa, NULL, 0)); 2165 2166 vd->vdev_degraded = 1ULL; 2167 if (!vdev_is_dead(vd)) 2168 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2169 aux); 2170 2171 return (spa_vdev_state_exit(spa, vd, 0)); 2172 } 2173 2174 /* 2175 * Online the given vdev. If 'unspare' is set, it implies two things. First, 2176 * any attached spare device should be detached when the device finishes 2177 * resilvering. Second, the online should be treated like a 'test' online case, 2178 * so no FMA events are generated if the device fails to open. 2179 */ 2180 int 2181 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 2182 { 2183 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 2184 2185 spa_vdev_state_enter(spa, SCL_NONE); 2186 2187 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2188 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2189 2190 if (!vd->vdev_ops->vdev_op_leaf) 2191 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2192 2193 tvd = vd->vdev_top; 2194 vd->vdev_offline = B_FALSE; 2195 vd->vdev_tmpoffline = B_FALSE; 2196 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2197 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2198 2199 /* XXX - L2ARC 1.0 does not support expansion */ 2200 if (!vd->vdev_aux) { 2201 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2202 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2203 } 2204 2205 vdev_reopen(tvd); 2206 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 2207 2208 if (!vd->vdev_aux) { 2209 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2210 pvd->vdev_expanding = B_FALSE; 2211 } 2212 2213 if (newstate) 2214 *newstate = vd->vdev_state; 2215 if ((flags & ZFS_ONLINE_UNSPARE) && 2216 !vdev_is_dead(vd) && vd->vdev_parent && 2217 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2218 vd->vdev_parent->vdev_child[0] == vd) 2219 vd->vdev_unspare = B_TRUE; 2220 2221 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2222 2223 /* XXX - L2ARC 1.0 does not support expansion */ 2224 if (vd->vdev_aux) 2225 return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2226 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2227 } 2228 return (spa_vdev_state_exit(spa, vd, 0)); 2229 } 2230 2231 static int 2232 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2233 { 2234 vdev_t *vd, *tvd; 2235 int error = 0; 2236 uint64_t generation; 2237 metaslab_group_t *mg; 2238 2239 top: 2240 spa_vdev_state_enter(spa, SCL_ALLOC); 2241 2242 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2243 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2244 2245 if (!vd->vdev_ops->vdev_op_leaf) 2246 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2247 2248 tvd = vd->vdev_top; 2249 mg = tvd->vdev_mg; 2250 generation = spa->spa_config_generation + 1; 2251 2252 /* 2253 * If the device isn't already offline, try to offline it. 2254 */ 2255 if (!vd->vdev_offline) { 2256 /* 2257 * If this device has the only valid copy of some data, 2258 * don't allow it to be offlined. Log devices are always 2259 * expendable. 2260 */ 2261 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2262 vdev_dtl_required(vd)) 2263 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2264 2265 /* 2266 * If the top-level is a slog and it has had allocations 2267 * then proceed. We check that the vdev's metaslab group 2268 * is not NULL since it's possible that we may have just 2269 * added this vdev but not yet initialized its metaslabs. 2270 */ 2271 if (tvd->vdev_islog && mg != NULL) { 2272 /* 2273 * Prevent any future allocations. 2274 */ 2275 metaslab_group_passivate(mg); 2276 (void) spa_vdev_state_exit(spa, vd, 0); 2277 2278 error = spa_offline_log(spa); 2279 2280 spa_vdev_state_enter(spa, SCL_ALLOC); 2281 2282 /* 2283 * Check to see if the config has changed. 2284 */ 2285 if (error || generation != spa->spa_config_generation) { 2286 metaslab_group_activate(mg); 2287 if (error) 2288 return (spa_vdev_state_exit(spa, 2289 vd, error)); 2290 (void) spa_vdev_state_exit(spa, vd, 0); 2291 goto top; 2292 } 2293 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0); 2294 } 2295 2296 /* 2297 * Offline this device and reopen its top-level vdev. 2298 * If the top-level vdev is a log device then just offline 2299 * it. Otherwise, if this action results in the top-level 2300 * vdev becoming unusable, undo it and fail the request. 2301 */ 2302 vd->vdev_offline = B_TRUE; 2303 vdev_reopen(tvd); 2304 2305 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2306 vdev_is_dead(tvd)) { 2307 vd->vdev_offline = B_FALSE; 2308 vdev_reopen(tvd); 2309 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2310 } 2311 2312 /* 2313 * Add the device back into the metaslab rotor so that 2314 * once we online the device it's open for business. 2315 */ 2316 if (tvd->vdev_islog && mg != NULL) 2317 metaslab_group_activate(mg); 2318 } 2319 2320 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2321 2322 return (spa_vdev_state_exit(spa, vd, 0)); 2323 } 2324 2325 int 2326 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2327 { 2328 int error; 2329 2330 mutex_enter(&spa->spa_vdev_top_lock); 2331 error = vdev_offline_locked(spa, guid, flags); 2332 mutex_exit(&spa->spa_vdev_top_lock); 2333 2334 return (error); 2335 } 2336 2337 /* 2338 * Clear the error counts associated with this vdev. Unlike vdev_online() and 2339 * vdev_offline(), we assume the spa config is locked. We also clear all 2340 * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2341 */ 2342 void 2343 vdev_clear(spa_t *spa, vdev_t *vd) 2344 { 2345 vdev_t *rvd = spa->spa_root_vdev; 2346 2347 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2348 2349 if (vd == NULL) 2350 vd = rvd; 2351 2352 vd->vdev_stat.vs_read_errors = 0; 2353 vd->vdev_stat.vs_write_errors = 0; 2354 vd->vdev_stat.vs_checksum_errors = 0; 2355 2356 for (int c = 0; c < vd->vdev_children; c++) 2357 vdev_clear(spa, vd->vdev_child[c]); 2358 2359 /* 2360 * If we're in the FAULTED state or have experienced failed I/O, then 2361 * clear the persistent state and attempt to reopen the device. We 2362 * also mark the vdev config dirty, so that the new faulted state is 2363 * written out to disk. 2364 */ 2365 if (vd->vdev_faulted || vd->vdev_degraded || 2366 !vdev_readable(vd) || !vdev_writeable(vd)) { 2367 2368 /* 2369 * When reopening in reponse to a clear event, it may be due to 2370 * a fmadm repair request. In this case, if the device is 2371 * still broken, we want to still post the ereport again. 2372 */ 2373 vd->vdev_forcefault = B_TRUE; 2374 2375 vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2376 vd->vdev_cant_read = B_FALSE; 2377 vd->vdev_cant_write = B_FALSE; 2378 2379 vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 2380 2381 vd->vdev_forcefault = B_FALSE; 2382 2383 if (vd != rvd && vdev_writeable(vd->vdev_top)) 2384 vdev_state_dirty(vd->vdev_top); 2385 2386 if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2387 spa_async_request(spa, SPA_ASYNC_RESILVER); 2388 2389 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 2390 } 2391 2392 /* 2393 * When clearing a FMA-diagnosed fault, we always want to 2394 * unspare the device, as we assume that the original spare was 2395 * done in response to the FMA fault. 2396 */ 2397 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2398 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2399 vd->vdev_parent->vdev_child[0] == vd) 2400 vd->vdev_unspare = B_TRUE; 2401 } 2402 2403 boolean_t 2404 vdev_is_dead(vdev_t *vd) 2405 { 2406 /* 2407 * Holes and missing devices are always considered "dead". 2408 * This simplifies the code since we don't have to check for 2409 * these types of devices in the various code paths. 2410 * Instead we rely on the fact that we skip over dead devices 2411 * before issuing I/O to them. 2412 */ 2413 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 2414 vd->vdev_ops == &vdev_missing_ops); 2415 } 2416 2417 boolean_t 2418 vdev_readable(vdev_t *vd) 2419 { 2420 return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 2421 } 2422 2423 boolean_t 2424 vdev_writeable(vdev_t *vd) 2425 { 2426 return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2427 } 2428 2429 boolean_t 2430 vdev_allocatable(vdev_t *vd) 2431 { 2432 uint64_t state = vd->vdev_state; 2433 2434 /* 2435 * We currently allow allocations from vdevs which may be in the 2436 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2437 * fails to reopen then we'll catch it later when we're holding 2438 * the proper locks. Note that we have to get the vdev state 2439 * in a local variable because although it changes atomically, 2440 * we're asking two separate questions about it. 2441 */ 2442 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 2443 !vd->vdev_cant_write && !vd->vdev_ishole); 2444 } 2445 2446 boolean_t 2447 vdev_accessible(vdev_t *vd, zio_t *zio) 2448 { 2449 ASSERT(zio->io_vd == vd); 2450 2451 if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2452 return (B_FALSE); 2453 2454 if (zio->io_type == ZIO_TYPE_READ) 2455 return (!vd->vdev_cant_read); 2456 2457 if (zio->io_type == ZIO_TYPE_WRITE) 2458 return (!vd->vdev_cant_write); 2459 2460 return (B_TRUE); 2461 } 2462 2463 /* 2464 * Get statistics for the given vdev. 2465 */ 2466 void 2467 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2468 { 2469 vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2470 2471 mutex_enter(&vd->vdev_stat_lock); 2472 bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2473 vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2474 vs->vs_state = vd->vdev_state; 2475 vs->vs_rsize = vdev_get_min_asize(vd); 2476 if (vd->vdev_ops->vdev_op_leaf) 2477 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2478 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; 2479 mutex_exit(&vd->vdev_stat_lock); 2480 2481 /* 2482 * If we're getting stats on the root vdev, aggregate the I/O counts 2483 * over all top-level vdevs (i.e. the direct children of the root). 2484 */ 2485 if (vd == rvd) { 2486 for (int c = 0; c < rvd->vdev_children; c++) { 2487 vdev_t *cvd = rvd->vdev_child[c]; 2488 vdev_stat_t *cvs = &cvd->vdev_stat; 2489 2490 mutex_enter(&vd->vdev_stat_lock); 2491 for (int t = 0; t < ZIO_TYPES; t++) { 2492 vs->vs_ops[t] += cvs->vs_ops[t]; 2493 vs->vs_bytes[t] += cvs->vs_bytes[t]; 2494 } 2495 cvs->vs_scan_removing = cvd->vdev_removing; 2496 mutex_exit(&vd->vdev_stat_lock); 2497 } 2498 } 2499 } 2500 2501 void 2502 vdev_clear_stats(vdev_t *vd) 2503 { 2504 mutex_enter(&vd->vdev_stat_lock); 2505 vd->vdev_stat.vs_space = 0; 2506 vd->vdev_stat.vs_dspace = 0; 2507 vd->vdev_stat.vs_alloc = 0; 2508 mutex_exit(&vd->vdev_stat_lock); 2509 } 2510 2511 void 2512 vdev_scan_stat_init(vdev_t *vd) 2513 { 2514 vdev_stat_t *vs = &vd->vdev_stat; 2515 2516 for (int c = 0; c < vd->vdev_children; c++) 2517 vdev_scan_stat_init(vd->vdev_child[c]); 2518 2519 mutex_enter(&vd->vdev_stat_lock); 2520 vs->vs_scan_processed = 0; 2521 mutex_exit(&vd->vdev_stat_lock); 2522 } 2523 2524 void 2525 vdev_stat_update(zio_t *zio, uint64_t psize) 2526 { 2527 spa_t *spa = zio->io_spa; 2528 vdev_t *rvd = spa->spa_root_vdev; 2529 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2530 vdev_t *pvd; 2531 uint64_t txg = zio->io_txg; 2532 vdev_stat_t *vs = &vd->vdev_stat; 2533 zio_type_t type = zio->io_type; 2534 int flags = zio->io_flags; 2535 2536 /* 2537 * If this i/o is a gang leader, it didn't do any actual work. 2538 */ 2539 if (zio->io_gang_tree) 2540 return; 2541 2542 if (zio->io_error == 0) { 2543 /* 2544 * If this is a root i/o, don't count it -- we've already 2545 * counted the top-level vdevs, and vdev_get_stats() will 2546 * aggregate them when asked. This reduces contention on 2547 * the root vdev_stat_lock and implicitly handles blocks 2548 * that compress away to holes, for which there is no i/o. 2549 * (Holes never create vdev children, so all the counters 2550 * remain zero, which is what we want.) 2551 * 2552 * Note: this only applies to successful i/o (io_error == 0) 2553 * because unlike i/o counts, errors are not additive. 2554 * When reading a ditto block, for example, failure of 2555 * one top-level vdev does not imply a root-level error. 2556 */ 2557 if (vd == rvd) 2558 return; 2559 2560 ASSERT(vd == zio->io_vd); 2561 2562 if (flags & ZIO_FLAG_IO_BYPASS) 2563 return; 2564 2565 mutex_enter(&vd->vdev_stat_lock); 2566 2567 if (flags & ZIO_FLAG_IO_REPAIR) { 2568 if (flags & ZIO_FLAG_SCAN_THREAD) { 2569 dsl_scan_phys_t *scn_phys = 2570 &spa->spa_dsl_pool->dp_scan->scn_phys; 2571 uint64_t *processed = &scn_phys->scn_processed; 2572 2573 /* XXX cleanup? */ 2574 if (vd->vdev_ops->vdev_op_leaf) 2575 atomic_add_64(processed, psize); 2576 vs->vs_scan_processed += psize; 2577 } 2578 2579 if (flags & ZIO_FLAG_SELF_HEAL) 2580 vs->vs_self_healed += psize; 2581 } 2582 2583 vs->vs_ops[type]++; 2584 vs->vs_bytes[type] += psize; 2585 2586 mutex_exit(&vd->vdev_stat_lock); 2587 return; 2588 } 2589 2590 if (flags & ZIO_FLAG_SPECULATIVE) 2591 return; 2592 2593 /* 2594 * If this is an I/O error that is going to be retried, then ignore the 2595 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 2596 * hard errors, when in reality they can happen for any number of 2597 * innocuous reasons (bus resets, MPxIO link failure, etc). 2598 */ 2599 if (zio->io_error == EIO && 2600 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 2601 return; 2602 2603 /* 2604 * Intent logs writes won't propagate their error to the root 2605 * I/O so don't mark these types of failures as pool-level 2606 * errors. 2607 */ 2608 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 2609 return; 2610 2611 mutex_enter(&vd->vdev_stat_lock); 2612 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2613 if (zio->io_error == ECKSUM) 2614 vs->vs_checksum_errors++; 2615 else 2616 vs->vs_read_errors++; 2617 } 2618 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2619 vs->vs_write_errors++; 2620 mutex_exit(&vd->vdev_stat_lock); 2621 2622 if (type == ZIO_TYPE_WRITE && txg != 0 && 2623 (!(flags & ZIO_FLAG_IO_REPAIR) || 2624 (flags & ZIO_FLAG_SCAN_THREAD) || 2625 spa->spa_claiming)) { 2626 /* 2627 * This is either a normal write (not a repair), or it's 2628 * a repair induced by the scrub thread, or it's a repair 2629 * made by zil_claim() during spa_load() in the first txg. 2630 * In the normal case, we commit the DTL change in the same 2631 * txg as the block was born. In the scrub-induced repair 2632 * case, we know that scrubs run in first-pass syncing context, 2633 * so we commit the DTL change in spa_syncing_txg(spa). 2634 * In the zil_claim() case, we commit in spa_first_txg(spa). 2635 * 2636 * We currently do not make DTL entries for failed spontaneous 2637 * self-healing writes triggered by normal (non-scrubbing) 2638 * reads, because we have no transactional context in which to 2639 * do so -- and it's not clear that it'd be desirable anyway. 2640 */ 2641 if (vd->vdev_ops->vdev_op_leaf) { 2642 uint64_t commit_txg = txg; 2643 if (flags & ZIO_FLAG_SCAN_THREAD) { 2644 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2645 ASSERT(spa_sync_pass(spa) == 1); 2646 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2647 commit_txg = spa_syncing_txg(spa); 2648 } else if (spa->spa_claiming) { 2649 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2650 commit_txg = spa_first_txg(spa); 2651 } 2652 ASSERT(commit_txg >= spa_syncing_txg(spa)); 2653 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2654 return; 2655 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2656 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 2657 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2658 } 2659 if (vd != rvd) 2660 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2661 } 2662 } 2663 2664 /* 2665 * Update the in-core space usage stats for this vdev, its metaslab class, 2666 * and the root vdev. 2667 */ 2668 void 2669 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2670 int64_t space_delta) 2671 { 2672 int64_t dspace_delta = space_delta; 2673 spa_t *spa = vd->vdev_spa; 2674 vdev_t *rvd = spa->spa_root_vdev; 2675 metaslab_group_t *mg = vd->vdev_mg; 2676 metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2677 2678 ASSERT(vd == vd->vdev_top); 2679 2680 /* 2681 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 2682 * factor. We must calculate this here and not at the root vdev 2683 * because the root vdev's psize-to-asize is simply the max of its 2684 * childrens', thus not accurate enough for us. 2685 */ 2686 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2687 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 2688 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 2689 vd->vdev_deflate_ratio; 2690 2691 mutex_enter(&vd->vdev_stat_lock); 2692 vd->vdev_stat.vs_alloc += alloc_delta; 2693 vd->vdev_stat.vs_space += space_delta; 2694 vd->vdev_stat.vs_dspace += dspace_delta; 2695 mutex_exit(&vd->vdev_stat_lock); 2696 2697 if (mc == spa_normal_class(spa)) { 2698 mutex_enter(&rvd->vdev_stat_lock); 2699 rvd->vdev_stat.vs_alloc += alloc_delta; 2700 rvd->vdev_stat.vs_space += space_delta; 2701 rvd->vdev_stat.vs_dspace += dspace_delta; 2702 mutex_exit(&rvd->vdev_stat_lock); 2703 } 2704 2705 if (mc != NULL) { 2706 ASSERT(rvd == vd->vdev_parent); 2707 ASSERT(vd->vdev_ms_count != 0); 2708 2709 metaslab_class_space_update(mc, 2710 alloc_delta, defer_delta, space_delta, dspace_delta); 2711 } 2712 } 2713 2714 /* 2715 * Mark a top-level vdev's config as dirty, placing it on the dirty list 2716 * so that it will be written out next time the vdev configuration is synced. 2717 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2718 */ 2719 void 2720 vdev_config_dirty(vdev_t *vd) 2721 { 2722 spa_t *spa = vd->vdev_spa; 2723 vdev_t *rvd = spa->spa_root_vdev; 2724 int c; 2725 2726 ASSERT(spa_writeable(spa)); 2727 2728 /* 2729 * If this is an aux vdev (as with l2cache and spare devices), then we 2730 * update the vdev config manually and set the sync flag. 2731 */ 2732 if (vd->vdev_aux != NULL) { 2733 spa_aux_vdev_t *sav = vd->vdev_aux; 2734 nvlist_t **aux; 2735 uint_t naux; 2736 2737 for (c = 0; c < sav->sav_count; c++) { 2738 if (sav->sav_vdevs[c] == vd) 2739 break; 2740 } 2741 2742 if (c == sav->sav_count) { 2743 /* 2744 * We're being removed. There's nothing more to do. 2745 */ 2746 ASSERT(sav->sav_sync == B_TRUE); 2747 return; 2748 } 2749 2750 sav->sav_sync = B_TRUE; 2751 2752 if (nvlist_lookup_nvlist_array(sav->sav_config, 2753 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 2754 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 2755 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 2756 } 2757 2758 ASSERT(c < naux); 2759 2760 /* 2761 * Setting the nvlist in the middle if the array is a little 2762 * sketchy, but it will work. 2763 */ 2764 nvlist_free(aux[c]); 2765 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 2766 2767 return; 2768 } 2769 2770 /* 2771 * The dirty list is protected by the SCL_CONFIG lock. The caller 2772 * must either hold SCL_CONFIG as writer, or must be the sync thread 2773 * (which holds SCL_CONFIG as reader). There's only one sync thread, 2774 * so this is sufficient to ensure mutual exclusion. 2775 */ 2776 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2777 (dsl_pool_sync_context(spa_get_dsl(spa)) && 2778 spa_config_held(spa, SCL_CONFIG, RW_READER))); 2779 2780 if (vd == rvd) { 2781 for (c = 0; c < rvd->vdev_children; c++) 2782 vdev_config_dirty(rvd->vdev_child[c]); 2783 } else { 2784 ASSERT(vd == vd->vdev_top); 2785 2786 if (!list_link_active(&vd->vdev_config_dirty_node) && 2787 !vd->vdev_ishole) 2788 list_insert_head(&spa->spa_config_dirty_list, vd); 2789 } 2790 } 2791 2792 void 2793 vdev_config_clean(vdev_t *vd) 2794 { 2795 spa_t *spa = vd->vdev_spa; 2796 2797 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2798 (dsl_pool_sync_context(spa_get_dsl(spa)) && 2799 spa_config_held(spa, SCL_CONFIG, RW_READER))); 2800 2801 ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2802 list_remove(&spa->spa_config_dirty_list, vd); 2803 } 2804 2805 /* 2806 * Mark a top-level vdev's state as dirty, so that the next pass of 2807 * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2808 * the state changes from larger config changes because they require 2809 * much less locking, and are often needed for administrative actions. 2810 */ 2811 void 2812 vdev_state_dirty(vdev_t *vd) 2813 { 2814 spa_t *spa = vd->vdev_spa; 2815 2816 ASSERT(spa_writeable(spa)); 2817 ASSERT(vd == vd->vdev_top); 2818 2819 /* 2820 * The state list is protected by the SCL_STATE lock. The caller 2821 * must either hold SCL_STATE as writer, or must be the sync thread 2822 * (which holds SCL_STATE as reader). There's only one sync thread, 2823 * so this is sufficient to ensure mutual exclusion. 2824 */ 2825 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2826 (dsl_pool_sync_context(spa_get_dsl(spa)) && 2827 spa_config_held(spa, SCL_STATE, RW_READER))); 2828 2829 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 2830 list_insert_head(&spa->spa_state_dirty_list, vd); 2831 } 2832 2833 void 2834 vdev_state_clean(vdev_t *vd) 2835 { 2836 spa_t *spa = vd->vdev_spa; 2837 2838 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2839 (dsl_pool_sync_context(spa_get_dsl(spa)) && 2840 spa_config_held(spa, SCL_STATE, RW_READER))); 2841 2842 ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2843 list_remove(&spa->spa_state_dirty_list, vd); 2844 } 2845 2846 /* 2847 * Propagate vdev state up from children to parent. 2848 */ 2849 void 2850 vdev_propagate_state(vdev_t *vd) 2851 { 2852 spa_t *spa = vd->vdev_spa; 2853 vdev_t *rvd = spa->spa_root_vdev; 2854 int degraded = 0, faulted = 0; 2855 int corrupted = 0; 2856 vdev_t *child; 2857 2858 if (vd->vdev_children > 0) { 2859 for (int c = 0; c < vd->vdev_children; c++) { 2860 child = vd->vdev_child[c]; 2861 2862 /* 2863 * Don't factor holes into the decision. 2864 */ 2865 if (child->vdev_ishole) 2866 continue; 2867 2868 if (!vdev_readable(child) || 2869 (!vdev_writeable(child) && spa_writeable(spa))) { 2870 /* 2871 * Root special: if there is a top-level log 2872 * device, treat the root vdev as if it were 2873 * degraded. 2874 */ 2875 if (child->vdev_islog && vd == rvd) 2876 degraded++; 2877 else 2878 faulted++; 2879 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 2880 degraded++; 2881 } 2882 2883 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 2884 corrupted++; 2885 } 2886 2887 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 2888 2889 /* 2890 * Root special: if there is a top-level vdev that cannot be 2891 * opened due to corrupted metadata, then propagate the root 2892 * vdev's aux state as 'corrupt' rather than 'insufficient 2893 * replicas'. 2894 */ 2895 if (corrupted && vd == rvd && 2896 rvd->vdev_state == VDEV_STATE_CANT_OPEN) 2897 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 2898 VDEV_AUX_CORRUPT_DATA); 2899 } 2900 2901 if (vd->vdev_parent) 2902 vdev_propagate_state(vd->vdev_parent); 2903 } 2904 2905 /* 2906 * Set a vdev's state. If this is during an open, we don't update the parent 2907 * state, because we're in the process of opening children depth-first. 2908 * Otherwise, we propagate the change to the parent. 2909 * 2910 * If this routine places a device in a faulted state, an appropriate ereport is 2911 * generated. 2912 */ 2913 void 2914 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2915 { 2916 uint64_t save_state; 2917 spa_t *spa = vd->vdev_spa; 2918 2919 if (state == vd->vdev_state) { 2920 vd->vdev_stat.vs_aux = aux; 2921 return; 2922 } 2923 2924 save_state = vd->vdev_state; 2925 2926 vd->vdev_state = state; 2927 vd->vdev_stat.vs_aux = aux; 2928 2929 /* 2930 * If we are setting the vdev state to anything but an open state, then 2931 * always close the underlying device unless the device has requested 2932 * a delayed close (i.e. we're about to remove or fault the device). 2933 * Otherwise, we keep accessible but invalid devices open forever. 2934 * We don't call vdev_close() itself, because that implies some extra 2935 * checks (offline, etc) that we don't want here. This is limited to 2936 * leaf devices, because otherwise closing the device will affect other 2937 * children. 2938 */ 2939 if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 2940 vd->vdev_ops->vdev_op_leaf) 2941 vd->vdev_ops->vdev_op_close(vd); 2942 2943 /* 2944 * If we have brought this vdev back into service, we need 2945 * to notify fmd so that it can gracefully repair any outstanding 2946 * cases due to a missing device. We do this in all cases, even those 2947 * that probably don't correlate to a repaired fault. This is sure to 2948 * catch all cases, and we let the zfs-retire agent sort it out. If 2949 * this is a transient state it's OK, as the retire agent will 2950 * double-check the state of the vdev before repairing it. 2951 */ 2952 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 2953 vd->vdev_prevstate != state) 2954 zfs_post_state_change(spa, vd); 2955 2956 if (vd->vdev_removed && 2957 state == VDEV_STATE_CANT_OPEN && 2958 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 2959 /* 2960 * If the previous state is set to VDEV_STATE_REMOVED, then this 2961 * device was previously marked removed and someone attempted to 2962 * reopen it. If this failed due to a nonexistent device, then 2963 * keep the device in the REMOVED state. We also let this be if 2964 * it is one of our special test online cases, which is only 2965 * attempting to online the device and shouldn't generate an FMA 2966 * fault. 2967 */ 2968 vd->vdev_state = VDEV_STATE_REMOVED; 2969 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2970 } else if (state == VDEV_STATE_REMOVED) { 2971 vd->vdev_removed = B_TRUE; 2972 } else if (state == VDEV_STATE_CANT_OPEN) { 2973 /* 2974 * If we fail to open a vdev during an import or recovery, we 2975 * mark it as "not available", which signifies that it was 2976 * never there to begin with. Failure to open such a device 2977 * is not considered an error. 2978 */ 2979 if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 2980 spa_load_state(spa) == SPA_LOAD_RECOVER) && 2981 vd->vdev_ops->vdev_op_leaf) 2982 vd->vdev_not_present = 1; 2983 2984 /* 2985 * Post the appropriate ereport. If the 'prevstate' field is 2986 * set to something other than VDEV_STATE_UNKNOWN, it indicates 2987 * that this is part of a vdev_reopen(). In this case, we don't 2988 * want to post the ereport if the device was already in the 2989 * CANT_OPEN state beforehand. 2990 * 2991 * If the 'checkremove' flag is set, then this is an attempt to 2992 * online the device in response to an insertion event. If we 2993 * hit this case, then we have detected an insertion event for a 2994 * faulted or offline device that wasn't in the removed state. 2995 * In this scenario, we don't post an ereport because we are 2996 * about to replace the device, or attempt an online with 2997 * vdev_forcefault, which will generate the fault for us. 2998 */ 2999 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 3000 !vd->vdev_not_present && !vd->vdev_checkremove && 3001 vd != spa->spa_root_vdev) { 3002 const char *class; 3003 3004 switch (aux) { 3005 case VDEV_AUX_OPEN_FAILED: 3006 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3007 break; 3008 case VDEV_AUX_CORRUPT_DATA: 3009 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3010 break; 3011 case VDEV_AUX_NO_REPLICAS: 3012 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3013 break; 3014 case VDEV_AUX_BAD_GUID_SUM: 3015 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3016 break; 3017 case VDEV_AUX_TOO_SMALL: 3018 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3019 break; 3020 case VDEV_AUX_BAD_LABEL: 3021 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3022 break; 3023 default: 3024 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3025 } 3026 3027 zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3028 } 3029 3030 /* Erase any notion of persistent removed state */ 3031 vd->vdev_removed = B_FALSE; 3032 } else { 3033 vd->vdev_removed = B_FALSE; 3034 } 3035 3036 if (!isopen && vd->vdev_parent) 3037 vdev_propagate_state(vd->vdev_parent); 3038 } 3039 3040 /* 3041 * Check the vdev configuration to ensure that it's capable of supporting 3042 * a root pool. Currently, we do not support RAID-Z or partial configuration. 3043 * In addition, only a single top-level vdev is allowed and none of the leaves 3044 * can be wholedisks. 3045 */ 3046 boolean_t 3047 vdev_is_bootable(vdev_t *vd) 3048 { 3049 if (!vd->vdev_ops->vdev_op_leaf) { 3050 char *vdev_type = vd->vdev_ops->vdev_op_type; 3051 3052 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 3053 vd->vdev_children > 1) { 3054 return (B_FALSE); 3055 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 3056 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 3057 return (B_FALSE); 3058 } 3059 } else if (vd->vdev_wholedisk == 1) { 3060 return (B_FALSE); 3061 } 3062 3063 for (int c = 0; c < vd->vdev_children; c++) { 3064 if (!vdev_is_bootable(vd->vdev_child[c])) 3065 return (B_FALSE); 3066 } 3067 return (B_TRUE); 3068 } 3069 3070 /* 3071 * Load the state from the original vdev tree (ovd) which 3072 * we've retrieved from the MOS config object. If the original 3073 * vdev was offline or faulted then we transfer that state to the 3074 * device in the current vdev tree (nvd). 3075 */ 3076 void 3077 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3078 { 3079 spa_t *spa = nvd->vdev_spa; 3080 3081 ASSERT(nvd->vdev_top->vdev_islog); 3082 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3083 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3084 3085 for (int c = 0; c < nvd->vdev_children; c++) 3086 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3087 3088 if (nvd->vdev_ops->vdev_op_leaf) { 3089 /* 3090 * Restore the persistent vdev state 3091 */ 3092 nvd->vdev_offline = ovd->vdev_offline; 3093 nvd->vdev_faulted = ovd->vdev_faulted; 3094 nvd->vdev_degraded = ovd->vdev_degraded; 3095 nvd->vdev_removed = ovd->vdev_removed; 3096 } 3097 } 3098 3099 /* 3100 * Determine if a log device has valid content. If the vdev was 3101 * removed or faulted in the MOS config then we know that 3102 * the content on the log device has already been written to the pool. 3103 */ 3104 boolean_t 3105 vdev_log_state_valid(vdev_t *vd) 3106 { 3107 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 3108 !vd->vdev_removed) 3109 return (B_TRUE); 3110 3111 for (int c = 0; c < vd->vdev_children; c++) 3112 if (vdev_log_state_valid(vd->vdev_child[c])) 3113 return (B_TRUE); 3114 3115 return (B_FALSE); 3116 } 3117 3118 /* 3119 * Expand a vdev if possible. 3120 */ 3121 void 3122 vdev_expand(vdev_t *vd, uint64_t txg) 3123 { 3124 ASSERT(vd->vdev_top == vd); 3125 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3126 3127 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3128 VERIFY(vdev_metaslab_init(vd, txg) == 0); 3129 vdev_config_dirty(vd); 3130 } 3131 } 3132 3133 /* 3134 * Split a vdev. 3135 */ 3136 void 3137 vdev_split(vdev_t *vd) 3138 { 3139 vdev_t *cvd, *pvd = vd->vdev_parent; 3140 3141 vdev_remove_child(pvd, vd); 3142 vdev_compact_children(pvd); 3143 3144 cvd = pvd->vdev_child[0]; 3145 if (pvd->vdev_children == 1) { 3146 vdev_remove_parent(cvd); 3147 cvd->vdev_splitting = B_TRUE; 3148 } 3149 vdev_propagate_state(cvd); 3150 } 3151